Aerospace Science and Technology 37 (2014) 93109

Contents lists available at ScienceDirect

Aerospace Science and Technology

www.elsevier.com/locate/aescte

Winglet multi-objective shape optimization

Ali Elham a, , Michel J.L. van Tooren b
a
b

Delft University of Technology, Kluyverweg 1, 2629HS Delft, The Netherlands

University of South Carolina, Columbia, SC 29201, USA

a r t i c l e

i n f o

Article history:
Received 18 February 2014
Received in revised form 24 April 2014
Accepted 7 May 2014
Available online 5 June 2014
Keywords:
Winglet
Multi-objective optimization
Structural weight estimation

a b s t r a c t
A series of multi-objective winglet shape optimizations are performed to nd the Pareto front between
the wing aerodynamic drag and the wing structural weight for a wing equipped with a winglet. The
paper discusses the addition of winglets to existing aircraft designs. The outer shape of the wing is
therefore xed but the internal structure and associated weight are adjusted based on the change of
spanwise loading (magnitude and distribution) due to the addition of the winglet. In order to estimate the
aerodynamic and structural characteristics of a non-planar wing, a quasi-three-dimensional aerodynamic
solver is integrated with a quasi-analytical weight estimation method inside an optimization framework.
Using those tools, the aerodynamic drag and the structural weight of a wing equipped with various
winglets are estimated with a high level of accuracy. A multi-objective genetic algorithm is used to
determine the Pareto front for two objective functions: minimum wing drag and minimum wing weight.
In order to nd the best winglet shape among the winglets on the Pareto front, three gures of merit
are used: the aircraft Maximum Takeoff Weight (MTOW), the aircraft fuel weight and the aircraft Direct
Operating Cost (DOC). The optimization results showed that about 3.8% reduction in fuel weight and
about 29M$ reduction in 15 years DOC of a Boeing 747 type aircraft can be achieved by using winglets.
2014 Elsevier Masson SAS. All rights reserved.

1. Introduction
Beginning with early work of Whitcomb [34], using winglets for
improving aircraft aerodynamic eciency became topic of many
research. Detailed aerodynamic principles of winglets are discussed
in [6]. In general a winglet produces a side force, which has a forward component because of the sidewash produced by the wing
tip circulation. This forward component of the lift acts as a thrust
force reducing the aircraft induced drag, which is about 40% of the
total drag in cruise and 8090% of the total drag in the second
climb segment [20].
Some researchers have tried to design and optimize the winglet
shape for achieving minimum wing drag [1,32]. However, adding a
winglet to a wing affects the wing structural weight as well. Therefore a pure aerodynamic optimization may result in a winglet with
a huge weight penalty. This additional weight may compensate the
effect of drag reduction in an aircraft take-off weight or even fuel
weight minimization. Besides, the increase in wing weight should
be limited for add-on winglets to minimize the amount of required
modications in the wing structure.

Corresponding author. Tel.: +31 15 2784896.

E-mail address: [email protected] (A. Elham).

http://dx.doi.org/10.1016/j.ast.2014.05.011
1270-9638/ 2014 Elsevier Masson SAS. All rights reserved.

The wing root bending moment has been used as a constraint

to control the wing weight during a winglet design and optimization in some research [11,12,16]. In those cases the winglet has
been designed for a xed root bending moment to keep the wing
weight constant. Ning and Kroo [26] argued that using root bending moment is not a good indicator for the wing total weight, because it does not account for the effect of chordwise changes in the
planform on the bending weight. They suggested a semi-empirical
wing weight estimation method, which calculates the bending
weight as a function of the integrated bending moment over the
entire span. More sophisticated structural analysis has been used
by Jansen et al. [14], in which the wing weight is calculated by
performing a Finite Element Analysis (FEA) of the wing structure.
In their approach the wing structure is replaced with an equivalent
beam and the thickness of that beam is used as a design variable
in optimization. The results of the FEA is corrected using some empirical factors to yield a reasonable value for the wing weight.
An alternative approach for wing structural weight estimation is
quasi-analytical methods. In such methods, the primary weight is
estimated using elementary, physics based structural analysis and
the secondary weights are estimated using empirical method [5,18,
23,29,30]. Those methods are more accurate and also more sensitive to design variables than the empirical methods, faster than
FEM based methods, and can predict the structural weight with
high level of accuracy.

94

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

Nomenclature
C
CD
C Di
CDp
CD f
CL
Ct
C wr
d
deff
l
leff
lw
M
M dd
nmax

wing chord
wing drag coecient
induced drag coecient
pressure drag coecient
friction drag coecient
wing lift coecient
wing tip chord
winglet root chord
airfoil drag
airfoil effective drag
airfoil lift
airfoil effective lift
winglet length
bending moment
drag divergence Mach number
maximum load factor

The main effect of using winglet is the reduction of the wing

induced drag. The induced drag can be estimated with a good level
of accuracy using a Trefftz plane analysis [25] of the wake behind
a wing. This analysis can be done using a Lifting Line (LL) or a Vortex Lattice Method (VLM). Therefore in many cases LL or VLM are
used to calculate the induced drag. In order to calculate viscous
drag strip theory in combination of some semi-empirical relations
can be applied [14,26,32,33]. Effect of compressibility (wave drag)
is normally ignored. However ignoring some components of drag,
or using semi-empirical models may result in unrealistic designs.
For example Verstraeten and Slingerland [32] found the optimum
winglet height to be about 75% of the semi-span which is an unrealistic design. They discussed that the low level of delity aerodynamic model used by them is the cause of such an unrealistic
result.
In the present research the winglet shape has been optimized
using aerodynamic and structure models with higher level of accuracy. Various components of the wing drag are calculated using a quasi-three-dimensional aerodynamic solver. Instead of using
a constraint on the wing root bending moment, the total wing
structural weight is estimated using a quasi-analytical weight estimation method with high level of accuracy. Using the mentioned
solvers a series of multi-objective as well as single-objective optimizations are applied to investigate the effect of a winglet on the
performance of transonic passenger aircraft.
2. Analysis method
In order to steer the optimization toward more realistic results,
a high level of accuracy of the analysis methods is required. On
the other hand, the analysis methods should be fast and robust
enough to be called hundreds of time during an optimization. The
aerodynamic model should accurately estimate both the induced
drag and the prole drag of a wing equipped with a winglet. Since
the method is to be applied to winglet design for transonic passenger aircraft, the compressibility effects should be covered.
In addition, the weight estimation should be performed taking into account the effect of aerodynamic shape on the structural
weight, which is necessary for a wing multidisciplinary shape optimization. The weight model should be able to calculate the wing
weight due to all the applied loads: bending, shear and torsion. It
is also important to take into account the relief loads due to the
wing weight and the fuel weight (for the fuel stored inside the
wing).

q
Reeff
S
Sw
t max
V eff
V

i

wr
wt
t
w
w
w

dynamic pressure
effective Reynolds number
shear load
wing reference area
airfoil maximum thickness
effective velocity
free stream velocity
wing angle of attack
induced angle of attack
wing twist angle
winglet root twist angle
winglet tip twist angle
airfoil effective distance coecient
winglet taper ratio
winglet leading edge sweep angle
winglet cant angle

2.1. Aerodynamic model

A quasi-three-dimensional aerodynamic solver [24] is developed and validated to estimate the wing drag and also the applied
loads for structural analysis. The solver is based on strip theory
[8]. In this approach the total wing drag is decomposed into two
main components: the induced drag and the prole drag. The prole drag includes the pressure drag (including the wave drag) and
the friction drag. The drag estimation is achieved by a combined
use of a VLM to estimate the induced drag and a full potential
analysis integrated with the boundary layer model to estimate the
prole drag of airfoil sections at several wing spanwise positions.
In the quasi-three-dimensional approach for drag estimation,
the rst step is to determine the wing lift distribution for a given
lift coecient using a VLM. The required angle of attack for the
wing to produce a given amount of lift is also determined using
the same tool. Then the wing is divided into several sections (8
sections are enough [24]). The local lift at each section is determined from the lift distribution calculated by the VLM. The effective lift at each section can be determined if the downwash angle
and the effective drag at that section are known (see Fig. 1). The
following process is used to determine the drag of each 2D section:
1. Set i = 0 and deff = 0
2. leff = (l + deff sin i )/ cos i
3. Determine the effective velocity V eff = V / cos i and compute the effective Reynolds number Reeff

Fig. 1. Three-dimensional lift (l), three-dimensional drag (d), local effective lift (leff )
and local effective drag (deff ) of a strip.

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

Fig. 2. Various components of drag coecient for Fokker 100 wing in cruise condition (M = 0.77).

4. Find eff and deff by means of an airfoil analysis tool, using

leff , V eff and Reeff
5. Let i = +  eff
6. Repeat steps 25 using new values of i and deff until the
solution is converged
7. d = deff cos i + leff sin i
The drag force shown in step 7 has two components. The rst
component is the prole drag. The second component is in fact
the induced drag. However the induced drag can be calculated in
a more accurate way. In the proposed quasi-3D solver, the induced
drag is calculated using Trefftz plane analysis and only the prole
drag (rst term in right hand side of the drag equation in step 7)
is calculated using the 2D airfoil analysis tool. The ability of VLM
(and lifting line methods) for calculation of the induced drag of
a non-planar wing using Trefftz plane analysis is well studied in
the past. The works of Eppler [7], Kroo [19] and Craig and McLean
[2] are examples of such studies, indicating reliable induced drag
estimation. Therefore a VLM code is used in this research for induced drag calculation of wing-winglet combinations. For a swept
wing, the sweep theory [13] is used to translate the forces and
moments in streamwise direction to the direction perpendicular
to the sweep line. The airfoil aerodynamic analysis is executed for
the airfoil perpendicular to the sweep line and the aerodynamic
characteristics of that airfoil are translated again to the streamwise
direction using the sweep theory. Detailed information about this
process is presented in [24].
The solver is validated using a higher delity aerodynamic
solver called MATRICS-V [31]. MATRICS-V is based on fullyconservative full-potential outer ow solver in quasi-simultaneous
interaction with an integral boundary-layer method on the wing.
Ref. [31] compares the pressure distribution on the Fokker 100
wing-body conguration calculated by MATRICS-V with the results
of the wind tunnel tests and ight tests. The comparison shows
very high accuracy of the MATRICS-V code. The validation results
for the quasi-3D solver presented in [24] show an average error
of 2% for subsonic and 3% for transonic regime. The results of the
quasi-3D aerodynamic solver for the Fokker 100 wing is compared
with the MATRICS-V in Figs. 2 and 3.
2.2. Weight model
Quasi-analytical weight estimation methods [3] are very useful
techniques for weight estimation inside an MDO framework. They
can predict the structural weight with the same accuracy as the
FEA based methods but with a much shorter computational time.
The Elham Modied Weight Estimation Technique (EMWET) [5] is

95

Fig. 3. Comparing the drag coecient of a transonic wing calculated by the quasi-3D
solver and the MATRICS-V code.

Fig. 4. Equivalent panels representation of wing-box.

Fig. 5. Upper and lower at panels placed in effective distance.

used for wing structural weight estimation in this research. This

method models the wing-box skin, stringers, spar caps and webs
with equivalent panels (see Fig. 4), and applies elementary structural sizing methods to compute the amount of material required
for those panels to withstand the applied loads. In this approach
the upper and lower curved panels are replaced with two at
panels placed in an effective distance (see Fig. 5). The effective distance is a fraction of the airfoil maximum thickness (t eff = t t max ).
The factor t represents the effect of airfoil shape (panels curvature) on the wing structural weight. The value of t is determined
as a function of the airfoil geometry, wing-box structural parameters (e.g. stringers type, rib pitch) and material properties (e.g.
allowables). Ref. [4] shows that ignoring the effect of airfoil curvature on the wing weight may cause more than 13% error in wing
weight estimation of a Boeing 747 like aircraft. It should be mentioned that 13% of the wing weight of a Boeing 747 is more than
the weight of 50 passengers.
The secondary weights (the weight of aps, slates, etc.) and the
non-optimum weights (the weight penalties due to attachments,
cutouts, etc.) are estimated using a series of empirical equations.
EMWET has been validated using actual weight data of several
passenger aircraft. Table 1 from [5] shows the validation results
of EMWET.
EMWET showed an average error less then 2% for conventional
passenger aircraft made of aluminium alloys. Ref. [5] reports the
application of a power regression to calculate the total wing weight
as a function of the analytically calculated optimum wing box
weight. Fig. 6 from Ref. [5] shows this regression analysis. Results
is the following equation:

96

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

Table 1
Validation results of EMWET.
Aircraft

Maximum take-off
weight (kg)

Error of wing weight

estimation (%)

Fokker 50
Boeing 737-200
Boeing 727-300
Airbus A300-600R
Airbus A330-300
Boeing 777-200

20 820
52 390
95 028
170 500
217 000
242 670

0.72
0.15
2.71
1.86
2.18
2.66

Fig. 7. Winglet geometry and the design variables.

Fig. 6. Correlation between calculated wing-box weight and actual total wing
weight [5].

0.8162
W total = 10.147 W wingbox

R 2 = 0.9982

(1)

3. Winglet shape optimization

Using the mentioned aerodynamic drag and structural weight
estimation methods a series of winglet multi-objective multidisciplinary shape optimizations is applied to nd the optimum winglet
shape for both aerodynamic and weight points of view.
3.1. Winglet multi-objective optimization
In the rst step nding the Pareto front for the wing drag
and the wing structural weight is considered. Two test cases are
used for winglet design. The Fokker 100, a regional passenger aircraft, and the Boeing 747-100, a long range passenger aircraft. The
winglet geometry is dened using 7 variables: root chord C w r ,
taper ratio w , length l w , leading edge sweep angle w , cant
angle , twist (toe out) angle at root  w r and at tip  w t . The
winglet geometry is illustrated in Fig. 7. The airfoil designed by
Whitcomb [34] is used for the winglet (see Fig. 8) and it is kept
constant during the optimization.
A multi-objective optimization is applied with two objective
functions: minimizing the wing drag and minimizing the wing
structural weight. The optimization is formulated as follows:

min
where

F (X)

F=

W wing C d

W wing R and C D R represent the wing weight and wing drag coefcient of the reference aircraft respectively. Although in a winglet
design for an existing wing the amount of increase in the wing
structural weight should be limited to minimize the required modications of the wing structure, no constraint on the wing weight
is dened in this optimization. This decision has been made to determine the Pareto front for the wing drag and the wing structural
weight. The limitations on the allowed increase in wing weight
can be taken into account later for nal selection of the optimum
winglet shape among all winglet designs on the Pareto front.
The Q3D aerodynamic solver is used to calculate the wing drag
in the cruise condition. The aircraft lift coecient is assumed to
be the same as the wing lift coecient (the effect of tail and fuselage is neglected). The cruise lift coecient is determined using
the following equation:

C L cruise =

W wing R C d R

X = [C w r , w , l w , w , ,  w r ,  w t ]
X lower X X upper

Fig. 8. Winglet airfoil designed by Whitcomb.

(2)

MTOW ZFW
qS w

(3)

where ZFW is the aircraft zero fuel weight. The average aircraft
cruise weight, which is between the aircraft MTOW and the aircraft ZFW, is assumed to be equal to MTOW ZFW as suggested

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

97

Table 2
Upper and lower bounds for the design variables.
C wr

lw

wr

wt

Fokker 100

Lower bounds
Upper bounds

0.5C t a
Ct

0.3
1

0.5C t
2C t

25
45

5
90

0.5
5.5

0.5
5.5

Boeing 747

Lower bounds
Upper bounds

0.5C t b
Ct

0.3
1

1m
2.5 m

35
55

5
90

0.5
5.5

0.5
5.5

Winglet design variable

a
b

Wing tip chord of Fokker 100 is 1.26 m.

Wing tip chord of Boeing 747 is 4.06 m.

Fig. 9. The test results of response surfaces for Fokker 100 aircraft.

by Torenbeek [30]. For calculation of the wing weight the aerodynamic loads on the wing are calculated for the maximum load
factor. The related lift coecient is determined using Eq. (4).

C L weight =

nmax MTOW
qS w

(4)

nmax is selected equal to 2.5 for a pull up maneuver. Matlab multiobjective genetic algorithm is used as the optimizer. In order to
reduce the computational time a response surface methodology is
used. The latin hypercube technique [9] is used to generate 400
samples of the winglet geometry between the upper and lower
bounds (see Table 2). In order to ensure that the sample points are
well spaced, the Latin Hypercube sample points are generated using an optimization technique to maximize the minimum distance
between the points. The wing weight and the wing drag of each
sample are calculated using the EMWET and the Q3D solvers. The
Matlab toolbox DACE [22] is used to t response surfaces to the
sample data. This toolbox uses the Kriging method [17] to generate the response surfaces. In order to check the quality of the
generated response surfaces 10 sample winglets are generated randomly. The values of the wing drag and wing weight for the wings
equipped with those winglets are calculated using the response
surfaces and compared to the drag and weight values calculated
using Q-3D aerodynamic solver and EMWET. Fig. 9 shows these
comparisons. The average errors of drag and weight estimation are
0.27% and 0.35% respectively.
The Pareto front for the wing structural weight and the wing
aerodynamic drag for the Fokker 100 aircraft is shown in Fig. 10.
The geometry of the winglet resulting in minimum increase in
the wing weight and the winglet resulting in maximum reduction
in the wing drag are shown in Figs. 11 and 12 respectively. From
these gures one can observe that the minimum increase in weight
is achieved using a winglet with smallest allowed size and cant angle. However the maximum reduction in drag is resulted by using
a tip extension.

Fig. 10. Pareto front for the wing structural weight and the wing drag for Fokker
100 aircraft.

In order to select the best winglets among the ones on the

Pareto front, three gures of merit are used. The aircraft MTOW,
the aircraft fuel weight and the aircraft DOC. The aircraft fuel
weight is calculated using the method presented by Roskam [27].
In this method the Breguet equation is used to estimate the required fuel for the cruise phase of the ight, while a series of
statistical factors are used to estimate the required fuel for the
other ight segments (e.g. takeoff, climb etc.). Aircraft DOC is calculated using the method presented in Ref. [21]. In that method
the aircraft DOC is decomposed into 10 components:
1.
2.
3.
4.

Cost of ight crew

Cost of cabin crew
Landing fees
Navigation fees

98

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

Fig. 11. Winglet with minimum increase in the wing structural weight for Fokker 100 aircraft.

Fig. 12. Winglet with maximum reduction in the wing drag for Fokker 100 aircraft.

5.
6.
7.
8.
9.
10.

Maintenance cost airframe

Maintenance cost Engine
Fuel cost
Depreciation cost aircraft and spares
Insurance
Interest

Adding winglets to the aircraft wing changes the aircraft fuel

weight and also the aircraft structural weight. The fuel weight affects the fuel cost. However the aircraft structural weight affects
landing fee, navigation fee, airframe maintenance cost, depreciation cost, insurance and interest costs. The costs related to the
ight and cabin crew are assumed to be constant. All costs related to engines are neglected in this research, because they are
assumed to be independent of the winglet shape.
Fig. 13 shows the positions of the winglets resulting in minimum MTOW, minimum fuel weight and minimum DOC on the
Pareto front for the Fokker 100 aircraft. The geometry of those
winglets is shown in Figs. 14 to 16. From Fig. 13 one can observe that the position of the winglet resulting in minimum MTOW

Fig. 13. Location of the winglets result in minimum fuel weight and minimum DOC
for Fokker 100 aircraft.

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

is very close to the position of the winglet resulting in minimum wing weight. The same gure also shows that the position
of the winglet resulting in minimum fuel weight is very close
to the position of the winglet resulting in minimum drag. However for achieving minimum DOC, a winglet between those two
winglets corresponding to minimum MTOW and minimum fuel
weight should be used.
Table 3 summarizes the characteristics of the wings with ve
different winglets resulting in minimum wing structural weight,
minimum wing drag, minimum MTOW, minimum fuel weight and
minimum DOC. The geometrical parameters of those winglets are
summarized in Table 4.
Table 3 shows, that although the reduction in the wing total
drag is mainly achieved by reducing the induced drag, winglets
also signicantly reduce the wing pressure drag (mainly the wave
drag). The effect of winglet on the wing wave drag will be discussed later in this paper.
From Table 3 one can conclude that despite the high amount
of reduction in the wing drag achieved by using winglets (about

99

7%), the maximum amount of reduction in fuel weight is only

slightly above 2% (for the winglet with minimum fuel weight in
Table 3). Although 2% reduction in the fuel consumption of that
aircraft results in about 2.25M$ reduction in fuel cost in a 15 years
life cycle of that aircraft (assuming an aircraft utilization of 3750
hours/year [15]), taking into account the effect of the weight increase on the other components of the aircraft DOC, reduces the
amount of lift cycle cost saving to 660,000$ (see Fig. 17). This can
be explained by comparing the aircraft fuel weight with the wing
structural weight. The Fokker 100 aircraft is a mid-range narrow
body aircraft. The design fuel weight of this aircraft is about 20%
of the aircraft MTOW. In the case of maximum payload weight, the
fuel weight is limited to 17% of the MTOW. On the other hand the
wing structural weight is about 11% of the aircraft MTOW. Compared to a long range wide body aircraft (like Boeing 747), where
the design fuel weight is about 40% of the aircraft MTOW, while
the wing weight is still about 11% of the MTOW, it is expected that
the winglet is more effective in cost reduction for a long-range aircraft.

Fig. 14. Winglet results in minimum MTOW.

Fig. 15. Winglet results in minimum fuel weight.

100

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

Fig. 16. Winglet results in minimum DOC.

Table 3
Characteristics of the wings with various winglets for Fokker 100.

Initial wing
Minimum wing weight
Minimum wing drag
Minimum MTOW
Minimum fuel weight
Minimum DOC

CD

C Di

CDp

CD f

W wing

W fuel

MTOW

DOC

1
0.9878
0.9296
0.9845
0.9314
0.9596

1
0.9765
0.8452
0.9700
0.8491
0.9015

1
0.9894
0.9262
0.9867
0.9329
0.9669

1
1.0025
1.0580
1.0052
1.0486
1.0213

1
1.0140
1.1067
1.0140
1.1018
1.0479

1
0.9955
0.9805
0.9945
0.9798
0.9850

1
1.0007
1.0075
1.0006
1.0071
1.0023

1
0.9991
0.9985
0.9988
0.9980
0.9971

Table 4
Geometry of various winglets for Fokker 100.

Minimum
Minimum
Minimum
Minimum
Minimum

wing weight
wing drag
MTOW
fuel weight
DOC

C w r [m]

l w [m]

w [deg]

[deg]

 w r [deg]

 w t [deg]

0.76
1.15
0.67
1.08
0.72

0.45
0.32
0.31
0.30
0.31

0.87
2.52
0.80
2.52
2.25

21.39
15.26
29.37
16.29
15.56

8.84
79.55
23.77
66.99
36.64

2.69
1.14
2.86
1.25
1.57

5.11
1.73
2.46
1.88
1.13

Fig. 17. Effect of winglet on various components of DOC for the Fokker 100 winglet resulting in minimum fuel weight.

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

101

Fig. 18. The test results of response surfaces for Boeing 747 aircraft.

Fig. 19. Pareto front for winglet design for Boeing 747 aircraft.

In order to investigate the effect of winglets on the performance

of a long-range wide body aircraft, the same optimization formulated in Eq. (2) is used to nd the Pareto front for the wing weight
and the wing drag for Boeing 747-100 aircraft wing. Similar to the
Fokker 100 test case, two response surfaces are generated for wing
drag and wing weight. The validation results of those response surfaces are shown in Fig. 18. The error of wing drag and wing weight
estimation using those generated response surfaces is 0.36% and
0.26% respectively.
This Pareto front is shown in Fig. 19. The positions of winglets
resulting in minimum MTOW, minimum fuel weight and minimum
DOC are shown in Fig. 20.
The characteristics of the wings with winglets resulting in minimum wing weight, minimum wing drag, minimum MTOW, minimum fuel weight and minimum DOC are compared to the initial
wing (without winglet) in Table 5. The geometrical parameters
of those winglets are summarized in Table 6. From Table 5 one
can observe that for almost the same amount of reduction in the
wing drag as for the Fokker 100 test case, a higher amount of reduction in the aircraft fuel weight and DOC are achieved for the
Boeing 747.
From Table 5 it can be observed that the winglets resulting in
minimum fuel weight and minimum DOC are very similar to each
other and they are close to the winglet resulting in minimum drag.
The winglet resulting in minimum MTOW has almost equal dis-

Fig. 20. Positions of the winglets result in minimum fuel weight and minimum DOC
for Boeing 747 aircraft.

tance to the winglet resulting in minimum drag and the winglet

resulting in minimum structural weight. About 3.8% reduction in
the aircraft fuel weight is achieved by more than 6% reduction
in the wing total drag and about 8% increase in the wing structural weight. These weight reductions lead to almost half a percent
reduction in the aircraft MTOW. 0.5% reduction in MTOW of the
Boeing 747 aircraft is equal to 1700 kg, which is about the weight
of 19 passengers. The same winglet results in 1.7% reduction in
the aircraft DOC. This amount of reduction means about 29M$ cost
saving in a 15 years life cycle of the aircraft (assuming an utilization of 4800 hours/year [15]). All cost estimations in this research
are performed using the fuel price equal to 3$ per gallon. The geometry of the wings with winglets listed in Table 5 are shown in
Figs. 21 to 25. Table 6 summarizes the geometrical parameters of
those winglets.
3.2. Multi-objective winglet shape optimization with xed cant angles
In Section 3.1 the winglet cant angle is used as a design variable. However there is sometimes limitation on the wing span
(specially for a large aircraft like Boeing 747). Therefore in some
cases the winglet cant angle should be limited to avoid a large increase in the wing span.

102

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

Table 5
Characteristics of the wings with various winglets for Boeing 747.

Initial wing
Minimum wing weight
Minimum wing drag
Minimum MTOW
Minimum fuel weight
Minimum DOC

CD

C Di

CDp

CD f

W wing

W fuel

MTOW

DOC

1
0.9799
0.9364
0.9511
0.9376
0.9381

1
0.9668
0.8875
0.9083
0.8907
0.8916

1
0.9851
0.9260
0.9579
0.9334
0.9346

1
1.0026
1.0337
1.0116
1.0219
1.0203

1
1.0244
1.1031
1.0481
1.0824
1.0800

1
0.9900
0.9659
0.9695
0.9623
0.9624

1
0.9991
0.9996
0.9930
0.9952
0.9946

1
0.9956
0.9861
0.9852
0.9830
0.9829

Table 6
Geometry of various winglets for Boeing 747.

Minimum
Minimum
Minimum
Minimum
Minimum

wing weight
wing drag
MTOW
fuel weight
DOC

C w r [m]

l w [m]

w [deg]

[deg]

 w r [deg]

 w t [deg]

2.10
3.61
2.04
2.42
2.83

0.54
0.64
0.47
0.51
0.54

1.02
2.50
2.43
2.49
2.50

42.87
43.25
43.44
43.37
55.00

17.64
87.70
37.45
73.20
55.98

4.32
2.95
4.44
2.76
3.95

2.85
5.16
2.74
4.92
0.95

Fig. 21. Geometry of the wing with winglet results in minimum wing weight for Boeing 747 aircraft.

Fig. 22. Geometry of the wing with winglet results in minimum drag for Boeing 747 aircraft.

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

Fig. 23. Geometry of the wing with winglet results in minimum MTOW for Boeing 747 aircraft.

Fig. 24. Geometry of the wing with winglet results in minimum fuel weight for Boeing 747 aircraft.

Fig. 25. Geometry of the wing with winglet results in minimum DOC weight for Boeing 747 aircraft.

103

104

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

Fig. 26. Effect of cant angle.

In order to investigate the effect of the winglet cant angle on

the wing drag and the wing structural weight, three optimizations
are applied for winglet shape optimization with three different
xed cant angles: 0 (vertical winglet), 15 and 90 (tip extension).
The Pareto fronts for the wing drag and the wing structural weight
for Boeing 747 aircraft are shown in Fig. 26. This gure shows that,
for achieving maximum reduction in the wing drag without any
constraint on the wing weight, a tip extension is a better choice
than a winglet. However when the amount of increase in the wing
weight is limited, which is normally the case specially for add-on
winglets, the choice of tip extension vs winglet depends on the
amount of the allowed increase in the wing weight. From Fig. 26
it can be observed that for a weight increase less then about 5.5%
a winglet is a better choice than a tip extension.
If there is a constraint on the wing span, which can be the
case for such a large aircraft as Boeing 747, then winglets with
lower cant angles are better choices. Comparing winglets with 0
degree cant angle (vertical winglet) and winglets with 15 degrees
cant angle, it can be found that, for a wing weight increase less
than 6% a vertical winglet performs better. However if higher increase in the wing weight is allowed, a choice of 15 degrees canted
winglet seems to be better. It should be noted that the interference drag (mainly wave drag) is not taken into account. The
winglet cant angle has a signicant effect of the interference drag
between the wing tip and the winglet in high speeds. Taking into
account that effect, a winglet with higher cant angle may be a better choice [28].
3.3. Winglet optimization with constraint on lift coecient
Fig. 27 shows the lift distribution on the initial wing of the
Boeing 747 aircraft and the wing with winglet resulting in minimum fuel weight. This gure shows, that the optimizer has tried
to increase the lift produced by the winglet to reduce the lift coefcient of the main wing (up to 6% reduction in the local C l in the
inboard part of wing is resulted), which results in lower wave drag
in transonic speeds (see Section 4.1).
From Fig. 27 one can observe that the winglets are operating under very high values of lift coecients in cruise condition.
These values of C l seem unrealistic. These high values of the lift
coecient may cause buffet problems. Besides, in a pull up maneuver (see Fig. 28) the lift coecient of the winglet exceeds the
maximum lift coecient of the winglet airfoil, which results in a
winglet stall.
In addition to the mentioned problems associated with high C l
on the winglet, the interference drag between the main wing and

the winglet (mainly wave drag at high speeds) can be very high
if the winglets produce very high amount of super velocities on
their suction side as a result of working at high C l values. Although
the interference drag can be controlled using a carefully designed
blended winglet, it is desired to prevent the winglets operating at
such high values of C l .
A single objective optimization has been performed to optimize
the winglet shape for minimum fuel weight, while keeping the
cruise C l and the pull up C l values of the wing and winglet lower
than certain amount of lift coecients. The wing should be designed in a way to avoid buffet up to C l value 1.3 times of the
cruise C l . Based on this requirement the cruise lift coecient is
assumed to be less than 0.5. The maximum lift coecient of the
airfoil sections of the main wing and the winglet is assumed to
be 1.6. The winglet resulting in minimum fuel weight found by the
multi-objective optimization is selected as the starting point. The
wing weight of the new design is constrained to be equal or less
than the weight of the wing used as the starting point (which is
equal to 1.0824W wing R , see Table 5).
The SNOPT optimization algorithm [10] is used as the optimizer. The optimization is formulated as follows:

min

F (X) =

W fuel
W fuel R

X = [C w r , w , l w , w , ,  w r ,  w t ]
C L = C L des
C lcruise 0.5
C lpullup 1.6
W wing 1.0824W wing R
X lower X X upper

(5)

Tables 7 and 8 compare the wing characteristics and the geometry of the winglet optimized with constraints on the lift coefcient with the one optimized without those constraints. Table 7
shows that the wing with winglet optimized with constraints on
C l has slightly higher drag that the wing with winglet optimized
without any constraint on C l , which results in a slightly higher fuel
weight. The lift distribution on these mentioned wings are plotted
in Figs. 29 and 30. The geometry of the optimized winglet is shown
in Fig. 31.
4. Parametric studies
4.1. Effect of winglet on various drag components
The main reason behind the drag reduction achieved by using
a winglet is the effect of a winglet on induced drag. However the
winglet affects other drag components as well. Adding a winglet
to a wing increases the wing friction drag, because of the increase
in the wing wetted area. Winglet also affects the wing pressure
drag (mainly the wave drag) by changing the lift distribution on
the wing.
In order to investigate the effect of winglet on various drag
components, the wing of Boeing 747-100 aircraft is considered as a
test case. Variation of different drag components of that wing with
Mach number is calculated using the Q3D aerodynamic solver.
Then the wing is equipped with the winglet optimized for minimum fuel weight with constraints on C l (see Table 8) and the drag
calculation is repeated for the new wing.
Figs. 32 and 33 show the variation of the wing total drag and
the various drag components with Mach number for the Boeing 747 wing with winglet compared to the wing without winglet.
In this research the NACA 6 series airfoils with 13% thickness to
chord ratio at root and 8% at tip are used for the Boeing 747 wing.

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

105

Fig. 27. Lift distribution on the Boeing 747 wings with and without winglet in aircraft cruise condition.

Fig. 28. Lift distribution on the Boeing 747 wings with and without winglet in pull up maneuver.

Table 7
Characteristics of the wing with winglet optimized using constraints on C l .

Initial wing
Without constraint on C l
With constraint on C l

CD

C Di

CDp

CD f

W wing

W fuel

DOC

1
0.9376
0.9438

1
0.8907
0.8988

1
0.9334
0.9374

1
1.0219
1.0210

1
1.0824
1.0808

1
0.9623
0.9666

1
0.9838
0.9833

Table 8
Geometrical parameters of the winglet optimized using constraints on C l .

Without constraint on C l
With constraint on C l

C w r [m]

l w [m]

w [deg]

[deg]

 w r [deg]

 w t [deg]

2.42
2.83

0.51
0.54

2.49
2.50

43.37
55.00

73.20
55.98

2.76
3.95

4.92
0.95

106

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

Fig. 29. Lift distribution on the optimized wing and winglet of Boeing 747 in aircraft cruise condition.

Fig. 30. Lift distribution on the optimized wings with and without winglet of Boeing 747 in pull up maneuver.

Fig. 31. Geometry of the winglet optimized with constraint on C l .

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

107

Fig. 32. Effect of winglet on drag variation with Mach number.

Fig. 34. The amount of drag reduction or increase for each drag component resulted
by using winglet.

Fig. 33. Effect of winglet on variation of various drag components with Mach number.

Therefore the values of the drag coecients and the drag divergence Mach number presented in this paper can be different from
the actual values of the Boeing 747 wing with airfoils different
from the ones used for these analyses.
The wave drag appears after Mach number about 0.77 for
both wings (the wing with winglet and the one without winglet).
The drag divergence Mach number of both wings is calculated
to be about 0.82. This value is calculated based on the standard denition of the drag divergence Mach number, which deD
nes M dd as the Mach number in which dC
= 0.1 or C D =
dM
0.002.
The amount of drag reduction (or increase) achieved by using
a winglet is shown in Fig. 34 for various Mach numbers. From
Fig. 34 one can observe that the amount of reduction in the induced drag is almost constant with Mach number. The amount
of increase in friction drag slightly increases by increasing the
Mach number. However the maximum change with Mach number is found for the pressure drag. At low subsonic speeds, where
there is no wave drag, the presence of a winglet results in about
5% increase in the pressure drag. However this amount of change is
negligible because in those speeds the pressure drag is the smallest component of the wing total drag. After M = 0.77, where the
wave drag appears, in contrast to lower Mach numbers, a winglet
results in a reduction in the pressure drag. The amount of this
drag reduction increases rapidly by increasing the Mach number,

see Fig. 34. This drag reduction is due to the reduction in the
wave drag, which is a very important component of the pressure drag at transonic speeds. Fig. 29a shows the lift distribution
on the wings with and without winglet. From that gure it can
be observed that most of the inboard part of the wing with a
winglet is working at lower values of the lift coecient compared to the initial wing without winglet, which results in less
wave drag. Before M dd the pressure drag is a small fraction of
the total drag (20% of the total drag comparing to 47% for C D i
and 33% for C D f ). However after M dd it rapidly increases (40%
of the total drag at M = 0.85 compared to 35% for C D i and 25%
for C D f ). The cruise Mach number of the passenger aircraft is always lower than the drag divergence Mach number, therefore the
effect of winglet on the wave drag reduction is less important
compared to the effect of winglet on the reduction of the induced
drag. However the presence of a winglet can extend the aircraft
operating envelope by reducing the wave drag at dive Mach number or maximum operating Mach number, where the wave drag
can be much higher than the wave drag at the cruise Mach number.
4.2. Effect of winglet on the wing structural sizing loads
As mentioned in the introduction part of this paper, most of
the researchers use the wing root bending moment as an indicator
of the wing structural weight for a winglet design. In many cases
in literature the winglet has been optimized using a constraint on
the root bending moment. However this constraint is not a good
indicator of the wing weight, since the root bending moment of a
wing with a winglet can be the same as the root bending moment
of that wing without winglet, while the change in the bending moment distribution may cause some weight penalties. Figs. 35 to 37
show the load distribution on the Boeing 747 initial wing and the
wing equipped with the winglet resulted from the optimizating in
Section 3.3.
Fig. 35 shows that the wing with winglet produces higher lift in
the outboard sections compared to the initial wing, while lower lift
in the inboard sections. This results in a higher shear force compared to the initial wing in the outboard sections (after 50% of the
semi-span, see Fig. 36) and lower shear force in the inboard sections (before 50% of the semi-span, see Fig. 36). This kind of shear
force distribution results in a spanwise bending moment distribution, in which the ratio of the bending moment of the wing with
winglet to the bending moment of the initial wing is very close
to one in the inboard wing sections close to the wing root, while

108

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

Fig. 35. Spanwise lift distribution in a pull up maneuver.

Fig. 38. Ratio of the shear force of the wing with winglet to the shear force of the
reference wing.

Fig. 36. Spanwise shear force distribution in a pull up maneuver.

Fig. 39. Ratio of the bending moment of the wing with winglet to the bending
moment of the reference wing.

The root bending moment of the wing with winglet is just

1.0014 times of the root bending moment of the reference (initial) wing, while the total wing weight of that wing is more than
8% higher than the wing weight of the initial wing. This simple
analysis shows that, using a constraint on the wing root bending
moment is not enough to control the wing structural weight in a
winglet design and optimization.
5. Conclusions

Fig. 37. Spanwise bending moment distribution in a pull up maneuver.

much higher than one in the outboard wing sections. The ratio of
the shear force and bending moment of the wing with winglet to
those ones for the initial wing is plotted in Figs. 38 and 39 as a
function of the wing spanwise position.

The use of optimization techniques for design of winglets for

existing aircraft designs is showed in this paper. A series of multiobjective and single-objective optimizations has been applied to
optimize the winglet shape for a mid-range and a long-range passenger aircraft. The outer shape of the wings are kept xed but
the internal structure and associated weight is adjusted based on
the change of spanwise loading (magnitude and distribution) due
to the addition of the winglet. The results of the optimizations
show that, although the amount of drag reduction achieved by using winglets is almost the same for both the test case aircraft, the
amount of fuel saving resulting from the application of a winglet
is more considerable for a wide body long range aircraft. The main
reason is the difference between the ratio of the fuel weight to the
MTOW and the wing structural weight to the MTOW. In a narrowbody short to medium range aircraft those ratios are close to each

A. Elham, M.J.L. van Tooren / Aerospace Science and Technology 37 (2014) 93109

other, however in a wide-body long range aircraft, the fuel weight

is much higher than the wing structural weight.
The cost analysis showed that, although reducing the fuel
weight by using winglets will result in lower fuel cost, the increased wing weight due to the presence of a winglet increases
some other components of the aircraft DOC. This increase may
compensate the effect of winglet on the fuel cost reduction for
small aircraft, where the fuel weight is in the same order of magnitude as the wing structural weight. However for long range passenger aircraft, the use of winglets may result in a DOC reduction
about 2%.
A quasi-three dimensional aerodynamic solver is used to calculate the wing drag. Using this solver various drag components can
be calculated separately. Therefore the effect of winglets on different drag components was investigated. This study showed that
the winglet can result in a high amount of wave drag reduction
at high speeds by reducing the lift coecient on the wing inboard
sections.
In this research a quasi-analytical wing weight estimation
method has been used to estimate the wing structural weight.
Instead of using constraints on the wing root bending moment,
which is a popular approach in winglet optimization, the winglet
shape is optimized directly for minimum wing structural weight.
This research has showed that using the maximum root bending
moment as indicator of the wing total structural weight is not
a good approach for winglet multidisciplinary design optimization. An example was provided to show that although by a careful
winglet design the root bending moment can be very close or even
equal to the root bending moment of the wing without winglet,
the changes in the bending moment distribution may result in a
non-negligible amount of weight increase.
Eventually some recommendations are provided for future research. As shown in Fig. 7, there is a hard corner between the
wing and winglet. In terms of interference drag this seems undesirable. A design resembling a blended winglets geometry is
recommended for further research. The second consideration in
a winglet design is the effect of a winglet on the wing utter.
Because of the winglet placement largely behind the wings elastic axis, wing-winglet combinations can be prone to utter. This
sometimes necessitates the use of a ballast weight. In this research
this weight increase is neglected. However by using a higher delity aeroelastic model, the effect of winglet on the wing aeroelasticity can be investigated.
Conict of interest statement
The authors declare that no conict of interest is involved in
this research.
References
[1] A. Bscher, R. Radespiel, T. Streit, Modelling and design of wing tip devices
at various ight conditions using a databased aerodynamic prediction tool,
Aerosp. Sci. Technol. 10 (8) (2006) 668678.
[2] A. Craig, J.D. McLean, Spanload optimization for strength designed lifting surfaces, AlAA Paper 88-2512, 1988.
[3] A. Elham, Structural Design: Weight Estimation, in: Encyclopedia of Aerospace
Engineering, John Wiley & Sons Ltd., June 2012.
[4] A. Elham, G. La Rocca, R. Vos, Rened preliminary weight estimation tool for
airplane wing and tail, SAE Technical paper No. 2011-01-2765, 2011.

109

[5] A. Elham, G. La Rocca, M. van Tooren, Development and implementation of

an advanced, design-sensitive method for wing weight estimation, Aerosp. Sci.
Technol. 29 (2013) 100113.
[6] Engineering Science Data Unit (ESDU), 98013, Aerodynamic Principles of
Winglets, June 1998.
[7] R. Eppler, Induced drag and winglets, Aerosp. Sci. Technol. 1 (1997) 315.
[8] G.A. Flandro, H.M. McMahon, R.L. Roach, Basic Aerodynamics: Incompressible Flow, Cambridge Aerospace Series, Cambridge University Press, Cambridge,
New York, 2012.
[9] A.I.J. Forrester, A. Sobester, A.J. Keane, Engineering Design via Surrogate Modeling: A Practical Guide, J. Wiley, Chichester, West Sussex, England, Hoboken, NJ,
2008.
[10] P. Gill, W. Murray, M. Saunders, SNOPT: an SQP algorithm for large-scale constrained optimization, SIAM Rev. 47 (1) (2005) 99131.
[11] J.-L. Hantrais-Gervois, R. Grenon, A. Mann, A. Bscher, Downward pointing
winglet design and assessment within the M-DAW research project, Aeronaut.
J. 113 (1142) (2009) 221232.
[12] H. Heyson, G. Riebe, C. Fulton, Theoretical parametric study of the relative advantages of winglets and wing-tip extensions, NASA TP-1020, September 1977.
[13] D.R. Holt, Introduction to transonic aerodynamics of aerofoils and wings, Tech.
rep., ESDU (Engineering Sciences Data Unit) 90008, April 1990.
[14] P. Jansen, R. Perez, J. Martins, Aerostructural optimization of nonplanar lifting
surfaces, J. Aircr. 47 (5) (2010) 14901503.
[15] L.R. Jenkinson, P. Simpkin, D. Rhodes, Civil Jet Aircraft Design, AIAA Education
Series, American Institute of Aeronautics and Astronautics, Reston, VA, 1999.
[16] R. Jones, T. Lasinsky, Effect of winglets on the induced drag of ideal wing
shapes, NASA TM-81230, September 1980.
[17] D. Jones, M. Schonlau, W. Welch, Ecient global optimization of expensive
black-box functions, J. Glob. Optim. 13 (4) (1998) 455492.
[18] R. Kelm, M. Lapple, M. Grabietz, Wing primary weight estimation of transport
aircrafts in the pre-development phase, SAWE Paper No. 2283, May 1995.
[19] I. Kroo, A general approach to multiple lifting surface design and analysis, AlAA
Paper 84-2507, 1984.
[20] I. Kroo, Nonplannar wing concept for increased aircraft eciency, in: Innovative Congurations and Advanced Concepts for Future Civil Aircraft, in: VKI
Lecture Series, von Karman Institute for Fluid Dynamics, Rhode-St-Genese, Belgium, June 2005.
[21] R.H. Liebeck, et al., Advanced subsonic airplane design & economic studies,
NASA CR-195443, April 1995.
[22] S. Lophaven, H.B. Nielsen, J. Sondergaard, DACE, A Matlab kriging toolbox, Technical report IMM-TR-2002-12, Information and Mathematical Modeling Institute, Technical University of Denmark, August 2002.
[23] S.H. Macci, Semi-analytical method for predicting wing structural mass, SAWE
Paper No. 2282, May 1995.
[24] J. Mariens, A. Elham, M. van Tooren, Quasi-three-dimensional aerodynamic
solver for multidisciplinary design and optimization of lifting surfaces, J. Aircr.
51 (2) (2014) 547558.
[25] M. Mheut, D. Bailly, Drag-breakdown methods from wake measurements,
AIAA J. 46 (4) (2008) 847862.
[26] S. Ning, I. Kroo, Multidisciplinary considerations in the design of wings and
wing tip devices, J. Aircr. 47 (2) (2010) 534543.
[27] J. Roskam, Airplane Design, DAR corporation, Lawrence, Kan, 1986.
[28] K. Takenaka, K. Hatanaka, W. Yamazaki, K. Nakahashi, Multidisciplinary design
exploration for a winglet, J. Aircr. 45 (5) (SeptemberOctober 2008) 16011611.
[29] E. Torenbeek, Development and application of a comprehensive, design sensitive weight prediction method for wing structures of transport category
aircraft, Report LR-693, Delft University of Technology, Delft (Netherlands),
September 1992.
[30] E. Torenbeek, Advanced Aircraft Design, Conceptual Design, Analysis and Optimization of Subsonic Civil Airplanes, John Wiley & Sons Ltd, West Sussex, UK,
2013, p. 410.
[31] A. van der Wees, J. van Muijden, J. van der Vooren, A Fast and Robust ViscousInviscid Interaction Solver for Transonic Flow about Wing/Body Congurations
on the Basis of Full Potential Theory, AIAA-93-3026, in: AIAA 24th Fluid Dynamics Conference, July 69, 1996.
[32] J. Verstraeten, R. Slingerland, Drag characteristics for optimally span-loaded
planar, wingletted, and C wings, J. Aircr. 46 (3) (2009) 962971.
[33] J. Weierman, J. Jacob, Winglet design and optimization for UAVs, AIAA Paper
No. 2010-4224, 2010.
[34] R. Whitcomb, A design approach and selected wind tunnel results at high subsonic mounted speeds for winglets, NASA TN-D-8260, July 1976.