Research On Weld Size Prediction Of Low Carbon Steel Multi Layer Single Pass GMAW Based On Deep Learning

Multi-layer single-pass welding is often used for welding thick and wide components,and has a wide range of applications in fields such as machinery,aerospace,and construction.In the application of multi-layer single-pass welding,it is necessary to strictly ensure the welding quality of each layer of welds.Therefore,achieving realtime detection of weld quality is of great significance.The weld seam size is an important criterion for judging the quality of the weld.The weld seam size can be used to accurately determine defects such as underfilled,unfilled and excessive residual height,thus effectively improving the quality of the weld seam.The melt pool image contains a large amount of information,which can visually reflect the dynamic changes of the melt pool,and is often used to judge the weld size information.However,a single pool image is limited in characterizing weld information when inspecting weld quality.This research introduces the electrical signal from the welding process and the pool image together to predict weld size information.This research adopts data normalization and data enhancement methods to avoid dimensional differences between the data and enrich the sample data;optimizes the melt pool classification model based on the AlexNet model to achieve the classification of different welding processes;constructs the weld seam size prediction model based on the PyTorch framework,extracts features from images and electrical signals through the residual network and residual shrinkage network respectively,and adopts the feature fusion module to achieve the fusion of multiple information features.After testing,the model effectively combines the characteristics of the two information sources,and the prediction results are stable and the error level can be kept low.After experimental verification,the welding process classification model proposed in this topic can achieve accurate classification of the welding process,and the weld size prediction model has high prediction accuracy when predicting the weld size of each process.Classification model prediction accuracy reached 99.70%.And the verification set of bottoming,filler and cover welds were identified with an accuracy of 99.74%,98.86%and 98.6%,respectively.The minimum average absolute error of weld thickness,back reinforcement and weld width of backing weld in the prediction model is 0.0267 mm,0.0246 mm and 0.0257 mm respectively.The minimum average absolute error of weld thickness,back reinforcement and weld width of filler welding is 0.092 mm,0.144 mm and 0.084 mm respectively.The minimum average absolute error of weld thickness,back reinforcement and weld width of cover welding is 0.123 mm,0.133 mm and 0.148 mm respectively.