Abnormal Detection Gram (andgram): an Informative Frequency Band Selection Method Using Composite Index for Bearing Incipient Fault Diagnosis

Zhiwen Liu,Lei Wang,Yulin Jin,Hao Xu
DOI: https://doi.org/10.1016/j.ymssp.2024.112033
IF: 8.4
2025-01-01
Mechanical Systems and Signal Processing
Abstract:This paper presents a novel method called abnormal detection gram (Andgram) for bearing incipient fault diagnosis. A challenge in bearing incipient fault diagnosis is how to accurately identify the informative frequency band and extract the periodic transients caused by faults. The kurtosis, Gini index, reciprocal of smoothness index, averaged negentropy and harmonic-to-noise ratio are well-known fault information measures to select the informative frequency band by characterizing the impulsiveness and/or cyclostationarity of periodic transients. First, a numerical investigation is conducted to reveal the discriminability, or lack thereof, of these fault information measures for periodic transients in presence of sporadic impulses, harmonics and strong Gaussian white noise, respectively. Then, considering the complementarity of these fault information measures, a composite index named abnormal index is constructed by using variational autoencoder models. The proposed Andgram incorporates redundant second generation wavelet transform with the constructed abnormal index to accurately and robustly extract periodic transients under complex interference and strong Gaussian white noise. Finally, three simulation case studies and two applications to a laboratory dataset and a real-world wind farm dataset verify the effectiveness and robustness of the proposed method for bearing incipient fault diagnosis.
What problem does this paper attempt to address?