Automated Functional Development for Automatic Transmissions Using Deep Reinforcement Learning
- PDF / 1,754,444 Bytes
- 6 Pages / 595.276 x 790.866 pts Page_size
- 27 Downloads / 307 Views
tomated Functional Development for Automatic Transmissions Using Deep Reinforcement Learning A current project at Mercedes-Benz AG addresses the question of whether artificial intelligence methods are capable of learning how to perform gearshifting in an automatic transmission. By the use of deep reinforcement learning, new control concepts can be generated and optimization potentials of existing control concepts can be identified, thus enabling function developers to be supported by artificial intelligence in the development process.
8
A U THOR S
Katja Deuschl works at Mercedes-Benz AG in the Division of eDrive D evelopment, Innovations eDrive Services, in Stuttgart (Germany).
Thomas Doster works at Mercedes-Benz AG in the Division of eDrive Development, Functional Development for C ontrollability of Automatic Transmissions, in Stuttgart (Germany).
Dr. Gerd Gaiselmann works at Mercedes-Benz AG in the Division of Group Research, Artificial Intelligence Research, in Sindelfingen (Germany).
© Daimler
DEEP REINFORCEMENT LEARNING
Latest challenges in automatic transmissions are causing transmission experts, in cooperation with pre-development and research groups, to consider unconventional solutions. The experts raised the question of how Artificial Intelligence (AI) can be used to support engineers in developing the optimal control of a gearshift. The aim of the current project carried out at Mercedes-Benz AG is to use Deep Reinforcement Learning (DRL) methods to induce a Software (SW) agent to find a functional control ATZ electronics worldwide 09|2020
program for the demanding gearshifts in the automatic transmission from Mercedes-Benz. CHALLENGES IN FUNCTIONAL DEVELOPMENT
A modern automatic transmission is a complex overall system in which various subsystems such as hydraulics, electronics, actuators, sensors, mechanics and the tribology of friction discs in the shift elements interact with each other. During a simple gear change in the 9G-Tronic automatic transmission,
Stefan Studer works at Mercedes-Benz AG in the Division of Group Research, Artificial Intelligence Research, in Sindelfingen (Germany).
two shift elements are involved. In order to set a new gear ratio in the gearbox, one shift element that has previously transmitted torque must be opened and another previously opened shift element must be closed. Via the control currents of the valves, the control unit adjusts the pressure on the friction discs and thus the torque to be transmitted [1]. The challenge is to achieve comfortable gearshifts reliably, and which do not create a noticeable jolt in the vehicle (time derivate of acceleration). In addition, mechanical and thermal loads must remain within the specified limits
9
C OVER STORY C ontrol Technolo gy
FIGURE 1 Automatic transmission 9G-Tronic (© Daimler)
and meet the criteria with regard to duration and agility. In order to achieve this, a modelbased software is created that controls or actuates the transfer of the transmitting torque from one shift element to the other as
Data Loading...
