You are on page 1of 13

Aerospace Science and Technology 43 (2015) 152164

Contents lists available at ScienceDirect

Aerospace Science and Technology

Supercritical natural laminar ow airfoil optimization for regional

aircraft wing design
Yufei Zhang a,1 , Xiaoming Fang a,2 , Haixin Chen a,,3 , Song Fu a,3 , Zhuoyi Duan b,4 ,
Yanjun Zhang b,5

Tsinghua University, Beijing, 100084, China

AVIC The First Aircraft Institute, Xian, 710089, China

a r t i c l e

i n f o

Article history:
Received 29 September 2014
Received in revised form 26 February 2015
Accepted 27 February 2015
Available online 4 March 2015
Natural laminar ow
Pressure gradient constraint
Favorable pressure gradient

a b s t r a c t
An optimization design method of supercritical natural laminar ow airfoil based on Genetic Algorithm
and Computational Fluid Dynamics is tested in this paper. Class Shape Transformation method is
adopted as geometry parameterization method. Constraints on pressure distribution are applied to gain
appropriate ow eld in addition to the L / D performance. A xed transition computation method
is used in the optimization process to save computation time while giving the reasonable friction
drag estimation and predicting the inuence of the laminar boundary layer on airfoil performances.
Specied favorable pressure gradient constraints are used to guarantee the expected laminar length.
Objective of optimization is set to weaken the shock wave and minimize the pressure drag. Such a
simplied NLF optimization process is veried by natural transition computation. The optimal setting
of the favorable pressure gradient constraint, which is important for the trade-off between drag
reduction and laminar stability, is then studied via numerical investigation. Results show that the airfoil
optimized by constraining a favorable pressure gradient larger than 0.2 is good for both cruise eciency
and robustness. A natural laminar wing is then designed based on the optimized airfoil. Numerical
verications show that the wing has good natural laminar performance and low speed behavior.
2015 Elsevier Masson SAS. All rights reserved.

1. Introduction
Natural Laminar Flow (NLF) airfoil has already been studied for
several decades [15]. However, it still draws high attention in recent years. As the friction drag is about half of the total drag
for modern civil aircrafts [10], laminar technology has great potential to increase lift to drag ratio. Although a lot of ight tests
had successfully validated the eciency of NLF airfoil on modern large civil aircrafts [16,6], the technology is only realized on
wings of several light business aircrafts in the commercial market
until now, such as the Honda Jet [11,13] and the Aerion Super-

Supported by National Key Basic Research Program of China (2014CB744801)

and National Natural Science Foundation of China (11102098 and 11372160).
Corresponding author.
E-mail addresses: (Y. Zhang), (H. Chen).
Assistant professor, School of Aerospace Engineering.
Master student, School of Aerospace Engineering.
Professor, School of Aerospace Engineering.
Research professor, General Design and Aerodynamic Department.
Senior engineer, General Design and Aerodynamic Department.
1270-9638/ 2015 Elsevier Masson SAS. All rights reserved.

sonic Business Jet [14,31]. Parameter analysis results of Lammering

et al. [19] showed that, NLF design of a Boeing 777-size airplane
could not show improvements on airplane direct operating costs
than conventional turbulent design unless the drag reduction is
more than 40 counts. In order to preserve laminar region, a lower
leading edge sweep angle is adopted in the NLF wing [19]. Consequently, the cruise Mach number is quite lower than turbulent
wing. Increasing cruise Mach number is as important as reducing
skin friction for NLF wing design. Benet and penalty of the NLF
technology need to be clearly quantied.
Airfoil is a fundamental element of a wing. Many investigators
have focused on the supercritical NLF airfoil design because of its
importance for the high-subsonic NLF wing. Biber and Tilmann [4]
developed a supercritical NLF design method based on the panel
and Euler codes coupled with boundary layer equation, and attempted to increase the drag bucket of the NLF airfoil in order to
extend the operational speed range. Eggleston et al. [9] showed
that the peak Mach number, pressure gradient, and aft loading
were critical factors of a favorable pressure distribution of an NLF
airfoil. Cella et al. [5] successfully used the rule of cosine to design a high-subsonic NLF wing with a multi-objective optimization
method. They separately designed the root/kink/tip airfoils and got

Y. Zhang et al. / Aerospace Science and Technology 43 (2015) 152164




Turbulence intermittency factor

Reynolds number based on boundary layer momentum thickness
Mach number
Pressure coecient
Friction coecient
Lift coecient
Drag coecient

an NLF wing with good laminar performance. Khalid and Jones [17]
showed the supercritical NLF airfoils with different thicknesses designed at the National Aeronautical Establishment, and the experiment validated good performances of the airfoils at Reynolds number up to 12.5 million. Streit et al. [30] provided a new method
of converting pressure distribution of two-dimensional NLF airfoil
to three-dimensional wing by considering the sweep and tapered
effects. The pressure distribution of the Honda jet [12] provided
some new concepts of supercritical NLF airfoil design, on which
the tailing edge bubble of the upper surface was adopted to suppress low speed ow separation, and the leading-edge shape was
carefully designed to cause transition at high angles of attack (AoA)
to obtain higher maximum lift coecient. Shockwave/boundary
layer interaction is a major cause of transonic drag rising. Aircraft
designers would have an opportunity to raise cruise Mach number
if they could decrease shock wave drag. That is a main objective of
supercritical airfoil design. Apparently, another objective of supercritical NLF airfoil design is to decrease the friction drag. In realistic
high-subsonic design practice, aerodynamic designer usually has a
target (or an expectation) of laminar length for a certain condition
based on experience or literature survey. Consequently, the potential of friction drag reduction is approximately conrmed. The
design problem becomes how to achieve laminar length and how
to reduce shockwave drag. Laminar ow length could be achieved
through maintaining Favorable Pressure Gradient (FPG, or negative
pressure gradient) [8]. However, the FPG should not be so great as
to avoid excessive shock strength [8]. Nevertheless, the quantitative inuence of FPG on drag of supercritical NLF airfoil is not so
clear. Trade-off between wave drag and friction drag is a problem
of NLF airfoil design, which is closely related to the FPG.
With the help of modern optimization methods, the application of NLF technology could be pushed forward. Genetic algorithm
[35,2] and adjoint method [21,22] are two kinds of widely used
optimization methods on airfoil design. Both methods have their
inherent problems. The latter is lack of global optimization ability and dicult to treat realistic design constraints. The former
has the probability to achieve global optimization solution, but requires lots of computation costs. Computation cost of CFD must
be carefully controlled in genetic algorithm optimization. Manin-loop design process is a practical compromise for engineering
applications, for example, introducing some pressure distribution
constraints in a design problem to guide optimization direction
[34,33] and articially adjusting the constraints and objectives during design iteration.
In this paper, supercritical NLF airfoil is optimized for the highsubsonic NLF wing of a regional jet. A Reynolds Averaged Navier
Stokes CFD solver is used as aerodynamic analysis tool. An inhouse developed optimization platform [27] based on the Nondominated Sorting Genetic Algorithm-II (NSGA-II) [7] is adopted as
background scheduling software. Class shape transformation (CST)
method is employed as airfoil parameterization method. Optimization process is controlled by a series of realistic constraints. Supercritical NLF airfoil is obtained through optimization based on

C d, p
C d, f
dC p /d X
t /c

Pressure drag coecient

Friction drag coecient
Pitching moment
Lift to drag ratio
Pressure coecient gradient of airfoil
Airfoil relative thickness
Chord length
quantity of favorable pressure gradient, dC p /d X

RAE2822 supercritical airfoil. Effect of FPG on laminar characteristics is investigated by six airfoils which are optimized by different pressure distribution constraints. A good compromise between
pressure drag and friction drag is achieved when the FPG on the
upper surface before 50% chord is larger than 0.2. A high-subsonic
NLF wing is got by assembling and simply modifying the optimized
airfoil. Numerical results demonstrate that the wing has both large
laminar region and good robustness.
2. Numerical method and validation
2.1. Turbulence modeling
Aerodynamic analysis in this paper is based on a Reynolds Averaged NavierStokes CFD code. It is used to compute xed transition
ow eld in the optimization, and to calculate natural transition
ow eld after optimization.
In NLF wing design, the accuracy of transition prediction is an
important factor of design quality. Based on Shear Stress Transport (SST) model [23], a transition model had been developed by
Menter et al. [24,25] through adding an intermittency factor ( )
equation and a momentum thickness based Reynolds number (Re )
equation to the turbulence models, called as SST Re model.
Because of the strong source terms in the SST Re model, the
computation time of the SST Re model is much longer than
the SST model, as the CourantFriedrichsLewy number must be
smaller. However, the computation time is a critical factor of genetic algorithm optimization. An alternative method is used in the
present optimization process to reduce computation cost. The pressure distribution of airfoil is predicted by the SST turbulence model
with xed transition when optimizing the airfoil shape and the
accurate transition location is validated by the SST Re model
after optimization. The xed transition location is located based
on the design expectation of laminar length. The xed transition
computation could consider the inuence of the laminar boundary
layer and accurately predict the pressure drag. The laminar length
is achieved through maintaining the FPG. In the next sub-section,
the code is validated by two cases with experimental data. The
pressure distributions of the xed transition and natural transition
computations are also compared.
2.2. Validation
In this paper, we mainly focus on transition prediction capability of the SST Re transition model for supercritical NLF airfoil.
The transition prediction accuracy is validated by two test cases.
The rst is a low speed NLF airfoil, NLF 0416. It is used to test
grid convergence, as well as xed transition computation to ensure it as a cheaper substitution in the optimization process. The
second case NLR 7301 is used to validate the transition prediction
accuracy of transonic ow.


Y. Zhang et al. / Aerospace Science and Technology 43 (2015) 152164

locations. It employed microphones, which connected to the orices on the airfoil, to determine the transition location. Transition
locations of present computation are all located between the laminar and turbulence orices of the experiment, which shows the
Re transition models good capability of capturing natural transition.

Fig. 1. Computation grid of NLF 0416 airfoil (257 97 points).

2.2.1. NLF 0416 airfoil

NLF 0416 is a low speed natural laminar ow airfoil. The National Aeronautics and Space Administration provided a lot of experimental data of the airfoil [29]. In the present study, ow conditions are free stream Mach number = 0.1 and Reynolds number =
1.0 106 . Free stream eddy viscosity is set as 0.1 times of dynamic
viscosity, and turbulence intensity is 0.1%. Fig. 1 shows a grid with
257 97 points. The grid is generated by an in-house developed
grid code. Generated by solving an elliptic equation with a source
term, the mesh is in good orthogonal with the solid wall. The rst
layer in the normal direction is less than 3.0E6 to ensure Y+ less
than 1.0, and increasing rate of the boundary layer grid height
is 1.15. Far-eld boundary is set at 80 times of the chord length
away from the airfoil.
Three meshes with different grid number are used to test grid
convergence. Fig. 2 shows pressure and friction distributions of the
three meshes at the same AoA by SST Re transition model,
as well as the xed transition result of 257 97 points grid by
SST model. The pressure distributions of the different grids computed by the SST Re transition model are almost the same,
and the friction distributions also show a convergent tendency. The
xed transition positions of upper and lower surfaces are located
at 40% and 60%, respectively. The pressure distribution of the xed
transition matches well with the others. The friction distribution
is a little different with the natural transition because of inaccurate xed transition location. The xed transition computation has
quite a little inuence on the drag prediction, as in Table 1. The
xed transition computation of the 257 97 grid costs about 2
minutes on an Intel 2.8 GHz CPU core. However, the natural transition computation on this grid needs about 10 minutes. Therefore
in the optimization, the xed transition computation could be used
to minimize the shock drag in order to increase optimization eciency.
Fig. 3 shows lift and lift-drag polar curves of the computation
results and experimental data. The CFD results are computed by
the 257 97 points grid. Results of transition computations are
in good agreement with experimental data. However, the drag of
full turbulence computation is much higher than experiment. Fig. 4
shows transition locations of the computation compared with experimental data. The experiment could not provide exact transition

2.2.2. NLR 7301 airfoil

NLR 7301 airfoil is a typical supercritical airfoil with large thickness. The Advisory Group for Aerospace Research and Development
provided a series of transition experimental data [26]. Computation
results are also available in publications [32,36]. In this section,
the airfoil is adopted to test the transition prediction accuracy
for transonic ow, especially with shock/boundary layer interaction. Computation grid is the same as the 361 137 grid of Section 2.2.1. Several free stream Mach numbers are calculated to test
drag rising. Reynolds numbers and AoAs in the experiment are all
2.2 106 and 0.85 . However, AoAs in the present computation
are corrected according to corrections in the experiment, referring
to literature [32]. Free stream eddy viscosity is set as 0.1 times of
dynamic viscosity, and turbulence intensity is 0.1%.
Fig. 5 shows the aerodynamic coecients compared with experiment, including lift, drag and pitching moment coecients. All
of the curves match quite well before Mach number 0.75. However,
the shockwave/boundary layer interaction becomes very strong after drag rising (Ma = 0.75), leading to deviation of the RANS results. Nevertheless, the Mach number of drag divergence is well
captured. Fig. 6 shows transition locations compared with experimental data at different Mach numbers. On both upper and lower
surfaces, the computed transition locations match well with the
experimental data. Results of NLR 7301 validated the transition
prediction capability for transonic ow, and the accuracy of both
pressure and friction drag. Such a capability provides the basis for
supercritical NLF validation.
The 257 97 points grid is having a good accuracy together
with an acceptable computation cost. In the following section, this
grid is chosen as the design grid in the optimization process, and
the 361 137 points grid is employed as a natural transition validation grid after optimization.
3. Natural laminar airfoil optimization
3.1. Design requirements of NLF airfoil
The NLF wing in this paper is designed for a high-subsonic
regional jet. The objectives and constraints of NLF airfoil are derived from the needs of the wing. Cruise Mach number of the
regional jet is 0.76, and ight altitude is 37 000 ft. Fig. 7 shows
the planform of the wing. Sweep angles of leading edge and 1/2
chord are 17.50 and 13.60 , respectively. Airfoil optimization is
focused on proles of the outboard wing. Rule of cosine [5] can be

Fig. 2. Results of three different grids of NLF 0416 (Ma = 0.1, Re = 1.0E6, AoA = 1 ).

Y. Zhang et al. / Aerospace Science and Technology 43 (2015) 152164


Table 1
Aerodynamic coecients of different computations of NLF 0416 airfoil (Ma = 0.1, Re = 1.0E6, AoA = 1 ).

Lift coecient, C l
Drag coecient, C d
Pressure drag, C d, p
Friction drag, C d, f
Transition positions (upper/lower surfaces)

187 69 grid
Natural transition

257 97 grid
Natural transition

361 137 grid

Natural transition

257 97 grid
Fixed transition




40%/60% (xed)

Fig. 4. Transition locations of the computation compared with experiment.

Fig. 3. Lift and lift/drag polar curves of NLF 0416 airfoil (Ma = 0.1, Re = 1.0E6, 257
97 grid).

used to transform wing parameter to airfoil requirement. In general speaking, any sweep angle from leading edge to trailing edge
of a tapered wing could be used for 3D to 2D parameter transformation. However, Streit et al. [30] suggest that the sweep angle at
shock wave location could be a good choice for transonic airfoil. In
this paper, the shock wave location is found to be near the 1/2c, so
the sweep angle of the 1/2 chord is used to decide the airfoil design Mach number. Relative thickness of the airfoil is transformed
from the kink location. The parameters of airfoil design are listed
in Table 2.
There are three types of transition mechanisms on sweep wing,
which are TollmienSchlichting (TS) instability, crossow instability and attachment line instability or contamination. TS instability
could be suppressed by FPG. In this paper, the TS instability is
controlled by pressure distribution constraints in optimization and
predicted by the SST Re transition model after optimization.
Crossow instability and attachment line contamination could not
be directly captured in 2-D computation. Referring to ight tests of
Anderson and Meyer [1], the transition location is almost the same
when the sweep angle is less than 20 . Therefore, crossow instability could be excluded in the present planform. Attachment line
contamination is also a cause of premature transition. It is primarily inuenced by leading edge radius. A leading edge momentum
Reynolds number is often used to estimate the leading edge contamination [28,3]. In the present optimization, the leading edge
radius is restricted to be less than 0.035c because the leading edge
momentum Reynolds number should be less than 100 [28,3].
3.2. Class-shape transformation method
Parameterization method should have fewer variables and
larger design space. In recent years, Class-Shape Transformation
(CST) [18,20] method is widely used in aerodynamic optimization.
In the method, an airfoil with a blunt trailing edge is dened as
the product of a class function and a shape function, plus a linear term of tail thickness, as in Eq. (1). The = Y /c, = X /c are
the normalized coordinates of the airfoil. The te is the tail thickN
ness. The C N 21 () is a class function, as in Eq. (2). Round nose and

pointed aft end airfoils could be dened by setting the N 1 = 0.5

and N 2 = 1.0. In this paper, the shape function S () is dened by
a combination of Bernstein polynomials B ni (), as in Eq. (3). The
b i is design variable vector, which represents weights of Bernstein
polynomials. Upper and lower surfaces of airfoil are generated by
two independent polynomials. Thus it is dicult to explicitly x
the airfoil thickness in the method. When designing an airfoil with
a certain thickness, Y coordinate of the airfoil is scaled to control

() = C NN21 () S () +
C N 21 ()

S () =



= N 1 (1 ) N 2


b i B ni ()

i =0

which B ni () = K ni i (1 )ni , K ni =

i !(n i )!


Sixth order (n = 6) CST polynomial is used for both upper and

lower surfaces in this paper. Table 3 shows ranges of the CST variables in the optimization. The rst variables of both upper and
lower surfaces are different with the others in order to obtain a
large leading edge radius. The sixth and seventh variables of the
lower surface are adjusted to obtain aft loading. Each variable is
separated by 201 steps in the optimization, thus possible parameter combinations are about 20114 1.76 1032 .
3.3. Airfoil optimization with pressure distribution constraint
For supercritical NLF airfoil, wave drag is primarily determined
by the Mach number before the shock wave, and friction drag is
controlled by the length of laminar region. Sucient FPG length in
the front part of NLF airfoil is needed to maintain laminar ow.
However, FPG would raise the Mach number ahead of shock and
increase wave drag. Therefore the appropriate FPG is the key issue
of NLF supercritical airfoil design.
In the authors previous publications [34,33], pressure distribution constraints can be applied on the airfoil or wing design
process to gain a good overall supercritical performance. Such a
pressure distribution constraint provides us with a tool to obtain
airfoils with different pressure gradients and compare their performances.
In order to not only design an airfoil with minimum drag, but
also separately study the effects of the FPG on pressure drag and


Y. Zhang et al. / Aerospace Science and Technology 43 (2015) 152164

Fig. 5. Aerodynamic coecients of the NLR 7301 airfoil varying with Mach number (Re = 2.2E6).

Fig. 6. Transition location of NLR 7301 airfoil varying with Mach number (Re =

Fig. 7. Planform of the natural laminar ow wing.

Table 2
NLF airfoil design parameters.
Cruise Mach number
Reynolds number
Lift coecient
Relative thickness


friction drag, a strategy is designed in this paper. First, by setting

different pressure gradient constraints, based on xed transition
computation and a reasonable transition location, corresponding
low pressure drag airfoils are designed by optimization [34,33].
Then the transition computations are conducted on these airfoils
to validate the real laminar ow length, and to investigate the effects of the pressure gradient, and at the same time, get overall
optimal design.
In this section, the xed transition points of both upper and
lower surfaces are set to be at X = 0.52c, as we suppose that the
laminar region of the nal design should be longer than 50%. It
also assumes that the shock wave should locate after X = 0.52c,
because the computation would not be stable if ow after the
shock wave is still laminar. Such a setting is reasonable for supercritical NLF airfoil design. Objective of optimization is to minimize the pressure drag. Shock wave drag will be implicitly minimized with this objective. NSGA-II is employed as the optimization
method. The 257 97 points grid discussed in Section 2.2.1 is used
in the optimization process.
The optimization problem is dened as in Eq. (4). The design
conditions are from Section 3.1. Leading edge radius R LE is limited
to be less than 0.035 in order to avoid leading edge contamination, and larger than 0.010 to ensure a good enough low speed
stall behavior. Pitching moment C m is expected to be larger than
0.10 to restrict trim drag. Because of xed transition computation, the friction drag variation of airfoils is quite small. However,
if ow separation exists in the ow, friction drag would be small
because of the reverse ow direction. Consequently, friction drag
constraint C d, f > 0.0025 is employed to rule out designs with
ow separation. In order to maintain laminar ow, and to improve optimization eciency, the quantity of the FPG, denoted as
Q FPG = dC p /d X , are applied in the optimization. On the front
part of airfoil, where X < 0.50, for both upper and lower surfaces
Q FPG should be larger than 0.0. dC p /d X is limited to be less than
3.0 on the pressure recovery region ( X > 0.70) to avoid trailing
edge separation [8].

Table 3
Parameter ranges of the CST design variables.
Upper surface parameter number
Parameter range

[0.05, 0.3]

[0.0, 0.3]

Lower surface parameter number

Parameter range

[0.3, 0.0]

[0.3, 0.05]

[0.2, 0.1]

[0.0, 0.3]

Y. Zhang et al. / Aerospace Science and Technology 43 (2015) 152164


Fig. 8. Iteration history of the NLF supercritical airfoil optimization. (For interpretation of the references to color in this gure, the reader is referred to the web version of
this article.)

Fig. 9. Airfoil shapes and pressure distributions of the airfoils.


C d, p


Ma = 0.739;

Re = 1.0E7;

C l = 0.55

t /c = 12.35%
R LE < 0.035;

R LE 0.010

C m 0.10
C d, f 0.0025
Q FPG |Upper and Lower, X <0.50 > 0.0

dC p /d X Upper and Lower, X >0.70 < 3.0


In the genetic algorithm, 32 individuals form the population of

each generation. The rst generation contains the CST parameters
of the RAE2822 airfoil and 31 random parameter combinations.
Optimization is terminated after 100 generations. Total number of
individuals generated in this process is 3200. However, only about
2100 individuals are computed by CFD because duplicated individuals could be generated due to the elite strategy of NSGA-II [7].
Total computation time is about 6 hours on a 32-core 2.8 GHz
Fig. 8 shows the iteration history of the optimization. Each airfoil is given a Design ID, which is equal to generation population + individual number. Fig. 8(a) shows all results of the
optimization process. The red points are unfeasible designs which
do not fully satisfy all constraints of Eq. (4). The black points are
feasible designs that satisfy all of the constraints. The results show
that feasible designs rst appear after ID > 250, and a lot of un-

feasible designs have less pressure drag than the feasible designs.
Fig. 8(b) shows the history of feasible designs. A clear convergent
tendency could be seen when ID > 1500. Furthermore, feasible designs become less and less when ID > 2500 because individuals
identical to already exist designs are produced more and more frequently, which means convergent solution is approached.
Three airfoil shapes and pressure distributions in the optimization process are shown in Fig. 9. Corresponding Mach number
contours are presented in Fig. 10. The original design (called as
ORD) in the gures is the scaled RAE 2822 airfoil of the rst generation. The ORD is unfeasible because FPG on the lower surface is
not sucient for NLF design, and the pitching moment is also too
The rst feasible design (called as FFD) is the airfoil which rst
satises all constraints. The maximum thickness location of FFD
moves towards the trailing edge to obtain a longer region of FPG
on the lower surface, as in Fig. 9(b). However, Mach number ahead
of shock exceeds 1.2, as in Fig. 10(b), and the shock wave is very
strong. The optimized design (called as OPD) is the design which
has the minimum pressure drag in all of the feasible designs. It is
from the 97th generation. Lower surfaces of the FFD and OPD are
similar. The FPG region is longer than 50% of the chord on both
upper and lower surface. Result shows that, at the current design
Mach number and lift coecient, shock wave might be unavoidable. However, the shock wave of the OPD is obviously weaker than
the ORD and FFD. The Mach number ahead of shock is 1.12, as in
Fig. 10(c), which is an acceptable value for transonic airfoil.


Y. Zhang et al. / Aerospace Science and Technology 43 (2015) 152164

Fig. 10. Mach number contours of the three airfoils in optimization process. (For interpretation of the colors in this gure, the reader is referred to the web version of this
Table 4
Aerodynamic coecients of the three airfoils (Ma = 0.739, Re = 10.0 million, C l = 0.55).
Computation method

Aerodynamic coecient

Original design (ORD)

First feasible design (FFD)

Optimized design (OPD)

Fixed transition

Pressure drag C d, p
Friction drag C d, f
Pitching moment C m




Natural transition

Pressure drag C d, p
Friction drag C d, f
Total drag C d
Lift to drag ratio L / D
Transition positions (upper/lower surfaces)




Fig. 11. Pressure distributions and friction coecient distributions of the ORD and OPD.

The three airfoils are validated by natural transition computation. Results of the xed transition and natural transition computations are compared in Table 4. It could be seen that the pressure
drag from both computation methods are close to each other. After
optimization, the L / D is increased by about 20. Most of the drag
reduction comes from the pressure drag. The friction drag is also
reduced by a little, although such a reduction is not expected by
using such a process.
Fig. 11(a) shows pressure distributions of natural transition
computation and xed transition computation. Pressure distributions of the two methods show only a little difference. The results
prove that the xed transition computation could well predict the
pressure distribution of the airfoils. Fig. 11(b) shows friction coef-

cient distributions of the ORD and OPD airfoils which are both
computed by the SST Re transition computations. The transition location is closely related to the length of FPG region. As in
Fig. 11(b), laminar region of the lower surface is increased by a lot
after optimization. However, laminar region on the upper surface is
a little bit reduced because the shock wave location is moved back
towards the leading edge. However, it is necessary penalty paid
to weaken the shockwave. Nevertheless, laminar region length of
the OPD is more than 55% for both upper and lower surfaces. It
is a little longer than the pre-assumed transition locations for the
xed transition computation, which means that the FPG constraint
is adequate to generate such a laminar ow region at this Reynolds
number. The optimization method with xed transition computa-

Y. Zhang et al. / Aerospace Science and Technology 43 (2015) 152164


Fig. 12. Design histories of drag coecients of different FPG constraints.

Fig. 13. Shapes and pressure distributions of the airfoils with different Q FPG constraints.

Fig. 14. Friction coecient distributions of the six airfoils at different Reynolds numbers (Ma = 0.739, C l = 0.55).

tion and pressure gradient constraint can produce reasonable and

satisfactory designs.
3.4. Effect of Q FPG on transition
In last section, the expected pressure gradient can be obtained
by setting a constraint on Q FPG through the optimization process,
and the expected laminar length can be achieved by maintaining a

reasonably FPG. A question is raised, what should be a good Q FPG

constraint. Is it the larger the better? Is it related to Reynolds numbers?
With the pressure gradient constrained optimization, we are
able to design and compare a series of airfoils with different pressure distribution gradients, or different length of FPG. With the
pressure drag minimized by the optimization, such a process could
provide an insight into how the pressure gradient is affecting the
transition performance, taking into account not only the friction
drag, but also the robustness and Reynolds number effects.


Y. Zhang et al. / Aerospace Science and Technology 43 (2015) 152164

The optimization problem is dened as in Eq. (5). Different

pressure gradients from Q FPG > 0.0 to Q FPG > 1.0 are set before
the 50% chord of upper surface. The length of FPG on the lower
surface is expected to be more than 60%. In the optimization process, the xed transition locations of the upper and lower surfaces
are set at 52% and 62%, respectively. The optimized design in Section 3.3 is put into the rst generation as the original design in the
following optimizations.


C d, p


Ma = 0.739;

Re = 1.0E7;

C l = 0.55

t /c = 12.35%
R LE < 0.035;

R LE 0.010

C m 0.10
C d, f 0.0025
Q FPG |Upper, X <0.50 > 0.0, 0.2, 0.4, 0.6, 0.8, or 1.0

Fig. 15. Friction drag and total drag of the six airfoils varying with Reynolds number
(Ma = 0.739, C l = 0.55).

Q FPG |Lower, X <0.60 > 0.05

dC p /d X 

Upper and Lower, X >0.70

< 3.0


After several days of running on a 32-core workstation, we

got six optimized airfoils corresponding to different upper surface
pressure distribution constraints, which are designated as Q FPG >
0.0, 0.2, 0.4, 0.6, 0.8, and 1.0 airfoils, respectively. Fig. 12 shows
the drag coecient histories of the different FPG constraints. Only
the feasible designs are plotted in the gures. All cases are iterated
about 90100 generations. With the increase of Q FPG , feasible designs are less and less. It is clear that the drag increases with Q FPG
increasing, because the shock wave drag is increased, while the
friction drag is almost unaffected in the xed transition computation.
Fig. 13 compares the shapes and the pressure distributions of
four airfoils out of the six and the original airfoil. Shapes and lower
surface pressure distributions are similar. Lengths of the FPG on
the lower surface are more than 60% chord. The shock wave becomes stronger and stronger as the specied limitation of Q FPG
Based on the above six airfoils, ve Reynolds numbers ranging
from 10 million to 20 million are computed by transition model to
test the sensitivity of pressure distribution to the FPG constraints.
Different Reynolds numbers could represent ow conditions of
different types of aircrafts with different size, conguration and
speed. Reynolds number less than 1.0 107 could be the ight
condition of business jet, and Reynolds number 1.5 107 represents the typical condition of regional jet, while Reynolds number
higher than 2.0 107 might correspond to the narrow body airliners. In the present computation, the free stream eddy viscosity is
0.1 times of dynamic viscosity, and turbulence intensity is 0.1%, as
the airfoils are supposed to be in ight rather than in wind tunnel.
Fig. 14 shows friction coecient distributions of the six airfoils at three Reynolds numbers with the same lift coecient.
As shown in Fig. 14, all of the airfoils have quite large areas of
laminar ow, which indicates that the constraints of Q FPG are adequate to obtain laminar ow. From all the ve Reynolds number
studied, we could gure out some transition tendencies from the
(1) When the Reynolds number is 1.0 107 , transition locations
on the upper surface are all later than 50% chord. Remember that
the FPG region is constrained to be longer than 50%, actual length
of FPG on the upper surface is determined by shock wave location,
so are the transition locations. Transition locations on the lower
surfaces keep the same for different airfoils, no shock is detected
on the lower surface. All these facts indicate that the transition
location is not sensitive to the quantity of FPG at this Reynolds
number range, but only to the shock location.

(2) For Reynolds number Re = 1.5 107 , transition location of

the upper surface is mostly before 50% chord. The differences are
hence not produced by the shock, but by the intensity of FPG.
Transition locations move downstream with the Q FPG increasing.
The only exception is the Q FPG > 1.0 airfoil. The shock wave of this
airfoil is very strong and induces a small separation bubble, which
might inuence the transition location. From airfoils of Q FPG > 0.0
to Q FPG > 0.8, transition locations of the upper surfaces move from
35% to 53% chords. For lower surfaces, transition locations are also
all before 60% chord, which is also less than the assigned FPG
range. At this Reynolds number, the pressure gradient should be
carefully designed to get adequate laminar length.
(3) When the Reynolds number is 2.0 107 , the effect of Q FPG
becomes small. Variations of the upper surface transition locations
are less than 6% chord. The longest laminar region is less than 30%
chord, which is on the Q FPG > 0.8 airfoil. For such a large Reynolds
number, a large range of NLF is dicult to realize. Hybrid NLF and
laminar control measures must be considered.
Fig. 15 shows friction drag and total drag of the six airfoils at
different Reynolds numbers. Friction drag of the Q FPG > 0.0 airfoil is the largest. By increasing the Q FPG from 0.0 to 1.0, benet
of friction drag achieved is about 4 drag counts with Reynolds
number less than 1.5 107 . However, the benet drops to less
than 2 drag counts when Reynolds is 2.0 107 . Total drag of
the Q FPG > 1.0 airfoil is much larger than other airfoils, as the
shock wave is quite strong. The total drags of the Q FPG > 0.0 airfoil and the Q FPG > 0.4 airfoil are almost the same when Reynolds
numbers less than 1.5 107 , which means the variations of friction drag and pressure drag are balanced on these two points.
However, when Reynolds number is 2.0 107 , the total drag of
Q FPG > 0.0 airfoil and Q FPG > 0.2 airfoil are about the same. The
Q FPG > 0.2 airfoil is the best one which has the least total drag for
all Reynolds numbers. It worth to be mentioned that constraints
of the recovery region ( X > 0.70) and the pitching moment greatly
limit the application of aft loading, which has the potential of further reducing shock wave drag.
Robustness of transition location is also an important criterion
for the evaluation of NLF airfoil. Flight could be unsafe if aerodynamic forces are not stable. Flight control problem would arise if
the aerodynamic derivatives change in cruise range. In the present
computation, the airfoils are computed at different AoAs to test
robustness. Fig. 16 shows polar curves of the Q FPG > 0.0, 0.2 and
0.4 airfoils, whose cruise drag coecients are close. Fig. 17 shows
the corresponding transition locations. Clear drag buckets can be
seen on the curves of Fig. 16. The lower extent (C l about 0.1
to 0.2) of the drag bucket of Q FPG > 0.0 airfoil, which is dom-

Y. Zhang et al. / Aerospace Science and Technology 43 (2015) 152164


Fig. 16. Polar curves of three airfoils at Reynolds number 10.0M and 12.0M.

Fig. 17. Transition locations varying with angles of attack.

Fig. 18. Pressure distributions varying with angles of attack (Ma = 0.739, Re = 10.0M).

inated by the lower surface transition, is larger than the other

two airfoils; however, the lift coecients may not appear in real
ight. The upper extents of the drag buckets are quite different.
Drag of the Q FPG > 0.0 airfoil shows a sudden rise when C l > 0.4
(AoA > 0.75 ), which demonstrates that premature transition occurs on the upper surface. As shown in Fig. 17, the upper surface
transition locations of Q FPG > 0.0 airfoil at AoAs from 0.5 to 1.0
are not stable, which means poor robustness. The other two airfoils
show better performance near the lift coecient of cruise range
(0.6 > C l > 0.4).

The cause of the premature transition could be explained by

looking into the pressure distribution. Fig. 18 shows the pressure
distributions varying with AoAs. With the AoA deviating from the
design point, the pressure distribution cannot always keep the favorable gradient. There will be an adverse region on the suction
roof-top. The adverse pressure gradient should be the cause of
premature transition. For the Q FPG > 0.0 airfoil, adverse pressure
gradient is strong when AoA is less than 1 ; However, for the
Q FPG > 0.2 and 0.4 airfoils, the adverse pressure gradients only appear at 0.5 and are much weaker. Besides the increase in friction


Y. Zhang et al. / Aerospace Science and Technology 43 (2015) 152164

drag, the premature transition also changes the slope of lift coefcient, as shown in Fig. 19. From the above results, we could see
that the Q FPG > 0.2 airfoil is the best airfoil with both low cruise
drag and good robustness.
4. High-subsonic NLF wing design

Fig. 19. Lift coecient curves of the Q FPG > 0.0 and Q FPG > 0.2 airfoils.

Fig. 20. Control airfoil locations of the NLF wing.

Fig. 21. Control airfoils of the NLF wing.

In this section, the Q FPG > 0.2 airfoil is used as the baseline airfoil of high-subsonic wing. Based on the conclusions in Section 3,
the FPG region of the lower surface is expected before only 50%
chord. The baseline airfoil is optimized again with the new design
constraints. Fig. 20 shows the control proles, which are located at
the root, kink and tip locations. The relative thicknesses of the root,
kink and tip locations are 15%, 12% and 10%, respectively. Rule of
cosine [5] is used to obtain the 2-D airfoil thickness. Spline-based
surface is adopted to ensure smoothness in the span-wise direction. The baseline airfoil is installed at the kink location. The wing
is installed onto a wing-body conguration to validate the performance. Cut and try method is employed to adjust the root and
tip airfoils. At the beginning of wing design, the upper surfaces
of the root and tip airfoils are copied from the kink airfoil, while
the lower surfaces are scaled from the kink airfoil to t the thickness requirements. The kink airfoil is kept unchanged during the
design process. After about 30 iterations of manual modications
and CFD computations, an NLF wing is obtained. Fig. 21 shows the
control airfoils of the NLF wing. The upper surface shapes of the
airfoils are similar with each other. In contrast, the lower surfaces
are quite different because of the differences of the relative thickness.
Fig. 22 shows surface pressure contour and pressure distributions at the cruise condition Ma = 0.76, C l = 0.42, Re = 1.6E7
(based on mean aerodynamic chord). The computation grid involves 15 million points, and the Y+ of the rst grid layer is less
than 1.0. The isolines of the pressure in the span-wise direction
are almost straight, as shown in Fig. 22(a). Clear FPG can be seen
in the stream-wise direction, as in Fig. 22(b). Fig. 23 shows the
skin friction contour and friction contours of the wing/body. Both
the upper and lower surfaces have large areas of the laminar ow.
Performances of the cruise Mach number 0.76 and low Mach
number 0.2 are also validated by the CFD method. Fig. 24 shows
lift and lift-drag ratio curves of the two Mach numbers. At the
cruise Mach number, the lift coecient remains linearly increasing
until AoA = 3.5 , on which the lift coecient is about 0.77, that is
larger than 1.3 times of the cruise lift coecient 0.42. Moreover,
the lift-drag polar does not have unexpected uctuation. The lift

Fig. 22. Surface pressure contour and pressure distributions at four sections (Ma = 0.76, C l = 0.42, Re = 1.6E7). (For interpretation of the colors in this gure, the reader is
referred to the web version of this article.)

Y. Zhang et al. / Aerospace Science and Technology 43 (2015) 152164


Fig. 23. Surface friction contour and friction coecient distributions at four sections (Ma = 0.76, C l = 0.42, Re = 1.6E7). (For interpretation of the colors in this gure, the
reader is referred to the web version of this article.)

Fig. 24. Lift coecient and lift-drag polar curves of the wing/body conguration.

to drag ratio is 22.95 when C l = 0.42 for the wing/body conguration. Low speed stall behavior also shows good performance. The
maximum lift coecient is larger than 1.25, which shows that the
constraints of minimum leading edge radius are effective to guarantee low speed performance.
5. Conclusions
An optimization method of supercritical NLF airfoil is introduced in this paper. The work could be concluded as:
(1) A supercritical NLF airfoil optimization method is developed based on genetic algorithm and CFD. With design expectation
of laminar length, the optimization method uses a xed transition computation to evaluate the friction drag and consider the
inuence of laminar boundary layer, rather than accurately predicting the transition location. With the aid of FPG constraint, the
design expectation of laminar length could be guaranteed, which
is validated by the SST Re transition model. The optimization can then be focused on minimizing the pressure drag (mainly
the shock drag), and the computational consumption is greatly reduced. With the help of FPG constraint in optimization, airfoils
with a specied quantity of favorable gradient, which is believed
to be the key issue of preserving laminar ow of NLF airfoil, can
be designed.
(2) Effects of the FPG on transition location are studied in this
paper. The results show that for moderate Reynolds number, the
variation of FPG has the most signicant effect on the laminar
ow region. When the Reynolds number is small, instead of natural

transition, the length of the laminar region is mainly determined

by the shock wave. Moreover, when the Reynolds number is large,
the transition location is only slightly affected by the FPG.
(3) Favorable pressure gradient of dC p /d X > 0.2 is a good
compromise for supercritical NLF airfoil in the design condition of
present paper. With such a constraint, the optimized airfoil can
have good performance on both cruise drag and robustness. Airfoil with lower FPG will be easy to endure a premature transition
at off-design AoA. The premature transition, causing sudden rise of
drag and change of the lift slope near the cruise condition, needs
to be carefully treated when designing an airfoil.
(4) The optimized supercritical NLF airfoil is applied to an NLF
wing design as its baseline airfoil. The good supercritical NLF properties of the airfoil can be well preserved in the wing.
Conict of interest statement
The authors declared that they have no conicts of interest to
this work.
[1] B.T. Anderson, R.R. Meyer, Effects of wing sweep on in-ight boundary-layer
transition for a laminar ow wing at Mach numbers from 0.60 to 0.79, NASA
Technical Memorandum 101701, 1990.
[2] A.P. Antunes, J.L.F. Azevedo, R.G. Silva, A framework for aerodynamic optimization based on genetic algorithms, AIAA paper 2009-1094, 2009.
[3] D. Arnal, J. Perraud, A. Sraudie, Attachment line and surface imperfection
problems, VKI course Advances in Laminar-Turbulent Transition Modeling,
912 June 2008, Brussels, EN-AVT-151-09.


Y. Zhang et al. / Aerospace Science and Technology 43 (2015) 152164

[4] K. Biber, C.P. Tilmann, Supercritical airfoil design for future high-altitude longendurance concepts, J. Aircr. 41 (1) (2004) 156164.
[5] U. Cella, D. Quagliarella, R. Donelli, B. Imperatore, Design and test of the
UW-5006 transonic natural-laminar-ow wing, J. Aircr. 47 (3) (2010) 783795.
[6] F.S. Collier, An overview of recent subsonic laminar ow control ight experiments, AIAA Paper 1993-2987, 1993.
[7] K. Deb, A. Pratap, S. Agarwal, T. Meyarivan, A fast and elitist multiobjective
genetic algorithm: NSGA-II, IEEE Trans. Evol. Comput. 6 (2) (2002) 182197.
[8] P. Edi, J.P. Fielding, Civil-transport wing design concept exploiting new technologies, J. Aircr. 43 (4) (2006) 932940.
[9] B. Eggleston, R.J.D. Poole, D.J. Jones, M. Khalid, Thick supercritical airfoils with
lowdrag and natural laminar ow, J. Aircr. 24 (6) (1987) 405411.
[10] A. Filippone, Data and performances of selected aircraft and rotorcraft, Prog.
Aerosp. Sci. 36 (2000) 629654.
[11] M. Fujino, Design and development of the Honda jet, J. Aircr. 42 (3) (2005)
[12] M. Fujino, Y. Yoshizaki, Y. Kawamura, Natural-laminar-ow airfoil development
for the Honda jet, AIAA paper 2002-2932, 2002.
[13] M. Fujino, Y. Yoshizaki, Y. Kawamura, Natural-laminar-ow airfoil development
for a lightweight business jet, J. Aircr. 40 (4) (2003) 609615.
[14] G.A. Garzon, J.R. Matisheck, Supersonic testing of natural laminar ow on sharp
leading edge airfoils. Recent experiments by Aerion corporation, AIAA paper
2012-3258, 2012.
[15] B.J. Holmes, C.J. Obara, Observations and implications of natural laminar ow
on practical airplane surfaces, J. Aircr. 20 (12) (1983) 9931006.
[16] R.D. Joslin, Overview of laminar ow control, NASA/TP 1998-208705, 1998.
[17] M. Khalid, D.J. Jones, A summary of transonic natural laminar ow airfoil development at NAE, Aeronautical note NAE-AN-65, NRC No. 31608, 1990.
[18] B.M. Kulfan, A universal parametric geometry representation method CST,
AIAA paper 2007-62, 2007.
[19] T. Lammering, E. Anton, K. Risse, K. Franz, R. Hoernschemeyer, Gains in fuel efciency: multi-stop missions vs. laminar aircraft, AIAA paper 2011-6885, 2011.
[20] K.A. Lane, D.D. Marshall, Inverse airfoil design utilizing CST parameterization,
AIAA paper 2010-1228, 2010.
[21] J.D. Lee, A. Jameson, Natural-laminar-ow airfoil and wing design by adjoint
method and automatic transition prediction, AIAA paper 2009-897, 2009.

[22] J.Q. Luo, J.T. Xiong, F. Liu, Aerodynamic design optimization by using a continuous adjoint method, Sci. China, Phys. Mech. Astron. 57 (7) (2014) 13631375.
[23] F.R. Menter, M. Kuntz, R. Langtry, Ten years of industrial experience with the
SST turbulence model, Turbul. Heat Mass Transf. 4 (2003) 625632.
[24] F.R. Menter, R.B. Langtry, S.R. Likki, et al., A correlation-based transition model
using local variables Part I: Model formulation, J. Turbomach. 128 (2006)
[25] F.R. Menter, R.B. Langtry, S.R. Likki, et al., A correlation-based transition model
using local variables Part II: Test cases and industrial applications, J. Turbomach. 128 (2006) 423434.
[26] NLR 7301 Airfoil, Experimental data base for computer program assessment,
AGARD Rept. AR-138, Neuilly sur Seine, France, 1979.
[27] A. Ni, Y. Zhang, H. Chen, An improvement to NSGA-II algorithm and its application in optimization design of multi-element airfoil, Acta Aerodyn. Sin. 32 (2)
(2014) 252257.
[28] H.L. Reed, W.S. Saric, Attachment-line heating in a compressible ow, AIAA paper 2011-3242, 2011.
[29] D.M. Somers, Design and experimental results for a natural-laminar-ow airfoil for general aviation applications, Technical report NASA TP-1861, Langley
Research Center, 1981.
[30] T. Streit, G. Wichmann, F. Knoblauch, R.L. Campbell, Implications of conical ow
for laminar wing design and analysis, AIAA paper 2011-3808, 2011.
[31] P. Sturdza, Extensive supersonic natural laminar ow on the Aerion business
jet, AIAA paper 2007-685, 2007.
[32] G. Wang, H.H. Mian, Z.Y. Ye, J.D. Lee, Numerical study of transitional ow
around NLR-7301 airfoil using correlation-based transition model, J. Aircr.
51 (1) (2014) 342349.
[33] Y. Zhang, H. Chen, S. Fu, et al., A practical optimization design method
for transport aircraft wing/nacelle integration, Acta Aeronaut. Astronaut. Sin.
33 (11) (2012) 19932001 (in Chinese).
[34] Y. Zhang, H. Chen, M. Zhang, et al., Supercritical wing design and optimization
for transonic civil airplane, AIAA paper 2011-27, 2011.
[35] Y. Zhang, H. Chen, W. Zhang, et al., Wing/engine integrated optimization based
on NavierStokes equations, AIAA paper 2012-1046, 2012.
[36] K. Zhao, Z.H. Gao, J.T. Huang, Robust design of natural laminar ow supercritical airfoil by multi-objective evolution method, Appl. Math. Mech. 35 (2)
(2014) 191202.