Learning Interface Breakup: A Geometry-Conditioned Latent Surrogate for Spray Formation

2026-06-15 · Source: Artificial Intelligence · Field: Science & Research — Engineering & Applied Sciences, Mathematics & Computational Sciences, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

A new geometry-conditioned latent surrogate model addresses the challenge of predicting transient two-phase breakup in spray nozzles, a process critical for nozzle design but computationally expensive with high-fidelity volume-of-fluid (VOF) simulations using adaptive mesh refinement (AMR). Traditional surrogate models falter due to the dynamic nature of liquid-gas interfaces and adaptive discretizations. This novel model, trained on 797 two-phase nozzle simulations, encodes the AMR cell-density field as a compact proxy for solver resolution, rather than the full flow state. It then reconstructs transient density evolution and nozzle geometry, with a second stage recovering other flow variables. The method accurately captures key interface dynamics on held-out simulations, achieving an inference time of 0.045 seconds per trajectory, representing a speed-up exceeding 6×10^4 compared to Basilisk CFD.

Key takeaway

For research scientists developing computational fluid dynamics (CFD) models for complex two-phase flows, this work offers a significant pathway to overcome simulation bottlenecks. You should consider adopting geometry-conditioned latent surrogates that utilize adaptive mesh refinement (AMR) structures as compact representations. This approach drastically reduces inference time, enabling rapid design exploration and optimization of systems like spray nozzles, which was previously impractical due to high computational costs.

Key insights

AMR refinement structure can serve as a compact, learnable representation for geometry-conditioned surrogate modeling of transient two-phase flows.

Principles

• High-fidelity VOF simulations are too expensive for iterative design.
• Adaptive discretization challenges standard surrogate models.
• Encoding AMR cell-density field offers compact proxy.

Method

Train a geometry-conditioned latent surrogate on two-phase simulations, encoding AMR cell-density as a proxy, then reconstruct density and geometry, and recover other flow variables in a second stage.

In practice

• Accelerate spray nozzle design iterations.
• Reduce CFD simulation time for two-phase flows.
• Develop surrogates for evolving adaptive discretizations.

Topics

• Spray Nozzle Design
• Two-Phase Flow
• Surrogate Modeling
• Adaptive Mesh Refinement
• Latent Representations
• Computational Fluid Dynamics

Best for: AI Scientist, Research Scientist

Related on AIssential

• Ten Years of Trying to Learn from a Mesh That’s Too Large — Large CFD meshes pose representational challenges for surrogate models, requiring specialized architectures to capture localized physical features.
• Neural Operator-Based Surrogate Model for CFD:Helical Coil Steam Generator in Small Modular Reactor — Neural operator-based surrogates, combined with ROMs and multi-scale techniques, enable real-time CFD for SMR digital twins.
• The Bug That Ruined Game Physics For Decades — A new fluid simulation method prevents volume loss and handles complex dynamics by design, making advanced theory practical.
• Neural surrogates for crystal growth dynamics with variable supersaturation: explicit vs. implicit conditioning — Explicit supersaturation conditioning in CRNNs offers superior crystal growth simulation fidelity over implicit methods.
• AirfoilGen: A valid-by-construction and performance-aware latent diffusion model for airfoil generation — AirfoilGen ensures geometrically valid and performance-aware airfoil designs using a novel representation and conditional diffusion.
• A Geometry-Aware Triplane Field Network for Vehicle Aerodynamic Prediction — GTF-Net combines triplane representations with explicit geometric cues for accurate vehicle aerodynamic prediction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.