A robust failure detection and isolation algorithm

A robust failure detection and isolation (RFDI) algorithm for linear dynamic systems that is insensitive to failure mode, noise, and plant model uncertainties is presented. For robustness to failure mode uncertainties, the algorithm assumes that the failure is the output of a shaping filter, or specifically, a Gauss-Markov model. Such a model embraces a large class of failures. For robustness to noise and plant model uncertainties, the algorithm relies on either a robust risk sensitive (exponential Gaussian) or a robust H_∞ filter for the synthesis of failure residuals. It is shown that the algorithm is a generalization of likelihood ratio tests for failure detection and isolation in dynamic systems. Numerical results are presented for an underwater vehicle application, demonstrating that the RFDI algorithm can provide good performance over the desired range of model uncertainties. By contrast, the performance of the nominally optimal likelihood ratio test can considerably degrade in the presence of the same uncertainties