Research On The Technology Of Faulty Insulator Identification Based On Infrared Thermal Images

The porcelain insulators sometimes operate under harsh environment conditions such as high temperature,rain and snow.Long term mechanical and electrical forces cause insulators' insulation and mechanical properties to fall easily,thus forming d eteriorated insulators.In recent years,with the progress of infrared thermal image technology and the use of portable thermal imager,the fault diagnosis technology based on infrared thermal image has been used more and more widely in the fault identification of transformer,insulator and high voltage circuit breaker,while its application level is still in the primary stage,and there are many problems,such as qualitative detection,imperfect detection standard and blind areas of detection.In view of the shortcomings of the infrared detection technology at present,this paper proposes an insulators identification model based on the convolution neural network,which combines the simulation analysis and the t est verification.According to the equivalent circuit and temperature rise model of insulator string,the electric network method is used to simulate the temperature rise law of insulators,which is used to clarify the blind areas phenomenon in infrared detection,and to discuss the correlation between the blind areas range and the environment temperature,wind speed,the position of the insulators and the measurement error.In order to solve the problem of impulse noise and Gauss noise in the original infr ared image of insulators,the adaptive median filter is used to denoise the image.The angle correction of ins ulators is corrected by Hof transform,and the single insulator region is segmented by projection statistics.On the basis of image processing,an insulators identification model based on convolution neural network is established.The infrared image after image processing is pretreated with background temperature homogenization,size normalization and gray linear enhancement.The pre processed image is directly used as the input of the model,avoiding explicit feature extraction and learning from the training data implicitly,so the extracted image texture features can reflect the essence of the image more.Finally,the model is trained and verified with the infrared image collected by the test and field.The test results show that the existing conventional infrared detection method for insulators has a blind detection area.The scope of the blind area will be affected by the environment temperature,wind speed,the position of insulators and the meas-urement error in varying degrees.The Identification model of deteriorated insulators based on the convolution neural network can effectively identify the deteriorated insulators,which can provide effective technical support for the electric power inspection department to detect deteriorated insulators.