Temperature prediction of multi-plate wet clutch by artificial neural network with heat transfer differential equation

This study proposes a novel methodology for predicting the temperature of a multi-plate wet clutch by integrating artificial neural networks with heat transfer differential equations. Traditional approaches for clutch temperature prediction, such as empirical models and finite element methods, are often limited by high computational costs and simplified assumptions. In contrast, this research combines physical laws with neural networks to enhance prediction accuracy. Using experimental data obtained from a wet clutch, LSTM-based neural network was trained to estimate convective heat transfer coefficients dynamically, incorporating factors such as friction plate internal and external diameter, rotational speed, lubricant flow rate, inlet lubricant temperature, and groove type. The predicted coefficients were then applied to solve the heat transfer differential equation, ensuring the physical validity of the temperature predictions. Comparative evaluations demonstrated that the proposed neural network model with heat transfer integration outperformed existing empirical and neural network-only models, achieving a 69.68% reduction in MSE and a 64.27% reduction in RMSE compared to the empirical model.