Fault diagnosis and vibration control of flexible satellite via boundary observer and Bayesian deep learning

Abstract

This paper presents a reliable active fault-tolerant control (AFTC) strategy for vibration control in flexible satellites. The proposed AFTC framework integrates three key modules: fault detection via a boundary observer, fault identification using Bayesian deep learning, and controller reconfiguration. A boundary observer is developed to generate residual signals sensitive to actuator faults. The residuals are analyzed for fault detection, while Bayesian neural networks, enhanced with Monte Carlo dropout, are employed for fault identification with uncertainty quantification. Based on the identified fault information, the vibration suppression controller is reconfigured to maintain system performance. Case studies involving multiple actuator fault scenarios validate the effectiveness and robustness of the proposed method in achieving reliable fault diagnosis and vibration mitigation.