Research On The Application Of Metal Oxide Arrester Fault Detection Based On Infrared Imaging

Lightning arrester is an important equipment of power system safe and stable operation, which is an important guarantee to avoid impact of lightning over-voltage and operating over-voltage harm to power grid security. Because of its excellent volt-ampere characteristic and protection performance, the metal Zinc Oxide arrester(MOA) has replaced many other products of the same type, and has been widely used in power grid. At present, the online monitoring technology of MOA is mainly by detecting the size of resistive component in leakage current to determine whether the fault occurs. Compared with the existing detection technology, infrared detection technology has the advantages of non-contact, easy operation, high coordination.This paper studies new methods for fault identification of MOA, which can efficiently and accurately determine whether the presence of MOA faults and defects exist.First, the fault characteristics of MOA and the infrared thermal imager error are analyzed, and this paper summarizes the basic conditions infrared detection need to meet and the methods and principles of the infrared detection of MOA. After collecting infrared image, this paper uses the median filter to preprocess the image, and then uses the Otsu method to extract the binary image. At last, based on the shape feature of the upper and lower ends of the MOA, we filter the bottom base and top cover parts to extract the positions of the arrester in the image.After the completion of infrared image pre-processing and segmentation, this paper puts forward two fault diagnosis method of MOA.1. After extracting the MOA position information in the image, this article intercepts arrester's binary image inscribed rectangle, abandons the arrester edge part, then extracts temperature information through the internal rectangular and calculates the maximum temperature and average value. Based on ?Specification for infrared diagnosis of charged equipment?, this paper puts forward a fault identification method combining maximum temperature and average temperature, which is very accurate and easy to operate, and can effectively overcome the false positives caused by the occlusion of the umbrella group and the large surface temperature difference.2. After the segmentation, this paper selects the image texture feature as the feature value. Using gray gradient co-occurrence matrix to extract 13 texture features of the image of the target area and then applying K-L transformation method for feature selection. After removing redundant characteristic value, this article chooses small gradient, gradient distribution inhomogeneity, uneven distribution of gray level as input variables of the neural network. The recognition model of metal MOA based on improved BP neural network is obtained through training. Then selecting the data to test the mode, results show that the method is accurate and effective, and is an effective technique for MOA condition monitoring of substation.