Physics-Informed Machine Learning for Real-Time Residual Stress Evolution and Fatigue Life Prediction in Multi-Layer Functional Coatings

Traditional Finite Element Analysis (FEA) for predicting residual stress (RS) evolution in multi-layer coatings is often computationally prohibitive for real-time industrial applications. This study introduces a novel hybrid framework that integrates Physics-Informed Neural Networks (PINNs) with stochastic surface morphology data to predict RS distribution and subsequent fatigue life. Unlike purely data-driven models, our approach incorporates elastoplastic partial differential equations directly into the loss function, ensuring the model adheres to fundamental mechanics.

We utilized a dataset combining high-resolution RS mapping (contour method) and validated FEA simulations of nano-silica/epoxy protective layers. Results demonstrate that this hybrid model achieves a 30x speed-up in stress-field estimation compared to beam-scale FEA, with a relative nodal temperature error below 5% (Frontiers in Materials, 2022). Furthermore, the model accurately captures the impact of stochastic surface roughness common in advanced manufacturing on compressive stress depth, which significantly influences crack initiation. By coupling RS evolution with the Walker mean stress model, we provide a robust tool for predicting the remaining useful life of coated assets. This research offers a transformative pathway for the end-to-end digital twin assessment of material performance under complex environmental loading.