Fault prediction of thermal warpage in friction components of multi-disc wet clutches based on sequence decomposition and multi-scale feature extraction

Abstract

Predicting thermal warpage faults of multi-disc wet clutches is crucial for ensuring the reliability and safety of transmission system shifting. When the local thermal stress caused by the temperature rise of clutch friction components exceeds the threshold, thermal warpage faults will occur. Existing methods have difficulty solving the problems of feature redundancy and the disconnection between mechanism and data-driven approaches in temperature rise prediction. This paper proposes a thermal warpage fault prediction method based on sequence decomposition and multi-scale feature extraction, which integrates sequence reconstruction, deep feature extraction, dual-branch model prediction and mechanism model prior. The proposed method first separates non-stationary information and decomposes the sequence through sequence reconstruction. Then, an adaptive weight multi-time-scale convolutional neural network is used to extract local features and reduce information redundancy. Next, a dual-branch prediction module is designed for temperature rise prediction. Finally, the prior temperature distribution and fault threshold are obtained through a thermal warpage fault numerical model. Thermal warpage fault prediction is achieved by combining the numerical model with temperature rise prediction. Temperature rise data under typical working conditions are collected through multi-disc wet clutch sliding friction bench tests, and the model is trained with this data to realise thermal warpage fault prediction based on historical data. Comparative experiments confirm the proposed method’s superior temperature trend capture and accuracy. Its temperature rise prediction accuracy is improved by at least 15.38%, and the fault index prediction error does not exceed 3.5%. The effectiveness of core modules such as the dual-branch hybrid model and feature extraction is verified through ablation experiments.