You are on page 1of 43

Aerodynamics Glossary

Wing Design Tip


A most difficult aspect of wing design can be choosing the correct airfoil cross sectional
shape. Although most airfoil shapes can support flight, only the right one will save
thousands of dollars in operational costs over the life of the aircraft. MultiSurface
Aerodynamics is a digital wind tunnel that can compare the performance of many airfoil
shapes to make the airfoil selection process easy.
Aerodynamic Center
The aerodynamic center is a point along the airfoil or wing about which the moment
coefficient does not vary with an angle of attack change.
Airfoil
An airfoil is the cross section of a wing. The airfoil shape and variations in angle of
attack are primarily responsible for the lift and profile drag of the wing.
Angle of Attack
The angle of attack is defined as the angle between the plane of the wing (airfoil chord)
and the direction of motion (free stream velocity). The angle of attack can be varied to
increase or decrease the lift acting on the wing. An increase in lift often results in an
increase in drag.
Center of Pressure
A point along the airfoil about which the moment due to the lift is ero, i.e., it is the point
of action of the lift. The center of pressure will change its position when the angle of
attack changes.
Chord
The chord is the dimension of the airfoil from its leading edge to trailing edge.
Circulation
!irculation is a measure of the vorticity in the flow field. "or an inviscid flow field, the
lift is e#ual to the product of the circulation about the airfoil, the density and the velocity.
Computational Fluid Dynamics (CFD)
!omputational fluid dynamics is the term given to a variety of numerical mathematical
techni#ues applied to solving the e#uations that govern fluid flows and aerodynamics.
Modern CFD results can rival the accuracy of wind tunnels in testing airfoils, wings and
entire airplanes for certain test configurations.
Density
The mass of a substance contained in a given volume divided by the volume. "or a
incompressible fluid, the density is considered to be constant throughout the flow field.
$owever, for a compressible fluid, the density can vary from one location to the ne%t in
the flow field. The speed of sound in a fluid depends on the ratio of pressure changes to
density changes in the fluid.
Drag
&rag is an aerodynamic force opposing the direction of motion. &rag can be due to
surface viscosity (friction drag), pressure differences due to the shape of an ob'ect (form
drag), lift acting on an finite wing (induced drag) and other energy loss mechanisms in
the flow such as wave drag due to shock waves and inefficiencies in engines.
Drag Coefficient
The drag coefficient is defined as the drag((dynamic pressure ) reference area). The
reference area is usually the plan*form or flat pro'ection (the wing+s shadow at noon) area
of the wing.
Dynamic Pressure
The dynamic pressure is defied as the product of the density and the s#uare of the
velocity divided by two. The dynamic pressure has units of pressure, i.e. "orce(Area. The
dynamic pressure is used to non*dimensionalie forces and pressures in aerodynamics.
Flap Deflection Angle
The flap deflection angle is the angle between the deflected flap and the chord line. The
angle is positive for a downwards deflection of the flap. &eflect the flap downwards to
increase the airfoil+s lift.
Lift
The lift is a force acting perpendicular to the direction of flight. The lift is e#ual to the
fluid density multiplied by the circulation about the airfoil and the free stream velocity. ,n
level flight, the lift developed by an airplane+s must be e#ual to the weight of the entire
airplane.
Lift Coefficient
The lift coefficient is defined as the lift((dynamic pressure ) reference area). The
reference area is usually the plan*form area of a wing or horiontal pro'ection of the
wing.
ean aerodynamic chord
This chord is located along the wing and has the aerodynamic property of the two*
dimensional wing.
!ACA Airfoils
-A!A airfoils are wing cross section designs invented by the !ACA organiation.
-A!A eventually became -ASA (-ational Aeronautics and Space Administration). $ere
are a few popular airplanes that have -A!A airfoil wings.
Airplane "oot Airfoil Tip Airfoil
/eech 01 Twin /onana -A!A 23145.4 -A!A 23142
/*46 "lying "ortress -A!A 1142 -A!A 1141
!essna 402 -A!A 2542 -A!A 1142
!essna 462 4763*later -A!A 2542 -A!A 2542 mod
!essna 001 !itation ,, -A!A 23145 -A!A 23142
&ouglas &!*3 -A!A 2240 -A!A 2218
"airchild A#$% Thunderbolt ,, -A!A 8648 -A!A 8643
Sikorsky S*84 S$*3 Sea 9ing -A!A 1142 -A!A 1142
Panel ethod
This numerical method places singularities along the airfoil. ,n the case of :isual"oil or
3DFoil , the singularities are vortices. The vorticity is distributed linearly along the panel.
Plain Flap
A plain flap is a hinge attachment near the trailing edge of an airfoil. The length of the
flap is measured as a percentage of the chord and the deflection is measured in degrees.
Pressure Coefficient
The pressure coefficient is a non*dimensional form of the pressure. ,t is defined as the
difference of the free stream and local static pressures all divided by the dynamic
pressure.
"eynolds !um&er
The ;eynolds number is a non*dimensional parameter that compares the inertia to
viscous forces. ,f the ;eynolds number is low, then viscosity plays an importatant part in
the simulations.
'tall
At low angles of attack, the lift developed by an airfoil or wing will increase with an
increase in angle of attack. $owever, there is a ma%imum angle of attack after which the
lift will decrease instead of increase with increasing angle of attack. This is know as stall.
9nowing the stall angle of attack is e%tremely important for predicting the minimum
landing and takeoff speeds of an airplane.
'treamlines
!ontours in the flow field that are tangent to the velocity vector.
Wing Loading
The total weight of the airplane divided by the plan form area of the wing.
Wing 'pan
The span is the total length of the wing.
A&out Dr( )anley
&r. <atrick =. $anley, is the owner and founder of $anley ,nnovations, a small business
specialiing in the development of aerodynamics and fluid dynamics simulation software
for education and industry. &r. $anley earned his /.S. degree (summa cum laude) in
aerospace engineering from <olytechnic ,nstitute of -ew >ork and his S.M. and <h.&.
degrees from the department of Aeronautics and Astronautics of Massachusetts ,nstitute
of Technology (M,T). $e also completed a minor in the area of management of
innovation and technology at M,T+s Sloan School of Management.
After graduating from M,T, &r. $anley 'oined the Mechanical =ngineering faculty at the
?niversity of !onnecticut where he formulated and taught courses in aerodynamics,
compressible fluids, introductory fluid mechanics and heat transfer. As a faculty member,
he won the highly competitive -ational Science "oundation ;esearch ,nitiation Award,
the -ASA*AS== Summer "aculty "ellowship and three consecutive research awards
from -ASA @ewis ;esearch center to study compressible viscous flows in
turbomachinery using pseudospectral methods. This research led to the successful
education of four (5) <h.& students and four (5) Masters degree students. ,n addition &r.
$anley can be credited with a number of publications including the pioneering work in
multi*domain pseudospectral methods for compressible viscous flows entitled AA
Strategy for the =fficient Simulation of :iscous !ompressible "lows using a Multi*
domain <seudospectral MethodA which can be found in Bournal of !omputational
<hysics, :ol 41C, -o. 4, pp. 403*40C, September 4773.
ulti'urface Aerodynamics
TM
is the leading !A= (computer aided
engineering) product for multiple wing design D analysis on the <!.
MultiSurface turns your <! into a powerful wind tunnel D diagnostic
tool for creating D testing airfoils D 3& wing systems.The user*
interface is self*contained and has tools for design, analysis,
presentations and prototyping. MultiSurface computes the lift, total
drag, moments and stability derivatives of your custom wing system
in 'ust seconds on a Eindows <!. ?se different airfoils and wing
shapes to optimie your design on the fly. -o advanced aerodynamics
or computer programming skills are re#uired to use the software.
Design * Test Wings +n ,our
PC-
MultiSurface Aerodynamics features include.
stand*alone application (built*in pre D post processing tools)
easily enter and manipulate comple% wing systemF instantly
compute lift, drag (profile and vorte%), moments and stability
derivativesF calculate trim, neutral point and ground effect.
ability to read*in, modify and analye custom airfoil shapes to test
the effectiveness of original designs.
users do not need to have advanced aerodynamics or computer
programming skills
accurate solvers for realistic simulations
computer aided design interface to model and edit 3& wing
systems ('oined wing for e%ample)
interactive visualiations
graphs D tables for comparative analysis
ability to e%port 2& D 3& !A& files for rapid prototyping.
Lease./uy +ptions
/uy !o0
ore 12amples
Sailplane Analysis
Boined Eings
=ndplates (.pdf)
Sprint !ar Eing (.pdf)
MultiSurface :alidation
$ydrofoil Analysis
=lliptical Eing
Gther Eing Shapes
Applications
MultiSurface Aerodynamics is the ultimate desktop wind tunnel software. ?se it to design and test comple%
wing systems directly on your Eindows <!. ,t was developed by <atrick $anley, <h.&. for accuracy,
efficiency and ease of use and is by far the 3uickest method on the market for developing wings for
aircraft, water craft and automobiles. MultiSurface is used in the following areas.
@ight aircraft wing design D analysis. This includes kites, powered parachutes, parafoils and other
sporting e#uipment.
?A: design D analysis
Multi*plane wing configurations including 'oined wings
Sailboat(yacht keel and rudder design D analysis
?p*wind sail design D analysis
Ship(sailboat hydrofoil design D analysis
$uman powered hydrofoil design D analysis
Eater sport e#uipment design. This includes sailboard fins and sail configurations.
=#uipment for marine industry such as foils for stabiliing vessels D trawl doors.
Automobile wing, spoilers and surfaces designed for safety and fuel efficiency.
@ectures and presentations in the area of aerodynamics design and analysis.
+nline Tutorials * Demos
Tutorial # $ow to enter and analye a wing. :iew tutorial now.
&emo * $ow to improve a sailboat keel. :iew demo now
&emo * $ow to compare the performance of two keels. :iew demo now.
&emo * ?sing the mirror imaging feature for wing design. :iew demo now.
12amples
&esign and analye 4oined 0ings for ?A: applications. More information ...
?nmanned aerial vehicles design and analysis. More information ...
Sailboat keel D rudder analysis and design. More information ...
Analye hydrofoil with front and rear foils. More information ...
Analye hydrofoils in shallow water. More information ...
Analye an elliptical wing. More information ...
Analye a full*sied sailplane. More information ...
&esign D analye full*sied race car wings and spoilers. More information ...
Analye a full*sied sprint car wing with endplates (pdf file). More information ...
1ducational 5se
,ntroduce engaging e%periments, design content D capstone pro'ects into your lectures D labs. More
information ...
Who 5ses ulti'urface Aerodynamics
MultiSurface Aerodynamics is best suited for users who possess a good understanding of aerodynamics,
hydrodynamics and related engineering principles. Gur customers leverage their design D engineering
knowledge with MultiSurface Aerodynamics to develop outstanding products and services. The following
companies, consultants and universities are among the users of MultiSurface Aerodynamics.
AeroScience
!&= &anish Marine &esign ApS
!lyde /ooth (race car wing engineer)
$ugh Eelbourn (/ritish yacht designer)
,sland =ngineering, ,nc
9imokeo =ngineering
9orea Gcean ;esearch D &evelopment ,nstitute
Mike Mageria (race car wing designer)
Multi$ull <ower
<aradis -autica AS
Sea State
?nited States Air "orce
?niversitat <olitHcnica de !atalunya
?niversity of !alifornia, San &iego
?niversity of =dinburgh
Computer "e3uirements
The perpetual licence of MultiSurface Aerodynamics re#uires a <! running Eindows 70 or @ater. The
leased (online) versions re#uires a <! running Eindows 70 or @ater and a connection to the internet. The
leased software accesses the internet for login purposes only
+6er6ie0
:isual"oil <lus is used by engineers and designers to compute the lift, drag and moment coefficients for
airfoils in subsonic, transonic and supersonic flows. ,t generates accurate aerodynamics data for
e%perimental or e%isting airfoils where the information might be incomplete or unknown. The software can
analye and select airfoils for aircraft, water craft, industrial machinery and other products. :isual"oil<lus
produces graphs of lift, drag and moment coefficient versus angle of attackF surface pressure coefficient,
velocity, Mach number and temperature. The software produces contour plots of pressure, Mach number,
temperature and total pressure. :isual"oil <lus has a large library of built*in airfoils which includes -A!A
5, 0 D 8*digit airfoils. ,n addition, the user can enter custom airfoils for analysis and modify e%isting
airfoil shapes.
Please click here to purchase 7isualFoil Plus.
/enefits
=liminate steep learning curve for new engineers D reduce training costs.
Share airfoil data D results across your entire department
?nderstand and make critical design decision in transonic D supersonic flow regime.
?nderstand the behavior of airfoils used in the design of wings, helicopter blades, hydrofoils, fans,
propellers and other e#uipment.
=asy*to*use tool for designing more efficient wing D machine blade cross*sections.
Eindows based interfaces re#uires little or no training.
8ncompressi&le Flo0 'ol6er
@inear strength vorte% panel method
!oupled boundary layer solver
@ift, drag D moment coefficient calculations
<rediction of transition D separation points
Stall D ma%imum lift prediction
Surface graphs of pressure D velocity ratio
!ontour graphs of !p
Iraphs of !l, !d D !m versus angle of attack
Iraphs of !l vs !d.
Compressi&le Flo0 'ol6er
2*dimensional compressible =uler e#uations using a finite*volume solver
2nd order flu% vector splitting
2nd order flu%*difference splitting
5*stage ;unge*9utta time marching scheme using local time steps
Automatic o*grid generation
!ontour plots of pressure, Mach number, temperature D total pressure
Surface plots of !p, Mach number, velocity ratio D temperature
+ther Features of 7isualFoil Plus
,ncompressible flow solver using vorte% panel method coupled with boundary layer solver
!ompressible flow solver based on =uler =#uations
/uilt*in library of -A!A 5, 0 D 8*digit airfoils
!ustom airfoil analysis
!omputes !l, !d D !m
Stall Model
/uilt*in graphs
/uilt*in tables
=%port tables to .csv files
=%port airfoil to .d%f files
Airfoil plotting
Articles
$igh Speed /lade Screening (.pdf file)
:isual"oil <lus for <ropeller Analysis
Computer 'ystem "e3uirements
:isual"oil <lus is a stand*alone software package that re#uires Eindows 70 or later. Ee recommend a
pentium based <! running Eindows J<. The leased version of :isual"oi <lus also re#uires an internet
connection.
Purchase
)o0 to Purchase 7isualFoil Plus
!ote9 :isual"oil <lus is available e%clusively on hanleyinnovations.com.
7ersion Price Purchase
Lease for Three onths K870.11 /uy !& -ow
Lease for +ne ,ear K4,870.11 /uy !& -ow
Perpetual ('ingle 'eat) :;<==>(%% /uy CD !o0
Perpetual (? 5sers) :@<==>(%% /uy CD !o0
Telephone +rders. <lease call us at (?>;) @AB#CC@@ to place your credit card order by telephone. Ee can
provide you with a download ?;@ D password after the transaction is processed.
Fa2 ,our +rders. <lease click here to fa% your order. (.pdf order form)
+rder &y ail. !lick here to order by mail using a check, money order or credit card.
** The leased versions of VisualFoil Plus are identical to the full perpetual version. nli!e the perpetual
version, however, they re"uire an internet connection to enter a username # password..
!ote9 $ll prices su%&ect to change without notice.
Other Payment Methods
Ee also accept payment by /ank Transfer (SE,"T for e%ample). <lease email us for details.
"e3uest ore 8nformation9
"or more information, please email us at salesLhanleyinnovations.com. Ee can also be reached by
telephone at (302) 8C6*5588. Gur mailing address is $anley ,nnovations, Attn. :isual"oil <lus, <G /o%
C34045, Gcala, "l 355C3*4345.
'ee Also9
Multi=lement Airfoils, MultiSurface Aerodynamics
A&out Dr( )anley
&r. <atrick =. $anley, is the owner and founder of $anley ,nnovations, a small business specialiing in the
development of aerodynamics and fluid dynamics simulation software for education and industry. &r.
$anley earned his /.S. degree (summa cum laude) in aerospace engineering from <olytechnic ,nstitute of
-ew >ork and his S.M. and <h.&. degrees from the department of Aeronautics and Astronautics of
Massachusetts ,nstitute of Technology (M,T). $e also completed a minor in the area of management of
innovation and technology at M,T+s Sloan School of Management.
After graduating from M,T, &r. $anley 'oined the Mechanical =ngineering faculty at the ?niversity of
!onnecticut where he formulated and taught courses in aerodynamics, compressible fluids, introductory
fluid mechanics and heat transfer. As a faculty member, he won the highly competitive -ational Science
"oundation ;esearch ,nitiation Award, the -ASA*AS== Summer "aculty "ellowship and three
consecutive research awards from -ASA @ewis ;esearch center to study compressible viscous flows in
turbomachinery using pseudospectral methods. This research led to the successful education of four (5)
<h.& students and four (5) Masters degree students. ,n addition &r. $anley can be credited with a number
of publications including the pioneering work in multi*domain pseudospectral methods for compressible
viscous flows entitled AA Strategy for the =fficient Simulation of :iscous !ompressible "lows using a
Multi*domain <seudospectral MethodA which can be found in Bournal of !omputational <hysics, :ol 41C,
-o. 4, pp. 403*40C, September 4773.
As owner and chief software author of $anley ,nnovations, &r. $anley has written a number of software
packages including Airfoil/rowser, Airfoil Grganier, Science Iraphs, :isual"oil, Model"oil,
Aerodynamics in <lain =nglish, !enter of Iravity !alculator, EingAnalysis, SmockSoft, <erpetural<aper
amongst other
This is the html version of the file http://www.geodise.org/files/Papers/Man20441.pdf.
G o o g l e automatically generates html versions of documents as we crawl the we.
To lin! to or oo!mar! this page" use the following url: http://www.google.com/search?q=cache:IAhUq6ly_CIJ:www.geodise.org/files/Papers/a!"#$$%.pdf&'o(r!als&
)performa!ce&of&differe!t&airfoils&i!&cfd*+hl=e!+gl=i!+ct=cl!,+cd=%6
'oogle is neither affiliated with the authors of this page nor responsi%le for its content.
These search terms have een highlighted: performance different airfoils cfd
These terms only appear in lin!s pointing to this page: journals
Page 1
A Study of Shape Parameterisation Methods for
Airfoil Optimisation
Wenbin Song

and Andrew J. Keane

University of Southampton, Highfield, Southampton, SO17 1BJ, United Kingdom


This paper presents a study on parameterisation methods for airfoil shape optimisation
within a CAD-based design optimisation framework. The objective of the paper is to study
the effect of different methods on airfoil shape optimisation when using computational fluid
dynamics (CFD). Parameterisation of geometry is one of the essential requirements in shape
optimisation, and it presents further challenges when carrying out multidisciplinary design
optimisation, as it is critically important to maintain shape consistency between analysis
domains, while providing different analysis models from the same CAD definition. It is
usually the case that there are numerous possibilities in defining the parametric model,
and it will prescribe to a large extent the scope of the search space and landscape of the
objective function. This paper adopts design of experiments and optimisation approaches to
study several representative parameterisation methods in terms of flexibility and accuracy
of the methods for aerodynamic shape optimisation.
I. Introduction
A
ircraft design is a complex decision making process and according to Raymer,
1
can usually be broken
down into three phases, conceptual design, preliminary design, and detailed design. Aerodynamic design
occurs throughout these steps. Two different approaches are often employed in the aerodynamic design: 1)
inverse design and 2) direct numerical optimisation. The first method tries to solve for a geometry that
produces a prescribed pressure distribution. On the other hand, direct numerical optimisation methods
couple a geometry definition and aerodynamic analysis code in an iterative process to produce optimum
designs subject to various constraints. Depending on whether the goal is to improve on an existing design or
to create a completely new design, different parameterisation methods are often required. If the new design
only requires small changes to the initial geometry, a localized parameterisation approach is often used. But
when conducting a study of a radically new concept, the parameterisation method needs to accommodate a
wider range of new shapes.
Airfoils have been represented in a number of different ways in the past. For example, coordinates have
been directly used to fit airfoil shapes using B-splines and Bzier curves
2
via interpolation methods. Analytical
functions have also been derived to represent families of airfoils, for example, in the work reported by Hicks
and Henne.
3
In a more recent work,
4
Non-uniform rational B-splines (NURBS) were used first to approximate
existing airfoils, then adopted as a general parameterisation method to be used in optimisation. The concept
of using relatively few orthogonal functions to represent a large number of functions has also been exploited,
for example in a work reported by Robinson and Keane,
5
where a set of orthogonal functions was developed
using numerical methods. These functions were then used to represent a family of airfoils in a wing design
study. However, the basis functions derived by Robinson and Keane
5
were believed to be dependent on the
particular familiar of airfoils. Although these numerically derived basis functions can be used in the design
of a particular set of airfoils, other airfoils may not be adequately represented using them.
The choice of parameterisation method, when coupled with optimisation techniques to find desirable
shapes in terms of user-defined objective functions and constraints, has a major effect on the final results,
efficiency and effectiveness of particular search strategies. Giving the same CFD models, the parameterisation

Research Fellow, School of Engineering Sciences, University of Southampton, Highfield, SO17 1BJ, United Kingdom, and
AIAA Member.

Professor, School of Engineering Sciences, University of Southampton, Highfield, SO17 1BJ, United Kingdom.
1 of 8
American Institute of Aeronautics and Astronautics
Page 2
effectively defines the optimisation problem formulation, the topology of the design space, and the landscape
of the objective functions. Although it is vitally important, it is also very difficult to come up with a set of
effective criteria that can be readily used to evaluate the pros and cons of different parameterisation methods.
Wu
6
compared three geometric representations in three case studies of cascade blade design using adjoint
methods. After carrying out the optimisation, it has concluded that one of the methods using geometry
parameters is not suitable for two occasions.
There are a number of key issues that need to be addressed in the choise of parameterisation methods.
The first issue is the flexibility of any parameterisation method. Flexibility is interpreted here as the ability
to represent a wide range of different shapes. Some parameterisation methods, for example, coordinate-
based methods, can accurately represent a variety of dramatically different shapes and can also reflect subtle
changes in local areas, however it would be very difficult to use such an approach for optimisation problems
using high fidelity codes due to the large number of design variables and complexity of the design space. On
the other hand, methods using fewer variables may not be capable of generating shapes with high accuracy,
especially when used in inverse design problems where a target pressure distribution is sought. The second
issue when considering parameterisation methods is the accuracy or the optimum objective functions that
the final shape can achieve either in an inverse design study or direct optimisation work, respectively. The
accuracy should be measured in both geometric and aerodynamic senses. However, the optimal objective
function cannot be obtained without actually carrying out the optimisation, therefore, here an inverse design
approach is adopted to compare different parameterisations.
In this work, three datum airfoils, two from the NACA supercritical airfoils family (NACA0406 and
NACA0610) and the third being the RAE2822, are used as reference shapes to compare different param-
eterisation methods for airfoil design. The paper is organised as follows. Section two describes different
parameterisation methods for airfoil design. The geometry modelling and flow analysis of the airfoil prob-
lems are described in section three. Results and discussions are presented in section four, with conclusions
given in section five.
II. Parameterisation of Airfoil Geometry
Figure 1. The first three numerically derived or-
thogonal basis functions.
Geometry parameterisation methods have attracted
renewed interests in recent years, especially in the con-
text of multidisciplinary design optimisation (MDO).
Samareh
7
identified three categories of parameterisation
methods in the context of MDO. These include the dis-
crete approach, CAD-based approaches, and free-form de-
formation methods. Indeed, all these different approaches
could be implemented in most modern CAD systems.
Several parameterisation methods have been proposed
in previous papers for airfoil geometry, for example, the
NACA supercritical airfoils are defined as a series of y-
coordinates at prescribed chord wise locations.
8
The sec-
ond approach models the geometry as the linear combina-
tion of a basis airfoil and a set of perturbation functions,
defined either analytically
4
or numerically,
5
as shown in
Eq. (1). The coefficients of the perturbation functions in-
volved are then considered as the design variables. A set
of such orthogonal basis functions derived from a group
of base airfoils was developed by Robinson and Keane
5
to
provide an efficient means to define the airfoil for optimi-
sation study in preliminary design, for example.
y(x) =
0
y
0
(x) +
w
i
f
i
(x)
(1)
A third, and more geometrically intuitive method, is to use geometric parameters such as leading edge
radius, thickness-to-chord ratio or maximum thickness to define the airfoil shape. An airfoil parameterisation
using 11 geometry parameters was presented by Sobieczky
9
and used by Oyama etc.
10
A fourth method uses
the control points of Non-uniform Rational B-splines (NURBS) curves to define the airfoils.
4
This method is
2 of 8
American Institute of Aeronautics and Astronautics
Page 3
also used by Li
11, 12
in which a B-pline interpolation through 35 points is used to define the airfoil geometry.
The advantage of this approach is that free-form geometrical shapes can be accommodated with fewer design
variables compared to the direct use of coordinates. However, due to difficulties in controlling the relative
positions of the control points, free-form parameterisations are usually used in the inverse design approach,
where only a subset of control points are allowed to change in a relatively small range to meet a target
pressure distribution.
One of the key issues in deciding the parameterisation method is the balance the requirements of ro-
bustness and flexibility, and these decisions are also strongly dependent on the goal of the design activity.
Although free-form parameterisations may well be able to generate radical new shapes, this is not suitable
for designs where the aim is to meet a specific pressure distribution, due to the poor efficiency caused by the
large search space that arises in the optimisation process. Another disadvantage of free form parameterisa-
tion is the inherent difficulties encountered when trying to generate airfoil-like shapes: Usually additional
geometrical constraints need to be imposed. Two different parameterisation methods are implemented in
the current work to compare their effectiveness. The first approach is to use a set of numerically derived
basis function to define the airfoil,
5
in this case, only a small number of design variables are involved. The
basis functions used are illustrated in Figure 1. The second approach uses a B-spline interpolation based on
34 points as shown in Figure 2. The y coordinates of the points are used as design variables while the chord
wise coordinates of the points are fixed. Both methods are implemented in the CAD system ProEngineer.
13
Airfoil shapes from the family of supercritical airfoils
8
and RAE2822 are chosen in the current work as ref-
erence airfoils in the comparison. The number of parameters involved in the two parameterisation methods
are summarized in Table 1.
Table 1. Summary of different parameter methods for airfoils
Method
Number of parameters Description of the parameters
Numerical Basis Functions
5
Weights for the basis airfoil functions
B-spline interpolation
34
Point coordinates
III. Geometry Modelling and Flow Analysis
X/C
Y
/
C
0
0.25
0.5
0.75
1
-0.03
-0.025
-0.02
-0.015
-0.01
-0.005
0
0.005
0.01
0.015
0.02
0.025
0.03
Figure 2. Airfoil Modelling using B-spline inter-
polation.
To compare the flexibilities of different parameterisa-
tion methods in representing different airfoil shapes, three
existing airfoils NACA0406, NACA0610, and RAE4822
are chosen as the modelling targets for these parameter-
isations. The difference between the target airfoil and
approximated airfoil is defined as
diff =
abs(f
target
(x
i
) f(x
i
))
(2)
where x
i
(i = 1, ..., n) are the chordwise coordinates used
in the definition of the target airfoils. This quantity is
minimised using a global optimisation algorithm for dif-
ferent parameterisation methods and then used as an in-
dication of the flexibility of the methods. To remove the
effect of the optimisation techniques that may not pro-
duce the optimum results, a large number of iterations
has been carried out for the minimisation problem.
The airfoil models are all implemented using ProEngi-
neer and exported in the form of a STEP file, which is
then imported into Gambit. A boundary layer is attached
to the lower and upper surface of the airfoils, and size
functions are also used to give better control of the mesh and to reduce the computational time for the
problem. An unstructured mesh generated using Gambit
14
for solving the N-S equations is shown in Figure
3. The mesh contains 11356 cells (compared with 87305 cells without using the size function). In both cases,
3 of 8
American Institute of Aeronautics and Astronautics
Page 4
the node spacing on the airfoil surfaces and farfield circle are the same, with size functions giving better
control of the transition of size of the cells in between. The computation time is reduced from around 40
minutes to less than 20 minutes for most geometries on a Xeon 2.4Ghz compute node with 1Gb memory.
The flow model used in the current work is based on the Navier-Stoke model from Fluent.
14
The
pressure distribution of the upper and lower surfaces are used in the comparison. Here, the cruise condition
(M

= 0.73) is used when calculating the lift and drag values. The Spalart-Allmaras viscosity model is
used.
IV. Results and Analysis
Figure 3. Unstructured mesh used for solving the N-S
equations by Fluent
It is not straightforward to compare alternative
parameterisation methods. There are two impor-
tant considerations when a parameterisation model
is built around an existing geometry: the first is
the flexibility of the model, i.e., how many different
shapes can this model represent. The second is be
the robustness of the model, i.e., can the model gen-
erates the desired shape for large number of different
designs. In general models with more design vari-
ables will be able to represent more complex shapes,
and will be more likely to produce novel designs us-
ing optimisation. However, that will be more expen-
sive in the search as the design space will have higher
dimensions, and chances of failure or not generating
desired shapes will be higher.
The best results for approximations of the tar-
get airfoils are shown in Table 2. The results are
produced by minimising the objective function computed using (2). A genetic algorithm (GA) from OP-
TIONS
15
is used in the current work, however, the first population is not generated randomly, rather, it
is generated using a Design of Experiment (DoE) method plus one point describing a user specified base
design to obtain more uniform coverage of the design space as well as to provide the best possible guess.
The DoE method used here is a Latin Hyper Cube method. The base design specified by the user can play
an important role in accelerating the search process as the GA used in this work always maintains the best
solution in the population. A single base design is used in the orthogonal basis function approach for all
three target airfoils and in the B-spline approach for RAE2822 approximation. The two base designs used
in B-spline interpolation for NACA0406 and NACA0610 are NACA0403 and NACA0606, respectively.
It can be seen that the approximations using orthogonal basis function can produce better results for
the NACA supercritical airfoils NACA0406 and NACA0610 than for the RAE2822, this is not surprising,
as the set of basis functions used were derived from a family of airfoils containing these two. The errors are
believed to be caused by the smoothing process in the derivation of these basis functions.
5
For NACA0406
and NACA0610, the orthogonal basis functions also produce better results than the B-spline interpolation
approach, this is because more variables are used in the B-spline interpolation and so it would be much more
expensive to obtain the optimum if a comparable number of iterations were used for both cases. However, a
different story arises for RAE2822. Since it was not included in the process of deriving the basis functions,
this leads to greater error when compared with the B-spline interpolation approach. This indicates the wider
applicability of the B-spline approach, at the higher cost of reaching the optimum.
Table 2. Best Approximation of target airfoils for different parameterisations
Method
RAE2822 NACA0406 NACA0610
Numerical Basis Functions
0.2217
0.0582
0.1595
B-spline interpolation
0.1552
0.0758
0.1993
However, similarity in geometry does not always guarantee similar pressure distribution, especially for
4 of 8
American Institute of Aeronautics and Astronautics
Page 5
transonic flows where small perturbations in shape will lead to large variations in pressure. Therefore, the
pressure distributions of the approximated shape are also compared to that of the original airfoils, as shown
in Figure 4. It can be seen from Figure 4 that the orthogonal basis function approach always produces
smooth pressure distributions; also errors are generally bigger in the leading and trailing edge areas than in
the middle section of the airfoils.
Figure 5 shows the results of approximation using the B-spline interpolation approach. It can be seen that
close agreement can be achieved using B-spline interpolation through 34 points apart from the leading edge
area, which indicates that more points need to be placed within this area to achieve better results. Moreover
the chordwise coordinates can also be varied, but this would involve higher computational cost while not
necessarily increasing the accuracy of the approximation. Another advantage of the B-spline approach is its
ability to carry out local shape tunning by varying a subset of the coordinates, while it would be difficult
to perform this with the basis function approach, in which, any changes in the coefficients will change the
shape globally.
The approach adopted in the current work is essentially an inverse design method. However, it is not
used to seek a prescribed pressure distribution, as the definition of the pressure distribution itself is a design
problem and accurately re-generating the prescribed pressure distribution often leads to degradation of
performance in other conditions. This method can be used to evaluate different parameterisations before
carrying out optimisations using the high fidelity codes.
V. Conclusions
Two airfoil parameterisation approaches are studied in this paper to analyse their flexibility and robust-
ness in producing optimal shapes when used in optimisation studies. Global optimisation methods are used
to analysis the accuracy these two parameterisations can achieve when used to model three target airfoils.
The B-spline approach produces better results in terms of accuracy at a higher computational cost while
the basis function approach is more efficient while producing less accurate results. Further work will in-
volve the combination of the basis function approach in the initial stages of design combined with B-spline
interpolation for the final tuning of the shapes.
Acknowledgement
The work described here is supported by the UK e-Science Pilot project: Grid-Enabled Optimisation and
Design Search for Engineering (Geodise) (UK EPSRC GR/R67705/01). Financial support from EPSRC is
greatly acknowledged.
References
1
Raymer, D.P., Aircraft Design: A Conceptual Approach, AIAA Educational Series 1999.
2
Cosentino, G. B. and Holst, T. L., Numerical Optimisation Design of Advanced Transonic Wing Configurations, May,
1984.
3
Hicks, R. M. and Henne, P. A., Wing Design by Numerical Optimisation, Journal of Aircraft, Vol. 15, No. 7, 1978, pp.
407-413.
4
Lepine, J., Guibault, F., Trepanier, J.-Y., and Pepin, F., Optimized Nonuniform Rational B-spline Geometrical Repre-
sentation for Aerodynamic Design of Wings, AIAA Journal, Vol. 39, No. 11, 2001.
5
Robinson, G. M. and Keane, A. J., Concise Orthogonal Representation of Supercritical Airfoils, Journal of Aircraft, Vol.
38, No. 3, 2001, pp. 580-583.
6
Wu, H.Y., Yang, S.C., Liu, F. and Tsai, H.M., Comparison of Three Geometric Representations of Airfoils for Aero-
dynamic Optimization, AIAA 2003-4095, 16th AIAA Computational Fluid Dynamics Conference, June 23-26, Orlando, FL,
2003
7
Samareh, J. A., Survey of shape parameterization techniques for high-fidelity multidisciplinary shape optimisation, AIAA
Journal, Vol. 39, No. 5, 2001, pp. 877-884.
8
Charles D.H., NASA Supercritical Airfoils, A Matrix of Family-Related Airfoils, NASA Technical Paper 2969, Mar. 1990.
9
Sobieczky, H., Parametric Airfoils and Wings, Notes on Numerical Fluid Mechanics, edited by K. Fujii and G. S. Du-
likravich, Vol. 68, Vieweg Verlag, 1998, pp. 71-88.
10
Oyama, A., Obayashi, S., and Nakamura, T., Real-coded Adaptive Range Genetic Algorithm Applied to Transonic Wing
Optimisation, Springer, Paris, France, 2000, pp. 712-721.
11
Li, W., Huyse, L., and Padula, S., Robust airfoil optimisation to achieve drag reduction over a range of Mach numbers,
Structural and Multidisciplinary Optimisaiton, Vol. 24, 2002, pp. 38-50.
5 of 8
American Institute of Aeronautics and Astronautics
Page 6
X/C
Y
/
C
0
0.25
0.5
0.75
1
-0.06
-0.05
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
0.04
0.05
0.06
RAE2822
Orthfoil-Approximation
(a) Approximation of geometry to RAE2822 by basis functions
approach
X/C
P
r
e
s
s
u
r
e
C
o
e
f
f
i
c
i
e
n
t
s
0
0.25
0.5
0.75
1
-1
-0.5
0
0.5
1
RAE2822
Orthfoil-approximation
(b) Comparison of pressure distributions for RAE2822 using
basis functions
X/C
Y
/
C
0
0.25
0.5
0.75
1
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
NACA0406
Orthfoil-Approximation
(c) Approximation of geometry to NACA0406 by basis func-
tions approach
X/C
P
r
e
s
s
u
r
e
C
o
e
f
f
i
c
i
e
n
t
s
0
0.25
0.5
0.75
1
-1
-0.5
0
0.5
1
NAC0406
Orthfoil-approximation
(d) Comparison of pressure distributions for NACA0406 using
basis functions
X/C
Y
/
C
0
0.25
0.5
0.75
1
-0.05
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
0.04
0.05
NACA0610
Orthfoil-Approximation
(e) Approximation of geometry to NACA0610 by basis func-
tions approach
X/C
P
r
e
s
s
u
r
e
C
o
e
f
f
i
c
i
e
n
t
s
0
0.25
0.5
0.75
1
-1
-0.5
0
0.5
1
NACA0610
Orthfoil-approximation
(f) Comparison of pressure distributions for NACA0610 using
basis functions
Figure 4.
Comparisons of geometry and pressure distributions for approximations using orthogonal basis
functions
6 of 8
American Institute of Aeronautics and Astronautics
Page 7
X/C
Y
/
C
0
0.25
0.5
0.75
1
-0.06
-0.05
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
0.04
0.05
0.06
RAE2822
B-Spline interpolation
(a) Approximation of geometry to RAE2822 by B-spline in-
terpolation
X/C
P
r
e
s
s
u
r
e
C
o
e
f
f
i
c
i
e
n
t
s
0
0.25
0.5
0.75
1
-1
-0.5
0
0.5
1
RAE2822
B-Spline Interpolation
(b) Comparison of pressure distributions for RAE2822 using
B-spline interpolation
X/C
Y
/
C
0
0.25
0.5
0.75
1
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
NACA0406
B-Spline Interpolation
(c) Approximation of geometry to NACA0406 by B-spline in-
terpolation
X/C
P
r
e
s
s
u
r
e
C
o
e
f
f
i
c
i
e
n
t
s
0
0.25
0.5
0.75
1
-1
-0.5
0
0.5
1
NAC0406
B-Spline Interpolation
(d) Comparison of pressure distributions for NACA0406 using
B-spline interpolation
X/C
Y
/
C
0
0.25
0.5
0.75
1
-0.05
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
0.04
0.05
NACA0610
B-Spline Interpolation
(e) Approximation of geometry to NACA0610 by B-spline in-
terpolation
X/C
P
r
e
s
s
u
r
e
C
o
e
f
f
i
c
i
e
n
t
s
0
0.25
0.5
0.75
1
-1
-0.5
0
0.5
1
NACA0610
B-Spline Interpolation
(f) Comparison of pressure distributions for NACA0610 using
B-spline interpolation
Figure 5. Comparisons of geometry and pressure distributions for approximations using B-spline interpolations
7 of 8
American Institute of Aeronautics and Astronautics
Page 8
12
Li, W., Profile Optimisation Method for Robust Airfoil Optimisation in Viscous Flow, NASA /TM-2003-212408, 2003.
13
ProEngineer, http://www.ptc.com, 2004.
14
Fluent, http://www.fluent.com, 2004.
15
Keane, A.J., OPTIONS Design Exploration System, http://www.soton.ac.uk/ajk, 2004
8 of 8
American Institute of Aeronautics and Astronautics
This is the html version of the file http://aero#comla.stanford.edu/Papers/!yoto$4.pdf.
G o o g l e automatically generates html versions of documents as we crawl the we.
To lin! to or oo!mar! this page" use the following url: http://www.google.com/search?q=cache:w!-alq./CowJ:aero-
comla/.sta!ford.ed(/Papers/,yoto0$.pdf&'o(r!als&)performa!ce&of&differe!t&airfoils&i!&cfd*+hl=e!+gl=i!+ct=cl!,+cd="#
'oogle is neither affiliated with the authors of this page nor responsi%le for its content.
These search terms have een highlighted: performance different airfoils cfd
These terms only appear in lin!s pointing to this page: journals
Page 1
A Two-Dimensional Multigrid-Driven Navier-Stokes
Solver for Multiprocessor Architectures
Juan J. Alonso, Todd J. Mitty, Luigi Martinelli, and Antony Jameson
Department of Mechanical and Aerospace Engineering
Princeton University
Princeton, New Jersey 08544 U.S.A.
Abstract
A two-dimensional unsteady Navier-Stokes solver has been parallelized using a domain decomposition
approach and the PVM message passing library. Several options for the treatment of multigrid and
implicit residual smoothing are examined. Results for the unsteady flow over a pitching NACA 64A010
airfoil are presented.
1 INTRODUCTION
In recent years, computational fluid dynamics (CFD) has been gaining acceptance as a design tool in industry.
Advancements in algorithm development and computational hardware have led to more complex modeling
of fluid flows. Although current inviscid models can accurately predict the coefficient of lift for an airfoil
in transonic flow, viscous effects such as shock wave/boundary layer interaction can significantly modify
important aspects ofa flow. Furthermore, unsteady phenomena ofpatent viscous character, such as buffeting,
can not be predicted with the help ofinviscid models. Such viscous phenomena directly impact the design
ofengineering configurations, and therefore, it is necessary to enhance the viscous prediction capability of
CFD tools.
Increasingly complex fluid flow models require high performance computing facilities. A cost effective
solution for problems of this type requiring fast CPUs and large internal memory is the use of a parallel
computing paradigm. For computational efficiency, one typically incorporates convergence acceleration tech-
niques such as multigrid and implicit residual smoothing. Message passing becomes necessary in this new
environment, and severely limits the performance of processes that are inherently communication intensive.
In this paper we present a parallelized version ofa well established Navier-Stokes solver, FLO103 [1].
This computer program has recently been enhanced to include Jamesons implicit multigrid approach [2] for
the efficient calculation ofunsteady viscous flows. Calculations are performed on an IBM SP1 multiproces-
sor computer with a domain decomposition approach, and message passing is handled by PVM (Parallel
Virtual Machine) software [3]. Several methods for implementing convergence acceleration techniques such
as multigrid and implicit residual smoothing are studied.
2 NAVIER-STOKES EQUATIONS DISCRETIZATION
The two-dimensional, unsteady, compressible Navier-Stokes equations may be written in divergence form for
a Cartesian coordinate system (x, y) as

w
t
+

f
x
+

g
y
=

R
x
+

S
y
,
(1)
where w is the vector offlow variables, f and g are the convective fluxes, and R and S are the viscous fluxes
in each ofthe coordinate directions. With Reynolds averaging, turbulence effects can be taken into account
with a turbulence model. In this work, a Baldwin-Lomax model was used. In integral form, Equation 1 can
1
Page 2
be applied to each finite volume ofa computational domain to yield a set ofcoupled first-order differential
equations ofthe form
d
dt
(w
ij
V
ij
) + E(w
ij
) + NS(w
ij
) + D(w
ij
) = 0
,
(2)
where E(w
ij
) are the convective Euler fluxes, NS(w
ij
) are the Navier-Stokes viscous fluxes, and D(w
ij
) are
the artificial dissipation fluxes added for numerical stability. For unsteady problems, Equation 2 is modified
by introducing a pseudo-time formulation to improve computational performance [2].
3 PARALLELIZATION STRATEGY
FLO103P is parallelized using a domain decomposition model, a SIMD (Single Input Multiple Data) strategy,
and the PVM Library for message passing. Flows were computed on a C-mesh of size n
i
n
j
= 102464. This
domain was decomposed into subdomains containing
n
i
N
p
n
j
points, where N
p
is the number ofsubdomains
used. Communication between subdomains was performed through halo cells surrounding each subdomain
boundary. A two-level halo was sufficient to calculate the convective, viscous, and dissipative fluxes for all
cells contained in each processor. In the coarser levels ofthe multigrid sequence, a single level halo suffices
since a simplified model ofthe artificial dissipation terms is used.
For problems with a low task granularity (ratio ofthe number ofbytes received by a processor to the
number of floating point operations it performs), large parallel efficiencies can be obtained. Unfortunately,
convergence acceleration techniques developed in the 1980s base their success on global communication in
the computational domain. Thus, current multigrid and implicit residual smoothing techniques [4] are bound
to hinder parallel performance in traditional mesh sizes. In order to effectively deal with the parallelization
ofthese two techniques, we propose several ideas.
In this paper, static load balancing is performed at the beginning of the calculation. Since the number
ofcells remains constant throughout the calculations, the domain can be partitioned into subdomains with
an equal number ofcells. Except for some domains where an additional message across the wake is required,
this provides for perfect load balancing.
4 PARALLEL MULTIGRID
The full approximation multigrid technique enhances the convergence rate of a scheme by performing com-
putations on a series ofincreasingly coarser meshes. The calculations performed in each ofthese meshes are
driven by the residuals at the previous finer level, and the results obtained are interpolated to the corre-
sponding finer mesh. The explicit time-stepping acts as a smoothing operator for the high frequency errors
in each level, thus damping the dominant error mode at the finest level and accelerating convergence. A
more detailed discussion of this procedure can be found in [4]. If domain decomposition is used for parallel
calculations, the size ofthe meshes contained in each processor at the coarser levels ofthe multigrid cycle
is quite small (typically 8 or 16 cells). Most ofthe CPU time is then wasted sending information back and
forth between processors. This situation worsens for multigrid W-cycles, where equilibrium in the coarser
meshes is established repeatedly before traversing the series back to the finest level. Previous authors [5]
have attempted to deal with this problem in the Euler equations by limiting message passing to only some
stages in the Runge-Kutta time-stepping scheme, or passing the boundary data exclusively at the finest level
in the series. This usually led to a decrease in the convergence rate ofthe numerical algorithm, presenting a
clear tradeoff between the improved parallel performance and the increasing number of cycles required for a
similar level ofconvergence. In this work, we examine three different approaches to the implementation ofthe
multigrid algorithm. First, the full multigrid algorithm is implemented with message passing at all required
points such that the parallel program exactly recovers the results ofthe serial code. Second, multigrid is
applied independently within each subdomain to completely avoid inter-processor communication. Third,
and last, at coarse multigrid levels where communication overhead dominates CPU time, a single processor
will be used to gather, compute, and scatter information. Parallel speed-up curves for these three cases are
also presented.
2
Page 3
"DUMB" PROCESSOR W-CYCLE
Parallel
T
T
T
T
T
T
T
Fine
Coarse
Single Proc.
"LAZY" PROCESSOR W-CYCLE
Parallel
T
T
T
T
T
T
T
Fine
Coarse
No work
Figure 1: Lazy-Dumb Multigrid Approach.
4.1 Full multigrid approximation
This first approach was an attempt to implement a parallel program that exactly reproduced the output of
the single processor code in which message passing is not necessary. In order to achieve this goal, boundary
information was passed among processors on all multigrid levels at the beginning of all five stages of the
Runge-Kutta time-stepping. Additional messages were required in order to process convergence information,
calculate force coefficients, and compute the eddy viscosity in the turbulence model. This procedure recovers
the serial version convergence rate, at the expense ofpoor parallel performance.
4.2 Implicit multigrid within subdomains
The second approach used to deal with the multigrid algorithm was to completely decouple the subdomains
on the coarser levels ofthe series in order to minimize the number ofmessages passed when very little
computation is being performed. As expected, this approach restores parallel efficiency to higher levels, but
it suffers from a decreased convergence rate. Clearly, the success of the multigrid technique is due to the
increased communication between different parts ofthe computational mesh at all levels, and the transfer of
this information is limited by the isolation of the coarser levels in each subdomain. The degradation of the
convergence rate is especially bad when interdomain boundaries lie close to regions ofhigh gradients (such
as shock waves and stagnation points). In some ofthese cases, a converged solution cannot be obtained.
4.3 Lazy-Dumb multigrid approach
In order to preserve the convergence rate ofthe original multigrid algorithm, and somehow improve the
parallel efficiency ofthe first approach we propose the following procedure: calculations on the finer levels are
performed as in 4.1; when the algorithm shifts the solution to a user-specified coarse mesh, all the processors
in the calculation pass their flow variables and grid locations to a single processor which computes the coarser
levels of multigrid without the need for message passing. This processor is termed dumb since it performs
everyone elses work. The rest ofthe processors in the computation wait until the calculation needs to be
performed on the finer levels, at which point they receive the information from the dumb processor and
proceed once more in parallel. These processors are called lazy since they avoid carrying out part ofthe
work that was, in theory, assigned to them. In our program, the level at which the transfer is done can be
specified as an input, allowing for investigations of the optimal location of this transition point. Figure 1
presents this procedure graphically. Note that this construction can be extended to a hierarchy ofdumb
processors for larger calculations involving more processors.
3
Page 4
5 IMPLICIT RESIDUAL SMOOTHING
Implicit residual smoothing (or averaging) is a technique that couples the residuals at any given point with
those ofall the other cells in the domain, increasing the support ofthe scheme, and allowing a larger time step
than that permitted by the Courant-Friedrichs-Lewy (CFL) restriction. Once more, this is a communication
intensive procedure that negatively impacts parallel performance in a multiprocessor environment.
5.1 Fully implicit residual smoothing
Serial implementations ofthis technique usually split the problem into the implicit coupling along each of
the two coordinate directions. The direction which is normal to the airfoil surface presents no difficulties
since all the required data resides in the appropriate processors (in this decomposition), allowing calculations
to proceed in parallel. In the coordinate direction that is parallel to the airfoil surface, the residuals that
are to be coupled reside in different processors. Moreover, the solution procedure (Thomas algorithm for a
tridiagonal system) is inherently serial. A brute force method allows only one processor to be active at a
time in the forward elimination and back substitution phases, while the remaining processors are idle. Since
the residual smoothing procedure consumes about 30% ofthe time ofthe total calculation, this idle time
causes parallel performance to drop-off considerably.
5.2 Implicit residual smoothing within subdomains
In a very similar fashion to the multigrid performed implicitly within blocks, and in an attempt to reduce
the amount ofmessages passed, residual smoothing was performed implicitly within blocks, with no global
coupling ofthe residuals. Once more, as expected, although parallel performance improves, the convergence
rate degrades to unacceptable levels. As in the corresponding multigrid procedure, this lack ofcoupling
between domains often led to instability in the calculations.
5.3 Iterative implicit residual smoothing
An alternative to the previous approach is to perform an iterative solution of the smoothing problem.
Messages containing the boundary residuals are passed at the beginning ofeach iteration, and the relaxation
process proceeds in parallel. It has been observed in practice that in order to obtain the increase in CFL
number provided by the fully implicit version (from 3 to about 6), at least two iterations are required.
In these calculations, three iterations were performed and a CFL number of 6 was used without stability
problems. The matrix problem is setup in each subdomain and a Jacobi relaxation procedure is performed
on all subdomains concurrently.
6 RESULTS
A summary of the results for the different multigrid approaches is presented in Figure 2. The full multigrid
approach produces results with an optimum convergence rate, but with a parallel performance that degrades
for a large number of processors. One must take into account, that for these meshes (n
i
n
j
= 1024 64)
granularity becomes too large as the number ofprocessors increases. This makes efficient parallel calculations
not viable for a number of processors larger than 8. The multigrid performed implicitly within blocks (see
Section 4.2) exhibits better parallel performance, but at a very high cost. When the total cost to achieve
a solution converged to the same level through these two procedures is computed, the second technique
requires about twice the number ofcycles making this approach less than desirable. Finally, the lazy-
dumb version ofthe algorithm performs slightly better when the dumb processor computes the two
coarsest levels ofthe sequence. When the three coarsest levels are assigned to the dumb processor, parallel
performance degrades since the amount of time required by the dumb processor exceeds that needed for
all processors (with poor parallel performance). Notice that wall time was used to compute the parallel
efficiencies in all cases. Ifthe unused CPU time ofthe lazy processors were to be factored in, the gain
would be larger. It must also be mentioned that the implementation ofthis approach introduces a higher
level ofcomplexity to the multigrid algorithm, since the memory management on both types ofnodes differs
4
Page 5
considerably from the traditional one. As a payoff for the additional work, this approach exhibits the exact
same rate ofconvergence as the full multigrid algorithm.
Evaluating the performance of multigrid in meshes that are relatively small places a strong restriction
on the parallel efficiencies that can be achieved. Calculations using very large meshes (three-dimensional
calculations, two-dimensional LES calculations, etc) will benefit from the full multigrid algorithm and will
achieve parallel performances on the order of 90% since most of the time will be spent on the finer levels of
the sequence. Nevertheless, for engineering calculations, reasonable speed-ups can be obtained through this
method.
The performance results of the different methods used to implement residual smoothing are presented in
Figure 3. The fully implicit approach is clearly unacceptable even for a small number of processors. While
the approach that uses implicit residual smoothing within blocks exhibits a parallel performance of 96% for
8 processors, the degradation ofthe convergence rate disqualifies it as a possible candidate. Finally, the
iterated residual smoothing preserves both the convergence rate and a reasonable parallel speed-up, and is
chosen as the candidate for engineering calculations.
Figure 4 presents a detail of the domain decomposition for a calculation involving four processors for a
typical airfoil section. Figure 5 shows the pressure contours at different points on the oscillation period of
a NACA 64A010 airfoil at a Reynolds number of 10
6
, M

= 0.796, and a reduced frequency, k


c
= 0.202.
These results were obtained with the O-mesh version ofthe program. One can clearly see how the shock
waves strengthen and weaken as the airfoil pitches up and down.
The lines that intersect the pressure contour lines are the interprocessor boundaries ofthe O-mesh.
Perfect continuity ofthese contour lines validates the accuracy ofthe code even for the cases where strong
shocks are close to the interprocessor boundaries. For more details, please refer to [6].
7 CONCLUSIONS
A parallelized, two-dimensional, unsteady Navier-Stokes flow solver has been developed. Parallelization
was realized using a domain decomposition approach to achieve proper load balancing and computational
efficiency. PVM communication software was used for message passing between processors. Strategies for
dealing with two convergence acceleration techniques, namely implicit residual smoothing and multigrid, and
the performance ofeach ofthese techniques have been evaluated.
It is observed that for meshes of sizes typically used in engineering calculations, acceptable parallel
performances can be achieved with up to 8 processors. A larger number of processors is not suitable for this
type ofcalculation. For larger meshes, the multigrid technique is still quite favorable even in multiprocessor
architectures. Implicit residual smoothing can be performed in an iterative fashion without an observable
impact in the convergence rate while retaining good parallel performance. All calculations used the public
distribution ofthe PVM software developed at Oak Ridge National Labs. Preliminary results with the IBM
optimized version ofthis message passing standard (PVMe), on both the SP1 and SP2 platforms, confirm
the expected trends in which parallel performances improve considerably. Finally, results for the unsteady
pressure field around a pitching NACA 64A010 at M

= 0.796 were computed on the IBM SP1 system.


References
[1] Martinelli, L., Jameson, A., Validation ofa Multigrid Method for the Reynolds Averaged Equations,
AIAA Paper 88-0414, AIAA 26th Aerospace Sciences Meeting, Reno, January 1988.
[2] Jameson, A., Time Dependent Calculations Using Multigrid, with Applications to Unsteady Flows
Past Airfoils and Wings, AIAA Paper 91-1546, AIAA 10th Computational Fluid Dynamics Conference,
Honolulu, June 1991.
[3] Geist, A., Beguelin A., Dongarra J., Jiang W., Manchek, R., Sunderam V., PVM 3 Users Guide and
Reference Manual, Oak Ridge National Laboratory, May 1993.
5
Page 6
1
2
3
4
5
6
7
8
1
2
3
4
5
6
7
8
Number of Processors
Parallel Speedup
Comparison of Parallel Speedups of Multigrid Methods
solid Full multigrid
Implicit within blocks
.. LazyDumb 2 levels in dumb processor
..... LazyDumb 3 levels in dumb processor
Figure 2: Summary ofParallel Speed-ups for Differ-
ent Approaches to the Multigrid Technique.
1
2
3
4
5
6
7
8
1
2
3
4
5
6
7
8
Number of Processors
Parallel Speedup
Comparison of Parallel Speedups of Residual Smoothing Methods
solid fully implicit
.. implicit within blocks
explicit iteration
Figure 3: Summary ofParallel Speed-ups for Differ-
ent Implicit Residual Smoothing Approaches.
129x6 5
129x6 5
129x6 5
129x6 5
Figure 4: Domain Decomposition for a NACA 0012 Airfoil at a 10 degree Angle of Attack.
[4] Jameson, A., Transonic Flow Calculations, Princeton University Report 1651, March 1984, in Numerical
Methods in Fluid Dynamics, edited by F. Brezzi, Lecture Notes in Mathematics, Vol. 1127, Springer-
Verlag, 1985, pp. 156-242.
[5] Yadlin, Y. and Caughey, D. A., Block Multigrid Implicit Solution ofthe Euler Equations ofCompressible
Fluid Flow, AIAA Journal, 29(5):712-719.
[6] Alonso, J. J., Martinelli, L., and Jameson A., Multigrid Unsteady Navier-Stokes Calculations with
Aeroelastic Applications, 33rd AIAA Aerospace Sciences Meeting, AIAA Paper 95-0048, Reno, NV,
January, 1995.
6
Page 7
1
2
3
4
5
6
Figure 5: Mach Number Contours. Pitching Airfoil Case. Re = 1.0 10
6
, M

= 0.796, K
c
= 0.202. Read
figures by lines.
7
This is the html version of the file http://www.aoe.vt.edu/%mason/Mason&f/'onfig(ero)i*ift.pdf.
G o o g l e automatically generates html versions of documents as we crawl the we.
To lin! to or oo!mar! this page" use the following url: http://www.google.com/search?q=cache:/h12et/e1a1J:www.aoe.3t.ed(/4maso!/aso!_f/Co!figAero5i6ift.pdf&'o(r!als&
)performa!ce&of&differe!t&airfoils&i!&cfd*+hl=e!+gl=i!+ct=cl!,+cd=$7
'oogle is neither affiliated with the authors of this page nor responsi%le for its content.
These search terms have een highlighted: performance different airfoils cfd
These terms only appear in lin!s pointing to this page: journals
Page 1
W.H. Mason
4/5/06
8. High-Lift Aerodynamics
8.1 Introduction: Why high lift?
For transonic transports, the high-lift system design is a critical part of the configuration
design.
To achieve reasonable field performance while also obtaining efficient transonic
cruise the
design will require a fairly sophisticated high lift system. From a paper by Boeing
aerodynamicists
1
: (presumably referring to the B-777)
A 0.10 increase in lift coefficient at constant angle of attack is equivalent to reducing
the
approach attitude by one degree. For a given aft body-to-ground clearance angle, the
landing gear may be shortened for a savings of airplane empty weight of 1400 lb.
A 1.5% increase in maximum lift coefficient is equivalent to a 6600 lb increase in
payload
at a fixed approach speed
A 1% increase in take-off L/D is equivalent to a 2800 lb increase in payload or a 150
nm
increase in range.
For fighters, devices are also scheduled allow efficient maneuver.
High-lift systems are also critical for STOVL and V/STOL aircraft. They also use the
propulsion
system to help generate the lift. It always seemed to me peculiar to design fighter aircraft,
or
virtually any military aircraft, to operate from traditional runways. The one thing the
adversary is
going to know is the exact location of your runways. So a STOVL capability seems to be
critical
in a serious confrontation.
Current status: Typical values of C
Lmax
are shown in Table 1. They come from papers by Brune
and McMasters,
2
Roskam and Lan,
3
and Sanders.
4
Table 1. Values of C
Lmax
for some airplanes.*
* Note that there is a significant variation of values from different sources.
Clearly the 727 emphasized short fields, and thus required a higher C
Lmax
Anyone who ever
looked out the window while landing in a 727 noticed the elaborate high lift system
employed.
Model
C
Lmax
B-47/B-52
1.8
367-80/KC-135
1.78
707-320/E-3A
2.2
727
2.79
DC-9
3.0
737-200
3.2
747/E-4A
2.45
767
2.45
777
2.5
Page 2
8-2 W. H. Mason, Configuration Aerodynamics
Some Key Aspects:
Compressibility can be important early.
Reynolds number scaling from WT to flight may be problematic.
Today simple high lift systems are critical, the high manufacturing cost for high lift
systems
is important.
Classes of problems
High lift for a single element airfoil
Multi-element airfoils
Use of blowing in some form: Powered Lift
Computing:
Requires consideration of viscous immediately. (unlike typical cruise airfoil analysis
and
design, where some insight can usually be gained ignoring viscous effects).
Predicting high-lift is done almost entirely with Navier-Stokes (RANS) codes
(exception,
Prof. Mark Drelas MSES
5
code). A recent summary of the computational capability is by
Rumsey and Ying.
6
Single element airfoils
The key example of how to obtain high lift on a single element airfoil is the story of
Liebecks high lift airfoil
7
and the Stratford pressure recovery shape of the pressure
distribution. This introduces the classic paper by A.M.O. Smith.
8
See section 8.5 below.
Multi-element airfoils:
Understanding the physics: Section 6.3 of A.M.O. Smiths paper
8
is critical to understanding
the physics. Know what is meant by: 1. The slat effect, 2. The circulation effect, 3., The
dumping effect, 4. Off-the-surface pressure recovery, and 5. The fresh boundary layer
effect.
Note: some people combine the circulation and dumping effects and call it the vane
effect.
See section 8.5 below for details.
Design: See the recent survey by C.P. van Dam,
9
Existing systems on commercial transports: Rudolph has surveyed the high-lift systems
on
current subsonic transport aircraft.
10
8.2 Types of Trailing Edge Devices
To begin we include a number of examples of high-lift devices as drawn by Dick Kita of
Grumman.
11
These are fairly realistic drawings, as opposed to many of the drawings in
textbooks, which are cartoonish. Kita worked in high lift for many years. Im aware of his
work
on the Gulfstream II and F-14, but he worked on many other Grumman aircraft high-lift
systems
as well.
Of the large number of papers addressing high lift, several deserve mention. Pepper, et al
12
provides a more current look at high lift, while the Boeing 777 high lift system
development, as
well as the overall design process, is available in the paper by Nield.
13
A valuable description of
high lift on transports in contained in Gratzer.
14
The use of powered lift is covered in the survey
by Korbacher.
15
Somewhat dated but valuable resources are the book by Hoerner and Borst,
16
and the book by McCormick
17
(recently reissued unchanged from the 1967 edition as a Dover
paperback). Perhaps the best chapter on high lift in a basic text is the chapter in Shevell.
18
Page 3
High-Lift Aerodynamics 8-3
a) basic devices
Figure 8-1. Examples of typical trailing edge devices. Note that the Fowler flap also adds
area,
so that part of the C
Lmax
increase is simply due to the use of the original reference area
in computing the C
L
. (from Dick Kitas Grumman talk, Feb. 1985)
Page 4
8-4 W. H. Mason, Configuration Aerodynamics
b) other trailing edge devices
Figure 8-1. Examples of typical trailing edge devices.
(from Dick Kitas Grumman talk, Feb. 1985)
Page 5
High-Lift Aerodynamics 8-5
8.3 Types of Leading edge devices
Figure 8-2. Typical leading edge device concepts.
(from Dick Kitas Grumman talk, Feb. 1985)
Page 6
8-6 W. H. Mason, Configuration Aerodynamics
Figure 8-3. The actual high-lift system employed on the F-14.
(from Dick Kitas Grumman talk, Feb. 1985)
Notice that the F-14 has a fairly elaborate scheme, where the cove region is smoothed
with a
moving flap, and the upper surface of the slat also has a movable piece to fair the flap.
Also note
the spoilers. Many fighter airplanes use spoilers instead of ailerons for roll control,
although their
use varies from company to company.
Page 7
High-Lift Aerodynamics 8-7
8.4 Aerodynamics of Leading and Trailing Edge Devices
Figure 8-4. Typical effect of flaps on lift (from Dick Kitas Grumman talk, Feb. 1985)
This figure illustrates the effects of flap deflection on lift. Note that the angle of attack for
C
Lmax
actually decreases as the flap is deflected.
Page 8
8-8 W. H. Mason, Configuration Aerodynamics
Figure 8-5. Effect of flap extension on lift.
(from Dick Kitas Grumman talk, Feb. 1985)
Flaps that extend in a Fowler motion also benefit from additional wing area. Because we
continue to use the same reference area, the C
Lmax
value increases.
Page 9
High-Lift Aerodynamics 8-9
Figure 8-6. Typical variation of C
Lmax
with Mach and Reynolds Number.
(from Dick Kitas Grumman talk, Feb. 1985)
In general, we expect C
Lmax
to increase with Reynolds number as shown here for a clean airfoil.
However, sometimes the projection is not so straightforward, and C
Lmax
may even decrease.
19
Considering the adverse effect of Mach number on C
Lmax
, notice the low Mach numbers at which
the effects take place. Its rather surprising to the uninitiated.
Page 10
8-10 W. H. Mason, Configuration Aerodynamics
Figure 8-7. Effect of leading edge slats.
(from Dick Kitas Grumman talk, Feb. 1985)
Leading edge slats work to protect the leading edge from separation. Therefore, they
dont really
do anything until you reach the angle of attack where the leading edge flow would let
go
without the slats for protection. Thus the slats allow the lift to continue to rise to higher
angles of
attack.
Page 11
High-Lift Aerodynamics 8-11
Figure 8-8. Effects of various types of leading edge devices on C
Lmax
.
(from Dick Kitas Grumman talk, Feb. 1985)
Different leading edge devices differ in their effectiveness. This is Kitas estimate of
how each
type of device affects the performance of the wing.
Page 12
8-12 W. H. Mason, Configuration Aerodynamics
Figure 8-9. Estimated performance of various types of high lift systems.
(from Dick Kitas Grumman talk, Feb. 1985)
This is Kitas estimate of the typical best performance you can get from various types
of high-
lift systems. You can see that sweeping the trailing reduces the effectiveness of all the
systems.
The curve labeled advanced is typical of projections made in advanced departments,
where the
assumption is that an advanced technology development effort can improve the
performance of
any system. This may or may not be true.
Page 13
High-Lift Aerodynamics 8-13
Figure 8-10. Effects of flap deflection on drag.
(from Dick Kitas Grumman talk, Feb. 1985)
In addition to lift, flaps make a large change in drag. Clearly, you dont want the flaps
deployed
at low lift, and thus flaps arent deflected in cruise. As lift increases, there may be an
optimum
flap deflection schedule, and this is done, for example, on the F-18, where the flaps are
scheduled with angle of attack and Mach number. This was also done on the Grumman
X-29.
Page 14
8-14 W. H. Mason, Configuration Aerodynamics
Figure 8-11. Flap effects on pitching moment.
(from Dick Kitas Grumman talk, Feb. 1985)
Flap deflection also produces a large change in pitching moment. This is an important
consideration, since you need to be able to trim this pitching moment. In the case of the
Beech
Starship, the canards actually changed sweep to be able to generate the required force. So
this
effect cannot be ignored in developing the high lift system.
Page 15
High-Lift Aerodynamics 8-15
Figure 8-12. Definition of gap and overlap.
(from Dick Kitas Grumman talk, Feb. 1985)
In developing the high lift system, the selection of the gap between the slot and main
element,
and the overlap have been found to be two of the key parameters. A lot of wind tunnel
time,
and more recently computer resources, are spent trying to identify the values of these
parameters
that produce the highest lift. Figure 8-12 provides Kitas definition of these parameters.
8.5 Physics of high lift: A.M.O. Smiths analysis of the high lift aerodynamics
Now that weve surveyed the characteristics of high lift systems, we need to examine the
physical basis for the operation and limits of high lift systems. A.M.O. Smith wrote the
book
on the physics of high-lift systems
8
and is required reading. His message: you need to carry as
much lift (load) as you can on the airfoils upper surface without separating the boundary
layer.
His classic paper describes the physics associated with the high-lift characteristics we
described
above, and how to achieve the available high lift performance. We summarize his
description
here.
To obtain insight into the characteristics of pressure distributions as they affect boundary
layer separation, Smith introduced the use of a canonical pressure distribution. He felt
strongly
that this was necessary to understand and compare possible separation on different
airfoils. It is
Page 16
8-16 W. H. Mason, Configuration Aerodynamics
essentially another type of dimensionless or scaled pressure distribution. For boundary
layer
investigations it is found that the best scaling factor is the velocity just before the
deceleration
begins. In Smiths canonical system, C
p
= 0 represents the start of the pressure rise and C
p
= +1
the maximum possible value, that is, u
e
= 0.The velocity at the start of the pressure rise is u
0
.
Thus, the canonical pressure distribution is defined as
C
p
= 1
u
e
u
0

2
The next step is to examine the best way to specify the pressure distribution to allow the
pressure to recover to as close as possible to C
p
= +1. Smith made a parametric study of various
possibilities to gain insight into the best way to prescribe a pressure distribution to
delay
separation. However, here we look at his limiting case. It makes use of an analysis by
Stratford
that was done to estimate separation before the days when boundary layer computer
programs
were available and their use routine (1959). Using Stratfords analysis, it is possible to
define a
pressure distribution where the boundary layer is everywhere just on the verge of
separation. To
do this, we manipulate the Stratford criteria:
C
p
x
dC
p
dx

10
6
R
(
)
1
10
1
2
= S
Note that originally, Stratford used this relation to say that separation occurred when the
quantity
on the LHS of the equation reached the value of S (typically 0.35). However, we can
define a Cp
distribution using this relation that is everywhere equal to S, just on the verge of
separation, and
this pressure distribution is the best way to achieve a very large pressure recovery without
separation. Figure 8-13 shows the resulting pressure distribution. Examining this pressure
distribution, several key observations can be made. The initial slope is infinite, and then
decreases. Thus, when the boundary layer is thin, it can withstand a very large pressure
gradient.
As the boundary layer thickens (either when it starts to recover, or as it recovers) it
cannot
sustain the large pressure gradient, and the pressure gradient to maintain attached flow
decreases.
This illustrates the idea that thick boundary layers are more likely to separate than thin
boundary
layers. Note also that the Reynolds number effect is relatively weak. Finally, the
boundary layer
could recover all the way to C
p
= +1, but to attain this, x would need to go to infinity. These
shapes are the best possible pressure distributions to use to recover the pressure without
separating the boundary layer. as Smith notes, the only way to do better is to use some
sort of
active boundary layer control (suction or blowing).
Page 17
High-Lift Aerodynamics 8-17
-0.20
0.0
0.20
0.40
0.60
0.80
1.0
0.0
0.50
1.0
1.5
2.0
Canonical
Cp
x - feet
solid line U
0
/v = 10^6
dashed line U
0
/v = 10^7
Figure 8-13. Stratford Limiting flows for two different Reynolds numbers.
Single element airfoils: The key is how to obtain high lift on a single element airfoil, and
is
essentially the story of Liebecks high lift airfoil
7
and the Stratford pressure recovery shape of
the pressure distribution described above, as told by Smith.
8
The question of how much lift you
can obtain on a single element airfoil involves how low the pressure can be on the upper
surface,
and how the pressure can recover to a positive pressure coefficient at the trailing edge
and keep
the boundary layer attached. Smith describes two aspects of the problem. In the first case,
he
explains the limit of the pressure coefficient in terms of the vacuum C
p
when a zero pressure is
specified on the airfoil upper surface. Thus, using the definition of C
p
,
C
p
=
p p

1
2

2
we can obtain an alternate form using q = 1
2

2
=

2
P

2
, that is
C
p
=
p p

2
p

2
Page 18
8-18 W. H. Mason, Configuration Aerodynamics
and vacuum C
p
occurs when the pressure is zero:
C
p
vac
=
2
M

2
Then, he points out that only a value of 70% of the vacuum C
p
has been achieved in practice.
This results in a C
p
limit of M

2
C
p
= 1 for of 1.4.
Next, Liebeck and Smith used the analytical analysis by Stratford illustrated above to
specify
a pressure distribution that allows the most lift to be obtained. Given this pressure
distribution,
an inverse method is used to obtain the associated airfoil shape. The result is the Liebeck
family
of high lift airfoils.
Multi-element airfoils: Understanding the physics: read section 6.3 of A.M.O. Smiths
paper.
The five ideas are:
1. The slat effect. The slat protects the leading edge of the main element. Thats why its
effect
is only observed near C
Lmax
of the single element. Thought of as a point vortex, the slat velocity
acts to reduce the velocity around the leading edge of the main element.
2. The circulation effect. The downstream element causes the upstream element to be in a
high
velocity region, inclined to its mean line. To meet the Kutta condition, the circulation has
to be
increased. Instead of the airfoil deflecting as a plain flap, the trailing edge is placed in an
inclined
flow, something else (the downstream element) turns the flow.
3. The dumping effect. The trailing edge of the forward element is in a region of velocity
appreciably higher than the freestream velocity. Thus, the boundary layer can come off
the
forward element at a higher velocity. You dont have to recover back to Cp = +0.2 for
attached
flow, relieving the pressure rise on the boundary layer, alleviating separation problems
and
permitting increased lift. The suction lift can be increased in proportion to U
TE
2
for the same
margin against separation.
4. Off-the-surface pressure recovery. The boundary layer leaves the trailing edge faster
than the
freestream, and now becomes a wake (a viscous phenomena). The recovery back to
freestream
velocity can be more efficient way from contact with the wall. Wakes withstand more
than
boundary layers. Note that the wake can actually separate out in the flowfield. Note: for
a well
designed high lift system the local boundary layers and wakes remain separate. If they
merge,
everything is more complicated.
5. The fresh boundary layer effect. Thin boundary layers can sustain a greater pressure
gradient
than a boundary layer. Thus, three thin boundary layers (on three airfoil elements) are
more
effective than one thick boundary layer (single element).
Note: some people combine the circulation and dumping effects and call it the vane
effect.
8.6 Computational methods for high lift
Significant effort has been devoted to improving prediction capabilities for high lift
systems. As
stated in the introduction, the best recent survey is by Runsey and Ying.
6
In the meantime, for
low speed predictions of the maximum lift for a single element airfoil, XFOIL can be
used.
experience shows that its predictions are slightly higher than experimental results.
Nevertheless,
this is rea remarkable capability of a code than be run on a laptop PC.
Page 19
High-Lift Aerodynamics 8-19
8.7 Passive and active boundary layer control
Passive Boundary Layer Control: The boundary layer can be prevented from separating
by the
use of vortex generators, snags and fences.
20
Active boundary layer control: If suction or blowing is used to suppress boundary layer
separation, the blowing (which is generally preferred to suction) is known as boundary
layer
control (BLC), if the amount of blowing exceeds the value required for BLC, then the
blowing is
know as powered lift. The key parameter used to describe the amount of blowing is the
blowing coefficient, defined as:
C

=
m
j
V
j
qc
where the subscript j refers to the jet, and q is the dynamic pressure. Blowing for BLC
was used
often in early fighters, but is not used nearly as much today. The F-4 Phantom originally
had
blowing on both the leading edges and over the trailing edge flap. However, to improve
transonic
maneuver characteristics and to improve resistance to departure, the leading edge
blowing was
replaced by leading edge slats.
21
8.8 Powered Lift
A huge class of concepts have been tried to increase maximum lift using high pressure air
from
the engine. Some examples of powered lift concepts are:
propeller slipstream deflection (Brequet 941/McDonnell Model 188)
externally blown flaps (McDonnell DouglasYC-15/C-17)
internally blown flaps
upper surface blowing (Boeing YC-14, NASA QSRA, Ball-Bartoe JetWing)
vectored thrust (AV-8 Harrier)
jet flaps (Hunting H.126)
jet augmentor wings (NASA-deHavilland Augmentor Wing Aircraft)
circulation control (advocated by Hokie Bob Englar
22
, A-6 CCW)
8-9 Configuration Integration issues
The best airplane C
Lmax
you can achieve with a mechanical high lift system is about 3 3.5
The military and civil air regulations require a margin between C
Lmax
and the operating C
L
of the airplane. This must be accounted for during design. For example, the approach
speed must be 1.3 times the stall speed. This would suggest that the maximum approach
C
L
is only 59% of the C
Lmax
. However, some relief is available because the measured stall
speed, V
smin
is usually about 0.94 times the stall speed in 1g steady flight (the wind tunnel
case). This means that you can use a C
L
of about 67% of the C
Lmax
.
9,23
Increased span can be used to reduce the induced drag, so a bigger flap angle can be
used
before a climb limit is encountered, but this is a heavy solution.
Sweep decreases max lift
Page 20
8-20 W. H. Mason, Configuration Aerodynamics
2D to 3D: lots of losses dont be mislead by 2D C
Lmax
values, the real 3D value will be
much less.
The maximum C
L
available for takeoff and landing for many swept-wing airplanes is
actually the limit on angle of attack to avoid tailscrape.
The best high-lift configuration integration description also involves powered lift. The
YC-14 AIAA Case Study
24
is highly recommended.
Finally, issues not covered but worth mentioning: the concept of the Gurney flap
7
and the need
for accuracy around the leading edge.

8-10 Exercises
1. Examine the predictive capability of XFOIL for C
Lmax
. Use the data from Abbott and von
Doenhoff supplied previously for the NACA 0012 and 4412 airfoils at a Reynolds
number of
6 million and free transition. Comment on your results.
2. Read the high lift paper by A.M.O. Smith. Summarize what you learned in one page.
Pay
special attention to the details of single and multielement airfoils described in Sections 3-
6.
8-11 References
1
Garner, P.L., Meredith, P.T., and Stoner, R.C., Areas for Future CFD Development as
Illustrated by Transport Aircraft Applications, AIAA Paper 91-1527, 1991.
2
G.W. Brune and J.H. McMasters, Computational Aerodynamics Applied to High Lift
Systems, Chapter 10 of Applied Computational Aerodynamics, P. Henne, Ed. Progress
in
Astronautics and Aeronautics, Vol. 125, AIAA, Washington, 1990.
3
Jan Roskam and C-T Edward Lan, Airplane Aerodynamics and Performance,
DARcorporation,
Kansas, 1997, pp. 343.
4
Karl L. Sanders, High-Lift Devices, A Weight and Performance Trade-Off
Methodology,
Society of Allied Weight Engineers (SAWE) Technical Paper 761, May 1969.
5
Mark Drela, Design and Optimization Method for Multi-Element Airfoils, AIAA Paper
93-
0969, Feb. 1993.
6
Christopher L. Rumsey and Susan X. Ying, Prediction of high lift: review of present
CFD
capability, Progress in Aerospace Sciences, Vol. 38, pp. 145-180, 2002. (note that
articles in
Progress in Aerospace Sciences are available for download through the Virginia Tech
University
Library if you search Addison and have a vt.edu address)
7
Robert H. Liebeck, Design of Subsonic Airfoils for High Lift, Journal of Aircraft, Vol.
15,
No. 9, Sept. 1978, pp. 547-561.
8
A.M.O. Smith, High-Lift Aerodynamics, 37
th
Wright Bothers Lecture, Journal of Aircraft,
Vol. 12, No. 6, June 1975
9
C.P. van Dam, The aerodynamic design of multi-element high-lift systems for transport
airplanes, Progress in Aerospace Sciences, Vol. 38, pp. 101-144, 2002. (note that
articles in
Progress in Aerospace Sciences are available for download through the Virginia Tech
University
Library if you search Addison and have a vt.edu address)
10
Peter K.C. Rudolph, High-Lift Systems on Commercial Subsonic Airliners, NASA CR
4746, Sept. 1996.

Maintaining an accurate leading edge contour is critical at high lift conditions. Once after a navy depot had
repainted an F-14 wing, a small ridge was left on the leading edge where the upper and lower surface paint
overlapped. This was enough to cause early stall. Ed Heinemann reported a similar experience on a
Douglas airplane
during World War II in his autobiography.
Page 21
High-Lift Aerodynamics 8-21
11
Dick Kita, Mechanical High Lift Systems, Grumman Aerodynamics Lecture Series,
Grumman Aerospace Corporation, Feb. 1985.
12
C.P. van Dam, J.C. Vander Kam, and J.K. Paris, Design-Oriented High-Lift
Methodology for
General Aviation and Civil Transport Aircraft, Journal of Aircraft, Vol. 38, No. 6,
November-
December 2001, pp. 12076-1084.
13
B. N. Nield, An overview of the Boeing 777 high lift aerodynamic design,
Aeronautical
Journal, Nov. 1995. pp. 361-371.
14
L.B. Gratzer, Analysis of Transport Applications for High Lift Schemes, AGARD LS
43,
1971.
15
G.K. Korbacher, Aerodynamics of Powered High-Lift Systems, Ann. Rev. of Fluid
Mech.,
1974.
16
S.F. Hoerner and H.V. Borst, Fluid Dynamic Lift, 1975.
17
B.W. McCormick, Jr., Aerodynamics of V/STOL Flight, Dover, 1999.
18
Shevelle, Fundamentals of Flight, 2
nd
ed., Prentice-Hall, 1989.
19
John H. McMasters and M.D. Mack, High Reynolds Number Testing in Support of
Transonic
Airplane Development (Invited Paper), AIAA Paper 92-3982, July 1992.
20
D.G. Mabry, Design features which influence flow separations on aircraft,
Aeronautical
Journal, Dec. 1988, pp. 409-415.
21
D.H. Bennett and W.A. Rousseau, Seven Wings the F-4 Has Flown, Evolution of
Aircraft
Wing Design Symposium, Dayton, OH, AIAA Paper 80-3042, March 1980.
22
Englar, Robert J., Smith, Marilyn J., Kelley, Sean M., Rover, Richard C., III,
Development of
circulation control technology for application to advanced subsonic transport aircraft,
AIAA
Paper 93-0644.
23
A. Flaig and B. Hilbig, High-Lift design for large civil aircraft, in AGARD High-Lift
System Aerodynamics, R 415, 1993.
24
John K. Wimpress and Conrad F. Newberry, The YC-14 STOL Prototype: Its Design,
Development, and Flight Test, AIAA Case Study, AIAA, Reston, 1998.