ANN vs. theoretical models for predicting crack path in fretting fatigue

Abstract

Fretting fatigue (FF) involves multiaxial and non-proportional stress states in addition to strong stress gradients, making crack propagation modeling challenging. In this work, we evaluate different models based on stress intensity factors (SIFs) to predict crack propagation kink angles under FF conditions, including both theoretical and artificial neural network (ANN) models. To train and test the ANN model, we collected FF data on Al alloys from six different sources in the literature, where the crack paths observed in the experiments were available. Testing data was also used to assess the performance of the theoretical models in predicting crack propagation kink angles. We demonstrate that the ANN model outperforms the theoretical models in predicting crack propagation kink angles. Finally, we compared the investigated models by simulating the crack propagation path under FF conditions using finite element modeling. The results indicate that, although the ANN model was more accurate in predicting the kink angle for a given crack configuration observed in the experiments, its performance was comparable to the theoretical models that account for non-proportional effects when estimating the crack propagation path. However, under high compressive states, the ANN model predicted more stable crack paths than the theoretical models.