Research on Failure Prediction Model for Wet Friction Components Based on GWO-TCN-BiGRU-Attention

As a key component in power transmission systems of mechanical equipment, wet friction components play a crucial role in ensuring operational safety and stability. In most cases, failure of these components results from coupled effects under multiple conditions and varying loads, with wear failure and thermal failure being the primary modes. Therefore, this study proposes three critical failure indicators: the rate of change in surface roughness (Rs), the critical radial temperature difference (Tr), and the critical circumferential temperature difference (Tθ). The failure datasets for these indicators are augmented using the Synthetic Minority Oversampling Technique (SMOTE) algorithm. Subsequently, by integrating thermal field failure features with critical surface roughness data, a failure prediction model is developed based on a Grey Wolf Optimizer-Temporal Convolutional Network-Bidirectional Gated Recurrent Unit-Attention (GWO-TCN-BiGRU-Attention) neural network architecture. Finally, the effects of sample size, rotational speed, and pressure on the failure indicators are also examined. The results show that the mean relative errors for Rs, Tr, and Tθare 2.78%, 3.89%, and 2.24%, respectively. Compared to other models, the proposed approach demonstrates superior predictive performance in failure forecasting of wet friction components.