Research On Welding Quality Monitoring And Early Warning Technology Based On Weld Pool Morphology

The change of weld pool morphology reflects the quality of the weld formation,and contains a lot of information related to welding quality.At present,the monitoring of welding quality is mainly based on the current weld pool morphology,and there is lack of research on early warning of welding quality.In this paper,the morphology of weld pool and its evolution trend are studied.The monitoring method of hump,penetration and the depth of the weld bead based on the current morphology of weld pool and the early warning method of weld quality based on the morphology prediction of weld pool are proposed.The research contents are as follows:(1)Welding quality monitoring technology based on current weld pool morphology:firstly,the molten pool images of hump,penetration and weld bead depth are collected by the vision acquisition device.Secondly,for the qualitative analysis of hump and penetration,Le Net-5 network with channel attention module is proposed to realize the qualitative classification of hump and penetration;for the quantitative analysis of the weld bead depth,Res Net network with global attention module is proposed to achieve quantitative analysis of weld bead depth.Finally,the performance evaluation experiment of the technology is carried out.The results show that the accuracy of the qualitative analysis algorithm of hump and penetration is 99.7%,and the average absolute error of the quantitative analysis algorithm of the weld bead depth is 0.064 mm.(2)Early warning technology of welding quality based on prediction of weld pool morphology: by modeling the changes of the weld pool morphology that have occurred,the video prediction algorithm is used to predict the future images of the weld pool.In order to overcome the shortage of single loss function in video prediction network,and combined with the characteristics of weld pool image,the idea of multiple loss function is proposed,that is,multiple loss function using MSE,SSIM and perceptual loss function.The validity of this algorithm is verified by calculating PSNR and other image quality evaluation indexes of real image and predicted image.Based on the prediction image,the qualitative analysis algorithm of weld pool prediction and the quantitative analysis algorithm of weld pool prediction are proposed.The qualitative analysis of hump,penetration and the quantitative analysis of depth of weld pool are carried out respectively.The experimental results are equivalent to the results of monitoring technology,and the early warning of weld quality is realized.