Wavelet neural network application in helicopter gearbox gear fault diagnosis

Abstract

This paper focuses on developing an artificial neural network (ANN) model designed to assess the condition of helicopter gearboxes in the presence of specific faults namely, shaft unbalance and bearing inner race wear. Monitoring these faults is crucial, as they can lead to significant performance issues or failures if undetected. By analyzing various input indicators, the model diagnoses faults based on subtle patterns and anomalies within the gearbox’s vibration signals. Unlike conventional diagnostic methods, which often rely on fixed thresholds or rigid rule based systems, learning and recognition algorithms used in ANN models provide a highly adaptable, data-driven approach. This adaptability enables the model to refine its accuracy over time, learning from new data and updating its fault classification capabilities. A robust dataset is essential to ensure the model can accurately classify different types and severities of faults. During the training phase, the ANN model learns to associate specific input patterns with known fault states, enabling it to generalize this knowledge and detect faults in new, unseen data. Simulation results demonstrate the ANN model’s success in accurately identifying gearbox gear vibration faults and effectively distinguishing between healthy and faulty states, even in the presence of random noise disturbances. This approach offers a scalable and efficient solution with broad applicability across aerospace and other high-stakes fields.