Friction performance prediction of automotive pads under operating conditions using attention-based CNN-BiLSTM deep learning framework

Abstract

In long-term operation, the gradual degradation process of automotive friction pads significantly affects the expected performance of mechanical equipment. In addition, the intrinsic correlations between friction properties and the multi-stage degradation process have been mostly ignored, leading to less accurate prediction of results under multifactorial influences on working conditions. In this paper, we propose a novel prediction method using the CNN-BiLSTM-Att model to overcome the problem. The model uses CNN to extract the friction features in the processed data, and combines with BiLSTM to evaluate the time series features hidden in the friction data. To improve the prediction accuracy, the attention mechanism is fed into the proposed model, which has the advantage of automatically assigning appropriate weights to the hidden layer states to distinguish the importance of different data features. Compared with other machine learning algorithms, the method has high prediction accuracy and can provide reference for braking.