A novel feature extraction method for roller bearing using sparse decomposition based on self-Adaptive complete dictionary

Sparse decomposition based on complete dictionary can effectively extract impulse features from weak fault signals. However, compared with the over-complete dictionary, the complete dictionary no longer has redundancy features, and its robustness is reduced, which makes it difficult for sparse signals to extract fault features under weak faults. To overcome this problem, a fault diagnosis method using adaptive complete dictionary via sparse signal is proposed. In particular, in order to improve the adaptability of a complete dictionary, we introduce an adaptive Q-factor wavelet transform (TQWT) algorithm to extract atoms. In the process of extracting atoms, according to the different oscillation characteristics of different Q factors, the adaptive TQWT algorithm is used to extract the atoms which accord with the vibration characteristics of faults. In the process of dictionary constructing, the atom can be extended to a complete dictionary with adaptive characteristics by Toplitz transformation, and then sparse signal with vibration characteristics can be obtained by sparse decomposition. The simulation and experimental results show that the proposed method can extract the frequency domain and time domain characteristics of impulse characteristics more effectively than the sparse signal diagnosis method based on discrete cosine transform (DCT) and discrete Hart transform (DHT) dictionary. (C) 2019 Elsevier Ltd. All rights reserved.