Complexity reduction in parametric flow problems via Nonintrusive Proper Generalised Decomposition in OpenFOAM
31 January, 2020 @ 11:00 am - 12:00 pm
Vasilis Tsiolakis received his Diploma in Mechanical Engineering in the Mechanical Engineering Department of the National Technical University of Athens (NTUA) and specialised in “Air and Ground Transport Vehicles”. He participated in the ITN AdMoRe project funded within the framework of EU H2020 Marie Sklodowska Curie Actions as Early Stage Researcher in Volkswagen Group Research. His PhD work between Universitat Politècnica de Catalunya and Swansea University deals with non-intrusive reduced order modelling for flow problems of industrial interest in the automotive sector.
The present thesis explores the viability of the proper generalised decomposition (PGD) as a tool for parametric studies in a daily industrial environment. Starting from the equations modelling incompressible flows, the separated formulation of the equations, the development of a parametric solver, the implementation in a commercial computational fluid dynamics (CFD) soft- ware, OpenFOAM, and a numerical validation are presented. The parametrised Stokes and Oseen flows are used as an initial step to test the applicability of the PGD to flow problems. The rationale for the construction of a separable approximation is described and implemented in OpenFOAM. For the numerical validation of the developed strategy analytical test cases are solved. Then, the parametrised steady laminar incompressible Navier-Stokes equations are considered. The non- intrusive implementation of PGD in OpenFOAM is formulated, focusing on the seamless integration of a reduced order model (ROM) in the framework of an industrially validated CFD software. The proposed strategy exploits classical solution strategies in OpenFOAM to solve the PGD spatial iteration, while the parametric one is solved via a collocation approach. Such non-intrusiveness represents an important step towards the industrialisation of PGD-based approaches. The capabilities of the methodology are tested by applying it to benchmark tests in the literature and solving a parametrised flow control problem in a realistic geometry of interest for the automotive industry. Finally, the PGD framework is extended to turbulent Navier-Stokes problems. The separable form of an industrially popular turbulence model, namely Spalart-Allmaras model, is formulated and a PGD strategy for the construction of a parametric turbulent eddy viscosity is devised. Different implementation possibilities in the nonintrusive PGD for parametrised Navier-Stokes equations are explored and the proposed strategy is applied to well-documented turbulent flow control benchmark cases in both two and three dimensions.