Professional Documents
Culture Documents
NASA
Technical
ICOMP-92-27;
Memorandum
105993
CMOTT-92-14
A Realizable Reynolds
Equation Model
Tsan-Hsing
Institute
Shih
for
and Jiang
Stress Algebraic
Zhu
Computational
Mechanics
in Propulsion
(NASA-TM-I05993)
REYNOLDS
STRESS
MODEL
Prepared
Kyoto,
(NASA)
A REALIZABLE
ALGEBRAIC
EQUATION
36
N93-16596
L. Lumley
Cornell
Ithaca,
Ninth
Transition
Ohio
and
John
and
llnc I a s
University
New York
0139889
for the
Symposium
Japan,
on Turbulence
August
10 -18,
Shear
1993
Flows
/',"/
* *
\@.2
2_
\_._,
..-,=,,/oe
\%X "- _----_
A REALIZABLE
REYNOLDS
STRESS
ALGEBRAIC
Tsan-Hsing
Institute
for Computational
and Center
for Modeling
NASA
in Propulsion
of Turbulence
Lewis
Research
Ohio
John
MODEL
Zhu
Mechanics
Cleveland,
Corneli
EQUATION
and
Transition
Center
44135
L. Luraley
University,
Ithaca,
New
York
14853
Abstract
The
invariance
Reynolds
ysis
leads
stresses
the
theory
stresses
in
to the
mean
number
constitutive
velocity
isotropic
mechanics
Reynolds
to a turbulent
classical
continuum
in high
relation
gradients
eddy" viscosity
is
just
to
flows.
that
relates
in a more
mode/is
applied
turbulent
the
general
analyze
The
Reynolds
form
the linear
anal-
in which
approximation
to the
coefficients
kinetic
will ensure
energy
satisfy.
--
ficiently
ison
capability
model
rotation
that
of each
most
based
which
and
proposal
equation
rotational
all conventional
step
models.
viscosity
finite-
in detail
addition,
shear
models
solutions
accuracy
improves
In
fail to
are taken
diffusion-free
quantities.
homogeneous
turbulent
a conservative
compared
turbulent
These
models
configurations
with
significantly
eddy
of the
of second-order
are
rate.
turbulence
and numerical
results
rotation
component
existing
schemes
mean
two
mean
are performed
calculated
present
to simulate
the
over backward-facing
for both
the
of K-e
rates
positivity
differencing
The
data
that
is able
and
Grid-independent
fine grids.
shows
rate
calculations
by using
experimental
the
flows
The
method.
are obtained
strain
realizability
Separated
as applications.
volume
mean
The
the
the
flows
on sufwith
the
comparpredictive
proposed
with
large
fail to simulate.
1.
Introduction
Numerics]
simulation
tational
fluid dynamics
without
any
flows
rectly
(CFD).
turbulence
within
the
large-eddy
of turbulence
simulation
computed
Reynolds
end
number
conditions.
most
practical
calculations
the
In turbulence
the
Reynolds
turbulence
models
equation
group
models
Reynolds
cosity
K-e
of the
most
zero-,
are
often
however,
correlations
of higher
underlying
order
mechanism)
In the
standard
termined
from
a set
turbulence
conditions
has shown
that
that
weak
the average
of a constent.
into account
(Launder
second
1980).
they
shear
are universal.
flows
ratio of Pie
the effect
of annular
transport
Rodi
of strearaline
and
models,
Reynolds
inaccuracy
Numerical
experience
has a broad
(1981)
improved
jets.
that
over
about
unphysical
their
results.
by using
ability
to
C_ as a function
of
kinetic
a function
and obtained
Recently,
Reynolds
Inappropriate
equilibrium
of the turbulent
proposed
curvature
twin parallel
under
applicability,
found
stresses
turbulent
are constant
flows
(1980)
and
is one
closures
coefficients
vis-
calculations.
individual
equations.
the
eddy
Second-order
of the
for simple
(P is the production
and
The
model
former
in which
flow
dynamical
algebraic
in this group
the
to
treated,
The
in practical
transport
in a serious
Rodi
stress
field.
1974)
equations.
model
can be significantly
Leschziner
today
are
1980)
flow
in terms
According
models.
(Rodi,
mean
closure
(often
Reynolds
Spalding,
used
expressed
correlations)
models
moment
all the
of experiments
Therefore,
Navier-Stokes
information.
equation
to the
the
are
the
transport
second-order
result
model,
(Rodi,
groups:
and
models
correlations
could
K-e
are di-
problems
in DNS.
on averaged
moment
to consistently
in these
is the
for high
has serious
correlations
two-equation
can simulate
it is difficult
appearing
modeling
calculation
they
two
stress
own dynamical
because
to DNS
applicable
also exist
are based
number
of turbulence
and
related
called
by their
are attractive
predict
model
time
(DNS)
Reynolds
though
expensive
difficulties
into
and
turbulence
group,
very
simulation
low
scales
LES,
second-order
Reynolds
one-
the large
turbulent
with
algebraically
popular
are determined
signifies
the
end
of compu-
modeling.
unknown
(the
numerical
A compromise
modeled.
of these
be divided
two-equation
In the latter
stresses;
may
the
stresses
are
at the present
stresses
and
includes
in which
aid of turbulence
flow quantities
way the
computers.
is usually
Some
modeling,
of determinable
to simple
scales
turbulence,
boundary
with
is restricted
approach
the small
with
equations
of direct
of current
(LES)
in the development
The approach
models
capabilities
is a bottleneck
improved
Yakhot
energy)
instead
for C_ which
and
results
co-workers
takes
in the
have
model
fidents
is of the same
take
model
different
(Speziale
term,
turbulent
a backward-facing
and
Thangam
gradients.
mean
the
The
flows
mean
needed
for more
terms
by algebraic
perform
better
stress
flow
turbulence-driven
secondary
ASM
its generality.
sure based
normal
stresses
merical
from
this
special
becoming
negative
of ASM
closure
numerical
care
may
general
relation
is capable
to prevent
especially
difficulties
of ASM
Leschziner,
first
example
K-e model
may
1985),
such
clo-
turbulent
and
than
three
nicely
the
the nu-
that
dimensional
resolved
is
limit
second-order
complicated
in general
was
does
order correlations,
to be taken
model,
are cal-
of generating
which
needs
be more
(1980)
model
viscosity
of the
even
Rodi
the
form,
is
and diffusion
stresses
deficiencies
and
due to
dependence
cause
(Huang
velocity
a well known
eddy
of higher
concept
linear
nature,
flows;
in the
stresses
stresses,
invariant
modeling
will also
is an
others
convection
ASM
in a tensorial
inappropriate
Moreover,
second-order
Recently,
appear
where
which
mean
the Reynolds
the isotropic
Speziale
in two-dimensional
a more
Reynolds
for certain
ducts
flow
normal
to its anisotropic
while
to a separated
that
the
in which
K-e model
motions
does not
implementation
parent
Owing
correlations,
ASM.
(ASM)
in non-circular
In addition,
as pressure-strain
model
for the
ratio
to the
suggests
that
scale
eddy-viscosity
stresses
by eliminating
equations
the isotropic
is fully-developed
However,
and
time
like many
is insufficient
expressions.
than
shear
This
transport
an algebraic
but
In fact,
one),
K-e
of calculations.
are proportional
of the concept.
flows.
accuracy
coef-
based
length
isotropic
type,
gradients
modeled
culated
not.
does
nature
complex
Boussinesq's
stresses
boundary-layer
velocity
in the
developed
usually
isotropic
on the
uses
RNG
et M. (1978),
reattachment
based
methods.
to the production
model
by Kim
overall
the RNG
Reynolds
concept
of the
erroneous
(including
group,
the
studied
the
of the
17 is the
this
of the
(RNG)
C1, related
of 71, where
In applying
to assess
but
version
coefficient
prediction
used
model
that
the
rate.
a good
K-_ model
assumes
1992),
group
K-e model,
In the latest
experimentally
often
equation
renormalization
to be a function
step,
parameter
algebraic
values.
strain
obtained
The standard
which
is set
to mean
over
important
constant
using
as the standard
and Thangam,
of dissipation
of the
form
by
of its
flows.
Taulbee
(1992).
There
els.
are other
For example,
a two-scale
terms
esting
Yoshizawa
direct
point
and
to developing
(1984)
interaction
of the mean
Rubinstein
approaches
velocity
Barton
in these
Reynolds
derived
a relation
approximation.
gradients.
(1990)
two methods
Yokhot
is that
relation
and
Orszag's
the values
algebraic
equation
mod-
stresses
using
It contains
A similar
using
stress
both
linear
and
of the model
quadratic
recently
method.
coefficients
An
by
inter-
can all
be determined
on the
analytically.
principle
of material
valid
are
strain
stress
not fully
turbulent
that
the
velocity
stresses
conditions,
of the
scale
model
shear
facing
step
flows
eta/.
(1978).
turbulence
vative
models
together
Modeling
simulated
type
sufficiently
and
free
final
studied
by Driver
has
Calculations
a second-order
fine grids,
numerical
data
the
realizability
carried
the
two
diffusion.
The
as well as with
backwardand
with
Kim
the
a conser-
differencing
solution
calculated
those
ho-
in validating
and bounded
that
they
rate.
rotational
(1985)
out
to ensure
functions
rotation
to the
and
up
In general,
Seegmiller
accurate
Reynolds
to be at least
rate.
the
is truncated
as a benchmark
are
is used
for the
to the mean
(1983)
As usual,
gradients,
form
strain
and
served
Reynolds
Using
are found
of flows
realizable
velocity
gradients.
eta/.
non-linear
relation
The
it is not
these
mechanics.
of applications
by Bardina
from
and
a constitutive
basis
hence
is
tested.
on the mean
relation
deriva-
frame-indifference
of rational
velocity
Oldroyd
In addition,
a general
theory.
based
turbulence,
to the mean
on the
with experimental
turbulent
of material
flows.
and
of Reynolds
Incompressible
continuity
then
turbulence
is made
method,
grid-independent
depend
expression,
the
extensively
method
invariance
scale
for complex
are compared
in detail
standard
K-e model.
2.
of the
latter
with
the
of the mean
experimentally
The
finite-volume
scheme,
both
flows
not been
in the obtained
of the time
validation
mogeneous
the
terms
ratio
scales,
contains
flows.
is to develop
stresses
by using
which
turbulent
with
Reynolds
quadratic
time
The
model
the coefficients
study
a different
incompressible
and have
and length
is derived
to tensorial
realizable
equation
proposed
the principle
for general
of the present
algebraic
we assume
However,
of two-dimensional
constraint
The purpose
(1987)
frmne-indifference,
rates.
in the limit
appropriate
models
Speziale
is
results
obtained
using
Stresses
flows
and Nsvier-Stokes
are governed
by the following
=o
v',,,+
Reynolds
averaged
equations:
(1)
= _r,,2
P
(2)
where
Ui are the
mean
velocity
respect
tensor
which
The
must
oldest
nesq's
viscosity
t and
(i = 1, 2, 3), p is the
and density,
co-ordinate
zj,
Ui,, and
respectively,
mean
pressure,
uiuj
is the
of Ui
turbulent
stress
be modeled.
and
isotropic
stresses
form
to time
components
simplest
proposal
eddy-viscosity
in laminar
flows
of this concept
for modeling
concept
and the
that
turbulent
the
turbulent
stress
assumes
an analogy
between
stresses
in turbulent
flows.
is Boussithe viscous
The
general
is
2
_iu$ = -_K 61j - 2vtSij
where
vt is called
the
eddy-viscosity
(3)
strain
rate
defined
by
1U,
Sij = _( ,j + Uj.,)
Equation
(3)
extensively
2.1
constitutes
used
(1970)
constitutive
discussed
those
on the
mean
problem
and time
flow
so that
field
developing
mean
flows
exists
turbulent
correlation
derived
relation
memory
we can
point
effects
of view
that
the
the
velocity
energy
models
that
turbulent
and
K and
stress
length
uluj
scales
its dissipation
wall.
In other
of walls,
are
with
if it exists,
is always
expansion
of turbulence
rate
aid
of the
the
smaller
than
conditions
situations
such
as
whether
however,
relation
of such
from
for any
a formally
of experiments.
of functional
form.
we neglect
at the present
of the functional
is a function
much
under
it is questionable
of application,
the relationship
only
correlation;
The validity
the
Lumley
boundary
a "constitutive"
problem.
and consider
are
and
derive
exist
of initial
the
to be verified
in the time
relations
the closure
correlations?
of turbulence
effect
far from
relation,
in the relation
turbulence
turbulent
such
scales
formally
of course,
constitutive
for
that
the
relation
to solve
needs,
A turbulent
a modeling
most
or in the vicinity
such a constitutive
of view,
exist
found
the length
significant
point
kinetic
and
is not
practical
Uij,
relation
turbulence
rapidly
there
this
in which
in the
for
relation
a general
situation
basis
today.
Constitutive
Does
a common
(4)
mean
characterized
form.
From
the
time
time
as the
Therefore,
deformation
by the
we
tensor
turbulent
e, i.e.,
(5)
5
Note
that
the molecular
here only
The
to high
Reynolds
axguments
According
viscosity
number
of equation
to the lr theorem
independent
v is not included
turbulent
(5)
contain
we restrict
our attention
flows.
ten
of dimensional
non-dimensional
because
quantities
analysis,
they
bearing
may
two
dimensions.
be grouped
into eight
quantities:
(6)
By normalizing
the
turbulent
stress
as
m
uiuj
vivjequation
(5)
can be written
2--K"
(7)
as
=
The
form
invaxiance
theory
be a function
have
tensors
of the tensor
shown
(Lumley,
of other
invaxiants.
with
the
The basic
B (also
tensor
principle
and
we may
is that
an invariant
a set of independent
1992)
its
Following
vectors,
see Shih,
_,j
relations.
arbitrary
function
In determining
A and
Cayley-Hamilton
two non-dimensional
isotropic
1978).
in Appendices
can be formed
generalized
valued
(8)
transpose
Lumley
form
V_,,Vh_A,
that
only
_,i
(1978),
th, _. following
Bj,
by using
can only
invariants,
we
18 independent
according
to the
let A_ and
Bj be
18 invariants:
Vh2,V_.jA, Bj,
(9)
In addition,
we have
other
invariants:
5_A_B_,
where
I, II, HI
A_AI,
invarlants
BiB,
I,
II,
of the tensor
III,
]_j:
...
(10)
I "- V/,i
1
_(_,d
Zl :
V_
jj-
_,j_,_)
(11)
111 = l(_,,Vj,jVk,k
-- 3_,,_,h_a
3.
and
-..
represents
Note
that
the
in the
invaria_ts
above
of this type
Cayley-HamUton
because
dent,
other
possible
hence
they
However,
the
invariants
in (9)
for
example,
_j_j.
we do not include
_,z_.kV;_jAiBj
quantities
and
according
theorem:
V_.tV_.kVh.j - I.
Any
17 tensors,
of other
llst of invariants
+ 2V,j_,_,,)
terms,
V_.kVkd + II.
for example,
V_.j - IIl&j
Vk.iVk.lV,,,.tV,,.iAiB
= 0
(12)
j, are
also not
indepen-
list
(10),
can
be extended
by
including
any
combination
of
for example,
f1(1,
VI,jAiBj"
1I,
III,
V_aVid,
...)
(13)
V_,kV_,.jA,Bj..f2(I,
where
.fa and
dependent,
(16)
As
.f2 are
but
scalar
they
are, in general,
a result,
invariants
listed
the
functions.
II,
Of course,
are useful
in explaining
functions
of various
invarlant
in (9),
v-_AiBj
_-_jAiBj
(10)
and
Ill,
Vi_iV_,i,
these
...)
types
of invariants
may
in question.
as a function
VLIAiB
j,
Vi,hVhjAiBj,
Vj,kVh,iAiBj,
VI,,Vo,
Vk,iV_jAiBj,
VlyiAIBi,
V_iVk,jAIBj,
K'.,%A,
Viyk_AiBj,
Bj,
A_A_,
of the
( 13):
f(Vi,jAiBj,
B_B_,
I,
11,
111,
V_,jV_._,
not in-
of the tensors
be written
are
...,
above
vTv'jA_Bj
right
equation
hand
(14)
is biline_r
side
can
in axbitrary
of equation
(14)
be reduced
_A_Bj
(14)
tensors
be
also
At and
bilinear
Bj,
we must
in At
and
Bj.
require
that
Therefore,
to
= a16tjAtBj
+ a2VijA_Bj
+a4V_.h VhjAtBj
+ asVj._A_Bj
+ asVj.k Vk.iA4Bj
+ aeV_,h Vj,kAiBj
+a,V_.,V_aA,B_ + asV,._Vj_A,B_
+ _.V,;_Vj,hA,
Bi
+a,oVk,tV_,2_liBj
+ aI,V_j, V_2_,VI_AtB
that
At and
B_ are
the
arbitrary
(15)
we obtain
_t-al2Y/,ky/,hy/_
-_- a l S gk,i
.._
V:l
al3Vi_/_
gl2h
.31- al4Yk',tYk',j
Vl2 j
9ff al,Yk,tYk,1521
3v a l S gi,_,
Vl,I,
Vl2rn
_?ra
the
coefficients
V_jV_j
...,
al -
(16)
due
to (13),
functions
of the
invariants
I, II,
i.e.,
ai = It(I,
Equation
an, are,
(16)
is a general
lI,
III,
V_iV_z,
relationship
...),
between
i=
1,2,..-,19
two second
rank
(17)
tensors.
The normalized
turbulent
stresses
have
two
properties
_;vj = vjv;
(18)
v_v_ = 1
(19)
and
Using
the
properties
of (18)
and
(19)
in equation
(16),
we obtain
the
following
relations:
aS
a12
a2_
QlS
g 1 =
[1
aS
--
a16
a4_
a9
=" a17
2a,I
- 2a4D
0-18
as_
Qll
"--- a19
QlO
(20)
- (a6 + aT)b
2(as
+ alo)/_
= V_jVI_j
(als
+ al,)b]
where
After
form,
D = VIjVj.,
b = V_jVI.j
introducing
equations
(20)
,
into
(16)
and
converting
= V_jV_d=
3
to the
(21)
dimensional
we obtain
K=
tti_t
K s
+2_4-_-(u,._u_
+ uj.ku_.,
- gnu,j)
K s
1-
2ae--_--(U,.kUj.k
gn6,_)
K s
2a.t--_-( Uk.iUhj
_ IISij )
K 4
+2as--_-(Ui,kU1j
. q- U_hUj,k -- gIISij)
K 4
+2alo---_(Uk.,Ul5
_k2als_s
(22)
2 :
+ U_jU_.,
(U_U_
2:
-gnu,
j)
1:
K 4
1--
q-2al4--_-(U_,iU_,
--
giI_ij)
where
n = U;,jUj,,,
_I = u,,w,,j ,
fI = v,_v2,j ,
9
fi = u.:.u.:.,_,
,_,
(23)
Equation
(22)
is the most
assumption
(5).
Interestingly
of equation
(22)
axe of the
scale
direct
method
interaction
when
smaller
however,
1990).
result
indicates
the
quadratic
form
2.2
Realizability
Reallzability
of equation
non-negativity
any
fluctuating
requirement
flow).
Lumley,
is a basic
stresses
physical
continuity
model
rate tensor
equation
(1)
from
equation
(22),
ulu---_
2K Since
the
time
scale
the
different
of equation
may
twogroup
theoretical
(22).
be sufficient,
time
espescale
is
consider
defined
and
Schwarz'
requirement
inequality
mathematical
obey.
from
as the
producing
to the constitutive
between
principle
It also represents
unphysical
relation
of
that
the
a minimal
results.
(22)
In
to derive
U2,2
0
0
0
0)
0
(24)
gives
the normal
1
KUI,_
3 + 2a2--e
ratio
both
side
of the form
U2,2
and
three
the
right-hand
the renormallzatlon
of (22)
and
should
0
U1,1
0
The
at the
U_j under
1978),
equation
a turbulence
a deformation
form
and
through
the
Therefore,
the following,
this principle
will be applied
constraints
on its coefficients.
Consider
and
the rationality
normal
model
to prevent
terms
derived
1984)
(which
(22).
of turbulent
turbulence
as those
tensorlal
1977,
quantities,
of any
five
unity
of mean
(Schumann,
the
solution
scale
between
first
(Yoshizawa,
is less than
the time
form
a quadratic
[JU_jJiK/e
than
same
the
Barton,
lead to a similax
In practice,
clally
and
relationship
enough,
formalism
(Ruhinstein
analyses
general
of the
-=
(25)
-U1,1
stress
1
+ _(2a4
turbulent
can be written
+ a6 + aT).
to the
mean
as
( KUI,_ _ =
/e
strain
rate
is defined
(26)
by
KS
= --
(27)
where
S = (2S_jSij)
10
I/=
(28)
equation
(26)
can further
Physically,
we know
in U1,1, but
be written
2K
that
cannot
as
3 -4-827/+
will decrease
be driven
by a vortex
to negative
(29)
values.
stretching
Therefore,
with
an increase
we must
require
that
UI_I
2K
>
0,
0,
if
0,
if 7/--,oo
_1R1
2K
ulul
These
are
of which
called
the
the
simplest
,,.
realizability
way
if
conditions.
is perhaps
the
0 < T/< oo
7/--.
They
(30)
oo
(31)
(32)
can be satisfied
in various
ways
following:
2/3
(33)
2G 2
A1 +_/
284
2as
(34)
/(7)
C.2
(35)
/(7)
G'.3
287
where
f(T/)
simplest
is in general
form
--
a polynomial
(36)
/(7)
of _/ of order
higher
than
2.
We
take
its
as
f()
A1,A_,G.I,G.2
and
C.3 are
adjustable
= A2 + _s
constants,
A1 > 0 ,
(37)
but
they
must
satisfy
A2 > 0 ,
(38)
2C'_1 -4- C,-2 + C',-s > 0
Similar
be mentioned
strain
rate
analysis
that
on _t2u2 and
equations
_
(33)
also leads
to (38)
tensor
11
to equations
also
hold
(33)
to (38).
for a three-dimenslonal
It should
pure
0
UI,I
0
and
that
principal
any
deformation
axes
of deformation
model
model
and
and
in the
weak
Reynolds
Further
deformation
sense,
can be
analysis
in the
realizabillty
such
models
(1990).
is, they
that
form
only
ensure
cannot
as those
of Speziale
In fact,
of (39)
in the
these
in the
(1987),
models
the positivity
be fully
satisfy
satisfied
standard
Yoshizawa
reallzability
K-e
(1984)
only
stresses.
constraints
on the
rate tensor
mode]
coei_cients
0
0
0
which
(39)
written
as constant,
anisotropic
Barton
that
O
0
1
Us,s
rate tensor.
are taken
recent
and
tensor
the above
coefficients
Rubinstcin
in the
rate
U,,
0
0
corresponds
0
UI.=
0
to a fully-developed
UlUl
2K
3
_3_3
flow.
In this case,
s)(2C's
- c.2
)
(41)
_'K
_K
we have
C_s)
.'K
] K + 3(A_-__
the
(40)
,/'K
+ 3(A, + ,f) (2C_,
by considering
0
0)
0
channel
2
_2_2
can be obtained
C_s)
2,1K
3(A1 -I_7)
where
KS
Experiments
indicate
= lUl,=l
s=
(42)
that
2
UlUl
>
u2u2
<
which
requites,
from
equation
(41),
that
12
(43)
2.3
Rotation
effect
The parameter
the
mean
where
_7 represents
rotation
_,
rate
In the present
the effect
x-
ij
ijJ
the rotation
study,
T/and
_ can be easily
2.4
coefficients
justified
algebraic
By introducing
rate,
and
the
effect
of
f_j
= (U,.j
- Uj,,)/2
+ 4emjlwm
(45)
it is sufficient
to simply
include
the paxameter
a2, i.e.,
the other
Realizable
strain
of the frame.
2/3
2a2 -keeping
mean
by _:
'
we find that
in the coefficient
while
of the
can be r _presented
12 =
t:
w,_ represents
only
(44)
equation
A1 + r/+
the same.
by equation
The dependance
of the
coefficients
on
(17).
equation
model
( 33)-(37)
into
2 g_
(46)
equation
(22),
we obtain
- v,(U,,j+ Uj,_)
-t A2 C-rl
+ 77a K_ s (U,.kUko
+ U.i:,Uk,,
II6,j)
(47)
U.,
K s
C.s
K s
1-
-eA, + ,? : (Uk,iUk,_
--_u&j)
Two
quantities,
be determined
standard
the turbulent
in equation
K-e model
(47).
are used
K:
kinetic
To this end,
which
+ [U.iK
energy
the
two
transport
rate
e, remain
equations
to
in the
read:
(!., + _)Kj]j
= P -
(48)
_r K
//t
_2
_.,
+ [u:- (v + _)_.j]._
= c,-_P- c_-_
13
(49)
where
g
2/3
C_,=
(50)
A_ +_+a_
P = --u, uiUi J
The
coefficients
C1,C2,o'K
and _
C1 =1.44,
and
the
additional
C,1=-4,
These
coefficients
have
their
C2=1.92,
ag=l,
C,3=-2,
been
standard
values:
_,=1.3
(52)
assume:
C,2=13,
values
assume
(51)
found
A1=1.25,
to work
well
_=0.9,
for both
A2--1000.
test
cases
(53)
considered
in this
work.
3.
Rotating
The
Homogeneous
present
model
turbulence.
A test
by Bardina
eta/.
the
flow being
energy
0,
K/Ko
0.5,
mean
were
-0.5
starting
with
the
initial
from
"equilibrium"
order
isotropic
three
cases.
c).
seen
nonequilibrium
to simulate
gives
the
data.
the effect
the
same
initial
used
results
from
of the
that
at
fI/S
of rotation
of turbulent
much
2(b)
and
rotation
as for the
on the evolution
14
present
0 the
better
2(c)
than
show
the
the
of turbulence.
the
cannot
very
up with
ability
case
and
model
s-K-e
was
LES results
energy
rate on turbulence.
no rotation
condition
model
present
catch
S is the
= 0.296
with
kinetic
it does
of f_/S
initial
eo/(SKo)
the
of
kinetic
energy,
are compared
However,
large
results
both
studied
The calculations
The
in LES with
to as s-K-e)
performs
Figures
rates
kinetic
on the
configuration
rotation
scheme.
was
turbulent
frame.
development
and
1 is the
of the
at the
Runge-Kutta
The
be
Figure
St,
rate
flow which
turbulent
referred
It can
shear
rotation
turbulence
(hereafter
mean
the evolution
time,
K0 is the
development
model
account
nondimensional
where
of the
method.
- c) show
turbulence.
overpredicts
model
2(a
effect
homogeneous
LES
an isotropic
highly
s-K-e
the
a fourth
to the
2(a
the
f2 is the rotation
K-e model
in figures
predict
the
standard
using
respectively,
corresponding
to mimic
Figures
rate and
Flow
is the rotating
(1983)
tested.
performed
adopted
is able
case
with
strain
Shear
the
model
of the
Note
because
well
later
which
present
that
the
it cannot
4.
Backward-Facing
4.1
Step
Numerical
procedure
For computational
tions
Flows
is solved,
convenience,
the non-dimensional
form
of the governing
equa-
in which
z_
U_
<x'>=r,,---7'
<u'>-u,,,'
<P>=pu2,----7'
(54)
< K >=
_L,, l
< e >-
u2, '
where
length
and velocity,
respectively.
< vt >=
u2,t'
quantity,
and
Accordingly,
ut
L,et,
V,,jL,,s
and
number
is defined
by
Re-
L,,tU,
ej
(55)
12
Hereafter,
dropped
steady-state
equations
(1),
and
(2),
two
(48)
and
1
stands
source
gradient
terms
equations
the
from
equation
same
volume
procedure.
It
mentum
usually
is handled
To ensure
both
and
associated
accuracy
are approximated
1991a),
all the
to solve
center
with
and
terms
system
with
control
all
grid,
algorithm
(Van
of numerical
accurate
and
conventional
15
the
volume
the non-staggered
by the
and
equations,
quadratic
the
velocity
non-dimensional
of equations
(1983)
stability
(56)
counterparts,
Chow
the SIMPLEC
form
except
that
by l/Re.
grid
and
general
dimensional
of each
of Rhie
by a second-order
other
the
transport
= s.
as their
the
12t
diffusion
12 is replaced
procedure
with
(47).
= y),
K and
a non-staggered
geometric
interpolation
oscillations
coupling
at the
U2),
form
used
uses
= z,xz
in the following
cross-derivative
viscosity
method
stored
can be written
V(=
arising
numerical
being
U1),
the
The
(zx
+ [v -
includes
molecular
cases
S,
kinematic
(49)
for 1, U(=
term
dimensional
v_
[uthe
so that
for simplicity.
In the
where
form
is used
and
solution,
central
is a finite-
dependent
(Figure
Doormal
bounded
(56)
variables
3).
The
to avoid
mo-
spurious
the pressure-velocity
and
the
Raithby,
convection
differencing
differencing
1984).
terms
scheme
scheme.
(Zhu,
As
a result,
lineexized
the discretized
form
counterpart
of equation
(56)
can be cast
A_e = A__+ Sc
(57)
where
the
coefficients
to its
neighbours
terms
of equation
the resulting
procedure
_bE(Figure
(56).
coefficient
of Stone
iterative
solution
residue
of all the
numerical
4.2
shows
scheme
matrix
is always
diagonally
is used
to solve
is considered
dependent
variables
are given
model
studied
from
the
the
relate
the
principal
discretization
of the
used
the system
is less than
eta/.
The strongly
the
(1989)
and
implicit
equations.
maximum
10 -4 . The
side
AI __ 0 so that
of algebraic
when
left-hand
that
dominant.
converged
in Rodi
ensures
unknown
The
normalized
details
of the
present
Zhu (1991b).
results
present
(1985),
from
convection
(1968)
procedure
imentally
3), result
which
The
process
Numerical
The
At (l = W, E, S, N),
by Kim,
here
flow
1 gives
the
flow
stream
velocity
is then
applied
Kline
on referred
and
parameters
and
and
Johnston
to as KKJ-
configuration
U,e!
to the two
the
for both
step
height
(1978)
and
the
taken
step
Driver
and
system
exper-
Seegmiller
Figure
used.
experimental
as the
flows
respectively.
co-ordinate
here
H0 are
and
DS-cases,
Cartesian
cases;
backward-facing
Table
reference
reference
free-
quantities
for
non-dimensionalization.
Table
case
Three
inlet,
the
types
the
L,
L_ 11. Hd U.S
DS
37423
1.5
10
40
KKJ
44737
0.6
10
40
outlet
and
of boundaries
assumption
normal
parameters
experimental
the turbulent
1. Flow
Re
data
stresses
are
present,
i.e.
inlet,
are
available
for the
streamwise
from
solid
wall.
At the
mean
velocity
U and
these
_-_ and
_'_ with
that
1
_-_ = _(_
and
(58)
+ _)
e by
e =
C_I4KSl2
L
'
L = min(0.41Ay,
16
0.0856)
(59)
where
Ay is the
in Table
1.
to zero.
At the
Influences
by changing
the
the
distances
interest.
of them
outlet,
the
inlet
Shih
was found
also reported
in Avva
in this
we use the
1974)
to bridge
Two
dence
eta/.
and
and
5(b)
s-K-e model
data
for the
seen
that
present
show
the
sufficiently
This
close
have
conducted
that
quadrupling
the present
DS-case,
data
results
Cray YMP
predict
of the
near-wall
is seen
from
the negative
turbulence
figure
peak
recirculation
The computed
are determined
in Table
5(a)
that
and
those
modeling
both
of Cy, pointing
findings
were
Therefore
and
Spalding,
only
the
this
to limited
Figures
with
KKJ-case.
for the
is also
results
the
Thangam
KKJ-case.
case
for the
are not
They
Hut
(1991)
have
found
improvement.
Therefore,
as grid-independent.
model
7.1 and
8.3 minutes
of CPU
is very
low Reynolds
and present
model
accuracy
of the
time
703
on the
to the
number
However,
in the
For the
required
are presented.
sensitive
restricted
of the
grids
and
and present
that
the
It can be
on the coarse
wall functions.
is only
for the
using
points
s-K-e
the various
s-K-e
(fine)
wall calculated
5(a)
a minimal
the
Recently,
C! is a parameter
than
but none
Lai (1988).
differences
obtained
to only
with
It is Cy that
worse
Similar
for the
can be considered
In the following,
much
(1992),
(Launder
in figure
for the
took approximately
modeling.
and Shih
points
KKJ-case,
stage.
grid leads
coefficient
Launder
and 199x91
some
calculation
computer.
(1981),
available
the solutions
of
and Bremhorst
(fine)
produce
a highly-resolved
a 166x73
(coarse)
are
in the
that
region
number
approach
also included
does
noticeable
the
cases,
Reynolds
C! at the bottom
to the grid-independent
in the
model;
indicates
from
110x52
but no such
more
results.
in both
coefficient
grid refinement
are set
are examined
low
and So and
function
and 201x109
the friction
far away
severa]
given
variables
that
solutions.
(1992)
computational
model,
s-K-e
sublayer
(coarse)
DS-case,
wall
flow
found
and Yang
satisfactory
Shuen
contain
Lam
(1992),
to yield
standard
they
106x56
sufficiently
we tested
(1982),
thickness
on the solution
it has been
work,
(1990),
viscous
of the solution;
and
and Lumley
of the
conditions
already
of Chien
sets of non-uniform
KKJ-case
5(a)
the
Le,
of this
to be able
derivatives
and outlet
1 are
stage
those
(1974),
streamwise
Lo and
in Table
earlier
including
work,
of both
given
Sharma
from
locations
In the
K-e models
and
distance
K-e models
the
near-wall
largely
near-
influence
regions.
It
underpredict
wall function
approach
region.
and
measured
in the calculation
reattachment
points
17
where
obtained
are compared
in Table
Cy goes to zero.
with the Reynolds
2. They
Also included
stress
model
(RSM)
and
that
reattachment
of Sindir
point
overall
performance
dearly
demonstrates
It is important
makes
the
that
model,
and that
a marginal
pressure
coefficient
predict
C_ along
premature
reattachment
well.
The
and
Sindir
almost
lengths,
good
The streamwise
mean
present
model
model
predicts
are
Here,
recovery
in the vicinity
recovery
also exists
say at z-20
other.
Finally,
u--Oare
perimental
and
7(a),
the comparisons
in figures
data
for the
continuously
uncertainty
some
uncertainty
other
reliable
a smaller
hand,
because
step
in the
the
unsteadiness
measured
smaller
of the
in both
works
in figures
between
quite
of Obi et a/.
present
to
of the
rises
(1989)
model
are
predictions
s-K-e,
but
results
DS-case
with
slower
such a slow
Further
downstream,
with
turbulent
stresses
u 2, v 2 and
In the
in the
KKJ-case,
of the
length.
should
also
recovery
region,
and back-
step,
This
in the
the
forward
leaving
an
points
to
region.
be considered
location,
s-K-e
results
each
no ex-
recirculation
to move
of the reattachment
As compared
The present
coincide
downstream
in the
and
nearly
experiment
quantities
s-K-e
enough,
(1989).
models
reattachment
in somewhat
Interestingly
z-locations.
heights
7(a)
flow variables.
are available
18
is seen
pressure
the results
to other
the
uncertainty
flow.
these
are shown
turbulent
data
s-K-e
static
of C,.
in the
for the
the
the improved
stresses
experimental
of the
of
stresses
measured
Again,
various
step
cases,
captures
and measured
was found
and
and ASM.
of the two
9 at
height
behavior
of Obi eta/.
the results
seven
model.
of the
point.
prediction
turbulent
around
of 4-0.5
than
of predicted
point
In both
as compared
8 and
ASM
5.66
results
differences
better
in the RSM
present
The
of the reattachment
the reattachment
ward
the
flows
in figure
shown
ASM.
U profiles
the
of computed
reported
variation
not substantial,
reverse
6.44
model
and
to the
velocity
cross-sections.
7.35
is consistent
of the RSM
attributed
the
Table
of the turbulent
points
RSM
wall.
present
RMS
to those
of C_ are mainly
which
the
the
behavior
comparison
of C_ were
using
comparable
different
while
the
due to the
of reattachment
present model
5.82
bottom
rises,
predictions
(1982),
the
the
pressure
The
to assess
to it.
6.35
show
(ASM).
used
procedures.
with
is mainly
the anisotropic
model
been
obtained
this improvement
7 4-0.5
6(b)
stress
often
as well as numerical
Table 2. Comparison
experiment
s-K-e
6.1
4.99
and
has
improvement
contribution
KKJ
6(a)
significant
algebraic
which
models
to mention
case
DS
Figures
with a modified
parameter
of turbulence
O, in the present
only
(1982)
is a critical
On
more
indicating
in figures
8 and
9, it can be seen
leading
the
to significant
anisotropic
behaviours
the
the
anisotropic
improvements
terms
are
that
have
clearly
flows considered
here.
in both
little
reflected
impact
on the
v 2 results.
shear
which
also hold
obtained
decreasing
On
turbulent
(41)
The improvement
u s while
v 2,
the
other
hand,
stress
_-_.
These
qualitatively
by the present
model
for
in figure
in Cu.
Conclusions
A constitutive
ance
theory.
because
relation
analysis.
can
be expressed
classical
approximation.
The
_, which
ensure
condition
that
has
been
shear
flows
and
are
cases,
has
little
but
impact
calibrated
the LES
significant
is achieved
coefficients
the
vaiue
given
other
The
step
The
improvement
of the latter.
computed
homogeneous
flows.
at an insignificant
simplicity
This
shear
penalty
inhomogeneous
shear
the
far downstream
of the reattachment
simple
parabolic
flows,
model
for simple
region
show
and
that
K-e model,
can
as evidenced
nature.
19
point
also
that
all flow
normal
gradients.
terms
C_1 may
be further
compared
in detail
the experimental
the
The
for both
quadratic
present
data
model
does
be expected
by its improved
where
homogeneous
appropriate
been
present
turbulent
velocity
have
r/ and
The
show
the
that
ap-
- a realizability
rotating
quite
to the computational
present
ratios
only
flows
The
scale
cross-derivative
results
comparisons
time
results
up
a first-order
to satisfy.
indicates
stress
quadratic
flows:
in
of order
only
stresses
seem
to the
considered.
turbulent
the
mean
in this paper
numbers
to use
unable
that
invari-
account
gradients,
normal
and
into
of the
of the
nonlinear
of C1 related
flows.
suffice
The calculated
containing
Reynolds
constitutes
are
of C_,
by using
taken
form
two different
flows.
at high
velocity
turbulent
models
variation
to the terms
mean
functions
the
step
for backward-facing
provide
to calculate
on the flows
against
are
derived
not been
it may
turbulence
to the
flows
representation
of the
backward-facing
been
general
of the
coefficients
existing
sensitive
has
has
the
calculations,
positiveness
applied
of the model
test
with
the
are sensitive
the
tensor,
eddy-viscosity
model
most
model
values
rank
For practical
proximation.
stresses
a second
viscosity
as a s, ries, in terms
the
stresses
of the molecular
Being
to 4, while
variables
The relation
the influence
the
mic
increase
u s and
in equations
5.
terms
and
algorith-
to work
well
prediction
in
to be of
Acknowledgements
The
authors
are grateful
ing homogeneous
shear
of J.L.
Lumley
0226,
funded
by the
Aerospace
No.
Programs),
Shabbir
and
for may
helpful
was
U.S.
supported
Airforce
F49620-92-J-0038,
(Aerospace
Aaralr
flows
contribution
jointly
to Dr.
funded
in part
Office
the
U.S.
of Scientific
No.
(Control
and in part
Office
The
AFOSR
Research
Research,
Airforce
rotat-
and discussions.
by Contract
Office of Naval
by
of the
89and
by Grant
of Scientific
Research
Program).
References
R.K.
Awa,
C.E.
J. Bardina,
models
lent
J.H.
based
flows."
K.Y.
number
1982,
D.M.
Driver
shear
layer
P.G.
Huang
and
Cornell
.
M.A.
with
S.J. Kline
reattachment
"Features
1985,
second
turbulence
incompressible
Stanford,
flows
turbulent
pp.163-171.
of recirculating-flow
closures
on
with
pp.33-38.
of a reattaching
"Stabilization
turbu-
Ca.
J., Vol.20,
5th Svraposium
of high
90-0246.
"Improved
J., Vol.23,
moment
paper
boundary-layer
1985,
AIAA
study
and
third
Turbulent
order
Shear
dis-
Flows,
pp.5.19-5.24.
and
J.P.
of a turbulent
MD-37,
and
AIAA
flow",
of the
1983,
model",
Leschziner,
AIAA
University,
of channel
channel
"Comparative
of homogeneous
Stanford
SeegmiUer,
Proceedings
University,
J. Kim,
Rept.
and
Reynolds,
simulation
turbulence
H.L.
1990,
of K-e models",
W.C.
"Predictions
performed
cretization',
and
No.TF-19,
in divergent
computations
Singhal,
versions
Ferziger
low-Reynolds-number
.
and A.K.
on large-eddy
Rept.
Chien,
Smith
Johnston,
shear
Thermosciences
1978,
layer:
Div.,
"Investigation
Flow
Dept.
over
of Mech.
of separation
a backward-facing
Eng.,
Stanford
and
step",
Univer-
sity.
.
C.K.G.
predicting
Lam
and
K.
Bremhorst,
wall turbulence",
1981,
J. Fluids
2O
"A modified
Eng.,
Vol.103,
form
of K-e
pp.456-460.
model
for
8.
B.E.
Launder
model
9. B.E.
and Mass
Launder
M.A.
Leschziner
and
various
Eng.,
Vol.1,
1970,
"Application
of the
of a flow near
energy-dissipation
a spinning
disk",
Letters
pp.131-138.
Spalding,
1974,
Meths.
App.
W. Rodi,
1981,
discretization
Vol.103,
J.L. Lumley,
Vol.41,
"The
Mech.
numerical
Eng.,
"Calculation
schemes
computation
Vol.3,
pp.269-289.
of annular
and turbulence
of tur-
model
variations",
J.
pp.352-360.
"Toward
a turbulent
constitutive
relation."
J. Fluid
Mech.,
pp.413-434.
J.L. Lumley,
Mech.,
13.
D.B.
Cornput.
Fluids
12.
and
1974,
calculation
transfer
flows",
using
Sharma,
to the
bulent
jets
11.
B.I.
of turbulence
in Heat
10.
and
1978,
Vol.18,
"Computational
modeling
of turbulent
flows",
Adv.
Appl.
pp.124-176.
14. R.S.
Report.
15. C.M.
276/N/89,
Lehrstuhl
ffir StrSmungsmechanik,
_it Erlangen-Nfirnberg.
Rivlin,
isotropic
LSTM
1955,
"Further
materials",
Rhie
J. Arch.
and W.L.
an isolated
airfoil
remarks
Chow,
with
Ratl.
on the
Mech.
stress
Anal.,
trailing
edge
deformation
Vol.4,
study
separation",
relations
for
pp.681-702.
of the turbulent
AIAA
flow past
J., Vol.21,
pp.1525-
1532.
16. W.
Rodi,
state
1980,
Hydraulic
17. W. Rodi,
Research,
incompressible
App.
Eng. , Vo1.75,
and J.M.
the
renormalization
U.
Schumann,
Phys.
Fluids,
1977,
Vol.20,
and
Publication
flows
application
in hydraulics
of the International
- A
Association
for
1989,
with complex
"Finite-volume
boundaries",
method
Cornput.
for two
Meths.
pp.369-392.
Barton,
group",
their
the Netherlands.
and B. SchSnung,
dimensional
Mech.
models
Book
Delft,
S. Majumdar
18. R. Rubinstein
19.
"Turbulence
Phys.
1990,
"Nonlinear
Fluids
"Realizability
21
stress
models
and
A 2, pp.1472-1476.
of Reynolds
pp.721-725.
Reynolds
stress
turbulence
models",
20. T.H.
Shih,
NASA
21.
1992,
T.H.
Shih
lence
and
and J.L.
on turbulent
Lumley,
its application
Int.
J. Comput.
J.S.
Shuen,
M.
"Remarks
constitutive
relations."
to appear
in
TM.
Sindir,
Dynamics,
Private
1982,
step
behavior
modeling".
NASA
of near-wall
turbu-
TM
also in
105663,
Vol.1.
communication.
"Numerical
backward-facing
"Kolmogorov
in turbulence
Fluid
1992,
1992,
study
of separating
geometry",
Ph.D.
and
Thesis,
reattachin8
University
flows
in a
of California
at
Davis.
24.
R.M.C.
So and Y.G.
a backward-facing
25.
26.
C.G.
Speziale,
Fluid
Mech.,
C.G.
Speziale
model
27.
H.L. Stone,
30.
1987,
D.B.
"Low-Reynolds-number
J. Appl.
"On
Vo1.178,
1968,
flows",
"Iterative
1992,
nonlinear
S. Thangam
NASA
K-e
of flows
Vol.39,
models
over
pp.13-27.
of turbulence",
SIAM
"An improved
model",
J.
for predicting
G.D.
Raithby,
1984,
incompressible
No.92-3.
Vol.5,
J. Eng.
Num.
and corre-
pp.2555-2561.
study
of turbulent
Sci., Vol.29,
"Enhancements
fluid flows",
pp.530-558.
model
No.ll,
numerical
Int.
turbulence
of multidimen-
stress
A, Vol.4,
"A hlghly-resolved
Doormal
Report.
Anal.,
Reynolds
Fluids
step",
based
approximations
J. Num.
algebraic
Phys.
J.P.
and
of an RNG
ICASE
of implicit
equation",
stress
"Analysis
CR-189600,
solution
method
(ZAMP),
K-1 and
1992,
separated
615.
Van
Phys.
modelling
pp.459-475.
differential
Taulbee,
Math.
nonlinear
and S. Thangam,
partial
sponding
29.
step",
for separated
sional
28.
Led, 1988,
of the
Heat
pp.607-
SIMPLE
Trans.,
Vol.7,
pp.147-163.
31.
Z. Yang
and
T.H.
wall turbulence",
32.
A. Yoshizawa,
from
33.
Shih,
NASA
1984,
its eddy-viscosity
J. Zhu,
1991a,
Comm.
App.
1992,
TM
"Statistical
new time
Methods.,
scale
based
K-e
model
for near
105768.
analysis
representation",
"A
of the deriation
Phys.
Fluids,
and osciUation-free
Vol.7,
22
pp.225-232.
of the Reynolds
Vol.27,
convection
stress
pp.1377-1387.
scheme",
34.
J. Zhu,
1991b,
two-dimensional
Rept.
No.690,
"FAST-2D:
A computer
incompressible
Institute
flows
program
with
for Hydromechanics,
23
complex
for numerical
simulation
boundaries",
University
of Karlsruhe.
of
Appendix
Generalized
Cayley-Hamilton
Rivlin
(1955)
showed
that
relating
matrices
(product
matrices
of lower
extension.
ABC
+ ACB
-B(trCA
-(Be
-$rC
Repl_ing
trB
trC
trAB
Some
of them
- trA
+ CAB
C(trAB
trCBA)=
ABA=-A_B
here
+ CBA
trA
trB)
of higher
for latter
A(trBC
trB
formulas
extension
to
use.
trB
- CAB + BA)trC
trBC
B in Eq.(A.1),
Cayley-Hamilton
A, B, C ...)
are listed
+ AC)trB
trABC
generalized
matrices
+ BAC
trA)
A and
are several
of several
- (CA
trC
C with
there
+ BCA
+ UA)trA
-l(tt"A
Formulas
trC)
(A.1)
trCA
0
respectively,
BA _ +
A(trAB
we obtain
-
trA
trB)
1
+_
B(trA
z -
trA
trA)
(AB+
BA)trA
A_trB
(A.2)
1
+I[trA2B
trA
trAB
+ _trB(trA
trA-
trA2)]
B(trBA
trA)
(BA
+ AB)trB
sad
BAB=-B_A
+21 A(trB2
AB 2 +
trB
trA)
trB
B_trA
(A.3)
i_dieate
by polynomials
that
the
of matrices
matrices
ABA
of extension
and
BAB
2 or less.
24
t Bof extension
trB )]
3 can be expressed
by B, and
= ABA
trA+
A 2 trAB
+ A(trASB
trA
trAB)
(A.4)
-B
detA
l detA
trB
and
BAB
s + BsAB
= BAB
trB
B s trBA
+ B(trB2A
trB
trBA)
(A.5)
-A
Replacing
detB
B with
I detB
trA
B 2 in Eq.(A.4)
ABSA
and
s + AsB_A
A with
= ABsA
A s in Eq.(A.5)
trA
A s trAB
give
s
(A.6)
+A(trA=B
_ -
trA
trAB
s)-
B s detA
I detA
trB s
and
BA2B
2 + BsAsB=
BAsB
trB
B s trBA
s
(A.7)
+B(trBSA
Replacing
B with
s -
trB
B 2 in Eq.(A.2)
AB2A=-A2B
s -
trBA
and
s)-
A s detB
A with
BSA 2 +
I detB
A s in Eq.(A.3)
A(trAB
2 -
trA
trA s
yield
trB s)
1
+_
BS(trA
+I[trA2B
= 2 -
trA
trA
trA)
trAB
(ABS+
BSA)trA
_ + ltrB2(trA
trA-
A=trB
(A.8)
trA2)]
_5
and
BA2B=-B=A
A2B 2 +
B(trBA
trA s)
(BA'
s -
trB
trA s)
1
+2
A'(trBS
+I[trBSA
Eqs.(A.8)
be expressed
hand
sides
trB
and (A.9)
2 -
trB
indicate
by polynomials
of Eqs.(A.6)
trBA
that
1
2 + _trA
the matrices
of matrices
and (A.7)
+ A=B)trB
2
(trB
trB
AB2A
of extension
2 or less.
25
trBS)]
and
BASB
2 or less.
(A.9)
of extension
3 can
B2trA
Therefore,
of matrices
the
of extension
right
of
Appendix
Number
of Independent
Let
us show
tensor
A sad
Rivlin
the
number
Formed
by Two
of independent
Tensors
tensors
formed
with
two
general
B is 18.
(1955)
u
a m_trix
that
Tensors
showed
that
a polynomial
product
Luy
m,_trlx
in these
II of extension
product
matrices
in two
of extension
3 x 3 matrices
4 or less.
may
Suppose
we have
5:
II = ABASB2A
This
cam be written
C = BA=B=.
From
in A m_l
C of extension
3 so that
II may
Therefore,
byA
(B.2)
II may be viewed
2 or less.
C itself
is a matrix
by a polynomial
to consider
as a polynomial
in A and
in A and
the possible
tensors
of matrices
B of extension
B of extension
of extension
4 or less.
4 or less formed
B.
We may
possible
ACA
Eq.(A.2),
be expressed
we only need
ud
(B.1)
as
II--
where
be
show
tensors
that
there
are
of extension
only
4 are
ABA=B
two
independent
the following
2, BAB2A
tensors
of extension
4.
The
8 tensors:
2, A2BAB
=, B2ABA
2,
(B.3)
AB2A_B,
With
Eq.(AA),
Eq,(A.5),
+
...;
B=ABA=
with
A=B2AB
where
A=BAB=
...
only two
Eq.(A.6),
ABA_B
represents
tensors
BA_B2A,
can
= -BAB2A
BA2B2A
= + ...;
with
L polynomial
of extension
A2B=AB,
be expressed
= + ...;
=
by ABA_B
with
-BAB2A
Eqs.(A.4)
Eq.(A.7),
2 + ...;
and
in A and
B2A2BA.
(A.5),
with
Eqs.(A.5)
B_A=BA
B of extension
are independent,
ABA2B
BAB_A
Similarly,
AB=A=B
4 in Eq.(B.3)
2,
....
-ABA2B
and
= BAB=A=
3 or less.
and
with
=
(A.4),
...;
As a result,
we select
them
as
(B.4)
26
OF POOR
QUALITY
Now
possible
we show
tensors
ABA
Using
that
are only
of extension
2, A2BA,
Eqs.(A.4),
independent.
there
Let
BAB
(A.5),
(A.6)
us select
them
ABA
Furthermore,
there
AB,
and
four independent
tensors
and
two general
we have
proved
tensors
of extension
3.
The
8 tensors:
2, A2B_A,
BA2B
(A.T),
we find
2, AB2A
2, BA2B
that
=, B=A=B.
only
four
(B.5)
of them
are
as
2, BAB
are eight
BA,
AB2A
independent
AB=,
B_A,
that
tensors
A2B,
of extension
A,
Therefore,
four independent
of extension
(B.6)
2:
BA 2, A2B _, B=A=
(B.7)
1:
A s, B,
only
2.
B 2.
18 tensors
tensors.
27
(B.8)
can be formed
independently
by
Figure
lO
1. Rotating
,,,,,,,,.,,
homogeneous
shear
_,,,I,,_1
(o)o/s=o
(b)
0/S=0.5
0
",,t"
_'_1_''_1
'
(:) o/s=-o.5
'-
/
I
/
I
/
4
flow
/
J
J
v_
....
10
....
I,
10
|
i
--._
st
Figure
2. Evolution
-----:
s-K-_;
of turbulent
--:
kinetic
present
28
energy
model;
,:
with
LES.
time.
10
,.
A
i
A
_W
Figure
3. Typical
control
volume
centered
at C and
Hd
Hs
Ls
Figure
4. Backward-facing
2g
step
geometry
related
nodes
IIl"WllVV|ll.lVrlWl|f.llWll.lJlJlVl.
WWl
0
(a) DS-case
-1
-2
ltJttitlJ_ll|ttlJl|JllllllJlJJt|t|lJtJl
10
20
30
40
X
3
,IWlllllVJflWlrlll|Jg||fllllW]rlV|lVrWl
O
(D
,/
0
(b)
KKJ-case
-1
-2
A |
10
it
20
J _
30
40
Figure
5.
- --:
m
Friction
s-K-e,
__:
fine
present
coefficient
grid;
model,
C/along
_:
present
co_rse
grid;
3O
the
bottom
model,
wall.
fine
e: experiment.
grid;
.2
....
'
'
'
'
'
....
'
'
'
'
'
'
'
'
'
'
'
'
'
'
'
'
'
'
'
'
'
'
'_
0.1
0.0
(a)
-0.1
.........
' .........
OS-case
' ..................
10
20
30
40
0.4
0.,,3
CL
(J
0.2
0.1
0.0
-0.1
0
X
Figure
6. Static
-----:
pressure
s-K-e;
--:
coe_cient
present
31
C'p along
model;
the bottom
e: experiment
wall.
X =20
3
|l
I
2
.
IJl
,,lllL
1.0-0.5
0.0
0.5
ii
1.0
X=2.667
3
X=5.333
i
''''1
....
X=6.220
i"_
rill
,,,,|,,,l_,_
''''I'''''P'_I
i,,
''.
(b) KKJ-cas4
"
tt
j-
./."
_ .
,i_lii,,l,liil,
0
-0.50.0
'
0.5
1.0-0,5
__
'lllll'lllll
0.0
li
ilia|l!
-----:
0.5
7. Streamwise
s-K-e;
||1
v
1.0-0.5
0.0
0.5
1.0-0.5
Figure
--:
mean
present
32
velocity
model;
U-profiles.
,: experiment
0.0
0.5
1.0
i|,|1,,,1|,..11,|,1
.|
v|
..
'"X=20
_1111|111111
0
3
.
X=8
w|l|,,,,ll.,.l.,,|
0
3
....,,.,.|..,.,l..
100uu
Figure
---:
'
'
'
'
'
2 -0.5
100vv
8. Turbulent
s-K-e;
'
........
'''''''''l'''''''''
_:
stress
present
33
0,0
0.5
- 100uv
profiles
model;
in DS-case.
,: experiment
1.0
''I''''I
....
I ....
..*.-,.-r'TT,,,I,,,,I,,,
''1
....
I ....
I ....
''1''''1''''1'''
'''_l''''l''''l'''
X=2.667
2
>,
lOOuu
Figure
-----:
lOOw
9. Turbulent
s-K-e;
_:
0.0
0.5
-lOOuv
stress
profiles
present
model;
34
4 -0.5
in KKJ-case.
*: experiment
1.0
1.5
Form
REPORT
Public mporth_
gathering
and
DOCUMENTATION
PAGE
per
of
collection
of information,
including
suggestions
for reducing
this burden,
to Washington
Davis
Highway,
Suite
1204,
Arlington,
VA
22202-4302,
anti to the Office
of Management
1.
AGENCY
USE
ONLY
(Leave
blank)
2.
REPORT
response,
information.
including
Send
the time
comments
for reviewing
instructions,
regarding
this burden
3.
REPORT
TYPE
AND
1993
DATES
5.
Reynolds
Stress
Algebraic
Equation
leta sources,
aspect
of this
1215 Jetferson
20503
COVERED
Technical
searching
existing
estimate
or any other
Headquarters
Services,
Directorate
for information
Operations
and Repo_ls,
and
Budget,
Paperwork
Reduction
Project
(0704-0188).
Washington,
DC
DATE
January
Approved
OMBNo.0704-0188
Memorandum
FUNDING
NUMBERS
Model
WU-505-62-21
6.
AUTHOR(S)
Tsan-Hsing
Shih,
7. PERFORMING
Aeronautics
Research
Cleveland,
g.
Zhu,
ORGANIZATION
National
Lewis
Jiang
John
NAME(S)
and
Space
L. Lumley
AND ADDRESS(ES)
g. PERFORMING
ORGANIZATION
REPORT NUMBER
Administration
E-7525
Center
Ohio
44135-3191
SPONSORING/MONITORING
National
and
AGENCY
Aeronautics
Washington,
and
D.C.
NAMES(S)
Space
AND ADDRESS(ES)
10. SPONSORING/MONITORING
AGENCY REPORT NUMBER
Administration
NASA
20546-0001
TM-
105993
ICOMP-92-27
CMOTT-92-14
11.
SUPPLEMENTARY
Tsan-Hsing
NOTES
Transition,
NASA
Lewis
and John
L. Lumley,
Research
Cornell
for Computational
Center
(work funded
University,
121l, _)ISTRIBIJTIONIAVAILABILI'I'Y
Ithaca,
New
Mechanics
by Space
York.
in Propulsion
Act
Agreement
Resixmsible
and Center
C-99066-G).
person,
for Modeling
Space
Tsan-Hsing
Shih,
STATEMENT
of Turbulence
Act Monitor:
(216)
and
Louis
A. Povinelli,
433-5698.
12b.
DISTRIBUTION
CODE
Unclassified-Unlimited
Subject
13.
Category
ABSTRACT
The
(Msxlmum
invariance
flows.
The
gradients
the
in continuum
leads
ratios
of
On
of
each
turbulent
form
the
the
basis
of
step
volume
Grid-independent
method.
second-order
for
both
mean
homogeneous
SUBJECT
and
capability
shear
a set
to the
strain
on
flows
rotation
that
are
and
quantities.
K-e
based
with
taken
fine
large
rate.
and
These
coefficients
existing
turbulence
The
diffusion-free
comparison
shows
rotation
rates
which
results
that
rate.
are
are
the
present
the
conventional
performed
by
eddy
model
viscosity
approximation
are
with
each
lS.
component
flows
of
over
finite-
schemes
the experimental
significantly
improves
rotational
to simulate.
NUMBER
OF PAGES
36
modeling
A03
17.
SECURITY
CLASSIFICATION
OF REPORT
Unclassified
NS'N 7540-01-280-5500
18.
SECURITY
OF
THIS
CLASSIFICATION
PAGE
Unclassified
19.
SECURITY
OF
CLASSIFICATION
20.
LIMITATION
OF
ABSTRACT
ABSTRACT
Unclassified
StanOard
Form
Prescribed
29e-lO2
by
ANSI
298
(Rev
Std.
of
data
the
to simulate
fail
the
the
a conservative
with
models
of
ensure
differencing
is able
of
fuctions
will
of
using
turbulent
velocity
Separated
in detail
proposal
proposed
linear
which
positivity
are
TERMS
Turbulence
the
to satisfy.
compared
the
number
mean
coefficients
fail
obtained
to the
obtained
The
ensure
calculations
In addition,
all
will
is just
are
models
solutions
calculated
models.
rotation
Reynolds
stresses
model
coefficients
mean
in high
Reynolds
viscosity
model
the
stresses
the
eddy
The
equation
Reynolds
relates
of
as applications.
grids.
The
two
rate
mosl
numerical
sufficiently
turbulent
of
mean
mean
configurations
accuracy
predictive
isotropic
analysis,
realizability
backward-facing
that
classical
the
the
to analyze
relation
realizability
turbulence
energy--
is applied
constitutive
in which
of
component
kinetic
mechanics
to a turbulent
general
form.
scale
words)
theory
in a more
positivity
14.
200
analysis
general
time
the
02
Z39-18
2-89)