ABSTRACT
This chapter discusses the measurement faults cause even more serious problems unless the R matrix is adapted in these cases. It investigates the adaptation methods for the measurement noise covariance matrix of the attitude filter. Fault tolerant attitude estimation focuses on the R-adaptation methods against the measurement faults in particular. Experimental attitude sensors or commercial off-the-shelf devices, for which it is difficult to get accurate noise characteristics, are used. The unscented Kalman filter works accurately when there is no fault in the measurement system. On the contrary, in case of a fault such as abnormal measurements, step-like changes or sudden shifts in the measurement channel etc. the filter deteriorates and the estimation outputs become faulty. The robustness of the filter is secured by scaling the measurement noise covariance matrix in case of fault. In this sense two different approaches may be used: Scaling by a single scale factor or scaling by a scale matrix built of multiple scale factors.
