Machine Learning Approach for Predicting Bead Geometry of Stainless Steel in Wire arc Additive Manufacturing

Abstract

Wire arc additive manufacturing (WAAM) employs an electric arc to melt wire feedstock, making it a method within additive manufacturing (AM). It deposits material layer by layer to build up a part. The present study investigated the application of machine learning classification-based models for estimating bead width and bead height of stainless-steel parts fabricated using WAAM. The input parameters (voltage, current, wire feed rate, and travel speed) were considered as input to algorithms. Training and testing were performed for 98 experimental data sets from peer-reviewed literature. The machine learning classification models, K-nearest neighbors, decision tree with gini index as criteria, and random forest were evaluated. The ML model performance was evaluated utilizing statistical metrics, including accuracy, F1 score, precision, and recall. The decision tree classifier exhibited the highest accuracy of 87.8% for bead width and 84.7% for bead height. The findings offer valuable insights into leveraging ML techniques to enhance the performance and accuracy of predictive models within WAAM-based AM.