Wire-arc additive manufacturing (WAAM) with cold metal transfer (CMT) has attracted considerable attention owing to its ability to produce complex shapes with minimal geometric defects, particularly in the fabrication of high-value, large-scale components. This study aimed to optimize bead geometry variations in WAAM-CMT by employing a full factorial design of experiments to generate a comprehensive dataset for the training of machine learning algorithms. The RSM method and NSGA-II algorithm were utilized to optimize key process parameters, such as the wire feed rate, welding speed, and stand-off distance. The objective was to achieve different bead geometry configurations for testing and evaluating each optimization method. Two optimization settings were employed for each method: one aimed to achieve wetting angles close to 90°, maximizing the bead height and minimizing the bead width, which is crucial for printing straight, thin-walled structures with precise dimensions and minimal defects, while the second aimed to achieve a larger bead with minimal height and low angles suitable for thick printing parts. The results indicate that RSM outperforms NSGA-II in minimizing bead height and maximizing bead width, with deviations of 3.17% and 13.29%, respectively, compared to NSGA-II’s 3.70% and 38.58%, respectively. However, NSGA-II excels at minimizing bead width and controlling alpha 2 and maximum temperature, achieving lower deviations of 6.40%, 0.80%, and 4.0%, respectively. This study demonstrates the potential for integrating machine learning techniques to refine WAAM processes and enhance the quality and reliability of additive manufacturing.
