Detection, isolation, and identification techniques for noisy degradation in linear discrete time systems

This paper summarizes some current and new results pertinent to the detection, isolation, and identification of noisy degradation in linear discrete time systems. Both noisy sensor degradation and process noise degradation can be treated. This problem is disjoint from the problem of state bias shifts. Simple algorithms discussed are based on the mean, covariance, and correlation properties of the measurement residuals. More structured algorithms that afford isolation and identification as well as detection can be based upon the ideas of sequential probability ratio and generalized log likelihood ratio hypothesis testing. Some preliminary computer simulation results are presented.