Multi-bandwidth Mode Manifold Fusion For Fault Diagnosis Of Rolling Bearings

With continuous advancement in Chinese rail transit system,rail vehicle equipment has rapid development in integration degree and scale,which needs more strict requirements on manufacture,installation and daily maintenance of the equipment system.A minor failure or dislocation of any part may affect the normal operation of rail transit system and cause serious traffic accidents finally.Rolling bearings,which usually operate under heavy-load and high-speed conditions,are key components of rail vehicle equipment,thus bearings inevitably suffer from performance degradation,leading to failure.The key for bearing fault diagnosis is the extraction of the transient components,which is generated by periodic shock excitations.However,the vibration signals collected from rail vehicle equipment are typically nonstationary,nonlinear,and mixed with heavy noise,adding difficulties to accurate bearing fault detection.Therefore,it is of great significance to carry out rolling bearing fault diagnosis research and accurately extract fault transient components for safe and reliable operation of rail transit vehicles.This thesis aims at the accurate and efficient diagnosis of bearing failure types in rail vehicle equipment,a multi-bandwidth mode manifold fusion method and its improved strategy are proposed in this thesis and successfully applied to bearing fault diagnosis.The main content of work is as follows(1)Introduction of basic theory.The theoretical basis of variational mode decomposition(VMD)is first explained,and the influence of key parameters on the output results of VMD is analyzed.By summarizing main improvement methods and application strategies based on VMD,this thesis discovers the theoretical limitations of the VMD method in terms of denoising capabilities and computational efficiency.Then,based on the characteristic that different modes obtained by VMD have complementary information,the idea of information fusion is introduced and the development of manifold learning technology is summarized.The local tangent space arrangement(LTSA)algorithm is explained in detail,which provides a theoretical basis for the research of full text(2)Establishment of multi-bandwidth mode manifold fusion method.In order to accurately extract the transient components of bearing faults,a new multi-bandwidth mode manifold fusion method which applies manifold learning to nonlinearly fuse multiple fault modes with different bandwidths is proposed.First,aiming at the problem of computational redundancy and difficult to determine the number of modes in VMD method,an efficient decomposition strategy,named recycling VMD(RVMD),is proposed to realize fast,complete and accurate location of fault information.Furthermore,based on Gini index,a screening strategy for fault-related mode is proposed to construct multi-bandwidth modes which contain fault information.Finally,the low-dimensional nonlinear inherent structure is revealed from the high-dimensional fault data matrix by applying manifold learning,realizing accurate extraction of bearing fault characteristic components.The effectiveness and superiority of the proposed multi-bandwidth mode manifold fusion method is verified by analyzing the simulated and experimental bearing fault signals.(3)Improvement of multi-bandwidth mode manifold fusion method.Aiming at the problems that the number of neighborhood points is difficult to determine when constructing the distribution of local data and the manifold feature would probably be asymmetric in the up-and-down direction,the improvement strategies are proposed respectively.First,a natural nearest neighbor algorithm is introduced to construct differentiated local data distribution and adaptively specify the neighborhood size based on the regional density of data,which overcomes the non-adaptive problem of traditional k-nearest neighbor algorithm.Furthermore,a waveform compensation strategy based on eigenvalue weights is proposed to construct the manifold feature with symmetrical waveforms,which is more suitable for characterizing bearing fault transient components.Through the analysis of the simulated and experimental bearing fault signals,it is proved that the improved method can achieve more efficient and accurate extraction of bearing fault transient components.In summary,this thesis proceeds from the goal of improving the accuracy of fault signal feature recognition and the efficiency of fault diagnosis.Through systematically studying the construction of multi-bandwidth mode matrix and manifold feature learning strategies,this thesis proposes multi-bandwidth mode manifold fusion and its improvement.The proposed method makes traditional bearing fault diagnosis technology get rid of the dependence on parameter selection,and improves the accuracy and efficiency of transient component extraction,which have certain theoretical guidance and engineering application value for rail vehicle bearing fault diagnosis.