Research On Recognition And Fault Detection Of Typical Fittings In Transmission Line Based On Machine Vision

Transmission lines are mainly used for power transmission and are an important part of the power grid.Most of the lines are operated in the field,and are greatly affected by external forces such as lightning strikes,wind dances,and ice coatings.Faults such as component damage and running positions often occur,which seriously affects the safety of power system operation.With the construction of smart grids,it has become an inevitable trend to use drones,robots and other equipment to replace manual inspection methods.Modern equipment has provided a broad platform for the application of machine vision technology in the field of electric power.This thesis focuses on the identification and positioning methods of stockbridge dampers,tension clamps and suspension clamps for transmission lines.Firstly,based on the cascade classifier,the visual detection of typical fittings of transmission line was realized.According to the characteristics of the target,the HOG features are extracted to train the first-level classifier to obtain the target suspect area.The Haar features are extracted to train the second-level classifier to classify the suspected areas again.The GWO algorithm is used to optimize the classifier parameters.Then,a visual detection algorithm based on SSD networks was proposed.The MobileNetV3 is used to build basic network,and the inverted residual structure is used to build auxiliary network.The K-means++algorithm is used to obtain the clustering center of the labeled box aspect ratio,which is used to improve the default boxes of network.Furthermore,based on this,the thesis analyzes the characteristics and installation rules of stockbridge damper,the key point detection network was trained,and fault detection methods of stockbridge damper position slip and bending based on the key points was proposed.Aiming at the problem of stockbridge damper position slip,the scheme is divided into two cases:hanging multiple objects and hanging single objects.When multi-targets are suspended within the range,the geometric relationship between the two stockbridge damper bodies is mainly used for judgment.When a single target is hung within the range,the actual installation distance is calculated by the EPnP pose estimation algorithm to determine whether it exceeds the threshold.Aiming at the bending problem of the stockbridge damper head,the geometrical constraint between the two damper heads is mainly used to realize fault detection.Finally,experiments show that the cascade classifier can effectively integrate the advantages of single classifiers at all levels,and reduce the false detections.This method can achieve target recognition and location under certain conditions.However,compared with the multi-target detection network obtained by training,the traditional machine learning algorithm has low accuracy,and it still has poor applicability and low efficiency when facing complex scene images.The fault detection method designed in this thesis can accurately diagnose the abnormality and provides new ideas for intelligent detection of transmission lines.