A new autocorrelation-based strategy for multiple fault feature extraction from gearbox vibration signals

Abstract

A gearbox plays a key role in rotating machines, and its fault diagnosis is of great significance to reduce maintenance costs, unnecessary casualties and property losses. However, due to the concurrence of heavy Gaussian and non-Gaussian background noises, the multi-fault diagnosis of gearboxes becomes a challenging task in practice. Besides, the components of multiple faults may be easily coupled with each other, thus making the weak fault unnoticeable or even undetectable. To address these issues, a new strategy for multi-fault detection of gearboxes called LEASgram, along with the cyclic feature index (CFI) which is a quantitative indicator for evaluation of the impulsive fault components in the log envelope’s auto-spectrum (LEAS) of a signal, is proposed in this study. In the proposed method, a series of frequency sub-bands are required to be divided and the one with the largest CFI value would be selected as the optimal frequency band for further demodulation analysis. Considering the merits of autocorrelation, log envelope, and the moving average process, the main advantage of the proposed LEASgram is that it has significant capability of multi-fault identification and separation under Gaussian and non-Gaussian interferences, whether the fault-related resonance frequency bands are overlapped or not. Two simulated vibration signals and two experimental scenarios were applied for testing the performance of the proposed method. The results reveal that the proposed strategy is superior to other existing methods in separating and extracting the impulsive fault features, thus demonstrating its effectiveness for gearbox multi-fault diagnosis.