Simulation-guided predictive machine learning framework for optimizing directed energy deposition of Fe-Mn alloys

Abstract

This study proposes a hybrid Finite Element Modeling (FEM) and Machine Learning (ML) framework for predicting melt pool geometry in the Directed Energy Deposition (DED) of Fe-Mn alloys. A thermal FEM model based on the Goldak double-ellipsoidal heat source was developed and validated using 24 single-track experiments performed at different laser powers and axial speeds. The experimental dataset was obtained from a full-factorial DOE, with laser power and axial speed as input parameters and melt pool width and penetration depth as output responses. The validated FEM model was then used to generate additional data within the investigated process window, resulting in a combined experimental–simulation dataset of 56 samples. Two ML models, Extreme Gradient Boosting (XGBoost) and Gaussian Process Regression (GPR), were trained and evaluated using this hybrid dataset. The FEM model showed good agreement with experimental measurements, with a maximum error of approximately 7%. GPR provided better predictive performance than XGBoost, achieving lower RMSE and MAPE values for both width and depth, while also providing uncertainty estimates. The proposed framework reduces the need for extensive experimental trials and enables rapid process-window evaluation for Fe-Mn DED. However, the present model is limited to single-track deposition and requires further validation before being extended to multi-layer builds and broader processing conditions.