Robust estimation in fault detection

Modeling errors present a significant and difficult challenge in the design of analytic fault detection mechanisms. The authors discuss the sensitivity to model uncertainty of estimator-based failure detection techniques. In particular, they discuss desired statistical properties for the decision variable, and why these characteristics are difficult to achieve in situations involving significant uncertainty in the noise, fault, or plant dynamic modeling assumptions. This discussion motivates the use of robust estimation techniques in failure detection. An aircraft example is presented to illustrate the effect of modeling error on the failure detection performance of detection test designs based on a Kalman filter and an H_∞/μ estimator