10 Simulations of Turbulence (Cancelled)

Turbulent Flows Simulations
1. Theory of Turbulent Flows
Experimental Observations
Highly unsteady, random
3D
A great deal of voriticity
Turbulent diffusion
Coherent (organized) structures
Fluctuate on a broad range of length and time scales
Mixing layer, Brown and Roshko(1956)
Aluminum flakes
suspended in water
Re=4300
Cantwell(1981)
Turbulent boundary layer Re=4000
A fog of tiny oil droplets + sheet of light
Falco(1977)
Kline(1967)
Low- and high-speed streaks at y+=2.7
Outer region
Wallace(1972)
Near-wall region
Robinson(1990)
Ejection / Sweep / Burst
Hinze(1975)
Friction
uv 60uv, y 30.5
during bursting
Willmarth(1972)
Iso-surfaces of the second invariant of the velocity gradient tensor

Flat plate boundary layer (DNS)
Wu and Moin(2009)
slow
Practical engineering method
Coherent Structures
Not yet
Truly prediction theory
Phenomenological Theories
Energy cascade (Richardson,1922)

Energy is transferred to successively smaller and smaller eddies
until the smallest eddies where molecular viscosity is effective
in dissipating the kinetic energy.
2
Kolmogorov hypotheses (1941)
u0 0
Three hypotheses
I. Kolmogorovs local isotropy hypothesis :
At sufficient high Re, the small scale turbulent motions (l<<l0) are
statistically isotropic.
Direction information or geometry of the large eddies is lost
as approaching smaller and smaller eddies
small scales local isotropic
II. Kolmogorovs first similarity hypothesis:

At sufficient high Re, the statistics of the small scale turbulent motions (l<<l0)
have a universal form that is uniquely determined by and .
Universal equilibrium range: small eddies can adapt quickly to maintain
a dynamic equilibrium with the energy-transfer rate imposed by the large eddies.
Small scales are statistically similar, or identical when scaled by Kolmogorov scales.
Kolmogorov scales
( 3 )1/ 4
( / )1/ 2
Considering a high Re flow,
u ( )1/ 4
(Dimensional analysis)
Re l0u0
Kinetic energy Time scale Rate of transfer of KE

2
0
Largest eddies
Smallest eddies
u2
l0
Re 3/4
0 l0 u0
u02 0
Re1/2
u u0
Re1/4
III. Kolmogorovs second similarity hypothesis:

At sufficient high Re, the statistics of the small scale turbulent motions
(<<l<<l0) have a universal form that is uniquely determined by ,
independent of .
Inertial subrange: <<l<<l0

Motions are determined by inertial effects.
l eddy length
ul ( l )
ul u0
(l l0 )1/3
l (l 2 )1/3
l 0
(l l0 )2/3
1/3
ul2 l
Viscous subrange: l~
Motions are determined by viscous effects.

Energy-containing range: l~l0
The bulk of energy is contained in the larger eddies in this size range.
The energy spectrum

How the turbulent kinetic energy is distributed among the eddies of
different sizes.
turbulent kinetic energy
k E ( )d
Wave number
With dimensional analysis and the Kolmogorovs local isotropic

and similarity hypotheses
Turbulent energy spectrum
E ( ) C 2 / 3 5/ 3
C is constant.
Isotropic turbulence,
Symbol: DNS
(Gotoh,2002))
Universal equilibrium range
Energy spectrum of turbulence

in function of wave number k, with
indication of the range of application
of the DNS, LES and RANS models.
The Taylor length scales lT

and integral scale lI are
associated with the LES and
RANS approximations,
respectively
2. Turbulence modelling
Dr.T.Gatski,
The biggest progress in turbulence
research is in turbulence modeling. CFD
software industry is based on turbulence
modeling. Many people now make their
livings on CFD softwares.
History of Turbulence modelling

K-model(1972)
Second moment
Closures-LRR(1975)
LES(1963)
SA-DES(1997)
Second moment
Closures(1945)
Log law(1930)
Mixing-length model(1925)
Time-averaged(1889)
Two-Eqs. RANS-LES
(2001)
DNS (Direct Numerical Simulation )

Turbulent flows are governed by Navier-Stokes equations.
How to solve NS directly?
(1) Resolution
Largest scale
the distance over which the fluctuating component of the

velocity remains correlated.
integral scale L
Smallest scale
largest turbulent eddy
on which viscosity is active, i.e., kinetic energy dissipation occurs
Kolmogorov scale
L Re3/L 4
ReL is based on the magnitude of the velocity fluctuation and the integral scale.
Re is Reynolds number in practice.

ReL 0.01 Re
The number of grid points ( L )3
Re9L/ 4
3/ 4
Nondimensional sizes of the space and time Re L
To obtain enough information in time consequence,

the time steps needed is Re 3L/ 4
3
Total operations Re L
DNS achieved now:
Applicable size:
simple flow (homogenous turbulent flow, channel flow,

free shear flow at low Re=104~105)
5123 (108) grid points
(2) Numerical Methods

Explicit time advance methods----------accurate time history
2nd--4th order Runge-Kutta method

Implicitly treated for viscous terms----------numerical instability
Spatial differecing-------energy conservative and low dissipation

compared to physical viscous (central scheme)
Step sizes of the space and time need to be related and the errors
in spatial and temporal discretizations should be balanced.
For equi-spaced grid and simple geometry, spectral methods is used.
Fast Fourier transform algorithm
Accuracy is difficult to measure in DNS and LES.
A small change in the initial state is amplified exponentially in time.
Another difficulty is the treatment of initial and boundary conditions.
Initial BCs is obtained from the close similar simulations
On solid wall, very fine grids resolve streaks
Symmetry BCs are not applicable instantaneous flow
A simulation must be run for some time before the flow develop all of the
correct characteristics of the flow------monitor some quantity
(3) The role of DNS: research tool

The results of a DNS contain very detailed information about a flow.
understand the physics
construct model
3. LES (Large Eddy Simulation )
Observation:
idea:
large structures
most energy transport
small structures
little energy transport

more universal
large structures
compute
small structures
model: subgrid-scale model
RANS
LES : resolve large structures

DNS: resolve all kinds of structures
RANS: resolve mean flow
(1) Resolution
High grid resolution is also required.
The number of grid points
Outer layer
Re0.4
L
Viscous sublayer
Re1.8
L
can be applied at Re at least one order of magnitude higher

To further reduce the number of grid points, approximate wall model is used.
more accurate than RANS and less expensive as DNS
(2) Numerical methods
is almost same as in DNS
(3) Spatial filtering

Decompose
Filtered part
or resolved part
Sub-filter part
or unresolved part
The filtered variable at the location r0
G represents the filter function
r is the position vector
When the sizes of the turbulent structures are less than ,they are cut off.
The mostly used filter functions:
Notice that filter functions in physical space limit

both spatial and temporal resolution.
(4) Filtered governing equations

The filtered incompressible Navier-Stokes equations-----Newtonian fluid
Describe the spatial and temporal evolution of the

large, energy-carrying scales of motion.
Subgrid-Scale Stress (SGS) tensor
Describe the effects of the unresolved scales.

The SGS tensor has to be modeled to close the equations.
Subgrid-Scale Reynolds-stress tensor

-----interactions between the small-scale structures
The filtered compressible Navier-Stokes equations

Favre(1965) averaging together with the spatial filtering.
Favre-averaged subgrid-scale stress tensor
(5) Subgrid-Scale modeling

Subgrid-Scale model is used to simulate energy transfer between the large
and the subgrid scales.
Transport from the large to the small ones---------cascade
small to the large ones--------backscatter
Eddy-viscosity model
incompressible
compressible
Very easy to implement in existing code
Kinetic energy dissipation is always positive----robust
Smagorinsky SGS model

Based on equilibrium hypothesis that small scales dissipate entirely
and instantaneously all the energy they received from large scales.
Eddy viscosity
Smagorinsky constant
In order to account for the reduced growth of the small scales near the wall
is modified as
numerically cheap and easy to implement

too dissipative in laminar regions with mean shear
require special provisions near wall and at laminar-turbulence transition
backscatter is not modeled.
Dynamic SGS model

Based on scale similarity, i.e., the smallest resolved scale motions
can provide information that can be used for largest subgrid scale
motions (Germano et al. 1990).
Smagorinsky constant is replaced by a parameter,
which evolves dynamically in space and time.
automatically decrease the parameter near the wall

automatically change the parameter from much smaller in shear flows
to larger in isotropic turbulence
(6) Wall models

The costs of LES for wall-bounded flows at high Re (>106) are still too
high for engineering purposes.
Excessively large number of grid points required to resolve the wall layer
The idea: model the wall layer by specifying a correlation between the
velocity in outer flow and the stress at the wall
Basic assumption: weak interaction between the near wall and outer region.
A new zonal approach proposed by Balaras et al.(1996)

Allows it to place the first point in a region
Remarks:
At present, reasonable subgrid-scale models exist and produce
good simulations, however, the models are not sufficiently precise to
be trusted to simulate a flow that has never been treated before.
To reduce the cost of LES, with retained accuracy, reliability and

versatility, hybrid LES/RANS approaches, have recently been
developed in which RANS and LES are combined to make the
most of both techniques. The most well-known model of this type
is the detached eddy simulation (DES) model, combining RANS
modeling for the attached eddies with LES computations for the
detached eddies.
Further Reading:
Towards the use of large eddy simulation in engineering
C. Fureby
Progress in Aerospace Sciences 44 (2008) 381396
4. RANS (Reynolds-Averaged Navier-Stokes equations )

------ turbulence models
(1) Basic equations of turbulence
Viscous stress tensor
Strain-rate tensor
Total energy
Total enthalpy
Incompressible flow
Reynolds averaging -------incompressible flow
mean value
turbulent fluctuations
There are three different forms of Reynolds averaging
Time averaging -------statistically steady turbulence
spatial averaging -------homogeneous turbulence
ensemble averaging -------general turbulence
In cases where the turbulence flow is both stationary and homogeneous,

all three averaging forms are equivalent-------ergodic hypothesis
Favre (mass) averaging -------compressible flow
The most convenient way,

density and pressure--------Reynolds averaging
velocity, internal energy, enthalpy, temperature--------Favre averaging
Favre averaging:
Favre decomposition
mean value
turbulent fluctuations
Reynolds-Averaged Navier-Stokes equations

-------incompressible flow
Reynolds-stress tensor
The sum of the normal Reynolds-stresses divided by density is called

Turbulent kinetic energy
Favre- and Reynolds-Averaged Navier-Stokes equations
------compressible flow
Favre-averaged Reynolds-stress tensor
In turbulence modeling, Morkovins hypothesis

Turbulent structure of a boundary layer is not notably influenced by
the density fluctuations if
This is true generally for Ma<5
in a wall-bounded flow.
Density fluctuations must be considered
Hypersonic flow
Compressible free shear layers
Flows with combustion or significant heat transfer
Eddy-viscosity hypothesis
Boussinesq (1877) observation that the momentum transfer in turbulent
flow is dominated by the mixing caused by large energetic turbulent eddies.
Boussinesq hypothesis : turbulent shear stress is related linearly
to mean rate of the strain as in a laminar flow.
An equilibrium between the turbulence and mean strain field
The proportionality factor is eddy viscosity
eddy viscosity
In compressible flow
KEY: how to model
Boussinesq hypothesis----------first order closures

Limitations:
flow with sudden change of the mean strain rate
flow with significant streamline curvature
flow with rotation and stratfication
flow with boundary layer separation and reattachment
secondary flow in ducts and in turbomachinery
Nonlinear Eddy-viscosity ----------first order closures
Lumley proposed to extend the linear Boussinesq approach by high-order
products of strain and rotation tensor like a Taylor series expansion.
Computation work only slightly increased, but can offer a substantially
improved prediction capabilities for complex flows
Reynolds-Stress transport equation ---------second order closures
It is possible to derive the exact equation for Reynolds-stresses

Taking time average
For incompressible flow
New unknowns
higher-order
correlations
(2) First order closures

Boussinesq hypothesis or nonlinear eddy-viscosity
Zero-equation models
Turbulent fluctuations
Mean flow quantities
Algebraic relations
Underlying assumption: local rate of production and dissipation of
turbulence are approximately equal (equilibrium), and do not include
the convection of turbulence (no history effect)
Equilibriumspecify the length and velocity scale in terms of mean flow
Baldwin-Lomax (1978)
Cebeci-Smith (1974)
Johnson-King (1984) ODE
adverse pressure gradient
Half-equation models
Wilcox (1988)
Flow with no separation, good results for pressure distribution

Not accurate for friction drag and rate of heat transfer.
Simple and easy to incorporate into a numerical code
One-equation models
length scale is specified algebraically
velocity scale is specified using a partial differential equation
namely, one transport equation is derived based on the NS equations.
Baldwin-Barth
Spalart-Allmaras(1994)
No need fine grid near the wall

local easy extend to unstructured grid
Two-equation models
Two transport equation is derived based on the NS equations.
K- model
K- model
K kinetic energy of turbulence

dissipation of turbulence
specific turbulence dissipation rate
Comparison
zero equation model (Baldwin-Lomax)
one equation model (Spalart-Allmaras)
two equation model (K- , K-)
Attached boundary layer:
K- < BL SA < K-
Free shear layer:
K- < K- <= SA
Moderate Separated flow: K-

The most important is to make comparison with the experiment results
no matter what model is applied.
Example: An asymmetric plane diffuser, known as the OBI diffuser.
Effect of different turbulence model on the length of recirculation zone.
Pressure distribution at the bottom wall
Wall shear stress at the bottom wall
Velocity profile at one position x/H=-5.87
Velocity profile at one position x/H=19.53
The last figure is noteworthy, as it demonstrates a clear weakness of

all the tested turbulence models, in that the velocity profile in the
downstream duct of the diffuser is experimentally fully recovered,
while the calculated profiles still show remaining effects of their
earlier separation.
We can hope that the gained knowledge

on turbulence from advanced DNS and
LES simulations will contribute to the
improvement of current turbulence models.

10 Simulations of Turbulence (Cancelled)

Uploaded by

Copyright:

Available Formats

10 Simulations of Turbulence (Cancelled)

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

10 Simulations of Turbulence (Cancelled)

Uploaded by

Copyright:

Available Formats

Turbulent Flows Simulations

1. Theory of Turbulent Flows

Mixing layer, Brown and Roshko(1956)

Turbulent boundary layer Re=4000

A fog of tiny oil droplets + sheet of light

Ejection / Sweep / Burst

Iso-surfaces of the second invariant of the velocity gradient tensor

Practical engineering method

Energy cascade (Richardson,1922)

Kolmogorov hypotheses (1941)

II. Kolmogorovs first similarity hypothesis:

Considering a high Re flow,

Kinetic energy Time scale Rate of transfer of KE

III. Kolmogorovs second similarity hypothesis:

Inertial subrange: <<l<<l0

Motions are determined by viscous effects.

The energy spectrum

With dimensional analysis and the Kolmogorovs local isotropic

Turbulent energy spectrum

Universal equilibrium range

Energy spectrum of turbulence

The Taylor length scales lT

History of Turbulence modelling

DNS (Direct Numerical Simulation )

the distance over which the fluctuating component of the

largest turbulent eddy

on which viscosity is active, i.e., kinetic energy dissipation occurs

Re is Reynolds number in practice.

The number of grid points ( L )3

Nondimensional sizes of the space and time Re L

To obtain enough information in time consequence,

simple flow (homogenous turbulent flow, channel flow,

(2) Numerical Methods

2nd--4th order Runge-Kutta method

Spatial differecing-------energy conservative and low dissipation

(3) The role of DNS: research tool

3. LES (Large Eddy Simulation )

most energy transport

little energy transport

model: subgrid-scale model

LES : resolve large structures

can be applied at Re at least one order of magnitude higher

(3) Spatial filtering

The filtered variable at the location r0

G represents the filter function

r is the position vector

The mostly used filter functions:

Notice that filter functions in physical space limit

(4) Filtered governing equations

Describe the spatial and temporal evolution of the

Describe the effects of the unresolved scales.

Subgrid-Scale Reynolds-stress tensor

The filtered compressible Navier-Stokes equations

Favre-averaged subgrid-scale stress tensor

(5) Subgrid-Scale modeling

Smagorinsky SGS model

numerically cheap and easy to implement

Dynamic SGS model