Cleveland State University

EngagedScholarship@CSU

ETD Archive

2018

Limits of Small Scale Pressure Swing Adsorption

Aaron A. Moran
Cleveland State University

Follow this and additional works at: https://engagedscholarship.csuohio.edu/etdarchive

Part of the Chemical Engineering Commons

How does access to this work benefit you? Let us know!

Recommended Citation
Moran, Aaron A., "Limits of Small Scale Pressure Swing Adsorption" (2018). ETD Archive. 1050.
https://engagedscholarship.csuohio.edu/etdarchive/1050

This Dissertation is brought to you for free and open access by EngagedScholarship@CSU. It has been accepted
for inclusion in ETD Archive by an authorized administrator of EngagedScholarship@CSU. For more information,
please contact [email protected].
LIMITS OF SMALL SCALE PRESSURE SWING ADSORPTION

AARON A. MORAN

Bachelor of Science in Chemical Engineering

Miami University

May 2012

Master of Science in Chemical Engineering

Cleveland State University

June 2014

Submitted in partial fulfillment of requirements for the degree

DOCTOR OF ENGINEERING

at the

CLEVELAND STATE UNIVERSITY

May 2018
 ACKNOWLEDGEMENTS

First and foremost, I would like to offer my sincerest gratitude to Dr. Talu for

encouraging me to pursue my doctoral degree and for all his support along this journey. I

am extremely grateful for his willingness to allow me to pursue a topic that excited me,

despite it not always being his favorite. Without a doubt, I would not be the

scientist/engineer I am today without him. Not only has he taught me how to be a better

experimentalist and researcher, but he also significantly enhanced my teaching ability.

The wide ranging advice he has given me over the years has been invaluable and I will

never forget it.

Next, I would like to thank my committee, Dr. Gatica, Dr. Wirth, Dr. Ungarala,

and Dr. Fodor for their input and critique of my thesis. Dr Ungarala deserves additional

acknowledgement for allowing use of the PSA model he developed, which was

instrumental in the data analysis and improving my understanding of the PSA process.

Dr. Thrash, Dr. Shah, and Dr. Gumma have also provided valuable input on various

papers and presentations along the way of which I am very grateful. Finally, I would like

to thank Dr. Belovich for her guidance and mentoring during my three years as a teaching

assistant under her.

Additionally, I wish to thank Jim Barker for his help with the electrical side of the

setup of the PSA process used in this thesis and wish him all the best in his retirement.

Becky Laird and Darlene Montgomery deserve extra thanks for all the help they have

provided during my time at CSU. I have also had the pleasure of mentoring several

graduate students who have worked extremely hard and did excellent work (Mihir Patel,
Himabindu Gandra, Muhammet Köksal, and Philipp Schandelmaier). I would not have

been able to finish this dissertation as quickly as I did without their assistance and

patience with me as I tried my best to teach them what I know. Mihir and Qian Qian

Zhou deserve special mention as well as some of the data used in the work was collected

by them. Dustin Bowden also deserves acknowledgment for helping take the SEM

images in this study.

Lastly, I would like to thank my family and friends for their love and support over

the years I have undertaken this project. A special shout out goes out to my roommates

Steve Cook and Mike Salem who put up with my often strange schedule and lack of

social life. I finally must express my utmost gratitude to my parents for their love and

encouragement throughout this process. Without their support in my times of need, I

would not be in the position I am today.

 LIMITS OF SMALL SCALE PRESSURE SWING ADSORPTION

AARON A. MORAN

ABSTRACT

Portable or small scale pressure swing adsorption (PSA) systems have gained

increasing popularity in both industry and literature due to the commercial success of

personal oxygen concentrators (POCs). While these processes have much in common

with larger PSA systems, significant differences exist that make understanding process

limitations difficult. These include faster cycle times, smaller adsorbent particles, and a

reduced column size.

Macropore diffusion is traditionally assumed to control the mass transfer rate in

columns packed with zeolite particles in an oxygen production process. While numerous

studies have confirmed this assumption for the particle size used in industrial size PSA

processes, it has not been validated for the much smaller particle size used in small scale

PSA. Smaller particles improve the mass transfer rate by increasing interfacial area per

volume as well as decreasing diffusion distance. Despite this reduction, small scale PSA

simulations often still assume a mass transfer rate solely limited by macropore diffusion.

This approach fails to adequately account for the influence of other mass transfer

mechanisms whose impact increases due to particle size reduction. This study

experimentally demonstrates the dominant mass transfer mechanism is no longer

macropore diffusion for the particle size used in small scale PSA for oxygen production.

Depending on the gas velocity, axial dispersion effects either become the limiting

mechanism or equally as important as macropore diffusion. It also shows that improperly

iv
accounting for axial dispersion effects has a significant impact on the mass transfer

coefficient estimation, often measured with breakthrough experiments.

An important limitation for small scale PSA processes is the limit on adsorbent

utilization. Decreasing cycle time for a PSA process typically results in a gain in

adsorbent utilization, often represented in industry by the bed size factor (BSF).

Increasing adsorbent utilization is represented by a decrease in BSF. A low BSF is

desirable because it represents a smaller overall process size, which is highly attractive

for portable systems. Currently, there is no consensus in literature if a lower limit for the

BSF exists and what may cause it.

In this study, a two column small scale PSA process was used to measure the

cycle time of a minimum BSF. It represents the first experimental literature example of a

minimum BSF for a two column air separation process. The data was then used with a

literature model to better understand why the minimum was occurring and what was

primarily causing it. It was determined that macropore diffusional resistance is the

primary cause of a minimum BSF.

v
 TABLE OF CONTENTS
Page
ABSTRACT…………..…………………..……………………………………………. iv

LIST OF TABLES………………………...……………………………………………. xi

LIST OF FIGURES………..…………...……………………………………………... xiii

NOMENCLATURE…………….…………………………………………………….. xvi

CHAPTER

I. INTRODUCTION……………………………………...…………………... 1

1.1. Purpose…...….....…………………………………………………. 1

1.2. Scope of Work…………………………………………………….. 2

1.3. Intellectual Contributions…….…………………………...………. 4

II. BACKGROUND…………………..………………………………………... 5

2.1 Zeolites…………………………………………………………….. 5

2.1.1 Zeolite History…………………………….………….. 6

2.1.2 Zeolite Composition and Structure…………………... 7

2.1.3 Zeolite Applications………………….…………….... 11

2.2 Adsorption………………………………………………………. 13

2.2.1 Adsorption Fundamentals……………..…………….. 13

2.2.2 Adsorption Thermodynamics………….……………. 15

2.2.3 Diffusion in Microporous Adsorbents….…………… 16

2.3 Air Separation through Pressure Swing Adsorption……………. 19

2.3.1 PSA Principles……………………………….……… 19

2.3.2 History and Growth of PSA…….…………………… 21

2.3.3 Improvements to PSA Cycles……………………….. 25

vi
 2.3.4 Air Separation Applications…......…………………... 27

2.3.5 Future Challenges for PSA…….........…..….………... 28

III. THEORY………………………………………………………………... 30

3.1 Intermolecular forces…………………………………………... 30

3.1.1 Electrostatic Forces...………….……………………... 31

3.1.2 Induction and Dispersion Forces...…...……………… 32

3.1.3 N2 and O2 Interaction with Zeolites…..…..………….. 33

3.1.4 Exchangeable Cations………………………………... 35

3.2 Adsorption Equilibrium Modeling….……………………..…... 36

3.2.1 Henry’s Law……………………………………….... 37

3.2.2 The Langmuir Isotherm…………………....………… 37

3.2.3 Dual Site Langmuir Model…………………………... 39

3.3 Adsorption Column Dynamics………………………………... 41

3.3.1 Adsorption Rate...…………………………………..... 41

3.3.2 Axial Dispersion..........…………………………….… 43

IV. CHALLENGES OF SMALL SCALE PSA……………………..………. 48

4.1 Effect of Particle Size Reduction in Small Scale PSA..……..… 48

4.1.1 Current Assumptions………………………………… 49

4.1.2 MTC Estimation for Small Particles……………….… 51

4.1.3 Axial Dispersion Coefficient Estimation….……….… 53

4.1.4 Impact of Higher Axial Dispersion Effects with Small…..

Particles…...……………………………..…………… 55

4.2 Small Scale PSA…..……….……………………………….…. 58

vii
 4.2.1 Evolution of Small Scale PSA………………………… 59

4.2.2 Limitations of Small Scale PSA………………………. 60

4.3 Possible Causes of a Minimum BSF……………..………….... 61

4.3.1 Pressure Drop…………………..…………...………… 61

4.3.2 Heat Transfer Limitations…….….……………………. 63

4.3.3 Mass Transfer Limitations….…….……………...……. 63

4.3.4 BSF Model…………….………………………………. 64

4.4 Study Aims and Experimental Plan Overview………………… 65

4.4.1 Study Aims……………………………………………. 65

4.4.2 Overview of Experimental Plan………………………. 66

V. MATERIAL AND METHODS………………….……………………….. 67

5.1 Type of Zeolite………………….………………..……………. 67

5.2 Isotherms……………………….…………………...…………. 68

5.3 Column Packing and Activation….………………..……….… 70

5.4 Breakthrough Study…………………………...…..……….… 70

5.4.1 Breakthrough System..................................................... 70

5.4.2 Breakthrough Measurements………………………..… 71

5.4.3 Constant Pattern……………………………………….. 73

5.4.4 MTC Calculation……………………………………… 74

5.4.5 van Deemter Model…………………………………… 77

5.5 PSA System…………………………………………………… 79

5.5.1 PSA Flow Diagrams….....…………………………..… 80

viii
 5.5.2 PSA Instrumentation, Process Control, and Data………….

Collection……………………………………...…..... 83

5.5.3 System Design……...…………………………………. 85

5.5.4 Material Balances…………...…..…………………….. 87

5.5.5 Design Parameters……………………………….……. 89

5.5.6 Performance Parameters……….………………..…….. 91

5.5.7 Pressure Drop Study……….……..……………...…… 93

5.5.8 Minimum BSF Study……...……………………….…. 96

5.6 PSA Model Framework………………………………….… 100

5.6.1 Mass Balance……………...…………………….....… 101

5.6.2 Equilibrium Model………..…….………………..….. 101

5.6.3 Momentum Balance………………………………..… 102

5.6.4 Energy Balance……………………………………..... 102

VI. RESULTS AND DISCUSSION………………………………….…….. 103

6.1 Axial Dispersion Effects in Small Scale PSA……………… 103

6.1.1 Experimental MTC Measurement…..………………. 104

6.1.2 Mass Transfer Parameter Estimation…..………..…. 106

6.1.3 Rate Limiting Contribution…………..………….…. 109

6.1.4 Skin Resistance…………………………...………… 112

6.2 Role of Pressure Drop in Small Scale PSA………………… 114

6.2.1 Experimental Results……………,,………………… 114

6.2.2 Simulation Results………………..……………….... 117

6.3 Adsorbent Utilization Limit in Small Scale PSA………..…. 127

ix
 6.3.1 Measurement of a Minimum BSF…………...……… 127

6.3.2 What Causes a Minimum BSF?.................................. 133

7 VII. CONCLUSIONS…………………….……………………………….… 136

7.1 On the Effects of Axial Dispersion in Small Scale PSA…….. 136

7.2 On the Role of Pressure Drop in Small Scale PSA………..… 137

7.3 On the Existence of a Minimum BSF………………………... 139

7.4 Summary of Intellectual Contributions….………………...… 139

7.5 Future Work……………..………………………………...… 140

REFERENCES……………………………………………………………………….. 141

APPENDIX………………………………………………………………………….… 153

A. PSA Model….…………………………………………..…………….. 153

B. Breakthrough Experiments.….…...……………...………...……..….. 159

C. PSA Pressure Drop Experiments……………..………….…..…....…. 167

D. Minimum BSF Experiments……………...………..……...…………. 170

x
 LIST OF TABLES

Table Page

3.1 Quadrupole Moment and Polarizabilities of Nitrogen and Oxygen Molecules…. 34

4.1 Comparison of parameters used for estimates of koverall in Figure 4.2…........…... 57

5.1 Nitrogen and oxygen DSL parameters for LiLSX particles used in simulations.. 69

5.2 Breakthrough column specifications……..……………………………..…...…... 72

5.3 PSA schematic symbol description……….…………..…………..……….…...... 82

5.4 Experimental column specifications for pressure drop study……………............ 95

5.5 Cycle specifications for short column cycles in pressure drop study………........ 95

5.6 Cycle specifications for long column cycles in pressure drop study ………….... 96

5.7 Experimental column specifications for minimum BSF study…………..…….... 97

5.8 Cycle specifications for long column cycles with pressure ratio 3.5 in minimum…..

BSF study……..………………………………..…………………………… 98

5.9 Cycle specifications for long column cycles with pressure ratio 3 in minimum BSF

study …………………………………………………………..…..….….… 98

5.10 Cycle specifications for long column cycles with pressure ratio 2.5 in minimum.....

BSF study …………….……………………………………………………. 99

5.11 Cycle specifications for short column cycles with pressure ratio 3.5 in minimum....

BSF study …………………………………………………………………. 99

5.12 Cycle specifications for long column cycles with pressure ratio 3 in minimum...….

BSF study ……………………………………………………...……..….. 100

6.1 Comparison of van Deemter constants determined from experimental data and…..

previously cited literature values………………………………………… 108

xi
6.2 Comparison of equation 4.1 contributions for a nitrogen overall MTC at a….….…

Reynolds number of 20 and two different particle sizes……………………..111

6.3 Comparison of experimental and simulation results at similar cycle times and….…

different column dimensions (Pressure drop study)…………………….….. 118

6.4 Comparison of experimental and simulation results at different cycle times and.….

same column dimensions (Pressure drop study)………………………..….. 122

6.5 Slope and intercept of k’ vs. superficial velocity in Figure 6.18 for long…….....…..

columns…….................................................................................................. 131

6.6 Slope and intercept of k’ vs. superficial velocity in Figure 6.18 for short….….…...

columns……………………….…………………………………………… 131

xii
 LIST OF FIGURES

Figure Page

2.1 Examples of zeolite basic secondary units…………….…………………..……… 9

2.2 Sodalite unit with Si, Al atoms…......……………….………..….……………….. 9

2.3 Structure of Type A and Type X or Y zeolites…….………………,,……….….. 10

2.4 Potential energy diagram for adsorption……...…….……………………...……. 14

2.5 The five types of isotherms……………………...….………………………...…. 15

2.6 Depiction of pores and mass transfer resistances in adsorbent particle............…. 18

2.7 Basic two-column pressure swing adsorption system …......………………….... 24

2.8 Column pressure during Skarstrom cycle …………..……………………….….. 25

3.1 Calculated potential energy curves of nitrogen and oxygen…....……………….. 34

3.2 Potential energy curves of Na+N2 and Li+N2………….….................................... 36

3.3 Mass transfer zone inside a column…..……………………………..……………………. 45

3.4 Shapes of isotherms in determining the sharpness of concentration wavefront.... 46

3.5 Self-sharpening wavefront and dispersive wavefront………………………..….. 46

3.6 Column breakthrough curve………………..………………………………….... 47

4.1 Limiting Peclet number vs particle diameter for flow through packed beds..….. 54

4.2 Comparison of predicted overall nitrogen MTC using constants and correlations….

according to Table 1……………………………………………………..... 58

5.1 Oxygen isotherms for LiLSX zeolite……..….......………………………........... 68

5.2 Nitrogen isotherms for LiLSX zeolite …..…….………..…………………….... 69

5.3 Breakthrough experimental setup…………………………………………….… 72

5.4 Comparison of experiment and simulation breakthrough curve……..………..... 77

xiii
5.5 PSA middle exit manifold schematic……….……………...……………..……... 80

5.6 PSA system feed side schematic…………….………………………………...… 81

5.7 PSA system product side schematic ……………..…………………….…..….... 81

5.8 Comparison of isothermal capacity and working capacity in a PSA process….... 93

6.1 Experimental data vs. correlation predictions of the nitrogen overall MTC…... 105

6.2 HETP vs. superficial velocity from experimental breakthrough experiment..… 108

6.3 Experimental overall MTC plotted against overall nitrogen MTC predictions... 109

6.4 Comparison of koverall predictions for a 0.5 mm particle and 2 mm particle….... 111

6.5 SEM images of an extruded particle edge………………………...…….……… 113

6.6 SEM images of LiLSX particle used in this study……..…………………….... 113

6.7 Comparison of oxygen recovery decline with increasing production step superficial

velocity ………………………………………………..…………………… 116

6.8 Comparison of BSF with increasing production step superficial velocity for…..…...

columns of different dimensions……………….………..………………… 117

6.9 Comparison of gas phase nitrogen composition at the end of the production step for

the short columns and long columns ……………….…………………….. 119

6.10 Comparison of gas phase nitrogen composition at the end of the purge step for the..

short columns and long columns………………………………………….. 119

6.11 Comparison of solid nitrogen loading at the end of the production and purge steps..

for the short columns and long columns ……….………………...……….. 121

6.12 Comparison of gas phase nitrogen composition at the end of the production step.....

for the slow cycle and fast cycle ……………..……..…………………….. 124

xiv
6.13 Comparison of gas phase nitrogen composition at the end of the purge step for the.

slow cycle and fast cycle ……………..………………………………...… 124

6.14 Comparison of solid nitrogen loading at the end of the production and purge steps..

for the slow cycle and fast cycle ……………………..……….……...…… 125

6.15 Comparison of solid temperature profiles at the end of the production and purge.…

steps……………………….………………………..……….……...…… 126

6.16 Oxygen recovery results for long, thin columns and short, wide columns...…. 128

6.17 BSF results for long, thin columns and short, wide columns………………… 128

6.18 k’ plotted against superficial velocity of high pressure feed step for long, thin……

columns and short, wide columns………………………..……………..… 130

6.19 BSF plotted against superficial velocity of high pressure feed step for long………

columns ……………………………………………………………...….… 132

6.20 BSF plotted against superficial velocity of high pressure feed step for long………

columns (pressure ratio 3)….………………………………………...…..… 134

xv
 NOMENCLATURE

b adsorption equilibrium constant

pre-exponential constant (DSL model) for gas i for first site

ci gas phase concentration of species i,

d equilibrium constant for second site (DSL model)

pre-exponential constant (DSL model) for gas i for second site

dp adsorbent particle diameter

Dc micropore diffusivity

DL axial dispersion coefficient

Dk Knudsen gas diffusivity

Dm molecular gas diffusivity

Dp macropore diffusivity

E potential energy

e electron charge

F instantaneous flow rate

average flow rate

f force

G Gibbs free energy

g ionic valences

H enthalpy

h distance between charges

j number of components

K dimensionless Henry’s law constant

KH Henry’s law constant

xvi
K0 Henry’s Law constant at a reference state

k mass transfer coefficient

k’ empirical constant

ka adsorption rate constant

kb desorption rate constant

kf film mass transfer coefficient

koverall overall mass transfer coefficient

L column length

LMTZ mass transfer zone length

M quadrupole moment

MTZtime mass transfer zone time

Ni total number of moles of component i

NT total number of moles of adsorbate

ni* equilibrium value of adsorbed phase concentration of component i,

ni average adsorbed phase concentration over an adsorbent particle of component i,

ns saturation capaity

rc crystal radius

rp adsorbent particle radius

P pressure

Ps saturated vapor pressure

p partial pressure

Pé∞ limiting Peclet number

Q heat of adsorption

R ideal gas constant

xvii
Re Reynolds number

S entropy

s selectivity

Sc Schmidt number

Sh Sherwood number

T temperature

t time

tc time stoichiometric center exits column

tct cycle time

s selectivity

u interstitial gas velocity

uw wave velocity

w magnitude of charge

WCideal isothermal working capacity

x, mole fraction in the adsorbed phase

y mole fraction in the fluid phase

z axial coordinate

Greek Symbols

ε0 dielectric permittivity

α polarizability

θ fractional surface coverage

ρb column bulk density

ρp particle density

xviii
εb bed/column porosity

εp particle macroporosity

γ1, γ2 axial dispersion constant

error of material balance

μg gas viscosity

ρg gas density

τp pore tortuosity

Commonly Used Abbreviations

BSF = bed size factor

CSS = cyclic steady state

DSL = dual-site Langmuir

LDF = linear driving force

LiLSX = lithium low silica X zeolite

MTC = mass transfer coefficient

MTZ = mass transfer zone

POC = personal oxygen concentrator

PSA = pressure swing adsorption

TPD = tons per day

xix
 CHAPTER I

INTRODUCTION

1.1 Purpose

The abundance of life on Earth depends on oxygen, which composes around 21%

of our atmosphere. However, demand for high purity oxygen has risen tremendously to

support new manufacturing processes and treat various medical conditions. Industrial

uses for purified oxygen include steel production, chemicals, petrochemicals, glass,

ceramics, and paper. The primary medical application is oxygen therapy for patients

suffering from conditions such as asthma and COPD (Chronic Obstructive Pulmonary

Disease).

Generating nearly pure oxygen from air is possible through either cryogenic

distillation or pressure swing adsorption (PSA). The high energy cost of cryogenic

distillation limits its economic viability to large scale oxygen demand or when > 96%

oxygen is required. PSA cannot separate oxygen and argon, which limits the maximum

oxygen purity to 96%. Since purity above 90% is acceptable for most personal

1
applications, PSA is the logical choice for small medical devices, often labeled personal

oxygen concentrators (POCs). These small scale devices have numerous advantages over

oxygen cylinders, the alternative option for oxygen therapy, including portability and

permitted use on airplanes. However, the technology is not fully mature and opportunity

remains to improve the size and efficiency of these devices. Creating a superior device is

lucrative because the market for oxygen therapy in North America alone is expected to

grow from around $2.7 billion in 2014 to over $7 billion in 2024. Moreover, COPD has

been projected to become the third leading cause of death in the world by 2030, which

further increases future need for oxygen therapy.1

While large scale PSA processes are well researched, less is known about small

scale operations. Significant differences exist between large and small scale processes

that potentially change process limitations. The goal of this dissertation is characterizing

and understanding these differences to improve the future design of POCs.

1.2 Scope of Work

The most significant difference between large and small scale processes, other

than overall process size, is the smaller particle size used in the packed column(s). In

literature, it is well known that diffusion in the macropores of the adsorbent is the

primary mass transfer resistance in columns packed with large zeolite particles (> 2mm). 2

Despite little proof this applies to the particle size used in small scale processes, it is still

often assumed as a reasonable approximation. As this study will demonstrate, this

assumption is not correct, which has significant implications on how the mass transfer

rate is measured or estimated. A suitable approximation for the mass transfer rate is

critical for designing and simulating a small scale PSA process. Hence, a primary goal of

2
this dissertation is to demonstrate what limits the mass transfer rate in columns of small

particles.

Limitations of small scale processes are currently not well understood because a

commercial market for these processes has not existed until recently. Large scale

processes operate much slower than small scale processes. Furthermore, the size of small

scale processes is miniscule compared to industrial units where the production rate is

orders of magnitude larger; hence limitations of larger units will not apply to these small

units. Maximizing process performance relies on understanding these limitations. One

particular challenge highlighted in this study is determining what limits adsorbent

utilization. Decreasing cycle time for a PSA process typically results in a gain in

adsorbent utilization, often represented in industry by the bed size factor (BSF).

Increasing adsorbent utilization results in a lower BSF and smaller overall process size,

which is highly attractive for portable systems. Currently, there is no consensus in

literature if a lower limit for the BSF exists and what causes it. Furthermore, if a limit

does exist, the primary cause of it remains unknown. Hence, the other main goal of this

dissertation is to improve the understanding on what limits the size and speed of small

scale PSA processes.

This dissertation is organized in the following manner. Chapter 2 provides a

broad overview of zeolites, adsorption, and pressure swing adsorption for those readers

unfamiliar with these subjects. Chapter 3 provides specific details about the zeolite and

equilibrium model used in this dissertation and introduces concepts related to column

dynamics that are useful to understand this study. Chapter 4 reviews literature related to

small scale PSA and introduces the challenges currently facing the field that are

3
addressed through this dissertation. Chapter 5 details the methods and materials used in

this dissertation for the breakthrough and pressure swing adsorption experiments.

Finally, Chapter 6 discusses the results of the study and offers further discussion on what

conclusions can be drawn from the data collected.

1.3 Intellectual Contributions

 Experimentally demonstrates the importance of axial dispersion effects for small

LiLSX particles for the first time in literature

 Determined axial dispersion effects are the dominant contribution to an overall

MTC for Reynolds numbers applicable to breakthrough experiments and most

small scale PSA processes.

 Experimentally demonstrates pressure drop does not have a significant effect on

small scale PSA performance

 First literature experimental measurement of a minimum BSF for a two column

small scale process

 The minimum BSF was proposed to be determined by macropore diffusional

resistance

4
 CHAPTER II

BACKGROUND

This chapter introduces the ideas and concepts required to understand the methods and

results of this dissertation. In general, adsorption is a separation process that involves

the attraction of molecules to an adsorbent surface based on its size or molecular

interactions with the solid. This dissertation will focus on a specific adsorbent, LiLSX

zeolite, and specific adsorption process, pressure swing adsorption, for the application of

oxygen production from air.

2.1 Zeolites

Zeolites are one of many adsorbents used for separation and purification

processes. Adsorbents are microporous materials that permit mobility of adsorbate

molecules (molecules that adsorb to an adsorbent surface) within the adsorbent and

5
provide a high surface area to enhance interactions of guest molecules with the solid

surface. Adsorbents vary in structure and composition enabling them to interact

differently with various types of molecules. Selection of the right type of adsorbent is

critical to any adsorption process to achieve the desired separation.

2.1.1 Zeolite History

A Swedish man named Baron Cronstedt first depicted zeolites in literature in the

1750’s. He proposed the name “zeolite” after heating a natural zeolite and witnessing it

bubble and dance as steam released from the zeolite pores.3 Natural occurring zeolites

usually contain impurities and an irregular chemical composition that limit their impact

scientifically and industrially.4 Consequently, zeolites went largely unstudied for close to

200 years as they were viewed as rare minerals without a purification method. In 1905 in

Germany, a synthetic zeolite with a larger capacity compared to natural zeolites was

manufactured, which permitted the first commercial use of zeolites as a way to soften

water. 3

Organized research into zeolites did not begin until the late 1930’s when Richard

Barrer classified zeolites by their pore size. His research at Union Carbide lead to the

synthesis of the first synthetic zeolites (i.e. zeolites A, X, and Y) for industrial use.3, 5, 6

Union Carbide used readily available raw materials along with a lower synthesis pressure

and temperature compared to former methods to create these zeolites. The key features

of synthetic zeolites compared to the natural version were a larger pore size to

accommodate larger molecules, an enhanced pore volume to increase the capacity, and a

higher purity of the crystalline phase.6 The ability to create a zeolite with a structure and

6
function tailored to a specific process allowed the industrial potential of zeolites to

expand.3 This has produced zeolites today that are crucial to many industrial processes as

adsorbents and catalysts.

2.1.2 Zeolite Composition and Structure

Of the approximately 40 naturally occurring zeolites, only chabazite, faujasite,

and mordenite are primarily used in industry. Commercially, the most important

synthetically created zeolites are Type A, Type X or Y, synthetic mordenite, and all of

their ion-exchanged variations.7 The composition of a zeolite consists of microporous

crystalline aluminosilicates with a chemical formula of the form seen in equation 2.1:

where M is a metal cation of valence n, x is 2.0 or more, and y is the moles of water in the

pores. Cavities (or cages) within the zeolite structure are linked to other cavities by pores

that allow adsorbate molecules to permeate into the structure. Zeolites differ from other

adsorbents through their uniform crystalline structure that provides a well defined pore

size for molecules to travel through, while also allowing them to act as effective

molecular sieves. IUPAC convention classifies pores by their size as follows:8

1) Micropores: Less than 2 nm

2) Mesopores: Between 2-50 nm

3) Macropores: Greater than 50 nm

Micropores, unlike larger mesopores and macropores, are able to trap guest molecules

using their solid surface force field. The larger pores serve to assist with diffusion by

allowing molecules to travel to the micropores easily. Zeolites have pore sizes less than 2

7
nm and are therefore completely microporous. The porosity of a zeolite provides a much

larger total surface area compared to its external surface, which produces a high

adsorption capacity solid where adsorption only occurs in the micropores. Molecular

characteristics specific to a zeolite (i.e. pore size, shape, and properties like polarity)

control how adsorption and desorption (opposite of adsorption) occur within a zeolite.

The main building blocks of zeolites consist of SiO2 and Al3O2 units that connect

tetrahedrally through oxygen atoms. Several of these units form larger secondary units

that serve as building blocks for the zeolite structure. These secondary units, shown in

Figure 2.1, illustrate silicon and aluminum atoms at the apices with lines representing

oxygen bridges between them. More secondary units exist, but the units in Figure 2.1

represent the units used to build the more common zeolite types A, X, and Y. These

secondary units are linked in 3-D space to create a porous crystalline structure. For

example, shown in Figure 2.2 is a sodalite unit formed from S4R and S6R units (Figure

2.1). Eight sodalite units form the eight-membered oxygen ring of type A zeolites and

connect to the final crystal (Figure 2.3) through D4R units (Figure 2.1). Type X and Y

zeolites are formed by ten sodalite units connected by D6R units (Figure 2.1) organized

in 3-D space to form a twelve-membered oxygen ring as the smallest pore diameter

(Figure 2.3). These oxygen rings are responsible for providing entry for adsorbate

molecules into the cavity of the zeolite. They determine pore size and what molecules

can enter the structure. Since zeolites are 3-dimensional structures, accurately

representing them on a 2-D surface is difficult. Figure 2.3 is one of the better illustrations

in texts and literature.9 By comparison, a single nitrogen molecule is approximately 3.6

Å.

8
Figure 2.1 Examples of zeolite basic secondary units5

Figure 2.2 Sodalite unit with Si, Al atoms10

9
Figure 2.3 Structure of Type A (left), nitrogen molecule (center), and Type X or Y

zeolites (right)7

The silicon and aluminum atoms within zeolites are interchangeable (although it

can be difficult), allowing for a range of Si/Al ratios between one and infinity.9 If the

ratio is reduced below one, the structure collapses. An aluminum atom induces a net

negative charge on the structure, requiring an exchangeable cation to preserve structure

electroneutrality. Cations are electrostatically bonded near an Al atom and are therefore

not part of the zeolite structure; hence cation exchange is possible. Adjusting the Si/Al

ratio and type of exchangeable cation permits zeolite modification for a specific purpose.

A higher Si/Al ratio increases the hydrophobic nature of a zeolite, which is useful for

removal of organics from water and catalytic applications where water is undesirable.

The Si/Al ratio differentiates Type X and Y zeolites. It is between 1 and 1.5 for Type X

zeolites, while the ratio is between 1.5 and ∞ for Type Y zeolites. A lower Si/Al ratio

increases cation exchange capacity and the ability to adsorb polar molecules. Examples

of commonly used exchangeable cations include Na+, K+, and Ca2+.

10
 The exchangeable cation determines the pore size and properties of the zeolite.

For example, a Type 4A zeolite utilizes the exchangeable cation Na+ and all twenty three

available sites within a unit cell. For a Type A zeolite, Na+ cations located on the eight

membered ring that serves as the opening into the central cavity will restrict the free

diameter of the ring to 3.8 Å (commonly referred to as a 4A zeolite). Forming a 3A

zeolite requires substitution of Na+ with K+, which further reduces the free diameter since

K+ is larger than Na+. A Ca2+ or Mg2+ cation substitution creates a relatively

unobstructed ring since one bivalent cation replaces two univalent cations. This causes

the cation to shift location from the main cavity window to inside the main cavity. The

number of cations per unit cell controls the location of the exchangeable cations in the

zeolite. A sodalite structure contains Type I, Type II, and Type III active exchangeable

ion location sites, each with a different cation capacity. Type I sites are the most readily

accessible and fill up first while Type III sites are the most difficult to access and fill up

last.

2.1.3 Zeolite Applications

The benefit of zeolites to industry has grown enormously over the last sixty

years.6 Primary industrial applications include petroleum refining, petrochemical

processing, detergent production, and separation and purification processes. Consumer

applications include portable oxygen concentrators, stationary refrigerant drying, and air

brake dryers for trucks and trains.6 Perhaps the largest industrial application is catalysis,

which first became important in 1962 with the addition of zeolites as catalysts in a

petroleum process called fluid catalytic cracking. Zeolites developed during this time

11
were significantly more reactive in comparison to their predecessors, which decreased

processing costs and substantially increased gasoline production.4

Examples of zeolites used in separation and purification processes include: 1)

Removal of CO2, chlorides, and mercury from different process streams in petroleum

refining; 2) Drying of hydrocarbon liquids, hydrogen, and cracked gas in petrochemical

processes; 3) Drying and desulfrizing natural gas; 4) Removal of water, N2, and CO2

from air for cryogenic distillation or PSA processes; 5) Hydrogen purification; and 6)

Xylene isomer separation, etc.

Three types of separation are possible with zeolites. First, equilibrium separation

occurs if there exists a difference in molecular interactions between the zeolite surface

and adsorbate molecules. Equilibrium separation usually hinges on the polarity of the

zeolite surface and adsorbate molecules. Another possibility, kinetic separation, exists if

there is a difference in the transport rate (i.e. diffusion) of the adsorbate into the internal

cavity of the zeolite. This requires the adsorbent micropore size to be similar to that of

the adsorbate molecules undergoing separation. Finally, molecular sieving occurs if the

size of a zeolite pore is too small for one molecule but not another in a fluid mixture. Air

separation, the topic of this dissertation, involves equilibrium separation where the zeolite

preferentially adsorbs nitrogen allowing for production of nearly pure oxygen at feed

pressure. 5A and 13X zeolites are commonly used to accomplish this separation in a

PSA process.

12
2.2 Adsorption

Adsorption by definition is the accumulation of fluid molecules at a surface. This

phenomenon occurs because of an attraction between adsorbate molecules (either a gas or

liquid) and an adsorbent surface (porous solid).7 There are two adsorption mechanisms,

physical adsorption and chemisorption. Physical adsorption is of interest because it is

easily reversible and can form more than one molecular layer. Chemisorption involves

the formation of bonds between the adsorbate molcule and adsorbent surface. However,

it can only form one molecular layer on a solid surface and it is not easily reversible,

which limits its industrial impact for separation and purification processes. Air

separation occurs via physical adsorption, hence all references to adsorption from this

point forward in this dissertation refer to physical adsorption only.

2.2.1 Adsorption Fundamentals

The attraction between an adsorbent and adsorbate molecules results from

molecular interactions, which are a combination of permanent dipoles, induced dipoles,

and London dispersion forces. Permanent dipoles occur in polar molecules as a result of

an uneven distribution of charge in the electron cloud. Polar molecules can also induce

an uneven charge distribution (i.e. polarity) in nonpolar molecules if they are close

enough to interact. Nonpolar molecules do not have permanent dipoles when their charge

is averaged over time. However, at any instantaneous moment they will have a dipole

that has potential to induce a dipole on another nonpolar molecule, creating London

Dispersion forces. Repulsion forces occur when molecules are too close to each other

and their electron clouds overlap. When adsorption occurs, equilibrium exists between

13
repulsive and attractive forces. Figure 2.4 shows potential energy (sum of all the

interactions that exist between the adsorbate and adsorbent) as a function of the distance

of the adsorbate molecule from the adsorbent surface. The high positive repulsive

potential energy near the adsorbent surface is where the electron cloud overlap would

occur. The potential well depth, U(r0), is related to the interaction strength between the

adsorbate and solid surface. The larger the potential energy difference, the greater the

attraction. At zero Kelvin (no kinetic energy), a molecule would settle at the bottom of

the well. At all other finite temperatures, the molecule oscillates around the minimum

potential energy.

Figure 2.4 Potential energy diagram for adsorption7

14
2.2.2 Adsorption Thermodynamics

Adsorption equilibrium determines how much of a fluid molecule is trapped in an

adsorbent at a defined pressure, temperature, and composition. Adsorption causes a gas

molecule to lose at least one of its translational degrees of freedom because it restricts the

gas molecule to move along the adsorbent surface. This causes a decrease in entropy

(ΔS) and since adsorption is a spontaneous process, Gibbs free energy (ΔG) is reduced as

well. According to the thermodynamic expression in equation 2.2, ΔH (enthalpy) also

must decrease (heat is released), making adsorption an exothermic process.

Isotherms are a typical method of representing equilibrium for single component

adsorption. Isotherms are measured at a constant temperature while the pressure is varied

and the amount of adsorbate is measured. They form different shapes as shown in Figure

2.5. Type I isotherms are most relevant to this study and represent adsorbents that have

pores similar in size to the adsorbate molecules and fill up as a saturation limit is

approached. The other types of isotherms have larger pores compared to the adsorbate

and hence do not experience this saturation limit, except for type IV and V isotherms near

Ps. Type III isotherms represent systems where an adsorbate interacts more strongly with

other adsorbates than with the adsorbent. Type II isotherms can have multiple adsorbed

layers because of their larger pore size distribution and represents the other type of

isotherm sometimes used in a PSA process.

15
Figure 2.5 The five types of isotherms7

2.2.3 Diffusion in Microporous Adsorbents

Diffusional resistances affect the approach to equilibrium in an adsorption

process. There are three resistances to mass transfer related to an adsorbent particle. The

first occurs around adsorbent particles as film resistance if the fluid is a mixture, due to

accumulation of the less adsorbed component on the surface. The other two occur within

the particle in the macro/mesopores between the crystals and in the micropores within the

crystals themselves. These resistances are in series, but typically only one is rate

controlling. Zeolites are also often synthesized with a porous binder material that holds

the crystals together, especially for separation processes where a sudden change in

pressure occurs. An example of a zeolite particle structure is represented in Figure 2.6.

Macropore diffusion occurs in adsorbent pores significantly larger than the

diameter of the diffusing sorbate. In zeolite particles, these exist between microporous

crystals. When pore size is much greater than the mean free path of the sorbate

molecules, diffusion occurs by bulk gas diffusion. At low pressures and with smaller

16
pores, collisions between the diffusing molecule and pore wall become important

resulting in Knudsen diffusion. Furthermore, Poiseuille flow (pressure driven flow inside

a particle) becomes relevant with larger pore and particle sizes and at higher pressures,

such as the pressurization step of a PSA process. Under high adsorbate concentrations

and if the adsorbed phase is mobile, adsorbed molecules may diffuse on the pore surface

as well.8

Micropore diffusion occurs in pores of comparable size to the adsorbate

molecules. For a zeolite particle, this occurs within the crystals where pore size into the

cavity of the zeolite is limited by the oxygen windows. If a binder is present, micropore

diffusion will occur through the binder as well. The “adsorbed phase” in this type of

diffusion refers to the combination of molecules adsorbed on the pore wall and trapped in

the middle of the pore.8 Micropore diffusion is an activated process because unlike

macropore diffusion, it is not possible for the adsorbate molecules to escape the forces of

the pore wall; hence temperature strongly influences micropore diffusion. The type and

distribution of the exchangeable cation will affect adsorbate diffusivity into the

micropores, along with how strongly the adsorbate molecules are attracted on to the

surface.

17
Figure 2.6 Depiction of pores and mass transfer resistances in an adsorbent particle7

18
2.3 Air Separation by Pressure Swing Adsorption

Pressure swing adsorption (PSA) is a process designed for efficient gas separation

and purification by utilizing the ability of molecules to adsorb on the surface of an

adsorbent. This study focuses only on equilibrium separation for oxygen production.

Air separation utilizing equilibrium separation is feasible because nitrogen is selectively

adsorbed over oxygen by X-type zeolites.

2.3.1 PSA Principles

The basic premise of a PSA process involves one or more columns packed with

an adsorbent (zeolite, carbon molecular sieve, etc.) that preferentially adsorbs one type of

gas molecule over another. This normally occurs at some pressure above atmospheric

pressure until the gas nearly saturates the column with the more strongly adsorbed gas

molecule (heavy component). The “raffinate” product is the gas molecule type that

adsorbs less (light component) and exits the product end of the column. Once a column

is nearly saturated with the heavy component, it needs to be removed through

regeneration (desorption) in order to reuse the column later in the process. Desorption is

critical to process efficiency. As previously mentioned, adsorption equilibrium is

determined by the gas composition, temperature, and pressure. A change in any of these

properties can be used to regenerate the adsorbent in the column. Since desorption is the

opposite of adsorption, the process is endothermic in nature. Desorption in a PSA

process occurs through a change in column pressure and composition as they provide

quicker methods of regeneration compared to a change in temperature. Desorption

occurs at either atmospheric or vacuum pressure causing the pressure to swing from high

19
pressure during adsorption to a low pressure during desorption, hence the name “Pressure

Swing”.

While most separation processes operate under steady state, a PSA process is

dynamic as conditions within the column are constantly changing at every location. The

process operates as cycles in which a column repeatedly experiences a series of

pressurization, adsorption, and regeneration steps. PSA processes ideally operate under

cyclic steady state (CSS), which occurs when column conditions at the end of each step

do not change from cycle to cycle. At CSS, the cycle performance remains the same over

time.

PSA performance is typically evaluated based on product purity, recovery, and

productivity. Adsorbent selectivity primarily determines the possible product purity.

Selectivity, s, of a zeolite (or any adsorbent) is the degree to which a zeolite can adsorb

one type of molecule over another. It is numerically described as:

where xi and yi represent the mole fractions of component i in the adsorbed and fluid

phases respectively. Selectivity depends on the temperature, pressure, type of zeolite

used and the molecules undergoing separation.

Product recovery measures the amount of desired component in the high pressure

product stream compared to that in the feed stream. Purity and recovery indirectly

measure separation efficiency and determine if a process design is capable of producing

the desired results.11 There is a trade-off between recovery and purity; i.e. high purity

20
usually results in lower recovery. Maximum potential recovery is established by the

affinity of the solid for the heavy component over the light component through

equilibrium. Recovery determines the energy efficiency of the process since it

determines how much high pressure feed is utilized per product rate.

Productivity is a measure of adsorbent utilization and is related to the speed and

size of a process. The mass transfer rate in the column primarily determines the size,

cycle time, and amount of adsorbent, all of which impact the process productivity. All of

these performance parameters are mutually dependent on each other and changing one

will impact the others.

2.3.2 History and Growth of PSA

Pressure swing adsorption is a technology that has developed primarily within the

last 50 years. It has become a valuable part of processes such as air drying, hydrogen

purification, n-paraffin removal, xylene isomer separations, and air separation.9 Several

important factors make PSA more feasible compared to other separation techniques.

First, an extra thermodynamic degree of freedom to define the adsorption process

provides greater design flexibility compared to processes like distillation, extraction, and

absorption. Second, the large number of adsorbents including zeolites, activated carbons,

silica gels etc., with different properties allow for various separations. Finally,

optimizing a process for a specific adsorbent or end product allows for creativity,

engineering, and continuous growth in PSA technology.12 For example, there are

numerous ways to design and operate an air separation process to achieve essentially

equivalent results. Improvement of PSA processes occur through either developing new

21
adsorbents (material science) or developing more efficient steps in the cycle

(engineering).11

PSA processes were originally researched for air separation in the 1950’s and

1960’s.9 Production of ultrapure oxygen through cryogenic distillation is economically

feasible only in large scale applications because of the extremely low temperature (96 K)

required. However, for smaller scale operations or when a high purity is not required,

PSA processes are more economical than cryogenic distillation. The low adsorption

selectivity of early solids rendered air separation before the 1950’s challenging.

However, with the development of synthetic zeolites (i.e. 5A, 13X), air separation was

thrust into the forefront of adsorption research.9 This lead to the development of two

different types of PSA cycles in the late 1950’s that served as the basis for later designs

of PSA systems. The first patent was for the Skarstrom cycle assigned to Esso Research

and Engineering Company.13 The other was for the Guerin-Domine cycle assigned to

L’Air Liquide.14 The regeneration method is the main difference between the cycles.8

Figure 2.7 illustrates an example of a traditional two column Skarstrom cycle,

which is the base cycle used in this study. A cycle consists of both columns experiencing

the following steps: 1) pressurization, 2) production, 3) countercurrent (opposite of feed

flow) blowdown, 4) countercurrent purge. The pressurization and production steps are

22
similar in that they both involve the input of feed gas and hence are referenced in the

future as a combined “feed” step where the flow direction is defined as the same as the

feed flow. The cycle works such that one column undergoes the feed step while the other

is counter-currently blowing down and purging. The purge step involves passing a

fraction of the purified light product through the column at low pressure after the

blowdown step. Figure 2.8 illustrates the pressure variation in the column during each

step. Skarstrom cycles usually operate their feed step at a pressure above atmospheric

pressure while the blowdown and purge steps occur at pressure slightly above

atmospheric to prevent a vacuum pump. Regeneration in a Skarstrom cycle occurs

primarily by lowering the column partial pressure of nitrogen. The blowdown step

accomplishes this by dropping the overall pressure to atmospheric pressure. The exit

stream from the blowdown step is primarily composed of the heavy component

concentration or “extract”. The purge step further lowers the nitrogen partial pressure by

countercurrently flowing part of the purified oxygen product through the column. The

purge step is perhaps the most important step because it pushes back the heavy

component concentration front towards the feed side of the column, preventing negative

effects on product purity in the next cycle. Product purity rises when the purge amount

increases, but only to a maximum purity of 96%. Raising the purge amount also has an

obvious negative impact of reducing recovery since it reduces the amount of product left

available. At a higher column operating pressure, the portion of the product gas required

for the purge step reduces because of the volume increase going from higher to lower

pressure.

23
However, higher pressure creates a greater energy requirement and eventually, any extra

adsorption that occurs does not outweigh the energy costs of the higher column pressure.

Also, higher column pressure increases the amount of raffinate product lost during the

blowdown step, which eventually limits recovery at high pressures.8 Part of designing a

PSA system deals with finding the optimum pressure ratio (ratio of adsorption pressure to

desorption pressure) and critical purge. Critical purge occurs when just enough purge gas

is taken from the product to achieve a specified product purity. This is useful information

to determine the optimal operating conditions of the process. Finding the optimal purge

amount that provides the necessary purity and highest possible recovery is a major

challenge when engineering a PSA process.

Figure 2.7 Basic two-column pressure swing adsorption system8

24
Figure 2.8 Column pressure during Skarstrom cycle8

2.3.3 Improvements to PSA Cycles

The Skarstrom process is effective for air drying, but air separation only produced

90% oxygen at a recovery of 10% with a 13X zeolite adsorbent.8 Enhancements to the

cycle were clearly necessary in order for it to become economical for air separation.

Several key improvements to the original Skarstrom cycle have allowed it to become a

viable option for smaller scale air separation operations. These improvements aim to

increase recovery, lower energy requirements, and reduce process size. They include

adding additional steps to the original 4-step cycle, operating at lower than atmospheric

pressure, and cycling faster.

One method of improving recovery uses product gas to pressurize a column after

the purge step. Often called product pressurization, this increases product purity because

the product gas helps keep the product end clean of the heavy component by pushing

back the heavy component front further towards the column feed side. A nearly full

pressurization with product gas requires a large tank, hence a full product pressurization

is not always desirable. However, a partial pressurization is still advantageous to

25
improve recovery while maintaining a high purity. The product gas used for partial

pressurization reduces the purge gas requirement and is mostly recovered during the feed

step.

Another important improvement is the addition of a pressure equalization step,

first suggested in a patent filed in 1964 by Marsh.15 The concept is intended to conserve

what is normally waste gas during the blowdown step, and use it to partially pressurize

another column before the feed step. Instead of an immediate blowdown step after the

end of the feed step, two columns are connected and pressure equalizes between the two

columns. This saves energy because the column that is partially pressurized now needs

less feed gas achieve the desired adsorption pressure. If raffinate is the desired product,

then the product side of the columns are connected so any leftover raffinate product from

the depressurizing column travels to the pressurizing column. This conserves separative

work during the following feed step, which increases recovery.11 For large capacity

operations, using pressure equalization with more than two columns further improves the

recovery and allows for a more continuous product stream flow without requiring an

extra storage tank.8, 11

Another possible improvement to the Skarstrom cycle is blowing down the

column to vacuum instead of atmospheric pressure. If the high pressure step is

atmospheric, the process is called vacuum swing adsorption (VSA). When adsorption

pressure is above atmospheric, then it is considered a hybrid PVSA process. Under

vacuum, less purge flow is required to regenerate the adsorbent. The extra energy

requirement to attain vacuum is offset to some extent by a lower adsorption pressure

requirement in the column, which makes VSA and PVSA economical in some

26
applications. Vacuum operation in air separation usually lies in the Henry’s law region

of the isotherm where pressure ratio determines the possible purity and recovery.8 Under

vacuum, adsorbent selectivity is always higher which improves recovery of the raffinate

product and up to 30% in energy savings.7

By definition, there are three ways to increase PSA process productivity. First,

employing an adsorbent with an increased capacity allows for either increased production

of raffinate or a reduction in the amount of sorbent needed in the column, both of which

will increase productivity.11 Another method is cycle time reduction. Typical industrial

PSA processes have a cycle time as long as 10 minutes. When the cycle time is lowered

to 30 seconds or below, it often referred to as rapid PSA (RPSA).

2.3.4 Air Separation Applications

The largest application of air separation is the production of 90-94% pure oxygen

gas for use in industrial processes and medical applications. The advancements discussed

earlier produced improved PSA/VSA units in the mid-1980’s and 1990’s that used five

times less adsorbent and two times less power.6 A simple two-column VSA system was

able to produce over 100 tons of oxygen per day, which provided an alternative to

cryogenic air separation. Compared to cryogenic systems with similar production, the

VSA system had a higher capital cost, but allowed for a significant energy savings.6

A significant application for air separation through small scale PSA is personal

oxygen concentrators (POC) for oxygen therapy. These oxygen concentrators utilize

rapid cycling to significantly reduce the device weight to less than 4.5 kilograms and

provide up to 6 liters per minute of oxygen. Originally POC’s were designed for use in

27
ambulances and short term travel. Currently they are found in nursing homes and for

everyday use in place of oxygen cylinders. Numerous companies have entered POC

market including Inogen One, Invacare, AirSep, and Respironics. POCs all have a

similar design because they all operate on the same engineering and design fundamentals.

The difference in performance comes from trade-offs made in the design process to

provide different specifications and characteristics that reduce the energy requirement

and target specific patient needs.16

2.3.5 Future Challenges for PSA

Modeling of a PSA process remains difficult because the complex and dynamic

nature of a PSA process; hence the design and optimization of the process remains

principally experimental in nature. Modeling the process requires numerous partial

differential equations describing the number of different process steps, initial conditions,

and boundary conditions for each step. Solving these equations is both time consuming

and difficult to do with the accuracy and reliability.12 Further adding to the problem is

the difficulty in understanding multi-component gas-solid interactions that are needed for

solving the equations in the model. Predicting these interactions from limited

experimental data is difficult because of the range of conditions (pressure, temperature,

and composition) that occur within PSA system throughout the process. Currently the

best solution is developing simplified models for predicting a PSA process and then

improving the process through a pilot plant.12 Improving the knowledge of multi-

component adsorption will remain the focus of research for years to come because of the

challenges it provides. While air separation is a fairly established PSA process, utilizing

the technology for other applications is another challenge for researchers. Current work

28
is directed at improving PSA process for CO2 capture from flue gases, olefin-paraffin

separation, and CH4-CO2 separation.7

29
 CHAPTER III

THEORY

This section introduces the theory and concepts behind gas adsorption on a surface

through a brief discussion of intermolecular forces. Using these concepts, the interaction

of oxygen and nitrogen molecules with zeolites is then explained with consideration for

how different exchangeable cations affect selectivity. Finally, isotherm models are

reviewed along with the column dynamics that occur within a PSA process.

3.1 Intermolecular Forces

Intermolecular forces were briefly discussed in the previous section. However, a

more rigorous description of these forces is required to understand zeolite selectivity. In

a PSA process for air separation, interaction of oxygen and nitrogen molecules with a

zeolite is dependent on electrostatic, induction, and dispersion forces. These forces are

reviewed to the extent necessary to understand air separation. Additional resources are

available that specifically discuss intermolecular forces in greater detail.17, 18

30
3.1.1 Electrostatic Forces

An electrostatic force is a broad term encompassing several types of

intermolecular forces. Coulomb’s law illustrates the simplest type where two ions

approximated as point charges have a force, f, between them, defined as:

where w is the magnitude of the point charges, h is the distance between the point

charges, and ε0 is the dielectric permittivity. Integration of equation 3.1 gives the

potential energy, E, between the two ions as:

where gi and gj are ionic valences of the two point charges and e is the electron charge.

Potential energy between the ions is inversely related to the distance between the ions.

When the ions are not approximated as point charges, their charge is shielded by ionic

electron clouds, which causes potential energy to vary inversely with distance (between

the ions) to a power greater than one.18

Electrostatic forces are not solely between ions, interactions between molecules

with dipole and quadrupole moments are also included in these forces. Polar molecules

exhibit dipole moments because their asymmetrical shape creates an unbalanced

distribution of electrons around the molecule, resulting in a separation of effective charge

between any two locations in the molecule. Non-polar molecules have a symmetrical

geometry, which does not produce an effective charge distribution resulting in a dipole

31
moment. Molecules may also exhibit higher order charge distribution. For example, a

quadrupole signifies an effective charge difference at four points in the molecule and an

octupole signifies an effective charge difference at eight points in the molecule.17 For

oxygen generation via air separation, the difference between quadrupoles of N2 and O2 is

exploited (see Table 3.1).

3.1.2 Induction and Dispersion Forces

Intermolecular interactions also take the form of induction forces when electrons

of a molecule become disturbed in an electrical field. The molecule (polar or non-polar)

essentially gains an instantaneous charge distribution from the electrical field of another

polar (or quadrupolar) molecule or ion in the immediate vicinity; hence the name

“induced” electrostatic interaction. This causes the molecules to experience an attractive

force with polar (or quadrupolar) molecules or ions. The ease with which electrons are

dislocated in a molecule is called polarizability, α, of the molecule. Accordingly, induced

electrostatic interactions are a strong function of polarizability.17

Finally, intermolecular interactions also occur due to instantaneous coupling of

electron clouds of two molecules approaching each other. This type of interaction, called

a dispersion force, exists for all molecules in nature, even if the molecules are non-polar

since the effective charge distribution for a non-polar molecule is only zero when

averaged over time. At any given instant, the molecule (polar or nonpolar) has a

momentary dipole moment capable of inducing an effective charge distribution on a

neighboring molecule. Hence, electrons in the interacting molecules essentially move in

tandem resulting in a net attractive (negative) potential energy. The ionization potential

32
and polarizability of the involved molecules determine the level of interaction.18

Dispersion forces, together with dipole-dipole (and higher order charge distributions),

and induction forces are collectively called van der Waals forces.

3.1.3 N2 and O2 Interaction with Zeolites

A simple way to view the interaction of oxygen and nitrogen molecules with

zeolites is to treat the exchangeable cations of zeolites as point charges.19 With this

model, the important intermolecular forces are due to interaction of the oxygen or

nitrogen quadrupole moment with the point charge and the always present dispersion

energy. When the point charge and one of the molecules are arranged linearly (most

energetically stable orientation), interaction due to the quadrupole moment depends on

1/h3, while the interaction energy due to the induced dipole depends on 1/h4. Hence,

quadrupole interaction energy is effective further away from the point charge compared

to the induced dipole interaction energy. Table 3.1 shows a significant difference

between the quadrupole moment of a nitrogen and oxygen molecule, while their

polarizability is nearly identical. A similar polarizability indicates nitrogen and oxygen

have similar induction and dispersion forces with the point charge. Therefore, a zeolite

attracts nitrogen over oxygen largely because of the quadrupole moment disparity

between them. Beyond this simple demonstration, additional factors also affect zeolite

selectivity including the relative position of the cations in the zeolite, orientation of the

nitrogen and oxygen molecules relative to the exchangeable cation, and the zeolite

structure.19 Figure 3.1 illustrates the difference in potential energy between the two

molecules as they approach the point charge (y-axis).

33
 Table 3.1 Quadrupole moment and polarizabilities of

nitrogen and oxygen molecules17, 18

Molecule M x 1040 (C-m2) α x 1025 (cm3)

O2 -1.3 16.0
N2 -5.0 17.7

h (angstroms)
Figure 3.1 Calculated potential energy curves of N2 (a) and O2 (b) in the linear

arrangement as a function of distance, h, between the point charge and midpoint of the

diatomic molecules19

34
3.1.4 Exchangeable Cations

The type of exchangeable cations in a zeolite significantly impacts its selectivity

for nitrogen. Papai et al.19 studied how different exchangeable cations affect zeolite

selectivity. Their conclusion was that substituting Li+ for Na+ increases the binding

energy for nitrogen molecules more than it does for oxygen molecules, which increases

selectivity for nitrogen. Figure 3.2 compares the potential energy curve of Li +N2 with

that of Na+N2. The potential energy difference is explained by the lack of core electrons

for Li+. The core electrons of Na+ provide a lower charge density than Li+, which also

has a smaller ionic radius. The higher charge density of Li + enhances interaction with

nitrogen molecules. A smaller ionic radius also enhances the attraction due to short range

interaction forces. The smaller cation increases potential well depth and decreases

separation between the ion and nitrogen molecule (i.e. collision diameter). Figure 3.2

compares the location of the minimum potential for Li+N2 to the minimum potential of

Na+N2.

Use of Li+ ions in zeolites was studied as early as 1964 by Mckee.20 Chao further

explored the extent of Li+ cation exchange necessary for effective nitrogen adsorption.21

His invention showed a greater Li+ ion exchange, preferably around 90%, increased

zeolite capacity and selectivity for nitrogen. He further noted that a Si/Al ratio near 1.0

significantly increased adsorption capacity and selectivity.

Ion exchange using Li+ does have some limitations. Highly Li+ exchanged

zeolites are expensive to produce since ion exchange with Li+ is less thermodynamically

favorable than with Na+ or Ca2+. However, the cost of producing highly exchanged Li-X

35
zeolites has recently been reduced and they are now considered the state of the art

material for oxygen generation via air separation for large and small scale processes.

Figure 3.2 Potential energy curves of Na+N2 (a) and Li+N2 (b).19 r(X-Y) is the distance

between the center of the positive ion and the closer end of the N2 molecule

36
3.2 Adsorption Equilibrium Modeling

Adsorption equilibrium models are highly important not only for PSA processes,

but other applications as well. Collection of equilibrium data is tedious and time

consuming; hence an accurate model avoids the need for excessive lab work. Numerous

models for both single and multi-component adsorption currently exist in literature.

Equilibrium modeling has been well studied and written about in numerous texts,

specifically ones by Crittenden & Thomas, Ruthven et al., Suzuki, and Yang.7-10 This

section provides a brief summary of equilibrium models relevant to this study starting

with the most basic and then increasing in complexity.

3.2.1 Henry’s Law

Equilibrium between a solid adsorbent and adsorbate molecules is determined

from thermodynamics. The adsorbed layer(s) is considered a separate phase in a

thermodynamic sense. At low adsorbate concentrations, the equilibrium relationship

closely resembles a linear function, which is defined as Henry’s Law:

where n* is the equilibrium amount adsorbed of component, KH is the “Henry’s Law”

constant, and p is the partial pressure of adsorbate in the gas phase. An Arrhenius

relationship is used for the temperature dependence of the Henry’s law constant.

In equation 3.4, ΔQ is the heat of adsorption, R is the ideal gas constant, K0 is related to

adsorption entropy change, and T is temperature. At low pressure as density approaches

37
zero, all adsorption systems must thermodynamically approach Henry’s law with a finite

slope at the origin. It is exact as pressure approaches zero and thermodynamically

describes the slope at the origin.

3.2.2 The Langmuir Isotherm

The simplest and most commonly used adsorption model for microporous solids

that includes a saturation capacity with a finite number of sites (each site can only

accommodate one molecule) is the Langmuir isotherm. The Langmuir model assumes

monolayer surface coverage of the adsorbate on the adsorbent surface.22 It also assumes

the surface is energetically uniform and adsorbed molecules are isolated with no

interaction forces between them. The Langmuir model has wide use for describing

dynamic equilibrium of physical adsorption systems at low surface concentrations. Many

systems show at least relative consistency with the Langmuir model. Furthermore, it

reduces to Henry’s Law at low concentrations, which is a requirement for thermodynamic

consistency in any physical adsorption system. This allows it to at least serve as a

starting point for design of PSA systems.8 Key assumptions of the Langmuir isotherm

are as follows:

1) The heat of adsorption is constant and independent of coverage (due to

assumption of isolated adsorbed molecules with no lateral interactions)

2) Only one adsorbate molecule is located at a site

3) Localized adsorption (adsorbed molecules remain adsorbed until desorption

occurs)

4) The solid possesses finite number of adsorption sites

38
 Langmuir contended the rate of adsorbate gas molecules colliding with the

adsorbent surface is proportional to the partial pressure of the gas, and the probability of

adsorption occurring is proportional to the fraction of empty sites. Additionally, he

asserted the desorption rate is directly proportional to the fraction of occupied sites.

Finally, he contended that at dynamic equilibrium, the rates of adsorption and desorption

are equal.7 For a single adsorbate system, the rate equation is represented as:

where pi is the adsorbate gas pressure, θ is the fractional surface coverage, and ka and kd

are rate constants for adsorption and desorption respectively. θ is the ratio n*/ns where ns

is the adsorption saturation capacity assuming only monolayer coverage. When ka and kd

are combined together (since they cannot be experimentally determined individually), the

Langmuir isotherm equation is represented as:

where b is the ratio ka/kd (adsorption equilibrium constant). Equation 3.6 correctly

approximates the asymptotic behavior that appears as sites fill up during adsorption for

microporous systems. The adsorption equilibrium constant, b, is often found

experimentally, and is directly related to the Henry’s constant ( =bns). Both b and

decrease as temperature increases because adsorption is exothermic. Hence, isotherms at

lower temperatures possess more curvature and appear higher than those at higher

temperatures.

39
3.2.3 Dual Site Langmuir Model

An extension of Langmuir is the “dual-site Langmuir (DSL)” model, which

accounts for an additional adsorption site not present in the Langmuir model.23 With two

adsorption “sites”, the DSL model has greater mathematical flexibility and can represent

almost any pure component data. It represents zeolitic systems very well because the two

sites capture the essence of the two types of interactions with the surface. One site

represents the electrostatic interaction with exchangeable cations, while the other site

represents the much smaller difference in the attraction of nitrogen and oxygen molecules

to the zeolite structure due to dispersion forces. In essence, it is rationalized there is one

site representing the electrostatic and induced electrostatic interactions with the

exchangeable cations, and another site representing dispersion interactions with the

zeolite structure. The DSL model for a pure component gas is expressed as:

where and b are the saturation capacity and equilibrium constant respectively for the

exchangeable ion sites while and d are the same parameters due to the dispersion

interaction effect. It is assumed b is greater than d because exchangeable ions have

higher affinity for adsorbate molecules than the neutral zeolite structure. If the saturation

capacity of each gas molecule is assumed equal for each site, the DSL multicomponent

model is represented as:

40
The equilibrium constants assume the usual Arrhenius temperature dependencies:

where and are the two pre-exponential constants for gas i, and and are

the heat of adsorption of gas i on the two sites (or vertical interaction energy). Mathias et

al. have demonstrated the accuracy of the DSL model for nitrogen/oxygen mixtures and

zeolite 5A.23 LiLSX is similar enough to zeolite 5A to justify using DSL to describe

binary equilibrium for this dissertation.

3.3 Adsorption Column Dynamics

The ultimate separation efficiency is determined by adsorption equilibrium, which

takes infinite time to achieve. Yet processes operate within a finite time frame.

Adsorption kinetics determines the approach to equilibrium at any location in a column at

a given time. This section reviews how adsorption kinetics affects column dynamics. An

abundance of literature exists that provides far greater detail on this subject, specifically

ones written by, Yang, Motoyuki, and Ruthven.9, 10, 24

3.3.1 Adsorption Rate

An axially dispersed plug flow model is used in PSA modeling as the material

balance for each component in a packed adsorption column.

41
The four terms in the model describe the following phenomena in order: 1) axial

dispersion 2) convective flow 3) gas phase accumulation 4) adsorption rate (i.e. solid

phase accumulation). The adsorption rate is often described using a linear driving force

(LDF) model expressed as: 25

where ni* is the equilibrium value of component i in the adsorbed phase at a given fluid

phase concentration, and ni is the adsorbed phase concentration of component i, both

averaged over an adsorbent particle. The mass transfer coefficient (MTC), ki, of a

spherical particle is typically approximated as:24

where kf, is the film transfer coefficient, rp and rc are the particle and crystal radius

respectivley, Dp and Dc are the macropore and micropore diffusivity of component i

respectively, εp is the particle macroporosity, and K is the dimensionless equilibrium

constant related to the Henry constant, KH, by K= KHRTρp. The three terms on the right

side of the equation account for resistance to mass transfer in the film, macropore and

micropore respectively. Equation 3.13 is defined from a flux equivalency through the

three resistances in series and is exact for linear isotherms in the Henry’s law region.

While only strictly applicable to linear isotherms, it is a useful approximation for non-

linear isotherms, especially to understand the relative importance of different resistances.

If the macropore term is considered controlling, equation 3.13 simplifies to:

42
The MTC for a zeolite adsorbent particle typically assumes macropore diffusional

resistance as the limiting resistance.2 Under this assumption, equation 3.14 is used to

estimate ki either using literature values to estimate Dp, or through a combination of

breakthrough and numerical experiments.

3.3.2 Axial Dispersion

Axial dispersion acts to disperse the concentration wave in packed column

operations, just like mass transfer resistances associated with the particle. At low gas

velocity, it is caused by molecular diffusion, while at high gas velocities, it is caused by

eddy turbulence. In PSA models, the combined effects of axial dispersion are

approximated with the axial dispersion coefficient, DL, often expressed as: 24

where Dm is molecular gas diffusivity, u is interstitial velocity, dp is particle diameter and

γ1 and γ2 are constants. The first term is the molecular diffusion contribution to

concentration gradient spreading and the second term is the eddy turbulence contribution.

The constant γ1 is the inverse of the bed tortuosity factor and 0.7 is a well-accepted

estimation for columns of non-adsorbing particles.24 The γ2 constant is the inverse of the

limiting Péclet number (Pé∞), which is the value Pé (udp/DL) approaches at high gas

velocity. The common estimation of 0.5 comes from the observation that large particles

43
approach a maximum Pé number of two at high Reynolds numbers, common in large

PSA applications.26

3.3.3 Mass Transfer Zone

In an adsorption process, a feed mixture is introduced at one end of the column

and product exits at the other end. As the adsorbate moves from the fluid phase into the

adsorbed phase, a concentration wave forms in the column. The concentration wave, or

MTZ, travels through the column and eventually reaches the opposite end, illustrated in

Figure 3.3. As it exits the column, the adsorbate outlet concentration increases and

eventually reaches the inlet concentration. The shape of the exit (“breakthrough”) curve

is heavily dependent on the adsorption isotherm shape and whether equilibrium is

favorable or unfavorable. Figure 3.4 illustrates how to determine adsorption isotherm

favorability. If favorable, the concentration wave is compressive or self-sharpening

(Figure 3.5), since the high adsorbate concentration front travels faster than the low

adsorbate concentration front because of a material balance in the MTZ. The limit of this

wave is a shock wave or a step discontinuity, which cannot occur in reality because

adsorption is not instantaneous. A sharp wave front produces a small MTZ and indicates

a high adsorption rate and separation efficiency. An unfavorable isotherm produces a

dispersive wavefront, the opposite of a compressive front.9 This occurs during the

desorption step of a PSA process and is illustrated in Figure 3.5. In realistic adsorption

processes, mass and heat transfer limitations will spread the MTZ. Mass transfer

limitations include diffusional and axial dispersion, as previously discussed. Since

adsorption is exothermic and finite heat transfer resistances exist in any packed column,

44
the adsorbent temperature will vary across the MTZ, which also causes additional

spreading.

Breakthrough experiments are often used to determine the adsorption rate in an

adsorbent. This test involves inducing a step change in the inlet concentration to a

packed column and monitoring the column exit concentration over time. Eventually the

MTZ “breaks through” and forms a breakthrough curve as it exits the column, illustrated

in Figure 3.6. In essence, a breakthrough curve is mirror image of the MTZ in the

column. The shaded region between point b and point c represent the unused portion of

the column. If mass and heat transfer resistances are small and the isotherm is favorable,

the breakthrough curve is steep and a sharp MTZ leads to a small length of unused

column.

Figure 3.3 Mass transfer zone inside a column. c0 is the inlet concentration, ce is the exit

concentration, and L is the column length7

45
Figure 3.4 Shapes of isotherms in determining the sharpness of concentration

wavefronts: (A) linear; (B) favorable; (C); unfavorable; (D) irreversible. q* =

equilibrium amount adsorbed and C = concentration of adsorbate in the gas phase9

Figure 3.5 Self-sharpening wavefront (top) and dispersive wavefront (bottom). c is

adsorbate concentration in the gas phase and z is distance from the column entrance8

46
Figure 3.6 Column breakthrough curve. c0 is the inlet concentration, ce is the exit

concentration, tb represents the time breakthrough occurs, ts is the time stoichiometric

center reaches the end of the column, to is the end of the MTZ 7

47
 CHAPTER IV

CHALLENGES OF SMALL SCALE PSA

This chapter demonstrates the key differences between PSA on a large and small scale. It

also provides background on the current state of the literature in this research area and

discusses the challenges in understanding the limitations of small scale processes.

Finally, it highlights current knowledge gaps in literature to demonstrate how this

dissertation fits into the field of PSA.

4.1 Effect of Particle Size Reduction in Small Scale PSA

Decreasing adsorbent particle size in packed columns is known to increase axial

dispersion effects.26 However, the extent to which it affects the mass transfer rate in PSA

processes remains uncertain. Knowledge of the limiting mass transfer resistance in a

PSA process is important for process design. Furthermore, PSA simulation models rely

on assumptions about mass transfer rate limitations for an adsorption rate approximation.

Hence, improving simulation accuracy depends on properly estimating relevant mass

transfer limitations in a PSA process.

48
4.1.1 Current Assumptions

Large diameter zeolite particles (> 2 mm) are well known to exhibit significant

mass transfer resistance in their macropores.2, 27

Since it represents the dominant

resistance in columns packed with these particles, a mass transfer rate controlled by

macropore diffusional resistance is typically assumed. Other resistances external to the

particle such as axial dispersion and film transfer resistance are usually considered small

or negligible. These assumptions generally provide a reasonable estimation of mass

transfer limitations for simulations involving large zeolite particles.

Current small scale PSA processes often operate with small particles (~ 0.3 mm to

0.7 mm in diameter) to permit rapid cycling. Smaller particles provide both a higher

surface area for diffusional flux into the particle and a shorter diffusional path inside the

particle. Decreasing macropore diffusional resistance through particle size reduction

increases the influence of other factors affecting the mass transfer rate, particularly

effects external to the particle. While this notion is generally accepted in literature, it

remains unclear at what particle size these external factors become significant. Zhong et

al.28 used a simulation study to determine the effect of LiLSX particle size on process

performance. They found axial dispersion effects begin to dominate the mass transfer

rate at a particle size < 1 mm for velocities often used in small scale PSA processes.

However, the study was never confirmed experimentally. Alpay et al.29 experimentally

studied the effect of 5A particle size on a single bed air separation process and found an

optimal particle size to maximize product purity. They also found axial dispersion effects

were more significant for small particles than predicted by common correlations. Rao

and Farooq30 came to a similar conclusion about axial dispersion effects for their single
49
column small scale PSA process using very small 5A zeolite particles (63-75 μm in

diameter). Other studies on packed columns of non-adsorbing particles further confirm

that as particle size decreases, the influence of axial dispersion increases significantly.26,31

Other resistances also require consideration when using small particles. While the

contribution of a film resistance is typically insignificant for large zeolite particles, this

assumption also requires confirmation for small particles. Furthermore, it was recently

suggested an additional skin resistance plays a prominent role in the mass transfer rate

with small LiLSX particles.32 Skin resistance is attributed to a crystal density increase at

the particle surface, which limits diffusion in the outer portion of the particle. A skin

may result from shaping methods used during particle manufacture. Further evaluation

of the mass transfer rate in columns of these particles is needed to confirm if this applies

to all LiLSX particles.

Axial dispersion effects may be overlooked when using a smaller particle size

since particles in large scale adsorption processes rarely approach the size where they

become significant. A rise in numerical and experimental studies utilizing small

particles28, 33-38 increases the need to determine the controlling mass transfer resistance in

these processes. It will also provide a better understanding of how to improve particle

manufacturing methods. Recent advances in adsorbent production techniques target

reducing the macropore resistance contribution. For large particles, pore diffusion has

been demonstrated to improve significantly through a reduction in binder content using

caustic digestion or by incorporating alternative binders.39-42 However, these same

methodologies may not have the same effect on a process where macropore diffusional

resistance no longer controls the mass transfer rate.

50
4.1.2 MTC Estimation for Small Particles

The MTC for large particles is typically found using equation 3.14, first

introduced in section 3.3.1.24

While equation 3.14 indicates a significant adsorption rate increase is possible for small

particles, it fails to account for the effect of particle size on factors external to the

particle. Axial dispersion is naturally not in the adsorption rate equation (equation 3.12)

as its effect is explicitly included in the material balance in equation 3.11. However, its

effect on column dynamics is similar to diffusional limitations in the particle since it acts

to disperse the MTZ. For a linear isotherm, axial dispersion effects are approximately

combined with resistances associated with the particle as described in equation 4.1,

sometimes referred to as the linear addition approximation for a bimodal pore size

distribution.24

In equation 4.1, the first term on the right side of the equation accounts for the resistance

equivalent of axial dispersion effects and is linearly added to macropore, film, and

micropore resistance respectively. All four effects on MTZ stretching are clearly

indicated, which makes it useful to compare resistances and their dependency on gas

velocity (u). Micropore diffusional resistance is usually negligible compared to other

contributions for type-1 adsorption in zeolites,2, 27 which was confirmed by Wu et al.32 for

51
LiLSX particles; hence, it was considered insignificant for the analysis in this study.

When axial dispersion effects are described through an overall MTC in the adsorption

rate term in equation 3.12, a plug flow model replaces the axially dispersed plug flow

model in equation 3.11. The linear addition approximation is useful when the axial

dispersion term is significant in comparison to the other terms in equation 4.1. It has

been demonstrated to provide a suitable MTC estimate even for highly non-linear

systems.24, 43

For zeolites, the macropore resistance term in equation 4.1 is typically assumed to

dominate koverall. However, for small particles, it has been demonstrated the axial

dispersion and pore diffusion rate constants in equation 4.1 become comparable.28, 44

Furthermore, Wu et al.32 measured koverall experimentally for nitrogen on ~0.5 mm in

diameter LiLSX particles and found a much smaller koverall increase than expected based

on particle size reduction. While the increase in koverall was high enough to provide the

increase in productivity necessary for oxygen generation using small scale PSA, an

additional resistance is clearly present that is not accounted for using traditional

correlations. The macropore term only accounts for ~1.4% of the overall resistance at a

pressure of 2 atm and temperature of 303.1 K, while axial dispersion (24.8%) and film

resistance (11.4%) play much larger roles. The authors suggest the remainder of the

resistance (62.3%) is due to a skin. However, as demonstrated in a later section, using a

literature correlation that properly accounts for higher axial dispersion effects in columns

of small particles also explains the difference.

52
 4.1.3 Axial Dispersion Coefficient Estimation

No consensus exists in literature for describing axial dispersion effects in columns

packed with small particles as estimates of the axial dispersion coefficient (DL) vary

depending on the correlation used. Some studies have continued to use a correlation

common for large particles.34, 45-47

Others have incorporated an alternative correlation

that increases axial dispersion effects for small particles.30, 38

It is currently unknown

how the estimate of DL affects simulation accuracy, which further contributes to a lack of

understanding of its importance.

As discussed in section 3.3.2, equation 3.15 is typically used to estimate the axial

dispersion coefficient, DL, with γ1 and γ2 assumed as 0.7 and 0.5 respectively.

While these estimates of γ1 and γ2 are suitable for columns of large particles, their

applicability to columns of small particles is questionable. As early as 1968, Edwards and

Richardson48 demonstrated the Pé∞ for small particles (< 2 mm) was much lower than 2

(Pé∞ = 1/ γ2), which requires γ2 to be much higher than 0.5. Suzuki and Smith31 further

confirmed this using small glass beads to determine the effect of particle size on

dispersion. They found a clear change in Pé∞ as particle diameter decreased for small

particles. Langer et al.26 also found a similar relationship, illustrated in Figure 4.1.

Although the reason Pé∞ is a function of particle size for small particles is not fully

understood,24 Moulijn and Van Swaaij49 reason the difference is due to channeling within

the column (rather than just near the wall). They postulate that small particles tend to

53
form agglomerates, which act as larger particles when the strong interaction forces

between the particles overcome the forces of gravity and drag.

It has also been suggested the estimate of γ1 in equation 3.15 is much larger (~50

for a rectangular isotherm) for strongly adsorbing particles. The proposed higher value

accounts for direct transport through the particle due to an asymmetric concentration

profile surrounding the particle at low gas velocity.50 As a result, the amount of effective

axial dispersion increases in the laminar and transitional flow regimes. While this

apparently has an impact in liquid chromatography where the influence of surface

diffusion is significant at low velocity,51 it has not been demonstrated to be applicable for

gas adsorption systems. The value of ~ 0.7 for γ1 currently remains the best estimate for

such systems.

Figure 4.1 Limiting Peclet number vs. particle diameter for flow through packed beds26

54
4.1.4 Impact of Higher Axial Dispersion Effects with Small Particles

The estimate of DL is much higher for small particles than large particles if the

estimate of γ2 is much larger than 0.5. Since the traditional method of matching simulated

breakthrough curves to experimental data relies on a reasonable estimate of DL to

determine Dp experimentally, an artificially low estimate of Dp is possible if DL is not

estimated correctly. Furthermore, a higher DL increases the contribution of axial

dispersion in equation 4.1 while the particle size decrease reduces the contribution of

macropore diffusional resistance. Hence, a macropore resistance limited system may no

longer be a suitable assumption.

To better understand how significant the effects of axial dispersion might be, a

case study was used to compare different mass transfer rate assumptions using a 0.5 mm

particle. In case 1, film resistance was considered negligible and DL was approximated

using equation 3.15 with γ1 and γ2 assumed as 0.7 and 0.5 respectively. In case 2, DL was

estimated with the same equation, but now the effect of film resistance was added

according to equation 4.2.24

In case 3, film resistance was considered negligible and γ2 was estimated using equation

4.3:26

55
where dp is in cm. Finally, case 4 used the same correlations as case 3 to determine DL,

but now film resistance was included according to equation 4.2. In all cases, Dp for

nitrogen was found from equation 4.4 using typical literature correlations for Dm and

Dk.24

A pore tortuosity (τp) of 3 was assumed since studies on larger LiLSX particles have

estimated it between 2 and 4.28, 39, 42 The dimensionless Henry’s law coefficient (K) was

found from previously determined DSL isotherm paramters52 assuming conditions of T =

313 K, P = 267 kPa, and a gas composition of 0.75/0.25 N2/O2 so the predictions are

comparable to future experiments in this study. Table 4.1 summarizes the constants and

correlations used in all four cases.

Figure 4.2 demonstrates the koverall varies significantly depending on the

assumptions made. As velocity increases, the controlling mass transfer limitation

switches from axial dispersion control to macropore diffusion control. This is illustrated

by the approach of koverall in Figure 4.2 to the horizontal “macropore only” line. As the

koverall approaches the horizontal line, the macropore term in equation 4.1 becomes

controlling. When koverall is far from the horizontal line, axial dispersion effects are

controlling. It is evident the inclusion of film resistance has a noticeable impact on koverall

if axial dispersion effects are low (Case 1 and 2). However, a higher axial dispersion

coefficient has a much greater effect on koverall, especially at lower Reynolds numbers

(Case 3 and 4) Small scale PSA processes often operate at low to moderate Reynolds

56
numbers where the difference between the predictions is the greatest. Additionally,

breakthrough experiments using small particles are often conducted at low Reynolds

numbers since the response time of gas analyzers make high velocity waves difficult to

accurately measure. Since such a difference in koverall exists depending on the correlation

used to estimate DL, it is critical to determine which one is more appropriate for use in

process simulators.

Table 4.1 Comparison of parameters used for estimates of koverall in Figure 4.2. Pressure

= 267 kPa dp = 0.05 cm, Dp = 0.02 cm2/s,

K=16.5

γ1 of DL γ2 of DL kf

Case 1 0.7 0.5 ∞

Case 2 0.7 0.5 Eq. 4.2

Case 3 0.7 3 ∞

Case 4 0.7 3 Eq. 4.2

57
Figure 4.2 Comparison of predicted overall nitrogen MTC using constants and

correlations according to Table 4.1. Overall gas pressure = 267 kPa, gas composition

(N2/O2) = 75/25

4.2 Small Scale PSA

Small scale PSA is an advancement of PSA technology for the purpose of process

size reduction. It is typically distinguished from traditional PSA by rapid cycling (< 30

seconds) of small adsorbent particles (< 1 mm). Larger scale PSA typically operates with

cycle times on the order of minutes and a particle size > 1.5mm. Cycle time reduction

requires an increase in gas velocity for a given amount of adsorbent. This raises several

issues such as sorbent fluidization, increase in column pressure drop, and higher gas

dispersion.12 In addition, a higher gas velocity reduces contact time with the solid,

necessitating smaller adsorbent particles (<1mm). For a finite adsorption rate, the result

58
is a stretched mass transfer zone (MTZ), which causes premature breakthrough and a

reduction in product recovery. However, provided the productivity increase is greater

compared to the loss in column working capacity, process size reduction is possible with

higher cycling frequency.11 This is ideal for certain applications like a POC. Small scale

processes typically increase solid productivity with faster cycling. The tradeoff is a lower

recovery, which leads to higher overall power consumption for a desired product flow

rate.53 Furthermore, it reduces the solid working capacity, which limits the achievable

process productivity.44

4.2.1 Evolution of Small Scale PSA

Small scale PSA originally was called rapid PSA (RPSA) and involved a single

column process using very small particles and high gas velocities to create a large

pressure gradient across the column.54, 55

This permitted a simple 2-step process with

column self-purging and a continuous product flow rate at one end of the column, while

swinging pressure on the other end. A small separation unit was possible by operating

with a high cyclic frequency to reduce the amount of solid adsorbent. Parameters such as

an optimal particle size and benefits of an additional delay step have been studied for the

application of air separation.29

POCs have recently renewed interest in small scale, fast cycling PSA, also often

labeled RPSA. However, these systems operate with 4-step Skarstrom cycles, more

similar to larger scale processes than the original RPSA definition given above. To

facilitate a portable process unit, column length is much shorter (~10-20 cm) compared to

the traditional RPSA concept (>100 cm). Even with short cycle times, pressure drop

across columns of this length is not high enough to permit RPSA operation as it was

59
originally designed. These systems operate with distinct steps of constant pressure or

pressure change in the column.

4.2.2 Limitations of Small Scale PSA

Experimental small scale PSA studies for POC applications often focus on how

purge to feed ratio, product purity, and adsorbent type affect process performance.36, 56, 57

Other studies have numerically investigated the effect of rapid cycling on a specific PSA

step.45, 46 Limitations of small scale systems have received less attention. One method of

studying the limits of small scale PSA entails increasing cyclic frequency while keeping

other process parameters (i.e. pressure ratio, product purity) constant. A cyclic frequency

limit is represented by a minimum bed size factor (BSF) at a specific cycle time. Since

BSF is inversely related to productivity, a minimum represents an adsorbent utilization

limit and defines a lower limit on operating speed; hence the minimum size of an oxygen

concentrator.

Several recent experimental studies35, 38 have observed an adsorption utilization

limit for their small scale processes. While the exact cause of a minimum remains

unclear, it has been suggested the combined effects of pressure drop, mass/heat transfer

resistances, and non-isothermal operation eventually cause it to occur at a specific

cycling rate.58 A recent numerical study demonstrated how the cumulative effect of

mass, heat, and momentum resistances affected process performance using a 5A zeolite

to separate nitrogen and helium.47 The study demonstrates non-isothermal operation and

mass transfer resistances are the most significant contributions to a minimum BSF, while

pressure drop and finite heat transfer resistance represent much smaller contributions.

60
However, the authors mention these conclusions will not necessarily apply to other

systems since they depend on adsorption properties and process design. Another

numerical study reached an alternative conclusion suggesting system dynamics (i.e.

fastest achievable cycle) limited adsorption utilization in their process.34

The limited experimental data studying limitations of small scale processes

largely involve single column concepts.30, 35, 38, 59 While some advantage is gained from a

single column setup with respect to equipment size, multiple column systems are much

more power efficient. An additional column(s) permits a pressure equalization step,

which significantly increases product recovery; hence reducing the power requirement

and battery size for a given production rate. This study seeks to determine if a minimum

BSF also exists for a dual bed small scale PSA process.

4.3 Possible Causes of a Minimum BSF

Increasing process speed reduces the solid inventory necessary for a desired

product flow rate. While this decreases the BSF, the tradeoff is a loss in column working

capacity. With increasing cyclic frequency, working capacity loss eventually overcomes

the productivity gain of rapid cycling, causing the BSF to go through a minimum. While

this minimum is generally attributed to non-idealities mentioned in the previous section

(i.e. pressure drop, non-isothermal operation, mass transfer resistances), an advantage in

process design will be gained if a specific cause is identified.

4.3.1 Pressure Drop

A common concern when using small particles is column pressure drop will

reduce process performance. However, the extent of this performance decline is difficult

61
to quantify. While it has been recognized that the importance of pressure drop increases

under faster cycling conditions,60-63 it has yet to be determined if column pressure drop

impacts small scale PSA performance enough to justify considering it in process design.

Pressure drop can reduce product recovery through the premature breakthrough of a

stretched MTZ.60, 64 It may also reduce recovery through a higher purge step operating

pressure, resulting in a larger loss in purge gas during this step.45, 65 The reason for a

reduction in working capacity is more intuitive. Column pressure drop reduces the

adsorption step nitrogen loading at the product end of the column. During the

regeneration steps, the nitrogen unloading at the product end of the column is also

reduced. Additionally, the pressure drop during regeneration steps is a larger fraction of

absolute pressure; hence properties in the column vary more substantially from one end

to the other during these steps.

The previously cited studies largely focus on pressure drop effects during a

specific step of a PSA cycle and not on the cumulative effect on process performance. A

better understanding on how column pressure drop affects overall process operation is

needed to better design small scale PSA processes. Several computational studies have

attempted to accomplish this. A recent study provided an individual and cumulative

assessment of how various mass, heat, and momentum resistances affect product

recovery and productivity of a PSA process separating a mixture of nitrogen and

helium.47 It demonstrated pressure drop had a small effect on recovery and productivity

at cycle times under 4 seconds. Yang et al.66 also used a simulated multi-bed PSA

process to show pressure drop has only a small influence on the performance of a PSA

process separating a H2/CO mixture. While these computational efforts indicate pressure

62
drop has a limited effect on process performance, experimental confirmation of this

notion is lacking.

4.3.2 Heat Transfer Limitations

Heat transfer is significant in PSA processes since adsorption is exothermic and

desorption is endothermic. Various heat transfer resistances in the column increase or

decrease the adsorbent temperature relative to isothermal conditions. Heat transfer from

the gas phase to the column wall and from the column wall to the ambient environment

may be important for small columns, but their effect is reduced greatly as column

diameter increases. Hence, large scale columns typically operate adiabatically while

small scale columns may approach isothermal operation. However, unless working with

dilute concentrations and a low heat of adsorption, even small scale processes do not

approach isothermal operation.24 Heat transfer between the particle and flowing gas is

quite fast according to typical Nusselt number correlations and is generally considered

instantaneous.

4.3.3. Mass Transfer Limitations

Mass transfer limitations that would affect a minimum BSF have already been

discussed in previous sections. The primary potential contributors are axial dispersion

effects and mass transfer resistance in the macropores of the adsorbent. Other potential

limitations include film and micropore diffusional resistance and the formation of a skin

on the adsorbent particle.

63
4.3.4 BSF Model

Rezaei and Webley44 previously developed a relationship between productivity

and cycle time that accounts for various non-idealities:

where K is an equilibrium constant (i.e. Henry constant for a linear system), ΔP is column

pressure drop, WCideal is isothermal working capacity, L is column length, LMTZ is the

mass transfer zone (MTZ) length, and U is gas velocity. Although their application was

comparing structured and non-structured adsorbents, the relationship also helps explain

why a minimum BSF exists in small scale PSA as well. The expression relates working

capacity to productivity (or BSF) and U through a proportionality constant k’. ΔP and

LMTZ both increase as a function of gas velocity (U ∝ 1/cycle time) and reduce isothermal

working capacity. Pressure drop reduces working capacity because the adsorbent at one

end of the column no longer experiences the same swing in pressure as the other end. A

longer MTZ increases the unused column length, which also limits available working

capacity.

Equation 4.5 exhibits a minimum BSF as gas velocity increases. The velocity and

absolute value of the minimum BSF depends on the extent to which pressure drop and

MTZ spreading contribute to working capacity reduction. At the limit of isothermal

operation and local equilibrium, the MTZ length is not a function of velocity and pressure

drop determines the velocity and depth of the minimum BSF. When mass and heat

transfer limitations are significant, the MTZ length is a function of velocity and a

minimum may develop even if pressure drop is negligible.

64
4.4 Study Aims and Experimental Plan Overview

4.4.1 Study Aims

It is evident from the literature discussed earlier that axial dispersion effects are

more significant for the particle size used in small scale PSA processes. However, there

are no current studies that demonstrate how much his affects the adsorption rate of small

LiLSX particles. Since the adsorption rate significantly affects process performance,

properly estimating it for process simulators is critical. The first aim of this study is to

fill this knowledge gap in literature. Understanding the impact of axial dispersion effects

on small scale PSA process design will also help fulfill the other aims in this study.

While small scale PSA processes operate in a similar fashion to larger scale

processes, several key differences exist including a much smaller particle size, shorter

column length, and a faster cycling rate. These differences increase the complexity of

understanding the limits of small scale PSA processes. While single column studies have

demonstrated a minimum BSF exists, it has never been demonstrated for a dual-column

process. Furthermore, a true understanding of why the minimum occurs is still lacking.

The second main aim of the study is to increase understanding in this area by measuring a

minimum BSF for a two-column process. The final aim of the study is then to determine

what causes a minimum BSF. Through a better understanding of small scale PSA

process limitations, it is hoped the design of these processes may be improved in the

future.

65
4.4.2 Overview of Experimental Plan

To fulfill the aims of this study, three primary studies on two different

experimental systems were conducted. Axial dispersion effects are best evaluated using a

breakthrough system, explained later in Chapter 5, rather than an experimental PSA

system. This study is hereafter referred to as the “breakthrough study”. Equation 4.5

indicates that pressure drop is a potential limitation on adsorbent utilization; hence, an

experimental study was designed to better understand the effect it has on PSA

performance. A small scale PSA process was used to measure process performance at

different levels of pressure drop. This study is referred to hereafter as the “pressure drop

study”. Finally, in order to understand what causes a minimum BSF, it needed to be

measured experimentally. The final study focuses on experimentally measuring a

minimum BSF using the same two column, small scale PSA process as the pressure drop

study. This study is referred to hereafter as the “minimum BSF study”.

66
 CHAPTER V

MATERIALS AND METHODS

This chapter introduces the type of zeolite, instruments, and devices utilized to conduct

the experiments for this dissertation. Data collection methods for breakthrough and PSA

experiments are also discussed in detail along with the conditions of the experiments.

Finally, the dynamic PSA model used in this dissertation to support and analyze data is

introduced.

5.1 Type of Zeolite

The zeolite in this dissertation was a LiLSX zeolite with a Si/Al ratio near 1.0, a

high Li+ exchange, and manufactured to increase the adsorption rate, specifically through

a particle size ~ 0.5 mm in diameter and a proprietary clay binder.

67
5.2 Isotherms

Process work requires knowledge of the adsorbent equilibrium properties.

Equilibrium data was collected volumetrically by another graduate student in our lab

(Qian Qian Zhou) at three different temperatures for both pure oxygen and nitrogen. The

data along with the DSL fits are shown in Figure 5.1 and 5.2. The DSL parameters

regressed from the data are shown in Table 5.1.

Figure 5.1. Oxygen isotherms for LiLSX zeolite.67

68
Figure 5.2. Nitrogen isotherms for LiLSX zeolite.67

Table 5.1 Nitrogen and oxygen DSL parameters for LiLSX particles used in

simulations.67

Nitrogen Oxygen

ns,1 (mol/kg) 2.025 1.426

b0 x 10-6 (1/kPa) 0.124 0.158

Qb /R (K) 3365 2373

ns,2 (mol/kg) 2.402 6.177

d0 x 10-6 (1/kPa) 0.513 3.386

Qd /R (K) 2263 1211

69
5.3 Column Packing and Activation

Any adsorption system requires a careful column packing to ensure the amount of

zeolite in the column is accurately known. Future process design calculations are highly

dependent on these values since the amount of processed gas is dependent on adsorption

capacity. The packing procedure was similar for all columns in this dissertation.

Initially, a column was packed with a known amount of zeolite prior to activation.

Activation involves heating the adsorbent to eliminate contaminants such as water and

carbon dioxide that would reduce the selectivity and capacity. Column activation was

performed in a furnace under vacuum and low helium flow. The temperature was

increased 1 °C every minute until the temperature reached 120 °C. The furnace was then

held at this temperature for an hour before again increasing 1 °C every minute until the

temperature reached 345 °C, at which the temperature was held for at least seven hours.

The column in the furnace was then pressurized with helium to atmospheric pressure and

allowed to cool to room temperature. Lastly, the column was weighed to determine the

amount of contaminants lost during activation so a “dry” zeolite content could be

determined before installation into the PSA or breakthrough system. The weight of the

left over helium gas in the column was neglected in this calculation.

5.4 Breakthrough Study

5.4.1 Breakthrough system

Measuring the MTZ length and kinetic parameters of the LiLSX adsorbent was

completed using a breakthrough column. The breakthrough system used in this

70
dissertation is shown in Figure 5.3. It is a simple apparatus with a column filled with

spherical particles of LiLSX. The column diameter was chosen to keep the column to

particle diameter ratio above the commonly accepted threshold of 20 to prevent wall

effects.68 Other relevant column specifications are listed in Table 5.2. Specifics related

to instruments and operating details may be found elsewhere.69

5.4.2 Breakthrough Measurements

Breakthrough experiments were conducted by another graduate student in our

lab (Mihir Patel), however the data analysis was primarily done by the author of this

dissertation. Pure nitrogen (99.999% purity) and oxygen (99.98% purity) was supplied

by Matheson Gas Products. The outlet gas stream of the column was monitored by a

Hiden Analytical process mass spectrometer with a response time ~ 300 ms, which

detected the molecular ions of O2 and N2. The column was initially saturated at room

temperature with pure oxygen at 267 kPa before introducing a ~75/25 mixture of nitrogen

and oxygen respectively at the same pressure. The pressure was chosen in the range of

the adsorption step of typical vacuum PSA processes. Experiments were run until the

outlet concentration reached the inlet concentration at a range of gas velocities within the

limits of the system. The effect of external void volume after the column was determined

to be negligible in the range of the experiments run. Further discussion on this topic is

available in Appendix B.

71
Figure 5.3 Breakthrough experimental setup. Note: adsorption column includes a

thermocouple at the exit of the column for monitoring exit gas temperature

Table 5.2 Breakthrough column specifications

diameter 1.09 cm

length 15 cm

LiLSX zeolite 8.64 g

avg. spherical particle size 0.05 cm

bulk density 0.62 g/cm3

εb 0.38

εp 0.35

72
 Most breakthrough experiments in literature are performed by an adsorbable

component displacing helium. Two reasons are usually cited for this choice: 1) the

resulting data can be integrated to determine equilibrium information (as a point on an

isotherm). Since we measured pure and binary equilibrium data for N2/O2 in a separate

volumetric instrument, there was no need to measure equilibrium data with this

apparatus; 2) From a kinetics stand point, which is the main reason for performing

breakthrough experiments, helium is so small and light that it offers little resistance to N2

diffusion in the pores, hence a “true” nitrogen mobility is measured experimentally.

From a purely scientific point of view, this is correct; however we are trying to generate

correlations for engineering use in process simulators where N2 is displacing O2. If a

process model is going to use only one MTC for N2 (irrespective of what the other

component is), experiments to determine the MTC should be closer to reality, which

requires N2 to displace O2. A similar procedure was used in a previous study on extruded

mordenite pellets.70 Using O2 as the initial condition has an additional advantage of

limiting the inevitable temperature increase in the column during breakthrough

experiments. This is simply because of the heat removed through the desorption of O2.

5.4.3 Constant Pattern

Breakthrough analysis in this dissertation relies on the assumption of a constant

pattern forming in the column during the experiment. A constant pattern forms for any

favorable system, such as the one in this dissertation (N2-LiLSX). More detail on why

this occurs is provided in section 3.3.3. A constant pattern arises when the concentration

profile in the MTZ does not change as the MTZ travels through the column. It typically

develops a short length into the column, sometimes referred to as the “entrance length”.

73
The column length was sufficiently long such that entrance length had a negligible

impact on the breakthrough results.

5.4.4 MTC Calculation

In this study, breakthrough experiments were used to measure a MTC at a fixed

pressure and room temperature while varying the velocity. Since axial dispersion and the

film transfer coefficient both depend on velocity while intraparticle transport mechanisms

(i.e. macropore/micropore diffusion and surface diffusion) do not, this provides a

convenient method of determining the significance of diffusional limitations external to

the particle. The overall MTC was estimated from experimental breakthrough

experiments by integrating equation 3.12 between the limits defined as the MTZ.32 This

gives equation 5.1:

where t0.9 and t0.05 is the experimental time the dimensionless outlet nitrogen

concentration, c/c0, reaches 0.9 and 0.05 respectively, c0 is the inlet nitrogen

concentration, nN2 is the average adsorbed phase concentration of nitrogen over an

adsorbent particle, and nN2* is the equilibrium value of the adsorbed phase concentration

of nitrogen. Defining the MTZ in this manner is common practice since it is often

difficult to detect where the MTZ exactly starts and stops, especially at higher gas

velocities. It additionally avoids the long tail caused by deviation from isothermal

operation due to temperature effects. Since the experiments were run at ambient

conditions, some temperature rise in the column is inevitable due to the heat of

74
adsorption. For a nitrogen-LiLSX system, the thermal wave travels behind the

concentration wave. This means that the heat effects primarily affect the tail end of the

MTZ. A thermocouple placed at the exit of the column indicated that the temperature

increased around 17 °C for the lower velocity runs, and 19 °C for the higher velocity

runs. While the response time of the thermocouple was not fast enough to capture the

true thermal wave, it does provide an estimate of the thermal gradient that exists across

the MTZ. Since the increase in temperature was relatively constant, the heat effects do

not change significantly for the range of gas velocities used in this study. To avoid

including heat effects in the MTZ, 0.9 of the inlet concentration was used to define the

end of the MTZ. This more reasonably approximates an isothermal MTZ defined

between 0.05 and 0.95, which is the more common MTZ definition. The reasoning

behind this is further described in the appendix. This does not affect the calculation in

equation 5.1 because a non-isothermal model is used to determine n and n* as described

in the next paragraph. However, this change is important when the data is transformed to

MTZ data to facilitate use of the van Deemter model (described in section 5.4.5) since

that model does not account for heat effects in describing the amount of MTZ spreading.

A dynamic PSA model described in a later section (5.6) was adapted for

breakthrough experiments and used to match a simulated breakthrough curve to an

experimental curve. In brief, the model is a non-isothermal, non-isobaric model that

accounts for equilibrium and the adsorption rate using the dual-site Langmuir and linear

driving force (LDF) model respectively. The only change made to the model for this

study was to use plug flow and account for DL through the linear addition approximation

rather than axially dispersed plug flow. The primary reason for using the model was to

75
improve the estimation of the integral in equation 5.1. While the nitrogen component of

n* and n may be found from an equilibrium model and experimental data, the number of

points in the MTZ becomes limited as gas velocity increases. The simulated curve

increases the amount of points in the MTZ to improve the accuracy of the integration.

Furthermore, since the experiments were run near adiabatic conditions, thermal effects

are better described through use of the model rather than attempting to measure them

experimentally. An example of a typical breakthrough curve along with the model

representation of the curve is presented in Figure 5.4. Since the model is non-isothermal,

it is able to reasonably represent the experimental curve and account for the shift in

equilibrium due to the temperature increase. The predicted rise in temperature in the

outlet stream ranges from 20 °C at the lower gas velocity experiments to 23 °C at the

higher gas velocity experiments. Both n and n* were calculated from the exit molar flow

rate, pressure, and temperature of the model. The variable n was found from a

differential balance of the adsorbate as described by Wu et al.32 and in Appendix B. The

variable n* was determined from gas properties at the simulated column exit conditions

at time t.

76
Figure 5.4 Comparison of experiment and simulation breakthrough curve at 267 kPa,

room temperature and a inlet N2/O2 gas mixture of 0.75/0.25. The average superficial

velocity across MTZ was 2.8 cm/s

5.4.5 van Deemter Model

While the linear addition approximation is useful for measuring an overall

MTC, it is not a good model to estimate relevant mass transfer parameters such as τp and

DL using regression analysis. The van Deemter model provides a convenient method for

distinguishing between axial dispersion effects and mass transfer resistances due to the

particle. Although developed for linear chromatography, the van Deemter expression is

often applicable as an engineering approximation for moderately non-linear systems.24, 71

The van Deemter model72 says the height equivalent to a theoretical plate (HETP) for a

linear system is expressed as:

77
where A = 2γ2rp (eddy diffusivity term of the axial dispersion coefficient), B = 2γ1Dm

(molecular diffusivity term of the axial dispersion coefficient) and C ≈ 2εb/[(1- εb)(Kki)]

accounts for particle mass transfer resistance. Other variations of the model exist that

include factors such as film resistance through a velocity dependency of the A term.73, 74

Differentiating between the effects of eddy turbulence and film diffusion is notoriously

difficult since they both act to disperse the concentration wave in a similar manner.74

Since significant film diffusion effects were not expected, it was beyond the scope of this

study to use more complex models.

HETP was estimated from the height of a transfer unit (HTU) assuming HETP

≈ HTU.75 In linear chromatography, HTU = L/NTU where NTU is the number of

transfer units. However, for an adsorption column, HTU is found from the expression

HTU = MTZ/NTU where the calculation is performed across the defined limits of the

MTZ and NTU is the integral of equation 5.1. While this approach is only strictly

applicable to linear systems, it is a reasonable approximation for non-linear systems,

provided it is not a strongly non-linear system. The MTZ length was found using the

length of unused bed (LUB) approach.71 This approach is applicable for systems with a

favorable isotherm that develop a constant profile within the column. With this method,

the MTZ length is approximated from breakthrough experiments as:

where uw is the wave (constant profile) velocity. The wave velocity is determined using

equation 5.4:

78
where L is the column length and tc is the time the stoichiometric center exits the column.

MTZtime and tc were found based on the same 0.05 and 0.9 fraction of the inlet

composition used in the MTC calculation.

5.5 PSA System

The apparatus used to conduct PSA experiments is a complex system with many

valves so the user can produce and control a cyclic process. The system is made of 1/8

inch stainless steel tubing and fittings that connect various types of valves and flow

controllers. It is highly instrumented so the physical changes occurring at different points

in the process can be measured to provide information on process performance. The

system is capable of measuring pressure, temperature, flow rate, and gas composition.

The data is collected by a data acquisition system and systematically displayed on a

computer screen.

79
5.5.1 PSA Flow Diagrams

The system is divided into three sections relative to the columns. Figure 5.5

shows a schematic of the exit manifold section of the system, Figure 5.6 shows a

schematic of the feed section, and Figure 5.7 shows a schematic of the product section.

Table 5.3 provides a key for the symbols found in the figures. The feed side of the

columns is where feed gas enters and waste gas exits. The product end is where product

gas exits the column and purge gas enters the column. The exit manifold lies in between

the feed end and the product end. It contains a manifold of valves that control how gas

exits the system.

Figure 5.5 PSA middle exit manifold schematic

80
 Figure 5.6 PSA system feed side schematic

Figure 5.7 PSA product side schematic

81
Table 5.3 PSA schematic symbol description
Acronym Symbol Meaning

Pneumatic valve
PV#

Adsorption column
COLUMN#

Back pressure controller, needle

BPC#,
valve
NEEDLVAL
Switch valve: first number is
physical valve on system, second
number is the port number on the
SV#-#
valve

Flow controller/transducer
FC#,
FLOWTRANS

Hand valve
HV#

Mass spectrometer sample port,

MS#, P#, pressure transducer port, pressure
PTRANSD transducer

Vacuum pump
PUMP

Oxygen product tank

O2TANK

82
5.5.2 PSA Instrumentation, Process Control, and Data Collection

There are three types of valves used in the system: hand valves, pneumatic valves,

and switch valves. The hand and switch valves direct flow into the system manually.

The pneumatic valves are switched on/off by solenoid coils that are either controlled

manually or by an EZPLC programmable logic controller (PLC). The PLC provides

automated control of the pneumatic valves through a programmed repeated sequence of

events that open and close valves at specific intervals to create a cyclic PSA process. It is

located on a controller switch board that contains sixteen manual switches. The switch

position determines the controller switch board mode. The up position delegates control

to the controller, the down position switches the valve on, and the middle position turns

the valve off.

Two different pressure transducers placed at two separate points in the process

collect pressure data. The first is an Omega Dyne transducer (PTRANSD on Figure 5.7),

which has a very small dead volume (i.e. flush membrane transducer) and is used

primarily to determine the column pressure. The other pressure transducer is a MKS

Baratron Type 722A absolute pressure transducer that is connected to a switch valve (not

pictured in the figures). The switch valve has seven ports and is connected to six

different points in the process (P1-P6 on Figures 5.6 and 5.7) with one port left open to

measure the atmospheric pressure. This provides flexibility to check pressure at different

points in the apparatus. However, this pressure transducer has a significant amount of

dead volume and is normally only used on the product end of the process as to not

impose extra dead volume to the columns. Pressure is controlled by back pressure

83
controllers (BPC) placed at various points in the process. BPC1 was used to control the

adsorption pressure and BPC4 was used to control the desorption pressure.

Flow rate in the system was measured by three MKS flow meters located at three

separate points in the process. One is on the feed line (FC1 on Figure 5.6), another on the

purge line (FC2 on Figure 5.7), and a third on the exit manifold (FC4 on Figure 5.5).

There is an additional Omega FDMA-1600A Gas Flow Meter (FLOWTRNS on Figure

5.7) used to measure mass flow rate, pressure, and temperature for the equalization flow

exclusively. Due to slow response time, all flow controllers were left open and used only

as flow meters. Inlet flow was controlled through a needle valve after FC1. A needle

valve (NEEDLVAL on Figure 5.7) was placed before FC2 to control purge flow into the

columns.

Exit flow composition is measured by an AMI (Advanced Micro Instruments)

Oxygen Analyzer (O2ANALYZ on Figure 5.5) located on the product end of the process

within the exit manifold. There are also ports (MS1, MS2, MS3, MS6) at both ends of

each column for capturing a small amount of gas to analyze in a mass spectrometer.

These ports were only used if the mass spectrometer was required for fine tuning of a

process.

Data collection from pressure transducers, flow controllers, and the oxygen

analyzer was done with a National Instruments 7025E data acquisition system. The data

was then displayed on a computer using Lab Windows software. The software was also

used to record the data into a comma-separated-value file for further analysis.

84
5.5.3 System Design

Gas into the system can be supplied to either the feed side or product side of the

system through a switch valve (SV2-5 or SV2-3 on Figure 5.6). Four types of gases are

connected to a separate switch valve (not shown on figures) which controls the type of

feed gas supplied to the system. The gas options are nitrogen, oxygen, helium, and dry

air, which were all stored in compressed gas cylinders. A vacuum pump (Thomas WOB-

L piston pump, model # 817CA22) was connected to the product end to allow for vacuum

pressure in the columns during the purge/blowdown step. It was connected to a switch

valve (SV1) that takes the pump offline if it was not used in the process. An industrial

back pressure controller (BPC4) with pressure control above and below atmospheric

pressure is located upstream of the pump and is responsible for controlling the

purge/blowdown pressure. A product tank is connected on the product end of the system

if product collection is needed for cycles involving product pressurization. Two hand

valves (HV1 and HV2) control whether flow enters or bypasses the tank during the

process. The pneumatic valves (PV1-16) that control the flow in and out of the

adsorption columns are housed together in four valve blocks on both sides of each

column.

On the feed side of the system (Figure 5.6) during a PSA process, feed gas enters

during the feed/production step through SV2-5. It then flows through the line labeled

Feed 1 and Feed 2 to PV8 and PV16 which lead to the adsorption columns. During the

blowdown and purge steps, flow travels from the columns through PV6 and PV14 into

the lines labeled Blow1 and Blow2, which lead to the vacuum pump and exit D. PV7 and

PV15 were not used for this study and hence the lines leading to them were not displayed

85
in order to simplify the diagram (they are only used if a rinse step is included in the PSA

cycle). Pressure on the feed side of the process is measured primarily at P6, which

provided the pressure during the blowdown and purge steps. P1 and P2 are not advised

to be used for an extended period of time because of the dead volume induced on the

columns as was discussed earlier.

The product side of the system (Figure 5.7) is more complex. During normal

operation, product flowed through a combination of PV1 and PV9 if the product tank was

bypassed during process operation. PV3 and PV11 allow for purge flow into the

columns. As previously mentioned, the needle valve located before FC2 controlled the

flow to PV3 and PV11, while FC2 was only used to measure the flow rate to the

columns. PV4 and PV12 were only used for processes with a pressure equalization step.

BPC3 is located on this line connecting these valves, but remained closed during the

entirety of the experiments. Pressure on the product side was measured in two places,

immediately before column 2 with PTransd, or at P3.

The middle manifold section of the system is represented in Figure 5.5 and is

where gas was analyzed prior to exiting the system. The manifold inlets correspond with

the exit locations located in Figure 5.6 and 5.7. Gas entered the manifold either through

exit A (product flow) from the product side or exit D (waste flow) from the feed side.

Gas can enter through exit B and C, but for this study, only exit A and exit D were used.

The manifold consists of four hand valves that direct the flow out of the system either

through FC4 and the oxygen analyzer, or directly to the vent. The function of this

manifold was for the calculation of a material balance for the process. It provided

measurement of the flow rate and oxygen concentration out of the system before the

86
gases were vented to the atmosphere. The manifold had the flexibility to either allow one

of the streams to be analyzed separately while the other was directed straight to the

atmosphere, or allow both product and waste streams to be analyzed together.

5.5.4 Material Balances

Material balances are important for PSA processes to validate process

performance. A general balance for any system with no chemical reaction is represented

as:

To calculate material balances, the flow rate and composition of the entrance and exit

streams were measured. The instantaneous molar flow rate, F, at different points in the

process is useful for plotting a flow rate versus time chart. This visual representation of

the data is useful for real time analysis and for determining CSS. However, for a process

that involves constantly changing flow rates, a steady flow in and out of the process

rarely occurs. When there is an absence of steady flow, material balances are conducted

using a time averaged molar flow rate, , which for each flow meter was calculated as:

where t is the time interval over which the average is taken. Using a multiple of cycle

time for the time interval provides a better flow average. Material balance calculations

are based on the amount of gas into and out of the process on a per cycle basis. The

87
amount of gas used for each step in the cycle was calculated by multiplying the average

flow rate with the cycle time, expressed as:

where ΔNstep is the amount of gas used/produced during a specific step in the cycle and tct

is the cycle time. Calculating the amount of gas per step in a cycle is a convenient way to

conduct a material balance to validate process performance.

For a material balance on a PSA process, the input is the amount of gas in the feed

stream, while the exit streams consist of the product stream and exit streams of the purge

and blowdown step. For this study, the streams from the exit of the purge and blowdown

steps are measured with the same flow meter, so this stream is collectively referred to as

the waste stream. Thus the overall material balance for the cycle is calculated as:

Where is the material balance error, NF is the amount of feed gas used, NP is the

amount of product gas produced, and NW is the amount of waste gas generated, all on a

per cycle basis. The individual component balances can then be solved by equations 5.9

and 5.10:

where , and are the mole fractions of either nitrogen or oxygen in the feed,

product, and waste streams respectively. At CSS, values of must be zero; however for

88
experimental work this almost never occurs due to uncertainty in measurement and data

collection. Nevertheless, it is important to minimize the amount of error and close

material balances as closely as possible to validate process calculations. All material

balances were closed within 5% for the data included in this study.

5.5.5 Design Parameters

When designing a cycle for a PSA process, several parameters are set based on

the product purity desired and allowable power consumption for the process. They

include purge to feed ratio, pressure ratio, and feed stream velocity (cycle time).

Product purity is the amount of oxygen in the product stream divided by the sum

of all the components in the product stream. It can be represented as:

where N is number of moles and j is the number of components in the product. For a

PSA process, it is useful to express oxygen purity as a concentration averaged over both

flow and time because the concentration exiting the process varies with both. This

involves dividing a time average of the oxygen flow rate by a time average of the total

flow rate. This may be represented as:

The purge/feed ratio controls product purity and the amount of product the

process generates. The purge/feed ratio may be represented as:

89
It should be noted that in this study, the “Feed Step” is the combination of the feed

pressurization and production steps. A higher P/F ratio means a greater amount of

product gas is used in column regeneration, which further pushes back the MTZ towards

the feed end of the columns. The minimum P/F ratio is the ratio that gives the desired

product purity, but a further decrease in the ratio causes undesired product purity. In

order to ensure the desired product purity is maintained, it is not advisable to operate at

the minimum P/F ratio, but rather at a ratio slightly higher than the minimum.

Pressure ratio is set based on the power allowances for the process. It is defined

as:

For a PSA process, the blowdown/purge step occurs at atmospheric pressure and a pump

is required to pressurize the columns to the desired adsorption pressure during the

feed/production steps. For a VSA and combined PVSA processes, the blowdown/purge

step occurs at vacuum pressure and the adsorption pressure is either atmospheric pressure

for VSA cycles, or some pressure above atmospheric for PVSA cycles. Regardless of

cycle type, increasing pressure ratio leads to higher recovery at the cost of additional

power consumption. However, the gain in recovery is diminishing as pressure ratio

increases; hence pressure ratio is normally chosen near the point where the gain in

recovery no longer justifies the required power increase.

90
 The last design parameter is cycle time, which was controlled by the individual

step times set by the PLC. These step times may be changed, however, it was required

that the combined time of the purge and blowdown steps equal the combined time of the

feed pressurization and production steps.

5.5.6 Performance Parameters

Process performance is primarily evaluated through two different performance

parameters. Recovery is the amount of desired component fed into the system that is

available as product. In essence, it is related to the energy efficiency of a process. For

air separation, it can be calculated based on the product stream or the waste stream. If it

is based on the product stream, it is calculated from equation 5.15 using the amount of

gas per cycle of the product and feed streams.

For a cycle at CSS, recovery is inversely related to purity. An increase in product purity

decreases recovery since a higher purity requires more purge gas, which leaves less gas

available as product. Recovery is dependent on separation efficiency just like purity.

Hence, it is possible to increase recovery with a more selective zeolite or inclusion of an

additional step like pressure equalization that increases the regeneration efficiency.

The other performance parameter is BSF which is a measure of adsorbent

utilization and process productivity, often expressed as:

91
BSF is related to column length, feed flow rate, and cycle time. It corresponds to the

system size and hence the process capital costs. Adsorption rate also plays a significant

role in the BSF as a faster adsorption rate allows for smaller columns, shorter cycle times,

and higher product flow rates.

Another parameter useful in evaluating a PSA process is working capacity, which

represents the amount of adsorbate that adsorbs/desorbs per cycle. This capacity is found

from the difference in the amount of gas at the end of the feed and purge step, which

corresponds to the “cleanest” and “dirtiest” conditions of the solid adsorbent. Figure 5.8

visually demonstrates what working capacity represents on an isotherm plot. At the end

of the feed step, the column temperature has risen (since adsorption is exothermic), which

changes the isotherm that determines adsorption equilibrium. Conversely, at the end of

the purge step, temperature in the column has decreased (since desorption is

endothermic), which again changes the isotherm that determines adsorption equilibrium.

While this is a useful metric, it is very difficult to measure zeolite temperature in the

columns to see this transpiring. That is why an average temperature isotherm is

generated to estimate the working capacity. One of the advantages of process simulators

is that they are able to calculate working capacity directly since they can create an

adsorption profile across the column.

92
Figure 5.8. Comparison of isothermal working capacity and dynamic working capacity

in a PSA process

5.5.7 Pressure Drop Study

The first PSA experimental study was conducted to better understand how

pressure drop across the column affects process performance. For this study, column

pressure drop was manipulated by changing both the total flow rate and column diameter.

Column length was chosen in the range of typical small scale processes designed for air

separation. Short, wide columns with a small, but measurable pressure drop under rapid

cycling conditions were designed for the first set of experiments in this study. Longer

columns with the same volume, but half the cross sectional area were designed for the

next set of experiments. The length was adjusted to maintain the same amount of zeolite

as the first column set. With an equivalent amount of adsorbing solid, a comparison was

possible between cycles operating with similar cycle times and flow rates, but at different

93
velocities, hence changing the amount of column pressure drop. Further column details

are provided in Table 5.4. The cycle type was a traditional feed pressurization cycle with

an additional product end equalization step. The cycle steps included: 1) feed

pressurization 2) production 3) co-equalization 4) blowdown 5) purge 6) counter-

equalization. For Tables 5.5 and 5.6, the time of steps 1 and 2 are combined and called

the “Feed press./Production” step. The switch from pressurization to production

(traditional feed step) is automatically controlled by a set back pressure controller.

PSA experiments were designed to study the effect of cycle time and column

pressure drop on process performance. To accomplish this, all other process parameters

(e.g. pressure ratio, purge/feed ratio) were held constant as cycle time was decreased.

Experimental conditions for each cycle are detailed in Tables 5.5 and 5.6. It should be

noted the synthetic air used in the experiments does not contain argon, water, or carbon

dioxide, making pure oxygen product possible. In comparison, the maximum attainable

oxygen purity of an ambient air fed system is around 96% since LiLSX zeolites cannot

separate oxygen and argon. With synthetic air, an average oxygen purity of 96% was

chosen as the purity requirement because it approximately corresponds to an oxygen

purity of 92-93% with ambient air, commonly used for personal oxygen concentrators..

The P/F ratio was selected based on these purity requirements. Equalization times were

chosen by pressure measurements on the two columns. Overall and individual material

balances for a cycle were closed within 5%. All data was taken after CSS. CSS was

assumed when the column pressure at the end of each step, inlet/exit flowrates, and

oxygen purity varied less than 1% for 15-20 cycles. The pressure ratios were selected in

the range of a typical small scale air separation process. The cited pressure ratios

94
correspond to feed end measurements. The pressure ratio on the product end of the

column is lower when pressure drop is significant. The superficial production step

velocities cited in Table 5.5 and 5.6 are at the production step pressure. Cycle times were

decreased until the apparatus could no longer maintain the desired pressure ratio at the

feed end of the process (combination of flow, valve CV, and pressure limitations).

Table 5.4 Experimental column specifications for pressure drop study

Short Long
columns columns

Column diameter (cm) 1.09 0.77

Column length (cm) 9.8 19.6

Amount of adsorbent per

5.7 5.7
column (g)

Table 5.5 Cycle specifications for short column cycles in pressure drop study

Production step Total

Feed
superficial Blowdown (s) Purge (s) Eq. (s) cycle time
Press./Production (s)
velocity (cm/s) (s)
Cycle 1 7.5 6.75 4.75 2 0.5 14.5

Cycle 2 10.9 4.3 3.3 1 0.5 9.6

Cycle 3 15.8 2.9 1.9 1 0.5 6.8

Cycle 4 28.8 1.4 1.1 0.3 0.5 3.8

Pressure ratio 4.5, P/F ratio 0.4, product purity: 96% oxygen , production step pressure ~195 kPa, synthetic air

95
Table 5.6 Cycle specifications for long column cycles in pressure drop study

Production
Feed Total cycle
step superficial Blowdown (s) Purge (s) Eq. (s)
Press./Production (s) time (s)
velocity (cm/s)

Cycle 5 11.6 7.2 5 2.2 0.3 15

Cycle 6 16.6 4.7 3.5 1.2 0.3 10

Cycle 7 22.9 3.2 2.2 1 0.3 7

Cycle 8 39.7 1.7 1 0.7 0.3 4

Cycles 5-8: Pressure ratio 2.5, P/F ratio 0.6, product purity: 96% oxygen , production step pressure ~195 kPa,
synthetic air
Cycle 9 20.7 4.5 3.5 1 0.5 10

Cycle 10 29.8 3 2 1 0.5 7

Cycle 11 48.1 1.5 1 0.5 0.5 4

Cycle 12 52.6 1.3 1 0.3 0.5 3.6

Cycles 9-12: Pressure ratio 4.5, P/F ratio 0.4, product purity: 96% oxygen , production step pressure ~195 kPa,
synthetic air

5.5.8 Minimum BSF Study

The next experimental PSA study was conducted to determine if a minimum BSF

occurs in the system. The main difference between these experiments and the pressure

drop study is that the process was simplified to a 4-step cycle in order to avoid the

complications of an equalization step on process performance. While a significant

advantage is gained through an equalization step, it adds additional uncertainty into the

analysis since it is difficult to measure. Considering the goal of this study was measuring

and explaining the existence of a minimum BSF, not maximizing performance, an

equalization step was not necessary. The cycle steps include: 1) feed pressurization 2)

production 3) counter-current blowdown 4) counter-current purge.

96
 Cyclic steady state (CSS) was achieved over a range of cycle times permitted by

our system. Process parameters such as product purity, purge/feed ratio, and pressure

ratio were kept constant at each cycle time. The experimental procedure was exactly the

same as the pressure drop study and the same product purity requirement was used. The

column dimensions used in this study were very similar to the pressure drop study (Table

5.7). Cycle details are further shown in Tables 5.8-5.12.

Table 5.7 Experimental column specifications for minimum BSF study

Short Long
columns columns

Column diameter (cm) 1.09 0.77

Column length (cm) 10.2 19.6

Amount of adsorbent per

5.3 5.7
column (g)

97
Table 5.8 Cycle specifications for long column cycles with pressure ratio 3.5 in

minimum BSF study.

Production step
Feed Total cycle
superficial Blowdown (s) Purge (s)
Press./Production (s) time (s)
velocity (cm/s)

Cycle 1 45.1 3 2.5 0.5 6

Cycle 2 69.8 1.95 1.62 0.33 3.9

Cycle 3 81.1 1.65 1.35 0.3 3.3

Cycle 4 82.4 1.6 1.3 0.3 3.2

Cycle 5 44.0 3 2.5 0.5 6

Cycle 6 62.6 2.1 1.75 0.35 4.2

Cycle 7 78.4 1.68 1.4 0.28 3.36

Cycle 8 84.8 1.56 1.3 0.26 3.12

Cycle 9 50.1 2.64 2.2 0.44 5.28

P/F ratio 0.45, product purity: 96% oxygen , production step pressure ~195 kPa, synthetic air

Table 5.9 Cycle specifications for long column cycles with pressure ratio 3 in minimum

BSF study

Production step
Feed Total cycle
superficial Blowdown (s) Purge (s)
Press./Production (s) time (s)
velocity (cm/s)

Cycle 10 39.7 3.1 2.5 0.6 6.2

Cycle 11 80.4 1.55 1.25 0.3 3.1

Cycle 12 93.1 1.36 1.1 0.26 2.72

Cycle 13 85.6 1.46 1.18 0.28 2.92

Cycle 14 50.0 2.45 2 0.45 4.9

Cycle 15 63.0 1.96 1.6 0.36 3.92

P/F ratio 0.47, product purity: 96% oxygen , production step pressure ~195 kPa, synthetic air

98
Table 5.10 Cycle specifications for long column cycles with pressure ratio 2.5 in
minimum BSF study

Production step
Feed Total cycle
superficial Blowdown (s) Purge (s)
Press./Production (s) time (s)
velocity (cm/s)

Cycle 16 34.5 3.2 2.5 0.7 6.4

Cycle 17 42.1 2.65 2 0.65 5.3

Cycle 18 70.3 1.6 1.25 0.35 3.2

Cycle 19 89.4 1.3 1 0.3 2.6

Cycle 20 78.8 1.45 1.1 0.35 2.9

Cycle 21 89.5 1.3 1 0.3 2.6

Cycle 22 48.9 2.2 1.75 0.45 4.4

Cycle 23 58.0 1.9 1.5 0.4 3.8

P/F ratio 0.5, product purity: 96% oxygen , production step pressure ~195 kPa, synthetic air

Table 5.11 Cycle specifications for short column cycles with pressure ratio 3.5 in
minimum BSF study

Production step
Feed Total cycle
superficial Blowdown (s) Purge (s)
Press./Production (s) time (s)
velocity (cm/s)

Cycle 24 22.6 3 2.5 0.5 6

Cycle 25 40.7 1.8 1.5 0.3 3.6

Cycle 26 49.3 1.55 1.25 0.3 3.1

Cycle 27 45.4 1.65 1.35 0.3 3.3

Cycle 28 28.7 2.42 2.02 0.4 4.84

Cycle 29 33.5 2.1 1.75 0.35 4.2

Cycle 30 16.4 4.2 3.5 0.7 8.4

Cycle 31 37.9 1.95 1.62 0.33 3.9

Cycle 32 42.6 1.75 1.45 0.3 3.5

Cycle 33 47.8 1.6 1.3 0.3 3.2

P/F ratio 0.45, product purity: 96% oxygen , production step pressure ~195 kPa, synthetic air

99
Table 5.12 Cycle specifications for short column cycles with pressure ratio 3 in
minimum BSF study

Production step
Feed Total cycle
superficial Blowdown (s) Purge (s)
Press./Production (s) time (s)
velocity (cm/s)

Cycle 34 20.0 3.2 2.6 0.6 6.4

Cycle 35 19.6 3.2 2.7 0.5 6.4

Cycle 36 36.8 1.85 1.5 0.35 3.7

Cycle 37 44.3 1.6 1.3 0.3 3.2

Cycle 38 49.5 1.47 1.2 0.27 2.94

Cycle 39 28.5 2.32 1.9 0.42 4.64

Cycle 40 14.1 4.64 3.8 0.84 9.28

P/F ratio 0.47, product purity: 96% oxygen , production step pressure ~195 kPa, synthetic air

5.6 PSA Model Framework

The primary aim of this work was experimental in nature. However, in order to

better illustrate the effects of pressure drop and more accurately measure the MTZ, an in-

house simulator developed by Dr. Sridhar Ungarala was used to provide further

information not available from experimental data, particularly column profiles. Only a

brief description of the more relevant aspects of the model is presented here, the complete

model is provided in Appendix A. The model is non-isothermal and assumes axially

dispersed plug flow with ideal gas behavior and negligible radial concentration and

temperature gradients. Pressure drop is included through the Ergun equation.

100
5.6.1 Mass Balance

The column component mass balances are represented by the axially dispersed

plug flow model introduced in section 3.3.1:

The adsorption rate was described with a linear driving force (LDF) model

described in section 3.3.1.24

The LDF model simplifies the mass transfer rate to permit faster computation time, which

explains its popularity in literature. The LDF approximation should provide reasonable

accuracy for the system used. Higher accuracy models exist, but their complexity

increases computation time significantly, often without a substantial difference in

accuracy.24 The MTC, ki, is typically approximated through literature data or

breakthrough experiments by assuming the resistance to macropore diffusion in the

adsorbent particle as the limiting resistance.2

5.6.2 Equilibrium Model

Adsorption equilibrium was formulated by the dual-site Langmuir (DSL) model.

Mathias et al.23 have demonstrated the accuracy of the DSL model for both pure

component gases and nitrogen/oxygen mixtures using zeolite 5A. A LiLSX zeolite is

similar enough to zeolite 5A to justify using the DSL model to predict mixture

101
equilibrium in the PSA model. For simulation studies, the parameters were used

according to equation 3.8, previously described in section 3.2.3.

The DSL parameters used in the model may be found in Table 5.1.

5.6.3 Momentum Balance

Column pressure drop was modeled using the Ergun equation (equation 5.17). It

has been found to provide a sufficient approximation of pressure drop in packed

adsorbing columns.76

5.6.4 Energy Balance

The model also considers three energy balances for the gas, solid particle, and column

wall with natural convection to room temperature air. All parameters were estimated

from well-established correlations from literature. Details are provided in Appendix A.

102
 CHAPTER VI

RESULTS AND DISCUSSION

This chapter first presents the MTCs found from breakthrough experiments and

demonstrates that axial dispersion is much more important for small particles compared

to large particles. It next experimentally demonstrates how pressure drop affects small

scale PSA performance with support from a brief simulation study. Finally, a minimum

BSF was measured and an explanation for why it is occurring is explored.

6.1 Axial Dispersion Effects in Small Scale PSA

To demonstrate the significance of axial dispersion in columns of small particles,

an experimental study was performed to measure the overall MTC as a function of gas

velocity at constant pressure and temperature. The experiments were conducted

according to the procedure outlined in section 5.4.2. The data is evaluated with the linear

addition approximation (equation 4.1) to provide a suitable method to combine the effects

of axial dispersion and particle mass transfer resistance, as computationally done for

Figure 4.2. The measured breakthrough curves were also converted to MTZ length data

103
to determine the relevant mass transfer parameters for the particles used in this study.

Finally the extracted kinetic data was used to better understand what the rate limiting

mechanism is for columns packed with small LiLSX particles.

6.1.1 Experimental MTC Measurement

In Chapter 4 a case study was used to demonstrate how the overall MTC

(equation 4.1) was affected by the estimation of axial dispersion effects. Case 1 and

Case 2 estimate koverall using assumptions applied to large particles to determine DL, while

Case 3 and Case 4 used assumptions more applicable to small particles. Figure 6.1

directly overlays the results of the breakthrough study on the results of the case study in

Figure 4.2 to determine the case that best describes the data. It is immediately clear the

experimental MTC is not constant, but increases with Reynolds number as expected

based on the predictions. It is also clear that either the case 3 or case 4 prediction most

closely matches experimental data. These cases reflect higher axial dispersion effects

through a higher estimate of DL compared to the case 1 or case 2 prediction. Results

from a similar study with the same particle size are also represented in Figure 6.1.32 In

that study, the focus was to determine the effect of temperature and pressure on the MTC,

which limits the amount of data that can be compared to this study. The higher reported

MTC is a linear addition model prediction using similar correlations as the case 2

prediction and the lower value is an experimentally measured MTC. The predicted

values and experimental data are different between the two studies primarily because of

the small variation in pressure (267 kPa for this study vs. 200 kPa for Wu et al.), inlet gas

composition, and isotherm parameters used. However, both studies are consistent in that

104
the case 2 prediction is ~ 3.6 times greater than the experimental data at a Reynolds

number ~ 1.6.

Figure 6.1. Experimental data vs. correlation predictions of the nitrogen overall MTC

using τp = 3, γ2 = 3 and γ1 = 0.7. Reynolds number of experimental data points is based

on average velocity across MTZ. Overall pressure is 267 kPa for this study and 200 kPa

for the literature study. Both studies have a particle size ~ 0.5 mm. Uncertainty for

experimental data ~ ± 5%

105
 The results of this study support the notion that axial dispersion effects

contribute significantly to the overall MTC of small particles. More importantly, the

results suggest the γ2 estimate required to determine DL in equation 3.15 is an order of

magnitude higher than the value of 0.5 used for larger particles. The value of 0.5 results

in a large overestimation of the MTC (case 1 and case 2), which will predict a sharper

MTZ than what is measured experimentally (example provided in Appendix B). The

results also indicate that despite its limitations, the linear addition approximation provides

a reasonable estimation for how the MTC varies with gas velocity when proper literature

correlations for pore diffusion, film transfer, and axial dispersion are used.

6.1.2 Mass Transfer Parameter Estimation

The van Deemter model (equation 5.2) was next used to determine relevant

mass transfer parameters useful for process simulations, namely τp and the γ2 coefficient

to estimate DL. The MTZ length for each experimental run was converted to HETP as

described in section 5.4.5 and plotted against the average superficial velocity across the

MTZ as demonstrated in Figure 6.2. A least squared fit of the data was completed using

the van Deemter model to determine values for the A, B and C constants in the model.

From the constants, the γ1 and γ2 coefficients of equation 3.15 and τp of equation 4.4 were

found assuming a particle size of 0.5 mm. Table 6.1 lists the calculated values along with

the standard error associated with each value and a comparison to reported literature

values or correlations.

Despite the limitations of this approach, the rate parameters are reasonable

when compared to available literature data or correlations. The value for γ1 used to

106
estimate DL is higher than the usual value of 0.7. More data at lower gas velocities is

necessary to confirm this value as demonstrated by the large standard error. However,

the γ1 estimation was not of great concern in this study, as it only has a small effect on the

MTC estimation and is primarily important at very low gas velocities not applicable to

PSA systems. More importantly, the results confirm γ2 is an order of magnitude higher

than the often used 0.5 value. The experimental value of 3.2 is similar to what was

predicted by the literature correlation in equation 4.3 for the particle size used in this

study. Finally, the estimate of τp for nitrogen is consistent with most X-type zeolites and

studies of larger LiLSX particles cited in section 4.1.4. Since these particles have been

manufactured to be rate enhanced for small scale processes, it is not surprising the value

is on the lower end of what has been previously reported in literature for larger particles.

Figure 6.3 uses the parameters for γ2 and τp found from the van Deemter

analysis to reconstruct the predictions from the earlier case study. The literature value of

0.7 was used to estimate γ1 since the data collected did not reach a low enough velocity to

achieve a reliable estimate using the van Deemter analysis. It is evident from the figure

that the experimental data is still best described by either the case 3 or case 4 prediction.

Furthermore, these predictions are now closer to the experimental values than those in the

original case study.

107
Figure 6.2. HETP vs. superficial velocity from experimental breakthrough experiments.

Superficial velocity is based on average velocity across MTZ. Superficial velocity = uεb

Table 6.1. Comparison of van Deemter constants determined from experimental data and

previously cited literature values. i = nitrogen

Standard
This study Literature
error

γ1 2.3 ± 1.1 0.7

γ2 3.2 ± 0.6 3

τ = Dm/Dpi 1.9 ± 0.5 2-4

108
Figure 6.3 Experimental overall MTC plotted against overall nitrogen MTC predictions

with τp = 1.9, γ2 = 3.2 and γ1 = 0.7. Reynolds number is based on average Reynolds

number across MTZ

6.1.3 Rate Limiting Contribution

Table 6.2 lists the individual contributions in equation 4.1 using the

experimental parameters from Table 6.1 for γ2 and τp. The table also compares the same

contributions for a larger (2 mm) particle using the same τp and equation 4.3 to determine

γ2. The table is designed to demonstrate how the rate limiting contribution changes for

the particle size used in small scale PSA at a Reynolds number applicable to process

work. The limiting contribution for the larger particle is clearly the macropore

contribution as it represents 67.7% of the total resistance. For the smaller particle, the

contributions from dispersion and macropore diffusion are about the same, which

109
demonstrates that both are important to model a PSA process accurately. For

breakthrough experiments, where even lower Reynolds numbers are used, the dispersion

term is predominant for the smaller particle.

Figure 6.4 further demonstrates the significance of axial dispersion effects for

small particles when determining the overall MTC experimentally. In the figure, MTC

predictions for the two different particle sizes are compared. It is clear the MTC

predictions for the large particle do not vary significantly, which indicates the DL

estimate does not greatly impact the overall MTC estimation. Furthermore, macropore

diffusion is the limiting resistance except at very low Reynolds numbers where the MTC

varies significantly because of axial dispersion effects. This is not the case for small

particles as the MTC is a much stronger function of Reynolds number. The switch from

axial dispersion control to macropore diffusion control is much more gradual compared

to larger particles.

Recognizing the mass transfer rate is significantly impacted by axial dispersion

effects increases the importance of properly accounting for it in process models with

small particles. Simulation accuracy may be significantly affected by using the wrong

correlation to approximate DL. A mass transfer rate controlled by axial dispersion effects

additionally influences typical particle production methods aimed at reducing macropore

resistance. These methods will have a more limited impact on process performance since

pore diffusion is no longer the only limiting mass transfer mechanism.

110
Table 6.2 Comparison of equation 4.1 contributions for a nitrogen overall MTC at a

Reynolds number of 20 and two different particle sizes. Constants used: τp =1.9, γ1 =0.7

and γ2 = 3.2 (small particle) γ2 = 0.75 (large particle). Mass transfer resistance is

proportional to 1/ki

% of total % of total % of total

Particle kdisp kmacro kfilm koverall
resistance resistance resistance
size (cm) (1/s) (1/s) (1/s) (1/s)
(disp) (macro) (film)

0.05 14.4 16.0 74.5 6.9 48 43 9

0.2 3.8 1.0 4.7 0.68 18 68 15

Figure 6.4. Comparison of koverall predictions for a) 0.5 mm particle (left) and b) 2 mm

particle (right). τp = 1.9 and γ1 = 0.7 for all predictions for both particle sizes. γ2 = 0.5 for

both case 1 and 2 predictions for both particle sizes. For case 3 and 4, γ2 = 3.2 for 0.5

mm particle and γ2 = 0.75 for 2 mm particle

111
6.1.4 Skin Resistance

Skin resistance has also been cited as a possible reason for a lower than

expected MTC based on typical correlations.32 A skin, or change in

density/porosity/composition of a particle near the external surface, is the consequence of

some particle manufacturing methods. In particular, extrusion is a popular method for

particle formation, which inevitably results in a skin on the particles. The SEM images in

Figure 6.5 provide a visual example of a skin formed through extrusion in our lab using a

material similar to LiLSX and a common binder. However, the current state of the art for

manufacturing spherical particles on an industrial scale does not involve extrusion.

Particles are rather formed by the addition of water to a mixture of zeolite and binder in a

rotating drum.42, 77 While this does not completely eliminate the existence of a skin, it

significantly reduces skin formation since there are no high stress flow of the extrusion

mixture. SEM images of a spherical particle used in this study are included in Figure 6.6.

No noticeable difference in density is visible at the particle surface where a skin would

form. It has been suggested a skin resistance accounts for up to 65% of the total

resistance for LiLSX particles.32 If skin resistance was this significant, it would be

visible in an SEM image like in Figure 6.5. While an SEM image cannot eliminate the

possibility of a skin resistance, it does justify considering the contribution negligible as

suggested by our experimental results. A further indication skin resistance is not

significant is that it would primarily affect the estimation of τp and would not be a

function of gas velocity. However, the results of this study clearly show the MTC varies

significantly with velocity and the estimation of τp, hence Dp, is reasonable.

112
Figure 6.5 SEM images of an extruded particle edge

Figure 6.6 SEM images of LiLSX particle used in this study. Particle was cut in half

prior to imaging

113
6.2. Role of Pressure Drop in Small Scale PSA

Pressure drop is a potential process limitation for small scale PSA that is not well

understood. Hence, the purpose of first set of PSA experiments is to demonstrate how

pressure drop affects small scale PSA performance. Pressure drop across the column was

manipulated by changing column dimensions such that the cross sectional area between

the columns sets varied by a factor of two while the amount of adsorbent remained

constant, as outlined in section 5.5.7. This permitted a comparison between the process

performance of two different sets of columns with a column pressure drop per unit

column length that varied by a factor of two (or total pressure drop across the column that

varied by a factor of four). The results were then compared to simulated cycles to further

explain the experimental results.

6.2.1 Experimental Results

Pressure drop measurements for the experimental portion of this study were

restricted to the constant high pressure production step due to experimental limitations.

As expected, the cycles of the longer columns exhibited about twice the pressure drop per

centimeter of column length as the cycles of the shorter columns at the same pressure

ratio. Column pressure drop during the production step was generally consistent with

Ergun equation predictions within 10%. The highest recorded pressure drop during the

production step only approached 1 kPa/cm during the fastest cycle of the longer columns

(pressure at column exit was 10% less than the inlet). For small scale air separation

applications, the column length used in this study is similar to what is used in industry for

a POC.

114
 The expectation based on previously cited studies was column pressure drop

would negatively affect the recovery and BSF of the longer column cycles more than

those of the shorter columns. Figure 6.7 demonstrates this was not the case for oxygen

recovery by plotting it as a function of production step superficial velocity. In Figure 6.7,

two comparisons are made. One is between cycles of the short and long columns at an

equivalent pressure ratio. The other is between cycles with different pressure ratios using

the longer set of columns. Despite differences in pressure drop between the long and

short columns, the slope of the recovery decline is comparable. If pressure drop was

affecting product recovery, the decline of the recovery for the longer columns would be

faster. Furthermore, the lower pressure ratio cycles for the longer columns exhibit a

similar recovery decline as the higher pressure ratio cycles. Pressure drop effects

increase with a higher pressure ratio. The results indicate for a pressure ratio increase

from 2.5 to 4.5, the effects are minimal. Figure 6.8 compares the BSF of the long and

short columns at the same pressure ratio as a function of production step velocity. The

figure illustrates that cycles operating with a similar cycle time have almost the same

BSF. If column pressure drop was affecting the recovery and working capacity of the

process, then the BSF of cycles operating with a similar cycle time was expected to differ

more.

Perhaps the most interesting aspect of these results is that they suggest for a given

pressure ratio and similar cycle time, operating with a long, thin column design provides

the same BSF with a higher oxygen recovery than a short, wide column design. This was

not an expected result if pressure drop was playing a large role in process performance

since the long, thin column design has a much greater amount of pressure drop. Clearly,

115
this suggests alternative factors play larger roles in determining process performance

other than pressure drop.

Figure 6.7 Comparison of oxygen recovery decline with increasing production step

superficial velocity. Data shown involves columns with different dimensions and cycles

with different pressure ratios (lines are drawn to guide the eye). Dashed lines connect

cycles of similar cycle time

116
Figure 6.8 Comparison of BSF with increasing production step superficial velocity for

columns of different dimensions (Lines are drawn to guide the eye). Dashed lines

connect cycles of similar cycle time. (TPD = tons per day)

6.2.2 Simulation Results

The dynamic response model was briefly reviewed in section 5.6 and is fully

described in the appendix. CSS was determined when the oxygen purity, recovery, and

BSF were within 0.5% for at least 15 cycles. Simulated and experimental values for two

cases (Cycles #3 and #10 of the pressure drop study) are summarized in Table 6.3. The

simulator provides process performance parameters comparable to experimental results.

Simulated concentration and loading profiles are therefore considered reasonable

approximations of experimental profiles. Figure 6.9 compares the gas composition

against column length at the end of the high pressure production step. The figure

demonstrates the concentration profiles are nearly identical even though the longer

117
columns have twice the pressure drop per unit column length as the shorter columns.

This indicates column pressure drop is not affecting the high pressure concentration

profile. Figure 6.10 compares the concentration profiles between the two cycles at the

end of the purge step. The profiles are once again nearly the same despite the higher

pressure drop of the longer columns. As with the production step, column pressure drop

has little impact on the low pressure concentration profile.

Table 6.3 Comparison of experimental and simulation results at similar cycle times and

different column dimensions. (TPD = tons per day)

Short columns: cycle #3 Long columns: cycle #10

experiment simulation experiment simulation

Cycle time (s) 6.8 6.83 7 7.08

Product purity (% oxygen) 96 96.2 96 96

Oxygen recovery (%) 45.5 45.5 50.3 50.2

BSF (lbs. solid/TPD O2) 81 83.6 80 76.7

Production step pressure

0.19 0.16 0.29 0.34
drop (kPa/cm)

Total pressure drop as %

of production step 0.95 0.80 2.9 3.4
pressure

Purge step pressure drop

- 0.28 - 0.54
(kPa/cm)

Total pressure drop as %

- 6.4 - 24.6
of purge step pressure

118
Figure 6.9 Comparison of gas phase nitrogen composition at the end of the production

step for the short columns (cycle #3) and long columns (cycle #10)

Figure 6.10 Comparison of gas phase nitrogen composition at the end of the purge step

for the short columns (cycle #3) and long columns (cycle #10)

119
 The nitrogen solid loading profiles (Figure 6.11) at the high pressure production

step and low pressure purge step indicate only slight differences between the two cycles.

The variation in loading at the end of the production step is primarily related to heat

transfer due to the difference in column diameter. For the low pressure purge step, the

difference in nitrogen loading at the feed end (z = 0) of the column is also adequately

explained by thermal effects. However, the difference in nitrogen loading at the product

end of the column is best explained by pressure drop effects since the loading of the

longer columns is greater than the shorter columns in this section. The difference due to

pressure drop is not significant and its impact is reduced because it affects the side of the

column undergoing the least amount of nitrogen loading/unloading. The solid nitrogen

working capacity was determined from the difference in solid nitrogen loading at the end

of the production and purge steps. In Figure 6.11, this amounts to the difference of the

area under the curves between the production and purge steps. Both columns had an

average working capacity around 0.495 mol/kg. If thermal effects were not as

considerable, the working capacity of the longer column would be less than the shorter

columns due to pressure drop. However, this difference would be relatively small since

the overall pressure drop across the column is not significant.

Since the P/F ratio was maintained between the two cycles, the amount of gas

used during the purge step was nearly the same. As previously mentioned, the loading

profile after the purge step does indicate pressure drop reduces the amount of nitrogen

desorbed on the product end of the longer columns. A greater amount of gas would be

needed to achieve a similar desorption duty as the shorter columns. However, even

though the pressure drop per unit length increased by a factor of two (total pressure drop

120
increased by a factor of 4), the difference in the amount desorbed from the solid was only

about 2% of the solid nitrogen working capacity. For columns with less pressure drop,

the difference would be even more insignificant.

Figure 6.11 Comparison of solid nitrogen loading at the end of the production and purge

steps for the short columns (cycle #3) and long columns (cycle #10)

121
 The purpose of the next comparison was to determine the effect of increasing

column pressure drop due to a higher cyclic frequency. For this comparison, the

performance of cycles #3 (slower cycle time) and #4 (faster cycle time) are compared to

experimental results in Table 6.4. The increase in pressure drop per unit column length is

comparable to the previous cycle comparison, except in this case, the rise in pressure

drop is due to an increase in cycle speed rather than a change in column diameter.

Table 6.4. Comparison of experimental and simulation results at different cycle times

using short, wide columns. (TPD = tons per day)

Slow cycle: cycle #3 Fast cycle: cycle #4

experiment simulation experiment simulation

Cycle time (s) 6.8 6.83 3.8 3.8

Avg. product purity (% oxygen) 96 96.2 96 96.1

Oxygen recovery (%) 45.5 45.5 41.0 42.7

BSF (lbs. solid/TPD O2) 81 83.6 51 53.3

Production step pressure drop

0.19 0.16 0.39 0.31
(kPa/cm)

Total pressure drop as % of

0.95 0.80 2.0 1.6
production step pressure

Purge step pressure drop (kPa/cm) - 0.28 - 0.45

Total pressure drop as % of purge

- 6.4 - 10.3
step pressure

122
 Figures 6.12 and 6.13 once again illustrate how the gas phase composition varies

over the column length at the end of the production and purge steps. After the production

step, there is a noticeable difference in the profiles as the faster cycle exhibits a greater

amount of MTZ spreading. This is at least partially responsible for the reduction in

recovery with increasing cycle speed. After the purge step, Figure 6.13 illustrates the

concentration profiles have nearly the same shape. Although there is a noticeable vertical

shift between the profiles, there is no apparent impact on the solid loading profile after

the purge step as indicated by Figure 6.14. As with the previous comparison, the P/F

ratio was held constant for the two cycles, thus the amount of purge gas used was again

about the same.

The columns achieving nearly identical loading profiles after the purge step seems

to contrast the previous simulation comparison. However, Figure 6.14 also shows less

nitrogen was adsorbed at the end of the production step for the faster cycle, which means

less desorption is required to achieve the same loading profile after the purge step. Thus,

even though the purge step of the faster cycle is less efficient due to column pressure

drop, the loading profiles after the purge step remain similar. The reduction in nitrogen

loading during the production step is likely the result of a combination of mass and heat

transfer resistances. At the feed end of the column where most of the

adsorption/desorption occurs, the difference in working capacity is caused by a resistance

to heat transfer. At the product end of the column, the difference in loading is the result

of a resistance to mass transfer. In the middle of the column, it is a combination of both

resistances.

123
Figure 6.12 Comparison of gas phase nitrogen composition at the end of the production

step for the slow cycle (cycle #3) and fast cycle (cycle #4)

Figure 6.13 Comparison of gas phase nitrogen composition at the end of the purge step

for the slow cycle (cycle #3) and fast cycle (cycle #4)

124
Figure 6.14 Comparison of solid nitrogen loading at the end of the production and

purge steps for the slow cycle (cycle #3) and fast cycle (cycle #4)

The previous conclusion is reasonably inferred by a combination of the previous

simulation comparison and Figure 6.15. The previous simulation comparison indicated

column pressure drop had a negligible impact on working capacity. Since a similar

increase in column pressure drop per length of column exists in this comparison, the

effect of pressure drop on working capacity is considered insignificant. Mass and heat

transfer resistances are the remaining primary contributors to the working capacity

reduction. Figure 6.15 shows the solid temperature profile along the length of the

column. At the feed end, the temperature difference reasonably accounts for a majority

of the difference in nitrogen loading between the two cycles. In the middle of the

column, the temperature profiles converge while the difference in nitrogen loading

125
between the cycles diverges. This indicates mass and heat transfer resistances both

contribute to the difference in loading in this region. At the product end of the column,

the solid temperature of the faster cycle is nearly the same as the slower cycle. Clearly,

the difference in loading at this end of the column is primarily due to mass transfer

resistances.

The simulation results therefore suggest the largest variation in the concentration

and loading profiles as cycle speed increases occurs during the production step. This

leads to the conclusion that for this study, mass and heat transfer resistances during the

production step are critical to the recovery and BSF of the process. It also confirms the

experimental results, which suggested column pressure drop was not a significant source

of process performance decline as cycling frequency increases.

Figure 6.15. Comparison of solid temperature profiles at the end of the production and

purge steps

126
6.3. Adsorbent Utilization Limit in Small Scale PSA

An adsorbent utilization limit (minimum BSF) represents another important

limitation of small scale PSA that is not well understood, especially for a two column

process. While some single column studies have indicated it exists,30,35,38,59 it has never

been demonstrated for a dual-column process. Hence, it is valuable to determine if a

minimum also occurs for a dual-column process, and more importantly what causes it.

The purpose of the last set of experiments was to measure a minimum BSF for a small

scale dual-column PSA process so that the cause of it may be better understood. The

experimental system and column size is the same as the pressure drop study, as outlined

in Chapter 5. The primary difference between the two studies is the removal of the

equalization step to simplify the analysis.

6.3.1 Measurement of a Minimum BSF

Figure 6.16 illustrates that oxygen recovery for both sets of columns declines with

faster cycling, which is consistent with the results from the pressure drop study. At each

pressure ratio, the decline is linear with cycle time until ~ 3.8 seconds, where the slope of

the decline increases. Figure 6.17 shows a minimum BSF around the same cycle time for

each pressure ratio, which demonstrates that the cycle time of a minimum BSF is not a

strong function of pressure ratio in our experimental range. Previous experimental results

for single column processes are mixed on whether pressure ratio affects the cycle time a

minimum BSF develops. One study showed pressure ratio affects the cycle time of a

minimum BSF (although no trend was found),78 while another shows it does not.38 The

minimum BSF developing at the same cycle time for both column sets confirms the

127
pressure drop study that showed pressure drop had an insignificant impact on process

performance. System limitations only permitted measurement of a minimum at two

pressure ratios for the short, wide column set.

Figure 6.16 Oxygen recovery results for long, thin columns (left) and short, wide

columns (right)

Figure 6.17 BSF results for long, thin columns (left) and short, wide columns (right)

128
 Experimental results were next used in equation 4.5 to evaluate its usefulness in

understanding the existence of a minimum BSF. To use equation 4.5, the empirical

constant, k’, was found using the following assumptions: 1) Gas velocity was estimated

from the conditions of the high pressure production step for each experiment 2) WCideal

was estimated from isotherm data at 20 °C, 195 kPa, and a nitrogen/oxygen composition

of 79/2152 3) Pressure drop was estimated using the Ergun equation76 using the velocity

of the high pressure productions step 4) K was determined through the expression

K=WCideal/(PH-PL) where PH and PL are the high and low pressure during the PSA

process respectively 5) LMTZ was estimated using the van Deemter model (equation 5.2);

hence, for practical PSA operation, the MTZ length is a linear function of gas velocity

with the slope determined by mass transfer resistance in the particle. For columns with

large particles, the eddy diffusion contribution (“A” term) is often relatively insignificant

compared to the particle resistance contribution (“C” term) and is sometimes ignored.

However, this term is significant for the particle size used in small scale PSA as

demonstrated in section 6.1. The particle resistance contribution is primarily controlled

by macropore diffusion since it represents the greatest mass transfer limitation for large

X-type zeolite particles.2, 24

While more complex models exist to estimate MTZ

spreading,51, 73, 74
they all have the same basic form and the van Deemter model was

demonstrated earlier to provide a suitable estimation. HETP was converted to MTZ by

assuming HETP ≈ HTU75 and LMTZ = NTU*HTU. The parameters in Table 6.1 were

used in the van Deemter model to determine HTU using the average velocity across the

MTZ; NTU was estimated ~ 4.5 from the results of the breakthrough experiments.

129
 k’ was found for each experiment using the above assumptions and plotted against

velocity in Figure 6.18. Although k’ is supposed to be constant, it clearly varies linearly

with velocity for both column sets. According to the regression analysis listed in Tables

6.5 and 6.6, the slope does not vary appreciably with pressure ratio; however, the y-

intercept is a function of pressure ratio. The regression analysis shows excellent

correlation of the data for each pressure ratio. k’ varying with velocity is not surprising

since LMTZ is found by estimating NTU from breakthrough experiments, which operate

with a completely regenerated column. Air separation via PSA operates with partial

regeneration, which decreases the driving force during adsorption relative to a fully

regenerated column and increases NTU. To account for this difference, k’ evidently must

vary with velocity. It is not clear if other factors also play a role in the variation of k’

with velocity.

Figure 6.18 k’ plotted against superficial velocity of high pressure feed step for long,

thin columns (left) and short, wide columns (right). Adsorption pressure = 195 kPa,

desorption

130
Table 6.5 Slope (m) and intercept (b) of k’ vs. superficial velocity in Figure 6.18 for

long columns. Std. error and R2 are from linear regression analysis

m*106 (s/cm2) Std. error*106 b*104 (1/cm) Std. error*104 R2

Pressure ratio 2.5 -1.3 0.1 1.71 0.066 0.967

Pressure ratio 3 -1.4 0.1 1.98 0.04 0.993

Pressure ratio 3.5 -1.4 0.1 2.17 0.073 0.958

Table 6.6 Slope (m) and intercept (b) of k’ vs. superficial velocity in Figure 6.18 for

short columns. Std. error and R2 are from linear regression analysis

m*106 (s/cm2) Std. error*106 b*104 (1/cm)Std. error*104 R2

Pressure ratio 3 -5.1 0.1 3.48 0.049 0.991

Pressure ratio 3.5 -5.3 0.2 3.87 0.066 0.996

131
 Figure 6.19 plots BSF against the production step velocity of every experiment.

At each pressure ratio, the plotted line represents the BSF from equation 4.5 using the

relationship between k’ and velocity from Tables 6.5 and 6.6. It is clear from Figure 6.19

that the minimum BSF shifts to a higher velocity as pressure ratio increases. This trend is

more pronounced for the long, thin column set.

Figure 6.19 BSF plotted against superficial velocity of high pressure feed step for long

columns. Ads. Pressure = 195 kPa, desorption pressure adjusted to achieve pressure

ratios listed. Solid line represents equation 4.5 predictions using experimentally

determined k’

132
6.3.2 What Causes a Minimum BSF?

Understanding what is causing the minimum to occur in our process would be

useful for future small scale PSA design. The primary potential contributors according to

equation 4.5 are pressure drop and MTZ spreading. Pressure drop was determined to

only have a very negligible impact on process performance through the study in section

6.2. Figure 6.20 further confirms this using equation 4.5. The dashed line plots equation

4.5 with pressure drop = 0 and the solid line plots equation 4.5 with pressure drop

predicted by the Ergun equation. There is no difference between the dashed line and

solid line until a gas velocity ~ 30 cm/s where a small deviation appears. However, the

minimum BSF is nearly identical in both cases, which indicates the pressure drop term in

equation 4.5 only minimally affects the minimum BSF.

Since the pressure drop contribution is so small, equation 4.5 indicates that the

MTZ length increasing with velocity is responsible for a minimum BSF. While mass and

heat transfer limitations increase the MTZ length, not all of them are a function of

velocity. Heat transfer resistances increase adsorbent temperature relative to isothermal

conditions, which stretches the MTZ. However, as cycle time decreases under rapid

cycling conditions, the time for heat transfer is reduced and the column approaches

adiabatic operation. When this occurs, the adsorbent temperature swing becomes nearly

constant. Hence, as gas velocity increases at the adiabatic limit, further MTZ spreading

due to heat effects is expected to be minimal. For very short cycle times (such as those in

this study), it is reasonable to assume the adiabatic limit is reached (or is very close to

being reached) and heat effects do not additionally spread the MTZ with increasing

velocity.

133
Figure 6.20 BSF plotted against superficial velocity of high pressure feed step for long

columns. Ads. Pressure = 195 kPa, desorption pressure adjusted to achieve pressure ratio

listed. Solid line represents equation 4.5 predictions using experimentally determined k’.

Dashed line represents Equation 4.5 prediction with ΔP = 0

134
 The van Deemter model (section 5.4.5) indicates the two dominant mass transfer

limitations that spread the MTZ in the range of our experiments are eddy diffusion axial

dispersion effects (“A” term) and mass transfer resistance in the particle (“C” term).

Axial dispersion effects due to eddy diffusion have previously been demonstrated to be

significant for the particle size used in small scale processes.26, 28, 31 However, according

to the van Deemter model, the axial dispersion contribution to MTZ spreading is constant

as cycle time decreases, similar to the contribution of heat effects under adiabatic

conditions. This is not the case for the “C” term in the van Deemter model, dominated by

macropore diffusional resistance, whose contribution to MTZ spreading does increase

with velocity. Earlier, it was demonstrated a transition from axial dispersion control to

macropore diffusion control exists for small LiLSX particles as gas velocity increases.

This transition occurs because the contribution to MTZ spreading from eddy diffusion

axial dispersion effects are not a function of gas velocity while the contribution from

macropore resistance does depend on velocity. As velocity increases (especially in the

range of fast PSA experiments), the axial dispersion contribution to the MTZ becomes

less important relative to the macropore diffusion contribution. Since the macropore

diffusional resistance contribution to MTZ spreading is the only limitation causing

additional MTZ spreading in the experimental range of this study, it is reasonable to

conclude the existence of a minimum BSF is primarily due to macropore resistance.

135
 CHAPTER VII

CONCLUSIONS

The conclusions of this dissertation are only for small scale processes with short

columns (10-30 cm in length) and particles similar in size to this study (0.5 mm).

Columns of much longer length or use of much smaller particles might invalidate these

conclusions as this may significantly increase the effect of pressure drop at similar cycle

times as this work.

7.1. On the Effects of Axial Dispersion in Small Scale PSA

While axial dispersion is known to increase with small adsorbent particles, it is

largely not considered significant for columns packed with small LiLSX particles. The

results of this study suggest this is no longer a good assumption as the overall mass

transfer rate in the column is significantly impacted by axial dispersion effects. The

implication of this observation for LiLSX particles is twofold in nature. First, the success

136
of typical methods to improve pore diffusion characteristics will be greatly limited since

macropore diffusion is no longer the only rate limiting mechanism. The other is the

estimation of axial dispersion effects in process modeling and breakthrough curve

analysis is critical for accurate model predictions and mass transfer rate parameter

estimations.

The results of this study further suggest measuring axial dispersion effects using

breakthrough curve analysis to determine an overall MTC can provide an accurate

estimation of axial dispersion effects. Literature correlations26 which correctly account

for the increase in axial dispersion for small particles also provide a reasonable

estimation. Not using one of these methods to describe axial dispersion effects either

leads to erroneous rate parameter estimation from breakthrough curves or overestimating

the MTC in process simulations.

7.2. On the Role of Pressure Drop in Small Scale PSA

While considering the effect of column pressure drop on a specific step of a PSA

cycle can be informative, it is difficult to determine the influence of individual steps on

the overall process performance. By studying the cumulative effect of column pressure

drop on process performance, it was determined its impact is minimal as cycling speed

increases. The results instead suggest mass and heat transfer resistances have a much

greater influence and limit operation performance before any significant impact of

column pressure drop. Several contributing factors are thought to be responsible for this

observation. The primary factor is the overall column pressure drop is small due to the

short column length often used in small scale PSA processes designed for a POC. The

137
other is that the end of the column most affected by column pressure drop (product end)

is the least loaded/unloaded with nitrogen, which somewhat mitigates the effect of

pressure drop on the column working capacity. This study includes and goes slightly

beyond the flow regime where Darcy’s law is valid. Thus, these conclusions are only

appropriate for this flow regime using similar size particles (dp ~ 0.5 mm). This regime is

often where small scale PSA processes using these particles operate. It should be noted

that cycles using much smaller particles, such as those described by Galbraith et al 79 (dp ~

0.15 mm), may have high enough pressure drop to invalidate these conclusions.

Column pressure drop is sometimes cited as a concern when using small particles

and small scale PSA. However, the experimental results suggest that within our

experimental range, a long, thin column design achieved a similar BSF with a higher

oxygen recovery than a short, wide column design. Since the long, thin column design

operated with roughly double the pressure drop per unit column length, pressure drop

clearly had little effect on process performance. Therefore, pressure drop concerns do

not seem justified for small scale air separation processes that operate in a flow regime

and with a column length and particle size similar to this study. While it should not be

ignored, especially for modeling purposes, its effect on process performance is negligible

compared to other more likely explanations, such as mass and heat transfer resistances.

7.3. On the Existence of a Minimum BSF

This study experimentally demonstrates a minimum BSF exists for a two column,

small scale PSA process. The cycle time a minimum BSF occurred was not found to be a

strong function of pressure ratio. The experimental data was further used to demonstrate

the usefulness of an available literature BSF model.44 It was found that the data was well

138
correlated using the model. Regression of the data was used to determine an empirical

parameter that permitted the model to be used to predict process performance.

It was finally proposed that macropore diffusional resistance was primarily

responsible for the existence of a minimum BSF in this dissertation as it was the only

mass/heat transfer limitation spreading the MTZ as a function of velocity in the range of

our experiments. While heat and axial dispersion effects are important, its contribution to

MTZ spreading is not a function of gas velocity in the experimental range of this work;

hence they cannot cause a minimum BSF to occur. These conclusions are likely to apply

to other small scale processes with a similar column length.

7.4. Summary of Intellectual Contributions

This dissertation experimentally demonstrates for the first time the significance of

axial dispersion effects for small LiLSX particles, especially for breakthrough

experiments. This is important because these particles are popular for POCs. The current

consensus in literature is that macropore resistance dominates the mass transfer rate,

however, this study clearly demonstrates this is no longer a sufficient assumption for

small particle breakthrough experiments where kinetic effects are evaluated.

Limitations of small scale PSA processes was the other focus of this dissertation.

While small scale PSA processes have much in common with larger PSA processes,

several key distinctions (e.g. rapid cycling and much smaller particles) make

understanding process limitations difficult. One potential limitation, pressure drop, was

explored further because its effects on overall process performance for small scale

systems has never been experimentally demonstrated. It was found not to have a

significant effect on process performance, which provides a wider design range for the

139
column dimensions. This is important to POC design because packaging all the

components into a portable process is tricky, and greater flexibility in column dimensions

is always desirable.

Another potential limitation, adsorbent utilization was also studied. A limit on

adsorbent utilization (minimum BSF) is important because it represents the limit on the

size and speed of a POC. There is no consensus currently on why a minimum BSF

occurs. A minimum BSF was experimentally found for a two-column small scale PSA

process for the first time in this study. The reason a minimum BSF occurred in this work

was proposed to be due to macropore diffusional resistance.

7.5. Future Work

The primary recommendation for future work is to vary particle size for each of

the studies in this dissertation. Currently in literature for small LiLSX particles, a

nitrogen MTC has been measured as a function of temperature, pressure, and velocity

(this study). Understanding how the MTC varies as a function of particle size is the last

variable needed to develop a generalized correlation for LiLSX particles. Shrinking

particle size may also increase pressure drop enough such that it impacts the minimum

BSF. Furthermore, running a small scale process with a different particle size would help

confirm that macropore diffusional resistance causes the minimum BSF. Other possible

future work includes measuring a minimum BSF with larger diameter columns (more

representative of industrial small scale systems) and demonstrating how the estimation of

the axial dispersion coefficient affects process simulation results.

140
 REFERENCES

1. Oxygen Therapy Market Size & Share, Industry Report, 2024.

https://www.grandviewresearch.com/industry-analysis/oxygen-therapy-market (accessed

1/19/2018).

2. Ruthven, D. M.; Xu, Z., Diffusion of oxygen and nitrogen in 5A zeolite crystals

and commercial 5A pellets. Chemical engineering science 1993, 48 (18), 3307-3312.

3. Masters, A. F.; Maschmeyer, T., Zeolites–From curiosity to cornerstone.

Microporous and Mesoporous Materials 2011, 142 (2), 423-438.

4. Weitkamp, J., Zeolites and catalysis. Solid State Ionics 2000, 131 (1), 175-188.

5. Chica, A., Zeolites: promised materials for the sustainable production of

hydrogen. ISRN Chemical Engineering 2013, 2013.

6. Sherman, J. D., Synthetic zeolites and other microporous oxide molecular sieves.

Proceedings of the National Academy of Sciences 1999, 96 (7), 3471-3478.

7. Thomas, W. J.; Crittenden, B. D., Adsorption technology and design.

Butterworth-Heinemann: 1998.

141
8. Ruthven, F.; Farooq, S., Knaebel, Pressure Swing Adsorption. VCH Publishers,

New York: 1994.

9. Yang, R. T., Gas separation by adsorption processes. Butterworth-Heinemann:

2013.

10. Motoyuki, S., Adsorption engineering. Eles. Sci. Pub 1990, 5-61.

11. Grande, C., Advances in Pressure Swing Adsorption for Gas Separation. ISRN

Chemical Engineering 2012, 2012, 13.

12. Sircar, S., Pressure Swing Adsorption. Industrial & Engineering Chemistry

Research 2002, 41 (6), 1389-1392.

13. Skarstrom, C. W., Method and apparatus for fractionating gaseous mixtures by

adsorption. US Patent 2,944,627. 1960.

14. Daniel, D.; De, M. P. G., Process for separating a binary gaseous mixture by

adsorption. US Patent 3,155,468. 1964.

15. Pressure equalization depressuring in heatless adsorption. US Patent 3,142,547A.

1964.

142
16. Chatburn, R. L.; Williams, T. J., Performance comparison of 4 portable oxygen

concentrators. Respiratory care 2010, 55 (4), 433-442.

17. Koretsky, M. D., Engineering and chemical thermodynamics. Wiley Hoboken,

NJ: 2004; Vol. 2.

18. Prausnitz, J. M.; Lichtenthaler, R. N.; de Azevedo, E. G., Molecular

thermodynamics of fluid-phase equilibria. Pearson Education: 1998.

19. Papai, I.; Goursot, A.; Fajula, F.; Plee, D.; Weber, J., Modeling of N2 and O2

adsorption in zeolites. The Journal of Physical Chemistry 1995, 99 (34), 12925-12932.

20. McKee, D. W., Separation of an oxygen-nitrogen mixture. US Patent 3,140,933

1964.

21. Chao, C. C., Process for separating nitrogen from mixtures thereof with less polar

substances. US Patent 4,859,217. 1989.

22. Langmuir, I., The adsorption of gases on plane surfaces of glass, mica and

platinum. Journal of the American Chemical society 1918, 40 (9), 1361-1403.

143
23. Mathias, P. M.; Kumar, R.; Moyer, J. D.; Schork, J. M.; Srinivasan, S. R.; Auvil,

S. R.; Talu, O., Correlation of Multicomponent Gas Adsorption by the Dual-Site

Langmuir Model. Application to Nitrogen/Oxygen Adsorption on 5A-Zeolite. Industrial

& Engineering Chemistry Research 1996, 35 (7), 2477-2483.

24. Ruthven, D. M., Principles of adsorption and adsorption processes. John Wiley

& Sons: 1984.

25. Glueckauf, E., Theory of chromatography. Part 10.—Formulæ for diffusion into

spheres and their application to chromatography. Transactions of the Faraday Society

1955, 51, 1540-1551.

26. Langer, G.; Roethe, A.; Roethe, K. P.; Gelbin, D., Heat and Mass-Transfer in

Packed Beds 3. Axial Mass Dispersion. International Journal of Heat and Mass Transfer

1978, 21 (6), 751-759.

27. Haq, N.; Ruthven, D. M., A chromatographic study of sorption and diffusion in

5A zeolite. Journal of colloid and interface science 1986, 112 (1), 164-169.

28. Zhong, G.; Rankin, P. J.; Ackley, M. W. High frequency PSA process for gas

separation. US Patent 7,828,878. November 2010.

144
29. Alpay, E.; Kenney, C. N.; Scott, D. M., Adsorbent particle size effects in the

separation of air by rapid pressure swing adsorption. Chemical Engineering Science

1994, 49 (18), 3059-3075.

30. Rao, V. R.; Farooq, S., Experimental Study of a Pulsed-Pressure-Swing-

Adsorption Process with Very Small 5A Zeolite Particles for Oxygen Enrichment.

Industrial & Engineering Chemistry Research 2014, 53 (33), 13157-13170.

31. Suzuki, M.; Smith, J. M., Axial dispersion in beds of small particles. The

Chemical Engineering Journal 1972, 3, 256-264.

32. Wu, C.-W.; Kothare, M. V.; Sircar, S., Column Dynamic Study of Mass Transfer

of Pure N2 and O2 into Small Particles of Pelletized LiLSX Zeolite. Industrial &

Engineering Chemistry Research 2014, 53 (45), 17806-17810.

33. Chai, S. W.; Kothare, M. V.; Sircar, S., Rapid Pressure Swing Adsorption for

Reduction of Bed Size Factor of a Medical Oxygen Concentrator. Industrial &

Engineering Chemistry Research 2011, 50 (14), 8703-8710.

34. Rao, V. R.; Farooq, S.; Krantz, W. B., Design of a two-step pulsed pressure-swing

adsorption-based oxygen concentrator. AIChE Journal 2010, 56 (2), 354-370.

145
35. Rao, V. R.; Kothare, M. V.; Sircar, S., Novel design and performance of a

medical oxygen concentrator using a rapid pressure swing adsorption concept. AIChE

Journal 2014, 60 (9), 3330-3335.

36. Santos, J. C.; Portugal, A. F.; Magalhães, F. D.; Mendes, A., Simulation and

Optimization of Small Oxygen Pressure Swing Adsorption Units. Industrial &

Engineering Chemistry Research 2004, 43 (26), 8328-8338.

37. Santos, J. C.; Portugal, A. F.; Magalhães, F. D.; Mendes, A., Optimization of

Medical PSA Units for Oxygen Production. Industrial & Engineering Chemistry

Research 2006, 45 (3), 1085-1096.

38. Zhu, X.; Liu, Y.; Yang, X.; Liu, W., Study of a novel rapid vacuum pressure

swing adsorption process with intermediate gas pressurization for producing oxygen.

Adsorption 2017, 23 (1), 175-184.

39. Ackley, M. W.; Barrett, P. A.; Stephenson, N. A.; Kikkinides, E. S., High rate

compositions. US Patent 9,533,280 B2. 2017.

40. Ackley, M. W.; Smolarek, J.; Leavitt, F. W. Pressure swing adsorption gas

separation method, using adsorbents with high intrinsic diffusivity and low pressure

ratios. 14 January 2003.

146
41. Weston, K.; Jaussaud, D.; Chiang, R. L. Lithium exchanged zeolite X adsorbent

blends. US Patent 7,300,899 B2. 2007.

42. Zheng, J.; Barrett, P. A.; Pontonio, S. J.; Stephenson, N. A.; Chandra, P.;

Kechagia, P., High-rate and high-density gas separation adsorbents and manufacturing

method. Adsorption 2014, 20 (1), 147-156.

43. Garg, D. R.; Ruthven, D. M., Performance of molecular sieve adsorption

columns: combined effects of mass transfer and longitudinal diffusion. Chemical

Engineering Science 1975, 30 (9), 1192-1194.

44. Rezaei, F.; Webley, P., Optimum structured adsorbents for gas separation

processes. Chemical Engineering Science 2009, 64 (24), 5182-5191.

45. Chai, S. W.; Kothare, M. V.; Sircar, S., Numerical study of nitrogen desorption by

rapid oxygen purge for a medical oxygen concentrator. Adsorption 2012, 18 (2), 87-102.

46. Rao, V. R.; Kothare, M. V.; Sircar, S., Numerical simulation of rapid

pressurization and depressurization of a zeolite column using nitrogen. Adsorption 2014,

20 (1), 53-60.

47. Vemula, R. R.; Kothare, M. V.; Sircar, S., Anatomy of a rapid pressure swing

adsorption process performance. AIChE Journal 2015, 61 (6), 2008-2015.

147
48. Edwards, M. F.; Richardson, J. F., Gas dispersion in packed beds. Chemical

Engineering Science 1968, 23 (2), 109-123.

49. Moulijn, J. A.; Van Swaaij, W. P. M., The correlation of axial dispersion data for

beds of small particles. Chemical Engineering Science 1976, 31 (9), 845-847.

50. Wakao, N.; Kaguei, S.; Nagai, H., Effective diffusion coefficients for fluid

species reacting with first order kinetics in packed bed reactors and discussion on

evaluation of catalyst effectiveness factors. Chemical Engineering Science 1978, 33 (2),

183-187.

51. Gritti, F.; Guiochon, G., General HETP equation for the study of mass-transfer

mechanisms in RPLC. Analytical chemistry 2006, 78 (15), 5329-5347.

52. Moran, A.; Talu, O., Role of Pressure Drop on Rapid Pressure Swing Adsorption

Performance. Industrial & Engineering Chemistry Research 2017, 56 (19), 5715-5723.

53. Chai, S. W.; Kothare, M. V.; Sircar, S., Rapid Pressure Swing Adsorption for

Reduction of Bed Size Factor. Industrial & Engineering Chemistry Research 2011, 50,

8703-8710.

148
54. Turnock, P. H.; Kadlec, R. H., Separation of nitrogen and methane via periodic

adsorption. AIChE Journal 1971, 17 (2), 335-342.

55. Jones, R. L.; Keller, G. E., Pressure swing parametric pumping—a new

adsorption process. J. Separ. Proc. Technol 1981, 2 (3), 17.

56. Jee, J.-G.; Lee, J.-S.; Lee, C.-H., Air Separation by a Small-Scale Two-Bed

Medical O2 Pressure Swing Adsorption. Industrial & Engineering Chemistry Research

2001, 40 (16), 3647-3658.

57. Rama Rao, V.; Wu, C. W.; Kothare, M. V.; Sircar, S., Comparative perfomances

of two commercial samples of LiLSX zeolite for production of 90% oxygen from air by a

novel rapid pressure swing adsorption system. Separation Science and Technology 2015,

50 (10), 1447-1452.

58. Chai, S. W.; Kothare, M. V.; Sircar, S., Rapid Pressure Swing Adsorption for

Reduction of Bed Size Factor of a Medical Oxygen Concentrator. Industrial &amp;

Engineering Chemistry Research 2011, 50 (14), 8703-8710.

59. Todd, R. S.; Webley, P. A., Mass-transfer models for rapid pressure swing

adsorption simulation. AIChE Journal 2006, 52 (9), 3126-3145.

149
60. Buzanowski, M. A.; Yang, R. T.; Haas, O. W., Direct observation of the effects of

bed pressure drop on adsorption and desorption dynamics. Chemical Engineering Science

1989, 44 (10), 2392-2394.

61. Lu, Z. P.; Loureiro, J. M.; Rodrigues, A. E.; LeVan, M. D., Pressurization and

blowdown of adsorption beds—II. Effect of the momentum and equilibrium relations on

isothermal operation. Chemical engineering science 1993, 48 (9), 1699-1707.

62. Sundaram, N.; Wankat, P. C., Pressure drop effects in the pressurization and

blowdown steps of pressure swing adsorption. Chemical Engineering Science 1988, 43

(1), 123-129.

63. Hart, J.; Battrum, M. J.; Thomas, W. J., Axial pressure gradients during the

pressurization and depressurization steps of a PSA gas separation cycle. Gas Separation

& Purification 1990, 4 (2), 97-102.

64. Kikkinides, E. S.; Yang, R. T., Effects of bed pressure drop on isothermal and

adiabatic adsorber dynamics. Chemical Engineering Science 1993, 48 (9), 1545-1555.

65. Sircar, S.; Golden, T. C., Isothermal and Isobaric Desorption of Carbon Dioxide

by Purge. Industrial & Engineering Chemistry Research 1995, 34 (8), 2881-2888.

150
66. Yang, J.; Park, M.-W.; Chang, J.-W.; Ko, S.-M.; Lee, C.-H., Effects of pressure

drop in a PSA process. Korean Journal of Chemical Engineering 1998, 15 (2), 211-216.

67. Zhou, Q., Pure Component Adsorption of Methane, Ethylene, Propylene and

Carbon Dioxide in Silicalite. Masters Thesis. 2013.

68. Delgado, J. M. P. Q., A critical review of dispersion in packed beds. Heat and

Mass Transfer 2006, 42 (4), 279-310.

69. Patel, M. S., Evaluation of Mass Transfer Rate in Column of Small LiLSX

Particles. Masters Thesis. 2017.

70. Kumar, R.; Sircar, S., Skin resistance for adsorbate mass transfer into extruded

adsorbent pellets. Chemical engineering science 1986, 41 (9), 2215-2223.

71. Wankat, P. C., Large-Scale Adsorption and Chromatography (Volumes 1-2).

1986.

72. van Deemter, J. J.; Zuiderweg, F. J.; Klinkenberg, A., Longitudinal diffusion and

resistance to mass transfer as causes of nonideality in chromatography. Chemical

Engineering Science 1956, 5 (6), 271-289.

151
73. Knox, J. H., Band dispersion in chromatography—a universal expression for the

contribution from the mobile zone. Journal of Chromatography A 2002, 960 (1), 7-18.

74. Knox, J. H., Band dispersion in chromatography–a new view of A-term

dispersion. Journal of Chromatography A 1999, 831 (1), 3-15.

75. Cussler, E. L., Diffusion: mass transfer in fluid systems. Cambridge university

press: 2009.

76. Todd, R. S.; Webley, P. A., Pressure Drop in a Packed Bed under Nonadsorbing

and Adsorbing Conditions. Industrial & Engineering Chemistry Research 2005, 44 (18),

7234-7241.

77. Ackley, M. W.; Leavitt, F. W. Rate-enhanced gas separation. US Patent

6,500,234. 2002.

78. Vemula, R. R.; Kothare, M. V.; Sircar, S., Performance of a medical oxygen

concentrator using rapid pressure swing adsorption process: effect of feed air pressure.

AIChE Journal 2016, 62 (4), 1212-1215.

79. Galbraith, S. D.; McGowan, K. J.; Baldauff, E. A.; Galbraith, E.; Walker, D. K.;

LaCount, R. B., Ultra rapid cycle portable oxygen concentrator. U.S. Patent 8,894,751.

2014.

152
 APPENDIX A

PSA MODEL

A.1. Simulation model description

The model is derived under the assumptions of ideal gas behavior, negligible radial

concentration and temperature gradients, and axially dispersed plug flow. MATLAB

R2014a was used to solve the system of equations presented in this section. The Crank-

Nicholson finite differencing method was used to solve the equation through a time and

space discretization of the column. Pressure equalization steps are modeled using the

same time and space discretization method used for other pressure changing steps in the

process.

153
A.2. Mass balance

The column material balance is represented by the axial dispersed plug flow model:

Axial dispersion is found using a model developed by Wakao and Funazkri1:

where Dm is the molecular diffusivity, u is the interstitial velocity, dp is the particle

diameter, and γ1 and γ2 are constants typically assumed to be 0.7 and 0.5 respectively.2

The adsorption rate was described with a linear driving force (LDF) model described as:2

LDF coefficients were found from performing breakthrough experiments on the same

material in a separate breakthrough apparatus.

A.3. Equilibrium model

Adsorption equilibrium was formulated by the dual-site Langmuir (DSL) model.3 The

pure component amount adsorbed, ni*, was represented by equation A4 where the

equilibrium constants show an Arrhenius relationship to temperature involving

adsorption energy and entropy, as expressed by:

154
For simulation studies, the parameters in Table A1 were used according to equation A7

for binary equilibrium.

Table A1. Nitrogen and oxygen DSL parameters for LiLSX particles used in

simulations.

Nitrogen Oxygen

ns,1 (mol/kg) 2.025 1.426

b0 x 10-6 (1/kPa) 0.124 0.158

Qb /R (K) 3365 2373

ns,2 (mol/kg) 2.402 6.177

d0 x 10-6 (1/kPa) 0.513 3.386

Qd /R (K) 2263 1211

155
A.4. Energy balance

The energy balance is broken down into three separate differential equations. The first

is described by equation A8 and represents the energy balance of the gas phase.

Equation A9 describes the solid phase energy balance:

where Qi is found by:

(A10)

The gas solid heat transfer coefficient (hs) is found from equation A11.

(A11)

The last energy balance equation is for the column wall and is described as:

The gas-wall heat transfer coefficient (hw) is found from equation A13.

156
 (A13)

The external heat transfer coefficient (hext) was found by using natural convection from

the surface of the column.

A.5. Momentum balance

Column pressure drop was modeled using the Ergun equation.4

A.6. Boundary conditions

The mass balance boundary condition for steps with an inlet gas flow is described

in equation A15:

where z = 0 is the end of the column where gas is entering and the velocity is set by

equation A16 depending on the step.

For steps with an outlet gas flow, the boundary condition described by equation A17:

where z=L is the end of the column where the gas is exiting. For constant pressure steps,

the total gas concentration at the inlet of the column is fixed. For varying pressure steps,

the gas velocity is fixed at the outlet of the column.

157
 The energy balance boundary condition for steps with an inlet gas flow is described in

equation A18:

where z = 0 is the end of the column where gas is entering For steps with an outlet gas

flow, the boundary condition is described by equation A19:

where z=L is the end of the column where gas is exiting.

A.7. Model parameters

Table A2. Model parameters used for simulations.

Value used in
Parameter
model
dp (cm) 0.05

ρb (g/cm3) 0.58

ρw (g/cm3) 7.74

εb 0.35
εp 0.65

cp,w (J/kg-K) 490

cp,s (J/kg-K) 830

kw (W/m-K) 16
ks (W/m-K) 1.3

Tamb (K) 296.15

158
 APPENDIX B

BREAKTHROUGH EXPERIMENTS

B.1. Tabulated data

Table B1. Results of breakthrough experiments. Inlet conditions: pressure = 267 kPa,
ambient temperature, N2/O2 gas composition 75/25. Note: superficial velocity and
Reynolds number based on average value across MTZ

Superficial MTC
Trial # Reynolds number MTZ length (cm)
velocity (cm/s) (1/s)
1 1.1 0.9 0.48 1.06
2 1.4 1.2 0.64 1.02
3 1.7 1.4 0.80 0.96
4 1.9 1.6 1.03 1.01
5 2.2 1.9 1.10 1.03
6 2.2 1.9 1.25 0.97
7 2.5 2.1 1.24 0.93
8 2.8 2.3 1.33 0.92
9 3.3 2.8 1.61 1.00
10 4.4 3.7 2.27 1.07
11 5.5 4.7 3.29 0.98
12 6.6 5.6 3.29 1.10
13 7.7 6.5 4.28 1.10
14 8.9 7.5 3.79 1.27
15 10.0 8.4 4.45 1.26

159
B.2. Examples of breakthrough profiles

0.8

0.6
C/C0

0.4
trial 14
0.2 trial 11
trial 9
0
0 20 40 60
Time (s)

Figure B1. Breakthrough profiles representative of the velocity range covered in the
study. Experiments are at inlet conditions described in Table S1. Average Reynolds
number across MTZ: trial 14 = 7.5, trial 11 = 4.7, and trial 9 = 2.8

B.3. Heat effects

Figure B2 illustrates the heat effects present in our system. At the exact same

conditions as the matched experimental curve, the model was run under isothermal and

adiabatic conditions. It is clear that the matched experiment is very near adiabatic

conditions. This figure also illustrates how the 0.9 cutoff was determined for defining the

MTZ zone. For an isothermal MTZ, 0.95 is the more common cutoff. However, since

the adiabatic response is stretched compared to the isothermal one, the location is

matched. In other words, a 0.9 cut-off for the adiabatic case provides a reasonable

estimate for a 0.95 cut-off on the isothermal curve. Since heat effects are relatively

constant in the range of experiments, the difference between the experimental and

isothermal curve should also be relatively constant, which justifies using the 0.9

160
approximation for all the experiments. This approximation is not valid in the gas velocity

range below our experimental velocities where heat transfer with the environment is more

significant.

0.8

0.6
C/C0

0.4
Isothermal
0.2 Adiabatic
Matched experiment
0
0.8 0.9 1
z/z0

Figure B2. Example of how heat transfer affects experimental results. Matched
experiment is run #11. Dashed lines represent 0.95 and 0.9 cutoffs

B.4. Model framework

The amount adsorbed (nN2) in Equation 5.1 of the main text is found based on the

assumption of a constant pattern forming in the column. The constant pattern velocity

defined by uw can be expressed as:5

(B1)

where M can be molar flux, gas composition, amount adsorbed or temperature. If Ni =

ρbni+ɛbCi, Equation 3.11 of the main text can be written as Equation B2 assuming plug

flow. A differential mass balance at z and t gives:5

161
 (B2)

Equation B1 and B2 are combined to give:

(B3)

Assuming the concentration and amount adsorbed (Ci,f, and ni,f) ahead of the MTZ equal

zero, Equation B3 can be integrated to get the following relationship inside the MTZ:

(B4)

uw was found experimentally from Equation 5.4 in main text and the term uCi can be

measured experimentally at the exit (which was numerically found using the simulated

exit breakthrough profile that was matched to the experimental curve). The ni* in

Equation 5.1 is the equilibrium value corresponding to exit gas conditions at that instant.

An example of integrand is shown in Figure B3 for one of the runs.

162
 12

10
1/(n*-n) (kg/mol)
8

0
0 0.3 0.6 0.9 1.2
n (mol/kg)

Figure B3. Plot of nitrogen integrand for trial 8 between limits of MTZ (C/C0 =.05 and
C/C0 = 0.9)

B.5. Void volume analysis

External void volume is known to affect chromatography results and

breakthrough column analysis.6,7 However, this primarily affects low gas velocity

experiments with a dilute feed composition. As has been demonstrated by Rajendran et

al.,7 it is necessary to correct for the external void volume effects under these conditions.

However, they also demonstrate that this correction loses significance for a concentrated

feed at high gas velocities, which are the conditions used in this study. Furthermore, the

residence time in the void volume for our experimental system was estimated to be over a

magnitude lower than the residence time of the MTZ in the column. For example, at the

conditions of run #13, it is estimated the residence time in the void volume is ~ 1s, while

the residence time for the MTZ in the column is ~ 23s.

163
 Despite the previous evidence that suggests the void volume effect is negligible, a

brief study on the effect of void volume in our system was first performed for the

experimental system. Correcting for external void volume effects usually involves using

a zero void bypass column to estimate the effect of the external void volume on the

response to breakthrough. This response is then subtracted from the breakthrough curve

when the adsorbent material is used. A different and simpler procedure was used just to

show that external volume effects in our system were negligible. In our experimental

setup, we have the ability to move the mass spectrometer closer to the exit of the column,

effectively eliminating half of the external void volume. This prevents us from using the

exit flow meter, however, it does allow us to compare the composition response under the

same conditions using two different external void volumes.

Two different gas velocities are highlighted to illustrate how the void volume

effects change with velocity. The lower velocity example (Figure B4) shows the void

volume has a very slight effect on the breakthrough response, but that this difference is

not detectable at the higher velocity example in Figure B5. As previously mentioned,

void volume effects should affect lower gas velocity experiments more, so this was not a

surprising result. Considering all of the experiments in this study was run at gas

velocities near the velocity in Figure B4 or much higher, it is reasonable to assume that

the void volume affects are negligible for these experiments.

164
 1
0.9
0.8
0.7
0.6
C/Co

0.5
0.4
0.3
0.2
Full void volume
0.1 Reduced void volume
0
0.95 1 1.05 1.1 1.15 1.2
t/tc

Figure B4. Breakthrough response for two different void volumes. Conditions:
superficial velocity: 1.5 cm/s, N2/He = 70/30, pressure: 240 kPa, room temperature

1
0.9
0.8
0.7
0.6
C/Co

0.5
0.4
0.3
Full void volume
0.2
0.1 Reduced void volume
0
0.95 1.05 1.15
t/tc

Figure B5. Breakthrough response for two different void volumes. Conditions:
superficial velocity: 7.4 cm/s, N2/He = 85/15, pressure: 155 kPa, room temperature

165
B.6. Effect of using wrong MTC on breakthrough profile

The consequence of using the wrong correlation to predict an overall MTC is

visually illustrated in Figure B6. Using the Case 1 prediction from Section 4.1 of the

main text results in a significantly steeper breakthrough curve compared to the more

accurate Case 4 prediction. The Case 1 prediction results in a nearly two-fold reduction

in MTZ time compared to the Case 4 prediction when the 0.05 and 0.9 limits used to

define the MTZ are used. Assuming the wave velocities are approximately equivalent,

this results in a MTZ length prediction that is nearly twice as small as demonstrated

experimentally.

0.8

0.6
C/Co

0.4

0.2 Experiment
Case 1 Prediction
Case 4 Prediction
0
45 46 47 48 49 50
Time (s)

Figure B6. Comparison of breakthrough profiles using different predictions for an

overall MTC. Experimental breakthrough curve at 267 kPa, room temperature and an
inlet N2/O2 gas mixture of 0.75/0.25. The average superficial velocity across MTZ was
2.8 cm/s

166
 APPENDIX C

PRESSURE DROP EXPERIMENTS

C.1. Column pressure profile

Figure C1 provides both simulation and experimental pressure profiles for Cycle #10 at

the product end of the process. It demonstrates the simulator captures the pressure

change reasonably well when plotted against experimental data. There is some

discrepancy between the model and experiment during the high pressure production step.

It is difficult to determine if this is due to the experimental setup or a simulation

limitation. Nevertheless, it seems to have little effect on the ability of the model to

reasonably approximate the oxygen recovery and BSF as evidenced by the comparison of

experimental and simulation process results.

167
Figure C1. Experimental and simulated pressure profile for Cycle #10.

C.2. Tabulated experimental results

Table C2. Short columns experimental results.

% Oxygen BSF (lbs. solid/TPD

Recovery O2)
Cycle 1 48.2 164

Cycle 2 47.3 114

Cycle 3 45.5 81

Cycle 4 41.0 51
Pressure ratio 4.5, P/F ratio 0.4, product purity:
96% oxygen, production step pressure ~195 kPa

168
Table C3. Long columns experimental results.

% Oxygen BSF (lbs. solid/TPD

Recovery O2)
Cycle 5 34.5 303

Cycle 6 33.2 219

Cycle 7 30.9 172

Cycle 8 26.3 116

Cycles 5-8: Pressure ratio 2.5, P/F ratio 0.6, product
purity: 96% oxygen , production step pressure ~195 kPa

Cycle 9 51.3 113

Cycle 10 50.3 80

Cycle 11 45.4 55

Cycle 12 43.9 52
Cycles 9-12: Pressure ratio 4.5, P/F ratio 0.4, product
purity: 96% oxygen , production step pressure ~195 kPa

169
 APPENDIX D

MINIMUM BSF EXPERIMENTS

D.1. Tabulated experimental results for minimum BSF study

Table D1. Long columns experimental results, pressure ratio 3.5

Total cycle % Oxygen BSF (lbs.

time (s) Recovery solid/TPD O2)

Cycle 1 6 18.8 140

Cycle 2 3.9 14.3 119

Cycle 3 3.3 12.7 116

Cycle 4 3.2 11.0 132

Cycle 5 6 19.4 138

Cycle 6 4.2 15.3 123

Cycle 7 3.36 12.2 123

Cycle 8 3.12 11.1 126

Cycle 9 5.28 17.0 140

Pressure ratio 3.5, P/F ratio 0.45, product purity 96%, oxygen
production step pressure ~195 kPa

170
Table D2. Long columns experimental results, pressure ratio 3

Total
% Oxygen BSF (lbs.
cycle time
Recovery solid/TPD O2)
(s)
Cycle 10 6.2 15.3 195

Cycle 11 3.1 8.8 167

Cycle 12 2.72 6.6 193

Cycle 13 2.92 7.5 184

Cycle 14 4.9 13.4 176

Cycle 15 3.92 11.6 161

Pressure ratio 3, P/F ratio 0.47, product purity: 96% oxygen , production
step pressure ~195 kPa

Table D3. Long columns experimental results, pressure ratio 2.5

Total cycle % Oxygen BSF (lbs.

time (s) Recovery solid/TPD O2)

Cycle 16 6.4 10.8 318

Cycle 17 5.3 10.1 276

Cycle 18 3.2 7.0 240

Cycle 19 2.6 3.8 353

Cycle 20 2.9 5.4 276

Cycle 21 2.6 4.2 318

Cycle 22 4.4 9.3 259

Cycle 23 3.8 8.9 231

Pressure ratio 2.5, P/F ratio 0.5, product purity: 96% oxygen , production
step pressure ~195 kPa

171
Table D4. Short columns experimental results, pressure ratio 3.5

Total cycle % Oxygen BSF (lbs.

time (s) Recovery solid/TPD O2)

Cycle 24 6 14.5 168

Cycle 25 3.6 9.0 151

Cycle 26 3.1 6.4 175

Cycle 27 3.3 7.6 159

Cycle 28 4.84 12.7 153

Cycle 29 4.2 10.9 151

Cycle 30 8.4 17.0 199

Cycle 31 3.9 9.5 153

Cycle 32 3.5 7.6 170

Cycle 33 3.2 6.3 185

Pressure ratio 3.5, P/F ratio 0.45, product purity 96%, oxygen
production step pressure ~195 kPa

Table D5. Short columns experimental results, pressure ratio 3

Total cycle % Oxygen BSF (lbs.

time (s) Recovery solid/TPD O2)

Cycle 34 6.4 12.1 226

Cycle 35 6.4 12.2 230

Cycle 36 3.7 7.1 210

Cycle 37 3.2 5.3 235

Cycle 38 2.94 4.3 255

Cycle 39 4.64 9.2 210

Cycle 40 9.28 13.4f 287

Pressure ratio 3, P/F ratio 0.47, product purity: 96% oxygen , production
step pressure ~195 kPa

172
 NOMENCLATURE

b0 = adsorption equilibrium constant for site 1 of DSL model

Ci = gas phase concentration of species i

Ci,f = Ci ahead of MTZ

CT = total gas phase concentration

cp.g = specific heat of gas

cp,s = specific heat of solid adsorbent

cp,w = specific heat of column wall

cv,a = specific heat of adsorbed phase

cv,g = specific heat of gas

d0 = adsorption equilibrium constant for site 2 of DSL model

DL= axial dispersion coefficient

Dm= molecular diffusivity

db,i = inner column diameter

db,o = outer column diameter

dp = adsorbent particle diameter

hext = wall-ambient heat transfer coefficient

hs = gas-solid heat transfer coefficient

hw = gas-wall heat transfer coefficient

ki = mass transfer coefficient of component i

kg = gas thermal conductivity

ks = adsorbent thermal conductivity

kw = wall thermal conductivity

ni = adsorbed phase concentration over an adsorbent particle of component i

173
ni* = equilibrium amount adsorbed of component i

ns,1 = adsorption equilibrium capacity constant for site 1 of DSL model

ns,2 = adsorption equilibrium capacity constant for site 2 of DSL model

ni,f = ni ahead of MTZ

P = pressure

Pr = Prandtl number

Qb = heat of adsorption for site 1 of DSL model

Qd = heat of adsorption for site 2 of DSL model

Qi = heat of adsorption of component i

R = ideal gas constant

Re = Reynolds number

T = temperature

t = time

tc = time for stoichiometric center to leave column

Tamb = ambient temperature

Tg = gas temperature

Ts = solid temperature

Tw = wall temperature

u = interstitial gas velocity

uw = wave (constant profile) velocity

yi = gas phase composition of component i

z = axial coordinate

Greek Letters

γ1 = axial dispersion constant

γ2 = axial dispersion constant

174
εb = bed/column porosity

εp = particle porosity

λw = ratio of column outer diameter to inner diameter

μg = gas viscosity

ρb = column bulk density

ρg = gas density

ρw = wall density

Subscripts

i,j = component

175
 REFERENCES

1. Wakao, N.; Funazkri, T., Effect of fluid dispersion coefficients on particle-to-

fluid mass transfer coefficients in packed beds: correlation of Sherwood numbers.

Chemical Engineering Science 1978, 33 (10), 1375-1384.

2. Ruthven, D. M., Principles of adsorption and adsorption processes. John Wiley

& Sons: 1984.

3. Mathias, P. M.; Kumar, R.; Moyer, J. D.; Schork, J. M.; Srinivasan, S. R.; Auvil,

S. R.; Talu, O., Correlation of Multicomponent Gas Adsorption by the Dual-Site

Langmuir Model. Application to Nitrogen/Oxygen Adsorption on 5A-Zeolite. Industrial

& Engineering Chemistry Research 1996, 35 (7), 2477-2483.

4. Todd, R. S.; Webley, P. A., Pressure Drop in a Packed Bed under Nonadsorbing

and Adsorbing Conditions. Industrial & Engineering Chemistry Research 2005, 44 (18),

7234-7241.

5. Wu, C.-W.; Kothare, M. V.; Sircar, S., Column Dynamic Study of Mass Transfer

of Pure N2 and O2 into Small Particles of Pelletized LiLSX Zeolite. Industrial &

Engineering Chemistry Research 2014, 53 (45), 17806-17810.

176
6. Gritti, F.; Felinger, A.; Guiochon, G., Influence of the errors made in the

measurement of the extra-column volume on the accuracies of estimates of the column

efficiency and the mass transfer kinetics parameters. Journal of Chromatography A 2006,

1136 (1), 57-72.

7. Rajendran, A.; Kariwala, V.; Farooq, S., Correction procedures for extra-column

effects in dynamic column breakthrough experiments. Chemical Engineering Science

2008, 63 (10), 2696-2706.

177