Clutch Pressure Plate Temperature Prediction Based on Bi-LSTM and Migration Learning

Abstract

Clutch pressure plate temperature prediction is crucial for the structural design and performance evaluation of the clutch. However, due to the complexity of the clutch structure and the non-linear characteristics of temperature changes, accurate temperature prediction of the pressure plate has always been a difficult task, especially when considering cost factors. Aiming at this problem, this paper proposes a pressure plate temperature prediction method based on Bi-directional Long Short-Term Memory (Bi-LSTM) and transfer learning. First, the actual temperature data of the pressure plate under different experimental conditions is collected to establish a Bi-LSTM neural network temperature model, and then, a migration learning method is introduced to migrate the temperature experimental data to obtain a migration model. The migration model is finally applied to predict the pressure plate temperature using three samples and validated by the test. The results show that for the temperature prediction of the same type of pressure plate under different experimental conditions, the MSE (Mean Squared Error) of the approach is 7.08 °C, the R2 (R-squared) is 0.90, the maximum error is 8.47 °C, and the maximum relative error is 3.14%. For the temperature prediction of different types of pressure plates, the MSE of the approach is 3.64 °C, the R2 is 0.97, the maximum error is 5.94 °C, and the maximum relative error is 1.78%. It shows that the proposed approach achieves high-precision prediction of the clutch pressure plate temperature in the case of small samples, which is difficult to achieve with previous methods. The proposed approach can be used for the temperature prediction of the other clutches of models and working conditions and has broad application prospects.

Keywords

• temperature
• transfer learning
• neural network
• small sample