Study On Mechanical Equipment Fault Diagnosis Method Based On Vibration Signal Feature Extraction

With the continuous advancement of modern science and technology,machinery toward the high-speed,large-scale,complex direction,mechanical systems toward the direction of high-speed development.The important rotating machinery and equipment such as rolling bearings,gears,steam turbine rotor and other rotating machinery equipments can lead to huge economic losses and safety problems when the system is running in the event of failure.The vibration signal analysis is the most commonly technique.The nonstationarity vibration signal analysis method is the key to improve the level of fault diagnosis technology.Therefore,studying the vibration signal feature extraction and fault diagnosis method of the rotating machinery and equipment have great practical significance.In this paper,vibration signal extraction and recognition methods of rotating machinery and equipment are studied.In particular,the main contents of this paper are as follows:(1)In view of the computational efficiency problem in the MSE.(Multiscale Sample Entropy),MFE(Multiscale Fuzzy Entropy)and MPE(Multiscale Permutation Entropy)models.A method named MBSE(Mutilscale Base-Scale Entropy)was proposed.Firstly,the advantages and disadvantages of SE(Sample Entropy),FE(Fuzzy Entropy)and PE(Permutation Entropy)are compared and analyzed by simulation signal.From the view of run time,the number of basic arithmetic operations,the different parameters in vaious models to compare the computation efficiency,and analyzes the reasons why the computation efficiency of MSE/MFE/MPE models are not good.Finally,the different combinmation methods based on MSBE/MSE/MFE/MPE,SVM(Support Vector Machine)by using PSO(Particle Swarm Optimization)/GA(Genetic Algorithm)and RF(RandomForest)were used to identificate the vibration signal under different conditions for roller bearings.The result shows that the computation efficiency and the recognition accuracy of MBSE model under different parameters is better than the MSE/MFE/MPE.So it is improve the diagnostic efficiency and accuracy,and better provide decision support for the diagnosis system.(2)In order to solve the problem of instability and poor overall consistency in the ME model,the CMPE(Compound Multiscale Permutation Entropy)method based on MPE by introduce average entropyunder each scale factor was proposed.Therefore,the overall smoothness and consistency of MPE/CMPE were compared under different parameters and various length of timeseries in detail.The combination method based on CMPE,PSO/GA,SVM and RF were build.Finally,using the mentioned above combination models to fulfill the rolling bearing and gears vibration signals identification under different conditions.Experiments show that the diagnosis effect of CMPE model is better than MPE on overall.(3)Aiming at the data in practical engineering has characteristics of no labels characteristics and its high dimension problems.The correlation coefficient method and PCA(Principal Component Analysis)were used to choose the number eigenvector and its dimension reduction.A combination method based on EMD(Empirical Mode Decomposition),EEMD(Ensemble Empirical Mode Decomposition),LMD(Local Mean Decompoeiton),SE/FE,FCM(Fuzzy-C-Means)/GK(Gustafson-Kessel)?GG(Gath-Geva)were used to finish the roller bearings and gears vibration signals identification under different conditions,then the mentioned above twelves models were compared by using Clustering evaluation index.The experimental results show that the recognition effect of EEMD-FE-GG is better than other eleven kinds of combination model(EMD/EEMD-SE/FE-FCM/GK,EMD/EEMD-SE-GG).(4)To slove the problem of preconfigured the number of clustering center points existed in roller bearings fault recognition by using cluster method,a method based on EEMD/LMD(Local mean decomposition),SE/FE/PE(Permutation Entropy)and CFS(Clustering by Fast Search)/AP(Affinity Propagation)was presented for roller bearings diagnosis recognition.The CFS method was analyzed according to the cut-off distance parameter in detail.But the condition of human choose the clustering center points exists in CFS model,the CFS/AP models were used to fulfill the roller bearings fault diagnosis.