Distributed computing for physics-based data-driven reduced modeling at scale: Application to a rotating detonation rocket engine
🤖 AI Summary
To address the insufficient real-time performance of high-fidelity multiphysics simulations for rotating detonation rocket engines (RDREs), this work proposes a distributed, physics-informed reduced-order modeling (ROM) framework. The method integrates scalable distributed computing with a physics-constrained neural operator (PCNO) that intrinsically embeds conservation laws, augmented by multiscale spatiotemporal convolutions and an MPI-based parallel ROM solver. Evaluated on a 256-GPU cluster, the framework achieves 92% weak scaling efficiency. It enables millisecond-scale online prediction on meshes exceeding ten million cells, yielding transient RDRE flow-field errors below 3.7%. Compared to full-order CFD, inference is accelerated by a factor of 2800×, while preserving physical consistency. This advancement significantly enhances the feasibility of rapid design exploration and closed-loop combustion control simulation for RDREs.
Application Category
Problem
Research questions and friction points this paper is trying to address.
Reducing computational cost of large-scale RDRE simulations
Enabling scalable physics-based reduced-order modeling
Accelerating design exploration for rocket engines
Innovation
Methods, ideas, or system contributions that make the work stand out.
Distributed memory algorithm for scalable ROM construction
Physics-based ROMs from sparse, large-dimension datasets
Fast predictive modeling for complex aerospace systems
🔎 Similar Papers
No similar papers found.
