Research On Vibration Signal Analysis And Fault Diagnosis Methods Of Rolling Bearing

Rotating machinery is an important large category of mechanical equipment.They are widely used in the national pillar industries,such as electric power,metallurgy,mining,carrying tools,and the military industry related to the national security.With the progress of science and technology,rotating machinery is developing towards the direction of large-scale,complex and precision,and its quantity is also increasing rapidly.However,the working environment of rotating machinery is bad and usually runs continuously full load,which determines that the probability of failure of rotating machinery is higher.As an important part of the rotating machinery,the fault diagnosis of the rolling bearing is of great theoretical and practical significance to its fault diagnosis.The vibration signal of rolling bearing is often accompanied by noise,and has the characteristics of nonlinear and non-stationary,which leads to the acquisition of the signal that can not directly obtain accurate fault information.Therefore,this paper takes the vibration signal of the rolling bearing as the research object,and studies the fault diagnosis method of the rolling bearing.The main research contents include:The basic principles and properties of Local Mean Decomposition(LMD)method and empirical mode decomposition(Empirical Mode Decomposition,EMD)are introduced,and their similarities and differences are compared in detail,especially from the point of view of the algorithm steps and calculation results.And the advantages of local mean decomposition are highlighted,such as PFs can reflect the local fluctuation of signal more accurately,there is no over envelope and under envelope problem,the endpoint effect is relatively slight and the instantaneous frequency is more precise.This paper introduces the causes of endpoint effect,and puts forward the idea of solving the problem--extension endpoint.Then we study the properties of Support Vector Machine(SVM)and cosine window respectively,and propose an endpoint extension method combining SVM with cosine window.In view of the modal aliasing problem in the LMD method,a complete overall local mean decomposition method(Complete Ensemble Local Mean Decomposition with Adaptive Noise,CELMDAN)is proposed based on the predecessors' method of EMD method modal mixing.The experiment proves that the method can deal with the nonlinear and non-stationary vibration signal better.The basic principle of envelope spectrum analysis is introduced,and Stochastic Resonance(SR)is introduced to solve the problem of poor resolution.Because of the shortage of system parameter selection,after elaborating the principle and nature of SR model,the improved SR model of particle swarm optimization(Particle Swarm Optimization,PSO)is proposed.The signal to noise ratio(SNR)of output signal is taken as the optimization objective,and the excellent global search ability of PSO is used to maximize the SNR of output signal.By combining the improved model with the envelope spectrum analysis method,based on the rolling bearing signal of West reservoir University,the recognition ability of envelope spectrum analysis to the characteristic frequency is effectively improved.Aiming at the lack of accuracy in time-frequency analysis,an intelligent recognition method,neural network,is introduced.Firstly,the principles and properties of sample entropy,permutation entropy and fuzzy entropy are introduced in detail.And fuzzy entropy is selected as the training feature of neural network.Through the comparative study of Extreme Learning Machine(ELM)and traditional neural network,ELM is selected as the training model of this paper.Finally,the verification of the experimental signal shows that the method can accurately identify the fault characteristics of rolling bearings.At the end of the paper,the work is summarized and the related research technology is prospected.