Using genetic algorithms to optimize airfoils in

incompressible regime

Adrian DINA*,1,2, Sterian DANAILA2, Mihai-Victor PRICOP1, Ionut BUNESCU1,2

*Corresponding author
*,1INCAS – National Institute for Aerospace Research “Elie Carafoli”,
B-dul Iuliu Maniu 220, 061126 Bucharest, Romania,
[email protected]*, [email protected], [email protected]
2
“POLITEHNICA” University of Bucharest, Faculty of Aerospace Engineering,
Department of Aerospace Sciences “Elie Carafoli”,
Str. Polizu 1-7, sector 1, 011061 Bucharest, Romania,
[email protected]
DOI: 10.13111/2066-8201.2019.11.1.6

Received: 07 January 2019/ Accepted: 03 February 2019/ Published: March 2019

Copyright © 2019. Published by INCAS. This is an “open access” article under the CC BY-NC-ND
license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

International Conference of Aerospace Sciences “AEROSPATIAL 2018”

25 - 26 October 2018, Bucharest, Romania, (held at INCAS, B-dul Iuliu Maniu 220, sector 6)
Section 1 – Aerodynamics

Abstract: Aerodynamic optimization is a very actual problem in aircraft design and airfoils are basic
two-dimensional shape forming cross sections of wings. Traditionally, the airfoil geometry if defined
by a very large number of coordinates. Nowadays, in order to simplify the optimization, the airfoil
geometry is approximated by a parametrization, which enables to reduce the number of needed
parameters to as few as possible, while effectively controlling the major aerodynamic features. The
present work has been done on the Class-Shape function Transformation method (CST) [1, 2]. Also,
the paper introduces the concept of Genetic Algorithm (GA) to optimize a NACA airfoil for specific
conditions. A Matlab program has been developed to implement CS into the Global Optimization
Toolkit. The pressure distribution lift and drag coefficients of the airfoil geometries have been
calculated using two programs. The first one is an in-house code based on the Hess-Smith [3] panel
technique and on the boundary layer integral equations, while the second is an XFOIL program. The
optimized airfoil has improved aerodynamic characteristics as compared to the original one. The
optimized airfoil is validated using the Ansys-Fluent commercial code.
Key Words: optimization, genetic algorithms, parametrization, XFOIL, aerodynamic models

1. INTRODUCTION
The airfoil optimization remains an actual topic of research in the frame of multidisciplinary
aircraft optimization. Due to the large number of coordinate values needed to define the
shape of an airfoil, a different types of parameterization have been developed [1, 2, 4, 5, 6].
Genetic Algorithm is a robust and accurate method for global aerodynamic shape
optimization and this has been suggested in the literature [7, 8, 10]. This paper refers to the
capitalization of low cost aerodynamic computational methods based on potential flow in
aerodynamic optimization. Such methods can serve as a preliminary stage for optimization
using more accurate techniques. The second section describes an approach to the airfoil
INCAS BULLETIN, Volume 11, Issue 1/ 2019, pp. 79 – 90 (P) ISSN 2066-8201, (E) ISSN 2247-4528
 Adrian DINA, Sterian DANAILA,Mihai-Victor PRICOP, Ionut BUNESCU 80

optimization based on evolutionary algorithms. The third section includes a discussion of

different designs parameterization. The fourth section deals with finding the
appropriate/optimal solution of the flow around airfoils and explains the basic theoretical
concepts of two-dimensional flows also describing the assessment of the XFOIL program
performance. Finally, the numerical results and main conclusions are presented.

2. GENETIC ALGORITHM
The basic rule of the Genetic Algorithm (GA) is to search for optimal solutions using an
analogy with the theory of evolution [9, 10]. Starting with an initial population composed by
a number of candidate solutions (designated as chromosomes), these parents are manipulated
using various operators (combination, crossover, or mutation) to create a new set of
chromosomes for the next generation. While the genetic operators are random, the genetic
algorithm is not completely random. During the evolution of the solution the chromosomes
are ranked in respect to the optimization criteria (the fitness). Only the higher-ranking
chromosomes are selected to continue to the next generation. Once the new generation is
created, its chromosomes are then evaluated for fitness and the process continues until an
imposed convergence condition is satisfied or until the quasi steady population was reached.
The basic genetic algorithm important steps are presented in Fig. 1.
STAR
SELECTION OF
PARENTS
GENERATE INITIAL
POPULATION
CROSSOVER TO
PRODUCE
“CHILDREN”
CALCULATE
FITNESS OF
INDIVIDIALS
MUTATION OF
“CHILDREN”

YE NO
SATISFY CALCULATE
STOP FITNESS OF
CRITERION ? CHILDREN

NEW GENERATION
“ELITISM”
END
Fig. 1 - Flow Chart of the Matlab Genetic Algorithm
The main genetic operators are: selection of parents, recombination and the mutation.
We will focus on the functional description of each operator implemented in Matlab code.
The roulette selection method refers to the fact that the best individuals are preferred, but not
always selected. The worst individuals, which are not always excluded, are kipped in order
to maintain the variability in each generation. Cross over is performed to combine the
INCAS BULLETIN, Volume 11, Issue 1/ 2019
 81 Using genetic algorithms to optimize airfoils in incompressible regime

desirable characters of two different parents which are selected for mating. The method of
cross over depends on the kind of problem to be solved and the method of encoding. In this
work, a single point, randomly chosen, was chosen to cut the string. So, two strings and two
queues are produced. Then the queues were changed to produce two new individuals.
Mutation is the second way through which GA explores the search domain. It can
introduce features that are not in the initial population and avoid premature convergence.
The mutation points are randomly selected. Increasing the number of mutations increases the
freedom of the algorithm to look outside the workspace region. It also tends to distract the
convergence algorithm from a local solution.

3. AIRFOIL PARAMETRIZATION
The Airfoil parameterization method is extremely important for aerodynamic optimization
due the important influence on the nonlinearity of the optimization problem. There are
several main criteria for selecting the most representative parameterization type: a) the
number of parameters used for the geometric representation should be as small as possible;
b) the method should be able to reproduce a variety of profiles; c) any constrain imposed on
profile geometry should be easy to formulate and applied; d) parameterization should be
effective in the optimization process. Several types of parameterizations have been studied
such as:
• NACA parametrization.
Early airfoil design was based on approximate theoretical models, the entire NACA 4
and 5 digits families were created using this method. For example, NACA 4 digits airfoils
are describe by the equation:
y t = 5t[0.2969 x − 0.1260 x − 0.3516 x 2 + 0.2843 x 3 − 0.1015 x 4 ] (1)
 m x x
(2 p( ) − ( ) 2 ) 0 ≤ x ≤ pc
 p 2 c c
yc =  . (2)
m x x
 ((1 − 2 p) + 2 p( ) − ( ) 2 ) pc ≤ x ≤ c
 (1 − p 2 ) c c
where t is the maximum thickness as fraction of the chord, m represents the maximum
camber as 1 / 100 from the chord, p represents the position of the maximum chamber as
1 / 10 from the chord, c represents the chord, y c represents the equation of curvature and y t
represents the equation for thickness.
• Bezier parametrization.
The Bezier parametrization uses the piecewise Bezier polynomials approximations of
curves, which in addition ensure some smoothness of the approximating curve. The Bezier
curve can be represented as:
n
B (t ) = ∑B
t =0
n
i (t ) Pi
, (3)

where n is the polynomial degree, i is the index, and t is the variable.

• Hicks-Henne parametrization.
Hicks and Henne (1978) introduced a compact formula for modeling small or moderate
perturbations of “baseline” airfoil shapes. Given an airfoil, the method generates new shapes
with a generic disturbance function called “bumps”. The bumps function is defined as
follows:
INCAS BULLETIN, Volume 11, Issue 1/ 2019
 Adrian DINA, Sterian DANAILA,Mihai-Victor PRICOP, Ionut BUNESCU 82

ln(0.5)
f i ( x) = sin t ( πx mi ) mi = (4)
ln( x M i )

Where x M i is the maximum position of the bumps function that can vary between one
and zero and t is the thickness of the jump.
• PARSEC parametrization.
This type of parameterization was first proposed by Sobieczky [4]. The key idea is
expressing the airfoil shape as an unknown linear combination of suitable base function, and
selecting 11 important geometric characteristics of the airfoil as the control variables, in such
a way that the airfoil shape can be determined from these control variables by solving a
linear system. To approximate the shape of the airfoil, 11 parameters are needed. The upper
side and lower side of the airfoil are represented as:
6 1 6 1
n− n−
y up = ∑
n =1
an x 2 , y lo = ∑ bn x
n =1
2
(5)

where the coefficients a n and bn can be determined by imposing geometric characteristics

conditions.
• Class Shape Transformation (CST) parametrization:
In this technique, introduced by Kulfan and Bussoletti [1, 2], the representation of the
airfoil is mapped as:
y ( x) = C ( x) S ( x) + Δz te 0 ≤ x ≤ 1. (6)
where S (x) is the shape function, C (x) represents the class function and ∆z te is. the trailing
edge thickness. The class function is defined by:
C ( x) = x N1 (1 − x) N 2 (7)
Where N1 and N 2 define a specific class of shapes. For airfoils having the rounded
leading edge, N1 = 0.5 and N 2 = 1 . For shape function S (x) any polynomial can be chosen,
but it is an advantage to select a family of Bernstein polynomials. The general form of a
Bernstein polynomial of the n order is:
n
BPn( x) = ∑K
r =0
r ,n x
r
(1 − x) n −r . (8)

Fig. 2 - Bernstein polynomials function

where K r ,n is a binomial coefficient

INCAS BULLETIN, Volume 11, Issue 1/ 2019

 83 Using genetic algorithms to optimize airfoils in incompressible regime

n n!
K r ,n =   = , (9)
 r  r!(n − r )!
and the shape function S (x) yields:
n
S ( x) = ∑AK r =0
i r ,n x
r
(1 − x) n − r . (10)

For flexibility, it is convenient to represent the upper and the lower side independently as:
n
yup ( x) = x N (1 − x) N
1 2
∑ Au K
r =0
i r ,n x r (1 − x) n−r + ∆zte 0 ≤ x ≤1 (11)

n
ylo ( x) = x N1 (1 − x) N 2 ∑ Al K
r =0
i r ,n x
r
(1 − x) n − r + Δz te 0 ≤ x ≤ 1. (12)

where the coefficients Aui and Ali can be selected as parameters in the optimization process. In
our applications the order of Bernstein polynomials was n=6.

4. AERODYNAMIC MODEL
In an optimization process, an objective function that must be minimized has to be defined.
In our case the ratio of drag to the lift, CD / CL , was chosen. Then the fitness evaluation in
the optimization algorithm requires the prediction of these aerodynamic characteristics of the
each chromosome of successive populations.
Obliviously, a fast and relatively accurate aerodynamic model has to be implemented.
Consequently, the linear potential model completed with the boundary layer correction was
considered.
The panel method, which is the numerical method to solve the incompressible potential
equation uses a superposition of particular solutions representing sources, doublets and
vortices. The solution procedure for the panel technique consists of discretizing the surface
of the airfoil into straight line segments or panels (Fig. 3).
There are many choices as how to formulate a panel method but the simplest and
practical method was due to Hess and Smith [3]. We consider N+1 points equally distributed
over the airfoil. The numbering system starts at the lower surface trailing edge and proceeds
forward, around the leading surface and aft to the upper surface trailing edge [11].
node
N -1
N
N+1
2 1
4 3
panel
Fig. 3 - Representation of an airfoil with straight line segments [6]

Sources and vortices with constant intensity are distributed along the panels. Imposing
the slip condition on panel control points (usually the middle points) and the Kutta-Jukovski
condition on the trailing edge a linear system of equations for the singularity intensities
yields:

INCAS BULLETIN, Volume 11, Issue 1/ 2019

 Adrian DINA, Sterian DANAILA,Mihai-Victor PRICOP, Ionut BUNESCU 84

 A11 … A1i … A1N … A1, N +1   Q1   b1 

 ………    
      
 Ai1 … Aii … AiN … Ai , N +1   Qi   bi 
   =  , (13)
 ………      
 AN 1 … ANi … ANN … AN , N +1  Q   b 
  N   N 
 AN +1,1 … AN +1,i … AN +1, N … AN +1, N +1   γ  bN +1 
where the influence coefficients Aij and the right hand terms bi are given by the following
equations:
 rij +1  1
sin (θi − θ j ) ln  cos (θi − θ j ) β ij , i=1,N. j=1,N,
1
Aij = +
 rij  2π
(14)
2π  
  rij +1  
( ) ( )
N
1
=
Ai , N +1
2π
∑
j =1 
 cos θ i − θ j ln 
 rij   − sin θ i − θ j β ij  , i=1,N, (15)
   

sin (θ1 − θ j ) β1 j + sin (θ N − θ j ) β Nj −

1 
AN +=
2π 
1, j

(16)
 r1 j +1   rNj +1  
− cos (θ1 − θ j ) ln   − cos (θ N − θ j ) ln  ,
 r1 j   rNj  
   

  r1 j +1   rNj +1 
∑ sin (θ − θ j ) ln  + sin (θ N − θ j ) ln 
N
1
AN=  +
+1, N +1
2π
1
 r1 j   rNj 
j =1
     (17)
+ cos (θ1 − θ j ) β1 j + cos (θ N − θ j ) β Nj  ,

=bi V∞ sin sin (θi − α ) , bN +1 =

−V∞ cos (θ1 − α ) − V∞ cos (θ N − α ) . (18)
where the geometrical parameters are sketched in Fig. 4, and α is the angle of attack.
Once the system is solved, the induced velocities and the pressure coefficient on control
points can be calculated. Then, the lift coefficient results by summing the pressure forces on
all panels.

Fig. 4 - Geometrical parameters

INCAS BULLETIN, Volume 11, Issue 1/ 2019

 85 Using genetic algorithms to optimize airfoils in incompressible regime

Thwaites integral method [12] is used to calculate the laminar boundary layer parameters
starting from the stagnation point to the transition onset, according to the following relation:
0.45ν
x

θ2 = ∫U
5
e dx (19)
U e6 x =0

where U e ( x) is the velocity distribution along the airfoil surface determined by the panel
method and ν is the kinematic viscosity. After θ is found, the following correlations are
used to compute the shape factor H:
H= 2.61 − 3.75λ + 5.24λ 2 0 ≤ λ ≤ 0.1; (20)
0.0147
=
H 2.472 + −0.1 ≤ λ ≤ 0, (21)
0.107 + λ
where:
θ 2 due
λ= , (22)
ν dx
The shear stress and the friction coefficient are estimated by the following empirical
relation:
µU e τw ,
=τw (λ + 0.09)0.62 , C f =
θ 1 (23)
ρU e 2
2
The empirical criteria reported by Michel [14] are used in the present work to describe
the location of the transition due to the growth of Tollmien-Schlichting assumed to occur
when the local Reynolds number based upon the momentum thickness exceeds a critical
value determined by the equation,
Uex Uθ
Re θ,tr = 2.9 Re 0x.,4tr , Re x = , Reθ = e (24)
ν ν
where Reθ and Re x are the local Reynolds numbers based on momentum thickness and the
distance from the airfoil leading edge, respectively.
In the turbulent region of the boundary layer, the integral Head [13] method is employed to
predict the turbulent flow parameters. Head suggested a new shape parameter H 1 , given by
δ − δ*
H1 ≡ (25)
θ
and the evolution of H 1 along the boundary layer is given by the equation:
1 d
(UθH1 ) = 0.0306( H1 − 3) −0.6169 (26)
U dx
Equation (26) and Von Karman Momentum global equation:
Cf dθ θ dU e
= + (2 + H ) (27)
2 dx U e dx
are solved by moving from the transition location to the trailing edge. For closure Head
proposed:

INCAS BULLETIN, Volume 11, Issue 1/ 2019

 Adrian DINA, Sterian DANAILA,Mihai-Victor PRICOP, Ionut BUNESCU 86

H 1 = 3.3 + 0.8234( H − 1.1) −1.287 H ≤ 1 .6 (28)

H 1 = 3.3 + 1.5501( H − 0.6788) −3.064 H ≥ 1.6 (29)

Friction coefficient in Head method is calculated by Ludwig-Tillman formula:
−0.268 (30)
C f = 0.246(10 −0.678 H ) Re θ

The previously presented methods (the panel method and the boundary layer correction
method) were applied to develop a Matlab program to estimate the fitness of the
chromosomes in genetic algorithm. This variant of the resulting optimization program will
be denoted in the following as Optaero.
A second code was developed in order to? check the results and to quantify the
influence of the accuracy in evaluation of the aerodynamic characteristics. This code denoted
Optx, uses the XFOIL program to calculate the lift and drag of a given airfoil. XFOIL is a
free software aerodynamic code released under the General Public License. The flow
solution in XFOIL is based on linear vortex panel method, coupled with a boundary layer
model.

5. NUMERICAL RESULTS AND CONCLUSIONS

In the following we will present some numerical solutions prescribed by the two codes:
OptAero and Optx. Both codes are based on the Global Optimization Toolkit of Matlab, but
with different flow solvers, as previously mentioned. Assuming a CST parametrization of the
airfoil shape, equations (11) and (12), the design parameters were the coefficients Aui and
Ali , i=1,6. The airfoil was presumed having the rounded leading edge, the flow
incompressible and a small angle of attack (0-0.5 deg). The data required by the GA were:
the population number of 40, the recombination factor of 0.4, the mutation factor of 0.5, and
the convergence criterion of 10−3 . Four cases are presented, corresponding to four values of
the Reynolds number Re = U ∞ c / V : 105, 5 105, 106, 5 106.
The results of the OptAero program are presented in Table 1. In this table, Cl is the lift
coefficient, C d is the drag coefficient, Δz te is the trailing edge thickness, p represents the
position of a maximum curvature, m is the maximum curvatures, b is the position of a
maximum thickness and t - the maximum thickness for the optimum airfoil.
Table 1 - Result from OptAero
OPTAERO
AIRFOIL 1 Cl = 0.29 m = 0.03 b = 0.232
α=0 Cd = 0.021 t = 0.15 ∆zte = 0.003
Re = 105 p = 0.406
AIRFOIL 2 Cl = 0.35 m = 0.031 b = 0.232
α=0 Cd = 0.010 t = 0.16 ∆zte = 0.003
Re = 5 ⋅105
p = 0.406

INCAS BULLETIN, Volume 11, Issue 1/ 2019

 87 Using genetic algorithms to optimize airfoils in incompressible regime

AIRFOIL 3 Cl = 0.34 m = 0.029 b = 0.375

α=0 Cd = 0.0058 t = 0.147 ∆zte = 0.003
Re = 106 p = 0.376
AIRFOIL 4 Cl = 0.36 m = 0.032 b = 0.345
α=0 Cd = 0.0058 ∆zte = 0.003
t = 0.147
Re = 5 ⋅106
p = 0.345

In Fig. 5 and 6 the airfoil polars are represented. We note the extremely large values of
the drag coefficient for low Reynolds number flows. These values are prescribed by the in-
house aerodynamic code.

Fig. 5 - Cl vs Alpha (code Optaero) Fig. 6 - Cl vs Cd (code Optaero)

The results of the second program, in which the aerodynamic performances are
predicted by XFOIL, are presented in Table 2. In Fig. 7 and 8 the polar of the optimum shape
are traced.
Table 2 - Result from OptX
OPTX
AIRFOIL 1 Cl = 0.4 m = 0.026 b = 0.287
α=0 Cd = 0.016 t = 0.130 ∆zte = 0.032
Re = 105 p = 0.206

AIRFOIL 2 Cl = 0.395 m = 0.0322 b = 0.345

α=0 t = 0.146 ∆zte = 0.032
Cd = 0.0073
Re = 5 ⋅ 105 p = 0.376

AIRFOIL 3 Cl = 0.4 m = 0.0321 b = 0.316

α=0 Cd = 0.0075 t = 0.150 ∆zte = 0.032
Re = 10 6 p = 0.469

AIRFOIL 4 Cl = 0.4 m = 0.030 b = 0.406

α=0 Cd = 0.0047 t = 0.139 ∆z = 0.032
te
Re = 5 ⋅106
p = 0.468

INCAS BULLETIN, Volume 11, Issue 1/ 2019

 Adrian DINA, Sterian DANAILA,Mihai-Victor PRICOP, Ionut BUNESCU 88

Again we made the polars for each airfoil for the same case as before α ∈ (−2...6) o and
Re = 106 :

Fig. 7 - Cl vs Alpha (code OptX) Fig. 8 - Cl vs Cd (OptX)

Fig. 9 - Airfoils geometry comparison between OptAero and OptX programs

Fig. 9 presents a comparison of results of programs.

6. CFD RESULTS
To verify, the obtained results for Airfoils-4 case from both programs were analyzed using
Ansys-Fluent. Analyzing the two profiles in ANSYS, for the case of a viscous flow in
incompressible regime at zero incidence using the k-omega SST method, we obtained the
values shown in Table 3.4.
Table 3 - Comparison with the Ansys-Fluent

Airfoil4 - OptAero Airfoil4 – OptX

- ANSYS Fluent Optimizer - ANSYS Fluent Optimizer
Cl 0.3469 0.3685 Cl 0.38002 0.4018

Cd 0.00579 0.00585 Cd 0.00573 0.00477

Cm -0.05062 -0.064 Cm -0.06789 -0.078

INCAS BULLETIN, Volume 11, Issue 1/ 2019

 89 Using genetic algorithms to optimize airfoils in incompressible regime

Fig. 10 - Pressure Coefficient OptAero

Fig. 11 - Pressure Coefficient OptX

7. CONCLUSIONS
Following the optimizations obtained, the following conclusions can be drawn:
• Designing an airfoil is a major activity in the aerodynamic design of an aircraft.
• Five methods for parameterization of the shape of an aerodynamic profile were
selected. A method called Class Forms (CST) was chosen to generate the curve, due to the
simplicity of the implementation and to the very small number of design parameters.
• Using a simplified aerodynamic model can accelerate the optimization process, but
the results will not be the most satisfactory. Using a more advanced aerodynamic coefficient
computation model, it can delay the process with an order of magnitude, but the results are
of better quality.
• By optimizing the airfoils at different Re numbers, we found that airfoils with an
increase in the number of Re profiles tend to become laminar, the maximum thickness and
curvature moving to the trailing edge of the profile.
• The genetic algorithm uses constraints that can be imposed both in the geometric
definition of the airfoil and in the aerodynamics characteristic.

INCAS BULLETIN, Volume 11, Issue 1/ 2019

 Adrian DINA, Sterian DANAILA,Mihai-Victor PRICOP, Ionut BUNESCU 90

ACKNOWLEDGEMENT
This paper is a part of the work in the bachelor’s thesis “Using genetic algorithm for airfoil
optimization in incompressible regime”, by Adrian DINA, awarded with the “Nicolae
TIPEI” prize during The International Conference of Aerospace Sciences
“AEROSPATIAL 2018”, 25 - 26 October 2018, Bucharest, Romania, organized by the
INCAS – National Institute for Aerospace Research “Elie Carafoli”. I would like to express
my deep gratitude to Professor PhD. Eng. Sterian DANAILA and Eng. Mihai-Victor
PRICOP, my research supervisors, for their patient guidance and useful critiques of this
research work. I would also thanks to Professor Sever TIPEI the son of Nicolae TIPEI and to
the INCAS institute, led by PhD. Eng. Catalin NAE, for their support and encouragement for
young researchers.

REFERENCES
[1] B. Kulfan, J. Bussoletti, “Fundamental” Parameteric Geometry Representations for Aircraft Component
Shapes, in: 11th AIAA/ISSMO multidisciplinary analysis and optimization conference, p. 6948, 2006.
[2] B. M. Kulfan, Universal parametric geometry representation method, Journal of Aircraft, 45.1: 142-158,
2008.
[3] S. Dănăilă, C. Berbente, Metode numerice în dinamica fluidelor, Bucureşti, Editura Academiei Romane,
2003.
[4] H. Sobieczky, Parametric airfoils and wings, in: Recent development of aerodynamic design methodologies,
Vieweg+ Teubner Verlag, p. 71-87, 1999.
[5] H. Sobieczky, Aerodynamic design and optimization tools accelerated by parametric geometry preprocessing,
in: European congress on computational methods in applied sciences and engineering, ECCOMAS, 2000.
[6] H.-Y. Wu, et al., Comparisons of Three Geometric Representations of Airfoils for Aerodynamic Optimization,
in: 16th AIAA computational fluid dynamics conference, p. 4095, 2003.
[7] D. A. Masters, et al., Review of aerofoil parameterisation methods for aerodynamic shape optimization, in:
53rd AIAA Aerospace Sciences Meeting, p. 0761, 2015.
[8] J. H. Holland, Genetic algorithms and the optimal allocation of trials, SIAM Journal on Computing, 2.2: 88-
105, 1973.
[9] J. H. Holland, Adaptation in natural and artificial systems: an introductory analysis with applications to
biology, control, and artificial intelligence, MIT press, 1992.
[10] L. D. Chambers, The practical handbook of genetic algorithms: applications, Chapman and Hall/CRC, 2000.
[11] J. Katz and A. Plotkin, Low-Speed Aerodynamics: From Wing Theory to Panel Methods, McGraw-Hill
Series in Aeronautical and Aerospace Engineering, McGraw-Hill, Inc., ISBN 0-07-100876-4, 1991.
[12] B.Thwaites, Approximate calculation of the laminar boundary layer, The Aeronautical Quarterly, 1.3: 245-
280, 1949.
[13] H. Schlichting and K. Gersten, Boundary-Layer Theory, Book, Publisher: Springer Berlin Heidelberg (Nov
1999), 8th Revised and Enlarged Edition, 2008.
[14] R. Michel, Etude de la transition sur les profile d’ aile; Edtablishment d’un critere de determination de point
de transition et calcul de la trainee de profile incompressible, ONERA Report 1/1578A, 1951.

INCAS BULLETIN, Volume 11, Issue 1/ 2019