Original Article

Proc IMechE Part G:

J Aerospace Engineering
Aerodynamic design optimization 2018, Vol. 232(6) 1047–1062
! IMechE 2017

for low Reynolds tandem airfoil Reprints and permissions:

[Link]/[Link]
DOI: 10.1177/0954410017704219
[Link]/home/pig

Fangzheng Chen, Jianqiao Yu and Yuesong Mei

Abstract
A series of aerodynamic design optimization studies are performed to investigate the performance of two-dimensional
tandem airfoil at low Reynolds number of 220,000. A total of 23 design variables, including decalage, gap, stagger, and
airfoil profile variables are considered. Hierarchical evolutionary algorithm together with dynamic metamodel optimiza-
tion is used to achieve the maximum lift-to-drag ratio for tandem airfoil. The studies investigate the impact of various
design variables on optimized tandem airfoil. The results show that tandem airfoil in lower Reynolds flow has better
aerodynamic performance at high angle of attack. And the lift-to-drag ratio can be further increased by 11.9% when airfoil
profile variables are introduced. Morphing tandem airfoil can overcome low lift-to-drag ratio at lower lift coefficient, and
is attractive for unmanned aircraft.

Keywords
Low Reynolds flow, tandem airfoil, aerodynamic optimization, laminar separation bubble, metamodel

Date received: 6 December 2016; accepted: 20 March 2017

Introduction
methods and analyzed the NACA 62-215
With the advancement of unmanned aerial vehicles and NACA0012 airfoils. He analytically confirmed
(UAVs) technology, many new aerodynamic the aerodynamic improvements measured by
configurations have been adopted to improve the Nenadovitch, and discussed the mechanisms causing
aerodynamic performance of UAVs at low Reynolds the reduction of drag. Rhodes7 performed two- and
number. Closely coupled tandem airfoil systems have three-dimensional numerical analysis to investigate
shown aerodynamic advantages over single-wing optimal dual wing configurations. And ‘‘baseline’’
counterpart in low Reynolds flow, in which tandem business airplanes with these configurations were
airfoil aircrafts exhibits stronger maneuverability, designed and compared. His results showed that the
longer endurance and higher fuel efficiency.1,2 closely coupled dual wing aircraft has lower cruise
The study for tandem airfoil started as early as in drag and engine weight. Rokhsaz and Bruce8,9
1930s, with Nenadovitch being the first one to dis- improved the tandem airfoil prediction algorithms
cover improved two-dimensional characteristic.3 by adding compressibility effects, and they studied
And it became research hotspot from 1970s to the optimal spanwise distribution of the decalage
1990s, aiming to improve the performance of angle for tandem airfoil. Their results showed a fur-
manned aircrafts, for which the Reynolds number is ther improvement on aerodynamic performance for
usually greater than 106. Several researchers have ‘‘baseline’’ airplane.
investigated the tandem airfoil system in this condi- In recent investigations, the research has focused
tion. Olson and Selberg4 compared monoplane and on tandem airfoil aerodynamic characteristic in low
dual wing plane by performing wind tunnel experi- Reynolds flow. Daniel and Thomas10 performed
ments with three-dimensional models. Their results experiments with two identical Wortmann Fx63-137
showed that the dual wing configurations could
achieve substantially higher lift-to-drag ratio than School of Aerospace Engineering, Beijing Institute of Technology,
single wing designs. Addoms and Spaid5 performed Beijing, China
the aerodynamic analysis of high-performance dual
Corresponding author:
wings and their study claimed that the biplane can
Jianqiao Yu, School of Aerospace Engineering, Beijing Institute of
have superior low-speed maneuverability and short Technology, 5 South Zhongguancun Street, Haidian District, Beijing
field performance. Rokhsaz6 performed a numerical 100081, China.
two-dimensional analysis by using vortex panel Email: jianqiao@[Link]
1048 Proc IMechE Part G: J Aerospace Engineering 232(6)

airfoils in closely coupled tandem configurations at

low Reynolds number. His results showed that some
Methodology
advantages over a single-wing design could be gained This section describes the numerical tools used in this
by utilizing a closely coupled tandem airfoil arrange- paper. These tools, which contain geometric paramet-
ment. Jones et al.11 used wind tunnel to study the rization, computational fluid dynamics (CFD)
aerodynamic of biplane and tandem airfoil at low method and optimization algorithm, are the compo-
Reynolds numbers. Their measurements showed that nents of aerodynamic optimization.
the two-wing configuration was a viable method to
overcome the challenges of low Reynolds number
flight. Aerodynamic benefits were observed in the
Geometric parameterization and grid modification
post-stall regime of the single wing. Bramesfeld2 In this paper, we use the Sculptor software to per-
compared the flight performance of biplane and form the geometric parameterization and grid modi-
monoplane, and his results showed that under limit fication of tandem airfoil. The Sculptor is based on
wingspans and restricted maneuvering, biplane con- the so-called arbitrary shape design (ASD),12,13
figurations have minimum power requirements that which allows the user to perform shape optimization
potentially lead to 30% longer endurance than for the CFD design. The ASD volume parameterizes
single-wing counterpart. Aerodynamic characteristics the changes of the embedded geometry rather than
of tandem airfoil in low Reynolds number flow were the geometry itself, resulting in a more efficient and
attractive. Although many studies have been done in compact set of geometry design variables, and
this filed, optimization is not performed in low making it easier to manipulate complex geometries.
Reynolds number flow condition and airfoil profile All geometric transformations are performed by
is not considered as optimal design variables. changing control points. In the optimization pro-
The objective of this study is to investigate the opti- gress, the software performs a smooth free-form
mal configurations and aerodynamic characteristics of deformation (FFD)14 of the geometry model based
tandem airfoil in low Reynolds flow. We perform a on the change of control points and automatically
series of optimal designs that not only consider deca- performs CFD grid modification, which eliminates
lage, gap, and stagger, but also the airfoil profile for the extremely costly process of having to remesh
both canard and wing. This paper extends the prelim- the grid for every shape change. The key assumption
inary researches of tandem airfoil optimization to low of the ASD approach is that the geometry has con-
Reynolds number. The results show the benefits of stant topology throughout the optimization process,
considering airfoil as design variables. Morphing which is usually the case for aerodynamic design
tandem airfoil conceptual design is presented to min- optimization.
imize tandem airfoil drag in larger lift coefficient Figure 1 shows the ASD volume and the geometric
range. This paper is organized as follows. In the control points for the aerodynamic shape optimiza-
upcoming section, the numerical methods used in tion for the tandem airfoil configuration. The whole
this work are described. The optimization formula- tandem airfoil is enveloped by 126 ASD volumetric
tion, mesh, and the baseline geometry are discussed blocks. Each airfoil profile is parameterized by 10
later. A series of aerodynamic design optimization control groups, which can change the profile in a dir-
cases are presented and discussed subsequently. At ection perpendicular to the chord. We implement the
last, the conclusions are presented in the last section. variable of decalage by rotating a sub-ASD block that

Figure 1. FFD volume and control points.

Chen et al. 1049

surrounds the upward wing, as defined in Figure 3. region criterion.19 In this paper we use the
The axis for the rotation of the wing is at the Kriging model, which is provided by Design and
1/4-chord length and is normal to the plane. The vari- Analysis of Computer Experiments (DACE,
ables of gap and stagger are implemented by moving Version 2.0) Matlab toolbox, as dynamic metamo-
the ASD block around the downward wing separately del. For the Kriging model proposed in this
along the y and x coordinates. paper, the ‘‘regression models’’ is zero order poly-
nomial, the ‘‘correlation models’’ is Gaussian model,
and the data is normalized between 0 and 1. In the
CFD method optimization progress, the application of GA is to
We use ANSYS 14.5 Fluent to solve the CFD prob- find the optimum solution of metamodel, so as to
lem. As the flow is at low Reynolds number, the air is ensure the convergence of the whole solution pro-
almost incompressible but the viscosity cannot be cess. In our proposed optimization algorithm, the
ignored. We use the steady Navier–Stokes model genetic algorithm is provided by matlab 2012b opti-
coupled with the shear-stress transport (SST) -Re mization toolbox. In the optimization progress, we
model (named transition SST turbulence model in set the ‘‘population size’’ as 50. ‘‘Stopping criteria’’
ANSYS) to solve the flow. The SST-Re model cou- option is 300, and the other parameters remain
ples the SST turbulent model with transport equations default. The flow chart of optimization loop is
for the intermittency and momentum-thickness shown on the right side of Figure 2.
Reynolds number. The turbulence model only relies Hierarchical evolutionary algorithm20 is a com-
on local parameters, which makes it suitable for monly used method for complex optimization prob-
modern CFD methods and complicated transitional- lems with large number of design variables. It divides
flow simulations. The SST-Re model has been used these design variables to several groups to improve the
to predict laminar separation bubbles (LSB) by accuracy of metamodel and can reduce the coupling
researchers15,16 and it shows good performances. among parameters. In this study, the design variables
In this paper, we keep the default settings for the are divided into three groups. Decalage, gap, and
transition SST model. As heat exchange is not con- stagger, which are the main factors affecting the per-
sidered in the flow, energy equations are turned off. formance of the tandem airfoil, are the first design
The boundary is set as velocity inlet and pressure group. The 10 canard shape control variables are
outlet. The inlet turbulent intensity is set 0.08%, the second group, and the 10 wing shape variables
depending on the simulation result17 and wing are the last design group. As shown in nearside of
tunnel experiments10 and the inlet viscosity ratio Figure 2, for each group optimization, the other two
is set as 5. All equations are solved with steady design groups remain unchanged. And in the follow
pressure-based coupled method, and the pseudo time- group optimization, the result of the previous optimal
scale factor is set to 0.01. design is adopted.
The computational time of an optimization
method depends on single CFD simulation time, com-
Optimization algorithm
puter performance and the algorithm efficiency. As
Because of the high computational cost of CFD solu- the number of design variables of each study is not
tions, it is critical to choose an efficient optimization same, the computational time also varies. In this
algorithm that requires reasonable low number of paper, we perform the optimization on the
function calls. In this paper, we apply the gradient- ThinkStation S20 computer with Windows 7
free optimization algorithm coupled with dynamic Enterprise, 64 bits operating system. The computer
metamodel method to solve the optimization prob- CPU version is Intel(R) Xeon(R) CPU E5620
lem. The gradient-free method treats the aero- @2.4 GHz, and RAM version is 4.00GB, DDR3,
dynamic function as black box, which makes it 1333 MHz. The optimization information for each
easier to realize but generally requires large study is shown in Table 1.
number calls of function to converge the solution
While the adoption of dynamic metamodel can cut
down the computational cost18 with adding more
Problem formulation
samples near the minima solutions, the dynamic The advantage of tandem airfoil over single wing has
metamodel can have a higher accuracy in specific been confirmed by many researchers.3–6 But few of
area. In this way, the samples filed the solution them performed the optimization of the tandem air-
domain, which make this algorithm efficient for foil, and the airfoil profile variables were almost
small aerodynamic optimization with small numbers barely considered. The primary focus of this study is
of variables. In this paper, we use the maximin to maximize the CL/CD of the tandem airfoil, and
Latin hypercube sampling method to generate initial discuss the tradeoff between design variables. In this
and subsequent sampling points. And in each opti- section, we discuss the problem setup and the opti-
mization loop, the metamodel is updated by newly mization formulation for the aerodynamic shape of
added sampling points within the updated trust the tandem airfoil.
1050 Proc IMechE Part G: J Aerospace Engineering 232(6)

Figure 2. Flow chart of optimization method.

Table 1. Computational information for each study. below the canard. The chord of both the airfoils is
200 mm. Both of these two geometry variables are
Study case Study 2 Study 3 Study 4
non-dimensionalization by the chord. The decalage
Number of CFD 53 324 220 is defined as the relative angle of attack between the
simulation calls canard and wing, and decalage is positive if the
Computational time 17.6 h 108 h 73.3 h canard incidence is greater than that of the wing.
The initial airfoil profile is Fx63-137, which is
designed as a low Reynolds number airfoil. For initial
model A, decalage is 0, gap is 0, and the initial stagger
Initial geometry is 1.5, as the investigate model described in Scharpf
The initial geometry is shown in Figure 3. The geom- and Mueller;10 and for initial model B, decalage is 6,
etry variables of the tandem airfoil are described as gap is 0.26, and the initial stagger is 1.5. The refer-
decalage, gap, and stagger, which are the main factors ence area is 0.4 m2.
affecting the performance of the tandem airfoil. The
stagger is defined as the absolute horizontal distance
between the 1/4-chord of canard and wing. The gap is
Grid details
the vertical distance between the 1/4-chord of canard We generate the mesh for tandem airfoil using the
and wing, and is defined negatively if the wing is ICEM software. Structural mesh is adopted in the
Chen et al. 1051

Figure 3. Parameter definition of the canard and wing.

Figure 4. CL comparison of experimental data and CFD data. Figure 5. CL and CD comparison of experimental data and
CFD: computational fluid dynamics. CFD data.
CFD: computational fluid dynamics.

whole flow field, and O-type grid is used in the vicinity As pointed out by Marchman,22 in wind tunnel experi-
of each airfoil. The grid is matched out to a far field ments to study the two-dimensional flow of airfoils, for
that is located at a distance of 20 times of the single small-aspect-ratio models the gap effect may cause more
chord. The flow speed is 16 m/s, and the Reynolds errors in low Reynolds number wing aerodynamic
number is 220,000 based on the single airfoil chord behavior compared to high Reynolds flow. As presented
length, which is within the range of low Reynolds in Figure 5, the drag of CFD results is close to the
number (<5105). To achieve accuracy prediction experimental data variations with CL. Although there
of LSB, there is a higher requirement for the grid are small errors between these results, the error is within
which is close to the wing surface as described in the acceptable range, and the mesh is still useable.
Tang.17 In this paper, there are 180 points on each
airfoil surface, and the spacing on the first layer is
106 of the chord length, with an average growth
Optimization problem formulation
ratio of 1.2 in the O-type grid range. The mesh used Object function. The objective of this optimization
for optimization has reached 500,000 cells. problem is to maximize the CL/CD when a ¼ 8.5 .
The reliability of numerical method must be con- According to the previous studies,7,23 tandem airfoil
firmed. Comparisons for CFD results and experiment can pull off stalling  and show improved aero-
data21 of single Fx63-137 airfoil are shown in dynamic performance than single wing at high .
Figures 4 and 5. The experiment freestream Performing optimization at high  will helps us under-
Reynolds number is 200,000 based on 12 in stand the aerodynamic characteristics of the tandem
(304.8 mm) chord length, which is close to our calcu- airfoil better, and higher CL and CL/CD are able to
lation condition. As shown in Figure 4, the CFD improve the aircraft performance in some ways. These
results are in good agreement with the wind experi- will be further explained in following chapter. The
ment data in lift curve slope and stalling . But the lift flow velocity is set to be 16 m/s, which is within the
curve of CFD results shift higher than the experiment speed range for most small UAVs.
data, the difference between CL is about 0.06 ( ¼ 0 ).
The distinct may be due to the small gap between the Design variables and constraints. Design variables and
wing and end plate in the wind tunnel experiments. constraints need to be carefully selected in order to
1052 Proc IMechE Part G: J Aerospace Engineering 232(6)

yield a physically feasible design. In this study, we the effects of design variables and constraints.
consider airfoil profile, decalage, gap, and stagger as The optimization algorithms, described in the previ-
design variables, and all these variables have been ous chapter, are used to solve the optimization prob-
constrained. Airfoil profile is a main factor to improve lems. By combining different sets of design variables
aerodynamic performance. For each airfoil, there are and constraints, we explore the tradeoffs and benefits
five control points spreading on the upper surface and of each variable. The initial study discusses the aero-
five points on the lower surface. These variables allow dynamic performance of the tandem airfoil in low
us to fine-tune each airfoil to obtain better aero- Reynolds flow and the influence of interference
dynamic performance. At the same time, we impose between canard and the wing (Study 1). After that
thickness constraint along the chord. The constraints we carry out the optimization of tandem airfoil
of airfoils have a lower bound of 70% of the baseline with design variables of decalage, gap, and stagger
thickness and the upper bound is 120% thickness of (Study 2). Then we progress by adding airfoil profile
baseline. These constraints ensure the bending design variables (Study 3). At last, morphing tandem
strength and the deflection of wing tip acceptable. airfoil is presented, as a result of a series of optimiza-
Decalage, gap, and stagger are considered as the tions at different  (Study 4).
three main factors affecting the aerodynamic perform-
ance, such as separation, lift, and drag of tandem air- Study 1. Comparison between artificial and
foil.7 The definitions of these variables are described
in previous section. The variables of gap and stagger
initial tandem airfoil
define the relative position of canard and wing. When To have a better understanding of the tandem airfoil
the two airfoils move away from each other, the aero- aerodynamic performance in low Reynolds flow and
dynamic interference between them decreases. So the the interference between canard and wing, and also to
upper bound of stagger is set as three times of the achieve reasonable comparison for the optimization
chord, at which location the aerodynamic interference studies, we perform a baseline study of two different
becomes unapparent. If these two wings move too initial tandem airfoils and the relevant artificial
close, it leads to intervention as the rudder of the models. The CL and CD of initial model A, in which
upward wing rotates. So the lower bound of stagger S ¼ 1.5, D ¼ 0, G ¼ 0 with Fx63-137 airfoil profile, and
is set to be 1.35 of the chord, at which location the initial model B, in which S ¼ 1.5, D ¼ 6, G ¼ 0.026
horizontal distance between canard trailing edge and with Fx63-137 airfoil profile, are calculated variations
the wing leading edge is only 0.35 of the chord. As with . The calculation results of single airfoil Fx63-
shown in the previous studies,7,8 the tandem airfoil 137 are used to set up the artificial model, as a com-
shows improved performance when the wing is parison without interference. The single airfoil with
under the canard and the decalage is negative, so the same incidence angle of canard is described as
the gap is set from 0.6 to 0 of the chord. And the the ‘‘canard’’ of artificial model, and similarly the
bound of decalage is set from 10 to 0 , which prob- single airfoil with the same incidence angle of wing
ably covers the optimization point as shown in the is described as the ‘‘wing’’ of artificial model.
previous studies. The summary of design variables The ‘‘decalage’’ of the tandem airfoil is defined as
and constraints used in the optimization are summar- the relative angle of incidence between the ‘‘canard’’
ized in Table 2. and ‘‘wing’’. The CL and CD of the artificial model is
the sum of ‘‘canard’’ and ‘‘wing’’, and the calculation
Aerodynamic design optimization formulas are presented in the following equation
studies ðCL Þu þ ðCL Þd
ðCL Þartificial ¼
We perform a series of RANS-based aerodynamic 2 ð1Þ
shape optimizations of the tandem airfoil to examine ðCD Þu þ ðCD Þd
ðCD Þartificial ¼
2

Table 2. Design variables for the tandem airfoil aerodynamic Initial model A and artificial model A are included
shape optimization. in model A. Similarly, initial model B and artificial
model B are classified as model B. The aerodynamic
Count Lower Upper characteristics of model A and model B are shown in
Design variable point boundary boundary Figures 6 to 8. And Figure 9 shows the flow field plots
Airfoil profile 20 70% 120% in the form of velocity magnitude with streamlines.
thickness thickness The plots for the artificial model are splicing by the
Decalage (deg) 1 10.00 0.00 field plots of ‘‘canard’’ and ‘‘wing’’. As shown in
Stagger (c1) 1 1.35 3.00 Figure 6, CL curves of these two artificial tandem
Gap (c1) 1 0.60 0.00 airfoils are quite different, because of different ‘‘dec-
alages’’. For artificial model A, the ‘‘decalage’’ equals
Total 23 – –
to 0 , while for artificial model B it is 6 . The canard
Chen et al. 1053

Figure 6. Lift coefficient (CL) variation with . Figure 8. Lift-to-drag ratio (CL/CD) variation with CL.

model B. This trend has also been mentioned in the

previous studies (not the data),7,8 and is discussed in
the following section. The difference between the ini-
tial model and artificial model is a result of the inter-
ference between the canard and wing, and its influence
on the flow field can be seen when comparing the ini-
tial model and corresponding artificial model as
shown in Figure 9. The interference has great relation-
ship with decalage, gap, and stagger as described by
former researchers.7–9 It can be inferred from the
above description that the change of aerodynamic
characteristic is partly result from the difference
layout in decalage and partly result from the aero-
dynamic interference, which can be confirmed in the
following discussion.
Figure 7. Drag coefficient (CD) variation with .
In Figure 7, the drag difference between the artifi-
cial models is obvious. Because of the ‘‘decalage’’,
artificial model B shows lower drag compared with
of artificial model B is at relatively low angle of inci- artificial model A when  > 6 . While  < 2 , artificial
dence, leading the artificial model B CL curve below model A shows advantages, as the separation occurs
that of the artificial A. When the ‘‘wing’’ of artificial on the ‘‘canard’’ lower surface of artificial model B
model B is over the stalling , ‘‘canard’’ is still below (Figure 9(d)). When the  > 5 , the CD of initial model
that, which leads to the slightly putting off of A is obviously higher than the corresponding artificial
stalling . This can be seen in field plots, when we model A, which can be contributed to the interference
compare Figure 9(h) and (f) where the trailing edge between these two airfoils. As shown in Figure 9(e)
separation on the ‘‘canard’’ disappears. While at small and (f), the presence of wing makes the canard easier
, the ‘‘canard’’ incident angle of artificial model B is to separate. While, when comparing model B, it looks
lower, resulting in the lower surface of ‘‘canard’’ like that the bottom of CD curve for initial model B
easier to separate, which can be confirmed by compar- has enlarged, as the interference between tandem air-
ing Figure 9(b) and (d). At a lower , the CL of initial foil delayed the lower and upper separation. The flow
model and contrast artificial one are adjacent, leading field change can be seen by comparing model B flow
the CL of initial model A over initial model B, owing field plots in Figure 9.
to the different ‘‘decalage’’ for each tandem airfoil. Figure 8 shows that CL/CD of these four models
When we compare the initial and artificial models, changes with CL. The results show both the beneficial
initial model A shows lower CL than artificial model and the invidious influence of interference of tandem
A at the same , and the slope of lift curve is lower. airfoil. It is obvious that artificial model A has higher
Meanwhile, the stalling  and maximum CL only CL/CD than artificial model B. And the max CL/CD is
increases slightly. But for initial model B, CL is over even higher than initial model B. If there is no inter-
that of the artificial one and the stalling  increased ference between the tandem airfoil, artificial model A
with the maximum CL 20% higher than artificial shows best aerodynamic performance when CL ¼ 1.3,
1054 Proc IMechE Part G: J Aerospace Engineering 232(6)

Figure 9. Flow field plots for model A and model B: (a) initial model A  ¼ 0 ; (b) artificial model A  ¼ 0 ; (c) initial model B  ¼ 0 ;
(d) artificial model B  ¼ 0 ; (e) initial model A  ¼ 12 ; (f) artificial model A  ¼ 12 ; (g) initial model B  ¼ 12 ; (h) artificial model B
 ¼ 12 .

and this can be realized by enlarging the stagger of the The influence of tandem airfoil interference can be
tandem airfoil in order to minimize the influence of seen in the Cp distribution figures of canard
interference. While as a result of the considerable (Figure 10) and wing (Figure 11). The single wing
interference, initial model A shows lower CL/CD, ( ¼ 2.5 ) is chosen as the ‘‘canard’’ of artificial geom-
because of the lower CL and higher CD than artificial etry, which has the same angle of incidence as the
model A at the same . For initial model B, at the canard of initial tandem airfoil. And the single wing
same CL, the tandem airfoil exhibits a higher CL/CD ( ¼ 8.5 ) is chosen as the ‘‘wing’’ to compare with the
than artificial model B, which means lower drag at the wing of initial tandem airfoil. We can see from these
same CL. It should be noticed that the stalling  is put figures that the interference between the airfoils has
off in initial model B, which leads to higher CL and changed the Cp distribution on each airfoil greatly. In
low CD. Therefore, initial model B has greater CL/CD Figure 10, the Cp on the upper and lower surfaces of
than artificial model A when CL is higher than 1.5. canard (initial model B) has increased, which contrib-
It indicates that the tandem airfoil could have bet- utes lift increases for canard airfoil. Figure 11 shows
ter aerodynamic characteristic at high CL. Initial that the Cp distribution on the leading edge of upper
model B, which has higher CL/CD than another initial surface of backward wing has decreased greatly, as a
model, will be chosen as the contrast in the following result of the down wash of the canard. And the Cp on
chapter. the lower surface has decreased slightly. The changes
Chen et al. 1055

Table 3. Comparison of optimization results.

Study Study Study Study

1A 1B 2 3

Angle of attack (deg) 8.50 8.50 8.50 8.50

Decalage (deg) 0.00 6.00 8.44 9.02
Stagger (c1) 1.50 1.50 1.35 1.35
Gap (c1) 0.00 0.26 0.41 0.22
Airfoil profile * * * 
CL 1.4623 1.4094 1.3522 1.3448
CD 0.0349 0.0215 0.0198 0.0176
ðCL =CD ÞMAX 41.9 65.61 68.29 76.41
* Airfoil profile design variables not considered.  Airfoil profile design
variables considered.
Figure 10. Pressure coefficient (Cp) distribution of canard.
because the downwash of forward wing decreases
the local  of wing, on the other hand the tandem
airfoil increases the air speed on the upper surface
of wing. This can be confirmed by comparing with
Figure 9(g) and (h).
As shown in Figure 11, the peak of pressure coef-
ficient disappears at the leading edge of the backward
wing, which makes the Cp more flat than a single wing
at the same high . And as described above, the LSB
is pushed backward, which means the region where
flow remains laminar is greater than the single wing.
This makes the backward wing, which at high ,
harder to separate. While at the same time, because
of the decalage, the canard of tandem airfoil is at
relative low . All these phenomena lead to the
delay of stalling .

Study 2. Decalage, gap, and stagger

Figure 11. Pressure coefficient (Cp) distribution of wing.
In this study, we investigate the effect of decalage,
gap, and stagger. These three design variables are con-
of Cp distribution for the wing of tandem airfoil lead sidered to be the most important for the tandem air-
to the decrease of lift. While the overall result from foil, and have been widely studied by many
the lift force measurement shows a slight increase in researchers. The formulation is to get the maximum
CL compared with artificial model B. CL/CD when  ¼ 8.5 . The optimizer can change deca-
LSB does exist on the airfoil of Fx63-137 in low lage, gap, and stagger within the boundary constraint
Reynolds number flow.10 The location of the LSB can in order to get the object. The airfoil profile is not
be extracted from the Cp distributions. The beginning considered in this study.
of the LSB is the point at which a plateau region Table 3 shows the comparison of the results of
begins in the Cp distribution and the transition loca- Study 2 and initial model B. Figures 12 to 14 show
tion is identified by a sharp increase in pressure from the aerodynamic characteristics of Study 2 optimal
plateau region. When the air flows over the LSB, the configuration and initial model B. Figure 15 shows
laminar flow turns to turbulence flow, which may the flow field plot in the form of velocity magnitude
increase the airfoil skin friction drag. The location with streamlines, and the Cp distributions of canard
and size of LSB has great relation with pressure gra- and wing are shown in Figures 16 and 17. The optimal
dient, , Reynolds number, and freestream turbu- tandem airfoils configuration shows further CL/CD
lence.24 In Figure 10, the LSB on canard moves improvement.
forward than the compared airfoil, leading greater As presented in Table 3, the decalage decreases from
portion of turbulence than the compared wing, 6.00 to 8.44 . While the gap decreases from 0.26
which is mainly because of the upwash of wing.23 to 0.41, and stagger also decreases from 1.50 to 1.35,
While the backward wing shows the contrary charac- which means the wing of tandem airfoil moves down
ters. The location of LSB has been pushed backward and forward. The CL and CD has decreased about
compared with the single airfoil. This is mainly 6.2% and 8.0% separately, and as a result leads
1056 Proc IMechE Part G: J Aerospace Engineering 232(6)

to about 4% increase in CL/CD. Reduction in CD

mainly contributes to the decrease of pressure drag
(0.0159 for Study 1; 0.0140 for Study 2), where the
change in viscous drag is not significant (0.00553 for
Study 1; 0.00558 for Study 2). The improvement
of CL/CD is very small, as initial model B is very
close to the optimal solution.
As presented in Figure 12, the CL curve of Study 2
is slightly lower than that of initial model B, at rela-
tively lower . While as  increases, the optimal CL
curve moves over the compared model, and the max-
imum CL also increases. From Figure 13, we can see
that the bottom of CD shifts to higher . And the CD
( ¼ 8.5 ) is lower than the initial model, which is con-
sistent with Table 3. Figure 14 shows the comparison
of CL/CD between Study 2 and initial model B varies Figure 12. Lift coefficient (CL) variations with .
with CL. The CL/CD improvement of the optimal
aerodynamic configuration can be seen in this figure
when CL is above 0.9. At lower CL, initial model B
shows better performance, because CD is almost the
same while CL is slightly higher. This indicates that
tandem airfoils with different configuration show per-
formance improvement in different CL ranges.
As shown by Figure 15, the close coupled tandem
airfoil makes the surrounding air flows faster, and no
trailing-edge separation can be seen. The location of
laminar LSBs is also pointed out in this figure.
Figure 16 shows a comparison between Cp for
canard of initial model B and Study 2 optimal config-
uration. The significant drop in canard CL seen in the
force under this condition is a direct result of the
decrease in Cp. The LSB that exists on the canard of
Study 2 optimal has been pushed backward compared
with the canard of initial model B. The Cp distribution Figure 13. Drag coefficient (CD) variations with .
over the wing of initial model B and Study 2 config-
uration is shown in Figure 17. From this figure it can
be seen that the Cp on the lower surface of the wing
has increased due to the change of aerodynamic inter-
ference between two airfoils, which is the main factor
for increases of wing CL. The Cp on the upper surface
has remained relatively unaltered for Study 2 config-
uration. The length and the position of the LSB seems
not to be affected by the change of optimal variables.

Study 3. Airfoil design variables

In this study, we progress the optimization by adding
airfoil profile design variables. Airfoil profile shows
great importance for improving the aerodynamic per-
formance, while compared with decalage, gap, and
stagger, it has barely been considered by the research-
ers. The object formulation is the same as that of
Study 2, with the addition of airfoil design variables Figure 14. Lift-to-drag ratio (CL/CD) variations with CL.
and shape constraints. A total of 10 shape design vari-
ables are used for each airfoil profile, with five vari-
ables controlling the upper surface and five variables As presented in Table 3 the decalage decreases by
controlling the lower surface. The constraints of about 0.58 . The gap increases from 4.10 to 0.22,
thickness are imposed. The optimization result, but and the stagger is the same as the previous study,
not the airfoil profile variables, is proposed in Table 3. which means the wing of tandem airfoil moves
Chen et al. 1057

Figure 18. Lift coefficient (CL) variations with .

Figure 15. Velocity magnitude and streamlines for Study 2

optimal configuration.

Figure 19. Drag coefficient (CD) variations with .

upward, compared with Study 2. The CL/CD increases

Figure 16. Pressure coefficient (Cp) distribution of canard. to 76.25, which is 11.9% higher than that of Study 2,
with the CD decreasing by about 12.5%, while the CL
is almost unchanged. The CD drop of Study 3 results
from the decrease of both pressure drag (0.01248 for
Study 3) and viscous drag (0.00515 for Study 3). The
result concludes that CL/CD can be further improved
by introducing airfoil profile consideration and
adjusting the configurations.
As presented in the CL curve, the lift characteristics
for Study 2 and Study 3 are almost the same when
 ¼ 8.5 , while on other conditions the CL of Study 3
is slightly lower (Figure 18). In Figure 19, the CD for
Study 2 and Study 3 are almost same, while when
 ¼ 8.5 , the CD of Study 3 configuration is lower.
When  increases, the CD also increases slightly
higher than Study 2. This has caused the improvement
of CL/CD shown in Figure 20 at the design . The
CL/CD for Study 3 is steeper than that of Study 2,
which makes Study 2 show better performance only
in a small lift range. When  increases or decreases,
Figure 17. Pressure coefficient (Cp) distribution of wing. Study 2 shows better characteristics.
1058 Proc IMechE Part G: J Aerospace Engineering 232(6)

Figure 20. Lift-to-drag ratio (CL/CD) variations with CL.

Figure 22. Velocity magnitude and streamlines for Study 3

optimal configuration.

Figure 21. Airfoil geometry.

The optimal airfoils of canard and wing for Study 3

are presented in Figure 21. Comparing with the initial
airfoil Fx63-137, the optimal canard and wing is thin- Figure 23. Pressure coefficient (CP) distribution of canard.
ner, and both the upper and lower surfaces are more
flat. It is also shown in this figure that the optimal
shapes of canard and wing are almost the same
except the difference near the leading edge, which
makes the wing camber greater. The flow field plot
and Cp distribution comparison of both the canard
and wing are presented in Figures 22 to 24.
There are slight differences in the flow field plots
between Figure 15 and 22. As shown by Figure 22, the
LSB on the upper surface of Study 3 canard is pushed
backward compared with Study 2, which is confirmed
by Figure 23. It means the flow keeps laminar in
larger range on the upper surface of canard. This
will tend to decrease the skin friction drag. As
shown by Figure 23, the size of the LSB has been
enlarged from 15.6% of the chord to 25%, which
can lead to the increase of form drag. But in general,
the viscous drag of canard decreases. The upper sur-
face Cp distribution of Study 3 canard decreases Figure 24. Pressure coefficient (CP) distribution of wing.
Chen et al. 1059

compared with Study 2, while the lower surface Cp Study 4. Optimal configurations with the
contributes more force in normal vector, which
makes the normal force of canard almost unchanged.
change of 
The Cp on the wing shows smaller change than the In this study, we discuss the optimal configurations of
canard, but LSB size enlarging and location backward the tandem airfoil variation with . The decalage, gap,
can also be detected. and stagger of tandem airfoil are considered as opti-
mization variables, while the airfoil profile remains
the same as Study 3. The object formulation of this
study is to get the max CL/CD when  equals to 0.0,
Table 4. Aerodynamic configuration morphing variations with 4.0 , 6.0 , 8.5 , and 10.0 .
angle of attack. The changes of optimal tandem airfoil aero-
dynamic configuration variables variations with 
Serial numbers 1 2 3 4 5
are shown in Table 4 and the morphing phase coupled
Angle of attack (deg) 0.00 4.00 6.00 8.50 10.00 with flow field plots is shown in Figure 25. The results
Decalage (deg) 0.00 1.98 5.10 9.02 10.00 show that at small , the tandem airfoil attempts to
Gap (c1) 0.16 0.31 0.51 0.22 0.14 improve the aerodynamic performance by enlarging
Stagger (c1) 3.00 3.00 2.16 1.35 1.35 the stagger (reduce the airflow interference) and redu-
cing the decalage. When  ¼ 0 , decalage and stagger

Figure 25. Tandem airfoil morphing process variations with .

1060 Proc IMechE Part G: J Aerospace Engineering 232(6)

reach the maximum boundary as was set. As  show the best performance. The morphing tandem
increases, the wing moves forward to the canard and airfoil can also exhibit performance improvement
the decalage also decreases to adapt to the new flow at high . Although the CL/CD of morphing
condition. When  reaches 10 , lower boundaries of tandem airfoil is higher than Study 2 when the two
decalage and stagger are reached. It seems that the are at the same , the fall of CL causes the CL/CD
flow interference becomes an important factor for curve lower than Study 2 when the CL is high, which
improving the aerodynamic performance as CL is the result from the changes in airfoil profile. Power
increases. factor (PI), which is defined as in equation (2), can
To have a better understanding of the perform- be used to estimate the power consumption in low
ance of morphing tandem airfoil, the CL/CD curve is Reynolds number range level flight. Higher PI means
calculated, and comparisons between the previous lower level flying power consumption, as described in
studies and the single airfoil Fx63-137 are presented Simons.25 And it is obvious from Figure 26 that
below. The 0 optimal configuration is used to cal- morphing tandem airfoil configuration has highest
culate the CL and CD when  < 0 , as two of its PI than other configurations.
design variables have reached the upper boundaries.
Similarly, the aerodynamic coefficient is calculated
ðCL Þ1:5
by using the 10 optimal configuration, PI ¼ level
ð2Þ
when  > 10 . ðCD Þlevel
These results are shown in Figure 26. As shown in
this figure, the CL/CD curve of morphing tandem
airfoil almost envelops the others, except the Study The morphing airfoil shows improved aerodynamic
2 opt at high CL, which means the morphing tandem characteristics according to foregoing descriptions
airfoil has lower CD at most CL points. At lower CL, and can be realized on small UAVs, as the flow
single wing exhibits higher CL/CD than all tandem speed is low and the aerodynamic loading is relatively
airfoils except the morphing one, which is mainly small, which can reduce the structural strength
because of the separation occurred on the lower sur- requirements, thereby reducing the weight of the
face of canard. As the decalage of initial B, Study 2 deformation mechanism. An unusual morphing
opt and Study 3 opt are respectively 6.00 , 8.44 , tandem airfoil UAV conceptual design is proposed
and 9.02 . The canard incident angles of these con- in Figure 27 for extended mission. The UAV is
figurations are much lower at , and leading separ- designed to perform cruise with low  optimal
ation occurs early on the lower surface of the tandem configuration while morphs to closely coupled
canard. The separation leads to lower lift and more tandem configuration when long time detection mis-
drag, making the aerodynamic characteristics of sion is required. The results of Study 4 can be adapted
these tandem airfoil configurations poor. Morphing for this craft to take deformation. The fuselage can
tandem airfoil could improve this condition by keep low  by adjusting the mounting angle when the
increasing decalage and reducing the interference tandem airfoil is morphing.
between airfoils. As the CL increases, the canard of Morphing tandem aircraft shows performance
morphing tandem rotates to maintain a small improvement in several aspects. Initially, the
incident angle. And at the same time, the relative tandem airfoil pulls off the stalling  of aircraft
position of wing and canard is also adjusted to than single wing. Furthermore, its higher maximum
CL can reduce the take-off/landing speed and dis-
tance of morphing aircraft, making the UAV to be
used in the limited space area or platform, such as
on the flight deck of destroyer. Although take-off
and landing is a small part of the mission envelope,
it is very useful for expanding the scope of UAV
applications. Landing gear strength design require-
ments can be reduced as the take-off/landing speed
decreases. At same time, the low-speed maneuver-
ability of the craft can also be improved by higher
CL. Additionally, as morphing tandem airfoil shows
higher CL/CD in large lift range, the morphing
tandem aircraft shows extended flight range based
on Breguet formula.26 As the aircraft could
have higher PI, the level flight power consumption
can also be reduced, and the aircraft can be
fuel saving and flight time extended. Other advan-
tages for the morphing aircraft can be studied in
Figure 26. Lift-to-drag ratio (CL/CD) variations with CL. further work.
Chen et al. 1061

Figure 27. Conceptual design of morphing tandem airfoil.

Conclusions
necessity of adding airfoil profile to the tandem opti-
In this work, we set out to study the aerodynamic mization design. We study the optimal configuration
characteristics of tandem airfoil configuration in of tandem airfoil at different , and present the
Reynolds number flow. We perform a comparison morphing tandem airfoil concept. Morphing
study and a series of optimizations of the two- tandem airfoil can have higher CL/CD than the
dimensional tandem airfoil. The formulation of opti- fixed one in larger CL range, as it can make up the
mizations is set to get the maximum CL/CD within the defect at small  by adjusting configurations to put
geometric constraints. These optimizations are per- off the separation at the lower surface of canard.
formed on structural mesh, using decalage, gap, and Then we propose the conceptual design of morphing
stagger to define the configurations of tandem airfoil tandem wing vehicle, which shows the capacity of
and another 20 shape design variables adjust the performance improvement in take-off/landing per-
shape of canard and wing. formances, low-speed maneuverability, fuel saving,
The comparison of the initial model and artificial range increasing and others, but many other details
model of tandem airfoil indicates that specific about the aircraft remain to be further studied.
tandem airfoil can delay the separation point to
higher , and shows increase in maximum CL and Declaration of Conflicting Interests
CL/CD. This performance improvement is partly The author(s) declared no potential conflicts of interest with
due to the difference in decalage and partly due to respect to the research, authorship, and/or publication of
the aerodynamic interference between airfoils. In the this article.
optimization, which only includes the configuration
design variables, the optimal result improves 4% in Funding
CL/CD compared with initial model B, which already The author(s) received no financial support for the research,
has good aerodynamic performance. The result con- authorship, and/or publication of this article.
cludes that tandem airfoil should be optimized to get
higher CL/CD at given , and with different config- References
urations, different tandem airfoils show performance 1. Zhang GQ and Yu SCM. Unsteady Aerodynamics of a
improvement in different CL ranges. When adding morphing tandem-wing unmanned aerial vehicle.
the airfoil profile design variables to the optimiza- J Aircraft 2012; 49: 1315–1323.
tion, the CL/CD of the tandem airfoil is further 2. Bramesfeld G and Malik A. Micro aerial vehicles in con-
increased to 11.9%, with the changes in airfoil pro- fined spaces: are two wings better than one? J Aircraft
files and configurations. The result shows the 2015; 5: 1575–1585.
1062 Proc IMechE Part G: J Aerospace Engineering 232(6)

3. Nenadovith M. Recherches sur les Cellules Biplane basis function metamodel. In: 17th AIAA/ISSMO multi-
Rigides d’envergure Infine. Institut Aero’ – technique disciplinary analysis and optimization conference,
de Saint-Cyr, Paris, 1936. Washington, D.C., 13–17 June 2016.
4. Olson EC and Selberg BP. Experimental determination 20. Li JZ and Gao ZH. Multivariable aerodynamic design
of improved aerodynamic characteristics utilizing based on multilevel collaborative optimization. Acta
biplane wing configurations. J Aircraft 1976; 13: Aeronaut Astronaut Sin 2013; 34: 1–65.
256–261. 21. Selig MS, Donovan JF and Fraser DB. Airfoils at low
5. Addoms RB and Spaid FW. Aerodynamic design of speeds. Soartech8, Virginia Beach, VA: Soar Tech
high-performance biplane wings. J Aircraft 1975; 12: Publications, 1989.
629–630. 22. Marchman JF. Aerodynamic testing at low Reynolds
6. Rokhsaz K. Analytical Investigation of the aerodynamic numbers. J Aircraft 1987; 24: 107–114.
characteristics of dual wing systems. Thesis, University 23. Zhang GQ and Yang SX. Experimental investigation of
of Missouri, USA, 1980. the aerodynamic characteristics of tandem-airfoil based
7. Rhodes MD and Selberg BP. Benefits of dual wings on low Reynolds number. Chin J Comput Mech 2010;
over single wings for high-performance business air- 27: 733–737.
planes. J Aircraft 1984; 21: 116–127. 24. Lin JCM and Pauley LL. Low-Reynolds-Number sep-
8. Rokhsaz K and Selberg B. Dual and single wing design aration on an airfoil. AIAA J 1996; 34: 1570–1576.
integration-optimized for decalage angle, R(e), and 25. Simons M. Model aircraft aerodynamics. London:
M(cr). In: 2nd applied aerodynamics conference, Special Interest Model Books Ltd, 1999.
Seattle, Washington, USA, August 1984. 26. Range (aeronautics) - Wikipedia. [Link]
9. Rokhsaz K and Selberg B. Dual-wing systems with org/wiki/Range_(aeronautics) (2016, accessed 10
decalage angle optimization. J Aircraft 1986; 23: February 2017).
444–448.
10. Scharpf DF and Mueller TJ. Experimental study of a
low Reynolds number tandem airfoil configuration.
J Aircraft 1992; 29: 231–236. Appendix
11. Jones R, Cleaver DJ and Gursul I. Aerodynamics of
biplane and tandem wings at low Reynolds numbers. Notation
Exp Fluids 2015; 56: 1–25.
c length of chord
12. NASA Technical Reports Server (NTRS). Arbitrary
Cp static pressure coefficient
shape deformation in CFD design, [Link]
gov/[Link]?R¼20140002362&qs¼N%3D42949672 CD drag coefficient
58%26Nn%3D4294963897%257CdOrganization% (CD)artificial artificial tandem airfoil drag coefficient
257CNASA%2BLewis%2BResearch%2BCenter (2014, (CD)u single airfoil drag coefficient with the
accessed 12 February 2017). same incidence angle of canard
13. Sculptor User Manual, version 3.2., Chapter 1 (CD)d single airfoil drag coefficient with the
Introduction. same incidence angle of wing
14. Lyu Z and Martins JRRA. Aerodynamic design opti- (CL)d lift coefficient
mization studies of a blended-wing-body aircraft. (CL)artificial artificial tandem airfoil lift coefficient
J Aircraft 2014; 51: 1604–1617. (CL)u single airfoil lift coefficient with the
15. Counsil JNN and Boulama KG. Low-Reynolds-
same incidence angle of canard
number aerodynamic performances of the NACA
0012 and Selig-Donovan 7003 airfoils. J Aircraft 2013;
(CL)d single airfoil lift coefficient with the
50: 204–216. same incidence angle of wing
16. Counsil JNN and Boulama KG. Validating the (CL)level level flight lift coefficient
URANS shear stress transport g-Rey model for low- (CD)level level flight drag coefficient
Reynolds-number external aerodynamic. Int J Numer CL/CD lift-to-drag ratio
Meth Fluids 2011; 69: 1141–1432. D decalage (deg)
17. Tang L. RANS simulation of low-Reynolds-number G gap
airfoil aerodynamics. In: 44th AIAA aerospace sciences S stagger
meeting and exhibit, Reno, Nevada, 1–14 January 2006. U X velocity
18. Han Z. Kriging surrogate model and its application to U1 freestream velocity magnitude
design optimization: a re-view of recent progress. Acta
V Y velocity
Aeronaut Astronaut Sin 2014; 35: 1–28.
19. Jian L, Long T, and Shi R, et al. Composite structure  angle of attack (deg)
optimization for satellite using discrete dynamic radial