POD-DEIM based model order reduction for speed-up of flow parametric studies

Abstract

The model order reduction (MOR) is a useful tool for accelerating calculations connected to parametric studies or optimizations of complex systems. In this paper, we provide a MOR method directly applicable to industrial scale problems. We consider a posteriori MOR based on the proper orthogonal decomposition (POD) with Galerkin projection. The problems arising from the nonlinearities in the original model are addressed within the framework of the discrete empirical interpolation method (DEIM). The reduced order model (ROM) construction is newly implemented as a part of the open-source OpenFOAM CFD library. The new implementation is validated on a number of small scale tests and applied to a real-life problem of acceleration of a parametric study of flow in a complex geometry of an absorption column. Based on our tests, the application of reduced order model as a predictor for standard SIMPLE solution can speed-up the solution approximately two-fold. Furthermore, higher speed-ups may be achieved if the full order model may be completely replaced by ROM.