Professional Documents
Culture Documents
Nuclear Engineering Department, Tokyo Institute of Technology, 2-12-1-N1-5 Ookayama, Meguro-ku, Tokyo 152-8550, Japan
Japan Nuclear Energy Safety Organization (JNES), 3-17-1 Toronamon, Minato-ku, Tokyo 105-0001, Japan
a r t i c l e i n f o
a b s t r a c t
Article history:
Received 29 October 2010
Received in revised form
8 February 2011
Accepted 16 February 2011
The correct evaluation of ows at transitional Reynolds number in nuclear reactors is gaining higher
importance in relation to the accident analysis for buoyancy-driven ows which dominate the heat decay
removal process. In the present paper a comparative study of different turbulence modeling and wall
treatment for the evaluation of a uid ow in transitional Reynolds number, is presented employing
computational uid dynamics (CFD). The relative performance of the models is assessed through
benchmarking of fully developed pipe ow at Reynolds number 4900 and of a 90 bend pipe at Reynolds
number 5000. Predictions of velocity proles at different locations are compared to both experimental
and accurate numerical simulations.
It has been found that the predictions between the models can vary considerably in particular in
relation to the different wall treatment employed on the wall. The results show the concerns about the
employment of the available turbulence models and wall treatments in low Reynolds number ow
regimes and explanation is provided in relation to their formulation.
2011 Elsevier Ltd. All rights reserved.
Keywords:
Transitional Reynolds number
Fully developed pipes
Bent pipes
CFD
1. Introduction
During protected accident in sodium fast reactors the establishment of coolant ow inside the plant, created either by natural
convection or by forced convection, is mandatory for the safety
removal of the decay heat released by the core. In such events,
sodium experiences low ow characteristics inside the piping
system which is likely to bring the system in condition of transition
turbulence. In CFD applications, the employment of Reynolds
Average Navier Stokes (RANS) equations with two equation
turbulence models, has always been considered a reliable tool and
widely employed for the solution of ow inside pipes. The
mentioned modelization neglects the effects of viscosity, so that it
cannot be applied in the vicinity of the wall and therefore, for wallbounded ows often empirical wall functions are employed to
bridge this gap, in order to avoid modeling the viscous effects.
However the universality of such functions breaks down for
complex ows and not fully turbulent ows, as shown by Eggels
et al. (1994) and den Toonder and Nieuwstadt (1997). In this
direction near-wall k-3 turbulence models (hereafter dened as low
Reynolds number models), which attempt to model the direct
inuence of viscosity, have been proposed (Jones and Launder,
1972). Those models are generally regarded as a more reliable
* Corresponding author.
E-mail address: pellegrini.m.aa@m.titech.ac.jp (M. Pellegrini).
0149-1970/$ e see front matter 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.pnucene.2011.02.005
2. Numerical analysis
In the following study two different benchmarks are analyzed
through the commercial CFD code, Star-CCM 5.02.009. In the
present work the RANS equations,
vuj
0
vxj
Duj
vu
1 vP
v
n j ui uj
r vxj vxi vxi
Dt
(1)
ui uj
2
k
1
kdij nt Sij C1 nt Sik Skj dij Skl Skl
3
3
3
k
1
k
1
C2 nt Uik Skj dij Ujk Ski C3 nt Uik Ujk dij Ukl Ukl
3
3
3
3
i
k2 h
C4 nt 2 Ulj Ski Skl Uli Skj Skl
3
k2
2
C5 nt 2 Uil Ulm Smj Umj Ulm Sil Slm Umm Unl dij
3
3
k2
k2
C6 nt 2 Skl Skl Sij C7 nt 2 Ukl Ukl Sij
(2)
and is employed in this analysis, where Sij and Uij are the mean
strain rate and mean vorticity tensor, respectively.
k-u turbulence model does not present any need of special treatments close to the wall, as this model acts effectively as a low Reynolds
number model without need of further modications and therefore,
even though two-layer modeling could be formulated, their
employment in the present analysis would not act as a further
comparison but as a lack of generalization. The model employed in the
present work follows the implementation given by Menter (1994),
and dened Shear-Stress Transport (SST) model where, for the
present case, the low Reynolds number modications of the coefcients, as introduced by Wilcox (Wilcox, 1993) and demonstrated to
improve results in these conditions, are employed. For the present
model the linear formulation of the eddy viscosity was employed
since the non linear formulation is not available in the employed code.
Finally the RST model, which solves the transport equations for
the components of the Reynolds stress tensor is employed with twolayer modelization which acts in a similar way as already described
for k- 3 model two-layer. This model, solving the equations for the
Reynolds stresses, accounts for ow swirl and anisotropy effects, and
therefore considered mandatory in the present study in order to
accomplish a complete comparison of the most suitable turbulence
models. Even though regarded as a general model, one of the RSTs
drawback is the introduction of modelization for all the Reynolds
917
Table 1
Summary of the turbulence models and wall treatment employed in the comparative analysis.1
Turbulence Model
Wall Treatment
low-Re
Two Layer
k-3 cubic
k-u SST
RST
X
X
X
X
1
It has to be noticed that the k-u low Reynolds number modeling has actually
a different meaning compared to the k-e low Reynolds number modeling where
damping functions are added to the standard formulation of nt and e. For the k-u
model instead it means that it is solved through the boundary layer, i.e. at low
Reynolds effects but no damping functions are added.
918
us [m/s]
0.001282
0.001190
0.001312
0.001315
0.001313
Fig. 1. Non dimensional velocity prole for Re 4900 in comparison with the fully
turbulent log law.
Fig. 2. Fully developed velocity prole for different turbulence modeling and wall treatment. a) shows the k-3 model comparing between low Re and two layer wall treatment.
b) shows comparison between RST two layer and k-u low Re.
919
Fig. 3. Geometry and mesh for the bent pipe. Dashed arrows represent the ow direction. b) represents the grid on a pipe section and on the symmetry plane.
Fig. 4. Velocity proles at 1D on the bend exit tangent. a) shows the location of the line probes, dashed arrows represent the ow direction; b) axial velocity prole on the
symmetry plane; c) radial velocity prole and d) radial velocity prole on the symmetry plane.
2.2.2. Results
Due to the results achieved in the previous section, the evaluation
of the present case was not performed for the low Reynolds number
k-3 model, whose inlet conditions for the specied geometry would
result in a laminarized prole. Therefore three different models
were compared and the results are shown in Fig. 4 b), c) and d). The
comparison demonstrates that all the models are able to predict the
creation of secondary ows and separation region, however the k-3
model in the cubic formulation and RST are able to predict, with
a reasonable grade of accuracy the axial velocity prole in comparison with the experiment and LES data. On the other hand instead
k-u shows an underprediction in the recirculation region which is
typical for turbulence model not accounting for the streamline
curvature and with a linear description of the eddy viscosity, as
shown in Fig. 4 b). Moreover k-3 cubic model gives a better estimation of the peaks of velocity of the secondary ows in the pipe
section respect to the RST, while k-u fails completely to reproduce
the behavior of the Dean vortices predicting a wrong size of the
recirculation region showing two peaks of velocity which dont join
in the centre of the domain as claried in Fig. 4 c) and d).
The underpredicted velocity prole at the pipe inlet is in some
way hidden in a 90 degree bent pipe, where the most important
feature seems to be the ability to model the anisotropy, task achieved in the present evaluation through cubic formulation (i.e. k- 3)
or Reynolds stress denition (i.e. RST). Therefore the employment
of the above mentioned modelization shows a fair ability to provide
information about bent pipe ows at low Reynolds number.
Moreover another important aspect is that the three models
above have the characteristic to achieve results in steady state
solution which is in contradiction with the experience. All of them
show a characteristic symmetry which seems averaging the instability as shown in Fig. 4 b). This aspect is indeed extremely relevant
in the present evaluation since the modelization should be able to
provide information on the velocity proles as well as the unsteady
nature of the phenomena. Extending the considerations to those
gained for the fully developed pipe therefore, we can assert that
even though the anisotropic nature of the problem is correctly
modeled, on the other hand the non proper modelization of the
viscous affected region, which is responsible in the fully developed
pipe for inaccurate modelization of the turbulent quantities
reecting in an underpredicted velocity prole, is here responsible
for the inability in the prediction of the oscillations which, in
particular during safety analysis, can cause serious underestimation
of the effects of the transients on the structure.
920
3. Conclusions
Two benchmarks for the evaluation of transitional ows were
analyzed through comparison with experiments and accurate
numerical solutions. The fully developed pipe ow shows that
damping functions, employed with k-3 models, predict a laminar
ow which does not satisfy the turbulent characteristics found both
experimentally and numerically (DNS). On the other hand the
employment of a two layer approach on the wall assures the
reproduction of the turbulent prole even though some discrepancies for the evaluation of the us are introduced. Turbulent prole
but with unpredicted normalized velocity is then shown for the k-u
model also.
The 90 bent pipe evaluation instead shows how the anisotropic
models (k-3 cubic and RST) are able to predict with reasonable
accuracy the behavior of the secondary ows which inuence the
velocity prole downstream the bend. The presence of the anisotropic model seems to be the main feature to be able to predict the
phenomenon. These considerations are validated by the behavior of
the SST model which, independently from the accurate inlet
conditions, provide not accurate results for the downstream
velocity prole and secondary ow predictions.
The study shows that for the state of art turbulence models the
one which shows a high generality of application in case of bent
pipes in transitional turbulence seems to be the k-3 model together
with the employment of two layer approach for the wall treatment
which, even though underestimating the prole of the fully
developed ow, is able to capture the main characteristics of the
ow downstream an elbow with strong curvature.
As a nal remark the authors would add that the employment of
low Reynolds number k-3 model, despite the achieved results in
Section 2.1.2, is regarded to be potentially generic and able to
improve the general predictions of the ow, such as the prediction
of the instability which is not accessed via two layer treatment of
the wall. The reasons of its failure are to be imputed to the applied