Font Size: a A A

Fault Diagnosis Method Based On Sparse Representation And Dictionary Learning

Posted on:2019-11-16Degree:DoctorType:Dissertation
Country:ChinaCandidate:H D YuanFull Text:PDF
GTID:1362330590970285Subject:Mechanical design and theory
Abstract/Summary:PDF Full Text Request
With the rapid development of science and technology,modern mechanical equipment are increasingly large-scale,high-speed,heavy and intelligent.However,the operating conditions of the equipment are more and more complex and harsh,thus high operational reliability is required.Failure of some key components of mechanical equipment will not only cause huge economic losses to enterprises and countries,but also cause serious casualties and environmental pollution as well as bad social impact.Therefore,the condition monitoring and fault diagnosis of mechanical equipment is of great significance to protect the economic benefits of enterprises and the safety of workers.Based on sparse representation theory and dictionary learning method,the key components of rotating machinery including rolling bearings,gears and rotors are studied in this dissertation.The dictionary learning method is based on K-SVD algorithm and four aspects are studied surrounding the dictionary learning based on K-SVD algorithm as well as the sparse representation method: the weak feature extraction based on shift invariant K-SVD;the single channel compound fault analysis based on shift invariant K-SVD;the locality constrained sparse feature extraction based on K-SVD and improved LLC algorithm;intelligent diagnosis based on time-frequency images and discriminant K-SVD.The main research contents of this paper are as follows:(1)In order to solve the problem that the signal has periodically recurring patterns,a feature extraction method of mechanical fault signal based on shift-invariant K-SVD dictionary learning is proposed.The method mainly includes two major steps,that is,the learning of multiple fault patterns and the selection of the optimal latent components,which can effectively extract the periodically recurring fault features in mechanical fault signals.Through the simulation and experimental analysis and compared with the signal feature extraction methods based on K-SVD dictionary learning and wavelet dictionary matching pursuit method,the effectiveness of the proposed method is verified.(2)With regard to mechanical compound fault,a single channel blind source separation method based on shift-invariant K-SVD dictionary learning and adaptive cluster is proposed.The method first uses the shift-invariant K-SVD dictionary learning method to adaptively learn single channel compound fault signal to obtain a set of basis functions and corresponding latent components,and then cluster analysis is conducted based on the structure similarity of latent components and the mean value of between-class correlation coefficient is employed to determine the optimal number of clusters,and finally the separation of different fault source signals is achieved.The effectiveness of the proposed method is verified by the simulation and experimental analysis of the compound fault of rolling bearing.(3)In order to make signals under different conditions have better distinguishability so as to improve the accuracy of fault diagnosis,based on the locality-constrained linear coding(LLC)algorithm,a locality constrained sparse feature extraction method based on K-SVD dictionary learning and improved LLC sparse codes is proposed.Firstly,the feature of the mechanical vibration signal is extracted by the feature extraction method based on time and frequency domain and so on and used as the preliminary feature.Then,K-SVD dictionary learning is performed to obtain an over-complete dictionary containing each state category.Afterwards,the improved LLC algorithm is used to obtain the locality constrained sparse features based on improved LLC sparse codes.Finally,the improved LLC sparse codes are utilized as feature vector and optimized SVM based on improved PSO algorithm is employed for mechanical fault diagnosis.The feasibility and effectiveness of the proposed method are verified by the experiments of single fault and compound fault of rolling bearings.(4)The time-frequency images of mechanical vibration signals contain abundant feature information,in order to realize automatic classification and recognition of the time-frequency features,a fault diagnosis method based on time-frequency images of mechanical vibration signal and discriminant dictionary learning is proposed.Firstly,the wavelet transform of mechanical vibration signal is used to obtain the wavelet time-frequency images,and then the gray level co-occurrence matrix is used to extract the texture features of the wavelet time-frequency images.Finally,label consistent K-SVD algorithm based on discriminative dictionary learning is employed to realize the intelligent fault diagnosis.The effectiveness of the method is verified by the diagnosis of rolling bearing fault and rotor fault.
Keywords/Search Tags:fault diagnosis, sparse representation, dictionary learning, shift-invariant K-SVD dictionary learning, blind source separation, discriminative dictionary learning
PDF Full Text Request
Related items