lies in using data-driven technologies to automate the data processing and evaluation of static and kinematic coefficients of friction, by combining tribological knowledge and experience with the best data science practices, minimizing this way the data evaluation errors. The present paper is structured as follows: Section 2 analyses literature regarding the automated evaluation of coefficients of friction and reviews applications of data science methods in tribology. Section 3 presents the methodologies applied for data acquisition and signal processing. Section 4 discusses the results obtained with the applied automated methodology and outlines future work. Finally, the paper presents concluding remarks in Section 5. 2 State of the art In recent years, the use of artificial intelligence in various case studies in tribology has been increasingly appraised and considered beneficial [5-8]. Furthermore, the practice of applying semi-automated and statistical methods, such as planning the experimental matrices by Aus Wissenschaft und Forschung 22 Tribologie + Schmierungstechnik · 70. Jahrgang · 3/ 2023 DOI 10.24053/ TuS-2023-0014 1 Introduction The complexity of tribological interactions and various boundary conditions make tribology a highly experimental field. The evaluation of experimentally measured friction data can be a challenging task due to the large volume of data that must be processed to extract relevant signals. This is often done manually or semi-automatically, especially when it comes to extracting static and kinematic coefficients of friction measured on an oscillating test rig, a common test type in tribology. Since for the separate evaluation of static and kinematic coefficients of friction, high-frequency data acquisition is needed (large file size) and the data processing often requires significant computational power, this is often not performed by researchers, but rather an average out of raw data points per stroke or per unit time (e.g., an average value per second) is calculated by the software of the testing rig. However, the separate evaluation of the static and kinematic coefficients of friction provides additional insight into the system of interest and can often carry crucial information for the understanding of the relevant tribological mechanisms and their correlations with the properties of the tested materials [1,2]. With the development of measuring equipment and computational power, the possibility to introduce this option as a standard evaluation procedure in tribological testing is now closer and more relevant than ever before. Applying data science methods to tribological problems for enabling automated workflows and practices has already been reported [3,4]. In this respect, Bitrus et al. [3] presented an adapted data mining methodology that applied existing procedures to account for the specifics of the field of tribology. The key focus of the present study Enhanced Efficiency in Coefficient of Friction Evaluation through Automated Data Processing Igor Velkavrh, Katharina Dimovski, Fevzi Kafexhiu, Thomas Wright* In this work, an approach towards automated extraction and evaluation of static and kinematic coefficients of friction is presented. By replacing manual evaluation with an automated process, this approach yields promising results, simplifies data handling, and saves significant time. The methodology shows potential for application to a wide range of experimental data and provides advanced processing capabilities using the extracted and mathematically evaluated data points. However, for data with high signal-tonoise ratios, automatic detection still requires further optimization to improve accuracy. Keywords Automated Data Processing, Coefficient of Friction, Zero Detection, Savitzky-Golay Filter, Data Evaluation, Data Presentation Abstract * DI Dr. Igor Velkavrh (federführender Autor) Orcid-ID: https: / / orcid.org/ 0000-0002-4293-9978 Katharina Dimovski, BSc, MSc Orcid-ID: https: / / orcid.org/ 0000-0002-3107-6791 DI Dr. Fevzi Kafexhiu Orcid-ID: https: / / orcid.org/ 0000-0001-6848-9293 Dr. Thomas Wright: Orcid-ID: https: / / orcid.org/ 0000-0002-3475-0016 V-Research GmbH, Stadtstrasse 33, 6850 Dornbirn, Austria using the design of experiments (DoE) has been increasing [9]. Although digitization has a significant advantage over the current (mainly) manual approach in tribology [10-13], the possibility of digitizing laboratories is still very rudimentary. Both the in-laboratory and cross-institute data collection is not yet designed for efficient collaboration. The main problem with the processes lies in the connection between the heterogeneous measuring equipment and their partially non-digitized interfaces. In addition, cross-institute collaboration also requires flexibility in the execution of experiments [14]. The data evaluation methods are however specific to individual experimental setups as well as laboratories, and if a distributed test is carried out, there is no systematic approach that can, for example, directly compare the friction data evaluated in different laboratories. A distant goal in tribology is to unify the data processing procedures as for example in the field of medical analytics solutions the automated acquisition of standardized tests already exist [15]. 3 Methodology 3.1 Experimental setup and measurements Measurement data for the present study were generated on a modular friction and wear test rig RVM 1000 (Werner Stehr Tribologie GmbH, Germany). Tests were performed in oscillating sliding motion where the lower specimen is fixed, and the upper specimen is moving with a defined oscillating frequency. To generate exemplary data files for the evaluation, load step tests were carried out. In these tests, the normal load (force) is successively adjusted to several levels starting with the highest value. Between normal load levels, there is a pause time of 60 seconds to allow for the adjustment of the normal load. After the pause, the samples are set in motion for a short time, to reach 10 oscillating cycles, as shown in Figure 1. Such tests provide frictional forces for each load level and allow for the determination of the contact pressure dependency of the friction in tribological pairings. During the test, the normal load, the tangential force and other parameters such as temperature, oscillating frequency, etc., are continuously recorded. The data acquisition frequency is 1 kHz. The duration of the load step test is around 10 minutes. 3.2 Data understanding and pre-processing 3.2.1 Data understanding from a tribology perspective Friction and wear tests performed in the form of oscillating (reciprocating) sliding motion are used to replicate the nature of engineering component contacts such as the piston-cylinder contact in the crank-slider mechanism, fretting vibrations in drilling machines, etc. At the laboratory scale, the reciprocating sliding tests are performed using carefully controlled parameters such as normal load depending on the desired contact pressure between the bodies in the sliding contact, sliding velocity, stroke (displacement amplitude), temperature, contact interface (dry or lubricated), etc. The most important parameter that is derived from these tests is the coefficient of friction (CoF), which is the ratio between the tangential (friction) force in the sliding contact and the applied normal load. The high acquisition rate (1 kHz) of the tests in the present work provides detailed friction signals such as those presented with red and green dashed lines in Figure 2. In the CoF signal shown with a turquoise dashed line (Figure 2), the periodic minimum values represent the points where the oscillating body changes the sliding direction, i.e., from left to right or vice-versa. In these points, the sliding velocity equals zero, and the threshold friction value that must be overcome to re-initiate the sliding is called static CoF. Typically, it is characterised by the first peak (maximum) Aus Wissenschaft und Forschung 23 Tribologie + Schmierungstechnik · 70. Jahrgang · 3/ 2023 DOI 10.24053/ TuS-2023-0014 Figure 1: Schematic representation of the normal force and friction force signals during the load step test. In the third step, the semi-cycle length (defined using the extracted starting points) is utilized to determine the intervals in which the coefficients of friction reside. As shown in Figure 3 the definition of the intervals requires determining the points a, b, and c, where: - a represents the starting point minus 10 % of the previous semi-cycle length, - b represents the starting point plus 10 % of the current semi-cycle length, - c represents the starting point plus 90 % of the current semi-cycle length. The percentages can vary depending on the tested materials; the intervals are defined relative to the minimum values within each semi-cycle as well. Note that this method is sensitive to the sliding velocity (v) as the number of points per semi-cycle varies with v for a constant data acquisition rate. Aus Wissenschaft und Forschung 24 Tribologie + Schmierungstechnik · 70. Jahrgang · 3/ 2023 DOI 10.24053/ TuS-2023-0014 value after each minimum value. As soon as the sliding begins, the resistance created between the surfaces in sliding contact is known as sliding friction represented by the kinematic CoF. The kinematic CoF changes with the sliding velocity, which has a sinusoidal profile (simple harmonic motion) and is evaluated as a mean value between the recorded data points in a chosen range. 3.2.2 Pre-processing from a data science perspective To automatically extract static and kinematic CoF values from the turquoise dashed curve in Figure 2, the following algorithm was applied: in the first step, the recorded friction forces (red dotted line in Figure 2) were smoothed with a Savitzky-Golay filter (red solid line in Figure 2) to roughly define the position of the semi-cycles (inflexion points of the oscillating friction force). In the second step, starting points of the semi-cycles, defined as the consecutive minima of the measured friction force were extracted. Figure 3: Insight into a semi-cycle: the static CoF is located in the interval a: b (blue area), while the kinematic CoF represents the average of the interval b: c (grey area). Figure 2: Cutout of the data acquired in the load step test: the measured friction force (red dotted line), the friction force smoothed using the Savitzky-Golay filter (red solid line) and the measured CoF (green dashed line). The inflexion point of the smoothed friction force curve shows the beginning of a semi-cycle. In the fourth step, static and kinematic CoF values are extracted for each semi-cycle. The maximum value within the region a: b (blue area) is defined as the static CoF, while the average value of points in the region b: c (grey area) is defined as the kinematic CoF. In the fifth step, the average values and standard deviations for each load step are calculated. Since there are multiple data points per semi-cycle from which the kinematic CoF is derived and there are multiple cycles in a single experiment, the combined mean and standard deviation method [16] is applied. First, using Equations (1) the mean (x̅ ) and the standard deviation (σ) are calculated for each semi-cycle (subgroup) consisting of n points: (1) Next, the parameters from Equations (2), are calculated for the entire data set consisting of m semi-cycles (subgroups). These are the overall number of analysed points (N), the sum of products of averages and the number of points per subgroup (δ s ), and the sum of variances per subgroup (∆ s ): (2) Finally, using Equations (3) the combined mean (X¯) and overall standard deviation (σ s ) are calculated: (3) * - / 5 6 7 869 : 7 >? - @ 6 A * - B CD E F G E H E - B G E H E I * - B E JG E K LM N CD E H E - O P7 B JC E K CD M P E7 CD - P B C E P E For the static CoF, only a single point per each semi cycle is obtained, therefore the mean and standard deviation values from a single experiment, can be calculated using Equations (1), where n represents the number of semi-cycles. 4 Results and discussion The algorithm results in a series of extracted CoF values for each measurement separately, which can be further analysed using visual representations, such as the one shown in Figure 4. The static and kinematic CoF values are also available as a separate table, allowing for comparison between experiments or for further use in correlative analyses, related to the relevant material parameters such as the chemical composition, physical properties (hardness, roughness), etc. This provides a comprehensive and convenient way to analyse the results and gain valuable insights into the experiments. The algorithm has shown to work well for most of the experiments, especially when the friction signals have a distinct maximum at the beginning of motion. However, there are limitations with visco-elastic materials such as rubbers and foams where the static CoF does not appear as a typical peak at the beginning of motion, but rather the frictional force gradually increases and reaches its peak during the sliding motion. In these cases, the algorithm may not be able to detect static CoF accurately. Additionally, integrating machine learning methods or neural networks has the potential to yield a more detailed insight into the experiments by being trained with the frictional behaviour of each material and experimental setup. By taking into account the material properties and tribological system information, a more extensive analysis can be accomplished. Aus Wissenschaft und Forschung 25 Tribologie + Schmierungstechnik · 70. Jahrgang · 3/ 2023 DOI 10.24053/ TuS-2023-0014 Figure 4: Static and kinematic CoF values for a lubricated and dry contact in dependence of the applied contact pressure. Data was evaluated using the developed algorithm. Überwachung und optimierung tribologischer systeme. In: Tribologie in Industrie und Forschung: Effizienter durch Kooperation: Vortragsunterlagen Symposium 2020 der Österreichischen tribologischen Gesellschaft. pp. 40- 41. Österreichische tribologische Gesellschaft (2020) [5] Ciulli, E.: Tribology and industry: from the origins to 4.0. Frontiers in Mechanical Engineering 5, 55 (2019) [6] Kolodziejczyk, T., Toscano, R., Fouvry, S., Morales-Espejel, G.: Artificial intelligence as efficient technique for ball bearing fretting wear damage prediction. Wear 268- (1-2), 309-315 (2010) [7] Nowell, D., Nowell, P.: A machine learning approach to the prediction of fretting fatigue life. Tribology International 141, 105913 (2020) [8] Xu, X., Zhao, Z., Xu, X., Yang, J., Chang, L., Yan, X., Wang, G.: Machine learning-based wear fault diagnosis for marine diesel engine by fusing multiple data-driven models. Knowledge-Based Systems p. 105324 (Dec 2019). https: / / doi.org/ 10.1016/ j.knosys.2019.105324 [9] Antony, J.: Design of experiments for engineers and scientists. Elsevier (2014) [10] Lhymn, C.: Statistical Analysis of Wear Rate Data. Journal of Tribology 109(4), 594-597 (Oct 1987). https: / / doi.org/ 10.1115/ 1.3261514 [11] Lüchinger, M., Velkavrh, I., Kern, K., Baumgartner, M., Klien, S., Diem, A., Schreiner, M., Tillmann, W.: Development of a constitutive model for friction in bulk metal forming. Lubricants 6(2), 42 (Apr 2018). https: / / doi.org/ 10.3390/ lubricants6020042 [12] Borjali, A., Monson, K., Raeymaekers, B.: Predicting the polyethylene wear rate in pin-on-disc experiments in the context of prosthetic hip implants: Deriving a data-driven model using machine learning methods. Tribology International 133, 101-110 (May 2019). https: / / doi.org/ 10.1016/ j.triboint.2019.01.014 [13] Pimenov, D.Y., Bustillo, A., Mikolajczyk, T.: Artificial intelligence for automatic prediction of required surface roughness by monitoring wear on face mill teeth. Journal of Intelligent Manufacturing 29(5), 1045-1061 (Jun 2018). https: / / doi.org/ 10.1007/ s10845-017-1381-8 [14] Breu, R., Dustdar, S., Eder, J., Huemer, C., Köpke, J., Langer, P., Mangler, J., Mendling, J., Neumann, G., Rinderle-Ma, S., Schulte, S., Sobernig, S., Weber, B.: Towards living inter-organizational processes. In: Proceedings - 2013 IEEE International Conference on Business Informatics, IEEE CBI 2013 (2013). https: / / doi.org/ 10.1109/ CBI.2013.59 [15] Tong, M., Li, Y., Deng, C.,Wu, X., Li, Z., Hang, C., Sheng, H., Ding, H.: Enhancing laboratory efficiency with total laboratory automation. J Clin Chem Lab Med 3, 137 (2020) [16] Altman, D.G., Machin, D., Bryant, T.N., Gardner, M.J.: Statistics with Confidence Second Edition. BMJ Books (Jun 2013). ISBN 978-1-118-70250-5 Aus Wissenschaft und Forschung 26 Tribologie + Schmierungstechnik · 70. Jahrgang · 3/ 2023 DOI 10.24053/ TuS-2023-0014 5 Conclusions The present study demonstrates that the developed algorithm provides researchers with a notable reduction in time by enabling the automated evaluation of static and kinematic CoF from large data sets, paving the way for broader-scale analyses and machine learning implementation. Additionally, this research underscores the continued prevalence of manual evaluations in scientific research, while the industry has already embraced digitalization. Nevertheless, it is important to recognize that data science is a distinct field, encompassing essential process steps of data understanding and preparation before any implementation of machine learning or artificial intelligence methods. Therefore, only through accurate data interpretation, valuable information can be gleaned from research. Acknowledgements The work presented was funded by the Austrian Cooperative Research (Project SlipIT, no. SP-2021-05) and carried out in collaboration between V-Research GmbH, Austrian Research Institute for Chemistry and Technology (OFI) and Graz Centre for Electron Microscopy (ZFE). Parts of the work presented were funded by the Austrian COMET Program (Project InTribology, no. 872176) and carried out at the “Excellence Centre of Tribology” (AC2T research GmbH) in collaboration with V-Research GmbH. References [1] Voyer, J., Klien, S., Velkavrh, I., Ausserer, F., Diem, A.: Static and dynamic friction of pure and friction-modified PA6 polymers in contact with steel surfaces: influence of surface roughness and environmental conditions. Lubricants 7(2), 17 (2019) [2] Llavori, I., Zabala, A., Aginagalde, A., Tato, W., Ayerdi, J., Gómez, X.: Critical analysis of coefficient of friction derivation methods for fretting under gross slip regime. Tribology International 143, 105988 (2020) [3] Bitrus, S., Velkavrh, I., Rigger, E.: Applying an adapted data mining methodology (DMME) to a tribological optimisation problem. In: Data Science - Analytics and Applications: Proceedings of the 3 rd International Data Science Conference - iDSC2020. pp. 38-43. Springer (2021) [4] Fleisch, R., Velkavrh, I., Bitrus, S., Eigger, E., Voyer, J.: Entwicklung einer intelligenten plattform zur effizienten