Physics-guided FEM–ML framework for melt pool geometry prediction and CET analysis in DED-processed in-situ alloyed FeMn–CoCrNi HEA

Understanding the columnar-to-equiaxed transition (CET) and predicting melt-pool dimensions under different fabrication conditions are essential for tailoring the microstructure and mechanical properties of materials produced by directed energy deposition (DED). In the present work, single tracks of an in-situ alloyed FeMn–CoCrNi high-entropy alloy (HEA) were fabricated by DED, and experimentally measured melt-pool dimensions were used to calibrate a finite element (FE) model to predict melt-pool geometry under the investigated processing conditions. The calibrated FE model showed good agreement with the experimental measurements, with relative errors below 6%. The validated FE model was subsequently used to generate an extended physics-assisted dataset for machine learning, enabling melt-pool geometry prediction over the investigated processing conditions. The GPR model exhibited superior predictive accuracy compared with XGBoost, with R 2 values of 0.992 and 0.996 for melt pool width and penetration depth, respectively. Experimental observations revealed that CET occurred along the deposition height and was associated with the morphological transformation of cellular substructures. The finite element simulations were employed to calculate the thermal gradient (G) and solidification rate (R), which were subsequently used with a Hunt–Gäumann-based CET criterion to analyze the spatial tendency toward columnar and equiaxed grain formation. The results show that increasing LED is associated with a lower G/R ratio and a greater tendency toward equiaxed grain formation. Overall, the present work establishes an integrated experimental–FEM–ML framework for predicting melt pool geometry and a FE-CET base analysis in DED-processed in-situ alloyed FeMn–CoCrNi HEA.