Reduced-order models are applied to the laminar vortex-shedding flow around a circular cylinder. The models rely on approximations of the solutions in the space spanned by a set of POD or CVT reduced basis vectors, which are computed from snapshots of a numerical solution. To enable the computation of drag and lift forces, the modeling of the velocity and pressure is realized via a one-way coupling. A comparison of the results of the CVT and POD reduced-order models is presented.
Reduced-order modelingA spatially semi-discretized incompressible Navier-Stokes problem can be obtained by inserting functions of a finite element space in the weak form of the governing equations. The resulting finite element model is given bywhere the vectors U (t) and P (t) contain the values of the velocity and pressure at the mesh nodes, and U (0) = U 0 . Reducedorder models can be derived in an analogical way by inserting functions of a reduced-order subspace in the weak form or, alternatively, by projecting the finite element model on the space spanned by a set of reduced basis vectors. Velocity reduced basis vectors Φ 1 , . . . , Φ R are computed from a set of snapshots U 1 , . . . , U N , where typically R ≪ N . The proper orthogonal decomposition (POD) [1] generates basis vectors by solving the minimization problemwhere M indicates norms and inner products induced by the finite element mass matrix. The solution of the minimization problem can be found by singular value decomposition. The centroidal Voronoi tessellation (CVT) assigns each snapshot to one of the clusters V 1 , . . . , V R and defines Φ 1 , . . . , Φ R as the cluster centers which solveSolutions of this minimization problem can be approximated in an iterative way using Lloyd's method [2]. Reduced basis vectors Φ 1 , . . . , Φ R are now computed usingwhere U ref is a reference velocity. Additionally, reduced basis vectors Ψ 1 , . . . , Ψ R are computed from a set of pressure snapshots P 1 , . . . , P N . The discrete velocity and pressure fields can then be approximated usingFor the velocity reduced-order model, a Galerkin projection of (1) on the space spanned by the velocity reduced basis vectors is performed, which gives Φ T r M˙ U R = Φ T r N ( U R ) U R + Φ T r K U R + Φ T r CP, r = 1, . . . , R.