The Design Of A Diagnostic System For Overheating Defects In Composite Insulators Based On Infrared Thermography

Infrared thermal imaging technology has been widely used in the field of power equipment defect diagnosis and is also one of the key projects in the daily inspection of composite insulators.However,there are two main shortcomings in the current infrared diagnosis of insulator defects:(1)It heavily relies on manual judgment and analysis,resulting in low efficiency when faced with a large amount of infrared data and inability to achieve onsite diagnosis,leading to poor timeliness in maintenance;(2)For suspected defective insulators,the usual strategy is direct replacement,resulting in significant waste of manpower and material resources.To address these challenges,this paper is based on a large number of on-site collected composite insulators and their infrared thermal images,and takes the perspectives of defect diagnosis and maintenance diagnosis.With the aim of improving the automation and real-time level of the model,a composite insulator overheating defect diagnosis system based on infrared thermal imaging is designed.Firstly,to address the problem of high noise in infrared thermal imaging,an improved median filtering algorithm is used to preprocess the enhanced dataset of insulators.The causes and mechanisms of composite insulator overheating defects are studied through finite element analysis and testing experiments,with a focus on three typical overheating defects including core rod decay,surface area pollution,and moisture at the high-voltage end.Targeted diagnostic and maintenance recommendations are summarized.The infrared thermal imaging characteristics of different defects are revealed,and the relative area and position are proposed as effective feature parameters for insulator infrared thermal imaging analysis,laying down technical indicators for insulator defect diagnosis.Secondly,an improved cascade of the YOLOv5(You Only Look Once version 5)object detection model and the Deeplabv3+(Deep Lab version 3 Plus)semantic segmentation model is used to globally locate all insulators in the original image.The predicted boxes of insulators with an area proportion less than a certain threshold are extracted from the original image for pixel-level segmentation,solving the problem of ineffective diagnosis of some overheating defects due to small pixels.In the global object detection stage,the K-means++ clustering algorithm is introduced to fit the prior knowledge of insulators,resulting in a 2.65% increase in recall rate.In the partial insulator segmentation stage,the ECA-Mobile Netv3 is used as the backbone network of Deep Labv3+ to significantly improve the real-time performance of the model.The dilated convolution in the Atrous Spatial Pyramid Pooling(ASPP)is replaced by separable convolution to improve the inference speed.Furthermore,to overcome the problem of blurry object edges in infrared thermal imaging,a Point Rend module is added to optimize the sampling distribution,achieving an average Intersection over Union(Io U)of 86.36%.Finally,to address the issue of insufficient samples of overheated defects in composite insulators,an artificial defect method was utilized to augment the samples of core rod decay defects,while a style transfer algorithm was used to augment the overheated defect dataset.Additionally,multi-angle feature extraction was applied to the identification area,and an optimized extreme random forest algorithm was employed for feature selection.The algorithm determined the three most distinctive features,relative position,relative size,and relative energy,as diagnostic indicators for overheated defects.An automatic diagnosis model for overheated defects in composite insulators was designed for on-site diagnosis.It achieved a diagnostic accuracy of 92.8%,a defect detection rate of 96.3%,and a frame rate of 34.6frames per second,thereby demonstrating the feasibility and effectiveness of the method proposed in this paper.A visual intelligent interface for the diagnostic system was developed using Py Qt,which integrated the trained model and diagnostic algorithm code,facilitating onsite defect diagnosis on portable host computers.When combined with diagnostic recommendations for timely maintenance,this system effectively prevents further development and deterioration of defects.