Bridging scales: Enhanced practical modeling of rolling bearings by AI-powered prediction of EHL contacts

Prof. Dr.-Ing. Max Marian

Summary

Accurate prediction of elastohydrodynamic lubrication (EHL) is a cornerstone for understanding friction, wear, and efficiency in rolling bearings. While conventional EHL solvers provide highly resolved pressure and film thickness distributions, their computational demands hinder large-scale applications such as drivetrain-level optimization. In contrast, simplified regression-based models are fast but lack physical fidelity.

This contribution presents recent advances in artificial intelligence (AI)-powered modeling of EHL contacts, where artificial neural networks (ANNs) trained on high-fidelity datasets predict locally resolved film thickness and pressure distributions at negligible computational cost. These models can be applied in two ways: (i) as efficient initializations for EHL solvers, drastically reducing computation time, and (ii) directly within system-level bearing and drivetrain simulations, replacing oversimplified regression fits with physically grounded, data-driven predictions. Case studies demonstrate both accuracy and efficiency, enabling friction-, lubrication-, and wear-related phenomena to be modeled across scales.