Reduced Order Models
3D computational fluid dynamics (CFD) simulations typically require hours or even days of computing time on a high-performance cluster. In contrast, reduced-order models (ROMs) can deliver fast yet accurate approximations of quantities of interest. Such models hold independent value and can accelerate uncertainty quantification and optimization studies. Physics-based ROMs simplify assumptions about the model physics, e.g., 0D and 1D fluid dynamics models (see picture). Data-driven models leverage data from past CFD solutions to train computationally lightweight models (e.g., neural networks) that predict the flow fields produced by new anatomies and flow conditions.