Vibration Fault Diagnosis Of Multistage Centrifugal Pump Based On SVM And Experimental Verification

As an important energy conversion and fluid transfer device,multistage centrifugal pump is the key equipment in large industries such as petroleum,chemical industry,steel and so on.In order to ensure the safe and efficient operation of enterprise production,and to deal with the two major problems which are high maintenance cost and long repair time in current operation and maintenance of multistage centrifugal pump,it is of practical engineering significance to carry out condition monitoring and fault diagnosis research on multistage centrifugal pump.This paper focuses on the fault mechanism analysis and pattern recognition of the three typical faults in the rotor system of multistage centrifugal pump,such as misalignment,dynamic unbalance and rubbing impact between the stator and the rotor.An intelligent fault diagnosis method of multistage centrifugal pump based on optimized Support Vector Machine(SVM)is proposed and verified with actual products,and the condition monitoring and diagnosis system of multistage centrifugal pump is designed and developed.The main contents of the paper are as follows:(1)For the three types of faults,such as misalignment,dynamic unbalance and rubbing impact between stator and rotor,the vibration characteristics of each single fault system are obtained through force analysis and solving dynamic equations based on simplified mechanical model according to the actual situation.It can provide theoretical guidance for fault diagnosis and intelligent fault diagnosis in practical engineering.Among them,the vibration characteristics of misalignment and dynamic imbalance faults are respectively characterized by significant double frequency(2x)and single frequency(1x)component in the vibration signal spectrum.The vibration characteristics of rubbing impact fault are mainly manifested by the "clipping" phenomenon of time-domain waveforms and the presence of high fundamental frequency components(2x,3x,4x,5x)in the frequency spectrum.(2)Aiming at the low accuracy and efficiency of fault diagnosis for multistage centrifugal pumps,an intelligent diagnosis method based on optimized SVM is proposed using the unique advantages of SVM in dealing with small samples,nonlinear,and high-dimensional problems.This method uses Ensemble Empirical Mode Decomposition(EEMD)to denoise the measured vibration signal of multistage centrifugal pumps,and high-dimensional feature is extracted and vector samples is constructed based on denoised signal.Sequential Feature Selection(SFS)or Principal Component Analysis(PCA)methods are used for feature dimensionality reduction to optimize the quality of input samples.And Particle Swarm Optimization(PSO)and Grid Search(GS)algorithms are combined to optimize the parameters of the model itself.Finally,the optimal classification performance of SVM model is achieved.The experimental results show that the optimized SVM fault classification model can effectively identify actual product faults,and has good applicability in practical engineering.Compared with Back Propagation(BP)Neural Network,the classification effect of the optimal SVM model is slightly better,and the physical interpretation is stronger.(3)According to the actual requirements of on-line condition monitoring and fault diagnosis of multistage centrifugal pump,the condition monitoring and fault diagnosis system of multistage centrifugal pump is designed and developed based on the feature extraction and SVM model optimization methods in this paper.The effectiveness and engineering application value of the system are verified by the measured analysis and off-line pattern recognition of the vibration signals of the multistage centrifugal pump products.