Unsupervised Pattern Classifier for Abnormality-Scaling of Vibration Features for Helicopter Gearbox Fault Diagnosis

A new unsupervised pattern classifier is introduced foron-line detection of abnormality in features of vibrationthat are used for fault diagnosis of helicopter gearboxes.This classifier compares vibration features with theirrespective normal values and assigns them a value in[0, 1] to reflect their degree of abnormality. Therefore,the salient feature of this classifier is that it does notrequire feature values associated with faulty cases toidentify abnormality. In order to cope with noise andchanges in the operating conditions, an adaptationalgorithm is incorporated that continually updates thenormal values of the features. The proposed classifieris tested using experimental vibration features obtainedfrom an OH-58A main rotor gearbox. The overallperformance of this classifier is then evaluated byintegrating the abnormality-scaled featares for detectionof faults. The fault detection results indicate that theperformance of this classifier is comparable to theleading unsupervised neural networks: Kohonen's Fea-ture Mapping and Adaptive Resonance Theory (ART2).This is significant considering that the independence ofthis classifier from fault-related features makes ituniquely suited to abnormality-scaling of vibrationfeatures for fault diagnosis.Keywords: Condition monitoring; Diagnostics; Gear-box; Vibration