Towards a robust parallel solver for large-scale industrial flow simulations
In this work a CFD analysis is done on incompressible viscous flows using Finite Volume schemes implemented in the open-source software OpenFOAM. The objective of this study is twofold: to gain experience with the software and to define a set of best practices when running large-scale cases in OpenF...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2022 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/376517 |
| Acceso en línea: | https://hdl.handle.net/2117/376517 |
| Access Level: | acceso abierto |
| Palabra clave: | High performance computing High-performance computing OpenFOAM Finite Volume scheme large-scale industrial flows Càlcul intensiu (Informàtica) Àrees temàtiques de la UPC::Informàtica |
| Sumario: | In this work a CFD analysis is done on incompressible viscous flows using Finite Volume schemes implemented in the open-source software OpenFOAM. The objective of this study is twofold: to gain experience with the software and to define a set of best practices when running large-scale cases in OpenFOAM using parallel architectures. The first objective is obtained by testing three academic benchmarks, namely the lid- driven cavity, the flow over a backward facing step, and flow past a circular cylinder. The validation of these results is made by contrasting them to the data available in the literature. The second objective was fulfilled by studying two large-scale industrial problems, laminar flow inside an S-bend and turbulent external flow around a car. For the latter, the DriveAer geometry has been used. The analysis of these high- performance computing studies has been defined in terms of the relative efficiency and speed up for the two problems. The studied cases have been scaled up until 84 CPUs for the S-bend, and until 224 CPUs for the vehicle geometry. Furthermore, the performance of three partitioners, namely the simple, hierarchical, and scotch decomposers have been evaluated. |
|---|