Failure detection via recursive estimation for a class of semi-Markov switching systems

The authors apply the recursive state estimation algorithm for dynamic systems whose state model experiences jumps according to a sojourn-time-dependent Markov (STDM) chain to the problem of failure detection. The algorithm, which is of the interacting-multiple-model (IMM) type, uses noisy state observations. Two simulation examples are presented. The first indicates that the use of the STDM-based IMM estimator can give a substantial improvement in state estimation over a Markov-based IMM. The second example shows that for the particular system under consideration, the STDM-based IMM estimator, which is a hypothesis-merging technique, compares favorably in terms of the probability of error to the detection-estimation-algorithm-based estimator, which discards the unlikely parameter history hypothesis