Auxiliary Input Design for Active Fault Detection via Deep Reinforcement Learning

The paper addresses a novel algorithm of active fault detection and control for closed-loop system with stochastic noise, where an auxiliary input signal generator is trained by deep reinforcement learning algorithm. The input signal consists of optimal control input and auxiliary input. In order to obtain the optimal control policy, a linear quadratic tracker is designed by using the model-based policy iterative algorithm. The problem of auxiliary input signal design is transformed into an optimization problem with the aim of minimizing a discounted cost, which is a trade-off between fault detection and control performance. A simulation example is carried out to demonstrate the effectiveness of the proposed method.