24th International Colloquium Tribology - January 2024 159 Towards the Prediction of Lubricated Contacts by Machine Learning Max Marian 1* 1 Department of Mechanical and Metallurgical Engineering, School of Engineering, Pontificia Universidad Católica de Chile, Macul, Chile * Corresponding author: max.marian@uc.cl 1. Introduction The prediction of lubricated tribo-contacts plays a critical role in optimizing mechanical system performance and durability. However, accurately forecasting their behaviour is a complex and numerically costly task due to the intricate interplay between contacting surfaces, lubricants, and the environment. Artificial Intelligence (AI) Machine learning (ML) techniques have emerged as powerful tools to enhance prediction accuracy and efficiency [1]. This contribution explores the utilization of ML algorithms, such as artificial neural networks, to model and predict the behaviour of lubricated tribo-contacts. 2. Prediction of hydrodynamic contacts using Physics-Informed Machine Learning Utilizing physics-informed machine learning (PIML), one can seamlessly integrate foundational physics knowledge into ML models, particularly for hydrodynamically lubricated (HL) contacts. Methods like physics-informed neural networks (PINN) leverage established physical principles and equations specific to lubricated contacts, such as the Reynolds equation, see Fig. below. This integration serves to guide the learning process, resulting in precise and interpretable models. During the training phase, the relevant equations become an integral part of the neural network‘s loss functions, introducing a dual aspect that blends data-driven and physics-driven elements into the loss function. This is achieved by sampling input training data, encompassing spatial coordinates and/ or time stamps, and processing them through the neural network. Subsequently, the network‘s output gradients are computed with respect to these inputs at designated locations. These gradients can often be efficiently calculated using auto differentiation (AD) and are then employed to determine the residual of the underlying differential equation. This residual is incorporated as an additional term in the loss function. The inclusion of this „physics loss“ in the overall loss function serves to ensure that the solution acquired by the network adheres to the established laws of physics. This approach has rapidly evolved in just two years, progressing from the 1D Reynolds equation for converging sliders to more intricate scenarios, including the 2D Reynolds equation, addressing journal bearings with load balance, variable eccentricity, and accounting for cavitation effects. [2] 3. Data-driven prediction of elastohydrodynamic contacts using Machine Learning Apart from PIML, ML algorithms can be trained to learn patterns in data sets. Thus, approaches, such as support vector machines, Gaussian process regressions and artificial neural networks can predict relevant film parameters or even film thickness distributions more efficiently and with higher accuracy and flexibility compared to sophisticated elastohydrodynamic lubrication (EHL) simulations and analytically solvable proximity equations, respectively. By optimizing the ML algorithm hyperparameters, it is possible to achieve high prediction accuracies. However, the data set to train the ML models is of crucial importance. Graphical representation of a PINN approach. Reprinted from [2] with permission by CC BY 4.0. 160 24th International Colloquium Tribology - January 2024 Towards the Prediction of Lubricated Contacts by Machine Learning Data-based predicting EHL contact parameters using artificial neural networks. Reprinted from [3] with permis-sion by CC BY 4.0. Acknowledgement The support from the Vicerrectoría Académica (VRA) of the Pontificia Universidad Católica de Chile within the Programa de Inserción Académica (PIA) is greatly acknowledged. References [1] M. Marian, S. Tremmel: Current Trends and Applications of Machine Learning in Tribology—A Review, Lubricants, 9, 2021, 86, DOI: 10.3390/ lubricants9090086. [2] M. Marian, S. Tremmel: Physics-Informed Machine Learning—An Emerging Trend in Tribology, Lubricants, 11, 2023, 463, DOI: 10.3390/ lubricants11110463. [3] M. Marian, J. Mursak, M. Bartz, F. J. Profito, A. Rosenkranz, S. Wartzack: Predicting EHL Film Thickness Parameters by Machine Learning Approaches, Friction, 11, 2023, 6, DOI: 10.1007/ s40544-022-0641-6.