Prediction of the shift in melting mode during additive manufacturing of 316 L stainless steel

Abstract

In the additive manufacturing (AM) of 316 L stainless steel, the keyhole and conduction mode of melting significantly influence the quality of the final product. The present work proposes a novel criterion based on AM process parameters like laser power, scan speed, beam width, and powder bed thickness to indicate the shift in a melting mode. The proposed criterion avoids the limitation of the normalized enthalpy factor as an indicator for predicting a shift in the melting mode. The values of the proposed criterion for various experimental data clearly indicate a gradual transition from conduction to keyhole mode rather than a sharp shift, which complies with the actual process. Further, classification-type machine learning (ML) techniques (Logistic Regression, Random Forest Classifier, and XGBoost Classifier) are used to predict the shift in the melting mode based on the data collected from the literature. The evaluation of the performance of the models for the train and test datasets and K-Fold validation showed satisfactory results. The parametric ML model based on logistic regression with an accuracy of 85% and a physics-informed ML model based on the XGBoost algorithm with an accuracy of 88.75% are reliable for predicting the melting modes. Based on the results of the ML model, process maps were developed showing the transition from conduction to keyhole mode for different beam widths and powder bed thicknesses. It is further observed from the process maps that the tendency of keyholing decreases with an increase in the beam size and powder bed thickness for 316 L stainless steel.