Robust Detection and Identification of Simultaneous Sensor and Actuator Faults

In this work, we address the problem of fault detection, identification, and recovery (FDIR) for simultaneous actuator and sensor degradation. Prior studies have considered sensor degradation to some extent, but results for systems that do not have sensor redundancy while experiencing simultaneous actuator degradation are lacking. We present a novel method for robustly detecting and identifying sensor degradation in the presence of potential bounded actuator degradation modes under the influence of process and output noise. Subsequently, we develop novel theory for guaranteed forward reachability in the case of stochastic differential equations, as well as theory for robust fault detection for dynamical systems. Our approach enables in robust state estimation under sensor degradation, namely in the framework of zonotopic-Gaussian Kalman filters (ZGKF), providing an end-to-end FDIR framework for simultaneous sensor and actuator faults in a multi-agent setting. We apply our approach, which can be run in real time, to a realistic model of rigid-body satellite attitude dynamics in the presence gyroscope bias and gain changes, as well as changes in the thruster efficacy. We also present a second examples based on a multi-agent rover mission with limited periodic information sharing.