Case-Based Fault Detection - A Method for Parallel Processing

Abstract

Fault detection in technical systems is a difficult task, since there is only poor prior information about how installation and row material disorders or operator errors affect the process working. However, technical systems as artefacts, have well-known structures and well known components' behaviour - from the designer and the producer respectively. Disorders occur at some components and the effects spread along the flux, so. the values from sensors fonn patterns in time. The paper presents a method to generate cases as pairs of the event and the intelligent encoded observation, for multiple phase processes. The human expert knowledge about normal and faulty installation behaviour is structured using qualitative models for normal running, and semi-qualitative modelling of the functional component's behaviour for some faults. The representation of the cases enables parallel hypothesis generation. A distributed monitoring and fault detection application for the hydraulic installation of a rolling mill plant is shortly presented.