ALTERNATIVE MODELS IN FAULT DETECTION - MONTE CARLO APPROACH

摘要

State estimation problem together with alternative models approach is used to solve the problem of fault detection. Interesting idea - based on [1] is to represent all conditional probability density functions (c.p.d.f.) as a set of random samples. Usual approach is to represent c.p.d.f. as a function over the state space. Using large number of samples an equivalent representation of c.p.d.f. is obtained. From these samples the estimates of moments like mean and covariances can be obtained. In [1] bootstrap filter is described for updating these samples for discrete time system. In this way any nonlinearity in the system and non-normality of noise can be handled. The contribution of this article is to use this approach to fault detection. It is obvious, that information about a fault is obtained in the future data only. So it is necessary to create the tree of alternative models which describe the history of state development. In this way it is possible to compute the probability of the fault models based on the future data (but in real time only with some delay).