Robust fault detection of linear systems using a computationally efficient set-membership method

In this paper, a computationally efficient set-membership method for robust fault detection of linear systems is proposed. The method computes an interval outer-approximation of the output of the system that is consistent with the model, the bounds on noise and disturbance, and the past measurements. If the output of the system does not belong to this interval, a fault is detected. To compute the output interval, we propose using support functions. Only two support functions for each output must be computed which results in a computationally efficient algorithm. Moreover, the method is trivially parallelizable. The method is demonstrated for fault detection of a hydraulic pitch actuator of a wind turbine. We show the effectiveness of the proposed method by comparing our results with two zonotope-based set-membership methods.