In Situ Studies Of Full Field Strain And Residual Stress Of Welding Heat Affect Zone Using Digital Image Correlation

In the practical application process of automatic robot welding automation,seam tracking and weld pool monitoring are the two key problems to be resolved,in the existing technologies of vision sensing,relevant commercial products are now mostly used for seam tracking,and the core technology is monopolized by a few foreign companies,thus the price is extremely expensive and makes large-scale promotion hard to carry out.In addition,because the laser vision sensor can not accurately obtain the image information of welding pool,it can not be used in the fusion control.Therefore,develop a practical,cheap and high-performance vision sensing system which can realize these two functions at the same time with completely independent intellectual property rights is particularly important,and is of great significance to break the foreign technical barriers,realize localization,lower the prices and improve the research of intelligent robot welding technology.In this paper,a practical new 3D-printed vision sensing system combining the laser active vision and passive vision in the way of single binocular was independently developed,simultaneous acquisition of laser scanning information of weld and dynamic image information of welding pool can realized by only one vision sensor,furthermore a set of robot welding system combining the function of seam tracking and weld pool monitoring is built.By using this active and passive visual sensing system,the Steger weld image processing method based on the Hessian matrix aimed at the working condition of complex weld seam tracking control was studied,central line of laser stripe and characteristic of welding seam can be accurately extracted in the process of robot GMAW,also precision verification experiments of weld seam tracking was conducted towards v-type groove weld,and its maximum error is 0.68 mm while the average error is 0.22 mm,good weld formation can be realized and the basic precision meet the demand of actual welding robot.Real-time monitoring of the welding pool has always been the short slab of slaser vision sensor,this paper,by using the developed the active passive vision sensors,typical images of pool both in robot GTAW and GMAW were separately collected.The collected images were processed by using DCNN and Mask – RCNN.As a result,the edge of pool can be accurately extracted,and positive fusion width and half weld length can be obtained,this is very important for predicting welding pool width on the back of molten pool.The purpose of real-time acquisition and processing of weld pool images is to predict the fusion width of pool on the back side.In this paper,multiple linear regression,support vector regression and XGBoost algorithms are respectively used for modeling and prediction for the weld pool.The results show that the average relative error of the back weld width prediction model based on the XGBoost algorithm is the smallest(6.62%),and its accuracy is higher than the other two algorithms.Therefore,the XGBoost algorithm is selected for the prediction model of the back weld width of the passive visual sensing system.