Novelty detection for machinery condition monitoring

This paper is concerned with novelty analysis for fault detection in machinery. The detection procedure employed here uses novelty detection based on two approaches: an auto-associative neural network and kernel density estimation. The methodology is illustrated on the detection of local tooth faults in pseudo-experimental gearbox vibration data. The study shows the possibility of automatic signalling of failure. The method can be extended to any data type representing normal and abnormal conditions.