Research On Detection Technology And Visual Guidance System Of Ship Component Line Based On Deep Learning

With the comprehensive development of intelligence,intelligent machines gradually replace manual operation,automatic welding has become an important research content.Workpiece weld extraction plays an important role in automatic welding.In this paper,the edge detection algorithm is studied and improved to improve the accuracy of workpiece weld and contour detection.Aiming at the research status of automatic welding technology,a visual guidance system of automatic welding based on image feature extraction and 3D model matching was studied.The main research work is as follows:(1)In view of the characteristics of weldment image and the problems of low detection accuracy of existing algorithms on weldment image,an edge line detection algorithm based on skip layer decoding fusion is proposed.This algorithm is improved based on RCF network,which combines void space convolutional pooling pyramid with hop layer decoding fusion module to fully mine the semantic information of network.In addition,the attention mechanism module is introduced to make the network more focused on the learning of weld and profile characteristics of weldments.The experimental results show that,compared with the RCF algorithm,the proposed model improves by 0.020 on ODS and 0.013 on OIS,respectively,for the data set of welds and profiles,which indicates that the proposed model has higher accuracy in the extraction of welds and profiles of welds.(2)Aiming at the current situation that it is difficult to realize automatic welding for large workpieces and large-scale production lines,a visual guidance system for automatic welding based on feature extraction and 3D model matching was studied.To solve the problem that it is difficult for a single camera to cover large workpieces and large production lines,the field of view features of multiple cameras were combined.Aiming at the hand-eye calibration error caused by the Angle between the direction of the gantry and the coordinate axis of the robot,mathematical compensation method was used to erase the Angle error.Aiming at the problem that it is difficult to extract complete and accurate welds with the existing technology,a complete and high-precision welds can be extracted finally by identifying various features of the workpiece and registering them with the standard workpiece model.The actual verification shows that the welding seam error extracted by the system is less than 4mm(including ± 1mm deviation between the workpiece and the model),and the system has high efficiency and strong generalization ability,which can be stably applied in the actual industrial welding environment.