Aerospace Science and Technology 92 (2019) 766–776

Contents lists available at ScienceDirect

Aerospace Science and Technology

www.elsevier.com/locate/aescte

Multidisciplinary design optimization of aircraft wing using

commercial software integration
Abdelkader Benaouali a,b,∗ , Stanisław Kachel a
a
Faculty of mechatronics and aviation, Military university of technology, Warsaw, Poland
b
LCSM, Ecole Militaire Polytechnique, Algiers, Algeria

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, a fully automated framework dedicated to the high-fidelity multidisciplinary design
Received 30 October 2018 optimization of aircraft wing is developed. This design framework integrates a set of popular commercial
Received in revised form 25 June 2019 software using their programming/scripting capabilities. It goes through geometric modeling in SIEMENS
Accepted 26 June 2019
NX, aerodynamic meshing in ICEM CFD, flow solution using ANSYS FLUENT, structural finite element
Available online 2 July 2019
modeling in MSC.PATRAN and structural sizing in MSC.NASTRAN. By adopting a parametric modeling
Keywords: methodology, the structural and aerodynamic metrics reflecting the wing performance can be evaluated
Multidisciplinary design optimization given a description of its shape and dimensions. In order to overcome the high cost of simulation models
High fidelity and allow the efficient solution of high-fidelity optimization problems, a surrogate-based optimization
Aircraft wing strategy is adopted. The reliability of the proposed approach is investigated through its application to the
Surrogate-based optimization design of a high-speed passenger aircraft wing. The optimization objective is to maximize the aircraft
range, given by the Breguet equation, while maintaining the lift coefficient and the structural safety. The
case study results in a 8.9% increase in the range by considering shape and structural design variables.
© 2019 Elsevier Masson SAS. All rights reserved.

1. Introduction ters, and does not guarantee the achievement of the best possible
performance. These issues are overcome by the design optimiza-
With the continued growth in air traffic, reflected by the in- tion, which is an approach that combines mathematical optimiza-
crease in the passenger transport demand by 6.3% in 2016 com- tion algorithms with simulation tools in order to automatically
pared to 2015 [1] and by 7.6% in 2017 compared to 2016 [2], the search the design space for the optimal solution [3]. Accordingly, it
need to improve aircraft performance and efficiency is an ongo- is certain that the obtained design, which is precisely determined,
ing concern. In order to meet these requirements, constraints have
is the best possible candidate that fulfills the problem specifica-
been imposed on the development of future aircraft concepts, the
tions and respects all the constraints.
evolution of design tools and the improvement of the design pro-
The application of design optimization in the case of aircraft
cess quality.
systems, however, presents a big challenge [4]. One reason is re-
The designer’s aim is to improve the efficiency of the system
in study so that it performs its function in the best possible way. lated to the complexity of aircraft systems, as they involve lots of
In order to determine the best design, numerical analysis is used parts and a large number of parameters and constraints to be con-
to evaluate the behavior of multiple candidates, from which the sidered. The other reason is the multiple conflicting performance
one with the top performance in terms of the desired objective criteria that the aircraft design entails. These criteria are related
is selected. For that purpose, the traditional approach was based to different disciplines, including structures, aerodynamics, stabil-
on a trial-and-error procedure which is a manual iterative process ity, control, and propulsion. For the wing, which is the aircraft’s
of analyses and modifications, according to the designer’s experi- major lift-generating surface, the aerodynamic shape is respon-
ence, until obtaining a satisfactory design. This approach has many sible for the intensity of loads, which are required to fulfill the
drawbacks, including that it requires an excessive effort, does not flight mission requirements, while the structural configuration de-
have the ability to consider simultaneously several design parame- scribes how well the wing behaves when subjected to extremal
flight conditions. These two disciplines are strongly coupled [5],
and therefore taking into account only one of them while perform-
*Corresponding author at: Faculty of mechatronics and aviation, Military univer-
ing the design optimization may likely comes at the expense of the
sity of technology, Warsaw, Poland.
E-mail address: [email protected] (A. Benaouali). other.

https://doi.org/10.1016/j.ast.2019.06.040
1270-9638/© 2019 Elsevier Masson SAS. All rights reserved.
 A. Benaouali, S. Kachel / Aerospace Science and Technology 92 (2019) 766–776 767

The simultaneous consideration of multiple disciplines falls applying it to maximize the range of a high-speed passenger air-
within the scope the Multidisciplinary Design Optimization (MDO) craft in the class of the Airbus A320.
[6]. MDO is one of the most effective tools used in the design
process because it permits to cover all aspects of the system un- 2. Software-integration-based wing design framework
der study [7]. The application of MDO improves the design in a
global sense and leads to an excellent estimate early in the design In this section, the development of a multidisciplinary design
process. Without the use of MDO, discrepancies may emerge dur- framework which conducts the different wing modeling and anal-
ing the late design steps, for which the level of detail reaches its ysis activities is presented. Given the wing parametric description,
maximum, resulting in a strong requirement for expensive modifi- the framework allows to automatically evaluate the structural and
cations [8]. aerodynamic performance metrics which are used as optimiza-
To implement and solve wing MDO problems, a variety of mod- tion objectives and constraints. The CAD/CAE integration is made
eling and simulation methods are available in the literature. These possible by means of data exchange translators which allow im-
methods can be classified according to their level of fidelity, being porting/exporting the model data to/from neutral file formats. The
the degree to which a model or a simulation reproduces the state design automation is realized by writing scripts in the respective
and behavior of a real-world object [9]. For aerodynamic analysis, programming languages of the software used, and execute them in
the fidelity levels are the result of different simplification degrees batch mode.
applied to the Navier-Stockes (NS) equations, from the Reynolds- The developed design framework, whose work-flow is depicted
averaged NS (RANS) down to the potential flow theory methods in Fig. 1, starts with the geometric modeling module which gen-
[10,11], such as the vortex lattice method and the lifting line the- erates the wing geometry and exports the resulting CAD model.
ory. On the other hand, the structural analysis methods range from Using the CAD model, the aerodynamic meshing module carries
the high fidelity Computational Structural Mechanics (CSM) [10, out the process of domain discretization and exports the result-
12], which use the Finite Element Method (FEM) for discretiza- ing mesh. Next, the CFD numerical calculations are executed on
tion, to the reduced order models, namely the equivalent plate the imported mesh to evaluate the aerodynamic coefficients and
and beam theories. Although modern computational capabilities the load distribution. After being interpolated from the CFD mesh
make it feasible to combine optimization procedures with simpli- nodes to the CSM mesh nodes by the load transfer module, the
fied analysis models, the requirement to increase the precision of aerodynamic loads are used along with the CAD model by the
complex system representations, such as the structural weight and structural modeling module to build the wing FEM model. Finally,
the aerodynamic drag at transonic and supersonic regimes, results the structural analysis module executes numerical calculations to
in a growing interest for high fidelity MDO applications [13–15]. solve the finite element equations and perform the wing structural
The current state of development in software technology al- sizing. These modules are discussed separately in more detail in
lowed to implement high fidelity Computational Structural Me- the next subsections. Tested and commented version of the scripts
chanics (CSM) and Computational Fluid Dynamics (CFD) methods used to automate the design framework are made publicly avail-
into computer codes. Evaluating the cost functions in the MDO able in the authors’ repository [19].
process requires that the software tools used collaborate and ex- In addition to the modeling and analysis tools, the framework is
change the model data, especially when the optimization consid- managed by a C# application acting as the core of the framework.
ers shape design parameters. In this case, the construction of the It is the main driver of the integration process since it is responsi-
different analysis models, particularly the process of domain dis- ble for the modification of design parameters in the scripts, the
cretization, must be updated for each design alternative. Thus, execution of simulation software in batch mode and the post-
implementing a high fidelity MDO requires the development of processing of results.
a generic design environment. On another side, the costly simu-
lations carried out by these tools, reflected by the required CPU 2.1. Geometric modeling
time, obstruct the use of traditional optimization procedures and
stresses the need for more efficient optimization techniques. One The geometry generation is the first task carried out in the de-
veloped design framework. The CAD component accepts the para-
other key problem while developing an MDO environment is the
metric description of the wing shape and automatically generates
data exchange between software tools [16]. Attempts to create a its geometry. The present process chain uses the commercial CAD
“common language” that reduces the number of necessary links software SIEMENS NX 7.5 (formerly Unigraphics). From the various
have been and are being addressed by various research teams, such automation tools offered by NX, the graphic programming inter-
as OpenVSP [17], CPACS [18], ...etc. active language (GRIP) is selected because of its flexibility and
A literature survey on the progress made up to date in the convenience for the present design problem. The three steps of the
field of aircraft design, related to studies based on single-discipline script development, namely compilation; linking and execution, are
or multidisciplinary optimization approaches, using low or high automated using a set of system commands operating on a batch
fidelity models, points out that the application of numerical op- file and can be executed from the C# application:
timization for the design of aircraft wing structures has been de-
gripbatch -c -dev:osfile -name:<output_file_name>
voted to a single discipline, based partially/totally on low fidelity
-dir:<directory_name> filespec
methods, or based fully on high fidelity models however using
gripbatch -l -dev:osfile -name:<output_file_name>
in-house codes which are unavailable for public use. The present -dir:<directory_name> filespec
paper presents the development of a generic design framework gripbatch -r -dev:osfile -name:<output_file_name>
aimed at evaluating the structural and aerodynamic performances -dir:<directory_name> filespec
of aircraft wing configurations. The complexity and yet the power
of the presented approach lies in the fact that it is totally based on where output_file_name specifies the file name to which the
high fidelity models using well-known commercial software. The output is printed, directory_name is the directory where the
whole process of modeling and calculation is executed in a fully script is located, and filespec is the script name.
automated manner. To alleviate the high simulation cost, the de- In this study, the geometry of a wing is described by its sec-
veloped framework is coupled with surrogate modeling techniques, tional and planform shapes. The double tapered planform of most
resulting in an efficient environment for solving MDO problems. transport aircraft wings can be seen as a trapezoidal wing plan-
The functionality of the proposed procedure is demonstrated by form with an extra triangular area along the inboard trailing edge.
768 A. Benaouali, S. Kachel / Aerospace Science and Technology 92 (2019) 766–776

Fig. 1. Workflow of the wing design framework.

Therefore, its parameterization can be obtained by adding two which are preferred to be used by the CFD meshing software and
more parameters to the basic parameters of a trapezoidal wing structural modeling software, respectively.
which are:
2.2. Aerodynamic meshing
• C r - root chord length
• b/2 - semi-span For this study, the popular meshing software ICEM CFD 15.0,
• λ - taper ratio module of ANSYS commercial package, is used to generate a multi-
•  L E - leading edge sweep angle block structured mesh around the wing geometry. To meet the
automation requirement, scripting is performed using the Tool
The additional parameters are Command Language/Tool Kit (Tcl/Tk) through a script written in
a so-called replay file (.rpl) which can be interpreted by ICEM CFD.
• γc - ratio of the trapezoidal chord C tpz to the root chord The batch mode execution is realized through the following system
• γs - ratio of the kink position to the semi-span command:

For the airfoil parameterization, the Class/Shape Transformation <ICEMCFD_path> -batch -script <script_file_path>
(CST) method [20] is chosen because of the interesting features,
including efficiency; flexibility and intuitiveness. A C# function is As the meshing process is carried out without the designer
developed to compute the point coordinates given a set of CST intervention, the grid generation script must be compatible with
parameters. Using this function, files containing the point coordi- various wing shapes while ensuring a converged and correct solu-
nates of the upper and lower curves are generated before the GRIP tion for each case studied. Moreover, it must ensure the absence
script is executed. In addition to the airfoil and planform parame- of any line crossings between the blocks since they cause nega-
ters, other parameters describing the number and configuration of tive volumetric elements, which may yield a poor quality mesh. On
the wing spars and ribs, which are required for the finite element that account, the script must be based on the geometric design pa-
modeling, must be specified. rameters as well as a set of characteristic point coordinates, angles
The CAD modeling starts by reading the coordinate files and and lengths related to the wing surfaces. The values of these char-
creating the corresponding points. Spline curve definition is then acteristic entities are obtained through the geometry manipulation
used to generate the airfoil upper and lower curves from the cre- tools available within ICEM. The aforementioned requirements em-
ated points. Next, the obtained curves are scaled using the value phasize more the necessity of an extreme manual labor to program
of the root chord to define the root section. A series of geometric the grid generation so that the resulting mesh for any configu-
transformations are then applied in order to define the kink and ration yields a converged and correct solution. Thus, writing the
tip sections. These transformations include translation to the kink ICEM script presents the most complex task in the development of
and tip locations, scaling by the taper ratio at the kink and the the design framework.
tip, and rotations using the sweep back angle as well as the twist The grid generation process, automated by the written script,
distribution at multiple spanwise stations. The wing skin is cre- involves the following steps:
ated by extruding surfaces through the root, kink and tip curves.
The GRIP program execution is terminated by exporting the CAD (i) Import the wing geometry from the neutral CAD file;
model to the common data translation formats Parasolid and IGES, (ii) Create the fluid domain and define the boundary surfaces;
 A. Benaouali, S. Kachel / Aerospace Science and Technology 92 (2019) 766–776 769

it is not involved in flow simulation. The edges of this block are

associated with the wing curves so that the mesh captures the
wing surface shape. This is done using ICEM tools, namely “Block
splitting”, “Ogrid block” and “Edge/curve association”. In order to
generate a more refined mesh near the wing boundary layer while
avoiding large dimension cells at the far field boundary, a two
stage distribution is realized by duplicating the block around the
wing.
Fig. 3 shows the wing surrounding blocks for the developed
O-O mesh topology. The edges of these blocks are adjusted per-
pendicularly to the airfoil curves so that the mesh lines follow the
normal direction. This avoids high skewness levels which may re-
sult in a poor orthogonal quality especially at the surface boundary
layer where the aspect ratio levels are relatively high.
In addition, the developed script ensures a smooth spacing vari-
ation across the mesh volume. This variation is achieved by match-
ing the mesh spacings at the different block interfaces using the
ICEM tool “Match edge spacing”. Fig. 4 depicts an example of the
CFD mesh generated within the automated meshing module of the
developed design framework.

2.3. CFD calculations

Fig. 2. Overall computation domain for the O-O mesh topology.
This module is responsible for the flow field calculation around
the wing surface. The RANS commercial package ANSYS FLUENT
15.0 is used as the flow solver. The CFD process is automated by
executing a journal (.jou) file in FLUENT batch mode through the
following system command:

<FLUENT_path> 3d -hidden -t4 -i <journal_file_path>

where -t4 specifies the number of CPU launched for a FLUENT

parallel session. The journal file is written using the Text User
Interface (TUI) and Scheme commands which automate the case
Fig. 3. Wing surrounding blocks for the O-O mesh topology. setup, the grid operations, the solver settings, the calculation pro-
cess and the result post-processing.
(iii) Define domain blocking; The pressure-based solver within FLUENT is used to account
(iv) Set spacing parameters; for the compressibility of the flow, which is assumed to be steady
(v) Generate mesh; and viscous. The one-equation Spalart-Allmaras model is selected
(vi) Export the mesh to a “.msh” file. to simulate the turbulence regime. The fluid is assumed to be air
modeled by the ideal gas law, while the dynamic viscosity is gov-
In the present study, the O-O mesh topology is adopted. The erned by the three-coefficient Sutherland law.
initial fluid domain, as depicted in Fig. 2, is first divided to yield The FLUENT journal accepts as input the reference wing length,
a block around the wing geometry which is later deleted since the reference area S re f , the angle of attack (AoA) α and the flow

Fig. 4. An example of the generated O-O topology mesh.

770 A. Benaouali, S. Kachel / Aerospace Science and Technology 92 (2019) 766–776

Fig. 5. An example of the generated wingbox FEM mesh.

conditions, namely the Mach number M, the far-field temperature materials as well as the 1D element properties for spar caps and
and the far-field pressure. The temperature, pressure, density and stringers and 2D element properties for shell surfaces. The spar cap
speed of sound are calculated as a function of the altitude H , us- sections are of “L” shape, while the stringer sections are of “HAT”
ing an implemented program based on the standard atmospheric type.
model (US atmosphere 1976) [21]. Next, to prepare for the structural mesh generation, mesh seeds
The “Coupled” scheme is set for the pressure-velocity coupling are created along the wing surface edges. The seeds allow to affine
and the Green-Gauss Node-based method is used for the gradi- the mesh at desirable areas, and ensure that each two shared
ent information. The second-Order Upwind interpolation scheme is edges have the same number of nodes at the same locations.
selected for the spacial discretization of all variables, namely pres- Defining the mesh seeds is performed using loops that go through
sure, density, momentum, modified turbulent viscosity and energy. all the surface edges.
The pseudo transient under relaxation method available in FLUENT The skin, spars and ribs, being thin walled structures, are
is used in order to avoid numerical oscillations and achieve a sta- meshed into CQUAD4 shell elements, while the spar caps and
ble solution. stringers, distributed on the upper and lower wing box skin, are
To make sure that the flow solver achieves convergence at a modeled as beam elements. Node equivalence, which represents
minimal computational time, the simulation of the different de- the elimination of duplicate nodes, is then applied at a certain tol-
sign alternatives are initialized using a fully converged solution erance to yield an entirely enclosed mesh that does not involve
obtained beforehand for a chosen configuration. The initialization any free edges. Fig. 5 shows an example of the wingbox FEM mesh
is therefore performed by importing the data file (.dat) of this cho- generated using the design framework with a close look on the
sen solution before starting the iterating process. This strategy is
stringers.
applicable if the mesh arrangement and the number of cells are
Since the loads calculated by the aerodynamic module are given
held constant regardless of the geometrical changes. The struc-
on the nodes of the CFD mesh, they have to be transfered to the
tured mesh used for the present design framework fulfills these
nodes of CSM mesh before they can be applied. This process is per-
requirements. Accordingly, the optimization process is significantly
formed by the load transfer module (Section 2.6) which requires
accelerated.
the CSM skin mesh as input. For that purpose, a neutral PATRAN
The outputs of this module are the lift and drag coefficients as
file (.out) containing the node coordinates and element connec-
well as the pressure distribution over the wing surface. The CFD
tivities of the wing skin is exported. Once the transfer process is
model is validated by the present authors in a previous paper [22]
completed, the loads are written to a another neutral file, which is
considering the ONERA M6 wing geometry [23]. The comparison
imported by PATRAN to apply the nodal force distribution to the
between the obtained pressure coefficient c p distribution and the
experimental results available in the literature prove the accuracy corresponding skin mesh nodes.
of the developed CFD model. Therefore, it is qualified to be used At this point, the structural model is ready for the sizing op-
for the evaluation of the wing aerodynamic performance and to be timization, which will be described in details in the next section.
integrated in the optimization procedure. For that purpose, the PCL script requires more parameters in or-
der to formulate the optimization problem. These parameters are
2.4. Structural modeling the optimization constraint values as well as the upper and lower
variable bounds. A design study which defines the optimization
Once the aerodynamic pressure distribution over the wing sur- objective, constraints and design variables is created before the
face is calculated, the structural modeling is started. The com- analysis job is submitted for calculation. The job translation yields
mercial software MSC.PATRAN 2012.2 is used as the preprocessing a NASTRAN input file (.bdf) ready for running the solution process.
tool for structural analysis. To meet the automation requirements,
scripting is performed by executing a session file (.ses) using the 2.5. Structural analysis
Patran Command Language (PCL). The session file is executed in
PATRAN’s batch mode by the C# application through the following
system command: The structural sizing of the wing is performed using the FE
solver MSC.NASTRAN 2012.2 through its design optimization mod-
<Patran_path> -b -sfp <session_file_path> ule SOL 200. The process is automatically executed from the PCL
script. Once the analysis is successfully completed, a result file
The PATRAN script takes as input the material properties, the with a (.f06) extension is exported. From this file, the C# appli-
number of stringers, the dimensions of stringer and spar cap sec- cation extracts the optimal values of the objective function and
tions, and the thicknesses of wing surfaces. It starts by creating the design variables.
 A. Benaouali, S. Kachel / Aerospace Science and Technology 92 (2019) 766–776 771

Fig. 6. Flowchart of the optimization strategy adopted.

The aim of the structural sizing is to determine the wing layout and inversion, the FSI process is executed using the computing and
and sizes that minimize its weight while preventing the structure programming environment MATLAB.
from failure. In this study, the stress criterion is considered as the
optimization constraint, while the thicknesses of shell elements are 3. Optimization strategy
taken as the design variables. The stresses are limited to be lower
than the material allowable stress σallow given by: When dealing with a simulation-driven-optimization, the direct
approach is to incorporate the simulation directly in the search
σ yield process in order to provide the optimizer with the model response.
σallow = (1)
FS The cost of the optimization task depends on the number of evalu-
ations of the objective function, and the average simulation time. A
where σ yield is the yield Von-Mises stress and F S is the safety
single run of the developed design framework can be costly due to
factor. The thicknesses are assumed to vary continuously within
the high fidelity models used, particularly the RANS simulations, as
the specified range although in the real world restrictions related
well as the sizing process performed within the structural analy-
to cost and manufacturing may force discrete values. The surfaces
sis module. Therefore, the use of derivative-free algorithms, which
subjected to sizing are subdivided into multiple design zones each
require a population size of 10-100 with hundreds or thousands
of which consists of a number of finite elements that all have the
of iterations, is not practical [30]. On the other hand, the use of
same thickness.
derivative-based algorithms, which require the differentiation of
To solve the sizing optimization problem, an efficient technique
governing equations to implement analytical sensitivity methods,
known as the Fully Stressed Design (FSD) is employed. This method
is hampered by the incapacity to access the simulation codes of
allows to implement structural design optimizations without the
commercial software.
requirement to excessive sensitivity calculations or sophisticated
The common approach to overcome the aforementioned limi-
optimization algorithms [24]. One more advantage of the FSD that
tations is the Surrogate-based optimization (SBO), for which the
makes it suitable for structural sizing is that it can handle literally
high-cost simulation model is replaced by a fast, cheap, continu-
thousands of independent design variables [25]. It resizes the de-
ous and reasonably accurate analytical model, called the surrogate
sign variables so that each variable is at the limit value for-which
model [31,32]. The SBO process adopted in this study (Fig. 6) starts
the constraints are activated. The implementation of FSD within
by carrying out a number of high fidelity computer simulations
MSC.Nastran is embedded within the SOL 200 Design Optimization
on a set of sample points selected from the whole design space
module by setting a positive value to the maximum number of FSD
following a design of experiment (DoE) strategy. Then, the func-
cycles (FSDMAX).
tional relation between the input variables and simulation results
is extracted in a form of a response surface. In order to ensure
2.6. Load transfer module
whether the true optimum has been reached or not, the surro-
gate model accuracy has to be inspected. The simulation code is
For the efficiency, convenience and resolution requirements of run using the predicted optimal design  xopt to compute the true
their own calculations, the structural and aerodynamic domains function value and compare it to its approximated value. If the
are discretized in different ways. Therefore, the resulting CFD and two values are close enough, the optimization process can be ter-
CSM meshes do not coincide along the wet boundary of the struc- minated; if not, the design  xopt is taken as an additional sample
ture [26], called the Fluid–Structure Interface (FSI). Since the flight point to update the surrogate model with a better approximation
loads are calculated by the CFD solver at the nodes of the aerody- near the optimum. The process is repeated until the true opti-
namic mesh, they must be interpolated to those of the structural mum is achieved. Accordingly, the total number of high fidelity
mesh. In this study, an interpolation scheme based on radial bass simulations, which represent the evaluations of the cost and con-
functions (RBFs) [27–29] is used. Accordingly, the forces at the CSM straint functions, is equal to the number of sampling points plus
nodes fs are computed from those at the CFD nodes fa as follows: the number of database updates. The methods chosen for the DoE
strategy and the surrogate modeling technique are, respectively,
fs = [ H ] T fa (2) the improved Latin Hypercube Sampling (IHS) and the RBF. These
methods have been implemented in MATLAB and linked to the de-
where [ H ] is called the coupling matrix. Using this technique, the sign framework to build an efficient MDO environment.
need for connectivity information is totally removed as the grids
are considered as an arbitrary set of point clouds, which makes 4. MDO of an A320 class aircraft
it compatible with any mesh type without requirements. More-
over, the method is based mainly on matrix multiplication and In this section, the use of the developed design framework for
does not require an excessive computational cost. Because it in- the efficient solution of high fidelity MDO problems is investigated
volves large size matrix operation such as multiplication, transpose through a simple case study of a high-speed aircraft wing.
772 A. Benaouali, S. Kachel / Aerospace Science and Technology 92 (2019) 766–776

4.1. Case specifications Table 1

Reference aircraft specifications.
The reference aircraft is a short-to-medium range high-speed Parameter Value
commercial passenger aircraft in the Airbus A320 class, assumed to Maximum take-off weight MTOW 73500 kg
be flying with the A320 typical mission conditions, i.e. at a Mach Fixed weight fraction W fixed / W 0 0.35
number M = 0.78 and an altitude H = 11000 km. In order to per- Crew + Payload weight fraction 0.24
Wing reference area S re f 122.4 m2
form an optimization that considers simultaneously both structural Initial cruise altitude H 11000 m
and aerodynamic performances, the objective function must com- Cruise Mach number M 0.78
bine the wing aerodynamic and structural properties. In this study, Cruise Lift coefficient C L 0.58
Thrust specific fuel consumption (c T ) 15 g/(kNs)
the aircraft range, commonly used in multidisciplinary optimiza-
Additional drag coefficient (C D ,add ) 0.005
tion [33,34], is chosen. It is the maximum distance an aircraft can
fly, given by the Breguet equation [35]:
W c1
  = g ( W wing ) (9)
V L W c1 W c2
R= ln (3)
g cT D W c2 Accordingly, the aircraft range evaluation reduces to the evaluation
where R is the range, V is the flight speed, c T is the thrust-specific of the wing weight and the lift-to-drag ratio:
fuel consumption, L / D the lift-to-drag ratio, g is the acceleration  
L
of gravity and W c1 , W c2 are the initial and final cruise weights, R= f , W wing (10)
respectively. This formulation is resulted from a general equation D
by assuming that c T , V and L / D are constant [36]. Equation (3) and therefore the range maximization yields a trade-off between
describes the three main disciplines of aircraft physics as c T rep- L
W wing minimization and maximization (L/D corresponding to
resents the propulsive efficiency, L / D represents the aerodynamic D
performance while W c1 / W c2 represents the structural optimiza- cruise C L coefficient).
tion term. The maximization of the range for a given Mach number Table 1 lists the key parameters of the reference aircraft config-
M and a specific fuel consumption leads to the simultaneous max- uration used for the optimization. Since there are multiple variants
imization of L / D and W c1 / W c2 . of the Airbus A320, the parameter values presented in Table 1 for
According to [37], the maximum takeoff weight (MTOW) is the MTOW, S re f , altitude H , c T and M are averaged from multiple
sum of crew weight W crew , payload weight W payload , fuel weight references, such as [38,39]. The additional drag coefficient C D ,add ,
W fuel and empty weight W empty : which represents the sum of the drag factor of all aircraft com-
ponents except the wing, is added to the wing drag coefficient
M T O W = W crew + W payload + 1.06W fuel + W empty (4) C D to compute the lift-to-drag ratio. Here, the lift is assumed to
be generated only by the wing. For the additional drag coefficient
For an aircraft mission composed of five segments: takeoff, climb, as well as the weight fractions, typical values have been assumed
cruise, descent for landing and taxi, the aircraft weight at the end for demonstration purposes. As mentioned before, the application
of the total mission W 5 is expressed as follows: of the Breguet equation assumes constant L / D and V during the
  cruise segment, and therefore the values of V and C L are obtained
W1 W2 W3 W4 W5
W5 = W0 (5) by the flight conditions at the start of the cruise segment.
W0 W1 W2 W3 W4
where W 0 = MTOW and W i is the aircraft weight after the mis- 4.2. Problem formulation
sion segment i. The segment weight fractions were estimated his-
torically by [37]. The goal of this optimization is to maximize the aircraft range R
The cruise weight fraction W 3 / W 2 , same as the inverse of by controlling the wing geometric parameters, including the plan-
W c1 / W c2 in Equation (3), can be determined from: form shape parameters P and the airfoil shape parameters A, the
W5 wingbox structural thicknesses T and the angle of attack α . The
constraints are related to the cruise lift coefficient C L , the refer-
W c2 W3 W0
= =  (6) ence wing area S re f and the structural failure F . The optimization
W c1 W2 W1 W2 W4 W5
problem is formulated as follows:
W0 W1 W3 W4
maximize R ([P, A, T, α ])
Assuming that the wight loss is due to fuel burning only, the air-
craft weight at the end of the total mission W 5 is given by: subject to C L − C L ,0 = 0,

W5 W fuel W 0 − W crew − W payload − W empty S re f − S re f ,0 = 0, (11)

=1− =1− (7)
W0 W0 1.06W 0 F ≤0
The empty weight is obtained by summing the weight of the two xl ≤ x ≤ xh
wings and the aircraft structural weight without the wings, i.e.
W empty = 2W wing + W fixed , thus Equations (6) and (7) yield: The selected planform design parameters are the aspect ratio
  AR, the taper ratio λ and the leading edge sweep angle  L E . The
W1 W2 W4 W5 root chord ratio γc and the kink position ratio γs are assumed
1.06
W c1 W0 W1 W3 W4 constant and equal to their respective baseline values. They are
= (8) used along with the wing reference area, which is constrained to
W c2 W crew W payload W fixed W wing
0.06 + + + +2 its baseline value S re f ,0 , to evaluate the wing semi span b/2 and
W0 W0 W0 W0
the wing root chord C r . The CST method is adopted for the pa-
Given the values of W crew / W 0 , W payload / W 0 and W fixed / W 0 , rameterization of the airfoil shape, which is assumed to be the
which can be estimated from historical trends, the only unknown same along the whole span. For that, a 4th order Bernstein poly-
left in Equation (8) is W wing / W 0 : nomial has been selected for the upper and lower curves resulting
 A. Benaouali, S. Kachel / Aerospace Science and Technology 92 (2019) 766–776 773

Table 2 Table 3
Baseline wing geometry parameters. Structural properties of the aluminum alloy 2024-T351.

Parameter Value Unit Property Value

Aspect ratio AR 9.40 - Young’s Modulus E (MPa) 73.1 E+3
Taper ratio λ 0.3 Material density ρ (Kg/m3 ) 2.78 E+3
Leading edge sweep angle  L E 27.5 deg Yield stress σY ield (MPa) 324
Semi span b/2 16.96 m Ultimate stress σUltimate (MPa) 469
Root chord C r 7.01 m
Trapezoidal chord ratio γc 0.71 -
Kink position ratio γs 0.37 -

in 10 CST parameters. For simplicity reasons and due the limita-

tion of the available computational capability, only the first and
last coefficients for each curve, namely A u0 , A u4 , Al0 , Al4 , are in-
cluded as design variables. To account for the airfoil thickness
relatively to its chord length, a parameter which scales the airfoil
point y-coordinates, denoted by t c , is added. Therefore the shape
parameters are P = [AR, λ,  L E ] and A = [ A u0 , A u4 , Al0 , Al4 , t c ]. In
addition, the design variables include the vector T representing the
thicknesses of the wingbox shell elements, and the angle-of-attack
α which offers an additional degree of freedom to fulfill the lift
constraint.
Since the reference wing area as well as the aircraft weights
and flight conditions are kept constant during the optimization, the
lift must remain the same as prescribed in Table 1. The constraint
F ≤ 0 describes that a condition must be satisfied so that the wing
structure does not fail under extremal flight loads. This condition
is translated to a limitation of the von-Mises stress to be less than Fig. 7. Design zones for the wingbox structural sizing.
the allowable stress.
A fixed layout is considered for the structural modeling of the
4.3. Baseline wing geometry candidate wing designs during the optimization. The wingbox con-
sists of two spars: a front spar and a rear spar. The front spar of
The geometric dimensions of the reference aircraft wing were the reference aircraft wing is assumed straight at 15% of the local
extracted based on the general dimension schemes provided in the chord from the leading edge, while the rear spar is supposed to
AIRBUS A320 manual [40], as well as on values used in references extend at 75% of the local chord following the trailing edge shape
[39,41,42]. The wing dimensions used for the baseline configura- (having a kink). On the other hand, the wing comprises 27 ribs
tion are given in Table 2. [43] including root and tip ribs, assumed to be equidistantly dis-
Since the A320 wing sectional profiles are not available in the tributed along the span and parallel to the root section. The finite
literature, the widely known airfoil RAE2822 is assumed for the element model also considers 4 spar caps, attached to the edges of
whole span of the baseline wing. The values of the CST parame- each spar, and 16 stringers for each of the upper and lower wing-
ters defining the RAE2822 airfoil are obtained by carrying out a box skin surfaces.
numerical optimization for which the objective is to minimize the In this case study, the structural sizing is applied exclusively to
difference between the CST-generated airfoil and the target airfoil. the wingbox skin and spars. The thicknesses of the wing ribs as
The optimization gives as results the following CST parameters: well as those of the skin leading edge and trailing edge parts are
assumed constant, equal to 5 mm, 10 mm and 5 mm, respectively,
{ 0.1277 , 0.1399 , 0.1903 , 0.1927 , 0.2019 , −0.1299 , in order to represent the weight of high lift devices. Similarly,
the spar caps and stringers are considered with constant section
− 0.1345 , −0.2247 , −0.0783 , 0.0412} dimensions. All structural components of the wing are supposed
made of a 2024 − T 351 aluminum alloy with the metrial proper-
4.4. Simulation models ties listed in Table 3. A single manoeuver load case with a critical
load factor n = 2.5 is taken into account for simplicity. Due to the
The CFD computational domain extends to 15 spans upstream limited computational resources available, the sizing loads are de-
from the leading edge of the root airfoil, 15 spans downstream rived by scaling the aerodynamic cruise loads using the load factor
from the trailing edge of the tip airfoil, and 5 chords from the tip value 2.5.
section in the spanwise direction. For all meshes generated during The design zones of the sizing optimization are obtained by
the optimization procedure, a constant number of cells of 4 million subdividing the wing box skin and spars at the rib locations, as
is considered. shown in Fig. 7. Accordingly, there are (27 − 1) × 4 = 104 zones,
To make sure that the flow solver achieves convergence at a corresponding also to the number of design variables. For the
minimal computational time, the convergence criterion is based stress constraint, a safety factor F S = 1.5 is used.
on monitoring the lift and drag coefficients. Moreover, the flow
calculations of the optimization design alternatives are initialized 4.5. Optimization results
using a fully converged solution obtained for the baseline configu-
ration. This strategy is applicable if the mesh arrangement and the Because the design framework includes a structural sizing mod-
number of cells are held constant regardless of the geometrical ule, the outputs of each execution are the lift-to-drag ratio along
changes. The structured mesh used for the present design frame- with the weight of a sized wing with a safe stress levels. There-
work already fulfills these requirements. fore, the structural thicknesses T are iteratively considered during
774 A. Benaouali, S. Kachel / Aerospace Science and Technology 92 (2019) 766–776

Fig. 8. Geometry comparison between the baseline and optimized designs.

Table 4 Table 5
Design variable and their ranges. Comparison of the baseline and optimized wing configurations.

Design variable Lower bound Upper bound Baseline Optimized

AR 8 12 Aspect ratio AR 9.4 8.56
λ 0.2 0.4 Taper ratio λ 0.3 0.28
First set
L E 20 40 Sweep angle L E 27.5 33.37
α 2 4 Angle of Attack α 2.73 3.10
A u0 0.1277 0.1248
A u0 0.0777 0.1777
A u4 0.2019 0.2519
A u4 0.1519 0.2519 CST parameters
Al0 -0.1299 -0.1232
Second set Al0 -0.1799 -0.0799
Al4 0.0412 0.0193
Al4 -0.0088 0.0912
Thickness ratio tc 0.121 0.1166
tc 0.08 0.14
Lift coefficient CL 0.58 0.58 (+0.0%)
Range R 6311.8 Km 6870.8 Km (+8.86%)

the optimization, which reduces the design variables to the vector

[P, A, α ], and eliminates the structural failure constraint F ≤ 0.
drag and therefore the improvement of the lift-to-drag ratio. This
The SBO is implemented using the genetic optimization algo-
can be explained by the higher sweep angle and the lower taper
rithm (GA) through its MATLAB implementation. A number of 140
ratio with comparison to the baseline configuration.
samples is set for the DoE, and four (4) database updates were
The results of the structural sizing for the baseline and opti-
required to achieve the optimization convergence. The global com-
mized wings are depicted in Fig. 10 through the optimal element
putational cost to evaluate the performance of these samples is
thicknesses of the upper and lower wingbox skin. For both wings,
approximately 3 days using a 3.4 GHz i7 processor. The design
the skin thicknesses of the lower surfaces are slightly larger than
variables and their ranges are given in Table 4. The upper and
those of the upper surface. The thicknesses increase from the wing
lower bound of the CST parameters are obtained by perturbing the
tip to the kink, whereas they decrease from the kink to the wing
original values by ±0.05.
A comparison of the baseline and optimal designs, showing root. This is because the areas near the wing kink experience the
the values of design variables and performance metrics, is given highest bending moments and shear stresses which are probably
in Table 5. The planform and the airfoil shapes are compared in due to the sudden change in the trailing edge shape (kink break).
Figs. 8(a) and 8(b), respectively. The results show that the aspect The optimal thicknesses obtained for the optimized wing are lower
ratio and the taper ratio are decreased, while the sweep angle and than those of the baseline wing, and thus the wing weight is re-
the AoA are increased. Reducing the aspect ratio while constrain- duced. This is an effect of lower stress levels resulting from the
ing the reference area leads to an increase of the root chord length. decrease in the aspect ratio.
The optimal airfoil differs from the baseline one mainly in the rela- The improvement in each of the structural and aerodynamic
tive thickness being decreased, and in the trailing edge shape being performance did not reach its maximum due to the existing com-
thickened ( A u4 reached its maximum value). On the other hand, promise. The optimization objective to maximize the aircraft range
the leading edge is practically the same, which indicates that it is realized by the decreasing the wing weight and increasing the
is already optimal for the RAE2822 airfoil. The changes in the air- lift-to-drag ratio as much as possible simultaneously. The wing as-
foil shape do not seem to be important, yet their contribution in pect ratio, for example, is preferred to be maximal to improve the
the performance is considerable. The value of the lift coefficient lift-to-drag ratio, however the aim to minimize the wing weight
is maintained for the optimized wing, and thus the optimization forces its reduction.
constraint is respected. The objective function, that is the range, is
improved by 8.86%. 5. Conclusions
Fig. 9 shows the contours of the pressure coefficient C p over
the upper and lower surfaces of the baseline and optimized wings. The developed framework is a powerful and generic tool that
As one can notice by comparing the contours of the upper surfaces, can be used to solve a wide range of wing MDO problems. The
the shock wave intensity has been alleviated, especially in the area proposed approach highlights the outcome from using high fidelity
near the wing root. This reduction explains the decrease of the methods in the design optimization instead of low fidelity meth-
 A. Benaouali, S. Kachel / Aerospace Science and Technology 92 (2019) 766–776 775

Fig. 9. Comparison of C p contours for the baseline and optimized wings.

Fig. 10. Comparison of the wingbox skin thickness for the baseline and optimized wings.

ods. Indeed, capturing complex behavior, such as the wave drag take into account more optimization constraints, such as the buck-
in transonic flow conditions and the load-dependent structural ling failure modes.
weight, is demonstrated to have an impact on the performance
of a wing and therefore on its optimal design. In addition, it has Declaration of Competing Interest
been proved that the global performance of an aircraft, such as the
range, can be significantly improved based only on changes applied There is no competing interest.
to the geometric and structural configuration of its wing.
In the future work, the authors are planning to consider the References
wing structure flexibility which allows the implementation of
[1] International Air Transport Association, Air passenger market analysis, avail-
an aerostructural optimization. It can be realized by integrating
able on http://www.iata.org/publications/economics/pages/index.aspx, Decem-
a fluid-structure coupling loop, also called the Multidisciplinary ber 2016. (Accessed 11 November 2017), All Rights Reserved.
Analysis (MDA) loop, which seeks the achievement of static aeroe- [2] International Air Transport Association, Air passenger market analysis, avail-
lastic equilibrium at each iteration. Moreover, the design frame- able on http://www.iata.org/publications/economics/pages/index.aspx, Decem-
ber 2017. (Accessed 20 March 2018), All Rights Reserved.
work can be extended to consider composite materials, which have [3] Panos Y. Papalambros, Douglass J. Wilde, Principles of Optimal Design: Model-
very interesting weight and strength characteristics, as well as to ing and Computation, Cambridge university press, 2000.
776 A. Benaouali, S. Kachel / Aerospace Science and Technology 92 (2019) 766–776

[4] Shigeru Obayashi, Multidisciplinary design optimization of aircraft wing plan- ment, Report of the Fluid Dynamics Panel Working Group 04, AGARD AR 138,
form based on evolutionary algorithms, in: Proceedings of the IEEE Interna- May 1979.
tional Conference on Systems, Man, and Cybernetics, San Diego, CA, USA, vol. 4, [24] Erwin H. Johnson, Fully stressed design in MSC. Nastran, in: MSC 3rd World-
October 1998, pp. 3148–3153. wide Aerospace Users Conference and Technology Showcase, Toulouse, France,
[5] James Reuther, Juan Alonso, Joaquim R.R.A. Martins, Stephen Smith, A coupled April 2002, Paper No. 2001-42.
aero-structural optimization method for complete aircraft configurations, in: [25] MSC Nastran, Design Sensitivity and Optimization User’s Guide, MSC. Software
Proceedings of the 37th Aerospace Sciences Meeting and Exhibit, Reno, NV, Corp, 2012.
USA, January 1999, pp. 187–217. [26] Charbel Farhat, Michael Lesoinne, P. Le Tallec, Load and motion transfer al-
[6] Ilan Kroo, Steve Altus, Robert Braun, Peter Gage, Ian Sobieski, Multidisciplinary gorithms for fluid/structure interaction problems with non-matching discrete
optimization methods for aircraft preliminary design, in: Proceedings of the interfaces: momentum and energy conservation, optimal discretization and ap-
AIAA 5th Symposium on Multidisciplinary Analysis and Optimization, Panama plication to aeroelasticity, Comput. Methods Appl. Mech. Eng. 157 (1–2) (1998)
City Beach, FL, USA, September 1994, pp. 697–707. 95–114.
[7] Jaroslaw Sobieszczanski-Sobieski, Raphael T. Haftka, Multidisciplinary [27] Marilyn J. Smith, Carlos E.S. Cesnik, Dewey H. Hodges, Evaluation of some data
aerospace design optimization: survey of recent developments, Struct. Optim. transfer algorithms for noncontiguous meshes, J. Aerosp. Eng. 13 (2) (2000)
14 (1) (1997) 1–23. 52–58.
[8] Jaroslaw Sobieszczanski-Sobieski, Alan Morris, Michel van Tooren, Multidisci- [28] Armin Beckert, Holger Wendland, Multivariate interpolation for fluid-structure-
plinary Design Optimization Supported by Knowledge Based Engineering, John interaction problems using radial basis functions, Aerosp. Sci. Technol. 5 (2)
Wiley & Sons, 2015. (2001) 125–134.
[9] C. Gross David, Report from the fidelity implementation study groups, in: Pro- [29] T.C.S. Rendall, C.B. Allen, Unified fluid–structure interpolation and mesh mo-
ceedings of 1999 Spring Simulation Interoperability Workshop, Orlando, FL, tion using radial basis functions, Int. J. Numer. Methods Eng. 74 (10) (2008)
USA, March 1999, pp. 231–240. 1519–1559.
[10] Jacek Mieloszyk, Tomasz Goetzendorf-Grabowski, Introduction of full flight dy- [30] Slawomir Koziel, Leifur Leifsson, Simulation-Driven Design by Knowledge-
namic stability constraints in aircraft multidisciplinary optimization, Aerosp. Based Response Correction Techniques, Springer, 2016.
Sci. Technol. 68 (2017) 252–260. [31] Nestor V. Queipo, Raphael T. Haftka, Wei Shyy, Tushar Goel, Rajkumar
[11] Stephen A. Andrews, Ruben E. Perez, Comparison of box-wing and conventional Vaidyanathan, P. Kevin Tucker, Surrogate-based analysis and optimization, Prog.
aircraft mission performance using multidisciplinary analysis and optimization, Aerosp. Sci. 41 (1) (2005) 1–28.
Aerosp. Sci. Technol. 79 (2018) 336–351.
[12] Odeh Dababneh, Timoleon Kipouros, Influence of high fidelity structural mod- [32] Alexander Forrester, Andy Keane, et al., Engineering Design Via Surrogate Mod-
els on the predicted mass of aircraft wing using design optimization, Aerosp. elling: a Practical Guide, John Wiley & Sons, 2008.
Sci. Technol. 79 (2018) 164–173. [33] Gaetan W. Kenway, Graeme J. Kennedy, Joaquim R.R.A. Martins, Aerostructural
[13] Joaquim R.R.A. Martins, Andrew B. Lambe, Multidisciplinary design optimiza- optimization of the common research model configuration, in: Proceedings of
tion: a survey of architectures, AIAA J. 51 (9) (2013) 2049–2075. the 15th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference,
[14] Nenad Vidanović, Boško Rašuo, Gordana Kastratović, Stevan Maksimović, Dušan Atlanta, GA, USA, June 2014, AIAA paper 2014-3274.
Ćurčić, Marija Samardžić, Aerodynamic–structural missile fin optimization, [34] Tobias F. Wunderlich, Multidisciplinary wing optimization of commercial air-
Aerosp. Sci. Technol. 65 (2017) 26–45. craft with consideration of static aeroelasticity, CEAS Aeronaut. J. 6 (3) (2015)
[15] Odeh Dababneh, Timoleon Kipouros, A review of aircraft wing mass estimation 407–427.
methods, Aerosp. Sci. Technol. 72 (2018) 256–266. [35] Louis Bréguet, Calcul du poids de combustible consommé par un avion en vol
[16] Tomasz Goetzendorf-Grabowski, Multi-disciplinary optimization in aeronauti- ascendant, Gauthier-Villars, 1920.
cal engineering, Proc. Inst. Mech. Eng., Part G, J. Aerosp. Eng. 231 (12) (2017) [36] Colonel Kip P. Nygren, Major Robert R. Schulz, Breguet’s formulas for aircraft
2305–2313. range & endurance an application of integral calculus, in: SEE Annual Confer-
[17] OpenVSP – NASA open source parametric geometry, http://www.openvsp.org. ence Proceedings, Washington D., WA, USA, June 1996, pp. 1901–1905.
(Accessed 17 February 2019). [37] D. Raymer, Aircraft Design: A Conceptual Approach, fourth edition, AIAA Edu-
[18] Common parametric aircraft configuration schema (CPACS), https://software.dlr. cation Series, AIAA, 2006.
de/p/cpacs/home/. (Accessed 17 February 2019). [38] Élodie Roux, Avions civils à réaction: Plan 3 vues et données caractéristiques,
[19] Abdelkader Benaouali, Wing design: commercial software scripts, https:// Elodie Roux, 2007.
github.com/benaoualia/WingDesign_CommercialSoftwareScripts, 2019. [39] Ed Obert, Aerodynamic Design of Transport Aircraft, IOS press, 2009.
[20] Brenda M. Kulfan, Universal parametric geometry representation method, J. [40] Airbus, SAS. A320 Aircraft Characteristics Airport and Maintenance Planning,
Aircr. 45 (1) (2008) 142–158. 2005.
[21] NOAA, NASA, US Air Force, U.S. Standard Atmosphere, 1976, NASA report No. [41] Maria Pester, Multi-disciplinary Conceptual Aircraft Design Using Ceasiom,
TMX–74335, NOAA-S/T-76-1562, October 1976. Master’s thesis, Hamburg university of applied sciences, 2010.
[22] Abdelkader Benaouali, Stanisław Kachel, A surrogate-based integrated frame- [42] Jan E.K. Hoogervorst, Ali Elham, Wing aerostructural optimization using the in-
work for the aerodynamic design optimization of a subsonic wing planform dividual discipline feasible architecture, Aerosp. Sci. Technol. 65 (2017) 90–99.
shape, Proc. Inst. Mech. Eng., Part G, J. Aerosp. Eng. 232 (5) (2018) 872–883. [43] Jean-Paul Pourtau, Sharklet entry into service, in: A320 Femily Regional Semi-
[23] V. Schmitt, F. Charpin, Pressure Distributions on the Onera-m6-Wing at Tran- nar, Manila, Philippines, 2012.
sonic Mach Numbers. Experimental Data Base for Computer Program Assess-