High-low frequency features fusion and integrated classification SCNs for intelligent fault diagnosis of rolling bearing

For fault diagnosis of the rolling bearing, it is difficult to accurately extract and recognize fault features of the vibration signal. In this paper, a novel feature extraction method based on optimized Maximum Correlated Kurtosis Deconvolution (MCKD) by an improved Black Widow Optimization Algorithm (IBWOA) and Singular Value Decomposition (SVD) method was proposed, and an integrated classification Stochastic Configuration Networks (ISCNs) model was designed for fault recognition. Firstly, SVD was used to denoise and reconstruct the signal, in which fault features of the reconstructed signal were highlighted by MCKD; however, the filtering effect of MCKD was seriously affected by accurate value of some parameters, so IBWOA was proposed to realize optimized selection of them. Then, Wavelet Packet Decomposition (WPD) was used to deal with the signal after the feature enhancement, and the high-frequency and low-frequency singular values were extracted as the feature vectors. Finally, the ISCNs model was designed to train and classify the low-frequency and high-frequency feature vectors several times, which were then decided by the principle of ”minority obeying majority”. Simulation experiments show that the proposed method can effectively highlight and extract bearing fault features, and the average diagnostic accuracy can maximum reach 99.66% according to multiple experiments.