Research On Fault Diagnosis Of Rolling Bearing Based On Improved Singular Spectrum Decomposition And Dispersion Entropy

Rolling bearing is known as one of the "joint" parts of rotary machinery and is widely used in modern industrial equipment.However,bearing is often under severe conditions such as heavy load and high-speed during work.Once a failure occurs,it may cause equipment damage,and even threaten the lives of workers.Therefore,research on rolling bearing condition monitoring and fault diagnosis technology has far-reaching significance for ensuring safe and stable operation of rotating machinery.In this paper,the rolling bearing is taken as the research object,the vibration signal is taken as the analytical medium,and the existing rolling bearing fault diagnosis methods are deeply studied and optimized on the basis of singular spectrum decomposition(SSD)algorithm and dispersion entropy(DE)theory.The main contents of this paper are as follows:(1)The basic theory of the SSD algorithm is sorted out,and two limitations of the SSD in processing rolling bearing fault signals are summarized through simulation signal.They are the applicability of bandwidth estimation model and the problem of improper selection of embedding dimension.Additionally,the influence of embedding dimension on the decomposition results was studied through the simulation signals,which laid a theoretical foundation for subsequent improvement.(2)Aiming at the limitations of SSD in processing rolling bearing fault signals,an improved singular spectrum decomposition(ISSD)algorithm is proposed.Compared with the conventional SSD algorithm,this method uses the order statics filter(OSF)to calculate the upper envelope of the power spectrum of vibration signal,and then fits the upper envelope through the Gaussian superposition spectrum model to improve the accuracy of the main peak bandwidth estimation.Based on the proposed mode mixing index and over-decomposition index,the optimal embedding dimension of each iteration is selected to improve the decomposition result and suppress mode mixing.The proposed method and the comparison method are quantitatively evaluated by simulation and experimental analysis.The results show that the proposed method has a stronger capability of fault feature extraction,and can effectively reduce the number of false modes compared with the conventional SSD algorithm,and the computational efficiency is more stable.(3)Aiming at the problem that the DE does not consider the amplitude relationship of adjacent points of the same category,a refined dispersion entropy(RDE)algorithm is proposed.This RDE counts the distribution of embedded vectors belonging to the same dispersion pattern,and redefined the calculation method of the probability of dispersion mode according to the distribution situation,and its probability is proportional to the degree of concentration of the embedding vector.Simulation results show that the proposed method has a stronger ability to detect small complexity changes of signal.Additionally,in order to enhance the feature extraction ability of multi-scale DE,a composite multi-scale RDE weighted based on root mean square and spectral negative entropy(SNE-w CMRDE)is proposed,this method introduces weights to enhance or offset the estimated value of the DE at certain scales.Simulation experiments show that the method is more in line with people’s expectations than conventional multi-scale entropy in the estimation of signal complexity.(4)To enhance the anti-noise ability of multi-scale DE,a rolling bearing fault diagnosis method based on ISSD and SNE-w CMRDE is proposed.The method first uses the ISSD to decompose and reconstruct the vibration signal to increase SNR,and then extract the SNEw CMRDE features of the signal to form a feature vector.The application results of two experimental data sets of rolling bearings show that the proposed method has a stronger ability to distinguish fault severity and fault type than the conventional multi-scale DE,and is more robust.