Multi Information Fusion Inspection For 5754/7075 Aluminum Alloys Laser Lap Welding Joints

As a lightweight alloy material,aluminum alloy is widely used in some parts of car body and is an important choice for lightweight of vehicle.Laser welding has high power density,high welding efficiency and high automation level.It is gradually replacing other traditional joining methods in the automotive industry.Different materials are used in different parts of car body,and defects are easily occurred in laser welding of dissimilar aluminum alloys,affecting the quality of welded joints,and reducing production efficiency.Currently,there is no effective online quantitative welding depth inspection and surface defect online inspection methods.In this paper,a metal vapor high-speed photography image acquisition system is built based on Phantom VEO 710 s high-speed photography camera,and a surface defect inspection image acquisition system is built based on Basler ac A780-75 gc industrial camera,realizing the image acquisition of metal vapor in laser welding process and weld surface after welding.Metal vapor images are acquired by high-speed photography,and total21 features in terms of brightness,volume,location and texture are extracted from metal vapor images.A 10 mm weld is used as the basic unit for weld depth inspection.A weld depth regression model is established based on machine learning.According to the features extracted from metal vapor images,the on-line weld depth inspection is effectively implemented,with an average absolute error of 0.165 mm and an error rate of 8.25%.The surface defect inspection system is established by combining the laser triangulation method and the image processing method,which effectively combines the complementarity of the two methods,improves the inspection accuracy rate and the inspection speed,and reduces the high hardware requirements of the inspection system.The three-dimensional contour of the surface was reconstructed based on laser triangulation,and18 contour features were extracted.Based on image processing,70 image features were extracted from the surface image.Based on feature importance,feature selection was performed,and 37 features with feature importance accounting for the first 70% were selected.Based on XGBoost,a surface defect classification model was established,which effectively realized the online inspection of surface defects,with a detection speed of up to 9.84 m / min and classification accuracy of 95.71%.The defect and defect-free two-class classification accuracy is 98.57%,with 100% recall ratio and 97.20% precision.Through the established weld depth inspection system and surface defect inspection system,the quality inspection of 5754/7075 aluminum alloys laser lap welding joints is effectively realized based on the metal vapor images,surface contour and surface images.