This project provides the official implementation of a Spatio-temporal Resolution-adaptive Fourier Neural Operator for accurate and efficient prediction of phase-change interface evolution. The work is developed to address long-standing limitations in numerical phase-change rate modeling, where traditional approaches often rely on empirical parameters, and exhibit limited predictive accuracy.
A key contribution of this project is the demonstration of spatio-temporal resolution adaptability. The model generalizes robustly across unseen spatial mesh configurations and prediction intervals, maintaining over 95% accuracy with inference times below 4 ms. This highlights its suitability for integration into computational fluid dynamics (CFD) workflows. Such capability enables efficient surrogate modeling of phase-change rates, paving the way for faster and more accurate numerical simulations in advanced thermal management applications.
The authors of this work are:
Fanshuo Meng, Xiaoyang Li, Xun Zhu, Haipeng Xie, Peixue Jiang, Ruina Xu.
We have opened the codes and data involved in this article, which include: