Heat-Stress Resistance & Aging By: Scott Scholz Phillips Lab ’09 Mentor: Rose Reynolds.
Reynolds Stress ARSM
-
Upload
mohamed-yassin -
Category
Documents
-
view
226 -
download
0
Transcript of Reynolds Stress ARSM
-
8/10/2019 Reynolds Stress ARSM
1/114
Helsinki University of Technology. Laboratory of Aerodynamics.
Series B
Teknillinen korkeakoulu. Aerodynamiikan laboratorio.
Sarja B
Espoo 2001, FINLAND
IMPLEMENTING AN EXPLICIT ALGEBRAIC REYNOLDS STRESS
MODEL INTO THE THREE-DIMENSIONAL FINFLO FLOW SOLVER
Report B-52
Ville Hmlinen
TEKNILLINEN KORKEAKOULU
TEKNISKA HGSKOLAN HELSINKI UNIVERSITY OF TECHNOLOGY
-
8/10/2019 Reynolds Stress ARSM
2/114
-
8/10/2019 Reynolds Stress ARSM
3/114
Helsinki University of Technology. Laboratory of Aerodynamics.
Series B
Teknillinen korkeakoulu. Aerodynamiikan laboratorio.
Sarja B
Espoo 2001, FINLAND
IMPLEMENTING AN EXPLICIT ALGEBRAIC REYNOLDS STRESS
MODEL INTO THE THREE-DIMENSIONAL FINFLO FLOW SOLVER
Report B-52
Ville Hmlinen
Approved by Seppo Laine
Helsinki University of Technology
Department of Mechanical Engineering
Laboratory of Aerodynamics
Teknillinen korkeakoulu
Konetekniikan osastoAerodynamiikan laboratorio
-
8/10/2019 Reynolds Stress ARSM
4/114
Distribution:Helsinki University of TechnologyLaboratory of AerodynamicsP.O.Box 4400FIN-02015 HUTTel. +358-9-451 3421Fax +358-9-451 3418
Ville Hmlinen
ISBN 951-22-5605-3ISBN 951-22-6824-8 (PDF)ISSN 1456-6990
Printed in OtamediaEspoo 2001, FINLAND
-
8/10/2019 Reynolds Stress ARSM
5/114
iii
Abstract
This thesis describes the implementing of an explicit algebraic Reynolds
stress model (EARSM) into the three-dimensional FINFLO flow solver. The
EARSM replaces the Boussinesq eddy-viscosity assumption by a more
general constitutive relation for the second-order correlation in the Rey-
nolds averaged Navier-Stokes equations.
The thesis begins by describing the Navier-Stokes equations, in both their
time-accurate and Reynolds averaged form. Some general aspects of tur-
bulence are also discussed and a differential Reynolds stress model is pre-
sented. Then an algebraic Reynolds stress model (ARSM) is formulated
and simplified to its explicit form. After that some programming aspects of
the model are discussed. Finally, four test cases are selected, calculated
and analysed to verify the correct implementation of the model.
As the first test case a two-dimensional boundary layer over a flat plate is
calculated to verify the connection between the EARSM and the rest of the
FINFLO. The second order effects of the model are studied with the secondtest case, a fully developed flow inside a rectangular duct. Thirdly, a fully
developed flow inside a rotating pipe is calculated to examine the third or-
der effects of the model. As the last test case an axial flow over a cylinder,
part of which rotates, is calculated to further study the behaviour of the
EARSM when a sudden strain component is applied to the flow field. From
the results of the test cases it is then concluded that the EARSM has been
correctly implemented. The model gives significant improvements in most
cases compared to standard eddy-viscosity models without a large in-
crease in the computational effort.
-
8/10/2019 Reynolds Stress ARSM
6/114
-
8/10/2019 Reynolds Stress ARSM
7/114
v
Contents
Abstract .................................................................................................... iii
Contents .................................................................................................... v
Nomenclature.......................................................................................... vii
1 Introduction......................................................................................1
2 Governing equations and eddy-viscosity turbulence models ......3
2.1 Navier-Stokes equations ........................................................3
2.2 Reynolds averaging................................................................5
2.3 General facts of turbulence.....................................................62.4 Eddy-viscosity turbulence models...........................................8
2.4.1 Algebraic models.........................................................9
2.4.2 One-equation models................................................10
2.4.3 Two-equation models................................................11
3 Differential Reynolds stress models ............................................15
3.1 Reasons for using the Reynolds stress models ....................15
3.2 Exact equation for the Reynolds stress tensor......................16
3.3 Modelling the Reynolds stress tensor ...................................18
4 Formulation of an explicit algebraic Reynolds stress model .....21
4.1 Algebraic Reynolds stress model..........................................21
4.2 Explicit algebraic Reynolds stress model..............................26
4.3 Simplified EARSM for two-dimensional mean flow................29
4.4 Simplified EARSM for three-dimensional mean flow.............31
4.5 Solution of the general quasi-linear ARSM equation.............33
5 Programming approach and written subroutines........................37
5.1 Subroutine EARSM1 ............................................................37
5.2 Subroutine TTIMES..............................................................41
5.3 Subroutine ANISFLUX..........................................................42
5.4 Subroutine FLUXP................................................................43
5.4 Subroutine PEEKOO............................................................43
5.5 Subroutine VELGRAD ..........................................................44
5.6 Boundary routines ................................................................44
-
8/10/2019 Reynolds Stress ARSM
8/114
Contents
vi
6 Test cases ......................................................................................47
6.1 Flow over a flat plate.............................................................48
6.1.1 Grid ...........................................................................48
6.1.2 Boundary conditions..................................................50
6.1.3 Results ......................................................................51
6.2 Flow inside a rectangular duct ..............................................55
6.2.1 Grid ...........................................................................55
6.2.2 Boundary conditions..................................................56
6.2.3 Results ......................................................................57
6.3 Flow inside a rotating pipe ....................................................63
6.3.1 Grid ...........................................................................64
6.3.2 Boundary conditions..................................................656.3.3 Results ......................................................................65
6.4 Flow over a rotating cylinder .................................................69
6.4.1 Grid ...........................................................................70
6.4.2 Boundary conditions..................................................71
6.4.3 Results ......................................................................71
7 Summary and conclusions............................................................79
Bibliography ............................................................................................81
Appendix A, Subroutine EARSM1..........................................................87
Appendix B, Subroutine TTIMES ...........................................................91
Appendix C, Subroutine ANISFLUX.......................................................93
Appendix D, Subroutine FLUXP .............................................................95
Appendix E, Subroutine PEEKOO .........................................................97
Appendix F, Subroutine VELGRAD........................................................99
-
8/10/2019 Reynolds Stress ARSM
9/114
vii
Nomenclature
Roman alphabet
A Model coefficient
ija Reynolds stress anisotropy tensor
C Model coefficient
fC Skin friction coefficient( )aijC Coriolis tensoreffC Effective eddy-viscosity coefficient
E Total energye Specific internal energy
if Body force vector
H Matrix defined by the equation (4.44)
J Matrix defined by the equation (4.45)
k Turbulent kinetic energy, Heat transfer coefficient
l Turbulent length scale
N Symbol defined bytheequation(4.25), Non-dimensional rotation rate
in Cell face unit normal vector
P Production of turbulence
ijP Production of turbulence tensor
p Pressure
iq Heat flux vector
Re Reynolds number
S Cell face area
ijS Mean strain tensor
T Temperature
t Time
iu Velocity vector
u Friction velocity
u Fluctuating component of the u
V Cell volume
ix Cartesian coordinate vector
+y Dimensionless wall distance
-
8/10/2019 Reynolds Stress ARSM
10/114
Nomenclature
viii
Greek alphabet
Coefficient for the terms of the ija
ij Kroneckers delta
Dissipation rate
ij Dissipation rate tensor
ijk Permutation tensor
ij Redistribution tensor
Molecular viscosity
Density
Turbulent time-scale
w Wall shear stress
ij Viscous stress tensorT Kinematic eddy-viscosity
ij Mean vorticity tensor
*ij Absolute mean vorticity tensor
Rij Effective mean vorticity tensor
Sij System rotation tensor
kspecific dissipation rateSk Constant system rotation rate vector
Invariants
SII Second invariant of the ijS
II Second invariant of the ij
III Third invariant of the ijS
IV Third invariant defined as kijkijS
V Fourth invariant defined as likljkijSS
-
8/10/2019 Reynolds Stress ARSM
11/114
1
Chapter 1
Introduction
In earlier days fluid dynamics, like other physical sciences, was divided into
theoretical and experimental branches. The equipment and vehicles in-
volving fluid flow were designed and analysed by these two methods. With
the evolution of the digital computer, a third method called ComputationalFluid Dynamics (CFD) has become available. In this computational ap-
proach the equations that govern a process of interest are solved numeri-
cally at certain discrete points of space.
The evolution of numerical methods for solving ordinary and partial differ-
ential equations began approximately at the beginning of the twentieth
century. The automatic digital computer was invented in the early 1940s
and was used from nearly the beginning to solve problems in fluid dynam-
ics. The explosion in computational activity did not, however, begin until the
high-speed digital computers began to be generally available in the 1960s
(Ref. [1]).
The three key elements of CFD are algorithm development, grid generation
and turbulence modelling. In this study only the third element is scrutinised.
Turbulence is inherently three-dimensional and time dependent, and an
enormous amount of information is thus required to completely describe a
-
8/10/2019 Reynolds Stress ARSM
12/114
Introduction
2
turbulent flow. This is beyond the capability of the existing computers for
virtually all practical flows. Thus, some kind of approximate and statistical
method, called a turbulence model, is needed.
Helsinki University of Technology has quite a long history in the field of
CFD. The development of the parallel multi-block Navier-Stokes flow solver
called FINFLO was initiated in 1987. The original two-man development
group has grown along with new applications. Nowadays, in 1995 estab-
lished CFD group consists of about 15 researchers from the Laboratory of
Aerodynamics, the Laboratory of Applied Thermodynamics and the Labo-
ratory of Ship Hydrodynamics, all utilising the FINFLO code. The FINFLO is
nowadays suitable for compressible, time dependent, laminar and turbulent
flows and has been applied to dozens of demanding research and devel-
opment projects. The code is able to handle structured multi-block grids
and the equations are solved by an implicit pseudo-time integration scheme
using Roes flux splitting.
The CFD-group has always tried to keep up with the most recent turbu-
lence modelling. Prior to this work the FINFLO included the following tur-
bulence models; Baldwin-Lomax [2], Cebeci-Smith [3], Chiens low Rey-
nolds number
k model [4] and different variants of Menters
k
model (BSL), (SST) [5] [6] and (RCSST) [7]. The aim of this thesis has
been to implement into the FINFLO a new turbulence model called the Ex-
plicit Algebraic Reynolds Stress Model (EARSM) developed by Stefan Wal-
lin and Arne V. Johansson from Sweden. The formulation of the model is
presented briefly in this thesis, but the reader is encouraged to find the
complete description in reference [8]. The EARSM was programmed using
Fortran 77 and 90 as they are used throughout the main code. After the
programming was completed, four test cases were selected, calculated and
analysed to verify the correct implementation of the model.
-
8/10/2019 Reynolds Stress ARSM
13/114
3
Chapter 2
Governing equations and eddy-viscosity turbu-lence models
In this chapter the equations governing the flow field are presented in their
time-accurate and Reynolds averaged form. Then some general remarks of
turbulence are made and importance for the turbulence modelling is em-phasised. At the end of the chapter the basic aspects of the so-called eddy-
viscosity turbulence models are discussed.
2.1 Navier-Stokes equations
The equations of viscous flow have been known for more than 100 years.
The exact number of basic equations depends upon personal preference,
but some relations are, however, more basic than others. Usually the basicsystem of equations is considered to be the three laws of conservation for
physical systems:
- Conservation of mass (i.e. the continuity equation)
- Conservation of momentum (i.e. the Newtons second law)
- Conservation of energy (i.e. the first law of thermodynamics)
-
8/10/2019 Reynolds Stress ARSM
14/114
Governing equations and eddy-viscosity turbulence models
4
The continuity equation simply states that the mass must be conserved. In
the Cartesian coordinates ix this equation can be written as
( ) 0=
+
i
i
xu
t (2.1)
where is the density of the fluid, ttime and iu the velocity vector. Ten-
sor notation will be used throughout the text whenever practical. The sec-
ond equation, conservation of momentum, states that momentum must be
conserved. It can be written in the Cartesian coordinates as
( )
j
ij
ii
i
jii
xx
p
fx
uu
t
u
+
=
+
(2.2)
where if is a body force, p the pressure. The viscous stress tensor ij is
+
=k
kij
i
j
j
iij
x
u
x
u
x
u
3
2(2.3)
where is the molecular viscosity and ij the Kroneckers delta. The third
equation, conservation of energy, states that the energy must be con-served. It can be written as
( )[ ] ( ) 0=
+
+
jiji
j
j
j
qux
pEuxt
E (2.4)
whereEis defined as the total energy and can be written as
+= iiuueE2
1 (2.5)
where e is the specific internal energy. Assuming Fouriers heat transfer
law the heat flux vector can be written as
i
ix
Tkq
= (2.6)
where kin the heat transfer coefficient and T the temperature.
-
8/10/2019 Reynolds Stress ARSM
15/114
Governing equations and eddy-viscosity turbulence models
5
2.2 Reynolds averaging
In a turbulent flow the local pressure, density, velocity components and
temperature vary randomly with time. A reasonable approach is to separate
the flow quantities into stationary and random parts. The quantities are thus
usually presented as a sum of the mean flow value and the fluctuating part
as ii uu + . This formulation is inserted into the equations (2.1), (2.2) and(2.4); a process called Reynolds averaging is performed. There are three
forms of averaging present in the turbulence-model research, which are a
time average, a spatial average and an ensemble average. The general
term used to describe these averaging processes is mean. The most
commonly used averaging is the first, time averaging. It can be used onlyfor stationary turbulence, i.e. for a turbulent flow that does not vary with
time on the average. For such a flow the mean flow value is defined as
( )+
=
Tt
t
iT
i dttuT
u 1lim (2.7)
In practice taking the limit to infinity means that the integration time T
needs to be long enough relative to the maximum period of the assumed
velocity fluctuations.
The Reynolds averaged Navier-Stokes equations (RANS) are obtained
from the continuity and momentum equations, (2.1) and (2.2), by taking the
time average of all the terms in the equations. The continuity equation does
not change since it is linear in terms of the velocity. However, the momen-
tum equation is non-linear, which means that all the fluctuating components
do not vanish. An extra term, called Reynolds stress jiuu , appears in the
momentum equation (2.2). The result can be written as
( ) ( ) ( )j
jiij
i
i
i
jii
x
uu
x
pf
x
uu
t
u
+
=
+
(2.8)
where the overbar has been dropped from the mean values. This conven-
tion will be used throughout the text whenever obvious. The equation (2.8)
presents the fundamental problem of turbulence. In order to compute all the
-
8/10/2019 Reynolds Stress ARSM
16/114
Governing equations and eddy-viscosity turbulence models
6
mean-flow properties of the turbulent flow we need a reasonably accurate
way to compute the Reynolds stress jiuu . This is the fundamental reasonfor the need of the turbulence models.
The complete set of the Reynolds averaged Navier-Stokes equations are
not presented in this text, because they can be found from many refer-
ences, for example [9]. Scalar transport equations are also needed, for ex-
ample to describe the transport of the concentration of species or the mass
fraction of species. Their exact formulation can be found from reference [9]
or [10]. The presentation of the exact formulation of the FINFLO along with
the solution methods used is beyond the scope of this text. The reader is
encouraged to consult references [11], [12] and [13].
2.3 General facts of turbulence
Many basic definitions of turbulence have been written, but probably the
most accurate was formed by Bradshaw in 1974:
"Turbulent fluid motion is irregular condition of flow in which the various
quantities show a random variation with time and space co-ordinates, so
that statistically distinct average values can be discerned. Turbulence has a
wide range of scales."(Ref. [14])
Analysis of the solutions to the Navier-Stokes equations shows that turbu-
lence develops as an instability of the laminar flow. At low Reynolds num-
bers, fluid layers slide smoothly past each other and the molecular viscosity
dampens the randomly occurring small-scale, high frequency disturbances.
The flow is laminar and exactly predictable. When Reynolds number in-
creases, the restraining effects of viscosity are too weak to prevent such
disturbances from amplifying. The disturbances grow gradually and be-
come non-linear. They also interact with neighbouring disturbances. At high
Reynolds number, the flow reaches a chaotic and non-repeating, i.e. tur-
bulent, state. Turbulent motions are characterised by the following proper-
ties: three dimensional, unsteady, random, strongly vorticial, strongly dissi-
pative and strongly diffusive. Turbulence increases friction, heat transfer
-
8/10/2019 Reynolds Stress ARSM
17/114
Governing equations and eddy-viscosity turbulence models
7
and mixing and spreading rate. It also reduces the separation tendency of
the flow by energising the boundary layer (Ref.[15]).
The non-linearity of the Navier-Stokes equations leads to interactions be-tween turbulent fluctuations of different wavelengths and directions. Ac-
cording to reference [14] the wavelengths of motion usually extend all the
way from a maximum comparable to the width of the flow to a minimum
fixed by viscous dissipation of energy. The main physical process that
spreads the motion over a wide range of wavelengths is called vortex
stretching. The turbulence gains energy if the vortex elements are primarily
oriented in a direction in which the mean velocity gradients can stretch
them. This is called production of turbulence. The turbulent kinetic energy is
then convected, diffused and dissipated.
The larger-scale turbulent motion carries most of the energy and is mainly
responsible for the enhanced diffusivity and attending stresses. The large
eddies have memory and their orientation is sensitive to the mean flow.
The large eddies randomly stretch the smaller eddies, cascading energy to
them. Small eddies lose their orientation preference and become statisti-
cally isotropic. Energy is dissipated by viscosity in the shortest wave-
lengths, although the long-wavelength motion at the start of the cascadesets the rate of dissipation of energy. The shortest wavelengths simply ad-
just accordingly. This kind of turbulence is called equilibrium turbulence.
The ratio of largest to smallest scales increases rapidly as the Reynolds
number increases (Ref. [14]).
If the unsteady Navier-Stokes equations were calculated, a vast range of
time and length scales would have to be computed. In other words, a very
fine grid and very small time steps would be required. This is possible with
todays computers for only very simple problems with small Reynolds num-
bers. Thus a turbulence model is needed to account at least some of the
fluctuating motion by a statistical approach. The model should be, in princi-
ple, simple, broadly general and applicable, based on physics rather than
intuition, computationally stable and coordinately invariable. This is, of
course, an impossible list of requirements. The next sections of this chapter
describe traditional ways to model the turbulence using a so-called Boussi-
nesq eddy-viscosity approximation.
-
8/10/2019 Reynolds Stress ARSM
18/114
Governing equations and eddy-viscosity turbulence models
8
2.4 Eddy-viscosity turbulence models
In this section the so-called eddy-viscosity turbulence models are briefly
discussed. The models can be divided in three categories; zero-, one- and
two-equation models. These models use the Boussinesq eddy-viscosity
approximation. The word approximation is a little misleading here because
it is really more of a hypothesis than an approximation. It can be written for
one- and two-equation models as
kx
u
x
u
x
uuu ij
k
kij
j
i
i
j
Tji 3
2
3
2
+
= (2.9)
For the zero-equation models, which do not make use of the turbulent ki-
netic energy 2iiuuk , the last term of the equation is zero. The Boussi-nesq approximation is used to compute the Reynolds stress tensor as the
product of the kinematic eddy-viscosity T and the mean strain-rate tensor.
The kinematic eddy-viscosity will be referred hereafter as the eddy-viscosity
for simplicity. The Reynolds stress components are calculated from the
eddy-viscosity for an incompressible flow as
kSuu ijijTji 3
22 = (2.10)
where the mean strain-rate tensor is defined as
+
=i
j
j
iij
x
u
x
uS
2
1(2.11)
The eddy-viscosity is a scalar field variable and the different Reynolds
stress components of a certain point of the flow field thus scale with the
mean strain-rate tensor, i.e. the Reynolds stress components are linearly
proportional to the mean strain-rate tensor. Different ways to calculate the
eddy-viscosity are described in the next subsections.
The equations (2.10) and (2.11) as well as all the following equations in this
text are presented for an incompressible flow. The FINFLO calculation rou-
tine calculates the flow as compressible, of course, while all of the turbu-
-
8/10/2019 Reynolds Stress ARSM
19/114
Governing equations and eddy-viscosity turbulence models
9
lence models approximate the flow as incompressible. This approximation
is a so-called Morkovins hypothesis meaning that the compressibility af-
fects the turbulence quantities only at hypersonic speeds.
2.4.1 Algebraic models
Models where the eddy-viscosity is completely determined in terms of the
local mean flow variables are referred to as zero-equation or algebraic
models. Algebraic models are the simplest of all turbulence models. They
often use some form of Prandtls mixing length hypothesis in order to com-
pute the eddy-viscosity using terms of mixing length in analogy to the mo-
lecular mixing phenomenon. Whereas the molecular viscosity is an intrinsic
property of the fluid, the eddy-viscosity (and hence the mixing length) de-
pends upon the flow. Because of this, the eddy-viscosity and mixing length
must be specified in advance by an algebraic relation between eddy-
viscosity and length scales of the mean flow. Thus, algebraic models are,
by definition, incomplete models of turbulence. (Ref. [14]).
The eddy-viscosity is usually calculated as
j
i
mixT
dy
dul 2= (2.12)
where mixl is the mixing length analogous with the molecular mean free
path in molecular mixing.
Well-known and much used algebraic models are the Baldwin-Lomax [2]
and the Cebeci-Smith [3] models, both of which are two-layer models with
separate expressions for the computation of the eddy-viscosity in eachlayer. The Cebeci-Smith model is simple and easy to implement. Most
computational effort goes to the calculation of the boundary layer velocity
thickness. The Baldwin-Lomax model was formulated for use in computa-
tions where boundary layer properties such as the boundary layer and the
boundary layer velocity thickness are difficult to determine.
Algebraic models are conceptually very simple, economical in terms of
computer resources and rarely cause any unexpected numerical difficulties.
-
8/10/2019 Reynolds Stress ARSM
20/114
Governing equations and eddy-viscosity turbulence models
10
The user of the algebraic models must always be aware of the issue of in-
completeness. Both the Cebeci-Smith and Baldwin-Lomax models work
well only for the boundary layer flows for which they have been fine-tuned.
They reproduce skin friction and velocity profiles well for incompressible
turbulent boundary layers provided the pressure gradient is not too strong.
However, they can not be applied to other flow cases, such as wake flows
or free jets.
2.4.2 One-equation models
In one-equation models, one of the two turbulence scales or a combination
of both is determined from a transport equation. The transport equation for
the turbulence kinetic energy can be written as
+
= jjiijjj
iij upuuu
x
k
xx
u
Dt
Dk
1
2
1 (2.13)
where the tensor notation jj xutDtD + is used to denote the rateof change following the mean flow. The term jiij xu is the production
of the turbulence, P , the term jxk is the molecular diffusion, the term2jii uuu is the turbulent flux of the turbulent kinetic energy, k, and the last
term jup is the so-called pressure diffusion term. The last term is usu-ally neglected because its contribution is very small.
The is the dissipation rate per unit mass, defined as
k
i
k
i
x
u
x
u
= (2.14)
The eddy-viscosity is usually calculated as
klC mixT = (2.15)
where the C is a model coefficient. As for algebraic models, additional
information from the mean flow field or geometrical measures is needed.
Most commonly used one-equation models are the Spalart-Allmaras [16]
-
8/10/2019 Reynolds Stress ARSM
21/114
Governing equations and eddy-viscosity turbulence models
11
and Baldwin-Barth [17] models. Actually, in the models the transported
quantity is 2k , which follows from the definition of the turbulent Reynolds
number TTT luRe , since 21~kuT and 23~klT .
According to reference [15], the one-equation models give usually more
accurate results than algebraic models because more realism is obtained
as a result of deducing the turbulent velocity scale from a turbulence quan-
tity k rather than from gradients of mean velocity. This also implies that,
by obtaining the solution for the turbulent velocity scale from a differential
transport equation, some account is taken of the history effects, which be-
come important in non-equilibrium flows. The differential transport equation
also removes the problems associated to the flow field points where the
local mean velocity gradient is zero.
The main drawbacks with one-equation models are largely the same as for
the mixing-length models. The history effects are not accounted for the
length scale l, which is still prescribed as an algebraic function of local
quantities. According to reference [18] the Spalart-Allmaras model performs
better than the Chiens k model in a decelerating flow with adversepressure gradient because it is specially designed for aerodynamical pur-
poses. It is one of the most popular turbulence models in the field of aero-nautical applications, especially in the U.S.A.
2.4.3 Two-equation models
Two-equation models have served as the foundation for much of the tur-
bulence model research until very recent years. The most significant differ-
ence between the other eddy-viscosity models and the two-equation mod-
els is that the latter are complete, i.e. can be used to predict properties of agiven turbulent flow with no prior knowledge of the turbulence structure or
flow geometry. These models provide an equation not only for computation
of k, but also for the turbulence length scale or equivalent.
The starting point for virtually all linear two-equation eddy-viscosity models
is the Boussinesq approximation (2.9) and the turbulence kinetic energy
equation (2.13). The choice of the second variable is arbitrary and many
-
8/10/2019 Reynolds Stress ARSM
22/114
Governing equations and eddy-viscosity turbulence models
12
proposals have been presented. By far two of the most popular dependent
variables for the second variable have been the dissipation rate and the
k-specific dissipation rate . Nowadays there is also increasing interest in
the use of the turbulent time-scale . The most common proposals are
presented in the table 2.1
Table 2.1. Proposals for the second variable (Ref. [15]).
Proposer Year Variable Symbol Physical meaning
Kolmogorov 1942 121
lk Frequency
Rotta 1951 l l Length scale
Rotta 1968 kl kl k times the length scale
Harlow & Nakayama 1968 123
lk Energy dissipation rate
Spalding 1969 2
kl Vorticity fluctuations squared
Nee & Kovasznay 1969 21
lkT Eddy viscosity
Speziale 1992
21
lk Time-scale
The k model is the best-known two-equation turbulence model be-cause it is simple to understand and use and relatively easy to program.
The model dates back to the late 1960s but the most used formulation, re-
ferred as the standard k model, was presented in 1972 by Jones andLaunder [19]. The eddy-viscosity used with the Boussinesq approximation
is calculated as
2k
CT= (2.16)
where C is a model coefficient. The k model is used widely in practi-cal engineering calculations even though the standard model can be used
only with attached flows with thin shear layers. The model fails to reproduce
the correct flow behaviour in many important flow situations, such as: ad-
-
8/10/2019 Reynolds Stress ARSM
23/114
Governing equations and eddy-viscosity turbulence models
13
verse pressure gradients, bluff-body flows, separation, streamline curva-
ture, swirl, buoyancy, turbulence driven secondary motion, compressibility
and unsteadiness.
The k model is based on the choice of the specific dissipation rate asthe second variable. The eddy viscosity is calculated as
kT= (2.17)
Even though the wall boundary condition for the is more difficult to for-
mulate and program than for the , the k model has found its way to
many engineering, especially aerodynamic, applications. This is mainly dueto the fact that it can also reproduce the flow field behaviour with adverse
pressure gradient. It usually also predicts separation better than the other
linear two-equation eddy-viscosity models. Actually, the k SST model[7] predicts the separation fairly well, but it is no longer really a linear
model.
-
8/10/2019 Reynolds Stress ARSM
24/114
-
8/10/2019 Reynolds Stress ARSM
25/114
15
Chapter 3
Differential Reynolds stress models
In this chapter a different approach to model the turbulence is reviewed.
The Boussinesq approximation is abandoned and the transport equation for
the Reynolds stresses is presented instead. In the end of the chapter mod-
elling propositions of the different terms of the transport equation are dis-cussed.
3.1 Reasons for using the Reynolds stress models
Eddy-viscosity models perform reasonably well in attached boundary layer
flows as long as only one component of the Reynolds stress tensor is of
significant importance. In these cases the eddy-viscosity can be thought as
a representative for the significant Reynolds stress component. However, ifthe flow becomes more complicated the eddy-viscosity assumption fails.
There is thus not much hope for a more general validity of the eddy-
viscosity approach (Ref. [8]).
The development of the so-called differential Reynolds stress models (also
called Reynolds stress transport, second-order closure, second-
moment closure and second-order modelling) started officially in 1968,
-
8/10/2019 Reynolds Stress ARSM
26/114
Differential Reynolds stress models
16
when C. Donaldson lectured in Stanford of the need to model the Reynolds
stress terms. The Boussinesq eddy-viscosity hypothesis is abandoned and
the unknown Reynolds stress components are obtained directly from the
solution of differential transport equations in which they are the dependent
variables. The Reynolds stress tensor is symmetric meaning that four
equations need to be solved in two-dimensional and six equations in three-
dimensional flows. An additional equation for the length scale or equivalent
is also solved in each case.
The Reynolds stress models are thus more complicated than the eddy-
viscosity models. Still, they are attractive because they provide a more ac-
curate representation of the turbulence and are valid over a wider range of
flows. They can capture many of the complex effects encountered in nature
and in engineering practice without recourse to the ad-hoc modifications
found necessary in lower-order models. According to reference [15], exam-
ples of flow situations where Reynolds stress models were found to give
accurate predictions include flows with streamline curvature, rotation, swirl
and buoyancy.
3.2 Exact equation for the Reynolds stress tensor
The exact transport equation for the Reynolds-stress tensor jiuu is ob-tained from the momentum equation (2.2) by multiplying the instantaneous
iu component equation by ju , the ju component equation by iu , addingthe two and then time-averaging the result. Following reference [20], the
result can be written for constant-density flows with no body forces after
some rearrangement as
=
t
uuD ji
+
k
ikj
k
j
kix
uuu
x
uuu Terms I and II
( )
++
k
ji
ikjjkikji
k x
uuupupuuu
x
1Term III
+
+i
j
j
i
x
u
x
up
k
j
k
i
x
u
x
u2 Terms IV and V
(3.1)
-
8/10/2019 Reynolds Stress ARSM
27/114
Differential Reynolds stress models
17
Term I is the convection term. It represents the rate of change of jiuu
along a streamline. In steady flows this is equal to the rate at which Rey-
nolds stresses are convected by the mean fluid motion.
Term IIis the production term. It represents the rate of production of jiuu
by mean shear, which is the reason for turbulence as discussed earlier.
The shear stress is generated by interaction of transverse normal stress
and shear strain. The production term is both large and exact, which is a
partial reason for the success of the Reynolds stress models. It will be
hereafter referred to as ijP .
Term III is the diffusion term. It represents the rate of spatial transport of
jiuu by the action of turbulent fluctuations, pressure fluctuations and mo-lecular diffusion.
Term IV is the redistribution term, which is also called the pressure-strain
term. It represents the redistribution of the available turbulent kinetic energy
amongst the fluctuating velocity components. The mean flow direction be-
ing 1x , only 11uu is generated in the shear layer, meaning that it is usuallymuch larger than the other Reynolds stress components. The redistribution
term shares 11uu s energy out to 22uu and 33uu . It has no effect on theoverall level of turbulent kinetic energy since it has zero trace. The redistri-
bution term just drives turbulence towards isotropy by redistributing energy.
This term will be hereafter referred to as ij .
Usually 21uu has the opposite sign to the boundary layer shear strain. Theredistribution term 12 reduces also 21uu since isotropic turbulence mustbe shear-free. The same applies to the possible shear strains to other di-
rections.
Term V is the dissipation term. It represents the dissipation rate of jiuu
due to molecular viscous action. The dissipation term will be hereafter re-
ferred to as ij .
-
8/10/2019 Reynolds Stress ARSM
28/114
Differential Reynolds stress models
18
3.3 Modelling the Reynolds stress tensor
With the exception of the convection and production terms of the equation
for the Reynolds stress tensor, all the other terms introduce new unknown
correlations that must be modelled in terms of known or knowable quanti-
ties in order to close the equations.
The diffusion term is a sum of three parts, which are called the viscous,
pressure and turbulent diffusion. The contribution of the viscous diffusion
term to the total rate of transport of jiuu is small at high values of the tur-bulence Reynolds number. This term is thus often neglected in high Rey-
nolds number applications even though it could be included in the numeri-
cal simulations without any difficulty.
Little is known about the pressure diffusion term since direct measurements
can not be conducted. Estimates of its magnitude are done by indirect
methods, which contain all the errors made in the measurements of the
other terms. The consensus of several experiments, however, suggests
that this term is relatively unimportant and may therefore be neglected (Ref.
[15]).
Daly and Harlow [21] were the first to model the turbulent diffusion term
using the gradient transport hypothesis, meaning that the diffusion of a
quantity is assumed to be proportional to the spatial gradient of the same
quantity. According to reference [15] the turbulent diffusion can be written
as
l
ji
lkSkjix
uuuu
kCuuu
=
(3.2)
where SC is an empirical model coefficient typically set equal to 0.22,
which is obtained by computer optimisation. The ratio k represents a
characteristic time scale of the energy containing eddies.
Alternative models for the turbulent diffusion exist, but they all are consid-
erably more complex than Daly & Harlows model without necessarily pro-
-
8/10/2019 Reynolds Stress ARSM
29/114
Differential Reynolds stress models
19
ducing better overall performance. Examples of such models are Hanjalic &
Launder [22] and Lumley & Khajeh-Nouri [23].
The redistribution term modelling is based on the Poisson equation forthe instantaneous pressure. It is obtained by differentiating the Navier-
Stokes equations in a manner that is beyond the scope of this text. Then,
by subtracting the mean, the following equation for the fluctuating pressure
is obtained
( )
+
=
j
i
j
i
ij
jiji
i x
u
x
u
xx
uuuu
x
p2
1 2
2
2
(3.3)
where a sum of two very distinct terms is identified in the parentheses. The
first term of the sum contains only turbulence quantities and second term
contains mean-velocity gradients. When the equation is solved for homo-
geneous turbulence, the so-called Chous integral for the pressure-strain
correlation is obtained
+
+
=
wij
jl
im
m
l
jml
iml
j
i
r
dvol
rr
uu
x
u
rrr
uuu
x
up,
*2**3
24
1
(3.4)
where * denotes quantities evaluated at position *x and xxr = * . wij ,is a surface integral, which is important only when the typical size of the
energy containing eddies is of the same order as distance from a wall (Ref.
[15]).
Launder, Reece and Rodi (LRR) proposed that the first and the second
term in the parentheses of the equation (3.4) are modelled separately [24].
This is the traditional approach to modelling Chous integral and the great
majority of Reynolds-stress model developers have followed it.
Speziale, Sarkar and Gatski (SSG) proposed that the Chous integral could
be modelled as a whole [25]. This approach is rapidly becoming more
popular than (LLR) as it does not appear to require a specific model for the
surface integral wij , .
-
8/10/2019 Reynolds Stress ARSM
30/114
-
8/10/2019 Reynolds Stress ARSM
31/114
-
8/10/2019 Reynolds Stress ARSM
32/114
Formulation of an explicit algebraic Reynolds stress model
22
Also a transport equation for the Reynolds stress anisotropy tensor can be
construed in a similar was as the transport equation (3.1) for the Reynolds
stress component tensor. According to reference [8] it can be presented in
a rotating Cartesian coordinate system as
( )aij
ijijijji
l
lji
l
ljiijC
PP
k
uu
x
ku
k
uu
x
uuu
Dt
Dak+++
=
1
1
(4.2)
where the dissipation rate tensor ij and the redistribution tensor ij need
to be modelled. On the other hand, the production terms ijP and 2iiPP=and the Coriolis term ( )aijC do not need any modelling since they can be
calculated directly from the Reynolds stress tensor.
In flows where the Reynolds stress anisotropy varies slowly in time and
space, the transport equation for the Reynolds stress anisotropy tensor is
reduced to an implicit algebraic relation. Many inhomogeneous flows of en-
gineering interest consist of a steady flow and equilibrium turbulence as-
sumption can thus be made. Therefore the convection and diffusion of the
Reynolds stress anisotropy may be neglected. This is equivalent to the as-
sumption made in reference [27] that the convection and diffusion of each
Reynolds stress component scale with the convection and diffusion of the
turbulent kinetic energy. This is indeed the traditional ARSM idea, to ne-
glect convection and diffusion terms in the exact transport equation for the
Reynolds stress anisotropy. This means that the left-side terms in the
equation (4.2) can be neglected and set to zero
01
=
l
lji
l
ljiij
x
ku
k
uu
x
uuu
Dt
Dak
(4.3)
The convection term DtDaij is exactly zero for all stationary parallel mean
flows, such as fully developed channel and pipe flows. For inhomogeneous
flows the assumption of negligible diffusion effects can cause problems,
particularly in regions where the production term is small or where the in-
homogeneity is strong. However, the ARSM assumption incorporates in a
natural way the effects of rotation, effects of streamline curvature and
three-dimensionality of the flow. The assumption results in an implicit alge-
-
8/10/2019 Reynolds Stress ARSM
33/114
Formulation of an explicit algebraic Reynolds stress model
23
braic equation for the Reynolds stress anisotropy tensor, which can be
written as
( )aij
ijijijjiCPP
kuu ++=
1 (4.4)
The production terms ijP and 2iiPP= and the Coriolis term ( )aijC do notneed any modelling since they can be calculated directly from the Reynolds
stress tensor. However, the dissipation rate tensor ij and the redistribution
ij need to be modelled. In a non-rotating coordinate system, the produc-
tion term is normally written as implied in the equation (3.1). It is rewritten
here for the sake of completeness as
k
ikj
k
j
kiijx
uuu
x
uuuP
= (4.5)
To illustrate the natural way in which rotational effects enter in this type of
formalism it is convenient to split the mean velocity gradient tensor into a
mean strain and a mean vorticity tensor. Symbols ijS and ij are usedhere to represent these tensors, normalised with the turbulent time-scale as
=
+
=
i
j
j
iij
i
j
j
iij
x
u
x
u
x
u
x
uS
2
2
(4.6)
where the turbulent time-scale is defined as
k
(4.7)
A consistent formulation of the equation (4.4) can then be obtained by re-
placing the mean vorticity tensor by the absolute vorticity tensor. According
to reference [8], this can formally be done introducing the absolute mean
vorticity tensor
Sijijij +=* (4.8)
-
8/10/2019 Reynolds Stress ARSM
34/114
Formulation of an explicit algebraic Reynolds stress model
24
where the system rotation tensor Sij is defined as
Skjik
Sij = (4.9)
where ijk is the permutation tensor and Sk is the constant rotation rate
vector of the system. The permutation tensor is unity when the indexes are
123, 231 or 312. It is minus unity when the indexes are 213, 321 or 132. In
other cases it is zero. Now, it is possible to rewrite the production term (4.5)
using the equations (4.1), (4.6) and (4.8). It is presented here normalised
by the dissipation rate as
( ) ( )kjikkjikkjikkjikij
ij
aaaSSaS
P**
3
4
+++= (4.10)
The Coriolis term ( )aijC arises from the transformation of the convection
term. It is presented in this text for completeness, even though it was not
programmed into the FINFLO, as the EARSM was only implemented into a
non-rotating coordinate system. For rotational purposes, FINFLO uses a
semi-rotational formulation, which is presented in reference [28]. The
Coriolis term can be written according to reference [8] as
( )kj
S
ik
S
kjik
a
ij aaC = (4.11)
For the present modelling purpose the dissipation rate tensor is assumed to
be isotropic. The equation is written here normalised by the dissipation rate
as
ijij
3
2= (4.12)
The redistribution term is modelled in two subparts, slow and fast redistri-
bution, as proposed in reference [8]. According to reference [29], the slow
redistribution rate can be though to be linear in terms of the anisotropy ten-
sor as
( )
ij
sij
aC1=
(4.13)
-
8/10/2019 Reynolds Stress ARSM
35/114
Formulation of an explicit algebraic Reynolds stress model
25
where 1C is a model constant. For the rapid redistribution rate the general
linear Launder, Reece and Rodi (LRR) model [24] is chosen. It is normally
written for a non-rotating system as
( )
+
+=
PDC
x
u
x
uk
CPP
C
ijij
i
j
j
iijij
rij
3
2
11
28
55
230
3
2
11
8
2
22
(4.14)
where the 2C is a model constant and ijD is calculated as
i
kkj
j
kkiij
x
uuu
x
uuuD
= (4.15)
A simple way of obtaining a consistent, frame-independent formulation of
the rapid pressure-strain rate model is to apply the same methodology as
for the production term, i.e. to use the mean strain and vorticity tensors de-
fined in equation (4.4.) and to normalise by . The result can be written as
( )
( )kjikkjik
ijmkkmkjikkjikij
rij
aaC
SaaSSaC
S
**2
2
11
107
3
2
11
69
5
4
+
++
+=
(4.16)
When the equations for the production term (4.5), the Coriolis term (4.11),
the dissipation term (4.12), the slow redistribution term (4.13) and the fast
redistribution terms (4.16) are inserted to the equation (4.4), the implicit al-
gebraic equation for the Reynolds stress anisotropy tensor is then obtained
as
( )
+
+
+=
+
kiikijkjikkjik
kjRik
Rkjikijij
SaaSSaC
aaC
SaP
C
3
2
11
95
11
17
15
81
2
2
1
(4.17)
Here the effective mean vorticity tensor Rij is dependent on the modelconstant 2C and therefore on the choice of the redistribution model.
-
8/10/2019 Reynolds Stress ARSM
36/114
Formulation of an explicit algebraic Reynolds stress model
26
The effective mean vorticity tensor Rij of the equation (4.17) can be writtenas
Sijij
Sijij
Rij
CC
C
+++=
++=
17127
1711
2
2
2
* (4.18)
It should be noted, however, that in the present FINFLO formulation the
effective mean rotation rate tensor is the same as the vorticity tensor
ijRij = , for the reasons discussed earlier. The dissipation rate must be
modelled by a transport equation similar to the one used by the two-
equation eddy-viscosity models. It should be noted that equation (4.17) rep-
resents a non-linear relation, since the production to dissipation ratio is de-
fined as
kiikSaP
(4.19)
The algebraic Reynolds stress model has been formulated in the form of
equation (4.17). In the next section this is further simplified.
4.2 Explicit algebraic Reynolds stress model
The implicit ARSM relation for the Reynolds stress anisotropy tensor has
been found to be numerically and computationally cumbersome since there
is no diffusion or damping present in the equations. This usually means
convergence problems. In many applications the computational effort has
been found to be excessively large and the benefits of using ARSM instead
of the full Reynolds stress transport form are then successively lost, as
Wallin discusses [8]. Therefore a more simple and straightforward way to
calculate the Reynolds stress anisotropy is needed. A model, where the
Reynolds stresses are explicitly related to the mean flow field, is called an
explicit algebraic Reynolds stress model. It is much more numerically ro-
bust and has been found to have almost a negligible effect on the compu-
tational effort as compared to the linear k or the linear k model.
The procedure used here to obtain an explicit form is to avoid the non-
linearity by considering the production to dissipation ratio P as an extra
-
8/10/2019 Reynolds Stress ARSM
37/114
Formulation of an explicit algebraic Reynolds stress model
27
unknown. The resulting linear equation system can then be formally written
as
( ) 0,,, = PSa RklklklijL (4.20)
and may, in principal, be solved directly. However, by the inspection of the
equation (4.17) it is realised that the anisotropy tensor is dependent on only
two other tensors, ijS andRij , which can be used to form a complete base
for the anisotropy. The most general form for ija in terms of ijS andRij
consists of ten tensorially independent groups to which all higher-order ten-
sor combinations can be reduced with the aid of the Caley-Hamilton theo-
rem. Various techniques have been developed with different level of simpli-
fications in references [26], [30], [31], [32] and [33], for example. The Rey-
nolds stress anisotropy tensor is, however, written here following reference
[8] as
( )SSIISSa +
+
+= 423221
3
1
3
1 IIIIS
( )
++
+++ ISSISSSS VIV
3
2
3
2 22227
22
6
22
5 (4.21)
( ) ( ) ( )+++ 222210
22
9
22
8 SSSSSSSS
For simplicity, boldface has been used to denote second-rank tensors as
ija=a , ijS=S and Rij= . The inner product of two matrices is definedas ( ) ( ) kjikijij SS
2SSS and I represents the identity matrix. The
coefficients may be functions of the five independent invariants of S
and , which can be written as
( )( )
( )
( )
( ) likljkij
kijkij
kijkij
jiij
jiijS
SSV
SIV
SSSIII
II
SSII
====
====
==
22
2
3
2
2
trace
trace
trace
trace
trace
S
S
S
S
(4.22)
Other scalar parameters may also be involved.
-
8/10/2019 Reynolds Stress ARSM
38/114
-
8/10/2019 Reynolds Stress ARSM
39/114
Formulation of an explicit algebraic Reynolds stress model
29
The procedure to solve this equation system is the following:
1) The general form of the anisotropy, the tensor polynomial (4.21), is in-
serted into the simplified ARSM equation (4.24) where N is not yet
determined. Now N is considered to be a known parameter. This re-
sults in a linear equation system for the coefficients.
2) The linear equation system can be solved using the fact that higher-
order tensor groups can be reduced with the aid of the Cayley-
Hamilton theorem. The solution is unequivocal because the ten groups
in the general form (4.21) form a complete basis for the system. The
solution consist of many pages of complicated tensor algebra and it is
not essential to reproduce it here. The -coefficients are now functions
of N , or the production to dissipation ratio.
3) The next step is to formulate and solve the non-linear scalar equation
for N . This is done by inserting the solution of the coefficients into
the tensor polynomial (4.21) for the Reynolds stress anisotropy tensor.
4) The final step is to insert the resulting equation to the simplified ARSM
equation (4.24). Then a non-linear scalar equation for N is obtained,
which for a general three-dimensional flow field is of sixth order.
The solution for the simplified ARSM equation (4.24) is presented in the
following section 4.3 for two- and 4.4 for three-dimensional mean flows. In
section 4.5, the solution for the model without the simplification made from
the equation (4.17) to (4.23) is also presented.
4.3 Simplified EARSM for two-dimensional mean flow
For two-dimensional mean flows the simplified solution of the ARSM equa-
tion is reduced to only two non-zero coefficients, which can be expressed
according to reference [8] as
=
=
IIN
IIN
N
2
1
5
6
25
6
24
21
(4.27)
-
8/10/2019 Reynolds Stress ARSM
40/114
Formulation of an explicit algebraic Reynolds stress model
30
In the equation (4.27) it is clearly seen that the denominator cannot become
singular since II is always negative. The non-linear equation for Nin two-
dimensional mean flow can be derived by inserting the tensor polynomial
(4.21) for the Reynolds stress anisotropy tensor with the coefficients from
the equation (4.27), into the definition of N, equation (4.24). The resulting
equation is cubic and can be written following reference [8] as
02210
271
21
3 =+
+ IICNIIIINCN S (4.28)
The equation can be solved in a closed form with the solution for the posi-
tive root being
( ) ( )
( )