CFD for Aerospace Engineers

CFD Miguel Angel

Computational Fluid Dynamics BY :
Saez Ortuno

Programs Used : Check for these

•
AN SYS -
CFX and ICEM
and
practice if
possible

History :

1st Man Richardson in 1922

- a
Initial conditions are most important .

Prandtl in 1925
→ Turbulence model

Why using CFD

?

Experiments
too
Experiments can be achieved with similar condition , but sometimes is too expensive
or

dangerous .

Same conditions The ad Ma the

only mater
if are same

CFD Colors For Directors

All be extracted without error but there is uncertainty

.
qualities ca
,

a NO measurement errors ,
cheap ,
hazard free
LES GOOD ONE

analytical Choose wisely

¥#
Sometimes XFOIL better
XX
Is or
.

° .

a RAMS can not

predict all He flows

RAMS
be x MODEL h MODEL To
-

-
E

=
ear
Er #
.

SPALAR

÷L
-

All MARAS

too small separation -

big separation
combining
NO REATACUMENT
both

MENTER

-
 BASIC EQUATIONS AND NUMBERS

PROPERTIES OF FLUIDS

Fluids deform under He influence of shear forces

functional
o

The deformation with to the unbounded and the deformation

respect initial state is
velocity is a

He
of shear
forces
↳ function of He derivative ! SOLID NEWTONIAN FLUID
Solids deformation itself function of He sheer force
°
: is a :

I s a I a
It
In He
fluids there
large number
of
is
,
a
very molecules .
-
-
a
too costly to use direct integration of eauatron
-

of motion
molecules
the fluid instead modelled Contini elements
/
a is
as um
,
with feud
element
→ fluid
a
Macroscopic properties : obtained molecules
averaging over
many fluid

T P P
VV
Knudsen Number : ,
, , , M
X free path ten
kn I I Centrum
= mean •
cc
mechanics Ratio between molecule
J
= a

L L length of transport ten Molecular hitting another

=

dynamics he length
so
and
> so
of the
transport .

TWO OPTIONS TO EVALUATE THE FLOW

Lagrangian approach :
Eulerian approach

s
Reference frame Observer
moves with
fluid at fixed position
a

element remains a

°
Material Field
description
function of
a
description function of and time
as
fluid space
as
element and time ,

RE WOLDS TRANS THEOREM

Tort : Eulerian

Used to
express temporal change of material
a volume integral as He temporal Change of he
integral of Suarhty fixed V He
and
flux Su of this
a
over a domain over He boundary domain

of Sff If
#
old V = out 0 win ios

J(t) CONSERVATION LAWS Lagrangian

- - -

temporal temporal change flux over he Usually formulated as He temporal

change the
og integral
of boundary of He domain derivative material volume
material volume
of
Sv .

of integrals
a
gratify
integral with
over a
fixed a
dynamic boundary
domain V .

Fe = o
Ze DV
fgpcx.tl
= O
 CONSERVATION LAWS

SIMPLIFIED

Inviscid flows :
µ=o
→ Euler equations

Viscous effects can be often neglected at high reach and Re .

Dimensional Analysis

INCOMPRESSIBLE FLUID :
pressure independent of density .

Mass conservation implies volume conservation D. U = O

Energy conservation follows directly from momentum conservation

BARO TROPIC
fluids :
pressure depends only on Ochs Hy

Energy equation is not needed .

DOTEN IAC
flows
-

:
inviscid and rotation
free

Velocity field can be expressed He

as
gradient of scalar potential field .
U =
DOI
Incompressible potential flows I = O

Dimensionless numbers

When He flow mores faster than

wares .

sina.ee
:c
:
.ae .

f. Turbulent water bubles .

REYNOLDS NUMBER :

Navier-Stokes in dimensionless
form

7¥
D. C Kk )
If ¥0 fluid constant density
)
t
t
Op Oy O incompressible with
-
.
=

and constant viscus Hd .

D. I = O

Re I flow The
flow
cc :
creeping
> > s : turbulent

>
Viscous forces are dominant Inertial forces dominant

Nonlinearity leads small scales

>
The linear Stoles
to production of
valid
.

equation is

The Laminar flow

for flows in Porous media .
i O C 2 ) :
 CONSERVATION LAWS Derivations fully in 02 -
BASIC - EQUATIONS .
.
.

Continuity Cauchon Conservation of Conservation

mass
of Momentum

In differential form

Conservation
of energy
work flux heat flux I integral form

In divergence form heat conduction

Gi =
-

he
I
2 Xi

GRID GENERATION

DISCRETIZATION !

Numerical methods based discrete

representation of solution and operators Discrete cells
of
are a
on .
megh .

Various methods to He solution

CFD represent :

Cell averages finite volume discretization robust and fast ICT

-

Point values discretization

:
finite difference at center
of vertex difficult for complex geometry

Coefficients :
finite element dis roti ration of basis functions high order .
but slow .

FINITE VOLUME
finite volume ,
Hey do not overlap .

get average on each finite volume

evaluated he
conservation laws at each FV with
fluxes .
determine evolution

of ke cell average .

Volver integrals → surface integrals

Gauss theorem

crease !

Conservation law

Gauss theorem applied

Momentum conservation example

 FINITE VOLUME

information transferred
Several approximation methods to calculate
numerically .

boundary boundary
→ → → → →

anti hors
Evaluate flux
conditions
at cell faces : reconstruction or interpolation

Surface integral of fluxes : rules

quadrature .

After substituting Suitable boundary conditions ,

we obtain a
system of equations .

DEFINE FINITE VOLUME CELLS

Topology .

.
Relations between
neighboring elements keels

Geometry : Shape and size

of cells and entire domain .

TOPOLOGY :

a Subdomains ca be projected
o
Generally not
possible to
project
onto Cartesian domains

Manly rectangles
land
XD
>
a
Many triangles and tetrahedrons
hexahedrons C 3 -

D)

°
Simple data structure and o
Complex data structure and access

direct access
through connectivity matrix
"
Allows
°

for by
"
hand
grid
a
Straightforward application to
-

Generation good quality complex

,
geometries .

control .

Local grid refinement by dividing

>

a
Simple topology and implementation elements .

Efficient solution algorithms Higher computing t

per element
a
a me

Local grid
requires
refinement
s

STRUCTURED GRIDS
multi bloch cords

The to
napping from physical space
TYPES OF

STRUCTURED GRIDS Computational space is done by

coordinate transformation
 GRID REQUIREMENTS

Grid Enes between All cells volume

must not
overlap or intersect nodes . must have positive .

Should be easy to refine coarsening He grid in selected regions .

The distribution and He boundary elements

spacing of should be fine enough to represent geometry .

The should be smooth than before

grid no more I. s than .

mesh a
pipe flow
The grid shoved be as orthogonal as
possible .

NOT good good

Algebraic distribution function :
He center avoid
in
singularities

Defined o , bloch boundaries

Popolo used :

CONFORMAL MAPPING
geometric stretching
°

a
exponential stretching Defined for an entire block , or a 2b plane
>
hyperbolic stretching The complex transformation
Es = Elt )
maps He
complex 2 -
domain C 2- =
Xtiy )
onto He
complex E domain ( 5=5
IN
-

ELLIPTIC GRID GENERATION

Formulate grid generation value

as a
boundary problem .

Lines of equal potential ( to =

const ) and streamers ( Y -
. const )

form an
orthogonal grid

Either solve Laplace or Red

space
Grid

Poisson equations .

This method for each (5 y) corresponding (

searches pair ,
a x.
y) ,
inverse laplace .

Used to improve manually generated grids .

Refinement aliens point or Are :
using control functions Paro Q .

AUTOMATIC BLOG GENERATOR

I CEM CFD
Does not work for complex
}
cases
Define blocking resolution
.

topology .
Specify requirements and

Works well for simple geometry block boundaries .

Ioem generates automatically

He mesh .
 UNSTRUCTURED GRIDS

o
They offer much better
flexibility .
Tetaheorons as be easier fitted to a
geometry than

hexahedrons .

ADVANCING FRONT METHOD DELAUNAY TRIANGULAR ON

8BggggoG

Sometimes requires final smoothing step : elliptic grid

Also He cell site can be controlled with background mesh
,
.

STRUCTURED GRIDS ARE BETTER

NUMERICAL DIFFUSION :

• Reduce the effect by :

Afon
structured
'T
! flow
'
On
grids wont align with

)
>
Mahi cell faces orthogonal
's to glow direction ,
.

leg will be less precise

Align
.

lines
flow
°
grid with direction .

More numerical error .

Prism Layers :

Combine He advantages of unstructured and

structured
RESOLUTION Of BOUNDARY LAYERS grids Improves accuracy at boundaries
resolutions
.

Recommended
.

There Case bloats

are
gradients at walls

Turbulence production happens mostly at t a 20

y .

The wall shear can be computed accurately and

reliably only
if He first cell is within
Yt e 2
Resolution Wall Models

- Wall functions improve boundary conditions

for simple equilibrium boundary layers .

Allows

for much coarser '

resolution
y

a
CFX includes one based
,
Cee
RANS

Not
>
working for strongly curved walls

at
large pressure gradients and for
non .

equilibrium turbulence

The octree method

Data can be stored

efficiently in a tree data structure

Every leave of the tree corresponds to a structured

grid block .

At block boundaries unstructured grid interpolation

-

rules
necessary
are
.

TURBULENCE

Osborne Reynolds first investigated It .

Laminar flow recovers from Oishi voices at low

velocities and becomes unstable at high speeds .

The .
-

Ludwig Prandtl : discovered stream wise

instability
waves in a boundary layer .

Toll men : found that Peat depends on wavelength

T S and
waves unstable evolve X
-

:
To and hairpin
vortices and
eventually to turbulence .
ORIGIN OF TURBULENCE

°
Very small perturbation Re
grow in shear layers if he exceeds a critical value

The Reet
°

depends on wavelength of he perturbation

the Kelvin Helmholtz instability

°

if Here is
velocity difference across
-

a
appears

He boundary between two

fluid layers

EFFECTS OF TURBULENCE

Enhances Delays flow

mixing separation

Increases Wale friction

increases lift at low speeds
reduce bodies golf ball
pressure drag of buff

TURBULENCE ENERGY CASCADE

Turbulence energy is generated on He largest C

integral ) scales , transferred to medium -

small
Scale
vortices C inertial subrange ) and dissipated in ( sub I .

micro scale vortices C dissipative

scales )

) More

high Rc
pronounced differences
.
at

LARGE
SCALES : Vorhees are
induced
by and strongly depend on geometry and boundary conditions of He

flow .

SMALL SCALES Obtain their from the

large scales through energy cascade
:
energy .

Indirectly affected by geometry and b. c .

Easier to model than loser scales .

CHARACTERISTICS OF TURBULENT FLOW :

Unsteady Rotational Viscous Breaking of symmetries Chaotic

,
, ,
,
,
Wide range of last h

and time scales Potential rotational therefore

Coherent flows
"
.

structures , ,
are no
,

potential will be laminar flows .

SMALLEST
VORTICES

The smallest uonex structures characteristics can be obtained with :

 LARGEST VORTICES :

Integral length of vortex

scale Be the largest structures can be estimated Osias flow geometry .

two point correlation function

The turbulence of two neighbouring points are correlated this function measures he time
correlation between fluctuating Quantities
"

average distance
"

,
at a n
.

If this correlation is
computed for different the Rover will be He Integral
r .

integral of r

Length scale .

DIRECT NUMERICAL SIMULATION ( DNS )

without
Captures all length ad time scales of a turbulent
flow Yu .
ad The directly

assumptions .
Using Nano
stock equations .

These hold for laminar and turbulent flows .

COMPUTATIONAL COST
" "
Number Re
"

than
"
of grid points Nc ~
→ 3 Dimensions NP ~ Re

Number of time
Nt Nc low
steps v we can compute small things at

3 Reynolds number
Total Re
.

cost -
s
Nt .
Nc =

USES :

Useful for fundamental turbulence research but for

not
every
-

day engineering flow

Simulations .
 ( FD FOR AEROSPACE ENGINEERS

Requirements high Re M Complex Transition turbulence

:
,
moderate , geometry , , and separation .

Derive simpler equations from conservation laws that

capture He most important flow
Qualities

METHODS : RAUS steady mean Solution

URANS include effect

unsteady
LES large and energetic scales

DES hybrid methods .

REYNOLDS AVERAGED NAVIER STOKES

REYNOLDS AVERAGING

Ensemble averaged solution

9

↳ integral , average .

For statically stationary processes :

The solution is decomposed in He mean value and He

fluctuation .

Reynolds averaging is an
orthogonal projection :

Average the
average ,

Remind Wavier Stokes

same value
dimensionless
.
 RULES ON REXNOCDS AVERAGING DANS DERIVATION
he
Substitute he sum of mean value and fluctuation into

N S
-

equation .

Apply averages operator , simplify as much as

possible .

mass incompressible momentum conservation

F. u =
O
O
Ju
accuse
to u 't > > C
Je
> =

I 284£
P .

( Kud tcu 'S ) =

O
2L
t
~
t
pressure and density at
Cu O
>

op >
LFtf
Taps > t

tf OLAP
's
Tcu
=

Ip
=

peps > = 0

test
Reynolds
Can )
Q
fkn tulku 'D >
¥00
s > cu '
> u > tu
.

a
-

+
-

( Ee 00 w
> → - u >

=p ⇐ u > sus >

'
.LU
' '
pcw.su > pea >
.
+
> a > u s to . u

=p cu > cu To adhonal
.

> t Cu > Lu
'
> used as independent
.cuc
t 00 * ,

- ↳
-
⑧

COMBINATION OF ALL DANS

Icu >

>
+08k
-u f- OSPD
¥00548
'DOcana
+ '
.
+

at o
-
=

D .

L a > -

o
RAIS
.

More unknowns than equations

Turbulence models provide approx of stress tensor

TRANSPORTATION EQUATION RE RST

FOR STRESS Reynolds Stress Transport .

momentum caution multiply : The arithmetic mean

of both equations is the RST with th
i component with
'

foreal form
Uj
-

j -

component with Ui
'

ther K
apply averaging operator
The order of new unknowns increases during every
step Thus we can not solve it this have to
.

way .
We use empirical approximations TURBULENCE

MODELS

TURBULENCE MODELS

EDDY Viscosity Hi ponies is - EDDY VISCOSITY MODEL ( EVM )

Turbulence leads to momentum

exchange between fluid elements

The deviator to he
of uj proportional
'
RS .

cu i > is mean

IGggG••
,

shear rate .

Proportionality factor is
Cody viscosity Vt .

The turbulence kinetic energy adds turbulence to flow .

Turbulence

Viscosity to add mix ins .

A or dog

Substituting EVM on TANS yields :

NEW pans

turbulence
•

intensity .

Problem reduced to one scalar field :

eddy "
scoots

einaudi
d , ghncfgeow direction
Strong assumption but confirmed DNS for flows
by with boundary lays .

Furthermore Vt can be .

next
page
EDDY VISCOSITY MODELS I CONTINUED )

Dimensional arguments lead to a variety of expressions of eddy viscosity .

Models are characterized by

Young
:b
:
Teamed transport

| .

ZERO -

EQUATION MODELS -
BOUNDARY LAYER METHOD

Mixing Length -
em : typical length scale of turbulence chose properly .

⑦ Turbulent mixing IS only

important if have veto at Eddy viscosity is approx

constant in he outer part of
you

gradients .

the b l

#
-

characteristic
Turbulence length scales smaller hoards He walls .

→
mixing length
is he dstace

from wall
fm
around
Not use in complex flows
he walls .

ONE EQUATION MODEL

Includes turbulence iiersity in eddy viscosity

he turbulence
Transport equation for
" "

kinetic h
energy .

one
half of the trac of ke Re

Thess tensor

- -
 ONE EQUATION : MODEL TRANSPORTATION EQUATION

REVIEW :

JONES I LAUNDER Two EQUATION MODEL h -

E model .

Equilibrium between turbulence production and dis , patron .

Isotropic turbulence .

em = CD
h Vc =
Cy
Partial
differential equation

E E oral transport
adit , equation
because
of complexity of E .
) Postulate E

Same unknowns model But solved E

as in one equation : dis ,
patron is on .

with
Obtained flows for h cancel out
and E
in .

REVIEW
•

@
 Wilcox Two .
EQUATIOU MODEL h .

w model ,
one of most used

Solve the transport equation of turbulence kinetic u and postulated

energy
a

transport equator for He dis

patron W .

Similar To h -

E but ues different

results .

Transport equation for turbulence heretic

energy
CD = 0.09 x =
51g
Pru - 2
Huo
p
-

Pro =
2

dissipation rate
Transport equation for specific
.

Review

S -
A EDDY VISCOSITY TRANSPORT MODELS Spal art .

All maras model

Based postulated transport

on equation for a
functional of he
eddy viscosity .

With NOT Important probably but

anyways
 Review

REYNOLDS STRESS MODELS

Directly solve model transport equations for all components of He unknown The tensor .

Approximations of
higher order correlations
and 0181 patron rate tensor .

REYNOLDS STRESS MODELS ( RSM )

Pressure
diffusion is rather small , neglected Dgp
Triple correlations in He turbulent transport term are approximated by : hanja lie I Launder .

Triple correlation is a combination

of double correlation .

Turbulence dis patron acts small scales scale and decoupling

separation
: on .
we assume

He tub dace cascade

through energy .

The dismal tensor modelled as isotropic tensor represented by a scalar

rate
Eij
on
is an ,

wants .

Eiji } Site
 Pressure correlation disrate turbulence energy
-
Strain cannot produce or .

{ : Katie
Yeon
to isotropic

rapid : immediate effects of mean flow

gradients and external forces .

He He Re
This ten redistributes turbulence between components of stress tensor .

The strain correlation only affects the Re stress lessor

pressure anisotropy aij of He
-

LRR ad 56M differences In

pg 41 42 I
.

Of RANS
.

RECOMMENDATIONS & EXPERIENCE

O s

0
s

LES : LARGE EDDY SIMULATION

male He
The Oo we
main
question is ,
where

separation between large and small ?

Using a
grid or a
filter Big. solved by simulation ad

Small by modelling .
SMALL SCALES
LARGE SCALES

SIMULATION CLASSIFICATION
FILTERS or GRIDS TO Separate scales :

Spatial filter :

.
captures large energetic scales

.
removes small scales .

D :
filter width = grid width

FILTES : SCALE SEPARATION

Convolution integral
"
hence filter Spectral Dutfall
basically
"

average
.

Convolution theorem ,
convolution in real Space I
physical )

corresponds to
.
multiplication in fourier (spectral ) space

COMMON FILTER KERNELS

Aggregators Fleeing produces ke lose -

scale part C the resolved )

Ii =
Go Ui

From He identity hi Tri "

t
hi trivially follow
-

He sub
grid scales

Ii
"

hi =
Ui
-

Filtering is no
orthogonal projection

Ei t Tic Ui to
LES EQUATIONS

dx 6. Substitute
'

ni
"
Wc assume a homogeneous filter 6. due = h hi =
hi , in Tij

Applying this filter to He momentum equation goes .

He
And N S ess for day scales
.

Sub grid Scale Modeling

Receive mats : exact is

impossible .

want to model ke He scales

effect large resolved
we
>
of he sub grid .

scale turbulence on .

>
Observations .

Sub grid
•

scales receive
energy through energy cascade

• Backscatter is much smaller than forward arrow Site

!
•
Dis patron
, happens on unresolved scales .

°
Sobs rid scale models have to He
provide correct
energy drain from He presented scales . If not
,
this can
happen :

SMAGORWSUY MODEL
Turbulence causes momentum

exchange analogously to

He molecular
diffusion
( gas line tics )

Cs 20.2 from energy spectrum of

homogeneous
.

30 tropic turbulence

Using Praotls (o equation ) LES obtain

mixing or we
.

Close to the walls Cs has to be corrected

by van -
Driest damping .
 SMAGORINSUY MODEL REVIEW
DSM : DYNAMIC SMAGORINSKY MODEL

The problem with this model is with he

Cs The dynamic ( DSM )

.
Smagorinsuy proposes
a Cs = Cs Cx , t )

hypothesis :

the
Most of effect of He SGS stress tensor

'S doe to tee interaction btw th Small and

large scales .

Turbulence on those scales are similar

DSM :
CONTINUED
.
The filtered velocity feed is filtered usain lest
using a

filler SI 2B

Tig ad Iij can be modeled

)
~
with he same value for He
dii

neooooe.e.agssigt.no?iiomshs

The germano identity :

Icj =Iij t
Tej connects He models on filler and test filter
level .

LT .

=
Fits - Ei .

E, →
Leonard stress

If we substitute He values :

CD = C CST can not

REVIEW be extracted drag .

Cos C x ,
t ) is filtered by
test filler
If spatial
-

th
.

we neglect variation of Cb we obtain :

Solved by least optime cation

square
-

This particular form proposed by Lilly Haas ) should be

was and proffered
He original Germano
of
over
proposal

SCALE SIMILARITY
MODEL

Approach without He edge viscosity approximation

The resolved filtered clouts substituted into He definition of He Sos tensor

is used as unfiltered velocity and .

Numerically unstable , not enough dissipation .

 APPROXIMATE DECONVOLUTION MODEL ( ADM )

Generalized high -
order scale -

similarity model

Sets Deconvolution of He filtered velocity returns He

expansion of inverse filter operation :

unfiltered velocity with arbitrary accuracy .

More information how it works in

pg .
33 of 06 - LES

model
MIXED MODELS ZANG : combined method with a
dynamic eddy viscosity

one of He best

CURRENT TRENDS

Combining DANS and LES

•
tonal coupling : some a ones use RANS and others LES problems at
coupling conditions .

Detached simulation based on 80 resolution and wall distance

eddy
•
: .

Implicit large eddy

-
simulation ( ILES )

SUMMARY
FINITE VOLUME METHOD

The computational decomposed into

overlapping control volumes : finite volumes C i v )
Space
=
a
is non
.

Every is control volume ke

FV considered as C cu ) for which compute evolution of
s
a we

He mean valves .

The
°

solution represents the cell value

average .

The conservation low is integrated over He FV

Gauss '
theorem applied to transform volume to
is
integrals surface Integrals .

The fluxes THE the FV

over surface Su determine He time evolution of He cell average

balance He
The numerical evaluation of He flux over FV
resoles

approximation methods .

The fluxes have to be approximated from he known cell average

Solutions y

For each FV !

APPROXIMATIONS :

Quadrature :
surface integral af fluxes is a sun of discrete values at one or Several
points
athe cell surface .

Interpolation : Values of 4 at He cell surfaces are reconstructed from He values of 9 at He All

centers
 APPROXIMATION OF
SURFACE INTEGRALS

MID -

point role : ( 2nd order )

Fe =
fgehlds = Ie Ae Ae -

fgesos Ie =
fee fseulds = mean value .

The integral is approximated as He

product of
the integrant Ye with the area
of He cell face .

ERROR :

"

Yen -
-

Yo I 2¥ lot 3¥ lot .
.
.
.
F- x -

xo
,

! WINE -

-
h .
Yo
-12¥ 7¥40
,

If
-
EEI.to u .

error truncation
-
→

Of )

Midpoint rule

The He he
order
of method determines
"
rate of grid convergence for a
sufficiently smooth solution "
.

we car observe that the higher order methods are

actually not so
good for low resolution and large cells .

TRAPEZOIDAL RULE ( )
:
2Nd order
€ogE↳g#②gL€gSfhth order )

APPROXIMATION OF VOLUME INTEGRALS

 INTERPOLATION

To evaluate He
quadrature rules we need to
compute He flux at one or several points at he

cell surface .

These points are reconstructed from He volume integrals by spatial interpolation .

He =L ( Yw ,
Qp , HE . . . )

UPWIND INTERPOLATION Upwind

( differencing scheme .

VDS ) Linear INTERPOLATION ( central Offerman

scheme -
CDS )

TRUNCATION Error DIFFUSION

-

NUMERICAL

Analysed with Taylor series expansion :

Truncation error
of UDS

Proportional to width He xp )
grid
-

EXAMPLE : ODS discretization of linear advection equation :

Effect of numerical approximations on numerically computed solution ?

Upwind approximation :
Taylor series e. isms , on
of y

numerical method

let alone
error
what we wat to compute
EXAMPLE ( CONTINUED )

Substitution
truncation
We neglect He
error

be he truncation
The exact solution will including error .

This error is an ad tonal

01
fusion term . ( Numerical

diffusion )
-

negate I ND ) 's unstable

A numerical method with
XII
.

expressed as constant .

. If be original
At E DX IV term
equation has diffusion
a
already
Max
accuracy
at at - SHU
CFL =

SEE ,
Hn both summed
diffuse ions are .

The =

¥ The
effective
=

Y¥w
Central scheme
TRUNCATION Error ON CCDS ) differencing

He He cell
approx proportional to Square of size

( Xe -

Xp ) C XE -

Xe ) 2nd order method

NUMERICAL DIFFUSION AND DISPERSION

DBtggd

DISCRETIZATION OF GRADIENTS

To evaluate the
diffusion terms we have to approximate gradients of 4 at FV surface .

Viscous stress is
approximated with a Cbs method because M 's closely

corresponds to He
diffusion
isotropic character of a :

Joe
(2×-12
XE k
-

XP
DISCRETIZATION OF GRADIENTS ( CONTINUED )

From the Taylor series expansion follows :

Soooooo
24
( ⇒ = 4E-YP_ Es
e
, g
XE -

Xp

Non Sst method He leading scales with

uniform grids ,
order Since order error Sx

For
uniform grids ,
2nd order method .
The order increases by one .

p e E
• • a

NON UNIFORM GRID

Conclusions
1. Refire cells by splitting them Into equal intervals :
Non He
•
uniform grids car give same

rate of grid convergence as refinement

of uniform grids ,

The tons where He He be

first order appear distance to both sides is not sane .

o Grids should smooth

truncation will than 2nd order method

e global error converge only slightly slower a .

2. Refine cells by splitting hem into Sob intervals stretching

with equal .

The first term faster than He second error tem 20 order

error converges convergence
.

HIGHER ORDER METHODS

Can solution
accurate grid
-
give more on same .

Can be obtained by including

information from cells
-

more
neighbor .

•
Simple for structured grids

• Difficult computationally expensive for unstructured grids .

AN Sys -
CFX INTERPOLATION

1st order Upwind Differencing Scheme VDS a 2Nd order Central Differencing Scheme CDS

2nd order Upwind Differencing Scheme UD S2

1st -
2nd order blend factor ( VDS a c .
> Ubsz ,
O
Ep Es )
 UNSTEADY PROBLEMS

For unsteady time be discretized 4th dimension

has to aswell making it a
problems
°
.

the
⇒
Unsteady problems ,
He future has no influence on past ,
thus they are
parabolic in time .

value solution depends initial

°

Unsteady problems = Initial boundary problems I on conditions and boundary

conditions )
a Same method as spatial discretization

EXAMPLE conservation
law What the solution
: is
y at tries th = to + St

Integration

ME MARCHING METHODS

For small It all methods to He sane solution

a
converge

fast it to IS At small
The order of a method bees how converses too is sufficiently .

CFL
Styx cos is good enough for time integration
=
errors .

IMPLICIT Values of 4 at t > En EXPLICIT 4 at

tn
only
ADVANTAGES
ADVANTAGES
No iteration necessary efficient
,

Stable much large time steps

for straight forward implementation
DISADVANTAGES Low requirements
memory

Large memory refinements DISADVANTAGES

iterative G , complex implementation Unstable for large time steps .

Careful with and

implicit time marching LES CFL
gives max step
 LOCAL us PHYSICAL TIME ST ED

limited
The maximum reasonable time step she At is by flow physics and or numerical stability .

The solution must not

physically large distance than what covered by the
propagate over a is

influence domain of He
computational steal .

Influence domain : h Cz cell width )

Improvements by

Signal speed : S =
lol , c compressible St = Stix )

S -

-
I U I incompressible

time
Max step :
Its men his → Appropriate time
step Is often limited by smallest cell in domain .

the transl hat

Local time step But is
Conversed steady state results identical but obtained faster with .

in are

be used accelerator time dual

stepping methods
not good . Can as on
in .

DUAL TIME STEPPING METHOD

Split discrete time oeiuohue into :

•
Large physical time step At , using an
implicit method .

Pseudo
part
•
-
time se
Ltda

Discretization with Euler implicit for physical time and Euler explicit with local time step she for pseudo time .

-
-

time time
pseudo physical

AN SYS -
CFX

Dual time stepping method Local or physical time step size

Impose CFL if needed

implicit )
.

1st order Euler C

2nd order Euler C implicit )

 BOUNDARY CONDITIONS

EXAMPLE : ID Diffusion - advection equation Time : 1st order backward Euler c implicit )
god

order
Space : 2nd central
Offences c uniform grid )

Discrete equation :

As St TED

The all advection and diffusion

Sumatran of terms leads to a linear
algebraic
Order by index of i
system

Set of algebraic es . for ever consecration

law and
every cell .

Index P PDE
:
post where we
approx

Index i : runs over all cells

Ap ,
Ai : depend on
gnd ,
geometry
and
fluid properties .

Gp : includes all hows terms at

previous
LINEAR 17A -

ION : how to treat nonlinear PDEs time level .

Navier Stoles we a
system of nonlinear PDEs

Interaction of vortices o.O turbulence result from He grad rake

nonlinearity .

N
Implicit time discretization methods applied to linearized version of Sean
-

are a
.

BOUNDARY CONDITIONS

Discrete operators He flow evolution in a cell as a Possible choices for linear ,

approx try :

function He Solution in neighbor cells But do

of .

,
what we

do at He boundaries
domain
?

o N .

s describe initial boundary value

problems

well
° A
posed problem requires correct

initial conditions and correct boundary

conditions

TYPES OF BOUNDARY CONDITIONS :

I Diriohet boundary condition

Robbin
:
3 boundary condition
.
:
.

Impose value of value ( Y ) on boundary

Combination of Dirichlet and Neumann

Neumann
Us Periodic boundary condition :
2 . boundary condition :

variable should be
Chose bonding surface such that values of
Impose gradient of variable ( doin ) on
He
boundary He same at two opposite walls .
EXAMPLE : NO SLIP WALL

Advection I convection through wall is NULL

No slip yield
-
condition plus incompressible continuity equation in relative wall coordinates .

Pressure gradient normal to wall is NULL

Neumann boundary condition for pressure .

Viscous stress :

Viscous normal :
Viscous tangential : one side
differences approximation .

we can also use wall -

stress model to approx

viscous shear . Incorporate modeled steer stress

as source ten .

EXAMPLE : SYMMETRY

that confines the other He

Assume a mirrored solution on side of bondage :

No flow through He boundary

ftp.T#ffAqgggdlSapccrs Tangential viscous shear stress of U comp

-

Normal viscous before

is
approximated as

EXAMPLE : INFLOW EXAMPLE : Outflow

SUPERSONIC :
SUPERSONIC :

Single N Sea independent variables

species 5 No dirichlet b.
-

upstream influence No
-

, c
to be specified
.
.

have as Dvichet boundary

conditions . Only Neumann be .

3x
density momentum and
energy
.

, ,

temperature
'

pressure ,
velocity vector
SUBSONIC I
INCOMPRESSIBLE
.

SUBSONIC I b.
:
variable as Dirichlet c .
usually static

U pressure
independent variables Dirichlet
.

as boundary

Density and 3xmomentum

*•
Po
,
To
, flow direction .
 BEST Practice For BOUNDARY CONDITIONS

Boundary conditions approximate the

>
reality

> To limit un
physical effects , hey should be located as far away as
possible from region of interest .

o
Interference flow can lead to un physical oscillations and reflections .

The He
choice of boundary conditions significantly affect rate
•
can
convergence .

°
Avoid "

conditions and He
Opening never determine problem
"
over
-
.

PRESSURE VELOCITY COUPLING

COMPRESSIBLE N S EQ INCOMPRESSIBLE N S EQ
-
-
.

Tosefta

DISCRITTATLON PRESSURE
OF THE Poisson

Pressure term
Momentum equation x -
dir

Central differences scheme on

uniform grid

Pressure is effectively discretized on coarser grid

Velocity and pressure of cell P are decoupled .

Pressure Poisson with

equation CDS
Neither mass nor momentum couple pressure
ceg .

and speed at any point Oscillations on

pressure
.

can occur and remain undetected

by ice numerical

scheme !
RULE & Chow INTERPOLATION

Pressure
oscillations can occur and are not damped by th numerical scheme if we di soothe
pressure
and velocity with CDs on a common grid .

>
Coupling is restored by :

different grids
C
staggered ) for pressure and
velocity .

Special interpolation techniques for He Rhee

•
s I Chow
pressure
-
.
:

METHOD RHI 't & CHOW :

Tay cored interpolation method for the mass flux through cell surface .

Include th
gradient interpolation rules for velocity finite volume
pressure into be advection across -

cell faces .
.
.
NOT important probably .

EXAMPLE

K¥1
c-

SOLUTION ALGORITHMS FOR LARGE LINEAR SYSTEMS

LINEAR ALGEBRAIC SYSTEM

Discretization leads to a linear algebraic system

two solution methods

'
4 = A .
b direct or iterative
DIRECT SOLUTION METHOD

Invert the matrix A to obtain solution

=
UIbe
AI
? Problem :
memory requirements -
I
She of inverse matrix A ,

A is tipi call , lot of zeros in He entries

sparse : a

but A
-2
does not have to be .

GAUSS -

ELIMINATION

Multiply tee
first row of A with
Says , ,
and Subs tract
from second row .

Last line of U has

only one nonzero
city

Requires OCN 3) operation for Ouse C Non -

sparse )

matrices

LU ( lower -

upper ) factorization

Matrix A be decomposed product of 2 triangular

can in
matrices L and U .

|%%
!
%
" " " " " " " " " : " " " "

"

independent of hand
Advantage : decomposition is right side

Disadvantage : d and U are dense also if A is a

sparse matrix

DIRECT METHODS ITERATIVE METHODS

 ITERATIVE SOLUTION METHODS

Split A as a
part N that can be inverted easily ad a
part
P for which computing he inverse is difficult

The linear
system is re -
written accordingly Use symple iteration Scheme

Alternative iteration scheme ,

which is better in terms of round -

off errors :

RESIDUUM : Error .

be complex He
In
"
be solved but it will as as
problem
They related It E
.

are :
= can
,

Residuum converses before

ITERATION matrix 6 : example

Change of error :

Necessary for convergent is that spectral radius of G

IS smaller than A he less radius

faster .

JACOBI METHOD

Spectral radius of iteration matrix for Poisson es on

Decomposition of A
.

uniform grid
•

Diagonal elements into N

All the rest into P

goes
•

CHARACTERISTICS :

Simple and robust

Ves slow convergence for large problems .

 GAUSS -
SE DEL METHOD

Decomposition of A :

•
Diagonal elements into N

• Elements in lower triangle go formally into N

, however ,
not inverted but multiplied with Ints
•
Elements In triangle are Pao will be multiplied with
"
upper in
Y .

Jacobi but need to remember yn

Implementation same as
,

Spectral radius same as Jacob '

relaxation
Accelerate it
faster with successive over -

INCOMPLETE LU FACTORIZATION

a A is approximated by product of L and

U .

too
that non
only computes elements
are
Iw
-

>

in A .

a dad U are
sparse if A is sparse .

Compared with complete factorization L and well manageable

: U are
sparse as
.
Memory is .

CFX

Coupled Iw solver for u ,

V , w and p

Requires than uncoupled

more
memory and operations
methods

More robust and Faster Less iterations

convergence . .

MULTI GRID METHODS

computational time
Adapt he grid to match He begin ad decrease
.

OBSERVATION : error

Iterative solvers quickly reduce errors wth a small wave length ,

which is on the order of
the cell size .

Errors with Cage wave length converse veg slowly .

IDEA .

Converge acceleration by adapting snd to error wavelength :

Solve for pot short length fire Sid

Sylve
grid with wave on

for part with lose length On a

.
.
• wave coarse

.
Interpolate solution back to fire grid .
 MULTI GRID METHOD ( CONTINUED )

By
He grid Case ke
projet ing error on a coarser , ware lengths appear smaller be Soho .

thus computational time decreased

Algebraic bigger is
system grid much
.

is
°

in coarse

a Direct solvers are used on coarsest

grid level .

•
Information exchange :
.
Method :

It

•
Fine → coarse : = Restriction

•
Coarse →
fine :-. Prolongation

Many ways
for walking through the grid hierarchy .

Most popular the Vad w

are
cycles .

MULTI GRID METHODS :

defined automatically serrated grid

grid based or
Geometric Multi Coarsening
•
is on a user

•
Algebraic Multi grid Coarser in
, is based on coefficient matrix .

ALGEBRAIC MULTI 6 RID

The
He original ( fine ) grid
discretization is done only once on

The are obtained by summation from He fine grid

°
discrete equations ( coefficient matrix ) for He coarser grids .

lines and columns C cells ) are nosed

value of coefficients determines which
•
Absolute

Irregular shapes
 VERIFICATION & VALIDATION

VERIFICATION with known solutions VALIDATION : comparisons

with experimental data
comparison

Is th code what
°
Simulation good representation of reality ?
doing supposed to do ?
°

is

assumptions
O

Inference of model
Influence of numerical
°

approximations

Physical
°

issue
> Mathematical issues .

Modelling error C validation )

Numerical error C verification )

Types OF Errors
METHODOLOGY FOR VENFICATON

Requires with
Sources Numerical errors :
2 comparison :

.
analytical solutions

.
semi analytical solutions

.
benchmark Solutions

2. Code comparisons
=
verification activity !

3 that
. Accuracy recs .
is
after more stishtent in

validation activities .

TUERM NOLO 67 4 Document verification

.

5 . Evidence gathered from He users .

6 Must be repealed for He

.

ay change in code

Consistent schemes It U exact .

Uh ,
ell = ① ( HD ,
JP )

DISCRETIZATION ERROR numerical diffusion .

Depends on
grid quality , cell size and time step .

If possible select a high order method depends cell resolution

enables He separation btw numerical

Grid convergence study
CONVERGENCE ERROR truncation and error of physical model .

Residual 20-3 are not enough

 COMPUTATIONAL GRID

The resolution of He boundary layers is important for :

.
Friction coefficient

.
Flow separation

Transition and turbulence

The all relevant qualities should be resolved

gradient of
The
geometry should be resolved .
When this is not
possible simplify
,
He geometry before meshing .

The quality He grid critical for

of is :
Ctx
.

•
Accuracy of discreet ta ton Wall functions switched on if :

behaviour Laminar sub lager unresolved

Convergence is
- -

An epsilon -

based turbulence model IS selected .

Time step SIZE

Critically analyse vale of yt , use yet CI at wall -

Error ESTIMATION

Asymptotic convergence should be verified using at least 3 solutions D

,
Diz ,
$ , u

Singularities and discontinuities Omphale He analysis .

METHODOLOGY FOR VALIDATION

MODELLING ERROR check with model assumptions
:
consistency

Beguines
experimental data .

> Can be performed on subsystems and unit tests .

MODELLING ERROR TURBULENCE :

Choice of model ca affect He results .

LES is
good

MODELLING ERROR -

THERMODYNAMICS

The standard cello under dry and

-
parameters are

lukewarm conditions .

The transition In humid temperature charges

btw Camino turbulent flow reality
'
and
point and ,
air .

should be well captured Cows have temp

Empirical workin range
.

o a on

SUMMARY a
Ideal Gas also has a raise

DOMAIN SIZE AND BOUNDARY CONDITIONS

CFD b. c .
are not realistic

Experiments have errors

Inflow and far freed critical

Echota
of b. c
influences convergence