A Fault-Tolerant Estimator for Redundant Systems

This paper proposes a digital estimator for redundant systems which is superior to Kalman Filtering if a failure is present and reduces to Kalman Filtering if no failure is present. Fault tolerant estimation is achieved by defining the non-stationary weighting matrix associated with the nominal least squares estimator (Kalman filter) as a continuous ous nonlinear function of the measurements. Despite the nonlinear character of the failure detection and isolation feature, the estimator equations have closed form and hence require no iterative computations ions or approximations for implementation.