Clamping Force Estimation Method for Electro-Mechanical Brakes Based on CNN-BiLSTM

The electro-mechanical brake (EMB), as a brake-by-wire system that generates clamping force through the torque motor and mechanical mechanism, features compact structure, fast response, and high control accuracy. However, due to the strong nonlinear characteristics of EMB and disturbances such as friction, the clamping force in EMB exhibits a hysteresis effect, making it difficult to accurately control the clamping force directly based on the motor rotation angle signal. To address this issue, a neural network learning mechanism is introduced into the clamping force estimation of EMB, and a low-cost estimation method based on convolutional neural network and bidirectional long short term (CNN-BiLSTM) is proposed. First, through the analysis of hysteresis effects, it is concluded that the EMB clamping force is related to the current motor rotation angle, historical motor rotation angles, and historical clamping force outputs. Subsequently, by integrating the bidirectional learning capability of BiLSTM with the strong feature extraction capability of CNN, the CNN and BiLSTM are combined to reduce the interference of redundant input information on prediction results, thereby effectively improving the accuracy of EMB clamping force estimation. Finally, under given motor rotation angle reversal conditions, experimental tests demonstrate that the clamping force estimation model based on CNN-BiLSTM proposed in this paper has better estimation accuracy compared with LSTM and BiLSTM.