| Rolling bearings,as key parts of rotating machinery,are widely used in the field of machinery.Once they fail,they will bring huge economic losses to people’s production and life,and even endanger people’s life safety.Therefore,fault diagnosis of rolling bearings is the key technology to ensure the reliable operation of mechanical equipment.Due to the complex working environment of rolling bearings,bearing vibration signals contain a lot of noise.At the same time,the continuous operation of equipment and the uninterrupted status monitoring produce massive data,which brings a great burden to the transmission,storage and fault diagnosis of bearing signals.Compressed sensing,as a new way of information acquisition,takes advantage of the sparse characteristics of the signal and can complete the accurate reconstruction of the signal with a small number of observed values.Under the framework of compressed sensing,this paper carries out research from the following three aspects: sparse representation,signal observation and fault diagnosis.(1)In the process of sparse representation of fault signals,dictionary learning using K-singular value decomposition algorithm(K-SVD)is vulnerable to noise interference and cannot effectively represent the impact characteristics of fault signals.Therefore,a sparse representation method based on K-SVD dictionary in wavelet domain is proposed.The original signal is projected into the wavelet domain by wavelet transform,and the optimal coefficient is screened by the normalized autocorrelation function as the training sample of dictionary learning,and the sparse dictionary is constructed by combining K-SVD algorithm for dictionary learning.The effectiveness of the proposed method in sparse representation is verified by fault simulation signals.(2)Aiming at the problem that the observation values extracted from the deterministic measurement matrix cannot be accurately reconstructed,an optimized measurement matrix based on sparse Toeplitz matrix is proposed.Firstly,the Toeplitz matrix is sparse on the basis matrix,and then the elements in the sparse Toeplitz matrix are optimized by equiangular compact frame,and the optimal measurement matrix is obtained.Finally,the advantages of the matrix in signal observation are verified by the reconstruction experiments of simulated signal and actual signal.(3)Under the compressed sensing framework,firstly,the proposed wavelet domain K-SVD dictionary is used to conduct sparse representation of inner circle,outer circle and roller fault signals,and signal observation is carried out by optimizing the measurement matrix.At the same time,signals are reconstructed by combining the orthogonal matching tracking algorithm to obtain reconstructed signals.Then twelve characteristic indexes of reconstructed signal are calculated and six indexes with low aliasing degree are selected and input into extreme learning machine for classification.Finally,the accuracy of fault identification method is verified by experimental analysis of actual bearing signals. |
