Investigation On Ultrasonic Detection And Pattern Recognition Of Partial Discharge In Liquid Insulation

The partial discharge is one of the major failure mechanisms for insulationof power equipment. In order to monitor the service conditions of insulatingmaterials, it is very important to investigate on partial discharge characteristic inelectrical apparatus. So in order to prevent occurrence of serious accidents inpower system, it is important to study partial discharge detection and recognitiontechnology. The acoustic emission technology is not interference byelectromagnetic interference, what is more important, this method can realize thelocalization of partial discharge source. But the waveform of acoustic signal isvery complicated and easily influenced by the transmission path. After studyingcharacteristic of partial discharge signal deeply, this paper presents a methodbased on wavelet transform and neural network to classify different types ofpartial discharge signal.According to the characteristic of partial discharge, the pin-plane, plane-plane, surface discharge, sliding discharge and floating discharge model isdesigned. The acoustic detection system is built after finishing the design ofdischarge model. A software based on Labview platform is designed to detect thepartial discharge in oil insulation. In order to get the most appropriate location ofsensor, the comsol software is used to simulate the acoustic pressure of partialdischarge. In order to get the characteristic of partial discharge, the waveletdenoising technology is used to decrease the influence of noise. Then the waveletenergy distribution is used to classify pin-plane, surface discharge, slidingdischarge model. In the end the modified BP network is used to classify plane-plane and surface discharge model, the classifying result is good.After the above analyzing, some results can be gotten. The using of wavelet denoising technology can effective erase the effect of environment noise. Thecombining of wavelet distribution and modified BP network can effectivelyclassify pin-plane, plane-plane, surface discharge, sliding discharge and floatingdischarge model, which cannot be realized by using the traditional BP network.