Improving fault detection abilities of extended Kalman filters by covariance matrices adjustment

The problem of model-based fault detection is studied with application of the Kalman filter for residual generation. The filter has two important incoming parameters, the state noise and the output noise covariance matrices, which tuning is analyzed in order to optimize the fault detection performance. The problem is formulated through an appropriate optimization criteria and applied to the oscillatory failure case detection in aircraft control surfaces. The results of simulation illustrate efficiency of the proposed technique.