Prediction of weld area based on image recognition and machine learning in laser oscillation welding of aluminum alloy

Abstract

Laser oscillation welding is an advanced aluminum alloy welding technology which can effectively reduce weld porosity induced by physical and chemical properties of aluminum alloy. Weld area is often as an evaluation index of geometric characteristics to evaluate the welding quality since it has great effect on the mechanical properties of welded joints. In this paper, a prediction method of weld area based on process parameters is proposed for laser oscillation welding of 6061 aluminum alloy. From the metallographic micrographs of welding experiments, the cross-sectional area of weld is calculated by image recognition technology and the error of recognized weld area is less than 8.8%. Additionally, various prediction models for weld area under different process conditions are established by machine learning algorithms including linear regression, polynomial regression and back propagation neural network (BPNN). The established prediction models are compared by the mean square error (MSE) and coefficient of determination (R2). The results show that the BPNN prediction model indicates the highest accuracy and robustnessand can be adopted to predict the cross-sectional area of weld. The proposed method can be used for selecting the optimal process condition for the ideal welded joints with the desired weld geometry to improve the quality of laser oscillation welding.