Machine learning approach for predicting and understanding fatigue crack growth rate of austenitic stainless steels in high-temperature water environments

A machine learning (ML) approach is proposed to predict and understand the corrosion fatigue (CF) crack growth rate of austenitic stainless steels (SSs) in high temperature water. Six commonly used supervised ML algorithms were considered here and shown to perform exceptionally well. Among ML models, categorical boosting (CB) model was shown to perform best. The considered ML models were also compared to and shown to substantially outperform the existing empirical models. The CB model, which has accurately learned and captured important hidden patterns in the data, was explained/interpreted using the Shapley Additive explanation (SHAP) method. Such an approach allowed to unearth various meaningful patterns hidden in the studied data, including non-linearities and interactions of feature effects on the CF crack growth rate, which were overlooked by the existing empirical models. These findings would be helpful to improve the understanding of the CF crack growth behavior. Finally, a validation testing on the data out of the original data set confirmed the applicability of the CB model.