Fault Detection Filter Design and Optimization for Switched Systems with All Modes Unstable

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Fault Detection and Diagnosis

Background:

• Switched systems with unstable subsystems pose significant challenges for fault detection.
• Traditional fault detection filters often struggle with complex system dynamics and transient performance.

Purpose of the Study:

• To design an intelligent switched fault detection filter for unstable switched systems.
• To enhance filter performance using a hybrid approach combining dynamics-based and learning-based methods.
• To guarantee system stability and prescribed attenuation performance.

Main Methods:

• Mode-dependent average dwell time (MDADT) for generating time-dependent switching signals.
• A hybrid filter combining dynamics-based (using linear matrix inequalities) and learning-based (deep reinforcement learning, actor-critic) components.
• Multiple Lyapunov function (MLF) method for stability and performance guarantees.
• Deep deterministic policy gradient algorithm and nonfragile control for robustness.

Main Results:

• The proposed switched fault detection filter effectively generates residual signals.
• Stability and prescribed attenuation performance are guaranteed by MDADT and MLF methods.
• Deep reinforcement learning improves the transient performance of the filter.
• Simulation results demonstrate the effectiveness of the proposed intelligent fault detection method.

Conclusions:

• The developed intelligent switched fault detection filter is effective for systems with unstable subsystems.
• The hybrid filter design, incorporating deep reinforcement learning, enhances fault detection capabilities.
• The method provides a robust solution for fault detection in complex switched systems.