Research On Bearing Fault Diagnosis Method Based On Multi-feature Extraction And Extreme Learning Machine

In recent years,Chinese science and technology has been further promoted.The development of information technology promoting the progress of industrialization that makes mechanized production becoming more and more popular.In the production process of machinery,the unplanned shutdown and fault shutdown of the mechanical system will bring serious damage to the development of production and economic benefits,even there will be risks to the personal safety,resulting in security incidents.In many parts of rotating machinery,the highest and most vulnerable part is the bearing.Its normal operation directly affects the performance and life of the entire mechanical system.Therefore,it is significant to research the method of rolling bearing fault diagnosis.The vibration signals of rolling bearing are mostly non-stationary and nonlinear,at the same time,due to the influence of the noise from the peripheral equipment and the short-term impact component containing in the fault information,it is difficult to extract the fault features.Feature extraction is not comprehensive and characteristic value is not obvious,the above two points will affect the bearing fault identification accuracy which can bring miscalculation or even false negative phenomenon.Aiming at this problem,this paper comes up with the bearing fault diagnosis method research based on multi-feature extract and extreme learning machine.The main research work of this paper are as follows:(1)Studied the reasons for the formation of rolling bearing faults and vibration frequency.The structure and dynamic characteristics of the bearing are analyzed in detail,and putting forward the basic parameters of several typical vibration.Learning and researching the rolling bearing fault data acquisition system too in order to obtain the bearing vibration signal under different working conditions.Finally,taking the typical nonlinear system of Logistic as the verification object that verified the permutation entropy can be used to detect the dynamic catastrophe behavior of nonlinear systems.(2)A fault diagnosis method of rolling bearing based on permutation entropy and extreme learning machine is proposed.The original acceleration vibration signal is decomposed by using multi-resolution singular value decomposition which can get three different detail components D1～D3.Combining the advantage of permutation entropy in information extraction constructs the characteristic vector which can represent the fault characteristics of the original acceleration vibration signal.At last,the method of extreme learning machine is used to identify the type of bearing fault,which can verify the feasibility and validity of the method.(3)A fault diagnosis method of rolling bearing based on optimized maximal relevance and minimal redundancy and extreme learning machine is proposed.With it,time domain,frequency domain and time-frequency domain feature sets was generated using the original signal,which is denoised by multi-resolution SVD.On feature selection,a selection method based on weight maximal relevance minimal redundancy(WMRMR)was used,finally,it could select 8 major feature vectors from the mixed domain containing 18 features based on the classification accuracy of extreme learning machine(ELM).By analyzing the experimental data,the fault type can be accurately identified with 97.5%accuracy.The result proved that the method presented here can be realize diagnosis of bearing fault type efficiently.By analyzing the actual rolling bearing fault data,results show that the rate of fault diagnosis based on optimized MRMR and extreme learning machine is higher than permutation entropy and extreme learning machine.Comparing with the permutation entropy which the characteristic value is single and characteristics isn’t obvious,hybrid domain feature can represent the inherent characteristics of the vibration signal versatilely,the feature subset selected by optimized MRMR is the most representative.