Vous êtes sur la page 1sur 11

Aero-Structural Airfoil Design Optimisation for Cruising

Conguration using Tabu Search


Pedro Pablo Guerrero Vela

Escuela T ecnica Superior de Ingenieros de Sevilla, C/Camino de los Descubrimientos S/N, Espa na
Multi-Disciplinary Design Optimisation is the best approach for nding optimum solutions to real-world
problems, where conictive requirements from different nature converge. Thus, the optimisation of such
systems represents a really challenging task from a technical point of view. In this paper, the development
of a fully integrated automatic tool for continuous aero-structural wing design optimisation purposes will
be introduced. Five different modules have been congured and interconnected to make the tool a reality,
representing each of them a state-of-the-art application: a) Multi-Objective Tabu Search (MOTS) a stochastic
optimiser enhanced with local search strategies, b) a parameterisation tool based on Free Form Deformation
(FFD) techniques, c) ow analyser congured for automatic execution, d) in-house uid-structure interface
and e) structural analyser. An initial optimisation study of a simplied wing based on the Airbus Test Case-
A airfoil has been performed. Two objective functions were assessed, these are: drag to lift coefcient and
the maximum Von Mises stresses that appear at any point on the structure. The preliminary results reveal the
potential of the tool for the successful bi-objective optimisation, demonstrating the integrity and functionality of
the design tool. Outstanding improvements for both objective functions are found and three different solutions
from the pareto front are analysed and compared against the datum conguration.
Nomenclature
C
l
lift coefcient CFD Computational Fluid Dynamic
C
d
drag coefcient FSI Fluid Structural Interface
C
p
pressure coefcient FEA Finite Element Analysis
VM Von Mises RANS Reynolds-averaged Navier-Stokes
GUI Graphical User Interface FFD Free Form Deformation
I. Introduction
T
HIS research is part of a major Multi-Objective Optimisation programme of work under development in the
Department of Power and Propulsion at Craneld University. The project considered represents a rst Aero-
Structural Optimisation study on the way towards a more complete multi-disciplinary design capability.
From an engineering point of view, design is the rst and most important step for the materialisation of a product
that meets a human need. Before numerical analysis was a reality only simple analytical models were used to explain
the physics involved in a problem and eventually to reach a solution. The main disadvantage of using this approach is
the lack of accuracy that those models achieve.
Recently, the quality of these models has been increased thanks to advances in computational data processing as
well as simulation techniques allowing engineers to achieve better designs. Nevertheless, most current methods to
produce good designs still rely primarily on engineers experience and capabilities. However, the means to enhance
these, using computers, has become ever clearer and more feasible in recent years. Computational design optimisation
techniques can now in principle allow designers to explore vastly more design parameters simultaneously, and also to
consider multi-dimensional aspects that could not previously be envisaged.
In order to achieve such an ambitious objective, the optimisation algorithm
1
plays the main role, followed by the
multi-physics numerical simulation. Accuracy of the numerical models is a key parameter, always in conict with

MSc, double degree student at Department of Power & Propulsion, School of Engineering, Craneld University, Craneld MK43 0AL, United
Kingdom
1 of 11
American Institute of Aeronautics and Astronautics
the computational cost that it requires. One of the main constrains is the wall-clock time required for the whole
optimisation, as the solution has to be obtained in a sensible gap of time.
The present project has built on earlier multi-objective, but single-discipline, Aerodynamic Optimisation work
performed and similarly reported by a previous student
2
within the Power & Propulsion Sciences Group, to develop
a fully integrated and automated Aero-Structural Optimisation tool. This tool, in principle, enables the optimisation
of a wing shape for cruise conguration with an steady analysis (non-aeroelastic), with the objective of minimising
two criterias of different nature, i.e. aerodynamic (
C
d
C
l
) and structural (VM stresses). The datum geometry has been
generated using the Airbus Test Case-A airfoil. Several simplications have been used to model the real physics and
geometry of the problem. These are necessary for the automation of the process and for the generation of results in
the time available for the research. Commercial simulation packages have been used, exploiting their batch mode
capability (no GUI displayed).
II. Methodology
An overview of the whole optimisation process can be seen in Figure 1. MOTS (Multi Objective Tabu Search) is
the optimiser, a kind of brain that controls the new design that is going to be analysed using the design variables,
its decisions are based on the information collected from the simulation results. Basically, once the new set of design
variables have been produced by the optimiser, the parameterisation tool (FFD) generates the new airfoil that, once
extruded, creates the whole wing. Within the CFD module two important results are produced,
C
d
C
l
and pressure eld.
The rst one will be directly used as an objective function (to be minimised), whereas the second one is used for the
FEA. Due to the different nature of each problem, different meshes are required for each analysis. This increases the
complexity of the problem as another module is necessary to send the information between modules, i.e. FSI (Fluid
Structure Interface). Once the structural data is available, the second objective function can be assessed.
Figure 1: Optimisation procedure
A. Datum Geometry
The starting point of the optimisation process is the datum conguration, see Figure 2. The data available for this was
an Airbus Test Case-A airfoil prole in a deployed conguration. The rst step was therefore to construct a stowed
airfoil section from this, keeping smoothness and continuity of the external surface, as illustrated in Figure 2 (a)-(c).
Once the resulting cruise conguration was created, it was extruded to create a quasi-3D datum wing shape, with a
constant sectional prole to initially avoid wing tip shape end effects. Two beams located at 25% and 50% of chord
from the leading edge are the last elements added to the datum conguration, these are necessary for the structural
analysis.

Note: Due to condentiality reasons, none of the absolute values of the geometry are presented in this paper.
2 of 11
American Institute of Aeronautics and Astronautics
(a) Slat (b) Stowed vs. Deployed conguration (c) Flat
(d) 3D extrusion
Figure 2: Datum Wing.
B. Parameterisation
In this work, FFD
3
(Free Form Deformation) was chosen for the airfoil shape parameterisation. This technique is
widely used in computer assisted geometry design and animation. The method enables the deformation of an embedded
object by modifying the shape around it (this space is controlled by a set of lattice points in a grid). This technique
was chosen due to its ability to modify a complex geometry with a low number of variables and high degree of
freedom. In addition, this approach enables local deformation with derivative continuity of any degree with adjacent
and undeformed regions of the model. The last characteristic is highly valuable in this type of problems, where slight
deformation of critical regions of the airfoil has a huge impact on its performance, e.g. leading and trailing edge shape
and suction peak point intensity and relative position.
The practical implementation to the datum airfoil can be seen in the Figure 3. The polygon that embedded the
airfoil is highlighted in red, 8 lattice points dened it. However, only the four interior lattice points can be used for
controlling the shape. The other must be xed to keep the leading and trailing edges in their xed positions. Hence, 8
design variables dene the shape (horizontal and vertical movement of the four interior lattice points).
Figure 3: Airfoil Design Variables.
The variability of the design variables were chosen in the range 0.3 to 0.3. This design space was exible enough
to account for major changes in the airfoil, for example, it can be seen in Figure 4 eight randomly deformed airfoil
which show the suitability of the technique for airfoil optimisation. The design variables values are written in the top of
each prole with the arrangement: x
1
y
1
y
3
y
2
x
4
y
4
x
3
x
2
. Changes in most of the parameter
that dene an airfoil prole are appreciated; thickness changes (cases 4, 6 and 8); leading edge shape modication
(cases 2, 5, 6 and 7); trailing edge morphology (cases 1, 3 and 7); suction peak region displacement (cases 1, 5, 6, 7
and 8). This is not an exhaustive list of examples but it shows the potential of the method. The number of combinations
that can be achieved is innite.
C. Computation Fluid Dynamic (CFD) analysis
CFD was one of the most important tasks to be congured within the optimisation process. This step required 90%
of the wall-clock time within each optimisation loop, therefore, a large effort was done to reduce the computational
3 of 11
American Institute of Aeronautics and Astronautics
0 0 0 0 0.3 0 0 0
1
0.3 0 0 0 0 0 0 0
2
0 0 0 0 0 0.3 0 0
3
0 0.3 0 0 0 0 0 0
4
0 0 0.3 0 0 0 0 0
5
0 0 0.3 0 0 0 0 0
6
0.2 0.05 0.2 0.29 0 0.25 0.1 0.3
7
0 0 0 0.3 0 0 0 0
8
Figure 4: FFD Airfoil (Datum in black).
cost without any major loss in accuracy. That is why the wing is modelled as an extrusion of an airfoil, hence, a 2D
uid analysis can be performed, avoiding a costly 3D one. In the same way of thinking, a steady state analysis was
congured (aeroelastic effects are neglected). The analysis was performed using the commercial package Ansys Inc.
CFX software V.13.
Both simplications did not affect the accuracy of the solution, but moved the model away from reality. Nev-
ertheless, they were necessary for making the optimisation a feasible problem constrained by the time available.
The validation of the aero-structural tool was not affected by these simplications as the simplications only mean
a reduction in computational time. In the Table 1 the parameters for the uid analysis are presented.
Compressibility effects are important in this problem as the Mach number can duplicate its far-eld value in the
peak-suction region of the upper camber of the airfoil, therefore, heat transfer phenomena was taken into account
within the RANS. Two options are available in Ansys for compressibility effects to be taken into account, i.e. Thermal
Energy and Total Energy. The rst one does not take into account kinetic energy effects throughout the ow simulation,
therefore, fewer equations need to be solved, whereas the second one is a full analysis. The assessment of both models
led to the conclusion that only 5% of CPU time is saved with the rst model, and the solution is 30% less accurate
(proved in the Validation and Verication analysis). The conclusion was to use the Total Energy alternative.
In this problem the main interest was to solve the boundary layer with high accuracy, as drag and lift dene one of
the objective functions. Hence, k w based methods were the most suitable for the nature of the problem, as Trapani
2
showed the good performance of SST (Shear Stress Transport) to catch these effects. The particular benets that this
model provided were the automatic near-wall treatment that switched from wall-functions methods to low-Re ones
when the renement of the grid was not good enough, therefore, was a robust model for the variable airfoil that was
assessed. In the Table 2 the solver settings are shown.
Table 1: CFD modelling.
Material: Air ideal gas at 1 atm and 298.16 K
Heat Transfer Model: Total Energy (compressible ow)
Turbulence Model: Menter Shear Stress Transport k w combined with transitional model
Flight conditions: Re = 6x10
6
Mach number=0.24 Inlet axial speed=82.3m/s
Table 2: Solver control settings.
Advection Scheme: High Resolution
Maximum: number of iteration and Run time: 200 3 hours
Time-scale control: Auto time-scale
Residual Target 1x10
6
4 of 11
American Institute of Aeronautics and Astronautics
1. Grid generation
The keystone in the production of an automated numerical ow simulation is the meshing stage. The uid domain has
to be discretized in a grid of points where the RANS equations can be solved. The accuracy of the solution is strongly
inuenced by the distribution of points, renement is needed in regions where higher velocity gradients are found, e.g
boundary layer or sudden geometry changes. This problem is even more difcult when the geometry under study is not
known before hand, therefore, the meshing approach has to be congured with this degree of freedom. It was found
quite useful to generate this automatic meshing strategy using Ansys ICEM CFC V.13 mainly thanks to the Blocking
Strategy available for Structured meshes.
The Structured
4
approach was chosen for three reasons: 1) it saves memory (the grid points can be calculated)
2) the better alignment of the elements with the ow increases the quality and the convergence to the solution 3) the
Blocking Strategy is feasible. However, there is an important drawback, i.e. it generates useless highly rened regions.
Overall, the performance of the Structured mesh is preferred in this type of applications. The combination of this type
of mesh and O-grid deformation of the elements generated

were used in the grid strategy. Smoothness techniques
were applied to the resultant grid, see Figure 5 (b).
In the Figure 5 (a) a diagram of the Blocking Strategy can be seen. The main advantage of this strategy is the
possibility to dene a mesh generation, setting the parameters that dene it without the knowledge of the real geometry.
Instead, these parameters are attached to the edges and vertex

that dene the different blocks. In the Figure 5 (c) the
details of the mesh produced (in an automatic way), in the boundary layer is shown.
(a) Blocking Strategy (b) Structure Mesh, O-grid Strategy, Fine Mesh.
(c) Detail of the boundary layer renement.
Figure 5: Automatic Mesh Generation
2. Validation and Verication
In order to use CFD results for engineering purposes, the CFD code must be checked to ensure its level of condence
and reliability. Some guidelines for best practice in CFD already exist, see AIAA G-077-1998. According to these
guidelines, two concepts are assessed: Verication: the process of determining whether a model accurately represents

The O-grid strategy improved the adaptation of the elements to the curved camber of the airfoil, aligning in a better manner the elements with
the ow.

Not to be confused with lines and points that dene the geometry
5 of 11
American Institute of Aeronautics and Astronautics
the users concept of the case under study and its solution. Validation: the process of determining the degree to which
a model represents the real world.
For the CFD verication, quantication of the errors produced in the simulation has to be assessed. In high-quality
CFDwork, monotonic decrease of the discretization error should be demonstrated on two or three levels of renement
5
.
Validation assessment deals with the uncertainty of CFD simulation, i.e. all the potential deciencies caused by a lack
of knowledge
In order to analyse the quality of the solution, three different levels of renement were studied, all of them
following the same pattern. Hence, coarse, medium, and ne meshes with 55000, 125000 and 220000 elements
respectively were analysed. The geometry used for the Validation and Verication purposes is the asymmetric airfoil
Naca-23012. It was selected from the wide variety available in the literature as it represents a general case for a cruise
airfoil. Additionally, there is no doubt about the denition of its geometry as it is a NACA prole, signifying that
there is an algebraic way to generate the geometry
6
, or indeed concerns regarding the experimental data, which has
been used in numerous works
7
. In the Table 3 the errors between the numerical calculation using the aforementioned
strategy and the data available in the literature is shown. The error reduction with improved grid renement proves the
verication of the analysis. In addition the validation is proved using the ne mesh, as the relative errors are smaller
than 5%.
Table 3: Validation Results, relative errors.
Mesh Type Cl [%] Cd [%]
Fine 0.39 3.72
Medium 0.41 15.02
Coarse 3.75 40.52
The Validation and Verication are extended to the automatic analysis of the variable geometries thanks to: a) the
monotonic behaviour for the mesh generation provided by Blocking Strategy, b) the narrow design space, which only
allows for small variations of the geometry under consideration, c) same ow conditions for all cases.
D. Fluid-Structure Interaction (FSI) considerations
Once the pressure distribution over the airfoil is obtained in the previous step it has to be mapped to the wing (extruded
2D airfoil prole). Due to the different nature of aerodynamic and structural problems, the grids built to analyse them
are totally different. The best alternative to solve this disagreement is using interpolation procedures which transfer
data from both media
8
. In this optimisation project, the uid analysis was performed in 2D, whereas the structural
analysis was in 3D, therefore, none of the previous alternatives were feasible.
The use of an in-house FSI code developed in Matlab, version R2011a, was necessary. Basically, the pressure
distribution was approximated with a polynomial interpolation. Hence, the coefcients of the interpolation law could
be transferred to the structural analyser. The errors introduced in the model were monitored and the results proved
that they were kept under the same average value (low dispersion), thus, the inuence in the optimisation process was
minimum.
E. Finite Element Analysis (FEA)
Abaqus Standard V6.10 was utilised for the structural analysis. Once again, the automation of the whole structural
analysis represents a challenge. The mathematical model of the wing incorporates: a) skin, extruded airfoil prole with
a thickness of 15mm, modelled with shell elements, and b) two spars, modelled as shells elements (due to automation
constrains), with a thickness of 170mm. The material used was Alclad 2024 T3
9
. The discrete model used S4R
10
as elements types (Abaqus nomenclature) and 88000 of them

. These elements were reduced integration elements,


therefore the wall-clock time per analysis was reduced by half, whereas, the accuracy was only reduced by 1% (from
the Safe-Side). In the Figure 6 the nal model is shown. The detail of the elliptical pressure distribution prole can be
seen. That prole is a simplication that accounts for the 3D effects that appear when considering the fuselage and the
wing tips.

Quadrilateral, 1 point of integration, general-purpose, conventional nite-membrane-strain shell elements.

The number of element used was chosen after a convergence analysis as the one done for the CFD Validation and Verication
6 of 11
American Institute of Aeronautics and Astronautics
Figure 6: Pressure distribution on the wing.
F. Multi Objective Tabu Search (MOTS)
Several optimisation algorithms have been developed over the years to deal with different kinds of optimisation
problems. No Free Lunch (NFL) theorems
11
show that there is no one individual optimiser better than another,
it just depends on the nature of the problem they are trying to solve. Real-word problems, in particular aerodynamics,
are highly constrained. Therefore, their design space has a high proportion of unfeasible solutions. Tabu Search has
shown a particularly good performance in dealing with this kind of problem as Harveys work has shown
12
, which is
mainly thanks to the combination of local search paradigm and the different types of memory it uses. This particular
MOTS has been developed by Jaeggi et al.
13
, adapting a single-objective Tabu Search software implemented by Connor
and Tilley
14
. Kipouros T. et al.
1
successfully used it for turbomachinery blade optimisation.
III. Results
The optimisation results are shown in Figure 7. They represent the non-dominant solutions found by the optimiser
after 373 CFD and FEA evaluations. To reduce wall-clock time, the design tool was run on a nine-node parallel PC
cluster of 3.0 GHz Intel 5160 Xeon dual-core machines, exploiting the multi-level parallelization capability of the tool.
The whole optimisation was coded in C++, but a section of Matlab code was necessary for the FSI module and Python
code for the extraction of the maximum VM stresses from the Abaqus database, both with the purpose of reducing
running time. However, each whole loop of the optimisation need about 60 minutes to be performed. The results
shown are the solutions found after 16 days running in the above mentioned cluster.
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
1.05
Normalised

Cd
Cl

N
o
r
m
a
l
i
s
e
d
M
a
x
.
V
o
n
M
i
s
e
s
s
t
r
e
s
s
Optimisation Result


Datum
CASE 1

CASE 2

CASE 3

Tabu points
Optima points
Figure 7: MOTS results: Pareto front and Search Pattern.
7 of 11
American Institute of Aeronautics and Astronautics
From the search pattern it can be seen that almost all the points visited improve the performance of both objective
functions, which proves the efciency of the optimiser. Although the number of optimisation iterations is not sufcient
to say that the design space has been entirely explored, the tendency shows very successful solutions. Eleven Pareto
Points

were found. In order to choose one of the Pareto equivalent solutions, preferences or relative weighting of the
objective functions must be introduced. These would act as the criteria in the task of choosing a single solution from
the trade-off that represents the Pareto-optimal, however, this is not the purpose of this project.
Three relevant points from the pareto-front were chosen for deeper analysis: Case 1 (best aerodynamic perfor-
mance), Case 2 (trade-off solution), and Case 3 (best structural performance). The geometries of these three points
can be seen in the Figure 8 and the relative improvements to the datum performance can be seen in Table 4. It is
noticeable that the aerodynamic objective function improved by 54.6%, mainly due to a huge improvement in drag
coefcient. In Figure 9 the Cp are shown for the three cases. In Figures 10 and 11, the pressure contour and velocity
stream lines are shown. The pictures reveal that the optimiser is moving the solution to regions where the suction-peak
is reduced, which reduces the adverse pressure gradient, delaying the detachment of the ow, thus, reducing the drag
coefcient. Geometrically it is obtained improving the alignment of the leading and trailing edges with the ow as well
as smoothing the upper camber curvature, see Figure 8.
Table 4: MOTS results for cases 1, 2 & 3 (
1
Maximum Von Mises stress).
Improvements [%] Norm. C
d
/C
l
Norm. MVM
1
C
l
/C
l,datum
C
d
/C
d,datum
Case 1 54.6 3.6 6.5 51.6
Case 2 39.1 21.58 2.5 37.7
Case 3 29.1 32.1 0.6 28.6
Datum
Case 1
(a) Case 1
Datum
Case 2
(b) Case 2
Datum
Case 3
(c) Case 3
Figure 8: Airfoil proles
In the gure 12 the maximum VM stress is shown. The tendency of the optimiser in this case is totally reasonable.
Considering the wing as a clamped beam (very rst order), the optimiser moves the solution toward geometries with
higher second moment of inertia and increasing skin length. These new geometries provide higher bending and
torsional stiffness, therefore, the stress state in the wing is reduced, even though the lift coefcient is slightly increased.
Improvement until 32.1% has been achieved in reducing the maximum VM stress. In the case 1 the point where the
maximum stress appears is in the junction between the forth main spar and the lower camber, whereas in the case 2
& 3 that point is moved toward the rear main spar, which proves the tendency of the optimiser to nd solutions that
improve the exural behaviour of the wing (as the rear spar is in the middle of chord of the airfoil).

The Pareto-optimal set are the design vectors whose objective functions are Pareto equivalent, i.e. any of the design vectors are dominant over
the other, in this way, a boundary denes the set of solutions that cannot be improved, without degrading the solution in another aspect.
8 of 11
American Institute of Aeronautics and Astronautics
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1.5
1.2
0.9
0.6
0.3
0
0.3
0.6
0.9
1.2
1.5
1.8
2.1
2.4
x
chord
-Cp
Pressure distribution comparison


Datum
Case1
Case2
Case3
Figure 9: Pressure Coefcients
(a) Datum (b) Case 1
(c) Case 2 (d) Case 3
Figure 10: Pressure Contour
9 of 11
American Institute of Aeronautics and Astronautics
(a) Datum (b) Case 1
(c) Case 2 (d) Case 3
Figure 11: Velocity Streamlines
(a) Datum (b) Case 1
(c) Case 2 (d) Case 3
Figure 12: Von Mises Stress
10 of 11
American Institute of Aeronautics and Astronautics
IV. Conclusions
The successful application of aero-structural airfoil design optimisation for cruise conguration has been presented.
The objective functions of the optimisation process were the drag to lift coefcient and the maximum VM stress that
appears at any point of the wing. The design variables were presented by moving the lattice points that dene the
polygon used to deform the airfoil under the FFD technique. In the test case analysed, the MOTS optimiser has been
able to produce newdesigns that increase the performance of the datumconguration of the wing, showing the potential
of using numerical optimisation methods for coupled aero-structural design.
More realistic mathematical models can be built in future works. Aeroelastic effects and 3D analysis can be
congured in the aerodynamic study. More structural components can be added to the structural analysis. Improved
models for the spars can be assessed as well. Multi-physics software can be analysed to improve the FSI data transfer.
Always keeping in mind the need for automation of the process.
Acknowledgements
The lead author gratefully acknowledges his MSc project supervisors Professor Mark Savill

and Dr. Timos


Kipouros

for the opportunity to work on their project, in developing multi-disciplinary optimisation. Special thanks
must go to Murray Cross, Airbus Technology Product Leader, FuSim, Giuseppe Trapani, Airbus/EPSRC EngD student
Craneld University for their collaboration in completing this project and Airbus Operations Ltd for providing technical
data for the Test Case A. Finally, last but not least, I am thankful to Relaciones Exteriores, Escuela T ecnica Superior de
Sevilla, for giving me the opportunity to undertake this MSc in Thermal Power at Craneld University, and specically
to Dr. Mara de Los

Angeles Martn Prats for her continued support and trust.
References
1
Kipouros, T., Jaeggi, D., Dawes, W., Parks, G., Savill, A., and Clarkson, P., Biobjective Design Optimization for Axial Compressors Using
Tabu Search, AIAA journal, Vol. 46, No. 3, 2008, pp. p. 701711.
2
Trapani, G., Multi-Objective Optimization of 2D High-Lift Airfoil Conguration using Tabu Search, in: VI Pegasus-AIAA Student
Conference, 28th - 30th April, Seville, 2010.
3
Sederberg, T. W. and Parry, S. R., Free-form deformation of solid geometric models, SIGGRAPH, 1986, pp. 151160.
4
Thompson, J., Soni, B., and Weatherill, N., Handbook of grid generation, CRC Press, NY, United States, 1999.
5
Versteeg, H. and Malalasekra, W., An Introduction to Computational Fluid Dynamics: The Finite Volume Method., Pearson Education
Limited, Harlow, UK, 2007.
6
Abbott, I. and Von Doenhoff, A., Theory of wing sections: including a summary of airfoil data, Dover Pubns, UK, 1959.
7
Lowry, J., Wind-Tunnel Investigation of an NACA 23012 Airfoil with Several Arrangements of Slotted Flaps with Extended Lips, NACA
Langley Memorial Aeronautical Laboratory, VA, United States, 1941.
8
Hounjet, M. and Meijer, J., Evaluation of Elastomechanical and Aerodynamic Data Transfer for Non-Planar Congurations in CAE
Analysis, Proceedings of the International Forum on Aeroelasticity and Structural Dynamics, Manchester, UK, published by the Royal Aeronautical
Society, pp. 1011.
9
Kaufman, J. and Rooy, E., Aluminum alloy castings: properties, processes, and applications, Asm Intl, UK, 2004.
10
ABAQUS, I., ABAQUS Analysis Users Manual, Version 6.7, Inc., Rising Sun Mills, United States, 2007.
11
Wolpert, D. and Macready, W., No free lunch theorems for optimization, Evolutionary Computation, IEEE Transactions on, Vol. 1, No. 1,
1997, pp. p. 6782.
12
Harvey, S., The design optimisation of turbomachinery blade rows, Ph.D. thesis, Cambridge University, Departament of Engineering, 2002.
13
Jaeggi, D., Parks, G., Kipouros, T., and Clarkson, P., The development of a multi-objective Tabu Search algorithm for continuous
optimisation problems, European Journal of Operational Research, Vol. 185, No. 3, 2008, pp. p. 11921212.
14
Connor, A. and Tilley, D., A tabu search method for the optimization of uid power circuits, Proceedings of the Institution of Mechanical
Engineers, Part I: Journal of Systems and Control Engineering, Vol. 212, No. 5, 1998, pp. p. 373381.

Senior Member AIAA, Professor of Computational Aerodynamics Design and Head of Power & Propulsion Sciences Group, Department of
Power & Propulsion.

Visiting Research Fellow, Power & Propulsion Sciences Group, Department of Power & Propulsion.
11 of 11
American Institute of Aeronautics and Astronautics

Vous aimerez peut-être aussi