Speakers
Description
We consider a model-order reduction approach to approximate scalar transport
and mixing in a T-junction using spectral-element-based large-eddy simulation
(LES) for the full-order model. One such LES simulation of 100 convective-time
units can cost hundreds of thousands of core-hours on a supercomputer. For the
reduced-order model, we apply Galerkin projection using POD-generated global
basis functions to reproduce the quantity of interest, the concentration
profile of the passive scalar. For turbulent flows, the Galerkin formulation
applied to a small number of dominant POD modes may not be sufficient to remain
in the basin of attraction of the underlying dynamics. The reason that is
typically cited for this behavior is the inability of this combination to
sufficiently dissipate energy since the POD modes lack the small-scale
structures found in such flows. By introducing a stabilization mechanism,
significant improvement in accuracy can be achieved. Here, we describe recent
results using POD-Galerkin projection with constrained optimization to ensure
that the flow evolves within the dynamical range observed in the full-order
model. With this approach, we address the turbulent flow reproduction problem,
which is a first step towards parametric model order reduction for turbulent
thermal transport.
