Study On Bearing Fault Diagnosis Of Motor By Combined Wavelet Fractal With Neural Network

In modern production life, people increasingly pay more attention to Machinery fault diagnosis technology. Rolling bearing is the most common components in the Machinery and equipment, as the most common weak link, whose state has a direct impact on the security status of the machine. The key to fault diagnosis is to research the bearing vibration signal generated during Rolling bearing's running.In this paper, we take rolling bearing as the research object, mainly research on the vibration mechanism and fault characteristics of the bearing, and introduce several common fault types of motor bearing. All the tests are based on collection of typical fault vibration signals.Neural network fault diagnosis of rolling element bearing can reduce the request of expertise of the operator, making the fault diagnosis of rolling element bearing increasingly intelligent.This paper presents a bearing fault diagnosis method, which is based on the wavelet fractal theory and neural network PNN. PNN neural network has many advantages as follows:simple training process, fast convergence, less parameters to be adjusted, and compared with BP neural network, its local minimum does not exist.There are many different ways of Fault Feature Extraction based on Vibration Signal Analysis, but each method only reflects one aspect of the failure characteristics. Take energy values extracted by wavelet packet as training samples of neural networks, network effects of convergence and diagnostic accuracy rate are not very satisfactory. Fractal theory is emerging in recent years, a subject which studies the irregular and nonlinear systems in the nature, therefore fractal dimension can effectively describe the complexity and irregularity of the vibration signal.In order to make the extracted value better reflect the state features of the rolling bearing, this paper studies two methods of extracting signal characteristic value which are wavelet theory and fractal dimension.One of Fractal dimension is the correlation dimension, which is sensitive to the attractor irregularity, hence it is easy for it to reflect the dynamic structure of attractors. Besides, the correlation dimension is easily measured directly from experimental data by a simple algorithm of high reliability. Therefore the correlation dimension is adopted to extract another eigenvalue of rolling bearing fault signal.Rolling bearing fault signal, firstly denoised by wavelet packet, adopts wavelet packet's multiple decomposition and recon extracted energy value from different frequency bands. And at the same time, the denoised signal correlation dimension value of rolling bearing is obtained by using the solving method of correlation dimension. The using of wavelet packet theory and correlation dimension method is studied in rolling bearing fault diagnosis.Finally, we use the correlation dimension value as well as signal energy value as neural network input, the results show that the method is feasible and has better diagnosis through the rolling bearing fault diagnosis example.We also use a ActiveX technology to help carry out the mixed program between VB and MATLAB to improve the ability of human-computer interaction, to further simplify the fault diagnosis.