Modeling Residual Stresses in Coated Components Using FEM and Machine Learning Hybrid Approaches

Abstract

Residual stresses in coated components critically influence their fatigue performance and durability. This study develops a hybrid computational framework combining finite element modeling (FEM) with machine learning algorithms to predict residual stress distribution in epoxy-based coatings. Simulations were performed under different coating thicknesses and thermal load conditions. FEM results provided detailed stress contours, while artificial neural networks (ANNs) were trained on FEM datasets to predict stress fields across new parameter ranges with high accuracy. Validation against experimental X-ray diffraction results confirmed predictive accuracy within 5%. The ANN model significantly reduced computational time compared to traditional FEM-only approaches. Feature importance analysis indicated coating thickness and thermal gradient as the most influential parameters. The developed hybrid framework demonstrates the feasibility of combining physics-based and data-driven approaches for efficient prediction of residual stresses, with direct implications for coating design and reliability assessment in aerospace and automotive industries.