Research On Fault Diagnosis And Performance Degradation Evaluation Method Of Rolling Bearing

Rolling bearing is the core component of major equipment in petrochemical industry,wind power generation,high-speed railway,aerospace and other important industrial fields.In industrial production,it is prone to failure and damage in long-term continuous high-speed operation process.As a result,the whole equipment is in potential safety hazard.In order to ensure the safe and stable operation of the equipment,it is great significance of timely fault diagnosis for equipment maintenance.In this dissertation,the main technical route is to analyze and process the vibration signals of rolling bearing.In constant condition,variable condition,early single,early composite and degradation assessment of rolling bearing,respectively,the fault diagnosis was studied.The main research contents are as follows:(1)Amining at the low accuracy of fault diagnosis which the fault feature information is difficult to effectively extract under constant conditions,a fault diagnosis method of rolling bearing based on the hierarchical dispersion entropy(HDE)and K-Nearest Neighbor(KNN)classifier is proposed.Firstly,vibration signals are decomposed hierarchically and then the dispersion entropy(DE)of different nodes is calculated to realize the effective extraction of high and low frequency band features of vibration signals.This overcomes the problems that DE can not extract fault features at multi-scale and MDE can not extract fault features at high frequency band.The feasibility and stability of HDE is verified by analyzing MDE and HDE of Gaussian white noise.Finally,HDE and KNN are combined to realize fault diagnosis of rolling bearing under constant condition.(2)Focusing on the problem that the diagnosis ability of fault diagnosis method for rolling bearing based on traditional shallow machine learning is reduced under variable condition,a fault diagnosis method of rolling bearing under variable condition based on one dimensional convolutional neural network(1D-CNN)is proposed.Based on 1D-CNN,this method replaces full connection(FC)of traditional 1D-CNN with global average pooling(GAP).Then in the GAP layer of the improved 1D-CNN combined,domain adaptation(DA)is introduced to realize the transfer learning of fault features in variable condition.Finally,the fault data of bearing under different conditions in Case Western Reserve University are taken as the verification object.The proposed method is compared with the traditional signal processing methods of SVM+EMD+Hilbert,BPNN+EMD+Hilbert,Residual neural network(Res Net)and no improvement 1D-CNN with DA.The results prove that the fault diagnosis ability of the proposed method is stronger whether noise is added or not.(3)Contraposing the problem that the vibration signals with early fault are weak and vulnerable to noise pollution,which leads to the difficulty of fault diagnosis,an early fault diagnosis method of rolling bearing is proposed by optimizing variational mode decomposition(VMD)and improving threshold denoising.This method establishes the optimal mode component selection criteria of L-kurtosis and correlation coefficient and whale optimization algorithm(WOA)is used to optimize VMD to realize decomposition of early fault vibration signals to detect characteristics of early fault impact.Then the optimal component signals are denoised by improving threshold de-noising.The fault impact characteristic frequency is extracted by envelope spectrum analysis so as to realize early single fault diagnosis.Finally,the effectiveness of the proposed method is verified in the simulation signals and early fault tests of bearings.At the same time,it is compared with the method based on optimized VMD decomposition with the noise reduction of Teager energy operator and the method based on envelope entropy criterion optimized VMD with improved threshold noise reduction,which indicate that the proposed method can better realize early single fault diagnosis.(4)It is hard to draw early composite fault features of rolling,an optimization swarm decomposition(OSWD)method is proposed to draw the early composite fault features of rolling bearing.Firstly,the optimization criterion based on square envelope spectrum negentropy is constructed.Then taking this as the fitness function,the swarm decomposition(SWD)is optimized by an improved grasshopper algorithm(IGOA)to decompose the early composite fault vibration signals to retain the energy of composite fault impulse features.Secondly,the envelope spectrum analysis is used to extract the different fault impulse feature frequency of composite faults so as to realize the early composite fault diagnosis of rolling bearing.Finally,the proposed method is applied to the simulation signals and a fan bearing to verify.The advantages of OSWD are demonstrated by comparing with SWD and VMD methods.(5)The fault diagnosis of rolling bearing only belongs to the post-fault diagnosis,which cannot meet the problem of life prediction and predictive maintenance of bearing.Considering global-local feature extraction and support vector data description(SVDD),a degradation assessment method of rolling bearing is proposed.In this method,the maximum variance function is introduced into neighborhood preserving embedding(NPE)algorithm.The global-local objective function of maximum variance preserving and minimum neighborhood preserving is established to realize the dimensionality reduction of high-dimensional degradation features of vibration signals.Then the performance degradation assessment of rolling bearing is realized by combining SVDD.Finally,the method is tested on the whole life bearing experiment.Compared with the related literature,local or global feature dimension reduction with SVDD,the proposed method has better performance.