2011 International Nuclear Atlantic Conference - INAC 2011

Belo Horizonte,MG, Brazil, October 24-28, 2011

ASSOCIAÇÃO BRASILEIRA DE ENERGIA NUCLEAR - ABEN
ISBN: 978-85-99141-04-5

VERIFICATION AND VALIDATION OF A NUMERIC PROCEDURE

FOR FLOW SIMULATION OF A 2X2 PWR ROD BUNDLE

André A. C. Santos1, José Afonso Barros Filho1 and Moysés A. Navarro1

1
Brazilian Nuclear Energy Commission - CNEN
Av. Pres. Antônio Carlos, 6627 - Pampulha
30270-901 Belo Horizonte, Brazil
[email protected], [email protected], [email protected]

ABSTRACT

Before Computational Fluid Dynamics (CFD) can be considered as a reliable tool for the analysis of flow
through rod bundles there is a need to establish the credibility of the numerical results. Procedures must be
defined to evaluate the error and uncertainty due to aspects such as mesh refinement, turbulence model, wall
treatment and appropriate definition of boundary conditions. These procedures are referred to as Verification
and Validation (V&V) processes. In 2009 a standard was published by the American Society of Mechanical
Engineers (ASME) establishing detailed procedures for V&V of CFD simulations. This paper presents a V&V
evaluation of a numerical methodology applied to the simulation of a PWR rod bundle segment with a split
vane spacer grid based on ASMEs standard. In this study six progressively refined meshes were generated to
evaluate the numerical uncertainty through the verification procedure. Experimental and analytical results
available in the literature were used in this study for validation purpose. The results show that the ASME
verification procedure can give highly variable predictions of uncertainty depending on the mesh triplet used for
the evaluation. However, the procedure can give good insight towards optimization of the mesh size and overall
result quality. Although the experimental results used for the validation were not ideal, through the validation
procedure the deficiencies and strengths of the presented modeling could be detected and reasonably evaluated.
Even though it is difficult to obtain reliable estimates of the uncertainty of flow quantities in the turbulent flow,
this study shows that the V&V process is a necessary step in a CFD analysis of a spacer grid design.

1. INTRODUCTION

The spacer grids exert great influence on the thermal hydraulic performance of the PWR fuel
assembly. The presence of the spacer grids promote two antagonist effects on the core: a
desirable increase of the local heat transfer downstream the grids and an adverse increase of
the pressure drop due the constriction on the coolant flow area. Most spacer grids are
designed with mixing vanes which cause a cross and swirl flow between and within the
subchannels, enhancing even more the heat transfer performance in the grid vicinity. Due to
this significant fluid dynamic influence on the nuclear fuel assembly performance, the spacer
grids are often improved aiming to obtain an optimal commitment between pressure drop and
enhanced heat transfer.

Aiming to improve the knowledge of the flow characteristics downstream the spacer grids
and thus subsidize the nuclear fuel bundle designs, several experimental and theoretical
investigations have been conducted in the past years [1 - 9].

In recent theoretical evaluations, the use of CFD (Computational Fluid Dynamic) analysis
using three dimensional Reynolds Averaged Navier Stokes (RANS) has greatly increased,
which is due to several reasons, from the ease of use of commercial codes and development
of low costs computational systems of reasonable processing capacity, to the speed at which
results are obtained.

However, before CFD can be considered as a reliable tool for the analysis of flow through rod
bundles there is a need to establish the credibility of the numerical results. Procedures must
be defined to evaluate the error and uncertainty due to aspects such as mesh refinement,
turbulence model, wall treatment and appropriate definition of boundary conditions. These
procedures are referred to as Verification and Validation (V&V) processes [10]. In 2009 a
standard was published by the American Society of Mechanical Engineers (ASME)
establishing detailed procedures for V&V of CFD simulations [11].

According to the Standard for Verification and Validation in Computational Fluid Dynamics
and Heat Transfer – V&V 20 [11], the objective of validation is to estimate the modeling
error within an uncertainty range. This is accomplished by comparing the result of a
simulation (S) and an experiment (D) at a particular validation point. The discrepancy
between these two values, called comparison error (E), can be defined by Equation 1 as the
combination of the errors of the simulation (δ  S  True Value) and experiment
(δ  D  True Value) to an unknown True Value.

E  S  D  δ  δ (1)

The simulation error can be decomposed in input error (δ ) that is due to geometrical and
physical parameters, numerical error (δ ) that is due to the numerical solution of the
equations and modeling error (δ ) that is due to assumptions and approximations.
Splitting the simulation error in its three components and expanding Equation 1 to isolate the
modeling error gives Equation 2.

δ  E  δ  δ  δ  (2)

The standard then applies to this analysis the same concepts of error and uncertainty used in
experimental data analysis, defining a validation standard uncertainty, uval as an estimate of
the standard deviation of the parent population of the combination of the errors in brackets in
Equation 2, in such a way that the modeling error falls within the range E  u  , E  u  ",
or using a more common notation:

δ  E # u  (3)

Supposing that the errors are independent, uval can be defined as Equation 4.

u   $u%  u%  u% (4)

The estimation of these uncertainties is at the core of the process of validation. The
experimental uncertainty can be estimated by well established techniques [12]. Input
uncertainty is usually determined by any propagation techniques [11] or analytically. The
numerical uncertainty, on the other hand, poses greater difficulties to access.

INAC 2011, Belo Horizonte, MG, Brazil.

The estimation of the numerical uncertainty is called verification and is usually split into two
categories: code and solution verification. Code verification evaluates the mathematical
correctness of the code and is accomplished by simulating a problem that has an exact
solution and verifying if that solution is obtained. This activity requires extensive
programming access to the core of the code which is not available in commercial codes, due
to this it is common practice to take commercial codes as verified by the supplier.

Solution verification is the process of estimating the numerical uncertainty for a particular
solution of a problem of interest. The two main sources of errors here are the discretization
and iteration processes. The discretization error is the difference between the result of a
simulation using a finite grid and that obtained with an infinitely refined one. The methods
developed to evaluate it are based on a systematic grid refinement study where the solution is
expected to asymptotically approximate the exact value as the grid is refined, at a rate
proportional to the discretization order of the solution. The iteration error is present in codes
that use iterative solvers, where the result must converge to the exact value as the iterations
develop. It is usually estimated using the residual root mean square (RMS) between
subsequent iterations of a variable over all the volumes of the domain.

This paper presents a V&V evaluation of a numerical CFD methodology applied to the
simulation of a PWR rod bundle segment with a split vane spacer grid based on ASMEs
standard [11]. The rod bundle geometry and flow conditions similar to the used in the
experimental studies performed by Karoutas et al. [1] were assumed in the simulations.

2. THE NUMERICAL METHODOLOGY

The analysis was performed using the commercial CFD code CFX 13.0 [13] that is based on
the finite volume method. The RANS equations for mass, momentum and turbulence model
were solved. Six parallelized workstations with two 4 core processor and 24 GB of RAM
were used for all simulations amounting to ~25 hours of computing time per simulation. The
model and flow conditions were inferred from the experimental studies performed by
Karoutas et al. [1] without heat generation.

The numerical procedure used in the simulations is presented as follows.

2.1. The Model

Figure 1 shows the dimensional details of the model and of the split vane spacer grid used in
the simulations. The flow cross section of the model represents four sub-channels of a real
size fuel element. The 2 x 2 rod bundle has a total length of 660 mm with one spacer grid.
Each rod is 9.53 mm in diameter with a bundle pitch of 12.7 mm. The rod bundle has a total
flow area of 359.84 mm2 in the bare region with hydraulic diameter (Dh) of 12.02 mm. The
springs and dimples were disregarded as their form and arrangement were unknown. As
shown in Fig. 1 the model includes 0.48 mm thick straps.

INAC 2011, Belo Horizonte, MG, Brazil.

 O A B C D E F G
12.7 25.4
50 102.7 50.8 103.9 127 146
A

Inlet A’ Outlet
100 40 520
Z
Periodic boundaries
A - A’

12.7 25º

9.53 10

Measurement path 4.5

Figure 1. Computational domain with geometrical details of the spacer grid and the
vanes and positioning of the boundary conditions (units in mm).

2.2. The Mesh Parameters

A solution verification study was performed according to ASME CFD Verification and
Validation standard [11] to evaluate mesh uncertainty and proper mesh parameters for the
assembly’s simulation.

Six gradually refined non-structured tetrahedral meshes with prismatic near wall elements
(inflated) were generated for the model presented in Figure 1. Progressive grid refinements
with a factor of 1.3 were applied to edge sizing on the spacer grid and rod bundle. The ratio
between the height of the last hexahedral layer and the first prism was kept the same for all
meshes. To do this, the number of prismatic layers was adjusted for each case, since the
growth factor between layers was maintained constant with a value of 1.35. The first
prismatic layer was fixed with a height of 0.0067 mm in order to maintain a constant y+ of
~11 for all simulated meshes. This was done to avoid misleading mesh convergence
conclusions due to changes in the properties calculated by the wall function that is a directly
related to y+.

The characteristics of the generate meshes are shown in Table 1. The table includes the
resulting grid refinement ratio (r) and representative grid edge size (h) defined by Equations 5
and 6, respectively. Figure 2 shows some details of the generated meshes.

r   h  ' ( ) *+ ⁄h( )  (5)

h  Model volume⁄Number of elements of i mesh+/: (6)

INAC 2011, Belo Horizonte, MG, Brazil.

 Table 1. Meshes characteristics.

No. of Element Edge Length [mm]

No. of elements /
Mesh i hi ri prismatic
nodes Rod Bundle Spacer Grid
layers
102.891.783 /
1 0.13 1.27 4 0.296 0.135
22.636.557
50.512.494 /
2 0.17 1.29 5 0.385 0.175
12.087.648
23.791.151 /
3 0.21 1.24 6 0.500 0.225
6.159.610
12.381.326 /
4 0.26 1.27 7 0.650 0.296
3.768.827
6.083.296 /
5 0.33 1.24 8 0.845 0.385
2.000.671
3.187.324 /
6 0.42 1.27 9 1.100 0.500
1.201.629

hi = 0.42 0.33 0.26

hi = 0.21 0.17 0.13

Figure 2. Mesh cross section at the bare region of the rod bundle.

INAC 2011, Belo Horizonte, MG, Brazil.

2.3. Boundary Conditions

The entrance boundary conditions assumed in the simulations correspond to the inlet
conditions of the experiments performed by Karoutas et al. [1]. The temperature and the
pressure were set to 26.67o C and 4.83 bar, respectively, and a mean axial velocity of 6.79 m/s
was defined at the entrance of the bundle. Uniform velocity profile and 5% turbulence
intensity were assumed at the inlet as no information of these conditions was provided in
Karoutas et al. [1] study. At the outlet of each simulated section a relative average pressure of
0 Pa was defined. The surfaces of the rods and spacer grid were assumed hydraulically
smooth. Translational periodic boundary conditions were defined at the lateral extremities of
the domain as shown in Figure 1.

2.4. Numerical Parameters

The Shear Stress Transport-ω (SST) turbulence model [14] was used in this study. The SST
model includes a function to modify the turbulent eddy viscosity to account for the transport
of the turbulent shear stress. In CFX 13.0 [13] this model is employed together with an
automatic wall treatment that gradually switches from a log-law to a viscous sub–layer wall
function as the mesh is refined allowing that any degree of mesh refinement be used.

All simulations were performed for steady state condition using the high resolution numerical
scheme (formally second order) for the discretization of the conservation and BSL turbulence
model equations terms. The importance of utilizing a high resolution scheme for the
turbulence model equation terms can be seen in Figure 3. The figure shows the total pressure
loss variation as the mesh is refined for rod bundles with a spacer grid using different
numerical schemes for the turbulence equation terms. It can be seen that as the mesh is
refined the difference between results increase reaching up to 10%.

41 12
40
10
Total pressure loss [kPa]

39
Difference [%]

8
38
37 6
Upwind
36
High resolution 4
35 Difference
2
34
33 0
0 0.1 0.2 0.3 0.4 0.5
hi

Figure 3. Influence of turbulence equation discretization scheme as the mesh is refined.

INAC 2011, Belo Horizonte, MG, Brazil.

A root mean square (RMS) residual target value of 10-7 was defined as the convergence
criteria for the simulations in double precision. This target was selected after analyzing a
typical convergence behavior of the pressure loss (DP), local velocity (u) and turbulent
kinetic energy (k) during the simulation of a rod bundle with a spacer grid show in Figure 4.
It was observed that full convergence was only obtained with RMS residuals of 10-7 for the
equation solution and that residuals between 10-4 and 10-5 surprisingly show very high
residuals that if used as a convergence target can lead to incorrect conclusions.

450
1
Variable variation [%]

400
0.8
350
Variable variation [%]

0.6
300
0.4
250
0.2 DP
200
0 u
150 1.E-07 1.E-06 k
100
50
0
1.E-07 1.E-06 1.E-05 1.E-04 1.E-03 1.E-02
RMS

Figure 4. Variable variation with different target residual RMS.

By using a RMS target of 10-7 the interactive error is minimized and could be neglected in the
uncertainty evaluation as its contribution are usually many orders lower that of other sources
like discretization [10].

3. RESULTS AND ANALYSIS

Solution verification was performed using the six generated meshes based on the Grid
Convergence Index method (GCI) of the ASME V&V 20 standard [11]. The theoretical basis
of the method is to assume that the results are asymptotically converging towards the exact
solution of the equation system as the mesh is refined with an apparent order of convergence
(p) that is in theory proportional to the order of the discretization scheme. The objective of
the method is to determine p utilizing three systematically refined meshes and determine
relative to the finest grid result a 95% confidence interval (±Unum 95% = ±GCI) where the
exact solution is. In other word, the objective is to determine the expanded uncertainty
interval due to the grid for the fine mesh.

Considering the representative grid edge sizes h;+ < h < h*+ and grid refinement ratios
r  h*+ ⁄h , the apparent order of convergence p can be determined by Equations 7, 8 and 9.

INAC 2011, Belo Horizonte, MG, Brazil.

 1
p  |ln|ε*+ ⁄ε |  qp | 7
lnr 

r C  s
qp   ln B C D 8
r*+  s
s  1 · sgnε*+ ⁄ε  (9)

where εI*+  J*%  JI*+, εI  JI*+  JI , Jk denotes the variable solution on the kth grid
and sgn is the signal function (sgn(x) = -1 for x < 0; 0 for x = 0 and 1 for x > 0).

It is recommended by the standard [11] that the obtained value of p be limited to the
maximum theoretical value, which for the high resolution discretization scheme is 2. Also the
value of p can be limited to a minimum of 1 to avoid exaggerations of the predicted
uncertainty, however when limited it is recommended that the obtained value is presented for
comparison.

With the value of p the expanded uncertainty GCI can be calculated using Equation 10 using
an empirical Factor of Safety, Fs, equal to 1.25, that is recommended for studies with more
than three meshes [11].

Fs · εI
GCI   10
r C  1

When the presented procedure is applied to obtain the GCI for local variables, such as a
velocity profile, an average value of p should be used as to represent a global order of
accuracy.

In this study the pressure loss, secondary flow ratio (SF) and velocity profiles along the
bundle length were evaluated. Pressure loss was calculated between positions O and A,
shown in Fig.1, to evaluate the loss through the spacer grid (DPspacer grid) and between
positions A and G, also shown in Fig. 1, to assess the loss through the rod bundle
(DProd bundle).

The secondary flow rate (SF) was calculated at positions A to G along the measurement path
of two pitch lengths, shown as the red dotted line in Fig. 1, with the same number of points
and total length as performed by Karoutas et. al. [1] to enable proper comparisons in the
subsequent validation process. The SF is defined by Equation 11, where L is the total integral
length equal to 1.8 mm, Vlat is the lateral velocity perpendicular to the integral path, Vaxial is
the axial velocity and dx is the distance between points equal to 0.1 mm.

1 |V  |
SF  Q dx 11
L V  

Table 2 shows the results obtained applying the V&V 20 [11] method as recommended by
the standard for the integral variables pressure loss and SF. Small variances were observed
between the three most refined meshes as can be seen by the values of ε1 and ε2. However,
high variations of GCI were observed between these meshes.

INAC 2011, Belo Horizonte, MG, Brazil.

 Table 2. Verification process results for integral variables.
Parameter DPspacer grid [kPa] DProd bundle [kPa] SF at position A SF at position G
J+ 19.276 20.643 0.2166 0.0124
J% 19.522 20.413 0.2086 0.0119
J: 19.727 20.260 0.2011 0.0113
JS 22.371 20.479 0.1969 0.0049
JT 22.172 18.741 0.1868 0.0028
JU 21.824 18.484 0.1732 0.0011
p1 1.02 2.00 1.00 (0.63*) 1.48
p2 2.00 1.85 1.55 2.00
p3 2.00 2.00 2.00 2.00
p4 2.00 2.00 1.71 1.00 (0.52*)
ε1 0.246 -0.230 -0.00809 -0.00044
ε2 0.205 -0.154 -0.00741 -0.00066
ε3 2.644 0.220 -0.00424 -0.00636
ε4 -0.199 -1.738 -0.01012 -0.00212
ε5 -0.348 -0.257 -0.01354 -0.00170
GCI1 (%) 1.129 (5.9) 0.480 (2.3) 0.0381 (17.6) 0.0013 (10.6)
GCI2 (%) 0.427 (2.2) 0.439 (2.2) 0.0210 (10.1) 0.0014 (11.5)
GCI3 (%) 5.498 (27.9) 0.457 (2.3) 0.0088 (4.4) 0.0132 (117.2)
GCI4 (%) 0.415 (1.9) 3.614 (17.7) 0.0255 (13.0) 0.0100 (202.9)
* Calculated value.

The Vlat and Vaxial profiles were evaluated at positions A, D and G along the measurement
path of two pitch lengths, shown as the red dotted line in Fig. 1, as by Karoutas et. al. [1].
Table 3 shows the results obtained through the solution verification process.

Table 3. Verification process results for the velocity profiles.

Vlat [m/s] at positions: Vaxial [m/s] at positions:
Parameter A D G A D G
p 1* 1.59 1.60 1.82 1.65 1.46 2.00
p 2* 1.76 1.54 1.94 1.74 1.58 1.98
p 3* 1.72 1.85 1.73 1.69 1.88 1.58
p 4* 1.69 1.42 1.65 1.71 1.28 1.72
GCI1* 0.023 0.012 0.0065 0.029 0.019 0.073
GCI2* 0.036 0.021 0.0026 0.057 0.023 0.017
GCI3* 0.042 0.026 0.0216 0.072 0.011 0.170
GCI4* 0.034 0.078 0.0093 0.051 0.072 0.134
* Average value obtained along the measuring path.

INAC 2011, Belo Horizonte, MG, Brazil.

Figure 5 shows the profiles at position A for all simulated meshes as an example of the
behavior observed for all evaluated profiles as the mesh was systematically refined. It is
observed that the profiles tend to smoothly converge to a solution that in theory is the exact
one for this model.

0.6 hi 1.3
Position A
0.13
0.4 0.17 1.2
0.21
0.26 1.1
0.2 0.33

/ Vrefref
/ Vref

0.42 1.0
u / Vref

w/V
0.0

VVaxial
VVlat

0.9
-0.2
0.8

-0.4 0.7

-0.6 0.6
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
x /zpitch
/p x /zpitch
/p

Figure 5. Velocity profiles along measuring path position A for the six meshes.

The obtained uncertainty prediction through the solution verification process proposed by
ASME V&V 20 standard [11] showed very variable and sometimes incoherent results for the
uncertainty prediction. The method only takes in account three meshes for the estimate of
GCI and requires that the results between these meshes be well behaved to produce coherent
uncertainty estimates. Convergence must be “well behaved” due to the core assumption made
by the method that the solution is converging asymptotically as the mesh is refined. This is a
very strong assumption as it has been concluded in recent studies that it is safest to assume
that the numerical data are not within the asymptotic regime [15]. It is in fact questionable
that even the finest meshes in use today can produce solutions that are in this regime [16].
However, the obtained method gives a good insight as to how results are varying as the mesh
is refined and the estimated uncertainty may not be accurate but is a quantification on how
“well behaved” is the solution. In the present work as all three most refined meshes presented
well behaved convergence, the predicted uncertainty for the finest grid can be considered as a
more reliable estimate.

Following the solution verification, a validation process was performed. To evaluate the
pressure drop empirical correlations were used based on the works by Chun and Oh [17] and
In et. al. [18]. An uncertainty of 15% is estimated for the used correlation as indicated by the
authors. In order to assess the secondary flow (SF) and velocity profiles the LDA (Laser
Doppler Anemometry) results obtained by Karoutas et. al. [1] was used. According to the
author, LDA measurements had 0.5% of uncertainty. It must be highlighted that the
experimental results used for this study’s validation process does not match proper criteria as
discussed by Oberkampf and Trucano [19] and that for proper validation new experimental
data will be acquired in the near future.

INAC 2011, Belo Horizonte, MG, Brazil.

To determine the validation expanded uncertainty, Uval, only estimated numerical and
experimental uncertainties were considered neglecting the input contribution. Although the
input uncertainty is in fact non-neglectable, its evaluation is beyond the purpose of this study
as it is extremely complex requiring hundreds of simulations taking in account fabrication
tolerances and uncertainties in all measurable variables.

Figures 6 shows the validation error (E = S – D) and uncertainty for pressure loss as mesh is
refined and Figure 7 shows E for SF for the most refined mesh. The already discussed erratic
behavior of predicted uncertainty can be observed in Figure 7. Analyzing the results of the
most refined mesh allows to conclude that the presented model predicts with an uncertainty
of ~15% the pressure loss through the spacer grid and rod bundle and that SF is well
predicted with uncertainties lower than15% at positions A to D and G.

25
Spacer Grid Rod Bundle
15
5
E (DP) [%]

-5
-15
-25
-35
-45
0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4
hi [mm] hi [mm]

Figure 6. Pressure loss error to empirical correlation as mesh is refined.

AB C D E F G

40
30
20
E (SF) [%]

10
0
-10
-20
-30
-40
A B C D E F G
Axial position after the spacer grid (Fig. 1)

Figure 7. Error of calculated secondary flow downstream the spacer grid.

INAC 2011, Belo Horizonte, MG, Brazil.

Figure 8 shows the calculated lateral velocity profiles in positions A, D and G of Figure 1
compared to the experimental data of Karoutas et. al. [1] and the error of comparison between
them. It can be observed that a good qualitative agreement to experimental data although
quantitative errors are high compared to the amplitude of observed velocities.

0.6
Position A EXP
Error
0.4 Mesh 1

0.2
refref
/V
Vulat/ V

0.0
V

-0.2

-0.4

-0.6
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
x /zpitch
/p x /zpitch
/p
0.3 Position D EXP
Error
0.2 Mesh 1

0.1
ref
VVlatu / Vref

0.0

-0.1

-0.2

-0.3
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
x /zpitch
/p x /zpitch
/p
0.05
Position G EXP
0.04 Mesh 1
Error
0.03
0.02
ref
Vu/ /VVref

0.01
0
Vlat

-0.01
-0.02
-0.03
-0.04
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
x /zpitch
/p x /zpitch
/p

Figure 8. Lateral velocity profiles downstream the spacer grid compared to

experimental data.

INAC 2011, Belo Horizonte, MG, Brazil.

Figure 9 compares experimental data of Karoutas et. al. [1] to calculated axial velocity
profiles in positions A, D and G of Figure 1. It can be observed that the qualitative agreement
to experimental is not as good as for lateral velocities, especially at position A. However,
quantitative error is smaller compared to the amplitude of observed velocities.

1.3 Position A 0.3

Error
1.2 0.2

1.1 0.1

])
refref
refref

/V
/V

1.0

w/V
w/V

[Vaxial
-0.1
VVaxial

0.9

EE(V
-0.2
0.8
-0.3
0.7 EXP
-0.4
Mesh 1
0.6 -0.5
0.0 0.5 1.0 1.5 2.0 0 0.5 1 1.5 2
x /zpitch
/p x /zpitch
/p
1.15 Position D EXP 0.16
Mesh 1
1.10 0.12
])
refref
refref

/V

1.05 0.08
/V

w/V
w/V

[Vaxial
VVaxial

EE(V

1.00 0.04

0.95 0

0.90 -0.04
0.0 0.5 1.0 1.5 2.0 0 0.5 1 1.5 2
x /zpitch
/p x /zpitch
/p
1.3 Position G 0.2
1.2
0.1
1.1
])
ref
refref

ref
/V

0
/V

1.0
w/V
w/V

[Vaxial
VVaxial

0.9
EE(V

-0.1
0.8
EXP -0.2
0.7
Mesh 1 Error
0.6 -0.3
0.0 0.5 1.0 1.5 2.0 0 0.5 1 1.5 2
x /zpitch
/p x /zpitch
/p

Figure 9. Axial velocity profiles downstream the spacer grid compared to experimental
data.

INAC 2011, Belo Horizonte, MG, Brazil.

 4. CONCLUSIONS

A V&V evaluation of a numerical CFD methodology using CFX 13 [13] applied to the
simulation of a PWR rod bundle segment with a split vane spacer grid based on ASMEs
standard [11] was performed. The rod bundle geometry and flow conditions similar to the
used in the experimental studies performed by Karoutas et al. [1] were assumed in the
simulations.

Solution verification based on the Grid Convergence Index method (GCI) of the ASME V&V
20 standard [11] was performed using six progressively refined meshes. The pressure loss,
secondary flow ratio (SF) and velocity profiles along the bundle length were evaluated. Small
uncertainty predictions were obtained by the 3 most refined meshes. Good convergence
behavior could be observed for the velocity profiles.

The obtained uncertainty prediction through the solution verification process proposed by
ASME V&V 20 standard [11] showed very variable and sometimes incoherent results for the
uncertainty prediction that is due to the core assumption made by the method which presumes
that the solution is converging asymptotically as the mesh is refined. This is a very strong
assumption as it has been concluded in recent studies that it is safest to assume that the
numerical data are not within the asymptotic regime [15]. It is in fact questionable that even
the finest meshes in use today can produce solutions that are in this regime [16]. However,
the obtained method gives a good insight as to how results are varying as the mesh is refined
and the estimated uncertainty may not be accurate but is a quantification on how “well
behaved” is the solution. As all three most refined meshes presented well behaved
convergence, the predicted uncertainty for the finest grid could be considered as a more
reliable estimate and its results were used in the validation process.

A validation process was performed evaluating the pressure drop through empirical
correlations were used based on the works by Chun and Oh [17] and In et. al. [18] and
secondary flow (SF) and velocity profiles through LDA measurements performed by
Karoutas et. al. [1].

The validation results showed that the presented model predicts with an uncertainty of ~15%
the pressure loss through the spacer grid and rod bundle and that SF downstream the spacer
grid is well predicted with uncertainties lower than 15% at positions near the grid.

The V&V processes performed resulted in a quantifiable uncertainty and modeling error that
gives a certain degree of reliability to the proposed model for simulation of spacer grids.
Although proper validation was not obtained due to the used experimental results that do not
match proper criteria [19], this study is a step forward towards a more mature stage of CFD
development and application in the nuclear field.

ACKNOWLEDGMENTS

The authors would like to thank INB (Indústrias Nucleares do Brasil) for the motivation and
support for this study and the Fundação de Amparo à Pesquisa do Estado de Minas Gerais –
FAPEMIG for the financial support.

INAC 2011, Belo Horizonte, MG, Brazil.

 REFERENCES

1. Karoutas, Z., Gu, C., Sholin, B., “3-D Flow analyses for design of Nuclear Fuel Spacer”,
Proceedings of the 7th International Meeting on Nuclear Reactor Thermal-hydraulics
NURETH-7, New York, USA, pp. 3153-3174 (1995).
2. Holloway, M. V., McClusky, H. L., Beasley, D. E., Conner, M. E., “The effect of support
grid features on local, single-phase heat transfer measurements in rod bundles”, Journal
of Heat Transfer, vol. 126, pp. 43-53 (2004).
3. Holloway, M. V., Conover, T. A., McClusky, H. L., Beasley, D. E., Conner, M. E., “The
effect of support grid design on azimuthal variation in heat transfer coefficient for rod
bundles”, Journal of Heat Transfer, vol. 127, pp. 598-605 (2005).
4. Holloway, M. V., Beasley, D. E., Conner, M. E., “Single-phase convective heat transfer in
rod bundles”, Nuclear Engineering and Design, vol. 238, pp. 848-858 (2008).
5. Ikeda, K., Hoshi, M., “Flow Characteristics in spacer grids measured by rod-embedded
fiber laser Doppler velocimetry”, Journal of Nuclear Science and Technology, vol. 44,
no. 2, pp. 194-200 (2007).
6. Ikeda, K., Makino, Y., Hoshi, M., “Single-phase CFD applicability for estimating fluid
hot-spot locations in a 5 x 5 fuel rod bundle”, Nuclear Engineering and Design, vol. 236,
pp. 1149-1154 (2006).
7. Lee, C. M., Choi, Y. D., “Comparison of thermo-hydraulic performances of large scale
vortex flow (LSVF) and small scale vortex flow (SSVF) mixing vanes in 17x17 nuclear
rod bundle”, Nuclear Engineering and Design, vol. 237, pp. 2322-2331 (2007).
8. In, W. K., Chun, T. H., Shin, C. H., Oh, D. S., “Numerical computation of heat transfer
enhancement of a PWR rod bundle with mixing vane spacers”, Nuclear Technology, vol.
161, pp. 69-79 (2008).
9. Tóth, S., Aszódi, A., “CFD analysis of flow field in a triangular rod bundle”, Nuclear
Engineering and Design, doi:10.1016/j.nucengdes.2008.08.020 (2008).
10. Roache, P. J., “Fundamentals of Verification and Validation”, Hermosa Publishers, (2010)
11. ASME, “Standard for Verification and Validation in Computational Fluid Dynamics and
Heat Transfer - V&V 20”, ASME, (2009).
12. ISO, “Guide to the Expression of Uncertainty in Measurement”, ISO, Geneva,
Switzerland (1993).
13. ANSYS, “CFX-13.0 User Manuals”, ANSYS, (2010).
14. Menter, F. R., “Two-equation eddy-viscosity turbulence models for engineering
applications”, AIAA-Journal, vol. 32, pp. 269-289, (1994).
15. Oberkampf, W. L., Trucano, T. G., “Verification and Validation Benchmarks”, Nuclear
Engineering and Design, vol. 238, n. 3, pp. 716-743, (2008).
16. Eça, L., Hoekstra, M., Roache, P., Coleman, H., “Code verification, solution verification
and validation: an overview of the 3rd Lisbon workshop”, AIAA, (2009).
17. Lockard, D. P., “In search of grid converged solutions”, Procedia Engineering, vol. 6, pp.
224-233, (2010).
18. Chun, T. H., Oh, D. S., “A pressure drop model for spacer grids with and without flow
mixing vanes”, Journal of Nuclear Science and Technology, vol. 35, pp. 508-510 (1998).
19. In, W. K., Oh, D. S. and Chun, T. H., “Empirical and Computational Pressure Drop
Correlations for Pressurized Water Reactors Fuel Spacer Grids”, Nuclear Technology, vol.
139, pp. 72-79, (2002).

INAC 2011, Belo Horizonte, MG, Brazil.