Actual microstructure-based three-point bending finite element simulation coupled with FE-SAFE for machine learning fatigue life prediction of SiCp/Al composites

The fatigue performance of silicon carbide particle-reinforced aluminum matrix (SiCp/Al) composites is critical for their reliable service under cyclic loading in engineering structural applications. This work establishes an integrated computational and machine learning (ML) approach to estimate the fatigue response of SiCp/Al composites reinforced with 10, 20, 30, and 40 vol% SiC. Python-assisted finite element models were developed to reconstruct actual microstructures from SEM images and simulate uniaxial tensile and three-point bending mechanical responses, while FE-SAFE was utilized to predict fatigue life under cyclic loading. Experimental tensile and fatigue tests were performed to validate the FE simulations, with cyclic loading applied at 15 Hz under load-controlled conditions. Four ML models, namely Random Forest Regressor, CatBoost Regressor, XGBoost, and Artificial Neural Network, were developed using the selected input features and optimized through standard scaling, recursive feature elimination, and grid search hyperparameter tuning. All models demonstrated excellent predictive capability, with coefficient of determination (R²) values exceeding 0.98 for both training and testing datasets. Among them, the RFR model exhibited the best results (R² = 0.9978, MAE = 121.68, RMSE = 183.61) for the test dataset. Parity plot analysis further confirmed the close agreement between the ML-predicted fatigue life and the FE-SAFE simulation-based fatigue life, with most data points falling within the ±20% error tolerance. This integrated simulation-experimental-data-driven methodology offers a reliable and computationally efficient fatigue life prediction technique, facilitating the design of high-performance SiCp/Al composites for cyclic loading service conditions.