Conventional fatigue tests on complex components are difficult to sample, time-consuming and expensive. To avoid such problems, several popular machine learning (ML) algorithms were used and compared to predict fatigue life of gray cast iron (GCI) with the complex microstructures. The feature analysis shows that the fatigue life of GCI is mainly influenced by the external environment such as the stress amplitude, and the internal microstructure parameters such as the percentage of graphite, graphite length, stress concentration factor at the graphite tip, matrix microhardness and Brinell hardness. For simplicity, collected datasets with some of the above features were used to train ML models including back-propagation neural network (BPNN), random forest (RF) and eXtreme gradient boosting (XGBoost). The comparison results suggest that the three models could predict the fatigue lives of GCI, while the implemented RF algorithm is the best performing model. Moreover, the S-N curves fitted by the Basquin relation in the predicted data have a mean relative error of 15% compared to the measured data. The results have demonstrated the advantages of ML, which provides a generic way to predict the fatigue life of GCI for reducing time and cost.
KeywordsGray cast iron • Microstructure feature • Machine learning • High-cycle fatigue life * Jianchao Pang