Classifying helicopter gearbox faults using a normalized energy metric

A normalized energy metric is used to classify seeded faults of the OH-58A main transmission. This gearbox comprises a two-stage transmission with an overall reduction of 17.44:1. Loaded gearbox test runs are used to evaluate the sensitivity of a non-stationary fault metric for early fault detection and classification. The non-stationary fault metric consists of a simple normalized energy index developed to account for a redistribution of sideband energy of the dominant mesh frequency and its harmonics in the presence of actual gearbox faults. This index is used to qualitatively assess the presence, type and location of gearbox faults. In this work, elements of the normalized energy metric are assembled into a feature vector to serve as input into a self-organizing Kohonen neural network classifier. This classifier maps vibration features onto a two-dimensional grid. A feedforward back propagation neural network is then used to classify different faults according to how they cluster on the two-dimensional self-organizing map. Gearbox faults of OH-58A main transmission considered in this study include sun gear spalling and spiral bevel gear scoring. Results from the classification suggest that the normalized energy metric is reasonably robust against false alarms for certain geartrain faults.