Fault Detection for Reaction Wheels Using Convergent Zonotopic Thresholds with Experimental Validation

This paper addresses the challenge of model-based fault detection for satellite reaction wheels, which are critical actuators in attitude control systems. A fault detection framework is developed by combining a robust fault detection observer (FDO) with zonotope-based residual evaluation. The FDO is designed to be input-to-state stable and achieves an L_∞-gain performance, ensuring robustness against unknown-but-bounded disturbances and measurement noise. Both the observer gain and the parameters for zonotope order reduction are computed offline. The convergence and boundedness of the dynamically computed thresholds are rigorously proven. The methodology, from the experimental identification of the nonlinear friction model to the fault detection strategy, is successfully validated on a reaction wheel experimental platform. Experimental results demonstrate that the proposed zonotope–based method achieves earlier fault detection and higher sensitivity compared to both an extended interval observer and an ellipsoidal set–membership method.