Machine learning-based approach for predicting the fatigue crack growth in the radial direction in steel pipe under pure bending

Abstract The present investigation elucidates the prediction of crack length ( a ) and fatigue crack growth rate (FCGR), in the TP316L stainless steel pipe during four-point bending, using different approaches— ridge regression (RR), random forest (RF), and polynomial regression (Poly.) machine learning (ML) modelling. The FGCR i.e., d a /d N , in the radial direction of the cylindrical pipe was predicted considering 70% of the data for training and 30% for the testing. Initially, a was predicted considering the number of loading cycles ( N ) as the independent variable. The Poly. method provided the best mean value of the coefficient of determination while the RR method provided the least value. Considering the mean squared error (MSE), it was the RF method that provided the best prediction. The prediction of d a /d N was done considering stress intensity factor range (Δ K ), a , and N as independent variable. When Δ K was considered, the RF method provided the best coefficient of determination, Poly. method provided a conservative value and the RR method could not predict the d a /d N . A similar prediction trend was obtained considering a and N as the independent variables. The optimum coefficient of determination for d a /d N was obtained when the a was considered as the independent variable. The MSE for the prediction of d a /d N provided the best results using the RF method considering Δ K , a , and N as the independent variables.