Acoustic bearing fault diagnosis based on LCMV beamformer and MGIgram via an MEMS microphone array

Abstract

Acoustic signals have advantages over vibration signals, such as non-contact measurement and high spatial resolution, but are also easily affected by environmental noise, which leads to their insufficient exploration in the field of bearing fault diagnosis. To tackle this problem, this paper proposes a hybrid acoustic feature extraction technique for bearing fault diagnosis by combining the linearly constrained minimum variance (LCMV) beamforming with a fast filtering method guided by the median Gini index (MGI). First, a low-cost linear MEMS microphone array is employed to collect the acoustic signals of faulty bearings. Second, the LCMV beamforming algorithm is used to convert array signals into a one-dimensional signal, which can also suppress interference and enhance signals from the desired direction. Finally, the MGIgram is proposed to select the informative frequency band and extract weak fault-induced repetitive transients from heavy background noise. Experimental results verify the effectiveness of the proposed technique in bearing fault diagnosis and demonstrate its superior robustness compared to traditional methods such as the Kurtogram, Infogram, and GIgram.