Research On Key Techniques Of Vision Control For Catenary Insulator Cleaning Robot

Insulators are important supporting and insulating devices in catenary of electrified railways.Recently,with the rapid construction of electrified railways in China,the number of catenary insulators has been increasing.However,the surface of insulators is very easy to be polluted because of long-term exposure in outdoor and the influence of haze,dust and other effects.With the increase of pollution degree,it is easy to induce pollution flashovers,which can cause serious accidents such as power outages,and bring safety hazards and economic losses to railway operations.The main way to solve this kind of problem is to clean the insulators by hand regularly,which has some limitations such as low efficiency,high strength,poor safety and so on.In recent years,with the rapid development of artificial intelligence technologies,the research on cleaning robot for catenary insulators has been attracting more and more attention.Therefore,the dissertation mainly focuses on the visual control technologies of insulator cleaning robot,and research on key technologies of insulator identification,contamination detection,spatial positioning,robot visual servo control,etc.At the same time,a test platform for insulator cleaning robot is designed and developed,on which the design,development and deployment test of visual control related algorithms are completed,and some factors are also considered such as the equipment cost,the real-time performance of algorithms and so on.The dissertation is organized as follows.Firstly,the common neural network models for target detection are compared and analyzed.In order to ensure the real-time performance of the algorithm on ordinary PC or embedded platform,the lightweight model of Mobilenet-SSD is employed as the basic model of the insulator identification.The identification experiment of insulator is accomplished through datasets making,model training and model verification and other processes,on which the average precision can reach 83.36%.To improve the detection accuracy further,the MobileNetSSD-FF model is proposed based on the improvement and optimization through three feature fusion schemes,on which the average precision can be increased to 86.11%in the basically same detection speed.By this way,the catenary insulators can be accurately identified under complex working conditions.The type and location information of insulators can be obtained.Secondly,a scheme is proposed to judge the type and degree of the contamination based on insulator segmentation algorithm and the contamination features on the surface of the catenary insulators.The lightweight neural network model Mobilenet-DeepLabv3 and image post-processing technologies are employed to accomplish segmentation of insulators through data annotation,model training and model verification and other processes.In the HSV color space,the type of the contamination is judged by extracting the color features of the segmented insulator surface.And the degree of the contamination is judged through using ROI(region of interest)to extract the uniform contamination area on the insulator surface,combining with histogram analysis and similarity calculation.By this way,the problems of contamination detection on the surface of the catenary insulators can be solved under complex working conditions.Thirdly,a matching method based on center feature points of prediction frame of MobileNet-SSD-FF network is proposed to solve some problems of traditional stereo matching algorithm,such as complex process,large amount of computation,difficult to fuse with neural network algorithms and so on.Through camera calibration,target recognition,stereo matching,three-dimensional coordinate calculation and other processes,the central position information of the target insulator is obtained.Based on this,the cleaning range of the robot can be planned according to the actual size of the insulator,which provides data support for the control of the cleaning robot.Fourthly,an insulator cleaning manipulator is designed in the research of the visual servo control.The robot controller based on FPGA chip is designed by using design method based on Simulink model.A circular interpolation algorithm based on three points in space is proposed,and a series of driver functions are developed.A position-based visual servo control scheme for insulator cleaning robot is designed,which solves the problems of limited interface and inconvenient secondary development of the existing visual servo systems.Finally,a test platform of insulator cleaning robot is built,which includes six-degree-offreedom robot,visual servo control system,lifting platform,mobile chassis,water spray device and so on.The Algorithms of insulator identification,insulator contamination detection,insulator binocular vision location and robot servo control are designed and deployed.Web control software compatible with PC platform and embedded platform is programmed.The visual control function of the cleaning robot of insulators is tested in the laboratory environment,and the regular cleaning and precise cleaning tests of the contaminated insulators are also carried out.The test results show that the visual servo system designed in the dissertation can basically meet the real-time requirements,and the overall design has achieved the desired goal.