Period-refined CYCBD using time synchronous averaging for the feature extraction of bearing fault under heavy noise

• Published in: Structural health monitoring 2024-03, Vol.23 (2), p.1071-1088
• Main Authors: Miao, Yonghao, Shi, Huifang, Li, Chenhui, Hua, Jiadong, Lin, Jing
• Format: Article
• Language: English
• ISSN: 1475-9217; 1741-3168
• DOI: 10.1177/14759217231181514

Deconvolution methods have been widely used in machinery fault diagnosis. However, their application would be confined due to the heavy noise and complex interference since the fault feature in the measured signal becomes rather weak. Time synchronous averaging (TSA) can enhance the periodic components and suppress the others by the comb filter function. And in the iteration process of the deconvolution methods, the filtered signal after each iteration can be further processed using TSA, and the time delay with maximum Gini index value is refined as the iterative period for the next iteration. Benefitting from these advantages, a period-refined maximum second-order cyclostationarity blind deconvolution (PRCYCBD) using TSA is proposed for the weak fault detection of rolling element bearings (REBs) in this paper. Firstly, without any prior knowledge, the proposed method which can estimate the period more accurately is more suitable for the weak fault detection of REBs, especially incipient fault. Secondly, TSA is firstly applied to estimate the iterative period rather than just depending on the Signal Noise Ratio (SNR) of the filtered signal in the iterative process . Furthermore, the new improvement frame can be expanded to other deconvolution methods using iterative algorithms, especially under heavy noise. Finally, a simulation with a slight bearing fault as well as two real experimental data including the vibration signal with the wind turbine bearing fault and the acoustical signal with the locomotive wheel bearing fault is used to verify the superiority of the proposed PRCYCBD compared with the traditional minimum entropy deconvolution and the traditional autocorrelation-improved cyclostationarity blind deconvolution.