Design Of Railway Automatic Unhook Robot And Visual Inspection Of Couplers

At present,domestic freight train marshalling stations mainly use manual operations to achieve hump picking.However,with the increasing growth of railway freight,the disadvantages of manual picking are low efficiency and poor safety.Based on the existing hook picking robot,a new design scheme is proposed.This solution realizes the problem of recognition and classification of couplers by hooking robots by adding a computer vision module.Compared with traditional coupler recognition methods,it has the disadvantages of low automation,low recognition accuracy,and poor environmental adaptability.The recognition classification can fully meet the design requirements.After a series of optimization and debugging of the algorithm,the accuracy,response speed and fault tolerance rate of the coupler recognition classification have reached a high practicality.The main research contents of this paper are:Structural design part.According to the actual situation of the hump picking platform,the mechanical structure part of the track-type hook picking robot is designed.This part includes the motion walking module,the visual detection of the coupler,the posture adjustment module and the end the execution module.This part adopts SolidWorks software entity modeling,and then import into ADAMS for kinematics analysis by adding relevant constraints and driving conditions,especially the analysis of the kinematics of the hook manipulator module,to obtain the motion simulation results of key research components and draw the kinematics curve,Analyze the stability of the motion state of the automatic hook picking robot and the feasibility of the structural design of the hook picking robot during the hook picking process.Coupler identification detection section.Firstly,several target detection algorithms that are widely used are introduced.The Faster R-CNN algorithm and YOLO_V2 are selected as the experimental basis.The widely used open source library is used to improve the algorithm to a certain extent.The improvements include Using the K-means clustering algorithm to adjust and optimize the number of anchors,find the optimal number of anchors sui Tab for this coupler identification and the width and height dimensions of the target frame;optimize the loss function and improve the algorithm model in harsh environments Factors,the ability to recognize couplers under different light conditions;the algorithm structure is rebuilt and optimized by adding residual recognition effects;finally,Faster R-CNN,YOLO_V2 and The improved YOLO_V2 algorithm performs comparative experiments on the accuracy rate,recall rate and real-time detection speed of the coupler identification.The final experimental results show that the improved YOLO_V2 algorithm proposed in this paper can effectively meet the requirements of coupler identification and classification,and is superior to traditional positioning methods and other Target detection algorithm.The last part summarizes the problems encountered in the study and research process of this subject and the related solutions,and at the same time prospects for the future application of computer vision to coupler recognition.