Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Refined turbulence modelling for reactor thermal-hydraulicsPart of UK’s “Keeping Nuclear Options Open” project
School of Mechanical, Aerospace & Civil Engineering (MACE)The University of Manchester
Dr. Marc Cotton, Dr. Yacine Addad, Amir Keshmiri, Stefano Rolfo,
+ F. Billard, D. Laurence
Code Saturne
User’s Annual Meeting, EDF R&D Chatou
FR, Nov. 2007
http://www.knoo.org/
- to maintain and develop skills relevant to nuclear power generation.- 4 years, 6M £, 50 PhDs and Post-docs, - largest commitment to fission reactor research in UK for over 30 years, - collaboration with industrial and governmental stakeholders and international partners.
KNOO WORKPACKAGES
Ex: Molecular dynamics of radiation enhanced helium re-solution
Droplets on hot solid surfaces: Modelling and Experimentation
Ballooning fuel pins
Heat transfer through crud
LES with adaptive FE meshing
CFD @ Imperial College
CFX
STAR
FE (Fetch
Test Case 1 :Test Case 1 : Mixed convection in vertically flowing heated pipeMixed convection in vertically flowing heated pipe
(buoyancy aiding or opposing)(buoyancy aiding or opposing)
Problem specifications:Re=2650
Pr=0.71
Wall constant heat flux
Boussinesq approximation
Heat transfer Regimes:Gr/Re2=0.000 Forced Convection
Gr/Re2=0.063 Forced/Mixed Convection
Gr/Re2=0.087 Re-Laminarization
Gr/Re2=0.241 Recovery
KNOO CFD @ Manchester: Heat transfer test cases
AGR working schemeAGR working schemeRelevance to AGR and VHTR
DNS Relaminarization
point, Impairment of heat transfer
Buoyancy aided heated pipe flow
(Buoyancy number)
K-omega
or SST worst !
Range of RANS models
Buoyancy aided heated pipe flow
LES Addad Manch.
best models: Launder Sharma, V2F
GrGr/Re**2 = 0.000/Re**2 = 0.000
Buoyancy aided heated pipe flow
GrGr/Re**2 = 0.063/Re**2 = 0.063Buoyancy aided heated pipe flow
GrGr/Re**2 = 0.087 (/Re**2 = 0.087 (relaminarizationrelaminarization))
Buoyancy aided heated pipe flow
GrGr/Re**2 = 0.241 (recovery)/Re**2 = 0.241 (recovery)Buoyancy aided heated pipe flow
The model and Code_Saturne• A low-Reynolds (near-wall integration) eddy viscosity
model derived from second moment closure models• No damping functions, no wall functions, less empirical
assumptions• Best results on range of test cases, heat transfer and
natural convection in particular. • The original model is stiff (requires coupled solver or very
small time-step)• Degraded version available in StarCD, Fluent, NUMECA..• Long collaboration Stanford, Delft, Chatou, Manchester
(Durbin, Parneix, Hanjalic, Manceau, Uribe) => “several code friendly” versions since 1995.
• Present: Reconsider all historical choices with numerical stability and known asymptotic states as principal objectives
v2− fAccuracy
Robustness
Stanford 1991
TU-Delft 2004
Manchester 2004
Stanford 1996(Fluent, STAR-CD)
Revisiting the V2F model
Durbin's original model
the walls affect the whole domain through an elliptic operator : near walleffects (mainly wall echo and wallblocking effect)
Classic near wall modelν t = fμ Cμk T ; T = k /ε
fμ =ν t you wantν t you have
fμ = f (y +,ν t /ν )
Elliptic Relaxation
ν t = Cμ v 2 T
Dv 2
Dt= φ22 − ε22 +∇((ν +ν t )∇v 2 )
φ22 = −2ρ
v ∂p' /∂y
Revisiting the V2F model
The usual second moment closure• Usual closure for the source term of : v2
Dv 2
Dt= φ22 − ε22 −
23
ε⎛ ⎝ ⎜
⎞ ⎠ ⎟ −
23
ε +Diffν +ν tv 2
Term to be modelled
ϕ 22
Launder Reece Rodi (LRR) :
ϕ22h = −
1T
C1v2
k−
23
⎛
⎝ ⎜ ⎜
⎞
⎠ ⎟ ⎟ +C2
Pk
Speziale, Sarkar, Gatski
(SSG) :
ϕ22h = −
1T
C1 +C'2Pε
⎛ ⎝ ⎜
⎞ ⎠ ⎟
v2
k−
23
⎛
⎝ ⎜ ⎜
⎞
⎠ ⎟ ⎟ −
C4
3− C5
⎛ ⎝ ⎜
⎞ ⎠ ⎟
Pk
εh = 23
εModel for the dissipation :
wall
2u
2u
2v
2w
2w
2v
2u
2v2w
2w
2v
Wall blocking effect:
2u
The redistribution is anisotropic near the wall
Revisiting the V2F model : original
The asymptotic behaviour
• Near wall limit of the k
–eps model
DkDt
= (ν + ν t ) Δk + P − ε ≈ νΔk − ε ≈ νΔk n +1 −εk
⎛ ⎝ ⎜
⎞ ⎠ ⎟
n
k n +1
( O 1( )+O y 3( ) ).ΔO y 2( )
( )3yO
)1(O
( )2yO=k ( )1O=ε
No stability problem
Negative but implicit
source
Numerical stability
The asymptotic behaviour
• Taylor series expansion : ( )42 yO=v
Dv 2
Dt= ν ∂2 v 2
∂xk2 −
∂v 2ui
∂x i
+ φ22 + D22p − ε22 −
εk
v 2⎛ ⎝ ⎜
⎞ ⎠ ⎟ −
εk
v 2
( )2yO ( )5yO ( )2yO ( )2yO( )2yO
( )22 yO=u ( )22 yO=w
two-component limit of the turbulence
term to be modelled
The balancebetween
quadratic terms must be ensured
( )yO ( )yO
2Ay fkBy =22Cy
0)1()5()6(0
=−+−++=++ wallthenearCBA
2 42 2 2
hom 4 40 0
( )20lim lim( )
!y y
v O yL f f = f with f = =y O y
νε→ →
⎛ ⎞∇ − − => −⎜ ⎟⎜ ⎟
⎝ ⎠
Revisiting the V2F model
Stanford “code friendly” model (1)
It involves a change of variable so that gf=f + 0lim0
=fy→
The new equation for f reads :
L2∇2 f − f = C1 −1T
v 2
k−
23
⎛
⎝ ⎜ ⎜
⎞
⎠ ⎟ ⎟ − C2
Pε
+g − L2∇2gneglected term
Original Durbin : 2220 CyByAy ++=
fk
Stanford
: 222220 DyCyDyByAy ++−+=gkfk − gk
0=− DB
Lien and Durbin (1996)
Numerical stability vs. accuracy
Stanford “code friendly” model (2)Numerical stability vs. accuracy
Delft / UMIST approach
• A new change of variable to reduce the stiffness of the B.C. : kv=
2
ϕ
UMIST (Laurence et al. (2004)) and TU-Delft (Hanjalic et al. (2004))
DϕDt
= f − P ϕk
+2k
ν + ν t
σ k
⎛
⎝ ⎜
⎞
⎠ ⎟ ∇ϕ∇k+∇((ν + ν t
σϕ
)∇ϕ)
A B C
wallthenearCBA )( +−=
fw = −2ε ϕy 2
UMIST
modelϕϕνϕν 22
∇+∇∇+= kk
ff
0=fw
Delft
modelς
fw = −2ε ςy 2
Numerical stability vs. accuracy
Numerical instabilities of the model
2 options :•
f is forced to be equal to 0 at the wall : excellent robustness,
similar to the one
of Stanford's model, but bad prediction of the turbulent viscosity near the wall.• the turbulent viscosity is removed : numerical instabilities
kv=
2
ϕ
Numerical stability vs. accuracy
Natural convection in a heated cavity (non convergence in C_S V 1.2 !)
HOTCOLD
Very low value of k
Very low value of k
ϕνϕν 22∇+∇∇+= k
kff
ϕ model
Code_Saturne
Numerical stability vs. accuracy
The model• Using the elliptic blending of Manceau and Hanjalic 2002 (with Re stress model)
ϕ −α
122 −−∇ =L ααDϕDt
= α 3 fhom + 1−α 3( )fw − P ϕk
+ 2ν t
k∇ϕ∇k +∇((ν + ν t
σϕ
)∇ϕ)
Successfully tested on channel flows for many Re numbers, flow around airfoil trailing edge, heated pipe, heated channel flow, heated cavityNormal time-step (external flow CFL values as for k-omega)Unlike, code friendly Stanford model, no term has been neglected hereUnlike UMIST and Delft model, the correct asymptotic behaviour of and
is accurately predicted without impairing the numerical robustnessv2
tν
0=wα
2yfw
ϕε−= )1(2 ofw =∇− ϕν
Final model
Results (Channel flow, Re*=395
tν
v2
Results (Channel flow, Re*=395) (2)
V2v2
tν
y+
Conclusions and further work
• 4 different versions of V2F revisited => numerical stability improved while respecting known asymptotic states,
• Relaminarisation OK, next: prediction of laminar-turbulent transition,
input some ingredients of Launder and Sharma model (extra viscous source terms in epsilon equation)
• Recalibration of source terms in the ε
equation (all literature focuses only on near wall layer but prediction of the core region can be improved)
• An accurate and robust near-wall low-Reynolds RANS model also suitable for RANS/LES coupling
[1] Launder, B.E. and Sharma, B.I., 1974, “Application of the energy dissipation model of turbulence to the calculation of flow near a spinning disc”, Lett. Heat Mass Transfer, 1, pp. 131-138.[2] Cotton, M.A., Ismael, J.O., 1998, “A strain parameter turbulence model and its application to homogeneous and thin shear flows”,
Int. J. Heat Fluid Flow 19, pp.
326–337.[3] Chen, Y.S., and Kim, S.W. 1987, “Computation of turbulent flows using an extended k-ε
turbulence closure model”,
NASA CR-179204.
[4] Wilcox, D.C. 1998, “Turbulence Modelling for CFD”,
2nd edition, DCW Industries, Inc.[5] Durbin, P.A. 1995, “Separated flow computations with the k-ε−v2 model”
, AIAA
Journal, 33(4), pp. 659–664.[6] Laurence, D., Uribe, J.C. and Utyuzhnikov, S., 2004, “
A robust formulation of the
v2-f model”, Flow, Turbulence and Combustion, 73, pp. 169-185[7] You, J., Yoo, J.Y. and Choi. H., 2003, “Direct Numerical Simulation of Heated Vertical Air Flows in Fully Developed Turbulent Mixed Convection”, Int. J. Heat Mass Transfer, 46, pp.1613-1627