Wavelet Analysis of Acoustic Signals for Gearbox Vibration Monitoring and Diagnostics

ALPHATECH has recently developed an integrated software environment for developing and testing prototype fault detection and identification (FDI) systems. It has been applied with excellent results (zero false and missed alarms, with 4% probability of deferral to the next decision time) to helicopter gearbox test-stand accelerometer data. The front-end of our environment extracts novel feature sets offered by the continuous wavelet transform (CWT), on which we train and evaluate a classical artificial neural network classifier. We propose to apply this approach to two forms of acoustic data collected in Phase 1: field data collected using an external acoustic sensor focused on the gearbox areas of an actual SH-60 helicopter, and synthetic data from an acoustic sensor in a controlled acoustic space containing loudspeakers driven by gearbox test-stand accelerometer signals plus recordings of helicopter rotor downwash and engine noise. We will compare the FDI results for the synthetic data with that previously obtained using the accelerometer signals directly, and determine dominant CWT features associated with the actual unfailed gearboxes. We will also develop the plan for Phase II, in which we will obtain and analyze sufficient data to demonstrate the reliability and robustness of the approach.