Microstructure Design and Fatigue Life Prediction of High-Fatigue-Resistance A356 Alloy via Machine Learning

A356 aluminum alloy is widely used in automotive applications, yet its fatigue property influenced by microstructural features still needs improvement. This study integrated machine learning (ML) with microstructural characterization to design fatigue-resistant A356 alloys. A microstructure-fatigue database was established, correlating fatigue life with 16 microstructural parameters and stress amplitudes. Six ML models were evaluated, with the BP-ANN model demonstrating superior accuracy (R² = 0.84, RMSE = 0.36, MAE = 0.29). Objective optimization via NSGA-II genetic algorithm determined the range of optimal microstructural parameters, and experimental validation confirmed that the optimized alloys achieved a pre-eminent fatigue limit with 118 MPa and without failure under 120 MPa. SHAP analysis combined with microstructural characterization were used to reveal the mechanisms of microstructure features’ influence on fatigue life. Pores are the most detrimental features and will become fatigue crack sources and reduce fatigue life significantly when larger than the critical threshold (32 μm). EBSD and TEM results show that, below this threshold, SDAS, β-Fe, and eutectic Si phases are the important features influencing fatigue life in sequence. Refining the secondary dendrites, β-Fe, and eutectic Si particles reduces the stress concentrations and prevents intergranular crack propagation, thereby increasing the fatigue life. This study provides a theoretical foundation for the high fatigue-resistant design of low-pressure die casting A356 alloys.