Prediction of deposition bead geometry in wire arc additive manufacturing using machine learning

Abstract

The deposition bead geometry in the arc strike zone is often abnormal compared with that in the middle deposition zone. The accumulation of these errors in the deposition process results in defects in the overall geometry of the deposition bead. To prevent the accumulation of such errors, this study aimed to address the irregular deposition bead shape and deviation of the deposition bead geometry in the arc strike zone without changing the deposition path or introducing external devices. A support vector machine classifier, which is a typical application of machine-learning classification models, was used to obtain the deposition condition ranges for a uniform deposition bead shape. The deposition conditions of the arc strike and middle deposition zones that could reduce the deviation of the deposition bead geometry in the arc strike zone were obtained using support vector machine regression. These conditions were applied in the form of variable conditions in single-path deposition. Five-layer deposition was conducted to validate the regression model. As a result of the validation experiment, the average height and width errors in the arc strike zone were 0.36% and 1.28%, respectively, and the average height and width errors in the middle deposition zone were 0.64% and 1.26%, respectively. Therefore, the regression model can be used for accurate deposition, thereby reducing the post-processing cost.

Keywords

• Additive manufacturing
• Wire arc additive manufacturing
• Machine learning
• Support vector machine
• Bead geometry optimization