Data-driven Based Dual-Clutch Transmission Shift Control

Dual-Clutch Transmission (DCT) has been one of the most popular types of trans-missions these days. Due to the high efficiency and the no power interruption during shift process, the market share of DCT has been grown more and more.Compared with the traditional manual transmissions, the biggest difference of DCT in mechanical structure is that it has two clutches, and a clutch-to-clutch method of shift is applied in DCT. The shift process can be thought as two phases:the torque phase and the inertia phase. During the torque phase, the power transition of the engine is transferred from the current working clutch to the on-coming clutch. During the inertia phase, the oncoming clutch begins to engage and slip, and finally locked-up. The determination of the finish time of the torque phase and the engaging and disengaging speed of the clutches influence the shift quality greatly. Besides, the control of the power source of the transmission is what we have to consider, which also has an influence on the shift quality. For DCT the lack of torque converter and one-way clutch also makes the shift control even harder.To improve shift quality, different gear shift control methods of DCT have been discussed intensively in many researches, such as the calibration control method and PID control methods and so on. But the biggest limitation of current control method for the shift process is the ignorance of the vehicle power source, the engine, while considers only the transmission itself. Another limitation is that the close-loop control method we used is just for the inertia phase while for the torque phase only open-loop control method is used, what we have to point out here is that the exactness of finish time determination of the torque phase will also have influence on shift quality. To solve these two problems, in this paper we designed a controller based on data-driven predictive control method. A vehicle equipped with a6-speed DCT is modeled in AMESim and by using subspace identification method a input-output data based model is obtained. In order to obtain offset-free control for the control process, the predictor equation is refined into incremental inputs and outputs. In consideration of the vehicle physical characteristics, the inputs and outputs constraints are analyzed separately in the problem formulation. The contradictory requirements of shorter shift time and less shift jerk are included in the objective function. Furthermore, considering the controller is actually working in a realtime environment, Particle Swarm Optimization method is adopted to resolve the objective function. Finally, the controller is tested in the offline environment and xPC-dSPACE based online environment, simulations results under different driving condition show the effectiveness of the shift controller.