Академический Документы
Профессиональный Документы
Культура Документы
Corresponding author.
E-mail address: roelofs@nrg.eu (F. Roelofs).
and transported to a coolant. Almost all reactor cores consist of
many fuel assemblies which in turn consist of a large number of
fuel rods. For the design and safety analyses of such reactors, sim-
ulations of the heat transport within the core are essential. During
the sixties, seventies, and eighties of the twentieth century, when
the current light water reactor technology was developed, the only
possibility to compute the ow inside a fuel bundle was to use
one-dimensional sub channel analysis. The sub channel scale could
not be resolved. For simulations of the whole reactor core either
system codes or homogenization was employed. In system codes,
resolutionof individual assemblies was the state of the art. Homog-
enization used porous media equations and averaged the thermo
hydraulics on reactor core scale. Current potent computing power
allows using Computational Fluid Dynamics (CFD) to simulate indi-
vidual fuel assemblies. Detailed CFD analyses ranging from single
0029-5493/$ see front matter 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.nucengdes.2012.05.029
F. Roelofs et al. / Nuclear Engineering and Design 250 (2012) 548559 549
subchannels as described by Chandra et al. (2009), Baglietto et al.
(2006) and Chang and Tavoularis (2008) to multiple subchannels
as reported by Benhamadouche and Le-Maitre (2009), Ikeno and
Kajishima (2010), Pointer et al. (2008) and Fischer et al. (2007)
and even complete fuel assemblies as described by Gajapathy et al.
(2009) and Pniguel et al. (2010) using different CFD turbulence
modeling techniques, like Reynolds AveragedNavier Stokes (RANS)
(Chandra et al., 2009; Baglietto et al., 2006; Chang and Tavoularis,
2008; Pointer et al., 2008; Gajapathy et al., 2009; Pniguel et al.,
2010), Large Eddy Simulation (LES) (Benhamadouche and Le-
Maitre, 2009; Ikeno and Kajishima, 2010; Fischer et al., 2007) and
Direct Numerical Simulation (DNS) (Baglietto et al., 2006) have
been reported. Yet the large number of fuel assemblies within the
core limits exploiting such detailed CFD simulations for core wide
or even in most cases fuel assembly wide simulation.
In order to prot from the fast development of commercial
CFD software, newapproaches of applying CFD to fuel bundles are
assessed by different teams. This paper will describe the meth-
ods used by the Nuclear Research and consultancy Group (NRG)
in the Netherlands and the Karlsruhe Institute of Technology (KIT)
in Germany. Both approaches are based on resolving the geometry
of the fuel assembly to a certain degree. However, the resolution of
the employedmeshes is muchlower thaninthe traditional detailed
CFD analyses. Therefore, the methods are called Low Resolution
Geometry Resolving (LRGR) CFD and Coarse-Grid-CFD (CGCFD).
No new models have been added to existing CFD codes and the
approaches infact just usetheexistingpossibilities of theemployed
codes. Therefore, the described developments are referred to as
approaches, rather than new models. These methods are based on
low resolution meshes and can be applied with and without a sub
grid model (SGM) as will be explained below.
This paper will describe the newly developed LRGR and CGCFD
with and without SGM in Section 1. Section 2 explains how the
grid resolution of the low resolution meshes was selected. Some
applications of both the LRGR CFDand the CGCFDwith and without
SGM will be discussed in Section 3. Finally, Section 4 will present
the conclusions.
2. Approaches
As explained in the introduction, the approaches seek to prot
from the fast development of commercial CFD software and are
developed independently by NRG and KIT. The two methods are
similar in their use of lowresolution grids meanwhile maintaining
the important geometrical features. However, theydiffer inthe way
the owphysics are taken into account. Within NRG, the focus has
been directed towards LRGR CFD without any model for the non-
resolved sub grid physics, whereas at KIT, for the CGCFD sub grid
models are used to involve these particular owphysics.
A similar simulation method was recently reported by
researchers from Argonne National Lab in the USA (Hu and
Fanning, 2010). They try to ll the gap between the detailed
CFD simulations and the traditional subchannel codes by propos-
ing an intermediate-resolution modeling method. Their objective
is to develop scalable intermediate-resolution assembly model-
ing methods in which the anisotropic ow behavior is resolved,
to support full-core transient modeling coupled to existing or
emerging neutronics and fuel performance models. In fact, they
examined two ways to reduce the number of computational vol-
umes compared to their detailed simulations reported by Pointer
et al. (2008) and Fischer et al. (2007). The rst way is use a spi-
ral n to replace the wire wrap. The second way is to simulate a
bare rod bundle and introduce a momentumsource termthat spi-
rals around each rod. The latter method is elaborately described in
Hu and Fanning (2011). In fact, these two ways are similar to the
LRGR CFD and CGCFD approaches from the authors as described
below.
At the Commissariat lnergie atomique (CEA) inFrance, Bieder
et al. (2010) studied the way a wire can be modeled in a numerical
model. They distinguish between a real wire, a wire radially dis-
placedinthe rod, a wire radially displacedinthe rodusing blending
at the edges, and a square or nned wire (see Fig. 1). They qualita-
tively compare the different methods concluding that the meshing
becomes easier for the respective approaches but onthe other hand
the temperature predictions in the zone of contact between the
wire and the rod are worse.
2.1. LRGR approach without SGM
At NRG, in order to bridge the gap between traditional fuel
assembly simulations using system codes, subchannel codes or
porous medium approaches and the detailed CFD simulations to
analyze single sub-channels, an LRGR CFDapproach is applied. This
approach does not take into account the physics which is included
at scales withresolutionlower thanthe meshresolution. Therefore,
this method will be denoted as LRGR CFD without sub grid model
(SGM). TheLRGRCFDwithout SGMallows capturingmediumscale
ow features such as large scale recirculation zones (e.g. behind
fuel assembly blockages), which cannot be captured by the system
code, subchannel code andporous media approaches, andcanserve
e.g. to ne-tune the porous parameters which are important input
for a porous medium approach. However, it should be noted that
the prediction of detailed ow features such as secondary ows is
not feasible. The importance of secondary ows which are depend-
ent on the specic design parameters and conguration has been
reportedextensivelybye.g. Grtzbach(2003, 2007), Biemller et al.
(1996), Meyer (1994), Meyder (1975) and Nijsing and Eier (1973).
2.2. CGCFD approach with SGM
At KIT it is proposed to combine ideas of sub channel analysis
and CFD and develop a method which takes advantage of the fast
development of commercial CFDsoftware and the efciency of sub
channel analysis (Class et al., 2011a, 2011b).
The motivation for this approach was the fact, that due to com-
putational restraints it is not possible to simulate a complete fuel
assembly or a complete reactor core with state of the art CFD soft-
ware and computational power within reasonable time. This is due
to the numerous length scales that have to be resolved inside a
reactor core. The resolution must be chosen such, that the differ-
ent physical effects that can occur inside a thermo hydraulic ow
can be resolved. In case of the reactor core the numerous geome-
tries like spacers, wire-wraps, veins, rod bundles and complete
fuel assemblies makes it difcult to perform core wide simula-
tions. In such large cases, code coupling to subchannel codes like
F-COBRA-TF, NASCA, etc. is needed. The subchannel codes use
empiric correlations and case-specic model constants that have
to be adjusted or adapted for each new case or geometry. The
goal of the CGCFD is to provide an approach that is as efcient
as subchannel codes and becomes independent from experimen-
tal or empirical data. The method should close the gap between
subchannel codes and RANS simulation. Furthermore, advantage
should be taken of desirable features of modern CFD solvers, i.e.
reliability, robustness and graphical user interfaces. The aim of
the CGCFD is to compute very large and complex geometries that
cannot be tackled with a traditional CFD approach due to need
for exceedingly large computational resources. In contrast to sub-
channel analysis the CGCFD technique can be adopted to many
applications beyond the typical subchannel geometry. In Himmel
(2009) this methodology was rst applied to simulate a wire-wrap
fuel bundle of the High Performance Light Water Reactor (HPLWR)
550 F. Roelofs et al. / Nuclear Engineering and Design 250 (2012) 548559
Fig. 1. Wire wrap modeling (Bieder et al., 2010).
(Schulenberg et al., 2009). Computations using an inviscid Euler
solver on an extremely coarse grid were tuned to simulate the
thermo hydraulics by adding volumetric forces. For this reason the
approach is called the CGCFD with SGM in this paper. The vol-
umetric forces represent the non-resolved sub-grid physics. The
volumetric forces cannot be measured directly. However, they can
be accessed from detailed CFD simulations resolving all relevant
physics. Parameterization of these subgrid forces can be realized
analogous to models in sub channel codes.
2.2.1. Methodology of the CGCFD
The CGCFD method applies to problems where similar owsit-
uations repeat frequently in a large domain. Instead of simulating
the whole domain with many similar owpatterns at distinct loca-
tions, the appearing owpatterns are computed just a single time.
The global ow is simulated by assembling local behavior. Such a
method works best if the geometry can be subdivided into repre-
sentative small blocks. Fig. 2 describes the general methodology of
a CGCFDsimulation. The rst stepis the identicationof a represen-
tative block of the geometry. This step requires physical knowledge
onthe owsince the selectionof the representative blockhas major
impact on the results of the CGCFD. In the next two steps detailed
CFDsimulationof the representative block are performedincluding
parametric studies for potential owconditions. Fromthe detailed
CFD simulations volumetric forces are extracted which represent
the most important input for the CGCFD. Note that a very careful
detailed CFD must be performed since all inaccuracies will directly
enter the CGCFDresults. Details on the derivation of the volumetric
forces are explainedinSection2.2.2. Since a frictionless Euler solver
without turbulence model is used to performthe CGCFD effects of
turbulence and friction are completely embedded in the volumet-
ric forces. The next step is parameterization of the volumetric force
terms with relevant ow conditions. Here advantage can be taken
fromthe subchannel analysis which suggests a robust parameteri-
zation for CGCFDof fuel assemblies. In the last two steps the coarse
mesh of the complete geometry is setup and the Euler equations
with implemented volumetric force terms are solved.
2.2.2. Theoretical background: derivation of the sub-grid source
terms
In systemcodes or subchannel analysis the conservation equa-
tions are integrated over rather large control volumes and models
are used for friction losses and transport between neighboring
channels. When using CGCFD the required terms can be evaluated
directly from fully resolved elds. The momentum equation from
a fully resolved RANS simulation reads:
( v v) = p + +
(1)
Fig. 2. Methodology of the Coarse-Grid-CFD.
F. Roelofs et al. / Nuclear Engineering and Design 250 (2012) 548559 551
With the viscous stress tensor :
=
v +( v)
T
2
3
( v)
ij
(2)
Andtheturbulent stress tensor
t
=
t
v +( v)
T
2
3
(k
ij
)
(3)
For the CGCFD-simulation the frictionless Euler equation is used:
( v v) = p (4)
Obviously the Euler equations do not resolve the subgrid physics,
i.e. viscous dissipation or turbulence. Therefore the lost physi-
cal information is taken into account by an additional volumetric
source term
FC (5)
Upon Stokes integration of Eqs. (1) and (5) we nd for (1):
o
( v v) ndO =
o
(p) ndO +
o
() ndO +
o
(
t
) ndO (6)
And for Eq. (5):
o
( v v) ndO =
o
(p) ndO +
V
FC dV (7)
To obtain the same results fromRANS simulation with the CGCFD
that uses the Euler equation with an additional volumetric force
term, the coarse resolution should be chosen such that the corre-
sponding terms are well approximated, so that v and p are almost
equal in both representations. Then Eqs. (6) and (7) are consistent
and we can subtract Eq. (6) fromEq. (7). The reason why the values
for v and p should be almost equal is that the forces by themselves
are not able to inuence the convective terms and the pressure gra-
dient in ta way it is necessary to obtain the correct results. That
means that the values of v and p in the coarse control volume
shouldbealignedandalmost equal tothoseof thedetailedCFDsim-
ulation. In particular it is not allowed to use a coarse mesh where
appreciable ows of opposite direction cancel. These conditions
lead to an equation for the volumetric force that reads:
V
FC dV =
o
() ndO
o
(
t
) ndO (8)
Discretizationof (8) withthecorrespondingcontrol volumeleads to
a conditional equation for the volumetric force term
FC that reads:
FC V =
j
n
oj
A
j
t
n
oj
A
j
(9)
Note that these forces include turbulence and non-resolved form
losses. In general the fully resolved elds are not available. In fuel
assembly simulation, typically the same geometry with near iden-
tical ow patterns repeats all over the core region. Therefore, the
detailed simulation can be computed for a generic representative
small section and possible variations can be tabulated or parame-
terized.
3. Meshing
3.1. Low resolution meshes for LRGR CFD without SGM
Typically a detailed CFD ow analysis in a single sub channel
of a fuel assembly requires a mesh resolution of 1 million com-
putational cells, see e.g. Chandra et al. (2009). A representative
fuel assembly normally employs about 200400 suchsub channels.
Thus, nowadays a detailed simulation of an entire fuel assembly
is very challenging or only feasible with extremely large compu-
tational power. Therefore, it is important to determine a suitable
computational grid, that has appropriate convergence capabilities
and sufciently captures the main ow features and thereby the
temperature.
Simulations are carried out on a single sub-channel of a bare
square fuel assembly. A Reynolds Averaged NavierStokes (RANS)
based turbulence model is used. For the current low resolution
grid analyses ANSYS CFX 11 using the SST k turbulence model
is employed in combination with wall functions. Although higher
order models and anisotropic models have been applied for the
detailed CFD analyses, such models will not lead to improved
results whenlowresolutiongrids are employed. As the inlet section
of the European Lead cooled System(ELSY), described by Alemberti
et al. (2009), square fuel assembly was not designed yet, alterna-
tively a constant uniform ow velocity of 1.5m/s is used at the
inlet, azerostatic pressureat theoutlet andspeciedheat uxes are
applied on the walls. Lead is used as the working uid with temper-
ature dependent properties fromthe OECD handbook (OECD/NEA,
2007).
Fig. 3 shows the cross-sectional mesh for the reference grid and
two lowresolution grids used in the grid resolution study. This g-
ure also shows the lead temperature and the vorticity distribution
at the outlet for different grids. The vorticity eld demonstrates
that the grid coarsening has a strong inuence on the analyzed
ow eld. Relative to the cross-ow seen in the reference grid,
the lowest resolution grid (number 5) does not capture any sec-
ondary ow (cross-ow) at all. However, from Fig. 3 it can be
qualitatively observed that lower resolution grids do not have a
remarkable effect on the analyzed temperature eld relative to the
velocity eld. This is probably due to the fact the main mechanism
for heat transport in bare rod bundles is dominated by the axial
ow. The well knownsecondary ows whichoccur inbare rodbun-
dles and which are subject of many detailed CFD analyses have a
limited inuence on the heat transport. Nuclear reactor designers
are mostly interested in the temperature eld and pressure drop.
In the reported analyses, the main focus has been the tempera-
ture eld as geometrical details which inuence the pressure drop
largely, like e.g. the inlet and outlet header and spacers, were not
yet available. Therefore, it is proposed to neglect the deciencies
clearly detected in the less resolved ow eld and to work with
approximated owelds.
Table 1 summarizes the different grids that are used in this
grid resolution study, including the y+ values which provide an
indication of the wall treatment and their effect on the maximum
temperature at the outlet of the sub-channel. The relative error in
temperature is dened as:
Relative error =
T
max grid
T
max ref
T
max ref
T
min ref
Outlet
(10)
Here temperatures (T) are considered at the outlet of a sub-
channel at whichtheir correspondingmaximaoccurs. Thesubscript
max gridmeans the calculatedmaxima withdifferent grids, max ref
means the calculated maximum for the reference ne grid and
min ref means the calculatedminimumvalue for the reference grid.
These values are provided in Table 1.
Fromthe percentage relative error inthe RANS basedtime mean
temperatures with respect to the reference grid it is concluded
that all grids showa sufciently accurate temperature eld for the
current evaluation of the ELSY fuel assembly design which is still
subject to changes (e.g. the inlet header and spacer design have not
yet beenxed). Eventhe lowest resolutiongridwith9500elements
may beusedfor simulationof anentirefuel assemblywithsufcient
accuracy to study the effects of blockages. Furthermore, it should
be noted that the lowresolution grids all showover predictions of
552 F. Roelofs et al. / Nuclear Engineering and Design 250 (2012) 548559
Fig. 3. Meshes and temperature distributions for 3 employed grids (Reference, Grid 1, and Grid 5).
the maximumtemperatures. This will lead to conservative results
with respect to the calculated coolant and cladding temperatures.
3.1.1. Test geometry for the CGCFD
The geometry that is used for the CGCFD and the detailed RANS
CFD simulation is part of a water rod bundle experiment that was
performed at the KALLA Laboratory in Karlsruhe. The experiment
consisted of a hexagonal water rod bundle and was conducted with
isothermal conditions (Fig. 5A). The geometrical information and
owconditions can be found in Table 2. For more details about the
experiment see Litn et al. (2008).
3.1.2. Mesh creation for the CGCFD
In the assessment of a quadratic wire-wrap fuel assembly
(Himmel, 2009) (Fig. 4A), it is shown that the grid resolution should
be chosen such that at least ve control volumes cover each sub
channel. The center cell contains most of the axial ow, while the
transverse transport between neighboring sub channels is concen-
trated in the four surrounding cells. A closer look to the set up and
results of the High Performance Light Water Reactor (HPLWR) rod
bundle CGCFD can be found in Himmel (2009).
The same strategy is employed for a hexagonal assembly where
now a wedge central cell is surrounded by three hexahedral cells
represents each sub channel as shown in Fig. 4B. The spacer is
equipped with dimples which are in point contact with the rods,
so that the spacer structure is located in the center of the subchan-
nel, forcing the owto pass the spacer structure on both sides. The
spacer grid also subdivides the owinto separate owstreams, so
that no transverse transport between the subchannels is possible
in this region (Figs. 5B and 6A). For the coarse mesh 33 cells are
Table 1
Grid specications and accuracy of results.
Grid No. of elements No. of elements in
horizontal cross section
y+ range Max. clad temperature
(Tmax in
C)
% Relative error with
respect to temperature rise
Reference 1,100,000 6100 3442 519.2 0
Grid 1 37,000 750 63133 523.71 10.0
Grid 2 19,000 750 63133 523.45 9.4
Grid 3 21,000 430 98168 521.27 4.6
Grid 4 11,000 430 98168 521.37 4.8
Grid 5 9500 380 115300 525.85 14.8
F. Roelofs et al. / Nuclear Engineering and Design 250 (2012) 548559 553
Fig. 4. Employed meshes for square (A) and hexagonal fuel assembly arrangement (B).
Fig. 5. Employed geometries for the rod bundle experiment: (A) water rod bundle and (B) implemented spacer grid.
Fig. 6. Spacer region: (A) blueprint spacer and (B) obstruction due to spacer grid in CGCFD (black).
554 F. Roelofs et al. / Nuclear Engineering and Design 250 (2012) 548559
Table 2
Technical data and owconditions of the water rod bundle.
Geometry Hexagonal
Total power
Fuel pins 19
Pin diameter 8.2mm
Pich 11.48mm
Pin length 1272mm
Active height 870mm
Grid spacer 3
Mean velocity 10m/s
Mass ow 13kg/s
Max. Heat ux
Inlet temp. 25
C
Outlet temp. 25
C
used in the 60
C is used at the
inlet, a zero static pressure at the outlet and specied uniformheat
uxes are applied on the walls of the fuel rods. Lead is used as
the working uid with temperature dependent properties fromthe
OECD handbook (OECD/NEA, 2007).
Fig. 6shows the geometry andthe computational gridemployed
for the closed hexagonal ELSY fuel assembly. The closed hexagonal
assembly has a length of 1200mm. More details of the geometrical
set-up are provided in Chandra et al. (2009). As a spacer design was
not yet xedfor the ELSYrodbundle, andas the designof the spacer
should be such that the pressure drop would be minimal and no
mixing enhancers were foreseen, the simulations were performed
for a bare rod bundle, i.e. without spacers. Inlet blockages up to
16% were assumed. Furthermore, an internal blockage is modeled,
in which 40% of the height of the fuel assembly around the cen-
ter and 60% of the cross-sectional area around the central fuel pin
is blocked. This case resembles a blockage which has occurred in
the sodiumcooled KNK2 reactor (the Kompakte Natriumgekhlte
Kernreaktoranlage2
sector of the
rod bundle containing two spacer grids and a 60
sector of the
rod bundle that contains only one spacer grid where considered.
Within the investigation of these ow domains it was noticed
that the pressure drop across both spacer grids and the pin xer
are nearly identical so that only a single spacer grid was used as
representative section for the CGCFD simulation of the rod bundle.
For more details of the simulations with the half ow domain and
the 60
sector with two spacer grids see Batta et al. (2010) and
Litn et al. (2009). In bare subchannels, secondary ow may lead
to additional friction that cannot be described with the isotropic
k turbulence model. However, in the considered bundle form
Fig. 8. Velocity magnitude and temperature contours in a vertical cross-section and temperature at the outlet of the closed hexagonal ELSY fuel assembly for 16% inlet
blockage.
556 F. Roelofs et al. / Nuclear Engineering and Design 250 (2012) 548559
Fig. 9. Computational domain and grid of the square open ELSY fuel assembly.
Fig. 10. Axial velocity and temperature contours in a vertical cross-section for a complete open square ELSY fuel assembly blockage.
Fig. 11. (A) Cross-sectional plane just behind the blockage indicating the highest temperatures and (B) temperature contours at the outlet.
F. Roelofs et al. / Nuclear Engineering and Design 250 (2012) 548559 557
Table 3
Numerical- and owconditions detailed RANS CFD.
Solver Bulk velocity Turbulence
model
Number
of cells
Discretization
schemes
RANS CFD 10m/s k 500,000 Second order
Fig. 12. Mesh showing deformation of rods in spacer region and boundary condi-
tions.
losses dominate over the friction losses allowing for the use of
isotropic models at reasonable accuracy (Fig. 11).
The CGCFD simulation was performed using the open source
CFD code OpenFOAM. In a rst attempt the RANS CFD simulations
of the 60