Fault diagnosis in industry using sensor readings and case-based reasoning

Fault diagnosis of industrial equipments becomes increasingly important for improving the quality of manufacturing and reducing the cost for product testing. Developing a fast and reliable diagnosis system presents a challenging issue in many complex industrial scenarios. The major difficulties therein arise from contaminated sensor readings caused by heavy background noise as well as the unavailability of experienced technicians for support. In this paper we propose a novel method for diagnosis of faults by means of case-based reasoning and signal processing. The received sensor signals are processed by wavelet analysis to filter out noise and at the same time to extract a group of related features that constitutes a reduced representation of the original signal. The derived feature vector is then forwarded to a classification component that uses case-based reasoning to recommend a fault class for the probe case. This recommendation is based on previously classified cases in a case library. Case-based diagnosis has attractive properties in that it enables reuse of past experiences whereas it imposes no demand on the size of the case base. The proposed approach has been applied to fault diagnosis of industrial robots at ABB Robotics and the results of experiments are very promising.