Research On The Key Technology Of Quadruped Robot Substation Inspection Based On Machine Vision

As an important place in the power system,substation is responsible for the exchange of voltage and current and the distribution of electric energy,which relates to the power demand and economic development of the whole country and society.Therefore,it is particularly important to develop a comprehensive regular inspection of the substation.With the continuous development of robot and computer vision technology,intelligent robot provides a new idea and opportunity for substation inspection.However,transformer substation inspection robots mostly use wheeled robots as moving mechanisms.Due to the motion characteristics of the wheeled mechanism,the robot is not strong in adapting to the terrain of the substation,so it is difficult to realize the inspection task in the complex terrain.Special inspection roads must be laid,which greatly increases the inspection cost.Quadruped robot has good terrain adaptability.However,due to its motion characteristics of multiple degrees of freedom and complex structural characteristics,it often has the problem of poor positioning accuracy,resulting in poor quality of the acquired image and low inspection efficiency.Robot substation inspection mainly relies on a variety of image sensors to collect image information,through the analysis of the image to determine whether there are abnormal points.With its strong learning ability,object detection algorithm based on deep learning has gradually become the mainstream method of object detection and recognition.However,there are some problems such as large model and low real-time performance,which are not conducive to deployment on mobile devices.Based on the above background,this paper carries out the research on the key technologies of robot inspection for quadruped substation based on machine vision technology.Based on the research status of substation inspection robot technology,it is determined to use quadruped robot instead of wheeled robot to realize substation inspection task.On this basis,the rapid positioning of substation equipment and the control technology of gimbals correction are researched,which solves the technical difficulties of obtaining high quality equipment image caused by inaccurate positioning of quadruped robot.The research of substation equipment detection and identification technology based on machine vision is carried out to solve the problems of low detection accuracy,poor real-time performance,and not suitable for deployment on embedded intelligent hardware.At the same time,experiments are designed to verify the substation inspection system based on quadruped robot.The main research work is as follows:.1)We completed the overall system scheme design of the quadruped inspection robot.In order to meet the requirements of substation electrical equipment inspection,related hardware selection and software system design were carried out.2)We carry out research on rapid positioning and correction technology of substation equipment.we established the kinematics model of the cradle head,using the improved fast positioning algorithm based on Single Shot Multi Box Detector(SSD)to locate the devices to be detected in the camera’s field of view,using the kinematic model of the cradle head to carry out coordinate changes,and controlled the cradle head to adjust the device to be detected to the center of the camera field of vision,so as to realize the cradle head deviation correction control.3)We carried out research on substation target detection and recognition technology based on convolutional neural network.We designed Adaptive Multi-head Structure and Lightweight Feature Pyramid Network(AHLNet)for substation multi-target detection.In the feature extraction stage,we designed the adaptive multi-structure network(AMHNet).In the feature fusion stage,we designed a lightweight feature pyramid network(LFPN)and introduced the spatial attention mechanism(SA).Through the above improvement,the design of target detection network is improved,which is light weight,simple and high precision.