GreenMelt: Part-Scale Ultra-Fast Melt Pool Prediction in Laser Powder Bed Fusion Accounting for Latent Heat and Marangoni Convection
Abstract
Predicting microstructure in Laser Powder Bed Fusion (LPBF) or laser surface scanning requires accurate thermal analysis—including melt pool shape resolution at the part scale—yet remains computationally prohibitive due to the coupling between macro-scale heat accumulation and meso-scale melt pool phenomena. Thermal history governs solidification microstructure, making its rapid and accurate prediction a prerequisite for process parameter optimization. In this work, we propose a two-scale thermal model, named GreenMelt, that couples a macroscopic module, capturing part-scale heat accumulation, with an ultra-fast mesoscopic module resolving the melt pool under the conduction-mode assumption and especially the zone where solidification takes place. The approach is formulated using Green's functions supplemented by physically based distributed heat sources that account for the principal nonlinearities of the problem: latent heat of solidification, and temperature redistribution induced by outward or inward Marangoni convection. Determining these sources from physical arguments is the main novelty of this contribution. The framework supports arbitrary laser scan paths and nonlinear thermal contributions without compromising computational efficiency and without requiring calibration of internal parameters. Experimental validation was conducted on five single laser traces on a bare IN625 plate, demonstrating agreement within 2.5% on average on cross-section profiles. Comprehensive code-to-code verification was performed for single-track numerical experiments under various process parameters and across four different materials (IN718, IN625, 316L, and Ti6Al4V). The temperature field in the melt pool is computed in less than 20 seconds on average on a laptop, compared to more than 5 hours on average for equivalent computational fluid dynamics (CFD) simulations. Good quantitative agreement is maintained, with an average relative error in the mushy zone lower than 2% on average for single-track cases. The model's capabilities are further demonstrated on a full three-dimensional LPBF numerical experiment, confirming its applicability at the part scale. This approach makes the rapid numerical optimization of process parameters — including laser power, velocity, and scan strategy — tractable, paving the way for fine control over microstructure and mechanical properties in industrial LPBF components.
Related
- Marangoni flow induced by crystallization in saline droplets
- Surfactant-induced Marangoni stresses and surface viscosity in Faraday instability
- A contaminant-concentration-dependent surface tension does not explain the absence of solutal Marangoni flow in evaporating droplets
- The effects of surfactant solubility on inertial Marangoni flow: theory and numerics
- Thermosolutal Marangoni instabilities in a weakly viscoelastic falling film with soluble surfactant
BibTeX
@misc{weiszpatrault2026greenmelt,
title = {{GreenMelt: Part-Scale Ultra-Fast Melt Pool Prediction in Laser Powder Bed Fusion Accounting for Latent Heat and Marangoni Convection}},
author = {Daniel Weisz-Patrault},
howpublished = {HAL},
year = {2026},
month = oct,
url = {https://hal.science/hal-05682678v3},
note = {CC BY-NC-SA 4.0},
}
RIS
TY - UNPB
TI - GreenMelt: Part-Scale Ultra-Fast Melt Pool Prediction in Laser Powder Bed Fusion Accounting for Latent Heat and Marangoni Convection
AU - Daniel Weisz-Patrault
PY - 2026
DA - 2026/10/27
PB - HAL
UR - https://hal.science/hal-05682678v3
AB - Predicting microstructure in Laser Powder Bed Fusion (LPBF) or laser surface scanning requires accurate thermal analysis—including melt pool shape resolution at the part scale—yet remains computationally prohibitive due to the coupling between macro-scale heat accumulation and meso-scale melt pool phenomena. Thermal history governs solidification microstructure, making its rapid and accurate prediction a prerequisite for process parameter optimization. In this work, we propose a two-scale thermal model, named GreenMelt, that couples a macroscopic module, capturing part-scale heat accumulation, with an ultra-fast mesoscopic module resolving the melt pool under the conduction-mode assumption and especially the zone where solidification takes place. The approach is formulated using Green's functions supplemented by physically based distributed heat sources that account for the principal nonlinearities of the problem: latent heat of solidification, and temperature redistribution induced by outward or inward Marangoni convection. Determining these sources from physical arguments is the main novelty of this contribution. The framework supports arbitrary laser scan paths and nonlinear thermal contributions without compromising computational efficiency and without requiring calibration of internal parameters. Experimental validation was conducted on five single laser traces on a bare IN625 plate, demonstrating agreement within 2.5% on average on cross-section profiles. Comprehensive code-to-code verification was performed for single-track numerical experiments under various process parameters and across four different materials (IN718, IN625, 316L, and Ti6Al4V). The temperature field in the melt pool is computed in less than 20 seconds on average on a laptop, compared to more than 5 hours on average for equivalent computational fluid dynamics (CFD) simulations. Good quantitative agreement is maintained, with an average relative error in the mushy zone lower than 2% on average for single-track cases. The model's capabilities are further demonstrated on a full three-dimensional LPBF numerical experiment, confirming its applicability at the part scale. This approach makes the rapid numerical optimization of process parameters — including laser power, velocity, and scan strategy — tractable, paving the way for fine control over microstructure and mechanical properties in industrial LPBF components.
ER -