Online Learning Control of Automatic Transmission: Torque Control in Shifting Phase

Online learning control;Automatic transmission;Echo state Network;Recursive least square;621

공과대학 기계항공공학부 ;

University:서울대학교 대학원

关键词: Online learning control;    Automatic transmission;    Echo state Network;    Recursive least square;    621;   

Others  :  http://s-space.snu.ac.kr/bitstream/10371/141397/1/000000150480.pdf

美国|英语

来源: Seoul National University Open Repository

【 摘 要 】

Automotive transmission is an equipment to convert engine power to necessary speed of revolution and torque according to demands of a vehicle. Among the transmissions, the automatic transmission can change gear ratio automatically without manual control for driving. In the automatic transmission, a component which provides various gear ratio is a planetary gear consisting of sun, ring, carrier, planet gears. By restricting or releasing rotation of the sub-gears, a planetary gear can generate multiple gear ratio. In order to restrict or release the gears, clutches are used by hydraulic systems. How to engage the clutches on each gear is an important subject to decide performance of the transmission. The engagement control is important to satisfy desirable performance such as shift impact and time, while the transmission provides suitable rpm and torque. To move a clutch, a hydraulic system is used. However, the hydraulic system is influenced by fluid dynamics according to surrounding condition, and its characteristic determines control performance. So, previous engagement control of transmissions is a feed-forward method by pre-defined pressure profiles depending on vehicle state. Because of not only fluid characteristics, but also nonlinear time-variant properties such as wear of clutch surface and thermal deformation, the model-based control approaches cannot guarantee the stability of the system. Also, the previous control approach needs engineers to tune the optimal pressure profiles of each vehicle.Therefore, we propose online learning control based on Echo State Network on the clutch engagement so as to satisfy desired output torque of a transmission. Contrary to model-based control methods, online learning control does not require mathematical dynamics equations and keeps learning the system under control at the same time. These features are allowed to cope with time-variant systems such transmission control problem. In this paper, we are going to validate the online learning control algorithm on the transmission model made into Simulink, before applying online learning control to a real car.