Abstract: The control accuracy of the clutch torque has significant impact on the control performance of the dual clutch transmissions, where the clutch torque is mainly affected by the changes of friction coefficient. To improve the adaptability of starting control to clutch friction coefficients, an intelligent control method, which can adapt to different friction coefficients, is proposed by integrating the advantages of model-based and model-free control. Firstly, the dynamic model of dual clutch transmission system is established. The clutch desired speed during starting process is optimized by pseudo-spectral method. Afterwards, an adaptive control strategy of speed closed loop for the starting process is developed. The clutch desired speed is obtained in real time based on back propagation neural network. To track the clutch desired speed, the sliding mode controller is designed based on the estimated clutch torque, which is generated by radial basis function neural network. Finally, simulation and real-vehicle experiments are carried out to verify the proposed method. The results show that satisfied starting quality is achieved under the complex condition of variable starting intentions, unknown friction coefficients, and disturbances. Compared with model-free control method, the adaptive sliding mode control has superior adaptability and robustness, and the adaptive control for friction coefficient is realized.

Key words: dual clutch transmission, starting process, friction coefficient, neural network, adaptive control