NASA Technical

Memorandum

110446

Turbulence Modeling Validation,

Testing, and Development
J. E. Bardina, Caelum Research Corporation,
Moffett Field, California
P. G. Huang, University of Kentucky, Lexington, Kentucky
T. J. Coakley, Ames Research Center, Moffett Field, California

April 1997

National Aeronautics and

Space Administration
Ames Research Center
Moffett Field, California 94035-1000

CONTENTS

Page
General

viii

Nomenclature

Summary

Introduction

Part

Section

A. Governing

1. Governing

Equations

b. Turbulent

d. Boundary

eddy viscosity

closure

of transport

equations

Part

2. Classification

Section

3. k-E Two-Equation

of Turbulence

Model

classification:

B. Turbulence

Models

Models

(Launder-Sharma)

9
9

32011

b. Summary
c. Model

equations

d. Model

constants

10

conditions

f. Numerical

implementation

4. k-c0 Two-Equation

10

and parameters

e. Boundary

a. Model

model

conditions

Section

Flow
4

equations

c. Transformation

Section

for a Turbulent

Equations

a. Conservation

a. Model

of Motion

Model

classification:

10

12

(Wilcox)

12

32011

12

b. Summary
c. Model

equations

d. Model

constants

12
13

and parameters

iii

e.Boundaryconditions

13

f. Numericalimplementation

13

Section5. SST

Two-Equation

a. Model

Section

Model

classification:

14

(Menter)

14

32011

b. Summary

14

c. Model

equations

14

d. Model

constants

15

and parameters

e. Boundary

conditions

16

f. Numerical

implementation

16

6. One-Equation
a. Model

Model

classification:

17

(Spalart-Allmaras)
31011

17

b. Summary

17

c. Model

17

equations

d. Boundary

19

conditions

e. Numerical

19

implementation
Part

C. Turbulent

Free-Shear

20

Summary
Mathematical

Section

Flows

Formulation

20

Layer

22

a. Introduction

22

7. Mixing

b. Experimental

22

data

23

c. Results
1) Velocity
2) Spreading
d. Sensitivity
1) Freestream

profiles

23

rate

23

analyses

24

turbulence

2) Grid resolution

sensitivity

sensitivity

study

iv

study

24
24

3) Initial

conditions

4) Code

Section

8. Plane

sensitivity

sensitivity

study

and Mach

number

25
sensitivity

28

a. Introduction
b. Experimental

28

data

28

c. Results
1) Velocity

d. Sensitivity

29

rate

30

analyses

1) Freestream

turbulence

sensitivity

3) Initial

9. Round

29

profiles

2) Spreading

Section

conditions

sensitivity

study

sensitivity

30

study

32
32

a. Introduction
b. Experimental

32

data

32

c. Results
1) Velocity

33

profiles

2) Spreading
d. Sensitivity

33

rate

34

analyses

1) Freestream

turbulence

sensitivity

3) Initial

Section

30
30

study

Jet

2) Grid

25

28

Jet

2) Grid

study

conditions

sensitivity

study

34

study
sensitivity

34

study

34

Wake

36

a. Introduction

36

10. Plane

b. Experimental

36

data

36

c. Results
1) Velocity

37

profiles

rate

37

analyses

38

2) Spreading
d

Sensitivity
1) Freestream

turbulence

2) Grid sensitivity
3) Initial

Section

38

study

38

study

conditions

11. Compressible

sensitivity

sensitivity

Mixing

38

study

40

Layer

a. Introduction

40

b. Experimental

40

data

c. Results

41

1) Velocity

profiles

41

2) Spreading

rate

42

analyses

43

d. Sensitivity
1) Freestream
2) Grid

sensitivity

3) Initial
Part
Section

turbulence

12. Incompressible
a. Empirical

43

study

43

study

conditions

D. Attached

sensitivity

sensitivity

and Separated

Turbulent

Boundary

Layer

Flows

Layer

48

of data

48

Boundary
correlations

44

study

b. Results

49

1) Grid sensitivity
2) Sensitivity

to the distance

3) Freestream

sensitivity

4) Effects
5) Mach

Section

49

study

13. Compressible
a. Empirical

correlations

5O
50

and code-independence

study

50
50

effects

Flat Plate

from the wall, Y+I

study

of inlet conditions
number

of the first point

Flow

56

of data

56

vi

b. Results

57

Section14.AxisymmetricBoundaryLayerwith Adverse
a. Experimental

data and description

Pressure

of numerical

Gradient

procedure

b. Results

Section

15. Transonic

Separation

Flow over

an Axisymmetric

"Bump"

67

Airfoil:

a. Experimental

RAE

2822

74
74

data

b. Results

74
Part

Section

Appendix.

17. Conclusions

Self-Similar

E. General

Conclusions

and Recommendations

Equations

a. Compressible

of variables

d. Self-similar

equations

81

Flows

81

layer equations

two-dimensional

c. Separation

79

for Free-Shear

boundary

b. Incompressible

e. Boundary

67
67

data

b. Results

16. Transonic

59
59

a. Experimental

Section

59

and axisymmetric

and coordinate

transformation

equations

82
82
83
85

conditions

References

86

vii

General

Nomenclature

The following
included
have

list shows

the nomenclature

in this list and are explained

more than
* General

one meaning

of the general

in immediate

and they are described

symbols.

Special

of the particular

in the particular

text.

Symbols

Cd

Drag

coefficient

cy

Skin

friction

Cl

Lift coefficient

Specific

heat at constant

pressure

Cv

Specific

heat at constant

volume

Drag

Specific

Sub-index;

i = 1,2,3

Sub-index;

j = 1,2,3

Turbulent

kinetic

energy

Turbulent

mixing

length

L_

Characteristic

Mass

Static

Pr

context

coefficient

force
energy,

p/(y-

1) + pujuj/2

length

for a perfect

gas

scale

pressure

Prandtl

number,

pl.tCp/_C

Pr t

Turbulent

qj

Total

qlj

Molecular

heat-flux

qtj

Turbulent

heat-flux

Magnitude

Re

Reynolds

Magnitude

SU

Mean

Time

Temperature

uj

Mean

Prandtl

heat-flux

number,

rate in the xj coordinate

rate in the xj coordinate

number,

rotation

strain

rate tensor,

Characteristic

xj

Space

coordinate

Space

distance

rate,

direction
direction

(2f2/jC26) 1/2

pUL/g

of mean

flow velocity

direction

rate in the xj coordinate

of mean

strain

pt.ttcp/_: T

rate, (2SijSij) 1/2

(_gui/_xj +3uj/Oxi)/2

component

velocity

in the xj coordinate

scale

component;
to nearest

j = 1,2,3

no-slip

surface

..o

Vlll

direction

symbols

models.

are not

Few symbols

* General

Greek

Symbols

Ratio

Boundary

5"

Displacement

6ij

Kronecker

of specific

heats,

layer

thickness
thickness

second-order

Turbulent

_ijl

Alternating

third-order

Momentum

thickness

Thermal

tensor

dissipation

rate
tensor

conductivity

Turbulent

eddy thermal

Molecular

viscosity

_t

Turbulent

eddy

Kinematic

Vt

Turbulent

Density

_ij

Total

viscosity,

eddy viscosity

tensor
stress

zt ij

Turbulent

Reynolds

lg

Stream

O3

Specific

turbulent

O3i

Vorticity

vector

tensor
stress

tensor

function

Magnitude
Mean

m/r

kinematic

stress

conductivity

viscosity

Molecular

"_1 ij

Cp/c v

dissipation
component,

of mean

rotation

tensor,

rate, turbulent

frequency

Eijl(O_Ul/OXj)

vorticity
(3Oui/OOx j - _Ouj/_Oxi)/2

ix

Turbulence

Modeling

Testing,

Validation,

and Development

J. E. Bardina,*

R G. Huang, t and T. J. Coakley

Ames Research

Center

Summary

The
fields

primary

objective

and to compare

of this work

and evaluate

is to provide

the performance

accurate

numerical

solutions

of selected

turbulence

models

for selected

flow

with experimental

results.
Four popular
turbulence
models have been tested and validated
against
experimental
data
of ten turbulent
flows. The models are: 1) the two-equation
k-co model of Wilcox, 2) the two-equation
k-_ model
equation

of Launder
model

a mixing

of Spalart

layer,

boundary
layer
rated boundary
transonic
used

in model

parameters,
experimental
flows,

a round

3) the two-equation

and Allmaras.
jet,

a plane

The flows

jet,

a plane

k-co/k-E SST model

investigated
wake,

and

The experimental
developments.

of motion
boundary
data

data for these flows

The results

and boundary
conditions,

and

and the performance

are shown

conditions;
numerical

a compressible

are well established

in the following

implementation;

and 4) the one-

flows

mixing

consisting

layer;

and

of
five

a Mach 5 adiabatic
flat plate, a sepalayer interaction,
and an RAE 2822
and have been

four

Part B describes

of the models

of Menter,

are five free shear

flows consisting
of an incompressible
flat plate,
layer, an axisymmetric
shock-wave/boundary

airfoil.

the equations

and Sharma,

sections:

the model
and

in the free-shear

Parts
flows

Part A describes

equations,
C and

extensively
constants,

D describe

and the boundary

respectively.

*Caelum Research Corporation, Moffett Field, California. Formerly at MCAT Inc., San Jose, California.
?University of Kentucky, Lexington, Kentucky. Formerly at MCAT Inc., San Jose, California.

the
layer

Introduction
Turbulenceis oneof the key phenomenain fluid
designs
ment

is the accuracy
of improved

ments

of

different
zation

limitations
models

of advanced

of present

is necessary

and updating

turbulence

models.

to accurately

predict

of metrics

for developing

try, and
research

by defining

with specific

accurate

universities
(refs. 1-3).

A unified

testing

to achieve

models

advances

process.

to understand
evaluation

flows.

in this important

criterion

of engineering

fluid dynamics

aim of this effort

(CFD)

codes

is the development

of-the-art
turbulence
models
and advanced
models.

of a methodology

which

can be used

aerodynamic

flows

for the validation

and applied

former

enables

is interested

a user to select

in predicting

the model

performance

in a few

free jet or a separating

metrics

required

study,

code-

four of the currently

two-equation

5 and 6), the two-equation

model

of Spalart
obtained

section.
codes.
their

and Allmaras
with other

It is suggested
The

implementations
to reduce

2% when

comparing

This initial
and separated
used

basic

posted

involved,

with the standard

solutions

addresses
layers.

layer

basic

flows

own

it is expected
provided
flows.
data

free-shear

in this report,

selected

of

such

criteria

as
and

in the benchmark

k-e model

of Launder

are encouraged

be repeatable
CFD

code

application

by other

developers

codes.

that the errors

to validate

illustrated

With

in each
computer

to validate
some

be reduced

careful

to less than

in this report.
The flows

include

are well established

flows

considered

in this validation

if one

judgement

to determine

the guidelines

should

stateof new

(ref. 7), and the one-equation

and users

enables

round jet, plane jet, plane

considered

were

of Menter

developers

in their

ten experimental

layer,

solutions.

solutions

The experimental
The

mixing

numerical

in this report

model

errors

illustrated

that they follow

of current

For example,

(ref. 4), the two-equation

8 and 9). Model

The specific

and testing

on a personal

and users

models

and

is to evaluate
the perforsolutions for the models. The

modelers

k-co SST model

of "well-controlled"

developments.

boundary

of Wilcox

of interest.

and testing

based

comparisons

indus-

with effective
turbuand implemented
in

for his application.

selection

most popular

but it is advised

of the turbulence

effort

incompressible

(refs.

models,

the possible

boundary

in model

similar

k-c0 model

that the results

establishment

measures

benchmark

and grid-independent

and Sharma
results

model

layer. The later enables

comparisons--the
(refs.

turbulence

jet, one may make

relevant

boundary

to produce

In the present

a "desirable"

an impinging

provides

development

to the development

The results and discussion

presented
here serve two purposes-one
of turbulence
models; the other is to establish
standard numerical

mance

The

used to solve

and

the defining

for NASA,

The general
objective
of this research
is to provide
NASA and industry
lence models
to be used with the Reynolds-averaged
Navier-Stokes
equations
computational

models.

for turbulence

involves

a framework

area

minimi-

the capabilities

The methodology

and constitutes

under

The

turbulence

This process

procedures.

simulation

their performance

for evaluating

and validation

state-of-the-art

about

criteria
effective

Developrequire-

numerical

are necessary

turbulent

in aerodynamic
flows.

to technological
and

and risks in the design

and

complex

due

of information

standard

models

challenge
turbulent

in computers

uncertainties

turbulence

A major

of complex

in the last decade

advances

and the lack

additional

risks can be achieved

and testing

and

of models

create

dynamics.

for simulations

has increased

systems

The proliferation

of these

models

models

aerodynamic

flow conditions

Validation

basis

turbulence

present

capabilities.

of turbulence

free-shear

flows,

and has been

in this validation

wake,

and compressible

are

an incompressible

attached

extensively
are

a self-

mixing

layer.

flat plate,

Mach5 adiabaticflat plate,a separated

boundarylayerwith adversepressuregradient,anaxisymmetric shock-wave/boundary
layerflow, andanRAE 2822airfoil.
The model validation efforts concentrateon the predictionsof the experimentaldataand the
sensitivityto physicalandnumericaleffects.Theseeffectsincludesensitivityanalysisto grid resolution, freestreamturbulence,initial conditions,andcodeselection.
Weintroduceeachtestcasewith a generaldiscussionof theflow anddataselectedfor comparison,which may involve actualmeasureddatasetsor maybe a well-establishedexperimentalcorrelation. A secondsectionaddressesissuesregardingsensitivity of the solution to establish standard
solutions.Theseissuesinvolve the assessment
of numericalandboundaryconditionerrors and the
measuresto reducethem.In particular,the "standard"solutionswill be reportedin this section.In
somecases,the computationalresultsare reportedin termsof percenterrorsagainstexperimental
correlations.However,this doesnot indicatethatthe experimentalcorrelationsarethe"law of nature"
but the purposeis to provide a convenientway to distinguishthe modeldifferences.The last section
providesparticularinstructionsof thecasetoestablishmodelsolutionsthat arenot only grid independentbut alsocodeindependent.
The presentreportis divided in four mainsections.PartA describesthe govemingequationsof
turbulent flow which in this casearethe Reynolds-averaged
Navier-Stokesequations.In sometest
cases,theequationsarereducedto the two-dimensionalor axisymmetricboundarylayerequations,or
the self-similar free-shear-layerequations.PartB describesthe turbulencemodels,the model equations,boundaryconditions,andnumericaltreatment.PartC describesthe free-shearflows andPartD
describesthe boundarylayer flows. Eachflow casedescribesthe experimentaldata,model predictions, andsensitivitystudy.

Part A. Governing

Section

1. Governing

a. Conservation
The

of Motion

Flow

equations
equations

compressible
laws and the

of motion

for a continuous

Navier-Stokes
fluid property

equations
laws. The

energy and the equation

of state for a perfect
variables
and compact
tensor notation are
Mass

for a Turbulent

Equations

governing

time-dependent
conservation

Equations

conservation

viscous

fluid

are the three-dimensional

(ref. 3). These equations

represent
the flow
conservation
laws of mass, momentum,
and

gas expressed

in terms

of Reynolds

density-averaged

equation:

0p
0
0_ +b-_xj(PuJ ) = 0
Momentum

conservation

(1.1)

equation:
_PUi
= 0
_---_ + "_j

Energy

conservation

(fJUjUi

+ P_ij

equation:

+ p)
Perfect

(1.2)

-- "Cij)

gas equation

- ui'cij-qj]

(1.3)

of state:
p = RpT

Repeated
indices
in any term denote summation
dimensional
flows and j = 1,2 for two-dimensional
For

isothermal

and momentum

eddy

in terms

proportional
which

viscosity

to the

the pressure

compressibility

closure

quantities

mean

will deal exclusively

molecular
and turbulent stresses
bulent viscosities
as follows:

fluids,

spatial

these

index

equations

range

(j = 1,2,3

reduce

field is obtained

from

the mass

(as in the INS2D

code,

for example)

requires

the definition

of the turbulent

(ref. 4). For eddy viscosity

strain-rate
kinetic

over the
flows).

to the

for three-

set of mass

or continuity

equation

or other

numerical

model

tensor,

in terms of additional

are the turbulent

This report

incompressible

of these equations

of known

is modeled

lent scales

and

of artificial

The closure
fluxes

with

equations,

using the method

methods.

b. Turbulent

flows

(1.4)

and the factor

turbulent

energy

models,

of proportionality

scale variables.

and its rate of energy

with isotropic

and heat fluxes

stresses

and heat

is modeled

is the eddy

Two of the more widely

as

viscosity

used turbu-

dissipation.

eddy viscosity

are expressed

Reynolds

the stress tensor

models.

as functions

In these models,

the total

of the molecular

and tur-

 Total

molecular

and Reynolds

stresses:

"Cij= T'lij + "_tij

"elij = 2_ ( Sij - Snn_)ij/3)
'_tij = 2gt (Sij - Snng)ij/3)

(1.5)

- 29k80/3

where

The turbulent

stresses

ity, gt, in analogy

Heat-flux

are modeled

with Stoke's

using

the Boussinesq

law for the molecular

approximation

in terms

of the eddy

viscos-

stresses.

rates:
(1.6)

_T
qj = qlj +qtj
The turbulent

heat-transfer

rates

eddy thermal

conductivity,

nt-

In most
stresses

application

and heat-flux

and heat-flux
follows:
Total

are modeled

CFD codes,

rates

equations

= -(l_+Kt)

the equations

are expressed

1.5 and

following

in terms

are written
of Reynolds

1.6 may be written

in terms

Heat-flux

(Sij

- Snn_)ij/3

with all variables

representing

and poo (or u=) which

The closure

turbulent-scale

energy
in detail,

including

numbers.

and the shear

The shear-stress

nondimensional

variables

) - 2pk_)ij/3

as

(1.7)

equations

governing
varies

from

simple

to represent

transformed

into

geometries.

The

quantities

generalized
equations

(1.8)

and reference

on inlet or freestream

viscosity

model

with their
In sections

a one-equation

The

simplified

eddy

quantities,

based

and its rate of dissipation.

of transport

and

defined

a turbulent

c. Transformation

geometry

and Prandtl
of general

form,

of the

_-r+_rt)_.

nondimensional

are usually

using

additional

described

in nondimensional

flow in terms

rates:

qJ=-(_-l)-_e

lent kinetic

law for laminar

stresses:
_ij = 2 ( (g + gt)/Re,)

ties and

Fourier's

and three

defines

respective

variables

denoted

as L=,

p=,

conditions.

gt in terms
transport

of known
equations,

3-6 of Part B, four eddy

two-equation

turbulence

mean

quanti-

such as turbu-

viscosity

models

are

models.

equations
of motion
to very

these

complex.

problems

curvilinear
may

are applied

The equations
more

coordinates

be simplified

to different

efficiently.

flows.

of motion
For

are frequently

example,

or unstructured

to represent

The complexity

axisymmetric

the

systems

of the flow
transformed

equations
to simulate

or two-dimensional

are

often

complex
flows,

andmay be simplified further to representboundarylayers.The equationsmay be transformedinto

self-similarcoordinatesandvariablesto representequilibriumboundaryandfree-shearlayers.
The free-shearlayerstestedin this reporthavebeensolvedusing the Navier-Stokesequations
and similarity equations.In particular,the similarity equationsto representthe free-shearlayersare
given in the appendixof this report.Theseequationsare reportedfor the purposeof indicating the
preciseassumptionsandapproximationsusedin the formulations.
d. Boundary conditions
The particular boundaryconditionsimposedon the flow variablesare flow dependent,and
they will be describedin PartsC and D. In general,most boundaryconditions are Dirichlet or
von Neumannconditions.Dirichlet conditionswith specificfunctionvaluesaretypically specifiedat
inflow boundaries,while von Neumannconditionswith specifiedzerogradientsaretypically specified at symmetryboundaries.No slip (mixed)conditionsarespecifiedat wall boundaries.Normally,
outflow conditionsareextrapolated.

Part
The
reference
dated

following

and

sections

a detailed

give

B. Turbulence

a general

description

of each

models

classification

of turbulence

one of the four

in this study.
Section

2.

Classification

of Turbulence

Section

3.

k-e Two-Equation

Section

4.

k-co Two-Equation

Model

(Wilcox)

Section

5.

SST Two-Equation

Model

(Menter)

Section

6.

One-Equation

Model

Model

Models
(Launder-Sharma)

(Spalart-Allmaras)

turbulence

models
eddy

as a framework

viscosity

models

of
vali-

Section

2. Classification

of Turbulence

Models

The general classification

of the turbulence
models using one-point closure
framework
of reference
for the turbulence
models and is defined as follows:
Major

classification

determines

1) Reynolds

stress

model

2) Algebraic

stress

model

3) Eddy viscosity

the modeling

Subclassification

the number

2 determines

model

Subclassification

equations

the

required

3 determines

2) Compressible

number

model

of ordinary-differential

to define the turbulent

equations

or other

required

to

non-partial-

scales.

type of fluid of the model:

fluid model

fluid model

3) Incompressible

fluid model
4 determines

1) Integration

as a

stresses:

of partial-differential

the general

1) Incompressible/compressible

Subclassification

below

model

Subclassification
1 determines
define the turbulent scales.

differential

of the turbulent

is shown

the treatment

of the near wall-boundary

flow:

to the wall

2) Wall function
3) Algebraic
4) Switch
Other

For example,
viscosity

to one-equation

models

described

models,

with matching

point
near the wall

that may not be easily

by the developer
the standard

classifiable

in the preceding

format should

be clearly

and so noted.
two-equation

with two partial-differential

k-e and k-c0 turbulence

equations

models

for incompressible

are classified

and compressible

as eddy
fluids,

and integration
capability
to the wall, i.e., 32011. (The number 32011 denotes that the major classification is 3, the subclassification
1 is 2, the subclassification
2 is 0, the subclassification
3 is 1, and the
subclassification

4 is 1.) The one-equation

model, with one partial-differential

equation
tion capability
to the wall, i.e., 31011.

Spalart-Allmaras
for incompressible

model

is classified

and compressible

as an eddy
fluids,

viscosity

and integra-

Section 3.
a. Model

k-e Two-Equation
classification:

Two-equation
gration

Model

(Launder-Sharma)

32011
eddy

viscosity

model

for incompressible/compressible

4-6)

is the most

turbulent

flows

with inte-

to the wall.

b. Summary
The
viscosity

k-e model
model.

(refs.

Different

versions

widely

of this model

known
are found

and extensively

used

two-equation

in the literature.

The

main

eddy

references

on

this model are described

by Jones and Launder (ref. 5) and Launder and Sharma (ref. 6). The formulation used here is the one of Launder
and Sharma. This model is sometimes
referred
in this report as
the standard
mercially

k-e model.
available,

The
algebraic

was

prescription

This

formulations

and should

k-e model

for two scalar

bulent kinetic

Other

be included

originally

of the turbulent

model

has given

(ref. 4). The predictions

model

requires

reasonably

explicit

The

(refs. 4 and

in the literature

flows.

or are com-

10).

the mixing-length

scale in complex

good

results

flows, the
gradients,

of the model

wall-damping

model

Transport

and to avoid

equations

the

are solved

for free-shear-layer
model gives good
but is less accurate

are insensitive

functions

flows

with

relatively

agreement
with
for large adverse

to freestream

values

small

experimental
pressure
gra-

of the turbulence.

and the use of fine grid spacing

near solid

The

walls.

equations
Reynolds

stresses

are modeled

gt is the eddy viscosity,

turbulent

to improve

length

in terms

"r,tij = 21.tt ( Sijwhere

work

are reported

properties
of turbulence.
The k-equation
is a model of the transport equation
for the turenergy, and the e-equation
is a model for the dissipation
rate of turbulent
kinetic energy.

dients

The

in future

developed

pressure
gradients.
For wall bounded
results for zero and small mean pressure

c. Model

of the k-e model

kinetic

energy,

eddy

viscosity

dissipation

of the eddy viscosity

SnnSij/3)

is defined

as a function

(3.1)

- 2pk_ij/3

Sij is the mean-velocity

and 5ij is the Kronecker

as follows:

strain-rate

tensor,

p is the fluid density,

k is the

delta.
of the turbulent

kinetic

energy,

k, and the turbulent

rate, e, as
2

IXt = c_f _pk

/e

(3.2)

The edd/Y2viscosity
scale, k 3 /e, based
rium analysis
lence

Reynolds

at high Reynolds
number,

The turbulence
Turbulence
3pk

1/2

is scaled with the fluid density, p, the turbulent

velocity scale, k
, and the lengthon local dimensional
analysis. The model coefficient,
cla, is determined
by equilibnumbers,

and the damping

function,f_t,

is modeled

in terms

of a turbu-

Re t = pk2/el.t.

transport

equations

energy

transport

3(

3k

_'_" + _'_xj(puj_'_Xj--

for the k-e Launder-Sharma

model

are defined

below.

equation:
I't_13k

( I'l" + _)_"_Xj)

(3.3)
=

TIijSiJ-PE

+*k

 Energy

dissipation

transport

OP_+Ot _
where

Ouje-

the right-hand-side

d. Model

constants

The model

terms

equation:

g+<

= CEl_'r.tqSij-ce2f20-_+_e

represent

production,

dissipation,

and wall terms,

constants

are defined

_k

below

fit

ce

functions

ce2 = 1.92

= 1.3

( l +O.O2Ret

)2)

and
f2

= 1-

wall terms

Re t = p.___k
ga

0.3exp(-Re_)

and

simpler

u s is the flow velocity

parallel

_t_2Us')2

_e =

laTt

7)

(3.7)

to the wall.

conditions

Boundary
conditions
relationships:

at no-slip

surfaces

k = 0
when

(3.6)

are
2

e. Boundary

(3.5)

0.9

Pr t =

are

exp(-3.4/

as

C_l = 1.45

= 1.0

The near wall damping

where

respectively.

and parameters

cit = 0.09

and the explicit

(3.4)

the above

wall terms

e - qbJp. Zero-gradient
ditions are discussed
f. Numerical

are included

with integration

and

to the wall are given

by the following

e = 0

(3.8)

(see eq. 3.7) and the effective

conditions
are applied at symmetry
further for each flow case.

boundaries,

dissipation

rate is expressed

and additional

boundary

as
con-

implementation

The turbulence
recommended

transport

numerical

equations

treatment

are similar

to the transport

of each term of the turbulence

equations

of the mean flow. The

transport equations

is the numerical

treatment
used in this work for the mean flow equations
of each particular numerical code. Although
not including
any specific reference, these statements
reflect the general treatment used in different
numerical

methods

Second-order
terms

and codes

developed

numerical

differences

of the flow and turbulence

ers, second-order

upwind

model

flux-difference

and applied

at NASA

are recommended
equations.
splitting

Ames

recommended,
although in some cases first-order
methods
bulence
equations.
Implicit second-order
central-difference

I0

Center.

and used for the convective

For the convective

methods

Research

terms using Navier-Stokes

with flux-limiters
have been
methods

and diffusive

have

been

used successfully
for the diffusion

solv-

used and are

with the turterms of the

momentum,
methods

energy,
provide

The
tance

and

turbulence

efficient

numerical

numerical

treatment

for the stability

terms

and

increase

in the diagonal

lent kinetic

energy

of source

all positive
dominance

representation
equation

where

Pk and D k represent

subscript
additional

n and n+l
details

represent

terms

have

and good

It is recommended

used

agreement

portion

of the approximate

= [Pk-Dk]n-[_'ff

the sum of the positive

of the scheme

employed

procedure

and thereby

for the source

leads

enhances
terms

impor-

(destruction)
to an
its sta-

of the turbu-

here.
(3.9)

Jn(kn+l-kn)

and negative

11

is of major

all negative

This

These

and experiment.

equations

implicitly

and the next time steps,

in the procedure.

with theory

terms.

linearization

of the method

and are recommended.

model

to treat

right-hand-side

of the implicit

the current

been

of the turbulence

(production)

is indicative

[Pk-Dk]n+l

equations

procedures

of this scheme.

explicitly

bility. The following

model

source

terms,

respectively.

respectively,
See reference

and the
32 for

Section

4. k-co Two-Equation

a. Model

classification:

Two-equation
gration

Model

(Wilcox)

32011

eddy viscosity

model

for incompressible/compressible

turbulent

flows

with inte-

to the wall.

b. Summary
The k-co model

(refs.

4 and 11) is a well known

and widely

tested

two-equation

eddy

viscosity

model. The main reference

to this model is given by Wilcox (ref. 4), and his formulation
is used here.
This model is referred to here as the standard k-c0 model. The roots of this model can be traced to Kolmogorov,

Prandtl,

Saffman,

This model
model
solved

marily

was

and Wilcox

originated

in collaboration

at about the same

numerical
results

wall

boundary

off the wall.

model

conditions

In the

for mild adverse

In free-shear

freestream
c. Model

to small

it is difficult

models

in parallel

with the k-e

Convective
transport equations
rate, k and cA respectively.

due to the large values of co in the wall region.

region,

gradient

freestream

to exercise

the model

of the distance
gives

good

boundary

values of co (refs.

enough

control

in the predicted

from

agreement

The

the wall to the first

with

experimental

layer flows,

the results

11 and 12). In complex

over the local

freestream

of the k-co

Navier-Stokes

turbulence

com-

to avoid

small

results.

equations
stresses

are modeled

Ztij

eddy

rate of dissipation,

viscosity

in terms of the eddy viscosity

= 21.t t (Sij - SnnSij/3)

is defined

as a function

as follows:

- 2pk_)ij/3

of the turbulent

(4.1)
kinetic

energy,

k, and the specific

co, as follows:
P-t = pk/co

The

are

flows.

adverse-pressure-gradient

co and ambiguities

The Reynolds

The

function.
dissipation

require the specification

logarithmic

pressure

layer and

are sensitive

putations

and was developed

of Wilcox has proven to be superior in numerical stability to the k-e model prisublayer
near the wall. This model does not require explicit wall-damping
func-

tions as do the k-e and other two-equation

point

time

as an alternative
to define the eddy viscosity
for the turbulent kinetic energy and its specific
The k-co mode!
in the viscous

with other scientists.

two transport

model

equations

pujk

(4.2)

for the k and co scalar

turbulence

scales

are defined

below.
(4.3)

-_

Ot

--

pujco _

(! t + G,l.tt ) 3k

(it

+ (Ygt)

12

oco

"_tijS ij

= o_xtijSij

- _pco

(4.4)

d. Model

constants

Following

and parameters

Wilcox

(ref. 4), the model constants

5
9

_=

e. Boundary

[3

are defined

3
=4-0

= 0.5

_*

values

recommended

13"

= 0.5

as
9
10-'--0

Pr t = 0.9

conditions

The choice

of freestream

03

>

X u oo
L

gtoo < 10

-2

for boundary

_tmax

layer

L is the approximate

The factor

needed

of the computational

_, = 10 has been

domain

recommended

(ref. 3) are

(4.6)
Poo

length

of proportionality

flows

oo

where

(4.5)

and

oo

Uo_ is the characteristic

velocity.

(ref. 3).

Free-shear
layers are more sensitive to small freestream
values of 03_ and larger values of 03 are
in the freestream.
Following
a similar analysis to the one referenced
above, and based on the

analysis
of self-similarity,
a value of _, may be determined.
This analysis
is not included
here; however, it shows a value of at least X = 40 for mixing layers, increasing
up to _, = 80 for round jets.
Boundary

conditions

at no-slip

surfaces

k = 0

and

are given

by the following

co = 10

61.t

relationships:
(4.7)

1313(Yl)

where

tions

Yl is the distance
Zero-gradient
are discussed

f. Numerical
The

away

from the wall and yl + < 1.

conditions
are applied
at symmetry
further for each flow case.

boundaries,

and

additional

boundary

condi-

implementation
numerical

pages 10 and
difference--a
co-equation

of the first point

implementation

of the

k-03 model

equations

is similar

to that

11 for two-equation
models.
The implicit treatment
of the source
factor of 2 is introduced
in the implicit treatment
of the dissipation

due to the proportionality

of this term to 032.

13

described

on

terms shows one

term 13p032 of the

Section

5. SST Two-Equation

a. Model

classification:

Two-equation
gration

Model

(Menter)

32011

eddy viscosity

to the wall. Following

model

Menter,

for incompressible/compressible

this model

is expressed

turbulent

flows

in terms of a k-co model

with inte-

formulation.

b. Summary
The k-co SST shear-stress-transport
ments

of existing

two-equation

model

models.

(refs.

7 and 12-14)

The two major features

combines

of this model

several

desirable

ele-

are a zonal weighting

of

model coefficients
and a limitation on the growth of the eddy viscosity
in rapidly strained flows. The
zonal modeling uses Wilcox's k-co model near solid walls and the standard k-e model (in a k-co formulation)

near boundary

function

layer edges

of the model

ity by forcing

and in free-shear

coefficients.

the turbulent

The shear

shear

stress

differences

between

term appears

transport

this formulation

is achieved

and the original

k-co model

the eddy viscos-

the turbulent

improves

energy
of flows

into a k-c0 formulation.

are different.

k-8 model

kinetic

the prediction

are that an additional

constants

F 1 and the transformed

with a blending

also modifies

times

the latter is transformed

and that the modeling

by a function

modeling

by a constant

constraint). This modification

and separation.

the k-co and the k-8 model,

in the co-equation

is then multiplied

stress

to be bounded

inside boundary layers (a realizability

with strong adverse pressure gradients
In order to blend

layers. This switching

cross-diffusion

The original

by a function

The

k-co model

(1 - F1) and the cor-

responding
equations of each model are added together. The function F 1 is designed to be a value of
one in the near wall region (activating
the original model) and zero far from the wall. The blending
takes place

in the wake

The SST model

region

of the boundary

also modifies

layer.

the turbulent

eddy viscosity

separated

flows.

Two-equation

models

generally

boundary

layer due to adverse

pressure

gradients.

mation

of the effects

mance

estimates

do not
model

account
(ref.

models

for the

energy.

of the eddy
c. Model

important

viscosity

effect

using

which

The reason

effects

of the shear

is achieved

a blending

function

deficiency,

generally

stress

results

as being

and

separation

in too optimistic

stresses.

The

can be obtained
to that

by a modification

F 2 in boundary

of

of the

to an underesti-

is that two-equation

proportional

model

the prediction

leading

results

of the turbulent

improved

in the present

to improve

the retardation

for this deficiency

of transport

that significantly

the transport

A similar

underpredict

This is a serious

interaction

bodies.

15) has demonstrated

by modeling

kinetic

of viscous-inviscid

for aerodynamic

function

performodels

Johnson-King
with algebraic
of the turbulent

in the formulation

layer flows.

equations

The eddy viscosity

is defined
specific dissipation
rate or turbulent
_
_t In turbulent

boundary

bulent

stress

shear

layers,

function

of the turbulent

kinetic

energy,

k, and

pk/co
max[1;f2FE/(atco)]

the maximum

to be bounded

an auxiliary
function
defined as a function

as the following
frequency,
co:

value

by the turbulent

F 2 and the absolute

of wall distance y as

value

al
of the eddy
kinetic

(5.1)

viscosity

is limited

times

a 1. This effect

energy

of the vorticity,

14

= 0.31

2. The

auxiliary

by forcing

the tur-

is achieved

with

function,

F 2, is

F 2 = tanh
The two transport
model
kinetic

coefficients
energy is

"_7

equations

pujk

500} /2

;--'--5--"
k 00 coy
py

of the model

of the original

at-

,max,z_

are defined

co and e model

(_.1 + _kgt)

below

equations.

"- "CtijS_j

with a blending

The

transport

(5.2)
function

equation

FI for the

for the turbulent

(5.3)

- fJ*pcok

and the transport

equation

_Pco
(pu;co-_t
+_,_
OXj\
where

for the specific

(_ + t/'tt)_-_xmj]

the last term of equation

co-equation
proportional

from

dissipation

5.4 represents

the original

to the absolute

e-equation.
value

of turbulence

= Pc-13Pco2

+ 2 (1-F1)

the cross-diffusion

The

of vorticity

production
(ref.

is

P_2_k03
_Xj_Xj_co

term that appears

term

of co is sometimes

(5.4)

in the transformed
approximated

14):

Pc0 - 2yp (Sis - OOSn,,Sij/3 ) Sij -- 79f22

d. Model

constants

and

as

(5.5)

parameters

The function F 1 is designed to blend the model coefficients

of the original k-co model in boundary layer zones with the transformed
k-e model in free-shear
layer and freestream
zones. This function takes

the value

of one on no-slip

and goes to zero at the boundary

surfaces

layer

edge. This auxiliary

[(
F
= tanh_lmtnlmaXlo0-_
L_
k

and near one over

[-

a large portion

blending

function,

500g];4PGco2k]_4_
;-----7I
D
2
coypy
coJ c
kcoy J

of the boundary
F l, is defined

layer,

as

)f
(5.6)

F29coz_k

= max
where
where

CDko _ stands
The constants

CDko _

for cross-diffusion

The model

of the SST model

denoted

coefficients

of the

original

co

in the k-co model.

are

a I = 0.31

coefficients

_co ,,,-20]
.
_ ;_v
,
OxjOxj
J

13" = 0.09

13,7, ok, and Cycodenoted

k-co model,

denoted

_: = 0.41
with the symbol

as _1, with

those

(5.7)
_ are defined
of the

by blending

transformed

k-e

the

model,

as _2.
= F_q_ 1 + (1-F1)_

with the coefficients

of the original

models

where
defined

as

15

q_ = {ok, Oo:,13,7}

(5.8)

 Innermodelcoefficients:
= 0.85

(Ykl

_oJl

= 0.5

131 = 0.075
(5.9)

_,_ = _1/_* - %1_:2/,4_

Outer

model

coefficients:

Gk2

1.0

0.856

[32 = 0.0828

[_2/_*-

Oto2 K2/'f_-_

0.440

conditions

The boundary
conditions
tion 4 for the k-03 model.
f. Numerical
The

(30.)2

(5.1o)
_2

e. Boundary

= 0.553

of the k-03 SST model

equations

are the same

as those described

implementation
numerical

implementation

of the SST model

equations

is similar

to the one described

pages 10 and 1 1 for two-equation

models. The implicit treatment
of the source
tion includes a factor of 2 for the dissipation term [3p032 due to the proportionality
The
with

production

a limiter

numerically
eddy

term

of o_ written

on the production
in some

viscosity

turbulent

in sec-

cases

(ref.

in the stagnation

and nonturbulent

as a function

of turbulent

kinetic

14). This procedure

region

interfaces.

of airfoils
These

of the absolute
energy,
can help

vorticity

Pk, have
to prevent

and can help to suppress

effects

have

been

also been

(see eq. 5.5) together

found

16

buildup

oscillations

of
near

with other two-equation

models. The limiter recommended

on the production
of turbulent
kinetic energy
tion of turbulent
kinetic energy to a value not larger than 20 times the dissipation
energy.

to be beneficial

the unrealistic
numerical

observed

on

terms of the co equaof this term to 032.

bounds the producof turbulent

kinetic

Section6. One-Equation
a. Model

classification:

One-equation
gration

Model

(Spalart-Allmaras)

31011

eddy

viscosity

model

for incompressible/compressible

turbulent

flows

with inte-

to the wail.

b. Summary
The

Spalart-Allmaras

model

(refs.

4, 8, and 9) is a relatively

recent

eddy

viscosity

model

based

on a transport
equation
for the turbulent
viscosity.
This model was inspired
from an earlier model
developed
by Baldwin and Barth (ref. 16). Its formulation
and coefficients
were defined using dimensional

analysis,

development
The
models

invariance,

aim of this model

to develop

equation

a local

turbulence

The

model

turbulent

uses

distance

2-D mixing

layers,

results.
wakes,

the predictions

for complex

to the nearest

capabilities,

convergence

The model

provided

flows,

The empirical

and fiat-plate

obtained

and

with

to provide

results

boundary
algebraic

a simpler

does

layers,

not give good

wake

flows,

wall in its formulation,

that the location

in wall-bounded

in simpler

tion of flows with adverse

much as the SST model.
c. Model

model

empirical

mixing

is to improve

as fine a grid resolution

good

and selected
(2-D)

used
layer

mixing-length

alternative

to two-

and provides

smooth

of the start of transition

flows

as two-equation

laminar-

is given.

turbulence

It does

models,

and it

flows.
predictions

and flat-plate

in jet flows,
boundary

pressure

gradients

function

is defined

but gives

layers

compared

reasonably

and shows

good

improvements

with the k- and k-r.o models,

predictions

of

in the predicalthough

not as

equations

The eddy
function,fv

viscosity

in terms

of an eddy viscosity

variable,

v,

and a wall

1, as follows:
vt = vfvl

In zones

in its
flows.

models.

transition

not require
shows

Galilean

were two-dimensional

far from

wall boundaries,

The convective

Ot

transport

+ -_

the functionfv

equation

= cb

(6.1)

1 is equal

of the eddy

to one and

viscosity

( 1 - f t2) pSv + 1

vt = v .

is modeled

as

a?e

a a -]

-[ %,.I 2
Cwlfw-----_at2

+ ftlOAU

(6.3)

where

the right-hand-side

sion, nonconservative

terms
diffusion,

and transition

source

for "viscous,"

and t stands

represent

turbulence

eddy

near-wall

turbulence

destruction,

of turbulence.
for "trip"

The subscript
(start

b stands

of transition).

17

viscosity

production,
transition

for "basic,"

conservative

damping

w stands

diffu-

of production,

for "wall,"

v stands

The

model

constants

and

auxiliary

functions

shear flows, the wall model for boundary

layers,
transition
model for laminar-turbulent
transition.
The basic model constants
lent eddy viscosity
are

for flee-shear

Cbl = 0.1355

are defined

the viscous

flows

Cb2

0.622

The additional
model constants
and auxiliary
viscosity
in the boundary
layer zone are
Cwl = Cbl/_c2+

the Reynolds
Modeling

functions
shear

allow

stress

and

transition

reduction

functions

to turbulence

of production

are

where

03t is the vorticity

velocity at the field point

trip, and d is the distance

to the wall,

and diffusion

and the

of turbu-

(6.4)
destruction

of turbulent

eddy

a2
(6.5)

-6-"7"
1 +c6w3 11/6
g + Cw3
the log layer,

agreement

fv2

although

the particular

with experimental

2_--SiijSij

model

= 1

constants

defined

budget

with a source

= ctlgt

" exp

ft2

= ct3.

expl-ct4_

Ct2

at the boundary
and the velocity
from the wall.

term

data.

are given

by

Cv 1

(6.6)
= 7.1

the laminar
controlled

region
with

of the shear

the

function

layers

ftl

and

ft 2.
)t

-C,2A--_roEd2.

AU/

of

_-

Z
1 + Zf vl
to control

with the function

f ,l

= 1

for free-

gt = min [0.1,
ctl

for the

r-_

= g

S=

3
+ cvl

controlled

the production

for the near wall flow regions

and

functions

to predict

_ 2fv2
(rid)
3

f vl
auxiliary

the model

and constants

7S=_S+

The

fw

is not in quantitative

functions

for integration

model

g = r+ Cw2 (r 6 - r )

Cw3 = 2

auxiliary

of the basic

6 = 2/3

(1 +Cb2)/(_

Cw2 = 0.3

These

model

to control

in terms

(gtd,)

2)

(6.7)

(03tAxt)
Ct3

layer trip point,

at the trip point,

18

]
1.2

Ct4

---- 0.5

AU is the norm

of the difference

Ax t is the grid spacing

along

between

the

the wall at the

d. Boundary

conditions

The ideal value of the freestream

with zero freestream values, and values

eddy viscosity
variable is zero. Some solvers
have trouble
smaller than v/2 have been recommended.
Small freestream

values

flows;

are recommended

in velocity

profiles

e. Numerical
The
similar

for free-shear

and spreading

otherwise,

the solutions

rates. At wall boundaries

show

the eddy viscosity

freestream

dependence

is zero.

implementation

numerical

implementation

to the one described

on pages

of the Spalart-Allmaras
10 and 11 for two-equation

19

model

equations

models.

that

we used

here

is

Part

C. Turbulent

Free-Shear

Flows

Summary
Turbulent
These
and

flows

flee-shear

provide

calibration

the results

a good balance

of model

flows.

as turbulent

for testing

coefficients.

of each flow are shown

Four turbulence
shear

flows are defined

the predictive

Five turbulent

in the following
Mixing

Section

8.

Plane Jet

Section

9.

Round

Jet

Section

10.

Plane

Wake

Section

11.

Compressible

included

models

capability

with solid boundaries.

of models

flows

on diffusion

are included

Mixing

are validated

in the present

study

Layer

for the prediction

of these" turbulent

are the two-equation

k-f.o model

These

in Part B of this report.

The validation
of each model is mainly based on the ability of the models
profile and spreading
rate of each one of these fully developed
free-shear

velocity
observed
stants

are described

in the predictions

have

been

defined

Sensitivity

of these
or selected

analyses

flows

may be explained

during

the model

of the validation

results

in part because

development
to freestream

based

particular,

the

k-03 model

showed
no sensitivity
mean velocity
profiles
flow

beyond

ffeestream
pressible

the
Mach

mixing

Mathematical

showed

sensitivity

to low

initial

development
used

These

zone.

The

results

showed

and ini-

and

the k-_ model

of the shapes of nondimensional

to different
initial profiles
in each

of the

codes

is also shown

good consistency

con-

flows.

grid resolution,

03 values,

sensitivity

with compressible

of the model

great sensitivity
to grid resoluwith some of the models.
In

freestream

to low freestream
_ values. The predictions
and their spreading
rates were insensitive

numbers
layer.

large

k-03 SST
8 and 9).

to predict the mean

flows. Differences

on one of these

turbulence,

tial profiles are included in each section. In general, the results showed
tion with all the turbulence
models,
and to freestream
turbulence

some

free-

of Wilcox

and Sharma (refs. 5 and 6), the two-equation

model of Spalart and Allmaras
(S-A) (refs.

and their equations

and

Layer

(ref. 4), the two-equation

k-e model of Launder
model of Menter
(ref. 7), and the one-equation
models

effects

in this study,

sections:

7.

eddy viscosity

not interacting

free-shear

Section

The four models

flows

results

to numerical

in section

codes

and

7 for the incom-

from flow to flow.

Formulation

The

governing

equations

are the Navier-Stokes

equations

in Part A of this report.

For self-similar

free-shear

form

transformations

and the boundary

dix gives
similarity

the self-similar
free-shear
flow equations
and boundary
conditions
expressed
in terms of
coordinates.
These equations
were used to obtain most of the free-shear
flow results shown

here. The agreement

between
incompressible
mixing layer.

Navier-Stokes

and self-similar

20

layer

predictions

equations

and are

a simpler

similarity

these

numbers

described

through

flows,

at high Reynolds

can be reduced

approximation.

is given

to

The appen-

in section

7 for the

The boundaryconditionsimposedon the equationsof motion are definedby the respective

statesof the freestreams,andreflectionconditionsin symmetryplanesof planejets andaxesof symmetry in round jets. Integralconstraintsbasedon the conservationof total momentumandthe displacementthickness of the layer were usedfor the jet and wake calculations,as describedin
reference4.
For the self-similarequations,theequationsweresolvedin deltaform usinganimplicit relaxation procedureuntil the residualsof the dependentvariablesconvergedto zero within computer
machineaccuracy.The numericalmethodusedto solvetheseequationswasanimplicit upwind algorithm for the convectiveterms,second-order-implicitcentraldifferencesfor thediffusion terms,and
implicit treatmentof the destructionsourcetermsof the turbulencemodel equations.This methodis
describedfurther in the appendix.The solutionsof theseequationsshowedfast rateof convergence.
Very small andlargevaluesof freestreamvariablesrequiredupperandlower limiters in theeddyviscosity during thetransientrelaxationtime to achievefastconvergence.
No significant differenceswere observedbetweenresults obtainedwith self-similar and
Navier-Stokessimulations with similar correspondingboundary conditions. However, different
freestreamconditionsare normally usedwith different methods.For example,self-similar calculations are usually done with fixed boundaryconditions.Navier-Stokessimulationswith the oneequationmodelof SpalartandAllmarasare alsonormally donewith fixed freestreamstates.On the
other hand, Navier-Stokessimulationswith two-equationturbulencemodels normally produce a
decayof freestreamturbulence,andthis decayis inverselyproportionalto the freestreamspeed.This
differenceis not anissuein modelswhoseresultsareinsensitiveto freestreamturbulence;for example, the k-E and the k-03 SST models. However, the results of mean velocity profiles obtained with the
k-03 model
Thus

are very

under

03 using

these

the

self-similar

Spalart-Allmaras
freestream

sensitive
conditions

eddy

model
viscosity,

to low values

of freestream

the k-03 model

equations

and the Navier-Stokes

are not sensitive

which

to freestream

may become

03, and 03 decays

can give different

an issue

21

results

equations.
turbulence,
in complex

rapidly

for low
The

except
flows.

in streams

freestream

results

obtained

for very

large

at rest.
values

of

with

the

values

of

Section

7. Mixing

Layer

a. Introduction
Turbulent
class

of flows

are defined
properties

mixing

layers

encountered

as the region
of fully

are of considerable

in many
between

developed

aerodynamic

two parallel

free-shear

interest

in engineering

applications
streams

flows

have

(refs.

moving
been

design

1-4 and

at different

extensively

speeds.

used

and represent

17). Mixing

layer

flows

The self-preserving

to validate

model

coeffi-

cients in almost all well-known

turbulence
models (ref. 4). Single phase, incompressible,
fully developed turbulent
mixing layer flows are considered
fundamental
flows in turbulence
modeling.
At high
layer

Reynolds

develops

numbers

an asymptotic

and

equilibrium

AFOSR-HTTM-Stanford
Conference
of 0.115 for a fully developed
mixing
profile

and its spreading

b. Experimental

and

that even
boundary

mixing

layer

velocity

numerical

simple

and

initial

effects,

of the experimental

the mixing

data in the 1980-81

Flows (ref. 17) reported

a spreading
rate
and moving streams. The mean velocity

the most significant

is fully

(ref.

The experimental
y/x = [-0.180,
-0.116,
-0.084,
-0.050,
-0.017,

formulation

parameters

in this flow.

19) where

of the governing

have

varied

between

is sensitive

by the

the available
spreading

the predictions
one freestream

-0.160,
-0.113,
--0.080,
-0.044,
-0.010,

to initial
and

experimental
rate value

are compared

data

(refs.

-0.147,
-0.106,
-0.077,
-0.036,
-0.005,

along

of motion.
experiments

conditions
Reynolds
show

against

with the proper

Experimental
in the initial

a unique

effects.
self-similar

data
zone

of flow structures
number

the experimental
freestream

and U/U 1, where

-0.145,
-0.102,
-0.069,
-0.034,
-0.004,

data,

of the

After

the

mean-

-0.135,
-0.100,
-0.065,
-0.033,
-0.002,

data

is moving

the data arrays

of Liepmann
at speed

are

-0.132,
-0.097,
-0.065,
-0.028,
0.008,

-0.129,
-0.095,
-0.063,
-0.026,
0.011,

-0.126,
-0.092,
-0.062,
-0.021,
0.011,

-0.120,
--0.089,
-0.053,
-0.019,
0.015,

0.021,
0.057,
0.094,

0.025,
0.058,
0.094,

0.027,
0.062,
0.096,

0.030,
0.062,
0.098,

0.033,
0.069,
0.106,

0.039,
0.074,
0.110,

0.042,
0.078,
0.113,

0.044,
0.079,
0.115,

0.045,
0.081,
0.118,

0.046,
0.083,
0.121,

0.054,
0.090,
0.126]

U/U 1= [0.015,
0.060,
0.114,
0.300,
0.450,
0.670,
0.885,

0.050,
0.065,
0.170,
0.285,
0.445,
0.695,
0.890,

0.055,
0.055,
0.145,
0.295,
0.475,
0.715,
0.865,

0.040,
0.100,
0.145,
0.310,
0.485,
0.708,
0.905,

0.040,
0.075,
0.137,
0.335,
0.515,
0.748,
0.923,

0.075,
0.085,
0.177,
0.380,
0.525,
0.790,
0.950,

0.055,
0.088,
0.195,
0.350,
0.530,
0.815,
0.920,

0.050,
0.090,
0.255,
0.375,
0.595,
0.815,
0.958,

0.075,
0.125,
0.220,
0.390,
0.622,
0.825,
0.970,

0.045,
0.107,
0.210,
0.435,
0.633,
0.800,
0.967,

0.060,
0.105,
0.245,
0.437,
0.635,
0.865,
0.980,

0.960,

0.980,

0.980,

0.985,

0.990,

0.985,

0.990,

0.998,

1.000,

1.000,

0.995]

22

of

17 and 18).

is at rest and the other

as 1"1= y/x + 0.007

-0.150,
-0.113,
-0.077,
-0.040,
-0.009,

equations
different

flow geometry,

data are defined

---0.176,
-0.115,
-0.081,
-0.048,
-0.016,

set of experimental

imposed

developed,

study,

selected

layer. This zone

with an asymptotic

In the present

a carefully

measurements

conditions

profile

and Laufer

requires

of the mixing

layer,

The analysis

on Complex Turbulent
layer between stagnant

rate are considered

validation

mathematical
development

state.

of transition

data

Model
reveal

far downstream

U 1.

The
mean

displacement

velocity

of 0.007

ratio profile,

is used

to set the coordinate

origin,

n = 0, at the midpoint

of the

U/Uloo -- 1/2.

c. Results

The
spreading
tions,

validation

of the turbulence

rate of the mixing

these

described

calculations

layer
were

in the appendix.

models

is shown

below.

performed

A uniform

for the prediction

by

In order

solving

distribution

of the mean

to ensure

the

velocity

accuracy

of the model

nondimensional

of 501 points

profile

similarity

was defined

within

and

predicequations

the grid domain

-0.3 < q - y/x < 0.2, the freestream

value of the nondimensional
turbulent
kinetic energy was defined
as K 1 - kl/Ul 2 - 10 .-6 (nearly zero) for the two-equation
turbulence
models, and the freestream
value
of the nondimensional
used

turbulent

to vary the freestream

velocity

profile

streamwise

was

value

defined

coordinate

viscosity

was varied

10 -30< N < 1. This was

of o3 or _ for the two-equation

as U/U 1 = 0.5 +0.5

direction

from

with its origin

tanh(10q).

turbulence
The

at the virtual

origin

models.

coordinate

Figure

of the self-similar

7.1 shows
with its origin

The results
good

agreement

difference

the comparison

of the predictions

data of Liepmann
located

where

and Laufer

the mean

of the k-_ and k-to SST models

in the middle

between

initial

defined

mean
as the

flow, the coordi-

as the freestream

speed

profiles

q = y/x and the experimental

was defined

The

x was

nate y was defined as the transverse

coordinate
direction,
and U 1 was defined
of one boundary
stream while the other freestream
was set at rest.
1) Velocity

also the method

of the mixing

the predictions

of these

of mean

profile

U/U 1 against

(ref. 5). The nondimensional

velocity

ratio was

are insensitive

zone and sharp edge

two models

velocity

to freestream
profiles

near the edge

coordinate

1/2.
turbulence

and show

at the boundaries.

The small

of the freestrearn

at rest is due

to the different
value of their diffusion
model constant,
gE- The results of the Spalart-Allmaras
model
show very good agreement
with the experimental
data, but show a wider mean velocity
profile with
very

large

values

of freestream

eddy

large values of dimensionless

eddy
errors can be controlled
by limiting
model
values

viscosity

(10 -3 <_N 1 =- vt/Ulx;

see fig. 7.2).

In practice,

viscosity
are much larger than the molecular
viscosity,
the eddy viscosity
in the freestream,
(gt/l.tl _< 10-3Reo_).

and the
The k-co

shows two different

results of mean velocity profiles, one for low values and another for high
of freestream
co, and a range of profiles in between these two values, (10 -2 <_W- o_/U 1 <_ 10).

The profiles
show significant
underprediction
in the low-speed
side and overprediction
speed side of the mixing layer with low freestream
co, (W _< 10-2), and underprediction
speed

these

side with high freestream

2) Spreading

co, (W>

in the highin the higher

10).

rates

The spreading
rate provides an estimate of the width of the mixing region and is considered
one of the most significant
parameters
that turbulence
models attempt to predict with accuracy.
The
spreading
mean

rate was defined

velocity

respectively.

profile

where

This definition

in the literature

as the distance

to measure

the

square

of spreading
the spreading

between

the points

of the nondimensional
rate is one of several
rate. The

23

main

of the nondimensional
mean

velocity

formulations

shortcoming

variable
ratio

that have

of this definition

is 0.9
been

rl in the
and 0.1,
proposed

is reported

to

be the large differencesthat the modelspredict nearthe edgezonesof the mixing layer (refs. 17
and 18). The main advantageof this definitionis its widespreadusein turbulencemodeling(refs. 17
and 18).
The following tablereportsthe spreadingratevaluesobtainedwith theturbulencemodelsand
the recommendedexperimentalvalue.It is to be notedthattheexperimentalvalue of 0.115showsan
uncertaintyof about+10%.
Mixing

layer

Spreading

Experiment

0.115

Launder-Sharma

k-e model

Wilcox

k-co model

Menter

SST model

Spalart-Allmaras
The

range

values.

of values

These

reported

effects

d. Sensitivity

0.099
0.068-0.143
0.100

model

0.109

for the k-co model

are shown

in detail

is due to the effects

in the following

turbulence

sensitivity

Figure 7.2 shows the spreading

rates predicted
N = vt/UlX = 10 -n. For each turbulence
model,

values

of the exponent

n within

the range

defined
within the grid domain -0.3
Allmaras
model for very large values
up to -5 <rl

=y/x<_4

forN=

The predictions
and k-co SST models

with

sensitivity
and

high freestream

and in figure

co

7.2.

study

cosity,

0.068

paragraphs

of low and

analyses

1) Freestream

strong

rate

small

< n < 0. A uniform

distribution

of 501 points

was

< 1"1= y/x < 0.2 in all these simulations,

except for the Spalartof freestream
eddy viscosity where the grid range was increased

of the spreading

rate, velocity

profile,

are quite insensitive

to freestream

turbulence.

values

values of eddy visdone with different

1.

to freestream

the far wake

of -13

with different
freestream
a set of simulations
was

turbulence.

of freestream

predictions

Large

values

co give a spreading

presented

by Menter

(ref.

and eddy

viscosity

profile

The k-co model

of freestream
rate of 0.143.

co give
These

12) in his discussion

with

predictions

show

a spreading
results

over

the k-e
rate

a
of

are consistent

the sensitivity

of

the k-co model to freestream

conditions.
The one-equation
model of Spalart-Allmaras
shows no sensitivity to freestream
turbulence
and gives a spreading
rate of 0.109 with N < 10 -5. For very large
freestream
N=I.

eddy

viscosity,

2) Grid
A grid
needed
with

resolution

sensitivity

to capture
50,

100,

t50,

it predicts

sensitivity
study

an accurate
300,

larger

500,

up to 1.61 with a nondimensional

eddy viscosity

of

study

for each
solution
and

values

model

was

and is shown

1000

points

with

24

made

to provide

in figure
a uniform

an estimate

7.3. The computations

distribution

inside

of the

resolution

were performed
the grid

domain

-0.3 _<r1_<0.2. The freestreamvaluesof the turbulentkinetic energy and eddy viscosity were set
equalto K 1 = 10 -6 and N 1 - 10 -8, respectively. The results are presented based on the percent error of
spreading
rate with respect
to the solution
obtained
using 1000 grid points. In general,
the results
show that errors can be controlled
within less than 3% if 100 grid points are used in the calculations
of the mixing

layer

grid resolution,

with the k- and the k-co SST models.

while

3) Initial

the k-c0 model

conditions

The models

shows

sensitivity

all appeared

a much

The S-A model

larger

sensitivity

shows

a smaller

sensitivity

to

to grid resolution.

study

to show

zero sensitivity

to different

initial

profiles,

as expected

from

the usual concept of self-similar

profiles. For example, the following
table shows the spreading
rates
with the same six significant
figures computed
with two different initial velocity profiles, a step function and a hyperbolic

tangent

defined

as U/U I = 0.5 + 0.5 tanh(10q).

Initialkmodel

k-_.

k-co

SST

S-A

tanh/step

0.098594

0.067572

0.100237

0.108637

function

4) Code

sensitivity

A study

of code

obtained
fications,

profile

independent
such
as

and Mach

number

independence

sensitivity

is needed

of the numerical
full Navier-Stokes

to demonstrate

code--even
methods,

current

Figure

turbulence

significant

differences

they all show

500 grid

an accurate

solution

similar

points.

The

7.4 shows

model

(S-A).

were

found

predictions.

the

comparison

The mean

of the results

velocity

profiles

in the predictions
The

incompressible

can

be

codes. Five codes with three different

numerical
equilibrium
method
(SSE), one incompressible

method (INS) (ref. 20), and one compressible

Navier-Stokes
methods
solutions
for each code were established
following
the guidelines

section.

Allmaras

that

among codes involving

different physical
simpliboundary
layer
approximations,
self-similar

approaches,
and incompressible
versus compressible
methods
were used in this study: three self-similar
Navier-Stokes
The "standard"

study

obtained

self-similar

the

are similar

one-equation

show

solution

similar

obtained

Spalart-

and code-independent.

with the five different

solutions

Navier-Stokes

using

(CNS) (ref. 21).

illustrated
in the

results
with

the

numerical
with

No
codes;

102, 200,

INS2D

code

and
used

101 grid points.

The effect
Navier-Stokes
compressible

of the freestream

codes.

The

Mach

comparison

Navier-Stokes

code,

equilibrium

approximations

differences.

One SSE predicted

number

in figure

is presented

7.4 of the velocity

the incompressible

for the incompressible

a spreading

for the benefit

profiles

Navier-Stokes
mixing

rate about

layer

using

code,

of users
using

of compressible

M 1 - 0.5 with the

and the three

M 1 = 0 showed

2% less than the consensus

self-similar

no significant

of the other

pressible
methods,
and the CNS prediction
was only 3.7% low even at M 1 - 0.5. These
are much smaller
than the 10% observed
in the experimental
data. Simulations
based
Stokes codes should make sure to determine
the self-similar
state independent
of initial
and should
rate becomes

define

the virtual

a linear

function

origin

of the mixing

of the streamwise

layer
distance.

25

based

on the flow region

where

incom-

differences
on Navierconditions,

the spreading

0.16

....

"

"

0.08
0.00

Q'"
--0.08

.,_:_-_/
::/._-o

SpreoOing

---

k--e

_r_-
--0.!6

......
......

z_

/
/ ,

0.0

SST
S-A

0.2

"I

o.0_66-?._,7

model
mode!

0.099

_o_

o/

--0.2

rote_

model

0.100
0.lOg

O. _-

0.6

-4
I

0.8

I .0

U/U

Figure

7.1. Comparison

of velocity

profiles

for mixing

layer

0.20

flow.

t
-- _-

-e_l
-

_-

L;epmon
k--([

mode

ond

k-_
SST
S--A

_o_el
model
model

Loufer
l

0.15

k/U_

=10

1 O"

wjU_x

_,-

_dr

__ ......

m. "

-t

/
z

.......

_ .......

-_/
/
/

0.05

,
--

,
L ,
-- i 0

k ,
--8

, ,
--6

,
--4

J
--2

,
0

Figure

7.2. Comparison
turbulence

of spreading
rate sensitivity
for mixing layer flow.

2o!
1

,....

to freestream

----..o--

k--E
/--w
SST

model
model
model

S--A

model

-!
_

10,

200

4-00
number

Figure

600
of

800

7.3. Comparison
of spreading
rate sensitivity
resolution
for mixing layer flow.

26

1000

points

to grid

[.
/

N_lCode

SSE

'-//_0
,._

__
C

---....

Spreading
(Bardmal

rate

.q

0.109

1NS(Bardma)

0.109

SSE
CNS

0.107
O. 105

(Coaklcy)
[Huang')

i
J

0.2

0.4

0.6

0.8

1.0

U/UI

Figure

7.4. Comparison
of code sensitivity
in the prediction
velocity profile for mixing layer flow.

27

of the

Section

8. Plane

Jet

a. Introduction
Turbulent

plane

jets or nearly

22, and 23). Experimental

high Reynolds
mental

data

developed

U/UI=[1.000,

0.646,
0.256,
the mean

located

shows

a spreading
free-shear

significant

study,

data of Bradbury
[0.000,
0.088,
0.150,

where

of transition

an asymptotic

and initial
rate

flows,

parameters

in engineering

value

between

and

at

The experi-

0.11

profile

3, 4,

profile

effects.

0.10

velocity

model

(refs.

nondimensional

development

the mean

for turbulence

design

for a fully

and its spreading

predictions.

data

In the present

coordinate

interest

that the jet develops

to the other

the most

b. Experimental

y/x=

(ref. 23)

jet. Similar

rate are considered

show

and far downstream

of Bradbury
plane

imental

results

numbers

plane jets are of great

the predictions

(ref. 23), whose

of the turbulence
nondimensional

models
mean

are compared

velocity

profile

against

the exper-

is

0.009,
0.096,
0.158,

0.018,
0.097,
0.167,

0.026,
0.106,
0.169,

0.034,
0.110,
0.176,

0.035,
0.110,
0.185,

0.044,
0.114,
0.188,

0.053,
0.123,
0.194,

0.061,
0.123,
0.202,

0.062,
0.132,
0.211,

0.070,
0.141,
0.220]

0.079,
0.142,

0.996,
0.553,
0.210,

0.983,
0.588,
0.168,

0.962,
0.529,
0.174,

0.916,
0.500,
0.131,

0.933,
0.468,
0.099,

0.898,
0.471,
0.084,

0.856,
0.400,
0.072,

0.784,
0.414,
0.050,

0.809,
0.358,
0.033,

0.758,
0.305,
0.020]

0.703,
0.284,

speed

direction
at the virtual

of one freestream

is U 1 and the other

and y is the transversal

coordinate

origin

flow.

of the self-similar

freestream

direction,

is at rest, x is the streamwise

and

the center

of coordinates

is

c. Results
The validation
and

spreading

these

calculations

the appendix,
mensional

metrics

of the turbulence

rate of the plane

were

performed

achieving

turbulent

fast

kinetic

models

jet are shown

below.

by solving

convergence
energy

for the prediction

Consistent

the nondimensional

to machine

was also defined

within

The

as K 1 = kl/Ul

within

were

predictions

increased
with

the

rl < 0.7 for freestream

model

eddy viscosities

N<

also used

described

value

2 = 10 -6 (nearly

profile

simulations,

zero)

for the two-

The number

of grid points

with low freestream

300

grid

points

W-

10 -2, and used 600 grid points

and

transverse
freestream

its origin

at the virtual

coordinate
direction,
was set at rest.

origin

of the self-similar

U 1 was defined

flow, the coordinate

as the mean

28

speed

distributed

_max

= 1.4 for

y was defined

of one freestream,

and

mx/U 1 < 1.

uniformly

N = 0.1, and I000 grid points and rlmax = 6 for N = 1, respectively.

The initial mean velocity
was defined as U/U 1 = exp(-811"1 2 ). The coordinate
x was defined as the streamwise
coordinate
tion with

in

of the nondi-

turbulent
viscosity
was varied within
was defined inside the grid domain

to 300 and rl < 0.7 for the k-c0 model

Spalart-Allmaras

velocity

layer

equations

freestream

0 < 11 = y/x < 0.35 for the k-e, the k-co SST, and the k-co models.

the domain
The

similarity

accuracy.

equation turbulence
models, and the nondimensional
freestream
the range 10 -15 < N < 1. A uniform
distribution
of 150 points

of the mean

with the mixing

profile
direcas the

and the other

1) Velocity
Figure

profiles

8.1 shows

experimental

data

the comparison

of Bradbury

of the mean

velocity

(ref. 23). The k-co SST model

profiles
profile

U/U 1 against
gives

the experimental
data and is also insensitive
to low freestream
values
model is very similar, except near the freestream
at rest, and is insensitive
_. The

small

difference

their diffusion
model
thickness;
the results
larger

freestream

practice,

these

between

of these

two models

excellent

and the

agreement

of co. The profile

to low freestream

with

of the k-e
values of

is due to the different

value

of

constant,
oe. The Spalart-Allmaras
model overpredicts
the mean velocity profile
are insensitive
to freestream
eddy viscosity
for N =- vt/Ulx < 10 -3. Results with

eddy
errors

the predictions

rl -y/x

viscosities

give much

can be controlled

(_t/_tl < 10 -3 Re_). The

the other underpredicts

larger

by limiting

overpredictions
the values

and are not shown

of the eddy

viscosity

in figure

8.1. In

in the freestream,

results of the k-co model show two predictions;

one largely
the thickness
of the mean velocity profile, corresponding

overpredicts
to low and

and
high

freestream
W- o3x/U 1 values, (W < 10 -4 and W >_ 103), respectively.
This model gives a set of intermediate solutions
(not shown in fig. 8.1) depending
on the values of freestream
03, (10 .-4 _<W _< 103).
2) Spreading

between

rates

The spreading
the centerline

U/U 1 = 1/2, respectively.

been

rate S is defined as the distance

and the freestream,
where the
This

definition

of spreading

rate is one

of several

formulations

that have

proposed.
The

following

the recommended

table gives

experimental

0.11 and is given

the spreading
value.

only as reference

The range

rate values

Launder-Sharma
Wilcox

k-co model

Menter

SST model

Spalart-Allmaras

Spalart-Allmaras
to the effects

with the turbulence

values

is reported

models

between

and

0.10 and

values.
Spreading

give close

rate

0.100-0.110

Experiment

k-e and the k-co SST models

obtained

of experimental

Plane jet

The

of the nondimensional
coordinate
rl -= y/x
nondimensional
velocity
is U/U 1 = 1 and

k-8 model

0.108
0.092-0.132
0.112

model

0.143

prediction

of the experimental

spreading

rate,

while

the

model overpredicts
the spreading
rate. The k-co model predicts a range of values due
of low and high freestream
03. These effects are further
discussed
in the following

paragraphs.

29

d. Sensitivity

analyses

1) Freestream
Figure
values
-15

8.2 shows

of eddy

flows.

shows

rate of 0.143;
of values.

A grid
500,

of the

spr011ding

10 -n, and

shows

a strong

rate of 0.092,
are predicted

small values
in between

no sensitivity

to freestream

even

larger

values

sensitivity

the values

with

of the exponent
turbulence

different
n was

is similar

03. Large

varied

within

in all free-shear

to low freestream

to freestream

freestream

values

values

of e, or 03,

of freestream

co

of freestream
co give a spreading
rate of 0.132, and interthese two limits. The one-equation
model
of Spalartturbulence

are predicted

with greater

with N < 10 -5 and gives

nondimensional

a larger

freestream

spreading

eddy

viscosity

study
study

for each

model

solution.

The

1000 points

predicted

are insensitive

sensitivity

an accurate

with a uniform

rlmax = 0.35 was defined

rates

rate to freestream

with the k-e, and k-03 SST models

to capture

and

study

of the spreading

2) Grid sensitivity

needed

sensitivity

The k-03 model

give a spreading
mediate
values

sensitivity

N =- vt/Ulx=

sensitivity

The results

Allmaras

the

viscosity,

< n < 0. The

respectively.

turbulence

was

done

to provide

computations

distribution

inside

for the k-e., SST, and k-03 models,

were

an estimation

performed

the grid domain

and the value

with

of the resolution
50, 100,

150,

0 < ri < rlmax. The

300,

value

of qmax = 0.7 was defined

of
for

the S-A model. The freestream

values of the turbulent
kinetic energy
and eddy viscosity
were set
equal to K 1 = 10 -6 and N 1 = 10 -8, respectively.
The results are presented
based on the percent
error
with
model

respect

to the

solutions

solution

obtained

to the number

using

1000

of grid points

grid points.

used in these

Figure

calculations.

8.3 shows

the sensitivity

In general,

errors

trolled within less than 2% with 50 through

1000 grid points used in the calculations,
k-03 model when the number of grid points inside the boundary
layer is less than 200.
3) Initial

conditions

The predictions

sensitivity

of the different

except

study
models

showed

3O

zero sensitivity

to the initial

profiles.

of the

can be confor the

0.30

0.20
II

0.10

_-I

__

0.00
0.0

0.2

0.4

0.6

0.8

1.0

U/U,
Figure

8.1. Comparison

of velocity

profiles

for plane jet flow.

0.18
--

Bradbury
k--(
model

0
O. 1 6

k-_
SST
S--A

o.14_

T_

_'

model
model
model

-_
I

lO-_

L/tl
O.

- - /'
_-_)

1 2
.........

_:: : - - : _..._

......

O- .....

-o- .....

7___--

o. lo

'

1
!

0.08
--15

-i0

--5

Figure

8.2. Comparison
turbulence

of spreading
rate sensitivity
for plane jet flow.

to freestream

.e-----z------e----

k--e
model
/_--_) model
SST
model

----49--

$--A

model

ul

&
_"__
o

eo i

--2r--4

t
0

200

400

600

number

Figure

of

8.3. Comparison
of spreading
for plane jet flow.

31

_
800

I
i000

points

rate sensitivity

to grid resolution

Section

9. Round

Jet

a. Introduction

Turbulent

round

or axisymmetric

and 24). At high Reynolds

the round
nanski
oped

jet develops

and

Fiedler

round

considered

data

equilibrium

to other

free-shear

rate value

0.086

velocity

in turbulence

flows

(refs.

and 0.095

profile

model

3, 4, 22,

development

of the experimental

between

flows, the mean

parameters

aerospace
and initial

state. The analysis

a spreading

significant

in many

of transition

effects,

data

of Wyg-

for a fully devel-

and its spreading

rate are

predictions.

data

In the present

y/x=

an asymptotic

jet. Similar

jets are encountered

and far downstream

(ref. 24) shows

to be the most

b. Experimental

mental

numbers

study,

of Wygnanski

[0.000,
0.032,
0.068,
0.107,
0.149,
0.210]

U/UI=[1.O00,

0.900.
0.641.
0.375.
0.145.
0.027]

the predictions
and Fiedler

of the turbulence
models
(ref. 24) whose nondimensional

are compared
mean

velocity

profile

with the experiis

0.004,
0.036,
0.068,
0.114,
0.154,

0.005,
0.038,
0.076,
0.115,
0.160,

0.007,
0.042,
0.078,
0.118,
0.161,

0.009,
0.042,
0.080,
0.125,
0.163,

0.015,
0.049,
0.081,
0.128,
0.173,

0.017,
0.049,
0.087,
0.131,
0.176,

0.017,
0.054,
0.088,
0.134,
0.182,

0.019,
0.058,
0.096,
0.137,
0.185,

0.029,
0.060,
0.097,
0.137,
0.192,

0.029,
0.064,
0.099,
0.144,
0.192,

0.030,
0.066,
0.102,
0.145,
0.197,

0.995,
0.850,
0.623,
0.332,
0.136,

0.991,
0.855,
0.559,
0.314,
0.109,

0.989,
0.832,
0.555,
0.300,
0.114,

0.984,
0.809,
0.550,
0.268,
0.109,

0.977,
0.791,
0.532,
0.255,
0.082,

0.973,
0.775,
0.500,
0.241,
0.073,

0.961,
0.739,
0.486,
0.214,
0.073,

0.973,
0.727,
0.432,
0.218,
0.055,

0.918,
0.700,
0.439,
0.205,
0.050,

0.909,
0.700,
0.432,
0.182,
0.041,

0.893,
0.636,
0.405,
0.182,
0.045,

where the centerline

speed of the jet is U 1 and the freestream
is at rest, x is the streamwise
coordinate
direction
and v is the transversal
coordinate
direction,
and the center of coordinates
is located at the
virtual

origin

of the self-similar

flow.

c. Results

The
spreading

validation

of the turbulence

rate of the round

calculations

were

appendix,

performed

achieving

models

jet is shown

below.

by solving

fast convergence

the

for the prediction

Consistent

nondimensional

to machine

accuracy.

similarity
A uniform

was defined in the domain 0 < rl - y/x < 7/10. The freestream
kinetic energy was also defined as K 1 ---kl/U12 = 10 -6 (nearly
models,

and the nondimensional

the models.

The

Spalart-Allmaras
600 grid
mean

points

velocity

streamwise

number

freestream

of grid points

models

for large

turbulent

values

profile
coordinate

was

defined

direction

its origin

eddy

1000 grid

as U/U 1 = 1.0with

viscosity

of freestream

32

virtual

from

were increased

viscosity

profile

these

described

in the

of 300 grid points

10-

15<

turbulent
turbulence

_ N <_ 1 for all

with the k-o3 and the

due to freestream

sensitivity,

and llmax = 6 for N = 1. The

The

coordinate

origin

of the

and

simulations,

equations
distribution

was varied

points

tanh(10q).
at the

velocity

layer

value of the nondimensional

zero) for the two-equation

and the radial jet distance

and rlmax = 1.4 for N = 0.1 and

of the mean

with the mixing

x was

defined

self-similar

initial
as the

flow,

the

coordinate

y was defined

U 1 was defined

as the transverse

as the centerline

1) Velocity

profiles

Figure

shows

9.1

mean

speed

the comparison

q = y/x and the experimental

data

coordinate

direction

with its origin

of the jet, and the freestream

of the predictions

of Wygnanski

of mean

and Fiedler

at the center

was set at rest.

velocity

profile

(ref. 4). All the models

thickness
of the experimental
mean velocity profile. This classical
models
that have been fine-tuned
with empirical
data of mixing

of the jet,

U/U 1 against
overpredict

the

anomaly
is well known in these
layer, plane jet, and/or far wake

experiments.
The
insensitive
tions

results

of the k-c

of these two models

fusion

and k-o3 SST models

to low freestream

model

constant,

values

are closer

of e or co, respectively.

near the edge of the freestream

eye. The Spalart-Allmaras

to the experimental

The small

model

gives

a considerably

in figure

9.1. In practice,

these

large

values

and

between

is also due to the different

the mean velocity' profile thickness,

and the results are insensitive
N- vt/Ulx < 10-L Results with larger freestream
eddy viscosities
and are not shown

data

difference

value

larger

are also

the predicof their

dif-

overprediction

of

to freestream
eddy viscosity
for
give much larger overpredictions

of dimensionless

eddy

viscosity

are

much larger than the molecular

viscosity, and the errors can be controlled
by limiting the eddy viscosity in the freestream,
(_t/_l.l < 10 -3 Re_,). The results of the k-co model show two overpredictions
of the
thickness

of the mean

velocity

profile,

corresponding

to low and high freestream

(W _< 10 -4 and W > 104), and a set of intermediate

values

of freestream

where
round
provide

rate is defined

speed

jet and is widely

The
recommended

is half

that have been

information

its centerline

proposed.

This definition

modeling.

However,

reports
value.

the spreading
The range

overpredict

with the

values

Spreading

k-e model

k-c0 model

Menter

SST model

the spreading

S = y/(x-

Xo),

rate is one of several

of the thickness

of the

and it does not

turbulence

is between

0.086

models

and

the

and 0.095.

rate

0.086-0.095

Wilcox

of low and high freestream

an estimate

it is only one parameter

rate obtained

of experimental

jet

Launder-Sharma

the effects

on the

profile.

Experiment

Spalart-Allmaras

jet radius,

of spreading

rate provides

the shape of the velocity

table

experimental

value.

of the nondimensional

The spreading

used in turbulence

about

following

as the value

Round

All models

in fig. 9.1) depending

rate

spreading

the mean

formulations

(not shown

co, (10 --4 < W < 104).

2) Spreading
The

solutions

W = coX/UlX values,

0.120
0.169-0.356
0.127

model

0.253

rate. The range

co values.

33

of values

reported

for the k-co model

is due to

d. Sensitivity

analyses

1) Freestream

ues

turbulence

sensitivity

study

Figure 9.2 shows the sensitivity

of the spreading
rates predicted
with different freestream
valof eddy viscosity,
N-vtlUlx=
10 -n, and the values
of the exponent
n was varied
within

-15 < n < 0. A uniform

distribution
of 300 grid points inside the domain 0 < T! - y/x < 0.7 was
all these simulations,
except for the models with large freestream
sensitivity
(see fig. 9.1).
For a given
all freeshear
flows.

model, the sensitivity

of the spreading
rate to freestream
The results with the k-E and k-co SST models are insensitive

ues of E or co, respectively.

freestream

co give

The

k-co model

a spreading

shows

rate of 0.068,

a strong

small

values

sensitivity

turbulence
is similar in
to low freestream
val-

to freestream

of freestream

set in

co give

co, large

values

a spreading

of

rate of

0.143, and intermediate

values are predicted
in between these two limits. The one-equation
model of
Spalart-Allmaras
shows no sensitivity
to freestream
turbulence
with N < 10 .5 and gives a spreading
rate

of 0.109,

but

N > 10-3; in other

2) Grid
Figure

9.3 shows

needed

150,

500,

where

rlmax

with

points

= 0.35

except

The results
used

Similar
initial

a grid sensitivity
an accurate

kinetic

with

very

large

freestream

eddy

viscosity

values,

study

1000

3) Initial

ferent

predicted

to capture

grid

study

for each

solution.

with

and eddy
based

In general,

viscosity

which

distribution

model

where

were set equal

on the percent

errors

model,

The computations

a uniform

for the Spalart-Allmaras

energy

are presented

1000 grid points.

grid points

are

gt/gl > 10 .3 Rexoo.

and

of the turbulent
tively.

values
when

sensitivity

resolution
300,

larger
words,

error

can be controlled

were
inside

an estimate

performed
the

rlmax = 0.70.

domain

with

of the
50, 100,

0 < _ < Tlmax,

The freestream

values

to K 1 = 10 -.6 and N 1 = 10 -8, respec-

with respect
within

provides

to the spreading

rate obtained

less than 2% with 50 through

1000

in the calculations.
conditions

sensitivity

study

to the results

of other

free-shear

flows,

profiles.

34

all the models

showed

zero

sensitivity

to dif-

0.40

_..

,...

\
\

'

"\'_

0,30

k--_

r_oCel

O.

",/_--_
_'S.,T

r'_odel
,"_odel

C.169--0.356
Spreccfimc
O. 127

......

\\

_...._.
x

S--_\model
o',

"_

-._

,
0.2

0.0

-_

Fiedle'_

""_" ..

\\

'
0.6

0.4

'

0.086--0095

_-._.

0.001

rote

0.253

Wygnah_ski
'_,

I 20

0.8

1.0

U/U,

Figure

9.1. Comparison

0.38

of velocity

profiles

for round

jet flow.

!
-o

Wygnanski
k--_
model

,5.
0
[]

k--_
SST
S--A

&

Fiedler

"-'_.-

Ar
_ _

-_-

,'

0.3_

_--

L
0.26

_ .....

0,20

_-

0.14

_:_ ._ _ __

model
model
model

,J

/
/

-c- .....

,
a - /Z- - -

e ......

0.08

::_: _ _ ____

i
--15

_=_= = _ = ___

--i0

__

_ _ =_

_ _ _ = =

--5

Figure

9.2. Comparison
turbulence

of spreading
rate sensitivity
for round jet flow.

'

'

6
'_

_q

cz

to freestream

i
I
e

}c--(

model

._

/c--_

model

SST

model

S--A

model

I
2

--,4

Figure

200

400
number

9.3. Comparison
of spreading
for round jet flow.

35

of

600
points

rate sensitivity

800

I000

to grid resolution

Section 10. Plane

Wake

a. Introduction
Turbulent
Similar

plane

to other

free-shear

nondimensional
modeling,
parameters.

profile

study,

(ref. 25) whose

experimental

aerodynamic

results

of transition

show

flows

that the

and initial

(figs. 3, 4, 22, 25, and 26).

wake

develops

an asymptotic

effects.

For turbulence

development

velocity

profile

are compared
against
is defined as

the experimental

data

of Fage

and

0.109,
0.373,
0.688,

0.118,
0.444,
0.730,

0.139,
0.452,
0.737,

0.214,
0.456,
0.778]

0.221,
0.489,

0.231,
0.556,

0.252,
0.568,

0.331,
0.609,

0.335,
0.669,

[0.923,
0.481,
0.052,

0.928,
0.444,
0.044,

0.916,
0.323,
0.030,

0.899,
0.290,
0.020,

0.725,
0.300,
0.014]

0.715,
0.243,

0.686,
0.150,

0.719,
0.145,

0.479,
0.107,

0.467,
0.075,

m and

(U-

of both freestreams

origin

wake.

1"1=[0.101,
0.344,
0.669,

11 = y(pUJDx)
speed

plane

the predictions

mean

(U-Umin)/(U_-Urnin)=

virtual

in many

data

In the present

mean

flows,

far downstream

for a fully developed

b. Experimental

where

are encountered

the mean velocity profile and its spreading

rate are also considered
the most significant
The analysis
of the experimental
data of Fage and Falkner (ref. 25) shows an spreading

rate of 0.365

Falkner

wakes

Umin), the centerline

is U.., x is the streamwise

of the self-similar

sity, and D is the total wake

Urnin)/(U._-

coordinate

flow and y is the transversal

drag or momentum

deficit

speed

direction

coordinate

(integrated

of the wake

is Umi n, the

with its origin

direction,

at the

P is the fluid den-

over the full wake

width).

c. Results

The validation
and
these

spreading

rate

calculations

the appendix,
mensional

metrics

were

by solving

the nondimensional

energy

distribution

was also defined

freestream

eddy

of 300 grid points

viscosity,

flows with lower

lowing exponential

freestream
function

the displacement

thickness

was defined

N > 0.01,
eddy
(U -

and

Urain)/(Uo_

similarity

accuracy.

The

velocity

layer

equations

freestream

value

= 10 -.6 (nearly

zero)

in the domain
points

Umin)

constraint

""

described

within

the

domain

in

of the nondifor the two-

flows

within
a uni-

with high

0 < 11 < 1 to simulate

mean velocity
profile was defined as the folexp(-2132). See reference
4 for details on using

in computing

36

0 < 11 < 2 to simulate

profile

simulations,

freestream
turbulent
viscosity
was varied
spreading
rate of the mean velocity profile,

The initial
-

of the mean

with the mixing

as K 1 -= kl/U12

150 grid

viscosity.

as an integral

Consistent

to machine

equation
turbulence
models, and the nondimensional
the domain
10 -15 < N < 1. Based on the predicted
form

for the prediction

below.

fast convergence

kinetic

models

are shown

performed

achieving

turbulent

of the turbulence

of the far wake

the mean

velocity

profile.

1) Velocity
Figure
sional

profiles

10.1 shows

coordinate

the mean

rl =y(pUJDx)

velocity

profiles

1/2 compared

(ref. 25). The k-e and k-o SST models

(U-Umin)/(Uoo-Umi

with

give thinner

the

n) versus

experimental

profiles

data

than experiment

the nondimen-

of Fage

and

Falkner

and are insensitive

to low

freestream
values of e or co, respectively.
The small difference
between the predictions
of these
models near the freestream
is due to the different value of their diffusion
model constant,
_.
Spalart-Allmaras
model gives best agreement
with the experimental
sitive to freestream
eddy viscosity
for N- vt/U_,x <_ 10 -4. Results
viscosities
trolled

give overpredictions

by limiting

the eddy

and are not shown

viscosity

in figure

in the freestream,

two
The

profile, and the results are insenwith much larger freestream

eddy

10.1; in practice,

these

errors

(I.tt/la I _< 10 -3 Re=).

The

results

can be conof the

k-co

model show two predictions;

one largely overpredicts
and the other underpredicts
the thickness
of the
mean velocity
profile, corresponding
to low and high freestream
W- cox/U= values, (W <__10-4 and
W>_ 103),
depending

respectively.
This
on the intermediate

2) Spreading
The

rl between

This

definition

widely

rate S is defined

the points

where

of spreading

used

in turbulence

The

following

recommended

as the difference

the nondimensional

rate

is one

speed

formulations

table

reports

the

value 0.365

spreading
of Fage

rate obtained
and Falkner

with

Spreading

been

zero,

respectively.

proposed

and

it is

k-E model

0.209-0.494

Menter

SST model

0.257-0.260

gives

a range

models

and the

0.255

k-co model

the k-co SST models

the turbulence

rate

Wilcox

Spalart-Allmaras

model

is one half and

that have

coordi-

0.365

Launder-Sharma

and

10.1)

(ref. 25).

Experiment

k-e

in fig.

modeling.

experimental

Spalart-Allmaras

shown

S = r10.5 - rl0 of the nondimensional

mean

of several

Far wake

The

(not

rate

spreading

nate

model gives a set of intermediate

solutions
values of freestream
co, (10 --4 _< W <_ 103).

model

underpredict

a value

k-co model

predicts

of values

prediction

of 43% and an overprediction

much

0.339
the experimental

closer

due to the effects

of 35%.

37

to the

spreading

experimental

rate by
spreading

of low and high freestream

30%,
rate

co, within

while
(7%).

the
The

an under-

d. Sensitivity

analyses

1) Freestream
Figure

turbulence

10.2

shows

sensitivity

the sensitivity

study

of the spreading

rates

predicted

with

different

freestream

values of eddy viscosity,

N=vt/UlX=
10 -n, and the
-10 < n < 0. The results with the k-E and k-o3 SSTmodels

values of the exponent

n was varied
within
are insensitive
to low freestream
values of

or _

sensitivity

respectively.

freestream

The

co give

k-co model

a spreading

shows

a strong

rate of 0.209,

small

values

to freestream

of freestream

o), large

c0 give

0.494, and intermediate

Spalart-Allmaras
shows

values are predicted

in between these two limits. The one-equation
no sensitivity
to freestream
turbulence
with N < 10 --4 and predicts

ing rate of 0.339,

gives best agreement

2) Grid

which
sensitivity

needed

300, 500,
of/'lmax

and 1000

were defined

within

used

to the number
less

k-co model

turbulent

for all the models.

points

to the solution

solutions

study

than
when

3) Initial

2%

obtained

kinetic

using

with

50 through

conditions

number

distribution

energy

Figure

The

study

of the different

models

Figure

grid

points

model of
a spread-

showed

38

in the

viscosity
solutions

to the num-

errors

calculations,

error

to the initial

with

of the model

can be controlled
except

400.

zero sensitivity

150,

Noo = 10 -8

on the percent
the sensitivity

In general,

100,

0 < 1"1< rlmax- The value

of the model
based

of the reso-

with 50,

eddy

10.3 shows

used

is about

an estimation

performed

the domain

are reported

in the calculations.

of grid points

sensitivity

inside

the sensitivity

results

1000 grid points.

1000

were

K._ = 10 -6, and freestream

10.3 shows

used

to provide

The computations

with a uniform

of grid points

is recommended

solution.

in the calculations.

the minimum

The predictions

of

with experiment.

for each model

an accurate

grid points

= 1, freestream

ber of grid
respect

to capture

of

rate

study

A grid sensitivity
lution

values

a spreading

profiles.

for

the

1 .0_.

__

_-E

.....
.....

0.8

).
r).g

irate

model
model

0.257--0,260
0,339

/
/

Falkner

0.6

./-_

0.2

0.365

_ C._- -

-{

0.209-0.494

Fage
&

0.255

model

SST
S--A

I_)r e

model

_-_

i""

t.

_-

_ _ _
-I,

0.0

....
C,.O

0.2

,
0.6

0.4-

(u-u
Figure

10.1. Comparison

0.8

o)/(u-u

of velocity

, , , , j
1.0
_ .2

o)

profiles

for plane

wake

0.60

0.50

Fage
&
k--c
model

"O
_-

k--_
SST
S--A

Falkner

model
model
model

'
- "_

10

utoo =

j
J

k. = _o-:
:n

flow.

,_

- -

_-

--

-_
-_-

- -

,'

,'

_J

0.40

_-

.....

-__'m-

--

r-

--

-3

-_

F
0.30
,_ .....

==========================
i

0.20
lO

--8

-6

-4

--2

Figure

10.2. Comparison
turbulence

of spreading
rate sensitivity
for plane wake flow.

to freestream

._

k--a_
SST

model
model

S--A

model

---e.--

/_--e

model

200

400
number

600
of

800

i000

points

Figure 10.3.Comparison of spreadingratesensitivity

to gridresolution
forplane wake How.

39

Section

11. Compressible

Mixing

Layer

turbulent

mixing

a. Introduction

The compressible
siderable

interest

in engineering

in the modeling
transition

of turbulence

and initial

analysis

condition

of the experimental

Research

Center

spreading

rate

was

the mixing

by Birch
the

flow for studying

layer develops

Mach

shape

aerodynamic

At high Reynolds
Shear

and Eggers

convective

and

in many

It is a fundamental

18 and 27-29).

effects,

increasing

curve."
The spreading
rate
parameters
for prediction.

Model

is present

data in the Free Turbulent

reported

with

b. Experimental

design.
(refs.

layer

(refs.

of the mean

these

equilibrium

velocity

shows

are

profile

are

The

Langley

a decrease

known

of

state.

held at NASA

28) and
data

effects

and far downstream

an asymptotic

18 and

and is of con-

compressibility

numbers

Flow conference

number;

flows

as the

of the

"Langley

considered

significant

data
validation

requires

a careful

comparison

of predictions

and

experimental

data

and

proper mathematical
and numerical
formulation
of the governing
equations.
Experimental
data reveal
that the initial development
of the mixing layer is sensitive
to initial conditions
of flow structures,
boundary
ment

conditions

nolds

numbers

Settles

show

a unique

results

and Dodson

with increasing
velocity,

different

freestream

or static

freestream

convective
velocities

at rest

freestream

M 1 -- [0.0,
S=

number
0.5,

0.0446,

Mc=[0,

with constant

0.247,

number,

speed,

0.477,

total

profiles

show

of the thickness

experimental

layer,

results

of

with

and with constant

with

by the "Langley

the low-speed
data"

2.5,

3.0,

3.5,

4.0,

4.5,

0.08275,

0.0632,

0.04052,

0.03798,

0.03563,

0.03395]

and the convective

Mach number
reduces
R 2 = 92/Pl and R 2 = [1 + 0.5(),1)M22]/[1

in terms

with

high

rate S as

2.0,

and total temperature,

0.854,

of

of the mixing

data are available

0.1106,

0.681,

the self-similar

in the database

0.115,

pressure

28 and 29).

flows.

Experimental

are represented

develop-

at high Rey-

rate (refs.

and tabulated

sides of the mixing

of different

temperature

1.5,

flows

M c. M e = (U 1 - U2)/(a I + a2) is defined

M 1 and spreading
1.0,

velocity

a decrease

a, in each freestream.

compilation

In the later

layer

spreading

on free-shear

rate show

effects.

mixing

are compiled

in the low- and high-speed

[0.115,

For perfect gases

numbers
are

results

of 45 experiments

Mach

The

and constant
Mach

and Elliot on mean

of spreading

U, and the sound

number

with an asymptotic

experimental

results

temperature.

profile

of Samimy

(ref. 29) in a study

experimental

and Reynolds

data for fully developed

self-similar

layer. These

the mean

speed

experimental

of the mixing

The

total

by flow geometry,

The experimental
profiles

layer

imposed

or far field, the available

1.000,

1.122,

the corresponding

1.225,

1.312,

5.0]
0.05182,

convective

1.386,

Mach

1.450]

to M c = (U 1 - U2)/(a 1 + a2) = (MIRM2)/(R + 1) where

+ 0.5(],1)Ml2]. The spreading
rate S is defined based on

the energy thickness

definition
of Birch and Eggers
y/x where (U- u2)Z/(u1
- U2) 2 is 9/10 and 1/10.

40

(ref. 28) as the difference

between

the values

of

The experimentalresultsof LiepmannandLaufer(ref. 19)for the mixing layerof an incompressiblefluid arereportedin section8 andarealsousedhereasreferencedata.
c. Results
The validation
and spreading

results

of the turbulence

rate of the compressible

models

mixing

layer

for the prediction

are shown

of the mean

below.

velocity

In the present

profile

simulations,

uniform distribution
of 501 grid points was defined within the domain -0.3 < rl = y/x < 0.2, where x is
the streamwise
coordinate
direction with its origin at the virtual origin of the self-similar
flow, and y is
the transverse
coordinate
direction.
The nondimensional
mean variables
U, M, a, K, and N denote
mean

velocity,

Mach

number,

sound

speed,

turbulent

kinetic

The subscripts
1 and 2 stand for the high- and low-speed
freestream
value of the nondimensional
turbulent
kinetic

energy,

(nearly zero) for the two-equation

turbulence
models. The
turbulent
viscosity
was varied between
10 -15 <_N-vt/(UlX
equal

to N=

Mach

10 .9 (also

number

speed

side

set

Figure

shows

(Udata

system

with constant

the range

zero,

U 2 = 0.

0 and
The

the

of Samimy

is used

thickness

comparison

between

U2)/(U 1 - U 2) of the mixing

in order

rl - y/x, the coordinate

vorticity

between
to

freestream
value of the nondimensional
) -< 1 for freestream
analysis
or was set
freestream

conditions.

6.5, and the freestream

initial

velocity

profile

The convective

velocity
ratio

and Elliot
to show

against

the point

layer

where

predictions

The numerical

a wide range
and 1.6. The
number.

nondimensional

turbulence

coordinate

at the midpoint.
sional

shape

as

predictions

plot. The

the nondimensional

nondimensional
speed

5co = (U 1 - U2)/(dU/drl)max.

with the four different

profiles
system

The results

show

degree

of the plot. These

turbulence

of agreement

plot coordinates

of the k-E and k-co SST models

of the experimental

agreement
with
velocity
profiles

some

profile.

The results

mean

models

and the

models

is 0.5, and 5co is the

show

the range

good

agreement

of the Spalart-Allmaras

data

are

data of Liepin this figure as

were

obtained

over

with M c = 0, 0.8,
convective
Mach

with the experimental

bound

coordi-

coordinate

The experimental

of convective
Mach numbers,
and figure 11.1 shows the predictions
vertical arrows indicate
the trend of the predictions
with increasing

All velocity
particular

defined

(rl - TI0.5)/Seo. This particular

in a simpler

where

of the

the standard

the coordinate

all the data

TI0.5 represents

of the mixing

the

layer using

shown with convective

Mach numbers
of M c = 0.51, 0.64, and 0.86. The experimental
mann and Laufer (ref. 19) for the incompressible
mixing layer, M c = 0, is also shown
a reference.

in the low-

was

profiles

11.1

profile

experimental

equal

for analysis

respectively.

+ tanh(lOrl)).

1) Velocity

nate

zero)

M c was varied
was

U/U 1 = 0.5(1.0

velocity

nearly

and eddy viscosity,

side of the freestreams,

respectively.
The
energy was defined as K 1 - kl/U12 = 10 .6

data

due to the

and collapse

all data

with the nondimenmodel

also show

good

the experimental
data, except for very high freestream
eddy viscosity.
The
of the k-co model show sensitivity
to low freestream
co values. The relative

mean
good

agreement
of prediction
with data is due to the use of 5co in the nondimensional
sional profiles
and spreading
rates show a much stronger
dependence
on Mach
seen in subsection
d.

41

plots. The dimennumber,

as will be

2) Spreading

rate

In the self-similar

region,

distance

and

systems,

(x,y), the spreading

the spreading

In nondimensional
transformed

The

most

1980-81

well-known

have

formulation

There

Conference

as the difference

are other

well-known

of them

is Bogdanoff's

simulations

the

spreading

to measure
thickness

vorticity

with

downstream

with Cartesian

coordinate

rate

the

of the mixing

S becomes

thickness

used by Birch
Turbulent

the values

definitions

linearly

5(x) is the thickness

on Complex

between

grows

layer.

simply

the

S = 5.

proposed

is the energy

AFOSR-HTTM-Stanford

30). One

(x, rl = y/x),

of the layer

been

layer

as dS/dx where

coordinates,
thickness

formulations

of the mixing

In Navier-Stokes

rate S is defined

self-similar

thickness
5 is defined
and 1/10.

and

rate is constant.

nondimensional

Several

the thickness

ofy/x

of the

and Eggers

Flows.

where

mixing

(U-

(ref.

layer.

3) in the

In this definition,
U2)2/(U1

to measure

the thickness

thickness,

6co = (Ul - U2)/(dU/drl)max,

of the mixing

the

- U2) 2 is 9/10

layers
and

(refs.

19

another

is

Roshko's
pitot thickness,
where 5pt is defined as the difference
between
the values of y/x where the
pitot pressure
is 0.95 and 0.05. Two other correlation
techniques
for comparing
spreading
rates were
also assessed
by Viegas and Rubesin (ref. 30). One is based on Roshko's
pitot thickness
and is defined
as

G(Mc)=S(UI+

Bogdanoff's
definitions
between

U2R)/((UI-U2)(I+R)

R2=p2/Pl,

and

the

other

one

is

the four model

Figure

predictions

11.2 shows

with

lines

that all models

convective

been

proposed

corrections

number.

of spreading

known

data

that present

by Sarkar

Mach

rates

are shown

the experimental

on

turbulence

numbers

et al., Zeman,

data.

predicted

with the four

Langley

data.

with symbols.

data on the decrease

This is a well-known

convective

Langley

and the experimental

and the experimental

It is well
increasing

and the experimental

the comparison

fail to predict

Mach

with

of present

models

fail to predict

shown

by experiment.

and Wilcox

The

most

between

the

dilitation-dissipation
are several factors
results

of the

dicted

by each

ing rate values

predicted

results

of the k-e,

k-co, and

SST

result

is

models.

the decrease

(see ref. 4). These

are

rate with increasing

turbulence

Recently,

mod-

results

significant

of spreading

weakness

turbulence

The predicted

of spreading

rate

corrections

have

model

corrections

add an additional

term for the dilitation-dissipation

of turbulent
kinetic energy and improve the predictions
rates in better agreement
with experiment
(ref. 4). Figures 11.3a, 11.3b, and 11.3c show
son

based

vorticity
thickness
and is defined
as Cco(Mc)= S(U l + U2)(U ! -U2).
These different
of spreading
rates were tested and showed no significant
differences
in the comparison

els with no compressibility

shown

where

k-co turbulence

models

of spreading
the compariincluding

the

model of Sarkar et al. and the compilation

of experimental
Langley
data. There
that have to be considered
in the analysis of these results. Figure
11.3a shows the

spreading
model

rate

nondimensionalized

for the incompressible

for the compressible

with
mixing

mixing

layer.

the corresponding

layer;

spreading

this is normally

All models

show

done

very

rate

value

in reporting

similar

predictions

pre-

spreadwith

decrease
of spreading
rate and overprediction
of the experimental
data with increasing
M c. One additional effect that has to be considered
is the sensitivity
of the predictions
of the k-co model with lower
freestream
values,

values

of co. Figure

and an intermediate

11.3b shows

value

the range

of freestream

of spreading

co. The agreement

rates

predicted

with experiment

with high
seems

and low

to improve

with lower values of freestream

co. However,
all the models predict very different values of spreading
rate for the incompressible
mixing layer, and this fact should be considered
to obtain an accurate
representation

of the predictions.

Figure

11.3c

shows

42

the

comparison

of the

actual

predictions

of

spreadingratesandthe compilationof experimentalLangleydata.The spreadingratespredictedwith

the k-e and SST k-co models show an intermediate
value of the experimental
data range and much
less

sensitivity

to M c than

the

experimental

data.

The

predictions

freestream

sensitivity

in the presence

of low M c, and the predictions

freestream

sensitivity

in the presence

of high M c.

d. Sensitivity

Figure
eddy

of the k-co model

show

better

show

strong

agreement

and less

analyses

1) Freestream

turbulence

The

of the model

sensitivity

11.4 shows
viscosity.

the variation

sensitivity

study

predictions

to their

of the spreading

The simulations

are based

rates

freestream

obtained

on the conditions

boundary

values

with different

is shown

freestream

of the experiment

below.

values

of Samimy

of the

and Elliot

(refs. 4, 29, and 31) with Me = 0.86, freestream

Mach numbers of 3.07 and 0.46, freestream
density
ratio of 0.36, and located at x = 210 mm downstream
of the initial mixing zone. The simulations
use a
uniform

distribution

of 300 grid points

the nondimensiona/turbulent
The spreading
turbulence.
those

Both

reported

predicts

rates predicted

models

underpredict

in figure

an spreading

freestream

turbulence.

freestream

eddy

within

viscosity

the experimental

of 0.063-0.064.

Small

viscosity

fleestream

values

-1/10

< r I < 1/10. The freestream

with -13

The

an spreading

k-co model

eddy viscosity

predict

are quite

data S = 0.105,

predicts

a spreading

values

insensitive

results

to freestream

that are consistent

rate of 0.062,
show

predict

a spreading

rate of 0.042,

and intermediate

values

rate of 0.154,

with

and the SST model

predictions

values

of

_<n _<0.

with the k-_ and SST models

11.1. The k-_ model

rate

the domain

N = 10 -n varied

a strong

sensitivity

to
large

are pre-

dicted in between
these two limits. The one-equation
model of Spalart-Allmaras
predicts
a spreading
rate of 0.059 with N _< 10 -3 and increases only with very large values of fleestream
turbulence,
up to a
value of 0.136

with a nondimensional

The sensitivity
mixing

layer

incompressible

eddy viscosity

of the turbulence

of compressible

fluids

models

of N = 1.

to fleestream

is consistent

turbulence

with the results

on the spreading

obtained

for flee-shear

rate of the
flows

of

fluids.

2) Grid sensitivity
A grid sensitivity

study

study

for each model

is recommended

to provide

an estimation

of the resolu-

tion needed
to capture
an accurate
solution.
The computations
were performed
with 50, 100, 150,
300, 500, and 1000 grid points with a uniform distribution
inside the domain -1/10 < rl < 1/10. The
simulations

are based

Figure

11.5 shows

calculations.
cent errors

on the conditions
the sensitivity

The predictions
with respect

obtained

to the solution

given

in the previous

of the model

solutions

with the four turbulence

obtained

using

subsection

c.

to the number
models

1000 grid points.

of grid points

are reported
In general,

based
errors

used

on the percan be con-

trolled within less than 2% if 100 grid points or more are used in the calculations
of the mixing
with the k-_, SST, and S-A models. The k-co model shows larger sensitivity
to grid resolution,
errors

greater

than 2% were obtained

in the simulations

43

with less than 500 grid points.

in the

layer
and

3) Initial

conditions

The results
lence
figures
from

models
11.1,

sensitivity

of mean

and
11.2,

velocity

different
and

a step function

profiles

initial

11.3. The

shape

and spreading

profiles

initial

defined

file, U/U 1 = 0.5 + 0.5 tanh(10rl)

study

showed

velocity

1.0

profiles

by the freestream

with different

were

---

k--,

.....
- ....

/cIW
SST

._

'rmmoCdel

modet
model

with each
with

varied

using

the

one of the turburesults

different

up to an smooth

shown

initial

hyperbolic

'

'

'
OJ
_@I
o_ _l_l]

'

'

.I'____

_--__

fi
q

/_i

-0.5

0.0

0.2

0.4
( U-

Figure

11.1.

Comparison

0.12

of

0.6
U2)/

velocity

( U,

0.8

1.0

U2)

profiles

for

compressible

mixing

0.10

--..................

"_-':'--'-:'_'"---._.-r

..........
.........

SST
model.
Z-'.T_.- "_.--A model
_----'-'--_-_ model

0.08

__--_

mo_ei

0.06

0.04

-_
o

oLomgley

Ooto

0.02
0

Ml

Figure

11.2. Comparison
of spreading
for compressible
mixing

44

rate of the mean

layer.

velocity

profile

layer.

in

profiles,

tangent

conditions.

M=0.6 4M =0.86

[]

"

obtained

differences

states,

freestream

L_epmonr_
& Loufer
O
M_=O
Somlmy
& Elliot

0.5

rates

no

pro-

".O

0.8
q
4
4

-_ o.6

Lcmgley
/C--_

---

0.4.[
0.2

--

dote

.....

moclel

S'ST
moc_e
(rtmQdelS
with

Sorkor

0.5

-]

correctiom)
,

C,0

mode

k--_

1 .0

1 .5

2.0

M=

Figure

11.3a. Comparison
of spreading
for compressible
mixing

rate of the mean

layer.

velocity

profile

1.O
-,.-_,

- _.,
',kk\

_,..

0.8

"

-.

._- o.6

"'.

\?.o.
""-

_-

_..

"l

_"_

W_>IO

0.4

Langley
G--co

---(models

0.2

dot

"_-.

model

with

Sorkor

_.
"'_-..

"o-_.

correction)

....

_..._

W_0.37

p--< 10 -_

_
i

0.0

0.5

1 .0

2.0

1 .5

M_

Figure

11.3b.

Comparison
of spreading
for compressible
mixing

rate of the mean

layer.

velocity

profile

0,!5

Lw.=o.37
.___._._._ ""

...

0.10

N
"-.

_..\>_

"ong!ey

-----

_-E
k--_

......

SS'T
(m

dot

mode:
model
model
ode:,s

Sorkcr

with

c ....

ctlon)

0.05

0.00

0.0

0.5

1 .0

1 .5

2.0

M=

Figure

11.3c.

Comparison
of spreading
for compressible
mixing

45

rate of the mean

layer.

velocity

profile

0.20

0.15

--

Samimy

-o-_- e -_-

]--_
k--_a
SST
S--A

/U

&

=10

Elliot

model
model
model
model
-e

_,'_?u2 = 1o

_--

,'

- _

0.10

//

0.05
__ _=__= _= _== __ _=_ ___ _=0.00

!
-10

--8

._ .......

_ .........

--6

--4

--2

Figure

4. Comparison
turbulence

of spreading
rate sensitivity
to freestream
for compressible
mixing layer.

200

number

Figure

4-00

5. Comparison
of spreading
for compressible
mixing

46

600
of

800

1000

points

rate sensitivity
layer.

to grid resolution

Part
The

turbulent

pressible
metric
RAE

D. Attached
flows

and Separated

included

fluid and a compressible

cylinder,

2822
Unless

of Huang

in this study
fluid at Mach

the shock/boundary

airfoil.

The present

it is otherwise

and Coakley,

Turbulent

layer

list of flows

stated,

Boundary

5, the adverse

is shown

12.

Incompressible

Boundary

Section

13.

Compressible

Flat Plate

Section

14.

Axisymmetric

Boundary

Section

15.

Transonic

Section

16.

RAE

Separation

gradient

bump,

47

of an incom-

flow on an axisym-

and a transonic

is the compressible

of this code

is given

flow on the

Navier

Stokes

in reference

Layer
Flow
Layer

with Adverse

Pressure

Flow over an Axisymmetric

airfoil

flat plate

below.

discussion

Section

2822

pressure

flow on an axisymmetric

A detailed

Flows

are the flow over an adiabatic

the code used in this study

TURCOM.

Layer

Gradient

"Bump"

32.

code

Section

12. Incompressible

a. Empirical
The
stream

correlations

of data

incompressible

"flat

of transition
Reynolds

empirical

correlations

number
laws

velocity

can

5 is the boundary

tive constant
asymptote
mulas

layer

to 0.55
been

profile

(_1

= lln

gradient)

boundary

parameter:

predictions

layer

the standard

will be compared

far

enough

down-

is the momentumwith the following

and skin friction.

displayed

+C+-- _(

by Coles'

mean

velocity

profile

(refs.

" (nY))2
sm
_

33

(12.1)

thickness.

in the logarithmic

have

pressure

be explicitly

u, t

where

(zero

by any one thickness

Re o -- UeO/V. The model

of mean

similarity

Layer

plate"

can be defined

thickness

The
and 34) as

Boundary

Coles

took the von KLrm_in constant

law as 5.0 and tabulated

at Re 0 > 5000.
used for higher

Coles'

tabulation

Reynolds

The von KLrm_in-Schoenherr

the "wake

_c as 0.41

parameter"

and the addi-

H, which

he chose

(ref. 34) only goes up to Re 0 = 29000

number,

assuming

H = constant

to

but his for-

= 0.55.

correlation
1

Cf

"-

17.08. .(loeRe
_
is a good
+2%,

approximation

and this correlation

and Re o are determined

for skin friction.

agrees

+25.111ouRe0,_

+6.012

The

experimental

uncertainty

Coles'

to within

for skin

this uncertainty

friction

(see table

is at least

12.1) when

12.1. Skin friction

coefficients

5000

10000

20000

50000

100000

cf(K-S)

0.003636

0.003007

0.002633

0.002326

0.001996

0.001790

cf(Coles)

0.003658

0.002980

0.002630

0.002342

0.002028

0.001829

1.404

1.358

1.324

1.297

1.269

1.251

Note,

however,

or so. Fortunately,

high Reynolds

numbers.

that several

authors

the skin friction

A 10% decrease

have

suggested

coefficient

in FI will result

in only

sensitive
1.43%

as Re 0 increases
to the exact

and

1.18%

value

increase

above
of FI at
in skin

coefficient

in table

Values of cfand H --- _5"/0 obtained

from Coles and KArm_in-Schoenherr
correlations
are listed
12.1. In addition,
values of u versus y+ for Re 0 = 10000 and 100000
are provided
in
12.2 and

and 100000,

that FI decreases

is not very

friction

tables

for Re o = 10000

law.

2000

H=-5*/O (Coles)

15000

o)

with that of Coles

by integrating
Table

Re o

(12.2)

respectively.

12.3, respectively.

48

Table12.2.u + vs.

y+ at Re o = 10,000--Coles'

velocity

law

y+

10

20

50

100

200

500

1000

2000

3000

u+

9.999

1.998

4.871

8.336

11.5

14.46

16.24

17.94

20.29

22.36

25.19

27.08

velocity

law

Table

12.3. u+ vs. y+ at Re 0 = 100,000---Coles'

y+

100

200

500

1000

2000

5000

10000

20000

30000

u+

16.23

17.92

20.16

21.85

23.56

25.92

28.03

30.93

32.77

Although

the computational

conditions

become

experiments

unimportant,

these

have

suggested

results

have

been

that for Re o < 5000

obtained

using

the influences

an inlet boundary

of inlet
layer

with

Re o = 1000.
b. Results
The
number,

results

herein

it is found

show

that the

the comparison

solution

of the skin friction

is not affected

at Re 0 = 10000.

by the inlet conditions

used

At this Reynolds
(see

fig.

12.4 for

effects of inlet conditions).

In this section, all calculations
were performed
for the following
conditions unless otherwise
stated. The ratio of turbulent
viscosity to molecular
viscosity
in the freestream,
_tt/l.t l, and the ratio of the square root of the freestream
turbulent
kinetic
velocity,
",]k/Uoo, were both taken to be 0.1% (see more details in paragraph
tional
grid

box

was defined

points

were

ratio determined

by Re L = 2.10 7 (Re 0 up to 2.5.104)

expanded

exponentially

by the choice

from

of the value

between

adequate

a boundary

layer

to use a boundary

interested

in validating

code

layer

the model

code

of y+ at the first grid point.

and a Navier-Stokes
to assess

be kept on the order

1) Grid

sensitivity

The computations
tion.

This

corresponds

layer

at Re o = 10000.

The value

ratio

of 0.02.

the model

implementations

code

as will be seen

performance.

of y+ at the first point

be used
in solu-

later. Therefore,

However,

in their own application

The

with an expansion

has shown that at least 60 points should

results. There is virtually no difference

least 100 grids in the streamwise

direction
with an expanding
mended
to achieve
accurate
solutions.
For use of compressible
should

to length

the wall to the top of the domain

was kept approximately

0.1. Grid sensitivity
study
inside the boundary
layer to obtain grid-independent
tions

with a height

energy
to the freestream
3 below). The computa-

for those

(Navier-Stokes)

it is

who

are

codes,

at

grid-spacing
ratio of 1.05 are recomcodes, the freestream
Mach number

of 0.1.
study
were performed

to approximately
The results

with 50, 100, 250, 500, and 1000 grid points

35, 65, 145, 265,

are reported

based

and 465 grid points

on the percent

errors

inside

with respect

in the y directhe boundary

to the solution

obtained
using 1000 grid points. Figure 12.1 shows the sensitivity
of the model
ber of grid points used in the calculations.
As can be expected,
the one-equation

solutions to the nummodel is less sensi-

tive to the grid refinement.

In general,

2% if 100 grid

are used

(corresponding

in the calculations

errors

can be controlled
to 60 grid points

49

within
inside

less than

the boundary

layer).

points

2) Sensitivity
The
500 points

to the distance

computations

were

in the y-direction.

solution

obtained

distance

of the first point

using

from

of y+of the first grid

ones.

In general,

ratio

at 0.1%

sensitivity

Figure

while

freestream

varying

the value

solutions

dard"

solutions

shows

the comparison

the region
ences

Re o < 5000

Mach

were

seen

that the

(ref.

obtained

by the three

the comparison

than

friction

the

difference

coefficients

Stokes Mach
layer solution

0.1 solution is about

and the Navier-Stokes

between

and 0. Figure

12). In general,

one

set by Menter

to have

codes

described

using

above.

the SST model.

inlet conditions
Re o > 5000.

Comparisons

obtained

with the k-e model

obtained

from

of inlet conditions

4%

are larger
while

the

for Moo = 0.5 are

slightly

incompressible

lower

boundary

0.15% while the difference

between
Mach 0.5 solution is about 1.7%.

k-E and

numbers;

k-0) models

Figure

12.4

runs.

based

in

Differon other

freestream

than

layer

boundary

not investigated.
those

solution

It can be

for Moo = 0.1. At

and

the incompressible

the

Navierboundary

at high Reynolds

models
predict
the skin friction

numbers,

the SST k-c0 and the Spalart-Allmaras

underpredict

50

using

shape factor H are shown

in figures
12.6
12.8 and 12.9 are reported based on the per-

to the von KLrm_in correlation

at low Reynolds

the

freestream-

the incompressible

were

100 and (H-H(Coles))/H(Coles)

100. All
within +2% for Re 0 > 104 and tend to overpredict

differences

sensitive

choose

conclusions.

cent errors,
(cf-ct(K-S))/cl(K-S)
Coles' correlation
for skin friction

overpredict

the

Differences

used in the three

skin friction
cf and
the results in figures

The

is very

should

The comparisons
of both
and 12.7, respectively.
In addition,

1% to 3% with respect

12.3 shows

layer code (ref. 35), an incompressible

Navier-Stokes
code (ref. 32). The "stanthe guidelines

of the results

the

-1,

study

1% after

similar

In this case the effects

skin

-3,

the criteria

following

are less

Re 0 = 10000,

within

solutions.

of n. It can be seen that the k-o) model

coefficients

shown.

sensitive

viscosity

turbulent

were -6,

of Moo = 0.1 and Moo = 0.5. The solution

is also

are the least

to molecular

of freestream

established

effects

numbers

to the

viscosity

freestream

and code-independence

5) Mach

code

to the

sensitive

upstream

is well within

and they all show

layer

using
to the

solutions

is very

models

1.4

"_ko,,/Uoo at the

due to the different

12.5 shows

of the model

The k-e model

and

with respect

value

the

by Menter

are also available

Figure

1.1,

by fixing

to the choice

of the results

number

0.7,
errors

accurate

were mainly

in skin friction

models

the sensitivity

were used in this study: a boundary

(INS2D)
(ref. 20), and a compressible

for each code

0.4,

of Y+l to be less than 0.3 to have

of n chosen

as discussed

of inlet conditions

codes
code

0.14,

on the percent

study

value of n to be less than -3. This choice

independent
results.

Three
Navier-Stokes

12.2 shows

performed

turbulence

4) Effects

based

the zero and the one-equation

to _ttoo/lAloo = l0 n. Values

of the model

to high

Y+l =0.014,

are reported

limit the value

were

while

with

the wall, Y+I

the wall used in the calculations.

sensitivity

computations

according

results

point

one should

3) Freestream

boundary

performed

The

Y+l = 0.014.

value

The

of the first point from

3.5%

and

5%,

respectively,

(Re o = 105).
models
the

skin

friction
of the von Kirrn_in correlation
at Re O = 2000. The SST k-o,) and the Spalart-Allmaras
tend to agree better than the other models with the values obtained
from Coles' correlation,
agree

almost
Plots

respectively.

exactly

with Coles'

of u + against
The

results

shape

factor

y+ for Reo=

for Re o > 5000.

10000

are also presented

models
and they

and

100000

are shown

in term of percent

errors

in figures

with

respect

12.10

and

to Coles'

12.11,
profiles,

shown in figures 12.12 and 12.13. All models show agreement

with the empirical
correlation.
In addition, the Re o = 100000 run provides information
on the predicted
values for K and C in the inner layer.
Use y+ between 50 and 3000 (y/8 = 0.1) to extract
C evaluated
from the benchmark
model solutions.

Table

values

12.4. Predictive

values

Model

_:

0.4096

5.525

k-0) 2-eqn

0.3831

4.011

0.3816

3.939

S-A 1-eqn

0.4073

5.060

B-L 0-eqn

0.4101

5.2034

2-eqn

12.4 shows

of _: and C

k-e 2-eqn

SSTk-co

of _; and C. Table

4
o
o

z-

2
g

G-

o''

-2

SST

-4<
"

number of points
Figure

12.1. Grid sensitivity

10 3

10 2

study--flat

51

plate boundary

layer.

values

of _: and

25

''

''

I''''

o-15

''

--e-_

k-_
k-o.}

_
_

SST
S-A I eqn

20

I''

....

''''

I''

''

.._

10

g
0

-5

l,

Figure

2 I,
0.25

_,

I ....
0.50

12.2. Sensitivity
.....

o
o
I-

I ....

I ....

I,
1.00

----------_.

q i I i i
1.25

i
.50

of the first point to the wall.

I ....

I ....

....

Re e = 10000
k-
---W--

I ....
0.75

to the distance

_D

.9

c_

S-A
koe

1 eqn

SST

o
01

"

_ ,,

-6

Figure
10 -2

....

I ....

-5

12.3.
}-

Freestream
'

-4

'

sensitivity
'

x I

-3

'

'

_ i

-2

-1

_
0

study.

' ' 'I

'

'

'

'

' '

SST model
---

--

Incomp. b-I code

Comp. N-S code

--'--"

(M=0.1)

6"

10"3
103

T I , I
104

z , , ,
105

Re O

Figure

12.4.

Effects

of inlet conditions

52

and code-independent

study.

10 -2

k- model
-Incomp. b-I code
-- -- Comp. N-S code (M=0.1)
__

....

Comp. N-S code (M=0.5)

o'-

10-3

10 3

104

105

Reo

Figure

12.5. Freestream
I

Mach
I

number
I

effects.
I

4.0
-

-_

K-S Correlation

-k

6B-L 0eqn
S-A 1eqn
k-_
SST

V
/_
O
1.0

, L ill

103

105

104
Reo

Figure

12.6. Skin friction

1.5

'

'

comparison.
'

' ' '''1

/k

1.4

-r-

1.3

o
-- O

1.2

1.1
10 3

Coles
law
k-E
B-L Oeqn
_
S-A leqn
z_
k-to
o
SST
n
l
I I I t Ill
10 4
Ree

Figure

12.'7. Shape factor comparison.

53

10 5

10

i,

Coles law
k-e

[]

B-L Oeqn
S-A leqn

o
A

SST
k-{o

--<>

_t

/"

__,

0
o"-

<>

[]

-I0

'

'

I I

10 3

104

105

Re O
Figure

12.8. Percentage
errors of sl<in friction with respect
von K_irm_in-Schoenherr
correlation.
10

-[]
_
A
0

SIAI
k-e)
SST

o
o

"-

5
A

(fJ
O

(,D

"1_.

Coles law
k-e
B-L 0eqn

to the

eqn

0
o
O
V

(o
'3i

-5

-10

TI[

10 3

10 4

105

Re o

Figure

12.9. Percentage
''

errors of shape factor with respect to Coles' velocity

30

'

25

Re e = 10 4
I.
Coles law
<>
[]

20

'''"1

....

''"1

k-_
B-L 0eqn
S-A leqn

....

_ _

''"l

'

' '

I
I

I
I

"

,0
5
0:

I
I

I
I

IIIlll
I IIl1,l

I
,

I
,

IIIIII
, 1,1Ill

10

I
J

I
I

IIIIII
I I I Illl

102

103

y+
Figure

12.10. Comparison

of the velocity

54

profilesIReo

= 104.

profile.

40

'

' ''""1

'

' ''""1

....

'"'1

.....

'"1

'

' '

Iltl

Re e = 105
3O

-_

-----

Coles

law

k-E

=_

B-L Oeqn
S-A

20
A

__/

1eqn

k-_

10
I

10

III111

102

II

103

104

y+

Figure

12.11. Comparison
2O

'

''

of the velocity

'''"1

'

'

profiles--Re

''''"1

'

''

o = 105.

'''"1

'

' '

10
O
+O

O
o_

0 I

<>

-5
+cO

Coles

i
+

Re e = 104

-10

-2O

, ,,,,,i

10

k-i_

[]

B-L 0eqn

S-A leqn
k-_

, ,,,,,I

law

ss?,,,,,, , ,
10 3

102
y+

Figure

12.12. Percentage
20

........

errors
'

.....

with respect
"1

'

to Coles'

' ''""1

'

velocity

' ''""1

'

profile--Reo

= 104.

' '

10
+cJ

0_

J
+

-_

Coles

--

+o

-20

B-L 0eqn

Re e = 105

-10

V
z&
I

I ,,,,,,I

law

k-E

, , , ,,_,,1

10

S-A 1 eqn
k-o)

, ,,,,,,I o, ,SST,,

102

103

, , ,

10 4

y-l-

Figure

12.13. Percentage

errors

with respect

55

to Coles'

velocity

profile--Reo

= 105.

Section

13. Compressible

a. Empirical
The
velocity

correlations
model

profile

eral consensus

Flat Plate

Flow

of data

predictions

will be compared

and skin friction

of a Mach

is that the Van Driest

I velocity

the following

5 boundary

layer

I transformation

the velocity profile in the inner layer,

the experimental
data of skin friction.
The Van Driest

with

flow over

correlations

an adiabatic

(ref. 36) is a good

and the Van Driest

transformation

empirical

of mean

surface.

fit to the experimental

II transformation

(ref. 37) offers

The gendata of

a good

fit to

is

Uc+=

lny++

(13.1)

where
Uc+=Uc/Ut

Uc=

y+=pwUty/I.t

,,/B [asin(

_: = 0.41

U,= %,fi-J w

- asin(A)]

ADU)

C=5

(13.2)

A = qw/'Cw

The general

formula

B = (2CpTw)/Pr

for Van Driest

D=

II skin friction

_--_+B

transformation

is
(13.3)

cfF c = f (FReoRe O)
where

Taw/ T _ - 1
(asinc_

+ asin _) 2

(Taw/T._)
!
_[

ERe =

+ (Tw/T**)

-2
(13.4)

2
(TawT**)

(Tw/T..)

(TawT**)
J[

l.t

and

(TawT**)

Taw = To.J1

_ (Tw/T

+ 0.89(_--_)M

+4(Tw/T
-

(Tw/Z

) ] 2 +4(Tw/T

2]

56

) [(TawT**

) - 1]

) [ (Taw/T

) -11

For a flow with freestreamtemperatureof 300K, table 13.1givesthe skin friction valuesat
selectedReynoldsnumbers:
Table13.1.Skin friction coefficients
Re o

5000

10000

20000

50000

cy

0.001077

0.0009285

0.0008087

0.0006830

The velocity
Van Driest
This

profile

Reynolds

profile

I transformation
family

numbers.

family

of Huang,

applied

to Coles'

profile

fit to boundary

layer

is a good

It is also displayed

tion of the compressible

boundary

layer

Bradshaw,

and Coakley

that includes
velocity

in this comparison

profiles

0.0006065

(ref. 38) is an extension

the sublayer

but should

100000

and the wake

for a wide

range

not be treated

of the
regions.

of Mach

as an exact

and
solu-

flow.

b. Results

values

Figure
13.1 shows the comparison
listed in table 13.1. The calculations

of the skin friction versus Reynolds

number
for the Re 0
were performed
with y+ at the first grid point at approxi-

mately
0.1 and 120 points
inside the boundary
layer. The freestream
temperature
was set at
Too = 300 K. All models
seem to follow the empirical
trend well. The k-e model, however,
shows a
significant
underprediction
sensitive
to low Reynolds

of skin friction coefficient.

This is due to the fact that the model is very
number effects and tends to overpredict
compressible
log law as shown in

figure l3.2, where comparisons

of the mean velocity and temperature
seen from the comparison
of the dimensionless
transformed
velocity
predict a large
for compressible

slope in the inner

flow reasonably

profiles
profiles,

are shown. As can be

the k-e model tends to

layer while all the other models seem to follow the law of the wall
well. The S-A model appears to give the best overall predictions
of

cf and u+.

57

%
--

--van
Driest lht II
k-e model
S-A 1-eqn model
z_
k-(o model

SSTk-_) model

10-1

-- -'_

L A z Lzl

o
@,

_.>

-_o

o
<>
o

,c,

-2o

-30 0 3

i ]
10
4

i
10
5

Ree
Figure

13.1. Comparison

'
3O

'

' '''"1

of the skin friction

'

' ' '''"1

....

coefficients.

''"1

'

' ' '''____

......

2O

-- -.....

k-comodel
SSTk-_ model

....

S-A l-eqn model

.-__
-"" "_'_

-{
--

.-_._

10

0
"

I I [I

'

_ _''_l

nnl

_ _ '''_'1

'

_ ''_J'l

ii

n[I

'

' ' _ '"_

1.0

0.8

0.6

0.4

0.2

0
10

10 2

10 3

10 4

y+

Figure

13.2. Comparison

of velocity

58

and temperature

profiles.

Section

14. Axisymmetric

a. Experimental

Boundary

data and description

The experiment
superimposed
wind tunnel

several
outer

the flow to remain

Experimental
locations.

method

performed

with Adverse

of numerical

by Driver

Pressure

data,

Since

attached

procedure

(ref. 39) consists

including

along

the tunnel

velocity

flow separation

was performed
boundary

of an axial

and Reynolds

was observed

and is recommended.

such

as a streamline.

walls

flow along

stress

a cylinder

the surface

of the cylinder

(slip condition).
x(mm)

x(mm)

The

with

-330.2,

152.4,
= [

condition)

streamline

-228.6,

-152.4,

requires

velocity

and other

outer

been

profiles

measured

boundary

defined

h(mm)

been

with

of an

integrated

to

point of view,
defined with

an outer

as a function

in

prediction

the specification
have

From a computational
duct with one boundary

distance

pressure

streamline

of the coordinate

is

= [-457.2,

h(mm)

(no-slip

recommended

have

a full Navier-Stokes

procedure

experimental

of the strong

profiles,

experimentally,

The solution

The

in the presence

obtain the stream function

and corresponding
outer streamline.
this method allows the flow to be treated as flow in an annular

distance

Gradient

adverse
pressure
gradient.
Boundary
layer suction was applied
through
slots on the
walls, and this mass flow removal (about 10% of the incoming
mass flow trough the tun-

nel) allowed
gradient.

Layer

228.6,

37.08,
67.89,

in which

the

outer

measured

experimental

304.8,

37.10,
70.40,

streamline

38.51,
71.20,
profile

velocity

-12.70,

50.8,

500.0,

600.0,

700.0,

42.66,
69.0,

between

profiles,

-76.2,

400.0,

48.53,
63.04,

x = -457.2

mm

and for x > 400 mm

55.44,
56.04,

101.6,
800.

61.88,
48.04,

and x = 304.8
it is estimated

mm

0]

65.08,
42.08]

is estimated

by matching

the

from
mea-

sured pressure
coefficients,
cp, along the cylindrical
surface. It should be noted that since the flow
reattached
around x --- 250 mm, the exact shape of the streamline
after x = 304.8 mm does not affect
the flow field in the region
The inlet profile
obtained
layer

The

interest--a

is a good match

by matching

(using

of current

the momentum

the same Navier-Stokes

following

solutions

y directions,

respectively.

in the radial

direction

fact that will be demonstrated

to an equilibrium
thickness

code

or a boundary

are obtained

The grid is uniform

boundary

layer

from a calculation

from

layer

a mesh

with Re o = 2760

of a spatially

evolving

and can be
boundary

code).
with

in the streamwise

later.

201 x 81 grid

direction

points

and exponentially

in the

x and

expanded

with the first y+ less than 1/10.

b. Results
Figure
same

outer

14.1

shows

boundary.

all, the SST model

comparisons

With the exception

gives

of the pressure
of the k-e model,

and

skin friction

all models

coefficients

predict

based

flow separation.

on the
Over-

the best performance.

Comparisons
of the velocity, turbulent kinetic energy, and shear
sured locations
are displayed
in figures 14.2, 14.3, and 14.4, respectively.

59

stresses at some specific meaAgain, the figures show that

the SST

model
are in between.

gives

There
conditions,
and

the best overall

are a few uncertainties

and effects

skin

friction

89 89--alI

of outflow

coefficients

being

uniform

using

are presented.

As can be seen

identical.

INS2D

Results

x/R o approximately
streamlines

associated

boundary
using

code

-1

using
and

for such a small

As mentioned

earlier,

the worst,

with this test case--grid

Figure

different

obtained

101 x81,
only results

using

boundary

grid system

in this region

in the first subset

of inlet

of pressure

301 x 161,

and

(89 x 89)

for the SST model

301 161 and 201 81 were

101 81 show a slight undershoot

1. This is because

effects

for the last grid system

of illustration,

the results

two models

the comparison

x81,

The results

and the other

resolution,

14.5 shows

grids--201

direction.

(ref. 20). For purposes

(see the plot of the outer

grid arrangement

four

from the figure,

obtained

between

the k-e model

conditions.

in the streamwise

were obtained
almost

performance,

of pressure

there

coefficient

for

expansion

of

is a rapid

in fig. 14.7),

and

hence

a uniform

may not be sufficient.

the inlet conditions

were obtained

by matching

experimental

Re o of a

Navier-Stokes
flat-plate
boundary
layer calculation.
All models show a good match of velocity, turbulent kinetic energy, and shear stress profiles at the first measured
location,
as shown in the first subset
of figures
ments

14.2,

14.3, and

in the inlet.

In figure

14.4, respectively.
14.6, effects

There

is some

uncertainty

of inlet conditions

were

associated

studied---in

with the measure-

this case,

results

compared
with those obtained
by matching
an Re 0 value being 10% less than the experimentally
gested value. As can be seen from the figure, the effects seem smaller than differences
between

were
sugmodel

predictions.
The last location
for measuring
up to this position,
the outer streamline
the outer
ments
face

boundary

and the wall

to satisfy

after x/R 0 = 4.4, and therefore

pressure

predictions
x/R o -4.4

coefficients

profiles was at x/R 0 = 4.4 and R 0 = 6.927 mm. From the inlet
was estimated
by allowing
the mass flow rate (or 0") between

against

a prescribed

the outer boundary

the

measured

value.

There

was estimated

ones

(fortunately,

Finally,
Figure

a test was
14.8

shows

_tt/la 1 = i and the other

conducted
the

to show

comparison

_tt/_l = 0.001

were

no profile

by matching
the

differences

seem small in this region). To ensure that the exact position

has little effect in the region of interest,
the current
results

obtained
by using an outer boundary
of fixed height
the results indeed show very little effect.

model.

given

of two

of freestream
calculations

at the inlet boundary

boundary
using

with q(k/U,_)

the predicted
between

of the outer boundary

were compared
with

after x/R o = 4.4. As can be seen

effects

measure-

from figure

conditions

the

after
ones
14.7,

for the k-co

k-co model--one

= 0.1%.

surmodel

with

As can be seen from

the results, as expected,

effects of freestream
conditions
do have some important
impact on the predictive behavior
of the k-co model for this flow as well. As can be seen from the comparison,
the skin
friction

predicted

and the pressure

with

gt/_t = 1 is slightly

coefficient

larger

is also overpredicted

than

that predicted

for the region

60

with

_/_t = 0.001

of x/R o between

0 and 5.

everywhere,

'

....

'

'

'

'

....

0.6

0.4

0.2

__j

.....k_.

'

'

'

'

'

'

'

t
"

'

'

'

'

A
0
0
0

\o___I
-

.S-"

"I"",

-5

14.1. Comparison

1.0

O.B

--

'

....

of surface

....

....

....

....

'

10

pressure

and skin friction

coefficients.

'__

__:-

x!R 0 = -6.531

5
x/R

Figure

",_',

x/R o = -2.177
o
o

'

0.6
0.4

O.2
z
0

1.0

0.8,

0.6
_

0.4
'''l

O.2

1.0

' I ....

I ....

....

I ....

[ ....

0.8

.t/R o = -3.266

0.6
_

0.4

0.2
....

0
o

1 ....
0,2

0,6

I ....

i ....

i ....

I '
o
o

__ x/R o = -0.181

o
o

0.4

I ....

oo

_L_'

]
]
I
o.8

,1

1 .o

0.2

0.4

0.6

....

0,8

I ,
1.o

U/U.

Figure

14.2a.

Comparison

of velocity

profiles.

61

See symbol

notations

in figure

14.1.

,.o:"

....

E....

, ....

" o ' '

"2

'

o.a

x/Ro = 0"181

....

....

.... o

....

I'

....

,,11 o

x./R o = 2.177

il

0.4

....

!iib

"_

0.2

oh

_-_'._
'

.,.

....

....

_ ....

_ ....

_ ....

_,

....

....

1.0

x/R o = 0.726

O.EI

0,4

0.2

' ....

' ....

"5

'

....

....

0.4

-_)7

i_,jd_

....

....

0.2

....

0.4

....

0.6

....

0.8

14.2b. Comparison

1.0

_.'
--

I ....
[ .: l

'

"

.....................

1.0

0.2

0.4

u/u_
Figure

....

t:!_

o.o
2:

x/R o = 3.266

li!_

oB :

....

' ....o ' ....

_,R0 = 1.451

i 31111?

LI

:" ....

1,0

i
0i

iI:l_
:!1o

0,6

....

'

'

'I

....

,
0.8 _.

of velocity

....

0.6

0.8

1.0

U/U**

....

profiles.

'

--

-"

See symbol notations

I ....

....

- o

xJR 0 = -6.531

....

I ....

in figure 14.1.

....

'

'

'

'__

x 0---2177

7:>

0.6

..
0.4

0.2

I.OE._'

'

'

....

F ....

....

x/R 0 = -4.717

0.8

x/R o = - 1.089

._

0.8

-o

=i

_]

0.2

)o
0

"'*

'

_-

;
0.8 _-

x./R 0 = -3.266

'

'

'

_,

....

o
,_

....

x/R0

....

....

....

....

....

'

'

'

'-

= -0.181

;r-

O.4

_-_

0.2

,
0

2.S

5.0

7.S

10.0

12.5

1S.00

103k_/U2,,

Figure

_'

_,

,:,

0.8

....

14.3a. Comparison

_-Qz"["9
2.S

IILllLll'''J

',0

103k-/U

of the turbulent-kinetic

62

energy profiles.

....
10,0

7.S

12.5

1S.I

2"

See symbols in figure 14.1.

_'''_

o.o

....

Y_
7.o

x/Ro = 0"181

0.6

___o

....

....

'

'

15

__--

"_.

o.4

_-_ ---_ o

0.2

....q

ocL-_-_r_

x/R 0 = 0.726

o' ' = '

OO

o.2L-

I ....
x./RI 0.... = 3.266

....

....

/._

,.o k'_''_r- ....

0.6

' ....

, ....

, ....

.....

''"'7

....

'

'

x/Ro

'

'

'

,=

'

__

....

x/R 0 = 2.177

....

....

'

'

'

'

'

= 4.354

1
I,
0

Figure

2.5

5.0

7.5

14.3b. Comparison

1.0_

- ....

....

10.0

iI
12.5

15.00

2.5

5.0

of the turbulent-kinetic

....

....

'

'

7.5

energy

10.0

12.5

15.1

profiles.

'
I

'

x/R 0 = -6.531

'

....

oil

.dRo=

__

....

....

'

'

'

....

....

....

-2.177

0.6

0.4

O.

_" ....

....

....

'''

'

'

'

F
o8 "_

x/Ro

= -4.717

x/R o = - 1.089

__

r---

0.4

"

0.2

L
"

1.0_-

'

'

'

....

....

....

'Jr)

....

x/R o = -0.181

x/R 0 = -3.266
06
I

'

'

....

....

"

0.4

0,2
0.6_

14.4a.

Comparison

103-_-v/ U2oo

103-_/U2
Figure

of the shear-stress

63

profiles.

See symbols

in figure

14. l.

1 .O

,.....

....

.;

0.8

....

....

'

'

'

x/R0 = 0.181

__

x/R 0 = 2.177

0.6cc

0.4

0.2

oa

x/R

= 0.726

_-

x/R0

= 3.266

0.6

_
\

0.4

--''"

0.2
o_,,

_.

___._.
',

.....
1.0

-----_.

....

....

'
....

.
'

'

_,

r_'_, _ , _ ....

_ .....

'
....

<

:o
-

x/R o = 1,451

....

....

....

....

x/R 0 = 4.354

0.8

0.6

o o

0.4

0.2
_

"I"_

_._,

....
1

_ ....

103__/U

Figure

14.4b.

Comparison
-

'

,,.,

103-_/U2

of the shear-stress
....

profiles.

....

....

-"
I

0.6

0.4

0.2

O
o

,
",'

....

'

'

'

'

....

I
I

A
0

2
..

SST

-5

0
x/R

Figure

14.5. Grid sensitivity

study.

64

10

0.6

0.4.

0.2

'

O0

-5

10

x/R o

Figure

14.6. Effects

of inlet boundary

'

'

....

'

layer profiles.

'

'

'

Case 1
Suggested
,Case 2

'

'

1.0

0.8

0.6

-E
0.4

by Menter

'

0.6

0.4

0.2

_"

'

'

'

'

'

'

'

'

'

f/

c-,
_"

-5

-_

10

x/R o

Figure

14.7. Effects

of position

of the outer streamline

65

in the outflow

region.

'

....

....

'

'

'

0.6

ata

0.4

0.2

_"-'---'_

k-c0 m0del

,I

//f_

/j

x
v

-5

5
x/a

Figure

14.8. Effects

of freestream

boundary

66

10

conditions

on the k-o3 model.

Section

15. Transonic

a. Experimental

Separation

was

The

chosen

dimensionality
0.0762

can

m in outside

the

occur

problem

radius

and extended
natural

to allow

was

of sidewall

61 cm upstream

arated
tively.

a shock

determination

in comparison

intensities,

ment

locations

wave

was generated

obtained

through

and shear
were

layer

tests.

The

of sufficient

layer

edge.

at its two end points.

number of 13.1.106/m.

reattachment.

stresses

determined

with conventional

These

pressure

sec-

of the bump
However,

that

the

separation

of

bump had a 20.32 cm

by a smooth circular

Test conditions
At this freestream

a relatively

large

were a
Mach

region

of sep-

at approximately
0.7 and 1.1 chords, respecby the laser velocimeter
technique
from

data consist

in the streamwise

from oil-flow

to produce

was

The straight

information.

on airfoils,

of two-

cylinder

layer just ahead

interaction

with the
configura-

contamination

leading

boundary

of boundary

strength

aligned

thin-walled

occur than is representative

of full scale. The circular-arc
of 1.905 cm. Its leading edge was joined to the cylinder

of separation

lence

boundary

with the

cylinder

arc. The axisymmetric

of the bump

flow. The separation

and reattachment
points were
Boundary
layer measurements
were obtained

upstream

on a circular

and a turbulent

arc that was tangent

to the cylinder
and the bump
freestream
Mach number of 0.875 and unit Reynolds
number,

bump

was a circular

two-dimensional

transition

accurate

not so thick,

greater severity would

chord and a thickness

of an annular
of the bump

in full-span

permitted

thickness
layer

section

to circumvent

tion of the cylinder

boundary

consisted

longitudinal

that

of sufficient

"Bump"

(ref. 40) was conducted

in the Ames 2- by 2-Foot Transonic
Wind Tunnel
and total pressure of 302 K and 9.5.10 4 N/m 2, respectively.
The axisymmetric

of this investigation

flow direction.
tion

over an Axisymmetric

data

The experiment
with total temperature
flow model

Flow

of profiles

and normal

visualizations,

of mean

direction.

and local

velocities,

Separation

surface

turbu-

and reattach-

static

pressures

were

instrumentation.

b. Results
Figure
metric

15.1 shows

"bump."

the comparison

dict the size of the flow separation.

one-equation

model

Comparisons
sured

of the pressure

Both k-e and k-co models

positions

comes

coefficients

along

a delay of the shock

The SST model

provides

the surface

position

the best overall

of the axisym-

and therefore
performance

underpreand the S-A

second.

of the velocity,

are shown

predict

in figures

shear

stress, and turbulent

15.2,

15.3, and

kinetic

15.4, respectively.

energy

profiles

Since

at specific

experimental

mea-

data

pro-

vide only two components

of normal stresses, the turbulent
kinetic energies shown in figure 15.4 were
obtained
by setting
w '2 = (u '2 + v'2)/2. The SST k-m model gives the best agreement
of the mean
velocity

sents

profile

and closer

agreement

of mean

shear-stress

profiles.

Figure
15.5 shows the grid-independent
comparison
for the SST model. The solid line repreresults obtained
using 181 x 108 grid points. The grid distribution
in the streamwise
direction
is

nonuniform
points

by allowing

in the cross-stream

the wall being

The

with experiment,

grid

between

points

denser

grids near regions

direction

less than 0.1. The

in the

two points

streamwise

of the original

is expanded
dashed

of high pressure
exponentially

line represents

direction
grid system

were

results

increased

gradients.

obtained

at the points

using

by placing

(181 x 108). As can be seen

67

The distribution

with y+ values

one

of grid

adjacent

to

382 x 201 grid points.

additional

grid

from the comparison,

point
the

solution is almostunchanged.It should,however,be notedthat Menter,who usedCFL3D (ref. 21)

anda grid systemsimilar to the currentone(181 101), obtained a slightly different solution. This is
the only
found.

case

out of all the cases

Although
(x/c =-3.2),

the

profiles

fit the data reasonably

well,

obtained

thickness

not have

cients
and

were

that

were

momentum
does

calculations

it appears

(x/c = -0.25)
culations

the

much

for the results

the other

before

by assuming
of 200,

impact

performed

from

in the

regions

obtained

differences

by using

between

the leading
just

edge

before

found

15.6 shows

flow

of the freestream

there

coefficients

conditions

layer

is slight
is almost

the

rig
bump

15.2. The caledge

at the leading
of pressure

profiles,

change

were

experimental

of figure

assumption

codes

reaches

at the leading

the comparison

inlet boundary

that although

of the pressure

profile

that this profile

Figure

two different

It can be seen

layer

the two

of the

the

as can be seen from the first subset

that the inlet boundary

but it has been

the comparison
effects

so far in which

on the solutions.

Re 0 = 1000.

the bump
Finally,

studied

has
edge
coeffi-

one with Re o = 200

of the velocity

profile

unchanged.

of the k-co model

are presented

in figure

15.7. As

can be seen in the figure, by decreasing

the value of the freestream
co 1000-fold,
the results are almost
unaffected;
both predict a delay of the shock position and underpredict
the size of the flow separation.

0.8

0.6

'

'

'_

._

_o

....

....
O

!_

/_

I !,_

!!

....

....

data

---

SST

.....

S-Aleqn

i_
0.4
o=
0.2

0
_'_,.

-0.2

....

0.50

-_

0.75

..,..4

1.00

1.25

1.50

x/c

Figure

15.1. Comparison

of surface

pressure

68

coefficient--transonic

bump

flow.

_Jc = 0688
0.100

0.075

_"

x/c . -0.25
I

--

...

....

'

....

iii.

"_I!'

0.025

0,100

....

, -

o11
l

"

0.075

0.050
0.125
0.025

--

I X]C

0.563

X/C

0.625

0,100

'

_'

'i

o o_5

5
L
2;

__.--_-.._-_

0.025

I
0.5

i
1.0

1.5

0,5

0,00_

_c.0875

,i.... 25

l!_P

0075

1.0
U/U

U/U

IJ

_;

0.050

0.125

O.lOO

0.075

0.050

0.025

o.125
....
,,,
....
,Ii
.....
,
o.o E--

_'_

0.050

0.025

_'"

0.5

1.0

1.5

U/U

Figure

15.2. Comparison

0.5

1.0
U/U

of mean

velocity

69

profiles.

See symbols

in figure

15.1.

.....

0.125

....

....

....

'-

0.100

'c_

'

'

x/c,
' I

....-0.25

....

....

....

0.075

O,OSO

0.025

....

0.125

0.100

x/c

x/c,

0.563

....

,_

,_

0.75

0.075

0.050

0.025

0.125

....

x/c

0.625

0.100

x/c,

0.813

0.075

Ei?

0,050

0,025

o
I

0
0

0,5

1.0

1.5

2.0

0.5

-uv/U 2 (" loo)

1.0
--uv/U

1.5
(-

2.0

100)

0.125
....

--'_

0.100

X/

....

....

....

_-

x/c,

0.875

1,125

0,075

0.050

i_

---- _-------_
",

0.025
i
,

....

....

....

....

....

0.125

0.100

'

....

x/c,

'_

1.25

0.075

%
i_,--

x/c

0.938

0,050

o -

0.025

. _._o

_-

/"

0,125

0,100

:o'.

:,O'C

1.375

0.075

0,050

0.025

0
0

0.5

1.0
--uv,'q.J

Figure

15.3.

Comparison

I
1.5

("

//j

2,0

0.5

1,0

1.5

2.0

-uv/U 2 (- 1oo)

1001

of mean

o o

shear-stress

7O

profiles.

See symbols

in figure

15.1.

0.125

:I
x/c,

0.100

____

,_c

0.688

-o.2s

0.075

c
o

0.050

0.025

%
I

....

....

....

....

0.125

x/c

0.563

....

....

E ....

....

'..-

":"

0.100
I

0.075

x/c.

0.75

-40
o

-'o

0.050

0.025

O.

125

0.100
o
0.075

0.050

0.025

....

....

]=,,,t,,,,

I=

o
1

."J

0.125

....

....

....

'

k/U_ (-lOO)

wu 2 (- loo)

'

'

'

....

_ ....

....

....

'-

'

" io

0.100

x/c

_c

0.875

1.125

0.050
0.075

o
o

.,

o.o2s

__

(_

'

'

'

'

'
0.125

_-_

0.100

_o

'

'

'

....

....

'

....

....

I'--.'

....

x/c

(,;,

X/C

1.25

....

:/"

....

I ....

'-

....

'.-

0.938

o o7s_\_
o

0.125

._

O. 1 O0

._o

....

Io

....

....

'

,,I,,,,I,

'

,-_o

....

....

x/c
J

x/c

'

'

....

o
.......

....

....

....

1.375

' i

0.075

_._

o
0.050

.-

"_

)
o

0.025

2-

_._
I
0

I
S

wu_ 2 (- loo)

Figure 15.4. Comparison

of the turbulent

....

I
5

wu 2 (-1oo)

kinetic-energy

71

profiles. See symbols

in figure 15.1.

'-

'

'-

0.8
-"

--

0.6

.Zo

li
u

tl

data

--

181-1o8

--

382"201

: L sTmo+-Z"

0.4

0.2

.....

181"101

(CFL3D)

_o._

-0.2

0.50

0.75

1.00

1.25

1.50

X/C

Figure

"

0.8

'

0.6

....

15.5. Grid refinement

'

'

'

'

'

'

'

'

study.

data

....

: _-

Re0=2oo

-/o-

"-

Ree = 1000

'

'

'

0.4

0.2

0
o
-0.2

",

0.50

0.75

oo
, , , I

1.00

1.25

-=
Z
--

1.50

x/c
.....
0.125

....

....

....

IC

x/c = -0.25

0.100

_.

....

"-'-

0.075
0.050

--t

0.025

+ ,

I ,
0.2

I
0.4

0.6

0.8

U/U e

Figure

15.6. Effects

of inlet boundary

72

layer profiles.

1.0

'

0.8

-"

'

'_,.]'

_
_

....

....

---

}
_

....

....

--

data
ptt/I..l = 0.001

0.6

0.4
/___

0 O0
_

k-co model

0.2
co

0
0
0000

-0.2

0.50

0.75

O0
_'._.

0000

1.00

1.25

1.50

X/C

Figure

15.7. Effects

of freestream

73

conditions

for the k-co model.

Section

16. Transonic

a. Experimental

airfoil:

selected

and flow field measurements

dom (ref. 41). The detailed
1980-81

data
0.319,

of the solution
included

Conference

0.498,

(x/c = 1, 1.025,
layer

and

to the assumed
surface

boundary

0.404,

boundary

for the RAE 2822

airfoil

from the boundary

on Complex

Turbulent

Flows

10 are M = 0.75 and Re = 6.2 106. The recommended

measured

coefficients,

results

layer
0.574,

pressure

profiles
0.650,

2). It should

freestream

layer, wake,

made in the RAE 2.438 x 1.829 m tunnel at Farnborough,

United Kingdescription
of this test problem can be obtained
from FILE 1120 of the

AFOSR-HTTM-Stanford

for Case

sensitivity

2822

data

This data set contains

tions

RAE

angle

of attack

coefficients

on the upper
0.75,

and

be noted

conditions,

will be presented

on both upper
surface

0.9),

angle

and

condi-

was 2.79 (the

surfaces,

locations

velocity

that the skin friction

not by the incoming

17). Flow

in a later discussion).

and lower

at selected
three

(ref.
of attack

profiles

coefficient

The

skin friction

(x/c = 0.152,

0.179,

in the

region

wake

is defined

by the local

flow conditions.

b. Results

The calculations

were

Company.

Far

circulation

due to lift. The grid is shown

tion coefficients

field boundary

performed

is given

conditions

in figure

using

a 386 82 grid provided

on the velocity
in figure

field took into account

16.1. The comparison

16.2. As can be seen

by L. Wigton

from

figure

of the Boeing

the effect

of the pressure
16.2,

of airfoil

and skin fric-

the SST model

shows

very

good agreement
with the experimental
data; both the k-e and k-co models exhibit a shift of shock position in the downstream
direction,
and the S-A model result is in between.
It should be noted that the
k-E model
mentioned

sented

fails to predict
flow
locations
is presented

separation.
The
in figure 16.3.

comparison

of the velocity

profiles

at the

above-

The comparison
of results obtained
using the Boeing grid and grids used by others is prein figure 16.4 for the SST model. As can been seen from figure 16.1, Boeing's
grid condenses

many points near the stagnation

and the shock regions.
the front of the airfoil and a better shock resolution.
Effects
turbulent

kinetic

of freestream
energy

conditions

is fixed

according

to 1. The results show that the k-c0 model

this particular
test case.
Finally,
the change

by slightly

of the solution

Thus

on the k-c0 model

it results

are shown

to k/Uoo 2 = 10 --6 while

the angle

of attack

is smaller

than the differences

74

by +10%,
found

in figure

the value

does not seem to be affected

adjusting

in a better

pressure

16.5.

of _/_t

The
varies

by the freestream

it can be seen from

in the models.

peak

freestream
from

conditions

figure

near

10 -3
in

16.6 that

Figure 16.1.

....

"_)-

Computational

....

grid with 386 82 points.

....

....

1.0__
0.5

....

'

'

'

'

'

'

..)

-0.5

---

-1.0
' ,

k-E:

.....
....
-I
,

S-A
k-ee
SST
, ,

'

'

'

'

'

'

'

I eqn

....

....

....

'

'

'

'

O-2

0.2

-,_,"

0.4

0.6

0.8

1.0

X/C

Figure

16.2. Comparison

of surface

pressure

75

and skin friction

coefficients.

10

it' 3

x/c

0.152

'

....

....

....

....

....

....

_'_

.//_.
/

_-

._-. j:o
.I"
2

/.-

._

._-

o o

0
0.85
5

_--'

"

0.90
'

'

0.95

....

x./c

....

,
0.7

0.8

J
o%

o./-

_2-

o_
o

1.0

o#._,,,j

_,-_

0.9

'--.

0.179

3-

,_-

1.00

....

o
.

I
0.7

0.8

0.9

1.0

0.7

0.8
'

=.'

'

0.9

....

_"

X/C

1.0

....

....

0.574

10.0

,.s,2.s:-

0.7

0.8

0.9

U/U

'

'

'

....

x/C

1.0

0.4

0.6

....

0.8
U/U

....

.....

'_

'

'

....

eo-

0_65

,oL

,!

b
10

o ;'

'_

....

x/c=

1.0

....

.....

'''I

40

20

0
0.2

40"--

''

3.0

_x'_

0.4

....

'I

--

x/c

0.6

....

1.0

0.2

0.4

0.6

0.8

1.0

....

--

....

0.75

20.--

10

0.8

iI .................
' ' '

oo_

2
.40

_ X/C

1.02.R

-20
I
0

0.2

so40

0.4

' ....

....
--

0.6

x/c

0.8

' ....

1.0

0.2

0.4

0.6

0.8

1.0

600

'''''_

0.9

400
o
200

_ok

_
C

10

_4I_, _, .... , .... ,

....
0.2

0.4

0.6

U/U

Figure 16.3. Comparison

0.8

,':
-20C

1.0

0.80

0.85

0.9O

0,95
U/U

1 .CO

of velocity profiles. See symbol notations

76

I .(

in figure 16.2.

....

....

'

'

'

'

....

....

1.0
0.5

do
(Do 0

-0,5

t"

-1.o

--_

Wigton's

....

Coakley's

-till

--

grid 386 x 82
grid

Menter's

grid

0.2

242

257

x
L

0.4

_Z-

62
97

(GFL3D)

0.6

0.8

x/c
i

'

'

'

'

0.0100

r-

, 4z_4 _

=St

0.0075

_o

0.0050

= _
._

Coakley's

r_

Wigton's

grid

Menter's

grid

grid

0.0025

-2
1

0.002

0,004

0.006

0.008

O.C

X/C

Figure

16.4. Grid sensitivity

study.

1.O

02

0
-0.5

--

-1

--

0_

Dtata

O.

I_.t/l.t

.O

-'

'

'

'

'

'

'

'

'

'

'

'

'

'

'

'

'

'

'

'

'

i.

_ ,

I ,
1.0

i-

k-o) model
0
0
0

x
0-

0.2

0.4

I ,
0.6

,
0.8

X/C

Figure

16.5. Effects

of freestream

boundary

77

conditions

on the k-co model.

I.C_)

i '

'

'

'

'

'

'

'

'

'

'

'

'

'

'

'

....

'

'-

1.0
0.5
00000

_"

O,
-0.5

----....

-1.0

or. = 2.79'-'
c,. = 2.5 o
o_ = 3.1

, , , , I , , , , , , , , , I , , , , I , ,
'

'

'

'

....

'

'

'

'

....

SST

o-

....

i i I , 1I

'

model

__

.....
0

!,I',
,

0.2

I _'_I

0.4

0.6

0.8

I-

1.0

X/C

Figure

16.6. Effects

of slight

variation

in the angle

78

of attack

of the mean

flow.

Part
Section

17. Conclusions

E. General

Conclusions

and Recommendations

In this section,

we will give our summary

of the work

performed

and our conclusions

regard-

ing the relative performance

of the various models tested. Four turbulence
models were tested on a
series of ten flows. The models included the one-equation
model of Spalart and Allmaras
and the twoequation

models

models

of Wilcox,

are k-co models

Launder-Sharma,

while

seven were relatively

simple free-shear
were relatively
complex
flows involving
the models
mance
tivity

by comparing

of the models
to initial

predictions

were

lows.

The

model,

considered

while

For the simple

flows

conditions

give improved

flows,

flows,

and Wilcox

to ffeestream

regard

flows,

although

followed

the Wilcox

perforand sensi-

the complex
models

of solutions.

flows

are about

in their

and plane
The

of its sensitivity

of the unmodified

with compressibility

flows.

of the jet flows.

because

None

equal
layer

is

involv-

for the simple

of the mixing

flows

is as fol-

The SST model

the best predictions

for these

studied

by the Spalart-Allmaras

model

best predictions
giving

there

to

models

modifications

do

they do

Menter

was sensitivity.

model

did not appear

a Navier-Stokes
While

to offer a definitive

performance

SST and

SST model

on the jet flows.

not discussed

of reattachment.
Finally,

The Wilcox
(where

was the SST model

there

solver

appear

explanation

of the models,
for accurate

of

k-E model
to be as sen-

was used)

to be several

as it

possible

at this time.

the Spalart-Allmaras

and then the Launder-Sharma

required

because

was the Launder-Sharma

solutions

model

and Wilcox

and the maximum

was

models.
y+ allow-

off the wall.

with the other models,

the

model

in this regard

flows

by the SST model,

compared
above,

the best overall

model

in between.

on grid spacing

the

Although

attention.

above,

falling

Although

downstream

the numerical

of Wilcox.

job in predicting

layer, although

for the complex

numerical

was based

improvement
flows.

SST
tested,

flows and three

performance
of

models

followed

the k-co model

to be poor

The worst

models

to the

to be the best,

stated

Menter
flows

of grid refinement

of the various

except

giving

unreliability

mixing

as stated

conditions

able at the first grid point

As

is judged

for this, we are not prepared

This evaluation

mance

model

and SST models

model

separation.

was for the free-shear

found

and the

Of the ten

predictions.

with the Spalart

With

studies

with the best of the other

all of the models

with the resulting

in predicting

explanations

and finally

comparable

the compressible

For the complex

sitive

performance

it does the best overall

results

of the Wilcox

well in predicting

model

data, tests to determine

tests included

to be the k-co SST model,

k-_ model,

and the Launder-Sharma

freestream

its ability

is judged

with the Spalart-Allmaras

performance

These

the relative

because

giving

free-shear

performance,
wake

model

the best model

Wilcox

is a k-_ model.

conditions.

the Launder-Sharma

ing separation

The

model

with experimental

regarding

best overall

then

Menter.

and zero-pressure-gradient
boundary
layer
separation.
In addition to testing the relative

also conducted.

and boundary

Our conclusions

and

the Launder-Sharma

the SST

needs

Spalart-Allmaras
there

is considerable

improvement

All of the models

in this report,
Corrections

and

S-A models

models
room

need better

are

flow,

compressibility
appear

and curvature

not coordinate-invariant

79

to give

for improvement

on the wake

none of the models

for rotation

are found

and

the

superior

perfor-

of these

models.

S-A model

corrections

needs

for free-shear

to do well on recovering

flows

are still another

area

requiring

in that

use

functions

they

dependenton distancefrom nearestwalls, whereasthe modelsof LaunderandSharmaandWilcox

areinvariant.Coordinateinvarianceis a desirablefeaturein modelssinceit removesambiguitiesand
uncertaintiesin the definition of a distancefunction and is more in tune with the original intent
of higher orderturbulencemodeling.It would be desirableif the two newermodelscould be made
coordinate-invariantas well. Thesetopics, as well as additional testing of the currently popular
modelson a variety of flows, will bethe subjectof future studies.

8O

Appendix.

Self-Similar

Equations

for Free-Shear

Flows

There is a large variety of self-similar

flows that are of great importance
from an engineering
and a turbulence
modeling
point of view. These flows include a large variety of physical
phenomena
of practical
flows (ref. 3), compressible
and incompressible
flows, attached boundary
layers, and freeshear flows. Mixing layers, jets, and far wakes belong to this class of turbulent
flows. The principle
of
self-similarity
flow
(refs.

applications,
modeling

flows

at high Reynolds

at all sections
can be solved

and using

simpler

and

numbers

asserts

the similarity

numerically
methods

using

with

that far enough

is independent
Navier-Stokes

transformed

downstream

of initial
codes

self-similar

the

conditions

in complex

equations

engi-

in turbu-

validations.

a. Compressible

boundary

The two-dimensional
pressible

perfect

equation

of state.

conservative

flows

similar

3 and 4). These

neering
lence

of free-shear

is geometrically

layer

equations

equations
of motion

for the self-similar

gas are given by the conservation

The static

pressure

form of the equations

is assumed

neglecting

Ox(PU) +
O

of mass,
constant

lower

turbulent

momentum,
and equal

order

terms

mixing

layer

and energy

to the freestream

is written

of a com-

equations

and the

pressure.

The

as follows:

=0

,(or)

tuff

(1)
O--_(puH) +

(pvH)
P = RpT
2
u

H = cpT+-f+k

where

x and v are the coordinate

the temperature,
energy
perfect

p is the static

directions,
pressure,

u and v are the velocity

and H is the total enthalpy

components,

is k, the absolute eddy viscosity

is gt, the specific heat of the gas at constant
gas constant is R, and the turbulent
Prandtl number is Pr r

Equations
1 for total enthalpy
and energy conservation
This quantity is not available
and is omitted in zero-equation
excluded

in two-equation

turbulence

should be included,
although little
simulations
of free-shear
flows.
In most computational
dimensional

form.

Equations

models.
difference

fluid dynamics

The

(CFD)

1 may be written

codes,

when

81

approach
these

indicates

terms

the equations

in nondimensional

pressure,
and density leading to a similar set of equations
3t/('),- 1). The factor _' is the ratio of specific heats, which

pressure

T is

kinetic
is cp, the

include the turbulent

kinetic energy,
turbulence
models; it can be included

rigorous

is found

p is the density,

of the fluid. The turbulent

that

are included

of motions

form using

all quantities
in the present

are written

a reference

k.
or

length

in nonscale,

where the factor Cp is replaced

by the factor
is equal to 7 = 1.4 for a diatomic
perfect gas

accordingto classicalstatisticalmechanics.The referencevaluesareusuallyselectedto be ffeestream

valuesin the high-speedsideof the mixing layer.
b. Incompressible
The

equations

two-dimensional
of motion

and axisymmetric

for two-dimensional

and axisymmetric

pressible
fluid are given by the conservation
of mass
constant
and equal to the freestream
pressure.
The
lower order terms is written as follows:
Ou

equations
free-shear

flows

and momentum.
The static pressure
conservative
form of the equations

of an incomis assumed
neglecting

_v
(2)

-_(

In equations
flow.

2, the subscript
eddy

c. Separation

of variables

self-similar

_-f

j = 0 stands

The kinematic

The

u21+

viscosity

equations

(uv

l?-ff-fy_
(

jyVt-_

for two-dimensional

is defined

flow and j = 1 stands

for axisymmetric

as v t = laJp.

and coordinate

transformation

can be written

in compact

from (x,y) coordinates

to (rl,x) self-similar
coordinates,
components
and turbulence
variables,
for mixing layer

11 = -y

)_U")

form

with the following

transformations

and separation
of variables
for the velocity
and jet flows (refs. 3 and 4) as

and

x-'x

u=aU

v=aV

p=rp

T = T/u_

(3)

J = 2_Jfu2yJdy

k = a2K

e = aE

o=aw

and with the following

transformation

(ref. 2) for far wake

rl = yu../,,/-D-x
= u -aU

v,_ = axN

flows

and

k = a2K

as

x = x
D = 2Ipu(u,

-u)dy
(4)

2
Uoo

03 = m W

uooa

e -

where
symbols

the self-similar
Tand

variables

p are repeated

E
X

v. = a x N
U

U, V, K, W, N, E, T, and p are functions

in the definition

of the separation

only of rl (for simplicity

of variables);

the variables

are functions
only of the transformed
coordinate
x; and u_, is the constant
velocity
wake flow. D is the total drag per unit mass on the body upstream
of the wake

of the freestream
flow, and J is the

momentum
flux per unit mass of the jet. The variables k, e, and 03 are the turbulent
kinetic
sipation rate, and specific dissipation
rate variables
defined in the turbulence
models.

82

the

a, u 1, and r

energy,

dis-

The transformationcoefficient, a(x),

turbulent

free-shear

flows

is chosen

without

Flow
layer

The

Mean

(J/x) 112

Round jet

j1/2[x312

Far wake

(Dlpx)

1/2

boundary

layer

equations

of motion

for two-dimensional

and

axisymmetric

below.

momentum

and mass conservation

The self-similar
of a compressible

of the

equations

transformed

are shown

one

u1

Plane jet

flows

for each

a(x)

Mixing

d. Serf-similar

loss of generality

as follows:

mean

momentum,

fluid are shown

equations

energy,

and mass conservation

mixing
equations

layer
for flee-shear

flows

below.
oV, ClV

d(

dn

,dH
9 v -_

for compressible

"_ pNdT
_--_( 7 - 1 Prtdrl

. dU_

=o

pNudU+
drl

cYknN"_dX) = 0

(5)

pV*-pV-rlpU

= -I p Udc
0

U2
H = --_T
where
ponent

the transformed
is defined
The mean

equations

as V* = Venergy

have

no source

+--_-+
terms,

K
and the transformed

transverse

velocity

com-

11U.

equation

can be reduced

to the following

equation

for the mean

temperature

profile:
. ,dT
pv _

where

E is the transformed

the high-speed

dissipation

d ( pNdT')
-_\-fi--Tt-_)

rate of turbulent

side of the layer. In equation

6, the source

83

(7-1)

2
MlpE

kinetic

energy,

(6)

and M 1 is the Mach

term has been transformed

in order

number

of

to set the

rangeof meanvelocity, temperature,anddensitybetween0 and 1 If the turbulentkinetic energyis

neglectedin the definition of total enthalpyin equations5, thenequation7 for the meantemperature
is obtained

ov, dr d (oUdr
. = . (clrA:
all.
i
= (7-1)M1
where

the source

Mean

shown

term

momentum

For incompressible
below.

is a direct

function

of the production

and mass conservation

fluids,

equations

the self-similar

of turbulent

(7)

kinetic

for incompressible

equations

for the mean

energy.

flows

momentum

and continuity

are

(8)
V* - V-flU
The transverse
page

velocity

Two-equation
The
form as

component

turbulence

self-similar

V* is defined

models

equations

dF.._

_a-fi
l d(

nJPcY u _

rl p_I_N-_

flow in the table

on the following

2--4)

for the two-equation

_j&"

p v,

(refs.

for each free-shear

turbulence

= pN

Cf l P glV " (_'-_

dU'_ )F2

models

-pE

-cf2P_2EF

can be written

in compact

+ SkoK

+ Sf 9F + 2_

q--

(9)

p dK dF
Wdndn

,)

N and the model

where

constants

K
W -

K"
c_t'--E

for free shear

Model

K-F

k-E (ref. 6)

K-E

k-co (ref. 4)

K-W

SST (ref.

K-W

14)

the generalized

E = _3*WK
flows

are shown

c_ = [3" = 0.09
below

as

c t2

_f

Yk

(Yc

ce1=1.44

ce2 = 1.92

_e= 1.3

1.0

[3*WK

y=5/9

13/13"=516

c=0.5

0"*=0.5

_3*WK

'_0.44

13/13"=0.92

_Jo_=0.856

1.0

13o_

nomenclature

of model

K-F stands

84

for K-E or K-W models

 One-equationturbulencemodel
The self-similarequationsfor the one-equationmodelof Spalart-Allmaras(ref. 7) can be written in compactform as
pV ,dN

1 1 d(rlj

%1

Flow

= 0.1355

layer

Round

SspNJ
(10)

o = 2/3

jet

So_

Ss

-U

2U

2U

4U

2.5U

1.5U

Far wake

boundary

conditions
value)

(conservation

0.5

on the

in freestream

of the total
of total

to achieve

numerical

methods

velocity

and

boundaries,

Sorl U(c_) do

-0.5 SOrl U(o)

0_

do

-0.5rl
i

turbulence

variables

and Neumann

convergence
to impose

by explicitly

momentum

drag

in the table on page 83). These

in order

numerical
be relaxed

-(1/rl)

do

are Dirichlet

conditions

(specified

zero normal

conditions

boundaries.

conservation

is obtained

-J0n u(o)

conditions

function

The

as
g

in symmetry

flows

0.5U

are defined

e. Boundary

(specified

equations

5 8

The

of the model
Sk

jet

Plane

dure

+ Cb2okdrdN)2+
lp-(

Su

J ]

Mixing

shown

CblPN-_
= 0.622

cb2

flow, the parameters

Flow

shear

Parameters

For each flee-shear

gradient)

NdN'_

qJocz,a , P

drl

necessary

from

integral
scaling

leads

D in far wake

to an additional
flows

conditions

initial

profiles

constraints
the mean

and total

must be satisfied
to self-similar

into the numerical

velocity

integral

profile

relaxation
until convergence
is achieved;
the system
until the error reaches machine
accuracy. However,

in free-

J in jet flows

by the numerical

profiles.
method.

after

constraint

momentum

There
One

are

successful

as

method
different
proce-

each

numerical

iteration

of the

of ordinary
this method

differential
decreases

equations
may
the rate of con-

vergence
and increases
the total computational
time. For plane jet and far wake flows, the "inverse
method of relaxation"
provides
a much faster convergence
to the self-similar
solution. In this method,
instead

of specifying

a symmetry

or Neumann

boundary

condition

at the center

of the flow, the veloc-

ity magnitude
in the symmetry
plane is specified
after each iteration
subject to the conservation
of
total momentum
in the streamwise
direction.
The use of such an adaptive
Dirichlet
condition
instead
of Neumann

condition

at the center

of these

flows

same final solution.

85

leads

to a very

fast relaxation

procedure

and the

References
1. Marvin,

J. G.; and Holst,

AIAA

Paper 90-2995,

2. Marvin,

T. L.: CFD Validation

AIAA

J. G.: Accuracy

No. 2, AGARD-CP
3. Schlichting,
London,

Requirements

437,

vol.

H.: Boundary

Verlag

8th Applied

Karlsmhe,

4.Wilcox,
D. C.: Turbulence
La Cafiada, Calif., 1993.
5. Jones,

W. P.; and Launder,

Turbulence.
6. Launder,

Journal

of Flow

and Benchmark

Experiments

of Computational
McGraw-Hill

for

1990.
for CFD

Fluid Dynamics,

Book Company,

CFD.

DCW

of Heat and Mass

a Spinning

Industries,

of Laminarization

B. I.: Application

Near

for the 90's.

Validation.
May

Paper

1988.

Inc., New York,

Toronto,

1960.

Modeling

B. E.; and Sharma,

the Calculation

Conference,

B. E.: The Prediction

International

Flows--Challenge

Aerodynamics

1, Validation

Layer Theory.

G. Braun,

for Aerodynamic

Transfer,

Disc.

Letters

Palm

Drive,

Model

of

15, 1972, pp. 301-314.

Dissipation

in Heat

5354

with a Two-Equation

vol.

of the Energy

Inc.,

Model

and Mass

of Turbulence

Transfer,

vol.

to

1, no. 2,

1974, pp. 131-138.

7. Menter,
AIAA

F. R.: Two-Equation
J., vol. 32, Nov.

Eddy

8. Spalart,
P. R.; and Allmaras,
AIAA Paper 92-0439,
1992.
9. Spalatt,

P. R.; and Allmaras,

La Recherche
10. Rodi,

Aerospatiale,

F. R.: Influence

F. R.:
93-2906,

14. Menter,

Turbulence

Model

for Aerodynamic

Flows.

Attached

Zonal

17. Kline,

Two

Equation

Fla.,

Values

Equation

and

S. J.; Cantwell,
Turbulent

18. Birch,

King,

k-c0 Turbulence

25th AIAA

Models

Turbulent

Boundary

Flows,

B. J.; and Lilley,

Mixing

Stanford
Layer.

S. J. Kline,

vol. 1, 1981,

Turbulence

Model

Predictions.

for Aerodynamic

of Two-Equation

Fluid

Dynamics

L. S.: A Mathematically

Turbulent

Flows.

S.: Planar

a One-Equation

Models.

AIAA

J.,

Flows.

AIAA

1993.

Layers.

G. M.:

University,

In 1980-81

B. J. Cantwell,

Turbulence

Conference,

Simple
AIAA

1980-81
Stanford,

Calif.,

for Transonic
Springs,

Closure

Colo.,

Model

for

1985, pp. 1684-1692.

Model

for High

AFOSR-HTTM-Stanford

and G. M. Lilley,

86

Turbulence

Transport

AFOSR-HTTM-Stanford

pp. 170-177.

Models

Colorado

J., vol. 23, Nov.

B. S.; and Barth, T. J.: A One-Equation

Turbulence
Wall-Bounded
Flows. NASA TM- 102847, 1990.

Complex

with

for Advance

on k-co Turbulence

L. C.: Assessment

94-2343,

and Separated

16. Baldwin,
Number

the k-e Model

1991.

1988, pp. 1299-1310.

of Freestream

Orlando,

D. A.;

Combining

Jan.

of the Scale-Determining

F. R.; and Rumsey,

15. Johnson,

Models

91-0216,

1992, pp. 1657-1659.

Flows. AIAA Paper

June 20-23,
1994.

Calif.,

S. R.: A One-Equation

Paper

D. C.: Reassessment

13. Menter,
Paper

Flows.

no. 1, 1994, pp. 5-21.

J., vol. 26, no. 11, Nov.

vol. 30, no. 6, June

Applications.

for Aerodynamic

with Two-Layer

12. Menter,

for Engineering

Model

AIAA

11. Wilcox,

Models

Turbulence

Near

the Wall.

Turbulence

S. R.: A One-Equation

W.: Experience

Model

AIAA

Viscosity

1994, pp. 1299-1310.

Reynolds

Conference

on

1981.
Conference

eds., Stanford

on Complex

University,

Stanford,

19. Liepmann,H. W.; and Laufer, J.: Investigationsof Free Turbulent Mixing. NACA TN-1257,
1947.
20. Roger,S. E.; andKwak, D.: An Upwind DifferencingSchemefor the Steady-StateIncompressible Navier-StokesEquation.NASA TM-103911,Mar. 1992.
21. Rumsey,C. L.; andVatsa,V. N.: A Comparisonof the PredictiveCapabilitiesof SeveralTurbulenceModelsusing UpwindandCentral-DifferenceComputerCode.AIAA Paper93-0192,1993.
22. Foss,J. E: The Effect of the Laminar/TurbulentBoundary Layer Stateson the Development
of a Plane Mixing Layer. Proc. Symposiumon Turbulent ShearFlows, April 18-20, 1977,
pp. 11.33-11.42.
23. Bradbury, L. J. S.: The Structure of a Self-Preserving
Turbulent
Plane Jet. J. Fluid Mech., vol. 23,
pt. 1, Sept.

1965, pp. 31-64.

24. Wygnansld,
vol.

I.; and Fiedler,

38, pt. 3, Sept.

Science
25. Fage,

Laboratory,

26. Chevray,
AIAA

J, vol. 7, no. 8, Aug.

Stanford
28. Birch,
Langley
29. Settles,

30. Viegas,

Free

Research

Center,

NASA,

Layers.
32. Huang,
AIAA

AIAA

Elliot,

AIAA

Mech.,

Labs,

Flight

in the Wake

1932, pp. 702-705.

in the Wake

of a Thin Flat Plate.

Layers.

In 1980-81

AFOSR-HTTM-Stanford

B. J. Cantwell,

and

G. M. Lilley,

eds.,

of the Experimental
Flows,

Volume

Va., July 20-21,

Turbulent

Center,

Data

for Developed

Free

I--Conference

Proceedings,

1972, pp. 11-40.

Boundary-Layer

and Free

Shear

Database.

Apr. 1993.
of Compressibility

Corrections

to the k-e Model

in

J., vol. 30, no. 10, Oct. 1992, pp. 2369-2370.

of Compressibility

on the Characteristics

of Free

Shear

J., vol. 28, no. 3, Mar. 1990, pp. 439--445.

92-0547,

33. Coles, D.: The Law

pp. 191-226.

of the Wakes

D.: The Turbulent

35. Huang,

P. G.; Schwarz,

ary Layer Code.

June 1990.

T. J.: An Implicit

30th Annual

34. Coles,
1962.

36. Van Driest,

Hampton,

G. S.: Effects

P. G.; and Coakley,

Paper

Shear

M. W.: Assessment

Layers.

M.; and

Review

Turbulent

Research

J. R.; and Rubesin,

31. Samimy,

J. Fluid

and Velocity

vol. A135,

Measurements

S. J. Kline,

L. J.: Hypersonic

Ames

Shear

Jet.
Research

vol. 1, 1981, pp. 364-368.

J. M.: A Critical

Layers.

High-Speed

Roy. Soc., London,

Flows,

Calif.,

Shear

CR-177610,

on the Temperature

on Free-Shear

Turbulent

G. S.; and Dodson,

NASA

Effects

Stanford,

S. E; and Eggers,

Turbulent

Scientific

1969.

on Complex
University,

Proc.

L. S. G.: Turbulence

P.: Compressibility

Conference

in Boeing

D 1-82-0712.)

Obstacle.

R.; and Kovasznay,

in the Self-Preserving

(Also

V. M.: Note on Experiments

Cylindrical

27. Bradshaw,

Measurements

pp. 577-612.

Document

A.; and Falkner,

of a Heated

H. E.: Some

18, 1969,

Boundary

Meeting,

E. R.: Turbulent

Report,

Reno,

in the Turbulent

Layer

Dept.

Boundary

Boundary

P.: L-BL-SW--

of Mech.

Layer

Engrg.,

in Compressible

no. 3, 1951, pp. 145-160.

87

Code

for Turbulence

Nev., Jan. 6-9,

in a Compressible

W. R.; and Bradshaw,

Internal

Navier-Stokes

Modeling.

1992.

Layer.

Fluid.

Flow

J. Fluid

RAND

A General
Stanford

Fluids.

Mech.,

Corp

vol.

Rept. R-403-PR,

Reynolds-Stress

University,

1, 1956,

Stanford,

J. Aeronaut.

BoundCalif.,

Sci., voi.

18,

37. Hopkins, E. J.;

and

Heat

and Inouye,

Transfer

M.: An Evaluation

on Flat Plates

of Theories

at Supersonic

for Predicting

and Hypersonic

Mach

Turbulent
Number.

Skin Friction
AIAA

J., vol. 9,

Family

for Com-

no. 6, 1971, pp. 993-1003.

38. Huang,

P. G.; Bradshaw,

pressible
39. Driver,
Paper

Turbulent

D. M.: Reynolds
91-1787,

40. Bachalo,

Shear

W. D.; and Johnson,

Flow

P.; McDonald,

ary Layer

Wake

Layers.
Stress

T. J.: Skin Friction

AIAA

and Velocity

Profile

J., vol. 31, no. 9, 1993, pp. 1600-1604.

Measurements

in a Separated

Boundary

Layer

Flow.

AIAA

1991.

on an Axisymmetric
41. Cooke,

P.; and Coakley,

Boundary

D. A.: Transonic

Model.

AIAA

M.; and Firmin,

Measurements.

AGARD

Turbulent

Boundary-Layer

Separation

Generated

J., vol. 24, 1986, pp. 437--443.

M.: Airfoil
AR-138,

88

RAE2822--Pressure
1979.

Distributions

and Bound-

REPORT
Public

reporting

burden

for

this

DOCUMENTATION

collection

Of reformation

is

estimated

to average

Form

PAGE
1 hour

per

Approved

OMBNoo7o4-oloo

response,

including

the

time

for

reviewing

instructions,

searching

existing

data

sources,

gathering and maintaining the data needed, and completingancireviewing the collectionof information. Send comments regarding thisburden estimate or any other aspect of this
collectionof information,includingsuggestionsfor reducingthis burden, to Washington HeadquartersServices. Directoratefor information Operationsand Reports. 1215 Jefferson
Davis

Highway,

1. AGENCY

Suite

1204,

USE

Arlington,

ONLY

VA

(Leave

22202-4302.

blank)

and

to the

2.

REPORT

Office

of Management

and

Budget,

DATE

TITLE

AND

REPORT

Reduction

TYPE

Technical

April 1997
4.

Paperwork

3.

Pr _ect

AND

5. FUNDING

Modeling

Validation,

R G. Huang,:

PERFORMING
ORGANIZATION
NAME(S)ANDADDRESS(ES)

8. PERFORMING
ORGANIZATION
REPORT NUMBER

A-976276

9. SPONSORING/MONITORING
AGENCYNAME(S)ANDADDRESS(ES)

10.

National Aeronautics
and Space Administration
Washington, DC 20546-0001

SUPPLEMENTARY

NASA TM- 110446

T. J. Coakley, Ames Research

(415) 604-6451

Research

Center, MS 229-1, Moffett Field, CA 94035-1000

Corp., Moffett Field, California

DISTRIBUTION/AVAILABILITY

Unclassified

13.

SPONSORING/MONITORING
AGENCY
REPORT NUMBER

NOTES

Point of Contact:
*Caelum

NUMBERS

and T. J. Coakley

Ames Research Center

Moffett Field, CA 94035-1000

12a.

20503

505-59-50
505-59-53

J. E. Bardina,*

11.

DC

Testing, and Development

6. AUTHOR(S)

7.

Washington,

COVERED

Memorandum

SUBTITLE

Turbulence

(0704-0188),

DATES

--

Subject

Category

ABSTRACT

(Maximum

*Universit), of Kentucky,

STATEMENT

12b.

Lexington,

Kentucky'

DISTRIBUTION

CODE

Unlimited
34

200 words)

The primary objective of this work is to provide accurate numerical solutions for selected flow fields and
to compare and evaluate the performance of selected turbulence models with experimental results. Four popular
turbulence models have been tested and validated against experimental data of ten turbulent flows. The models
are: l) the two-equation k-03 model of Wilcox, 2) the two-equation k-e model of Launder and Sharma, 3) the twoequation k-co/k-e SST model of Menter, and 4) the one-equation
model of Spalart and Allmaras. The flows
investigated
are five free shear flows consisting of a mixing layer, a round jet, a plane jet, a plane wake, and a
compressible
mixing layer; and five boundary layer flows consisting of an incompressible
flat plate, a Mach 5
adiabatic flat plate, a separated boundary layer, an axisymmetric
shock-wave/boundary
layer interaction, and
an RAE 2822 transonic airfoil. The experimental
data for these flows are well established
and have been
extensively used in model developments.
The results are shown in the following four sections: Part A describes
the equations of motion and boundary conditions; Part B describes the model equations, constants, parameters,
boundary conditions, and numerical implementation;
and Parts C and D describe the experimental data and the
performance

14.

SUBJECT

of the models in the free-shear

flows and the boundary

layer flows, respectively.

TERMS

Turbulence

modeling,

Computational

fluid dynamics,

15.

NUMBER

16.

PRICE

OF PAGES

98

CFD, Code validation

CODE

A05
17.

SECURITY
CLASSIFICATION
OF REPORT

Unclassified
NSN

7540-01-280-5500

18.

SECURITY
CLASSIFICATION
OF THIS PAGE

19.

SECURITY
CLASSIFICATION
OF ABSTRACT

20.

LIMITATION

OF ABSTRACT

Unclassified
Standard
Prescribed
298-102

Form
by

ANSI

298
Sld.

(Rev.

Z39-18

2-89)