79th Annual Meeting of the APS Division of Fluid Dynamics (Nov 22 — 24, 2026)

V4397727: A Real-Time Interactive Digital Twin of Particle-Resolved Multiphase Flow via Compositional AI

Authors
  • Kai Liu, University of Tennessee Knoxville
  • Haochen Li, University of Tennessee
  • S. Balachandar, University of Florida

Particle-resolved multiphase-flow simulations provide rich information about the flow around individual particles, but their computational cost normally restricts them to offline analysis. This creates a long-standing tradeoff between high-fidelity physical prediction and the responsiveness needed for interactive exploration, rapid what-if studies, and real-time engineering decisions. This work presents a real-time interactive digital twin that uses compositional AI to bridge this gap, turning particle-scale flow prediction from a calculation that users wait for into an environment they can explore and modify while it is running. A Neural Particle Model (NPM) provides reusable particle-scale flow representations, while Modular Operator Superposition (MOS) couples these building blocks to account for many-particle interactions and reconstruct the collective flow without requiring a fine online CFD mesh. The video demonstrates direct hand interaction, on-the-fly particle reconfiguration, real-time flow visualization and control, fluid-mediated particle interactions, and polydisperse systems of triaxial ellipsoidal particles with arbitrary size, shape, orientation, and 6-DOF motion. The broader goal is to provide high-fidelity flow insight at the speed of engineering decisions.

This work is licensed under the Creative Commons Attribution-NonCommercial 4.0 International License. Any reuse must credit the author(s) and provide a link back to this page.