Research On Sparse Representation Of Dictionary In Compressed Sensing Of Rotating Machinery Vibration Signals

Applying the theory of compressed sensing to vibration signal can solve the problem of storage and transmission of a large number of data in the field of condition monitoring and fault diagnosis of rotating machinery,but the precondition of the theory is that sparse representation technology is limited by the structure of dictionary.Therefore,this paper focuses on the research on sparse representation of dictionary in compressed sensing of rotating machinery vibration signals.In the process of dictionary generation,the common sparse estimation methods need to go through many times to complete the final prediction,which leads to low efficiency.In this paper,BP neural network is proposed to realize the automatic sparse estimation.A cascading dictionary based on k-singular value decomposition is proposed to solve the problem that the learning dictionary can not guarantee the sparsity efficiency at the same time,although it has high reconstruction accuracy.By comparing the above algorithms with the vibration signals of typical parts(bearings,gears and rotors)of rotating machinery,the superiority of the algorithm proposed in this paper is verified,which provides a theoretical basis for the engineering application of vibration signal compressed sensing.The main contents of this paper include:(1)The method of estimating the sparsity using SVD is used to explore the relationship between the sparseness of vibration signals and the eigenvalues.Based on the nonlinear relationship between the two,a method of estimating the sparsity of vibration signal by BP neural network is proposed,which realizes the automatic estimation.Using vibration signals from rolling bearings,gears and rotors to train and test the network,the test results show that the estimated sparsity obtained by this method has lower error compared with the real sparsity,and the error has less impact on the reconstruction quality and sparsity efficiency.This method makes the estimation of sparsity more concise and fast.(2)In order to solve the problem that the learning dictionary has strong adaptability but the dictionary generation is slow,a cascaded dictionary structure is proposed for sparse representation of vibration signals.Based on the comparison of the efficiency and the corresponding reconstruction accuracy of the common dictionary generation algorithm,ksingular value decomposition algorithm is determined as the main structure of the cascading dictionary.(3)Through the established simulation signal and the fault vibration signals of rotor,rolling bearing and gear,which are from the fault experiment platform,the results show that the sparse efficiency of the cascading dictionary is higher than that of the common redundant base dictionary,and the reconstruction quality of the original algorithm can be achieved within the effective compression ratio range.(4)Based on the Matlab GUI,the proposed sparsity prediction method,the proposed dictionary structure and the common learning dictionary structure are systematically integrated to build a sparse reconstruction system of vibration signal of rotating machinery which can be applied to Windows platform.