Stochastic Reduced-Order Modeling and Data Assimilation for Near Real-Time Digital Twins of Turbulent Flows
Résumé
Digital Twins (DT) for engineering systems require models capable of real-time execution. While Reduced Order Modeling (ROM) offers significant acceleration, traditional methods often suffer from instabilities and predictive drift during time extrapolation. This work presents a robust framework to produce DT using a stochastic ROM with physics-driven generative closure. Our DT is coupled with 12 sparse, non-coplanar measurements through Particle Filter data assimilation every 0.5 dimensionless time. Validated on a Re = 3900 cylinder wake, our Digital Twin demonstrates high structural fidelity, stability, and accuracy while running up to 4 orders of magnitude faster than high-fidelity LES. This work is a step toward near real-time monitoring and prediction of complex turbulent flows.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |
