Research Of Extracting Fault Model Based On Wavelet

Advanced fault diagnosis technology is a prerequisite for the development of equipments. It will be impossible with no fault diagnosis technology. It not only provides power for equipment development, but also for the device to provide security and avoid loss. In the technology of fault diagnosis, the most important is fault feature extraction. Wavelet analysis is a new type of it.Wavelet analysis has been widely applied in extracting characteristic patterns of high-dimensional data. It not only can reduce data dimension, but also can extract the data signal excitation characteristics through data decomposition in time domain and frequency domain. However, wavelet analysis in the fault pattern abstract application has four main aspects:the wavelet basis function, the wavelet decomposition level, wavelet coefficients selection and the selection of feature generation algorithm. These four aspects directly restrict the wavelet analysis to extract the fault mode advantages and disadvantages, and affect the final circuit board fault diagnosis rate. On the study of wavelet, this paper presents an evaluation standard of wavelet to better plays its characteristics.According to the needs of the time domain and frequency domain, wavelet analysis can pull away or closer the lens of the microscope, so as to achieve the purpose of extracting signal failure characteristics. So it's more suitable for extracting the contour and singular information. The paper describes three feature extraction algorithm and BP neural networks, support vector machine, information fusion technology. Base on this, the volatility function, signal to noise ratio, time complexity, the accuracy rate are put forward.Select the logic output control circuit as a test circuit board. According to the working principle of the circuit and common fault types, choose38circuit state, and25sampling node to collect signal data of three times. The fault diagnosis system is composed of wavelet analysis for feature extraction, neural network and support vector machine for classification. It verified the evaluation of wavelet analysis.