The Research Of Motor’s Fault Diagnosis Of Based On High-order Spectrum And Frequency Entropy

With the improvement of industry automation technology and integration level between factory equipment gradually, industrial intelligent is becoming the development tendency. The motor acts as the important roles of producer and driver in the complex industry producing environment,the malfunction of which has a direct influence on the production status of the entire system. To make sure the continues work of the motor is fine,it is an important method that the motor state is monitored uninterruptedly and quickly without loss by collecting the relevant electrical signals, analyzing the characteristics of the signal,extracting the related feature parameters, judging the state of system and predicting the potential hazards.Firstly,the feature extraction method for bis-pectrum entropy,which used in higher order statistics,is chosen after the analysis of the common method for the motor faults. Then,this paper introduces some signal analysis methods, including the time analysis and frequency analysis. Combining with simulational signals,the simple time domain characteristics of the strategy used in this paper are analyzed. Aiming at the applicable range of higher order statistics,the effectiveness of the bis-pectrum analysis for the realization of white noise signal is proved by grouped experimental signals,before the analysis of the experimental data by the method of higher order statistics. Meanwhile,this paper analyzes the generation mechanism and damages of the motor faults and samples the signals of electric field strength of DC motor by setting DC motor operating on three mode,including two faults. Time domain analysis is first used and then bis-pectrum three-dimension graph is achieved by bis-pectrum analysis. The coupling characteristics between fault signals and normal signals are determined through bis-pectrum coherence analysis of the frequency of the system. The bispectrum value is the input of obtention of spectrum entropy. Finally,according to band entropy theory combining with the computation model of bis-pectrum entropy,this paper gets the bis-pectrum entropy at the full frequency domain,and the result was divided into eight frequency bands. Then this paper treats the entropy in eight frequency bands of five groups as characteristic quantity,and inputs these characteristic quantities to input layer of BP neural network, that is regarded as learning sample and determinant sample to be learned and recognized.By analyzing the experimental results about identified the test sample of 45 groups,it can be concluded that the structure of the network can be determined through constant adjustment BP neural network structure. Finally,there are 38 groups identified correctly,and five groups errors. The remaining two groups ambiguity recognition, so the accuracy rate can reached 84.4%. Taking into account the possibility of existing the sampling error and calculation error,so this paper ultimately determine the higher order entropy spectral bands can be used as a reference feature amount of fault identification,and it can successfully identified the malfunction.