Aerodynamics Glossary
Transcript of Aerodynamics Glossary
-
8/11/2019 Aerodynamics Glossary
1/43
Aerodynamics Glossary
Wing Design TipA 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 savethousands of dollars in operational costs over the life of the aircraft. MultiSurfaceAerodynamics is a digital wind tunnel that can compare the performance of many airfoilshapes to make the airfoil selection process easy.
Aerodynamic Center
The aerodynamic center is a point along the airfoil or wing about which the momentcoefficient does not vary with an angle of attack change.AirfoilAn airfoil is the cross section of a wing. The airfoil shape and variations in angle ofattack are primarily responsible for the lift and profile drag of the wing.
Angle of AttackThe 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 toincrease or decrease the lift acting on the wing. An increase in lift often results in anincrease in drag.Center of PressureA point along the airfoil about which the moment due to the lift is ero, i.e., it is the pointof action of the lift. The center of pressure will change its position when the angle ofattack changes.ChordThe 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, thelift 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 mathematicaltechni#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 andentire 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 inthe flow field. The speed of sound in a fluid depends on the ratio of pressure changes todensity changes in the fluid.Drag
&rag is an aerodynamic force opposing the direction of motion. &rag can be due tosurface viscosity (friction drag), pressure differences due to the shape of an ob'ect (formdrag), lift acting on an finite wing (induced drag) and other energy loss mechanisms in
-
8/11/2019 Aerodynamics Glossary
2/43
the flow such as wave drag due to shock waves and inefficiencies in engines.Drag CoefficientThe drag coefficient is defined as the drag(dynamic pressure reference area). Thereference area is usually the plan*form or flat pro'ection (the wing+s shadow at noon) areaof the wing.
Dynamic PressureThe dynamic pressure is defied as the product of the density and the s#uare of thevelocity divided by two. The dynamic pressure has units of pressure, i.e. "orceArea. Thedynamic pressure is used to non*dimensionalie forces and pressures in aerodynamics.Flap Deflection AngleThe flap deflection angle is the angle between the deflected flap and the chord line. Theangle is positive for a downwards deflection of the flap. &eflect the flap downwards toincrease the airfoil+s lift.LiftThe lift is a force acting perpendicular to the direction of flight. The lift is e#ual to thefluid 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 entireairplane.Lift CoefficientThe lift coefficient is defined as the lift(dynamic pressure reference area). Thereference area is usually the plan*form area of a wing or horiontal pro'ection of thewing.ean aerodynamic chordThis chord is located along the wing and has the aerodynamic property of the two*dimensional wing.!ACA Airfoils
-A!A airfoils are wing cross sectiondesigns invented by the !ACAorganiation.-A!A eventually became -ASA (-ational Aeronautics and Space Administration). $ereare a few popular airplanes that have -A!A airfoil wings
Airplane "oot Airfoil Tip Airfoil
/eech 01 Twin /onana -A!A 23145.4 -A!A 23142
/*46 "lying "ortress -A!A 1142 -A!A 1141
!essna 402 -A!A 2542 -A!A 1142
!essna 462 4763*later -A!A 2542 -A!A 2542 mod
!essna 001 !itation -A!A 23145 -A!A 23142
&ouglas &!*3 -A!A 2240 -A!A 2218
"airchild A#$%Thunderbolt -A!A 8648 -A!A 8643
Sikorsky S*84 S$*3 Sea 9ing -A!A 1142 -A!A 1142
http://www.hanleyinnovations.com/whatsanairfoil.htmlhttp://www.hanleyinnovations.com/whatsanairfoil.html -
8/11/2019 Aerodynamics Glossary
3/43
Panel ethodThis numerical method places singularities along the airfoil. n the case of :isual"oilor3DFoil, the singularities are vortices. The vorticity is distributed linearly along the panel.
Plain FlapA plain flap is a hinge attachment near the trailing edge of an airfoil. The length of theflap is measured as a percentage of the chord and the deflection is measured in degrees.Pressure CoefficientThe pressure coefficient is a non*dimensional form of the pressure. t is defined as thedifference of the free stream and local static pressures all divided by the dynamicpressure."eynolds !um&er
The ;eynolds number is a non*dimensional parameter that compares the inertia toviscous forces. f the ;eynolds number is low, then viscosity plays an importatant part in
the simulations.'tallAt low angles of attack, the lift developed by an airfoil or wing will increase with anincrease in angle of attack. $owever, there is a ma%imum angle of attack after which thelift 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 minimumlanding and takeoff speeds of an airplane.'treamlines!ontours in the flow field that are tangent to the velocity vector.Wing LoadingThe total weight of the airplane divided by the plan form area of the wing.
Wing 'panThe span is the total length of the wing.
http://www.hanleyinnovations.com/air_16.htmlhttp://www.hanleyinnovations.com/3dfoil.htmlhttp://www.hanleyinnovations.com/air_16.htmlhttp://www.hanleyinnovations.com/3dfoil.html -
8/11/2019 Aerodynamics Glossary
4/43
A&out Dr anley
&r.
-
8/11/2019 Aerodynamics Glossary
5/43
MultiSurface Aerodynamics features include
stand*alone application (built*in pre D post processing tools)
easily enter and manipulate comple% wing systemF instantlycompute lift, drag (profile and vorte%), moments and stabilityderivativesF 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 computerprogramming skills
accurate solvers for realistic simulations
computer aided design interface to model and edit 3& wingsystems ('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
ApplicationsMultiSurface Aerodynamics is the ultimate desktop wind tunnel software. ?se it to design and test comple%wing systems directly on your Eindows
@ight aircraft wing design D analysis. This includes kites, powered parachutes, parafoils and othersporting e#uipment.
?A: design D analysis
Multi*plane wing configurations including 'oined wings
Sailboatyacht keel and rudder design D analysis
?p*wind sail design D analysis
Shipsailboat hydrofoil design D analysis
$uman powered hydrofoil design D analysis
http://www.hanleyinnovations.com/msajwing.htmlhttp://www.hanleyinnovations.com/multisurface/#How%23Howhttp://www.hanleyinnovations.com/multisurface/msa_glider.htmlhttp://www.hanleyinnovations.com/multisurface/msa_glider.htmlhttp://www.hanleyinnovations.com/msajwing.htmlhttp://www.hanleyinnovations.com/msajwing.htmlhttp://www.hanleyinnovations.com/endplates.pdfhttp://www.hanleyinnovations.com/sprintwing.pdfhttp://www.hanleyinnovations.com/sprintwing.pdfhttp://www.hanleyinnovations.com/multisurface/msaexperiment01.htmlhttp://www.hanleyinnovations.com/multisurface/msaexperiment02.htmlhttp://www.hanleyinnovations.com/multisurface/msaellipse.htmlhttp://www.hanleyinnovations.com/multisurface/msurf.htmlhttp://www.hanleyinnovations.com/whyhanley2.htmlhttp://www.hanleyinnovations.com/whyhanley2.htmlhttp://www.hanleyinnovations.com/whyhanley2.htmlhttp://www.hanleyinnovations.com/msajwing.htmlhttp://www.hanleyinnovations.com/multisurface/#How%23Howhttp://www.hanleyinnovations.com/multisurface/msa_glider.htmlhttp://www.hanleyinnovations.com/msajwing.htmlhttp://www.hanleyinnovations.com/endplates.pdfhttp://www.hanleyinnovations.com/sprintwing.pdfhttp://www.hanleyinnovations.com/multisurface/msaexperiment01.htmlhttp://www.hanleyinnovations.com/multisurface/msaexperiment02.htmlhttp://www.hanleyinnovations.com/multisurface/msaellipse.htmlhttp://www.hanleyinnovations.com/multisurface/msurf.htmlhttp://www.hanleyinnovations.com/whyhanley2.html -
8/11/2019 Aerodynamics Glossary
6/43
Eater sport e#uipment design. This includes sailboard fins and sail configurations.
=#uipment for marine industry such as foils for stabiliing vessels D trawl doors.
Automobile wing, spoilers and surfaces designed for safety and fuel efficiency.
@ectures and presentations in the area of aerodynamics design and analysis.
+nline Tutorials * Demos
Tutorial #$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 analye4oined 0ingsfor ?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. Moreinformation ...
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 engineeringknowledge with MultiSurface Aerodynamics to develop outstanding products and services. The followingcompanies, 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
-
8/11/2019 Aerodynamics Glossary
7/43
?nited States Air "orce ?niversitat
5, 0 D 8*digit airfoils. n addition, the user can enter customairfoils for analysis and modifye%istingairfoil 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
http://www.wpafb.af.mil/http://www.upc.es/http://www.ucsd.edu/http://www.ed.ac.uk/http://www.hanleyinnovations.com/air_16.html#Purchase%23Purchasehttp://www.wpafb.af.mil/http://www.upc.es/http://www.ucsd.edu/http://www.ed.ac.uk/http://www.hanleyinnovations.com/air_16.html#Purchase%23Purchase -
8/11/2019 Aerodynamics Glossary
8/43
@ift, drag D moment coefficient calculations
http://www.hanleyinnovations.com/transonic.pdfhttp://www.hanleyinnovations.com/transonic.pdfhttp://www.hanleyinnovations.com/vfprop1.htmlhttp://www.hanleyinnovations.com/transonic.pdfhttp://www.hanleyinnovations.com/vfprop1.html -
8/11/2019 Aerodynamics Glossary
9/43
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 ;) @B#@@to place your credit card order by telephone. Ee canprovide you with a download ?;@ D password after the transaction is processed.
Fa2 ,our +rders
-
8/11/2019 Aerodynamics Glossary
10/43
As owner and chief software author of $anley nnovations, &r. $anley has written a number of softwarepackages includingAirfoil/rowser, Airfoil Grganier,Science Iraphs, :isual"oil, Model"oil,Aerodynamics in
-
8/11/2019 Aerodynamics Glossary
11/43
Airfoilshave been represented in a number of differentways in the past. For example, coordinates have
been directly used to fit airfoil shapes using B-splines and Bzier curves2
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 airfoilsin a wing design
study. However, the basis functions derived by Robinson and Keane5
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 airfoilsmay 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 CFDmodels, 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 2effectively 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 differentparameterisation methods.
Wu6
compared three geometric representations in three case studies of cascade blade design using adjointmethods. 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 differentshapes. Some parameterisation methods, for example, coordinate-
based methods, can accurately represent a variety of dramatically differentshapes 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 designapproach is adopted to compare differentparameterisations.
In this work, three datum airfoils, two from the NACA supercritical airfoilsfamily (NACA0406 and
NACA0610) and the third being the RAE2822, are used as reference shapes to compare differentparam-
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
-
8/11/2019 Aerodynamics Glossary
12/43
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).
Samareh7
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 differentapproaches
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 airfoilsare 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 analytically4
or numerically,5as 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 airfoilswas developed by Robinson and Keane5
to
provide an efficient means to define the airfoil for optimi-
sation study in preliminary design, for example.
y(x) =0
y0
(x) +
wi
fi
(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 Sobieczky9
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 is2 of 8
American Institute of Aeronautics and Astronautics
Page 3also used by Li11, 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
-
8/11/2019 Aerodynamics Glossary
13/43
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 differentparameterisation 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 airfoils8
and RAE2822 are chosen in the current work as ref-
erence airfoilsin 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 AnalysisX/C
Y
/
C0
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 differentparameterisa-
tion methods in representing differentairfoil shapes, three
existing airfoilsNACA0406, 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(ftarget
(xi
) f(xi
))
-
8/11/2019 Aerodynamics Glossary
14/43
(2)
where xi
(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 theeffect 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 Gambit14
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 4the 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 AnalysisFigure 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 higherdimensions, and chances of failure or not generating
desired shapes will be higher.
The best results for approximations of the tar-
get airfoilsare shown in Table 2. The results are
produced by minimising the objective function computed using (2). A genetic algorithm (GA) from OP-
TIONS15
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
-
8/11/2019 Aerodynamics Glossary
15/43
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 airfoilsand 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 airfoilsNACA0406 and NACA0610 than for the RAE2822, this is not surprising,as the set of basis functions used were derived from a family of airfoilscontaining 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
differentstory 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 Functions0.2217
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 for4 of 8
American Institute of Aeronautics and Astronautics
Page 5transonic 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 shownin 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 ofperformancein other conditions. This method can be used to evaluate differentparameterisations before
carrying out optimisations using the high fidelity codes.
V. ConclusionsTwo 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-
-
8/11/2019 Aerodynamics Glossary
16/43
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.
AcknowledgementThe 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.
References1Raymer, D.P., Aircraft Design: A Conceptual Approach, AIAA Educational Series 1999.2
Cosentino, G. B. and Holst, T. L., Numerical Optimisa tion 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 Airfoilsfor Aero-
dynamic Optimization, AIAA 2003-4095, 16th AIAA Computational Fluid Dynamics Conference, June 23-26, Orlando, FL,
20037
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 Airfoilsand 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 6X/C
Y
/
C0
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
0.06RAE2822Orthfoil-Approximation
(a) Approximation of geometry to RAE2822 by basis functions
approachX/C
P
r
e
s
s
ur
e
C
o
e
f
f
i
c
i
e
n
t
s0
0.25
0.5
0.75
1
-1
-
8/11/2019 Aerodynamics Glossary
17/43
-0.5
0
0.5
1RAE2822
Orthfoil-approximation
(b) Comparison of pressure distributions for RAE2822 using
basis functionsX/C
Y
/
C
0
0.25
0.50.75
1
-0.03
-0.02
-0.01
0
0.01
0.02
0.03NACA0406
Orthfoil-Approximation
(c) Approximation of geometry to NACA0406 by basis func-
tions approachX/C
P
r
e
s
s
u
r
e
C
o
e
f
f
i
ci
e
n
t
s0
0.25
0.5
0.75
1
-1
-0.5
0
0.5
1NAC0406Orthfoil-approximation
(d) Comparison of pressure distributions for NACA0406 using
basis functionsX/C
Y
/
C0
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
0.05NACA0610Orthfoil-Approximation
(e) Approximation of geometry to NACA0610 by basis func-
tions approachX/C
P
r
e
s
s
u
r
e
C
o
e
f
f
ic
i
e
n
t
s0
0.25
0.5
0.75
1
-1
-0.5
0
0.5
1NACA0610
Orthfoil-approximation
(f) Comparison of pressure distributions for NACA0610 using
basis functions
-
8/11/2019 Aerodynamics Glossary
18/43
-
8/11/2019 Aerodynamics Glossary
19/43
0
0.5
1NAC0406
B-Spline Interpolation
(d) Comparison of pressure distributions for NACA0406 using
B-spline interpolationX/C
Y
/
C0
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
0.05NACA0610B-Spline Interpolation
(e) Approximation of geometry to NACA0610 by B-spline in-
terpolationX/C
P
r
e
s
s
u
r
e
C
o
e
f
f
i
c
i
e
n
t
s0
0.25
0.5
0.75
1
-1
-0.5
0
0.5
1NACA0610
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 812
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$4.pdf.Googleautomatically generates html versions of documents as we crawl the we.To lin! to or oo!mar! this page" use the following url: http://www.google.com/search?q=cache:w!-comla.sta!ford.ed(/Papers/,yoto0$.pdf&'o(r!als&)performa!ce&of&differe!t&airfoi
'oogle is neither affiliated with the authors of this page nor responsi%le for its content.
http://aero-comlab.stanford.edu/Papers/kyoto94.pdfhttp://aero-comlab.stanford.edu/Papers/kyoto94.pdf -
8/11/2019 Aerodynamics Glossary
20/43
These search terms have een highlighted: performance different airfoils cfdThese terms only appear in lin!s pointing to this page:journals
Page 1
A Two-Dimensional Multigrid-Driven Navier-Stokes
Solver for Multiprocessor ArchitecturesJuan 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 INTRODUCTIONIn 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
CFDtools.
Increasingly complex fluid flow models require high performancecomputing 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 performanceof 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 DISCRETIZATIONThe 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
=
-
8/11/2019 Aerodynamics Glossary
21/43
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 2be applied to each finite volume ofa computational domain to yield a set ofcoupled first-order differential
equations ofthe form
d
dt
(wij
Vij
) + E(wij
) + NS(wij
) + D(wij
) = 0
,
(2)
where E(wij
) are the convective Euler fluxes, NS(wij
) 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 STRATEGYFLO103P 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 ni
nj
= 102464. This
domain was decomposed into subdomains containingni
Np
nj
points, whereNp
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,
-
8/11/2019 Aerodynamics Glossary
22/43
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 performancein 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 MULTIGRIDThe 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 levelin the series. This usually led to a decrease in the convergence rate ofthe numerical algorithm, presenting a
clear tradeoff between the improved parallel performanceand the increasing number of cycles required for a
similar level ofconvergence. In this work, we examine three differentapproaches 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-CYCLEParallelT
T
T
T
T
T
T
Fine
Coarse
Single Proc.
"LAZY" PROCESSOR W-CYCLEParallel
T
T
T
T
T
T
T
Fine
Coarse
No work
Figure 1: Lazy-Dumb Multigrid Approach.
4.1 Full multigrid approximationThis 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, boundaryinformation 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 subdomainsThe 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
-
8/11/2019 Aerodynamics Glossary
23/43
increased communication between differentparts 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 approachIn 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 SMOOTHINGImplicit 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 performancein a multiprocessor environment.
5.1 Fully implicit residual smoothingSerial 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 differentprocessors. 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 performanceto drop-off considerably.
5.2 Implicit residual smoothing within subdomainsIn 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 performanceimproves, 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 smoothingAn 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 stabilityproblems. The matrix problem is setup in each subdomain and a Jacobi relaxation procedure is performed
on all subdomains concurrently.
6 RESULTSA summary of the results for the differentmultigrid approaches is presented in Figure 2. The full multigrid
approach produces results with an optimum convergence rate, but with a parallel performancethat degrades
for a large number of processors. One must take into account, that for these meshes ( ni
nj
-
8/11/2019 Aerodynamics Glossary
24/43
= 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, parallelperformancedegrades 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 5considerably 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 performanceof 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 willachieve 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 performanceresults of the differentmethods 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 performanceof 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 differentpoints on the oscillation period of
a NACA 64A010 airfoil at a Reynolds number of 106
,M
= 0.796, and a reduced frequency, kc
= 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 CONCLUSIONSA 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, andthe performanceofeach 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
-
8/11/2019 Aerodynamics Glossary
25/43
pressure field around a pitching NACA 64A010 atM
= 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 Airfoilsand 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 61
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 5129x6 5
129x6 5129x6 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
-
8/11/2019 Aerodynamics Glossary
26/43
5
6Figure 5: Mach Number Contours. Pitching Airfoil Case.Re = 1.0 106
,M
= 0.796, Kc
= 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.Googleautomatically 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:h1)performa!ce&of&differe!t&airfoils&i!&cfd*+hl=e!+gl=i!+ct=cl!,+cd=$7
'oogle is neither affiliate
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 Aerodynamics8.1 Introduction: Why high lift?
For transonic transports, the high-lift system design is a critical part of the configuration
design.
To achieve reasonable field performancewhile also obtaining efficient transonic
cruise the
design will require a fairly sophisticated high lift system. From a paper by Boeing
aerodynamicists1
: (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-offL/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.
http://www.aoe.vt.edu/~mason/Mason_f/ConfigAeroHiLift.pdfhttp://www.aoe.vt.edu/~mason/Mason_f/ConfigAeroHiLift.pdf -
8/11/2019 Aerodynamics Glossary
27/43
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 isgoing 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 CLmax
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 CLmax
for some airplanes.*
* Note that there is a significant variation of values from differentsources.
Clearly the 727 emphasized short fields, and thus required a higher CLmax
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
737-2003.2
747/E-4A
2.45
767
2.45
777
2.5
-
8/11/2019 Aerodynamics Glossary
28/43
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 MSES5
code). A recent summary of the computational capability is by
Rumsey and Ying.6
Single elementairfoils
The key example of how to obtain high lift on a single element airfoil is the story of
Liebecks high lift airfoil7
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-elementairfoils:
Understanding the physics: Section 6.3 of A.M.O. Smiths paper8
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 vaneeffect.
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
-
8/11/2019 Aerodynamics Glossary
29/43
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 al12
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 McCormick17
(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 3High-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 CLmax
increase is simply due to the use of the original reference area
in computing the CL
. (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
-
8/11/2019 Aerodynamics Glossary
30/43
-
8/11/2019 Aerodynamics Glossary
31/43
may even decrease.19
Considering the adverse effect of Mach number on CLmax
, 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 CLmax
.
(from Dick Kitas Grumman talk, Feb. 1985)
Differentleading edge devices differ in their effectiveness. This is Kitas estimate of
how each
type of device affects the performanceof the wing.
Page 12
8-12 W. H. Mason, Configuration Aerodynamics
Figure 8-9. Estimated performanceof various types of high lift systems.
(from Dick Kitas Grumman talk, Feb. 1985)
This is Kitas estimate of the typical best performanceyou 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
performanceof
any system. This may or may not be true.
Page 13
High-Lift Aerodynamics 8-13
-
8/11/2019 Aerodynamics Glossary
32/43
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
optimumflap 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 systems8
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
-
8/11/2019 Aerodynamics Glossary
33/43
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
stronglythat 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, Cp= 0 represents the start of the pressure rise and Cp
= +1
the maximum possible value, that is, ue
= 0.The velocity at the start of the pressure rise is u0
.
Thus, the canonical pressure distribution is defined as
Cp
= 1ue
u0
2
The next step is to examine the best way to specify the pressure distribution to allow thepressure to recover to as close as possible to Cp
= +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
-
8/11/2019 Aerodynamics Glossary
34/43
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. Todo this, we manipulate the Stratford criteria:
Cp
x
dCp
dx
106
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 ofseparation, 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.
-
8/11/2019 Aerodynamics Glossary
35/43
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
boundarylayers. Note also that the Reynolds number effect is relatively weak. Finally, the
boundary layer
could recover all the way to Cp
= +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 17High-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 - feetsolid line U0
/v = 10^6
dashed line U0
/v = 10^7
Figure 8-13. Stratford Limiting flows for two differentReynolds numbers.
Single elementairfoils: The key is how to obtain high lift on a single element airfoil, and
isessentially the story of Liebecks high lift airfoil7
and the Stratford pressure recovery shape of
the pressure distribution described above, as told by Smith.8
The question of how much lift you
-
8/11/2019 Aerodynamics Glossary
36/43
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,
heexplains the limit of the pressure coefficient in terms of the vacuum Cp
when a zero pressure is
specified on the airfoil upper surface. Thus, using the definition of Cp
,
Cp
=
p p
1
2
U
2
we can obtain an alternate form using q = 1
2
U
2
=
2
P
M
2
, that is
Cp
=
p p
2
p
M
2
-
8/11/2019 Aerodynamics Glossary
37/43
Page 18
8-18 W. H. Mason, Configuration Aerodynamics
and vacuum Cp
occurs when the pressure is zero:
Cp
vac
=
2
M
2
Then, he points out that only a value of 70% of the vacuum Cp
has been achieved in practice.
This results in a Cp
limit ofM
2
Cp
= 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-elementairfoils: Understanding the physics: read section 6.3 of A.M.O. Smithspaper.
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 CLmax
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 hasto 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
-
8/11/2019 Aerodynamics Glossary
38/43
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 UTE
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 liftSignificant 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
-
8/11/2019 Aerodynamics Glossary
39/43
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 theblowing coefficient, defined as:
C
=
mj
Vj
qc
where the subscriptj refers to the jet, and q is the dynamic pressure. Blowing for BLC
was usedoften 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
fromthe 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 Englar22
, A-6 CCW)8-9 Configuration Integration issues
The best airplane CLmax
you can achieve with a mechanical high lift system is about 3 3.5
The military and civil air regulations require a margin between CLmax
and the operating CL
-
8/11/2019 Aerodynamics Glossary
40/43
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
CL
is only 59% of the CLmax
. However, some relief is available because the measured stallspeed, Vsmin
is usually about 0.94 times the stall speed in 1g steady flight (the wind tunnel
case). This means that you can use a CL
of about 67% of the CLmax
.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 CLmax
values, the real 3D value will be
much less.
The maximum CL
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 Study24
is highly recommended.
Finally, issues not covered but worth mentioning: the concept of the Gurney flap7
and the need
for accuracy around the leading edge.
8-10 Exercises
1. Examine the predictive capability of XFOIL for CLmax
. Use the data from Abbott and von
Doenhoff supplied previously for the NACA 0012 and 4412 airfoilsat 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
-
8/11/2019 Aerodynamics Glossary
41/43
special attention to the details of single and multielement airfoilsdescribed in Sections 3-
6.
8-11 References1
Garner, P.L., Meredith, P.T., and Stoner, R.C., Areas for Future CFDDevelopment as
Illustrated by Transport Aircraft Applications, AIAA Paper 91-1527, 1991.2G.W. Brune and J.H. McMasters, Computational Aerodynamics Applied to High Lift
Systems, Chapter 10 ofApplied 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 andPerformance,
DARcorporation,
Kansas, 1997, pp. 343.4
Karl L. Sanders, High-Lift Devices, A Weight and PerformanceTrade-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 Airfoilsfor High Lift, Journal of Aircraft, Vol.
15,
No. 9, Sept. 1978, pp. 547-561.8
A.M.O. Smith, High-Lift Aerodynamics, 37th
Wright Bothers Lecture,Journal of Aircraft,
Vol. 12, No. 6, June 19759
C.P. van Dam, The aerodynamic design of multi-element high-lift systems for transportairplanes, 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
-
8/11/2019 Aerodynamics Glossary
42/43
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 8-2111
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, 2nd
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
-
8/11/2019 Aerodynamics Glossary
43/43
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 Tes