Nuclear Engineering and Design 240 (2010) 1548–1557

Contents lists available at ScienceDirect

Nuclear Engineering and Design

journal homepage: www.elsevier.com/locate/nucengdes

Computational study of conjugate heat transfer in T-junctions

Simon Kuhn a , Olivier Braillard b , Bojan Ničeno a,∗ , Horst-Michael Prasser a,c
a
Paul Scherrer Institute, 5232 Villigen, PSI, Switzerland
b
Commissariat à l’Énergie Atomique, CEA Cadarache 13108 St Paul lez Durance Cedex, France
c
ETH Zürich, Sonneggstrasse 3, 8092 Zürich, Switzerland

a r t i c l e i n f o a b s t r a c t

Article history: In this work we focus on the numerical prediction of temperature fluctuations induced in solid materials
Received 2 December 2009 through turbulent mixing processes. As test case we use the mixing of two streams of different tempera-
Received in revised form 18 January 2010 ture in a T-junction. Due to the turbulent mixing of the two streams temperature fluctuations occur which
Accepted 19 February 2010
are also transferred to the solid walls in contact with the fluid. Such fluctuations in the solid material
may lead to thermal fatigue and are therefore relevant for the lifetime management of components used
in nuclear power plants (NPP).
We investigate the mixing in T-junctions made of different materials and having different pipe wall
thicknesses. The temperature difference between the streams in the main and side branch is set to 75 ◦ C
and the mass flow rate in the main pipe is three times larger than in the side branch. In a first step we
perform a set of simulations by using different formulations of the large-eddy simulation (LES) subgrid
scale model, i.e. classical Smagorinsky model and dynamic procedure, to identify the influence of the
modeled subgrid scales on the simulation results. The comparison between available experimental data
and the numerical results reveals a good agreement when using the dynamic procedure. In a second step
we address the temperature fluctuations in the solid wall subject to the wall thickness. The influence of
the wall thickness is represented as a damping effect on the temperature fluctuations in radial direction
in the pipe material. This study shows the capability of LES to predict thermal fluctuations in turbulent
mixing.
© 2010 Elsevier B.V. All rights reserved.

1. Introduction (Muramatsu and Ninikata, 1996). However, this issue is relevant

for various types of nuclear reactors as the incident at the light
In T-junctions, especially in the regions where hot and cold water reactor Civaux-1 revealed (Jungclaus, 1998). With respect to
streams are not completely mixed, significant temperature fluc- plant life management several experimental and numerical stud-
tuations can occur in the whole mixing region and even close to ies were conducted to address thermal fatigue of components in
the solid walls. The presence of these fluctuations may also induce nuclear power plants (e.g. Hu and Kazimi, 2003, 2004; Lee et al.,
temperature fluctuations in the solid walls, which leads to cyclic 2004, 2009).
thermal stresses and eventually to fatigue cracking. To be able to The present study applies LES to predict the fluid velocity field
conduct material, structural and damage analysis, knowledge about and the temperature field in the fluid and the solid in two different
the exact location, amplitude and frequency of those temperature mixing T-junctions using the commercial CFD software FLUENT.1
fluctuations is crucial. In this respect computational fluid dynam- The two cases differ by the wall material and its thickness: (i) 1 mm
ics (CFD) is used to predict the mixing between hot and cold water of brass and (ii) 5 mm of steel. The properties of both materials
in a T-junction and recent studies have shown large-eddy simula- are listed in Table 1. The first design is based on CEA’s FATHERINO
tions (LES) to be successful (Hu and Kazimi, 2006; Coste et al., 2008; “Skin of Fluid Mockup” experiment and these results are used for
Niceno et al., 2008). the validation of the computations. For both cases the mixing T-
Thermal striping leading to thermal fatigue was first investi- junction has an inner pipe diameter of 54 mm. We computed for
gated in the context of liquid-metal-cooled fast breeder reactors both designs a case where the hot fluid enters the main pipe with
a mean velocity of 2.55 m/s and a temperature of 83 ◦ C. The mean
velocity of the cold fluid in the branch pipe was set to 0.85 m/s and
∗ Corresponding author at: Paul Scherrer Institute, Thermal-Hydraulics Labora-
tory, Nuclear Energy and Safety Department, 5232 Villigen PSI, Switzerland.
Tel.: +41 563104149.
1
E-mail address: [email protected] (B. Ničeno). FLUENT is a trademark of ANSYS Inc. (http://www.ansys.com).

0029-5493/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.nucengdes.2010.02.022
 S. Kuhn et al. / Nuclear Engineering and Design 240 (2010) 1548–1557 1549

perature difference reaches 75 ◦ C with a maximum mixing flow rate

Nomenclature of 50 m3 /h. The facility is able to create a very turbulent mixing in 2
T-junction mockups. The heat transfer coefficient is estimated to be
˛ heat transfer coefficient 14,000 W/(m2 K) with Re = 4.9 × 105 . These physical parameters
ˇ volumetric thermal expansion coefficient are similar to industrial mixing T-junctions.
Bi Biot number, ˛t/ The FATHERINO 2 test procedure consists of mixing the flu-
cp heat capacity ids from two vessels in the mockup. These vessels are initially
CS Smagorinsky constant filled with water and conditioned at low (5 ◦ C) and high temper-
 LES filter width ature (80 ◦ C). There is a motor pump for each line (cold and hot
ˆ LES test filter width branches) and the flow rate is controlled by operating valves. To
d wall distance ensure the boundary conditions in each inlet a homogenization
g acceleration due to gravity device is installed to control the turbulent flow and to provide
 von Kármán constant a known velocity profile. A thin brass mockup (“Skin of Fluid
 heat conductivity Mockup”) was designed to visualize the field of mean and fluc-
 dynamic viscosity tuating temperature in the fluid with minimized attenuation in
 kinematic viscosity real time. The basic idea is to use the mockup as an envelope of
t turbulent viscosity the fluid, and to benefit from its thermal properties to efficiently
p̄ filtered pressure visualize the thermal information. Only the interface between the
Pr Prandtl number fluid and the structure will be observed, but this represents the
Prt Turbulent Prandtl number most interesting part of the mixing domain which contributes
 fluid density to the assessment of the thermal load for mechanical stress
Re Reynolds number for pipe flow, Um D/ calculation.
Sij rate of strain tensor The mockup has to exhibit a good thermal transparency; tem-
t pipe wall thickness perature fluctuations in the fluid must be visualized with a very
T temperature low level of attenuation. To obtain such a property the structure
 
ij = ui uj subgrid turbulent stress must have a very low Biot number (ratio between the heat transfer
  coefficient and the conductivity of the material). As a consequence
 i = T ui subgrid turbulent heat flux
the selected material for the mockup needs to have a large ther-
ū, v̄, w̄ resolved velocity components
mal conductivity and a thin thickness. In the same time, a too
Um mixed velocity
thin mockup increases the risk to be damaged by the inertia of
x, y, z Cartesian coordinates
the fluid and pressure effects in the flow. Because of all these
demands the mockup was finally constructed of brass with a wall
thickness of 1 mm, which yields a Biot number of Bi=0.126 (see
the temperature to 8 ◦ C. This results in a temperature difference Fig. 2).
of 75 ◦ C of the pre-mixed fluid streams and a Reynolds number The mockup consists of 3 pipes and 1 T-join, the different parts
(Re) in the mixed stream of 2.7 × 105 . On the outer pipe surface are joined together without any additive material (brazed joint or
we applied a convective boundary condition with a heat transfer cement) to avoid undesired thermal effects in the area of interest.
coefficient of 2 W/(m2 K) and a free stream temperature of 293 K. A mat black color covers the parts used for infrared visualization
These values were chosen to mimic the experimental conditions of (see Fig. 2).
the FATHERINO setup (Braillard, 2008).
The practical relevance of the performed simulations is given 3. Subgrid scale models and numerical details
by the large value of Re and the varying fluid properties due to the
large temperature difference. We focus in detail on the dynamics The equations describing turbulent flows and heat transfer are
of the mixing and the temperature fluctuation in the solid and the given by the conservation of mass, momentum and thermal energy.
fluid with respect to the wall thickness. The generated data are The spatially filtered form of the conservation of momentum equa-
relevant for the subsequent prediction of thermal stresses and the tion can be written as
corresponding life span of the T-junctions.    
∂ ( ui ) ∂ ui uj ∂p̄ ∂ ∂ ( ui )
+ =− +  − ij + Fi , (1)
∂t ∂xj ∂xi ∂xj ∂xj
2. The FATHERINO 2 facility
and the conservation of thermal energy equation becomes
The FATHERINO 2 facility (see Fig. 1) is located at CEA Cadarache        
research center. The FATHERINO facility is designed to perform ∂ T̄ ∂ uj T̄ ∂  ∂T̄
thermo-hydraulic tests in mixing T-junctions. The available tem- + = −  j , (2)
∂t ∂xj ∂xj Pr ∂xj

where p̄ is the filtered pressure field, Fi is an external force, 

Table 1
Properties of the brass and steel mockup (Braillard, 2008). denotes the dynamic viscosity of the fluid, Pr is the Prandtl number
defined as
Properties of brass
cp
Density  (kg/m3 ) 8522 Pr = , (3)

Thermal conductivity  (W/(mK)) 111
Heat capacity cp (J/(kg K)) 385 and ij ,  j represent the unresolved turbulent stress and turbu-
lent heat flux, respectively. The coupling between the temperature
Properties of steel and the velocity fields is accomplished by applying the Boussinesq
Density  (kg/m3 ) 7821 approximation:
Thermal conductivity  (W/(mK)) 16  
Heat capacity cp (J/(kg K)) 440 FB = − gi ˇ T̄ − Tref , (4)
1550 S. Kuhn et al. / Nuclear Engineering and Design 240 (2010) 1548–1557

Fig. 1. Schematic of the FATHERINO 2 facility.

Fig. 2. Detailed view of the T-junction and the “Skin of Fluid Mockup” used for the FATHERINO experiment.

where ˇ is the thermal expansion coefficient of the fluid, and gi the The eddy viscosity is defined as
acceleration due to gravity. The working fluid for this test case is 2
t = (CS ) |S̄|, (6)
water, its properties are outlined in Table 2.
The LES is performed by using the classical and the dynamic where  denotes the length scale of the unresolved motion related
Smagorinsky model for the subgrid scales (SGS) (Germano et al., to the volume of the computational cell as: V
1991). For both models the unresolved traceless subscale stresses
1/3
ij are related to the rate of strain Sij of the resolved velocity field  = (V ) , (7)
by employing the Boussinesq eddy-viscosity concept:
and |S̄| is the magnitude of the strain rate defined as:
  
1 1 ∂ui ∂uj
ij − kk ıij = −2t Sij . (5) |S̄| = 2Sij Sij , Sij = + (8)
3 2 ∂xj ∂xi

In the classical formulation of the Smagorinsky model CS is a

constant, which is set to a value of 0.1 in our case (e.g. Piomelli
Table 2 et al., 1988). To ensure correct near wall behavior of the turbulent
Properties of water at 20 ◦ C (Wagner and Kruse, 1998). viscosity wall-damping functions must be applied when using this
Property classical Smagorinsky approach. In FLUENT this is implemented by
computing the turbulent viscosity as
Density  (kg/m3 ) 998.2
Kinematic viscosity  (m2 /s) 1 × 10−6
t = LS2 |S̄|, (9)
Thermal conductivity  (W/(m K)) 0.597
Heat capacity cp (J/(kg K)) 4182 where the characteristic length scale LS is computed from
Thermal expansion coefficient ˇ (1/K) 2.07 × 10−4
Prandtl number Pr 7 LS = min(d, CS ), (10)
 S. Kuhn et al. / Nuclear Engineering and Design 240 (2010) 1548–1557 1551

where  denotes the von Kármán constant ( = 0.42) and d the

distance to the closest wall. The subgrid turbulent heat flux is cal-
culated from the simple gradient diffusion hypothesis as

t ∂T̄
 i=− (11)
Prt ∂xi
where Prt = 0.85 is the turbulent Prandtl number.
For the dynamic procedure used with the Smagorinsky model
CS is no longer a constant but evaluated with the expression
1 Lij Mij
CS = − , (12)
2 Mkl Mkl

where Lij = ui uj − ui uj represents the resolved turbulent stress

of the scales between  and a coarse  ˆ (where  ˆ = 2), and
ˆ 2 
Mij =  |S̄|Sij −  |S̄|Sij represents the contribution of the mod-
2

eled stress of those scales. The value of CS is clipped at zero to

avoid numerical instabilities. The dynamic procedure to obtain
the Smagorinsky constant ensures correct near wall behavior
of the turbulent viscosity eliminating the necessity to include
the wall-damping functions. The SGS turbulent Prandtl num-
ber is obtained by applying the dynamic procedure to the SGS
flux.
The computations were performed by using the commercial
software package FLUENT, which provides a finite-volume based
second-order accurate solver. The PRESTO algorithm was used for
pressure–velocity coupling, the diffusive and convective terms of
the discretised equations were approximated by a second-order
bounded central differencing scheme (BCD), and time integration
was performed by a fully implicit second-order method. The
modeled flow domain covered 8 hydraulic diameters of both inlet
branches before the junction and 8 hydraulic diameters down- Fig. 3. Contour of the mean streamwise velocity in the mixing zone of the T-junction.
stream. The mesh for this domain consisted of 1,900,800 volume
elements for the fluid and 288,000 volume elements for the solid els. The two branches intersect at an angle of 90◦ , and the cold
and yielded a maximum y+ of 5 in the mixing zone. The inlet fluid (blue in Fig. 3(a) and (b)) enters the junction from the bot-
boundary conditions were specified as velocity inlets with a vortex tom. Consequently the side flow has zero momentum in x-direction
method to add random fluctuations, on the outer pipe surface (streamwise direction of the main branch) and exerts blockage
a convective boundary condition was applied (˛ = 2 W/(m2 K), effects on the main flow. This blockage can be observed in Fig. 3(a)
T∞ = 293 K). A total of 4 s of flow time (real time) was computed and (b) by the distortion of the mean streamwise velocity compo-
for each case. nent distribution. Both SGS models predict the same velocity field
upstream of the intersection of the pipe branches. However, a major
4. Comparison to the FATHERINO experiment of CEA difference is observed downstream: For the classical Smagorinsky
model (Figs. 3(a) and 4) a zone of flow separation and reversal
In a first step we address the influence of the SGS model on is found between the streamwise coordinates x ≥ 0.04 mm and
the simulation results. Therefore we apply LES with the classical x ≤ 0.065 mm in the vicinity of the lower pipe wall. This region of
Smagorinsky model and dynamic procedure to the “Skin of Fluid flow reversal is not predicted by the dynamic procedure (Fig. 3(b)).
Mockup”, which is made of brass with a wall thickness of 1 mm, and This difference between the two SGS models in the near wall
compare the predictions with the experimental results. The mea- region can be explained by the usage of the turbulent viscos-
surements are performed by applying infrared thermography to the ity wall-damping function which must be applied in the classical
pipe surface, i.e. an ensemble of infrared images is recorded and Smagorinsky approach.
statistically analyzed to obtain the distribution of the mean surface To investigate the effect of the presence or absence of flow rever-
temperature and the root mean square (RMS) of the temperature sal on the momentum and scalar field we plot instantaneous and
fluctuations. mean velocity components together with the temperature field in
For the case of mixing two fluid streams in a T-junction differ- a cross-section downstream of the junction at x = 120 mm. Fig. 5
ent flow regimes are observed. Upstream of the junction the flow depicts an instantaneous realization of the velocity components
is fully developed in the main branch and the side branch. Behind in y and z directions together with a contour of the instantaneous
the intersection of the two branches intensive mixing occurs and temperature for both models in this cross-section. In case of the
further downstream the flow returns to the fully developed state. classical Smagorinsky model (Fig. 5(a)) this region of mixing is char-
In the next section this region of intense mixing is addressed, fol- acterized by a pair of counter-rotating vortices at the bottom of the
lowed by an investigation of the mean temperature field and the pipe and two separate larger vortices in the upper part. In addi-
temperature fluctuations in the solid pipe wall. tion the two fluid streams of different temperature are still sharply
distinguishable, the cold fluid is accumulated in the lower pipe sec-
4.1. Velocity and temperature field in the mixing zone tion and a steep temperature gradient is found. These results are
attributed to the zone of flow reversal upstream which entraps the
Contours of the mean streamwise velocity component in the cold fluid in the vicinity of the bottom wall. For the dynamic proce-
region of intensive mixing are depicted in Fig. 3 for both SGS mod- dure (Fig. 5(b)) a more chaotic picture is observed. There are several
1552 S. Kuhn et al. / Nuclear Engineering and Design 240 (2010) 1548–1557

Fig. 4. Mean streamwise velocity profiles at positions x = −0.05 and x = 0.05.

vortices of different sizes in the instantaneous velocity field which predicts a region of flow reversal downstream of the junction which
contribute to the mixing process. Consequently the two streams of also affects the scalar mixing. In the next section we will address the
different temperature are mixed more efficiently, fluid packets of resulting temperatures and temperature fluctuations in the wall for
elevated temperature are found in the lower part of the pipe and both models.
there is no pronounced temperature gradient.
Fig. 6 depicts the mean velocity components in y and z direc- 4.2. Mean and fluctuating temperature fields on the outer wall
tions and the contour of the mean temperature. The statistical mean
verifies the observations made for the instantaneous cases. For the The failure mechanism of thermal fatigue is connected to tem-
classical Smagorinsky model (Fig. 6(a)) the velocity field is charac- perature fluctuations in the solid induced by the mixing processes
terized by a pair of counter-rotating vortices in the lower section in the fluid. This means that the computations need to capture these
of the pipe and two separate vortices in the upper section. The temperature fluctuations and their frequency properly in order to
latter transport some cold fluid into the stream of high temper- calculate the resulting thermal stresses in the pipe material accu-
ature, however the lower part of the pipe is still filled with fluid of rately. In addition, some available results from the FATHERINO
low temperature and a large temperature gradient between these experiments carried out by CEA could be used for qualitative code
regions is observed. In case of the dynamic procedure (Fig. 6(b)) the validation (Braillard, 2008). In case of FATHERINO’s “Skin of Fluid
scalar mixing is more efficient. The temperature gradient between Mockup” (1 mm brass wall) special attention is paid to the cou-
the two fluid streams is not so steep as for the classical Smagorin- pling with the heat conduction in the T-junction wall. Infrared
sky model. The mean velocity field is similar to the results obtained thermography images are available for the mean and fluctuating
with the classical Smagorinsky model, the only difference is that the temperature on the outer pipe surface.
two vortices in the upper section of the pipe are less pronounced Fig. 7 depicts contours of the mean temperature on the outer
but still present. pipe wall for both models. By comparing the results obtained with
The findings of this section revealed a major difference in the the classical Smagorinsky model (Fig. 7(a)) with the dynamic pro-
results of the two SGS models. The classical Smagorinsky model cedure (Fig. 7(b)) major differences in the distribution of the mean

Fig. 5. Instantaneous snapshots of the velocity and the temperature field in a cross-section located 120 mm after the junction.
 S. Kuhn et al. / Nuclear Engineering and Design 240 (2010) 1548–1557 1553

Fig. 6. Mean velocity field and mean temperature in a cross-section located 120 mm after the junction.

temperature are found, which can be explained by the prediction of due to this separated region predicted by the classical Smagorin-
the separated flow region using the classical Smagorinsky model. sky model. This finding is verified for the mean temperature on the
As already observed in the cross-section at x = 120 mm the mixing outer surface where the imprints of the flow reversal are seen in
of the two fluid streams with different temperature is less efficient the amount of fluid with lower temperature accumulating in the

Fig. 7. Contours of the mean temperature on the outer pipe wall for the two SGS Fig. 8. Contours of the temperature fluctuations on the outer pipe wall for the two
modeling approaches. SGS modeling approaches.
1554 S. Kuhn et al. / Nuclear Engineering and Design 240 (2010) 1548–1557

Fig. 9. Contours of the mean temperature on the pipe surface. Top: numerical solu-
tion. Bottom: infrared thermography. Fig. 10. Contours of the RMS of the fluctuating temperature on the pipe surface.
Top: numerical solution. Bottom: infrared thermography.

lower pipe section after the junction. The dynamic procedure pre-
dicts a more efficient mixing at the location and downstream of the found closer to the intersection. In addition, the numerical results
junction. predict another location of increased RMS at the upstream round
Due to these mechanisms we also observed differences in the edge of the T-junction. This is also observed in the infrared ther-
location of the maximum temperature fluctuations, which are mography, but not as pronounced as for the LES. As it was shown
depicted as contours on the outer pipe wall in Fig. 8. For the clas- by Coste et al. (2008) the RMS of temperature in this region is sen-
sical Smagorinsky model (Fig. 8(a)) we observe the maximum of sitive to the inlet boundary conditions. The maximum value of the
temperature fluctuations downstream of the junction in a region RMS is found to be 3.5 in the simulations and 3.0 in the experi-
close to the upper pipe section and further downstream in a region ment, thus also a good quantitative agreement in the magnitude
near the center of the pipe. In the zone where the flow reversal is of the fluctuating temperature is obtained. Therefore we conclude
predicted no temperature fluctuations occur since the fluid of lower that LES with the dynamic SGS model is able to capture both mean
temperature is entrapped in the vicinity of the wall. The dynamic and fluctuating temperatures in the mixing region and in the pipe
procedure (Fig. 8(b)) predicts the maximum of temperature fluctu- wall for the mixing in T-junctions.
ations in the junction in the region where the two fluid streams of
different temperature meet. 5. Influence of wall thickness
We validate the numerical results with data available from the
FATHERINO experiment of CEA (Braillard, 2008). Fig. 9 depicts con- 5.1. Mean and fluctuating temperature fields on the outer wall
tours of the mean temperature field on the pipe surface obtained
by LES using the dynamic SGS model (top) and by infrared ther- In this section we address the influence of wall thickness on
mography (bottom). As it can be observed from this figure a good the resulting temperature fields. Therefore we computed the thin
agreement between the measured mean temperature and the (brass, 1 mm) and the thick wall (steel, 5 mm) case of the T-junction.
numerical result is obtained. Fig. 10 depicts contours of the RMS Fig. 11 depicts the comparison of the mean temperature on the
of the fluctuating pipe wall temperature. The numerical results outer pipe wall for both cases. For the thin wall (left in Fig. 11) a
(top in Fig. 10) agree well with the experiment (bottom in Fig. 10). sharp distinction between the two streams of different temperature
The region of maximum RMS downstream of the intersection pre- is observed. In the whole simulation domain the upper part of the
dicted by LES corresponds to the region of maximum RMS in the pipe surface remains at the main branch inlet temperature of 356 K.
experiment. However, a slight difference in the exact location of the A temperature gradient is only found in the mixing region after the
maximum RMS value is observed, for the LES results this location is intersection of the two branches. Thus for the thin wall the mixing

Fig. 11. Contours of the mean temperature on the outer pipe wall.
 S. Kuhn et al. / Nuclear Engineering and Design 240 (2010) 1548–1557 1555

Fig. 12. Contours of the temperature fluctuations on the outer pipe wall.

process inside the pipe is reflected in the temperature contours on Fig. 12 shows the RMS values of the fluctuating temperature on
the pipe surface. This is different for the thick wall case (right in the pipe wall for both cases. The locations of maximum RMS differ
Fig. 11). Due to conduction in the solid material the mean temper- between the thin (left in Fig. 12) and the thick wall (right in Fig. 12)
ature is smeared over the pipe surface in the region of mixing after case. For the latter, the region of maximum fluctuations is shifted
the branch intersection. Thus the mean quantity is affected by the in positive z-direction, which is also caused by conduction in the
pipe wall thickness and measurements of the pipe surface temper-
ature yield no information about the mixing processes inside the
junction for thick walls.

Fig. 13. Fluctuating temperatures on the pipe inner and outer wall 120 mm down-
stream of the junction. Fig. 14. Amplitude spectra of the temperature fluctuations for the thin wall case.
1556 S. Kuhn et al. / Nuclear Engineering and Design 240 (2010) 1548–1557

solid wall. In addition, the magnitude of the obtained RMS values fluctuations in the wall. Therefore we plot the amplitude spectra of
is different. The damping effect of the thicker wall is seen in values the fluctuations in the fluid close to the wall (i.e. the center of the
of the temperature RMS that are an order of magnitude lower than first computational cell adjacent to the wall) as well as at the inner
in the case of the thin brass wall. and outer surfaces of the wall in top of Fig. 14 for the thin case and
To visualize this damping effect further we plot the fluctuating top of Fig. 15 for the thick wall. The data for these plots are extracted
temperature on the inner and outer pipe wall on the perimeter in the cross-section 120 mm downstream of the junction. For both
of the pipe in a cross-section located 120 mm downstream of the cases it is observed that the amplitude of the temperature fluctu-
junction. Fig. 13(a) shows the RMS profiles for the thin wall and ations in the fluid close to the wall is larger than inside the wall
Fig. 13(b) for the thick wall. The distribution and the magnitude of which is due to the presence of the thermal boundary layer. In the
the RMS values on the inner pipe surface are similar for the two thin case the influence of the wall thickness itself is negligible, since
different wall thicknesses. However, for both cases the damping the amplitude spectra for the inner and outer surface collapse and
effect of the wall is observed in the lower RMS values on the outer the shape of the spectra in the wall follows the shape of the spectra
wall. This effect is more pronounced for the thick wall. in the fluid. The influence of the thick wall is observed in the cor-
responding spectra, the amplitude of the spectra on the outer pipe
5.2. Amplitude spectra of the temperature fluctuations surface is lower compared to the inner surface and it is completely
smoothed out, i.e. the fluctuations are damped throughout the pipe
In this section we address the amplitude spectra of the tem- wall.
perature fluctuations in the solid pipe material. In addition to To address the influence of the wall thickness on the temper-
the location of the maximum temperature fluctuations their fre- ature fluctuations in more details, we plot the amplitude spectra
quency and amplitude is also of interest for the analysis of thermal of the temperature in 6 points throughout the wall (P1–P6) in the
fatigue. bottom of Fig. 14 for the thin case and in the bottom of Fig. 15
As first step we investigate the coupling between the tempera- for the thick wall. These data are also extracted in the cross-section
ture fluctuations in the fluid close to the wall and the temperature 120 mm downstream of the junction and P1 denotes a point located
at the inner surface and P6 at the outer wall, respectively. For the
thin case the spectra for the different points throughout the wall
collapse, i.e. no influence of the wall thickness on the heat transfer
can be extracted. For the thick wall the damping effect is clearly
visible in the amplitude and shape of the temperature spectra. The
amplitude decreases with increasing radial position and the spectra
gradually level out.

6. Conclusions

In this study we numerically investigated the mixing of hot

and cold water streams in a T-junction with a temperature dif-
ference of 75 ◦ C of the pre-mixed fluid streams and a Reynolds
number in the mixed stream of 2.7 × 105 . We studied two differ-
ent designs of the T-junction: (i) a thin wall of 1 mm made of brass
and (ii) a thick wall of 5 mm made of steel. These two different
designs are aimed at identifying the influence of the wall thickness
on conjugate heat transfer. We applied a LES with different SGS
models, i.e. the classical Smagorinsky model and the dynamic pro-
cedure. The main difference between these SGS models lies in the
near wall treatment. The classical Smagorinsky model needs wall-
damping functions to correct the turbulent viscosity in the vicinity
of walls. In addition, the unresolved turbulent heat flux is modeled
by the simple gradient diffusion hypothesis which aligns it with
the gradient of the mean temperature field. In the dynamic pro-
cedure the need of wall-damping functions is eliminated and the
dynamic procedure to obtain CS ensures correct near wall behav-
ior of the turbulent viscosity. The subgrid scale turbulent Prandtl
number is obtained by applying the dynamic procedure to the
SGS flux.
This difference in the SGS models is seen in the results of the
simulations. The T-junction of the FATHERINO setup features round
edges at the connection of the main and the side branch. As a con-
sequence there is no sharp edge which clearly defines the point of
flow separation, this has to be treated by the near-wall behavior
of the model. This prediction of flow separation and reversal is the
main difference between the two models. The results of the clas-
sical Smagorinsky model show a separated flow region between
the streamwise coordinates x ≥ 0.04 mm and x ≤ 0.065 mm in the
vicinity of the lower pipe wall. This is not the case when using
the dynamic procedure. This difference in the computed momen-
Fig. 15. Amplitude spectra of the temperature fluctuations for the thick wall case. tum field affects the scalar mixing efficiency downstream of the
 S. Kuhn et al. / Nuclear Engineering and Design 240 (2010) 1548–1557 1557

junction, which results in different predictions of the mean tem- References

perature distribution and the location of maximum temperature
fluctuations of the two models. In comparison with available exper- Braillard, O., 2008. FATHERINO experiment for thermal fatigue studies in a mix-
ing tee and nozzle—support of TH and mechanical analysis of thermal load
imental data the results of the dynamic procedure agree well with determination. In: Älvkarleby, Sweden, 5, June, 2008.
measurements. Thus we conclude that the quality of the numeri- Coste, P., Quemere, P., Roubin, P., Emonot, P., Tanaka, M., Kamide, H., 2008. Large eddy
cal results depends heavily on the proper choice of the SGS model, simulation of highly fluctuational temperature and velocity fields observed in a
mixing-tee experiment. Nucl. Technol. 164, 76–88.
which needs to be adapted to the geometrical constraints of the Germano, M., Piomeli, U., Moin, P., Cabot, W.H., 1991. A dynamic subgrid-scale eddy
particular test case. viscosity model. Phys. Fluids A 3, 1760–1765.
The influence of the wall on the temperature fields is well rep- Hu, L.-W., Kazimi, M.S., 2003. Large eddy simulation of water coolant thermal strip-
ing in a mixing tee junction. In: Proceedings of the 10th International Topical
resented in the numerical solutions. For the thin wall case the Meeting on Nuclear Reactor Thermal Hydraulics (NURETH-10), Seoul, Korea.
values of the RMS of the temperature fluctuations are reduced by Hu, L.-W., Kazimi, M.S., 2004. A simplified method for determination of high cycle
a factor of 3 between the inner and outer pipe surface, however fatigue caused by thermal striping. In: Proceedings of the ICAPP’04, Pittsburgh,
USA.
the distribution of maximum and minimum values on the pipe
Hu, L.-W., Kazimi, M.S., 2006. LES benchmark study of high cycle temperature fluc-
perimeter is identical. The amplitude spectra of the temperature tuations caused by thermal striping in a mixing tee. Int. J. Heat Fluid Flow 27,
fluctuations inside the thin wall at different radial locations are sim- 54–64.
ilar. For the thick wall case the wall influence is more expressed. Jungclaus, D., 1998. Common IPSN/GRS safety assessment of primary coolant
unisolable leak incidents caused by stress cycling. In: NEA/CSNI Specialists’
The magnitude of the RMS of the temperature fluctuations reduces Meeting on Experience with Thermal Fatigue in LWR Piping Caused by Mixing
by a factor of 10 between inner and outer pipe surface and the and Stratification, OECD Nuclear Energy Agency, Paris, France, 7–12 June.
distribution on the pipe perimeter is smeared out. This effect is Lee, J.I., Saha, P., Kazimi, M.S., 2004. A parametric study of high cycle thermal fatigue
caused by thermal striping. In: Proceedings of the ICAPP’04, Pittsburgh, USA.
also seen in the shape of the amplitude spectra throughout the Lee, J.I., Hu, L.-W., Saha, P., Kazimi, M.S., 2009. Numerical analysis of thermal striping
wall. induced high cycle thermal fatigue in a mixing tee. Nucl. Eng. Des. 239, 833–839.
The comparison with this single experiment in the present study Muramatsu, T., Ninikata, H., 1996. Development of thermohydraulic computer pro-
grams for thermal striping phenomena. Nucl. Technol. 113, 54–72.
shows the potential use of LES to predict the location of maximum Niceno, B., Smith, B., Prasser, H.-M., 2008. Computational fluid dynamics (CFD) as
temperature fluctuations in the solid wall material induced by the a tool for prediction of thermal fatigue in T-junctions. In: International Topical
mixing processes inside the pipe. In addition their frequency and Meeting on Safety of Nuclear Installations (TOPSAFE), Dubrovnik, Croatia.
Piomelli, U., Moin, P., Ferziger, J.H., 1988. Model consistency in the large eddy sim-
amplitude can also be studied. The generated results can be passed ulation of turbulent channel flows. Phys. Fluids 31, 1884–1891.
on to a structural analysis tool for the subsequent prediction of Wagner, W., Kruse, A., 1998. Properties of water and steam. The industrial standard
thermal stresses and the corresponding life span of the considered IAPWS-IF97 for thermodynamic properties and supplementary equations for
other properties. Springer-Verlag, Berlin.
T-junctions.