Soft Computing-Based FDI

Contents

Challenging design problems arise regularly in modern fault diagnosis systems. Unfortunately, the classic analytical techniques often cannot provide acceptable solutions to such difficult tasks. Indeed, as has already been mentioned, the design complexity of most observers for non-linear systems does not encourage engineers to apply those in practice. Another fact is that the application of observers is limited by the need for non-linear state-space models of the system being considered, which is usually a serious problem in complex industrial systems. This explains why most of the examples considered in the literature are devoted to simulated or laboratory systems, e.g., the known two- or three-tank system, the inverted pendulum, etc. [27, 96, 198]. Another serious difficulty is that there are examples for which fault directions are very similar to that of an unknown input. This may lead to a situation in which the effect of some faults is minimised and hence they may be impossible to detect. Other approaches that make use of the idea of an unknown input also inherit these drawbacks, e.g., the robust parity relation (see Section 2.3).