Professional Documents
Culture Documents
Ou
Ou
GUO, SHUJIE. Numerical Study of T-junction Thermal Mixing using PHASTA. (Under the
direction of Dr. Igor A. Bolotnov).
Thermal fatigue is an important topic in nuclear engineering and design. It affects the
reactor safety and lifetime of plant components. The T-junction thermal mixing is typically used
to evaluate the code performance as it is a good example where the thermal fatigue can be
observed. Thus, a lot of effort was offered to predict flow behaviors in T-junction thermal mixing
by both experiment and simulation methods. CFD codes achieved good validation based on
Vattenfall T-junction thermal mixing experiment data. The objective of the present study is to
validate PHASTA heat transfer capabilities with implicit LES approach for T-junction thermal
mixing problem. To control computational cost adiabatic case was tested based on both under and
highly resolved mesh, and the mixing phenomenon was presented. Simulations show the same
streamwise velocity distribution features with the experiment and the results agree well with the
experimental data. Considering the effect of heat transfer between the fluid and pipe wall,
conjugate heat transfer (CHT) code was implemented in PHASTA and CHT case was tested with
both Plexiglas and stainless steel pipe walls. Error analysis was developed to quantify the error of
the simulation and experimental results. Based on the experiment uncertainty and fluctuation of
simulation results, different factors were tested and compared. The most appropriate one was
chosen to evaluate the error of simulations. The results prove that with the growth of computational
costs, the factor fluctuates around a value which could measure heat transfer capabilities of
PHASTA code.
© Copyright 2021 by Shujie Guo
by
Shujie Guo
Nuclear Engineering
APPROVED BY:
_______________________________ _______________________________
Dr. I. A. Bolotnov Dr. N. Dinh
Committee Chair
_______________________________
Dr. M. A. Diaconeasa
BIOGRAPHY
The author graduated from Xi’an Jiaotong University (XJTU) with a Bachelor of
worked on a research project with professors in developing experiments focusing on studying the
leakage due to a micro crack in the steam generator heat transfer tubes. While studying in XJTU,
she was awarded the Siyuan Scholarship and attended a summer exchange program in University
of Wisconsin, Madison, where she decided to apply for a graduate program in the United States.
She joined the Department of Nuclear Engineering at North Carolina State University in 2019
focusing on thermal hydraulics, especially the field of computational fluid dynamics under the
supervision of Dr. Igor A. Bolotnov. As hobbies, she enjoys reading, travelling and music.
ii
ACKNOWLEDGMENTS
I would like to thank my family. I would like to give my ultimate thanks to my parents who
give me the strength and confidence when I was facing the difficulties and support me persistently
whatever decision I made. My family is always the most important energy source to me especially
during this unusual period with the threat of Covid-19. They are the ones who give me abundant
love and unconditional support throughout my life and help me to realize my dream as an engineer.
I would like to express my deepest gratitude to Dr. Igor A. Bolotnov. Without his help I
would not be able to complete my master study. He offers much patience and conscientiousness
to train me for research. I would like to thank Dr. Dalin Zhang, who lead me into the nuclear
engineering and her passion and perseverance to the nuclear engineering affect me during my
undergraduate study. I would also like to thank the professors giving me help during my master
study, Dr. Jason Hou, Dr. Djamel Kaoumi, Dr. Pramod Subbareddy, Dr. Mihai A. Diaconeasa, Dr.
I would like to thank Joy Fan, who helped me to spend the hard time of culture shock after
I came to the United States and offered me much help when I was facing the difficulties. I would
like to thank all of my friends in the department of nuclear engineering. Thanks to the other
groupmates who gave much advice for my research. Saini Nadish and Matthew Zimmer gave me
awesome tutorial and introduction of the PHASTA code and helped me to solve numerious issues.
Thanks to Xinya Tao, one of my best friends, whose company is one of the most important energy
iii
TABLE OF CONTENTS
iv
5.2 CHT case results analysis .................................................................................................... 39
REFERENCES .............................................................................................................................. 58
APPENDICES ............................................................................................................................... 62
Appendix A ............................................................................................................................... 63
Appendix B ................................................................................................................................ 64
v
LIST OF TABLES
Table 3. The energy conservation validation results of PreCase 1 and PreCase 2 ........................ 12
Table 4. The fluid properties and wall properties for PreCase 3 and PreCase 4 ........................... 14
Table 5. Heat flux calculation based on the Fig. 7 results of PreCase 3&4 .................................. 15
Table 7. Mesh resolutions of Case 1&2 along the streamwise direction ...................................... 19
Table 8. Friction factor and Corresponding ∆𝑦 + Value of sections in Case 1&2 ....................... 20
Table 10. Mesh resolutions of Case 4&5 along the streamwise direction .................................... 24
Table 11. Friction factor and Corresponding ∆𝑦 + Value of sections in Case 4&5 ..................... 25
Table 13. Mesh resolutions of Case 7 along the main inlet streamwise (z axis) and branch inlet
Table 14. Friction factor and Corresponding ∆𝑦 + Value of sections in Case 7 .......................... 30
Table 17. The areas of the analytical, artificial and overlapped for three comparisons ................ 52
vi
LIST OF FIGURES
Fig. 1. The schematic of the Vattenfall T-junction thermal mixing experiment [2] ....................... 3
Fig. 2. The schematic of WALTON T-junction thermal mixing experiment [11] .......................... 6
Fig. 3. The mesh design of (a) side wall (b) cross-section in laminar pipe case ........................... 11
Fig. 4. The temperature distributions at (a) vertical section and (b) cross section in PreCase 1;
Fig. 6. Temperature distributions at vertical section of (a) PreCase 3 and (b) PreCase 4 at
Fig. 7. Temperature profiles at the outlet cross section of (a) PreCase 3 and (b) PreCase 4 at
Fig. 8. Vattenfall T-junction experimental test section reproduced for simulations shown with
Fig. 9. The boundary layer mesh design of the under-resolved case (a) the overall mesh of the
cross-section within the most refined region; (b) zoom into the boundary layer region ... 18
Fig. 10. The mesh transition near the outlet of the under-resolved case ....................................... 19
Fig. 12. Model size of shortened inlets with mesh (unit: mm) ...................................................... 22
Fig. 13. Velocity distribution of Case 1 at (a) z/D=1.6 and (b) z/D=4.6; of Case 3 at (c) z/D=
Fig. 14. Streamwise velocity profiles comparison of Case 1 and Case 3 at z/D=1.6 along (a) x
axis and (b) y axis; at z/D=4.6 along (c) x axis and (d) y axis ......................................... 23
vii
Fig. 15. The boundary layer mesh design of the well-resolved case (a) overall mesh of cross-
section within the most refined region; (b) zoom into the boundary layer region ........... 24
Fig. 16. Mesh transition near the outlet of the well-resolved case ................................................ 24
Fig. 17. Case 6 with mesh generated for CHT code validation ..................................................... 26
Fig. 18. The transient (a) temperature and (b) velocity magnitude distributions of Case 6 .......... 27
Fig. 19. The temperature profile in the midplane at (a) y=0.15; (b) z=-0.4 .................................. 27
Fig. 20. the schematic of the innovative model with gap applied with periodic condition ........... 28
Fig. 21. CHT parallel plate case with mesh (unit: mm) ................................................................ 29
Fig. 22. The model schematic at the junction upper corners ......................................................... 30
Fig. 23. (a) Unsteady-state and (b) Steady-state examples of three-window method used to
Fig. 24. Average (a) temperature and (b) streamwise velocity fields of Case 1; average (c)
Fig. 25. Average streamwise velocity fields for (a) Case 1 (b) Case 2 at location 𝑧/𝐷 = 1.6
Fig. 26. Average streamwise velocity profiles of Case 1 and Case 2 at location (a) z/D=1.6 and
(b) z/D=4.6 along x axis; at cross-section location (c) z/D=1.6 and (d) z/D=4.6 along y
axis ................................................................................................................................... 35
Fig. 27. The streamwise velocity fields of (a) Case 4 and (b) Case 5 at location z/D=1.6 and (c)
Fig. 28. Average streamwise velocity profiles of Case 4 and Case 5 at location (a) z/D=1.6 and
(b) z/D=4.6 along x axis; at cross-section location (c) z/D=1.6 and (d) z/D=4.6 along y
axis ................................................................................................................................... 37
viii
Fig. 29. Temperature comparisons between the results of Case 5 and experiment data at (a)
top, (b) bottom, (c) left and (d) right walls along streamwise directions ......................... 39
Fig. 30. Average (a) temperature and (b) velocity fields of vertical section in Case 6 ................. 40
Fig. 31. Average temperature fields of cross sections at (a) 𝑧/𝐷 = 1.6 and (b) 𝑧/𝐷 = 4.6 ..... 41
Fig. 32. Average temperature profiles at locations 𝑧/𝐷 = 1.6 and 𝑧/𝐷 = 4.6 along y axis .... 41
Fig. 33. At location 𝑧/𝐷 = 2, the temperature file over time of the (a) top, (c) bottom, (e)
left and (g)right wall and the corresponding temperature spectrum (b)(d)(f)(h) ............. 42
Fig. 34. At location 𝑧/𝐷 = 4, the temperature file over time of the (a) top, (c) bottom, (e)
left and (g)right wall and the corresponding temperature spectrum (b)(d)(f)(h) ............. 43
Fig. 35. Average (a) streamwise velocity and (b) temperature fields of vertical section in
Case 7 ............................................................................................................................... 45
Fig. 36. Average (a) streamwise velocity and (b) temperature fields of cross section at
Fig. 37. Average (a) streamwise velocity and (b) temperature fields of cross section at
Fig. 38. Average streamwise velocity profiles of Case 7 along y axis at location (a) z/D = 1.6
Fig. 39. Average temperature profiles at locations 𝑧/𝐷 = 1.6 and 𝑧/𝐷 = 4.6 along y axis .... 47
Fig. 40. Average temperature profile of top wall along the streamwise direction in Case 7 ........ 48
Fig. 41. At location 𝑧/𝐷 = 2, the temperature profiles over time of the (a) top and (c) bottom
Fig. 42. At location 𝑧/𝐷 = 4, the temperature profiles over time of the (a) top and (c) bottom
ix
Fig. 43. The comparison of (a) 10 samples, (b) 100 samples, and (c) 1000 samples of artificial
Fig. 45. The values of three indexes with the computational cost ................................................ 53
x
CHAPTER 1. Introduction
plants and chemical processing plants. One important function of T-junction is to mix flows with
different physical properties including thermal mixing. However, at the center of the junction
component where hot and cold fluids mix with each other, significant temperature fluctuations
could be introduced and may lead to thermal fatigue on the pipe structures. With respect to nuclear
plant, thermal fatigue, which may cause pipe crack and even leak, is always a non-negligible
concern to system safety and could affect the ageing and lifetime of plant components. Several
accidents occurred because of T-junction thermal fatigue, including the accident of Civaux-1 in
France in 1998 mainly due to the thermal fatigue occurring in T-junction of Residual Heat Remove
For monitoring and detecting thermal fatigue are limited by sensor response time,
prediction with computational fluid dynamic (CFD) codes obtains great interests to control the
risk. The objective of this work is to validate PHASTA heat transfer capabilities based on the
Vattenfall T-junction thermal mixing experiment [2]. The work includes the validation to the
single-phase turbulent thermal mixing under adiabatic condition and the study of conjugate heat
Chapter 3 gives an introduction to the simulation tool, PHASTA code and describes the
validation effort to study the heat transfer capabilities of the PHASTA code with and without CHT
under simple geometry and laminar flow. The two tests prove the potential of heat transfer
1
Chapter 4 describes the simulation setup, which contains: (i) adiabatic case setup and mesh
designs and (ii) CHT case setup and mesh designs. The detailed model descriptions and the
Chapter 5 presents the results of both diabatic and CHT cases. It includes result discussion
on the phenomenon of T-junction thermal mixing simulation, the comparison between the
experiment and simulation results and the analysis of conjugate heat transfer results, and error
analysis method.
2
CHAPTER 2. Literature Review
of Vattenfall Research and Development [2] in Sweden and groups of reliable experiment results
were generated. The experiment facility shown in Fig. 1 with the cold water importing by the
horizontal pipe and the hot water by the vertical pipe is manufactured by Plexiglas. The
temperatures and volumetric flow rates of two inlets are shown in the Table 1. Along the
downstream, Particle Imaging Velocimetry (PIV) was used to measure the fluid velocities at the
location 0.224m and 0.644m, and thermal couples were set for the temperature measurements at
the top, left, bottom and right sides. Various CFD codes such as ANSYS and Nek5000 are
Fig. 1. The schematic of the Vattenfall T-junction thermal mixing experiment [2]
3
A great deal of benchmark research work [3][4][5][6] was done based on Vattenfall
experiments and achieved good validation with the data, which indicates that Vattenfall
Since Reynolds-averaged Navier–Stokes (RANS) equations are not designed to capture the
fluctuation features of the velocity and temperature profiles, Large Eddy Simulation (LES) is the
most common method utilized for these simulations. Kim and Jeong [3] developed Vattenfall T-
junction simulation using LES based on the dynamic Vreman model [7]. The simulation results
show that Reynolds shear stress, turbulence intensity and mean velocity profiles are well predicted.
The T-junction thermal mixing was also tested with ANSYS code. Höhne [4] performed a
prediction with ANSYS CFX-12 in 2014. Two-equation turbulence model, Shear Stress Transport
Model [8], was utilized for the steady-state calculation and LES was for the transient calculation.
The velocity field and mean temperature agree well with the experiment results, and the
fluctuations and frequencies are close to the experimental data, while the simulation failed to
provide a good prediction for a realistic mixing. In 2019, Kang et al. [5] performed the T-junction
thermal mixing based on ANSYS CFX-16 with detached eddy simulation (DES) model. The
simulation results of velocity agree well with experimental data while at the separation region, a
large temperature difference was observed between the simulation and experiment results. In 2017,
Obabko et al. [6] tested Nek 5000 with the T-junction thermal mixing using LES model and
More effort was given to achieve good prediction for the temperature fluctuation of the
pipe wall and conjugate heat transfer (CHT) is considered. CHT capability involves a fully coupled
resolution of the heat conduction phenomenon in the solid region and heat convection between the
solid and the fluid, which is referred as coupled fluid-solid heat transfer. The Vattenfall
4
experiments [2] give the conclusion that the wall temperature is close to ambient (≈ 15℃) due to
the low conductivity of the pipe wall material, Plexiglas. Besides, other materials such as stainless
steel are also tested in the simulation to validate the effect of materials with different physical
properties.
In 2011, Hannink and Blom [9] coupled CFD and FEM analysis to perform the T-junction
thermal mixing phenomenon for both fluid and solid behavior. The test was based on the Vattenfall
T-junction thermal mixing experiment but the temperature difference between two inlets was
increased from 17K to 80K and AISI 304L stainless steel was chosen to replace the original pipe
material, Plexiglas. Temperature results from the CFD are used in FEM analysis as thermal load,
which induces a linear relationship between the stress and temperature. The stress intensity
fluctuations were provided by the measurements in the test setup and the temperature fluctuations
were obtained since it has linear relationship with the stress intensity fluctuations which is similar
Timperi [10] performed the temperature distributions of T-junction thermal mixing with
conjugate heat transfer LES. The Smagorinsky SGS model of commercial Star-CD version 4 code
in 2014 and the SIMPLE pressure-correction algorithm were used to solve the flow equations. The
stainless steel was assumed to be the wall material. Timperi compared the results of CHT case and
adiabatic case, and the comparison indicates that different wall boundary conditions could have
similar temperature fluctuations in the logarithmic layer and upward, while at the wall surface, the
WATLON experiment [11] also focus on the T-junction thermal mixing phenomenon and
the experiment setup shown in Fig. 2. Hot fluid goes through the horizontal main pipe and mixes
5
with the cold fluid from the vertical branch pipe. The differences between the WALTON and
Vattenfall experiment setup are not only the facility size but also the flow directions of the hot and
cold flow. In the Vattenfall experiment, the side flow is hot water down along the branch pipe,
while for WALTON experiment, cold fluid becomes the side flow and goes up in the branch pipe.
Kamaya and Nakamura [12] simulated the thermal mixing phenomenon and fluctuations in wall
by ANSYS CFX 10 using the detached eddy simulation (DES). Besides, relatively large thermal
stress along the symmetry line was discovered and was supposed to result from membrane
constraint. Based on the experimental data, Cenk Evrim and Laurien [13] completed the simulation
on the T-junction thermal mixing prediction using LES method with the OpenFOAM code using
finite volume method. The simulation results demonstrate that the code has good predictive
capabilities for the mean temperature and velocity distributions, as well as the spectral peak of
Several validation papers of the conjugate heat transfer were also performed based on the
other experimental data. In 2010, Kuhn et al. [14] demonstrated the capability of LES to predict
6
the T-junction turbulent thermal mixing with dynamic Smagorinsky model for the subgrid scales
(SGS) has good agreement with experimental data. The effect of the wall thickness on the
temperature fluctuations is also revealed as a damping effect in the radial direction. Jayaraju et al.
[15] performed a large eddy simulation with WALE sub-grid-scale model based on the
experimental setup of Andersson et al. in 2006 [16]. With the solver STAR-CCM+, the mean
velocity and temperature profiles were captured while the RMS heat-flux in the stream-wise
direction was under-estimated. In 2015, Selvam et al. [17] presented the investigations on the
temperature fluctuations of T-junction turbulent thermal mixing using LES method with ANSYS
CFX 14.0, based on the experimental data [18] of Fluid–Structure Interaction test facility at the
Materials Testing Institute, University of Stuttgart. The flow directions in the experiment are
different from the Vattenfall T-junction experiment, but the simulation predicts the mean
temperatures well and has good agreement with experimental data, while the root mean square
temperatures have small differences. The energy and amplitude of temperature fluctuations are
also obtained by the numerical simulation. In 2016, Selvam et al. [19] completed a further analysis
on the numerical performance of T-junction thermal mixing behavior. Three inflow temperature
differences were tested within LES method, and stratified flow condition was observed in all three
cases due to the incomplete thermal mixing. The energy of temperature fluctuations in all three
cases is further confirmed that is mainly in the frequency range of 0.1–2 Hz, and the amplitude of
7
CHAPTER 3. Numerical Approach
PHASTA code [20] is a “Parallel Hierarchic Adaptive Stabilized Transient Analysis” CFD-
solver for both incompressible and compressible Navier-Stokes equations. The application of
PHASTA code on the single-phase flow for both laminar and turbulent conditions has been
developed and the results proved its capability to capture and fully resolve the detailed flow
features. In 1999, Jansen [20] showed that the stabilized finite element method using LES is an
effective approach to simulate the complex flow such as turbulent flow. Later in 2001, Whiting
and Jansen [21] demonstrated the potential benefits using the stabilized finite element method
applied to the simple geometries and laminar flow, and also indicated the positive expectation to
the outcome of turbulent flow using both LES and RANS. In 2006, Tejada-Martínez and Jansen
[22] further validated a new dynamic Smagorinsky subgrid-scale model based on the code using
LES method. The parameter-free model is not necessary to compute the filter width ratio and could
improve the computation of LES with irregularly connected topologies. In 2009, Trofimova et al.
[23] performed turbulent single-phase channel flows with PHASTA code successfully and
demonstrated that this stabilized finite element method using at direct numerical simulation
resolutions could fully resolve the turbulent channel flow. In 2017, Li et al. [24] performed the
heat transfer study of turbulent flow under low Prandtl number, which provide the foundation of
The PHASTA code is extended to the two-phase flow simulation with level set method for
interface tracking. In 2013, Bolotnov [25] obtained Reynolds stress distribution for single-phase
and low void fraction two-phase using PHASTA and isotropic turbulence models were analyzed
based on the simulation results. The work indicates that direct numerical simulation (DNS) has the
8
capability to evaluate the applicability of various turbulent models (such as turbulent viscosity
based or Reynolds stress based) by evaluating the flow anisotropy. Mishra and Bolotnov [26]
studied the effect of surface roughness density on the turbulent flow using PHASTA with DNS
level simulation. Cases with different wall roughness were analyzed and provided the basis for the
boundary conditions of the two-phase flow model. In 2018, Fang et al. [27] summarized the
achievement of PHASTA code on the two-phase flow simulation using high performance
computing, including subcooled boiling model and bubble coalescence. The results of single-phase
and bubbly flow in the PWR subchannel were also analyzed. Zimmer and Bolotnov [28] developed
the transition study on the slug-to-churn vertical two-phase flow in 2019. The results are consistent
with the experimental data and theory, which indicate the capabilities of PHASTA code for not
only bubbly flow but also more complex two-phase flow conditions.
Certain capabilities of PHASTA code have been validated and verified, however, the
complex pipe geometry such as T-junction haven’t been tested, and the heat transfer between
different materials neither yet. But based on the previous work, PHASTA code has good potential
Incompressible Navier-Stokes (INS) equations are used in the simulations. The spatial and
temporal discretization of INS used in the PHASTA are described by Whiting and Jansen [21],
Continuity:
𝛻∙𝒖=0 (1)
Momentum:
𝜕𝑢! (2)
𝜌 + 𝜌∇ ∙ (𝑢! 𝒖) = −𝑃,! + 𝜏#!,! + 𝑓!
𝜕𝑡
9
where u is velocity, 𝑢! is the 𝑖 $% component of velocity, 𝜌 is density, 𝑃,! is pressure gradient, 𝜏#!
is viscous stress tensor, and 𝑓! is the 𝑖 $% component of body force. The viscous stress tensor is
given as:
where 𝜇 is the dynamic viscosity of the fluid, and 𝑆!# is the strain rate tensor.
The objective of the current work is to validate PHASTA heat transfer capabilities within
T-junction thermal mixing phenomenon. Thus, heat transfer simulations without CHT are
introduced first and compared with the experiment to test the heat transfer capabilities of the
PHASTA code. Second step is to evaluate CHT capabilities. Before introducing the simulation
setup and results, some validation effort is required to test the PHASTA code in the simple
geometry such as pipe or parallel plates geometries with laminar flow conditions. The effort could
help to understand the potential and possible limitation of the PHASTA code in the following
simulations.
To validate the heat transfer capability of PHASTA code on thermal boundary conditions
(Appendix A), a laminar pipe case was set up. The diameter of the pipe is 0.013 m, and the length
is 0.04 m. The mesh designs of the side wall and cross-section are presented in Fig. 3. The
thicknesses of two boundary layers are 0.3mm and 0.4mm, and the surface mesh resolution is
0.5mm. The number of the elements is 316,160. The simulations took around 30 minutes to reach
10
(a) (b)
Fig. 3. The mesh design of (a) side wall (b) cross-section in laminar pipe case
The fluid properties applied are shown in Table 2. The parabolic velocity profile at the inlet
is set as 𝑢(𝑦, 𝑧) = 2 × (1 − (𝑦 & + 𝑧 & )⁄𝑅& ) and the averaged velocity is 1.0m/s. The
corresponding Reynolds number is 1000 which indicates the laminar flow. The laminar flow pipe
was tested with two thermal boundary conditions of constant temperature 3500K in PreCase 1, and
heat flux 1000W/m& in PreCase 2. The inlet temperature of both cases is set to be 500K.
The temperature distributions of PreCase 1 and PreCase 2 are shown in Fig. 4. The energy
where 𝑞** is the heat flux applied to the wall, A is the area of the side wall, 𝑚̇ is the mass flow rate,
ℎ+,!$ is the enthalpy of the outflow and ℎ!- is the enthalpy of the inflow. The equation (4) whose
results are shown in Table 3 is satisfied by both the cases, which give the conclusion that both the
11
(a) (b)
(c) (d)
Fig. 4. The temperature distributions at (a) vertical section and (b) cross section in PreCase 1; at
(c) vertical section and (d) cross section in PreCase 2
CHT capability is first implemented in the PHASTA code (Appendix B) by Nadish Saini
[29]. Since PHASTA could read the set surface to be solid wall and generate the distance of
element to the wall, the elements whose distance to the wall is smaller than the wall thickness
would be recognized as wall elements, and the velocities of these elements would be zero. Based
on his work I further modified code to recognize and apply the different properties for solid and
fluid. The parameters of the solid would be used for calculation when the solid region is read by
the code.
There are two principles to be satisfied in the CHT case at the vicinity of the interface
To have a preliminary validation on the CHT capability of PHASTA code, a parallel plate
case was created. The length of the plates (along x axis) is 0.1m and the width (along z axis) is
0.02m. The thickness of the plates is 0.001m and the distance between the plates is 0.02m. Both
the top and bottom wall are applied with boundary layers shown in Fig. 5, and the thickness of the
first boundary layer is 0.1mm and boundary layer growth factor is 1.2. The bulk resolution is 1mm.
The number of elements is 286,651. The simulations took around 16 hours to reach steady state
with 64 processors.
Periodic condition is set for the front and back surface. Parabolic velocity profile is applied
as 𝑢(𝑦) = 0.02 × (1 − 𝑦 & ⁄0.01& ) at the inlet and the mean velocity is 0.01667m/s. The
corresponding Reynolds number is 1126, thus the flow is laminar flow. The inlet temperature is
373.15K and heat flux 1000W/m& is applied at the top and bottom wall. The properties of 304L
stainless steel and Plexiglas are selected to be used in PreCase 3 and PreCase 4 show in Table 4,
respectively.
13
Table 4. The fluid properties and wall properties for PreCase 3 and PreCase 4
Wall
Fluid
PreCase 3 PreCase 4
Viscosity/ (Pa ∙ s) 2.8573 × 10(. ------------- -------------
Density/ (𝑘𝑔/𝑚' ) 965.201 7910.86 1200.0
Specific heat/(𝐽/𝑘𝑔 ∙ 𝐾) 41.8 510.0 1460.0
Thermal conductivity/ (𝑊/𝑚 ∙ 𝐾) 0.687221 15.5 0.190
The temperature distributions of the vertical section are shown in Fig. 6 and the temperature
profile of the cross section at the outlet in Fig. 7. It can be observed that at the interface between
the wall and fluid, the temperature profile is continued, which support that Equation (5) is satisfied.
𝑑𝑇
𝑞** = −𝑘 (7)
𝑑𝑦
(a)
(b)
Fig. 6. Temperature distributions at vertical section of (a) PreCase 3 and (b) PreCase 4 at
simulation time 47.1s
14
(a) (b)
Fig. 7. Temperature profiles at the outlet cross section of (a) PreCase 3 and (b) PreCase 4 at
simulation time 47.1s
The results for the heat flux of PreCase 3 and PreCase 4 are shown in Table 5. Currently
PHASTA code is more appropriate to be used in the CHT case whose solid material has the same
value of the 𝜌𝐶/ product with the fluid. For the future work, more effort will be given to expand
the CHT capability of PHASTA code to be applied to various solid material properties.
Table 5. Heat flux calculation based on the Fig. 7 results of PreCase 3&4
Heat flux/ (W/m& ) In the wall near wall (in the fluid)
PreCase 3 951.13 980.85
PreCase 4 1016.93 3113.64
15
CHAPTER 4. Simulation Setup
With the validation effort discussed in CHAPTER 3, the simulation cases are set to
demonstrate the heat transfer capability of PHASTA code for the T-junction thermal mixing. The
simulation cases contain the cases with the adiabatic thermal boundary which the heat flux through
the boundary is set to zero, and the cases with the CHT in which solid wall is built based the data
from the Vattenfall experiment and the heat transfer in the fluid is fully coupled with heat
Two different mesh designs are applied in adiabatic cases. The coarser mesh under-resolves
the flow and studied as the initial step of the T-junction thermal mixing simulations, and the
comparison between the results of two mesh designs can also help to understand the influence of
the mesh resolution on the result. The second mesh with refined resolution can get much more
accurate simulation data. The comparison between the simulation results and experimental data
CHT cases are tested with two models. Considering to fully solve the flow, the boundary
layers have to satisfy the LES requirement. However, limited by the current mesh generation
capability, the solid and fluid regions have to be separate to set the required boundary layer mesh
structure for the fluid and solid regions. Thus, a pipe case with coarse mesh and another case of
parallel plates (forming T-junction geometry) with refined mesh are set.
Adiabatic cases are test prior to the CHT cases. Heat conduction in the solid is not involved
in the adiabatic cases, and zero heat flux is applied as thermal boundary conditions at all wall
boundaries. Although adiabatic condition is not the physical condition of the Vattenfall experiment
[2], the heat loss from water is not considered as important due to the low heat conductivity of the
16
pipe material. Thus, based on the previous research [3][4][5][6], adiabatic condition can achieve
good agreement with the experiment. The objective of the adiabatic case is to validate the heat
Because of the large domain size and high Reynolds number, under-resolved cases were
tested prior to the refined case to have a basic understanding to the T-junction thermal mixing. To
compare with the experimental data, the model geometry shown in Fig. 8 is set to be the same as
the Vattenfall experiment facility size. The main inlet temperature is 292.15K and branch inlet
temperature is 309.15K. The uniform velocities are applied to the inlets, which for main inlet is
0.584 m/s and for branch inlet is 0.764 m/s, respectively. Due to single-fluid limitation, the fluid
properties of the outlet temperature which is approximately 27.5ºC shown in Table 6 are applied
to the case based on the material properties table provided by the Älvkarleby Laboratory [2]. Two
cases with different viscosities were tested to observe the influence of the viscosity on the flow
and also control the computational cost. Case 1 with a larger viscosity (0.003 Pa ∙ s) would be run
first to reach the steady state and Case 2 with the physical viscosity (0.0008693 Pa ∙ s) could
utilize the data of Case 1, which would save the simulation time cost to reach steady state.
Fig. 8. Vattenfall T-junction experimental test section reproduced for simulations shown with
computational mesh (unit: mm) [2]
17
Table 6. Fluid properties [2]
parameter Main inlet pipe Branch inlet pipe
Temperature /℃ 19 37
Diameter /m 0.14 0.1
'
Density /(kg/m ) 996.79
Heat capacity /(kJ/kg ∙ K) 4.18
Thermal conductivity /(W/m ∙ K) 0.6
Velocity /(m/s) 0.584 0.764
Case 1:Re (µ = 0.003 Pa ∙ s) 27165 25385
Case 2:Re (µ = 0.0008693 Pa ∙ s) 93751 87600
The boundary layer mesh is only applied in the region from inlet to 0.72m to prevent the
backflow. For the under-resolved cases, there are 10 layers of boundary mesh shown in Fig. 9. The
thickness of the first boundary layer mesh is 1.6 × 10(. m. The boundary layer mesh growth factor
(a) (b)
Fig. 9. The boundary layer mesh design of the under-resolved case (a) the overall mesh of the
cross-section within the most refined region; (b) zoom into the boundary layer region
18
To prevent the backflow and divergency, the outlet section is elongated from 0.72m up to
1.22m and applied with coarse mesh. The mesh transitions are also developed to reduce numerical
error induced by the difference between the bulk mesh and outlet mesh. The mesh of the transition
region is shown in Fig. 10 and the mesh resolution along the streamwise direction is presented in
Table 7. The total number of tetrahedral elements is 5,721,545 and the number of nodes is 995,851.
Fig. 10. The mesh transition near the outlet of the under-resolved case
In order to study the refinement of the mesh, ∆y 0 of both the bulk and boundary mesh were
𝜌𝑢1 ∆𝑦 (8)
∆𝑦 0 =
𝜇
where 𝜌 is the density of the fluid, μ is the fluid viscosity, y is the distance to the wall, and 𝑢1 is
friction velocity:
𝜏2
𝑢1 = h (9)
𝜌
19
where τ3 represents wall shear stress and given the pipe flow it could be obtained by:
1 2.51 1 𝜖 (11)
= −2.00 𝑙𝑜𝑔45 m + p
k𝑓 𝑅𝑒k𝑓 3.7 𝐷
where for the smooth pipe, the roughness ϵ = 0. Given the Reynolds number and friction factor
could be solved iteratively. The estimated friction factors and ∆y 0 at three sections for both Case
1 and Case 2 are shown in Table 8. The ∆y 0 values indicate that the flow in Case 1 is well-resolved
near the wall region while in the bulk region it’s under-resolved, and the flow in Case 2 is under-
Since the accurate velocity distributions of the simulation results are required to compare
with the experiment results, at the two downstream locations z/D=1.6 and z/D=4.6 (D is the
diameter of outlet section), 3073 virtual probes shown in Fig. 11 were put at either cross-section
to record flow data. The distance of the outset probes to the wall is 2.75 × 10() m which is less
than the thickness of first boundary layer, 1.6 × 10(. m, thus, the whole cross-section is covered
20
8.00E-02
6.00E-02
4.00E-02
2.00E-02
0.00E+00
-8.00E-02 -6.00E-02 -4.00E-02 -2.00E-02 0.00E+00 2.00E-02 4.00E-02 6.00E-02 8.00E-02
-2.00E-02
-4.00E-02
-6.00E-02
-8.00E-02
To get more accurate results and compare with the experimental data, well-resolved cases
were tested. Considering that the computational cost would be extremely large using the
experiment model geometry shown in Fig. 8, a model with shortened inlets was created and shown
in Fig. 12. The outlet region is also elongated to prevent the backflow. To guarantee that the
turbulent flow is fully developed in the shortened inlets, 1/7th power law is used as inlet velocity
profiles:
𝑢 𝑟 4/9
= s1 − u (12)
𝑈67, 𝑅
where U:;< is the maximum velocity and R is the radius of pipe. A validation case, Case 3, was
generated first to test the influence of the shortened pipe length with 1/7th power law on the
turbulent flow velocity development. Case 3 has the same mesh design with Case 1, including
boundary layers and bulk mesh resolutions shown in Table 9. The fluid properties and boundary
21
Fig. 12. Model size of shortened inlets with mesh (unit: mm)
After reaching the steady state, the velocity distributions of Case 3 were recorded by the
same group of virtual probes shown in Fig. 11 to be compared with the distributions of Case 1.
The visible velocity distributions at two locations, z/D=1.6 and z/D=4.6, of both cases are shown
in Fig. 13. The result comparison between Case 1 and Case 3 is shown in Fig. 14, which indicate
that with 1/7th power law, the turbulent flow in Case 3 is fully developed and the shortened inlet
Based on the results of the validation case, two well-resolved cases with different
viscosities were developed, which are similar to the under-resolved cases. Case 4 is built with a
larger viscosity (0.003 Pa ∙ s) and Case 5 with the physical viscosity (0.0008693 Pa ∙ s).
The boundary layer mesh is only applied in the region from inlet to 0.72m to prevent the
backflow. For the well-resolved cases, there are 10 layers of boundary mesh shown in Fig. 15. The
thickness of the first boundary layer mesh is 4.5 × 10() m. The boundary layer mesh growth factor
22
(a) (b)
(c) (d)
Fig. 13. Velocity distribution of Case 1 at (a) z/D=1.6 and (b) z/D=4.6; of Case 3 at (c) z/D=1.6
and (d) z/D=4.6
v z /v(bulk)
1 1
0.5 0.5
Case 1
Case 3
0 0
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
x/R y/R
(a) (b)
1.5 1.5
v z /v(bulk)
v z /v(bulk)
1 1
0.5 0.5
Case 1 Case 1
Case 3 Case 3
0 0
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
x/R y/R
(c) (d)
Fig. 14. Streamwise velocity profiles comparison of Case 1 and Case 3 at z/D=1.6 along (a) x
axis and (b) y axis; at z/D=4.6 along (c) x axis and (d) y axis
23
(a) (b)
Fig. 15. The boundary layer mesh design of the well-resolved case (a) overall mesh of cross-
section within the most refined region; (b) zoom into the boundary layer region
More refined mesh presented in Table 10 is used to fully resolve the flow. The mesh
transitions shown in Fig. 16 are also applied to reduce numerical error induced by the difference
between the bulk mesh and outlet mesh. The total number of tetrahedral elements is 12,704,554
Table 10. Mesh resolutions of Case 4&5 along the streamwise direction
Streamwise [-0.10, 0.30] [-0.61, 0.72] [0.72, 0.82] [0.82, 1.02] [1.02, 1.12] [1.12, 1.22]
location / m
Mesh 2.5E-03 5.0E-03 1.0E-02 2.0E-02 4.0E-02 8.0E-02
resolution
Fig. 16. Mesh transition near the outlet of the well-resolved case
24
In order to study the refinement of the mesh, ∆y 0 of both the bulk and boundary mesh were
calculated based on Equation (8)(9)(10) and (11). The estimated friction factors and ∆y 0 at three
sections for both Case 4 and Case 5 are shown in Table 11. The ∆y 0 values indicate that the flow
in Case 4 is well-resolved, and the flow in Case 5 is a little under-resolved in the bulk region.
Table 11. Friction factor and Corresponding ∆𝑦 0 Value of sections in Case 4&5
Section Main inlet Branch inlet Outlet
Mean velocity / (m/s) 0.584 0.764 0.974
Friction factor 0.02404 0.02443 0.02136
Case 4 Wall ∆y 0 0.48 0.63 0.75
0
Bulk ∆y 26.59 35.07 41.81
Friction factor 0.01823 0.01850 0.01642
Case 5 Wall ∆y 0 1.44 1.90 2.28
0
Bulk ∆y 79.92 105.32 126.50
Virtual probes shown in Fig. 11 were put at either cross-section to record flow data. The
distance of the outset probes to the wall is 2.75 × 10() m which is less than the thickness of first
boundary layer, 4.5 × 10() m, thus, the whole cross-section is covered by the virtual probes.
Conjugate heat transfer (CHT) capabilities in the T-junction thermal mixing simulation are
then tested. Wall will be generated, the heat conduction in the solid and the heat convection
between the flow and solid are coupled and induced into simulations. The objective of CHT
simulations is not only to validate the CHT capabilities of PHASTA code, but also to demonstrate
the influence of the CHT procedure to the T-junction thermal mixing results.
To validate whether CHT code could work for T-junction pipe, Case 6 shown in Fig. 17 with
coarse mesh was performed. According to the T-junction thermal mixing experiment [2], a wall
25
region with the thickness of 10mm is generated for the domain except the region near the outlet.
Ambient temperature (≈ 15℃) is set for the wall boundary conditions and the fluid properties are
the same as the Case 2 and Case 5. The physical properties of 304 stainless steel under room
temperature is used as the parameters of the wall shown in Table 12. The wall region is covered
by the 4 uniform boundary layers whose thickness is 2.5mm and the bulk resolution is 10mm. The
Fig. 17. Case 6 with mesh generated for CHT code validation
The transient temperature and velocity magnitude distributions are shown in Fig. 18. The
velocity distribution in the wall region is kept being zero, which indicates that CHT code works
well on the velocity. With respect to temperature distribution, the temperature profiles in the
midplane at location y=0.15 and z=-0.4 are presented in Fig. 19. It is observed that the left side of
the branch inlet has lower temperature than the right side and the top side of the main inlet has
higher temperature than the bottom. The phenomenon is due to the heat conduction in the wall,
which transfer the heat from the hot leg to the cold one.
26
(a) (b)
Fig. 18. The transient (a) temperature and (b) velocity magnitude distributions of Case 6
(a) (b)
Fig. 19. The temperature profile in the midplane at (a) y=0.15; (b) z=-0.4
To fully solve both the wall and fluid domain, it is necessary to set the boundary layer meshes
for wall and fluid appropriately. The thickness of the wall is 0.01mm while the well-resolved
boundary layers for the flow based on LES requirement need the thickness of first boundary layer
mesh in the fluid domain to be approximate 4.5 × 10() m. It indicates that the boundary layer
meshes for the two regions have to be separated while the current mesh generation executable file
only enables one design of boundary layers for each surface. Therefore, a gap is inserted to separate
the regions as shown in Fig. 20. The periodic condition is applied on the two faces of the gap to
27
connect the temperature across the gap. Based on this innovative model design, parallel plates are
solid
gap periodic
fluid condition
Fig. 20. the schematic of the innovative model with gap applied with periodic condition
The model schematic with mesh of Case 7 shown in Fig. 21 was created. The depth of the
model is 50mm. Outlet section is also elongated to prevent backflow. Considering the result
comparison, the Reynolds number of the case is set to be the same as the experiment. Since the
hydraulic diameter of the parallel plates is 2D where D is the distance between two parallel plates,
the distances between parallel plates of main inlet and branch inlet are set to be half of diameter of
The boundary layers are only applied in the region from inlet to 0.72m to prevent the
backflow. The thickness of the first boundary layer mesh is 4.5 × 10() m. The boundary layer
mesh growth factor is 1.4 and the total thickness is 0.00314m. Bulk mesh resolution are shown in
28
Fig. 21. CHT parallel plate case with mesh (unit: mm)
Table 13. Mesh resolutions of Case 7 along the main inlet streamwise (z axis) and branch inlet
streamwise (y axis) directions
z axis location / m [-0.10, 0.30] [-0.61, 0.72] [0.72, 0.82] [0.82, 1.02] [1.02, 1.22]
y axis location / m [-0.045, 0.055] [0.055, 0.25] [0.25, 0.328] ---------- ----------
Mesh resolution 2.5E-03 5.0E-03 1.0E-02 2.0E-02 4.0E-02
To separate the solid and fluid region, five thin gaps which has the thickness of 0.01mm
are inserted into the locations of interfaces. At the upper corners the two conjunction parts are
utilized to connect the solid and fluid regions as shown in Fig. 22 to keep one integral part. Periodic
condition is set for each couple of surfaces in a gap to transfer the heat across the gaps to get
continuous temperature profiles. The gaps with periodic conditions are only able to be created for
parallel plate case since the areas of faces applied with periodic conditions is strictly required to
be the same.
In order to study the refinement of the mesh, ∆y 0 of both the bulk and boundary mesh were
calculated based on Equations (8) - (11). The estimated friction factors and ∆y 0 at three sections
for Case 7 are shown in Table 14. The ∆y 0 values indicate that the flow in Case 7 is a little under-
29
Fig. 22. The model schematic at the junction upper corners
30
CHAPTER 5. Results and Discussion
Summary tables, Table 15 and Table 16, for both the adiabatic cases and CHT cases are
shown as described in CHAPTER 4. Case 1 and Case 2 are tested prior to the Case 4 and Case 5
to have an initial step for the T-junction thermal mixing simulation. To reduce the computational
cost, model with shortened inlet lengths is developed. Case 3 is compared with Case 1 to
demonstrate that the shortened inlet lengths have little influence on the downstream results. Then
Case 4 and Case 5 are studied to validate that the PHASTA code has the good heat transfer
capability. CHT cases are then studies to compare the effect of the heat conduction in the solid to
the T-junction thermal mixing. Case 6 is present to have a verification of the performance of CHT.
Then Case 7 is developed to evaluate the effect of CHT module to the T-junction thermal mixing
behavior.
Since simulation data is recorded by the virtual probes, described in section 4.1.1, a three-
window comparison method is used to decide whether the simulations the statistical steady-state.
31
The method compares the data of the adjacent time period to observe if the simulation data will
converge. Examples of the unsteady-state and steady-state simulations are shown in the Fig. 23.
1.6
1.4
1.2
0.8
vz
0.6
0.4
15500-15900
0.2
15900-16300
16300-16700
0
-1 -0.5 0 0.5 1
(a)
x/R
1.4
1.2
0.8
0.6
0.4
3.920-4.917 s
0.2 4.917-5.907 s
5.907-6.826 s
0
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
(b)
Fig. 23. (a) Unsteady-state and (b) Steady-state examples of three-window method used to
determine the simulation state
32
5.1.1 Results of the under-resolved cases
Once T-junction thermal mixing simulation was fully developed and reached statistical
steady state, the mean results were obtained by averaging the data in 1.76 seconds simulation time.
The simulation cost around 60 wall clock hours for either case (while running on 128 computing
cores). Fig. 24 shows the average temperature streamwise velocity fields for Case 1 and Case 2.
There is a kidney-shape recirculation region near corner due to the effect of the hot side flow as
The averaged streamwise velocity fields for Case 1 and Case 2 of cross sections 𝑧/𝐷 = 1.6
and 𝑧/𝐷 = 4.6 are shown in Fig. 25. At location 𝑧/𝐷 = 1.6, the velocity fields show that the side
flow has significant effect that near the end of the recirculation zone the flow is pushed down to
the bottom so that the velocity near the top wall is much less than the one near the bottom, while
at location 𝑧/𝐷 = 4.6, the effect is much slighter, and the dissipation of the effect can be observed.
(a) (b)
(c) (d)
Fig. 24. Average (a) temperature and (b) streamwise velocity fields of Case 1; average (c)
temperature and (b) streamwise velocity distributions of Case 2.
33
(a) (b)
(c) (d)
Fig. 25. Average streamwise velocity fields for (a) Case 1 (b) Case 2 at location 𝑧/𝐷 = 1.6 and
for (c) Case 1 (d) Case 2 at location 𝑧/𝐷 = 4.6.
The recirculation zone causes the two-peak characteristic of the streamwise velocity
distribution along x axis and uneven distribution along y axis as the streamwise velocity profiles
shown in Fig. 26. Since the cross-section at 𝑧/𝐷 = 1.6 is closer to the sharp edge and in the
recirculation zone, the streamwise velocity profile along x axis has obvious two-peak feature,
while at location 𝑧/𝐷 = 4.6 the streamwise velocity profile is much flatter due to the dissipation
of the side flow effect. Besides, the streamwise velocity profiles along y axis also show the effect
of the side flow. In Fig. 26, the results of Case 1 and Case 2 are compared with experimental data.
At location 𝑧/𝐷 = 1.6, the velocity profiles of Case 2 show the same feature as the experimental
34
data, but there is reasonable error since the flow is under resolved. However, at location 𝑧/𝐷 =
4.6 especially for the velocity profile along x axis, the results of both cases miss the two-peak
feature.
1.6
1.5
1.4
1.2
1 1
v z /v(bulk)
v z /v(bulk)
0.8
0.6
0.5
0.4 Case 1
Case 1
Case 2 Case 2
0.2 Experiment
Experiment
0 0
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
x/R x/R
(a) (b)
1 1
v z /v(bulk)
v z /v(bulk)
0.5 0.5
Case 1
Case 2
Experiment
0 0
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
y/R y/R
(c) (d)
Fig. 26. Average streamwise velocity profiles of Case 1 and Case 2 at location (a) z/D=1.6 and
(b) z/D=4.6 along x axis; at cross-section location (c) z/D=1.6 and (d) z/D=4.6 along y axis
Based on the results of the under-resolved cases, the capability of predicting the flow
behaviors in T-junction thermal mixing were demonstrated and the simulation results show good
agreement with the experimental data given the computational cost limitation. Thus, further
validation based on the well resolved cases are completed based on the results.
35
5.1.2 Well-resolved case
Since the result comparison between Case 3 and Case 1 discussed in section 4.1.2 has
shown that the shortened pipe has negligible influence on the results, the simulations of Case 4
and Case 5 (simulation setup is described in section 4.1.2) were performed. The simulation data is
recorded by the virtual probes once the cases reach the steady state. The simulation of Case 4 costs
approximately 100 wall clock hours and of Case5 costs 80 wall clock hours (while running on 192
computing cores).
The streamwise velocity fields of cross sections at locations 𝑧/𝐷 = 1.6 and 𝑧/𝐷 = 4.6
of both Case 4 and Case 5 are shown in Fig. 27. Compared with the velocity fields shown in Fig.
25, it is observed that the velocity difference on the same cross section is greater since the flow is
better resolved. Two-peak feature at location 𝑧/𝐷 = 4.6 is also presented in Fig. 27 (c) (d).
(a) (b)
(c) (d)
Fig. 27. The streamwise velocity fields of (a) Case 4 and (b) Case 5 at location z/D=1.6 and (c)
Case 4 and (d) Case 5 at location z/D=4.6.
36
The streamwise velocity profiles of Case 4 and Case 5 at locations 𝑧/𝐷 = 1.6 and 𝑧/𝐷 =
4.6 along x and y axis are shown in Fig. 28. Compared with the experimental data, the results of
Case 5 show the same two-peak feature of the streamwise velocity along x axis at the two locations.
The streamwise velocity profiles of Case 5 along both x and y axes agree well with the
experimental data and demonstrate the good capability of the PHASTA code to predict the
1.6
1.5
1.4
1.2
1 1
v z /v(bulk)
v z /v(bulk)
0.8
0.6
0.5
0.4 Case 4
Case 4 Case 5
0.2 Case 5 Experiment
Experiment
0 0
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
x/R x/R
(a) (b)
1.6
1.5 Case 4
Case 5 1.4
Experiment
1.2
1
v z /v(bulk)
1
v z /v(bulk)
0.8
0.6
0.5
0.4 Case 4
Case 5
0.2 Experiment
0 0
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
x/R y/R
(c) (d)
Fig. 28. Average streamwise velocity profiles of Case 4 and Case 5 at location (a) z/D=1.6 and
(b) z/D=4.6 along x axis; at cross-section location (c) z/D=1.6 and (d) z/D=4.6 along y axis
37
The temperature result comparisons between the simulations and experiments are shown
𝑇 − 𝑇>?@A
𝑇∗ = (13)
𝑇%?$ − 𝑇>?@A
where T represents the temperature of fluid, 𝑇%?$ is the temperature of the hot inflow and 𝑇>?@A is
of the cold inflow. In the Vattenfall T-junction thermal mixing experiment, the temperatures are
obtained by thermocouples in the flow located 1 mm from the wall of four directions, top, right,
left and bottom. In Fig. 29, it can be observed that at the locations close to the center, the
normalized temperatures are well predicted and have good agreements with the experimental data,
while at the locations that close to the outlet, the temperatures at the bottom, left and right walls
are underpredicted and the temperature at the top wall is overpredicted. It indicates that the thermal
Therefore, based on the comparison of both the velocity and temperature profiles between
the simulations and experiments, in which velocity profiles have good agreements with the
experimental data and capture the features shown in the experiments, temperature profiles show
the small difference at the locations near the outlet, an assumption is given that introducing CHT
into the simulation may help to enhance the heat transfer since with the CHT the heat conduction
38
1
Case 5 0.5 Case 5
Experiment Experiment
0.9
0.4
0.8
0.3
T*
T*
0.7
0.2
0.6
0.1
0.5
0
0.4
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
z/D z/D
(a) (b)
0.4 0.4
0.3 0.3
T*
T*
0.2 0.2
0.1 0.1
0 0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
z/D z/D
(c) (d)
Fig. 29. Temperature comparisons between the results of Case 5 and experiment data at (a) top,
(b) bottom, (c) left and (d) right walls along streamwise directions
The objective of CHT tests is to demonstrate the effect of the heat conduction procedure in
the solid to the temperature fluctuations and the result accuracy of the simulations. Fourier
transform approach is used to get the amplitude and frequency of the temperatures recorded by
probes.
The results of Case 6 are analyzed to test the CHT results based on the under-resolved flow.
It took 6 wall clock hours (while running on 64 computing cores) for Case 6 to reach the statistical
39
steady state and collect data. Fig. 30 shows the average temperature and velocity fields of vertical
section in Case 6. The recirculation zone presented in Case 6 indicates that the simulation with
CHT has the expected feature of the temperature and velocity profiles observed in the experiment.
The zero-velocity region generated for the solid demonstrated that CHT code has successfully set
the zero velocity for the solid element. The average temperature fields of cross-sections at locations
𝑧/𝐷 = 1.6 and 𝑧/𝐷 = 4.6 are shown in Fig. 31 and the corresponding temperature profiles
along x and y axis are presented in Fig. 32. The temperature profiles presented that for the solid
element, there is linear relationship between the temperature and the distance to wall, which
(a)
(b)
Fig. 30. Average (a) temperature and (b) velocity fields of vertical section in Case 6
40
(a) (b)
Fig. 31. Average temperature fields of cross sections at (a) 𝑧/𝐷 = 1.6 and (b) 𝑧/𝐷 = 4.6
306
z=1.6D
304 z=4.6D
302
300
temperature/K
298
296
294
292
290
288
-0.08 -0.06 -0.04 -0.02 0 0.02 0.04 0.06 0.08
y/m
Fig. 32. Average temperature profiles at locations 𝑧/𝐷 = 1.6 and 𝑧/𝐷 = 4.6 along y axis
Since the objective to study the flow behavior of T-junction thermal mixing is to simulate
the thermal stress which may cause thermal fatigue, Fourier transform is used to get the
temperature spectrum with frequency and amplitude of the temperature profiles based on the
simulation time recorded. the temperature over simulation time and the corresponding temperature
spectrum of top, bottom, left and right wall at location 𝑧/𝐷 = 2 and 𝑧/𝐷 = 4 are shown in Fig.
41
10 0
-5
10
Amplitude
10 -10
0 1 2 3
10 10 10 10
Frequency(Hz)
(a) (b)
0
10
-5
10
Amplitude
-10
10
0 1 2 3
10 10 10 10
Frequency(Hz)
(c) (d)
0
10
-5
10
Amplitude
-10
10
0 1 2 3
10 10 10 10
Frequency(Hz)
(e) (f)
0
10
10 -5
Amplitude
-10
10
10 0 10 1 10 2 10 3
Frequency(Hz)
(g) (h)
Fig. 33. At location 𝑧/𝐷 = 2, the temperature file over time of the (a) top, (c) bottom, (e) left
and (g)right wall and the corresponding temperature spectrum (b)(d)(f)(h)
42
10 0
10 -5
Amplitude
10 -10
10 0 10 1 10 2 10 3
Frequency(Hz)
(a) (b)
0
10
-5
10
Amplitude
-10
10
0 1 2 3
10 10 10 10
Frequency(Hz)
(c) (d)
0
10
10 -5
Amplitude
10 -10
10 0 10 1 10 2 10 3
Frequency(Hz)
(e) (f)
0
10
10 -5
Amplitude
10 -10
10 0 10 1 10 2 10 3
Frequency(Hz)
(g) (h)
Fig. 34. At location 𝑧/𝐷 = 4, the temperature file over time of the (a) top, (c) bottom, (e) left
and (g)right wall and the corresponding temperature spectrum (b)(d)(f)(h)
43
The Fourier transforms of the experiment temperature spectrum show a distinct peak at
around 3.5 Hz at the left and right walls of both two locations, while the temperature spectrums of
the top and bottom don’t show the peak [2]. The temperature spectrums in Fig. 33 and Fig. 34
shows that for the left and right wall at 𝑧/𝐷 = 2 and the right wall at 𝑧/𝐷 = 4 has the peak near
3-10 Hz and the difference between the simulation results and experimental data is acceptable due
The results of CHT parallel plate case, Case 7, are studied based on the effort of Case 6.
The Reynolds number of Case 7 is set to be the same as the Case 6 and Case 7, while due to the
transformation of the geometry, the phenomenon shown in the Case 7 would has the difference
with the experiment. Once Case 7 reached the statistical steady state, the results are recorded. It
took approximately 90 wall clock hours (while running on 128 computing cores).
The averaged velocity and temperature fields of vertical section are shown in Fig. 35 and
of cross section at 𝑧/𝐷 = 1.6 and 𝑧/𝐷 = 4.6 are shown in Fig. 36 and Fig. 37, respectively. It
can be observed that the recirculation zone still exists in the parallel plate case, while the shape of
the recirculation zone is different from the one in the pipe as shown in Fig. 24 and Fig. 25. The
explanation is that in the pipe, the recirculation zone is influenced by the effect of the existence of
the wall at the right and left side, while in parallel plate, since the back and front surface is set to
44
(a)
(b)
Fig. 35. Average (a) streamwise velocity and (b) temperature fields of vertical section in Case 7
(a) (b)
Fig. 36. Average (a) streamwise velocity and (b) temperature fields of cross section at 𝑧/𝐷 = 1.6
in Case 7
45
(a) (b)
Fig. 37. Average (a) streamwise velocity and (b) temperature fields of cross section at 𝑧/𝐷 =
4.6 in Case 7
Therefore, only the comparison of velocity profiles between the results of Case 7 and
experimental data along y axis are completed, and the comparison of temperature focus on the top
and bottom wall. The streamwise velocity profiles along y axis at location 𝑧/𝐷 = 1.6 and 𝑧/𝐷 =
4.6 are presented in Fig. 38, and the comparison between the experimental data is also shown. In
Fig. 38, the streamwise velocity profiles at both locations have the same feature with the
experimental data, while the difference between values is obvious due to the transformed shape of
the recirculation zone. The average temperature profiles at locations 𝑧/𝐷 = 1.6 and 𝑧/𝐷 = 4.6
along y axis are shown in Fig. 39 and the linear ship can be observed in the solid. In the top wall,
the temperature profiles at locations 𝑧/𝐷 = 1.6 are higher than the profiles at 𝑧/𝐷 = 4.6 since
heat is conducted from the hot inlet to the outlet, while in the bottom wall, the temperature profiles
at locations 𝑧/𝐷 = 1.6 are lower than the profiles at 𝑧/𝐷 = 4.6 since heat is conducted from the
outlet to the cold inlet. The simulation results perform the good capability of the PHASTA code
46
1.6 1.6
Case 7 Case 7
1.4 experiment 1.4
experiment
1.2 1.2
1 1
v z /v(bulk)
v z /v(bulk)
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
y/R y/R
(a) (b)
Fig. 38. Average streamwise velocity profiles of Case 7 along y axis at location (a) z/D = 1.6 and
(b) z/D = 4.6
302
300
298
temperature/K
296
294
292
z=1.6D
290 z=4.6D
The temperature profile along the streamwise direction at the top wall is presented in Fig.
40 and has the good agreement with the experimental data. Fourier transform is used to get the
temperature spectrum with frequency and amplitude of the temperature profiles based on the
simulation time. The temperature profiles over simulation time and the corresponding temperature
spectrum of top and bottom wall at location 𝑧/𝐷 = 2 and 𝑧/𝐷 = 4 are shown in Fig. 41 and Fig.
42, respectively. The temperature spectrums in Fig. 41 and Fig. 42 have no obvious peak, which
47
agrees with the experiment result analysis that the temperature spectrums of the point located on
1
Case 7
Experiment
0.9
T* 0.8
0.7
0.6
0.5
0.4
0.1 0.2 0.3 0.4 0.5 0.6 0.7
z/m
Fig. 40. Average temperature profile of top wall along the streamwise direction in Case 7
303
10 0
302
301
temperature/K
300
Amplitude
299
10 -5
298
297
296
295 10 -10
0 200 400 600 800 1000 1200 1400
10 0 10 1 10 2 10 3
time step
Frequency(Hz)
(a) (b)
306
10 0
304
302
temperature/K
Amplitude
300
298 10 -5
296
294
292 10 -10
0 200 400 600 800 1000 1200 1400
10 0 10 1 10 2 10 3
time step
Frequency(Hz)
(c) (d)
Fig. 41. At location 𝑧/𝐷 = 2, the temperature profiles over time of the (a) top and (c) bottom
wall and the corresponding temperature spectrums (b)(d)
48
303
0
10
302
301
temperature/K
Amplitude
300
10 -5
299
298
297 -10
10
0 200 400 600 800 1000 1200 1400 0 1 2 3
10 10 10 10
time step
Frequency(Hz)
(a) (b)
302
10 0
301
300
temperature/K
Amplitude
299
298
10 -5
297
296
295 10 -10
0 200 400 600 800 1000 1200 1400
10 0 10 1 10 2 10 3
time step
Frequency(Hz)
(c) (d)
Fig. 42. At location 𝑧/𝐷 = 4, the temperature profiles over time of the (a) top and (c) bottom
wall and the corresponding temperature spectrums (b)(d)
5.3 Error analysis
In the comparisons mentioned in section 5.1 and 5.2, the profiles of experimental data and
simulation results are presented to observe the agreement between two set of data. However, since
in the experiment, the averaged results are usually companied with the uncertainty induced by the
measurement. To quantify the error between the experimental data and simulation result, an error
Since the experimental data is given with the uncertainty, a range of convinced
experimental data can be obtained. Thus, to quantify the error between the experimental data and
simulation result, simulation results could also be given as a range of convinced data to overlap
with the experimental data, and overlapped area with the areas of simulation and experiment results
49
can represent the agreement of the simulation. Since the experiment uncertainty area is obtained
by the averaged experiment results and the uncertainty, to get the simulation uncertainty area,
A test was developed to demonstrate the effectiveness of the method. Considering the
laminal flow in a pipe, which has the mean velocity of 2 m/s and the radius of the pipe is 1 m, the
analytical velocity profile is given as 𝑢 = 2 × (1 − 𝑟 & ). To imitate the error analysis for the results
of experiments and simulations, the analytical profile is regarded as the result of “experiment”,
and the uncertainty is set as 5% of the velocity value. To imitate the simulation results, a group of
artificial results is generated. The artificial results are randomly produced by standard normal
distribution at each point along the diameter of the pipe based on the analytical result. The
Different number of samples of the artificial results are generate and the comparison between
the artificial and analytical results is shown in Fig. 43. Since with more samples, the averaged
result generated by the standard normal distribution is closer to the analytical result, and the
The areas of the analytical, artificial and overlapped for all three comparisons are shown in
Table 17. It could be demonstrated that when the number of samples increases, the overlapped
area is closer to the analytical area, which indicates this error analysis method indeed can obtain
and represent the error of the simulation. Results of Case 5 is chosen as a sample to present the
method as shown in Fig. 44. The uncertainty of the simulation depends on the three-window
described in section 5.1. The maximum and minimum value of the three-window results is
50
2.5
1.5
velocity
1
0.5
0 analytical result
artificial result
-0.5
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
r/R
(a)
2.5
1.5
velocity
0.5
0 analytical result
artificial result
-0.5
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
r/R
(b)
2.5
1.5
velocity
0.5
0 analytical result
artificial result
-0.5
-1 -0.5 0 0.5 1
r/R
(c)
Fig. 43. The comparison of (a) 10 samples, (b) 100 samples, and (c) 1000 samples of artificial
results and analytical results
51
Table 17. The areas of the analytical, artificial and overlapped for three comparisons
The number of samples 10 100 1000
Analytical area 0.2665 0.2665 0.2665
Artificial area 0.3940 0.4115 0.4127
Overlapped area 0.2338 0.2610 0.2653
1.6
1.4
1.2
1
v z /v(bulk)
0.8
0.6
0.4
A measurement index representing the error of based on the experiment area 𝐴+ , simulation
area 𝐴B and overlapped area 𝐴? should be obtained to quantity the quality of the simulation. Three
indexes are tested, including 𝐴? /𝐴+ , 𝐴? /𝐴B , and (𝐴B − 𝐴? )/𝐴+ . The relation between value of
three indexes with the computational cost are presented in Fig. 45. The computational cost is
represented by the time steps used to get the results. With larger computational cost, the simulation
results should be closer to the experiment data, and the overlapped area should increase, and non-
overlapped simulation area should decrease. Based the relationship shown in Fig. 45, (𝐴B −
𝐴? )/𝐴+ is the most appropriate index to quantity the simulation error since it shows the clear
52
feature that with the increasing of computational cost, after the Case 5 reaches the steady state, the
value of (𝐴B − 𝐴? )/𝐴+ will be stable and fluctuate around 0.26, which could be used to represent
Overlapped/Experiment
Overlapped/Simulation
2.5 unover(Sim)/Experiment
2
Agreement
1.5
0.5
0
0 1000 2000 3000 4000 5000 6000 7000
Computational Cost
Fig. 45. The values of three indexes with the computational cost
It can be observed that in Fig. 45, value of (𝐴B − 𝐴? )/𝐴+ starts to fluctuate when time step
is larger than 3000, the corresponding simulation time of approximately 1 second and 1.5 flow
through time, which represent when after reaching the steady-state the simulation time recorded
for data is larger than it the simulation results reach statistical stable. Based on it, the method can
be used for other practical simulations to select the simulation time required for data recording,
which is that the value of (𝐴B − 𝐴? )/𝐴+ could be calculated and plotted over simulation time and
the minimum simulation time required for data analysis is obtained once the value of (𝐴B −
53
CHAPTER 6. Conclusions and Future work
6.1 Conclusions
A study focusing on the heat transfer capability of PHASTA code based on the Vattenfall
T-junction thermal mixing experiment is conducted. The thermal boundary and CHT application
were tested before the T-junction thermal mixing simulations. The issue of the heat flux boundary
calculation was fixed and the tests with both heat flux and constant temperature boundary
conditions performed with good agreement with the analytical results. The energy conservation is
verified. CHT is more appropriate to be used in the case which has the same value of the 𝜌𝐶/
The simulations were performed for both adiabatic and CHT cases. Adiabatic cases
concentrate on the heat transfer process within the fluid and apply zero heat flux for the boundary
of the fluid, including two different mesh designs. Considering the large size of the model, the
mesh design which has larger ∆y 0 was tested as the initial step of the T-junction thermal mixing
simulations with two different viscosities aiming at studying the effect of the viscosity to the
simulation result. The simulation with physical viscosity could also be developed based on the one
with larger viscosity to save the computational cost. The results of Case 1 and Case 2 have shown
that the velocity profiles have the same features with the experimental data, while due to the under-
resolved flow the difference between the simulation results and experimental data is acceptable.
The well-resolved cases developed based on the results of Case 1 and Case 2. To reduce
the computational cost, the lengths of T-junction inlets are reduced. Case 3 was developed to study
whether the shortened inlet lengths have observable influence on the downstream results and the
comparison between Case 1 and Case 3 shows negligible difference. Then Case 4 and Case 5 with
different viscosities of the refined mesh design which satisfies the LES requirements are performed.
54
The simulation results of Case 5 have good agreement of mean velocity profiles with the
experiment results, and for the most temperature values, the simulation also shows the good
agreement. However, at the location close to the outlet, the temperatures at the bottom, left and
right walls are underpredicted and the temperature at the top wall is overpredicted.
Thus, the next step is to include the CHT capability to evaluate if it can help to get more
accurate temperature prediction in the simulations. Two cases are set for the CHT simulation, a
pipe case with coarse mesh, Case 6, and a parallel plate case with refined mesh, Case 7. Case 6
was studied as an initial step for T-junction thermal mixing simulation with conjugate heat transfer.
The results of temperature profiles of Case 6 present that the heat conduction in the solid works
well. However, limited by the current mesh generation capability, the model had to be transformed
to parallel plates to have different boundary layer mesh designs for the solid and fluid region. Gaps
were inserted to separate the wall and flow and periodic conditions were applied to connect the
temperatures and velocities of two regions, so the velocity and temperature are solved to be
continuous even with the gap. The velocity profiles show the general feature of the experiment
velocity profiles but have differences with the experiment data which is acceptable since the
geometry is not exactly the same. The temperature spectra at the top and bottom wall show the
To quantity the error of the simulation, since the experiment results are always obtained
with uncertainty, an error analysis method was introduced. The simulation fluctuations are
regarded as the uncertainty of simulation, and in current case, since the three-window method is
used, the maximum and minimum values obtained from the three-window method is recognized
as the simulation fluctuations. Thus, both the experiment and simulation results are shown with
the statistical average results and uncertainty. The overlapped area between the experiment and
55
simulation results can be obtained, and three indexes based on experiment area 𝐴+ , simulation area
𝐴B and overlapped area 𝐴? are tested over the computational cost. (𝐴B − 𝐴? )/𝐴+ is chosen to be
the most appropriate index to quantity the heat transfer capability of the PHASTA code.
The application range of CHT module should be expanded as the future work. Since the
energy conservation issue is observed when the solid has different value of the 𝜌𝐶/ product, there
is an apparent inconsistency in the code formulation when dealing with energy equation. Thus,
when different properties of materials are applied, the heat transfer through the interface is not
calculated correctly. Detailed study of the PHASTA code is required to address this issue in the
future work.
Currently limited by the boundary layer mesh generation capability described in section
4.2.2, mesh in the wall should be relative coarse while the boundary layer mesh is required to be
refined enough to satisfy the LES requirement, the pipe was transformed to the parallel plates.
Although the Reynolds number is set to be the same as the experiment and periodic boundary is
set to connect the two regions to transfer the heat, the geometry transformation has influence on
the results especially for the velocity profile. Mesh design should be improved or consider using
other mesh generation tool which could be adapted for the PHASTA code to generate separate
Further study on the temperature fluctuations in the T-junction thermal mixing will be
developed. The work performed is based on the flow ratio applied in the experiment, while in the
benchmark report, it is mentioned that with different flow ration the peak of the temperature
56
spectrum would also change. Thus, the temperature fluctuations related to the different flow ratio
To study the thermal stress caused by the temperature fluctuations, Finite Element Method
such as ANSYS and ABUQUS can be utilized to get the influence of the temperature on the
lifetime of the T-junction. A relationship between the temperature fluctuation and the
corresponding thermal stress will be discovered to transfer the results of the computational fluid
57
REFERENCES
influence of surface condition and weld local geometry," Welding in the World, 46(1-2),
23-27 (2002).
Validation of Nek5000 for T-junction, Matis, SIBERIA, and Max Experiments,” Nuclear
fatigue due to turbulent mixing,” Nuclear Engineering and Design, 241(3), 681-687 (2011).
10. A. TIMPERI, “Conjugate heat transfer LES of thermal mixing in a T-junction,” Nuclear
58
11. H. KAMIDE, M. IGARASHI, S. KAWASHIMA, N. KIMURA, K. HAYASHI, “Study on
mixing behavior in a tee piping and numerical analyses for evaluation of thermal striping,”
12. M. KAMAYA, A. NAKAMURA, “Thermal stress analysis for fatigue damage evaluation
(2020).
conjugate heat transfer in T-junctions,” Nuclear Engineering and Design, 240(6), 1548-
1557 (2010).
Large Eddy Simulation for thermal fatigue in a T-junction,” Nuclear Engineering and
Vattenfall Research and Development AB Model tests Report U 06: 66, 68 (2006).
mixing of fluids in a T-junction with conjugate heat transfer,” Nuclear Engineering and
59
19. P. K. SELVAM, R. KULENOVIC, E. LAURIEN, "Experimental and numerical analyses
junction," International Journal of Heat and Fluid Flow, 61, 323-342 (2016).
20. K. JANSEN, “A stabilized finite element method for computing turbulence,” Computer
numerical simulation of turbulent channel flows using a stabilized finite element method,”
24. M. LI, J. FENG, I. A. BOLOTNOV, “Heat Transfer Study in Turbulence Flow with Respect
26. A.V. MISHRA and I. A. BOLOTNOV, “DNS of turbulent flow with hemispherical wall
60
28. M. D. ZIMMER, I. A. BOLOTNOV, “Slug-to-churn vertical two-phase flow regime
http://www.lib.ncsu.edu/resolver/1840.20/37328
reference to the transition region between the smooth and rough pipe laws. (includes plates),”
61
APPENDICES
62
Appendix A
This appendix describes the modified codes about heat transfer boundary calculated in
PHASTA.
E3B.F: This is the code file where the values of boundary nodes are calculated. In the
subroutine e3bSclr thermal boundary conditions are calculated for the energy equation. Before the
correction, the heat flux boundary condition cannot lead to the correct results since the value flux
is not solved correctly. The modification is made and currently PHASTA has the ability to develop
if (ibb.eq.6) then
do n = 1, nshlb
nodlcl = lnode(n)
flux(iel) = flux(iel)
+ shape(iel,nodlcl)*BCB(iel,n,ibb)*nsurf
/(datmat(1,1,1)*datmat(1,3,1))
enddo
endif
63
Appendix B
This appendix contains the code parts related to the CHT modifications implemented in
CHTBLOCK.F: In this subroutine edited by Saini [29], the elements that located in the
solid part are recognized based on its distance of the element to the wall. Once an element is judged
do i=1,numnp
if(d2wal(i,1) .le. cht_dist)then
if(ibits(iBC(i),3,3) .eq. 0)then
iBC(i) = iBC(i) + 56
BC(i,3:5) = 0.0
endif
endif
enddo
PROCES:F: In this subroutine edited by Saini [29], the wall thickness is read and if the
wall exists, the CHT would be active and the subroutine CHTBLOCK will be called.
64
GETDIFF.F: In this subroutine first edited by Saini [29] and modified by me, the solid
elements is recognized based on the distance to the wall of the elements, which is calculated by
the distance of each node in the element. The solid properties will be assigned to the element once
it’s recognized as the solid elements. The arraies rho, cp and k store density, specific heat capacity
diffus(:) = k_T(:)/(rho(:)*cp(:))
65
ProQuest Number: 28688460
This work may be used in accordance with the terms of the Creative Commons license
or other rights statement, as indicated in the copyright statement or in the metadata
associated with this work. Unless otherwise specified in the copyright statement
or the metadata, all rights are reserved by the copyright holder.
ProQuest LLC
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346 USA