A weak fault feature extraction of rolling element bearing based on attenuated cosine dictionaries and sparse feature sign search. (February 2020)

• Record Type: Journal Article
• Authors: Zhou, Haoxuan; Li, Hua; Liu, Tao; Chen, Qing
• Is Part Of: ISA transactions. Volume 97(2020)
• Page Start: 143
• Page End: 154
• Publication Date: 2020-02
• Subjects: Sparse representation theory -- Basis pursuit -- Attenuated cosine dictionary -- Bearing fault feature extraction -- Feature Sign Search; Engineering instruments -- Periodicals; Engineering instruments; Periodicals; Electronic journals; 629.805
• DOI: 10.1016/j.isatra.2019.08.013
• Languages: English
• ISSNs: 0019-0578

Abstract

Abstract: The time domain signal of bearing pitting/spalling fault always presents shock and modulation, and it is often submerged by strong noises, especially in the early stage. the conventional fault feature extraction method may have insufficient feature extraction accuracy, and even in some extreme cases, the fault feature frequency cannot be extracted because of the strong noise interference. Aiming at overcoming the noise interference problem encountered in this kind of weak fault feature extraction, a novel weak fault feature extraction algorithm termed as ACFSS of rolling bearing is proposed. The ACFSS is based on an overcomplete dictionary (or overcomplete atomic library) of Attenuated Cosines(AC) basis, which is highly matched to the bearing fault waveforms, and an improved Basis Pursuit algorithm with Feature Sign Search(FSS) is introduced into the ACFSS to improve the calculating speed. In order to select the suitable parameters of the attenuated cosine dictionary, some methods such as peak resonance frequency (PRF), power variation peak (PVK), time shift parameter (TSP), etc. are introduced. These parameters span the sparse overcomplete dictionary. Finally, the bearing fault data of Case-Western University and full life accelerated IMS bearing data are utilized to verify the validation of ACFSS. Compared with the ordinary envelope spectrum analysis(ESA) method /the ordinary Basis Pursuit Denoising(BPDN) method/ Wavelet package transform(WPT) Kurtogram method and Empirical Mode Decomposition(EMD) combining Singular Value Decomposition(SVD) method, The experiment show that the proposed method are more redundant and robust when facing strong noise interference, and it can be used to extract the weak fault feature frequency efficiently and accurately. Highlights: A novel sparse dictionary based on attenuated cosine basis is proposed. Feature Sign Search algorithm is introduced to calculate the sparse coefficient. A new algorithm ACFSS is proposed for weak fault feature extraction of rolling element bearing. Experiments demonstrate the effectiveness of the proposed method.