Research On Target Identification And Localisation Methods For Insulator Replacement Robots

Insulators are exposed to acid rain,high voltage fields and lightning strikes for long periods of time,which can lead to degradation of their electrical properties and insulation performance,resulting in breakdown or damage.The degraded and damaged insulators need to be replaced in a timely manner,otherwise serious economic losses and casualties can be incurred,and in some cases the stability of the power system can be threatened.The development of insulator replacement robots is an urgent need in the power system due to the safety risks and the difficulty and intensity of traditional manual replacement methods.The identification and localisation of multi-attitude targets under complex outdoor light disturbances is a key issue in the development of insulator replacement robots,and is the focus of this paper.This research is supported by the Southern Power Grid Key Project and the Guangxi Zhuang Autonomous Region Key R&D Project "Research and Development of a Robot for Insulator Inspection and Replacement".Firstly,through literature review and field research,the theoretical knowledge about insulator replacement robot and insulator pin identification and positioning methods is acquired,the existing research hotspots and difficulties in the target identification and positioning methods are studied in depth,and the background significance and key research problems of this topic are clarified.Secondly,the YOLOv5 algorithm is improved and the YOLOv5-n S algorithm is proposed to address the difficulties and real-time problems of multi-attitude insulator pin recognition under complex outdoor light perturbations in the rim replacement robot.The main optimisations are as follows: firstly,the Backbone part is reconstructed using the n SNet model proposed in this paper;secondly,a feature fusion layer is added to the Neck end and a light weighting improvement is made;again,the loss function of the boundary regression frame is improved by using the Alpha-Io U loss function;finally,the K-means++ algorithm is improved based on the Anchor scale reselection strategy is improved based on the K-means++ algorithm.At the same time,the deep learningbased network model optimization training method was studied and analysed,and the YOLOv5-n S algorithm model was trained.The final experimental results show that the YOLOv5-n S algorithm model can quickly and effectively identify insulator pins with multiple poses in different lighting scenarios,is not easily affected by complex lighting and multiple poses,and has good robustness.Then,after in-depth research and analysis,a vision system for an insulator replacement robot consisting of an industrial camera and LIDAR is designed.The transformation relationship between the coordinate systems of each device is modelled for the connection between the industrial camera,the Li DAR and the robot.A method for locating insulator pins based on the fusion of industrial camera and LIDAR information is proposed,which effectively compensates for the inability of a single industrial camera to accurately acquire depth information.Finally,an experimental platform is built to simulate the insulator replacement task,and the reliability and effectiveness of the identification and localisation algorithm based on the fusion of industrial camera and Li DAR information are verified and analysed.In addition,the reliability of the proposed algorithm is also verified in a real insulator replacement task.