This action might not be possible to undo. Are you sure you want to continue?
OSCILLATORY COMPRESSIBLE FLOW AND HEAT
TRANSFER IN POROUS MEDIA – APPLICATION TO
CRYOCOOLER REGENERATORS
A Dissertation
Presented to
The Academic Faculty
by
Jeremy Paul Harvey
In Partial Fulfillment
of the Requirements for the Degree
Doctor of Philosophy in Mechanical Engineering
Georgia Institute of Technology
December 2003
UMI Number: 3117935
________________________________________________________
UMI Microform 3117935
Copyright 2004 by ProQuest Information and Learning Company.
All rights reserved. This microform edition is protected against
unauthorized copying under Title 17, United States Code.
____________________________________________________________
ProQuest Information and Learning Company
300 North Zeeb Road
PO Box 1346
Ann Arbor, MI 481061346
OSCILLATORY COMPRESSIBLE FLOW AND HEAT
TRANSFER IN POROUS MEDIA – APPLICATION TO
CRYOCOOLER REGENERATORS
APPROVAL
Prateen V. Desai, Chairman, Mechanical Engineering
S. Mostafa Ghiaasiaan, Mechanical Engineering
Minami Yoda, Mechanical Engineering
Carl S. Kirkconnell, Raytheon Space and Airborne Systems
Jeffrey F. Morris, Chemical Engineering
Date Submitted: November 25, 2003
iii
ACKNOWLEDGEMENTS
I have been blessed to have been involved in this research with so fine a group.
And it is indeed the group of individuals whom have assisted me that have made this
research so enjoyable and productive. Beginning with those most directly involved in the
research, I would like to thank my committee. Dr. Prateen Desai, my thesis advisor,
fortunately thought of me when Dr. Carl Kirkconnell was searching out an undergraduate
student to perform this research for, at the time, Hughes Aircraft Company. The
Woodruff School and the Georgia Tech community have been so fortunate to have Dr.
Desai as a dedicated professor, and I as a friend and advisor. Dr. Mostafa Ghiaasiaan
quickly took an interest in the research after conversing with me and he now will lead the
cryocooler research at Georgia Tech. Drs. Minami Yoda and Jeffrey Morris completed
my team of advisors by providing breadth of knowledge in their fields. Of these
individuals, Dr. Kirkconnell deserves special thanks for being the catalyst for the research
at Hughes Aircraft , and now Raytheon Space and Airborne Systems. I hope that I can
provide the same enthusiasm and energy that you have provided me.
This research was performed due, in no small part, to the research grants provided
by Hughes Aircraft Company and Raytheon Company to the Georgia Tech Foundation.
Additional thanks must go to the Hughes and Raytheon Doctoral Fellowship Program and
the Georgia Institute of Technology Presidential Scholarship for supporting me directly
throughout this program.
At Raytheon, special thanks to Mr. Alberto Schroth, Mr. Kenneth Price, Mr.
Thomas Pollack, Mr. William Croft, and Mr. David Mc Gorrin. At Georgia Tech, I have
iv
been so fortunate to have several undergraduate and graduate students work with me in
my lab; Mr. Matthieu Coutaudier, Mr. Robert Hon, and Mr. Jeesung “Jeff” Cha, the next
torch bearer.
I thank my whole family for their support and love, especially my parents who are
so far away, but so near my heart, my sister Janelle, who I am so thankful is close, and to
my wife’s family who I wish were close. My wife Chrystine has been my rock during
this undertaking. She has given me so much hope and now our new life begins.
v
TABLE OF CONTENTS
Approval ............................................................................................................................. ii
Acknowledgements............................................................................................................ iii
List of Figures......................................................................................................................x
List of Tables ................................................................................................................... xiv
List of Tables ................................................................................................................... xiv
Nomenclature.....................................................................................................................xv
Summary............................................................................................................................xx
1. Introduction and Background ..........................................................................................1
1.1 Introduction ................................................................................................................1
1.2 Background.................................................................................................................4
1.2.1 History of Pulse Tube Cryocoolers ......................................................................4
1.2.2 Survey of Efforts to Improve the Regenerator.....................................................7
1.2.3 Development of the Volume Averaging Technique ..........................................15
1.2.4 Overview............................................................................................................16
2. Theoretical Development...............................................................................................19
2.1 Governing Equations and Constitutive Relations.....................................................19
2.1.1 Conservation of Mass Equation .........................................................................20
2.1.2 Balance of Momentum Equations......................................................................20
2.1.3 αPhase Conservation of Energy Equation........................................................21
2.1.4 βPhase Conservation of Energy Equation ........................................................22
vi
2.1.5 Equations of State ..............................................................................................22
2.1.6 αPhase Entropy Generation Equation ..............................................................23
2.1.7 βPhase Entropy Generation Equation...............................................................23
2.1.8 Summary of Equations.......................................................................................24
2.2 VolumeAveraged Equations ...................................................................................25
2.2.1 VolumeAveraged Conservation of Mass Equation ..........................................25
2.2.2 VolumeAveraged Balance of Momentum Equation.........................................26
2.2.3 VolumeAveraged αPhase Conservation of Energy Equation .........................26
2.2.4 VolumeAveraged βPhase Conservation of Energy Equation..........................27
2.2.5 VolumeAveraged αPhase Entropy Generation Equation................................28
2.2.6 VolumeAveraged βPhase Entropy Generation Equation................................28
2.3 Simplifying Assumptions .........................................................................................29
2.3.1 Density Spatial Deviation ..................................................................................29
2.3.2 Negligible Mechanical Dispersion.....................................................................31
2.3.3 Thermal Dispersion............................................................................................32
2.3.4 Entropy Generation Due to Thermal Dispersion ...............................................33
2.3.5 Negligible Brinkman Effect ...............................................................................34
2.3.6 One Dimensional Model on the Macroscopic Length Scale..............................34
2.4 Simplified Equations ................................................................................................35
2.4.1 Simplified Conservation of Mass Equation .......................................................35
2.4.2 Simplified Balance of Momentum Equation......................................................36
2.4.3 Simplified αPhase Conservation of Energy Equation......................................37
vii
2.4.4 Simplified βPhase Conservation of Energy Equation ......................................37
2.4.5 Simplified αPhase Entropy Generation Equation.............................................39
2.4.6 Simplified βPhase Entropy Generation Equation.............................................39
2.4.7 Summary of Equations.......................................................................................44
2.5 Exact Solutions.........................................................................................................46
2.6 Scale Analysis ..........................................................................................................49
2.6.1 Conservation of Mass Equation Scale Analysis.................................................50
2.6.2 Balance of Momentum Equation Scale Analysis...............................................50
2.6.3 αPhase Conservation of Energy Equation Scale Analysis ...............................52
2.6.4 βPhase Conservation of Energy Equation Scale Analysis................................54
2.6.5 Scaled Equation Summary .................................................................................55
2.6.6 Limiting Cases ...................................................................................................59
3. Computational Models...................................................................................................60
3.1 Problem Definition ...................................................................................................60
3.2 Numerical Method....................................................................................................61
3.2.1 Time Integration.................................................................................................62
3.2.2 Spatial Discretization .........................................................................................62
3.3 QuasiSteady Convergence via Cyclic Time Relaxation .........................................63
3.4 Boundary and Initial Conditions ..............................................................................66
3.5 The Constant Temperature Model (CTM)................................................................70
3.6 The Local Thermal Equilibrium Model (LTEM) .....................................................72
3.7 Dual Energy Equation Model (DEEM) ....................................................................74
viii
3.8 Model Verification ...................................................................................................75
4. Experimental apparatus and measurements ...................................................................84
4.1 Overview and Experimental Apparatus....................................................................84
4.2 Experimental Results................................................................................................94
5. Results and Discussion ................................................................................................105
5.1 Sage System Level Modeling.................................................................................106
5.2 Model Comparison .................................................................................................110
5.2.1 Net Enthalpy Flowrate  The Perfect Regenerator ...........................................110
5.2.2 The Mean Temperature Profile ........................................................................113
5.2.3 Baseline Regenerator Solutions .......................................................................114
5.2.4 First Law Results..............................................................................................118
5.2.5 Second Law Results .........................................................................................120
5.2.6 Sage and REGEN Comparisons.......................................................................122
5.2.7 Permeability and Forchheimer Coefficient Predictions ...................................125
6. Conclusions and Recommendations ............................................................................129
6.1 Conclusions ............................................................................................................129
6.2 Contributions ..........................................................................................................131
6.3 Future Work............................................................................................................132
Appendix 1 – Derivation of the VolumeAveraged Governing Equations......................136
A1.1 Volume Averaging Theory..................................................................................136
A1.1.1 Definitions.....................................................................................................136
A1.1.2 Transport Theorem........................................................................................138
ix
A1.1.3 Spatial Averaging Theorem ..........................................................................138
A1.1.4 Modified Averaging Theorem.......................................................................139
A1.2 Application to the Governing Equations .............................................................140
A1.2.1 VolumeAveraged Conservation of Mass Equation......................................140
A1.2.2 VolumeAveraged Balance of Momentum Equation....................................142
A1.2.3 VolumeAveraged αPhase Conservation of Energy Equation ....................145
A1.2.4 VolumeAveraged βPhase Conservation of Energy Equation.....................146
A1.2.5 VolumeAveraged αPhase Entropy Generation Equation...........................147
A1.2.6 VolumeAveraged βPhase Entropy Generation Equation ...........................148
Appendix 2 – Derivation of Differentiation Operators Using MatLab............................149
Appendix 3 – Compressor ElectroMechanical Modeling ..............................................154
A3.1 The electromechanical system.............................................................................154
A3.2 The Mechanical System.......................................................................................155
A3.3 The Electrical System..........................................................................................158
Bibliography ....................................................................................................................162
x
LIST OF FIGURES
Figure 1  Comparison Diagram of the PTC and Stirling Cryocooler.................................1
Figure 2 – A) Basic Pulse Tube Cryocooler and B) Orificed Pulse Tube
Cryocooler ........................................................................................................5
Figure 3  Bypass Pulse Tube Cryocooler ...........................................................................6
Figure 4  Thermohydraulic system of a typical porous media. The αphase is a
Newtonian fluid and the βphase is a solid.....................................................19
Figure 5 – Conceptual diagram illustrating the local velocity, velocity deviation,
and the volumeaveraged velocity. .................................................................30
Figure 6 – Regenerator computational domain diagram....................................................60
Figure 7 – Test case solution results. .................................................................................78
Figure 8 – The net enthalpy flux satisfies the test case to within visual accuracy
(0.2 mW deviation).........................................................................................79
Figure 9 – Maximum density error 8x10
6
kg/m
3
..............................................................80
Figure 10 – Maximum mass flux error 1x10
5
kg/m
2
s......................................................80
Figure 11 – Maximum energy error 2 kJ/m
3
......................................................................81
Figure 12 – Maximum matrix temperature error 6x10
6
K................................................81
Figure 13 – Maximum gas temperature error 8x10
6
K.....................................................82
Figure 14 – Maximum pressure error 1.2 Pa .....................................................................82
Figure 15  Experimental Apparatus..................................................................................85
xi
Figure 16 – Regenerator/Pulse Tube expander module. Regenerator is the larger
diameter section. .............................................................................................85
Figure 17 – Experimental apparatus with complete instrumentation. ...............................86
Figure 18 – Closeup view of the compressor. ..................................................................86
Figure 19 – Turbo vacuum pump station. ..........................................................................87
Figure 20  Wire Mesh.......................................................................................................90
Figure 21  Perforated Disk................................................................................................90
Figure 22  Foam Metal .....................................................................................................90
Figure 23 – Data acquisition system. .................................................................................91
Figure 24  Data acquisition and control program. Main panel view. ..............................92
Figure 25  Data acquisition and control program. Compressor panel view. ...................92
Figure 26  Data acquisition and control program. Pressure panel view. .........................93
Figure 27  Steady Flow Pressure Drop.............................................................................95
Figure 28  Load Curves ....................................................................................................97
Figure 29  Oscillatory Pressure Drop ...............................................................................99
Figure 30  Pressure Wave Phase Angle..........................................................................100
Figure 31  Pressure Ratio Attenuation............................................................................101
Figure 32 – Sintered glass regenerator.............................................................................104
Figure 33 – Sage diagram of the laboratory pulse tube apparatus. ..................................107
Figure 34  System level energy flow diagram; 400 mesh, 76 K, 0 W............................108
xii
Figure 35 – Comparison of regenerator loss calculated with limiting case models
based on identical operating conditions (ideal gas assumes constant
properties.) ....................................................................................................112
Figure 36 – Comparison of mean temperature profiles calculated with the
different models based on identical operating conditions. ...........................114
Figure 37 – Solutions plotted versus time and position. The middle plots are max
and min of density and velocity. Pressure and mass flow phase shifts
are apparent...................................................................................................115
Figure 38 – Boundary solutions for the gas and matrix temperatures and enthalpy
flow rates. .....................................................................................................116
Figure 39 – Cycleaveraged gas temperature (red) versus a linear profile (blue). ..........117
Figure 40 – Temperature difference versus position (multiple lines) and time. ..............118
Figure 41 – Mean Reynolds Number...............................................................................118
Figure 42 – Cycleaveraged mass flow rate.....................................................................119
Figure 43 – Cycleaveraged energy flows. Regenerator total energy flow is
constant along regenerator. ...........................................................................119
Figure 44 – Entropy generation due to viscous and inertial losses (top), and
entropy generation due to conduction and dispersion in the gas
(bottom). Multiple curves represent different locations in the
regenerator, plotted versus cycle time. Results for 400 mesh baseline
case. ..............................................................................................................120
xiii
Figure 45 – Entropy generation due to matrix conduction (top), and entropy
generation due to interfacial convection (bottom). Results for 400
mesh baseline case. .......................................................................................121
Figure 46 – Cycleaveraged volumetric entropy generation rates plotted versus
position in the regenerator. Results for 400 mesh baseline case. ................121
Figure 47 – Comparison of Sagepredicted regenerator loss versus GT model. .............123
Figure 48 – Friction factor comparison for 400 mesh screens.........................................128
Figure 49  Demonstration calculation for flow through wire mesh screens.
Visualization of surface pressure..................................................................133
Figure 50 – Demonstration calculation for flow through wire mesh screens.
Visualization of pathlines. ............................................................................133
Figure 51 – Voice coil compressor schematic .................................................................155
Figure 52 – Compressor piston freebody diagram (FBD)..............................................156
Figure 53 – Leach’s lossy inductor voice coil model ......................................................158
xiv
LIST OF TABLES
Table 1Summary of scale analysis ...................................................................................58
Table 2  Descriptive Summary of the Regenerators under Study ....................................88
Table 3 – Porous Media Parameter Summary (64)............................................................96
Table 4 – 400 mesh baseline operating conditions ..........................................................109
Table 5 – Sage Comparison Summary.............................................................................124
Table 6 – Permeability and Forchheimer Coefficient (compressible model) ..................126
Table 7 – Permeability and Forchheimer Coefficient (incompressible model)...............126
Table 8 – Permeability and Forchheimer Coefficient (neglecting advective
acceleration)..................................................................................................126
xv
NOMENCLATURE
Notation:
α Gas Phase
β Solid Phase
α
a Volume average of quantity a in the α phase
α
α
a Intrinsic volume average of quantity a in the α phase
aˆ Spatial deviation of quantity a
v
r
Vector of v
Variables:
αβ
A Area of the β α − interface (within the averagingvolume, V) [m
2
]
f
A Flow area [m
2
]
A Total cross sectional area, gas and solid [m
2
]
v
a Surface area per unit volume [1/m]
α
b
~
Thermal dispersion tensor closure variable
2 1
, C C Integration constants
p
c Gas constant pressure specific heat [J/kgK]
v
c Gas constant volume specific heat [J/kgK]
vs ps
c c , Specific heat scales [J/kgK]
xvi
f
c Forchheimer inertia coefficient
α
D
~
Thermal dispersion tensor
2
~
D Mechanical dispersion tensor, rank 2
r
D Regenerator diameter [m]
dS Differential surface area of
αβ
A [m
2
]
E Gas total energy per unit volume [J/m
3
]
E
~
Gas total energy temporal scale [J/m
3
]
3
~
E Mechanical dispersion tensor, rank 3
e e ,
α
Specific internal energy [J/kg]
F
~
Forchheimer tensor
F DarcyForchheimer surface integral
4 3 2 1
, , , F F F F Source terms for exact solution
f Friction factor, frequency [Hz]
4
~
G Mechanical dispersion tensor, rank 4
H Convective heat transfer coefficient [W/m
2
K]
h Enthalpy [J/kg]
h Enthalpy spatial scale [J/kg]
I
~
Identity tensor
K
~
Permeability tensor [m
2
]
K Permeability, in axial direction [m
2
]
xvii
β α
k k , Conductivity [W/mK]
r
L Regenerator length [m]
s
L Macroscopic length scale [m]
m m ,
α
Mass flux [kg/m
2
s]
m m
~
, Mass flux spatial and temporal scales [kg/m
2
s]
c h
m m , Hot and cold boundary mass flux amplitudes [kg/m
2
s]
c h
m m & & , Hot and cold boundary mass flow rates [kg/s]
k
N Axial conductivity enhancement
Nu Nusselt number
β α
n n , Normal vector
Pe Reference Peclet number
p p ,
α
Gas pressure [Pa]
p Pressure spatial scale [Pa]
b
p System baseline pressure [Pa]
m g
q q , Gas and matrix heat flux [W/m
2
]
R Gas constant, or radius of averaging volume [Pam
3
/kgK]
K
Re Reynolds number based on permeability length scale
St Reference Stanton number
β α , ,
,
gen gen
s s ′ ′ ′ ′ ′ ′ Volumetric entropy generation [W/m
3
K]
β α
s s , Entropy [J/kgK]
xviii
β α
T T , Temperature [K]
T Temperature spatial scale [K]
T
ˆ
Temperature difference scale [K]
c h
T T , Hot and cold boundary mean temperatures [K]
o
T Reference temperature, for lost power [K]
t Time [s]
s
t Time scale [s]
α
u
r
Velocity vector [m/s]
u u ,
α
Axial velocity [m/s]
V V V , ,
β α
Volume (of the averagingvolume) [m
3
]
lost
W
&
Lost power [W]
x Axial location [m]
Greek:
ρ ρ
α
, Density [kg/m
3
]
ρ ρ
~
, Density spatial and temporal scales [kg/m
3
]
( )
β
ρc Solid heat capacity [J/K]
mc mh
φ φ , Hot and cold boundary mass flux phase angles [rad]
α
φ Viscous dissipation function
µ µ
α
, Gas viscosity [ms]
xix
ε ε ε
β α
, , Porosity
ω Angular frequency [rad/s]
τ Time period [s]
β
τ Solid tortuosity
2 1
, Γ Γ Leading order dimensionless parameters
2 1
, ε ε Leading order dimensionless parameters
γ Specific heat ratio
xx
SUMMARY
In this study the phenomena of compressible flow and heat transfer in porous
media are modeled based on fundamental principles. The conservation equations for the
two phases are transformed by the method of volume averaging which is an analytical
method used to unite the microscale and macroscale effects characteristic to porous
media flows. Unique to this analysis is the fact that the model is valid for oscillatory,
cryogenic flows such as that occurring in a regenerative cryogenic refrigerator such as a
Pulse Tube Cryocooler (PTC.)
In a PTC the forced flow driven oscillations in the regenerator create Reynolds
numbers high enough such that microscale inertial effects dominate the momentum
equation. This phenomenon, known as the Forchheimer Effect, can be predicted and
modeled based solely on fundamental principles and the method of volume averaging.
The coefficients that characterize the Forchheimer momentum equation are determined
experimentally.
Heat transfer within a porous medium occurs due to temperature gradients in the
gas and solid phases. Conduction within the solid and fluid phases is made evident by
volume averaging, but the determination of the conductivity coefficients requires
numerical experiments and is unique to the geometry and conductivities of the two
phases. Convection between the two phases is the dominant mode of heat transfer within
the porous media. Determination of the convective heat transfer coefficient for a porous
media requires physical experiments.
xxi
Heat transfer due to temperature gradients and flow friction in the regenerator are
always competing effects leading to a model which requires coupling of the momentum
and energy equations. These competing effects are united with the concept of entropy
generation which relies on the second law of thermodynamics. All real processes
generate entropy, and the most efficient processes which balance flow friction and heat
transfer generate minimum entropy.
The theoretical model is presented with a numerical solution technique. These
numerical solutions are compared with similar solutions existing in the literature. The
uniqueness of this model is the completeness of the theoretical development and the
flexibility of use for a variety of applications. Numerical solutions are compared with
experimental data for an operating cryocooler.
1
CHAPTER 1
1. INTRODUCTION AND BACKGROUND
1.1 Introduction
Cryocoolers (refrigerators capable of cooling to temperatures below 120 K) have
long been classified into two categories based on the type of heat exchange process:
recuperative and regenerative (1). The Pulse Tube Cryocooler (PTC) and the Stirling
Cryocooler are two examples of regenerative cryocoolers (Figure 1), both operating on a
R
e
g
e
n
e
r
a
t
o
r
A
f
t
e
r
c
o
o
l
e
r
C
o
l
d
H
e
a
t
E
x
c
h
a
n
g
e
r
P
u
l
s
e
T
u
b
e
R
e
j
e
c
t
H
e
a
t
E
x
c
h
a
n
g
e
r
S
u
r
g
e
V
o
l
u
m
e
Pulse Tube Cryocooler
Stirling Cryocooler
C
o
m
p
r
e
s
s
o
r
O
r
i
f
i
c
e
W
comp
Q
aftercooler
Q
net
H
regen
H
pt
Q
reject
Pulse Tube Expander
R
e
g
e
n
e
r
a
t
o
r
C
o
m
p
r
e
s
s
o
r
W
comp
H
regen
Stirling Expander
Q
compressor
Q
net
W
expander
Figure 1  Comparison Diagram of the PTC and Stirling Cryocooler
2
modified Stirling thermodynamic cycle using helium as the working substance. Unlike
the Stirling thermodynamic cycle, the PTC and Stirling cryocooler cycles are not steady
flow processes. Instead, the gas flow oscillates in a quasisteady fashion due to the
motion of a compressor piston (and an additional expander piston in the Stirling
cryocooler).
In the Stirling cooler, refrigeration is produced by driving the expander piston
such that it receives mechanical work from the working fluid. If this work is sufficiently
large, the heat exchanger can absorb heat from the surrounding, thus producing
refrigeration. In the case of the PTC, the active expander piston is replaced with a
passive pulse tube, orifice, and surge volume. The pulse tube gas motion can be
controlled such that there is a gas “piston” which acts like the Stirling piston. If the
motion of the gas piston has the proper phase, it accepts work from the cold heat
exchanger and delivers this work to the reject heat exchanger where it is converted into
heat. This heat is rejected from the system in the reject heat exchanger.
The PTC is a unique type of regenerative cryocooler in that it does not have any
moving parts in the cold region. This is distinctly different from the Stirling cryocooler
that operates by an oscillating displacer directly in the cold region, resulting in the
potential for mechanical wear that increases mechanical complexity and can limit the life
of the cooler. Vibration in the Stirling cooler caused by the displacer also presents
problems for applications like sensitive infrared sensors, which cannot tolerate vibration.
Due to the absence of an expander piston in the cold region and the additional volume
3
after the regenerator, the pulse tube experiences much higher mass flow rates through the
regenerator. This results in a larger regenerator pressure drop.
The regenerator is a duct packed with some porous material. This porous material
is selected such that it has sufficient thermal heat capacity, high heat transfer coefficient,
and low flow friction. The designer of a cryocooler is chiefly concerned with achieving a
specified net refrigeration at a given temperature with a minimum input power. In the
design of a pulse tube cooler, the net refrigeration is the heat transfer rate from the
cryogenic device being cooled, which is equal to the total gross refrigeration produced
less internal parasitic losses in the cooler. Minimizing parasitic heat loads is critical to
achieving the design goals. One main system loss is the regenerator loss, or the cycle
averaged enthalpy flow at the cold end plus conduction losses in the gas and matrix.
These losses can be considerable, and may be quantified only via an accurate
mathematical model. The enthalpy loss is the most difficult quantity to estimate,
requiring an accurate prediction of the mass flow and temperature waveforms at the cold
end. It is not unrealistic to have an enthalpy loss on the order of one Watt with a peak
enthalpy flow rate of 1000 Watts, or 0.1%.
Regenerative cryocoolers are used in a variety of applications. The types of
regenerators being studied in this work are typically found in Stirling and pulse tube
cryocoolers, and other types of regenerative cryocooler applications. These devices are
typically used in applications which demand small net refrigeration (on the order of a few
watts) at temperatures below 100 K. Applications which require this type of refrigeration
are superconducting electronics, magnetic resonance imaging, and infrared focal plane
4
arrays. Several other applications include gas liquefaction of nitrogen,
magnetocardiography using Superconducting Quantum Interference Devices (SQUIDs),
minesweeping magnets, nondestructive evaluation using SQUIDs, outer space
experiments and instruments, and military weapon systems.
1.2 Background
The experimental apparatus used in this research is a Pulse Tube Cryocooler. For
this reason, it is necessary to provide a brief history of the PTC including efforts to
improve the performance of the PTC. The component being studied in the PTC is the
regenerator. A review of the efforts to improve regenerator performance includes studies
of materials and geometry. To understand the regenerator, investigators have used
experimental measurements and numerical modeling. The following discussion is a
review of these efforts.
1.2.1 History of Pulse Tube Cryocoolers
In 1966, Gifford and Longsworth first detailed the construction of a pulse tube
refrigerator (2, 3). The design of their pulse tube was essentially a tube with one end
closed and the other end open (in addition to a compressor and heat exchangers.) Both
ends had heat exchangers, and the open end was subjected to an oscillating pressure
through a regenerator, causing the open end to cool. This refrigerator is commonly
known as the Basic Pulse Tube Cryocooler (BPTC) (Figure 2.A). Not until 1984 was the
modern pulse tube cryocooler developed by Mikulin, Tarasov, and Shkrebyonock (4).
5
This cryocooler was equipped with an orifice and a surge volume on the warm end of the
pulse tube. The effect of this enhancement was to create an advantageous phase
difference between the oscillating pressure and velocity. This new design is commonly
called the Orifice Pulse Tube Cryocooler (OPTC) (Figure 2.B). Other enhancements
have been made such as porting the highpressure gas from the outlet of the compressor
to the warm end of the pulse tube which led to improved performance due to improved
phase control. This type of cryocooler is called a Bypass Pulse Tube Cryocooler (Figure
3). The drawback of this design is that the flow bypass can create a DC flow circulation
if the system is not designed properly. A very small flow circulation can lead to a huge
reduction in performance.
Compressor
Qcomp
Qreject
Qrefrig
Pulse
Tube
Regenerator
Compressor
Qcomp
Qreject
Qrefrig
Pulse
Tube
Regenerator
Surge
Volume
Orifice
(A)
(B)
Wcomp Wcomp
Figure 2 – A) Basic Pulse Tube Cryocooler and B) Orificed Pulse Tube Cryocooler
6
Figure 3  Bypass Pulse Tube Cryocooler
Developments in pulse tube technology have been occurring continuously up to
the present. Currently, three main areas of improvement are minimization of pulse tube
flow losses, phase control, and improving regenerator effectiveness. Reducing pulse tube
flow losses begins by being able to completely model the oscillating flow in the pulse
tube. The flow oscillations result in mass streaming and radial heat transfer, both of
which reduce the refrigeration possible with a pulse tube. Recently, several investigators
discovered that a proper tapering of the pulse tube leads to a complete elimination of
mass streaming (5).
As already mentioned in the preceding, the basic, orifice, and bypass pulse tube
all seek to manipulate the phase angle between the pressure and mass flow to an optimum
angle at which the maximum refrigeration is produced. One recent improvement is the
Compressor
Qcomp
Qreject
Qrefrig
Pulse
Tube
Regenerator
Surge
Volume
Orifice
Wcomp
7
inertance tube (6). The inertance tube improves the orifice design by adding an inertance
to the fluidic system. This improvement not only increases the phase angle, but also
eliminates the possibility of flow circulations which can occur with the bypass design.
Design of an optimum phase control device and reducing pulse tube losses will
eventually make the pulse tube cryocooler more competitive than the Stirling cryocooler
for most applications due to significantly lower cost and comparable efficiency.
1.2.2 Survey of Efforts to Improve the Regenerator
Extensive efforts have been focused on improvement in regenerator technology
since the development of regenerative cryocoolers. These efforts have been categorized
into areas of materials and geometry, modeling, and measurement.
The problems with designing the optimum regenerator have interesting
complications as the temperature of the cold end decreases. These complications arise in
factors such as variation of thermal properties over the huge temperature range of the
regenerator. Orders of magnitude decrease in thermal capacity of the material from
300 K to the cold temperature make different materials attractive in different areas of the
regenerator. The basic problem focuses on increasing the heat transfer effectiveness
between the gas and solid in the regenerator. The heat transfer effectiveness is a function
of fluid properties, solid properties, and the flow geometry. Any change in these
parameters will affect the regenerator performance.
At temperatures below 20 K, the solid properties appear to be a major contributor
to the overall regenerator performance. This has led to a development of exotic materials
8
and processes so that the regenerator has sufficient thermal capacity in the cold region.
Several erbium alloys have been tested, but these magnetic alloys are brittle at cryogenic
temperatures. Pecharsky, et al. at Ames Laboratory have investigated titanium alloying
of a popular erbium alloy, Er
3
Ni (7). This alloy has been investigated by many as a
regenerator material for GiffordMcMahon (GM) and Stirling coolers operating below
10 K. The significant result was that this particular alloying increased the ductility
without significantly decreasing the specific heat of the alloy. Gshneidner, et al. also at
Ames Laboratory have suggested Er
6
Ni
2
Sn alloy for the first stage of a GM cooler (8).
Their main effort was to develop a process to generate this alloy. The results showed that
practical issues such as particle escape from the regenerator and settling will prevent this
alloy from being used in the near future. Again, the brittle properties of this alloy have
limited its use. Bradshaw, et al. have performed experiments using a variety of different
materials in several combinations to study their effects in a two stage Stirling cooler
operating in the 15 K range (9). The materials used were gold wire, lead wire, Er
3
Ni and
stainless steel mesh. At the lowest operating temperatures, they found that the lead and
Er
3
Ni performed the best, and these materials were also found to have the highest specific
heat at that temperature. Takashi, et al. have used a slightly different erbium alloy
(ErNi
0.9
Co
0.1
) along with Er
3
Ni in a large 2.2 W cooling capacity, 4.2 K GM cooler with
12 kW compressor input (10). Their conclusion was that the combination of ErNi
0.9
Co
0.1
and Er
3
Ni increased the cooling capacity by a factor of 1.2 from a regenerator with just
Er
3
Ni. The ErNi
0.9
Co
0.1
has a dramatic increase in specific heat around 6 K of
approximately 1 J/Kcm
3
in contrast to 0.4 J/Kcm
3
for Er
3
Ni. Chafe, et al. have replaced
9
lead balls with Neodymium plates and balls in the second stage of a 10 K GM cooler
application to utilize the higher heat capacity at the low temperatures (11). They have
also taken advantage of a perforated plate geometry that reduces pressure drop in
comparison to spheres. As a result, they were able to progress from 8 K down to 4 K
under similar operating conditions.
The geometry of the pores in the regenerator matrix determine the pressure drop
within the regenerator. Since pressure drop and heat transfer are coupled, a certain
amount of pressure drop is required to achieve effective regeneration. Efforts to
minimize the pressure drop while still having adequate heat transfer have been reported
in the literature. Measurements of steady pressure drop and correlating these
measurements to friction factors have been investigated by many. However, many
investigators put too much confidence in the importance of steady flow friction factors
(12, 13, 14). The correlations given by Kays and London, for example, do not correlate
well for oscillatory flow pressure drop and heat transfer due to additional effects such as
enhanced dispersion due to the oscillations (15). Organ in Thermodynamics and Gas
Dynamics of the Stirling Cycle Machine states this argument repeatedly (16).
There is something less than satisfactory about the way in which Stirling machine
analysis handles flow within the regenerator:
(1) The flow case is treated by the method traditional for steady, twodimensional
(or axisymmetric), incompressible viscous flow in pipes, i.e., in terms of a
friction factor, C
f
, correlated with geometry and Reynolds number, N
re
. When
analysis and computer simulation based on such correlations yield pressure
distributions which do not tally with measurement from running machines, it is
common practice to 'improve' matters by arbitrarily adjusting the correlations.
The technique is part of a process which has become known as 'validation'.
Exercises in validation have been reported
13
which called for C
f
at given Nre to
be multiplied by factors between 4 and 7.
10
(2) The discrepancy between experimental measurement and theoretical prediction
has come to be attributed to the fact that steadyflow correlations do not take
into account the unsteady effects which arise from the cyclic nature of the flow
processes in the Stirling machine. An enquiry into the role of unsteadiness is
certainly called for. At the same time, usage and interpretation of the steady
flow correlations has been parochial, having in most instances looked no further
than the incompressibleflow cases documented by Kays and London.
4
Some investigators have made the realization that steady flow friction factor correlations
are not adequate, but there is still a gap between their measurements and the physical
phenomena occurring within the regenerator (16, 17, 18, 19, 20, 21). Organ makes a
clear argument that the 'steady' flow through the porous media is not a true description
due to inherent oscillatory fluctuations at the natural frequency of the pore structure.
This indicates that there are local accelerations within the fluid even for unidirectional,
'steady' flow. This type of argument shows that there are effects other than the inertial,
½ρu
2
, effect which is the conventional basis for correlating friction factor with Reynolds
number. Organ claims that the pressure drop is a result of independent viscous, inertial,
and compressibility effects. In a separate reference, Organ develops the concept of the
regenerator flow impedance, which is based on a linear wave model (22). This concept
provides a method to better explain the physical phenomena in the regenerator. The
research being conducted by Roberts and Desai (23) are motivated by the concerns
voiced by Organ.
It will be seen in later Chapters that steady flow measurements can be used
effectively to predict friction factors which are accurate for steady flow and oscillating
flow. The compressible nature of the gas flowing through the regenerator matrix leads to
a nonconstant pressure gradient. This requires that experimental data be used as
11
boundary conditions for a differential model rather than a lumped model which is
typically used.
Theoretical modeling of the regenerator and general porous media flows have
been conducted by many, and can be found under a variety of headings. Watson used an
exact solution for incompressible flow in a tube to show how diffusion can be enhanced
(24). While Watson was not interested in the regenerator problem, his research illustrates
the basic phenomenon which occurs in the regenerator and the pulse tube. For zero net
flow, the flow oscillations produce a positive net energy transport from the cold end to
the warm end of a channel. Two authors have made direct extensions of Watson’s initial
investigation. Siegel’s analysis applied this directly to the problem of the regenerator
(25, 26). Although the flow in the regenerator is not incompressible, the results are
interesting since he has focused explicitly on porous media flows. Siegel attributes the
axial transport to the transverse conduction which occurs between adjacent fluid layers.
Kaviany has investigated the effect of oscillatory flows on heat exchangers composed of
tube bundles (27, 28). In this analysis, Kaviany considers the case of enhanced heat
diffusion between two reservoirs due to the oscillations. His analysis is an important
extension of Watson’s work since he considers the coupled problem of the gas and the
wall energy. While Watson assumes zero wall heat flux, Kaviany solves for the wall
temperature directly. The solutions obtained are exact solutions for the velocity and
temperatures. Zhang, et al. have applied the method of characteristics to solve the 1D
pulse tube governing equations (29). Preliminary results indicate that their method over
predicts the performance based on experimental results.
12
The concept of a complex Nusselt number is developed by Kornhauser and
Smith (30). In this study, the instantaneous heat flux in a compression cylinder is
observed to be out of phase with the bulk temperature difference. This indicates that the
traditional Newton’s law of cooling does not hold for oscillating flow above a critical
oscillation frequency, which is used to define an oscillating Peclet number. They find
that a complex formulation of Newton’s law of cooling is applicable. This requires
measurement of the real and imaginary component of the Nusselt number. These studies
were conducted in a compression cylinder, but the authors allude to the fact that this can
be extended to the compression and flow process occurring in the regenerator. The
critical oscillation frequency increases proportional to the inverse of the square of the
hydraulic diameter. The hydraulic diameter of the regenerator can be 5 to 6 orders of
magnitude below that of the compression space used. Thus, it is expected that the
imaginary component of the regenerator Nusselt number would be quite small for typical
frequencies used in Stirling and PTC cooler.
Bauwens has proposed a 2D model for the pulse tube and regenerator that can be
approximated only for small fluctuations using perturbation methods (31). While the
approximate solution that he obtains is interesting, the model could yields invalid results
for a pulse tube cooler due to relatively large pressure oscillations. At large pressure
ratios, the linearity assumption is no longer valid, requiring a fully nonlinear model. The
fully nonlinear model ultimately requires a numerical approximation rather than the
analytical solution which he obtains.
13
Roach, et al. also propose a perturbation solution to the regenerator problem (32).
Their results indicate that the velocity and pressure fluctuations do not depend on the
thermal interaction between the gas and solid. Conversely, the thermal interaction does
depend on the velocity and pressure fluctuation. This allows solving for the velocity and
pressure and then using those solutions to solve for the temperature. It does not appear
that they have provided any experimental comparison. This model has been described in
a separate publication by Kashani and Roach (33). The method has been incorporated
into a program called ARCOPTR.
A finite difference program called REGEN3.2 (34) has been developed by the
National Institute of Standards and Technology, and has been used to study the effects of
regenerator geometry by Kuriyama, et al. (35). This model assumes that the pressure in
the regenerator is uniform and oscillates in time. The effect of the pressure gradient is
included as a correction. The mathematical model proposed by REGEN involves solving
a system of equations for the density, velocity, gas temperature, and matrix temperature.
The velocity field is found using an explicit equation derived from the continuity
equation assuming zero pressure gradient. As a result, the velocity and density are
essentially determined from a single equation. This leads to a nonconservative model.
While the negligible pressure gradient assumption may be accurate for some cases such
as Stirling regenerators, the higher flow rates found in pulse tube regenerators lead to
inaccuracies in the REGEN model. The pressure phase shift across an optimally
designed Stirling regenerator is typically small compared to a pulse tube regenerator.
This leads to a smaller pressure gradient in the Stirling regenerator.
14
Currently, the most comprehensive model for cryocooler systems is Sage (36).
This model provides the user the ability to model the entire cryocooler and carry out
optimization studies. Although this model appears complete, Harvey, et al. were
concerned with the accuracy of assumptions and numerical methods in Sage (37).
Apparently, the pulse tube correlations lack accuracy for small pulse tube. Additionally,
the numerical scheme employed by Sage involves approximating the solutions using
Fourier series. The solution is found such that the equations are satisfied at as few as 6
time nodes in the cycle. If the actual solution cannot be accurately described using 6 time
nodes, then the Sage model will be limited in accuracy. Further inaccuracies can be
attributed to a non conservative spatial discretization method based on first order control
volumes. While this model may have some inaccuracies, it has been found to be a
valuable tool for the cryocooler designer because of the builtin optimization tool.
Measurements within the regenerator matrix are inherently difficult due to the
small geometry of the pores. High frequency response sensors must be used to accurately
resolve the oscillating temperatures and velocities. Several individuals have made some
interesting and important measurements on the regenerator and pulse tube. Yuan and
Dybbs have developed a method to measure both the gas and solid temperature
fluctuations in the regenerator of a Stirling engine (38). This method utilizes small
thermocouples whose responses are compensated to correct for attenuation and lag.
Their results are well matched to numerical predictions. Rawlins, et al. have used hot
wire anemometer probes in key locations in a pulse tube cooler (39, 40). Oscillatory
velocities and temperatures have been measured at both the warm and cold ends of the
15
pulse tube and regenerator. These measurements allowed them to calculate from
pressure, temperature, and mass flow rates, the instantaneous energy and entropy flows in
the pulse tube. The efforts to make these measurements are remarkable since the
conditions, especially in the cold region, are certainly not hospitable to making hotwire
measurements. In fact, these measurements are some of the most sophisticated
measurements made with hotwires. Typical measurements in wind tunnels have only
small temperature fluctuations about the ambient. These measurements were performed
at temperatures below 90 K with fluctuations of 5 K. Installing the probes in a high
pressure leak free fitting was also a feat. Direct measurement of the flows in the
cryocooler provides very valuable data for validating an numerical model.
1.2.3 Development of the Volume Averaging Technique
The volume averaging technique is an analytical tool for describing the flow and
heat transfer in a porous media. This technique has found extensive uses in ground water
and pollution transport science, petroleum reservoir modeling, catalytic reactors, and
fluidized beds to mention a few.
Hassanizadeh and Gray point out that there are at least three methodologies for
describing the flow and heat transfer in multiphase systems, some of which rely mainly
on intuition and empirical observations (41). While these methodologies have led to
some of the original models, such as the Darcy model, the volume averaging technique
provides a formal framework for improving the science of porous media. All of the
current regenerator models in the open literature, such as REGEN and Sage, rely on the
16
intuitive and empirical knowledge of the flow in the regenerator without any application
of the volume averaging technique. As a result, these models fall short of describing in
an exact fashion the flow and heat transfer in the regenerator.
Whitaker has been fundamental in the development of the volume averaging
technique and it application to a variety of problems; diffusion and dispersion in a
reactor, conduction in multiphase systems, development of conditions for
nonhomogeneous porous media, local numerical studies and experimental validations to
investigate the validity of volume averaging closure conditions (42). Whitaker’s studies
have mainly focused on incompressible flow, but he has briefly talked about the case of
slightly compressible flow (43). Use of the volume averaging in this dissertation is
applied to a problem which is highly compressible due mainly to the large pressure
oscillations and the large temperature gradient across the regenerator.
1.2.4 Overview
To this end, the abundance of assumptions and modeling techniques for
regenerators creates the need for a systematic study of the phenomenon based on
fundamental conservation principles. This dissertation describes the derivation of the
macroscopic equations which govern the regenerator problem. These equations are
derived from the local governing equations for a generalized, compressible, real fluid.
These equations are referred to as local equations since they describe the flow of the fluid
within the pores. Thus, they are also referred to as the microscale equations. The
microscale equations are then transformed into a set of macroscale governing equations
17
using volumeaveraging. In this form, the governing equations describe the macroscopic
flow behavior in addition to the effect that the microscale flow has on the macroscopic
flow. The theory and details of the generalized volumeaveraging technique is included
in Appendix 1. Chapter 2 begins by summarizing the microscopic and macroscopic
governing equations as developed in Appendix 1. Assumptions are then developed and
discussed which allow the governing equations to be simplified. Closure relationships
for friction, heat transfer, and dispersion are then developed to reduce the equation set to
a tractable problem definition. The chapter concludes with a presentation of a set of
illustrative exact solutions and scale analysis.
Chapter 3 details the development of a series of computational models. Several
models are developed which can be used to study several limiting assumptions such as
constant temperature and local thermal equilibrium. The numerical method used to solve
these models is the Method of Lines. Of significant importance is the development of the
artificial convergence technique which allows the problem to be converged rapidly. The
chapter concludes with the development and results of an exact solution verification.
Chapter 4 details the development of the experimental apparatus and data.
Regenerator steady flow data and system level cryocooler performance data is
summarized and discussed.
Results and discussion are included in Chapter 5. A system level model is
discussed which allows the regenerator boundary conditions to be estimated. The
numerical results of the detailed regenerator model are presented and compared with the
system level model. The limiting models which were developed in Chapter 3 are
18
compared with the full regenerator model. A comparison of two regenerator models from
the literature, including the system level model, is provided for the baseline regenerator.
Steady flow pressure drop data is used to predict friction factors using a compressible
flow exact solution to the governing equations. This friction factor is compared with
friction factors measured in oscillating flow yielding excellent agreement.
Chapter 6 concludes this dissertation. The important results are summarized.
Future research topics are motivated and discussed.
19
CHAPTER 2
2. THEORETICAL DEVELOPMENT
2.1 Governing Equations and Constitutive Relations
The development of a theoretical model for flow and heat transfer in a generalized
porous medium begins with a set of governing equations and constitutive relations for the
thermohydraulic system illustrated in Figure 4. In this system, there is a solid and a fluid
phase. The solid phase is assumed to be stationary and rigid with known thermal
properties which are functions of temperature. The fluid phase is assumed to behave as a
linearly viscous fluid. The fluid flow is assumed to be compressible, and the fluid
properties are known functions of temperature and pressure, or another combination of
state properties. The governing equations are derived in any fluid mechanics text and are
repeated here.
α
• x
α
β
• x
β
n
α
Averaging volume, V
Figure 4  Thermohydraulic system of a typical porous media. The αphase is a
Newtonian fluid and the βphase is a solid.
20
2.1.1 Conservation of Mass Equation
The differential statement of mass conservation for the αphase (in strong
conservation form) is
( ) , 0 = ⋅ ∇ +
∂
∂
α α
α
ρ
ρ
u
t
r
(21)
where ρ
α
is the gas density and
α
u
r
is the gas velocity vector. This equation is a scalar
equation with 4 unknowns; the fluid density and three components of the fluid velocity.
2.1.2 Balance of Momentum Equations
The differential statement of the balance of momentum for a Newtonian fluid with
no body forces (in strong conservation form) is
( ) ( ) ( ) ( ) , 0
3
1
= ∇ ⋅ ∇ − ⋅ ∇ ∇ − ∇ + ⋅ ∇ +
∂
∂
α α α α α α α α α α
µ µ ρ ρ u u p u u u
t
r r r r r
(22)
where
α
µ is the gas viscosity coefficient, and p
α
is the mechanical pressure. This
equation is a vector equation with three components corresponding to the three
components of velocity. The balance of momentum equation produces 1 additional
unknown, the fluid pressure. The viscosity is a material property and is thus known from
experimental data.
21
2.1.3 αPhase Conservation of Energy Equation
The differential statement of the conservation of energy for a Newtonian fluid in
terms of the gas specific internal energy, e
α
, is
( )
( ) , 0 = ∇ − ⋅ ∇ +
∂
∂
α α α α α
α α
ρ
ρ
T k h u
t
e r
(23)
where h
α
is the gas enthalpy defined as
.
α
α
α α
ρ
p
e h + = (24)
It should be noted that Equation 23 is in strong conservation form. This is the preferred
form for the conservation equations. In this form, the gradient of the enthalpy flow
includes viscous dissipation, although it is not immediately obvious. Expanding this term
gives
( ) ( )
( ) .
α α α α α α α
α α α α α α α α
ρ
ρ ρ
p u u p e u
u p e u h u
∇ ⋅ + ⋅ ∇ + ⋅ ∇ =
+ ⋅ ∇ = ⋅ ∇
r r r
r r r
(25)
Using Equation (22), the pressure gradient can be eliminated by solving in terms of the
viscous and acceleration terms. It can then be shown that
α α α α
φ µ − = ∇ ⋅ p u
r
(26)
where
α
φ is the viscous dissipation function defined as
( ) ( ) ( ) ( ). : : 2 :
2
1
3
2
2 T T
u u u u u u u
α α α α α α α α
φ
r r r r r r r
∇ ∇ + ∇ ∇ + ∇ ∇ + ⋅ ∇ − = (27)
The double dot notation represents the scalar product of two tensors, and the superscript
“T” denoted the tensor transpose.
22
2.1.4 βPhase Conservation of Energy Equation
For the βphase, the conservation of energy equation is written as
( ) ( ) , 0 = ∇ ⋅ ∇ −
∂
∂
β β
β
β
ρ T k
t
T
c (28)
where (ρc)
β
is the solid heat capacity per unit volume. Both energy equations are scalar
equations introducing an additional 3 unknowns; the fluid and solid temperatures, and the
fluid internal energy.
2.1.5 Equations of State
Thus far, the system of equations consists of six equations for eight unknowns,
requiring two additional equations to close the problem. These equations are the gas and
caloric equations of state which can be generally expressed as
( )
α α α
ρ e f p ,
1
= (29)
and
( ). ,
2 α α α
ρ e f T = (210)
The particular equations of state are arbitrary, and do not effect the form of the governing
equations. Ideal gas equations or real gas equations can be used.
23
2.1.6 αPhase Entropy Generation Equation
The differential statement of the second law of thermodynamics is given as
( )
( ) 0
,
≥


.

\

⋅ ∇ + ⋅ ∇ +
∂
∂
= ′ ′ ′
α
α
α α α
α α
α
ρ
ρ
T
q
u s
t
s
s
gen
r
r
(211)
where
α , gen
s ′ ′ ′ represents the gas volumetric rate of entropy generation. The entropy, s
α
, is
a thermodynamic property which is fundamental to optimizing any thermodynamic
process. The inequality indicates that the entropy generation is always positive except for
totally reversible processes, in which case, it is zero. It will be shown later that the
entropy generation for the case of a porous media can be represented by three effects;
local conduction due to molecular diffusion and dispersion, gastomatrix convective heat
transfer through a finite film temperature difference, and flow losses due to viscous and
inertial effects.
2.1.7 βPhase Entropy Generation Equation
For the βphase, the entropy generation equation is written as
( )
. 0
,
≥


.

\

⋅ ∇ +
∂
∂
= ′ ′ ′
β
β β β
β
ρ
T
q
t
s
s
gen
r
(212)
The entropy generation in the solid phase is caused by local conduction due to molecular
diffusion and gastomatrix convective heat transfer through a finite film temperature
difference.
24
2.1.8 Summary of Equations
The system of equations represents 8 equations and 8 unknowns. Repeated here
is this system of equations.
( ) 0 = ⋅ ∇ +
∂
∂
α α
α
ρ
ρ
u
t
r
(213)
( ) ( ) ( ) ( ) 0
3
1
= ∇ ⋅ ∇ − ⋅ ∇ ∇ − ∇ + ⋅ ∇ +
∂
∂
α α α α α α α α α α
µ µ ρ ρ u u p u u u
t
r r r r r
(214)
( )
( ) 0 = ∇ − ⋅ ∇ +
∂
∂
α α α α α
α α
ρ
ρ
T k h u
t
e r
(215)
( ) ( ) 0 = ∇ ⋅ ∇ −
∂
∂
β β
β
β
ρ T k
t
T
c (216)
( )
( ) 0
,
≥


.

\

⋅ ∇ + ⋅ ∇ +
∂
∂
= ′ ′ ′
α
α
α α α
α α
α
ρ
ρ
T
q
u s
t
s
s
gen
r
r
(217)
( )
. 0
,
≥


.

\

⋅ ∇ +
∂
∂
= ′ ′ ′
β
β β β
β
ρ
T
q
t
s
s
gen
r
(218)
( )
α α α
ρ e f p ,
1
= (219)
( ). ,
2 α α α
ρ e f T = (220)
The unknowns are the local, instantaneous, nonvolume averaged gas density, gas
velocity, gas internal energy, gas temperature, matrix temperature, gas and matrix
entropy generation, and gas pressure. These equations represent the microscale flow in
the porous media. The flow geometry is far too complicated to allow for a direct
application of these equations for any large scale porous system such as the regenerator.
25
The flow is best analyzed in terms of volumeaveraged quantities. The derivation of the
volumeaveraged governing equations is the topic of Section 2.2.
2.2 VolumeAveraged Equations
The details of the volume averaging method are included as Appendix 1 in this
document. The resulting equations are repeated here.
2.2.1 VolumeAveraged Conservation of Mass Equation
The volume average of Equation (213) is
( ) . 0
ˆ
ˆ
1
= ⋅ ∇ + ⋅ ∇ +
∂
∂
α α
α
α
α
α
α
α
α
ρ
ε
ρ ρ u u
t
r r
(221)
where
α
ε is the gas phase volume fraction, or porosity as defined in Appendix 1. The
notation indicates a volume average which is also defined in Appendix 1. The “hat”
notation indicates a spatial deviation quantity. The volume averaging results in an
additional term to the standard continuity equation representing mass dispersion. This
term will be addressed later.
26
2.2.2 VolumeAveraged Balance of Momentum Equation
The volume average of Equation (214) is
( ) ( ) ( ) ( )
( ) ( )
( ) . 0
ˆ ˆ
ˆ
1
ˆ
ˆ
1
ˆ
ˆ
1
ˆ ˆ
1
ˆ
ˆ
1
ˆ ˆ
3
ˆ
~ 1
3
1
= ⋅ ∇ + ⋅ ∇ +
⋅ ∇ + ⋅ ∇ +
∂
∂
+
(
¸
(
¸
∇ −

.

\

⋅ ∇ − ⋅ +
∇ ⋅ ∇ − ⋅ ∇ ∇ − ∇ + ⋅ ∇ +
∂
∂
∫
α α α
α
α
α α α
α
α α
α
α
α
α α
α
α
α
α α
α
α α α
α
α α
α
α
α α
α
α α
α
α
α
α
α
α
α
α
α
α
α
α
ρ
ε
ρ
ε
ρ
ε
ρ
ε
ρ
ε
µ
µ
µ µ ρ ρ
αβ
u u u u
u u u u u
t
dS u u p I n
V
u u p u u u
t
A
r r r r
r r r r r
r r
r r r r r
(222)
The volume averaging has produced multiple terms which need to be eventually
represented as functions of the volumeaveraged variables or eliminated justifiably.
2.2.3 VolumeAveraged αPhase Conservation of Energy Equation
The volume average of Equation (215) is
( )
( )
( )
( ) ( ) . 0
ˆ ˆ
ˆ
1
ˆ
ˆ
1
ˆ ˆ
1
ˆ
ˆ
1
ˆ ˆ
1 1
ˆ
1
= ⋅ ∇ + ⋅ ∇ + ⋅ ∇ +
⋅ ∇ +
∂
∂
+ ∇ ⋅ −
(
¸
(
¸


.

\

+ ∇ ⋅ ∇ − ⋅ ∇ +
∂
∂
∫
∫
α α α
α
α
α α α
α
α α
α
α
α
α
α α α
α
α α
α
α α α
α
α α
α
α
α α
α
α
α
α
α
α
α
α
α
α
ρ
ε
ρ
ε
ρ
ε
ρ
ε
ρ
ε
ρ
ρ
αβ
αβ
h u h u h u
h u
t
e
dS T k n
V
dS T n
V
T k h u
t
e
A
A
r r r
r
r
(223)
Again, the volume averaging has produced many additional terms which will be address
in the following pages. The terms of the original energy equation have survived, except
these terms are now in the form of volumeaveraged quantities.
27
2.2.4 VolumeAveraged βPhase Conservation of Energy Equation
The volume average of Equation (216) is
( )
. 0
1
ˆ
1
= ∇ ⋅ −
(
(
¸
(
¸


.

\

+ ∇ ⋅ ∇ −
∂
∂
∫
∫
αβ
αβ
β β β
β
β β
β
β
β β
β
β
β
ρ
A
A
dS T k n
V
dS T n
V
T k
t
T
c
(224)
Note that that the two energy equations, (223) and (224), are coupled via surface
conduction heat transfer terms in the form of surface integrals over the αβ interface area.
Later it will be shown that the surface integrals,
∫
∇ ⋅ −
αβ
α α α
α
A
dS T k n
V
1
(225)
and
,
1
∫
∇ ⋅ −
αβ
β β β
β
A
dS T k n
V
(226)
represent the convection between the two phases.
28
2.2.5 VolumeAveraged αPhase Entropy Generation Equation
The volume averaging of Equation (217) is
( )
( ) ( ) ( )
( )
∫
∇
⋅ −
(
(
¸
(
¸
∇
⋅ ∇ −
⋅ ∇ + ⋅ ∇ +
⋅ ∇ + ⋅ ∇ + ⋅ ∇ +
∂
∂
+
∂
∂
= ′ ′ ′
αβ
α
α α
α
α α
α
α
α
α α α
α
α
α α α
α
α α
α
α
α
α
α α α
α
α
α
α
α
α
α
α α
α
α
α
α
α
α
α
ε
ρ
ε
ρ
ε
ρ
ε
ρ
ε
ρ
ρ
ε
ρ
A
gen
dS
T
T k
n
V T
T
k
u s s u
u s u s u s
t
s
t
s
s
1 1
ˆ
ˆ ˆ
1
ˆ
ˆ
1
ˆ
ˆ
1
ˆ ˆ
1
ˆ ˆ
1
,
r r
r r r
(227)
2.2.6 VolumeAveraged βPhase Entropy Generation Equation
The volume average of Equation (218) is
∫
∇
⋅ −
(
(
¸
(
¸
∇
⋅ ∇ −
∂
∂
= ′ ′ ′
αβ
β
β β
β
β β
β
β
β
β
β
β
β
β
ε
ρ
A
gen
dS
T
T k
n
V T
T
k
t
s
s
1 1
,
. (228)
As with the energy equations, the entropy generation equations now contain terms which
will be shown later to represent entropy generation due to solidtogas convection
through a film temperature difference,
∫
∇
⋅ −
αβ
α
α α
α
α
A
dS
T
T k
n
V
1
(229)
and
,
1
∫
∇
⋅ −
αβ
β
β β
β
β
A
dS
T
T k
n
V
(230)
as well as entropy generation due to molecular diffusion,
29
(
(
¸
(
¸
∇
⋅ ∇ −
α
α
α
α
ε T
T
k
1
(231)
and
.
1
(
(
¸
(
¸
∇
⋅ ∇ −
β
β
β
β
ε T
T
k (232)
2.3 Simplifying Assumptions
At this point in the development of the volumeaveraged equations, the assumptions
are:
1) The solid phase is stationary and nondeforming.
2) The porosity is constant.
3) The fluid phase satisfies the noslip condition on the fluidsolid interface.
4) Fluid and solid properties, such as conductivity and viscosity, can be treated as
locally constant with respect to the averaging volume. Properties are assumed to
vary with the intrinsicaveraged state variables.
To facilitate the equations to be in a tractable form, all terms containing deviation
quantities must be expressed as functions of the averaged variables.
2.3.1 Density Spatial Deviation
The pore scale velocity varies across the cross section of a pore due to the noslip
Dirichlet boundary condition at the fluidsolid surface (see Figure 5.) This results in an
order of magnitude estimate for the velocity deviation, . ~
ˆ
α
α α
u u
r r
30
( ) t x u ,
r r
( ) t x u ,
ˆ
r r
α
( ) t x u ,
r r α
α
α
β
α
α α
u u
r r
~
ˆ
( ) t x u ,
r r
( ) t x u ,
ˆ
r r
α
( ) t x u ,
r r α
α
α
β
α
α α
u u
r r
~
ˆ
Figure 5 – Conceptual diagram illustrating the local velocity, velocity deviation, and
the volumeaveraged velocity.
Since density does not satisfy a Dirichlet boundary condition at the fluidsolid interface,
the density deviation is small compared to the volumeaveraged density (43). The
consequence of this assumption is
{
. ˆ
α
α α
α
α α
ρ ρ ρ ρ ≅ + =
negligible
(233)
The consequences of this assumption can be illustrated by considering the volume
average of the ideal gas law which gives
31
.
ˆ
ˆ
α α
α
α
α
α
α
α
α
ρ
ε
ρ T
R
T R p + = (234)
Neglecting the density deviation eliminates the last term in Equation (234), even though
the temperature deviation is not negligible. This assumption will effectively eliminate a
large number of terms in the volumeaveraged equations. This conclusion is true for
ideal or real gases.
2.3.2 Negligible Mechanical Dispersion
The mechanical dispersion term appearing in the momentum and energy
equations is
( ) .
ˆ ˆ
α α
α
α
ρ u u
r r
⋅ ∇ (235)
Gray and O’Neill (44) identify the mechanical dispersion term appearing in the
momentum equation. They propose that this term can be expressed as
α
α
α
α
α
α
α
α α α
u u G u E u D u u
r r r r r r
:
~ ~ ~
ˆ ˆ
4 3 2
+ ⋅ = = (236)
where the numerical subscripts represent the rank of the tensors. Whitaker (45) shows by
scale analysis that the dispersion term for such flows is, in general, negligibly small in
comparison to the surface integral which represents the Darcy and Forchheimer effects.
Thus, this analysis will neglect the mechanical dispersion terms.
32
2.3.3 Thermal Dispersion
The thermal dispersion term in the fluid energy equation is
( ) .
ˆ ˆ
α α
α
α
ρ h u
r
⋅ ∇ (237)
This term can be at least an order of magnitude larger than the molecular diffusion term.
Whitaker shows in an analogous fashion, that this dispersion is diffusive at the
macroscopic scale and the dispersion coefficient is proportional to the Peclet number
(42). Taking the molecular diffusion term and the dispersion term together, Whitaker
suggests that these terms reduce to
( )
α
α α
α
α α
α
α α α α α
ρ ε ε
αβ
h D T dS b n
V
k
A
∇ ⋅ ⋅ ∇ −
(
¸
(
¸
∇

.

\

+ Ι ⋅ ∇ −
∫
~ ~ 1 ~
(238)
where the dispersion tensor is
α
D
~
. The closure variable,
α
b
~
, must be solved for
numerically. The effective conductivity is then defined as
.
~ 1 ~
,

.

\

+ Ι =
∫
αβ
α α α α
A
eff
dS b n
V
k k (239)
Equation (238) can then be written as
( )
α
α α
α
α α
α
α α α
ρ ε ε h D T k
eff
∇ ⋅ + ∇ ⋅ ∇ −
~
,
(240)
which can be simplified assuming
α
α α
α
α
T c h
p
∇ = ∇
,
(241)
to
( )  .
~
, ,
α
α α α
α
α α α
ρ ε T D c k
p eff
∇ ⋅ + ⋅ ∇ − (242)
33
Gedeon also recognizes this Peclet dependence (36). He calls this phenomenon axial
conductivity enhancement, which is an essential definition of thermal dispersion. He
suggests that
( )
α α α α
α
α α
ρ k N c c k D c k
k
n m
xx p eff
= + = + Pr Re
2 1 , , ,
(243)
and he reports coefficients for several matrices of interest for regenerators such as wire
mesh screens and felts. This functional form for the dispersion agrees with other
empirical predictions (42).
2.3.4 Entropy Generation Due to Thermal Dispersion
It was shown that the terms in the gas energy equation representing molecular diffusion
and dispersion,
( )
(
(
¸
(
¸


.

\

+ ∇ ⋅ ∇ − ⋅ ∇
∫
αβ
α α
α
α
α α α α
α
α
α
ρ
ε
A
dS T n
V
T k h u
ˆ
1
ˆ ˆ
1 r
, (244)
reduce to
( )
α
α α
T k N
k
∇ ⋅ ∇ − .
The same terms in the gas entropy generation equation,
( )
(
(
¸
(
¸
∇
⋅ ∇ − ⋅ ∇
α
α
α
α
α α
α
α
α
ε
ρ
ε T
T
k u s
1
ˆ
ˆ
1 r
, (245)
represent entropy generation due to dispersion and molecular diffusion, respectively.
They reduce similarly to


.

\

∇
⋅ ∇ −
α
α
α
α α
T
T k N
k
.
34
2.3.5 Negligible Brinkman Effect
The Brinkman effect is represented by the term
( ) ( )
α
α α
α
α α
µ µ u u
r r
∇ ⋅ ∇ + ⋅ ∇ ∇
3
1
(246)
and is generally negligible in comparison to the Darcy and Forchheimer effects (45). The
Brinkman effect will be neglected in this analysis, yielding a spatially first order
momentum equation.
2.3.6 One Dimensional Model on the Macroscopic Length Scale
It is improper to impose a noslip boundary condition on the macroscopic problem
of a porous medium bounded by a solid surface. The wall can only satisfy a noslip
boundary condition for the microscopic flow problem. The solid wall that contains the
regenerator matrix has minimal effect on the macroscopic flow field. The microscale
flow is affected by viscous shear and interfacial heat transfer. Both of these effects are
related to surface area. For an averaging volume containing the wall, the ratio of the wall
surface area to the matrix surface area will be quite small. It is reasonable to conclude
that the wall will have minimal effect on the macroscopic flow in the axial direction. For
a 2D axisymmetric flow, the wall boundary condition is simply zero normal velocity,
which reduces to zero radial velocity for a cylindrical regenerator.
Typically the flow passages at the ends of the regenerator are designed with a
contraction in diameter in an effort to minimize dead volume. This will lead to flow
jetting at the ends of the regenerator necessitating the need for a 2D model. It is not clear
how this flow jetting will affect the regenerator performance, and this is an area of
35
ongoing research. For the current model, it is assumed that the flow approaches the
regenerator with a uniform profile. In this situation, there is no driving force to create
anything other than a one dimensional macroscopic flow field.
The regenerator is bounded on the ends by a homogeneous fluid in which the
differential continuum equations are valid. The transitional region between the
homogeneous fluid and homogeneous porous medium has been analyzed by OchoaTapia
and Whitaker (46, 47, 48). They find that the transition creates a jump condition in the
momentum and energy equations. These conditions lead to additional parameters which
need to be measured experimentally. Further analysis is needed to determine the
magnitude of these effects.
2.4 Simplified Equations
The preceding assumptions lead to a large reduction in the equation complexity.
Several additional terms remain which need to be represented in terms of volume
averaged quantities. These relations are developed in the following discussion.
2.4.1 Simplified Conservation of Mass Equation
Under these assumptions, the volumeaveraged conservation of mass equation (2
21) becomes
( ) . 0 =
∂
∂
+
∂
∂
α α
α
ρ
ρ
u
x t
(247)
where the volume averaging notation has been dropped where appropriate. This is the
standard one dimensional continuity equation in terms of volumeaveraged quantities.
36
2.4.2 Simplified Balance of Momentum Equation
The momentum equation (222) becomes
( ) ( )
0
ˆ ˆ
3
ˆ
~ 1
=
(
¸
(
¸
∇ − 
.

\

⋅ ∇ − ⋅ +
∂
∂
+
∂
∂
+
∂
∂
∫
4 4 4 4 4 4 4 4 3 4 4 4 4 4 4 4 4 2 1
r r
effects r Forchheime and Darcy
A
dS u u p I n
V
x
p
u u
x
u
t
αβ
α α α
α
α α
α
α
α α α α α
µ
µ
ρ ρ
(248)
which still contains the surface integral of the deviation quantities. Following the
analysis of Whitaker (45), the surface integral in Equation (248) can be written such that
( ) ( )   { } . 0
~ ~ ~
1
= + ⋅ +
∂
∂
+
∂
∂
+
∂
∂
−
α α α
α
α α α α α
ε µ ρ ρ u F I K
x
p
u u
x
u
t
xx
(249)
The permeability, K, and form drag coefficient, c
f
, are defined as
 .
~
~
Re
2 / 1
α
α
α α
α
µ
ρ
ε u sign
K u
c F
K K
K
f xx
xx
43 42 1
=
=
(250)
Note that the Forchheimer correction is expressed in terms of a Reynolds number based
on the permeability length scale. The permeability has units of length squared. The
momentum equation is now written as
( ) ( )   . 0
2 / 1
2
2
= + +
∂
∂
+
∂
∂
+
∂
∂
α
α
α α α
α α α
α α α α α
ε
ρ
ε µ
ρ ρ u sign
K
c
u u
K x
p
u u
x
u
t
f
(251)
37
2.4.3 Simplified αPhase Conservation of Energy Equation
The volumeaveraged conservation of energy equation for the αphase (223) can
be simplified to
( )
. 0
1
= ∇ ⋅ −


.

\

∂
∂
−
∂
∂
+
∂
∂
∫
αβ
α α α
α
α
α
α
α
α
α
α
α
α
α
α
α
α
ρ
ρ
A
k
dS T k n
V
x
T
k N h u
x t
e
(252)
2.4.4 Simplified βPhase Conservation of Energy Equation
The βphase conservation of energy equation (224) can be simplified to
( )
. 0
1
ˆ
1
= ∇ ⋅ −
(
(
¸
(
¸



.

\

+
∂
∂
∂
∂
−
∂
∂
∫
∫
αβ
αβ
β β β
β
β β
β
β
β
β
β
β
β
ρ
A
A
dS T k n
V
dS T n
V x
T
k
x t
T
c
(253)
The last term in Equation (253) is a surface integral representing the volumetric heat
transfer between the two phases. The same integral appears in Equation (252). These
two integrals are exactly of the same magnitude and are opposite in sign. Whitaker (42)
has proposed that these integrals be represented as
( ),
1 1 α
α
β
β α α α
α
β β β
β α
β
αβ αβ
ε
ε
T T H a dS T k n
V
dS T k n
V
v
A A
− = ∇ ⋅ = ∇ ⋅ −
∫ ∫
(254)
where a
v
H is the volumetric heat transfer coefficient. Substitution of Equation (254) into
Equation (252) and Equation (253) results in a twoequation energy model
38
( )
( ) 0 = − −


.

\

∂
∂
−
∂
∂
+
∂
∂
α
α
β
β
α
α
α
α
α
α
α
α
α
α
α
α
α
ρ
ρ
T T H a
x
T
k N h u
x t
e
v
k
(255)
and
( )
( ) . 0
ˆ
1
= − +
(
(
¸
(
¸



.

\

+
∂
∂
∂
∂
−
∂
∂
∫
α
α
β
β
β
α
β β
β
β
β
β
β
β
β
ε
ε
ρ
αβ
T T H a
dS T n
V x
T
k
x t
T
c
v
A
(256)
The conduction term in Equation (256) can be written as
(
(
¸
(
¸
∂
∂
∂
∂
− =
(
(
¸
(
¸
∂
∂

.

\

+ Ι
∂
∂
−
∫
x
T
k
x x
T
dS b n
V
k
x
A
β
β
β β
β
β
β β β
τ
αβ
~ 1 ~
(257)
where τ
β
is referred to as the “tortuosity” by Gedeon (36). The terminology for tortuosity
can be confusing since tortuosity is also used in the context of the fluid phase. The
closure variable,
β
b
~
, must be solved for numerically. Experiments have suggested that
the tortuosity is less than or equal to one. Tortuosity equal to one represents parallel path
geometries such as tube bundles. Geometries such as wire mesh, felts, and sintered
metals have tortuosities less than one. Unlike dispersion, the tortuosity is thought to be a
function of geometry only, and not the flow field. Using this result, and dropping the
volume averaging notation, the energy equations can be respectively written in final form
as
39
( )
( ) 0 = − − 
.

\

∂
∂
−
∂
∂
+
∂
∂
α β
α
α α α α
α α
ρ
ρ
T T H a
x
T
k N h u
x t
e
v k
(258)
and
( ) ( ) . 0 = − +
(
¸
(
¸
∂
∂
∂
∂
−
∂
∂
α β
β
α
β
β β
β
β
ε
ε
τ ρ T T H a
x
T
k
x t
T
c
v
(259)
2.4.5 Simplified αPhase Entropy Generation Equation
The simplified gas entropy generation equation (227) is
( )
( )
0
1
,
≥
∇
⋅ −
(
(
¸
(
¸
∂
∂
∂
∂
−
∂
∂
+
∂
∂
= ′ ′ ′
∫
αβ
α
α α
α
α
α
α α
α
α
α
α
α
α
α
α
α
α
α
α
α
α
ρ
ρ
A
k
gen
dS
T
T k
n
V
T
x
T
k N
x
u s
x t
s
s
r
(260)
2.4.6 Simplified βPhase Entropy Generation Equation
The simplified gas entropy generation equation (228) is
0
1
,
≥
∇
⋅ −
(
(
¸
(
¸
∂
∂
∂
∂
−
∂
∂
= ′ ′ ′
∫
αβ
β
β β
β
β
β
β β
β
β β
β
β
β
β
β
τ
ρ
A
gen
dS
T
T k
n
V
T
x
T
k
x t
s
s (261)
The last term in Equation (260) is a surface integral representing the entropy generation
due to volumetric heat transfer between the two phases. The same integral appears in
Equation (261). By a similar method as that with the energy method, these terms can be
expressed as
40
( )
α
α
β
β α
α
β
β β
β
β
αβ
T T
T
H a
dS
T
T k
n
V
v
A
− =
∇
⋅ −
∫
1
(262)
and
( ).
1 α
α
β
β β
β
α
α α
α
α
αβ
T T
T
H a
dS
T
T k
n
V
v
A
− =
∇
⋅
∫
(263)
Substitution of Equation (262) and Equation (263) into the entropy generation model
and dropping the volume averaging notation results in
( )
( ) ( )
α β
β
α
α
α
α α α
α α
α
ρ
ρ
T T
T
H a
x
T
T
k N
x
s u
x t
s
s
v k
gen
− −
(
¸
(
¸
∂
∂
∂
∂
−
∂
∂
+
∂
∂
= ′ ′ ′
,
(264)
and
( ) .
, α β
α
β
β
β β β
β β
τ
ρ T T
T
H a
x
T
T
k
x t
s
s
v
gen
− +
(
(
¸
(
¸
∂
∂
∂
∂
−
∂
∂
= ′ ′ ′ (265)
At this point, it is possible to write the entropy generation rate equation for the gasmatrix
system, which is
. 0
, , ,
≥ ′ ′ ′ + ′ ′ ′ = ′ ′ ′
gen gen sys gen
s s s
β β α α
ε ε (266)
By using the Bridgman tables (49) the Maxwell relation for a pure substance can be
written in a more convenient form as
( ) ,
1
ρ
ρ
hd dE
T
ds − = (267)
which can then be used to simplify the entropy generation equations for the gas and the
matrix. After considerable simplification, using the energy and momentum equations, the
gas entropy generation equation reduces to
41
.
2
2
2 ,
x
p
T
u
T T
T T
H a
x
T
T
k N
s
v k
gen
∂
∂
− − + 
.

\

∂
∂
= ′ ′ ′
α
α
α
α β
β α
α
α
α
α
(268)
For an incompressible substance, the Maxwell Relation reduces to
, dT
T
c
ds = (269)
which, together with the matrix energy equation, and the matrix entropy generation
equation reduces to
.
2
2
2 , α β
β α
β
β
β β
β
τ
T T
T T
H a
x
T
T
k
s
v
gen
− +


.

\

∂
∂
= ′ ′ ′ (270)
Now the system entropy generation equation can be written
. 0
2
2
2
2
2 ,
≥
∂
∂
− − +


.

\

∂
∂
+ 
.

\

∂
∂
= ′ ′ ′
43 42 1
4 43 4 42 1 4 4 3 4 4 2 1
4 4 3 4 4 2 1
losses inertial and viscous
transfer heat Film
v
conduction Matrix
conduction Gas
k
sys gen
x
p
T
u
T T
T T
H a
x
T
T
k
x
T
T
k N
s
α
α
α
α α β
β α
β
β
β β
β
α
α
α
α
ε
τ
ε ε (271)
It should be noted that all terms are positive definite, including the last term representing
entropy generation due to viscous and inertial losses. This requires that the sign of the
pressure gradient always be opposite to the sign of the velocity. This is an interesting
result, but not one that is immediately obvious. Using the momentum equation, this term
can be written as
0 ≥ 
.

\

+
∂
∂
+
∂
∂
=
∂
∂
− F
x
u u
t
u
T
u
x
p
T
u
α α α α α
α
α α
α
α
ρ ρ
(272)
where F is the DarcyForchheimer surface integral,
42
.
ˆ ˆ
3
ˆ
~ 1
∫ (
¸
(
¸
∇ − 
.

\

⋅ ∇ − ⋅ =
αβ
α α α
α
α α
α
µ
µ
A
dS u u p I n
V
F
r r
(273)
This indicates that if F IS NOT a strict function of velocity as proposed then it must
combine with the acceleration terms such that the entropy generation is positive definite.
For positive velocity
x
u u
t
u
F
∂
∂
−
∂
∂
− ≥
α α α α α
ρ ρ
(274)
and for negative velocity
.
x
u u
t
u
F
∂
∂
−
∂
∂
− ≤
α α α α α
ρ ρ
(275)
The other implication which is possible to extract from this result is that if F IS a strict
function of velocity, as proposed, then the acceleration terms should be eliminated from
the momentum equation. This produces the positive definite result
. 0
1
2 / 1
2
3 2
≥
¦
)
¦
`
¹
¦
¹
¦
´
¦
+ = =
∂
∂
−
K
c
u u
K T
F
T
u
x
p
T
u
f
ε
ρ
µε
α α α
α α
α α
α
α
(276)
There is an additional requirement on the functionality of the interfacial
convection terms. The assumptions imposed by Equation (254) must satisfy the
additional requirement that
43
. 0
ˆ
1
ˆ
1
2
≥ ∇ ⋅



.

\

− =
−
∫
αβ
α α α
β
β
β
β
β
α
α
α
α
α
α
α
β
β β
β
α
α
A
v
dS T k n
T T
T
V
T T
T
V
T T
T T
H a
(277)
where the volume averaging notation is reapplied for clarity. This statement is produced
by retaining the surface integrals representing the interfacial heat transfer through the
simplification process for the gas and matrix entropy generation equations. In this form,
it is clear that the entropy generation due to this effect is directly dependent on the
temperature deviation quantities,
α
T
ˆ
and
β
T
ˆ
, which are defined in the Appendix 1. These
quantities are proportional to the volumeaveraged temperature difference.
The entropy generation can be used to calculate lost available power by
integrating the volumetric entropy generation rate over the entire regenerator. The lost
power is then the total entropy generation rate times the reference temperature giving
.
0
2
2
2
2
2
0
,
∫
∫
(
(
¸
(
¸
∂
∂
− − +


.

\

∂
∂
+ 
.

\

∂
∂
=
′ ′ ′ =
r
L
v k
o
L
sys gen o lost
Adx
x
p
T
u
T T
T T
H a
x
T
T
k
x
T
T
k N
T
Adx s T W
α
α
α
α α β
β α
β
β
β β
β
α
α
α
α
ε
τ
ε ε
&
(278)
The reference temperature is the lowest naturally occurring temperature in the system.
The lost power represents the additional input power that is required to perform the same
thermodynamic function as compared to an internally reversible refrigerator. The lost
power is a scalar value which can then be used as an optimization parameter. This idea is
not investigated beyond this level in the dissertation other than calculating this value in
Chapter 5. This form of the lost power represents an internal method of calculation. An
44
alternative method can be considered based on an external control volume. The lost
power for a quasisteady system with only mass flow interactions with the external
environment can be found by cyclical integration of the entropy flux at the boundaries.
These two methods of calculating the lost power will give identical results for an
analytical system, but differences will be notices for a numerical approximation. This
provides a good method for evaluating the accuracy of a numerical scheme. Values of
this lost power discrepancy are reported in Chapter 5 for the numerical model presented
in Chapter 3.
2.4.7 Summary of Equations
The system of equations representing the one dimensional regenerator is repeated
here with the volume average notations removed. The mass flux is defined as m, and the
volumetric gas internal energy is defined as E. The αsubscripts are dropped for
simplicity. The porosity of the solid phase is defined to be one minus the gas porosity, ε.
0 =
∂
∂
+
∂
∂
x
m
t
ρ
(279)
  0
2 / 1
2
2 2
= + +


.

\

+
∂
∂
+
∂
∂
m sign
K
c
m m
K
p
m
x t
m
f
ε
ρ ρ
µε
ρ
(280)
( ) 0 = − − 
.

\

∂
∂
−
∂
∂
+
∂
∂
T T H a
x
T
k N mh
x t
E
v k β
(281)
( ) ( ) 0
1
= −
−
+
(
¸
(
¸
∂
∂
∂
∂
−
∂
∂
T T H a
x
T
k
x t
T
c
v β
β
β β
β
β
ε
ε
τ ρ (282)
45
x
p
T
m
T T
TT
H a
x
T
T
k N
s
v k
gen
∂
∂
− − + 
.

\

∂
∂
= ′ ′ ′
ρ
β
β
2
2
2
(283)
2
2
2 ,
1
T T
TT
H a
x
T
T
k
s
v
gen
−
−
+


.

\

∂
∂
= ′ ′ ′
β
β
β
β
β β
β
ε
ε
τ
(284)
( ) . 0 1
2
2
2
2
2
,
≥
∂
∂
− − +


.

\

∂
∂
− + 
.

\

∂
∂
= ′ ′ ′
x
p
T
m
T T
TT
H a
x
T
T
k
x
T
T
k N
s
v k
sys gen
ρ
ε
τ
ε ε
β
β
β
β
β β
(285)
( ) E f T ,
1
ρ = (286)
( ) E f p ,
2
ρ = (287)
This system of PDEs represents 9 equations to be solved for
ρ Gas density
m Gas mass flux
E Gas total energy per unit volume
T Gas temperature
p Gas pressure
β
T Matrix temperature
gen
s ′ ′ ′ Gas volumetric entropy generation rate
gen
s
, β
′ ′ ′ Matrix volumetric entropy generation rate
sys gen
s
,
′ ′ ′ Total volumetric entropy generation rate
The volumetric total energy, E , is chosen as the conserved quantity in the gas energy
equation. This is simply the product of the internal energy and density. These equations
require appropriate boundary and initial conditions which will be discussed.
46
These equations represent a fully compressible model for flow and heat transfer in
a porous medium. The porous medium is completely characterized by the specification
of the porosity, friction factor, dispersion coefficient, solid tortuosity, and heat transfer
coefficient in addition to fluid and solid thermal properties. At this point in the
development, these equations apply to onedimensional steady, unsteady, or oscillating
flow. For limiting cases, important exact solutions exist, and these are discussed in
section 2.5 that follows.
2.5 Exact Solutions
Experimental measurement of the permeability and inertia coefficient is
conducted by measuring the steady mass flow rate and pressure drop through a one
dimensional channel. Under steady state conditions, the first term in Equation (251) is
zero, and the momentum equation reduces to
  . 0
2 / 1
2
2
= + + +
α
α
α α α
α α α α
α α
ε
ρ
ε µ
ρ u sign
K
c
u u
K dx
dp
dx
du
u
f
(288)
The continuity equation reduces to
,
1
1
α
α α α
ρ ρ
u
C
C u = ⇒ = (289)
and the momentum equation becomes (for positive velocity and ideal gas)
,
2 / 1
2
1
3
2
(
(
¸
(
¸
+ − =
(
¸
(
¸
−
K
c
KC dx
du
u
RT u
f α
α α α
α
α
ε
ε µ
(290)
which can be integrated to give
47
( ) .
2
2
2 / 1
2
1
2
C x
K
c
KC u
RT
u Ln
f
+
(
(
¸
(
¸
+ − = +
α
α α
α
α
ε
ε µ
(291)
The two constants can now be solved to give
f
A
m
C
&
=
1
(292)
and
,
2
2
2
0
2
0
2
RT m
p A
p A
RT m
Ln C
f
f
&
&
+


.

\

= (293)
where p
0
is the pressure at the inlet. The result is an implicit equation for velocity which
apparently needs to be solved numerically,
.
2
1
2
1 1
2 / 1
2 2
0
2
0
x
RTK
c
K m RT
A
m RT
A p
u RT m
p A u
Ln
RT
f f f f
(
(
¸
(
¸
+ −


.

\

= +


.

\

α α α
α
α
ε ε µ
& & &
(294)
Equation (294) represents a steady, compressible, isothermal Forchheimer momentum
equation. This equation allows the selection of the mass flow rate and the inlet pressure.
By solving for the velocity at the outlet allows for calculating the density at the outlet,
and thus the pressure, as outlined in what follows. In such a manner, experimental data
can be used to solve for the two Darcy and Forchheimer coefficients iteratively.
If the convective acceleration term is neglected in the momentum equation, an
additional exact solution can be obtained for the case of steady, isothermal, compressible
flow. The problem is defined by
48
dx
du
u A
m
dx
d
A u
m
f f
α
α
α
α
α
ρ
ρ
2
& &
− = ⇒ = (295)
,
1
2 / 1
2
3
RTK
c
K m RT
A
dx
du
u
f f α α α
α
α
ε ε µ
+ =
&
(296)
which can be integrated to give
.
2
1
2 / 1
2
2
x
RTK
c
K m RT
A
C
u
f f


.

\

+ − =
α α α
α
ε ε µ
&
(297)
The constant can be solved to give
.
2
1
2
0


.

\

=
m RT
A p
C
f
&
(298)
The result is an equation for velocity,
,
2
1
2
1
2 / 1
2
2
0
2
x
RTK
c
K m RT
A
m RT
A p
u
f f f


.

\

+ −


.

\

=
α α α
α
ε ε µ
& &
(299)
which is identical to Equation (294) after dropping the natural logarithm term. This
equation can be solved directly for the velocity giving
. 2
1
2 / 1
2
2
0
−
(
(
¸
(
¸


.

\

+ −


.

\

= x
RTK
c
K m RT
A
m RT
A p
u
f f f α α α
α
ε ε µ
& &
(2100)
If the flow is steady, isothermal, and incompressible, the continuity equation reduces to
0 =
dx
du
α
(2101)
and the Forchheimer Momentum Equation reduces to
49
  , 0
2 / 1
2
2
= + +
α
α
α α α
α α α
ε
ρ
ε µ
u sign
K
c
u u
K dx
dp
f
(2102)
which is the standard “incompressible” Forchheimer Equation. This equation implies
that the pressure gradient and density are known constants and not related through an
equation of state. Thus, the pressure gradient must be constant giving
  0
2 / 1
2
2
0
= + +
−
α
α
α α α
α α
ε
ρ
ε µ
u sign
K
c
u u
K L
p p
f
r
L
(2103)
The pressure at the outlet can be solved in terms of the inlet pressure and mass flow rate
giving
. 2
2 / 1 2
2
2
0


.

\

+ − =
K A
c m
KA
m
RT L p p
f
f
f
r L
α
α α
ε
ε µ
&
&
(2104)
These three models are compared in Chapter 5. The compressible model and the
incompressible model differ by 31% in predicting friction factor. The effect of the
advective acceleration term is shown to be negligible in predicting friction factor.
2.6 Scale Analysis
A nontraditional scale analysis method is adopted for scaling the partial
differential system of equations. Since a partial differential equation relates the change of
a variable in more than one dimension, it is sensible to expect that there are separate
scales for each dimension. Thus, separate temporal and spatial scales are used.
50
2.6.1 Conservation of Mass Equation Scale Analysis
The scale analysis begins by scaling the continuity equation. By choosing
independent temporal and spatial scales for density and mass flux, the scaled continuity
equation is written as
0
~
=
∂
∂
(
¸
(
¸
+
∂
∂
x
m t
L
m
t
s
s
ρ
ρ
(2105)
where the tilde overbar represents the temporal scale and the dash overbar represents the
spatial scale. Arbitrary length and time scales,
s
L and
s
t , have been used. The result of
this scaling is the single balance
m
L
t
s
s
ρ
~
~ . (2106)
The scaled continuity equation then reduces to the familiar form
0 =
∂
∂
+
∂
∂
x
m
t
ρ
. (2107)
2.6.2 Balance of Momentum Equation Scale Analysis
Next, the momentum equation is similarly scaled to give
 
0
~ ~
~
~
~
~ 2 2
=
(
¸
(
¸
+
∂
∂
(
¸
(
¸
+


.

\

∂
∂
(
¸
(
¸
+
∂
∂
f
m sign m
K
L
m
m
x
p
m m
p m
x m
m
t
m
s
ρ ρ
ρ ρ
ρ ρ
ρ
. (2108)
where the friction factor and the Reynolds number are, respectively
f
K
c f
2
Re
ε
ε
+ = (2109)
and
51
*
*
Re
µ
K m
K
= . (2110)
where the asterisks denote dimensional quantities. The length scale in the Reynolds
number is the square root of the permeability, not the macroscopic length scale, L
s
. The
macroscopic length scale represents the length scale of the macroscopic flow while the
permeability length scale represents the length scale of the microscopic flow which is
what drives the Darcy and Forchheimer effects. The fact that two length scales have been
defined is evident since the volume averaging method requires such a result. The two
length scales are necessary to model a porous media as a continuum where the
macroscopic and microscopic effects are united. The basic assumption of the volume
averaging method is that the macroscopic length scale is orders of magnitude larger than
the microscopic length scale. Thus, it is necessary to define these two length scales, and
it is expected that the model will eventually depend on two length scales.
This scaling leads to two balances. The first balance is proposed for the temporal
and spatial acceleration terms, which gives
ρ
ρ
~
~
~
m
m
. (2111)
The remaining balance between the pressure gradient and the friction terms results in
K
L m
p
s
ρ
2 ~
~ . (2112)
These two scales lead to the scaled momentum equation,
52
 
0
2
1
2
=
(
¸
(
¸
+
∂
∂
Γ +


.

\

∂
∂
+
∂
∂
f
m sign m
x
p m
x t
m
ρ ρ
, (2113)
which is characterized by a leading order dimensionless parameter,
K
L
s
2
1
~


.

\

= Γ
ρ
ρ
. (2114)
Since this parameter is much larger than the remaining acceleration terms, which are O(1)
in this form, it is tempting to consider dropping the acceleration terms. This results in an
algebraic equation for the velocity given the pressure field.
2.6.3 αPhase Conservation of Energy Equation Scale Analysis
Using an additional scale for the interfacial temperature difference, T
ˆ
, the scaled
gas energy equation is
( )   ( ) 0
ˆ
~
~
2
= − − 
.

\

∂
∂
∂
∂
(
¸
(
¸
−
∂
∂
(
¸
(
¸
+
∂
∂
(
¸
(
¸
T T H T H a
x
T
k N
x L
T k
mh
x L
h m
t
E
L
m E
s v k
s
s
s s
β
ρ
(2115)
which can be reduced to
( )
( ) 0 4
~
ˆ
~ ~
= −
(
(
¸
(
¸
−

.

\

∂
∂
∂
∂
(
(
¸
(
¸
−
∂
∂
(
¸
(
¸
+
∂
∂
T T H
c m
H
K
L
T c
T c
x
T
k N
x K c m
k
L
K
T c
T c
mh
x T c
T c
t
E
ps
s s
vs
ps
k
ps
s
s vs
ps
vs
ps
β
(2116)
if
53
T c E
vs
~
~
~
ρ = , (2117)
T c h
ps
=
, (2118)
and
vs
ps
c
c
= γ . (2119)
By defining the reference Peclet number and the reference Stanton number respectively
as
s
ps
k
K c m
Pe = (2120)
and
m c
H
St
ps
s
= , (2121)
and the surface area per unit volume as
K
a
v
4
= , (2122)
the gas energy equation becomes
( ) ( ) 0 4
~
ˆ
1
~ ~
= −
(
¸
(
¸
− 
.

\

∂
∂
∂
∂
(
¸
(
¸
−
∂
∂
(
¸
(
¸
+
∂
∂
T T H St
K
L
T
T
x
T
k N
x Pe L
K
T
T
mh
x T
T
t
E
s
k
s
β
γ γ γ . (2123)
The proper balance is between the interfacial heat transfer term and the enthalpy
advection term which gives
s
L
K
T
T
St
4
1
ˆ
~ . (2124)
This leads to the final scaled gas energy equation,
54
( ) 0
1 2
=
(
¸
(
¸
− − 
.

\

∂
∂
−
∂
∂
Γ +
∂
∂
T T H
x
T
k N mh
x t
E
k β
ε (2125)
which is characterized by an additional leading order dimensionless parameter,
T
T
~ 2
γ = Γ (2126)
and a second order dimensionless parameter
Pe L
K
s
1
1
= ε . (2127)
The last parameter, which characterizes the diffusive flux, is very small in comparison to
the first parameter. This indicates that the gas energy equation is dominated be advection
and interfacial heat transfer.
2.6.4 βPhase Conservation of Energy Equation Scale Analysis
The matrix energy equation can be scaled to produce
( )
( )
( )
( ) . 0
~
ˆ
4
~
1
1
~
~
= −
(
(
¸
(
¸
−
+
(
¸
(
¸
∂
∂
∂
∂
(
(
¸
(
¸
−
∂
∂
T T H St
T
T
K
L
c
c
x
T
k
x Pe L
K
k
k
T
T
c
c
t
T
c
s
s
vs
s s
s
s
vs
β
β
β
β β
β
β
β
β
γ
ρ
ρ
ε
ε
τ γ
ρ
ρ
ρ
(2128)
There is only one relevant balance with this scaling, which is between the interfacial heat
transfer term and the accumulation term. This is represented by the balance
( )
vs
s
c
c
T
T
ρ
ρ
ε
ε
γ
β
~
1 1
~
~
−
. (2129)
Using this balance, the matrix energy equation can be written in its final form,
55
( ) ( ) 0
2
= − +
(
¸
(
¸
∂
∂
∂
∂
−
∂
∂
T T H
x
T
k
x t
T
c
β
β
β β
β
β
τ ε ρ , (2130)
which is characterized by an additional second order dimensionless parameter,
Pe L
K
k
k
s s
s
1 1
2
β
ε
ε
ε
−
= . (2131)
Again, this parameter is much smaller than the other terms which are O(1) which
illustrates that the interfacial heat transfer and accumulation dominate over the diffusive
flux.
2.6.5 Scaled Equation Summary
The scaled equations are:
0 =
∂
∂
+
∂
∂
x
m
t
ρ
(2132)
 
0
2
1
2
=
(
¸
(
¸
+
∂
∂
Γ +


.

\

∂
∂
+
∂
∂
f
m sign m
x
p m
x t
m
ρ ρ
(2133)
( ) 0
1 2
=
(
¸
(
¸
− − 
.

\

∂
∂
−
∂
∂
Γ +
∂
∂
T T H
x
T
k N mh
x t
E
k β
ε (2134)
( ) ( ) 0
2
= − +
(
¸
(
¸
∂
∂
∂
∂
−
∂
∂
T T H
x
T
k
x t
T
c
β
β
β β
β
β
τ ε ρ (2135)
which are characterized by four dimensionless parameters resulting from the scaling
balances. The boundary conditions provide the remaining information to complete the
scale analysis. The time scale can be chosen as the angular frequency of the mass flow
rate as
56
ω
1
=
s
t . (2136)
The temporal mass flux scale can be chosen based on the boundary mass flux amplitudes
as
2
~ c h
m m
m
+
= . (2137)
The spatial mass flux scale can be expressed by assuming that the amplitude and phase
varies linearly across the 1D domain. Then, the RMS value of the gradient is
( )
2
cos 2
2 2
mh mc c h c h
m m m m
m
φ φ − − +
= . (2138)
The spatial temperature scale can be expressed as
c h
T T T − = (2139)
which then leads to the spatial scale for density;
( )
c h
c h b
T RT
T T p −
= ρ . (2140)
The five scales can now be simplified to
m
m
L
s
~
1
2
ρ ω
= , (2141)
m
m
~
~
ρ ρ =
, (2142)
( )
( )
s
vs
c h
c h b
c
c
m
m
T RT
T T p
T
β
ρ ε
ε
γ
−
−
=
1
~
~
2
, (2143)
57
K
m
p
ω ρ
1
~
~
2
3
=
, (2144)
and
2
~
4
ˆ
m
m
St
K
p
T RT
T
b
c h
ω
= . (2145)
The importance of this scaling is that it has produced expressions which give insight into
the amplitude of the temperature oscillation as well as the temperature difference between
the matrix and gas. These amplitudes are seen to depend on the boundary conditions.
Finally, the dimensionless parameters can be written in terms of known quantities
as
ρ ω K
m
~
1
= Γ (2146)
( )
vs
s
c
c
m
m
ρ
ρ
ε
ε
β
~
1
2
−
= Γ (2147)
Pe m
K
1
~
1
ρ
ω ε = (2148)
1 2
1
ε
ε
ε
ε
β
s
s
k
k
−
= (2149)
Table 1 summarizes the numerical results of the scale analysis for a representative
regenerator.
58
Table 1Summary of scale analysis
F 40 [Hz] A
f
5.969E05 [m^2]
L
r
7.30E02 [m] w 251 [rad/s]
D
r
0.01048 [m] R 2078 [Pam^3/kgK]
Ε 0.6920 [] c
ps
5190 [J/kgK]
K 1.005E10 [m^2] c
vs
3112 [J/kgK]
m
h
19.58 [kg/s] γ 1.67 []
m
c
29.62 [kg/s] ( )
ms
c ρ 3.5E+06 [J/m^3K]
φ
mh
0 [rad] k
b
12.5605 [W/mK]
φ
mc
1.62 [rad] k 0.1146 [W/mK]
h
T
300 [K]
c
T
70 [K]
b
p
3.40E+06 [Pa]
t
s
0.0040 [s] H
s
35482 [W/m^2K]
L
s
5.92E03 [m] St 2.67E01 []
ρ
~
17.21 [Kg/m^3] Pe 12 []
ρ
17.92 [Kg/m^3] N
k
7.41 []
m
~
24.60 [kg/s]
m 25.6 [kg/s]
p
2.00E+04 [Pa]
E
~
707388 [J/m^3] Γ
1
545 []
T
~
13.21 [K] Γ
2
29.0 []
T 230 [K] ε
1
1.46E04 []
T
ˆ
1.14E03 [K] ε
2
7.10E03 []
59
2.6.6 Limiting Cases
For the parameters chosen, the dimensionless parameters satisfy the assumptions
imposed on the problem, i.e.
1
1
>> Γ (2150)
1 2
ε >> Γ (2151)
1
1
<< ε (2152)
1
2
<< ε (2153)
There are several limiting cases of interest when these assumptions break down. For
example, and the frequency approaches zero, it is expected that the upstream influence of
the temperature will have an important effect.
60
CHAPTER 3
3. COMPUTATIONAL MODELS
3.1 Problem Definition
The problem to be solved, as defined in Chapter 2, is the unsteady, periodic, one
dimensional regenerator problem. The geometry is a cylindrical domain. Figure 6
illustrates the computational domain.
Figure 6 – Regenerator computational domain diagram
L
r
D
r
( ) t m
h &
( ) t m
c &
x
61
The solutions which are desired are the quasisteady solutions, i.e. the solutions are
identical from one cycle to the next. To accomplish this solution, a numerical method
was employed to approximate the solutions. The details of the temporal and spatial
representation are discussed in the following sections. This is followed by a discussion
of a unique technique which was developed to artificially advance the solution in time to
the quasisteady solution. This technique relies on the proper specification of boundary
conditions which is discussed below. Several asymptotic models are then developed
which illustrate the mechanisms which contribute to a net energy transport in the
regenerator.
3.2 Numerical Method
The numerical solution of partial differential equations (PDEs) is a broad field.
Typical solution techniques in Computational Fluid Dynamics involve temporal and
spatial discretization. A unique method of solving PDEs is a semianalytical technique
call the Method Of Lines (MOL.) MOL is a general technique in which all but one
domain of a multidomain system of PDEs is discretized (50, 51, 52, and 53). The
remaining domain remains analytical, thus the meaning of semianalytical. The result of
this discretization is a system of ordinary differential equations (ODEs.) In physical
systems, this usually involves discretizing the spatial domain, which can be
multidimensional. At this level, all MOL approaches are identical. To proceed, the
analyst must choose a unique method of time integration as well as spatial discretization.
62
3.2.1 Time Integration
Time integration can in general be an analytical technique or a numerical
technique. Numerical techniques are generally classified as implicit and explicit. For
hyperbolic systems, implicit time integration is necessary for stability. MatLab was used
as the platform for solving the equations. The implicit ODE integrator, ODE15S, was
found to be the most stable and efficient. ODE15S is a variable order solver which can
use numerical differentiation formulas (NDFs) or backward differentiation formulas
(BDFs.) The BDF formulas are commonly known as Gear’s method and they suffer from
lower efficiency. ODE15S is especially efficient in solving stiff systems (54).
3.2.2 Spatial Discretization
Several spatial discretization methods exist in the literature such as finite element,
finite volume, finite difference, and collocation (55, 56, 57, 58, 59). Finite differences
were used to solve the regenerator equations. Experimentation indicated that first order
accurate central finite differences were the most stable and efficient if the equations were
formulated in strong conservation form. Five point central differences yield fourth order
accuracy, but this produces a larger banding in the Jacobian leading to reduced stability
in the time integrator. Seven point differences were too unstable to use at all. Upwind
differences did not appear to be as stable as central differences. As will be shown later in
the discussion of boundary conditions, the continuity equation reduces to a second order
equation in density in the limit of explicit velocity formulation. In this context, it would
be understandable that central differences would yield the most stable scheme. To
63
calculate second derivatives, first order central differences using a three point stencil
were used recursively. Appendix 2 contains a detailed discussion of the derivation of
difference operators and their use with MatLab. For the equation set defined by
Equations (279) – (282), the MOL discretized equations are
,
t
x ij
t
x
j
i
m D
t
− =
∂
∂ρ
(31)
( ) ( )
  ,
2 / 1
2
2 2
t
x
f
t
x
t
x
t
x
t
x
t
x t
x t
x
t
x
ij
t
x
i
i
i
i
i i
j
j
j
i
m sign
K
c m m
K
p
m
D
t
m ε
ρ ρ
ε µ
ρ
− −


.

\

+ − =
∂
∂
(32)
( )   ( ) ,
t
x
t
x
t
x v
t
x jk
t
x
t
x k
t
x
t
x ij
t
x
i
i
i k j j j j
i
T T H a T D k N h m D
t
E
− + − − =
∂
∂
β
(33)
and
( )
( )   ( ) .
1
1
)
`
¹
¹
´
¦
−
−
− =
∂
∂
t
x
t
x
t
x v
t
x
jk
t
x
ij t
x
t
x
i
i
i
k j
i
i
T T H a T D k D
c t
T
β β β β
β
β
ε
ε
τ
ρ
(34)
for i=1..n where n is the number of spatial grid nodes. In these discretized equations, the
central difference operator,
ij
D , is an nxn square matrix. Equations (31) – (34) are now
a coupled system of 4n ODEs which can be integrated in time from a set of 4n initial
conditions.
3.3 QuasiSteady Convergence via Cyclic Time Relaxation
Time integration begins with an initial condition for each of the solution
variables. The equation set is integrated over a complete cycle. The solution at the final
time step becomes the initial condition for the next period of integration. The integration
64
continues until the solution for two consecutive periods are equal. When this condition is
met, the net change of internal energy of the system over a cycle is identically zero and
the system is said to be in quasisteady state. The regenerator system suffers from a long
time constant with respect to the cycle period. As a result, the system of equations needs
to be integrated for thousands of cycles until quasisteady state is achieved unless the
initial condition can be advanced artificially. Cyclic analysis of the governing equations
provides an interesting and valuable technique called cyclic time relaxation. Cyclic
averaging of the gas and matrix energy equations results in
  , 0
, ,
=
)
`
¹
¹
´
¦
+ +
∂
∂
∫
dt q q h
z
ss m ss g ss
(35)
where the “ss” subscript indicates steady state. Let
.
1
, ,
,
, ,
x
T
k q
x
T
k N q T uc h
ss
ss m
ss
k ss g ss p ss
∂
∂
−
− =
∂
∂
− = =
β
β β
τ
ε
ε
ρ (36)
A simplification in terms of temperature gives
. 0
1
,
=
)
`
¹
¹
´
¦
(
¸
(
¸
∂
∂
−
−
∂
∂
−
∂
∂
∫
dt
x
T
k
x
T
k N T uc
x
ss
ss
k ss p
β
β β
τ
ε
ε
ρ (37)
The temperature field at the end of a cycle may not satisfy this relationship. In this case,
this relation provides a means of calculating a correction for the initial condition of the
following cycle. By defining the quasi steadystate temperature fields which satisfy
Equation (37) exactly as
65
( ) ( ) ( ) ( ) , , , ,
,
t x T x T T t x T x T T
ss ss β β
+ ′ = + ′ = (38)
where the primed temperature is an axial correction. T(x,t) and T
β
(x,t) are the
temperatures of the gas and matrix, respectively, from the current cycle. These quantities
can be substituted into Equation (37) and T’(z) can be solved to obtain
( ) ( ) ( ) , 0
3 2 2
2
1
= +
∂
′ ∂
+
∂
′ ∂
z f
z
T
z f
z
T
z f (39)
where
( )
( ) { }
( )
∫
∫
∫
)
`
¹
¹
´
¦
(
¸
(
¸
∂
∂
−
−
∂
∂
−
∂
∂
=
=
)
`
¹
¹
´
¦ −
+ − =
dt
x
T
k
x
T
k N T uc
x
x f
dt uc x f
dt k k N x f
k p
p
k
β
β β
β β
τ
ε
ε
ρ
ρ
τ
ε
ε
1
1
3
2
1
(310)
For the case of constant specific heat,
( ) { } 0
2
= =
∫
dt u c x f
p
ρ (311)
which follows from a cyclicaverage of the continuity equation. Equation (39) is a
second order ODE which can be solved for the temperature correction function. The
boundary conditions are appropriately chosen as
( )
( ) , 0
0 0
= ′
= ′
r
L T
T
(312)
since the temperature field at the ends does not need a correction due to the boundary
conditions which fix those values. T’ can then be solved numerically with a finite
difference technique. Then the initial condition for the temperatures for the next period is
66
( ) ( ) ( )
( ) ( ) ( ). , , 0
, , 0
x T x T x T
x T x T x T
′ + =
′ + =
λ τ
λ τ
β β
(313)
A relaxation factor, λ, is used to aid stability. Experimentation has indicated an optimum
value of 0.300.35.
3.4 Boundary and Initial Conditions
Equations (279)  (282), in their present form, are a mixed system of first and
second order (spatially) partial differential equations. The continuity and momentum
equations are both first order spatially while the energy equations are both second order.
These equations in strict mathematical terms require six boundary conditions for the
problem to be wellposed.
Two boundary conditions are required by the continuity and momentum
equations. Stable solutions are found by imposing pressure at both ends or mass flux at
both ends. Mass flux boundary conditions are chosen such that zero net mass flux can be
achieved. Boundary conditions are needed at both ends for stability and this can be
understood by considering the limiting case when the acceleration terms in the
momentum equation are neglected. This results in the explicit equation for velocity,
  . 0
2 / 1
2
2
= + +
∂
∂
u sign
k
c
u u
k x
p
f α
α α
ε
ρ
ε µ
(314)
It is then clear that velocity is a function of the pressure gradient, which, in turn, is a
function of the density gradient and temperature gradient. When the velocity in the
continuity equation is eliminated in terms of temperature and density, the result is a
spatially second order continuity equation,
67
. 0 , =


.

\


.

\

∂
∂
∂
∂
∂
∂
+
∂
∂
x
T
x
f
x t
ρ
ρ
ρ
(315)
In this limiting case, the continuity equation is diffusive. Thus, it seems natural to
impose mass flux boundary conditions at both ends of the domain.
In actuality, the momentum equation is a spatially first order equation. The
advective acceleration term which makes it first order is quite small and can be neglected
producing a zeroth order momentum equation. Together with the first order continuity,
only one boundary condition is required. Since two mass flow boundary conditions have
already been specified, the problem is overspecified by one boundary condition at this
point. The boundary condition requirements of the energy equations will absorb this
overspecification.
Referring back to the scaled energy equations, the diffusion terms are
characterized by a small dimensionless parameter. This indicates that there is the
possibility for a thermal boundary layer at the ends of the regenerator. Numerical
experiments suggest that resolving such a boundary layer does not have a significant
impact on the regenerator solution since the temperature fluctuation in the regenerator is
largely a result of the compression process. Outside of this boundary layer, the
regenerator is largely unaffected by the incoming gas temperature. This motivates a
simplified application of boundary conditions based on a leading order simplification of
the energy equations.
Scaling and numerical experiments suggest that these second order diffusion
terms are essentially steady, spatial sources, i.e. the temporal fluctuations are orders of
68
magnitude smaller than the spatial variations. This suggests that the boundary conditions
that are required mathematically can be neglected with little effect on the solution. The
gas energy equation is still first order due to the enthalpy advection term, but the matrix
energy equation is zeroth order. As a result, the gas energy equation requires a single
boundary condition while the matrix energy equation requires no boundary conditions.
To summarize, the solutions of interest can be found by neglecting the strict
mathematical requirement of six boundary conditions. For the limiting problem, the
equations consist of a first order continuity and gas energy and zeroth order momentum
and matrix energy. Thus two boundary conditions are needed which satisfy the important
corollary condition of zero net mass flux. The most satisfactory method of achieving this
is by imposing harmonic mass flux boundary conditions, i.e.
( ) ( )
mh
f
h
h
t
A
m
u φ ω ρ + = cos
&
(316)
and
( ) ( ) . cos
mc
f
c
c
t
A
m
u φ ω ρ − =
&
(317)
These mass flux boundary conditions are actually boundary conditions for the momentum
equation, not the continuity equation. Imposing mass conserving boundary conditions for
the continuity equation is difficult since simple sinusoidal densities at the ends will not in
general lead to mass conservation. If the entire cryocooler is modeled as a closed system,
then density boundary conditions could be used.
Since the solution of interest is the quasisteady solution, the proper initial
condition is one that produces a quasisteady solution after one cycle of simulation. This
69
is essentially the purpose of the artificial convergence technique discussed above.
Additionally, this technique is responsible for producing a solution which has the desired
mean pressure and mean warm and cold end temperatures. Thus, the converged solution
satisfied seven boundary parameters;
h
m& Warm end mass flow amplitude
h
m& Cold end mass flow amplitude
mc mh
φ φ − Mass flow phase shift
ω Angular frequency
h
T Warm end mean temperature
c
T Cold end mean temperature
b
p Mean pressure
These parameters are in addition to the seven geometric and empirical parameters which
define the shape of the regenerator and the pores;
r
L Regenerator length
r
D Regenerator diameter
ε Porosity
f Friction factor
Nu Nusselt number
k
N Dispersion coefficient
β
τ Matrix tortuosity
70
3.5 The Constant Temperature Model (CTM)
A variety of asymptotic models can be considered. Each asymptotic model
provides insight into the mechanism of net energy transport in the regenerator. Two
asymptotic cases are presented, beginning with the most restrictive and an intermediate
model. Each model can be considered with real or ideal gas, and constant or variable
properties, which provide additional permutations. Results of numeric computations with
these models are presented in Chapter 5. There it is revealed that the regenerator net
energy transport increases with each additional relaxation.
The simplest idealized regenerator is represented by the Constant Temperature
Model (CTM.) This model assumes 1) that the heat transfer between the gas and the
matrix is perfect and 2) that the matrix heat capacity is very large. The consequence of
the first assumption is that there is an infinitesimal temperature difference between the
gas and matrix. The consequence of the second assumption is that the temperature field
is steady, i.e. there is no temperature oscillation, only a spatial temperature field.
Mathematically, this requires that the temperature field be specified a priori. The matrix
gas energy equation can be rearranged to give
( ). T T H a
x
T
k
x
v
− =
(
¸
(
¸
∂
∂
∂
∂
−
β
β
β β
τ (318)
which can then be inserted into the gas energy equation to give
. 0 =


.

\

∂
∂
−
∂
∂
−
∂
∂
+
∂
∂
x
T
k
x
T
k N uh
x t
E
k
β
β β
τ ρ (319)
The cycle average of this equation is
71
. 0 =


.

\

∂
∂
−
∂
∂
−
∂
∂
∫
dt
x
T
k
x
T
k N uh
x
k
β
β β
τ ρ (320)
Note that the cycle average of the time rate of change of the volumetric internal energy is
identically zero for steady, periodic operation. For ideal gas with constant properties, this
becomes
( ) . 0 =


.

\

∂
∂
−
∂
∂
−
∂
∂
+
∂
∂
∫ ∫
dt
x
T
k
x
T
k N
x
dt u
x
T c
k p
β
β β
τ ρ (321)
The first integral, which is the net enthalpy flow, can be shown to be zero by cyclic
integration of the continuity equation. This is the only condition under which the net
enthalpy flow is zero. The diffusion fields drive the steady state temperature field for this
case. Beginning with a linear temperature field, the cyclic time relaxation procedure is
applied to advance the solution to steady state. If a real gas equation of state is used, then
the net enthalpy flow is positive. This is obvious since the enthalpy is a function of
temperature and density. The steady temperature field is then affected by the diffusion
and net enthalpy flow fields. The regenerator loss due solely to the density dependence
of the enthalpy can now be assessed.
This model is thermally perfect, but the effects of pressure drop and viscous
dissipation are still captured via a coupled solution of the continuity and momentum
equation, which remain unchanged. Thus, the system of equations being solved is
72
0 =
∂
∂
+
∂
∂
x
m
t
ρ
(322)
  0
2 / 1
2
2 2
= + +
∂
∂
+


.

\

∂
∂
+
∂
∂
m sign
K
c
m m
k x
p m
x t
m
f
ε
ρ ρ
ε µ
ρ
α
(323)
This system of PDEs requires two mass flux boundary conditions, as defined previously
in addition to initial conditions for density and mass flux. Cyclic time relaxation is
applied to find a temperature field which yields a steady solution.
3.6 The Local Thermal Equilibrium Model (LTEM)
Local thermal equilibrium should not be confused with local thermodynamic
equilibrium. In this model, the assumption of very large matrix heat capacity is relaxed
while retaining the assumption of perfect gastomatrix heat transfer. The result is a
single energy equation with a single temperature. This energy equation is formed by
combining Equations (281) and (282) such that
( ) . 0
1 1
=


.

\

∂
∂
−
−
∂
∂
−
∂
∂
+
∂
∂
−
+
∂
∂
x
T
k
x
T
k N uh
x t
T
c
t
E
k
β
β β
β
β
τ
ε
ε
ρ
ε
ε
ρ (324)
For ideal gas, this combined energy equation can be rewritten as
( ) . 0
1 1
=


.

\

∂
∂
−
−
∂
∂
− +
∂
∂
+
∂
∂
−
+
∂
∂
+
∂
∂
x
T
k
x
T
k N Pu ue
x t
T
c
t
T
c
t
e
k v
β
β β
β
β
τ
ε
ε
ρ
ε
ε
ρ ρ
ρ
(325)
Using the fact that the temperatures of the gas and matrix are identical, the combined
energy equation is now written as
73
( ) ( )
. 0
1
1
=


.

\

∂
∂
−
−
∂
∂
−
∂
∂
+
∂
∂
+
(
¸
(
¸
∂
∂
+
∂
∂
+
∂
∂
(
¸
(
¸
−
+
x
T
k
x
T
k N Pu
x
x
T
uc u
x t
e
t
T
c c
k
v v
β
β β
β
τ
ε
ε
ρ ρ
ρ
ε
ε
ρ ρ
(326)
Continuity can be used to eliminate the second term, and the final form of the LTE
energy equation is
. 0
1 1
=


.

\

∂
∂
−
−
∂
∂
−
∂
∂
+
∂
∂
+
∂
∂
x
T
k
x
T
k N Pu
x C x
T
C
uc
t
T
k
LTE LTE
v
β
β β
τ
ε
ε ρ
(327)
where the combined volumetric heat capacity is defined as
( ) .
1
ε
ε
ρ ρ
β
−
+ = c c C
v LTE
(328)
Implementing an equivalent combined energy equation for a real gas is more difficult due
to the additional density dependence of the gas internal energy. Expanding Equation (3
24) for a real gas gives
( )
0
1
1
=


.

\

∂
∂
−
−
∂
∂
−
∂
∂
+
∂
∂
−
+
∂
∂


.

\

∂
∂
+
∂
∂

.

\

∂
∂
+
∂
∂
x
T
k
x
T
k N uh
x
t
T
c
t
e
t
T
T
e
t
e
k
T
β
β β
β
β
ρ
τ
ε
ε
ρ
ε
ε
ρ
ρ
ρ
ρ ρ
ρ
(329)
which can be simplified to
( )
( )
. 0
1
1
=


.

\

∂
∂
−
−
∂
∂
−
∂
∂
+
∂
∂
(
(
¸
(
¸


.

\

∂
∂
+ −
∂
∂
(
(
¸
(
¸
−
+

.

\

∂
∂
x
T
k
x
T
k N uh
x
x
u e
e
t
T
c
T
e
k
T
β
β β
β
ρ
τ
ε
ε
ρ
ρ
ρ
ρ
ε
ε
ρ ρ
(330)
This is wellposed mathematically. The additional complexities include the calculation
of the partial derivatives of the internal energy. The boundary conditions using either
74
energy equation are the same; two mass flux BCs. Again, cyclic time relaxation is
applied to find a temperature field which produces a steady solution.
3.7 Dual Energy Equation Model (DEEM)
The Dual Energy Equation model solves the complete equation set, repeated here
as
0 =
∂
∂
+
∂
∂
x
m
t
ρ
(331)
  0
2 / 1
2
2 2
= + +


.

\

+
∂
∂
+
∂
∂
m sign
K
c
m m
K
p
m
x t
m
f
ε
ρ ρ
µε
ρ
(332)
( ) 0 = − − 
.

\

∂
∂
−
∂
∂
+
∂
∂
T T H a
x
T
k N mh
x t
E
v k β
(333)
( ) ( ) 0
1
= −
−
+
(
¸
(
¸
∂
∂
∂
∂
−
∂
∂
T T H a
x
T
k
x t
T
c
v β
β
β β
β
β
ε
ε
τ ρ (334)
This model captures the effects of finite heat transfer and matrix heat capacity as well as
flow friction. The characterization of this model depends on several parameters which
can be categorized.
Boundary mean temperatures: T
h
and T
c
Mass flowrates:
h
m& ,
c
m& , w, and
mc mh
φ φ −
Mean pressure: p
b
Macroscopic geometry: L
r
and D
r
Additionally, the model depends on several correlations which are functions of the
microscopic geometry and/or the flow field: N
k
,
β
τ , Nu, f , and ε .
75
3.8 Model Verification
The accuracy of the Dual Energy Equation model was verified by using a test
solution in a method similar to Kirkconnell (60). The particular solution was chosen such
that it retained the important characteristics of the actual solution. An arbitrary solution
can be made to satisfy the system of equations if appropriate source terms are added to
each equation, i.e.
( ) x t F
x
m
t
,
1
=
∂
∂
+
∂
∂ρ
(335)
  ( ) x t F m sign
K
c
m m
K
p
m
x t
m
f
,
2
2 / 1
2
2 2
= + +


.

\

+
∂
∂
+
∂
∂
ε
ρ ρ
µε
ρ
(336)
( ) ( ) x t F T T H a
x
T
k N mh
x t
E
v k
,
3
= − − 
.

\

∂
∂
−
∂
∂
+
∂
∂
β
(337)
( ) ( ) ( ) x t F T T H a
x
T
k
x t
T
c
v
,
1
4
= −
−
+
(
¸
(
¸
∂
∂
∂
∂
−
∂
∂
β
β
β β
β
β
ε
ε
τ ρ . (338)
The test solution was chosen as follows:
( )
3 2
cos M t M m − = ω (339)
( )
3 2 1
cos T t T T T − + = ω (340)
( )
3 2
cos P t P p P
b
− + = ω (341)
RT
P
= ρ (342)
ρ
m
u = (343)
76
T c E
v
ρ = (344)
T T =
β
(345)
( )
r
h c h
L
x
m m m M − + =
2
(346)
( )
r
mh mc mh
L
x
M φ φ φ − + =
3
(347)
( )
r
h c h
L
x
T T T T − + =
1
(348)
( )
r
h c h
L
x
T T T T
~ ~ ~
2
− + = (349)
2
3 3
π
− = M T (350)
( )
r
h c h
L
x
P P P P
~ ~ ~
2
− + = (351)
( )
r
Ph Pc Ph
L
z
P φ φ φ − + =
3
(352)
All of the parameters in the preceding equations come from the boundary conditions with
the exception of the boundary temperature and pressure oscillation amplitudes,
h
T
~
,
c
T
~
,
h
P
~
,
c
P
~
, and the pressure phase angles,
Ph
φ and
Pc
φ . These parameters are chosen
from a representative numerical solution. Equation (350) was chosen such that the net
enthalpy flux would be identically zero. Alternatively, a condition could be constructed
to produce a constant, nonzero net enthalpy flux. The zero enthalpy flux case was
chosen for simplicity. Adding a nonzero enthalpy flux will produce a similar solution
with a shift in the net enthalpy field. The source terms, F
i
, were then analytically solved.
77
The numerical solution was then calculated using the exact numerical method detailed
above. Since the initial condition can be set using the known exact solution, the model
converges in one cycle.
The model numerically solves the test equations to very high accuracy as
illustrated in Figure 8Figure 14. The net enthalpy flux is visually identically zero.
There is less that 0.2 mW deviation. The solution fields show very small errors. The
normalized errors are:
Density 5x10
7
Mass flux 3x10
7
Energy 4x10
7
Matrix temperature 3x10
8
Gas temperature 4x10
8
Pressure 4x10
7
The small normalized errors indicate that the numerical scheme is not introducing
unrealistic errors into the solution and that the solver can solve similar equations. The
source terms should not dramatically affect the character of the equations. Additionally,
the question of the proper set of boundary conditions is answered. This solution was
obtained using just two mass flux boundary conditions even though the strict
mathematical requirement is six boundary conditions.
78
0 0.005 0.01 0.015 0.02 0.025
0
10
20
30
Density [kg/m
3
] vs. Cycle Time [radians]
0 0.005 0.01 0.015 0.02 0.025
4
2
0
2
4
Velocity [m/s] vs. Cycle Time [radians]
0 0.02 0.04 0.06 0.08
0
10
20
30
Density [kg/m
3
] vs. Position [m]
rho
max
rho
mean rho
min
0 0.02 0.04 0.06 0.08
4
2
0
2
4
Velocity [m/s] vs. Position [m]
u
max
u
mean
u
min
0 0.005 0.01 0.015 0.02 0.025
2
1
0
1
2
Mass Flow Rate [g/s] vs. Cycle Time [radians]
0 0.005 0.01 0.015 0.02 0.025
2.5
3
3.5
4
Pressure [MPa] vs. Cycle Time [radians]
Figure 7 – Test case solution results.
79
0 0.02 0.04 0.06 0.08
3
2.5
2
1.5
1
0.5
0
0.5
x 10
8
Cycleaveraged mass flow [g/s]
Position
0 0.02 0.04 0.06 0.0
2
0
2
4
6
8
10
x 10
6
Cycleaveraged mass flow gradient [g/s]
Position
0 0.02 0.04 0.06 0.08
20
0
20
40
60
80
Cycleaveraged energy flows [mW]
Position
H
Qg
Qm
Total
0 0.02 0.04 0.06 0.0
40
20
0
20
40
60
Cycleaveraged enthalpy flow gradient [mW]
Position
Figure 8 – The net enthalpy flux satisfies the test case to within visual accuracy
(0.2 mW deviation)
80
Figure 9 – Maximum density error 8x10
6
kg/m
3
Figure 10 – Maximum mass flux error 1x10
5
kg/m
2
s
81
Figure 11 – Maximum energy error 2 kJ/m
3
Figure 12 – Maximum matrix temperature error 6x10
6
K
82
Figure 13 – Maximum gas temperature error 8x10
6
K
Figure 14 – Maximum pressure error 1.2 Pa
83
Before proceeding on to Chapters 4 and 5, it is worth reviewing what has been
accomplished up to this point. A mathematical model for the regenerator was developed
based on fundamental principles and guided by intuition. A numerical method was then
developed using the Method of Lines with central finite differences. An artificial
convergence technique was developed which dramatically improves convergence speed.
To provide a comparison with the full model, two asymptotic models were developed.
Chapter 4 describes the development of an experimental apparatus which was used to
investigate this model. Chapter 5 presents the modeling results and compares these
results to the experimental data from Chapter 4.
84
CHAPTER 4
4. EXPERIMENTAL APPARATUS AND MEASUREMENTS
4.1 Overview and Experimental Apparatus
The apparatus used is an Orifice Pulse Tube Cryocooler (OPTC) design and is
illustrated in Figure 15. With the exception of the regenerator materials, the identical
cooler assembly was used by Kirkconnell, et al. to characterize overall cryocooler
performance as a function of pulse tube aspect ratio and volume (61). Towards that end,
the apparatus was designed to accommodate five interchangeable regenerator/cold heat
exchanger/pulse tube assemblies (Figure 16), hereafter referred to as “expanders.” Using
the numbering scheme of the previous paper, expanders 1 through 3 are of identical
volume, and expanders 1, 4, and 5 are of constant length. The performance of the
constant volume pulse tubes (13) were found to be virtually indistinguishable, hence the
use of these three is ideal for the present effort in which the unique impact of regenerator
design on performance is sought. An unfortunate oversight resulted in the testing of one
of the regenerators in expander 5, which has a 17% larger pulse tube volume than the
constant volume set and performed slightly better in the previous experiments (no load
temperature of 76 K vs. 79 K). The differences between the performances of the various
regenerators were much larger than this pulse tube volume effect, however, so the data
from expander 5 are still considered relevant. The reader is referred to ref. 61 for
additional details on the apparatus.
85
Va c u u m En c lo s u r e
Co ld HX
P u ls e T ub e Re gen e r a t o r
T o Comp r e s s o r
Or i fi c e Va lve
H
2
O Co o la n t Lo o p H
2
O Co ola nt Lo o p
Fle x i b le Me mb r a n e
Ad a p t o r Pla t e
Re j e c t i o n HX In le t HX
T o Su r ge T a nk
Figure 15  Experimental Apparatus
Reproduced with authors’ consent (61)
Figure 16 – Regenerator/Pulse Tube expander module. Regenerator is the larger
diameter section.
Figure 17 shows the apparatus integrated with the vacuum system and data
acquisition system. Figure 18 shows a closeup view of the compressor. The compressor
used is a nominal 3 cc swept volume, dual opposed piston, Hughes Condor on loan from
Night Vision & Electronic Sensors Directorate (NVESD).
86
Figure 17 – Experimental apparatus with complete instrumentation.
Figure 18 – Closeup view of the compressor.
The chamber surrounding the expander is evacuated using a turbomolecular vacuum
pump, shown in Figure 19. This vacuum system is capable of achieving pressures as low
as 2x10
6
torr. These pressures are necessary to eliminate convective and conductive heat
87
transfer through the surrounding gas. Typically, natural convection will occur at
atmospheric pressures down a few torr. At these pressures, the heat carried by circulation
currents is very small, but conduction will continue at levels comparable to atmospheric
pressure. Around 10 torr, the conductivity of the low pressure gas will begin to decrease
until it is essentially zero around 10
4
torr (62). At this pressure and lower, the only mode
of heat transfer is radiation. Further reduction of the pressure is limited by slow moisture
release from the chamber and MLI. MLI (multilayer insulation) is a material used as a
radiation shield. It is composed of layers of aluminized mylar with woven Dacron spacer
layers. Typically, it is recommended that a vacuum level of 10
6
torr be achieved and
maintained before cooling a cryogenic system. At this pressure, the residual moisture is
low enough that it will not form significant frost and alter the MLI performance.
Figure 19 – Turbo vacuum pump station.
88
Four regenerators were tested in a previous study (63, 64). They consisted of
quite different geometry, but the same material. Since that report, an additional 45
micron regenerator and a diced foam metal regenerator were tested. All six regenerators
are made of stainless steel fabricated into different forms of porous materials. The
regenerators are summarized in Table 2 with their relevant data. The porosity of each
regenerator was measured experimentally by measuring the total matrix mass. The
density of the homogeneous material and the total regenerator volume can be used to
calculate the matrix porosity.
Table 2  Descriptive Summary of the Regenerators under Study
# Name / Figure Description
1 325 Mesh / Figure 20 Wire mesh screens  325 wires per inch
Wire diameter: 27.9 µm
Measured Porosity: 0.696
2 400 Mesh / not shown Wire mesh screens – 400 wires per inch
Wire diameter: 25.4 µm
Measured Porosity: 0.692
3 60 Micron / Figure 21
Alabama Cryogenic
Engineering (65)
Perforated disks
Pore size: 60 microns
Measured Porosity: 0.717 (spacers)
4 45 Micron
Alabama Cryogenic
Engineering
Perforated disks
Pore size: 45 microns
Measured Porosity: 0.644 (no spacers)
5 Foam Metal / Figure 22 Sintered foam metal plug
Measured Porosity: 0.614
In packing wire mesh screens, the actual porosity can vary by several percent
based on the degree of nesting. For wire screens perfectly aligned, the porosity can
approach 50%. For a sample of 30 400 mesh screens, the total thickness of the stack was
measured at 1.536 mm. This gives an average screen thickness of 51.2 microns, or a half
89
thickness of 25.6 microns. The half thickness is very close to the nominal wire diameter
suggesting that the screens were not nesting.
The 60 micron perforated disk thickness was measured at 211 microns. This
results in a perforation aspect ratio of 3.5. The 45 micron disks have an aspect ratio of
4.7 and 7.1 for the 30 micron disks. As the aspect ratio increases, the flow begins to
resemble that of a tube bundle.
The foam metal regenerator was fabricated from 100 micron particles which were
sintered. The result is a rigid matrix with random pore geometry and variable pore size.
The process is able to produce sintered matrices for a range of porosity. The process
used to make the sintered foam metal is proprietary, but shrinkage of the matrix during
sintering most likely requires beginning with a higher porosity. The target porosity of the
foam metal regenerator was chosen such that it was in the same range as the wire mesh
screens. The actual porosity was 8% less than the wire mesh regenerators. System level
modeling suggests that the regenerator performance is sensitive to the matrix porosity.
An 8% porosity difference may cause a significant performance change. The
morphology of the pore and solid structure is not studied in this dissertation, but the other
matrices have a much better understood geometry in terms of wire diameter and pitch or
perforation diameter and aspect ratio. The performance test with the foam metal
regenerator is the critical measure of the matrix performance. It is possible that sintered
matrices can provide reasonable regenerator performance, especially for a low cost
application where performance is not as critical.
90
The first three regenerators are composed of a stack of disks made from the
particular material (Figure 20 and Figure 21). These disks are packed into the
regenerator tube making sure that the disks are not bent. Each disk is then lightly tamped
to assure that it is snugly packed into the tube. The fourth regenerator is drastically
different from the others. This foam metal regenerator is a single plug of porous metal
(Figure 22). This allows significant time savings in packing the regenerator. Whereas
the single disk regenerators each take as much as eight hours to hand pack, the foam
metal regenerator takes only a few minutes. The perforated disks take considerably less
time to pack since they are about ten times as thick as 400 mesh screens. The disks, like
the screens, have to be restrained in the regenerator tube due to the spring force of the
stack. This makes the screens and the disks more difficult to pack at low porosities
which are desirable. Additionally, the perforated disks were packed using spacer rings
which create a 0.0015” gap between the disks to allow for flow redistribution between
disks.
Figure 20  Wire Mesh
(200X magnification)
Figure 21  Perforated Disk
(200X magnification)
Figure 22  Foam Metal
(200X magnification)
91
An extensive data acquisition system was designed to provide test control and
automated data collection. This system consists of a series of lab instruments which
provide a variety of functions and a PC (Figure 23). The instruments and PC are
connected using a GPIB 488 interface. GPIB is a laboratory instrumentation standard
which provides communication and control via a PC. This provides a realtime data
acquisition environment which provides for much more sophisticated measurements due
to automation. Agilent VEE Pro software was used to create the data acquisition and
control program. Several detail panels are shown in Figure 24, Figure 25, and Figure 26.
Figure 23 – Data acquisition system.
92
Figure 24  Data acquisition and control program. Main panel view.
Figure 25  Data acquisition and control program. Compressor panel view.
93
Figure 26  Data acquisition and control program. Pressure panel view.
The data acquisition system has been in development for six years. Initially the
objective was to measure a series of DC voltages such as thermocouples, diodes,
pressure, etc. As the program developed, communication with most of the test
instruments was incorporated allowing the program to monitor and display test data.
Further developments included doing real time data analysis which can be used to
automate and control the test. Pressure waveforms are acquired and stored for analysis.
Discrete Fourier transform analysis is used to calculate phase angles and to analyze the
frequency content of the various waveforms. The compressor electrical power input is
measured with a wattmeter, and controlled digitally with a feedback control algorithm by
the PC. The compressor voice coil voltage and current are measured as waveforms.
Considerable effort has been invested to use these waveforms to infer the piston motion
94
from a kinematic model for the compressor pistons. This would allow calculation of the
compressor PV power output, which would allow the compressor to be decoupled from
the rest of the system from an efficiency viewpoint. However, the results are yet
inconclusive. Tests need to be performed on a compressor which has piston position
indicators. Appendix 3 has been included to summarize the development of the
compressor model.
4.2 Experimental Results
Experimental data can be separated into two distinct categories. The first
category consists of data measured under steady flow and essentially isothermal
conditions for the purpose of determining friction factors. By varying the mass flow rate
of gas through the regenerator and measuring the pressure drop, the two coefficients that
characterize the momentum balance in the porous medium can be determined (ref. Figure
27 and Table 3). These are the Darcy permeability, K, and the Forchheimer inertia
coefficient, c
f
. The Darcy term, as discussed in previous chapters, is a measure of the
viscous effects while the Forchheimer term is a measure of the inertial effects. Both of
these terms are due to microscale, or pore scale, flow phenomenon which contributes to a
macroscopic pressure gradient. The hypothesis of this research is that the Darcy and
Forchheimer coefficients depend solely on the geometry of the porous media. Thus they
can be measured by any convenient method, i.e. it is not necessary to conduct a series of
mass flow versus pressure drop measurements at various temperatures. This hypothesis
is tested in Chapter 5 by comparing the calculated pressure ratios with the experimentally
95
measured pressure ratios during actual cryogenic operation. This does not provide
conclusive proof of the hypothesis. Direct pore scale simulations are required to test the
hypothesis, and this work is suggested for future investigations. The volume averaging
method produces a consistent model that can be tested via the pore scale simulations.
The data in Table 3 was calculated using a control volume model rather than a
differential model. As a result, the data is flawed and leads to artificially low friction
factors. This data is revisited in Chapter 5.
0
200
400
600
800
1000
1200
0.0 0.2 0.4 0.6 0.8 1.0 1.2
Flowrate [g/s]
P
r
e
s
s
u
r
e
D
r
o
p
[
k
P
a
]
400 Mesh
325 Mesh
60 Micron
45 Micron
Foam Metal
Figure 27  Steady Flow Pressure Drop
400 Mesh
96
Table 3 – Porous Media Parameter Summary (64)
ε K c
f
[%] [m
2
] 
Foam
Metal
61.37 1.473E10 0.5315
400
Mesh
69.20 1.005E10 0.5163
325
Mesh
69.61 1.060E10 0.3917
60
Micron
71.73 1.573E10 0.4296
Some of the previously calculated pressure ratios using steady flow data did not
correlate with experimentally observed pressure ratios (17, 17, 18, 19, 20, and 21). It is
the opinion of this writer that either their model was incomplete (erroneous simplifying
assumptions, presence of real gas effects, etc.) or that the Darcy and Forchheimer
coefficients are not only dependent on geometry but possibly frequency and gas
properties. Indeed, there are references in unrelated applications that address the issue of
frequency dependent permeability (66). The authors assume that this frequency
dependence is in fact a result of increasing inertial effects, which would indicate the
Forchheimer effect.
A second set of data consists of measurements under actual operating conditions
of the cryocooler. Of prime importance is the ability of the cryocooler to perform net
refrigeration at varying temperatures. This is measured by constructing load curves (ref.
Figure 28). These load curves provide an overall picture of the system performance.
These curves are also used to estimate the total parasitic load on the cooler. From this
data and the system level model, the pulse tube loss and the regenerator loss can be
estimated. However, these estimates are only as accurate as the model. By comparing
97
two load curves for the same cryocooler with only the regenerator changed, and under the
same operating conditions, the effect of the regenerator can be isolated in a sense.
0.0
0.5
1.0
1.5
2.0
2.5
70 90 110 130 150
Cold Tip Temperature [K]
C
o
l
d
T
i
p
A
p
p
l
i
e
d
L
o
a
d
[
W
]
400 Mesh, 3.1 MPa, 50 W
325 Mesh, 3.4 MPa, 55 W
60MPD, 3.0 MPa, 55 W
45MPD, 3.4 MPa, 60 W
Foam Metal, 3.1 MPa, 50 W
Figure 28  Load Curves
As mentioned previously, the 400 mesh regenerator was tested with a larger
volume pulse tube leading to a slightly more efficient expander. The temperature
difference at noload for the 400 mesh regenerator tested with the larger volume pulse
tube was only 23 K (61) compared to the nominally 15 K differences. Therefore the
comparison among the regenerators is still valid. The 400 mesh regenerator outperforms
all of the other regenerators even if it is tested with the smaller volume pulse tube. In
Chapter 5, the 400 mesh regenerator is modeled in Sage. This model accurately reflects
the pulse tube geometry.
98
The performance tests provide a direct comparison between the regenerator
matrices (63,64). The 400 mesh regenerator performed the best under the conditions of
fixed input power, i.e. lowest noload temperature. The fact that the 400 mesh
regenerator outperformed the other matrices is not immediately obvious from the
pressure drop data since the 400 mesh has one of the highest friction factors. This fact
illustrates that the heat transfer efficiency is significantly better for 400 mesh than the
other matrices.
The model will be shown to be driven by the pressures at the warm and cold end
of the regenerator. Thus it is important to measure the effect that the regenerator has on
the pressure wave. Figure 29, Figure 31, and Figure 30 illustrate that the pressure wave
is attenuated and phase shifted by the regenerator and pulse tube expander. The
oscillatory pressure drop is the magnitude of the instantaneous pressure difference across
the regenerator/pulse tube which is measured experimentally. This parameter is
influenced both by the difference in pressure amplitude as well as the pressure phase
shift. The pressure phase shift as measured is primarily occurring in the regenerator even
though the pulse tube is physically between the two points where the pressure is being
measured. The pulse tube has minimal effect on the pressure because of its relatively
short length and large diameter, i.e. it has a small friction factor. The ratio of the pressure
ratios is a normalized measure of the frictional losses occurring in the regenerator.
99
100
120
140
160
180
200
220
240
60MPD
3.0 MPa, 55 W
45MPD
3.4 Mpa, 60 W
Foam Metal
3.1 MPa, 50 W
325 Mesh
3.4 MPa, 55 W
400 Mesh
3.1 MPa, 50 W
O
s
c
i
l
l
a
t
o
r
y
P
r
e
s
s
u
r
e
D
r
o
p
,

P
1

P
2

[
K
p
a
]
Figure 29  Oscillatory Pressure Drop
The 400 mesh regenerator, which performed the best, has the largest oscillatory
pressure drop and the largest pressure phase shift. The large phase shift is partly
responsible for the magnitude of the oscillatory pressure drop. The 400 mesh and foam
metal regenerators were similarly restrictive for steady flow. However, the oscillatory
pressure drops are significantly different. The foam metal exhibited a smaller pressure
phase shift which can partly explain this smaller pressure drop. It cannot be concluded
from this data that the friction factor for the 400 mesh is higher for the oscillating flow.
The 400 mesh regenerator is modeled in Chapter 5. The pressure data for this
regenerator is used to validate the system level model. The pressure amplitudes and
phase shift are found to accurately characterize the model. The phase shift is found to be
very sensitive to the porosity, increasing by up to 10 degrees with a 2% decrease in the
100
porosity. The packing density of wire mesh screens has a large influence on the
regenerator performance.
The porosity of the 400 mesh and the 325 mesh were almost the same. The
percent open area is higher for 325 mesh screens (36% for 400 mesh versus 41.3% for
325 mesh.) Percent open area is based on the wire diameter and spacing. This helps to
explain why the 400 mesh has higher pressure drop.
8
10
12
14
16
18
20
22
24
60MPD
3.0 MPa, 55 W
45MPD
3.4 Mpa, 60 W
Foam Metal
3.1 MPa, 50 W
325 Mesh
3.4 MPa, 55 W
400 Mesh
3.1 MPa, 50 W
P
h
a
s
e
A
n
g
l
e
(
P
1
v
s
.
P
2
)
[
D
e
g
r
e
e
s
]
Figure 30  Pressure Wave Phase Angle
101
1.00
1.01
1.02
1.03
1.04
1.05
1.06
1.07
1.08
1.09
60MPD
3.0 MPa, 55 W
45MPD
3.4 Mpa, 60 W
Foam Metal
3.1 MPa, 50 W
325 Mesh
3.4 MPa, 55 W
400 Mesh
3.1 MPa, 50 W
P
1
P
r
e
s
s
u
r
e
R
a
t
i
o
/
P
2
P
r
e
s
s
u
r
e
R
a
t
i
o
Figure 31  Pressure Ratio Attenuation
The remaining 30 micron perforated disk regenerator remains to be tested.
Testing is planned for the near future. Completion of this test will provide a complete set
of data for this type of regenerator. It was a bit of a surprise to see the 60 micron and 45
micron regenerators perform almost the same. The expectation was that the smaller
perforations would provide better heat transfer for similar pressure drop. The results
show that the pressure drop was lower slightly higher than the 60 micron and
significantly lower than 325 mesh. However, the 325 mesh provided better cooling
performance.
The foam metal regenerator was retested by dicing it into 5 equal lengths using an
electron discharge machining (EDM) saw. The EDM was found to be the only process
which could cut the foam metal without closing the pores. Any abrasion or shearing
102
process produces a cut which completely closes the pores while EDM erodes the material
away forming a clean cut. The diced sections were then precisioncleaned and installed
into the regenerator using spacer rings made of laser cut 316 stainless steel shim stock.
These rings were intended to create a gap between the foam metal plugs of approximately
0.0015” to further reduce conduction. The spacer rings were cut with an OD equal to the
tube ID and a thickness of 0.013”.
Testing of the diced foam metal regenerator produced no measurable difference in
performance. This indicates that the poor performance was not due to the matrix
conduction. In Chapter 5, the regenerator conduction is found to be the smallest loss in
the regenerator, even smaller than the total conduction heat flux in the gas with the
addition of dispersion heat flux. Reducing the solid conduction will only reduce the total
regenerator loss by a few milliwatts. Dicing the foam metal should not affect any
characteristic of the regenerator other than the solid tortuosity. Locally, the solid
tortuosity approaches zero at the cut, but is quite high away from the cut.
The proposed Pyrex glass fiber regenerators did not turn out as expected. Fibers
with an average diameter of 19 microns were procured. These fibers were then packed
into glass sleeves in a variety of different manners. The best methodology was to chop
the fibers to produce a fiber felt which could be packed uniformly and to desired porosity.
This required pressing the glass felt to the desired porosity and then holding the felt into
the tube. The tubes were then furnaced. The complex phenomenon of the sintering
process made this a frustrating process. The furnace temperature needed to be high
enough such that the fibers would soften and sinter but not flow or slump. The first
103
attempt produced a well sintered slug of glass fiber but the overall dimensions had shrunk
by ~25% leaving an annular gap between the fibers and the tube. The second attempt
was performed at a lower temperature, but still within the softening range of the fibers.
Shrinkage was reduced to ~10% which is still too much. Any annular gap will destroy
the regenerator performance. The third attempt was performed at the same temperature
but with the fibers in compression using a dead weight and a piston. The piston length
was chosen such that the final fiber column would produce the desired porosity. Since
the fibers were in compression, they should expand radially to fill the tube during
shrinkage. The piston did not compress completely on this attempt.
The sintered fiber samples from all three attempts were examined under a
microscope to assess the sintering and fiber deformation (Figure 32.) It became quite
obvious that the fibers would shed very easily from the sintered plug. At this point, there
was no testable regenerator, and it was concluded that the fibers were not a viable
alternative for a regenerator.
104
Figure 32 – Sintered glass regenerator.
105
CHAPTER 5
5. RESULTS AND DISCUSSION
In this chapter, the volume averaged model developed in Chapter 2 will be shown
to be a flexible, accurate model. It is however a model of just the regenerator. A system
level model is needed to estimate the boundary conditions for the regenerator model, and
this model is developed in this chapter. The regenerator model is then used to obtain an
accurate simulation of the flow and heat transfer in the regenerator. The solution is
presented in terms of plots of important aspects of the solution, most importantly the
regenerator loss. The regenerator loss is caused by several effects such as temperature
oscillations due to finite matrix heat capacity, real gas properties, and temperature
difference due to finite heat transfer coefficient. Asymptotic models developed in
Chapter 3 are used to investigate the magnitude of these individual effects on the
regenerator loss and the mean temperature profile. The accuracy of this model is
considered by comparing it to other regenerator models and by calculating the entropy
generation discrepancy as defined by the Sage model. A last comparison shows that a
friction factor based on oscillating flow data is essentially identical to a friction factor
based on steady flow data if a compressible flow model is used. The effect of
compressibility is shown to account for a significant error in the measured friction factor.
106
5.1 Sage System Level Modeling
The regenerator model being developed is decoupled from the rest of the system
via the particular boundary conditions. To evaluate the accuracy of the model, accurate
boundary conditions for mass flow rate need to be applied. Ideally, these mass flow rates
would be measured experimentally at both ends of the regenerator via hot wire
anemometers or a similar method. With the apparatus used, these measurements were
not possible at the cold end of the regenerator due to the integral design of the regenerator
and pulse tube. The wall thickness of both tubes is only 0.007”. If the mass flow
boundary conditions are known, then the regenerator solution can be obtained with the
additional knowledge of the regenerator geometry, friction factor, heat transfer
coefficient, and dispersion coefficient.
The system performance is not completely known from just the regenerator
solution though. The expander, whether a Stirling piston or pulse tube, will produce a net
energy flow out of the cold volume which must carry the regenerator loss, other parasitic
loads, and any net refrigeration. Thus, to compare the system level experimental data
available from the laboratory pulse tube apparatus, there is a need for a system level
model. To this end, a commercially available modeling tool called Sage was used.
Sage is a graphical programming package. The different components are
represented by icons which are then connected to allow flow and energy transmission.
Figure 33 illustrates the diagram of the lab cryocooler. After building the Sage model
which contains all of the geometric aspects of the lab cooler, the model can be exercised
over the range of experimental data. Pressure data at the transfer line and before the
107
orifice provide the primary method of correlating the Sage model. To drive the model,
the compressor piston stroke needs to be specified. This data is unavailable
experimentally with the current compressor. However, the Sage optimizer can be used to
adjust the compressor stroke and orifice setting until the two pressure ratios match. The
additional degree of freedom is the pressure phase difference. This phase shift in the
pressure is due primarily to flow friction, most of which occurs in the regenerator. The
Sage model initially gave incorrect pressure phase shift predictions. Experimentation
indicated that the friction in the regenerator was too low, either due to an inaccurate
friction factor or incorrect porosity. The porosity was adjusted until the exact phase shift,
which was measured experimentally, was achieved. This porosity was then kept constant
for the 0.5 and 1.0 W cases. The resulting porosity was found to be 0.663 for 400 mesh.
The measured porosity was 0.692 calculated based on an assumed density and known
mass. This difference was found to be attributable to a density difference for the stainless
steel alloy used to make the screens which results in a 2% density difference. This is
comparable to the differences in the porosity.
Figure 33 – Sage diagram of the laboratory pulse tube apparatus.
108
Sage provides useful system level details such as the energy flow map in Figure
34. The critical energy balance for any cryocooler is based on a control volume
containing the cold heat exchanger. This map is based on the 400 mesh regenerator
operating at its coldest temperature of 76 K. At this temperature, and with 50 W of
electrical input power, the cooler produces zero net refrigeration. Considering the cold
heat exchanger control volume, the pulse tube enthalpy flow is consumed totally by the
regenerator loss and the regenerator wall conduction loss. The model illustrates that the
gross refrigeration, in this case 2.09 W, is rejected at the warm end of the pulse tube. The
surge volume accounts for a large portion of the total heat rejection which seems rather
contrary. The critical aspect of the warm end heat transfer is that energy is conserved.
1.884 W
76.51 K
2.09 W 2.09 W
293.9 K
1.86 W 44.45 W 1.884 W
292.6 K
42.57 W 0.0 W 0.233 W
1.86 W
37.9/6.54 W
0.206 W
Compressor Regenerator
Surge Volume
Cold HX Comp. HX/
Transfer Line
W
pv
/Q Q
rej,comp
Q
cond,regen
Q
net
Q
reject
Q
sv
Pulse Tube Reject HX
400 Mesh – 76 K, 0.0 W
Figure 34  System level energy flow diagram; 400 mesh, 76 K, 0 W
The additional experimental data which was used to correlate the Sage model was the
measured net refrigeration. The 400 mesh regenerator was operated with three
refrigeration loads; 0.0, 0.5, and 1.0 W. For each refrigeration load, the pressure ratios
and phase shifts were measured. The Sage predicted refrigeration was not accurate
109
initially. This is probably due to inaccuracies in the empirical pulse tube correlations
which Sage uses. Several phenomenons contribute to reduced pulse tube refrigeration
such as mass streaming and boundary convection, according to the Sage model. These
effects are modeled in Sage as enhanced diffusion effects based on some theoretical and
experimental correlations. These correlations were constructed for larger capacity pulse
tubes, so their accuracy in this size range is unknown. Sage provides a scale factor to
adjust these pulse tube losses. This factor was adjusted at the noload point and then it
was used for the 0.5 W and 1.0 W cases. The correlation was reasonably good; 1.072 W
predicted versus 1.0 W measured. This error is within experimental error and certainly
within the range of unknown radiation load.
Table 4 – 400 mesh baseline operating conditions
0.0 W 0.5 W 1.0 W
h
m& [g/s]
0.949 0.913 0.904
c
m& [g/s]
0.720 0.679 0.652
Frequency [Hz] 34 34 34
c h
m m & &
φ φ − [deg.]
15.4 16.4 17.0
T
h
[K] 292.6 292.6 292.7
T
c
[K] 76.51 88.44 100
Charge Pressure
[MPa] 3.1 3.1 3.1
The regenerator boundary parameters can then be extracted from the correlated
model. Table 4 summarizes the 400 mesh regenerator operating conditions. These
parameters were then used to perform comparison calculations with the model proposed
in Chapter 3. In this table, the mass flow rates decrease with increasing cold tip
temperature. This decrease in mass flow rate is accompanied by an increase in the mass
110
flow phase angle. This is a result of the decreasing regenerator loss which is proportional
to the phase difference between the pressure and mass flow waves.
5.2 Model Comparison
In this section, the results of the limiting case models presented in Chapter 3 are
illustrated in terms of the net regenerator enthalpy flow rate and the mean temperature
profile. The full regenerator model is then discussed beginning with a presentation of the
solution fields. Several postprocessing results such as net energy flows and net entropy
generation rates are then presented. The model is then compared with the Sage and
REGEN model solutions. Finally, the results of the analytic method of determining the
permeability and inertia coefficient from Chapter 2 are presented.
5.2.1 Net Enthalpy Flowrate  The Perfect Regenerator
The net energy flow rate in the regenerator represents a loss mechanism since this
energy must be carried by the gross refrigeration effect of the expander whether it be a
pulse tube or a Stirling displacer. The regenerator loss is a result of conduction and
dispersion in the gas, conduction in the matrix, and net enthalpy flow due to the mass flux
and enthalpy being partially in phase. Only in the limit of zero conduction or dispersion
and zero net enthalpy flux is the regenerator loss zero.
Figure 35 is a comparison of the calculated regenerator loss for the three models
(CTM, LTEM, and DEEM) with additional calculations to compare the effect of real gas
properties. The ideal gas cases additionally assume constant properties evaluated at the
111
mean temperature. The operating conditions were kept constant for all models. The
Constant Temperature model, as described in Chapter 3, is the most idealized model.
The only nonidealities included in this model are pressure drop, gas conduction and
dispersion, and matrix conduction. The resulting regenerator loss is only 0.33 W, due
mainly to conduction and dispersion. The addition of real gas properties results in a 0.8
W increase. This increase in the regenerator loss is due to the fact that the enthalpy for a
real gas depends on both temperature and pressure. For the ideal gas, enthalpy depends
only on temperature, which is constant for this model. As a result, the real gas enthalpy
oscillates partially in phase with the mass flux. For the case of ideal gas, constant
properties evaluated at the average of the end point temperatures were used. This has a
significant impact on the viscosity which is strongly dependent on temperature over the
range of interest.
The Local Thermal Equilibrium model was written only for ideal gas. So a good
comparison of the increase in the regenerator loss due to finite thermal capacity is the
ideal gas CTM and the LTEM. In this case, there is a 0.86 W increase. This increase in
regenerator loss is due to additional net enthalpy flow due to the local temperature
oscillation which is allowed in this model. As a result, the enthalpy oscillates locally in
phase with the temperature (and partially in phase with the mass flux.)
The Dual Energy Equation Model captures all of the nonidealities in the
regenerator, specifically the addition of finite temperature difference between the gas and
matrix. This temperature difference leads to an increase in net enthalpy flow of nearly a
112
watt (comparing the ideal gas LTEM and ideal gas DEEM.) This increase is due to a
more adverse (smaller) phase angle between enthalpy and mass flux.
Being able to decompose the total regenerator loss provides insight into the
relative importance of the competing design challenges with regenerators. Although
there are no direct simulations provided in this report, as the cold temperature decreases
below the 20 K point the problem of matrix heat capacity becomes a severely limiting
factor. The reason is made evident by comparing the CTM and LTEM results. A factor
that is present at any temperature is the inverse relationship between heat transfer
effectiveness and pressure drop. This analysis does not explicitly illustrate the effect of
pressure drop. Pressure drop represents a loss of available energy for the pulse tube or
expander piston to convert to refrigeration.
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
Constant T
Ideal Gas
Constant T
Real Gas
LTE
Ideal Gas
Dual E
Ideal Gas
Dual E
Real Gas
R
e
g
e
n
e
r
a
t
o
r
L
o
s
s
[
W
]
Figure 35 – Comparison of regenerator loss calculated with limiting case models
based on identical operating conditions (ideal gas assumes constant properties.)
113
5.2.2 The Mean Temperature Profile
At quasisteady state, the net diffusion and enthalpy fields are in a balance such
that the total energy flow through the regenerator is constant. The shape of the net
diffusion and enthalpy fields is strongly affected by the mean temperature profile. Model
convergence is primarily limited by the convergence of the steady mean temperature
profile. Figure 36 illustrates the effect of the different models on the mean temperature
profile. The CTM profile is nearly linear, with a slight, 3 K, deviation at the midpoint.
A dramatically different profile occurs with the addition of real gas properties, with the
midpoint temperature 36 K higher. This difference between the ideal gas and real gas
cases is partly due to the fact that the viscosity is temperature dependent has a significant
variation over the range of temperatures. The ideal gas case uses a constant viscosity
evaluated at the average temperature. The LTEM profile is higher at the midpoint by 9 K
and the ideal gas DEEM profile higher by 17 K. The real gas DEEM also shows a
dramatic deviation especially at the cold end, where the gradient is almost zero. An
interesting comparison exists between the two real gas cases. The mean temperature
profiles for the real gas CTM and DEEM represent the extremes of all the profiles. The
likely cause of this deviation is the finite matrix heat capacity in the DEEM. The CTM
assumes that the heat capacity is essentially infinite such that no temperature fluctuations
occur.
114
0
50
100
150
200
250
300
350
0 10 20 30 40 50 60 70 80
Position [mm]
M
e
a
n
T
e
m
p
e
r
a
t
u
r
e
[
K
]
Constant T
Ideal Gas
Constant T
Real Gas
LTE
Ideal Gas
Dual E
Ideal Gas
Dual E
Real Gas
Linear
Figure 36 – Comparison of mean temperature profiles calculated with the different
models based on identical operating conditions.
5.2.3 Baseline Regenerator Solutions
The regenerator model solves for density, mass flux, gas total energy, and matrix
temperature at periodic quasisteady state. From these solutions, pressure, velocity,
enthalpy, and entropy can be calculated. The local, instantaneous solutions are shown in
Figure 37. In the upper right, the density field is shown versus time (horizontal axis) and
position (family of curves with red and blue corresponding to hot and cold ends
respectively.) Below this figure, the density maximum and minimum profiles are plotted
versus position in the regenerator. This illustrates the relative difference in the spatial
and temporal scales for the density. The temporal scale (max versus min) is small
115
compares to the spatial scale (cold end versus warm end.) The velocity is given similar
treatment. The velocity at the cold end is much smaller than the velocity at the warm end
due to the large density difference. The mass flow rates (bottom left) are comparable in
magnitude but with a noticeable phase shift. Pressure is totally driven by the mass flow
boundary conditions (or vise versa with pressure boundary conditions.) The pressure
wave is attenuated and shifted in phase as it passes through the regenerator. Both of
these phenomenon are measured in the apparatus.
0 0.005 0.01 0.015 0.02 0.025
5
10
15
Density [kg/m
3
] vs. Cycle Time [radians]
0 0.005 0.01 0.015 0.02 0.025
5
0
5
Velocity [m/s] vs. Cycle Time [radians]
0 0.01 0.02 0.03 0.04 0.05
5
10
15
Density [kg/m
3
] vs. Position [m]
rho
max
rho
mean
rho
mi n
0 0.01 0.02 0.03 0.04 0.05
5
0
5
Velocity [m/s] vs. Position [m]
u
max
u
mean
u
mi n
0 0.005 0.01 0.015 0.02 0.025
0.5
0
0.5
Mass Flow Rate [g/s] vs. Cycle Time [radians]
0 0.005 0.01 0.015 0.02 0.025
2.8
3
3.2
3.4
Pressure [MPa] vs. Cycle Time [radians]
Figure 37 – Solutions plotted versus time and position. The middle plots are max
and min of density and velocity. Pressure and mass flow phase shifts are apparent.
The temperature field is calculated from the density and total energy using the
appropriate caloric equation of state (ideal or real gas.) In this problem, the temperature
116
is not specified on the boundaries. The predicted gas and matrix temperature waves at
the warm and cold ends are shown in Figure 38. The gas temperature leads the matrix
temperature, as would be expected. The gas temperature has a distortion which occurs at
the flow reversal. This is due to the reduction in the heat transfer coefficient at low
velocities. The flow reversal is also associated with the beginning of the expansion
process which causes a reduction in the gas temperature.
0 0.005 0.01 0.015 0.02 0.025
291.5
292
292.5
293
293.5
Warm end temperatures , Gas=r, Matrix=b, Th=g phiTh=75.2995
T
e
m
p
e
r
a
t
u
r
e
[
K
]
0 0.005 0.01 0.015 0.02 0.025
1000
500
0
500
1000
Warm end enthalpy flux phiHh=2.012
E
n
t
h
a
l
p
y
f
l
u
x
[
W
]
0 0.005 0.01 0.015 0.02 0.025
74
75
76
77
78
79
Cold end temperatures , Gas=r, Matrix=b, Tc=g phiTc=115.3045
Cycle time [sec.]
T
e
m
p
e
r
a
t
u
r
e
[
K
]
0 0.005 0.01 0.015 0.02 0.025
200
100
0
100
200
Cold end enthalpy flux phiHc=22.3202
Cycle time [sec.]
E
n
t
h
a
l
p
y
f
l
u
x
[
W
]
Figure 38 – Boundary solutions for the gas and matrix temperatures and enthalpy
flow rates.
The enthalpy is also calculated from the total energy and density. In the figure,
the enthalpy flux at the warm and cold end have extremely large amplitudes (1500 W and
117
300 W) compared to the DC component which is less than 2 W. It is this DC component
which is critical to be able to predict accurately.
The mean temperature profile for this baseline case is shown in Figure 39 in
comparison to a linear profile. It is this temperature profile which converges the solution
to periodic quasisteady state.
0 0.01 0.02 0.03 0.04 0.05
80
100
120
140
160
180
200
220
240
260
280
Position [m]
T
e
m
p
e
r
a
t
u
r
e
[
K
]
Figure 39 – Cycleaveraged gas temperature (red) versus a linear profile (blue).
The temperature difference (Figure 40) is a critical aspect to the regenerator
problem. This temperature difference is responsible for a large percentage of the
regenerator loss. The temperature difference is noticeably smaller at the cold end due to
considerably larger mean Reynolds number at the cold end as shown in Figure 41.
118
0 0.005 0.01 0.015 0.02 0.025 0.03
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
Cycle Time [sec.]
T
e
m
p
e
r
a
t
u
r
e
D
i
f
f
e
r
e
n
c
e
,
T

T
m
[
K
]
Figure 40 – Temperature difference versus position (multiple lines) and time.
0 0.01 0.02 0.03 0.04 0.05 0.06
42
44
46
48
50
52
54
56
58
60
62
Position [m]
M
e
a
n
R
e
y
n
o
l
d
s
N
u
m
b
e
r
Figure 41 – Mean Reynolds Number
5.2.4 First Law Results
Cycle averaging of the solutions during postprocessing provides the most
important aspects of the solution such as the net mass flow rate and enthalpy flow rate.
Figure 42 illustrates that at steady state, the net mass flow rate is very small, essentially
numerical noise. The net enthalpy and heat fluxes due to conduction and dispersion are
119
not zero and are shown in Figure 43. The critical point is that the total energy flux
through the regenerator is constant.
0 0.01 0.02 0.03 0.04 0.05 0.06
3.5
3
2.5
2
1.5
1
0.5
0
0.5
x 10
6
Position
C
y
c
l
e

a
v
e
r
a
g
e
d
m
a
s
s
f
l
o
w
[
g
/
s
]
Figure 42 – Cycleaveraged mass flow rate.
0 0.01 0.02 0.03 0.04 0.05 0.06
0
200
400
600
800
1000
1200
1400
1600
Position
H
Qg
Qm
Total
C
y
c
l
e

a
v
e
r
a
g
e
d
e
n
e
r
g
y
f
l
o
w
s
[
m
W
]
Figure 43 – Cycleaveraged energy flows. Regenerator total energy flow is constant
along regenerator.
120
5.2.5 Second Law Results
Entropy generation is calculated in postprocessing as well. The individual
sources of entropy generation are illustrated in Figure 44 and Figure 45 as functions of
position and time. The entropy generation due to viscous and inertial losses is the
dominant source, contributing almost 80% of the total available energy loss in this case.
The entropy generation due to the gas conduction and dispersion is characteristically
different than the entropy generation due to the matrix conduction. Since the dispersion
is proportional to the Peclet number, it is oscillatory as can be seen. The matrix
conduction only shows slight oscillation. The entropy generation due to interfacial heat
transfer is the second largest contributor to available energy loss, but it is the dominant
contributor to regenerator net enthalpy flow.
0 0.005 0.01 0.015 0.02 0.025 0.03
1
0
1
2
3
4
5
dP entropy generation rate [W/mK]
0 0.005 0.01 0.015 0.02 0.025 0.03
0
0.02
0.04
0.06
0.08
0.1
0.12
kg entropy generation rate [W/mK]
Cycle Time [sec]
Figure 44 – Entropy generation due to viscous and inertial losses (top), and entropy
generation due to conduction and dispersion in the gas (bottom). Multiple curves
represent different locations in the regenerator, plotted versus cycle time. Results
for 400 mesh baseline case.
121
0 0.005 0.01 0.015 0.02 0.025 0.03
0
0.005
0.01
0.015
0.02
km entropy generation rate [W/mK]
0 0.005 0.01 0.015 0.02 0.025 0.03
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
h entropy generation rate [W/mK]
Cycle Time [sec]
Figure 45 – Entropy generation due to matrix conduction (top), and entropy
generation due to interfacial convection (bottom). Results for 400 mesh baseline
case.
0 0.01 0.02 0.03 0.04 0.05 0.06
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Position [m]
E
n
t
r
o
p
y
g
e
n
e
r
a
t
i
o
n
r
a
t
e
[
W
/
m

K
]
Total
Viscous and inertial
Gas conduction and dispersion
Matrix conduction
Interfacial convection
Figure 46 – Cycleaveraged volumetric entropy generation rates plotted versus
position in the regenerator. Results for 400 mesh baseline case.
122
Figure 46 provides an interesting picture of the entropy generation mechanisms. The
viscous ad inertial entropy generation decreases to almost nothing at the cold end, but the
interfacial heat transfer component increases sharply at the cold end due to the 1/T
nature. The total entropy generation has a surprising shape with a minimum in the
interior of the regenerator. The area under these curves represents the net total entropy
generation, with units W/K.
5.2.6 Sage and REGEN Comparisons
Sage provides the solution grid for the regenerator. From this, the net energy flow
profiles can be constructed. Figure 47 is a comparison of Sage, the Georgia Tech model
proposed in this thesis, and profiles from a third model known as REGEN3.2.
REGEN3.2 was developed at the National Institute of Standards and Technologies
(NIST.) It models only the regenerator, but with a simplified model. All three models
were run with identical inputs. The profiles plotted for REGEN are after 100,000 cycles
of simulation time which took several hours. The solution is still not satisfactorily
converged as can be seen by the total energy flow profile which has a 10% variation
along the regenerator. The Sage profiles correlate roughly with the Georgia Tech model
which is the most visibly accurate. The Sage model apparently forces the matrix
conduction flux to zero at the ends giving the profiles an abrupt transition at the ends.
This produces a sharp corresponding increase in the enthalpy flux at the ends. The
REGEN seems to grossly over predict the total conduction which is counterintuitive since
the model wholly neglects dispersion.
123
0 0.01 0.02 0.03 0.04 0.05
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Position [m]
N
e
t
e
n
e
r
g
y
f
l
o
w
[
W
]
Sage h
Sage q
Sage h+q
GT h+q
GT q
GT h
REGEN h+q
REGEN h
REGEN q
0 0.01 0.02 0.03 0.04 0.05
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Position [m]
N
e
t
e
n
e
r
g
y
f
l
o
w
[
W
]
Sage h
Sage q
Sage h+q
GT h+q
GT q
GT h
REGEN h+q
REGEN h
REGEN q
Figure 47 – Comparison of Sagepredicted regenerator loss versus GT model.
REGEN, as mentioned previously, assumes that the pressure gradient is
negligible. The GT model does not make this assumption. The REGEN model as a
result of this assumption produces a nonconservative scheme while the GT model solves
a fully conservative set of equations. This may be the cause of the REGEN model not
converging to a constant energy flow rate as seen in the figure.
Table 5 is a summary of the comparison calculations with Sage. Sage predicts a
moderately higher regenerator loss. The regenerator loss is seen to decrease with
increasing cold end temperature. This source of error is approximately 8%. The effect of
this error directly translates into an error in the net refrigeration prediction. However,
this error is small in comparison to the uncertainty in the pulse tube net enthalpy flow.
124
Table 5 – Sage Comparison Summary
GT SAGE
Net refrigeration 0.0 0.5 1.0 0.0 0.5 1.0
Regenerator Loss 1.73 1.60 1.54 1.84 1.744 1.671
Viscous and inertial
lost power 15.3 14.2 14.1 15 14.18 14.02
Interfacial heat
transfer lost power 3.17 2.36 1.87 3.15 2.34 1.85
Gas conduction and
dispersion lost power 0.618 0.477 0.381   
Matrix conduction lost
power 0.129 0.105 0.087   
Total gas and matrix
conduction and disp. 0.747 0.582 0.469 0.713 0.553 0.441
Total lost power 19.2 17.2 16.4 18.9 17.1 16.3
Lost power based on
external calculation 19.2 17.2 16.4 19.0 17.2 16.4
Lost power
discrepancy 0.029% 0.012% 0.060% 0.734% 0.481% 0.270%
Warm end pressure
ratio 1.30 1.31 1.33 1.30 1.31 1.33
Cold end pressure
ratio 1.22 1.22 1.24 1.21 1.22 1.24
Warm end pressure
phase angle 42.6 43.1 44.9 42.7 43.27 45.11
Cold end pressure
phase angle 62.8 62.2 63.4 63.1 62.5 63.6
The available energy predictions correlate very well between the two models.
The Georgia Tech model demonstrates better available energy discrepancies (0.03%
versus 0.5%.) This discrepancy is calculated, in both models, by summing the internal
sources of entropy generation due to conduction, convection, dispersion, and flow
friction. The sum of the internal entropy generation rates is then compared to the entropy
generation rate determined by an external calculation, i.e. integrating the entropy flux at
the boundaries. Ideally, these two methods of calculating the entropy generation rate are
identical. However, due to numerical errors, these two methods will produce different
results. This discrepancy is then a measure of the accuracy of the numerical scheme.
125
The GT model is an order of magnitude more accurate than the Sage model. The
REGEN model does not provide this data, so a comparison cannot be made.
Pressure ratios and phase angles correlate well with each other as well as with
experimental data. The GT model requires that the mass flux be specified at both ends.
The pressure waves which develop are an output from the model. Ideally these pressure
waves should match experimental data in both amplitude and phase. The mass flux at the
ends are unknown experimentally, so the results of the Sage system level model are
heavily relied upon for this information. The pressure waves are sensitive to accurate
modeling of the friction factor in the regenerator. The following section provides a
discussion of the friction factor prediction from experimental data. These predictions
made using a compressible model and are shown to be comparable to friction factors
measured in oscillatory flow.
5.2.7 Permeability and Forchheimer Coefficient Predictions
Chapter 2 introduced a method to predict the permeability and Forchheimer
inertial coefficient using three analytical solutions to the continuity and momentum
equations. The results of this methodology are presented in Table 6  Table 8. The
compressible model is the most realistic since this model predicts the increase in pore
velocity as the pressure decreases in a steady flow test. The decrease in pressure, which
occurs in incompressible flow as well, leads to a density decrease for a gas flow. The
same predictions were made with the incompressible model. The compressible inertial
term is consistently 30% higher for all of the matrices tested while the permeability is
126
almost identical. The permeabilities should be identical since this parameter
characterizes the low Reynolds number regime of the flow where compressibility is much
less important.
The final comparison assesses the importance of the advective acceleration term
in the momentum equation. Both the permeability and inertial coefficient are identical
for all of the matrices tested. This indicates that the inclusion of the advective
acceleration term, at least for steady flow, is not necessary.
Table 6 – Permeability and Forchheimer Coefficient (compressible model)
K c
f
400 Mesh 2.69E11 0.407
325 Mesh 3.53E11 0.376
Foam Metal 2.80E11 0.445
60 Micron 4.91E11 0.372
45 Micron 2.21E11 0.259
Table 7 – Permeability and Forchheimer Coefficient (incompressible model)
K c
f
400 Mesh 2.69E11 0.282
325 Mesh 3.53E11 0.260
Foam Metal 2.80E11 0.308
60 Micron 4.91E11 0.257
45 Micron 2.21E11 0.179
Table 8 – Permeability and Forchheimer Coefficient (neglecting advective
acceleration)
K c
f
400 Mesh 2.69E11 0.407
325 Mesh 3.53E11 0.376
Foam Metal 2.80E11 0.445
60 Micron 4.91E11 0.372
45 Micron 2.21E11 0.259
127
A comparison of the friction factor predicted from measurements using 400 mesh
yield close agreement with the friction factor used in Sage as shown in Figure 48. The
friction factor and Reynolds number are defined using different length scales in Sage.
The friction factor and Reynolds number, as defined in Chapter 3, use the permeability
length scale which is measured experimentally. The length scale used in Sage is a
hydraulic diameter based on an approximate geometrical definition which depends on the
porosity and the wire diameter in the case of screens.
There is only a slight deviation in the two friction factors at high Reynolds
number. This is interesting in light of the methodology in which the Sage friction factors
were measured. The friction factor, heat transfer coefficient, and dispersion coefficient
for screens and felts were measured using an oscillatory flow test rig, in hopes of gaining
some improved predictions (19). It is obvious from this comparison that the flow
oscillations may not be nearly as important as the compressibility. It is suspected that the
friction factor depends only on the microscale geometry while the flow friction depends
on the actual macroscopic flow field and fluid properties. The compressible model
provides more motivation that this hypothesis is true.
128
1E01
1E+00
1E+01
1E+02
1E+03
1E+04
1E03 1E02 1E01 1E+00 1E+01 1E+02 1E+03
Reynolds Number
F
r
i
c
t
i
o
n
F
a
c
t
o
r
f_GT
f_sage*K1/
2/(2*dh)
Figure 48 – Friction factor comparison for 400 mesh screens
h
Sage
d
K
f
2
2 / 1
129
CHAPTER 6
6. CONCLUSIONS AND RECOMMENDATIONS
6.1 Conclusions
The governing equations for the flow and heat transfer which hold for a point in
space and time were volumeaveraged, producing a set of continuum equations for the
porous regenerator. These equations contained several source terms which represent the
viscous and inertial friction, gastomatrix heat transfer, and thermal dispersion. Several
additional terms were neglected based on scaling arguments. The simplified equations
describe unsteady compressible flow and heat transfer in a porous medium.
The dimensionless regenerator model was found to depend on two leading order
dimensionless parameters and two second order parameters. These parameters indicate
that the second order diffusion terms are small and should not be used as justification for
requiring additional boundary conditions. The resulting problem requires only two mass
flow boundary conditions and an appropriate method for finding the quasisteady initial
condition.
The oscillatory regenerator flow solution was found to depend strongly on the
mean temperature profile. Finding this mean profile with a semianalytical technique
allows the solution to be advanced in time much faster than with direct simulation. This
provides a huge computational advantage.
130
To validate the numerical method, an exact solution was found which required
additional source terms to be added to the governing equations. The numerical solution
agreed with the exact solution within O(10
6
) normalized errors.
A method for experimentally testing a cryocooler regenerator has been developed.
Pressure waves are measured at the warm ends of the regenerator/pulse tube. The net
refrigeration is measured with an applied heat load over a range of refrigeration
temperatures. A data acquisition system provides the ability to make high speed
measurements and for test control. Six regenerators were tested with a nearidentical
apparatus: 400 mesh, 325 mesh, 60 and 45 micron perforated disks, foam metal, and
diced foam metal. An additional 30 micron perforated disk regenerator is planned for
testing. An optional glass fiber regenerator was investigated but abandoned due to
manufacturability and reliability issues.
System level modeling gives detailed insight into the regenerator operation.
Matching the measured pressure amplitudes and phases in addition to the measured net
refrigeration with the system level model allows the regenerator boundary conditions to
be extracted.
The steady, compressible flow friction factor measurement using an analytical
solution matches with that measured in oscillating flow. The higher friction factor is
apparently due to the velocity acceleration which occurs with a gas flowing through a
porous media at high Reynolds number; it does not appear to depend on the flow
unsteadiness which several investigators have claimed.
131
6.2 Contributions
Beginning with the most significant contributions first, the application of the
volume averaging technique to the regenerator problem is the most noteworthy. This
approach removes the ambiguity which has existed in the regenerator literature for
decades. While the simplifications which were made reduce the problem to a similar
formulation found in the literature, the framework for future refinements and
investigations are solidly laid.
The numerical solution of the one dimensional model using a conservative
scheme is a significant contribution. This method is more accurate and reliable than any
existing method in the open literature due to the direct solution of a conservative system
of equations with a conservative numerical scheme. The lack of assumptions make this
model a test case for other models such as REGEN and Sage. Sage is a valuable system
level model, but the regenerator poses the largest source of irreversibility in the entire
cryocooler system. As a result, any inaccuracies in the regenerator can manifest
themselves as large net refrigeration errors. Having an accurate prediction tool allows the
Sage model to be verified.
The test apparatus and accompanying data acquisition system represents six years
of effort. The diagnostic capability of the system allows for real time data analysis.
Optimizing the cryocooler operation is assisted by data averaging, filtering, and trending.
Troubleshooting is much improved. With the addition of the Sage system level modeling
tool, experimentation and simulation can occur simultaneously giving the researcher the
ability to immediately understand the effect of a change in the operating conditions.
132
6.3 Future Work
The volumeaveraged equations were simplified in a manner which reduced the
macroscopic equations to a form which could be solved. The form of the surface
integrals representing the friction factor and heat transfer correlations reduced to familiar
forms. The strict assumptions under which these simplifications are allowed may not be
exactly true in the case of fully compressible flow with the large temperature and density
variations which are observed in the regenerator. A possible discrepancy which the
simplifications produce is due to the flow reversals which are occurring in the
regenerator. Investigators have long supposed that the friction factor and heat transfer
correlations from steady measurements deviate for oscillating flow. While it is more
likely that the friction factor depends only on the microscale geometry, direct, porescale
numerical simulations could be made to evaluate the surface integrals and assess the
effect of the flow oscillations. These types of simulations have been made for steady
flow, but none have been seen in the literature for oscillating, or even unsteady, flow. A
demonstration calculation was performed using Fluent, which is a commercially available
computational fluid dynamics software suite. The simulation consisted of constructing a
solid model for a representative, periodic cell of the porous media, in this case wire mesh.
A steady flow through the mesh was then simulated. Figure 49 and Figure 50 show some
preliminary results of this type of simulation. These simulations were performed for
steady flow only. Oscillating flow simulations are eventually a possibility.
133
Figure 49  Demonstration calculation for flow through wire mesh screens.
Visualization of surface pressure
Figure 50 – Demonstration calculation for flow through wire mesh screens.
Visualization of pathlines.
The surface and volume integrals representing the friction factor, heat transfer coefficient
and dispersion coefficient can be evaluated knowing the pore scale flow. More
importantly, oscillating flow and heat transfer can be modeled.
134
Admittedly, the heat transfer coefficient does depend on the flow, especially the
frequency. For “significantly high” frequencies, Newton’s Law of Cooling fails to hold
due to significant phase shifting between the bulk temperature and the interfacial
temperature. Kornhauser and Smith made this realization and adopted a complex Nusselt
number to model this phenomenon. In light of the Reynold’s Analogy, the friction factor
may also exhibit a similar effect at some frequency limit.
The macroscopic regenerator flow is commonly assumed to be onedimensional
although several investigators have noticed definite deviations. The flow passage
geometry leading into the regenerator will typically create a jet. The jet should be
dispersed rapidly, but it is not possible at this point to model such a phenomenon. Small
aspect ratio regenerators have lead to thermal instability problems producing, apparently,
stagnation zones due to flow asymmetry. These deviations are assumed to dramatically
deteriorate the regenerator performance. Proper system design should alleviate these
problems, but a model capable of capturing these effects would be useful. Such a model
would require, as a minimum, an additional radial coordinate. For most random media,
the friction factor could be assumed, initially, to be constant in the radial direction. For
such a model, the wall boundary conditions would need to be established. The radial
velocity would obviously be zero at the wall, but the axial velocity does not have an
obvious boundary condition at the wall (if one is required at all.) This needs to be studied
carefully.
Several improvements to the test apparatus will represent a significant
augmentation in ability to research more advanced concepts. Improved instrumentation
135
such as piezoelectric pressure transducers will greatly affect the reliability of the test
results. These piezoelectric pressure transducers which do not require the repeated
calibration and are less affected by temperature can be used in the cryogenic regions of
the cooler allowing for direct measurement of the pressure. This is increasingly
important for multistage cryocoolers. Anemometers can be designed with low helium
leakage and can be operated in cryogenic regions as well providing mass flow and
temperature measurements. A larger vacuum dewar will allow for calorimetric heat
exchanger measurements to be made which will provide better system level
understanding. It is possible to make modular components such as regenerators, pulse
tubes, heat exchangers, instrumentation blocks, inertance tubes, surge volumes, etc. This
gives more flexibility to investigate new concepts without redesigning the entire system.
Compressor LVDTs are a necessity for continued testing. These LVDTs are used to
measure the piston position as a function of time. With this information, the mechanical
input PV power can be calculated. This is an important piece of data not currently
available.
136
APPENDIX 1 – DERIVATION OF THE VOLUMEAVERAGED
GOVERNING EQUATIONS
A1.1 Volume Averaging Theory
The method of volume averaging is applied to the governing equations for flow in a
porous media. At the outset, the averaging theorems and principles are discussed. The
generalized governing equations of mass conservation, momentum balance and energy
conservation are developed for a single fluid phase flowing and interacting with a single
stationary solid (matrix) phase. The only assumptions that are made initially are that the
two phases do not react chemically with eachother and that the noslip assumption at the
fluidsolid interface is valid.
A1.1.1 Definitions
Within a porous media, an arbitrarily given volume, V, can contain portions of different
phases of material. Consider for example, a porous media with a single fluid phase, α,
and a single solid phase, β such as the system illustrated in Figure 4. Then the total
volume, V(t) is given as
. ) ( ) ( ) ( t V t V t V
β α
+ = (A11)
A property in the αphase, ω
α
, at a given position, x, in the porous media is written as a
function of space and time as
137
. 0 ) , (
, ) , (
=
=
t x
t x
β α
α α α
ω
ω ω
(A12)
where the subscript on the position vector indicates the phase at that position.
The phase average of a property is defined as the volumetric average over the
total volume. Explicitly, the phase average is
.
1
∫
=
V
dV
V
α α
ω ω (A13)
The intrinsic phase average is the volumetric average of a property over the phase
volume, i.e.
.
1 1 1
0
∫ ∫ ∫ ∫
=
(
(
(
(
¸
(
¸
+ = =
=
α β α
α
α
α α
α
α
α
α
α
ω ω ω ω ω
V V V V
dV
V
dV dV
V
dV
V
43 42 1
(A14)
By defining the volume fraction or porosity of the βphase, ε
α
, as
,
V
V
α
α
ε = (A15)
the phase average and intrinsic phase average can be related by
.
α
α α α
ω ε ω = (A16)
To transform the governing equations of mass, momentum, and energy for the porous
media, it is necessary to note a few theorems that relate averages of derivatives to
derivatives of averages. These theorems are given without proof. The reader is referred
to the literature for proofs of these theorems (67, 68).
138
A1.1.2 Transport Theorem
The volume average of a temporal derivative is related to the temporal derivative of an
average by the Transport Theorem given by
dS n w
V t t
A
α αβ α α
α
αβ
ω ω
ω
⋅ −
∂
∂
=
∂
∂
∫
1
(A17)
where A
αβ
is the area of the αβ interface in V, w
αβ
is the velocity of the αβ interfacial
surface in V, and n
α
is the unit normal vector on the αβ interface pointing into the β
phase. If the noslip assumption is valid, then
0 = ⇒ =
αβ β α
αβ
αβ
w v v
A
A
(A18)
This result requires the surface integral in Equation (A27) to be zero, and this leads to a
simplified transport theorem valid for a noslip porous media given by
.
α
α
ω
ω
t t ∂
∂
=
∂
∂
(A19)
A1.1.3 Spatial Averaging Theorem
The volume average of a spatial derivative is related to the spatial derivative of an
average by the Spatial Averaging Theorem given by
.
1
dS n
V
A
∫
+ ∇ = ∇
αβ
α α α α
ω ω ω (A110)
This transformation generates an additional term represented by the surface integral even
for a noslip porous media. This additional term represents a macroscopic contribution
caused be the microscopic interaction of the two phases.
139
A1.1.4 Modified Averaging Theorem
Equation (A210) can be modified by application of Equation (A26). Note that
( )
( )
dS n
V
dS n
V
dS n
V
dS n
V
A A
A
A
∫ ∫
∫
∫
+ + ∇ + ∇ =
+ + ∇ + ∇ =
+ ∇ = ∇
αβ αβ
αβ
αβ
α α
α
α α α
α
α
α
α α
α
α
α α α
α
α
α
α α
α α
α
α α α
ω ω ε ω ω ε
ω ω ε ω ω ε
ω ω ε ω
ˆ
1 1
ˆ
1
1
(A111)
where
α
ωˆ is the deviation of ω
α
from its intrinsic phase average defined by
. ˆ
α
α α α
ω ω ω − = (A112)
The intrinsic average that appears in the first surface integral is essentially constant over
the surface (42). As a result, the intrinsic average can be removed from the surface
integral giving
dS n
V
dS n
V
dS n
V
A
A A
∫
∫ ∫
+ ∇ =
+ + ∇ + ∇ = ∇
−∇
αβ
αβ
α
αβ
α α
α
α α
α α
ε
α
α
α α
α
α
α
α α α
ω ω ε
ω ω ε ω ω ε ω
ˆ
1
ˆ
1 1
43 42 1
(A113)
This leads to the Modified Averaging Theorem given by
dS n
V
A
∫
+ ∇ = ∇
αβ
α α
α
α α α
ω ω ε ω ˆ
1
(A114)
This is referred to as the scale decomposition since the parameter is decomposed into a
macroscopic and microscopic component. Note that several consequences arise from this
140
term. First, in the αphase, the deviation can be nonzero, but in the βphase, this term is
identically zero by Equation (A22). Also note that
. 0 ˆ ˆ = =
α
α α
ω ω (A115)
These principles and theorems can now be applied to the governing equations to develop
a generalized set of averaged governing equations.
A1.2 Application to the Governing Equations
The preceding volume averaging theorems can now be applied to the governing
equations. In doing so, the continuum flow which exists at the microscopic level is
transformed into a continuum flow at the macroscopic level. The result is the production
of various terms representing the microscale influence on the macroscopic flow.
A1.2.1 VolumeAveraged Conservation of Mass Equation
The phase average of Equation (213) is
( ) . 0 = ⋅ ∇ +
∂
∂
α α
α
ρ
ρ
u
t
r
(A116)
In developing this volumeaveraged equation, we will assume that the noslip assumption
is valid at the fluid solid interface. We will also neglect gradients in the porosity.
Application of the Transport Theorem (A29) to the first term in Equation (A216) yields
.
α
α
ρ
ρ
t t ∂
∂
=
∂
∂
(A117)
Definition (A26) can be used to obtain
141
.
α
α α
α
ρ ε
ρ
t t ∂
∂
=
∂
∂
(A118)
Application of the Averaging Theorem (A210) to the second term in Equation (A216)
yields
( ) .
1
dS u n
V
u u
A
∫
⋅ + ⋅ ∇ = ⋅ ∇
αβ
α α α α α α α
ρ ρ ρ
r r r
(A119)
The surface integral in the last equation evaluates to zero due to the no slip assumption
reducing the equation to
( ) .
α α α α
ρ ρ u u
r r
⋅ ∇ = ⋅ ∇ (A120)
Definitions (A212) and (A215) can be used to obtain
( ) ( ) .
ˆ
ˆ
α α
α
α
α
α α α α
ρ ρ ε ρ u u u
r r r
⋅ ∇ + ⋅ ∇ = ⋅ ∇ (A121)
Thus, the volumeaveraged continuity equation becomes
( ) . 0
ˆ
ˆ
1
= ⋅ ∇ + ⋅ ∇ +
∂
∂
4 43 4 42 1
r
4 4 3 4 4 2 1
r
43 42 1
dispersion mass
convection
on accumulati
u u
t
α α
α
α
α
α
α
α
α
ρ
ε
ρ ρ (A122)
This volumeaveraged form of the continuity equation contains an additional term which
was not in the point equations. This term represents the dispersive mass transport. This
is a generalized continuity equation for 3D flow in a porous media. The momentum
equation can be averaged in the same manner.
142
A1.2.2 VolumeAveraged Balance of Momentum Equation
The phase average of Equation (214) is
( ) ( ) ( ) ( ) . 0
3
1
= ∇ ⋅ ∇ − ⋅ ∇ ∇ − ∇ + ⋅ ∇ +
∂
∂
α α α α α α α α α α
µ µ ρ ρ u u p u u u
t
r r r r r
(A123)
As with the continuity equation, we will assume that the noslip assumption is valid and
that gradients of the porosity are negligible. Application of the Transport Theorem (A2
10) to the first term in Equation (A223) yields
( ) .
α α α α
ρ ρ u
t
u
t
r r
∂
∂
=
∂
∂
(A124)
Definitions (A212) and (A215) can be used to obtain
.
ˆ
ˆ
α α
α
α
α
α α α α
ρ ρ ε ρ u u u
r r r
+ = (A125)
Substitution of Equation (A225) into Equation (A224) yields
( ) ( ) .
ˆ
ˆ
α α
α
α
α
α α α α
ρ ρ ε ρ u
t
u
t
u
t
r r r
∂
∂
+
∂
∂
=
∂
∂
(A126)
The Averaging Theorem (A210) can be applied to the second term in Equation (A223)
to obtain
( ) .
α α α α α α
ρ ρ u u u u
r r r r
⋅ ∇ = ⋅ ∇ (A127)
Definitions (A212) and (A215) can be used to expand
α α α
ρ u u
r r
to the form
.
ˆ ˆ
ˆ
ˆ
ˆ
ˆ
ˆ
ˆ ˆ
α α α
α
α α α
α α
α
α α α
α
α
α
α
α
α
α
α α α α α
ρ ρ
ρ ρ ρ ε ρ
u u u u
u u u u u u u u
r r r r
r r r r r r r r
+ +
+ + =
(A128)
Thus Equation (A227) becomes
143
( ) ( ) ( ) ( )
( ) .
ˆ ˆ
ˆ
ˆ
ˆ
ˆ
ˆ
ˆ ˆ
α α α
α
α α α
α α
α
α α α
α
α
α
α
α
α
α
α α α α α
ρ ρ
ρ ρ ρ ε ρ
u u u u
u u u u u u u u
r r r r
r r r r r r r r
⋅ ∇ + ⋅ ∇ +
⋅ ∇ + ⋅ ∇ + ⋅ ∇ = ⋅ ∇
(A129)
The Modified Averaging Theorem (A214) can be applied to the third term in Equation
(A223) to produce
. ˆ
1
∫
+ ∇ = ∇
αβ
α α
α
α α α
ε
A
dS p n
V
p p (A130)
Now proceed to expand the viscous terms in Equation (A223). The viscosity can be
considered to be constant within the averaging volume, and the last term in Equation
(A223) can be written as
( ) ( ) .
1
dS u n
V
u u
A
∫
∇ ⋅ + ∇ ⋅ ∇ = ∇ ⋅ ∇
αβ
α α α α α α α
µ µ µ
r r r
(A131)
Applying the modified averaging theorem again produces
( ) .
1
ˆ
1
dS u n
V
dS u n
V
u u
A A
∫ ∫
∇ ⋅ +


.

\

(
(
¸
(
¸
+ ∇ ⋅ ∇ = ∇ ⋅ ∇
αβ αβ
α α α α α
α
α α α α α
µ ε µ µ
r r r r
(A132)
Again the surface integral, dS u n
V
A
∫
αβ
α α
ˆ
1 r
, evaluates to zero by the noslip assumption
which simplifies the last equation to
( ) ( ) dS u n
V
u u
A
∫
∇ ⋅ + ∇ ⋅ ∇ = ∇ ⋅ ∇
αβ
α α α
α
α α α α α
µ µ ε µ
r r r 1
(A133)
which can be expanded to
144
( ) ( )
.
ˆ
1 1
dS u n
V
dS u n
V
u u
A A
∫ ∫
∇ ⋅ + ∇ ⋅ +
∇ ⋅ ∇ = ∇ ⋅ ∇
αβ αβ
α α α
α
α α α
α
α α α α α
µ µ
µ ε µ
r r
r r
(A134)
The first surface integral also evaluates to zero if the porosity is constant. The final form
of the viscous term is
( ) ( ) .
ˆ
1
dS u n
V
u u
A
∫
∇ ⋅ + ∇ ⋅ ∇ = ∇ ⋅ ∇
αβ
α α α
α
α α α α α
µ µ ε µ
r r r
(A135)
The compressible viscous term in Equation (A223) can similarly be expanded to
( ) ( ) .
ˆ
3
1
3 3
1
dS u n
V
u u
A
∫
⋅ ∇ + ⋅ ∇ ∇ = ⋅ ∇ ∇
αβ
α α
α
α
α α
α
α α
µ
µ
ε
µ
r r r
(A136)
Substitution of Equations (A226), (A229), (A230), (A236), and (A234) into Equation
(A223) yields
( ) ( ) ( ) ( )
( ) ( )
( ) . 0
ˆ ˆ
ˆ
1
ˆ
ˆ
1
ˆ
ˆ
1
ˆ ˆ
1
ˆ
ˆ
1
ˆ ˆ
3
ˆ
~ 1
3
1
= ⋅ ∇ + ⋅ ∇ +
⋅ ∇ + ⋅ ∇ +
∂
∂
+
(
¸
(
¸
∇ −

.

\

⋅ ∇ − ⋅ +
∇ ⋅ ∇ − ⋅ ∇ ∇ − ∇ + ⋅ ∇ +
∂
∂
∫
α α α
α
α
α α α
α
α α
α
α
α
α α
α
α
α
α α
α
α α α
α
α α
α
α
α α
α
α α
α
α
α
α
α
α
α
α
α
α
α
α
ρ
ε
ρ
ε
ρ
ε
ρ
ε
ρ
ε
µ
µ
µ µ ρ ρ
αβ
u u u u
u u u u u
t
dS u u p I n
V
u u p u u u
t
A
r r r r
r r r r r
r r
r r r r r
(A137)
In the last step, three surface integrals were combined and the whole equation was
divided by porosity. We can take advantage of the continuity equation (A222) to re
write the momentum equation in weak conservation form as
145
( ) ( )
( ) ( )
. 0
ˆ ˆ
ˆ
1
ˆ
ˆ
1
ˆ
ˆ
1
ˆ ˆ
1
ˆ
ˆ
1
ˆ ˆ
3
ˆ
~ 1
3
1
= ⋅ ∇ + ∇ ⋅ +
⋅ ∇ + ⋅ ∇ +
∂
∂
+
(
¸
(
¸
∇ −

.

\

⋅ ∇ − ⋅ +
∇ ⋅ ∇ − ⋅ ∇ ∇ − ∇ + ∇ ⋅ +
∂
∂
∫
α α α
α
α
α α α
α
α α
α
α
α
α α
α
α
α
α α
α
α α α
α
α α
α
α
α α
α
α α
α
α
α
α
α
α
α
α
α
α
α
α
ρ
ε
ρ
ε
ρ
ε
ρ
ε
ρ
ε
µ
µ
µ µ ρ ρ
αβ
u u u u
u u u u u
t
dS u u p I n
V
u u p u u u
t
A
r r r r
r r r r r
r r
r r r r r
(A138)
A1.2.3 VolumeAveraged αPhase Conservation of Energy Equation
The phase average of Equation (215) is
( )
( ) ( ) . 0 = ∇ ⋅ ∇ − ⋅ ∇ +
∂
∂
α α α α α
α α
ρ
ρ
T k h u
t
e r
(A139)
The same assumptions for noslip and gradients of porosity will be applied to the energy
equation as well. Application of the averaging theorems to the accumulation term in
Equation (A239) yields
( )
( )
.
ˆ ˆ
t
e
t
e
t
e
t
e
∂
∂
+
∂
∂
=
∂
∂
=
∂
∂
α α
α
α
α
α
α
α α
α α
ρ ρ
ε
ρ
ρ
(A140)
Application of the averaging theorems to the convection term in Equation (A239) yields
( ) ( ) ( ) ( )
( ) .
ˆ ˆ
ˆ
ˆ
ˆ
ˆ ˆ ˆ
ˆ
α α α
α
α α α
α α
α
α
α
α α α
α
α
α
α
α
α α α α α
ρ ρ
ρ ρ ρ ε ρ
h u h u
h u h u h u h u
r r
r r r r
⋅ ∇ + ⋅ ∇ +
⋅ ∇ + ⋅ ∇ + ⋅ ∇ = ⋅ ∇
(A141)
Finally, the conduction term in Equation (A239) can be simplified to
146
( )  
.
1
ˆ
1
1
∫ ∫
∫
∇ ⋅ −
(
(
¸
(
¸


.

\

+ ∇ ⋅ −∇ =
∇ ⋅ − ∇ ⋅ −∇ = ∇ ⋅ ∇ −
αβ αβ
αβ
α α α α α
α
α α α
α α α α α α α
ε
A A
A
dS T k n
V
dS T n
V
T k
dS T k n
V
T k T k
(A142)
The last step in Equation (A242) involves decomposing the point temperature according
to the decomposition definition. Substituting Equations (A240), (A241), and (A242)
into Equation (A239) gives
( )
( )
( )
( ) ( ) . 0
ˆ ˆ
ˆ
1
ˆ
ˆ
1
ˆ ˆ
1
ˆ
ˆ
1
ˆ ˆ
1 1 1
ˆ
1 1
= ⋅ ∇ + ⋅ ∇ + ⋅ ∇ +
⋅ ∇ +
∂
∂
+ ∇ ⋅ −
(
(
¸
(
¸


.

\

+ ∇ ⋅ ∇ − ⋅ ∇ +
∂
∂
∫
∫
α α α
α
α
α α α
α
α α
α
α
α
α
α α α
α
α α
α
α α α
α
α α
α
α α α
α
α
α
α
α
α
α
α
α
α
α
ρ
ε
ρ
ε
ρ
ε
ρ
ε
ρ
ε ε
ε
ε
ρ
ρ
αβ
αβ
h u h u h u
h u
t
e
dS T k n
V
dS T n
V
T k h u
t
e
A
A
r r r
r
r
(A143)
A1.2.4 VolumeAveraged βPhase Conservation of Energy Equation
The phase average of Equation (216) is
( ) ( ) 0 = ∇ ⋅ ∇ −
∂
∂
β β
β
β
ρ T k
t
T
c
v
(A144)
which can be readily simplified to
( )
. 0
1
ˆ
1
= ∇ ⋅ −
(
(
¸
(
¸


.

\

+ ∇ ⋅ ∇ −
∂
∂
∫
∫
αβ
αβ
β β β
β
β β
β
β
β β
β
β
β
ρ
A
A
v
dS T k n
V
dS T n
V
T k
t
T
c
(A145)
147
A1.2.5 VolumeAveraged αPhase Entropy Generation Equation
The phase average of Equation (217) is
( )
( ) . 0
,
≥


.

\

⋅ ∇ + ⋅ ∇ +
∂
∂
= ′ ′ ′
α
α
α α α
α α
α
ρ
ρ
T
q
u s
t
s
s
gen
r
r
(A146)
The first term on the right in Equation (A246) representing the time rate of change of
entropy can be expanded to obtain
( )
( )
.
ˆ ˆ
t
s
t
s
t
s
t
s
∂
∂
+
∂
∂
=
∂
∂
=
∂
∂
α α
α
α
α
α
α
α α
α α
ρ ρ
ε
ρ
ρ
(A147)
In developing this volumeaveraged equation, the noslip assumption is valid at the fluid
solid interface. Gradients in the porosity will also be neglected. Now applying the
averaging theorems to the second term on the right of Equation (A246) representing the
convection of entropy gives
( ) ( ) ( ) ( )
( ) .
ˆ
ˆ ˆ
ˆ
ˆ
ˆ
ˆ ˆ ˆ
α α α
α
α α α
α α
α
α
α
α α α
α
α
α
α
α
α α α α α
ρ ρ
ρ ρ ρ ε ρ
u s s u
u s u s u s u s
r r
r r r r
⋅ ∇ + ⋅ ∇ +
⋅ ∇ + ⋅ ∇ + ⋅ ∇ = ⋅ ∇
(A148)
Using Fourier’s Law, the third term on the right of Equation (A246) can be expanded to
∫
∇
⋅ −
(
¸
(
¸
∇
⋅ −∇ =


.

\
 ∇
⋅ ∇ −
αβ
α
α α
α
α
α
α
α
α α
A
dS
T
T k
n
V T
T
k
T
T k 1
(A149)
Substituting Equation (A247) through Equation (A249) into Equation (A246) produces
148
( )
( ) ( ) ( )
( )
∫
∇
⋅ −
(
¸
(
¸
∇
⋅ ∇ −
⋅ ∇ + ⋅ ∇ +
⋅ ∇ + ⋅ ∇ + ⋅ ∇ +
∂
∂
+
∂
∂
= ′ ′ ′
αβ
α
α α
α
α α
α
α
α
α α α
α
α
α α α
α
α α
α
α
α
α
α α α
α
α
α
α
α
α
α
α α
α
α
α
α
α
α
α
ε
ρ
ε
ρ
ε
ρ
ε
ρ
ε
ρ
ρ
ε
ρ
A
gen
dS
T
T k
n
V T
T
k
u s s u
u s u s u s
t
s
t
s
s
1 1
ˆ
ˆ ˆ
1
ˆ
ˆ
1
ˆ
ˆ
1
ˆ ˆ
1
ˆ ˆ
1
,
r r
r r r
(A150)
A1.2.6 VolumeAveraged βPhase Entropy Generation Equation
The phase average of Equation (218) is
( )
0
,
≥


.

\

⋅ ∇ +
∂
∂
= ′ ′ ′
β
β β β
β
ρ
T
q
t
s
s
gen
r
(A151)
which can be readily expanded to
∫
∇
⋅ −
(
(
¸
(
¸
∇
⋅ ∇ −
∂
∂
= ′ ′ ′
αβ
β
β β
β
β β
β
β
β
β
β
β
β
β
ε
ρ
A
gen
dS
T
T k
n
V T
T
k
t
s
s
1 1
,
(A152)
The volume averaged equations have been developed in sufficiently generalized
form. In the present form, the equations do not present a tractable problem. Eventually,
these equations will be represented entirely in terms of volume averaged quantities. The
details of this simplification are the subject of Chapter 2.
149
APPENDIX 2 – DERIVATION OF DIFFERENTIATION
OPERATORS USING MATLAB
Spatial differentiation can be expressed as a linear operator using a variety of
discrete differentiation schemes. In general, the i
th
derivative of a quantity, a, can be
calculated as
j ij
i
a D
x
a
≅ 
.

\

∂
∂
(A21)
where D
ij
is the linear differentiation operator. Higher order differentiation can be
performed using either a higher order operator or via recursive operations, i.e
( )
k jk ij
i
a D D
x
a
≅


.

\

∂
∂
2
2
(A22)
These operators can be of variable order or method. In this research, it was found
that central finite difference operators performed well due to the diffusive nature of the
continuity equation. Second order central and fourth order central differences were
employed. It was noticed that the second order central operators were more stable than
fourth order, but the fourth order provided more accuracy. It is possible to construct
upwind or biased difference operators. These gave little success for this problem.
150
Any finite difference operator can be constructed by forming a Taylor series
expansion:
( )
( )
( )
( )
( )
( )
( )
( )
...
! 4 ! 3 ! 2 ! 1
4
4
3
3
2
2
1
1
+
∆ −
+
∆ −
+
∆ −
+
∆ −
+ =
− i i i i i n i
u
x n
u
x n
u
x n
u
x n
u u
M
( )
( )
( )
( )
( )
( )
( )
( )
...
! 4
2
! 3
2
! 2
2
! 1
2
4
4
3
3
2
2
1
1
2
+
∆ −
+
∆ −
+
∆ −
+
∆ −
+ =
− i i i i i i
u
x
u
x
u
x
u
x
u u
( )
( )
( )
( )
( )
( )
( )
( )
...
! 4
1
! 3
1
! 2
1
! 1
1
4
4
3
3
2
2
1
1
1
+
∆ −
+
∆ −
+
∆ −
+
∆ −
+ =
− i i i i i i
u
x
u
x
u
x
u
x
u u
( )
( )
( )
( )
( )
( )
( )
( )
...
! 4
1
! 3
1
! 2
1
! 1
1
4
4
3
3
2
2
1
1
1
+
∆ +
+
∆ +
+
∆ +
+
∆ +
+ =
+ i i i i i i
u
x
u
x
u
x
u
x
u u
( )
( )
( )
( )
( )
( )
( )
( )
...
! 4
2
! 3
2
! 2
2
! 1
2
4
4
3
3
2
2
1
1
2
+
∆ +
+
∆ +
+
∆ +
+
∆ +
+ =
+ i i i i i i
u
x
u
x
u
x
u
x
u u
M
( )
( )
( )
( )
( )
( )
( )
( )
...
! 4 ! 3 ! 2 ! 1
4
4
3
3
2
2
1
1
+
∆ +
+
∆ +
+
∆ +
+
∆ +
+ =
+ i i i i i n i
u
x n
u
x n
u
x n
u
x n
u u (A23)
By forming a truncated linear series, the desired operator can be created. For example,
the three point central differences for the first and second derivative are formed by taking
a linear series of the i+1 and i1 points:
( )
( )
( )
( )
( )
3 2
2
1
1
1
! 2
1
! 1
1
x O u
x
a u
x
a au au
i i i i
∆ +
∆ −
+
∆ −
+ =
−
(A24)
( )
( )
( )
( )
( )
3 2
2
1
1
1
! 2
1
! 1
1
x O u
x
b u
x
b bu bu
i i i i
∆ +
∆ +
+
∆ +
+ =
+
(A25)
Summing the threepoint approximation equations gives
151
( )
( ) ( )  
( )
( ) ( )  
( )
( )
3 2
2
2 2 1
1
1 1
1 1
! 2
1 1
! 1
1 1 x O u
x
b a u
x
b a
bu u b a au
i i
i i i
∆ +
∆
+ − +
∆
+ − =
+ + −
+ −
(A26)
which can be solved for the first and second derivatives,
( )
( )
( ) ( )  
) (
! 1
1 1
2
1
1 1
1 1
1
x O
x
b a
bu u b a au
u
i i i
i
∆ +
∆
+ −
+ + −
=
+ −
,
( ) ( )
( ) ( )
(
¸
(
¸
=
(
¸
(
¸
(
¸
(
¸
−
−
0
1
1 1
1 1
2 2
1 1
b
a
(A27)
( )
( )
( ) ( )  
) (
! 2
1 1
1
2
2 2
1 1
2
x O
x
b a
bu u b a au
u
i i i
i
∆ +
∆
+ −
+ + −
=
+ −
,
( ) ( )
( ) ( )
(
¸
(
¸
=
(
¸
(
¸
(
¸
(
¸
−
−
1
0
1 1
1 1
2 2
1 1
b
a
. (A28)
The matrix equations are chosen such that the unwanted derivatives are eliminated. The
end node approximations are calculated using biased formulae. For i=1
( )
( )
( )
( )
( )
3 2
1
2
1
1
1
1 2
! 2
1
! 1
1
x O u
x
u
x
u u ∆ +
∆ +
+
∆ +
+ = (A29)
( )
( )
( )
( )
( )
3 2
1
2
1
1
1
1 3
! 2
2
! 1
2
x O u
x
u
x
u u ∆ +
∆ +
+
∆ +
+ = (A210)
summing
( )
( ) ( )  
( )
( )
( ) ( )  
( )
( )
( )
3 2
1
2
2 2 1
1
1
1 1
3 2 1
! 2
2 1
! 1
2 1 x O u
x
b a u
x
b a
bu au u b a
∆ +
∆
+ + + +
∆
+ + +
= + + + −
(A211)
which can be solved for both the first and second order approximations as
152
( )
( )
( ) ( )  
( )
( )
2
1
1 1
3 2 1
1
1
! 1
2 1
x O
x
b a
bu au u b a
u ∆ +
∆
+ + +
+ + + −
= ,
( ) ( )
( ) ( )
(
¸
(
¸
=
(
¸
(
¸
(
¸
(
¸
0
1
2 1
2 1
2 2
1 1
b
a
(A212)
( )
( )
( ) ( )  
( )
( )
1
2
2 2
3 2 1
2
1
! 2
2 1
x O
x
b a
bu au u b a
u ∆ +
∆
+ + +
+ + + −
= ,
( ) ( )
( ) ( )
(
¸
(
¸
=
(
¸
(
¸
(
¸
(
¸
1
0
2 1
2 1
2 2
1 1
b
a
. (A213)
Similarly, the formulae for i=n are
( )
( )
( ) ( )  
( )
( )
2
1
1 1
2 1
1
! 1
2 1
x O
x
b a
bu au u b a
u
n n n
n
∆ +
∆
− + −
+ + + −
=
− −
,
( ) ( )
( ) ( )
(
¸
(
¸
=
(
¸
(
¸
(
¸
(
¸
− −
− −
0
1
2 1
2 1
2 2
1 1
b
a
(A214)
( )
( )
( ) ( )  
( )
( )
1
2
2 2
2 1
2
! 2
2 1
x O
x
b a
bu au u b a
u
n n n
n
∆ +
∆
− + −
+ + + −
=
− −
,
( ) ( )
( ) ( )
(
¸
(
¸
=
(
¸
(
¸
(
¸
(
¸
− −
− −
1
0
2 1
2 1
2 2
1 1
b
a
. (A215)
For a system with three spatial nodes, the operators are
(
(
(
¸
(
¸
−
−
− −
∆
=
3 4 1
1 0 1
1 4 3
2
1
) 1 (
2
x
D (A216)
and
(
(
(
¸
(
¸
−
−
−
∆
=
1 2 1
1 2 1
1 2 1
1
2
) 2 (
2
x
D . (A217)
These operators are both O(∆x
2
) accuracy. Applying this same technique for a five point
stencil results in operators which are O(∆x
4
) accuracy. These are given as
153
(
(
(
(
(
(
¸
(
¸
− −
− −
− −
− − −
− − −
∆
=
25 48 36 16 3
3 10 18 6 1
1 8 0 8 1
1 6 18 10 3
3 16 36 48 25
12
1
) 1 (
4
x
D (A218)
and
(
(
(
(
(
(
¸
(
¸
− −
− −
− − −
− −
− −
∆
=
35 104 114 56 11
11 20 6 4 1
1 16 30 16 1
1 4 6 20 11
11 56 114 104 35
12
1
2
) 2 (
4
x
D . (A219)
154
APPENDIX 3 – COMPRESSOR ELECTROMECHANICAL
MODELING
The compressor piston position, and the resulting PV power due to this motion,
can be predicted by measuring the voltage and current waveforms delivered to the
compressor. The compressor electrical power input and PV power output are not
equivalent due to losses which occur in the compressor. The largest loss is due to
resistive power dissipation, which is simple to measure. Some smaller losses are due to
eddy currents and friction. These losses require a complete model to allow the prediction
of piston motion.
A3.1 The electromechanical system
The compressor model consists of a mechanical model for the piston moving
mass which is coupled to an electrical model for the voice coil circuit. The coupling
occurs through the electromotive force in the mechanical model and the back emf voltage
in the electrical model. Experimental measurement of the voice coil current and voltage
in addition to the compression space pressure then allows the piston position to be
predicted. Several unknown mechanical and electrical design parameters must be
determined before this is possible. The experimental measurement of these design
parameters, and the basis for this model are developed by Leach who has modeled a
loudspeaker electromechanical system (69).
155
Compression
Space, p
c
To cryocooler
Housing plenum, p
h
Piston moving mass
Voice coil
Magnetic structure
Suspension
spring
stiffness, k
Figure 51 – Voice coil compressor schematic
The compressor schematic illustrated in Figure 51 is representative of the
compressors commonly used to drive Oxford class cryocoolers including Stirling and
Pulse Tube Cryocoolers. This type of compressor utilizes a “voicecoil” to create an
electromotive force to drive the piston motion. The term “voicecoil” comes from audio
load speakers which operate very similarly to reciprocating compressors. The
compressor is essentially a loudspeaker designed to operate at a very high sound pressure
level (SPL). Most compressors use a suspension spring of some sort. Flexures allow the
piston to be centered in the cylinder which reduces frictional drag. Other pistons are
designed to have a sliding contact seal between the piston and cylinder.
A3.2 The Mechanical System
The compressor model can be constructed by considering a freebody diagram of
the compressor piston, similar to Figure 52. In this freebody diagram, the reaction
forces due to pressure, friction, electromotive force, spring stiffness, and inertia are
156
shown. The electromotive force constant, Bl , is the product of the magnetic flux in the
air gap, B , and the effective length of wire that cuts the flux, l . It is assumed that Bl is
a nonlinear function of the piston position and possibly other parameters such as
frequency or current. The pressure in the compression space on the positive x face of the
piston is
c
p and the pressure in the housing plenum space is
h
p . These pressures act on
the piston cross sectional area,
piston
A . The suspension spring stiffness, k , resists the
piston movement. In general, k is a function of the piston position. The moving piston
experiences a frictional force due to sliding contact between the piston and cylinder or
due to viscous shearing of the gas in the pistoncylinder clearance gap in the case of
flexuremounted pistons. In either case, the frictional force can be approximated by the
friction coefficient, b , times the velocity of the piston.
I Bl ⋅
m
piston c
A p
piston h
A p
kx
x
piston
A
x b&
Figure 52 – Compressor piston freebody diagram (FBD)
157
Considering all of the forces acting on the piston, Newton’s Law can be written
for the piston moving mass, m , as
( )
piston c h
A p p I Bl kx x b x m − + ⋅ = + + & & & (A31)
Equation (A41) is a differential equation which defines the piston position. This
equation can be solved if the pressures,
h
p and
c
p , and current, I , are known as
functions of time. Additionally, the coefficients, m , b , k , Bl , and
piston
A must be
known. The compression space pressure,
c
p , can be measured experimentally. The
pressure in the housing plenum,
h
p , can be modeled using a simple adiabatic
compression process for the plenum volume. This results in an expression for the
housing pressure as a function of the piston position,


.

\

− =
g hou
piston
b h
V
A
x p p
sin
1 (A32)
where
b
p is the mean, or baseline, pressure. When this relation for the housing pressure
is substituted into Equation (A41), the mechanical equation becomes
piston c
g hou
piston b
A p I Bl x
V
A p
k x b x m
~
sin
2
− ⋅ =


.

\

+ + + & & & (A33)
where the compression space pressure has been decomposed into a mean pressure and the
fluctuating component,
c b c
p p p
~
+ = . (A34)
The adiabatic assumption for the housing volume pressure reduces the information
needed. More accurate modeling of the housing pressure is possible. Heat transfer
158
between the gas and the housing and piston clearance seal flow will affect the housing
pressure. The adiabatic model should be sufficient for a first estimate. The AC
component of the compression space pressure,
c
p
~
, still remains, and this is measured
experimentally. At this point, the current can be measured experimentally, and the piston
position can be calculated, provided the parameters are known.
A3.3 The Electrical System
x Bl & V
I
coil
R
( ) ω
E
L
( ) ω
E
R
R
I
L
I
Figure 53 – Leach’s lossy inductor voice coil model
The experimentally observed voicecoil current presents large harmonic content
with a pure sinusoidal voltage drive. The first and second harmonics are typically 10% of
the fundamental. This indicates that there are significant nonlinearities in the
electromechanical system. An electrical model can be considered such that the source of
the nonlinearities can be studied. This is important for the current objective of
determining the piston motion. If the nonlinearity is occurring entirely within the
electrical system, then the piston motion is unaffected. However, if the nonlinearity is
related to a nonlinear Bl , then the piston motion is affected. Leach (69) has proposed a
159
possible circuit model for the voicecoil which is illustrated in Figure 53. This circuit is
driven by a sinusoidal voltage, V. Current, I, flows through a constant resistance, R
coil
,
which is the coil DC resistance. The current then flows through a parallel circuit
representing the lossy inductor created by the voice coil. Losses occur due to eddy
currents in the magnetic pole structure. The lossy inductance, L
E
, and the eddy current
resistance, R
E
, are empirically determined and functions of frequency. Leach proposes a
method for measuring these two parameters using a small drive voltage at a frequency
significantly above the resonance frequency (69). This ensures that there is negligible
piston motion. The circuit also contains an additional voltage source, known as the back
emf. This voltage source is created by the voice coil traveling through the magnetic field.
This voltage is proportional to the electromotive constant, Bl , and the voice coil
velocity.
The circuit diagram can be written mathematically as
( ) x Bl I L
dt
d
R I V
L E coil
& + + ⋅ = . (A35)
The node current law gives
R L
I I I + = . (A36)
The loop voltage law around the parallel inductor and resistor can be written as
( )
L E R E
I L
dt
d
I R = . (A37)
Finally, the current flowing through the inductor is defined by the nonlinear differential
equation
160
( ) ( )
E L L E
R I I I L
dt
d
− = . (A38)
The drive current can now be written as
E coil
E L
R R
R I x Bl V
I
+
+ −
=
&
(A39)
and the inductor current equation can be written as
( ) x Bl V
R R
R
dt
dL
R
R R
R
I
dt
dI
L
E coil
E E
E
E coil
E
L
L
E
& −
+
=


.

\

− −
+
−
2
. (A310)
The inductor current must be solved for simultaneously with the piston motion which can
now be written as
piston c
E coil
E L
g hou
piston b
E coil
A p
R R
R I V
Bl x
V
A p
k x
R R
Bl
b x m
~
sin
2
2
−
+
+
⋅ =


.

\

+ +


.

\

+
+ + & & & (A311)
An effective spring stiffness and friction coefficient can be defined as
g hou
piston b
eff
V
A p
k k
sin
2
+ = (A312)
and
E coil
eff
R R
Bl
b b
+
+ =
2
(A313)
respectively. It is clear that the housing gas acts as a spring which increases with the
square of the piston area. The gas spring stiffness also depends on the baseline pressure
and the housing volume as expected. The effective friction coefficient now contains a
term which depends on the electromotive force constant squared. This term represents
the apparent dissipative force due to coil and eddy current losses.
161
The compressor electromechanical model can be used for a variety of purposes.
First, it can be used as a design tool to predict compressor performance based on a choice
of design parameters. Secondly, it can be used to determine unknown parameters based
on experimentally measured voltage, current, pressure, and position. In the case of a few
unknown parameters, this should be tractable. However, if there are many unknown
coefficients, then this presents a difficult problem. For nonlinear parameters, a
correlation function needs to be chosen which increases the number of unknown
parameters. The model can be tested using a compressor with piston position sensors. In
this case, the current, voltage, pressure, and position can be measured. The coil
resistance can easily be measured. Using the methodology described by Leach, the lossy
inductance and resistance can be measured. From the mechanical design, the piston area
and housing volume can be calculated. The suspension stiffness can be relatively easily
measured or calculated based on the design. The moving mass can also be measured
easily. This leaves the electromotive force constant, Bl , as the remaining unknown.
Once the model parameters have been determined, the compressor piston position,
and the resulting PV power due to this motion, can be predicted by measuring the voltage
and current waveforms delivered to the compressor and the compression space pressure.
This is an attractive development for compressors which are not or cannot be equipped
with piston position sensors. The accuracy of the model remains to be determined, but
the ability to monitor the compressor piston position without a dedicated sensor which
can leak provides significant motivation.
162
BIBLIOGRAPHY
1. Walker, G. 1983. Cryocoolers. Plenum Press, New York and London.
2. Gifford, W. and Longsworth, R. 1965. Pulse tube refrigeration progress. Advances
in Cryogenic Engineering, Vol. 10, Plenum Press, New York, 69.
3. Longsworth, R. 1967. An experimental investigation of pulse tube refrigeration heat
pumping rates. Advances in Cryogenic Engineering, Vol. 12, Plenum Press, New
York, 608.
4. Mikulin, E., Tarasov, A., and Shkrebyonock, M. 1984. Lowtemperature expansion
pulse tubes. Advances in Cryogenic Engineering, Vol. 29, 629.
5. Olsen, J. and Swift, G. 1997. Acoustic streaming in pulse tube refrigerators: tapered
pulse tubes. Cryogenics, Vol. 37, No. 12, 769.
6. Radebaugh, R., Lewis, M., Luo, E., Pfotenhauer, M., Nellis, G., and Schunk, L.
2004. Inertance tube optimization for pulse tube refrigerators. Advances in
Cryogenic Engineering, Vol. 49, submitted for publication.
7. Pecharsky, V., Gschneidner, K., McCallum, R., and Dennis, K. 1997. Influence of
alloying on the behavior and properties of Er
3
Ni. Cryocoolers 9, Plenum Press, New
York and London, 663.
8. Gshneidner, K., Pecharsky, V., Osborne, M., Moormen, J., Anderson, I., Pasker, D.,
and Eastwood, M. 1997. Processing and testing of the lowtemperature stage
Er
6
Ni
2
Sn cryogenic regenerator alloy. Cryocoolers 9, Plenum Press, New York and
London, 669.
9. Bradshaw, T., Orlowska, A., Jewell, C., Jones, B., and Scull, S. 1997. Improvements
to the cooling power of a space qualified twostage Stirling cycle cooler.
Cryocoolers 9, Plenum Press, New York and London, 79.
10. Takashi, I., Masashi, N., Kouki, N., and Hideto, Y. 1997. Development of a 2W
class 4K GiffordMcMahon cycle cryocooler. Cryocoolers 9, Plenum Press, New
York and London, 617.
163
11. Chafe, J., Green, G., and Hendrix, J. 1997. A neodymium plate regenerator for low
temperature GiffordMcMahon refrigerators. Cryocoolers 9, Plenum Press, New
York and London, 653.
12. Kuehl, H., Schulz S., Walther, C. 1998. Theoretical models and correlations for the
flow friction and heat transfer characteristics of random wire regenerator materials.
Proceedings of the 33
rd
Intersociety Engineering Conference on Energy Conversion,
American Institute of Chemical Engineers, New York, 207.
13. Chafe, J. and Green, G. 1998. Neodymiumribbonregenerator cooling performance
in a twostage GiffordMcMahon refrigerator. Advances In Cryogenic Engineering,
Vol. 43, Plenum Press, New York, 1589.
14. Shull, C., Ravikumar, K., and Frederking, T. 1994. Hydrodynamic characterization
of perforated plate flow passages. Advances In Cryogenic Engineering, Vol. 39,
Plenum Press, New York, 1615.
15. Kays, W. and London, A. 1964. Compact Heat Exchangers. McGrawHill, Inc.
New York.
16. Organ, A. 1992. Thermodynamics and Gas Dynamics of the Stirling Cycle Machine.
Cambridge University Press, Cambridge 7678.
17. Jiang, Y., Ju, Y., and Zhou, Y. 1998. A study of oscillating flow characteristics of
the regenerators in high frequency pulse tube refrigerators. Advances In Cryogenic
Engineering, Vol. 43, Plenum Press, New York, 1635.
18. Yuan, Z. and Dybbs, A. 1992. Oscillating flow and heat transfer in a Stirling engine
regenerator. Fundamentals of Heat Transfer in Porous Media, Vol. 193, ASME,
New York, 73.
19. Gedeon, D. and Wood, J. 1997. Oscillatoryflow regenerator test rig: hardware and
theory with derived correlations for screens and felts. NASA Contractor Report
198442.
20. Helvensteijn, B., Kashani, A., Spivak, A., Roach, P., Lee, J. and Kittel, P. 1998.
Pressure drop over regenerators in oscillatory flow. Advances In Cryogenic
Engineering, Vol. 43, Plenum Press, New York, 1619.
21. Zhao, T. and Cheng, P. 1996. Oscillatory pressure drops through a wovenscreen
packed column subjected to cyclic flow. Cryogenics, Vol. 36, 333.
22. Organ, A. 1997. The Regenerator and the Stirling Engine. Mechanical Engineering
Publications Limited, London and Bury St Edmunds.
164
23. Roberts, T. and Desai, P. 2003. Periodic porous media flows in regenerators.
Cryocoolers 12, Kluwer Academic, New York, 555561.
24. Watson, E. 1983. Diffusion in oscillatory pipe flow. Journal of Fluid Mechanics,
Vol. 133, 233244.
25. Siegel, R. 1987. Influence of oscillationinduced diffusion on heat transfer in a
uniformly heated channel. Transactions of the ASME, Vol. 109, 244247.
26. Siegel, R. 1987. Effect of flow oscillations on axial energy transport in a porous
media. Transactions of the ASME, Vol. 109, 242244.
27. Kaviany, M. 1990. Performance of a heat exchanger based on enhanced heat
diffusion in fluids by oscillation: analysis. Journal of Heat Transfer, Vol. 112, 49
55.
28. Kaviany, M. and Reckker, M. 1990. Performance of a heat exchanger based on
enhanced heat diffusion in fluids by oscillation: experiment. Journal of Heat
Transfer, Vol. 112, 5663.
29. Wu, P. Zhang, L., Qian, L., and Zhang, L. 1994. Numerical modeling of orifice
pulse tube by using the method of characteristics. Advances In Cryogenic
Engineering, Vol. 39, Plenum Press, New York, 1417.
30. Kornhauser, A. and Smith, J. 1989. Heat transfer with oscillating pressure and
oscillating flow. Proceedings of the 24th Intersociety Energy Conversion
Engineering Conference IECEC89. Vol. 5, 2347.
31. Bauwens, L. 1995. Twodimensional nearly isothermal pulsetube and regenerator
model. Proceedings of the 10
th
Intersociety Cryogenic Symposium. American
Institute of Chemical Engineers, New York, NY, 119.
32. Roach, P., Kashani, A., and Lee, J. 1996. Theoretical analysis of a pulse tube
regenerator. Advances in Cryogenic Engineering, Vol. 41. Plenum Press, New
York, 1357.
33. Kashani, A. and Roach, P. 1998. An optimization program for modeling pulse tube
coolers. Advances in Cryogenic Engineering, Vol. 43. Plenum Press, New York,
1903.
34. Gary, J. and O’Gallagher, A. 2000. REGEN3.2 User Manual, National Institute of
Standards and Technology.
165
35. Lewis, M., Kuriyama, T., Xiao, J., and Radebaugh, R. 1998. Effect of regenerator
geometry on pulse tube refrigerator performance. Advances in Cryogenic
Engineering, Vol. 43. Plenum Press, New York, 1999.
36. Gedeon, D. 1999. Sage Pulse Tube ModelClass Reference Guide, Gedeon
Associates.
37. Harvey, J., Desai, P., and Kirkconnell, C. 2003. A comparative evaluation of
numerical models for cryocooler regenerators. Cryocoolers 12, Kluwer Academic,
New York, 547554.
38. Yuan, Z. and Dybbs, A. 1992. High frequency temperature measurement in porous
metals. Fundamentals of Heat Transfer in Porous Media, Vol. 193, ASME, New
York, 67.
39. Rawlins, W. 1992. The measurement and modeling of regenerator performance in
an orifice pulse tube refrigerator. Ph.D. thesis, University of Colorado, Boulder,
CO.
40. Rawlins, W., Radebaugh, R., Bradley, P., and Timmerhaus, C. 1994. Energy flows
in an orifice pulse tube refrigerator. Advances in Cryogenic Engineering, Vol. 39,
Plenum Press, New York, 1449.
41. Hassanizadeh, S. and Gray, W. 1979. General conservation equations for multi
phase systems: 2. Mass, momenta, energy, and entropy equations. Advances in
Water Resources. Vol. 2, 191.
42. Whitaker, S. 1999. The method of volume averaging. Kluwer Academic Publishers.
Netherlands.
43. Quintard, M. and Whitaker, S. 1996. Transport in chemically and mechanically
heterogeneous porous media. I: Theoretical development of regionaveraged
equations for slightly compressible flow. Advances in Water Resources, Vol. 19,
No. 1, 2947.
44. Gray, W. and O’Neill, K. 1976. On the general equations for flow in porous media
and their reduction to Darcy’s law. Water Resources Research, Vol. 12, No. 2,
American Geophysical Union, 148154.
45. Whitaker, S. 1996. The Forchheimer equation: a theoretical development. Transport
in Porous Media, Vol. 25, Kluwer Academic Publishers, Netherlands, 2761.
166
46. OchoaTapia, J. and Whitaker, S. 1995. Momentum transfer at the boundary
between a porous medium and a homogeneous fluid  I. theoretical development.
International Journal of Heat and Mass Transfer, Vol. 38, No. 14, 26352646.
47. OchoaTapia, J. and Whitaker, S. 1995. Momentum transfer at the boundary
between a porous medium and a homogeneous fluid  II. comparison with
experiment, International Journal of Heat and Mass Transfer, Vol. 38, No. 14,
26472655.
48. OchoaTapia, J. and Whitaker, S. 1997. Heat transfer at the boundary between a
porous medium and a homogeneous fluid. International Journal of Heat and Mass
Transfer, Vol. 40, No. 11, 26912707.
49. Bejan, A. 1997. Advanced Engineering Thermodynamics, Second Edition. John
Wiley & Sons, Inc, Chapter 11.
50. Schiesser, W. 1991. The numerical method of lines : integration of partial
differential equations. Academic Press, San Diego.
51. Hicks, J. and Wei, J. 1967. Numerical solution of parabolic partial differential
equations with twopoint boundary conditions by use of the method of lines. Journal
of the Association for Computing Machinery, Vol. 14, No. 3, 549562.
52. Sincovec, R. and Madsen, N. 1975. Software for nonlinear partial differential
equations. ACM Transactions on Mathematical Software, Vol. 1, No. 3, Association
for Computing Machinery, Inc., 232260.
53. Madsen, N. and Sincovec, R. 1979. ALGORITHM 540 PDECOL, general
collocation software for partial differential equations. ACM Transactions on
Mathematical Software, Vol. 5, No. 3, Association for Computing Machinery, Inc.,
326351.
54. Shampine, L. and Reichelt, M. 1997. The MATLAB ODE Suite. SIAM Journal on
Scientific Computing, Vol. 18, 122.
55. Pareschi, L. 2001. Central differencing based numerical schemes for hyperbolic
conservation laws with relaxation terms. SIAM Journal of Numerical Analysis, Vol.
39, No. 4, 13951417.
56. Naldi, G. and Pareschi, L. 2000. Numerical schemes for hyperbolic systems of
conservation laws with stiff diffusive relaxation. SIAM Journal of Numerical
Analysis, Vol. 37, No. 4, 12461270.
167
57. Morton, K. 2001. Discretization of unsteady hyperbolic conservation laws. SIAM
Journal of Numerical Analysis, Vol. 39, No. 5, 15561597.
58. Roach, R. and Yanping, S. 1998. Unconditional stability, monotonicity and accuracy
of the linear exponential interpolation function for the Burgers Equation. Computers
and Fluids, Vol. 27, No. 8, Elsevier Science, Ltd., 963983.
59. Weideman, J. and Reddy, S. 2000. A MATLAB differentiation suite. ACM
Transactions on Mathematical Software, Vol. 26, No. 4, ACM, 465519.
60. Kirkconnell, C. 1995. Numerical analysis of the mass flow and thermal behavior in
highfrequency pulse tubes. Ph.D. thesis, Georgia Institute of Technology, Atlanta,
GA.
61. Kirkconnell, C., Soloski, S., and Price, K. 1997. Experiments on the effects of pulse
tube geometry on PTR performance. Cryocoolers 9, Plenum Press, New York and
London, 285.
62. Gilmore, D. 1994. Satellite thermal control handbook. The Aerospace Corporation
Press, El Segundo, CA.
63. Harvey, J. 1999. Parametric study of cryocooler regenerator performance.
Master’s Thesis, Georgia Institute of Technology, Atlanta, GA.
64. Harvey, J., Kirkconnell, C., and Desai, P. 2000. Regenerator performance evaluation
in a pulse tube cryocooler. Advances in Cryogenic Engineering, Vol. 45, Kluwer
Academic, New York and London, 373382.
65. Hendrix, J. 1996. A new method for producing perforated plate recuperators.
Advances in Cryogenic Engineering, Vol. 41, Plenum Press, New York, 1329.
66. Knackstedt, M., Sahimi, M., and Chan, D. 1993. Cellularautomata calculation of
frequencydependent permeability of porous media. Physical Review E, Vol. 47, No.
4, The American Physical Society, 2593.
67. Whitaker, S. 1967. Diffusion and dispersion in porous media. AIChE Journal, VOl.
13, No. 3, 420.
68. Whitaker, S. 1973. The transport equations for multiphase systems. Chemical
Engineering Science, Vol. 28, 139.
69. Leach, M. 2002. Loudspeaker voicecoil inductance losses: circuit models,
parameter estimation, and effect on frequency response. Journal of the Audio
Engineering Society, Vol. 50, No. 6, 442.
UMI Number: 3117935
________________________________________________________
UMI Microform 3117935 Copyright 2004 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ____________________________________________________________ ProQuest Information and Learning Company 300 North Zeeb Road PO Box 1346 Ann Arbor, MI 481061346
OSCILLATORY COMPRESSIBLE FLOW AND HEAT TRANSFER IN POROUS MEDIA – APPLICATION TO CRYOCOOLER REGENERATORS
APPROVAL
Prateen V. Desai, Chairman, Mechanical Engineering S. Mostafa Ghiaasiaan, Mechanical Engineering Minami Yoda, Mechanical Engineering Carl S. Kirkconnell, Raytheon Space and Airborne Systems Jeffrey F. Morris, Chemical Engineering Date Submitted: November 25, 2003
to the research grants provided by Hughes Aircraft Company and Raytheon Company to the Georgia Tech Foundation. And it is indeed the group of individuals whom have assisted me that have made this research so enjoyable and productive. Kirkconnell deserves special thanks for being the catalyst for the research at Hughes Aircraft . Mr. Mr.ACKNOWLEDGEMENTS I have been blessed to have been involved in this research with so fine a group. Dr. At Raytheon. Alberto Schroth. Minami Yoda and Jeffrey Morris completed my team of advisors by providing breadth of knowledge in their fields. Mr. Dr. Prateen Desai. Dr. my thesis advisor. This research was performed due. and now Raytheon Space and Airborne Systems. David Mc Gorrin. Beginning with those most directly involved in the research. At Georgia Tech. Hughes Aircraft Company. Kenneth Price. at the time. Carl Kirkconnell was searching out an undergraduate student to perform this research for. William Croft. I have iii . and I as a friend and advisor. fortunately thought of me when Dr. special thanks to Mr. and Mr. The Woodruff School and the Georgia Tech community have been so fortunate to have Dr. I hope that I can provide the same enthusiasm and energy that you have provided me. in no small part. Desai as a dedicated professor. Drs. Mostafa Ghiaasiaan quickly took an interest in the research after conversing with me and he now will lead the cryocooler research at Georgia Tech. Of these individuals. I would like to thank my committee. Additional thanks must go to the Hughes and Raytheon Doctoral Fellowship Program and the Georgia Institute of Technology Presidential Scholarship for supporting me directly throughout this program. Thomas Pollack.
She has given me so much hope and now our new life begins. and Mr. the next torch bearer. iv . who I am so thankful is close. Mr. I thank my whole family for their support and love. Robert Hon. Matthieu Coutaudier. but so near my heart. especially my parents who are so far away. Mr. my sister Janelle. Jeesung “Jeff” Cha. and to my wife’s family who I wish were close.been so fortunate to have several undergraduate and graduate students work with me in my lab. My wife Chrystine has been my rock during this undertaking.
.........................................................TABLE OF CONTENTS Approval ..................................................2......................1..........................................................1 Governing Equations and Constitutive Relations................................................1.............2.............................................................1 Introduction ........................2 Balance of Momentum Equations ............1.........................................21 2.......4 βPhase Conservation of Energy Equation ..............................................................................................20 2......................................................................................................................2................................................................................................................ Introduction and Background ...........2...........................7 1....................15 1.....................................................1 Conservation of Mass Equation ..3 αPhase Conservation of Energy Equation .....................................................................................................................xv Summary ............22 v ...1 1....................................x List of Tables .............................................................. Theoretical Development............................ xiv List of Tables .....................1 1.....................4 1............... iii List of Figures ...1.19 2............... ii Acknowledgements.4 Overview ...........................................................................................................................................................20 2....................................................................................................................................................19 2....................1 History of Pulse Tube Cryocoolers .................................................................................................................xx 1..................................................16 2.................................................................3 Development of the Volume Averaging Technique ............................ xiv Nomenclature................................................................2 Background...2 Survey of Efforts to Improve the Regenerator..........4 1..............
..3...25 2..............................1...............5 Negligible Brinkman Effect .................................................26 2.............................................28 2....................23 2..................4 VolumeAveraged βPhase Conservation of Energy Equation.............2.......35 2.......4.........................1 Simplified Conservation of Mass Equation ......4 Entropy Generation Due to Thermal Dispersion .3 VolumeAveraged αPhase Conservation of Energy Equation ................1......................................2 Simplified Balance of Momentum Equation.............................2 Negligible Mechanical Dispersion .....4.......3.......32 2.........3.............2................4 Simplified Equations ...................3 Simplified αPhase Conservation of Energy Equation .......................6 VolumeAveraged βPhase Entropy Generation Equation ...............................................29 2....................................3 Thermal Dispersion.......25 2.......3............................34 2........................................6 One Dimensional Model on the Macroscopic Length Scale.....................1................2 VolumeAveraged Balance of Momentum Equation.............................................33 2........6 αPhase Entropy Generation Equation .................................8 Summary of Equations ........29 2........23 2......................5 VolumeAveraged αPhase Entropy Generation Equation.......................2 VolumeAveraged Equations .....................................24 2......5 Equations of State ................................2.....26 2.................28 2....27 2...............................4.......1 VolumeAveraged Conservation of Mass Equation .34 2................................................3.................................1..................................................................................2..2..22 2.............................7 βPhase Entropy Generation Equation...............................1 Density Spatial Deviation ............................36 2..............3 Simplifying Assumptions .31 2..........................2................................................................37 vi ......35 2......2......................................3......................................
..................................60 3.........................52 2...4..............................4.................................5 The Constant Temperature Model (CTM)..............................................................................................................................39 2............................5 Scaled Equation Summary .2 Balance of Momentum Equation Scale Analysis......................................................1 Conservation of Mass Equation Scale Analysis..................................................................54 2.................60 3...................6 The Local Thermal Equilibrium Model (LTEM) ................................................................................4 Simplified βPhase Conservation of Energy Equation ..........5 Simplified αPhase Entropy Generation Equation..........6..........................74 vii .............................................66 3...........2 Spatial Discretization ................................ Computational Models......7 Dual Energy Equation Model (DEEM) ....................4.................6..2............6 Scale Analysis ....................................................................6 Limiting Cases .............................................1 Problem Definition ..............................................................................46 2.................37 2................................................6........................................................................................6.....................................7 Summary of Equations ..................................................63 3.....50 2.............................................................70 3........6 Simplified βPhase Entropy Generation Equation ....62 3..2 Numerical Method.........................................1 Time Integration....4...3 αPhase Conservation of Energy Equation Scale Analysis ...............61 3.2.......62 3...2.....................................3 QuasiSteady Convergence via Cyclic Time Relaxation ..........................................55 2............5 Exact Solutions .....................59 3..........6....6...72 3.....................................44 2..............................49 2..................................................4 βPhase Conservation of Energy Equation Scale Analysis.....................39 2.................4 Boundary and Initial Conditions .......50 2.....
.................1 Definitions..........2.....................2.........................................84 4..138 viii ................110 5...2 Transport Theorem ......3...............................................1 Overview and Experimental Apparatus.............................................................................................................The Perfect Regenerator .......................................................106 5............................................75 4................................................................84 4.........................5 Second Law Results ...................2 Model Comparison .........................................................................................................................94 5...................118 5........................2 Experimental Results..............................................120 5.......................136 A1.............................................................................................132 Appendix 1 – Derivation of the VolumeAveraged Governing Equations.........................4 First Law Results......................3 Baseline Regenerator Solutions ................................1 Sage System Level Modeling ..........................2.. Conclusions and Recommendations ..................6 Sage and REGEN Comparisons....................................... Experimental apparatus and measurements ................7 Permeability and Forchheimer Coefficient Predictions .................................1 Conclusions .....2 The Mean Temperature Profile .....................................................113 5..... Results and Discussion ..................................................2..............131 6........2 Contributions ...110 5..................................................8 Model Verification ................................1 Net Enthalpy Flowrate ..114 5.........2............2..............................................................................3 Future Work.......................136 A1...............122 5..............................105 5....................................................................................................................................................................................................1.125 6...................1........................129 6....................................1 Volume Averaging Theory ...........................129 6.....................2...............................................................136 A1........
...........................................154 A3..2...................3 Spatial Averaging Theorem ..................162 ix .................1..............................................147 A1.......3 VolumeAveraged αPhase Conservation of Energy Equation ..........................................5 VolumeAveraged αPhase Entropy Generation Equation..............A1...............158 Bibliography ......2 The Mechanical System......138 A1.............139 A1...............2..142 A1........................................................2 Application to the Governing Equations ..................................................148 Appendix 2 – Derivation of Differentiation Operators Using MatLab...........................2 VolumeAveraged Balance of Momentum Equation.........3 The Electrical System .......2.........146 A1......................6 VolumeAveraged βPhase Entropy Generation Equation ............................154 A3.............................................2....................2.................................................................155 A3.4 VolumeAveraged βPhase Conservation of Energy Equation.............................................1 VolumeAveraged Conservation of Mass Equation..........1 The electromechanical system...............140 A1...........................................145 A1........................149 Appendix 3 – Compressor ElectroMechanical Modeling ...................................140 A1......1........4 Modified Averaging Theorem.....................2.............................
.....................................2 mW deviation)............................6 Figure 4 ...........80 Figure 10 – Maximum mass flux error 1x105 kg/m2s.......... and the volumeaveraged velocity......Experimental Apparatus.................................... ..19 Figure 5 – Conceptual diagram illustrating the local velocity............60 Figure 7 – Test case solution results...2 Pa .........................................................................................................................................1 Figure 2 – A) Basic Pulse Tube Cryocooler and B) Orificed Pulse Tube Cryocooler .....81 Figure 12 – Maximum matrix temperature error 6x106 K ...........................................................5 Figure 3 ........................................82 Figure 14 – Maximum pressure error 1......................................................................................... The αphase is a Newtonian fluid and the βphase is a solid...................................................78 Figure 8 – The net enthalpy flux satisfies the test case to within visual accuracy (0.........................................................................................................................................82 Figure 15 ....................Comparison Diagram of the PTC and Stirling Cryocooler...Bypass Pulse Tube Cryocooler ...... velocity deviation................................................................................................................85 x ......................................................80 Figure 11 – Maximum energy error 2 kJ/m3 .....................81 Figure 13 – Maximum gas temperature error 8x106 K .LIST OF FIGURES Figure 1 ........................................................................79 Figure 9 – Maximum density error 8x106 kg/m3 ...................Thermohydraulic system of a typical porous media............ ..30 Figure 6 – Regenerator computational domain diagram.................
...........107 Figure 34 ............93 Figure 27 ...........92 Figure 25 ...........91 Figure 24 ........................................................................................... Main panel view................................90 Figure 22 .....................................................................................108 xi ....90 Figure 23 – Data acquisition system...................... 76 K...Steady Flow Pressure Drop.............................................. ...............................85 Figure 17 – Experimental apparatus with complete instrumentation....System level energy flow diagram...................Data acquisition and control program.........................Oscillatory Pressure Drop .....................................86 Figure 18 – Closeup view of the compressor..................................................................................................................................................95 Figure 28 ................90 Figure 21 ........................................92 Figure 26 ..................... ........97 Figure 29 ................................Perforated Disk........Load Curves ...... ...................Data acquisition and control program.......Pressure Wave Phase Angle............. 400 mesh.......... Compressor panel view............................................99 Figure 30 .......... .......................................................87 Figure 20 .................................... ................................101 Figure 32 – Sintered glass regenerator.....................................................Pressure Ratio Attenuation....................... 0 W.............Figure 16 – Regenerator/Pulse Tube expander module....86 Figure 19 – Turbo vacuum pump station......................... ................... Regenerator is the larger diameter section......Foam Metal ............................................................Wire Mesh............................................ Pressure panel view.............................104 Figure 33 – Sage diagram of the laboratory pulse tube apparatus.................100 Figure 31 ................................................Data acquisition and control program.....
........................................................................ plotted versus cycle time.....................................................117 Figure 40 – Temperature difference versus position (multiple lines) and time....................... and entropy generation due to conduction and dispersion in the gas (bottom)...................................... Regenerator total energy flow is constant along regenerator...... .............. Pressure and mass flow phase shifts are apparent.................................................... The middle plots are max and min of density and velocity.....116 Figure 39 – Cycleaveraged gas temperature (red) versus a linear profile (blue)..........................................114 Figure 37 – Solutions plotted versus time and position.. Results for 400 mesh baseline case................................................118 Figure 42 – Cycleaveraged mass flow rate...................................115 Figure 38 – Boundary solutions for the gas and matrix temperatures and enthalpy flow rates.......... ...................... .....119 Figure 43 – Cycleaveraged energy flows.......................120 xii ....................................................119 Figure 44 – Entropy generation due to viscous and inertial losses (top)..................) .........................112 Figure 36 – Comparison of mean temperature profiles calculated with the different models based on identical operating conditions..............................Figure 35 – Comparison of regenerator loss calculated with limiting case models based on identical operating conditions (ideal gas assumes constant properties.....................................118 Figure 41 – Mean Reynolds Number....................... Multiple curves represent different locations in the regenerator......................... .......................................
................................. Visualization of surface pressure.......................................................121 Figure 46 – Cycleaveraged volumetric entropy generation rates plotted versus position in the regenerator................Demonstration calculation for flow through wire mesh screens.......................................155 Figure 52 – Compressor piston freebody diagram (FBD).....133 Figure 51 – Voice coil compressor schematic ..........123 Figure 48 – Friction factor comparison for 400 mesh screens...Figure 45 – Entropy generation due to matrix conduction (top)....... Results for 400 mesh baseline case.........................128 Figure 49 ...............133 Figure 50 – Demonstration calculation for flow through wire mesh screens......... .............................. and entropy generation due to interfacial convection (bottom).....................................................121 Figure 47 – Comparison of Sagepredicted regenerator loss versus GT model...... Visualization of pathlines...................................................156 Figure 53 – Leach’s lossy inductor voice coil model ........................158 xiii ............................................................................. Results for 400 mesh baseline case... ........ ......
..............................126 xiv ..........Descriptive Summary of the Regenerators under Study ................................126 Table 7 – Permeability and Forchheimer Coefficient (incompressible model)..............124 Table 6 – Permeability and Forchheimer Coefficient (compressible model) ....................126 Table 8 – Permeability and Forchheimer Coefficient (neglecting advective acceleration)...............................LIST OF TABLES Table 1Summary of scale analysis ........................................................................................................................................................................96 Table 4 – 400 mesh baseline operating conditions ............................................................................................109 Table 5 – Sage Comparison Summary........88 Table 3 – Porous Media Parameter Summary (64)....58 Table 2 ......................................
cvs Specific heat scales [J/kgK] xv .NOMENCLATURE Notation: α β aα Gas Phase Solid Phase Volume average of quantity a in the α phase α aα ˆ a r v Intrinsic volume average of quantity a in the α phase Spatial deviation of quantity a Vector of v Variables: Aαβ Af Area of the α − β interface (within the averagingvolume. V) [m2] Flow area [m2] Total cross sectional area.C2 cp cv c ps . gas and solid [m2] Surface area per unit volume [1/m] Thermal dispersion tensor closure variable Integration constants Gas constant pressure specific heat [J/kgK] Gas constant volume specific heat [J/kgK] A av ~ bα C1 .
F3 .cf ~ Dα Forchheimer inertia coefficient Thermal dispersion tensor Mechanical dispersion tensor. rank 4 Convective heat transfer coefficient [W/m2K] Enthalpy [J/kg] Enthalpy spatial scale [J/kg] Identity tensor Permeability tensor [m2] Permeability. e ~ F F F1 . rank 3 Specific internal energy [J/kg] Forchheimer tensor DarcyForchheimer surface integral Source terms for exact solution Friction factor. F2 . F4 f ~ G4 H h Mechanical dispersion tensor. rank 2 Regenerator diameter [m] Differential surface area of Aαβ [m2] Gas total energy per unit volume [J/m3] Gas total energy temporal scale [J/m3] ~ D2 Dr dS E ~ E ~ E3 Mechanical dispersion tensor. in axial direction [m2] xvi h ~ I ~ K K . frequency [Hz] eα .
mc & & mh .kα . k β Lr Conductivity [W/mK] Regenerator length [m] Macroscopic length scale [m] Mass flux [kg/m2s] Mass flux spatial and temporal scales [kg/m2s] Hot and cold boundary mass flux amplitudes [kg/m2s] Hot and cold boundary mass flow rates [kg/s] Axial conductivity enhancement Nusselt number Normal vector Reference Peclet number Gas pressure [Pa] Pressure spatial scale [Pa] System baseline pressure [Pa] Gas and matrix heat flux [W/m2] Gas constant. β gen gen sα .α . m ~ m. p p pb q g . s′′′ . nβ Pe pα . or radius of averaging volume [Pam3/kgK] Reynolds number based on permeability length scale Reference Stanton number Volumetric entropy generation [W/m3K] Entropy [J/kgK] xvii Ls mα . qm R Re K St s′′′ . m mh . s β . mc Nk Nu nα .
Tc To t ts r uα uα . µ .Tα .V & Wlost x Greek: ρα . u Vα .Vβ . ρ Density [kg/m3] Density spatial and temporal scales [kg/m3] Solid heat capacity [J/K] Hot and cold boundary mass flux phase angles [rad] Viscous dissipation function Gas viscosity [ms] xviii (ρc )β φmh . ρ ~ ρ. Tβ T Temperature [K] Temperature spatial scale [K] Temperature difference scale [K] Hot and cold boundary mean temperatures [K] Reference temperature. φmc φα µα . for lost power [K] Time [s] Time scale [s] Velocity vector [m/s] Axial velocity [m/s] Volume (of the averagingvolume) [m3] Lost power [W] Axial location [m] ˆ T Th .
εα . ε β . ε ω τ Porosity Angular frequency [rad/s] Time period [s] Solid tortuosity Leading order dimensionless parameters Leading order dimensionless parameters Specific heat ratio τβ Γ1 . Γ2 ε1 . ε 2 γ xix .
Convection between the two phases is the dominant mode of heat transfer within the porous media. Unique to this analysis is the fact that the model is valid for oscillatory. Conduction within the solid and fluid phases is made evident by volume averaging. cryogenic flows such as that occurring in a regenerative cryogenic refrigerator such as a Pulse Tube Cryocooler (PTC. The coefficients that characterize the Forchheimer momentum equation are determined experimentally.) In a PTC the forced flow driven oscillations in the regenerator create Reynolds numbers high enough such that microscale inertial effects dominate the momentum equation. but the determination of the conductivity coefficients requires numerical experiments and is unique to the geometry and conductivities of the two phases. Heat transfer within a porous medium occurs due to temperature gradients in the gas and solid phases. xx . known as the Forchheimer Effect.SUMMARY In this study the phenomena of compressible flow and heat transfer in porous media are modeled based on fundamental principles. Determination of the convective heat transfer coefficient for a porous media requires physical experiments. The conservation equations for the two phases are transformed by the method of volume averaging which is an analytical method used to unite the microscale and macroscale effects characteristic to porous media flows. can be predicted and modeled based solely on fundamental principles and the method of volume averaging. This phenomenon.
Heat transfer due to temperature gradients and flow friction in the regenerator are always competing effects leading to a model which requires coupling of the momentum and energy equations. These competing effects are united with the concept of entropy generation which relies on the second law of thermodynamics. The uniqueness of this model is the completeness of the theoretical development and the flexibility of use for a variety of applications. Numerical solutions are compared with experimental data for an operating cryocooler. All real processes generate entropy. and the most efficient processes which balance flow friction and heat transfer generate minimum entropy. These numerical solutions are compared with similar solutions existing in the literature. The theoretical model is presented with a numerical solution technique. xxi .
CHAPTER 1 1.1 Introduction Cryocoolers (refrigerators capable of cooling to temperatures below 120 K) have long been classified into two categories based on the type of heat exchange process: recuperative and regenerative (1). INTRODUCTION AND BACKGROUND 1. The Pulse Tube Cryocooler (PTC) and the Stirling Cryocooler are two examples of regenerative cryocoolers (Figure 1).Comparison Diagram of the PTC and Stirling Cryocooler 1 . both operating on a Pulse Tube Cryocooler Qaftercooler Wcomp Aftercooler Compressor Wcomp Compressor Hregen Regenerator Hregen Regenerator Cold Heat Exchanger Qnet Hpt Orifice Pulse Tube Reject Heat Exchanger Surge Volume Qreject Pulse Tube Expander Stirling Cryocooler Qcompressor Qnet Wexpander Stirling Expander Figure 1 .
which cannot tolerate vibration. In the Stirling cooler. In the case of the PTC. If the motion of the gas piston has the proper phase.modified Stirling thermodynamic cycle using helium as the working substance. orifice. Instead. The PTC is a unique type of regenerative cryocooler in that it does not have any moving parts in the cold region. This is distinctly different from the Stirling cryocooler that operates by an oscillating displacer directly in the cold region. Unlike the Stirling thermodynamic cycle. thus producing refrigeration. the PTC and Stirling cryocooler cycles are not steady flow processes. it accepts work from the cold heat exchanger and delivers this work to the reject heat exchanger where it is converted into heat. If this work is sufficiently large. The pulse tube gas motion can be controlled such that there is a gas “piston” which acts like the Stirling piston. This heat is rejected from the system in the reject heat exchanger. Due to the absence of an expander piston in the cold region and the additional volume 2 . the gas flow oscillates in a quasisteady fashion due to the motion of a compressor piston (and an additional expander piston in the Stirling cryocooler). Vibration in the Stirling cooler caused by the displacer also presents problems for applications like sensitive infrared sensors. and surge volume. resulting in the potential for mechanical wear that increases mechanical complexity and can limit the life of the cooler. refrigeration is produced by driving the expander piston such that it receives mechanical work from the working fluid. the heat exchanger can absorb heat from the surrounding. the active expander piston is replaced with a passive pulse tube.
It is not unrealistic to have an enthalpy loss on the order of one Watt with a peak enthalpy flow rate of 1000 Watts. The types of regenerators being studied in this work are typically found in Stirling and pulse tube cryocoolers. Applications which require this type of refrigeration are superconducting electronics. high heat transfer coefficient. Minimizing parasitic heat loads is critical to achieving the design goals. and infrared focal plane 3 . This porous material is selected such that it has sufficient thermal heat capacity. Regenerative cryocoolers are used in a variety of applications.after the regenerator. or the cycleaveraged enthalpy flow at the cold end plus conduction losses in the gas and matrix. These losses can be considerable. and low flow friction. The enthalpy loss is the most difficult quantity to estimate. and other types of regenerative cryocooler applications. the net refrigeration is the heat transfer rate from the cryogenic device being cooled. which is equal to the total gross refrigeration produced less internal parasitic losses in the cooler. and may be quantified only via an accurate mathematical model. These devices are typically used in applications which demand small net refrigeration (on the order of a few watts) at temperatures below 100 K.1%. The designer of a cryocooler is chiefly concerned with achieving a specified net refrigeration at a given temperature with a minimum input power. requiring an accurate prediction of the mass flow and temperature waveforms at the cold end. One main system loss is the regenerator loss. magnetic resonance imaging. This results in a larger regenerator pressure drop. The regenerator is a duct packed with some porous material. In the design of a pulse tube cooler. the pulse tube experiences much higher mass flow rates through the regenerator. or 0.
and Shkrebyonock (4). Several other applications include gas liquefaction of nitrogen.1 History of Pulse Tube Cryocoolers In 1966. minesweeping magnets. 1.2 Background The experimental apparatus used in this research is a Pulse Tube Cryocooler. This refrigerator is commonly known as the Basic Pulse Tube Cryocooler (BPTC) (Figure 2. causing the open end to cool. Gifford and Longsworth first detailed the construction of a pulse tube refrigerator (2.2. it is necessary to provide a brief history of the PTC including efforts to improve the performance of the PTC. 1. The following discussion is a review of these efforts. investigators have used experimental measurements and numerical modeling. nondestructive evaluation using SQUIDs. The design of their pulse tube was essentially a tube with one end closed and the other end open (in addition to a compressor and heat exchangers. 3).A). 4 .arrays. For this reason. Tarasov. A review of the efforts to improve regenerator performance includes studies of materials and geometry. Not until 1984 was the modern pulse tube cryocooler developed by Mikulin. and military weapon systems.) Both ends had heat exchangers. outer space experiments and instruments. To understand the regenerator. magnetocardiography using Superconducting Quantum Interference Devices (SQUIDs). The component being studied in the PTC is the regenerator. and the open end was subjected to an oscillating pressure through a regenerator.
The effect of this enhancement was to create an advantageous phase difference between the oscillating pressure and velocity. A very small flow circulation can lead to a huge reduction in performance. This type of cryocooler is called a Bypass Pulse Tube Cryocooler (Figure 3). This new design is commonly called the Orifice Pulse Tube Cryocooler (OPTC) (Figure 2.This cryocooler was equipped with an orifice and a surge volume on the warm end of the pulse tube. The drawback of this design is that the flow bypass can create a DC flow circulation if the system is not designed properly.B). Wcomp Wcomp Compressor Qcomp Surge Volume Compressor Qcomp Orifice Qreject Pulse Tube Regenerator Qreject Pulse Tube Regenerator Qrefrig (A) Qrefrig (B) Figure 2 – A) Basic Pulse Tube Cryocooler and B) Orificed Pulse Tube Cryocooler 5 . Other enhancements have been made such as porting the highpressure gas from the outlet of the compressor to the warm end of the pulse tube which led to improved performance due to improved phase control.
Currently. orifice. One recent improvement is the 6 .Bypass Pulse Tube Cryocooler Developments in pulse tube technology have been occurring continuously up to the present. The flow oscillations result in mass streaming and radial heat transfer. Reducing pulse tube flow losses begins by being able to completely model the oscillating flow in the pulse tube. the basic. both of which reduce the refrigeration possible with a pulse tube. several investigators discovered that a proper tapering of the pulse tube leads to a complete elimination of mass streaming (5).Wcomp Surge Volume Orifice Compressor Qcomp Qreject Pulse Tube Regenerator Qrefrig Figure 3 . As already mentioned in the preceding. Recently. and improving regenerator effectiveness. phase control. three main areas of improvement are minimization of pulse tube flow losses. and bypass pulse tube all seek to manipulate the phase angle between the pressure and mass flow to an optimum angle at which the maximum refrigeration is produced.
The inertance tube improves the orifice design by adding an inertance to the fluidic system. This has led to a development of exotic materials 7 Any change in these . The heat transfer effectiveness is a function of fluid properties. and the flow geometry. the solid properties appear to be a major contributor to the overall regenerator performance. Design of an optimum phase control device and reducing pulse tube losses will eventually make the pulse tube cryocooler more competitive than the Stirling cryocooler for most applications due to significantly lower cost and comparable efficiency. The basic problem focuses on increasing the heat transfer effectiveness between the gas and solid in the regenerator. The problems with designing the optimum regenerator have interesting complications as the temperature of the cold end decreases.2. Orders of magnitude decrease in thermal capacity of the material from 300 K to the cold temperature make different materials attractive in different areas of the regenerator. solid properties. At temperatures below 20 K. These efforts have been categorized into areas of materials and geometry. parameters will affect the regenerator performance. but also eliminates the possibility of flow circulations which can occur with the bypass design.2 Survey of Efforts to Improve the Regenerator Extensive efforts have been focused on improvement in regenerator technology since the development of regenerative cryocoolers. 1.inertance tube (6). This improvement not only increases the phase angle. and measurement. modeling. These complications arise in factors such as variation of thermal properties over the huge temperature range of the regenerator.
Er3Ni (7). Bradshaw.9Co0. Several erbium alloys have been tested. The materials used were gold wire. et al. have performed experiments using a variety of different materials in several combinations to study their effects in a two stage Stirling cooler operating in the 15 K range (9). This alloy has been investigated by many as a regenerator material for GiffordMcMahon (GM) and Stirling coolers operating below 10 K.1 and Er3Ni increased the cooling capacity by a factor of 1.and processes so that the regenerator has sufficient thermal capacity in the cold region. at Ames Laboratory have investigated titanium alloying of a popular erbium alloy. they found that the lead and Er3Ni performed the best. The ErNi0. At the lowest operating temperatures. Their main effort was to develop a process to generate this alloy. lead wire. also at Ames Laboratory have suggested Er6Ni2Sn alloy for the first stage of a GM cooler (8). Takashi. Gshneidner.9Co0. et al. Their conclusion was that the combination of ErNi0. Again.4 J/Kcm3 for Er3Ni. et al. the brittle properties of this alloy have limited its use.2 K GM cooler with 12 kW compressor input (10). The results showed that practical issues such as particle escape from the regenerator and settling will prevent this alloy from being used in the near future. et al. 4. have replaced 8 . have used a slightly different erbium alloy (ErNi0. Er3Ni and stainless steel mesh.1 has a dramatic increase in specific heat around 6 K of approximately 1 J/Kcm3 in contrast to 0. The significant result was that this particular alloying increased the ductility without significantly decreasing the specific heat of the alloy. and these materials were also found to have the highest specific heat at that temperature.2 from a regenerator with just Er3Ni. Chafe. et al.1) along with Er3Ni in a large 2.9Co0. Pecharsky. but these magnetic alloys are brittle at cryogenic temperatures.2 W cooling capacity.
The technique is part of a process which has become known as 'validation'. 13. The geometry of the pores in the regenerator matrix determine the pressure drop within the regenerator. Measurements of steady pressure drop and correlating these However.. they were able to progress from 8 K down to 4 K under similar operating conditions. twodimensional (or axisymmetric). Efforts to minimize the pressure drop while still having adequate heat transfer have been reported in the literature. Cf. They have also taken advantage of a perforated plate geometry that reduces pressure drop in comparison to spheres.lead balls with Neodymium plates and balls in the second stage of a 10 K GM cooler application to utilize the higher heat capacity at the low temperatures (11). in terms of a friction factor. for example. many measurements to friction factors have been investigated by many. When analysis and computer simulation based on such correlations yield pressure distributions which do not tally with measurement from running machines. Organ in Thermodynamics and Gas Dynamics of the Stirling Cycle Machine states this argument repeatedly (16). Since pressure drop and heat transfer are coupled. do not correlate well for oscillatory flow pressure drop and heat transfer due to additional effects such as enhanced dispersion due to the oscillations (15). The correlations given by Kays and London. it is common practice to 'improve' matters by arbitrarily adjusting the correlations. investigators put too much confidence in the importance of steady flow friction factors (12.e. 14). Exercises in validation have been reported13 which called for Cf at given Nre to be multiplied by factors between 4 and 7. correlated with geometry and Reynolds number. i. 9 . a certain amount of pressure drop is required to achieve effective regeneration. Nre. There is something less than satisfactory about the way in which Stirling machine analysis handles flow within the regenerator: (1) The flow case is treated by the method traditional for steady. As a result. incompressible viscous flow in pipes.
17. The research being conducted by Roberts and Desai (23) are motivated by the concerns voiced by Organ. This type of argument shows that there are effects other than the inertial. effect which is the conventional basis for correlating friction factor with Reynolds number. 19. 20. Organ claims that the pressure drop is a result of independent viscous. usage and interpretation of the steadyflow correlations has been parochial. Organ makes a clear argument that the 'steady' flow through the porous media is not a true description due to inherent oscillatory fluctuations at the natural frequency of the pore structure. In a separate reference. This requires that experimental data be used as 10 . which is based on a linear wave model (22). The compressible nature of the gas flowing through the regenerator matrix leads to a nonconstant pressure gradient.4 Some investigators have made the realization that steady flow friction factor correlations are not adequate.(2) The discrepancy between experimental measurement and theoretical prediction has come to be attributed to the fact that steadyflow correlations do not take into account the unsteady effects which arise from the cyclic nature of the flow processes in the Stirling machine. 'steady' flow. This concept provides a method to better explain the physical phenomena in the regenerator. ½ρu2. having in most instances looked no further than the incompressibleflow cases documented by Kays and London. An enquiry into the role of unsteadiness is certainly called for. 18. It will be seen in later Chapters that steady flow measurements can be used effectively to predict friction factors which are accurate for steady flow and oscillating flow. and compressibility effects. but there is still a gap between their measurements and the physical phenomena occurring within the regenerator (16. inertial. This indicates that there are local accelerations within the fluid even for unidirectional. 21). At the same time. Organ develops the concept of the regenerator flow impedance.
Two authors have made direct extensions of Watson’s initial investigation. 11 . Zhang. and can be found under a variety of headings. Theoretical modeling of the regenerator and general porous media flows have been conducted by many. Kaviany has investigated the effect of oscillatory flows on heat exchangers composed of tube bundles (27. 28). the flow oscillations produce a positive net energy transport from the cold end to the warm end of a channel. Watson used an exact solution for incompressible flow in a tube to show how diffusion can be enhanced (24). 26). For zero net flow. While Watson was not interested in the regenerator problem. His analysis is an important extension of Watson’s work since he considers the coupled problem of the gas and the wall energy. Although the flow in the regenerator is not incompressible. et al. Preliminary results indicate that their method overpredicts the performance based on experimental results. the results are interesting since he has focused explicitly on porous media flows. Kaviany considers the case of enhanced heat diffusion between two reservoirs due to the oscillations.boundary conditions for a differential model rather than a lumped model which is typically used. Siegel’s analysis applied this directly to the problem of the regenerator (25. have applied the method of characteristics to solve the 1D pulse tube governing equations (29). In this analysis. The solutions obtained are exact solutions for the velocity and temperatures. his research illustrates the basic phenomenon which occurs in the regenerator and the pulse tube. Siegel attributes the axial transport to the transverse conduction which occurs between adjacent fluid layers. Kaviany solves for the wall temperature directly. While Watson assumes zero wall heat flux.
The fully nonlinear model ultimately requires a numerical approximation rather than the analytical solution which he obtains. it is expected that the imaginary component of the regenerator Nusselt number would be quite small for typical frequencies used in Stirling and PTC cooler. which is used to define an oscillating Peclet number. the model could yields invalid results for a pulse tube cooler due to relatively large pressure oscillations. The hydraulic diameter of the regenerator can be 5 to 6 orders of magnitude below that of the compression space used. The critical oscillation frequency increases proportional to the inverse of the square of the hydraulic diameter. the instantaneous heat flux in a compression cylinder is observed to be out of phase with the bulk temperature difference. the linearity assumption is no longer valid. In this study. They find that a complex formulation of Newton’s law of cooling is applicable. These studies were conducted in a compression cylinder. requiring a fully nonlinear model. While the approximate solution that he obtains is interesting. Bauwens has proposed a 2D model for the pulse tube and regenerator that can be approximated only for small fluctuations using perturbation methods (31). This requires measurement of the real and imaginary component of the Nusselt number. At large pressure ratios. Thus.The concept of a complex Nusselt number is developed by Kornhauser and Smith (30). This indicates that the traditional Newton’s law of cooling does not hold for oscillating flow above a critical oscillation frequency. but the authors allude to the fact that this can be extended to the compression and flow process occurring in the regenerator. 12 .
et al.Roach. A finite difference program called REGEN3. The pressure phase shift across an optimally designed Stirling regenerator is typically small compared to a pulse tube regenerator. 13 . The effect of the pressure gradient is included as a correction. The velocity field is found using an explicit equation derived from the continuity equation assuming zero pressure gradient. The method has been incorporated into a program called ARCOPTR. velocity. the thermal interaction does depend on the velocity and pressure fluctuation. and matrix temperature. the higher flow rates found in pulse tube regenerators lead to inaccuracies in the REGEN model. Conversely. also propose a perturbation solution to the regenerator problem (32). and has been used to study the effects of regenerator geometry by Kuriyama.2 (34) has been developed by the National Institute of Standards and Technology. This allows solving for the velocity and pressure and then using those solutions to solve for the temperature. This leads to a smaller pressure gradient in the Stirling regenerator. This model has been described in a separate publication by Kashani and Roach (33). While the negligible pressure gradient assumption may be accurate for some cases such as Stirling regenerators. It does not appear that they have provided any experimental comparison. Their results indicate that the velocity and pressure fluctuations do not depend on the thermal interaction between the gas and solid. (35). This model assumes that the pressure in the regenerator is uniform and oscillates in time. gas temperature. et al. This leads to a nonconservative model. the velocity and density are essentially determined from a single equation. As a result. The mathematical model proposed by REGEN involves solving a system of equations for the density.
Several individuals have made some interesting and important measurements on the regenerator and pulse tube. High frequency response sensors must be used to accurately resolve the oscillating temperatures and velocities. Apparently. et al. Further inaccuracies can be attributed to a non conservative spatial discretization method based on first order control volumes.Currently. Rawlins. were concerned with the accuracy of assumptions and numerical methods in Sage (37). This model provides the user the ability to model the entire cryocooler and carry out optimization studies. If the actual solution cannot be accurately described using 6 time nodes. it has been found to be a valuable tool for the cryocooler designer because of the builtin optimization tool. the numerical scheme employed by Sage involves approximating the solutions using Fourier series. then the Sage model will be limited in accuracy. Their results are well matched to numerical predictions. the pulse tube correlations lack accuracy for small pulse tube. Oscillatory velocities and temperatures have been measured at both the warm and cold ends of the 14 . The solution is found such that the equations are satisfied at as few as 6 time nodes in the cycle. 40). et al. Although this model appears complete. the most comprehensive model for cryocooler systems is Sage (36). Yuan and Dybbs have developed a method to measure both the gas and solid temperature fluctuations in the regenerator of a Stirling engine (38). This method utilizes small thermocouples whose responses are compensated to correct for attenuation and lag. While this model may have some inaccuracies. Additionally. Harvey. Measurements within the regenerator matrix are inherently difficult due to the small geometry of the pores. have used hotwire anemometer probes in key locations in a pulse tube cooler (39.
especially in the cold region. catalytic reactors.3 Development of the Volume Averaging Technique The volume averaging technique is an analytical tool for describing the flow and heat transfer in a porous media. are certainly not hospitable to making hotwire measurements. some of which rely mainly on intuition and empirical observations (41). and mass flow rates. Hassanizadeh and Gray point out that there are at least three methodologies for describing the flow and heat transfer in multiphase systems.pulse tube and regenerator.2. All of the current regenerator models in the open literature. the volume averaging technique provides a formal framework for improving the science of porous media. Direct measurement of the flows in the cryocooler provides very valuable data for validating an numerical model. the instantaneous energy and entropy flows in the pulse tube. petroleum reservoir modeling. This technique has found extensive uses in ground water and pollution transport science. 1. While these methodologies have led to some of the original models. These measurements allowed them to calculate from pressure. these measurements are some of the most sophisticated measurements made with hotwires. such as the Darcy model. rely on the 15 . Typical measurements in wind tunnels have only small temperature fluctuations about the ambient. such as REGEN and Sage. temperature. These measurements were performed at temperatures below 90 K with fluctuations of 5 K. In fact. Installing the probes in a high pressure leak free fitting was also a feat. The efforts to make these measurements are remarkable since the conditions. and fluidized beds to mention a few.
development of conditions for nonhomogeneous porous media. these models fall short of describing in an exact fashion the flow and heat transfer in the regenerator. local numerical studies and experimental validations to investigate the validity of volume averaging closure conditions (42). Thus. they are also referred to as the microscale equations. Use of the volume averaging in this dissertation is applied to a problem which is highly compressible due mainly to the large pressure oscillations and the large temperature gradient across the regenerator.intuitive and empirical knowledge of the flow in the regenerator without any application of the volume averaging technique. As a result. Whitaker’s studies have mainly focused on incompressible flow. These equations are derived from the local governing equations for a generalized. This dissertation describes the derivation of the macroscopic equations which govern the regenerator problem. real fluid. the abundance of assumptions and modeling techniques for regenerators creates the need for a systematic study of the phenomenon based on fundamental conservation principles. Whitaker has been fundamental in the development of the volume averaging technique and it application to a variety of problems.4 Overview To this end. diffusion and dispersion in a reactor. compressible.2. conduction in multiphase systems. but he has briefly talked about the case of slightly compressible flow (43). These equations are referred to as local equations since they describe the flow of the fluid within the pores. 1. The microscale equations are then transformed into a set of macroscale governing equations 16 .
using volumeaveraging. Results and discussion are included in Chapter 5. and dispersion are then developed to reduce the equation set to a tractable problem definition. In this form. The chapter concludes with a presentation of a set of illustrative exact solutions and scale analysis. the governing equations describe the macroscopic flow behavior in addition to the effect that the microscale flow has on the macroscopic flow. Regenerator steady flow data and system level cryocooler performance data is summarized and discussed. Of significant importance is the development of the artificial convergence technique which allows the problem to be converged rapidly. The limiting models which were developed in Chapter 3 are 17 . The theory and details of the generalized volumeaveraging technique is included in Appendix 1. Chapter 2 begins by summarizing the microscopic and macroscopic governing equations as developed in Appendix 1. The numerical results of the detailed regenerator model are presented and compared with the system level model. A system level model is discussed which allows the regenerator boundary conditions to be estimated. Closure relationships for friction. The chapter concludes with the development and results of an exact solution verification. Chapter 4 details the development of the experimental apparatus and data. Assumptions are then developed and discussed which allow the governing equations to be simplified. Several models are developed which can be used to study several limiting assumptions such as constant temperature and local thermal equilibrium. heat transfer. The numerical method used to solve these models is the Method of Lines. Chapter 3 details the development of a series of computational models.
Future research topics are motivated and discussed. is provided for the baseline regenerator. Steady flow pressure drop data is used to predict friction factors using a compressible flow exact solution to the governing equations. including the system level model.compared with the full regenerator model. This friction factor is compared with friction factors measured in oscillating flow yielding excellent agreement. The important results are summarized. A comparison of two regenerator models from the literature. 18 . Chapter 6 concludes this dissertation.
CHAPTER 2 2. and the fluid properties are known functions of temperature and pressure. The governing equations are derived in any fluid mechanics text and are repeated here. The αphase is a Newtonian fluid and the βphase is a solid. In this system. The fluid phase is assumed to behave as a linearly viscous fluid.1 Governing Equations and Constitutive Relations The development of a theoretical model for flow and heat transfer in a generalized porous medium begins with a set of governing equations and constitutive relations for the thermohydraulic system illustrated in Figure 4. there is a solid and a fluid phase. The solid phase is assumed to be stationary and rigid with known thermal properties which are functions of temperature. V Figure 4 . 19 . The fluid flow is assumed to be compressible.Thermohydraulic system of a typical porous media. THEORETICAL DEVELOPMENT 2. • xα nα α β • xβ Averaging volume. or another combination of state properties.
2 Balance of Momentum Equations The differential statement of the balance of momentum for a Newtonian fluid with no body forces (in strong conservation form) is r r r r r ∂ (ρα uα ) + ∇ ⋅ (ρα uα uα ) + ∇pα − 1 ∇(µα ∇ ⋅ uα ) − ∇ ⋅ (µα ∇uα ) = 0.2.1. 20 . 2. This equation is a vector equation with three components corresponding to the three components of velocity. ∂t (21) r where ρα is the gas density and uα is the gas velocity vector. The balance of momentum equation produces 1 additional unknown. the fluid density and three components of the fluid velocity.1. The viscosity is a material property and is thus known from experimental data. and pα is the mechanical pressure. the fluid pressure.1 Conservation of Mass Equation The differential statement of mass conservation for the αphase (in strong conservation form) is r ∂ρα + ∇ ⋅ (ρα uα ) = 0. This equation is a scalar equation with 4 unknowns. ∂t 3 (22) where µα is the gas viscosity coefficient.
3 2 ( ) (27) The double dot notation represents the scalar product of two tensors. It can then be shown that r uα ⋅ ∇pα = − µα φα where φα is the viscous dissipation function defined as (26) φα = − r r r r r r r 2 (∇ ⋅ uα )2 + 1 ∇uα : ∇uα + 2∇uα : (∇uα )T + (∇uα : ∇uα )T .3 αPhase Conservation of Energy Equation The differential statement of the conservation of energy for a Newtonian fluid in terms of the gas specific internal energy. eα. although it is not immediately obvious. In this form. the pressure gradient can be eliminated by solving in terms of the viscous and acceleration terms. (24) It should be noted that Equation 23 is in strong conservation form. ∂t where hα is the gas enthalpy defined as hα = eα + (23) ρα pα . 21 .2. and the superscript “T” denoted the tensor transpose.1. is r ∂ (ρα eα ) + ∇ ⋅ (ρα uα hα − kα ∇Tα ) = 0. (25) Using Equation (22). Expanding this term gives r r r ∇ ⋅ (ρα uα hα ) = ∇ ⋅ (ρα uα eα + pα uα ) r r r = ∇ ⋅ (ρα uα eα ) + pα ∇ ⋅ uα + uα ⋅ ∇pα . the gradient of the enthalpy flow includes viscous dissipation. This is the preferred form for the conservation equations.
2. eα ). 2. Both energy equations are scalar equations introducing an additional 3 unknowns. requiring two additional equations to close the problem. Ideal gas equations or real gas equations can be used. and do not effect the form of the governing equations. the fluid and solid temperatures. the conservation of energy equation is written as (ρc )β ∂Tβ − ∇ ⋅ (k β ∇Tβ ) = 0. eα ) and Tα = f 2 (ρα .4 βPhase Conservation of Energy Equation For the βphase. These equations are the gas and caloric equations of state which can be generally expressed as pα = f1 (ρα . the system of equations consists of six equations for eight unknowns.1.5 Equations of State Thus far. ∂t (28) where (ρc)β is the solid heat capacity per unit volume. 22 . (210) (29) The particular equations of state are arbitrary. and the fluid internal energy.1.
local conduction due to molecular diffusion and dispersion.1. gastomatrix convective heat transfer through a finite film temperature difference.α represents the gas volumetric rate of entropy generation.α gen r qα r ∂ (ρα sα ) = + ∇ ⋅ (ρα sα uα ) + ∇ ⋅ T ∂t α ≥0 (211) where s′′′ .7 βPhase Entropy Generation Equation For the βphase. the entropy generation equation is written as s′′′ . is gen a thermodynamic property which is fundamental to optimizing any thermodynamic process. The entropy.2. The inequality indicates that the entropy generation is always positive except for totally reversible processes. it is zero. sα.6 αPhase Entropy Generation Equation The differential statement of the second law of thermodynamics is given as s′′′ . β = gen ∂ (ρ β sβ ) ∂t r qβ +∇⋅ Tβ ≥ 0. 2. It will be shown later that the entropy generation for the case of a porous media can be represented by three effects. in which case. 23 . and flow losses due to viscous and inertial effects. (212) The entropy generation in the solid phase is caused by local conduction due to molecular diffusion and gastomatrix convective heat transfer through a finite film temperature difference.1.
The flow geometry is far too complicated to allow for a direct application of these equations for any large scale porous system such as the regenerator. gas temperature. (218) (219) (220) pα = f1 (ρα . β = gen ∂ (ρ β sβ ) ∂t r qβ + ∇⋅ T β ≥ 0. Repeated here is this system of equations. matrix temperature. ∂ρα r + ∇ ⋅ (ρα uα ) = 0 ∂t r r r r r ∂ (ρα uα ) + ∇ ⋅ (ρα uα uα ) + ∇pα − 1 ∇(µα ∇ ⋅ uα ) − ∇ ⋅ (µα ∇uα ) = 0 ∂t 3 ∂ (ρα eα ) r + ∇ ⋅ (ρα uα hα − kα ∇Tα ) = 0 ∂t (213) (214) (215) (ρc )β s′′′ . These equations represent the microscale flow in the porous media.1.8 Summary of Equations The system of equations represents 8 equations and 8 unknowns.2. and gas pressure. nonvolume averaged gas density. The unknowns are the local. instantaneous. 24 . eα ) Tα = f 2 (ρα . eα ).α gen ∂Tβ ∂t − ∇ ⋅ (k β ∇Tβ ) = 0 ≥0 (216) r qα r ∂ (ρα sα ) = + ∇ ⋅ (ρα sα uα ) + ∇ ⋅ T ∂t α (217) s′′′ . gas velocity. gas internal energy. gas and matrix entropy generation.
2. The derivation of the volumeaveraged governing equations is the topic of Section 2.1 VolumeAveraged Conservation of Mass Equation The volume average of Equation (213) is ∂ ρα ∂t α + ∇ ⋅ ρα ( α r uα α )+ ε1 ∇ ⋅ ρˆ urˆ α α α = 0. 25 . The resulting equations are repeated here. or porosity as defined in Appendix 1.2.The flow is best analyzed in terms of volumeaveraged quantities. The notation indicates a volume average which is also defined in Appendix 1. This term will be addressed later.2.2 VolumeAveraged Equations The details of the volume averaging method are included as Appendix 1 in this document. The volume averaging results in an additional term to the standard continuity equation representing mass dispersion. 2. (221) where ε α is the gas phase volume fraction. The “hat” notation indicates a spatial deviation quantity.
The volume averaging has produced multiple terms which need to be eventually represented as functions of the volumeaveraged variables or eliminated justifiably. except these terms are now in the form of volumeaveraged quantities.3 VolumeAveraged αPhase Conservation of Energy Equation The volume average of Equation (215) is ∂ ρα 1 Vα 1 ( α eα α ∂t − + )+ ∇⋅( ρ α α α r uα α hα α )− ∇ ⋅ k ∇ T α α α + 1 Vα ∫ Aαβ ∫ Aαβ nα ⋅ kα ∇Tα dS + εα ∇ ⋅ ρα ( r ˆ ˆ uα hα ) ˆ ˆ r 1 ∂ ρα eα 1 α ˆ ˆ + ∇ ⋅ ρα uα hα εα εα ∂t r r 1 1 α ˆ ˆ ˆ ˆ ˆ + ∇ ⋅ ρα uα hα + ∇ ⋅ ρα uα hα = 0. the volume averaging has produced many additional terms which will be address in the following pages. 2.2.2.2. εα ( ( ) ε ) ˆ nα Tα dS (223) α Again. 26 .2 VolumeAveraged Balance of Momentum Equation The volume average of Equation (214) is ∂ α r α α r α r α ρα uα + ∇ ⋅ ρα uα uα + ∇ pα ∂t r r ~ 1 µ ˆ ˆ ˆ + ∫ nα ⋅ I pα − α ∇ ⋅ uα − µα ∇uα dS Aαβ 3 Vα r 1 ∂ 1 ˆ ˆ + ρα uα + ∇ ⋅ ρα ε α ∂t εα + 1 ( ) ( ) α r 1 − ∇ µα ∇ ⋅ uα 3 ( α )− ∇ ⋅ (µ ∇ ur ) α α α ( α r r r 1 ˆ ˆ uα uα + ∇ ⋅ uα ) εα r r ˆ ˆ ∇ ⋅ ρα uα uα ( α )+ ε1 ∇ ⋅ ρˆ urˆ urˆ α εα ( α r ˆ ˆ ρα uα ) (222) α α α = 0. The terms of the original energy equation have survived.
4 VolumeAveraged βPhase Conservation of Energy Equation The volume average of Equation (216) is (ρc )β 1 − Vβ ∂ Tβ ∂t β − ∇ ⋅ k β ∇ Tβ β + 1 Vβ ∫ ˆ nβ Tβ dS Aαβ (224) ∫ Aαβ nβ ⋅ k β ∇Tβ dS = 0. 27 . (226) represent the convection between the two phases. (223) and (224). − 1 Vα ∫ Aαβ nα ⋅ kα ∇Tα dS (225) and − 1 Vβ ∫ Aαβ nβ ⋅ k β ∇Tβ dS . are coupled via surface conduction heat transfer terms in the form of surface integrals over the αβ interface area. Later it will be shown that the surface integrals. Note that that the two energy equations.2.2.
28 .2.6 VolumeAveraged βPhase Entropy Generation Equation The volume average of Equation (218) is s′′′. gen β β = ρβ ∂ sβ ∂t β − ∇Tβ ∇ ⋅ k β εβ Tβ 1 1 − Vβ ∫ Aαβ nβ ⋅ k β ∇Tβ Tβ dS . (230) as well as entropy generation due to molecular diffusion. (228) As with the energy equations.2. gen α = ∂ ρα ( α sα α α + ∇ ⋅ ρα + − 1 ( ∂t sα ) + 1 ∂ ρˆ sˆ εα ∂t α α α r uα α α εα 1 r ˆ ˆ ∇ ⋅ ρα uα sα ( )+ ε )+ ε1 ∇ ⋅ ( ρˆ sˆ α α α r uα α )+ ε1 ∇ ⋅ ( ρ α α α r ˆ ˆ sα uα ) (227) 1 α r ˆ ˆ ˆ ∇ ⋅ ρα sα uα ∇Tα 1 ∇ ⋅ kα − Tα Vα εα ∫ Aαβ nα ⋅ kα ∇Tα dS Tα 2.2.5 VolumeAveraged αPhase Entropy Generation Equation The volume averaging of Equation (217) is ′ sα′′. − 1 Vα ∫ Aαβ nα ⋅ kα ∇Tα dS Tα (229) and − 1 Vβ ∫ Aαβ nβ ⋅ k β ∇Tβ Tβ dS . the entropy generation equations now contain terms which will be shown later to represent entropy generation due to solidtogas convection through a film temperature difference.
2) The porosity is constant. such as conductivity and viscosity. uα ~ uα 29 α .− ∇Tα ∇ ⋅ kα εα Tα 1 (231) and − ∇Tβ ∇ ⋅ k β εβ Tβ 1 .) This results in an r r ˆ order of magnitude estimate for the velocity deviation.3. can be treated as locally constant with respect to the averaging volume. all terms containing deviation quantities must be expressed as functions of the averaged variables. 2. 4) Fluid and solid properties.3 Simplifying Assumptions At this point in the development of the volumeaveraged equations. the assumptions are: 1) The solid phase is stationary and nondeforming.1 Density Spatial Deviation The pore scale velocity varies across the cross section of a pore due to the noslip Dirichlet boundary condition at the fluidsolid surface (see Figure 5. . To facilitate the equations to be in a tractable form. Properties are assumed to vary with the intrinsicaveraged state variables. 3) The fluid phase satisfies the noslip condition on the fluidsolid interface. (232) 2.
the density deviation is small compared to the volumeaveraged density (43). t ) r uα α r (x . Since density does not satisfy a Dirichlet boundary condition at the fluidsolid interface. and the volumeaveraged velocity. t ) α β Figure 5 – Conceptual diagram illustrating the local velocity. t ) r r ˆ uα ~ uα α rr u ( x. velocity deviation. { negligible α (233) The consequences of this assumption can be illustrated by considering the volume average of the ideal gas law which gives 30 . consequence of this assumption is The ρα = ρα α ˆ + ρα ≅ ρα .r r ˆ uα (x.
this analysis will neglect the mechanical dispersion terms. Thus. (235) Gray and O’Neill (44) identify the mechanical dispersion term appearing in the momentum equation. Whitaker (45) shows by scale analysis that the dispersion term for such flows is. 31 . in general.2 Negligible Mechanical Dispersion The mechanical dispersion term appearing in the momentum and energy equations is ∇ ⋅ ρα ( α r r ˆ ˆ uα uα ).3. This assumption will effectively eliminate a large number of terms in the volumeaveraged equations. 2. even though the temperature deviation is not negligible. negligibly small in comparison to the surface integral which represents the Darcy and Forchheimer effects. This conclusion is true for ideal or real gases. (234) Neglecting the density deviation eliminates the last term in Equation (234).pα α = R ρα α Tα α + R εα ˆ ˆ ρα Tα . They propose that this term can be expressed as r r ~ r ˆ ˆ uα uα = D2 uα α ~ r = E3 ⋅ uα α ~ r + G4 : uα α r uα α (236) where the numerical subscripts represent the rank of the tensors.
~ The closure variable.3 Thermal Dispersion The thermal dispersion term in the fluid energy equation is ∇ ⋅ ρα ( α r ˆ ˆ uα hα ).α Dα ⋅ ∇ Tα ) α ]. bα . that this dispersion is diffusive at the macroscopic scale and the dispersion coefficient is proportional to the Peclet number (42). The effective conductivity is then defined as ~ 1 kα . (239) Equation (238) can then be written as − ∇ ⋅ ε α kα .eff ∇ Tα ( α + ε α ρα α ~ Dα ⋅ ∇ hα α ) (240) which can be simplified assuming ∇ hα α = c p . 32 (242) . Taking the molecular diffusion term and the dispersion term together. Whitaker shows in an analogous fashion. Whitaker suggests that these terms reduce to ~ 1 − ∇ ⋅ ε α kα Ι + V ∫ ~ nα bα dS ∇ Tα Aαβ α − ∇ ⋅ ε α ρα ( α ~ Dα ⋅ ∇ hα α ) (238) ~ where the dispersion tensor is Dα .α ∇ Tα α (241) to − ∇ ⋅ ε α kα .eff + ρα [ ( α ~ c p .eff = kα Ι + V ∫ Aαβ ~ nα bα dS .2. (237) This term can be at least an order of magnitude larger than the molecular diffusion term.3. must be solved for numerically.
They reduce similarly to N k ∇T − ∇ ⋅ k α α α Tα α . which is an essential definition of thermal dispersion. 1 ∇ ⋅ ρα εα ( α r ∇Tα 1 ˆ ˆ sα uα − ∇ ⋅ kα εα Tα ) . 1 ∇ ⋅ ρα εα ( α r ˆ ˆ uα hα − ∇ ⋅ kα ∇ Tα ) α + 1 Vα ∫ ˆ nα Tα dS . 33 . The same terms in the gas entropy generation equation.eff + ρα α c p . empirical predictions (42).3.Gedeon also recognizes this Peclet dependence (36). This functional form for the dispersion agrees with other 2.4 Entropy Generation Due to Thermal Dispersion It was shown that the terms in the gas energy equation representing molecular diffusion and dispersion. (245) represent entropy generation due to dispersion and molecular diffusion. He calls this phenomenon axialconductivity enhancement. xx = kα (c1 + c2 Re m Pr n ) = N k kα (243) and he reports coefficients for several matrices of interest for regenerators such as wire mesh screens and felts.α Dα . He suggests that kα . Aαβ (244) reduce to − ∇ ⋅ N k kα ∇ Tα ( α ). respectively.
5 Negligible Brinkman Effect The Brinkman effect is represented by the term r 1 ∇ µα ∇ ⋅ uα 3 ( α )+ ∇ ⋅ (µ ∇ ur ) α α α (246) and is generally negligible in comparison to the Darcy and Forchheimer effects (45). It is reasonable to conclude that the wall will have minimal effect on the macroscopic flow in the axial direction. the wall boundary condition is simply zero normal velocity. The microscale flow is affected by viscous shear and interfacial heat transfer. The solid wall that contains the regenerator matrix has minimal effect on the macroscopic flow field.6 One Dimensional Model on the Macroscopic Length Scale It is improper to impose a noslip boundary condition on the macroscopic problem of a porous medium bounded by a solid surface. Typically the flow passages at the ends of the regenerator are designed with a contraction in diameter in an effort to minimize dead volume. yielding a spatially first order momentum equation. For an averaging volume containing the wall. The Brinkman effect will be neglected in this analysis. Both of these effects are related to surface area. For a 2D axisymmetric flow. which reduces to zero radial velocity for a cylindrical regenerator.3. It is not clear how this flow jetting will affect the regenerator performance. The wall can only satisfy a noslip boundary condition for the microscopic flow problem. 2.3.2. This will lead to flow jetting at the ends of the regenerator necessitating the need for a 2D model. the ratio of the wall surface area to the matrix surface area will be quite small. and this is an area of 34 .
These conditions lead to additional parameters which need to be measured experimentally. They find that the transition creates a jump condition in the momentum and energy equations. In this situation. The transitional region between the homogeneous fluid and homogeneous porous medium has been analyzed by OchoaTapia and Whitaker (46.ongoing research. This is the standard one dimensional continuity equation in terms of volumeaveraged quantities. These relations are developed in the following discussion. For the current model. it is assumed that the flow approaches the regenerator with a uniform profile.1 Simplified Conservation of Mass Equation Under these assumptions. 35 . 48). 2. magnitude of these effects. there is no driving force to create anything other than a one dimensional macroscopic flow field. the volumeaveraged conservation of mass equation (221) becomes ∂ρα ∂ + (ρα uα ) = 0. Several additional terms remain which need to be represented in terms of volume averaged quantities. ∂t ∂x (247) where the volume averaging notation has been dropped where appropriate.4 Simplified Equations The preceding assumptions lead to a large reduction in the equation complexity. The regenerator is bounded on the ends by a homogeneous fluid in which the differential continuum equations are valid. 47. Further analysis is needed to determine the 2.4.
are defined as ~ K xx = K ρ u K 1/ 2 ~ Fxx = ε α c f α α sign[uα ].2 Simplified Balance of Momentum Equation The momentum equation (222) becomes ∂ (ρα uα ) + ∂ (ρα uα uα ) + ∂pα ∂t ∂x ∂x ~ 1 r r µ ˆ ˆ ˆ + ∫ nα ⋅ I pα − α ∇ ⋅ uα − µα ∇uα dS = 0 Aαβ 3 Vα 3 14444444 244444444 4 Darcy and Forchheimer effects (248) which still contains the surface integral of the deviation quantities. the surface integral in Equation (248) can be written such that ∂ ~ ~ ~ (ρα uα ) + ∂ (ρα uα uα ) + ∂pα + µα ε α K −1 ⋅ I + F ∂x ∂x ∂t { [ ]} u xx α = 0. ∂t ∂x ∂x K K 2 (251) 36 . The permeability has units of length squared. (249) The permeability.4. µα 1 24 4 3 Re K (250) Note that the Forchheimer correction is expressed in terms of a Reynolds number based on the permeability length scale. cf. Following the analysis of Whitaker (45). and form drag coefficient.2. K. The momentum equation is now written as ε c ∂ (ρα uα ) + ∂ (ρα uα uα ) + ∂pα + µα εα uα + ρα uα 2 α 1 / 2f sign[uα ] = 0.
(254) where avH is the volumetric heat transfer coefficient. The last term in Equation (253) is a surface integral representing the volumetric heat transfer between the two phases.4. These two integrals are exactly of the same magnitude and are opposite in sign.3 Simplified αPhase Conservation of Energy Equation The volumeaveraged conservation of energy equation for the αphase (223) can be simplified to ∂ ρα 1 − Vα ( α eα α ∂t ) + ∂ ρ ∂x α α uα α hα α − N k kα ∂ Tα ∂x α (252) ∫ Aαβ nα ⋅ kα ∇Tα dS = 0.4 Simplified βPhase Conservation of Energy Equation The βphase conservation of energy equation (224) can be simplified to (ρc )β − 1 Vβ ∂ Tβ ∂t β ∂ ∂ Tβ − kβ ∂x ∂x β 1 + Vβ ∫ Aαβ ˆ dS nβ Tβ (253) ∫ Aαβ nβ ⋅ kβ ∇Tβ dS = 0. Whitaker (42) has proposed that these integrals be represented as − εβ 1 ε α Vβ ∫ Aαβ n β ⋅ k β ∇Tβ dS = 1 Vα ∫ Aαβ nα ⋅ kα ∇Tα dS = av H Tβ ( β − Tα α ).2. Substitution of Equation (254) into Equation (252) and Equation (253) results in a twoequation energy model 37 . The same integral appears in Equation (252). 2.4.
The terminology for tortuosity can be confusing since tortuosity is also used in the context of the fluid phase. the energy equations can be respectively written in final form as 38 . The ~ closure variable. felts. Experiments have suggested that the tortuosity is less than or equal to one. must be solved for numerically. Tortuosity equal to one represents parallel path geometries such as tube bundles. the tortuosity is thought to be a function of geometry only. bβ .∂ ρα ( α eα β α ∂t − av H Tβ ) + ∂ ρ ∂x α α α uα α hα α − N k kα ∂ Tα ∂x α ( − Tα )= 0 β (255) and (ρc )β + ∂ Tβ ∂t β εα av H Tβ εβ ( ∂ ∂ Tβ kβ − ∂x ∂x β 1 + Vβ − Tα α ) = 0. ˆ dS ∫Aαβ nβ Tβ (256) The conduction term in Equation (256) can be written as ∂ ~ 1 − k β Ι + ∂x V ~ ∂ Tβ ∫Aαβ nβ bβ dS ∂x β ∂ Tβ = − ∂ k β τ β ∂x ∂x β (257) where τβ is referred to as the “tortuosity” by Gedeon (36). and sintered metals have tortuosities less than one. and not the flow field. and dropping the volume averaging notation. Geometries such as wire mesh. Unlike dispersion. Using this result.
4. The same integral appears in Equation (261).4.6 Simplified βPhase Entropy Generation Equation The simplified gas entropy generation equation (228) is s′′′.5 Simplified αPhase Entropy Generation Equation The simplified gas entropy generation equation (227) is ′ sα′′.∂(ρα eα ) ∂ ∂T + ρα uα hα − N k kα α ∂t ∂x ∂x and − av H (Tβ − Tα ) = 0 (258) (ρc )β ∂Tβ ∂t − ∂Tβ ε α ∂ kβτ β ∂x + ε av H (Tβ − Tα ) = 0. ∂x β (259) 2. By a similar method as that with the energy method. gen α = ∂ ρα ( α sα α α ∂t )+ ∂ ( ρ ∂x α α sα α r uα α ) (260) ∂ N k ∂ − k α Tα ∂x Tα α ∂x 1 − Vα kα ∇Tα ∫Aαβ nα ⋅ Tα dS ≥ 0 2. gen β β = ρβ ∂ sβ ∂t β ∂ k βτ β ∂ − Tβ ∂x Tβ β ∂x β 1 − Vβ ∫ Aαβ nβ ⋅ k β ∇Tβ Tβ dS ≥ 0 (261) The last term in Equation (260) is a surface integral representing the entropy generation due to volumetric heat transfer between the two phases. these terms can be expressed as 39 .
gen = ρ β β ∂s β ∂t − ∂ k β τ β ∂Tβ av H (Tβ − Tα ) . ρT (267) which can then be used to simplify the entropy generation equations for the gas and the matrix.− 1 Vβ ∫ Aαβ nβ ⋅ k β ∇Tβ Tβ dS = av H Tα α (T β β − Tα α ) (262) and 1 Vα ∫ Aαβ nα ⋅ kα ∇Tα aH dS = v β Tβ Tα Tβ ( β − Tα α ). gen ≥ 0. sys = ε α sα′′. + ∂x Tβ ∂x Tα (265) At this point. using the energy and momentum equations. gen = ∂ N k ∂T a H ∂ (ρα sα ) ∂ + (ρα uα sα ) − k α α − v (Tβ − Tα ) ∂x Tα ∂x Tβ ∂x ∂t (264) and s ′′′. gen β (266) By using the Bridgman tables (49) the Maxwell relation for a pure substance can be written in a more convenient form as ds = 1 (dE − hdρ ) . the gas entropy generation equation reduces to 40 . gen + ε β s′′′. (263) Substitution of Equation (262) and Equation (263) into the entropy generation model and dropping the volume averaging notation results in ′ sα′′. After considerable simplification. which is ′ s′′′ . it is possible to write the entropy generation rate equation for the gasmatrix system.
gen = β k βτ β ∂Tβ 2 Tβ ∂x 2 av H + Tβ − Tα . Tα ∂x ∂x Tα Tβ 2 (268) For an incompressible substance. the Maxwell Relation reduces to ds = c dT . This is an interesting result. sys gen k τ ∂T a H 2 N k ∂T u ∂p = ε α k 2α α + ε β β 2β β + v Tβ − Tα − ε α α α ≥ 0. ∂x T T ∂x Tα ∂x Tβ Tα 144 44 144 44 14 244 14243 2 3 3 2 3 α β4 2 Gas conduction Matrix conduction Film heat transfer viscous and inertial losses 2 (271) It should be noted that all terms are positive definite.′ sα′′. this term can be written as − uα ∂pα uα ∂ρα uα ∂ρα uα uα +F≥0 + = ∂x Tα ∂x Tα ∂t (272) where F is the DarcyForchheimer surface integral. T (269) which. TT α β 2 (270) Now the system entropy generation equation can be written s′′′ . including the last term representing entropy generation due to viscous and inertial losses. This requires that the sign of the pressure gradient always be opposite to the sign of the velocity. 41 . gen = N k kα 2 Tα 2 u ∂p ∂T a H α + v Tβ − Tα − α α . and the matrix entropy generation equation reduces to s′′′. together with the matrix energy equation. Using the momentum equation. but not one that is immediately obvious.
as proposed. This produces the positive definite result − 2 uα ∂pα uα 1 µε 2 3 ε cf F= uα + ρα uα = ≥0 . − ∂t ∂x (275) The other implication which is possible to extract from this result is that if F IS a strict function of velocity. 1/ 2 Tα ∂x Tα Tα K K (276) There is an additional requirement on the functionality of the interfacial convection terms.F= 1 Vα ∫ Aαβ ~ µ r r ˆ ˆ ˆ nα ⋅ I pα − α ∇ ⋅ uα − µα ∇uα dS . 3 (273) This indicates that if F IS NOT a strict function of velocity as proposed then it must combine with the acceleration terms such that the entropy generation is positive definite. For positive velocity F ≥− ∂ρα uα ∂ρα uα uα − ∂t ∂x (274) and for negative velocity F ≤− ∂ρα uα ∂ρα uα uα . then the acceleration terms should be eliminated from the momentum equation. The assumptions imposed by Equation (254) must satisfy the additional requirement that 42 .
which are defined in the Appendix 1. TT Tα ∂x α β (278) The reference temperature is the lowest naturally occurring temperature in the system. An 43 .av H Tα α Tβ β Tβ β − Tα α 2 1 ˆ ˆ Tβ Tα 1 =∫ − α β Aαβ V T T Vβ Tβ Tβ α α α n ⋅ k ∇T dS ≥ 0. This idea is not investigated beyond this level in the dissertation other than calculating this value in Chapter 5. In this form. Tα and Tβ . This statement is produced by retaining the surface integrals representing the interfacial heat transfer through the simplification process for the gas and matrix entropy generation equations. These quantities are proportional to the volumeaveraged temperature difference. α α α (277) where the volume averaging notation is reapplied for clarity. This form of the lost power represents an internal method of calculation. The lost power is then the total entropy generation rate times the reference temperature giving & Wlost = To ∫ s′′′ . The entropy generation can be used to calculate lost available power by integrating the volumetric entropy generation rate over the entire regenerator. sys Adx gen 0 L N k ∂T 2 k τ = To ∫ ε α k 2α α + ε β β 2β Tβ Tα ∂x 0 Lr ∂Tβ ∂x 2 2 aH u ∂p + v Tβ − Tα − ε α α α Adx. The lost power is a scalar value which can then be used as an optimization parameter. The lost power represents the additional input power that is required to perform the same thermodynamic function as compared to an internally reversible refrigerator. it is clear that the entropy generation due to this effect is directly dependent on the ˆ ˆ temperature deviation quantities.
ε. The αsubscripts are dropped for simplicity. and the volumetric gas internal energy is defined as E. These two methods of calculating the lost power will give identical results for an analytical system. The lost power for a quasisteady system with only mass flow interactions with the external environment can be found by cyclical integration of the entropy flux at the boundaries.7 Summary of Equations The system of equations representing the one dimensional regenerator is repeated here with the volume average notations removed. This provides a good method for evaluating the accuracy of a numerical scheme.4.alternative method can be considered based on an external control volume. ∂ρ ∂m + =0 ∂t ∂x 2 µε m m 2 ε c f ∂m ∂ m 2 + p + + K ρ + ρ K 1 / 2 sign[m] = 0 ∂t ∂x ρ (279) (280) ∂E ∂ ∂T + mh − N k k − av H (Tβ − T ) = 0 ∂t ∂x ∂x (281) (ρc )β ∂Tβ ∂t − ∂Tβ ε ∂ k βτ β ∂x + 1 − ε av H (Tβ − T ) = 0 ∂x (282) 44 . The mass flux is defined as m. The porosity of the solid phase is defined to be one minus the gas porosity. but differences will be notices for a numerical approximation. Values of this lost power discrepancy are reported in Chapter 5 for the numerical model presented in Chapter 3. 2.
s′′′ = gen
2 N k k ∂T av H m ∂p Tβ − T − + 2 T ∂x TTβ ρT ∂x 2
2
(283)
s′′′, gen β
k τ ∂T = β 2β β Tβ ∂x
ε av H + 1 − ε TT Tβ − T β
2
2
(284)
2
s′′′ , sys gen
k βτ β N k ∂T = ε k2 + (1 − ε ) 2 T ∂x Tβ
∂Tβ ∂x
2 av H m ∂p + TT Tβ − T − ε ρT ∂x ≥ 0. β
(285) (286) (287)
T = f1 (ρ , E ) p = f 2 (ρ , E )
This system of PDEs represents 9 equations to be solved for
ρ
m
Gas density Gas mass flux Gas total energy per unit volume Gas temperature Gas pressure Matrix temperature Gas volumetric entropy generation rate
E T p Tβ s′′′ gen
s′′′, gen Matrix volumetric entropy generation rate β s′′′ , sys Total volumetric entropy generation rate gen The volumetric total energy, E , is chosen as the conserved quantity in the gas energy equation. This is simply the product of the internal energy and density. These equations require appropriate boundary and initial conditions which will be discussed. 45
These equations represent a fully compressible model for flow and heat transfer in a porous medium. The porous medium is completely characterized by the specification of the porosity, friction factor, dispersion coefficient, solid tortuosity, and heat transfer coefficient in addition to fluid and solid thermal properties. At this point in the
development, these equations apply to onedimensional steady, unsteady, or oscillating flow. For limiting cases, important exact solutions exist, and these are discussed in section 2.5 that follows.
2.5 Exact Solutions Experimental measurement of the permeability and inertia coefficient is conducted by measuring the steady mass flow rate and pressure drop through a onedimensional channel. Under steady state conditions, the first term in Equation (251) is zero, and the momentum equation reduces to
ε c µε du dp ρα uα α + α + α α uα + ρα uα 2 α 1 / 2f sign[uα ] = 0. dx dx K K
2
(288)
The continuity equation reduces to
ρα uα = C1 ⇒ ρα =
C1 , uα
(289)
and the momentum equation becomes (for positive velocity and ideal gas) µα ε α ε α 2 c f uα 2 − RT duα = − + 1/ 2 3 K KC1 dx uα which can be integrated to give ,
(290)
46
Ln(uα ) +
µ ε ε 2c RT = − α α + α 1 / 2f 2 K 2uα KC1
x + C2 .
(291)
The two constants can now be solved to give
C1 =
& m Af
(292)
and mRT A f 2 p0 2 & + , C 2 = Ln A p 2m 2 RT & f 0
(293)
where p0 is the pressure at the inlet. The result is an implicit equation for velocity which apparently needs to be solved numerically,
2 εα 2c f uα A f p0 1 1 1 p0 A f A f µα ε α + − Ln = + x. & & & RT mRT 2uα 2 2 RTm RTmK RTK 1/ 2
(294)
Equation (294) represents a steady, compressible, isothermal Forchheimer momentum equation. This equation allows the selection of the mass flow rate and the inlet pressure. By solving for the velocity at the outlet allows for calculating the density at the outlet, and thus the pressure, as outlined in what follows. In such a manner, experimental data can be used to solve for the two Darcy and Forchheimer coefficients iteratively. If the convective acceleration term is neglected in the momentum equation, an additional exact solution can be obtained for the case of steady, isothermal, compressible flow. The problem is defined by
47
ρα =
& & m dρ m duα ⇒ α =− 2 uα Af dx Af uα dx
2
(295)
εα c f 1 duα A f µα ε α = + , 3 & RTmK RTK 1 / 2 uα dx
which can be integrated to give A f µα ε α εα 2c f 1 x . =C− + 2 RTmK & RTK 1 / 2 2uα The constant can be solved to give 1 p0 A f . C= & 2 RTm The result is an equation for velocity,
2 εα 2c f 1 1 p0 A f A f µ α ε α = 2 RTm − RTmK + RTK 1 / 2 x , & & 2 2uα 2
(296)
(297)
(298)
(299)
which is identical to Equation (294) after dropping the natural logarithm term. This equation can be solved directly for the velocity giving
p0 A f 2 A µ ε ε 2c − 2 f α α + α 1f/ 2 x uα = RTmK & & RTK RTm
−1
.
(2100)
If the flow is steady, isothermal, and incompressible, the continuity equation reduces to duα =0 dx and the Forchheimer Momentum Equation reduces to
(2101)
48
KA f Af K 2 (2104) These three models are compared in Chapter 5. 49 .6 Scale Analysis A nontraditional scale analysis method is adopted for scaling the partial differential system of equations. separate temporal and spatial scales are used. Since a partial differential equation relates the change of a variable in more than one dimension. This equation implies that the pressure gradient and density are known constants and not related through an equation of state. the pressure gradient must be constant giving p L − p0 µ α ε α 2 εα c f + uα + ρα uα sign[uα ] = 0 Lr K K 1/ 2 2 (2103) The pressure at the outlet can be solved in terms of the inlet pressure and mass flow rate giving pL = µα ε α m m 2ε α c f & & p0 − Lr 2 RT + 2 1/ 2 . Thus. dx K K 1/ 2 2 (2102) which is the standard “incompressible” Forchheimer Equation. The compressible model and the incompressible model differ by 31% in predicting friction factor. The effect of the advective acceleration term is shown to be negligible in predicting friction factor. 2.dpα µα ε α 2 εα c f uα + ρα uα + sign[uα ] = 0 . it is sensible to expect that there are separate scales for each dimension. Thus.
2. By choosing independent temporal and spatial scales for density and mass flux.2 Balance of Momentum Equation Scale Analysis Next. the scaled continuity equation is written as ∂ρ m t s ∂m + ~ =0 ∂t Ls ρ ∂x (2105) where the tilde overbar represents the temporal scale and the dash overbar represents the spatial scale.6. ∂t ∂x (2107) 2. Ls and t s .1 Conservation of Mass Equation Scale Analysis The scale analysis begins by scaling the continuity equation. Arbitrary length and time scales. the momentum equation is similarly scaled to give ~ ~ ~ ~ ∂m ρ m ∂ m 2 ρp ∂p ρ m Ls m 2 sign[m] + ~ + f =0.6. respectively f = and ε Re K + ε 2c f (2109) 50 . + ~ ∂t ρ m ∂x ρ mm ∂x ρ m K ρ (2108) where the friction factor and the Reynolds number are. (2106) The scaled continuity equation then reduces to the familiar form ∂ρ ∂m + = 0. have been used. The result of this scaling is the single balance ts ~ ~ ρLs m .
ρ K (2112) These two scales lead to the scaled momentum equation. The length scale in the Reynolds number is the square root of the permeability. (2110) where the asterisks denote dimensional quantities. The fact that two length scales have been defined is evident since the volume averaging method requires such a result. The basic assumption of the volume averaging method is that the macroscopic length scale is orders of magnitude larger than the microscopic length scale.Re K = m* K µ* . This scaling leads to two balances. and it is expected that the model will eventually depend on two length scales. m ρ (2111) The remaining balance between the pressure gradient and the friction terms results in p~ ~ m 2 Ls . The macroscopic length scale represents the length scale of the macroscopic flow while the permeability length scale represents the length scale of the microscopic flow which is what drives the Darcy and Forchheimer effects. Thus. not the macroscopic length scale. The first balance is proposed for the temporal and spatial acceleration terms. which gives m ρ ~ ~ ~. Ls. it is necessary to define these two length scales. 51 . The two length scales are necessary to model a porous media as a continuum where the macroscopic and microscopic effects are united.
(2114) Since this parameter is much larger than the remaining acceleration terms. ~ 2 ρ Ls Γ1 = ρ K . ∂p m 2 sign[m] ∂m ∂ m 2 + Γ1 + f =0. + ∂t ∂x ρ ρ ∂x (2113) which is characterized by a leading order dimensionless parameter. the scaled gas energy equation is ~ Em ∂E m h ∂ (mh ) − k sT + ~ 2 ρLs ∂t Ls ∂x Ls which can be reduced to c T K ∂ ∂E c psT ∂ ks ∂T + ~ (mh ) − ps ~ Nk k ∂t cvsT ∂x ∂x cvsT Ls m c ps K ∂x ˆ c T L Hs − ps ~ 4 s H (Tβ − T ) = 0 K m c ps cvsT ∂ ∂T ˆ − av H sT H (Tβ − T ) = 0 Nk k ∂x ∂x [ ] (2115) (2116) if 52 . 2. T . which are O(1) in this form. This results in an algebraic equation for the velocity given the pressure field. it is tempting to consider dropping the acceleration terms.3 αPhase Conservation of Energy Equation Scale Analysis ˆ Using an additional scale for the interfacial temperature difference.6.
(2119) By defining the reference Peclet number and the reference Stanton number respectively as Pe = m c ps K ks (2120) and St = Hs . (2117) (2118) h = c psT . 53 . c ps m (2121) and the surface area per unit volume as av = 4 . and γ= c ps cvs . K (2122) the gas energy equation becomes T K 1 ∂ ∂T ∂E T ∂ + γ ~ (mh ) − γ ~ − γ Nk k ∂x ∂t T ∂x T Ls Pe ∂x ˆ T Ls St H (Tβ − T ) = 0 . ~4 T K (2123) The proper balance is between the interfacial heat transfer term and the enthalpy advection term which gives St ~ T 1 K .~ ~ ~ E = ρcvsT . ˆ T 4 Ls (2124) This leads to the final scaled gas energy equation.
T Γ2 = γ ~ T (2126) and a second order dimensionless parameter ε1 = K 1 . is very small in comparison to the first parameter. + K T 1 − ε (ρc )βs ∂Tβ (2128) There is only one relevant balance with this scaling. ∂ ∂E ∂T + Γ2 mh − ε1 N k k − H (Tβ − T ) = 0 ∂t ∂x ∂x (2125) which is characterized by an additional leading order dimensionless parameter. ~~ ~ T γ ε ρcvs (2129) Using this balance. 54 . This indicates that the gas energy equation is dominated be advection and interfacial heat transfer. which characterizes the diffusive flux. which is between the interfacial heat transfer term and the accumulation term.6. 2. This is represented by the balance T 1 1 − ε ( ρc )βs .4 βPhase Conservation of Energy Equation Scale Analysis The matrix energy equation can be scaled to produce (ρc )β ~ ρc T k ∂Tβ K 1 ∂ − vs ~ βs γ k βτ β ∂x ∂t (ρc )βs T ks Ls Pe ∂x ~ ˆ ε ρcvs L T γ 4 s ~ St H (Tβ − T ) = 0. Ls Pe (2127) The last parameter. the matrix energy equation can be written in its final form.
The boundary conditions provide the remaining information to complete the scale analysis.(ρc )β ∂T ∂Tβ ∂ − ε 2 k βτ β β + H (Tβ − T ) = 0 . this parameter is much smaller than the other terms which are O(1) which illustrates that the interfacial heat transfer and accumulation dominate over the diffusive flux. The time scale can be chosen as the angular frequency of the mass flow rate as 55 .5 Scaled Equation Summary The scaled equations are: ∂ρ ∂m + =0 ∂t ∂x ∂p m 2 sign[m] ∂m ∂ m 2 + Γ1 + + ∂t ∂x ρ ρ ∂x f=0 (2132) (2133) ∂ ∂E ∂T + Γ2 mh − ε1 N k k − H (Tβ − T ) = 0 ∂t ∂x ∂x (2134) (ρc )β ∂Tβ ∂T ∂ − ε 2 k βτ β β + H (Tβ − T ) = 0 ∂t ∂x ∂x (2135) which are characterized by four dimensionless parameters resulting from the scaling balances. 2.6. ε ks Ls Pe (2131) Again. ε2 = 1 − ε k βs K 1 . ∂x ∂x ∂t (2130) which is characterized by an additional second order dimensionless parameter.
(2136) The temporal mass flux scale can be chosen based on the boundary mass flux amplitudes as ~ m + mc . Then. (2143) 56 . ρ= pb (Th − Tc ) . m= h 2 (2137) The spatial mass flux scale can be expressed by assuming that the amplitude and phase varies linearly across the 1D domain. the RMS value of the gradient is m= mh + mc − 2mh mc cos(φmc − φmh ) . RThTc (2140) The five scales can now be simplified to Ls = 1 m2 ~ .ts = 1 ω . 2 2 2 (2138) The spatial temperature scale can be expressed as T = Th − Tc (2139) which then leads to the spatial scale for density. (2142) 2 ~ ε cvs ~ p (T − T ) m γ T = b h c RThTc m 1 − ε (ρc )βs . ωρ m (2141) ~ ρ=ρ ~ m m.
the dimensionless parameters can be written in terms of known quantities as Γ1 = ~ m ω Kρ 1 − ε m (ρc )βs ~ ε m ρcvs (2146) Γ2 = (2147) ε1 = K ω ε2 = ~ ρ 1 m Pe (2148) 1 − ε k βs ε ε ks 1 (2149) Table 1 summarizes the numerical results of the scale analysis for a representative regenerator. T= h c pb 4 St m 2 (2145) The importance of this scaling is that it has produced expressions which give insight into the amplitude of the temperature oscillation as well as the temperature difference between the matrix and gas. Finally. (2144) and ~ ˆ RT T ω K m .~ m3 1 p = ~2 ρ ω K . These amplitudes are seen to depend on the boundary conditions. 57 .
92E03 17.1146 [m^2] [rad/s] [Pam^3/kgK] [J/kgK] [J/kgK] [] [J/m^3K] [W/mK] [W/mK] ts Ls ~ ρ ρ ~ m Hs St Pe Nk 35482 2.01048 0.005E10 19.40E+06 0.21 230 1.5605 0.62 0 1.30E02 0.969E05 251 2078 5190 3112 1.Table 1Summary of scale analysis F Lr Dr Ε K mh mc φmh φmc Th Tc pb 40 7.67E01 12 7.46E04 7.10E03 [] [] [] [] 58 .60 25.6920 1.5E+06 12.92 24.62 300 70 3.0 1.0040 5.21 17.67 3.00E+04 707388 13.41 [W/m^2K] [] [] [] m p ~ E ~ T T ˆ T Γ1 Γ2 ε1 ε2 545 29.6 2.14E03 [Hz] [m] [m] [] [m^2] [kg/s] [kg/s] [rad] [rad] [K] [K] [Pa] [s] [m] [Kg/m^3] [Kg/m^3] [kg/s] [kg/s] [Pa] [J/m^3] [K] [K] [K] Af w R cps cvs γ (ρc )ms kb k 5.58 29.
6 Limiting Cases For the parameters chosen. it is expected that the upstream influence of the temperature will have an important effect. i.6. 59 . For example. the dimensionless parameters satisfy the assumptions imposed on the problem. Γ1 >> 1 Γ2 >> ε1 (2150) (2151) (2152) (2153) ε1 << 1 ε 2 << 1 There are several limiting cases of interest when these assumptions break down.e. and the frequency approaches zero.2.
1 Problem Definition The problem to be solved. Lr x mh (t ) & Dr mc (t ) & Figure 6 – Regenerator computational domain diagram 60 . onedimensional regenerator problem.CHAPTER 3 3. Figure 6 illustrates the computational domain. periodic. COMPUTATIONAL MODELS 3. as defined in Chapter 2. The geometry is a cylindrical domain. is the unsteady.
a numerical method was employed to approximate the solutions. this usually involves discretizing the spatial domain. 52. all MOL approaches are identical. The remaining domain remains analytical. the analyst must choose a unique method of time integration as well as spatial discretization. This technique relies on the proper specification of boundary conditions which is discussed below.2 Numerical Method The numerical solution of partial differential equations (PDEs) is a broad field.) In physical systems. which can be multidimensional. Several asymptotic models are then developed which illustrate the mechanisms which contribute to a net energy transport in the regenerator. the solutions are identical from one cycle to the next.) MOL is a general technique in which all but one domain of a multidomain system of PDEs is discretized (50. 51. 61 . i.e. This is followed by a discussion of a unique technique which was developed to artificially advance the solution in time to the quasisteady solution. A unique method of solving PDEs is a semianalytical technique call the Method Of Lines (MOL. The result of this discretization is a system of ordinary differential equations (ODEs. thus the meaning of semianalytical. Typical solution techniques in Computational Fluid Dynamics involve temporal and spatial discretization. 3. To accomplish this solution. At this level.The solutions which are desired are the quasisteady solutions. and 53). The details of the temporal and spatial representation are discussed in the following sections. To proceed.
ODE15S. For hyperbolic systems.2. Numerical techniques are generally classified as implicit and explicit. As will be shown later in the discussion of boundary conditions. was found to be the most stable and efficient. finite difference. MatLab was used as the platform for solving the equations. 58. Experimentation indicated that first order accurate central finite differences were the most stable and efficient if the equations were formulated in strong conservation form. The implicit ODE integrator.3. To 62 . and collocation (55. but this produces a larger banding in the Jacobian leading to reduced stability in the time integrator. 56. 59). In this context. 3. ODE15S is a variable order solver which can use numerical differentiation formulas (NDFs) or backward differentiation formulas (BDFs. it would be understandable that central differences would yield the most stable scheme. implicit time integration is necessary for stability. finite volume. Five point central differences yield fourth order accuracy. Finite differences were used to solve the regenerator equations.2.1 Time Integration Time integration can in general be an analytical technique or a numerical technique. ODE15S is especially efficient in solving stiff systems (54). the continuity equation reduces to a second order equation in density in the limit of explicit velocity formulation. Upwind differences did not appear to be as stable as central differences. Seven point differences were too unstable to use at all.2 Spatial Discretization Several spatial discretization methods exist in the literature such as finite element.) The BDF formulas are commonly known as Gear’s method and they suffer from lower efficiency. 57.
3. t ij β x j β k i 1− ε (ρc )β x i [ ( )] ( ) (34) for i=1. Dij . the central difference operator. i [ t ( )] ( t ) (33) i = 1 ε t t t t D k τ D jkTβ x − av H xi Tβ x − Txti . is an nxn square matrix. ∂t t ∂mxi (31) t 2 t t 2 mt 2 x j + p t − µ xi ε mxi − mxi ε c f sign mt . Equations (31) – (34) are now a coupled system of 4n ODEs which can be integrated in time from a set of 4n initial conditions.3 QuasiSteady Convergence via Cyclic Time Relaxation Time integration begins with an initial condition for each of the solution variables. In these discretized equations.calculate second derivatives. The solution at the final time step becomes the initial condition for the next period of integration. first order central differences using a three point stencil were used recursively. The equation set is integrated over a complete cycle. the MOL discretized equations are t ∂ρ xi t = − Dij mx j . = − Dij t xi xj t t ρ ρ xi K 1 / 2 K ρ xi ∂t xj ( ) ( ) [ ] (32) t ∂E xi ∂t and ∂Tβ x ∂t t t t t t = − Dij mx j hx j − N k x j k x j D jkTxtk + av H xi Tβ x − Txti . Appendix 2 contains a detailed discussion of the derivation of difference operators and their use with MatLab. The integration 63 .n where n is the number of spatial grid nodes.. For the equation set defined by Equations (279) – (282).
ss . ss = − N k k ∂T ∂Tss 1− ε . Cyclic averaging of the gas and matrix energy equations results in ∫ ∂ hss + q g . the system of equations needs to be integrated for thousands of cycles until quasisteady state is achieved unless the initial condition can be advanced artificially.continues until the solution for two consecutive periods are equal.ss + qm. Cyclic analysis of the governing equations provides an interesting and valuable technique called cyclic time relaxation. q g . In this case. the net change of internal energy of the system over a cycle is identically zero and the system is said to be in quasisteady state.ss dt = 0. ∂x ∂x ε (36) A simplification in terms of temperature gives ∫ ∂ ∂T ∂Tss 1 − ε − k βτ β β .ss dt = 0 . The regenerator system suffers from a long time constant with respect to the cycle period. this relation provides a means of calculating a correction for the initial condition of the following cycle. When this condition is met. Let hss = ρuc pTss . As a result. By defining the quasi steadystate temperature fields which satisfy Equation (37) exactly as 64 . ∂z [ ] (35) where the “ss” subscript indicates steady state. qm. ss = − k β τ β β . ρuc pTss − N k k ∂x ε ∂x ∂x (37) The temperature field at the end of a cycle may not satisfy this relationship.
T’ can then be solved numerically with a finite difference technique. respectively. f1 (z ) 2 + f 2 (z ) ∂z ∂z where 1− ε f1 ( x ) = − N k k + k βτ β dt ε f 2 (x ) = f 3 (x ) = uc p }dt (39) ∫ ∫ {ρ ∫ (310) ∂ ∂T ∂T 1 − ε − k βτ β β dt ρuc pT − N k k ∂x ∂x ε ∂x For the case of constant specific heat.Tss = T ′(x ) + T ( x.t) and Tβ(x.t) are the temperatures of the gas and matrix.ss = T ′( x ) + Tβ (x. (38) T(x. t ). Tβ . f 2 (x ) = c p ∫ {ρ } u dt = 0 (311) which follows from a cyclicaverage of the continuity equation. t ) . where the primed temperature is an axial correction. (312) since the temperature field at the ends does not need a correction due to the boundary conditions which fix those values. Equation (39) is a second order ODE which can be solved for the temperature correction function. The boundary conditions are appropriately chosen as T ′(0) = 0 T ′(Lr ) = 0 . from the current cycle. These quantities can be substituted into Equation (37) and T’(z) can be solved to obtain ∂T ′ ∂ 2T ′ + f 3 (z ) = 0 . Then the initial condition for the temperatures for the next period is 65 .
Boundary conditions are needed at both ends for stability and this can be understood by considering the limiting case when the acceleration terms in the momentum equation are neglected. λ. x ) = Tβ (τ .T (0. 3. When the velocity in the continuity equation is eliminated in terms of temperature and density.4 Boundary and Initial Conditions Equations (279) . Mass flux boundary conditions are chosen such that zero net mass flux can be achieved. x ) + λT ′( x ). ∂x k k 2 (314) It is then clear that velocity is a function of the pressure gradient. Stable solutions are found by imposing pressure at both ends or mass flux at both ends. is a function of the density gradient and temperature gradient. 66 . ε c ∂p µα ε α + u + ρu 2 α1 / 2 f sign[u ] = 0. in turn. x ) + λT ′( x ) Tβ (0. This results in the explicit equation for velocity.300. These equations in strict mathematical terms require six boundary conditions for the problem to be wellposed. the result is a spatially second order continuity equation. (313) A relaxation factor. Two boundary conditions are required by the continuity and momentum equations.(282). in their present form. The continuity and momentum equations are both first order spatially while the energy equations are both second order. is used to aid stability. which.35. are a mixed system of first and second order (spatially) partial differential equations. x ) = T (τ . Experimentation has indicated an optimum value of 0.
In actuality.∂ρ ∂ ∂ρ ∂T + ρf .e. Referring back to the scaled energy equations. This motivates a simplified application of boundary conditions based on a leading order simplification of the energy equations. the regenerator is largely unaffected by the incoming gas temperature. The boundary condition requirements of the energy equations will absorb this overspecification. the diffusion terms are characterized by a small dimensionless parameter. the momentum equation is a spatially first order equation. Together with the first order continuity. only one boundary condition is required. the problem is overspecified by one boundary condition at this point. the continuity equation is diffusive. Scaling and numerical experiments suggest that these second order diffusion terms are essentially steady. Outside of this boundary layer. the temporal fluctuations are orders of 67 . i. This indicates that there is the Numerical possibility for a thermal boundary layer at the ends of the regenerator. Thus. experiments suggest that resolving such a boundary layer does not have a significant impact on the regenerator solution since the temperature fluctuation in the regenerator is largely a result of the compression process. spatial sources. ∂t ∂x ∂x ∂x (315) In this limiting case. The advective acceleration term which makes it first order is quite small and can be neglected producing a zeroth order momentum equation. it seems natural to impose mass flux boundary conditions at both ends of the domain. = 0. Since two mass flow boundary conditions have already been specified.
This suggests that the boundary conditions that are required mathematically can be neglected with little effect on the solution. This 68 . For the limiting problem. not the continuity equation. Thus two boundary conditions are needed which satisfy the important corollary condition of zero net mass flux.e. the proper initial condition is one that produces a quasisteady solution after one cycle of simulation.magnitude smaller than the spatial variations. the gas energy equation requires a single boundary condition while the matrix energy equation requires no boundary conditions. To summarize. & Af (317) These mass flux boundary conditions are actually boundary conditions for the momentum equation. but the matrix energy equation is zeroth order. Imposing mass conserving boundary conditions for the continuity equation is difficult since simple sinusoidal densities at the ends will not in general lead to mass conservation. the solutions of interest can be found by neglecting the strict mathematical requirement of six boundary conditions. then density boundary conditions could be used. the equations consist of a first order continuity and gas energy and zeroth order momentum and matrix energy. The gas energy equation is still first order due to the enthalpy advection term. i. Since the solution of interest is the quasisteady solution. The most satisfactory method of achieving this is by imposing harmonic mass flux boundary conditions. As a result. If the entire cryocooler is modeled as a closed system. (ρu )h = mh cos(ωt + φmh ) & Af (316) and (ρu )c = mc cos(ωt − φmc ) .
Additionally. Thus. Lr Dr Regenerator length Regenerator diameter Porosity Friction factor Nusselt number Dispersion coefficient Matrix tortuosity ε f Nu Nk τβ 69 . & mh & mh Warm end mass flow amplitude Cold end mass flow amplitude Mass flow phase shift φmh − φmc ω Th Tc pb Angular frequency Warm end mean temperature Cold end mean temperature Mean pressure These parameters are in addition to the seven geometric and empirical parameters which define the shape of the regenerator and the pores. this technique is responsible for producing a solution which has the desired mean pressure and mean warm and cold end temperatures.is essentially the purpose of the artificial convergence technique discussed above. the converged solution satisfied seven boundary parameters.
which provide additional permutations.3. Each model can be considered with real or ideal gas. this requires that the temperature field be specified a priori. ∂x (318) which can then be inserted into the gas energy equation to give ∂T ∂T ∂E ∂ − k βτ β β + ρuh − N k k ∂x ∂x ∂t ∂x = 0. Two asymptotic cases are presented. Results of numeric computations with these models are presented in Chapter 5. There it is revealed that the regenerator net energy transport increases with each additional relaxation.e. The simplest idealized regenerator is represented by the Constant Temperature Model (CTM. Mathematically. only a spatial temperature field.5 The Constant Temperature Model (CTM) A variety of asymptotic models can be considered. i. (319) The cycle average of this equation is 70 . there is no temperature oscillation. Each asymptotic model provides insight into the mechanism of net energy transport in the regenerator. and constant or variable properties. The matrix gas energy equation can be rearranged to give − ∂Tβ ∂ kβτ β ∂x = av H (Tβ − T ). The consequence of the first assumption is that there is an infinitesimal temperature difference between the gas and matrix. The consequence of the second assumption is that the temperature field is steady. beginning with the most restrictive and an intermediate model.) This model assumes 1) that the heat transfer between the gas and the matrix is perfect and 2) that the matrix heat capacity is very large.
can be shown to be zero by cyclic integration of the continuity equation. which is the net enthalpy flow. This model is thermally perfect. The steady temperature field is then affected by the diffusion and net enthalpy flow fields. the system of equations being solved is 71 . The regenerator loss due solely to the density dependence of the enthalpy can now be assessed. For ideal gas with constant properties. This is obvious since the enthalpy is a function of temperature and density. this becomes c pT ∫ ∂ (ρu )dt + ∂x ∫ ∂T ∂T ∂ − Nk k − k βτ β β dt = 0. periodic operation. If a real gas equation of state is used. ∂x ∂x ∂x (320) Note that the cycle average of the time rate of change of the volumetric internal energy is identically zero for steady. ∂x ∂x ∂x (321) The first integral. which remain unchanged. This is the only condition under which the net enthalpy flow is zero. Thus. Beginning with a linear temperature field. then the net enthalpy flow is positive.∫ ∂T ∂T ∂ ρuh − N k k − k βτ β β dt = 0. The diffusion fields drive the steady state temperature field for this case. but the effects of pressure drop and viscous dissipation are still captured via a coupled solution of the continuity and momentum equation. the cyclic time relaxation procedure is applied to advance the solution to steady state.
In this model. this combined energy equation can be rewritten as e ∂Tβ 1 − ε ∂Tβ ∂ ∂T 1 − ε ∂T ∂ρ = 0.∂ρ ∂m + =0 ∂t ∂x 2 ∂m ∂ m 2 ∂p µα ε m m 2 ε c f + + + + sign[m] = 0 ∂t ∂x ρ ∂x k ρ ρ K 1/ 2 (322) (323) This system of PDEs requires two mass flux boundary conditions.6 The Local Thermal Equilibrium Model (LTEM) Local thermal equilibrium should not be confused with local thermodynamic equilibrium. Cyclic time relaxation is applied to find a temperature field which yields a steady solution. − + ρuh − N k k + ( ρc )β k βτ β ε ∂x ε ∂t ∂x ∂x ∂t (324) For ideal gas. − + ( ρc )β + ρue + Pu − N k k + ρcv k βτ β ε ε ∂t ∂x ∂x ∂x ∂t ∂t (325) Using the fact that the temperatures of the gas and matrix are identical. 3. This energy equation is formed by combining Equations (281) and (282) such that ∂Tβ 1 − ε ∂Tβ ∂ ∂T 1 − ε ∂E = 0. The result is a single energy equation with a single temperature. as defined previously in addition to initial conditions for density and mass flux. the assumption of very large matrix heat capacity is relaxed while retaining the assumption of perfect gastomatrix heat transfer. the combined energy equation is now written as 72 .
∂T 1 − ε ∂T ∂ρ ∂ ρcv + (ρc )β ε ∂t + e ∂t + ∂x (ρu ) + ρucv ∂x ∂T ∂ ∂T 1 − ε k β τ β β = 0. and the final form of the LTE energy equation is ∂Tβ 1 ∂ ∂T 1 − ε ∂T ρucv ∂T Pu − N k k − + + k βτ β ε ∂x ∂x ∂t C LTE ∂x C LTE ∂x = 0. ∂x ∂x ∂x ε (330) This is wellposed mathematically. Expanding Equation (324) for a real gas gives e ∂e ∂ρ ∂ρ 1 − ε ∂Tβ ∂e ∂T + ρ + ρ ∂ρ ∂t + (ρc )β ε ∂t ∂t ∂T ρ ∂t T ∂T ∂ ∂T 1 − ε k βτ β β = 0 + ρuh − N k k − ∂x ∂x ∂x ε (329) which can be simplified to ∂e ∂e ∂ (ρu ) 1 − ε ∂T − e + ρ + (ρc )β ρ ∂ρ ε ∂t T ∂x ∂T ρ ∂T ∂ ∂T 1 − ε + ρuh − N k k − k β τ β β = 0. (328) Implementing an equivalent combined energy equation for a real gas is more difficult due to the additional density dependence of the gas internal energy. The boundary conditions using either 73 . (327) where the combined volumetric heat capacity is defined as C LTE = ρcv + (ρc )β 1− ε ε . + Pu − N k k − ∂x ∂x ∂x ε (326) Continuity can be used to eliminate the second term. The additional complexities include the calculation of the partial derivatives of the internal energy.
Boundary mean temperatures: Th and Tc & & Mass flowrates: mh .energy equation are the same. and φmh − φmc Mean pressure: pb Macroscopic geometry: Lr and Dr Additionally. τ β . the model depends on several correlations which are functions of the microscopic geometry and/or the flow field: Nk. cyclic time relaxation is applied to find a temperature field which produces a steady solution. The characterization of this model depends on several parameters which can be categorized. repeated here as ∂ρ ∂m + =0 ∂t ∂x 2 µε m m 2 ε c f ∂m ∂ m 2 + + p + + sign[m] = 0 1/ 2 ∂t ∂x ρ K ρ ρ K (331) (332) ∂E ∂ ∂T + mh − N k k − av H (Tβ − T ) = 0 ∂t ∂x ∂x (333) (ρc )β ∂Tβ ∂t − ∂Tβ ∂ ε k βτ β ∂x + 1 − ε av H (Tβ − T ) = 0 ∂x (334) This model captures the effects of finite heat transfer and matrix heat capacity as well as flow friction. mc . and ε . f . 74 . 3. two mass flux BCs. w .7 Dual Energy Equation Model (DEEM) The Dual Energy Equation model solves the complete equation set. Nu. Again.
e. x ) ∂t ∂x µε m m 2 ε c f ∂m ∂ m 2 + p + + K ρ + ρ K 1 / 2 sign[m] = F2 (t . ∂ρ ∂m + = F1 (t . x ) ∂t ∂x ∂x (337) (ρc )β ∂Tβ ∂t − ∂Tβ ∂ ε k βτ β ∂x + 1 − ε av H (Tβ − T ) = F4 (t . x ) . i.8 Model Verification The accuracy of the Dual Energy Equation model was verified by using a test solution in a method similar to Kirkconnell (60). ∂x (338) The test solution was chosen as follows: m = M 2 cos(ωt − M 3 ) T = T1 + T2 cos(ωt − T3 ) P = pb + P2 cos(ωt − P3 ) (339) (340) (341) (342) ρ= u= P RT m ρ (343) 75 . x ) ∂t ∂x ρ 2 (335) (336) ∂E ∂ ∂T + mh − N k k − av H (Tβ − T ) = F3 (t . The particular solution was chosen such that it retained the important characteristics of the actual solution. An arbitrary solution can be made to satisfy the system of equations if appropriate source terms are added to each equation.3.
These parameters are chosen from a representative numerical solution. Pc . Fi. Equation (350) was chosen such that the net enthalpy flux would be identically zero. nonzero net enthalpy flux. 76 . ~ ~ ~ ~ Th . Adding a nonzero enthalpy flux will produce a similar solution with a shift in the net enthalpy field. The source terms. Ph . were then analytically solved.E = ρ cvT Tβ = T M 2 = mh + (mc − mh ) x Lr x Lr (344) (345) (346) M 3 = φmh + (φmc − φmh ) T1 = Th + (Tc − Th ) x Lr (347) (348) ~ ~ ~ x T2 = Th + Tc − Th Lr T3 = M 3 − ( ) (349) π 2 (350) ~ ~ ~ x P2 = Ph + Pc − Ph Lr P3 = φ Ph + (φ Pc − φ Ph ) z Lr ( ) (351) (352) All of the parameters in the preceding equations come from the boundary conditions with the exception of the boundary temperature and pressure oscillation amplitudes. and the pressure phase angles. The zero enthalpy flux case was chosen for simplicity. φ Ph and φ Pc . Alternatively. a condition could be constructed to produce a constant. Tc .
There is less that 0. Additionally. 77 .The numerical solution was then calculated using the exact numerical method detailed above. The net enthalpy flux is visually identically zero. The solution fields show very small errors. the model converges in one cycle.2 mW deviation. The normalized errors are: Density Mass flux Energy Matrix temperature Gas temperature Pressure 5x107 3x107 4x107 3x108 4x108 4x107 The small normalized errors indicate that the numerical scheme is not introducing unrealistic errors into the solution and that the solver can solve similar equations. Since the initial condition can be set using the known exact solution. the question of the proper set of boundary conditions is answered. The source terms should not dramatically affect the character of the equations. The model numerically solves the test equations to very high accuracy as illustrated in Figure 8Figure 14. This solution was obtained using just two mass flux boundary conditions even though the strict mathematical requirement is six boundary conditions.
06 0.01 0.5 0 0.5 0 0. Cycle Time [radians] 2 1 0 1 2 3 4 Pressure [MPa] vs.Density [kg/m3 ] vs.015 0.025 2.02 0.08 0 0.01 0.06 0.025 Figure 7 – Test case solution results.02 0. Position [m] 20 um ax um ean um in 0 0 0.04 0.02 0.005 0.005 0.02 0. Cycle Time [radians] 30 4 2 0 10 2 4 Velocity [m/s] vs. Cycle Time [radians] 3.01 0.015 0. Cycle Time [radians] 20 0 0 0.015 0.01 0.04 0.08 Mass Flow Rate [g/s] vs.005 0.02 0.02 0.005 0.025 Density [kg/m3] vs.015 0. 78 .025 0 0. Position [m] 30 rho rhom ax rhom ean m in 4 2 0 10 2 4 Velocity [m/s] vs.
04 Position 0.06 0.0.5 1 x 10 8 Cycleaveraged mass flow [g/s] 10 8 6 4 x 10 6 Cycleaveraged mass flow gradient [g/s] 1.5 3 0 0.0 Cycleaveraged energy flows [mW] 80 Total 60 Qm 40 60 Cycleaveraged enthalpy flow gradient [mW] 40 20 20 0 Qg H 0 20 20 0 0.06 0.04 Position 0.04 Position 0.08 40 0 0.06 0.02 0.5 0 0.02 0.04 Position 0.02 0.0 Figure 8 – The net enthalpy flux satisfies the test case to within visual accuracy (0.5 2 2.08 2 0 2 0 0.2 mW deviation) 79 .02 0.06 0.
Figure 9 – Maximum density error 8x106 kg/m3 Figure 10 – Maximum mass flux error 1x105 kg/m2s 80 .
Figure 11 – Maximum energy error 2 kJ/m3 Figure 12 – Maximum matrix temperature error 6x106 K 81 .
2 Pa 82 .Figure 13 – Maximum gas temperature error 8x106 K Figure 14 – Maximum pressure error 1.
two asymptotic models were developed. A numerical method was then developed using the Method of Lines with central finite differences. To provide a comparison with the full model. A mathematical model for the regenerator was developed based on fundamental principles and guided by intuition. it is worth reviewing what has been accomplished up to this point. Chapter 5 presents the modeling results and compares these results to the experimental data from Chapter 4. 83 .Before proceeding on to Chapters 4 and 5. Chapter 4 describes the development of an experimental apparatus which was used to investigate this model. An artificial convergence technique was developed which dramatically improves convergence speed.
the apparatus was designed to accommodate five interchangeable regenerator/cold heat exchanger/pulse tube assemblies (Figure 16). to characterize overall cryocooler performance as a function of pulse tube aspect ratio and volume (61). expanders 1 through 3 are of identical volume.CHAPTER 4 4. With the exception of the regenerator materials. 4. which has a 17% larger pulse tube volume than the constant volume set and performed slightly better in the previous experiments (no load temperature of 76 K vs. et al. An unfortunate oversight resulted in the testing of one of the regenerators in expander 5. 61 for additional details on the apparatus. hereafter referred to as “expanders. The differences between the performances of the various regenerators were much larger than this pulse tube volume effect. The reader is referred to ref. and 5 are of constant length. so the data from expander 5 are still considered relevant. and expanders 1. 84 . the identical cooler assembly was used by Kirkconnell.” Using the numbering scheme of the previous paper.1 Overview and Experimental Apparatus The apparatus used is an Orifice Pulse Tube Cryocooler (OPTC) design and is illustrated in Figure 15. Towards that end. EXPERIMENTAL APPARATUS AND MEASUREMENTS 4. however. hence the use of these three is ideal for the present effort in which the unique impact of regenerator design on performance is sought. The performance of the constant volume pulse tubes (13) were found to be virtually indistinguishable. 79 K).
85 .V a c u u m E n c lo s u r e H 2 O C o o la n t L o o p H 2 O C o o la n t L o o p C o ld H X O r ific e V a lv e T o C o m p resso r T o S u rge T ank In le t H X R e g en e r a to r P u ls e T u b e R e je c tio n H X A d a p to r P la te F le x ib le M e m b r a n e Figure 15 . Figure 18 shows a closeup view of the compressor. Hughes Condor on loan from Night Vision & Electronic Sensors Directorate (NVESD). Figure 17 shows the apparatus integrated with the vacuum system and data acquisition system. dual opposed piston. Regenerator is the larger diameter section.Experimental Apparatus Reproduced with authors’ consent (61) Figure 16 – Regenerator/Pulse Tube expander module. The compressor used is a nominal 3 cc swept volume.
This vacuum system is capable of achieving pressures as low as 2x106 torr.Figure 17 – Experimental apparatus with complete instrumentation. The chamber surrounding the expander is evacuated using a turbomolecular vacuum pump. Figure 18 – Closeup view of the compressor. shown in Figure 19. These pressures are necessary to eliminate convective and conductive heat 86 .
it is recommended that a vacuum level of 106 torr be achieved and maintained before cooling a cryogenic system. Figure 19 – Turbo vacuum pump station. Further reduction of the pressure is limited by slow moisture release from the chamber and MLI. 87 . Around 10 torr. At this pressure. MLI (multilayer insulation) is a material used as a radiation shield. Typically. At these pressures. but conduction will continue at levels comparable to atmospheric pressure. the heat carried by circulation currents is very small. natural convection will occur at atmospheric pressures down a few torr. It is composed of layers of aluminized mylar with woven Dacron spacer layers. the residual moisture is low enough that it will not form significant frost and alter the MLI performance. Typically.transfer through the surrounding gas. the conductivity of the low pressure gas will begin to decrease until it is essentially zero around 104 torr (62). the only mode of heat transfer is radiation. At this pressure and lower.
Four regenerators were tested in a previous study (63. For a sample of 30 400 mesh screens. the total thickness of the stack was measured at 1. Since that report. but the same material. an additional 45 micron regenerator and a diced foam metal regenerator were tested.717 (spacers) 4 45 Micron Perforated disks Alabama Cryogenic Pore size: 45 microns Engineering Measured Porosity: 0.614 In packing wire mesh screens. the porosity can approach 50%. The regenerators are summarized in Table 2 with their relevant data.325 wires per inch Wire diameter: 27.9 µm Measured Porosity: 0.Descriptive Summary of the Regenerators under Study # Name / Figure Description 1 325 Mesh / Figure 20 Wire mesh screens .2 microns. or a half 88 . The density of the homogeneous material and the total regenerator volume can be used to calculate the matrix porosity. 64). Table 2 .536 mm. the actual porosity can vary by several percent based on the degree of nesting. All six regenerators are made of stainless steel fabricated into different forms of porous materials. The porosity of each regenerator was measured experimentally by measuring the total matrix mass. For wire screens perfectly aligned.4 µm Measured Porosity: 0. This gives an average screen thickness of 51.644 (no spacers) 5 Foam Metal / Figure 22 Sintered foam metal plug Measured Porosity: 0.692 3 60 Micron / Figure 21 Perforated disks Alabama Cryogenic Pore size: 60 microns Engineering (65) Measured Porosity: 0.696 2 400 Mesh / not shown Wire mesh screens – 400 wires per inch Wire diameter: 25. They consisted of quite different geometry.
It is possible that sintered matrices can provide reasonable regenerator performance. 89 . The target porosity of the foam metal regenerator was chosen such that it was in the same range as the wire mesh screens. The performance test with the foam metal regenerator is the critical measure of the matrix performance.1 for the 30 micron disks. The process is able to produce sintered matrices for a range of porosity. An 8% porosity difference may cause a significant performance change. but the other matrices have a much better understood geometry in terms of wire diameter and pitch or perforation diameter and aspect ratio. The result is a rigid matrix with random pore geometry and variable pore size. especially for a low cost application where performance is not as critical.7 and 7.6 microns. The half thickness is very close to the nominal wire diameter suggesting that the screens were not nesting. The actual porosity was 8% less than the wire mesh regenerators. but shrinkage of the matrix during sintering most likely requires beginning with a higher porosity.thickness of 25. The morphology of the pore and solid structure is not studied in this dissertation. The 60 micron perforated disk thickness was measured at 211 microns. System level modeling suggests that the regenerator performance is sensitive to the matrix porosity. The process used to make the sintered foam metal is proprietary. the flow begins to resemble that of a tube bundle. As the aspect ratio increases. The foam metal regenerator was fabricated from 100 micron particles which were sintered. This results in a perforation aspect ratio of 3. The 45 micron disks have an aspect ratio of 4.5.
The fourth regenerator is drastically different from the others. Figure 20 . Each disk is then lightly tamped to assure that it is snugly packed into the tube. the foam metal regenerator takes only a few minutes. Additionally. Whereas the single disk regenerators each take as much as eight hours to hand pack.0015” gap between the disks to allow for flow redistribution between disks.Foam Metal (200X magnification) 90 . This allows significant time savings in packing the regenerator. These disks are packed into the regenerator tube making sure that the disks are not bent. like the screens. This makes the screens and the disks more difficult to pack at low porosities which are desirable. the perforated disks were packed using spacer rings which create a 0. This foam metal regenerator is a single plug of porous metal (Figure 22). The perforated disks take considerably less time to pack since they are about ten times as thick as 400 mesh screens. The disks. have to be restrained in the regenerator tube due to the spring force of the stack.Wire Mesh (200X magnification) Figure 21 .The first three regenerators are composed of a stack of disks made from the particular material (Figure 20 and Figure 21).Perforated Disk (200X magnification) Figure 22 .
This system consists of a series of lab instruments which provide a variety of functions and a PC (Figure 23). Figure 25. The instruments and PC are connected using a GPIB 488 interface. GPIB is a laboratory instrumentation standard which provides communication and control via a PC. Figure 23 – Data acquisition system. and Figure 26. 91 . This provides a realtime data acquisition environment which provides for much more sophisticated measurements due to automation. Agilent VEE Pro software was used to create the data acquisition and control program.An extensive data acquisition system was designed to provide test control and automated data collection. Several detail panels are shown in Figure 24.
Data acquisition and control program. Figure 25 . 92 .Data acquisition and control program. Main panel view. Compressor panel view.Figure 24 .
The compressor electrical power input is measured with a wattmeter. Further developments included doing real time data analysis which can be used to automate and control the test.Data acquisition and control program. The compressor voice coil voltage and current are measured as waveforms. Discrete Fourier transform analysis is used to calculate phase angles and to analyze the frequency content of the various waveforms. Considerable effort has been invested to use these waveforms to infer the piston motion 93 . etc. Pressure waveforms are acquired and stored for analysis. Pressure panel view. diodes. Initially the objective was to measure a series of DC voltages such as thermocouples. As the program developed.Figure 26 . communication with most of the test instruments was incorporated allowing the program to monitor and display test data. pressure. and controlled digitally with a feedback control algorithm by the PC. The data acquisition system has been in development for six years.
the results are yet inconclusive. the two coefficients that characterize the momentum balance in the porous medium can be determined (ref. However. These are the Darcy permeability. Appendix 3 has been included to summarize the development of the compressor model. is a measure of the viscous effects while the Forchheimer term is a measure of the inertial effects. By varying the mass flow rate of gas through the regenerator and measuring the pressure drop. it is not necessary to conduct a series of mass flow versus pressure drop measurements at various temperatures. This hypothesis is tested in Chapter 5 by comparing the calculated pressure ratios with the experimentally 94 . or pore scale.e. Both of these terms are due to microscale. The hypothesis of this research is that the Darcy and Forchheimer coefficients depend solely on the geometry of the porous media. Tests need to be performed on a compressor which has piston position indicators.from a kinematic model for the compressor pistons. Thus they can be measured by any convenient method. The first category consists of data measured under steady flow and essentially isothermal conditions for the purpose of determining friction factors. This would allow calculation of the compressor PV power output. The Darcy term. i. as discussed in previous chapters. and the Forchheimer inertia coefficient. which would allow the compressor to be decoupled from the rest of the system from an efficiency viewpoint. flow phenomenon which contributes to a macroscopic pressure gradient.2 Experimental Results Experimental data can be separated into two distinct categories. 4. K. cf. Figure 27 and Table 3).
2 Figure 27 . The data in Table 3 was calculated using a control volume model rather than a differential model.0 1. Direct pore scale simulations are required to test the hypothesis.4 0.measured pressure ratios during actual cryogenic operation. This does not provide conclusive proof of the hypothesis.0 400 Mesh 325 Mesh 60 Micron 45 Micron Foam Metal 400 Mesh 0.2 0.Steady Flow Pressure Drop 95 .6 0. and this work is suggested for future investigations. the data is flawed and leads to artificially low friction factors. This data is revisited in Chapter 5. The volume averaging method produces a consistent model that can be tested via the pore scale simulations. 1200 Pressure Drop [kPa] 1000 800 600 400 200 0 0.8 Flowrate [g/s] 1. As a result.
These load curves provide an overall picture of the system performance. However. Of prime importance is the ability of the cryocooler to perform net refrigeration at varying temperatures. there are references in unrelated applications that address the issue of frequency dependent permeability (66).5315 0.4296 Some of the previously calculated pressure ratios using steady flow data did not correlate with experimentally observed pressure ratios (17. The authors assume that this frequency dependence is in fact a result of increasing inertial effects.73 K [m2] 1. 17.5163 0. 19. This is measured by constructing load curves (ref.473E10 1. 20. Figure 28). and 21). A second set of data consists of measurements under actual operating conditions of the cryocooler.37 69.005E10 1. the pulse tube loss and the regenerator loss can be estimated. 18.61 71. From this data and the system level model. which would indicate the Forchheimer effect. These curves are also used to estimate the total parasitic load on the cooler.3917 0.20 69.) or that the Darcy and Forchheimer coefficients are not only dependent on geometry but possibly frequency and gas properties. It is the opinion of this writer that either their model was incomplete (erroneous simplifying assumptions. By comparing 96 .Table 3 – Porous Media Parameter Summary (64) Foam Metal 400 Mesh 325 Mesh 60 Micron ε [%] 61. etc. Indeed.060E10 1. presence of real gas effects.573E10 cf 0. these estimates are only as accurate as the model.
0 1. 3. 50 W 90 110 130 Cold Tip Temperature [K] Figure 28 . 60 W Foam Metal. 97 .5 1.0 MPa. 50 W 325 Mesh. Therefore the comparison among the regenerators is still valid. the 400 mesh regenerator was tested with a larger volume pulse tube leading to a slightly more efficient expander.1 MPa.4 MPa. This model accurately reflects the pulse tube geometry. and under the same operating conditions.0 0. 3. the 400 mesh regenerator is modeled in Sage. 3. 2.1 MPa. 55 W 45MPD. the effect of the regenerator can be isolated in a sense.Load Curves 150 As mentioned previously.5 0.4 MPa. 3. The 400 mesh regenerator outperforms all of the other regenerators even if it is tested with the smaller volume pulse tube. In Chapter 5. The temperature difference at noload for the 400 mesh regenerator tested with the larger volume pulse tube was only 23 K (61) compared to the nominally 15 K differences.0 70 400 Mesh.5 Cold Tip Applied Load [W] 2. 55 W 60MPD. 3.two load curves for the same cryocooler with only the regenerator changed.
The model will be shown to be driven by the pressures at the warm and cold end of the regenerator. The fact that the 400 mesh regenerator outperformed the other matrices is not immediately obvious from the pressure drop data since the 400 mesh has one of the highest friction factors.e. This parameter is influenced both by the difference in pressure amplitude as well as the pressure phase shift. it has a small friction factor.The performance tests provide a direct comparison between the regenerator matrices (63. The ratio of the pressure ratios is a normalized measure of the frictional losses occurring in the regenerator. The pressure phase shift as measured is primarily occurring in the regenerator even though the pulse tube is physically between the two points where the pressure is being measured. Figure 31. The pulse tube has minimal effect on the pressure because of its relatively short length and large diameter. The oscillatory pressure drop is the magnitude of the instantaneous pressure difference across the regenerator/pulse tube which is measured experimentally. Thus it is important to measure the effect that the regenerator has on the pressure wave. Figure 29. i.64). The 400 mesh regenerator performed the best under the conditions of fixed input power. i.e. 98 . This fact illustrates that the heat transfer efficiency is significantly better for 400 mesh than the other matrices. and Figure 30 illustrate that the pressure wave is attenuated and phase shifted by the regenerator and pulse tube expander. lowest noload temperature.
It cannot be concluded from this data that the friction factor for the 400 mesh is higher for the oscillating flow.0 MPa.Oscillatory Pressure Drop. the oscillatory pressure drops are significantly different.1 MPa. which performed the best. 50 W 325 Mesh 3. The foam metal exhibited a smaller pressure phase shift which can partly explain this smaller pressure drop. The pressure amplitudes and phase shift are found to accurately characterize the model. 55 W 45MPD 3.1 MPa. 50 W Figure 29 . increasing by up to 10 degrees with a 2% decrease in the 99 .4 Mpa. has the largest oscillatory pressure drop and the largest pressure phase shift. 55 W 400 Mesh 3. The pressure data for this regenerator is used to validate the system level model. P 1P2 [Kpa] 240 220 200 180 160 140 120 100 60MPD 3.Oscillatory Pressure Drop The 400 mesh regenerator. The 400 mesh regenerator is modeled in Chapter 5. The large phase shift is partly responsible for the magnitude of the oscillatory pressure drop. The phase shift is found to be very sensitive to the porosity. The 400 mesh and foam metal regenerators were similarly restrictive for steady flow. However.4 MPa. 60 W Foam Metal 3.
The porosity of the 400 mesh and the 325 mesh were almost the same. 60 W Foam Metal 3.porosity. The packing density of wire mesh screens has a large influence on the regenerator performance. 55 W 45MPD 3.4 MPa.1 MPa.4 Mpa. The percent open area is higher for 325 mesh screens (36% for 400 mesh versus 41. This helps to explain why the 400 mesh has higher pressure drop. P2) [Degrees] 24 22 20 18 16 14 12 10 8 60MPD 3.Pressure Wave Phase Angle 100 .0 MPa.) Percent open area is based on the wire diameter and spacing. 50 W Figure 30 . 55 W 400 Mesh 3.1 MPa.3% for 325 mesh. Phase Angle (P1 vs. 50 W 325 Mesh 3.
Completion of this test will provide a complete set of data for this type of regenerator. 50 W Figure 31 . The results show that the pressure drop was lower slightly higher than the 60 micron and significantly lower than 325 mesh. However.00 60MPD 45MPD Foam Metal 325 Mesh 400 Mesh 3. It was a bit of a surprise to see the 60 micron and 45 micron regenerators perform almost the same.P1 Pressure Ratio/P2 Pressure Ratio 1.03 1.Pressure Ratio Attenuation The remaining 30 micron perforated disk regenerator remains to be tested.04 1.05 1.08 1. Any abrasion or shearing 101 .4 Mpa. Testing is planned for the near future. the 325 mesh provided better cooling performance. 55 W 3.06 1. The expectation was that the smaller perforations would provide better heat transfer for similar pressure drop. 55 W 3.1 MPa.4 MPa.1 MPa. The EDM was found to be the only process which could cut the foam metal without closing the pores. 60 W 3.01 1.07 1.0 MPa. 50 W 3.02 1.09 1. The foam metal regenerator was retested by dicing it into 5 equal lengths using an electron discharge machining (EDM) saw.
the regenerator conduction is found to be the smallest loss in the regenerator. These fibers were then packed into glass sleeves in a variety of different manners.013”. even smaller than the total conduction heat flux in the gas with the addition of dispersion heat flux.0015” to further reduce conduction. Fibers with an average diameter of 19 microns were procured. This required pressing the glass felt to the desired porosity and then holding the felt into the tube. The complex phenomenon of the sintering process made this a frustrating process. Dicing the foam metal should not affect any characteristic of the regenerator other than the solid tortuosity. but is quite high away from the cut. the solid tortuosity approaches zero at the cut. The proposed Pyrex glass fiber regenerators did not turn out as expected.process produces a cut which completely closes the pores while EDM erodes the material away forming a clean cut. These rings were intended to create a gap between the foam metal plugs of approximately 0. This indicates that the poor performance was not due to the matrix conduction. In Chapter 5. Locally. The tubes were then furnaced. The first 102 . Reducing the solid conduction will only reduce the total regenerator loss by a few milliwatts. The best methodology was to chop the fibers to produce a fiber felt which could be packed uniformly and to desired porosity. Testing of the diced foam metal regenerator produced no measurable difference in performance. The spacer rings were cut with an OD equal to the tube ID and a thickness of 0. The diced sections were then precisioncleaned and installed into the regenerator using spacer rings made of laser cut 316 stainless steel shim stock. The furnace temperature needed to be high enough such that the fibers would soften and sinter but not flow or slump.
The second attempt was performed at a lower temperature. Since the fibers were in compression. Any annular gap will destroy the regenerator performance. Shrinkage was reduced to ~10% which is still too much. but still within the softening range of the fibers. and it was concluded that the fibers were not a viable alternative for a regenerator.) It became quite obvious that the fibers would shed very easily from the sintered plug. 103 . there was no testable regenerator.attempt produced a well sintered slug of glass fiber but the overall dimensions had shrunk by ~25% leaving an annular gap between the fibers and the tube. The sintered fiber samples from all three attempts were examined under a microscope to assess the sintering and fiber deformation (Figure 32. they should expand radially to fill the tube during shrinkage. The piston length was chosen such that the final fiber column would produce the desired porosity. The third attempt was performed at the same temperature but with the fibers in compression using a dead weight and a piston. At this point. The piston did not compress completely on this attempt.
Figure 32 – Sintered glass regenerator. 104 .
the volume averaged model developed in Chapter 2 will be shown to be a flexible. The solution is presented in terms of plots of important aspects of the solution. A system level model is needed to estimate the boundary conditions for the regenerator model. It is however a model of just the regenerator. The regenerator model is then used to obtain an accurate simulation of the flow and heat transfer in the regenerator. accurate model. most importantly the regenerator loss. The effect of compressibility is shown to account for a significant error in the measured friction factor. and temperature difference due to finite heat transfer coefficient. real gas properties. 105 . and this model is developed in this chapter. A last comparison shows that a friction factor based on oscillating flow data is essentially identical to a friction factor based on steady flow data if a compressible flow model is used. The regenerator loss is caused by several effects such as temperature oscillations due to finite matrix heat capacity. The accuracy of this model is considered by comparing it to other regenerator models and by calculating the entropy generation discrepancy as defined by the Sage model. Asymptotic models developed in Chapter 3 are used to investigate the magnitude of these individual effects on the regenerator loss and the mean temperature profile.CHAPTER 5 5. RESULTS AND DISCUSSION In this chapter.
Pressure data at the transfer line and before the 106 . a commercially available modeling tool called Sage was used. other parasitic loads. The different components are represented by icons which are then connected to allow flow and energy transmission. to compare the system level experimental data available from the laboratory pulse tube apparatus. these measurements were not possible at the cold end of the regenerator due to the integral design of the regenerator and pulse tube. heat transfer coefficient.007”. To this end.1 Sage System Level Modeling The regenerator model being developed is decoupled from the rest of the system via the particular boundary conditions. friction factor. The system performance is not completely known from just the regenerator solution though. If the mass flow boundary conditions are known. With the apparatus used. these mass flow rates would be measured experimentally at both ends of the regenerator via hot wire anemometers or a similar method. whether a Stirling piston or pulse tube.5. Thus. To evaluate the accuracy of the model. will produce a net energy flow out of the cold volume which must carry the regenerator loss. The expander. Sage is a graphical programming package. and any net refrigeration. the model can be exercised over the range of experimental data. Figure 33 illustrates the diagram of the lab cryocooler. there is a need for a system level model. then the regenerator solution can be obtained with the additional knowledge of the regenerator geometry. Ideally. and dispersion coefficient. accurate boundary conditions for mass flow rate need to be applied. The wall thickness of both tubes is only 0. After building the Sage model which contains all of the geometric aspects of the lab cooler.
This data is unavailable experimentally with the current compressor. The resulting porosity was found to be 0. The additional degree of freedom is the pressure phase difference. The porosity was adjusted until the exact phase shift. which was measured experimentally. either due to an inaccurate friction factor or incorrect porosity. To drive the model. the Sage optimizer can be used to adjust the compressor stroke and orifice setting until the two pressure ratios match. Figure 33 – Sage diagram of the laboratory pulse tube apparatus. was achieved. However. The measured porosity was 0.663 for 400 mesh. This porosity was then kept constant for the 0.5 and 1. 107 . Experimentation indicated that the friction in the regenerator was too low. This difference was found to be attributable to a density difference for the stainless steel alloy used to make the screens which results in a 2% density difference.orifice provide the primary method of correlating the Sage model. This phase shift in the pressure is due primarily to flow friction. The Sage model initially gave incorrect pressure phase shift predictions. the compressor piston stroke needs to be specified.0 W cases. most of which occurs in the regenerator. This is comparable to the differences in the porosity.692 calculated based on an assumed density and known mass.
0.0 W W pv/Q 37.0 W 44. Considering the cold heat exchanger control volume. HX/ Transfer Line Regenerator 76.884 W 1. This map is based on the 400 mesh regenerator operating at its coldest temperature of 76 K. in this case 2.9 K Reject HX Surge Volume Figure 34 .884 W 2.86 W 292.System level energy flow diagram.09 W. For each refrigeration load. The 400 mesh regenerator was operated with three refrigeration loads.comp 42. The critical aspect of the warm end heat transfer is that energy is conserved.206 W 0. At this temperature. 400 Mesh – 76 K. The surge volume accounts for a large portion of the total heat rejection which seems rather contrary.6 K Compressor Comp. 400 mesh.regen 0.57 W Q net Qcond.Sage provides useful system level details such as the energy flow map in Figure 34.54 W Qsv Qreject 0.09 W 2. 0. 0 W The additional experimental data which was used to correlate the Sage model was the measured net refrigeration.09 W 1. The Sage predicted refrigeration was not accurate 108 . 0. the pressure ratios and phase shifts were measured. is rejected at the warm end of the pulse tube. 76 K. the pulse tube enthalpy flow is consumed totally by the regenerator loss and the regenerator wall conduction loss. The model illustrates that the gross refrigeration.86 W Qrej.51 K Cold HX Pulse Tube 293.9/6. and 1. and with 50 W of electrical input power.233 W 1. the cooler produces zero net refrigeration.45 W 1. The critical energy balance for any cryocooler is based on a control volume containing the cold heat exchanger. 0.5.0 W.
0 W cases. In this table.44 3. the mass flow rates decrease with increasing cold tip temperature.5 W and 1.] & & Th [K] Tc [K] Charge Pressure [MPa] 0. 1. This is probably due to inaccuracies in the empirical pulse tube correlations which Sage uses. Table 4 summarizes the 400 mesh regenerator operating conditions.1 1.6 76.0 W measured. These parameters were then used to perform comparison calculations with the model proposed in Chapter 3.720 34 15.949 0. This factor was adjusted at the noload point and then it was used for the 0.4 292.5 W 0.6 88.0 W 0. These correlations were constructed for larger capacity pulse tubes.0 292.652 34 17.072 W predicted versus 1.7 100 3. The correlation was reasonably good.679 34 16.913 0.initially.4 292. so their accuracy in this size range is unknown.1 & mh [g/s] & mc [g/s] Frequency [Hz] φmh − φmc [deg.51 3. Sage provides a scale factor to adjust these pulse tube losses. according to the Sage model.904 0. This error is within experimental error and certainly within the range of unknown radiation load.1 The regenerator boundary parameters can then be extracted from the correlated model. This decrease in mass flow rate is accompanied by an increase in the mass 109 . Several phenomenons contribute to reduced pulse tube refrigeration such as mass streaming and boundary convection.0 W 0. Table 4 – 400 mesh baseline operating conditions 0. These effects are modeled in Sage as enhanced diffusion effects based on some theoretical and experimental correlations.
The full regenerator model is then discussed beginning with a presentation of the solution fields. and DEEM) with additional calculations to compare the effect of real gas properties. The regenerator loss is a result of conduction and dispersion in the gas.2. the results of the limiting case models presented in Chapter 3 are illustrated in terms of the net regenerator enthalpy flow rate and the mean temperature profile. 5. Figure 35 is a comparison of the calculated regenerator loss for the three models (CTM.flow phase angle. The model is then compared with the Sage and REGEN model solutions. Only in the limit of zero conduction or dispersion and zero net enthalpy flux is the regenerator loss zero. 5. Finally. and net enthalpy flow due to the mass flux and enthalpy being partially in phase.1 Net Enthalpy Flowrate . conduction in the matrix.The Perfect Regenerator The net energy flow rate in the regenerator represents a loss mechanism since this energy must be carried by the gross refrigeration effect of the expander whether it be a pulse tube or a Stirling displacer. Several postprocessing results such as net energy flows and net entropy generation rates are then presented. The ideal gas cases additionally assume constant properties evaluated at the 110 . the results of the analytic method of determining the permeability and inertia coefficient from Chapter 2 are presented.2 Model Comparison In this section. This is a result of the decreasing regenerator loss which is proportional to the phase difference between the pressure and mass flow waves. LTEM.
the real gas enthalpy oscillates partially in phase with the mass flux. This temperature difference leads to an increase in net enthalpy flow of nearly a 111 . The resulting regenerator loss is only 0. and matrix conduction. This has a significant impact on the viscosity which is strongly dependent on temperature over the range of interest. The only nonidealities included in this model are pressure drop. The Constant Temperature model. As a result. enthalpy depends only on temperature. as described in Chapter 3.8 W increase. there is a 0. As a result.) The Dual Energy Equation Model captures all of the nonidealities in the regenerator. So a good comparison of the increase in the regenerator loss due to finite thermal capacity is the ideal gas CTM and the LTEM.33 W. The Local Thermal Equilibrium model was written only for ideal gas. constant properties evaluated at the average of the end point temperatures were used. This increase in regenerator loss is due to additional net enthalpy flow due to the local temperature oscillation which is allowed in this model. For the case of ideal gas. For the ideal gas.mean temperature. The addition of real gas properties results in a 0. specifically the addition of finite temperature difference between the gas and matrix. which is constant for this model.86 W increase. The operating conditions were kept constant for all models. due mainly to conduction and dispersion. the enthalpy oscillates locally in phase with the temperature (and partially in phase with the mass flux. gas conduction and dispersion. This increase in the regenerator loss is due to the fact that the enthalpy for a real gas depends on both temperature and pressure. In this case. is the most idealized model.
5 3.0 Regenerator Loss [W] 2. 3.5 2.) This increase is due to a more adverse (smaller) phase angle between enthalpy and mass flux. This analysis does not explicitly illustrate the effect of pressure drop.0 0.0 1. The reason is made evident by comparing the CTM and LTEM results. as the cold temperature decreases below the 20 K point the problem of matrix heat capacity becomes a severely limiting factor.) 112 . Being able to decompose the total regenerator loss provides insight into the relative importance of the competing design challenges with regenerators. A factor that is present at any temperature is the inverse relationship between heat transfer effectiveness and pressure drop.5 0.watt (comparing the ideal gas LTEM and ideal gas DEEM. Pressure drop represents a loss of available energy for the pulse tube or expander piston to convert to refrigeration.0 Constant T Ideal Gas Constant T Real Gas LTE Ideal Gas Dual E Ideal Gas Dual E Real Gas Figure 35 – Comparison of regenerator loss calculated with limiting case models based on identical operating conditions (ideal gas assumes constant properties.5 1. Although there are no direct simulations provided in this report.
where the gradient is almost zero.2 The Mean Temperature Profile At quasisteady state. with the midpoint temperature 36 K higher. Model convergence is primarily limited by the convergence of the steady mean temperature profile. Figure 36 illustrates the effect of the different models on the mean temperature profile. with a slight. The real gas DEEM also shows a dramatic deviation especially at the cold end. The CTM profile is nearly linear. The ideal gas case uses a constant viscosity evaluated at the average temperature.2. deviation at the midpoint. The LTEM profile is higher at the midpoint by 9 K and the ideal gas DEEM profile higher by 17 K. 3 K. The mean temperature profiles for the real gas CTM and DEEM represent the extremes of all the profiles. The shape of the net diffusion and enthalpy fields is strongly affected by the mean temperature profile. The likely cause of this deviation is the finite matrix heat capacity in the DEEM. the net diffusion and enthalpy fields are in a balance such that the total energy flow through the regenerator is constant. An interesting comparison exists between the two real gas cases. 113 . A dramatically different profile occurs with the addition of real gas properties.5. This difference between the ideal gas and real gas cases is partly due to the fact that the viscosity is temperature dependent has a significant variation over the range of temperatures. The CTM assumes that the heat capacity is essentially infinite such that no temperature fluctuations occur.
instantaneous solutions are shown in Figure 37. 5. mass flux. This illustrates the relative difference in the spatial and temporal scales for the density. pressure. From these solutions. and entropy can be calculated. velocity. gas total energy. In the upper right. the density field is shown versus time (horizontal axis) and position (family of curves with red and blue corresponding to hot and cold ends respectively. and matrix temperature at periodic quasisteady state. enthalpy. the density maximum and minimum profiles are plotted versus position in the regenerator.350 300 Mean Temperature [K] 250 200 150 100 50 0 0 Constant T Ideal Gas LTE Ideal Gas Dual E Real Gas Constant T Real Gas Dual E Ideal Gas Linear 10 20 30 40 Position [mm] 50 60 70 80 Figure 36 – Comparison of mean temperature profiles calculated with the different models based on identical operating conditions. The local. The temporal scale (max versus min) is small 114 .2.) Below this figure.3 Baseline Regenerator Solutions The regenerator model solves for density.
the temperature 115 .015 0.015 0.025 3 2.005 0.04 0.02 0.) In this problem. Density [kg/m ] vs. Both of these phenomenon are measured in the apparatus.) The pressure wave is attenuated and shifted in phase as it passes through the regenerator.015 0.4 Pressure [MPa] vs.01 0.005 0.05 0 3. Cycle Time [radians] 0. Pressure is totally driven by the mass flow boundary conditions (or vise versa with pressure boundary conditions. Position [m] 15 0 10 5 0 Mass Flow Rate [g/s] vs.01 0.015 0.) The velocity is given similar treatment.01 0. Position [m] rhomax rhomean rhomin 5 0 5 umax umean umin 0. The middle plots are max and min of density and velocity. Cycle Time [radians] 5 15 0 10 5 0 0.005 0.02 0.02 0.025 Velocity [m/s] vs. The velocity at the cold end is much smaller than the velocity at the warm end due to the large density difference.03 0.01 0.2 0 0.02 0.03 0.01 0.04 0.5 3. The temperature field is calculated from the density and total energy using the appropriate caloric equation of state (ideal or real gas.8 0 0. The mass flow rates (bottom left) are comparable in magnitude but with a noticeable phase shift. Cycle Time [radians] 5 0. Cycle Time [radians] 3 Velocity [m/s] vs.compares to the spatial scale (cold end versus warm end.01 0. Pressure and mass flow phase shifts are apparent.5 0 0.02 0.02 0.025 3 Density [kg/m ] vs.005 0.025 Figure 37 – Solutions plotted versus time and position.05 0.
] Figure 38 – Boundary solutions for the gas and matrix temperatures and enthalpy flow rates.02 0. The gas temperature leads the matrix temperature. the enthalpy flux at the warm and cold end have extremely large amplitudes (1500 W and 116 . The flow reversal is also associated with the beginning of the expansion process which causes a reduction in the gas temperature.015 0.025 Enthalpy flux [W] Temperature [K] 1000 500 0 500 1000 0 0.015 0.005 0.2995 Warm end enthalpy flux phiHh=2.025 Cold end enthalpy flux phiHc=22.01 0.5 292 291. Matrix=b. In the figure.005 0.3045 79 78 77 76 75 74 0 0.015 0.005 0.012 293. Gas=r. Tc=g phiTc=115.025 Cycle time [sec.01 0.025 Cycle time [sec.] Enthalpy flux [W] Temperature [K] 200 100 0 100 200 0 0. Warm end temperatures .is not specified on the boundaries.015 0. Gas=r.01 0. The enthalpy is also calculated from the total energy and density.02 0.5 293 292. The gas temperature has a distortion which occurs at the flow reversal.5 0 0. as would be expected.02 0. Matrix=b.02 0. Th=g phiTh=75. The predicted gas and matrix temperature waves at the warm and cold ends are shown in Figure 38.01 0.3202 Cold end temperatures .005 0. This is due to the reduction in the heat transfer coefficient at low velocities.
300 W) compared to the DC component which is less than 2 W. The temperature difference (Figure 40) is a critical aspect to the regenerator problem.01 0. This temperature difference is responsible for a large percentage of the regenerator loss. 117 . It is this temperature profile which converges the solution to periodic quasisteady state. The mean temperature profile for this baseline case is shown in Figure 39 in comparison to a linear profile.02 0. 280 260 240 Temperature [K] 220 200 180 160 140 120 100 80 0 0.03 0. It is this DC component which is critical to be able to predict accurately.04 0.05 Position [m] Figure 39 – Cycleaveraged gas temperature (red) versus a linear profile (blue). The temperature difference is noticeably smaller at the cold end due to considerably larger mean Reynolds number at the cold end as shown in Figure 41.
The net enthalpy and heat fluxes due to conduction and dispersion are 118 .5 0.05 0.4 First Law Results Cycle averaging of the solutions during postprocessing provides the most important aspects of the solution such as the net mass flow rate and enthalpy flow rate. essentially numerical noise. the net mass flow rate is very small.03 Position [m] 0.005 0.01 0.2 0.] 0.02 Cycle Time [sec. 62 60 58 Mean Reynolds Number 56 54 52 50 48 46 44 42 0 0.3 0.1 0 0.0.03 Figure 40 – Temperature difference versus position (multiple lines) and time.1 0.2.04 0. Figure 42 illustrates that at steady state.02 0.4 0.01 0.3 0.2 0.4 Temperature Difference.06 Figure 41 – Mean Reynolds Number 5.5 0 0. TTm [K] 0.015 0.025 0.
5 0 0.05 0.5 3 x 10 6 3. 0.04 0.5 2 2. 1600 Cycleaveraged energy flows [mW] 1400 Total H 1200 1000 800 600 400 200 0 0 0. 119 .5 1 1. The critical point is that the total energy flux through the regenerator is constant.not zero and are shown in Figure 43.01 0.02 0.05 Qg Qm 0.06 Figure 42 – Cycleaveraged mass flow rate.5 0 Cycleaveraged mass flow [g/s] 0. Regenerator total energy flow is constant along regenerator.03 Position 0.03 Position 0.06 Figure 43 – Cycleaveraged energy flows.04 0.02 0.01 0.
12 0.03 kg entropy generation rate [W/mK] 0.025 0.5 Second Law Results Entropy generation is calculated in postprocessing as well. 120 . The matrix conduction only shows slight oscillation. plotted versus cycle time. Since the dispersion is proportional to the Peclet number.06 0. contributing almost 80% of the total available energy loss in this case. Results for 400 mesh baseline case. The entropy generation due to the gas conduction and dispersion is characteristically different than the entropy generation due to the matrix conduction. it is oscillatory as can be seen.04 0.005 0.01 0. Multiple curves represent different locations in the regenerator.015 0.02 0. The entropy generation due to interfacial heat transfer is the second largest contributor to available energy loss.02 0 0 0.025 0.2. The entropy generation due to viscous and inertial losses is the dominant source. dP entropy generation rate [W/mK] 5 4 3 2 1 0 1 0 0.01 0.1 0.03 Figure 44 – Entropy generation due to viscous and inertial losses (top).5.02 Cycle Time [sec] 0. and entropy generation due to conduction and dispersion in the gas (bottom).08 0. but it is the dominant contributor to regenerator net enthalpy flow.005 0. The individual sources of entropy generation are illustrated in Figure 44 and Figure 45 as functions of position and time.015 0.
8 0.02 Cycle Time [sec] 0.2 0. 121 .06 Figure 46 – Cycleaveraged volumetric entropy generation rates plotted versus position in the regenerator.01 0.2 0 0 0.03 h entropy generation rate [W/mK] 0.02 0.01 0.04 0.025 0. Results for 400 mesh baseline case.km entropy generation rate [W/mK] 0. Results for 400 mesh baseline case.6 1.05 0.4 0.6 0.02 Total Viscous and inertial Gas conduction and dispersion Matrix conduction Interfacial convection 0.005 0 0 0.6 0.015 0.02 0.4 0.7 0.015 0.015 0.01 0. and entropy generation due to interfacial convection (bottom).3 0. 2 1.005 0.03 Position [m] 0.025 0.03 Figure 45 – Entropy generation due to matrix conduction (top).2 1 0.01 0.1 0 0 0.8 Entropy generation rate [W/mK] 1.005 0.4 1.5 0.
The total entropy generation has a surprising shape with a minimum in the interior of the regenerator. Figure 47 is a comparison of Sage.Figure 46 provides an interesting picture of the entropy generation mechanisms. with units W/K. the net energy flow profiles can be constructed. but with a simplified model. The profiles plotted for REGEN are after 100. The Sage model apparently forces the matrix conduction flux to zero at the ends giving the profiles an abrupt transition at the ends. This produces a sharp corresponding increase in the enthalpy flux at the ends.2.) It models only the regenerator.2. REGEN3. The solution is still not satisfactorily converged as can be seen by the total energy flow profile which has a 10% variation along the regenerator.6 Sage and REGEN Comparisons Sage provides the solution grid for the regenerator.2 was developed at the National Institute of Standards and Technologies (NIST. but the interfacial heat transfer component increases sharply at the cold end due to the 1/T nature. 122 . 5. The viscous ad inertial entropy generation decreases to almost nothing at the cold end. The REGEN seems to grossly over predict the total conduction which is counterintuitive since the model wholly neglects dispersion. and profiles from a third model known as REGEN3. From this. The Sage profiles correlate roughly with the Georgia Tech model which is the most visibly accurate. the Georgia Tech model proposed in this thesis. The area under these curves represents the net total entropy generation.000 cycles of simulation time which took several hours. All three models were run with identical inputs.
6 1. Sage predicts a moderately higher regenerator loss. The GT model does not make this assumption. as mentioned previously.2 0 Net energy flow [W] Sage h Sage q Sage h+q GT h+q GT q GT h REGEN h+q REGEN h REGEN q 0 0. REGEN.03 0.4 1. 123 . assumes that the pressure gradient is negligible. However.01 0.2 1. This may be the cause of the REGEN model not converging to a constant energy flow rate as seen in the figure. The regenerator loss is seen to decrease with increasing cold end temperature.02 0.04 0.05 Position [m] Figure 47 – Comparison of Sagepredicted regenerator loss versus GT model. this error is small in comparison to the uncertainty in the pulse tube net enthalpy flow.4 0.2 1 0. The REGEN model as a result of this assumption produces a nonconservative scheme while the GT model solves a fully conservative set of equations.8 1. This source of error is approximately 8%. Table 5 is a summary of the comparison calculations with Sage. The effect of this error directly translates into an error in the net refrigeration prediction.8 0.6 0.
and flow friction. This discrepancy is then a measure of the accuracy of the numerical scheme.Table 5 – Sage Comparison Summary Net refrigeration Regenerator Loss Viscous and inertial lost power Interfacial heat transfer lost power Gas conduction and dispersion lost power Matrix conduction lost power Total gas and matrix conduction and disp.5 1. The sum of the internal entropy generation rates is then compared to the entropy generation rate determined by an external calculation.84 15 3.) This discrepancy is calculated.618 0.481% 1.6 62.3 16.5%.747 19.60 14.0 1.02 1. convection.36 0.087 0. due to numerical errors. integrating the entropy flux at the boundaries.477 0. The Georgia Tech model demonstrates better available energy discrepancies (0.85 0.441 16.4 0. by summing the internal sources of entropy generation due to conduction.22 43.22 42.30 1.0 1.34 0.671 14.2 1.1 17.18 2.15 0. Total lost power Lost power based on external calculation Lost power discrepancy Warm end pressure ratio Cold end pressure ratio Warm end pressure phase angle Cold end pressure phase angle 0. However.2 0.105 0.713 18.22 43.744 14.553 17. dispersion.270% 1.1 SAGE 0.6 The available energy predictions correlate very well between the two models. in both models.e.27 62.54 14.2 0.30 1.11 63.5 1.8 GT 0.73 15.87 0.1 62.1 1.9 19.2 19.3 3.9 63.2 17. i.2 2.469 16.5 1.7 63.734% 1.012% 1.129 0.0 0.24 44.33 1. these two methods of calculating the entropy generation rate are identical.029% 1.17 0.03% versus 0.4 16.381 0.0 1.4 0.24 45.31 1.33 1.0 1.582 17. these two methods will produce different results.21 42. 124 .060% 1.4 0.31 1.2 0. Ideally.
These predictions made using a compressible model and are shown to be comparable to friction factors measured in oscillatory flow. The compressible model is the most realistic since this model predicts the increase in pore velocity as the pressure decreases in a steady flow test. leads to a density decrease for a gas flow. 5.2. The pressure waves are sensitive to accurate modeling of the friction factor in the regenerator. The REGEN model does not provide this data. The same predictions were made with the incompressible model. Ideally these pressure waves should match experimental data in both amplitude and phase.The GT model is an order of magnitude more accurate than the Sage model. which occurs in incompressible flow as well. The results of this methodology are presented in Table 6 . Pressure ratios and phase angles correlate well with each other as well as with experimental data. The mass flux at the ends are unknown experimentally. The decrease in pressure. The following section provides a discussion of the friction factor prediction from experimental data.7 Permeability and Forchheimer Coefficient Predictions Chapter 2 introduced a method to predict the permeability and Forchheimer inertial coefficient using three analytical solutions to the continuity and momentum equations. The pressure waves which develop are an output from the model. The GT model requires that the mass flux be specified at both ends. The compressible inertial term is consistently 30% higher for all of the matrices tested while the permeability is 125 . so a comparison cannot be made. so the results of the Sage system level model are heavily relied upon for this information.Table 8.
21E11 126 cf 0.372 0. The final comparison assesses the importance of the advective acceleration term in the momentum equation.80E11 4.308 0. The permeabilities should be identical since this parameter characterizes the low Reynolds number regime of the flow where compressibility is much less important. This indicates that the inclusion of the advective acceleration term.69E11 3.407 0.almost identical.376 0.80E11 4. at least for steady flow.282 0.445 0.21E11 cf 0.257 0.260 0.69E11 3.407 0.53E11 2. is not necessary.69E11 3.376 0.53E11 2.80E11 4.91E11 2.91E11 2.21E11 cf 0.259 .445 0.179 Table 8 – Permeability and Forchheimer Coefficient (neglecting advective acceleration) 400 Mesh 325 Mesh Foam Metal 60 Micron 45 Micron K 2.259 Table 7 – Permeability and Forchheimer Coefficient (incompressible model) 400 Mesh 325 Mesh Foam Metal 60 Micron 45 Micron K 2.372 0.53E11 2. Table 6 – Permeability and Forchheimer Coefficient (compressible model) 400 Mesh 325 Mesh Foam Metal 60 Micron 45 Micron K 2.91E11 2. Both the permeability and inertial coefficient are identical for all of the matrices tested.
It is obvious from this comparison that the flow oscillations may not be nearly as important as the compressibility. The friction factor. use the permeability length scale which is measured experimentally. There is only a slight deviation in the two friction factors at high Reynolds number. heat transfer coefficient. It is suspected that the friction factor depends only on the microscale geometry while the flow friction depends on the actual macroscopic flow field and fluid properties. The length scale used in Sage is a hydraulic diameter based on an approximate geometrical definition which depends on the porosity and the wire diameter in the case of screens.A comparison of the friction factor predicted from measurements using 400 mesh yield close agreement with the friction factor used in Sage as shown in Figure 48. The compressible model provides more motivation that this hypothesis is true. The friction factor and Reynolds number are defined using different length scales in Sage. This is interesting in light of the methodology in which the Sage friction factors were measured. 127 . and dispersion coefficient for screens and felts were measured using an oscillatory flow test rig. as defined in Chapter 3. in hopes of gaining some improved predictions (19). The friction factor and Reynolds number.
1E+04 f_GT 1E+03 f_sage*K1/ f Sage 2/(2*dh) h 2d Friction Factor 1E+02 K 1/ 2 1E+01 1E+00 1E01 1E03 1E02 1E01 1E+00 1E+01 Reynolds Number 1E+02 1E+03 Figure 48 – Friction factor comparison for 400 mesh screens 128 .
CHAPTER 6 6. CONCLUSIONS AND RECOMMENDATIONS
6.1 Conclusions The governing equations for the flow and heat transfer which hold for a point in space and time were volumeaveraged, producing a set of continuum equations for the porous regenerator. These equations contained several source terms which represent the viscous and inertial friction, gastomatrix heat transfer, and thermal dispersion. Several additional terms were neglected based on scaling arguments. The simplified equations describe unsteady compressible flow and heat transfer in a porous medium. The dimensionless regenerator model was found to depend on two leading order dimensionless parameters and two second order parameters. These parameters indicate that the second order diffusion terms are small and should not be used as justification for requiring additional boundary conditions. The resulting problem requires only two mass flow boundary conditions and an appropriate method for finding the quasisteady initial condition. The oscillatory regenerator flow solution was found to depend strongly on the mean temperature profile. Finding this mean profile with a semianalytical technique allows the solution to be advanced in time much faster than with direct simulation. This provides a huge computational advantage.
129
To validate the numerical method, an exact solution was found which required additional source terms to be added to the governing equations. The numerical solution agreed with the exact solution within O(106) normalized errors. A method for experimentally testing a cryocooler regenerator has been developed. Pressure waves are measured at the warm ends of the regenerator/pulse tube. The net refrigeration is measured with an applied heat load over a range of refrigeration temperatures. A data acquisition system provides the ability to make high speed
measurements and for test control. Six regenerators were tested with a nearidentical apparatus: 400 mesh, 325 mesh, 60 and 45 micron perforated disks, foam metal, and diced foam metal. An additional 30 micron perforated disk regenerator is planned for testing. An optional glass fiber regenerator was investigated but abandoned due to manufacturability and reliability issues. System level modeling gives detailed insight into the regenerator operation. Matching the measured pressure amplitudes and phases in addition to the measured net refrigeration with the system level model allows the regenerator boundary conditions to be extracted. The steady, compressible flow friction factor measurement using an analytical solution matches with that measured in oscillating flow. The higher friction factor is apparently due to the velocity acceleration which occurs with a gas flowing through a porous media at high Reynolds number; it does not appear to depend on the flow unsteadiness which several investigators have claimed.
130
6.2 Contributions Beginning with the most significant contributions first, the application of the volume averaging technique to the regenerator problem is the most noteworthy. This approach removes the ambiguity which has existed in the regenerator literature for decades. While the simplifications which were made reduce the problem to a similar formulation found in the literature, the framework for future refinements and investigations are solidly laid. The numerical solution of the one dimensional model using a conservative scheme is a significant contribution. This method is more accurate and reliable than any existing method in the open literature due to the direct solution of a conservative system of equations with a conservative numerical scheme. The lack of assumptions make this model a test case for other models such as REGEN and Sage. Sage is a valuable system level model, but the regenerator poses the largest source of irreversibility in the entire cryocooler system. As a result, any inaccuracies in the regenerator can manifest
themselves as large net refrigeration errors. Having an accurate prediction tool allows the Sage model to be verified. The test apparatus and accompanying data acquisition system represents six years of effort. The diagnostic capability of the system allows for real time data analysis. Optimizing the cryocooler operation is assisted by data averaging, filtering, and trending. Troubleshooting is much improved. With the addition of the Sage system level modeling tool, experimentation and simulation can occur simultaneously giving the researcher the ability to immediately understand the effect of a change in the operating conditions. 131
Investigators have long supposed that the friction factor and heat transfer correlations from steady measurements deviate for oscillating flow. The simulation consisted of constructing a solid model for a representative. or even unsteady.6. direct. which is a commercially available computational fluid dynamics software suite. These simulations were performed for steady flow only. in this case wire mesh. Oscillating flow simulations are eventually a possibility. periodic cell of the porous media. porescale numerical simulations could be made to evaluate the surface integrals and assess the effect of the flow oscillations. 132 . but none have been seen in the literature for oscillating. The strict assumptions under which these simplifications are allowed may not be exactly true in the case of fully compressible flow with the large temperature and density variations which are observed in the regenerator. A steady flow through the mesh was then simulated. A demonstration calculation was performed using Fluent. The form of the surface integrals representing the friction factor and heat transfer correlations reduced to familiar forms. flow. Figure 49 and Figure 50 show some preliminary results of this type of simulation.3 Future Work The volumeaveraged equations were simplified in a manner which reduced the macroscopic equations to a form which could be solved. These types of simulations have been made for steady flow. A possible discrepancy which the simplifications produce is due to the flow reversals which are occurring in the regenerator. While it is more likely that the friction factor depends only on the microscale geometry.
Demonstration calculation for flow through wire mesh screens. importantly. 133 More . Visualization of pathlines. The surface and volume integrals representing the friction factor.Figure 49 . Visualization of surface pressure Figure 50 – Demonstration calculation for flow through wire mesh screens. heat transfer coefficient and dispersion coefficient can be evaluated knowing the pore scale flow. oscillating flow and heat transfer can be modeled.
Proper system design should alleviate these problems. For “significantly high” frequencies. the friction factor could be assumed. In light of the Reynold’s Analogy. The radial velocity would obviously be zero at the wall. stagnation zones due to flow asymmetry. Several improvements to the test apparatus will represent a significant augmentation in ability to research more advanced concepts. For most random media. Kornhauser and Smith made this realization and adopted a complex Nusselt number to model this phenomenon. to be constant in the radial direction. apparently. These deviations are assumed to dramatically deteriorate the regenerator performance. the heat transfer coefficient does depend on the flow. Small aspect ratio regenerators have lead to thermal instability problems producing. initially. The jet should be dispersed rapidly. the friction factor may also exhibit a similar effect at some frequency limit. For such a model.Admittedly. but it is not possible at this point to model such a phenomenon. The flow passage geometry leading into the regenerator will typically create a jet. Improved instrumentation 134 . The macroscopic regenerator flow is commonly assumed to be onedimensional although several investigators have noticed definite deviations. but the axial velocity does not have an obvious boundary condition at the wall (if one is required at all. but a model capable of capturing these effects would be useful. an additional radial coordinate. especially the frequency.) This needs to be studied carefully. Such a model would require. Newton’s Law of Cooling fails to hold due to significant phase shifting between the bulk temperature and the interfacial temperature. as a minimum. the wall boundary conditions would need to be established.
With this information. These piezoelectric pressure transducers which do not require the repeated calibration and are less affected by temperature can be used in the cryogenic regions of the cooler allowing for direct measurement of the pressure. This gives more flexibility to investigate new concepts without redesigning the entire system. inertance tubes. etc. This is an important piece of data not currently available. instrumentation blocks. heat exchangers. Anemometers can be designed with low helium leakage and can be operated in cryogenic regions as well providing mass flow and temperature measurements. Compressor LVDTs are a necessity for continued testing. It is possible to make modular components such as regenerators. A larger vacuum dewar will allow for calorimetric heat exchanger measurements to be made which will provide better system level understanding. 135 . pulse tubes. This is increasingly important for multistage cryocoolers. These LVDTs are used to measure the piston position as a function of time. surge volumes.such as piezoelectric pressure transducers will greatly affect the reliability of the test results. the mechanical input PV power can be calculated.
can contain portions of different phases of material. A1. β such as the system illustrated in Figure 4. an arbitrarily given volume. Then the total volume. the averaging theorems and principles are discussed. a porous media with a single fluid phase. The generalized governing equations of mass conservation. and a single solid phase.1 Definitions Within a porous media. x.1. The only assumptions that are made initially are that the two phases do not react chemically with eachother and that the noslip assumption at the fluidsolid interface is valid. (A11) A property in the αphase. in the porous media is written as a function of space and time as 136 . at a given position.1 Volume Averaging Theory The method of volume averaging is applied to the governing equations for flow in a porous media. At the outset. Consider for example. V. α. V(t) is given as V (t ) = Vα (t ) + Vβ (t ). momentum balance and energy conservation are developed for a single fluid phase flowing and interacting with a single stationary solid (matrix) phase. ωα.APPENDIX 1 – DERIVATION OF THE VOLUMEAVERAGED GOVERNING EQUATIONS A1.
α V (A13) The intrinsic phase average is the volumetric average of a property over the phase volume. 137 . it is necessary to note a few theorems that relate averages of derivatives to derivatives of averages. 1 1 1 ωα dV = ωα dV + ωα dV = = Vα V Vα Vα Vα V 1β2 3 4 4 =0 ωα α ∫ ∫ ∫ ∫ω Vα α dV . V (A15) the phase average and intrinsic phase average can be related by ωα = ε α ωα . α (A16) To transform the governing equations of mass. Explicitly. and energy for the porous media. ωα ( x β . 68).ωα ( xα . These theorems are given without proof. (A14) By defining the volume fraction or porosity of the βphase. The reader is referred to the literature for proofs of these theorems (67. as εα = Vα . t ) = 0. the phase average is ωα = 1 V ∫ ω dV . i. εα.e. where the subscript on the position vector indicates the phase at that position. t ) = ωα . momentum. (A12) The phase average of a property is defined as the volumetric average over the total volume.
1.1.3 Spatial Averaging Theorem The volume average of a spatial derivative is related to the spatial derivative of an average by the Spatial Averaging Theorem given by ∇ωα = ∇ ωα + 1 V ∫ nα ωα dS . wαβ is the velocity of the αβ interfacial surface in V. and nα is the unit normal vector on the αβ interface pointing into the βphase.A1. If the noslip assumption is valid. then vα Aαβ = vβ Aαβ ⇒ wαβ = 0 (A18) This result requires the surface integral in Equation (A27) to be zero. (A110) Aαβ This transformation generates an additional term represented by the surface integral even for a noslip porous media. This additional term represents a macroscopic contribution caused be the microscopic interaction of the two phases. 138 . ∂t ∂t (A19) A1.2 Transport Theorem The volume average of a temporal derivative is related to the temporal derivative of an average by the Transport Theorem given by 1 ∂ωα ∂ = ωα − ∂t ∂t V ∫ω Aαβ α wαβ ⋅ nα dS (A17) where Aαβ is the area of the αβ interface in V. and this leads to a simplified transport theorem valid for a noslip porous media given by ∂ωα ∂ = ωα .
Note that ∇ωα = ∇ ε α ωα = ε α ∇ ωα = ε α ∇ ωα ( α 1 )+ V ∫ n ω dS α α Aαβ α + ωα + ωα α ∇ε α + ∇ε α α α ∫ 1 + n V∫ 1 V Aαβ Aαβ nα ωα ( α ˆ + ωα dS 1 V ) (A111) α ωα dS + α ∫ n ωˆ dS α α Aαβ ˆ where ωα is the deviation of ωα from its intrinsic phase average defined by ˆ ωα = ωα − ωα α . (A112) The intrinsic average that appears in the first surface integral is essentially constant over the surface (42). As a result. the intrinsic average can be removed from the surface integral giving ∇ωα = ε α ∇ ωα α + ωα α ∇ε α + ωα α 1 1 nα dS + V Aαβ V 1 24 4 3 −∇εα ∫ ∫ ˆ nα ωα dS Aαβ (A113) = ε α ∇ ωα α + 1 V ∫ ∫ ˆ nα ωα dS Aαβ This leads to the Modified Averaging Theorem given by ∇ωα = ε α ∇ ωα α + 1 V ˆ nα ωα dS (A114) Aαβ This is referred to as the scale decomposition since the parameter is decomposed into a macroscopic and microscopic component. Note that several consequences arise from this 139 .1.A1.4 Modified Averaging Theorem Equation (A210) can be modified by application of Equation (A26).
We will also neglect gradients in the porosity.2 Application to the Governing Equations The preceding volume averaging theorems can now be applied to the governing equations. we will assume that the noslip assumption is valid at the fluid solid interface. A1. A1. Also note that ˆ ˆ ωα = ωα α = 0. (A115) These principles and theorems can now be applied to the governing equations to develop a generalized set of averaged governing equations. Application of the Transport Theorem (A29) to the first term in Equation (A216) yields ∂ ∂ρα = ρα . ∂t (A116) In developing this volumeaveraged equation. the deviation can be nonzero. The result is the production of various terms representing the microscale influence on the macroscopic flow. in the αphase.1 VolumeAveraged Conservation of Mass Equation The phase average of Equation (213) is r ∂ρα + ∇ ⋅ (ρα uα ) = 0.term. this term is identically zero by Equation (A22). In doing so. the continuum flow which exists at the microscopic level is transformed into a continuum flow at the macroscopic level. but in the βphase.2. ∂t ∂t Definition (A26) can be used to obtain (A117) 140 . First.
(A120) Definitions (A212) and (A215) can be used to obtain r ∇ ⋅ (ρα uα ) = ε α ∇ ⋅ ρα ( α r uα α )+ ∇ ⋅ ρˆ urˆ α α . ∂t ∂t (A118) Application of the Averaging Theorem (A210) to the second term in Equation (A216) yields r r 1 ∇ ⋅ (ρα uα ) = ∇ ⋅ ρα uα + V ∫ r nα ⋅ ρα uα dS . (A119) Aαβ The surface integral in the last equation evaluates to zero due to the no slip assumption reducing the equation to r r ∇ ⋅ (ρα uα ) = ∇ ⋅ ρα uα . The momentum equation can be averaged in the same manner. 2 3 εα ∂4 3 t 1 24 144 44 14 244 4 3 convection accumulation mass dispersion ( ) (A122) This volumeaveraged form of the continuity equation contains an additional term which was not in the point equations. This is a generalized continuity equation for 3D flow in a porous media. (A121) Thus. 141 . the volumeaveraged continuity equation becomes r ∂ 1 α α r α ˆ ˆ ρα + ∇ ⋅ ρα uα + ∇ ⋅ ρα uα = 0.∂ρα ∂ α = εα ρα . This term represents the dispersive mass transport.
Application of the Transport Theorem (A210) to the first term in Equation (A223) yields r r ∂ (ρα uα ) = ∂ ρα uα . (A125) Substitution of Equation (A225) into Equation (A224) yields r ∂ (ρα uα ) = ε α ∂ ρα ∂t ∂t ( α r uα α )+ ∂∂t ˆ ˆ ρα uα .A1.2 VolumeAveraged Balance of Momentum Equation The phase average of Equation (214) is r r r r r ∂ (ρα uα ) + ∇ ⋅ (ρα uα uα ) + ∇pα − 1 ∇(µα ∇ ⋅ uα ) − ∇ ⋅ (µα ∇uα ) = 0.2. ∂t ∂t (A124) Definitions (A212) and (A215) can be used to obtain ρα uα = ε α ρα r α r uα α r ˆ ˆ + ρα uα . ∂t 3 (A123) As with the continuity equation. r r Definitions (A212) and (A215) can be used to expand ρα uα uα to the form (A127) ρα uα uα = ε α ρα r r α r uα α r uα α + ρα α r r r ˆ ˆ uα uα + uα α α ˆ ˆ ρα uα r r r ˆ ˆ + ρα uα uα Thus Equation (A227) becomes r r ˆ ˆ ˆ + ρα uα uα . we will assume that the noslip assumption is valid and that gradients of the porosity are negligible. r (A126) The Averaging Theorem (A210) can be applied to the second term in Equation (A223) to obtain r r r r ∇ ⋅ (ρα uα uα ) = ∇ ⋅ ρα uα uα . (A128) 142 .
r r ∇ ⋅ (ρα uα uα ) = ε α ∇ ⋅ ρα ( α r uα α r uα α )+ ∇ ⋅ ( ρ α α r ˆ ˆ + ∇ ⋅ ρα uα ( r r r ˆ ˆ uα uα + ∇ ⋅ uα α α ) ( ρˆ urˆ ) r r r ˆ ˆ ˆ u )+ ∇ ⋅ ρ u u . 1 V α 1 V ∫ 1 r ˆ nα uα dS + V Aαβ ∫ µα nα ⋅ ∇uα dS . α α α α α α (A129) The Modified Averaging Theorem (A214) can be applied to the third term in Equation (A223) to produce ∇pα = ε α ∇ pα α + 1 V ∫ ˆ nα pα dS . r (A132) Aαβ ∫ r ˆ nα uα dS . and the last term in Equation (A223) can be written as r r 1 ∇ ⋅ (µα ∇uα ) = ∇ ⋅ (µα ∇uα ) + V ∫ + µα nα ⋅ ∇uα dS . (A130) Aαβ Now proceed to expand the viscous terms in Equation (A223). The viscosity can be considered to be constant within the averaging volume. r (A131) Aαβ Applying the modified averaging theorem again produces r r ∇ ⋅ (µα ∇uα ) = ∇ ⋅ µα ε α ∇ uα Again the surface integral. evaluates to zero by the noslip assumption Aαβ which simplifies the last equation to r r ∇ ⋅ (µα ∇uα ) = ε α ∇ ⋅ µα ∇ uα ( α 1 )+ V ∫ µ n Aαβ α α r ⋅ ∇uα dS (A133) which can be expanded to 143 .
(A136) Aαβ Substitution of Equations (A226). (A230). (A229). r (A134) Aαβ The first surface integral also evaluates to zero if the porosity is constant. three surface integrals were combined and the whole equation was divided by porosity. The final form of the viscous term is r r ∇ ⋅ (µα ∇uα ) = ε α ∇ ⋅ µα ∇ uα ( α 1 )+ V ∫ µ n Aαβ α α r ˆ ⋅ ∇uα dS .r r ∇ ⋅ (µα ∇uα ) = ε α ∇ ⋅ µα ∇ uα ( α ) α 1 + V ∫ r 1 µα nα ⋅ ∇ uα dS + V Aαβ ∫ ˆ µα nα ⋅ ∇uα dS . and (A234) into Equation (A223) yields ∂ α r α α r α r α ρα uα + ∇ ⋅ ρα uα uα + ∇ pα ∂t r r ~ 1 µ ˆ ˆ ˆ + nα ⋅ I pα − α ∇ ⋅ uα − µα ∇uα dS 3 Vα Aαβ ( ) ( ) α r 1 − ∇ µα ∇ ⋅ uα 3 ( α )− ∇ ⋅ (µ ∇ ur ) α α α ∫ r 1 ∂ 1 α ˆ ˆ + ρα uα + ∇ ⋅ ρα ε α ∂t εα r r α 1 1 ˆ ˆ + ∇ ⋅ ρα uα uα + ∇ ⋅ ( r r r 1 ˆ ˆ uα uα + ∇ ⋅ uα ) εα ( ) εα ( α r ˆ ˆ ρα uα ) (A137) εα ˆ ˆ ˆ ρα uα uα = 0. We can take advantage of the continuity equation (A222) to rewrite the momentum equation in weak conservation form as 144 . (A236). r r In the last step. (A135) The compressible viscous term in Equation (A223) can similarly be expanded to r r 1 ε ∇(µα ∇ ⋅ uα ) = α ∇ µα ∇ ⋅ uα 3 3 ( α 1 )+ V ∫ µα 3 r ˆ nα ∇ ⋅ uα dS .
Application of the averaging theorems to the convection term in Equation (A239) yields r ∇ ⋅ (ρα uα hα ) = ε α ∇ ⋅ ρα r ˆ ˆ + ∇ ⋅ ρα uα hα ( ( α )+ ∇ ⋅ ρˆ urˆ hˆ hα α )+ ∇ ⋅ ( ρˆ urˆ α α α α α hα α )+ ∇ ⋅ ( ρ α α r ˆ ˆ uα hα ) . r r A1. (A141) Finally.2.3 VolumeAveraged αPhase Conservation of Energy Equation The phase average of Equation (215) is r ∂ (ρα eα ) + ∇ ⋅ (ρα uα hα ) − ∇ ⋅ (kα ∇Tα ) = 0. the conduction term in Equation (A239) can be simplified to 145 .ρα + 1 Vα α ∂ r uα ∂t α + ρα α r uα α r ⋅ ∇ uα α + ∇ pα α r 1 − ∇ µα ∇ ⋅ uα 3 ( α )− ∇ ⋅ (µ ∇ ur ) α α α ∫ r r ~ µ ˆ ˆ ˆ nα ⋅ I pα − α ∇ ⋅ uα − µα ∇uα dS 3 Aαβ r 1 ∂ 1 α ˆ ˆ + ρα uα + ∇ ⋅ ρα ε α ∂t εα r r α 1 1 ˆ ˆ ρα uα ⋅ ∇ uα + ∇ ⋅ + ( r r r 1 ˆ ˆ uα uα + ∇ ⋅ uα ) εα ( α r ˆ ˆ ρα uα ) (A138) εα εα ˆ ˆ ˆ ρα uα uα = 0. ∂t (A139) The same assumptions for noslip and gradients of porosity will be applied to the energy equation as well. Application of the averaging theorems to the accumulation term in Equation (A239) yields ∂ ρα eα ∂ (ρα eα ) = ∂t ∂t = εα ∂ ρα ( α eα α ∂t ) + ∂ ρˆ eˆ ∂t r uα α α (A140) α α .
A1.− ∇ ⋅ (kα ∇Tα ) = −∇ ⋅ [kα ∇Tα ] − = −∇ ⋅ kα ε α ∇ Tα 1 V α ∫ n ⋅ k ∇T dS 1 1 ˆ n T dS − n ⋅ k ∇T dS . Substituting Equations (A240). + V ∫ V∫ α α α Aαβ (A142) α α α α α Aαβ Aαβ The last step in Equation (A242) involves decomposing the point temperature according to the decomposition definition. Aαβ 146 . (A241).2.4 VolumeAveraged βPhase Conservation of Energy Equation The phase average of Equation (216) is ( ρc v ) β ∂Tβ ∂t − ∇ ⋅ (k β ∇Tβ ) = 0 (A144) which can be readily simplified to (ρcv )β 1 − Vβ ∂ Tβ ∂t β − ∇ ⋅ k β ∇ Tβ β + 1 Vβ ∫ ˆ nβ Tβ dS Aαβ ∫ (A145) nβ ⋅ k β ∇Tβ dS = 0. and (A242) into Equation (A239) gives ∂ ρα ( α eα α ∂t − + 1 1 εα V 1 )+ ∇⋅( ρ α α r uα α hα α ∫ ( nα ⋅ kα ∇Tα dS + α Aαβ εα ∇ ⋅ ρα r ˆ r 1 ˆ ˆ ˆ uα hα + ∇ ⋅ ρα uα hα ) ˆ ˆ 1 ∂ ρα eα εα ∂t r 1 ˆ ˆ + ∇ ⋅ ρα uα hα α )− ε1 ∇ ⋅ k ε ∇ T α α α α + 1 V ∫ εα ( εα ( α ) ˆ nα Tα dS Aαβ (A143) α )+ ε1 ∇ ⋅ ρˆ urˆ hˆ α α α α = 0.
T ∂t α (A146) The first term on the right in Equation (A246) representing the time rate of change of entropy can be expanded to obtain ∂ ρ α sα ∂ ( ρ α sα ) = ∂t ∂t = εα ∂ ρα ( α sα α ∂t ) + ∂ ρˆ sˆ ∂t (A147) . (A148) Using Fourier’s Law. α α In developing this volumeaveraged equation. Gradients in the porosity will also be neglected.5 VolumeAveraged αPhase Entropy Generation Equation The phase average of Equation (217) is s′′′ .α = gen r q r ∂ (ρα sα ) + ∇ ⋅ (ρα sα uα ) + ∇ ⋅ α ≥ 0. the third term on the right of Equation (A246) can be expanded to k ∇T − ∇ ⋅ α α Tα ∇Tα 1 = −∇ ⋅ kα − Tα V ∫ nα ⋅ Aαβ kα ∇Tα dS Tα (A149) Substituting Equation (A247) through Equation (A249) into Equation (A246) produces 147 . the noslip assumption is valid at the fluid solid interface.2.A1. Now applying the averaging theorems to the second term on the right of Equation (A246) representing the convection of entropy gives r ∇ ⋅ (ρα sα uα ) = ε α ∇ ⋅ ρα r ˆ ˆ + ∇ ⋅ ρα uα sα ( ( α sα α α )+ ∇ ⋅ ρˆ sˆ urˆ r uα α )+ ∇ ⋅ ( ρˆ sˆ α α α α α r uα α )+ ∇ ⋅ ( ρ α α r ˆ ˆ sα uα ) .
these equations will be represented entirely in terms of volume averaged quantities. In the present form. gen β β = ρβ ∂ sβ ∂t β − ∇Tβ ∇ ⋅ k β εβ Tβ 1 1 − Vβ ∫ nβ ⋅ k β ∇Tβ Tβ dS (A152) Aαβ The volume averaged equations have been developed in sufficiently generalized form. The details of this simplification are the subject of Chapter 2.2.6 VolumeAveraged βPhase Entropy Generation Equation The phase average of Equation (218) is s′′′. 148 . gen = β ∂ (ρ β sβ ) ∂t r qβ + ∇ ⋅ T β ≥0 (A151) which can be readily expanded to s′′′. gen α = ∂ ρα ( α sα α α + ∇ ⋅ ρα + − 1 ( ∂t sα ) + 1 ∂ ρˆ sˆ εα ∂t α α α r uα α α εα 1 r ˆ ˆ ∇ ⋅ ρα uα sα ( )+ ε )+ ε1 ∇ ⋅ ( ρˆ sˆ α α α r uα α )+ ε1 ∇ ⋅ ( ρ α α α r ˆ ˆ sα uα ) (A150) 1 α r ˆ ˆ ˆ ∇ ⋅ ρα sα uα nα ⋅ kα ∇Tα dS Tα ∇Tα 1 ∇ ⋅ kα − εα Tα Vα ∫ Aαβ A1. the equations do not present a tractable problem. Eventually.′ sα′′.
It is possible to construct upwind or biased difference operators. 149 . In general. it was found that central finite difference operators performed well due to the diffusive nature of the continuity equation. It was noticed that the second order central operators were more stable than fourth order. a. In this research. but the fourth order provided more accuracy. can be calculated as ∂a ≅ Dij a j ∂x i (A21) where Dij is the linear differentiation operator. Second order central and fourth order central differences were employed. i. the ith derivative of a quantity. Higher order differentiation can be performed using either a higher order operator or via recursive operations.APPENDIX 2 – DERIVATION OF DIFFERENTIATION OPERATORS USING MATLAB Spatial differentiation can be expressed as a linear operator using a variety of discrete differentiation schemes. These gave little success for this problem.e ∂ 2a 2 ≅ Dij (D jk ak ) ∂x i (A22) These operators can be of variable order or method.
+ 1! i 2! i 3! i 4! i ui −1 = ui (− 1∆x )1 u (1) + (− 1∆x )2 u (2 ) + (− 1∆x )3 u (3) + (− 1∆x )4 u (4 ) + . the desired operator can be created. the three point central differences for the first and second derivative are formed by taking a linear series of the i+1 and i1 points: aui −1 = aui + a (− 1∆x )1 u (1) + a (− 1∆x )2 u (2 ) + O (∆x 3 ) 1! i 2! i (A24) bui +1 = bui + b (+ 1∆x )1 u (1) + b (+ 1∆x )2 u (2 ) + O (∆x 3 ) 1! i 2! i (A25) Summing the threepoint approximation equations gives 150 .. + 1! i 2! i 3! i 4! i ui + 2 = ui M (+ 2∆x )1 u (1) + (+ 2∆x )2 u (2 ) + (+ 2∆x )3 u (3) + (+ 2∆x )4 u (4 ) + .Any finite difference operator can be constructed by forming a Taylor series expansion: ui − n = ui M (− n∆x )1 u (1) + (− n∆x )2 u (2 ) + (− n∆x )3 u (3) + (− n∆x )4 u (4 ) + ....... + 1! i 2! i 3! i 4! i ui + n = ui + (+ n∆x )1 u (1) + (+ n∆x )2 u (2 ) + (+ n∆x )3 u (3) + (+ n∆x )4 u (4 ) + . 1! i 2! i 3! i 4! i (A23) By forming a truncated linear series. + 1! i 2! i 3! i 4! i ui +1 = ui (+ 1∆x )1 u (1) + (+ 1∆x )2 u (2 ) + (+ 1∆x )3 u (3) + (+ 1∆x )4 u (4 ) + . + 1! i 2! i 3! i 4! i ui − 2 = ui (− 2∆x )1 u (1) + (− 2∆x )2 u (2 ) + (− 2∆x )3 u (3) + (− 2∆x )4 u (4 ) + ...... For example.
For i=1 u 2 = u1 (+ 1∆x )1 u (1) + (+ 1∆x )2 u (2 ) + O (∆x 3 ) + 1! 1 2! 1 (A29) u 3 = u1 (+ 2∆x )1 u (1) + (+ 2∆x )2 u (2 ) + O (∆x 3 ) + 1! 1 2! 1 (A210) summing − (a + b )u1 + au2 + bu3 = [a(+ 1) + b(+ 2) ](∆1x! ) u ( ) + [a(+ 1) + b(+ 2) ](∆2x!) u ( ) + O(∆x ) 1 1 1 1 2 2 2 2 3 1 1 (A211) which can be solved for both the first and second order approximations as 151 . ui (1) (− 1)1 aui −1 − (a + b )ui + bui +1 2 = + O ( ∆x ) . = 2 ∆x 2 2 2 (− 1) (− 1) a + (1) b 2! [ ] (1)1 a = 0 .aui −1 − (a + b )ui + bui +1 = (− 1) a + (1) b 1 1 [ ] ∆1x ! 1 ui + (− 1) a + (1) b (1) 2 2 [ ] ∆2x! 2 ui (2 ) + O (∆x 3 ) (A26) which can be solved for the first and second derivatives. (1)2 b 1 (A28) The matrix equations are chosen such that the unwanted derivatives are eliminated. The end node approximations are calculated using biased formulae. 2 ∆x 1 1 1 (− 1) (− 1) a + (1) b 1! [ ] (1)1 a = 1 (1)2 b 0 (A27) ui (2 ) (− 1)1 au i −1 − (a + b )u i + bu i +1 1 + O ( ∆x ) .
the formulae for i=n are un (1) = − (a + b )u n + au n −1 + bu n − 2 1 [a(− 1) + b(− 2) ] (∆x ) 1 1 (− 1)1 + O (∆x ) . Applying this same technique for a five point stencil results in operators which are O(∆x4) accuracy.u1 (1) = (∆x )1 [a(+ 1) + b(+ 2) ] 1 1 − (a + b )u1 + au 2 + bu 3 1! (1)1 + O (∆x 2 ) . 2 (1) (1)1 + O (∆x ). (− 2 )2 b 1 (A214) 1! un (2 ) = − (a + b )u n + au n −1 + bu n − 2 (∆x )2 [a(− 1) + b(− 2) ] 2 2 (− 1)1 + O (∆x 1 ) . 2 (1) 1 (2)1 a = 1 (2)2 b 0 (2)1 a = 0 . the operators are − 3 4 − 1 1 = 1 −1 0 2∆x 1 −4 3 ( D21) (A216) and 1 − 2 1 1 = 2 1 − 2 1 . These are given as 152 . 2 (− 1) (A215) 2! For a system with three spatial nodes. ∆x 1 − 2 1 D ( 2) 2 (A217) These operators are both O(∆x2) accuracy. 2 (− 1) 2 (− 2)1 a = 1 (− 2 )2 b 0 (− 2)1 a = 0 . (2)2 b 1 (A212) u1 (2 ) = 2 [a(+ 1) + b(+ 2) ] (∆x ) 2 2 − (a + b )u1 + au 2 + bu 3 2! (A213) Similarly.
( D41) − 25 48 − 36 16 − 3 − 3 − 10 18 1 −6 1 1 = 0 8 −8 − 1 12∆x 6 3 − 18 10 −1 3 − 16 36 − 48 25 (A218) and 35 − 104 114 − 56 11 − 20 6 4 1 − 1 16 = − 30 16 12∆x 2 4 6 − 20 − 1 11 − 56 114 − 104 11 − 1 − 1 . 11 35 ( D42) (A219) 153 .
The coupling occurs through the electromotive force in the mechanical model and the back emf voltage in the electrical model. The compressor electrical power input and PV power output are not equivalent due to losses which occur in the compressor. 154 . and the resulting PV power due to this motion. can be predicted by measuring the voltage and current waveforms delivered to the compressor. which is simple to measure. Several unknown mechanical and electrical design parameters must be The experimental measurement of these design determined before this is possible. The largest loss is due to resistive power dissipation. These losses require a complete model to allow the prediction of piston motion.1 The electromechanical system The compressor model consists of a mechanical model for the piston moving mass which is coupled to an electrical model for the voice coil circuit. and the basis for this model are developed by Leach who has modeled a loudspeaker electromechanical system (69). parameters. Some smaller losses are due to eddy currents and friction.APPENDIX 3 – COMPRESSOR ELECTROMECHANICAL MODELING The compressor piston position. Experimental measurement of the voice coil current and voltage in addition to the compression space pressure then allows the piston position to be predicted. A3.
To cryocooler Housing plenum. This type of compressor utilizes a “voicecoil” to create an electromotive force to drive the piston motion. the reaction forces due to pressure. A3. ph Compression Space. Most compressors use a suspension spring of some sort. The term “voicecoil” comes from audio load speakers which operate very similarly to reciprocating compressors. In this freebody diagram. electromotive force. k Magnetic structure Voice coil Figure 51 – Voice coil compressor schematic The compressor schematic illustrated in Figure 51 is representative of the compressors commonly used to drive Oxford class cryocoolers including Stirling and Pulse Tube Cryocoolers. and inertia are 155 . The compressor is essentially a loudspeaker designed to operate at a very high sound pressure level (SPL). friction.2 The Mechanical System The compressor model can be constructed by considering a freebody diagram of the compressor piston. Flexures allow the piston to be centered in the cylinder which reduces frictional drag. similar to Figure 52. pc Piston moving mass Suspension spring stiffness. spring stiffness. Other pistons are designed to have a sliding contact seal between the piston and cylinder.
The pressure in the compression space on the positive x face of the piston is pc and the pressure in the housing plenum space is ph . b . These pressures act on the piston cross sectional area. The moving piston experiences a frictional force due to sliding contact between the piston and cylinder or due to viscous shearing of the gas in the pistoncylinder clearance gap in the case of flexuremounted pistons. l . times the velocity of the piston. is the product of the magnetic flux in the air gap. In either case. resists the piston movement. the frictional force can be approximated by the friction coefficient. k . Bl ⋅ I & bx pc Apiston Apiston m ph Apiston kx x Figure 52 – Compressor piston freebody diagram (FBD) 156 . B . Apiston . The electromotive force constant. Bl . and the effective length of wire that cuts the flux. It is assumed that Bl is a nonlinear function of the piston position and possibly other parameters such as frequency or current. The suspension spring stiffness. In general.shown. k is a function of the piston position.
m . Apiston ph = pb 1 − x Vhou sin g (A32) where pb is the mean. More accurate modeling of the housing pressure is possible. Bl . k . I . (A31) This equation can be solved if the pressures. ph . are known as functions of time. can be measured experimentally. This results in an expression for the housing pressure as a function of the piston position. m . Additionally. the mechanical equation becomes 2 p A k + b piston x = Bl ⋅ I − ~c Apiston m&& + bx + x & p Vhou sin g (A33) where the compression space pressure has been decomposed into a mean pressure and the fluctuating component. the coefficients. p (A34) The adiabatic assumption for the housing volume pressure reduces the information needed. and current. or baseline. pc = pb + ~c . b . can be modeled using a simple adiabatic compression process for the plenum volume. When this relation for the housing pressure is substituted into Equation (A41). as m&& + bx + kx = Bl ⋅ I + ( ph − pc )Apiston x & Equation (A41) is a differential equation which defines the piston position. Newton’s Law can be written for the piston moving mass. pc . The compression space pressure. and Apiston must be known. ph and pc . Heat transfer 157 . The pressure in the housing plenum.Considering all of the forces acting on the piston. pressure.
A3. then the piston motion is unaffected. provided the parameters are known. and the piston position can be calculated. The first and second harmonics are typically 10% of the fundamental. if the nonlinearity is related to a nonlinear Bl . and this is measured experimentally. Leach (69) has proposed a 158 . determining the piston motion. The AC p component of the compression space pressure. still remains. This indicates that there are significant nonlinearities in the electromechanical system.3 The Electrical System IL LE (ω ) I V Rcoil RE (ω ) IR & Blx Figure 53 – Leach’s lossy inductor voice coil model The experimentally observed voicecoil current presents large harmonic content with a pure sinusoidal voltage drive. then the piston motion is affected. At this point. The adiabatic model should be sufficient for a first estimate. However. ~c .between the gas and the housing and piston clearance seal flow will affect the housing pressure. the current can be measured experimentally. An electrical model can be considered such that the source of the nonlinearities can be studied. This is important for the current objective of If the nonlinearity is occurring entirely within the electrical system.
This voltage source is created by the voice coil traveling through the magnetic field. This circuit is driven by a sinusoidal voltage. Losses occur due to eddy currents in the magnetic pole structure. The current then flows through a parallel circuit representing the lossy inductor created by the voice coil. This ensures that there is negligible piston motion. RE. the current flowing through the inductor is defined by the nonlinear differential equation 159 . (A36) The loop voltage law around the parallel inductor and resistor can be written as RE I R = d ( LE I L ) . Leach proposes a method for measuring these two parameters using a small drive voltage at a frequency significantly above the resonance frequency (69). V. which is the coil DC resistance. and the voice coil velocity. are empirically determined and functions of frequency. flows through a constant resistance. The circuit diagram can be written mathematically as V = I ⋅ Rcoil + d & (LE I L ) + Blx . This voltage is proportional to the electromotive constant. Rcoil. Bl . I. The circuit also contains an additional voltage source. and the eddy current resistance.possible circuit model for the voicecoil which is illustrated in Figure 53. dt (A35) The node current law gives I = IL + IR . Current. LE. dt (A37) Finally. The lossy inductance. known as the back emf.
It is clear that the housing gas acts as a spring which increases with the square of the piston area. 160 . dt The drive current can now be written as (A38) I= & V − Blx + I L RE Rcoil + RE (A39) and the inductor current equation can be written as LE RE 2 dL RE dI L & (V − Blx ) .d (LE I L ) = (I − I L )RE . The gas spring stiffness also depends on the baseline pressure and the housing volume as expected. This term represents the apparent dissipative force due to coil and eddy current losses. The effective friction coefficient now contains a term which depends on the electromotive force constant squared. − IL − RE − E = R +R dt Rcoil + RE dt E coil (A310) The inductor current must be solved for simultaneously with the piston motion which can now be written as 2 p A Bl 2 k + b piston x = Bl ⋅ V + I L RE − ~c Apiston x + & m&& + b + x p Rcoil + RE Vhou sin g Rcoil + RE (A311) An effective spring stiffness and friction coefficient can be defined as keff = k + and pb Apiston Vhou sin g 2 (A312) beff = b + Bl 2 Rcoil + RE (A313) respectively.
the current. and the resulting PV power due to this motion. voltage. This is an attractive development for compressors which are not or cannot be equipped with piston position sensors. if there are many unknown coefficients. pressure. as the remaining unknown. The model can be tested using a compressor with piston position sensors. the compressor piston position. Using the methodology described by Leach. the piston area and housing volume can be calculated. and position. From the mechanical design. In this case. The coil resistance can easily be measured. First. the lossy inductance and resistance can be measured. However. can be predicted by measuring the voltage and current waveforms delivered to the compressor and the compression space pressure. Once the model parameters have been determined. it can be used to determine unknown parameters based on experimentally measured voltage. In the case of a few unknown parameters. Bl . then this presents a difficult problem. The suspension stiffness can be relatively easily measured or calculated based on the design. a correlation function needs to be chosen which increases the number of unknown parameters. current. For nonlinear parameters. but the ability to monitor the compressor piston position without a dedicated sensor which can leak provides significant motivation. 161 . The moving mass can also be measured easily. this should be tractable. The accuracy of the model remains to be determined. This leaves the electromotive force constant. pressure. it can be used as a design tool to predict compressor performance based on a choice of design parameters.The compressor electromechanical model can be used for a variety of purposes. and position can be measured. Secondly.
M. Tarasov. New York. and Scull. Radebaugh. 2004. and Longsworth. 617. Takashi. Pulse tube refrigeration progress.. Lowtemperature expansion pulse tubes. Pecharsky. Walker. 2. Acoustic streaming in pulse tube refrigerators: tapered pulse tubes. Advances in Cryogenic Engineering. K. New York and London. 79. V. Olsen.. Anderson. and Hideto. I. C. Pasker. and Schunk.. 8. New York and London. Osborne. 629. Vol. submitted for publication. G. An experimental investigation of pulse tube refrigeration heat pumping rates. Advances in Cryogenic Engineering. Y.BIBLIOGRAPHY 1. Gshneidner. G. 3. 10. Cryocoolers 9. Vol.. Plenum Press. Cryocoolers. J. 669.. Lewis. 769. Plenum Press. A. Bradshaw.. McCallum. Advances in Cryogenic Engineering. 1997. R. 10. K... K. Luo.. Jones. 1997. M. Cryocoolers 9. N. Masashi. Cryocoolers 9. 9. T. Mikulin. Development of a 2W class 4K GiffordMcMahon cycle cryocooler. Plenum Press. Vol.... 1983. L. Pfotenhauer. 5. E. Orlowska. and Shkrebyonock. S. Vol. M. 1967. Pecharsky. New York and London. Cryocoolers 9. Moormen.. J. Plenum Press. New York and London... Plenum Press. 1997.. E. Jewell. 29. Gifford. Longsworth. W.. New York and London. A. 1997. Plenum Press.. and Dennis. 4. New York. 663. R. Nellis... and Swift. R.. G. Processing and testing of the lowtemperature stage Er6Ni2Sn cryogenic regenerator alloy. Vol. 162 . 1984. Inertance tube optimization for pulse tube refrigerators. Gschneidner. N. 37. Improvements to the cooling power of a space qualified twostage Stirling cycle cooler. B. 608. 49. D.. Advances in Cryogenic Engineering. 12. 1965. I. Influence of alloying on the behavior and properties of Er3Ni. 6. 69. R. 12. Cryogenics. Kouki. No. M. and Eastwood. Plenum Press. V.. 1997. 7. M.
11. Chafe, J., Green, G., and Hendrix, J. 1997. A neodymium plate regenerator for lowtemperature GiffordMcMahon refrigerators. Cryocoolers 9, Plenum Press, New York and London, 653. 12. Kuehl, H., Schulz S., Walther, C. 1998. Theoretical models and correlations for the flow friction and heat transfer characteristics of random wire regenerator materials. Proceedings of the 33rd Intersociety Engineering Conference on Energy Conversion, American Institute of Chemical Engineers, New York, 207. 13. Chafe, J. and Green, G. 1998. Neodymiumribbonregenerator cooling performance in a twostage GiffordMcMahon refrigerator. Advances In Cryogenic Engineering, Vol. 43, Plenum Press, New York, 1589. 14. Shull, C., Ravikumar, K., and Frederking, T. 1994. Hydrodynamic characterization of perforated plate flow passages. Advances In Cryogenic Engineering, Vol. 39, Plenum Press, New York, 1615. 15. Kays, W. and London, A. 1964. Compact Heat Exchangers. McGrawHill, Inc. New York. 16. Organ, A. 1992. Thermodynamics and Gas Dynamics of the Stirling Cycle Machine. Cambridge University Press, Cambridge 7678. 17. Jiang, Y., Ju, Y., and Zhou, Y. 1998. A study of oscillating flow characteristics of the regenerators in high frequency pulse tube refrigerators. Advances In Cryogenic Engineering, Vol. 43, Plenum Press, New York, 1635. 18. Yuan, Z. and Dybbs, A. 1992. Oscillating flow and heat transfer in a Stirling engine regenerator. Fundamentals of Heat Transfer in Porous Media, Vol. 193, ASME, New York, 73. 19. Gedeon, D. and Wood, J. 1997. Oscillatoryflow regenerator test rig: hardware and theory with derived correlations for screens and felts. NASA Contractor Report 198442. 20. Helvensteijn, B., Kashani, A., Spivak, A., Roach, P., Lee, J. and Kittel, P. 1998. Pressure drop over regenerators in oscillatory flow. Advances In Cryogenic Engineering, Vol. 43, Plenum Press, New York, 1619. 21. Zhao, T. and Cheng, P. 1996. Oscillatory pressure drops through a wovenscreen packed column subjected to cyclic flow. Cryogenics, Vol. 36, 333. 22. Organ, A. 1997. The Regenerator and the Stirling Engine. Mechanical Engineering Publications Limited, London and Bury St Edmunds. 163
23. Roberts, T. and Desai, P. 2003. Periodic porous media flows in regenerators. Cryocoolers 12, Kluwer Academic, New York, 555561. 24. Watson, E. 1983. Diffusion in oscillatory pipe flow. Journal of Fluid Mechanics, Vol. 133, 233244. 25. Siegel, R. 1987. Influence of oscillationinduced diffusion on heat transfer in a uniformly heated channel. Transactions of the ASME, Vol. 109, 244247. 26. Siegel, R. 1987. Effect of flow oscillations on axial energy transport in a porous media. Transactions of the ASME, Vol. 109, 242244. 27. Kaviany, M. 1990. Performance of a heat exchanger based on enhanced heat diffusion in fluids by oscillation: analysis. Journal of Heat Transfer, Vol. 112, 4955. 28. Kaviany, M. and Reckker, M. 1990. Performance of a heat exchanger based on enhanced heat diffusion in fluids by oscillation: experiment. Journal of Heat Transfer, Vol. 112, 5663. 29. Wu, P. Zhang, L., Qian, L., and Zhang, L. 1994. Numerical modeling of orifice pulse tube by using the method of characteristics. Advances In Cryogenic Engineering, Vol. 39, Plenum Press, New York, 1417. 30. Kornhauser, A. and Smith, J. 1989. Heat transfer with oscillating pressure and oscillating flow. Proceedings of the 24th Intersociety Energy Conversion Engineering Conference IECEC89. Vol. 5, 2347. 31. Bauwens, L. 1995. Twodimensional nearly isothermal pulsetube and regenerator model. Proceedings of the 10th Intersociety Cryogenic Symposium. American Institute of Chemical Engineers, New York, NY, 119. 32. Roach, P., Kashani, A., and Lee, J. 1996. Theoretical analysis of a pulse tube regenerator. Advances in Cryogenic Engineering, Vol. 41. Plenum Press, New York, 1357. 33. Kashani, A. and Roach, P. 1998. An optimization program for modeling pulse tube coolers. Advances in Cryogenic Engineering, Vol. 43. Plenum Press, New York, 1903. 34. Gary, J. and O’Gallagher, A. 2000. REGEN3.2 User Manual, National Institute of Standards and Technology.
164
35. Lewis, M., Kuriyama, T., Xiao, J., and Radebaugh, R. 1998. Effect of regenerator geometry on pulse tube refrigerator performance. Advances in Cryogenic Engineering, Vol. 43. Plenum Press, New York, 1999. 36. Gedeon, D. 1999. Associates.
Sage Pulse Tube ModelClass Reference Guide,
Gedeon
37. Harvey, J., Desai, P., and Kirkconnell, C. 2003. A comparative evaluation of numerical models for cryocooler regenerators. Cryocoolers 12, Kluwer Academic, New York, 547554. 38. Yuan, Z. and Dybbs, A. 1992. High frequency temperature measurement in porous metals. Fundamentals of Heat Transfer in Porous Media, Vol. 193, ASME, New York, 67. 39. Rawlins, W. 1992. The measurement and modeling of regenerator performance in an orifice pulse tube refrigerator. Ph.D. thesis, University of Colorado, Boulder, CO. 40. Rawlins, W., Radebaugh, R., Bradley, P., and Timmerhaus, C. 1994. Energy flows in an orifice pulse tube refrigerator. Advances in Cryogenic Engineering, Vol. 39, Plenum Press, New York, 1449. 41. Hassanizadeh, S. and Gray, W. 1979. General conservation equations for multiphase systems: 2. Mass, momenta, energy, and entropy equations. Advances in Water Resources. Vol. 2, 191. 42. Whitaker, S. 1999. The method of volume averaging. Kluwer Academic Publishers. Netherlands. 43. Quintard, M. and Whitaker, S. 1996. Transport in chemically and mechanically heterogeneous porous media. I: Theoretical development of regionaveraged equations for slightly compressible flow. Advances in Water Resources, Vol. 19, No. 1, 2947. 44. Gray, W. and O’Neill, K. 1976. On the general equations for flow in porous media and their reduction to Darcy’s law. Water Resources Research, Vol. 12, No. 2, American Geophysical Union, 148154. 45. Whitaker, S. 1996. The Forchheimer equation: a theoretical development. Transport in Porous Media, Vol. 25, Kluwer Academic Publishers, Netherlands, 2761.
165
No. J. Schiesser. 14. 232260. Pareschi. Heat transfer at the boundary between a porous medium and a homogeneous fluid. OchoaTapia. Momentum transfer at the boundary between a porous medium and a homogeneous fluid . 50. Numerical schemes for hyperbolic systems of conservation laws with stiff diffusive relaxation. 4. ACM Transactions on Mathematical Software.I. 1997. 1975. A. 1997. 3. OchoaTapia. 122. J. No. Vol. Naldi. G. 37. Journal of the Association for Computing Machinery. 56. L. Second Edition. International Journal of Heat and Mass Transfer. and Madsen. 13951417. No. J. Vol. Momentum transfer at the boundary between a porous medium and a homogeneous fluid . 1995. SIAM Journal on Scientific Computing. 40. 39. 53. SIAM Journal of Numerical Analysis. Association for Computing Machinery. Madsen. comparison with experiment. 14. The numerical method of lines : integration of partial differential equations. and Pareschi. 51. OchoaTapia. San Diego. 48. Chapter 11. J. 1995. Vol. 11. Inc. International Journal of Heat and Mass Transfer. 55. 2001. R. 38. and Sincovec. 3. ACM Transactions on Mathematical Software. International Journal of Heat and Mass Transfer. Vol. No. and Whitaker. 12461270. and Wei. ALGORITHM 540 PDECOL. and Whitaker. Academic Press. Numerical solution of parabolic partial differential equations with twopoint boundary conditions by use of the method of lines. 1991. Central differencing based numerical schemes for hyperbolic conservation laws with relaxation terms. 166 . 47. S. and Reichelt. Bejan. 326351. Vol. 1967. 1979. 49. Hicks. 2000. Inc. John Wiley & Sons.46. 1. J. Advanced Engineering Thermodynamics. No. Association for Computing Machinery. Sincovec. 5.. 549562. S. 4. 26912707. Shampine. 1997. theoretical development. L. N. M. 26352646. 54. Vol. Vol. No. The MATLAB ODE Suite. Software for nonlinear partial differential equations. 26472655. SIAM Journal of Numerical Analysis. 18. 14. S. N. 38. Vol. W. L.. and Whitaker. 52. general collocation software for partial differential equations. No. No. R. 3. Inc. Vol.II.
C. 1995. Cellularautomata calculation of frequencydependent permeability of porous media. No. Diffusion and dispersion in porous media. 442. Roach. The Aerospace Corporation Press. 3. 50. 1996.. Harvey. 67. 4. Soloski. No. Kirkconnell. GA. J. 4. 8.. 5. S. 1994. D. Advances in Cryogenic Engineering. 1329. and Yanping. 63. parameter estimation. 68. ACM.D. Ltd. Master’s Thesis. 2593. Vol. Chemical Engineering Science. Ph. and Chan. Weideman. 6. ACM 60. Vol. 62. Vol. The transport equations for multiphase systems. S. Journal of the Audio Engineering Society. Kirkconnell. Georgia Institute of Technology. New York and London. 285. Plenum Press. Satellite thermal control handbook. P. Morton. 465519. 58. Vol. The American Physical Society. Atlanta. and Reddy. No. 47. 1999. Leach. New York and London. New York. Vol. 420. Loudspeaker voicecoil inductance losses: circuit models. and Price. D. 1973. 64. CA. VOl. El Segundo. 61. Whitaker. 66. 69. S. Knackstedt. Numerical analysis of the mass flow and thermal behavior in highfrequency pulse tubes.. J. S.. 1998. Vol. Vol. K. AIChE Journal. Harvey. No. M. Cryocoolers 9. Hendrix. Computers and Fluids. 28. 45. Sahimi. J. 65. 373382. 167 . Unconditional stability. monotonicity and accuracy of the linear exponential interpolation function for the Burgers Equation. C. and Desai. 1993. Gilmore. SIAM Journal of Numerical Analysis. 2000. thesis. No. M. 59. Kirkconnell. 963983. Kluwer Academic. 2000. A new method for producing perforated plate recuperators. Advances in Cryogenic Engineering. 13. 39. R. 1997.. 2002. GA. 27.57. Whitaker. No. K. and effect on frequency response. Vol. 2001. 26. S.. Regenerator performance evaluation in a pulse tube cryocooler. A MATLAB differentiation suite. Elsevier Science. 15561597. Parametric study of cryocooler regenerator performance. Transactions on Mathematical Software. Physical Review E. Plenum Press. Experiments on the effects of pulse tube geometry on PTR performance. C. Discretization of unsteady hyperbolic conservation laws. 139.. J. Atlanta. 1967. 41. Georgia Institute of Technology. M.