In order to assess the accuracy and validity of subchannel, system, and computational fluid dynamics codes, the Paul Scherrer Institut has participated in the OECD/NRC PSBT benchmark with the thermalhydraulic system code TRACE5.0 developed by US NRC, the subchannel code FLICA4 developed by CEA, and the computational fluid dynamic code STARCD developed by CDadapco. The PSBT benchmark consists of a series of void distribution exercises and departure from nucleate boiling exercises. The results reveal that the prediction by the subchannel code FLICA4 agrees with the experimental data reasonably well in both steadystate and transient conditions. The analyses of singlesubchannel experiments by means of the computational fluid dynamic code STARCD with the CDadapco boiling model indicate that the prediction of the void fraction has no significant discrepancy from the experiments. The analyses with TRACE point out the necessity to perform additional assessment of the subcooled boiling model and bulk condensation model of TRACE.
An international benchmark program, namely, the OECD/NRC NUPEC BWR FullSize Bundle Test (BFBT) Benchmark [
Meanwhile, the Paul Scherrer Institute (PSI) is engaged in strengthening its subchannel analysis capability for light water reactors in Switzerland. Following the experience from the European Union NURESIM and NURISP projects, PSI has opted for the subchannel analysis code FLICA4 developed at CEA in France [
This paper reviews the different analyses of the PSBT Benchmark that were conducted at PSI, and discusses and compares the results obtained with the different thermal hydraulic codes.
The OECD/NRC PSBT benchmark aims at encouraging advancement in subchannel analysis of fluid flow in rod bundles under the conditions typical for PWRs. The benchmark is aimed at assessing the capabilities of systemcodes, subchannel codes, and CFD codes to predict void distributions, including DNB, in PWR rod bundle geometry on the basis of experimental data measured at the NUPEC test facility [
NUPEC test facility [
The benchmark consists of two phases: phase I for void distribution benchmark and phase II for DNB benchmark. Benchmark phase I includes four exercises: steadystate single subchannel exercise, steadystate and transient bundle exercises, and pressure drop exercise. Phase II consists of three exercises, which are steadystate fluid temperature exercise and steadystate and transient DNB exercises for bundle geometries.
Three different codes have been employed for the PSBT benchmark: the subchannel analysis code FLICA4 v1.10.8, the system code TRACE v5.0 Patch 2, and the CFD code STARCD v4.14.
FLICA4 is a threedimensional (3D) twophase flow analysis code developed for subchannel analysis by CEA in France. The twophase flow model in FLICA4 is based on a 4equation model, combined with a drift flux model to describe the relative velocity between phases. The driftflux model developed by Chexal et al. [
TRACE is the latest bestestimate thermalhydraulic system code developed by US NRC for analyzing steadystate and transient neutronic/thermalhydraulic behavior of light water reactors. In TRACE, a twofluid 6equation model of the steamwater flow is employed. The Gnielinski correlation [
STARCD is a CFD code developed by CDadapco. The governing transport equations solved are the conservation laws of mass, momentum, and energy for each of the two phases. These equations are solved in 3 dimensions. The numerical algorithm used for this benchmark is IPSA (Interphase Slip Algorithm) [
Total wall heat flux is made up of three components, as follows:
For the nucleation site density, the Hibiki and Ishii model [
pressure: 0.101–19.8 MPa;
mass flux: 0–886 kg/m^{2}sec;
contact angle: 5°–90°;
Number density: 1.0E+04 –1.51E+10 sites/m^{2}.
The bubble departure diameters were calculated using Kocamustafaogullari’s correlation [
The numerical algorithm and its parameters (i.e., relaxation factors, difference scheme, etc.) were kept the same for all geometries of the benchmark exercise and were not adjusted from run to run.
A series of steadystate experiments were performed to measure the void distribution in a single subchannel. The test section consists of four different geometries, which simulate various subchannels in a bundle as described in Table
pressure: 4.90–16.6 (MPa),
mass Flux:
power: 12.5–90.0 (kW),
inlet temperature: 164.1–345.0 (°C).
Geometry and power shape for singlechannel tests [
Item  Data  

Assembly (subjected subchannel) 




 
Subchannel type  Center (typical)  Center (thimble)  side  corner 
Number of heaters  4 × 1/4  3 × 1/4  2 × 1/4  1 × 1/4 
Axial heated length (mm)  1555  1555  1555  1555 
Axial power shape  Uniform  Uniform  Uniform  Uniform 
■: subjected subchannel, white circle: heated rod, gray circle: thimble rod.
Geometrical parameters of each subchannel [
Subchannel type  

Typical (S1)  Thimble (S2)  Side (S3)  Corner (S4)  
Flow area, mm^{2}  107.098  107.098  68.464  42.592 
Heated perimeter, mm  29.845  22.384  14.923  7.461 
Wetted perimeter, mm  54.645  54.645  44.923  33.161 
Crosssectional view of subchannel test assembly [
Identical nodalizations were used for the singlechannel analyses carried out with FLICA4 and TRACE. The singlechannel test section was nodalized by means of a onedimensional pipe divided into 32 axial nodes. Four different geometries were prepared, to take into account the difference in crosssection and hydraulic diameter of each case. The ChexalLellouche driftflux model was employed for the FLICA4 calculations. Multipliers for the turbulent diffusivity and viscosity,
In order to reduce the needs in computational resources, all subchannel geometries were modeled utilizing subchannel symmetry. 1/8 symmetry could be employed for the central subchannel, while the half symmetry could be exploited for the central (thimble) side and corner subchannels. Heated rods were not modeled explicitly, instead heat fluxes were applied as boundary conditions on the channel side walls. The height of the computational model is the same as the heated length of the experimental test section (1555 mm). A flat velocity profile was used for the inlet boundary condition, since no additional data about the inlet manifold geometry was made available to the benchmark participants. A summary of the boundary conditions applied to the CFD model is shown in Figure
Computational cells distribution and mesh size were kept the same for each run in the test series. The CFD meshes used are shown in Figure
53600 hexahedral cells for S1;
214400 hexahedral cells for S2;
114800 hexahedral cells for S3;
62200 hexahedral cells for S4.
CFD model.
Hexahedral computational mesh.
The number of axial cell layers was set to 100 for each subchannel geometry. Mesh sensitivity studies performed on the 1/8 symmetry sector model of the type S1 geometry showed that there is no significant effect on the section averaged void fraction value after reaching 250 cells per axial cell layer.
The results of the void fraction for S1 to S4 are depicted in Figure
The bias of the calculation results from FLICA4 and TRACE has been assessed by using the linear regression method. Figure
Void fraction predictions for singlechannel exercise.
Results from linear regression analyses.
Figure
The red lines indicate error bars of ±4%. Figure
Section average void fraction results.
Case S1
Case S2
Case S3
Case S4
Void fraction distribution in a measuring section (CFD versus experimental data).
Axial void profile for 4 runs with S1 type of geometry.
A series of steadystate experiments were performed to measure the void distribution in bundle geometry. The void fraction from the experiment is the one averaged over the four central subchannels. The three different 5 × 5 bundles as given in Table
Geometry and power shape for each bundle [
Item  Data  

Assembly 



 
Rods array  5 × 5  5 × 5  5 × 5 
Number of heated rods  25  25  24 
Number of thimble rods  0  0  1 
Heated rod outer diameter (mm)  9.50  9.50  9.50 
Thimble rod outer diameter (mm)  —  —  12.24 
Heated rods pitch (mm)  12.60  12.60  12.60 
Axial heated length (mm)  3658  3658  3658 
Flow channel inner width (mm)  64.9  64.9  64.9 
Radial power shape  A  A  B 
Axial power shape  Uniform  Cosine  Cosine 
Number of MV spacers  7  7  7 
Number of NMV spacers  2  2  2 
Number of simple spacers  8  8  8 
MV spacer location (mm)  471, 925, 1378, 1832, 2285, 2739, 3247  
NMV spacer location (mm)  2.5, 3755  
Simple spacer location (mm)  237, 698, 1151, 1605, 2059, 2512, 2993, 3501 
White circle: heated rod, gray circle: thimble rod.
MV: mixing vane. NMV: no mixing vane.
Spacer location is distance from bottom of heated length to spacer bottom face.
Cosine power profile [
Node  Relative Power 

Cosine  
(Bottom)  
1  0.42 
2  0.47 
3  0.56 
4  0.67 
5  0.80 
6  0.94 
7  1.08 
8  1.22 
9  1.34 
10  1.44 
11  1.51 
12  1.55 
13  1.55 
14  1.51 
15  1.44 
16  1.34 
17  1.22 
18  1.08 
19  0.94 
20  0.80 
21  0.67 
22  0.56 
23  0.47 
24  0.42 
(Top) 
Bundle average spacer pressure loss coefficients [
Spacer type  Loss Coefficient 

Simple spacer (SS)  0.4 
Nonmixing vanes spacer (NMV)  0.7 
Mixing vanes spacer (MV)  1.0 
Radial power distribution [
Type A
Type B
Analyses of the steadystate bundle experiments have been conducted by means of FLICA4 and TRACE only.
Since the geometry and the radial power distribution adopted for the bundle tests are fully symmetric, the symmetric boundary condition was employed so that each test assembly was described by means of a 1/8 symmetrical model as depicted in Figure
The ChexalLellouche driftflux model was employed and a value of
1/8 symmetrical model for FLICA4.
The bundles are described by using a 3D rectangular VESSEL component of TRACE as depicted in Figure
Nodalization for TRACE.
Overview of TRACE Model
Topdown view of test section
The void fraction was measured at three different elevations: 2216 mm (lower), 2669 mm (middle), and 3177 mm (top) from the bottom of the test section. The calculated void fractions at the four central subchannels were averaged and compared with the experimental data. The results of both codes presented in Figure
Mean absolute error of void fraction.
Code  Lower  Middle  Upper  Overall 

FLICA4  
Averaged error (%)  1.18  3.54  1.59 

Mean absolute error (%)  2.95  5.29  4.04 

TRACE  
Averaged error (%)  9.54  7.45  3.35 

Mean absolute error (%)  10.1  9.49  6.49 

Void fraction of bundle exercise.
Results from linear regression analysis.
Axial void fraction profile of case B5.2442.
The transient experiments were conducted with assemblies B5, B6, and B7. Four different transient experiments were performed for each assembly: a power increase, a flow reduction, a depressurization, and a temperature increase. The initial conditions for each transient test are summarized in Table
Initial conditions for transient bundle tests [
Initial conditions  

Test series  Assembly  Pressure (kg/cm^{2} a)  Mass flux (10^{6} kg/m^{2} h)  power (kW)  Inlet temperature (Celsius)  Transients 
5T  B5  154.2  11.95  2282.0  300.4  Power increase 
153.8  11.93  2244.0  301.2  Flow reduction  
153.0  11.92  2236.0  300.4  Depressurization  
152.5  11.94  2230.0  301.7  Temperature increase  
 
6T  B6  158.2  11.55  2621.0  288.1  Power increase 
158.4  12.03  2574.0  288.8  Flow reduction  
154.6  12.02  2556.0  288.2  Depressurization  
157.2  11.92  2603.0  288.8  Temperature increase  
 
7T  B7  158.2  12.02  2500.0  291.9  Power increase 
158.1  12.04  2405.0  292.0  Flow reduction  
155.0  11.99  2577.0  291.8  Depressurization  
158.8  11.99  2496.0  290.2  Temperature increase 
The transient experiments were analyzed by means of FLICA4 only. The FLICA4 models for the transient experiments are the same as the ones used for the steadystate analyses. As for the steadystate cases, since the void fraction is averaged over the four central subchannels in the experiment, the void fraction of the subchannel 6 in the FLICA4 model can be considered as the averaged void fraction.
The analysis has been carried out for all transient cases specified as part of the benchmark. As an example, the calculation results from transient exercise 5T are reported in Figure
Axial void fraction of transient exercise 5T.
Axial void fraction with original and modified time.
5T—temperature increase
6T—temperature increase
7T—temperature increase
PSBT benchmark phase II aims at developing and assessing mechanistic models for DNB prediction. Both steadystate and transient DNB exercises are included in phase II and all the tests were carried out with bundles. In the experiments, the occurrence of DNB was detected by a sudden increase of the surface temperature measured by thermocouples attached to the heater rods. The heating power was increased in fine steps to the vicinity of DNB power estimated by preliminary analysis and experience. In the experiment, a sudden surface temperature increase of more than 11°C confirmed the occurrence of DNB and the corresponding DNB power was defined as the power just before the sudden temperature increase.
As listed in Table
Test assemblies for phase II [
Assembly  Reference fuel type  Rods 
Type of cell  Power distribution  
Radial  Axial  
 
A0  17 × 17 M  5 × 5  Typical cell  A  Uniform 
A1  Typical cell  C  Uniform  
A2  Typical cell  A  Uniform  
A3  6 × 6  Typical cell  D  Uniform  
A4  5 × 5  Typical cell  A  Cosine  
A8  Thimble cell  B  Cosine  
A11  Typical cell  A  Cosine  
A12  Thimble cell  B  Cosine 
Radial power distribution [
Type A
Type B
Type C
Type D
FLICA4 is employed for the analyses of PSBT benchmark phase II. Thermal hydraulic models for the FLICA4 calculations are generated on the basis of information on the geometry and the power distribution. Unlike in phase I, it was not always possible to adopt 1/8 symmetry in the models for phase II due to the radial power distribution C, which allows implementing 1/2 symmetry only. Therefore, for consistency, all models have been generated by using a 1/2 symmetry, as depicted in Figure
1/2 symmetry model.
5×5 assembly
6×6 assembly
The ChexalLellouche driftflux model was employed and a value of
However, due to limitation in the application ranges, the W3 correlation could not be used for all the cases in phase II. Sensitivity studies carried out with both correlations, which will be discussed in more detail in Section
Selected cases from the test series highlighted in Table
Test series for exercise 2 of PSBT benchmark phase II [
Test series  Test 
Assembly  Test mode  Measurement  

Steady 
Transient  DNB  Fluid  
0  A0  Y  Y  
1  5 × 5  A1  Y  Y  
2  A2  Y  Y  
 
3  6 × 6  A3  Y  Y  
 
4  A4  Y  Y  
8  A8  Y  Y  
11T  5 × 5  A11  Y  Y  
12T  A12  Y  Y  
13  A4  Y  Y 
Results of steadystate DNB exercises.
The transient DNB benchmark was conducted for test series 11T and 12T, which include four transient scenarios in each test as indicated in Table
Test series for transient DNB exercise [
Test 
Assembly  Initial conditions  Transients  

Power 
Mass flux (10^{6} kg/m^{2}h)  Pressure 
Inlet temperature (°C)  
1 1T  A4  2.50  11.18  156.2  291.0  Power increase 
2.50  11.19  156.1  293.1  Flow reduction  
2.52  11.28  156.3  291.7  Depressurization  
2.48  11.04  154.6  291.6  Temperature  
12T  A8  2.51  11.40  156.1  291.3  Power increase 
2.51  11.71  156.3  292.5  Flow reduction  
2.50  11.42  156.2  290.6  Depressurization  
2.50  11.38  155.8  291.2  Temperature 
Results from the transient DNB benchmark exercise are depicted in Figure
Results of transient DNB exercise.
FLICA4 includes three models to predict the CHF: the W3 correlation, the Groeneveld lookup table, and the SUDO correlation [
The W3 correlation is one of the most widely used correlations for evaluation of DNB in PWRs and it is applicable to circular, rectangular, and rod bundle geometries. The correlation has been developed for axially uniform heat flux, with a correction factor for nonuniform flux distribution. In addition, local spacer effects can be taken into account by specific correction factors [
The Groeneveld lookup table was developed jointly by AECL (Canada) and IPPE (Russia) and has a very wide range of applications. Compared against the combined AECLIPPE CHF database, it is known that the Groeneveld lookup table can predict the CHF data with an overall rootmeansquare (RMS) error of 7.82%.
Selected cases from test series 0 and 8 of the benchmark are employed to assess the two CHF models. Each test series represents tests with or without a guide thimble at the center. The calculation results presented in Figure
DNB Results with different CHF models.
Test series 0
Test series 8
In order to assess the range of validity and the accuracy of the subchannel analysis code FLICA4 and the commercial CFD code STARCD including a boiling model recently developed by CDadapco, PSI has participated in an international benchmark based on the NUPEC PWR subchannel and bundle tests (PSBT) organized by OECD/NEA and US NRC. The tests have been analyzed by employing the US NRC code TRACE as well, in order to assess the applicability of TRACE to a subchannel analysis.
The results from the void distribution benchmark indicate that a reasonable agreement with the experimental data is obtained with FLICA4. The void fraction prediction by STARCD shows no significant discrepancy for the single Subchannel experiments. It is worthwhile mentioning that all the benchmark cases were calculated without any tuning of the boiling model and numerical algorithms. This fact demonstrates that the boiling model used in STARCD is able to predict void fractions over a wide range of void fraction values with acceptable accuracy. TRACE instead tends to overpredict the void fraction, especially for values lower than 40%. The analysis of the axial void fraction profile reveals that overprediction by TRACE is caused by an earlier increase of the void fraction along the axis of the channel, pointing out the necessity of additional assessments for the subcooled boiling model and bulk condensation model currently implemented in the TRACE code.
The DNB benchmark exercises have been analyzed with FLICA4 only. The steadystate benchmark exercise results indicate that FLICA4 slightly underpredict the DNB power. However, considering the accuracy of Groeneveld lookup table and the uncertainties in the experimental data, it can be concluded that the prediction of the DNB power by FLICA4 is acceptable and furthermore conservative. The transient DNB prediction reveals that FLICA4 predicts DNB earlier than in the experiments, which is consistent with the result from the steadystate DNB benchmark. In addition, an assessment of the CHF models of FLICA4 has been carried out by using the benchmark data. It is found that the Groeneveld lookup table predicts more conservative DNB powers with higher accuracy than the W3 correlation.
The authors would like to thank the benchmark organizers and colleagues participating in the PSBT benchmark for the cooperation and meaningful discussion. This work has been performed within the STARS project (