An interval-based approach for fault isolation and identification in continuous dynamic systems

Diagnosis of faults is a very important task because a fault can lead to a reduction in performance, or even to break-downs or catastrophes. In this paper, a diagnosis system, which takes into account the uncertainties in the model and measurements by means of an interval scheme, is shown. It states the fault detection, fault isolation and fault identification problems as a constraint satisfaction problem (CSP) with continuous domain. The fault isolation scheme which consists of hypothesis generation and refinement uses the fault detection results, and the quantitative fault identification scheme refines the fault hypothesis test and estimates the fault magnitude. As a difference to more traditional approaches, in this paper we use overconstrained Subsystems of the model, ARRs, to tackle the fault identification problem in smaller subproblems, and for the fault identification, instead of the least squares estimation, the interval-based consistency techniques prune the initial domains of the parameters associated with the fault hypothesis. In order to illustrate the different steps of the proposed approach, an application example composed by a well known hydraulic plant is presented.