Finite-element and machine-learning framework for dissipated-energy prediction and slip-regime classification in fretting contacts

Fretting contacts are a major source of energy dissipation, surface degradation, and fatigue damage in mechanical assemblies subjected to small-amplitude oscillatory motion. Predicting fretting energy dissipation and identifying the corresponding slip regime remain challenging because of the complex relationship among contact mechanics, frictional response, and slip evolution. This study proposes an integrated finite element and machine learning framework for predicting dissipated energy and classifying slip regimes in a ball-on-flat configuration. The finite element model was analytically verified against classical Hertzian contact theory and the Cattaneo-Mindlin partial-slip solution, with an error of 0.43% and 1.99% at two different normalized tangential force ratios. A dataset of 170 simulations was generated by varying normal load, displacement amplitude, and coefficient of friction. Additional features were derived using contact-mechanics based approach and evaluated using four regression and four classification algorithms. The models were optimized using repeated five-fold cross-validation with 10 repetitions. XGBoost Regression with derived features achieved the strongest held-out regression performance, with R 2 of 0.966 and RMSE of 0.510, and the highest mean cross-validation R 2 of 0.912 ± 0.063. Logistic Regression with derived features achieved test accuracy and macro-F1 values of 1.000, while the baseline Support Vector Classifier achieved the highest mean cross-validation macro-F1 of 0.914 ± 0.115. The framework enables rapid approximation of FEM-derived fretting responses within the investigated parameter domain.