Early-stage fault isolation based on frequency response fitted by small-size samples for cryogenic cold compressors with active magnetic bearings

Abstract

A model-based method for fault detection and early-stage isolation, applicable when unfaulty conditions can be identified only by a reduced number of trials (even only one), is presented. The basic idea is to model analytically the uncertainty of the unfaulty frequency response and express the fault condition in terms of the noise power variance. A preliminary fault isolation is carried out by sensitivity analysis in order to identify the most influencing model parameters and assess their influence on the estimated noise. Then, during maintenance tests, the noise power is checked to detect the faulty condition. This technique is conceived to check the quality of a critical component in an experimental installation (fault detection and early-stage isolation), as well as to detect its faulty dynamic behaviors over a long horizon maintenance test campaign (condition monitoring). The method was applied to four cold compressors with active magnetic bearings at CERN by proving to be able to detect an actual faulty condition in one of such compressors.