Research On Fault Diagnosis Method And Application Of Rolling Bearing Based On Sparse Representation

Modem mechanical equipment toward large-scale,high-speed,precision,digitization and intelligence,which put forward more stringent requirements for the safety and stability of mechanical equipment.As a kind of common mechanical parts.rolling bearing is faced with complex and changeable working environment in actual operation,which is one of the main fault sources of mechanical equipment.Research on fault feature extraction and fault type diagnosis of rolling bearings under complex working conditions is carried out,which is to avoid unpredictable accidents,solve the problems of insufficient and excessive maintenance,promote industrial production and improve social and economic benefits.It is difficult to extract fault feature components of bearing vibration signals hidden in a large number of redundant noise.Sparse representation uses over-complete dictionary bases to replace traditional orthogonal bases,and uses a small number of atomic libraries to represent more feature information,so as to capture and express the essence of fault features efficiently.In this paper,in view of the shortcomings of traditional rolling bearing fault diagnosis methods based on sparse representation in theoretical and practical applications,.Rolling bearings of printing press is taken as the main research object,and the weak signal fault diagnosis under the complex working scenarios such as strong background noise and variable working conditions are taken as the application background.Based on the sparse representation theory of signals,four aspects are studied in depth,namely,dictionary learning improvement,mixed noise model establishment,fault features transfer under variable working conditions,and fault feature subspace construction.A series of methods arc applied to multiple experimental cases for verification.The research contents include the following aspects:(1)Aiming at the problems of noise interference,false atoms mixing and improper parameters setting in the process of K-means singular value decomposition(K-SVD)dictionary learning,which lead to the introduction of noise or the loss of feature components.An improved rolling bearing fault feature extraction method based on K-SVD dictionary learning and variational modal decomposition method(VMD)is proposed.VMD method is used to extract the main components related to fault features from the nonlinear and nonstationary vibration signals.The initialization dictionary matching the fault shock components is established from the original data.The spectral negative entropy index is introduced as the constraint condition of K-SVD,the kurtosis and residual error are used to verify the correctness of the method.The experimental results show that the proposed method can effectively extract the potential fault feature information.Compared with the traditional K-SVD method,the interference components are obviously weakened.It has advantages in sparse representation effect and feature extraction ability.(2)Vibration signals of rolling bearings are mixed with various noise in complex environments.It is difficult to extract fault features by traditional dictionary learning methods that obey Gaussian distribution.A sparse representation fault diagnosis method based on mixed noise dictionary learning is proposed.A mixed noise model based on Gaussian distribution and Laplace distribution is established.The optimization problem of the model is decomposed into sub-problems with closed solutions by alternating direction method of multipliers.The fault types of rolling bearings are identified by calculating the minimum redundancy error between the measured signals and the reconstructed sparse signals in the dictionary learning model.The experimental results show that the robustness of mixed noise dictionary to complex noise representation is better than that of single distribution dictionary learning model.Compared with four representative methods,the proposed method has better feature extraction and classification diagnosis ability under complex environmental noise.(3)The traditional fault diagnosis method based on dictionary learning is not suitable for the problem of unbalanced fault data distribution under actual variable conditions.A sparse representation fault diagnosis method based on transfer dictionary learning under variable conditions is presented.The maximum average difference is calculated by the linear combination of Gaussian kernel function,and the distribution distance between source and target domains is effectively measured to solve the transfer learning problem between fault feature information with different distributions.The status of rolling bearings is diagnosed by the minimum redundancy error fault identification method,12 transfer tasks with different fault damage degrees and rotation speeds under variable working conditions are established,all of which obtain high accuracy.Compared with traditional signal processing methods and deep learning methods,it is verified that the proposed method can not only extract fault features from complex noise environment,but also has the ability to identify fault types across domains.(4)Under variable working conditions,the fault diagnosis method based on transfer dictionary learning cannot directly apply source domain data to target domain task learning.It is difficult to identify rolling bearing fault types.A fault diagnosis method based on mixed noise dictionary and transfer subspace learning is proposed.A transfer subspace model based on dictionary learning is established.According to the similarity of fault features in different domains,the source domain and the target domain are transferred to the constructed common subspace by transformation matrix.The distribution difference between the two domains is reduced by combining the distribution adaptation method and reducing the source domain classification error.Low rank and sparse constraints are imposed on the reconstruction matrix to preserve the structural relationship between data.The experimental results under different damage degrees,rotating speeds and load conditions show that the proposed method can effectively improve the fault feature transfer ability and classification accuracy.