Você está na página 1de 212

Origin of Fluid Forces

We are often particularly interested in the forces and moments applied to bodies moving through the fluid. These can be divided into just two types: pressures and shears.

Pressures are created at the surface of a body due to (nearly) elastic collisions between molecules of the fluid and the surface of the body. Shearing forces are produced by fluid viscosity. This quantity is a measure of how well momentum is transferred between adjacent layers of the fluid. Although both types of forces are important in applied aerodynamics, pressures are usually the dominant type of force. Each of the shapes below, drawn to scale, have the same drag. The reason that the streamlined airfoil can be so much larger is that most of the force is due to shear stress, not pressure forces.

The cylinder, with its separated airflow, has large pressure forces that give rise to high drag. In fact, it is quite amazing how much force can be generated by differences in pressure: Many airliners have wing loadings (weight / wing area) of over 100 lbs/ ft2 (4.8 KPa). This means that it takes a section of wing only as large as a book to lift a large dog (for example). This is possible because the normal atmospheric pressure is 2116 lb/ft2 (or 101 KPa) at sea level. So, in fact 100 psf (4.8 KPa) represents only a 5% change in the pressure on the upper side of the wing.

At 68,000 ft (20.7 km) you would have to create a complete vacuum on the wing upper surface to lift that much weight.

The Origin Of Pressure Forces

Pressure arises because each molecule that bounces off the surface transfers momentum to the body. If a particle of mass, m, hits the body "straight-on" and bounces off, it transfers momentum of the amount 2mc where c is the speed of the molecule. The pressure is then proportional to the number of molecules striking a unit area of the surface per unit time, (Number density*c), times the momentum transfer per particle, (~mc) or: p = k1 c2

Note that since temperature is defined as proportional to the mean kinetic energy of the molecules, T = k2 c2. So we expect: p = k T, the perfect gas relation. The component of molecular velocity normal to the surface is what is really needed in the above expression, and if the body is moving, we must add its velocity to the molecular velocity measured in the "fluid-fixed" reference frame. Typically, we do not consider these direct interactions, but rather model the molecules as a continuous fluid. This works well for most flows of interest. However, for very rarefied flows such as those associated with initial re-entry of space vehicles, it is sometimes possible to analyze aerodynamics with kinetic theory, keeping track of the molecular interactions. The figure

below shows the results of one such calculation.

The Origin Of Shear Forces

As molecules in adjacent layers with different average velocities collide, they transfer momentum between the layers. The rate of change of momentum produces a shear stress in the fluid. At the surface of a body, molecules transfer momentum to the surface as they collide, resulting in a tangential, shear force. When molecules hit the surface of a body, they bounce around among the surface molecules and finally leave with a tangential velocity which is, on average, that of the surface itself. Thus, the average tangential velocity near the surface of a body is zero with respect to the body. This is the so-called no-slip condition.

This layer of slow moving fluid near the body surface is called the boundary layer, and the viscosity of the fluid causes a distribution of tangential velocity above the surface as shown here. As the tangential momentum of the air molecules is transferred to the surface, a shear stress is produced. This shear stress is related to the viscosity and velocity gradient by the expression: = dU/dy

We can see more quantitatively how the transfer of tangential momentum between fluid layers leads to this relation by considering a small section in the boundary layer.

Molecules starting near the top of the box and moving to the bottom, lose momentum in the amount: m h dU/dy Since the shear stress, , is the rate of change of momentum: = n m h dU/dy where n is the number of molecules passing through this area per unit time. Now n is related to the average molecular velocity, c, and the density, so: mn=c and the shear stress is: = c h dU/dy so that: = c h The height, h, over which molecules transfer their momentum is related to the mean free path, , with more detailed calculations showing that: = 0.49 c Now the mean free path decreases almost in proportion to density and the average molecular speed varies with T, so we expect that varies with T and does not depend on pressure. This is approximately true for most fluids.

Dimensionless Groups
The forces on a body, moving through a fluid, depend on the body velocity (V), the fluid density (), temperature, and viscosity (), the size of the body (l), and its shape. Using the speed of sound, a, rather than temperature (they are directly related) we can then make the following table that shows the units associated with each of the parameters. Here the numbers indicate the power to which the mass, length, or time units are raised:

F mass length time

= f 1 1

(V, , 0 1 1

a, 0

, l, shape) 1 0 0

-3 1 0

-1 1 0

-2 -1

-1 -1 0 0

The Buckingham pi theorem states that the number of dimensionless parameters is equal to the number of parameters minus the rank of the above matrix. In this case 7 - 3 = 4. So, there exists a functional relationship among the four dimensionless groups. We can express the force on a body, for instance, by a relationship between the following four dimensionless parameters: F / ( V2 l2) Vl/ V/a shape

Dimensionless Force Coefficient Reynolds Number Mach Number Geometry The relation is: F / ( V2 l2) = f (Vl/, V/a, shape) We will discuss each of these dimensionless groups in a moment, but let's first look at the functional relationship between them. Much of applied aerodynamics involves finding the function f, but there is a great deal we can say, even without knowing it. For example, we can see that a wide variety of similar flows exist. The forces on a large, slow-moving body could be predicted from tests of a small higher-speed model as long as the speed of sound were sufficiently high. Also, the flow around a small insect could be represented by a large model in a very viscous fluid. The idea behind model testing is to simulate the flow over one body by matching the dimensionless parameters of another.

This is not always easy -- or possible. The following figure, from J. McMasters of Boeing shows the Mach and Reynolds number range of several wind tunnels. Why can't wind tunnels be designed to more fully cover this range of parameters? What alternatives exist to wind tunnel tests? (See assignments.) Subsequent pages consider each of the dimensionless groups in a bit more detail. First note that we could have included other fluid properties such as specific heats. This would lead to additional dimensionless parameters such as the Prandtl number which is important in the study of compressible boundary layers with heat conduction. We have also left out gravity which is often important in the flow of water around ships. This would lead to an additional dimensionless parameter called the Froude number. There are often several ways of combining the parameters to form dimensionless groups, but these are commonly used in aerodynamics.

Dimensionless Forces
We use dimensionless force and moment coefficients defined by: L = 0.5 V2 S CL(Re, M, shape) M = 0.5 V2 S c Cm(Re, M, shape) CL is called the lift coefficient, Cm the moment coefficient. The length2 term in our first dimensionless parameter has been replaced by the area, S. This area could be anything we choose (the contact area of the nose wheel, the wing planform area, the fuselage cross-sectional area). In a particular application people generally agree on a reference area. For car drag coefficients the frontal area is often used. For aircraft the wing area is a common reference area. The c (for chord) in the moment coefficient definition is similarly agreed upon. This "agreement" on reference area is very important as can be seen in advertisements for cars. (Automobile drag coefficients are usually based on frontal area and numbers like 0.4 are sometimes mentioned in car ads. But, the drag coefficient means nothing by itself. If we chose the reference area to be the floor area of the Fremont GM plant, we would have very low drag coefficients.)

Reynolds Number
The quantity: V l / is called the Reynolds number. is the fluid density, V is the speed, is the fluid viscosity, and l is some characteristic length. This length is, like the areas in the definition of dimensionless force coefficients, agreed on as a standard by whoever is using it. So, we have chord Reynolds numbers which are based on wing chord lengths or Reynolds numbers based on the diameter of a sphere, or any other characteristic length that can be devised. The Reynolds number is one of the most important and strange dimensionless numbers. It varies over many orders of magnitude and expresses the importance of viscosity: high Reynolds numbers can be achieved by decreasing the viscosity or making the length or speed very large. The Reynolds number, in a sense, represents a ratio of pressure to shear forces: V l / = V2 / (V/l) V2 is related to the pressure while V/l is related to dU/dy, the shear stress.

The range of Reynolds number -- from McMasters.

Viscosity, and hence Reynolds number, strongly affects the performance of wings and airfoils, making it an important parameter to match in wind tunnel tests. It is often not possible to match these dimensionless parameters precisely. The plot below shows the effect of Reynolds number on maximum lift to drag ratio for two dimensional airfoil sections. Note the plight of insects.

The plot here shows the effect of Reynolds number on the maximum section lift coefficient of a few typical airfoil sections. Note that these are not necessarily the best sections for high lift, though. Recent studies have shown that substantial changes in CLmax are seen even at quite high Reynolds numbers, making it difficult to extrapolate data on small wind tunnel models.

Mach Number
The Mach number is the ratio of flow speed, V, to the speed of sound, a. It reflects the importance of the compressibility of the fluid. This ratio is important because pressure disturbances propagate in a fluid at the local speed of sound and the compressibility of a fluid permits a sound wave to travel. The speed of sound in a fluid is related to the way in which density and pressure vary: a2 = dp/d Assuming isentropic flow and a perfect gas: a2 = R T The flow pattern and pressures can change dramatically with Mach number as the applicable differential equation changes form. (See later sections.) At low subsonic speeds, the effect of compressibility is not large and the flow behaves almost as if pressure disturbances traveled with infinite speed. As the flow speed is increased, but remains subsonic the effects of compressibility start to appear slowly. As the flow velocity approaches Mach 1 (transonic flow) more significant compressibility effects appear quickly. Because of the increase in local velocity over parts of an airfoil, the local Mach number can be much higher than the freestream Mach number. In fact, compressibility effects can be important for high-lift sections at freestream Mach numbers as low as 0.3. As the flow velocity increases beyond Mach 1.0, it becomes supersonic and its characteristics change greatly. Very high velocity flows (usually above Mach 5 or 6) are called hypersonic. These types of flow are of great importance in the aerodynamics of rockets and reentry vehicles, which achieve Mach numbers as high as 25.

Conservation Laws
To derive the equations of motion for fluid particles we rely on various conservation principles. These principles are entirely intuitive. They are a statement of the fact that the rate of change of mass, momentum, or energy in a certain volume is equal to the rate at which it enters the borders

of the volume plus the rate at which it is created inside. The first two of these will be used extensively here.

These integral expressions are combined with the divergence theorem and the fact that they hold over arbitrary volumes to obtain the differential form of the equations:

We can use the momentum theorem by itself to obtain useful results. In this example, we apply the momentum theorem to relate the force on a body to the properties of the flow some distance from the body. This technique is useful in wind tunnel tests and is the basis of several fundamental theorems related to lift and induced drag of wings. We take the control volume shown below, bounded by the single surface, S which we divide into 3 parts: the outer surface (Souter), the inner surface (Sinner), and the pieces of the surface connecting the two (S*). We can write the integral form of the momentum equation for steady flow with no body forces as shown.

Note that the contribution from the part of the surface connecting Sinner and Souter to the integrals is zero because as the two pieces of S* are made close together, the unit normals point in opposite directions while p and V are equal.

Simplifying Approximations
The equations of motion for a general fluid are extremely complex and even if the problem could be formulated it would be impractical to solve. Thus, from the outset, certain simplifying approximations that are often very accurate, are made. These may include the following assumptions. Continuity and Homogeneity

We assume that the fluid is composed of particles which are so small and plentiful that the statistically-averaged properties of interest are the same at any scale. This works well for gases and fluids under most conditions. It does not work for studying the flow of sand. It does not work when the fluid is so rarefied that the mean free path is of the same order as the dimensions of interest in the problem. The mean free path varies with altitude as shown in the plot. We further assume that the medium can be treated as a single type of fluid -- no suspensions of oil and



The effect of viscosity may sometimes be neglected or modeled indirectly. For many aerodynamic flows of interest, the region of high shear and vorticity is confined to a thin layer of fluid. Outside this layer, the fluid behaves as if it were inviscid. Thus the simpler equations of an inviscid fluid are often solved outside of the shear layers. There are some fluids which seem to be almost completely inviscid. Tests in superfluid helium have given results similar to inviscid calculations.

Incompressible (constant density)

When the fluid density does not change with changes in pressure, the fluid is incompressible. Water density changes very little with changes in pressure and is generally treated as an incompressible fluid. Air is compressible, but if pressure changes are small in comparison with some nominal value, the corresponding changes in density are small also and incompressible equations work quite well in describing the flow. The degree to which the fluid density changes with pressure is related to the speed of sound in the fluid. Thus, assuming that the flow is incompressible is equivalent to assuming that the speed of sound is infinite. When the local Mach number is less than 0.2 to 0.5 compressibility effects can often be ignored. The reason for this is discussed further in the chapter on compressibility, but one can see qualitatively that in order to make an appreciable change to the nominal 2116 lb/ft^2 air pressure at sea level, substantial speeds are required.


Circulation is defined as: It is a measure of the rotation of an area of fluid. As the integration contour is shrunk down to a point, the ratio of circulation to the area enclosed by the curve is called the vorticity.

Fluid that starts out without rotational motion will not develop it unless there has been some shear stress acting on it*. And if the shear is confined to a small region, the vorticity will be also. Thus, for many cases, especially in inviscid flow, much of the flow field may be treated as irrotational: curl V = 0 When this is the case, the vector field, V, may be written as the gradient of a scalar field, : V = grad where is the called the potential. This simplifies many of the equations discussed in subsequent sections. The velocity components are then: u = d / dx and v = d / dy

When the variables describing the fluid properties at a given point do not change in time, the flow may be treated as steady and the time derivatives in the equations of motion are zero. This condition depends on the chosen coordinate system. If the system is at rest with respect to a body in uniform motion through a fluid the equations in that system are steady, but expressed in a system fixed with respect to the undisturbed fluid, the flow is unsteady. It is often convenient to transform the coordinate system to one in which the flow is steady. This is, of course, not always possible. We will assume that the flow is steady in most of the discussions in this course but unsteady effects are often important in the study of bird flight, propellers, aircraft gust response, dynamics, and aeroelasticity as well as in the study of turbulence. We will always apply the first of these assumptions and will sometimes adopt one or more of the latter in the following discussions.

*Some important exceptions to the idea that without viscosity irrotational flow remains irrotational: Vorticity can be created in a gravitational field when density gradients exist or in a rotating

system (such as the earth) due to Coriolus forces. These are important sources of vorticity in meteorology.

Equations of Fluid Flow

The conservation laws may be used to derive the equations of fluid flow. These are supplemented with constitutive relations such as the perfect gas law: p=RT or the isentropic relation between pressure and density: p2 / p1 = (2 / 1) Some of the most commonly-solved equations are shown in the following table along with the corresponding assumptions. Equation Navier-Stokes ReynoldsAveraged Navier-Stokes Euler Full Potential Transonic Small Disturbance Prandtl-Glauert Acoustic Laplace Inviscid Irrotational X X X X X X X X X X X X X X Small Perturbations Incompressible X Linearized Linearized Notes Homogeneous Modeled Turbulence

Navier-Stokes Equations

The Navier-Stokes equations describe the flow of a continuous, Newtonian fluid. They may be derived from the principal of conservation of momentum. (For more details see note or Kuethe and Chow Appendix B, Moran Ch. 6, Anderson Ch. 15).

where X, Y, Z are the body forces per unit mass in each direction and t is the stress tensor. X,Y, and Z are often associated with gravitational forces and are often neglected. The equations become more usable when the stress tensor is expressed in terms of viscosity and pressure. The pressure and shear forces may be expanded so that the NS equations are:

is the "bulk viscosity" , relating the normal stress to the rate of change of volume, div(V). If pressure is a function only of density and not of the rate of change of density, then: = - 2/3 In the simplest case, with no body force, these equations become:

Solutions of the full Navier-Stokes equations show the onset of turbulence, the interaction of shear layers, and almost all of the interesting aerodynamic phenomena (with the exception of interacting or rarefied gas flows). Unfortunately, the equations are very difficult to solve. As the Reynolds number is increased, the scale of the interesting dynamics gets smaller so that most solutions of the full NS equations are done at Reynolds numbers of 1 to 1000. One of the most recent solutions of a flat plate boundary layer pushed the calculations to a Reynolds number of 1410 based on boundary layer thickness.

These calculations took hundreds of hours on the Cray computer at NASA Ames. The video shows the results of a direct simulation of turbulence with color-coded vorticity contours, from S. Robinson, NASA. Even at very small Reynolds numbers, the geometries which can be analyzed using the full NS equations are quite simple and it currently does not make sense to consider solving these equations for realistic aircraft configurations. One reason that this is the case is that many of the approximate equations work quite well in such cases and are much more easily solved. When the time averaged Navier-Stokes equations are not a sufficient description of the problem, one may resort to "large eddy simulations". This is a numerical solution of the time-dependent Navier-Stokes equations, with only the smaller scales of turbulence modeled in an averaged way. Larger scale turbulent motion can be included in this way. While this is faster than solving the full equations, it is still very slow. The figure below shows results from a large eddy simulation of the flow over a 2D circular cylinder. Each simulation required approximately 300 CPU hours and about 10 megawords of core memory on the Cray C-90. Figure from NASA / Parviz Moin.

Navier-Stokes Derivation

This is the basic idea. We look at the momentum equation in the x direction: where Fx represents a general force on a fluid element that might include viscous forces or gravity. Noting that:

conservation of mass permits us to substitute:


The momentum equation in the x-direction may then be written:



and canceling terms leads to:

Reynolds Averaged Navier-Stokes Equations

One of the most popular simplifications made to the Navier-Stokes Equations is "Reynolds Averaging". This simplification to the full Navier-Stokes equations involves taking time averages of the velocity terms in the equations. Writing: u = <u> + u ', v = <v> + v', etc. (where <> represents a time average) with the fluctuations having zero mean value: <u'> = 0

we have: <u2> = <u>2 + <u'2>, <uv> = <u><v> + <u'v'> This allows us to write the time-averaged NS equations as: and similarly for the y and z components. This looks just like the more general Navier Stokes equations for incompressible flow* which hold for steady, laminar flow except that there are additional terms that act as additional stresses on the right hand side. These terms represent the effect of turbulence on the mean flow. They are called "Reynolds stresses" and are sometimes said to be caused by "eddy viscosity". These terms are generally much larger than the normal viscous terms. The business of predicting these stresses and relating them to the computed mean flow properties is called turbulence modeling. This is usually accomplished empirically or by using the results of detailed time-dependent simulations. Reynolds averaged NS solvers are appropriate for the analysis of viscous, compressible flows and have been applied to rather general configurations, but one must be careful that the assumptions of the turbulence model are compatible with the characteristics of the flow of interest.

From NAS Technical Summaries, High-Lift Configuration CFD, Karlin Roth, NASA Ames Research Center

Euler Equations
The momentum equation is sometimes called Euler's equation. (There are lots of equations called Euler equations!) But when people talk of solving the Euler equations these days, they are referring to the inviscid equations of motion given by:

With some work*, the equation in the x direction becomes:

or in vector notation:

These are combined with the equations of energy and continuity. The equations are often solved by finite differences whereby the values of each velocity component, the density, and the internal energy are computed at each point in the flow. From these quantities constitutive relations such as the perfect gas law or the isentropic pressure relation are used to find pressure. Since Euler equations permit rotational flow and enthalpy losses (through shock waves), they are very useful in solving transonic flow problems, propeller

or rotor aerodynamics, and flows with vortical structures in the field.

* Recall DF/Dt, the substantial, or particle derivative of F is defined by: DF/Dt = dF/dt + V grad F Also see the note in the derivation of the NS equations. It looks as though we have assumed constant density, but this is not the case.

Full Potential Equation

The full potential equation is derived from the assumption of irrotational flow and the equations of continuity and momentum. The pressure and density terms in the Euler equations can be combined when use is made of the perfect gas law and the isentropic relation between pressure and density. Ashley and Landahl show how we may derive the following vector form of the unsteady full potential equation:

This may be simplified for the case of steady flow in 2-D to:

About the notation: When flow is irrotational curl V = 0 and by definition of curl and gradient: curl (grad ) = 0 where is a scalar field. Thus we can define a nonphysical scalar potential, , that describes the velocity field. is related to the velocities by the relation: V = grad The equations can thus be written in terms of the unknown scalar rather than the 3 components of the velocity. This simplifies their solution. In the above expressions: a is the local speed of sound, x is the streamwise coordinate, and V is the vector velocity. Subscripts denote partial derivatives with respect to the subscripted variables (e.g. Ux = du/dx)

Derivation of the Potential Equation

The derivation of the full potential equation is easily seen in the case of 2-D steady flow. In this case, the continuity equation is:

Also, for steady, inviscid flow, Euler's relation between p and is: dp = - V dV = - d(u2 + v2)/2 If the flow is also isentropic, then another relation between p and is: dp = a2 d

Combining the last two expressions:

We can thus write: Then, substituting into the continuity equation, we obtain:

Note that the local speed of sound may be written in terms of some constants and the local velocities:

Transonic Small Disturbance Equation

When the full potential equation is simplified by assuming that perturbation velocities are small and we relate the local speed of sound to the freestream value by making use of the isentropic relations we obtain the small disturbance equation (derivation):

When we let the freestream Mach number go to one and ignore the last term, the equation becomes the classic transonic small disturbance equation:

A great deal has been written about this nonlinear equation and its variants. (See Nixon.) It is used less frequently these days since finite difference methods can be used to solve the full potential equation directly.

TSD Derivation
We begin with the 2-D full potential equation:

Ignoring terms of second order in the perturbation velocities and with:

The local speed of sound, a, may be related to the local velocity by isentropic relations. After some algebra, and again dropping terms of second order in the perturbations:

Substituting: The final term is sometimes neglected.

Prandtl-Glauert Equation
The Prandtl-Glauert equation is a linearized form of the full potential equation. Full potential:

If the velocity perturbations are much smaller than the freestream velocity, this expression becomes:

or in the unsteady case:

The 3-D version is easily constructed with the addition of z derivatives corresponding to the y derivatives shown here.

Note that this linearized form of the equation does not hold near the nose of an airfoil where the velocity perturbation is of the same order as the freestream, unless the freestream Mach number is itself small. Also note that this expression holds for subsonic and supersonic flow (but not transonic flow). It forms the basis for many aerodynamic analysis methods.

Analysis of P51 Mustang from Analytical Methods, Inc. using VSAERO, a code that solves the Prandtl-Glauert equations.

Acoustic Equation

The acoustic equation may be obtained from the full potential equation by assuming that there is no freestream velocity, and that all perturbation velocities are small. Or, by changing from a coordinate system fixed to the body to one fixed with respect to the undisturbed fluid, the Prandtl-Glauert equation may be transformed to the acoustic equation. This equation is often used in the study of sound propagation and sometimes for rotor aerodynamics; thus the name.

Laplace's Equation

Laplace's equation is the Prandtl-Glauert equation in the limit as the freestream Mach number goes to zero. It was actually first derived by Euler. The derivation is very simple, requiring only the equation of continuity, and the assumptions of irrotational and constant density flow. The continuity equation becomes then:

Since the flow is irrotational: Substitution into the continuity equation yields: It is interesting to note that Laplace's equation does not require the assumption of small perturbations, while the Prandtl-Glauert equation does. In fact, near the stagnation point of an airfoil where velocities become small, the full potential equation reduces to Laplace's equation, not the Prandtl-Glauert equation. Note also that all of the time dependent terms in the full potential equation are multiplied by 1/a2 so that this form of the equation holds for unsteady phenomena as well.

Bernoulli Equations
The Equations

Some of the equations we have discussed are posed in terms of state variables that do not include

pressures. In these cases (e.g. the potential flow equations) the differential equations and boundary conditions allow one to compute the local velocities, but not the pressures. Once the velocities are known, however, the momentum equation can be used to find the local pressure. Such equations are known as Bernoulli equations and they come in various forms, depending on the assumptions that can be made about the flow. The conservation of momentum principle is the source of the relation between pressure and velocity. It can be used very simply to derive the Bernoulli equation. To illustrate the basic physics behind the Bernoulli equations, we can derive a simple form: that for steady, incompressible flow. In this case we show that along a streamline:

When the flow is not steady, the Euler equations can be integrated to obtain a more general form of this result: Kelvin's equation, the Bernoulli equation for irrotational flow.

Where f is a body force per unit mass (such as gravity) and F is an arbitrary function of time. If we do not assume that the flow is irrotational, we cannot introduce the potential and the expression is not so nicely integrable. If, however, we assume that the flow is steady with no "body forces", but not necessarily irrotational we can write the following expression that holds along a streamline:

While the above equations hold for steady flows along a streamline, for irrotational flows they hold throughout the fluid.

We can derive a more useful form of the Bernoulli equation by starting with the expression for steady flow without body forces shown just above. If the flow is assumed to be isentropic flow (no entropy change or heat addition): p = constant * Substitution yields the compressible Bernoulli equation:

This actually works for adiabatic (no heat transfer) flows as well as isentropic flows. In summary, we often deal with one of two simple forms of the Bernoulli equation shown below.

The Pressures

In both the incompressible and compressible forms of Bernoulli's equation shown above there are 3 terms. The quantity pT is the total or stagnation pressure. It is the pressure that would be measured at points in the flow where V = 0. The other p in the above expressions is the static pressure. Note that in incompressible flow, the speed is directly related to the difference in total and static pressure. This can be measured directly with a pitot-static probe shown below.

The dynamic pressure is defined as:

The static pressure coefficient is defined as: where p is the freestream static pressure. In incompressible flow, the expression for Cp is especially simple:

If the local velocity is expressed as a small perturbation in the freestream: Then the incompressible Cp relation can be written: Be careful with this expression! It is often not a good approximation and the correct expression is not very difficult.

The expression for Cp in compressible isentropic flow (sometimes called the isentropic pressure rule) is derived from the compressible Bernoulli equation along with the expression for the speed of sound in a perfect gas. In terms of the local Mach number the expression is:

In air with gamma = 1.4

Some interesting results follow from this expression... We can tell if the flow is supersonic, just by looking at the value of Cp. The critical value of Cp, denoted Cp* is found by setting M = 1 in the above expression:

Also, we see that there is a minimum value of Cp, corresponding to a complete vacuum. Setting the local Mach number to infinity yields:

Cp cannot be any more negative than this. Experiments show that airfoils can get to about 70% of vacuum Cp. This can limit the maximum lift of supersonic wings.

Solution Methods
This chapter is a brief overview of methods used to investigate fluid flows. It includes a bit of discussion on the role of experimental, analytical, and computational methods and outlines the basic ideas behind the computational approaches. 1. The Role of Theory and Experiment 2. Analytic Methods 3. CFD Overview

4. Panel Methods 5. Nonlinear CFD 6. References

Panel Methods -- Introduction

Since the equations solved by panel methods are linear, we can multiply a known solution by a scalar and add these results together to form more general solutions. This can be made to work in both subsonic and supersonic cases.

Panel methods may be based on one or more fundamental solutions to the Prandtl-Glauert equation or Laplace's equation. These commonly include source, vortex, and doublet flows, discussed in the section on potential theory. The basic idea is to add up known solutions... ... such as a uniform flow... ...and a point source.... ... to produce a streamline pattern that matches the flow of interest.

Here we add a freestream, a source, and a sink (negative source strength) to produce the flow

over an oval (called a Rankine Oval).

We could superimpose many sources and sinks to get nearly any flow pattern we desired:

Panel methods are based on this idea. Sources (or doublets or vortices) of some strength are located in the flow such that their combined solutions satisfy the boundary conditions of the problem. The boundary conditions are typically that the combined flow does not go through the surface, and that far from the body, the flow approaches the freestream solution.

Panel Methods -- Introduction

Since the equations solved by panel methods are linear, we can multiply a known solution by a scalar and add these results together to form more general solutions. This can be made to work in both subsonic and supersonic cases.

Panel methods may be based on one or more fundamental solutions to the Prandtl-Glauert

equation or Laplace's equation. These commonly include source, vortex, and doublet flows, discussed in the section on potential theory. The basic idea is to add up known solutions... ... such as a uniform flow... ...and a point source.... ... to produce a streamline pattern that matches the flow of interest.

Here we add a freestream, a source, and a sink (negative source strength) to produce the flow over an oval (called a Rankine Oval).

We could superimpose many sources and sinks to get nearly any flow pattern we desired:

Panel methods are based on this idea. Sources (or doublets or vortices) of some strength are located in the flow such that their combined solutions satisfy the boundary conditions of the problem. The boundary conditions are typically that the combined flow does not go through the surface, and that far from the body, the flow approaches the freestream solution.

Panel Methods -- AIC Matrix

The next step, after dividing the geometry into panels is to compute the flow pattern at each panel, i, associated with a source or doublet or vortex of unit strength at panel j. The component of the velocity normal to the panel is denoted AIC(i,j), an element of the aerodynamic influence coefficient matrix.

The flow at panel i associated with a singularity of unit strength at panel j can be computed from the basic singularity solution. The result depends on the vector distance between the panels, R. More specifically, the vector from the jth singularity to the control point of panel i (often at the panel centroid). The fundamental solution for the flow field some distance from the singularity is discussed in following sections, but the jth panel usually contains not just a single point source or vortex or doublet, but several vortex lines or a distribution of such singularities. The form of this distribution is one of the things that differentiates various panel methods. The distribution of singularity strength over a panel may be a constant value, or may vary linearly or quadratically in both directions. The list below shows some of the panel codes and the choice of singularity types and distributions.

Panel Methods -- Boundary Conditions

After the AIC matrix is computed, we specify the boundary conditions. The total normal velocity at panel i is then given by the expression below. This must be zero if the flow is tangent to the surface of the body and constitutes the boundary conditions of the problem. Here, {n} is a vector of surface unit normals {sigma} represents the unknown singularity strengths. Each element of these vectors is associated with one panel of the geometry. The boundary conditions for panel methods must express the requirement that streamlines follow the surface contour. But they do not have to explicitly set Vn = 0. In fact, the method currently more in vogue is to specify the B.C.'s in terms of the potential. This is called a Dirichlet (as opposed to the von Neumann) type of boundary condition. It works as follows on the doublet panel method. The total potential in the interior of the section is set to 0. If the total potential is 0 everywhere inside the body (in practice it is set to 0 just inside at each panel control point) then the velocity there is 0 also. In particular the velocity normal to the panel, on the inside of the panel is 0. Since doublets produce no jump in normal velocity (see next section) then Vn = 0 in the external flow as well. This form of the B.C.'s is often better behaved (numerically) than the direct (Neumann) type of B.C..

Nonlinear Methods

Nonlinear CFD methods can be used to predict complex flow fields such as those associated with transonic or separated flows. They have been used recently for predicting flows from air over hypersonic aircraft to blood through artificial hearts. The list below includes links to internet sites with example applications: NAS Technical Summaries (NASA Ames) A gallery of CFD examples from Fluent, Inc. Analytical Methods, Inc.

Here are some suggestions about using these general methods, which seem in principle to be capable of doing anything:

Use the simplest method or model adequate for the job. One need not always use the most sophisticated method to solve fluid mechanics problems. It is sometimes fashionable to use the latest technology, but one need not submit a Cray job to find the roots of a quadratic. Sometimes the "outdated" 1960's technology is just what is needed. Evaluate CFD results critically before accepting them. Know assumptions, limitations of the method. Do an order of magnitude analysis. Do results make sense qualitatively?

Neither computation nor experiment is infallible. "Nobody believes theoretical predictions but the engineer who computed them; everybody believes experimental results but the engineer who conducted the test."

The basis of modern CFD techniques for the solution of the nonlinear equations of fluid flow is illustrated here. We start with the differential equation such as:

This partial differential equation may be solved in two fundamentally different ways: Finite Difference: Discretization of differential form of equations Solutions for all unknowns computed at node points Finite Volume: Discretization of integral form of equations Solution computed at cell centroids To do this requires that the flow field be first divided into a grid. This is often difficult as the grid must not only conform to the body but also be dense in regions of large flow gradients. The two grid directions must be relatively orthogonal so that the difference equations are good approximations to the real PDE. The process of generating such a grid is one of the more difficult aspects of nonlinear CFD. The following examples illustrate some of the approaches.

The intersection of the grid with the surface of a wing/nacelle is shown here. The grid is divided into a number of pieces to better accommodate the complex geometry. (From NAS Summary: Flow Simulation for Subsonic Transports, Pieter G. Buning, NASA Langley Research Center) Although grids with a fixed topological structure are often used, unstructured adaptive grids are

also used, especially over complex, multiply-connected domains as might be found on a multielement airfoil.

From NAS Summaries: Adaptive Unstructured Flow Computations, Dimitri J. Mavriplis, NASA Langley

Shown below is the computer power required to solve various non-linear problems in a reasonable period of time (~15 min). Some of the points are a bit optimistic.

The vast amounts of data generated by these codes are often displayed using computer graphics. Sometimes it is illuminating to use simulated experimental techniques such as oil flow patterns, tufts, or smoke particle traces in order to visualize the results.

From: NAS Summaries -- Viscous Unstructured-Grid Computations, William K. Anderson, NASA Langley Click here for a short Quicktime video clip of a streamline simulation over a wing-body model. Here is another clip from NASA Ames of the flow over a vertical take-off aircraft in groundeffect. Note that a 128 x 128 x 128 grid requires 2.1 M grid points (25 MB just for velocities with 4 bytes per number). A wing with 20 x 40 panels might be modeled with a 60x80x80 grid requiring only 384K points.

Finite Difference Methods

We start with the differential equation such as:

A matrix with these equations (tridiagonal system) is then solved at each time step. The matrix is huge (perhaps a million by a million) but sparse.

Finite Volume Methods

We start with the differential equation such as:

2-D Potential Flow

This chapter starts the description of solution methods in detail. Beginning with the simplest flows: two-dimensional, inviscid and irrotational, the chapter describes the basic theoretical results. These are applied to airfoil problems in later chapters and then modified to include the effects of compressibility and viscosity. 1. Basic Theory 2. Sources and Vortices 3. Interactive Calculations 4. References

5.Basic 2-D Potential Theory

6. 7. We outline here the way in which the "known" solutions used in panel methods can be generated and obtain some useful solutions to some fundamental fluid flow problems. Often the known solutions just come out of thin air and can be applied, but sometimes other approaches are possible. The simplest case, two-dimensional potential flow illustrates this process. We shall discuss 2-D incompressible potential flow and just mention the extension to linearized compressible flow. For this case the relevant equation is Laplace's equation: There are several ways of generating fundamental solutions to this linear, homogeneous, second order differential equation with constant coefficients. Two methods are particularly useful: Separation of variables and the use of complex variables. Complex variables are especially useful in solving Laplace's equation because of the following: We know, from the theory of complex variables, that in a region where a function of the complex variable z = x + iy is analytic, the derivative with respect to z is the same in any direction. This leads to the famous Cauchy-Riemann conditions for an analytic function in the complex plane. Consider the complex function: W = + i The Cauchy-Riemann conditions are: Differentiating the first equation with respect to x and the second with respect to y and adding gives: Thus, analytic function of a complex variable is a solution to Laplace's equation and may be used as part of a more general solution. W = + i is called the complex potential. It consists of the usual velocity potential as the real part and the stream function as its imaginary part. The flow velocities can be then be written as a single complex number: dW/dz = u - iv (Try deriving this.) We consider some simple analytic functions for W that are of great use in applied aerodynamics: Uniform flow:

Line Source or Vortex: Doublet:

Uniform Flow

If U is real the flow is in the x direction with a speed U. The flow direction can be adjusted by changing real and imaginary parts. This is a good example of the fact that the potential is not defined apart from an arbitrary constant. Although the flow is uniform everywhere, the potential depends on our choice of the origin. Differences in the potential are physically meaningful, though and do not depend on the choice of the origin.

Line Source or Vortex

The same expression describes a "point" source or vortex in 2D (which can be thought of as a vortex line or line of sources in 3-D). When K is real the expression describes a source with radially directed induced velocity vectors; imaginary values lead to vortex flows with induced velocities in the tangential direction. Further discussion of these flows is given in the next section.


A doublet is formed by superimposing a source and a sink along the x-axis. The doublet strength is given by S dx. The fundamental doublet singularity with the potential shown above is formed by taking the limit as dx goes to zero and S goes to infinity while keeping the product constant. The doublet is commonly used as one of the fundamental singularities in many panel methods.

Sources and Vortices

Notice that many of the solutions to the 2-D potential equation that we proposed are singular. In fact, the source solution seems the ultimate way of violating continuity while the vortex is the essence of rotational (not irrotational as we assumed) flow.

These solutions are indeed singular at a point and do not satisfy the differential equation at that point. Away from the singularity, however, they are perfectly adequate solutions as can be seen by evaluating the integral forms of the continuity and irrotationality conditions. Why the flow field near a source satisfies continuity:

Why the flow field near a vortex satisfies irrotationality:

The solutions are singular at a point, but even near the singular point strange things happen: the velocity gets very large. In real life, the large velocities in this region give rise to compressibility effects; viscous effects smear the discrete vortex into a distribution of vorticity in a viscous core. The actual velocity distribution near the core of a free vortex behaves more like a solid body with a velocity distribution V(R) = kR. (This is the result obtained by assuming a Gaussian distribution of distributed vorticity in the core region. The size of the viscous core depends on the Reynolds number, often taken as /.)

This 1/r behavior of the vortex induced velocity is not just a mathematical result. It is essential for the flow to exist in equilibrium. We can easily see that the velocity must vary as 1/r for the pressure gradients to balance the centrifugal force acting on the fluid. The derivation is given

here. We can combine these singularities in different locations to produce the desired flow pattern. Since the solution to Laplace's equation is uniquely determined in regions without singularities when the solution on the boundaries is specified, we can use combinations of singularities to model many flows of interest.

Method of Images: Ground Effects, Wall Interference

Source Doublets, Circular Cylinder, Ellipses, Blasius Theorem

Groups of Vortices, Far Field Flow, Stokes Theorem

Free Vortex Motion

Method of Images

The flow field created by singularities in the presence of solid boundaries can be simulated by superimposing "image vortices". This works because the symmetry of the problem on the right ensures that there is no flow through the plane of symmetry. The boundary does the same thing for the problem on the left. Since both of these problems have the same boundary conditions and satisfy the same linear differential equation, the flow must be the same.

This technique is useful for simulating the effects of the ground on the aerodynamics of cars or airplanes at low altitude. It can also be used in more complex situations. Here, three images are required to simulate the boundary conditions associated with a corner.

This technique is used to predict the effects of wind tunnel walls on the flow field of models being tested. Imagine the system of image vortices that would be required to simulate wall effects on a 2D airfoil test. Yes, more than 2 images are required. The 3-D situation cannot in general be solved with images.

The flow on a circular cylinder may be computed from a uniform stream and a doublet. (See previous section.)

Some interesting conclusions and generalizations follow from the expressions for the velocity and the potential on a circular cylinder shown above.

Note that on the surface of the cylinder, the tangential velocity is: V = 2U sin , so the maximum velocity is twice the freestream value.

The more general forms of these results hold for all ellipsoids: Vmax = V (1 + t/c) and V at surface = n x (n x Vmax)) Notice that this holds exactly in incompressible potential flow, even if the ellipse has a t/c much larger than 1. Of course, in such a case, the real flow will probably look quite different from the potential flow solution.

The force on a general 2-D cylinder can be computed by calculating the velocities, using Bernoulli's law to compute pressures, then integrating the surface pressures. However, the total forces and moments can be derived directly from the complex potential. The result is called the Blasius theorem. It is not derived here, but the result follows from the theory of residues, the complex potential, and the incompressible Bernoulli equation. (Or one might just use the momentum equation and compute the net force by far field integrals.)

where is the total circulation and S is the net source strength. In the case of no net source strength, the net force exerted on a collection of sources and vortices in a flow with freestream velocity U is perpendicular to the freestream and proportional to U and the total circulation.

Circulation, Vorticity, and Stokes Theorem

Stokes' theorem is an integral identity that may be written:

When the vector function F is taken to be the velocity field, V, then this relation in 2-D may be restated as:

This result implies that the circulation around a contour that contains a group of vortices is just equal to the sum of the enclosed vortex strengths.

This allows application of the Blasius theorem to find the force acting on a group of vortices. The result is sometimes called the Kutta-Joukowski law:

We can also treat the flow field far from a group of vortices as if it were created by a single vortex with a strength equal to the sum of the individual vortices. Such far field solutions can be especially simple and useful as a check of more complex results. Far field solutions can also be used as boundary conditions for the more complex near field solution, reducing the required extent of computational grids. We should note here that just because we find a superposition of singularities that satisfies the boundary conditions and the differential equation, it does not mean that we have found the only solution to the problem. For example, we could add a vortex to the doublet that was used to model the circular cylinder, and we would still find that the flow went around the cylinder. These non-unique solutions are problemsome and we appeal to additional considerations to find the one(s) that actually will appear in nature. Just such an auxiliary condition, the Kutta condition, is provided by viscous effects which then determine the value of circulation.

Free Vortices
Singularities that are free to move in the flow do not behave in response to F = ma (what is m?). Rather they move with the local flow velocity. Thus, vortices and sources are convected downstream with the flow. And interacting singularities can produce complex motions due to their mutual induced velocities.

A pair of counter-rotating vortices moves downward because of their mutual induced velocities.

Co-rotating vortices orbit each other under the influence of their mutual induced velocities.

Streamlines Past Sources and Vortices

Drag any of the singularities from the well on the right into the main computation area. Set the freestream speed (the flow is from left to right), then click Compute. The marks on the page simulate small tufts and indicate the direction of the local flow. Experiment with multiple singularities to simulate a pair of wing trailing vortices, a source/sink doublet, or a spinning baseball.

Airfoils, Part I: Introduction

In this chapter, an introduction to airfoils and airfoil theory is followed by the application of potential flow methods to the analysis of airfoils. The purpose of this section is to discuss the relation between airfoil geometry and airfoil performance. To do this we will discuss the methods that are used to compute the distribution of pressures over the airfoil surface. Then we will discuss the relation between these pressures and the airfoil performance.

Outline of this Chapter The chapter is divided into several sections. The first of these consist of an introduction to airfoils: some history and basic ideas. The latter sections deal with simple analyses and results that relate the airfoil geometry to its basic aerodynamic characteristics.

History and Development Airfoil Geometry Pressure Distributions Relation between Cp and Performance Relating Geometry and Cp

Methods of Airfoil Analysis References

History of Airfoil Development

The earliest serious work on the development of airfoil sections began in the late 1800's. Although it was known that flat plates would produce lift when set at an angle of incidence, some suspected that shapes with curvature, that more closely resembled bird wings would produce more lift or do so more efficiently. H.F. Phillips patented a series of airfoil shapes in 1884 after testing them in one of the earliest wind tunnels in which "artificial currents of air (were) produced from induction by a steam jet in a wooden trunk or conduit." Octave Chanute writes in 1893, "...it seems very desirable that further scientific experiments be be made on concavo-convex surfaces of varying shapes, for it is not impossible that the difference between success and failure of a proposed flying machine will depend upon the sustaining effect between a plane surface and one properly curved to get a maximum of 'lift'."

At nearly the same time Otto Lilienthal had similar ideas. After carefully measuring the shapes of bird wings, he tested the airfoils shown here (reproduced from his 1894 book, "Bird Flight as the Basis of Aviation") on a 7m diameter "whirling machine". Lilienthal believed that the key to successful flight was wing curvature or camber. He also experimented with different nose radii and thickness distributions.

Airfoils used by the Wright Brothers closely resembled Lilienthal's sections: thin and highly cambered. This was quite possibly because early tests of airfoil sections were done at extremely low Reynolds number, where such sections behave much better than thicker

ones. The erroneous belief that efficient airfoils had to be thin and highly cambered was one reason that some of the first airplanes were biplanes. The use of such sections gradually diminished over the next decade.

A wide range of airfoils was developed, based primarily on trial and error. Some of the more successful sections such as the Clark Y and Gottingen 398 were used as the basis for a family of sections tested by the NACA in the early 1920's.

In 1939, Eastman Jacobs at the NACA in Langley, designed and tested the first laminar flow airfoil sections. These shapes had extremely low drag and the section shown here achieved a lift to drag ratio of about 300.

A modern laminar flow section, used on sailplanes, illustrates that the concept is practical for some applications. It was not thought to be practical for many years after Jacobs demonstrated it in the wind tunnel. Even now, the utility of the concept is not wholly accepted and the "Laminar Flow True-Believers Club" meets each year at the homebuilt aircraft fly-in.

One of the reasons that modern airfoils look quite different from one another and designers have not settled on the one best airfoil is that the flow conditions and design goals change from one application to the next. On the right are some airfoils designed for low Reynolds numbers. At very low Reynolds numbers (<10,000 based on chord length) efficient airfoil sections can look rather peculiar as suggested by the sketch of a dragonfly wing. The thin, highly cambered pigeon wing is similar to Lilienthal's designs. The Eppler 193 is a good section for model airplanes. The Lissaman 7769 was designed for human-powered aircraft.

Unusual airfoil design constraints can sometimes arise, leading to some unconventional shapes. The airfoil here was designed for an ultralight sailplane requiring very high maximum lift coefficients with small pitching moments at high speed. One possible solution: a variable geometry airfoil with flexible lower surface.

The airfoil used on the Solar Challenger, an aircraft that flew across the English Channel on solar power, was designed with an totally flat upper surface so that solar cells could be easily mounted.

The wide range of operating conditions and constraints, generally makes the use of an existing, "catalog" section, not best. These days airfoils are usually designed especially for their intended application. The remaining parts of this chapter describe the basic ideas behind how this is done

Airfoil Geometry

Airfoil geometry can be characterized by the coordinates of the upper and lower surface. It is often summarized by a few parameters such as: maximum thickness, maximum camber, position of max thickness, position of max camber, and nose radius. One can generate a reasonable airfoil section given these parameters. This was done by Eastman Jacobs in the early 1930's to create a family of airfoils known as the NACA Sections.

The NACA 4 digit and 5 digit airfoils were created by superimposing a simple meanline shape with a thickness distribution that was obtained by fitting a couple of popular airfoils of the time: y = (t/0.2) * (.2969*x0.5 - .126*x - .3537*x2 + .2843*x3 - .1015*x4) The camberline of 4-digit sections was defined as a parabola from the leading edge to the position of maximum camber, then another parabola back to the trailing edge.

NACA 4-Digit Series: 4 4 max camber position in % chord of max camber in 1/10 of c

1 2 max thickness in % of chord

After the 4-digit sections came the 5-digit sections such as the famous NACA 23012. These sections had the same thickness distribution, but used a camberline with more curvature near the nose. A cubic was faired into a straight line for the 5-digit sections.
NACA 5-Digit Series: 2 3 0 approx max position camber of max camber in % chord in 2/100 of c 1 2 max thickness in % of chord

The 6-series of NACA airfoils departed from this simply-defined family. These sections were generated from a more or less prescribed pressure distribution and were meant to achieve some laminar flow.
NACA 6-Digit Series: 6 3, Sixlocation 2 half width 2 ideal Cl 1 2 max thickness


of min Cp in 1/10 chord

of low drag in tenths bucket in 1/10 of Cl

in % of chord

After the six-series sections, airfoil design became much more specialized for the particular application. Airfoils with good transonic performance, good maximum lift capability, very thick sections, very low drag sections are now designed for each use. Often a wing design begins with the definition of several airfoil sections and then the entire geometry is modified based on its 3-dimensional characteristics.

Airfoil Pressure Distributions

The aerodynamic performance of airfoil sections can be studied most easily by reference to the distribution of pressure over the airfoil. This distribution is usually expressed in terms of the pressure coefficient:

Cp is the difference between local static pressure and freestream static pressure, nondimensionalized by the freestream dynamic pressure. (See discussions of Cp and the Bernoulli equation.) What does an airfoil pressure distribution look like? We generally plot Cp vs. x/c. x/c varies from 0 at the leading edge to 1.0 at the trailing edge. Cp is plotted "upside-down" with negative values (suction), higher on the plot. (This is done so that the upper surface of a conventional lifting airfoil corresponds to the upper curve.) The Cp starts from about 1.0 at the stagnation point near the leading edge... It rises rapidly (pressure decreases) on both the upper and lower surfaces... ...and finally recovers to a small positive value of Cp near the trailing edge. Various parts of the pressure distribution are depicted in the figure below and are described in the following sections.

Upper Surface The upper surface pressure is lower (plotted higher on the usual scale) than the lower surface Cp in this case. But it doesn't have to be. Lower Surface The lower surface sometimes carries a positive pressure, but at many design conditions is actually pulling the wing downward. In this case, some suction (negative Cp -> downward force on lower surface) is present near the midchord. Pressure Recovery This region of the pressure distribution is called the pressure recovery region. The pressure increases from its minimum value to the value at the trailing edge. This area is also known as the region of adverse pressure gradient. As discussed in other sections, the adverse pressure gradient is associated with boundary layer transition and possibly separation, if the gradient is too severe. Trailing Edge Pressure The pressure at the trailing edge is related to the airfoil thickness and shape near the trailing edge. For thick airfoils the pressure here is slightly positive (the velocity is a bit less than the freestream velocity). For infinitely thin sections Cp = 0 at the trailing edge. Large positive values of Cp at the trailing edge imply more severe adverse pressure gradients. CL and Cp When the chord is taken as 1 unit, the section lift coefficient is related to the Cp by:

(It is the area between the curves with Cpu = upper surface Cp and recall Cl = section lift / (q c) )

Stagnation Point The stagnation point occurs near the leading edge. It is the place at which V = 0. Note that in incompressible flow Cp = 1.0 at this point. In compressible flow it may be somewhat larger.

We can get a more intuitive picture of the pressure distribution by looking at some examples and this is done in the follo

Airfoil Pressures and Performance

The shape of the pressure distribution is directly related to the airfoil performance as indicated by some of the features shown in the figure below.

Most of these considerations are related to the airfoil boundary layer characteristics which we will take up later, but even in the inviscid case we can draw some conclusions. We may compute the maximum local Mach numbers and relate those to lift and thickness; we can compute the pitching moment and decide if that is acceptable. Whether we use the inviscid pressures to form qualitative conclusions about the section, or use them as input to a more detailed boundary layer calculation, we must first investigate the close relation between the airfoil geometry to these pressures.

Relating Airfoil Geometry and Pressures

Before discussing in detail the methods used to predict airfoil pressure distributions let's consider, more intuitively the relationship between airfoil geometry and airfoil pressure distributions. We first look at the effect of changes in surface curvature (Click on figure to look in more detail.)

The figure below shows how the airfoil pressures vary with angle of attack. Note that the "nose peak" becomes more extreme as the angle increases.

To make this a bit more clear, you may use the small java program below to change the angle of attack and see its effect on Cp, Cl, and Cm. Click on the upper half of the plot to increase the angle of attack, alpha, and on the lower portion to decrease it.

Let's consider, in more detail the relationship between airfoil geometry and airfoil pressure distributions. The next few examples show some of the effects of changes in camber, leading edge radius, trailing edge angle, and local distortions in the airfoil surface. A reflexed airfoil section has reduced camber over the aft section producing less lift over this region. and therefore less nose-down pitching moment. In this case the aft section is actually pushing downward and Cm at zero lift is positive.

A natural laminar flow section has a thickness distribution that leads to a favorable

pressure gradient over a portion of the airfoil. In this case, the rather sharp nose leads to favorable gradients over 50% of the section.

This is a symmetrical section at 4 angle of attack. Note the pressure peak near the nose. A thicker section would have a less prominent peak.

Here is a thicker section at 0. Only one line is shown on the plot because at zero lift, the upper and lower surface pressure coincide.

A conventional cambered section.

An aft-loaded section, the opposite of a reflexed airfoil carries more lift over the aft part of the airfoil. Supercritical airfoil sections look a bit like this.

The best way to develop a feel for the effect of the airfoil geometry on pressures is to interactively modify the section and watch how the pressures change. A Program for ANalysis and Design of Airfoils (PANDA) does just this and is available from Desktop Aeronautics. A very simple version of this program, is built into this text and allows you to vary airfoil shape to see the effects on pressures. (Go to Interactive Airfoil Analysis page by clicking here.) The full version of PANDA permits arbitrary airfoil shapes, permits finer adjustment to the shape, includes compressibility, and computes boundary layer properties.

Interactive Airfoil Analysis


The program built into this page allows you to experiment with the effect of airfoil shape and angle of attack on the pressure distribution.


Click on the top part of the plot to increase the angle of attack; clicking on the lower portion reduces alpha. Drag the handles shown on the upper or lower surfaces to modify the shape of the section and watch the effects on Cp.
Suggested Exercises

Change the airfoil thickness and note the effect on upper and lower surface pressures. Notice how thickness affects the Cp at the trailing edge. Create a pressure peak near the nose on the upper or lower side by changing the angle of attack. Change the camber near the nose to remove the pressure peak. Try to create a positive pitching moment section, a very thin, highly cambered section, and a symmetrical section.

Technical Details

This program uses a combination of thin airfoil theory and conformal mapping to very quickly compute pressures on an airfoil. A method like this was used in the 1950's to compute airfoil pressure distributions before Java was invented. The section shape is very simple as well: upper and lower surfaces consist of a quadratic in sqrt(x) and a quadratic in (1-x), patched together at the control points. This provides just 4 degrees of freedom, but does lead to curves that look like airfoils.

Conformal Mapping

Any analytic function of a complex variable satisfies the equation for incompressible, irrotational flow: Why? We can, therefore, relate one flow field to another by setting: where z' is related to z by an analytic function of z, z' = f(z). (Recall z = x + iy.) The idea behind airfoil analysis by conformal mapping is to relate the flow field around one shape which is already known (by whatever means) to the flow field around an airfoil. Most often a circle is used as the first shape. The problem is to find an analytic function that relates every point on the circle to a corresponding point on the airfoil.

Joukowski found that the simple function: z' = z + 1/z transforms a circle to a shape which looks a bit like an airfoil. By taking the origin of the circle at various points, different airfoil-like shapes are produced.

With just 2 degrees of freedom (the coordinates of the circle origin), the number of airfoil shapes that can be represented with the simple Joukowski transformation is limited.

Furthermore, the thickness is greatest at 25% chord -- rather far forward. This constraint has led to a number of generalizations of the Joukowski mapping that produce more practical shapes. But all of these methods are based on the same basic idea which is illustrated in the simple mapping. We begin with a circle. The center of the circle is at Xc, Yc.

Every point z is mapped to the point z' by the relation: z' = z + 1/z. If W(z) = F(z) + i G(z) is the complex potential function, then the velocity is given by: w = u - iv = dW/dz. We set W(z') = W(z) so that the velocity on the airfoil may be related to the velocity on the cylinder by: w(z') = dW/dz' = (dW/dz) (dz/dz') = w(z) / ( 1-z-2). Notice that this relates the velocity on the airfoil directly to that on the circle, but the relation blows up when z2 = 1. We know that the velocity on the airfoil does not go to infinity anywhere. The reason that the mapping does not work at these points is that the mapping is not analytic here. This does not mean it cannot be used, it just means that we must make sure that such points are not in the flow: they must be inside or on the airfoil. Here we choose the circle so that it encloses the point -1,0, and we choose the circulation so that the velocity is 0 at the point 1,0. The point 1,0 maps to the airfoil trailing edge.

If we are to have a stagnation point on the cylinder at 1,0 we must have a certain amount of circulation. The origin of the circle thus determines the lift on the airfoil at a given angle of attack. (Also note that the point -1,0 must be enclosed.) One of the troubles with conformal mapping methods is that parameters such as xc and yc are not so easily related to the airfoil shape. Thus, if we want to analyze a particular airfoil, we must iteratively find values that produce the desired section. A technique for doing this was developed by Theodorsen. Another technique involves superposition of fundamental solutions of the governing differential equation. This method, discussed in subsequent sections, is called thin airfoil theory

Thin Airfoil Theory Derivation

We start with the analysis of a very thin cambered plate and will build up the solution to a more arbitrary airfoil.

A distribution of vorticity on the airfoil will be a solution to Laplace's equation. It will satisfy the boundary conditions if the combination of the velocity induced by the vortices cancels the component of the freestream normal to the plate: (where small angle approximations have been introduced) The basic approximation of thin airfoil theory is that the velocity induced at some point x due to the vorticity at x'...

... may be approximated by the velocity induced at the same x position on the x axis due to a vortex on the x axis:

The velocity induced by this bit of vorticity is computed from the basic vortex singularity. The formula is known as the Biot-Savart Law and in 2-D for the element of vorticity at x', it reads:

So, the total induced velocity at the point x is given by:

Combining this expression with the flow-tangency boundary condition, we have the basic

integral equation to be solved for the unknown vorticity distribution:

The approach to solving this equation is to change variables: and to write as a Fourier series:

Substituting (4) into (2) and using the trigonometric relations:


Finally, we multiply both sides by cos m and integrate from 0 to :

We can substitute these coefficients into the expression for (4), to find the vorticity, pressures, lift, and moment as a function of the surface slope distribution, dz/dx. In particular, the expressions for the local pressure difference and integrated lift and moment about the leading edge are:

(Note that the expression relating Cp and applies only to thin airfoils. )

This method of computing the circulation distribution permits us to compute the pressure difference across the airfoil. The pressures on the upper and lower surfaces may also be computed by noting that at the surface the perturbation velocities in the x direction caused by the singularities are zero except for those due to the local vorticity:

The only contribution to du(x) is from the local vorticity (at x). It can be shown that this perturbation velocity is du = / 2, with the + sign for the upper surface and the - sign for the lower surface.

Thin Airfoil Theory Results

The basic equations derived from thin airfoil theory are repeated below:

Several important results are derived from these expressions and are described in the following sections. Ideal angle of attack

The constants An just depend on the airfoil shape -- except for A0 which depends also on the angle of attack. The A0 term is a strange term in the Fourier series for since it leads to a singularity at the leading edge ( -> infinity). Thus, there is one angle of attack, called the ideal angle of attack, at which A0 = 0 and the vorticity goes to 0 at the leading edge. This angle is called the ideal angle of attack.

Of course in real flows, the vorticity would not become infinite (why?), but the concept of ideal angle of attack is still important, identifying the flow conditions for which leading edge pressure peaks are avoided.

Lift curve slope

The rate of change of lift coefficient with angle of attack, dCL/d can be inferred from the expressions above. The result, that CL changes by 2 per radian change of angle of attack (.1096/deg) is not far from the measured slope for many airfoils. The effects of thickness and viscosity which are ignored here cancel each other out to some extent with the result that most airfoils have a lift curve slope within 10% of the 2 value given by thin airfoil theory.
Aerodynamic center and pitching moment at L = 0

The expression for pitching moment coefficient measured about the leading edge is given above. If we measure the moment about another reference center at a position x0/c, the expression becomes:

Note that if we choose the point x0 = 0.25c, then the lift dependence drops out and the moment coefficient measured about this point is independent of the angle of attack*. The point about which dCm/dCL = 0 is called the aerodynamic center and according to thin airfoil theory it is the quarter chord point of the airfoil. Experiments show this to be quite close.
Results for a general example + Rules of Thumb

The parabolic camber meanline is used as an example of thin airfoil theory. Results for this case serve as useful first approximations for any thin cambered airfoil. Assume: z(x) = 4 h x (1-x)

Thin airfoil theory can also be used to estimate the effect of flap deflection on airfoil lift and moment. It also provides an estimate of the hinge moments vs. the deflection angle and the angle

of attack. This is a good problem to work on your own.

The results are:

where: Because of the effects of viscosity, these results tend to overestimate, to some extent, the lift and moment due to flap deflections.

*The Cm about this point, denoted Cma.c., is therefore equal to the Cm at zero lift. Cm0, as it is written, is also independent of the moment reference location and so is a particularly useful quantity: Cm0 = Cma.c.

Thin Airfoil Inverse Design

The basic thin airfoil theory formulation can be used to design airfoils with a desired pressure distribution. This process is actually easier than the direct analysis. The integral equation used to obtain the circulation distribution given the airfoil shape:

can be evaluated directly if we know the circulation distribution. We evaluate the integral at several values of x to find the surface slope dz/dx along the airfoil. which can be integrated to obtain:

C and can be chosen to make z(0) = z(1) = 0.

Thickness Effects

The same ideas used in predicting the characteristics of very thin airfoils with camber can be applied to study airfoils with some thickness.

Just as we placed vortices on the airfoil chord line to model a camber line, we can place sources on the x axis to model a thick symmetric airfoil. The source strengths can be computed much more simply than the vortex strengths, however.

To satisfy continuity in the region ABCD:

where dx is the source strength contained in the box.

To first order then: The induced velocity on the chord line is:

Note that the condition for tangent flow, is automatically satisfied by the given solution for source strength in (2). The total x component of velocity is then:

Near the nose of the airfoil our approximation of small angles is no longer valid and the answers given by the previous expression are not accurate. In the 1950's Riegels introduced a correction to thin airfoil theory that greatly improves the accuracy. In fact, the Riegels correction produces results that work for ellipses up to and exceeding 100% thick, while the previous theory only works well up to 10 or 15%. The velocity on the surface of the airfoil

with Riegels modification is:

Moderately Thin Airfoils

A general cambered airfoil may be represented by sources and vortices with the source strength of a symmetrical airfoil with the same thickness distribution and the vorticity distribution of a cambered plate at the desired angle of attack.

This solution will lead to a singularity at the leading edge except at the ideal angle of attack. A better estimate of the velocity distribution can be obtained by adding the velocity distribution around a thick symmetrical airfoil at angle of attack to the perturbations associated with the camberline at the ideal angle of attack. This is not as simple since thin airfoil theory does not provide the velocity distribution over a thick symmetrical airfoil at angle of attack. However, if we can find this distribution, from experiment or some other theory, we can combine it with thin airfoil results to provide a more general solution.

Surface Panel Methods for Airfoils

Rather than make the assumption of thin airfoil theory, that the singularities and boundary conditions are evaluated on the x-axis, we can use a panel method to compute the forces and vorticity on a thin airfoil.

One approach is to place vortices () on the mean line as shown above, with boundary conditions of tangential flow satisfied at the control points (x). Vortices are often placed at the 1/4 chord point of each panel with control points at the 3/4 chord point. This is chosen solely to make the lift and moment come out right with very few panels. The details of this method are left as a homework exercise. One may also place vorticity on the actual surface (not the mean line) of a thick section. A common approach involves the distribution of linearly-varying vorticity panels. One solves for the vorticity strength at panel edges and requires that the boundary conditions and Kutta condition are satisfied. More detail on this method is given in Kuethe and Chow or Moran.

Recall that the PDE and BC's are not sufficient to produce a unique solution. We have gotten away without an explicit extra condition until now by using some tricks. In thin airfoil theory, we assumed a form of the circulation distribution that had no vorticity at the trailing edge. For the discrete vortex panel model, the selection of the control points one-half panel from the vortices effectively did the same thing. In a true surface panel method, this auxiliary condition must be included explicitly. By forcing the vorticity on the upper surface to be equal in magnitude but of opposite sign to that on the lower surface, one obtains flows that satisfy the Kutta condition. This condition is enforced in reality by viscosity: if the flow did continue around the sharp trailing edge (at very high speeds), the boundary layer would separate on the upper surface. Here is a Fortran subroutine that computes the pressure distribution on arbitrary airfoils using a linear surface vorticity distribution.
subroutine LVFoil(XB,YB,n, alphaD, X, Y, Cp) C C Linear vorticity surface panel method for airfoils. C C Adapted from Kuethe and Chow 4th Edition C This version by I. Kroo 2-2-87


No attempt has been made to make this particularly fast or efficient. Inputs: ------XB, YB n alphaD Outputs: -------X, Y Cp

x and y coordinates of panel edges starting at trailing edge proceeding forward on lower surface, wrapping around leading edge and then running back to the trailingedge. Number of panels Angle of attack in degrees

coordinates of panel centers incompressible pressure coefficient at panel center

Declarations: ------------implicit none integer ndim parameter (ndim = 100) real XB(ndim), YB(ndim), X(ndim), Y(ndim), S(ndim), SINT(ndim) real COST(ndim), THETA(ndim), V(ndim), RHS(ndim), CP(ndim) real CN1(ndim,ndim), CN2(ndim,ndim) real CT1(ndim,ndim), CT2(ndim,ndim) real AN(ndim,ndim), AT(ndim,ndim) real pi, alpha, A, B, C, D, E, F, G, P, Q, P1, P2, P3, P4 real COSA, SINA, ANmax, alphaD integer n, np1, i, j, IP(ndim), ip1 logical err


Constants --------pi = 3.14159265 Calculation of geometric data: -----------------------------np1 = n+1 do 1 i = 1,n ip1 = i+1 X(i) = .5 * (XB(i) + XB(ip1)) Y(i) = .5 * (YB(i) + YB(ip1)) S(i) = SQRT( (XB(ip1)-XB(i))**2 + (YB(ip1)-YB(i))**2) THETA(i) = ATAN2( (YB(ip1)-YB(i)), (XB(ip1)-XB(i)) ) SINT(i) = SIN(THETA(i)) COST(i) = COS(THETA(i)) continue alpha = alphaD *pi/180.

SINA = SIN(alpha) COSA = COS(alpha) C C C C Calculation of influence coefficients ------------------------------------do 3 i = 1,n do 2 j = 1,n if (i.eq.j) then CN1(i,j) = -1. CN2(i,j) = 1. CT1(i,j) = .5*pi CT2(i,j) = CT1(i,j) else A = -(X(i)-XB(j))*COST(j) - (Y(i)-YB(j))*SINT(j) B = (X(i)-XB(j))**2 + (Y(i)-YB(j))**2 C = SINT(i)*COST(j)-COST(i)*SINT(j) D = COST(i)*COST(j)+SINT(i)*SINT(j) E = (X(i)-XB(j))*SINT(j) - (Y(i)-YB(j))*COST(j) F = ALOG( 1. + S(j)*(S(j)+2.*A)/B ) G = ATAN2(E*S(j), B+A*S(j)) P1 = 1.-2.*SINT(j)*SINT(j) P2 = 2.*SINT(j)*COST(j) P3 = SINT(i)*P1 - COST(i)*P2 P4 = COST(i)*P1 + SINT(i)*P2 P = (X(i)-XB(j)) * P3 + (Y(i)-YB(j)) * P4 Q = (X(i)-XB(j)) * P4 - (Y(i)-YB(j)) * P3 CN2(i,j) = D + ( .5*Q*F - (A*C+D*E)*G )/S(j) CN1(i,j) = .5*D*F + C*G - CN2(i,j) CT2(i,j) = C + ( .5*P*F + (A*D-C*E)*G )/S(j) CT1(i,j) = .5*C*F - D*G - CT2(i,j) end if 2 3 continue continue ANmax = 0. do 5 i = 1,n AN(i,1) = CN1(i,1) AN(i,np1) = CN2(i,n) AT(i,1) = CT1(i,1) AT(i,np1) = CT2(i,n) do 4 j = 2,n AN(i,j) = CN1(i,j) + CN2(i,j-1) AT(i,j) = CT1(i,j) + CT2(i,j-1) if(ABS(AN(i,j)).gt.ANmax)ANmax = ABS(AN(i,j)) continue continue The Kutta Condition is imposed by the relation: A*gamma(1) + A*gamma(n+1) = 0. A would be 1 but for matrix conditioning problems. AN(np1,1) = 1 AN(np1,np1) = 1

4 5 C C C

do 6 j = 2,n AN(np1,j) = 0. continue RHS(np1) = 0.

C C C C C 11

Decompose and solve the system: ------------------------------call DECOMP(np1, ndim, AN, IP) RHS(i) = SIN( THETA(i)-alpha ) do 11 i = 1,n RHS(i) = SINT(i)*COSA - COST(i)*SINA continue RHS(np1) = 0. call SOLVE(np1, ndim, AN, RHS, IP)


Compute derived quantities: --------------------------V(i) = COS( THETA(i) - alpha ) do 8 i = 1,n V(i) = COST(i)*COSA + SINT(i)*SINA do 7 j = 1,np1 V(i) = V(i) + AT(i,j) * RHS(j) continue Cp(i) = 1.-V(i)*V(i) continue return end

7 8

C MATRIX TRIANGULARIZATION BY GAUSSIAN ELIMINATION C C C Variables: C N The size of the system to be solved C NDIM The dimension of the maxtrix as declared in the calling program C C(NDIM,NDIM) The matrix to be decomposed by L-U decomposition. C On return C is the decomposed form of the matrix. C IP(NDIM) A vector of pivots which is used by the routine SOLVE. C If IP(1)=0 then the matrix is singular. C SUBROUTINE DECOMP(N,NDIM,C,IP) REAL C(NDIM,NDIM),T INTEGER IP(NDIM) IP(N)=1

3 4 5 6


C SOLUTION OF LINEAR SYSTEM CX = B C C Variables: C N The size of the system to be solved C NDIM The dimension of the maxtrix as declared in the calling program C C(NDIM,NDIM) The matrix as returned by the routine DECOMP C IP(NDIM) A vector of pivots. If IP(1)=0 then the matrix is singular. C B(NDIM) The right hand side vector. On return, the solution vector. C C SUBROUTINE SOLVE(N,NDIM,C,B,IP) REAL C(NDIM,NDIM),B(NDIM),T INTEGER IP(NDIM) IF(N.NE.1)THEN NM1=N-1 DO 2 K = 1,NM1 KP1=K+1 M=IP(K) T=B(M) B(M)=B(K) B(K)=T DO 1 I=KP1,N

1 2


3 4

Compressibility in 2D
In this section we discuss the basic equations of compressible flow, the effect of compressibility on airfoil performance, as well as the analysis and design of transonic and supersonic airfoils. 1. Basic Results from Compressible Flow Theory 2. Effects on Airfoils 3. Linear Theory 4. Transonics 5. Supersonic Flow 6. References

Basic Results from Compressible Flow Theory

Compressibility alters the results of the previous theories. It is an important effect when the density variation is significant - and this situation can be related to several parameters. First, it is most important for gasses because most liquids require large pressure changes to cause even mild density variations. But the effect is of great importance to most aircraft, propulsion systems, rockets, and any vehicle in air

with large pressure variations. Pressure variations produce density changes that can be related to the speed of sound:

Since the sea level air pressure is 2116 lb/ft2 (101 KPa), changes of order 150 lb/ft2 (7.2 KPa), required for an airplane to fly, produce a very small change in the overall pressure and hence density. At 35,000 ft (10.7 km) the pressure is 500 lb/ft2 (24 KPa), so the disturbances created by an airplane are much greater and compressibility effects can become important. The connection with Mach number is clear from the following: At sea level, the air pressure is 2116 lb/ft2 (101 KPa), and the density = .002377 sl/ft3 (1.23 kg/m3), so: 1/2 V2 = .1 p when M = .378 At 35,000 ft (10.7 km) the air pressure is 499 lb/ft2 (24 KPa) and the density is .000738 sl/ft3 (0.38 kg/m3), so: 1/2 V2 = .1 p when M = .378 (Same number!) Thus, the importance of density changes can be directly related to the Mach number. When M < .3 the effects are small, when M > .7 the effects can be large. (But be careful about this rule of thumb. It is not uncommon to see a Cp of -7 on an airfoil at M=0.3. This implies a local Mach number of 1.0.) When the density is not constant the assumptions we made in deriving the incompressible Bernoulli equations are no longer valid. The effects of density changes can be incorporated as follows: We consider adiabatic (no heat addition) and reversible (no friction or dissipation) flow: isentropic flow. Many flows outside the boundary layer are isentropic. From basic thermodynamics we have:

together with the equation of state for a perfect gas: p = R T We obtain the isentropic relations:

Combining these with the Euler equation: dp = - V dV leads to the compressible Bernoulli eqn.

Note that the dynamic pressure can be written:

The effect of compressibility can be seen directly in the way a flow responds to changes in the distances between streamlines:

from the continuity equation: S V = constant together with Euler's equation and the expression for the speed of sound, we can derive:

This shows that as M gets closer to 1.0, small changes in area produce large changes in velocity. This is reflected in the Prandtl-Glauert compressibility correction factor. While at M=0, a 1% reduction in area produces a 1% increase in speed, at M = 0.9, a 1% reduction in area produces a 5% increase in speed. At M>1, the sign of the velocity increment changes, so while converging flows speed up subsonically, they slow down if supersonic. This group of equations may be combined in various ways to predict many characteristics of a flow, knowing some of the other characteristics. This is demonstrated in the isentropic flow calculator below. Isentropic Flow Computations

Enter known values in the table below. Select the checkbox if the value is a fixed input.

Compressibility Effects on Airfoils

Compressibility produces changes to the flow patterns and in the resulting airfoil forces and moments.

At low Mach numbers, the local flow velocities over the airfoil are changed because of compressibility, but the overall shape is not changed.

As the Mach number is increased, a local zone of supersonic flow may develop. This changes the pressure distribution more significantly. The airfoil shown here has some lift and therefore the upper surface becomes supersonic first.

Although the transition from subsonic to supersonic flow occurs smoothly, the transition back to subsonic flow occurs through a shock. When the local Mach number becomes appreciably greater than 1, the shock wave becomes strong and the pressure distribution is radically altered. This happens very quickly (small changes in M). The characteristic lambda shape is caused by the subsonic flow in the boundary layer and is exaggerated here.

As the Mach number is further increased, shocks develop on the lower surface as well.. The sudden increase in pressure after the shock can cause the boundary layer to separate, with correspondingly large increases in drag. The shock also moves back as the Mach number is increased.

The shock wave gradually moves back to the trailing edge and almost all of the airfoil is surrounded by supersonic flow. In this condition the drag is very high and the airfoil is operating far from the design point. Further increases in Mach number (supersonic incident flow) change the flow character totally. When the freestream flow is supersonic, an airfoil of the type shown below behaves very differently. The supersonic flow is not able to suddenly change direction when it happens onto the airfoil leading edge.

A bow wave is formed, extending far into the flow field. Only a small region of the airfoil is subsonic.

The bow wave is bent back as the Mach number is increased further. At this speed only a very small area near the nose remains subsonic.

One finds from experiments an increase in the wing lift curve slope as the Mach number is increased. The increase in Cl with Mach number reverses itself when strong shocks form

on the wing.

Other effects include a moving aft of the center of pressure that causes a nose-down pitching moment, and a rapid increase in drag as shocks form.

This figure shows how the drag increases with Mach number at a given CL. The data are based on flight tests of a medium size airliner.

Linear Compressibility
Some of the effects of compressibility can be predicted quite simply in the Mach number range for which the flow still satisfies the linear differential equations. This can be done using the socalled compressibility corrections. The simplest of these is the Prandtl-Glauert correction. It is derived from the governing differential equation.

One can make a change of variables:

to obtain the incompressible potential equation in the new variable x':

The boundary conditions then need to be revised in terms of x' and the resulting velocity will be:

A more revealing approach comes from introducing the variables:

In terms of these variables we can write the Prandtl-Glauert equation as: which describes an incompressible potential flow, phi'(x,y')

The original (linearized) boundary conditions are: If we imagine a fictitious airfoil with coordinates: y'a(x) in the incompressible flow field described by the potential phi', the boundary conditions are:

Noting that it follows that at a given x location, the velocity component in the y direction at the airfoil surface must be equal in both the compressible and fictitious incompressible flows and that the fictitious airfoil has the same shape as the actual section. We can thus solve for the incompressible potential, phi', then compute the compressible potential, , at the desired location. The velocity perturbation component in the x direction u, is then computed by: where u' is the x velocity perturbation component for the same airfoil in incompressible flow. If one assumes that the velocity perturbations are small compared with the freestream velocity, then the pressure coefficient can be written approximately* as:

Hence, the compressible Cp may be related to the Cp for the same airfoil in incompressible flow by:

This is the Prandtl-Glauert compressibility correction. A variety of alternative compressibility corrections have been proposed. These may include higher order approximations in the Cp formula or an attempt to include some nonlinear effects from the full potential equation. All of these more complex corrections lead to a nonlinear relation between the compressible Cp and the incompressible one. One of the more useful nonlinear relations is the Karman-Tsien compressibility correction which is only slightly more complex than the Prandtl-Glauert rule:

This expression is easily derived by incorporating the increased u perturbation velocity into the second order expression for Cp. In the following plots, the incompressible Cp distributions are compared with those predicted using these two correction rules. Cp's predicted from the full potential equation and from experiments are also shown. Note that both simple rules do quite well until the minimum local Cp approaches the critical value, Cp*. For air with = 1.4:

Plots of the effect of Mach Number on predicted and measured airfoil pressures are given here.

* The linearized expression for Cp in terms of the perturbation velocity u: Cp = - 2u/U is easily derived for incompressible flow, but it also follows from the isentropic expression for Cp. The derivation is not so obvious (see Anderson). This assumption of small perturbations is consistent with the assumptions that led to the PG equation in the first place.

Compressibility and Airfoil Cp

Effect of Mach number on the pressure distribution on an NACA 0012 airfoil. Comparison of various compressibility correction methods.

Transonic Airfoil Pressures

Transonic airfoils are designed to achieve the desired thickness and lift coefficient without developing strong shock waves on the upper surface. The larger the t/c and the larger the Cl, the lower the freestream Mach number must be in order to avoid shocks and the accompanying drag increase. Actually, one need not entirely avoid supersonic flow to avoid high transonic drags. By keeping the maximum local Mach number to something less than about 1.2 and trying to keep the maximum speeds forward of the airfoil "crest", a low drag airfoil with some supercritical flow can be designed. One way to keep the upper surface Cp's from getting too low is to carry positive pressures on the lower surface. This can be done easily near the trailing edge. This technique of aft-loading together with minor changes to the airfoil nose can lead to achievable low drag Mach numbers as much as 0.06 above conventional sections. These are called "supercritical" airfoils and are discussed in more detail in the section on transonic airfoil design. Here we look in a bit more detail at how well the Karman-Tsien correction works at transonic

speeds. PANDA is incompressible with KT correction, FLO-6 is a full potential program.

Supersonic Airfoils
The flow over bodies at supersonic speeds is very different from that at subsonic speeds. As discussed earlier, the differential equations for inviscid flow become hyperbolic, rather than elliptic as the Mach number exceeds 1.0. As disturbances propagate at a speed slower than the freestream speed, waves form.

The angle of the waves generated when the disturbances are small is called the Mach angle:

Sin = a dt / V dt = 1/M

The Mach wave is a boundary between areas that are affected by the presence of very small disturbances and those that are unaffected. A shock wave, however, is produced by larger disturbances. As the flow is compressed, the temperature changes, the speed of sound changes, and the shock angle becomes greater than the Mach angle. When the flow is forced to turn a corner, a shock wave is created. The relationship between the initial Mach number, the Mach number after the shock, the turning angle , and the shock angle , can be computed from the shock jump conditions: relationships derived from continuity, momentum, and energy.

The relationship between the turning angle and the shock angle is given in many textbooks on compressible flow. The important point here is that there are two solutions, one corresponding to weak shock waves and one corresponding to strong shocks. The weak shock solution is the one that actually occurs in most external aerodynamics problems. This solution has a maximum value of and for any incident Mach number. This means that there is a maximum value of turning angle that is possible. Attempting to turn the flow by an angle greater than this results in a detached, bow shock and much higher drag. This is why the leading edges of supersonic airfoils are sharp. These sharp leading edges lead to great simplifications in the analysis of supersonic airfoils because the entire flow may be quite well approximated with small perturbations. Further details are left to courses in compressible flow theory. You should be familiar with oblique shocks, shock jump conditions, Prandtl-Meyer expansions, etc. Additional information on sonic booms and linear interference effects will be included in the next edition of these notes.

Linear Supersonic Theory

Solutions to the Prandtl-Glauert equation in supersonic flow are particularly simple. Since the differential equation is still linear, we can still superimpose known solutions, such as vortices, sources and doublets. For example, a 3-D supersonic source can be described by:

But because of the change in the character of the equation, we can write down a simple, general solution to the equation (in 2-D for now):

One can verify that the solution:

does satisfy the differential equation. Where F and G are arbitrary functions! This solution is very different in character from the subsonic solution because of the fact that disturbances cannot propagate ahead of the wave front. This is known as the law of forbidden signals. The two functions F and G in our solution represent characteristic lines moving away from the upper and lower surfaces of the body respectively. Consider the part that describes the upper surface flow field:

The expressions for the perturbation velocity components are then:

If the boundary condition is approximated by:



Thus, the total velocity at the airfoil surface is given approximately by:

The pressure may then be computed using the compressible Bernoulli equation. Since we have assumed small perturbations in the freestream, the linearized form of the Bernoulli equation is appropriate:

Airfoil Results
The pressure on the surface of the airfoil is thus:

Note that is positive when the flow is being turned away from the freestream direction, and negative when it is being turned back. Since the same solution applies all along the characteristic line, this is the pressure in the flow field as well, with changes due to dispersion and dissipation. Unlike the incompressible case, we require supersonic airfoils to produce small perturbations if they are to satisfy the equations. This means small surface slopes everywhere: no blunt leading edges. This means that we can express the surface slope as:

Note that the Cp equations are linear in , thus the net Cp is the sum of that due to thickness, camber, and angle of attack.

The net lift and moment is then also the sum of those associated with each of the 3 components.

Note that the flat plate at angle of attack is the only component that contributes to the integral. The result for any supersonic airfoil is then:

(The zero lift angle of attack is always 0) The camber line contributes to the moment however with the result that the moment coefficient about the leading edge is:

All three components contribute to the drag:

There are cross terms such as angle of attack * thickness slope, but a useful canceling of cross terms occurs so that the total drag is the sum of the drags of each component (the drag due to a flat plate at angle of attack, the drag of the symmetrical thickness form at zero lift, and the drag of the camber line at zero lift.

The drag of the flat plate is:

This implies that the net force on the plate acts normal to the plate. Well, this seems obvious: we are integrating surface pressures which act normal to the surface. How could it be any other way? In fact how can we explain that in subsonic flow Cd = 0, even for a flat plate at angle of attack? The reason that we can get zero drag in subsonic flow is that for thick airfoils, there is indeed some surface area facing in the freestream direction.

As the airfoil is made thinner, this forward facing area gets smaller, but the Cp gets more negative (a pressure peak develops). In the limit as t/c goes to 0, a singularity develops. The combined effects always cancel, so Cd = 0. For supersonic sections no singularity can form; in fact, the pressure peak has a finite limit, so Cd is greater than zero. The forward facing pressure component is called leading edge suction and the amount of leading edge suction that can be achieved depends on the angle of attack, Mach number, and sharpness

of the leading edge. To appreciate the effect of airfoil shape and angle of attack on performance at supersonic speeds, try changing the airfoil below and examine the effect on lift, moment, and L/D. Click on the upper half of the plot to increase angle of attack and on the lower half to decrease it. The pitching moment is measured about the 50% chord point. You may also visualize complete flow field pressures by clicking the second button, but note that this may take a while on slower computers.

Viscosity and Boundary Layers

This chapter deals with the effects of viscosity in two dimensions. The sections describe the basic phenomena and some simple theory that may be used to estimate boundary layer properties. Boundary layers appear on the surface of bodies in viscous flow because the fluid seems to "stick" to the surface (*see note). Right at the surface the flow has zero relative speed and this fluid transfers momentum to adjacent layers through the action of viscosity. Thus a thin layer of fluid with lower velocity than the outer flow develops. The requirement that the flow at the surface has no relative motion is the "no slip condition." The velocity in the boundary layer slowly increases until it reaches the outer flow velocity, Ue. The boundary layer thickness, , is defined as the distance required for the flow to nearly reach Ue. We might take an arbitrary number (say 99%) to define what we mean by "nearly", but certain other definitions are used most frequently. (see theory section).

The boundary layer concept is attributed primarily to Ludwig Prandtl (1874-1953), a professor at the University of Gottingen. His 1904 paper on the subject formed the basis for future work on skin friction, heat transfer, and separation. He subsequently made fundamental contributions to finite wing theory and compressibility effects. (His name appears about 30 times in these notes.) Theodore von Karman and Max Munk were among his many famous students. R.T. Jones was a student of Max Munk and I have subsequently learned a great deal from R.T. Jones --

which makes readers of these notes great-great grandstudents of Prandtl. The character of the boundary layer changes as it develops along the surface of the airfoil. Generally starting out as a laminar flow, the boundary layer thickens, undergoes transition to turbulent flow, and then continues to develop along the surface of the body, possibly separating from the surface under certain conditions.

In laminar flow, the fluid moves in smooth layers or lamina. There is relatively little mixing and consequently the velocity gradients are small and shear stresses are low. The thickness of the laminar boundary layer increases with distance from the start of the boundary layer and decreases with Reynolds number.

As the fluid is sheared across the surface of the body, instabilities develop and eventually the flow transitions into turbulent motion. Turbulent boundary layer flow is characterized by unsteady mixing due to eddies at many scales. The result is higher shear stress at the wall, a "fuller" velocity profile,and a greater boundary layer thickness. The wall shear stress is higher because the velocity gradient near the wall is greater. This is because of the more effective mixing associated with turbulent flow. However, the lower velocity fluid is also transported outward with the result that the distance to the edge of the layer is larger.

Several fundamental effects are produced by viscosity: Drag: Skin friction drag caused by shear stresses at the surface contribute a majority of the drag of most airplanes. The pressure distribution is changed by the presence of a boundary layer, even when no significant separation is present. This changes CL and Cm. Flow separation: Viscosity is responsible for flow separation which causes major changes to the flow patterns and pressures.

To compute these characteristics some basic boundary layer theory is described here with more detailed computational methods for laminar and turbulent boundary layers. *Actually, the zero slip condition at the surface arises from the roughness of the surface on a molecular scale. Fluid molecules hitting the surface impart a net momentum to the surface and the mean velocity of molecules hitting the surface is about the same as the surface velocity. Even when the surface is extremely smooth, electrostatic forces exist between the surface and the air molecules, introducing the shear stress at the surface. If this interaction could be reduced, a reduction in skin friction would result, but no one has found a way to do this. (Yet.)

Viscous Drag
Skin Friction The shearing stresses at the surface of a body produce skin friction drag. We define the skin friction coefficient, Cf, by:

The shear stress is then related to the viscosity by:

Cf is related to the drag coefficient by CD (skin friction) = Cf*Swetted/Sref. where Swetted is the area "wetted" by the air and Sref is the reference area used to define the drag coefficient. This expression applies to a flat plate. When the body has thickness, the local velocities on the surface may be higher than the freestream velocity and the skin friction is increased. We usually write: CD = k * Cf * Swetted / Sref where k is a "form factor" that depends on the shape of the body. The skin friction coefficient varies with Reynolds number, Mach number, and the character of the boundary layer. The momentum transferred between the air and the body surface appears as a velocity deficit in the viscous wake behind the body. The plot below shows how Reynolds number and the location of the transition from laminar flow to turbulent flow, affects the skin friction coefficient.

From the basic boundary layer theory combined with experimental fits, the following results are obtained: For laminar boundary layers on flat plates:

For fully-turbulent flat plate boundary layers:

Pressure Drag In addition to direct skin friction, the presence of the boundary layer creates a pressure or form drag on bodies. This does not appear in the flat plate results because pressure always acts perpendicular to the drag direction in this case. In an adverse pressure gradient, the skin friction drag is reduced, but pressure drag increases. This increase in pressure drag compensates for some of the reduction in skin friction. The combined drag may be estimated by a handy expression derived by Squire and Young (see Thwaites, Incompressible Aerodynamics) and gives amazingly good estimates of the total profile drag:

where is nondimensionalized by the chord length and the velocity outside the boundary layer at the trailing edge, Ue is normalized by the freestream U. Hte is the shape factor of the boundary layer at the trailing edge. (See the section on Boundary Layer Theory) Note that when Ue is 1.0, the drag is just twice the momentum thickness on upper and lower surfaces. When H = 2 and Ue = 0.9 (Cp = .2), CDu = 1.38

Effect of Boundary Layers on Pressures

In addition to direct skin friction drag, the presence of the boundary layer changes the effective shape of the body, leading to changes in the pressure distribution and to the overall lift and drag.

The effective shape can be used to approximate the effect of the boundary layer using inviscid analysis methods combined with the boundary layer equations. Outside the boundary layer, the flow behaves much like an inviscid (and usually irrotational) fluid. Thus, we can use the simpler analysis methods outside the boundary layer, and if we could compute the streamlines just outside of the boundary layer we could use these as boundary conditions. (This is actually easier said than done as the coupled boundary layer -- inviscid solutions are poorly conditioned numerically.)

This leads to changes in the lift, drag, and moment compared with the inviscid solution. Note that this change in pressure distribution leads to a non-zero pressure drag in addition to the skin friction drag discussed previously. The sum of the skin friction and pressure drag is often termed "profile drag." As the angle of attack changes, the boundary layer shape changes, with thicker boundary layers developing toward the aft part of the airfoil at higher angles of attack (because of the more severe adverse pressure gradients).

The effective shape of the airfoil thus changes with angle of attack. If we look at the mean line of the effective shapes, it is clear that viscous effects cause an effective decambering of the airfoil shape. This leads to changes in the lift curve slope (up to a 10% reduction in Cl at Reynolds numbers in the millions) and an aerodynamic center that is usually farther forward than is predicted by inviscid theory.

The effect is of increasing importance as Reynolds number is reduced.

When the flow near the surface reverses its direction and flows upstream, there must be a place, generally a bit farther upstream, where streamlines meet and then leave the surface. This is separation and it is caused by the presence of an adverse pressure gradient. When this occurs, the assumptions that the u component of velocity is larger than the v component and that certain derivatives in the x direction may be ignored, no longer are valid. Thus, coupling an inviscid analysis with a simple boundary layer calculation does not work. One must resort to experiment or Navier-Stokes solutions. The changes in the flow pattern, and associated forces and moments are large. Drag usually increases substantially and airfoil lift usually drops. The effect is generally Reynolds number


The figure above shows how the flow pattern and pressure distribution is affected by separation. On the left, the pressures are modified slightly by the boundary layer; in the center image, separation near the trailing edge has reduced the Cp and lift; leading edge separation dramatically reduces the suction peaks and reduces lift.

The presence of an adverse pressure gradient (increasing pressure) causes a deceleration of the fluid. Just as when one coasts uphill, the fluid that starts up the (pressure) hill with little speed, starts rolling backward after a while.

This picture explains why flow does not separate as readily at higher Reynolds numbers. In that case, the velocity profile is "fuller" with the high external velocities extending down closer to the surface. Turbulent boundary layers also have greater velocity near the surface and are therefore better able to handle adverse pressure gradients. Since the velocity near the surface in a laminar boundary layer has lower velocity than its turbulent counterpart, the laminar boundary layer is more likely to separate. When this occurs, the laminar boundary layer leaves the surface and usually undergoes transition to turbulent flow away from the surface. This process takes place over a certain distance that is inversely related to the Reynolds number, but if it happens quickly enough, the flow may reattach as a turbulent boundary layer and continue along the surface.

This phenomenon has significant effects on airfoil pressure distributions at low Reynolds numbers.

To compute when separation will occur, we can solve the Navier-Stokes equations or apply one of several separation criteria to solutions of the boundary layer equations. Laminar Separation Criteria Turbulent Separation Criteria

Boundary Layer Theory and Definitions

Boundary Layer Thickness The boundary layer thickness, , is defined as the distance required for the flow to nearly reach Ue. We might take an arbitrary number (say 99%) to define what we mean by "nearly", but certain other definitions are used most frequently. The displacement thickness and momentum thickness are alternative measures of boundary layer thickness and are used in the calculation of various boundary layer properties. The displacement thickness is defined by considering the total mass flow through the boundary layer. This mass flow is the same as if the boundary layer were completely at rest, with a thickness, *:

For laminar boundary layers * is about one third of the distance to the edge of the boundary layer, . The momentum thickness, , is defined similarly, using the momentum flux rather than the mass flux:

For laminar boundary layers, tends to be about an order of magnitude greater than . The ratio of * to is termed the shape factor, H:

The Boundary Layer Equations Newton's law applied to a fluid element in 2-D is:

and, in the x direction:

This leads, directly to:

This boundary layer equation is combined with the continuity equation and dp/dy = 0 to obtain the 3 equations in the unknowns: p, u, and v. One of the basic results (assumptions?) in this development of boundary layer theory is that the static pressure is constant through the boundary layer (dp/dy = 0). This is a rather important result, but it is not exactly true*. We start with the 2-D boundary layer equation shown above. For steady flow, this reduces to:

The approach to the solution of this equation is to assume that the pressure does not vary with y, so it is specified by the external velocity distribution. v is computed from the continuity equation, leaving a PDE in u to be integrated. However, this holds only for laminar flow since turbulent boundary layers are inherently unsteady. Subsequent sections deal with the solution of these equations in more detail.

*If one considers the balance of normal forces on a fluid element we can see that dp/dn = V2/R where R is the radius of curvature. This holds even for viscous fluids since we expect viscosity to produce shear stresses and not normal stresses. Now, if Cp = (p-p0)/(.5 U02) then, dCp/dn = 2 (V/U0)2 / R Thus, the change in Cp through the boundary layer is: Cp ~ /R * (V/U0)2 So, as long as the radius of curvature is much larger than the boundary layer thickness, and the local velocities are not too large, this is true. Such conditions are not always met, though!

Laminar Boundary Layer Theory

Flat Plate Flow From the previous section the boundary layer equations reduce to the following when the boundary layer is assumed to be steady:

This is especially simple in the case of incompressible flow with no pressure gradient. This laminar, flat plate, boundary layer flow satisfies:

These equations are easily solved by introducing the variable :

also introducing the function: From continuity, then:

and the boundary layer equation becomes: f''' + ff'' = 0 This ordinary differential equation is accompanied by the boundary conditions which state that the velocity right at the surface is 0 and that far away the velocity approaches the specified Ue. These B.C.'s

are written: f = f' = 0 at h = 0 f' = 2 at h = infinity The problem was first formulated in this way by Blasius in 1908. The equation seems simple, but it is nonlinear and no closed form solution is known. The problem can be solved by assuming a series approximation for f. When all is said and done, the following relations are found:

Note that Cf is about twice the momentum thickness for a flat plate. Is this what you would expect?

Thwaites Method Thwaites method is used for computing the boundary layer characteristics in laminar flow with pressure gradients. It is based on the steady boundary layer equations with a specified external pressure gradient, and gives an approximate solution. Starting with the Navier-Stokes Equations we make the usual boundary layer assumptions, that the flow is steady, incompressible, and we ignore higher order terms (See Kuethe and Chow pg. 461 or 330 or 314) In the x-direction:

This may be rewritten in terms of the boundary layer variables, , H, Ue, and Cfl:

where is the momentum thickness:

H is the shape factor:

and Cfl is the local skin friction coefficient:

The basic idea behind many laminar boundary layer methods method is to assume a particular form for the variation of u with y: i.e. u/ue = f(y). Pohlhausen's method is based on a quartic polynomial for u(y). Thwaites' method is not quite so direct. It uses results from exact solutions to certain laminar boundary layers to obtain an approximate relationship between H, Cf, Re, and .

Substituting these results for and H into the expression above Thwaites obtained:

This is easily integrated for (x).

The boundary layer changes character from smooth, steady motion (laminar flow) to turbulent motion with unsteadiness and eddies on a very small scale. Several things may be responsible for the transition between these two states. These include: surface roughness, adverse pressure gradients, and wing sweep. NASA TP 2256 contains flight data on the allowable surface roughness and waviness for laminar flow. NACA TN 4363, A Simplified Method for Determination of Critical Height of Distributed Roughness Particles for Boundary Layer Transition, contains another discussion. A rule of thumb from Hoerner is that a roughness element should be at least 25% of the boundary layer displacement thickness to cause transition, although this depends on the form of the roughness elements. Two-dimensional trips, such as tape strips must be much larger than 3-D trips. Wing sweep also encourages transition through the cross-flow instability. Some discussion appears in the above NASA TP, but methods for predicting this are, in general, very complex. People often run large computer codes that take quite some time to master and their results are often questioned. The usual reason for transition on airfoils is the presence of an adverse pressure gradient. (Increasing Cp with x.) When this becomes severe enough, or long enough, transition is likely to

occur. As a rule of thumb, boundary layers at high Reynolds numbers can withstand very little adverse gradient and one may assume that once the gradient is adverse transition will occur. On a more quantitative basis, several transition criteria have been proposed and are in wide use. These are empirically-based and work over a restricted Reynolds number range. Two of the most popular are Michel's and Granville's criteria. Michel's criteria states that transition will occur when the Reynolds number based on momentum thickness exceeds a certain value which depends on the local Reynolds number:

The relationship holds for Rex values between 105 and 40x106. Unfortunately, in practice, the left side increases at about the same rate as the right hand side, making the determination of the transition location inaccurate. But when an adverse gradient exists, the left side grows much more rapidly and transition is reasonably predicted. (Note that when gradients are not large transition can be expected at values of Rex near 3 million. This is very much a ballpark estimate, of course.)

A somewhat more refined approach to transition prediction involves an analysis of the stability of the boundary layer equations to disturbances. The basic ideas for this approach were described by Osborne Reynolds (of Reynolds number fame) and Lord Rayleigh (of Rayleigh number fame). Their hypothesis was that, above a critical Reynolds number, small disturbances to the laminar boundary layer, are not damped by viscosity but rather amplified until the laminar character of the flow disappears. One starts with the unsteady Navier-Stokes equations and assumes that the unsteady terms are very small and are written as a Fourier series with unknown coefficients. Upon substitution of these equations into the unsteady part of the NS equations, one obtains a 4th order, linear ODE, known as the OrrSommerfeld equation. If the mean velocity profile and Reynolds number are specified, this leads to an eigenvalue problem for each assumed value of a disturbance wavelength. In 1957, A.M.O. Smith did several experiments to correlate the location of transition with the value of the real part of this eigenvalue. He found that transition generally occurred when the disturbances had grown by about 8000 times (e9) their value at the point of neutral stability. This sort of linear stability theory (en method) is now routinely used to estimate transition location with n between 9 and 11.

Turbulent Boundary Layer Theory

The equations for turbulent flow are the full Navier-Stokes equations. We can simplify the equations by using empirical models for the more complex aspects of the turbulence and making the assumption of a thin layer. The method described here is used to compute the characteristics of a "steady" turbulent boundary layer with a pressure gradient. In this case we solve two coupled first order ODE's. The first is the von Karman integral equation:

where and Ue are made dimensionless with the chord length and freestream velocity and H is the shape factor, */. The second is an expression describing the entrainment of flow into the boundary layer:

Here, H1 is the mass flow shape factor:

and F is an entrainment parameter. These variables are related to each other by the following expressions. The skin friction coefficient is related to H and by the Ludwieg-Tillman skin friction law:

The mass flow shape factor, H1 can be related to H by the following fits:

Finally, the entrainment parameter, F, is approximated by:

We then just integrate this system of equation numerically to obtain the variation of and H with

position along the airfoil.

These expressions must be evaluated numerically, but some useful results for flat plates are given below. (Re = 0.5 to 10 million):

The formulas involving Re.2 are based on the assumption of a 1/7th power law shape for the boundary layer profile u/Ue = (y/)1/7. Results agree with experiments up to Re = 20 million. Above this, the skin friction is underestimated. The results based on the logarithmic distribution agree well with limited test data out to Re = 500 million. In many cases, the flow starts laminar, undergoes transition, and continues as a turbulent boundary layer. The details of the transition process are still a matter of current research, but we may estimate the gross features of the flow by matching assuming a virtual start of the turbulent layer upstream of transition so that the momentum thicknesses of the laminar and turbulent layers match at the transition "point".

Boundary Layer Theory and Definitions

Boundary Layer Thickness The boundary layer thickness, , is defined as the distance required for the flow to nearly reach Ue. We might take an arbitrary number (say 99%) to define what we mean by "nearly", but certain other definitions are used most frequently. The displacement thickness and momentum thickness are alternative measures of boundary layer thickness and are used in the calculation of various boundary layer properties. The displacement thickness is defined by considering the total mass flow through the boundary layer. This mass flow is the same as if the boundary layer were completely at rest, with a thickness, *:

For laminar boundary layers * is about one third of the distance to the edge of the boundary layer, .

The momentum thickness, , is defined similarly, using the momentum flux rather than the mass flux:

For laminar boundary layers, tends to be about an order of magnitude greater than . The ratio of * to is termed the shape factor, H:

The Boundary Layer Equations Newton's law applied to a fluid element in 2-D is:

and, in the x direction:

This leads, directly to:

This boundary layer equation is combined with the continuity equation and dp/dy = 0 to obtain the 3 equations in the unknowns: p, u, and v. One of the basic results (assumptions?) in this development of boundary layer theory is that the static pressure is constant through the boundary layer (dp/dy = 0). This is a rather important result, but it is not exactly true*. We start with the 2-D boundary layer equation shown above. For steady flow, this reduces to:

The approach to the solution of this equation is to assume that the pressure does not vary with y, so it is specified by the external velocity distribution. v is computed from the continuity equation, leaving a PDE in u to be integrated. However, this holds only for laminar flow since turbulent boundary layers are inherently unsteady. Subsequent sections deal with the solution of these equations in more detail.

*If one considers the balance of normal forces on a fluid element we can see that dp/dn = V2/R where R is the radius of curvature. This holds even for viscous fluids since we expect viscosity to produce shear stresses and not normal stresses. Now, if Cp = (p-p0)/(.5 U02)

then, dCp/dn = 2 (V/U0)2 / R Thus, the change in Cp through the boundary layer is: Cp ~ /R * (V/U0)2 So, as long as the radius of curvature is much larger than the boundary layer thickness, and the local velocities are not too large, this is true. Such conditions are not always met, though!

Laminar Boundary Layer Theory

Flat Plate Flow From the previous section the boundary layer equations reduce to the following when the boundary layer is assumed to be steady:

This is especially simple in the case of incompressible flow with no pressure gradient. This laminar, flat plate, boundary layer flow satisfies:

These equations are easily solved by introducing the variable :

also introducing the function: From continuity, then:

and the boundary layer equation becomes: f''' + ff'' = 0 This ordinary differential equation is accompanied by the boundary conditions which state that the velocity right at the surface is 0 and that far away the velocity approaches the specified Ue. These B.C.'s are written: f = f' = 0 at h = 0 f' = 2 at h = infinity The problem was first formulated in this way by Blasius in 1908. The equation seems simple, but it is nonlinear and no closed form solution is known. The problem can

be solved by assuming a series approximation for f. When all is said and done, the following relations are found:

Note that Cf is about twice the momentum thickness for a flat plate. Is this what you would expect?

Thwaites Method Thwaites method is used for computing the boundary layer characteristics in laminar flow with pressure gradients. It is based on the steady boundary layer equations with a specified external pressure gradient, and gives an approximate solution. Starting with the Navier-Stokes Equations we make the usual boundary layer assumptions, that the flow is steady, incompressible, and we ignore higher order terms (See Kuethe and Chow pg. 461 or 330 or 314) In the x-direction:

This may be rewritten in terms of the boundary layer variables, , H, Ue, and Cfl:

where is the momentum thickness:

H is the shape factor:

and Cfl is the local skin friction coefficient:

The basic idea behind many laminar boundary layer methods method is to assume a particular form for the variation of u with y: i.e. u/ue = f(y). Pohlhausen's method is based on a quartic polynomial for u(y).

Thwaites' method is not quite so direct. It uses results from exact solutions to certain laminar boundary layers to obtain an approximate relationship between H, Cf, Re, and .

Substituting these results for and H into the expression above Thwaites obtained:

This is easily integrated for (x).

The boundary layer changes character from smooth, steady motion (laminar flow) to turbulent motion with unsteadiness and eddies on a very small scale. Several things may be responsible for the transition between these two states. These include: surface roughness, adverse pressure gradients, and wing sweep. NASA TP 2256 contains flight data on the allowable surface roughness and waviness for laminar flow. NACA TN 4363, A Simplified Method for Determination of Critical Height of Distributed Roughness Particles for Boundary Layer Transition, contains another discussion. A rule of thumb from Hoerner is that a roughness element should be at least 25% of the boundary layer displacement thickness to cause transition, although this depends on the form of the roughness elements. Two-dimensional trips, such as tape strips must be much larger than 3-D trips. Wing sweep also encourages transition through the cross-flow instability. Some discussion appears in the above NASA TP, but methods for predicting this are, in general, very complex. People often run large computer codes that take quite some time to master and their results are often questioned. The usual reason for transition on airfoils is the presence of an adverse pressure gradient. (Increasing Cp with x.) When this becomes severe enough, or long enough, transition is likely to occur. As a rule of thumb, boundary layers at high Reynolds numbers can withstand very little adverse gradient and one may assume that once the gradient is adverse transition will occur. On a more quantitative basis, several transition criteria have been proposed and are in wide use.

These are empirically-based and work over a restricted Reynolds number range. Two of the most popular are Michel's and Granville's criteria. Michel's criteria states that transition will occur when the Reynolds number based on momentum thickness exceeds a certain value which depends on the local Reynolds number:

The relationship holds for Rex values between 105 and 40x106. Unfortunately, in practice, the left side increases at about the same rate as the right hand side, making the determination of the transition location inaccurate. But when an adverse gradient exists, the left side grows much more rapidly and transition is reasonably predicted. (Note that when gradients are not large transition can be expected at values of Rex near 3 million. This is very much a ballpark estimate, of course.)

A somewhat more refined approach to transition prediction involves an analysis of the stability of the boundary layer equations to disturbances. The basic ideas for this approach were described by Osborne Reynolds (of Reynolds number fame) and Lord Rayleigh (of Rayleigh number fame). Their hypothesis was that, above a critical Reynolds number, small disturbances to the laminar boundary layer, are not damped by viscosity but rather amplified until the laminar character of the flow disappears. One starts with the unsteady Navier-Stokes equations and assumes that the unsteady terms are very small and are written as a Fourier series with unknown coefficients. Upon substitution of these equations into the unsteady part of the NS equations, one obtains a 4th order, linear ODE, known as the OrrSommerfeld equation. If the mean velocity profile and Reynolds number are specified, this leads to an eigenvalue problem for each assumed value of a disturbance wavelength. In 1957, A.M.O. Smith did several experiments to correlate the location of transition with the value of the real part of this eigenvalue. He found that transition generally occurred when the disturbances had grown by about 8000 times (e9) their value at the point of neutral stability. This sort of linear stability theory (en method) is now routinely used to estimate transition location with n between 9 and 11.

Turbulent Boundary Layer Theory

The equations for turbulent flow are the full Navier-Stokes equations. We can simplify the equations by using empirical models for the more complex aspects of the turbulence and making the assumption of a thin layer. The method described here is used to compute the characteristics of a "steady" turbulent boundary layer with a pressure gradient.

In this case we solve two coupled first order ODE's. The first is the von Karman integral equation:

where and Ue are made dimensionless with the chord length and freestream velocity and H is the shape factor, */. The second is an expression describing the entrainment of flow into the boundary layer:

Here, H1 is the mass flow shape factor:

and F is an entrainment parameter. These variables are related to each other by the following expressions. The skin friction coefficient is related to H and by the Ludwieg-Tillman skin friction law:

The mass flow shape factor, H1 can be related to H by the following fits:

Finally, the entrainment parameter, F, is approximated by:

We then just integrate this system of equation numerically to obtain the variation of and H with position along the airfoil.

These expressions must be evaluated numerically, but some useful results for flat plates are given below. (Re = 0.5 to 10 million):

The formulas involving Re.2 are based on the assumption of a 1/7th power law shape for the boundary layer profile u/Ue = (y/)1/7. Results agree with experiments up to Re = 20 million. Above this, the skin friction is underestimated. The results based on the logarithmic distribution agree well with limited test data out to Re = 500 million. In many cases, the flow starts laminar, undergoes transition, and continues as a turbulent boundary layer. The details of the transition process are still a matter of current research, but we may estimate the gross features of the flow by matching assuming a virtual start of the turbulent layer upstream of transition so that the momentum thicknesses of the laminar and turbulent layers match at the transition "point".

Summary of Boundary Layer Results

Here are some useful expressions for flat-plate boundary layers: Laminar


Airfoils, Part II: Design

This chapter is an introduction to the problem of airfoil design. Some typical design goals and constraints are discussed, along with a brief discussion of design methodology. 1. Design Methods and Objectives

2. Some Typical Design Problems 3. High Lift Systems 4. References

Airfoil Design Methods

The process of airfoil design proceeds from a knowledge of the boundary layer properties and the relation between geometry and pressure distribution. The goal of an airfoil design varies. Some airfoils are designed to produce low drag (and may not be required to generate lift at all.) Some sections may need to produce low drag while producing a given amount of lift. In some cases, the drag doesn't really matter -- it is maximum lift that is important. The section may be required to achieve this performance with a constraint on thickness, or pitching moment, or off-design performance, or other unusual constraints. Some of these are discussed further in the section on historical examples. One approach to airfoil design is to use an airfoil that was already designed by someone who knew what he or she was doing. This "design by authority" works well when the goals of a particular design problem happen to coincide with the goals of the original airfoil design. This is rarely the case, although sometimes existing airfoils are good enough. In these cases, airfoils may be chosen from catalogs such as Abbott and von Doenhoff's Theory of Wing Sections, Althaus' and Wortmann's Stuttgarter Profilkatalog, Althaus' Low Reynolds Number Airfoil catalog, or Selig's "Airfoils at Low Speeds". The advantage to this approach is that there is test data available. No surprises, such as a unexpected early stall, are likely. On the other hand, available tools are now sufficiently refined that one can be reasonably sure that the predicted performance can be achieved. The use of "designer airfoils" specifically tailored to the needs of a given project is now very common. This section of the notes deals with the process of custom airfoil design. Methods for airfoil design can be classified into two categories: direct and inverse design. Direct Methods for Airfoil Design The direct airfoil design methods involve the specification of a section geometry and the calculation of pressures and performance. One evaluates the given shape and then modifies the shape to improve the performance. The two main subproblems in this type of method are 1. the identification of the measure of performance

2. the approach to changing the shape so that the performance is improved The simplest form of direct airfoil design involves starting with an assumed airfoil shape (such as a NACA airfoil), determining the characteristic of this section that is most problemsome, and fixing this problem. This process of fixing the most obvious problems with a given airfoil is repeated until there is no major problem with the section. The design of such airfoils, does not require a specific definition of a scalar objective function, but it does require some expertise to identify the potential problems and often considerable expertise to fix them. Let's look at a simple (but real life!) example.

A company is in the business of building rigid wing hang gliders and because of the low speed requirements, they decide to use a version of one of Bob Liebeck's very high lift airfoils. Here is the pressure distribution at a lift coefficient of 1.4. Note that only a small amount of trailing edge separation is predicted. Actually, the airfoil works quite well, achieving a Clmax of almost 1.9 at a Reynolds number of one million.

This glider was actually built and flown. It, in fact, won the 1989 U.S. National Championships. But it had terrible high speed performance. At lower lift coefficients the wing seemed to fall out of the sky. The plot below shows the pressure distribution at a Cl of 0.6. The pressure peak on the lower surface causes separation and severely limits the maximum speed. This is not too hard to fix.

By reducing the lower surface "bump" near the leading edge and increasing the lower surface thickness aft of the bump, the pressure peak at low Cl is easily removed. The lower surface flow is now attached, and remains attached down to a Cl of about 0.2. We must check to see that we have not hurt the Clmax too much.

Here is the new section at the original design condition (still less than Clmax). The modification of the lower surface has not done much to the upper surface pressure peak here and the Clmax turns out to be changed very little. This section is a much better match for the application and demonstrates how effective small modifications to existing sections can be. The new version of the glider did not use this section, but one that was designed from scratch with lower drag.

Sometimes the objective of airfoil design can be stated more positively than, "fix the worst things". We might try to reduce the drag at high speeds while trying to keep the maximum CL greater than a certain value. This could involve slowly increasing the amount of laminar flow at low Cl's and checking to see the effect on the maximum lift. The objective may be defined numerically. We could actually minimize Cd with a constraint on Clmax. We could maximize L/D or Cl1.5/Cd or Clmax / Cd@Cldesign. The selection of the figure of merit for airfoil sections is quite important and generally cannot be done without considering the rest of the airplane. For example, if we wish to build an airplane with maximum L/D we do not build a section with maximum L/D because the section Cl for best Cl/Cd is different from the airplane CL for best CL/CD.
Inverse Design Another type of objective function is the target pressure distribution. It is sometimes possible to specify a desired Cp distribution and use the least squares difference between the actual and target Cp's as the objective. This is the basic idea behind a variety of methods for inverse design. As an example, thin airfoil theory can be used to solve for the shape of the camberline that produces a specified pressure difference on an airfoil in potential flow. The second part of the design problem starts when one has somehow defined an objective for the airfoil design. This stage of the design involves changing the airfoil shape to improve the performance. This

may be done in several ways: 1. By hand, using knowledge of the effects of geometry changes on Cp and Cp changes on performance. 2. By numerical optimization, using shape functions to represent the airfoil geometry and letting the computer decide on the sequence of modifications needed to improve the design.

Thick Airfoil Design

The difficulty with thick airfoils is that the minimum pressure is decreased due to thickness. This results in a more severe adverse pressure gradient and the need to start recovery sooner. If the maximum thickness point is specified, the section with maximum thickness must recover from a given point with the steepest possible gradient. This is just the sort of problem addressed by Liebeck in connection with maximum lift. The thickest possible section has a boundary layer just on the verge of separation throughout the recovery. The thickest section at Re = 10 million is 57% thick, but of course, it will separate suddenly with any angle of attack.

High Lift Airfoil Design

To produce high lift coefficients, we require very negative pressures on the upper surface of the

airfoil. The limit to this suction may be associated with compressibility effects, or may be imposed by the requirement that the boundary layer be capable of negotiating the resulting adverse pressure recovery. It may be shown that to maximize lift starting from a specified recovery height and location, it is best to keep the boundary layer on the verge of separation*. Such distributions are shown below for a Re of 5 million. Note the difference between laminar and turbulent results. The thickest section at Re = 10 million is 57% thick, but of course, it will separate suddenly with any angle of attack.

For maximum airfoil lift, the best recovery location is chosen and the airfoil is made very thin so that the lower surface produces maximum lift as well. (Since the upper surface Cp is specified, increasing thickness only reduces the lower surface pressures.) Well, almost. If the upper surface Cp is more negative than -3.0, the perturbation velocity is greater than freestream, which means, for a thin section, the lower surface flow is upstream. This would cause separation and the maximum lift is achieved with an upper surface velocity just over 2U and a bit of thickness to keep the lower surface near stagnation pressure.

A more detailed discussion of this topic may be found in the section on high lift systems. *This conclusion, described by Liebeck, is easily derived if Stratford's criterion or the laminar boundary layer method of Thwaites is used. For other turbulent boundary layer criteria, the conclusion is not at all obvious and is almost surely not generally valid.

Laminar Airfoil Design

As discussed in the section on boundary layers, laminar flow may be useful for reducing skin friction drag, increasing maximum lift, or reducing heat transfer. It may be achieved without too much work at low Reynolds numbers by maintaining a smooth surface and using an airfoil with a favorable pressure gradient. The section below shows how the pressures may be tailored to achieve long runs of laminar flow on upper and/or lower surfaces. Again, the Stratford-like pressure recovery is helpful in achieving the maximum run of favorable gradient on either upper or lower surfaces.

Transonic Airfoil Design

The transonic airfoil design problem arises because we wish to limit shock drag losses at a given transonic speed. This effectively limits the minimum pressure coefficient that can be tolerated. Since both lift and thickness reduce (increase in magnitude) the minimum Cp, the transonic design problem is to create an airfoil section with high lift and/or thickness without causing strong shock waves. One can generally tolerate some supersonic flow without drag increase, so

that most sections can operate efficiently as "supercritical airfoils". A rule of thumb is that the maximum local Mach numbers should not exceed about 1.2 to 1.3 on a well-designed supercritical airfoil. This produces a considerable increase in available Cl compared with entirely subcritical designs. Supercritical sections usually refer to a special type of airfoil that is designed to operate efficiently with substantial regions of supersonic flow. Such sections often take advantage of many of the following design ideas to maximize lift or thickness at a given Mach number:

Carry as much lift as is practical on the aft potion of the section where the flow is subsonic. The aft lower surface is an obvious candidate for increased loading (more positive Cp), although several considerations discussed below limit the extent to which this approach can be used. Make sure that sufficient lift is carried on the forward portion of the upper surface. As the Mach number increases, the pressure peak near the nose is diminished and without additional blunting of the nose, possible extra lift will be lost in this region. The lower surface near the nose can also be loaded by reducing the lower surface thickness near the leading edge. This provides both lift and positive pitching moment. Shocks on the upper surface near the leading edge produce much less wave drag than shocks aft of the airfoil crest and it is feasible, although not always best, to design sections with forward shocks. Such sections are known as "peaky" airfoils and were used on many transport aircraft. The idea of carefully tailoring the section to obtain locally supersonic flow without shockwaves (shock-free sections) has been pursued for many years, and such sections have been designed and tested. For most practical cases with a range of design C L and Mach number, sections with weak shocks are favored.

One must be cautious with supercritical airfoil design. Several of these sections have looked promising initially, but led to problems when actually incorporated into an aircraft design. Typical difficulties include the following.

Too much aft loading can produce large negative pitching moments with trim drag and structural weight penalties. The adverse pressure gradient on the aft lower surface can produce separation in extreme cases. The thin trailing edge may be difficult to manufacture. Supercritical, and especially shock-free designs often are very sensitive to Mach and CL and may perform poorly at off-design conditions. The appearance of "drag creep" is quite common, a situation in which substantial section drag increase with Mach number occurs even at speeds below the design value.

The section with pressures shown below is typical of a modern supercritical section with a weak shock at its design condition. Note the rooftop Cp design with the minimum Cp considerably greater above Cp*.

Low Reynolds Number Airfoil Design

Low Reynolds numbers make the problem of airfoil design difficult because the boundary layer is much less capable of handling an adverse pressure gradient without separation. Thus, very low Reynolds number designs do not have severe pressure gradients and the maximum lift capability is restricted. Low Reynolds number airfoil designs are cursed with the problem of too much laminar flow. It is sometimes difficult to assure that the boundary layer is turbulent over the steepest pressure recovery regions. Laminar separation bubbles are common and unless properly stabilized can lead to excessive drag and low maximum lift. At very low Reynolds numbers, most or all of the boundary layer is laminar. Under such conditions the boundary layer can handle only gradual pressure recovery. Based on the expressions for laminar separation, one finds that an all-laminar section can generate a CL of about 0.4 or achieve a thickness of about 7.5%. (Try this with a simple airfoil design code such as PANDA.)

Low Moment Airfoil Design

When the airfoil pitching moment is constrained, it is not always possible to carry lift as far back on the airfoil as desired. Such situations arise in the design of sections for tailless aircraft, helicopter rotor blades, and even sails, kites, and giant pterosaurs. The airfoil shown here is a Liebeck section designed to perform well at low Reynolds numbers with a positive Cm0. Its performance is not bad, but it is clearly inferior in Clmax when compared to other sections without a Cm constraint. (Clmax = 1.35 vs. 1.60 for conventional sections at Re = 500,000.)

Multiple Design Point Airfoils

One of the difficulties in designing a good airfoil is the requirement for acceptable off-design performance. While a very low drag section is not too hard to design, it may separate at angles of attack slightly away from its design point. Airfoils with high lift capability may perform very poorly at lower angles of attack. One can approach the design of airfoil sections with multiple design points in a well-defined way. Often it is clear that the upper surface will be critical at one of the points and we can design the upper surface at this condition. The lower surface can then be designed to make the section behave properly at the second point. Similarly, constraints such as Cm0 are most effected by airfoil trailing edge geometry. When such a compromise is not possible, variable geometry can be employed (at some expense) as in the case of high lift systems.

High Lift Systems

A wing designed for efficient high-speed flight is often quite different from one designed solely for take-off and landing. Take-off and landing distances are strongly influenced by aircraft stalling speed, with lower stall speeds requiring lower acceleration or deceleration and correspondingly shorter field lengths. It is always possible to reduce stall speed by increasing wing area, but it is not desirable to cruise with hundreds of square feet of extra wing area (and the associated weight and drag), area that is only needed for a few minutes. Since the stalling speed is related to wing parameters by:

It is also possible to reduce stalling speed by reducing weight, increasing air density, or increasing wing CLmax . The latter parameter is the most interesting. One can design a wing airfoil that compromises cruise efficiency to obtain a good CLmax , but it is usually more efficient to include movable leading and/or trailing edges so that one may obtain good high speed performance while achieving a high CLmax at take-off and landing. The primary goal of a high lift system is a high CLmax; however, it may also be desirable to maintain low drag at take-off, or high drag on approach. It is also necessary to do this with a system that has low weight and high reliability.

This is generally achieved by incorporating some form of trailing edge flap and perhaps a leading edge device such as a slat.

The triple-slotted flap system used on a 737 is shown below.

Below is a double-slotted flap and slat system (a 4-element airfoil). Here, some of the increase in CLmax is associated with an increase in chord length (Fowler motion) provided by motion along the flap track or by a rotation axis that is located below the wing.

Modern high lift systems are often quite complex with many elements and multi-bar linkages. Here is a double-slotted flap system as used on a DC-8. For some time Douglas resisted the temptation to use tracks and resorted to such elaborate 4-bar linkages. The idea was that these would be more reliable. In practice, it seems both schemes are very reliable.

Current practice has been to simplify the flap system and double (or even single) slotted systems are often preferred. Flaps change the airfoil pressure distribution, increasing the camber of the airfoil and allowing more of the lift to be carried over the rear portion of the section. If the maximum lift coefficient is controlled by the height of the forward suction peak, the flap permits more lift for a given peak height. Slotted flaps achieve higher lift coefficients than plain or split flaps because the boundary layer that forms over the flap starts at the flap leading edge and is "healthier" than it would have been if it had traversed the entire forward part of the airfoil before reaching the flap. The forward segment also achieves a higher Clmax than it would without the flap because the pressure at the

trailing edge is reduced due to interference, and this reduces the adverse pressure gradient in this region.

The favorable effects of a slotted flap on Clmax was known early in the development on high lift systems. That a 2-slotted flap is better than a single-slotted flap and that a triple-slotted flap achieved even higher Cl's suggests that one might try more slots. Handley Page did this in the 1920's. Tests showed a Clmax of almost 4.0 for a 6-slotted airfoil.

Leading edge devices such as nose flaps, Kruger flaps, and slats reduce the pressure peak near the nose by changing the nose camber. Slots and slats permit a new boundary layer to start on the main wing portion, eliminating the detrimental effect of the initial adverse gradient.

Today computational fluid dynamics is used to design these complex systems; however, the prediction of CLmax by direct computation is still difficult and unreliable. Wind tunnel tests are also difficult to interpret due to the sensitivity of CLmax to Reynolds number and even freestream turbulence levels.

Navier Stokes computations of the flow over a 4-element

3-D Potential Flow

This chapter introduces some basic concepts, important in the analysis of three dimensional flow fields, including finite span wings and slender bodies. 1. General Theory 2. Finite Wings 3. Aerodynamics of Slender Bodies 4. Reference

General Theory
We start our discussion of three dimensional aerodynamics by considering the simple case of irrotational, inviscid flow. When compressibility can also be neglected, we consider solutions to Laplace's equation in 3D:

Just as in the 2-D case, since this equation is linear, we might construct solutions by superimposing known solutions. In 2-D we used sources and vortices extensively to construct the flow over airfoils. In 3D we also use sources and vortices to model the flow over wings and bodies. This section shows how this is done, starting with some results for some fundamental 3-D singularities.

Fundamental Singularities in 3D Potential Flow

One may derive fundamental solutions to Laplace's equation in 3-D, just as we did in 2-D (although complex variables are not quite so useful). 3-D Source It was easily discovered that the potential: = -k/r satisfied Laplace's equation in 3-D. Since V = grad , the velocity associated with this solution is directed radially with a magnitude: V = k/r2. It is easily shown that the constant k is related to the volume flow rate, S by: k = S / 4, so: V = S / 4 r2. The velocity distribution associated with this 3-D source dies off as r2 rather than r as in the 2-D case. Point Doublet Another basic solution, that has been used with some success in supersonic aerodynamics programs is the point doublet, obtained by moving a point source and sink together while keeping the product of their strength, S, and separation, L, constant. With = SL, the velocity associated with the point doublet is:

Vortex Filament One of the most useful fundamental solutions to the 3-D Laplace equation is that of a vortex filament. A vortex filament may be visualized as a thin tube in which the flow has vorticity, . In the limit as the diameter of the tube is made small, but the circulation, , is held fixed, this region of vorticity is called a vortex filament.

Helmholtz Vortex Theorems

Helmholtz summarized some of the properties of vortex filaments, or vortices, in 1858 with his vortex theorems. These three theorems govern the behavior of inviscid three-dimensional vortices:

1. Vortex strength is constant 2. Vortices are forever (end on boundaries or form a closed path) 3. Vortices move with the flow Vortex strength is constant: A vortex line in a fluid has constant circulation.

This can be proved by imagining a closed 3-D loop around the vortex line as shown:

The integral around the closed loop from a to b to c to d to a cuts through no vorticity so from Stokes theorem the integral is zero. But as the slit is made very small the integral approaches the sum of the integral from b to c and the integral from d to a. These are the local circulations around the vortex line and so, the circulations must be constant along the line. Since the vortex strength is constant along the vortex line the strength cannot suddenly go to zero. Thus, a vortex cannot end in the fluid. It can only end on a boundary or extend to infinity. Of course in an real, viscous fluid, the vorticity is diffused through the action of viscosity and the width of the vortex line can become large until it is hardly recognized as a vortex line. A tornado is an interesting example. One end of the twister is on a boundary; but at the other end, the vortex diffuses over a large area with vorticity. As discussed in the section on sources and vortices, singularities such as vortices in the flow move along with the local flow velocity. Here, interactions of the vortices in the trailing wake, cause them to curve around each other and to form the nonplanar wake shown below.

Image from Head 1982 in van Dyke, An Album of Fluid Motion, used with permission.
Biot-Savart Law

The Biot-Savart law relates the velocity induced by a vortex filament to its strength and orientation. The expression, used frequently in electromagnetic theory, can be derived from the basic equations for the 3D potential. The result is:

In the simple case of an infinite vortex we obtain the 2-D result:

In the case of a horseshoe vortex, the two trailing legs contribute:

Here is a simple subroutine to compute the velocity components due to a vortex filament of length Gx, Gy, Gz with the start of the vortex rx, ry, rz from the point of interest.

Biot-Savart Derivation

To determine the velocity associated with a vortex line, we consider the expression for vorticity (see the definition here and related theorems.): If the flow is incompressible, then so we can write , so then, This is a Poison equation for A which has the well-known solution: , ,

where A is called the vector potential. We are free to choose A so that it satisfies

This expression may be integrated along the vortex line for the velocity induced by the filament to obtain the Biot-Savart law:

Vortex-Induced Velocity
C+---------------------------------------------------------------------C Subroutine for computing the velocity induced by a vortex element. C Ilan Kroo 3-1-86 SUBROUTINE VXYZ(GX,GY,GZ,RX,RY,RZ,VX,VY,VZ) C----------------------------------------------------------------------IMPLICIT NONE REAL VX,VY,VZ

REAL GX,GY,GZ,RX,RY,RZ,GXRX,GXRY,GXRZ,R2,G2,GR,V,E1,E2 REAL GR2, RX2 REAL RX1, GX1 REAL TOL, TOL2, PI LOGICAL ERROR TOL = 1.E-10 PI = 3.14159265 R2=RX*RX+RY*RY+RZ*RZ G2=GX*GX+GY*GY+GZ*GZ TOL2 = TOL * G2 GXRX=GY*RZ-GZ*RY GXRY=GZ*RX-GX*RZ GXRZ=GX*RY-GY*RX C Check to see if control point lies in line with vortex: E1=GXRX*GXRX+GXRY*GXRY+GXRZ*GXRZ GR=GX*RX+GY*RY+GZ*RZ E2=(GX+RX)*(GX+RX)+(GY+RY)*(GY+RY)+(GZ+RZ)*(GZ+RZ) if(R2.le.TOL2 .or. E2.le.TOL2) ERROR =.true. if(E1.le.TOL2*R2) ERROR =.true. IF (.not. ERROR) THEN V = GR/SQRT(R2) - (G2+GR)/SQRT(E2) V = V/(4.*PI*E1) VX = V*GXRX VY = V*GXRY VZ = V*GXRZ ELSE C Error trap if control point and vortex are colinear: VX=0. VY=0. VZ=0. END IF RETURN END

Finite Wing Theory

This section deals with several aspects of wing theory, from the development of theoretical models of the finite wing to simple computational methods: 1. Wing Models 2. Lifting Line Theory 3. Induced Drag and the Trefftz Plane 4. Computational Models 5. Simple Sweep Theory

Wing Models
One may apply the results of 3-D potential theory in several ways. We first consider the theory of finite wings. We might start out by saying that each section of a finite wing behaves as described by our 2-D analysis. If this were true then we would still find that the lift curve slope was 2 per radian, that the drag was 0, and the distribution of lift would vary as the distribution of chord. Unfortunately, things do not work this way. There are several reasons for this:

One explanation is that the high pressure on the lower surface of the wing and the low pressure on the upper surface causes the air to leak around the tips, causing a reduction in the pressure difference in the tip regions. In fact, the lift must go to zero at the tips because of this effect. We will next see how and why we must model the 3-D wing differently from 2-D. If we were to take the naive view that the 2-D model would work in 3-D, we might have the picture shown on the right. If each section had the distribution of vorticity along its chord that it had in 2-D, the lift would be proportional to the chord, and would not drop off at the tips as we know it must. This sort of model does not conform to our physical picture of what happens at the wing tips. And indeed, it does not satisfy the equations of 3-D fluid flow. The reason that this does not work is that in this case the streamlines are not confined to a plane. They move in 3-D and the flow pattern is quite different.

We could go back to the governing equations and start simply with the linear Laplace equation. By superimposing known solutions we could obtain a simple model of a 3-D wing. We might start by superimposing vortices on the wing itself:

But this is no more than the strip theory model that did not work. The reason that this model (which seems just to be a superposition of known solutions) is not adequate is that it violates the governing equations in certain regions. The model does not satisfy the Helmholtz laws since vorticity ends in the flow near the tips. Some additional requirements must be imposed on the model. The requirements for such a model are just the Helmholtz vortex theorems, discussed previously. Our simple 3-D model above may be modified as shown below to satisfy the first of the Helmholtz theorems.

In fact, as can be seen from the picture here, this vortex model is not too far from reality.

The downwash field and the existence of trailing vortices are not just some strange mathematical result. They are necessary for the conservation of mass in a 3-D flow.

Air is pushed downward behind the wing, but this downward velocity does not persist far from the wing. Instead it must move outward. The outward-moving air is then squeezed upward outboard of the wing and the flow pattern shown above develops. The trailing vortex is visualized by NASA engineers by flying an agricultural airplane through a sheet of smoke.

The main effect of this vortex wake is to produce a downwash field on the wing. This downwash field has several very significant effects:

It changes the effective angle of attack of the airfoil section. This changes the lift curve slope and has many implications. Induced drag: Lift acts normal to flow in 2D. This accounts for about 40% of the fuel used in a commercial airplane, and as much as 80% of the drag in the critical climb segments. It produces interference effects that are important in the analysis of stability and control.

The magnitude of the downwash can be estimated using the Biot-Savart law, discussed previously. When applied to our simple model with two discrete trailing vortices, the equation predicts infinite downwash at the wing tips, a result that is clearly wrong. In fact, the induced downwash is not even very large. The failure of this simple model led Prandtl to develop a slightly more sophisticated one in 1918. Rather than representing the wing with just one horseshoe-shaped vortex, the wing is represented by several of them:

In this way the circulation on the wing can vary from the root to the tip. The strength of the trailing vortex filaments is related to the circulation on the wing then by: wake = wing A vortex is shed from the wing whenever the circulation changes. In the limit as the number of horseshoe vortices goes to infinity, the trailing wake is a sheet of vorticity.

The trailing vortex strength per unit length in the y direction (vorticity) is the derivative of the total circulation on the wing at that station. From this model, we can derive the basic relations for finite wings.

The vorticity strength in the trailing vortex sheet is given by: = d /dy and since the wing circulation changes most quickly near the tips, the trailing vorticity is strongest in this region. This is why we see tip vortices, and not a complete vortex sheet, as in this NASA photo of an F-111 in a 4-g turn. The vortices are visible in this picture because the low pressure in this region lowers the temperature and we see the condensed

Lifting Line Theory

Basic Theory We could try using 2-D flow results for each section, but correct them for the influence of the trailing vortex wake and its downwash. This is the idea of lifting line theory. We use the 2-D result that: together with the relation: to obtain:

But the angle of attack used here is reduced through the effects of downwash so that the effective angle of attack is the true angle* minus the downwash angle:

Where the induced downwash, Wind, is given by the Biot-Savart Law:

Combining the expression for gamma: with the expression for the downwash angle:

provides an integral equation for the circulation distribution along the wing. Just as in thin airfoil theory, the integral equation can be solved by assuming a Fourier series representation for the distribution.

Substitution of the expression for circulation into the integral equation leads to:

After integrating we have:

The solution of this equation for all values of y is not quite so easy as in the case of thin airfoil theory where we could get closed form expressions for the An's. This is generally done numerically. However, several interesting and simple results appear from this analysis without ever actually computing the An's from the distribution of local angle of attack. Some of these are discussed in the next section. Elliptic Wing Results If, for example, we represent the lift distribution with only a single term in the Fourier series, then:

This represents an elliptic distribution of lift. The downwash angle is, in this case:

The integral is constant when |y| < b/2. In this domain:

Since the downwash distribution is constant the Cl distribution is just:

If the angle of attack is also constant along the wing (no twist) then the Cl is constant and since:

Then in this case the section Cl is equal to the wing CL and:


Recall that this holds for unswept elliptical wings. General Lift Distributions If we are given the lift distribution we can compute the An's as we would with any Fourier expansion. And once we know the Fourier coefficients, we may compute the downwash distribution and the induced drag:

Substitution and evaluation of the definite integral** leads to:

This formula gives the downwash in the plane of the wing for arbitrary load distributions. For the simple elliptical case, closed form solutions for the downwash and sidewash at the start of the wake sheet exist. The simple relation for the velocity induced by an elliptic wing tailing vortex sheet is:

Here, the variable Z is the complex coordinate y + iz and wo is the downwash at the wing root: y = z = 0. This formula permits computation of induced velocities behind a wing as they effect downstream surfaces such as horizontal tails.

Note that the downwash is only constant in the plane of the wing and behind the wing. As we move outboard of the wing or out of the plane of the wake, the downwash varies considerably and there is a rather large upwash beyond the wing tips. This downwash field produces several important effects. It changes the lift of surfaces in other surfaces wakes. This is important in the analysis of airplane stability and the effectiveness of horizontal tails. As can be seen from the downwash plot, the interference of a canard wake with a wing is extreme: the wing lift is reduced behind the canard and the part of the wing outboard of the canard has increased lift. The downwash also produces induced drag as discussed in the next section.

*Note that what we have called the geometric angle of attack is just that for flat plates but it is, in general, the angle of attack from zero lift. **The integral is not in all tables of integrals. It is sometimes called the Glauert integral and is given, for example, in Kuethe and Chow page 146.

Induced Drag and the Trefftz Plane

Fundamentals The 2-D paradox that surfaces in inviscid flow produce no drag no longer applies in 3-D. The downwash created by the trailing wake changes the direction of the force generated by each section:

In three dimensions the force per unit length acting on a vortex filament is Here, the local velocity V includes the component from the freestream and a component from the induced downwash. This latter component produces a component of force in the direction of the freestream: the induced drag. The induced drag is related to the lift by: From the results of lifting line theory for lift and downwash in terms of the Fourier coefficients of the lift distribution:

so we have:

The induced drag is often written as:

This may be written in coefficient form as:

with the same definition of e. Note that e simply depends on the shape of the lift distribution. It is called the span efficiency factor or Oswald's efficiency factor. Note also that the induced drag force depends principally on the lift per unit span, L/b. We can determine quickly, from the expression for induced drag above that drag is a minimum for a given lift and span when all of the Fourier coefficients except the A1 term (which produces lift) are zero. This corresponds to the elliptic loading case mention previously. In this case the downwash is constant and e = 1.

Far Field Analysis and the Trefftz Plane The analysis above works quite well for analyzing the drag of wings given the distribution of lift. It was invented by Prandtl and Betz around 1920. It does involve a bit of hand-waving, though, and requires some very approximate idealizations of the flow. For example: when we talk about the downwash induced by the wake on the wing, just where on the wing do we mean? We assume a single bound vortex line and compute velocities there, but a real wing does not have a single bound vortex and the velocity induced by the wake varies along the chord. Fortunately, the answers from this model are more general than the model itself appears. The induced drag formulas can also be derived from very fundamental momentum considerations.

If the box is made large, contributions from certain sides vanish. In the limit as the box sides go to infinity we obtain the following expressions for lift and drag:

Here, u, v, and w are the perturbation velocities induced by the wing and its wake. Note that the drag only depends on the velocities induced in the "Trefftz Plane" -- a plane far behind the wing. The drag can be expressed as the integral over the infinite plane of the perturbation velocities squared. But, using Gauss' theorem (derivation) it can be expressed as a line integral over the wake itself:

This simplifies the calculation of the drag. The normalwash, Vn, is just the downwash if the wake is flat, but the downwash far behind the wing, not at the wing itself.

Thus we would obtain the same expression as from lifting line theory if the downwash due to the wake at the wing is half the downwash at infinity. This is indeed the case for unswept wings modeled with a lifting line.

Nonplanar Wakes All of the comments above apply to nonplanar wings as well as to simple planar ones. But we must be careful about the assumed wake shape when evaluating forces in the far field. The integrals above actually give the force on the wing and wake combination. Of course, in reality, there is no force on the wake sheet, but if we assume a shape a priori, it is not likely to be a force-free wake. However, since forces act in a direction perpendicular to the vortex, extending wakes streamwise always yields a dragfree wake and nearly correct answers for drag using far field methods.

Far-field velocities can also be used to compute the lift. The results are subtle, but rather interesting.
Munk's Stagger Theorem The result that the drag of a lifting system depends only on the distribution of circulation shed into the wake leads to some very useful results in classical aerodynamics. Perhaps the most useful of these is called Munk's stagger theorem. It states that: The total induced drag of a system of lifting surfaces is not changed when the elements are moved in the streamwise direction. The theorem applies when the distribution of circulation on the surfaces is held constant by adjusting the surface incidences as the longitudinal position is varied. This implies that the drag of an elliptically-loaded swept wing is the same as that of an unswept wing. It also is very useful in the study of canard airplanes for which the canard downwash on the wing is quite complicated. Moving the canard very far behind the wing does not change the drag, but makes its computation much easier. One may use the stagger theorem to prove several other useful results. One of these is the mutual induced drag theorem which states that: The interference drag caused by the downwash of one wing on another is equal to that produced by the second wing on the first, when the surfaces are unstaggered (at the same streamwise location). These results are especially useful in analyzing multiple lifting surfaces.

Trefftz Plane Drag Derivation

Why does the contribution to drag from all but the front and back sides vanish? The pressure terms are clearly 0 since the faces are parallel to the x direction. But the momentum terms are not so easy. We argue that the contribution goes to zero as the walls recede faster than the area goes to infinity. Here is why: In the far field, the induced velocities of a lifting system may be represented as the velocities induced by a single transverse vortex filament and two trailing vortices. The two trailing vortices cancel each other out in the far field, leaving only the piece of bound vorticity. This piece induces a velocity that varies as 1/r2 while the area is increasing as r2. But because the flux (Vn) through the top and bottom are opposite, no first order term exists on these surfaces. Furthermore, this vorticity induces no Vn through the sides. We start with the expression for the drag in terms of the perturbation velocities:

This result comes directly from the application of conservation of momentum and the incompressible Bernoulli equation. Actually, it also assumes that the wake extends infinitely far downstream and trails back from the wing in the freestream direction. If we did not go very far downstream or the wake were not assumed to be straight, the more general expression would be:

But, if we assume that the streamwise perturbation velocities are small, great simplifications are possible:


and outside the wake:


Gauss' theorem states that:


The contour integral is taken as shown below:

We thus obtain:

The jump in potential at the location y in the wake is just the integral of Vds from a point above the wake to a point below. Since the normal velocity is continuous across the wake, the integral is just equal to the circulation enclosed in the loop. This is just the circulation on the wing at the point where this part of the wake left the trailing edge. Similarly, the derivative of normal to the wake, is the induced normalwash, Vn. So:

The last expression may be recognized as the result of lifting line theory, but it has been derived in a much more general way.

Trefftz Plane Lift Derivation

We have discussed the calculation of drag based on the velocities induced in the Trefftz plane, but can lift be calculated in a similar way? The answer is not so easy. We start with the expression for force based on the momentum equation.

Let's assume (naively, for now) that the contribution of each of these integrals goes to zero on each side of the box, except for the back side, as the dimensions of the box are increased. This leaves the contribution in the Trefftz plane due to the wake. In the Trefftz plane, if we assume that all of the induced velocities are normal to the plane, the lift becomes:

The evaluation of this integral seems straightforward. But, it is not.

Consider the integral

when the wake is modeled simply by a pair of vortices. The induced velocity, w, is given by:

Thus, the inner part of the above integral becomes:

So, the integral for lift is:

This looks exactly right; however, let's consider the same integral when the order of integration is reversed:

The inner part of the above integral now becomes:

This integrand is antisymmetric as shown in the plot below. So, the integral for lift is: L = 0.

Now we have a paradox. We get two values for the same integral. Actually, this is not a paradox; it is rather a function that is not Lebesgue integrable. In order to evaluate an integral unambiguously, the function must satisfy two conditions:

1. It must be continuous, except at a countable number of points. 2. The integral of its absolute value must be finite. To avoid this problem, the integral for lift can be first evaluated over finite limits. Taking z from -A to A and y from -B to B we find: L = (2 U / ) {(B+s) atan(A/(B+s)) - (B-s) atan(A/(B-s)) + A/2 ln[ (A2+(B+s)2) / (A2 + (B-s)2)] } The limit as A and B get large depends on the ratio of A to B. When A>>B the value goes to 2 U s, when B>>A the value is 0 and when A = B the value is U s. Thus the integral remains ambiguous when evaluated over an infinite domain. As noted by Larry Wigton of Boeing, this dilemma is resolved by using a different model of the flow field. When the vertical velocity associated with this vortex system is integrated over the Trefftz plane, no ambiguities arise. But, the results are surprising.

The result is that the contributions from the finite length trailing vortices goes to zero. The contribution from the bound vortex is found to be independent of the length of the trailing vortices and is:

The starting vortex contribution is similarly independent of the trailing vortex length and is equal to the bound vortex contribution. Thus, this lift is due to momentum flux, but not from the trailing vortices. We finally need to look at how the pressure term on the upper and lower sides of the control volume is involved. As might be expected, integrals once again are not unambiguous. They depend on the relative sizes of the box sides, even though everything is infinite. A careful analysis leads to the following basic results:

1. If the wake length is small compared with the box width and height then the lift is associated with the momentum term of the starting and bound vortices. 2. If the wake length is large, and the box height is large compared with the width, then the lift is associated with the momentum term of the trailing vortices. 3. If the wake is long and the width is large compared with the height, then the lift is associated with the pressure terms on the top and bottom. Nonplanar Wings Even after the issues with infinite-domain integrals have been resolved, we must worry about the assumed wake position. Although we could argue that streamwise wakes can usually be used for farfield drag computataions, streamwise wakes can still support lift forces. When the wing or wake is substantially nonplanar, these effects can be significant. In fact, the vortex lift generated by highly swept wings can be estimated by far-field methods only when the roll-up of the wake sheet is accurately computed. The alternatives in such cases are to use near field methods or to compute the wake shape. Details: The analysis is a discrete vortex Weissinger computation. Pitching moment is based on the mean geometric chord and is measured about the root quarter chord point. The twist is assumed linear and is taken to be positive for washout (tip incidence less than root incidence).

Simple Sweep Theory

Lifting line theory works only for unswept wings. Weissinger theory provides a means for computing the distribution of lift on swept wings, but not the chordwise distribution of pressures. Vortex lattice models, panel methods, and nonlinear CFD provide pressure distributions on swept-back wings, but do not provide some of the insight that we can obtain with the simpler models. It is mostly for this reason, and partly for historical reasons that simple sweep theory is interesting. It was invented by Buseman around 1935 and independently by R.T. Jones. Consider an infinite wing as shown below.

If we ignore the effects of viscosity, and the wing is painted white so there is nothing to distinguish one section from another, we can slide the wing sideways and we could not tell that it was moving sideways. The air could not tell either, so the pressure distribution would remain unchanged.

We have just created an infinite, obliquely-swept wing that is moving with respect to the air at a speed: We can use this fact to design a wing which can fly at a high speed with a pressure distribution associated with a lower speed. The main idea behind sweeping the wing is to reduce the effects of compressibility. The component of the flow parallel to the wing is not effected by the presence of the wing; the normal component is decoupled from the tangential component.

This is true not only according to linear flow theory, but also in the case of nonlinear compressible flow with shock waves. It is an interesting exercise to show how the full potential equations decompose into a normal term and a tangential term when one asserts that nothing changes in the tangential direction. This idea is called simple sweep theory. We can consider sections normal to the wing edges as operating in a flow with lower Mach number and dynamic pressure. The effective normal Mach number is then: but because of the reduced normal dynamic pressure, the section lift coefficient based on this component of the freestream velocity must be increased if the total lift is fixed:

Furthermore, at a given angle of attack, the lift is reduced. The reduction of lift at a given angle of attack for swept wings has important implications: the airplane incidence angle must be higher, causing several problems for some aircraft on landing approach (e.g. Concorde's drooped nose and long nose gear, F-8 variable incidence wing). The reduced lift curve slope due to sweep can improve the ride quality in gusty air, however.

This basic idea permits subsonic sections to be used at supersonic freestream Mach numbers or transonic airfoils to be used at Mach numbers higher than they would otherwise be able to operate. This works quite well even up to Mach numbers well over 1.0. A recent airplane design that I worked on had a design Mach number of 1.4. The airfoils were designed to operate at a Mach number of 0.7 (normal Mach number) with the wing swept 60. Although the airplane lift coefficient was not meant to exceed 0.25, the airfoils had a design Cl of 1.0 because of the reduced normal dynamic pressure from simple sweep theory. The reason for designing a supersonic wing with sweep and subsonic airfoil sections can be seen in the results of thin airfoil theory which predicted no drag for subsonic sections, but did indicate that supersonic airfoil sections would produce drag due to thickness, camber, and lift. Since the effect varies with cosine of the sweep angle, we expect that either forward, or aft swept wings would realize similar benefits. This is basically true, although, as discussed in the section on wing design and forward-swept wings, there are some important differences. Similarly, wings with oblique sweep have been designed and tested. Further discussions of oblique wings are given in the section on supersonic wings.

Forward-Swept Wings
Since sweep produces effects that vary with cos(sweep), we might expect that either forward or aft sweep would yield the same results. To a first approximation, this is true; but, many other considerations can be important in comparing designs with forward and aft sweep. Historically these have led designers to adopt aft-swept wings for most aircraft, but this was not universally true. The Hansa Jet was a forward-swept wing business jet designed in the 1960's. Its forward swept wing permitted a larger cabin without a wing spar interrupting the floor. Some sailplanes have slight forward sweep to provide better visibility. Recently there has been renewed interest in the forward swept wing concept for aerodynamic reasons and a demonstrator / research aircraft, the X-29 was built by Grumman for NASA, DARPA, and the Air Force.

Several aerodynamic advantages of the forward swept wing have been suggested. One of the more interesting of these is illustrated below. The claim is that the lower surface of a swept forward wing contributes a larger share of the total lift than the lower surface of an aft-swept wing.

Although exaggerated in this figure, this effect is predicted and observed. It is due in part to perturbation velocities induced by the 3-D thickness distribution and in part to the velocities induced by streamwise vorticity. Some of the advantages and disadvantages of forward sweep: Advantages

Better off-design span loading (but with less taper: Cl advantage, weight penalty) Aeroelastically enhanced maneuverability Smaller basic lift distribution Reduced leading edge sweep for given structural sweep

Increased trailing edge sweep for given structural sweep - lower CDc Unobstructed cabin Easy gear placement Good for turboprop placement Laminar flow advantages?


Aeroelastic divergence or penalty to avoid it Lower |Cl| (effective dihedral) Lower Cn (yaw stability) Bad for winglets Stall location (more difficult) Large Cm0 with flaps Reduced pitch stability due to additional lift and fuse interference Smaller tail length???

Aerodynamics of Slender Bodies

Sometimes we are not interested in generating lift, just reducing drag, and when we have to reduce the drag of a given volume, the best shape is often a slender body -- and often nearly a body of revolution. The aerodynamics of such shapes is quite different from airfoils and wings, but follows some of the the same basic principles. Bodies including fuselages are important because they produce drag, lift and moment. They also produce important interference effects with wings and substantially change the stability of an airplane. The flow over more general fuselages and bodies can be predicted in much the same way as the flow over airfoils and wings. Superposition of sources and doublets in the form of panel methods or simpler forms (ala thin airfoil theory) are often used. Navier Stokes equations are used when flow separation is suspected.

In the first part of this section, we examine the results of such calculations rather than the methods themselves, while the second part briefly introduces the concept of slender body theory.

Flow over Bodies

Some closed-form solutions for the potential flow over bodies of revolution are available and are useful as reference results.

We noted that in 2-D the maximum velocity on an ellipse was given by: umax/U = 1 + t/c In 3-D the surface velocity over an ellipsoid of revolution is given by:

x is the distance from the midpoint in units so that the length is 2.0 and r is the radius (or in these units, the ratio of diameter to length.) The maximum velocity is given by:

The figure below (from Schlichting) illustrates the pressure distribution on bodies of revolution. D/L = 0.1

Note that perturbation velocities are much smaller than in 2-D. These velocities may be estimated by superimposing point sources. In this case for an ellipsoid:

Note that the maximum velocity is sensitive to the actual shape, with a paraboloid having about 50% larger perturbations. The results from such a distribution of sources on the axis, slightly underpredicts the velocity perturbations.

The figure below shows the pressure distribution on a typical fuselage shape with D/L = 0.09 computed by a source distribution on the x-axis. Note how the pressure falls near the center of the cylindrical portion of the fuselage.

As indicated below, fuselages in inviscid flow produce a nose-up pitching moment when the angle of attack is increased. This effect is destabilizing and is an important consideration in the sizing of the horizontal tail.

Although the inviscid picture suggests that no lift is produced, the viscous flow actually separates at the back of the fuselage, making the moment somewhat smaller and the lift larger than predicted by inviscid theory. This lift produces induced drag and the fuselage behaves as a low aspect ratio wing. The figure below shows the effect of angle of attack on fuselage lift, drag, and moment based on experimental data. Also shown is the estimated moment based on inviscid


Slender Body Theory

A simple theory can be used to estimate the aerodynamic characteristics of bodies that vary slowly in X-direction: wings of very low aspect ratio and high sweep or slender fuselages.

In such cases, the rate of change of all quantities in the x direction is small and the governing equation becomes:

Note that the (1-M2) du/dx term has been dropped because we assume that everything varies slowly in the X direction. The remaining equation is an equation for 2 dimensional incompressible flow in the cross-plane, and since the Mach dependence has dropped out, it is valid for both M<1 and M>1.

The 2-D cross flow can be computed using a conformal mapping method or even a 2-D panel method. Note that in the cross flow plane, the flow is unsteady as the span of the swept wing or the diameter of the fuselage changes as it cuts through the plane. So while the solution to Laplace's equation still provides the correct velocity distribution, the pressures must be computed using the unsteady Bernoulli equation.

One particularly simple and useful case is that of a highly-swept, low aspect ratio wing.

Using the slender body concept we can solve for the lift of a low aspect ratio wing. Such low aspect ratio wings tend to produce constant downwash and thus nearly elliptic loading until the angle of attack gets large. The rate of change of momentum of the air in the cross-flow plane is equal to the force per unit length on the wing:

If we consider a certain area in the cross plane given a velocity w ( ) then the force becomes: It can be shown from 2-D unsteady theory (not discussed here) that the effective area , A is given by a circle around the plate of diameter equal to the local span, Y. Thus,

The total force on the wing is then given by:

This force acts normal to the wing plane. So for small angles of attack:

This shows that in the low aspect ratio limit, the lift curve slope differs from what lifting line theory predicted by a factor of 2. An expression for lift curve slope derived from second order corrections to lifting line theory is given by Jones:

where p is the ratio of wing semi-perimeter to wing span.

For a rectangular wing p = (b+c) / b = 1 + 1/AR so CL = 2AR / (AR+3). This expression is in close agreement with experiment over a wide range of aspect ratios. A similar analysis can be done for slender bodies of revolution.

This leads to the result that the lift produced by a body with a cut-off base is given by:

Note again that these results are independent of Mach number. The Prandtl-Glauert correction still applies, but the reduction in forces due to effective stretching in the xdirection just cancels the increase in the velocities (1/1-M2) that would have been applied in 2-D.

Compressibility in 3-D
The role of compressibility in the aerodynamics of wings and bodies is introduced in this chapter, including a brief discussion of linear compressibility effects and supersonic wave drag. 1. Subsonic Effects 2. Supersonic Aerodynamics 3. References

4.3D Subsonic Compressibility Effects

5. 6. The effect of compressibility in 3-D flows is somewhat less dramatic than with 2-D flows, but many of the same effects become important. Many of the same techniques for predicting linear compressibility effects work in 3-D too. For example, we can transform the 3-D Prandtl-Glauert equation into the 3-D Laplace equation for incompressible flow by changing variables just as in 2-D. Whereas in 2-D the effects of stretching the x coordinate could be easily analyzed, in 3-D, stretching the x coordinate results in a lower aspect ratio wing with larger sweep -- effects that produce nonlinear changes in wing forces and moments. Nonetheless, changing the lift curve slope just by the Prandtl-Glauert factor does not do too badly:

A somewhat better approximation is obtained by applying the Prandtl- Glauert correction to the 2-D lift curve slope, then applying the downwash correction from lifting line theory:


Applying the 2-D Prandtl-Glauert correction to sections based on the normal component of the freestream Mach number is better yet and this is the basic idea behind a formula for lift curve slope that is widely used for preliminary design calculations. The DATCOM formula shown below includes the effects of finite aspect ratio, sweep, and Mach number as well as a correction factor for viscous effects on the 2-D lift curve slope of the section.

Results are illustrated below.

Of course, solving the correct 3D Prandtl-Glauert equation is the preferred approach for 3D linear compressibility, and transonic computations are better yet, but often the result is quite close, even for cases in which we have no right to be expecting much from the linear theory.

3-D Supersonic Aerodynamics

Sweep may be used to produce subsonic characteristics for a wing, even in supersonic flow. At some point, though, sweep is no longer very effective in delaying the effects of compressibility. That is, the difficulties associated with sweep outweigh the advantages as the required sweep angle gets very large. When the Mach number normal to the leading edge becomes greater than 1, the airfoil sections behave according to linear supersonic theory, with the associated wave drag.

As in 2D, such supersonic wings are more easily analyzed than their subsonic counterparts, though. Consider the point (A) on the wing shown below. Its effect on the flow cannot propagate upstream because disturbances travel at the speed of sound and the freestream is traveling faster than this. This fact is called the law of forbidden signals and implies that disturbances originating at (A) can only affect the darker shaded area. Similarly, points outside the forward-going Mach cone (lightly shaded area) cannot affect the flow at point A.

This means that points on the tips of a supersonic wing can only affect a small part of the wing. The rest of the wing behaves as if it did not know about the wing tips and (except for the effects of sweep and taper) the rest of the wing may be treated as a set of 2-D sections. More detailed analysis shows that in the tip regions behave very much like 2-D sections with their lift curve slope reduced by 50%.

To avoid this loss of lift, the tip sections of supersonic wings are sometimes truncated so that no part of the wing is affected by the tips:

Supersonic Drag As in 2-D, the supersonic wing has lift and volume-dependent wave drag in addition to the skin friction and induced drag terms:

This approximate expression was derived by R.T. Jones, Sears, and Haack for the minimum drag of a supersonic body with fixed lift, span, length, and volume. The expression holds for low aspect ratio surfaces. Notice that unlike the subsonic case, the supersonic drag depends strongly on the airplane length, l. While the equation on the above gives the minimum wave drag for a given length, the actual wave drag may be computed by calculating surface pressures (by solving the Prandtl-Glauert equation with a panel method for example). One can compute the wave drag on a body of revolution relatively easily. For a paraboloid of revolution the drag coefficient based on frontal area is:

For a body with minimum drag with a fixed length and maximum diameter, the result is:

Note that even with a fineness ratio (L/D = length / diameter) of 10, the drag coefficient is about 0.1 -- a large number considering that typical total fuselage drag coefficients based on frontal area are around 0.2. More on Sears-Haack Bodies... Computing Supersonic Volume Wave Drag When the body is does not have the Sears-Haack shape, the volume dependent wave drag may be computed from linear supersonic potential theory. The result is known as the supersonic area rule. It says that the drag of a slender body of revolution may be computed from its distribution of crosssectional area according to the expression:

where A'' is the second derivative of the cross-sectional area with respect to the longitudinal coordinate, x. For configurations more complicated than bodies of revolution, the drag may be computed with a panel method or other CFD solution. However, there is a simple means of estimating the volume-dependent wave drag of more general bodies proposed by Whitcomb.

Transonic drag rise for three configurations: fuselage, fuselage+wing, and body of revolution with same area distribution as wing+body -- from Whitcomb. This involves creating an equivalent body of revolution. At Mach 1.0, this body has the same distribution of area over its length as the actual body; at higher Mach numbers the distribution of area is evaluated with oblique slices through the geometry. A body of revolution with the same distribution of area as that of the oblique cuts through the actual geometry is created and the drag is computed from linear theory.

The angle of the plane with respect to the freestream is the Mach angle, Sin = 1/M, so at M=1, the plane is normal to the flow direction, while at M = 1.6 the angle is 38.7 (It is inclined 51.3 with respect to the M = 1 case.) The actual geometry is rotated about its longitudinal axis from 0 to 2 and the drag associated with each equivalent body of revolution is averaged.

A comparison of actual and estimated drags using this method is shown below from Whitcomb.

Wave Drag Due to Lift

The expression given for wave drag due to lift: ratio.

holds for wings of low aspect

A more general expression is derived by R.T. Jones in "Minimum Drag of Airfoils at Supersonic Speeds", J. of Aero Sciences, Dec. 1952. The combined vortex and wave drag may be written:

This expression approaches the correct limits as M-> 1 and as AR ->infinity. The assumption here is that the lift distribution is elliptical in all directions, an assumption that is not realized exactly in practice, but which is central to understanding how supersonic wings should be shaped. An interactive example in the form of a supersonic wing game is provided here. Jones also gives an expression for the wave drag due to lift for a yawed ellipse, showing that there is an optimum sweep angle. At M = sqrt(2), a 10:1 yawed ellipse at 55 has less than 1/2 the wave drag of the ellipse with 0 or 90 of yaw. These basic ideas can be used to formulate a simple method for estimating wing supersonic drag, and also lead to some interesting configuration concepts such as R.T. Jones' oblique wing.

Sears-Haack Bodies
Sears and Haack derived the shapes for bodies of revolution that produce minimum wave drag due to volume. There are several solutions: 1. Given maximum diameter and length:

2. Given volume and length:

Supersonic Wing Design Game

The purpose of this game is to distribute lift over the length and span of a wing to minimize drag. The idea is that there are several approaches tro obtaining a desired lift distribution. Click on the squares to add or remove lift from a particular place. The goal is to achieve an elliptic distribution of lift over the length and span of the wing. The score represents the deviation from the ideal loading. To assist in designing your wing the ideal and actual loadings are shown as row and column totals. Also each cell is labeled with the amount by which adding or removing lift will change the score. Clicking on the design button will automatically select those cells that help most, starting with your current design and proceeding for a number of generations

Supersonic Drag Estimation

Although inviscid wing drag may be estimated from Euler or full potential analyses, it is sometimes useful to employ simple formulas that provide bounds on the achievable minimum drag. The next pages describe such a method. Lift-dependent wave drag Based on R.T. Jones' expression for ellipses, we may approximate the lift-dependent wave drag of a general shape by considering an ellipse of the same area,S, and length, l:

This choice preserves the average wing pressure difference and agrees well with experimental data for well-designed supersonic wings.

Supersonic Drag Due To Lift Computed by Present Method (*) and Boeing Optimization Results Volume-dependent wave drag Although the area rule may be used to estimate the volume-dependent wave drag of a wing, the equivalent ellipse idea is useful here too. J.H.B. Smith derived an expression for the volume-dependent wave drag of an ellipse. For minimum drag with a given volume:

where t is the maximum thickness, b is the semi-major axis, and a is the semi-minor axis. Note that in the limit of high aspect ratio (a -> infinity), the result approaches the 2-D

result for minimum drag of given thickness: CD = 4 (t/c)2 / . For an ellipse of given area and length the volume drag is:

The figure below shows how this works.

Volume-dependent wave drag for slender wings with the same area distribution. Data from Kuchemann. A program for computation of these drag components is provided here. It may be used to compute the various components of supersonic drag for a wing of given area, length, span, and t/c at a specified Mach number and CL. The program uses the formulas for drag of an equivalent ellipse. Here the length ratio is the aspect ratio based on overall length (length2 / area)

Oblique Wings
These results suggested to R.T. Jones that obliquely-swept wings would be the ideal shape for supersonic aircraft wings. He first proposed the concept in the 1940's and flew flying wing models at the first ICAS meeting in Madrid in 1958. A great deal of work has been done since on oblique wing aircraft including design work by Boeing, General Dynamics, and Lockheed, wind tunnel testing and analysis by NASA, and flight testing of models and piloted aircraft. The picture below shows the AD-1 low speed oblique wing demonstrator.

Although one of the principal advantages involves reduced supersonic wave drag, the concept has other merits.

When compared to an aircraft with symmetric variable sweep (such as the B-1 or F-111), the oblique wing has little change in the aerodynamic center position. This keeps the stability at reasonable levels and avoids large trim changes or complex fuel-pumping schemes.

In addition, several structural advantages have been suggested, especially for variable-sweep aircraft: bending loads on the pivot are avoided, only a single pivot is required, the actuator loads are reduced, and the straight carry-through structure reduces weight in other portions of the aircraft.

An all-wing version of the oblique wing was first proposed by G. H. Lee in 1962. The idea has been revived with the advent of active control systems and a recent artist concept of the oblique flying wing is shown below.

Despite several questions about stability and control, this concept can be made to fly as illustrated in this 4MB video clip of the first flight of an oblique all-wing testbed aircraft.

Viscosity in 3-D
This brief chapter deals with some of the viscous effects peculiar to three-dimensional boundary layers, including separation on high and low aspect ratio surfaces. 1. 3-D Boundary Layer Effects 2. Wings at High Angles of Attack 3. References

4.3-D Boundary Layers on Wings

5. 6. The presence of lateral induced velocities and spanwise pressure gradients on 3-D wings means that one cannot simply use the 2-D boundary layer results along streamwise strips. This is actually done quite frequently today since 3-D boundary layer codes are still rather slow and cumbersome. In some cases 2-D boundary layer equations are solved along the direction of inviscid streamlines. This may be a bit better than using streamwise strips, but one still misses important spanwise pressure gradient effects on the boundary layer.

7. 8. These are especially significant on swept wings. In such cases strong pressure gradients
cause the boundary layer to flow outward, piling up tired, slow air near the tips and contributing to premature tip stall. The streamwise growth of the boundary layer tends to cause early stall near the tips. This is because the boundary flows in the direction of the pressure gradients as shown in the oil flow photographs. This means that it has to travel farther than it would have it it just traveled streamwise. Consequently the boundary layer gets more and more tired and less capable of surviving the adverse gradients without separating. Techniques for improving this problem include the use of vortilons, fences, and other vortex generators.

This photograph shows the outboard flow of the boundary layer and shock-induced separation on the outer portion of this wing at Mach = 0.85. The spanwise flow on the wing also tends to create streamwise vorticity in the boundary layer. This cross-flow instability is very damaging to laminar boundary layers and quickly causes transition to turbulent flow. Wings with sweep angles in excess of 30 to 40 require some sort of boundary layer control (e.g. suction) to maintain laminar flow. In 3-D the approximate boundary layer equations appear:

We cannot just align the x-coordinate with the flow direction because the flow direction may change substantially through the boundary layer. So, three dimensional boundary layer calculations require much more work than the 2-D equations.

9.Wings at High Angles of Attack


11. 12. At high angles of attack, several phenomena usually distinct from the cruise flow appear. Usually part of the wing begins to stall (separation occurs and the lift over that section is reduced). An approximate way to predict when this will occur on well-designed high aspect ratio wings is to look at the Cl distribution over the wing and determine the wing CL at which some section (the critical section) reaches its 2-D maximum Cl. In the example below the outboard sections have Clmax = 1.5 so the wing begins to stall near the tip when CL = 1.24

The effects of wing sweep must be taken into account when using critical section theory as the outboard flow of the boundary layer acts to reduce the maximum Cl available over the outboard sections. When the sweep is very large, separation tends to occur near the leading edge of the wing, but unlike in the low sweep situation, the separated region is not large and does not reduce the lift. Instead, the flow rolls up into a vortex that lies just above the wing


13. Rather than reducing the lift of the wing, the leading edge vortices, increase the wing lift in a nonlinear manner. The vortex can be viewed as reducing the upper surface pressures by inducing higher velocities on the upper surface. The net result can be large as seen on the plot here.

The effect can be predicted quantitatively by computing the motion of the separated vortices using a nonlinear panel code or an Euler or Navier-Stokes solver. This figure shows computations from an unsteady non-linear panel method. Wakes are shed from leading and trailing edges and allowed to roll-up with the local flow field. Results are quite good for thin wings until the vortices become unstable and "burst" - a phenomenon that is not well predicted by these methods. Even these simple methods are computation-intensive.

However, a simple method of estimating the so-called "vortex lift" was given by

Polhamus in 1971. The Polhamus suction analogy states that the extra normal force that is produced by a highly swept wing at high angles of attack is equal to the loss of leading edge suction associated with the separated flow. The figure below shows how, according to this idea, the leading edge suction force present in attached flow (upper figure) is transformed to a lifting force when the flow separates and forms a leading edge vortex (lower figure).

The suction force includes a component of force in the drag direction. This component is the difference between the no-suction drag: CDi = Cn sin , and the full-suction drag: CL2 / AR where is the angle of attack. The total suction force coefficient, Cs, is then: Cs = (Cn sin - CL2/ AR) / cos where is the leading edge sweep angle. If this acts as an additional normal force then: Cn' = Cn + (Cn tan - CL2/ AR) / cos = Cn + (Cn sin - CL2/ AR) / cos so, Cn' = Cn + (Cn sin - CL2/ AR) / cos and in attached flow: CL = CLa sin with Cn = CL cos Thus, Cn' = CL cos + (CL cos sin - CL2/ AR) / cos = CLa sin cos + (CLa sin cos sin - (CLa sin )2/ AR) / cos = CLa sin cos + CLa/ cos sin2 cos - CLa2/( AR cos ) sin2 a CL' = CLa [sin cos2 + sin2 cos2 /cos - CLa/( AR cos ) cos sin2 ] = CLa sin cos (cos + sin cos / cos - CLa sin /( AR cos )) If we take the low aspect ratio result: CLa = AR/2, then: CL '= AR/2 sin cos (cos + sin cos / cos - sin /(2 cos ) ) The plot below shows this computation compared with experiment for a 80 delta wing (AR = 0.705)

Attached flow computations, Polhamus suction analogy, and experiment for lift on a 80 delta wing.

A flow pattern, similar to that of the highly swept delta wing, is found at the tips of low aspect ratio wings and over fuselages. The vortex formation significantly increases the lift in these cases as well. Especially in the case of fuselage vortices, the airplane stability is affected and interaction with downstream surfaces is often important and hard to predict.

Vortices generated by the fuselage and leading edge stakes of an F-18 are visible in the photo below and this QuickTime video clip of NASA's High Alpha Research Vehicle,

used to investigate these phenomena and ways to control them.

Wing Design

There are essentially two approaches to wing design. In the direct approach, one finds the planform and twist that minimize some combination of structural weight, drag, and CLmaxconstraints. The other approach involves selecting a desirable lift distribution and then computing the twist, taper, and thickness distributions that are required to achieve this distribution. The latter approach is generally used to obtain analytic solutions and insight into the important aspects of the design problem, but is is difficult to incorporate certain constraints and off-design considerations in this approach. The direct method, often combined with numerical optimization is often used in the latter stages of wing design, with the starting point established from simple (even analytic) results.

This chapter deals with some of the considerations involved in wing design, including the selection of basic sizing parameters and more detailed design. The chapter begins with a general discussion of the goals and trade-offs associated with wing design and the initial sizing problem, illustrating the complexities associated with the selection of several basic parameters. Each parameter affects drag and structural weight as well as stalling characteristics, fuel volume, offdesign performance, and many other important characteristics. Wing lift distributions play a key role in wing design. The lift distribution is directly related to the wing geometry and determines such wing performance characteristics as induced drag, structural weight, and stalling characteristics. The determination of a reasonable lift and Cl distribution, combined with a way of relating the wing twist to this distribution provides a good starting point for a wing design. Subsequent analysis of this baseline design will quickly show what might be changed in the original design to avoid problems such as high induced drag or large variations in Cl at off-design conditions. A description of more detailed methods for modern wing design with examples is followed by a brief discussion of nonplanar wings and winglets. 1. Wing Design Parameters 2. Lift Distributions 3. Wing Design in More Detail 4. Nonplanar Wings and Winglets 5. References

Wing Design Parameters

Span Selecting the wing span is one of the most basic decisions to made in the design of a wing. The span is sometimes constrained by contest rules, hangar size, or ground facilities but when it is not we might decide to use the largest span consistent with structural dynamic constraints (flutter). This would reduce the induced drag directly. However, as the span is increased, the wing structural weight also increases and at some point the weight increase offsets the induced drag savings. This point is rarely reached, though, for several reasons. 1. The optimum is quite flat and one must stretch the span a great deal to reach the actual optimum. 2. Concerns about wing bending as it affects stability and flutter mount as span is increased. 3. The cost of the wing itself increases as the structural weight increases. This must be included so that we do not spend 10% more on the wing in order to save .001% in fuel consumption. 4. The volume of the wing in which fuel can be stored is reduced. 5. It is more difficult to locate the main landing gear at the root of the wing. 6. The Reynolds number of wing sections is reduced, increasing parasite drag and reducing maximum lift capability.

On the other hand, span sometimes has a much greater benefit than one might predict based on an analysis of cruise drag. When an aircraft is constrained by a second segment climb requirement, extra span may help a great deal as the induced drag can be 70-80% of the total drag. The selection of optimum wing span thus requires an analysis of much more than just cruise drag and structural weight. Once a reasonable choice has been made on the basis of all of these considerations, however, the sensitivities to changes in span can be assessed. Area The wing area, like the span, is chosen based on a wide variety of considerations including: 1. Cruise drag 2. Stalling speed / field length requirements 3. Wing structural weight 4. Fuel volume

These considerations often lead to a wing with the smallest area allowed by the constraints. But this is not always true; sometimes the wing area must be increased to obtain a reasonable CL at the selected cruise conditions. Selecting cruise conditions is also an integral part of the wing design process. It should not be dictated a priori because the wing design parameters will be strongly affected by the selection, and an appropriate selection cannot be made without knowing some of these parameters. But the wing designer does not have complete freedom to choose these, either. Cruise altitude affects the fuselage structural design and the engine performance as well as the aircraft aerodynamics. The best CL for the wing is not the best for the aircraft as a whole. An example of this is seen by considering a fixed CL, fixed Mach design. If we fly higher, the wing area must be increased by the wing drag is nearly constant. The fuselage drag

decreases, though; so we can minimize drag by flying very high with very large wings. This is not feasible because of considerations such as engine performance. Sweep Wing sweep is chosen almost exclusively for its desirable effect on transonic wave drag. (Sometimes for other reasons such as a c.g. problem or to move winglets back for greater directional stability.) 1. It permits higher cruise Mach number, or greater thickness or CL at a given Mach number without drag divergence. 2. It increases the additional loading at the tip and causes spanwise boundary layer flow, exacerbating the problem of tip stall and either reducing CLmax or increasing the required taper ratio for good stall.

3. It increases the structural weight - both because of the increased tip loading, and because of the increased structural span.

4. It stabilizes the wing aeroelastically but is destabilizing to the airplane. 5. Too much sweep makes it difficult to accommodate the main gear in the wing.

See section 9.2.5 for more detail on simple sweep theory and the effects of sweep. Much of the effect of sweep varies as the cosine of the sweep angle, making forward and aft-swept wings similar. There are important differences, though as discussed further in the section on forward swept wings. Thickness The distribution of thickness from wing root to tip is selected as follows: 1. We would like to make the t/c as large as possible to reduce wing weight (thereby permitting larger span, for example). 2. Greater t/c tends to increase CLmax up to a point, depending on the high lift system, but gains above about 12% are small if there at all. 3. Greater t/c increases fuel volume and wing stiffness.

4. Increasing t/c increases drag slightly by increasing the velocities and the adversity of the pressure gradients. 5. The main trouble with thick airfoils at high speeds is the transonic drag rise which limits the speed and CL at which the airplane may fly efficiently. Taper The wing taper ratio (or in general, the planform shape) is determined from the following considerations: 1. The planform shape should not give rise to an additional lift distribution that is so far from elliptical that the required twist for low cruise drag results in large off-design penalties. 2. The chord distribution should be such that with the cruise lift distribution, the distribution of lift coefficient is compatible with the section performance. Avoid high Cl's which may lead to buffet or drag rise or separation. 3. The chord distribution should produce an additional load distribution which is compatible with the high lift system and desired stalling characteristics. 4. Lower taper ratios lead to lower wing weight. 5. Lower taper ratios result in increased fuel volume. 6. The tip chord should not be too small as Reynolds number effects cause reduced Cl capability. 7. Larger root chords more easily accommodate landing gear.

Here, again, a diverse set of considerations are important. The major design goal is to keep the taper ratio as small as possible (to keep the wing weight down) without excessive Cl variation or unacceptable stalling characteristics. Since the lift distribution is nearly elliptical, the chord distribution should be nearly elliptical for uniform Cl's. Reduced lift or t/c outboard would permit lower taper ratios. Evaluating the stalling characteristics is not so easy. In the low speed configuration we must know something about the high lift system: the flap type, span, and deflections. The flaps- retracted stalling characteristics are also important, however (DC-10). Twist The wing twist distribution is perhaps the least controversial design parameter to be selected. The twist must be chosen so that the cruise drag is not excessive. Extra washout helps the stalling characteristics and improves the induced drag at higher CL's for wings with additional

load distributions too highly weighted at the tips. Twist also changes the structural weight by modifying the moment distribution over the wing. Twist on swept-back wings also produces a positive pitching moment which has a small effect on trimmed drag. The selection of wing twist is therefore accomplished by examining the trades between cruise drag, drag in second segment climb, and the wing structural weight. The selected washout is then just a bit higher to improve stall.

Wing Lift Distributions

As in the design of airfoil sections, it is easier to relate the wing geometry to its performance through the intermediary of the lift distribution. Wing design often proceeds by selecting a desirable wing lift distribution and then finding the geometry that achieves this distribution. In this section, we describe the lift and lift coefficient distributions, and relate these to the wing geometry and performance.

About Wing Lift and Cl Distributions Relating Wing Geometry and Lift Distribution Lift Distributions and Performance

About Lift and Cl Distributions

The distribution of lift on the wing affects the wing performance in many ways. The lift per unit length l(y) may be plotted from the wing root to the tip as shown below.

In this case the distribution is roughly elliptical. In general, the lift goes to zero at the wing tip. The area under the curve is the total lift. The section lift coefficient is related to the section lift by:

So that if we know the lift distribution and the planform shape, we can find the Cl distribution. The lift and lift coefficient distributions are directly related by the chord distribution. Here are some examples:

The lift and Cl distributions can be divided into so-called basic and additional lift distributions. This division allows one to examine the lift distributions at a couple of angles of attack and to infer the lift distribution at all other angles. This is especially useful in the process of wing design. From the discussion of lifting line theory and Weissinger theory, we saw that the distribution of lift could be written:

where: r is the angle of attack at the root is the twist angle {l1} is the wing lift distribution with no twist and with r = 1 {l2} is the lift distribution at zero angle of attack and unit twist. The lift distribution may also be written in a more conventional way:

Here, the distributions {la} and {lb} are the wing lift distributions with no twist at CL = 1 and with unit twist at zero lift respectively. The first term, CL {la}, is called the additional lift. It is the lift distribution that is added by increasing the total wing lift. {lb} is called the basic lift distribution and is the lift distribution at zero lift. Why is this useful? Consider the following example.

We can use the data at these two angles of attack to learn a great deal about the wing. From the expression above:

or: The additional lift distribution, CL {la} may be interpreted graphically as shown below.

The additional lift coefficient distribution at CL = 1.0 is plotted below. Note that it rises upward toward the tip -- this is indicative of a wing with a very low taper ratio or a wing with sweep-back.

The basic lift distribution is negative near the tip implying that the wing has washout.

Wing Geometry and Lift Distribution

The wing geometry affects the wing lift and Cl distributions in mostly intuitive ways. Increasing the taper ratio (making the tip chords larger) produces more lift at the tips, just as one might expect:

But because the section Cl is the lift divided by the local chord, taper has a very different effect on the Cl distribution.

Changing the wing twist changes the lift and Cl distributions as well. Increasing the tip incidence with respect to the root is called wash-in. Wings often have less incidence at the tip than the root (wash-out) to reduce structural weight and improve stalling characteristics.

Since changing the wing twist does not affect the chord distribution, the effect on lift and Cl is similar. Wing sweep produces a less intuitive change in the lift distribution of a wing. Because the downwash velocity induced by the wing wake depends on the sweep, the lift distribution is affected. The result is an increase in the lift near the tip of a swept-back wing and a decrease near the root (as compared with an unswept wing.

This effect can be quite large and causes problems for swept-back wings. The greater tip lift increases structural loads and can lead to stalling problems. The effect of increasing wing aspect ratio is to increase the lift at a given angle of attack as we saw from the discussion of lifting line theory. But it also changes the shape of the wing lift distribution by magnifying the effects of all other parameters. Low aspect ratio wings have nearly elliptic distributions of lift for a wide range of taper ratios and sweep angles. It takes a great deal of twist to change the distribution. Very high

aspect ratio wings are quite sensitive, however and it is quite easy to depart from elliptic loading by picking a twist or taper ratio that is not quite right.

Note that many of these effects are similar and by combining the right twist and taper and sweep, we can achieve desirable distributions of lift and lift coefficient. For example: Although a swept back wing tends to have extra lift at the wing tips, washout tends to lower the tip lift. Thus, a swept back wing with washout can have the same lift distribution as an unswept wing without twist. Lowering the taper ratio can also cancel the influence of sweep on the lift distribution. However, then the Cl distribution is different.

Today, we can relate the wing geometry to the lift and Cl distributions very quickly by means of rapid computational methods. Yet, this more intuitive understanding of the impact of wing parameters on the distributions remains an important skill.

Lift Distributions and Performance

Wing design has several goals related to the wing performance and lift distribution. One would like to have a distribution of Cl(y) that is relatively flat so that the airfoil sections in one area are not "working too hard" while others are at low Cl. In such a case, the airfoils with Cl much higher than the average will likely develop shocks sooner or will start stalling prematurely. The induced drag depends solely on the lift distribution, so one would like to achieve a nearly elliptical distribution of section lift. On the other hand structural weight is affected by the lift distribution also so that the ideal shape depends on the relative importance of induced drag and wing weight. With taper, sweep, and twist to "play with", these goals can be easily achieved at a given design point. The difficulty appears when the wing must perform well over a range of conditions.

One of the more interesting tradeoffs that is often required in the design of a wing is that between drag and structural weight. This may be done in several ways. Some problems that have been solved include:

Minimum induced drag with given span -- Prandtl Minimum induced drag with given root bending moment -- Jones, Lamar, and others Minimum induced drag with fixed wing weight and constant thickness -- Prandtl, Jones Minimum induced drag with given wing weight and specified thickness-to-chord ratio -Ward, McGeer, Kroo Minimum total drag with given wing span and planform -- Kuhlman ... there are many problems of this sort left to solve and many approaches to the solution of such problems. These include some closed-form analytic results, analytic results together with iteration, and finally numerical optimization. The best wing design will depend on the construction materials, the arrangement of the high-lift devices, the flight conditions (CL, Re, M) and the relative importance of drag and weight. All of this is just to say that it is difficult to design just a wing without designing the entire airplane. If we were just given the job of minimizing cruise drag the wing would have a very high aspect ratio. If we add a constraint on the wing's structural weight based on a trade-off between cost and fuel savings then the problem is somewhat better posed but we would still select a wing with very small taper ratio. High t/c and high sweep are often suggested by studies that include only weight and drag. The high lift characteristics of the design force the taper ratio and sweep to more usual values and therefore must be a fundamental consideration at the early stages of wing design. Unfortunately the estimation of CLmax is one of the more difficult parts of the preliminary design process. An example of this sensitivity is shown in the figure below.

The effect of a high lift constraint on optimal wing designs. Wing sweep, area, span, and

twist, chord, and t/c distributions were optimized for minimum drag with a structural weight constraint. (Results from work of Sean Wakayama.

Wing Design in More Detail

The determination of a reasonable lift and Cl distribution, combined with a way of relating the wing twist to this distribution provides a good starting point for a wing design. Subsequent analysis of this baseline design will quickly show what might be changed in the original design to avoid problems such as high induced drag or large variations in Cl at off-design conditions. Once the basic wing design parameters have been selected, more detailed design is undertaken. This may involve some of the following:

Computation or selection of a desired span load distribution, then inverse computation of required twist. Selection of desired section Cp distribution at several stations along the span and inverse design of camber and/or thickness distribution. All-at-once multivariable optimization of the wing for desired performance.

Some examples of these approaches are illustrated below.

This figure illustrates inverse wing design using the DISC (direct iterative surface curvature) method. The starting pressures are shown (top), followed by the target (middle), and design (bottom); light yellow = low pressure and green = high pressure. This is an inverse technique that has been used very successfully with Navier-Stokes computations to design wings in transonic, viscous flows. Below is an example of wing design based on "fixing" a span load distribution. When the 737 was re-engined with high bypass ratio turbofans, a drag penalty was avoided by changing the effective wing twist distribution.

The details of the pressure distribution can then be used to modify the camber shape or wing thickness for best performance. This sounds straightforward, but it is often very difficult to accomplish this, especially when it takes hours or days to examine the effect of the proposed change. This is why simple methods with fast turnaround times are still used in the wing design process.

As computers become faster, it becomes more feasible to do full 3-D optimization. One of the early efforts in applying optimization and nonlinear CFD to wing design is described by Cosentino and Holst, 1986. In this problem, a few spline points at several stations on the wing were allowed to move and the optimizer tried to maximize L/D.

Although this was an inviscid code, the design variables were limited, and the objective function simplistic, current research has included more realistic objectives, more design degrees of

freedom, and better analysis codes.

--but we are still a long way from having "wings designed by computer."

Nonplanar Wings and Winglets

One often begins the wing design problem by specifying a target Cp distribution and/or span loading and then modifying the wing geometry (either manually, by direct inverse, or by nonlinear optimization). In the case of planar wings, the elliptic loading results provide a useful benchmark in the creation of target loadings. (For high aspect ratio wings, 2D airfoil results may provide a useful reference for the chordwise loading.) More complex methods for creating target Cp's are beyond the scope of this discussion, but we have little guidance at all when the wing is nonplanar. This section deals with the problem of optimal loading for nonplanar lifting surfaces. It is easily generalized to multiple surfaces. When the wing is not planar, many of the previous simple results are no longer valid. Elliptic loading does not lead to minimum drag and the span efficiency can be greater than 1.0. Here we will describe a method for computing the minimum induced drag for planar and nonplanar wings. First, consider the distribution of downwash for minimum drag. This can be obtained by using the method of restricted variations as follows.

We consider an arbitrary variation in the circulation distribution represented by 1 and 2 which do not change the lift: L = U 1 + U 2 = 0. This implies: 1 = - 2 If the drag was minimized by the initial distribution: D = /2 w1 1 + /2 w2 2 = 0. So, w1 = w2 That is, the downwash is constant behind a planar wing with minimum drag.

In the general case, with multiple surfaces or nonplanar wings, the same approach may be used. In this case, the condition for constant lift is: L = U 1 cos 1 + U 2 cos 2= 0. where is the local dihedral angle of the lifting surface. For minimum drag: D = /2 Vn1 1 + /2 Vn2 2 = 0. where Vn is the induced velocity in the Trefftz plane in a direction normal to the wake sheet (the normalwash). In this case, 1 cos 1 = - 2 cos 2 so, Vn = k cos .

The normalwash is proportional to the local dihedral angle. Thus, the sidewash on optimally-loaded winglets is 0, for example. We may then solve for the distribution of circulation that produces this distribution of normalwash. Alternatively, one may use a more direct optimization approach. With the circulation distribution represented as the row vector, {} and the wake modeled as a collection of line vortices of strength {w}, we may write the wake vorticity in terms of the surface circulation, based on a discrete vortex model as shown below.

The drag is then given by: D = /2 {Vn} {} where Vn is the normal wash in the Trefftz plane computed using the Biot-Savart law. {Vn} is related to the circulation strengths by: {Vn} = [VIC] {} where [VIC] is a function of the geometry. So, D = /2 [VIC] {} {} The lift is also a function of the circulations: L = U {} {cos } with the local dihedral angle. Finally, the optimal values of {} are given by setting d ( D + (L-Lref) ) / di = 0 where is a Lagrange multiplier. This problem is sometimes done as homework, but some results are summarized below: When the wing/winglet combination is optimized for minimum drag at fixed span, it achieves about the same drag as a planar wing with a span increased by about 45% of the winglet height. The wing lift distribution is as shown below with increased lift outboard compared with the no winglet case.

This increased tip loading along with the extra bending moment of the winglet leads to increased structural weight. When a bending moment constraint replaces the span constraint, wings with winglets are seen to have about the same minimum drag as the stretched-span planar wings. This is shown below. Induced drag of wings with winglets and planar wings all with the same integrated bending moment (related to structural weight). Note that solutions to the left of the span ratio = 1.0 line are not meaningful.

The same approach may be taken for general nonplanar wake shapes. The figure below summarizes some of these results, showing the maximum span efficiency for nonplanar wings of various shapes with a height to span ration of 0.2.

Several points should be made about the preceding results. 1. The result that the sidewash on the winglet (in the Trefftz plane) is zero for minimum induced drag means that the self-induced drag of the winglet just cancels the winglet thrust associated with wing sidewash. Optimally-loaded winglets thus reduce induced drag by lowering the average downwash on the wing, not by providing a thrust component. 2. The results shown here deal with the inviscid flow over nonplanar wings. There is a slight difference in optimal loading in the viscous case due to lift-dependent viscous drag. Moreover, for planar wings, the ideal chord distribution is achieved with each section at its maximum Cl/Cd and the inviscid optimal lift distribution. For nonplanar wings this is no longer the case and the optimal chord and load distribution for minimum drag is a bit more complex. 3. Other considerations of primary importance include: Stability and control Structures Other pragmatic issues More details on the design of nonplanar wings may be found in a recent paper, "Highly Nonplanar Lifting Systems," accessible here.

Configuration Aerodynamics
This chapter deals with a few of the considerations required when one is dealing not with a single wing or body, but with a more complex configuration. In particular, we look at multiple

lifting surfaces and introduce some of the stability and control issues that are key elements of configuration aerodynamic performance. These sections represent introductions to some of these issues; more complete discussions are given in the aircraft design text. 1. Multiple Lifting Surfaces 2. Longitudinal Stability and Trim 3. Horizontal Tails 4. Canard Aircraft 5. Tailless Aircraft 6. References

Multiple Lifting Surfaces

Introduction to Multiple Lifting Surfaces
Sometimes a combination of several lifting surfaces can do more than just a single surface. Applications include:

Balancing the moments: Trim and tail surfaces Structures: Biplanes for increasing effective structural depth Stability: Tail surfaces can also be used to increase the stability of an airplane (or boat or arrow) Drag: Additional surfaces can sometimes be used to reduce the overall drag of a design. Winglets are a case in point.

The aerodynamics of multiple lifting surfaces are more than just the sum of the isolated parts. There are several types of interference that can be important.

In this case velocities are induced by the bound vorticity of each wing. This effect can be seen in the 2-D image effects discussed previously. In the case pictured here, the rear wing increases the velocity on the forward wing thus increasing its lift.

In this case, the velocities induced by the wake of one surface affect the lift on the other. When the forward wing has a smaller span the downwash inboard reduces the lift while the upwash outboard increases the lift in this region. One common example of multiple surface interference is the interaction of a wing and a tail surface.

The downwash from the wing at the tail reduces the tail lift from its value without the wing...

(it is the incidence of the zero lift line of the tail with respect to the angle of attack reference line) ... to its value in the wing's presence:

The rate at which tail lift changes with airplane angle of attack is thus reduced by the downwash effect to:

This has a very important effect on the aircraft stability, reducing the effectiveness of the tail by about 40% with a wing of AR = 8.

Induced Drag
The Prandtl Biplane Equation, derived by integrating the wing downwash, is used to compute the induced drag of two elliptically-loaded wings including the interference effects.

where the interference factor, , depends on the span ratio and the vertical gap between the surfaces. It varies from b2/b1 when the gap is 0, to 0 when the gap is infinite. Note when comparing the induced

drag of two systems it is often necessary to fix the total lift: L= L1+L2. From the biplane equation the ratio of the drag of 2 wings with a given total lift and maximum span to a single wing with the same total lift and span is:

We can define the span efficiency, e, with the usual equation for induced drag:

where b1 is the span of the surface with the largest span. Then:

Although the biplane equation provides a simple method for computing the induced drag of two planar wings, it makes the assumption that the wings are elliptically-loaded. This leads to minimum drag for a single planar wing, but not for multiple lifting surfaces or nonplanar wings, such as wings with winglets. To compute the minimum induced drag for a general arrangement of nonplanar lifting surfaces we must compute the optimal circulation distribution and then compute the minimum drag. The method for doing this is just a revised version of that described in the section on induced drag. When this is done, we find a similar expression for the drag of two surfaces:

Where the constant *, like , depends on the span ratio and vertical gap. * also goes to 0 as the vertical gap is made large, and approaches (b2/b1)2 in the limit of small vertical gap. Thus, for minimum drag with zero gap, the combined loading is elliptical and the span efficiency is 1.0.

Longitudinal Stability and Trim

The drag of the system is dependent on the distribution of loads between the surfaces. In order to determine this, and to properly size the tail surface, we must consider the aircraft's stability and trim. Stability is the tendency of a system to return to its equilibrium condition after being disturbed from that point. Two types of stability or instability are important. A static instability: A dynamic instability:

An airplane must be a stable system (well, with some exceptions ) with acceptable time constants. To assure this, a careful analysis of the dynamic response and controllability is required, but here we look only at the simplest case: static longitudinal stability and trim. This will tell us something about the aerodynamic design of the surfaces -- the load they must carry, the effect of airfoil properties, and the drag associated with the surfaces. If we displace the wing or airplane from its equilibrium flight condition to a higher angle of attack and higher lift coefficient:

we would like it to return to the lower lift coefficient. This requires that the pitching moment about the rotation point*, Cm, become negative as we increase CL:

Note that: where x is the distance from the system's center of additional lift to the c.g.

So, If x were 0, the system would be neutrally stable. x/c represents the margin of static stability and is thus called the static margin. Typical values for stable airplanes range from 5% to 40%. The airplane may therefore be made as stable as desired by moving the c.g. forward (by putting lead in the nose) or moving the wing back. One needs no tail for stability then, only the right position of the c.g..

Although this configuration is stable, it will tend to nose down whenever any lift is produced. In addition to stability we require that the airplane be trimmed (in moment equilibrium) at the desired CL. This implies that: With a single wing, generating a sufficient Cm at zero lift to trim with a reasonable static margin and CL is not so easy. (Most airfoils have negative values of Cm0.) Although tailless aircraft can generate sufficiently positive Cm0 to trim, the more conventional solution is to add an additional lifting surface such as an aft-tail or canard. The following sections deal with some of the considerations in the design of each of these configurations.

* The rotation point is the center of gravity (c.g.) for freely flying bodies.

Horizontal Tails

Trim can be achieved by setting the incidence of the tail surface (which adjusts its CL) to make Cm = 0.

Stability can simultaneously be assured by appropriate location of the c.g.:

Thus, given a stability constraint and a trim requirement, we can determine where the c.g. must be located and can adjust the tail lift to trim. We then know the lifts on each interfering surface and can compute the combined drag of the system using the theory of interfering surfaces, discussed in the previous section. Horizontal tails are generally used to provide trim and control over a range of conditions. Typical conditions over which tail control power may be critical and which sometimes determine the required tail size include: take-off rotation (with or without ice), approach trim and nose-down acceleration near stall. Despite the drawing above, many tail surfaces are normally loaded downward in cruise. For some commercial aircraft the tail download can be as much as 5% of the aircraft weight. As stability requirements are relaxed with the application of active controls, the size of the tail surface and/or the magnitude of tail download can be reduced.

Canard Aircraft

Voyager: unrefueled flight around the world. Image courtesy B. Rutan. The same stability and trim equations may be satisfied with negative values of the tail length. Such designs are called "canard" designs. (Canard is the French word for duck. Some of the early canard designs bore a strange resemblance to their namesakes. Oddly enough, canard in English means a gross exaggeration or hyperbole.) These notes deal with some of the basic analyses required in the design of canard aircraft. These analyses differ in several ways from those methods which are usually used in the initial design of aft-tail configurations. Although several differences are apparent we do not address in detail here the general question of the relative merits of canard and aft-tail designs. (See Refs. 1 and 2 for this.) The large variety of canard designs illustrates that, unlike the case of aft-tail designs, a consensus on optimal canard size has not yet emerged.

Some of Burt Rutan's canard designs. One possible explanation for this lies in the fact that aircraft performance is much more sensitive to canard size than to tail size. The proper canard size depends strongly on the most critical design condition (e.g. climb unimportant, high speed cruise critical vs. a long endurance design with high speeds unnecessary.) Here we consider simple methods of estimating key performance parameters for a given design. For a particular project, several canard sizes may be tried to achieve the best compromise.

Stability and Trim Drag Summary: Canard Advantages and Disadvantages Interactive Calculations of Canard Performance

Canard Stability and Trim

The performance of a canard design depends strongly on the amount of lift that the canard must carry. This is set by stability and trim requirements. It is first necessary to determine the position of the c.g. and the relative loads carried by the wing and the canard. The ratio of lift carried by the canard to that carried by the wing:

depends on several parameters listed below.


If moving the c.g. back by sm * cref makes the airplane neutrally stable then:



The value of the static margin, sm, should be large enough to provide acceptable handling qualities at the most aft center of gravity position. This may require an analysis of the aircraft dynamics at various flight conditions. Since the destabilizing effect of the fuselage is not included explicitly here the value of sm used in the above expressions should be increased appropriately. Lift Curve Slopes A difficulty with the above equations is that a, the ratio of canard to wing lift curve slopes must be calculated. If the wings did not interfere with each other one could use the approximate relations:

However, the surfaces produce upwash and downwash on each other so that the effective lift curve slope is changed. Unless the canard and wing are very close together, the major effect is that of the canard on the wing. The canard produces downwash on the inner part of the wing and upwash outboard of the canard tip vortices. The net effect, though, is a reduction in wing lift which can be estimated roughly by the following formula which is based on the Hayes Reverse Flow Theorem (see Ref. 3):

where kappa is a correction which is applied if the canard is very close to the wing or does not lie in the plane of the wing. kappa should be computed from a 2-surface lifting line or lifting surface program.

Canard Drag
Standard methods for parasite drag prediction work quite well for canards: Compute the drag of fuselage, etc., at the desired Reynolds number. Compute profile drag of the wing based on airfoil data at the appropriate Cl. The canard profile drag is also computed from airfoil data at CL=CLc. The tricky part of drag prediction is induced drag which includes most of what is generally termed "trim drag". (A part of the trim drag will also appear as profile drag.) Given the lift on each of the surfaces, the induced drag may be computed from the expressions derived earlier for multiple lifting surfaces.

e is the overall span efficiency and is given by:

if the surfaces are individually elliptically-loaded. By twisting the wings in just the right way the induced drag can be reduced from this value somewhat (see Ref. 5) but the preceding equations (which assume an elliptical distribution of lift over each of the surfaces) provide a good practical estimate. This is especially true when structural design is considered: the distribution of wing lift for minimum drag with fixed span carries a large fraction of the total lift on the outboard wing sections, leading to larger bending moments and greater weights. When drag is minimized at fixed weight, the optimal loading is more nearly elliptic.

Results for a particular set of cases is shown below.

Unrelated to canard system drag, the maximum CL of these systems based on total area is shown below. These results are very sensitive to many of the assumed parameters and are

included here only as representative results, not general conclusions.

A Summary of Canard Advantages and Disadvantages


Possibility for very good stalling characteristics without elevator stops. Sometimes a desirable layout from the packaging standpoint: Main wing carry-through behind cabin, pusher engine installation simplified. Synergistic use of winglets for directional stability. In certain cases a simplified control linkage is possible. When wing flaps are not desired (for simplicity as in ultralights, or competition rules as with standard class sailplanes for example) the CLmax of a canard may exceed that of an aft-tail airplane. For unstable aircraft, canard designs may have a CLmax and/or drag advantage. Control authority is larger for unstable canard aircraft at high CL than for unstable aft-tail designs.


Fuel center of gravity lies farther behind aircraft c.g. than in conventional designs. This means that a large c.g. range is produced or that the fuel must be held elsewhere (e.g. strakes near the wing root.) CLmax problems with flaps or margin on the entire wing: Flaps produce a larger pitching moment about the c.g. on a canard aircraft. This results in the need for both large canard aerodynamic incidence change and high maximum canard lift coefficient. Note that since the value of a S is usually larger for canard designs, Cm0 has a greater impact on L than it does on aft-swept designs. Induced drag / CLmax incompatibility: Canard designs can achieve equal or better CLmax values than conventional designs, and similar values of span efficiency. However, the configurations with high CLmax values have terrible values of e and those with respectable e 's have low maximum lift coefficients. Directional stability: The distance from the aircraft c.g. to the most aft part of the airplane is usually smaller on canard aircraft. This poses a problem for locating a vertical stabilizer and may result in very large vertical surfaces. (Note, however, that winglets may be used to advantage in this case.) Wing twist distribution is strange and CL dependent: The wing additional load distribution is distorted by the canard wake. Power effects on canard - deep stall: Accidents have been associated with tractor canard configurations for which the propeller slipstream has prevented canard stall before wing stall. The result is a possible deep-stall problem. Finally, and perhaps most importantly, canard sizing is much more critical than aft tail sizing. By choosing a canard which is somewhat too big or too small the aircraft performance can be severely affected. It is easy to make a very bad canard design.

The appendices include a JavaScript-based calculator for computing several quantities related to the standard atmosphere, a page of useful vector identities, and (soon) other interactive tools.

Standard Atmosphere Unit Conversions Vector Identities List of Video Clips

List of Pages with Interactive Computation

Vector Identities
Subscripts denote partial derivatives w.r.t. given coordinate. u, v, w are x, y, z components of V. Definitions of differential operators:

Miscellaneous identities:

Integral relations:

Problem Set #1

1. a. Show that in an unpressurized (atmospheric) wind tunnel, if one would like to duplicate full-scale Reynolds numbers and dimensionless force coefficients with a scale model, the actual (dimensional) forces on the model will be the same as the forces on the full-scale vehicle. Assume sea level conditions. b. A DC-9 weighs 100,000 lbs (45455 kg), has a wing area of 1000 sq ft (92.9 sq m) and is flying at 30,000 ft (9144 m) at M = .80. Find the lift coefficient, CL. If we tested a 1/5th scale model of this airplane in a large, unpressurized wind tunnel, what Mach number would have to be used if we were to match Reynolds numbers? Since this is not an attractive situation (why?) what alternatives do we have, i.e. how could we change the test conditions to achieve the same Re and M?

2. It has recently been suggested that some of the problems discussed in problem 1 could be avoided with tests on remotely-piloted vehicles. If we built a scale flying model of the DC-9 with the same wing loading (Weight / Wing Area) as the real DC-9, at what altitude and speed would the model have to be flown in order to match Re and M with the smallest possible model and what would be the model scale? What would the lift coefficient be in this case? If we wanted to match full-scale lift coefficient also, we could fly in a turn. What turn radius and 'g' - load would be required?

This concept has recently been proposed by John McMasters of Boeing Commercial Airplanes. Do you think this is a good idea? Why or why not?

3. Using the principle of momentum and mass conservation, derive a formula for the drag of a

wing in a wind tunnel in terms of the measured velocity components u(y,z), v(y,z), w(y,z) near the back of the test section. Assume inviscid, incompressible flow where the pressures and velocities are related by the Bernoulli equation:

4. a. For a movie flight sequence, a scale radio-controlled model is used with a wing span that is 1/10 of the full-scale airplane, and with a weight that is 1/2000 of the full-scale weight. Reynolds number effects will be ignored, but the model will be flown at the proper (full-scale) angle of attack. If the full-scale aircraft flies at 200 knots (371 km/hr) and the power required is 300 H.P. (224 kw), calculate the speed at which the model should be flown when the movie is filmed and the power for the model. (Hint: At a given angle of attack, the CL's of the model and full-scale airplane are the same. Remember: Power = Speed * Drag and assume everything is at sea level ) b. By considering the time for the vehicle to fly one body length, calculate the speed-up (or slow-down) ratio at which the movie must be projected to provide a life-like simulation.

AA 200A Problem Set #2

1. A two-dimensional incompressible flow field is described by:

a. Show that this field satisfies continuity and is irrotational. b. Find a formula for the velocity potential. c. Sketch the streamlines of this flow.

2. An airfoil is traveling through sea level air at 100 km/hr. a. Find the pressure and the Cp at the stagnation point and at a point on the airfoil where the local velocity is 50 m/sec.

b. If the airfoil is traveling at 700 feet per second what is the value of the stagnation Cp?

3. If a complete vacuum is produced on the upper surface of a wing and the lower surface is at stagnation pressure, what is the wing CL when the freestream Mach number is 2.5?

4. An airfoil is tested in a low-speed wind tunnel at the correct Reynolds number and found to have a minimum value of Cp of -2.0 at the design angle of incidence. The airfoil is to be used as a hydrofoil, operated 4 ft below the surface of the ocean. How fast can the boat travel before cavitation becomes a problem. (Cavitation occurs when the local pressure drops below the vapor pressure of the liquid. Assume a temperature of 25C at which the vapor pressure is 3170 N/m2)

AA 200A Problem Set #3

1. It is sometimes claimed that airfoils generate lift because the flow on the top of a thick, cambered airfoil has a greater distance to travel than the flow on the lower surface. If one considers two particles that are near each other at the leading edge, the particle traveling over the upper surface must travel faster if it is to meet its partner (which travels over the lower surface) at the trailing edge. This concept is entirely wrong. It is based on the assumption that the two particles do indeed reach the trailing edge at the same time. Consider a thin airfoil designed so that = constant and CL = 1.0. How far has a particle on the lower surface traveled by the time a particle on the upper surface reaches the trailing edge? 2. A line vortex of strength is located a distance h above a wall. a. Compute the motion of the vortex (its position as a function of time). b. If a uniform flow of speed U0 is imposed and the vortex is held in place, compute the force on the wall, assuming that the pressure on the other side is freestream static. 3. Four singularities are arranged in a line as shown below. They include a source, a vortex with counter-clockwise circulation, then a sink and another vortex.

The singularities are equally spaced, each a distance dx from its neighbors. a. Find circulation around a large circle of radius R. b. Find the total force on the system of singularities and indicate its direction. 4. Sketch the image system and write the expression for the force on the upper vortex in the system shown below.

5. A closed body can be simulated by a pair of vortices aligned at right angles to the flow, as sketched. Find the distance from the stagnation point to the center of the body, x, and make a rough sketch of the streamlines outside the body. Express the answer in terms of U, L, and .

AA 200A Problem Set #4

1. An airfoil pressure distribution at M=0 is shown below. a. Use the results from thin airfoil theory to estimate the CL at = 0. b. Estimate the pitching moment coefficient when CL = 0. c. Does this airfoil have any thickness? How can you tell?

2. Here is the incompressible Cp distribution over an NACA 0012 airfoil at zero angle of attack. Compute and plot the Cp distribution of the NACA 2512 airfoil at the ideal angle of attack and M = 0 by adding thickness and camber perturbations. (2% parabolic camber line)

NACA 0012 Pressures x/c Cp 0 0 .005 .360 .025 -.241 .05 -.378 .10 -.411 .20 -.399 .30 -.350 .40 -.288 .50 -.228 .6 -.166 .7 -.109 .8 -.044 .9 .044 .95 .094 1.00 .30

3. The 2-D airfoil shown below is moving through the air at M = 1.6 at an angle of attack of 2. The thickness to chord ratio is 0.05 with the maximum thickness 30% aft of the leading edge (x/c = 0.3).

Use linear theory to predict the following: a. pitching moment about the aerodynamic center b. center of pressure location c. angle of zero lift d. drag due to thickness, camber, lift, and total inviscid Cd. e. If the angle of attack is allowed to vary and the skin friction coefficient is .002, what is the maximum L/D?

4. Show that that the Kutta-Joukowski formula for lift: holds for supersonic flow over a 2D flat plate at small angles of attack. Do this by computing the velocities, lift, and circulation around the plate using linear supersonic theory. 5. Use thin airfoil theory and the Prandtl-Glauert compressibility correction to estimate the freestream Mach number at which supersonic flow appears on thin airfoils. Assume that the camber line is parabolic and is at the ideal angle of attack. Plot the critical Mach number vs. CL. An interesting extra problem (not graded): Use thin airfoil theory to compute the shape of the airfoil that produces the constant vorticity distribution used in the last problem set (CL=1.0). Plot the camberline shape. (Use the same scale for x and y axes so you can see the true shape.)

AA 200A Problem Set #5

1. One method of estimating the drag on a flat plate with some laminar and some turbulent flow is to assume that the flow is laminar up to the transition location, and then to assume that from there back, the flow is turbulent. The turbulent layer does not start out with 0 thickness at the transition location, but rather acts as though it started upstream of the transition and developed to have the same momentum thickness as the laminar layer at the transition point. Using the formulas for flat plate skin friction, estimate the drag coefficient of a flat plate at a Reynolds number of 1,000,000 with transition at 20%, 40%, and 80% of the chord. 2. A very thin flat plate is set at an angle of attack of 2 degrees with a Reynolds number of 10 million. Assume that the plate is two dimensional (AR = infinite) so there is no vortex drag associated with three-dimensional effects and ignore any variation of skin friction with lift. a. Find the ratio of lift to drag, L/D at M = 2.0 b. Find the ratio of lift to drag at M = 0.5. 3. A group of ants has discovered that they can get out of the wind by staying in the boundary layer. They try to go no farther from the surface than 10% of the boundary layer thickness. If the ants are .1 inches tall, how far downstream from the leading edge of a flat plate would they have to stand in a 20 mi/hr (32.2 km/hr) wind if the boundary layer is: a. laminar b. turbulent An interesting extra problem (not graded): Based on Stratford's laminar separation criterion, what is the maximum Cl for a very thin airfoil that has a laminar boundary layer over its entire surface. This problem may be of interest to bugs.

AA 200A Problem Set #6

1. An elliptically-loaded wing (A) with 5% more span than another elliptic wing (B) with the same total lift has 5% higher bending moment at the wing root. If the lift distribution is changed so that the lift at the tips is reduced, we can build the higher span wing with the same root bending moment as the one with the lower span. Find the span efficiency (e) of wing A when it is designed with the same lift and root bending moment as wing B and the induced drag is minimized. How does the induced drag (in lbs) compare? Use lifting line theory and include only the first two symmetric terms in the Fourier series.

2. The distribution of Cl on a wing at two different angles of attack is shown below. The airfoil section Clmax is 1.4. a. Estimate the wing maximum lift coefficient. b. Does the wing have washout? c. Is the wing is probably swept backward or forward?

3. a. What is the section lift coefficient at the root of an unswept wing with taper ratio = 0 (triangular) with an elliptic distribution of lift at CL = 1.0? b. Does the wing have washout or wash-in? (Give a reason for your answer.) c. Would the wing be expected to stall first at the root or at the tip? (Do not use the computer program to answer this question.) 4. The simplest form of lifting line theory is shown below. Find the value of k so that the downwash, produced by the two trailing vortices, at the center of the wing is the same as the downwash produced by the trailing vortex wake of an elliptically-loaded wing with the same total lift.

An interesting extra problem (not graded): Use the Weissinger wing analysis program to examine the effect of aspect ratio, sweep, twist, taper, and angle of attack on the lift and Cl distributions of wings.

AA 200A Problem Set #7

1. A wing with 32 degrees of sweep is designed to fly at a lift coefficient of 0.45 and a Mach number of 0.85. You are to design the airfoil sections for this wing, defined in the direction normal to the leading edge. What would you choose for the section design CL and Mach number?

2. A fuselage of fineness ratio 8.0 is attached to a wing as shown below. Model the fuselage with a line of sources and estimate the effective flow velocity over the wing from the fuselage side out to a distance of 5 diameters. (Plot u/U0 vs. y at x = L/2.) Assume the fuselage is a parabola of revolution, r = x(L-x) / 2L . In incompressible flow the perturbation velocities are related to the source strength by:

3. Some canard aircraft have been designed with unusual twist distributions which enable the wing to achieve an elliptical load distribution in spite of the non-uniform downwash from the canard surface. a. Assume no vertical gap and compute the span efficiency factor (the ratio of induced drag of an elliptical wing to the induced drag of this system) if the canard span was 40% of the wing span and carried 30% of the total lift? Assume each surface is elliptically-loaded. b. What is the span efficiency if each surface is twisted to obtain minimum total induced drag? 4. A tail surface has 20% of the wing span. The designers are trying to decide whether to use a Ttail with a very large vertical gap (assume it is infinite) or a conventional tail with almost no vertical gap. Compute the span efficiency if both surfaces are elliptically loaded. Do this in the following cases and think about the results: a. Very large vertical gap, tail carries 5% of total lift. b. Very large vertical gap, tail carries -5% of total lift. c. No vertical gap, tail carries 5% of total lift. d. No vertical gap, tail carries -5% of total lift. 5. The tail of a simple model airplane is an all-moving surface whose angle of incidence, i, can be adjusted. Assume the center of gravity is located at the wing aerodynamic center (25% chord for this unswept wing). a. How much does the trimmed angle attack of the wing change when the tail incidence is changed by 1 degree? b. What is the static margin if the wing has AR = 8, the tail has an aspect ratio of 5, and the tail a.c. is located 3 wing chords behind the wing a.c.?

An interesting extra problem (not graded): The Polhamus suction analogy suggests that if we assume that the leading edge suction force acts in the lift direction rather than the chordwise direction, we will compute the correct vortex lift on a highly swept wing. For small angles of attack: If the AR is low, Kp is given by slender body theory: Kp = AR / 2. Use the Polhamus idea to show that for very low AR delta wings, Kv = .