Fault Data Sets' Reduction And Classification Based On Manifold Learning

With the rapid development of information technology,it has become an important trend for the development of mechanical fault diagnosis to apply information technology to rotating machinery,and explore a new method based on data-driven thinking mode that can excavate the intrinsic attributes of the operation state of machinery and reflect it in time.Manifold learning is a non-linear data editing algorithm.It can construct data structure in high-dimensional space,and has been successfully applied to data dimension reduction and data pattern recognition.Kernel principal component analysis(Kernel principal component analysis,KPCA)has a strong global distance ability but poor local clustering ability to data structure,while Local tangent space alignment(Local tangent space alignment,LTSA)has the opposite clustering ability to data structure.Therefore,this article proposed an improved KPCALTSA dimension reduction algorithm which can integrate the characteristics of the two algorithms.In addition,the article proposed a fault diagnosis method based on Variational Mode Decomposition(VMD),Local Tangent Space Arrangement and Gath-Geva(GG)clustering for the fusion of manifold learning and clustering classification algorithm for fault diagno sis composed.The main progress and research achievements are as follows:(1)Six typical global and local manifold learning algorithms were comprehensively expounded,and their advantages and disadvantages were compared and analyzed.Therefore,it is a meaningful idea to combine the two algorithms to complement their advantages and disadvantages.(2)In view of the problem that traditional data dimension reduction methods c ouldn't keep both global distance information and local structure information,a d ata set dimension reduction algorithm based on KPCA-LTSA fusion was proposed.The algorithm combines KPCA and LTSA with kernel method.Firstly,the expression of the core matrix of KPCA and LTSA is solved,and then the new kernel matrix is solved with the help of the equilibrium parameter,and finally the low-dimensional coordinates are calculated.The experimental results of swissroll on artificial data set show that the algorithm has successfully realized dimension reduction,and the effect of dimension reduction is good.2(3)The original fault feature data set was constructed based on the state signals acquired on the two-span rotor test bench,and then dimension reduction analysis and comparison were carried out using KPCA-LTSA,KPCA and LTSA algorithms.The recognition rate of three-dimensional graph and input KNN was used as the index to measure the effect of dimension reduction.The experimental results show that the fusion algorithm can successfully reduce the dimension of the rotor fault data set,an d the fault classification is clearer,and the corresponding recognition accuracy has been significantly improved.(4)Aiming at the problem that the type and damage degree of rolling bearing is difficult to be identified,a fault classification method bas ed on the combination of Variational Mode Decomposition(VMD)? Local Tangent Space Alignment(LTSA)and Gath-Geva(GG)fuzzy clustering is proposed.Firstly,the original signal of the known rolling bearing fault samples is decomposed by VMD.Secondly,LTSA algorithm is used to extract quadratic features from high-dimensional data.And finally,the membership degree matrix and clustering center of known fault samples are obtained by GG clustering algorithm.The fault type and the degree of damage of the rolling bearing are determined by calculating the Hamming proximity between the initial membership matrix and the known fault sample clustering center.The effectiveness of the method is verified by the vibration signal of the rolling bearing of the American University of Western Reserve,the experimental results show that the proposed method can accurately extract the fault information contained in the vibration signal,and realize the accurate clustering of different faults.It provides a reliable basis for fault diagnosis of rolling bearings.