Reinforcement learning-based active fault diagnosis with tracking performance guaranteed

Due to the consideration of control performance and the uncertainty of the dynamic characteristics of nonlinear systems, designing the auxiliary signal for active fault diagnosis presents significant challenges. This paper presents a novel data-driven approach for auxiliary signal design in the active fault diagnosis of nonlinear systems while ensuring guaranteed control performance. Specifically, we introduce a double actor-critic network to generate tracking and diagnostic signals, respectively. Subsequently, a two-objective optimization method based on deep reinforcement learning is proposed to address the tradeoff between tracking performance and fault diagnosis. Finally, the effectiveness of this method is verified through a cart-pole system with stochastic noise.