DEGREE WITH HONOURS IN MSC AEROSPACE ENG

DEGREE WITH HONOURS IN
MSc. AEROSPACE ENGINEERING
MSc Final Year Project Report

School of Engineering and Technology
University of Hertfordshire
MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

(AERODYNAMICS & STRUCTURAL)
Report by
Akinwande O. Cole
Supervisor
Dr. Jason Knight
Date
10/09/2013
DECLARATION STATEMENT
I certify that the work submitted is my own and that any material derived or quoted from the
published or unpublished work of other persons has been duly acknowledged.
Student Full Name: Akinwande O. Cole
Student Registration Number: 11489561
Signed: …………………………………………………
Date: 10/09/2013
School of Engineering and Technology MSc Final Year Project Report
ABSTRACT
The quest to operate at the cutting edge of technology is a worthwhile effort but requires
intensive research and brilliant implementation of logical thoughts aimed at an achievable
objective. This gives distinct results with excellent achievements and impeccable brilliance in
technology. This applies to optimization and multidisciplinary design optimization in engineering
designs.
The objective is to achieve an optimum wing configuration defined by three design objectives
within the aerodynamic and structural discipline – minimizing the maximum Von Mises stress;
minimizing the maximum wing tip displacement; and maximizing the lift-drag coefficient – by
varying the design sweep angle, dihedral angle, and the aspect ratio of the wing. This was
achieved adopting a methodology that satisfies the NACA requirements for Multidisciplinary
Design Optimization which highlights a framework that allows: automation; analyses and re-
analyse; parameterization for easy modifications of design configurations; and leveraging
existing tools and methods.
This was done using: CATIA for designing, parameterization and modelling; STAR CCM+ for
computation of wing pressure loading; ANSYS for static structural analysis; and MATLAB as the
central executive for the multi-objective optimization using a genetic algorithm. The result was
not unique but was represented on a Pareto optimum on which any set of design parameter is
optimum. This leaves the designer the flexibility on the Pareto optimum to trade off other
objective(s) for objective(s) of higher importance.
MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING (AERODYNAMICS

& STRUCTURAL) ii
ACKNOWLEDGEMENTS
I reverence God for the grace for making this a reality.

I sincerely appreciate my project supervisor, Dr. Jason Knight for giving me the opportunity to
explore my desire to investigate this project topic and for the constant support and assistance in
ensuring that this has been a wonderful learning experience. I say a big thank you to him for
trusting my capability right from the start. My appreciation on supervision wouldn’t be complete
without appreciating my second supervisor too, Dr Rachael Cunliff, you have both been great
supervisors to me. I also acknowledge the effort of all my lecturers, whom I ran to when I
needed expertise in all the major and minor disciplines involved in my project. Big thanks to the
University of Hertfordshire at large for giving me an opportunity to undertake this piece of work.
Resounding thanks to the CD Adapco group at large, Dr. Konstantinos and co. they have been
so good with support services. They have facilitated the completion of this project with a double
500 hours licence of STAR CCM+ which has gone a long way towards the completion of the
project – This includes training sessions that have made my use of the program so vast.
The support team of MathWorks has also been very helpful with the use of MATLAB and have
both responded to my troubles during the project.
To all my friends, the Afolabi’s, all the family and friends who have helped in one way or the
other with my academic endeavour, I say thanks to you all. Generally, I feel I have been blessed
with the best people around me. Kudos to you all and I wish you the very best.
Thanks to my parents and siblings; Oluseyi, Olubisi; Eniola and Enitan. I give bunch of
numerically meshed and countless appreciation to you all for making this all a reality. This
means so much to me. You helped restore sanity when aerospace engineering shook my world
at first. This would not have been possible had I not a combination of you all in my life. You
have trusted and believed in me at times I didn’t believe it was possible; that’s what I call
“family”. I Love you all.
Lastly I say thanks to my humble self for not letting myself or anyone down. It was an excellent
experience overall. Thank God for bringing this work to a wonderful end.

& STRUCTURAL) iii
TABLE OF CONTENTS
DECLARATION STATEMENT ........................................................................................................i

ABSTRACT .................................................................................................................................... ii
ACKNOWLEDGEMENTS ............................................................................................................. iii
TABLE OF CONTENTS ................................................................................................................ iv
LIST OF FIGURES ....................................................................................................................... vii
LIST OF TABLES ...........................................................................................................................x
GLOSSARY ................................................................................................................................... xi
1 INTRODUCTION ................................................................................................................ 12
1.1 Multidisciplinary Design Optimization .......................................................................... 13
1.1.2 Definition.............................................................................................................. 14
1.2 MDO Methods ............................................................................................................. 14
1.3 CO or BLISS ................................................................................................................ 15
1.4 Approximation Methods............................................................................................... 15
1.5 Rudiments of MDO ...................................................................................................... 16
1.5.1 Objective functions .............................................................................................. 16
1.5.2 Design variables .................................................................................................. 17
1.5.3 Constraints .......................................................................................................... 17
1.5.4 Baseline design of the wing................................................................................. 17
1.5.5 Multiple disciplines .............................................................................................. 17
1.6 MOTIVATION .............................................................................................................. 18
1.7 OBJECTIVES .............................................................................................................. 19
2 LITERATURE REVIEW ...................................................................................................... 20
2.1 FSI and MDO .............................................................................................................. 20
2.2 Adoption of CO ............................................................................................................ 21
2.3 Parameterization ......................................................................................................... 21
2.4 MDO Framework ......................................................................................................... 22
3 MODELLING AND PARAMETERIZATION ........................................................................ 24
Introduction: ............................................................................................................................ 24
3.1 Modelling: .................................................................................................................... 24
3.2 PARAMETERIZATION ................................................................................................ 28
4 FLUID-STRUCTURAL INTERACTION (FSI) ..................................................................... 31
Introduction ............................................................................................................................. 31
4.1 Coupling ...................................................................................................................... 31
4.2 Mapping ....................................................................................................................... 32
4.3 Coupling Time-Step Classifications ............................................................................ 32
4.3.1 Uncoupled ........................................................................................................... 32
& STRUCTURAL) iv
4.3.2 Implicit coupling ................................................................................................... 33
4.3.3 Explicit coupling ................................................................................................... 33
4.4 Coupling Data Exchange Classifications .................................................................... 33
5 MDO OF THE MODEL AIRCRAFT WING ......................................................................... 35
5.1 Modelling the geometry ............................................................................................... 36
5.1.1 STAR CCM+ Model ............................................................................................. 36
5.1.2 ANSYS Model ..................................................................................................... 37
5.2 File management......................................................................................................... 38
5.3 File nomenclature system ........................................................................................... 38
5.4 Design of Experiments ................................................................................................ 38
5.5 Simulations .................................................................................................................. 39
5.6 Importing for analysis .................................................................................................. 39
5.7 Meshing ....................................................................................................................... 40
5.8 Automation .................................................................................................................. 41
5.9 Finite volume method – Finite element method .......................................................... 41
6 IMPLEMENTATION OF FSI (FVM & FEM) ........................................................................ 43
6.1 Phase I: ....................................................................................................................... 43
6.2 Phase II: Writing the CAE file (.cdb) ............................................................................ 45
6.3 Phase III: First-way coupling and mapping ................................................................. 46
6.4 Phase IV: Simulating static and modal analyses ........................................................ 48
6.5 Phase V: Second-way coupling and mapping............................................................. 50
6.6 Phase VI: Second cycle displacement simulation ....................................................... 52
6.7 Phase VII: Second cycle pressure simulation ............................................................. 53
6.8 Phase VIII: Subsequent mappings .............................................................................. 53
6.9 Phase IX: Post-processing results .............................................................................. 53
7 IMPLEMENTATION OF FSI (FVM only) ............................................................................ 55
8 RESULTS ........................................................................................................................... 56
8.1 Airfoil analysis ............................................................................................................. 56
8.2 FVM & FEM aerodynamic analysis ............................................................................. 57
8.3 Metamodelling ............................................................................................................. 59
8.3.1 Function formation ............................................................................................... 59
8.3.2 Evaluation of the Function of Maximum Nodal Displacement............................. 60
8.3.3 Evaluation of the Function of Maximum Von Mises Stress ................................. 64
8.3.4 Evaluation of the Function of – Lift/Drag ............................................................. 68
8.3.5 Evaluation of the Functions ................................................................................. 71
8.4 Optimised Result ................................................................................................. 72
VALIDATION ............................................................................................................................... 73
DISCUSSION .............................................................................................................................. 75
Parameterization ..................................................................................................................... 75
Modelling ................................................................................................................................. 76

& STRUCTURAL) v
Computational fluid dynamics ................................................................................................. 76
Finite element analysis ........................................................................................................... 76
Coupling .................................................................................................................................. 76
Mapping .................................................................................................................................. 77
File management .................................................................................................................... 77
Simulation results .................................................................................................................... 77
Objectives and Design Variables ............................................................................................ 78
Metamodelling ......................................................................................................................... 79
Multidisciplinary design optimization ....................................................................................... 79
CONCLUSION ............................................................................................................................ 80
RECOMMENDATION ................................................................................................................. 81
FUTURE WORK.......................................................................................................................... 82
REFERENCES ............................................................................................................................ 83
APPENDICES ............................................................................................................................. 88

& STRUCTURAL) vi
LIST OF FIGURES
11
Figure 1-1 Concurrent Subspace Optimization ........................................................................................ 15
11
Figure 1-2 Collaborative Optimization or BLISS ....................................................................................... 15
5
Figure 1-3 Central executive overview ...................................................................................................... 18
6
Figure 1-4 Involved disciplines .................................................................................................................. 18
1
Figure 1-5 Top-level wing MDO capability ................................................................................................ 19
26
Figure 2-1 Least-squared solution to lattice deformation of constrained airfoil ..................................... 21
27
Figure 2-2 Representation of geometric surface ..................................................................................... 21
Figure 2-3 Basic framework of MDO .......................................................................................................... 22
Figure 2-4 2-D representation of Pareto front ........................................................................................... 23
Figure 2-5 Response surface result of the MDO ......................................................................................... 23
30
Figure 3-1 Cutaway drawing of the 767 Aircraft ..................................................................................... 25
30
Figure 3-2 Upper skin sub-assembly ........................................................................................................ 25
30
Figure 3-3 Decomposition of component parts ....................................................................................... 26
30
Figure 3-4 Colour coded discipline components ...................................................................................... 26
Figure 3-5 Far field of NACA 5412 airfoil with chord length, C .................................................................. 27
Figure 3-6 Hybrid mesh of the NACA 5412 fluid region in C-grid ............................................................... 27
Figure 3-7 Airfoil analysis pressure scene of NACA 5412 ........................................................................... 27
Figure 3-8 Fully constrained airfoil............................................................................................................. 28
Figure 3-9 Fully parameterized airfoil ........................................................................................................ 29
Figure 3-10 Design variable parameters shown on wing model ................................................................ 30
Figure 4-1 Data mapping route ................................................................................................................ 32
Figure 5-1 The MDO framework processes ................................................................................................ 35
Figure 5-2 The MDO framework tools stages ............................................................................................ 36
Figure 5-3 Rectangular block; model for STAR CCM+ ................................................................................ 37
Figure 5-4 Wing model in rectangular block; model for STAR CCM+ ......................................................... 37
Figure 5-5 Wing model without block model for ANSYS ............................................................................ 37
47
Figure 5-6 +ve hedral angle (dihedral) .................................................................................................... 39
47
Figure 5-7 -ve hedral angle (anhedral) .................................................................................................... 39
Figure 5-8 Imported geometry in STAR CCM+ ........................................................................................... 40
Figure 5-9 transparent geometry ............................................................................................................... 40
Figure 5-10 Imported geometry appear as lines in ANSYS ......................................................................... 40
Figure 5-11 Line plotted in colour code in ANSYS ....................................................................................... 40
49
Figure 5-12 A tetrahedral mesh element ................................................................................................ 41
50
Figure 5-13 A Polyhedral mesh cell ......................................................................................................... 41
Figure 6-1 Mesh cells showing polyhedral cells ......................................................................................... 44
Figure 6-2 Zoomed-in mesh cells................................................................................................................ 44

& STRUCTURAL) vii
Figure 6-3 Controlled line mesh in ANSYS .................................................................................................. 46
Figure 6-4 Tetrahedral mesh with fine elements at both edges ................................................................ 46
23
Figure 6-5 Cycle of one-way/two-way coupling ...................................................................................... 47
Figure 6-6 Wing skin pressure distribution ................................................................................................ 48
Figure 6-7 Pressure mapped onto CAE model in STAR CCM+ .................................................................... 48
Figure 6-8 Mapped pressure ...................................................................................................................... 49
Figure 6-9 Mapped pressure, wing with fixed end ..................................................................................... 49
Figure 6-10 Contour plot of Von Mises Stress on wing .............................................................................. 49
Figure 6-11 Plot of wing under displacement in ANSYS ............................................................................. 50
Figure 6-12 View of deformed wing through transparent block mesh ...................................................... 50
Figure 6-13 Result of static nodal solution in ANSYS ................................................................................. 51
Figure 6-14 Scene of imported displacement corresponding to UY in ANSYS (Fig 6-13) ............................ 52
Figure 6-15 Small displacement ................................................................................................................. 53
Figure 6-16 Large displacement ................................................................................................................. 53
Figure 7-1 Questionable maximum displacement on wing (FVM only) ..................................................... 55
Figure 8-1 Airfoil in C-grid designed in CATIA ............................................................................................ 56
Figure 8-2 Airfoil in C-grid imported into STAR CCM+ ................................................................................ 56
Figure 8-3 Scene summary of airfoil 5412 results ...................................................................................... 57
Figure 8-4 AR vs UXYZ curve fitting ............................................................................................................ 60
Figure 8-5 Hedral angle vs UXYZ curve fitting ............................................................................................ 61
Figure 8-6 Sweep angle vs UXYZ curve fitting ............................................................................................ 62
Figure 8-7 AR vs SEQV curve fitting............................................................................................................ 64
Figure 8-8 Hedral angle vs SEQV curve fitting ........................................................................................... 65
Figure 8-9 Sweep angle vs SEQV curve fitting ............................................................................................ 66
Figure 8-10 Aspect ratio vs -Lift/Drag coefficient curve fitting .................................................................. 68
Figure 8-11 Hedral angle vs -Lift/Drag coefficient curve fitting ................................................................. 69
Figure 8-12 Aspect ratio vs Lift/Drag coefficient curve fitting ................................................................... 70
Figure 8-13 Optimum design variable ........................................................................................................ 72
Figure 8-14 Optimum design objective ...................................................................................................... 72
47
Figure 0-1 Airplanes showing wing configurations ................................................................................. 89
47
Figure 0-2 enlarged view of the wing ...................................................................................................... 89
Figure 0-3 A number of possible disciplines ............................................................................................... 90
Figure 0-4 Parameterized airfoil (enlarged) ............................................................................................... 90
Figure 0-5 Cell set within STAR CCM+ ........................................................................................................ 91
Figure 0-6 Cell set within STAR CCM+ (zoomed in) .................................................................................... 91
Figure 0-7 Cell set around wing within STAR CCM+ ................................................................................... 91
Figure 0-8 Controlled line mesh within Ansys (Zoomed-in leading edge) .................................................. 92
Figure 0-9 Controlled line mesh within Ansys (Zoomed-in trailing edge)................................................... 92

& STRUCTURAL) viii
Figure 0-10 STEP file model defect (geometry scene) ................................................................................ 93
Figure 0-11 STEP file model defect (transparent mesh scenee) ................................................................. 93
Figure 0-12 STEP file model defect (velocity scalar scene) ......................................................................... 93
Figure 0-13 Wing showing displacement through the transparent mesh ................................................. 94
Figure 0-14 Wing showing displacement through the transparent mesh ................................................. 94
Figure 0-15 Response surface of design variables (Scattered) ................................................................... 95
Figure 0-16 Response surface of design variables ..................................................................................... 95
Figure 0-17 Pareto front of Objectives (Scattered) .................................................................................... 96
Figure 0-18 Response surface of Objectives............................................................................................... 96

& STRUCTURAL) ix
LIST OF TABLES
Table 3-1 Design parameters and their effect on the wing model _______________________________ 28
Table 6-1 Command for preparing the CAE model in ANSYS ___________________________________ 45
Table 8-1 Aerodynamic characteristics of the airfoil _________________________________________ 57
Table 8-2 Results with partial variation of Aspect Ratio _______________________________________ 58
Table 8-3 Results with partial variation of Hedral Angle ______________________________________ 58
Table 8-4 Results with partial variation of Sweep Angle ______________________________________ 58
Table 8-5 UXYZ dependency on Aspect Ratio _______________________________________________ 60
Table 8-6 UXYZ dependency on Hedral Angle _______________________________________________ 61
Table 8-7 UXYZ dependency on Sweep Angle _______________________________________________ 62
Table 8-8 SEQV dependency on Aspect Ratio _______________________________________________ 64
Table 8-9 SEQV dependency on Hedral Angle _______________________________________________ 65
Table 8-10 SEQV dependency on Sweep Angle ______________________________________________ 66
Table 8-11 Lift/Drag dependency on Aspect Ratio ___________________________________________ 68
Table 8-12 Lift/Drag dependency on Hedral Angle ___________________________________________ 69
Table 8-13 Lift/Drag dependency on Sweep Angle ___________________________________________ 70
Table 8-14 20 Aircraft wing configurations on the Pareto optimum ___________________________ 72
Table 17 Comprehensive simulation results ________________________________________________ 97

& STRUCTURAL) x
GLOSSARY
AIAA American Institute of Aeronautics and Astronautics

AIC Aerodynamic Influence Coefficient
ANSYS FEA tool used
BLISS Bi-Level Subspace optimisation
CAD Computer Aided Design
CAe Computational Aeroelasticity
CAE Computer Aided Engineering
CATIA CAD tool used
CFD Computational Fluid Dynamics
CO Collaborative Optimisation
CSSO Concurrent Subspace Optimisation
DFRC Dryden Flight Research Centre
DoE Design of Experiments
DV Design Variable
FAR US Federal Aviation Regulation
FSI Fluid Structural Interaction
GUI Graphic User Interface
MDAO Multidisciplinary Design Analysis and Optimisation
MDO Multidisciplinary Design Optimisation
NASA National Aeronautic and Space Administration
OF Objective Function
RAM Random Accessory Memory
RMSE Root Mean Squared Error
SEQV Von Mises Stress
STAR CCM+ FVA tool used
UXYZ Resultant Displacement magnitude

& STRUCTURAL) xi
1 INTRODUCTION
Ever since the inception of aircrafts, till date, it has been a transition from one generation to
the next with changes in the aircraft industry. These changes became inevitable as human
wants for the best helps with the improvement of flying technology. This is evident of the lag
in technology as human wants are always one step ahead of the ever increasing technology.
Human imperfections also lag by a step from the optimum. This optimum is ideally infinite as
there are infinite numbers of design variables so can only be defined by the immediate future
description of the optimum. This optimum search approach which has in the past been found
using a trial-and-error approach has metamorphosed into an approach presently known as
the Multidisciplinary Design Optimization. This project thus limits the definition of the optimum
to that within two disciplines, aerodynamics and structural interaction as being multiple
disciplines. A solid wing made of an unsymmetrical aerofoil, NACA 5412 was considered all
through the experimentation.
Basically for an aircraft wing, the objective functions are numerous and of the order of
hundreds to thousands according to research. This objective functions vary with aircrafts
since the purpose for which an aircraft is design also varies. Some are for commercial
purposes and meant to lift large payload, some are for cargo, hence prone to very heavy pay-
loading, while some are designed for just fewer passengers on-board, some are fighter jet
that requires high manuoverability. These are just to mention a few of the numerous
categories of aircrafts. A common objective among these aircrafts is the use of the wing to
generate lift (priorities on this vary among these various categories). Another common
objective is the consideration for a strong aircraft. The materials must be of enough strength
to withstand an amount of deformation without failure and must be durable and fit for
structural purposes to withstand the aerodynamic loading within the flight conditions.
The first two major disciplines for consideration in a Multidisciplinary Design Optimization,
1; 2; 3
MDO on an aircraft wing are the Aerodynamics and the Structural . This is not to say that
others are less important as no feasible MDO will base its optimization on just these two
functions since they do not totally encapsulate the involved disciplines on the aircraft wing. A
typical example is the case of an MDO on a fighter plane in which manuoverability, range and
many other objectives are also very essential to the performance and functionality of the
aircraft. This leads to a scientific prioritization of the involved objective functions in a MDO
within a number of disciplines – this would have been easier if these objective functions are
known and (or) linear. None of the objective functions done on an aircraft wing has been
4; 5
reported to be linear up till date . So also is the fact that no known relationship exists to
express the behaviours of the objective functions with respect to changes in the design
variables. This calls for a paradigm in modelling and simulation to investigate the dependency
of the design variables on the objective functions with consideration of the required
constraints - the approach used is the Design of Experiments DoE to establish the
relationship between the variables and the objectives.
(AERODYNAMICS & STRUCTURAL) 12
A parametric study was done on the aerofoil. The result of which shows that the aerofoil was
good to design a finite wing at some particular flight condition which was used all through the
simulation. Then, the DoE was done following the framework of the MDO as a tool to
achieving a multi-objective optimisation. This involves Parameterization, DoE, Modelling and
Simulation, and Automation. An FSI Modelling and Simulation was done in STAR CCM+, the
pressure load was then imported into ANSYS to solve for the structural objectives. Then, a
genetic function approximation approach was used to determine the non-linear function
approximation which is a quantitative structure-property relationship, of simply put, a �(�) for
the considered objective functions. These objective functions are then entered into the
optimization toolbox within MATLAB to determine the design variables to give the optimized
wing model.
It might be interesting to know that this is a long term project as reported in journals but the
best has been made of the MDO process in simplifying the problem and still maintaining the
logical concept and objective of the purpose of the MDO. These simplification process is
however more difficult than it sounds - “simple”. Overall, one understands the following:
• What needs to be done
• How it should be done and
• Why it should be done the way it’s done.
1.1 Multidisciplinary Design Optimization

Taking a curious look at the shape of the aircraft, the wing in particular, known as the part of
the plane that generates lift, the inquisitive nature of human turns in raising the questions of
why the wings of all aircrafts do not have the same topology. The shapes and additional
features physically make the difference such that it seems it is meant to serve a range of
aerodynamic purpose of beauty. The structural members therein also changes from one wing
to the other. Actually, these variations are designed to suit the purpose for which an aircraft is
designed. This wing serves several purposes which include but not limited to aerodynamics,
structural, flight stability and control, etc. all these disciplines are considered during the design
of the aircraft together with the consideration for cost of operation, payload capacity, and the
list goes on as much as possible. These variations are known to affect these multiple
disciplines been considered which creates an awareness within every single disciplines to
approach a set of design variables that best suits the objective of the discipline. To achieve
this every discipline seeks an approach to an optimum set of design variables. However, the
optimum set of design variables favourable to an objective is unique to every discipline
therefore; there exists no common set of design variables for an overall optimum. This raises
consideration for a systemic search for optimum known as the multidisciplinary design
optimization which is a methodology for the design of complex engineering systems and
6
subsystems that intelligibly feats the synergy of mutually interacting disciplines .

1.1.2 Definition
NASA defined MDO as “a methodology for design and analysis of complex engineering
systems and subsystems which coherently exploits the synergy of mutually interacting
1; 6 7
phenomena” . While Venter describes the methodology as a design of systems in which
interaction between multiple disciplines is considered having design variables affecting more
7
than one of the inter-woven disciplines . The AIAA version can be explained more easily to
mean how to decide what is meant to be changed and to what extent it is meant to be
1
changed when everything influences every other thing . In all definitions, the one thing that
holds in the MDO methodology are cultural reminders that must be strictly adhered to if the
MDO it to be termed potent for possible synergetic result – the contributing disciplines must
be genuinely interdependent. In this case of an aircraft wing, the aerodynamics depends on
the stiffness and hence the deflected displacement of the wing in the structural discipline
while the structural discipline depends on the pressure distribution on the wing in the
aerodynamic discipline. This might seem to be a bit of trouble but ideally sets in place a
perfect scenario to have a meaningful MDO. If on the contrary this interdependency does not
exist, there will be no multidisciplinary optimization. It would have then been a working group
of single disciplinary optimization on a design in which the team leader only merely collates
the disciplinary results from all single disciplines.
1.2 MDO Methods

An efficient MDO framework will be one in which a systematic approach is applied to explore
the synergy opportunities in achieving the optimized model. Doing a topological optimization
requires that the model should be easily updateable and allows an easy update of the FEA
and CFD models for every configuration of the design. Presently, there are two basic
approaches to achieving the optimum of a design using MDO – these are the concurrent
3; 4; 8; 9
subspace optimization (CSSO) and the collaborative optimization (CO) or (BLISS). The
CSSO finds its application at the high level of application of MDO due to the availability of
8; 10
high fidelity computers (further knowledge about this topic can be found in references).
This makes it possible to run all analyses simultaneously on a computer and feed the result
into the design space as a data even in a less amount of time. This approach has an
advantage of exploring space region closer to the desired optimum hence having a better fit
for the Pareto optimum front. It uses a gradient-based algorithm to pursuer the direction of
8
feasible direction . In cases where the MDO is limited in terms of computing power, the
approach becomes highly unrealistic. In general cases where this or similar limitation sets in,
the use of the use of a high approximation algorithms, CO or BLISS then becomes invaluable.
In this case, several disciplinary analyses are run on a number of dedicated computers which
communicates (coupling) the inter-related and dependent field functions (say pressure and
displacement) among the involved disciplines. This is a way to protect the synergy of the
multidisciplinary design optimization.

11
Figure 1-1 Concurrent Subspace Figure 1-2 Collaborative Optimization or BLISS
11
Optimization
As a means to exploiting the limited computational facilities within reach, the CO was adopted
as a means to reaching the goals of achieving an optimum wing design. Several
approximation techniques were explored and exploited. This is in accordance with the
1; 6; 10; 12; 13
conclusion of several research in MDO as a means to reducing the large
computational time involved in acquisition of data in a simulation-based multidisciplinary
10; 14; 15
design optimization .
1.3 CO or BLISS
Aerodynamic and structural disciplines form the baseline of any MDO within the aerospace
industries; this continues to hold even far outside the aerospace industries. Prominently, the
potency of an MDO on the wing on an aircraft is highly vital because all other functionality of
the wing depends on at least one of the two disciplines. CFD result provides the pressure
distribution which is needed to solve the static analysis in the FEA. This pressure distribution
on the wing has been agreed to be very promising when computed by a CFD analysis unlike
other methods which are mainly analytical and provides lesser sensitivity to modelling the
physics which includes the effect of turbulent. Running a CFD simulation on the wing in this
case offers the highest degree of accuracy (this is the reason for considering CFD rather than
the other methods) but the time spent on a single CFD analysis take the order of hours on a
Core i7 computer when running on the serial mode. This is ridiculously unacceptable with the
CSSO because every next model for MDO analyses has to wait till the present finishes which
will take unnecessary duration to complete an MDO that would require analyses on hundreds
of models. Additionally, this is why CO was preferred for the MDO.
1.4 Approximation Methods

Rather than having to run simulations on every of the prescribed configuration by the multi-
objective tool, the method of systematically creating set of data from a vector of design
variable within a design space has been tailored to be of high importance in simulation-based
multidisciplinary design optimization. The compact surrogate model of approximation method
was used to predict the performance of the configuration models with respect to design

16
variables . This design space which can be referred to as a black box represents the
boundaries of the degrees of freedom of the design model. The number of occurrence within
this space for exploration is indeed infinite but can be made finite by employing a design of
experiment DoE. This produces a finite data set that systematically samples the whole space
and uses a passive constraint defined by the ranges of the limits of the design variables which
forms the black box.
These sets of data selected for the DoE are then been analysed (possibly on several
computers at a time) in a coupled manner so as to preserve the multidisciplinary nature of the
optimization. The product is the behavioural information of the DoE which is then being
approximated by some methods to define an approximated function of the product behaviour
10
known as Metamodels . These metamodels are created by the combination of information
about the product behaviour, which is an expression of the dataset and the interpolation
information, which is expressed using algorithms. There are several algorithms which work
approximations differently with different local adaptive accuracy (that is, they are specific to
approximation) which depends on the accuracy of the dataset and the space between them.
These metamodels or surrogate models (are also called response surface models or
10; 16
replacement models) include :
• Polynomial models
• Kriging models
• Radial basis functions
• Multivariate adaptive regression splines
• Support vector regression
• Artificial neural network (ANN)
The evaluation of accuracy is however not a practical concept because an attempt to test the
quality of the metamodels results in addition to the population of the dataset which invariably
increases the quality of the approximation.
1.5 Rudiments of MDO

A lot is yet to be unravelled about the MDO as a means to achieving the objectives of this
project which is finding the optimum design for the model aircraft wing within the two
disciplines of interest. It is necessary at this juncture to enumerate the basics of the MDO.
1.5.1 Objective functions

The objectives are variable that are meant for optimization. This are restricted to constraints
and depends on the design variables. A relationship between every single one of these
objectives and the design variable defines the objective functions. Those of concern in this
17
report are :
• − �� ⁄��
18; 19
Negative lift to drag coefficient ratio
•
19; 20
Maximum Von Mises Stress factor

•
17; 19; 20
Maximum wing tip nodal displacement
1.5.2 Design variables

The design variables consist of parameters that will be varied while causing a variation in the
design configuration. This can be represented as a vector whose variation can be controlled
by the result of the parameterized model. The list of possible design variables is countless but
will be limited to the topological variable in this case. It is however a requirement that the
design variables should be uncoupled when an approximated metamodel is meant to be used
in the multi-objective optimization as a genetic algorithm. They are:
•
17; 19; 20
The wing sweep
•
19
The wing hedral angle
•
19; 20
And the wing aspect ratio
1.5.3 Constraints
This represents a failure mode in which the variables are being constrained. These variables
could be the objective or the design variables. However in this case, since the aircraft wing is
a model solid structure with material properties not driven by the possible available materials
but by the feasibility of the analytical tools. The properties are not real even though they are
close to reality so constrain was not imposed on the objectives but on the design variables
only. Hence the constraints are defined by the limits of the design variables as passive
constraints.
1.5.4 Baseline design of the wing

The baseline design is defined as the unsweep wing without a tapper or a hedral angle with
root and tip chord length of1.2 ��. Material properties were the same all through the
optimization process.
1.5.5 Multiple disciplines

It is cultural that the multiple disciplines involved must be inter-related as the definition
explains that changing one thing causes a change in other things. Here, the aerodynamic
pressure load distribution is dependent on the structural analysis and the structural thus
determines the pressure loading invariably. However, it should be noted that even though the
two disciplines were considered, there are many other disciplines and sub-modules that must
be considered for an actual aircraft.

6
Figure 1-3 Central executive overview
5 Figure 1-4 Involved disciplines
1.6 MOTIVATION
21
Reviewing the report on computational fluid dynamics of the whole-body aircraft revealed
the present state-of-the-art development as at 1999 and also highlights the empowerment
achievable by computational analyses. This wetted the inquisition to understand what drives
the shapes of aerodynamic structures in an ever changing design world. Knowledge of the
advanced optimization method appeared to be vast across all ramifications of optimization. It
was then only logical to register into this method which has been recognized to be the key
thing in technological advancement of the future. The possibilities are endless and have also
been reported to be best practiced on an aircraft wing.

1
Figure 1-5 Top-level wing MDO capability
1.7 OBJECTIVES
This project aims at performing MDO using ANSYS and STAR CCM+ as tools for structural
analysis and CFD analysis respectively on a baseline wing model. Modifications of the design
will be done in CATIA and optimization in MATLAB. The primary objective is to achieve an
optimum for a wing design with respect to the aerodynamic and structural characteristics of
the wing model taking the baseline design described above as the datum. This will answer to
the question “what is the best topological design for an aircraft when the aerodynamics and
structural characteristics are considered”.
5
Furthermore, it is meant to follow the MDO approach of optimum search described by NASA
which invariably will facilitate the design of a MDO tool in future research.

2 LITERATURE REVIEW
The theory of multiobjective optimization dates far back as 1776 during the publication of
Adam Smith’s treatise The Wealth of Nations. Vilfredo Pareto and others also worked on this
22
concept between 1874 and 1906 . However, multidisciplinary design optimization has been
in development for about two decades which makes it a relatively new concept to achieving
an optimum design. It involves the search for an optimum; using a methodology for design
and analysis of complex engineering systems and subsystems which coherently exploits the
6
synergy of mutually interacting phenomena . Also included in this project report is the concept
of Fluid-Structure Interaction which has been in existence for a relatively longer period and
still undergoing development. Several achievements have been recorded on this but it has not
23
been as successful as multidisciplinary design optimization . However, the results are quite
reliable for present day application to technology.
This chapter highlights the review of previous and present work of other publications in close
relation to this topic. It therefore splits the review into FSI and MDO categories. It might be a
bit difficult separating this chapter from the methodology chapters in order to avoid direct
repetition. However, the practicality of the review is adequately outlined in the methodology
chapters.
2.1 FSI and MDO

1
Kesseler and Vankan appraise the capability of optimizing an aircraft wing using MDO and
described the interactions between aerodynamics and structural analyses in the aerospace
industry. Even though several works has been done on an aircraft wing in the past, several
approaches to MDO has been explored by researcher which attempts solving MDO of a wing
using different analytical means to determine the various disciplinary objectives. The primary
objectives for consideration in an MDO of the wing will be those within the aerodynamics and
the structural disciplines which interacts with each other, satisfying the cultural requirement of
17
the MDO . While many works focused on the wing as the chief structure for the MDO, much
more of the work in this field concentrates on the framework and methodology, highlighting
2
and attempting simplification of the process at large .
20
Davide worked on optimizing the wing of a supersonic aircraft using curvilinear sparib
simulating the objectives on different tools which was in accordance with the NASA
requirement that the MDO tool should be capable to leverage existing analytical tools. In the
work, ZAERO was used alongside with other tools to compute the FSI of the wing during the
flutter and divergence speed analysis as part of the considered objectives. This trend of
24
leveraging existing tools was also used by Sirdhar and others to analyse the FSI. Although
this project is limited to two disciplines, some of the other disciplines were mentioned by
9
Sobieski . This justifies the consideration of the two disciplines in this project.

2.2 Adoption of CO
The FSI aspect of this project was a phase of acquisition of data from the results of various
11
analyses involved in this project. Kroo mention two basic approaches to MDO namely, the
4
CSSO and the CO which has been earlier discussed in the introduction. Sobieski , 2000
looked into BLISS which is a method not so embraced nowadays because it divides
20
optimization into sub-optimization like the two-way optimization investigated by David which
may invariably loosen the interdependency that exist between the involved disciplines.
Presently, the CO has found more application due to the advantages over the other methods
1
and was recommended in 2006 by Vankan . It uses a non-gradient based search for
8;
optimization which makes it less venerable to local optimum convergence unlike the CSSO
25
. On the other hand, CSSO uses a smaller discretization as the search draws closer to
convergence. This increases the accuracy of the optimum.
2.3 Parameterization
Parameterization was described as an art of scientific modelling for the purpose of ease of
26
design modification. The two forms of geometry deformation were discussed by Anderson
namely: constrained-based deformation and the lattice deformation. These forms of
27
deformation applied to 2-D and 3-D CAD models respectively. The work of Edwin also
bolstered the analytical methods and processes involved in the deformations. This works
were more specific to the B-spline parameterization as it dominates the real world of
designing. Several types of curves and their advantages during parameterization were
analysed. The clues deduced from these works were valuable for the parameterization of the
wing as a constrained-based deformation type of parameterization. Several other authors
mentioned parameterization and very few mentioned the tool used for the parameterization
but none did the parameterization in CATIA. The closes work on the application of
20
parameterization was that of Davide in which three forms of the wing parameterization was
implemented in MSC.PATRAN. The implementation of parameterization in the tool was not
really applicable to that in CATIA. However, the knowledge therein was helpful together with
28
the CATIA user guide . A present study at the University of Cranfield emphasizes on
achieving easier and quicker parameterization.
Figure 2-1 Least-squared solution to

lattice deformation of constrained 27
26 Figure 2-2 Representation of geometric surface
airfoil

2.4 MDO Framework

NASA and AIAA have recorded quite an amount of success on the MDO framework. Works of
9
Jaroslaw at AIAA; Chan’s work at NASA; and that of Kesseler at the National Aerospace
Laboratory, Netherland has been more embraced by many other researchers. The logics in
the framework were adopted in this project. The framework is controlled by a central
5
executive that handles analyses and other processes within the MDO . This central executive
controls the modelling and modification of the parameterized model. It then analyse the model
with respect to the number of disciplines involved therein. The possible modules of algorithm
th 5
were mention during the 26 international congress of the aeronautical sciencs . The MDO
tool must also be capable of leveraging the existing tools for the various disciplinary
5
analyses . Within this disciplines are a number of possible objectives or sub-modules. Some
of which has to be sequential in order of analyses like the modal analysis that would be
completed before analysing phenomenon like flutter and divergence. Vankan and Kessler did
apply a designed MDO tool in the Multidisciplinary design optimization of a three-bar truss
5
and the Ikhana with fire pod . A number of challenges and approaches were discussed in the
work. Results were then given on a Pareto front for individual objectives on a response
surface.
Figure 2-3 Basic framework of MDO

9
Sobieszczanski went further stating the components of the MDO as: mathematical modelling
29
of the system which Vankan and Laban described to be metamodelling of fitness function;
design oriented analysis; approximation techniques like the Kriging model and its derivatives;
system sensitivity analysis to response surface; MDO procedures; and the human interface of

communicating and interpretation of optimum result. The simultaneous optimization of
aerodynamics and structural, and that of structural and control disciplines combinations were
elucidated by Sobieszczanski in which Kroo’s work was appraised too.
Figure 2-4 2-D representation of Pareto Figure 2-5 Response surface result of the MDO
front

3 MODELLING AND PARAMETERIZATION
Introduction:
In order to perform analysis of whatsoever on the wing, a representation of the actual wing
has to be created in a format that is readable by the computer. This format depicts the
geometry of the actual aircraft wing to the computational tools meant to be used for simulation
and makes it perceive the computational format of the aircraft wing in a manner human and
physics will perceive the actual wing. The process of creating this representation is known as
modelling. A set of design variables is needed to run simulation on a model which is just one
of the numerous models that will be processed for simulations during the MDO process. That
brings about the need for parameterization in order to allow easy of modification from a model
to another with similar but different design variables.
3.1 Modelling:
Modelling which has been introduced as a process of replicating the actual aircraft wing in a
format that the computer recognises for possible simulation is an art of scientific applications
because there is no hard and fast rule to modelling all problems. Hence, human discretion is
highly invaluable in order to create an actual replica of the problem being modelled. This is
the case with the model aircraft wing. The model has to be created in a way such that it will
be readable by the programmes for which it’s being modelled. It also calls for an improvisation
of the available tools as this is essential in all engineering organisation. In this case, ANSYS
is used as the tool for the Finite Element Analysis simulations of the wing structures, STAR
CCM+ for the Finite Volume Method of analysis on the fluid, while the design modelling is
done in CATIA. For simplicity and easy of automation in STAR CCM+, the wing model was
designed separately for analysis. Every unique model has two geometries prepared for the
structural and aerodynamic analysis. This art of modelling in simulation based paradigms
goes beyond the boundaries of the geometry. It also encompasses the physics modelling
which will also be discussed furthered in this chapter.
The present state of the art technology in computational fluid dynamics has recorded a
number of achievements, landmarks and developments. Even though some of the challenges
of modelling geometry and physics remains a high concern among researchers, modelling
21
geometry has been relatively quite successful compared to that of turbulent flow physics .
Aforementioned that MDO process requires an effective means of numerous geometrical
modifications. These modifications are being evaluated by means of simulation which is
relatively less expensive, flexible to modifications, and less time consuming when compared
to experimentation. In addition to these advantages, it can be equally as valid as the real
physics interaction on the actual model with adequate modelling effort. This approach to
solving engineering problems is however similar for all branches of engineering analysis in
which discretization and iteration process is the means to reaching a converged solution.

Ideally, the aircraft wing is made of composite materials and aluminium alloy with a large
volume of void inside which is often meant to serve as a reservoir for fuel. The wing structure
is made of several parts which can be majorly grouped to be the lower and upper wing skin,
the spars, stringer and the ribs. It is however difficult to generalise the definition of an aircraft
wing as it is meant to serve different purposes in several types of aircrafts.
Figure 3-1 Cutaway drawing of the 767

30 30
Aircraft Figure 3-2 Upper skin sub-assembly
The cross-section of the wing is defined by an aerofoil at any plane from the wing root to the
wing tip. It forms a cantilever with the tip as the free end and the root as the connection that
joins the whole wing to the fuselage. Several modifications can be implemented on the aircraft
wing as requirement, performance and functionality varies with planes. These include
31
winglets at the tip, tapper, sweep, dihedral angle or a slight twist . Other parts used for
mechanical flight control also makes the list of possible unique wing model numerously
31
countless. A case study on aircraft wing manufacture shows that the amount of parts meant
for structural consideration during the design of an aircraft is actually vast and not as simple
as the wing may seem to be to an aerospace amateur. In order to have a logically reasonable
case for consideration for MDO, a simplification approach has to be devised within the limited
time frame. This simplification however requires an adequate understanding and appreciation
of the whole functionality of the aircraft wing. In order words this simplification is non-trivial as
the simplified problem may be. It requires an understanding of the aerodynamic and structural
understanding of the aircraft wing as well as the aeroelastic effect because this is a major
phenomenon that can’t be neglected in a vast process as MDO.
Looking at the wing, it is made of several structures among which some functions as
structural parts only, some function as both aerodynamic and structural parts. Since fluid to
be considered for aerodynamics has to be bounded by boundaries (either real or imaginary),
no part functions for aerodynamics only. Parts that share both structural and aerodynamics
purposes has to be considered for aerodynamic loading and aeroelastic effects. This forms a
suitable instance for the simulation of fluid-solid interaction, FSI. This has been neglected in
several cases of MDO on an aircraft wing due to the difficulty and high computational time
32
associated with automating MDO with incorporated CFD . However, it has been
33
recommended to be the ideal approach in modelling MDO of an Aircraft wing . Furthermore,
6
recommendation has it that more accurate results and opportunities for synergy will be
achieved if the loading on the wing was determined by CFD rather than using the vortex-
lattices method which is being used by several researchers that would not tolerate
interdependency which is the culture of the MDO in determining the pressure loading
distribution on the wing during a turbulent flight condition.
Figure 3-3 Decomposition of component Figure 3-4 Colour coded discipline

30 30
parts components
The logical simplification approach followed in the quest for an optimised wing was carried out
on a model solid wing since it combined the aerodynamic and structural characteristics of the
aircraft wing. This further requires a level of discretion since the void and a number of
component parts have been replaced by a homogeneous solid wing. The material used for
the wing model is aluminium alloy with properties as stated below. The Young’s modulus has
been adjusted to suit the solid displacement and morphing tolerance allowed in STAR CCM+.
• Young’s modulus 120000 ��
• Poisson’s ratio 0.3
• Density 8.5 × 10−06 �� −3
• Wing reference area 3.6 �2
The NACA 5412 airfoil was considered to be used after a parametric study on airfoils was
conducted for four other transonic airfoils. The study revealed that the NACA 5412 was a
transonic airfoil with the highest value of lift to drag ratio at a true air speed of Mach 0.7,
approximately 240�/� and angle of attack of 5 degrees. This is an airfoil with maximum
camber of 5% positioned at the 40% chord length from the leading edge with a thickness
percentage of 12. The coordinates were generated by a web application known as Airfoil
34
Tools . It produces a .DAT file which has a chord length of unity. This file was prepared in
Microsoft Excel and a macro was played in Excel which automatically plotted the coordinates
and created splines which form the upper and lower perimeter of the airfoil in CATIA. These
splines were fixed which made them non feasible for MDO since the MDO process requires
full parameterization. Although the splines could not be used for parameterization, they
served as guidelines on which unconstrained splines were constructed carefully. The result
was a good and feasible transonic airfoil. The airfoil was given an angle of rotation of 5
degrees with the leading edge fixed and the trailing edge rotated. A c-grid was then formed
around the airfoil such that it was at the centre and the far field was averagely 15 times the
35; 36
chord length . The turbulence model was set-up with the one-equation Spalart–Allmaras
model at a Mach number of 0.7.
A bit of theoretical assumption was also made such that the wing is fixed at the root to the
fuselage and the wing deformation due to loading is linear. Ideally, this is not true as the wing
structure is nonlinear undergoing a nonlinear loading both in the form of reaction at the
connection to the fuselage and the action pressure loading computed by CFD. It would have
been of great interest to consider a real wing but the requirements and level of design
parameterization for an actual aircraft wing is so vast and can be considered as a discipline
on its own. However, the model wing serves as an ideal test case for MDO.
Figure 3-5 Far field of NACA 5412 airfoil with Figure 3-6 Hybrid mesh of the NACA 5412 fluid
chord length, C region in C-grid
Figure 3-7 Airfoil analysis pressure scene of NACA 5412

3.2 PARAMETERIZATION
The framework of the MDO which will be illustrated and discussed later in this chapter
requires that the wing model should be designed flexibly enough to allow faster modification
of the model configuration. This can only be efficiently achieved by allowing a set of control
parameters to drive the modification from a baseline model to a desired model with the
desired design variables for the next model configuration. As mentioned earlier on,
parameterization is non-trivial since the efficiency of the MDO depends on the accuracy and
easy of automation of the rudiments therein. By the definition and objectives of
parameterization, the control dimensions are the geometric parameters that can be varied
and allowing design modifications while preserving the primary shape of the intended design.
This should be flexible enough to allow automation such that changing a control dimension
results in the model being recreated automatically while still preserving the relationships and
26
geometric constraints . The aim of the modification during parameterization will therefore be
27
to create a change in design variables while still preserving the relashinships .
Table 3-1 Design parameters and their effect on the wing model
Control dimensions Effect of modification
Root chord Changes the chord length of the root airfoil without moving the
leading edge
Tip chord Changes the chord length of the tip airfoil without moving the leading
edge
Tip offset plane Changes the semi-span of the model wing
X constraint Controls the sweep angle of the model wing
Y constraint Controls the dihedral angle of the model wing
Figure 3-8 Fully constrained airfoil

27
According to Edwin , the approach to design parameterization used on the wing model is
26
termed CAD-based shape design parameterization . Modelling with feature-based
parameterization was not possible with the licence of CATIA V5 used. This means that the
wing could not be modified after the 3-D solid wing has been created. This was a bit of a work
27
around since the only fully parameterization achieved was on the 2-D airfoil . Two airfoils
26; 37
were created and were fully parameterized by using the f(x) command within CATIA. The
driving dimension was the chord length which was unity and at an angle of attack of 5
degrees. In order to achieve this every points that made up the splines were constrained in
both x and y directions bring the airfoil to a full constrain. The x and y distance was measured
from the leading edge of the airfoil. This leading edge which has been fully constrained to the
neutral x and y axis then serves as a reference point. The x and y constrained values of every
other point were then individually linked to the chord length which was unity using the f(x)
command mention above. This resulted in a full constrained and parameterized airfoil.
Figure 3-9 Fully parameterized airfoil
This airfoil which now forms the root of the wing model was then copied and pasted onto
another plane which was created and defined by an offset from the plane on which the airfoil
was originally parameterized. This newly pasted airfoil formed the tip of the wing. The wing
tip airfoil had the same properties as the root airfoil and was driven directly by the
modifications on the control dimension on the root airfoil. This means changing the chord
length of the root airfoil will result in changes in the wing rip airfoil which contradicts the
objective of parameterization. To preserve this objective, the tip airfoil was then isolated from
the root airfoil which resulted in two independent airfoils. This resulted in the root chord, the
tip chord, the offset, the x-axis constraint on the leading edge and the y-axis constrain on the
leading edge as the control parameter that drives the parameterization of the wing model.
When the two airfoils are in position, the multi-section solid command was then used to create
a solid wing whose boundary is defined by the position and hence constraints on the root and
tip airfoils. The table below highlights the control dimensions and their modification effects on
the model.

Figure 3-10 Design variable parameters shown on wing model
Many were the challenges encountered during modelling and parameterization. Modelling is
the creation of a geometric description of the configuration or model in discretised form.
Secondly, a grid is then generated for the volume, space or material on which solution is
required. This grid is referred to as mesh and forms the mesh continuum. It defines the
coherent whole quantitative and qualitative discretization of the space or material involved.
Even though it is less time consuming, the real time it takes to complete a simulation is way
too large for a high fidelity simulation and can be worse when simulation is limited to run on a
serial core machine. This is due to the amount of time needed for geometrical modelling,
surface preparation and meshing. This can be reasonable improved as the parallel
computation is enabled. It is however more expensive to purchase a licence for parallel
computation. This was the case with this dissertation and created a major drawback which
was tackled by a change of method and switching from the sequential analytical approach of
the MDO to the concurrent analytical approach in which simulations were done on several
computers at a time.
One other hindrance with the modelling of the wing geometry is the incompatibility and lack of
good interfaces to computer-aided design. CATIA was able to do the design and save it in a
variety of file formats. The licence on STAR CCM+ was able to read STEP and IGES file
formats only while that of ANSYS could not import files in these formats. The aerofoil section
at the root of the wing was also not in-plane after been exported from CATIA into STAR
CCM+.
The wing geometry was inconsistent on the root plane of some of the wing configurations as
the design variables changes hence failing to meet the requirement of continuity and
smoothness needed for computational fluid simulation. This was not acceptable because the
process was meant to be automatic as much as possible.

4 FLUID-STRUCTURAL INTERACTION (FSI)
Introduction
The surging improvement in analyses involving Computational Aeroelasticity (CAe) has made
it possible to simulate phenomena involving different CAE tools such as simulating a high-
fidelity aeroelastic loading of solid geometry. In reality, where fluids and solids are involved in
engineering, they do so interactively by co-existing. But the levels to which they co-exist differ
for different cases. This call for an intuitive decision making based on previous knowledge or
experience to tell if it is best to analyse both as being coupled together or to un-couple them
and analyse them distinctively. Even while coupling both together, the level of coupling varies
as some cases require an implicit coupling while some are just perfect case to be simulated
explicitly.
The solution of both fluid and structural analysis are usually being solved by iterative process
in a range of simulation software codes. Usually but not in all cases, these codes are specific
to solving either the structural part or the fluid part of the engineering case. This is true in
most cases because the discretisation and cell representation are different for either the solid
or the fluid analyses using the finite analysis method. Finite element analysis is used to solve
for structures in which the discretised cells are represented and defined as elements with user
defined material properties while the Finite Volume Method is used in solving for fluids in a
region discretised by volume cells. In some cases, either of the discretized analysis methods
are used to solve either of solid or structural problems but nothing compares to using the
appropriate finite analysis method as the case may be. The accuracy also varies from one
code to the other. So does the availability too. So the choice of which code to use is always
worth a number of considerations which narrows down to having a skilled user to model and
run the simulations.
4.1 Coupling
Within the engineering community is several CAD/CAE software which is in most cases
specific to disciplines. These sub-disciplines that make up the community do not exist
singularly. They co-exist; which makes communication and standardization really important
38
among the member disciplines . These disciplines include structures, fluids, thermal,
39
electrical, magnetism and so on . They also interact with one another. The fidelity of these
tools also varies as much as there are many in the industry. This therefore sometimes
requires solving a problem with two or more of this software. This can only be achieved when
the software of interest are compatible. This then requires the user to manually or
automatically set up the connection logically between according to an established framework
40; 41; 42
to imitate the practical physics or phenomenon .

.cdb CAE model used for data

transfer
.mac Mapped data imported into
ANSYS
.Lis Mapped data imported into
STAR CCM+
STAR CCM+ Fluid analysis tool
ANSYS Structural analysis tool
Figure 4-1 Data mapping route
4.2 Mapping
Coupling in most cases involve the complete process of setting up the data connection and
also transferring the data within the involved software. Whereas, In some cases especially
when the mode of transfer is the file-based coupling during which the user has the flexibility to
choose an appropriate interpolation technique to be used for the mapping operation, it can be
considered separately so as to have adequate control over the mapping method in use.
4.3 Coupling Time-Step Classifications

Aforementioned, there are basically three categories of coupling solver setting between fluid
and structural analysis. This classification is typically based on the time step in-between
which data is been exchanged. These classifications are:
• Uncoupled
• Implicit coupling
• Explicit coupling
These depends on if they are being coupled at all and to what extent they are coupled with
respect to iteration time stepping. This interaction could be pressure from the fluid to the solid
or even temperature and on the other side usually would be displacement from the solid to
the fluid. The result of the analysis from one is always due to the action from the other and
hence forms a coupled simulation case.
4.3.1 Uncoupled
For an uncoupled case, the fluid and the solid has little or no interaction across their
boundaries. This usually is with either one of the solid or the fluid as the continuum of interest.
A typical example would be a stationary aircraft in a very slow moving airstream. The aircraft
interfere with the moving air but the moving air is slow enough not to cause any displacement
on the aircraft. In this case, when the aircraft is of interest, the FSI effect can be considered
negligible and hence treated as an uncoupled case.

4.3.2 Implicit coupling

In the case of an implicitly coupled simulation of the aircraft wing in a free stream of air at a
Mach number of 0.7 where the inner iterations of both fluid and solid solvers are strongly
coupled. That is, they exchange relevant data at the boundary at every iteration time step or
thereabout. This exchange of data is done periodically on the computer RAM since it is a
randomly continuous process. In order to ensure that the structure is solved by a FEM
software code and the fluid by a FVM software code, it may be necessary to run a co-
simulation where the FEM and the FVM solves of different software codes are made to
exchange data implicitly using the computer RAM. This is however not possible with some
certain FEM-FVM code combination (say STAR CCM+ and ANSYS) or highly cumbersome or
even less accurate. Compatibility problems usually sets in which in most cases may be due to
the difference in programming language, meshing capabilities, processes or representations.
In possible cases, this method can be used to compute flutter if allowed by the combination of
software codes.
4.3.3 Explicit coupling

This barrier is in some cases broken by employing the File-based coupling which is chiefly an
explicit coupling process. It involves a large iteration time step whereby the converged
solution of one of the software codes is saved in a format readable by the other code. This
other code then imports the saved data in other to run its own simulation and then save the
result in the format that is readable by the first software code. This goes on and on until a
43
staggered convergence is reached between the two software codes . In this case, the data
storage is done on the computer hard disk. This process requires the user to know which
format is readable and which one is writeable by both programmes. The writing and reading
process may be tedious to the user, prone to errors and also reduces the possibility and ease
of automation. The large time step involved cannot capture the enormously small time steps
in flutter but may be relatively cheaper and affordable to solve general FSI problems.
Depending on the level of coupling required, a one-way or a two-way coupling can be
implemented. In a one-way coupling, data and field functions are being imported only in one
direction which means just one convergence is required. It may then be required in some
instances to have a high-fidelity CAe which can be achieved by the staggered convergence
method described above which is a two-way coupling method.
4.4 Coupling Data Exchange Classifications

Another method of classification of coupling in based on the route through which data is being
exchanged. Simulations on these analyses involved in the use of a dedicated computer in
most cases which processes the iterations by assigning the task to the core(s) therein. This
uses the RAM or the hard disk for storage of data during the iteration runs. Using the RAM
allows a faster and more periodic exchange of data than using the computer hard disk. So

data exchange that goes through the RAM is known as co-simulation while that involving the
38; 44
hard disk is termed file-based coupling . Both the co-simulation and the file-based
coupling could be set to exchange implicitly or explicitly but co-simulation offers more
flexibility because file-based coupling will so much of a pain or practically impossible to be
adopted as a method for tightly coupled simulation like flutter and divergence. However, it
offers a method to achieving staggered convergence among two or more simulations that are
44
loosely coupled .

5 MDO OF THE MODEL AIRCRAFT WING
The workflow simply defines the methodology of a MDO with particular interest to the path of
operation as a complex system. This is in accordance with the cultural attributes of a good
MDO. In particular, it relates the baseline design to the optimum design through an
interdependency loop of FSI analyses.
Figure 5-1 The MDO framework processes
A simplified version of the framework above can be seen in the figure below as it depicts the
MDO framework in terms of the familiar tools used for the analyses. It also considers the
initial stage during which metamodelling was done by using these tools to analysis and define
the fitness function of the objectives in terms of the design variables. These two figures depict
the major difference between the CSSO and the CO.

Button
CATIA
OPTIMUM DESIGN
STAR
Button Button
ANSYS [END]
CCM+
SEEDS FROM
SIMULATIONS
METAMODELS CENTRAL EXECUTIVE

MATLAB
PHASE II
CENTRAL EXECUTIVE
MATLAB STAR
Button Button
ANSYS
PHASE I CCM+
OPTIMUM
Button
CATIA
DATUM
GENETIC ALGORITHM GRADIENT-

OPTIMIZER BASED
OPTIMIZER
Figure 5-2 The MDO framework tools stages
5.1 Modelling the geometry

Modelling the aircraft wing seems to be the easiest of all tasks during the project. However, it
becomes more tasking when the only available solution when problems arises in other stage
of work is to come back to the CATIA design to for possible modification of the wing model or
even a redesign of the model. A number of lessons were learnt about batch modelling for
possible automation in other CAE codes like STAR CCM+ and ANSYS. A general
observation is that when geometry models look the same, have the same material model, and
even measure the same dimensions, the path of modelling or how they have been modelled
does determines to a large extent how easily they can be managed in other CAE codes. It
can’t be over emphasised how important it is to get the model right is the simulation or
analysis is to have any hope of success.
5.1.1 STAR CCM+ Model

The parameterized solid wing was designed to have its root fixed to one side of a rectangular
block which represents the wind tunnel this side was called the ‘fuselage’, and that which is
directly opposite it was called ‘far side’. To the top is another surface called the ‘roof’ and that
beneath it was called ‘floor’. The other two sides were called the ‘inlet’ and the ‘outlet’ with the

wing closer to the inlet than the outlet. The leading edge of the wing root which is on the
coordinate (0, 0, 0) was used as a reference to dimension the sides of the block. The roof and
floor were both 5 meters from the reference point which was just a little above 4 times the
chord length of 1.2 meters. And the inlet and outlet were 3 meters and 12 meters from the
reference point respectively. Placing the wing in a block was a remedy for problems with the
root of the aircraft wing not precisely staying on the reference xy-plane after importing the
wing into STAR CCM+ and then attempting to enclose it in a block within STAR CCM+. The
problem persisted no matter how much effort was given to automate the correction and
prevented the execution of the automated mesh.
Figure 5-3 Rectangular block; model for STAR Figure 5-4 Wing model in rectangular block;
CCM+ model for STAR CCM+
5.1.2 ANSYS Model

The model for ANSYS needed no cause to model the surrounding of the model or its
boundary. The parameterized wing was imported into ANSYS without a modification to the
geometry. This posed no form of additional requirements apart from the importation into
ANSYS which was limited by the available license.
Figure 5-5 Wing model without block model for ANSYS

5.2 File management

The method adopted is a file-based coupling in which files are periodically saved on the
computer hard-disk either manually or automatically. During the process, files of the same
format were being saved repeatedly in the same file name for all configurations and also in
the same file name for different load steps. It is so easy to confuse file of different
29
configuration or load steps for one another as the case may be . A simple, efficient but
manually repetitive management approach was adopted in which a folder was created to save
all files in. This alone was insufficient to avoid the problem so as soon as a single-way
coupling was completed; all the generated files are sent to a compressed zip folder which was
named after the configuration. This FBCoupling file was not made the ANSYS work directory
so as to prevent ANSYS from saving log files in it. So the files from the general ANSYS work
directory was cut and saved in the FBCoupling folder. These prevented a mismatch of result
and saved data.
5.3 File nomenclature system

A nomenclature system was adopted as an additional management tool. The files were
assigned names consistently with an adopted pattern such as ‘2 sweep 5deg 1200 800 3600
317 0’. The first entry is the serial number which is unique for a design variable; the second
entry represents the design variable which is identical for all configurations within a design
variable; the third describes the magnitude of the configuration with reference to the baseline
design as the datum; the fourth which is the root chord length was constant all through the
optimization as a constraint. The fifth is the length of the chord of the wing tip; the sixth is the
aircraft semi-span. The seventh is the distance in millimetres of the leading of the wing tip
along the x-axis; and the last is the distance in millimetres of the leading of the wing tip along
the y-axis.
5.4 Design of Experiments

The Collaborative Optimization which has been discussed earlier requires a systemic and
statistical approach to collection of result data within a black box. The DoE allows extraction
of relationships between objectives and design variables even with limited simulations within
29; 45
the design space . These simulations involved three topological design variables so a
parametric modelling was coined out based on: Tip chord; Semi-span; Sweep angle (only aft
swept and un-swept were considered); and the hedral angle (dihedral and anhedral). The
wing reference area was kept constant all through the experiment as a constraint which was
necessary for a feasible optimization. The baseline design configuration was ‘1 AR 110 ratio
1200 1200 3000 0 0’. It has an aspect ratio of 1:1 with the same chord length of 1.2 meters at
the root and the tip. It has a rectangular plan-form with a semi-span of 3 meters. There were 9
different configurations for each of the design variables among which 1 coincides for the three
6
design variables . This much of configurations were adopted in order to have a more accurate

fit of data and hence a higher quality approximation which makes a total of 25 different
8; 10
configurations to be considered for the design of experiments . It would have been easier
16; 46
to generate the DoE using MATLAB but the licence was unavailable so it was formed
manually as a means of improvisation as:
• Aspect ratio: 1: [1.00: 0.25: 3.00] ratio

• Hedral angle: [+ve12: 3: -ve12] degrees
• Sweep angle: [0: 5: 55] degrees
. . . A list of the seeds of the design of experiments can be seen in the appendix.
47 47
Figure 5-6 +ve hedral angle (dihedral) Figure 5-7 -ve hedral angle (anhedral)
5.5 Simulations
The first stage of both the structural and the aerodynamic simulations was to import the CAD
models into the ANSYS and STAR CCM+ software respectively followed by the other stages
of the simulation. This was run on multiple computers at the same time since the collaborative
optimization method allows this and in order to save time. At a time, the numbers of
computers used for the STAR CCM+ simulation did not exceed 15 computers which was a
limitation in running the CFD simulation because the complete 25 sets of simulation had to be
done in 2 days in order to get the simulations to convergence. The ANSYS simulation on the
other hand was done one after the other because the time requirement to complete a solution
is minimal.
5.6 Importing for analysis

The CAD file of the wing fixed to the fuselage side of the wing box produced in CATIA was
saved as a STEP file because the licence on the STAR CCM+ only allows STEP or IGES.
And also STEP was more preferred because it is being referred to as a newer international
21
standard to be an improved format for saving CAD file generally .

Figure 5-8 Imported geometry in STAR Figure 5-9 transparent geometry

CCM+
The ANSYS importation was a bit of a work around due to the limitation which in this case
was an issue with product the licence. The university has got an academic licence which does
not support import of any of the file formats writable by CATIA. The wing was first saved as a
STEP file and then imported into Solidworks. It was then saved as IGES which was the file
that was imported into ANSYS.
Figure 5-10 Imported geometry appear as Figure 5-11 Line plotted in colour code in ANSYS
lines in ANSYS
5.7 Meshing
Meshing is the representation given to the discretized continuum. This could be representing
either a structural or fluid continuum. The accuracy and fidelity of the simulations to be carried
out on these continuums depends on the type of meshing with which the continuum has been
discretized. This could be a cell in a FVA or an element in an FEA. Apart from being specific
to the material of the continuum, the properties also determine to a large extent how close the
simulation result will be to the actual result. For this MDO analysis, the major meshes used
48
are the tetrahedral and the polyhedral for the FEA and FVA respectively .

49 50
Figure 5-12 A tetrahedral mesh element Figure 5-13 A Polyhedral mesh cell
5.8 Automation
The need for the MDO to be automated can’t be over emphasised due to the nature of
complexity and inter-dependency within the entire framework. This level of automation within
the system depends on what extent the system is allowed computation and logical decision
making within the MDO tool. Designing a tool of this level of complexity requires time for
designing and testing which is a limitation to designing an overall automated tool or system
for the optimization. However, the objective of this project was achieved by improvising
means and writing macros scripts in phases that enabled the automation within STAR CCM to
a large extent. Processes in other programs require not as much time as that of STAR CCM+
so preference was given for automation in STAR CCM+ while others was done manually to a
large extent. Even though this project focuses on achieving a multidisciplinary design
optimized aircraft wing, the framework was followed in accordance to the NASA requirements
for a MDO – creating a path to designing a standard MDO tool as a follow-up to the
framework followed in this project. However, the fully automated MDO tool will require a full
5
automation of all the software as one of the criteria described by NASA which was un-
achievable within the limited time. Asides this requirement, manual repetition of a long and
cumbersome process involving a staggering convergence between STAR CCM+ and ANSYS
is prone to human error which was actually solved by automation. This involved a lot of logical
reasoning, tests, corrections and physical time consumption but was worthwhile for
consistency and avoidance of systematic and random errors
5.9 Finite volume method – Finite element method

Both ANSYS and STAR CCM+ allow the simulation of FSI independently. ANSYS has a finite
element solver as well as a finite volume solver while STAR CCM+ has only the finite volume
solver. However the analysis has been simulated using ANSYS to solve for the structure and

STAR CCM+ to solve for the fluid. This is because the available licence on ANSYS is limited
to fewer numbers of mesh which would have greatly reduced the accuracy of the simulation
since the fluid is of a large volume and needs a higher number of mesh cells at the wall
boundaries. Besides, further simplifications such as the use of symmetry were not possible
since there is no form of symmetry in the model. Also, the use of polyhedral mesh cells in
STAR CCM+ has no equal when compared to the available meshing methods in ANSYS.
This prioritizes the use of STAR CCM+ for the fluid simulation. Asides solving for fluid, STAR
CCM+ also has a solid stress solver that uses the FVM to solve for structures. This lead to
weighing the differences between solving for structures in ANSYS with the usual FEM and
solving for the structures in STAR CCM+ using the FVM.
The major difference between FVA and FVM is the location of the displacement, strain, and
stress in the discretized continuum. The displacements in the FEA are nodal during
computation while the stress and strain are evaluated at the gauss points whereas in FVA,
the displacements and strain are computed at the cell centres and then interpolated to the
face centres. The stresses are then computed at the cell centre. FVA is however less
accurate than its counterpart because it then has to use some interpolation technique other
than the shape function of the element used by the FEA. Apart from the method of
approximations available, FEA allows the selection of element types, the number of nodes,
availability of mid-nodes and several other properties to better define the element to soothe
the deformation the structure will be subjected to. Many more differences are discussed in the
23; 51
references .

6 IMPLEMENTATION OF FSI (FVM & FEM)
The physics in reality is the interaction of the air at a Mach number of 0.7 with the solid
aircraft wing model. This was imitated cycles of simulation of the pressure distribution on the
wing and solving the structural analysis with the load from the pressure distribution
continuously until a staggered convergence was reached between both solvers. This is
represented in the cycle below.
6.1 Phase I:
The first phase of the FSI analysis involves the computation of the pressure load on the
aircraft wing. The essence was to get the converged pressure distribution on the surface of
the wing. This is meant to be a repeated process so a macro was made to automate the first
phase and used to simulate the pressure distribution on the 25 wing configurations. Following
the explained process of FSI above, the CFD analysis was computed first in STAR CCM+
using a macro script that was compatible with all the configurations. The processes are as
follows:
With the STEP file wing model already imported into STAR CCM+, a node of new geometry
parts was created. An operational command was then given to split parts by angle into
surfaces. Choosing a particular angle was important because it determines if the macro will
be feasible for all wing configurations, which was important since the feasibility of the other
stages of automation depends on the initial. The volume of the void in the block housing the
wing was then extracted which was used in creating a region that described the air around the
wing. The region was bounded and named according to the names of the faces described
above. The area joining the wing and the fuselage was missing as a surface in the subtract
(this surface was created in another attempt to simulate the entire CAe within STAR CCM+
using the FVM to solve for solid stress but not needed for the file based coupling). The
surfaces were named according to the convention given earlier. The automated part-based
method of meshing was used with polyhedral mesh, prism layer mesh, and the surface
remesher. This is because the polyhedral mesh is the most accurate and easily automated
unstructured meshing technology for volume region discretization presently available with a
large amount of time saving. The prism layer on the other hand was used near walls to
capture and damp the boundary layer effect to preserve the accuracy of the simulation while
the surface remesher and automatic surface repair automatically fixed issues with bad quality
mesh cell faces. The walls at the boundary which was meant to be defined as a slip wall was
assigned a no-slip wall property which was one of the simplifications that was applied to
speed-up the rate of convergence. This also gave a steadier convergence within a shorter
time of about 4.5 hours on a single core i7 computer converging at an average of 600
iterations compared to when the walls were having a slip condition. The outlet was defined to
be a pressure outlet and the inlet was defined as a velocity inlet with a velocity of 240 �� −1

which was approximately equal to a Mach number of 0.7. These were the processes within
the first phase of the automation which excluded running the simulation so the mesh can be
manually checked for errors before execution the run operation (the mesh settings were
carefully selected such that it needed no manual repair for all the configurations). The macro
was saved as “��”.
The physics continuum of the region was defined to be air at:
• Temperature of air: 300 �
• Density of air: 1.18415 ��−3
• Dynamic viscosity: 1.85508 × 10−5 �� −1
The solvers and models of the physics continuum are:

• Spalart-Allmaras Turbulent
• Reynolds-Average Navier-Stokes
• All y+ wall treatment
• Constant density
• Steady flow
• Segregated flow
The new technique of meshing within STAR CCM+ known as Part-Based Meshing was used
to discretise the mesh continuum of the region. The polyhedral mesh which makes use of an
unstructured gridding technology, reported to be of high accuracy and containing about five
times lesser cells than a tetrahedral mesh for the same initial surface. Prism layer mesh was
also used near walls to capture the boundary layer effect which makes the combination of a
hybrid mesh. Then the surface of the wing was controlled to be finer for a higher degree of
accuracy in computing the pressure distribution in particular. The mesh execution was
excluded from the automation so as to allow for inspection so was the running of the
simulation too. So after the completion of the compilation of the macro, the mesh is executed
followed by the run at the click of the “execute” and “run” buttons respectively. On the
average, running every simulation takes 700 iterations to confirm a steady convergence which
lasted hours per simulation.
Figure 6-1 Mesh cells showing

polyhedral cells
Figure 6-2 Zoomed-in mesh cells

6.2 Phase II: Writing the CAE file (.cdb)

The simulation from first phase converged to a steady state with pressure distribution
developed on the wing. This pressure load has to be exported by STAR CCM+ that writes
pressure load on faces, and also imported by ANSYS that solves at the element nodal level
and can import pressure loaded on faces. That is, STAR CCM+ writes the pressure on faces
and ANSYS reads it from faces. In order to transfer this data into ANSYS, it has to be written
on ANSYS faces. A face in this case does not mean the wing surface but the mesh faces that
form the surface of the wing. In order to achieve this, ANSYS creates a mathematical
representation of the wing surface discretised by the meshed elements. This describes the
geometric location and intersection among the elements that forms the surface using the
element nodes. This was achieved by creating a CAE file in ANSYS which represents the
wing. To achieve this, the wing model also had to be discretised in ANSYS just like it was in
STAR CCM+. The wing had to be meshed with 20-nodes 186 solid element type using a
tetrahedral cell which was the only element type compatible with STAR CCM+ for a FSI
coupling simulation. Controlling the mesh size at the trailing edge and most importantly at the
leading edge was achieved by line controlled meshing within ANSYS. This was necessary in
order to allows gradual transition of pressure magnitude from one mesh cell face to the
nearest neighbour cell face. This is because the tetrahedral mesh has no mid node which
makes interpolation a problem which can in this case only be solved by creating fine cell
meshes at high transition areas on the wing. The file was then saved by using the command
interface within ANSYS to write and save the CAE model. Below are the commands that
saved the file as a “.cdb” file in the ANSYS working directory.
Table 6-1 Command for preparing the CAE model in ANSYS

Command Action
/POST1 This initializes the postprocessor
CDWRITE,DB,FSI_auto.cdb This writes the file and saves it in the database of the ANSYS
working directory with the name “FSI_auto” in the .cdb file
format
The physics continuum of the region was defined to be air at:

• Young’s modulus: 1.2 × 105
• Density of air: 1.18415 ��−3
• Dynamic viscosity: 1.85508 × 10−5 �� −1
The .cdb file format is the only CAE compatible format writable by ANSYS and also readable
by STAR CCM+ which requires including “FSI” somewhere in-between the file name in order
to recognise it and treat it as a fluid-solid interaction representation from foreign CAE
software. The file name “FSI_auto” is common to all the configurations in order to allow an
adequate automation within the STAR CCM+ environment for all configurations too. This
requires adequate file management which was adopted in a method described earlier in
which the “.cdb” file is cut out of the working directory and saved in the FBCoupling folder
periodically.
Figure 6-3 Controlled line mesh in ANSYS Figure 6-4 Tetrahedral mesh with fine
elements at both edges
6.3 Phase III: First-way coupling and mapping

Now the CAE representation of the wing is residing in the FBCoupling folder of the computer
awaiting importation into STAR CCM+. It is important at this stage to ensure that the
“FSI_auto.cdb” in the folder is for the particular wing configuration. With this confirmed, the
next macro, “��” is being written carried out to execute the first-way, file-based
coupling and mapping of the pressure distribution onto the CAE representation. To achieve
this, the CAE model was imported from the FBCoupling folder into STAR CCM+ which
automatically recognises it as a CAE model meant for FSI simulation. This creates an
imported node within STAR CCM+ which is used for the mapping operation. This requires
setting the mapping direction to be from STAR CCM+ to the imported model and specifying
the boundary surfaces of the wing in both models.

23
Figure 6-5 Cycle of one-way/two-way coupling
Within a surface are several faces of the mesh cells. The surfaces in STAR CCM+ consist of
unstructured polyhedral cell faces while that of ANSYS consist of structured tetrahedral cell
faces that are triangular. In order to successfully map the pressure from the polyhedral; faces
unto the tetrahedral faces; an interpolation technique was adopted within STAR CCM+ known
as nearest neighbour interpolation technique which was also fine-tuned by adjusting the
setting to ensure a more accurate interpolation during mapping. The use of the nearest
neighbour interpolation technique was adopted all through the simulation because it has
shown to be generally most efficient on the wing model with different configurations compared
to the other interpolation methods available for mapping within STAR CCM+. It was
necessary to visualize the scalar scene after the mapping has been completed to determine
how efficient the mapping was and if it needed any alteration to the mapping technique
already used. So a scalar scene was created for visualization in which the wing boundary of
the STAR CCM+ simulation region was added.

Figure 6-6 Wing skin pressure distribution
Figure 6-7 Pressure mapped onto CAE model in STAR CCM+
While mapping the pressure from the wing surface in STAR CCM+ model unto the imported
ANSYS model surface, it is important to be consistent with unit; not just within STAR CCM+
but also with all other CAE/CAD tool involved because mapped data are been imported or
exported as the case may be in forms of magnitude so it is essential to get the units right.
“mm” goes with “MPa” in this case so the field function was set to be pressure and the unit
was set to be MPa. With all these settings in place, the mapped data already on the CAE
model was exported together with the nodal-position definition of the CAE model and saved
automatically in the FBCoupling folder of the computer hard disk as a “.mac” file (the only
CAE file format readable by ANSYS), awaiting to be read by ANSYS as a pressure load.
6.4 Phase IV: Simulating static and modal analyses

Again, this phase was not automated as it was done in ANSYS. The structural displacement
was solved in this phase during the static analysis after reading the pressure load from the
FBCoupling folder. The solution was saved in the same folder as a “.Lis” file which is the only
CAE file writable by ANSYS and readable by STAR CCM+.
In order to achieve this, a new analysis was started in ANSYS for static analysis. The root of
the wing was defined to be fixed at all degrees of freedom since it is fixed to the fuselage
which will have no displacement relative to the wing. Within the GUI, the command to read
data from file was selected and the FBCoupling folder was selected. This was visualized as
the pressure was been read all over the wing surface. The visualization was necessary so as

to make corrections in cases when the load was not read completely as this was the case
with some of the wing configuration and was one of the reasons the ANSYS simulation was
not being automated at the time. The solution was done and the nodal displacements were
saved in the FBCoupling folder waiting to be read by STAR CCM+. During the reading stage,
what ANSYS static load solver does is to redistribute, interpolate and convert the pressure
written on faces as pressure load into normal force vector at all wing mesh surface nodes.
The pressure loaded is been converted into force vector and then solved for the static
analysis. The Von Mises Stress was also recorded as this was one of the objective functions
in the multidisciplinary design optimization. The modal analysis was done too using the same
mesh but without the pressure loading condition since the boundary condition was the only
requirement for the modal analysis. The first two of the modal natural frequencies were also
recorded just for reference in case the computation of the flutter and divergence will be of
interest in future research while using programs like ZAERO.
Figure 6-8 Mapped pressure Figure 6-9 Mapped pressure, wing with fixed end
Figure 6-10 Contour plot of Von Mises Stress on wing

Figure 6-11 Plot of wing under displacement in ANSYS
At this stage in all the file-based coupling processes and simulations, it is important to save
the already generated file and clear the file that will no longer be uses for the subsequent
simulations. All the stored files apart from the CAE file (.cdb) and the nodal displacement
solution file (.Lis) were sent to a compressed zip folder which was named according to the
model configuration naming system. This was a measure adopted to avoid miss-match of
data among the wing model configurations.
6.5 Phase V: Second-way coupling and mapping

The first-way has been completed successfully and the next stage of the file-based coupling
was resumed in STAR CCM+ which was also automated using macros. Most care was
needed at this phase to ensure the logics and framework of two-ways file-based coupling was
strictly adhered to.
Figure 6-12 View of deformed wing through transparent block mesh

The written macro at this phase was named “��ℎ��”. What it does is importing the
nodal displacement data from the FBCoupling folder. After this step, the data has only been
imported. It has not yet been mapped even though it has been coupled in the earlier stage
during the importation of the CAE model into STAR CCM+. The settings and controls were
repeated during the second-way mapping as it was done during the first-way mapping. This
ensured the quality of the mapping is as efficient as it was at first. Two scalar scenes were
then created to visualize the imported displacement and the imported mapped displacement.
The CAE model and the STAR CCM+ wing boundaries were the respective parts for the
scenes. This was used to determine the accuracy of the mapping by visual inspection and
also to visualize the actual displacement of the wing boundary within STAR CCM+.
Figure 6-13 Result of static nodal solution in ANSYS
Again, this visualization is just for the sights and has not actually displaced the wing region
and the mesh cells around it. This was achieved by performing a technique within STAR
CCM+ known as morphing which is a form of motion in a way that the simulation regions will
understand. It involves step(s) displacement of the polyhedral mesh cells within the region.
The efficiency of this procedure is actually as accurate as re-designing the wing to the new
assumed position defined by the results from ANSYS, re-meshing and then running the
simulations. This would have been a problem but has been overcome by the feasibility of this
technique. The only limitation in this case is that even though it has a solver for large
displacement, there is limitation to the amount of displacement allowed by the solver which
does depend on the orientation of the mesh cells around the boundaries of the region
undergoing displacement. This gave a “negative or zero volume cell(s)” error in some cases
and limited number of ways a simulation can go during the file-based coupling. However, in
instances where the error was not observed, it was required to perform a pseudo-time

stepping in an unsteady solver with other solvers frozen. This involves giving the fuselage an
in-plane boundary condition and setting the wing boundary to morphing motion displacement.
This displacement was defined to be composite which was set to read the �, �, ��
components of the imported mapped displacements as a field function. The solvers which
were running initially was then set to change from steady to implicit unsteady while the
segregated flow, Spalart-Allmaras turbulence solver, and the Spalart-Allmaras turbulence
viscosity solver were frozen during a time step. This resulted in a morphed wing into the
shape and position defined by the nodal solution from ANSYS. It was compulsory to then
return back to the normal steady solution and unfrozen solvers. These were all change back
to the initial values apart from the implicit unsteady solver which could not be changed back to
the steady solver. So the implicit unsteady solver was then frozen and the simulation run
resumes as normal until convergence. All these process were all automated due to the
complexity which made the automation within STAR CCM+ strictly necessary and invaluable.
As a measure it is required to ensure that all file directly readable in the FBCoupling folder at
this moment are only those needed. Others were sent to the same compressed zip folder for
the particular wing configuration.
Figure 6-14 Scene of imported displacement corresponding to UY in ANSYS (Fig 6-13)
6.6 Phase VI: Second cycle displacement simulation

The procedures within this stage involve solving for static analysis in ANSYS as it was done
earlier on in the phase IV. To start with, the initially loaded condition which was read from the
FBCoupling folder on the wing model was deleted so as not to be considered for this stage of
simulation. With the boundary condition at the root still applied, the new pressure distribution
was read in a similar method as it was done before. The solution was done again but now just
the static analysis was done since the modal analysis is independent of loading condition.
The nodal displacement and nodal Von Mises Stress were recorded while the “.Lis” file was
again saved for the next STAR CCM+ simulation.

Figure 6-15 Small displacement
Figure 6-16 Large displacement
6.7 Phase VII: Second cycle pressure simulation

Automation was made for this stage too with process and steps similar to that in phase III.
The major difference is that coupling which has already been established between ANSYS
and STAR CCM+ was not repeated but it was mapping, freezing, and saving as usual. The
nodal displacement file in the FBCoupling file, on the hard disk was again imported into STAR
CCM+ and then mapped in a similar way as it was done in phase III with direction going from
imported model to STAR CCM+ model. Setting and controls were as usual and the solvers
were frozen and unfrozen as explained earlier in Phase III. The pseudo-time stepping was
completed at the end of the time step run and the pressure was checked and exported into
the FBCoupling folder.
6.8 Phase VIII: Subsequent mappings

A logical rule with tolerance of ±5.0 % of the initial maximum resultant displacement
magnitude was adopted to determine if the next cycle of mapping will hold or not. This was
because there was a higher consistency between mapped displacement and the computed
displacement than it was with the mapped pressure and computed pressure. There was no
need for a third cycle during any of the file-based coupling simulations. However, in cases
where the need be, phase VII and VIII were repeated successively.
6.9 Phase IX: Post-processing results

At the time, the simulations have both staggered to convergence or failed continuity of the
cycle. The results were all recorded for the objectives on all wing configurations. These
include that which has already been recorded in ANSYS for Von Mises Stress, Maximum
Nodal Displacement, First Natural Frequency of Vibration, and the Second Natural Frequency
of Vibration. The Coefficient of Lift and the Coefficient of Drag was also recorded in STAR
CCM+ as aerodynamic characteristics by an automated macro named “��”.
However, the results considered for modelling the metamodels were those after the first cycle
to achieve uniformity since most of the large displacements were large enough that the
morpher solver could not handle. The results are listed in the appendixes as FSI simulation
results.

7 IMPLEMENTATION OF FSI (FVM only)
After weighing the cons and pros of running the FSI simulation by using FVM only, and using
both FVM and FEM coupled together, an attempt was made to solve the FSI problem with
FVM only on STAR CCM+. Automation of the approach was the greatest merit as it was very
adaptable and allows a fully automated simulation.
The same STEP file wing design that was used in the previous implementation describe in
chapter six was used and the ANSYS part of the FSI simulation was omitted. The first macro
creates a new geometry part and then extracts the fluid volume from the geometry resulting in
a part with the inner geometry part surface boundaries. This extract volume which is now a
part is been duplicated and all surfaces were deleted apart from the wing skin. The wing skin
at this instance which has no surface at the root was then renamed and repaired to close the
root opening. This makes two regions: the extract volume region and the wing region. Both
parts were assigned to regions separately and were defined with a fluid and solid continuum
respectively. The two parts were meshed and unsteady simulation done on them after the
motions for both regions has been prepared for morphing and solid displacement motions.
The result was not as promising as it appears the displacement was too large for the solver to
handle and returned an error, “zero or negative volume cells error” for all simulations. The
time requirement for fixing the problem and the probability of a remedy was blik compared to
the inaccuracy of using a FVM only. However, mistakenly defining the wing region as fixed
gave a solution for all simulation which was clearly wrong as the largest displacement was not
at the wing tip as expected. A considerable amount of time was spent on this paradigm which
yielded questionable results, calling for a switch back to the first principle of FEM and FVM.
Figure 7-1 Questionable maximum displacement on wing (FVM only)

8 RESULTS
The work done was in several stages and types. They all contributed to the final solution of
the optimization problem on the wing model. Some of which were highly essential in the
methodology but have no concrete results, like the parameterization which has already been
discussed in chapter three and others whose result were landmarks that determined the next
stage of the methodology and eventually the final solution towards achieving an optimized
wing configuration. These include the.
• Airfoil analysis
• FVM & FEM aerodynamic analysis
• FVM & FEM structural analysis
• FVM only aerodynamic and structural analysis
• Metamodelling
• Multi-objective Genetic Algorithm optimization
8.1 Airfoil analysis

The results in this section are that of the concerned NACA 5412 airfoil only since it was the
airfoil considered in modelling the wing. The figures below show the influence of the 5 degree
on the pressure distribution.
Figure 8-2 Airfoil in C-grid imported into

Figure 8-1 Airfoil in C-grid designed in
STAR CCM+
CATIA

Table 8-1 Aerodynamic characteristics of the airfoil
Airfoil with 0 degree angle Airfoil with 5 degrees angle of
of attack attack
Coefficient of 50.0598 37.9970
Lift
Coefficient of 7.3765 7.6584
Drag
Lift:Drag 6.7864 4.9615
Ratio
The scalar scenes, residuals and the coefficient of lift and drag reports are attached to the
appendixes for both airfoil analyses.
Figure 8-3 Scene summary of airfoil 5412 results
8.2 FVM & FEM aerodynamic analysis

The results of the simulation on the 25 wing models are given below. It represents the
maximum pressure on the wing skin before and after mapping for all the wing configurations.
Also included are the recorded readings mentioned in the methodology. They include all
components of displacement magnitudes; the maximum Von Mises Stress; the first modal
natural frequency; and the second modal natural frequency. The complete table can be found
in the appendices as results for 9 AR went wrong and was neglected. The modal analyses
results too were not included in the optimization so as to limit the objectives and design
variables to three each in order to facilitate the visualization of the Pareto front.

Table 8-2 Results with partial variation of Aspect Ratio

STAR Mapped Co Of Lift Co of UXYZ SEQV Lift/Drag
(Pa) (Pa) Drag (mm) (MPa)
1 AR 34693 33818 192243 11204 13.828 64.482 17.15843

2 AR 34612 33445 206019 10475 23.355 75.887 19.66768
3 AR 34559 34034 214614 11196 34.533 83.127 19.16881
4 AR 34521 32863 222304 9443 46.704 90.784 23.54167
5 AR 34445 32461 226289 9002 58.813 98.194 25.13764
6 AR 34438 32474 229785 8684 71.096 104.17 26.46073
7 AR 34447 32837 232587 8439 83.008 104.45 27.56097
8 AR 34447 33072 234639 8228 93.823 111.57 28.51714
Table 8-3 Results with partial variation of Hedral Angle

STAR Mapped Co Of Co of UXYZ SEQV Lift/Drag
(Pa) (Pa) Lift Drag (mm) (MPa)
1 hedral 34403 33814 216366 9956 39.817 108.04 21.73222

2 hedral 34607 33729 216002 9859 37.359 96.201 21.90912
3 hedral 34456 33633 215793 9878 35.680 90.503 21.84582
4 hedral 34559 33647 215527 9880 34.771 87.852 21.81447
5 hedral 34559 34034 214614 11196 34.533 83.127 19.16881
6 hedral 34527 33387 213739 9774 34.808 83.521 21.86812
7 hedral 34498 32446 212318 9671 35.404 80.671 21.95409
8 hedral 34585 32321 210709 9541 37.561 81.336 22.08458
9 hedral 34644 32808 201408 13396 40.124 84.066 15.03494
Table 8-4 Results with partial variation of Sweep Angle

STAR Mapped Co Of Co of UXYZ SEQV Lift/Drag
(Pa) (Pa) Lift Drag (mm) (MPa)
1 sweep 34559 34034 214614 11196 34.533 83.127 19.16881

2 sweep 34478 33636 216143 9940 35.736 86.046 21.74477
3 sweep 34131 33554 217481 10039 38.199 91.056 21.66361
4 sweep 33743 30709 212951 9745 45.864 153.48 21.85233
5 sweep 32342 29654 209025 9499 52.264 168.32 22.00495
6 sweep 31583 28263 203059 9119 61.170 224.66 22.26768
7 sweep 29178 25610 185493 8006 91.427 419.23 23.16925
8 sweep 26358 21997 159938 6492 159.89 625.66 24.63617
9 sweep 25380 19897 144192 5616 227.82 914.41 25.67521

8.3 Metamodelling
The results derived from the simulations and analyses of the file-based coupling of the fluid-
structural interaction discussed in previous chapters and tabulated above is comprehensive
enough but insufficient for a multidisciplinary design optimization. It has to be modelled such
that it will be useful in the multi-objective optimization toolbox of the optimizer, MATLAB. To
1; 2; 29
achieve this, the result data was metamodelled . A partial variation of the three objectives
was analysed as they vary with the design variables individually to determine the fitness
function which defines the objective functions as a mathematical function of the design
25; 29; 52
variable . This is only possible since the cultural approach of identifying the design
18; 53
variable has been observed such that the design variables are partially independent .
8.3.1 Function formation

For the purpose of generalization, assume the objective to be y and the design variables to be
x. That is y is a function of x; where x has three vector elements that defines the vector y
which is characterized by three objectives functions. This can be expressed mathematically
as:
� = [�] = �[�] = ∑�� (�� )
[�] = �� (�� ) + �� (�� ) + �� (�� )
�ℎ��:
• � is the objective
• [�] is the objective vector
• �[�] is the objective function defined by the design variables vector, [�]
• ∑�� (�� ) is the summation of the individual objective functions common to an objective, �
�� (�� ) + �� (�� ) + �� (�� )

[y]=� �� =� �� (�� ) + �� (�� ) + �� (�� ) �
−��/�� (�� ) + �� (�� ) + �� (�� )
�ℎ��:
• �� is the maximum resultant nodal displacement
• �� is the Maximum Von Mises Stress
• ��/�� is the lift to drag ratio
• �� is the maximum resultant nodal displacement objective function
• �� is the maximum von Mises stress objective function
• �� is the lift/drag objective function

8.3.2 Evaluation of the Function of Maximum Nodal Displacement
8.3.2.1 Evaluation of the maximum nodal displacement as a

function of the wing aspect ratio
Table 8-5 UXYZ dependency on Aspect Ratio

AR Factor (*) UXYZ (mm)
1 AR 110ratio 1.2500 13.828
2 AR 125ratio 1.5429 23.355
3 AR 150ratio 1.8000 34.533
4 AR 175ratio 2.0248 46.704
5 AR 200ratio 2.2000 58.813
6 AR 225ratio 2.3965 71.096
7 AR 250ratio 2.5514 83.008
8 AR 275ratio 2.6800 93.823
9 AR 300ratio 2.8125 105.36
Goodness of fit:
SSE: 1.1650
R-square: 0.9999
Adjusted R-square: 0.9997
RMSE: 0.5397
Coefficients (with 95%

confidence bounds)
Figure 8-4 AR vs UXYZ curve fitting

Linear model Poly4:
�(�) = �. �� – ��. �� + ��. �� – ��. �� + ��. ��
�� = �. �� – ��. �� + ��. �� – ��. �� + ��. ��


function of the wing hedral angle
Table 8-6 UXYZ dependency on Hedral Angle

Hedral angle (degrees) UXYZ (mm)
1 hedral 12deg 12 39.817
2 hedral 9deg 9 37.359
6 hedral n3deg -3 34.808
9 hedral n12deg -12 40.124
Goodness of fit:
SSE: 0.1068
R-square: 0.9971
RMSE: 0.1634

confidence bounds)
Figure 8-5 Hedral angle vs UXYZ curve fitting

Linear model Poly4:
�(�) = �. �� − �. �� + �. �� + �. �� + ��. ��
�� = �. �� − �. �� + �. �� + �. �� + ��. ��


function of the wing sweep angle
Table 8-7 UXYZ dependency on Sweep Angle

Sweep angle (degrees) UXYZ (mm)
1 sweep 0deg 0 34.533
2 sweep 5deg 5 35.736
3 sweep 10deg 10 38.199
4 sweep 20deg 20 45.864
5 sweep 25deg 25 52.264
6 sweep 30deg 30 61.170
7 sweep 40deg 40 91.427
8 sweep 50deg 50 159.89
9 sweep 55deg 55 227.82
Goodness of fit:
SSE: 0.8215
R-square: 1.0000
RMSE: 0.5233

confidence bounds)
Figure 8-6 Sweep angle vs UXYZ curve fitting

Linear model Poly5:
�(�) = �. �� − �. �� + �. �� − �. �� + �. �� + ��. ��
�� = �. �� − �. �� + �. �� − �. �� + �. ��
+ ��. ��

8.3.2.4 Evaluation the maximum nodal displacement as a function

design variables
Fitness function for maximum nodal displacement can be mathematically written as:
�� = �. �� + �. �� + �. �� − �. �� – ��. ��
− �. �� + �. �� + ��. �� + �. ��
− �. �� – ��. �� + �. �� + �. �� + ��. ��

8.3.3 Evaluation of the Function of Maximum Von Mises Stress

8.3.3.1 Evaluation of the maximum Von Mises stress as a function
of the wing aspect ratio
Table 8-8 SEQV dependency on Aspect Ratio

AR Factor (*) SEQV (MPa)
1 AR 110ratio 1.2500 64.482
2 AR 125ratio 1.5429 75.887
3 AR 150ratio 1.8000 83.127
4 AR 175ratio 2.0248 90.784
5 AR 200ratio 2.2000 98.194
6 AR 225ratio 2.3965 104.17
7 AR 250ratio 2.5514 104.45
8 AR 275ratio 2.6800 111.57
9 AR 300ratio 2.8125 111.09
Goodness of fit:
SSE: 17.530
R-square: 0.9919
RMSE: 1.7100

confidence bounds)
Figure 8-7 AR vs SEQV curve fitting

Linear model Poly2:
�(�) = −�. �� + ��. �� + �. ��
�� = −�. �� + ��. �� + �. ��


of the hedral angle
Table 8-9 SEQV dependency on Hedral Angle

Hedral angle (degrees) SEQV (MPa)
1 hedral 12deg 12 108.04
9 hedral n12deg -12 84.066
Goodness of fit:
SSE: 11.0900
R-square: 0.9824
RMSE: 1.4890

confidence bounds)
Figure 8-8 Hedral angle vs SEQV curve fitting

Linear model Poly3:
�(�) = �. �� + �. �� + �. �� + ��. ��
�� = �. �� + �. �� + �. �� + ��. ��


of the wing sweep angle
Table 8-10 SEQV dependency on Sweep Angle

Sweep angle (degrees) SEQV (MPa)
1 sweep 0deg 0 83.127
2 sweep 5deg 5 86.046
3 sweep 10deg 10 91.056
4 sweep 20deg 20 153.48
5 sweep 25deg 25 168.32
6 sweep 30deg 30 224.66
7 sweep 40deg 40 419.23
8 sweep 50deg 50 625.66
9 sweep 55deg 55 914.41
Goodness of fit:
SSE: 5234.0
R-square: 0.9923
RMSE: 32.3500

confidence bounds)
Figure 8-9 Sweep angle vs SEQV curve fitting

Linear model Poly3:
�(�) = �. �� − �. �� + �. �� + ��. ��
�� = �. �� − �. �� + �. �� + ��. ��

8.3.3.4 Evaluation the maximum Von Mises stress as a function

design variables
Fitness function for maximum Von Mises stress can be mathematically written as:
�� = �. �� + �. �� − �. �� + �. �� − �. �� + ��. ��

+ �. �� + �. �� + ��. ��

8.3.4 Evaluation of the Function of – Lift/Drag

8.3.4.1 Evaluation of – Lift/Drag as a function of the wing aspect
ratio
Table 8-11 Lift/Drag dependency on Aspect Ratio

AR Factor (*) – Lift/Drag
1 AR 110ratio 1.2500 –17.15843
2 AR 125ratio 1.5429 –19.66768
3 AR 150ratio 1.8000 –19.16881
4 AR 175ratio 2.0248 –23.54167
5 AR 200ratio 2.2000 –25.13764
6 AR 225ratio 2.3965 –26.46073
7 AR 250ratio 2.5514 –27.56097
8 AR 275ratio 2.6800 –28.51714
9 AR 300ratio 2.8125 –29.32321
Goodness of fit:
SSE: 0.2585
R-square: 0.9980
RMSE: 0.2076

confidence bounds)
Figure 8-10 Aspect ratio vs -Lift/Drag coefficient curve fitting

Linear model Poly1:
�(�) = −�. �� − �. ��
– ��/�� = −�. �� − �. ��

8.3.4.2 Evaluation of – Lift/Drag as a function of the hedral angle
Table 8-12 Lift/Drag dependency on Hedral Angle

Hedral angle (degrees) – Lift/Drag
1 hedral 12deg 12 –21.73222
2 hedral 9deg 9 –21.90912
3 hedral 6deg 6 –21.84582
4 hedral 3deg 3 –21.81447
5 hedral 0deg 0 –19.16881
6 hedral n3deg -3 –21.86812
7 hedral n6deg -6 –21.95409
8 hedral n9deg -9 –22.08458
9 hedral n12deg -12 –15.03494
Goodness of fit:
SSE: 0.02727
R-square: 0.6378
RMSE: 0.07386

confidence bounds)
Figure 8-11 Hedral angle vs -Lift/Drag coefficient curve fitting

Linear model Poly1:
�(�) = �. �� − ��. ��
– ��/�� = �. �� − ��. ��

8.3.4.3 Evaluation of – Lift/Drag as a function of the sweep angle
Table 8-13 Lift/Drag dependency on Sweep Angle

Sweep angle (degrees) – Lift/Drag
1 sweep 0deg 0 –19.16881
2 sweep 5deg 5 –21.74477
3 sweep 10deg 10 –21.66361
4 sweep 20deg 20 –21.85233
5 sweep 25deg 25 –22.00495
6 sweep 30deg 30 –22.26768
7 sweep 40deg 40 –23.16925
8 sweep 50deg 50 –24.63617
9 sweep 55deg 55 –25.67521
Goodness of fit:
SSE: 0.0034
R-square: 0.9998
RMSE: 0.0292

confidence bounds)
Figure 8-12 Aspect ratio vs Lift/Drag coefficient curve fitting

Linear model Poly3:
�(�) = −�. �� − �. �� + �. �� − ��. ��
– ��/�� = −�. �� − �. �� + �. �� − ��. ��

8.3.4.4 Evaluation – Lift/Drag as a function design variables

Fitness function for maximum Von Mises stress can be mathematically written as:
– ��/�� = −�. �� − �. �� + �. �� − �. �� + �. ��
− ��. ��
8.3.5 Evaluation of the Functions

The fitness functions for the objectives are summarised as below:
�� = �. �� + �. �� + �. �� − �. �� – ��. ��

− �. �� + �. �� + ��. �� + �. ��
− �. �� – ��. �� + �. �� + �. �� + ��. ��
�� = �. �� + �. �� − �. �� + �. �� − �. �� + ��. ��

+ �. �� + �. �� + ��. ��
– ��/�� = −�. �� − �. �� + �. �� − �. �� + �. ��

− ��. ��

8.4 Optimised Result

The optimization was done using the GAMULTIOBJ function in the multi-objective toolbox of
MATLAB and the results were represented with twenty sets of optimized results, each
containing the three design objectives and three design variables. The Pareto fronts were
then produced. Table and Pareto fronts can be seen below. The enlarged views can be found
in the appendices.
Table 8-14 20 Aircraft wing configurations on the Pareto optimum
Design Objectives Design Variables
UXYZ (mm) SEQV Lift/Drag Sweep Aspect Hedral angle
(MPa) angle (deg) Ratio (deg)
79.2824 211.4182 59.0790 0.0 1.0 0.0866
188.5884 282.1724 74.3239 2.5302 2.9489 -0.1677
193.0739 273.5405 74.7747 0.2533 3.0000 -2.7723
79.2824 211.4182 59.0790 0.0 1.0 0.0866
88.8798 232.4145 62.5867 0.7598 1.4483 -0.2436
91.5037 236.7040 63.10178 1.1999 1.5149 -0.0603
80.2224 216.2298 59.9273 0.0534 1.1083 0.0495
115.0685 253.2734 66.8865 1.4032 1.9967 -1.1549
193.0739 273.5405 74.7747 0.2533 3.0000 -2.7723
82.9150 222.0004 61.0814 0.1137 1.2546 -0.7083
95.8796 242.5688 63.8941 1.7980 1.6167 -0.1044
192.9486 282.8181 74.6617 2.4879 2.9919 -0.2458
138.9821 266.5437 69.7416 2.2709 2.3609 -2.0688
166.5225 268.8482 72.5773 0.3919 2.7217 -1.3272
169.2115 271.4226 72.7843 1.0463 2.7471 -2.6433
94.2438 236.0695 63.7254 0.3726 1.5928 -0.5219
158.2232 273.8174 71.7089 2.2109 2.6148 -0.1164
146.7436 264.8001 70.6399 0.9180 2.4746 -1.6480
131.0411 259.1673 68.9169 0.8300 2.2561 -0.5632
84.4713 228.1977 61.4036 1.1400 1.2981 0.0017
Figure 8-13 Optimum design variable Figure 8-14 Optimum design objective

VALIDATION
The interdependency of the aerodynamic and the structural behaviour of the wing which
portrays a fluid-solid interaction problem is one branch of the CAE that the aeroservoelastic
6
community has consented not to have sufficient verification and validation . The confidence
on the results can be ascribed to the validation of the used methodology which would be
agreed to be insufficient at the cutting-edge of present day technology. With MDO, the
concern is not all about validation but a demonstration that deviation from the result of the
MDO is actually less efficient compared to optimised result. This creates another problem of
having to simplify the weight function among the involved disciplines in order to determine the
direction of efficiency of the deviation from the optimized result. This however forms the test
basis – “experimentally exploring the design space in a neighbourhood of the supposed
6
optimum to determine whether it is indeed an optimal design ” – of proving an optimal design
in FSI problems.
Two of the aircraft wing configurations were then selected from the Pareto optimum table for
re-simulations as it was done on the earlier twenty five configurations. The FSI analyses were
again carried out and the design objectives were objectives were compared to that predicted
by the Pareto optimum. They were in close proximity therefore validates the quality of the
metamodels generated for the objective functions within the Pareto optimum.
Table I Validation of optimum result

Validation (1) Validation (2)
Metamodel Simulation Metamodel Simulation
Sweep angle 2.53 2.53 Sweep angle 2.21 2.21
(deg) (deg)
Aspect Ratio 2.95 2.95 Aspect Ratio 2.61 2.61
Hedral angle -0.17 -0.17 Hedral angle -0.12 -0.12
(deg) (deg)
UXYZ (mm) 188.59 188.60 UXYZ (mm) 158.22 157.49
SEQV (MPa) 282.17 279.99 SEQV (MPa) 273.82 274.05
Lift/Drag 74.32 69.37 Lift/Drag 71.71 73.12
Despite the general lack of validation for FSI simulations and analyses, works has been done
on several instances which would serve as validation. One of such is that of the validation of a
29
transonic wing attached to the aircraft fuselage discussed in publications and the Ikhana
with fire pod optimization[Chan]. An attempt to validate these will be to attempt to optimize the
baseline design of an already validated model before commencing the validation on the
intended design. This requires so much time and therefore not ideal for a routine
multidisciplinary design optimization. Rather, the time and rigor required for the complete
validation was spent on elementary sub-modules meticulously. This includes refinement of
meshes both in the computational fluid dynamics and the finite element analysis. The grid
size of the DoE was also minimized so as to increase the quality of the fits during

53; 54
metamodelling . Again, the fitness function used was not identical all through the
metamodelling. The best was selected of the polynomial fitness degree for every of the
metamodels. The determinant being the Root Mean Squared Error of which the function with
the least RMSE was selected. This is because the Kriging function approximation which has
been used among most modellers has already been reported not to give the best fit in all
29
instances therefore requiring flexibility among a number of functions. This applies to fitting of
the Pareto optimum for the design variables and the objectives too.

DISCUSSION
Reviewing the methodology applied to this case of optimization will reveal that the simulations
and the analyses done for the FSI were of high fidelity so this will boost the confidence of the
results to a great extent. The prism layers helped capturing near wall, boundary layer effect
on the wing and the meshing was meticulously refined and defined to adapt to intricate edges
curves and surfaces. The use of polyhedral mesh for the fluid was also very valuable.
Coupling was also done automatically but checked manually by visually inspecting the
pressure and the mapped pressure scene for all wing configurations. Correction of the
mapping was also possible by checking several interpolation methods to determine the best
for the particular wing configuration. All forms of measures were also adopted in ANSYS to
ensure that the structural properties of the wing were feasible for the purpose of FSI using
file-based coupling between ANSYS and STAR CCM+. The CAE model was written after the
best discretization was achieved by using line controlled meshing. Lastly, an adequate file
management technique was adhered to in order to ensure there was no mis-match of data
during the file-based coupling.
The methodology was actually not as simple as it has been summarised above because this
is a project with numerous area of speciality. The experience was not only on how and what
methods were adopted but also proved that some were absolutely wrong and will lead only to
wrong results. A typical example was that the time spent on working out the parameterization
of the wing would have been sufficient to design the entire wing configurations for the DoE but
the merit is in the knowledge and appreciation of the art of designing and parameterization. A
number of approximations and methods were used to speed up the convergence of the
simulations involved in the project a typical example is defining the walls as no-slip wall for
the rectangular block that represents the wind tunnel. This gave faster convergence and a
more steady pressure distribution on the wing skin.
Parameterization
A lot of effort was spent during the parameterization stage because the wing cross-section
which is an aerofoil is mathematically represented by cubic B-splines. This spline is defined
by pilot points such that the relationship that exist on splines are dependent on the local pilot
points only and not on the entire pilot points that makes up the airfoil. So an attempt to
parameterize the airfoil by adjusting a parameter will only result to the alteration of the spline
around it without affecting the neighbouring pilot points. That causes distortion of the airfoil
which is not allowed in parameterization so the method of mathematically linking all the pilot
points together discussed in chapter three was adopted. This had to be done manually after
fully constraining the airfoil which was too tedious a task to do. An increase in the complexity
of the shape will therefore determine the level of difficulty of the shape parameterization which
may render the method less efficient. The inability to produce a fully parameterized wing may

again render the optimization process less adaptable for complex geometries requiring
several other stages of designing after the spline parameterization.
Modelling
The analyses was to be done on ANSYS and STAR CCM+, having a common model for the
two tools would have been possible if the entire coupling would have been done on just one
tool but have two different tools necessitated local modelling such that the models were
specific to the tool individually. The solid wing as a whole was initially modelled for both
analyses but it was slightly translated at the root of the wing on importing it into STAR CCM+
which made meshing impossible for some of the models as the faces were self-intersecting at
the root. This was why a rectangular block has to be created with it in CATIA. The ANSYS
model happened to be sufficient for the FEA so two different models were made for every
single configuration.
Computational fluid dynamics

The computational fluid dynamics simulation to compute the pressure distribution has been
reported to be the most accurate when computing a MDO but has also been suggested not to
be included in the early design stage because of the high time requirement. However, the
efficiency of the method was worth trading off the accuracy of using a slip wall at the
rectangular block. Ideally, this would have been best represented by a freestream which has
no wall or whatsoever as the boundary of the domain. Using a far field similar to that
described for the airfoil was also not possible because the fluid domain would have been too
expensive to mesh adequately with respect to time. Lastly, the absence of symmetry on the
airfoil made any further simplification impossible.
Finite element analysis

The use of tetrahedral mesh on the wing to run the static analysis was inadequate but was
compromised for higher accuracy by the use of finer mesh at the leading edge and trailing
edge of the wing. This was confirmed by mesh refinement. The root of the wing was also fixed
at all degrees of freedom which was a simplification to avoiding a non-linear analysis in
modelling the boundary non-linearity at the root. Besides, the root was just of a model wing
and is nothing close to the boundary condition of an actual aircraft wing.
Coupling
There are common methods and routes in the use of CAE tools within a discipline but the little
differences that exists among them has to be learned as a custom to a particular tool and to
particular version of the tool. This is why the user guide was a very valuable tool in learning
the requirements of the tools as regards performing coupling specifically between both tools.
It was essential to understand that ANSYS writes .cdb only for the purpose of coupling which

serves as the template for exchange of data. It also writes solution data as .Lis file which is
readable by STAR CCM+ and reads .mac data only for coupling. ANSYS was choosier while
STAR CCM+ was more flexible so the coupling was carried out in STAR CCM+ with the
various interpolation methods available therein.
Mapping
STAR CCM+ computes the pressure on the cell faces of the wing skin and then maps it using
interpolation techniques unto the triangular faces on the CAE model. ANSYS then converts
the pressure on the faces to nodal normal forces for every triangular face and then
interpolates the forces between the normal forces of corresponding nodes.
Mapping has to be done separately from coupling since the FSI simulation was done using
file-based coupling and the coupling is loose. This could have been combined in the case of a
co-simulation which is tightly coupled; especially when using just one tool or familiar tools
(combination of tools with very similar or the same programming language; example is STAR
CCM+ with Abaqus or ANSYS Workbench with ANSYS Fluent). This therefore requires an
appropriate interpolation between STAR CCM+ surface and the corresponding ANSYS
surface. This is because the mesh faces on the surface of the wing are made of polyhedral
face which are unstructured and that of ANSYS which are tetrahedral cells or triangular faces.
Using tetrahedral mesh for the fluid mesh was not even an option as the resulting wing
surface will still not be corresponding to that from ANSYS because the meshing control
techniques in both tools are different hence interpolation will still be requires despite having
both mesh as tetrahedral. It would have also resulted in the loss of accuracy and increase in
computational time which is unacceptable for MDO.
File management
MDO and the use of file-based coupling highly require an adequate file management to
prevent miss-match of data among design configuration. This is very important especially
when the full automation is not possible. This also saves time and prevents coupling failure
because an attempt to couple a CAE model with another model of different configuration
results in a failure which did happen at the initial stage of the analyses so this was actually
adopted as a requirement of the NASA but was perfected logically through learning from
mistakes at the earlier stage of the project.
Simulation results
Iterations were run in STAR CCM+, ANSYS, and MATLAB. Each of these gave results which
was either feedback into the MDO loop or reported as the results of the optimized wing. The
results of the FVM was neglected as it depicts a large variation from the reality even though a
considerable amount of time was given to attempting correction of the error messages since
the FSI simulation was completely automated therein. This was discussed in chapter seven

which made a switch back to the file-based coupling of the FSI problem. The pressure was
mapped, checked for possible error. It was noted that the difference in the pressure
magnitude was negligible when less than 1000 Pa because it in most cases affect the
maximum nodal displacement by a range of 1 – 3 percent. This most likely would have been
caused by having a local cell face with a high pressure gradient compared with the
neighbours. This was why nearest neighbour interpolation technique was most efficient for the
interpolation instead of the commonly used least square method. What it does is that faces
with a high pressure jump will be discarded. This proved to be efficient generally for the
interpolation during mapping. More than a half of the configurations failed morphing due to the
displacement so only the first way mapping results were considered and even those with
successful morphing till a staggered convergence was reached were just within a range
difference of 15% and convergence was confirmed at the second time step. This was
predominated by the configurations in the aspect ratio seeds.
The displacement considered for the objective function was the resultant displacement. This
is because the algebraic sum of pressure on the three laboratory coordinates was all non-
zero. That is, they all have displacement in the x, y, and z directions of the wing so it was
ideal to consider the resultant.
Objectives and Design Variables

The choice of the design variables is a totality of variables that affect the wing topology.
Having it to be three was an advantage because it gives it the limits of dimensions thereby
making visualization of results possible. The sweep angle, the hedral angle, and the aspect
ratio all define the shape and orientation of the wing. It was also required that the design
variables must be independent so other parameters like the wing tapper ratio could not be
considered since it is already linked to the wing aspect ratio. On the other hand, the
objectives must be a subset of disciplines that are dependent. Without this dependency, the
problem will in most cases be solvable if properly formulated but will not be term a MDO.
Minimizing the Von Mises Stress was considered because it is the most practical measure of
stress in an elastic structure and the lesser it is, the better. It would most likely be given an
inequality constrain in an actual aeroelastic structure MDO. Secondly, one other objective
was the lift-drag ratio. The most considered drags are the form drag and the skin friction drag,
both of which increases as the lift increases for a given structure. It is desired to have a higher
lift associated with a drag as little as possible. This was why the MDO problem was
formulated to minimize –lift/drag ratio. Lastly is the maximum nodal displacement. It is
inevitable and also required to have a little amount of displacement on an aeroelastic
structure such as the wing. However, having a large displacement means having a wing with
low torsional stiffness which is will easily lead to all forms of aeroelastic problems like flutter,
divergence and control reversal. This is because the torsion stiffness is inversely proportional
to the maximum displacement of a cantilever beam like the wing. So the third objective was to
minimize the maximum nodal displacement which is at the wing tip.

Metamodelling
It was essential to adjust and check for the most accurate fitness function that relates the
individual design variable to the objective function. It was interesting and fun while fitting the
functions and comparing to select the metamodel with the list value of the RMSE. That was
just nine fitness functions in all combining to form three objective functions. An increase in the
number of the objective functions and increasing the degree of freedom of the design variable
vector will entirely spoil the fun of metamodelling as it becomes cumbersome beyond manual
approach as the objective and design parameter vectors increases which will be the case in a
practical MDO of an actual aircraft wing where the number of objective functions are of up to
hundreds.
Multidisciplinary design optimization

Unlike what many novice and some engineers with little knowledge on MDO will expect, the
results of an MDO is not unique but lies on a front defined by the Pareto optimum which
separates the feasible region from the infeasible region. A lot of understanding and expertise
was needed at this stage but it took the least simulation time. The genetic algorithm
optimization toolbox in MATLAB takes a set of design variables from parent genes and
evaluate if they lie on the Pareto front or not. The combinations of genes from parents in a
logical tournament are somewhat complex, therefore not discussed in this report as it is
constantly undergoing customized improvement. This explains why running the same multi-
objective optimization problem with the same settings will never return the same results when
run in a number of times. The results of the MDO of a problem with more degrees of freedom
is however non-trivial and will require an MDO experienced user to interpreted and consider
the results for possible trade off.

CONCLUSION
There is no equivalence to solving structural problem with FEA and fluid dynamics with FVM.
The use of STAR CCM+ for CFD gives a high fidelity fluid simulation results for pressure
distribution but requires a number of simplifications to a large extent. The quality of such
simulation would be greatly attributed to the use of the polyhedral mesh cells and the prism
layer cells to capture the boundary layer effects. The FEA analysis was also sufficient with an
appropriate controlled meshing method that was adopted as discussed in the methodology.
The coupling and mapping techniques used were also very efficient but the aeroelastic
simulation of the FSI made the two-way coupling simulation difficult to automate which
resulted in less than 50% success. This limited the FSI results to a one-way coupling which
was considered for the MDO.
The MDO done in MATLAB was a great success using the genetic algorithm multi-objective
solver of the toolbox. It produced twenty generations on the optimum Pareto front and
reported that the Pareto optimum of the design objectives lies between a range of maximum
nodal displacement of 79.2824 − 193.0739 (��); a maximum Von Mises Stress
of211.4182 − 273.5405 (��); and a lift/drag ratio of59.0790 − 74.7747. While that of the
design variables has a rage of 0 − 2.4879 (��) for sweep angle; 1 – 2.9919 for aspect
ratio; and 0.0866 anhedral angle to 0.1677 dihedral angle. The expertise of the designer is
then need to adopt any of the optimum.
Table ii Optimum Pareto upper and lower limits

Design Objectives Design Variables
UXYZ SEQV Lift/Drag Sweep Aspect Hedral
(mm) (MPa) angle Ratio angle (deg)
(deg)
Optimum 79.2824 211.4182 74.7747 0.2533 3.0000 -2.7723
Pareto
upper
limit
Optimum 193.0739 273.5405 59.0790 0.0 1.0 0.0866
Pareto
lower
limit

RECOMMENDATION
The greatest limitation that tends to reduce the efficiency of the MDO process is the time
consumption in modelling and running the CFD analysis. This can be improved by running the
simulation on a high fidelity parallel computing machines in which several machines are
simultaneously running the simulations for different design configurations in order to reduce
6
the computational time. Also suggested by reference is the use of accurate but affordable
approximations like the variable-fidelity method to compute the aerodynamic analysis.
STAR CCM+ and ANSYS are very good tools for fluid and structural analyses respectively
and will be sufficient for a FSI with one-way coupling but it is recommended to solve FSI
problems in a single tool when more than one-way is required like it is for the aeroelastic
loading of the wing.
It will be a huge breakthrough to first develop a tool that will analyse the differences among
2
the various interpolation and approximation methods during metamodelling and while fitting
the response surface for the optimum Pareto. This will enable the use of the most accurate
approximation model during the MDO stages. This accuracy can then be improved by
furthering the optimization with the CSSO method if the need be to fine tune the optimum.

FUTURE WORK
An outline has been laid for the framework of the MDO process and a considerable amount of
experience and expertise has been acquired too. The prospects of MDO in designing is fast
becoming a necessity rather than just being optional. Present high fidelity MDO tools are
mostly specific to a set of disciplines and expensive to design, this restricts many medium
scale designing industries from implementing MDO. A suggested future work will be to design
a MDO tool for a large number of adaptable disciplines to facilitate the use of the tools at local
levels of engineering.
The objectives of the MDO tool will be to leverage the available tools for analyses within the
disciplines and useable at the GUI for a medium expertise designer. It should also focus on
improving the process of parameterization and optimum solution representation using an
unstructured interpolation and approximation methods.

REFERENCES
1
KESSELER, E.; VANKAN, W. J. Multidisciplinary design analysis and multi-objective
optimisation applied to aircraft wing. Life, v. 65, n. 85, p. 95, 2006.
2
VANKAN, W.; KOS, J.; LAMMEN, W. DOCUMENT CONTROL SHEET. 2003.
3
SOBIESZCZANSKI-SOBIESKI, J. et al. Advancement of bi-level integrated system
synthesis (BLISS). National Aeronautics and Space Administration, Langley
Research Center, 2000.
4
LAW SOBIESZCZANSKI-SOBIESKI, J. et al. Advancement of Bi-Level Integrated System
Synthesis (BLISS). 2000.
5
PAK, C.-G.; LI, W. Multidisciplinary design, analysis and optimization tool
development using a genetic algorithm. Proceedings of the 26th Congress of
International Council of the Aeronautical Science, 2008. 14-18 p.
6
A.ZANG, T. multidisciplinary design optimization techniques: implications and
oppurtunity for fluid dynamics research. 30th AIAA fluid dynamics conference. 99-
3798 1999.
7
VENTER, G.; SOBIESZCZANSKI-SOBIESKI, J. Multidisciplinary optimization of a
transport aircraft wing using particle swarm optimization. Structural and
Multidisciplinary Optimization, v. 26, n. 1-2, p. 121-131, 2004. ISSN 1615-147X.
8
KROO, I. Distributed multidisciplinary design and collaborative optimization. VKI
lecture series on optimization methods & tools for multicriteria/multidisciplinary
design, p. 1-22, 2004.
9
SOBIESZCZANSKI-SOBIESKI, J.; HAFTKA, R. T. Multidisciplinary aerospace design
optimization: survey of recent developments. Structural optimization, v. 14, n. 1, p.
1-23, 1997. ISSN 0934-4373.
10
TOPUZ, T. Quality assessment of data-based metamodels for multi-objective
aeronautic design optimisation. Second International Conference on CFD in the
Minerals and Process Industries, CSIRO, Melbourne, Australia, 2007. p.
11
KROO, I.; MANNING, V. Collaborative optimization: status and directions. Presented
at the 8th AIAA/NASA/ISSMO Symposium on Multidisciplinary Analysis and
Optimization, 2000. 8 p.

12
GANTOIS, K.; MORRIS, A. J. The multi-disciplinary design of a large-scale civil aircraft
wing taking account of manufacturing costs. Structural and Multidisciplinary
Optimization, v. 28, n. 1, p. 31-46, 2004. ISSN 1615-147X.
13
NAMURA, N.; OBAYASHI, S.; SHINKYU, J. Surrogate-based multi-objective
optimization and data mining of vortex generators on a transonic infinite-wing.
Evolutionary Computation (CEC), 2013 IEEE Congress on. 20-23 June 2013, 2013.
2910-2917 p.
14
BAKAR, A.; ZHANG, K.-S. Multidisciplinary optimization of transonic wing using
evolutionary algorithm. Applied Sciences and Technology (IBCAST), 2013 10th
International Bhurban Conference on. 15-19 Jan. 2013, 2013. 189-195 p.
15
QUAGLIARELLA, D.; VICINI, A. Sub-population policies for a parallel multiobjective
genetic algorithm with applications to wing design. Systems, Man, and Cybernetics,
1998. 1998 IEEE International Conference on. 11-14 Oct 1998, 1998. 3142-3147 vol.4
p.
16
GORISSEN, D. et al. A surrogate modeling and adaptive sampling toolbox for
computer based design. The Journal of Machine Learning Research, v. 99, p. 2051-
2055, 2010. ISSN 1532-4435.
17
MAALAWI, K.; NEGM, H.; EL SHEIKH, M. Aerodynamic/Structural Optimization of a
Training Aircraft Wing.
18
ALEXANDROV, N. M. Optimization with variable-fidelity models applied to wing
design. Hampton, Virginia. 1999. (99-49)
19
COLE, O. A. Multidisciplinary Design Optimization of an Aircraft Wing
(Aerodynamics and Structural). Hertfordshire. 2013
20
LOCATELLI, D. Optimization of Supersonic Aircraft Wing-Box Using Curvilinear
SpaRibs. 2012. Virginia Polytechnic Institute and State University
21
AGARWAL, R. computational dynamics of whole-body aircraft. Wichita, Kansas,
p.125 - 171. 1999. (67260-0093)
22
COELLO, C. A. C.; LAMONT, G. B.; VAN VELDHUISEN, D. A. Evolutionary algorithms
for solving multi-objective problems. Springer, 2007. ISBN 0387367977.
23
ADAPCO, C., 2013.

24
CHINTAPALLI, S. Preliminary structural design optimization of an aircraft wing-box.
2006. Concordia University
25
ALEXANDROV, N. M.; LEWIS, R. M. Comparative properties of collaborative
optimization and other approaches to MDO. Citeseer, 1999.
26
R, A. G. Constraint-based Shape Parameterization for Aerodynamic Designs.
ICCFD7-2001. Big Island, Hawaii. 2012
27
HARDEE, E. et al. A CAD-based design parameterization for shape optimization of
elastic solids. Advances in Engineering Software, v. 30, n. 3, p. 185-199, 3// 1999.
ISSN 0965-9978. Disponível em: <
http://www.sciencedirect.com/science/article/pii/S0965997898000659 >.
28
CATIA. CATIA User Guide. In: (Ed.). CATIA User Guide, 2012.
29
VANKAN, W.; MAAS, R.; LABAN, M. Fitting fitness in aircraft design. 25th
International Congress of the Aeronautical Sciences ICAS, 2006. p.
30
767 case study, Chapter 9 2003.
31
WHITNEY, D. 2003.
32
LI, L.; LIU, J. An efficient and flexible web services-based multidisciplinary design
optimisation framework for complex engineering systems. Enterprise Information
Systems, v. 6, n. 3, p. 345-371, 2012/08/01 2012. ISSN 1751-7575. Disponível em: <
http://dx.doi.org/10.1080/17517575.2011.651627 >. Acesso em: 2013/08/26.
33
HICKS, R. M.; HENNE, P. A. Wing Design by Numerical Optimization. Journal of
Aircraft, v. 15, n. 7, p. 407-412, 1978/07/01 1978. ISSN 0021-8669. Disponível em: <
http://dx.doi.org/10.2514/3.58379 >. Acesso em: 2013/08/26.
34
TOOLS, A. Airfoil tools 2013.
35
JAKIRLIĆ, S. et al. Near-wall, Reynolds-stress model calculations of transonic flow
configurations relevant to aircraft aerodynamics. International journal of heat and
fluid flow, v. 28, n. 4, p. 602-615, 2007. ISSN 0142-727X.
36
WITTEVEEN, J. et al. Uncertainty quantification of the transonic flow around the RAE
2822 airfoil. Center for Turbulence Research, Annual Briefs, Stanford University,
2009.

37
OLHOFER, M.; JIN, Y.; SENDHOFF, B. Adaptive encoding for aerodynamic shape
optimization using evolution strategies. Evolutionary Computation, 2001.
Proceedings of the 2001 Congress on: IEEE, 2001. 576-583 p.
38
ALLEN, E. B.; KHOSHGOFTAAR, T. M.; CHEN, Y. Measuring coupling and cohesion of
software modules: an information-theory approach. Software Metrics Symposium,
2001. METRICS 2001. Proceedings. Seventh International: IEEE, 2001. 124-134 p.
39
KUNTZ, M.; MENTER, F. Simulation of fluid-structure interactions in aeronautical
applications. European Congress on Computational Methods in Applied Sciences
and Engineering ECCOMAS 2004, 2004. 133-144 p.
40
SU, B.; QIAN, R.-J.; YUAN, X.-F. Advances in research on theory and method of data
exchange on coupling interface for FSI analysis [J]. Spatial Structures, v. 1, p. 002,
2010.
41
SU, B. et al. Data exchange method for fluid-structure interaction based on energy
conservation and interpolation algorithm adopting radial basis function []. Journal of
Xi’An Jiaotong University, v. 43, n. 9, p. 114-119, 2009.
42
SATHE, S. Enhanced-discretization and solution techniques in flow simulations and
parachute fluid–structure interactions. 2004. Ph. D. thesis, Rice University
43
LEAHY, J. Computational Aeroelasticity.
44
SCHÜLE, M.; HERRLER, R.; KLÜGL, F. Coupling gis and multi-agent simulation–
towards infrastructure for realistic simulation. In: (Ed.). Multiagent System
Technologies: Springer, 2004. p.228-242. ISBN 3540232222.
45
Sartorius 2013.
46
SIMULINK, M. MATLAB 2013.
47
CLEYNEN, O. Aspects of Aircraft Design and Control lecture 3. In: (Ed.). Aspects of
Aircraft Design and Contro, 2012.
48
SHEFFER, A.; DE STURLER, E. Parameterization of faceted surfaces for meshing using
angle-based flattening. Engineering with Computers, v. 17, n. 3, p. 326-337, 2001.
ISSN 0177-0667.
49
CALIFORNIA, U. O. UCDAVIS 2010.

50
LAZAR, E. A., MASON, J.K., MACPHERSON, R.D. AND SROLOVITZ, D.J. Complete
topology of cells, grains, and bubbles in three-dimensional microstructures.
51
ANSYS. 2013.
52
ALEXANDROV, N. M.; LEWIS, R. M. Analytical and computational aspects of
collaborative optimization for multidisciplinary design. AIAA journal, v. 40, n. 2, p.
301-309, 2002. ISSN 0001-1452.
53
SIMPSON, T. W. et al. Comparison of response surface and kriging models for
multidisciplinary design optimization. AIAA paper 98, v. 4758, n. 7, 1998.
54
______. Kriging Models for Global Approximation in Simulation-Based
Multidisciplinary Design Optimization. AIAA Journal, v. 39, n. 12, p. 2233-2241,
2001/12/01 2001. ISSN 0001-1452. Disponível em: <
http://dx.doi.org/10.2514/2.1234 >. Acesso em: 2013/08/26.

APPENDICES
Aspect ratio: 1: [1.00: 0.25: 3.00] ratio
Hedral angle: [+ve12: 3: -ve12] degrees
Sweep angle: [0: 5: 55] degrees
Seeds of Design of Experiments
S/N Simulation Magnitude Root-chord Tip-chord Semi-span X-trans Y-trans
Aspect Ratio Simulation
1 AR 110 ratio 1200 1200 3000 0 0

2 AR 125 ratio 1200 960 3333 0 0
4 AR 175 ratio 1200 686 3818 0 0
5 AR 200 ratio 1200 600 4000 0 0
6 AR 225 ratio 1200 533 4154 0 0
7 AR 250 ratio 1200 480 4286 0 0
8 AR 275 ratio 1200 436 4400 0 0
9 AR 300 ratio 1200 400 4500 0 0
Hedral Simulation
1 hedral 12deg 1200 800 3600 0 769

2 hedral 9deg 1200 800 3600 0 573
3 hedral 6deg 1200 800 3600 0 380
4 hedral 3deg 1200 800 3600 0 190
6 hedral n3deg 1200 800 3600 0 n190
7 hedral n6deg 1200 800 3600 0 n380
8 hedral n9deg 1200 800 3600 0 n573
9 hedral n12deg 1200 800 3600 0 n769
Sweep Simulation
1 sweep 0deg 1200 800 3600 0 0

2 sweep 5deg 1200 800 3600 317 0
3 sweep 10deg 1200 800 3600 638 0
4 sweep 20deg 1200 800 3600 1317 0
5 sweep 25deg 1200 800 3600 1687 0
6 sweep 30deg 1200 800 3600 2089 0
7 sweep 40deg 1200 800 3600 3035 0
8 sweep 50deg 1200 800 3600 4311 0
9 sweep 55deg 1200 800 3600 5166 0

47
Figure 0-1 Airplanes showing wing configurations
47
Figure 0-2 enlarged view of the wing

Figure 0-3 A number of possible disciplines
Figure 0-4 Parameterized airfoil (enlarged)

Figure 0-5 Cell set within STAR CCM+
Figure 0-6 Cell set within STAR CCM+ (zoomed in)
Figure 0-7 Cell set around wing within STAR CCM+

Figure 0-8 Controlled line mesh within Ansys (Zoomed-in leading edge)
Figure 0-9 Controlled line mesh within Ansys (Zoomed-in trailing edge)

Figure 0-10 STEP file model defect (geometry scene)
Figure 0-11 STEP file model defect (transparent mesh scenee)
Figure 0-12 STEP file model defect (velocity scalar scene)

Figure 0-13 Wing showing displacement through the transparent mesh
Figure 0-14 Wing showing displacement through the transparent mesh

Figure 0-15 Response surface of design variables (Scattered)
Figure 0-16 Response surface of design variables

Figure 0-17 Pareto front of Objectives (Scattered)
Figure 0-18 Response surface of Objectives

Table 15 Comprehensive simulation results

st nd
STAR mapped Co Of Lift Co of Drag UXYZ UX UY UZ SEQV 1 freq 2 freq Lift/Drag
1 AR 110ratio 34693 33818 192243 11204 13.828 1.1381 13.778 1.7075 64.482 0.28030 1.5031 17.15843
2 AR 125ratio 34612 33445 206019 10475 23.355 1.8762 23.275 1.5857 75.887 0.24844 1.3225 19.66768
3 AR 150ratio 34559 34034 214614 11196 34.533 2.7583 34.418 1.4648 83.127 0.22846 1.1049 19.16881
4 AR 175ratio 34521 32863 222304 9443 46.704 3.7242 46.552 1.3247 90.784 0.21514 0.96162 23.54167
5 AR 200ratio 34445 32461 226289 9002 58.813 4.6814 58.624 1.1698 98.194 0.20577 0.86121 25.13764
6 AR 225ratio 34438 32474 229785 8684 71.096 5.6494 70.869 1.0065 104.17 0.19903 0.78839 26.46073
7 AR 250ratio 34447 32837 232587 8439 83.008 6.5912 82.744 0.82161 104.45 0.19378 0.73261 27.56097
8 AR 275ratio 34447 33072 234639 8228 93.823 7.4743 93.523 -0.63340 111.57 0.18980 0.68914 28.51714
9 AR 300ratio 34420 32815 236521 8066 105.36 8.3859 105.03 -0.50148 111.09 0.18663 0.65458 29.32321
1 hedral 12deg 34403 33814 216366 9956 39.817 3.1128 38.818 -8.9591 108.04 0.21460 1.0386 21.73222
2 hedral 9deg 34607 33729 216002 9859 37.359 2.9498 36.765 -6.4953 96.201 0.22073 1.0680 21.90912
3 hedral 6deg 34456 33633 215793 9878 35.680 2.8353 35.372 -4.3624 90.503 0.22517 1.0891 21.84582
4 hedral 3deg 34559 33647 215527 9880 34.771 2.7755 34.624 -2.4504 87.852 0.22774 1.1014 21.81447
5 hedral 0deg 34559 34034 214614 11196 34.533 2.7583 34.418 0.82161 83.127 0.22846 1.1049 19.16881
6 hedral n3deg 34527 33387 213739 9774 34.808 2.7729 34.605 2.6349 83.521 0.22732 1.0994 21.86812
7 hedral n6deg 34498 32446 212318 9671 35.404 2.8046 35.006 4.5131 80.671 0.22441 1.0862 21.95409
8 hedral n9deg 34585 32321 210709 9541 37.561 2.9512 36.839 6.7236 81.336 0.21956 1.0626 22.08458
9 hedral n12deg 34644 32808 201408 13396 40.124 3.1222 38.942 9.2239 84.066 0 0 15.03494
1 sweep 0deg 34559 34034 214614 11196 34.533 2.7583 34.418 0.8216 83.127 0.22846 1.1049 19.16881
2 sweep 5deg 34478 33636 216143 9940 35.736 2.9359 35.613 -0.9225 86.046 0.22786 1.1031 21.74477
3 sweep 10deg 34131 33554 217481 10039 38.199 3.2401 38.062 -1.1997 91.056 0.22399 1.0854 21.66361
4 sweep 20deg 33743 30709 212951 9745 45.864 4.1019 45.675 -2.0560 153.48 0.20721 1.0045 21.85233
5 sweep 25deg 32342 29654 209025 9499 52.264 4.7787 52.028 -2.7185 168.32 0.19447 0.94293 22.00495
6 sweep 30deg 31583 28263 203059 9119 61.170 5.6894 60.863 -3.6569 224.66 0.17920 0.86914 22.26768
7 sweep 40deg 29178 25610 185493 8006 91.427 8.7014 90.826 -7.1587 419.23 0.14251 0.69261 23.16925
8 sweep 50deg 26358 21997 159938 6492 159.89 15.192 158.41 -16.599 625.66 0.10137 0.49195 24.63617
9 sweep 55deg 25380 19897 144192 5616 227.82 21.439 225.20 -27.694 914.41 0.08073 0.37103 25.67521
MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING (AERODYNAMICS & STRUCTURAL) 97

DEGREE WITH HONOURS IN MSC AEROSPACE ENG

Uploaded by

Copyright:

Available Formats

DEGREE WITH HONOURS IN MSC AEROSPACE ENG

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

DEGREE WITH HONOURS IN MSC AEROSPACE ENG

Uploaded by

Copyright:

Available Formats

DEGREE WITH HONOURS IN

MSc. AEROSPACE ENGINEERING

MSc Final Year Project Report

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

Student Full Name: Akinwande O. Cole

Student Registration Number: 11489561

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING (AERODYNAMICS

I reverence God for the grace for making this a reality.

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING (AERODYNAMICS

DECLARATION STATEMENT ........................................................................................................i

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING (AERODYNAMICS

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING (AERODYNAMICS

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING (AERODYNAMICS

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING (AERODYNAMICS

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING (AERODYNAMICS

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING (AERODYNAMICS

AIAA American Institute of Aeronautics and Astronautics

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING (AERODYNAMICS

1.1 Multidisciplinary Design Optimization

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

1.2 MDO Methods

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

1.4 Approximation Methods

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

1.5 Rudiments of MDO

1.5.1 Objective functions

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

1.5.2 Design variables

1.5.4 Baseline design of the wing

1.5.5 Multiple disciplines

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

2.1 FSI and MDO

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

Figure 2-1 Least-squared solution to

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

2.4 MDO Framework

Figure 2-3 Basic framework of MDO

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

3 MODELLING AND PARAMETERIZATION

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

Figure 3-1 Cutaway drawing of the 767

Figure 3-3 Decomposition of component Figure 3-4 Colour coded discipline

Figure 3-7 Airfoil analysis pressure scene of NACA 5412

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

Figure 3-8 Fully constrained airfoil

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

Figure 3-9 Fully parameterized airfoil

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

Figure 3-10 Design variable parameters shown on wing model

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

4 FLUID-STRUCTURAL INTERACTION (FSI)

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

.cdb CAE model used for data

4.3 Coupling Time-Step Classifications

MULTIDISCIPLINARY DESIGN OPTIMIZATION OF AN AIRCRAFT WING

4.3.2 Implicit coupling