Single Channel Blind Source Separation Based On Improved VMD Rolling Bearing Composite Fault Diagnosis

One of the main components of spinning machines is the rolling bearing,whose running state directly influences the entire equipment’s operation.It is also not a single fault,usually several faults coexist and reciprocal coupling,until a fault occurs,so the detection of bearing composite fault has a very significant functional engineering importance.This paper primarily discusses three facets of vibration signal noise reduction,signature elevation and blind source isolation,based on the main rolling bearing compound fault line.Firstly,there is the issue of threshold selection in the conventional wavelet denoising procedure,and the decomposition mechanism is split into inherent high and low frequency modes.The time-frequency resolution would become weak with the rise in decomposition levels.A wavelet packet adaptive threshold de-noising mechanism is proposed in this article.For the decomposition of the signal,the wavelet packet is used and the adaptive threshold is used to denoise the coefficients of each decomposed wavelet packet.This approach can not only efficiently filter highfrequency noise through the verification of experimental and simulation results,but also has a certain noise reduction effect on the low-frequency,effectively increasing the signal-to-noise ratio.Secondly,in view of the fact that the separation of single channel blind source has little known knowledge and can not be directly solved,the underdefined problem of blind source separation needs to be turned into a positive problem of definite blind source separation,and a method of feature dimension elevation based on improved decomposition of variational mode(VMD)is proposed.In order to adaptively evaluate the sum of modal decomposition and penalty factor,the energy factor and information entropy are added as constraints.Using the close value dominance process,the feature dimension of the de-noising signal is increased,the number of sources is calculated,and the IMF sections are filtered using the correlation analysis and kurtosis index,and the multi-channel feature collection is constructed as the blind source separation algorithm data.The approach suggested in this paper will decompose the signal adaptively,correctly approximate the number of origins of faults,and solve the problems of the aliasing mode of conventional signal decomposition effectively.Then,as the objective convergence function of FastICA,Tukey M calculation is proposed to target at the slow convergence speed and poor precision of the objective function of conventional FastICA.In order to reduce the effect of the individual residual value on the estimation outcome,and to increase its convergence speed and robustness,the Tukey m estimator can have sufficient weight according to the residual value.The fault frequency of the split signal is essentially compatible with the theoretical fault frequency,which ultimately realizes the composite fault diagnosis,taking the multi-channel characteristic matrix as the data.Finally,an experimental device is designed for composite fault diagnosis,and the single row angular touch ball bearing is used as the experimental entity to validate the effect of the proposed de-noising adaptive threshold wavelet packet and FastICA based on Tukey M estimation.