A comparative study of machine learning algorithms in the prediction of bead geometry in wire-arc additive manufacturing

In the past few years, machine learning (ML) has become a widely used computational tool in manufacturing. The present study conducted a comparative analysis of five machine learning (ML) algorithms to predict the bead geometry in wire-arc additive manufacturing (WAAM). A robotic-controlled gas metal arc welding (GMAW) based WAAM setup was employed to deposit the beads. For the purpose of training and testing the ML models, a dataset consisting of four independent variables (travel speed, torch angle, wire feed rate, and contact tube to workpiece distance) and two dependent variables (bead height and width) was collected. The study employed five ML models including decision tree, random forest, XGBoost algorithm, linear regression, and artificial neural networks (ANN). The performance of ML models was evaluated using statistical metrics such as index of merit (IM), mean square error (MSE), mean absolute error (MAE) and coefficient of determination (R² value). A comparison of ML algorithms revealed that ANN (R² value = 0.6505) performed better than the XGBoost (R² value = 0.5916) in predicting the bead height. The lowest value of MSE and MAE equal to 0.0611 and 0.1893 respectively was achieved for ANN in predicting the bead height. For the prediction of bead width, linear regression (R² value = 0.8486) performed better than ANN (R² value = 0.8316). The lowest value of MSE and MAE equal to 0.1730 and 0.3157 respectively was achieved for linear regression in predicting the bead width. Linear regression is most suitable for datasets having strong collinearity, however, ANN, Random Forest and XGBoost performed very well in prediction when the dataset shows high non-linear relations. The comparison of ML algorithms revealed that computational work using machine learning gives high-performance results with minimal use for resources and time. Hence, machine learning can be the most effective computational tool in modern manufacturing industries.