Computational Fluid Dynamics (CFD) is a fundamental tool for the analysis and optimization of aerodynamic designs for internal and external ows. Alongside with the research of formulations that can accurately predict experimental data, a fundamental goal is also represented by solvers computational e ciency. In this paper a GPU-accelerated density-based and a coupled ressure-based solvers are proposed as two possible solutions to accelerate the simulation of compressible viscous ows. The first strategy is supported by the nowadays GPUs single precision performances, reaching computational power of the order of TFLOPS at few hundreds of USD and providing high performance/Watt ratios. However they require speci cally designed algorithms and programming languages. The one adopted in this work is based on OpenCL.The second strategy has instead the aim to resolve the usual convergence deterioration of the SIMPLE family algorithms improving the variables coupling. In this work an introduction of the fundamental details of the two formulations and solver architectures is provided. A validation campaign using experimental measurements is then presented for the most important applications in propulsion, external aerodynamics and supersonic flows.
Comparison of Acceleration Techniques on CFD Open-Source Software for Aerospace Applications
ROMANELLI, GIULIO;GADDA, ANDREA;
2015-01-01
Abstract
Computational Fluid Dynamics (CFD) is a fundamental tool for the analysis and optimization of aerodynamic designs for internal and external ows. Alongside with the research of formulations that can accurately predict experimental data, a fundamental goal is also represented by solvers computational e ciency. In this paper a GPU-accelerated density-based and a coupled ressure-based solvers are proposed as two possible solutions to accelerate the simulation of compressible viscous ows. The first strategy is supported by the nowadays GPUs single precision performances, reaching computational power of the order of TFLOPS at few hundreds of USD and providing high performance/Watt ratios. However they require speci cally designed algorithms and programming languages. The one adopted in this work is based on OpenCL.The second strategy has instead the aim to resolve the usual convergence deterioration of the SIMPLE family algorithms improving the variables coupling. In this work an introduction of the fundamental details of the two formulations and solver architectures is provided. A validation campaign using experimental measurements is then presented for the most important applications in propulsion, external aerodynamics and supersonic flows.File | Dimensione | Formato | |
---|---|---|---|
MANGL01-15.pdf
Accesso riservato
Descrizione: Paper
:
Publisher’s version
Dimensione
790.13 kB
Formato
Adobe PDF
|
790.13 kB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.