Non-Invasive Detection of CR-46 AFT Gearbox Faults Using Digital Pattern Recognition and Classification Techniques

摘要

Currently, the United States Navy performs routine intrusive maintenance on CH-46 helicopter gearboxes in order to diagnose and correct possible fault condition (incipient fault) which could eventually lead to gearbox failure. This type of preventative maintenance is costly and it decreases mission readiness by temporarily grounding usable helicopter. Non-invasive detection of these fault conditions would save tine and prove cost-effective in both manpower and materials. This research deals with the development of a non-invasive fault detector through a combination of digital signal processing and artificial neural network (ANN) technology. The detector will classify incipient faults based on real-tine vibration data taken from the gearbox itself. Neural networks are systems of interconnected units that are trained to compute a specific output as a non-linear function of their inputs. For sons tine the United States Navy has been interested in the use of artificial neural networks in monitoring the health of helicopter gearboxes. In order to determine the detection sensitivity of this method in comparison with traditional invasive methods, the USN funded Westland Helicopters Ltd to conduct a series of CH-46 gearbox rig tests. In these tests, the gearbox was seeded with nine different fault conditions. This seeded fault testing provided the vibration data necessary to develop and test the feasibility of en artificial neural network for fault classification. This research deals with the formation of the pattern vectors to be used in the neural network classifier, the construction of the classification network, and an analysis of results.