Transformer Ignition Point Monitoring System Based On Infrared Technology

With the continuous expansion of the overall scale of the power grid,transformers are often subjected to long-term overload operation in harsh environments that can lead to frequent transformer fires as important equipment in the power grid.Transformer fires can cause not only cause equipment damage but also explosion accidents and casualties in severe cases because the characteristics of uncertainty and rapidity.Infrared thermal imagers can present the temperature state of equipment non-contact and have been widely used in the field of power equipment diagnosis.Therefore,this thesis proposes a transformer ignition point monitoring system based on infrared thermal imaging technology,which is used for timely detection and early prediction of transformer fires,and has significant significance in reducing economic losses and ensuring personal safety.The main research content includes:(1)This thesis clarified the common types of faults in transformers and the reasons for transformer fires,analyzed the technical difficulties of the transformer ignition point monitoring system based on infrared image technology,and summarized the overall system design process.Due to the presence of noise interference and low contrast during the imaging process of infrared images,this thesis adopts three filtering algorithms to denoise infrared images based on their noise characteristics.The experimental results show that the median filter can more effectively suppress the noise signal in the infrared image and improve the image quality.On this basis,the Contrast Limited Adaptive Histogram Equalization(CLAHE)algorithm is used to improve the image contrast and enhance the detail information,which lays the foundation for the subsequent better extraction of target features.(2)This thesis introduces the two manifestations of transformer ignition points: the presence of flames and abnormal high temperatures,analyzes their causes and dynamic characteristics and uses Gaussian Mixture Model(GMM)background modeling to extract the motion features of the ignition point to determine whether there is a suspected ignition point area in the image,And the image is put into the YOLOv5s-SE network for precise identification and positioning of fire points,which can effectively remove static background interference and improve network accuracy.(3)In order to make YOLOv5 s pay more attention to the ignition point area,the SE attention mechanism was introduced into the backbone network.The experimental results showed that the improved YOLOv5s-SE network effectively improved the average accuracy m AP by 96.0% that was 3.3% higher than the original YOLOv5 s and reduced missed detections.Its FPS was 95.8 frames per second and enabled real-time monitoring.Finally,the transformer ignition point detection system was built to achieve real-time monitoring and display as well as identification and alarm of transformer ignition points.