Helicopter gearbox fault detection: a neural network based approach

One of the most dangerous problems that can occur in both military and civilian helicopters is the failure of the main gearbox. Currently, the principal method of controlling gearbox failure is to regularly overhaul the complete system. This paper considers the feasibility of using a neural network to perform fault detection on vibration measurements given by accelerometer data. The details and results obtained from studying the neural network approach are presented. Some of the elementary underlying physics will be discussed along with the preprocessing necessary for analysis. Several networks were investigated for detection and classification of the gearbox faults. The performance of each network will be presented. Finally, the network weights will be related back to the underlying physics of the problem.