Enhanced Heat Transfer in Natural Convection by Means of Liquid Metals and

Partitioned Domains

By

Joel Bruce Smith

A thesis submitted in partial fulfillment

of the requirements for the degree of
Master of Science in Engineering
(Mechanical Engineering)
in the University of Michigan-Dearborn
2024

Master’s Thesis Committee:

Professor Oleg Zikanov, Chair

Assistant Professor Hugo Casquero
Assistant Professor Doohyun Kim
 Table of Contents

List of Tables………………………………………………………………………………. iii

List of Figures……..................……….…………………………………….……………... iv

List of Variables………………...…………………………………………………....…..... vi

Abstract………………………………………………...……………………….………..... vii

1. Introduction……………………………………………………………………………… 1

1.1. Literature Review……………………………………………………………………… 2

2. Methods………………………………………………………………………………….. 7

2.1. Governing Equations and Physical Parameters………………………………………... 7

2.2. Approach to Simulations……………………………………………………………… 10

2.3. Model Verification………………………………………………………………….… 12

2.4. Grid Sensitivity Study..……………………………………………………………….. 13

2.5. Simulation Procedure...……………………………………………………………….. 16

3. Results and Discussion…………………………………………………………………. 18

3.1. Effect of Partition on Heat Transfer…… ……………………………………………. 18

3.2. Effect of Partition on Flow Structure …………….………………………………….. 20

i
3.3. Relation Between the Optimal Gap Size and Thermal Boundary layer Thickness.….. 31

3.4. Further Heat Transfer Optimization………………………………………………….. 35

4. Conclusion……………………………………………………………………………… 37

References………………………………………………………………………………… 39

ii
 List of Tables

Table 1…………………………………………………………………………….….…...... 9

Table 2…………………………………………………………………………….….….... 13

Table 3…………………………………………………………………………….…...….. 14

Table 4…………………………………………………………………………….….….... 15

Table 5…………………………………………………………………………….….….... 18

Table 6…………………………………………………………………………….….….... 19

Table 7…………………………………………………………………………….….….... 35

Table 8…………………………………………………………………………….….….... 36

iii
 List of Figures

Figure 1……………………………………………………………………………..……… 4

Figure 2……………………………………………………………………………..……… 8

Figure 3……………………………………………………………………………..…….. 15

Figure 4……………………………………………………………………………..…….. 16

Figure 5……………………………………………………………………………..…….. 17

Figure 6……………………………………………………………………………..…….. 19

Figure 7……………………………………………………………………………..…….. 20

Figure 8……………………………………………………………………………..…….. 21

Figure 9……………………………………………………………………………..…….. 22

Figure 10……………………………………………………………………………..…… 23

Figure 11……………………………………………………………………………..…… 24

Figure 12……………………………………………………………………………..…… 25

Figure 13……………………………………………………………………………..…… 26

Figure 14……………………………………………………………………………..…… 27

Figure 15……………………………………………………………………………..…… 28

iv
Figure 16……………………………………………………………………………..…… 30

Figure 17……………………………………………………………………………..…… 30

Figure 18……………………………………………………………………………..…… 32

Figure 19……………………………………………………………………………..…… 33

Figure 20……………………………………………………………………………..…… 34

Figure 21……………………………………………………………………………..…… 36

v
 List of Variables

AR – Aspect Ratio Tmelt – Melting Temperature

c – Courant Coefficient U – Free-Fall Velocity

Cp – Specific Heat α – Ratio of Top and Bottom Hg

D – Cylinder Diameter β – Coefficient of Thermal Expansion

g – Acceleration of Gravity ΔT – Temperature Difference

Gr – Grashof Number Δt – Time Step Size

H – Cylinder Height Δx – Mesh Element Length

Hg – Gap Height Above and Below Partitions δ – Ratio Partition Gap and Cylinder Height

NE – Number of Elements of Compute Grid δB – Average Thermal Boundary Layer H

p – Pressure δBL,lower – Lower Thermal Boundary Layer H

pb – Fluid Bulk Pressure δBL,upper – Upper Thermal Boundary Layer H

Pr – Prandtl Number φ – Angle of Applied Gravity

Ra – Rayleigh Number χ – Thermal Diffusivity

Racr – Critical Rayleigh Number λ – Thermal Conductivity

STB – Top/Bottom Cylinder Surface Area ρ – Density

T – Temperature µ – Dynamic Viscosity

Tb – Fluid Bulk Temperature ϴ – Angle Between x and y Magnitudes

Tcold – Cold Sink Surface Temperature ν – Kinematic Viscosity

Thot – Heat Source Surface Temperature Ω – Frequency

vi
 Abstract

Heat generation by commonly used systems and components, such as the large batteries used for

energy storage, powerful instrumentation in computing, and advanced HVAC and climate

control systems, has continued to increase and is further augmented by technological

advancement. Assuming progress continues, the research of heat transfer efficiency remains a

meaningful and worthwhile endeavor. This study explores possible ways to increase the

effectiveness of heat transfer based on natural convection for systems at relatively low

temperatures, which increases the range of applications for which it can be applied. It is

hypothesized that the high energy density and high thermal conductivity of liquid metals and the

effects of vertical partitions on flow organization in a fluid cavity can positively impact the heat

transfer rate of a convective cell. The hypothesis is explored for a geometry of a cylindrical

cavity with a single partition using Ansys Fluent CFD simulations. The aspect ratio of the

cylinder, the Rayleigh number of the convective fluid flow, and the gap height between the top

and bottom cylinder surfaces and a partition are considered as factors of a parametric

optimization study. The results show manyfold enhancement of the heat transfer rate by a

partition and indicate a strong potential in heat transfer applications.

vii
 1. Introduction

Rayleigh-Benard Convection (RBC), a phenomenon in which thermal stratification of a

fluid within a cavity facilitates convection, can be observed improving heat transfer amongst a

myriad of thermal systems both natural and man-made. This can be observed prominently in a

vertically oriented thermal system circulating a working fluid such as air, water, or refrigerant. In

the interest of expanding on its potential, we explore alternative configurations of RBC that may

yield greater improvements in heat transfer.

Although not commonly used, liquid metals have a high specific heat and thermal

conductivity, consequently having a greater heat transfer potential in comparison to common

working fluid [1]. However, liquid metals have a relatively high melting point and are therefore

only considered for applications at higher temperatures. To optimize heat transfer for an array of

systems that operate at low temperatures, liquid gallium will be used as the working fluid in this

research. Its low melting point makes it liquid at just above room temperature which is within the

temperature range of typical water or air cooled systems. Moreover, data on the material

properties of liquid gallium is more readily available when compared to other liquid metals and

compounds. Further research into convective heat transfer improvements for RBC shows

potential with the implementation of partitions in a convective cell by assisting in the

organization of flow, although only observed for water and air [2], [8].

The hypothesis motivating this study is that a significant heat transfer enhancement can

be achieved by combining the effects of liquid metal used as a working fluid and a partition of

1
the flow domain. In the following sections, background information pertaining to liquid metal

convective heat transport, partitioned flow, methods used for analysis and simulation, and

outcomes of the simulations and resulting data will be discussed.

1.1. Literature Review

Understanding the effects of partitioned RBC for a liquid metal flow requires

comprehension of the factors affecting the heat transfer in RBC. Formation of a thermal

boundary layer is expected for flows with a temperature difference between a surface and fluid

flowing over it. This effect should be notable for fluids with a low Prandtl number fluid which

dissipate heat quickly, such as liquid metals [11]. For simulation of a vertically mounted

cylindrical convective cell, thermal boundary layers are predicted to form near the cold and hot

regions of the top and bottom surfaces, respectively. Identifying these thermal boundary layers,

which are a source of resistance to heat transfer, provides information on the effectiveness of

RBC and how much of an effect the thermal boundary layer resistance has. There exists a

plethora of research investigating methods for reducing this resistance, such as deformation of

the boundary layers. In one such study, boundary layer deformation of the standing-wave type,

that being a combination of two waves at the same amplitude and frequency, changes global

responses to convection turbulence, given that the deforming amplitude of the standing waves is

close to or larger than the boundary layer thickness of the flow in RBC [5].

The large-scale circulation caused by RBC in a convective cell also has a prominent

effect on the heat rate transfer [8]. Modifying the large-scale circulation may, therefore, be used

to increase the heat transfer rate. Due to a limited number of published studies on liquid metals,

value can be found in works published on more common working fluids. Research from a study

on large-scale circulation used empirical data produced through experimentation to study

2
different types of convective domains for modifying RBC bulk flow [4]. For two different flow

domain setups, type 1 with a square grid suspended in a convective cell and type 2 with the grid

fully extending to the top and bottoms of the convective cell, heat transfer efficiency within the

type 1 domain was enhanced by up to 14%. It was suggested that increased plume coherency

caused this effect. Furthermore, heat transfer efficiency in the type 2 domain, with longer

segments or “sub-units” of flow, increased by as much as 30%. These results confirm that the

organization of the flow is a component of heat transfer optimization.

Geometry and orientation of the convective cell cavity play a significant factor in RBC

heat transfer. For a cylinder, the ratio of the height to diameter, known as the aspect ratio AR,

influences RBC by defining the shape in which flow can occur. Research suggests an increase in

heat transfer can be observed in convective cells with a lower AR, caused by modification of

large-scale circulation. In a confined geometry, the amount and intensity of hot and cold plume

clusters increases and are more energetic, which has a significant influence in reducing the

thickness of the thermal boundary layers [6]. However, the organization of the flow aided by the

partitions may facilitate increased heat transfer even with an increase in the surface area of a

taller convective cell of the same diameter. Additionally, the effect of a smaller convective cell

must also be noted, which leads to a decrease in the volume of fluid and thus reduces the amount

of heat transport that can be accommodated.

Looking at the effects of just the partitions on the convective cell geometry shows

promise for improved heat transfer. Partitioned RBC may lead to reduction of heat exchange

between hot ascending and cold descending jets which reduce heat transfer [5], [7]. Further

research shows that adding vertical partitions in a convective cell with a high-Prandtl number

liquid increases convective heat transfer in a liquid medium by increasing Nu when compared to

3
non-partitioned cases [2]. Investigation of the causes of this increase suggests that partitions in a

large-scale circular fluid flow within a fluid cavity create a symmetry-breaking bifurcation,

causing the fluid to organize into a unidirectional flow along the partition walls. This can create a

disruption of the thermal boundary layer where the partitions extend close to the top and bottom

walls [2], [8]. Furthermore, it has been observed that mean velocity and temperature fields are

correlated due to the increased coherency of the flow as number of partitions increases, leading

to a meaningful, albeit small, improvement in heat flux. It should be noted that as the number of

partitions increases, the volume of fluid within the cell decreases and impedance from the no-slip

condition of the partitioned walls increases. Adding volume to the convective cell to compensate

for the loss of working fluid may prove to negate these effects.

Figure 1. Simulation results of the temperature distributions for Ra=106, AR=1 which are

discussed in detail in section 3. Instantaneous distributions of temperature in the vertical cross-

section are shown. Left: RBC in a convective cell without a partition showing the typical flow

organization. Right: RBC in a partitioned convective cell for δ=0.16 demonstrating stronger and

more coherent large-scale circulation.

4
 Additional investigation into the effects of partitions on a convective cell should yield an

optimized configuration for increased convective heat transfer. It has been observed that the

Nusselt number increases monotonically as the number of partitions aligned perpendicularly in a

convective cell increase [2]. Aided by the partitions, the working fluid is forced through the gaps

at the top and bottom of the partitions and leads to a pressure distribution at the gaps that sustain

flow, creating horizontal jets, which will sweep the thermal boundary layer, disturbing these

layers and further increase thermal efficiency [2]. The size of the gap height also influences the

fluid flow and heat transport and may influence the flow in relation to the cell height [7].

There are several challenges when looking at generating valuable data for liquid metal

flows. Liquid metals are opaque, and so measuring velocity using optical techniques, such as

PIV, is impossible. What is more, systems using liquid metal tend to operate at much higher

temperatures than conventional liquids such as air or water, creating difficult working conditions

and energy requirements [1]. Regardless, an advantage of liquid metals that cannot be

overlooked is their high thermal conductivity that facilitates more heat transfer.

It must be stressed that the effect of partitions on heat transfer in RBC is, while evident,

not very strong in conventional fluids such as water. Increase of the Nusselt number by up to

30% is reported in [2], [7], [8]. As we will see in the discussion of the results in section 3, the

effect is much stronger in fluids with low Prandtl numbers, e.g. liquid metals.

Understanding the advantages of partitioned RBC provides evidence that an optimized

AR for increased RBC, in combination with a specific number of partitions with a given gap

height, can be used to design a fluid model for which heat transfer can be optimized. It is

expected that the high thermal conductivity and low viscosity of the liquid gallium should further

add to optimized heat transport, outweighing the possibility of reduced heat transfer from fluid

5
volume reduction and increased surface area the partition will add. Thus, this thesis will report

on simulations combining the effects of liquid metal and partitioned flow in a convective cell.

6
 2. Methods

This section will detail methods used to set up and verify the accuracy of simulations.

The results will be reported and discussed in section 3.

2.1. Governing Equations and Physical Parameters

An unsteady three dimensional flow in a partitioned cylindrical cavity acting as a

convective cell is calculated. Several assumptions are made to simplify the model. The

Oberbeck-Boussinesq approximation is assumed in which all physical properties of the fluid are

assumed constant except density in the gravity force term. Density in this case is assumed to be a

linear function of temperature, thus providing the buoyant force effect. The sidewalls and the

wall of the partition are assumed adiabatic with heat transfer only occurring at the top and

bottom walls of the cell. No-slip boundary conditions are assumed at the walls of the cavity and

along the partition walls. Under these conditions, with the cylinder height H, free-fall velocity

U≡�gβ∆T H, and ∆T as the typical scales, flow within the cavity can be represented by the

governing equations:

∇ ∙ 𝐮𝐮 = 0 (1)
1 1
∂𝐮𝐮
�
∂t
+ (𝐮𝐮 ∙ ∇)𝐮𝐮 = −∇p + Pr 2 Ra−2 ∇2 𝐮𝐮 + T𝐞𝐞z (2)

∂T
∂t
+ 𝐮𝐮 ∙ ∇T = (RaPr)−1/2 ∇2 T (3)

7
 With boundary conditions

Top Surface: z = 1, T = T2 Bottom Surface: z = 0, T = T1

1 ∂T ∂T
Cylinder Side Wall: r = 2AR , ∂r
=0 Partition Walls: ∂r
=0

Figure 2. Geometry of the convective cell. Left: Top-down cross section of the center of the

convective cell. Right: Front facing perspective of the convective cell.

The control parameters for this model are the Rayleigh number Ra, the aspect ratio of the

cylinder AR, and the ratio of the partition gap height to the cylinder height δ. A change in the

rate of heat transfer is anticipated by modifying AR of the convective cell cylinder, a second

parameter identified as the ratio of the gap height Hg between the partition fixed to the center of

the cylinder δ, and the Ra. While AR and δ are directly correlated to the model geometry, Ra and

the Prandtl number Pr (ratio of kinematic viscosity to thermal diffusivity) represent design

parameters determined by the properties of the fluid and the temperature difference ΔT=T1-T2.

Specifically, Ra is the relationship between the Grashof Number Gr (ratio of buoyant force to

8
viscous force) and the Prandtl number Pr. A relationship between these properties and the design

parameters is shown in (1) and (2). H is the characteristic length represented by the height of the

cylinder, D is the cylinder diameter, g is the acceleration of gravity, Thot is the heat source

temperature along the bottom surface, and Tcold is the fluid bulk temperature.

Material Properties
Tmelt 302.95 K
χ 1.2x10-5 m2/s
β 1.2x10-4 K-1
λ 30 W/m*K
ρ 6040 kg/m3
ν 3.0x10-7 m2/s
µ 1.81x10-3 kg/m‧s
Table 1. Material properties of liquid gallium at 60°C [9]. Properties listed are as follows:

melting temperature Tmelt, thermal diffusivity χ, coefficient of thermal expansion β, thermal

conductivity λ, density ρ, kinematic viscosity ν, and dynamic viscosity µ.

g β(Thot −Tcold )H3 D Hg ν

Ra = , AR = H , δ= , Pr = χ (4)
χυ H

Ra = Gr ∗ Pr (5)

Research from section 1.1 is used to develop the modeled environment for simulation of

partitioned RBC with liquid gallium, the properties for which are listed in Table 1. The work

presented in this thesis is limited to the case of a single partition. The effect of multiple partitions

is left to future studies. It is anticipated that the actual width of the partition will have minimal

effects on the model assuming the partition width remains thin, and so the width of the partition

will be kept at a constant value of 10% of H for each model.

9
 In a fully developed flow, the heat fluxes through the top and bottom walls fluctuate

around the same constant mean Qconv. This value is produced by averaging instantaneous values

of Qconv,inst in time over a long period of evolution of fully developed flow. Qconv is then

calculated using (6) and is then used to calculate the Nusselt number Nu in (8). This value

represents the ratio of convective over conductive heat transfer. Conductive heat transfer Qcond

will be calculated for each simulation using (7) which is a function of the cylinder height,

temperature difference, and fluid properties. Higher values for Nu indicate increased convection,

indicating which models produce better results for convective efficiency.

Qconv,inst
Qconv = (6)
NT

λSTB ΔT
Qcond = H
(7)

Qconv
Nu = (8)
Qcond

2.2. Approach to Simulations

As described below, the problem is solved computationally using Ansys Fluent, which

requires the problem to be presented in dimensional units. The dimensional parameters are found

for a given set of AR and Ra by fixing the temperature difference between the hot and cold

surfaces ΔT and using the physical properties of gallium reported in table 1. Values for H, D, and

Hg are then calculated, which provides the desired values of AR, Ra, and δ. Simulations are

carried out for Ra=106, 107 and AR=1, 2, 3, 5 and the partition gaps for δ=0.04, 0.06, 0.08, 0.10.

This creates a minimum of five simulations for each combination of Ra and AR, including the

non-partitioned cases. The goal of the simulations is to determine for each combination of AR

and Ra, which values of δ produces the greatest heat transfer, resulting in the highest value for

10
Nu. Additional simulations for higher or lower δ are run in case a local maximum is not found

within the initial simulation set.

Beginning with the DesignModeler in Ansys, the geometry is built according to

calculated values for H and D. The geometry is then meshed using the verified refined meshes

discussed in Section 2.3. Setup for the simulations is done by setting a double precision and

assigning CPUs. The number of CPUs per simulation is set based on the number of mesh

elements, using an approximation of 1 CPU per 100,000 elements needed.

Ansys Fluent is capable of both pressure and density based solvers. These models will be

run with a pressure-based transient solver due to the incompressible nature of liquid metal fluid

flow and account for the change in flow over time from turbulent to fully developed flows.

Model parameters are then set within the simulation by inputting the materials properties for

gallium listed in Table 1, setting the flow to laminar, and having the energy equation turned on.

The convective cell boundaries are set at temperatures of Tcold=323.15K and Thot=363.15K for

the top and bottom surfaces, respectively, giving a value for ΔT=40K. Tcold and Thot are

nondimensionalized by the initial bulk temperature of the fluid Tb=343.15K, giving a

nondimensional temperature range of 0.94≤T≤1.06. The initial conditions are the distribution of

temperature and vertical velocity represented by (9) and (10), respectively.

2 2)
D2
uz (x, y) = −(x + y + (9)
4

∆T
T(z) = Thot + �− �∙z (10)
H

11
 The flow evolution is computed with the time step Δt=0.1s and up to 200 iterations

performed per Δt. The simulations fully developed flows used for final data acquisition run for a

minimum of 100s. This period is increased as needed to ensure convergence of the computed

statistical means to a steady value (see Figure 4). The data collected are graphed for Qconv at the

top and bottom surfaces and values recorded before Δt=40s ignored when calculating Nu to

account for the transient nature of the initial flow profile.

2.3. Model Verification

Before partitioned models are simulated, an initial non-partitioned model is run, and the

resulting data is compared to verified data of RBC at high-resolution simulations of RBC in an

inclined cylinder with low Pr [3]. The non-dimensional parameters are set at Ra=106 and Pr=0.1,

matching the non-dimensional values in the referenced study.

The geometry is built using calculated dimensions H and D and is then meshed with

number of elements NE=130,000 using hexahedrons as the element type. Quality of the mesh is

maintained by keeping the element aspect ratio below 6. These simulations are run at ΔT=10K

and at varying angles between applied gravity and the z axis of the cylinder φ between 0 and

0.3π for comparison to the referenced study [3].

These verification simulations are run at only 40 iterations for time steps Δt of 0.1s.

Collected data are averaged over a minimum time of 3s after steady state convergence is reached.

The resulting Qconv is used to calculate Nu, which is then compared to results of [3]. These

calculations show excellent qualitative agreement with Qconv increasing with φ as predicted in

[3]. The quantitative agreement is within 5% shown in Table 2.

12
 φ 0 0.1π 0.2π 0.3π
Nu[3] 7.25 8.10 8.50 8.70
Nu[This Study] 7.15 8.40 8.91 9.06
Percent Error 1.3% 3.5% 4.6% 3.9%
Table 2. Percent errors for Nu values in an incline cylinder between simulated results and the

data of high resolutions [3].

2.4. Grid Sensitivity Study

Additional testing of this model is performed to determine mesh independence, ensuring

that values for heat transfer are only a function of the flow parameters and not of the mesh

parameters. This sensitivity study will ultimately determine the element size, number of elements

NE, and element type which will provide accurate results. First, the simulation of a flow in a non-

partition cavity is run at Ra=105 and AR=1 for meshes with hexahedron elements at NE=133,104

and tetrahedral element meshes at NE=133,326. Increased mesh density spanning a distance from

the top and bottom surfaces at approximately 5mm is included for both models to account for the

increased activity near the thermal boundary layers. Nu is then calculated as mentioned in

Section 2.1, using the time average values of the instantaneous temperatures where both models

reach a fully developed flow, this being after 600 time steps, and using (6), (7), and (8). Values

of Nu for the hexahedron and tetrahedral models are found to be Nu=3.25 and Nu=3.23,

respectively. This gives a percentage error of less than 1% and validates that the tetrahedral

elements can be used in place of hexahedron elements. Although hexahedron elements can

provide improved accuracy with less elements, tetrahedral elements are better for complex

geometry which will be a factor once adding the partition to the model. Therefore, further

meshing will be done with the tetrahedral elements.

13
 Finding the minimum mesh resolution that provides a reasonably accurate solution is

determined by running multiple tests of a non-partitioned models at Ra=106, AR=1 and a

partitioned model at Ra=107, AR=3, δ=0.016 both with increasing NE. The accuracy is

determined by finding the point at which the value of Nu change minimally as NE changes. As

can be seen in Figure 3, Nu decreases as the resolution increases, with Nu starting to level out at

approximately NE=1.4 million elements. This number is taken as acceptable in providing

sufficient accuracy for the purposes of this study. Anticipated increased activity in and around

the thermal boundary layers near the top and bottom surfaces requires a higher mesh resolution,

which can be seen in Figure 5. This increased resolution is especially important when adding the

partition which falls within the thermal boundary for each partitioned model.

The accuracy of the numerical model used in the study is further analyzed by using the

computed maximum velocity found to calculate the Courant Coefficient c. The coefficient is

used to determine the stability of schemes for hyperbolic equations and establishes the distance

to which information is transported by velocity over a time step in relation to the mesh step [10].

Explicit schemes for purely hyperbolic equations are generally stable if the Courant-Friedrichs-

Lewy stability condition (11) is met. Ansys simulations use an implicit scheme which can be

considered stable even when the condition is not met, the values for which are listed in Table 3.

∆t
c = |umax | ≤1 (11)
∆xmin

Nondimensional Parameter Instantaneous Mesh Length at C

Combinations at N ~ 1.4 Million Max Velocity (m/s) Max Velocity (m) (Δt=0.1s)

Ra=106 AR=1 No Partition 3.72E-2 1.50E-4 24.8

Ra=107 AR=3 δ=0.016 1.59E-1 4.15E-4 38.3
Table 3. Courant coefficients c calculated for further mesh verification.

14
 Number of Nu for Nu for
Elements Ra=106 Ra=107
2900000 6.25 -
2230000 6.27 6.99
2000000 - 7.01
1350000 - 7.05
1200000 6.28 -
1060000 7.17
750000 6.37 -
690000 - 7.37
345000 - 7.58
80000 - 8.09
Table 4. Values of Nu as a function of NE which is represented graphically in Figure 3.

7.6

7.4
Ra=106
Ra=107
7.2

7
Nu

6.8

6.6

6.4

6.2
0 0.5 1 1.5 2 2.5 3
NE 10 6

Figure 3. Nu as a function of number of elements N. Results are shown for the non-partitioned

domain at Ra=106 AR=1 (red) and partitioned domain at Ra=107 AR=3 δ=0.016 (green).

It should be noted that implicit schemes are unconditionally stable but have a truncation

error heavily influenced by numerical dissipation; this can result in the amplification of a

rounding error. To mitigate this effect, further simulations continue with a small time step size at

0.1s. These verifications lead to the final design parameters for data collection simulations.

15
2.5. Simulation Procedure

Each model geometry is built using values for H based on set combinations of the non-

dimensional parameters Ra and AR. A non-partitioned case is first run to find a baseline value

for the heat transfer at the top and bottom of the cells and calculate Nu0. Instantaneous velocities

along the z-axis in each non-partitioned simulation are collected and averaged over time to

establish upper and lower thermal boundary layer thicknesses δBL,upper and δBL,upper.

dT −1
δBL,lower ≡ (Tb − Thot ) � dz � (12)
z=0

dT −1
δBL,upper ≡ (Tb − Tcold ) � dz � (13)
z=H

Where Tb is the bulk temperature of the fluid outside of the boundary layer. The models

for partitioned domains are then built for δ=0.04, 0.06, 0.08, and 0.10. If a local maximum of

Qconv is not found in these simulations, models for additional values for δ are built and simulated.

Figure 4. Graph for the total rate of heat transfer at the top and bottom of the cell during the

entire simulation for Ra=106 AR=1. The trend over time shows convergence of the heat transfer

rate at both top and bottom walls to approximately 275W.

16
Figure 5. Zoomed in region of the mesh at the partition gap showing increased mesh resolution

within the thermal boundary layers near the top and bottom surfaces.

17
 3. Results and Discussion

3.1. Effect of Partition on Heat Transfer

The main results of the study are summarized in Table 5, showing all but two simulations

presenting a notable increase in Nu over the non-partitioned counterparts, the exceptions being

δ=0.04 and δ=0.06 for Ra=106 AR=1. A value of δ was found for each combination of Ra and

AR which produced the highest Nusselt number δmax, resulting in the greatest convective heat

transfer amplification.

The data in table 5 present a trend in which δmax decreases as AR increases, which is

further decreased with an increase in Ra as shown in Figure 7. This demonstrates an inverse

relationship between δ and combinations of AR and Ra.

Ra 106 107
AR 1 2 3 5 1 2 3 5
No Partition 6.28 7.84 3.06 1.00 10.5 9.28 7.92 4.65
δ=0.01 - - - - - - - 17.45
δ=0.02 - - - 10.91 - - 18.69 25.65
δ=0.04 5.77 7.86 11.07 13.05 12.40 17.08 21.33 25.25
δ=0.06 6.20 9.32 11.49 11.40 12.50 17.66 19.32 20.60
δ=0.08 6.40 9.86 10.56 9.65 12.80 16.98 17.91 19.15
δ=0.10 6.80 9.56 9.57 8.35 11.80 16.38 16.38 17.70
δ=0.12 6.83 - - - - - - -
δ=0.14 6.90 - - - - - - -
δ=0.16 6.78 - - - - - - -
Table 5. Time-averaged values of Nu computed in all completed simulation.

18
 Ra 106 107
AR 1 2 3 5 1 2 3 5
δ=0.01 - - - - - - - 3.75
δ=0.02 - - - 10.90 - - 2.36 5.51
δ=0.04 0.92 1.00 3.62 13.10 1.18 1.84 2.69 5.43
δ=0.06 0.98 1.19 3.75 11.05 1.19 1.90 2.44 4.43
δ=0.08 1.02 1.26 3.45 9.65 1.21 1.83 2.26 4.12
δ=0.10 1.08 1.22 3.13 8.35 1.12 1.77 2.07 3.81
δ=0.12 1.09 - - - - - - -
δ=0.14 1.10 - - - - - - -
δ=0.16 1.08 - - - - - - -
Table 6. Normalized values of the Nusselt Number (Nu/Nu0).

Figure 6. Graphical Representations of normalized values for the Nusselt Number.

This trend stays consistent with the data for the Nusselt number of partitioned models

normalized with the Nusselt number for non-partitioned models Nu/Nu0, shown in Figure 6. This

figure also shows that higher values of Nu are produced for AR≤2 at Ra=107 and for AR≥3 at

Ra=106, presenting a relationship between AR and Ra that can be observed in Figure 7.

19
 Figure 7. Left: Graphical representation of the inverse relationship between δmax and

combinations of Ra and AR. Right: Normalized values of Nu at δmax as a function of Ra and AR

showing a direct relationship that becomes more extreme as Ra decreases.

3.2. Effect of Partition on Flow Structure

Figures 8a through 15g present the flow structure and discuss the results based on

temperature profiles with a nondimensional range of T = 0.94 (blue) to T = 1.06 (red). The

partition created changes in the flow structure in two significant ways. First, it can be observed

that adding the partition creates the large-scale circulation of the flow, which was noted to

influence heat transfer when modified [8]. Second, having δ smaller than the thermal boundary

layer leads to an increased heat flux by intensifying the heat transfer across the thermal boundary

layer. However, a decrease of δ also has a detrimental effect on heat transfer, because a small

gap creates obstruction to the flow which reduces its kinetic energy. The relation between δ and

the thickness of the thermal boundary layer is further discussed in section 3.3.

20
 a. No Partition b. δ=0.04

c. δ=0.06 d. δ=0.08

e. δ=0.10 f. δ=0.12

g. δ=0.14 h. δ=0.16

Figure 8. Typical distribution of temperature in the vertical axial cross-section and perpendicular

partition. Flows with Ra=106, AR=1 are shown.

21
 a. No Partition b. δ=0.04

c. δ=0.06 d. δ=0.08 e. δ=0.010

Figure 9. Typical distribution of temperature in the vertical axial cross-section and perpendicular

partition. Flows with Ra=106, AR=2 are shown.

22
 a. No Partition b. δ=0.04

c. δ=0.06 d. δ=0.08 e. δ=0.010

Figure 10. Typical distribution of temperature in the vertical axial cross-section and

perpendicular partition. Flows with Ra=106, AR=1 are shown.

23
 a. No Partition b. δ=0.02 c. δ=0.04

d. δ=0.06 e. δ=0.08 f. δ=0.010

Figure 11. Typical distribution of temperature in the vertical axial cross-section and

perpendicular partition. Flows with Ra=106, AR=5 are shown.

24
 a. No Partition b. δ=0.04

c. δ=0.06 d. δ=0.08 e. δ=0.010

Figure 12. Typical distribution of temperature in the vertical axial cross-section and

perpendicular partition. Flows with Ra=107, AR=1 are shown.

25
 a. No Partition b. δ=0.04

c. δ=0.06 d. δ=0.08 e. δ=0.010

Figure 13. Typical distribution of temperature in the vertical axial cross-section and

perpendicular partition. Flows with Ra=107, AR=2 are shown.

26
 a. No Partition b. δ=0.02 c. δ=0.04

d. δ=0.06 e. δ=0.08 f. δ=0.010

Figure 14. Typical distribution of temperature in the vertical axial cross-section and

perpendicular partition. Flows with Ra=107, AR=3 are shown.

27
 a. No Partition b. δ=0.01 c. δ=0.02

d. δ=0.04 e. δ=0.06 f. δ=0.08 g. δ=0.01

Figure 15. Typical distribution of temperature in the vertical axial cross-section and

perpendicular partition. Flows with Ra=107, AR=5 are shown.

28
 Organization of the flow due to the addition of the partition is evident in the simulation of

the temperature and velocity contours shown in Figures 16 and 17. Horizontal jets occur at the

top and bottom gaps with the addition of the partition and large-scale circulation is produced.

These jets flow through and disrupt the top and bottom thermal boundary layers, contributing to

thermal efficiency. Given the results in Table 4, it can be assumed that organization of the fluid

flow facilitates increased convective heat transfer. The one instance in which convective heat

transfer experienced a decrease in Nu compared to its non-partitioned case may speak to the

theory of increased heat transfer due to a more compressed model as this specific instance was

set at AR=1 [6]. Table 6 shows the results for the analysis of the thermal boundary layers and

can be observed in Figures 18 and 19.

The case of Ra=106 AR=5 with no partition is an outlier of the non-partitioned cases

having no velocity with no flow having been developed. This can be observed in Figure 11a

where the temperature gradient remains perfectly stratified due to the lack of fluid flow. It has

been shown through analysis [13] and demonstrated experimentally [14] that flows within

cylindrical containers of AR=5 for a fluid at Ra=106 have a stability that falls near the Critical

Rayleigh Number, Racr, the threshold for which convective flow begins to develop. Due to the

strong numerical dissipation produced by the finite-volume scheme used by Fluent, it is possible

that a weak convective flow just above Racr are not captured in simulation. The comparison of

analyzed and experimental data presented by Muller, Neumann, and Weber in their work on

natural convection in cylindrical cavities [14] shows that a flow with Ra=106 AR=5 is weak and

non-turbulent. From this, it can be assumed that flow produced at these conditions does not

manifest in Fluent simulation, where flow computed in a non-partitioned domain shows zero

velocity and only a purely conduction profile of temperature. This is not surprising, considering

29
Figure 16. Temperature and velocity contours of x-velocity showing evidence of horizontal jets

in the gaps between the top/bottom walls and the partitions. Instantaneous distributions for x-

velocity contours Left: Ra=106, AR=1, δ=0.14 for nondimensional velocity range -0.74≤u≤0.77;

Right: Ra=107, AR=3, δ=0.04 for nondimensional velocity range -0.84≤u≤0.81.

Figure 17. Horizontal cross sections of fluid temperature for Ra=106, AR=1, and δ=0.14. Left:

Cross sections above and below the partition. Right: Cross section through cylinder center.

30
that the sidewall closeness increases Racr above the point at which convection first occurs, to

about Racr=7x105 [14]. Therefore, convection not occurring in the simulations at a slightly higher

Ra=106 is attributable to the known sensitivity of instability to numerical dissipation of the

numerical method. The dissipation is relatively high in the finite-volume solution of the models.

Results presented in Figure 11 show that a partition may generate convection flow in geometries

where convection would not otherwise happen.

3.3. Relation Between the Optimal Gap Size and Thermal Boundary layer Thickness

The relationship between the size of the gaps between the partition and the top and

bottom walls and the thickness of the thermal boundary layer is further explored in this section.

For calculating the boundary layer thickness, flows computed for non-partitioned domains are

used to find the vertical profiles of time-averaged temperature along the cylinder axis. These

results are presented in Figures 19 and 20 and show that, except for the case Ra=106, AR=5, in

which no flow and, thus, no boundary layer exists, the boundary layer thicknesses can be

determined using (12) and (13) and are shown in Table 7.

The thicknesses obtained in each case for the upper and lower walls are averaged to

provide the final estimates for δBL and presented in Table 7, showing that the boundary layer

thickness drops about twofold as Ra increases from 106 to 107. The effect of AR is weaker. At

both values of Ra, δBL decreases as AR decreases from 1 to 2 and increases with AR at AR ≥ 2.

Analysis of the results in terms of the ratio between the partition gap size and the thickness of the

thermal boundary layer δ/ δBL is presented in Figures 20 and 21. Unfortunately, an hypothesis

that the amplification of heat transfer can be approximated as a function of the single parameter

δ/ δBL is not supported by the data. The curves of Nu/Nu0 obtained for various Ra and AR do not

collapse into one curve if plotted with δ/ δBL (see Figure 18). Even more disappointing is that the

31
ratio δmax/ δBL varies strongly with Ra and AR. The scatter plot of δmax vs δBL shown in Figure 21

also does not show any clear dependency. However, while the hypothesis that δmax/ δBL being a

defining parameter is disproved by the results, it can be concluded that the optimal gap size δmax

is always smaller than the boundary layer thickness. This means that the partition must penetrate

the thermal boundary layer to facilitate a strong heat transfer enhancement.

Figure 18. Top Left: Normalized values of the Nusselt number as a function of the single

parameter δ/ δBL. Top Right: δ/ δBL as a function of AR (dependence on the geometry of the

cylinder). Bottom: Average boundary layer thickness as a function of δmax.

32
 AR = 1
363

353

343

333

323
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045

AR = 2
363

353

343

333

323
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045

AR = 3
363

353

343

333

323
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045

AR = 5
363

353

343

333

323
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045

Figure 19. Axial profiles of time-averaged temperature in flows without partitions at Ra=106.

33
 AR = 1
363

353

343

333

323
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1

AR = 2
363

353

343

333

323
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1

AR = 3
363

353

343

333

323
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1

AR = 5
363

353

343

333

323
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1

Figure 20. Axial profiles of time-averaged temperature in flows without partitions at Ra=107.

34
 Ra 106 107
AR 1 2 3 5 1 2 3 5
Tb 343.08 343.19 343.13 343.15 343.40 343.16 342.90 343.63
𝐝𝐝𝐝𝐝
� 𝐝𝐝𝐝𝐝 �lower 2986 2641 1658 N/A 2390 3012 2272 1094
δBL,lower 6.70 6.52 7.74 N/A 8.08 6.20 6.66 9.07
𝛅𝛅𝐁𝐁𝐁𝐁, 𝐥𝐥𝐥𝐥𝐥𝐥𝐥𝐥𝐥𝐥
𝐇𝐇
0.158 0.154 0.183 N/A 0.088 0.068 0.073 0.099
𝐝𝐝𝐝𝐝
� 𝐝𝐝𝐝𝐝 �upper 2660 2549 1686 N/A 2368 2938 2303 1070
δBL,upper 7.51 6.79 7.70 N/A 8.74 6.43 6.52 8.85
𝛅𝛅𝐁𝐁𝐁𝐁, 𝐮𝐮𝐮𝐮𝐮𝐮𝐮𝐮𝐮𝐮
𝐇𝐇
0.177 0.160 0.181 N/A 0.096 0.070 0.071 0.097
𝛅𝛅𝐁𝐁𝐁𝐁, 𝐚𝐚𝐚𝐚𝐚𝐚𝐚𝐚𝐚𝐚𝐚𝐚𝐚𝐚
𝐇𝐇
0.167 0.157 0.182 N/A 0.092 0.069 0.072 0.098
Table 7. Thicknesses of the upper and lower thermal boundary layers δBL (in mm) computed with

data from time-averaged temperature fields in domains without partitions (see text for

explanations). Note that the mesh refinement zones set in these simulations as 20% of the

cylinder height completely cover the thermal boundary layer in all simulations.

3.4. Further Heat Transfer Optimization

In this section, we consider the possible effects of asymmetry between the top and bottom

gaps of the partition. The asymmetry is defined by the non-dimensional parameter α, or the ratio

of Hg,upper and Hg,lower. A sampling of data was taken for the flow at Ra = 106, AR = 1, and δ =

0.06, with α adjusted from 0.5 up to 1.25. A local maximum of the Nusselt number was found at

α=0.75.

It was decided to test α=0.75 with the combinations of AR and δ that produced the

highest values for Nu at Ra=106 and Ra=107. The results are presented in Table 8. While there is

a similar increase in Nu for optimized conditions at Ra=107, there is a slight decrease for

Ra=106. These findings suggest that there is a potential for further increase of heat transfer rate.

Further investigations are required to explore this effect.

35
 Ra δ α =0.75 α =1

106 0.06 7.00 6.20

106 0.14 6.85 6.90

107 0.08 14.2 12.8

Table 8. Comparison of Nu as a function of α for set combinations of non-dimensional variables.

Figure 21. Results of modifying α for Ra = 106, AR = 1, and δ = 0.06.

36
 4. Conclusion

This thesis presented the results of numerical simulations of the Rayleigh-Benard

convection in cylindrical cells with a vertical partition and liquid gallium as a working fluid. The

results show that the use of a partition leads to very strong (more than tenfold in one case)

amplification of the rate of heat transfer. The amplification is much stronger than what was

observed in earlier studies with water [2], [6], [7] which we attribute to the effect of the low Pr of

gallium. Moreover, we found that with a configuration close to convection stability limit

(Ra=106, AR=5) the presence of a partition may cause a convection flow, even while such a flow

is not observed in the non-partitioned case. Unfortunately, our hypothesis that the effect of heat

transfer amplification is largely determined by the ratio of the gap width between the partition

and the cylinder walls and the thickness of the thermal boundary layer was not supported by our

data. Nevertheless, it can be concluded that the optimal gap size δmax is always smaller than the

boundary layer thickness, meaning the partition penetrating the thermal boundary layer does

assist in heat transfer enhancement.

Supplementary exploration of the combined effects studied in this thesis includes

extending the analysis to higher Ra, higher and lower AR, and various values of α. Additionally,

it is anticipated that the effect of heat transfer enhancement is not limited to the geometry of a

cylinder with a single partition. Other cavity shapes (e.g. a cuboid) and the use of multiple

partitions may prove beneficial. Lastly, an experiment confirming the effect may be simple

enough to perform and would further validate the simulated data.

37
 Future work related to this research may include more accurate numerical simulations,

which would be free from numerical dissipation and other accuracy-detrimental features of a

commercial CFD model. Furthermore, research into the use of an asymmetrically positioned

partition indicates the possibility of additional heat transfer amplification.

The simulated data from this study may prove useful for thermal management of systems

in which a significant amount of convective heat transfer is required. An example of this would

be a horizontally mounted cpu with a hypothetical cold sink resting above it. A convective cell

using a low temperature liquid metal could be used in between to facilitate a large amount of

heat transfer from the cpu for cooling [12]. Other promising applications are stationary battery

energy storage or cooling for high-rate power electronics equipment. A provision patent

application on the use of the described technological innovation is planned for the near future.

38
 References

[1] J. Scheel and J. Schumacher, “Global and local statistics in turbulent convection at low

Prandtl numbers,” Journal of Fluid Mechanics, vol. 802, pp. 147–173, Aug. 2016

[2] Y. Bao, J. Chen, B.-F. Liu, Z. She, J. Zhang, and Q. Zhou, “Enhanced heat transport in

partitioned thermal convection,” Journal of Fluid Mechanics, vol. 784, Nov. 2015

[3] O. Shishkina and S. Horn, “Thermal convection in inclined cylindrical containers,” Journal

of Fluid Mechanics, vol. 790, Feb. 2016

[4] L. Zhang and K. Xia, “Achieving heat transfer enhancement via manipulation of bulk flow

structures in turbulent thermal convection,” Physical Review Fluids, vol. 8, no. 2, Feb. 2023

[5] L. Yuan, S. Zou, Y. Yang, and S. Chen, “Boundary-Layer disruption and Heat-Transfer

enhancement in convection turbulence by oscillating deformations of boundary,” Physical

Review Letters, vol. 130, no. 20, May 2023

[6] S. Di Huang, M. Kaczorowski, R. Ni, and K. Xia, “Confinement-Induced Heat-Transport

enhancement in turbulent thermal convection,” Physical Review Letters, vol. 111, no. 10, Sep.

2013

[7] J. Chen, Y. Bao, Z. Yin, and Z. She, “Theoretical and numerical study of enhanced heat

transfer in partitioned thermal convection,” International Journal of Heat and Mass Transfer,

vol. 115, pp. 556–569, Dec. 2017

39
[8] P. K. Kar, U. Chetan, J. Mahato, T. L. Sahu, P. K. Das, and R. Lakkaraju, “Heat flux

enhancement by regular surface protrusion in partitioned thermal convection,” Physics of Fluids,

vol. 34, no. 12, Dec. 2022

[9] O. Zikanov, I. A. Belyaev, Y. Listratov, P. Frick, N. G. Razuvanov, and V. G. Sviridov,

“Mixed convection in pipe and duct flows with strong magnetic fields,” Applied Mechanics

Reviews, vol. 73, no. 1, Jan. 2021

[10] O. Zikanov, Essential computational fluid dynamics, 2nd Edition. John Wiley & Sons,

2019.

[11] Y. Cengel, R. Turner, and J. Cimbala, Fundamentals of thermal-fluid sciences, 4th Edition.

McGraw-Hill, 2012.

[12] J. Liu, Advanced liquid metal cooling for chip, device and system. 2022.

[13] G. S. Charlson and R. L. Sani, “On Thermoconvective instability in a bounded cylindrical

fluid layer,” International Journal of Heat and Mass Transfer, vol. 13, no. 9, pp. 1479–1496,

Sep. 1970

[14] G. Müller, G. Neumann, and W. Weber, “Natural convection in vertical Bridgman

configurations,” Journal of Crystal Growth, vol. 70, no. 1–2, pp. 78–93, Dec. 1984

40