Sequential detection and estimation of soft failures in linear systems

An algorithm is reported for soft failure detection and estimation which has very modest computational requirements. Soft failures manifest themselves as deviations from the expected statistical behaviour of system inputs. The detection scheme utilizes only one Kalman filter and requires the implementation of a Wald Sequential Detector (WSD). The algorithm consists of applying the Sequential Probability Ratio Test (SPRT) to the nonwhite Kalman filter innovations for rapid failure detection. The Average Sample Number (ASN) is used for estimating failed parameters. Some computer simulation results are presented.