sirpa kallio, vtt - Åbo akademiusers.abo.fi/rzevenho/icfd sk abo_akademi2019_slides.pdf · vtt...
TRANSCRIPT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND1
Multiphase flow dynamicsSirpa Kallio, VTT
Boiling water
Turbulent fluidized bed
VTT TECHNICAL RESEARCH CENTRE OF FINLAND2
Outline
t Classification of multi-phase flows
t Modeling approaches
– Homogeneous flow model– Lagrangian particle tracking
– General description– Methods for dense suspensions (DPM, MP-PIC)
– Eulerian-Eulerian two-fluid models– Algebraic slip mixture model– Other approaches
t Example: Eulerian-Eulerian modeling of dense gas-solid flow
VTT TECHNICAL RESEARCH CENTRE OF FINLAND3
Multiphase flows can be classified according to
• the combination of phases:
• gas-solid (fluidized beds, conveying, combustors, etc.)• gas-liquid (sprays, nuclear reactors)• liquid-solid (crystallisators, sedimentation, fiber suspensions)• liquid-liquid (oil-water separation)• gas-liquid-solid (catalytic cracking)• etc.
VTT TECHNICAL RESEARCH CENTRE OF FINLAND4
Multiphase flows can be classified according to
• The volume (or mass) fraction of the dispersed phase• dilute (relaxation time tp < collision time tc)• dense
• The type of distribution of the dispersed phase• homogeneous• heterogeneous (e.g. clusters, sprays, flocks)
• The geometry of the phase interfaces• separated flows• mixed flows• dispersed flows
)1(Re3
4)1(Re
18
2
>=<= pfpDf
pppp
f
ppp C
dt
dt
urr
mr
VTT TECHNICAL RESEARCH CENTRE OF FINLAND5
• Multiphase flows can be treated as single phase flows when• the amount of the dispersed phase(s) is very small• the carrier phase is not affected• the dispersed phase(s) are of no interest
• In special cases single phase flow can be treated as multiphaseflow, e.g.
• in mixing of fluid streams of different temperature and/orconcentration• in the description of the different scales of turbulence
• A computational phase in multiphase CFD is not same as aphase in the physical sense
• computational phase represents a mass moving at a ‘single’velocity
VTT TECHNICAL RESEARCH CENTRE OF FINLAND6
TYPES OF PHASE INTERACTION
One-way coupling:
Fluid flow affects particles/dispersed phase
Two-way coupling:
Above + dispersed particles modify fluid flow
Four-way coupling:
Above + particle-particle interaction
VTT TECHNICAL RESEARCH CENTRE OF FINLAND7
Multiphase flow and turbulence
t multiphase flows are typically by nature ’turbulent’ (phaseinterfaces distort the flow field which leads to fluctuations invelocities and other properties)
t dispersed phase modifies carrier phase turbulence– big particles enhance turbulence– small particles damp turbulence– carrier phase turbulence becomes unisotropic
t carrier phase turbulence enhances dispersion and clustering of thedispersed phase
t in dense suspensions the turbulent fluctuations in the differentphases are strongly coupled and often of a different character thanin single-phase flows
VTT TECHNICAL RESEARCH CENTRE OF FINLAND8
Approach 1: Homogeneous flow model
- Simplest multiphase CFD model
- Assumes that dispersed phase travels at the same velocityas the carrier phase
- Mixture momentum equation
- Mixture continuity equation
- Phase continuity equations: transport of dispersed phase through ‘diffusion’
- Fast; describes well particle size and other distributions
- Describes well effects of turbulence on mixing of the dispersed phase
VTT TECHNICAL RESEARCH CENTRE OF FINLAND9
Approach 2: Particle tracking methods
• also called Lagrangian approach
• utilizes the equation of motion of a single particle
• initially used for dilute suspensions (one-way and two-way coupling)
• natural description for mixtures of particles with different properties
• usable for steady state simulation only in cases with distinct inlets(eg. sprays, burners)
• can be used for dense suspensions if collisions are accounted for(the so-called micro-scale models; complicated and slow)
VTT TECHNICAL RESEARCH CENTRE OF FINLAND10
Equation of motion
tt
trmpmp d
t
rdd
rr
r
rdtdm
dtd
mmmdt
dm
t
t
fp
fp
ffpfp
fpfp
fp
fpff
ffpp
p
ò -
÷÷ø
öççè
æÑ--
-÷÷ø
öççè
æÑ---
÷÷ø
öççè
æÑ---+-=
0
22
222
22
66
66
102)(
vvv
vvv
vvvv
gv
according to Maxey & Riley (Phys. Fluids, 26:883-889, 1983)Note: modified from the older BBO-equation
Forces acting on the particle: gravity + added mass +viscous drag + Basset history effects
VTT TECHNICAL RESEARCH CENTRE OF FINLAND11
Validity of Maxey & Riley equationt For laminar flow (and mean field of turbulent flow)
L is characteristic length scale, W characteristic relative velocity, U/Lthe scale of fluid velocity gradient for undisturbed flow.
t For turbulent flow
1,1,12
<<<<<<L
UrWrLr
f
p
f
pp
nn
1,1,12
<<<<<<kf
kp
f
p
k
p vrWrrhnnh
kh and kv are the Kolmogorow microscales of length and velocity:
4/1
4/13
)( ene
nh fkk vf =
÷÷ø
ö
ççè
æ=
e is the dissipation rate of turbulent kinetic energy
VTT TECHNICAL RESEARCH CENTRE OF FINLAND12
Equation of motion in Stokes regime (Rep<1)
( )fpfpfpp
p rmmdt
dm vvg
v---= mp6)(
Usually simplified:
For rapidly accelerating or oscillating flows this approximationcan be poor.
Problems can also be expected when density differencebetween the phases is small.
VTT TECHNICAL RESEARCH CENTRE OF FINLAND13
Drag models
t a large number of empirical correlations exist
t models that account particle shape and deformation
t for dilute fluid particle flows, the drag is given by
)(43
fpfpp
fD d
CF vvvv --=r
Schiller & Naumann correlation
[ ] 1000ReRe15.01Re24 687.0 £+= pp
pDC
1000Re44.0 >= pDC
• for dense fluid particle flows, the drag force dependsalso on volume fraction of dispersed phase
VTT TECHNICAL RESEARCH CENTRE OF FINLAND14
Dispersion of particles
t Eddy interaction model/discrete random walk model: a particle is assumed tointeract with an eddy over an interaction time
t The interaction time is the smaller of the eddy life time te and the eddycrossing time tc
t The velocity fluctuation during an eddy is obtained by sampling a normallydistributed random function:
úúû
ù
êêë
é
--=»
pfp
epce uut
Lttkt ln6.0e
)3/2()(' 2' kfufu ==
VTT TECHNICAL RESEARCH CENTRE OF FINLAND15
Numerical methods
One-way coupling Two-way coupling
VTT TECHNICAL RESEARCH CENTRE OF FINLAND16
Lagrangian approach, dilute example:
Pulverized coal boiler:Particle temperature alongparticle tracks
VTT TECHNICAL RESEARCH CENTRE OF FINLAND17
Lagrangian methods for dense gas-solids flow
t Lagrangian particle tracking methods have been applied e.g. to fluidizedbeds
t Several approaches that differ– in the way a computed particle corresponds to real particles: one particle can
present either one or a group of particles– in the way particle-particle collisions are treated
– Discrete Particle Method (DPM):– no collisions (used for dilute suspensions)– soft-sphere approach (the discrete element method, DEM)– hard-sphere approach
– Dense DPM, CPFD (computation particle fluid dynamic):– statistical approach
VTT TECHNICAL RESEARCH CENTRE OF FINLAND18
Lagrangian methods for dense gas-solids flow:soft- and hard-sphere models
t Soft-sphere model of particle-particle collisions:– the instantaneous inter-particle contact forces, namely, the normal,
damping and sliding forces are computed using equivalent simplemechanical elements, such as springs, dashpots and sliders.
– a very small time step is required (< 1e-6 s)
t Hard-sphere model of particle-particle collisions:– momentum-conserving binary collision– interactions between particles pair-wise and instantaneous– faster than soft-sphere method due to longer time steps
VTT TECHNICAL RESEARCH CENTRE OF FINLAND19
Lagrangian methods for dense gas-solids flow:multiphase particle-in-cell method (MP-PIC)
t Eulerian gas phase
t Lagrangian tracking of parcels of particles– Gas-particle and particle-particle interaction forces calculated in the mesh of
the gas phase from closures written for Eulerian-Eulerian models for densegas-particle suspensions (kinetic theory of granular flow, see later in thepresentation)
t Also called / similar models:– Dense DPM (Dense Discrete Particle Model; Ansys Fluent)– CPFD (Computational Particle Fluid Dynamic; Barracuda)
t Description of particle size distribution easy
t Coarse gas phase mesh leads to inaccuracies
VTT TECHNICAL RESEARCH CENTRE OF FINLAND20
Lagrangian simulation, dense example:(LES for carrier phase, collision models for particle-particle interaction; Helland et al., Powder
Technology, 110, pp. 210-221 (2000))
VTT TECHNICAL RESEARCH CENTRE OF FINLAND21
Coupling with direct numerical simulation (DNS) forcontinuous phase
t Lagrangian simulation of particles with hard or soft sphere models canbe coupled with direct numerical simulation of the continuous phase
– Fine mesh with several mesh elements per particle diameter– Short time steps, time consuming!– Produces accurately fluid-particle interactions– Used for submodel development
t Body-fitted grid: mesh is regenerated at each time step to fitparticle surfaces
t Non-body fitted grid: e.g. immersed boundary method in whichthe fluid flow calculations are performed by assuming that the fluidoccupies the entire flow field and the effect of particles is expressed bya body force into the momentum equation of fluid to constrain the no slipboundary condition at the nodes inside the particles.
VTT TECHNICAL RESEARCH CENTRE OF FINLAND22
Lagrangian approach, summary
• momentum and continuity equation for the carrier phase
• Newton equation for dispersed phase
• good for dilute suspensions, usable also for dense suspensions if collisionsare accounted for
• description of dispersion of particles in turbulent flow requireslarge number of particles
• fast for laminar flow; describes well particle size andother distributions
VTT TECHNICAL RESEARCH CENTRE OF FINLAND23
Approach 3: Eulerian-Eulerian approach
t single phase equations (momentum, continuity) are averaged overtime and/or volume and/or a particle assembly (Newtons equation)to obtain multiphase flow equations
t distribution of phases given by volume fraction
t models for phase interaction required
t all phases are described as interpenetrating continua
VTT TECHNICAL RESEARCH CENTRE OF FINLAND24
Volume averaging
Averaging volumePhase interface
VTT TECHNICAL RESEARCH CENTRE OF FINLAND25
Starting point in volume averaging: Single phase equations + jumpconditions
( ) ( ) 0=×Ñ+¶¶
qqqtvrr
( ) ( ) gτvvv qqqqqqqq pt
rrr +×Ñ+-Ñ=×Ñ+¶¶
( ) ( ) 0=×-+×- gggrr nvvnvv AqAqq
( ) ( ) AA
qqAqqqAqAqqq pp R
Rσ
σnτInτInvvvnvvv)g
gggggggg rr2
)()( +Ñ-×+-+×+-=×-+×-
qn Unit outward normal vector of phase q
Av Velocity of the interfaceAR
)AA RR /=
AR interface curvature vectorgqσ interface surface tension
AD Ñ×-Ñ AR)
= = surface gradient operator I unit tensor
VTT TECHNICAL RESEARCH CENTRE OF FINLAND26
Averaging
Volume averaging:
dVV qV qq ò= yy 1
V qq
q dVV q
ya
yy11~ == ò
V q
V qq
q dV
dV
q
q
ra
yr
r
yry ~==
òò
Volume average
Intrinsic volume average
Mass-weighted (Favre) average
VVqq /=a Volume fraction, 1=åq
qa
VTT TECHNICAL RESEARCH CENTRE OF FINLAND27
Eulerian-Eulerian balance equations
( ) ( ) qqqqqqtG=×Ñ+
¶¶ vrara ~~
( ) ( )
qqqqqqq
qqqqqqq
pt
draa
rara
τMgτ
vvv
×Ñ+++×Ñ+-Ñ=
×Ñ+¶¶
~)~(
~~
Mq is the interaction force.
Continuity:
Momentum:
VTT TECHNICAL RESEARCH CENTRE OF FINLAND28
0=Gåq
q
å òå ÷÷ø
öççè
æ+Ñ--=
g
gg
g,
221
q AA
A
qqA
qq dA
Vq
RRσ
σM)
Turbulent stress term:
Conservation of mass:
Surface tension:
))((~qqqqqqq vvvvτ ---= rad
VTT TECHNICAL RESEARCH CENTRE OF FINLAND29
Alternative averaging approaches
t volume averaging
t time averaging
t ensemble averaging
t volume-time averaging
t two-step volume-volume or volume-time averaging (first over thelocal scale, then over a larger scale)
t three-step averaging
VTT TECHNICAL RESEARCH CENTRE OF FINLAND30
Closures for turbulent stress termst no generally accepted turbulence models exist; parameters in any
models are probably suspension dependent
t several alternatives suggested
t mixture turbulence model:uses mixture properties to calculate viscous stress; applicable forstratified flows and when densities of the phases are of same order
t dispersed turbulence model– appropriate for dilute suspensions– turbulence quantities of the dispersed phase are obtained
using the theory of dispersion of particles by homogeneousturbulence
t turbulence model for each phase
VTT TECHNICAL RESEARCH CENTRE OF FINLAND31
Closures for drag force
t large number of models exist
t drag force is generally written in the form:
t Note: in dense flows, CD depends on αf
t parameters in drag models should be functions of theaveraging scale in inhomogeneous flows!
)(43
fqfqq
ffqDq d
CM vvvv ---=raa
VTT TECHNICAL RESEARCH CENTRE OF FINLAND32
Description of particle size distribution in the Eulerian-Eulerianmethod
t Simplest way would be to treat each size class as a separatecomputational phase (method of classes)
– Continuity and momentum equation for each size class– Computationally slow– Numerical difficulties: at least earlier versions than Ansys Fluent 14
produced non-physical results
t Particle size distribution described by methods of moments– QMOM (Quadrature Method Of Moments)– DQMOM (Direct QMOM)
– In principle requires solution of momentum equations for velocities ofthe moments
– Momentum equations can in some cases be replaced by algebraicequations
VTT TECHNICAL RESEARCH CENTRE OF FINLAND33
Numerical methods for the Eulerian-Eulerian approach
t several algorithms; usually extensions of single phase ones
t e.g. IPSA (Inter Phase Slip Algorithm) performs the following steps inone iteration loop:
– solve velocities of all phases from momentum equations– solve pressure correction equation (which is based on the joint
continuity equation)– update pressure– correct velocities– calculate volume fractions from phasic continuity equations
VTT TECHNICAL RESEARCH CENTRE OF FINLAND34
Example: simulated solids volume fractions in acirculating fluidized bed
Two-phase flow
Largeparticles
Smallparticles
’Three’-phase flow
VTT TECHNICAL RESEARCH CENTRE OF FINLAND35
Eulerian-Eulerian models, summary
• momentum and continuity equations for each phase
• several averaging methods used to derive the equations
• best alternative for dense suspensions with large velocitydifferences/acceleration
• often very slow
• particle size and other distributions difficult to account for
VTT TECHNICAL RESEARCH CENTRE OF FINLAND36
Approach 4: Mixture model
t A local equilibrium between the continuous and dispersed phases, i.e.at every point, the particles move with the terminal slip velocity relative tothe continuous phase
Þ velocity components for dispersed phases can be calculated fromalgebraic formulas
Þ significantly less equations to be solved
t Also called / similar models:– Algebraic slip (mixture) model– Diffusion model– Suspension model– Local-equilibrium model– Drift-flux model
VTT TECHNICAL RESEARCH CENTRE OF FINLAND37
Motivation for simplified multiphase models
Full Eulerian models - problemst Numerical problems - long computing times - small time steps
t Difficulties in convergence
t Only a few secondary phases possible
t More difficulties, if mass transfer and chemical reactions areconsidered
t Simplified models are recommended, if applicable
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND38
Detailed derivation of the mixture model (1)
t Starting point: Eulerian model equationst Continuity equation for phase k:
t Momentum balance for phase k:
t Mixture equations are derived by taking a sum over all phases
( ) ( ) kkkkkktG=×Ñ+ urara
¶¶
( ) ( ) ( )[ ] kkkTkkkkkkkkkkkk pt
Mgττuuu +++×Ñ+Ñ-=×Ñ+ raaarara¶¶
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND39
Detailed derivation of the mixture model (2)
t Continuity
t Define mixture density and velocity as
t Resulting equation
( ) ( ) ååå===
G=×Ñ+n
kk
n
kkkk
n
kkkt 111
urara¶¶
r a rmk
n
k k==
å1
åå=
==n
kkk
kkkk
mm c
1
1 uuu rar
( ) 0=×Ñ+ mmm
tur
¶¶r åå =×Ѻ×Ѻ×Ñ
kmk
kkk 0jjua
If all phases are incompressible
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND40
Detailed derivation of the mixture model (3)
t Momentum balance
t Diffusion velocity (not slip velocity!)
t Definitions
( ) åååå
åå
====
==
+++×Ñ+Ñ-=
×Ñ+
n
kk
n
kkk
n
kTkkk
n
kkk
n
kkkkk
n
kkkk
p
t
1111
11
Mgττ
uuu
raaa
rara¶¶
u u uMk k m= -
t tmk
n
k k==
å1
a
tTm k Ik Fk Fkk
n
= -=
å a r u u1
t Dmk
n
k k Mk Mk= -=
å1
a r u u
Viscous stress
Turbulent stress
Diffusion stress
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND41
Detailed derivation of the mixture model (4)
t Assumption: all phases share the same pressure,t Using the definitions of mixture density and velocity and the
stresses above, we obtain the resulting momentum balance for themixture
( ) ( ) mmDmTmmmmmmm pt
Mgτττuuu ++×Ñ++×Ñ+Ñ-=×Ñ+ rrr¶¶
ppk =
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND42
Detailed derivation of the mixture model (5)
t Eliminate the phase velocity in the continuity equation for a phase
t Or if phase densities are constants and mass transfer is zero:
which is a balance equation for phase k volume fraction andincludes the effect of slip. Note that this the same as the intuitiveresult except for the diffusion velocity instead of slip velocity.
( ) ( ) ( )Mkkkkmkkkktuu rarara
¶¶
×Ñ-G=×Ñ+
( ) ( )Mkkmkktuu aaa
¶¶
×-Ñ=×Ñ+
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND43
Detailed derivation of the mixture model (6)
t Relation between the slip velocity and the diffusion velocity
t For only one dispersed phase
t Note: so far no additional assumptions or approximations has beenmade (except the shared pressure, commonly used also in fullEulerian models)
u u uCk k c= -
å-=-=l
CllCkmkMk c uuuuu
( )u uMd d Cdc= -1
slip velocity (c denotes the continuous phase)
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND44
Slip velocity (1)
t The momentum equation for the dispersed phase is
t Momentum equation for the mixture (with Mm=0)
t Eliminate pressure gradient and solve for Mp
( ) ( ) ( )[ ]
( )[ ] ( ) ( )gτττττ
uuuuuuM
mppDmTmmpTppp
mmmppppm
mpMp
ppp tt
rraaa
rra¶
¶rr¶
¶ra
--++×Ñ++×Ñ-
Ñ×-Ñ×+úû
ùêë
é-+=
( ) ( )[ ] pppTppppppppp
pp pt
Mguuu
+++×Ñ+Ñ-=Ñ×+ rattaara¶
¶ra
( ) ( ) gτττuuu mDmTmmmmmmm pt
rrr¶¶
+×Ñ++×Ñ+-Ñ=×Ñ+
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND45
Slip velocity (2)
t Now, the following approximations are made:– All turbulent terms are omitted for the moment– Viscous and diffusion stress terms are small and are neglected– The local equilibrium assumption implies
t The momentum transfer term is given by
mmpp uuuu )()( Ñ×»Ñ×0»¶
¶tMpu
CpCpcDp
ppp C
VA
M uura
21=
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND46
Slip velocity (3)
t Finally:
t For small particles (Stokes drag) in gravitational and centrifugalfield
t This is the slip velocity. The diffusion velocity used in the modelequations must be calculated as shown above.
( ) ( ) úû
ùêë
é-+Ñ×-= guuuuu
tVCA m
mmmppCpCpDpc ¶¶rrr
21
úû
ùêë
é-
-= geu r
m
m
mppCd r
ud 22
18)( j
mrr
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND47
Closures for viscous stress in mixture models
[ ]Tmmmm )( vv Ñ+Ñ= mt
*
÷÷ø
öççè
æ-=
ma
aa
mmmax,5.2
max,
1p
p
pfm
max,pa
Viscous stress:
1=*mfp
fp
mmmm
m+
+=* 4.0
Mixture viscosity (Ishii & Zuver, 1979):
where is the maximum packing and
for bubbles or drops
for solid particlesBy Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND48
Treatment of turbulence in mixture models
t in dilute suspensions, mixture turbulent stress is assumed sameas carrier phase turbulent stress
t at higher concentrations, modeling required
t turbulence enhances spreading of dispersed phases -> takeninto account in continuity equation:
( ) ( ) ( ) ( )mpppppmpmpppp Dt
vv raarara ~~~ ×Ñ-G+Ñ×Ñ=×Ñ+¶¶
÷÷ø
öççè
æ+=
3/285.01
2
kD cp
Tcmp
vn
where Dmq is a dispersion (diffusion) coefficient:
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND49
Turbulent dispersion
t Turbulence is coupled in the mixture model in a straightforwardway, because the treatment has a single phase nature. A standardmodel, e.g. k-e model, can be coupled to the mixture equations.
t The turbulent stress and the fluctuating part of Mq are modeledusing a turbulent dispersion (diffusion) coefficient DT
t A suitable model is needed for the dispersion coefficient, forexample (Picart et al., 1986)
Tmqmq
Da
aÑ+®uu
DukT TCd= +
æ
èç
ö
ø÷
-
n 1 0 852 3
2 1 2
.
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND50
Mixture model - validity
t The essential approximation is the local equilibrium assumption:the particles are accelerated instantaneously to the terminalvelocity
t Consider again the simplified one-particle equation
t Solution of particle velocity as afunction of distance (Stokes regime)
t The characteristic length ofacceleration iswith
2pDpfp
pp uCAgV
dtdu
m rr21
-D=
0
0.2
0.4
0.6
0.8
1
0 0.5 1 1.5 2 2.5 3
px l
tpp ut=l
f
ppp
dt
mr
18
2
= pp
t gtur
rD=
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND51
Mixture model - validity
0.0001
0.001
0.01
0.1
1
1.0 2.0 3.0 4.0 5.0 6.0rp/rc
l p(c
m)
0.1 mm
0.2 mm
Virtual mass and history effects
Stokes drag only
u If the density ratio is small, the virtual mass and history terms cannot be ignored (botheffects increase the response time)
u A suitable requirement for the local equilibrium is lp<< typical system dimension
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND52
Validity range of mixture models, cont.
- Mixture model is usually not suitable for gas-particle flows,because the characteristic length can be large
- Not suitable for clustering flow, where average slip velocityis greater than single particle slip
- Typically used for liquid-solid flows and bubbly liquid-gasflows (not for very big bubbles)
VTT TECHNICAL RESEARCH CENTRE OF FINLAND53
Validating simulationsSolid particles in an agitated vessel
Glass beads of three different sizes in water
Experiments: Barresi and Baldi (1987)
D = 0.390 m, H = 0.464 m4 baffles (0.039 m)45° pitched 4-blade turbine(Dt = 0.130 m, Htb = 0.046 m).Impeller power number: 1.2
Average mass fraction: 0.5 %
dp,exp.(mm) 100-177 208-250 417-500dp,sim.(mm) 140 230 460r (kg/m3) 2670 2600 2600N (rev/s) 4.83 5.67 6.50
d = 230 mm
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND54
Validating simulations
Solid particles in an agitated vessel
0.0
0.5
1.0
0.0 0.5 1.0 1.5 2.0 2.5
c/cav e
z/z m
ax
ExperimentalMixture modelMultiphase model
dp = 140 mm
0.0
0.5
1.0
0.0 0.5 1.0 1.5 2.0 2.5
c/cav e
z/z m
ax
ExperimentalMixture modelMultiphase model
dp = 230 mm
0.0
0.5
1.0
0.0 0.5 1.0 1.5 2.0 2.5
c/cav e
z/z m
ax
ExperimentalMixture modelMultiphase model
dp = 460 mm
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND55
Bubble drag correlations
Tomiyama 1
Tomiyama 2
Ishii-Zuber
( )s
rr 2pqp d
Eo-
=a
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND56
Bubble terminal velocity
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0 0.002 0.004 0.006 0.008 0.01d (m)
Usl
ip
I-Z, gI-Z, 10gI-Z, 20gTomiyama 2, 20gSl
ipve
loci
tyd (m)
In gravitational field Effect of acceleration
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND57
Mixture model example
VTT TECHNICAL RESEARCH CENTRE OF FINLAND58
Mixture models (FLUENT): (N+3)t Continuity equation for the mixture and
all bubble phases (N)t Momentum equation for the mixture (3)t Bubbles: slip velocities from algebraic
balance equationst Numerically well behaved, a large
number of bubble classes possible
General Eulerian models(4xN balance equations)t Continuity equation for all phases (N)t Momentum equation for all phases (3N)t Numerically difficult and slow
Comparison of models for bubbly flows
By Mikko Manninen, VTT
VTT TECHNICAL RESEARCH CENTRE OF FINLAND59
Mixture model summary
- Length scale of particle relaxation must be small comparedto the mesh spacing
- Local equilibrium assumption yields the slip velocity
- Transport of the dispersed phase calculated from thephase continuity equation
- Fast; describes well particle size and other distributions
- Describes well effects of turbulence
VTT TECHNICAL RESEARCH CENTRE OF FINLAND60
Other approaches
t VOF (volume of fluid) –method for tracking phase interfaces inseparated flows or e.g. droplet formation in small scale
t Eulerian-Eulerian + Lagrangian
t Eulerian-Eulerian + mixture model
t Mesoscopic simulation methods (e.g. Lattice-Bolzmann method)
t MP-PIC (multi-phase particle in cell; Lagrangian transport ofparcels of particles + Eulerian kinetic theory description for solids)
VTT TECHNICAL RESEARCH CENTRE OF FINLAND61
Summary of multiphase CFD:
• Several ways to derive equations
• Simplifications in all models
• Jump conditions at phase interfaces complicated (drag force )
• Multitude of time and length scales -> difficult to choosesuitable averaging and simulation scales
• Flows often time-dependent also in macroscopic length scales(e.g. bubbling fluidized beds)
VTT TECHNICAL RESEARCH CENTRE OF FINLAND62
Summary of multiphase CFD, cont.:
• Never fully ‘laminar’
• Turbulence modelling difficult; no general models exist;unisotropy causes problems
• Dense suspensions dominated by particle-particle collisions
• Dilute suspensions dominated by particle-turbulence interaction
• Measurements still necessary for derivation of equation closuresand for validation; often not available
VTT TECHNICAL RESEARCH CENTRE OF FINLAND63
Summary of multiphase CFD, cont. :
• Choose the simplest possible model
• Follow the literature
• Compare with measurement data if available
• Results for dilute suspensions already good
• Even dense suspensions can be simulated
• Increasing computer speed will facilitate simulationof more demanding cases in the future
VTT TECHNICAL RESEARCH CENTRE OF FINLAND64
CFB (circulating) TFB (turbulent) BFB (bubbling)
- inhomogeneous flow pattern- flow characteristics change as a function of fluidization velocity
Ref.: U. Ojaniemi, S. Kallio, A. Hermanson, M. Manninen, M. Seppälä, V. Taivassalo, Comparison of simulated and measuredflow patterns: solid and gas mixing in a 2D turbulent fluidized bed, Fluidization XII, Harrison Hot Springs, BC, Canada, 2007
Example of Eulerian-Eulerian modeling: Fluidized beds
VTT TECHNICAL RESEARCH CENTRE OF FINLAND65
Eulerian-Eulerian approach based on KTGF– background
t Kinetic theory of granular flow (KTGF)
t A special model developed for dense gas-solid suspensions, seee.g. (Gidaspow D., 1994, Multiphase flow and fluidization:Continuum and kinetic theory descriptions, Academic press,Boston)
t Theory similar to gas kinetic theoryt Solid phase equations are derived by ensemble averaging over the
single particle velocity distribution functiont This averaging is done in a local scale assuming locally random
distribution of particles
VTT TECHNICAL RESEARCH CENTRE OF FINLAND66
Eulerian-Eulerian approach based on KTGF– limitations and simplifications
t Because of the randomness assumption, the theory is basicallyonly suitable for transient simulations in fairly fine meshes
t In derivation of the models, only binary collisions between particlesare accounted for, and thus KTGF is not valid in densesuspensions, like in bubbling beds
t In practise it is used from packing limit to very dilute conditions
t The theory was originally written for a single particle sizet Later rigorous extension of the theory to mixtures has been
developed
t Here we focus on models for a single particle size
VTT TECHNICAL RESEARCH CENTRE OF FINLAND67
Eulerian-Eulerian approach based on KTGF– software inplementations
t Kinetic theory of granular flow is today found in– commercial codes– open source codes– researchers’ and universities own codes
t Alternative closures based on different assumptions have beensuggested in the literature
t Here we mainly present models available in the commercial Fluentsoftware
VTT TECHNICAL RESEARCH CENTRE OF FINLAND68
Eulerian-Eulerian approach based on KTGF– balance equations
0)()( =×Ñ+¶¶
qqqqqtvrara
)()()()( gssggggggggggggg Kpt
vvgvvv -++×Ñ+Ñ-=×Ñ+¶¶ rataarara
)()()()( sgsgsssssssssssss Kppt
vvgvvv -++×Ñ+Ñ-Ñ-=×Ñ+¶¶ rataarara
Continuity equation for phase q:
Momentum equation for gas phase g:
Momentum equation for solid phase s:
VTT TECHNICAL RESEARCH CENTRE OF FINLAND69
Eulerian-Eulerian approach based on KTGF– balance equation for granular temperature
2'31
ss v=Q
Granular temperature is defined as an ensemble average
where vs´ is the fluctuating solids velocity.
First term on the right is the generation term, the second term is the diffusion term, the thirdthe dissipation term and the fourth expresses the energy exchange between gas and solids.
Alternatively, the granular temperature is computed from an algebraic equation obtained fromthe original transport equation by neglecting convection and diffusion terms.
Balance equation for Granular temperature:
( ) ( ) ( ) gssssssssssss sskp
tfgrara +-QÑ×Ñ+Ñ+-=úû
ùêëé Q×Ñ+Q¶¶
QQvτIv :)(23
VTT TECHNICAL RESEARCH CENTRE OF FINLAND70
Eulerian-Eulerian approach based on KTGF– closures for granular temperature
1
31
max,,0 1
-
úúú
û
ù
êêê
ë
é
÷÷ø
öççè
æ-=
s
sssg
aa
Radial distribution function related to the probability of collisions (Lun et al):
where αs,max expresses the solids volume fraction of a packed bed.
2/32,02 )1(12
ssss
ssss
dge
sQ
+=Q ra
pg
The collisional dissipation of energy can be computed from (Lun et al.)
and the energy exchange between gas and solids from (Gidaspow et al.)
sgsgs K Q-= 3f
VTT TECHNICAL RESEARCH CENTRE OF FINLAND71
Eulerian-Eulerian approach based on KTGF– closures for granular temperature, continued
The diffusion coefficient is computed, according to Syamlal et al., from
úûù
êëé -+-+
-Q
=Q ssssssssss gg
dk
s ,0,02 )3341(
1516)34(
5121
)3341(415
hahp
ahhh
par
where
)1(21
sse+=h
Gidaspow suggests instead:
praapr
/)1(2)1(561
)1(384150
,02
2
,0,0
sssssssssssssssss
sss gedegge
dk
sQ++úû
ùêëé ++
+Q
=Q
VTT TECHNICAL RESEARCH CENTRE OF FINLAND72
Eulerian-Eulerian approach based on KTGF– granular pressure
Lun et al.:
Syamlal & O’Brien:
Ahmadi & Ma:
sssssssssss gep Q++Q= ,02)1(2 arra
ssssssss gep Q+= ,02)1(2 ar
( ) úûù
êëé +-+++Q= )21)(1(
2141 ,0 fricsssssssssss eegp mara
VTT TECHNICAL RESEARCH CENTRE OF FINLAND73
Eulerian-Eulerian approach based on KTGF– solids shear stress based on KTGF
Ivvv qqqqTqqqqq ×Ñ-+Ñ+Ñ= )
32()( mlamatPhase q stress-strain tensor
Solids shear viscosity is given as a sum of the one predicted by the kinetic theoryand a frictional shear viscosity
frskinscolsfrsktgfss ,,,,, mmmmmm ++=+=
The collisional viscosity (Gidaspow et al) :
21
,0, )1(54
÷ø
öçè
æ Q+=
pram s
ssssssscols egd
Kinetic viscosity (Syamlal) (Gidaspow)
úûù
êëé -++
-Q
= sssssssss
sssskins gee
ed
apra
m ,0, )13)(1(521
)3(6
The granular bulk viscosity (Lun et al)
21
,0 )1(34
÷øö
çèæ Q
+=p
ral sssssssss egd
2
,0,0
, )1(541
)3(9610
úûù
êëé ++
-Q
= ssssssssss
ssskins ge
ged
aa
prm
VTT TECHNICAL RESEARCH CENTRE OF FINLAND74
Eulerian-Eulerian approach based on KTGF– frictional stress
( )Tssfrsfrfr p )(, vvI Ñ+Ñ+-= mt
frktgfs ppp +=
Johnson and Jackson (1987)
Shear stress:
Solids pressure:
Frictional pressure: , Fr=0.05, n=2, p=5.In FLUENT Fr=0.1αs
Frictional viscosity:
Schaeffer (1987)
Frictional viscosity:
pss
nss
fr Frp)()(
max,
min,
aaaa-
-=
fm sin, frfrs p=
D
frfrs I
p
2, 2
sinfm =
VTT TECHNICAL RESEARCH CENTRE OF FINLAND75
Eulerian-Eulerian approach based on KTGF– frictional stress
Syamlal et al. (1993):
Shear stress:
Solids pressure:
Frictional pressure:
VTT TECHNICAL RESEARCH CENTRE OF FINLAND76
Eulerian-Eulerian approach based on KTGF– drag force in dense gas-particle suspension
VTT TECHNICAL RESEARCH CENTRE OF FINLAND77
Eulerian-Eulerian approach based on KTGF– implentation in the software
t The models are often not implemented as such in numerical solvers.t Simplifications are introduced in the implementations to improve the
numerical stability.
t For example, the radial basis function needs to be restricted close tothe packing limit to avoid numerical problems
t The manuals of commercial codes don’t reveal details of thesemodifications.
t Sometimes it is possible to track down the actual equations used by atedious comparison of the values given by the commercial code anddifferent versions of the equations.
VTT TECHNICAL RESEARCH CENTRE OF FINLAND78
Validation of the KTGF model in a 2D CFBSimulation in a fine mesh (0.625 mm spacing)
Comparison of measured solids volume fraction and velocities with thesimulation results at 0.8 m and 1.2 m heights in the 2D CFB at ÅboAkademi. Particle size 0.44 mm and fluidization velocity 3.75 m/s.
VTT TECHNICAL RESEARCH CENTRE OF FINLAND79
Results without drag correction in a coarse mesh
1 s 2 s 3 s 4 s 5 s 6 s 7 s 8 s 9 s 10 s 11 s 12 s
-> the assumption of local homogeneity in the closure equations leads to problems-> subgrid scale closures required
VTT TECHNICAL RESEARCH CENTRE OF FINLAND80
Modified drag force model used in CFB and turbulentbed simulations
- larger particle size → smaller cluster correction
- finer mesh → smaller correction necessary
Voidage function used at slip velocity 1 m/s.The drag model by Gidaspow et al., recommendedin Fluent manuals, is shown for comparison.
Ref.: U. Ojaniemi, S. Kallio, A. Hermanson, M. Manninen, M. Seppälä, V. Taivassalo, Comparison of simulated and measuredflow patterns: solid and gas mixing in a 2D turbulent fluidized bed, Fluidization XII, Harrison Hot Springs, BC, Canada, 2007
VTT TECHNICAL RESEARCH CENTRE OF FINLAND81
Simulations of a 1 m X 7.3 m riser in 2D with macroscopic drag model
8000 cells sized 2.5 cm X 3.65 cm 4400, 17600 and70400 elements
Effect of the mesh: the finer the mesh, the finer structures are produced. Average voidage patterns are unaffected.
S. Kallio, The role of the gas-solid drag force in CFD modeling of fluidization,Åbo Akademi, Heat Engineering Laboratory, Report 2005-3, 2005
Solidsvolumefraction
VTT TECHNICAL RESEARCH CENTRE OF FINLAND82
Relationship between the number of elements and element size:
3D simulation, furnace volume 10 x 30 x 40 m3
Elements Element size20 000 0.853 m3
100 000 0.53 m3
1 000 000 0.233 m3
10 000 000 0.113 m3
100 000 000 0.053 m3
2D simulation, furnace ‘area’ 30 x 40 m2
Elements Element size20 000 0.252 m2
100 000 0.112 m2
500 000 0.052 m2
5 000 000 0.0152 m2
At least when mesh spacing > 0.5 cm, drag laws need to be adjusted.When mesh spacing > 10 cm, drag laws need further modifications andprobably also stress terms need to be modified ?!?Macroscopic/steady state/coarse mesh models should be developed!
3D simulation of large BFB’s and CFB’s
VTT TECHNICAL RESEARCH CENTRE OF FINLAND83
Macroscopic CFD models
t take into account large control volume sizet are written for steady statet large fluctuation terms in the equations → equation closure highly demandingt equation closure requires measurementst produces reasonable solids distributionst see e.g.
– S. Kallio, V. Poikolainen, T. Hyppänen, Report 96-4, Åbo Akademi/Värmeteknik– Taivassalo, V., Kallio, S., Peltola, J.: On time-averaged modeling of circulating
fluidized beds. 12th Int. conf. on multiphase flow in industrial plants, Ichia, Italy, 21.-23.9.2011, paper 69.
t coarse mesh models (filtered models) similar; meant for transient simulations oflarge processes
t no generally accepted macroscopic models available today!
VTT TECHNICAL RESEARCH CENTRE OF FINLAND84
Simulation of real fluidization processes- special features
t Complicated by a multitutude of particle sizes/types
t Complicated by differences in temperature → energy balance
t Complicated by homogeneous and heterogeneous reactions →species balances necessary
t Laboratory scale processes can be modeled in more details in thetransient mode
VTT TECHNICAL RESEARCH CENTRE OF FINLAND85
Additional equations/models required (examples)
t Energy balance equations for each phase– Heat capacity for each phase– Inter-phase heat transfer models, depend on particle size,
temperature, suspension density, velocities etc– Reaction enthalpies– Radiative heat transfer– Conductivity of each phase
t Species balance equations– Models for homogeneous and heterogeneous reactions– Inter-phase transfer rates– Diffusion/dispersion
t Comminution (attrition, fragmentation)t Solid-solid momentum transfer
VTT TECHNICAL RESEARCH CENTRE OF FINLAND86
Simulation of large industrial processes
t Complicated by a multitutude of particle sizes/types
t Complicated by differences in temperature → energy balance
t Complicated by homogeneous and heterogeneous reactions →species balances necessary
t Complicated by a large size → coarse meshes, long computations
t Simplifications/special approaches necessary
VTT TECHNICAL RESEARCH CENTRE OF FINLAND87
Examples of approaches for large industrial furnaces
t BFB modeling– Simplified models for the bed and 1-phase + Lagrangian particles in
the freeboard– Coarse mesh models
t CFB modeling– Empirical bulk solids distribution + balance equations for gas phase,
energy, species; fuel mixed using Lagrangian models or dispersionmodels
– Full two-phase model, transient (slow!!!) or steady-state (closuresbeing developed and validated e.g. at VTT)
– Full multi-phase model (slow!!!)– Particle-in-cell method (problems with coarse mesh)
VTT TECHNICAL RESEARCH CENTRE OF FINLAND88
Example: steady-state simulation of Chalmers boiler with a time-averaged model
t Gas phase: 5 componentst Solid phase: ash and sand as one Eulerian solid phaset Coal: particle tracking; drying, devolatilzation, combustion; two-way
coupled with gas phaset Solved using rigorous time-averaged equations for steady-state
hydrodynamics, with balance equs for Reynolds stressest Species balance equations solved for all the componentst Source terms calculated from homogeneous and heterogeneous
reactionst Mixing of enthalpy and chemical components described by
dispersion coefficients in energy and species balance equations;dispersion calculated from Reynolds stresses and time scales
Ref.: Taivassalo et al., FBC21, Naples 2012
VTT TECHNICAL RESEARCH CENTRE OF FINLAND89
Example2: steady-state simulation of Chalmers boiler with thetime-averaged model (Taivassalo et al., FBC21, Naples 2012)
a) Time-averaged solid volume fraction,. b) time-averaged mass-weightedvertical solid velocity, c) typical pathlines for a 1 mm fuel particle coloured bythe residence time , d) mole fraction of O2 and e) the time-averaged gastemperature (oC) .
a) b) c) d) e)
VTT TECHNICAL RESEARCH CENTRE OF FINLAND90
Conclusions on simulation of dense gas-solids flow
t Eulerian-Eulerian is a good approach; MP-PIC is also usable and good fordescription of particle size distributions
t Cluster formation is characteristic of dense gas-solid flows
t Transient simulations produce correct fluctuation patterns and averagedistributions, but as such usable only for small geometries
t Industrial processes large, large processes require coarse meshes
t Remember: closure model parameters for transient equations coarse-meshsimulations should be mesh-dependent! No such models available!
t Steady-state models now available at VTT and used for simulation of CFBboiler furnaces