Fatigue life prediction of aluminum alloy via knowledge-based machine learning

Fatigue life prediction based on small size of fatigue experimental datasets in specific materials is limited and easy to overfit. Herein, we devise a knowledge-based machine learning framework that combines empirical formulas and data-driven models to predict the fatigue life of seven different series of Al alloys. With the features designed by the proposed estimation and guesswork methods that transfer knowledge from empirical formulas, the machine learning model called gradient boost regression was constructed to predict the fatigue life of seven different series of Al alloys with a mean relative error of 140%, achieving a great improvement compared to the baseline model. The feature analysis shows that the logarithm of fatigue life linearly depends on the Stüssi and σmax1.5 features. The results have successfully demonstrated the advantages of knowledge-based machine learning, which provides a generic way to predict fatigue life for reducing experimental time and cost.