Machine learning framework for predicting the low cycle fatigue life of lead-free solders

This study explores an efficient and reliable machine learning framework for determining the low cycle fatigue life of lead-free solders, which does not necessarily separately test different series of lead-free solder. With 1387 datasets from the published experiments and formulae, five main-stream machine learning models to date are adopted for the first time to predict the low cycle fatigue life for four different series of tin-based solders by considering the composition, loading and geometry factors. Based on feature importance and Shapley values, it is confirmed that the Boosting model is capable of capturing the nonlinear relationships of factors to influence the low cycle fatigue life of lead-free solder by greatly emphasizing the effects of plastic strain amplitude and temperature.