Research On The Motor Fault Diagnosis Based On Wavelet And Fuzzy Neural Network

Asynchronous motor has a simple structure, which is not expensive in the practical application. It has stable operation in relatively compared with other products, therefore,plays a decisive role in our daily life. Motor fault leads to unstable operation caused economic losses, and even endanger the personal safety due to the influence of the mechanical load, the working environment, and other aspects of the motor itself. So, people put a lot of effort to study asynchronous motor at the same time in the study of the fault diagnosis. With the development of computer has been widely used in signal processing, fault diagnosis method of motor has made great breakthrough in the efforts of the people. In this paper, proposes the asynchronous motor common fault diagnosis method, studys the wavelet packet and fuzzy RBF neural network combined with genetic algorithm, optimization based on reading a lot of Chinese literature and English literature.The vibration signal can reflect the running state of the motor, and wavelet packet analysis has high resolution. Based on the wavelet energy feature extraction, the main research contents of this paper are as follows:Research on energy feature extraction of signals, choose three wavelet packet base that commonly used experiences, uses these wavelet packet base to analysis the signals, the peak-to-peak value eliminates the energy, select the maximum value which is db3, use it to do three layer wavelet packet decomposition can obtains the energy characteristic vector.Different vibration signals basically is different vibration energy, after wavelet packet decomposition the signal, each layer of the signal power spectrum is different. According to the coefficient of power spectrum to judge different faults.Research on the asynchronous motor fault diagnosis method combine wavelet packet and fuzzy RBF neural network. The method using wavelet packet analysis of vibration signal processing and extract energy feature vector used as training samples and testing samples.Regards the degree of failure as a fuzzy concept, and fuzzy the input signal. The hidden layer nodes get the number according to the empirical formula, and train them one by one, and then choose the one in the output with small error and the short time. Because the parameters of the network is not easy to determine, compare the three kinds of basis function center selection method, the final choice of the supervised method combined with the gradient descending method. Train the network with training sample, and do the test of the test sample after training.Research on the asynchronous motor fault diagnosis methods based on the wavelet packet and genetic optimization and fuzzy RBF neural network. In order to solve not let the decline method fall into a minimum problem when selects the basis function center,combined the genetic algorithm and the above algorithm together, use the same network structure. The number of hidden layer nodes through the choice of values of the less. The method uses network parameters such as genetic algorithm and descent method to optimizethe parameters of the network on the basis of last method, which avoids the minimal value,reduces the number of hidden layer nodes, shortens the time to convergence, and accuracy is also improved.