Research On Mechanical Fault Diagnosis Algorithm Based On Sparse Representation

Mechanical equipment usually operates in complex and changeable conditions such as high speed,hyperthermia,heavy load and severe corrosion.Once the failure occurs,the consequences will be disastrous.In order to maintain the normal status of mechanical equipment and progress of industrial production,the research on fault diagnosis technology of large and complex mechanical equipment gradually becomes the focus of the world.However,the collected vibration signal is liable to be interfered by noise components,which often influence the subsequent mechanical equipment diagnosis.Consequently,there is a huge challenge in the field of mechanical equipment diagnosis.It is always difficult to extract weak fault features accurately.Consequently,there is a huge challenge in the field of mechanical equipment diagnosis.It is always difficult to extract weak fault features accurately.From the aspect of sparse representation theory,this thesis aims to solve the problem of mechanical weak fault feature enhancement under complex working conditions and sparse fault feature extraction with cluster characteristics.Specifically,this thesis conducts extensive research for machinery fault diagnosis,wherein the diagnostic methods are developed based on feature extraction and sparse coding.The research work in this thesis is of great significance and practical value to fault diagnosis of machinery in engineering fields.The detailed contributions of this thesis are listed as follows.(1)The fault signals tend to be weak,since the mechanical failure occurs in the early stage.Besides,the failure feature is easily disturbed by noise and other signals.As a result,the performance of conventional sparse signal processing is limited.To solve this problem,a non-convex regular term that acts as a penalty function is introduced,which could both induce sparsity maximally and retain the cost function convexity.Compared with other processing methods,the proposed non-convex regularization is more accurate in estimating high-amplitude components.Besides,it is also adept in extracting much sparse fault features.Experimental results verify the effectiveness and advantages in enhancing the weak fault features and improving the accuracy of feature extraction.(2)In several damaged rotating machinery equipment,the fault generally exhibits the characteristic of clusters.Nevertheless,existing feature extraction methods usually ignore the feature of clusters when processing the signals,resulting in poor feature extraction performance.Aiming at this problem,this thesis selects the Tunable-Q Wavelet Transform as the basic transformation matrix,and processes signals in batches based on the Overlapping Group Shrinkage methodology.Then a cluster sparse feature extraction algorithm based on the Tunable-Q Wavelet Transform and the Overlapping Group Shrinkage is proposed to improve the accuracy of cluster sparse feature extraction.Experimental results show that the proposed method outperforms other methods with respect to cluster sparse feature extraction.(3)This thesis selects the Tunable-Q Wavelet Transform as the basic transform matrix,and then optimizes the methods of weak feature enhancement as well as cluster sparse feature extraction.Since the Tunable-Q Wavelet Transform essentially changes the characteristics of the wavelet filter by adjusting the internal quality factors,the oscillation characteristics of the wavelet basis function could match the signal to be analyzed.In order to achieve the best effect for an ideal process of feature extraction,the adopted wavelet basis function should be highly matched with the oscillation characteristics of the signal to be analyzed.However,it is difficult to set proper decomposition parameters of the Tunable-Q Wavelet Transform only by experience.Accordingly,it is hard to accurately obtain the optimal decomposition results.In this work,the amplitude weight of fault features is adopted as an index to adaptively select the parameters to obtain the optimal decomposition results,which is convenient to extract the fault features more accurately in the following steps.