| Welding technology is an important link in industrial production,weld identification technology as an important link in modern welding technology,has been widely paid attention to,intelligent welding technology is also one of the important trends of future development.However,there are still some problems such as low accuracy and noise interference,which need to be further studied and explored.In complex industrial production environment,due to the impact of industrial field environment such as arc light and smoke,the accuracy of weld identification is greatly reduced,which is prone to problems such as poor real-time performance and low accuracy,so as to obtain inaccurate weld information,thus affecting welding quality and reducing welding efficiency.In addition,due to the lack of efficient weld trajectory planning method,the weld automatic identification effect is poor in the welding process.Therefore,how to effectively improve the accuracy and effectiveness of automatic identification of welding seams in the welding process has become an important topic.This topic mainly carries out the following work:Aiming at the problem of low precision of laser fringe segmentation of welding seams caused by large amount of arc noise in complex welding environment,an improved U-Net robust weld recognition algorithm incorporating attention mechanism was proposed to build a deep learning model and improve the accuracy of feature extraction.Firstly,the super channel attention mechanism is used to achieve the weighted fusion of features.Then,after the encoder structure,a feature classification structure is added so that it can output the weld corresponding type name.Finally,since the imbalance of positive and negative samples in network training would affect the recognition results,Dice Loss and Focal Loss were added to the loss function of the model to modify it to improve the robustness and generalization of the model.Therefore,in the environment with arc light,smoke noise and other interference,this subject has obtained good experimental results,which can meet the requirements of accuracy and real-time detection,and has certain application prospects in the actual welding field with arc smoke and other interference.For discontinuous welds and continuous welds,feature loss is caused by smoke interference.Firstly,Steger laser fringe extraction algorithm is selected to accurately extract the center line of continuous and discontinuous welds.Then,an improved FAST corner detection algorithm with adaptive threshold is proposed to extract discontinuous welds,which can shorten the extraction time of feature points After that,for the continuous welds with occlusions,in order to solve the problem of occlusions,the lack of weld feature points caused by the welding fringe cannot be completely extracted from the network model.In this paper,the weld line segment is fitted with the Hough straight line detection,and the intersection point,namely the weld feature point,is calculated Subsequently determine the weld trajectory planning method to provide data support.In order to plan a smoother welding trajectory,D-H modeling of the welding robot was carried out first,the pose description and homogeneous transformation were carried out.After that,an interpolation algorithm based on NURBS spline curve was proposed to deduce the expressions of its velocity,acceleration and impact,and finally the butt type was adopted As an example,the joint Angle corresponding to each welding spot was calculated,and the NURBS spline curve interpolation fitting was carried out for each joint Angle,and the fitting results were analyzed. |
