=
+ =
n
i
i i
Airfoil
Initial i
x f x y x y
1
) ( ) ( ) , (
Introduction
Analytical Approach;
Control over Design Variables;
Cover Large Design Window;
Linearly Added to a Baseline Shape;
Participating Coefficient act as Design
Variables (
i
); and
Optimization Study to Evaluate Parameters
Population & Shape Functions
=
+ =
n
i
i i
Airfoil
Initial i
x f x y x y
1
) ( ) ( ) , (
i
Optimization
Shape Function Convergence Criteria
Convergence Measure Requirements
Flexibility & Accuracy; and
Library of Target Airfoils
Geometrical Convergence Process
Specify Base & Target Airfoil;
Select Shape Function;
Model Upper & Lower Surfaces;
Design Variable Population Size (2:10);
Perturbation of Design Variables;
Record Fitness  Geometrical Difference
of Target and Approximated Section;
Aggregate of Total Fitness; and
Geometrical Fitness vs. Aerodynamic
Performance
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.1
0.05
0
0.05
0.1
0.15
Comparison of Airfoil Shape Configuration for Geometrical Shape Parameterisation
x/c
y
/
c
Base: NACA 0015
Target 1: NASA LRN(1)1007
Target 2: NASA LS(1)0417Mod
Target 3: NASA NLF(1)1015
0.58c
0.2c
0.45c
0.010c
Camber
Location
0.022 0.3c 17% Target 2: NASA LS(1)0417Mod
0.06 0.4c 7% Target 1: NASA LRN(1)1007
0 0.3c 12% Base: NACA 0012
0.4c
Thickness
Location
0.047
Max.
Camber
15% Target 3: NASA NLF(1)1015
t/c Airfoil
0.58c
0.2c
0.45c
0.010c
Camber
Location
0.022 0.3c 17% Target 2: NASA LS(1)0417Mod
0.06 0.4c 7% Target 1: NASA LRN(1)1007
0 0.3c 12% Base: NACA 0012
0.4c
Thickness
Location
0.047
Max.
Camber
15% Target 3: NASA NLF(1)1015
t/c Airfoil
] ) / ( ) / ( [
. arg min i approx i et t
c x f c x f abs f = A
Intelligent Search Agent Particle Swarm Optimization
Swarm Approach
Models Natural Flocks and Movement
of Swarms;
Quick, Efficient and Simple
Implementation;
Ideal for NonConvex Discontinuous
Problems;
Solution Governed by Position of
Particle within Ndimensional Space;
Each Particle Records Personal
Fitness p
best
;
Best Global Fitness g
best
;
Velocity & Position Updates based on
Global Search Pattern; and
Convergence Particles Unite at
Common Location
J. Kennedy and R. Eberhart, "Particle Swarm Optimization, presented
at IEEE International Conference on Neural Networks, 1995.
Algorithm
1. Initialise Particle Swarm
2. Initialise Particle Velocities
3. Evaluate Fitness of Each Particle
4. Update according to:
i. Velocity Update
ii. Position Update
5. Repeat until Convergence Satisfied
Particle Swarm Optimization Set Up
PSO Structure / Inputs Definition
Velocity Update:
Position Update:
S
P
S
O
o 0.110% of
N
DIM
o c
1
= 2
o c
2
= 2
0.110% of N
DIM
 w Facilitates Global Search
+ w Facilitates Local Search
Determine pull of p
best
& g
best
c
1
Personal Experience
c
2
Swarm Experience
A

P
S
O
o 0.110% of N
DIM
Maximum Velocity
Inertia Weight (w):
o c
1
= 2
o c
2
= 2
Scaling Factors
Cognitive & Social
(c
1
& c
2
)
;
4 2
2
w
2
=
2 1
c c + = where
c +
=
ij ij
ij
best best
best ij
g p
p x
ij
ISA

.

\

+
=
ij
ISA ij
e
w
1
1
1 o
Standard vs. Adaptive PSO
( ) ( ) ( ) ( ) ( ) ( ) k x P rand c k x P rand c k v w k v
i g i i i i
+ + = +
2 1
1
( ) ( ) ( ) 1 1 + + = + k v k x k x
i i i
Particle Swarm Optimizer Search Agents
Particle Swarm Optimizer  Function Test
( )  
=
+
+ =
1
1
2 2
1
2
1 ) ( 100 ) (
n
i
i i i
x x x x f
n i x
i
,..., 2 , 1 , 100 100 = s s
0 ) ( ), 1 ,..., 1 (
* *
= = x f X
10
5
0
5
10
10
5
0
5
10
0
5
10
15
x 10
5
x
Rosenbrock Function
y
z
30 15 s s
i
x
Definition:
Search Domain:
Initialization Range:
Global Minima (Fitness):
Velocity Fitness Fitness
Low Velocity = Low Fitness
Particle Swarm Optimizer  Function Test
Definition:
Search Domain:
Initialization Range:
Global Minima (Fitness):
0 ) ( ), 1 ,..., 1 (
* *
= = x f X
=
=
n
i
i i
x x n x f
1
) sin ( 9829 . 418 ) (
n i x
i
,..., 2 , 1 , 500 500 = s s
500 250 s s
i
x
Velocity Fitness Fitness
Low Velocity = Low Fitness
Shape Parameterization Results
Summary of Results
Measure of Geometrical Difference
HicksHenne Most Favorable
Legendre Polynomials
Computationally Not Viable
Aerodynamic Coefficients
Convergence
1
0
1
2
3
4
5
6
7
8
Shape Function
C
o
s
t
Magnitude of Cost Function
Bernstein
HicksHenne
Legendre
NACA
Wagner
Geometrical Convergence Plots /
Animations
s
HicksHenne Geometrical
Convergence
s Bernstein Geometrical
Convergence
Aerodynamic Convergence Plots /
Animations
s
HicksHenne Aerodynamic
Convergence
s Bernstein Aerodynamic
Convergence
Shape Functions Limitations
Polynomial Function Limitation
Local Shape Information;
No Direct Geometry Relationship;
NURBS Require Many Control Points; and
Lead to Undulating Curves
PARSEC Airfoil Representation
6
th
Order Polynomial;
Eleven Variables
Equations Developed as a Function of
Airfoil Geometry; and
Direct Geometry Relationship
H. Sobieczky, Parametric Airfoil and Wings, in: Notes on Numerical
Fluid Mechanics, Vol. 68, pp. 7188, 1998
1
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
Shape Functions
F
i
t
n
e
s
s
M
a
g
n
i
t
u
d
e
Bernstein
HicksHenne
PARSEC
Legendre
NACA
Wagner
Fitness Magnitude of Shape Functions
2
1
6
1
=
=
=
n
n
n PARSEC
X a Z
PARSEC Airfoils
PARSEC Aerodynamic Convergence
Convergence to Target Lift Curve Slope Convergence to Target Drag Polar
Convergence to Target Moment Convergence to Target L/D
5 0 5 10 15 20
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
o()
C
L
Target
HicksHenne
PARSEC
0 0.02 0.04 0.06 0.08 0.1 0.12
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
C
D
C
L
Target
HicksHenne
PARSEC
5 0 5 10 15 20
0.11
0.1
0.09
0.08
0.07
0.06
0.05
0.04
0.03
o()
C
M
Target
HicksHenne
PARSEC
5 0 5 10 15 20
0
50
100
150
o()
L
/
D
Target
HicksHenne
PARSEC
PARSEC Design Variables Definition
Effect of Y
UP
on PARSEC Airfoil Aerodynamics
Lift Coefficient Drag Coefficient Moment Coefficient LifttoDrag Ratio
Effect of Y
UP
on PARSEC Airfoil Geometry
Y
UP
Nose Radius
t/c Camber
Low Y
UP
= Good C
D
Performance
Shape Function Modifications
Airfoil Surface Bumps
Aerodynamic Performance Improvements;
Rough Airfoils Outperform Smooth Sections at Low R
e
;
Control Flow Separation;
Passive & Active Methods for Bypass Transition;
Reduction in Turbulence Intensity; and
Bumps Delay Separation Point
Shape Functions  Further Developments
Local Curvature Control;
Roughness in Line with Boundary Layer Height; and
Control over NonLinear Flow Features
Airfoil Surface Bumps to Assist Flow Reattachment
Source: A. Santhanakrishnan and J. Jacob, Effect of Regular Surface
Perturbations on Flow Over an Airfoil,  University of Kentucky, AIAA20055145
Ideal Surface
Bumpy Surface
Flow Solver Computational Fluid Dynamics
Laminar Turbulent
6,000 Maximum Iteration Count
1.0 x 10
6
Residual Solution Convergence
0.32 Flow Mach Number
Turbulence Intensity = 0.5%; Viscosity Ratio = 5
Turbulence Intensity = 2%; Viscosity Ratio = 20
Boundary Conditions:
Inlet
Pressure Outlet
Air as an Ideal Gas Flow Medium
6.0 x 10
6
Reynolds Number
ke & SA Turbulence Modeling
Viscous Model
Second Order Upwind Discretization Scheme
1.055 Wall Cell Intervals
96,000 Total Mesh Size (approx.)
Segregated Implicit Formulation of RANS
Energy Equations also Solved
Solver
~1 Wall y
+
Range (approx.)
80 Circumferential Lines
100 Radial Lines
2D Structured CType Mesh
6,000 Maximum Iteration Count
1.0 x 10
6
Residual Solution Convergence
0.32 Flow Mach Number
Turbulence Intensity = 0.5%; Viscosity Ratio = 5
Turbulence Intensity = 2%; Viscosity Ratio = 20
Boundary Conditions:
Inlet
Pressure Outlet
Air as an Ideal Gas Flow Medium
6.0 x 10
6
Reynolds Number
ke & SA Turbulence Modeling
Viscous Model
Second Order Upwind Discretization Scheme
1.055 Wall Cell Intervals
96,000 Total Mesh Size (approx.)
Segregated Implicit Formulation of RANS
Energy Equations also Solved
Solver
~1 Wall y
+
Range (approx.)
80 Circumferential Lines
100 Radial Lines
2D Structured CType Mesh
Flow Solver Validation Case 1: NASA LS(1)0417 Mod
5 0 5 10 15 20
0.5
0
0.5
1
1.5
2
2.5
3
Fixed Boundary Layer Transition: Lift Curve Slope
o()
C
L
Exp
CFD
0.008 0.01 0.012 0.014 0.016 0.018 0.02
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Fixed Boundary Layer Transition: Drag Polar
C
D
C
L
Exp
CFD
Validation Data
C
P
Agreement at AOA 10;
Lift & Drag Convergence over Linear
AOA;
Lift ~ 2% ; Drag ~ 5%;
Solution Divergence at Stall; and
Fluid Separation Zone Effectively
Captures Boundary Layer Transition
0 0.1 0.2 0.3 0.4 0.5 0.6
4
3
2
1
0
1
2
Fixed Transition C
P
Distribution Comparison: Re=6.0e6, Mach=0.32
x/c
C
P
Exp
CFD
Flow Solver Validation Case 2: NACA 0012
Validation Data
C
P
Agreement at AOA 11;
Lift & Drag Convergence over Linear
AOA;
Lift ~ 5% ; Drag ~ 7%;
Solution Divergence at Stall; and
Fluid Separation Zone Effectively
Captures Boundary Layer Transition
0 2 4 6 8 10 12 14
0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
NACA 0012  Fixed Boundary Layer Transition Lift Curve Slope
o()
C
L
Exp.
CFD
0 0.1 0.2 0.3 0.4 0.5 0.6
6
5
4
3
2
1
0
1
2
x/c
C
P
NACA 0012  Fixed Boundary Layer Transition C
P
Distribution: Re = 6.0e6, Mach 0.35
Exp.
CFD
0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0.022
0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
NACA 0012  Fixed Boundary Layer Transition Drag Polar
C
D
C
L
Exp.
CFD
Sample Optimization Run
Objective Function
o = 2
C
L
= 0.40
Minimize C
D
Optimizer Inputs Final Solution
Swarm Size = 20 Particles
r
LE
= [0.001 , 0.04] 0.0368
Y
TE
= [0.02 , 0.02] 0.0127
T
eg
= [2.0 , 25] 19.5
T
EW
= [3.0 , 40.0] 29.10
X
UP
= [0.30 , 0.60] 0.4581
Y
UP
= [0.07 , 0.12] 0.0926
Y
XXU
= [1.0 , 0.2] 0.2791
X
L
= [0.20 , 0.60] 0.5120
Y
L
= [0.12 , 0.07] 0.1083
Y
XXL
= [0.2 , 1.20] 0.6949
Results
t/c = 20%
C
L
= 0.4057
C
D
= 0.0069
Total Iterations = 29
0 5 10 15 20 25 30
0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
Optimization History Plot
Optimization History Plot
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.25
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
0.2
0.25
Airfoil Optimization
x/c
y
/
c
Final Airfoil Shape
Aerodynamic Coefficient Database Artificial Neural Networks
Artificial Neural Networks Airfoil Training Database
Geometrical Inputs;
Aerodynamic Coefficient/s Output/s

;
Setup of Transfer Function within the Hidden Layer; and
Output RMS Evaluation
Coefficient of Lift NN Structure Coefficient of Drag NN Structure Coefficient of Moment NN Structure
R. Greenman and K. Roth Minimizing Computational Data Requirements for MultiElement Airfoils
Using Neural Networks, in: Journal of Aircraft, Vol. 36, No. 5, pp. 777784 SeptemberOctober 1999
Coupling of ANN & Swarm Algorithm
Conclusion
Geometry Parameterisation Method
Six Shape Functions Tested;
Particle Swarm Optimizer Validated / Utilized;
SOMs for Design Variable Definition; and
PARSEC Method for Shape Representation
Flow Solver
RANS Solver with Structured CGrid;
Transition Points Integrated;
Acceptable Solution Agreement; and
Transition Modeling and DES for HighLift
Flows
Airfoil Optimization
Direct PSO Computationally Demanding; and
ANN to Reduce Computational Data
www.cosmosmagazine.com www.mathworks.com
Acknowledgements
Viscovery Software GmbH
[http://www.viscovery.net/]
Mr. Bernhard Kuchinka
Kindly provided a trial copy of Viscovery SOMine