Risk assessment of initial crack propagation in bearing steel based on finite element analysis and machine learning

This study analyzes the initial subsurface crack propagation in bearing steel by utilizing a 3D Voronoi finite element model to simulate the bearing steel's grain structure. Subsurface stress calculations validate the model. Stress intensity factors were computed to determine crack propagation as a function of initial crack orientation, length, and depth. A novel aspect of this study is the integration of FE analysis with an Artificial Neural Network for predictive modeling. A grid search method was employed for hyperparameter tuning, and ten-fold cross-validation was used to evaluate the ANN's performance, ensuring robust and accurate predictions. This hybrid approach enables the prediction of SIFs based on various load and crack parameters, facilitating a rapid assessment of crack propagation risk. The results provide valuable insights into the reliability and lifespan of bearing steel, contributing significantly to the field of bearing failure analysis.