NUCLEUS-MoE: Unified Model of Pool Boiling for Liquid Cooling

2026-05-26

AI Summary

This work addresses the challenge of modeling pool boiling, a highly nonlinear process governed by strong coupling among phase change, turbulence, and transport phenomena, with performance critically dependent on fluid properties and thermodynamic conditions. Existing data-driven approaches exhibit limited generalization and typically require separate models for different operating regimes. To overcome this, we propose NUCLEUS-MoE, the first unified surrogate model based on a Mixture-of-Experts (MoE) architecture capable of simultaneously handling diverse fluids and boiling states within a single framework. The model integrates neighborhood attention and signed distance field reinitialization to preserve interface consistency and employs a dynamic routing strategy that enables unsupervised, spatially and physically aware specialization of experts. Experiments demonstrate that NUCLEUS-MoE matches or outperforms specialized baselines across saturated and subcooled boiling of dielectric liquids, refrigerants, and cryogenic fluids, while achieving zero- or few-shot cross-fluid generalization to the novel immersion coolant Opteon 2P50 without compromising physical fidelity.

Abstract

Two-phase boiling enables heat transfer rates an order of magnitude higher than single-phase cooling, but it remains difficult to model due to the strong coupling between phase change, turbulence, and transport, as well as extreme sensitivity to fluid properties and thermodynamic conditions. Existing learning-based surrogates are either condition- or fluid-specific, limiting generalization and requiring separate models. We present NUCLEUS, a mixture-of-experts model for pool boiling that replaces collections of specialized surrogates with a single architecture. NUCLEUS combines neighborhood attention, signed distance field reinitialization for interface consistency, and expert routing that exhibits emergent specialization across distinct boiling dynamics. Trained on high-fidelity simulations of pool boiling, NUCLEUS jointly models saturated and subcooled boiling across three fluid classes (dielectrics, refrigerants, and cryogens), resolving failure modes of prior models on extreme fluids. We show that expert routing exhibits coherent spatial structure and specialization without explicit supervision. Quantitatively, NUCLEUS matches or exceeds baselines while maintaining physical consistency across heterogeneous boiling configurations. We also show zero-shot and few-shot generalization capabilities on downstream tasks such as a new fluid (Opteon 2P50 developed for immersion cooling). These results demonstrate that mixture-of-experts models are a scalable pathway toward unified surrogate modeling of boiling dynamics and lay the groundwork for broader generalization across scientific ML.

Problem

• pool boiling
• surrogate modeling
• generalization
• two-phase flow
• fluid dynamics

Innovation

• Mixture-of-Experts
• pool boiling
• signed distance field
• neighborhood attention
• zero-shot generalization