Research On Mode Analysis Method For Undetermined Blind Source Separation Of Rotating Machinery

In order to separate the complex signal of rotating machinery and extract its fault characteristics,this paper presents a novel blind separation method based on the theory of the undetermined blind source separation and mode analysis method.Taking the roller bearings as research object,the complex fault signals collected by the acceleration sensors are used to separate the source signals.In addition,the paper compares the proposed methods with the traditional signal processing methods,which can verify the validity and practicability of the proposed methods.The main contents of the paper are as follows:First,a blind signals extracting method with the improved ensemble empirical mode decomposition(EEMD)method based on correlation coefficients is proposed.By decomposing the original signal with the improved EEMD algorithm,the present method can not only improve the accuracy of mode signals and solves the redundancy problem,but also successfully obtain multi-channel signals.Then the singular value decomposition method is used to estimate the sources number,and the appropriate mode signals are selected as the input matrix of ICA method to extract the fault sources.Experiments show that compared with the traditional method,this method can not only effectively separate the outer race fault and roller fault,but also improve the accuracy and reduce the running time.Second,a variational mode decomposition(VMD)method is used successfully to solve the blind source separation problem of roller bearings.Firstly,the original signal is decomposed into multi-channel mode signals by VMD method.Then,the singular value decomposition method is used to estimate the number of fault sources.According to the number of fault sources,appropriate mode signals are used as the input matrix of FastICA method so that the source signals can be separated.Experimental results show that this method can not only effectively separate the fault signals of the outer race,inner race and roller faults,but also improve the speed and accuracy of extracting the independent components compared with the traditional method.In addition,this method shows a good advantage in the low signal-noise ratio signals blind separation.Third,a blind separation model of mixed matrix estimation is established using VMD and normal vector of hyperplane(NVH)method.Firstly,the vibration signal of single channel is decomposed by VMD method.Then the BIC criterion is used to estimate the number of source signals.The mode vector is clustered by the NVH method so that the clustering result is obtained and the mixed matrix is estimated.Finally,the source signals can be restored to extract the fault source signal using the shortest path method.The experimental results show that this method can effectively separate the outer race,inner race and roller faults...