Data-Driven Surrogate Modeling Techniques to Predict the Effective Contact Area of Rough Surface Contact Problems

Kurzfassung

This paper addresses the problem that accurate methods for computing the effective contact area of rough surfaces, like the Boundary Element Method (BEM) are too slow for applications requiring many repeated evaluations e.g., uncertainty quantification, parameter identification, multi-scale simulations. The authors build surrogate models - machine learning models trained on a large precomputed dataset of BEM results. Given a surface's statistical descriptors and loading conditions, the model predicts the effective contact area almost instantly instead of running the full BEM solve.