Towards an ensemble machine learning framework for accurate prediction of WAAM bead geometry

Abstract

Wire Arc Additive Manufacturing (WAAM) has gained significant attention as an efficient technique for producing large-scale metallic structures with high material utilization and enhanced deposition rates. Despite its advantages, accurately predicting bead geometry remains challenging due to the highly nonlinear interactions between process variables and deposition behavior. In this study, an AI-based regression framework is proposed to predict bead geometry in stainless-steel WAAM processes using key input parameters, including voltage, current, wire feed rate, and travel speed. An experimental WAAM dataset was employed to train and evaluate several machine learning regression models, including Random Forest Regressor (RF), Support Vector Regressor (SVR), and Extreme Gradient Boosting Regressor (XGBoost). Model performance was analyzed using evaluation metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and coefficient of determination (𝑅 2 ). Among the developed models, XGBoost achieved the highest prediction accuracy, with an 𝑅 2 of 0.9560, indicating a strong correlation between predicted and actual bead geometry measurements. In addition, feature importance analysis identified wire feed rate as the most significant parameter influencing bead geometry. The proposed ensemble learningbased framework successfully models the nonlinear dynamics of the WAAM process while minimizing reliance on traditional trial-and-error methods. This work advances intelligent manufacturing, predictive process optimization, and smart WAAM technologies in accordance with Industry 4.0 principles.