National Technical University of Athens (NTUA)
Συζυγείς μέθοδοι για τυρβώδεις ροές, με εφαρμογή στη βελτιστοποίηση μορφής ή τοπολογίας και στο στιβαρό σχεδιασμό
Abstract
dc:descriptionThe present dissertation deals with the mathematical formulation, programming and validation ofadjoint methods for the computation of sensitivity derivatives of objective functions related toaerodynamics/hydrodynamics and the utilization of the latter in optimization algorithms. Methods basedon both the discrete and continuous adjoint approaches are presented. Academic and industrial casesare tackled in the fields of shape optimization, topology optimization and optimization underuncertainties (robust design).Regarding shape optimization, the continuous adjoint method is extended to cover incompressibleflows governed by the low-Re number Launder-Sharma k-ε and the high-Re number Spalart-Allmarasturbulence models, overcoming the implications of neglecting the differentiation of these models on theoptimization process. A significant part of the thesis is concerned with applications of the developedmethods to relevant industrial problems. In specific, the drag minimization of passenger cars designedby a European automotive industry is tackled, using the developed exact continuous adjoint method. Inaddition, surrogate aeroacoustic problems of the automotive industry, dealing with the noise perceivedby the driver are faced using the developed software. Moreover, a feasible continuous adjointformulation for drawing drag and lift sensitivity maps based on the DES Spalart-Allmaras model isproposed.In addition, the developed software is applied to the optimization of a Francis turbine runner. Thecontinuous adjoint method is used to calculate the sensitivity derivatives of various cost functionsincluding target velocity distributions at the outlet of the runner, changes in the hydraulic head andcavitation suppression. Regarding topology optimization, the present thesis extends the mathematicalformulation to cover 3D incompressible flows governed by the aforementioned turbulence models.Moreover, frequently used industrial constraints are tackled, making the procedure more user-friendlyand requiring less (or no) trial-and-error iterations. The developed software is used for designing an airconditioning duct placed on a passenger car as well as a school formula's plenum chamber.Aiming at lower turn-around optimization times, a truncated-Newton algorithm is developed for topology optimization problems. The proposed algorithm makes use of Hessian-vector products to approximatethe second-order design variables correction, at a feasible computational cost. The method is appliedto topology optimization problems with encouraging results.Robust design is the procedure of designing aerodynamic shapes with acceptable performance, notonly at the design point, but in a range of operating conditions. The Second- Order Second-Momentmethod is used to compute the mean value and standard deviation of an objective function, requiringthe computation of first- and second-order sensitivities of this function with respect to the uncertain variables. If the minimization of these metrics through a deterministic algorithm is targeted, derivativesup to third-order must be computed. The optimal combination of direct differentiation and the adjointmethod to compute the necessary derivatives is proposed and applied.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2013
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Papoutsis-Kiachagias, Evangelos
- Παπουτσής-Κιαχαγιάς, Ευάγγελος
Subjects
dc:subject × 18- Υπολογιστική ρευστοδυναμική
- Συνεχής και διακριτή συζυγής μέθοδος
- Συζυγή μοντέλα τύρβης
- Βελτιστοποίηση μορφής και τοπολογίας
- Μέθοδοι Newton με αποκοπή
- Στιβαρός σχεδιασμός
- Παράγωγοι ευαισθησίας υψηλής τάξης
- Computational fluid dynamics
- Continuous and discrete adjoint methods
- Adjoint turbulence models
- Shape and topology optimization
- Truncated newton methods
- Robust design
- High-order sensitivity derivatives
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Μηχανολόγου Μηχανικού
- Engineering and Technology
- Mechanical Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/38630