Span Efficiency Factor
Span Efficiency Factor
Span Efficiency Factor
Daniel Böhnke, Jonas Jepsen, Till Pfeiffer, Björn Nagel, Volker Gollnick
Air Transportation Systems, DLR e.V., Blohmstr. 18, 21079 Hamburg
Carsten Liersch
Institute of Aerodynamics and Flow Technology, DLR e.V., Lilienthalplatz 7, 38108 Braunschweig
Abstract 1 Introduction
part of the design space. In a simplified declara- tion 3. A trade study is then carried out and ex-
tion, historic models are categorized in concep- amined in Section 4. Finally, a discussion of the
tual design and physic based models in prelimi- results is given in Section 5 and some conclusions
nary design. and an outlook in Section 6.
The question arises whether it is possible to
close the gap between conceptual and prelimi- 2 State of the Art
nary design, i.e. historic and physic based analy-
sis. Preferably this should be accomplished in an In this section an overview is given for the calcu-
automated manner, to make the results of physic lation of the Oswald factor (e) in conceptual and
based analysis available at an earlier point of the preliminary design. The Oswald factor is needed
design. for the estimation of the lift induced drag (C DI ),
Requirements whereas drag due to lift (C DL ) in this paper is
meant to take into account lift induced drag and
drag due to viscous forces (C DV ). The calculation
Feedback
Loop of the wave drag (C DW ) is not considered any fur-
C ther. The lift induced drag in conceptual design
P
VAMPzero
Conceptual A Aerodynamics is usually noted as in Eq. 1:
C
S 1 C L2
Knowledge e= (1)
Base πAR C DI
Additionally, the drift for the zero lift angle of
Fig. 1: Multi-Fidelity Process attack (C L0 ) needs to be taken into account. Fur-
thermore, local lift distributions may result in lift
Within this study a central model approach
induced drag even if the overall lift sums up to
(CPACS) is proposed that combines a historic
zero, e.g. for twist distributions. Therefore a pa-
based conceptual design module (VAMPzero)
rameter for induced drag at zero lift (C DI0 ) is in-
with a physic based aerodynamic module (LIFT-
troduced. Equation 1 is then extended to Eq. 2:
ING_LINE), [3, 4]. As use-case the calculation
of the Oswald factor is chosen. On a first step, 1 (C L − C L0 )2
a set of requirements is defined and fed into the e= (2)
πAR (C DI − C DI0 )
conceptual design tool. With knowledge based
engineering techniques a detailed geometry defi- The first outlined approach originates from
nition is derived from the conceptual design and Raymer and is based on the aspect ratio (AR) and
stored in the central model data format. The aero- the leading edge sweep angle of the wing (Λ), [6].
dynamic module is linked to the central model Two different equations valid for unswept (Eq. 3)
and triggered without further user input. The re- and swept (Eq. 4) wings are defined.
sulting aerodynamic performance map is subse-
quently reduced to the Oswald factor and reused
e = 1.78(1 − 0.045AR0.68 ) − 0.64 (3)
within the conceptual design module. Provided
that a sufficient number of trades is carried out, e = 4.61(1 − 0.045AR 0.68
)(cos(Λ))0.15
− 3.1 (4)
a symbolic regression approach, [5] is applied to with Λ > 30
derive new conceptual design methods from the
multi-fidelity loop, as displayed in Figure 1. These equations can be evaluated quickly as
The paper is grouped into six sections: This the number of variables is low. Drawbacks exist
section gives a short introduction. Section 2 elab- because the design space is limited to a specific
orates on existing methods for the determination area via the limitations on the leading edge sweep
for the Oswald factor in conceptual design. The angle.
proposed design environment is described in Sec- An approach by Howe (Eq. 5) takes into ac-
count additional parameters like the flight Mach
number (MN ), the taper ratio (λ) and the thick- will be extended by a set of relvant vaiables with
ness to chord ratio (t/c), [7]. The equations de- their corresponding formulation.
rived by Howe are defined for drag due to lift and
it is hard to extract an exact formulation for the 3 Design Environment
Oswald factor. Nevertheless, they are included in
this overview to display complex symbolic equa- In a generalized manner, a distributed design en-
tions known in conceptual design as we will de- vironment consists of three components. The
rive such an equation at a later point of this study. basis for all communication is a common lan-
The functions for the taper ratio and aspect ratio guage that is achieved via a central model ap-
are given in Eq. 6 and Eq. 7: proach. The second component is a set of analy-
sis modules that can exchange information using
(1 + 0.12MN6 ) the common language. The benefits of using a
C DI = (5) central model are depicted in Figure 3. Due to
πAR
"
0.142 + AR f (λ)(10t/c)0.33
# the fact that communication now only exists be-
1 + g(AR) + tween the analysis modules and the central model
cos(Λ)2
h i the number of interfaces decreases. Additionally,
f (λ) = 0.005 1 + 1.5(λ − 0.6)2 (6) interaction between the analysis modules is de-
0.1 creased in a way that different modules can be
g(AR) = (7)
(4 + AR)0.8 utilized for similar tasks. For example a fuse-
lage weight prediction can be made using a his-
A semi physic based approach is described by toric based or a physic based, either beam or
Jenkinson for the induced drag utilizing lifting- CPACS
shell, model. Regardless from the chosen anal-
line theory, [8, 9]. The approach leads to a func- ysis strategy, the fuselage mass is included in the
tion (C1 ) depending on the taper ratio and aspect overall design and available for further modules.
ratio, as shown in Figure 2.
► n(n-1) ► 2n
Fig. 2: C1 depending on λ and AR, from [8] work also takes over general tasks like process
Aeroconf 2011 > Daniel Böhnke > 09.03.2011
3.3 VAMPzero
For the purposes of DLR’s distributed multi-
disciplinary design environment a new concep-
utal design module is developed. VAMPzero is
based on handbook methods taken from well-
known design literature, [6, 7, 12]. The key mod-
eling aspects of VAMPzero are described in more
detail in [4].
Fig. 4: Baseline Geometry In a design environment that is mostly based
on preliminary level, two possible tasks remain
In Figure 4 a sample geometry in CPACS for a conceptual design tool. The first one is the
is displayed. This geometry is generated by initialization of the datasets. Designers usually
VAMPzero (s. Section 3.3) and is the baseline need to start working with only the top level air-
for the trade study (s. Section 4). It shows a craft requirements (TLAR) at hand. In this case
conventional short range configuration in a quasi- the conceptual design code can close the gap be-
symmetrical view. tween the TLAR and sufficient information to
trigger preliminary analysis modules.
To overcome the gap, the results from the con-
ceptual design are transferred into CPACS via
3.2 LIFTING_LINE a knowledge based engineering approach. For
the conceptual design calculations and sizing
In the aerodynamic analysis the CPACS dataset is VAMPzero uses a single trapez wing shape. For
imported via an interface tool to trigger the mul- the export the wing shape can be mapped to a
tiple lifting line program LIFTING_LINE, [9]. double trapez with an additional rectangular sec-
The interface from LIFTING_LINE to CPACS tion up to the wing-fuselage intersection. Fig-
and the interaction with other analysis modules ure 6 shows the geometry. The wing planform
is outlined in more detail by Liersch et. al. in [3] is described by the wing span (bW ), the leading
edge sweep, the aspect ratio, the taper ratio, and The second task for a conceptual design tool
the kink ratio (ηk ). The trailing edge from root to in a preliminary design environment is the inte-
kink is perpendicular to the fuselage. The twist gration of results. Results from several higher
angle (ϑ) is applied in a linear distribution from level analysis modules need to be aggregated to
root to tip. Thickness to chord ratios at root and obtain the effects on the overall aircraft configu-
tip stations are a result from VAMPzero. For ration. For example, the wing mass in the mass
the empenage symmetrical airfoils are used. The breakdown is overwritten by higher level analysis
wing geometry is generated with NACA-6 series module (e.g. FEM) and VAMPzero takes these
airfoil. Further airfoil sets and their assortment changes into account when re-calculating the air-
is an ongoing topic within the development of craft. In the case of this study, VAMPzero inter-
VAMPzero. prets the results from LIFTING_LINE to derive
y the Oswald factor and calculate a new configura-
x tion.
bw
ηk bw
3.4 RCE
0.01 ◦ × SW ×
× × × ×
C DI ×
◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦
× × × × × × × × × ◦ ×
× ◦ ×
◦ × ◦ ×
◦ × ◦ ◦ ◦ ◦ ◦ ◦
4 Trade Study
0.005
The baseline configuration for this study is a sin-
gle aisle (nPAX = 150), medium range (R = -20 -10 0 10 20 30
3000km, Mach = 0.78) aircraft. Λ
To determine the effects of different parame-
◦
×
ters on the lift induced drag, design variables are ◦
× SR
◦
varied as shown in Table 2. Due to the fact that
×
◦
× BL ◦
◦
the number of dimensions and the range of the 0.01 ×
◦
× SW ×
C DI ◦
× ◦
× ◦ ◦
×
variables are large, a full factorial design is un- × ×
◦ ◦
◦ ◦
× ×
◦ ◦ ◦
× ×
× × ◦ ◦ ◦ ◦
suitable. Instead the design space is explored by × × ×
×
0.005
Latin-Hypercube sampling. Apart from the spec-
ified values for the baseline configuration the ref- 6 8 10 12 14 16
erence area for the wing is kept constant. AR
tails on the benefits from twist due to load reduc- trol data. It is assumed that reliable results can be
tion at the tip. The single trapez wing can still obtained within the area of the design space. A
benefit from the twist. The fitted curve is in ac- detailed assumption on the validity of the equa-
cordance with the baseline. The twist has a high tion outside of the design space is not yet given.
influence on C DI0 . The aerodynamic (in)efficency Finally, it is possible to automize large parts of
can therefore not be described by the Oswald fac- the analysis chain. Nevertheless the extraction of
tor alone. knowledge is not yet possible in a similar way.
As already mentioned, the change in the kink This is due to practical reasons, as the process
ratio only applies to the double trapez wing. The works also well regardless of faulty data. For val-
more the kink moves outward the higher the lift idation and verification the Engineer in the Loop
induced drag. This is due to the fact that the is a necessity.
inward area increases rapidly in comparison the
outer part of the wing. The symbolic regression
constantly over estimates the lift induced drag.
The plots show a good agreement between the 6 Conclusion
derived equations and the single parameter varia-
tions. The root mean squared error for the Os- The current study outlines a methodology to de-
wald factor is 0.011. This gives confidence to rive the Oswald factor in a multi-fidelity envi-
implement the derived equation (Eq. 10) into the ronment. This methodology can be extended for
conceptual design tool VAMPzero. As all param- different use cases targeting single higher fidelity
eters that are needed to generate the higher level analysis modules, e.g. for mass estimation. For
geometry are avaiblable in VAMPzero no addi- future research the target is to analyze couplings
tional customizations of the code are necessary. of higher fidelity tools, as for example in aeroe-
The outlined study focuses on the interaction lastics, and try to reflect these couplings also on
of analysis modules on different levels of fidelity. the level of a simple analysis module such as
As it was desired to derive a new calculation VAMPzero. These coulings are not necessarily
method in a large design space the number of de- limited to different disciplines but may also cou-
sign evaluations was high. In this case the study ple analysis modules of different fidelity levels.
benefitted from the fact that LIFTING_LINE of- Within the outlined approach the curse of di-
fers a physic based approach at low computa- mensionality is of crucial importance, [17]. Due
tional cost. It is questionable at which cost a sim- to the fact that the number of design variables
ilar appraoch is possible with higher detail mod- is usually beyond two, the number of calculation
els, e.g. CFD. runs increases rapidly. Determining the number
Integration of higher level analysis modules of- of required runs for a good fit is currently only a
ten is done using response surface models. In this smart guess. Integrating the symbolic regression
way the computational time is kept low, as most toolbox into the design loop may give the chance
of it is spent a-priori. Within the outlined method to optimize this process. The symbolic regression
this approach is possible as seen from the derived algorithm is capable of deriving new experiments
equation. Additionally, the direct link between at locations in the design space where the current
VAMPzero and the disciplinary analysis module error is high. Starting from a small number of
LIFTING_LINE is existent and needs to be trig- calculation, e.g. a full factorial design only on
gered as soon as the design space is enlarged the outer limits of the design space, the design
It is shown that the derived equation from the loop could be configured until the overall error
symbolic regression approach fits well to the con- reaches a certain minimum.
[1] Stark, J., “Product Lifecycle Management , [11] Koch, A., Lührs, B., Dahlmann, K., Linke,
21st Century Paradigm for Product Realisa- F., Grewe, V., Plohr, M., Nagel, B., Goll-
tion,” Springer London, 2005. nick, V., and Schumann, U., “Climate im-
pact assessment of varying cruise flight al-
[2] La Rocca, G. and van Tooren, M., “En-
titudes applying the CATS simulation ap-
abling distributed multi-disciplinary design
proach ,” CEAS European Air and Space
of Complex Products : a knowledge based
Conference, 2011.
Engineering Approach,” Journal of Design
Research, Vol. 5, 2007, pp. 333–352.
[12] Roskam, J., “Airplane Design,” DARCorpo-
[3] Liersch, C. and Hepperle, M., “A Uni- ration, Vol. I-VII, 1989.
fied Approach for Multidisciplinary Air-
craft Design,” CEAS European Air and [13] Buckingham, E., “On Physically Similar
Space Conference, 2009. Systems; Illustrations of the Use of Dimen-
sional Equations,” Physical Review, 1914,
[4] Böhnke, D., Nagel, B., and Gollnick, V., pp. 345–376.
“An Approach to Multi-Fidelity in Con-
ceptual Aircraft Design in Distributed De- [14] Jepsen, J., “Wissensbasierte Erstellung
sign Environments,” IEEE Aerospace Con- von Geometrien auf der Grundlage von
ference, 2011. Handbuchverfahren,” IB-328-2011-07, TU
[5] Schmidt, M. and Lipson, H., “Distilling Hamburg-Harburg, 2011.
Free-Form Natural Laws from Experimen-
tal Data,” Science, Vol. 324, 2009, pp. 81– [15] Litz, M., Bachmann, A., and
85. Kunde, M., “TIVA XML Interface,”
tixi.googlecode.com.
[6] Raymer, D., “Aircraft Design: A Concep-
tual Approach,” AIAA Education Series, [16] Litz, M., Bachmann, A., and Kunde,
1989. M., “TIVA Geometric Library,”
tigl.googlecode.com.
[7] Howe, D., “Aircraft Conceptual Design
Synthesis,” Professional Engineering Pub-
[17] Koch, P. N., Simpson, T., Allen, J. K.,
lishing, 2000.
and Mistree, F., “Statistical Approximations
[8] Jenkinson, L., Simkin, P., and Rhodes, D., for Multidisciplinary Design Optimization:
“Civil Jet Aircraft Design,” Butterworth The Problem of Size,” Journal of Aircraft,
Heinemann, 1999. Vol. 36, 1999, pp. 275–286.