Framework development for robust design of novel

aircraft concept
Marco Saporito, Andrea da Ronch, Peter Schmollgruber, Nathalie Bartoli

To cite this version:

Marco Saporito, Andrea da Ronch, Peter Schmollgruber, Nathalie Bartoli. Framework development
for robust design of novel aircraft concept. 3AF Aerospace Europe Conference 2020, Feb 2020, BOR-
DEAUX, France. ฀hal-02904365฀

HAL Id: hal-02904365

https://hal.archives-ouvertes.fr/hal-02904365
Submitted on 22 Jul 2020

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 FRAMEWORK DEVELOPMENT FOR ROBUST DESIGN OF NOVEL AIRCRAFT
CONCEPTS

Marco Saporito (1), Andrea Da Ronch (1), Peter Schmollgruber(2), Nathalie Bartoli(2)

(1) University
of Southampton, Southampton, S017 1BJ, United Kingdom.
Email: [email protected], [email protected]

(2) ONERA/DTIS, Université de Toulouse, F-31055 Toulouse, France.

Email: [email protected], [email protected]

KEYWORDS: Aircraft design, Multidisciplinary issues and affect stability, control strategies and
Analysis and Optimization, Aerodynamics. pilot coupling [7]. Therefore, the exploration of
disruptive concepts needs to be accompanied by
ABSTRACT: analysis and optimization frameworks as
multidisciplinary as possible, even at conceptual or
One of the challenges in Multidisciplinary Design
preliminary design stages [8]. Failing to provide
Analysis and Optimization is raised by the fidelity of
adequate multidisciplinary capabilities early in the
the methods and tools used within each single
design may lead to severe consequences such as
discipline. This element, introducing uncertainty in
expensive corrections later in the process, or even
predictions, is neglected generally. In this work, a
to the failure of the whole process [7]. To report a
framework is built to analyse the impact that the
relevant case, after finding that divergent flutter may
fidelity of the aerodynamic model has on four key
occur under certain conditions in the Boeing B747-
figures of merit, including Maximum Take-off
8 and B747-8F, reparations and software updates
Weight, maximum Lift-to-drag ratio, fuel weight and
had to be applied [9], with costs for the company
take-off length. It is found that the aircraft
and the operators and some damage to the
configuration with minimum fuel weight computed
company’s image.
with two aerodynamic models differs significantly,
with large variations in terms of predicted fuel weight
and wing planform.

1. INTRODUCTION
As environmental requirements become more and
more urgent, reduction of emissions in commercial
aviation is targeted with increasing pressure both by
research and industrial [1Erreur ! Source du
renvoi introuvable., 12, 26]. A large effort is put on
the exploration of disruptive technologies and
configurations that may lead to a new generation of
highly efficient aircraft [2, 3].
Most innovation strategies arise at the three levels
of propulsion, structures and aerodynamics.
Technologies such as electric/hybrid propulsion and
distributed propulsion are receiving large attention
[4], as well as cutting-edge structural solutions
including composites-rich structures, bio-inspired
materials, morphing structures [22], foldable wings
[23,24], to just name a few. Figure 1: Comparison of conventional and high
From the aerodynamic side, the tendency is to aspect ratio concepts, from [6]:
promote efficient layouts such as blended wing-
body configurations [5], box-wing configurations The added knowledge arising from a
[14], and very high aspect ratio wings [6] (see Fig. multidisciplinary approach has motivated a large
1). As usual in aerospace design, the effort in the research and industrial environments
implementation of such choices involves important towards the implementation of integrated tools for
implications on several disciplines. For instance, aerospace vehicles design. With the support of
high aspect ratio wings accentuate aeroelastic

1
increasingly powerful computers, multidisciplinary framework, discussing its capabilities, limits and
analysis and optimization (MDAO) for aircraft faced challenges. This is done by employing the
design has known significant progress. Together framework for design exploration and optimization
with the numerous achievements, though, several and comparing the effects of different fidelities of the
issues still remain challenging and stimulate aerodynamic models. To this purpose, two
ongoing efforts in the research community. consecutive objectives were established.
For instance, as the aircraft design involves several The first objective is to compare the effect of two
disciplines, each of which requires different tools different aerodynamic models on the overall results
and handling a large body of information, the of a conceptual multidisciplinary aircraft sizing
exchange of data between modules represents a process. To this end, a set of design candidates was
crucial question [15]. The definition of common evaluated with both the models and some relevant
programming languages and data storage formats performance indices were compared. The
becomes a key enabler for the effectiveness of the dispersion of the results was used to assess the
whole framework [16]. robustness of the sizing tool predictions with respect
Another critical point is the choice of an appropriate to the perturbation of the most relevant
model and numerical strategies for each of the aerodynamic design variables.
involved disciplines. The design team, as well as the The second objective is to perform an optimization
single specialist, have at their disposal a broad set study on a baseline configuration and compare the
of analysis methods of different fidelities. A proper optimal designs obtained for each of the two
design framework should be implemented in such a aerodynamic fidelity levels available within the MDA
way that the appropriate fidelity level is employed framework.
for each maturity status of the design process. The MDA framework and the aerodynamic models
Often, effort is put on introducing as much physics are presented in Sections 2 end 3. The adopted
as possible since the earliest stages by some clever methodology is described in Section 4, whereas the
multifidelity approaches. The strategy is to exploit selected test case is outlined in Section 5. Results
information from a few expensive higher-fidelity are presented and discussed in Section 6. Finally,
analyses to correct and enrich the results of the conclusions are given in Section 7.
lower-fidelity tool. This can be done either by directly
merging the two solvers in a new hybrid tool [17-19],
or by deriving a surrogate model through data sets 2. AIRCRAFT SIZING TOOL
from multiple tools [20, 21]. The adoption of such
strategies and the choice of the appropriate fidelity
The adopted aircraft sizing tool, FAST (Fixed-wing
levels for different design purposes are widely
Aircraft Sizing Tool) was provided by ONERA and
discussed in [13]. When developing an MDAO
ISAE-SUPAERO. It is conceived as a quick and
framework, the assessment of the applicability of all
effective conceptual design tool for traditional tube
the different analysis tools and their impact in terms
and wing configurations. The user specifies a series
of accuracy of the results is a necessary task. In
of Top Level Requirements and the framework
fact, it is desirable to provide the highest possible
estimates the required fuel consumption through a
flexibility towards the exploration of wide design
series of sizing loops involving modular analyses for
spaces, and all the integrated tools should prove
the key disciplines, namely flight mechanics,
capable to handle large ranges of variables and
aerodynamics, structures, propulsion, weight and
physical conditions.
balance. The original approach is based on a point
The complexity of the MDAO tasks is even
mass approximation together with semi-empirical
increased by the fact that any fidelity level is
equations for performance and aerodynamic
inherently affected by uncertainties, the most
predictions. This allows high computational
relevant being those related with mass and balance,
efficiency and accuracy to be achieved as long as
and aerodynamic performances. Uncertainty
traditional concepts are treated.
quantification and management in MDAO
The propulsion module can be based either on a
applications is an interesting challenge, at the
dataset from the CeRAS project [27], or an
centre of several research activities [10, 25]. If
analytical model that provides thrust and fuel
properly addressed, it would improve the quality of
consumption as function of altitude and flight speed
the design outcomes providing key information on
[28].
the robustness and reliability of the results.
The performance module gathers all the information
The long term aim of the project here discussed, still
from the disciplinary modules and performs a time-
ongoing, is to provide robust design capabilities in
marching simulation of the full mission.
the context of MDAO for modern aircraft concepts,
Sizing and positioning of components are iteratively
able to comply with all the discussed requisites,
updated during the design loops through dedicated
reliable against wide design spaces and robust
geometry, weight and balance modules. Overall
against the major sources of uncertainty.
aircraft design rules from [32] are used to initially
The present work focuses on the actual,
locate the main components, such as wing, tail,
intermediate state of implementation of such a

2
landing gear, etc.
Such features suit the purpose of the present study, The compressibility term is estimated by a semi-
as the interest here is to investigate the reliability of empirical formula taking as inputs only Mach
simplified models against perturbation of some number and lift coefficient [32]. Although its validity
relevant design variables. In fact, it is always is not general, the correction is considered
desirable in the conceptual stage to provide robust acceptable as far as the Mach number does not
predictions against possible changes later in more exceed 0.8 [38].
advanced phases of the design process. Moreover, The remaining two contributions, namely induced
the results obtained with the original knowledge- and trim drag, are evaluated in this study with two
based code architecture are used in this work in different approaches.
comparison with those from an extended version The first one still employs simple analytical
enabling a more physics-based estimation of the functions and semi-empirical corrections. The
aerodynamic performances. induced drag is given by Eq. 4 [35], and the Oswald
The framework is fully implemented in Python and factor 𝑒 is estimated as in [36]. The contribution due
all the modules are consistently interfaced through to trim is computed by Eq. 5 as indicated in [32].
a class hierarchy, whereas all the inputs/outputs are
standardized in XML or CPACS [29] format. Such 𝐶 = (4)
well-structured organization is a key feature to
properly manage multidisciplinary frameworks, and
it allows modules and models to be updated, 𝐶 = 5.89 × 10 𝐶𝐿 (5)
extended and switched, and, in turn, the interface
within the research community is also facilitated. The second approach aims at introducing more
With this philosophy, the aerodynamic module was physics in the aerodynamic predictions, and makes
upgraded with an additional model based on the use of an in-house developed VLM solver, which is
Vortex Lattice Method (VLM) that provides a more discussed in the next section. Following the
flexible, physics-based alternative to the original categorization given in [13], the first approach will
semi-empirical formulation. be hereinafter referred to as level zero (L0),
As the focus of this work is mainly on the impact of whereas the latter as level one (L1).
the aerodynamic model on the overall design Finally, as the wing cross-section is fixed since the
outcomes, the following section is dedicated to the beginning of each analysis, and its two-dimensional
description of the two aerodynamic models characteristics are known, the maximum lift
embedded in FAST. Further details about the coefficient is estimated both for L0 and L1 by Eq. 6
framework can be found in [30]. [34, 37].

𝐶 =𝑘 𝐶 𝑐𝑜𝑠𝛬 (6)
3. AERODYNAMIC MODELS
The purpose of the aerodynamic module is to 3.1. Vortex Lattice Method
compute the drag polar, 𝐶 = 𝑓(𝐶 ), for different
flight conditions. The total drag coefficient is given This method is based on a simplified form of the fluid
by three contributions arising from viscous drag, dynamic equations, which allow efficient three-
𝐶 , compressibility, 𝐶 , induced drag, 𝐶 , and trim dimensional computations to be performed with
drag, 𝐶 . The total drag build-up is obtained as fairly accurate prediction in most subsonic flight
in Eq. 1 [32]. conditions, making the method particularly suitable
for conceptual design purposes. The assumptions
𝐶 =𝑘 𝑘 𝐶 +𝑘 𝐶 +𝑘 𝐶 +𝑘 𝐶 (1) are of inviscid, incompressible and irrotational flow.
Under these hypotheses, valid with good
where the coefficients 𝑘 , when available, account approximation in the whole flow field except within
for additional corrections due to particular the boundary layer, it can be demonstrated [39] that
technologies such as winglets. the continuity equation reduces to the Laplace’s
The viscous drag is obtained by summing and equation (Eq. 7), where the only boundary condition
normalizing the friction contributions of all the is that the normal flow component must be zero at
wetted areas as in Eq. 2, with the friction coefficients the solid boundary (Eq. 8). Here 𝛷 is the velocity
given by the Prandtl-Schlichting correlation [33] as potential, 𝑛 is the normal to the local solid surface
in Eq. 3. and the subscripts ( ) and ( ) denote the body
and the undisturbed flowfield, respectively.
𝐶 =∑ 𝑐 𝑘 ,
(2)
𝛻 𝛷=0 (7)
.
𝑐 = . (3) 𝛻 (𝛷 − 𝛷 ) ⋅ 𝒏 = 0 (8)
( . )

3
 which is a standard Vortex-Lattice model used to
The basic idea behind the method is to obtain the check the accuracy of vortex lattice codes. Its
solution to the Laplace’s equation (Eq. 7) by a parameters can be found in [41]. Since for the scope
distribution of elementary solutions on the problem of this work multi-wing configurations have to be
boundaries (body surface and wake). In fact, modelled, this test case was repeated with two
because of the principle of superposition, if each different approaches. Initially the wing was
elementary function is solution of the Laplace’s modelled in the standard way as a single wing with
equation, their linear combination will also be a a single set of parameters and a single aerodynamic
solution for that equation. Among the several influence matrix. Then, the semispan was split in
suitable functions, the VLM is based on vortex two wings, each one with its own independent set of
segments following the Biot-Savart law. The method parameters. In this case also mutual influence
covers the solid surface with a lattice of vortices matrices need to be computed, and then correctly
whose circulations are initially unknown, and will be assembled in a global one. The results are reported
determined after enforcing the boundary condition in Tab. Table 1, showing good agreement in both
(Eq. 8) in some discrete locations, named cases.
collocation points. Once the circulations are known,
the aerodynamic force can be computed by means
of the Kutta-Jukowsky theorem: 𝑭 = 𝜌𝑉 × 𝜞, where
𝑭 is the aerodynamic force per unit length, 𝑽 the
free stream speed and 𝜞 the vortex circulation.
Reformulating the boundary condition (Eq. 8) in
terms of the normal velocity components induced by
the vortex distribution and the free-stream speed
yields the linear equation as in Eq. 9,

𝛻𝛷 ⋅ 𝒏 = 𝛻𝛷 ⋅ 𝒏 → 𝑨𝜞 = 𝛻𝑉 ⋅ 𝒏 (9)

where 𝑨 is the aerodynamic influence matrix, in

which any element (i,j) represents the velocity
induced in i by a unit vortex in j. In principle, for each
given rigid configuration the influence matrix needs
to be computed just once, and then can be used to
solve the problem for any desired flow condition,
specified by the right-hand side of Eq. 9. In the
present case, though, the solution was searched for Figure 2: Warren-12 wing VLM model decomposed
the trimmed aircraft, which requires that the in four wings to validate the interaction of multi-wing
geometry adapts to each flight condition in order to configurations.
keep a balanced load distribution. Therefore, the
geometry being a function of the loads, the problem
becomes nonlinear, and an iterative algorithm was
set up in order to efficiently find the equilibrium
𝑪𝑳𝜶 𝑪𝒎𝜶
configuration with a reduced number of evaluations.
The VLM solver, coded in Fortran, has been Theory (DATCOM ) [41] 2.743 -3.10
wrapped in Python using the F90wrap open source SURFACES solver [41] 2.790 -3.17
package [40] and included in the FAST
aerodynamic module as a switchable option. To Present solver - 1w 2.780 -3.12
speed up the loops, a Python interface script allows Present solver - 4w 2.778 -3.11
data to be exchanged between the modules without
writing and reading input and output files. Table 1: Results for the Warren-12 wing, compared
Before plugging the module, the code has been with theoretical and numerical data from [41].
tested and validated against available data from the
open literature and free software. Some relevant
validation cases are reported in the following Since the analyses presented in this work require
section. the capability to predict trim configuration, a few
tests were carried out to assess the accuracy of
3.2. Validation of the VLM solver pitching moment estimation with varying geometry.
To this purpose, a flat rectangular wing was
Some validation cases are here presented to attest analysed with sweep angles from 0° to 15°, and
the reliability of the employed solver. results were compared with those from the XFLR5
First, to verify the predictions in symmetric flight software [42] (see Fig. 3).
conditions. the Warren-12 wing was used (Fig. 2), The last case here reported is for a multi-wing three-

4
dimensional configuration in asymmetric flow (Fig. Table 2: Aerodynamic coefficients for the validation
4). Again, the aerodynamic coefficients were case of Fig. 4.
compared with those available from XFLR5. A good
matching was obtained, as shown in Tab. 2.
4. METHODOLOGY
The first objective of the present work is to assess
the robustness of the design tool when different
fidelity levels are available. To this purpose, the
proposed approach is to start from a reference
benchmark design and use FAST to evaluate the
scatter of the design outcomes when some relevant
variables are perturbed. Repeating this task with
each of the L0 and L1 fidelity levels (see Section 2)
gives further insights on how the agreement
between the two tools changes when exploring new
designs.
To accomplish this, a number of design variables
and their range of variation were fixed. The interest
here is to compare two different aerodynamic
models and investigate how their different
predictions are propagated through the
multidisciplinary architecture at a conceptual design
Figure 3: Pitching moment coefficient variation with level. Therefore, a few variables defining the main
increasing sweep angle, compared with results wing geometry are enough to accomplish the task.
from XFLR5 [42]. Case of at flat wing of wingspan The selected variables are reported in Tab. 3
20 m, chord 1 m, angle of attack 2°. together with their boundaries, which were chosen
in accordance with other literature on aircraft
conceptual design exploration as [8].

Design variables Ref values Boundaries

Aspect ratio 9.48 ±25%
Kink position 0.4 ±15%
Sweep at 25% chord 25 [deg] ±15%
Taper ratio 0.3 ±15%

Table 3: Design Variables and relative boundaries

with respect to the reference CeRAS baseline.

A Design of Experiments (DOE) was carried out

using a Latin Hypercube Sampling (LHS) algorithm
in order to generate a population of design
candidates with good projection properties [43].
Figure 4: Multi-wing VLM configuration in This data set was used as training points to build
asymmetric flow used for validation against the some surrogate models as approximation of the L0-
XFLR5 code[42]. Main wing: sweep angle 0°, based or L1-based MDA model. For this purpose,
dihedral angle = 10°. Horizontal stabilizer: the recent toolbox called Surrogate Model Toolbox
span 5m, chord 1m, tilt angle 3°. Fin: span 3,5m, (SMT [49]) was used. SMT is a Python package that
chord 1m. Angle of attack 2°, sideslip angle 2°. contains a collection of surrogate modelling
methods, sampling techniques, and benchmarking
functions. 150 points were used to train the kriging
XFLR5 [42] Present solver model interpolation (Gaussian process model, [49]).
The second objective of this work is to exploit the
𝑪𝑳 0.123 0.131 information learned from the design exploration to
0.001 0.001 perform an optimization on the reference baseline.
𝑪𝑫
The aim is to assess the impact of the L0 and L1
𝑪𝒎 0.384 0.377 models on the optimization outcomes.
0.007 0.006 To this purpose, three different optimization
𝑪𝒍
algorithms were adopted. Once the kriging model is
𝑪𝒏 0.009 0.009 built, it can be easily used within an optimization
process. A gradient free (COBYLA [44]) and a
gradient based (SLSQP [46]) algorithms from the

5
Scipy Python library [47] were selected to be used model, thanks to the viscous correction of Section
in this study. A multi-start approach was adopted to 3, is capable to capture the drag at low lift
avoid convergence to local minima. A number of 10 coefficient, which would not be possible for a pure
starting points, selected with the LHS method of VLM solver.
[43], was found adequate. Although the mentioned
algorithms were found satisfactory for the present
case, further studies with involving more complex,
high dimensional searches, which are envisioned
for future work, would require improved techniques
such as the Efficient Global Optimization (EGO [45])
algorithm available within the SMT package.
Further details on the adopted test case are given
in Section 5. Results are reported and discussed in
Section 6.

5. TEST CASE

The test case is for a small-medium range aircraft

reproducing the features of the Airbus A320 family
from available public data, mainly from the CeRAS
archive [27] with some corrections and additional
assumptions as in [31]. The baseline model Figure 5: CeRAS baseline planform, from [31].
planform is shown in Fig. Figure 5, whereas the
main parameters are reported in Tab Table 4:
CeRAS baseline parameters, from [31]..
The VLM representation of its configuration is also
reported (Fig. 6). The former shows that the
semispan of the main wing was modelled by
assembling three different wings, attached together
and discretized in a parametrical way such that the
connections remains consistent for any global
geometrical variation. Also, the number of span-
wise panels was parameterized as a function of the
chord-wise panels so that the two dimensions of
each cell are approximately equal. The number of
chord-wise panels was fixed to N = 15 after a
convergence study, reported in Fig. 7. The VLM
representations includes also the meanline camber
of the baseline airfoil, which remains unchanged for
any design change. This is visible in Fig. 8, which Figure 6: VLM model of the CeRAS baseline -
shows the pressure distribution during one step of top view.
the polar calculation. The drag polars of the full
trimmed vehicle at cruise conditions calculated with
both the L1 and the L0 aerodynamic models are
plotted in Fig. 9, against some reference data from
the CeRAS database. It is worth noting that the L1
Top Level Aircraft Requirements
Number of passengers 150
Passenger weight [lbs] 200
Design Range [NM] 2750
Operational Range [NM] 800
Cruise Mach number 0.78
Approach speed [kts] 132
Wing Geometry
Aspect ratio 9.48
Wing break 0.4
Sweep angle at 25% [deg] 25
Taper ratio 0.3 Figure 7: Convergence of the trimmed-aircraft lift
Propulsion coefficient with increasing number of chord-wise
Max thrust at sea level [N] 117880 panels N, conducted at cruise speed, and an
Table 4: CeRAS baseline parameters, from [31]. imposed angle of attack of 3°.

6
 Fig. 9, is the significant mismatch of fuel weight
estimations over most of the examined domain.
This phenomenon is clearly reproduced in Fig. 12,
where the Probability Distribution Functions (PDF)
of four performance indices are extracted from the
analysed data. It is found that, in three of the four
cases, the L0 and L1 distributions are shifted by
3÷10%. Since the L0 model is knowledge-based
and created for configurations close to the adopted
baseline, its results are more clustered in the
neighbourhood of the CeRAS data. Nevertheless,
its reliability for innovative configurations remains
questionable.
Fig. 13 shows that for design candidates featuring
more significant variations from the baseline, the
shift between the L0 and L1 predictions remains.
Also, it can be noted that the amplitude of the scatter
obtained by increasing the deviation from the
Figure 8: Pressure distribution over the trimmed baseline parameters increases in the same way for
CeRAS VLM model for polar computation at cruise both L0 and L1. No convergence of the results from
speed. the two models is found.
The outcomes of the optimization task, where the
aim is to minimize the mission fuel weight, follow the
same trend, with L0 predicting a more performing
minimum (see Tab. 5). It is interesting to note that
the two optimal designs are appreciably different
between them, as well as with respect to the
baseline configuration. The task therefore turns out
not to be robust enough, since considerably
different optimal values are estimated,
corresponding to considerably different design
points.

Figure 9: Cruise polars computed with the two

aerodynamic models. CeRAS data are reported for
comparison.

6. RESULTS

The MDA results obtained from the DOE are shown

in Fig. 10 for the L0-based case, and in Fig. 11 for
the L1-based case. In both cases the resulting
mission fuel weight is plotted against the four design
variables, and its value being proportional to the
size of the spheres. It is found that the L1
aerodynamic model provides worse performance
predictions compared to L0. It is not unexpected Figure 10: 5D visualization of fuel weight
that, in both cases, the best candidates in terms of predictions. The size of the spheres is proportional
fuel weight correspond in most cases to medium- to the normalized fuel weight. Results obtained with
high aspect ratios. This is due to the reduction of L0 aerodynamic model.
induced drag, and is in line with the current trend
towards higher aspect ratio configurations (see Fig.
1 and [6]).
A less predictable outcome, given the apparently
minor differences visible in the polar predictions of

7
 conventional concepts. FAST has been extended to
include two levels of aerodynamic models: to the
original one (L0), based on empirical methods
created around the Airbus A320 family of aircraft, a
steady/unsteady panel method was added (L1)
based on the vortex lattice approach. FAST was
employed to explore a design space around a
reference baseline using both the L0 and L1
models. The impact of four main wing design
parameters on four key figures of merit was studied
for a set of 150 candidates. The Design of
Experiments was performed with a Latin Hypercube
Sampling algorithm, and the scatter of the results
over the four-dimensional space revealed
appreciable differences in the predictions obtained
with the two fidelity levels.

Figure 11: 5D visualization of fuel weight

predictions. The size of the spheres is proportional
to the normalized fuel weight. Results obtained with
L1 aerodynamic model.

Figure 13: Scatter plot of four performance indices

against the root mean square deviation of the
design variables from the baseline parameters.
CeRAS data are reported when available.

AR wb s25 tr fsurr ftrue

L0 SLSQP 11.76 0.57 22.1 0.51 17,654 17,731
Figure 12: Probability Distribution Functions (PDF) COBYLA 12.58 0.47 24.4 0.61 17,628 17,574
of four performance indices, obtained with L0 and
L1 for the same population of candidates. CeRAS L1 SLSQP 11.24 0.30 28.7 0.36 19,041 19,248
data, when available, are reported for comparison. COBYLA 11.27 0.30 28.2 0.36 19,042 19,280
CeRAS baseline 9.48 0.40 25.0 0.38 - 18,678

Table 5: Fuel weight optimization results with two different

7. CONCLUSIONS optimization algorithms for the L0 and L1 data sets. fsurr is
the optimal prediction from the surrogate model, whereas
A Multidisciplinary Design Analysis and ftrue is the value computed by FAST. The best values are
Optimization framework is as computationally highlighted in bold. The CeRAS baseline data are
efficient and accurate as the least efficient and reported for comparison.
accurate method adopted for each discipline-
specific analysis. Therefore, the choice of an
The following step was to assess the impact of the
appropriate model and an appreciation of the
two aerodynamic models on the optimization of the
underlying assumptions is critical to define the
fuel weight. To this purpose, the gathered data were
intrinsic limitations of the framework. This work
fed to a kriging algorithm to generate a surrogate
leverages on a sizing tool for conceptual design,
model of the desired performance over the design
named FAST, originally developed to investigate
space. One gradient-based and one gradient-free

8
optimization algorithms were compared. After 1741.
optimization, it was found that: 1) the predicted
3. M. V. Bendarkar, A. Behere, S. I. Briceno, D. N.
optimal fuel weight shows significant variations -
Mavris, A Bayesian Safety Assessment
around 10 t for airplanes with MTOW between 75.5
Methodology for Novel Aircraft Architectures
t and 76.5 t; 2) the best configurations
and Technologies Using Continuous FHA.
corresponding to minimum fuel weight are also
AIAA Aviation Forum, 17-21 June 2019,
different in terms of all wing planform design
Dallas, Texas (AIAA Paper 2019-3123).
parameters apart from the AR (see Fig. 14). The
https://doi.org/10.2514/6.2019-3123.
study provides a lucid example that small changes
within the optimization chain may have a large 4. Sgueglia, A., Schmollgruber, P. Bartoli, N.,
impact on the results, calling for an approach that is Atinault, O., Bénard, E. & Morlier, J. (2018).
robust to deal with these changes. Exploration and Sizing of a Large Passenger
Moreover, both cases show that the trend for Aircraft with Distributed Electric Ducted Fans.
reduced fuel consumption is towards high aspect AIAA Scitech Forum, 8-12 January 2018,
ratio configurations. This indicates that in order to Kissimmee, United States (AIAA Paper 2018-
provide realistic predictions aeroelastic modules 1745). https://doi.org/10.2514/6.2018-1745.
need to be included within the MDAO architecture.
5. A. Sgueglia, P. Schmollgruber, E. Bénard, N.
The outcomes of this study pave the way for further
Bartoli, J. Morlier, Preliminary sizing of a
improvements towards effective robust analysis
medium range blended wing-body using a
capabilities for new generation aircraft concepts.
multidisciplinary design analysis approach,
MATEC Web of Conferences 233 (2018) 1–9.
doi:10.1051/matecconf/201823300014.
https://oatao.univ-toulouse.fr/21202/
6. G. Potter, Conceptual design of a strut-braced
wing configuration, UTIAS National
Colloquium on Sustainable Aviation, June 21-
23, 2017.
7. F. Bocola, V. Muscarello, G. Quaranta, P.
Masarati, Pilot in the loop aeroservoelastic
simulation in support to the conceptual design
of a fly by wire airplane, AIAA Atmospheric
Flight Mechanics Conference, 22-26 June,
Dallas,TX (AIAA Paper 2015-2557).
doi:10.2514/6.2015-2557.
8. V. Trifari, M. Ruocco, V. Cusati, F. Nicolosi, A.
De Marco, Multi-disciplinary analysis and
optimization java tool for aircraft design, 2018.
Figure 14: Optimal configurations for minimum fuel 9. Airworthiness Directives; The Boeing Company
weight obtained with L0 and L1, compared to the Airplanes. US Federal Aviation Administration,
baseline configuration. 14 CFR Part 39, Federal Register 136(80),
Rules and Regulations, 16 July 2015 4 pp
2012-42014.
10. L. Wang, C. Xiong, J. Hu, X. Wang, Z. Qiu,
8. REFERENCES Sequential multidisciplinary design
optimization and reliability analysis under
1. N. Cumpsty, D. Mavris, M. Kirby, Aviation and interval uncertainty, Aerospace Science and
the environment: Outlook, ICAO Technology, 80 (2018), pp. 508–519.
environmental report 2019, Chapter 1. doi:10.1016/j.ast.2018.07.029.
https://www.icao.int/environmental-
protection/Pages/environment- 11. J. Katz, A. Plotkin, Low-Speed Aerodynamics,
publications.aspx. Second Edition, Cambridge University Press,
2001.
2. Y.Cai, I. Chakraborty, D. N. Mavris, Integrated
assessment of vehicle-level performance of 12. Report of the Air Travel – Greener by Design
novel aircraft concepts and subsystem Mitigating the environmental impact of
architectures in early design, AIAA Aerospace aviation: Opportunities and priorities. (2005).
Sciences Meeting, SciTech forum, Kissimmee, The Aeronautical Journal (1968), 109(1099),
Florida, 8-12 January 2018 (AIAA Paper 2018- 361-416. doi:10.1017/S0001924000000841.
1741). doi:https://doi.org/10.2514/6.2018-

9
13. Piperni, P., DeBlois, A. & Henderson, R. design optimization of morphing aircraft.
(2013). Development of a multilevel Aerospace Science and Technology, Volume
multidisciplinary-optimization capability for an 67, pp 1-12,ISSN 1270-9638.
industrial environment. AIAA Journal 51, https://doi.org/10.1016/j.ast.2017.03.029.
2335–2352 doi: 10.2514/1.J052180.
23. Smith M. S., Sandwich, C.& NR Alley N. R.
14. A. Sieradzki, A. Dziubinski, C. Galinski. (2016). (2018). Aerodynamic analyses in support of
Performance comparison of the optimized the spanwise adaptive wing project. AIAA
inverted joined wing airplane concept and Aviation 2018, June 25-29, Atlanta, GA.
classical configuration airplanes. Arch. Mech.
24. Wilson, T., Castrichini, A., Azabal1,A.,
Eng. 63 (3), pp. 455-470.
Cooper2, J.E., Ajaj, R. & Herring1 M. (2017).
http://dx.doi.org/10.1515/ meceng-2016-0026.
Aeroelastic behaviour of hinged wing tips.
15. T. Goetzendorf-Grabowski,(2017). Multi- International Forum on Aeroelasticity and
disciplinary optimization in aeronautical Structural Dynamics (IFASD Paper 2017-216),
engineering. Proceedings of the Institution of 25-28 June, Como, Italy.
Mechanical Engineers, Part G: Journal of
25. Toal, D. J.J. (2014). On the potential of a multi-
Aerospace Engineering, 231(12), 2305–2313.
fidelity G-POD based approach for
https://doi.org/10.1177/0954410017706994
optimization & uncertainty quantification.
16. T. Goetzendorf-Grabowski, J. ASME Turbo Expo 2014: Turbine Technical
Mieloszyk,(2017). Common computational Conference and Exposition, Germany.
model for coupling panel method with finite
26. Rogers, H., Lee, D., Raper, D., Foster, P.,
element method. Aircraft Engineering and
Wilson, C., & Newton, P. (2002). The impacts
Aerospace Technology: An International
of aviation on the atmosphere. The
Journal 89(5), pp. 654–662. doi:
Aeronautical Journal (1968), 106(1064), pp.
10.1108/AEAT-01-2017-0044.
521-546. doi:10.1017/S0001924000018157.
17. Kharlamov, D., Drofelnik, J., Da Ronch, A. &
27. CeRAS - Central Reference Aircraft Data
Walker, S. (2018). Rapid Load Calculations
System, [on line database] URL:
Using an Efficient Unsteady Aerodynamic
https://ceras.ilr.rwth-aachen.de
Solver. AIAA AVIATION Forum, 25–29 June
2018, Atlanta, Georgia, USA. 28. E. Roux (2005). Pour un Approche Analytique
Doi:10.2514/6.2018-3621. de la Dinamique du Vol, PhD Thesis, ISAE-
SUPAERO, Toulouse.
18. O. Ş. Gabor, A. Koreanschi, R. M. Botez,
(2016). A new non-linear vortex lattice 29. B. Nagel et al. (2014),, “Communication in
method: applications to wing aerodynamic Aircraft Design: Can We Establish a Common
optimizations. Chinese Journal of Aeronautics, Language? ”, 28th Congress of the
Volume 29, Issue 5, pp. 1178-1195, ISSN International Council of the Aeronautical
1000-9361. Sciences.
https://doi.org/10.1016/j.cja.2016.08.001.
30. Schmollgruber P., Bedouet J., Sgueglia A.,
19. Paul, R. C., Gopalarathnam, A. (2014). Defoort S., Lafage R., Bartoli N., Gourinat Y. &
Iteration schemes for rapid post-stall Benard E. (2017). Use of a Certification
aerodynamic prediction of wings using a Constraints Module for Aircraft Design
decambering approach. International Journal Activities. 17th AIAA Aviation Technology,
for Numerical Methods in Fluids, Wiley Online Integration, and Operations Conference, 5-9
Library (wileyonlinelibrary.com). Doi: June, Denver, Colorado.
10.1002/fld.3931
31. Schmollgruber P. (2018). Enhancement of the
20. Da Ronch, A., Ghoreyshi, M. & Badcock, K. J. conceptual aircraft design process through
(2011). On the generation of flight dynamics certification constraints management and full
aerodynamic tables by computational fluid mission simulations. PhD Thesis. Université
dynamics. Progress in Aerospace Sciences, fédérale Toulouse Midi-Pyrénées, Toulouse,
47, pp. 597-620. France.
doi:10.1016/j.paerosci.2011.09.001
32. Dupont W. P., Colongo C. (2012). Preliminary
21. Choi, S., Alonso, J. J. & Kroo, H. M .(2009). design of commercial transport aircraft.
Two-level multifidelity design optimization ISAESupaero, Toulouse, France.
studies for supersonic jets. Journal of Aircraft
3(46), pp. 776-790. DOI: 10.2514/1.34362 33. R. J. Monaghan (1953). A review and
assessment of various formulae for turbulent
22. Afonso F., Vale, J., Lau F. & Suleman A. skin friction in compressible flow. Aeronautical
(2017). Performance based multidisciplinary Research Council Tech. Rep., C.P. 142

1
0
 (15.464). surrogate modeling framework with derivatives,
Adv. Eng. Softw., https://doi.org/10.1016/j
34. C. de Nicola (2015). Appunti per un corso di
.advengsoft.2019.03.005.
aerodinamica degli aeromobili. 1st ed. Class
notes, Università degli Studi di Napoli 49. Williams, C. K., & Rasmussen, C. E. (2006).
“Federico II”. Gaussian processes for machine learning
http://wpage.unina.it/denicola/AdA/DOWNLOA (Vol. 2, No. 3, p. 4). Cambridge, MA: MIT
D/Appunti_AdA_2014_2015.pdf. press.
35. J. D. Anderson Jr. (2011). Fundamentals of
aerodynamics. 5th ed. McGraw-Hill.
36. M. Niţă and D. Scholz (2012). Estimating the
Oswald factor from basic aircraft geometrical
parameters. In: Deutscher Lift- und
Raumfahrtkongress 2012, Document ID:
281424.
37. I. H. Abbott and A. E. von Doenhoff (1959).
Theory of wing sections, including a summary
of airfoil data. Dover publications Inc.
38. Sgueglia, A. (2019). Sizing and optimization
priorities applied to a Blended Wing-Body with
distributed electric ducted fans. PhD Thesis.
Université fédérale Toulouse Midi-Pyrénées,
Toulouse, France.
39. J. Katz, A. Plotkin. (2001). Low-Speed
Aerodynamics, Second Edition, Cambridge
University Press.
40. F90wrap: Fortran to Python interface generator
with derived type support.
https://github.com/jameskermode/f90wrap.
41. Surfaces-vortex lattice module, user manual.
(2009). Great OWL Publishing Engineering
Software.
42. XFLR software, http://www.xflr5.tech/xflr5.htm.
43. Jin, R. and Chen, W. and Sudjianto, A. (2005),
“An efficient algorithm for constructing optimal
design of computer experiments.” Journal of
Statistical Planning and Inference, 134:268-
287.
44. Powell, M. J. (1994). A direct search
optimization method that models the objective
and constraint functions by linear interpolation.
In Advances in optimization and numerical
analysis (pp. 51-67). Springer, Dordrecht.
45. Jones,D. R. Schonlau, M, Welch, W.J. (1998)
Efficient global optimization of expensive black-
box functions, J. Glob. Optim. 13(4) 455–492.
46. Kraft, D et al. (1988). A Software Package for
Sequential Quadratic Programming, DFVLR
Obersfaffeuhofen, Germany.
47. E. Jones, T. Oliphant, P. Peterson, et al. (2001).
SciPy: open source scientific tools for Python,
http://www.scipy.org/.
48. M.A. Bouhlel, J.T. Hwang, N. Bartoli, R. Lafage,
J. Morlier, J.R.R.A. Martins (2019). A Python

1
1