Prediction of white etching area damage based on machine learning model combined with crystal plasticity under rolling contact fatigue

• A ML model combined with CPFEM is proposed for butterfly WEA formation under RCF . • The model captured the impact of inclusion shape, size, and depth on WEA. • LightGBM shows a clear advantage over SVR and BPNN in prediction accuracy. • Number of cycles is identified as the most influential factor for WEA formation. A machine learning model combined with crystal plasticity finite element method (CPFEM) was developed to predict the formation of butterfly white etching area (WEA) damage under rolling contact fatigue (RCF). Initially, RCF tests were conducted to generate butterfly-wing WEA. Following this, a CPFEM model based on phase-field constitutive theory was established for simulations. Subsequently, machine learning (ML) models were trained and evaluated using data obtained from experiments, literature, and CPFEM simulations. The XGBoost algorithm and SMOTE-NC algorithm were employed to address missing data and augment the dataset, respectively. The results showed that the ML model, combined with CPFEM, can successfully simulate the formation of butterfly WEA damage under different non-metallic inclusion parameters, such as shape, size, and depth. The simulation is in good agreement with experimental data, confirming the viability of using machine learning to predict WEA.