Dynamic flux surrogate-based partitioned methods for interface problems
🤖 AI Summary
Traditional loosely coupled partitioned methods suffer from slow convergence, poor stability, and limited accuracy in multiphysics simulations. To address these limitations, this paper proposes a novel partitioned solution framework based on a dynamic flux surrogate model. The core method integrates a projection-based reduced-order model (PROM), online adaptive training, and an implicit–explicit hybrid time-integration scheme. Its key innovation is a dynamically updated boundary flux mapping mechanism—the first of its kind—that adaptively refines the surrogate during simulation, overcoming the failure of static surrogates under strong nonlinearity and transient coupling. Evaluated on canonical fluid–structure–thermal interaction benchmarks, the proposed approach achieves a 3.2× acceleration in convergence rate and reduces computational time by 67%, while rigorously preserving numerical stability and high-fidelity accuracy.
Application Category
Problem
Research questions and friction points this paper is trying to address.
Improves accuracy of loosely coupled multiphysics methods
Replaces field transfers with dynamic flux surrogates
Shifts computational burden to offline training phase
Innovation
Methods, ideas, or system contributions that make the work stand out.
Data-driven partitioned method for coupled PDEs
Dynamic mode decomposition for flux surrogate
Offline training shifts computational burden
🔎 Similar Papers
No similar papers found.
