Optimal state estimation with failed sensor discrimination and identification

A new estimation scheme is presented that combines a fixed-gain Kalman filter for optimal state estimation with a prefilter that discriminates against failed sensors and identifies a failed sensor in real time. This new scheme has features characteristic of systems with triple-redundant sensing and voting, but with fewer sensors. It is tested on second- and third-order plants with dual-redundant measurements of the system states and is shown to out perform the stand-alone Kalman filter by a factor of two or more in terms of the rms estimation errors. Strategies for application to systems higher than third order are discussed.