Evaluation and development of an in-house GPU based CFD solver
Ladda ner
Publicerad
Författare
Typ
Examensarbete för masterexamen
Master's Thesis
Master's Thesis
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
An in-house GPU based CFD solver has been validated with respect to total pressure loss on a turbine rear structure compared to results from ANSYS Fluent and the previous version of the code itself. The order of the normalized error was found to be O(10−2) and O(10−4) respectively. The solver has also been analyzed with respect to its code structure, finding it made up of three parts: a Python part, a C part and a CUDA C++ part. A critical memory error was found in the C part which pronounces from seemingly trivial source code changes. It does not present as a problem, however, if these trivial source code changes are reversed. Previous investigations on the solver found incorrect total pressure readings on the domain inlet, which were located and discussed in the source code. Finally, asymptotic convergence criteria were added and their settings optimized to give a compromise between solver speed and accuracy for a generic turbine rear structure simulation.
Beskrivning
Ämne/nyckelord
Validation, G3D, CFD, Compressible, Explicit, Runge-Kutta, CUDA, C/C++, Python, Convergence Criteria