UNSUPERVISED CONNECTIONIST NETWORK FOR FAULT DIAGNOSIS OF HELICOPTER GEARBOXES

A diagnostic method is introduced for helicopter gearboxes that uses knowledge of gearbox structure and characteristics of the features of vibration to define the influences of faults on features. To define the structural influences, the gearbox is represented by a lumped mass model and the root mean square value of vibration from this model is used to assign the influences. These structural influences are then converted to fuzzy variables to account for the approximate nature of the simplified gearbox model, and used as the connection weights of a connectionist network. Diagnosis in this Structure-Based Connectionist Network (SBCN) is performed by propagating the abnormal features obtained from an unsupervised pattern classifier through the connection weights of SBCN to obtain fault possibility values for each component in the gearbox. The performance of the SBCN is tested in application to an OH-58A helicopter gearbox. The results indicate a diagnostic accuracy rate of about 80%, which is remarkable considering that the SBCN performed diagnosis in the absence of any supervised training.