<?page no="0"?> ISBN 978-3-381-12991-1 6 th International Conference on Ignition Systems for SI Engines 7 th International Conference on Knocking in SI Engines MARC SENS (ED.) Ignition Systems for SI Engines Knocking in SI Engines With these two conferences, we at IAV have been offering interested researchers and developers the opportunity to exchange information on the latest developments in the field of ignition and combustion of Otto cycle engines for more than two decades. And this exchange is more urgently needed today than ever before, because the introduction of low and zero carbon fuels of synthetic or biological origin, such as hydrogen, ammonia, methanol, synthetic diesel or gasoline, etc. brings with it new challenges, which in turn require innovative solutions. Only through the exchange of knowledge and ideas will we be successful in making the combustion engine even cleaner and more efficient. Our clear goal is to maintain these two conferences as the “last man standing” for exchange in this highly interesting and important subject area! Marc Sens, IAV GmbH MARC SENS (ED.) <?page no="1"?> 6 th International Conference on Ignition Systems for SI Engines - 7 th International Conference on Knocking in SI Engines <?page no="3"?> Marc Sens (Ed.) 6 th International Conference on Ignition Systems for SI Engines - 7 th International Conference on Knocking in SI Engines <?page no="4"?> DOI: https: / / doi.org/ 10.24053/ 9783381129928 © 2024 expert verlag ‒ Ein Unternehmen der Narr Francke Attempto Verlag GmbH + Co. KG Dischingerweg 5 · D-72070 Tübingen Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlages unzulässig und strafbar. Das gilt insbesondere für Vervielfältigungen, Übersetzungen, Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen. Alle Informationen in diesem Buch wurden mit großer Sorgfalt erstellt. Fehler können dennoch nicht völlig ausgeschlossen werden. Weder Verlag noch Autor: innen oder Herausgeber: innen übernehmen deshalb eine Gewährleistung für die Korrektheit des Inhaltes und haften nicht für fehlerhafte Angaben und deren Folgen. Diese Publikation enthält gegebenenfalls Links zu externen Inhalten Dritter, auf die weder Verlag noch Autor: innen oder Herausgeber: innen Einfluss haben. Für die Inhalte der verlinkten Seiten sind stets die jeweiligen Anbieter oder Betreibenden der Seiten verantwortlich. Internet: www.expertverlag.de eMail: info@verlag.expert Elanders Waiblingen GmbH ISBN 978-3-381-12991-1 (Print) ISBN 978-3-381-12992-8 (ePDF) Bibliografische Information der Deutschen Nationalbibliothek Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet über http: / / dnb.dnb.de abrufbar. <?page no="5"?> 7 9 25 35 51 81 95 125 147 157 197 221 Contents Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jakub Kaleta, Danny Jäger, Sascha Gerhardt, Carsten Kluth, Stefan Angermaier, Mukunda Gopal / Robert Bosch Group Software-Features for optimization of ignition timing . . . . . . . . . . . . . . . . . . . . . . . . . . . Moritz Grüninger, Olaf Toedter, Thomas Koch, Ahmed Assabiki On the Origin of Pre-Ignition inside a Pre-Chamber Spark Plug - Gas Analysis . . . . . Matthias Biehl, Timo Rehm, Philipp Hahn, Carsten Kluth, Rama Krishnan Saravana Kumar Prechamber combustion and the challenges of knock detection . . . . . . . . . . . . . . . . . . . Chandelier, M., Papi, S., Serrano, D., Colliou, T., Duffour, F. Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines Dr. Jakob Ängeby, SEM AB, Andreas Ehn, Lund University Robust ignition and sparkplug wear for H2 SI-ICE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Papi, S., Ricci, F., Dal Re, M., Chandelier, M. Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection . . . . . . . . . . Ryosuke Shiina, Yusuke Shintani, Hirokazu Ando, Noritaka Kimura Consideration of the Relationship between Flame Formation and Fast Combustion in the Second Half of the Combustion Phase of Pre-chamber Jet Combustion . . . . . . . . . Alessandro Nodi, Lorenzo Sforza, Tommaso Lucchini 3D CFD modelling of TJI combustion achieved by active or passive pre-chamber . . . . Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek, Carmen Vesel Assessment of passive TJI technology on a mild hybrid powertrain and its performance on cold operating conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dario Di Maio, Pierpaolo Napolitano, Chiara Guido, Carlo Beatrice, Lorenzo Sforza, Tommaso Lucchini, Stefano Golini Virtual analysis of the Efficiency Gain with Pre-Chamber combustion systems in Heavy-Duty Natural Gas Engine for Long-Haul Truck Application. . . . . . . . . . . . . . . . . Dr.-Ing. Antonino Vacca, M. Sc. Cristian Tortorella, Dr.-Ing. Marco Chiodi, M. Sc. Sebastian Bucherer, Prof. Dr.-Ing. André Casal Kulzer, Dipl.-Ing. Florian Helmut Karl Sobek, M.-Sc. Paul Rothe, Dipl.-Ing Ivica Kraljevic, Dr.-Ing. Hans-Peter Kollmeier, Dipl.-Ing. Albert Breuer, Dr.-Ing. Ruhland Helmut Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation via Active Pre-Chamber System: Numerical Investigation and Experimental Validation . . <?page no="6"?> 257 279 295 313 325 335 351 Dr.-Ing. Thomas Emmrich Assessment of pre-ignition phenomena by thermodynamically approach . . . . . . . . . . . Jan Reimer (KIT-IFKM), Jürgen Pfeil (KIT-IFKM), Frank Altenschmidt (Mercedes-Benz Group AG), Thomas Koch (KIT-IFKM) A fundamental investigation of oil additives on pre-ignition in a high pressure combustion chamber . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Lu Li, Pradeep Sapkota, Pinaki Pal, Yee Chee See, Mingyi Liang, Josep Gomez-Soriano, Sameera Wijeyakulasuriya, Riccardo Scarcelli, Ricardo Novella Simulating Fuel Ignition and Combustion in IC Engines with Lagrangian-Eulerian Spark Ignition (LESI) Model and Detailed Chemistry . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yiqing Wang, Ricardo Scarcelli, Chao Xu, and Ales Srna Modeling the Impact of Mixture Formation on Ignition and Flame Propagation in a Hydrogen Direct-Injection Engine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hermsen, Philipp (TME)*; Plum, Lukas (TME); Günther, Marco (TME); Pischinger, Stefan (TME); Blomberg, Michael (FEV Europe GmbH) Determining the Transition from Auto-Ignition to Knock in Methanol Operation by Application of the Bradley Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Thomas E. Briggs, Jr., Ph.D., Institute Engineer, Powertrain Engineering Division, Southwest Research Institute Observations on pre-ignition in a port injected heavy duty hydrogen internal combustion engine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dr.-Ing. Michael Fischer/ Tenneco GmbH, Dr.-Ing. Michael Grill/ FKFS, Prof. Dr.-Ing. Andre Kulzer/ FKFS & IFS, University Stuttgart, Dr. Ing. Marco Guenther/ TME, RWTH Aachen University, Prof. Dr.Ing. Stefan Pischinger/ TME, RWTH Aachen University Higher Efficiency through model-based, predictive Knock Control . . . . . . . . . . . . . . . . 6 Contents 6 <?page no="7"?> Preface I would like to take this opportunity to thank my fantastic Scientific Advisory Board, which contributes the following thoughts to our conferences: “These conferences bring together all the major players and experts in Combustion devel‐ opment sharing technical developments and new technologies to improve the efficiency of the Internal Combustion Engines. The introduction and development of alternative fuels brings up challenges which need new solutions. Such will be presented and discussed, helping OEMs to work with partners to ensure rapid and efficient introduction into production.” Sandro Pino, Federal Mogul Powertrain, Tenneco Champion “While electrified technologies will play a pivotal role in the future of transportation and energy, internal combustion engines (ICEs) continue to offer key advantages in many applications. Their function is likely to evolve, incorporating hybrid systems, alternative fuels, and continuous improvements in efficiency and emissions control. Conferences such as 'Ignition Systems for SI Engines & Knocking in SI Engines' provide a platform for global technical experts to exchange innovative ideas for enhancing engine performance. As these discussions and innovations continue, ICEs will remain a critical component in the transition to a more sustainable energy landscape.” Dr. Kelly Senecal, Convergent Science “Even two decades after the establishment of this conference series, the topics of 'ignition and combustion' are more relevant than ever. New technologies such as pre-chamber ignition and the use of alternative fuels such as hydrogen, methanol and ammonia open the door to the further development of the internal combustion engine demanded by the society. This will not succeed without a deeper understanding of the engine combustion process, and these both conferences will make a substantial contribution to this.” Dr.-Ing. Michael Fischer, Tenneco GmbH “Reducing greenhouse gases in a short time period can only be reached by enabling all technological tools including a fast ramp-up of renewable fuels like methanol, ammonia or hydrogen. High ignitability causes pre ignition Problems and low ignitability causes ignition challenges. We show at these both conferences how to solve it.” Dr.-Ing. Olaf Toedter, Karlsruher Institut für Technologie (KIT) <?page no="8"?> “The mobility sector is in its greatest transition ever. To master this challenge on a global scale, we will need all technical solutions available. Advanced combustion processes and sustainable fuels still provide big benefits for the existing vehicle fleet as well as new vehicles. This conference is the best platform for researchers from all over the world to discuss latest research results for advanced combustion and ignition technologies.” Dr.-Ing. Michael Fischer, HONDA R&D Europe “The future will be sustainable by electrification and hybridization, hence renewable energy carriers and electricity will be of utmost importance! But renewable energy won’t be cheap. This means that the efficiency of hybridized powertrain systems has to be higher and further developed! And there, the complete powertrain has its optimization tasks as also combustion efficiency has to be further improved! These both conferences, organized in one event, addresses the core of ICE efficiency improvement, looking at the most important phenomena like ignition and knocking in the Otto engine! ” Prof. Dr. André Casal Kulzer, Stuttgart University In line with the statements of the Advisory Board, we look forward to the next edition of the conference in 2026. Marc Sens, IAV GmbH 8 Preface 8 <?page no="9"?> Software-Features for optimization of ignition timing Jakub Kaleta, Danny Jäger, Sascha Gerhardt, Carsten Kluth, Stefan Angermaier, Mukunda Gopal / Robert Bosch Group Abstract The continuously enhancement of the efficiency and reduction of emissions of spark-ignition engines is task of system and hardware development on one side, and function and software development on the other side. The increasing stringency of legal requirements has led to a continuous rise in engine complexity in recent years. For instance, significant efficiency improvements under stationary conditions on the test bench have been achieved through variable valve timing and low-pressure exhaust gas recirculation. Notably, the ignition timing itself significantly determines the efficiency of in-cylinder combustion. The conventional approach for ignition timing or ignition angle calculation, primarily based on calibratable maps depending on relevant physical input signals. But this approach is increasingly limited under real driving conditions, where frequent transient operating conditions occur. This is mainly due to cross-influences of various control parameters, whose actual values often deviate from their (stationary) set positions during transient operation. By utilizing Software-Features developed by the Robert Bosch Group the determi‐ nation of an optimal ignition timing, across the entire parameter space to further enhance the efficiency of modern, complex engines in real driving, can be improved (see Fig. 1): - Ignition setpoint determination with a data-based model (hereinafter called ASC@ECU IgnSp) - Ignition Advance by knock control (hereinafter called IadKnock) - Transient Ignition Setpoint Adaptation (hereinafter called TISA) <?page no="10"?> Figure 1: Overview of Software-Features for ignition timing optimization 1 Ignition setpoint determination with a data-based model (ASC@ECU IgnSp) 1.1 Introduction and Motivation To achieve the highest emission and efficiency standards, while fulfilling high requirements of power density and dynamic qualities, modern internal combustion engines need to be operated as close as possible to the thermodynamic optimum or the knock limit under all operating conditions. At the same time the complexity of the engines is still growing and an increasing number of actuators requires more and more input parameters and dependencies to be taken into account in the ignition angle model. Achieving these goals with conventional ignition angle models based on 1or 2-dimensional calibration maps and arithmetic calculations is increasingly difficult: - Nonlinear dependencies between input parameters cannot be modelled with sufficient precision as illustrated by Fig. 2. - Conventional models are accurate, but mostly under stationary conditions - under dynamic operation, when actuator positions are off the desired values, the models loose accuracy leading to increased fuel consumption and CO 2 emissions - Adding more parameters and dependencies to the model increases the software development effort, makes the software more complex and less maintainable and also increases the calibration effort 10 Jakub Kaleta, Danny Jäger, Sascha Gerhardt, Carsten Kluth, Stefan Angermaier, Mukunda Gopal 10 <?page no="11"?> Figure 2: The center of combustion (MFB50, 50-% of mass fraction burned) needed for maximal combustion efficiency varies depending on the operating conditions as does the combustion velocity. The resulting dependency between the ignition timing needed to achieve this optimal center of combustion and the operating conditions is nonlinear and not fully described by a conventional map-based approach. To address these points parts of the conventional ignition angle calculation chain can be replaced with a data-based Gaussian Process Regression model trained on engine test bench data. For better convenience the Robert Bosch Group provides a standardized workflow and a corresponding Toolsuite stretching from designing the model and planning the necessary measurements to deployment on the engine control unit (ECU). This workflow is referred to an Advanced Simulation and Calibration on the engine control unit (ASC@ECU) and is supported by the ETAS ASCMO calibration software. 1.2 Gaussian Process Regression (GPR) GPR is a nonparametric regression method, meaning that no explicit function structure is assumed. Rather, the model is constructed from basic assumptions about the smoothness of the unknown true function and about the measurement. Combined with measurement data this results in a probability distribution for model output. That means each model evaluation not only produces a predicted value but also gives a measure of confidence as shown in Fig. 3. Software-Features for optimization of ignition timing 11 11 <?page no="12"?> (1) (2) (3) Figure 3: The confidence interval of the GPR model prediction is smaller if more data points are used (right figure) compared to sparser data (left figure). The black dots represent measurement data, generated from an unknown true function plus an unknown amount of noise. The model prediction generated by the GPR method is shown as a blue curve. The shaded area shows the 1-sigma confidence interval about the model prediction. In the right panel, more data points were used than in the left panel, resulting in smaller confidence intervals of the model prediction. For x values around the value of 1, a gap in data was assumed. This results in smaller model confidence in this region. The same is true for x values outside of the region covered by training data (beginning and end of x axis). Another key feature resulting from the probabilistic approach is automatic handling of the tradeoff between model complexity and generalization, which addresses the problem of overfitting. The formula for the GPR model predictions given by: v = k * T K + σ n2 I −1 y Where K is the covariance matrix, K = k x 1 , x 1 k x 1 , x 2 … k x 2 , x 1 ⋱ … ⋮ ⋮ k x N , x N incorporating the squared exponential kernel function. k x i , x j = σ f2 • e − 12 (∑ d ( (xi, d − xj, d)2 ld2 )) The vector k(u) contains the covariances between the test point and the n training points. 12 Jakub Kaleta, Danny Jäger, Sascha Gerhardt, Carsten Kluth, Stefan Angermaier, Mukunda Gopal 12 <?page no="13"?> (4) k * T = k x 1 , u … k x N , u Here u is the input position where the model is evaluated, N is the number of training points with x = x 1 … x N (d-dimensional) and y = y 1 … y N . Furthermore, l, σ f , σ n are additional model parameters (hyperparameters), which are determined during model training, see (Rasmussen & Williams, 2006) for details. 1.3 ASC@ECU workflow from functional concept to ECU deployment 1.3.1 Functional structure Likewise to the conventional approach the functional concept is derived from the project specific requirements. The main goals are to select relevant input variables and design a suitable functional structure. Although a full data model calculating the final ignition angle directly from a set of input variables could be used, most real applications of ASC@ECU use a hybrid model. This structure combines a data-based model with conventional calculations based on calibration maps and arithmetic operations. This has several benefits: - More simple dependencies without significant cross-influence of other relevant input variables can still be considered in the classical mapand curve-based approach. - By reducing the model input parameters, the number of required training data / meas‐ urement points decreases. - offset calculations for special conditions and proven in use can be retained (e.g. ambient compensation or switching from HOM to HSP injection mode) The ASC based ignition angle calculation uses a model with typically four to six input variables like engine speed, cylinder air charge and camshaft positions, but other project specific variables like the air-fuel-ratio or EGR-rate could be added. The model is trained to calculate a calculation base for the best possible ignition angle under stationary basic conditions and then other influences like the intake air or coolant temperature are added similarly as in the conventional software. 1.3.2 Data acquisition The data needed for model training is usually obtained during a measurement campaign on the engine testbench. The ASCMO software can be used to design an optimized measurement campaign (DoE - Design of Experiments). Depending on the number and constraints of the input parameters on the one side and the desired model precision on the other side, ASCMO automatically creates a list of necessary measurement points optimally covering the multidimensional feature space according to a Sobol Sequence. On the other hand, the ASC model training can be also done based on existing data from previous measurements or calibration campaigns to achieve good results in an early project phase. In a later stage the model quality can be improved by adding additional data to areas with lower precision. Software-Features for optimization of ignition timing 13 13 <?page no="14"?> 1.3.3 Model training Once the measurements are available the ASCMO software is used to automatically train a GPR-model for the best possible ignition angle. The software provides convenient tools to assess the model quality and optimize the results, e.g. by identifying outliers. By using advanced AI (Artificial Intelligence) methods from the field of machine learning, it is possible to accurately model as well as analyze and optimize the behavior of complex systems based on a relatively small set of measurement data. 1.3.4 ECU deployment Almost all Bosch MDG1 ECUs feature a dedicated Hardware unit (AMU - Advanced Modelling Unit) optimized for the evaluation of exponential functions. The AMU is used to evaluate ASC models in real time. This would be not possible on the normal ECU hardware, as the evaluation of the exponential functions would take too long. The ASC model needs to be compressed with ASCMO for the usage in the ECU. This is done by resampling the fully trained model with a significantly lower number of artificial data points with minimal losses in model accuracy. 1.4 Demonstration of potential based on real measurements The ASC@ECU-based (hereinafter called ASC-based) ignition angle model is already implemented in several systems with which the benefit of the approach could be depicted. In Fig. 4 it is depicted that the model deviation for ignition angle for maximum engine efficiency of the conventional-based model is almost seven times higher than the model deviation of the ASC-based model. This means that the ignition timing is more precise and closer to the optimum leading at the end to a lower fuel consumption. The data was captured on test bench with an engine with 2.0-liter engine displacement, 4 cylinders, fully variable valve train, direct injection in homogeneous mode. Figure 4: Model deviation for ignition angle for maximum engine efficiency 14 Jakub Kaleta, Danny Jäger, Sascha Gerhardt, Carsten Kluth, Stefan Angermaier, Mukunda Gopal 14 <?page no="15"?> The ASC-based model shows benefits in stationary, but also in dynamic driving situations in which (for example) the air charge, engine speed and camshaft positions are changing dynamically. The measurement data for Fig. 5 to Fig. 7 are derived from a powertrain test bench with a 1.5-liter engine, 4 cylinders, fully variable valve train, external low pressure exhaust gas recirculation and direct injection. WLTC stands for Worldwide harmonized Light vehicles Test Cycle and represents a standardized test-cycle for light-duty vehicles for determining exhaust emissions (pollutant and CO 2 emissions) and fuel consumption. Fig. 5 depicts a WLTC sequence with partially knock-limited operating conditions. The black lines depict the measurement [1] with conventional ignition angle calculation, while the green lines represent the behavior with the ASC@ECU approach [2]. The following measured parameters are depicted: Engine speed, relative air charge in the cylinder, low-pressure exhaust gas recirculation rate, MFB50, mean knock intervention, and the ignition angle difference between both approaches calculated from measurement [2]. A more precise ignition angle calculation with ASC@ECU results in efficiency advantages, evident in the more favorable center of combustion positions (MFB50). Figure 5: WLTC sequence with conventional map-based compared to ASC-based ignition angle calcu‐ lation Fig. 6 is illustrating for a WLTC-cycle that the ignition angle determined by the ASC-model is closer to the optimum, especially for lower or medium relative cylinder air charges. Software-Features for optimization of ignition timing 15 15 <?page no="16"?> Figure 6: Measured MFB50 in WLTC-cycle. Conventional-based (black dots) versus ASC-based (green dots) ignition angle setpoints. Sampling rate: 100ms The usage of an ASC-based ignition angle determination leads to a lower fuel consumption what can be depicted by the indicated specific fuel consumption (ISFC) in WLTC-cycle (see Fig. 7). It shows that especially for low and medium air charge the fuel consumption is lower with the ASC-based model. Figure 7: Indicated specific fuel consumption (ISFC) in WLTC-cycle. Conventional-based (black dots) versus ASC-based (green dots) ignition angle determination 1.5 Summary of key benefits The ASC@ECU approach for the ignition setpoint determination has shown several key benefits over the conventional-based approach: Reduction of fuel consumption and thus CO 2 emissions by greater than 1 % due to an overall improvement of ignition angle accuracy. Reduced knocking risk under dynamic engine operating conditions. Robustness of the ignition angle model against calibration (setpoint) changes late in the project and improved software maintainability due to standardized functional design. Reduced software development effort for the introduction of new input parameters or new complex dependencies. 16 Jakub Kaleta, Danny Jäger, Sascha Gerhardt, Carsten Kluth, Stefan Angermaier, Mukunda Gopal 16 <?page no="17"?> Reduced calibration effort and nevertheless an increased degrees of freedom in the parameter optimization of the parameters influencing the ignition timing. The data models behave like maps at the edges of the parameter space. Unwanted extrapolation does not occur outside of these definable values. Visualization can be carried out in the usual way by plotting maps with the ASCMO tool. Applicable also for other engines, like hydrogen internal combustion engines. 2 Ignition Advance by Knock control (IadKnock) The features TISA and ASC@ECU IgnSp can be combined with an ignition advance algorithm based on knock detection. Target of this algorithm is to operate the engine at the knock border or the thermodynamically optimal set point. The algorithm considers the center of combustion, the acceptable mean or maximum cylinder pressure and the knock frequency. It is based on a holistic knock detection and control approach which evolves the classic knock detection further to a real optimizer of the ignition timing under all operation conditions, compensating environmental influences, as well as system tolerances and ageing effects of the engine. The algorithm and further information are presented in detail in the paper (Benzinger & Biehl, 2022). 3 Transient Ignition Setpoint Adaptation (TISA) 3.1 Introduction Not only under (quasi-) stationary engine operating conditions or combustion chamber temperature conditions is it important to pre-control the ignition angle optimally. Tran‐ sient, dynamic operating conditions can represent a significant, fuel consumption-relevant proportion depending on the powertrain concept and driving profile. Especially during acceleration phases, in which the engine must be operated at higher loads far from the MFB50-optimum due to knock-limited ignition angle, there is often untapped potential for further CO 2 reduction. Coming from part-load or coasting operation, it can take several seconds for the combustion chamber temperature to reach stationary level after a load jump. Since the combustion chamber temperature has a major influence on mixture formation and tendency to knock, it is obvious that the ignition angle suitable for stationary conditions cannot represent the efficiency optimum for non-stationary conditions. Depending on the temperature difference to the thermally stable condition, the knock limit shifts towards earlier MFB50 with significantly better efficiencies. With the feature TISA this dependency is considered by advancing the ignition angle. The concept is not novel. A patent from Mitsubishi dating back to 1992 describes an ignition angle calculation that considers the current cylinder wall temperature (USA Patentnr. 5150300, 1992). See excerpt in Fig. 8. Software-Features for optimization of ignition timing 17 17 <?page no="18"?> Figure 8: Extract from patent US5150300 At Robert Bosch Group, piston surface temperature is used as an equivalent to the combustion chamber temperature, instead of the cylinder wall temperature. The difference to the respective steady-state piston surface temperature is the reference variable for ignition setpoint correction. Over the years, valuable experience has been gained in the application of a corresponding temperature model for injection angle correction (TSA: Transient SOI-Adaptation) in various series projects. SOI stands for Start Of Injection. Here, a temporary SOI correction prevents or reduces fuel wetting on the colder piston surface at the beginning of a load jump, thereby reducing particle formation during subsequent combustion. Ideally, both software features (TSA & TISA) are combined for the best possible reduction in particle emissions and fuel consumption during dynamic engine operation. 3.2 Functional Approach Fig. 9 schematically illustrates the functionality of TISA. The suitable spark advance is promptly considered depending on the engine speed, cylinder air charge, and temperature difference. If there is a (very short term) dynamic intervention of the knock control (in re‐ tarded direction), this can be considered, and TISA is enabled with a slight delay. Depending on the gradually decreasing piston surface temperature difference, the positive ignition angle offset is stepwise reduced. In the event of detected knock control interventions, the spark advance can be prematurely reduced or terminated on a cylinder-individual basis, depending on the intensity of the knock control intervention. The speed of this ignition angle adjustment must be limited to the rate at which the air system can compensate for the corresponding change in desired air charge. This will prevent any unwanted noticeable torque fluctuations. 18 Jakub Kaleta, Danny Jäger, Sascha Gerhardt, Carsten Kluth, Stefan Angermaier, Mukunda Gopal 18 <?page no="19"?> Figure 9: Schematic representation of the functionality of TISA In combination with an advanced ignition angle correction for variable environmental or fuel conditions (engine temperature, intake air temperature, humidity, fuel octane rating), it can be achieved that the ignition timing is set as close as possible to the knock limit in order to optimize the fuel efficiency for different conditions. In this context, it becomes clear that simultaneously acting ignition angle offset correc‐ tions (in both directions) must be coordinated with each other. Different physical influences must be considered separately in the ECU software to ensure optimal system behavior under all conditions. A known example of this is the intake air temperature, which, like the combustion chamber or piston surface temperature, has a significant influence on the shifting of the knock limit. At a higher level, there is the functional option to apply a spark advance limit that is dependent on operating point and calibration parameters. This limit can, for example, represent the respective thermodynamic ignition angle optimum, a maximum permissible peak cylinder-pressure, or a maximum tolerable combustion pressure gradient. When using the ASC@ECU approach for the stationary base ignition angle, this data model-based advance limit is already included. Alternatively, a conventional functional structure (map-/ curve-based) can be used. All these options are available in the platform software of the Robert Bosch Group. 3.3 Piston surface temperature model The piston surface temperature is provided by an appropriate model in the engine control unit (ECU). The model itself will not be further discussed here. To calibrate the model, a suitable reference signal is required. The measurement of piston surface reference temperature is indirectly carried out using an infrared temperature measurement system from FOS Messtechnik GmbH (Fig. 10). For this purpose, a special spark plug with optical access is installed in a suitable cylinder. The signal recording can be crank angle-resolved using an existing indication system. Software-Features for optimization of ignition timing 19 19 <?page no="20"?> Figure 10: Measuring system used for piston surface temperature 3.4 Proof of concept for piston surface temperature model Regardless of the starting condition a positive load jump into the knock-limited range occurs, the actual temperature profile must be accurately mapped. Only then can a suitable spark advance be determined based on the temperature difference between the stationary and current state. Without this, efficient calibration of ignition angle adjustment for unsteady states becomes challenging. An underestimated piston surface temperature causes more difficulties than an overestimated temperature. If the piston surface is hotter than the model value suggests, the calculated ignition angle may be too early, leading to unwanted heavy knocking interventions on several cylinders at the beginning of the load jump. On the other hand, the ECU knock detection also offers the possibility to validate and adjust the piston surface temperature model if necessary. Fig. 11 shows a load jump sequence performed on the engine test bench at constant speed (3000 rpm), constant start load (1) and constant target load (3). The ignition angle for this target load was exemplarily advanced by 3°CA compared to the stationary. In this test, the initial value of the piston surface temperature has been varied by different durations of the Fuel Cut Off phase (2), hereinafter called FCO. The expectation was that the first knock events always occur above a certain threshold of the modelled surface piston temperature. All engine measurements shown below were performed on a 2.0L direct injection, turbocharged engine with variable intake and exhaust camshaft. 20 Jakub Kaleta, Danny Jäger, Sascha Gerhardt, Carsten Kluth, Stefan Angermaier, Mukunda Gopal 20 <?page no="21"?> Figure 11: Variation of FCO duration before a load jump into knock limited area at constant engine speed (3000-rpm) As expected, the first knock events provoked by the constant ignition angle offset always occurred at a very similar modelled piston surface temperature (475-478°C), which was reached after different durations at defined high load condition, depending on the preceding overrun time. This general behavior was also observed in other load jump sequences, demonstrating the fundamental applicability of the TISA functionality. 3.5 Demonstration of potential based on real measurements Figure 12: Load jump into full load at constant 2500-rpm after FCO phase; 2.0L GDI, TC, VVT engine Software-Features for optimization of ignition timing 21 21 <?page no="22"?> Fig. 12 exemplifies a load jump performed on the engine test bench from the FCO-operation to full load at a constant engine speed (2500 rpm). The ignition angle was adjusted by a maximum of 3 degrees advancing in the knock-limited range compared to the stationary optimum. The TISA intervention (green) was enabled for approximately 9 seconds. For comparison, all signals without TISA are shown in blue. By advancing the MFB50, the cylinder charge required to achieve the desired torque was significantly reduced, resulting in a 5 % decrease in fuel consumption during this transient phase. In addition, the exhaust gas temperature could be lowered by approximately 35K. A second example (Fig. 13) illustrates a sequence of the ADAC highway cycle BAB130. The activation conditions for TISA are frequently met through repeated deceleration and subsequent acceleration to 130 kph. The computed difference of the piston surface temperature (in blue), the TISA ignition angle intervention (in light green), and the relative fuel mass with (in green) and without TISA (in black) are depicted. The required fuel quantity could be significantly reduced during the TISA activation periods with comparable indicated mean pressure (IMEP), resulting in an overall 1.7 % reduction in CO 2 emissions through a maximum ignition timing advance of only 3°CA. Figure 13: Sequence of BAB130 test cycle, potential fuel consumption reduction through a max. spark advance of only 3°CA Not only fuel consumption and thus CO 2 emissions can be reduced during dynamic engine operation through TISA, but also the particle number (PN) of the raw exhaust gas can be decreased. As the engine can be operated at better ignition angle efficiency during load jumps, the cylinder charge and thus the injection quantity can be reduced. This often has a positive effect on soot particle number reduction, especially during dynamic engine operation. Example with a 20-% reduced relative cylinder filling see Fig. 14. 22 Jakub Kaleta, Danny Jäger, Sascha Gerhardt, Carsten Kluth, Stefan Angermaier, Mukunda Gopal 22 <?page no="23"?> Figure 14: Potential of PN-reduction during load jump at 1750 rpm through reduction of relative cylinder charge by 20-% 3.6 Summary of the benefits and positive side effects With the TISA software feature at the Robert Bosch Group, it is possible to close the gap between stationary ignition setpoint calibration in the knock-limited operating area and the consideration of actual temperature conditions inside the combustion chamber during dynamic engine operation. In summary, the advantages are as follows: - Improved fuel efficiency in non MFB50-optimal operating area means a considerable reduction of CO 2 emission under real driving condition. - Potential to reduce raw emissions (PN, CO). - Improved engine response and performance in dynamic driving. - Enhanced drivability in the low-end torque range; torque build-up is more reliant on ignition angle efficiency than on cylinder air charge build-up, resulting in a more harmonious acceleration. - More stable and robust detection of low octane (poor fuel quality) for ECU knock control. 4 Conclusion In the chapters before it has been demonstrated that there is still potential to increase the efficiency of modern complex spark-ignited combustion engines significantly just by adding and calibrating innovative software features for ignition control. These Software-Features for optimization of ignition timing 23 23 <?page no="24"?> features have been validated in several projects. The validation has shown benefits of the features individually but also in the combination. These and further software solutions and services are provided by the Robert Bosch Group. References Benzinger, M., & Biehl, M. (2022). Holistic knock detection and control as the key to optimum ignition timing. In: International Conference on Ignition Systems for Gasoline Engines - In-ternational Conference on Knocking in Gasoline Engines, pp 525---535 Rasmussen, C. E., & Williams, C. (2006). Gaussian Processes for Machine Learning. The MIT Press. Yoshiaki, D., Kazuhide, T., Hiromitsu, A., Jun, T., Eiichi, K., & Ichiyo, K. (1992). USA Patentnr. 5150300. Ignition timing controller for spark-ignition internal combustion engine using estimated cylinder wall temperature, p.-5 (fig.-4 and 5) 24 Jakub Kaleta, Danny Jäger, Sascha Gerhardt, Carsten Kluth, Stefan Angermaier, Mukunda Gopal 24 <?page no="25"?> On the Origin of Pre-Ignition inside a Pre-Chamber Spark Plug - Gas Analysis Moritz Grüninger, Olaf Toedter, Thomas Koch, Ahmed Assabiki Abstract Pre-chamber spark plugs are widely used to increase the efficiency of internal combustion engines in order to meet future CO 2 emission regulations. Unfortunately, pre-ignition at higher load operation points in combination with pre-chamber ignition systems has been reported in the past [1]. Two main assumptions for pre-ignition are (1) pre-ignition to be triggered either by hot pre-chamber surface components or (2) by hot and still reactive residual gases present inside the pre-chamber from previous combustion cycles. Assumption (1) was examined in a previous publication [2] and not found to be the leading cause of pre-ignition. However, with the help of optical pre-chamber spark plugs, it was discovered that the origin of the pre-ignition is located in the upper part (around the insulator) of the pre-chamber. For this reason, the gas composition in the upper part of the pre-chamber is to be examined in detail and the CFD calculations thereby validated. The objective of this study is to analyze the residual gas composition inside a pre-chamber spark plug while running in an engine. For this purpose, a heavy-duty natural gas engine, that is operated with a M14 pre-chamber spark plug, will be run purposely up to a point where pre-ignition occurs. A small gas probe is to be sampled from the internal pre-chamber spark plug volume at various times during the combustion cycle and gas analytical measurements are to be performed. The gas is extracted from the area identified in the previous publication as the source of the pre-ignition. The sampling is controlled by a fast-switching valve so that the sampling time and duration can be selected precisely. In order to achieve the necessary sensitivity for measurement of the low sampling volume flow, the sample is diluted with nitrogen in a controlled manner. In addition to carbon dioxide, unburned methane was also detected in the pre-chamber spark plug throughout the entire combustion cycle. These findings corroborate the hypotheses proposed in the previous publication, namely that the upper area of the pre-chamber spark plug only undergoes combustion at a relatively late stage, which allows for the presence of unburned gases and radicals that can lead to pre-ignition. However, further measurements are necessary to obtain a comprehensive analysis. <?page no="26"?> 1 Scope and Motivation The automotive industry has been engaged in research into pre-chamber combustion technology with the objective of enhancing the efficiency of gasoline combustion engines for a number of years now. The technology can be classified into two concepts: passive and active pre-chamber spark plug combustion. In passive pre-chamber spark plug technology, the fuel is mixed with air inside the main combustion chamber (engine cylinder) and then pushed into the pre-chamber during compression. In contrast, in active systems, fuel is introduced directly into the pre-chamber. The concept of passive pre-chamber combustion is regarded as the cost-optimal solution in the medium term due to its reduced complexity in comparison to the active pre-chamber system. For an overview of the literature on passive pre-chamber combustion, please refer to [3, 4]. The utilisation of passive pre-chamber spark plugs in series production presents a number of advantages, including the potential for faster combustion and lean-burn capa‐ bility. However, the necessity for their functionality in challenging marginal areas must be considered. Such conditions include the operation under high loads and the heating of the three-way-catalyst (operating the engine under low loads with delayed combustion center positions). At elevated loads, the potential operating range of pre-chamber plugs can be highly constrained, such that even a minor alteration in the ignition timing, by a single degree of crank angle, can result in the initiation of strong, consecutive pre-ignitions. It is only possible to control these pre-ignition series by cutting off the fuel supply. The two primary assumptions regarding pre-ignition are as follows: (1) pre-ignition can be triggered either by hot pre-chamber surface components or (2) by hot and still reactive residual gases present inside the pre-chamber from previous combustion cycles. In a previous paper [2], the hypothesis that component temperatures are the cause of pre-ignition was tested with the aid of thermal measuring spark plugs, which yielded results that did not support this hypothesis. However, investigations utilising optical pre-chamber spark plugs have demonstrated, that the source of the pre-ignition phenomenon is situated in the upper region of the pre-chamber spark plug. The aforementioned volume is insufficiently scavenged in the pre-chamber spark plug under examination. It is conceivable that combustion residues are not adequately flushed out, which could result in pre-ignition during the subsequent cycle. To validate this thesis and the preceding CFD calculations on scavenging, it is necessary to extract and analyze the gas in the upper area of the spark plug while the engine runs in high load close to pre-ignition operating point. Preceding CFD calculations indicated the presence of a prolonged combustion flame in the region of interest, well after top dead center. The objective of this study is to establish a measurement system for the analysis of gases in the upper region of a pre-chamber spark plug, with the aim of identifying the source of pre-ignitions that occur under conditions of increased load. 26 Moritz Grüninger, Olaf Toedter, Thomas Koch, Ahmed Assabiki 26 <?page no="27"?> The following working hypotheses have been formulated: 1. Unburned and reactive gases from the main combustion chamber are pushed into the pre-chamber volume. 2. The unburned/ reactive gases from inside the pre-chamber remain within the system until the beginning of the subsequent cycle. Both hypotheses assume that the compression of the remaining reactive gases and the fresh methane which is pushed into the pre-chamber leads to self-ignition. 2 Engine and Pre-Chamber Description 2.1 Engine and Test Bench Characteristics and Operating Point The engine utilized is a single-cylinder medium-duty diesel engine (type BR2000) manu‐ factured by MTU. The engine has been modified to function as an intake manifold-injected spark-ignited engine. In place of the diesel injector, a spark plug is positioned within the cylinder head, and the compression ratio is adjusted to 12.5: 1 through the modification of the piston. The fuel utilized is natural gas drawn from the local natural gas network. The utilization of a medium-duty engine allows for a detailed examination of pre-ignition phenomena due to its inherent durability. Consequently, the engine can be operated for a certain duration in areas, where pre-igniting combustion occurs. The operating point was selected to replicate the conditions of the previous test series conducted by Rosenthal et al. [1] in 2018, ensuring high reproducibility. The following table illustrates the operating point at which pre-ignitions can be generated in a reproducible manner. Speed / rpm 1500 Mean indicated pressure / bar 12.3 MFB50 / °CA-aTDCf ≤ 8 Energy from fuel per cycle / kJ/ cycle 5.9 Lambda / - 1.45 Tab. 1: Operating point with occurring pre-ignitions The engine is operated on a modern engine test bench, which allows for the setting of reproducible operating conditions. The fuel gas is subjected to continuous analysis by a gas chromatograph in order to enable the compensation for fluctuating calorific values, thus allowing the establishment of a reproducible operating point. The implementation of automatic, pressure-based safety monitoring ensures that short-term operation in pre-ignition conditions can be conducted in a safe manner. On the Origin of Pre-Ignition inside a Pre-Chamber Spark Plug - Gas Analysis 27 27 <?page no="28"?> 2.2 Pre-Chamber Spark Plug The passive pre-chamber spark plug utilized in this investigation was developed for medium-sized gas engines in which M14 pre-chamber spark plugs can be employed. A distinctive attribute of this passive pre-chamber spark plug is the incorporation of a ring-shaped ground electrode, comprising a flat disk with a central aperture containing a ring carrier with precious metal. In conjunction with an elongated and protruding center electrode situated within the center hole of the ground electrode disc, a ring-shaped spark gap measuring approximately 0.3 mm is formed. This electrode configuration was designed with the objective of ensuring a long service life, and a center electrode with a copper core allows rapid heat dissipation. The ground electrode disk is positioned between the end of the shell and the pre-chamber cap. It is attached using laser welding. Three supplementary kidney-shaped apertures are incorporated into the design of the ground electrode to facilitate enhanced gas exchange between the lower and upper sections of the pre-chamber volume. Prior research [1,2] has indicated, that the aforementioned holes are inadequate for the complete scavenging of the pre-chamber spark plug, and instead serve to impose an aerodynamic resistance. Furthermore, the pre-chamber cap is equipped with four radially arranged holes, each with a diameter of 1.2 mm. The center lines of the holes are oriented in a direction aligned with the central electrode tip. Additionally, the core nose of the insulator and the shell ridge on which the insulator rests are also situated in the upper part of the pre-chamber. The total volume of the pre-chamber is 813 mm³, with 322 mm³ between the inner surface of the cap and the ground electrode ring (lower section) and 491 mm³ behind the ground electrode ring (upper section). For further information regarding the spark plug characteristics, refer to [2]. Fig. 1: Pre-chamber spark plug 3 Measurement System for Gas Anaylsis 3.1 Gas-Extraction Pre-Chamber Characteristics In order to obtain reliable measurement results as close as possible to the original series system, the gas extraction channel has to be installed into a series M14 spark plug with only minimal changes to the geometry of the pre-chamber volume. Furthermore, the extraction process must be precisely controllable in order to facilitate the extraction of gas at a defined point in the combustion cycle. This necessitates the use of a fast-switching valve. To avoid any distortion of the gas composition, it is essential to maintain the shortest possible length and volume of the line between the pre-chamber 28 Moritz Grüninger, Olaf Toedter, Thomas Koch, Ahmed Assabiki 28 <?page no="29"?> and the valve. However, the diameter has not be too small, as otherwise the friction losses will impede the extraction process. In order to satisfy these requirements, a capillary with an internal diameter of 1.8 mm was attached to the body of an M14 pre-chamber spark plug by laser welding. Holes of 0.8 mm diameter are drilled through the spark plug body terminating in the upper volume surrounding the core nose. Due to packaging limitations of the cylinder head, the gas sampling valve cannot be positioned in closer proximity than ~9-cm to the sampling point in the pre-chamber. Fig. 2: Gas-extraction pre-chamber spark plug To ensure the validity of the operating conditions, it is necessary to extract only a minimal quantity of gas, as otherwise the spark plug would be actively scavenged, thereby falsifying the operating conditions. As a preliminary estimation, approximately 1-% of the pre-chamber volume, or 8 mm³, should be removed per cycle. Assuming an extraction pressure of 30 bar and a motor speed of 1500 rpm, the resulting extraction flow is 180 ml/ min. To maintain the maximum 1 % volume extraction, the extraction duration must be adjusted depending on the pressure at the time of extraction. The aforementioned sample gas flow is insufficient for the purposes of exhaust gas analysis. Accordingly, the sample gas is diluted with nitrogen. A dilution section has been implemented for this purpose, which regulates a nitrogen volume flow to which the sample gas is added. The total flow of sample gas and nitrogen is subsequently measured and controlled. The dilution is quantified using the following formula: f D = V˙ tot V˙ P C where f D is the factor of dilution, V˙ tot is the total extraction gas flow and V˙ P C the gas flow extracted from the pre-chamber spark plug. Subsequently, the diluted sample gas is fed into a FTIR (Fourier Transform Infrared Spectroscopy) spectrometer for analysis, with a focus on identifying and quantifying 28 specific components typical for CNG combustion. This analysis is conducted and documented at a rate of 5 Hz on a continuous basis. To obtain the concentration of the components in the undiluted gas, the individual concentrations are multiplied by the dilution factor. It is important to ensure that the gas transit time from the dilution section to the FTIR is considered. Using repetitive combustion cycles and pressure signals, the gas analysis measurements are projected onto the crank angle of the engine. On the Origin of Pre-Ignition inside a Pre-Chamber Spark Plug - Gas Analysis 29 29 <?page no="30"?> As the FTIR necessitates a specified sample gas flow, in this instance 3 l/ min, an overflow valve is incorporated into the system following the dilution section, whereby a proportion of the higher total flow is discharged after the dilution section. To evaluate the gas composition at different points in the combustion cycle, the extraction valve can be independently actuated by the single-cylinder engine control unit. Previous measurements have shown long combustion durations within the prechamber. The CFD simulation has also shown long burnouts in the upper prechamber volume [2]. To confirm this behavior of the current pre-chamber spark plug, gas is sampled at several points in time after top dead center. A decreasing methane concentration and increasing CO 2 concentration with advancing crank angle degrees after top dead center would confirm the previous observations. To answer the underlying question of how pre-ignition is triggered, gas is also sampled and analyzed at various points in the compression cycle prior to spark timing. 3.2 Measuring procedure After preliminary investigations and testing of the developed measurement equipment, the gas-extraction spark plug was operated in the above-mentioned operating point to investigate the gas composition inside the pre-chamber before pre-ignitions occur. A stationary engine operating point was established for the measurement (see Tab. 1). Subsequently, the sampling angle of the gas sampling valve is set to the desired point in time in the combustion cycle, and the opening duration is adjusted so that the sampling volume flow is approximately 1-% of the pre-chamber volume. This results in sampling times of 15-20°CA. Once the gas transport time from the gas sampling spark plug to the FTIR has elapsed, a data logger is activated, which saves all measurement data at 5-Hz. 4 Results The measurement system described has been set up and successfully tested. Exemplary measurements were performed and evaluated at the engine operating point described in chapter 2.1. The occurrence of pre-ignition was not observed during the course of this test series with activated gas sampling valve. The measurement of the gas composition during pre-ignition operation is therefore the subject of further investigation. In the following, the concentrations of individual species are evaluated and interpreted. Figure 3 illustrates the concentration of specific elements as a function of the extraction crank angle. Cycle-to-cycle fluctuations result in a certain fluctuation range of the concentrations, which is why the representation as a statistic plot with the 25-75 percentiles and medians is chosen here. The data obtained for each individual sampling angle is based on approximately 400-800 measurements. 30 Moritz Grüninger, Olaf Toedter, Thomas Koch, Ahmed Assabiki 30 <?page no="31"?> - 4 0 - 3 0 - 2 5 3 0 4 0 5 0 9 0 0 2 × 1 0 3 4 × 1 0 3 6 × 1 0 3 8 × 1 0 3 1 × 1 0 4 ← C H 4 * 0 . 1 ← C 2 H 6 ← C O 2 ← H 2 O → C O → C 4 H 1 0 C r a n k A n g l e / ° C A a T D C f C o n c e n t r a t i o n / p p m 0 1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 C o n c e n t r a t i o n / p p m Fig. 3: Concentrations of analyzed gas components As the representation of all measuring points in one image (Fig. 3) is comlpex, the medians of the individual measurements at the sampling angles are presented subsequently. Figure 4 shows the median concentrations of the individual species as a function of the extraction crank angle. It can be observed, that the concentration of combustion reactants (CH 4 ) and exemplary resulting intermediates (C 2 H 6 , C 4 H 10 ) are roughly constant over the observed part of the combustion cycle and up to approximately 40-50°CA aTDCf. This phenomenon may be attributed to the displacement of fresh air-fuel mixture from the main combustion chamber or the lower portion of the pre-chamber spark plug into the upper section of the pre-chamber spark plug during the compression stroke. These gases are then subjected to measurement at the extraction point. During ignition and combustion in the pre-chamber and the main combustion chamber, the flame does not yet reach the upper part of the pre-chamber spark plug, which is why the concentration remains constant. The late reduction in concentration levels suggests that a significantly delayed burn-through is occurring in this area, as also indicated by the previous simulations [2]. The CO emissions increase during the compression stroke and remain at a consistently low level following the combustion in the main combustion chamber. This phenomenon may be attributed to the displacement of unburned CO into the upper region of the pre-chamber spark plug during the compression stroke, analogous to the displacement of CH 4 and other similar compounds. Wippermann [5] has demonstrated the interaction of the main combustion chamber and pre-chamber volume due to pressure differences using a pressure-indexed pre-chamber spark plug. It is possible to oxidize CO with oxygen [6], oxygen and additional water (vapor) [7] or methane [8]. However, the concentration of CO 2 does not increase and the concentration of methane remains constant during the main combustion. For this reason, oxidation of CO is unlikely to be the only reason for the decrease in concentration between -30 and 30°CA aTDCf. For the explanation of this reduction, further measurements at more sampling angles are required. The consistent low level of CO in the later course of the examined On the Origin of Pre-Ignition inside a Pre-Chamber Spark Plug - Gas Analysis 31 31 <?page no="32"?> - 4 0 - 2 0 0 2 0 4 0 6 0 8 0 1 0 0 0 2 × 1 0 3 4 × 1 0 3 6 × 1 0 3 8 × 1 0 3 1 × 1 0 4 I g n i t i o n C H 4 * 0 . 1 C 2 H 6 C 4 H 1 0 C O C O 2 H 2 O C o n c e n t r a t i o n / p p m C r a n k A n g l e / ° C A a T D C f 0 1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 C o n c e n t r a t i o n / p p m Fig. 4: Medians of data shown in Fig. 3 part of the combustion cycle may be attributed to lean combustion in the upper part of the pre-chamber spark plug as discussed above. The observed course in CO 2 emissions lends support to the aforementioned assumption of late burning in the upper pre-chamber volume. From 30°CA-aTDCf, the CO 2 concentra‐ tion rises, as well as the concentration of H 2 O which is a product of combustion. The reason for the reduction in concentration of CO 2 between -30 and 30°CA aTDCf will be caused by several parallel reactions and can not be attributed to a singular effect and requires further investigation. The high fluctuations in the concentration of CO 2 and H 2 O at 90°CA aTDCf in Fig. 3 indicate strong cyclic fluctuations during late combustion in the upper pre-chamber volume. Species as ammonia, acetaldehyde or methanol were not detectable at any point in the measurement. 5 Summary and Outlook Engine tests with a gas extraction pre-chamber spark plug have shown, that it is possible to measure the concentration of gas species inside a pre-chamber spark plug at defined timings during the combustion cycle. The results support the CFD simulation from the previous paper in such a way that the combustion in the upper part of the pre-chamber spark plug occurs very late and slowly. This indicates limited scavenging inside the pre-chamber spark plug and unfavorable volume to wall ratios (quenching). The observed trends in CO and CO 2 over the sampling angle remain inconclusive and necessitate further investigation at additional sampling angles. Additionally, the extraction flow may require adjustment. 32 Moritz Grüninger, Olaf Toedter, Thomas Koch, Ahmed Assabiki 32 <?page no="33"?> Further investigations will also include measurements of engine operation during active pre-ignition events to see the changes of gas composition inside the pre-chamber. In order to achieve this, a further reduction in the extraction current may be necessary. The proposed hypotheses can both be partially confirmed. Firstly, it was observed that the concentration of some species increased during compression in the pre-chamber spark plug, which lends support to hypothesis 1. Secondly, species were observed for which the concentration remained constant over the observed part of the combustion cycle, thus supporting hypothesis 2 (gases and radicals remain in the pre-chamber). References [1] Rosenthal, F., Toedter, O., Kubach, H., Niessner, W., Janas, P. (2019) Optische Untersuchung der Zündeigenschaften einer Vorkammerzündkerze in einem HD-Gasmotors. In 11th Dessauer Gasmotorenkonferenz, Dessau, Germany, April 2019 [2] Grüninger, M., Janas, P., Toedter, O., Koch, T. (2022) On the Origin of Pre-Ignition inside a Pre-Chamber Spark Plug - Optical and Thermal Analysis; In 5th International Conference on Igntion Systems for Gasoline Engines, 6 th International Conference on Knoking in Gasoline Engines, Berlin, Germany, September 2022 [3] Zhu, Sipeng et al. (2022) A review of the pre-chamber ignition system applied on future low-carbon spark ignition engines. In Renewable and Sustainable Energy Reviews 154: 111872. [4] Janas, P., Niessner, W. (2018) Towards a thermally robust automotive pre-chamber spark plug for turbocharged direct injection gasoline engines. In Ignition Systems for Gasoline Engines: 4th International Conference, Berlin, Germany, December 2018. [5] Wippermann, N.; Thiele, O.; Toedter, O.; Koch, T. (2020) Measurement of the air-to-fuel ratio inside a passive pre-chamber of a fired spark-ignition engine. In Automot. Engine Technol. 5 (3-4), S.-147-157. DOI: 10.1007/ s41104-020-00067-w. [6] Lewis, B., VonElbe, G., (1951) Combustion, Flames and Explosions of. Gases, Academic Press, Inc., New York [7] Kim, T.J., Yetter, R.A., Dryer, F.L. (1994) New results on moist CO oxidation: high pressure, high temperature experiments and comprehensive kinetic modeling, In Symposium (International) on Combustion, Volume 25, Issue 1, Pages 759-766, DOI: 10.1016/ S0082-0784(06)80708-3. [8] Dryer, F.L., Glassman, I. (1973) High-temperature oxidation of CO and CH4, In Symposium (Inter‐ national) on Combustion, Volume 14, Issue 1, Pages 987-1003, DOI: 10.1016/ S0082-0784(73)80090-6. On the Origin of Pre-Ignition inside a Pre-Chamber Spark Plug - Gas Analysis 33 33 <?page no="35"?> Prechamber combustion and the challenges of knock detection Matthias Biehl, Timo Rehm, Philipp Hahn, Carsten Kluth, Rama Krishnan Saravana Kumar Abstract The implementation of prechamber ignition technology in gasoline engines offers the potential to enhance efficiency but leads to challenges for a reliable knock and pre-ignition detection. The ignition of air and fuel mixture by jets from the prechamber leads to turbulence and pressure oscillations at resonance frequencies similar to knock and preignition events in the main combustion chamber. These pressure oscillations coincide with the critical time range for knock detection, posing a significant challenge for reliable engine protection measures. The presence of pressure oscillations during regular combustions complicates the precise measurement of knock intensity, neces‐ sitating the development of optimized signal processing algorithms. Additionally, the calibration and verification of knock detection using structure-borne noise sensors are hindered by the effects of prechamber ignition. However, the Robert Bosch Group has enhanced its Model Based Knock Detection (MBKD) method to address these challenges, demonstrating the possibility of achieving reliable knock detection and control without compromising the efficiency advantages of prechamber ignition. This paper explores the impact of prechamber ignition on knock detection systems and presents potential solutions to ensure the reliable operation of engines equipped with prechamber ignition technology. 1 Introduction The implementation of prechamber ignition technology in gasoline engines has garnered significant attention for its potential to enhance engine efficiency and performance. By igniting the air and fuel mixture in the main combustion chamber through jets from the prechamber, this technology offers improved combustion characteristics such as advancing the knock limit. However, knock detection methods are challenged by the prechamber ignition as the in-cylinder pressure signal as well as the structure borne noise are disturbed by the prechamber ignited combustion. First, the in-cylinder pressure signal is analyzed and a new reference feature to quantify the knock intensity is introduced. This new feature is used to improve the knock detection with a structure borne noise sensor. <?page no="36"?> 2 Prechamber Combustion High Frequency Characteristics The prechamber ignition system in gasoline engines involves the use of a small combustion chamber located within the main combustion chamber, connected by small passages. The ignition of the air-fuel mixture in the prechamber creates a high-pressure jet of hot gases that rapidly ignites the main air-fuel mixture in the main combustion chamber. This system enables a more efficient combustion, leading to improved fuel efficiency and reduced emissions. This article includes measurement results of two turbocharged GDI multi cylinder engines with different per cylinder displacement and compression ratios. Both use a passive prechamber design. Project A uses a smaller prechamber volume of less than 2 % of the clearance volume, Project B has a larger prechamber volume of more than 2 % of the clearance volume. Due to the faster combustion of the jet ignited main charge, the knock limit shifts to more earlier ignition angles compared to combustions with conventional spark plugs. Both effects however result in an increased pressure rise which roughens the combustion sound and - alongside with the high-pressure jets - causes pressure oscillations in the main combustion chamber. The intensity of these oscillations varies for each combustion. Figure 1 shows the behavior of a combustion without and with pressure oscillations. Fig.-1: In-cylinder pressure trace (blue) and high pass filtered pressure trace (black) of combustions without (top) and with (bottom) oscillations in the same operation point (Project A) The maximum intensity of the pressure oscillations depends mainly on air charge, ignition angle and the prechamber volume. 36 Matthias Biehl, Timo Rehm, Philipp Hahn, Carsten Kluth, Rama Krishnan Saravana Kumar 36 <?page no="37"?> Applying a continuous wavelet transformation (CWT), shown in figure 2, on the pressure trace shows that the prechamber ignition stimulates the resonance frequency of the combustion chamber. Noticeable as well is the change in frequency of the oscillation due to geometricand temperature change. - Fig. 2: CWT of pressure oscillations with resonance frequency of Project A (left) and Project B (right) If the severity of the pressure oscillations increases, excitations up to the second harmonic could be observed. An additional pressure phenomenon was observed. An intense pressure spike that often gets followed by a high frequency oscillation at resonance frequency of the pressure sensor as shown in figure 3. The high frequency oscillation is not present in the bandpass filtered pressure trace. Fig. 3: Pressure trace (blue) and band pass filtered pressure trace (black) from 4 to 40 kHz of a combustion with slight knock close to the pressure sensor The root cause of these combustion could be identified as a minor knocking combustion located closely to the pressure sensor using an additional pressure sensor mounted in a different orientation and location of the original position. The pressure traces of a pressure spike of this double indication setup are shown in figure 4. Prechamber combustion and the challenges of knock detection 37 37 <?page no="38"?> Fig. 4: Pressure trace of pressure spike event with double indication setup The slightly knocking combustion starts closely to pressure sensor one resulting in a pressure spike which is not recognized by pressure sensor two. This shows that the pressure spike does not present in the entire combustion chamber and thus can be classified as harmless event. 3 Knock Detection Reference Feature The calibration of knock detectionand control functions of engine control units (ECUs) requires a quantifiable reference feature that describes the intensity of a knocking com‐ bustion. For spark ignited engines, the absolute maximum value of the bandpass-filtered cylinder pressure signal (peak pressure) has proven to be the most representative feature for the evaluation of the knock intensity. As knocking combustions excite pressure oscillations with the same frequencies, they now interfere with the pressure oscillations caused by the prechamber ignition. This interference affects all commonly used reference features that are based on high frequency pressure oscillations and thus requires a new reference feature for knocking combustion with prechamber ignition systems. The main requirement to the new knock feature is therefor to handle all permutations and intensities of knocking and pressure oscillations (see figure 5) recorded from a single in-cylinder pressure sensor. Additionally, the algorithm needs to be real time capable in a way, that it can be run by indication systems and thus be used for closed loop engine protection. Furthermore, the new knock feature needs to be calculated from a single combustion cycle to allow an adequate knock intensity determination under dynamic conditions. This prevents the usage of statistical measures such as rolling averaged oscillation peaks. 38 Matthias Biehl, Timo Rehm, Philipp Hahn, Carsten Kluth, Rama Krishnan Saravana Kumar 38 <?page no="39"?> Fig. 5: In-cylinder pressure trace (blue) and high pass filtered pressure trace (black) of combustions (top to bottom): knocking with pressure oscillation; knocking without pressure oscillation; no knocking with pressure oscillation; no knocking without pressure oscillation Prechamber combustion and the challenges of knock detection 39 39 <?page no="40"?> 3.1 Prechamber Knock Reference The identification and consideration of pressureand knock oscillations can either be done in time domain, frequency domain or a combination of both. This section describes each approach and highlights advantages and disadvantages. 3.1.1 Knock Peak with Reference Peak Subtraction This approach separates the combustion phase into two windows, a reference window and a knock window. The reference window is used to determine a reference peak pressure of the pressure oscillation before a possible knock event. It begins at firing top dead center and ends at the crank angle of the maximum of the low pass filtered pressure subtracted by a tolerance distance of up to 2 °CA. The tolerance distance is necessary to compensate the filter delay and to increase the distance to knocking oscillations. The reference peak pressure is determined by the absolute maximum of the band pass filtered pressure between 4 and 40 kHz within the reference window. The knock window is used for the determination of the knock peak pressure. Its begin is defined by the end of the reference window and has a total length of 60 °CA. The windowing is shown in figure 6. Fig. 6: Reference window (purple) and knock window (blue) for knock reference feature calculation The knock peak pressure is calculated by the absolute maximum of the filtered pressure with the same filter settings as the reference peak pressure. The knock reference feature KPEAK diff is the difference of knock peak pressure and reference peak pressure. 3.1.2 Frequency based Knock Peak Correction This approach utilizes the different frequency characteristics of pressureand knock oscillations. Within a single window defined by the start of combustion at 10% mass fraction burnt and a length of 60 °CA the absolute maximum pressure is calculated with two different band pass filters. The band pass filter (reference frequency filter) used to determine the reference peak has a lower cut-off frequency of 4 kHz and an upper cut-off frequency of 15 kHz. The knock frequency filter has a lower cut-off frequency of 15 kHz and an upper cut-off frequency of 45 kHz. In both frequency ranges, the absolute maximum peak pressure is determined. The knock peak gets subtracted by the reference peak to form the resulting knock reference feature KPEAK freq-corr . 40 Matthias Biehl, Timo Rehm, Philipp Hahn, Carsten Kluth, Rama Krishnan Saravana Kumar 40 <?page no="41"?> 3.1.3 Feature Combination This approach combines the time and frequency domain features described in the last sections. The in-cylinder pressure is filtered with both, knockand reference filter. The reference peak is determined by the absolute maximum of the pressure values within the reference window filtered with the reference filter. The knock peak is determined by the absolute maximum of the pressure values within the knock window filtered with the knock filter. Again, the knock peak gets subtracted by the reference peak to form the resulting knock reference feature KPEAK feat-comb . 3.2 Results From both examined engines, an example operation point is chosen to highlight the advantages and disadvantages of the new features. These features were simulated based on the same combustion cycles. To compare the results of the previously presented knock peak features, they are plotted versus the current knock feature: the peak pressure KPEAKKID. It is determined by the absolute maximum of the bandpass filtered pressure from 4 to 40 kHz within a fixed crank angle range from -10 to 60 °CA. Figure 7 shows the results of Project B. There are two distinct propagations in all three graphs visible. The first one correlates to the KPEAKKID values and indicate knocking combustions, the second one remains close to zero for all KPEAKKID indicating that the reference peaks for these combustions surpasses any knock peak. Fig.-7: Comparison of knock reference features (Project B) The green circle indicates a combustion with high pressure oscillations. All new features correctly classified the combustion as non knocking. The pressure trace of this combustion is shown in figure 8. Prechamber combustion and the challenges of knock detection 41 41 <?page no="42"?> Fig. 8: Detailed graph of example combustion marked with a green circle in fig. 7 with pressure oscillations The red circled combustion shows that the KPEAK freq-corr indicates a significantly lower knock intensity than the other features due to strong oscillations in the reference frequency band caused by the knocking combustion. The pressure trace of this combustion is shown in figure 9. Fig. 9: Detailed graph of example combustion marked with a red circle in fig. 7 with knocking The simulation results of project A, displayed in figure 10, also shows the two distinct propagations. However, the KPEAK diff has a significantly lower tendency to detect pressure oscillation than the other features. The KPEAK freq-corr seems to overcompensate the pressure oscillations and therefor results in lower values as the other features. 42 Matthias Biehl, Timo Rehm, Philipp Hahn, Carsten Kluth, Rama Krishnan Saravana Kumar 42 <?page no="43"?> Fig. 10: Comparison of knock reference features (Project A) The combustion circled in green describes a typical phenomenon of combustions of project A. At the angle of maximum pressure, the oscillation intensity of the resonance frequency is increasing spontaneous, not being caused by knocking. The KPEAK freq-corr and KPEAK feat-comb identify this combustion correctly as there is almost no frequency above 15 kHz present. The pressure trace of this combustion is shown in figure 11. Fig. 11: Detailed graph of example combustion marked with a green circle in fig. 10 with pressure oscillations The red circled combustion does again highlight the disadvantage of the KPEAK freq-corr as the knock causes oscillations at resonance frequency that hence overcompensate this knock feature. Prechamber combustion and the challenges of knock detection 43 43 <?page no="44"?> Fig. 12: Detailed graph of example combustion marked with a red circle in fig. 10 with knocking The KPEAK diff captivates with its simplicity in both, definition and calculation and the results correlate with knocking combustions for the projects A and B. A drawback is, that the algorithm fails as soon as the knock starts before the knock window opens. This behavior is seen especially in hydrogen ICE projects. This might be counteracted by adjusting the tolerance distance of the window definition. The KPEAK freq-corr demonstrated, that the purely frequency selective approach does not correspond to the characteristics of knocking making it indistinguishable from non knocking combustions and thus inappropriate as knock feature for knock detection. The KPEAK feat-comb shows the highest potential as knock feature in prechamber ignition systems. By varying the filter cut-off frequencies for reference and knock oscillations as well as introducing a weight factor applied to the reference peak, the feature can be adapted to increase its performance even more. But is subject to further investigation. A common drawback of all new reference feature is that neither of them is capable of detecting pressure spike events as their similarity to knocking combustions is high. 4 Structure Borne Noise of Combustion and Knock The in-cylinder pressure sensor itself could not establish for knock detection in series production, due to the comparatively high costs and the short service life. Thus, the knock detection with a structure borne noise sensor (knock sensor) is the commonly used method. 4.1 Frequency Characteristic of Combustion Noise The pressure oscillations caused by knocking are translated into structure borne noise at the cylinder walls and measured with the knock sensor. The same principle also applies for pressure oscillations caused by the prechamber ignition. The combustion noise can be seen on the knock sensor signal. A sample combustion is shown in figure 13. 44 Matthias Biehl, Timo Rehm, Philipp Hahn, Carsten Kluth, Rama Krishnan Saravana Kumar 44 <?page no="45"?> Fig. 13: Pressure trace and its CWT (left) and knock sensor signal and its CWT (right) of a non knock combustion with pressure oscillations 4.2 Frequency Characteristic of Knocking Combustions Knocking combustions in prechamber ignited engines often excite frequencies above 20 kHz as shown in figure 14. The impact of the combustion noise due to prechamber ignition can thus be reduce by omitting the frequencies below 20 kHz. - Fig. 14: Knock sensor signal and corresponding CWT of a knocking combustion of Project A (left) and Project B (right) Currently, the usable frequency range of the knock sensor is limited from 5 to 25 kHz. For Project B, this range limitation reduces the knock detection capability by missing additional knock information in the frequencies above. Therefore, the Robert Bosch GmbH started the release process of a knock sensor for large volume projects with an extended frequency range up to 30 kHz, enabling series projects to increase their knock detection capabilities. Prechamber combustion and the challenges of knock detection 45 45 <?page no="46"?> 4.3 Ignition Angle Dependency of Combustion Noise For the investigation of the influence of the combustion noise measurements were performed at a fixed operation point and varying ignition angle. The ignition variation was repeated with a hot intake air temperature to force a change in the knock limit. For each ignition angle, the knock sensor integrals of all combustions are statistically evaluated and displayed as box chart in figure 15. The integrals are calculated by the absolute sum of the band pass filtered knock sensor signal between 20 and 30 kHz. Fig. 15: Statistical evaluation of the knock sensor integrals (20 to 30 kHz) of two ignition variations with different knock limit The red line indicates the median value, the surrounding blue box the lower and upper quartiles. The dashed lines that extend above and below the blue boxes indicate the maximum and minimum values, which are no outliers. These are displayed individually as red crosses. The median value of the knock sensor integrals drops proportionally to the applied ignition retardation. Similar behavior is also observed in conventional spark ignition projects, but rather than being neglectable for conventional spark ignited engines, the reduction is considerably higher for prechamber ignited engines. Additionally, the spread of the combustion noise for non knocking combustions is large compared to its median value. 46 Matthias Biehl, Timo Rehm, Philipp Hahn, Carsten Kluth, Rama Krishnan Saravana Kumar 46 <?page no="47"?> 4.4 Effects On Knock Detection The Model Based Knock Detection (MBKD) is the most advanced knock detection method of engine control units (ECU) of the Robert Bosch Group. There are different variants of MBKD available to ensure a reliable knock detection according to the engine complexity and the required quality regarding the determination of the knock intensity. The MBKD version used for prechamber ignition projects utilizes a mean value of com‐ bustion intensities of the previous combustions to calculate a reference level, representing the base noise of the engine. A normalization of each combustion intensity to the reference level is used to compute the final relative knock intensity. A more detailed explanation is given in [1]. Figure 16 shows the base noise level dependency on ignition offset related to the base noise at knock limit for two engine speeds. Fig. 16: Ignition angle dependency of combustion noise at two different engine speeds With a drop in the base noise level, the resulting intensity of combustion noise will increase and can even cause false detections resulting in an unnecessary loss of efficiency. As the engine noise increases with engine speed, the ignition angle dependency of the base noise level is reduced. Therefore, the MBKD was improved to be able to compensate for loss in base noise level due to ignition changes enabling a robust knock detection even in lowest engine speeds. 5 Results For both projects, the correlation diagrams of the resulting MBKD calibrations are shown in figure 17. The relative knock intensity, i.e. the knock sensor based feature calculated by the ECU and the knock reference feature KPEAK feat-comb are normalized to their respective detection thresholds. Prechamber combustion and the challenges of knock detection 47 47 <?page no="48"?> Fig. 17: Knock detection correlation diagram of Project A (left) and Project B (right) Compared to the results of conventional spark ignited engines, the knock detection does not look as appealing as both structure borne noise and peak pressure are affected by the prechamber ignition. Despite the bad looks, the knock detection shows a robust behavior and ensures a safe engine operation. A further confirmation of a robust knock detection is given by the evaluation of the knock control ignition retardation. In figure 18, the resulting average of the knock control intervention for a given ignition offset is shown. For each degree of ignition advancing past the knock limit, the knock control retards accordingly maintaining the operation at the knock limit. This also applies for different knock limits. Fig. 18: Relation between ignition advance and knock control intervention for 3 different cylinders 48 Matthias Biehl, Timo Rehm, Philipp Hahn, Carsten Kluth, Rama Krishnan Saravana Kumar 48 <?page no="49"?> 6 Conclusion The prechamber ignition leads to additional challenges for knock detection as the reference pressure signal as well as the structure borne noise is affected by it. Methods have been presented to counteract these challenges for both signal paths enabling a robust knock detection. The improved MBKD is available for series development showing reliable knock detec‐ tion and control in current series calibration projects without compromising the efficiency advantages of pre-chamber ignition. Additionally, extending the usable frequency range of the knock sensor improves the knock detection quality noticeably. References [1] Biehl, M., Meister, M.: Model Based Knock Detection. In: 5, International Conference on Knocking in Gasoline Engines, pp. 257-266 (2017) Prechamber combustion and the challenges of knock detection 49 49 <?page no="51"?> 1 Tenneco / Champion® 2 Tenneco / Champion® 3 IFP energies nouvelles, Institut Carnot IFPEN transport Energie, 1 et 4 avenue de Bois Préau, 92852 Rueil-Malmaison Cedex, France 4 IFP energies nouvelles, Institut Carnot IFPEN transport Energie, 1 et 4 avenue de Bois Préau, 92852 Rueil-Malmaison Cedex, France 5 IFP energies nouvelles, Institut Carnot IFPEN transport Energie, 1 et 4 avenue de Bois Préau, 92852 Rueil-Malmaison Cedex, France Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines Chandelier 1 , M., Papi 2 , S., Serrano 3 , D., Colliou 4 , T., Duffour 5 , F. Abstract In the crucial context of climate change, it is vital to broaden our sources of decarbo‐ nated energy. In the transport sector, in addition to the electrification of vehicles and the use of advanced biofuels, hydrogen fuel is emerging as a promising solution to the challenges of reducing CO 2 emissions from internal combustion engines. This strategy preserves the advantages of existing engines, including high load capacity, well-established technology, durability, and manageable costs. However, the experimental evaluations already carried out on engines have high‐ lighted persistent challenges linked to the specific characteristics of hydrogen used as fuel. These include the management of abnormal combustion (pre-ignition and knocking), which is crucial for the durability of the engine and its ability to withstand high loads. It has also been shown that key components like the ignition system need to be adapted for use with hydrogen. The aim of this study is to understand the constraints on the ignition system when hydrogen is used as the main fuel, to establish the specifications required for this system. Tests were carried out on a single-cylinder engine specifically developed to operate with hydrogen, to understand and address the following topics: importance to use an ignition coil developed to overcome issues with combustion using hydrogen, performance of the ignition system (coil and spark plug), characterization of several spark plug designs (voltage demand, electrode temperatures…), impact of the spark plug on abnormal combustion (misfire, pre-ignition, knocking). <?page no="52"?> 1 Introduction Hydrogen is considered as a promising energy carrier for industry and mobility. Indeed, it appears to be an alternative solution for the decarbonation of ground transportation, specifically for all the applications where an electric powertrain solution is not suitable due to range limitation or high power demand. Obviously, these solutions will require a prohibitive mass of battery for energy storage, and a long refuelling time [1]. Consequently, the use of hydrogen energy carrier appears to be relevant for medium and heavy trucks, coaches, building and public works. Besides, the recent breaktrough innovations in hydrogen production such as carbon capture (leading to what is called blue H 2 ) or water electrolysis (leading to green H 2 ) makes the hydrogen more virtuous. The result is that political energy strategies now include in hydrogen as a part of the future of mobility [2]. Two technologies using hydrogen as a fuel are presently developed: the fuel cell and the Internal Combustion Engine (ICE). This latter has many interesting advantages such as its advanced maturity and low cost compared to the fuel cell solution. Indeed, the existing ICE technology and the current manufacturing plants can be reused. Moreover, the hydrogen ICE does not require high purity hydrogen and appears to be more robust in various ambient conditions (less sensitive to dust, vibrations and temperature) [3]. CO 2 emissions standards for heavy duty applications have been tightened in 2024 by the European commission [4]. A reduction of 43 % of CO 2 emissions for 2030 compared to 2019 baseline year is now required instead of the 30-% reduction presented in previous regulation. Besides, the definition of a zero-emission truck has also been revised from a 1 gCO 2 / kWh threshold to a 3 gCO 2 / ton-km for trucks. Thus, based on these regulations, H 2 ICE can be considered as a carbon-free technology that could play an important role for the next generation of heavy-duty vehicles. Hydrogen has intrinsic characteristics that are quite different from conventional fossil fuels such as methane or gasoline in terms of combustion, ignitability or behaviour relative to pre-ignition or knock. Consequently, the key components of an H 2 ICE must be adapted to properly operate this type of engine and ignition system is one of them. The aim of this study is to understand the constraints on the ignition system when hydrogen is used as the main fuel, to establish the specifications required for this system. 1.1 Impact of lean hydrogen combustion on ignition system requirements To ensure a proper understanding of the ignition system needs for adaptation, the hydrogen characteristics have to be clearly specified as mentioned in Table 1. This gas has a wide flammability range and a high laminar flame speed enabling to consider ultra-lean combustion, provided that an appropriate air loop is available. This ultra-lean combustion allows minimizing the NOx emissions [5]. Besides, hydrogen combustion flame has a very low quenching distance allowing to burn the hydrogen trapped in crevices within the combustion chamber and thus increasing the combustion efficiency. But hydrogen is not free from drawbacks. Indeed, the gaseous hydrogen has a very low density and a low energy density despite a very high lower heating value (mass energy content). These properties lead to mixing issues due to density difference with air and will require an additional compression work. Besides, the very low quenching distance of this fuel increases thermal losses as the flame comes closer to the walls. 52 Chandelier, M., Papi, S., Serrano, D., Colliou, T., Duffour, F. 52 <?page no="53"?> Fuel properties Hydrogen Methane Isooctane Flammability limits in air [λ] 10 → 0.14 2 → 0.6 1.51 → 0.26 Laminar flame speed [cm/ s] 290 48 45 Min quenching distance [mm] 0.64 2.03 3.50 Density (kg/ m 3 ) 0.09 0.65 692 Mass lower heat. value [MJ/ kg] 120 50 42 Vol. lower heat. value (MJ/ m 3 ) 10.8 (gas) 35.82 (gas) 32000 (liquid) Mass diffusivity in air [cm²/ s] 0.61 0.16 0.07 Stoichiometric A/ F ratio [kg/ kg] 34.4 17.1 14.5 Min ignition energy [m/ J] 0.02 0.28 0.28 Thermal conductivity [W/ m.K] 0.050 0.030 0.023 Table 1: Hydrogen physical and chemical properties at 1-bar, 300-K and λ = 1 [5] Hydrogen has a very low minimum ignition energy compared to other conventional carbon fuels such as methane or gasoline (Figure 1). This level of ignition energy remains very low even for highly diluted mixtures which makes H 2 / air mixtures easy to ignite in ultra-lean conditions. Figure 1: Minimum Ignition Energy for hydrogen, methane and gasoline versus λ [6] However, the very low density of hydrogen combined to ultra-lean mixture conditions leads to largely increase the intake pressure compared to the use of other liquid fuels (isooctane) or even gas fuels (methane) at stoichiometry. The left-hand chart in Figure 2 shows a comparison between hydrogen, methane and isooctane in terms of relative intake pressure requirement at λ = 1 and at same fuel energy introduced whatever the considered fuel. At λ = 1, the relative intake pressure requirement increases from 1 for gasoline, up to 1.3 for hydrogen. Indeed, this gaseous fuel has a larger volume than a liquid fuel as gasoline. The same chart shows the very low ignition energy requirement at different λ for hydrogen. When λ increases, the relative Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines 53 53 <?page no="54"?> intake pressure requirement also rises for hydrogen but the ignition energy requirement is relatively stable. The right-hand chart of Figure 2 compares the same full load operating point (IMEP = 20 bar) for two fuels (at constant fuel energy introduced): λ = 1 for gasoline and λ = 2 for hydrogen. These calculations have been performed by experiments carried out at IFPEN on different engines operated with these fuels. The peak cylinder pressure for gasoline is measured at 45 bar compared to 110 bar with hydrogen fuel. Consequently, in the case of hydrogen, higher intake pressures imply that cylinder pressure at spark timing drastically increases, requiring higher energy to ignite H 2 / air mixture. However, the resulting level of ignition energy in ultra-lean conditions for hydrogen is still largely inferior to those for gasoline or methane at stoichiometry (ten times inferior). Figure 2: Calculations of intake pressure, pressure at spark timing and ignition energy requirements for hydrogen, methane and gasoline As a matter of fact, the initiation of a combustion in spark ignited engine is based on two successive conditions. Firstly, the breakdown voltage of the ignition system must be sufficient to ionize the air/ fuel mixture and generate a consistent spark. Secondly, the energy initially contained in the coil and then released in the generated spark should be sufficient to ensure a proper flame propagation for the considered diluted conditions of air/ fuel mixture. In the case of hydrogen, the need for ignition energy is more related to spark generation (high breakdown voltage) than to air/ fuel mixture ignitability (low coil energy). The coil energy can be at a low level once the spark is properly generated on air/ H 2 diluted mixture. Figure 3: Paschen law curve and impact of gas type on breakdown voltage [8] 54 Chandelier, M., Papi, S., Serrano, D., Colliou, T., Duffour, F. 54 <?page no="55"?> The empirical Paschen law indicates that the generation of an electrical arc in a volume of gas is obtained for a breakdown voltage field function of the gas pressure “p” multiplied by the inter-electrode distance “d” of the spark plug as depicted in Figure-3. The curve is non linear in global view but it has a linear part for high values of “p.d”, range of values encountered inside the combustion chamber. The need for high breakdown voltage can be lowered by using a reduced inter-electrode distance in spark plugs. Nevertheless, in this case, other risks are amplified like misfiring if water condensation creates droplets in the inter-electrode volume in cold conditions. Additionnaly, reducing the gap electrode of spark plug can result in central electrode to wear rapidly as experienced Iwasaki et al [7], meaning that electrode materials should be adapted to operate with hydrogen. 1.2 Impact of hydrogen abnormal combustion propensity on ignition system requirements Hydrogen has a very low minimum ignition energy but it has also an on-off behaviour in terms of auto-ignition when reaching a threshold temperature level. Figure 4 displays the auto-ignition delay calculated using cinetic modelling for gasoline and hydrogen in pressure vs temperature conditions. The gradient between auto-ignition zone and no auto-ignition zone is smoother with gasoline than with hydrogen. These specificities induce for hydrogen a high propensity to abnormal combustion as pre-ignition, knock or backfiring. Figure 4: Auto-ignition delay maps for gasoline and hydrogen calculated using cinetic modelling by IFPEN. The presence of hot spots within the combustion chamber can generate run-away pre-igni‐ tion or backfiring. Run-away pre-ignition is a persistant pre-ignition in consecutive engine cycles with a rise issue in peak cylinder pressure as mixture is progressively ignited earlier and earlier. Hot spots are the main cause for this type of pre-ignition. Spark plug and more Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines 55 55 <?page no="56"?> specifically its electrodes can accumulate heat from gas compression or combustion and become a hot spot depending on their design, their internal temperatures and their location in the combustion chamber. One can assume that as the quenching distance of hydrogen is very low, the hydrogen flame can come closer to the central and ground electrodes of the spark plug and heat transfer can be drastically increased compared to gasoline or methane flames. However, the literature [9] mentioned the opposite trend meaning that electrode temperatures of the spark plug using hydrogen as fuel are around 200 °C lower than that of gasoline at same operating conditions. This is assumed to be due to the higher thermal conductivity of hydrogen, which causes the temperature of the mixture to increase more easily. Moreover, the positioning of the spark plug (more or less recession from cylinder head surface on combustion chamber side) into the combustion chamber also affects the ability of the spark plug to maintain its electrodes and its body as cold as possible [10]. At the end, the material of the spark plug electrodes is very important as it can more or less dissipate the heat transferred by the gases during compression and combustion. That is why in hydrogen combustion, the spark plug is generally chosen with a cold heat range, optimized for high heat dissipation. Another topic concerns the residual energy in the coil. A ghost spark can occur which is a non-controlled spark due to gas condition between electrodes and coil residual energy. This could lead to abnormal combustion called backfiring in the case of hydrogen indirect injection set up. Indeed, if this unexpected spark occurs during the intake phase when hydrogen is present in the vicinity of the spark plug and in the intake duct, then combustion appears in the combustion chamber, and the flame can propagate reversely into the intake ducts, leading to severe engine damage. As a remark, other ignition sources of backfire have been reported in the literature such as hot spots in the combustion chamber or high-temperature residual gases. The main contributing factors to the ghost spark are: • High residual energy in the coil after ignition [11]. The end of ignition occurs with a high arc breakdown voltage and pressure. Besides, the very low concentration of ions such as C 3 H 3+ (which are present with carbonaceous fuels) in an oxygen-hydrogen flame does not allow a gradual drop of the residual energy of the coil. • Low ionization voltage due to the presence of hydrogen which lowers dielectric properties of the mixture. The Paschen curves in Figure-3 show the impact of mixture composition between the spark plug electrodes on the breakdown voltage needed to create an electric spark. • Low intake pressure and high exhaust pressure prevent the coil from properly discharging during the exhaust phase. • Presence of high amount of water as a combustion by-product which promotes discharge of the ignition coil. Some counter measures to prevent ghost spark by unexpected coil discharge are provided in the literature such as modification of electrical circuit in the coil with diodes [11], increasing the resistance of the high voltage cable between coil and spark plug [11]. The pre-ignition can be in a runaway type in case of a hot spot (hot spark plug electode or hot exhaust valve for instance). But pre-ignition is more usually sporadic, meaning 56 Chandelier, M., Papi, S., Serrano, D., Colliou, T., Duffour, F. 56 <?page no="57"?> that it does not happen in consecutive engine cycles. In this last case, the ignition source changes from cycle to cycle but it always occurs when the piston is close to top dead center, showing that the mixture temperature increase during compression stroke is assumed to be this contributing factor to pre-ignition. Consequently, high temperature residual gases should be reduced as much as possible to mitigate pre-ignition. For instance, a study from Toyota [9] reveals that the spark design and especially the volume between the shell counterbore and insulator plays a key role in the apperance of sporadic pre-ignition in H 2 ICE. Reducing this volume decreases the high temperature of the residual gas in that region. Moreover, hot spots causing pre-ignition are also possible with pre-chamber spark plugs like in [12] depending on the design of this pre-chamber (volume, material, number of orifices, orifice diameters). 1.3 Impact of spark plug materials with hydrogen The materials used for spark electrodes can be the cause of abnormal combustion, specifically backfiring or pre-ignition. Table 2 describes the material properties commonly used for spark plug electrodes. Two effects are mentioned in the literature: the catalyst effect and the thermal effect on abnormal combustions: • Platinum is commonly used in gasoline engines to delay electrode wear, specifically for high energy sparks. Nevertheless, the spark plug electrodes that contain platinum can become a catalyst for mixture oxidation in presence of hydrogen. The publication [12] for pre-chamber spark plugs describes that the presence of platinum promotes pre-ignition. That is why this precious metal should be not used in electrode spark plug design for H 2 ICE. • Iridium enables lower surface temperatures and more homogeneous temperature dis‐ tribution on spark plug electrodes. This is mainly due to the high thermal conductivity of iridium compared to platinum [12]. The heat transferred from the gas during compression and combustion strokes is highly dissipated, limiting peak temperatures. Nickel also used in spark electrodes also has a high thermal conductivity. Material λ [W/ m.K] ρ [kg/ m 3 ] c [J/ kg.K] Al 2 O 3 30 3900 30 Ni 94 8910 515 Pt 7 21450 138 Ir 148 22400 135 Table 2: Material properties used for spark plug electrodes [7] Let us summarize some key issues pointed out in the literature for ignition systems adaptation to hydrogen operation: Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines 57 57 <?page no="58"?> • High breakdown voltage needed because of ultra-lean conditions leading to very high cylinder pressure at spark timing. • Reduced electrode gap to lower the breakdown voltage. But not too reduced to avoid misfire caused by water droplets bridging the gap in cold conditions. • Low energy coil required to ignite H 2 / air mixture as H 2 minimum ignition energy is very low even at high λ. • Electrode material with high thermal conductivity to enhance heat dissipation, meaning that very cold spark plug should be used. • Limitation of the volume between shell counterbore and insulator to avoid pre-ignition propensity. • Optimization of the positioning of the spark plug in the combustion chamber consid‐ ering a trade-off between proper mixture ignition capability and limitation of heat accumulation on spark plug electrodes. • Modification of electrical coil discharge circuit with diodes to prevent ghost spark leading to backfiring in some cases. • Avoiding certain materials in the spark plug such as platinum (promoter for hydrogen oxidation). 2 Experimental Setup 2.1 Engine desciption In this study, different ignition systems have been tested on a single cylinder engine which main characteristics are listed in Table 3. The displacement of this engine is 0.574 liter which corresponds to a 4-cylinder engine with a 2.3 liters displacement used for light duty vehicle. The maximum in-cylinder peak pressure is set to 180 bar for the engine integrity at full load. The compression ratio (CR) has been chosen at 12.0: 1. The combustion chamber architecture is a pent-roof cylinder head type with a piston with central shallow bowl. H 2 is directly injected in the combustion chamber with the injector in central location. The Hydrogen DI Injector technology is a PHINIA DI-CHG10 with an outwardly opening control valve. An injection pressure range of 20 to 40 bar was used for all the tests, yielding a maximum H 2 mass flowrate of 10 g/ s in steady state conditions. The spark plug is supplied by Tenneco / Champion® and is implemented vertically at the center of the combustion chamber, close to the injector as shown in Figure-5. The features of the different tested spark plugs are detailed in section §4. Bore x Stroke 85-mm x 101.3-mm Displacement 574 cm 3 Connecting rod length 157.25-mm Compression ratio 12,0 : 1 Valvetrain VVT on intake and exhaust camshafts Air motion Tumble 58 Chandelier, M., Papi, S., Serrano, D., Colliou, T., Duffour, F. 58 <?page no="59"?> H 2 injection PHINIA direct injector / Max mass flowrate = 10 g/ s Injection pressure = 40 bar Ignition Tenneco / Champion® cold spark plug with reduced air gap Max. peak cylinder pressure 180 bar Table 3: Engine main characteristics The air motion of the engine is based on the tumble concept. The two intake ports have been designed together with a pent-roof cylinder head adapted to spark-ignited combustion as shown in Figure-5. Hydrogen combustion is operated in homogeneous mode. It implies that high turbulence is required to enhance air/ H 2 mixing and burning velocity. A previous IFPEN study based on a heavy-duty engine basis for hydrogen [13] has compared different air motion concepts with a flat roof cylinder head and a pent-roof cylinder head. It was found that the latter leads to higher Turbulent Kinetic Energy (TKE) and better mixture homogeneity at the vicinity of the spark plug at the combustion ignition. The use of Miller intake valve improves hydrogen combustion behaviour. In a previous IFPEN study [13], it has been confirmed that Miller intake valve opening strategy was an interesting trade-off for improved air/ H 2 mixing, reduced temperature inside the cylinder at firing TDC and reduced knock tendency. Figure 5: Cylinder head with injector, spark plug, intake and exhaust ports of the tested single cylinder engine 2.2 Test cell description The single cylinder engine described in the previous paragraph has been installed on a hydrogen test cell at IFPEN in Solaize, France, to perform a test campaign as illustrated in Figure 6. Pressurized intake air was provided by an external compressor through a sonic flowmeter. For all the tests, the intake temperature was controlled with an air heater between 30 and 55 °C according to a map depending on engine speed and intake pressure. A flap is used in the exhaust line to simulate the backpressure of a real single-stage turbocharging system. An in-house real-time air loop model is implemented on the test bed supervisor software to evaluate at any time the exhaust backpressure and the corresponding Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines 59 59 <?page no="60"?> exhaust flap position. The inputs of this single-stage turbocharger system model are intake pressure and air fuel equivalence ratio (also referred as λ exhaust ). No exhaust gas recirculation (EGR) has been used for these tests and nor water injection has been implemented on this engine. Hydrogen is supplied from 20 to 40 bar to the direct injector and the hydrogen flow is measured by a Coriolis mass flowmeter located upstream the injector. Hydrogen is provided by Air Products with a purity > 99.95 %. All actuators (ignition, injection, VVT) are controlled with an in-house Electronic Control Unit (ECU). Multiple closed-loop controls are implemented on the supervisor software for controlling Indicated Mean Effective Pressure (IMEP), Crank Angle Degree for 50 % mass fuel burnt (CA50), air fuel equivalence ratio (λ exhaust ), air intake pressure and exhaust backpressure. - Figure 6: Architecture of the single-cylinder engine and implementation in IFPEN test cell Real time engine-out emissions (HC, CH 4 , CO, CO 2 , O 2 , NO and NO 2 ) have been measured by Horiba MEXA-7100DEGR analyser. Hydrogen engine-out emissions are measured by a ROSEMOUNT NGA2000 gas analyser. The reference relative air-fuel ratio λ exhaust is determined based upon the exhaust gas composition making an oxygen atomic balance. This value of λ exhaust is also validated with a measure done by a UEGO λ-sensor located in the exhaust line. Oil and coolant were supplied by electrically driven external pumps and temperatures were kept constant for both fluids at 90 °C ± 2 °C. The oil used for this test campaign is a Diesel engine oil referenced as Rubia Optima 1100 FE 10W30 from TotalEnergies. Combustion is monitored by an in-cylinder water-cooled pressure transducer recorded with an angular resolution of 0.1 CAD for 300 consecutive engine cycles. Apparent Heat Release Rate (HRR) and mass fraction burnt are obtained from in-cylinder pressure traces. Moreover, pressure transducers are present at the entrance of the intake port and at the 60 Chandelier, M., Papi, S., Serrano, D., Colliou, T., Duffour, F. 60 <?page no="61"?> exit of the cylinder head on exhaust port. The objective is to record the engine acoustics, i.e. the pressure oscillations on both intake and exhaust ports during engine tests. 3 Ignition Systems Tenneco / Champion® have developed and provided specific ignition components dedicated to H 2 ICE. Both coil and spark plug have been designed and adapted to hydrogen operation. 3.1 Coil The HY2Fire™ plug top coil supplied for this test is a high-energy coil equipped with an energy damper circuit. The objective is twofold: on one hand to prevent unexpected spark at start of charge and on the other hand, to prevent random uncontrolled sparks due to residual energy in the coil after the main spark ends and specifically when in-cylinder pressure decreases. The phenomena addressed with this technology are ghost sparks that can lead to backfiring. This is a must for a hydrogen PFI (Port Fuel Injection) engines and a plus for DI (Direct Injection) engines. In addition, the HY2Fire™ plug top coil can feature an optional diagnostic signal to get feedback about the ignition system condition. The electrical characteristics are reminded in Table 4. Input Voltage 14 V Nominal Primary Current 12.5 ± 0.3 A Primary Winding Resistance 0.3 Ω ± 10 % Dwell Time 2.4 - 2.6 ms Spark Energy ≥ 80 mJ Secondary Voltage ≥ 33 kV (Load: 25 pF / / 1 MΩ) Table 4: Electrical characteristics measured at nominal and at 23 ± 3°C The solution HY2Fire™ has already been validated through endurance tests and has entered mass production ramp-up early 2024. All the features of the HY2Fire™ plug top coil are presented in a specific paper in the same conference and entitled “Smart Ignition Coil Diagnostic System for H 2 ICE Combustion Detection”. 3.2 Spark Plug The spark plugs supplied are also adapted for hydrogen use. The main target was to keep the spark plug operating temperature as cold as possible to prevent any hot spot that could generate abnormal combustion. Temperatures over 600 °C will cause uncontrolled ignition with hydrogen. To do so, the heat range of the spark plug is selected from the racing scale and the ground electrode projection limited as much as possible. Tenneco / Champion® have designed a short line of spark plugs with several firing ends, available in both M10 and M12 long reach thread size. The main differences between the designs can be found in Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines 61 61 <?page no="62"?> the spark position (from a retracted spark position by 2.5 mm to a projection of 2 mm), the gap configuration (one or two ground electrodes with a side spark or an annular gap), or the electrode materials. The main characteristics of the designs are reminded in Figure-7. Figure 7: Spark plug design options It is essential to highlight that the spark plugs adapted for hydrogen used were designed with the considerations listed in table 5. Robustness to abnormal combustion Cold heat range (short core nose). Limited ground electrode projection. Reduced internal volume [9]. No platinum. Ignitability Spark position not too recessed. Durability Number of ground electrodes. Type of gap (side vs annular). Reduced air gap. Corrosion resistance Specific materials or platings. Dielectric strength Premium ceramic. Reduced air gap. Small center electrode. Table 5: Spark plug design considerations Knowing the carbon fouling risk is very limited with hydrogen used as the main fuel, the heat range of the spark plug can be cold or extra-cold. The potential drawback of this design feature is the risk of spark tracking on the core nose. Instead of sparking between the electrodes as expected, the spark can jump from the center electrode to the shell or inner gasket though the insulator core nose. This phenomenon called channeling cannot be neglected because there is an impact on the ignitability but also on the ceramic lifetime. 62 Chandelier, M., Papi, S., Serrano, D., Colliou, T., Duffour, F. 62 <?page no="63"?> Another sensible effect needs to be discussed. The energy required to ignite the mixture, even at high lambda, is low compared to gasoline. As a direct consequence, the starting gap between the electrodes can be set at a relatively low value. On all the designs, the nominal air gap is set around 0.4 mm. Compared to modern gasoline applications, the nominal air gap is approximately divided by two. Starting with a smaller air gap gives additional durability. However, there is disadvantage that should be considered. Water is one of the combustion products and produced in a significant quantity. When the engine is cooling down, the water contained in the combustion chamber is condensing and there is a risk of water droplets bridging the gap and causing a short-circuit to ground. To prevent this risk, the nominal air gap should not be set too small. The ignition coil, when equipped with the smart features, can help to detect the phenomenon, and provide feedback on the diagnostic signal. 3.3 Thermal modeling For firing ends A to F, a thermal analysis was carried out (firing end G was included later in the scope) using ANSYS Mechanical and custom material librairies (Figure 8). A same high heat load was applied to the different spark plugs to allow their thermal performance to be determined and compared with one another. The thermal load used included: • A sliding temperature load from the external gasket seat to the firing end. • A HTC (Heat Transfer Coefficient) on the internal faces of the spark plugs. • A HTC on the external faces of the spark plug. • A free stream gas temperature. • A radiation load. Due to the hypotheses of modeling, the absolute values are not discussed. The important is to compare the back-to-back results, and some points are crucial. A J-gap spark plug with a heat range 4 from the Champion® scale and a limited ground electrode projection will be critical. The tip of the core nose will be the hottest point of the spark plug wand will cause abnormal combustions under high loads. When comparing all the firing ends, firing ends A and C are showing the coolest temperatures while designs B, E and F have the hottest temperatures. The core nose temperature differences are mainly driven by the core nose position. The ground electrode temperature differences are due to a combination of projection, material, thermal conductivity, and ground electrode volume. It is important to note that both materials and geometries can be optimized to improve the heat transfer of the critical components. Figure 8: Thermal modeling results Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines 63 63 <?page no="64"?> 4 Spark Plug Measurement For the test campaign, several measurements were carried out on the spark plugs. We can distinguish temperature, voltage, and current measurements. For temperature, we had the possibility to measure either central electrode or ground electrode temperature. The measurement techniques are detailed below. It was also relevant to measure primary and secondary current, as well as secondary voltage. 4.1 Ground Electrode Temperature - Figure 9: Ground Electrode thermocouple - Firing end D 64 Chandelier, M., Papi, S., Serrano, D., Colliou, T., Duffour, F. 64 <?page no="65"?> Tenneco / Champion® have prepared specific samples equipped with a K-type thermo‐ couple to measure the ground electrode temperature as close as possible to the tip. After assembly, the spark plugs required a specific rework to get a thin thermo element inserted (Ø0.5 mm) inside the metal shell through a tiny hole drilled on the assembled spark plug. The tip of the thermocouple is brazed inside the ground electrode and the socket of the thermocouple is available above the barrel. Except a specific tooling, these samples are not requiring a specific measurement device and can be directly connected to the test cell measurement system (Figure-9). Ground electrode thermocouple samples were prepared for firing ends B to E. Firing end A was excluded from the build due to incompatible electrodes configuration. Firing end B was equipped with a thermo element but the result was not successful. The main problem was the small difference in size between the ground electrode itself and the thermocouple and the resulting approximative position. 4.2 Central Electrode Temperature The central electrode thermocouple temperature spark plugs require a special assembly. The specific components are prepared and then assembled together to obtain the final spark plug. The temperature is measured at the tip of the center electrode, just before the weld pool in case of precious metal usage. At this stage, there are two key remarks. First, the heat sink of the center electrode is modified due to the hole drilled to insert the thermo element. Second, the spark plug has no more internal resistance and the assembly will generate electro-magnetic interferences. A special shielding is used around the high-voltage cable to limit disturbances. To measure the central electrode temperature on a spark plug, a special equipment is needed because there is high voltage at the center electrode, and this must be separated from the thermo signal. The temperature transmitter has an internal battery and separates the high voltage signal from the thermo signal. The thermo signal goes via a fiber optic cable out directly to the laptop (Figure-10). The blue box on the right picture is not needed, it is just an option to change the signal to an analog signal. - Figure 10: Central Electrode thermocouple - Measurement setup Firing A to E were prepared with central eletrode thermocouple measurement. Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines 65 65 <?page no="66"?> 4.3 Voltage Demand The voltage demand measurement is required to evaluate the level of high voltage (HV) going through the spark plug during various operating conditions. For the voltage demand measurement, a Y-cable is used between spark plug and Ignition coil. The HV probe is connected to the Y-cable at the high voltage side. On the low voltage side, the HV probe is connected to the visualization system (Pico system - Figure-11 - or oscilloscope). Figure 11: Voltage Demand - Measurement setup 5 Test methodology and results 5.1 Spark plug temperatures During the test campaign at the test bench, the spark plug temperature has been measured for spark timing sweeps on the operating point at 2000 rpm engine speed and 16 bar IMEP for CA50 between 0 and 30 CAD ATDC. Two injection configurations were tested: • For SOI = 300 CAD BTDC (Start Of Injection) in order to test injection during intake valve opening. • For SOI = 180 CAD BTDC in order to test injection after intake valve closing. As a matter of fact, the temperature of the electrodes is the consequence of the heat transferred during compression and combustion strokes and its capability to dissipate this heat into the cylinder head. The results for ground electrode temperature measures are shown in Figure-12. The different designs can be classified from the hottest to the coldest: E / D / C. 66 Chandelier, M., Papi, S., Serrano, D., Colliou, T., Duffour, F. 66 <?page no="67"?> Figure 12: Ground electrode temperature at 2000 rpm engine speed, 16-bar IMEP with early injection (left) and late injection (right) Besides, these results show that the temperature of the ground electrode decreases with delayed combustion. A temperature reduction of around 80 °C is observed for combustion phasing delayed from the optimal value CA50 = 5 CAD ATDC to CA50 = 30 CAD ATDC. Full load tests show equivalent classification for the different designs as presented in Figure-13. Figure 13: Ground electrode temperature at 2000 and 4000 rpm engine speeds and high load. Concerning the central electrode temperature, measurements have been carried out only for design D spark plug (Figure 14). The results reveal that, whatever the operating point, the temperature of the central electrode is approximately 60°C colder than the ground electrode at 2000 rpm engine speed. At 4000 rpm, the temperature of the 2 electrodes is approximately equivalent. With the combined effect of load and combustion phasing, we can observe that the retarded combustion makes it possible to reduce significantly electrode temperature. Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines 67 67 <?page no="68"?> Figure 14: Ground and central electrode temperature comparison at 2000 and 4000 rpm engine speeds and high load. The Figure 15 presents the ground electrode temperature during an engine load increase with optimal combustion phasing (for CA50 = 5 CAD ATDC in normal combustion to delayed CA50 =11 CAD ATDC in the case of knock limitation) Figure 15: Load and combustion impact on ground electrode temperature The green dots on Figure 15 represent a spark timing sweep at constant IMEP: the CA50 varies from 0 to 20 CAD ATDC. When 20 CAD of CA50 change, the ground electrode temperature varies in a range of 100 °C. This trend can partially explain the ground electrode temperature decrease when engine load increases from 20 to 22 bar IMEP as CA50 increases from 5 to 11 Cad ATDC. 5.2 Effect on pre-ignition As described in the introduction, pre-ignition is one of the main issues for H 2 combustion. This abnormal combustion appears with the increase of engine thermal stress, in particular for delayed combustion and high loads. Pre-ignition limits the engine performance and the Figure-16 displays the simultaneous effect of both knock and pre-ignition on engine load limitation. On the one hand, when the load increases, the spark timing must be re-adapted 68 Chandelier, M., Papi, S., Serrano, D., Colliou, T., Duffour, F. 68 <?page no="69"?> to maintain optimal combustion phasing (corresponding to CA50 between 5 and 7 CAD ATDC). If engine load is further increased, knock occurrences force to drastically delay the combustion. On the other hand, if combustion is too much delayed, pre-ignition appears. It means that the possible range of spark timing narrows when engine load rises, limited on the early side by knock and on the late side by pre-ignition. The intersection point between these knock and pre-ignition fordbidden areas on Figure 16 determines the maximum achievable engine load. Figure 16: Maximum load limitation due to pre ignition and knocking The different parameters impacting pre-ignition apparition are: • Thermal load • Dilution rate • Residual burned gas (exhaust backpressure) • Spark plug design A spark timing has been carried out on the operating point 2000 rpm engine speed and 16 bar IMEP for λ exhaust = 2.25, to compare the impact of the different spark plug designs on the pre-ignition propensity. This engine load allows safe operation by limiting the risk of overpassing the maximum peak cylinder pressure for engine integrity (set to 180 bar). These tests have been done for the injection strategy during intake valve opening as shown in Figure 17. The used pre-ignition detection methodology is based on a statiscal approach for two combined indexes: the combustion initiation duration (CA10 - Spark Timing) and the peak cylinder pressure. The details of this methodology developed by IFPEN is explained in reference [14]. Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines 69 69 <?page no="70"?> Figure 17: Ratio of engine cycles in pre-ignition mode vs. combustion phasing CA50 for injection during intake valve opening. In the case of delayed combustion (for instance, CA50 = 24 CAD ATDC), the combustion initiation is slightly increased, and the pre-ignition frequency starts to increase. Very small discrepancies between the different spark plug designs are observed in terms of pre-ignition ratio. The spark plug with design D (double ground electrode) seems to be less advantageous. The pre-ignition phenomenon is precisely described in Figure 18 for the same spark variation test but for only two different spark timings with the cylinder pressure super‐ position of 300 consecutive engine cycles. On the one hand, it can be observed that for conventional combustion phasing like CA50 = 12 CAD ATDC, no pre-ignition is observed. The combustion initiation duration (CA10 - Spark timing) is around 11 CAD. On the other hand, for delayed combustion phasing such as CA50 = 24 CAD ATDC, numerous engine cycles pre-ignite. Figure 18: 300 consecutive engine cycles at optimal combustion phasing (left) and delayed combustion (right). 70 Chandelier, M., Papi, S., Serrano, D., Colliou, T., Duffour, F. 70 <?page no="71"?> Another test has been performed to determine the maximum achievable engine load for two different engine speeds 2000 and 4000 rpm with injection during intake valve opening and λ exhaust set to 2,0. The Figure 19 displays the results for the different design of spark plugs. Figure 19: Maximum achievable engine load at 2000 and 4000 rpm. Once again, very small discrepancies are observed for the different designs of spark plugs. It can be noted that this test does not allow to highlight differences in behavior between the designs of the ground electrodes. This is probably due to the chosen spark plug heat range and the fact that the electrode temperature measurements are lower than 650 °C. As the auto-ignition temperature of hydrogen is around 880 K at λ = 2, the electrode can not be a hot spot causing pre-ignition. 5.3 Spark plug with platinum Platinum has a high melting point and it is known to help preventing wear and erosion of the spark plug electrodes, extending their lifetime and maintaining optimal engine performance. However, this precious metal is also a catalyst for hydrogen oxidation. The design B (shown in Figure 20) of spark plug has been tested in a configuration with and without platinum on the ground electrode. For the comparison purpose, a spark timing sweep has been performed on the operating point for 2000 rpm engine speed and 16 bar IMEP. Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines 71 71 <?page no="72"?> Figure 20: Firing end B design The Figure 21 shows the pre-ignition rate depending on the nature of the ground electrode. Thus, we can note that the presence of platinum increases the ratio of engine cyles in pre-ignition mode by approximately 1 % compared with standard electrode made of nickel alloy. The presence of platinum increases pre-ignition frequency as soon as the combustion phasing is higher than CA50 > 10 CAD ATDC. Figure 21: Ratio of engine cycles in pre-ignition mode for design B spark plug with and without platinum. The Figure 22 presents a comparison of 300 consecutive engine cycles with a spark plug with ground electrode containing platinum and with ground electrode made of nickel alloy. In-cylinder pressure are plotted for 0 CAD BTDC spark timing. The presence of platinum on the ground electrode increases the number of cycles in pre-ignition mode. The combustion of pre-ignited engine cycles starts earlier, around 30 CAD before ignition, whereas it starts a few CAD before ignition for spark plug without platinum. Operating at a lower engine load does not show pre-ignited cycles despite the presence of platinum (IMEP < 12 bar). Finally, these results could be explained by a combination of the catalytic effect and low thermal conductivity of platinum metal. 72 Chandelier, M., Papi, S., Serrano, D., Colliou, T., Duffour, F. 72 <?page no="73"?> Figure 22: In-cylinder pressure traces at constant combustion phasing depending on electrode ground material with or without platinum. 5.4 Mixture dilution limitation With the aim of characterizing the spark plug ability to ignite a highly diluted air/ H 2 mixture, two methods to determine the flammability limits depending on the spark plug are possible. The first method is based on a dynamic test with a constant increasing λ gradient. In this case, the fuel quantity remains constant and intake air mass flowrate is increased. Initially, one might think this would lead to an increase in cylinder pressure at spark timing, but since the combustion heat release decreases, the spark timing is re-adjusted by setting it earlier. Consequently, the cylinder pressure at spark timing is approximately kept constant. As λ increases with a constant gradient, the maximum λ limit is then determined by the sudden increase of the IMEP standard deviation. For this test, the λ increment starts at λ = 1.8 (corresponding to intake pressure = 0.6 bar), which leads to 2.5 bar IMEP. These tests have been performed at 2000 rpm, with CA50 kept constant between 5 and 7 CAD ATDC as displayed in Figure-23. In this case, the maximum achievable λ is around 5,5. Figure 23: λ increasing gradient procedure until the increase in the IMEP standard deviation, to determine the maximum allowed λ Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines 73 73 <?page no="74"?> The second method is also a dynamic test with a constant reducing gradient of the discharge energy available to the spark plug. This is obtained by decreasing the coil charging duration (called dwell) for the chosen operating point as depicted in Figure 24. Similarly to the previous procedure, the operating limit is determined by the sudden increase of IMEP standard deviation when constant dwell gradient is applied. This test has been carried out at λ = 3, which corresponds to the common mixture dilution set for this type of operating point. Figure 24: Dwell reduction procedure until the increase in the IMEP standard deviation These two methods have been used to compare seven spark plug electrode configurations whose pictures are shown in Figure-25. Figure 25: Firing end tested configurations 74 Chandelier, M., Papi, S., Serrano, D., Colliou, T., Duffour, F. 74 <?page no="75"?> The Figure 26 presents the results obtained for these different designs. A repeatability test with concept C and more partially with configuration F has also been performed. The repeatability test shows a result with a low coefficient of variation in the value of maximum λ or minimum dwell between each test. Figure 26: Maxium achievable λ and minimum dwell values The results obtained in terms of minimum dwell remain relatively close and mainly depend on the capability of generating a spark between the electrodes. The fire-end G concept had a nonimal electrode gap of 0.7 mm instead of 0.4 mm for the other concepts. There is a first group of spark plugs (B, D, E, F) whose electrode protrusion is high and allow operation at λ > 5.5. The fire end A and G concepts have a recessed electrode and do not allow operation for λ > 5.5. Finally, the surface spark plug C presents a limitation for operation in very high dilution λ > 5 5.5 Voltage demand / ground electrode gap impact The objective is to evaluate the effect of the ignition energy on the diluted mixture and on the electrode gap. A λ variation has been performed between 1.7 and 2.7 at 2000 rpm engine speed and 16 bar IMEP. For each λ, two different combustion phasing were tested. The test matrix is presented on Figure-27 and shows the λ and CA50 tested ranges and the cylinder pressure at spark timing for each operating point. Figure 27: In-cylinder pressure depending on λ and combustion phasing CA50 Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines 75 75 <?page no="76"?> Thus, for these conditions, the cylinder pressure at ignition timing varies in a range between 60 and 100 bar. The analysis of the Figure 28 shows that the ionization voltage is independent of the mixture dilution but directly depends on the cylinder pressure at the ignition timing. Figure 28: Voltage demand depending on combustion timing, electrode gap, λ and cylinder pressure. It can also be noted that the combustion initiation duration determined by the CA05 - Spark timing is independent of the electrode gap. We can therefore note that the minimum ignition energy necessary to initiate hydrogen combustion is independent of the mixture dilution but proportional to the cylinder pressure at ignition timing. It has been seen previsouly that platinum can not be used for ground electrode to prevent wear. Besides, wear is related to the level of spark energy. Consequently, it seems appropriate to adapt the coil charging time close to the minimum ignition energy required to ignite the air/ H 2 mixture. This allows to prevent from electrode erosion and extend spark plug lifetime. 6 Conclusions This study has focused on the ignition system for a hydrogen internal combustion engine and aimed to understand the inherent constraints and establish the specifications for this gaseous fuel. To achieve this analysis, various ignition systems were tested on a single cylinder engine, specifically developed for hydrogen fuel operation and very close to what could be a future serial H 2 ICE application. A specific coil with energy damper and diagnostic feature was tested. Concerning the spark plug, several designs were tested with the capability to measure electrodes temperature and high voltage demand. Based on the test campaign results, different general conclusions can be drawn: • The electrode temperature is not sensitive to the injection strategy. From one design to the other, there is a good correlation between modeling and testing on the temperature classification. A delay in the combustion phasing leads to a temperature reduction. At full load and without knock limitation, the electrode temperatures are increasing while load increases. Higher engine speeds also result in more temperature on the spark plug electrodes. • Abnormal combustion is a limiting factor of H 2 ICE performance. On the one hand, for earlier combustion phasing, knock can be a limitation. On the other hand, for delayed 76 Chandelier, M., Papi, S., Serrano, D., Colliou, T., Duffour, F. 76 <?page no="77"?> combustion phasing, pre-ignition can occur in certain conditions. The maximum achievable engine load is at the intersection of the two areas. None of the fire end design was showing a significant advantages nor disadvantages regarding pre-ignition. It was possible to achieve the same maximum engine load whatever the tested fire end. • Platinum is always described as prohibited material when using hydrogen as a fuel. This was confirmed during this study with a back-to-back test using the same operating conditions. The exact same design was prepared with a nickel alloy ground electrode for one sample, and with a platinum ground electrode for another sample. The pre-ignition tendency is promoted with platinum compared to nickel alloy. • A special testing methodology was developed to characterize the spark plug designs versus the capability to ignite an ultra-diluted mixture. During this test, one design has shown a lower ignitability. • Air gap and voltage demand are also key elements for the ignition system. It has been highlighted that there is no air gap limitation to ignite the mixture. Besides, a direct correlation exists between voltage demand and cylinder pressure at start of ignition. The air gap could be reduced to the minimum without any side effect. Besides, the coil dwell should be optimized to extend the ignition system lifetime. Finally, this study focused on the ignition system confirms the main key issues of the H 2 ICE already stated in the literature. At the end, both coil and spark plug should be considered and optimized together to enhance the performance of the H 2 ICE. Considering the test conditions, it was not possible to identify a clear limiting factor of one design against the others. However, the passive pre-chamber presents promising results and should be further investigated. The next steps of the study will be to evaluate the ignition system under more severe conditions such as higher engine speeds, towards λ = 1 operation and also to identify the root cause of abnormal combustions within the current test conditions. 7 Glossary λ. Air-Fuel Equivalence Ratio ρ. Mass density ATDC. After Top Dead Center BTDC. Before Top Dead Center CA10. Crank Angle Degree for 10-% mass fuel burnt CA50. Crank Angle Degree for 50-% mass fuel burnt CAD. Crank Angle Degree CE. Center Electrode CR. Compression Ratio DI. Direct Injection EGR. Exhaust Gas Recirculation Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines 77 77 <?page no="78"?> GE. Ground Electrode HR. Heat Range HRR. Heat Release Rate HTC. Heat Transfer Coefficient HV. High Voltage ICE. Internal Combustion Engine IMEP. Indicated Mean Effective Pressure PFI. Port Fuel Injection SOI. Start Of Injection ST. Spark Timing TKE. Turbulent Kinetic Energy UEGO. Universal Exhaust Gas Oxygen sensor VVT. Variable Valve Timing 8 References [1] F O R R E S T , K., M A C K I N N O N , M., T A R R O J A , B., A N D S A M U E L S E N , S., Estimating the technical feasibility of fuel cell and battery electric vehicles for the medium and heavy duty sectors in California, Applied Energy 276: 115439, 2020, doi: 10.1016/ j.apenergy.2020.115439. [2] E U R O P E A N C O M M I S S I O N , A hydrogen strategy for a climate-neutral Europe, 2020. [3] A C A R , C. A N D D I N C E R , I., The potential role of hydrogen as a sustainable transportation fuel to combat global warming, International Journal of Hydrogen Energy 45(5): 3396-3406, 2020 [4] M U L H O L L A N D et al, The revised CO 2 standards for heavy-duty vehicles in the European Union, ICCT, 2024 [5] V E R H L E S T , S. and W A L L N E R , T., Hydrogen-fueled internal combustion engines, Progress in Energy and Combustion Science, Elsevier, Volume 35, Issue 6, Pages 490-527, 2009 [6] E U R O P E A N I N S T I T U T E F O R H Y D R O G E N S A F E T Y HYSAFE - Biennial Report on Hydrogen Safety, 2017, http: / / www.hysafe.net/ wiki/ BRHS/ BRHS [7] I W A S A K I , H., S H I R A K U R A , H. and I T O , A., A study on suppressing abnormal combustion and improving the output of hydrogen fueled internal combustion engines for comercial vehicles, SAE Technical Paper 2011-01-0674, 2011 [8] R A I Z E R , Y. P., Gas discharge physics, Springer Berlin, 1991 [9] T A K A H A S H I , D., M A T S U B A R A , N., Y A M A S H I T A , A., N A K A T A , K., Toyota’s Hydrogen-Engine Devleopment to Contribute to Carbon Neutrality, 2023-36, Vienna Motor Sympozium 2023 [10] G O L I S A N O , R., S C A L A B R I N I , S., S A C C O , N., R O S S I , R et al, System Optimization in a state-of-art V8 6.6L Hydrogen Engine, 2023-45, Vienna Motor Sympozium 2023 [11] T A K A S H I , K., S H U U I C H I , I., M A S A R U , H., A Study on the Mechanism of Backfire in External Mixture Formation Hydrogen Engines - About Backfire Occurred by Cause of the Spark Plug, SAE Technical Paper 971704, 1997 78 Chandelier, M., Papi, S., Serrano, D., Colliou, T., Duffour, F. 78 <?page no="79"?> [12] S ÖH N L E I N , S.O. et al, Effects of different prechamber spark plug geometries on combustion anomalies in an internal combustion engines, SAE Technical Paper SAE 2022-32-0023, 2022 [13] D U F F O U R , D., L A G E T , O., G I U F F R I D A , V., V A L I N , T., A N D R E , M., Optimizing a Heavy-Duty Hydrogen DI combustion system using experimental and numerical workflow, SIA International Congress: SIA Powertrain & Energy - Rouen, 2022 [14] S E R R A N O , D., G I U F F R I D A , V., V A L I N , T., D U F F O U R , F., and G A U T R O T , X., Development of a Dedicated Hydrogen Combustion System for Heavy Duty Application, Thiesel 2024 Conference on Thermo and Fluid Dynamics on Clean Propulsion Prowerplants, 2024 Development of Ignition Systems for Hydrogen-Powered Internal Combustion Engines 79 79 <?page no="81"?> Robust ignition and sparkplug wear for H2 SI-ICE J. Ängeby, A. Johnsson, J. Tidholm (SEM) K. Zhang, M. Richter, A. Ehn (Lund University) Foreword Abstract Results from nickel sparkplug electrode wear experiments are presented. Observations indicate that the wear of Ni sparkplugs from the breakthrough and arc phases is constant and independent of ignition system used. During the spark glow-phase, the wear depends linearly on the spark energy. It is demonstrated that the wear is accelerated in the presence of oxygen. The observations are important, especially for hydrogen fueled SI-ICE applications using lean combustion. It is concluded that the sparkplug lifetime can be significantly increased by using a capacitive ignition system with spark control to produce sparks of short duration and low energy. Such sparks have been shown to provide robust ignition not only for hydrogen SI-ICE applications, but also when using bioand natural gas. However, scientific research is needed to clarify how the wear phenomena depend on the spark characteristics. Ongoing research using laser-induced fluorescence technique is presented, which successfully measures Ni-electrode evaporation during a spark with a high temporal and spatial resolution. 1 Introduction Vehicles using hydrogen fueled SI-ICE is a zero-emission vehicle (ZEV) and a viable alternative to battery electric vehicles (BEV) and fuel cell electric vehicles (FCEV). The cost for H2 SI-ICE is relatively low and can be produced in large scale in established production processes using easily accessible materials. However, spark ignition of hydrogen fuel mixtures is a challenge, mainly due to the associated high spark plug electrode wear. To provide a robust ignition a high voltage is required, but only low spark energy. The sparkplug electrode wear is closely related to the spark energy which needs to be minimized to maximize the sparkplug lifetime. The key question is: What is required of the spark to provide robust ignition while keeping the electrode wear to a minimum? Numerous experiments on heavy duty H2 SI-ICE have shown that H2 fuel mixtures can be robustly ignited using a short spark (30 µs) with very little energy (3-4 mJ) at medium and high loads. A conventional inductive ignition system typically delivers sparks of energies in the range 60-110 mJ, i.e., 20-35 times higher energy than necessary. At idling and extremely diluted fuel mixtures, sparks of higher energies may be needed due to that long spark durations (300-600-µs) are needed for robust ignition. <?page no="82"?> A spark has three phases - breakthrough, arc, and glow. Breakthrough and arc cannot be controlled in closed loop. The properties of the arc phase depend mainly on parasitic capacitances and can only to some extent be affected by the hardware design. The glow phase, however, may be controlled in closed loop during the sparking. SEM, an ignition system provider, is currently developing such a system denoted FLEXISPARK ® . In the paper, results from experiments on how the electrode wear depends on the available degrees of spark control, i.e., spark duration and current, is presented together with an overview of on-going research based on Laser-induced Fluorescence (LIF). The paper is organized as follows. First, the fundamentals of a spark and electrode wear phenomena are described followed by a general description of spark ignition system design and the related degrees of spark control, for easy reference. Second, results from simple but efficient experiments on Nickel spark plug electrodes are presented that show that the wear from the breakthrough and arc phases is constant irrespective of the spark characteristics, but the wear from the glow phase (a few µ-seconds) depends linearly on the spark energy, indicating that the glow phase is the predominant wear phenomenon if an inductive or a capacitive ignition system without spark control is used. The extrapolation of such results to sparkplugs coated with precious metals such as Iridium (Ir) and/ or Platinum (Pt) is briefly discussed, which motivates a more scientific research approach. Third, an overview of ongoing scientific sparkplug wear research based on laser-induced fluorescence (LIF) is presented. Finally, conclusions are drawn from the findings and the need for future research discussed. 2 Spark, Electrode Wear and Ignition System Fundamentals To appreciate the complexity associated with spark ignition, it is important to understand the fundamental of spark discharge and resulting electrode wear. It is our objective to identify how and how much the different phases of a spark erode the electrodes. Also, it is our objective to identify spark control strategies such that robust ignition is achieved with minimal electrode wear. Therefore, it is important to understand basic ignition system design and the associated degrees of freedom that may be used to fulfil the objectives. Here, an overview is presented for easy reference. 2.1 Phases in Spark Discharge An electrical spark discharge can be divided into three phases: Breakdown, arc and glow. In terms of power and energy release, the breakdown phase has the highest power with the least energy release, the arc discharge is intermediate in both parameters, and the glow discharge has the lowest power but the highest energy release due to its long duration. 1 2.1.1 Breakdown When a high voltage is applied between electrodes of spark plug, existing free electrons and ions in the gas are accelerated by the electric field, causing collisions that ionize more gas molecules, creating an electron avalanche. A sufficiently high electric field (5 - 10 kV/ mm in air and ambient conditions) accelerates this process and the rate of electron generation surpasses the electron removal mechanisms consisting of electron 82 J. Ängeby, A. Johnsson, J. Tidholm, K. Zhang, M. Richter, A. Ehn 82 <?page no="83"?> transportation to the anode, diffusion, and attachment to positive ions. Secondary electrons, produced via secondary electron emission from ion impacts and thermionic emission from cathode heating, sustain the discharge. Breakdown occurs when the discharge becomes self-sustaining, characterized by a rapid voltage drop to about 100 V and a peak current of hundreds of amperes (A) for a few nanoseconds. The electron density can reach 10 19 -e/ cm 3 , causing significant energy transfer and gas temperatures up to 60,000 K. The rapid temperature rise leads to a hot spot on the cathode, initiating the transition to the arc phase. 2.1.2 Arc The arc phase is marked by a low potential difference (cathode fall) of about 15 V in a 1 mm gap at 1 bar pressure and a current in excess of 100 mA. Multiple small hot spots are formed on the cathode which emit thermionic electrons, matching the plasma current and sustains the arc. This phase can result in significant electrode erosion due to the high temperatures and relatively longer duration (a few µ-seconds) as compared to the breakdown phase. 2.1.3 Glow Glow discharge differs from arc discharge by having a ‘cold’ cathode. Ion bombardment, the primary mechanism for charge carrier generation, is inefficient, resulting in currents below 100 mA and high cathode fall voltages about 400 V. The ‘cold’ cathode leads to a decreased erosion rate, but the long duration (ranging from ~30 to ~2000 µs depending on ignition system) of glow discharge can result in a significant total erosion. 2.2 Models of electrode wear In general, electrode wear is caused by the interaction between the spark discharge plasma and the surface of metal alloy. To explain electrode wear four mechanisms have been developed: Sputtering, oxide removal, erosion by evaporation and particle ejection. 2.2.1 Sputtering Sputtering 2,3 involves multiple collisions between atoms. High-energy ions striking the target surface impart kinetic energy to surface atoms, causing further collisions within the target. If the incident ion energy is sufficient, atoms can be ejected from the target. In electrode erosion, most sputtered atoms leave the surface opposite to the incident ion direction, known as back-sputtering. Factors influencing sputtering include ion energy, mass, and incident angle. 2.2.2 Oxide Removal Oxide removal rather than metal itself is the primary reason for the wear of spark plug according to Goering et al. 4 The formation of oxide is necessary for a considerable wear. Shimanokami et al. 5 showed that a high melting point and good oxidation resistance can improve the lifetime of spark plugs. Rager et al. 6,7 showed that nearly no erosion occurs in pure nitrogen environment, further supporting this model. Robust ignition and sparkplug wear for H2 SI-ICE 83 83 <?page no="84"?> 2.2.3 Evaporation Jones et al. 8,9 proposed that a molten pool of cathode metal is formed during the discharge, with a metallic plasma layer above it. Ionization of evaporated metal atoms sustains the plasma channel. Thus, the most important properties for an electrode with a low wear-rate are the boiling point and thermal conductivity. A low erosion rate of platinum was observed. Higher surrounding gas pressure can reduce evaporation-induced erosion by confining the metal vapor closer to the cathode. 2.2.4 Particle Ejection Gray et al. 10-12 observed damage on electrode surfaces caused by discharges and proposed an erosion model through particle ejection. During an arc discharge, molten metal spots form on the cathode. Ion beams hitting these spots push the molten metal to crater edges, creating a recoil force. When the discharge ends, the recoil force remains, potentially overcoming the surface tension and ejecting metal droplets from the crater. 2.3 Ignition System Design and Control The two main ignition system designs are denoted capacitive and an inductive, where the names indicate how the energy for the spark is stored prior to the spark release. 13 The spark energy is stored as charge in a capacitance and as an induced magnetic field in a coil, respectively. The systems have different degrees of freedom to control the spark characteristics which has a major impact on the ignition and sparkplug wear properties. A superior spark ignition system needs to provide robust ignition while keeping the electrode wear to a minimum. A capacitive system in its simplest form is illustrated in Figure 1, and an inductive in Figure 2. Figure 1. A capacitive system - energy stored in capacitor. When the switch is closed, the voltage is transformed to the secondary coil and a spark is generated. The discharge of the capacitor may be controlled, enabling control of the spark characteristics such as current and duration (energy). 84 J. Ängeby, A. Johnsson, J. Tidholm, K. Zhang, M. Richter, A. Ehn 84 <?page no="85"?> Figure 2. An inductive system - energy stored in a magnetic field in the coil. The spark continues until the energy stored in the magnetic field has been emptied. The dwell time can be controlled, and then the maximum available spark voltage, current and duration follow. The degree of freedom when using an inductive ignition system is the time by which a current is drawn from the battery through the primary winding, denoted dwell time. The maximum available voltage to generate a spark increases with increased dwell time. However, increasing the maximum available voltage also increases the energy stored in the coil which leads to an increased spark current and duration. If the spark power/ energy is higher than necessary, then this leads to excess electrode wear and decreased sparkplug lifetime. With a capacitive design, the discharge can be controlled using µ-controlled power electronics, enabling control of the maximum available spark voltage independent of the spark current and duration. Such control can ensure robust ignition while minimizing the electrode wear. 3 Experimental evaluation of electrode wear A qualitative investigation of the sparkplug wear was performed using a motored SI-ICE and by using a bench test rig, respectively. The objectives were to evaluate the wear rate for different spark characteristics (spark control) and the impact of oxygen. 3.1 Electrode Wear for Different Spark Characteristics To experimentally investigate the wear when using different types of ignition systems and spark profiles, the authors used a motored (no combustion) spark ignited internal combustion engine (SI-ICE) and Nickel spark plug electrodes and measured the resulting wear using an ocular approach. The engine used was a 35 hp Briggs and Stratton Vanguard V-Twin 613477 with a displacement of 993 cc, bore 86 mm and compression ratio 8.2. The sparkplugs used were NGK BCPR5ES, 0.9mm gap, with Ni-electrodes. To accelerate the tests and facilitate easy measurements, the high voltage cylinder shaped electrode was machined down from the side to approx. 20 % of its initial volume before the tests, see Figure 3. Robust ignition and sparkplug wear for H2 SI-ICE 85 85 <?page no="86"?> Figure 3. The high voltage electrode before and after machining The eroded electrode volume was estimated by the difference in length of the electrode before and after the test. Three types of ignition systems were used: Capacitive with and without closed loop spark control, and inductive ignition coils, all provided by SEM. The systems are denoted ICD, FLEXISPARK and Inductive ignition coil, respectively. The basic properties of the ignition systems are summarized in Table 1. Name Type Current Control parameters Capability Inductive coil Inductive DC Maximum voltage V = f dwell ≤ 42kV I = g dwell T = ℎ(dwell) ICD Capacitive AC (7-kHz) Maximum voltage V ≤ 42kV I ≈ 200mA T ≈ 300µs FLEXISPARK Capacitive AC (~23-kHz) Closed loop spark control: - Maximum voltage - Glow current - Duration V ≤ 42kV I ∈ 40, 300 mA T ∈ 30, 2500 µs Table 1. The ignition systems used in the experiment with a motored engine. The inductive ignition coil is charged from the battery during a dwell time and a DC spark released at the end of dwell. The maximum available voltage, the current and duration are determined by said dwell time. The ICD system is a basic capacitive system without closed loop spark control for heavy-duty SI-ICE with up to 6 cylinders. The spark energy, current and duration is determined by the amount of charge stored in the capacitor. The FLEXISPARK ® system is a capacitive ignition system with a µ-controller based control module that controls an AC spark discharge in closed loop, enabling a full control of the 86 J. Ängeby, A. Johnsson, J. Tidholm, K. Zhang, M. Richter, A. Ehn 86 <?page no="87"?> available maximum spark voltage, current amplitude and duration for heavy-duty SI-ICE with up to 6 cylinders. Each experiment started with a new sparkplug and the engine motored for typically 300 h running at 1600 RPM generating ~15 M cycles. All three ignition systems were tuned to provide the same maximum available spark voltage of 40 kV. The spark current and voltage was measured using an oscilloscope and the spark energy during the glow phase computed as E = R * I 2 , where R is an estimate of the resistance between the electrodes during the glow phase (estimated to be approximately constant) and I is the spark current. The electrode size was measured before and after the test, providing a measure of the amount of eroded electrode material from which the wear per cycle was computed. Figure 4 shows the results from the experiments. The Ni electrode wear depends linearly on the spark glow-phase energy, and the wear can be modeled by wear = A + B * E where--A = 5 . 8, B = 1 . 3 and E is said spark energy during the glow-phase. Figure 4. The electrode wear vs spark energy [mJ] in the glow phase using capacitive ignition systems with and without closed loop spark control, and inductive ignition coils. Ni sparkplug. The experiment design and result deserve some consideration. As described in Section 2.2, the electrode wear can be modeled by four phenomena: Sputtering, oxide removal, evaporation and particle ejection, respectively. The estimated constant A approximates the wear during the break-through and arc-phases (wear through sputtering and oxide layer removal). The factor B is a “wear factor” capturing the rate of wear (evaporation and/ or particle ejection) as function of spark glow-phase energy, assuming a linear relationship. It is clear from Figure 4 that the model fits the measured data well indicating that the electrode Robust ignition and sparkplug wear for H2 SI-ICE 87 87 <?page no="88"?> wear is constant from the breakthrough and arc phases, independent of the wear from the glow phase. The wear depends linearly on the energy during the spark glow-phase. The observation makes sense since the spark cannot be controlled during the break-through (first few ns) and arc (3-4 µs) phases. The spark current and duration can, however, be controlled during the glow phase, i.e., from 3-4 µs until the end of spark. Increasing the spark glow-phase energy above what is necessary to sustain the plasma for robust ignition will excessively erode the Ni-electrodes, leading to a decreased sparkplug lifetime. This may become a serious issue, especially in H2 SI-ICE applications which are known to suffer from high sparkplug erosion. The ICD (capacitive without closed loop spark control) provided the highest glow-phase energy and the highest wear, 43. The inductive ignition system provided a spark with a comparatively lower glow-phase current (less power) and a wear of 28, despite a longer spark duration varying between 800 and 2000 µs depending on the cycle-to-cycle variation in breakthrough voltage and turbulence in the electrode gap. Using FLEXISPARK® (capac‐ itive with spark control), the wear was reduced to 11, when using a short duration of 40 µs. In summary, the results from the experiment indicate that the electrode wear can be re‐ duced by appr. 60 % using a high voltage, short duration, low energy spark (FLEXISPARK®) as compared to a spark from an inductive ignition coil or ICD which generates sparks with significantly higher glow-phase energies. It should be noted that a short spark with minimum energy has proven to be enough to provide robust ignition and combustion for not only H2 SI-ICE, but also for natural gas fueled SI-ICE 13 , hence indicating that sparkplug lifetime can be significantly increased for both methane and hydrogen applications. The observation holds for Ni-electrodes. Sparkplugs using rare-earth metal such as Iridium or Platinum will show a different wear. However, considering the significantly lower price for Ni sparkplugs, the observation is important, especially when considering hydrogen fueled SI-ICE where the spark plug electrode wear is known to be high and increases the maintenance cost significantly. Increasing the lifetime of Ni-sparkplugs by spark control can make such sparkplugs a cost efficient and attractive alternative to expensive sparkplugs using rare earth metal electrodes. 3.2 Electrode wear with/ without oxygen present Similar tests were also performed in bench test rigs where a sparkplug is mounted into a pressurized small chamber. The gas in the chamber flows to ensure that the electrodes are surrounded by “fresh” gas. An investigation was performed to clarify the impact of oxygen (O 2 ) motivated by that H2 SI-ICE applications typically run lean with an excess of oxygen in the combustion chamber. The same type of sparkplugs was used as in the experiment described associated with Figure 4. Figure 5 shows close-up pictures of the electrodes when run with 100-% nitrogen (N 2 ) and air (78-% N2 and 21-% O 2 ) exposed to ~53 M sparks. 88 J. Ängeby, A. Johnsson, J. Tidholm, K. Zhang, M. Richter, A. Ehn 88 <?page no="89"?> Figure 5. The electrode wear when exposed with and without O2 present. The wear increases dramatically with O2 present. In the presence of 100 % N 2 the wear from the inductive (duration ~2500µs, glow phase spark energy ~25 mJ) and FLEXISPARK (duration ~40 µs, glow phase spark energy ~4 mJ) is approximately the same. The wear changes dramatically when oxygen is present and becomes significantly higher in presence of O2 for the inductive system, whereas the FLEXISPARK ® wear increases much less. The presence of O 2 during the glow-phase clearly has a significant impact and increases the wear. Note than in the experiment leading to Figure 5 , there was oxygen present during the whole cycle since there was no combustions involved (the engine was motored). Under stoichiometric conditions the presence of O 2 during the glow-phase is expected to be negligible. However, in a lean-burn application there will be O 2 present, and that is expected to increase the spark plug wear as compared to stoichiometric SI-ICE. Of course, the hypothesis is a mere speculation. There is a need for scientific research. 3.3 On-going Scientific Research In conjunction with the traditional approach of studying spark plug wear through long-term engine tests at SEM AB, an ongoing research project at LTH (Lund University, Sweden) is utilizing laser-based optical diagnostic techniques to investigate the wear process during single or multiple sparks, which have been widely employed to examine instantaneous and complex processes with high temporal and spatial resolution, such as combustion and plasma phenomena. Robust ignition and sparkplug wear for H2 SI-ICE 89 89 <?page no="90"?> 3.3.1 Laser-induced fluorescence (LIF) Among all laser diagnostic techniques, laser-induced fluorescence (LIF) is particularly noteworthy for its high selectivity and sensitivity, allowing for the detection of trace wanted species. Furthermore, the use of pulsed excitation and detection in LIF effectively minimizes intense background interference, in our case, plasma emissions, enhancing the accuracy of measurements. Given the involvement of both evaporation and sputtering processes, the initial research focused on the detection of atoms. Nickel was selected due to its relatively low boiling point, which we hypothesized would result in a higher concentration of atoms generated by the sparks. Figure 6 is an illustration of the experimental setup of laser-induced nickel fluorescence. Shown in Figure 6 (a), the laser, as the excitation source, is a Nd: YAG pumped dye-based laser (Quantel YG 980E & TDL 90). Different kinds of spark discharges are generated by the abovementioned ignition systems developed by SEM AB. The laser pulse is guided into the spark plug gap in a shape of sheet with varied delay relative to the initiation of the spark discharge to excite the nickel atoms. The following fluorescence signals were captured by an ICCD camera (Princeton Instrument PI-MAX 2) equipped with a large aperture UV lens (B. Halle OUC 2.50, f = 100 mm, f/ # = 2.0). All equipment is synchronized by a digital pulse & delay generator (Berkeley Nucleonics Corporation BNC Model 577). Figure 6 (b) is a schematic drawing of the J-type nickel-based alloy spark plug with the blue arrow showing the direction of the laser irradiation. 90 J. Ängeby, A. Johnsson, J. Tidholm, K. Zhang, M. Richter, A. Ehn 90 <?page no="91"?> Figure 6. Illustration of LIF techinque. 3.3.2 Preliminary Results Figure 7 shows the preliminary result from the LIF measurements. Figure 7 (a) was taken about 25 nanoseconds after the breakdown without laser excitation. Between the center electrode and the grounding electrode, the plasma channel formed by the discharge can be clearly seen. Figure 7 (b) shows the LIF image ~25 µs after breakdown and the presence Robust ignition and sparkplug wear for H2 SI-ICE 91 91 <?page no="92"?> of evaporated Ni-atoms is clearly seen in the sparkplug electrode gap. Figure 7 (c) is the accumulation of nickel fluorescence in Figure 7 (b) along the spark gap to show the spatial distribution of the evaporated nickel atoms. Figure 7. Preliminary result of LIF technique. 92 J. Ängeby, A. Johnsson, J. Tidholm, K. Zhang, M. Richter, A. Ehn 92 <?page no="93"?> 4 Concluding Remarks It has been demonstrated that the wear of sparkplugs with nickel electrodes is constant from the breakthrough and arc phases, irrespective of the spark characteristics. The wear depends linearly on the spark energy during the glow-phase. It was also demonstrated that the wear is accelerated in the presence of oxygen. The observations indicate that an ignition system with a controllable spark is preferred, especially in H2 SI-ICE applications under lean condition where oxygen will be present and sparkplug lifetime is known to be a challenge. The spark energy should be enough for a robust ignition, but never more to avoid excessive wear. The observation holds for Ni-electrodes, which have a melting point of 1453 degC and boiling point of 2913 deg C in atmospheric pressure. Sparkplugs using a platinum-group metal (PGMs) such as Platinum (1774|3827 degC) or Iridium (2466|4428 degC) will show different wear due to their higher melting and boiling temperatures. Considering the significantly lower price for Ni sparkplugs it is desired to extend the corresponding lifetime using spark characteristics that minimizes the wear to make Ni-sparkplugs a cost efficient and attractive alternative to more expensive sparkplugs using rare earth metal electrodes. Scientific research is required to further understand the mechanisms of wear and a promising approach based on laser induced fluorescence was briefly described. It was demonstrated that eroded Ni-atoms can successfully be measured with a high temporal and spatial resolution. The observations presented here indicate that the main erosion of Ni-electrodes is due to evaporation and/ or particle ejection. Future research aim to clarify the dominating erosion phenomena in other alloys as well. Are there threshold values for erosion? For example, is there a minimum break-through voltage for sputtering, i.e., a minimum kinetic energy to tear out material? Are there current and/ or duration thresholds for the evaporation process (energy to reach boiling point)? It is reasonable to expect the evaporation to be a non-linear function of current (heating power) and time. Is there a minimum current/ duration to trigger particle ejections? Finally, what is the impact of AC vs DC spark on wear? Scientific research is on-going. We need to understand the basic wear phenomena to optimize the spark ignition in renewable fuel SI-ICE applications and make them viable and cost-efficient complements to electrification and fuel cells to mitigate the climate crisis. 5 References 1. R. Maly. “Spark Ignition: Its Physics and Effect on the Internal Combus-tion Engine”. In: J.C. Hilliard, G.S. Springer, editors. Fuel Economy. 1st ed. Springer US, Boston, MA, 1984. Pp. 91-148. 10.1007/ 978-1-4899-2277-9_3. 2. P. Sigmund. “Theory of Sputtering. I. Sputtering Yield of Amorphous and Polycrystalline Targets”. Phys. Rev. 1969. 184(2): 383-416. 10.1103/ PhysRev.184.383. 3. P. Sigmund, A. Oliva, G. Falcone. “Sputtering of multicomponent materi-als: Elements of a theory”. Nuclear Instruments and Methods in Physics Re-search. 1982. 194(1-3): 541-548. 10.1016/ 0029-554X(82)90578-X. 4. V.F. Zackay, W.A. Goering. “The Air Melting of Iron-Aluminum Alloys”. Trans. Met. Soc. AIME. 1958. 212(2): 203-204. Robust ignition and sparkplug wear for H2 SI-ICE 93 93 <?page no="94"?> 5. Y. Shimanokami, Y. Matsubara, T. Suzuki, W. Matsutani. Development of High Ignitability with Small Size Spark Plug. 2004. Pp. 2004-01-0987. 10.4271/ 2004-01-0987. 6. J. Rager, A. Flaig, G. Schneider, T. Kaiser, F. Soldera, F. Mücklich. “Oxi-dation Damage of Spark Plug Electrodes”. Adv Eng Mater. 2005. 7(7): 633-640. 10.1002/ adem.200500025. 7. J. Rager, J. Böhm, T. Kaiser, A. Flaig, F. Mücklich. Design and Materials for Long-Life Spark Plugs. SAE Technical Paper. SAE International, US, 2006. Pp. 2006-01-0617. 10.4271/ 2006-01-0617. 8. F.L. Jones. “Electrode Erosion by Spark Discharges”. Br. J. Appl. Phys. 1950. 1(3): 60-65. 10.1088/ 0508-3443/ 1/ 3/ 302. 9. F.L. Jones. “The Mechanism of Electrode Erosion in Electrical Discharg-es: Physical Basis of the Low Erosion Rate of the Platinum Metals”. Platinum Metals Review. 1963. 7(2): 58-65. 10.1595/ 003214063X725865. 10. E.W. Gray, J.R. Pharney. “Electrode erosion by particle ejection in low-current arcs”. J. Appl. Phys. 1974. 45(2): 667-671. 10.1063/ 1.1663300. 11. E.W. Gray, J.A. Augis, F.J. Gibson. “Plasma and electrode interactions in short gap dis‐ charges in air I. Plasma effects†”. International Journal of Elec-tronics. 1971. 30(4): 301-313. 10.1080/ 00207217108900329. 12. J.A. Augis, F.J. Gibson, E.W. Gray. “Plasma and electrode interactions in short gap dis‐ charges in air II. Electrode effects†”. International Journal of Electronics. 1971. 30(4): 315-332. 10.1080/ 00207217108900330. 13. J. Ängeby, J. Tidholm, B. Gustafsson, A. Johnsson. Ignition Systems for SI-ICE Fueled by Alternative and Renewable Fuels. ASME 2023 ICE Forward Conference. American Society of Mechanical Engineers, Pittsburgh, Pennsyl-vania, USA, 2023. P. V001T03A005. 10.1115/ ICEF2023-110153. 94 J. Ängeby, A. Johnsson, J. Tidholm, K. Zhang, M. Richter, A. Ehn 94 <?page no="95"?> 1 Tenneco / Champion® 2 University of Perugia / Department of Engineering Abstract 3 Tenneco / Champion® 4 Tenneco / Champion® Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection Papi 1 , S., Ricci 2 , F., Dal Re 3 , M., Chandelier 4 , M. Abstract The urgent imperative to combat climate change and mitigate air pollution has sparked rapid advancements in alternative fuel technologies, with a primary focus on decarbonizing the transportation sector. Meeting this objective necessitates modifying and enhancing various components within Internal Combustion Engines to overcome current challenges and achieve ambitious goals linked to alternative fuels. One crucial aspect of this adaptation is the ignition system, which must undergo adjustments to effectively handle the increasingly demanding thermal, mechanical, and electrical challenges. Furthermore, integrating sensors and diagnostic feedback for engine control has become increasingly crucial to enhance engine control strategies to optimize efficiency. This paper explores the utilization and correlation of diagnostic feedback integrated into ignition coil coupled with spark plug, both specifically designed for H2 ICE application. Tests were carried out on a single-cylinder research engine fueled with H2 at air excess coefficient equals to 2.0 by varying the engine speed and load, across four distinct operating points. Preliminary activities were carried out to refine the ignition system’s performance and to investigate its behavior in conventional and unconventional ignition strategies. Correlations between coil diagnostic signal characteristics and in-cylinder pressure were examined to identify the presence of abnormal combustion, misfire, and good combustion. These efforts enabled to correlate the diagnostic signals behavior with the development of the pressure inside the combustion chamber, make it possible to enhance the quality and consistency of diagnostic feedback related to combustion performance conditions. 1 Introduction To decrease carbon emissions in the transportation sector and address air quality issues, regulations targeting pollutant and greenhouse gas emissions are driving the advancement of cleaner and more efficient internal combustion engines (ICEs) [1]. To achieve sustainable mobility, it is essential to investigate modern combustion techniques [2-4], enhance the <?page no="96"?> hybridization of vehicles [5], and encourage the use of renewable and alternative fuels [6]. In this context, hydrogen (H 2 ) is acknowledged as the energy carrier leading towards a fossil fuel-free future of mobility. Hydrogen is a promising fuel for transportation, offering zero carbon emissions, high efficiency, and the potential to significantly reduce pollutants like NOx [32]. It is the only fuel capable of eliminating carbon, carbon monoxide, and carbon dioxide emissions, enabling high efficiencies under very lean combustion conditions and thereby reducing NOx emissions as well [7,8]. Hydrogen’s wide flammability limits and rapid flame propagation ensure stable combustion, especially in lean mixtures [8]. It can be used in internal combustion engines in dedicated fuel, bi-fuel, or dual-fuel configurations. Research promotes hydrogen as a sole fuel or additive to fossil fuels to improve brake thermal efficiency and reduce emissions [9,10]. Hydrogen engines tolerate higher compression ratios (up to 14.5: 1) due to highly dilute mixtures and high autoignition temperatures, enhancing thermodynamic efficiency and potentially achieving 52 % engine efficiency [11-13]. Shi et al. found that adding 6 % hydrogen to gasoline increased brake thermal efficiency from 10.0% to 16.7% in a retrofitted Wankel engine at an excess air ratio of 1.3 [14]. Dimitriou et al. reported a 3 % improvement in brake thermal efficiency with an 80 % hydrogen energy addition [15]. Using pure hydrogen nearly eliminates HC and CO emissions, with minimal contributions from lubricating oil combustion [16]. While hydrogen addition reduces CO emissions, it can increase NOx emissions, which can be mitigated by water injection [17]. The transformation of ICEs to operate with hydrogen as an alternative fuel is encountering developments but also a myriad of challenges. Several components necessitate adaptation and evolution to surmount existing issues and fulfill ambitious benchmarks [18-32]. Among these components, ignition systems must also undergo adaptation to tackle the increasingly demanding thermal, mechanical, and electrical challenges associated with this transition [19]. When hydrogen is utilized with advanced combustion strategies [2-4], conventional spark ignition systems often fail to effectively ignite or sustain stable combustion processes under critical operating conditions. Despite the numerous benefits of hydrogen, its use in ICEs presents significant challenges, particularly related to abnormal combustion phenomena. These include misfire in in-cylinder processes and backfire in port fuel injection (PFI) engines [20]. Backfire can also trigger engine knock [21], causing damage to cylinders and pistons [22]. To mitigate these challenges, it is crucial to prevent pre-ignition caused by ghost sparks or hot spots around the spark plug. This can be achieved by using a cooled ignition system or unconventional ignition methods, which not only prevent pre-ignition but also facilitate the ignition of highly diluted hydrogen-air mixtures [23]. This is essential to avoid or minimize misfire events. Previous activities of the same authors [24] showed the outcomes of an innovative ignition system called Hy2Fire®. Compared to traditional spark ignition systems, the Hy2Fire® system effectively mitigates backfire phenomena in port fuel injection (PFI) hydrogen engines by rapidly discharging residual energy trapped inside the coil at the end of the spark. This system’s unique discharge configuration, coupled with a higher energy release into the cylinder, significantly improves engine power output under identical operating conditions. This enhancement is critical to enhance good combustion and preventing or reducing misfires. However, misfires and other issues can still occur, making their detection essential. While monitoring internal cylinder pressure with sensors 96 Papi, S., Ricci, F., Dal Re, M., Chandelier, M. 96 <?page no="97"?> is feasible on a test bench, it is impractical for in-vehicle applications. Therefore, innovative and advanced diagnostic systems are needed to enhance engine control strategies and optimize engine efficiency. 1.1 Present contribution This study delves into the application and integration of diagnostic feedback within an ig‐ nition coil coupled with a spark plug Hy2Fire®, specifically designed for hydrogen-powered internal combustion engines. Tests were conducted on a single-cylinder research engine fueled with hydrogen, maintaining an air excess coefficient λ = 2.0, while the operating conditions. The study examined correlations between the diagnostic signals from the coil and in-cylinder pressure to detect abnormal combustion events such as misfires and assess overall combustion quality. These investigations aimed to establish a link between the behavior of diagnostic signals and the development of pressure within the combustion chamber. This correlation enhances the reliability and precision of diagnostic feedback pertaining to combustion performance conditions. Previous investigations have identified a correlation between diagnostic signals and in-cylinder pressure. However, due to the challenges in analytically determining this correlation directly, alternative instruments are necessary. Harnessing the potential of artificial intelligence (AI) [25,26] presents a promising avenue. It offers the capability to extract intricate patterns and relationships from complex datasets [27,28], which traditional analytical methods may struggle to discern. By applying deep learning algorithms to the diagnostic signals and corresponding in-cylinder pressure data, it becomes possible to uncover slight correlations and enhance the understanding of combustion dynamics in hydrogen-powered internal combustion engines. This approach not only facilitates more accurate diagnostic feedback but also paves the way for future advancements in optimizing ignition strategies and improving overall engine performance. 2 Experimental Setup 2.1 Igniter All tests described in this manuscript were conducted using an inductive spark ignition coil supplied by Tenneco, known as the Hy2Fire® ignition system. This coil features a power circuit with a high voltage transformer (14V to 40kV) and an ignition insulated gate bipolar transistor (IGBT) on the primary side. A diode on the secondary side prevents pre-ignition by avoiding positive current during transformer charging, and a suppressor limits EMI. The internal magnetic core can store up to 90 mJ of energy (Figure 1). Designed for hydrogen applications, this device can discharge residual energy in the transformer’s secondary side and measure both primary and secondary currents in the transformer windings. The ‘energy dumper’ feature protects against pre-ignition, while the ‘diagnostic’ feature generates a signal to detect abnormal behavior. Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection 97 97 <?page no="98"?> Figure 1: Hy2Fire® Design This diagnostic signal activates when the primary side current exceeds a threshold and deactivates when the secondary side current (spark current) drops to zero (Figure 2 - Hy2Fire® Design (right) Hy2Fire® signals (left) REA23065D-Wb H2 spark plug. Figure-2). The ECU uses this signal to calculate the time between the trigger rise and fall edges, helping to identify issues like open or short circuits and ensure proper coil charging. The time between the diagnostic signal’s fall and rise edges informs the ECU about the spark discharge duration, indicating possible issues like secondary short circuits, spark plug wear, liquid in the gap, and combustion quality. Additionally, it checks for backfires caused by unwanted sparks. The ignition coil was mounted on Tenneco’s prototype M12 cold spark plug REA23065D-WB (Figure 2 (right)), specifically designed for hydrogen ICE applications. This spark plug features a 0.5mm gap between the central electrode and two radial ground electrodes. This design was chosen for several key reasons: • The low temperature of the electrodes and insulator allows the engine to operate without the risk of preignition or abnormal combustion due to hot-spot ignition of hydrogen. • The smaller spark gap (0.5mm compared to the typical 0.7mm used with conventional fuels for this engine type) reduces the high ignition voltage that would otherwise be necessary. Not needed for H 2 ignition and unfavorable for electrode erosion. • The spark gap’s visibility from below the plug makes it ideal for optical investigations of the spark and early flame development. 98 Papi, S., Ricci, F., Dal Re, M., Chandelier, M. 98 <?page no="99"?> Figure 2: Hy2Fire® Design (right) Hy2Fire® signals (left) REA23065D-Wb H2 spark plug. 2.2 Single-cylinder research engine Bore x Stroke 85-mm x 88 mm Displacement 500 cm 3 Connecting rod length 139 mm Compression ratio 8.8 : 1 Valvetrain VVT on intake and exhaust camshafts H 2 injection Port Fuel Injection , Injection pressure = 4 bar Exhaust valve open 13 CAD bBDC Exhaust valve close 25 CAD aBDC Intake valve open 20 CAD bBDC Intake valve close 24 CAD aBDC Table 1: engine main characteristics. Measurements were carried out on a 500-cc single-cylinder engine (Figure 3Figure 3schematic representation of the engine setup) which is designed to operate in Port Fuel Injection (PFI) mode (Table 1). The tests were conducted at 1000 rpm and with a throttle valve opening of 10 %. Airflow rate was regulated by means of a throttle valve upstream of the intake manifold; its position was maintained fixed for all the test points, so that the airflow inside the combustion chamber, as well as the in-cylinder charge motion, did not change. The air-fuel ratio was controlled only by increasing or decreasing the hydrogen fuel injected quantity, which had a fixed pressure of 4 bar absolute. A research ECU controlled the energizing time of the injector and the ignition timing by sending a trigger signal to the igniter control unit. A piezoresistive transducer (Kistler 4075A5) on the intake measured the intake port pressure and a piezoelectric transducer (Kistler 6061 B) on the side of the chamber measured the in-cylinder pressure. A HIOKI MR6000 oscilloscope acquired in Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection 99 99 <?page no="100"?> (1) real-time the pressure signals, the ignition signal from the ECU and the diagnostic signal of the igniters and the oxygen concentration O 2 %. A Kistler KIBOX combustion analysis system (temporal resolution of 0.1 CAD), which acquired the same mentioned signals, has been also used for comparative purposes. It is worth highlighting that all signals acquired by the HIOKI are time-resolved, while those acquired by the KIBOX are CAD-resolved. This CAD resolution is achieved through the absolute crank angular position measured by an optical encoder (AVL 365C). Figure 3: schematic representation of the engine setup During operation, the λ value is adjusted in real-time based on the O 2 % concentration using the following formula [23] (Equation 1): λ = 1 + xο 2 1 − xο 2 yο 2 where x O2 and y O2 are the wet concentrations of oxygen in the exhaust gas and intake air respectively. 100 Papi, S., Ricci, F., Dal Re, M., Chandelier, M. 100 <?page no="101"?> 3 Methodology 3.1 Preliminary observation The following signals (Figure 4) have been utilized by the proposed Machine Learning (ML) algorithms to determine the combustion quality of each cycle: • In-cylinder pressure (Pcyl) • ECU trigger signals, used to determine the ignition timing (IT). • Diagnostic signals, with their lengths (Diag) estimated. The ignition timing (IT) and diagnostic signals (Diag) are initially computed based on time and subsequently converted to crank angle degrees (CAD) according to the engine’s operating speed. Figure 4: details of the signals acquired by HIOKI to determine, via ML, the combustion quality. 3.2 Analytical methodology This section provides a comprehensive description of the analytical methodology employed in the study to detect misfire events. It explores the limitations or challenges of the current analytical approach and presents reasons why alternative methods may be beneficial or necessary to address these issues. Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection 101 101 <?page no="102"?> (2) Overview Starting from considering the operating condition reported in Figure-5. Figure 5: distribution of maximum in-cylinder pressure for case 19. First step is considering as critical threshold for misfire individuation the one reported by the following Equation 2: P max, Threshold = k x P max, Motored with P max,Motored = 13 [bar] (Figure 6) and k equals to 1.4, in this preliminary activities, to identify as misfire events the combustion cases presenting maximum in-cylinder pressure equals to 40 % of the P max,motored value . It is important to highlight that this threshold is not fixed and can be adjusted based on specific needs or conditions. The chosen value has been selected with consideration of the general trend observed in maximum in-cylinder pressure values (see Figure 5). This approach allows for flexibility in refining the threshold to better align with varying engine characteristics and performance criteria, ensuring a more accurate identification of misfire events. 102 Papi, S., Ricci, F., Dal Re, M., Chandelier, M. 102 <?page no="103"?> (3) Figure 6: In-cylinder pressure trend. (left) motored case (right) firing case. The following optimized procedure (i.e. the one which allowed to obtain the best perform‐ ance) has been used to detect misfire events, based on the diagnostic length Diag of each combustion case. It is crucial to underscore that a substantial amount of preliminary work was required to determine the most effective procedure for detecting misfire events. This effort involved a comprehensive and time-consuming process that included extensive testing, analysis, and iterative refinement. The aim was to optimize the procedure for accuracy and reliability in identifying misfires. This involved evaluating various methods, adjusting parameters, and rigorously assessing performance to ensure that the final procedure met the highest standards of precision. The time invested and the depth of analysis were necessary to develop a robust and dependable approach for misfire detection. Here the optimized procedural steps: 1. With the engine firing, the lengths of the diagnostics for the first 10 cycles (i=10) are determined. 2. The mean (mean_firing) and standard deviation (std_firing) of these lengths are calculated. 3. If the standard deviation falls below a threshold value, the test continues; otherwise, the operating point is deemed unstable (Equation 3). Tℎresℎold = mean_firing x (k x std_firing) 4. Proceed to cycle i + 1 (i + 1 = 11) and update the moving parameters, namely the updated mean (mean_firing) and standard deviation (std_firing), until condition i + x1 (with x1 > 11) is met (Equation 4). Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection 103 103 <?page no="104"?> (4) Diag (i) < Tℎresℎold 5. If the above condition is not met, cycle i + x1 is identified as a misfire event, and the moving average and standard deviation calculated up to event i + x1 - 1 will be frozen until a combustion event with a diagnostic length below the threshold value occurs. 6. When this happens, the diagnostic of cycle i + x1 + x2 will be considered to update the moving average and standard deviation of stable combustion cycles, and so on. Figure 7 presents the results of the analysis. The black circles indicate events where the diagnostic signal analysis, based on the reported methodology, identified the event as a misfire. The method demonstrates an accuracy of approximately 47-% in detecting misfire events. Figure 7: result of the analytical method. Reasons to Adopt Alternative Methods The analytical procedure tends to underestimate misfire events and often detects regular combustion events as misfires. Given the large dispersion in the analyzed case, the proposed methodology still shows a good degree of accuracy. The challenge for each operating condition lies in finding the correct threshold each time, or rather, the appropriate value of k that multiplies the moving standard deviation. Other tests have shown the same limitations, making it difficult to extend this methodology across an operational plane of m x n x z (m = rpm; n = load; z = lambda). Therefore, it is necessary to resort to an advanced tool such as Machine Learning. 3.3 Machine Learning methodology Two different Artificial Intelligence (AI) structures has been developed. The first one (CLASS) is a classification algorithm used to create a decision tree for classification tasks. It builds a model to predict categorical responses based on input data, aiding in decision-making [30]. The second one (BP) is the second model (BP) is a feed-forward neural network that uses a back-propagation algorithm for optimization purposes [31]. This structure consists of multiple layers where data moves forward from input to output, 104 Papi, S., Ricci, F., Dal Re, M., Chandelier, M. 104 <?page no="105"?> and the back-propagation algorithm adjusts the weights by minimizing the error between predicted and actual outputs through iterative updates. For each case analyzed in this work, preliminary activities were performed to optimize the BP structures in terms of neurons and hidden layers to maximize performance. It is worth highlighting that the interesting finding is that the performance, in terms of RMSE [26] on predicted values of the tested structures, varies by approximately 5 %. This demonstrates that choosing the BP model is an excellent decision, as it consistently provides reliable performance. This choice of BP over other models is justified by its ability to maintain high accuracy and adaptability across different scenarios [27, 28]. Based on the mentioned AI structures, three distinct Machine Learning structures and methodologies have been developed and tested to differentiate anomalies from regular combustion processes. The first method, referred to as BP-P max , uses the ignition timing (IT) and diagnostic signals (Diag) as inputs to predict the maximum in-cylinder pressure P max . This predicted value is then classified according to Equation (2) to determine whether the event is a misfire or a regular combustion. The second method, referred to as BP-P cyl , uses the ignition timing (IT) and diagnostic signals (Diag) as inputs to predict the entire in-cylinder pressure traces. For the predicted traces is then computed the P max and then it is classified according to Equation (2) to determine whether the event is a misfire or a regular combustion. Figure 8: simplified representation of the AI structures used to predict the combustion quality. Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection 105 105 <?page no="106"?> The third method, referred to as CLASS, uses the ignition timing (IT) and diagnostic signals (Diag) as inputs to catalogue the combustion event as regular combustion (prediction = 1) or as anomalies (prediction = 0), according to the Equation (2). A specific magnitude has been assigned to each algorithm based on preliminary investigations and used to determine the probability of a misfire event. Figure 8 resumes the ML methodology for misfire prediction. 3.4 Test campaign Table 2 summarizes the main characteristic of the point tested. Test campaign was performed at fixed speed of 1000 rpm, at idle conditions (throttle valve opening = 10°) and fixed relative access ratio λ=2.0. The activation time of the diagnostic signals was fixed at all the operating tested conditions at 1500 μs. Table 2: main characteristics of the operating point tested. Figure 9: resume of the indicating results of the point tested. 106 Papi, S., Ricci, F., Dal Re, M., Chandelier, M. 106 <?page no="107"?> For the sake of completeness, the following Figure 9 summarizes the results of the indicating analysis, while Figure 10 shows the relationship between Diag and Pmax as the grade of misfire events increases. It is evident that there is a relationship between the diagnostic length and the pressure; however, this relationship is not sufficiently clear to rely solely on analytical methods for identifying misfire events. Consequently, the use of machine learning (ML) tools becomes essential for accurate misfire detection. Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection 107 107 <?page no="108"?> Figure 10: relationship between Diag and Pmax for the operating point from case 16 to 19, as misfire rate increases. 4 Results and Discussions In this section, we present the findings from the various analytical methods and Machine Learning (ML) models applied to distinguish anomalies from regular combustion processes. 108 Papi, S., Ricci, F., Dal Re, M., Chandelier, M. 108 <?page no="109"?> (5) We begin with the results obtained from the BP-P max methodology, which utilizes the IT and diagnostic Diag signals as inputs to predict the maximum in-cylinder pressure. Following this, we discuss the results from three combined ML structures designed to enhance the accuracy and reliability of misfire detection. 4.1 Results for BP-Pmax Firstly, the performance of BP-Pmax was evaluated on a case-by-case basis to determine the feasibility of the proposed machine learning methodology. Additionally, cases with an increasing number of misfires were analyzed to effectively investigate the behavior of the structure under progressively critical operating conditions. It is worth highlighting that: • Each analyzed case consists of 305 consecutive cycles. • 90-% of the dataset is allocated for training and validating the AI model. • The remaining 10 %, which the architectures have never encountered before, is reserved for testing and prediction purposes. • The AI’s output is compared to the known target values (clearly defined by the user) to assess the algorithm’s performance, by considering the Root Mean Square Error (RMSE) on the normalized data (Equation 5): RMSE = ∑(ypredicted - ytarget) 2 )/ n) where n is the number of samples or instances in the dataset test. The subsequent Figure 11 show the capability of BP-Pmax to reproduce the target maximum Pcyl. In all cases analyzed, the RMSE consistently remains below 10 %, indicating the accuracy of our predictive models in estimating target values compared to actual outcomes. Moreover, the methodology outlined in Paragraph 3.3 achieves a classification accuracy of 100-% in distinguishing between misfires and normal combustion events. Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection 109 109 <?page no="110"?> 110 Papi, S., Ricci, F., Dal Re, M., Chandelier, M. 110 <?page no="111"?> Figure 11: results for BP-Pmax. Note that for case 18, during the training phase, misfire events were presented to the AI. However, these events did not appear in the test dataset. The algorithm’s ability to accurately identify and handle such discrepancies is crucial for its performance. Based on the obtained results, the analysis can be continued by considering the same operating points previously analyzed but now incorporating four different possible targets and the three AI structures described in Paragraph 3.3. Figure 12 summarizes the RMSE for all the operating points tested and the accuracy level of the misfire detection. Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection 111 111 <?page no="112"?> Figure 12: RMSE (orange bars) for all the operating points tested and the accuracy level (blue bars) of the misfire detection. 4.2 Results for the three combined Machine Learning structures In this paragraph, four possible targets are hypothesized, similar to real-world applications. The performance of three algorithms is evaluated in determining the percentage probability of misfire presence. This assessment provides insight into the accuracy and reliability of each algorithm when applied to practical scenarios. The results demonstrate how well each machine learning structure can predict misfire probabilities, reflecting their potential effectiveness in real applications. Target 1 - known case with misfire events There is an interest in identifying real-time potential misfire events for a specific opera‐ tional case that is already known. It is assumed that misfires have already been observed for that operational point (Figure-13). The results of the analysis are presented in Figure-14. The figure includes examples of in-cylinder pressure predictions, with the predictions shown in red and the target values in black. Additionally, it displays the percentage of misfires as discussed in Paragraph 3.3. A comparison plot is also included, illustrating the prediction versus the target in terms of misfire identification, along with the corresponding global accuracy level. A critical threshold for accurate misfire identification has been established at a 35-% probability of misfire presence. The proposed method demonstrates an overall accuracy of approximately 95 % and exhibits a 93 % accuracy in detecting misfire events. This indicates that the method is highly effective at identifying misfires, with a strong capability to distinguish between normal and misfire conditions. The high accuracy rates suggest the robustness of the approach, making it a reliable tool for real-time misfire detection in various operational scenarios. 112 Papi, S., Ricci, F., Dal Re, M., Chandelier, M. 112 <?page no="113"?> Figure 13: case analyzed for target 1. Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection 113 113 <?page no="114"?> Figure 14: results of the ML analysis for target 1. Note that D1 stands for CLASS prediction. Target 2 - known case with no previous misfire event There is an interest in identifying real-time potential misfire events for a specific opera‐ tional case that is already known. It is assumed that misfires have not yet been observed for that operational point. In this case, unlike in Target 1, the motored parameters are necessary because the provided dataset does not contain misfire events. In fact, it is assumed that in the event of a misfire, the diagnostic signals exhibit the same behavior as a motored case under the same operating conditions (Figure-15). 114 Papi, S., Ricci, F., Dal Re, M., Chandelier, M. 114 <?page no="115"?> Figure 15: comparison between motored and misfire case in terms of Pcyl and Diag. The operating condition selected for this analysis is reported in Table 3. Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection 115 115 <?page no="116"?> Table 3: main characteristics of the tested case To valorize the observation reported via Figure 15, Figure 16 display the Diag trend against Pmax of case 17 (black markers) and for the corresponding motored case (red markers, Table 2). Figure 16: Diag trend against Pmax of case 17 and for the corresponding motored case In the current dataset (Table 3), only regular combustion events are considered for the analyzed case, and 150 of these events are randomly selected. Figure 17: (left) Diagnostic trend for each combustion case of target n.2 and (right) sorted from longest to shortest Diag. The green area represents the lengths used to train the AI, substituting for the length of misfire events. 116 Papi, S., Ricci, F., Dal Re, M., Chandelier, M. 116 <?page no="117"?> Following the method for selecting motored cases described in Figure 17, 150 motored cases are added to the 150 firing cases for training. Based on Figure 17, selecting 150 cases involves choosing data from the maximum Diag value recorded during the motored case down to 95 % of the maximum Diag value observed during the firing case. The test will then be conducted on the entire spectrum available (Figure 18) to simulate real-time acquisition and verify if the proposed method can identify misfire events despite never having encountered them before (results Figure-19). The proposed method demonstrates an overall accuracy of approximately 98 % and a misfire detection accuracy of about 63 % (5 out of 8 misfire events correctly detected). Iso‐ lated misfire events are detected with 100 % accuracy. Additionally, the method accurately identifies misfire zones. When multiple consecutive misfires occur, the ANN networks may struggle to recognize the continuity but can determine the beginning (156) and end (159) of the misfire zone marked by consecutive anomalous events. This behavior is attributed (as observed in Kibox data) to the intrinsic behavior of the first stroke diagnostics, which tend to show lengths comparable to regular combustion in the case of internal misfire events within a misfire zone. In other words, the ANN identifies the first and last misfire within a consecutive series and may not detect some within this range. This deficit is not attributable to the ANN networks but rather to the inherent behavior of the diagnostics on which the networks are trained. Figure 18: case analyzed for target 2. Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection 117 117 <?page no="118"?> Figure 19: results of the ML analysis for target 2. Note that D1 stands for CLASS prediction. Note that, for this case, it was decided not to plot the predicted in-cylinder pressure when the misfire probability exceeded 35-%. Target 3 - predicting case inside a known operating range (interpolation) The objective is to identify misfire events within a known operational range for unexplored operating points (green in Figure-20). It is assumed that no tests have been conducted on the specific operating point in question. This point is within the known operational range. The request is for pressure traces of known operating points and the corresponding motored traces. Additionally, the motored trace for the unexplored operating point is required. The results demonstrate the classification network’s ability to identify misfire zones within the test range with an accuracy of 22 correctly identified cases out of 29. The deficit is consistently associated with ‘internal’ misfire events within a specific misfire range. However, the algorithm is always capable of detecting isolated misfires and ‘zones’ of misfire where multiple anomalous events occur in succession. 118 Papi, S., Ricci, F., Dal Re, M., Chandelier, M. 118 <?page no="119"?> Due to the high amount of data required for the training, for this request, the focus is exclusively on the ability of CLASS to distinguish between regular events and misfires. Figure 20: case analyzed for target 3. Figure 21: results of the ML analysis for target 3. To improve visualization: Misfire target (=100% probability of misfire set as 90-% and Regular Combustion target (=0% probability of misfire) set as: = 10-%. Target 4 - predicting case outside a known operating range (extrapolation) The objective is to identify potential misfire events in real-time for a specific unknown operational scenario (Figure-22). No previous tests have been conducted at this particular operational point, which is beyond the known operational range. We need pressure traces from known operational points, both during firing and motored conditions. Additionally, we require motored condition data for the unknown operational scenario we intend to investigate. Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection 119 119 <?page no="120"?> Due to the high amount of data required for the training, for this request, the focus is exclusively on the ability of CLASS to distinguish between regular events and misfires (results in Figure-23). With an accuracy exceeding 90 %, one is able to predict misfire events for an operational case that falls outside of a known operating range. Figure 22: case analyzed for target 4. Figure 23: results of the ML analysis for target 4. To improve visualization: Misfire target (=100% probability of misfire set as 90-% and Regular Combustion target (=0% probability of misfire) set as: = 10-%. 5 Conclusions This study delves into the application and integration of diagnostic feedback within an ig‐ nition coil coupled with a spark plug Hy2Fire®, specifically designed for hydrogen-powered internal combustion engines. Based on the comprehensive analysis and results presented in this study, and related to the experimental test campaign performed on a single-cylinder 120 Papi, S., Ricci, F., Dal Re, M., Chandelier, M. 120 <?page no="121"?> engine at 1000 rpm and idle condition, several key conclusions can be drawn regarding the application of analytical methods and Machine Learning models for detecting anomalies, specifically misfire events, in hydrogen-fueled internal combustion engines (H 2 -ICE). Effectiveness of BP-Pmax Methodology: the BP-Pmax method, utilizing IT and diagnostic signals to predict maximum in-cylinder pressure, consistently demonstrated robust per‐ formance. With a Root Mean Square Error (RMSE) consistently below 10 %, it proved accurate in estimating target values. Furthermore, achieving 100-% classification accuracy in distinguishing between misfires and normal combustion events underscores its reliability under various operational conditions. Performance of Combined ML Structures: the integration of three ML structures aimed at enhancing misfire detection accuracy yielded promising results. Across different opera‐ tional targets, the models showcased high overall accuracy rates, particularly in scenarios where misfires were both known and unknown. Notably, the models achieved up to 98 % accuracy in identifying misfires, with capabilities extending to detecting isolated misfires and continuous misfire zones. Real-World Applicability and Reliability: the study’s findings highlight the practical applicability of advanced diagnostic tools in real-time misfire detection. With accuracies of approximately 95 % to 98 % in various operational scenarios, including cases beyond known operational ranges, the models prove reliable and effective, in the operating conditions tested in this work. Challenges and Future Directions: while the ML models demonstrated significant accu‐ racy, challenges remain, particularly in identifying consecutive misfire events within specific misfire zones. Future research could focus on refining the models to improve continuity detection and further enhancing their adaptability to unforeseen operational conditions. In conclusion, the study underscores the transformative potential of ML and advanced analytical methods in improving combustion analysis for hydrogen-fueled internal com‐ bustion engines. 6 Glossary λ. Air-Fuel Equivalence Ratio AI. Artificial Intelligence ATDC. After Top Dead Center BP. Back propagation BTDC. Before Top Dead Center CLASS. Classification CAD. Crank Angle Degree CO. Carbon monoxyde CR. Compression Ratio Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection 121 121 <?page no="122"?> DIAG. Diagnostic legth ECU. Engine Control Unit H 2 . Hydrogen HC. Hydrocarbons ICE. Internal Combustion Engine IGBT. insulated gate bipolar transistor IMEP. Indicated Mean Effective Pressure IT. Ignition timing ML. Machine Learning NOx. Nitrogen oxydes Pcyl. In-cylinder Pressure Pmax. Maximum in-cylinder pressure PFI. Port Fuel Injection RMSE. Root mean square error 7 References [1] D U A N , X.; X U , L.; X U , L.; J I A N G , P.; G A N , T.; L I U , H.; Y E , S.; S U N , Z. performance analysis and comparison of the spark ignition engine fuelled with industrial by-product hydrogen and gasoline. J. Clean. Prod. 2023, 424, 138899. [2] D E R N O T T E , J .; N A J T , P.M.; D U R R E T T , R.P. Downsized-Boosted Gasoline Engine with Exhaust Compound and Dilute Advanced Combustion. SAE Int. J. Adv. Curr. Pract. Mobil. 2020, 2, 2665- 2680. [3] D A D A M , S.R.; J E N T Z , R.; L E N Z E N , T.; M E I S S N E R , H. Diagnostic Evaluation of Exhaust Gas Recircu‐ lation (EGR) System on Gasoline Electric Hybrid Vehicle; SAE Technical Paper; SAE International: Warrendale, PA, USA, 2020. [4] S I N G H , A.P.; K U M A R , V.; A G A R W A L , A.K. Evaluation of comparative engine combustion, perform‐ ance and emission characteristics of low temperature combustion (PCCI and RCCI) modes. Appl. Energy 2020, 278, 115644. [5] P I E L E C H A , J .; S K O B I E J , K.; K U R T Y K A , K. Exhaust emissions and energy consumption analysis of conventional, hybrid, and electric vehicles in real driving cycles. Energies 2020, 13, 6423. [6] A N I K A , O.C.; N N A B U I F E , S.G.; B E L L O , A.; O K O R O A F O R , E.R.; K U A N G , B.; V I L L A , R. Prospects of low and zero-carbon renewable fuels in 1.5-degree net zero emission ctualization by 2050: A critical review. Carbon Capture Sci. Technol. 2022, 5, 100072. [7] C A M P O S -C A R R I E D O , F.; B A R G I A C C H I , E.; D U F O U R , J .; I R I B A R R E N , D. How can the European Ecodesign Directive guide the deployment of hydrogen-related products for mobility? Sustain. Energy Fuels 2023, 7, 1382-1394. 122 Papi, S., Ricci, F., Dal Re, M., Chandelier, M. 122 <?page no="123"?> [8] S E M E N T A , P.; A N T O L I N I , J .B. D .V.; T O R N A T O R E , C.; C A T A P A N O , F.; V A G L I E C O , B.M.; S ÁN C H E Z , J . J .L. Exploring the potentials of lean-burn hydrogen SI engine compared to methane operation. Int. J. Hydrogen Energy 2022, 47, 25044-25056 [9] L U O , Q.; S U N , B. Inducing factors and frequency of combustion knock in hydrogen internal combustion engines. Int. J. Hydrogen Energy 2016, 41, 16296-16305. [10] A Y D I N , K.; K U T A N O G L U , R. Effects of hydrogenation of fossil fuels with hydrogen and hydroxy gas on performance and emissions of internal combustion engines. Int. J. Hydrogen Energy 2018, 43, 14047-14058. [11] S R I N I V A S A N , C.B.; S U B R A M A N I A N , R. Hydrogen as a Spark Ignition Engine Fuel Technical Review. Int. J. Mech. Mechatron. Eng. IJMME-IJENS 2014, 14, 111-117. [12] C E V I Z , M. -A.; K A Y M A Z , I. Temperature and air-fuel ratio dependent specific heat ratio functions for lean burned and unburned mixture. Energy Convers. Manag. 2005, 46, 2387-2404. [13] V E R H E L S T , S.; S I E R E N S , R.; V E R S T R A E T E N , S. A Critical Review of Experimental Research on Hydrogen Fueled SI Engines. SAE Trans. 2006, 115, 264-274. [14] S H I , C.; J I , C.; W A N G , S.; Y A N G , J .; W A N G , H. Experimental and numerical study of combustion and emissions performance in a hydrogen-enriched Wankel engine at stoichiometric and lean operations. Fuel 2021, 291, 120181. [15] D I M I T R I O U , P.; K U M A R , M.; T S U J I M U R A , T.; S U Z U K I , Y. Combustion and emission characteristics of a hydrogen-diesel dual-fuel engine. Int. J. Hydrogen Energy 2018, 43, 13605-13617. [16] W U , H.; Y U , X.; D U , Y.; J I , X.; N I U , R.; S U N , Y.; G U , J . Study on cold start characteristics of dual fuel SI engine with hydrogen direct-injection. Appl. Therm. Eng. 2016, 100, 829-839. [17] S E R I N , H.; Y I L D I Z H A N , S., Hydrogen addition to tea seed oil biodiesel: Performance and emission characteristics. Int. J. Hydrogen Energy 2018, 43, 18020-18027. [18] W A N G , L. E T A L . “The effect of hydrogen injection parameters on the quality of hydrogen-air mixture formation for a PFI hydrogen internal combustion engine.” International Journal of Hydrogen Energy 42.37 (2017): 23832-23845. [19] C A T A P A N O , F. E T A L . “Analysis of energy efficiency of methane and hydrogen-methane blends in a PFI/ DI SI research engine.” Energy 117 (2016): 378-387. [20] G A O , J . E T A L . “Review of the backfire occurrences and control strategies for port hydrogen injection internal combustion engines.” Fuel 307 (2022): 121553. [21] Y E , Y., G A O ,W., L I ,Y., Z H A N G ,P., C A O ,X. Numerical study of the effect of injection timing on the knock combustion in a direct-injection hydrogen engine. IJHE 2020; 45(51): 27904-19. [22] C E S C H I N I ,L., M O R R I ,A., B A L D U C C I ,E., C A V I N A ,N., R O J O ,N., C A L O G E R O ,L. E T A L . Experimental observations of engine piston damage induced by knocking combustion. Materials Design 2017; 114: 312-25. [23] R I C C I , F E T A L . “Analysis of Hydrogen Combustion in a Spark Ignition Research Engine with a Barrier Discharge Igniter.” Energies 17.7 (2024): 1739. [24] P A P I ,S., R I C C I ,F., E T A L ., “Smart Ignition Coil Development for H2 ICE Application”, WTZ “13 th Dessau Gas engine conference”2024. [25] R I C C I , F. E T A L . “Investigation of a Hybrid LSTM+ 1DCNN Approach to Predict In-Cylinder Pressure of Internal Combustion Engines.” Information 14.9 (2023): 507. [26] R I C C I , F. E T A L . “NARX Technique to Predict Torque in Internal Combustion Engines.” Information 14.7 (2023): 417. Smart Ignition Coil Diagnostic System for H2 ICE Combustion Detection 123 123 <?page no="124"?> [27] R I C C I , F., P E T R U C C I , L., A N D M A R I A N I , F., “Using a Machine Learning Approach to Evaluate the NOx Emissions in a Spark-Ignition Optical Engine.” Information 14.4 (2023): 224. [28] R I C C I , F., P E T R U C C I , L., A N D M A R I A N I .F., “Hybrid LSTM+ 1DCNN Approach to Forecasting Torque Internal Combustion Engines.” Vehicles 5.3 (2023): 1104-1117. [29] R I C C I , F., P E T R U C C I , L., A N D M A R I A N I .F. “Comparative Analysis Of Thermal And Non-Thermal Discharge Modes On Ultra-Lean Mixtures In An Optically Accessible Engine Equipped With A Corona Ignition System.” Combustion and Flame 259 (2024): 113123. [30] D E V I L L E , B. “Decision trees.” Wiley Interdisciplinary Review Computational Statistics 5.6 (2013): 448-455. [31] S A N K H W A R , S. E T A L . “An anti‐phishing enterprise environ model using feed‐forward backpropa‐ gation and Levenberg‐Marquardt method.” Security and Privacy 4.1 (2021): E 132. [32] H U A N G ,M., L U O ,Q., S U N ,B., Experimental Investigations of the Hydrogen Injectors on the Com‐ bustion Characteristics and Performance of a Hydrogen Internal Combustion Engine , Sustainability (F E B 2024), 16(5): 1940. 124 Papi, S., Ricci, F., Dal Re, M., Chandelier, M. 124 <?page no="125"?> Consideration of the Relationship between Flame Formation and Fast Combustion in the Second Half of the Combustion Phase of Pre-chamber Jet Combustion Ryosuke Shiina, Yusuke Shintani, Hirokazu Ando, Noritaka Kimura Abstract Efforts to achieve carbon neutrality as a measure against global warming are acceler‐ ating. In the automotive industry, there is a continued effort to improve the thermal efficiency of internal combustion engines. The mitigation of knocking is a crucial technology for enhancing thermal efficiency. Fast combustion technology using pre- chamber jet combustion has been researched as one of the methods. Pre-chamber jet combustion consists of a complex series of combustion processes. Although there are many studies on pre-chamber jet combustion, there are few discussions on the relationship between the series of combustion processes and physical quantities. The purpose of this study is to elucidate the complex combustion mechanisms of passive pre-chamber jet combustion, which differs from conventional Spark Ignition (SI) combustion. In this study, the mechanisms were investigated by using combustion CFD Converge, and the relationships between physical quantities from combustion characteristics to pre-chamber specifications were contemplated. Com‐ bustion analysis was conducted on a 1.5-liter direct-injection turbocharged engine model. The analysis conditions were an engine speed of 2,000 rpm, IMEP 1,300 kPa, and stoichiometric conditions. Three different pre-chamber volume ratios were used in the analysis. As a result, the following insights were obtained: 1. The combustion phase where knocking occurred was predominantly in the second half of the combustion phase. This is thought to be because the tempera‐ ture of the unburned area is higher in this phase, resulting in a higher increasing rate of integral of induction time. 2. A long jet penetration distance to the bore end formed a complex flame surface, which increased the burning rate. This is believed to have shortened the second half of the combustion phase. 3. The jet penetration distance was dominated by the jet momentum, which is the product of the flow velocity and mass flow rate in the connecting nozzles. The flow velocity in the connecting nozzles was dominated by the pressure ratio between the main-chamber and the pre-chamber. 4. As the volume of the pre-chamber expanded, the inflow velocity into the pre- chamber escalated, resulting in an increase of the Turbulent Kinetic Energy <?page no="126"?> (TKE). This phenomenon contributed to an elevation in the rate of heat release within the pre-chamber. In addition, the outflow of working gas effects was reduced. These factors increased the pressure inside the pre-chamber, which in turn promoted jet penetration. 1 Introduction As a countermeasure against global climate change, efforts towards carbon neutrality are accelerating worldwide. According to the IPCC report, it is necessary to reduce CO 2 emissions 45 % from 2010 to 2030, and achieve net-zero emissions by 2050 [1]. In the automotive industry, improving the thermal efficiency of internal combustion engines and increasing the electrification ratio of products are effective [2]. To enhance thermal efficiency, increasing the expansion ratio is effective, but this requires the mitigation of knocking, for which there are several methods. One such method under consideration is the fast combustion technology using pre-chamber jet [3] [4] [5]. Currently, the conventional combustion method in gasoline internal combustion engines for four-wheeled vehicles is Spark Ignition combustion (SI combustion). SI combustion has simpler combustion processes than pre-chamber jet combustion: formation of the mixture, ignition, and flame propagation within the cylinder. The physical quantities governing flame propagation in SI combustion are related to turbulent intensity, which is associated with turbulent flame speed, and the air-fuel ratio, which is associated with laminar flame speed. Their impact on combustion characteristics has been investigated for a long time using experimental apparatus and CFD [6]. Therefore, it is now possible to estimate approximate combustion characteristics by analyzing the flow field and mixture distribution before ignition. Hence, combustion analysis using combustion models is not always necessary. The pre-combustion state can be analyzed by fluid models, which require fewer resources than combustion models. On the other hand, pre-chamber jet combustion follows a complex combustion process: formation of the mixture, introduction of fresh air into the pre-chamber, ignition and flame propagation within the pre-chamber, jet ejection and penetration into the main-chamber, and ignition and flame propagation within the main-chamber. This intricate combustion process complicates the estimation of combustion characteristics from the pre-combustion state. Therefore, to predict the combustion characteristics of the pre-chamber, analysis using combustion models is necessary. Factors influencing the combustion characteristics of internal combustion engines include engine specifications (such as displacement and compression ratio) and operating conditions (such as engine speed and ignition timing). Additionally, design factors specific to pre-chamber jet combustion include pre-chamber volume, diameter of connecting nozzles, and number of connecting nozzles. These multivariate factors necessitate exten‐ sive research and development through trial and error, making it challenging to obtain optimal design solutions within realistic development timelines using combustion analysis. Understanding the mechanism of pre-chamber jet combustion and the degree of influence of physical quantities is considered essential for efficiently designing multivariate systems. 126 Ryosuke Shiina, Yusuke Shintani, Hirokazu Ando, Noritaka Kimura 126 <?page no="127"?> Prior research on pre-chamber jet combustion encompasses studies that scrutinized the impact of pre-chamber design parameters on the combustion characteristics within the main-chamber experimentaly [7], as well as investigations into the correlation between the jet penetration depth and the jet cone angle [8]. However, there are few studies discussing the sequence of physical quantities from the specifications of the pre-chamber to the combustion characteristics of the main-chamber. In this research, the processes and obscure mechanisms of pre-chamber jet combustion were discussed using combustion CFD Converge. Subsequently, the relationships between the series of physical quantities were discussed. 2 Methodology Combustion analysis was conducted using a model of a mass-produced 1.5-liter direct-in‐ jection turbocharged engine with an added pre-chamber. Table 1 displays the specifications and operating conditions of the studied engine model. Figure 1 presents a schematic diagram of the engine model. The evaluation load point for this study was set at an engine speed of 2000rpm and an IMEP of 1,300kPa, with the air-fuel ratio at stoichiometric conditions. As shown in Figure 2, three different pre-chamber volumes were used. The connecting nozzles consisted of ten nozzles with a diameter of 0.8mm each. The presence of a spark plug in the pre-chamber complicates the flame propagation within it, resulting in heterogeneous jet distribution into the main-chamber and complicating the consideration of combustion characteristics. Consequently, the spark plug was not modeled, and the ignition energy was directly applied to the ignition location depicted in the figure. Combustion CFD was performed using Convergent Science’s Converge v3.0.10. Table 2 outlines the summary of the computational model. The Extended Coherent Flame Model (ECFM) [9] was used as the combustion model. This model calculates flame propagation combustion using the flame surface density equation. To account thermal dissipation to the prechamber and connecting nozzle walls, the wall temperature was ascertained utilizing the hardness method and was subsequently integrated into the computational model. The fidelity of these models had been substantiated through validation in both conventional SI combustion and pre-chamber jet combustion [10] [11]. Bore x Stroke, Displacement 73.0 x 89.5 mm,374.5 cc Fuel Supply, Place Direct Injection (10 MPa), Side Pre-chamber volume 1.7%, 3.2%, 4.7% (% of Total Chamber Volume) Nozzle Diameter x Number Φ0.8 mm x 10 Compression ratio 12.4 (Pre-chamber 3.2%), 10.6 (SI) Engine Speed, Load 2,000 rpm, IMEP 1,300 kPa Fuel, Equivalence ratio Regular Gasoline,1.0 Table 1: Single cylinder engine and Pre-chamber specifications Consideration of the Relationship between Flame Formation and Fast Combustion 127 127 <?page no="128"?> CFD software CONVERGE v3.0.10 Turbulence model ζ-f Spray model Kelvin-Helmholtz and Rayleigh-Taylor Combustion model Extended Coherent Flame Model Ignition model Imposed Stretch Spark Ignition Autoignition model Tabulated Kinetics of Ignition Table 2: Calculation method Figure 1: Schematic Diagram of Engine and Pre-chamber Figure 2: Variation of Pre-chamber Volume and Ignition Point 3 Results and Discussions Figure 3 shows a series of factor trees tracing from the combustion characteristics in the main-chamber to the specifications of the pre-chamber. It is hypothesized that the pivotal combustion characteristic for the knocking mitigation is the second half of the combustion duration (Mass Fraction Burned: MFB50-90), as the unburned region becomes high temperature towards the end of combustion, thereby shortening the ignition delay. The turbulent flame speed depends on both the turbulent flame surface area and the laminar 128 Ryosuke Shiina, Yusuke Shintani, Hirokazu Ando, Noritaka Kimura 128 <?page no="129"?> flame speed on the surface. However, the focus is solely on the turbulent flame surface area. This is because, under stoichiometric and non-diluted conditions, the differences in laminar flame speed are negligible. The turbulent flame surface area corresponds to the wrinkled flame surface area in flame propagation combustion [12]. Conversely, the average no-wrinkled flame surface, which does not account for the unevenness of wrinkles [13], is a significant factor in discussing MFB50-90. The turbulent flame can be conceptualized as a stretched average flame due to the elongating influence of turbulence. A distinctive aspect of pre-chamber jet combustion is that the jet, propelled by the pressure difference between the main-chamber and pre-chamber, penetrates the mixture in the main-chamber, creating a strong turbulent field and complex flame front. This is believed to significantly increase the turbulent flame surface area, thereby shortening the combustion duration. However, the pre-chamber is not a closed system; hence, it is necessary to consider not only the effect of increased pressure due to heat release but also the effect of decreased pressure due to the outflow of working gas through the connecting nozzles. The balance between these heat releases and the outflow of working gas is thought to vary with the specifications of the pre-chamber, such as the connecting nozzles and volume. In light of the aforementioned connections, this paper discusses cases where only the pre-chamber volume is changed. Figure 3: Combustion Phase Factor Diagram 3.1 Relationship between Knocking and Combustion Duration This section discusses that the second half of the combustion phase contributes more significantly to knocking than the first half. Figure 4 shows the relationship between the combustion center (Mass Fraction Burned 50%: MFB50) at knocking limit, the first half of the combustion duration (MFB10-50), and the second half of the combustion duration (MFB50-90). GT-Power, a one-dimensional fluid analysis tool, was used for this section. The main combustion duration (MFB10-90) was varied between 7 to 19 deg.CA. The combustion characteristic index, denoted as ‘m’ within the Wiebe function, was varied and Consideration of the Relationship between Flame Formation and Fast Combustion 129 129 <?page no="130"?> adjusted within the range of 0.6 to 9.6. The crank angle at which the Livengood-Wu integral of the unburned area exceeds 1.0 was defined as the onset of self-ignition. It is assumed that if more than 10 % of unburned fuel remains at this point, self-ignition transitions to knocking [14]. As indicated in figures 4(a) and (b), the correlation between MFB50 and MFB50-90 is evidently stronger than MFB10-50. Therefore, it is believed that shortening MFB50-90 is crucial for the mitigation of knocking through fast combustion. Figure 4: Relationships between MFB50% on Knocking and Combustion Duration Figure 5 depicts the relationship between the increasing rate of integral of normalized induction time, corresponding to the Livengood-Wu integral, and the temperature of the unburned area. This increasing rate is dominated by temperature and escalates rapidly in the region corresponding to the latter half of the combustion duration, above 880K. This rapid increase is believed to be the reason for the strong correlation observed between MFB50 at knocking limi and MFB50-90. Figure 5: Relationship between Unburned Zone Temperature and Increasing Rate of Integral of Normal‐ ized Induction Time 130 Ryosuke Shiina, Yusuke Shintani, Hirokazu Ando, Noritaka Kimura 130 <?page no="131"?> 3.2 Combustion Duration and Turbulent Flame Surface Area This section explains that the duration of the combustion process can be expressed by the integrated turbulent flame surface area with respect to the mass fraction burned (MFB) over that duration. Figure 6 shows the changes in the turbulent flame surface area against MFB for each pre-chamber volume. In the case of pre-chamber jet combustions, the peak values of the turbulent flame surface area are more than double that of Spark Ignition (SI) combustion. Figure 7(a, b) presents the combustion duration against the integrated turbulent flame surface area with respect to MFB over that duration. For MFB50-90 (Figure 7(a)), horizontal axis is corresponding to the shaded area in Figure-6. The plots for the pre-chamber include various specifications of pre-chamber designs. As discerned from Figure 7, it was found that the combustion duration can be characterized by the integral value of the turbulent flame surface area during that duration. Similarly, the first half of the combustion duration and the main combustion duration can also be represented. Therefore, to suppress knocking, it is crucial to maintain a high turbulent flame surface area during MFB50-90. Figure 6: Relationship between Turbulent Flame Surface Area and MFB Figure 7: Relationship between Turbulent Flame Surface Area and Combustion Duration Consideration of the Relationship between Flame Formation and Fast Combustion 131 131 <?page no="132"?> 3.3 Turbulent Flame Surface Area, Turbulent intensity and no-Wrinkled Flame Surface Area In this section, the relationship between the turbulent intensity and the no-wrinkled flame surface area, which are the constituents of the turbulent flame surface area, is discussed. Subsequently, the contributions of these factors to the turbulent flame surface area are contemplated, with a focus on how they vary between the initial and latter stages of the combustion process. The term ‘no-wrinkled flame surface’ is considered to be an average flame surface that does not take into account flame wrinkles caused by turbulence. For analysis, the no-wrinkled flame surface area was defined as the surface area where the fuel reaction rate at the leading edge of flame propagation reached 50 %. Figure 8 presents the schematic of the no-wrinkled flame surface and its area for three distinct pre-chamber volumes, plotted against MFB. It is observed that the no-wrinkled flame surface area in pre-chamber jet combustion significantly exceeds that of SI combustion. However, the increase in pre-chamber volume from 3.2% to 4.7% does not result in the corresponding increase in no-wrinkled flame surface area. The mechanisms and limitations of this increase in no-wrinkled flame surface area is discussed in subsequent Section 3.4. Figure 8: Relationship between No-wrinkled Flame Surface Area and MFB Figure 9 compares the relative temporal data of the turbulent flame surface area, as shown in Figure 6, with the no-wrinkled flame surface area depicted in Figure 8. In the latter combustion phase, a strong correlation exists between the turbulent flame surface area and the no-wrinkled flame surface area, with minimal differences observed between pre-chamber volumes of varying jet intensities. This suggests that, in the latter phase, the no-wrinkled flame surface area is more influential than turbulent intensity. In contrast, during the initial combustion phase, the turbulent flame surface area is greater than the no-wrinkled flame surface area, indicating a significant impact of turbulence-induced flame stretch. This inference is supported by the observation that the differences in pre-chamber volume have a minimal effect on the no-wrinkled flame surface area at the onset of combustion, as shown in Figure 8, whereas it has significant effects on the turbulent flame surface area, as depicted in Figure-6. 132 Ryosuke Shiina, Yusuke Shintani, Hirokazu Ando, Noritaka Kimura 132 <?page no="133"?> Figure 9: Relationship between Turbulent and No-wrinkled Flame Surface Area Figure 10 shows the turbulent kinetic energy (TKE) in the main-chamber relative to MFB. In pre-chamber jet combustion, larger pre-chamber volume correlates with increased turbulent intensity. Although the effect decreases towards the latter phase of combustion, the difference in turbulent intensity due to pre-chamber volume is maintained. Figure 11 shows TKE on the no-wrinkled flame surface area in relation to MFB. TKE at the onset of the main-chamber combustion is exceptionally high. However, as combustion progresses to the latter phase, TKE significantly decreases to levels comparable with SI combustion. These trends suggest that maintaining an expansive no-wrinkled flame surface area in the latter phase is instrumental in achieving fast combustion. Figure 10: Relationship between TKE and MFB in Main-chamber Figure 11: Relationship between TKE and MFB on No-wrinkled Flame Surface Consideration of the Relationship between Flame Formation and Fast Combustion 133 133 <?page no="134"?> Figure 12 delineates the distribution of turbulent intensity (TKE) and burning rate for the pre-chamber volume specification of 3.2%. It illustrates both the IN-EX direction cross-section at the bore center and the distribution on the gasket plane. It is advisable to consider the presence of flame at locations with high burning rate. Upon integrating the burning rate over the mass within the cylinder region, one obtains the flame propagation mass burning rate (same as OMEGA_CFM [g/ s] in Converge) [9] [15]. At MFB16, a strong turbulent field is generated by the jet injected into the main-chamber. It is believed that when a burning flame exists within this strong turbulent flow field, the surface area of the turbulent flame increases due to the elongation of the flame due to the turbulent flow, resulting in fast combustion. This suggests that the differences in turbulent flame surface area observed between MFB10 and 20 in Figure 6 are attributable to the variations in turbulent intensity produced by the jet flow. Post MFB52, the regions with high burning rate are predominantly located at the bore center, indicating that the flame is propagating towards the unburned area at the bore center. The high turbulence field generated by the jet flow retains the shape of the jet and does not diffuse widely. The high turbulence field has already burnt, and no flame is present. The flame burns at the bore center and bore end, where the influence of the jet flow is relatively minor, and turbulent intensity is low. This implies that in the latter combustion phase, the influence of turbulent intensity diminishes, and the no-wrinkled flame surface area becomes dominant. Initially, the flame forms along the jet. Subsequently, as the flame propagates between jets, it creates a complex flame surface shape. Eventually, the collision and annihilation of flames between jets reduce the complexity of the flame. This behavior is believed to be reflected in the relationship between MFB and no-wrinkled flame surface area in Figure-9. Figure 12: Distribution of TKE and Burning Rate 134 Ryosuke Shiina, Yusuke Shintani, Hirokazu Ando, Noritaka Kimura 134 <?page no="135"?> (1) 3.4 No-wrinkled flame surface Area and Jet Penetration Distance In this discussion, the relationship between jet characteristics, specifically penetration distance and flame surface area, is examined. It is shown that while a larger penetration distance contributes to an increase in no-wrinkled flame surface area, there is a limit to this increase. Figure 13 illustrates the burning rate distribution on the no-wrinkled flame surface and the head gasket cross-section at MFB30, 50, and 70. The third row of the diagram shows the burned ratio. In SI combustion (a), the flame propagates in a donut shape from the bore center towards the bore end. Conversely, pre-chamber jet combustion forms a complex, undulating flame surface around the ejected jet. Notably, for pre-chamber 3.2% (c) and pre-chamber 4.7% (d), the jet penetrates to the bore end, creating a more intricate flame front between the bore center and the jet. pre-chamber 1.7% (b) exhibits intermediate characteristics between SI combustion and large-volume pre-chamber jet combustion. At MFB30, the jet ejection forms the complex flame shape, yet the jet does not penetrate to the bore end, with most of the unburned area remaining at the bore end. At MFB50, the flames between jets have already collided and extinguished, transitioning to the donut-shaped flame propagation at MFB70. An increase in pre-chamber volume from 1.7% (b) to 3.2% (c) enhances the jet, extending the jet penetration distance. At MFB30 for 3.2% (c), the jet penetrates to the bore, thereby increasing the no-wrinkled flame surface area due to the more complex undulations created in the latter combustion phase. However, further strengthening the jet by increasing the pre-chamber volume to 4.7% (d) does not extend the penetration distance beyond the bore end, resulting in minimal differences in flame shape between 3.2% (c) and 4.7% (d). This is reflected in Figure 8, where there is little difference in the no-wrinkled flame surface area after MFB50 between 3.2% and 4.7%. In this section, the factors influencing jet penetration distance are examined. It is shown that the jet penetration distance can be organized by its momentum. The theoretical equation (1) [16] posits that penetration distance (pene) is determined by the main-chamber density, momentum (product of mass flow rate and velocity) at the connecting nozzles between the pre-chamber and main-chamber, jet cone angle (θ), and the time elapsed after ejection (t). pene = 4 π 1 ρ 2 0.25 m˙ × v 1 0.25 t tan θ 0.5 In order to confirm the correlation of this theoretical formula, discrepancy between the penetration distance deduced from CFD results and that calculated from theoretical formulas was observed. The state physical quantities featured in the theoretical equation were based on the CFD calculation results depicted in Figure 14. Figure 15 shows the theoretical calculation results of the penetration distance over time for a jet injected at a specific moment. The jet trajectory is considered in terms of whether it is unburned or burned, which affects the density variations within the main-chamber. Immediately after ignition in the pre-chamber, unburned gas is expelled, and an unburned jet (w) blows into the main-chamber. Once the flame in the pre-chamber reaches the connecting nozzles, the first burnt jet (x) is ejected. Both the unburned jet (w) and the first burnt jet (x) are injected into a combustion-chamber with the high main-chamber density corresponding Consideration of the Relationship between Flame Formation and Fast Combustion 135 135 <?page no="136"?> Figure 13: Distribution of Burnig Rate and Burned Ratio to the pressure and temperature (700K) at the end of compression, hence their penetration distance does not extend significantly. However, the situation differs for the second burnt jet (y) that follows the first burnt jet (x). The path of the second burnt jet (y) is heated to approximately 1300K by the first burnt jet (x), thus it travels through a main-chamber with the lower density corresponding to 1300K. The penetration distance of the second burnt jet (y) is expected to be longer, potentially overtaking the first burnt jet (x) near the piston surface or the bore end. Figure 16 shows the relationship between the penetration distance of the burnt jet calculated using the theoretical equation based on the above considerations, and the penetration distance deduced from CFD results. The CFD and theoretical equation align, regardless of differences in pre-chamber volume or jet intensity. This indicates that the jet penetration distance can be broadly summarized by the momentum, which is the product of velocity and mass flow rate at the connecting nozzles. 136 Ryosuke Shiina, Yusuke Shintani, Hirokazu Ando, Noritaka Kimura 136 <?page no="137"?> Figure 14: Physical State Quantities for Theoretical Equation Figure 15: Calculated Penetration Fomula Basis Figure 16: Correlation between CFD and Fomula Penetration Distance Consideration of the Relationship between Flame Formation and Fast Combustion 137 137 <?page no="138"?> (2) (3) 3.5 Momentum and Main and Pre-chamber Pressure Ratio This section shows that the momentum of the jet is governed by the pressure ratio between the main-chamber and the pre-chamber. It also shows that while the mass flow rate and the velocity at the connecting nozzles, which constitute momentum, vary significantly between unburned and burnt jets, they do not substantially impact the momentum. The mass flow rate and the velocity within the connecting nozzles are represented by the equations of nozzle flow for compressible fluids (2) and (3) [17]. These equations suggest that the flow rate and the velocity are determined by the pressure ratio between the pre-chamber and main-chamber, and the state physical quantities of the pre-chamber (connecting nozzles). m˙ = P 0 RT 0 πd 2 4 2γ γ − 1 P 2 P 0 2 γ − P 2 P 0 γ + 1 γ v 1 = 2γ γ − 1 P 0 ρ 0 1 − P 2 P 0 γ − 1 γ P 0 : Pre-chamber pressure, T 0 : Pre-chamber temperature, ρ 0 : Pre-chamber density, R: Gas constant, P 2 : Main-chamber pressure, d: Connecting nozzle diameter, γ: Specific heat ratio Figure 17(a) shows the relationship between the pressure ratio of the main-chamber to the pre-chamber and the mass flow rate within the connecting nozzles. After ignition in the pre-chamber, the pressure ratio decreases (pre-chamber pressure > main-chamber pressure), resulting in an increase the mass flow rate. The transition from unburned to burnt jet is marked by a significant reduction in the mass flow rate, likely due to a rapid rise in the temperature of the jet gas. Figure 17(b) displays the relationship between the pressure ratio of the main-chamber to the pre-chamber and the velocity within the connecting nozzle. Similar to the mass flow rate, the velocity increases as the pressure difference grows. The transition from unburned to burnt jet results in a substantial increase in the velocity, presumably due to a sharp decrease in density (ρ) associated with the rise in gas temperature. Figure 17(c) shows the relationship between the pressure ratio (P 2 / P 0 ) and momentum. Similar to the mass flow rate and the velocity, the momentum increases with the pressure difference. There is no significant change at the point of burnt jet transition. Ultimately, it is believed that the decrease in the mass flow rate and the increase in the velocity counterbalance each other, leading to the conclusion that the jet momentum is determined solely by the pressure ratio between the main-chamber and pre-chamber. Therefore, the jet penetration distance discussed in section 3.5 is also considered to be governed by the pressure ratio. 138 Ryosuke Shiina, Yusuke Shintani, Hirokazu Ando, Noritaka Kimura 138 <?page no="139"?> Figure 17: Relationship between Pressure Ratio and Nozzle Flow 3.6 Pre-chamber Pressure Rise and Heat Release Rate This section suggests that the pressure of the pre-chamber is contemplated to be determined by the equilibrium between the heat release rate and the outflow of working gas. The pre-chamber is an open system, unlike the closed system of a combustion- chamber in SI combustion. Consequently, as the pressure within the pre-chamber rises due to heat release, working gas flows out to the main-chamber (unburned jet), reducing the quantity of working gas in the pre-chamber. Therefore, the pressure inside the pre-chamber is considered to be dictated by the balance between ‘the pressure increase due to heat release’ and ‘the pressure decrease due to the outflow of working gas from the connecting nozzles’. Figures 18 and 19 depict the relationship between the heat release rate per pre-chamber volume, the outflow rate of working gas per pre-chamber volume, and the rate of change in the pre-chamber pressure. When the pressure change rate is positive, the pressure in the pre-chamber rises. The pressure reaches its maximum when the pressure change rate is zero. If the pressure change rate is negative, the pressure decreases. To facilitate comparison between different pre-chamber volumes, values per pre-chamber volume are used. In Figure 18, as the heat release rate per pre-chamber volume increases, the pressure rise rate also increases. The maximum values of the heat release rate for different pre-chamber volumes do not vary as much as the volume differences, suggesting that combustion is Consideration of the Relationship between Flame Formation and Fast Combustion 139 139 <?page no="140"?> slower in smaller pre-chambers and faster in larger ones. This is discussed further in Section-3.8. From Figure-19, it is observed that as the pre-chamber pressure rises, the outflow from the connecting nozzles also increases. Eventually, as the heat release rate decreases, the effect of the outflow becomes dominant, turning the pressure rise rate negative (pressure decreases). Furthermore, as the pressure rise in the main-chamber commences, the flow rate turns negative (backflow). The maximum flow rate is larger for smaller pre-chamber volumes. Figure 18: Relationship between Heat Release Rate in Pre-chamber per Volume and Rate of Pre- chamber Pressure Increase Figure 19: Relationship between Mass Outflow Rate per Volume and Rate of Pre-chamber Pressure Increase Figure 20 delineates the equilibrium between the heat release rate and the outflow rate of working gas, indicating that the upper left of the graph represents a greater pressure 140 Ryosuke Shiina, Yusuke Shintani, Hirokazu Ando, Noritaka Kimura 140 <?page no="141"?> increase due to the higher heat release rate compared to the pressure decrease caused by the outflow of working gas. The three points within the graph mark the transition of unburned jet to burnt jet. The trajectory up to these points reveals that smaller pre-chamber volumes are more significantly affected by the outflow of working gas relative to the heat release rate, suggesting that the pressure rise within the pre-chamber is less pronounced, resulting in a weaker jet ejection. Conversely, larger pre-chamber volumes are less susceptible to the influence of the outflow of working gas, achieving higher pressure increases, thereby strengthening the jet. Figure 20: Relationship between Heat Release Rate in Pre-chamber per Volume and Mass Outflow Rate per Volume 3.7 Heat Release Rate in Pre-chamber and Volume of Pre-chamber In this section, it is shown that an increase in the pre-chamber volume leads to a rise in the heat release rate due to an escalation in turbulent intensity. The flame ignited at the top of the pre-chamber propagates similarly to conventional SI combustion, eventually reaching the connecting nozzles at the bottom of the pre-chamber. Figure 21 represents the relative history of the turbulent flame surface area to the no-wrinkled flame surface area during the combustion within the pre-chamber. Despite changes in the pre-chamber volume, the maximum value of the no-wrinkled flame surface area does not significantly change, reflecting the constraint imposed by the inner diameter. As depicted in Figure 2, the volume of the pre-chamber varies with its height, while the inner diameter remains constant, thereby constraining the no-wrinkled flame surface area. Conversely, the turbulent flame surface area varies considerably with the pre-chamber volume, suggesting that larger pre-chamber volumes correspond to higher turbulent intensities. Consideration of the Relationship between Flame Formation and Fast Combustion 141 141 <?page no="142"?> Figure 21: Relationship between Turbulent and No-wrinkled Flame Surface Area in Pre-chamber Figure 22 illustrates the history of turbulent intensity within the pre-chamber during the latter phase of the compression stroke, while Figure 23 displays the distribution of turbulent intensity within three different pre-chamber volumes at the end of compression at 720deg.CA. Notably, in specifications with larger pre-chamber volumes, the turbulent in‐ tensity significantly increases towards the end of the compression stroke and subsequently attenuates. There is a high correlation between the differences in pre-chamber volume and the turbulent intensity at the end of compression at 720deg.CA. Figure 24 shows the history of flow velocity through the connecting nozzles during the latter phase of the compression stroke. The specifications of the connecting nozzles are the same regardless of the pre-chamber volume size. As mentioned in Section 3.7, larger pre-chamber volumes are less affected by changes in the quantity of working gas. Consequently, the pressure difference between the main-chamber and pre-chamber during the compression stroke is larger, resulting in increased inflow velocity. It is thought that this high flow velocity generated high turbulence intensity in the pre-chamber. Figure 22: TKE in Pre-chamber of Compression Stroke 142 Ryosuke Shiina, Yusuke Shintani, Hirokazu Ando, Noritaka Kimura 142 <?page no="143"?> Figure 23: Distribution of TKE in Pre-chamber Figure 24: Mass Flow Rate of Nozzles of Compression Stroke 4 Conclusion In the internal combustion engine with pre-chamber jet combustion, a distinct combustion process and its mechanisms were discussed using combustion CFD Converge, finding the following insights: 1. The second half of the combustion duration (MFB50-90) is characterized by a high increasing rate of integral of induction time. Therefore, in terms of suppressing knocking through fast combustion, MFB50-90 is more dominant than MFB10-50. 2. Pre-chamber jet combustion creates a jet stream with a long penetration distance that go through to the bore end, forming a complex and heterogeneous flame front. This increases the combustion speed, resulting in shorter duration for MFB50-90. Consideration of the Relationship between Flame Formation and Fast Combustion 143 143 <?page no="144"?> 3. The jet penetration distance can be summarized by the momentum, which is the product of the flow velocity and mass flow rate through the connecting nozzles. Both are determined by the pressure ratio between the main-chamber and pre-chamber. 4. Increasing the volume of the pre-chamber accelerates the inflow velocity into the pre-chamber, enhancing the Turbulent Kinetic Energy (TKE) within. This, in turn, increases the heat release rate. Meanwhile, the relative impact of the outflow of working gas diminishes. Consequently, the pre-chamber pressure rises, and the jet penetration is strengthened. References [1] Intergovernmental Panel on Climate Change (IPCC): Global Warming of 1.5degC Presentation to the wrapcup of the Talanoa Dialogue preparatory phase, COP24 Presentation, (2018) [2] Shuji Kimura, Hiromi Matsuura, Takashi Kikuchi, Kenji Tsuchiya : Examination of AICE’s Technical Scenario for Carbon Neutrality in Vehicles Equipped with Internal Combustion Engines, Society of Automotive Engineers of Japan 2021 Autum Convention Proceedings, 20216181, (2021) [3] Marc Sens, Emanuel Binder, Paul-Benjamin Reinicke, Michael Riess, Thorsten Stappenbeck, Marcus Wöbke: Pre-chamber Ignition and Promising Complementary Technologies,27th Aachen Colloquium Automobile and Engine Technology 2018, p.-957-998 (2018) [4] Michael Bassett, Adrian Cooper, Anthony Harrington, David Pates, Simon Reader, Martin Berger: Passive MAHLE Jet Ignition System Demonstrator,29th Aachen Colloquium Sustainable Mobility 2020, p.-395-420 (2020) [5] Ryohei Ono, Yuji harada, Kazuhiro Nagatsu, Hirotaka Suzuki, Kenji Uchida, Shinya Iida, Tatsuya Fujikawa : Study on the Pre-chamber technology application to gasoline engine combustion, 32th Internal Combustion Engine Symposium Proceedings, 20214885, (2021) [6] Taketo Yamada, Masayoshi Takahashi, Kenichiro Ikeya, Noriyuki Takegata : Intake Design for Reduction of Duration of Combustion, Honda R&D Technical Review, Vol. 27, No. 2, p.-67-75, [7] Hiroki Kobayashi, Kiminiori Komura, Keitaro Nakanishi, Atsushi Ohta, Hiroki Narumi, Noriyuki Takegat : Technology for Enhancing Thermal Efficiency of Gasoline Engine by Pre-chamber Jet Combustion,Honda R&D Technical Review, Vol.30, No.2, p.-57-64 (2018) [8] Marcus Wöbke, Paul-Benjamin Reinicke, Michael Rieß, Lorenz von Römer, Marc Sens: Charac‐ terization of the Ignition and Early Flame Propagation of Pre-chamber Ignition System in a High Pressure Combustion Cell, Ignition Systems for Gasoline Engines 4th International Conference 2018, p.-385-423 (2018) [9] Olivier Colin, Adlène Benkenida, Christian Angelberger: 3D Modeling of Mixing, Ignition and Combustion Phenomena in Highly Stratified Gasoline Engines, Oil & Gas Science and Technology - Rev. IFP, Vol.58, No.1, p.-47-62 (2003) [10] Olivier Laget, Stéphane Chevillard, Guillaume Pilla, Xavier Gautrot, Thierry Colliou: Investiga‐ tions on Pre-chamber Ignition Device Using Experimental and Numerical Approaches, JSAE Technical Paper, 20199342, (2019) [11] Hirokazu Ando, Yusuke Shintani, Hiroki Kobayashi, Ryosuke Shiina, Noritaka Kimura: Study of Knocking Mitigation and Thermal Efficiency Enhancement of Pre-chamber Jet Combustion in Stoichiometric Gasoline Engine, JSAE Technical Paper, 20239024, (2023) 144 Ryosuke Shiina, Yusuke Shintani, Hirokazu Ando, Noritaka Kimura 144 <?page no="145"?> [12] Mamoru Tanahashi, Yuzuru Nada, Toshio Miyauchi : Organized Structure in Turbulent Flames,NAGARE Journal of The Japan Society of Fluid Meckanics, Vol.23, No.5, P375-384 (2004) [13] Akihiro Hayakawa, Yukito Miki, Toshihiko Kubo, Yukihide Nagano, Toshiaki Kitagawa : Varia‐ tions of Turbulent Burning Velocity and Flame Front Shape of Spherically Propagating Premixed Turbulent Flame with Effective Turbulence Intensity,Journal of the Combustion Society of Japan, Vol.55, No.172, p.-202-209 (2013) [14] Toru Noda, Kazuya Hasegawa, Masaaki Kubo, Teruyuki Itoh: Development of Transient Knock Prediction Technique by Using a Zero-Dimensional Knocking Simulation with Chemical Kinetics, SAE World Congress 2004, 2004-01-0618, (2004) [15] Convergent Science: CONVERGE v3.0.19 manual, p.-430-437, p.-443, p.-445-446 (2021) [16] Yutaro Wakuri, Masaru Fujii, Tatsuo Amitani, Reijiro Tsuneya : Studies on the Penetration of Fuel Spray of Diesel Engine,Transactions of the Japan Society of Mechanical Engineers, Vol.25, No.156, p.-820-826 (1959) [17] Kazuyasu Matuso : Compressible Fluid Dynamics : Theory and Analysis in Internal Flow, Tokyo Japan, Ohmsha, 1994, p.-76-77 Consideration of the Relationship between Flame Formation and Fast Combustion 145 145 <?page no="147"?> 1 ICE Group - Energy Department, Politecnico di Milano 2 ICE Group - Energy Department, Politecnico di Milano 3 ICE Group - Energy Department, Politecnico di Milano 3D CFD modelling of TJI combustion achieved by active or passive pre-chamber Alessandro Nodi 1 , Lorenzo Sforza 2 , Tommaso Lucchini 3 Abstract The abatement of pollutants and CO 2 emissions is an essential requirement for the next generation of internal combustion engines. The Turbulent Jet Ignition (TJI) strategy for combustion initiation is a promising solution to obtain an efficiency increase for engines fed by both conventional and renewable fuels. The combustion duration can be reduced by igniting the air-fuel mixture inside a small volume connected to the cylinder, eventually obtaining a main chamber mixture ignition through highly turbulent hot jets released by the pre-chamber. In this work two different engine configurations are presented: a motorbike gasoline engine equipped with a passive pre-chamber and a research engine featuring a port-fuel injection of ammonia and an active pre-chamber fuelled with hydrogen. In both cases the Computational Fluid-Dynamics (CFD) simulations were used as a tool to achieve a deeper understanding of the fluid-dynamics inside the engine, in order to help with the comprehension of the experimental data. The first study is focused on the comparison of the engine operating with a conventional Spark-Ignition (SI) combustion and with a TJI combustion strategy. The application of a pre-chamber ignition technology on a motorbike engine can help to obtain a faster and more stable combustion, opening to the possibility of increasing the engine compression ratio or burning leaner air-fuel mixtures. 3D CFD simulation of the full engine cycle were performed. Specific care was given to the analysis of the passive pre-chamber scavenging process, which is crucial for a reliable estimation of the exhaust gases stratification and of the velocity flow field distribution inside the pre-chamber at the spark timing. The combustion process was simulated with both ignition strategies using a unique flamelet-based combustion model, regardless the selected combustion mode. The dependency of the hot-gases ejection process from the flow field inside the main and the pre-chamber was analysed. A numerical-experimental comparison was carried out in terms of pressure and heat release trends, demonstrating the reliability of the employed CFD methodology in the design of high-performance SI engines. <?page no="148"?> In the second study the investigation is focused on the impact of different hydrogen direct injection strategies on the main-chamber ammonia-air mixture combustion development. Ammonia is considered an ideal candidate as future energy carrier, due to the absence of carbon content. However, the premixed combustion of ammonia in engines calls into question the effectiveness of traditional ignition strategies, due to a higher minimum ignition energy and lower combustion speed compared to conven‐ tional fossil fuels. The application of a hydrogen-fuelled pre-chamber aims to settle the main drawbacks affecting SI engines fed with pure ammonia. A 3D CFD analysis was carried out to shed light on the complex phenomena affecting the dual fuel inhomogeneous premixed combustion process and highlight the challenges related to hydrogen injection strategies. The results show that the injection duration mainly impacts the hydrogen passing into the main-chamber, modifying the auto-ignition limit of the engine. This matter could be better investigated to obtain a controlled Spark-Assisted Compression Ignition (SACI) combustion, thanks to the low flame speed characterizing ammonia. 1 Introduction and motivation In the following years are expected more restrictive regulations concerning greenhouse gases and pollutant species emissions in all the countries of the European Union (European Union, 2021). One of the most promising solutions to increase the engine efficiency (hence reducing emissions) of conventional engines thanks to an improved combustion process is the TJI (Turbulent Jet Ignition). The advantages of the TJI combustion are already demonstrated by the available literature on several engine configurations (Attard et al., 2010; Novella et al., 2020). The two main remarkable improvements compared to the conventional SI combustion are the significant increase of the combustion speed and the considerable reduction of the cyclic variability (Hua et al., 2020; Sementa et al., 2019). The mentioned positive features of the TJI strategy application derive from the multi-site ignition in main chamber enabled by the previous combustion in pre-chamber and subsequent release of highly turbulent hot jets of burned gases and wrinkled flames (Bunce et al., 2014). These benefits can be exploited also in the context of fuels with zero carbon emissions, for instance by means of an actively fuelled pre-chamber with hydrogen to increase the ammonia turbulent flame speed and simultaneously reduce the required amount of hydrogen (Zhang et al., 2022), hence reducing the demand on on-board catalytic dissociation of ammonia for hydrogen generation, avoiding the cost and inconvenience of hydrogen storage. The goal of the presented research activities was to perform 3D CFD investigations of the complete engine cycle of two completely different engine configurations, both sharing the TJI concept. The first presented configuration is a naturally aspirated motorbike engine equipped with a passive pre-chamber and operated at high engine speed. The second presented configuration is a research (CFR) engine operated at low engine speed and equipped with a hydrogen-fuelled active pre-chamber aiming at providing a high ignition energy for the combustion initiation of a homogenous stoichiometric mixture of air and ammonia. 148 Alessandro Nodi, Lorenzo Sforza, Tommaso Lucchini 148 <?page no="149"?> The purpose of the CFD analyses is to clarify the mixture composition and flow field inside the pre-chamber, understanding their effect on the combustion propagation for the challenging operating conditions of the two different experimental applications presented. 2 Numerical methodology 2.1 Numerical models The 3D CFD simulations are carried out in a U-RANS context, to mimic the average-cycle gas exchange and combustion processes of the engine. The flame front propagation is simulated by the transport equation of the unburned gas fraction b, also called regress variable, according to the FAE (Flame Area Evolution) model from Weller (Weller et al., 1994, 1998): ∂ρb ∂t + ∇ • ρUb + ∇ • μ t ∇b = ρ u S u Ξ ∇b + ω˙ ign This model is based on the laminar flamelet concept where the enhancement of the flame front due to the turbulence is considered through the flame wrinkle factor Ξ, namely the turbulent-to-laminar flame speed ratio. The laminar-to-turbulent transition process occurring after the spark-ignition event is modelled according to the approach proposed by Herweg and Maly (Herweg & Maly, 1992). Briefly, it consists in evolving the Ξ in the regress variable equation from 1 (laminar flame) to its equilibrium value Ξ eq (fully turbulent flame), according to suitable models (Gülder, 1991; Peters, 2000). Clearly, the wrinkle factor Ξ is strictly linked to the turbulence intensity, which in this study is predicted by the k − ε turbulence model. The laminar flame speed value S u is estimated according to suitable correlations for the selected single-fuel or dual-fuel combustion (Gulder, 1984; Pessina et al., 2022). The second right-hand side term in the regress variable equation is the source term due to the ignition process. This stage is handled by means of a simplified deposition model (Sforza et al., 2019; Yang et al., 2012). Further details on the adopted methodology for the combustion process prediction can be found in the papers by Sforza et al. (Sforza et al., 2019, 2023). 2.2 Simulation setup 2.2.1 Initial and boundary conditions Crank-angle dependent boundary conditions of pressure and temperature retrieved from 1D simulations are imposed at the open surfaces of the 3D computational domain, on both the intake and the exhaust pipes. This allowed to include the dynamics of the engine breathing into the 3D simulations, without modelling the whole engine schematic. The initial conditions at the EVO in cylinder and pre-chamber are imposed according to the measured pressure values and to the temperature values retrieved from 1D simulations. The mixture composition in cylinder, pre-chamber and exhaust ducts is initialized as fully burned products from a prefect oxidation of the air-fuel mixture at the specified equivalence ratio. The port fuel injection is not simulated, hence in the intake duct a perfectly homogeneous premixed air-fuel mixture is assumed. 3D CFD modelling of TJI combustion achieved by active or passive pre-chamber 149 149 <?page no="150"?> 2.2.2 Mesh The complete engine-cycle simulation is handled with a multi-mesh approach. This method‐ ology consists in generating a series of mesh from a user defined start time, deforming each mesh until the minimum required mesh quality parameters are satisfied. The 3D simulation is automatically performed in series on each generated mesh, by systematically mapping the results from the final valid instant of a mesh to the next undeformed mesh. More details about this methodology are described in the paper by Lucchini et al. (Lucchini et al., 2007). An adequate base cell size (≈ 2mm) is chosen for the intake and exhaust ducts regions, then volume refinements of different levels are applied where required by the complex flow conditions, such as cylinder (0 . 5 − 1mm), pre-chamber (0 . 15 − 0 . 3mm), around the valves and at the pre-chamber nozzles outlet. 3 Validation 3.1 Passive pre-chamber engine 3.1.1 Experimental setup The experiments on the 4-stroke, 2-cylinder, naturally aspirated motorbike engine were conducted by Marmotors s.r.l. The main engine geometrical and operating data are reported in Table 1. Number of cylinders [-] 2 Displacement [cm 3 ] 660 Bore [mm] 81 Stroke [mm] 63.8 Compression ratio [-] 13.5 Engine speed [rpm] 10500 Fuel type [-] Gasoline RON 95 Air-fuel ratio λ [-] 0.866 Table 1: Motorbike engine geometrical and operating data. The spark plug seat on the cylinder head is slightly modified to host a pre-chamber, which can be easily mounted on the original engine. After the engine head modification, tests with the traditional SI configuration are still possible thanks to a dedicated spark plug adapter. 3.1.2 Results The analysis of the gas exchange process showed that a significant amount of exhaust residuals gases remains trapped inside the pre-chamber volume (Figure 1(a)). In particular, at the IVC the amount of residual gases inside the cylinder is 1.2% of its total trapped mass, while in the pre-chamber is the 36.3% of the passive pre-chamber trapped mass. The latter value reduces to 6.5% moving to the spark timing, proving that the pre-chamber filling mainly takes place during the compression stroke, when the air-fuel fresh mixture is forced into the 150 Alessandro Nodi, Lorenzo Sforza, Tommaso Lucchini 150 <?page no="151"?> pre-chamber by the piston (Figure 1(b)). An asymmetric flow field distribution was noticed inside the pre-chamber at the SA (Figure 1(c) and (d)). The causes were attributed to the tumble motion developed inside the cylinder during the intake stroke and to the inherent asymmetric geometry of the pre-chamber caused by the L-shape spark plug electrode. Figure 1: Pre-chamber scavenging process: (a) mass trapped inside the pre-chamber from EVO to SA; (b) oxygen concentration in cylinder and pre-chamber from EVO to SA; (c) EGR distribution in pre-chamber at the SA; (d) Velocity flow field inside the pre-chamber at the SA. Figure 2(a)-(b) compare the measured pressure traces with the numerical ones achieved during the combustion process inside the pre-chamber and cylinder, respectively. The cause of the underestimation of the first pressure peak inside the pre-chamber is attributed to an inaccurate prediction of the nozzles flow coefficient probably caused by their simplified geometry where sharp edges are considered at both ends of each nozzle, while from experimental observations the real nozzles ends are smoothed. Despite this aspect, the numerical-experimental agreement observed for the pressure evolution inside pre-chamber and cylinder can be considered satisfactory. 3D CFD modelling of TJI combustion achieved by active or passive pre-chamber 151 151 <?page no="152"?> Figure 2: Experimental and numerical pressure traces in pre-chamber and cylinder during the combus‐ tion phase. Analysing the flame front evolution inside both combustion chambers it is observed a preferential flame growth in one side of the pre-chamber and an asymmetric ejection process inside the main chamber. The causes of such behaviours are related to the existing flow velocity fields inside the pre-chamber, which is characterized by a small clockwise rotating tumble motion, and inside the cylinder, where a residual counterclockwise tumble motion influences the hot turbulent jets propagation. The comparison among the results of the simulations performed for both the conven‐ tional SI engine configuration and the TJI combustion strategy confirmed the faster flame speed provided by the pre-chamber ignition. In particular, a total combustion duration (i.e. MFB 0-90) reduction of 14.5° (22.5%) is detected when moving to the TJI configuration. The major benefits introduced by the passive pre-chamber layout are observed during the first half of the combustion process (i.e. MFB 0-50), while the second half is almost similar between the two configurations. This can be explained by the higher similarity of the flame front between the TJI and the traditional SI cases. 3.2 Active pre-chamber engine 3.2.1 Experimental setup Experiments were conducted using a Waukesha Cooperative Fuel Research (CFR) octane rating engine, modified for pre-chamber operation. The pre-chamber is actively fuelled by a hydrogen gas injector, while the main chamber is filled with a homogeneous, stoichiometric ammonia-air mixture. Engine geometrical data and operating conditions are summarized in Table-2. 152 Alessandro Nodi, Lorenzo Sforza, Tommaso Lucchini 152 <?page no="153"?> Number of cylinders [-] 1 Displacement [cm 3 ] 611 Bore [mm] 82.5 Stroke [mm] 114.3 Compression ratio [-] 16 Engine speed [rpm] 900 PFI fuel [-] NH 3 Pre-chamber DI fuel [-] H 2 Air-fuel ratio λ [-] 1 Baseline injection [°bTDC---ms] 70---1 Delayed injection [°bTDC---ms] 30---1 Longer injection [°bTDC---ms] 70---2 Table 2: CFR engine geometrical data and tested operating conditions. Three cases were considered in the present study to evaluate the effect of hydrogen injection on the subsequent combustion process. The baseline case refers to an early injection (70° bTDC) of 1 ms duration. Alternate cases included a delayed injection case (30° bTDC) where the same injection duration occurred with a shorter dwell between injection and spark timing, and a longer injection duration at the same early injection timing. The spark timing was kept constant at 20° before top dead center (bTDC) for all cases. 3.2.2 Results Figure 3: Baseline injection case (SOI70-1ms): comparison between measured and calculated average pressure traces inside both the combustion chambers. The coloured area includes the variability of 300 measured single cycles. 3D CFD modelling of TJI combustion achieved by active or passive pre-chamber 153 153 <?page no="154"?> Figure 3 shows the measured pressure traces evolution inside both pre-chamber and main-chamber, compared against the calculated ones, for the SOI70-1ms condition (baseline). It can be noticed that the hydrogen direct-injection causes a low-intensity pressure rise inside the pre-chamber, which is responsible for the leakage towards the cylinder of a significant share of the premixed ammonia-air mixture, together with a small amount of injected hydrogen. The satisfactory agreement observed between CFD results and experimental measurements allows to use the proposed numerical method as a diagnostic tool to better investigate the different proposed direct injection strategies. Figure 4 compares the experimental heat release rate of the three tested configurations, also considering the engine cyclic variability. Figure 4: Comparison between the measured in-cylinder apparent heat release rates of the average cycle for each investigated injection strategy: SOI70-1ms (baseline), SOI70-2ms (increased injection duration) and SOI30-1ms (delayed start of injection). Each coloured area includes the variability of 300 measured single cycles. The case with delayed injection (SOI30-1ms) shows some cycles experiencing a quite weak heat release. This is caused by a considerably richer mixture inside the pre-chamber consequence of the higher counter-pressure at the SOI compared to the baseline case that avoids the leakage of hydrogen towards the main chamber. The rich mixture inside the pre-chamber leads to a flame slowdown due to the Lewis number impact (Le ≫ 1) on the flame front stretch (Dai et al., 2023), consequently altering the subsequent jet-phase combustion in the main chamber. The case with longer injection (SOI70-2ms) shows instead a second sharp peak of the AHRR traces around 10 °aTDC. The cause should be attributed to a more abundant hydrogen leakage towards the main chamber leading to auto-ignition events caused by a combination of the high reactivity of hydrogen and the low laminar flame speed of ammonia, in agreement with previous studies (Mounaïm-Rousselle et al., 2022; Reggeti et al., 2023). 154 Alessandro Nodi, Lorenzo Sforza, Tommaso Lucchini 154 <?page no="155"?> 4 Conclusions The achieved numerical results for both the presented case studies demonstrated the robustness of the employed CFD methodology for two different engine configurations characterized by a TJI combustion strategy. These 3D CFD simulations can be considered a reliable tool to investigate both the injection and combustion processes, hence supporting the design of SI engines operating in unconventional conditions and equipped with a pre-chamber. References Attard, W. P., Fraser, N., Parsons, P., & Toulson, E. (2010). A Turbulent Jet Ignition pre-chamber combustion system for large fuel economy improvements in a modern vehicle powertrain. SAE Technical Papers, 3(2), 20-37. https: / / doi.org/ 10.4271/ 2010-01-1457 Bunce, M., Blaxill, H., Kulatilaka, W., & Jiang, N. (2014). The effects of turbulent jet characteristics on engine performance using a pre-chamber combustor. SAE Technical Papers, 1. https: / / doi.org/ 10.4 271/ 2014-01-1195 Dai, H., Wang, J., Cai, X., Su, S., Zhao, H., & Huang, Z. (2023). Lewis number effects on laminar and turbulent expanding flames of NH3/ H2/ air mixtures at elevated pressures. Proceedings of the Combustion Institute, 39(2), 1689-1697. https: / / doi.org/ 10.1016/ j.proci.2022.07.200 European Union. (2021). European Green Deal. Gulder, O. L. (1984). Correlations of laminar combustion data for alternative SI engine fuels. SAE Technical Papers. Gülder, Ö. L. (1991). Turbulent premixed flame propagation models for different combustion regimes. Symposium (International) on Combustion, 23(1), 743-750. https: / / doi.org/ 10.1016/ S0082-0784(06 )80325-5 Herweg, R., & Maly, R. (1992). A fundamental model for flame kernel formation in SI engines. SAE Transactions, 1947-1976. Hua, J., Zhou, L., Gao, Q., & Feng, Z. (2020). Effects on Cycle-to-Cycle Variations and Knocking Combustion of Turbulent Jet Ignition (TJI) with a Small Volume Pre-Chamber. SAE Technical Papers. https: / / doi.org/ 10.4271/ 2020-01-1119 Lucchini, T., D’Errico, G., Jasak, H., & Tukovic, Z. (2007). Automatic mesh motion with topological changes for engine simulation. SAE Technical Papers, 2007(724), 776-790. https: / / doi.org/ 10.4271/ 2007-01-0170 Mounaïm-Rousselle, C., Mercier, A., Brequigny, P., Dumand, C., Bouriot, J., & Houillé, S. (2022). Performance of ammonia fuel in a spark assisted compression Ignition engine. International Journal of Engine Research, 23(5), 781-792. https: / / doi.org/ 10.1177/ 14680874211038726 Novella, R., Pastor, J., Gomez-Soriano, J., Barbery, I., Libert, C., Rampanarivo, F., Panagiotis, C., & Dabiri, M. (2020). Experimental and Numerical Analysis of Passive Pre-Chamber Ignition with EGR and Air Dilution for Future Generation Passenger Car Engines. SAE Technical Papers. https: / / doi.org/ 10.4271/ 2020-01-0238 Pessina, V., Berni, F., Fontanesi, S., Stagni, A., & Mehl, M. (2022). Laminar flame speed correlations of ammonia/ hydrogen mixtures at high pressure and temperature for combustion modeling applications. International Journal of Hydrogen Energy, 47(61), 25780-25794. https: / / doi.org/ 10.10 16/ j.ijhydene.2022.06.007 3D CFD modelling of TJI combustion achieved by active or passive pre-chamber 155 155 <?page no="156"?> Peters, N. (2000). Turbulent Combustion. In Cambridge University Press. https: / / doi.org/ 10.1137/ 1.97 81611971064 Reggeti, S., Kane, S., & Northrop, W. (2023). Experimental Investigation of Spark-Assisted Compres‐ sion-Ignition with Ammonia-Hydrogen Blends. Journal of Ammonia Energy, 1(1), 91-105. https: / / doi.org/ 10.18573/ jae.21 Sementa, P., Catapano, F., Di Iorio, S., & Vaglieco, B. M. (2019). Effects of Prechamber on Efficiency Improvement and Emissions Reduction of a SI Engine Fuelled with Gasoline. SAE Technical Papers, October. https: / / doi.org/ 10.4271/ 2019-24-0236 Sforza, L., Lucchini, T., D’Errico, G., Gianetti, G., Beatrice, C., & Tunestål, P. (2023). A Numerical Methodology for the Design of Active Prechambers in Spark-Ignition Engines. Combustion Science and Technology, 00(00), 1-35. https: / / doi.org/ 10.1080/ 00102202.2023.2219450 Sforza, L., Lucchini, T., Gianetti, G., & D’Errico, G. (2019). Development and Validation of SI Combustion Models for Natural-Gas Heavy-Duty Engines. SAE Technical Papers. https: / / doi.org/ 1 0.4271/ 2019-24-0096 Weller, H. G., Tabor, G., Gosman, A. D., & Fureby, C. (1998). Application of a flame-wrinkling les combustion model to a turbulent mixing layer. Symposium (International) on Combustion, 27(1), 899-907. https: / / doi.org/ 10.1016/ S0082-0784(98)80487-6 Weller, H. G., Uslu, S., Gosman, a D., Maly, R. R., Herweg, R., & Heel, B. (1994). Prediction of Com‐ bustion in Homogeneous-Charge Spark-Ignition Engines. International Symposium COMODIA 94, July, 163-169. Yang, X., Solomon, A., & Kuo, T. W. (2012). Ignition and combustion simulations of spray-guided SIDI engine using Arrhenius combustion Zhang, T., Ji, C., Wang, Z., Wang, S., Yang, H., Wang, H., & Jiang, N. (2022). Experimental Investigation on the Combustion Characteristics of NH3/ H2/ air by the Spark Ignition and Turbulent Jet Ignition. Combustion Science and Technology, 196(1), 73-94. https: / / doi.org/ 10.1080/ 00102202.2022.2063687 Abbreviations BDC bottom dead center SA spark advance CAD crank angle degrees SI spark ignition CFD computational fluid-dynamics SOI start of injection EGR exhaust gas residuals TDC top dead center EVO exhaust valve opening TJI turbulent jet ignition IVC intake valve closing U-RANS unsteady-Reynolds averaged Navier-Stokes MFB mass fraction burnt 156 Alessandro Nodi, Lorenzo Sforza, Tommaso Lucchini 156 <?page no="157"?> 1 Aramco Overseas Company: 232 Avenue Bonaparte, Rueil-Malmaison, 92500, France 2 Renault Group / Ampere: 1 Av. du Golf, 78280 Guyancourt, France 3 Apside : 4 Place des Ailes, 92100 Boulogne-Billancourt, France 4 Renault Group / Horse: Bd. Preciziei Nr.-3G, Sector 6, Bucuresti, 062204 Romania Assessment of passive TJI technology on a mild hybrid powertrain and its performance on cold operating conditions Dimitrios Karageorgiou 1 , Thierry Prunier 2 , Matej Myslivecek 3, Carmen Vesel 4 1 Abstract The increasing global emphasis on CO 2 reduction, engine efficiency improvements, and the stringent mandates set forth by the Paris Agreement have catalyzed the need for advanced technological solutions. With the automotive industry challenged to meet CO 2 reduction targets and contribute to the global effort of mitigating climate change, there is a need to use engine technology that can simultaneously enhance performance and reduce environmental impact. In this respect, spark ignition engines development led to various solutions towards improving combustion with dilution. Passive pre-chamber ignition (Passive TJI technology) is considered as an effective enabling technology to operate in moderate EGR levels with good combustion stability while increasing the compression ratio of the engine without penalty in power density. While the operation of a passive pre-chamber is well demonstrated in medium and high loads, the low load and cold start operating points present several challenges in terms of ignitability and combustion stability. In this work, several passive pre-chamber designs were integrated into a 4-cylinder HR18 1.8L engine, replacing the standard M10 spark plug. The operation with the pre-chamber enabled the increase of the compression ratio from 14 to 15, while keeping a similar or better level of peak power output. Results from the multi-cylinder engine testing are presented at different engine speeds and loads with the engine able to run in the similar levels of dilution with EGR, while having more stable combustion and improved efficiency up to 1.5 %. With the support of CFD simulations, the importance of internal pre-chamber aerodynamic effects was highlighted and correlations between turbulence, local velocity level, local residual distribution and combustion performance were confirmed by testing. <?page no="158"?> Furthermore, dedicated testing under cold start conditions (up to -20 degC) was performed to demonstrate the startability of the engine and evaluate the parameters influencing the best operation, including the catalyst heating phase. The engine was also equipped with optical fibers in the pre-chamber to provide information about the ignitability of the pre-chamber mixture during the tested operating conditions. Additionally, optical access into the main chamber through an endoscope with a high-speed camera allowed for correlating the pre-chamber combustion with an onset of main charge combustion. The testing at -20 degC showed that even the non-op‐ timized PCSP designs were able to start the engine with good performance when enrichment increased. At -10 degC, although cold start performance and emissions were not at the level of the conventional spark plug, further optimization is possible and in combination with the HEV catalyst warm-up strategy, it seems possible to achieve the required performance. In conclusion, pre-chamber optimization measures are proposed and evaluated in CFD. 2 Introduction The transportation sector currently accounts for almost one quarter of the total global energy-related CO 2 emissions [1]. At the same time, passenger transport demand will further increase, making even more challenging the decarbonization of transport [2]. To meet the global climate objectives, sustainable and affordable transport solutions must be implemented which requires diverse policy and technological adaptations. In this context, it is essential to further develop clean and efficient internal combustion engines (ICEs), as they are about to represent an important part of the fleet mix, for the following decades. In combination with Low Carbon fuels, efficient Internal Combustion Engines can play a key role in fulfilling the objectives of the Paris Agreement and the Green Deal. Spark ignition engines development is oriented to more advanced combustion processes that can enhance the thermodynamic efficiency of the engine. An important technology towards this direction is Turbulent Jet Ignition (TJI) which uses a pre-chamber to initiate the combustion and generates turbulent jets through the nozzles of the pre-chamber, into the main combustion chamber. The high turbulence, high temperature and high radical composition of the jets, distributed into a wide volume in the cylinder, provide the ignition energy that is hundreds of times higher than the one of a conventional spark plug. TJI technology can appear in many variants which usually are categorized in two distinct approaches, known as active and passive pre-chamber. In the active approach, a spark plug and a secondary injector are positioned inside the pre-chamber. In the passive approach, only an ignition source is located inside the pre-chamber and the mixture is guided inside its volume through the piston movement. While active TJI can result in higher efficiency gains due to its ability to operate in extremely diluted conditions, passive TJI presents a very good benefit to cost ratio for an affordable engine. This publication presents an assessment of passive TJI technology on the HR18 engine of Horse which is used as a platform for development of the technology bricks for the next generation of Horse high efficiency engine. 158 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 158 <?page no="159"?> 3 Engine platform 3.1 HR NA engine family and future technology roadmap The main requirement for the next generation of engines is to improve efficiency, especially on driving profiles such as fast regional road or highway. During such use case, for HEV and PHEV applications, the electric mode is limited and after a certain mileage (depending on the battery size) the fuel consumption is mainly driven by the efficiency of the internal combustion engine. Some of the most important powertrain architectures are the affordable HEVs/ PHEVs implemented in the CMF-A entry vehicles plate-forms, such as the E-tech Horse architec‐ ture. - - Figure 1: E-tech Horse 2022 achitecture for MHEV and PHEV application These vehicles can represent an important part of the vehicle fleet during the next decades and their efficiency is crucial for achieving the regulation limits. For these powertrains, the implemented technical solutions, aiming to improve engine efficiency, must be at an excellent performance to cost ratio. Assessment of passive TJI technology on a mild hybrid powertrain and its performance 159 159 <?page no="160"?> Figure 2: Efficiency targets for future HR18 engine A technology roadmap for such an affordable but efficient Internal Combustion Engine can include well known solutions such as high stroke/ bore ratio, increased compression ratio and aggressive Atkinson valve profiles. Furthermore, the combination with a medium to high voltage electric traction system (>=200v) would allow to replace the accessory belt with an electrified accessory device, providing reduced friction losses. In addition to the previous, high EGR dilution is an important measure to achieve high levels of efficiency. In this respect, TJI combustion presents an important enabler for stable operation at high EGR due to its high ignition energy and jet-driven, volume combustion. As a result, TJI would be expected to increase combustion speed at lower EGR rate, providing better BSFC, and allow a lighter/ cheaper EGR system, easier to cool and to control. At this stage, a pragmatic approach has been chosen to assess the passive TJI technology and target moderate levels of EGR. Figure 3: Example of EGR sweep at 2500 rpm/ 8 bar BMEP with SSP showing BSFC improvement with EGR rate - expectation with TJI technology is to increase combustion speed at lower EGR rate providing better BSFC and allowing a lighter and cheaper EGR system, that is easier to cool and to manage with engine control system 160 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 160 <?page no="161"?> Figure 5: Engine test cell Figure 4: HR18 engine development roadmap 3.2 Test bench and Hardware setup The HR18 engine used for this study was an inline, four-cylinder, 1.8L, GDI, naturally as‐ pirated engine from Horse. Two variants of pistons were used during the testing campaign, resulting in two different compression ratios of 14: 1 and 15: 1. This engine serves as a develop‐ ment platform to develop TJI technology and defines the technology package of the for the future higher efficiency HR18evo engine. The engine is equipped with M10 SSP (standard spark plug) diameter and all PCSP (pre-chamber spark plug) variants tested were designed with the same external diameter. The engine was placed on a standard engine bench test cell. For specific testing in cold conditions, the test cell featured high cooling capability to allow to regulate the temperatures of the engine coolant, engine oil, charge of air cooler, fuel, intake air individually and up to -15 to -20degC for transient (cold start) conditions. Assessment of passive TJI technology on a mild hybrid powertrain and its performance 161 161 <?page no="162"?> Figure 6: Visualisation of optical accesses to PCSP and MC Table below shows the main temperatures for tests performed: - Air T o Oil T o - Water T o Hot tests 25 90 - 90 Late MFB50 potential 15 30 - 30 Cold start/ Warm-up tests -10 -10 - -10 Startability tests -15 -20 - -20 Table 1: Overview of the temperature conditions for engine testing In-cylinder pressure transducers were installed for each cylinder, with an additional transducer in the pre-chamber of one of the outer cylinders. Apart from the standard pressure indication measurement system, two dedicated optical accesses were used to investigate the pre-chamber and main chamber combustion. Emission measurement was performed with AVL gas analyzers. Simultaneous optical access inside the pre-chamber and the main chamber gave us the ability to trace the evolution of the flame and understand the main parameters that affect the interaction between the two. By using optical measurement techni‐ ques, it was possible to evaluate the quality of combustion in the pre-chamber and the main chamber separately. This lever allowed to drive the PCSP internal design optimization together with the PCSP hole layout optimization. The optical access inside the pre-chamber consisted of eleven optical fibers that were targeted in the region close to the spark plug electrodes, as seen in the Figure-8, and captured light in the near-infrared range of wavelengths. The sensor’s signal was passed to an amplifier, then to AVL’s X-ion and finally to AVL’s INDICOM. The optical fibers were able to capture the first stages of the flame kernel evolution and gave us insight about the direction and the speed of growth of the flame kernel but also provided a picture of the intensity of the combustion. In the Figure 9, an example of a heatmap created from the output of the optical fibers is visible. Downward propagation of the flame is highlighted. 162 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 162 <?page no="163"?> PCSP GDI injector MC Camera MC optical access Figure 7: Setup of the main chamber camera on the engine (left). Example of the MC camera view with combustion chamber components highlited (right). Figure 8: Scheme of PCSP optical fibers’ field of view targeting the spark plug electrodes (left) and their layout (right) - Figure 9: Visualisation of voltage (image of light in‐ tenstity) as captured by the PCSP optical acquisition system. Average of intensity of 250 cycles per each group of optical fibers. Non instantaneous flame prop‐ agation is highlighted by the arrow. The optical access of the main chamber was achieved with the use of XEVA XC-130 camera. The camera was able to capture photographs of the main chamber with a maximum acquisition rate of 100 Hz in the range of wavelengths from 900 to 1700 nm. Assessment of passive TJI technology on a mild hybrid powertrain and its performance 163 163 <?page no="164"?> 4 Validation of modelling 4.1 CFD setup Optimization of ICE applications often involves specific physical phenomena that cannot be assessed through experimental approaches. As a result, numerical simulation becomes a crucial tool for tackling issues related to innovative technologies development. This study aims at providing an understanding of key parameters monitoring performance of passive pre-chamber designs, in a context of highly EGR diluted, high tumble, gasoline engine. This study contains analysis based on two operating points - consistent with testing conditions (including load) - which are: 2500 rpm/ 8 bar 20 % EGR and 3000 rpm/ 6 bar 18% EGR. All simulations are conducted with Converge software (v3.0). The numerical domain represented solely one cylinder of the engine, while the intake and exhaust manifolds were limited to the close vicinity of the cylinder, as visible in Figure 10. Time-varying temperature/ pressure-imposed boundary conditions were applied at the extremities of the intake/ exhaust manifolds. The latter were generated from GT-POWER software. Temperatures of the walls of the engine were considered constant and estimated from GT-POWER. A base grid of 1.6mm was used in both ports, with a finer resolution of 0.4 mm in the MC, and 0.1 mm in the PCSP, ensuring a sufficient resolution in the more confined regions of the domain, such as the PCSP orifices. RNG k-ε approach was used for averaged turbulence modelling. Primary and secondary breakup of the injected fuel jet was modelled by KH-RT approach. The influence of engine leakage being considered negligible in the study, no blow-by modelling is employed. O’Rourke wall transfer model is used for heat losses. E10 fuel main properties used in experiment are reproduced by a single-component surrogate, ensuring consistency with experiments. Extended Coherent Flamelet Model (ECFM) model is used for modelling tur‐ bulent premixed combustion. In ECFM, turbulent flame propagation velocity is monitored by turbulence level (TKE) along with laminar flame speed at local conditions. Figure 10: Illustration of the computational domain, made of four regions: main chamber (blue), intake port (light green), exhaust port (darker green), PCSP (red). 164 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 164 <?page no="165"?> 4.2 Validation of the numerical setup 4.2.1 Spray modelling parameters Spray injection behavior is first validated on a closed volume vessel configuration, based on averaged liquid penetration length and averaged spray plumes parcels location. Liquid penetration is calculated along the axis of the injector and defined based on a threshold of 90 % evaporated mass. It is then averaged over the 6 spray plumes, respectively with the 6-holes injector. As visible in Figure 11, spray penetration displays deviations of about 10 % from the experimental data, which are considered acceptable for this study. The average position of parcels from each plume (i.e. each injector hole) is calculated on a plan perpendicularly situated 30 mm below the axis of the injector. This allows assessing the pattern of the spray plumes. The average position of each plume is calculated for 5 instants, evenly distributed between t = 1.5ms and t = 2.5ms (t=0 corresponding to injector opening), to avoid any non-stationarity effect from earlier stages of the injection. On Figure 11 (right), each plume is represented by a different color. For each plume, each bullet point represents an instant comprised in the t=1.5ms to t = 2.5ms observation timeframe. Minimal deviations between CFD and experimental target data are displayed, suggesting correct representation of the spray. These injection parameters are then fixed for the rest of the study. Figure 11: CFD results for spray validation. Spray penetration is on the left. Right plot displays spray plume center patterns: experimental plumes depicted by stars, and CFD plumes by bulletpoints. Both comparison metrics suggests acceptable deviations from experiments. 4.2.2 Mass transfer and combustion modelling parameters Trapped mass evolution, along with pressure evolution are validated against experimental data. Combustion model is validated based on the experimental data from PCSP1 and PCSP1A. Despite discrepancies between experimental and CFD pressure traces, experi‐ mental trends (i.e. better behavior for PCSP1) are considered correctly represented by simulation setup (Figure-12). Assessment of passive TJI technology on a mild hybrid powertrain and its performance 165 165 <?page no="166"?> Figure 12: 2500 rpm/ 8 bar 20-% EGR: Numerical setup is able to capture the combustion trends observed by experiment and correctly predicts the trapped in-cylinder mass. In this work, all PCSP designs are tested on the same engine configuration, for which the P, T trace in the cylinder is identical. Study of PCSP mixture homogeneity revealed minimal standard deviation value for both residual gas and equivalence ratio at ignition timing (therefore not presented). Thus, this study is based on TKE fields - quantity monitoring flame surface density in ECFM model used here - to assess the propensity of local conditions to promote faster turbulent flame propagation, ultimately affecting PCSP performance. 166 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 166 <?page no="167"?> 5 Prechamber performance across engine map The engine performance was tested on selected representative operating points to cover both the full load (1500 and 3500 rpm) and the EGR diluted part load operation (2500 rpm / 8-bar BMEP and/ or 3000-rpm / 6-bar BMEP). The stability criteria for partial load operating points were at COV IMEP = 3 %. During this study, the knock criteria used was KPPK3 (bar) - 3 rd percentile of ordered cycles according to KPPK metric (maximum amplitude of filtered pressure trace for each engine cycle). Due to the research nature of the activity, the knock limits for PCSP were established at the point of changing slope due to the varying base level of combustion noise according to each PSCP design tested. The final knock criteria for the production engine with pre-chamber will have to be defined in the future. The knock criteria for the SSP were taken according to the in-house standards. Test results were evaluated based on the following: Full Load: Main goal is BMEP (bar) and BSFC (g/ kWh) improvement under knock criteria (KPPK3 in bar): • 1500rpm low end torque area • 3500rpm max torque area Partial Load: Main goal is BSFC (g/ kWh) improvement with EGR dilution under COV (%) as combustion stability criteria: • 3000 rpm / 6 bar BMEP near peak efficiency region on rpm x BMEP engine map and/ or • 2500 rpm / 8 bar BMEP and constant EGR rate 5.1 Initial investigation on CR14 The original design of the engine was CR14 configuration, with initial tests of several PCSP designs (PCSP0, PCSP1, PCSP1a) in comparison with SSP. The differences between the pre-chambers are in number of orifices, their distribution and diameter. Internal design of the pre-chambers is kept constant for all three variants. Figure 13: Summary of the PCSP used for the firsts test on CR14 engine. All designs share the internal design and differ in the hole layout. Assessment of passive TJI technology on a mild hybrid powertrain and its performance 167 167 <?page no="168"?> 5.1.1 Full load results Reference absolute values for SSP are in dashed black. Each PCSP was added in absolute or relative (BSFC, BMEP). In full load condition, the parameter used to select each radar axis variation is the Knock criteria (KPPK3% level criteria) that is taken identical for each technical definition. Knock criteria is the main limiting parameter to ensure engine reliability. 5.1.1.1 Test bench results at 1500-rpm - Figure 14: Radar graph presenting comparison between SSP and PCSP candidates (CR14). PCSP1 shows improvement of BMEP by ~1% compared to SSP and earlier MFB50 by ~2 CAD. Full load at 1500 rpm, 0-%-EGR, engine Twater and Toil-= 90degC, corrected to Patm =1013 mbar and Tair = 20 degC For the 1500 rpm FL point, the best performing candidate was PCSP1, showing 1 % BMEP improvement compared to SSP and respecting the KPPK3 criteria. PCSP1 advanced the MFB50 by 2 CAD and provided improved combustion stability. Test bench results at 3500-rpm - Figure 15: Radar graph presenting comparison between SSP and PCSP candidates (CR14). PCSP1 shows improvement of BMEP and BSFC up to 1-% compared to SSP and earlier MFB50 by 2 CAD. Full load at 3500 rpm, 0 % EGR, engine T water and T oil -= 90degC, corrected to P atm =1013 mbar and T air = 20 degC The best pre-chamber candidate was PCSP1, showing up to BSFC and BMEP improvement compared to SSP and respecting the KPPK3 criteria. PCSP1 also advanced the MFB50 by almost 2 CAD and provided improved combustion stability (COV-%). 168 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 168 <?page no="169"?> 5.1.2 Partial load results In partial load condition was the combustion stability. - Figure 16: Radar graph presenting comparison between SSP and PCSP candidates (CR14). PCSP1 shows almost equivalent BSFC compared to SSP and shows the same MFB50 as SSP Partial load 2500-rpm/ 8-bar, EGR sweep, engine Twater and Toil-= 90degC For this partial load setpoint the best pre-chamber version is again PCSP1. Respects the COV criteria, having a BSFC (g/ kWh) almost equivalent to the SSP results with more EGR rate (+2-%) 5.1.3 Conclusion for CR14 Among the three PCSP designs considered, PCSP1 was the best candidate. In conclusion this design presents a gain in BMEP of approximately 1 % at full load compared to SSP, and stays without gain in BSFC at partial load operating points. 5.2 Investigation at CR15 For CR15 four new versions of pre-chamber were tested, with PCSP1 as reference from previous CR14 campaign, then PCSP2 to PCSP5 and reference SSP. The PCSP2 is an evolutionary step from PCSP1 in the optimization of the internal design - the cap definition stays the same. The design of PCSP3-5 with the increased A/ V ratio favorizes flame survival during the jet exit timing. Such design approach should be helpful especially during the cold start / catalyst heating operation, while at the same time performing in a satisfactory way across engine map. Figure 17: Main characteristics of PCSP for CR15 tests. Assessment of passive TJI technology on a mild hybrid powertrain and its performance 169 169 <?page no="170"?> Results are shown in two forms. Firstly, a more detailed analysis in the AVL CONCERTO software is presented to observe evolution of the most important variables (depending on operating point: BMEP (bar), BSFC (g/ kWh), MFB50 (CAD), KPPK3 (bar), and others). Secondly, radar graphs summarize the results per each operating point as for the CR14 investigation. 5.2.1 Full load 5.2.1.1 Test bench results at 1500-rpm Figure 18: Impact of combustion phasing on BSFC, KPPK3 and combustion for PCSP and SSP. Faster combustion and flame initialization for PCSP system shifts the knock limit which allows a better positioning of MFB50 with positive impact on BSFC. CR15, 1500-rpm at full load, 0-% EGR, SA sweep, T water and T oil = 90degC. BMEP corrected to P atm =1013 mbar and T air = 20 degC. Red cursor visualizes the reference values for the SSP. Compared to SSP all PCSP1-5 candidates offer faster flame initialization (SA→MFB10) and combustion duration (MFB10→MFB90) with improved stability level (COV). The faster combustion process of PCSP1-5 has a positive impact on knock behavior and allows a better positioning of combustion center (MFB50) which improves BSFC values with pre-chambers. The more efficient combustion process results in an increased BMEP level for pre-chamber application. 170 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 170 <?page no="171"?> Synthesis results Figure 19: Radar graph presenting comparison between SSP and PCSP candidates (CR15). PCSP4 shows ~4% gain in BMEP, MFB50 earlier by ~4CAD with faster flame initialization (SA→MFB10) and fast combustion process (MFB10→90). BMEP corrected to P atm =1013 mbar and T air = 20 degC. CR15, 1500-rpm at full load, 0-% EGR, SA sweep, T water and T oil = 90degC. Using cross-comparison on BSFC vs BMEP, the most efficient settings best candidate is PCSP4 offering +4-% BMEP and better flame initialization (delay SAà MFB10) than SSP. 5.2.1.2 Test bench results at 3500-rpm Figure 20: Impact of combustion phasing on BSFC, KPPK3 and combustion for PCSP and SSP. Faster combustion on PCSP shifts knock limitation allowing earlier MFB50 positioning with positive impact on BSFC. BMEP corrected to P atm =1013 mbar and T air = 20 degC. CR15, 1500-rpm at full load, 0-% EGR, SA sweep, T water and T oil = 90degC Red cursor visualizes the reference values for the SSP. Assessment of passive TJI technology on a mild hybrid powertrain and its performance 171 171 <?page no="172"?> Trends observed at previous 1500 rpm are equivalent at higher 3500 rpm with faster combustion process achieved by PCSP1-5 when compared to SSP counterpart. Synthesis results - Figure 21: Radar graph presenting comparison between SSP and PCSP candidates (CR15). PCSP1 and PCSP2 gain over 2-% BMEP and BSFC thanks to more centered combustion (MFB50). BMEP corrected to P atm =1013 mbar and T air = 20 degC. CR15, 1500 rpm at full load, 0 % EGR, SA sweep, T water and T oil = 90degC. For the comparison of the BMEP, PCSP1 and PCSP2 are the best candidates showing improvements of more than 2 %. Additionally, PCSP1 and PCSP2 offer better combustion centered behavior than SSP and others PCSP. 172 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 172 <?page no="173"?> 5.2.2 Partial load results 5.2.2.1 EGR sweep Figure 22: Impact of EGR dilution on combustion and stability for TJI and SSP. Dilution and lower load influence the filling of pre-chamber with negative effects on COV and BSFC. Design of PCSP5 and PCSP2 brings them close to SSP level of performance but with a limited EGR tolerance (PCSP limit at max 20 % dilution). Partial load 3000 rpm/ 6 bar BMEP, EGR sweep, T water and T oil = 90degC. Red cursor visualizes the reference values for the SSP. The tolerance of EGR dilution of all PCSP designs is lower than SSP. PCSP2 and PCSP5 are on the same level of fuel consumption during the EGR rate variation as the SSP. Compared to other candidates PCSP5 has more time on flame initialization (delay SA → MFB10) but with faster combustion process (delay MFB10→MFB90). Assessment of passive TJI technology on a mild hybrid powertrain and its performance 173 173 <?page no="174"?> Synthesis results Figure 23: Radar graph presenting comparison between SSP and PCSP candidates (CR15). PCSP2 and PCSP5 show 5-% less EGR tolerance than SSP, but similar BSFC. Despite PCSP5 longer SA→MFB10 delay, the MC combustion is relatively fast. PCSP2 shows shortened SA→MFB10 and MC combustion by 2 CAD. Partial load 3000-rpm/ 6-bar BMEP, EGR sweep, T water and T oil = 90degC. All PCSP designs show poorer stability as the EGR rate increases, with only two designs of PCSP (PCSP2 and PCSP5) reaching 20 % of EGR. Best candidate compared to SSP is PCSP2, with similar BSFC (+0.3%) at 5-% less total EGR than SSP. 174 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 174 <?page no="175"?> 5.2.2.2 Spark advance sweep at constant EGR rate Figure 24: PCSP geometry characteristics results at part load, constant EGR operation. Difficulties in filling the PCSP at part load has a negative impact on performance results. Designs changes of PCSP5 and PCSP2 offer marginal improvements which brings PCSP closer to SSP in terms of BSFC. Partial load 2500 rpm/ 8 bar BMEP, EGR rate at 17 %, SA sweep, T water and T oil = 90degC. Red cursor visualizes the reference values for the SSP. Assessment of passive TJI technology on a mild hybrid powertrain and its performance 175 175 <?page no="176"?> Synthesis results Figure 25: Radar graph presenting comparison between SSP and PCSP candidates (CR15). PCSP2 and PCSP5 similar in BSFC to SSP, PCSP2 with slightly faster combustion (MFB10→90) and improved MFB50 but with no gains in BSFC compared to SSP. Partial load 2500-rpm/ 8-bar BMEP, EGR rate at 17-%, SA sweep, T water and T oil = 90degC. Conclusion: The best candidate is PCSP2 due to faster combustion initilisation (delay SAà MFB10: -2.6 CAD vs SSP), combustion speed improvement (MFB10→90 reduced by 2 CAD compared to SSP) and MFB50 (-1 CAD vs. SSP). It nevertheless achieves the same BSFC as the SSP reference. 5.2.3 Conclusion for CR15 testing Overall, PCSP designs that offer advantage in full load (BMEP) have the same or increased BSFC in partial load. All PCSP designs tend to improve MC combustion duration in full load and COV across the studied operating points. The overall gains in full load are up to 3.5 % compared to SSP and depend on RPM. For the partial load, the best BSFC values are equal to the SSP performance. 176 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 176 <?page no="177"?> Figure 26: Overview of PCSP system compared to SSP at full load for CR15. There is up to 3.5-% gain on 1500-rpm (PCSP4) and up to 2.6-% gain on 3500-rpm (PCSP2) Figure 27: Overview of PCSP system compared to SSP at partial load for CR15. PCSP designs offer similar BSFC levels at part load (PL) with no improvements, the PCSP2 being the closest to the SSP performance. Assessment of passive TJI technology on a mild hybrid powertrain and its performance 177 177 <?page no="178"?> Figure 28: Comparison of combustion duration for PCSP designs vs SSP. In full load (FL) without EGR rate, every PCSP offer a better combustion speed, whereas in partial load (PL) at same EGR rate the performance of PCSP designs is similar to SSP. Figure 29: Combustion stability is improved in majority of cases with PCSP (at PL as in FL). 5.3 Investigation of the engine instability at EGR rate limit High EGR tolerance is the lever to reduce BSFC and, in the context of PCSP ignition, it depends on: (i) the level of residuals inside the pre-chamber, (ii) pre-chamber internal P, T conditions at the moment of ignition, (iii) local flame extinction phenomena due to conditions or design of the PCSP, (iv) MC conditions at jet exit timing. 178 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 178 <?page no="179"?> Figure 30: PCSP1 shows lowered tolerance to early SA at partial load (3000-rpm/ 6-bar). PCSP5 loses stability 10 deg of SA later than PCSP1, allowing to further increase the EGR rate by 3 %. Comparison at optimal MFB50, vertical lines mark the instability onset (COV limit at 3-%). The engine testing showed that the maximum achievable EGR rate at partial load (such as 3000 rpm/ 6 bar) depends on the PCSP design due to the differences in the aforementioned conditions i-iii. Since higher EGR rate increases PCSP ignition delay, a more advanced ignition timing is required to keep the phasing of MC combustion at optimum. The consequence of that is the ignition occurs right before or at the highest rate of gas flow from MC to PCSP (at 20 to 25 CAD bTDC), which tends to be the limit of stability for PCSP1-4 studied. The PCSP5 (different internal design compared to PCSP1-4) nevertheless showed such limitations much later and allowed for stable combustion at earlier SA of 30 to 35 CAD bTDC (Figure-30). A deeper look was therefore required to investigate the root causes of the inferior tolerance to early spark advance of the PCSP1-4 designs. If a more optimized pre-chamber design was capable of achieving earlier ignition timing while keeping a reasonable MC ignition delay, a higher EGR dilution would be possible, with the potential to further decrease the BSFC. 5.3.1 Investigation optical access to both chambers - PCSP 1-4 The objective of the optical investigation was to identify the source of the engine instability of the PCSP1-4 in its corresponding location or locations, understand its mechanism and potentially apply measures in the pre-chamber design to help the system reach earlier SA. For the sake of the analysis, the combustion chamber is divided in 4 locations: 1. The topmost part of the pre-chamber, where the field of view (FOV) of the PCSP optical access is located. 2. The pre-chamber body. 3. The PCSP cap and orifices. 4. The main chamber, where the camera of the MC is located. Assessment of passive TJI technology on a mild hybrid powertrain and its performance 179 179 <?page no="180"?> The flame is created in location 1, it propagates through 2, it reaches the cap and goes through the holes 3, and it reaches the main chamber 4 as hot jets. There is visual access to 1 and 4 only (Figure-31). Figure 31: Combustion chamber locations: [1-3]=PCSP, [4]=MC. There is visual access to 1 with FOV and to 4 with a camera. Optical fiber investigation suggests that despite an early onset of the engine instability for the PCSP1-4 and non-negligible proportion of cycles showing an almost complete MC misfire, the flame in the vicinity of the PCSP electrodes is developed in a way that is almost indistinguishable from regularly firing cycles. And despite the visibly regular PCSP combustion initiation at location 1 (Figure 32), the MC combustion is on average deteriorated (partial burn or misfire). Figure 32: 3000 rpm/ 6 bar instable operation with SA > 30 CAD. Left: Misfired cycles (< 0.5 bar IMEP n ) have similar PC light intensity as regularly firing cycles - Right: Averaged cycle light intensity during the PCSP combustion is stable and depends only a little on the produced IMEP n 180 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 180 <?page no="181"?> In addition, the combustion in the main chamber (location 4) was investigated by means of the MC camera. Observing a sample of worst performing cycles with IMEP n < 0.5 bar, no light was visible at the moment of average jet exit timing (Figure-33). Figure 33: 3000 rpm/ 6 bar instable operation with SA > 30 CAD. Left: Misfired cycle (< 0.5-bar IMEP n ) have no light perceived by the camera - Right: Fired cycle light. Both pictures were taken in the range of expected CA05: [-30,10] deg and a exposure of 1.8 deg. It was therefore concluded that the root cause for the PCSP misfire may come rather from the location 2 or 3 (see Figure 34), that is from the intermediate or lower part of pre-chamber volume or from the orifices. Since it was not possible to continue the investigation by means of testing, a CFD study was performed. Figure 34: 3000 rpm/ 6 bar with early SA / high EGR limit: PCSP is always firing, MC misfiring cycles (<0.5-bar-IMEP n ) show almost same PCSP combustion intensity as regularly firing cycles, with no jets in MC observed. Assessment of passive TJI technology on a mild hybrid powertrain and its performance 181 181 <?page no="182"?> 5.3.2 CFD investigation of the engine instability for PCSP1-4 The CFD investigation confirms that the internal design of the pre-chamber determines whether the dynamic fluid structures inside the PCSP shall promote the flame development or retard it. It could even result in inhibiting the PCSP jets from igniting the MC mixture. Study of velocity field distribution evolution suggests that a high velocity region in the lower part of the PCSP1-4 designs is created by a narrower flow section. This permits the high velocity to reach the vicinity of the spark plug promoting TKE. Nevertheless, straight upwards motion also ultimately leads to convection of the flame kernel towards more confined and less turbulent upper region at the top of the PCSP, consequently penalizing jet exit timing. This tendency is further amplified by the specific design of PCSP3, with a more permeable cap design, that does not create the tumble pattern observed in PCSP1 (Figure-35). Figure 35: 2500 rpm/ 8 bar EGR 20-% (IT=342 CAD). High velocity in the narrow bottom region of the PCSP1 and 3, with PCSP3 flow pattern towards the top (left) - slowing down the flame propagation. TKE decrease for PCSP3 compared to PCSP1 further worsening the conditions for flame development (right) Comparison of burnt gas mass fraction distribution after the ignition (342 CAD) around the sparkplug for PCSP1 and PCSP3 suggests faster flame kernel development for the first one, which is supported by HRR traces in Figure 36.This trend is linked to the observations made on velocity and TKE distribution: (i) onset of turbulent flame propagation is penalized for PCSP3 due to overall lower TKE level; (ii) upwards motion velocity more pronounced for PCSP3 tends to convect the flame kernel more into the upper - less turbulent - region of the PCSP. Figure 36: 2500 rpm/ 8 bar EGR 20-% (IT=342 CAD). Burnt gas mass fraction fields representing the flame front. Turbulent flame kernel appears (i) significantly smaller for PCSP3; (ii) more centered in the higher region of the PCSP, resulting in further delay of jet exit timing. 182 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 182 <?page no="183"?> These findings are coherent with the test results for 3000 rpm/ 6 bar EGR diluted operating point. PCSP1 shows shorter ignition delay than its versions with a larger maximum orifice diameter: PCSP3 (+50% in D hole,max ) and PCSP4 (+100% in D hole, max ) - visible in Figure 37. At the highest dilution levels (close to 20 %) PCSP1 ignition delay is 2~4 CA shorter than for PCSP3 and PCSP4. Therefore, PCSP3 and 4 show decreased maximum EGR rate tolerance due to earlier onset of instability at SA limit. Figure 37: 3000 rpm/ 6 bar EGR rate sweep - At the highest dilution levels (close to 20 %) PCSP1 ignition delay is 2~4 CA shorter than with PCSP3 and PCSP4 Overall, the PCSP1-4 show reduced tolerance to early IT as compared to PCSP5 and this can be directly attributed to the PCSP1-4 internal design with narrower flow section. Compared to PCSP1, the PCSP3-4 further amplified this tendency by having larger maximum orifice diameter. By combined analysis of the test bench results and the CFD study, it seems that exposing the developing flame kernel to a directly opposing flow should be avoided: instead, a more favorable and differently ordered fluid pattern should be created during engine compression stroke, by means of redefined PCSP internal design and orientation/ targeting of the PCSP orifices. These differently guided flow patterns should be established with the objective to (i) increase the resistance to early SA (similarly to PCSP5); (ii) to reduce or at least keep the same delay of MC combustion initiation SA→CA10 (like for PCSP1-4). 6 Prechamber performance at cold conditions A prerequisite for any commercialization of a combustion system is its compliance with emission regulations of the pollutants. Engines ignited by SSP use a late MC combustion to increase the exhaust gas temperature and thus heat up the catalyst beyond the light-off temperature. This allows to reduce the overall cumulative pollutant emissions. As already indicated in previous publications, pre-chamber concepts can pose signif‐ icant challenges at cold engine start-up and operation, mainly due to a) the difficulty to Assessment of passive TJI technology on a mild hybrid powertrain and its performance 183 183 <?page no="184"?> keep the engine at idling conditions and b) the difficulty to retard the MC combustion sufficiently. The latter is a key point to reduce the HC/ CO emissions and at the same time generate higher heat flux in the exhaust gases while keeping an acceptable engine stability. Since the engine is dedicated to hybrid powertrains, a new range of possible HEV catalyst warm-up strategies are feasible. It is no longer necessary to stay at idling conditions or be dependent on drivers’ request - the hybrid powertrain e-machine can be used to operate the engine at optimal conditions (load, RPM) for a certain period of time. This can be translated to facilitating the start-up or allowing the combustion system to work during the transient operation, at optimal trade-off between the emissions of pollutants and heat flux generation for the catalyst heat-up phase. 6.1 Startability at -20degC One of the typical constraints for engine operation in cold conditions is startability at very cold conditions. The user should not wait for too long before the first engine combustions occur and at the same time the misfire should not be present to achieve smooth engine operation. For this investigation the engine was conditioned at temperatures below zero (T intake = -15degC, T water- = -20degC ). The engine was then started with a partially optimized oper‐ ating strategy using the above-mentioned methodology for hybrid applications (without focus on retarding the MC combustion), with its speed set to 1000-rpm. Results in Figure 38 show that despite the use of an unfavorable PCSP1 design for such application (rather small pre-chamber orifice diameter of D < 1 mm that tends to quench the jets), the engine operation is relatively satisfactory with only one misfire during the first critical seconds. A comparison can be made to a more adapted pre-chamber design (PCSP1a) with maximum PCSP orifice diameter being 50 % larger than for the unfavorable case. A considerable difference was observed in the delay between the start of injection and the first firing cycles (0.6 seconds between PCSP1 and PCSP1a), favoring the latter design. This can be mainly explained by the different λ MC settings and the pre-chamber sensitivity to it. Investigation showed that enrichment of the MC charge helps to shorten the delay between first injection and the first firing cycles. At the same time, this effect cannot probably be attributed to the increased maximum orifice diameter size in the case of PCSP1a, since its performance in catalyst heating potential (stabilized engine operation to investigation the potential for MC combustion retard) stays very similar to the one of PCSP1 (not presented). 184 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 184 <?page no="185"?> Figure 38: PCSP1 startability at -20C is satisfactory with 1 misfire. Initial cranking delay between start of injection and first combustion can be further improved by increased enrichment during the first cycles (PCSP1a). λ MC trace is delayed in comparison with the IMEP n . Considering the difference in λ MC tuning, it can be concluded that the unfavorable (for cold start) PCSP1 design is capable of starting the engine in an almost satisfactory manner (one complete misfire), with further potential for optimization of the engine calibration (λ MC , engine load and RPM). 6.2 Tolerance to late combustion The objective of this steady-state operating point (1300 RPM / stable P intake ) was to effi‐ ciently evaluate the performance of PCSP designs at their capability of retarding the MC combustion (thus increasing the heat flux towards the catalytic converter) while keeping the engine stability satisfactory with no misfire. The engine temperature was kept at partially cold conditions to the limit of the test bench cooling capability (T intake = -5degC, T water = +5degC ). Tests were performed with an optimized MC fuel injection strategy with same pattern that allowed a partially optimized representative engine cold start with catalyst heating phase. Assessment of passive TJI technology on a mild hybrid powertrain and its performance 185 185 <?page no="186"?> Figure 39: Best candidate of the hot testing (PCSP2) shows poor tolerance to late MC combustion with the early onset of misfire. PCSP3-5, with higher A/ V ratio and larger maximum PCSP hole size, increased the late MFB50 by 20 CAD. 1300rpm / stable P intake , T intake = -5degC, T water = +5degC Comparison of several pre-chamber designs shows the latest achievable MC center of combustion (MFB50), with PCSP3-5 as most suitable designs for cold conditions due to their increased A/ V ratio and largest PCSP orifice being in the range of 1.5x-2.2x of the largest orifice of the PCSP2. As seen from the misfire overview (Figure 39), for all pre-chambers the non-zero misfire appeared early as compared to expected SSP performance (> 80 CAD). Figure 40: PCSP3-5 allow to reduce THC emissions by approximately 40-% compared to PCSP2 due to combined effect of improved combustion efficiency at iso MC phasing (1) and overall later phasing of MC combustion due to later onset of instability (2). Both effects are correlating with the increased A/ V ratio and larger maximum hole diameter. 186 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 186 <?page no="187"?> The comparison of emissions showed that PCSP3-5 were the best candidates due to of improved combustion efficiency at iso MC phasing and overall later phasing of MC combustion due to later onset of instability. These combined effects allowed to decrease the THC emissions by 40-% from the PCSP2 stability limit (Figure-40). 6.2.1 Investigation of the source of engine instability Same methodology and reasoning as for the partial load (Sections 5.3 and 5.3.1) was applied. The test bench acquisition allowed for investigating the locations with optical accesses: location 1 and location 4 (PCSP optical fibers and MC camera, respectively) and to correlate their results to the standard pressure indication result (IMEP n ). The methodology consisted in analyzing unstable late SA acquisitions, with misfire rates of around 10-%. Figure 41: PCSP3 and PCSP2 light intensity differs between firing and misfiring cycles (< 0.5 bar IMEP n ), averaged sample of 3000 cycles (left). Average PCSP intensity per cycle is weakly correlated to IMEP n (right). In other words, the combustion in the pre-chamber was always present, even when the main chamber misfired (IMEP n < 0.5 bar). Nevertheless, the comparison between the PCSP intensities recorded and the MC combustion shows that there is a weak link between the intensity of the PCSP combustion and the MC misfire: the PCSP combustion is statistically less light intensive than for the regularly firing cycles (see PCSP combustion in Figure 41). Assessment of passive TJI technology on a mild hybrid powertrain and its performance 187 187 <?page no="188"?> Figure 42: Catalyst heating potential: PCSP is always burning, even during complete MC misfires. Link between MC misfire and PCSP combustion (locations 1 and 4) is weak and depends on the internal pre-chamber design. At the same time, investigation of the MC camera images showed that no light was visible during the misfiring cycles, which suggests that the flame is quenched on the way from location 1 to location 4. Figure 43: Visio data from a test with PCSP2 and SA -15deg. Misfiring cycle: weak intensities observed in the PCSP and no light in the MC (Top). Firing cycle (Bottom). In catalyst heating, the misfiring cycles tend to have a weaker combustion in the PCSP. Camera pictures taken with 1.6 CAD of exposure time. 188 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 188 <?page no="189"?> Since a weak but existent correlation between the measured PCSP light intensity and the IMEP n was observed, a countermeasure such as differently structured PCSP internal flow pattern in combination with internal PCSP design changes could potentially reduce cycle-to-cycle variation in the PCSP combustion and help to increase the tolerance to late MFB50 by reducing the propensity to MC misfire. Another measure to improve the combustion repeatability would be to increase the TKE at later ignition timings by optimizing and adapting the inflow pattern. 6.3 Emissions performance at cold start conditions The tendencies observed on the stabilized catalyst heating investigation were confirmed for the cold start. Increasing the A/ V ratio and the maximum orifice diameter enabled later MFB50, thus increasing heat-flux to the exhaust, leading to the catalyst light-off temperature (~300 degC) being reached more quickly (PCSP3 and PCSP5, compared to PCSP2, see Figure-44). Figure 44: Cold start test at -10degC: PCSP5 and PCSP3 allow to retard the ignition more than PCSP2, thus reaching the catalyst light-off temperature faster. However, SSP can retard MFB50 by 50 CA more than these two (left). All prechambers produce higher THC emission overall, but especially during the first critical ~15s-20s. CO quickly decreases to comparable level with SSP. NO x stays higher due to quicker MC combustion. The best candidate at full and partial load PCSP2 shows inferior results in emissions (right). Specific HEV warm-up strategy applied. Nevertheless, the challenges of the PCSP designs are two-fold. Firstly, the time to heat-up to catalyst is still increased compared to SSP due to the inability to retard further the MC combustion. Secondly, the emissions of pollutants (mainly THC) are not only higher throughout the first critical 15s-20s, but also afterwards, where MC combustion phasing is closer for SSP and PCSP variants. This confirms the tendency of the pre-chambers to Assessment of passive TJI technology on a mild hybrid powertrain and its performance 189 189 <?page no="190"?> generate on average higher mount of unburnt hydrocarbons compared to standard spark plug at the same MC combustion phasing. Optical investigations during the cold start resulted in a new finding: it was for the first time that the camera detected light during a complete misfire in the main chamber. As visible in the Figure 45, the misfire occurs due to partial flame quenching in the holes and/ or the main chamber. - Figure 45: PCSP2 during cold start at -10degC. Camera pictures of two early cycles during the cold start: (Left) Normal combustion and (right) misfire with visible light emission in the prechamber holes suggesting quenching in the orifices or main chamber. Exposure of 5 ms between -5 CAD aTDC and 34 CAD aTDC. The cold start tests showed that further optimization is required: both in the engine calibration (rpm/ load) and in the pre-chamber design with the objective to decrease the level of the THC. 7 Optimization of pre-chamber design Two main approaches to PCSP design were tested and numerically simulated. The PCSP1-4 and PCSP5 are distinct pre-chamber designs with two different philosophies of the internal conception that drive the creation of the flow pattern inside the PCSP. The importance of the internal flow structures lies in its direct influence of the PCSP combustion and consequently the PCSP-MC interaction both at EGR limits of part load operation and cold start (heat dissipation in the PCSP body and in the orifices). The PCSP1-4 designs faced the challenge of early onset of instability during the part load EGR engine operation, with the advantage of overall faster SA-MFB10 delay than PCSP5. The PCSP3-5 designs faced the challenge of high A/ V ratio sometimes prohibiting the shortening of MC combustion duration to the same extent as PCSP2 on the full load operating points. Nevertheless, for this particular design, a higher A/ V ratio was required to increase the permeability of the pre-chamber to allow scavenging of residuals gases during the cylinder intake stroke. All designs confirmed that increasing the A/ V ratio and maximum PCSP orifice diameter is beneficial in improving the tolerance to late MC combustion, directly improving the cold start emission performance. 190 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 190 <?page no="191"?> It was concluded that the PCSP1-5 designs did not present the full potential of the TJI combustion and that there was a space for an improved trade-off between the increase of power at full load, BSFC decrease at EGR diluted part-load and the emission performance at cold start. The prior learnings lead to develop of a new generation of pre-chamber spark plugs, based on following objectives: (i) Creation of a stable internal flow pattern during the compression stroke, which despite the high rate of gas inflow into the PCSP keeps part of the gases in downward motion (Figure 46), facilitating the kernel development after ignition and flame propagation towards the orifices. Optimization of the momentum of the macroscopic gas movement is needed to find the optimal design to satisfy all conditions. Figure 46: Comparison of the PCSP1 and 3 flow structures (left), with PCSP designs that generate a negative velocity column along the centerline of the PCSP (right) (ii) Such bulk gas motion should furthermore develop higher TKE in the vicinity of the PCSP electrodes in expected ignition timing range (Figure 47) in order to accelerate the development of PCSP combustion. - Figure 47: development of higher TKE in the vicinity of the PCSP electrodes during IT range to accelerate the development of PCSP combustion (left). PCSP design features minimize the negative effect of the residual gas from previous cycle on the early stages of kernel development. Level of residuals at IT range is close to MC dilution. Compared to scavenged variant (scavenging during intake phase), the PCSP can be optimized for jet development and initiation of MC combustion (right). Assessment of passive TJI technology on a mild hybrid powertrain and its performance 191 191 <?page no="192"?> (iii) Application of PCSP design features to minimize the negative effect of the residual gases from previous cycle on the early stages of kernel development. Such a strategy minimizes the design constrains on PCSP scavenging properties during intake phase (Figure 47). The orifice layout can then be optimized for hot jet development and initiation of MC combustion. (iv) PCSP combustion duration and therefore jet exit delay (Figure 48) can be further improved by the reduction of the distance from PCSP electrodes to orifices. Further‐ more, by shortening the length of the PCSP orifices, the heat transfer losses during the passage of hot gases can be decreased. Figure 48: Effect of faster heat release rate for the PCSP gen2a example on reduced jet exit timing, compared to the PCSP1. Reducing the jet exit delay is the key leverage to further increase the EGR tolerance. (v) Focus was put on staying around an optimized A/ V ratio while reducing the number of orifices. Such design approach allowed to increase the size of the holes with the benefit of increasing the jet momentum (considered beneficial for MC penetration) and decreasing the propensity for heat transfer losses (improved surface to volume ratio of the orifices). (vi) Enabling to further retard the MC combustion during the catalyst warm-up by increasing the PCSP TKE level after TDC. New designs are capable of sustaining up to 2x the TKE level compared to PCSP1-4 (Figure 49). This suggests better ignitability properties of the PCSP charge at the beginning of the expansion stroke, potentially improving the repeatability of the PCSP combustion. Figure 49: Catalyst heating increased TKE level for the PCSP gen2(a), suggesting potentially improved ignitability at later stages of the expansion stroke 192 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 192 <?page no="193"?> 8 Summary and outlook Out of the candidates studied, the PCSP with the best trade-off between the full load and part load was the PCSP2. The summary will mainly focus on the performance of this pre-chamber. Full and part load CR15 On the presented (1500 and 3500 rpm) full load operating points at CR15, the knock mitigation potential of the PCSP2 increased the maximum engine power up to 3 % compared to SSP, depending on rpm and the exact knock limit used. At the same time, it offered the best performance from the studied PCSP designs on partial load EGR diluted operating points for CR15 (2500 rpm/ 8 bar and 3000 rpm/ 6 bar), showing same BSFC level as SSP (to the limit of the accuracy of the measurement) despite lowered EGR tolerance by up to 5 p.p. at 3000-rpm/ 6-bar. Increased CR effect The increased compression ratio by 1 point (CR14 to CR15) limited the performance of the full load operating points for SSP. Using the PCSP2 allowed to reduce the brake power losses at 1500 rpm (from -4.5 % for SSP CR15 to final -2 % of maximum brake power compared to SSP CR14) and even further advance the MC combustion as the rpm of the engine increased, which helped improving the maximum brake power at 3500-rpm by 1-% (Figure-50). Figure 50: PCSP2 full load performance on 1500 and 3500-rpm in relation to other PCSP candidates (left). Evaluated PCSP2 performance at CR15 compared to SSP baseline at CR14 - increasing the RPM of the engine enables to mitigate the knock further and increase the BMEP gains (right). Figure 51: PCSP2 with CR15 offers a BSFC reduction of 1~1.5% compared to SSP with CR14. Assessment of passive TJI technology on a mild hybrid powertrain and its performance 193 193 <?page no="194"?> The BSFC of partial load operating points (2500 rpm/ 8 bar and 3000 rpm/ 6 bar) was decreased by 1.5 and 1-%, respectively, when SSP CR14 and PCSP2 are compared. It is concluded that the use of higher compression ratio in combination with the PCSP2 design proved to be advantageous, as it allowed to decrease the BSFC of the engine by 1-1.5-% on studied partial load points while keeping the advantage of knock mitigation on full load, with the exception of the low-end torque point at 1500-rpm. PCSP performance at cold conditions As expected, PSCP2 offering the best performance overall on full and partial operating points showed limited potential for MC combustion retard in cold conditions, with latest MFB50 being around 30 CAD aTDC, falling short of the standard spark plug capabilities in HEV catalyst warm-up (> 80 CAD aTDC). Other PCSP designs were able to partially improve the results, nevertheless presented reduced break thermal efficiency on full and partial load operating points compared to the best PCSP design. As a resulting challenge remains the increased THC emissions (at the current stage more than 2x higher than target). Further optimization of the HEV catalyst warm-up strategy settings is possible (both load and rpm), it has to be nonetheless combined with improvements in the PCSP design. The startability result at -20degC shows that even the non-optimized PCSP designs were able to start the engine with good performance when enrichment of the main chamber increased. PCSP optimization outlook The objective is to further enhance the PCSP design that would be capable of sustaining the full load performance, while increasing the break thermal efficiency at partial load by additional 1-2 % compared to CR14 standard spark plug reference and while improving the cold start THC emissions. In order to achieve the better trade-off, several concrete pre-chamber spark plug design measures were proposed. 9 References [1] IEA, “Global CO 2 emissions by sector,” IEA, Paris, 2019-2022. [2] ITF, “Transport outlook,” OECD/ ITF, Leipzig, 2023. 10 Acknowledgement We would like to thank Maxime TAROT, Miguel MONSERRAT DIAZ, Odysseas BAKAT‐ SELOS and Cristian-Marian MIHAI for their work, support, collaboration and contribution to this publication. 194 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 194 <?page no="195"?> 11 Appendix 11.1 Abbreviations and symbols A/ F Air to Fuel ratio A/ V Total cross-section of all pre-chamber orifices divided by the pre-chamber volume BDC Bottom Dead Center BSFC Brake Specific Fuel Consumption BTE Brake Thermal Efficiency CAD Crank Angle Degree COV Coefficient of Variation ECFM Extended Coherent Flamelet Model EGR Exhaust Gas Recirculation EOI End of Injection FL Full Load FOV Field of View HEV Hybrid Electric Vehicle IMEP Indicated Mean Effective Pressure IT Ignition Timing ITE Indicated Thermal Efficiency ISFC Indicated Specific Fuel Consumption λ lambda (actual/ stoichiometric air-fuel ratio) MC Main Chamber MCE Multi-Cylinder Engine MFBxx Mass Fraction Burnt PCSP Pre-Chamber Spark Plug PL Partial Load RANS Reynolds-averaged Navier-Stokes RON Research Octane Number SA Spark Advance (Positive: bTDC, negative: aTDC) SSP Standard Spark Plug TCI Turbulence Chemistry Interaction TDC Top Dead Center TKE Turbulent Kinetic Energy Assessment of passive TJI technology on a mild hybrid powertrain and its performance 195 195 <?page no="196"?> TJI Turbulent Jet Ignition V/ A Inversed value of A/ V 11.2 Prefixes and subscripts a after b before g gross n net 196 Dimitrios Karageorgiou, Thierry Prunier, Matej Myslivecek , Carmen Vesel 196 <?page no="197"?> 1 CNR STEMS---Istituto di Scienze e Tecnologie per l’Energia e la Mobilità Sostenibili Via Guglielmo Marconi, 4---80125 Napoli, Italy 2 CNR STEMS---Istituto di Scienze e Tecnologie per l’Energia e la Mobilità Sostenibili Via Guglielmo Marconi, 4---80125 Napoli, Italy 3 CNR STEMS---Istituto di Scienze e Tecnologie per l’Energia e la Mobilità Sostenibili Via Guglielmo Marconi, 4---80125 Napoli, Italy 4 CNR STEMS---Istituto di Scienze e Tecnologie per l’Energia e la Mobilità Sostenibili Via Guglielmo Marconi, 4---80125 Napoli, Italy 5 Department of Energy---Politecnico di Milano---Via Lambruschini, 4a--- 20156 Milano, Italy 6 Department of Energy---Politecnico di Milano---Via Lambruschini, 4a--- 20156 Milano, Italy 7 FPT Industrial Spa, Via Puglia, 15---10156, Torino, Italy Virtual analysis of the Efficiency Gain with Pre-Chamber combustion systems in Heavy-Duty Natural Gas Engine for Long-Haul Truck Application. Dario Di Maio 1 , Pierpaolo Napolitano 2 , Chiara Guido 3 , Carlo Beatrice 4 , Lorenzo Sforza 5 , Tommaso Lucchini 6 , Stefano Golini 7 Abstract Spark Ignition Heavy-Duty engines with active pre-chamber combustion systems are a potential and viable technology for significantly boosting of the brake thermal efficiency (BTE). A considerable interest in this technology is also present in view of the zero carbon emission goals and the use of future sustainable e-fuels for high-energy demand vehicles. This work addresses the investigation on a preliminary analysis of the BTE gain shifting from a conventional spark-ignition stoichiometric HD engine fuelled with natural gas to an active pre-chamber configuration. To this aim, through a parallel and coupled 1D/ 3D simulation methodology, two crucial engine operating points were simulated and analysed: the Cruise and Rated Torque points. An iterative simulation process was adopted to carry out reliable results assuming the conventional multi-cylinder HD engine as baseline. The boundary conditions for the 3D-RANS numerical activity are provided by a 1D model of a single cylinder engine, properly tuned according to experimental data detected testing a pre-chamber single-cylinder research Heavy-Duty Gas engine with similar features to the reference multi-cylinder. To examine the alignment between the two simulation platforms, the combustion process evolution, as well as the turbulent flame propagation profiles acquired from <?page no="198"?> 3D-CFD simulations, are iteratively applied to the 1D engine model to determine in-cylinder pressures and heat release rates. Subsequently, an engine parameter sensitivity analysis was carried out to proper evaluate the influence of each parameter on the BTE. Compression ratio, lambda, spark advance, and intake and exhaust pressure drop were the selected parameters for the sensitivity analysis. In particular, two lean conditions with lambda values of 1.7 and 1.9 were identified, which have shown a significant decrease in NOx emissions compared to the starting stoichiometric conditions, and moving relevant steps toward the introduction of EURO VII regulation. An appreciable increase in BTE of 1.9% was found in Cruise conditions. In Rated Torque the increment in BTE is even more noticeable, reaching a value of +3.5% over the nominal reference engine. 1. Introduction Reducing greenhouse gas (GHG) emissions from the transportation industry is a chal‐ lenging undertaking. The combined use of different technologies is the most reasonable and sustainable path for cutting road transport sector environmental impact. The constant improvement of internal combustion engines (ICEs) with the possible exploitation of biofuels and future hydrogen applications will be the reply to the challenge together with growing electrification. As reported by the European Environmental Agency [1], the european passenger sector is moving towards electrification with the 22 % of newly registered cars getting battery electric vehicles (BEVs) and plug-in hybrid electric vehicles (PHEVs) in 2022; nevertheless, they still represent only the 1.2% of the total EU cars fleet. In addition, adjustments to the power system to meet the increased demand for electricity due to a high spread of such vehicles is an important element, both from an engineering and political point of view, related to the supply of raw materials. Some works, in particular, show that the threshold of available vehicles without an increase in net carbon emissions is quite limited compared to the total number of circulating car fleet [2-4]. Moreover, BEVs have some problems related to charging, such as long times and low number of charging stations, which are difficult for the heavy duty (HD) sector to overcome [5]. Manufacturers of HD vehicles operating in the EU are required to reduce CO 2 emissions by 15 % in 2025 compared to 2019 levels and by 45 % in 2030. To accomplish this goal, the main and foremost way is to significantly raise the thermal efficiency of ICEs, while increasing the use of fuels with a more favourable H/ C ratio. Natural gas (mainly made of methane, CH 4 ) has the highest H/ C ratio and has been commonly used as a fuel in HD vehicles for the past thirty years; moreover, as reported in Figure 1, biomethane production, a viable alternative to fossil natural gas, has been constantly growing in Europe [6]. 198 Di Maio, Napolitano, Guido, Beatrice, Sforza, Lucchini, Golini 198 <?page no="199"?> FIGURE 1: Biomethane production in Europe. Timeframe: 2011-2022. Renewable methane is particularly appealing in the short term since it is the only advanced, low-carbon alternative fuel pathway that can be produced using fully mature and readily available first-generation technology. It has been demonstrated that this sustainable production of biomethane, as an example from anaerobic digestion of livestock manure and sewage sludge, can sometimes more than offset the avoided emissions from burning fossil fuels (Figure 2) [7]. In the longer term, low-carbon renewable methane can be produced from a wider range of feedstocks using second-generation technologies, including gasification and power-to-methane, in which electricity and CO 2 are used to create methane. The Joint Research Center ( JRC) shows that the municipal waste is the most used path to produce biomethane in Europe and highlights its very good results. A compressed biomethane (CBM) vehicle provides well-to-wheel GHG emissions reduction up to 88 % compared to Diesel fuel [8]. FIGURE 2: GHG emissions from sustainable renewable methane pathways. Virtual analysis of the Efficiency Gain with Pre-Chamber combustion systems 199 199 <?page no="200"?> Recently, by virtue of the absence of carbon, HD vehicle manufacturers shown an increasing interest in the use of hydrogen, an energy carrier which is expected to be used in several energy systems and that can be employed as a fuel for currently produced ICEs [9][11]. However, some modifications are required to the existing engine architectures; the main ones involve the injection system and the materials of several components to handle the peculiar characteristics of this gas. Apart from being burned, pure hydrogen can be blended with natural gas to obtain hydromethane. Although this fuel is not carbon-free, it can significantly reduce CO 2 emission if compared to a standard diesel engine. Moreover, hydromethane can be used on existing HD gas engines with minor adjustment of the engine calibration, if an hydrogen content up to 30-% v/ v is employed [12]. Looking at the thermal efficiency of the engine, lean combustion is the most efficient combustion mode, as demonstrated in diesel engines, and its advantages are [13]: • Lower temperature and decreased heat losses; • Complete combustion with negligible levels of CO at the exhaust; • Higher isentropic efficiency due to higher exponent in the polytropic compression; • When applied to SI engines, lean combustion allows engine de-throttling, thus reducing pumping losses. By making the air-to-fuel ratio (AFR) leaner, however, NO x production during the combustion process increases, which, from a perspective of compliance with current regulations, would need a very efficient aftertreatment system (ATS). This phenomenon is particularly noticeable at least up to a λ of 1.6. Above this threshold, in the ultra-lean region (1.6 < λ < 2), a significant reduction in engine-out NO x is observed [13], due to lower combustion temperature. A drawback of operating SI engines in the ultra-lean region is combustion stability, which greatly deteriorates at such high values of AFR [14], easily reaching misfire conditions. One possible solution is the use of higher energy ignition sources for lean and ultra-lean combustion systems [15]. Another path is the use of a pre-chamber ignition system, which consists in igniting the mixture inside a small volume (pre-chamber) connected to the cylinder (main-chamber). When the flame propagates from the spark-plug to the surrounding pre-chamber volume, combustion products are then discharged to the main chamber through orifices, distributing high-temperature high-momentum gas jets into the main chamber. This strategy accelerates the ignition of the in-cylinder mixture, if compared to a single-point method, enabling the usage of higher dilutions than those usually employed in traditional SI engines [16]. However, as the charge becomes more and more diluted, ignition in the pre-chamber becomes increasingly difficult, if the spark-plug only is present inside this small volume (also known as “passive” pre-chamber configuration). To overcome this drawback, an auxiliary fuelling can be added to the pre-chamber, obtaining the “active” pre-chamber configuration. This solution allows to control the air to fuel ratio inside the pre-chamber, enabling the engine operation in the ultra-lean region, hence providing important enhancements for brake thermal efficiency (BTE) increase [17]. A pre-chamber system has been shown to provide greater repeatability, stability and combustion rate than conventional spark ignition systems operating under similar condi‐ tions because it presents a very high energy release in the mixture to be ignited and multiple ignition points for lean cylinder charge [18][19]. 200 Di Maio, Napolitano, Guido, Beatrice, Sforza, Lucchini, Golini 200 <?page no="201"?> NG HD engines, arising from diesel configuration, can benefit from consolidated components architecture, thus allowing economic advantage in their production. They also have in common the cylinder head, though some modifications (e.g., the cooling circuit and the cylinder head material) are necessary for NG configuration. The cylinder head layout (vertical valves, flat head), however, is derived from diesel case, so allowing the basic machining cycle for both engines. Numerous advantages of the application of a passive pre-chamber to a natural gas-fueled turbulent jet ignition (TJI) engine have been identified in [20]. The authors showed more stable combustion of TJI than conventional SI setup, highlighting greater tolerance to air dilution and combustion phasing changes for lean conditions operation. The introduction of a pre-chamber, moreover, is compatible with the cylinder head configuration currently used, requiring only minor modifications, thus making this solution viable also from an industrial and economic standpoint. 2. Description of the activity and objective This work is based on the results of a previous analysis, involving experimental activities on a HD engine powered by natural gas, as illustrated in the following [21]. Starting from the same experimental conditions and from the development of a predictive 1D “fractal” combustion model carried out by some co-authors [22], a new numerical activity was carried out. It has involved constant and simultaneous interaction between two research groups, CNR-STEMS for the 1D modelling part and PoliMI for the 3D-CFD modelling part, together with the industrial partner FPT Industrial, promoter of the activity. In particular, the first model provided the initial considerations of the combustion process evolution for the further design of a new 1D model, developed in the GT-SUITE platform by CNR-STEMS, distributed by Gamma Technologies [23]. That model represents a single-cylinder, heavy duty SI engine, which is a natural extrapolation of one of the six cylinders of the corresponding real engine, described by Table 1. This model was initially calibrated under two operating conditions, the Cruise point (1200 rpm - 800 Nm) and the Rated Torque condition (1200 rpm - 2000 Nm) in two λ lean conditions, with values of 1.7 and 1.9. The finalization of this model provided insights into the main boundary conditions and heat transfer indications for the simultaneous development of a 3D-CFD model by PoliMI. Several pre-chamber geometries were considered in agreement with the industrial partner. Specifically, within this work, only the one that showed the most interest in terms of system performance, appearing promising for this analysis, will be described. This activity allowed the detailed investigation of combustion processes within the main chamber and active pre-chamber, identifying the masses involved (air, fuel, burned/ unburned) in the tested operating conditions and the main parameters describing the turbulence evolution. Based on these data, the 1D model was appropriately re-calibrated, permitting to monitor the main characteristic parameters of this application and observing the increase in BTE obtained with respect to the relative stoichiometric configuration without pre-chamber. Virtual analysis of the Efficiency Gain with Pre-Chamber combustion systems 201 201 <?page no="202"?> As a final step, a sensitivity analysis was performed as the compression ratio changed up to a value of 14.0, which showed an additional increase in system efficiency, and driving the path toward using active pre-chamber to achieve the 50% BTE target. 3. Numerical 1D-3D models 3.1 1D Single-Cyl Engine Model with Pre-Chamber The 1D model developed in GT-Suite represents a single-cylinder equivalent of the HD 6-Cyl engine, detailed in Table 1. The 6-cyl engine model has been fully validated by CNR-STEMS in previous work, both in Steady-State conditions and during dynamic driving cycles [24]. The model diagram is shown in Figure 3. First, in addition to the main chamber, typical of the investigated single-cylinder, there is also a block describing the pre-chamber, which is present in the latest version of the software to specify the attributes of a pre-combustion for both indirect injection and spark ignited pre-chamber applications. Displaced volume 12.8 L Stroke 150 mm Bore 135 mm Number of Valves 4 Compression ratio 12: 1 Rated Power 338 kW @ 2000 rpm Torque 2000 Nm @ 1100-1620 rpm PFI Injector Natural gas TABLE 1: Main characteristics of corresponding bio-methane 6-cylidnder engine. FIGURE 3: 1D model schematic of the single-cylinder engine, equipped with pre-chamber. 202 Di Maio, Napolitano, Guido, Beatrice, Sforza, Lucchini, Golini 202 <?page no="203"?> It is defined through the indication of its volume and surface area. Heat transfer coefficients were defined from the data included in the main chamber by calculating an area weighed average of the head surface heat transfer coefficient. An equivalent area weighed average of the head surface temperature was also considered for the wall temperature model. The model for flow description is derived from CFD analysis, detailed in the next section. A connector describing the nozzle geometry is defined between the two chambers. In particular, the diameter of the nozzle hole and its number is identified in that block. Additional parameters provide a description of the forward/ reverse discharge coefficients, to determine the trapped masses involved in each condition. The extreme blocks of the engine diagram are represented by the Intake conditions, for which an instantaneous pressure and an intake temperature are defined, and the Exhaust block, representing the same conditions at the outlet of the cyclinder. Both are needed to provide the bounday conditions of the system. The exhaust pressure in particular is determined by a control logic, tunable based on a value identified by the user and entered in the “Delta_p” block. A sensitivity analysis on the optimal value, also by virtue of the final experimental application, was carried out in agreement with the OEM. The final identified value adopted in relation to the equivalent turbocharger, is not here reported, being a sensitive data, but it is easily assumed that the lower is the backpressure, the higher is expected to be the performance. A specific injector is associated with each chamber. The first, related to the main chamber, injects a quantity of fuel into the flowsplit runner present before the ramification that connects the intake line to the two valves. The second, instead, injects directly into the pre-chamber. The amount of fuel to be injected is regulated by additional control logic. First, the brake mean effective pressure (BMEP) value is determined by a controller that acts on the Intake pressure to reach the target expected from each operating condition tested. Based on this pressure, an intake air flow rate is determined. From an external signal, representing the target λ for the operating condition to be reproduced, together with the relevant formulas, the quantity of fuel to be injected into the system can be determined. An additional signal, representing the fuel distribution ratio between main chamber and pre-chamber sets the amount of fuel flow rate that reaches both chambers. Both the 1D and 3D-CFD simulations involve the use of 100-% methane fuel mixture. 3.2 3D-CFD NUMERICAL MODEL The modeling of the combustion process inside an active pre-chamber ICE is a challenging task, especially for 1D approaches. In fact, a proper prediction of the mixing and the reacting dynamics of the hot burned jets discharged by the pre-chamber is fundamental for a reliable description of the main chamber combustion process. Therefore, in this work, 3D-CFD simulations of the engine power-cycle are carried out to achieve a high-fidelity characterization of the combustion process inside the selected engine configuration. Then, the computed results are provided in terms of main-chamber heat release rate evolution to the 1D framework, to accomplish the full engine characterization, estimating its break thermal efficiency. The 3D-CFD modelling of the power-cycle is performed from intake valves closing (IVC) to exhaust valves opening (EVO) considering the closed-valves geometry of one cylinder, Virtual analysis of the Efficiency Gain with Pre-Chamber combustion systems 203 203 <?page no="204"?> namely its preand mainchambers only. All boundaries are walls, and the fields are initialized at IVC. In particular: • The flow field and the turbulence level inside the cylinder are retrieved from previous 3D-CFD gas-exchange simulations carried out on the same engine, without the pre-chamber. For more details the reader is referred to [22]. • Temperatures at walls are imposed according to results achieved from 1D simulations of a similar heavy-duty engine system, already validated against available experimental data [25]. • Different air/ fuel ratios are assumed inside both the pre- (λ P C, I V C ) and main- (λ MC, I V C ) chambers, according to preliminary 1D simulations of the pre-chamber scavenging. Homogeneous mixtures are considered during the initialization process. This hypothesis is a reasonable simplification for the main-chamber, considering the port-fuel injection (PFI) strategy, but also for the pre-chamber. In fact, here, the mixture stratification is mainly influenced by the cylinder-to-pre-chamber flow jets activated during the compression phase, as clarified by Kim and Sforza [25] [26]. FIGURE 4: 3D-CFD domain for the power-cycle simulation: a sector representing 1/ 6 of the closed-valves domain. The piston is positioned at the top dead center (TDC). A sector of 1/ 6 of the total closed-valves domain, as shown by Figure 4, is selected for the 3D-CFD simulations to minimize the computational effort. This choice is justified by the pronounced axis-symmetrical features of the engine analysed in the present study, which 204 Di Maio, Napolitano, Guido, Beatrice, Sforza, Lucchini, Golini 204 <?page no="205"?> is featured by: axis-symmetrical piston and nozzle geometry distribution, a flat cylinder head, and a central ignition position. The main features of the adopted 3D computational mesh are shown by Figure 5 and reported in Table 2. They are consistent with those employed in a previous work, where a similar actively fuelled pre-chamber engine was validated against available experimental data [25], to maximise the reliability of the numerical setup. In particular, the mesh is characterized by an almost jet-oriented structure, to minimize the CFD interpolation error related to strongly misaligned cells-jet orientations. The average cell size is 0.5 mm, which is reduced to 0.1 mm inside the nozzle and around their inlet/ outlet sections (Figure 5, zoom). The dynamic layering technique is employed to accommodate the piston motion [27], hence the total number of cells spans from 165k (TDC) to 474k (IVC) cells. Feature Value Cell n. @ TDC 165k Cell n. @ IVC 474k Cell size in PC 0.5 mm Cell size in MC 0.5 mm Cell size in nozzle 0.1÷0.15-mm TABLE 2: Main features of the 3D computational mesh. FIGURE 5: 3D sector mesh employed for the power-cycle simulation, showed on a cut-plane along the nozzle axis with the piston at TDC position. The zoom image shows the mesh refinement adopted inside the nozzle. Virtual analysis of the Efficiency Gain with Pre-Chamber combustion systems 205 205 <?page no="206"?> (1) All 3D-CFD simulations are performed with Lib-ICE, which is a set of solvers and libraries based on the OpenFOAM platform and where the numerical methods were previously implemented [25]. The RANS approach is used and the k-ε turbulence model is employed, with the suggested literature standard coefficients. The time-derivative is discretized with the Euler 1 st order scheme, while a limited 2 nd order scheme is used for convection terms. 4. 3D-CFD combustion model The 3D combustion process is modelled according to the methodology used and validated by Sforza [25]. In this section, a brief description of such approach is reported, to help the reader understanding. FIGURE 6: Schematic of the 3D combustion model. The structure of the selected 3D combustion model is reported in Figure 6. The flame front propagation is modelled by solving a transport equation for the regress variable b (unburned gas fraction) [28]: ∂ρb ∼ ∂t + ∇ • ρUb ∼ + ∇ • μ t ∇b ∼ = ρ u S ∼ u Ξ ∼ ∇b ∼ + ω˙ ign where the reaction rate (first right hand side term) is estimated according to the flame area evolution (FAE) model from Weller. In Eq. 1, ρ and ρ u are the mixture and unburned mixture densities, U the flow velocity, μ t the turbulent dynamic viscosity, S u the laminar flame speed and Ξ the wrinkle factor. Ignition is handled by means of a simplified deposition model 206 Di Maio, Napolitano, Guido, Beatrice, Sforza, Lucchini, Golini 206 <?page no="207"?> (2) (3) (4) (5) (6) ω˙ ign = C s ρ u b Δt ign where the ignition energy can be calibrated in terms of magnitude (C s ) and time-duration (Δt ign ), similarly to [29] . The flame-turbulence interaction effect is embedded in Eq. 1 by means of the Ξ parameter, which represents the ratio between the turbulent and laminar flame speeds (S t / S u ) and it is modelled as [28]. Ξ = f Ξ eq Here, the f parameter handles the laminar-to-turbulent transition process, taking place after the ignition event, and is modelled according to [30]. f = 1 − ex p − r k L t 1/ 2 • 1 − ex p − u ′ + S u L t • t ign 1/ 2 The kernel radius r k value is predicted through the simplified 0-D sub-model proposed in [29], while u ′ is the turbulence intensity and L t the integral length scale. The equilibrium wrinkle factor Ξ eq of Eq. 3 is modelled according to Peters [31]: Ξ eq = 1 − a 4 b 32 2b 1 L t δ l + a 4 b 32 2b 1 L t δ l 2 + a 4 b 32 u ′ S u L t δ l 1/ 2 where a 4 = b 3 = 0 . 78 and b 1 = 3 . 5 are model constants taken from [31] and slightly calibrated. The laminar flame thickness δ l and speed S u (Eq. 1, 4 and 5) are both retrieved from a lookup table (see Figure 6), which is created from 1-D laminar flame speed calculations at constant-pressure conditions, considering local thermodynamic conditions (P and T ) and mixture properties (λ). The chemical composition, instead, is computed as weighted average over the regress variable b of the burned Y b and unburned Y u mixture compositions: Y i = b ⋅ Y u, i + 1 − b ⋅ Y b, i The Y b value at equilibrium conditions is extracted from a dedicated lookup table, generated with homogeneous reactor calculations as illustrated in [32]. 5. Results and discussion In the context of 1D engine modeling treated, the combustion process refers to the transfer of a defined amount of unburned fuel mass and air (along with the associated enthalpy) from an unburned zone to a burned zone in the cylinder. Virtual analysis of the Efficiency Gain with Pre-Chamber combustion systems 207 207 <?page no="208"?> This process is governed by the “burn rate,” that is the rate at which fuel and air molecules are transferred to the burn zone. For a consistent modelling between the two platforms, a methodology to identify the masses involved in the combustion process from 3D-CFD investigations was identified. In fact, considering the total mass, trapped mass in pre-chamber, as well as the value of AFR and combustion efficiency, the percentage of fuel burned, both in Main Chamber and pre-chamber, was determined. This profile was then used in the 1D software to model the combustion process. As an example, Figure 7 shows the trend for the most challenging condition, in the extreme λ lean value of 1.9. FIGURE 7: Fuel burned rate in pre-chamber and Main Chamber in Cruise and Rated Torque conditions at λ = 1.9. Figure 8 shows the trends of in-cylinder pressure in the main chamber. Each graph shows both the profile obtained from 3D-CFD modeling (in blue) and the profile obtained from the 1D single-cylinder engine model, derived from the corresponding burn rate described above. Throughout the conditions tested, there is acceptable agreement between the results of the two simulation platforms. In particular, during the compression phase, the in-cylinder pressure curves are highly aligned with each other. This highlights the correct evaluation of the involved masses, especially for intake air. The fuel ignition process is also correctly reproduced by the 1D model. A minor, but consistent gap in all conditions tested concerns peak firing pressure (PFP). The 1D single-cylinder engine model underestimates PFP compared with the result of the 3D-CFD analysis. The main cause of this deviation could be traced to an inconsistent evaluation of the heat transfer model in the lean conditions. As anticipated, in fact, it represents an extension of the model obtained from the 6-cylinder engine under stoichiometric conditions, therefore, a validation against experimental data becomes necessary to better clarify any misalignment from the current results. However, in the context of this preliminary evaluation of the engine behavior when equipped with an active pre chamber, the achieved consistency between 1D and 3D results can be considered satisfactory. 208 Di Maio, Napolitano, Guido, Beatrice, Sforza, Lucchini, Golini 208 <?page no="209"?> FIGURE 8: In-cylinder pressure in Main Chamber in the tested operating conditions. Virtual analysis of the Efficiency Gain with Pre-Chamber combustion systems 209 209 <?page no="210"?> Regarding the pre-chamber in-cylinder pressure, shown in Figure 9, the typical first peak can be recognized as a consequence of the completion of the pre-combustion of the amount of fuel that is present in pre-chamber. 210 Di Maio, Napolitano, Guido, Beatrice, Sforza, Lucchini, Golini 210 <?page no="211"?> FIGURE 9: In-cylinder pressure in pre-chamber in the tested operating conditions. Figure 10 shows the heat release rate profile in main chamber for each simulated condition. Also in this instance, there is a small difference on the peak of this profile. It is attributable once again to a mismatch on the heat exchange in lean conditions and to the considerations previously made for the description of the in-cylinder pressure. Nevertheless, the 1D and 3D/ CFD profiles are aligned, both in terms of combustion ignition and timing. Virtual analysis of the Efficiency Gain with Pre-Chamber combustion systems 211 211 <?page no="212"?> FIGURE 10: Heat Release Rate in Main Chamber in the tested operating conditions. As shown in Figure 11, through the use of a lean blend the heat release rate profile is milder, by virtue of the lower amount of fuel involved, and with an earlier ignition. FIGURE 11: Heat Release Rate in Main Chamber in Cruise Operating Point. Both profiles show the 3D-CFD profile in the two λ lean conditions. 212 Di Maio, Napolitano, Guido, Beatrice, Sforza, Lucchini, Golini 212 <?page no="213"?> The discharge coefficient calibration in the block that identifies the connection between pre-chamber and main chamber allowed an adequate assessment of the total trapped mass involved throughout the entire cycle evolution. Two respective values were found for the forward and reverse discharge coefficients, which showed a reasonably good accuracy of the trapped mass profile to the variation of the crank angle compared with the 3D-CFD simulation. However, a smaller gap in the rated torque condition is found in the first part of the cycle, evident indication that the pre-chamber air filling is slightly overestimated in the 1D simulation. In absolute terms, the agreement is optimal in all the analyzed conditions. Virtual analysis of the Efficiency Gain with Pre-Chamber combustion systems 213 213 <?page no="214"?> FIGURE 12: 1D/ 3D-CFD Comparison of trapped mass in the pre-chamber to variation of crack angle in the tested operating conditions. With the aim of maximizing the efficiency of the system, and keeping an adequate ignition and stability of the combustion process, in addition to the use of a combustion pre-chamber, a sensitivity analysis to the compression ratio was carried out. This analysis was performed to provide a further indication of the efficiency benefits that the addition of the pre-chamber can lead to a HD engine as the one investigated. In Figure 12, in fact, in addition to an indication of the Indicated efficiency, consisting of brake efficiency and friction losses, additional BTE values have been presented again by applying the considerations of this study with a higher compression ratio. As expected, an increase in compression ratio promotes a higher in-cylinder pressure that nevertheless results in an increase in friction losses [33]. Therefore, within the 1D software, a design of experiments was developed to explore different CR values, up to the set limit of 14. This value was identified in agreement with the OEM to keep a correct combustion process, avoiding abnormal evolutions, within the entire engine workplan. It takes into account a prefixed COV IMEP limit of 4 %, especially in the highest load conditions. It must be considered that the value 14 is a limit that could also be exceeded through future developments, as also investigated in [34]. In that work, carried out similarly on an HD NG engine, the analysis was carried out up to a value of 15. The analysis carried out in this paper, could therefore hopefully not be limited to this CR, presenting even greater advantages. However, this purpose should be verified by means of appropriate experimental tests. The same authors also carried out an interesting analysis when varying other geometric parameters, showing promising results for the similar NG-fuelled engine category [35]. Starting from the reference condition of the HD 6-cyl engine under stoichiometric conditions, the increase in friction losses with respect to the total percentage of fuel energy increases by +1.4%t when reaching a CR equal to 14 and λ equal to 1.7 at the cruise point. Under rated torque conditions, given the high values of in-cylinder pressure, this increase is +2.5% with ultra-lean λ equal to 1.9 and maximum CR equal to 14. Excluding the change in compression ratio, the increase in BTE in the cruise point with the minimum λ value of 1.7 equals 1.3 %, which increases to 1.9 % if the CR goes from 12 to 14. In the rated torque condition, the advantages in terms of BTE are highly consistent and 214 Di Maio, Napolitano, Guido, Beatrice, Sforza, Lucchini, Golini 214 <?page no="215"?> appreciable. Excluding the change in compression ratio, the increase in BTE in the most extreme lean condition equal to λ 1.9 is 2.4 %, which goes up to 3.5% if the CR is increased from 12 to 14. The pre-chamber application in HD engines, therefore, is set in a context of absolute industrial interest both for the appreciable increase in system efficiency, which translates into a considerable benefit in terms of fuel economy without sacrificing the performance of conventional configurations, and also from an exhaust emission reduction point of view, given the achievement of a lean mixture. The deployment and subsequent developments resulting from this technology sets the stage for subsequent use towards the ambitious 50 % brake thermal efficiency target. FIGURE 13: System efficiency evaluation in Cruise and Rated Torque conditions respectively for extreme values of λ 1.7 and λ 1.9. The first represented bar shows the efficiencies of the reference HD 6-cylinder engine in stoichiometric conditions; the next bars show a λ lean and the change in compression ratio. 6. Conclusion The present work showed an extensive 1D/ 3D-CFD modelling activity, performed simul‐ taneously with synergy between the research groups to reproduce the use of an active pre-chamber for a heavy-duty engine powered by biomethane in two operating conditions. These engine points are the cruise and rated torque condition at a medium engine speed of 1200 rpm, respectively, in two lean λ conditions of 1.7 and 1.9. The technological solution showed substantial advantages over the reference application of the stoichiometric 6-cyl HD engine without the pre-chamber. In terms of BTE, the benefit is between a minimum value of +1.3% in the cruise point at lambda 1.7 and a value of +2.4% in rated torque condition at lambda 1.9. If in such conditions the compression ratio is further increased, these benefits reach respectively +1.9% for the first cruise point and 3.5% for the rated torque condition. Virtual analysis of the Efficiency Gain with Pre-Chamber combustion systems 215 215 <?page no="216"?> The outcomes from this investigation are particularly promising towards 50 % BTE challenging target, also for HD engines, particularly relevant due to the difficulty in their electrification and, at the same time, in compliance with legislative emission standards. This activity lays the foundation for further experimental work on a single-cylinder engine, contributing to a broader understanding of the combustion process and the benefits in the application of this technology, monitoring the exhaust gas composition under the above-mentioned lambda lean conditions. ACKNOWLEDGEMENTS This research has been partially supported by the European Union - NextGenerationEU - National Sustainable Mobility Center CN00000023, Italian Ministry of University and Research Decree n. 1033— 17/ 06/ 2022, Spoke 12, CUP B43C22000440001. ABBREVIATIONS AFR Air-to-Fuel Ratio ATS Aftertreatment System BEV Battery Electric Vehicle BMEP Brake Mean Effective Pressure BTE Brake Thermal Efficiency CBM Compressed Bio-Methane CFD Computational Fluid Dynamics CNG Compressed Natural Gas CR Compression Ratio EVO Exhaust Valves Opening GHG Greenhouse Gas HD Heavy Duty ICE Internal Combustion Engine IVC Intake Valves Closing MC Main Chamber NG Natural Gas OEM Original equipment manufacturer PC Pre Chamber PFP Peak Firing Pressure PHEV Plug-in Hybrid Electric Vehicle 216 Di Maio, Napolitano, Guido, Beatrice, Sforza, Lucchini, Golini 216 <?page no="217"?> RANS Reynolds Averaged Navier-Stokes SI Spark Ignition TDC Top Dead Centre TJI Turbulent Jet Ignition REFERENCES [1] European Environmental Agency, New registrations of electric vehicles in Europe, October 2023, https: / / www.eea.europa.eu/ en/ analysis/ indicators/ new-registrations-of-electric-vehicles? activeA ccordion=ecdb3bcf-bbe9-4978-b5cf-0b136399d9f8 [2] Cerruti, G., Chiola, M., Bianco, V., Scarpa, F., “Impact of electric cars deployment on the Italian energy system”, Energy and Climate Change, Volume 4, 2023, 100095, ISSN 2666-2787, doi: org/ 10.1016/ j.egycc.2023.100095. [3] Danielis, R., Rotaris, L., Giansoldati, M., Scorrano, M., “Drivers’ preferences for electric cars in Italy. Evidence from a country with limited but growing electric car uptake”, Transportation Research Part A: Policy and Practice, 137: 79-94, 2020, doi: org/ 10.1016/ j.tra.2020.04.004. [4] Więckowski, J., Wątróbski, J., Kizielewicz, B., Sałabun, W., “Complex sensitivity analysis in Multi-Criteria Decision Analysis: An application to the selection of an electric car”, Journal of Cleaner Production, 390, 2023, doi: 10.1016/ j.jclepro.2023.136051. [5] Kucukoglu, I., Dewil, R., Cattrysse, D., “The electric vehicle routing problem and its variations: A literature review”, Computers & Industrial Engineering, 161, 2021, doi: 10.1016/ j.cie.2021.107650. [6] EBA Statistical Report, 2023. [7] Searle, S., Baldino, C., Pavlenko, N., “What is the role for renewable methane in European decarbonization? ”, The International Council on Clean Transportation, October 2018. [8] Prussi, M., Yugo, M., De Prada, L., Padella, M., Edwards. JEC Well-To-Wheels report v5. EUR 30284 EN, Publications Office of the European Union, Luxembourg, 2020, ISBN 978-92-76-20109-0, doi: 10.2760/ 100379. [9] Rapetto N. at al., “H 2 ICE technologies as key enabler for the decarbonization of the heavy duty sector”, 32 nd Aachen Colloquium Sustainable Mobility, Aachen, October 2023. [10] Weber O. et al., “Optimized hydrogen injection strategies for lowest NOx emissions”, 11 th International Engine Conference, Baden-Baden, February 2024. [11] Ferrarese A. et al., “Hydrogen ICE with highest efficiency”, 11 th International Engine Conference, Baden-Baden, February 2024. [12] https: / / www.fptindustrial.com/ en/ media/ newsletter/ 2021_06_04-FPT-Industrial-and-Landi-Ren zo-Group-team-up-for-clean-fuels. [13] Heywood J., Internal Combustion Engines Fundamentals, 2 nd ed., Mc-Graw Hill, New York, 2018. [14] Quader A. A., “Lean Combustion and the Misfire Limit,” SAE Technical Paper 741055, 1974 [15] Qiao A. and Wu X., “Research on the New Ignition Control System of Leanand Fast-Burn SI Engines,” SAE Technical Paper 2008-01-1721, 2008, doi: 10.4271/ 2008-01-1721. [16] Attard, W., Fraser, N., Parsons, P., and Toulson, E., “A Turbulent Jet Ignition Pre-Chamber Combustion System for Large Fuel Economy Improvements in a Modern Vehicle Powertrain,” SAE Int. J. Engines 3(2): 20-37, 2010, doi: 10.4271/ 2010-01-1457. Virtual analysis of the Efficiency Gain with Pre-Chamber combustion systems 217 217 <?page no="218"?> [17] Peters N. at al., “Design and Development of a High-Efficiency Single Cylinder Natural Gas-Fueled Jet Ignition Engine”, SAE Technical Paper 2019-32-0565, 2019. [18] Toulson E, Schock HJ, Attard WP. A review of pre-chamber initiated jet ignition combustion systems. In: SAE technical paper. SAE International; 2010. https: / / doi: org/ 10.4271/ 2010-01-2263. [19] Desantes JM, Novella R, Morena JDL, Pagano V. Achieving ultra-lean combustion using a pre-chamber spark ignition system in a rapid compressionexpansion machine. In: SAE technical paper. SAE International; 2019. doi: 10.4271/ 2019-01-0236. [20] López, J.; Novella, R.; Gomez-Soriano, J.; Martinez-Hernandiz, P.; Rampanarivo, F.; Libert, C.; Da‐ biri, M. Advantages of the unscavenged pre-chamber ignition system in turbocharged natural gas engines for automotive applications. Energy 2021, 218, 119466. doi: 10.1016/ j.energy.2020.119466. [21] G. Onofrio, P. Napolitano, P. Tunestål, C. Beatrice, “Combustion sensitivity to the nozzle hole size in an active pre-chamber ultra-lean heavy-duty natural gas engine”, Energy 235, 2021, doi: org/ 10.1016/ j.energy.2021.121298. [22] Riccardi, M., De Bellis, V., Sforza, L., Tunestal, P. et al., “Experimental and Numerical Analysis of an Active Pre-Chamber Engine Fuelled with Natural Gas,” SAE Technical Paper 2023-01-0185, 2023, doi: org/ 10.4271/ 2023-01-0185. [23] Gamma Technologies, GT-Suite, Engine Performance Application Manual, US: Westmont, 2023. [24] Di Maio, D.; Stramaccioni, E.; Misul, D.A.; Napolitano, P.; Beatrice, C. A Multiphysics Co-Simu‐ lation Framework of a Gas Engine and Three-Way Catalyst toward a Complete Vehicle Design Model. Machines 2022, 10, 852. doi: org/ 10.3390/ machines10100852 [25] Sforza, L., Lucchini, T., D’Errico, G., Gianetti, G. et al., “A Numerical Methodology for the Design of Active Pre-chambers in Spark-Ignition Engines,”-Combustion Science and Technology, 2023; 1-35, doi: 10.1080/ 00102202.2023.2219450. [26] Kim, J., Scarcelli, R., Som, S., Shah, A. et al., “Numerical investigation of a fueled pre-chamber spark-ignition natural gas engine,” International Journal of Engine Research, 2023; 23 (9): 1475-94, doi: 10.1177/ 14680874211020180. [27] Lucchini, T., Della Torre, A., D’Errico, G., Montenegro, G. et al., “Automatic mesh generation for CFD simulations of direct-injection engines,” SAE Paper 2015- 01-0376, 2015. [28] Weller, H. G., Tabor, G., Gosman, A. D. and Fureby, C., “Application of a flame-wrinkling LES combustion model to a turbulent mixing layer,” Proceedings of the Twenty-Seventh Combustion Symposium (International), Boulder, Colorado, US; pp. 899-907; 1998. [29] Sforza, L., Lucchini, T., Gianetti, G., and D’Errico, G., “Development and Validation of SI Combustion Models for Natural-Gas Heavy-Duty Engines,” SAE Technical Paper 2019-24-0096, 2019, doi: 10.4271/ 2019-24-0096. [30] [Maly, R. R., and Herweg, R., “A fundamental model for flame kernel formation in SI engines,” SAE Paper 922243, 1992. [31] Peters, N., “Turbulent Combustion,” Cambridge University Press; 2000. doi: 10.1017/ CBO9780511612701 [32] Lucchini, T., D’Errico, G., Onorati, A., Frassoldati, A., et al., “Modeling non-premixed combustion using tabulated kinetics and different flame structure assumptions,” SAE Int. J. Engines, 2 (2): 593- 607; 2017. doi: 10.4271/ 2017-01-0556. [33] Pesic, Radivoje & Milojević, Saša & Veinovic, Stevan. (2010). Benefits and challenges of variable compression ratio at diesel engines. Thermal Science. 14. 1063-1073. 10.2298/ TSCI1004063P. 218 Di Maio, Napolitano, Guido, Beatrice, Sforza, Lucchini, Golini 218 <?page no="219"?> [34] Liu, X., Silva, M., Mohan, B., AlRamadan, A.S. et al. “Computational optimization of the performance of a heavy-duty natural gas pre-chamber engine”, Fuel, 352, 2023, doi: 10.1016/ j.fuel.2023.129075. [35] Liu, X., Marquez, M., Sanal, S., Silva, M et al. “Computational assessment of the effects of pre-chamber and piston geometries on the combustion characteristics of an optical pre-chamber engine”, Fuel, 341, 2023, doi: 10.1016/ j.fuel.2023.127659. Virtual analysis of the Efficiency Gain with Pre-Chamber combustion systems 219 219 <?page no="221"?> 1 FKFS - Forschungsinstitut für Kraftfahrwesen und Fahrzeugmotoren Stuttgart 2 Institut für Fahrzeugtechnik Stuttgart 3 Fraunhofer ICT 4 Ford Werke GmbH Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation via Active Pre-Chamber System: Numerical Investigation and Experimental Validation Dr.-Ing. Antonino Vacca, M. Sc. Cristian Tortorella, Dr.-Ing. Marco Chiodi 1 , M. Sc. Sebastian Bucherer, Prof. Dr.-Ing. André Casal Kulzer 2 , Dipl.-Ing. Florian Helmut Karl Sobek, M. Sc. Paul Rothe, Dipl.-Ing Ivica Kraljevic, Dr.-Ing. Hans-Peter Kollmeier 3 , Dipl.-Ing. Albert Breuer, Dr.-Ing. Ruhland Helmut 4 Preface Abstract As governments strive towards the mitigation of carbon dioxide emissions, methane-powered engine emerges as pivot catalyst in achieving this objective, since natural gas has a low carbon-to-hydrogen ratio and high knock resistance, compared to traditional liquid fuels. The 3D-CFD virtual development can play a crucial role in investigating a new engine design to exploit the potential of methane injection. For this purpose, a new cylinder head and combustion chamber are developed to host an active pre-chamber system and an additional direct injector operating at 26 bar. Through the 3D-CFD-Tool QuickSim different engine geometries are tested and a high-tumble concept is selected and manufactured as a single-cylinder engine. Two different active pre-chamber layouts are manufactured, one with a volume of 500 mm 3 and the possibility to be cooled by a separate channel, while the second one has a volume of 750 mm 3 and is designed to be hosted in the cylinder head of a three-cylinder engine, derived from the single-cylinder one, without a dedicated cooling system. In the current work, the experimental results for the single-cylinder engine at 2000 rpm 10 bar IMEP are discussed, with stoichiometric or lean operations and with different injection strategies in the pre-chamber. The validation of the simulation is realized through the comparison of the indicating pressure curves, especially against the measurements of the pressure in the pre-chamber through an additional sensor. The influence of the pre-chamber volume is presented and the results are correlated with the analysis of the flow field realized by the CFD investigations. The pre-chambers with an optimized injection strategy allow the engine to achieve lambda 1.6 with a COV of IMEP below 1.5 %. In the last part of the work, a load variation is presented. <?page no="222"?> Particularly at 17 bar IMEP and 2000 rpm, the engine could be operated with stable combustion at lambda 1.4 by achieving 44-% indicated efficiency. 1 Introduction The transportation sector is deeply committed to reducing emissions in the atmosphere, especially regarding carbon dioxide and greenhouse gases (GHG). In this scenario, the need for clean and efficient engines and fuels is nowadays stronger than ever. With this purpose, different fuels are considered as a possible replacement for traditional gasolines to accomplish this target. Between these, methane can be an immediate solution for the reduction of CO 2 emissions of the transport sector, since it has been for long employed as fuel for internal combustion engines (ICEs) and it can also take advantage of an already existing infrastructure, where the amount of renewable methane is considerably (for example 60 % of methane gas station are filled with bio methane in Germany [1]). Since Methane high hydrogen-to-carbon ratio, by burning it in an ICE, a CO 2 reduction of 24 % can be achieved compared to a gasoline engine with comparable efficiency, only considering the fuel replacement [2]. Moreover, natural gas has a higher Research Octane Number (RON) if compared to common gasolines (RON-Methane ≥120, RON Gasoline = 95-100), which leads to higher knock resistance [3]. Compressed natural gas (CNG) presents also good characteristics from the ignitability point of view if compared to gasoline, that ensure the adoption of diluted-combustion process via exhaust gas recirculation (EGR) or lean operation, leading to a further increase in efficiency. To enlarge the lean operation limit of methane-fuelled engines even more, another interesting solution is the adoption of a pre-chamber ignition system. The working principle of a pre-chamber spark plug (PC) is based on having a small volume in which the electrode is contained, linked to the main combustion chamber through small ducts. The geometry of the pre-chamber volume and the holes are designed purposely to introduce strong turbulence near the electrode to ensure a rapid combustion event. Within the ducts of the PC, turbulent jets reach the main combustion chamber resulting in high-pressure gradients near the top dead center (TDC), leading to very efficient combustion and so avoiding anomalous combustion forms such as knocking, via the short combustion process. Adopting an active pre-chamber (APC) allows the control of the stoichiometry into the pre-chamber volume, decoupling the mixture formation into the APC from the main combustion chamber. From these considerations, an engine operated with CNG, having a high efficiency requires a specific design and optimization. To realise such an engine, a consortium made by Ford-Werke GmbH, Fraunhofer-Institut für Chemische Technologie (ICT), Forschungsinstitut für Kraft‐ fahrwesen und Fahrzeugmotoren Stuttgart (FKFS), Rosswag GmbH and BRIGHT Testing GmbH (financed by BMWI) decided to run a 36 months project (Nr. 19I20014E), called MethMag (Methan Mager Motor). In previous works of the authors regarding the engine object of this work, all the steps of the design phase have been discussed, of which a resume is reported as follows: 222 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 222 <?page no="223"?> • The project aimed to develop a prototype three-cylinder engine derived from the combustion optimization of a single-cylinder engine. An evolution of the gasoline-pow‐ ered Ford Ecoboost (3-cylinder, 1.5 l, direct injection) engine has been used as a benchmark, to define the operating range for the new CNG engine. With the aim of exploiting methane potential as much as possible, 6 operating points are chosen for the investigation, as shown in Fig. 1. Particularly, the grey dots represent the operating points chosen for the analysis by means of 3D-CFD investigations [4] and for the later validation of the models through measurements at the test bench. Considering Fig. 1, the new methane engine achieves higher full load characteristics due to the increased resistance to knock of methane compared to gasoline and a higher combustion efficiency due essentially to pre-chamber ignition and the higher compression ratio. Fig. 1 also highlights the lean operations of the engine, being able to run with a stable combustion process up to 17-bar IMEP at lambda 1.4. Fig. 1: Operating map of the gasoline benchmark engine compared to the full load curve and the lean operating area of the new methane engine A detailed comparison between the benchmark engine and the new methane engine is provided in [4] and a resume of it is outlined in Tab.-1. Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 223 223 <?page no="224"?> Gasoline Benchmark Engine New-designed Methane En‐ gine Bore 84 84 Stroke 89.8 89.8 Injection PFI/ DI Pre-Chamber/ DI Valve overlap large (residual gas for de-throttling) small (high boost pressure, no throt‐ tling) Intake valve strategy Millerisation required (Knock) High lift to reduce the boost pressure Mixture Stoichiometric/ rich Stoichiometric / lean Knock resistance - ++ Exhaust enthalpy + - Compression ratio [-] 12.5 15 Max indicated efficiency [%] 40 44 Max. Pressure [bar] 130 180 Aftertreatment TWC TWC Tab. 1: Comparison of the benchmark gasoline engine with the new-designed methane engine. • In order to enlarge the lean operating zone of the new methane engine, an APC is chosen as ignition system, instead of a standard spark plug. After the design of a completely new cylinder head, purposely optimized to host the APC and a fully variable valve activation system [4], [5], the new cylinder head geometry and different APC designs have been tested [6] using 3D-CFD simulations. Two of the analysed concepts are shown in Fig. 2. They represent the first pre-chamber concept (APC1) versus the latest one (APC2), which was then manufactured and tested in the real hardware (to be highlighted is the new shape of the cap for APC2 to reduce thermal stress, the asymmetry of the holes and the injection channel for the connection of PC injector and main PC volume). 224 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 224 <?page no="225"?> Fig. 2: First APC design compared to the final manufactured one (APC2) used in the measurements and developed through 3D-CFD simulations [6]. Fig. 2 compares APC1 and APC2 (the right side of each section of Fig. 2 corresponds to the engine intake side). In APC1, the injector is placed on the top right corner of the pre-chamber volume with the spark plug next to it. This configuration results in a large inner diameter of the upper part of the pre-chamber, with the electrode too far away from the centre of the pre-chamber volume. 3D-CFD simulations have shown poor mixture formation for APC1, due to the shape of the pre-chamber volume. To ensure higher design flexibility (reduction of the pre-chamber space requirements in the cylinder head), improve the mixture formation and place the electrode in a more centred position, APC2 has been designed. The injector is heightened and it is connected to the pre-chamber volume by a small channel. This arrangement ensures a higher design flexibility for the pre-chamber volume to obtain a lower inner diameter. Moreover, for APC2 the electrode is moved closer to the main combustion chamber, resulting in a more compact pre-chamber design. More details can be found in other works of the authors [6], [4]. After investigations with 3D-CFD and CHT simulations, APC2 has been chosen as the best solution for the new methane engine. • A peculiarity of the new single-cylinder engine is an active pre-chamber with a dedicated cooling system and housing for a pressure transducer. The APC2 has been designed to have a coolant jacket completely separated from the coolant jacket of the cylinder. This solution ensures maximum flexibility, allowing the conditioning of the pre-chamber for every specific operating point. The APC2 with dedicated cooling system is used for research purposes to investigate nitrogen oxide (NO x ) emissions and reduce the risk of undesired glow ignition at high-load operating points. 3D-CFD Simulations coupled with conjugated heat transfer simulations (CHT) of the coolant flow related to the cylinder coolant jacket have shown a homogeneous flow distribution with a uniform temperature field for the cylinder head roof, without temperature peaks which can be very dangerous for the thermo-mechanical durability [5]. Such investigations have been conducted also for the coolant flow dedicated to the PC, showing that the flow through the channels is strictly dependent on the momentum of the incoming flow, and not on advantageous pressure gradients [5]. In Fig. 3 the result Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 225 225 <?page no="226"?> of the pre-chamber coolant flow simulations can be seen, while in Fig. 4 is outlined the temperature distribution for the cylinder head for OP 6 at the peak power point with 23-bar IMEP and 5500-rpm. Fig. 3: 3D-CFD coolant flow simulation of the pre-chamber coolant jacket [7]. Fig. 4: 3D-CHT simulation of cylinder head roof temperature field [7]. However, the cooling of the APC did not lead to any particular improvements in emissions and performance. This can be addressed to the small volume of the APC (500 mm 3 ). As a matter of fact, such a small volume can only condition a small quantity of mass, with respect to the total mass evolving in one engine cycle. The numerical investigations carried on through the 3D-CFD tool QuickSim have led to the production of a single-cylinder engine to be installed at the test bench (TB) by ICT. The focus of this work is the validation of the results obtained within 3D-CFD simulation, which took place months before the construction of the prototype methane engine in its single-cylinder configuration. Many operating points have been measured by ICT, nevertheless, this work will outline the results concerning 2 specific operating points: • 2000 rpm, 10-bar IMEP, 1 ≤ λ ≤ 1.6 • 2000 rpm, 17-bar IMEP, λ = 1.4 226 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 226 <?page no="227"?> Concerning the operating point 2000 rpm, 10 bar IMEP, the investigations focus on the pre-chamber injection strategy, while for the operating point 2000 rpm, 17 bar IMEP the analysis aims to show how to achieve the highest indicated efficiency, through the pre-chamber working as active concept (with injection). 2 Simulation setup and models The numerical investigations in this work are carried out with the 3D-CFD tool QuickSim, developed at IFS/ FKFS Stuttgart. QuickSim is specifically designed for ICEs simulation and within ICE-adapted models that are constantly improved, the tool ensures the possibility of reducing the computational time compared to other 3D-CFD approaches, without sacrificing the quality of the results. Coupling the shortened computational time and the employment of coarser meshes, QuickSim allows the extension of the computational domain up to a full engine configuration with a computational time of 4 hours for an operating cycle of a full engine with 32 CPUs. A detailed description of the models implemented in QuickSim can be found in other works of the Authors [8] [9] and a summary is reported in Tab.-2. Simulation Methodology Multiphase RANS Multi-phase flow Euler / Lagrange Turbulence κ-ε Combustion - Flame Propagation Two-zone Weller model adapted to GDI combus‐ tion for SI engines Time step Δt Const. = 0.5 degree crank angle (°CA) Fuel model Detailed chemistry solved separately in Cantera Tab. 2: Models implemented in QuickSim 2.1 Fuel and knock modelling The 3D-CFD tool QuickSim can reproduce any type of fuel composition through detailed chemistry mechanisms. During the engine cycle, QuickSim analyses the physical proper‐ ties of the working fluid without directly calculating chemical reactions. The detailed chemical processes are calculated in advance using Cantera. In Cantera, a wide range of λ, temperature, pressure, residual gas rate and composition combinations are considered. The resulting variables such as Laminar Flame Speed (LFS) and Ignition Delay Time (IDT) for different parameter combinations are reported in look-up tables, which are then read out by the 3D-CFD tool QuickSim during the simulation. The adopted chemical mechanism for the calculation of LFS and IDT, including 324 species and 5739 reactions, was developed by Lawrence Livermore National Laboratory (LLNL). In Fig.-5 are reported all the steps necessary for the virtual fuel development process. In case of methane, the detailed mechanism was reduced to few species, considering a standard composition of the Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 227 227 <?page no="228"?> gas. This is computationally not efficient but allows to use always the same mechanism for different fuel compositions whose limits and behaviour are well known to the authors. Fig. 5: Workflow for the calculation of the laminar flame speed and ignition delay time. Fig. 6 reports the evaluated LFS and IDT for a certain combination of λ, pressure, temperature and residual gas rate, for methane and a standard E10 (10 % ethanol) gasoline. Fig. 6: Laminar flame speed and ignition delay time calculated through detailed chemistry and integrated in QuickSim [10]. As can be seen, methane has a slightly lower LFS, if compared to gasoline, but the higher IDT makes methane effectively more resistant to knock. However, the IDT is evaluated considering full evaporation (in case of E10) with ideal homogenization. In real engine conditions, the higher resistance to knock of methane compared to gasoline is mitigated since the cooling effect of gasoline evaporation reduces the charge temperature, thus increasing the knock resistance in case of gasoline. Though, considering real conditions, methane has still higher knock resistance than gasoline. Furthermore, the knock model in the 3D-CFD tool QuickSim takes into account the following parameters [11]: • Pressure of unburned gas • Temperature of unburned gas • Residual gas content and relative composition (reactive or inert) • Local λ 228 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 228 <?page no="229"?> • Charge motion • Presence of water With respect to these parameters, a spatial and temporal evolution is considered. The described knock model was validated for various engine applications [11] [12] [13], as explained in previous work of the authors [4]. Measurements of the benchmark gasoline engine were used to calibrate the knock model for the development of the new engine through a virtual test bench. 2.2 Computational domain The meshing phase, with the resulting computational domain, is one of the most important parts of a 3D-CFD simulation. As mentioned, the 3D-CFD tool QuickSim allows extending the computational domain to a full-engine configuration and simulating many consequent cycles. Fig.-7 shows the computational domain calculated for every simulation, during the development of the new methane engine. The model reproduces the exact configuration of the test bench, including the damping vessels to faithfully reproduce the fluid exchange process of the single-cylinder engine (crucial for the reproduction of the knock onset). The 3D-CFD tool QuickSim is purposely designed to reduce the computational time needed for CFD calculations related to ICEs, by employing an innovative meshing process. In the software the moving parts are modelled in pre-processing considering only cell vertexes and many movement files are created, related to every crank angle step (Δϑ) of the simulation. With this approach, during the simulated cycle, no remeshing is required. At each Δϑ, the motion file containing the vertex positions of all the moving cells are read so that the cells can be placed in the correct position. Fig. 7: Single-cylinder engine computational domain, reproducing the test bench configuration. Two types of cells are used for QuickSim meshes: structured hexahedral cells and polyhedral cells, of which an example is provided in Fig. 8. Since the mesh motion is vertex-based, all the moving parts are meshed with hexahedral cells, while the rest of the mesh is discretized within polyhedral cells. The combustion chamber of the current engine model has roughly 300000 cells. Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 229 229 <?page no="230"?> Fig. 8: Hexahedral cells used for the combustion chamber (left side) and polyhedral cells used for the fluid domain of the pre-chamber (right side). This approach ensures also high modularity, allowing the test of different geometries in a short time (for example, if the pre-chamber geometry changes, the mesh of the combustion chamber and the motion of the elements have not to be generated again). Tab. 3 summarises all the geometries tested within 3D-CFD simulations in 18 months. Cylinder Head concepts 3 Channels 4 Intake 3 Exhaust Pistons 5 Injectors 3 Injector Positions 4 Valve Profiles 27 Pre-Chambers 10 Operating Points 5 3D-CFD Simulations (Geometry combinations) >200 Tab. 3: Tested geometries through 3D-CFD simulations. 3 Engine and test bench setup In addition to the construction of the single-cylinder engine shown in Fig. 9, the experi‐ mental activity was also carried out by Fraunhofer ICT. 230 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 230 <?page no="231"?> Fig. 9: Assembly of single cylinder research engine. Fig. 10: Fuel mass flow rate and injected quantity per operational cycle of the pre-chamber injector, as function of injection time duration at 16-bar. For this purpose, specific components were integrated into the ICT research engine test bench. Notably, a customized infrastructure for gaseous fuels, including temperature and pressure conditioning units, was implemented by Bright Testing and incorporated into the test bench setup. The fuel mass flow measurement was performed using a Gastron system Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 231 231 <?page no="232"?> from AVL List GmbH, which enables the totalized mass flow measurement, combining both pre-chamber injection and main combustion chamber injection. To determine the fuel mass flow into the pre-chamber via an HDEV 4 piezo injector, the main combustion chamber injection was deactivated at the relevant operating points, thus enabling an isolated measurement of the mass flow through the piezo injector. The results are shown in Fig. 10, which depicts the mass flow rates of methane through the HDEV 4 injector for several duration of injection (DOI). Considering Fig. 10, the piezo-injector exhibited an almost linear response, even with short injection durations, allowing the precise injection of very small quantities. The piezo-injector in the pre-chamber and the solenoid injector in the main combustion chamber were both regulated by a freely programmable rapid prototyping electronic control unit (ECU). One integrated function of this setup is to maintain the global lambda while effectively adjusting either the pre-chamber or main combustion chamber injection. The power amplifiers used were specifically designed for providing capable current profiles for gas injectors. In order to achieve a precise thermodynamic balancing, an additional identical M5 pressure-indicating sensor was integrated into the pre-chamber. Consequently, the additional pressure sensor was also implemented in the indicating measuring system, in conjunction with the in-cylinder pressure sensor. Fig. 11 presents the two manufactured and tested active pre-chambers. Particularly, on the left side of Fig. 11 it is visible the active pre-chamber (from now on defined as APC 1) with cooling system, as well as its lateral electrode and the indicating pressure sensor, while on the right side of Fig. 11 it is illustrated the pre-chamber with no cooling (from now on defined as APC 2). Fig. 11: 3D-CAD models of the PC 500-mm³ with integrated coolant jacket (left) and the PC 750-mm³ (right) without cooling jacket. The installation of two identical pressure sensors enables the comparison of the cylinder and the pre-chamber pressures, thus allowing the calculation of the pressure difference as a thermodynamic criterion for the evaluation of the pre-chamber performances. This 232 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 232 <?page no="233"?> is discussed in greater detail in chapter 5. As illustrated, the integration of a ring cooling system within the pre-chamber housing was only feasible for the smaller pre-chamber volume of 500 mm³ (APC 1). This is in accordance with the results of the 3D-CFD coolant jacket calculations reported in Fig.-3 and Fig.-4. 4 Measurements and validation of the simulations In the current paragraph, the new engine performances are discussed on several operating points. It is important to underline that almost all the 3D-CFD simulations were made months before the engine was produced, so that, a virtually designed engine was actually realised. Many strategies and different hardware variations were tested in the virtual environment and the final configuration of the engine with two different pre-chambers was manufactured. Below a summary of the results for 6 operating points is proposed [5] [6] [4]: • OP1 - 1500 rpm, 1.5 bar IMEP - OP1 represented a big challenge in terms of combustion stability and optimisation of fluid exchange phase. To enhance the engine efficiency also at low loads, the engine is designed to host a fully variable valve train that allows to reduce the maximum intake valve lift to 4.5 mm for OP1. By means of that, the engine is completely de-throttled, reducing pumping losses. The combustion stability is ensured by the direct injection into the pre-chamber, which guarantees close-to-stoichiometric condition into the pre-chamber. For OP1, all the fuel required (ca. 5-mg) is provided by the pre-chamber injector. • OP2 - 1500 rpm, 4 bar IMEP - OP2 is a typical low load point where lean mixtures are required to reduce emission and enhance efficiency. Due to very lean operating conditions targeted for OP2, the earliest pre-chamber design investigations have been carried out simulating this operating point, resulting in an asymmetrical placement of the pre-chamber holes, with the purpose of creating a good homogenization in the pre-chamber and realising a stratification with a slightly rich mixture near the electrode. All these geometry arrangements lead to the possibility of having 4 bar IMEP with an overall λ value of 1.8. • OP3 2000 rpm, 10 bar IMEP - OP3 is the operating point where different injection strategies in the pre-chamber have been tested. Stable operations are achieved through the injection in the pre-chamber up to λ<1.7, while above misfiring problems appeared. This operating point is used in the current paper for the validation of the simulation and the discussion of the results delivered by 3D-CFD simulations. • OP4 2000 rpm, 17 bar IMEP - OP4 represents the operating point where the engine is still capable of running lean (l=1.4) at high load and showing the highest indicated efficiency. This operating point is also used for the validation of the simulation and the discussion of the results delivered by 3D-CFD simulations. • OP5 2000 rpm, 23.5 bar IMEP - OP5 is a full-load operating point with a high tendency to knock, even if a millerization concept was already implemented. With the new methane engine, the high knock resistance of natural gas coupled with millerization strategies ensure lower problem of anomalous combustion. Here the pre-chamber works as passive concept, and it ensures a quick combustion with Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 233 233 <?page no="234"?> stoichiometric conditions and a boost pressure of 2.2 bar (absolute). For OP5 a pressure peak of 135 bar without knocking phenomena is reached, with a center of combustion 15 °CA earlier if compared to the petrol engine. • OP6 5500 rpm, 23 bar - OP6 is the nominal power operating point. The power achieved is 52.7 kW, obtained at lambda 1 with passive prechamber ignition. Higher knock resistance can be reached compared to the benchmark gasoline engine. The power can be achieved with almost ideal centre of combustion (9°CA) and with a pressure peak of 145-bar. For the validation phase, the focus lies on OP3, being interesting during homologation cycles for light-duty commercial vehicles, testing several strategies and hardware mod‐ ification. As mentioned before, two pre-chamber configurations have been tested and measured, a smaller pre-chamber with a volume of 500 mm 3 , and a bigger pre-chamber with a volume of 750 mm 3 . In Fig. 12 are reported the fluid volumes for both the pre-chamber configurations the smaller (APC1, left side) and the bigger one (APC2, right side). Fig. 12: Pre-chamber fluid volumes used for test bench experiments. 4.1 Validation of the simulation: OP3---stoichiometric In the current paragraph, the results at OP3 and stoichiometric mixture are highlighted. The comparison between simulations and experiments are realized considering that the simulation results were delivered before the engine manufacturing and testing. This means that the simulations presented are not calibrated with respect to the experiments, for the sake of highlighting the prediction capabilities of the CFD model and discussing eventually some improvements of them. The first measurements consider stoichiometric conditions and the small pre-chamber configuration (APC1). In Tab. 4 are reported the main values measured at the test bench in comparison with the 3D-CFD predictions calculated in the same conditions and with the same hardware. 234 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 234 <?page no="235"?> Test Bench 3D-CFD Simulation rpm 2000 2000 IMEP, bar 10.0 10.4 p2, bar 1.24 1.24 p3, bar 1.22 1.24 Ign. point, °CA -13 -13 PC injection, mg - - Ind. efficiency, % 39 38.6 Air consumption, kg/ h 26 27 Fuel Consumption, kg/ h 1.52 1.51 Max pressure, bar 73.6 73.0 10-% mfb, °CA 1 1 50-% mfb, °CA 8 12 90-% mfb, °CA 20 24 10-90% mfb, °CA 20 24 λ at ignition point, - 1 1 SOI-DI, °CA -160 -160 DOI-DI, °CA 20 21 SOI-PC, °CA - - DOI-PC, °CA - - Combustion eff., % - 93 COV IMEP, % 0.95 - Tab. 4: Comparison of the experiments with 3D-CFD simulation results for stoichiometric conditions at OP3. In this case, the engine was operated in stoichiometric condition with a Δp almost 0 between the manifolds (Δp=p 2 -p 3 ). As can be seen from Tab. 4, the simulation shows an overall good agreement with the experiments, whereas a noticeable difference lies in the combustion process. The first part of the combustion is very well reproduced having for both, the simulation and the experiments, the same value of 10-% mfb (mass fraction burned). From 10 % mfb to 50 % mfb, the 3D-CFD simulation shows a slower combustion event, ending up in a different value of the combustion centre (50 % mfb), 8 °CA (FTDC = 0 °CA) for the experiments compared to the 12 °CA for the 3D-CFD calculations. Considering then the second half of the combustion process, the behaviour is well reproduced again, with a Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 235 235 <?page no="236"?> mfb 50 %-90% of 12 °CA for both the experiments and the calculations. Fig. 13 shows the in-cylinder pressure curves. Fig. 13: In-cylinder pressure curves measured at the test bench and averaged over 200 cycles (colour magenta) and simulated by means of 3D-CFD simulation (colour blue) [7]. As it clearly depicted in Fig. 13, the delay in the first part of the combustion of the simulation determines also a delay for the pressure peak, resulting in a slightly different expansion curve. Nevertheless, the gradient of the expansion curves appears to be close for the experiments and the simulations. IMEP and peak pressure are well reproduced (see Tab. 4). Like in the reality, in the simulation the reached load of 10 bar IMEP is the result of the combustion process, and performance parameter such as air consumption. The air consumption matches perfectly by applying the test bench boundary conditions. In addition, having a look to Fig. 13, the simulated pressure curve has a good fitting with the experiments, except for a lower pressure gradient immediately after FTDC for the simulation, which is the results of lower flame speed propagation or generally of higher wall heat transfer in the pre-chamber than in the reality. Considering the expansion curve, with reference again to Fig. 13, the simulation is characterized by a slower drop down of the pressure which is an indication that it is underestimating the heat rejection to the wall with respect to the measurements. Finally, the two effects of overestimation of the heat rejection in the pre-chamber and underestimation of the wall heat transfer during the expansion are compensating and the simulation achieves very close IMEP compared to experiments. Potentially, these inaccuracies can be corrected running conjugated heat 236 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 236 <?page no="237"?> transfer simulation to provide the simulation with more faithful wall temperatures (the wall temperature provided to the current simulation are constant and estimated from experience on other engines). The reason for the slower 10-50% mass fraction burned in the simulation and the lower pressure gradient right after FTDC, can be also addressed to an underestimation of the mixture homogenization. An indication that the simulation is underestimating the mixture homogenization is the slightly low combustion efficiency which reaches only 91 % in the first simulation (see Tab. 4). The combustion efficiency is a direct evaluation of the homogenization of the mixture and, following the experience on methane engine [14], this value seems to be too low for such a load point. Indeed, methane if injected early and with sufficient high injection pressure (like 26 bar for the current engine), in combination with tumble channels should have a good diffusivity in the air, thus generating homogeneous mixture. One additional reason to support this theory (underestimation of the homogenization of the mixture) is that the COV of the IMEP is 0.9 % (see again Tab. 4) and it is relatively low for such load and speed. The relatively low COV of IMEP cannot be realized with a not sufficient mixture homogenization. For reproducing the mixture formation more accurately, an improvement of the injector model could be realized if more data of the injector would have been available, or some optical measurements could be have performed. Considering the available data, the differences in the combustion propagation at the early beginning of the process could be also corrected by analysing the detailed chemistry calculation of the laminar flame speed with respect to residual gas and mixture conditions. The propagation of the laminar flame can be set to be less affected by residual gas presence in the cylinder (5 % residual gas is calculated to be in the cylinder at OP3, while the pre-chamber has a value of 8 %). One possible cause for the underestimation of the laminar flame speed in certain condition could be addressed to the detailed kinetic mechanism used for modelling the fuel (LLNL with simplified surrogate for methane). Resuming, the differences in the first part of the combustion process must be addressed to the fuel model (sensitivity of the laminar flame speed calculation to residual gas rate) or to inaccuracy of the DI-injector model for lack of data about the DI-Injector generating an underestimation of the mixture homogenization. For further support to the previous considerations, the 3D-CFD simulation of the original Ford Ecoboost engine with gasoline injection and well-defined DI-injector delivered more accurate pressure curves with respect to the experiments provided by Ford Werke GmbH, as shown in [10]. 4.2 Validation of the simulation: OP3 - lean operations After the validation in stoichiometric condition, it was also important to test the reliability of the models at leaner mixture, being leaner conditions more critical for modelling ignition, flame propagation and heat exchange during the engine process. Following this path, the second validation has been focused on the same engine operating point, 2000 rpm, 10 bar, but run at λ 1.4, 1.5 and 1.6. Tab. 5 reports the comparison between simulation and for different engine parameters, while in Fig. 14 are shown the in-cylinder pressure curves for the three operating conditions considered, comparing test bench measurements (TB) with 3D-CFD results. Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 237 237 <?page no="238"?> TB λ = 1.4 CFD λ = 1.4 TB λ = 1.5 CFD λ = 1.5 TB λ = 1.6 CFD λ = 1.6 rpm 2000 2000 2000 2000 2000 2000 IMEP, bar 10.0 10.2 10.0 10.1 9.7 9.7 p2, bar 1.6 1.6 1.7 1.7 1.9 1.9 p3, bar 1.6 1.6 1.7 1.7 1.9 1.9 IP, °CA -17.5 -17.5 -21 -21 -22 -22 PC inj., mg - 0.55 - 0.55 - 0.55 Ind. eff., % 42 41.5 42 41.4 40 39.8 Air cons., kg/ h 33 33 36 37 40 39 Fuel cons., kg/ h 1.4 1.4 1.4 1.4 1.4 1.4 Max press., bar 78 75 81 79 81 84 10-% mfb, °CA -2 -3 -3 -4 -2 -5 50-% mfb, °CA 8 13 8 13 10 13 90-% mfb, °CA 23 34 24 38 30 43 10-90% mfb, °CA 24 37 27 42 32 48 λ at IP, - 1.4 1.4 1.5 1.5 1.6 1.6 SOI-DI, °CA -240 -240 -240 -240 -240 -240 DOI-DI, °CA 15 15 15 15 15 15 SOI-PC, °CA -360 -360 -360 -360 -360 -360 DOI-PC, °CA 10 10 10 10 10 10 Tab. 5: Comparison of the experimental measures with 3D-CFD simulation results for different lean mixtures at OP3. 238 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 238 <?page no="239"?> Fig. 14: In-Cylinder pressure curves for λ 1.4, 1.5 and 1.6. As can be seen in Tab. 5, for the three cases the same injection strategy is used, both for pre-chamber injection and for direct injection. The different lambda is achieved through a gradual boost pressure increase, by keeping a constant delta pressure between intake and exhaust, across the engine. The lean stoichiometric operations are achieved with a boost pressure of 1.6, 1.7 and 1.9 bar respectively. As depicted in Fig. 14, also for a leaner lambda, 3D-CFD investigations can reach a very good agreement with the experiments. In the combustion and expansion phases the same behaviour outlined for the lambda 1 case (see chapter 4.1) can be observed. The very first part of the combustion is well reproduced, since the 3D-CFD and the experiments show a very close 10 % burned mass fraction. In the middle part of the combustion (10 % mfb - 50% mfb) an average delay of 5 °CA results from the 3D-CFD evaluation, leading to a delayed combustion centre resulting in a delayed pressure peak. In this case is interesting to see how the 3D-CFD evaluates the 90 % mfb more than 10 °CA later compared to the measurements, resulting in a longer combustion event in the simulation. It is critical to compare the 90 % mfb between measurements and simulation because of the different methodology to evaluate it. At the test bench, the 90 % mfb calculation results from the in-cylinder pressure measurements and a model of heat transfer, thus deriving a burn rate. When the burn rate is close to 90 % mfb, the curve becomes very flat and small changes in the burned mass percentage generate high differences in the corresponding crank angle at which the burn rate is detected and resulting in the 90 % mfb. In addition, the heat transfer model comes close to its limit of application since the low temperature gradient close to the end of combustion. In the simulation the Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 239 239 <?page no="240"?> 90%mfb is evaluated exactly as the instant corresponding to the transformation of 90 % of the fresh fuel mass in burned mass. In many cases it is worth to compare simulation and test bench results at 80 % mfb instead of 90 % mfb where the burn rate evaluation through measurements can be more accurate and reproducible. In addition, in case of lean operation the heat transfer model used for the evaluation of the burn rate can come to its limit as well and being not anymore, a good prediction. After these considerations anyway, a review of the heat transfer model (a 3D phenomenological model similar to the model of Woschni [8]) in the simulation is ongoing, especially to study its sensitivity to different lean combustion processes. 5 Discussion of the results Within the project many strategies have been tested to exploit all the potential of this engine configuration. Particularly attention has been given to the active pre-chamber ignition system, to understand which modification can be effective. The active injection into the pre-chamber affects the lambda value near the electrode and the residual gas into the pre-chamber, two parameters that greatly influence the ignition of the mixture. Decoupling the mixture condition of the main combustion chamber, from the ones of the pre-chamber volume allows global leaner operating point to be performed, under the condition of good ignitability boundaries near the electrode. Following this path, in the next section are analysed two different strategies: a variation of the start of injection (SOI) in the pre-chamber, and a variation of its duration (DOI). The SOI variation, in the current paper is discussed in case of the engine equipped with the big pre-chamber (APC2, 750-mm 3 ), while the DOI variation is performed with both the big and the small pre-chamber. 5.1 SOI variation with APC2 (750 mm 3 ) The first analysis is carried out considering different SOIs in the pre-chamber to understand which timing performs better for the ignitability of the mixture. The investigation is carried out considering OP3 (2000 rpm - 10 bar IMEP), operated at λ 1.4, since in these conditions a low coefficient of variation (COV) of IMEP (1.39 %) allows to have consistent results for the comparison of the different SOIs. For all the SOIs considered, the same amount of fuel is injected by both the pre-chamber and the direct injector, with a duration of injection of 15 °CA for the direct injector and 20 °CA for the pre-chamber injector (DOI PC = 1600 µs). The analysis is performed parallelly at the test bench and by means of 3D-CFD simulations, with the engine equipped with APC2 (PC volume 750 mm 3 , Fig. 12). As can be seen from Tab.-6, three PC-SOIs were analysed: a very early SOI, -355 °CA (FTDC = 0 °CA), and then two delayed SOIs, -328 and -275 °CA, which are respectively named SOI 1, SOI 2 and SOI 3 as follows. 240 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 240 <?page no="241"?> TB SOI 1 CFD SOI 1 TB SOI 2 CFD SOI 2 TB SOI 3 CFD SOI 3 rpm 2000 2000 2000 2000 2000 2000 IMEP, bar 10 10 10 10 10 10 p2, bar 1.6 1.6 1.6 1.6 1.6 1.6 p3, bar 1.6 1.6 1.6 1.6 1.6 1.6 IP, °CA -12 -12 -11.5 -11.5 -12 -12 PC inj., mg - 1.3 - 1.3 - 1.3 Ind. eff., % 42.3 42.1 42.0 41.8 41.9 41.3 Air cons., kg/ h 34 34 35 34 35 34 Fuel cons., kg/ h 1.4 1.4 1.4 1.4 1.4 1.4 Max press., bar 79 79 79 77 79 79 10-% mfb, °CA 0 -1 0 0 0 -1 50-% mfb, °CA 8 11 8 12 8 12 90-% mfb, °CA 23 27 23 28 22 27 10-90% mfb, °CA 22 27 22 28 22 28 λ at IP, - 1.4 1.4 1.4 1.4 1.4 1.4 SOI-DI, °CA -240 -240 -240 -240 -240 -240 DOI-DI, °CA 15 15 15 15 15 15 SOI-PC, °CA -355 -355 -328 -328 -275 -275 DOI-PC, °CA 20 20 20 20 20 20 Tab. 6: Comparison of the experiments with 3D-CFD simulations for SOI sweep. Considering the in-cylinder conditions, the cycle-averaged values are the same for all the SOIs considered without noticeable differences in terms of performance and efficiency. For these measurements and the respective simulations, a difference in the centre of combustion is still present (see chapter 4), but the model seems to predict more accurately the combustion duration in this case. To understand better which parameter are affected by the SOI variation, the focus must be taken on the pre-chamber volume. In Tab. 7 are reported some parameters referred to the pre-chamber volume provided by the simulation and compared to the measured COV of IMEP. Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 241 241 <?page no="242"?> SOI 1 SOI 2 SOI 3 PC-λ at IP, - 0.90 0.85 0.77 PC-TKE at IP, m 2 / s 2 47 48 48 PC-Residual gas at IP, % 7 7 8 Combustion eff., % 95 96 95 COV IMEP, % 1.42 1.40 1.57 Tab. 7: Pre-chamber parameters for the three evaluated SOIs and comparison with the measured COV of IMEP. The major difference lies into the λ value at the electrode of the spark plug which becomes richer as the SOI is delayed, having the injected mass less time to be pulled out the APC before the fuel coming from the main chamber starts to be pushed into the APC by the piston during the compression stroke. Essentially, the SOI variation does not highlight major changes in the engine performance, since the indicated efficiency is comparable for all the three cases. On the other side, it is interesting that there is an optimal SOI with respect to the parameter COV of IMEP and it is worth to correlate it with the mixture in the pre-chamber provided by the simulation. The lower COV of IMEP for SOI 2 is the indication that at the electrode a lambda 0.85 is a good compromise in terms of ignition of the mixture. This consideration provides also a target for the future pre-chamber geometry development in terms of mixture formation at the electrode. Following this path, SOI 3 shows also a slightly higher residual gas, meaning that when the APC injection is performed too late, the scavenging of the PC volume is weaker, as can be seen in Fig.-15. The case with SOI 2 represents a good compromise between scavenging of the pre-chamber from residual gas of the previous cycle and slightly rich mixture at the electrode (λ~0.85). As result, SOI 2 is chosen as the best option for the PC injection strategy, having a slightly richer λ, which is a good aspect for the ignitability of the mixture. In the upper part of Fig. 16 is shown the λ distribution at IP, while in the lower part is reported the residual gas percentage at -180°CA for all three strategies (FTDC = 0°CA). 242 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 242 <?page no="243"?> Fig. 15: Residual gas evolution during the engine cycle for cylinder, prechamber volume and volume near the electrode for the three SOI considered. Fig. 16: Lambda distribution at ignition point (upper side) and residual gas percentage at -180 °CA (lower part) for different SOI of the APC injector, with APC2 configuration. The frames in Fig. 16 depict 2 important instants during the engine cycle for the mixture formation and the scavenging of the pre-chamber. In the lower part of Fig.-16 at -180°CA the injections for all the three SOIs are terminated and the later the injection the higher the residual gas in the pre-chamber. As consequence of the scavenging rate of the pre-chamber the mass in Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 243 243 <?page no="244"?> the pre-chamber is different and it has different pressure and temperature conditions. These conditions are at the origin of the differences in the mixture formation at ignition point (upper part of Fig. 16). It is not through the amount of the injected mass in the pre-chamber that is possible to control the mixture (for all the three SOIs the same fuel mass is injected into the PC), rather through the scavenging of the residual gas and its timing. 5.2 DOI variation with APC2 (750 mm 3 ) Keeping SOI 2 for the injection strategy, a further investigation has been made for the duration of the injection into the PC. Keeping the SOI point for the pre-chamber fixed at -328 °CA, two energizing times for the PC injector were considered: • DOI 1, 1600 μs energizing time, 20 °CA at 2000 rpm • DOI 2, 4000 μs energizing time, 48 °CA at 2000 rpm In Tab. 8 are shown the data related to the 3D-CFD results and measures for DOI-1 and DOI-2. - TB DOI 1 CFD DOI 1 TB DOI 2 CFD DOI 2 rpm 2000 2000 2000 2000 IMEP, bar 10.1 10.3 10.0 10.3 p2, bar 1.6 1.6 1.6 1.6 p3, bar 1.6 1.6 1.6 1.6 IP, °CA -11.5 -11.5 -11.5 -11.5 PC inj., mg - 1.3 - 3.6 Ind. eff., % 42 42.5 42 41 Air cons., kg/ h 35 34 35 34 Fuel cons., kg/ h 1.4 1.4 1.4 1.5 Max press., bar 79 77 79 76 10-% mfb, °CA 0 0 0 0 50-% mfb, °CA 8 12 8 13 90-% mfb, °CA 23 28 23 28 10-90% mfb, °CA 22 28 23 27 λ at IP, - 1.4 1.4 1.4 1.4 SOI-DI, °CA -240 -240 -240 -240 DOI-DI, °CA 15 15 15 15 SOI-PC, °CA -328 -328 -328 -328 DOI-PC, °CA 20 20 48 48 PC-λ at IP, - - 0.85 - 0.77 PC-TKE at IP, m 2 / s 2 - 48 - 48 PC-Residual Gas at IP, % - 7 - 7 Combustion. eff., % - 96 - 94.5 Tab. 8: Comparison of the experimental measures with 3D-CFD simulation results for DOI sweep analysis. 244 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 244 <?page no="245"?> As noticeable from Tab. 8, the longer the DOI in the pre-chamber the richer is the lambda in the PC volume at the ignition point (IP). The amount of residual gas at IP is the same for both strategies, nevertheless if the focus is brought on the evolution of the residual gas during the cycle (upper side of Fig. 17), it is clear that the scavenging phase of the PC is greatly affected by the DOI. Fig. 17: Residual gas evolution during the cycle (upper side) and injection strategy (lower side). Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 245 245 <?page no="246"?> Since the PC DOI is limited by the ignitability conditions in the pre-chamber volume and since the PC injector under investigation, operates at 16-bar, only small quantity of fuel can be injected (lower side of Fig. 17) and it is possible to state that the main advantage of having an active pre-chamber system is the possibility to influence the scavenging of the pre-chamber volume. During the compression stroke, indeed, when the mixture from the main chamber is pushed into the APC, the stoichiometry near the electrode is strictly dependant on the DOI. A better scavenging of the pre-chamber volume leads to minor percentage of residual gas in the pre-chamber, affecting the quantity of mixture which can be pushed back into the APC volume. As it can be seen in Fig. 18, the mixture formations in the pre-chamber are very similar at IP for both the DOIs, highlighting that the injected amount of fuel in the pre-chamber cannot determine directly the stoichiometry of the mixture in the pre-chamber (with DOI 2 twice the fuel mass is injected in the pre-chamber compared to DOI 1). Fig. 18: Lambda distribution at IP obtained with DOI 1 (upper side) and DOI 2 (lower side). 246 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 246 <?page no="247"?> The SOI and DOI variations at OP3 do not lead to substantial differences in the engine performances since this load point could be achieved with a valve profile and strategy to reduce residual gas in the main combustion chamber. The impact of the SOI/ DOI strategy on the scavenging of the pre-chamber resulted crucial in other operating points at IMEP < 2 bar where the amount of residual gas in the pre-chamber achieves 20 %. These results will be discussed in a later work of the authors. 5.2.1 DOI variation and pressure development for APC1 (500 mm 3 ) Another big advantage of using a pre-chamber ignition system is given by the turbulence generated by the jets expelled within the pre-chamber ducts toward the main chamber. The turbulence generated in this phase helps to speed up the combustion event, even if the overall air-to-fuel ratio in cylinder is very lean. A parameter that greatly influences the turbulence generated by the jets is the difference in terms of pressure between the pre-chamber and the main chamber during the ejection of the jets. To understand if the injection strategy can impact this parameter, a further analysis has been made. Another DOI variation has been carried out with the engine equipped with APC1 (APC volume 500 mm 3 , Fig. 12) parallelly at the test bench and within 3D-CFD simulations. Eight different energizing times have been considered for the TB analysis and then three of them were further investigate within 3D-CFD simulations. As can be seen from Fig.-19, the pressure difference between the pre-chamber and the main chamber decreases with the rise of the DOI in the APC. As already shown in this work, as the APC DOI is extended, the ignitability conditions in the APC worsen, leading to a non-homogeneous mixture at the ignition point (λ << 1). In these conditions, the combustion process is less efficient and determines a lower pressure in the PC, which leads to lower pressure difference between PC and main chamber. The Δp between the two chambers is strictly correlated to the efficiency of the turbulent jet propagation, thus a bigger Δp can generate a more efficient and quick combustion. Fig. 19: Measurements to evaluate the pressure difference between the pre-chamber and the main chamber with the engine equipped with APC1 (500-mm 3 Volume). Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 247 247 <?page no="248"?> From the experimental data of Fig. 19, the higher Δp is obtained with an energizing time between 800 and 1100 μs. Three energizing times have been chosen then for a further 3D-CFD investigation with the small pre-chamber (APC1): 1100 μs (DOI 1), 4000 μs (DOI 2) and 5000 μs (DOI 3), and the results in terms of pre-chamber flow field are reported in Tab.-9, in case the engine runs at λ =1.4. - 1100 μs 4000 μs 5000 μs PC-λ at IP, - 0.64 0.38 0.27 PC-TKE at IP, m 2 / s 2 31 32 30 PC-Residual gas at IP, % 7 6 6 Combustion. eff., % 95 89 87 Tab. 9: Pre-Chamber main parameters for the three DOI investigated, in APC1 configuration (500-mm 3 Volume). From the upper diagram of Fig. 20 is clear how DOI 1 (1100 μs) generates the highest pressure difference within this analysis since the DOIs of 4000 µs and 5000 µs produce such a locally rich mixture (Tab. 9) that the electrode is at the limit of not-ignitable conditions. Considering the combustion efficiency calculated for each case, by a DOI of 1100 µs the combustion is fast, and the global homogenization of the main combustion chamber contributes to a stable combustion event with global lean operations. For DOI 1 the spark plug can ignite the mixture with sufficient power to spread the jets towards the combustion chamber, consequently leading to a higher pressure peak. Fig. 20 shows a comparison of the pressure measured in the pre-chamber and the simulated one. The upper diagram in Fig. 20 presents the measured and the simulated delta pressure between pre-chamber and cylinder. The simulations can reproduce the trends and the influence of the duration of injection with respect to the pressure development, but with a delay of 4-5 °CA, when considering the peak pressure. This can be addressed to inaccuracies in the choice of the pre-chamber wall temperature in the simulations. 248 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 248 <?page no="249"?> Fig. 20: Pre-chamber pressure (lower side) and delta pressure between pre-chamber (APC1, 500m 3 Volume) and main chamber (upper side) evaluated with 3D-CFD simulation. Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 249 249 <?page no="250"?> From the analysis carried out using APC2 (paragraph 5.2) it was clearly depicted a good behaviour of the big pre-chamber in lean operation conditions. To compare the two pre-chamber layouts (different volume) in terms of generated pressure difference, the results shown for the DOI variation with APC1 can be compared with the results shown for the DOI variation with APC2 as depicted in Fig. 21. APC2 generates a higher Δp (13 bar for APC2), with respect to the small one which, in the best case, only provides 4 bar of pressure difference. A balance between volume of the pre-chamber and impulse of the jets (strictly linked to the delta pressure produced by the pre-chamber) has to be found to keep low the PC wall heat transfer, but at the same time to generate enough powerful jets. Based on the experience of the authors, for the actual engine characteristics (bore, stroke, lean operation and max pressure), a delta pressure above 7/ 8 bar has to be provided by the PC to achieve effective benefit in terms of speeding up the flame propagation with respect to a conventional spark plug. Therefore, APC1 has a too small volume or generally not a good balance between volume and number of holes and their dimension. It is possible to see as well from Fig. 21, that the simulation predicts with higher accuracy the pressure development in case of the big PC. It can be assumed that the heat transfer is better reproduced for the big PC providing an indication on the choice of the pre-chamber wall temperature with respect to the small pre-chamber. Fig. 21: Delta pressure between the two investigated pre-chamber configurations and main chamber evaluated with 3D-CFD simulation (continuous lines) and experiments (dashed lines). 250 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 250 <?page no="251"?> In Fig. 22, a comparison between the turbulent kinetic energy (TKE) for APC1 and APC2 with the same DOI (4000 µs) is reported at FTDC. Considering again Tab. 8 and Tab. 9, the condition at the electrode of both the PC can be compared at ignition point in case of DOI = 4000 µs. Particularly APC1 (see Tab. 9) has at the electrode λ = 0.38, TKE = 32 m 2 / s 2 and residual gas rate of 6 %, while APC2 (Tab. 8) has a λ = 0.77, TKE = 48 m 2 / s 2 and residual gas rate of 7 %. Therefore, the mixture conditions and the turbulence at the electrode of APC2 are favourable to generate a high delta pressure and the balance between the momentum of the jets and the loss in heat transfer through the PC walls is more favourable than in APC1. This results in the production of jets that can propagate faster in the main combustion chamber and generate higher turbulence for a rapid combustion process as shown in Fig. 22. Fig. 22: Turbulent kinetic energy at FTDC for small pre-chamber configuration (upper side) and big pre-chamber configuration (lower side) with 4000 μs duration of injection in the APC. Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 251 251 <?page no="252"?> 5.3 Load variation at Lambda 1.4 for APC1 In view of investigating the maximum engine indicated efficiency, a load variation has been run at the test bench with APC1, which, as discovered in the previous paragraph, is performing less effectively compared to APC2. In a future work of the authors, it will be discussed a load variation also in case of APC2. Nevertheless, by exploiting the analysis realized in the previous chapters, a SOI of -325 °CA and a DOI of 1100 µs for APC1 could be identified to run the engine at λ = 1.4 with stable combustion. The main objective of this measurements campaign is to achieve the highest possible indicated efficiency and using the measurements to again validate the respective 3D-CFD simulation. Fig. 23 presents the load variation at 2000-rpm, operating the pre-chamber as active. Fig. 23: Load variation for active pre-chamber injection operation considering APC1 at 2000-rpm and lambda 1.4 [7]. Few considerations can be driven starting from Fig. 23. The engine is running stably at λ = 1 showing at 17 bar IMEP a COV of IMEP below 1 %, thanks to the injection in the pre-chamber and the scavenging produced as explained in chapter 4 and 5. Thanks to the potential of methane in terms of avoiding anomalous combustions (such as knocking) it was possible to raise the boost pressure to 2.6 bar, without the occurrence of knocking phenomena. The injection strategy in the PC is particularly effective starting from 14 bar IMEP, where keeping the global stoichiometry of the mixture unchanged (λ = 1.4), the engine can operate with lower ignition delay (see parameter IP-mfb5 in Fig. 23) and lower COV of IMEP. Critical are anyway the emission production especially for NOx and THC 252 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 252 <?page no="253"?> emissions as can be seen in Fig. 23. Running the engine at l = 1.4 is not enough to cut down NOx emissions, but potentially the engine can run with further enleanment. The high HC emission can be explained by looking into the flow field through a 3D-CFD simulation. Particularly Tab. 11. reports the comparison between test bench and simulated parameters at 2000 rpm and 17 bar IMEP (λ=1.4). As said before, the model overpredicts the center of combustion and this trend is present also at this load point, especially in case of APC1 (see chapter 4 and 5). The other engine parameters are quite well predicted (see Tab. 11), without any calibration effort. - TB CFD rpm 2000 2000 IMEP, bar 17 17 p2, bar 2.6 2.6 p3, bar 2.6 2.6 IP, °CA -18 -18 PC inj., mg - 0.9 Ind. eff., % 43.8 43.4 Air cons., kg/ h 56.2 56.3 Fuel cons., kg/ h 2.3 2.3 Max press., bar 130 126 10-% mfb, °CA -1.2 0 50-% mfb, °CA 8.4 13 90-% mfb, °CA 22 28 10-90% mfb, °CA 23 28 λ at IP, - 1.4 1.4 SOI-DI, °CA -235 -235 DOI-DI, °CA 23 23 SOI-PC, °CA -325 -325 DOI-PC, °CA 13 14 PC-λ at IP, - - 1.12 PC-TKE at IP, m 2 / s 2 - 31 PC-Residual Gas at IP, % - 7 Combustion. eff., % - 97 Tab. 11: Comparison of the experimental measures with 3D-CFD simulation results for OP4 (max efficiency case). Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 253 253 <?page no="254"?> The simulation highlights 97 % combustion efficiency which is a quite high value if looking at the fuel distribution in Fig. 24. An almost stoichiometric mixture is present close to the electrode of the pre-chamber and the charge looks stratified close to the electrode, while the main combustion chamber is very lean and present an accumulation of fuel in the injector nose. Since the engine is running very stably (COV IMEP < 1 %) the effect of PC and the impulse of the jets must work effectively to produce a stable combustion with such an uneven mixture in the main combustion chamber. APC1 produces 5 bar delta pressure with respect to the pressure rise in the main combustion chamber which seems to be enough at this load point to trigger a fast combustion event. The reasons for the high THC emission can be addressed on the fuel accumulation at the injector nose as can be seen in Fig. 24. Unfortunately, the injector used as DI-Injector is an injector with wide spray angle more suitable for a central position in the main combustion chamber. Therefore, in the future work a new injector has to be identified to reduce emissions. Nevertheless, the engine could achieve an indicated efficiency of almost 44 %. Based on what assessed in chapter 5.2.1, further investigations will be made with the engine equipped with the big pre-chamber, to further enhance the indicated efficiency of the engine with a more convenient APC layout. Fig.-24: Lambda distribution at ignition point at 2000-rpm 17-bar IMEP in case of APC1 and l = 1.4. 254 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 254 <?page no="255"?> 6 Conclusions The current work resumes the analysis and the development of a new single-cylinder engine, operating with methane direct injection, with an active pre-chamber system. The geometry of the engine is the results of 20 months simulation and construction work realized by the partners of the project. This paper focuses on the validation of the 3D-CFD simulations, led by FKFS, well before the construction and on the testing of the real hardware by Fraunhofer ICT. In addition to the realization of an engine of high efficiency, scope of the project was to test the predictivity of the simulation models for an effective virtual development and the identification of weakness to be improved. Apart from some inaccuracies on the center of combustion, the simulation showed very accurate estimation of the overall engine behaviour, especially considering the pre-chamber behaviour. The predicted pressure development in the pre-chamber and its effectiveness for the speed up of the combustion process is aligned with the measurements being the simulation able to reproduce correctly the impact of different injection strategies in the pre-chamber. Nevertheless, a refinement of some models can improve the accuracy of the simulations. The differences in the first part of the combustion process can be addressed to the fuel model or to inaccuracies of the DI-injector model. A review of the laminar flame speed calculation with respect to residual gas rate concentration has already corrected partially the error in the center of combustion. It is also clear the underestimation of the mixture homogenization. To improve it more details about the DI-injector are needed. As main findings concerning the pre-chamber behaviour, the follow consideration can be driven: • The SOI and DOI variations at OP3 do not lead to substantial differences in the engine performances. The impact of the SOI/ DOI strategy on the scavenging of the pre-chamber results crucial in other operating points at IMEP < 2 bar where the amount of residual gas in the pre-chamber achieves 20 %. These results will be discussed in a later work of the authors. • The injection in the pre-chamber influences the scavenging of the pre-chamber and modifies the filling of the pre-chamber with fresh mixture during the compression stroke. There is no possibility to control directly through the injection strategy, the stoichiometry of the mixture at the electrode. • A volume of 750 mm 3 is a good compromise between impulse of the jets and wall heat transfer for an engine displacement of 500-cm 3 , generating at OP3 up to 13.5-bar delta pressure between PC and main combustion chamber. The PC with 500 mm 3 is less effective in the generation of the momentum of the jets. • The direct injector with a wide cone angle [10] is not suitable as lateral injector, producing an accumulation of methane in the injector nose, thus generating high HC emissions. Design Optimization of a CNG-Single-Cylinder Engine for Lean Mixture Operation 255 255 <?page no="256"?> 7 Bibliography [1] energy.ec.europa.eu, “BIOMETHANE FICHE---Germany (2021),” [Online]. [2] Soltic, P., Hilfiker, T., Hutter, R., Haenggi, S., “Experimental comparison of efficiency and emission levels of four-cylinder lean-burn passenger car-sized CNG engines with different ignition concepts. Combustion Engines. 2018, 176(1), 27-35.DOI: 10.19206/ CE-2019-104”. [3] Kramer, U., Lorenz, T., Hofmann, C., Ruhland, H. et al., “, “Methane Number Effect on the Efficiency of a Downsized, Dedicated, High Performance Compressed Natural Gas (CNG) Direct Injection Engine,” SAE Technical Paper 2017-01-0776, 2017”. [4] Vacca, A. et al. (2022)., Virtual Development of a New 3-Cylinder Natural Gas Engine with Active Pre-chamber, In: Bargende, M., Reuss, HC., Wagner, A. (eds) 22. Internationales Stuttgarter Symposium. Proceedings. Springer Vieweg, Wiesbaden. https: / / doi.org/ 10.1007/ 978-3-658-37009 -1_31. [5] Bucherer, S., Rothe, P., Kraljevic, I., Kollmeier, H.-P. et al., “Design of an Additive Manufactured Natural Gas Engine with Thermally Conditioned Active Prechamber,” SAE Technical Paper 2022-37-0001, 2022, doi: 10.4271/ 2022-37-0001. [6] Vacca A., Chiodi M., Casal Kulzer A., Bargende M., Bucherer S., Rothe P., Kraljevic I., Kollmeier H., Breuer A., Ruhland H., “Study of Different Active Pre-chamber Ignition Layouts for Lean Operating Gas Engine using 3D-CFD Simulations”. [7] Bucherer, S. “Use of additive additive manufacturing for concept, design and validation of a singlecylinder methane engine with conditioned active pre-chamber spark plug”, PhD thesis, University of Stuttgart, submitted 2024 [8] Chiodi, M., An Innovative 3D-CFD-Approach towards Virtual Development of Internal Combus‐ tion Engines”, PhD thesis, University of Stuttgart, 2010. [9] Vacca, A., Cupo, F., Chiodi, M., Bargende, M. et al., “The Virtual Engine Development for Enhancing the Compression Ratio of DISI-Engines Combining Water Injection, Turbulence Increase and Miller Strategy,”, SAE Technical Paper 2020-37-0010, 2020, doi: 10.4271/ 2020-37-0010. [10] Vacca. A, Chiodi M., Kulzer A. C., Bargende M., Bucherer S., Rothe P., Kraljevic I., Kollmeier H., “Study of Different Active Pre-chamber Ignition Layouts for Lean Operating Gas Engines using 3D-CFD Simulations”. [11] Cupo, F., “Modeling of Real Fuels and Knock Occurrence for an Effective 3D-CFD Virtual Engine Development,”, Ph.D. thesis, University of Stuttgart, 2021. [12] Vacca, A., Cupo, F., Chiodi, M., Bargende, M. et al., “The Virtual Engine Development for Enhancing the Compression Ratio of DISI-Engines Combining Water Injection, Turbulence Increase and Miller Strategy,” SAE Technical Paper 2020-37-0010, 2020, doi: 10.4271/ 2020-37-0010. [13] Jonas Villforth, et al., Methods for the Evaluation of eFuel Potentials on the Combustion and Emission Behavior of DISI Engines, 9th Int. Symposium on Development Methodology, 2021 Wiesbaden, Germany. [14] Chiodi M., Berner H. J., Bargende M., Investigation on different Injection Strategies in a Direct-Injected Turbocharged CNG-Engine, SAE International, 2006. 256 Vacca, Tortorella, Chiodi, Bucherer, Kulzer, Sobek, Rothe, Kraljevic, Kollmeier, Breuer, Helmut 256 <?page no="257"?> 1 IAV GmbH, Stollberg, Germany, Thomas.emmrich@iav.de Assessment of pre-ignition phenomena by thermodynamically approach Dr.-Ing. Thomas Emmrich 1 Abstract Regardless of the used fuel (mineral, synthetic, alternative) and the size of the engine, there is a development trend towards significantly higher efficiencies with significantly increasing compression ratios in gasoline engines. In principle, this leads - especially in combination with high mean pressure at low engine speed - to a higher pre-ignition tendency, well known as low speed pre-ignition (LSPI). The worldwide use of engine families with different fuel and oil quality represents an additional challenge, which has to be ensured within the scope of series development. The basis for this is extensive expertise and methodical approaches in order to minimize the risk of pre-ignition starting in the advanced development up to series calibration and to avoid engine damage in series application. The presentation examines in detail thermodynamic aspects for the occurrence of pre-ignition. On the one hand, the focus is on the definition and phenomenology of pre-ignition, and influences from operating materials and engine design are analyzed. Based on an enthalpy approach, a possibility is shown to objectively evaluate engines carried out by means of thermodynamic analyzes. If the enthalpy approach is consistently applied, the pre-ignition frequency can be minimized as early as the concept phase. Furthermore, the mechanisms of the 'pre-ignition protection function' on today's engines are explained. Finally, a test method is presented which is used for the final assurance of the operational stability even in the case of stochastically occurring pre-ignition. 1 Overview of irregular combustion The development of ever better, more comfortable and faster vehicles require ever more powerful engines. This demand for power can only be sensibly met by increasing the power density of the engines, with energy efficiency and thus efficiency playing an important role. To this end, turbocharging and/ or downsizing concepts are being pursued alongside technologies for innovative combustion processes and valve control strategies. Higher power density means higher mean pressure and, for the spark-ignition gasoline engine, an increase in <?page no="258"?> the risk of irregular combustion, i.e. undesired initialization of combustion by influences other than the ignition spark. During compression, chemical pre-reactions take place in the fresh gas at increasing pressure and temperature, which lead to self-ignition when a certain ther‐ modynamic state is reached. A distinction must be made here as to whether this auto-ignition starts before (pre-ignition) or after the ignition point (knocking, extreme knocking). The time available for pre-reactions combined with high pressure and high temperature increase the radical concentration in the fresh gas. The time interval for the pre-reactions in the fresh gas is large in combination with later ignition angle positions (ignition timing after TDC), especially at low engine speeds, which is why pre-ignitions generally occur only in the lower engine speed range. Since pre-ignition is only indirectly dependent on ignition timing, it can only be avoided to a limited extent by ignition interventions. Figure 1: Irregular combustion in the indicator diagram. Gasoline engine knock, so-called 'normal knock', can be clearly distinguished from this. Knocking always takes place in the final gas, defined as continuously decreasing fresh gas content during ongoing combustion, which has a strong tendency to radical formation due to increasing pressure and high combustion temperature. Knocking occurs when the radical concentration in the final gas exceeds a critical value, causing the final gas to ignite independently before the flame front arrives. According to the reaction kinetics, sufficient time must be available for this to occur, depending on pressure and temperature. Since knocking occurs only in the second part of combustion, the pressure amplitudes that occur are significantly lower. Knocking can be reliably avoided by adjusting the ignition angle and thus reducing the maximum combustion pressure. Extreme knocking, a special form of knocking, is caused by excessive, unwanted unequal distribution of residual gas, air and fuel in the cylinder with simultaneous high compression. The properties of the fuel 258 Thomas Emmrich 258 <?page no="259"?> itself have a decisive influence on these processes. A special form is glow ignition, which can be initiated by hot spots in the combustion chamber as a result of thermal overload with poor local cooling or by prolonged knocking combustion. In this case, ignition of the cylinder charge does not occur through the ignition spark and therefore cannot be avoided by adjusting the ignition timing. - Irregular combustion - Knocking Glow combustion Pre ignition charactis‐ tics self-ignition in the burned gas after the ignition point Self-ignition in the un‐ burned gas by hotspots Self-ignition in the un‐ burned gas before ignition Operating range • occurs primarily in the upper load range • especially at high engine speeds • occurs primarily in the upper load range • especially at high en‐ gine speeds • occurs primarily at low engine speeds and high loads Interven‐ tion • Adjustment of ignition timing • only by interrupting the fuel supply • cooling of the working gas by e.g. enrichment and/ or load reduction Damage • Thermal and mechanical influences lead to higher component loads and possibly to failure. • Thermal overheating of the piston or spark plug depends on exposure time and knock intensity. • In the case of ex‐ treme knocking, the toler‐ ance range between the knocking limit and the damage area is smaller. • Engine damage after short running time due to high heat input at high pres‐ sure in the com‐ bustion chamber-lim‐ iting components • Depending on number and severity of the PIs • Violent fracture of top land, cracks in piston or solid piston rings • Increase in blowby or crankcase pressure Table 1: Irregular burns and characteristics 2 Definition of pre-ignition and phenomenology Pre-ignition is a self-ignition phenomenon in highly supercharged gasoline engines, which is characterized by incipient combustion even before the external ignition source (e.g. electric ignition spark). Because of the early combustion, strong knocking usually occurs subsequently. Pre-ignition occurs spontaneously primarily at high mean pressures and low engine speeds and is stochastically distributed. Pre-ignition events often occur in sequences if no countermeasures are taken. These sequences are thermodynamically driven. If PI events are caused by oil droplets or particles, they are usually single events, which are also distributed in the combustion chamber, stochastically. The energy conversion is rapid and leads to significantly higher combustion chamber pressures (see Figure 2 on the right), which can reach 2 to 3 times the regular combustion Assessment of pre-ignition phenomena by thermodynamically approach 259 259 <?page no="260"?> chamber pressures. Acoustically, pre-ignition is audible as a distinct metallic-sounding noise (thump), which is due to the high pressure gradients. In the p-V diagram, pre-ignition can be clearly detected by the characteristic pressure rise (see Figure 2 on the left). The occurring maximum pressures often exceed the measuring range of the pressure transducers, which is why usually only a qualitative statement about the events can be made. Figure 2: Pressure curve and heat release for PI at n=1500-min-1 and pme=22-bar The consequence of these significant pressure overshoots is often damage to the piston. Typical damages are: • Fractures at ringland 1/ at the ringlands (see Figure 3) • Fractures/ cracks/ washouts at the top land (due to blowby) • Fractures of the piston pin hub • Hot and cold piston rings • Piston ring fractures These damages may occur after single, violent events (force fractures) or may develop after several pre-ignition events (endurance/ fatigue fractures). In addition to the mechanical robustness of the piston, the magnitude of the pressure surge or the severity of the PI event is decisive. The further the heat release moves away from the ignition point towards the early combustion position, the higher the pressure peaks. This results from the combination of the pressure increase due to energy release with simultaneous compression as a result of the volume reduction caused by the upward movement of the piston (compression) and the high-frequency pressure components from knocking combustion. Damage to the cylinder bore of the affected cylinder can remain minor if detected in time. Such pre-ignition damage can be detected by measuring the blowby or monitoring the crankcase pressure. Slight or incipient damage can only be reliably detected with high sensitivity of the measuring section and a great deal of experience. Since the resulting increase in crankcase pressure or blowby is only very slight or lies within the permissible tolerance, very narrow monitoring limits on the engine test bench are essential. In field operation without monitoring, PI damage is usually only detectable in the event of severe damage to the piston as a result of the loss of power that occurs. In multi-cylinder engines, compression measurement can be used to determine the affected cylinder. Piston designs with ring carriers have a particular tendency to cold and hot-strength clamping of the 2nd compression ring due 260 Thomas Emmrich 260 <?page no="261"?> to deformation of the lower groove flank. Consequential damage to other engine components such as the cylinder bore, valve train or ATL is very likely in this case. Figure 3: Typical damage pattern after several pre-ignitions. Figure 4: Worst-case scenario after pre-ignition. 3 Influencing factors on pre-ignition 3.1 Dependence on operating point/ thermodynamics 3.1.1 Engine speed influence Generally, the risk of pre-ignition is greatest at the low-end torque (LET) point. This point marks the point of maximum mean pressure (= maximum engine torque) at minimum en‐ Assessment of pre-ignition phenomena by thermodynamically approach 261 261 <?page no="262"?> gine speed. Below the engine speed of the LET, the maximum mean pressure (proportional to the boost pressure) is not reached, or above this engine speed the time for the necessary chemical pre-reactions in the combustion chamber decreases. Both conditions therefore have a reducing effect on the risk of pre-ignition. Figure 5: Schematic representation of the pre-ignition risk in the engine map In Figure 6 the PI frequency for a turbocharged 4-cylinder engine has been determined in the test. Figure 6: Engine speed influence 262 Thomas Emmrich 262 <?page no="263"?> 3.1.2 Mean pressure influence Assuming a stoichiometric fuel-air ratio (λ = 1), the increase in mean pressure (BMEP) is achieved by a higher degree of supercharging (increase in boost pressure) of the engine. Thus, the higher the boost pressure, the greater the internal energy (enthalpy) of the fresh gas at constant displacement. For a given compression ratio, the work of volume change during compression remains almost constant (neglecting wall heat exchange), meaning the enthalpy change of the working gas at TDC is proportional to the change in boost pressure. Figure 7 shows this relationship for 2 operating points. As described above, the increase in mean pressure is accompanied by the increase in boost pressure, which in turn leads to the increase in pre-ignition frequency. Figure 7: Mean pressure influence 3.1.3 Charge air temperature The temperature characterizes the internal kinetic energy of a substance, i.e. a high charge air temperature indicates a higher enthalpy state. The charge air temperature characterizes the starting conditions of the thermodynamic process. If we disregard the wall heat losses during compression, we find the enthalpy increase due to the increase of the charge air temperature at TDC (end of compression) (Figure 8). Assessment of pre-ignition phenomena by thermodynamically approach 263 263 <?page no="264"?> Figure 8: Charge air temperature influence. 3.1.4 Fuel-air ratio λ For the direct-injection, turbocharged gasoline engine, the fuel-air ratio is very important because the mixture is prepared in the cylinder and the fuel is injected into the combustion chamber in liquid form. Energy is required for the evaporation of the fuel (enthalpy of evaporation), which is extracted from the fresh gas. As a rule, the engine is operated stoichiometrically (λ = 1). In Figure 9, a variation of λ is shown. For the case of stoichiometric operation, more fuel is injected into the combustion chamber. As a result, the proportion of necessary evaporation enthalpy increases and additional heat is extracted from the fresh gas. This internal cooling reduces the pre-ignition frequency. Short-term sub-stoichiometric engine operation is actively used as the first engine protection measure when pre-ignition is detected, because injection is cycle-sequential. The duration of this engine protection function is usually limited to less than 10 operating cycles. 264 Thomas Emmrich 264 <?page no="265"?> Figure 9: Influence of the fuel-air ratio 3.2 Dependence on the designed engine 3.2.1 Constructional design - combustion process 3.2.1.1 Internal cylinder flow Starting from an external initialization of combustion, e.g. by an ignition spark, combustion progresses concentrically around this starting point in a flame front. The combustion causes the pressure and temperature in the combustion chamber to rise sharply, which shifts the thermodynamic framework conditions in the combustion chamber toward the self-ignition limit of the fresh gas. However, the chemical reactions are time-dependent, which is why pre-ignition only occurs at low engine speeds. Turbulence (kinetic energy of the particles) accelerates the propagation speed of this flame front and minimizes the risk of unwanted self-ignition (knocking), which is why the Otto combustion process is referred to as a turbulence-driven combustion process. As already stated in (Dahns, Han, & Magar, 2009), a low-knock combustion process also results in a lower risk of pre-ignition. 3.2.1.2 DI/ MPI mixture preparation The advantage of the duct-injected MPI process is the excellent homogenization of the fuel-air mixture in the intake manifold and the avoidance of lubricant film wash-off and wall wetting when the engine is warm (Brandt, et al., 2010). In addition to using fuel evaporation enthalpy for internal cooling, the DI process still offers the possibility of optimizing the effect of internal cooling via different injection patterns. The effect of these mechanisms is explained in more detail in section 5. Assessment of pre-ignition phenomena by thermodynamically approach 265 265 <?page no="266"?> 3.2.1.3 Combustion chamber geometry Due to the concentric propagation of the flame front, a central location of the ignition source should be aimed for. This results in combustion paths of equal length toward the cylinder wall and minimizes the risk of knock and pre-ignition. Due to the knock limitation in the LET and the upper load points at engine speeds < 2500 rpm, the ignition timing at these operating points is around TDC. The turbulent kinetic energy (TKE) generated by charge change and upward movement of the piston in the compression phase is then already dissipated. Therefore, to intensify turbulence at TDC, tumble flows and additionally punch points are useful. 3.2.2 Influence of the residual gas fraction Because the combustion gases are not completely expelled from the combustion chamber, the mixing temperature of the fresh gas charge is raised as a function of the residual gas content. High fresh gas temperatures in the cylinder lead to an increased tendency to pre-ignition. In addition, depending on the mixture quality of the fuel grade and the oil composition, there are radicals in the exhaust gas which increase the tendency to self-ignition. Especially at critical operating points, high residual gas contents in the cylinder and unfavorable residual gas composition due to incomplete combustion must be avoided. In particular, acetone and high NOx contents in the residual gas can have a catalytic effect on spontaneous combustion (Hoffmeyer, et al., 2009). 3.2.3 Constructive design for oil balance Since the number of possible influencing factors is very large, some typical causes are listed, but they do not claim to be exhaustive: • Washing off of liner (fuel wall buildup) • Entry into fresh gas (blowby recirculation in intake tract, ATL storage) • Storage in crevices or deposits (oil carbon in the top land, oil carbon, combustion residues) • Cylinder wall temperature (depending on coolant temperature) • Pockets/ sinks in the intake section, where oil emulsion can accumulate • Lubricant film thickness (e.g. depending on the tangential force of the piston rings) In Figure 10, the influence of crankcase pressure is shown as an example of these influencing factors. The increase in pressure was achieved externally via a pressure source and corresponds approximately to the doubling of the blowby value in the test engine. The 50 % change in PI frequency shows the relationship well. An increase in crankcase pressure can occur on the engine practically due to several causes: • Natural wear on the engine (wear of the piston rings - reduction of the tangential force) • Breakage of one or more piston rings • Wear on the liner • Fault in the crankcase ventilation system 266 Thomas Emmrich 266 <?page no="267"?> Figure 10: Influence of crankcase pressure 3.3 Influence of operating fluids 3.3.1 Influence of fuel Fuel quality has a very great influence on pre-ignition behavior. Within the EU, fuel quality is regulated in DIN EN 228 (German Federal Ministry for the Environment, Nature Conservation and Nuclear Safety, 2020). This standardization prescribes, according to the fuel quality, a certain minimum octane number, which characterizes the knocking behavior. As mentioned in (Spicher & Rothe, 2007), conclusions about the pre-ignition behavior can be drawn from the octane number. Fuels with a high octane number cause a lower risk of pre-ignition than fuels with a low octane number. The admixture of diesel fuel, which reduces the octane number and thus the knock resistance and increases the ignition readiness, relies on this effect. However, this method was not able to achieve a constant increase in the pre-ignition tendency over a prolonged period of time on the engine. Various mineral oil companies offer special fuels (referred to here as PI-fuel) that are specifically formulated to increase the tendency to ignite several times over compared with standard fuels (e.g. RON 95). The PI-fuel used for the tests contains approx. 1/ 3 more aromatics. Particularly long-chain and thus high-boiling aromatics are knock-resistant and thus realize the same octane number as in the standard fuel RON 95, but participate only to a limited extent in the mixture formation. The fuel components that really participate in mixture formation exhibit lower knock resistance and thus cause the significant increase in pre-ignition frequency (Figure 11). Evaporation behavior of the fuel is closely related to the fuel spray. The higher boiling point of PI-fuel prolongs the liquid phase of the spray. This increases the risk of wall wetting resulting in fuel ingress into the oil. Assessment of pre-ignition phenomena by thermodynamically approach 267 267 <?page no="268"?> Market-specific quality criteria for fuels require adapted test cycles to ensure the operational reliability of the engine. PI-Fuels are used specifically for the calibration of pre-ignition protection functions in the control unit and their escalation stages. On the other hand, the mechanical robustness of the engine with regard to pre-ignition can be checked within a reasonable time frame. Figure 11: Influence of fuel quality on pre-ignition behavior 3.3.2 Lubricating oil influence Engine oil is a formulation of long-chain hydrocarbons of mineral or synthetic origin. Additives are added to the basic formulation depending on the manufacturer and properties, which makes it very difficult to make a generalized statement regarding the influence of pre-ignition. In principle, however, engine oil has a high boiling point due to its components and thus poor evaporation properties. However, oil vapor mixed with air is very ignitable. Parallels can be drawn here with diesel fuel. It has been shown that the pre-ignition effect caused by the use of a particular oil can be engine-specific. In general, the focus is on oil consumption and certain additive components (Yasueda, Takasaki, & Tajima, 2011). From the large number of possible factors influencing engine oil on pre-ignition frequency, 2 typical representatives are presented: • In (Takeuchi, Fujimoto, Hirano, & Yamashita, 2012), the influence of calcium salt compounds as a typical D/ D additive component on pre-ignition tendency is presented. Particles and combustion residues bind to this salt and are thus kept in suspension. This property prevents the deposition of these particles in the engine and thus has a positive influence on cleanliness and mechanical function. However, as the content of this salt compound in the oil increases, so does the tendency of the engine to pre-ignite. In Figure 12, this influence is illustrated on a turbocharged 2-liter gasoline 268 Thomas Emmrich 268 <?page no="269"?> engine. Calcium was added in the form of calcium stearate. The experiment showed that doubling the calcium content from 1000 ppm to 2000 ppm led to a doubling of the pre-ignition frequency. The reason given by (Takeuchi, Fujimoto, Hirano, & Yamashita, 2012) for this behavior is that oil droplets/ oil vapor enter the combustion chamber through the wiping action of the piston rings and can oxidize there with oxygen from the fresh gas. This reaction provides the ignition source for pre-ignition. Figure 12: Effect of calcium compounds on pre-ignition behavior • As a 2nd influencing factor, oil dilution was investigated. This process is found in vehicles which are primarily operated in urban and short-distance traffic. In the case of under-stoichiometric engine operation as a result of cold-start enrichment, a certain amount of fuel is introduced into the engine oil via the wall coating, depending on the engine. When the engine is at operating temperature, most of this added fuel evaporates again. If the engine is not at operating temperature, higher molecular weight compounds remain in the engine oil and cause the engine oil level to rise. Due to lower viscosity and the lower boiling point of fuel, a risk potential can be derived here. In the test, 20 % fuel was added directly to the fresh engine oil. However, the test shows that oil dilution has no demonstrable influence on the pre-ignition risk. However, the high-boiling components that end up in the engine oil via the wall wetting due to a higher boiling point cannot increase the pre-ignition risk either, because these components have a high knock resistance (see section 3.3.1). Assessment of pre-ignition phenomena by thermodynamically approach 269 269 <?page no="270"?> Figure 13: Influence of oil dilution with fuel 3.4 Classification of pre-ignitions According to the theories formulated in section 3.1-3.3 and the influences presented, the following classification of influences on pre-ignition (see Figure 14) is made: Figure 14: Main factors influencing pre-ignition behavior 270 Thomas Emmrich 270 <?page no="271"?> 4 Measuring pre-ignition, evaluation parameters and limit values In contrast to knock analysis, unfiltered signals are analyzed and the absolute peak pressure is used as a measure. Since the absolute peak pressure is composed of low-frequency and high-frequency components, it is extremely important to record all components as unaltered as possible. Careful selection of the pressure sensors in conjunction with the associated measuring chain on the basis of their characteristic values must be observed: • Cut-off frequency f C : According to definition, this is reached when a signal attenuation of -3-dB occurs. • Time constant τ: It characterizes the response behavior of the measurement chain, i.e. the time that elapses until a signal change is realized. It is linked to the cutoff frequency via the following relationship: In knocking pre-ignitions, the frequency of the pressure signals change very quickly. In order to record frequency components above 10-kHz reliably and τ = 1 2π * f C without loss of information, the cut-off frequency of the entire measuring chain should be higher by a factor of 10. Quartz pressure transducers with a cutoff frequency below 100 kHz are therefore unsuitable. In addition, all bandpass filters in the measurement chain must be switched off. If it is not possible to install suitable high-frequency pressure transducers, the analysis data must be multiplied by a corresponding factor, which is determined in the laboratory with the measuring chain. With a correctly set up measuring system, all pressure curves are now recorded and classified with a peak value analyzer. The threshold of combustion chamber pressure, above which pre-ignition must be assumed, can be determined as follows: For the counting and evaluation of the pre-ignition phenomenon, only damage-relevant pre-ignitions are used, which exceed the designed reciprocating strength of the piston. This criterion is defined individually for each engine variant. In Figure 15 an analysis of the pre-ignition tendency is shown as an example. Here, the lower pressure class for PI detection was set at 135 bar. The distribution of peak pressures detected for this fuel variation shows that the vast majority of PI events are above 215 bar peak pressure. This information can be used as an indication for the mechanical design of the piston. Assessment of pre-ignition phenomena by thermodynamically approach 271 271 <?page no="272"?> Generally, pistons are designed for pressures that reflect the regular combustions in the map. Pre-ignition events occur statistically with low frequency, but with high damage potential. For this reason, a sensible compromise between fatigue strength and fatigue life (overstressing the piston for a limited time) is only possible with the involvement of the piston manufacturer's technical expertise. Figure 15: Statistical evaluation of pre-ignition with pressure classification 5 Evaluation of pre-ignition behavior with the IAV enthalpy approach As mentioned in section 3, to date there is no clear theory on the origin and mechanisms of pre-ignition. Nevertheless, new engines with higher mean pressures are being built all the time. Here, it is important to have an objective evaluation criterion for the pre-ignition problem at hand. For the characterization of thermodynamically based pre-ignition, the internal energy of the charge is a suitable criterion, as described in section 4.2.2. The IAV model approach (Günther, Tröger, Kratzsch, & Zwahr, 2011) is based on the consideration of the operating point-dependent variables of charge pressure, charge temperature and air ratio. Here, the enthalpy H describes the energy contained in the working gas, which is formed by • the enthalpy of the fresh gas • the enthalpy of the residual gas • the enthalpy of the fuel at the time 'inlet closes'. For the 'inlet closes' to 'Firing DTC' interval, energy is supplied to the fresh gas by the volume change work, an energy exchange occurs via the wall heat transfer, and energy is supplied via the injected fuel (DI) and removed again by its evaporation. Relating the enthalpy to the compression volume makes it possible to compare different engines and 272 Thomas Emmrich 272 <?page no="273"?> also characterizes the energy density at TDC, which is proportional to the reaction kinetics. All corresponding formulas are presented as follow: H = U + p * V = c V * m * T H T DC = H I C + dH I C T DC H I C = ℎ Air * m Air + ℎ RG * m RG + H F uel dH I C T DC = V d p + Q H T I C T DC +H F uel The enthalpy approach is based on purely thermodynamic factors. The limit enthalpy derived from this represents the pre-ignition limit (tolerable statistical frequency of pre-ignitions at a specific operating point, e.g. x/ 10000 ASP or x/ h). This limit enthalpy is to be determined in the engine test by parameter variation. The enthalpy curve for a charge air temperature variation on a 4-cylinder engine is shown in Figure 16. If one pre-ignition per cylinder and hour is declared permissible, the limit enthalpy can be read from this diagram. Figure 16: Determination of the enthalpy limit using the example of a charge air temperature variation Assessment of pre-ignition phenomena by thermodynamically approach 273 273 <?page no="274"?> The enthalpy change during compression (work input) is naturally strongly dependent on the volume ratio, whereby changes to the geometric compression ratio may require a larger fresh gas mass if the mean pressure remains unchanged. By influencing the effective compression ratio (IC to TDC), e.g. by late inlet closing (Atkinson process), significant enthalpy reduction effects can be achieved. In addition, enthalpy inputs can be easily influenced by residual gas mass or temperature. Processes for residual gas purging by means of positive purging slope have already proven their effectiveness in series production concepts. The enthalpy contribution of the residual gas is comparatively low. However, due to a high proportion of free radicals, especially in the case of entrained combustion, the critical enthalpy level is lowered at the same time, which increases the effect of the residual gas component on the tendency to pre-ignite (Hoffmeyer, et al., 2009). The enthalpy reduction due to fuel occurs, on the one hand, due to the location of the evaporation of the injected fuel with respect to the UT (Eichert, Günther, & Zwahr, 2005). On the other hand, the increasing fuel mass fraction in the cylinder directly increases the negative enthalpy fraction. The wall heat fraction must be differentiated into quantities that can be influenced, such as wall temperature and heat transfer due to charge movement. Although their effects on the total enthalpy appear small, under limited conditions they can make a significant contribution towards reducing the pre-ignition tendency. For pre-ignition investigations, the engine must be operated in this limit range in order to generate a statistically meaningful sample. It is advisable to equip the test vehicle with high-strength pistons to avoid early engine damage. A routine has been set up on the engine test bench for logging the pre-ignition events, which counts and classifies pressure transgressions of a limit value defined as a function of the operating point. The classification of the events offers the advantage of also including events with lower peak pressures in the evaluation and thus increasing the statistical significance. Any existing engine protection functions are partially or completely switched off for this purpose (depending on the objective of the investigation) and replaced by tailored protection routines from the test bench environment: • This protection routine registers pressure exceedances of the preset spray pressure p max_max , which represent pre-ignitions. At the same time, a signal is transmitted to the test bench automation system, which continuously compares the previous blowby or crankcase pressure mean value with the current value and shuts down the engine if there is a sudden increase. It can then be assumed that piston damage is beginning to occur. In this way, it is possible to react in good time and limit subsequent damage. • If an adjustable number of events occur in a sequence, the engine is switched off immediately to reduce the risk of engine damage. 274 Thomas Emmrich 274 <?page no="275"?> Figure 17: Principle test bench design for pre-ignition testing Two indexing systems are required for the test bench setup (see ). Since at IAV each PST is equipped with IAV's own system (IAV IndiCar), this is used to implement the engine protection function (shutdown) and event counting. An additionally installed system is used for thermodynamic analysis, enthalpy determination and statistical evaluation. For targeted modification of the enthalpy at TDC, conditioning of the charge air temperature or boosting of the charge pressure on the intake side is necessary on the test bench. An IAV calculation tool is used to evaluate the test results and determine the enthalpy, which includes a work process calculation based on the results of the process analysis. In addition to the specific enthalpy at TDC, the individual enthalpies can also be calculated as results according to the model approach. Particularly in the case of turbocharged gasoline engines, pre-ignition tests accompany the development of combustion processes and charge exchange strategies over the long term. By combining tests on suitably equipped test benches, the application of the IAV enthalpy approach and the associated simulation tools, engine development can thus be carried out effectively. 6 Pre-ignition protection function in the control unit In series production, a procedure analogous to that used for knock detection is used. Pre-ig‐ nition events are usually accompanied by knocking in addition to the high combustion chamber pressure. The burns, which are strongly shifted to the early stage, make it possible to distinguish them very well from classic knocking. By means of a 2nd detection window, immediately before the detection window of the knocking, the high-frequency signal components of the PI are detected. As an additional feature, the rotational irregularities resulting from the PI can also be used for PI detection. Assessment of pre-ignition phenomena by thermodynamically approach 275 275 <?page no="276"?> Extensive test bench investigations are necessary for the application of the PI protection function. The PI frequency in PI/ h or PI/ 10000 cycles, limit enthalpy in J are determined as evaluation parameters. The data obtained is used to derive the engine control unit parameters for regular engine operation and for the PI protection function. The PI protection function contains the following escalation levels: • Enrichment (reduction of the enthalpy of the charge mass due to a higher proportion of evaporation enthalpy + change of the limiting enthalpy for λ) • Change of injection strategy • Filling reduction • (cylinder deactivation) 7 Proposal of an endurance run for the evaluation of the application and mechanical robustness with regard to pre-ignition Defined acceptance methods are useful for the final acceptance of the engine, but also for the evaluation of as-built conditions. An endurance run covering all relevant load points that can occur in the field is particularly important for the pre-ignition problem. As already worked out, pre-ignition occurs in certain load situations and is stochastically distributed. Therefore, these load points must be approached in a targeted manner in the endurance run. In order to set statistically stable conditions, PI fuels are used in some cases. Figure 18 shows examples of safeguarding endurance runs. Figure 18: subdivision of endurance test run 7.1 Calibration test run Objective: • Verification of the PI protection function • Detection of PI • Response to PI → prevention of sequences • Synthetic replication of field-relevant extreme situations Fuel qualities: • Spec. Fuel (RON95/ 98) • PI-fuel 276 Thomas Emmrich 276 <?page no="277"?> Components: • Stationary test at LET speed + 100 rpm (safety reserve to prevent signs of aging during the test) • Dynamic engine speed change • Dynamic load change • Cold sooting followed by transient operation (low engine speed+high load) 7.2 Robustness test run Objective: • Verification of mechanical robustness • Generation of many PI’s in a reasonable amount of time • Achievement of the engine-specific number of PI’s according to manufacturer's specifications related to each cylinder or engine as a whole Fuel qualities: • PI-fuel Components: • Stationary test at LET speed + 100 rpm (safety reserve to prevent signs of aging during the test) • Dynamic engine speed change • Dynamic load change Source reference: Brandt, S., Knoll, G., Schlerege, F., Pischinger, S., Wittler, M., Stein, C… Gohl, M. (2010) Beeinflussung der Schmierölemission durch die Gemischbildung im Brennraum von Verbrennungsmotoren. [Influence of lubricating oil emission by mixture formation in the combustion chamber of internal combustion engines] 19th Aachen Colloquium on Vehicle and Engine Technology, (pp. 571-592) Aachen Federal Ministry of the Environment, Nature Conservation and Nuclear Safety (06 2020) Accessed from https: / / www.bmu.de/ themen/ luft-laerm-verkehr/ verkehr/ kraftstoffe/ kraftstoffqualitaet/ Dahns, C., Han, K.-M., & Magar, M. (2009) Vorentflammung bei Ottomotoren. [Pre-ignition in gasoline engines] Frankfurt/ M: FVV research report on project 931 Eichert, H., Günther, M., & Zwahr, S. (2005) Simulationsrechnungen zur Ermittlung optimaler Einspritzparameter an DI-Ottomotoren. [Simulation calculations to determine optimal injection parameters on DI gasoline engines] AEP no. 9-10. Günther, M., Tröger, R., Kratzsch, M., & Zwahr, S. (2011) Enthalpiebasierter Ansatz zur Quantifi‐ zierung und Vermeidung von Vorentflammungen. [Enthalpy-based approach to quantify and avoid pre-ignition] MTZ---Motortechnische Zeitschrift 72 (04), 296-301 Hoffmeyer, H., Montefrancesco, E., Beck, L., Willand, J., Ziebart, F., & Mauss, F. (2009) Catalytic Reformeated Exhaust Gases in Turbocharged DISI-Engines SAE Paper 2009-01-0503 Assessment of pre-ignition phenomena by thermodynamically approach 277 277 <?page no="278"?> Spicher, U., & Rothe, M. (2007) Extremklopfer - Ursachenforschung nach schadensrelevanten klopfenden Arbeitsspielen. [Extreme knocking - cause research after damage-relevant knocking working cycles] Frankfurt/ Main: FVV e.-V., Final Report FVV-Project 816 Takeuchi, K., Fujimoto, K., Hirano, S., & Yamashita, M. (2012) Investigation of engine oil effect on abnormal combustion in turbocharged direct injection - spark ignition engines SAE Int. J. Fuel Lubr. 5 (3) 2012-01-1615 Willand, J., & Daniel, M. e. (2009) Grenzen des Downsizing bei Ottomotoren durch Vorentflammung. [Limits of downsizing in gasoline engines by pre-ignition] MTZ-Motortechnische Zeitschrift 70 (05), 422-428 Yasueda, S., Takasaki, K., & Tajima, H. (2011) The abnormal combustion caused by lubrication oil on high BMEP gas engines 13th conference: 'The working process of the internal combustion engine', (pp.-324 - 336) Graz 278 Thomas Emmrich 278 <?page no="279"?> A fundamental investigation of oil additives on pre-ignition in a high pressure combustion chamber Jan Reimer (KIT-IFKM), Jürgen Pfeil (KIT-IFKM), Frank Altenschmidt (Mercedes-Benz Group AG), Thomas Koch (KIT-IFKM) Abstract The irregular combustion phenomenon of low speed pre-ignition (LSPI) is still an important issue in modern internal combustion engine development and research, despite numerous publications looking into different directions such as the influence of injection strategy, scavenging, engine oil and fuel properties, and solid particulate matter as a potential ignition source. As the occurrence of pre-ignition is heavily dependent on the test engine, a pressure chamber test bench was set up to enable experiments to be carried out under constant and reproducible conditions. Pressure and temperature can be set in such a way that ambient conditions similar to those in the combustion chamber can be created. The first step was to investigate whether a single droplet of oil in the combustion chamber can be the trigger for pre-ignition. The ignition delay times of pure PAO6 oil and PAO6 oil with different calcium and magnesium contents have been investigated. As it is quite conceivable that oil could accumulate in the combustion chamber and remain there for several operating cycles, different surface materials were also examined in order to investigate possible catalytic effects. Therefore, in a second step a collector for oil was developed with which it was possible to record the ignition temperature of the oil in the constant-volume chamber. This collector had to be optimized to be able to hold a very small amount of oil, heat up quickly and be coated with different materials. The oils were then examined with regard to their auto-ignition temperature at different chamber pressures and temperatures. In addition to metallic, ceramic and non-metallic materials, which have a chemically inert and possibly catalytic effect, materials that occur in the combustion chamber were also used as surfaces. This research has shown that the common mechanisms of action found in the literature can be refuted. A single oil droplet moving freely in the combustion chamber cannot be the trigger for pre-ignition due to the excessively long ignition delay time. In addition, catalytic effects between the oil and a surface made of gold, chromium nitride, graphite and aluminium oxide could not be observed. This is shown by the self-ignition temperature of the oil, which could not be reduced by the surfaces. <?page no="280"?> 1 Introduction As early as 1912, Heller [1] wrote in his textbook for self-instruction that the compression ratio must be lowered as the fuel quality decreases in order to avoid pre-ignition. Almost 10 years later, Thomas Midgley invented tetraethyl lead, which was used to reduce knocking and was dubbed the solution to the problem of knocking combustion [2]. However, lead pollution in the environment increased dramatically as a result and later studies proved that increased blood lead concentrations have a strong impact on human health. As a result, leaded petrol was banned in Germany in 1988 and in the European Union in 2000 [3]. In order to achieve the European Union's current climate targets, CO 2 emissions in the transport sector must be reduced. One way to achieve this is to reduce the engine displacement. However, in order to achieve the same performance, the intermediate pressure must be increased and the specific power must rise. This has more than doubled in the last 20 years (Figure 1). The resulting highly turbocharged petrol engines with direct injection then tend towards low speed pre-ignition (LSPI). This ignition before the actual ignition point leads to undesirable combustion anomalies, such as extremely knocking combustion, which can lead to serious engine damage due to high pressure amplitudes. The problem here is that LSPI only occurs sporadically and, unlike for knocking combustion, no limit can be defined. Figure 1: Development of specific power and consumption [4] For this reason, numerous investigations have been carried out for many years to find the cause of the sporadically occurring pre-ignition [5-22]. According to the literature, possible causes include oil droplets, fuel droplets, particles, residual gas or hot spots, whereby there 280 Jan Reimer, Jürgen Pfeil, Frank Altenschmidt, Thomas Koch 280 <?page no="281"?> is also the fundamental possibility that several influencing factors cause the combustion anomaly. As the occurrence of pre-ignition is heavily dependent on the test carrier, a basic test rig was set up in this work that is capable of generating combustion chamber-like ambient conditions. This offers the advantage that the tests can be carried out reproducibly and are therefore directly comparable with each other. What is new compared to previously used pressure chambers is that the chamber can generate high pressures and high temperatures at the same time. Part of this work is based on the results presented at the 5th and 6th International Conference of Knocking in Gasoline Engines [20, 23]. Together with the results on oil ignition on hot surfaces, an overview of the influence of oil additives on oil droplets in a high-pressure combustion chamber will now be given. Basically, pre-ignition with subsequent knocking combustion can be initiated by the cylinder head and piston (Figure 2). However, since hot components in the head area are more likely to be responsible for glow ignition, it is obvious that the cause is the piston. If we consider the gas and liquid phase on the piston as a possible cause, a distinction can be made between fuel and oil. In the case of fuel, it is conceivable that reactive deposits form, which then lead to pre-ignition at a later point in time [13, 16, 21, 22]. In the oil sector, the oil droplets and their composition are primarily considered with regard to additivation [6-8, 11, 12, 14-16, 18, 19, 22, 24]. However, despite numerous investigations, the cause of pre-ignition has not yet been found. As illustrated in Figure 3, it is possible for oil droplets to detach from the piston ring gap and then lead to pre-ignition through self-ignition. The results of this series of tests are presented in chapter 4.1. According to Schöler [25], it has been observed that the flame burns far into the top land area and thus leads to heating of the ring gap. With the knowledge gained, it is possible that an oil droplet can lead to pre-ignition not only in the air flow, but also on a hot surface, as high temperatures prevail here. To this end, various surface materials are analysed for catalytic reactions and presented in Chapter 4.2. Source for knocking Hot components Oil droplets Hot surface on piston Piston ring gap Glow ignition Cylinder Head Piston Figure 2: Possible causes of droplet ignition A fundamental investigation of oil additives on pre-ignition in a high pressure combustion chamber 281 281 <?page no="282"?> Engine Oil Self-Ignition Figure 3: Removal of oil from the piston ring gap 2 Test bench design The experiments are carried out in a pressurised chamber in which gas pressures of up to 1000 bar and temperatures of up to 500 °C can be displayed. The set-up is shown schematically in Figure 4. A defined quantity of air and fuel is fed into the heated premixing chamber via the air/ fuel management system. After a defined homogenisation time, the mixture flows into the combustion chamber by opening a valve. Then the oil management system located there feeds a small amount of tempered oil into the combustion chamber via an oil droplet generator. Of course, it is also possible to carry out the tests without fuel and/ or without oil injection. Combustion Chamber Fuel Management System Fuel Management System Oil Management System Air Management System Pre-conditioning Chamber Figure 4: Schematic representation of the test bench setup 282 Jan Reimer, Jürgen Pfeil, Frank Altenschmidt, Thomas Koch 282 <?page no="283"?> Glow plug access Op�cal access Oil dosing access Op�cal access Figure 5: Combustion chamber in open state In addition to the above-mentioned openings for air, fuel and oil, there are three further accesses, which are shown in Figure 5. They are located at the front and rear of the chamber and at a 90° angle to the upper access, which is occupied by the oil droplet generator. The front and rear access points are each fitted with a glass window so that the chamber is visually accessible and the introduced oil can be observed with a high-speed camera. An adapter is screwed into the side access, which carries a glow plug and a thermocouple. The glow plug can be used either to ignite the mixture introduced or to heat up an oil pick-up. The combustion chamber setup is controlled via a CompactRIO system from National Instruments and slow data (10 Hz) is recorded at the same time. A DEWE800 from Dewetron is used for fast data acquisition. Chamber pressure, glow plug current and temperature of the thermocouple located on the glow plug adapter are recorded by this. The image data is recorded with a Phantom v1612 high-speed colour camera from VisionResearch, on which the K2 DistaMax far-field microscope from Infinity is mounted. The fuel management system injects the fuel into the preconditioning chamber at up to 200 bar via a high-pressure injection valve. A defined quantity of oil is injected into the combustion chamber itself via the specially developed oil management system. This has a micro bore with a diameter of 100 µm at the tip and can inject a few drops of oil at temperatures of up to 245-°C into the combustion chamber. In order to be able to introduce the oil not only as drops into the chamber, but also to heat it on a surface, an adapter can be screwed onto the glow plug inserted into the side of the chamber, which is heated by this. The system consists of two parts and is shown in Fig. 6. The base body, which always remains screwed to the glow plug adapter, and the oil pick-up, which can be replaced via the upper access port. The volume of both components was reduced by 90 % to 205 mm 3 in several steps over the course of the project. This A fundamental investigation of oil additives on pre-ignition in a high pressure combustion chamber 283 283 <?page no="284"?> was necessary because otherwise the oil could not reach to the auto-ignition temperature quickly enough and would evaporate beforehand. This design allows the oil pick-up to be changed quickly and easily and therefore enables the use of cleaned oil pick-ups or oil pick-ups coated with different materials for the measurements. The temperature is not measured directly in the oil with a thin thermocouple. The reason for this is that catalytic effects are to be analysed and the thermocouple would have an influence on the measurement results. Therefore, there is a reference measuring point that is located close to the surface of the oil to be heated, but has no contact with it. The glow plug is controlled via a programmable power supply unit. This ensures, as has also been confirmed by measurements, that the glow plug behaves in the same way every time it is heated up, as long as the ambient conditions are not changed. This means that both the heating speed and the maximum temperature can be set or limited. Oil pickup Base Glow plug Temperature Measuring Point Figure 6: Design of the oil pick-up 3 Experimental methodology Four different oils were used for the tests (Table 1). They are based on the same base oil (PAO6) and differ in their content of calcium and magnesium detergent. The reason for this is that it is already known from the literature that the calcium content has an influence on the frequency of LSPI [7, 11, 15, 16, 18, 24]. Since magnesium has no influence on the auto-ignition temperature of the oil [6, 15, 16, 24], the results are directly dependent on the calcium content. - A B C D Calcium Detergent 0.8 0.6 0.4 0.2 Magnesium Detergent 0 0.33 0.67 1 PAO 6 99.2 99.07 98.93 98.8 Calcium / % 0,1 0.075 0.05 0.025 Magnesium / % 0 0.025 0.05 0.075 Table 1: PAO6 based oil with various additives 284 Jan Reimer, Jürgen Pfeil, Frank Altenschmidt, Thomas Koch 284 <?page no="285"?> Four different coatings are used for the oil pick-ups, which are listed in Table 2. Care was taken to ensure that the materials differ in their properties and partly occur as components in real engine combustion chambers. The aim is to find out whether the different materials have a influence on the auto-ignition temperature of the various oils and whether certain oil-surface pairings ignite earlier than others. Material Material Properties Gold Metallic chemically inert Aluminium oxide Ceramic chemically inert Chromium nitride Sample material of the combustion chamber - Graphite non-metal - Table 2: Coating materials and their material properties The analyses can be divided into three sub-areas: 1. Oil droplets in hot air 2. Homogeneous oil-fuel-mixtures in hot air 3. Oil on hot surfaces For parts 1 and 2, all measuring systems are triggered simultaneously by activating the oil dosing unit. In part 3, this is done via a limit value of the glow plug current. This means that all measuring systems can be synchronised with each other and start recording measurement data at the same time. 3.1 Oil droplets in hot air When the experiment is started, air flows into the combustion chamber until the final pressure is reached. Once the gas temperature has reached a defined limit value, a small amount of oil is introduced into the combustion chamber via the oil droplet generator. The camera recording is started when the oil droplet generator is opened. Since the recording frequency is known, the number of frames can be used to calculate the time that the droplet needs to ignite after leaving the oil droplet generator. The number of frames is therefore a measure of the ignition delay time of the oil droplet. If the oil droplet does not ignite spontaneously, the chamber temperature is increased and the test is repeated. In this way, the self-ignition limit of an oil can be determined at a defined pressure. 3.2 Homogeneous oil-fuel-mixtures in hot air In contrast to the previous chapter, an oil-fuel mixture is now introduced into the pre-chamber in addition to the air. A magnetic stirrer is located in the pre-chamber to achieve homogeneous distribution. After a sufficient homogenisation time, the air-oil-fuel mixture flows into the combustion chamber, which has the expected ignition temperature of the mixture. If the mixture does not ignite, a glow plug inserted in the combustion chamber A fundamental investigation of oil additives on pre-ignition in a high pressure combustion chamber 285 285 <?page no="286"?> is heated to determine whether the mixture is ignitable. The test is then repeated with a higher combustion chamber temperature. 3.3 Oil on hot surfaces In the third step, the oil is analysed on a surface. To do this, a defined amount of oil is added to the oil pick-up through the top access of the chamber using a pipette before the test begins. After closing the combustion chamber, it is filled with air up to a defined pressure. The pre-chamber cannot be used as in the previous experiments, as the oil would be ejected from the oil pick-up during the sudden overflow. In addition, high chamber temperatures cannot be applied in this experiment, as the oil would have already been vaporised in the time it takes to close the chamber and prepare for the experiment. The test begins with the heating of the glow plug, which then heats the surface of the oil pick-up. As can be seen in Figure 7, the oil begins to vaporise and starts to burn at a certain instant of time. As in the previous experiments, the ignition is detected optically and the ignition temperature is then determined. The thermocouple is located at a defined reference point in the oil pick-up (see Figure-6). Figure 7: Oil in the oil pick-up shortly before self-ignition (left), at the start of ignition (centre), after ignition (right) 4 Results 4.1 Reaction of oil droplets in hot air It is known from the literature that the calcium content in oils has an influence on the tendency to ignite [7, 11, 15, 16, 18, 24]. Higher calcium contents mean a lower auto-ignition temperature. Figure 8 shows that the oil with the highest calcium content ignites first at the same pressure. Oils B, C and D from Table 1 were compared here. The lowest temperatures at which a small amount of oil self-ignites in the combustion chamber are shown in each case. This confirmed the previous theory and showed that the chamber setup delivers the same results as an engine test bench. 286 Jan Reimer, Jürgen Pfeil, Frank Altenschmidt, Thomas Koch 286 <?page no="287"?> 280 285 290 295 300 305 310 315 320 25 30 35 40 45 50 Chamber temperature / °C Chamber pressure / bar 0,075 % Ca 0,050 % Ca 0,025 % Ca Figure 8: Ignition temperature for different chamber pressures with oil B, C and D If hot oil is introduced into hot air, it can be seen that the ignition delay time of the oil decreases with increasing chamber pressure and increasing chamber temperature (Figure 9). From a chamber pressure of 30 bar, ignition delay times of approx. 20 ms can be achieved at a corresponding chamber temperature. Figure 10 shows schematically what this means. For a speed of 1500 rotations per minute, 20 milliseconds results in a duration of 180 degrees crank angle. At 1750 rotations per minute, the value increases to 210 °KW and at 2000 rotations per minute to 240 °KW. This means that if an oil droplet should lead to pre-ignition in the same working cycle, it would have to detach from the piston at bottom dead centre for very low engine speeds in order to have sufficient time to ignite on its own. However, it is more likely that the droplet detachment occurs when the piston is decelerated during the upward movement and therefore there is not enough time for self-ignition. In addition, for such a short ignition delay time, the required temperature is only reached shortly before top dead centre in the compression phase and the oil is not yet much warmer than the coolant temperature when it separates. The only possibility that the single droplet triggers pre-ignition is if it originates from one of the previous cycles and remains in the combustion chamber. However, this is very unlikely, as it is very likely to be flushed out, burnt or deposited on a component due to the movement of the charge. A fundamental investigation of oil additives on pre-ignition in a high pressure combustion chamber 287 287 <?page no="288"?> 0 20 40 60 80 100 120 140 300 310 320 330 340 350 360 Ignition delay time / ms Chamber temperature / °C 20 bar 30 bar 47,5 bar 50 bar Figure 9: Ignition delay time for different chamber temperatures with oil A -270 -240 -210 -180 -150 -120 -90 -60 -30 0 30 60 90 Crank Angle / °CA aTDC 1500 rpm / 20 ms 1750 rpm / 20 ms 2000 rpm / 20 ms Ignition Delay / °CA aTDC 20 ms 50 ms 120 ms 1500 rpm 180 450 1080 1750 rpm 210 525 1260 2000 rpm 240 600 1440 Figure 10: Conversion of ignition delay time from milliseconds to °CA for different engine speeds; cylinder pressure curve for normal and knocking combustion [11] If the oil is mixed with fuel and evaporated beforehand, it can now be tested whether the mixture with added calcium starts to burn earlier. The results are shown in Figure 11. The flammability limit for various pressures of iso-octane was used as a reference. Oil B (0.075 % calcium content) was added to the fuel in two series of tests with 5 % and 10 %. With oil 288 Jan Reimer, Jürgen Pfeil, Frank Altenschmidt, Thomas Koch 288 <?page no="289"?> D (0.025 % calcium), a test series was carried out with 5 % admixture. It can be seen that with the same oil admixture, the mixture with the higher calcium content ignites earlier, but the ignition temperature increases with increasing oil content. In these mixtures, the iso-octane always triggers the ignition. This confirms the results of Ohtomo et al. [26]. 200 220 240 260 280 300 320 340 360 0 5 10 15 20 25 30 Chamber Temperature / °C Chamber pressure / bar iso-octane iso-octane + 5% Oil B (0,075 % Ca) iso-octane + 10% Oil B (0,075% Ca) iso-octane + 10% Oil D (0,025 % Ca) Figure 11: Ignition temperature for different chamber pressures with iso-octane and mixtures of iso-octane and oil The results to date rule out the possibility that a single drop of oil moving through the combustion chamber causes pre-ignition due to the excessively long ignition delay time or the excessively high ignition temperatures required. However, it is quite conceivable that the oil could be deposited on a hot surface and remain there until the next working cycle. Although a mixture of fuel and oil has a higher ignition temperature than pure fuel, Schöler [25] has shown that the flame burns in the piston ring gap. This suggests that very high temperatures must prevail here in order to ignite the mixture. 4.2 Reaction of oil droplets on hot surfaces Figure 12 shows a series of measurements for the gold-coated surface at 40 bar chamber pressure with oil D. It is easy to see that the measurements can be repeated very reproducibly and that the ignition temperatures differ in the range of 5-K. Figure 13 shows the results for different surface materials, chamber pressures, chamber temperatures and oil additives. Each measurement series consists of five individual measurements, whereby the average value and the standard deviation are shown. It is immediately apparent that there is no clear trend for any variation. All measurement points lie within a narrow ignition temperature range, which has to be regarded as measurement inaccuracy. In order to determine the temperature delta between the oil temperature and A fundamental investigation of oil additives on pre-ignition in a high pressure combustion chamber 289 289 <?page no="290"?> the reference point temperature, the measurements were repeated with the gold surface and the temperature was measured directly in the oil. In this configuration the ignition temperature is about 15-K higher than at the reference measuring point. 290 295 300 305 310 315 320 325 330 1 2 3 4 5 Ignition Temperature / °C Measuring Point / - Gold - 40 bar - Oil D Figure 12: Ignition temperature for gold surface at 40-bar chamber pressure with oil D 296 298 300 302 304 306 308 310 312 314 316 Temperature oil pick-up / °C Surface material / - 30 bar - Öl D - 50 °C 40 bar - Öl D - 50 °C 40 bar - Öl B - 50 °C 30 bar - Öl B - 75 °C Gold Chromium Nitride Graphite Aluminium oxide Figure 13: Ignition temperature for different surface materials, chamber pressures and temperatures 290 Jan Reimer, Jürgen Pfeil, Frank Altenschmidt, Thomas Koch 290 <?page no="291"?> 5 Summary & Outlook In this work, oils with various additives were analysed in a pressure chamber that can withstand both high pressures and high temperatures. It was shown that a small amount of oil moving freely in the combustion chamber cannot be the trigger for pre-ignition, as the ignition delay time or the auto-ignition temperature are too high. Calcium has a combustion-promoting effect in the oil and leads to lower auto-ignition temperatures at higher concentrations. However, if the oil is mixed with the fuel, the ignition temperature of the mixture is always higher than that of the pure fuel. This means that the oil cannot be the trigger for pre-ignition. A new hypothese for the ignition of oil is that oil could be deposited in hot areas such as the piston ring gap and remain there until the next work cycle. For this reason, a design has been developed that enables the smallest quantities of oil to be ignited spontaneously via a heatable surface and the temperature trace to be determined in the process. For gold, chromium nitride, graphite and aluminium oxide, no catalytic effect that significantly reduces the self-ignition limit could be demonstrated at various pressures. It will also be necessary to investigate the causes of pre-ignition in the future. So far, it has not been possible to clarify beyond doubt the development of pre-ignition, even through reproducible basic investigations. Another path for investigation would be to operate the combustion chamber with hydrogen. Since pre-ignitions also occur in hydrogen engines, the consideration is that the mechanism cannot come from the fuel. As a single drop of oil is unlikely to be the cause of pre-ignition, deposits come more to the fore. The next step would be to analyse the chemical composition and reactivity of the deposits. It is conceivable that the interaction of fuel and oil results in reactive deposits, which then lead to pre-ignition and the calcium has no direct influence on the pre-ignition, but on the properties (brittleness) of the deposits. These investigations are being carried out at the Institute as part of the project Deutsche Forschungsgemeinschaft (DFG) SFB/ TRR 150. References [1] A. Heller, Motorwagen und Fahrzeugmaschinen für flüssigen Brennstoff: Ein Lehrbuch für den Selbstunterricht und für den Unterricht an technischen Lehranstalten aus dem Jahre 1912, 1st ed. Moers: Steiger, 1985. [2] United States. Congress, Congressional Record: Proceedings and Debates of the 90th Congress: U.S. Government Printing Office, 1967. [3] H. Barth, K. Ernst, and P. Papatheodorou, Eds., Toxikologie für Einsteiger, 1st ed. Berlin, Heidelberg: Springer Berlin Heidelberg; Imprint: Springer Spektrum, 2022. [4] B. Heil, „Zukunftswege des Verbrennungsmotors,“ Motortechnische Zeitschrift (MTZ), 15/ 2014, pp. 22-27, 2014. [5] C. Dahnz, K.-M. Han, U. Spicher, M. Magar, R. Schiessl, and U. Maas, „Investigations on Pre-Ignition in Highly Supercharged SI Engines,“ SAE Int. J. Engines, vol. 3, no. 1, pp. 214-224, 2010, doi: 10.4271/ 2010-01-0355. [6] K. A. Fletcher, L. Dingwell, K. Yang, W. Y. Lam, and J. P. Styer, „Engine Oil Additive Impacts on Low Speed Pre-Ignition,“ SAE Int. J. Fuels Lubr., vol. 9, no. 3, pp. 612-620, 2016, doi: 10.4271/ 2016-01-2277. A fundamental investigation of oil additives on pre-ignition in a high pressure combustion chamber 291 291 <?page no="292"?> [7] K. Fujimoto, M. Yamashita, S. Hirano, K. Kato, I. Watanabe, and K. Ito, „Engine Oil Development for Preventing Pre-Ignition in Turbocharged Gasoline Engine,“ SAE Int. J. Fuels Lubr., vol. 7, no. 3, pp. 869-874, 2014, doi: 10.4271/ 2014-01-2785. [8] K. Gschiel, K. Wilfling, and M. Schneider, „Development of a method to investigate the influence of engine oil and its additives on combustion anomalies in hydrogen engines,“ Automot. Engine Technol., vol. 9, no. 1, 2024, doi: 10.1007/ s41104-024-00141-7. [9] T. Hülser, G. Grünefeld, T. Brands, M. Günther, and S. Pischinger, „Optical Investigation on the Origin of Pre-Ignition in a Highly Boosted SI Engine Using Bio-Fuels,“ in SAE Technical Paper Series, 2013. [10] G. S. Jatana, D. A. Splitter, B. Kaul, and J. P. Szybist, „Fuel property effects on low-speed pre-ignition,“ Fuel, vol. 230, pp. 474-482, 2018, doi: 10.1016/ j.fuel.2018.05.060. [11] M. Kassai et al., „Research on the Effect of Lubricant Oil and Fuel Properties on LSPI Occurrence in Boosted S. I. Engines,“ in SAE Technical Paper Series, 2016. [12] M. Ohtomo, T. Suzuoki, H. Miyagawa, M. Koike, N. Yokoo, and K. Nakata, „Fundamental analysis on auto-ignition condition of a lubricant oil droplet for understanding a mechanism of low-speed pre-ignition in highly charged spark-ignition engines,“ International Journal of Engine Research, vol. 20, no. 3, pp. 292-303, 2019, doi: 10.1177/ 1468087417751240. [13] Y. Okada, S. Miyashita, Y. Izumi, and Y. Hayakawa, „Study of Low-Speed Pre-Ignition in Boosted Spark Ignition Engine,“ SAE Int. J. Engines, vol. 7, no. 2, pp. 584-594, 2014, doi: 10.4271/ 2014-01-1218. [14] S. Palaveev, M. Magar, H. Kubach, R. Schießl, U. Spicher, and U. Maas, „Premature Flame Initiation in a Turbocharged DISI Engine - Numerical and Experimental Investigations,“ SAE Int. J. Engines, vol. 6, no. 1, pp. 54-66, 2013, doi: 10.4271/ 2013-01-0252. [15] A. Ritchie, D. Boese, and A. W. Young, „Controlling Low-Speed Pre-Ignition in Modern Auto‐ motive Equipment Part 3: Identification of Key Additive Component Types and Other Lubricant Composition Effects on Low-Speed Pre-Ignition,“ SAE Int. J. Engines, vol. 9, no. 2, pp. 832-840, 2016, doi: 10.4271/ 2016-01-0717. [16] K. Rönn et al., „Low-speed pre-ignition and super-knock in boosted spark-ignition engines: A review,“ Progress in Energy and Combustion Science, no. 95, 2023, doi: 10.1016/ j.pecs.2022.101064. [17] F. Steeger, M. Günther, E. Stitterich, and S. Pischinger, „Effect of external oil sources in the air path on abnormal combustion phenomena of a turbocharged gasoline direct injection multi-cylinder engine,“ in International Conference on Ignition Systems for Gasoline Engines - International Conference on Knocking in Gasoline Engines, M. Sens, Ed.: expert verlag, 2022, pp. 421-437. [18] K. Takeuchi, K. Fujimoto, S. Hirano, and M. Yamashita, „Investigation of Engine Oil Effect on Abnormal Combustion in Turbocharged Direct Injection---Spark Ignition Engines,“ SAE Int. J. Fuels Lubr., vol. 5, no. 3, pp. 1017-1024, 2012, doi: 10.4271/ 2012-01-1615. [19] B. Tormos et al., „Experimental assessment of ignition characteristics of lubricating oil sprays related to low-speed pre-ignition (LSPI),“ International Journal of Engine Research, vol. 23, no. 8, pp. 1327-1338, 2022, doi: 10.1177/ 14680874211013268. [20] I. Volz, J. Pfeil, T. Koch, and F. Altenschmidt, „Investigating the Cause of Initial Pre-ignition - A New Approach,“ in Knocking in Gasoline Engines, M. Günther and M. Sens, Eds., Cham: Springer International Publishing, 2018, pp. 37-54. 292 Jan Reimer, Jürgen Pfeil, Frank Altenschmidt, Thomas Koch 292 <?page no="293"?> [21] J. Willand, M. Daniel, E. Montefrancesco, B. Geringer, P. Hofmann, and M. Kieberger, „Grenzen des Downsizing bei Ottomotoren durch Vorentflammungen,“ Motortechnische Zeitschrift (MTZ), 05/ 2009, pp. 422-428, 2009. [22] N. Zöbinger, T. Schweizer, T. Lauer, H. Kubach, and T. Koch, „Experimental and Numerical Analysis on Two-Phase Induced Low-Speed Pre-Ignition,“ Energies, vol. 14, no. 16, 2021, doi: 10.3390/ en14165063. [23] J. Reimer, M. Zabihigivi, I. Volz, J. Pfeil, F. Altenschmidt, and T. Koch, „Basic investigations on the cause of initial pre-ignition in a constant volume combustion cell,“ in International Conference on Ignition Systems for Gasoline Engines - International Conference on Knocking in Gasoline Engines, M. Sens, Ed.: expert verlag, 2022, pp. 25-40. [24] M. Mayer, P. Hofmann, J. Williams, and D. Tong, „Vorentflammungseinfluss des Motoröls bei hochaufgeladenen Ottomotoren mit direkter Einspritzung,“ Motortechnische Zeitschrift (MTZ), 06/ 2019, pp. 42-47, 2019. [25] J. Schöler, „Visualisierung von Mehrphasen-Phänomenen in der Kolbengruppe eines direktein‐ spritzenden Ottomotors,“ Dissertation, Universität Duisburg-Essen, 2024. [26] M. Ohtomo, H. Miyagawa, M. Koike, N. Yokoo, and K. Nakata, „Pre-Ignition of Gasoline-Air Mixture Triggered by a Lubricant Oil Droplet,“ SAE Int. J. Fuels Lubr., vol. 7, no. 3, pp. 673-682, 2014, doi: 10.4271/ 2014-01-2627. A fundamental investigation of oil additives on pre-ignition in a high pressure combustion chamber 293 293 <?page no="295"?> Simulating Fuel Ignition and Combustion in IC Engines with Lagrangian-Eulerian Spark Ignition (LESI) Model and Detailed Chemistry Lu Li, Pradeep Sapkota, Pinaki Pal, Yee Chee See, Mingyi Liang, Josep Gomez-Soriano, Sameera Wijeyakulasuriya, Riccardo Scarcelli, Ricardo Novella Lu Li Convergent Science Inc, Madison, WI, USA Yee Chee See Convergent Science Inc, Madison, WI, USA Sameera Wijeyakulasuriya Convergent Science Inc, Madison, WI, USA Pradeep Sapkota Convergent Science Inc, Madison, WI, USA Mingyi Liang Convergent Science Inc, Madison, WI, USA Riccardo Scarcelli Argonne National Laboratory, Lemont, IL, USA Pinaki Pal Argonne National Laboratory, Lemont, IL, USA Josep Gomez-Soriano CMT---Clean Mobility & Thermofluids, Universitat Politècnica de València, Camino de Vera s/ n, 46022 Valencia, Spain Ricardo Novella CMT---Clean Mobility & Thermofluids, Universitat Politècnica de València, Camino de Vera s/ n, 46022 Valencia, Spain <?page no="296"?> Abstract The Lagrangian-Eulerian Spark Ignition (LESI) model developed by Argonne National Laboratory (ANL) has been integrated into the commercial CFD software CONVERGE. The current implementation contains several options to model the breakdown phase of the spark event, which was not included in the original model. Using Lagrangian particles to deposit energy during the spark event, LESI accurately captures the electrical spark discharge and its interaction with the surrounding flow. In a previous publication by the authors of the LESI model, the G-Eqn combustion model was used to demonstrate the usage of LESI in engine combustion. In this work, the usage of LESI has been demonstrated with detailed chemistry modeling using three different IC engine cases: a gasoline direct injection (GDI) case, a gasoline port fuel injection (PFI) case, and a hydrogen port fuel injection (H2-PFI) case. Multiple cycles simulated from each case show good agreement with measured cylinder pressure without any case-by-case model parameter tuning. The paper also compares predictions from the LESI model with a simple spherical energy deposition method to simulate the engine spark event. Keywords— Internal Combustion Engines, Lagrangian-Eulerian Spark Ignition, Detailed Chemistry Nomenclature AMR Adaptive Mesh Refinement ATDC After Top Dead Center CA Crank Angle CAD Crank Angle Degree CFD Computational Fluid Dynamics COV Coefficient of Variation dATDC Crank Angle Degree After Top Dead Center DI Direct Injection DOI Duration of Injection EGR Exhaust Gas Recirculation EVC Exhaust valve closing EVO Exhaust valve opening EPA Environmental Protection Agency GDI Gasoline Direct Injection ICE Internal Combuston Engines IMEP Indicated Mean Effective Pressure IVC Intake valve closing 296 Li, Sapkota, Pal, See, Liang, Gomez-Soriano, Wijeyakulasuriya, Scarcelli, Novella 296 <?page no="297"?> IVO Intake valve opening LESI Lagrangian-Eulerian Spark Ignition MOP Maximum Open Position PFI Port Fuel Injection PISO Pressure-Implicit with Splitting of Operators RANS Reynolds Averaged Navier-Stokes RNG Re-Normalization Group RPM Revolutions per Minute SA Spark Advance SI Spark Ignition SOI Start of Injection I. Introduction Accurately modeling ignition processes in internal combustion engines is crucial for optimizing performance, enhancing combustion and fuel efficiency, and meeting stringent emissions standards. Efficient energy deposition during the early stages of ignition, from spark initiation to flame propagation, directly impacts combustion efficiency and engine operation under varying conditions. High dilution rates and lean fuel-air mixtures are increasingly adopted to achieve lower emissions and higher thermal efficiency. However, these innovations pose challenges to ignition reliability and stable combustion due to increased combustion variability and instability under fuel lean conditions. These factors underscore the critical need for advanced ignition modeling techniques that can accurately predict energy deposition and ignition timing, which in turn affect flame propagation and combustion stability, in modern internal combustion engines. Spark ignition in internal combustion engines progresses through several distinct stages crucial for efficient combustion. It begins with the plasma energy deposition phase, where a high-voltage electrical discharge from the spark plug ionizes the air-fuel mixture, creating a conductive path for current flow. Following plasma deposition, the breakdown phase intensifies the spark discharge, expanding the ionized gases and further developing the spark channel [1]. Once established, the flame kernel interacts with the surrounding flow turbulence which results in transport, stretching, local quenching, and re-ignition. Phenomena like spark short-circuit, re-strike, and blowout also play significant roles in accurate spark ignition modeling. To accurately model the spark ignition process so that it can be used in limiting combustion phenomena as discussed above, it is crucial that ignition models include all these detailed phenomena and interactions. This ensures that simulations can predict ignition timing, flame propagation, and combustion stability with high fidelity. The Lagrangian-Eulerian Spark Ignition (LESI) model [2-5], developed by Argonne National Laboratory, takes advantage of both Lagrangian and Eulerian frameworks, where Simulating Fuel Ignition and Combustion in IC Engines 297 297 <?page no="298"?> the model tracks the arc using Lagrangian particles and deposits energy into Eulerian computational cells. The spark modeling begins with evenly distributed energy source particles arranged in a straight line between the two spark electrodes. The particles are advected with the local flow velocities, which results in arc/ spark kernel elongation. A secondary circuit predicts the voltage, current, and power based on the remaining secondary circuit energy and the spark channel length. This energy is then deposited into the computational cells that contains the particles. Blowout is a phenomenon in which the spark kernel gets extinguished. Re-strike is the phenomenon of the regeneration of the spark kernel after the blowout. These two phenomena are modeled in the glow and arc phase, if the secondary circuit current drops below a threshold. Re-strike will happen if the remaining secondary circuit energy is larger than the breakdown energy of the system; otherwise, a blowout will occur and the spark discharge process stops. With all of these features, the LESI model was found to be in good agreement with visualized experimental data at non-quiescent engine-like conditions [2]. A breakdown model is added to the LESI model in the current work to correctly simulate the breakdown phase of the spark ignition process. Two options are available: the constant breakdown voltage model, which requires a user-input breakdown voltage, and one that depends on the spark length, local pressure, and temperature as proposed by Briggs et al [6]. In this paper, the first option is selected. Three different engine cases are simulated using the LESI model and detailed chemistry in this study: a gasoline direct injection (GDI) case, a gasoline port fuel injection (PFI) case, and a hydrogen port fuel injection case. For each engine case, 10 consecutive engine cycles are simulated and compared against the experimental cylinder pressure data. The results demonstrate that the LESI model can accurately capture the engine spark energy deposition, which supports flame propagation and accurate cylinder pressure rise rate without case-by-case tuning of the model parameters. The paper is organized as follows: first, the LESI spark ignition model details are presented, followed by the information about the engine test cases and relevant experimental data used for validation. Numerical case setup details are discussed next along with the mesh strategies and the chemical kinetics mechanisms used for combustion modeling. The results, discussion, and conclusions of the study are presented last. II. Spark modeling Accurate spark modeling is essential for high fidelity combustion simulations in internal combustion engines. Traditionally, a shape-based (usually a sphere) uniform Eulerian source, combined with a refined mesh, is used to capture kernel formation and propagation [7]. However, this approach fails to account for the detailed breakdown description in the gas column, the stretch and elongation of spark arc, and the changes in deposited energy. In this study, we use the LESI model [2] with a secondary circuit model [5] to simulate the spark ignition process in the cases considered. The secondary circuit refers to the circuit connected to the secondary winding of the ignition coil, as shown in Figure-1. In the LESI model, the arc in the spark gap is represented as a segmented line of Lagrangian particles that are tracked in space and time. During ignition, energy is deposited to the gas-containing Eulerian cells occupied by these Lagrangian particles. To keep the 298 Li, Sapkota, Pal, See, Liang, Gomez-Soriano, Wijeyakulasuriya, Scarcelli, Novella 298 <?page no="299"?> Figure 1: Circuit diagram showing the primary cir‐ cuit and the secondary circuit for a simplified spark ignition system. (1) (2) (3) energy deposition more uniformly distributed, the amount of energy deposited by each particle is further weighted by the length of segmented line connected to the particle. These Lagrangian particles are advected using an averaged velocity field that also accounts for the velocities of the neigh‐ boring cells to simulate the spark arc elon‐ gation. To keep the particles moving only around the electrodes, this model affixes the endpoint of the segmented line to elec‐ trodes and truncates particles located too close to the electrodes. As the spark arc is being elongated, the particles can get farther apart, and the shape of the spark arc may not be well represented by the particles if they become too sparse. To alleviate this problem, new particles are added between two consecutive particles of i and i + 1 if the condition shown in Eq. (1) is satisfied: dx i, i + 1 > F dx 0 where dx i, i + 1 is the distance between particles i and i+1, F is a user supplied factor, and dx 0 is the initial distance between particles at the start of the ignition event. In the LESI model, the spark discharge starts with the electrical breakdown of gas within the spark gap. During the time duration of the breakdown phase (t bd ), the total energy deposited (E bd ) by the LESI model to the gas is calculated as shown in Eq. (2): E bd = 12 C plug V bd 2 where C plug is the capacitance of the spark plug and V bd is the breakdown voltage. In the current implementation of the LESI model in the CFD code CONVERGE [8], the breakdown voltage can either be an input parameter or be calculated following Eq. (3), as proposed by [6]: V bd = apl spk I E T gas V applied where a is a model coefficient, l spk is the spark channel length, I E is the ionization energy, V applied is the applied voltage, p is the cylinder pressure, and T gas is the averaged temperature across the spark gap. After the breakdown phase, the secondary circuit model [9] is used together with the LESI model to determine the instantaneous energy to be deposited to the gas during the glow and arc phases. The energy remaining in the secondary circuit (E sc ) is modeled by the ordinary differential equation shown in Eq. (4): Simulating Fuel Ignition and Combustion in IC Engines 299 299 <?page no="300"?> (4) (5) (6) (7) (8) (9) dE sc dt = − i 2 R plug − iBp m l spk i n − i V anode + V catℎode where i is the circuit current given in Eq. (5), R plug is the spark plug’s resistance, p is the average cylinder pressure, V anode is the anode voltage, V catℎode is the cathode voltage, and B, m, and n are model parameters. i = 2E sc / L s L s is the secondary circuit inductance. Eq. (4) is integrated using a forward Euler scheme, and the amount of energy deposited by Lagrangian particles, given by Eq. (6), is evaluated at the current time step, which is the second right-hand side term in Eq. (4) multiplied by the time-step size dt. E spk = iV spk dt = iBp m l spk i n During the glow and arc phase, the LESI model can predict blowout or re-strike if the circuit current i drops below the current threshold proposed by Sayama et al. [10], shown in Eq. (7): i tℎ = βl spk b 1 where β and b 1 are model parameters. Upon satisfying the above criteria, re-strike is predicted if the energy remaining in the secondary circuit exceeds the breakdown energy shown earlier in Eq. (2). Otherwise, a blowout occurs, and the spark discharge process stops. Short-circuit is another phenomenon that the LESI model can predict. To check for pos‐ sible short-circuits, the model first evaluates the voltage difference between all Lagrangian particles following V i, j = V spk l i, j l spk ∀i, jϵ 1, 2, … . N and i ≠ j where l i, j is the length of line segment connecting the particles i and j and N is the number of Lagrangian particles. If any of the V i, j exceeds a voltage threshold, the line segment between the particles i and j is replaced with a straight line. This voltage threshold, V tℎ , is proposed by Sayama et al. [10] and is shown in Eq. (9): V tℎ = Dδl d 1 i d 2 where δl is the length of a hypothetical straight line connecting the two particles and D, d 1 , and d 2 are model parameters. In the scenario where this threshold is satisfied by more than one V i, j , the shortening procedure is only performed to the pair of i, j with the largest V i, j . 300 Li, Sapkota, Pal, See, Liang, Gomez-Soriano, Wijeyakulasuriya, Scarcelli, Novella 300 <?page no="301"?> III. Engine experimental data for model validation In this paper, cylinder pressure data from three different engines are used to validate the LESI model: a gasoline GDI engine, a gasoline PFI engine, and a hydrogen PFI engine. Detailed engine specifications and experimental data are discussed in past publications and are summarized in this section. The GDI engine used in this study is a single-cylinder engine, with engine description and experimental pressure data over 500 cycles detailed by Scarcelli et al. [11]. Table I and Table II provide the engine specifications and test condition used for validation, respectively. Engine GDI Displacement 0.626 L Bore 89.04 mm Stroke 100.6 mm Compression Ratio 12.1: 1 Intake valve MOP 100 dATDC-compression Exhaust valve MOP 255 dATDC-compression GDI Injector 6 hole, solenoid Injection Pressure 150 bar Spark system Coil-based, 0.7 mm gap Fuel EPA Tier II EEE T A B L E I: GDI E N G I N E S P E C I F I C A T I O N S Engine GDI Engine Speed (RPM) 2000 Engine load---IMEP (bar) 6 EGR (%) 0 Relative air fuel ratio, λ 1 Start of Injection (dATDC) -300 Duration of Injection ( ◦ ) 58 Ignition Energy (mJ) 75 Spark Advance (dATDC) -24 Number of Pulses 1 Ignition Duration (ms) 0.5 Ignition Duration ( ◦ ) 6 Simulating Fuel Ignition and Combustion in IC Engines 301 301 <?page no="302"?> Engine GDI COV IMEP (%) (over 500 cycles) 1.5 COV PMAX (%) (over 500 cycles) 1.5 T A B L E II: S P E C I F I C A T I O N S O F T H E O P E R A T I N G P O I N T The gasoline port fuel injection engine used is a standardized single-cylinder CFR (coop‐ erative fuel research) engine at Argonne National Laboratory. Full details of the engine experiments can be found in Pal et al. [12]. The engine specifications are listed in Table III. This study focuses on a heavy knock case with a spark timing of SA = -13 dATDC,. Engine PFI Combustion chamber Cast iron, flat “pancake” Compression ratio Adjustable, 11.0: 1 Bore (mm) 82.55 Stroke (mm) 114.3 Connecting rod (mm) 254 Intake valve 180 shroud, no rotation Exhaust valve No shroud, rotating Valve overlap Positive 5 CAD Fuel system Carbureted Ignition Capacitive discharge coil to spark T A B L E III: G A S O L I N E P F I E N G I N E S P E C I F I C A T I O N S For the hydrogen PFI engine, the experimental measurements were obtained from the work by Molina et al. [13]. The engine uses a port fuel injection system. In this study, we compare the numerical results for a mid-load lean case with λ=2.4. The engine description along with the details of the test facility have been discussed in previous studies [14, 15]. The engine specifications are listed in Table IV. Engine PFI Number of cylinders 1 Displaced volume 454.2 cm 3 Cylinder diameter 82.0 mm Stroke 86.0 mm Compression ratio 10.7 302 Li, Sapkota, Pal, See, Liang, Gomez-Soriano, Wijeyakulasuriya, Scarcelli, Novella 302 <?page no="303"?> Engine PFI Connecting rod length 144.0 mm Injection systems PFI Ignition system Spark plug Valves per cylinder 2 intake, 2 exhaust Intake Valve opening (IVO) -380 CAD Intake Valve closing (IVC) -135 CAD Exhaust Valve opening (EVO) -600 CAD Intake Valve closing (EVC) -338 CAD T A B L E IV: H Y D R O G E N P F I E N G I N E S P E C I F I C A T I O N S IV. Numerical setup In this study, the computational framework for the three engine simulations uses the CONVERGE version 4.0 CFD solver [8]. An overview of the grid setup, numerical methods, and physical models used in this work is discussed in this section. A. Grid setup CONVERGE uses a cut-cell Cartesian approach to generate the volume mesh at runtime [8]. The base mesh size in CONVERGE indicates the coarsest cell used in the computational domain. With Adaptive Mesh Refinement (AMR), regions with large flow curvature (2nd derivative of flow variables with respect to spatial variables) can be refined automatically based on flow variables like velocity and temperature. In addition, local fixed embedding can be added to further refine specific areas of interest. In all three engine simulations, a base mesh size of 4 mm is used. Temperature-based AMR is activated to capture the flame propagation, while velocity-based AMR is used to capture the complex flow and turbulence evolution within the engines. For the gasoline engine cases (both GDI and PFI), the temperature AMR and velocity AMR have a minimum cell size of 0.5 mm, while for the hydrogen PFI engine, the temperature AMR has a minimum cell size of 0.25 mm due to the smaller flame thicknesses of hydrogen compared to gasoline. Boundary-based, cylindrical, and spherical shape-based fixed embeddings are used to locally refine the mesh when necessary. The minimum cell sizes used around the spark plug are 0.125 mm and 0.0625 mm for the gasoline and hydrogen cases, respectively, to accurately capture the flow and the spark kernel formation and development. The AMR and spark source embedding information is summarized in Table V. Simulating Fuel Ignition and Combustion in IC Engines 303 303 <?page no="304"?> Engines GDI Gasoline-PFI H2-PFI Velocity AMR 0.5 mm 0.5 mm 0.5 mm Temperature AMR 0.5 mm 0.5 mm 0.25 mm Spark embedding 0.125 mm 0.125 mm 0.0625 mm T A B L E V: MESH SETUP B. Physical models The renormalization group k-eps turbulence model [16] was selected to model the turbu‐ lence, the O’Rourke and Amsden wall heat transfer model [17] is used to account for wall heat transfer, and the standard wall function [18] is used to calculate the velocities around the walls. The LESI spark model is used to simulate the energy deposition process. The spark plug is divided into two parts: the ground and the electrode. The initial position of the spark line must be defined, typically from the electrode to the ground surface. Additionally, the number of particles to initiate in the spark kernel is also specified; in this case, 9 particles are used for all three scenarios. Initially, the particles are evenly distributed along the spark gap. The initial secondary circuit energy, the source start timing, and the reinitialization timing are also set. Between the source start and reinitialization times, the deposited energy at each time step is calculated by the model. If a blowout occurs and the remaining energy is less than the breakdown energy, the LESI energy deposition ceases (even if it is before the reinitialization time). Upon reaching the reinitialization time, the model parameters are reset to their initial values in preparation for the next engine cycle. The reinitialization time in all simulations is set to 25-35-CAD after the spark start timing for all three cases. The SAGE detailed chemistry model [8] is selected to simulate the ignition and combus‐ tion within the engines. The adaptive zoning method [19], in which the cells with similar temperature and equivalence ratio are grouped before solving for combustion, is used to speed up the chemistry calculations. For the GDI engine case, the liquid parcel injection model is included to simulate the gasoline injection, breakup, and evaporation. The blob injection model [20] is selected where the initial parent parcel has the same size as the effective nozzle hole diameter, and the Kelvin Helmholtz (KH) Rayleigh Taylor (RT) breakup model [21] is used for spray atomization. C. Chemical kinetics mechanisms The chemistry mechanism is important for the accurate prediction of ignition and combus‐ tion propagation. For the GDI engine simulation, a chemical mechanism with five fuel components to model gasoline was reduced starting from the C3 mechanism [22]. This mechanism includes 221 species and 1472 chemical reactions. For the gasoline PFI engine simulation, a primary reference fuel mechanism by Jia et al. [23] was used, which included 56 species and 168 reactions. For the H2-PFI engine simulation, the hydrogen chemistry is extracted from C3 mechanism [22] and includes 13 species and 25 reactions. 304 Li, Sapkota, Pal, See, Liang, Gomez-Soriano, Wijeyakulasuriya, Scarcelli, Novella 304 <?page no="305"?> D. Numerical methods Figure 2: Different stages of LESI spark modeling and flame propagation in the GDI engine case, compared to the conventioanl spherical modeling. Note: N_particles is the number of source particles, and l_spk is the spark kernel length. Simulating Fuel Ignition and Combustion in IC Engines 305 305 <?page no="306"?> Transport equations were solved using the pressure implicit with splitting of operators (PISO) methods [24]. A second-order central differencing scheme is used to discretize the mass, momentum, energy, and species transport equations, while a first-order for‐ ward differencing scheme is used to discretize the turbulence transport equations. A first-order-accurate Euler backward implicit scheme is used for time integration. For the H2-PFI case, the molecular mass diffusivities of each species are calculated using a mixture-averaged diffusion model which accounts for preferential species diffusion. For the two gasoline cases, a single-species molecular diffusion method is adopted. Using a preferential species diffusion model to calculate mass diffusivity is important for simulating hydrogen combustion in ICEs to obtain accurate fuel-air mixing and laminar flame speeds, owing to the higher diffusivity of hydrogen in air (compared to heavy hydrocarbons like gasoline). V. Results and discussion Matching cylinder pressure is one of the important ways to validate the accuracy of an ICE simulation. For ICE cylinder pressure predictions, accurate initial conditions, boundary conditions, and geometry information is important. Thus, high-fidelity ignition modeling is key to predicting the combustion process accurately, which in turn is necessary for accurate cylinder pressure predictions. Cylinder pressure predictions from the three engine cases are compared against measured data in this section to validate the use of the LESI model in ICE simulations. 10 consecutively run engine cycles are presented along with multiple cycle measured pressure data for each case. As long as the predicted cylinder pressure is within the cycle-to-cycle variation (CCV) of the measured data, they are considered as valid and accurate. Figure-2(a) depicts the LESI spark modeling process for the GDI engine, with the base mesh measuring 4-mm. Within the cylinder region, a level 2 fixed embedding refines the mesh size to 1 mm (4/ 2² = 1). Using level 3 temperature and velocity AMR, the mesh size is further reduced to 0.5 mm. Around the spark plug, levels 4 and 5 fixed embedding achieve cell sizes of 0.25 mm and 0.125 mm at the center for spherical regions with radii of 2.5 mm and 5 mm, respectively. In this case, the spark occurs at -24 crank angle degrees (CA) and is reinitialized at CA = 10. At the initial stage (CA = -23.8), LESI starts to deposit energy into the computational cells, with the particle positions shown in the figure. At this point, there are 8 particles (initially 9, controlled by user input) with a spark length of 6.7 × −4-m. By CA = -22.8, the flame kernel begins to form, and additional parcels are added due to the elongated spark length. At CA = -21.8, the parcel count decreases from the previous time instance, indicating a short circuit event happening between these times. The kernel begins to propagate, influenced by local fluid flow. At later stages, CA = -17.8 and CA = -16.8, more parcels are introduced, and the spark length continues to grow. At the same time, due to the flame propagation, the flame front gets out of the source embedding region, where the cell size is controlled by temperature and velocity AMR, resulting in a 0.5-mm mesh size. At CA = -15.8, a re-strike occurs, reducing the parcel count to the initial 9 parcels. The high-temperature cells at this step are due to energy deposited in previous steps. Finally, at CA = -14.8, a blowout occurs, ending the spark process. Figure-2(b) depicts the same scenario with a fixed spherical source, where energy is deposited into a sphere with a radius of 0.0004-m. This case utilizes two spherical sources to simulate the spark energy profile, each 306 Li, Sapkota, Pal, See, Liang, Gomez-Soriano, Wijeyakulasuriya, Scarcelli, Novella 306 <?page no="307"?> with a value of 0.0375-J. Two spherical sources with equal total energy are used to replicate the energy release profile. The first source operates from CA = -24 to CA = -23.5, while the second source spans CA = -24 to CA = -18. This arrangement ensures a short-duration, high-energy release at the beginning (CA = -24 to CA = -23.5) to represent the breakdown phase, followed by a longer-duration, lower-intensity energy release during the arc and glow phases. A spark kernel forms initially (CA = -23.8), spreading uniformly within the sourced region. The sourced region does not move with the local flow as in the LESI model. As the process continues, the flame propagates outward, influenced by local flow dynamics. In contrast to the LESI case, the temperature due to flame propagation appears more spatially uniform in this scenario. Figure 3 shows the in-cylinder pressure comparison for the GDI engine. The mean, minimum, and maximum cylinder pressures from 500 experimentally measured engine cycles are overlaid with the 10 consecutively run simulation cycles using the LESI model and the conventional spherical spark source model. The objective of this paper is not to compare the predictions from the two spark modeling approaches, but rather to show that the new LESI model, as implemented in the commercial code CONVERGE, is capable of predicting cylinder pressure accurately, as is achieved using the conventional spark modeling approach. It is also not expected to capture the CCV of cylinder pressure of the 500 experimental cycles using just 10 simulation cycles. A similar comparison for the CFR engine is shown in Figure 4. 300 cycles of experimental pressure data is overlaid with 10 consecutive cycles from the CFD simulations with LESI and the conventional spherical source. The peak cylinder pressure timing and magnitude matches well with the experimental data. Figure 3: GDI engine results for 10 consecutive simulation cycles with LESI and the conventional spherical source compared to 500 experimental data cycles. Note: A Sage-multiplier of 1.3 is used. Simulating Fuel Ignition and Combustion in IC Engines 307 307 <?page no="308"?> Figure 4: Gasoline PFI engine results for 10 consecutive simulation cycles with LESI and the conven‐ tional spherical source compared to 300 experimental data cycles. Figure 5: H2-PFI engine results for 10 consecutive simulation cycles with LESI and the conventional spherical source compared to 250 experimental data cycles. 308 Li, Sapkota, Pal, See, Liang, Gomez-Soriano, Wijeyakulasuriya, Scarcelli, Novella 308 <?page no="309"?> Figure 5 shows the comparison for the H2-PFI engine case. 250 measured cycles of cylinder pressure are overlaid with the 10 consecutively run simulation cycles using LESI and the conventional spherical source. Similar to the gasoline engine cases, the LESI model is able to predict the ignition process of this H2-PFI engine accurately. VI. Conclusion This paper applies the Lagrangian-Eulerian Spark Ignition (LESI) model to different IC engine cases to evaluate its usability in accurately predicting the ignition process. Developed by Argonne National Laboratory and recently implemented in the commercial CFD software CONVERGE 4.0.2, the LESI model is shown to capture the various phases of the ignition process accurately, which results in accurate combustion modeling. With the LESI model, the spark kernel elongation, short-circuiting of the kernel, re-strike, and blowout of the spark can be modeled accurately as shown in this work. Coupled with detailed chemistry modeling, the LESI model is shown to predict the cylinder pressure accurately for GDI, gasoline PFI, and H2-PFI engines. The paper also presents the cylinder pressure predictions when a conventional spark modeling approach is used to demonstrate that the LESI model is able to produce similar levels of accuracies. However, the advantages of using a detailed spark model like LESI over the conventional spherical spark modeling approach is not discussed in this paper. The authors intend to demonstrate the capabilities of the LESI model in capturing CCV of certain limiting engine combustion phenomena in a future publication. It should also be noted that the LESI model parameters were not tuned case-by-case to achieve the presented predictions. References [1] Heywood, J.B., Internal Combustion Engine Fundamentals (McGraw-Hill, 2018) [2] Scarcelli, R., Zhang, A., Wallner, T., Som, S., Huang, J., Wijeyakulasuriya, S., Mao, Y., Zhu, X., and Lee, S.-Y., 2019. Development of a hybrid lagrangian-eulerian model to describe spark-ignition processes at engine-like turbulent flow conditions. Journal of Engineering for Gas Turbinesand Power, 141(9). [3] Kazmouz, S. J., Scarcelli, R., Kim, J., Cheng, Z., Liu, S., Dai, M., Lee, S. Y. (2023). High-fidelity energy deposition ignition model coupled with flame propagation models at engine-like flow conditions. Journal of Engineering for Gas Turbines and Power, 145(5), 051022. [4] Kazmouz, S. J., Scarcelli, R., Cheng, Z., Dai, M., Pomraning, E., Senecal, P. K., Sjo¨berg, M. (2022). Coupling a Lagrangian-Eulerian Spark-Ignition (LESI) model with LES combustion models for engine simulations. Science and Technology for Energy Transition, 77. [5] Kazmouz, S. J., Scarcelli, R., Bresler, M., Blash, E., Hardman, K. (2023). A comprehensive model to capture electrical discharge and spark channel evolution during spark-ignition processes. Combustion and Flame, 248, 112589. [6] Briggs, T. and Gonway, G., A predictive model for spark stretch and mixture ignition in SI engines, SAE Paper 2023-010202, 2023. [7] Probst, D., Wijeyakulasuriya, S., Pal, P., Kolodziej, C., Pomraning, E. (2020, November). Accel‐ erating computational fluid dynamics simulations of engine knock using a concurrent cycles Simulating Fuel Ignition and Combustion in IC Engines 309 309 <?page no="310"?> approach. In Internal Combustion Engine Division Fall Technical Conference (Vol. 84034, p. V001T06A002). American Society of Mechanical Engineers. [8] Richards, K.J., Senecal, P.K. and Pomraning, E. “CONVERGE 4.” (2024). [9] Kim, J., Anderson, R. W. (1995). Spark anemometry of bulk gas velocity at the plug gap of a firing engine. SAE transactions, 2256-2266. [10] Sayama, S., Kinoshita, M., Mandokoro, Y., Masuda, R., Fuyuto, T. (2018). Quantitative optical analysis and modelling of short circuits and blow-outs of spark channels under high-velocity flow conditions (No. 2018-01-1728). SAE Technical Paper. [11] Scarcelli, R, Sevik, J, Wallner, T, Richards, K, Pomraning, E, and Senecal, PK. “Capturing Cyclic Variability in EGR Dilute SI Combustion Using Multi-Cycle RANS.” Proceedings of the ASME 2015 Internal Combustion Engine Division Fall Technical Conference. Volume 2: Emissions Control Systems; Instrumentation, Controls, and Hybrids; Numerical Simulation; Engine Design and Mechanical Development. Houston, Texas, USA. November 8-11, 2015. V002T06A010. ASME. [12] Pal, P., Kolodziej, C., Choi, S., Som, S., Broatch, A., Gomez-Soriano, J.,-… See, Y. C. (2018). Development of a virtual CFR engine model for knocking combustion analysis. SAE International Journal of Engines, 11(6), 1069-1082. [13] Molina, S., Ruiz, S., Gomez-Soriano, J., Olcina-Girona, M. (2023). Impact of hydrogen substitution for stable lean operation on spark ignition engines fueled by compressed natural gas. Results in Engineering, 17, 100799. [14] Gomez-Soriano, J., Sapkota, P., Wijeyakulasuriya, S., D’Elia, M., Probst, D., Viswanathan, V.,-… Novella, R. (2023). Numerical Modeling of Hydrogen Combustion Using Preferential Species Diffusion, Detailed Chemistry and Adaptive Mesh Refinement in Internal Combustion Engines (No. 2023-24-0062). SAE Technical Paper. [15] Sapkota, P., Wijeyakulasuriya, S., Liu, S., Zhang, Y., Parmar, B., Gomez-Soriano, J. (n.d.). Accelerating detailed chemistry simulations of hydrogen combustion in IC engines using Reynolds Averaged Navier Stokes (RANS) turbulence modeling and a thickened flame model. Manuscript submitted for publication to ASME [16] Yakhot, V., Orszag, S. A. (1986). Renormalization group analysis of turbulence. I. Basic theory. Journal of scientific computing, 1(1), 3-51. [17] Amsden, A.A., ” K IVA- 3 V: A B lock Structured K IVA Program for Engines with Vertical or Canted Valves,” Los Alamos National Laboratory Technical Report LA- 13313---M S, 1997 . [18] Launder, B. E., Spalding, D. B. (1983). The numerical computation of turbulent flows. In Numerical prediction of flow, heat transfer, turbulence and combustion (pp. 96-116). Pergamon. [19] Raju, M., Wang, M., Dai, M., Piggott, W. et al., Acceleration of Detailed Chemical Kinetics Using Multi-zone Modeling for CFD in Internal Combustion Engine Simulations, SAE Technical Paper 2012-01- 0135, 2012. [20] GT-Power User’s Manual GT-SuiteTM Version 6.1. Gamma Technologies; 2004 [21] Senecal, P. K., Pomraning, E., Richards, K. J., and Briggs, T. E., Multi-dimensional Modeling of Direct-injection Diesel Spray Liquid Length and Flame Lift-off Length Using CFD and Parallel Detailed Chemistry, SAE Technical Paper 2003-01-1043, 2003 [22] Dong, S., Wagnon, S. W., Maffei, L. P., Kukkadapu, G., Nobili, A., Mao, Q.,-… Curran, H. J. (2022). A new detailed kinetic model for surrogate fuels: C3MechV3. 3. Applications in Energy and Combustion Science, 9, 100043. 310 Li, Sapkota, Pal, See, Liang, Gomez-Soriano, Wijeyakulasuriya, Scarcelli, Novella 310 <?page no="311"?> [23] Liu, Y. D., Jia, M., Xie, M. Z., Pang, B. (2013). Development of a new skeletal chemical kinetic model of toluene reference fuel with application to gasoline surrogate fuels for computational fluid dynamics engine simulation. Energy & fuels, 27(8), 4899-4909. [24] Issa, R. I. (1986). Solution of the implicitly discretised fluid flow equations by operator-splitting. Journal of computational physics, 62(1), 40-65. Simulating Fuel Ignition and Combustion in IC Engines 311 311 <?page no="313"?> 1 Argonne National Laboratory, Lemont, IL, USA 2 Argonne National Laboratory, Lemont, IL, USA 3 Corresponding Author Email: rscarcelli@anl.gov 4 Argonne National Laboratory, Lemont, IL, USA 5 Sandia National Laboratories, Livermore, CA, USA Modeling the Impact of Mixture Formation on Ignition and Flame Propagation in a Hydrogen Direct-Injection Engine Yiqing Wang 1 , Ricardo Scarcelli 23 , Chao Xu 4 , and Ales Srna 5 1 Abstract Hydrogen (H 2 ) internal combustion engines continue to receive increasing attention to decarbonize heavy transport applications such as on-road, off-road, and rail. H 2 direct injection (DI) enables high power output density and thermal efficiency and ultimately helps to mitigate the occurrence of backfiring. In-cylinder mixture formation from the H 2 DI process typically does not lead to a perfectly homogeneous air/ fuel mixture even with an early injection timing, which significantly affects processes like flame early development and propagation. Furthermore, a late injection timing increases the degree of mixture stratification at the time of the spark, which enhances the impact on flame propagation. In this paper, the main challenges in modeling H 2 ignition and combustion in a DI engine are evaluated step by step. Both Reynolds-Average Navier-Stokes (RANS) and Large Eddy Simulations (LES) are used for computational fluid dynamics (CFD) simulations of H 2 injection and combustion. First, the accuracy of CFD calculations for mixture formation predictions is evaluated against optical engine data of mixture distribution. Then, the impact of varying the ignition location on flame early growth and propagation is investigated. Results show the significant impact of the degree of in-cylinder mixture stratification on combustion and highlight the need to properly model the mixture formation process in H 2 DI engines. Keywords: Internal Combustion Engines, Hydrogen Direct Injection, Mixture For‐ mation, Flame Propagation 2 Introduction Hydrogen-fueled internal combustion engines (H 2 ICEs) have been receiving increasing attention lately, particularly in applications such as heavy-duty on-road and off-road transportation. Due to the unique characteristics of H 2 , such as low volumetric density, low minimum ignition energy, and high flame speed, the application of H 2 ICEs is facing <?page no="314"?> technical barriers due to primarily low power density and high-propensity for pre-ignition, especially when the port-fuel injection (PFI) approach is adopted [1] . Direct Injection (DI) of H 2 helps to avoid the backfiring and mitigates the occurrence of pre-ignition while increasing volumetric efficiency [2] . Nevertheless, DI might introduce a large degree of mixture stratification in the cylinder, which consequently affects the ignition and combustion process. More specifically, ignition and complete combustion under stratified conditions could be challenging even for an easy-to-ignite fuel like H 2 . Many efforts have been devoted in the past to model H 2 DI and validate CFD results of mixture formation against experimental data. Scarcelli et al. developed a CFD methodology based on Reynolds-averaged Navier-Stokes (RANS) modeling which was validated against planar laser-induced fluorescence (PLIF) data from a light-duty optical engine [3] . These CFD results showed very good agreement with experimental data in terms of jet penetration and overall evolution, while the mixing between the injected gas (H 2 ) and the ambient gas (air or nitrogen) was under-predicted, leading to noticeable differences in mixture distribution during the later stage of injection. Targeting the same dataset from the same optical engine, similar performance of the CFD model - still based on RANS, but using a different CFD software - was also reported in the past by Lucchini et al. [4] and Le Moine et al. [5]. Very recently, several additional CFD studies ([6], [7], [8]) targeting the same optical engine dataset focused on developing best-practices for CFD simulations of H 2 DI, including mesh alignment and model constant tuning, nevertheless, without solving the challenge of the numerical under-prediction of the in-cylinder mixing. Also, as of today, the impact of mixing stratification on combustion processes, as well as the impact of CFD predictions of in-cylinder mixing on the related CFD combustion simulations for H 2 ICEs has not been assessed. Finally, while most of the CFD studies on H 2 DI so far have continued to focus on a relatively old dataset using a light-duty engine platform and an old generation of gaseous injectors, the expected future main field of application for H 2 ICEs (heavy-duty on-road and off-road transportation) and a new generation of gaseous injection system demand for newer and more relevant data to be targeted by CFD simulations. Accordingly, this study aims at showing the impact of the H 2 DI and mixture formation process on ignition, early flame propagation, and combustion in a heavy-duty optical engine using a new generation of H 2 injectors. RANS and LES simulations of H 2 injection and combustion are performed and evaluated against recently measured optical data ([9], [10]). 3 Experimental Setup The engine platform simulated in this work is the Sandia Cummins single-cylinder heavy-duty optical engine [9] with bore of 139.7 mm and stroke of 152.4 mm, yielding displacement of 2.34 L with a geometric compression ratio of 10.75. A PHINIA medium-pres‐ sure outward opening hollow-cone prototype injector was used to inject H 2 directly into the cylinder. The injector was centrally mounted in the diesel-injector bore of the cylinder head. The injector delivered a nominal flow rate of 10 g/ s of H 2 at the injection pressure of 40 bar. Planar laser-induced fluorescence (PLIF) imaging of fluorescent tracer seeded into H 2 314 Yiqing Wang, Ricardo Scarcelli, Chao Xu, and Ales Srna 314 <?page no="315"?> was used to quantify in-cylinder H 2 mixture formation. The imaging arrangement consists of a horizontal light sheet configuration for probing mixing state at the spark timing. For this horizontal plane imaging, the sheet entered the combustion chamber parallel to the cylinder head through a cylinder-liner window and a cutout in the bowl-rim (see Figure 1). Detailed information about the engine setup and the experimental measurement technique for mixture formation studies can be found in [9]. Two injection timings, namely SOI = -120 and -60°CA, are considered in this CFD study to represent intermediate SOI and late SOI direct injection strategies, respectively. Laser-induced plasma ignition was used to ignite the H 2 -air mixture. For this purpose, a pulsed laser beam was focused within the engine combustion chamber to create a plasma at the focal point 14 mm below the fire-deck. The laser beam was focused by a 250 mm focal-length spherical lens mounted on a translation stage to shift the spark location. Three spark locations were used to explore the role of spark location on subsequent flame evolution (see Figure 1): a central position (later indicated as Center), one shifted +20 mm to the intake side (later indicated as Left), and one shifted -25 mm towards the exhaust side (later indicated as Right). The flame evolution after ignition was tracked by high-speed OH* chemiluminescence imaging. The chemically excited hydroxyl (OH*) radical is one of the few species in hydrogen combustion emitting chemiluminescence. This emission was recorded by a high-speed camera with an image intensifier. Detailed information about the engine setup and the experimental measurement technique for ignition and combustion studies can be found in [10]. Figure 1: Optical engine laser light-sheet arrangement for tracer-PLIF measurement (violet) and for laser-induced plasma ignition (green) with the location of spark indicated. Modeling the Impact of Mixture Formation on Ignition and Flame Propagation 315 315 <?page no="316"?> 4 CFD Model Setup Three-dimensional (3-D) CFD simulations are carried out using CONVERGE CFD code v3.0 [26], a finite volume code with the capability of simulating incompressible and compressible, multiphase, reacting flows in complex geometries with stationary and moving boundaries. Automated meshing relies on a base Cartesian grid and a cut-cell approach to generate a computational grid at every time step, with a combination of adaptive mesh refinement (AMR) and fixed embedding (FE) to refine the grid locally. Two turbulence models are considered in this work. The first is RANS with the renormalization group (RNG) k − ϵ model, following the best-practices and guidance provided in a previous study [8]. The second is LES with the dynamic structure model. Figure 2: Mesh strategy (including base mesh, FE, and AMR) for the H 2 DI and mixture formation calculations For the CFD simulations, an optimum mesh strategy that compromises between accuracy and computational turnaround is used, which was derived by previous mesh sensitivity studies, not shown here. The base mesh size in the entire cylinder region is set as 1 mm. Furthermore, two additionally refined regions are created through FE in the proximity of the injector, where the mesh size is progressively reduced down to 0.25 mm and 0.125 mm respectively (see Figure 2). These two FE-based refinements are activated only during the injection phase. Still during injection, the minimum cell size in the cylinder is further reduced down to 0.0625 mm 316 Yiqing Wang, Ricardo Scarcelli, Chao Xu, and Ales Srna 316 <?page no="317"?> (62.5 μm) by using AMR criteria based on the local velocity and temperature, with minimum sub-grid values of 0.5 m/ s and 2.5 K, respectively. The resulting mesh has a very large cell count (peaking between 30 and 50 million cells) and small time-step (about 5 ns during the injection process), thus leading to very expensive calculations. The maximum convection, diffusion, and Mach-based Courant-Friedrichs-Lewy (CFL) number limits are 1.0, 2.5 and 50.0, respectively. During the injection, the Mach-based CFL is reduced to 10.0 to better capture the supersonic jet. The turbulent Prandtl and Schmidt numbers are 0.9 and 0.78, respectively. A second-order central difference scheme is used for the convective flux discretization in the momentum, species and energy transport equations, while a first-order upwind scheme is used for the turbulence equation, as suggested in [8]. For the DI calculations, the same outward-opening hollow-cone injector used in experi‐ ments is modeled. A schematic of this injector was provided in [9] and is shown in a very simplified version in Figure 2. The actual SOI is 2°CA after the nominal SOI as suggested by the experiments. The inlet boundary condition of the injector is specified by matching the injected fuel mass, injection pressure (40 bar), and expected injection duration (17°CA). A valve lift profile is applied for the needle motion, which is not shown here as it is proprietary information. Since the pressure-ratio across the needle is larger than the critical ratio, which creates a supersonic flow and choked (i.e., constant) mass flow rate, the in-cylinder pressure is not expected to impact the injected fuel amount, and the fuel flow rate is roughly proportional to the injection pressure. For the combustion calculations, a finite-rate chemistry combustion model is used, which relies on the assumption of each computational cell to be a well-stirred reactor (WSR). The FFCM-2 mechanism with 12 species and 34 reactions [11] is used in the present study. For the ignition modeling, a total of 50 mJ thermal energy is deposited in a sphere with 0.5 mm radius at different locations as in the experiments, i.e., Center, Left, and Right. The ignition timing is -20°CA and the ignition duration is set as 0.072°CA (i.e., 10 μs). During the combustion event, the minimum cell size in the combustion chamber, delivered by AMR, is 0.25 mm (250 μm), which is consistent with what is typically used in practical engine simulations. Besides, to better capture the ignition and flame kernel development processes, two spherical FEs are applied around the ignition location, one with 3 mm radius and 62.5-μm mesh size and the other with 6-mm radius and 125-μm mesh size. 5 Results 5.1 Mixture formation calculations CFD results shown in Figures 3 and 4 highlight the key differences between LES and RANS calculations of the H 2 DI and in-cylinder mixture formation processes. It is worth noting that while the meshing strategy is the same for LES and RANS, larger velocity gradients captured by LES enhance the AMR algorithm to create more refined mesh during simulations. As can be seen in these figures, while the jet penetration and overall evolution are similarly described by the two different turbulence modeling methodologies, the finer structures captured by LES significantly improve the characterization of local mixing between the injected gas (H 2 ) and the in-cylinder gas (air or N 2 ) Modeling the Impact of Mixture Formation on Ignition and Flame Propagation 317 317 <?page no="318"?> Figure 3: LES and RANS simulations of DI and mixture formation for SOI-= -120°CA (intermediate injection timing). 318 Yiqing Wang, Ricardo Scarcelli, Chao Xu, and Ales Srna 318 <?page no="319"?> Figure 4: LES and RANS simulations of DI and mixture formation for SOI = -60°CA (late injection timing). Figure 5 shows the comparison between experiments (PLIF data) and CFD simulations of the mixture distribution (equivalence ratio) on a horizontal plane at the time of the spark. Also, the spark locations are indicated and denoted as L (Left), C (Center), and R (Right). It can be seen that both RANS and LES simulations can correctly predict a relatively richer mixture in the intake region and relatively leaner mixture in the exhaust region. The equivalence ratio is generally and progressively reduced while going from the Left side to the Right side of the horizontal plane in Figure 5. As a consequence of the evolution of the gaseous jet described above and shown in Figures 3 and 4, RANS simulations cannot properly predict the degree of mixing between the injected H 2 and the in-cylinder gas. Rather than a larger well-mixed region as shown in the experiments, RANS predict a Modeling the Impact of Mixture Formation on Ignition and Flame Propagation 319 319 <?page no="320"?> richer core of the jet and much steeper gradients at the jet boundaries. All these results are consistent with the previous literature [3-8]. LES simulations do a much better job in matching the degree of mixture stratification from the experiments while, as shown above, the gaseous jet overall evolution during the compression stroke is very consistent between LES and RANS. Therefore, RANS simulations with adequate mesh resolution can equally well predict jet collapse, penetration, wall impingement, etc., but significantly lack in gaseous mixing predictions. It is also worth noting that, for late injection timing, both RANS and LES predict very lean or ultra lean mixture conditions at the center spark location and especially at the right spark location. This large discrepancy with experiments can make accurate combustion CFD predictions very challenging. Figure 5: Comparison between experiments (PLIF) and LES and RANS simulations in terms of mixture distribution at the time of the spark. 5.2 Ignition/ combustion calculations Results shown in Figure 6 show the impact of mixture formation CFD predictions on combustion CFD predictions for the intermediate injection timing (SOI = -120°CA). Both RANS and LES predict that combustion is faster if initiated from the Left or Center spark location as compared with the Right spark location. Qualitatively, both RANS and LES agree very well with experiments on this trend. RANS simulations predict much faster combustion than experiments at early stages for the left/ center spark location (likely due to less mixing and therefore higher equivalence ratio at L/ C locations), and much slower combustion than experiments for the right spark location (likely due to less mixing and therefore lower equivalence ratio at R location). Surprisingly, LES simulations, that predict mixing and the equivalence ratio distribution at the time of the spark much better than RANS, show poor performance and very slow combustion rates at every location. Future work will focus on better understanding what at a first look seems to be a deficiency of the LES combustion model with the finite rate chemistry approach. 320 Yiqing Wang, Ricardo Scarcelli, Chao Xu, and Ales Srna 320 <?page no="321"?> Figure 6: Comparison between experiments and combustion LES/ RANS simulations in terms of inte‐ grated HRR for SOI =---120°CA. Figure 7: Comparison between experiments and combustion LES/ RANS simulations in terms of inte‐ grated HRR for SOI =---60°CA. Results shown in Figure 7 show the impact of mixture formation CFD predictions on combustion CFD predictions for the late injection timing (SOI = - 60°CA). As it was anticipated in the previous section (mixture formation studies), RANS and LES calcula‐ tions can only predict very fast and complete combustion for the Left spark location, Modeling the Impact of Mixture Formation on Ignition and Flame Propagation 321 321 <?page no="322"?> which is characterized by much richer mixture compared to experiments. RANS and LES simulations significantly struggle with predicting the same combustion behavior as from the experiments when the spark is applied in the Center or Right locations. Still, the combustion simulation results (LES spark/ combustion successful only from the Left and Center locations, RANS spark/ combustion successful only from the Left and Right locations) clearly correlate well with the mixture distribution observations (Figure 5, bottom row). Also, both LES and RANS combustion simulations capture the experimental trend that show combustion rates for late injection timing (SOI = -60°CA) to be quite higher than for intermediate injection timing (SOI = -120°CA). 6 Conclusions This paper focuses on establishing a link between CFD predictions of H 2 DI and mixture formation and combustion calculations in a heavy-duty optical engine equipped with a new generation of H 2 injectors. Both LES and RANS simulations are performed, with the focus on emphasizing the impact of different mixing predictions on combustion calculations. For mixture formation calculations, both LES and RANS show to be able to capture the general behavior of the injected gas (overall evolution including jet penetration, impingement on the wall, recirculation, etc.). Nevertheless, the fine turbulent structures captured by LES are shown to be decisive to improve mixing predictions versus RANS and more closely match optical data. RANS simulations significantly under-predict mixing between the injected gas (H 2 ) and the in-cylinder gas. For combustion calculations, both LES and RANS capture key general trends shown by experiments, like the impact of the spark location as well as the SOI on the burning rate. However, there is a strong correlation between the inaccuracy of mixture formation predictions and the inaccuracy of CFD combustion simulations that often are off-target, mostly because the mixture at the specific spark location is predicted to be either too rich or too lean. Finally, this study is not conclusive on which CFD technique is optimal, including both mixture formation and combustion. Future work on LES should focus on improving the combustion model, especially for well-mixed conditions (early injection). For RANS, future efforts will continue to be put on developing mixing sub-models, since a such poor mixture stratification prediction from RANS will unavoidably affect all the related CFD calculations, including regular combustion as well as abnormal combustion events (pre-ignition and knock). 7 Acknowledgments The submitted manuscript has been created by UChicago Argonne, LLC, Operator of Argonne National Laboratory (“Argonne”). Argonne, a U.S. Department of Energy Office of Science laboratory, is operated under Contract No. DE-AC02-06CH11357. The U.S. Government retains for itself, and others acting on its behalf, a paid-up nonexclusive, irrevocable worldwide license in said article to reproduce, prepare derivative works, 322 Yiqing Wang, Ricardo Scarcelli, Chao Xu, and Ales Srna 322 <?page no="323"?> distribute copies to the public, and perform publicly and display publicly, by or on behalf of the Government. This research is funded by DOE’s Vehicle Technologies Program (VTO), Office of Energy Efficiency and Renewable Energy. The authors would like to express their gratitude to Gurpreet Singh, Program Manager at DOE VTO, for supporting this work. Numerical simulations were run on the Bebop and Improve Clusters at the LCRC facility, Argonne National Laboratory. The authors would like to thank PHINIA and in particular Gavin Dober and Walter Piock for providing the injector geometry and operating conditions. Sandia National Laboratories is a multi-mission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC (NTESS), a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy’s National Nuclear Security Administration (DOE/ NNSA) under contract DE-NA0003525. This written work is co-authored by an employee of NTESS. The employee, not NTESS, owns the right, title and interest in and to the written work and is responsible for its contents. Any subjective views or opinions that might be expressed in the written work do not necessarily represent the views of the U.S. Government. 8 Literature [1] Verhelst, Sebastian and Wallner, Thomas. “Hydrogen fueled internal combustion engines.” Progress in energy and combustion science Vol. 35 No. 6 (2009): pp. 490-527. [2] Verhelst, Sebastian, Demuynck, Joachim, Sierens, Roger, Scarcelli, Riccardo, Matthias, Nicholas S and Wallner, Thomas. “Update on the progress of hydrogen-fueled internal combustion engines.” Renewable hydrogen technologies (2013): pp. 381-400. [3] Scarcelli, Riccardo, Wallner, Thomas, Matthias, Nicholas, Salazar, Victor and Kaiser, Sebastian. “Mixture formation in direct injection hydrogen engines: CFD and optical analysis of single-and multi-hole nozzles.” SAE International Journal of engines Vol. 4 No. 2 (2011): pp. 2361-2375. [4] Lucchini, Tommaso, D’errico, Gianluca and Fiocco, Marco. “Multi-dimensional modeling of gas exchange and fuel-air mixing processes in a direct-injection, gas fueled engine.” Technical report no. SAE Technical Paper. 2011. [5] Le Moine, Jerome, Senecal, PK, Kaiser, Sebastian A, Salazar, Victor M, Anders, Jon W, Svensson, KI and Gehrke, CR. “A computational study of the mixture preparation in a direct-injection hydrogen engine.” Journal of Engineering for Gas Turbines and Power Vol. 137 No. 11 (2015): p.-111508. [6] Babayev, Rafig, Andersson, Arne, Dalmau, Albert Serra, Im, Hong G and Johansson, Bengt. “Computational characterization of hydrogen direct injection and non-premixed combustion in a compression-ignition engine.” International Journal of Hydrogen Energy Vol. 46 No. 35 (2021): pp. 18678-18696. [7] Addepalli, Srinivasa Krishna, Pei, Yuanjiang, Zhang, Yu and Scarcelli, Riccardo. “Multi-dimen‐ sional modeling of mixture preparation in a direct injection engine fueled with gaseous hydrogen.” International Journal of Hydrogen Energy Vol. 47 No. 67 (2022): pp. 29085-29101. [8] Wu, Bifen, Torelli, Roberto and Pei, Yuanjiang. “Numerical modeling of hydrogen mixing in a direct-injection engine fueled with gaseous hydrogen.” Fuel Vol. 341 (2023): p.-127725. Modeling the Impact of Mixture Formation on Ignition and Flame Propagation 323 323 <?page no="324"?> [9] Laichter, Judith, Kaiser, Sebastian A, Rajasegar, Rajavasanth and Srna, Ales. “Optical Investigation of Mixture Formation in a Hydrogen-Fueled Heavy-Duty Engine with Direct-Injection.” Technical report no. SAE Technical Paper. 2023. [10] Laichter, J., Kaiser, S., Rajasegar, R., and Srna, A., “Impact of Mixture Inhomogeneity and Ignition Location on Early Flame Kernel Evolution in a Direct-Injection Hydrogen-Fueled Heavy-Duty Optical Engine,” SAE Int. J. Adv. & Curr. Prac. in Mobility 6(3): 1624-1644, 2024, https: / / doi.org/ 10 .4271/ 2023-32-0044. [11] https: / / web.stanford.edu/ group/ haiwanglab/ FFCM2/ 324 Yiqing Wang, Ricardo Scarcelli, Chao Xu, and Ales Srna 324 <?page no="325"?> Determining the Transition from Auto-Ignition to Knock in Methanol Operation by Application of the Bradley Theory Hermsen, Philipp (TME)*; Plum, Lukas (TME); Günther, Marco (TME); Pischinger, Stefan (TME); Blomberg, Michael (FEV Europe GmbH) Abstract To achieve the climate targets, a further reduction in the CO 2 footprint of the com‐ bustion engine is necessary. One suitable option is the introduction of an alternative fuel, in this case methanol, which has a lower carbon content and can be used instead of gasoline fuel in spark-ignition engines and can be produced green using renewable energy and CO 2 capture (C O R M O S , 2023). Compared to gasoline, methanol has a significantly higher resistance to knocking (S U I J S E T . A L , 2024). However, under extreme conditions, e.g. high compression ratio, knocking is also possible here. In order to optimize efficiency during operation, a fundamental understanding of the knocking behavior of methanol is necessary and should therefore be researched. Simulation models that resolve the auto-ignition of a hotspot during engine operation in a three-dimensional environment are suitable for this purpose. The auto-ignition of a hotspot in the unburned mixture in the combustion chamber does not necessarily lead to a knocking event. In order to differentiate between auto-ignition and knock, and also to evaluate the knock intensity of methanol in engine operation, an appropriate knock criterion is defined in this investigation. The "Deflagration To Detonation Diagram" from the Bradley Theory can be applied for the evaluation. Bradley defines different types of auto-ignition and declares a transition from deflagration to detonation depending on the reactivity of a hotspot and the inhomogeneity of the unburned mixture (B R AD L E Y E T . A L , 2002). The theory states that the pressure wave originating from the auto-ignition detonates the unburned environment if it propagates supersonically in the combustion chamber. If the pressure wave propagates at subsonic speed, the energy is not sufficient to ignite the unburned surroundings and therefore does not trigger knocking operation. In order to be able to model knocking in 3D-CFD studies, the Bradley theory described above is applied to methanol as a fuel in this study and a 3D-CFD reactor model is created. This is used to determine the auto-ignition behavior and in particular the transition to the knocking event with the use of simulations. A clear transition from deflagration to detonation has been determined for methanol. A simplified Bradley approach is implemented to the 3D-CFD RANS engine process simulation model and a correlation between the unburnt mixture state to the knock intensity has been defined. <?page no="326"?> 1 Introduction With regard to the climate change, a reduction of CO 2 emissions of internal combustion engines is mandatory. Beside the increasement of peak efficiencies due to enleanment, high compression ratios or hybridization, the introduction of synthetic fuels is a promising solution. A fuel like methanol, which can be produced green using renewable energy and CO 2 capture, has a significantly lower carbon content than conventional gasoline and diesel fuels and therefore also produces fewer CO 2 emissions (C O R M O S , 2023). By use of such synthetic fuels in spark-ignition engines, an impact onto the knock occurrence and knock behavior must be examined. Knocking combustion is damaging to engine components and also limiting the compression ratio and early spark timings. Therefore, knocking is limiting the achievement of peak efficiencies and needs to be further investigated (W AN G E T . AL , 2017). To estimate the knock boundaries in early development stages in future industrial applications, simulation models capturing knock statistics are required. By use of such models, knocking can be identified and thus, the engine process optimized. With regard to 3D-CFD engine process simulations, various simulation approaches exist. The most well-known are the Large-Eddy-Simulation (LES) and Reynolds-Averaged-Navier-Stokes (RANS) approaches. While the LES resolves the total turbulent field and is able to cover cyclic variations, the RANS averages turbulent fluctuations (P O P E , 2004). Accordingly, RANS is not able to cover statistical phenomena. However, knock occurs stochastically and the RANS cannot resolve it with regard to cyclic variations of knock intensity and knock frequency. The LES requires a lot of computational time and computational costs and therefore, it is not used in the industry and the cost-effective RANS approach finds application. (R O B E R T E T . A L , 2014) In the research project by FVV, “EKIM - Engine Knock Intesity Modeling for Future Fuels”, we are combining both 3D-CFD approaches, to develop a cost-efficient simulation model, covering cyclic variations to resolve knock intensity and knock frequency. Hence, we will use a RANS based approach and develop a statistical model extension. The statistical approach is shown in the following graphic. Figure 1: Statistical extension of a 3D-CFD RANS approach to predict knock boundaries 326 Hermsen, Philipp; Plum, Lukas; Günther, Marco; Pischinger, Stefan; Blomberg, Michael 326 <?page no="327"?> The model is first extended by a transported Livengood-Wu integral. It depends on the inverted Ignition Delay Time (IDT) in the unburnt mixture and is integrated until a value of c I,u = 1, when a critical state for auto-ingition is reached, defining the knock onset for the mean cycle (C H E N E T . A L , 2021). Afterwards, a Probability Density Function (PDF) is implemented to the model, calculating the variance of temperature and pressure to cover cyclic variations of the engine operation point, differing from the mean cycle (D’A DAM O , 2017). A second PDF is required, also to calculate the auto-ignition probability for the cyclic variations. To get from auto-ignition to knock, a simplified Bradley approach is implemented, also enabling a direct derivation of the knock intensity by the state of the unburnt mixture. For predicting the knock frequency, a further PDF is implemented. 2 Deflagration to Detonation Transition The Bradley theory states that the propagation speed of a pressure wave caused by the ignition of a hotspot can be used to differentiate between different self-ignition modes (B R AD L E Y E T . A L , 2002, 2004). For this investigation, the relevant auto-ignition modes are the deflagration (no knocking operation) and the transition to detonation (knocking operation). Accordingly, deflagration occurs when the pressure wave created by the auto-ignition propagates through the combustion chamber at subsonic speed. However, if the pressure wave propagates at supersonic speed, the energy contained in the pressure wave is high enough to immediately ignite the unburned mixture surrounded by the hotspot, resulting in detonation. The differentiated auto-ignition modes are shown in Bradley’s introduced Diagram for Deflagration to Detonation Transition (DDT), which is shown in the following picture. Figure 2: Auto-ignition modes in the Deflagration to Detonation Transition diagram (BRADLEY E T . A L , 2002) The x-axis ε represents the reactivity of a hotspot. It can be calculated by ε = rH S a τ e . (eq. 1) (B R AD L E Y AT . A L , 2002, 2004) Determining the Transition from Auto-Ignition to Knock in Methanol Operation 327 327 <?page no="328"?> The reactivity depends on the hotspot radius r HS , the sound of speed a and the excitation time τ e . The y-axis ξ represents the inhomogenities of the unburnt gas temperatures in the combustion chamber and is calculated by ξ = a u a , (eq. 2) (B R AD L E Y AT . A L , 2002, 2004) while u a is u a = ∂τ ∂T ∂T ∂r −1 . (eq. 3) (B R AD L E Y AT . A L , 2002, 2004) Therefore, ξ depends on the sound of speed a and the difference of the IDT between hotspot and surrounding. 2.1 Investigation of the Bradley Theory for methanol In order to use the Bradley Theory as a knock criterion, we first check whether it can be applied to methanol fuel. For this, a LES reactor model is built up, to investigate the different auto-ingition modes. Artificial hotspot conditions are imposed to the simulation model and the transition from deflagration to detonation is carried out. 2.1.1 3D CFD LES Reactor model A simple 3D-CFD reactor model is created, that different boundaries can be imposed, enabling to solve ε and ξ of the DDT (B L O M B E R G E T . A L , 2022). The model is illustrated in the following graphic. Figure 3: 3D-CFD LES constant volume reactor model for evaluation of auto-ignition modes The model geometry is based on the investigated test bench single-cylinder engine and corresponds approximately to the volume at the ignition timing. In the center of the reactor, a hotspot with a radius r HS with the initial temperature T HS is located. The surrounding is initialized with a lower temperature T cyl . The pressure in the hotspot and surrounding is equal. The fluid is a homogeneous, stoichiometric air/ methanol mixture. Solving the 328 Hermsen, Philipp; Plum, Lukas; Günther, Marco; Pischinger, Stefan; Blomberg, Michael 328 <?page no="329"?> gradient in the mixture ξ, based on eq.2 and eq.3, the difference of IDT of hotspot and surrounding can be solved, depending on temperature, pressure and air ratio, by use of a kinetic reaction mechanism, in this case Shrestha (S H R E S THA E T . A L , 2019). The sound of speed a is estimated by the acoustic speed with a = κ mixtur e * R mixtur e * T H S . (eq.4) (R AA B E E T . AL , 2021) The polytropic exponent is estimated to κ mixture = 1.3 and the specific gas value R mixture = 280 J/ (kgK). The reactivity of the hotspot ε is solved with eq.1. For the radius r, the hotspot radius r HS is imposed and the excitation time τ e , which is defined as delay between 5 % to 100 % energy release rate (B R AD L E Y AT . A L , 2002, 2004), it is also precalculated by 0D kinetic investigations and imposed for the initial hotspot conditions. 2.1.2 Evaluation of deflagrative and detonative combustion The qualitative evaluation of deflagration is illustrated using exemplary results. Here, a pressure of p = 60 bar is imposed. The air ratio is λ = 1. The hotsport temperature is set to T HS = 1050 K, the surrounding temperature T cyl = 1000 K. This results in an ignition delay time of the hotspot IDT HS -=-1.95e-4-s and an excitation time τ e -=-5.82e-7-s. The ignition delay time of the surrounding is IDT cyl -=-4.42e-4-s. With the given boundary conditions, self-ignition of the hotspot is expected. This leads to a pressure wave propagating into the surrounding area, while its self-ignition time has not yet been reached. For the evaluation of temperature and pressure in the surrounding, a radial distance from the hotspot border dx-=-5-mm is used for varied hotspot radi r HS . In the following graphic, the pressure amplitude and temperature for an initial hotspot radius r HS- =-2.75-mm-(left) and r HS -=-3.75-mm (right) is shown. Figure 4: Pressure Difference and Temperature results at different times for T HS = 1050 K, T cyl = 1000 K, p-=-60-bar and r HS -=-2.75-mm (left) and r HS- =-3.75-mm (right) Determining the Transition from Auto-Ignition to Knock in Methanol Operation 329 329 <?page no="330"?> At a time of t = 2.1e-4 s and hotspot radius r HS = 2.75 mm, the hotspot already auto-ignitied, derived from the increased temperature inside the hotspot. This causes a pressure wave starting to propagate in radial direction. At a time of t = 2.16e-4 s, this pressure wave reaches the monitor point. The temperature at this location is even on initial conditions, no ignition is taking place. The energy inside the pressure wave is not high enough to cause a detonation. Therefore, for the boundaries, a deflagration can be derived. By increasing the hotspot radius to r HS = 3.75 mm, this will affect ε and ξ in the DDT. At a time of t = 2.14e-4 s, the pressure wave reaches the monitor point. Now, the temperature increases at this point. The energy of the pressure wave reaches a value, that the surrounding detonates. At the time t = 2.25e-4 s, the pressure wave further propagates in the surrounding. The increased temperature always follows the pressure wave immediately. For these conditions, a detonation is reached. In general, to differentiate between deflagration and detonation, the pressure, tempera‐ ture and propagation speed of the pressure wave are evaluated at the monitor point. For a variation of the hotspot radius r HS , the transient results for the given boundary conditions are shown in the following diagram. The pressure (full lines) and the temperature (dotted lines) are shown in the left diagram. The resulting pressure wave propagation speeds and the sound of speed a are shown in the right diagram. Figure 5: Temperature, pressure and pressure wave speed evaluation at 5 mm distance from hotspot border for T HS -=-1050-K, T cyl- =-1000-K, p-=-60-bar and varied r HS In a time period t = 2.1e-4 s and t = 2.2e-4 s, the pressure wave passes the monitor point. With increasing hotspot radius r HS , the hotspot reactivity ε increases. Therefore, also the pressure amplitude increases. For hotspot radi r HS = 2.75 mm to r HS = 3.25 mm, the pressure wave is too small, causing an ignition of the surrounding, leading to a deflagrative combustion. Here, an advancement of the surrounding auto-ignition perceptible with increasing hotspot radius r HS . For hotspot radi r HS = 3.25 mm and r HS = 3.5 mm, the pressure wave causes an immediate ignition of the surrounding, reaching a detonative combustion. For the deflagrative boundaries, the pressure wave propagation speed is lower than the speed of sound. For detonative boundaries, the sound of speed is exceeded, validating the Bradley Theory for these particular conditions. 330 Hermsen, Philipp; Plum, Lukas; Günther, Marco; Pischinger, Stefan; Blomberg, Michael 330 <?page no="331"?> 2.1.3 Transition from deflagration to detonation As it was illustrated in the previous chapter, further artificial conditions were imposed to the simulation. Hence, a wide range of hotspot reactivity ε and mixture gradient ξ is imposed to carry out the transition from deflagration to detonation. The boundary conditions and the resulting DDT is shown in the following graphic. Figure 6: Varied boundary conditions for investigation of auto-ignition modes (left). Data plotted into DDT and evaluated deflagration and detonation (right) For the given boundary conditions, deflagrative and detonative combustions are evaluated, using the approach of chapter 2.1.2. A clear and consistent transition is shown. Therefore, the Bradley Theory is suitable approach for the definition of a knock criterion. Thus, it will be used for further investigations and implemented in the 3D-CFD RANS engine process simulation model in the next step. 3 Implementation of simplified Bradley approach in 3D CFD RANS engine process simulation model For the engine process application, a 3D CFD RANS approach is used. As this approach averages the turbulent field, it is not able to resolve the inhomogenities in the unburnt mixture. Accordingly, the mixture gradient ξ cannot be solved. Moreover, it does not resolute hotspots precisely. Therefore, a simplifaction of the Bradley approach is required. A submodel is developed, solving ε in each cell of the unburnt for each time step. For this, the sound of speed is directly resolved by the CFD. The excitation time τ e is pre-tabulated for engine-like conditions and interpolated by the submodel. The hotspot radius r HS is unknown and assumed to r HS- =-1-mm. Afterwards, ε is averaged in the unburnt volume. Aim is, to correlate the knock intensity, represented by the Knock Peak to Peak value (KPP), with the state of the unburnt ε. From this correlation, a trend curve will be derived. Later in the project, when the PDF for covering of cyclic fluctuations is implement in the model, ε will be calculated for the cyclic variations. Therefore, the knock intensities KPP can be derived by the trend curve for the scatterband of an engine operation point. Here, Determining the Transition from Auto-Ignition to Knock in Methanol Operation 331 331 <?page no="332"?> the unburnt state is calculated by the PDF and the hotspot radius is again set to r HS = 1 mm, as it was used for the definition of the trend curve. 3.1 Ignition angle sweep The correlation between KPP and ε requires a variation of both values. Therefore, an ignition angle sweep is performed. Here, an engine operation point of a measurement campagne with the single cylinder test bench is simulated. The engine speed is n = 1000 1/ min, indicated mean effective pressure IMEP = 18 bar and air ratio λ = 1. On the test bench, the spark timing was advanced, until reaching a peak pressure limitation of p max = 200 bar, not to cause engine component damages. To evaluate KPP in the simulation, a monitor point is located in the near of the pressure sensor location of the test bench engine. The difference between monitor point pressure and average in-cylinder pressure results in a pressure oscillation. Additionally, a filtering of this signal between f = 2000 Hz and f = 20000 Hz is necessary, representing usual frequencies of knock signals. The filtered pressure wave signal is used to calculate the knock intensity KPP. The comparison of the mean cycle simulation and the measurement data is shown in the following graphic. Figure 7: Comparison of ignition angle sweep between measurement and simulation. KPP vs. combus‐ tion center (left). Cylinder pressure trace (right) The scattering of the measurement data is shown for three ignition timing. While the RANS only covers the mean cycle, also the mean values of the measurement data is plotted. A good alignment of the average knock intensity KPP mean is achieved by the simulation. The combustion center has an error of up to 1 bar, possibly caused by tolerances of the pressure sensor used on the test bench. The pressure trace is well matched, while the most important criteria is to match the pressure gradient during combustion. 332 Hermsen, Philipp; Plum, Lukas; Günther, Marco; Pischinger, Stefan; Blomberg, Michael 332 <?page no="333"?> 3.2 Correlation of hotspot reactivity and knock intensity Determination of the discussed trendcurve of KPP and ε requires even higher KPP values. Here, the simulation is beneficial, as no limitiations with regard to pressure and temperature exist and earlier spark timing can be realized. Consequently, the spark timing was further advanced to achieve such higher knock intensities and for each spark timing, the maximum hotspot reactivity in the unburnt ε was evaluated by the implemented submodel. The correlation of KPP and ε and the derived trend function are shown in the following diagram. Figure 8: Evaluated hotspot reactivity in the unburnt ε vs. knock intensity KPP of ignition angle sweep in engine process simulation and derived trend curve By advancing the spark timing, higher knock intensities were achieved and thus, ε increases. The first two points (red) were not showing classical knock but noise in the simulation. Therefore, these two points were not used for the creation of the trend curve, as they are predicting too high KPP values. The ignition angle was further advanced compared to the test bench, as high KPP also might be relevant for lower load conditions or other engines, for which higher knock intensities are also permissible. The correlation between the two variables can be successfully illustrated using the derived trend function. 4 Summary and outlook For differentiating between auto-ignition and knocking operation, the Bradley Theory was validated for methanol fuel by imposing artificial hotspot conditions in a 3D-CFD LES reactor model and analyzing deflagrative and detonative combustion. A consistent transition was carried out by simulations. Therefore, the theory is valid fort he investigated fuel and can be used as knock criterion. As the 3D-CFD RANS engine process model does not resolve the total turbulent field, a simplified Bradley approach was developed and implemented by a submodel into the simulation model. By performing a virtual ignition angle sweep, a correlation between the knock intensity KPP and hotspot reactivity ε has been derived and trend fitted with an exponential function. Determining the Transition from Auto-Ignition to Knock in Methanol Operation 333 333 <?page no="334"?> In later stages of the FVV EKIM project, the 3D-CFD RANS simulation model will be extended by a PDF to cover pressure and temperature traces of cyclic variations. The KPP vs. ε trend curve will be used to evaluate the knock intensity of those cyclic fluctuations, representing the scattering of engine operation points. Therefore, a cost-efficient simulation model is delevoped, capturing statistics including the prediction of knock intensities and frequencies. 5 Bibliography B L O M B E R G , M., F A J T , N., L E Y E N S , L. (2022): Fast Knocking Prediction for Gasoline Engines, Final Report on FVV Project 1370, Frankfurt am Main: Research Association for Combustion Engines (FVV) e.V. B R A D L E Y , D., M O R L E Y , C., G U , X. J., E M E R S O N , D. R. (2002): Amplified Pressure Waves During Autoignition: Relevance to CAI Engines. B R A D L E Y , D, G U , X., E M E R S O N D. R. (2004): Modes of reaction front propagation from hot spots in flammable gaseous primixtures. C H E N , X., Z H A O , P., D A I , P., C H E N , Z. (2021): On the prediction of hot spot induced ignition by the Livengood-Wu integral. C O R M O S , C.-C. (2023): Deployment of integrated Power-to-X and CO 2 utilization systems: Techno-eco‐ nomic assessment of synthetic natural gas and methanol cases. D’A D A M O , A., B R E D A , S., F O N T A N E S I , S., I R I M E S C U , A., M E R O L A , S. S., T O R N A T O R E , C. (2017): A RANS knock model to predict the statistical occurrence of engine knock. L I N S E , D., K L E E M A N N , A., H A S S E , C. (2014): Probability density function approach coupled with detailed chemical kinetics for the prediction of knock in turbocharged direct injection spark ignition engines. P O P E , S. B. (2004): Ten questions concerning the large-eddy simulation of turbulent flows. R A A B E , A., H O L S T E I N , P. (2021): Akustik und Raumklima. R O B E R T , A., R I C H A R D , S., C O L I N , O., M A R T I N E Z , L., D E F R A N C Q U E V I L L E , L. (2014): LES prediction and analysis of knocking combustionin a spark ignition engine. S H R E S T H A , K. P., S E I D E L , L., Z E U C H , T., M A U S S , F. (2019): Kinetic Modeling of NOx Formation and Consumption during Methanol and Ethanol Oxidation. S U I J S , W., B R O E K A E R T , S., D E C U Y P E R , T., V E R H E L S T , S. (2024): The sensitivity of pressure-based knock threshold values to alternative fuels: A comparison of methanol vs. gasoline. W A N G , Zhi, L I U A , Hui, R E I T Z , Rolf D. (2017): Knocking combustion in spark-ignition engines. Z H E N , X., W A N G , Y., X U , S., Z H U , Y., T A O , C., X U , T., S O N G , M. (2011): The engine knock analysis - An overview. 6 Acknowledgements The presented work was performed at TME of RWTH Aachen University within the scope of research projects #1478 “EKIM - Knock model for future fuels” undertaken by the FVV (The Research Association for Combustion Engines eV). The authors gratefully acknowledge the support received from the chairmen Dr. Jonas Villforth (Porsche AG), the FVV, the working groups as well as all others involved in the project. In addition, the authors would like to thank Convergent Science GmbH for providing academic licenses. The 3D-CFD Simulations were performed with computing resources granted by RWTH Aachen University under the projects rwth1374 and rwth1407. 334 Hermsen, Philipp; Plum, Lukas; Günther, Marco; Pischinger, Stefan; Blomberg, Michael 334 <?page no="335"?> Observations on pre-ignition in a port injected heavy duty hydrogen internal combustion engine Thomas E. Briggs, Jr., Ph.D., Institute Engineer, Powertrain Engineering Division, Southwest Research Institute Introduction It is accepted that the transportation industry needs to reduce carbon emissions to mitigate climate change associated with greenhouse gas generation. Decarbonizing some segments of the transportation sector are more difficult than others based on energy and power demands. Generally, the smaller the power demands the easier it is for low carbon technologies to be adopted. Such is the case of battery electric vehicles for light-duty passenger cars. However, heavy-duty segments such as Class 8 long haul trucks use much greater energy quantities in a day that would need larger battery sizes that reduce cargo capacity. Hydrogen as an energy carrier can potentially be beneficial in these applications. Hydrogen is a carbon free fuel that has a higher energy storage potential than current batteries. Fuel cells are one emerging technology that can power heavy-duty vehicles by utilizing hydrogen. Fuel cells do not produce emissions other than water and are accepted as a Zero Emissions Vehicle (ZEV). While that is a promising technology that will need to be pursued, internal combustion engines that use hydrogen as the fuel source have been overlooked by the ZEV regulatory frameworks. The purpose of this project was to build a demonstration Class 8 truck with a hydrogen internal combustion engine as an advocacy program that hydrogen engines should be included as zero emissions. To achieve this an SCR based aftertreatment system was matched to the engine to reduce oxides of nitrogen (NO X ) to well below the upcoming EPA/ CARB 2027 regulatory limits. With no CO 2 from fuel consumption, and ultra-low NO X and N 2 O emissions, it is believed that this type of platform should be considered a ZEV. The hydrogen internal combustion engine (ICE) is not meant to be in competition with fuel cells. The current state of the hydrogen economy is quite limited. While the US government is investing in increasing hydrogen production, hydrogen internal combustion engines will likely be the first products that will create a demand for hydrogen. With minor modifications and using many of the parts and assembly lines, OEM’s could produce vehicles with a low cost for entry into the zero-carbon fuel market. As more fueling stations and infrastructure are brought online both fuel cell vehicles and hydrogen ICE’s will benefit. <?page no="336"?> A natural gas engine was converted to run on hydrogen for this project to demonstrate the minor modifications to change fuels. Hydrogen is generally being utilized in spark ignited engines. As with most spark ignited engines, a hydrogen ICE is no different that its performance is limited by abnormal combustion. During the initial calibration of the engine for the demonstration vehicle it was observed that H 2 pre-ignites easily, and boundary conditions can impact its ignition. A list of potential pre-ignition sources for hydrogen engines has been outlined by [1] including hot spots, the catalytic effect of spark plugs, hot residuals, oil droplets, hot particles, or a slow burning flame in the top land of the crevice. The authors of [1] tested an engine with a controllable glow plug as a hot spot in an optical hydrogen engine. Their results provide evidence that hot spots as the source of pre-ignition is unlikely because the surface temperatures for hot spot ignition was over 1100 K, higher than any surfaces would be in an engine. As more evidence against hot spots, pistons with additional cooling jets and aluminum pistons were shown not to affect the pre-ignition rate compared to the same piston design of steel [2]. A conversion of a light-duty engine for hydrogen combustion mentions a change of spark plug because of the catalytic properties of hydrogen on platinum but no data directly shows changes in behavior [3] [4]. A more recent study could not definitively conclude that pre-ignition rates were reduced due to changes in spark plugs but show changes over time [5]. The authors were able to show a difference due to piston deposits, oil temperature and blow-by routing though. There is also a gap in literature showing the impact of residuals on pre-ignition rates. Residuals can be affected by valve timing or turbocharging configurations that change the exhaust pressures. A few publications have used simulations and experiments to show changes in valve timing on hydrogen power and emissions [6] [7] [8]. However, they do not discuss changes to pre-ignition behavior. One study used CFD to study pre-ignition based on injection timing and valve timing [9]. The authors found retarding the intake valve opening mitigated rich pockets of fuel near hot spots near the exhaust valves and spark plug. They concluded that pre-ignition was caused by poor mixing leading to rich mixtures near hot spots. Based on the previous mentioned publications hot spots seemed like an unlikely ignition source in direct contrast to this paper’s finding. Engine Platform A Cummins X15N natural gas engine was converted to run on hydrogen for this demon‐ stration program. In trying to show there are few engine changes required for conversion the engine was approached as a minimum viable product. The combustion chamber was not changed from the natural gas engine. It retained the same compression ratio, flat cylinder head, and re-entrant piston. The piston clearances and ring tension were not adjusted. The cam timing was also not optimized for the hydrogen version. It retained the same valve timing as the natural gas engine. Modifications that were made included adding PFI injectors, a Mahle High pressure impactor (HPI) crankcase ventilation system, a SEM capacitive discharge ignition system, and a SuperTurbo boosting system. More information on the hardware changes can be found in a previous publication [10]. A picture of the engine in the test cell is shown in Figure 1 and the corresponding specifications are provided in Tabelle 1. 336 Thomas E. Briggs 336 <?page no="337"?> For meeting US based emissions targets the lambda range for the calibration was maintained between lambda 2.5 and lambda 2.1. The lambda values for the final calibration are shown in Figure 2. Although hydrogen has a wide flammability range and operating at higher lambda was possible it would not be ideal for the aftertreatment system. In Figure 3 the brake thermal efficiency (BTE) is shown to be a maximum near lambda 2.4 to 2.5. At leaner conditions hydrogen slip reduced the combustion efficiency and the BTE. Testing the SCR system on a reactor rig showed increased selectivity to N 2 O with increased inlet hydrogen concentrations. On most engines hydrogen slip is not problematic. N 2 O has a much greater GHG potential than CO 2 so hydrogen slip could become problematic by generating N 2 O. The turbine out temperatures also decreased with leaner mixtures. At low loads during a cycle this could potentially lead to a condition where the SCR temperature was below its efficient conversion range. As a full system approach, lambda greater than 2.5 was not used in the calibration to avoid either NO X or N 2 O breakthrough. This was found to be a key calibration parameter for meeting the 2027 emissions limits. Boosting requirements also limit the lambda values that can be used at high loads. Richer mixtures lessen the burden on the boosting system enabling high torque and power output. However, lambda is limited on the rich side due to an increase in engine out NO X . For the best argument for inclusion as a ZEV the NO X emissions were desired to be under 10 mg/ hp-hr. This has a significant margin from the 35 mg/ hp-hr limit. Potentially the lambda values could have been richer and a slight penalty in engine out NO X accepted while remaining under the 2027 targets. Regardless of emissions though, pre-ignition becomes the limiting factor. As will be discussed in the next section pre-ignition rates can be worse at the leaner or richer end of the lambda range. Parameter Value Displacement 15 L Configuration Inline 6 cylinder Bore 135 Stroke 169 Compression Ratio 12: 1 Injection location PFI Injection Pressure 10-15-bar Boosting system Supercharged and turbocharged Table 1: Specifications of the X15N engine as converted to operate on hydrogen Observations on pre-ignition in a port injected heavy duty hydrogen internal combustion engine 337 337 <?page no="338"?> Figure 1: Cummins X15N engine converted to operate on hydrogen in the test cell Figure 2: Lambda map for the low NO X hydrogen engine demonstration 338 Thomas E. Briggs 338 <?page no="339"?> Figure 3: Turbine out temperature, BTE and engine out NO X at 900-rpm 8-bar BMEP versus lambda Observations on pre-ignition in a port injected heavy duty hydrogen internal combustion engine 339 339 <?page no="340"?> Steady State Knock and Pre-ignition The engine was converted from the production single-point injection system to a PFI system for the hydrogen conversion. Backfire was observed early in the calibration work. After the installation of capacitive discharge coils to eliminate ghost sparks, backfires were very rare. The focus of this paper will be on the pre-ignition phenomenon, that occurs late in the cycle after the intake valves are closed. Although this is a PFI engine and most OEM’s want to use direct injection (DI), the findings from this work will still be applicable to DI engines. The pre-ignition events as shown by an example trace in Figure 4 were late cycle near top dead center but before the spark event. For this type of event, there is little difference between a PFI and DI engine, except that a PFI system should eliminate any mixture inhomogeneity that could increase the likelihood of pre-ignition in a DI engine. All the pre-ignition counts ignore backfires that would be specific to a PFI engine. Figure 4: Example cylinder pressure traces at 1400-rpm, 14,5-bar BMEP at lambda 2.11 For determining calibration set points of air/ fuel ratio, load sweeps were performed at varying lambda values. The pre-ignition (PI) rate as a count of events per 1000 cycles during these load sweeps at 1200-rpm is shown in Figure 5. Generally, it shows that pre-ignition increased with load and that richer mixtures were more susceptible to pre-ignition. This is highlighted by focusing on the pre-ignition rate at 15 bar BMEP only. The PI rate is plotted versus lambda at 15 bar BMEP in Figure 6. At this load the PI rate was parabolic with lambda. At richer air/ fuel ratios the number of PI events increased sharply. There seems to be a minimum for this condition at around lambda 2.3. However, there was a stochastic nature to the PI rate that is evident from Figure 5 when the PI rate is less than five events per 1000 cycles. 340 Thomas E. Briggs 340 <?page no="341"?> Figure 5: Pre-ignition rate as at varying load and lambda at 1200-rpm Figure 6: Pre-ignition rate versus lambda at 1200-rpm, 15-bar BMEP To understand the sensitivity to boundary conditions the load sweeps were repeated with different coolant and manifold air temperatures. The results from this testing are shown in Figure 7 and Figure 8. The testing in Figure 5 was all with an engine coolant temperature (ECT) controlled to 80 °C. The thermostats were blocked open for this testing to allow the controlled temperature to be representative of the coolant in the head and block temperatures. Increasing the ECT to 90 °C increased the PI rate at 16 bar BMEP. There appears to only be a minor sensitivity to ECT though. The results from increasing the manifold air temperature (MAT) to 40 °C from 30 °C, showed a much larger sensitivity. Increasing the MAT by 10 °C, the engine generally experienced the same number of pre-ignition events as were seen at a load 4 bar BMEP greater. Compared to Figure 5 again, a change of lambda from 2.3 to 1.9 yielded the same number of PI events at a 2 bar BMEP lower load. The manifold temperature thus seems to be the most sensitive parameter. Observations on pre-ignition in a port injected heavy duty hydrogen internal combustion engine 341 341 <?page no="342"?> Figure 7: Pre-ignition rate at 1200 rpm with different coolant temperature and manifold air temperature at varying load Figure 8: Pre-ignition rate at 1200-rpm at different coolant and manifold air temperature vs lambda The load sweep testing with varying lambdas was also repeated at higher engine speeds. The results from testing at 1600 rpm, shown in Figure 9, was performed with the coolant temperature at 90 °C and the manifold temperature at 30 °C. The pre-ignition rate showed a similar exponential increase in events as at 1200 rpm with increases in load. However, examining the data at 12 bar BMEP in Figure 10 shows a parabolic shape with higher pre-ignition rates at higher lambda. Higher lambda requires higher manifold pressures and pre-turbine pressures. It was hypothesized at 1600 rpm that the higher pre-turbine pressures resulted in higher hot residual concentrations that caused more pre-ignition events. Because of this, even if a different boosting configuration could provide the required boost to run a target lambda and load level it may be limited by pre-ignition if the backpressure was too high. In this case it may be better to enrich the mixture rather than to increase the trapped residual fraction. This of course would require a careful study of trade-offs as both options have negative impacts on pre-ignition. 342 Thomas E. Briggs 342 <?page no="343"?> Figure 9: Pre-ignition rate at 1600-rpm with varying lambda for a sweep of BMEP Figure 10: Pre-ignition rate at 1600-rpm, 12-bar BMEP versus lambda Transient Knock and Pre-Ignition This engine has been tested on a variety of regulatory transient cycles as well as real world drive cycles adapted to engine cycles. Some hydrogen engines have derated torque targets because of pre-ignition during tip-ins. Lean burn engines may be calibrated to enrich during tip-ins to build exhaust energy to increase compressor speed and thereby engine power. As observed in the steady state results, mixtures that are too rich are prone to pre-ignition. The pre-ignition tendency can therefore limit the acceleration rate or even the maximum torque during the transient. For this engine the SuperTurbo could supercharge, providing power input from the crankshaft to the compressor shaft. This enabled the compressor speed to be independent of exhaust energy. The system was calibrated to provide a time to torque under three seconds while not overshooting the lambda target. An example of a throttle snap from 10 to 90 % at 1000 rpm is shown in Figure 11. At the very beginning of the Observations on pre-ignition in a port injected heavy duty hydrogen internal combustion engine 343 343 <?page no="344"?> tip-in there was a short lean spike. This was a result of the fast naturally aspirated response of fully opening the throttle. As the supercharging took over the lambda recovered and showed a quick recovery to lambda 2.1 without further overshoot in the rich direction. This calibration effort enabled the engine to be run over transient cycles at the same torque and power ratings throughout the cycle. Figure 12 shows the peak cylinder pressure for each cylinder during a portion of the hot HD FTP cycle. The corresponding cycle is shown in Figure 13 with a box highlighting the section where the in-cylinder peak pressure data is shown. This was a strong tip-in event during the cycle that would be a likely area for pre-ignition to occur. There were no high peak cylinder pressure events with a large deviation from the running average that would indicate a pre-ignition though. Throughout the variety of test cycles, the use of supercharging and lambda control seemed to be a key technology to enable fast transient response for hydrogen engines. Figure 11: BMEP and lambda response from 10 to 90-% throttle snap at 1000-rpm Figure 12: Peak cylinder pressures during a segment of the HD FTP cycle showing no abnormally high peak pressure events indicative of pre-ignition 344 Thomas E. Briggs 344 <?page no="345"?> Figure 13: Engine speed and BMEP from the HD FTP trace run as a hot start with the highlighted region corresponding to the data in Discussion The causes of pre-ignition in hydrogen engines are an ongoing area of research. Due to time constraints, this demonstration used a turbocharger sized for a 13 L engine; this led to it being undersized for the 15L displacement of the demonstration engine. The supercharging aspect of the SuperTurbo was a benefit on transient cycles. A future iteration of the hardware could easily be resized for this engine while retaining the same transient capabilities. For the results presented here, the smaller turbocharger size appeared to have caused the higher pre-turbine pressures that led to higher pre-ignition activity at higher engine speeds. The differences in pre-ignition rates between 1200 rpm and 1600 rpm are analyzed in more detail in this section. Specifically, it was desired to understand why the number of pre-ignition events increased with higher lambda even when the reactivity of the mixture should be lower. At 1200 rpm the turbocharger sizing was matched well for the airflow requirements. There was very little supercharging (torque input from the crankshaft) on the compressor. The pre-turbine pressure was about 20 kPa higher than the intake manifold pressure. This is shown by the head delta pressure in Figure 14. The data corresponds to the test conditions in Figures 8 and 10 presented above. Operating at leaner mixtures required more boost which caused a larger difference between the exhaust and intake manifold pressures. At 1600 rpm the turbine extracted more power than was necessary for the compressor to provide the boost. Instead of changing the vane opening or the mass flow through the turbine like traditional turbochargers, the excess power was supplied to the crankshaft via turbocompounding. While this improved efficiency by making useful work out of the exhaust energy it increased the pre-turbine pressure as shown by the much higher delta Observations on pre-ignition in a port injected heavy duty hydrogen internal combustion engine 345 345 <?page no="346"?> pressure at 1600 rpm in Figure 14. It was expected that the higher pre-turbine pressure would cause higher residual gas fractions, a potential source of pre-ignition. A GT-Power model was built and verified to match the test engine for airflow and efficiency. The simulation results for the residuals based on the operating conditions in Figure 8 and Figure 10 are presented in Figure 15. As expected, the residuals were much higher at 1600 rpm than at 1200 rpm. However, the change in residuals due to increasing lambda from 2.14 to 2.48 increased by just over 0.5%. This is a small increase that seems unlikely to be the cause of the higher pre-ignition rate at lambda 2.4 compared to lambda 2.14. Not only were the residuals fractions similar, but the exhaust gas temperatures were lower for higher lambda. Figure 16 shows the same change in lambda reduced the exhaust temperatures in the port for cylinder 4 by 30 °C. Figure 14: Pressure difference between the exhaust manifold minus the intake manifold. (Negative values are a higher exhaust manifold pressure than intake) Figure 15: Residual gas fraction at constant load with different lambda values based on a GT-Power Simulation 346 Thomas E. Briggs 346 <?page no="347"?> Figure 16: Exhaust port temperatures for cylinder 4 from engine test data at constant BMEP while varying lambda Another piece of information from the GT-Power simulations that is useful is the in-cylinder temperature estimation. This was used for qualitative comparison to ignition delay times for hydrogen combustion. The analysis followed the methods from [11]. Figure 17 shows the 8-millisecond ignition delay for hydrogen at lambda 2.5 and lambda 2 calculated from the C3 Mechanism version 3.3. The results from the C3 mechanism show that above 20 bar, hydrogen is primarily sensitive to temperature. For the most part the line is vertical suggesting that the ignition of hydrogen is not sensitive to pressure. However, at pressures less than 20 bar, hydrogen ignition is more sensitive to pressure than to temperature. It has been proposed that DI injection only after the cylinder pressure is above 20 bar would help to mitigate pre-ignition by avoiding this area of sensitivity [1]. Most of the pre-ignition events on this engine occurred when the cylinder pressure was between 50 and 75 bar, significantly higher than this area of high-pressure sensitivity. It is possible reactions start during this portion of the cycle and progress to the point of a sustained reaction later in the cycle. Or it is possible the reaction mechanisms at engine relevant conditions above 100 bar are not well understood. Most of these reaction mechanisms are tested on rapid compression machines that typically do not operate at pressures much above 20 bar. Also of note, the auto-ignition delay curves of lambda 2.5 and lambda 2 are only separated by a few degrees K. There are many references in literature at this point showing a significant sensitivity to lambda for hydrogen engines but that is not reflected by this mechanism. The simulated temperatures from 1600 rpm and 12 bar BMEP with varying lambda are compared to the autoignition delay lines. The peak cylinder pressure was about the same for each of the various lambda. The maximum temperature increased as the mixture was enriched. However, examining the temperature profiles over a crank angle basis showed each of the test points had the same in-cylinder temperature at the same crank angle during compression. The only difference in temperature was due to the combustion phase of the cycle. The intake manifold temperature was controlled to the same temperature for each case, so the start of compression temperature was the same. Without significant changes Observations on pre-ignition in a port injected heavy duty hydrogen internal combustion engine 347 347 <?page no="348"?> to residual fraction or the gamma of the mixture there was no significant change to the temperature during compression. This does not explain any pre-ignition events then. It is possible there is a chemical effect of the residuals from NO that is not captured by this thermodynamic analysis. Figure 17: Pressure and temperature history from GT-Power Simulations of constant load operation at 1600-rpm while varying lambda compared to the 8 ms ignition delay curves for hydrogen at lambda 2.5 and lambda 2 Conclusions A natural gas engine was converted to run on hydrogen as a demonstration of concept for the US market. An SCR based aftertreatment coupled with the engine successfully showed tailpipe NO X less than 10 mg/ hp-h. To achieve this the lambda range was limited between lambda 2.1 and lambda 2.5. This ensured the engine out NO X was not too high and that there was not hydrogen slip. Hydrogen that is unburned in the exhaust skews the selectivity of the SCR towards N 2 O, an even higher GHG potential gas than CO 2 . Within that lambda range selected for emissions the maximum torque and power of the engine was limited by pre-ignition. Boosting for hydrogen engines is a challenge to achieve the same power levels as the diesel and natural gas engines they could potentially replace. For this application a SuperTurbo was chosen for its ability to provide fast transient response. The supercharging capability of this turbocharger can provide boost with quick response times without the need for transient enrichment. As a result, the engine could operate at the same torque under transient conditions as in steady state without any derates. The lambda control was such that pre-ignition was not observed during heavy tip-in events of regulatory 348 Thomas E. Briggs 348 <?page no="349"?> or real-world cycles. However, there are potentially more boosting configurations to be investigated. Boosting systems for hydrogen engines have been studied in simulations for their ability to provide the necessary airflow and boost requirements [12]. This is valuable information, but they will likely need to be verified with actual hardware testing. As the testing results from this paper have shown, pre-ignition is difficult to predict based on thermodynamic results. In some conditions, the engine was more likely to have pre-ignition events at richer mixtures as would be expected of a more reactive mixture. However, at other conditions leaner operation was more likely to cause pre-ignition events. It was thought that residual gas would be the driving factor of pre-ignition especially at the leaner cases as the pre-turbine pressures are significantly higher to drive the required airflow. However, a GT-Power model showed minimal changes in both residual gas fraction and cylinder temperatures at the same load between a range of lambda values. From this analysis, a knock model that is based on cylinder temperatures and ignition delay times looks unlikely to capture the pre-ignition effect. There could be a chemical effect of residuals that is currently unaccounted for, or the root cause could be from oil droplets as has been proposed for pre-ignition in light-duty gasoline engines. There is still a need to understand the root cause of pre-ignition in hydrogen engines so that future engines can reach the same power output as current natural gas engines. Future work planned for this engine platform is to test variations in backpressure and various boosting configurations hardware. Various oil formulations should also be tested for their impact on pre-ignition rates. This should include oils with different volatility ranges as well as different detergent levels. References [1] R. Rajasegar, A. Srna, I. Barbery and R. Novella, “On the Phenomenology of Hot-spot Induced Pre-ignition in a Direct-Injection Hydrogen-Fueled, Heavy-Duty, Optical-Engine,” SAE TEchnical Paper 2023-32-0169, Vols. doi: 10.4271/ 2023-32-0169, 2023. [2] R. Meske, K. Schmidt, H. Shiba, R. Capellmann, M. Retzlaff, P. Zimmer, A. Boberic, S. Pischinger and L. Virnich, “Component and Combustion Optimization of a Hydrogen Internal Combustion Engine to Reach High Specific Power for Heavy-Duty Applications,” SAE Tech-nical Paper 2023-32-0038, 2023. [3] P. Huyskens, S. Van Oost, P. Goemaere, K. Bertels and M. Pecqueur, “The technical implementation of a retrofit hydrogen PFI system on a passenger car,” SAE Technical Paper 2011-01-2004. [4] A. Srna, T. Lee, G. Nyrenstedt and R. Rajasegar, “Towards Conceptual Understanding of Pre-Igni‐ tion Mechanisms in Hydrogen-Fueled Engines - Recent Progress at Sandia National Laboratories,” in 45th Int. Vienna Motor Symposium, Vienna, 2024. [5] P. Grabner, M. Schneider and K. Gschiel, “Formation Mechanisms and Characterization of abnormal combustion phenomena of hydrogen engines,” SAE Technical paper 2023-32-0168, 2023. [6] D. Kovacs, R. Rezaei, F. Englert, C. Hayduk and T. Delebinski, “High Efficiency HD Hydrogen Combustion Engines: Improvement Potentials for Future Regulations,” SAE TEchnical Paper 2022-01-0477, Vols. doi: 10.4271/ 2022-01-0477, 2022. [7] M. Mohamed, K. Longo, H. Zhao, J. Hall and A. Harrington, “Hydrogen Engine Insights: A compreshensive experimental examination of port fuel injection and direct injection,” SAE Technical paper 2024-01-2611, Vols. doi: 10.4271/ 2024-01-2611, 2024. Observations on pre-ignition in a port injected heavy duty hydrogen internal combustion engine 349 349 <?page no="350"?> [8] S. Verhelst, J. De Landstsheere, F. De Smet, C. Billiouw, A. Trenson and R. Sierens, “Effects of Supercharging, EGR and Variable Valve Timing on Power and Emissions of Hydrogen Internal Combustion Engines,” SAE Int. J. Engines 2008-01-1033, vol. 1, no. 1, pp. 647-656, 2008. [9] L. Rouleau, F. Duffour, B. Walter, R. Kumar and L. Nowak, “Experimental and Numerical Investigation on Hydrogen Internal Combustion Engine,” SAE Technical Paper 2021-24-0060, Vols. doi: 10.4271/ 2021-24-0060, 2021. [10] T. Briggs and R. Williams, “Zero Impact Engines: A Demonstration of H2-ICE Technologies for Zero-CO 2 and Near Zero NOX in the North American Class 8 Heavy-Duty Truck Market,” in 45th International Vienna Motor Symposium, Vienna, 2024. [11] J. P. Szybist and D. A. Splitter, “Pressure and Temperature effects on fuels with varying octane sensitivity at high load in SI engines,” Combus-tion and Flame, vol. 177, pp. 49-66, 2017. [12] F. Pucillo, F. Millo, A. Piano, S. Giordana, N. Rapetto and F. Paulicelli, “Turbocharging System Selection for a Hydrogen-Fuelled Spark-Ignition Internal Combustion Engine for Heavy-Duty Applications,” SAE Tech-nical Paper 2024-01-3019, pp. doi: 10.4271/ 2024-01-3019, 2024. [13] M. Amann, D. Mehta and T. Alger, “Engine Operating Condition and Gasoline Fuel Composition Effects on Low-Speed Pre-Ignition in High-Performance Spark Ignitied Gasoline Engines,” SAE Int. J. Fuels Lubr. 2011-01-0342, vol. 4, no. 1, do1: 10.4271/ 2011-01-0342, pp. 274-285, 2011. 350 Thomas E. Briggs 350 <?page no="351"?> Higher Efficiency through model-based, predictive Knock Control Dr.-Ing. Michael Fischer/ Tenneco GmbH, Dr.-Ing. Michael Grill/ FKFS, Prof. Dr.-Ing. Andre Kulzer/ FKFS & IFS, University Stuttgart, Dr. Ing. Marco Guenther/ TME, RWTH Aachen University, Prof. Dr.Ing. Stefan Pischinger/ TME, RWTH Aachen University Abstract Knocking combustion in spark-ignited internal combustion engines is a stochastic phenomenon. Today’s knock control systems generally accept the stochastic character of knocking and react to individual knocking work cycles with a retarded ignition timing - at the knock limit this allows 4 to 10 % knocking work cycles. A new predictive approach developed in the FVV project “Fast knocking prediction” avoids this and can increase the efficiency of the engine and reduce the exhaust gas temperature simply through better, innovative knock control. In a conventional knock control system, due to the retardation of the ignition timing after a working cycle with knocking and the ignition angle being advanced only slowly in smaller steps, a large number of cycles are operated with unnecessarily late combustion centers. Even approaches that include the intensity of the knock event in the decision to retard remain reactive approaches hereby. The predictive instead of the traditional reactive knock control enables an optimized center of combustion for operation at the knock limit, increases efficiency and thus reduces fuel consumption, CO 2 emissions and exhaust gas temperatures. However, the prerequisite for a predictive knock controller is a model that can predict the knock probability of the current engine operating state precisely and very quickly. Therefore, the aim of the FVV project “Fast Knocking Prediction” was to expand knowledge about knocking, to develop a model for predicting knock probabilities and to use this for a knock controller that is able to determine the ignition timing to control proactively. A related simulation model and corresponding real knock control software were developed and tested on the real engine in comparison to conventional reactive knock control. It was shown that the expected improvement potential could be tapped for stationary operating points. On one hand, the efficiency could be increased and at the same time, the exhaust gas temperature and the cylinder pressure fluctuations could be noticeably reduced. <?page no="352"?> 1 Introduction One major focus within the development of todays and new generation spark-ignition combustion engines is the continuous reduction of CO 2 emissions. Many different measures are required to comply with the increasingly stringent restrictions. Downspeeding and downsizing by increasing boost pressures and compression ratios are common approaches. However, such efficiency-increasing measures are limited primarily by the occurrence of knocking combustion. Operation of the engine as close as possible to the knocking boundary is therefore required to ensure the best possible efficiency. Related to this requirement, conventional knock control has some drawbacks. Due to the large spark timing retardation after knock detection and only slow subsequent advance in smaller increments, if no knock is detected, many cycles are operated at unnecessarily late center of combustions. Predictive instead of conventional reactive knock control enables an optimal center of combustion for operation at the knock limit, increases the efficiency and consequently reduces the fuel consumption and the CO 2 emissions. Hence, the goal of this project is to increase the knowledge on knocking and develop a knock controller that is able to adjust the spark timing predictively. The high complexity combined with the stochastic nature of knocking combustion make controller development challenging. Investigations of local auto-ignitions, preceding knock, are difficult to capture at the test bench and cannot be investigated in two-zone combustion models due to averaging over the entire burnt and unburnt zone. For that reason, in order to identify knock-relevant conditions, 3D-CFD simulations are employed to identify temperature and mixture inhomogeneities. Regime boundaries within the detonation diagram are verified using CFD simulations and the applicability of the detonation diagram in 0Dsimulations regarding the evaluation of auto-ignitions is investigated. Lastly, the role of combustion cycle-to-cycle variations regarding knocking is evaluated. Based on the gained insights from these investigations, a predictive knock control system is developed. The initial concept is subsequently compared to conventional knock control using 0D simulations to identify the emission reduction and efficiency increase potential. Finally, thermodynamic measurements are employed to investigate the ECU application of the novel control approach. 2 3D-CFD Investigations Within this project the LES methodology, established during “FVV Engine Knock Model” [1], has been refined to improve the knock prediction and feed the fundamental results into the development of a novel knock controller (see Chapter 4). Parallel to this investigation, research with respect to the Resonance Theory [2] has been carried out to prove the validity of the simulation approach as well as the detonation to deflagration border for gasoline. 352 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 352 <?page no="353"?> 2.1 Improved knock prediction and generation of datasets for 1D modelling Picking up on the results of FVV Engine Knock Model [1], Large-Eddy Simulations for the engine speed of n = 4000 1/ min and a load of pmi = 16 bar have been carried out (see full project report). While the cycle-to-cycle variations as well as the knock tendency is in good agreement, there is a significant spark advance of ΔαST = 5° CA necessary to realize that result. To understand this, a grid refinement study has been carried out. Significantly faster burning speeds have been realized with finer numerical grid. This is in agreement with theories demanding a resolution of the flame thickness with five or more grid points if underprediction of laminar flame speeds is to be avoided for LES [3]. This kind of flame resolution is not feasible for 3D, turbulent applications of high temperature and pressure as observed in combustion engines as due to extreme demands on computing power. Therefore, additional models to describe the turbulent flame speeds need to be implemented. One approach is the Thickened Flame Model (TFM) [4] [5], which artificially thickens the flame to allow for a sufficient resolution. The laminar flame speeds are tabulated beforehand and an efficiency factor E is needed to account for differences in diffusivity and turbulence/ chemistry interaction. The potential of such a model is demonstrated for a laminar 2D case in Figure 1. After calibration of a flame indicator (heat release was used), the results will be briefly discussed and compared to simulations with the currently utilized grid resolution “SAGE Coarse” with a base grid size of Δbase = 1.0 mm and a refinement in the vicinity of the flame of Δflame = 0.25 mm. As a reference point, a strong grid refinement to Δflame = 0.015625 mm has been carried out in order to achieve about 8 grid points across the flame, we will call this “SAGE Fine” (further refinement did not influence the outcome of the simulation). The results with respect to the flame speed are evaluated via the integrated heat release, see Figure 1 (left). To also evaluate the numerical effort of the approach, the total cell count is depicted in Figure 1 (right). Figure 1: Integrated heat release compared for two grid refinement levels with SAGE and the result for the Thickened Flame Model (depicted left). On the right, the influence of the total cell count of the simulation is shown, already demonstrating a factor of four for a 2D-case. Higher Efficiency through model-based, predictive Knock Control 353 353 <?page no="354"?> While SAGE TFM requires roughly the same cell count as SAGE Coarse, there is a distinct difference in the Integrated Heat Release with the SAGE Coarse burning significantly slower. SAGE TFM heat release is in alignment with the reference point, SAGE Fine, while requiring roughly a factor of four less cells (depending on the point in time for evaluation). As this is a 2D case and a 3D case will burn (idealized) spherically, the factor of four for 2D may become a factor of 16 for a 3D application. This does not include the influence of the time step on the overall runtime (due to the explicit solver scheme). Runtimes for the conducted simulations are shown in Table 1. Despite the slightly lower cell count, the TFM has about 10 % higher runtimes, indicating additional numerical effort. Between SAGE Coarse and SAGE Fine there is a factor of about five in the runtime, indicating a nearly neglectable influence of the time-step. In fact, the time step was about 4 times smaller for the SAGE Fine. However, it needs to be considered that all cases were run on the same setup with 44 cores. Evaluated at t = 0.02 s that makes around 900 cells/ core for SAGE Coarse and around 3600 cells/ core for SAGE Fine. With less cells/ core, most CFD solvers tend to get less effective as information needs to be exchanged between cores [6]. A realistic time-benefit would consequently be in the order of 14-16 for SAGE TFM over SAGE Fine while maintaining the same solution quality. For the research project this means, more individual cycles can be simulated, allowing for statistically valid evaluation of knocking combustion and its influence factors. Table 1: Impact on runtime for the conducted 2D laminar combustion cases. Simulations were carried out on 44 CPU, runtimes are affected by better efficiency of the solver with more cells/ core. After 2D calibration of the flame indicator, the LES TFM model has been successfully applied to an operating point with a high engine speed of n = 4000 1/ min, documented in the final report of the project. This was motivated by the results of [1], where the strongest deficiencies with respect to the turbulent flame speed were observed for that operating point, however this phenomenon has been observed throughout an engine speed variation. With the potential of the methodology being proven, the application to a more representative operating point (so a medium engine speed) is important in the scope of this project. Therefore, the operating point of n = 2500 1/ min, pmi = 16 bar with central direct injection of RON95E10 has been chosen. This represents a mean operating point within the experimental investigations of [Blo21], which also was the experimental foundation of this research project. To provide a sound statistical foundation, the number of investigated working cycles is increased to n = 100. 354 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 354 <?page no="355"?> Figure 2: Calibrated model and response to a spark advance in order to reach close to 50-% of the cycles above the knock limit. After carrying out 100 injection simulations (the overall approach is non-consecutive), 25 cycles are used to calibrate the Thickened Flame Model (Figure 2, left). With this calibration, a spark advance of ΔαST = 2° CA has been applied to the full dataset to generate a knock frequency of about 50-% to allow for a meaningful statistical evaluation of knock influence factors. The result of this computation is depicted in Figure 2 (right). With the scope of the overall research project in mind, it was of interest to find early indicators which can be utilized for knock prediction. Therefore, numerous quantities have been correlated both with knock intensities and combustion phasing. These have been correlated from linear to cubic fashion, including multiple correlations. The main influence factor has been identified as both local and global charge motion as well as turbulent quantities. A detailed overview is given in the final project report, in the following the charge motion and the local quantities at the spark plug will be discussed. Both the first and the second tumble peak do not reveal any significant correlation with the knock intensity KPP (Knocking Peak to Peak) with values of R2linear = 0.03. Evaluating the charge motion at spark timing (close to TDC), depicted on the right of Figure 3, a trend for higher knock intensities and strong tumble motion is observed. Within FVV “Engine Knock Model” [1] it has already been identified that a flame deflection toward the intake side reveals stronger knock intensities, often stemming from the exhaust side. As the tumble Higher Efficiency through model-based, predictive Knock Control 355 355 <?page no="356"?> evaluation is an integral method and the flame deflection is predominantly a result of early flame kernel development, a local evaluation of the flow properties at the spark plug (also at spark timing) is of great interest. This is shown for the U-Velocity (pointing toward the intake side) both for MFB50 and for KPP in Figure 4. Figure 3: Blindtext für eine Bildunterschrift Evaluation of Knock Peak to Peak versus global charge motion values. First and second tumble peak showing no correlation while the value close to spark timing reveals a minor trend for increased knock intensity with increased charge motion. The primary correlation can be found between center of combustion (MFB50) and the U-Velocity at the spark gap with a R2linear = 0.39, with faster combustion for stronger deflection towards the intake side. The correlation between knock intensity is reduced to about a half, R2linear = 0.19. Considering the correlation between MFB50 and KPP of R2linear = 0.64 this confirms the primary correlation is between center of combustion and the velocity component at the spark plug. Nevertheless, 35 out of the 47 (or 75 %) cycles about the knock limit have a positive U-velocity component at the spark gap, these cycles have been colored dark blue in Figure 4. Figure 4: A higher positive U-Velocity component (pointing toward the intake) at the spark gap at spark timing indicates a trend for earlier center of combustion as well as a weak trend for increased knock likelihood. 356 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 356 <?page no="357"?> 2.2 Resonance Theory: Validation of Deflagration to Detonation Diagram Boundaries As the resonance theory of [2] has also been investigated as a candidate to support novel knock controller (see section 3.2), the borders of the diagram needed to be confirmed for the applied fuel. This was carried out by the means of 3D-CFD simulation. As detailed LES combustion simulations, including several knocking cycles, exist, the data was evaluated and used to extract boundary conditions for a first evaluation using the DDT diagram. Due to the high data output frequency, the thermodynamic properties at the auto-ignition location can be analyzed, see Figure 5. The hotspot leading to auto-ignition is clearly visible at 748° CA, leading to a rapid consumption of the fuel in the vicinity and emitting a pressure wave. Figure 5: Detailed investigation close to knock onset in a vertical section cut. Hotspot is visible in the temperature distribution at 748° CA (on the upper left). The depicted auto-ignition has led to a knock intensity of KPP = 23-bar. From this simulation run, boundary conditions have been extracted, see Table 2. Table 2: Overview of extracted hotspot conditions, load point n = 2500-1/ min, pmi = 16-bar. A simplified simulation domain has been set up for evaluation of the ignition behavior of different hotspot conditions. The dimensions of the domain were selected from the reference engine with d Cyl = 75-mm and a height of h = 20-mm, leading to a volume of V Cyl = 8.8e-5-m 3 , a value corresponding to about 750° CA of the investigated engine (close to a typical knock onset). With residual gas and fuel assumed to be perfectly mixed (refer to the stratification evaluation of this operating point in [Blo21]), the temperature is the only influence factor on the thermodynamic properties needed for the evaluation with the resonance theory: ignition delay time τ i and excitation time τ e . Both have been tabulated for a wide range of operating conditions to allow for a semi-automated evaluation. No turbulence was applied and therefore Higher Efficiency through model-based, predictive Knock Control 357 357 <?page no="358"?> also no turbulence model was selected. The simulation domain is set up with a relative air/ fuel ratio of λ = 1 and no residual gas (EGR). To evaluate the temperature increase in the domain with a high temporal resolution, monitor points have been radially distributed throughout the system. The evaluation should be carried out in a constant distance from the hot spot border, at which the pressure waves will eventually be emitted. Starting with the conditions derived from Figure 5, a hotspot-size variation has been carried out to determine the boundaries of the DDT diagram. Figure 6 illustrates both the pressure (black line) and the temperature (red line) at a monitor point in 15 mm distance from the center for a hotspot size of r HS = 4 mm. Additionally, a reduced hotspot size of r HS = 2.5-mm is shown in grey. Figure 6: Evaluation of the temperature and pressure trace for a detonating case, as seen by the nearly instantaneous temperature increase upon arrival of the pressure wave. In grey, the same experiment is shown for a reduced hot spot size. Without any hotspot initialized, this experiment would just reflect the ignition delay time of the operating point. With the hotspot applied, the ignition delay time locally decreases and combustion is initialized from the center. For this set of boundary conditions, the ignition delay time of the hotspot is at around τ i,HS = 9.0e-2 ms while τ i,Cyl = 1.1e-1 ms. The pressure evolution depicted in Figure-6 (r HS = 2.5-mm, grey) shows a continuous, yet small increase until the hotspot auto-ignites and a pressure wave with an amplitude of around Δp = 15 bar arrives. After arrival of the pressure wave, the value falls back to the initial value. The pressure wave also increases the local temperature, however it also returns to the initial value after the pressure decreases. Eventually, the ignition delay time of the remaining domain is reached, accelerated by the temperature and pressure increase by the deflagrative flame propagation. Despite the short-term temperature and pressure increase, no coupling between the hotspot ignition and ignition in the remaining cylinder has been observed. For r HS = 4.0 mm, the arriving pressure wave (black line) has a much higher amplitude with Δp = 30 bar. After the pressure wave has travelled through the monitor point, the pressure is decreasing again, however with a strong interference. This can be understood 358 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 358 <?page no="359"?> (2.1) (2.2) when reviewing the temperature response at the hotspot which is also higher than for r HS = 2.5 mm, the main difference however is that the temperature is not falling back to the initial value. After a short decline, auto ignition conditions are reached at the monitor point. Therefore, auto-ignition shortly follows the pressure wave, coupling both and indicating a detonative combustion. The hotspot variation of the investigated operating point has been added to the DDT diagram, see Figure 7. In addition to the conditions of Table 2, a broader range of operating conditions has been investigated, to capture a larger section of the DDT diagram. The diagram is defined by the resonance parameter ξ and the reactivity parameter ε as per equations (2.1) and (2.2) with the speed of sound a and the excitation time τe, which is defined as the time from 5 % to maximum heat release rate. ξ = a δτ i δx ε = r H S α τ e Simplified, the resonance parameter defines if a coupling between the pressure wave and the auto-ignition front can take place. The reactivity parameter indicates, how rapid the ignition takes place and how much energy is released, so if the pressure wave amplitude can be raised. Theoretical background to the detonation theory (originally developed for syngas combustion) as well as the detailed, investigated conditions are documented in the final report of the project, the results are shown in Figure 7. Figure 7: Overview of the conducted simulations sorted into the DDT diagram. Overall, the boundaries of the DDT diagram can be confirmed. Higher Efficiency through model-based, predictive Knock Control 359 359 <?page no="360"?> Overall, it can be concluded, the originally defined boundaries [2] could be confirmed for gasoline, utilizing a 3D-CFD approach. Within the project, the application of the resonance theory to a novel knock controller has been investigated, this is documented in section 3.2. At this point it should be mentioned again that the resonance theory and the differentiation between denotation and deflagration is challenging when it comes to the evaluation if a knock event is harmful or not. While a single detonation event (mega or super knock [7]) might lead to engine damage, the differentiation of, for example, KPP = 2 bar and KPP = 15 bar is not possible for the given borders of the diagram, both would be categorized as deflagration. Nevertheless, the conducted investigation functioned as a validation for both the simulation methods as well as the theory itself, which might support further research on the topics of mega and super knock events. 3 Thermodynamic Investigations on Knock Occurrence and Knock Probability 3.1 Influence of Temperature and Mixture Inhomogeneities 3D CFD simulations are a convenient tool to investigate local phenomena that are difficult to capture on an engine test bench or that are not contained in typical 0D/ 1D models. Large-Eddy Simulations (LES) provide a good compromise between accuracy and the required computational effort for research purposes, in comparison to RANS and DNS. For this reason, results from Large-Eddy Simulations are analysed. General information about the investigated engine and the operating points is summarized in Table 3. Table 3: Basic engine data and investigated operating conditions. The two operating points at part load and a medium engine speed cover two different spark timings with different knock intensities and knock frequencies, while all other operating conditions are constant. At each operating point, data of 20 combustion cycles are available. Simulation of each cycle is based on an individual preceding simulation of injection. Thus, the resulting simulated engine cycles are non-consecutively. As a first step before the inhomogeneity can be analyzed, the unburnt zone has to be separated from the burnt zone, because in 3D CFD simulations the entire combustion chamber is discretized, which includes not only burnt and unburnt zones but also areas close to or in the flame front. This is contrary to the 0D simulations from the entrainment model, where the flame has no thickness and burnt and unburnt zone are already separated as they are treated as individual zones. 360 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 360 <?page no="361"?> If the flame propagation in 3D CFD simulations was to be investigated a simple separation by temperature is sufficient since temperatures of the burnt and unburnt mixture are significantly different and the steep temperature increase is confined to a thin flame front compared to the extent of the burnt and unburnt area. For evaluation of inhomogeneities however, this simple separation is insufficient. If a separation temperature is set too high, the temperature distribution in the unburnt zone will be dominated by cells with high temperatures near or in the flame front. The high temperatures are caused by heat conduction from the flame front into the unburnt mixture, but not the inhomogeneities that shall be evaluated. These cells are expected to be not knock-relevant, since the propagating flame would consume any occurring auto-ignition shortly after, due to the proximity to the flame front. A high separation temperature yields another problem for evaluations over time. As the flame propagates through the combustion chamber, already burnt areas start to cool down with increasing distance from the flame. Especially towards the end of the combustion, the temperature can drop below the separation limit, leading to a categorization of burnt cells as unburnt. If a low separation temperature is set, the unburnt cells will have a sufficient distance from the flame front and the elevated temperatures close to the flame are excluded, but due to the much smaller remaining temperature variation, there is a risk of classifying knock-relevant cells or areas with higher temperature as the burnt zone. It is therefore desired to have a separation criterion without direct influence on the temperature. For that reason, a separation based on the concentration of chemical species and a discrete flame front distance is developed. The instantaneous rise of the hydroxyl radical (OH) concentration during ignition is a well-suited indicator for the flame front and is used for initial separation of the burnt and unburnt zone. However, the hydroxyl radicals are only a temporary product of the combustion and the concentration decreases again behind the flame front. Therefore, as second species the concentration of the fuel (in this case iso-octane) is included. Burnt cells behind the flame front, where the OH concentration has decreased have only a very small or no amount of the initial iso-octane left. Thus, the additional criterion prevents burnt cells from being shifted back into the unburnt zone. Finally, after separation using the OH and the fuel concentration, cells with a minimum distance below 2 mm to the determined flame front are excluded from the unburnt zone to remove cells too close to the flame front that are not knock-relevant. Further details about the new separation approach are given in [8]. With the previously established separation criterion, inhomogeneities in the unburnt zone, as well as their cycle-to-cycle variation shall be quantified to identify knock-rele‐ vant conditions. Since both, average and maximum temperatures are increasing during compression and combustion inhomogeneities are not directly apparent from evaluation of absolute temperatures. Therefore, rather than absolute temperatures, the difference of cell temperatures to the average temperature of the unburnt zone at each time step is evaluated: Higher Efficiency through model-based, predictive Knock Control 361 361 <?page no="362"?> (3.1) ΔT = T ub, cell − T ub, mean Thus, similar positive ΔT values represent similar temperature elevations over the mean temperature. For the visualization and evaluation of the inhomogeneity, quantification is required. Evaluating the single hottest cell and its temperature elevation over the average temperature is not suitable, since auto-ignition does not originate from such a small volume and the included chemical energy. For that reason, inhomogeneity is defined as the relative temperature ΔT threshold where all cells with higher ΔT values combined have 5-% of the unburnt mass at the respective time step. The progress of this inhomogeneity indicator over time is shown in Figure 8 for one exemplary engine cycle. Figure 8: Temperature distribution and progress of 5 % mass limit with highest temperatures over time. During the first phase from the beginning of the evaluation until 725° CA, the inhomoge‐ neity increases almost linearly. This indicates uneven heating of the unburnt mixture. Since this linear increase already occurred during the compression, it is likely caused by varying wall temperatures, such as different valve temperatures or temperature differences between cylinder wall, piston, piston ring and spark plug. It further continues into the first phase of the combustion, in this example until 15° CA after ignition and an MFB of 3 %. Although the burnt mass fraction at the end of this first phase is relatively low, a significant amount of volume has already been burnt. After 725° CA, the temperature distribution transitions into a phase where the unburnt zone contains an inhomogeneity with unstable progress over time until the combustion is finished. A closer investigation reveals that shortly after 725° CA a significant heat release from the first ignition stage of the mixture occurs. It is concluded, that the temperature increase following the first ignition stage causes this significant disturbance of the initial temperature stratification and therefore the evaluated temperature inhomogeneity. In order to avoid the disturbances and evaluate the inhomogeneities before the beginning ignition of the mixture, the comparison of all engine cycles of each operating point is performed at top dead center (TDC), the last time step where none of the cycles has yet entered the first ignition stage. 362 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 362 <?page no="363"?> The results of all engine cycles at TDC are shown in Figure 9. Both ignition timings contain a general inhomogeneity between 13 K and 18 K. The cyclic variation is small with a variation range below 5 K, which is approximately a third of the general inhomogeneity. Also not specifically shown in Figure 9, the variation of 5 K is almost constant for the duration of 700° CA up to 720° CA. From the small cyclic variation of 5 K, a variation of the auto-ignition onset below 1° CA is expected. Therefore, the cycle-tocycle variations of the temperature inhomogeneity before occurrence of the first ignition stage in the unburnt mixture are expected to be negligible for the calculation of the auto-ignition onset. Figure 9: Temperature inhomogeneity with 5-% mass limit in relation to knock occurrence Besides the small cyclic variation, no obvious relation between the inhomogeneity of an engine cycle and the occurrence of knock can be observed. This observation was similar for varying mass fractions for the inhomogeneity definition, investigation of the mixture inhomogeneity and investigation of ΔT and the charge velocity for a spherical volume around the spark plug gap in relation to the unburnt mixture. Details about the additional investigations can be found in [8]. 3.2 Application of the Detonation Diagram to Evaluate Auto-Ignition The detonation diagram by Bradley et al. [2] and the contained limits that were confirmed by other researchers [9] [10] [7] was developed to identify developing detonations and categorize auto-ignitions in different regimes, starting from harmless deflagration over developing detonations with significant pressure waves up to thermal explosions. For the classification into the different regimes, two dimensionless parameters ξ and ε are used that contain local conditions and their stratification in a limited volume with elevated temperature, so-called hotspots, and boundary conditions of the surrounding unburnt mixture. Based on these two parameters, it can be evaluated if the propagating auto-ignition front can couple with the pressure wave leading to resonance with large pressure amplitudes in a developing detonation. This project aims to apply the detonation diagram to 0D simulations to identify knock-relevant boundary conditions. Netzer applied the detonation diagram to a quasi-dimensional (QD) simulation in [11] and found an apparent correlation between the variance of ε and the knock tendency. Following this observation, ε shall be determined at the time of auto-ignition for single working cycles within a 0D environment, with the aim to derive knock-relevant information. For Higher Efficiency through model-based, predictive Knock Control 363 363 <?page no="364"?> (3.2) the investigation, measurement data of single working cycles of the same single-cylinder engine as introduced in Table 3 is used. At each investigated operating point, measurement data and pressure trace analysis data (TPA) of 500 cycles is available. Covered operating conditions are summarized in Table 4. Table 4: Covered operating conditions for calculation of the reactivity parameter ε. The reactivity parameter ε is calculated as follows: ε = r 0 ατ e The acoustic velocity is determined for the unburnt zone based on data available from the TPAs at time of the auto-ignition. The excitation time is retrieved from a lookup table based on input parameters that are also directly available from the TPA results. Solely the hotspot size r 0 is not directly available from a two-zone calculation, which only contains mean values of the burnt an unburnt zone. Since it was shown in [8] that a constant radius of 10 mm, as proposed by Robert [12] is no suitable value, an operating point specific size distribution is modelled. The distributions are realized by a halved normal distribution with variable standard deviation to control the maximum occurring radii. Scaling of the standard deviation is implemented based on the margin of each cycle to the knock boundary, determined by a knock model [13]. The results, presented in Figure 10 (left) for a selection of operating points, show realistic levels of ε and a similar correlation as observed by Netzer in [Net17]. The variance of ε increases with advancing spark and thus the knock tendency. The correlation is further investigated by applying the determined ε distributions to a calculation approach from the literature [14] that enables calculation of the knock frequency. Comparison of the calculated knock frequency to the measured knock frequency (Figure 10 right side) revealed that the trend of the increasing knock frequency is correctly replicated by the calculated knock frequency. However, the mean deviation by over 6 % is significantly higher, compared to the 1.81 % deviation observed in [14] for the knock frequency calculation approach in its original form based on the distribution of auto-ignition onsets. Evaluation of a hotspot size distribution for 100 engine cycles from 3D CFD simulations revealed that a halved normal distribution is a good first approximation of a real distribu‐ tion, but could be further improved by a skewed normal or stretched beta distribution. 364 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 364 <?page no="365"?> The observed deviation of the calculated knock frequency is mainly attributed to the lack of validation possibilities for the hotspot size modeling approach and expected limited suitability of the detonation diagram in general to predict knock tendencies for operating points confined closely to the knock boundary. Figure 10: Mean and variance of ε (left) and knock frequency calculated from ε distribution vs. measured knock frequency (right). 3.3 Influence of Cycle-to-Cycle Variations on the Knock Probability Subsequent to the consideration of the cyclic variation of ε, the cycle-to-cycle variations are considered by application of the knock frequency prediction approach in its original form to 0D simulations. This so-called Three-Parameter-Approach enables determination of the knock frequency based on the auto-ignition onset distribution of single working cycles. The aim is to enable a prediction of the actual value of the knock frequency rather than the binary classification if an operating point is above or below the knock boundary, as provided by the current knock models. In total, three parameters AI Limit , AI mean and σ AI are required for the calculation of the knock frequency, which are derived from the auto-ignition onset distribution of an operating point (OP). The general calculation method is presented in Figure 11 at the example of a measurement set that contains five different operating points with various spark timings at otherwise similar operating conditions. At each operating point from TPAs and application of an auto-ignition model [15] the auto-ignition onset distribution is available. The first step towards the knock frequency calculation is to determine an auto-ignition limit (AI Limit ) for the operating point with the highest knock frequency. It is set in such way, that the relative amount of cycles with auto-ignition onset earlier than the limit match the measured knock frequency at this OP (12.8 % in the shown example). The AILimit is calibrated only at the operating point with the highest knock frequency and remains constant for the knock frequency calculations of all other spark timings. For the second step, the AI Limit is applied to all operating points. In Figure 11, this is illustrated for OP 4. Higher Efficiency through model-based, predictive Knock Control 365 365 <?page no="366"?> Figure 11: Three-Parameter-Approach for calculation of the knock frequency, according to [1] [14]. However, instead of evaluating the number of single working cycles with auto-ignition onset before AI Limit , the mean auto-ignition onset (AI mean ) and the standard deviation (σ AI ) of each auto-ignition onset distribution are determined. As the third step, the the difference between AI Limit and AI mean is calculated as the multiple of the standard deviation σ AI of the respective operating point. For the fourth and final step, a standard normal distribution is assumed for the auto-ignition onset distribution of all operating points. This allows calculation of the knock frequency as the probability of values of this Gaussian distribution deviating by more than the multiple of σ AI from AI mean . It is apparent, that the simulation of single working cycles is crucial in order to apply the approach to the simulation. This is a major difference to the application of current knock models, since the determination of the KLSA using knock models like [16] [17] [13] [18] is based on evaluation of the average working cycle. For this reason, two different approaches to simulate single working cycles are introduced. For the simulative investigation in total 15 operating points, covering two different engine speeds and two different indicated mean effective pressures, are investigated. The engine speed and load combinations are summarized in Table 5. Each combination includes five operating points with different spark timings, while the remaining operating conditions are similar. For the simulations of the single working cycles, the entrainment model [19] [20] is used. 366 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 366 <?page no="367"?> Table 5: Operating conditions covered by the simulations of single working cycles. To create the cyclic fluctuations, within the first simulation approach, a turbulence level distribution is modeled. This alters the flame propagation velocities and subsequently results in a variation of the engine load, the center of combustion, the pressure gradient, the maximum pressure and the combustion efficiency. A distribution of the turbulence level is realized by variation of the scaling factor implemented in the k-ε turbulence model. This factor allows adjustment of the start value of the turbulent kinetic energy. Within the second simulation approach, an available cycle-to-cycle variation (CCV) model [21] [22] is utilized. This phenomenological model alters available parameters in the combustion model to replicate the cyclic variations. Both approaches are solely calibrated to the engine at the OP with the highest knock frequency, where also AI Limit is calibrated, and are suitable for accurate replication of the CCVs. The simulation results of both approaches are used to determine the distribution-specific parameters AI mean and σ AI . Due to the significant reduction of simulated cycles from 500 to 15, using the CCV-model, AI Limit is only for the first simulation approach calibrated as described above. For the reduced number of simulated cycles, AI Limit is calibrated iteratively to the target knock frequency. Since all calibration is carried out solely at the OP with the highest knock frequency, the calculation is an actual prediction of the knock frequency for the other included OPs. Figure 12: Evaluation of the predicted knock frequency for the simulation approach using a turbulence level distribution (left) and using a CCV-model (right), according to [23]. In order to determine the accuracy of the predicted knock frequencies, not only the knock frequency but also the center of combustion for which it is predicted is relevant. Therefore, in Figure 12 the prediction accuracy is determined by comparing the center of combustion from simulation and measurement at similar knock frequencies. For both simulation approaches, all 15 operating points covering the three different engine speeds and engine Higher Efficiency through model-based, predictive Knock Control 367 367 <?page no="368"?> loads as well as the contained spark timing variations are included in the evaluation. Their small mean deviations of 0.36° CA and 0.58° CA for the first and second simulation method respectively, confirm a high accuracy of the predicted knock frequencies. The largest deviation of 1.6° CA was found for the second simulation method and could be attributed to inaccuracies of the CCV-model. 4 Knock Frequency Based Knock Control 4.1 Control Concept and Simulation Model Setup With the preceding work, that demonstrated an accurate prediction of the knock frequency, a control concept based on the knock frequency is introduced. Knock control based on the knock frequency/ probability would avoid the deterministic behaviour of conventional knock control, where the spark timing is retarded by a large increment each time after knock is detected and where the spark timing is gradually advanced in small increments if no knock is detected. Knowledge about and direct control of the knock frequency allows instantaneous and predictive adjustment of the spark timing, independent from the occurrence of knock, thus allowing the engine to be operated closer to knock boundary. Figure 13: Structure of the probability-based knock controller. The schematic structure of the new knock frequency based knock control concept is presented in Figure 13. The knock frequency Pcalc is calculated with the Three-Param‐ eter-Approach based on the parameters AI Limit , AI mean and σ AI . Within the control algorithm, the predicted knock frequency is directly compared to the target knock frequency P target 368 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 368 <?page no="369"?> in order to determine the appropriate spark timing adjustment. Spark timing adjustment is only initialized if the difference between calculated and target knock frequency exceeds the tolerance value P tolerance . This allows accounting for small modelling inaccuracies and prevents unnecessary adjustments close to the target knock rate. The final spark timing adjustment is determined by multiplication of the knock frequency deviation to the target frequency with the respective constant gain for spark advance K adv or retardation K ret . Including the difference of the input values in the adjustment allows precise control close to the desired limit and improves the fast transient response capabilities. For the simulation of conventional knock control, the identification of knocking cycles is crucial, since the controller is based on knock detection. However, in 0D simulations, no pressure oscillations are modeled that would allow similar knock detection as on the test bench by pressure transducers or structure-borne sensors. For that reason, a simple knock event simulation method is implemented in the simulation model. The method was originally introduced by Peyton et al. [24] and is based on the assumption of binomially distributed knock events, regardless of the underlying probability density function or the knock intensity [25]. Validity of the assumption for the investigated data was shown by evaluation of measurement data and comparison with results from the knock event simulation method and a theoretical calculation for a binomial distribution. Further details of the validation and the controller setup can be found in [8]. For the new predictive knock control approach the three parameters AI Limit , AI mean and σ AI have to be available at each engine cycle to allow the knock frequency prediction based on the Three-Parameter-Approach. σ AI and AI Limit are implemented as lookup tables. This is convenient, since AILimit represents an engine-specific calibration parameter for the knock frequency prediction. σ AI could in theory be predicted phenomenologically, similarly as described in Chapter 3.3 but would require the calculation of multiple cycles for simulation of one knock-controlled cycle. Therefore, due to performance reasons and since the parameter is directly available from measurements of the knock frequency which are required for calibration of the controller, it is convenient to store this parameter in a lookup table. Thus, both σ AI and AI Limit are determined based on engine speed, IMEP and the current spark timing. The third parameter AImean is determined as the auto-ignition onset of the mean simulated engine cycle. With IMEP being estimated for the subsequent engine cycle based on the intake pressure, the controller predicts the knock frequency of the upcoming engine cycle for unchanged spark timing and AI mean considering varying loads. The lookup tables as well as the knock frequency calculation are implemented into the GT-Suite simulation model via Python code. Both controllers are tuned to operate at a target knock frequency of 5 %. For the conventional controller this is achieved by adjusting the gains respectively. Within the knock frequency based controller, this is achieved by setting the target knock frequency to 5 %. The gains K adv and K ret of the knock frequency based controller account for the exponential behavior of the knock frequency increase over the spark timing, as a knock frequency decrease requires a smaller spark timing adjustment compared to a similar knock frequency increase. Due to this characteristics, two different gains are required to enable fast and accurate spark timing adjustment towards earlier and later timings. Finally, for Higher Efficiency through model-based, predictive Knock Control 369 369 <?page no="370"?> the simulation setup, both controllers are deactivated for the first five cycles to allow the simulation model to converge first. 4.2 Simulation Results In Figure 14 the results of both controllers under stationary operation are shown exempla‐ rily for one OP. In total, the engine speed and load combinations summarized in Table 5 are investigated. The spark timing is initialized in such way that the resulting knock frequency is very low and the initial control response is the advance of the spark timing. As can be seen, the predictive controller instantly adjusts the spark timing to yield a knock frequency close to the desired target value, whereas the conventional controller requires some cycles with the incremental adjustment, to reach within close range of the target value. Figure 14: Comparison of conventional and probability based knock control and the underlying pre‐ dicted knock frequency. The initial response and behavior over time is investigated in further detail. Figure 15 (left) shows the time difference between both controllers to reach within 0.5° CA to the center of combustion at 5 % knock frequency and the distribution of the contained centers of combustion over the entire duration for the same OP as presented in Figure 14. As the results for the other operating conditions reveal similar behavior they are not specifically shown here but can be found in [8]. As apparent, the probability based controller reaches faster within 0.5° CA to the target center of combustion compared to the conventional controller. In total, the new controller is between 6 to 22 cycles or 0.1-0.8 s faster. Regarding the behavior over time, the probability based control approach shows a significantly reduced variation of the contained center of combustion values and a mean center of combustion closer to the target value. Considering all operating points, the mean value for 370 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 370 <?page no="371"?> conventional control deviates by 0.6-0.9° CA from the target value whereas the deviation for the probability based controller is 0.02-0.3° CA. Stationary simulations were repeated for initialization at a high knock frequency before controller activation to evaluate the response for a spark timing retardation. The results comparing both controllers for initial response and response over time are similar to the presented ones and can be found in detail in [8]. Figure 15: Initial control response comparison (left) and comparison of MFB50 distribution over time (right). Transient simulations for varying loads show close agreement between the conventional and probability based controller, with a tendency of later center of combustions by the probability based controller for load decreases and earlier center of combustions for load increases (ref. to Figure 16). This characteristics could also be observed in a deviation of the predicted knock frequency from the target value during the transient changes. From further investigations two reasons were identified. The knock frequency prediction includes solely AI mean of the finished cycle but not an estimation of its variation, similarly to the estimation of the load for the upcoming cycle. This is a simulation-specific limitation, as AI mean could only be determined subsequent to a combustion calculation, but might not be relevant for an engine application, depending on the specific method to determine AI mean within an engine application. As second reason, the load estimation accuracy was identified. At different absolute values of the knock frequency, similar load estimation inaccuracies yield different knock frequency prediction inaccuracies. This effect could be demonstrated by varying the duration of the transient phase. For longer duration, the predicted knock frequency changes by a smaller amount between each cycle and thus the error is smaller and knock frequency prediction and spark timing adjustment more accurate. Higher Efficiency through model-based, predictive Knock Control 371 371 <?page no="372"?> Figure 16: Comparison of control response under transient load variation. Finally, from the stationary simulations, efficiency increase potential is quantified by means of fuel saving and CO 2 emission reduction potential. As shown in Figure 17 the new probability-based knock controller allows for a fuel consumption and CO 2 emission reduction of up to 1-%. Figure 17: CO 2 emission and fuel consumption reduction potential. 5 ECU Application of Predictive Knock Controller 5.1 Model integration and offline testing To test the potential of the novel developed predictive knock control model under real conditions on a single cylinder testbench, it must be translated from the model concept into a realtime capable, compilable RCP-Software model. For this application on the RCP Hardware, it must be considered, that the requirements for model complexity in terms of real-time capability are different from those for the 0D-Simulation model used previously. 372 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 372 <?page no="373"?> Figure 18: Functional Description of Predictive Knock Controller. The Predictive Knock Controller model developed for the 0D-Simulation needs to be adapted for the implementation on RCP-Hardware. To achieve real-time capability the boundaries for the model complexity and build are more restrictive than in the previously used 0D-Simulation model. Therefore, the previous model for simulation purposes had to be simplified in terms of the calculation complexity. Figure 18 shows the developed, simplified structure for the Predictive Knock Controller model which afterwards is converted into a compilable build in MATLAB/ Simulink. After the build, the model is tested regarding its real-time capability and, if necessary, adapted and improved. The Predictive Knock Controls model for the RCP application is depending on the main input for the calculation of its control parameters, namely the engine speed, the indicated mean effective pressure (IMEP) and the current ignition angle. Additionally, the initial pressure and temperature inside the cylinder is used as an input to the calculation, although it is not determined from the ECU sensors but taken from a calibrated map. Based on these inputs, the calculation of the three parameters will be performed in the main calculation block of the Predictive Knock Controls model. Hereby the current values for the AI Limit and the σ AI are determined by a map-based approach with lookup tables utilizing the three main input parameters of the model and a parametrization from the previous simulation. The calculation of the of the main parameter μ AI onset does not use the map-based approach, since this, in contrast to the other two parameters, is a lot of effort to be mapped via characteristic parameters. This approach greatly increases the computational effort, and the entire calculation must be carried out in a time-critical manner within a working cycle. Therefore, this calculation is to be seen as the critical point for the real-time capability. The calculation of the μ AI onset is composed of three main calculation steps: The crankshaft discrete calculation of the cylinder pressure and temperature curve based on the vibe-ap‐ proach, afterwards the determination of the ignition dwell time depending on the cylinder Higher Efficiency through model-based, predictive Knock Control 373 373 <?page no="374"?> (5.1) (5.2) (5.3) pressure and temperature, and finally the calculation of the two-stage integral according to Fandakov based on this. For the calculation of the Vibe function, the required vibe parameters a,m,φ VB ,φ VD and Q um are to be taken from a lookup table for the respective operating point. Then, with the help of the Vibe function, the burn rate and the pressure curve in the cylinder are to be determined with these parameters. This is then used to additionally determine the temperature in the cylinder. The functions used can be found in the formula below. dQ ℎ dφ = Q um φ V D * m + 1 * φ − φ V B φ V D m * ε −α * φ − φV B m + 1 φV D p(φ) = dQ ℎ + V (φ) * p φ φ − 1 κ p − 1 φ − φ φ − 1 κ p − 1 * dV + φ − φ φ − 1 * dV + 1 κ p − 1 * V T (φ) = T φ φ − 1 * p φ φ − 1 1 − n n p φ To represent the curve of the in-cylinder temperature also at higher temperatures, a cylinder pressure-dependent correction factor is multiplied for temperatures > 700° C. From the curve of the in-cylinder pressure and temperature the reciprocal value of the ignition delay time is derived for both stages of the Fandakov integration by use of precalibrated lookup-tables from the simulation. The Fandakov integration determines the μ AI onset by integration of the reciprocal value of the ignition dwell time. For the first stage, low stage value is integrated for each step in the calculation until the value of one is reached. Afterwards the second stage starts with an initial value of 0.3 and integrates the high stage value again until the calculation value equals one. The corresponding crank angle on which the second stage of the Fandakov integration reaches one is the value of the μ AI onset for this operating point and is then passed on to the calculation of the knocking frequency. The calculation of the knock frequency is then carried out by calculating the frequency of the normal distribution based on the parameters μ AI onset and σ AI at the AI Limit position and compared with a limit value for the frequency. If this value is exceeded by a certain threshold, the ignition angle is retarded by a presettable value. In case the value is below a certain threshold, the ignition angle will be advanced. The calculation of the Predictive Knock model is performed for every working cycle to determine the knock frequency for the actual operating point and adapt the ignition angle for the following workingcycle. This determines the available computing time for the performance of one calculation, which must be considered in the following integration of the Predictive Knock controller within the RCP toolchain. 374 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 374 <?page no="375"?> This functional description of the model was then converted into a simulink model. First, the framework for the model was set up with all the necessary settings for the subsequent compilation for the RCP hardware. The general structure was then implemented and the individual submodels created. During this, the individual characteristic maps were also generated with attention to a parametrization range, which has sufficient accuracy. In addition, the resolution accuracy for the calculation of the μ AI onset was determined. It was visible that the step size of 0.1° crankshaft angle used in the simulation leads to the fact that the real-time capability for a cycle is no longer given for a worst-case calculation range of 85° CA in the application on the RCP control unit due to the specified task time of at least 0.1 ms. Therefore, a step size of 1° CA was set for the calculation as a tradeoff between accuracy and real-time capability in the RCP model. Later it had to be validated that this step size has sufficient accuracy during testing on the engine test bench. As a fallback solution, a map-based approach was also integrated for the entire calculation of the μ AI onset , which, like the calculation of AI Limit and σ AI , is based on the engine speed, IMEP and ignition angle. To parameterize this map-based approach, it was necessary to run the calculation sub-model for the calculation of μ AI onset over the parameterization range of the entire model in order to generate sufficient data for the parameterization. To ensure that the model for implementation on the RCP control unit delivers the same results as the results of the previous simulation the model was tested open loop. For this purpose, the same input data was applied to the model in an open loop simulation test in Simulink as was already used in the simulation. The data from the RCP model and the simulation were then compared with each other. If deviations occurred, changes were made in the parameterization of the characteristic maps in the model or, where necessary, adjustments were made to the calculations of the model. To test the calculation of the μA I onset , the calculations of the cylinder pressure and temperature curve were first simulated in the RCP model and the results compared. Initially, there was a strong deviation in the calculation results. This could be greatly reduced by adjusting the MATLAB script. The remaining deviation indicated an insufficient parameterization quality of the characteristic maps. After two iteration steps in the parameterization of these, a sufficient calculation accuracy for the calculation of pressure and temperature in the cylinder could be achieved. The result of one point from this model validation can be seen in the following Figure 19. The example of the operating point n = 1500-1/ min, IMEP = 12-bar and an Ign angle = 6.9-°C BTDC shows that the deviation of the calculated values of the RCP model and the simulated values of the 1D simulation had a deviation of less than 0.1%. Afterwards the compiled model was tested on the RCP hardware in order to determine the possible task time for the model execution and again validate the accuracy of the model. The real-time capability of the model is directly dependent on the task time in which it is executed. Here, a faster task time means a faster calculation of the entire model and thus the calculation can be carried out more in a working cycle. However, it should be noted that the shorter the selected task time, the lower the complexity of the overall model can be, as the calculation performance decreases with decreasing task time. At this point, a suitable calculation speed must be found that is still capable of executing all the necessary models on the RCP hardware. Only task times of less than 1-ms Higher Efficiency through model-based, predictive Knock Control 375 375 <?page no="376"?> can be considered for a reliable calculation of the 85 calculation steps at a model resolution of 1° CA, otherwise the goal of a possible calculation in a working cycle cannot be achieved. Figure 19: Model testing validation for cylinder pressure and temperature in Simulink. For this purpose, the predictive knock model was first compiled and installed for task times of 1 ms and 0.1 ms for the RCP hardware. Then the model was run on the RCP hardware with fixed model inputs and the computing load of the RCP hardware was monitored. Figure 20 shows that with a task time of 1 ms only a low computing load of 4 % is caused by the Predictive Knock Control model, whereas with the lower task time of 0.1 ms a computing load of 45 % is occupied by the model. It should be noted that not only the predictive knock control model must be operated on the RCP hardware for operation on the single-cylinder test bench, but also all other models necessary for operating the engine. For this reason, a task time of 0.5-ms was used for further use of the model. Figure 20: tComputing load for different Task Times on RCP Hardware. 376 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 376 <?page no="377"?> Subsequently, values from the parameterization area of the model were applied to the model as input in order to test whether the model has the same calculation quality on the RCP control unit as in the simulation in Simulink. The comparison shows that the RCP model delivers the same. The final step of the model integration was the commissioning of the predictive knock control model on the single-cylinder test bench. First, the single-cylinder engine was set up according to the same specifications that were used for training the simulation model for the development and parameterization of the Predictive Knock Control model. Then, all the required input variables had to be made available to the model. The ignition timing, the IMEP and the knock peak-to-peak value were transmitted from the indication system of the test bench to the RCP hardware via a CAN bus. The last required value of the engine speed was evaluated directly by the RCP hardware and transferred to the Predictive Knock Control model. After the general software was put into operation, the first step was to test the model’s behavior on the test stand. For this purpose, the model was operated in load points that were already used for the parameterization of the simulation model, in order to be able to compare the behavior of the simulation and the RCP model on the test bench. As can be seen in Figure 21, no stable ignition angle was established in the model and it toggles during activation of the controller. Since this behavior made a validation of the model impossible, this problem first had to be investigated and solved. Figure 21: Model behavior with calculation approach for μ AI onset during commissioning. Higher Efficiency through model-based, predictive Knock Control 377 377 <?page no="378"?> A closer look at the measurement results showed that this unstable behavior was caused by large jumps in the calculation of the knock frequency. This was directly dependent on the three parameters of the normal distribution σ AI , μ AI onset and AI Limit . From the measured data, it quickly became apparent that the calculation in the model for μ AI onset causes the largest jumps due to the resolution of 1° CA and that the other values only follow suit due to the changed operating point. As already mentioned, this calculation accuracy could not be changed with the chosen approach, as this would have resulted in the model losing its real-time capability. For this case, the fallback was implemented in the model in advance with a map-based approach for calculating the μ AI onset , which could now be used. After switching to this approach, the operating points were again set on the SCE and the model behavior was checked. It was found that with this solution, the accuracy of the model could be increased significantly as the resolution of the μ AI onset was higher. As can be seen in Figure 22, this solution enabled the model to achieve a stable ignition timing and a knock frequency within the set threshold of 4-6%. Figure 22: Model behavior with calculation approach for μ AI onset during commissioning. After this problem was solved, the implemented conventional knock control model was verified and showed the desired behaviour. 378 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 378 <?page no="379"?> 6 Testing Results of Predictive Knock Controller 6.1 Testing Comparison of Predictive Knock Controller to Base Measurement and Conventional Controller For the final validation of the CO 2 reduction potential of the newly developed Predictive Knock Control model, the model was to be run on the engine test bench at the operating points previously used in the simulation. For this purpose, the knocking operating points were first measured at high load in the steady-state operating points n = 1500 1/ min, 2500 1/ min and 4000 1/ min in the base without knock control. Subsequently, these operating points were measured with both an activated conventional knock control and the newly developed predictive knock controller and the results were compared. As an example of this evaluation the operation point of n= 2500 1/ min with a load IMEP = 21 bar was examined. The start ignition angle for this operating point was at α ign = -1° CA BTDC. In the base measurement of this operating point, 55 knocking cycles occurred, resulting in fknocking = 12.2% over the measured cycles. Figure 23: Base measurement at 2500-1/ min with 21-bar IMEP. Higher Efficiency through model-based, predictive Knock Control 379 379 <?page no="380"?> Figure 24: Measurement with conventional knock controller at 2500-1/ min with 21-bar IMEP. Figure 25: Measurement with predictive knock controller at 2500-1/ min with 21-bar IMEP. 380 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 380 <?page no="381"?> The measurement with a conventional knock controller showed an average ignition angle α ign = -2.4° CA after activation, which subsequently resulted in 12 knocking cycles or fknocking = 4-% measured knocking frequency during activation. In the measurement before the controller was activated, the calculated knocking probability of the model was f probability = 11.8 % and thus 0.4 % different from the measured knocking frequency of the basic measurement. After activation, the model controlled a calculated knock probability f probability = 5.2 % with an average ignition angle of -1.5°. The deviation from the conventional controller was 0.9° CA earlier. Subsequently, 20 knocking cycles were measured, resulting in a knocking frequency f knocking = 6 %. The deviation from the model calculation was 0.8-%. 6.2 Evaluation of model performance The model shows good calculation accuracy in predicting the knocking probability in 2 of 3 tested operating points. In the operating points at n = 1500 and n = 2500 1/ min, a deviation between 0.4 % and 4.3 % before activation of the controller and after activation of the controller of 0.25 % and 0.8 % is shown. Compared to the conventional controller, the predictive knock controller sets a mean ignition angle between 0.2° CA and 1.5° CA earlier and thus has the potential of a higher engine efficiency. In this range, it can be said that the parameterization of the predictive controller reflects the real engine behavior well. At the operating point n = 4000 1/ min, the newly developed controller shows a higher deviation in its calculation compared to the measured values. This suggests that the parameterization in this range does not optimally represent the engine behavior. In conclusion, it can be said that the controller exhibits stable control behavior at all operating points investigated after the changeover to the map-based approach. Table 6 then serves as an overview of the measurements. Table 6: Measurement result overview. Higher Efficiency through model-based, predictive Knock Control 381 381 <?page no="382"?> 6.3 CO 2 Emission Reduction Potential Evaluation The previous simulation showed a maximum CO 2 emissions reduction potential up to ΔCO 2 = 1 % and a possible reduction in exhaust gas temperature ΔT exh = 20 K. This potential should now be confirmed in the evaluation of the real measurements on the single-cylinder test bench. The measurements were evaluated regarding the average exhaust gas temperature and the average cycle efficiency for operation with conventional controllers and the predictive knock controller model. In addition, the peak cylinder pressure was evaluated to classify the different mechanical loads between the two approaches. Figure 26: Evaluation of cycle efficiency and cylinder peak pressure of conventional knock controller at n = 2500-1/ min. In the operational point n = 2500 1/ min at IMEP = 21 bar the evaluation of the conventional controller showed an average cycle efficiency of η cycl.eng = 34.3 %. The average cylinder pressure w aspcyl,avg = 76.9 bar with a maximum of p cyl,max = 100.85 bar and a minimum of p cyl,min = 56.3 bar peak pressure. The measured exhaust gas temperature before turbine was Texh = 502.0-°C. 382 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 382 <?page no="383"?> Figure 27: Evaluation of cycle efficiency and cylinder peak pressure of predictive knock controller at 2500-1/ min. With the predictive knock controller, the cycle efficiency at this operating point improves by Δη cycl.eng = 0.54 % to η cycl.eng = 34.84 %. The average cylinder pressure increases by 2 bar to 80.87 bar. In addition, the exhaust gas temperature before the turbine is reduced by 20 K to 481.25-°C. This evaluation shows that the controller can also verify its potential shown in the simulation in real tests on the engine test bench and is able to increase the cycle efficiency by up to Δη eng =0.54-% and lower the exhaust energy by up to ΔT exh = 20-K. 7 Summary and conclusions 3D-CFD studies have been carried in order to generate a fundamental understanding and quantify certain thermodynamic properties unable to obtain from measurement data. A Thickened Flame Model (TFM) has been applied and validated to allow for cost-effective but accurate prediction of the knock tendency. An investigation for a virtual operating point over the knock limit, with 47 out of the 100 cycles above the knock limit, has been utilized for statistical evaluation with respect to relevant influence factors. Only local conditions at the spark plug have been identified to reveal solid trends for the center of combustion and thus also knock tendency. Due to a lack of current measurement techniques, this cannot directly be utilized for novel knock controllers. In parallel 3D-CFD combustion studies, the Higher Efficiency through model-based, predictive Knock Control 383 383 <?page no="384"?> boundaries of the detonation to deflagration diagram have been verified for gasoline. This allows the application of this theory for knock control with more confidence. To support the development of a novel knock controller, various thermodynamic investi‐ gations were conducted to gain a better understanding of the knocking phenomenon. The aforementioned 3D-CFD simulations were used to investigate the influence of temperature and mixture inhomogeneities on the occurrence of knock. No direct correlation could be found between globally hottest temperatures, globally richest fuel mixture, locally hottest temperatures and charge velocities close to the spark plug gap and knock occurrence. Application of the detonation diagram to 0D simulation enabled a trend prediction of the knock tendency with less accuracy compared to a knock frequency prediction based on the auto-ignition onset distribution. The lower accuracy was accounted to the lack of validation data for the crucial hotspot size and the low variability of the detonation parameter close to the knock boundary. Thus, limited suitability of the detonation diagram for operating conditions close to the knock boundary without excessive pressure oscillations was noted. Further, simulation of single working cycles to model the cycle-to-cycle variations enabled an accurate prediction of the knock frequency. Subsequently, the knock frequency calculation method was reduced to increase perform‐ ance to make it applicable for a knock frequency-based knock control concept. The simulative investigation of the new knock control concept revealed a fuel-saving and CO 2 emission reduction potential up to ΔCO 2 = 1-%. The previously developed novel knock frequency-based knock control concept was then translated from the concept phase into a Simulink model. After several test phases, both on the software and hardware side, the model was transferred into a fully-fledged, real-time capable RCP model in order to validate the previously determined potential for CO 2 emissions reduction in real test operation on a single-cylinder test bench. This validation on the single-cylinder test bench showed that the general concept of knock frequency-based knock control works and that a stable ignition timing with a specified knock frequency can be set. However, the calculation concept in the model had to be changed in order to increase the accuracy of the model. The evaluations showed that in the measured range, the calculated knock frequency deviated from the measured knock frequency by 0.2 % to 3.1 %. Through a revised parameterization or using a built-in adaptation function, a higher accuracy can be achieved here. In addition, the validation confirmed the CO 2 emission savings potential from the simulation when using the new knock control concept. Here, efficiency increases of up to Δη cycle = 0.54 % and a reduced exhaust gas temperature up to ΔT exh = 20 K were shown. All benefits could be realized without adding to the complexity of the engine, e.g. by adding an additional measurement system. 384 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 384 <?page no="385"?> 8 Literature References [1] Blomberg M., Hess M., Hesse R., Morsch P., “Engine Knock Model,” Final Report on FVV Project 1313, 2021. [2] Bradley D., Morley C., Gu X. J., Emerson D. R., “Amplified Pressure Waves During Autoignition: Relevance to CAI Engines,” SAE Technical Paper 2002-01-2868, 2002, doi: 10.4271/ 2002-01-2868. [3] Hasti, V., Liu, S., Kumar, G., Gore, J., Comparison of Premixed Flamelet Generated Manifold Model and Thickened Flame Model for Bluff Body Stabilized Turbulent Premixed Flame, AIAA Aerospace Sciences Meeting, AIAA SciTech Forum, 2018. [4] Colin, O., Ducros, F., Veynante, D., and Poinsot, T., “A thickened flame model for large eddy simu‐ lations of turbulent premixed combustion,” Physics of Fluids, 12(1843), 2000. DOI: 10.1063/ 1.870436. [5] Charlette, F., Meneveau, C. and Veynante, D. A power-law flame wrinkling model for LES of premixed turbulent combustion part I: non-dynamic formulation and initial tests. Combustion and Flame, 131(1): 159{180, 2002. [6] Sankalp, L., “Leveling Up Scaling with CONVERGE 3.0,” Convergent Science, 2020. [Online]. Available: https: / / convergecfd.com/ blog/ leveling-up-scaling-with-converge-3-0. [7] Peters N., “Mega-knock in super-charged gasoline engines interpreted as localized developing detonation,” 4th conference knocking gasoline engines, IAV Automotive Engineering GmbH, Berlin, 2013, pp. 23-40. [8] Blomberg M., Fajt N., Leyens L., “Fast Knocking Prediction for Gasoline Engines,” Final Report on FVV Project 1370, Frankfurt am Main: Research Association for Combustion Engines (FVV) e.-V., 2022. [9] Bates L., Bradley D., Paczko G., Peters N., “Engine hot spots: Modes of auto-ignition and reaction propagation,” Combustion and Flame Vol. 166, 2016, pp. 80-85. [10] Kalghatgi G. T., Bradley D., “Pre-ignition and ‘super-knock’ in turbo-charged spark-ig‐ nition engines,” International Journal of Engine Research 13(4), 2012, pp. 399-414, doi: 10.1177/ 1468087411431890. [11] Netzer C., Seidel L., Pasternak M., Klauer C., Perlman C, et al., “Engine Knock Prediction and Evaluation Based on Detonation Theory Using a Quasi-Dimensional Stochastic Reactor Model,” SAE Technical Paper 2017-01-0538, 2017, doi: 10.4271/ 2017-01-0538. [12] Robert A., Richard S., Colin O., Poinsot T., “LES study of deflagration to detonation mechanisms in a downsized spark ignition engine,” Combustion and Flame Vol 162, 2015, pp. 2788-2807, doi: 10.1016/ j.combustflame.2015.04.010. [13] Hess M., Grill M., Bargende M., “0D/ 1D Knock Criterion to Predict the Knock Boundary of SI Engines,” 6th International Conference on Knocking in Gasoline Engines, Berlin, 2022. [14] Hess M., Grill M., Bargende M., Kulzer A. C., “Two-Stage 0D/ 1D Knock Model to Predict the Knock Boundary of SI Engines,” 21th Stuttgart International Symposium, pp. 514-530, 2021, doi: 10.1007/ 978-3-658-33466-6_37. [15] Hess M., Grill M., Bargende M., Kulzer A., “Knock Model Covering Thermodynamic and Chemical Influences on the Two-Stage Auto-Ignition of Gasoline Fuels,” SAE Technical Paper 2021-01-0381, 2021, doi: 10.4271/ 2021-01-0381. [16] Fandakov A., Grill M., Bargende M., Kulzer A. C., “A Two-Stage Knock Model for the Development of Future SI Engine Concepts,” SAE Technical Paper 2018-01-0855, 2018, doi: 10.4271/ 2018-01-0855. [17] Franzke D. E., “Beitrag zur Ermittlung eines Klopfkriteriums der ottomotorischen Verbrennung und zur Vorausberechnung der Klopfgrenze,” Ph.D. Thesis, Technical University of Munich, 1991. Higher Efficiency through model-based, predictive Knock Control 385 385 <?page no="386"?> [18] Worret R., Bernhardt S., Schwarz F., Spicher U., “Application of Different Cylinder Pressure Based Knock Detection Methods in Spark Ignition Engines,” SAE Technical Paper 2002-01-1668, 2002, doi: 10.4271/ 2002-01-1668. [19] Grill M., Billinger T., Bargende M., “Quasi-Dimensional Modeling of Spark Ignition En‐ gine Combustion with Variable Valve Train,” SAE Technical Paper 2006-01-1107, 2006, doi: 10.4271/ 2006-01-1107. [20] Grill M., Bargende M., “The Development of an Highly Modular Designed Zero-Dimensional Engine Process Calculation Code,” SAE Int. J. Engines 3(1): 1-11, 2010, doi: 10.4271/ 2010-01-0149. [21] Wenig M., “Simulation der ottomotorischen Zyklenschwankungen,” Ph.D. Thesis, University of Stuttgart, 2013. [22] Wenig M., Grill M., Bargende M., “A New Approach for Modeling Cycle-to-Cycle Variations within the Framework of a Real Working-Process Simulation,” SAE Int. J. Engines 6(2): 1099-1115, 2013, doi: 10.4271/ 2013-01-1315. [23] Fajt, N., Grill M., Bargende M., “Knock Probability Prediction and its Potential for a Knock Control Application,” 6th International Conference on Knocking in Gasoline Engines, Berlin, 2022. [24] Peyton Jones J. C., Muske K. R., Frey J., “A Stochastic Knock Control Algorithm,” SAE Technical Paper 2009-01-1017, 2009, doi: 10.4271/ 2009-01-1017. [25] Peyton Jones J. C., Spelina J. M., Frey J., “Likelihood-Based Control of Engine Knock,” IEEE Transactions on Control Systems Technology Vol. 21, No. 6, 2013, pp. 2169-2180, doi: 10.1109/ TCST.2012.2229280. [26] FKFS Research in Motion, FKFS UserCylinder®, https: / / www.fkfs.de/ en/ competencies/ virtual-d evelopment/ virtual-engine-development, 2021. [27] Gamma Technologies, GT-Suite, https: / / www.gtisoft.com/ gt-suite, 2022. 386 Michael Fischer, Michael Grill, Andre Kulzer, Marco Guenther, Stefan Pischinger 386 <?page no="387"?> ISBN 978-3-381-12991-1 6 th International Conference on Ignition Systems for SI Engines 7 th International Conference on Knocking in SI Engines MARC SENS (ED.) Ignition Systems for SI Engines Knocking in SI Engines With these two conferences, we at IAV have been offering interested researchers and developers the opportunity to exchange information on the latest developments in the field of ignition and combustion of Otto cycle engines for more than two decades. And this exchange is more urgently needed today than ever before, because the introduction of low and zero carbon fuels of synthetic or biological origin, such as hydrogen, ammonia, methanol, synthetic diesel or gasoline, etc. brings with it new challenges, which in turn require innovative solutions. Only through the exchange of knowledge and ideas will we be successful in making the combustion engine even cleaner and more efficient. Our clear goal is to maintain these two conferences as the “last man standing” for exchange in this highly interesting and important subject area! Marc Sens, IAV GmbH MARC SENS (ED.)