Galerkin reduced-order models (ROMs) approximate computational fluid simulations by reducing snapshot data to a basis of proper orthogonal decomposition (POD) modes and solving for modal coefficients with ordinary differential equations. Galerkin ROMs reduce computational cost and can approximate flows with alternate Reynolds numbers, while parametric reduced order models allow adjustment of other system parameters. However, Galerkin ROMs do not explicitly define boundary conditions, limiting ROM results to the boundary conditions used by the original data. We demonstrate using boundary penalties to simulate alternate boundary conditions in a Galerkin ROM of an incompressible lid-driven cavity flow. We formulate proper orthogonal decompositions (POD) of a regularized boundary condition solution and measure the effect of enforcing an unregularized boundary condition. We also use mean reductions constructed from Gaussian functions to improve ROM accuracy. Using Morris screening and parameter subset selection, we measure the sensitivity and identifiability of integrated ROM solutions to Reynolds number, boundary condition, boundary penalties strength, and the mean reduction. The ROM solution is relatively insensitive to the mean reduction except for velocity magnitude in the top-right eddy, where the effect of boundary regularization is greatest. However, we find that almost all characteristics of the model reduction are identifiable.

Zoom link: https://ncsu.zoom.us/j/98139608681?pwd=Z3RnRGRYL3JCRnlNT0NFdmZLRzRtdz09