Machine learning-based prediction of bead geometry in AZ31 magnesium alloy using robotic cold metal transfer WAAM

Abstract

This research proposes a machine learning (ML)-assisted framework for bead geometry analysis and optimization in cold metal transfer (CMT)-processed AZ31 magnesium parts. This work aims to predict and validate the bead geometry of single beads using ML techniques. This approach can be used to fabricate thin-walled AZ31 magnesium alloy components via robotic wire arc additive manufacturing (WAAM). In this article, the impact of current, voltage, wire-feed rate, and travel speed on bead height, bead width, and contact angle was evaluated through experimental runs. Microstructural analysis revealed a preference for equiaxed grains as the dominant microstructural entities. The increased heat input initiated grain coarsening, while the bead edges exhibited a finer equiaxed structure due to rapid cooling, indicating that the local thermal history is the most significant controlling element. The experimental data were used to develop predictive models using polynomial regression (PR), decision trees (DT), random forests (RF), and gradient boosting regression (GBR). These models predict principal bead characteristics, such as width, height, and contact angle, to determine optimal process conditions that minimize defects and ensure uniformity. From the results, it is observed that the gradient boosting regressor consistently outperformed the other models, achieving coefficient of determination (R²) values of 95.16%, 91.11%, and 89.47% for bead height, width, and contact angle, respectively, along with the lowest mean squared error (MSE) values of 0.0788, 0.1146, and 28.92, respectively. The results demonstrated the effectiveness of the ML methodology in enhancing process stability, improving bead quality, and enabling reproducible manufacturing of AZ31 parts, with far-reaching implications for aerospace, automotive, and biomedical applications.