Fault Diagnosis Of Centrifugal Compressor Vibration Based On Support Vector Machine

The centrifugal compressor is the core equipment in the process industry such as air separation and refining.It requires running safely and efficiently.During equipment operation,it is very important to warn and diagnose faults.Two important rotating parts in centrifugal compressor are rotor and gearbox.This paper establishes fault diagnosis model for rotor and gearbox by support vector machine.Fault samples of centrifugal compressor rotor are collected.The frequency characteristics of rotor vibration fault are analyzed.The maximum amplitude of characteristic frequency domain of vibration signal is extracted as the features vector of fault.Membership functions of features vector are defined with BP neural network and expert scoring method.The identification models of single fault and multiple faults are established with support vector machine.These models have good recognition rate.In order to establish the gearbox running state recognition model,The gearbox vibration data of four points and six kinds of running state is collected.The trend term and noise of vibration are eliminated.Date features vectors of in view of three dimension(the time domain,energy and sample entropy)are extracted with the treated vibration data.The validity of each features vector is preliminarily analyzed by the error bar graph.For each measuring point,the fault diagnosis support vector machine(SVM)models are established using time,energy,entropy and comprehensive characteristic respectively.The model with comprehensive characteristics has the highest recognition accuracy.All the characteristic parameters of four points are used to establish gearbox running state identification model.The maximal correlation and minimum redundancy criterion is used to reducing the dimension of the integrated features vectors.At last,the running state identification model is established with support vector machine.It has good classification results among test samples.