Fault Diagnosis Approach Of Rolling Bearing Besed On Signal Sparse Representation

In face to the puzzle that traditional sparse representation model can not represent the signal under complex working conditions accurately,improve the accuracy of the Morphological Component Analysis(MCA)method to decompose multi-component complex signals,and realize the adaptive selection criteria of the Tunable Q-factor Wavelet Transform(TQWT)optimal wavelet basis under multiple parameters and the construction of sparse dictionary.Based on the related theories of sparse representation and resonance sparse decomposition,a non-convex regularized bearing fault diagnosis model based on the alternating direction multiplier method(ADMM),a MCA signal separation method based on K-SVD dictionary learning,and a fault diagnosis method based on adaptive TQWT atom dictionary were proposed in this paper studies.The main contents are as follows:(1)A non-convex regularized bearing fault diagnosis model based on the ADMM strategy was established.As we all known,the fault features of mechanical equipment under complex working conditions is extremely weak,the result of signal sparse characterizing under the traditional sparse model is not accurate,and it is difficult to accurately diagnose the operating state of the mechanical equipment.A non-convex regularized bearing fault diagnosis model based on the alternating direction multiplier method was established.First,according to the vibration signal characteristics and the theory of non-convex penalty function,construct a non-convex regularized sparse representation model;secondly,because the objective function of constructed non-convex regularized sparse representation model is too complex to solve using a single optimization algorithm.ADMM algorithm idea was introduced into the process of optimal solution of non-convex optimization problems;finally,the signal is reconstructed by the sparse solution and the feature extraction of weak signal fault is achieved precisely.(2)Research on MCA signal separation method based on K-SVD dictionary learning.According to the structural characteristics of the shock component of the vibration signal,the fixed dictionary in the traditional MCA algorithm is replaced with the K-SVD learning dictionary to fully learn the signal shock characteristics and realize the accurate separation of multiple components of the signal.Collecting laboratory rolling bearing fault signals to verify the effectiveness of the method,and the results show that the MCA signal separation method based on K-SVD dictionary learning algorithm can separate the impact characteristics from weak bearing fault signals,and realize the weak fault diagnosis of the bearing;Compared with the fault diagnosis algorithm of K-SVD dictionary learning,this method still shows stronger fault feature extraction ability.(3)An adaptive TQWT base atom dictionary algorithm applied to fault diagnosis was investigated.Based on the theory of TQWT algorithm and energy-weighted normalized wavelet entropy,an adaptive TQWT base atom dictionary model is established;the Orthogonal Matching Pursuit(OMP)algorithm was used to achieve the signal sparse representation under the dictionary.The laboratory rolling bearing fault signals were processed by the proposed method,and the conclusion was drawn from the experimental results:The signal fault features of rolling bearings were extracted accurately by adaptive TQWT base atom dictionary algorithm and the weak signal fault diagnosis of rolling bearings was achieved.