Abstract
Failure diagnosis is important from two aspects. The first one is an improvement on system availability, and the other, which is much more important, the protection against disasters. From this point of view, many failure detection methods have been developed until now. Especially, a generalized-likelihood-ratio (GLR) method is well-known for its most rapid failure detection speed for the dynamic systems whose mathematical models are known.
But, in its use, it has a drawback such that we must assume a priori several adequate hypotheses which are expected to model the actual anomalies or failures. Step hypothesis which models the anomaly function adding to the system by a step one has often been used for its convenience. But, we do not always have such a situation where the step hypothesis works well.
Thus, we propose here one approach which makes the most use of the information about the state and the input of the system to be able to detect even the anomaly which is hard to be detected by the step hypothesis, and finally show by numerical examples that the proposed method is effective.
