Sensitivity analysis and prediction of hoop stress in ceramic bearing washer based on a deep residual network surrogate model

To prevent unpredictable cracking failures induced by the thermal expansion mismatch of all-ceramic bearing washers at high temperatures, this study proposes a deep residual network (ResNet) surrogate model for predicting hoop stress. Through systematic simulations, this research reveals the underlying evolution mechanisms of hoop stress, thereby establishing fit precision design standards for ceramic bearings operating in high-temperature environments. First, a thermo-mechanical coupled finite element model of the bearing washers is developed and rigorously validated via high-temperature strain measurement experiments. Subsequently, this validated model is utilized to construct a high-fidelity dataset, which effectively trains the ResNet surrogate model. Finally, the trained model is employed to conduct global sensitivity analyses and comprehensive hoop stress predictions. The findings demonstrate that mechanical fit and temperature emerge as the dominant factors governing hoop stress, exhibiting a strong coupled amplification effect. The equivalent outer radius demonstrates a negative sensitivity; thus, increasing this dimension effectively suppresses the growth of hoop stress. Additionally, the post-assembly contact constraints between the raceway and the balls inhibit the maximum hoop stress within the bearing washers. Furthermore, machining errors introduce profound uncertainties into the hoop stress distribution. Consequently, to ensure structural reliability, the assembly clearance must be appropriately enlarged as the operating temperature increases.