Research On Nonconvex Regularized Sparse Representation For Rotating Machinery Fault Diagnosis

Rapid progress of science technology has promoted vigorous development of modern industry.Industrial mechanical equipment is developing in the direction of large-scale,precision,high-speed and intelligence.Widely used in in modern industrial equipment,the rotating machinery often operates in complex environments,such as heavy load,high speed and high temperature,and thus its key components inevitably suffer from performance degradation,leading to failure.Therefore,it is of great significance to carry out condition monitoring and fault diagnosis of mechanical equipment and reliably identify potential abnormalities or faults in these key components for ensuring safe and reliable operation of mechanical equipment.This dissertation focuses on the application of signal sparse representation method in fault feature identification,aiming at the accurate identification of fault features and fault diagnosis of key components in rotating machinery equipment.Aiming at two key problems of signal sparse representation,namely,the construction of sparse representation dictionary and the proposal of sparse decomposition algorithm,the nonconvex regularized sparse representation method is studied to improve the estimation accuracy of fault feature and reliability of fault diagnosis.The methods mainly involve two aspects:one is to construct a sparse representation dictionary matching fault components;the other is to propose an effective sparse decomposition algorithm to accurately estimate the sparse representation coefficients.The main work of this paper is as follows:Research on the matching pursuit method under the wavelet dictionary and its application in fault feature estimation.Based on the analysis of the local fault signal of key components,the over-complete analytical dictionary is constructed using tight-supported Laplacian wavelets and Morlet wavelets.In the aspect of sparse decomposition algorithm,we use LO-norm to construct an accurate sparse representation model.The model is solved using the averaging Randomized OMP algorithm,which adopts the idea of fusing multiple sparse representation results to improve the estimation accuracy.The proposed method is verified to be able to estimate fault components with high accuracy by simulation studies.The practical utility of the proposed method is verified by the application in fault feature identification of bearings and gearboxes.Research on the nonconvex regularized sparse representation method under the tight frame and its application in fault feature estimation.In terms of the sparse representation dictionary,considering the different oscillation attributes of components in fault signals,the tunable Q-factor wavelet transform is studied to construct the dictionary.In the sparse decomposition algorithm,considering that the commonly-used L1-norm underestimates the high-amplitude components and affects the estimation accuracy of fault components,the sparse representation model is constructed using GMC penalty function.The GMC penalty can improve the amplitude underestimation,and the sparse representation model can be guaranteed to be convex under specific conditions.Thus,the convex optimization algorithm can be used to solving the model and the solution can converge steadily.Through the numerical simulation and the experimental analysis,the proposed method is verified to be able to significantly improve the problem of amplitude underestimation,improve the estimation accuracy and increase the reliability of fault diagnosis.Research on the balanced nonconvex regularized sparse representation method under the tight frame and its application in fault feature estimation.Sparse priors of signals are the key to construct sparse representation models.Contrast with the extensively studied synthesis-based prior modeling used for fault feature estimation,relatively few studies on the analysis-based and balance-based priors modeling.The balance-based model establishes a connection between the synthesis-based model and analysis-based model by the balance parameters.To verify the performance of balance priors in fault feature estimation,and to meet requirements of accurate identification of fault feature,a sparse representation method based on balanced prior and MC nonconvex penalty functions is proposed.The performance of the proposed method and other sparse prior methods in fault feature estimation is systematically analyzed and verified.The proposed method is verified to be able to estimate fault components with high accuracy by simulation studies.The practical utility of the proposed method is verified by the application in fault feature identification of bearings and gearboxes.The research work in this dissertation has the theoretical and practical significance for improving the accuracy of fault feature estimation and increasing the reliability of fault diagnosis for key components in rotating machinery.