Carlos
- Updated: March 11, 2026
- 1 min read
Adaptive Uncertainty-Guided Surrogates for Efficient phase field Modeling of Dendritic Solidification
Adaptive Uncertainty-Guided Surrogates for Efficient Phase Field Modeling of Dendritic Solidification
In this article we explore a novel surrogate modeling framework that dramatically reduces the computational cost of phase‑field simulations for dendritic solidification, a critical challenge in additive manufacturing. The approach leverages uncertainty‑driven adaptive sampling, XGBoost, and convolutional neural networks (CNNs) to create high‑fidelity, low‑cost predictions of microstructural evolution.
Key concepts covered include:
- Monte Carlo dropout and bagging techniques for estimating model uncertainty.
- Local hyperspherical sampling to focus new simulations where uncertainty is highest.
- Comparative analysis of domain‑informed versus data‑driven surrogate models.
- Environmental impact assessment, including CO₂ emission reductions.
For more information on our adaptive surrogate methodology, visit our technology page and explore related resources on research.
Read the full pre‑print on arXiv for detailed methodology and results.