Off-policy Q-learning-based Output Feedback Fault-tolerant Tracking Control of Industrial Processes

This is the authors' abstract. We don't add key points for this paper.

In this paper, a data-driven Q-learning output feedback algorithm independent of system parameter information is proposed to solve the control problem of industrial processes with actuator faults. Firstly, an extended model is obtained by introducing tracking error into system state and output respectively. Secondly, the Bellman equation and GARE equation are acquired in the process of constructing the performance index and analyzing its relationship with value function. Since the solution of GARE equation requires knowing the system matrix information, so the Q-function is then described and an algorithm combining off-policy Q-learning and Kronecker product is used to determine the optimal controller with measurable external signals only. And the algorithm is proved to be unbiased. Finally, simulation experiments on the injection molding process verify the effectiveness of the proposed algorithm.