An Autonomous Approach To Condition Diagnosis Of HV Cable Accessories Based On Infrared Images

High-voltage cables have been widely used in power systems due to their excellent electrical,mechanical properties and environmental friendliness.Cable accessories are the weak links of the cable systems.The temperature of defective cable accessories is usually higher than that of the cable accessories under normal conditions.Therefore,the temperature measurement is an effective approach to condition diagnosis.Infrared thermography has many advantages,such as no physical contact,non-intrusiveness and high efficiency.Therefore,it has been widely used in regular inspection of HV cable accessories.However,the diagnosis based on infrared images has mainly relied on visual inspection.This is time-consuming and laborious on one hand,and on the other hand,it relies too much on expert expertise and is prone to erroneous diagnosis.Therefore,it is important to develop an autonomous approach to condition diagnosis of HV cable accessories based on infrared image processing.Infrared images have the characteristics of high noise and low contrast.In order to remove noise from the original infrared images,a local adaptive wavelet threshold denoising method,which based on maximum a posteriori estimation(MAP),was proposed.Experiments demonstrated that the proposed method was superior to the traditional wavelet threshold denoising method in terms of mean square error(MSE)and peak signal-to-noise ratio(PSNR).The proposed denoising method can further improve the denoising performance and the quality of infrared images.After achieving image denoising,in order to eliminate the influence of other interference information in images,an autonomous identification and localization method,which based on Faster RCNN network,was proposed.The Faster RCNN network consisted of the Convolutional Neural Network(CNN),the Region Proposal Network(RPN),Region of Interest(Ro I)pooling layer and classification layer.The infrared images,which had been captured during routine inspection activities,were used as samples to train the Faster RCNN network.After completing the training,the Faster RCNN network can realize autonomous identification and positioning of the cable terminations,grounding boxes and GIS terminations in the images under processing.After extracting the diagnostic objects,in order to assess the condition of HV cable accessories,it is necessary to extract the suspected overheating regions,reference regions and their temperature information.This paper used the Mean-Shift clustering algorithm to segment the images,so that the overheating regions could be extracted.Then based on the overheating regions,different recognition methods of reference areas were proposed for different HV cable accessories.Finally,according to the corresponding diagnostic criteria,the results of diagnosis could be obtained.The method proposed in this paper can accurately locate the overheating regions and normal operating areas of the diagnostic objects,and automatically divide the condition of the cable accessories into four classes,which including normal,general defect,major defect and emergency defect.In summary,this paper proposed an autonomous approach to condition diagnosis of HV cable accessories based on infrared images.This method included image denoising,location and recognition of diagnostic objects and temperature information extraction of interested regions.The proposed approach may potentially be productive as it helps reduce the dependence on human efforts and expertise and helps improve the practice of condition monitoring.