Aerodynamic Flow simulation using the vortex cloud Method to

Predict the Performance of Lifting Bodies

By

Julian D. T. Rimmer

B.Sc. (University of Manitoba) 2001

A dissertation submitted in partial satisfaction of the

requirements for the degree of

Master of Science

tn

Mechanical Engineering

at the

UNIVERSITY OF MANITOBA

Committee in charge:
Dr. R.W. Derksen, Chair
Dr, A. Gumel
Dr. M. Tachie

2006
 THE I]NIVERSITY OF M,{NITOBA

FACULTY OF GR,{DUATE STUDIES

,(***lr
COPYRIGHT PERMISSION

Aerodynamic Flow Simulation Using the Vortex Cloud Method to

Predict the Performance of Lifting Bodies

BY

Julian D.T. Rimmer

A ThesisÆracticum submitted to the Faculty of Graduate Studies of The University of

Manitoba in partial fulfillment of the requirement of the degree

of
MASTER OF SCIENCE

Julian D.T. Rimmer @ 2006

Permission has been granted to the Library of the Universify of Manitoba to lend or sell copies of
this thesis/practicum, to the National Library of Canada to microlilm this thesis and to lend or sell
copies of the film, and to University Microfilms fnc. to publish an abstract of this thesÍs/practicum.

This reproduction or copy of this thesis has been made available by authority of the copyright
owner solely for the purpose of private study and research, and may only be reproduced and
copied as permitted by copyright laws or with express written authorization from the copyright
o\ilner.
 Abstract

Aerodynamic Flow Simulation Using the Vortex Cloud Method to

Predict the Performance of Lifting Bodies

by

Julian D. T. Rimmer

Master of Science in Mechanical Engineering

University of Manitoba

Dr. Robert W. Derksen, Chair

The purpose of this work is to develop and evaluate a vortex cloud method

which simulates the flow over airfoil sections at low Reynolds numbers. The

current state of the aft consists of a good base of modern experimental data,

but relies on modeling techniques based on and validated with high Reynolds

number flow tools. The method continuously injects elements of vorticity

within the flow filed and tracks their movement over time. The convective

velocity is determined from the inviscid velocity component due to bodies

within the flow field, interaction with other free vortices as well as a random

component. The random component is based on the "random walk" model

and is scaled to the Reynolds number. An assessment of the vortex cloud

model is presented for a range of angles of attack and a range of airfoils

along with the implementation of a graphical user interface that allows the

user to watch the flow development.

 Acknowledgements

I would like to thank my advisor, Rob Derksen, for suggesting the problem

and for his advice, support and patience as I worked through it. Thanks to
E.H. Price for financial, moral and scheduling support, without which

completion of the project would not have been possible. Thanks go to my

Dad for reviewing this report and providing excellent feedback as well as

consultation on comma use. Thank you Jerry for taking the time to read this

report so quickly and for having provided feedback. Finally, to Ainslie, thank

you for the tremendous amount of support and patience you have shown

towards me over the course of the project.

 ilt

Contents
Contents ...... iii
List of Figures ....... iv
List of Tables .....,.,.v
Chapter 1 : Introduction. ......... 1
Chapter 2: State of the Art.. ............7
2.7 Background ..... ......7
2.2 The Panel Method ......,....9
2.2.L The Source Panel Method ........ 10
2.2.2 The Vortex Panel Method ......... 11
2.3 Vortex'Cloud Methods....,. .....L2
2.3.L Vortex Dynamics and Vortex Clouds .,.......13
2.3.2 Hybrid Vortex Cloud Methods ...........15
2.3.3 Viscous Diffusion in Vortex Clouds ....17
2.3.4 Other Modeling Considerations .........20
Chapter 3: Problem Formulation and Results. ..,.......22
3.1 Vortex Panel Method .....22
3.1.1 The vortex Panel Method - Theoretical Background .....23
3.t.2 The Vortex Panel Method - Implementation .......27
3.1,3 vortex Panel Method - considerations for Airfoils .,..,..36
3.2 The Vortex Cloud Method ....,...4g
3.2.L Vortex Cloud Method - Theoretical Background .......,..49
3.2.2 Vortex Convection..... .....,.49
3.2.3 Modified Euler Method ,....51
3.2.4 Viscous Diffusion .....59
3.2.5 Panel-Vortex Interaction. .......,66
3.2.6 Sub-Panels..... .......69
3.2.7 Absorption and Destruction of Vorticity ... ..........73
3.2,8 Vortex Shedding ....76
3.2.9 Full Vortex Cloud Simulation,.,.. .......80
3.3 Visual Vortex Cloud Program ..........g3
3.4.1 Test Cases..,... ........g7
3.4.2 Results .,....g9
Chapter 4: Discussion and Conclusions.... .......96
Chatper 5: Recommendations...,. .,.......101
Bibliography ... ..105
 IV

List of Figures
Figure 1) Velocity induced by vortex element at s, . ...........24
Figure 2) Discretization of a body surface ........26
Figure 3) Flow chart of a vortex panel program ........28
Figure 4) Invert coupling matrix was developed as a standalone program
Inveft_Test0.. ,...,....33
Figure 5) comparison of data set and computed panel endpoints ,....,..35
Figure 6) Airfoil profile for a NACA 0012 section ........37
Figure 7) Coupling coefficients for panel 9 of 28 ..,....3g
Figure 8) Trailing edge flow, ...........40
Figure 9) The effect of the back diagonal correction routine on the net
circulation due to panel 9. .... ........43
Figure 10) Flow chart of vortex panel program Flow 0 for airfoils .......44
Figure 11) coefficient of pressure for a NACA 0012 Airfoil .........46
Figure 12) Exact vortex convective motion ...,..4g
Figure 13) Spiral path and irreversibility due to forward differencing.......,..50
Figure 14) Modified Euler estimate of vortex motion ..........52
Figure 15) Flow chart for forward difterence method .....,...53
Figure 16) Flow chart modified Euler method ....54
Figure 17) Spiral effect of the forward difference method ..,..,,...56
Figure 18) Results from the implementation of the modified Euler method,.57
Figure 19) Error associated with the forward and modified Euler methods for
a circular path and varying the time steps to travel 2n radians.,.,,........,..,58
Figure 20) Flow chart for VorticesO with random walk, .....63
Figure 21) Bins with initial random vortex positions ....65
Figure 22) Distribution of vortices after 200 iterations due to drift ......66
Figure 23) Vortex interaction with the body surface ..........67
Figure 24) Influence of surface vorticity in crose proximity ..,..,,,..69
Figure 25) Votex path demonstrating unnatural pathline caused by surface
vorticity applied at the pivotal point. .........70
Figure 26) The use of sub-elements reduces the near wall effects of coarse
discretization.. ......7t
Figure 27) Effect of increasing the number of surface panels on the vortex
pathline. .........73
Figure 28) Vortex shedding by offset method ..........77
Figure 29) Free vortices shed from the surface of a NACA 0012 airfoil..,.....B0
Figure 30) Flow chart for program FlowaO .......82
Figure 31) Screen shot of program Visiflow0 .....,....,83
Figure 32) Input pane to set up flow simulation ,......,84
Figure 33) Vorticity shed from body surface, VisiflowO at 0oangle of attack g5
 Figure 34) Vorticity shed from body surface, Visiflow0 at l8Ooangle of attack
....,....85
Figure 35) Flow chart for program VisiflowO.... .,.....86
Figure 36) Airfoil profiles selected for this study, airfoil data from Selig ef a/.
(1ee6) .........88
Figure 37) Vortex Cloud Model simulation of flow over a NACA 24L4 airfoil at
Re = 149,503, with 250 free vortices shed into the flow field. ....89
Figure 38) Vortex Cloud Model simulation of flow overa NACA 24L4 airfoil at
Re = 149,503, with 500 free vortices shed into the flow field. ....90
Figure 39) Vortex Cloud Model simulation of flow over a NACA2414 airfoil at
Re = 199,337, with 500 free vortices shed into the flow field. ....91
Figure 40) Vortex Cloud Model simulation of flow over a CLARK-Y airfoil at Re
= L99,337, with 500 free vortices shed into the flow field. ..,......91
Figure 41) Vortex Cloud Model simulation of flow over a FX63-137 airfoil at
Re = 199,337, with 500 free vortices shed into the flow field. ...........92
Figure 42) Vortex Cloud Model simulation of flow overa NACA 2414 airfoil at
Re = 149,503,500 vortices shed, for a - 0o,50, 10o and 15o. .......g4
Figure 43) Vortex Cloud Model simulation of flow over a CLARK-Y airfoil at Re
= L99,337,500 vortices shed for ø = 15o. ...,.......95

List of Tables
Table 1) Data set for a ciriular cylinder ..........29
Table 2) Comparison of surface velocity predicted by FlowO and the exact
solution .......36
Table 3) Surface vorticity, bound circulation and lift coefficient for a NACA
0012 Airfoil .........47
Table 4) Comparison of the coefficient of lift calculated with program Flowo
to the empirical data. .........,.47
Table 5) Range of random numbers created by randomO.... ......64
Chapter 1

f ntroduction
To date, the development of airfoils suitable for use under a prescribed set of

application-specific parameters has been very time consuming and

expensive. Prototyping, testing and comparison have been the designer's

tools when it comes to airfoil optimization and aerodynamic design. For

example, a designer might require an airfoil for use in a compressor with a

high duty at a specific angular velocity or for an aircraft that is designed to

fly at a given Mach number and altitude. Each of these cases has an optimal
airfoil geometry that is specific to the application and, without the use of an

automated testing and selection algorithm, it can take years to find an

optimal shape, if one is found at all. Progress has been made over the past

century in the development of procedures to predict the performance of

airfoils under specific test conditions but the inherent complexities in the

governing equations of fluid mechanics has proven to be intractable difficult

and, at times, overwhelming.

 2

Aircraft design is one of the most expensive and exhaustive design processes

known to man. The design of the aerodynamic characteristics of an aircraft

is both the first design step to be completed and the most important as the

aerodynamic performance is what determines the remaining design details.

In today's aerospace industry, computational fluid dynamics (CFD) is

regularly used to predict the performance of aerospace structures. It is an

unfortunate reality, however, that the procedure for selecting an optimal

aerodynamic shape has progressed little from the days when a designer

would test a profile in a wind tunnel and manually change the profile until a

suitable candidate is found. The designer must still modify the geometry and

rerun the solution with the new cross section, this often includes the

requirement to generate a unique computational mesh, a time consuming

and operator dependant task, for each iteration. In the end, the design is

generally considered sufficient if ¡t met the required airflow characteristics

with little thought as to how the airfoil could be further optimized. An

optimization scheme that considers the application and can select a suitable

airfoil with minimal human intervention would be a great asset to the

aerospace community.

The recent advancement in computational methods and computer speeds has

been a catalyst for optimization schemes for all industries and has a

particular relevance to the aerospace industry with the extremely high cost of

development and prototyping.

computational Fluid Dynamics (cFD) today is the product of decades of

research that has developed ever-increasing efficiency in solving the

 governing equations of fluid mechanics. Recent developments of relatively

fast and accurate CFD methods have the capability to simulate viscous flows

around complex geometries under a variety of conditions, These Eulerian

methods are based on the numerical solution of Navier-stokes equations and

should have no difficulty with flow separation. These packages are of

considerable value for the aerospace design community, giving the designer

the ability to calculate consistently a variety of conditions more quickly than

is possible in the fluid mechanics lab with a wind tunnel, Further progress in

the form of coupled solvers, advanced turbulence and boundary models have

allowed researchers and industry alike to achieve relevant and accurate

solutions to real life problems. The main hindrance to the use of this type of

tool for optimization and large scale solutions, particularly when particle

tracking is of interest, is that the solvers that are used to analyze the

problem are computationally exhaustive and are unsuitable for use in an

iterative optimization process. A further limitation to the current methods is

that, in general, any result that can be achieved in a reasonable time frame
is steady state and gives little information about the formation of eddies or

other macroaerodynamic flow characteristics. These macroaerodynamic

phenomena play an important role in most simulations, particularly with high

Reynolds number flows. If this additional level of resolution is required or

particle tracking considered important then the designer must use the much

more computationally intensive transient solutions and the complex

turbulence models designed for such simulations. In order for these methods
 4

to be of use in an optimization scheme, a reduction in the time required to

solve a problem is needed, probably by at least an order of magnitude.

An iterative optimization scheme requires two parts: an efficient solution

algorithm which discretizes the problem and calculates the flow

characteristics and an optimization algorithm which monitors the

performance of the airfoils and makes geometric changes in an effort to

optimize certain, user specified design parameters.

This work focuses on the first component of optimization to extend the

substantial effort to date to find a faster way to predict the performance of

an airfoil under certain application parameters. This work looks at extending

a modern inviscid vortex panel method, from which aerodynamicists can

obtain relatively quick solutions for aerodynamic problems, by implementing

a relatively novel, yet relativey simple, numerical method, the vortex cloud

model. The developers of this method have suggested that it is suitable to

simulate viscous flows over complex geometries and is capable of predicting

flow separation. The vortex cloud model comprises a vortex panel method

coupled with a Lagrangian particle tracking algorithm. The particles in this

case are free stream vortices, which are shed from the surface of the

geometry under study. The important element of the vortex cloud method is

the ability to model viscous effects by way of a diffusion model, named

'random walk', which prescribes the way that the free stream vortices

interact and move through the flow field. This method has been used

effectively to demonstrate Brownian motion and is considered useful in

modeling viscosity in aerodynamic simulations. This model is advantageous

to this work because of the inherent computation efficiency of solving a

largely inviscid problem which has the capability of inferring viscous effects

from the movement of the free stream vortices.

The input parameters of the solver include the airfoil geometry and flow

application characteristics such as altitude and velocity. The Algorithm also

features tunable inputs such as the number of surface panels, the number of

free stream vortices and the time step size, Three essential elements

characterize the method: first, a method for receiving geometric information

and discretizing the aerodynamic flow problem; second, a method for solving

the flow regime, including the transport of vortices based on the local flow

characteristics; and third a method to model the viscous diffusion of the free

stream vortices, thereby introducing the effects of viscosity.

The aerodynamic calculations are performed using a vortex panel method

flow solver; the surface voriicity and flow field are used to calculate the

velocity of the free stream voftices. In order to model the viscosity, a

random perturbation is applied to the particle position, as one would see in

viscous flows. The surface vorticity is then recalculated from the flow field

and is used to calculate aerodynamic forces. The free stream vortices are

plotted at intervals throughout the analysis and give a visual indication of the

developing flow field.

Implementation of the vortex cloud method offers the following advantages:

1. Vortex methods consider only the small portion of the flow field where
vorticity is present
 2. Vortex flows can easily accommodate complex geometries as they do

not employ a computational grid

3. In the computation of the velocity field, mass conservation is satisfied

directly

4. When vortex methods are used for external flow simulation in

unbounded domains, boundary conditions can be applied such that the

computational domain is not limited to a finite size,

The aerodynamic simulation code developed in this thesis will be used to

produce a number of results. It will be used to demonstrate the

effectiveness of the viscous diffusion as well as the effect of the tuning

variables on the flow field. It will be used to demonstrate that this method

provides a good visual representation of the flow field and will demonstrate

the efficiency of the method by selectively omitting traditional solution

optimization routines. The current work extends the state of the art by

exploring the value of the vortex cloud for modeling low Reynolds number

flows. A novel method of shedding vorticity is presented to increase

computational stability and similarity to real flows. The voftex cloud method

is also examined for its suitability for use as the solution tool for an

aerodynamic optimization process.

 7

Chapter 2

State of the Art

Vortex cloud methods have been used over the past 20 years in varying

applications. As there has been limited work in the field, some discussion will

deal with vortex cloud methods which are not directly applicable to the

present work.

2.L Background
Early efforts at simulating aerodynamic flows were limited by our inability to

obtain analytical solutions to the Navier-Stokes equations to all but a few

idealized examples. Real world problems required the development of more

sophisticated design tools or a simplification of the Navier-stokes model that

would deliver accurate solutions to aerodynamic problems. The first such

simplifications came in the form of inviscid models which allowed us to obtain

reasonably accurate estimates of lift, pitching moments and induced drag for

flow over streamline bodies at high Reynolds numbers. While these methods

provide a good estimation of certain parameters, they are critically flawed in

 I

their inability to estimate viscous drag and are, therefore, inadequate for

obtaining the comprehensive performance information required to make

educated design decisions.

Progress was made by approximating an airfoil as an equivalent flat plate.

This approximation led to surprisingly accurate results in estimating

minimum drag, a big advancement over previous models. Although, these

results were promising, limitations were found when aerodynamicists used

the model at higher angles of attack. Viscosity proved to be a continuing

issue for airflow simulations and computational aero structure modeling by

way of the flow separation observed at high angles of incidence,

The problem of modeling the viscous effects in aerodynamic simulation can

be partially addressed through the use of Prandtl's boundary layer theory.

Prandtl's boundary layer theory provides a better understanding of the flow

characteristics in the near-wall regime. Researchers realized that this model,

iteratively coupled with the simulation models of the time, would produce

good results for flow over an airfoil. once the boundary layer flow was

solved, it could be used as a boundary condition for the free stream flow

domain. Unfortunately, as the boundary layer model provided no consistent

or universal method for dealing with flow separation, it gave way to the more

sophisticated methods of today, although the procedure is regularly used as

the foundation of modern aerodynamic simulation.

 9

2.2 The Panel Method

The underlying solution method used in the current work is a vortex panel

method (Martensen, 1959). This type of method has been around for many

years with its origins stemming from elementary airflow analysis, Panel

methods discretize the geometry under study into discrete line segments, or

panels, in order to simplify it to a manageable state. The panel velocity is

then used as an approximation of the velocity of the potential flow near the
geometry. Pressure distribution and lift can be derived from the surface

velocity distribution. While the panel method is less accurate than a full

Navier-Stokes solver, it has the key advantage of not requiring a

computational mesh. complex solvers require that the flow field be

discretized into volume elements whose properties are governed by the

fundamental equations of fluid mechanics and their relationship with their

neighboring elements. This group of elements, or mesh, is generated to

describe the flow field. When a panel method is used, it is the geometric
surface only that is discretized. It has much fewer computational points and

is, therefore, faster and more suitable for optimization schemes than a solver

employing a fully discretized flow field. The validity of the panel method, as

well as the usefulness of the method for computing aerodynamic forces, has

been demonstrated (Pfeiffer, 1989).

Kellog (1929) discussed the merits of modeling incompressible inviscid flows

by using a boundary integral solution to Laplace's equation,y,ó:0, for the

 10

velocity potential. In his book on potential theory, Kellog developed an

integral equation representing uniform flow over a body, The method most

commonly used by aerodynamicists to solve Laplace's equation is by

singularities. Singularities are elementary flows which have infinite velocity

at their center, whose flow functions satisfy Laplace's equation and may be

superimposed to build flow fields. Kellog's work represents the earliest form

of suface singularity model in which the body surface is replaced by a

surface singularity distribution. Panel methods solve the flow field by solving

the strength of the singularities over discrete sections of the geometry

surface, or panels. The most common types of singularities used in these

methods are the source, the doublet and the vortex. Generally, the

singularity is spread over the panel but is typically solved at a specific point

on the panel, the control point.

2.2.7 The Source Panel Method

The two common types of panel methods discussed here are the source

panel method and the vortex panel method. The fundamentals of the source

panel method can be traced back to the first half of the twentieth century.

Source panel methods place a point source on each panel that describes the

surface and contributes to the velocity of the potential flow on every other

panel. Kellog's integral equation can then be written for each pair of panels

in conjunction with the Neumann boundary condition, which states that the

velocity normal to the panel surface is zero. The system of equations is then
 11

solved with the source strength describing the potential flow around the

geometry. The major limitation to the source panel method is that it is

unsuitable for modeling flow around lifting bodies.

2.2.2 The Vortex Panel Method

In contrast with the source panel method, the vortex panel method describes
the flow field using surface vorticity. Unlike a source distribution, a surface

vorticity distribution provides a direct interpretation of the flow field by a

fundamental similarity to the physical problem. In fact, the surface vorticity

method is a good representation of the infinite Reynolds number flow of a

real fluid. Martensen (1959) extended his boundary integral theory for airfoil

cascades. Martensen represented the viscous boundary layer by a

distribution of vorticity adjacent to the body's surface. The Dirichlet

boundary condition is satisfied by imposing a zero tangential velocity on the

panel surface; this also leads to a direct calculation of the surface vorticity

and a solution of the potential flow field.

Jacob and Reigels (1963) first implemented this theory for numerical

modeling with a computer, although several issues needed to be resolved

before the method would garner acceptable solutions to physical problems.

Wilkinson (L967) did much work to alleviate computational problems and his

work has also led to the use of this method for modeling of airfoil cascades

(wilkinson, 1969). Further development of MarLensen's method, done by

 L2

Nyiri and Baranyi (1983), among others, led to the surface vorticity method

being a useful predictive tool for aerodynamic simulation.

Development of the vortex panel method continues today with recent

ínvestigations into the treatment of the trailing edge. Xu (1998) has

demonstrated that the addition of a control point downstream of the trailing

edge can help to deal with the different air velocities on the pressure and

suction sides of an airfoil joining at a stagnation point.

Where difficulties arise is the surface vorticity method's inability to model

boundary layer separation and for these problems a novel approach has been

proposed in the vortex cloud model. A good summary of panel methods and

their applications is available through two recent reviews by Hess (1990a &

1990b), the survey by Erickson (1990) and the book by Katz and plotkin

( lee1).

2.3 Vortex Cloud Methods

Separated flows and wake analysis pose serious problems to traditional

surface vofticity methods. Vortex cloud analysis attempts to model these

phenomena by discretizing the vorticity inherent to them into small discrete

voftex elements and observing the motion of these elements through the

flow field. Often referred to as'discrete vortex models', vortex cloud models

use a Lagrangian method to track the vortex elements as they interact with

the body surface as well as surrounding elements.

 13

2.3.L Vortex Dynamics and Vortex Clouds

Vortex dynamics have been studied by several groups but the first serious

attempt at discrete vortex modeling can be attributed to Rosenhead (1931)

who studied the Kelvin-Helmholtz instability of voftex sheets. Abernathy and

Kronauer (L962) extended Rosenhead's discrete vortex model to

demonstrate a progressive formation of a von Karman type vortex street as

the outcome of an unstable sheet vortex configuration.

The circular cylinder has played a large role ín the study of vortex dynamics.

Gerrard (t967) used the reflection system, where a vortex, equal in strength

and opposite in sign, is located inside the body surface, in one of the first

attempts at simulating a flow field using vortex dynamics for the solution of

the potential flow field due to a point vortex near a cylinder.

The vortex cloud model is compatible with the surface vorticity model and,

arguably, it is a natural extension of it, The governing equations for the

transport of the discrete vortices are analogous to the equations that solve

the potential flow. chorin (1973) suggested that, while the vortex cloud

method had come under some criticism (Takami, Lg64) and Moore (1971),

the large errors observed were due to a fundamental flaw in the

approximation of a vortex sheet or continuous vorticity by a point vortex with

an unbounded potential flow. chorin suggested that by smoothing the

analysis, the point vortex potential flow becomes bounded and delivers useful

results.
 14

Recent work by Kuwahara (L973) and Sarpkaya (1975) employs this model

to study separation around a flat plate set obliquely to a uniform flow,

Clements (1973) explored the similar problem of vortex street development

behind a rectangular body with a flat trailing face. one observes from

looking at these works that Kuwahara (1973), Clements (1973) and

sarpkaya (L975), along with Katz (1981), and cortelezzi et al. (L997), have

chosen to model the continual shedding of vorticity from a sharp edge by

introducing a new vortex element into the flow at the beginning of each time

step. The strength, position, and velocity of the shed vortex elements are
chosen, more or less arbitrarily, to satisfy the author's version of the

unsteady Kutta condition. In the work of Krasny (1991) and Nitsche and

Krasny (1994) the vortex elements are released at the edge of the body

instead of being placed into the flow at some arbitrary location. As a result

their comparison with the experimental work of Didden (L979), as presented

in Nitsche and Krasny (1994), is promising. Jones (2003) provides a solution

to the unsteady Kutta condition by using an alternate form of the conditions

associated with the imposition of the steady Kutta condition in his study of

flow around a moving flat plate.

The applications for vortex cloud models are quite broad. The work by Chan

et al. (2000) investigated the behaviour of a premixed turbulent flame. The

study used the vortex cloud model to solve the turbulent flow field and has

shown that the computed field and the turbulence scalar statistics are in

good agreement with the experimental results. Nakashima and Ono (2000)

used the vortex cloud model to predict the thrust, energy consumption, and
 15

propulsive efficiency of fish and cetaceans in water and found that the thrust

decreases due to the increase in the lift force as the normalized propulsive

speed increases when all the joints move in phase.

vortex cloud methods have been a useful and predictive tool for

aerodynamicists, providing insight into the evolution of jet and wake flows.

There are limitations, however, in the inviscid approximation employed by

the method. In many flows, viscous effects are responsible for the
generation of vorticity at the body boundaries and an approximation of

viscous effects, including diffusion, is necessary.

2.3.2 Hybrid Vortex Cloud Methods

Viscous effects have been modeled by fluid dynamicists for years in finite

difference and finite element discretized solution methods. These methods

typically solve the viscous Navier-stokes equations using numerical

approximations for the viscous terms. Traditionally, these have taken the

shape of time-averaged Reynolds Average Navier stokes (RANS) or shear

Stress Transport (SST) models. Recently, more computationally demanding

schemes such as Large Eddy Simulation (LES) or Detached Eddy Simulation

(DES) methodologíes have been available (Frohlich and Rodi 2002).

In an effort to model the viscous effects of the boundary layer and,

specifically, separation, there has been a significant effort to combine the

efficient, inviscid methodology of the vortex cloud method with viscous

Navier-Stokes solvers. This typically employs a grid in the near wall region
 16

for viscous analysis coupled with a grid-free analysis in the unbounded flow

field. One of the first successful attempts at this implementation was

completed by Spalart et al. (1983) in the study of various flow problems,

most of which had massive separation. spalart's results, when compared

with experimental results, demonstrate the reliability and the general

accuracy of the hybrid method, with little dependence on empirical

parameters, In Spalart's scheme, the vortex flow determines the

instantaneous pressure distribution, while the boundary layer flow

determines the separation points.

Many of the complex features of the flow past a circular cylinder, over a wide

range of Reynolds numbers, are correctly reproduced. spalart et al. (1983)

also considered flow over and airfoil in dynamic stall; this represents the first

application of the vorLex cloud method to simulate flows over moving

geometry. A NACA OOt2 airfoil as studied, oscillating in pitch about the

quarter chord position from 5o to 25o with Re = 2.6 x 106. Results were in
good agreement with experimental data and predicted the phenomenon of lift

coefficients being significantly higher than their steady state values at

equivalent angle of attack, as seen in experiment (Carr et al. Lg77).

The first blending of finite-difference and vortex methods was presented by

Chen et al. (2002)

There has been significant development of a vortex in cell method where the

vortex cloud method is used in conjunction with a computational grid which

discretizes the entire flow field for the sake of a direct numerical solution of
 L7

viscous diffusion. This is outside of the scope of this discussion but a

detailed account of the method can be found by Barber and Fonty (2005)

2.3.3 Viscous Diffusion in Vortex Clouds

The current work focuses on methods that are suitable for optimization

schemes and, while these models have been used to varying levels of

success, their demands on computational resources to present has been

prohibitive. The design of a particle scheme to handle viscous effects that

would directly model turbulence is compelled by the desire to maintain the

Lagrangian character of the vortex cloud method.

In a transient or time-stepping scheme, a natural approach is to consider the

inviscid and viscous parts of the equations separately. This is known as

viscous splitting and is rooted in classical fluid mechanics. Viscous splitting

for vortex methods was pioneered by chorin (1973). Chorin developed an

algorithm that split time steps into sub steps where:

1. Vortices move with local velocity - the inviscid portion

2. A diffusion algorithm is applied to the free vortices - the viscous
portion

In the method introduced, known as random walk, particles undergo a

Brownian-like motion and so simulate the effects of diffusion for high

Reynolds number flows. The crux of the popular method is to subject all of

the free vortex elements to a small perturbation of location, or small shift in

their position. This produces a diffusion-like behaviour of the fluid under

study and has been shown to emulate viscous eftects and calculate drag
 18

coeffic¡ents, for a high Reynolds number flow over a circular cylinder, which

are in good agreement with experimental data (Chorin, tg73). Chorin

(1978) extended his work to facilitate integration into hybrid algorithms and

to improve on the generation of vorticity at boundaries. Chorin was aware

that, for the random walk method to be useful as an alternative to Eulerian
grid-type Navier Stokes solvers, it would have to be able to model the

viscous boundary layer, while being suitable for larger scale phenomena such

as von Karman vortex street development in downstream wakes.

Porthouse and Lewis (1981) developed a random walk technique as a natural

extension of the vortex cloud modeling that was nearly identical to that of

chorin. The use of the vortex cloud method with random walk is often
referred to as the'random vortex method'. while Chorin's methodology has

been widely adapted for modeling viscous flows in combination with vortex

cloud methods, there was some skepticism about the validity of using viscous

splitting to compute adequately the solutions of the Navier Stokes equations.

Beale and Majda (1981) provide a rigorous proof that viscous splitting

algorithms, which are the underlying design principle for the random vortex

method, converge to solutions of the Navier Stokes equations at a rate which

improves as the viscosity becomes smaller.

Lewis (1991) did extensive work on the application of the random vortex

method to modeling flow fields around lifting bodies. In his thorough

discussion on random vortex methods, Lewis presents applications to bluff

body flows as well as to lifting bodies such as airfoils and cascades, Lewis

addresses the issue raised by Porthouse and Lewis (1981) and Spalartet al.
 19

(1983) and deals with the seeming inability of the random vortex method to

cope with boundary layer stability and the appearance of premature stall.

Lewis develops a hybrid model that stabilizes the upper surface of the airfoil

by enforcing potential flow and the lower surface being represented by a full

random vortex cloud model. Predicted flow pattern and lift and drag data

were presented for a NACA 0012 airfoil at 50 angle of incidence and Re = 106

and are compared to experimental results by Miley (1982). The predicted lift

coefficient is in reasonable agreement with the data; the drag coefficient is,

not unexpectedly, under predicted due to the virtual elimination of form drag

and associated separations. The predicted surface pressure distribution is in

very good agreement with a potential flow surface vorticity solution, Lewis'

seminal work also describes vortex cloud methods from first principles to

deal with shear layers, boundary layers, periodic wakes, bluff-body flows,

and cascades. Due to the sophistication of Lewis' models, the current work

will be fundamentally based on his airfoil simulation structure.

Borthwick and Barbe r (L992) describe a Biot-Savart vortex cloud model for

simulating the flow patterns which occur when a single high-velocity inflow

jet is used to stir the fluid within a circular container. Borthwick mapped the
circular perimeter of the container onto a rectangle by means of a Schwarz-

Christoffel transformation. A potential flow solution is then obtained and this

is transformed to give the potential flow inside the circle. Discrete vortices
are added at the inlet of the physical system in order to model the inflow

shear layers. Velocity components resulting from the discrete vortices and

their images in the walls of the cylinder are superimposed on the uniform
 20

potential flow solution. Viscous effects are incorporated through the use of

the random walk method, From the results it is shown that the discrete

vortex method does predict qualitatively the important features of jet-forced

reservoir flow.

Pereira (2003) presented an implementation and initial test of a sub-grid

scale model coupled with the random vortex cloud method under the same

test conditions as Lewis. Pereira demonstrated that the random vortex cloud

method with turbulence modeling can improve results obtained by use of the

voftex cloud method alone although the cost of the turbulence modeling for

marginal benefit still needs to be evaluated.

Vamos et al. (2003), in an effort to increase the computation speed and to

reduce the required memory of random vortex cloud methods, derived a

'global random walk' method in which the free vortices at a given site are

simultaneously scattered following the binomial Bernoulli repartition. Vamos

found that the computation time is reduced three orders of magnitude with

respect to individual random walk methods, while obtaining good simulation

of transport in unbounded domains, using normal size grids.

2.3.4 Other Modeling Considerations

As voftex particles are allowed to freely convect,'gaps' may appear in areas

where the flow strain is large and the model loses resolution of the flow field

as a result. One solution to this issue is the application of a 'remeshing'or

'redistribution'scheme, where the free vortices are periodically re-located

 2L

onto a regular lattice or grid, and their circulation weights are interpolated to

the new locations.

The addition of remeshing to vortex cloud methods made long-time unsteady

calculations possible (Koumoutsakos, 1993) and has been embraced by

various investigators who have produced remarkable results (Cottet et al.,

1999) and (Ploumhans and Winckelmans, 2000). Conversely, it can be

argued that the grid-free nature of vortex methods is undermined by the

requirements of remeshing

In an effort to decrease the convergence time of the random vortex cloud

method, Barba et al. (2005) introduced an alternate method of modeling

viscous effects, known as the core spreading method. In this method, the

vortex core size is adjusted in the spatial adaption, Numerical experiments

demonstrate considerable increase in accuracy in comparison with standard

remeshing schemes used with vortex methods.

An alternate method of modeling the viscosity in a vortex cloud method is

one where the voftex strength is adjusted according to a viscous relation.

There has been little work extending on this thought and it, therefore, will

not be examined.
 22

Chapter 3

Problem Formulation and

Results
The vortex cloud solver contains three elements. A vortex panel method

receives the geometrical information and computes the surface vorticity. The

surface vofticity is used to shed free vortices into the flow field where the

convection algorithm computes their movement through the computational

domain. A graphical user interface was developed to allow the user to input
simulation parameters easily as well as to watch the developing flow field.

3.1 Vortex Panel Method

An inviscid potential flow solver, Martensen's surface vorticity method, is

used to model the flow over the simulation geometry. There is a large body

of work in the area of panel method theory and its application to

aerodynamic simulation is well understood. An extended discussion of the

 23

theoretical background of panel methods can be found in many aerospace

and fluid mechanics textbooks and therefore only a brief summary of the

relevant theory and equations is given here. Special consideration will be

given to unique algorithms developed in the course of the present work.

3.1.1 The Vortex Panel Method - Theoretical

Background

In all real flows around a body, there is a boundary viscous shear layer, the
boundary layer, adjacent to the body surface. Outside of this layer, the fluid

will have a finite velocity, v,. It is the friction between the surface and the
airflow that causes the velocity parallel to the body surface to be zeroi this is

known as the Dirichlet boundary condition. The large velocity gradient in the

highly viscous layer causes substantial vorticity. As viscosity approaches

zero, and the Reynolds number approaches infinity, the layer becomes

infinitely thin and becomes a vortex sheet. It is therefore analogous and

physically representative of the airflow to model this layer by replacing the

body surface with a vortex sheet, y(s), where 7(s)is the strength per unit

length at location son the body surface. If we apply the Dirichlet boundary

condition to the problem to ensure zero velocity on the surface, we can say

that the velocity immediately outside of the viscous layer is equal to the

surface vorticity at that point.

v, : y(s) (3.1)
 24

We can also infer that the velocity at which the surface vorticity is convecting

downstream to satisfy Dirichlet is given by:

v"o,, =|rtt) (3.2)

For the two-dimensional body in Figure 1, the velocity dq,,,,, induced at m by

incremental vortex element n of strength Il, =y(s,,Þs,, at location s,, is given

by the Biot-Savart law in a simplified form, namely:

dQ,,, =+)ø-
zØru,
(3.3)

Where r,u, is the straight line distance from point n to point m.

Figure t) Velocity induced by vortex element at .str

For computational simplicity, it is beneficial to write the x and v components

of velocity dg,,, i

du,,, = Lk)4'' rin6,,,,, =(!,,-y,,)44 (3.4)

/ffu^ LTTT
 25

dv ,,u = +t
/ffu*
cosø,,,,¡ = (* ,,, - r,,)+4
Lffi,i,,,
(3.s)

The induced velocitydv,,,,is given by:

dv,,u, =
+l f,(,,,)a,,,

The Dirichlet boundary condition can then be satisfied by Martensen's

boundary integral equation for plane two dimensional flow (Lewis, 1991):

t,
-ir\,)+ftr(t,,,, r,,þ(r,,)dr,,+ zz*(cos d, cos 8,,,+sinasinfl,,)= 0 (3.6)

Where -)rG,,) is the self induced vorticity at n and represents the surface

velocity for non-convected surface vofticity and w*(cosri,cosp,,,+sinasinfl,,) is

the free stream velocity, parallel to the surface at s,, A(s,,,,s,,) is the coupling

coefficient and is given by:

k(s,,,,
1[ (y,, - y,,)cos 8,,, - (*,,, - x,,)sin 8,,, (3.7)
s,, )=
2ol (*,, - *,,)t * (y,,, - y,)'
The body geometry is divided into a finite number of discrete elements of

vorticity, or panels, of length the center of which is known as the control or

pivotal point (r,,,!,,) as seen in Figure 2.

 26

@,*r,Y,*r)
(x,,,Y,, As,
Panel Data
Point

Figure 2) Discretization of a body surface

In Figure 2, the body geometry is represented by a series of short straight

lines of length:

A,s,,=ffi (3.8)

The pivotal points, traditionally located at the center of each panel are given

by:

,,, : *{,.,)l
+(x,,
,Þ (3.e)
Y, =;(r,, +r*,)
)

If, for simplicity, we assume that the self-induced vorticity is zero,

integrating (3.6) and rearranging we arrive at:
M

I
n=l
K(r,,, , r,, Þ(",, ) - -(I * cos 8,,, - v* sin B ,,,
(3.10)

Equation (3.10) is the fundamental equation of the vortex panel method. We

expand (3.10) to develop a system of equations with Mequations and

 27

M unknowns of surface vorticity y(t,,) ov writing it for each of the pivotal

points. K(s.,s,) is known as the coupling coefficient and is expressed by:

K(s,,,r,,¡=4{o[(/"' -/")"otÉ"' -(""' -""]i"p"' l ,3.11)

., L (*,, - *u)' *(y,,, - y,,)' J

Representation of the surface by a series of straight line elements causes the

self-induced surface vorticity on m to be zero. While this is a reasonable

approximation, Lewis (1991) has developed an approximation for the self

inducing velocity due to surface curvature. As a result:

K(r^,r,,)= -i.+ K'(s,,,,r,)= -i-+ = -+-*@,*, Í þ,,-t) (s.rz)

where '/r(þ,,n t þ,,-t) is considered to be a reasonable estimate of the change

in slope Ap on m.

3.1.2 The Vortex Panel Method - Implementat¡on

The theory presented in section 3.1,1was used in development of the

current work. To this point, there are five major sections to the program as

outlined in Figure 3:
 28

Geometry Read in and Prepare Geometry

Data Data

Calculate the Coupling

Coefficients

lnvert Coupling Matrix

Calculate the Right Hand Sides

Solve

Figure 3) Flow chart of a vortex panel program

3.1,2.7 Input Data and Data Preparation

Data is read from a geometry file which describes the body surface in a

series of data coordinates moving clockwise from the leading edge. An

example data set, describing the surface of a circular cylinder is shown in

Table 1.

Whether the number of data points is odd or even is not important, so long

as they are sufficient in number to satisfy the user's required accuracy in

describing the body surface. The final point, however, should be equal to the

first point to ensure that the body is closed. The beginning of the data file
contains information describing the flow field. The angle of attack, free
 29

stream velocity, number of time steps and the number of surface panels,

which must be even, are all obtained from the data file. This was done to

speed up the development process and a more user-friendly manner of

inputting data was developed, discussed in section 3.3.

0.000000 0.000000
0.048943 0.309017
0.190983 0.587785
o.4L22r4 0.809017
0.690982 0.951056
1.000000 1.000000
1.309015 0.951056
r.587784 0.809017
1.809016 0.587785
1.951056 0,309017
2.000000 0.000000
1.951058 -0.309017
1.809016 -0.587785
1.587784 -0.809017
1.309015 -0.951056
1.000000 -1.000000
0.690982 -0.951056
o.412274 -0.809017
0.190983 -0.587785
0.048943 -0.309017
0.000000 0.000000
Table 1) Ðata set for a circular cylinder

This data is then prepared for the simulation. This preparation includes:

i. Calculation of the panel end- and pivotal-points

ii. Calculation of the sine and cosine of the panel slope as well as the

panel length

¡ii. Calculation of the change in slope for each panel

The panel end points are derived from the input data set and are spaced

around the surface of the body according to a cosine distribution:

 30

(",,)= 9+""or(, +ndo) (3.13)

where c is the chord length, d0 =4 and the

¡/2 9 t.m ensures that the

simulation profile starts at (O,O). This equation will concentrate panels, for

elliptical horizontal profiles, near the leading and trailing edge where it is

desirable to have increased resolution. The code implementation of this, for

an arbitrary number of panels, pDoc-)i_npanets is:

DTHETA : (2*pDoc->m_pl) /pDoc->i_npanels;

while ( (THETA/DTHETA) <pDoc->i npanels)
{

pDoc->f_l points IXI tD Xl : (pDoc->fI dnordl/2) +

( (pOoc->ff_cnoiOf/Z) J (pDãc->m pI+THETA) );
"o"
THETA=THETA+DTHETA,'

if (fabs(pDoc->fl pointstXl tD Xl )<MINVALUE)

pDoc->fI_polnts tXl tD Xl :O;

X++;
Ì

The code segment here determines the location of the points on the x-axis

only. The y-values are determined by using a linear interpolation scheme on

the original profile data and the newly computed x-values:

for (FN:0; FNcpDoc->m_NFILE; FX++)
{
for (N:0; N<pDoc->i npanels; X++)
{

XA:pDoc->fl_panelpts IFN] [D_X] ; / / oríqínaf dara

YA:pDoc->f1_paneJ_pts IFN] [D_Y] ; / / origínaf dara
if (FX::pDoc->m_NFILE)
{

¡g:pDoc->fJ- panelpts [1] [D X];

 31

YB:pDoc->fl_panelpts [1] [D y] ;

efse
{

¡B:pDoc->f l_pane j_pts I FN+ 1 ] t D_X l

YB:pDoc->fl paneJ_pts IFN+1] tD yl
Ì

if (N<: (pDoc->i_npanels/2) )

if (pDoc->fl_poinrs tNl tD_Xl > XA &&

pDoc->fJ- points tNl tD Xl < XB)
{

¡1:pDoc->f1_points IN] t D_Xl ;

pDoc->fl_points IN] t D_Yl
Inter(XA, XB, YA, YB, XI);

el- se
{

if (pDoc->f1_points tNl tD_Xl < XA &&

pDoc->fl_points IN] ID Xl > XB)
{

¡1:pDoc->fl_points tNl ID_X] ;

pDoc->f J-_points tNl I D_Yl
f nter (XA, XB, YA, YB, XI ) ;

Ì couxtl+;

This routine searches for the computed x-values of the panels pDoc-

>fl_points IN] lD_xl, which falls between two consecutive points from the
profile geometry, pDoc-)fl_paneJ-prs lFNl tD_xl êrìd pDoc->ft_panetprs

IFN+11 tD_xl, and then interpolates the y-value of the panel end point using

the function rnter O . This procedure is iterated a few times to ensure that
 32

all of the panel points are captured correctly. This routine is only required to

be called once in the simulation and ensuring that the points are computed

correctly in the initial set up is well worth the minor cost in processing time.

The pivotal points are calculated as the average of the panel end points:
X1:pDoc->fl_points IN] [D_X] ;
Y1=pDoc->f1_points IN] [D_y] ;
X2:pDoc->f1_points IN+1] [D_X] ;
Y2:pDoc->fJ-_points tN+11 tD_yl ;
if (N:=pDoc->i_npanels-1 )
{
X2:pDoc->fl points [0] tD_Xl ;
[0] [D_y];
, "r:pDoc->fl_points
pDoc->f] points tNl tD_Xl:(xI+X2) /2;
pDoc->fJ- polnrs tNl tD yl:(yl+y2) /2;

Finally, the change in slope, Lþ , on each panel was calculated according to

(3.12).

3,7,2,2 CouplingCoefficients
The coupling coefficientsK(s,,,,s,,) and K(s,,,,s,,) were calculated according to

(3.11) and (3.12), respectively.

3.7.2.3 Matrix Inversion

The program employs a Gauss-Jordan inversion scheme. The program

Invert-TestQ was written and tested by imporiing a matrix into the program,

inverting it and then crossing it with the original. The identity matrix was

found to be the cross product for all test cases ranging in size from a 3 x 3 to

a10x10.
 33

Geometry Read in and Prepare Geometry

Data Data

Calculate the Coupling

Coefficients

Calculate the Right Hand Sides

Solve

Figure 4) rnvert coupl¡ng matrix was developed as a standalone program

Inverr Testo

3,7.2,4 Right Hand Sr'des

The calculation of the right hand sides is a direct calculation of (3.10).

pDoc->fI _uinf : m_wfnf * cos (m_alpha);

pDoc->fl _vinf: m_wfnf * sin(m_alpha);
TEMP1 : -1 * pDoc->ff_cosineIS] * (pDoc->fl_uinf);
TEMP2 = -1 * pDoc->f1 sine [S] * (pDoc->fl vinf ) ,.

pDoc->fl rhs [S] : TEMP1 + TRMP2;

 34

3,7.2,5 Solver
The system of equations (3.6) is solved by multiplying the inverted matrix by

the right hand side vector, pDoc->fl-_rhs. The resulting vector is the

normalized surface vorticity which, when multiplied by the panel length gives

the suface velocity.

3.1,2,6 Validation

A validation of the vortex panel program FlowO was done using a circle as

the test profile. The geometry was positioned such that the leading edge

was located on the origin. The surface was represented by 18 panels and

follows the profile expressed by:

x = a(r_ cos /)
Y: asinú
Figure 5 shows the position of the circular profile in the domain as well as the

calculated panel end and pivotal points. From the figure it is evident that the

pivotal points calculated by the data preparation routine lie on the surface of

the body and are evenly distributed on the surface, as would be expected of

a cosine distribution on a circular profile.

 35

+lmported Data
Panel Endpoints

Figure 5) compar¡son of data set and computed panel endpoints

Batchelor (1970) states the exact solution for the surface velocity due to a

uniform stream U- as:

u, =2U-sinÓ

The surface velocity is compared for a circle of radius a=1, with its diameter
along the line Y=0 shown in Figure 5. Starting at (O,O), the panels are

numbered clockwise around the circle. Very good agreement is noticed in

Table 2 between Batchelor's exact solution and the output from the Flow0

program, indicating that the program is a useful tool for predicting surface

velocity. The percent relative error was found by subtracting the calculated
 36

velocity from the exact solution and then dividing by the exact solution for

each panel.

PANEL EXACT Visiflow % Error lRet.)

1 0.347296 0.349116 0.52o/o
2 1.000000 1.00095 0.10o/o
3 1.532089 1.53817 0.40o/o
4 1.879385 1.88944 0.54o/o
5 2.000000 2.01155 0.58%
6 1.879385 1.88947 0.54o/o
7 1.532089 1.53825 0.40%
8 1 .000000 1.00116 0.12o/o
9 0.347296 0.347868 0.16%
10 -0.347296 -0.347868 0.160/o
11 -1.000000 -1.00117 0.12o/o
12 -'1.532089 -1.53824 0.40o/o
13 -1.879385 -1.88947 0.54%
14 -2.000000 -2.01155 0.58o/o
15 -1.879385 -1.88944 0.54%
16 -1.532089 -1 .5381 7 0.40o/o
17 -1.000000 -1.00094 0.09%
18 -0.347296 -0.3491 15 0.52o/o

Table 2) Compar¡son of surface veloc¡ty pred¡cted by FlowO and the exact

solut¡on

3.1.3 vortex Panel Method - considerat¡ons for Airfoils

To this point, a vortex panel method has been developed which is applicable

to arbitrary geometric profiles. In order to apply the vortex panel method to

airfoils, there are some additional cons¡derations that must be addressed.

Airfoils generate lift and are generally quite thin, These can lead to

computational errors due to the proximity of opposing panels.

The airfoil used for the majority of the code development was 12 percent

thick NACA 0012 airfoil. Geometric data was obtained from the work of Selig

et al. (1996). Figure 6 is the airfoil profile for a NACA 0012 section.
 37

Figure 6) Airfoil profile for a NACA 0012 section

In order to examine lifting flows over a surface we apply the Kutta-Joukowski

theorem which states that the lift per unit span on a two-dimensional body is

directly proportional to the circulation around the body, or:

¡,'- p*V*l (3.L4)

We can determine the bound circulation by taking the line integral over the

body surface:

-t\
r=fz(")d'=I/G,,)4",, (3.1s)
n=l

Implementation of (3.13) has the unfortunate effect of increasing the

number of unknowns to N+1 with only N independent equations. Exploration

of methods to deal with this complexity will be discussed in section 3.2.L.2.

3.2.7,7 Back Diagonal Correction

From (3.11) it is noted that as (*,,-*u)'*(y,,,-y,,)'approaches 0,

K(s,,,s,)r*. For thin profiles, such as airfoils, the panel n, on the opposite

side of m,that will have the largest coupling coefficient,KG,,,s,,), will be the

panel N-¡ø *1, and therefore, the largest contribution to the surface vorticity.
 38

The coupling coefficients for panel 9 of a NACA 0012 airfoil represented by 28

panels are shown in Figure 7:

Ë
o
0.15
'õ
E
F
o
o
rì
; 0.10
.g
o.
a
o
O
0.05

. Figure 7) Coupling coefficients for panel 9 of 28

The considerably larger value of the coupling coefficient for the 20th panel is

evident and corresponds with the N-¡ø *1 location , i,e. 28-g*1=20.

Jacob and Riegels (1963) recommended a treatment to correct this

disproportionate influence by observing that, according to Kelvin's theorem,

the net circulation around the profile interior due to the surface vorticity of
 39

element m , y(s,,,)As,,, , should be zero. Implementation of this observation

takes the form of:

K,,.",,,r,,,
\ oppi tn,)= -IË
M*o ,,l,lro
K(s,,,s,,,)4s,, (3.16)

where opp=M -n+1. Equation (3.14) sets the opp element's coupling

coefficient to the negative sum of the other elements, for a panel z. It is

interesting to note that as n increases in value from 1 to M, opp decreases

in value from M to 1, This allows a computational simplification in that only

the back diagonal of the coupling coefficient matrix is affected.

once the back diagonal correction has been applied, the matrix has the

characteristic of any one equation being the minus of the sum of the other

equations and is therefore singular and non-invertable, This difficulty will be

addressed in the following section.

3,2,7.2 The Kutta Condition

In order to increase computation accuracy at the trailing edge, Wilkinson

(t967) suggested that the implementation of the Kutta condition should take
the form of an assurance that the static pressure, and therefore the surface

vorticity, at the two trailing edge segments on the upper and lower surfaces

should have the same magnitude. This is as opposed to ensuring a

stagnation point at the trailing edge implemented by a prescribed bound

 40

circulation. In order to ensure that the surface pressure approaches an equal

and opposite surface vorticity, the following constraint is imposed:

y(t,"):-Y6,"*t) r 3.L7)

The surface vorticity on the lower surface is negative in order to ensure

smooth flow leaving the trailing edge.

r(",,)------------>

Figure 8) Trailing edge flow

Wilkinson also noted that (3.15) eliminates one of the unknowns from the

system of equations given by (3.10) and if (3.10) is written for an element

n , we arrive at:

x(",t)y(',)*...* (r<(", ,")-x(r, te+rþ(s,,)+ ...*K(r, NÞG") =rhs,, (3.18)

where columns te and re+1 have been combined.

As the number of columns and unknown vorticity values has been reduced by

one, it is essential that the equations are reduced by an equivalent number.

There are three logical options that are considered for accomplishing this:

1. Simply eliminate the superfluous (te+I)th equation

2. Add the two equations that describe the trailing edge elements
 4t

3. Subtract these equations instead

Lewis (1991) suggests that the method of reducing the number of equations

by 1 that provides the best results is to subtract the two equations that

represent the trailing edge elements. The logic in this can be seen in Figure

8, where the induced velocity on the inner surface is represented by the

double arrows and flows positively clockwise around the profile, The effect of

the equation subtraction is the equivalent of determining the average

downstream flow in the region of the trailing edge.

The implementation of this requires that the (te+I)th coupling coefficient is

subtracted form that of the teth element and that the right hand side vector

is reduced in size accordingly. Whereas the matrix created during the back

diagonal correction was singular, the current procedure eliminates one of the

equations and the matrix ceases to be ill-conditioned.

For programming simplification, the implementation of the Kutta condition

was split into two parts, one for the coupling coefficients and one for the

right hand sides. This allows the coupling coefficients to be treated

completely independently, an advantage that will become clear when we

discuss the vortex cloud method

3,2.7,4 Solution
There are some additional considerations for the solution of the flow field and

induced body forces when airfoils are considered. The surface pressure

distribution is given by
 42

co:t (H)' (3.1e)

and is typically plotted against xllwhere x is the x-coordinate of the profile,

The lift coefficient can be calculated using the standard definition and (3.13):

,,=+;=# (3.20)
2'" *'
calculating the bound vorticity from (3,15) and substituting into (3.20)

arrives at:

,,:häl(",,)4,,, (3.21)

3.2,7.5 Implementation
The program Flowo was extended to include the considerations for airfoils

discussed in the previous sections and an updated flow chart is given in

Figure 10. The code implementation of the back diagonal correction and

equation (3.16) is:

for ( I:0,' I(pDoc-)l_npanels ; I++ )
{
SUM:O . 0,.

for (J:0; J<pDoc->i npaneJ_s; J++)

{

if (J!:(pDoc->i npanels-I-1) )

5g¡41:pDoc->fl_coupcoeff tIl tJl *pDoc->fJ_ dels IJ] ;

pDoc->f1_coupcoeff [ ]l [ (pDoc->i_npane]s-I-1 ) I :
-1*SUM/pDoc->f.l- dels [pDoc->i npane]-s-I-11 ;
 43

The effect of the back diagonal is shown in Figure 9 where the net circulation

due to panel 9 is shown for a NACA 0012 airfoil represented by 28 panels.

The back diagonal correction has the effect of ensuring zero net circulation

around the geometry interior by adjusting the contribution from panel 20.

-I Wifr Back Diagonal Correction

---e- Without Back Diagonal Co¡rection

Ê.
o
ñ
5ll
.=

{D

{=
o
5
E
J
o

j-t-r-r-r-r-r-¡-r'

8 10 12 14 16 '18 20 22 24
Panel

Figure 9) The effect of the back diagonal correction routine on the net

circulation due to panel 9,

 44

Geometry Read in and Prepare Geometry

Data Data

Calculate the Coupling

Coefficients

Backdiagonal Correction

Enforce Kutta Condition for

Coupling Coefficients

lnvert Coupling Matrix

Calculate the Right Hand Sides

Enforce Kutta Condition for Right

Hand Sides

Solve

Figure 1o) Flow chart of vortex panel program Flow O for airfoils

The code implementation for the Kutta condition's adjustment of the coupling

coeff¡cients is as follows:
 45

f or ( J:pDoc-)m_TE,' ¡4:pDoc- > i_npanel s - 1 ,. J++ )

{
for ( I:0; I<:pDoc->i_npanels-1; I++ )

if(J>pDoc->m_TE) COEFFtII tJl : COEFF[]l IJ+1];

else COEFFtII tJl : COEFFtTI tJl - COEFFII] [J+1];

In this algorithm, all of the coupling coefficients that are past the trailing

edge (except for te+i) are moved to the location n-1. The trailing edge

coefficients are subtracted, i.e. K,"=K,"-K,"*t, as indicated in (3.18).

The code implementation for the Kutta condition's right hand adjustment

sides is as follows and is largely the equivalent as the implementation of the

cou pling coefficients :

for (I:pDoc->m TE+1;T<:pDoc->i npaneÌs-1;I++)

{

if (I>pDoc->m_TE+1) pDoc->f1_rhsII-1]:pDoc->fl_rhs[]l ;

else pDoc->fl-_rhs II-1] =pDoc->fl_rhs II-1]-pDoc->ff rhs []l ;

pDoc->i_npanel s:pDoc- >i npane.l s- 1,.

3,2,7,6 Validation

The full flow solver, FlowO, with the above algorithms implemented shown in

the flow chart Figure 10, was validated using the NACA 0012 airfoil

previously discussed. The surface was represented by 30 panels using a

cosine distribution, The angle of attack was varied from 0 to 10 degrees.

Figure 11 shows the coefficient of pressure for several angles of attack.

 46

' Alpha = 0
o Alpha = 5
a Alpha = 10

o_
O

ÊÊtgÊÊÊeaanse

0.0 0.2 0.4 0.6 08 1.0

X/C

Figure 11) Coefficient of pressure for a NACA OO12 Airfoil

Table 2 shows the panel vorticity, the bounded circulation and the lift

coefficient calculated for 0, 5 and 10 degrees. With 30 surface elements, the

trailing edge is located at the 15th and 16th elements. Due to the

implementation of the Kutta Condition, it is noticed that the values at this

location are equal in magnitude and opposite in sign, even for the non-

symmetric flows of 5 and 10 degrees angle of attack,

The results in Table 3 were compared with the work of Abbott and von

Doenhoff (1959) and the predicted lift coefficients are in good agreement

with the empirical data as seen in Table 4.

 47

PanelVorticity
Angle of Attack
Element 5 10
1 11.9381 22.1485 32.1904
2 11.1252 16.7939 22.3347
3 '11.7985 15.3578 18.8002
4 11.8870 14.3760 16.7556
5 11.7761 13.6221 '15.3645
6 11.5785 12.9925 14.3077
7 11.3277 12.4285 13.4347
8 11.0570 11.9192 12.6906
I 10.7905 11.4639 12.0500
10 10.5267 11.0452 11.4796
11 10.2571 10.6443 10.9505
12 9.9670 10.2396 10.4343
l3 9.6365 9.8060 9.9009
14 9.4100 9.4830 9.4837
15 8.0544 8.0237 7.9320
16 -8.0544 -8.0237 -7.9320
17 -9.4100 -9.2655 -9.0505
18 -9.6365 -9.3937 -9.0794
19 -9.9670 -9.6'185 -9.1968
20 -10.2571 -9.7918 -9.2520
21 -10.5267 -9.9281 -9.2540
22 -10.7905 -10.0351 -9.2032
23 -11.0570 -10.1107 -9.0874
24 -11.3277 -10.1407 -8.8765
25 -11.5785 -10.0763 -8.4974
26 -11.7761 -9.8404 -7.8298
27 -11.8870 -9.3076 -6.6573
28 -11.7985 -8.1495 -4.4383
29 -11.1252 -5.3719 0.4223
30 -11.9381 -1.6367 8.6771

Predicted Circulation 0.0000 3.0022 5.9815

CL= 0.0000 0.6004 1.1963

Table 3) Surface vorticity, bound circulation and lift coefficient for a NACA

OO12 Airfoil

Angle of Attack FlowQ Abbott & v. Doenhoff

0 0.00 0
5 0.600 0.58
10 1.20 1.18

Table 4) Comparison of the coefficient of lift calculated with program Flowo

to the empirical data

 48

3.2 The Vortex Cloud Method

3.2.I Vortex Cloud Method - Theoretical Background

The vortex cloud, or discrete vortex, method comprises the surface vorticity

method, described in 3.1, modified to interact with free vortices which move

through the flow domain, The free vortices are convected through several

processes:

1. Interaction with the free stream velocity, W*

2. Interaction with the other free stream vortices
3. Random displacements or permutations of velocity

4. Interaction with the body surface

The first interaction is quite simple to model; the free vortex is assigned the

same velocity as the free stream velocity. The vortex will then move through

the flow field at a constant velocity. The other three mechanisms are quite

complex as their values are largely dependant on the local flow field and their

proximity to the body surface.

3.2.2 Vortex Convection

Consider the vortex pair of equal strength shown in Figure 12. If there is no
influence from an external system, these vortices will move in a direction

perpendicular to the straight line connecting them and will rotate about a

circle centered about the midpoint of this line.

Using a simplified form of (3.3),

 49

f (3.22)
U, =-
¿nr

the drift velocity for each free vortex can be calculated due to the strength of

the other.

f f
=-2nr

Figure 12) Exact vortex convective motion

If the simple system shown in Figure 12 is modeled using a time stepping

routine with the time elapsed between steps of a,t, the displacement that

vortex A will undergo is equivalent to

d: I at (3.23)
2tr
over several time steps, vortex A will experience a series of induced

displacements and will pass through (*,,y,), (xr,yr), (*r,yr) etc. This

displacement normal to the chord AB over time will lead to a spiraling

motion, outward from the circle. Furthermore, if an attempt is made to

retrace the path of motion by applying a negative time step-Âr, the motion
 50

is irreversible as is evident in Figure 13. Also evident is that the larger the

time step, A,t, and thus the induced displacement, d,,, the largerthe error

due to the spiral motion will be,

Figure 13) Spiral path and irreversibility due to forward differencing

Interestingly, this spiral motion can be considered to be similar, if not

analogous, to the effects of diffusion. We would expect that, if subjected to a

Brownian-type motion coupled with the induced velocity, the vortex A would

tend to move perpendicular to the chord AB and away from B, effectively a

spiral motion. It is for this reason that numerical error is commonly referred
to as "numerical viscosity". More discussion on this will be presented in the
section dealing with random walk.
 51

3.2.3 Modified Euler Method

3,2,3,7 Theoretica I Backg rou nd

In order to manage the errors evident in Figure 13, and to obtain a method
of convection that is reversible, a modified Euler method was employed. The

unsigned velocity induced on one vortex, m , by another, z, described by

(3.22) can be written, for unit strength f,,

(3.24)

If vortex m is in a cloud of N vortices, then the velocity induced by the cloud

on m is the sum of the velocity induced by the free vortices n .

uu,, =l l,u,,u
n=l
I+nt
(3.2s)
vu,, =L luv,,,
n=l
n+Dt

Using the forward differencing method, the vortex m would convect

according to:

Xrr.2=Xrr,liu^rdt\
(3.26)
!u,.2: !r,.t IVr.rdt)

Once all of the vortices have been convected, we can now recalculate the

convection step for the new position for each vortex. If, instead of moving
 52

the vortex to point

Q,,,s,!o,.t), tne vectors u,,,tdt and u,,,,rdtl and v,,,rdt and

v,,,tdt are averaged, a much better

sorution for the actuar motion of vortex

is obtained' This is the fundamental

element of the modified Euler method
and is physically described by Figure 14.

Figure 14) Modified Eurer estimate of vortex

motion
In this method, the effective displacement
is taken to be the average of the
displacements for the first and second steps.

+u'r.r)dtf
X nt+l = xu, *(ur,-,
2l (3.27)
! nt+t = !r, .þ^+ìol
 53

3,2,3,2 Implementation
Program VorticesO was written to explore voftex convection and examine

methods to increase the accuracy of the numerical simulation. The program

has the capacity to model many vortices and stores the vortex information,

x-location, y-location, x-velocity, y-velocity, vortex strength and a switch

variable that would enable the "turning on and off" of the voftex. This last

variable will become useful in the program written for the vortex cloud

model. The flow diagram for the program is shown in Figure 15 and 16, 15

being for the forward differencing method and 16 being for the central

differencing method.

Calculate the induced velocities

un and vn for each vortex

Calculate the displacement

according to the velocities un, yn

Figure 15) Flow chart for forward difference method

 54

Calculate the induced velocities

un and rn for each vortex

Calculate the temp vortex

locations according to the
velocities un, vn

Calculate the induced velocities

u'n and v'n for each vortex

Calculate the real vortex locations

according to the average
velocities Lln, ave, vn, ave

Figure 16) Flow chart modified Euler method

The program calculates the induced veloc¡ty on

each vortex by every other
vortex; it then stores this information in a temp array
and uses this data for
the intermediate time step. The program then calculates
the true velocity
and position of each vortex based on the average
of the first and
intermediate steps, i,e. voRrecEs lsl lul: (rEMp isl tul +vt) /2+IFSV for the
x-velocity' Although this calculation includes the addition
of the free stream
velocity term, its value was zero in the work presented
in this section. A
shortened section of the code is presented for N vortices:

for ( S:NI; S<:N_VORTECES; S++ ¡

{
f or ( I:NI; I<:N VORTECES,. I++ )
{-
if(r!:s)
{
 55

GAMA:VORTECES [ ] ] IGAMNAA] ;
W:_1*GAMA/ (2*pI*D);
UI+:sin (PHI ) *W,'
VI+:cos (PHI ) *Vrl;

TEMP IS] [U]:UI;

rEMPlSl Iv]:vr;

TEMP IS] [X] : VORTECES IS] tX] + (TEMP tS] iU] *DT]ME) ;
TEMP IS] tY] : VORTECES IS] IY]+ (TEMP tS] tV] *DTIME) ,.

f or (S:NI,'S<:N_VORTECES; S++ )
{
f or ( f :NI,' I<:N_VORTECES; I++ )
{
if (r !:s )

äo"o:uo*rEcËs trr tc^MlaAr ;

/ (2*pt*D) ;
hI=_1*GAljr\
UI+: ( y2-yl) *W;
VI+: (x2-x1 ) *W;
Ì
]

voRTECES tSl tUl: (TEMP tSl tUl +Ur) /2+UFSV;

VORTECBS tSl tvl : (TEMP tSl tVl +vr) /2+vFsv;
VORTECES tS] tX]:VORTECES IS] IX] + (VORTECES IS] tU] *DTTME) ;
VORTECES tSl Iyl =VORIECES tSl tyl + (VORTECES iSl tvl *DTTME) ;
)

3,2,3.7 Validation

Two vortices were chosen for the validation of the modified Euler method and

were arranged in a similar fashion to Figure 12 with vortex 1 located at (1,0)

and vortex 2 locateO at (-i,O). Each vortex has a vorticity of unit strength.

The path of the induced motion of each vortex for the forward difference

method is shown in Figure 17. The spiraling effect of the method is evident

as ¡s expected.
 56

----r- Vortex I
. -î:.'îî iiiil;ïì ;ï -
--.

,:.iiiîiJi'r.-r-r.-¡-'-;Tiï..1r*----r-vortex2
.¡-.r;\,
b.tL

a
a
a
a
a

t ,Í¿
'¡ j¡'
q'L -
-?
!i-.'i¡_ ì'l'
.lt
't.- .ir"
i'
--.t....*-. ,¡

.-.ï.1;:;:;:;:;ïï:.tîî-î.:t:':

Figure 17) Spiral effect of the forward difference method

The results for the central differencing method are presented in Figure 18.

Here, only one vortex is plotted but the quality of the path prediction is

clearly shown. The time step was equal for both test cases.
 57

tl -r_l -----r- Vonex '1

'r-r\r-r
a..-t-l'''t-t''

L
L
T
t¡
I
I
I
I
I

0.5 1i0
I
I
I
I
¡
I
I
T
i

t
ttl-l't-.-¡-¡-r

- a -a -a-a-a-t't't't't'

Figure 18) Results from the implementation of the modified Euler method

A logarithmic chart showing the error for each method is shown in Figure 19.

The error is defined as the difference between the final position of vortex 1

(initially at (t,o)) and the location (1,0) after the number of time steps

required to drift the path defined by 2n radians. The time step was
calculated based on the initial velocity of the vortex, known from the simple

calculation using (3.21) and was based on 100 steps to complete the path.
 5B

10 A

t_ ^.^,.
o
¡- \
¡-
LU
\o1
o\
,t-t-t'a.a \
'r tt'a,
o)
o tt+-
J
Euler Metl' od.a '.L
0.1 N/odifiod tr tler N/lnthnd '

10 100 1OO0

Log (Number of Time Steps)

Figure 19) Error associated with the forward and modified Euler methods

for a circular path and varying the time steps to travel 2n radians

The results shown in Figures 18 and 19 demonstrate that the vortex

convection model which employs the modified Euler method tracks very
well
and is representative of natural flows and will therefore be used in the
vortex
cloud model' Although both methods are of order Ar, the modified Euler

method is demonstrably more accurate, by an order of magnitude, than

the
accuracy of the forward difterencing scheme for the equivalent number
and
size of time step. This affords a couple of opportunities for implementation:
 59

1. The method can take advantage of the increased accuracy

2. The size of the time step may be increased to achieve a reasonable

level of accuracy while reducing the computational requirement

In the development of the Flow0 program, there was not investigation into

the second option.

3.2.4 Viscous Diffusion

The third process for vortex displacement described in 3.2.t is random

displacements of velocity. This is the implementation of viscous diffusion

which will be discussed here,

3.2,4,7 Theoretical Background

The "random walk" model, first proposed by Chorin (L973), has been used

with good success for the simulation of viscous diffusion in vortex cloud

models. The basic principle is to subject each element in the cloud to a

scaled perturbation of location or velocity that will model diffusion due to

viscosity. Such motion can be described by the two-dimensional Navier

Stokes equations.

(3.28)
#.fr.vþ=vY2a
where q is the velocity vector and V2 is the Laplacian operator. The second

term, (q Vþ, describes the convection orvorticity and the third term, vV2cù,

describes the diffusion of vorticity. writing (3.28) in an alternative

 60

dimensionless form, normalized by length and velocity scales, I and w*,

respectively

4*(a.v)a¡ = ' v2a=f-v', (3.2e)

ôt lW* Re

the influence of the Reynolds number is apparent. For high Reynolds number

flows, the diffusion term has little effect on the flow field. At low Reynolds

number flows, equation (3.28) reduces to:

ôa. vY'a (3.30)

=
-ôt
The solution of (3.30) for polar coordinates is given by Batchelor (1970),
/ )\
l-.-l
a:(r,t)= -l- ul+* ) (3.3 1)
4îtvt

Porthouse and Lewis (1981) have developed a random walk scheme based

on this solution. They argue that, if f is replaced by N discrete vortices of

strength f/ N initially located at the origin, there will be n number of

discrete vortices in the area rara.á after time r. The total amount of

vorticity, p, in this area can be expressed by:

n .I l-"1
lqr¡)
-"' /
P - --:
-N4nw e\ fA'A,g (3.32)

Equation (3.32) can also be interpreted to define the probability p that a

vortex will be in the area rLrLá at time r.

The symmetry of the radial system led to the conclusion that the probability

that a vortex will land within any arc Ad is equal and is independent of the

probability that the vortex will fall into the range between r and r+A,r. The
 61

angular portion of the random displacement is then assigned the simple

value of

0'=2nL (3.33)

where I is a random number between 0.0 and 1.0.

Porthouse and Lewis (1981) argue that the radial scattering of the vortices is

best found by first integrating (3.32) between 0 -0 and 2tt and then

integrating between r=0 and r:

-12
p(r):r- eã (3.34)

Equation (3.34) is a form of the standard normal curve from statistical

theory,

-'_2
1
p(x)ax=--e 2 dz
42n

and p(r) is equally likely to fall anywhere between 0.0 and 1.0. 4 is then

defined as a random number between 0.0 and 1.0 that will be used as a seed

to the radial displacement. If (3.34) is solved forthe 4th element it will

describe the radial portion of the random displacement

r, =,14v^¡tn(7Ð (3.3s)

The x and y components of displacement are then given by:

x' = r'cos dtJ

(3.36)
!' = r'sin9' )
The method proposed by Porthouse and Lewis (1981) is attractive because it

scales the random displacement in accordance to the system viscosity, v , as

 62

well as to the Reynolds number by way of the time step, ar and v , as will be
discussed,

Because the current work employs a modified Euler method to track

the
vortex motion, it must be determined at which step the random component

will be applied. It is considered to be overkill to apply this during the

intermediary time step as well as after the total velocity has been calculated

and it is therefore only applied after the central differencing step has been

completed' Besides computational simplicity, this also has the secondary

benefit of reducing the chances of randomly displacing a vortex into the body

surface during the central differencing step and thereby reducing

unnecessary occasions to treat this event. This will be further discussed in

section 3.2.7. Adding (3.36) to (3.27):

x nr+l =xu,*þ^+ìU.rl
(3.37)
! ¡,+l =!,,*ø#e.r]
3.2,4,2 Implementation
The VortexO program, discussed in 3.2.3.1 was also used to determine the

merit of the random walk component of vortex convection as a suitable

model for representing Brownian motion. The program was extended in

order to track N vortices through vortex convection and includes the random

walk according as seen in the program frow chart, Figure 20.

 63

Calculate the acutal vortex

velocities from the centreal
difference method core un, ave,
Vn, ave

Calculate the real vortex locations

according to the average
velocities un, ave, vn, ave

Calculate the random walk

components

Calculate the adjusted vortex

locations according random walk
components

Figure 2O) Flow chart for VorticesO with random walk

The implementat¡on of the random walk ¡s qu¡te simple. The following code

generates the random numbers and applies the positional shift to the

calculated x- and y-locations for each free vortex.

L:random( ) ;
K:random ( ) ;
L : 2*pDoc-)m_PI*L;
K : pow (4* (pDoc->m_viscosity/pDoc->m density) *
pDoc->DTIME*log (l/K) , .5) ;

pDoc->fl_vorteces tSl tD_Xl : pDoc->ff_vorteces IS] [D_X] +

(pDoc->fJ-_vorteces ISI tD_Ul *pDoc->DTIME)+ (K*cos (L) );
pDoc->fl_vorteces tSl tD_Yl : pDoc->fl-_vorteces IS] [D_Y] +
(pDoc->fJ-_vorteces tSl ID_Vl *pDoc-)DTIME)+ (K*sin (L) ) ;
 64

3,2,4.3 Validation

As the entire premise of the random walk method rests on the ability to

create quality random numbers, the first element to be validated was the

function random O . The following code was added to the VorticesO program
in order to determine whether the function would return a reasonable

distribution of random numbers:

int bin t10l:{0, 0, 0, 0, 0,0, 0, 0, 0, 0} ;
for (l:1,'I<:1000; I++)
{
K:random ( ) ;

bin[ ( (int) (K*10) ) ] ++;

]

The algorithm creates 1000 random numbers and sorts them into bins

according to their value. Table 4 shows that the output from this routine is

indeed reasonable as every bin has approximately the same number of

random numbers in it, as would be expected of a random distribution:

Number
Bin Ranqe Collected
1 0.0-0.1 111
2 0.1-0.2 97
3 0.2-0.3 86
4 0.3-0.4 89
5 0.4-0.5 98
6 0.5-0.6 115
7 0.6-0.7 93
I 0.7-0.8 93
I 0.8-0.9 105
10 0.9-1 .0 113
Table 5) Range of random numbers created by randomo
 65

The method employed to validate the motion is similar to that used to

determine the quality of the random numbers. A series of bins, shown in

Figure 21, was created and the program tracked the location of each vortex

at each time step.

Figure 21) Bins with initial random vortex positions

N vortices, of unit strength, were created and their initial location was

determined using the function randomO, i.e.their initial x- and y-locations

were random and between 0 and 1 in magnitude. The number of vortices

which fall into concentric annular "bir'ìs", where bin i will be the annular

region described between r¡ àîd r¡a1, is tracked and plotted in Figure 22 for

the 200th time step. One would expect that the random walk treatment

applied in the program Vortices0 would cause them naturally to drift apart.

If a reasonable representation of Brownian motion is achieved, the resultant

position of the voftices due to drift should follow a normal or Gaussian
 66

distribution. Figure 22 shows the results from the Binning test as well as a
trendline following a normal distribution and demonstrates that the results

appear to be Gaussian.

40
r Vortex Binning
Gaussian Curve Fit
35

30
U)
o
C) 25
o
2A
o
o 15
-o
E 10
z
5

0 I

40 60 80 100 124 140 160

Bin

Figure 22) Distribution of vortices after 2oo iterations due to drift

3.2.5 Panel-Vortex Interaction

As stated in 3.2,1, vortex convection, in a vortex cloud model, is dependant

on four factors. To this point, the first two have been discussed and the third
 67

source of induced velocity causing vortex convection, is the interaction

between the free vortices and the body surface. The voftex panel method

discretizes the body surface into a series of finite elements each with a

surface vorticity that is representative of the surface velocity. This surface

vorticity interacts with the free vortices in much the same way as was

presented in section 3.2.2. The relationship between the suface vorticity

/(Sr)^S of a panel, p, and the induced velocity of the free vortex is:

u.,.,=+ä/(s"bs,(;)
(3.38)
v,,,p=;äyß,bs,(;)

(*,,,,Y,,,).'o
l¡
,'i"
ii

r,r,p¡!

."' ii
:_6,bå," '"2 t'- i z(s^,, hs {þ,.r)^s

þo,ro)

Figure 23) Vortex interaction with the body surface

 68

The convective contribution from the body surface must be taking into

consideration during the central differencing method, Adding (3.26) to

(3.23) and (3.24) gives

u,, = u * .
+Ð,,
(s,
(tf). *Zr,E*)
)a,s,,

(3.3e)
v,, = v* .
+,2,
(^r, þ^s,
(;). *Ðr,(=)
and

u',,*t =, - .
*ä/(^r,, þs,, (*). +ä, (#) (3.40)
v,,,*, = n_ *
*ä, 6 .br .(+i!). +ä, (#)
Lewis (1991) presents the boundary integral equation for a body immersed in

a voftex cloud as:

r,)ri,)dr,,
- irí,). +f ¿(",,, , + w*ds +
*nn-, i)r, = 0 (3.41)

The first three terms are equivalent to the original system of equations

written for the voftex panel method. with the body immersed in cloud of

free voftices, there is significant velocity induced on the body surface by

these vortices. The fourth term in (3.a1) ensures that the Dirichlet boundary

condition is maintained and its numerical form is:

 69

M
3_f
I
n=l
K(",,, , t,)y$ ,) = -U * cos 8,,, -V*sin P,,, - L::cosþ,,,
j=t z,lLtil.
cos 8,,,

G.4Z)
sin8,,,
l"+sinþ,
where 4,u

3.2.6 Sub-Panels

One of the major concerns of vortex-surface interaction is the risk of a

disproportionate influence of a local pivotal point due to vortex proximity,

coarse discretization, or both. Figure 24 gives a good physical description of

the problem.

Figure 24) Influence of surface vorticity in close proximity

The issue is that the average surface vorticity for panel p+L, y1.¡o*,br*,, i,

applied at the control point, rather than spread over the panel surface. As

the free vortex interacts with this discrete point there is the unnatural effect

of dispropottionately large influence caused by the surface vorticity whereas

the panel vorticity should be averaged over the entire panel, causing an

even, distributed influence as the voftex passes near the surface. The

application of positive vorticity at the control point will cause vortex rn in

 70

Figure 24 to have a large velocity towards the body surface when it is near

but ahead of the pivotal point and away from the surface when it is near but

behind the pivotal point. This leads to the type of vortex motion around a

NACA OOLZ airfoil, seen in Figure 25.

0.08

o.07

0.06 . Pivotal Points

+ Free Vortex
0.0s

0.04

0.03

0.o2

0.01

0.00

Figure 25) Vortex path demonstrating unnatural pathline caused by surface

vorticity applied at the pivotal point.

Figure 25 is stretched in the vertical direction to give a better representation

of the vortex motion. It is also interesting to note that the irregular motion

is only present along the back side of the airfoil, starting just forward of the

quarter chord. This is due to cosine distribution of panel points, where the

panels are much coarser along the back side of the airfoil. Although the

leading edge panels appear larger in the figure, they are actually quite small
 77

and this reduces the near wall effect. The vortex shown in Figure 25 is

shown to oscillate as much as 8.3olo at the 50o/o chord position.

A common method to alleviate this problem is by using sub-panels, where

the panels that are less than a predetermined distance from a free vortex,

(Lewis (1991) suggests d <2.0As,,, as a reasonable distance), are divided into

a series of smaller panels each with a temporary control point where surface

vorticity is applied.

(r^,y,)r-^
f¡

Figure 26) The use of sub-elements reduces the near wall effects of coarse

discretization

The number of sub-elements will have a large impact on the computational

requirement and it is therefore prudent to make the number a binary-friendly

number (2, 4,8...) to reduce the computational impact. Lewis (1991)

suggests that a minimum of four sub-elements is required for general use.

The implementation of sub-panels requires that, when each distance, r,,,0, is

calculated, a test is performed to determine whether or not the local panel

qualifies for subdivision. If it does, then the panel in question must be

flagged for subdivision both in the convection process as well as in the

 72

calculation of the right hand sides. The panel right hand sides would then be

equal to the average of the right hand sides of the sub-panels.

Sub-paneling was considered in the development of the current work but the

complexities of tracking which panels need subdivision, additional

programming complexity and validation did not merit the incremental

decrease in computational resources, especially considering the speed of

modern computers. This element of program was explored and discarded in

favour of increasing the number of surface panels. The effect is essentially

the same and there is the double benefit of simplicity in programming and

standardization in the algorithms. Figure 27 depicts the effect of increasing

the number of surface panels from 74 to L74.

The path of the vortex over L74 panels is much more realistic than the path

of the vortex over 74 panels and the maximum oscillation has been reduced

to 0.160lo at37o/o chord, Even though a slight wavering in the path is

evident along the back side of the airfoil in the case of L74 panels, it is a

good representation of an actual streamline and is considered to be more

than sufficient for this algorithm. It is important to note that, as the user
may specify the number of surface panels, they would have to be aware of

this omission in order to specify a number of panels for a reasonable

accuracy. Expermimentation during the course of the developing the

algorithm has shown that 120 panels provide a good numerical compromise

between accuracy and speed, some variability in the path over the geometry

is acceptable as it only represents 1/q of the interactions causing motion and

the free voftices are being randomly displaced.

 73

74 Panels
+ 174 Panels

Figure 27) Effect of increasing the number of surface panels on the vortex

pathline

3.2.7 Absorption and Destruct¡on of Vorticity

There are two events that require that the free vortices be removed from the

simulation domain:

1. The vortex enters the body surface

2. The vortex moves so far from the body that it no longer has an

eftect on the analysis

In the first case, as the path of free vortices under convection routinely

comes near the body surface, it is possible for this vortex to pass through the

body surface due to a random movement or because of interaction with the

body itself. In this event, the vortex is no longer a viable contributor to the

flow field. There are several theories on how to deal with this:
 74

1. Porthouse (1983) recommends that these be retained for the

consequent convection iteration and then destroyed.

2. Chorin (1978) recommends that these voftices should be bounced

back into the flow.

3. Lewis (1991) argues that the vortex entering the body should be

absorbed into the surface vofticity, which can then be shed during

a future time step.

It is computationally exhaustive to continue tracking the free vortices as they

move far downstream from the body. These no longer contribute to the

domain under investigation and can be removed from the simulation.

However, an additional statement is required to ensure conservation of

vorticity and is expressed as:

Zrß,,)4",, *Ir, -Iì,i" =o (3.43)

Where I1,,. is the cumulative strength of all destroyed vorticity and is initially

0. Equation (3.44) must be added to the Martensen equations in order to

ensure that the correct amount of vorticity is shed at each time step:

zl
In=l
(r(r,,,, r,, ) * Â",, Þ(",, ) = -(I * cos 8,,, - V* sin 8,,, ->+cosþ,,u
L,/Ltili
j=t
cos 8,,,

G.44)

7'+ sinþ
""
sinP
"'
- f'' tl
"¡'"

3,2,7,7 Implementatíon
The current work employs Lewis' absorption scheme for vofticity entering the

body surface because of the natural manner in which this method handles
 75

vorticity, the numerical simplicity in adding vorticity to vorticity and for

computational convenience. Also there is little overhead required to

implement such a scheme. A sample of the code for this algorithm is given

for vortices that are above the line y=9.

if (pDoc->fl_vorteces tSl tD_Xl >pDoc-)m_XDATAII] &&
pDoc->fl_vorteces tSl tD_Xl ( pDoc->m_XDATA[]+11 )
{
1f (pDoc->f1_vorteces ISI tD_Yl >0 &&
pDoc->fl_vorteces tSl tD_Yl ( pDoc->m_YDATAIII )
{

pDoc->fl_vorteces IS ] [D_SVü]TCHI :0 ;
pDoc->f Ì_points t I I tD GAMMAI +:
pDoc->fJ-_vorteces I S ] [ D_GAMN{A] ;
pDoc->fJ-_vorteces tSl tD_Xl :0. 0 ;
pDoc->fl_vorteces ISI ID_Yl :0. 0 ;
pDoc->fJ-_vorteces tSl tD_Ul :0. 0;
pDoc->fl_vorteces tSl tD_Vl :0. 0;
pDoc->fl_vorteces ISI I D_GAMMA] :0 . 0 ;
pDoc- >VORT ï NFLOI¡I- -;

where pDoc->voRTrNFLovt is the number of vortices in the flow field at any

instant. The first statement is setting the switch for the deleted vortex
element to 0. This will cause the program to ignore this vortex in all

subsequent calculations until a housekeeping routine is called to allow these

dead vortices to be replaced by new ones.

Where the free vortices travel so far downstream as to be no longer

computationally significant, a similar cod¡ng is used except that the free

vorticity that is being destroyed is added to the accumulating circulation

variable pDoc->crRC. The variable oon is user defined and represents the

distance downstream after which the free vortices will be destroyed.

 76

if (pDoc->fI vortecestsl tD Xl>OOR)

{

pDoc->fl_vorteces IS ] [ D_SÍ,]ITCHI :0.

pDoc->CIRç 1: pDoc->fl__vorteces IS] [D GAMMA] ;
pDoc->f1_vorteces tSl tD_Xl :0. 0;
pDoc->f1_vorteces tSl iD_yl :0. 0;
pDoc->fJ-_vorteces tSl tD_Ul :0. 0 ;
pDoc->f1_vorteces tSl tD_Vl :0. 0;
pDoc->f1_vorteces tSl tD GAMMAI :0. 0;
pDoc->VORTINFLOW--;
return truei
]

The implementation of the coupling coefficients and the right hand side

additions according to (3.44) are simple and will therefore not be presented.

3.2.8 Vortex Shedding

3,2.8.7 Theoretical Background

The vortex cloud is formed by vorticity being shed from the body surface. In

a full vortex cloud model, the surface vorticity , yG,,)4",,, calculated in the

potential flow analysis is shed as a free vortex into the flow field. Lewis

presents two methods for delivering these free vortices into the flow field:

1. by random walk

2. by initial offset.

If the random walk method is used, Lewis recommends that the vortex
strength of the free vortices should be double the surface vorticity. This is

due to the statistical reality that, if random walks are applied to a free vortex

on the body surface, half will enter the surface and half will enter the flow

field. This doubling of the surface vorticity is an attempt to conserve the

vorticity at each time step.
 77

The initial offset method throws the free vortex a certain distance from the

panel surface, dramatically increasing the chance that the vortex will enter

the flow field. This is shown in Figure 28.

("o.þ,.'
( Ç6,,y,)
if't''¿
r(',)l"',
6,,r,)

Figure 28) Vortex shedding by offset method

Porthouse (1983) recommends that this offset distance is defined by:

/,h"d : tr-- (3.4s)

llvtt
There is some difficulty with using this value at high Reynolds number flows

as the value of rn,uo becomes infinitely small due to the decreasing time step,

This will be further discussed in the following sections.

3,2,8,2 Implementation
Vortex shedding was implemented using (3.45), ensuring that the distance,

r,r"o, is normal to and centered on the local panel, The number of vortices

shed in any given time step is equal to the difference between the maximum

number of free vortices in the flow, pooc->i_nvorteces, and the total

number of vortices in the flow at the instant that vorticity is being shed,

pDoc->VORTINFLOW, up to a maximum number, VORÄDD. Both pDoc-

 78

)i_nvorteces êt'Ìd voRADD are user specified. The panels to shed vorticity
are randomly selected using the randomO function. There is also a tuning

variable, OFFSETFACToR, that allows the offset distance to be adjusted by the

user if required.

The variable PERC was added to the algorithm as a way to adjust the

percentage of vorticity shed from a given surface. The addition of this

variable brings some coding difficulties due to the requirement of

conservation of vorticity. The reason for the variable is to promote program

stability and to reduce the surface vorticity fluctuations that are present if

the entire Martensen vorticity is shed into the flow field. As demonstrated in

the next section, a further advantage to shedding only a portion of the

surface vorticity is that the influence of the surface on the free stream

vortices is maintained. The method of ensuring that the total simulation

vorticity is conserved is to sum the shed vorticity, vonror, and compare this

with the circulation variable, pDoc->crRC. If pDoc->crRc ¡s not reduced to

an acceptable vorticity, vonr,lu, that will be carried forward to the next time

step, the program continues to shed vofticity until this condition is met. This

condition overrides the users limit on vorticity shed in a given time step. The

code implementation of this routine is:

pDoc->VORTDI FF : pDoc- >i_nvorte ces -pDoc->VORT INFLOW ;

if (pDoc->VORTDIFF > VORA,DD ) pDoc->VORTDIFF : VORADD

for ( I:1,' I(:pDoc->VORTDI FF; I++ )

RNDNUM : randomO;
RNDNUM : RNDNUM*pDoc->i_npaneÌs;
PANEL : (int)RNDNUM;
 79

int OFFSETFACTOR : 10;

GMA : pDoc->fl points IPANEL] [D GAM]aAl;

Rshed : OFFSETFACTOR*sqrt ( 4 * (pDoc->m_viscosity /

pDoc->m_dens ity ) *pDoc->DT tME / 3 ) ;

X1:pDoc->m_XDATA I PANEL ];
Y1:pDoc->m_YDATA I PANEL ];
X2:pDoc->m_XDATA I pANEL+
1] ;
Y2:pDoc->m yDATAIpANEL+11 ;

XI:X2-XI;
YL:Y2-Y7;

PANELLENGTH: sqrt (pow (X1, 2 ) +pow (y7,2) );

x2:x2- (x7/2); / / ttna center of panel

Y2:Y2- (YL/2);

pDoc-> f J__vorteces t VN I t D _y I :y 2- fabs (X*X1 / pANELLENGTH ) ;

pDoc->fl_vorteces tVNl tD Xl :¡21¡abs (X*Y1lPANELLENGTH) ;

pDoc->VORTINFLOW++;
pDoc->f l_vorteces tVNl I D_SWITCH I :1 ;
pDoc->fl_vorteces [VN] [D_GAMMÄ] :PERC * GMA;
VORTOT 1: pDoc->fl-_vorteces IVN] [D_GAMMA]
pDoc->fl_vorteces tVNl t D_Ul :0 ;
pDoc->fI_vorteces tVNl t D_Vl :0 ;
pDoc->f1_points IPANEL] [D_GAMMA]: ( 1-PERC) * GMA;
if fabs((VORTOT - pDoc->CIRC) > VORLTM) t--;
)

pDoc->CIRC -: VORTOT;

Figure 29 shows 30 vortices shed from a NACA 0012 airfoil as calculated by

the function addvortO:

 80

Figure 29) Free vortices shed from the surface of a NACA oo12 airfoil

3.2.9 Full Vortex Cloud Simulation

The algorithms presented in the last three sections are the final steps

required to implement a full vortex cloud model. The program Flowao was

written to encompass all of these algorithms, The program flow chart

indicating the adopted sequence for solution of the system of equations and

vortex tracking is shown in Figure 30.

The time interval, ar, is scaled to the Reynolds number by way of the free

stream velocity. For the FlowaO program, the time step was determined

using

Lt-- 2l (3.46)
4W*

where a is a user-defined variable. In the current case, a value of ø:20

was used. As w* increases, a¡ decreases, and therefore the vorticity

shedding distance, r,h"d, also decreases. The variable a may be used to

increase this distance at high Reynolds number flow, so long as the

numerical scheme retains resolution and validity.

 81

Another addition to the program was a small calculation that helped to

reduce the pure numerical dependence for the determination of the velocity

of the free vortices. As a vortex comes very near another, (3.3) suggests

that the induced velocity on each vortex approaches infinity. This is

discordant with natural vortex motion and therefore

l¡."or" =**t, (3.47)

2rr
was applied for vortices within As,nin / 2n of each other, The following code

was added to the vortex convection routines and the calculation of the right

hand sides:

if (RmjSQ < MGCheck)

{

GMA : GMA* (RmjSQ/MGCheck) ,'

where p.mjse is the straight line distance between vortices.

 82

Geometry
Data Output

_v
It"l aacto¡aoonalcorrect¡on l
Ì

I
v
I Enfor"" Kutta Condition tor I
It'-l Couolino Coefficients I

lnvert Coupling Matrix

Calculate the Right Hand Sides

Enforce Kutta Condition for Right

Hand Sides

Shed Surface Vorticity

Absorb Vorticity Entering Body

Martensen Potential Flow

Output

Figure 3O) Flow chart for program Flowao

 83

3.3 Visual Vortex Cloud Program

The FlowaO program, although a full vortex cloud simulation program, was

unable to show graphically the developing flow field without a large amount

of post-processing by the user. It was for this reason that the VisiflowO

program was written. This windows based program accepts user input and

shows the developing flow field on the screen.

BÞNffi F;i-'
ad . iiiËai¿àm*: fs¡
tuLd
l¡É------__l-
l&cbl I
lruKY t._,..__...
ud*nü
- lezlt I -l
lFr63-ll7 L-j:
!ñdP*4m*;í .lt¡=J
lsl?)1
tuddM'--_----T#
F@ stù%Hi F--
,,rq*1u n* úqo* [ãóõ--*
.

¡'Évaræt¡tu l0
r* *¡;;Vo.;,1-**-- T:
ouøru lrmo--- F:--l
lo¡Srø,.. :l:-
rddThsb; lm*.._-
tõl--
'le

PF-

Figure 31) Screen shot of program Visiflowo

Run information is input by the user into the pane on the left hand side. The

user may select the airfoil as well as the flow parameters such as the chord

length and the number of panels that will describe the body surface. The
 84

fluid used for the simulation as well as the number of free vortices and time

steps is also selectable here.

Fìun Name ,
NFunl
.

Airfoil, bffiäf*î Gol

:t a

Ehotd Leng!h; lT

Workinq Fluid IAIB ê STP r.l

Numberor paneb Jîäo'* [ó,úÈ'r,,+;l
AngleolAtt¡ck,. .,

Free;5leamVelociþ,,

Number of Free Vorleces

Free Vcrteces ln Flow,

llumber of Free Vorteces

output Fle iNãìi"'t tr]

Tim.q Step -
Total Time 5táps

dtime, ,

Show Messagesl

l. lnq*-l l. in"u+.l
Figure 32) Input pane to set up flow simulation

The VisiflowO program was built on the main code from the Flowa0 program

with some modifications that were requ¡red in order to interface with the

user, to accept input from the screen as well as graphically to show the flow

field, as seen in Figure 32. The free voftices are represented by circles and
 85

triangles, with the circles denoting a positive rotation, as seen in Figures 33

and 34.

Figure 33) Vorticity shed from body surface, VisiflowO at Ooangle of attack

Figure 34) Vorticity shed from body surface, VisiflowO at lS0oangle of

attack

In the case of zero angle of attack, the vorticity shed from the underside of

the airfoil is negative, as expected. When the flow field is reversed, and the

flow comes from the left over the airfoil, the vorticity shed is opposite in sign.

An updated flow chart for VisiflowO is shown in Figure 35.

 86

User Selects Run Parameters

Geometry
Data Output

Draw Aerofoil Simulation Profile

Calculate the Coupling

Coefficients

Backdiagonal Correction

l
v

I entor"" Kutta Condition for i

I Coupling Coefficients I

lnvert Coupling Matrix

I M".t"nr"n Potential Flow, W,¡

I ontv I

Shed Surface Vorticity

Convect Free Vortices, Step 1 Absorb Vorticity Entering Body

Convect Free Vortices, Step 2 Absorb Vorticity Entering Body

Destroy Downstream Vorticity

Martensen Potential Flow

Dedraw Flow Field

Figure 35) Flow chart for program Visiflowo

 87

3 .4.L Test Cases

The test cases chosen were based on the availability of high quality

experimental data from Selig et al. (1996). Selig's work details the results of

a carefully constructed experimental program to produce a database of

modern, low-Reynolds number airfoils. The measurements were

exhaustively examined for their accuracy; for example drag was measured

by the more accurate wake traverse method. This work filled a gap in

previously existing databases by focusing in on low-Reynolds number and is

of great use in the design of remote controlled aircraft and small unmanned

aerial vehicles.

The current work focused on a series of airfoils, the NACA 2414, the FX63-

137 and the CI-ARK-Y airfoils shown in Figure 36. These airfoils were
selected for this study as a complete data set could be found in Selig's work.

The data specified the ideal profile shape and the deviation of the

experimental model from the ideal.

 B8

a) NACA2414

b) FX63-137

__i--___

c) CLARK-Y

Figure 36) Airfoil profiles selected for this study, airfoil data from Setig et al.

(1ee6)

3.4.2 Results

A variety of test simulations were run using 120 panels to model each of the

selected airfoils. Example flow patterns are shown for the NACA2414 airfoil
 89

with 250 free vortices in Figure3T and for with 500 free vortices in Figure

38. The simulations were done for the airfoil at zero angle of attack and a
Reynolds number of approximately 150,000. The figures show similar

behaviour and, as expected, the free vorticity gathers into clusters in the

wake of the airfoil. The results show that vortices shed from the lower

surface are grouped to the lower half of the wake near the airfoil and, as we

move downstream, diffuse across the wake due to interaction with other

vortices and the application of random walk.

Figure 37) Vortex Cloud Model simulation of flow over a NACA 2414 airfoil

at Re = 149r5O3, with 25O free vortices shed into the flow field.

Of concern is the number of vortices that need to be shed in order to obtain

an accurate representation of the flow field.

 90

Figure 38) Vortex Cloud Model simulation of flow over a NACA 24L4 airfoil

at Re = 149¡5O3, with 5OO free vortices shed into the flow field.

As is evident in Figures 37 and 38, there is good similarity of the results for

both 250 and 500 shed vortices demonstrating that a sufficient number of

vortices were being shed to obtain an accurate representation and a

reasonable resolution of the bulk flow.

Figures 39 to 41 show the developed flow field for each of the airfoils with a

Reynolds number of 200,000. In each case the airfoils are set to have zero

angle of attack and 500 free vortices are active. The flow fields demonstrate

several properties that indicate that the results are of good quality.

1. There are no free vortices in the body; the elimination routine is

functioning as desired. The inclusion of these voftices can lead to

erroneous right hand side calculations and bizarre vortex dynamics.

2. The downstream wake is directed slightly downwards which is

expected for each of these non-symmetric airfoils. Also, for the FX63-

L37 airfoil, this behaviour is increased. As the zero lift angle for this
 91

airfoil is greater than the other two, the results are representative of

the real case.

Figure 39) Vortex Cloud Model simulation of flow over a NACA2414 airfoil at

Re = 1991337, with 5OO free vortices shed into the flow field

Figure 40) Vortex Cloud Model simulation of flow over a CLARK-Y airfoil at

Re = 1991337, with 5OO free vortices shed into the flow field
 92

Figure 41) Vortex Cloud Model simulation of flow over a FX63-t37 airfoil at

Re = 1991337, with 5OO free vortices shed into the flow field

In each of the above figures, there appears to be no separation of the flow

from the body surface and a reasonable picture of the flow field is presented.

Simulations of flow over the NACA 24L4 airfoil were made for a range of

angles of attack, a , and are shown in Figure 42. The VisiflowO program

displays the airfoil in its horizontal orientation and the free stream velocity is

inclined upward as ø is increased. There is no indication of flow separation

over the airfoil at d,:0' angle of attack. When a =5" the wake deflected

slightly upward in the direction of the free stream velocity and, as with the

case of a:0' / no separation is evident, These angles of attack are within

the linear portion of the lift curve and the flow would not be expected to have

significant separation regions. When the angle of attack is increased into the

nonlinear portion of the lift curve, near the maximum lift region of the lift

curve, say d =10', we would expectto see the presence of a significant

portion of separated flow on the suction surface near the trailing edge, The
 93

simulation represents this situation reasonably well with a separation zone

that is approximately 40o/o of the chord length. Finally, the flow over the

airfoil in the deep stall range, represented by a =15", clearly shows a

separated zone that extends over nearly the entire suction surface,

approximately 80 - 9O o/o of the chord length. In general, the Visiflowo

output indicates that the predicted flow appears to give a good qualitative

representation of the actual flow.

 94

d =I5"

a:I0'

a =5"

a=0'

Figure 42) Vortex Cloud Model simulation of flow over a NACA 2414 airfoil

at Re = t49t503, 5OO vortices shed, lor a = Oo, 50, l0o and 15o.
 95

The results in Figure 42 represent typical results from the VisiflowO program

and it was deemed superfluous to include the same results for each test

case. Use of the program can easily recreate the results. To demonstrate
this, Figure 43 shows the CLARK-Y airfoil at 150 angle of attack. There is

some separation noticed on the suction side of the airfoil as the vorticity is

spread from the typically narrow wake. There is a noticeable streamline

coming around the pressure side of the airfoil and resisting the free stream

velocity for nearly half a chord length downstream before angling upwards.

Figure 43) Vortex Cloud Model simulation of flow over a CLARK-Y airfoil at

Re = 1991337t 5OO vortices shed lor a = 15o.

 96

Chapter 4

Discussion and Conclusions

In this work, a vortex cloud aerodynamic flow simulation software package

has been developed which graphically demonstrates the developing flow field

over a body. The current work has seen the development of a vortex panel

method extended to include a vottex cloud model as well as the vortex

creation, interaction and destruction algorithms that are required. The

algorithms developed here provide a mechanism for the observation of fluid

dynamics over a body profile for low Reynolds number flows, The suitability

of the following algorithms will be discussed: the calculation of aerodynamic

forces, the method of vortex shedding and the visual representation of the

flow.

The vortex convection algorithm, with the implementation of the modified

Euler method, was shown to provide excellent prediction of vortex dynamics.

Further exploration of this algorithm demonstrated the modified Euler

method affords a 650/o reduction in computational requirement for the

equivalent accuracy when compared to the forward difference method.

 97

The random walk algorithm and the program VorticesO showed the validity

of simulated viscosity being able to reproduce vortex movement that is

similar to Brownian motion. The omission of the sub-paneling routine from

the current program is seen as an advantage. As previously discussed, the

programming simplicity and continuity in the solver algorithm maintained by

not implementing sub-paneling are two of the features that make the vortex

cloud model attractive. In fact, with the speed of modern computers, sub-

paneling is becoming more and more a technology of the past.

Our assessment is that the vortex cloud model is capable of simulating the

flow over an airfoil at low-Reynolds numbers. However, there are several

areas that need to be addressed before the method is suitable for use in an

optimization routine. There is some concern about the calculation of

aerodynamic forces. In the current work, the attempts made in estimating

the lift and drag forces were thwarted by numerical instabilities. There are

several potential causes for this difficulty:

1. The vortex cloud method is, by definition, a random process that

requires averaging. It may be that a large number of time steps are
required before the averages reach their asymptotic values.

2. The integration method was based on a simple trapezoid rule where a

more accurate method may be required. This method did, however,

generate reasonable results for a strictly inviscid estimate.

3. The act of shedding vorticity, particularly if the entire surface vorticity

is shed, will cause discontinuity in the predicted, instantaneous,

 98

surface pressure distribution. As a result, it may not be sufficiently

accurate to allow for the direct computation of these quantities

To help in the calculation of these forces, a control volume approach, where

a momentum analysis is performed fore and aft of the body profile, may be

more suitable for the estimation of these forces.

Furthermore, there are several issues with the vortex cloud model which

need to be addressed. The vortex shedding method that was recommended

by Lewis did not work as described. In that form the vorticity was wildly

flung around the flow field. This necessitated a 75o/o reduction of shed voftex

strength and the implementation of a routine that limits the induced velocity

for a given vortex in order for the free vorticity to evolve through the domain

in a reasonable manner. This indicates that a more formal, scientifically

based method for the determination of the shed vorticity needs to be

established.

Due to the difficulty realized in the calculation of the aerodynamic forces, lift

and drag measurements are not available. Thus, the validity of force

estimation from the random vortex displacement model could not be

assessed.

The graphical user interface is an invaluable tool in the analysis of fluid flow.

Real-time graphics are available to the user at any interval of the simulation

whereas with traditional CFD programs post-processing is required after the

entire flow field has converged. The transient nature of the algorithm is

another advantage of traditional CFD codes. Eulerian flow solvers require

that the flow field is fully meshed and it is typical for several iterations to be
 99

completed in order to converge the state at each time step for a transient

solution. The visual interface employed in VisiflowO allows the user to

observe the more minute detail in the transient flow model as well as the

more holistic characteristics such as flow separation.

The general flow characteristics predicted by the vortex cloud model were in

general agreement with natural flows, the observation and prediction of flow

separation being one of the key results from this work. The results shown

here are also generally consistent with Lewis's (1991), there is good

prediction of vortex dynamics and the wake development is graphically well

represented.

The vortex cloud msdel does have some significant advantages as a flow

simulation tool. The grid free nature of the Lagrangian voftex tracking

method allows a grid free approach. A significant amount of mesh

generation tool development ís eliminated. The lack of computational mesh

also reduces the discretization procedure from full mesh generation to simple

surface panel generation. This has real advantages when used in an

optimization scheme. It allows for quick geometrical changes with no change

to the fluid domain. As the optimization process typically requires the

simulation of a large number of candidate airfoils, it will benefit greatly from

the reduction of simulation effort.

The application of the method, aerodynamic forces not withstanding, was

simple to implement into a vorLex panel method. The model is also relatively

fast, making it quite competitive with other methods. The vortex cloud

model has merit in that it is a quick tool that gives, at least visually, a good
 100

representation of the flow field for low Reynolds number flows and is

suitable, if not ideal, for use in an aerodynamic optimization scheme.

 101

Chapter 5

Recommendations
Future development is required in several areas before this algorithm can be

implemented into a full scale optimization program, The issue most critical

to the advancement of this routine is the calculation of aerodynamic forces.

Without a validated method to predict the performance of the body under

investigation, there are few applications where this algorithm will find use.

The program Visiflow0 does require some extension before it can be used as

a stand-alone, autonomous program. First, there is currently no check for

aerodynamic convergence. Although watching the flow field develop is

interesting, the aerodynamic forces calculated while the flow is developing

will be unstable and unusable for averaging. It is necessary to start the

averaging of forces once the code determines that the domain is sufficiently

stable. Convergence for a transient solver is difficult to determine and

therefore the author recommends the use of the Strouhal number in order to

determine when the simulation has become sufficiently stable to commence

the performance analysis. The Strouhal number can identify the flow
 LO2

periodicity and provides a good time scale for averaging. The width of this

time scale can also be used to generate output graphics automatically at

predetermined, user-defined intervals. For example, if the user would like

twenty pictures generated for a simulation, the time scale divided by this

number will tell the program at which time steps it must generate this

output.

There is also some real opportunity for improvement in the algorithm that

sheds vorticity. In the current state, this algorithm has a methodology of

questionable stability. If shed vorticity is not opposite in sign to the

accumulated circulation, it is possible for this routine to cause the program to

stall in the function as it tries to satisfy an inappropriate condition. Because

the randomO function used to select the panel to shed vorticity has an

average output of 0.50 there is a 50/50 chance that the shed vorticity will be

opposite in sign from the accumulated circulation. Because of this, the

accumulated vorticity should be approximately equal to the average free

vorticity, or 0.0. The current implementation sheds only a portion of the

total surface vorticity and therefore the risk of stall is small but present. A

more suitable method of retaining a portion of surface vorticity while

ensuring zero net circulation is a scaled approach. It is possible to

implement a scaled vortex shedding process where the computer can target

the net circulation for shed voriicity and allow for more than the preset

amount of surface vorticity to be shed at any instant. This will compromise

the user-defined limit on shed vorticity but will ensure that the maximum

number of free vortices shed during the time step is enforced. Alternatively,
 103

the order of solution might be adjusted so that, if all of the surface vorticity

is shed and the noticed instabilities can be resolved, the surface vorticity is

regenerated before the convection processes, i,e. the calculation of the right

hand sides is completed before the convection routine, This would effectively

eliminate the discontinuity in surface velocity that arises from shedding the

entire surface vorticity.

The process of matrix inversion is a known source of numerical instability and

some exploration should be made into alternative solution methods.

Although this would mean that the matrix would have to be solved during

each time step as opposed to the current work where the matrix is inverted

only once, the speed of modern computers make this type of code

optimization less of a consideration than in the past.

The graphical user interface could be extended to provide additional

information to the user, such as plotting the pressure distribution at each

logical interval or providing additional controls for the simulation.

There is also some opportunity for optimization of the current program as

visiflowO is the amalgamation of four other programs. Algorithms may be

able to be simplified and the memory allocation for the program should be

examined. In the current implementation there was no investigation into

leveraging the increased accuracy of the modified Euler method for a longer

time step. This might reduce computational time as fewer calculations would

be required for the equivalent movement. The time step used in the analysis

also is somewhat arbitrary and does not consider all of the implications of its

size' Effort should be made to have each section of the code consider the
 LO4

code consider the effect of specifying a global factor. An example of this

would be the timescale calculation. If it is scaled to the velocity, the

program should also consider limits in order to ensure a user-specified

accuracy or resolution.
 105

Bibliography
ABBOTT, L and VON DOENHOFF, A. E., Theory of Wing Sections. Dover Publications, New
York, NY, 1959.

ABERNATHY, F. H,and KRONAUER, R. E., The Formation of Vortex Streets. J. Fluid Mech,
Vol. 13, 1962, pp. L-20.

ANDERSON, J.D., Fundamentals of Aerodynamics. McGraw-Hl/, New York, NY, 2001.

BARBER, R. W. and FONTY, A., Numerical Solution of Low Reynolds Number Flows
Using a Lagrangian Discrete Vortex Method.

http : //www.cse.cl rc.ac. uk/ceq/m isc/vortex/vortex.shtml, 2005.

BARBA, L. 4., LEONARD, A. and ALLEN, C. 8., Vortex Method with Meshless Spatial
Adaption for Accurate Simulation of Viscous, Unsteady Vortical Flows. Int. J. Numer.
Meth. Fluids Vol. 47, 2005, pp. 841-848.

BASUKI, L and GRAHAM, l. M. R., Discrete Vortex Computation of Separated AiÉoil

Flow. AIAA Journal,Vol.2S 7987 pp. 1409-1410.

 106

BATCHELOR, G. K., An Introduct¡on to Fluid Dynamics. Cambridge University Press, L97O.

BEALE, J. T. and MAJDA, 4., Rates of Convergence for Viscous Splitting of the Navier-
Stokes Equations. Mathematics of Computation, Vol. 37 , No. 156, 1981, pp. 243-259.

BORTHWICK, A. G. L. and BARBER, R. W., Numerical Simulation of Jet-Forced Flow in a

Circular Reservoir Using Discrete and Random Vortex Methods. International lournal
for Numerical Methods in Fluids Vol. 14, No. 12, 7992, pp. 7453-L472.

CARR, L. W., McALISTER, K. W. and McCROSBAY, W. J., Analysis of the Development of

Dynamic Stall Based on Oscillating Airfoil Experiments. /VÁSÁ TN D-8382, I977.

CHAN, C. K., LAU, K. S. and ZHANG, B. L., Simulation of a Premixed Turbulent Flame
with the D¡screte Vortex Method. International Journal for Numerical Methods in

Engineering, Vol. 48, 2000 pp. 6L3-627.

CHEN, 8., GUO, L. J. and YANG, X.G., Numerical Simulation of Flow Around Circular
Cylinder by Discrete Vortex Method. Progress in Natural Science,Vol. 12,2002, pp.964-
969.

CHORIN, A. J. and BERNARD, P. S., Discretization of a Vortex Sheet, with an Example of

Rolf-Up. J. Computational Physics Vol. 13, No. 3, 1973.

CHORIN, A. J., Numerical Study of Slightly Viscous Flows. J. Fluid Mech. Vol. 57, 1973,
pp. 485.

CHORIN, A. J., Vortex Sheet Approximation of Boundary Layers. J. Computational

Physics. Vol. 27, No. 3, 1978.
 L07

CLEMENTS, R. R., An Inviscid Model of Two-Dimensional Vortex Shedding. l. Fluid

Mech. Vol. 57, 1973, pp. 321-336.

CORTELEZZI, L., CHEN, Y. C. and CHANG, H. 1., Nonlinear Feedback Control of the Wake
Past a Plate: From a Low-Order Model to a High-Order Model. Phys. Fluids, Vol. 9,

L997, pp. 2009-2021.

COTTET, G. H. and KOUMOUTSAKOS, P. D., Vortex Methods. Theory and Practice.

Cambridge University Press, New York, N.Y., 2000.

COTTET, G. H., OULD SALIHI, M. L, and EL HAMRAOUI M., Multi-purpose Regridding in

Vortex Methods. ESAIM Proceedings, Giovannini A et al. (eds), Vol. 7. Societe de
Mathematiques Appliques et Industrielles, 1999, pp. 94-103.

DERKSEN, R. W. and RIMMER, J., Aerodynamic Flow Simulation, WIT Transactions on

Engineering Sciences, Vol. 52, 2006.

DIDDEN, N., On the Formation of Vortex Rings: Rolling-up and Production of

Circulation. Z. Angew. Math. Phys.30, 101 1979.

ERICKSON, L.L., Panel Methods - An Introduction. NASA TP-2995, 1990.

FROHLICH, J. and RODI, W., Introduction to Large Eddy Simulation of Turbulent Flows.
Closure Strategies forTurbulent and Transitional Flows, B. Launder and N. Sandham, eds.,

Cambridge University Press, Cambridge, 2002, pp. 267 -298.

GERRARD, J. H,, Numerical Computation of the Magnitude and Frequency of the Lift on
a Circular Cylinder. Phil Trans. Roy. Soc. A., Yol. 267, 1967, pp. L37-I62.
 108

HESS, J. L., Panel Methods in Computational Fluid Dynamics. Annual Review of Fluid

Mechanics, Vol.22, 1990, pp. 255-274.

HESS, J.L., Linear Potential Schemes. Applied Computational Aerodynamics, P.A. Henne,

ed., AIAA, Washington, 1990, pp. 21-36.

IACOB, K. and RIEGELS, F. W., The Calculation of the Pressure Distributions over
Aerofoil Sections of Finite Thickness with and without Flaps and Slats. Z. Flugwiss,
Vol. 11, No. 9, 1963, pp. 357-367.

JONES, M.4., The Separated Flow of an Inviscid Fluid Around a Moving Flat Plate. J.

Fluid Mech., Vol. 496, 2003, pp. 405-447.

KAO, H,C., A Note on Trapping Moving Vortices. Glenn Research Center, Cleveland, Ohio,

July 2000.

KATZ, J., A Discrete Vortex Method for the Non-Steady Separated Flow Over an
Airfoil. J. Fluid Mech. Vol. !O2, I98L, pp. 315-328.

KATZ, J., and PLOTKIN, 4., Low-Speed Aerodynamics From Wing Theory to Panel
Methods. McGraw-Hill,.Inc., New York, NY, 1991.

KELLOG, O., Foundations of Potential Theory. Frederick Ungar Publishing Corp., New York,
NY,1929.

KOUMOUTSAKOS, P., Simulation of Unsteady Separated Flows using Vortex Methods.

Ph.D. Thesis, California Institute of Technology, L993.
 109

KRASNY, R., Vortex sheet computations: roll-up, wakes, separation, Lectures in Applied
Mathematics. Am. Math. Soc. Vol. 28, 199I, pp. 385-401.

KUWAHARA, K., Numerical Study of Flow Past an Inclined Flat Plate by an Inviscid
Model. J. Phys. Soc. Japan Vol. 35, 1973, pp. 1545.

LIU L., JI, F., FAN J., and CEN K., Recent Development of Vortex Method in
Incompressible Viscous Bluff Body Flows. Journal of Zhejiang University SCIENCE Vol. 6A

No. 4, 2005, pp. 283-288.

LEWIS, R.I., Vortex Element Methods for Fluid Dynamic Analysis of Engineering
Systems. Cambridge University Press, New York, N.Y. 1991.

MARTENSEN, E., Berechnung der Druckverteilung an Gitterprofilen in ebener

Potentialstromung mit einer Fredholmschen Integralgleichung. Arch. Rat. Mech., Anal.

3, L959, pp.235-270.

MILEY, S. 1., A Catalogue of Low Reynolds Number Airfoil Data for Wind Turbine
Applications. Department of Aerospace Engineering, Texas A&M University, College Station,
Texas, RFP-3387, UC-60, 1982.

MOORE, D. W., The Discrete Vortex Approximation of a Finite Vortex Sheet. California
Institute of Technology Report AFOSR-1804-69. L977.

NAKASHiMA, M. and ONO, K., Numerical Study of the Thrust, Energy Consumption, and
Propulsive Efficiency of a Three Joint Bending Propulsion Mechanism. lournal of Fluids
Engineering Vol. 722,2000, pp. 614-618.
 110

NITSCHE, M. and KRASNY, R., A numerical study of vortex ring formation at the edge of
a circular tube. J. Fluid Mech. Vol. 97, t994, pp. 239-255.

NYIRI, A. and BARANYI, L., Numerical Method for Calculating the Flow around a

Cascade of Aerofoils. Proc. Of VII International J.S.M.E. Symposium of Fluid Machinery and

Fluidics, Tokyo, Vol. 2, 1983.

PEREIRA, L. A. 4., HIRATA, M. H. and SILVEIRA NETO, 4., Vortex Method with Turbulence
Sub-Grid Scale Modelling. J. Braz. Soc. Mech. Scl. & Eng,Vol.25, No. 2,2OO3, pp.74O-746.

PFEIFFER, N. l. and ZICKUHR, T. D., Validation of Computational Aerodynamics Applied

to General Aviation Configurations. Proc. of 7th Applied Aerodynamics Conference, AIAA-

1989-2169,1989.

PORTHOUSE, D. T. C. and LEWIS, R. I., Simulation of Viscous Diffusion for Extension of

the Surface Vorticity Method to Boundary and Separated Flows. J. Mech. Eng. Science,
L Mech. E. Vol. 23, No. 3, 1981, pp.I57-I67.

PORTHOUSE, D. T. C., Numerical Simulation of Aerofoil and Bluff Body Flows by Vortex
Dynamics. Ph.D. Thesis, Univ. of Newcastle up Tyne, 1983.

PLOUMHANS, P. and WiNCKELMANS G. S., Vortex Methods for High-Resolution

Simulations of Viscous Flow Past Bluff Bodies of General Geometry. Journal of
Computational Physics, No. 165, 2000, pp. 354-406.

ROSENHEAD, 1., The Formation of Vortices from a Surface of Discontinuity. Proc. Roy.

Soc. A., Vol, 134, 1931, pp. t7O-192.

 111

SARAPKAYA, T., An Inviscid Model of Two-Dimensional Vortex Shedding for Transient

and Asymptotically Steady Separated Flow over an Inclined Plate. l. Fluid Mech.Yol.

68, L975, pp. 109-128.

SELIG, M. S., LYON, C. A., GIGUERE, P., NINHAM, C. P. and GUGLIELMO, J. J., Summary of
Low Speeds Airfoil Data. Soarfech Publications, Virginia Beach, U.S.A., Vol. 2, 1996.

SPALART, P. R., LEONARD, A. and BAGANOFF, D., Numerical Simulation of Separated

Flows. NASA Technical Memorandum 84328, 1983.

TAKAMI, H., Numerical Experiment with Discrete Vortex Approximation, with

Reference to the Rolling Up of a Vortex Sheet. Dept. of Aero. And Astro., Stanford

University, Report SUDAER 202, 7964.

VAMOS, C. and SUCIU, N., Global Random Walk Simulations of Diffusion. Scientific
Computing, Validated Numerics, Interval Methods, Plenum Publishing, New York, NY, 2001,

pp, 343-354.

VAMOS, C., SUCIU, N. and VEREECKEN, H., Global Random Walk Simulations of
Diffusion. Journal Comp. PhysVol.186, 2003, pp.527-544.

WILKINSON, D.H., A Numerical Solution of the Analysis and Design Problems for the
Flow Past One or More Aerofoils or Cascades. A.E.C., R&M. No. 3545, L967.

WILKINSON, D.H,, The Analysis and Design of Blade Shape for Radial, Mixed and Axial
Turbomachines with fncompressible Flow. M.E.L. Report No. W/M(3F), English Electric
Co., Whetstone Leister, 1969.
 LL2

XU, C., SuÉace Vorticity Modeling of Flow Around Aitfoils. Computational Mechanics,

Vol. 21, No. 6, 1998, pp. 526 - 532.