A Fault Tolerant Approach to State Estimation and Failure Detection in Nonlinear Systems

A computationally feasible methodology will be presented for state estimation and failure identification in discrete-time nonlinear stochastic systems. The procedure involves using multiple hypothesis testing to isolate failures in both system inputs and outputs. The method is obtained by extending Friedland's results [1] for separate bias estimation and Caglayan's results [2] for simultaneous detection and estimation in linear systems to extended Kalman filtering. Although it is no longer possible to exactly factor out the no-fail estimation computations common to the hypothesis conditioned filters, suitable assumptions are made to obtain an estimator/detector structure similar to that of the linear case. The utility of the approach will be discussed in the context of a fault tolerant navigator for a terminal-configured vehicle simulation.