Gearbox fault diagnosis under fluctuating load conditions with independent angular re‐sampling technique, continuous wavelet transform and multilayer perceptron neural network

Abstract

Most research efforts in gearbox fault diagnosis thus far have focused on diagnosing gearbox faults under stationary conditions. Efforts in diagnosing gearbox faults under non-stationary conditions have mostly involved an analysis of gearbox vibration signals under the speed-up or run-down processes. This paper attempts to diagnose faults in a single stage spur gearbox under non stationary conditions arising from fluctuating loads at the output of gearbox. The vibration signal corresponding to each independent revolution is synchronized from the revolution point of view by converting into the angular domain. This is accomplished experimentally by a simple process referred to as the independent angular re-sampling (IAR) technique. The IAR technique is accomplished by employing a multiple pulse tachometer arrangement. Through the IAR process, non-stationary signals in the time domain are converted into quasi-stationary signals in the angular domain. The angular domain signals, each representing one revolution of the gearbox drive shaft, are then decomposed with continuous wavelet transform. Optimal scales are identified based on superior energy-Shannon's entropy ratio of continuous wavelet coefficients (CWCs). The classification accuracy of a multilayer perceptron neural network is compared when CWCs from all scales and when CWCs from the optimal scales are fed to the neural network.