Turbulence Modeling in Phoenics

Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 37

TURBULENCE MODELING IN PHOENICS

• The purpose of this 2 hour class is to give basic directions to


undergraduate students to use Phoenics to simulate turbulent
flows.

• The lecture gives a brief introduction to turbulent flows


features, to scales and to the law of the wall.

• The lecture applies to RANS models (algebraic and two


equation models) and the focus is on the selection of a proper
grid size near the wall to have a successful turbulent flow
simulation.
Turbulent flow features
• Fluctuatiation or Irregularity: turbulent
flow is random ie not deterministic.
The velocity fluctuates in all three
directions, so does the pressure and
temperature.
• Turbulent flow is intrinsically transient.

• Increased exchange of momentum: Diffusivity - spreading rate of jets,


boundary layers etc.
• Large Reynolds numbers;
• Dissipation of kinetic energy to internal energy
• Wide range of time and length scales
• Almost all practical flows are turbulent.
Turbulent flow features: Eddies
Side view of a turbulent boundary layer Water jet at symmetry plane, Re 2300

• Eddies are "packets" of fluid (identifiable flow structures) with different


sizes ranging from macroscopic dimensions to the microscopic scale also
known as Kolmogorov scale.
– The large eddies are responsible to the transport of mass, momentum and heat.
– The smallest eddies dissipate kinetic energy into heat through viscosity
• The largest eddies scale to the characteristic flow dimension.
– Boundary layer ~ boundary layer thickness;
– pipe flow ~ pipe diamenter;
– jet flow~ jet width and so forth.
Kolomogorov (1941) theory: the energy cascade
• The energy is transferred from the mean flow to produce the large eddies.
• The energy of the large eddies feed smaller eddies and these in turn transfer
energy to smaller eddies yet.
• This process results in a transfer of energy in the form of a cascade from
larger eddies to smaller ones.
• The smaller eddies dissipates the energy into heat due to the viscosity action.
• The energy cascade suggest the multiple scales (size and frequencies) present
in turbulent flow. This feature difficults the development of turbulence models.

Large eddies, Small eddies dissipates energy


contain kinetic energy into heat (viscosity)

“Big-size whirls have little whirls


that fed on their velocity
Little whirls have lesser whirls
and so on to viscosity”
Kolomogorov (1941) theory: length scales
• Turbulence phenomenon has multiples scales: the largest contain energy and
the smallest are responsible to dissipate energy.
• The scales refers to the time, frequency spectrum or length sizes.
• The ratio between the largest ‘l ‘ to the smallest scales ‘lk ‘is:
u
 Re3 4 where Re  and u*   w 
k 
• The length scale ratio impacts on the grid size definition necessary to
encompass the largest to the smallest length scales.
• Consider the turbulent flow in a plane channel:
Re Rel No
(U.l/) (u*l/) Nodes
2.00E+04 116 4E+04
1.00E+06 1623 4E+09
• For Re of 2.104 it is necessary a 40000 nodes grid to a channel volume
equivalent to a one hydraulic diameter; for Re of 106 it is necessary a billion
nodes grid!
The eddies’ sizes and Re
• The images below displays a jet flow with Re of 400 and 20000.
Observe as Re increase the refinement of the length scale of the
3 4
eddies due to Kolmogorov scale law: k  Re

• The eddies’ smallest scale lk ratio for Re of 400 and 20000 is of 19:1
Prediction Methods

l h = l/ReL3/4
Direct numerical simulation (DNS)

Large eddy simulation (LES)

Reynolds averaged Navier-Stokes equations (RANS)

The focus of this presentation rests on RANS models only


Numerical Simulations : estimates of cost
of fixed wing calculation

Sample fixed wing of AR=10, Re=5E+06


Boussinesq hypothesis
• Using the suffix notation where i, j, and k denote the x, y, and z
directions respectively, viscous stresses are given by:

 ui u j 
 ij   eij     

 x j xi 

• Many turbulence models are based upon the Boussinesq (1877)


hypothesis: Reynolds stresses are linked to the mean rate of
deformation.
 U i U j 
ij    u i ' u j '   t   
 x x
 j i 

• Exceptions are DNS and partially LES.

9
Predicting the turbulent viscosity
• The turbulent viscosity can be estimated by algebraic models or by
solving a system of PDE.
• Algebraic models (mixing length and LVEL) employs expressions
similar to: T = C(dU/dy). They are computationally cheap but limited
to very simple flows (pipe and boundary layers)
• The 2nd category usually solves two equations to get T . So they are
called by two equation models: k-, k-, k-l, k-  RNG, etc. For k-:
T = Ck2/ ; T derived by other 2 eq. models similarly follows k-.
• There is also the Reynolds stress model which instead of estimate T it
solves a system of six turbulent stress equations in order to get the
Reynolds turbulent stress tensor.
• The two equation models are largely applied to solve industrial
problems. Very often the constants are tuned to specific cases to
increase the accuracy.
10
Comparison of RANS turbulence models

Model Strengths Weaknesses


Spalart- Economical (1-eq.); good track
Not very widely tested yet; lack of submodels
record for mildly complex B.L.
Allmaras (e.g. combustion, buoyancy).
type of flows.
Mediocre results for complex flows with severe
Robust, economical, reasonably pressure gradients, strong streamline curvature,
STD k- accurate; long accumulated swirl and rotation. Predicts that round jets spread
performance data. 15% faster than planar jets whereas in actuality
they spread 15% slower.
Good for moderately complex
Subjected to limitations due to isotropic eddy
behavior like jet impingement,
RNG k- separating flows, swirling flows,
viscosity assumption. Same problem with round
jets as standard k-.
and secondary flows.

Realizable Offers largely the same benefits


Subjected to limitations due to isotropic eddy
as RNG but also resolves the
k- viscosity assumption.
round-jet anomaly.

Physically most complete model


Reynolds (history, transport, and anisotropy Requires more cpu effort (2-3x); tightly coupled
Stress Model of turbulent stresses are all momentum and turbulence equations.
accounted for).

11
Wall bounded flows
• The presence of a wall damps the velocity and the turbulence near the
wall region.
• To capture the wall effect the turbulence models have to be modified
near the wall.
• The embodied wall models on the turbulence models are grid
dependent.
• The success of turbulent flow simulation starts on a careful choice of
the grid spacing near the wall.
• This presentation focus on near wall grid spacing selection to a
successful simulation. To beginners this a potential area to make
mistakes
Wall bounded flows and velocity damping near the wall

• The presence of a wall damps the


velocity and the turbulence near the
wall region.
• To capture the wall effect the
turbulence models have to be
modified near the wall.
• The success of turbulent flow
simulation starts on a careful choice
of the grid spacing near the wall.

Experimental turbulent boundary layer


vel. Profiles for various pressure
gradients. Coles 1968
The meaning of the layers
• Inner layer: is the layer closest to the wall and is ruled by the fluid
viscosity. The viscous length scale /u* is greater than the wall
distance, or yu*/ = y+ << 1. On this region, u+ = y+ or  = u/y. For
modeling purposes the inner layer extends up to y+ = 5
• Overlap layer: is the region where the viscous action start loosing
influence and the inertial effects increases. For modeling purposes this
layer ranges from y+ = 5 up to 40.
• Log layer: it is sufficiently far from the wall such that it is completely
rulled by the inertial effects but close enough to the wall that the shear
stress is uniform and equal to the wall shear stress! Usually the log
layer ranges from y+ = 40 up to 200
• Outter layer: flow dominated by the eddies inertial effects, y+ > 200.
Inner + Buffer + Overlap Layers
The velocity profiles near the wall are universally expressed in a
dimensionless form:

 y  v*
y 

 u
u  *
u u
u 
v* where
u*   W 
is the eddy
velocity scale.
5 40 200
 y  v*
y 
Wall 
layers Inner Overlap Log Outter
Overlap: sensitivity to adverse pressure gradient
Phoenics turbulence models
• Algebraic models: constant turbulent viscosity, mixing length, lvel.
Lvel is the most popular, it can be applied starting from the inner layer
or from the log layer. It is an algebraic expression which fits the near
wall flow. These models are computationally cheap and delivers good
results for channel or boundary layer flows.
• Two equation models (standard) : refers to the k-e and variants. The
distance from the wall of the first volume center has to lay within the
log layer, 40 < y+ < 200.
• Two equation models (low Re) : turbulent low Reynolds flows usually
have the first node distance for 40 < y+ < 200 corresponding to a
significant percentage of the domain (> 15%). To avoid numerical
errors induced by the coarse grid it is preferred to starting integrating
the flow starting from the wall. The first node must lay at y+ ~ 1 and at
least 3 more volumes up to y+ < 5. This assures a smooth integration
along the inner, overlap and log layer.
How to get good results from turbulence models
• All models have limitations, one can not ask more than the model
can deliver.
• Algebraic models are the simplest and the most limited. Usually
work well for pipe flows and boundary layers.
• 2 equation models are more accurate than algebraic models but
their constants do not have universally, it is possible for certain types
of flow a better matching by tuning the constants.
• 2 equation models considers flow in equilibrium: all the
production of k is dissipated at the same point. Flow with sudden
changes in direction are prone to be out of equilibrium.
• In general, turbulent models demand a special care with the near
wall mesh. If you get this wrong your simulation will fail.
WKSP #1-Turbulent flow develop. 2D channel
• This WKSH deals with a turbulent flow development along a 2D
channel flow.
• The flow simulations will be done using Parabolic model (visit (1)
and (2) links to further information).
• Parabolic model apply only to one-way flows (no recirculation
zones present). It will used along the WKSH because it is far more
efficient than Elliptic model.
• Along the WKSH will be supplied specific hints to set up a problem
employing Parabolic model.
2D Channel dimensions & properties
y nwall (plate) (b) nooutlet Uniform GRID (c)
Inlet specification NX=1
NY=28
H/2=0.05m NZ=80
Center line 2H  U
z (a) Re 
L = 10m 

(a) For parabolic model the main flow direction has to be


aligned with the Z axis.
(b) Parabolic flow has w component always along z > 0
direction, if happens to have w along z<0 it will fail. Due
this feature it does not need specification of an outlet.
2D Channel dimensions & properties
y nwall (plate) (b) nooutlet Uniform GRID (c)
Inlet specification NX=1
NY=24
2H  U H = 0.05m
Re   105 NZ=80

Center line (a) Tolerance 10-6
z
L = 10m

S E T T I N G PA R A B O L I C M O D E L :
Model: choose parabolic (confined because is a channel)
Ground: Parabolic model visits only one slab at a time and stores only the last slab.
To provide full field storage go to VR-EDITOR box GRND and set IDSPA=1,
IDSPB=1 and IDSPC=80 (always equal to NZ), IDSPD=1

ACCESS THE VR-EDITOR B O X E S A N D S E T:


Properties Numerics Output Model Objects
(~air) Iterac =1000 Pause at end of run K-e Inlet: Win = 10 m/s
 = 1.0 kg/m3 Relax (manual) Storg.: YPLUS & STRS Re = 1.105
 = 10-5 m2/s 1; 1; 1; 0,5; 0,5    STRS W Inlet: 5% turbulence
YPLUS  W

 Plate: no slip, W=0
Phoenics variables
2D Channel Results –
W velocity at z/D of 2D, 10D and 40D

One may reproduce this figure using autoplot or get similar results
employing ‘Ploting variable’ in VRVIEWER.
2D Channel Results – W center-line velocity
overshoot

experimental
overshoot,
see link

16D 32D 48D 64D 80D


 1st volume distance correct: 40 < Y+ < 100.
2D Channel Results – Y+ & STRS
1st volume within Log Layer!

 STRS = w/
 STRS fully developed = 0.192
 STRS Colebrook-White = 0.195 (1.5% off)

16D 32D 48D 64D


2D Channel Results – T, k and 
• Viscosity is a flow property: T = Ck2/
• Near the wall the ‘turbulent viscosity’ is 16 times
greater than the molecular viscosity.
• The largest changes on k and  are near the wall.
• At the centerline T is nearly 300 greater than .

ENUT

KE

EP

16D 32D 48D 64D 80D


Phoenics adjustable constants for KE model

• The choice of one constant value by other depends on previous


knowledge about a specific flow and also knowledge about the
meaning of the constants within the model, for KE visit turbulence.
• Evaluating Cf thru Colebrook-White is possible to estimate, at the fully
Exploring the grid sensitivity to Y+ and STRS
developed region: (i) 1st node distance from the wall (), (ii) number of volumes
NY (uniformly distributed), (iii) STRS = w/. The last will be used to check the
numerical solution. Keep in mind that H = 50mm.
W U 2Cf
STRS  
 2

KE – Low Re

KE – Low Re
KE - Standard

KE - Standard
Workshop#1 – adjusting grids
• Based on the table below select between ‘standard’ or ‘low Re’ KE
models to cases where W = 1 and 50 m/s, also define a new NY. Do
these changes on the previous Q1.
• For reference, the previous case velocity is W = 10m/s or Re 105,
• STRS Cole is the w/ value estimated by Colebrook-White for a fully
developed flow.
• Y+ = 1 thru 100 are the estimated first node distance to the wall.

W Re2H Cf Cole STRS Cole Y+=1 Y+=40 Y+=100


(m/s) (---) (---) (m/s)2 mm mm mm
1 1.00E+04 0.0077 0.0039 0.2 6.4 16.1
10 1.00E+05 0.0045 0.2248 0.02 0.84 2.11
50 5.00E+05 0.0033 4.1116 0.005 0.20 0.49
Workshop – adjusting grids
W Re2H KE NY Q1
(m/s) (---)
1 1.00E+04 LowRe 250 unif 2 reg
10 1.00E+05 Standard 24 unif
50 5.00E+06 Standard 102 unif ellip

• Case W = 1m/s – ‘Low Re’ because the 1st volume for y+ = 40 would
be at 7mm from the wall which is nearly 14% of the Y length. One can
use a uniform grid or a more economical 2 region grid, the near wall
grid extends up to y+ = 40 (~7mm)
• Case W = 50m/s – Standard,
Workshop#2 – Flow in backward face step
• A simple backward step flow challenges the 2 equation turbulence
models because it has a shear layer, a recirculation zone, null wall
shear stress point and a redeveloping boundary layer!

• Compare the reattachment point position employing: k-e model, the


Chen-Kim k-e model, the RNG k-e model, the k-omega model and the
LVEL model.
Workshop#2 – Flow in backward face step
• Load case T103 from library. Step height, H = 0.038m and ReH =
45000.
Y
L = 0.762 m or 20H
Uin = 13 m/s
3H
H s = 0.1524 m
X
or 4H

• Change X grid regions to 10 and 60 cells (no power)


• Change Y grid regions to 8 and 10 cells (the last with pwr 1.5 sym)
• Start with the KE standard model
• Store STRS and YPLUS
Workshop#2 – Flow in backward face step

• Key flow features to


explore are the wall shear
stress and if possible the
pressure distribution at the
channel bottom wall.
• The reattachment point
is determined by the x
distance where the wall
shear is null. stress is null.
Workshop#2 – Flow in backward face step

• Fill in the table

x (m) x/H
k-e
Chen-Kim k-e
RNG k-e
k-omega
LVEL
Workshop#2 – Flow in backward face step

• Discuss the validity


Lower bound
of the results in view to two eq.
Standard
models
(k-e model) of the y+

values

• As the reattachment point is approached w 0 and Y+ 0


• Would you consider Low Re type models to capture the
reattachment point?
• If so use k-e low Re, same grid as before but modify Y 1st region to
60 cells, use pwr of + 1.6 and in Numerics set to 10000 sweeps
Workshop#2 – Flow in
backward face step
• The k-e low re apparently
estimates nearly the same
reattachment point position as
predicted by k-e standard!
• But the magnitude of the
STRS has changed
significantly!
• In fact the k-e standard made
wrong STRS estimations
because it employed an
inappropriate grid which
resulted in y+ values lower than
40 at the neighborhood of
reattachment point!
END
2D Channel dimensions & properties
y nwall (plate) (b) no
outlet Uniform GRID (c)
Inlet specification NX=1
NY=28
H/2=0.05m NZ=80
Center line 2H  U
z (a) Re 
L = 10m 

(a) For parabolic model the main flow direction has to be aligned with the Z axis.
(b) Parabolic flow has w component always along z > 0 direction, if happens to have
w along z<0 it will fail. Due this feature it does not need specification of an outlet.
(c) Parabolic model visits only one slab at a time and stores only the last slab. To
provide full field storage go to VR-EDITOR box GRND and set IDSPA=1, IDSPB=1
and IDSPC=80 (always equal to NZ), IDSPD=1

ACCESS THE VR-EDITOR B O X E S A N D S E T:


Properties Numerics Output Models Objects
(~air) Lsweep = 100 Pause at end of run K-E Inlet: Win = 10 m/s
 = 1.0 kg/m3 Relax (manual) Storg.: YPLUS & STRS Re = 2.105
 = 10-5 m2/s YPLUS 
  
W STRS 
 W Inlet: 5% turbulence
 
Plate: no slip, W=0
Phoenics variables

You might also like