Prediction of fatigue crack propagation behavior in elastic plastic region under block loading for type 316 steel via artificial neural network approach

This study investigated the fatigue crack propagation behavior from small defect in the elastic–plastic region under block loading conditions and clarified the influence of cycle ratio on the crack growth rate. Fatigue tests were conducted under constant and repeated two-step strain amplitude loading conditions using various cycle ratios. The results of the constant amplitude loading test indicated that the J integral can be employed to predict fatigue crack propagation rate in one master curve by considering the material constant, l₀. The results of the repeated two-step test showed that the fatigue life evaluated via the J integral had a larger scatter for test conditions at strain amplitudes of 1.0%/0.2% and 0.8%/0.2% with various cycle ratios. A highly accurate model was established to predict fatigue crack propagation behavior and investigate the effect of algorithms on the precision of models. To achieve this, three deep learning algorithms feed forward neural network (FFNN), cascade-forward neural network (CFNN) and function fitting neural network (FNN), were employed. It was observed that the precision of the constructed models was dependent on the algorithms and dataset split. The model constructed using the CFNN exhibited the highest prediction accuracy.