Research Of The Asynchronous Motor Fault Diagnosis Based On The Wavelet Neural Network

As power source, the AC induction motor is widely used in industrial production.Its operation status directly affects the performance of the entire system. Rollingbearing is one of the most crucial elements in motor yet with the characteristics ofhigh damage rate and large lifetime randomness caused by process and environmentalfactors. The fault of motor is various whereas about one third is resulted from rollingbearing fault. Therefore, to handle the working status and the fault formation as wellas development of rolling bearing is currently important topic in the area of motorfault diagnosis.Vibration analysis is an important means of rolling bearing condition monitoringand fault diagnosis. Bearing vibration signals contain a lot of information of operatingstatus, so it is an effective method to diagnose AC induction motor malfunction usingbearing vibration signals. Extracting features from the bearing vibration signals andfault type recognition according to the extracted fault features are the two key topicsin the field of fault diagnosis of bearing. This thesis systematically investigated thefault feature extraction and fault type recognition methods of the bearing applyingtesting technology, wavelet analysis, Hilbert envelope analysis, BP neural network,Elman neural network and RFB neural network. The main research work is asfollows:（1）This thesis has studied mechanism of bearing fault, analyzed time domaincharacteristics and frequency domain features of bearing fault, which can be used forfeature extraction and fault diagnosis of bearing vibration signal.（2）This thesis has studied application of the wavelet analysis in bearingvibration signal feature extraction. The Hilbert envelope analysis and the waveletpacket energy spectrum of bearing vibration signal under different states wereanalyzed, and the results showed that wavelet packet energy spectrum could becharacterized as fault features of bearing failure. When initial failure has happened,the Hilbert envelope demodulation could be characterized as fault features. But itcould not be characterized as fault features, when failure to deepen.（3）Taking into account the time-domain characteristic parameters of bearing vibration signals, this thesis put forward two methods to compose bearing faultfeature vector, including the wavelet packet energy spectrum, the wavelet packetenergy spectrum combined with signal time domain characteristics.（4）Bearing fault feature vector were constructed using the four methods(wavelet packet energy spectrum; wavelet packet energy spectrum combined withtime-domain characteristics;) based on wavelet analysis for two types ofabnormalities of bearing which are the normal operating conditions, Inner ring fault,rolling element fault, outer ring fault. Diagnostic classification has recognised by BPnetwork and Elman network and RFB network. The fault diagnosis results showedthat two types of feature vector constructing methods based on wavelet analysis canbe used in feature vector constructing for a bearing vibration signal. BP networkusing wavelet packet energy spectrum combined with time-domain characteristicsand classification diagnosis using support vector machine can achieve the bestdiagnosis effect. For the same method to feature vector constructing, the effect ofdiagnosis classification using BP neural network is better than using Elman neuralnetwork and RFB neural network.