Additive manufacturing (AM) processes like laser powder bed fusion (LPBF) offer advantages in producing complex parts. However, process variabilities and uncertainties lead to defects, hindering part qualification and certification. To address this, NIST organizes benchmark challenges to develop predictive simulations. This paper focuses on the NIST AM Bench 2022 Challenge 3, which involved predicting various parameters for the LPBF process of IN718 using varying laser power, scan speed, and spot diameter. The challenge included both single and multi-track (pad) printing setups. Previous modeling approaches ranged from simple heat conduction to complex thermal-fluid models. Simpler models, while computationally efficient, fail to capture crucial melt pool interactions like Marangoni convection and vaporization, leading to inaccuracies in melt pool geometry and cooling rate predictions. More complex models are computationally expensive and may still have challenges due to process uncertainties and non-linear material properties. This study explores model reduction techniques and calibration schemes to improve predictive accuracy while maintaining computational feasibility. The researchers developed two heat source models: a scaling law-based heat source and a physics-guided heat source model, and integrated them into an in-house AM-CFD code to predict the NIST challenge experiments.

The literature review covers various AM process modeling approaches, ranging from simple heat conduction models with limitations in capturing melt pool interactions to complex thermal-fluid models that account for Marangoni convection and vaporization. The authors note that while these more detailed models provide improved prediction, their computational demands are high. Previous work utilized reduced-order modeling techniques like HOPGD for more efficient predictions. The present work builds on these advances by proposing two novel heat source models and calibrating them against experimental data.

The study employed an in-house developed Additive Manufacturing Computational Fluid Dynamics (AM-CFD) code. This code is a finite volume method (FVM)-based C++ solver that includes a transient three-dimensional thermal-fluid model considering Marangoni effect driven liquid flow in the melt pool. The model uses a cylindrical heat source to represent the laser heating. The key aspect of the methodology lies in the calibration of the heat source parameters. Initially, a higher-order proper generalized decomposition (HOPGD) based surrogate model is used for initial calibration of the heat source parameters. This is followed by further refinement using a scaling law based on keyhole formation. A final physics-guided heat source model is also proposed, which relates the volumetric energy density (VED) to the melt pool aspect ratio. The AM-CFD code uses temperature-dependent material properties for IN718 obtained from literature. Adaptive meshing is employed to ensure accurate resolution of the melt pool region. The model incorporates several factors including heat loss due to radiation, convection, and evaporation. The mushy zone effect is included through momentum source terms. Three different calibration schemes are compared: heuristic heat source parameterization, scaling law-based heat source parameterization, and physics-guided heat source parameterization. The keyhole number (Ke), a dimensionless parameter, is used in the scaling law-based approach, while the physics-guided approach directly correlates the VED with the melt pool geometry.

The researchers found that the heuristic heat source parameterization showed limitations in predicting results at higher volumetric energy densities (VED). This was due to the simplified model not considering keyhole formation and heat loss from vaporization. The scaling law-based calibration, which incorporates keyhole formation, provided better results but still had limitations at lower VED. The physics-guided calibration scheme, which relates the heat source depth to the melt pool aspect ratio, significantly improved the predictions, particularly for melt pool geometry and cooling rates across all VED values. The study successfully predicted solid and liquid cooling rates, time above melt, and melt pool geometry for both single-track and multi-track experiments. For single-track experiments, the physics-guided method showed a significant reduction in the relative error compared to the heuristic and scaling law-based methods for almost all parameters. The differences were notably reduced for most parameters when comparing the experimental measurements to the physics-guided model. There was particularly good agreement for melt pool dimensions, especially with the physics-guided calibration. This was also true for multi-track (pad) laser scanning, although the morphology differed slightly, highlighting the need for improved handling of surface evolution physics in future models. While numerical results showed good agreement for most cases, the qualitative comparison of melt pool shape differed, mainly because of the absence of surface evolution physics. The varying depth of melt pools in the Y-direction scan could be attributed to the absence of gas flow modeling.

The findings demonstrate the importance of selecting an appropriate heat source parameterization scheme and heat source model for accurate prediction of LPBF processes. While sophisticated thermal-fluid models can improve prediction fidelity, their computational cost is substantial. This research effectively balances accuracy and computational efficiency through careful calibration and model reduction techniques. The use of a simple cylindrical heat source model, while a simplification, proves effective after proper calibration. The improvements achieved with the physics-guided heat source parameterization highlight the benefits of incorporating key physical relationships into the calibration process. The relatively good agreement between the simulation and experimental results for both single and multi-track cases validate the proposed approach. The discrepancies observed in multi-track simulations in qualitative aspects, such as morphology, reveal the need for integrating surface evolution and gas flow physics into future models.

The study successfully developed and validated a physics-guided heat source parameterization scheme for accurately predicting key parameters in IN718 LPBF. The results demonstrate the importance of considering keyhole formation and heat loss mechanisms in heat source modeling. Future work should focus on incorporating surface evolution and gas flow physics to improve the accuracy of multi-track simulations. The development of more sophisticated but computationally efficient models will enhance the predictive capabilities of AM simulations.

The study's limitations include the use of a simplified cylindrical heat source model, which neglects complex laser-material interactions like multiple reflections and detailed keyhole dynamics. The absence of surface evolution physics also affects the accuracy of melt pool morphology predictions, particularly in multi-track scenarios. The accuracy depends on the availability and quality of the material properties used in the simulation. Furthermore, the model does not account for the effects of gas flow, which could influence melt pool geometry. Despite these limitations, the proposed approach provides a significant improvement in quantitative prediction, offering a good balance between accuracy and computational feasibility.