Multi-wear characteristic parameter failure prediction of wet friction components based on data-mechanism hybrid drive

Abstract

Wet friction components are critical elements of transmission systems, playing a vital role in ensuring vehicle safety and operational stability. During clutch engagement, these components are susceptible to wear, and excessive wear can gradually degrade performance and ultimately lead to clutch failure. In this study, three failure-related parameters—the critical wear rate, the critical rate of change in copper concentration, and the critical rate of change in surface roughness—are selected to characterize the wear-induced damage behavior of wet friction components. Based on the experimentally determined critical wear rate, the material hardness H and wear coefficient K under different operating conditions are inversely identified, and an optimized Archard wear model is established. Subsequently, a Black Kite Algorithm–Convolutional Neural Network–Support Vector Machine (BKA-CNN-SVM) model is developed, with pressure p and rotational speed n as inputs, and failure time t along with the optimized K and H as outputs, to predict failure-related parameters. By integrating the predicted t, K, and H into the optimized Archard model, a hybrid data- and mechanism-driven framework is constructed for predicting wear-related parameters of wet friction components. The results demonstrate that the proposed BKA-CNN-SVM model achieves average relative errors of 4.56% and 3.11% in predicting K and H, respectively, with a maximum R² of 0.982. When applied within the hybrid-driven framework to solve wear parameters, the average relative error with respect to experimental data is 6.16%, with an R² of 0.997, indicating that the overall prediction accuracy is significantly superior to that of the unoptimized conventional Archard model. The proposed hybrid framework enables high-accuracy prediction of wear-induced failure and can provide valuable guidance for the safe operation of transmission systems under low- to medium-load real-vehicle conditions, demonstrating substantial engineering application potential.