Seamless shifting of a two-speed dual clutch transmission for electric vehicles using machine learning

Abstract

One of the challenges facing widespread electric vehicle (EV) adoption is the short driving range. To address this challenge, the development of various EV transmissions is underway, but the shock during EV shifting is more noticeable than for gasoline engines. In order to achieve seamless gear shifting, the motor and clutch must be controlled in response to the torque transmitted through the clutch. This torque is theoretically proportional to the friction coefficient and the thrust force but is difficult to estimate in reality as it changes according to various factors. Therefore, this study applied deep reinforcement learning to automatically learn a gear shifting control method for the two-speed dual clutch transmission of an EV that adapts to the control target (seamless transition of output torque) through trial and error. A model for learning this control method was developed using the Actor/Critic network to explore the possibility of automatically designing motor torque and clutch control rules to achieve seamless gear shifting through repeated learning of the resulting output torque. The gear shifting results with and without the developed control method were then compared. It was found that control rules could be automatically designed to achieve seamless gear shifting by applying deep reinforcement learning to the transmission in this study. Furthermore, it was possible to control the motor torque and clutch to enable seamless gear shifting without directly monitoring the clutch transmission torque.