A stability-guaranteed learning framework for fault-tolerant tracking control of nonlinear systems: applying recurrent equilibrium networks

This paper proposes a novel stability-guaranteed learning framework to address the fault-tolerant tracking control problem for nonlinear discrete-time systems with unmeasurable states and actuator faults. A descriptor observer aims to provide estimates of the states and fault signals. Subsequently, a controller is developed based on the Recurrent Equilibrium Network with significant freedom of parameter design. Theoretical analysis is conducted to derive conditions for the closed-loop stability. Improvement of tracking performance is then conducted by maximising the reward function with the stability condition used as a basic constraint. Numerical results are performed to illustrate the methodology's efficacy.