National Technical University of Athens (NTUA)
Development of shape parameterization techniques, a flow solver and its adjoint, for optimization on GPUs: turbamachinery and external aerodynamics applications
Abstract
dc:descriptionThis thesis is concerned with the development of methods and numerical tools for fast and cost-efficient aerodynamic/hydrodynamic shape optimization. Developments related to the ensemble of building blocks of a shape optimization framework, namely a) the Computational Fluid Dynamics (CFD) analysis software, b) the optimization methods and c) the shape parameterization techniques are performed. These reduce the overall optimization cost and make them appropriate for use in industrial design processes.The GPU-enabled in-house (U)RANS solver PUMA (Parallel Unstructured Multi-row Adjoint), is enriched with an incompressible flow solver based on the Artificial Compressibility (AC) approach. The RANS equations are expressed using the Multiple Reference Frame (MRF) approach. GPU-specific programming techniques and numerical schemes, developed in previous works on the PUMA software, are revisited, upgraded and re-evaluated, resulting to a GPU implementation which is up to 40 × faster than the equivalent CPU one. This is achieved by careful code restructuring, delicate GPU memory handling, the use of Mixed-Precision Arithmetics (MPA) and efficient utilization of the NVIDIA CUDA programming environment. A series of test cases are studied for validation and verification purposes of the PUMA compressible and incompressible flow solver. These concern the turbulent flow around two isolated airfoils, the flow through a convergent-divergent transonic diffuser and over a backward-facing step and the flow around a Horizontal Axis Wind Turbine (HAWT) blade. Furthermore, the solver is also validated/verified on the prediction of the flow field around the ONERA M6 wing, inside a propeller type water turbine and a high-pressure turbine stator.Throughout this thesis, the continuous adjoint method is exclusively used and its development for both the compressible and the AC-based incompressible flow solver is proposed. Two different expressions for the sensitivity derivatives are developed, namely the Surface Integral (SI) and Field Integral (FI) adjoint formulations. In both formulations, the variations of the eddy viscosity due to shape changes are taken into account by differentiating the Spalart-Allmaras turbulence model.In the field of shape parameterization, a parametric modeler for turbomachinery blade rows is developed, namely the GMTurbo software. The GMTurbo software, together with the PUMA flow solver and the in-house evolutionary algorithm based optimization tool (EASY), is employed for the optimization of a propeller type water turbine.Additionally, a Free Form Deformation technique, based on Volumetric NURBS, is developed to support optimization loops. A new strategy for extending the applicability of this technique to the shape optimization of turbomachinery components, based on intermediate coordinate system transformations, is proposed.The Volumetric NURBS parameterization is differentiated and used for the GB shape optimization of a 2D transonic airfoil and that of a linear compressor cascade. The method is also applied to the shape optimization of a 3D transonic wing and, finally, that of a peripheral high-pressure turbine nozzle guide vane.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2019
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Tsiakas, Konstantinos
- Τσιάκας, Κωνσταντίνος
Subjects
dc:subject × 20- Υπολογιστική ρευστοδυναμική
- Επεξεργαστές καρτών γραφικών
- Συνεχής συζυγής μέθοδος
- Βελτιστοποίηση μορφής
- Θερμικές και υδροδυναμικές στροβιλομηχανές
- Ανεμογεννήτριες
- Παραμετροποίηση μορφών
- Εξελικτικοί αλγόριθμοι
- Computational fluid dynamics
- Graphics processing units (GPUs)
- Continuous adjoint methods
- Shape optimization
- Shape parameterization
- Wind turbines
- NURBS
- Evolutionary algorithms
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Μηχανολόγου Μηχανικού
- Engineering and Technology
- Mechanical Engineering
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/45281