Physics-Informed Deep Learning for Lubricated Contacts with Surface Roughness as Parameter

Physics-informed neural networks (PINNs) are developed to solve variants of averaged Reynolds equations for accurately and time-efficiently modeling pressure build-up and cavitation in sealing contacts and journal bearings with rough surfaces. We use microscale coefficients provided through Patir and Cheng’s average flow model or homogenization to integrate roughness or texture height into these macroscale equations. Based on these equations we implement parameter-dependent PINNs to solve multi-case scenarios with varying roughness or texture heights, thus investigating the adaptability and generalizability of PINNs for modeling rough lubricated interfaces. The results demonstrate the promising potential of PINNs to accelerate tribological system computations.