Thesis Doctor of Philosophy GUO Yalou

Development of new pressure
swing adsorption (PSA) cycles
Yalou Guo
Student ID: 1063461
ORCID ID: 0000-0003-4052-9681
Department of Chemical Engineering
University of Melbourne
The dissertation is submitted for the degree of
Doctor of Philosophy
March 2023
Declaration
I hereby declare that this Ph.D. thesis entitled “Development of new pressure swing
adsorption (PSA) cycles” comprises only my original work towards to the degree of
Doctor of Philosophy in Chemical Engineering under the guidance and supervision of
A/Prof. Gang Kevin Li and Prof. Paul A. Webley. It is to certify that the thesis is less than
100,000 words in total, exclusive of tables, pictures, bibliographies, and appendix.
Yalou Guo
March 2023
I
Abstract
Pressure swing adsorption (PSA) is an adsorption-based process, which has been widely
used and studied for gas mixture separation due to its low investment and operating cost
and high automation. Numerous novel concepts based on basic features of traditional PSA
cycles, such as simulated moving bed (SMB)-PSA, dual reflux (DR)-PSA and layered
PSA, have been demonstrated for targeting specific separation requirements and
obtaining better separation performance.
Dual reflux pressure swing adsorption (DR-PSA), as a state-of-the-art process, uses a
lateral feed inlet and both light and heavy reflux strategy while keeping the basic features
of conventional PSA cycles, achieving the separation performance beyond the so-called
separation limitations constrained by pressure ratio. However, there are still some typical
problems of DR-PSA to be solved. The main objectives of this study are to develop new
cycles to overcome some key problems of the DR-PSA process. This study is divided into
three main sections, 1) parametric study of a dynamic-feed DR-PSA process for capturing
dilute methane (CH4) from nitrogen (N2) gas; 2) development of a triple-reflux pressure
swing adsorption (TR-PSA) for separating methane-nitrogen-helium ternary gas
mixtures; 3) development of the dual reflux vacuum swing adsorption (DR-VSA) process
for enrichment of low-grade CH4 which falls into the explosion range and demonstration
of a new dual-objective optimization method.
In the first section, it can be concluded that the dynamic-feed strategy can practically
solve the mixing problem caused by the lateral feed inlet of the DR-PSA process and the
performance of dynamic-feed DR-PSA elevates with the number of feed inlet positions
along the column. Parametric studies are conducted based on three key operating
parameters, heavy to feed flow ratio, light reflux flowrate and feed/purge step time,
indicating that the dynamic-feed DR-PSA can always obtain both higher purity and
II
recovery over the traditional DR-PSA process under same operating conditions; a typical
comparison of separation results between two cycles is 53.5% vs. 47.5% for purity and
81.1% vs. 72.2% for recovery, respectively.
In the second section, a triple-reflux (TR)-PSA process is demonstrated to separate a
ternary gas mixture composing of 10% helium (He), 20% CH4 and 70% N2 via a single-
stage process. The TR-PSA process can enrich 10% He up to 45.3 % with a recovery of
91.3% while achieving 60.0 % purity and 90.4% recovery for CH4 and 95.8 % purity and
68.9% recovery for N2 with a work duty of 49.6 kJ mol-1 (feed). The TR-PSA process can
obtain slightly better product purity and recovery of He (45.3% purity with a recovery of
91.3 %) than the 2-stage DR-PSA process (purity is 44.8% and recovery is 89.7%) while
leading to a mild decline in CH4 purity and recovery (1.4% and 1.7%, respectively)
compared to the two-stage DR-PSA. TR-PSA process only requires a cycle work of 49.6
kJ/mol (feed), which is significantly (30%) lower than the specific work of the 2-stage
DR-PSA process (64.3 kJ/mol (feed)) due to the use of one compressor in TR-PSA versus
two in the 2-stage DR-PSA system.
In the last section, vacuum swing adsorption (VSA) is integrated with the dual reflux
strategy as the DR-VSA cycle which can be operated within a pressure range lower than
the atmospheric pressure and avoid the usage of a compressor. Two types of DR-VSA
cycles (A- or B-type cycle indicates that pressure reversal step is carried out through the
heavy or light end, respectively) have been studied for enriching low-grade CH4 (CH4
molar fraction is between 5–20%) gas using activated carbon (AC) or ionic liquidic
zeolites (ILZ). The optimal configuration is A-type cycle packing with ILZ adsorbents,
which can enrich 20% CH4 to a purity of 80.2% with 95.5% recovery and a specific
energy consumption of 180.8 kJ/(mol CH4 captured). The final optimal results achieve a
good balance between purity and recovery by adopting the new optimization method
III
which uses a dual convergence algorithm to iteratively vary operating conditions and a
multiplicative score function to evaluate the separation performance of each case.
IV
Preface
i. Chapter 1 and 2 contains part of the work prepared for publication as a book chapter,
“Chapter 14: Swing technologies for carbon capture: pressure, thermal, vacuum,
electrical and mixed swing processes”, Adv. in G. Gases Vol. 4 - Emission, Capture
and Conversion, Carbon Capture Technologies, which is currently accepted and in
production stage.
ii. Chapter 3 is a paper “Capture of dilute methane with a novel dynamic‐feed dual‐
reflux pressure swing adsorption process” AIChE Journal, 2022, 68(1), e17390.
• I am responsible primarily for the planning, execution and preparation of the work
for publication, and all co-authors have authorized this publication to be
incorporated into this thesis.
iii. Chapter 4 is a paper “Separation of He/N2/CH4 ternary mixtures by a triple-reflux
pressure swing adsorption process” AIChE Journal, 2022, 68(5), e17569.
iv. Chapter 5 is a paper “Enrichment of low-grade methane by dual reflux vacuum swing
adsorption” Separation and Purification Technology, 2022, 301, 121907.
V
List of publications
1. Guo, Y., Gu, X., Hu, G., Webley, P.A. and Li, G.K., 2022. Enrichment of low-grade
methane by dual reflux vacuum swing adsorption. Separation and Purification
Technology, 301, p.121907.
2. Guo, Y., Jusko, V., Xiao, G., Hanekom, J., Hu, G., Webley, P.A., May, E.F. and Li,
G.K., 2022. Separation of He/N2/CH4 ternary mixtures by a triple‐reflux pressure
swing adsorption process. AIChE Journal, 68(5), e17569.
3. Guo, Y., Hu, G., Chen, K., Guo, J., Webley, P.A. and Li, G.K., 2022. Capture of dilute
methane with a novel dynamic-feed dual-reflux pressure swing adsorption process.
AIChE Journal, 68(1), e17390.
4. Hu, G., Xiao, G., Guo, Y., Manning, M., Chen, L., Yu, L., Li, K.G. and May, E.F.,
2022. Separation of methane and nitrogen using ionic liquidic zeolites by pressure
vacuum swing adsorption. AIChE Journal, e17668.
5. Chen, K., Singh, R., Guo, J., Guo, Y., Zavabeti, A., Gu, Q., Snurr, R.Q., Webley, P.A.
and Li, G.K., 2022. Electrical Regulation of CO2 Adsorption in the Metal–Organic
Framework MIL-53. ACS Applied Materials & Interfaces, 14(11), pp.13904-13913.
6. Guo, J., Zhang, Y., Zavabeti, A., Chen, K., Guo, Y., Hu, G., Fan, X. & Li, G. K. 2022.
Hydrogen production from the air. Nature Communications, 13, p.5046.
VI
Acknowledgements
Time flies. I can still remember the first day when I arrived in Australia and felt extremely
anxious and excited for the upcoming days to pursue my Ph.D. degree. I can also recall
the first day when I met with my principal supervisor A/Prof. Gang Kevin Li at his office
to discuss my possible research projects and luckily, we have finally realized most of
those concepts. The three and half years, mixing with countless challenges, anxieties,
satisfactions and passions would be the most unforgettable and meaningful time in my
life. Herein, I would like to send my best wishes to all people I met in University of
Melbourne and express my heartfelt gratitude to your kind help in my research and life.
I would like to thank my supervisor A/Prof. Gang Kevin Li for providing me the
opportunity to pursue this Ph.D. All my research work was completed based on your
guidance and support, which helped me to develop essential skills for academic study. I
also show my appreciation and respect to my co-supervisor Prof. Paul Webley, who
provided high-level and constructive insights and suggestions about my research work
which were critical to the completion of this dissertation. In addition, I would like to
acknowledge my committee chair, A/Prof. Gabriel da Silva, who provided me with
helpful feedback and suggestions on each annual progress meeting, especially during the
most challenging Covid-19 pandemic period. I would also like to sincerely thank Prof.
Guoping Hu for his selfless contribution, who gave me meticulous guidance at the start-
up of my experimental work, supported me in my simulation work, and helped me a lot
in writing and reviewing my papers.
I would like to thank my friendly colleagues in our Clean Energy Group. To Dr. Penny
Xiao, Dr. Ranjeet Singh, Dr. Ali Zavabeti, Jining Guo, Zhi Yu, Chao Wu, Yongqiang
Wang, Jianing Yang, Jiaming Goh, Linda Ye, Leila Dehdari, Hesam Mousavi. It is really
a pleasure to work with you guys. Your assistance is also crucial to my work. I would
VII
also like to express my appreciation to my friends Juncheng Yu, Xueqing Chen and Guo
Lyu. I must say the most amazing time I have spent in Australia is travelling with you
guys.
I must express my deepest gratitude to my family. I would like to thank my parents for
always supporting me, believing in me, and standing by my side with encouragement to
help me regain my confidence when I fell into self-doubt. My wife, Xuewei Gu, deserves
my sincerest appreciation and love. Thank you for accompanying me across the ocean to
a foreign country to pursue my dream. You are my lighthouse, no matter how thick the
fog is, when I see you, I know the ship has arrived at the safe harbour. Falling in love
with you was the easiest and most correct thing I’ve done in my life. To my love, this
thesis is dedicated to you.
VIII
Contents
Declaration......................................................................................................................... I
Abstract ............................................................................................................................. II
Preface ..............................................................................................................................V
List of publications ......................................................................................................... VI
Acknowledgements ....................................................................................................... VII
Contents .......................................................................................................................... IX
List of Tables ................................................................................................................. XII
List of Figures .............................................................................................................. XIV
Chapter 1: Introduction
1.1 Concept and mechanisms of adsorption ................................................................. 1
1.2 Fundamental PSA cycle .......................................................................................... 4
1.3 Dual-reflux PSA cycle ............................................................................................ 7
1.4 Thesis objectives and outline ................................................................................ 10
Reference .................................................................................................................... 12
Chapter 2: Literature review and DR-PSA model
2.1 Key adsorbent properties ...................................................................................... 20
2.1.1 Equilibrium adsorption capacity and selectivity ............................................ 20
2.1.2 Adsorption kinetics ........................................................................................ 23
2.2 Development of the PSA cycle ............................................................................. 25
2.2.1 Vacuum swing adsorption (VSA) .................................................................. 25
2.2.2 Pressure equalization step .............................................................................. 27
IX
2.2.3 Eliminating the impact of adsorption heat ..................................................... 28
2.2.4 Heavy purge (reflux) step............................................................................... 29
2.2.5 Displacement step .......................................................................................... 30
2.3 Advances in dual-reflux PSA process................................................................... 30
2.4. Simulation methods ............................................................................................. 37
2.4.1 Mathematical model ....................................................................................... 37
2.4.2 The flowsheet of DR-PSA model................................................................... 41
2.4.3 Benchmark results and discussion.................................................................. 44
Reference .................................................................................................................... 52
Chapter 3: Capture of dilute methane with a novel dynamic-feed dual reflux
pressure swing adsorption process
3.1 Chapter perspectives ..............................................................................................60
3.2 Main text ................................................................................................................61
3.3 Supplementary information ...................................................................................72
Chapter 4: Separation of He/N2/CH4 ternary mixtures by a triple-reflux pressure
swing adsorption process
4.1 Chapter perspectives ..............................................................................................82
4.2 Main text ................................................................................................................83
4.3 Supplementary information .................................................................................100
Chapter 5: Enrichment of low-grade methane by dual reflux vacuum swing
adsorption
5.1 Chapter perspectives ............................................................................................116
X
5.2 Main text ..............................................................................................................117
5.3 Supplementary information .................................................................................130
Chapter 6: Conclusions and general discussions
6.1 Conclusions ..........................................................................................................135
6.2 Future perspective ................................................................................................136
XI
List of Tables
Chapter 2
Table 1 Isotherm parameters of CH4 and N2 on Norit RB 3 activated carbon ...............37
Table 2 Adsorption column and operational parameters ................................................40
Table 3 Integrator configuration deployed in the model ................................................41
Table 4 Summarization of simulated and experimental results ......................................50
Chapter 3
Table 1 Operational and bed model parameters used in the simulation .........................65
Table S1 Governing equations used to describe the material, momentum and energy
balance in the adsorption column in this model ..............................................................74
Table S2 The detailed descriptions of the integrator configuration in the model...........75
Table S3 The mesh division of each column section......................................................76
Chapter 4
Table 1 Isotherm parameters of CH4, He and N2 ............................................................85
Table 2 Operational and adsorption bed parameters.......................................................86
Table 3 Product compositions obtained during the parametric study of L/F ratio .........94
Table 4 Product compositions obtained during the parametric study of RL/F ratio .......95
Table S.1 Valve logic for the TR-PSA cycle. P indicates that a PID controller directly
controls the valve ..........................................................................................................104
Table S.2 Volumes of tanks used in the simulation ......................................................105
Table S.3 Governing equations for the adsorption column in this model ....................107
XII
Table S.4 Purity and recovery of CH4, He and N2 obtained by nitrogen side-drawn DR-
PSA process under varying RL/F ratios .........................................................................112
Table S.5 Products obtained with varying RL from the first stage of the 2-stage DR-PSA
cycle. Operating parameters: ZF = 0.5, tFE/PU =30 s, F =1.250 SLPM, H =0.375 SLPM
.......................................................................................................................................113
Table S.6 Helium and nitrogen products obtained under the varying RL from the second
stage of the 2-stage DR-PSA cycle. Operating parameters: ZF = 0.5, tFE/PU =30 s, F=0.875
SLPM, H=0.625 SLPM .................................................................................................114
Chapter 5
Table 1 Langmuir isotherm parameters of AC and ILZ ...............................................119
Table 2 Adsorption bed and operational parameters ....................................................120
Table 3 Optimized operating parameters and separation results ..................................127
Table A.1 Governing equations for the adsorption column in this model ....................132
XIII
List of Figures
Chapter 1
Figure 1 Representation of gas adsorption isotherms and pathways for PSA and TSA (T2 >
T1 and p2 > p1, ΔQ indicates the working capacity) ..........................................................3
Figure 2 Scheme of the first 2-beds PSA unit and the four basic steps ............................5
Figure 3 Schematic diagram of separation limits in a) stripping and b) enriching PSA
configurations ...................................................................................................................8
Figure 4 Schematic diagram of the 2-bed DR-PSA process with four basic steps: 1) feed
(FE), 2) pressurization (PR), 3) purge (PU) and 4) blowdown (BD) ................................9
Chapter 2
Figure 1 The IUPAC classification of adsorption isotherms: red arrows indicate the
adsorption pathway and blue arrows indicate the desorption pathway. Reproduced with
permission from Ref. [2]. Copyright (2019) Royal Society of Chemistry .........................21
Figure 2 Schematic diagram of the 3-step VSA cycle....................................................26
Figure 3 Operating sequences of a 6-column 12-step PVSA cycle. AD: adsorption; PED:
pressure equalization (depressurization); Idle: no operation, totally isolated; EVA:
evacuation; PER: pressure equalization (pressurization) and LPR: light product
pressurization ...................................................................................................................28
Figure 4 Schematic of four basic DR-PSA configurations: DR-PL-A, DR-PH-A, DR-PL-
B and DR-PH-B. The whole DR-PSA cycle includes four steps: feed, pressurization,
purge and blowdown. Reprinted with permission from Ref. [39]. Copyright (2022) John
Wiley and Sons ................................................................................................................31
XIV
Figure 5 Flowsheet of the DR-PL-A model. Valves are coloured black or white to
indicate whether the valves are closed or opened during the feed/purge (FE/PU) step ..42
Figure 6 Comparison of pressure history profiles in simulation and experiment reported
by Saleman et al. [43]: dotted line and solid line indicate the pressure variation during the
first cycle and last cycle in the experiment, respectively while dashed line represents the
predicted pressure curve in a whole cycle at the CSS by simulation. Adapted with
permission from Ref. [43]. Copyright (2015) Elsevier ...................................................45
Figure 7 Comparison of temperature history profiles in simulation and experiment
reported by Saleman et al. [43]: T01 (blue), T06 (red) and T11 (green) indicate
temperature variations at top, middle and bottom positions of the adsorption column,
respectively. The solid lines represent experimental measurements while the dotted lines
represent the simulated results. Adapted with permission from Ref. [43]. Copyright
(2015) Elsevier ................................................................................................................46
Figure 8 Comparison of heavy (blue) and light (green) reflux flowrate profiles in the
experiment reported by Saleman et al. [43] and the simulation under the CSS. Adapted
with permission from Ref. [43]. Copyright (2015) Elsevier ...........................................48
Figure 9 Comparison of flowrate profiles of boundary gas streams: feed (red), heavy
(blue) and light (green) gases in the experiment reported by Saleman et al. [43] and
simulation at the CSS. Adapted with permission from Ref. [43]. Copyright (2015)
Elsevier ............................................................................................................................49
Figure 10 Deviation of the CH4 concentration in heavy (yH, marked as black) and light
(yL, marked as red) product gases between experiments and simulations .......................51
XV
Chapter 3
Figure 1 Schematic diagram of a complete DR-PSA cycle in four basic configurations:
feed to high pressure (PH) or low pressure (PL) adsorption column, and heavy (A) or light
(B) gases for pressure inversion ......................................................................................62
Figure 2 Mixing problem reported by Diagne (a) and McIntyre (b). The red line indicates
the sharp concentration increase, and the blue line exhibits a significant concentration
decrease. (a) CO2 concentration profiles at various values of the feed inlet position,
adapted with permission from (Ref. [12]). Copyright (1995) American Chemical Society.
(b) Ethane concentration profiles in the column at the end of the heavy reflux step,
adapted with permission from (Ref. [16]). Copyright (2010) American Chemical Society
.........................................................................................................................................63
Figure 3 Process flowsheet diagram for the numerical model to simulate the conventional
dual-reflux pressure swing adsorption process and the dynamic-feed dual-reflux pressure
swing adsorption process. The number of adsorption columns (N + 1) is determined by
the number of feed positions (n). PID1, PID controller for the compressor; PID2, PID
controller for the Light product valve; VF, system feed valve; VF1-n, column 1 feed valve
at feed point n; VF2-n, column 2 feed valve at feed point n; VH, system heavy product
valve; VH1, column 1 heavy product valve; VH2, column 2 heavy product valve; VHR1,
column 1 heavy reflux valve; VHR2, column 2 heavy reflux valve; VL, system light product
valve; VLR1, column 1 light reflux valve; VLR2, column 2 light reflux valve ..................64
Figure 4 Deviations of ysim predicted in the numerical simulation from the corresponding
values yexp measured in experiments22.............................................................................66
Figure 5 (a) Axial concentration distributions in the conventional DR-PSA column at the
time point of 0, 20, 40, 60, 80, 100 and 120 s during the feed step (operating parameters
XVI
are specified as RL=2.302, H/F≈0.042, and tFE/PU=120 s). z is the dimensionless axial
position along the column, z=0 is the enriching end, z=1 is the stripping end, feed position
is zF=0.7. (b) Enlarged diagram of the dynamic feed range ............................................66
Figure 6 The purity (a) and recovery (b) of methane by DF-DR-PSA cycles as a function
of n (the number of feed inlet points). The operating parameters are RL=2.302, H/F≈0.042,
and tFE/PU=120s). The dashed line represents the result using the conventional DR-PSA
process under the same operating conditions ..................................................................67
Figure 7 Axial concentration distributions in the DF-DR-PSA column at the time point
of 0, 20, 40, 60, 80, 100 and 120 during the feed step by the uniform time method (a) and
the uniform space method (b). The operating parameters are RL=2.302, H/F≈0.042, and
tFE/PU=120 s ......................................................................................................................67
Figure 8 The effect of H/F ratios on the purity (a) and recovery (b) of methane at constant
tFE/PU=120 s and RL=2.302 SLPM. The arrows indicate the direction in which the H/F
ratio should be varied to bias the separation performance favouring the purity or the
recovery. DF-DR-PSA, dynamic-feed dual-reflux pressure swing adsorption; DR-PSA,
dual-reflux pressure swing adsorption.............................................................................68
Figure 9 The correlation between methane recovery and heavy product purity. DF-DR-
PSA, dynamic-feed dual-reflux pressure swing adsorption; DR-PSA, dual-reflux pressure
swing adsorption ..............................................................................................................68
Figure 10 The purity (a) and recovery (b) of methane by two kinds of cycles as a function
of the RL. The results of the conventional DR-PSA process are picked from the optimum
fixed feed position. DF-DR-PSA, dynamic-feed dual-reflux pressure swing adsorption;
DR-PSA, dual-reflux pressure swing adsorption ............................................................69
XVII
Figure 11 The purity (a) and recovery (b) of methane by two processes as a function of
the tFE/PU at constant RL=2.302 SLPM and H/F=0.042. DF-DR-PSA, dynamic-feed dual-
reflux pressure swing adsorption; DR-PSA, dual-reflux pressure swing adsorption ......69
Figure 12 The productivity of methane by two kinds of cycles as a function of the tFE/PU
at constant RL=2.302 SLPM and H/F=0.042. DF-DR-PSA, dynamic-feed dual-reflux
pressure swing adsorption; DR-PSA, dual-reflux pressure swing adsorption.................70
Figure S1 The independence study of the simulated yH and yL results with the mesh
division of the adsorption column ...................................................................................77
Figure S2 The purity (a) and recovery (b) of methane in the heavy product at different
feed positions with varying feed step durations ..............................................................78
feed positions with varying H/F ratios ............................................................................79
feed positions with varying RL.........................................................................................80
Chapter 4
Figure 1 Schematic diagram of DR-PSA configurations: feed to high- (PH) or low-
pressure (PL) column and pressure reversal with the heavy (i.e., the more adsorbable, A)
or light (i.e., the less adsorbable, B) adsorbate ................................................................84
Figure 2 Schematic diagram of the TR-PSA process. FE: feed. PU: purge. PR:
pressurization. BD: blowdown. Valves in solid represent close .....................................84
Figure 3 Isotherms of pure CH4, He and N2 on Norit RB3 activated carbon at 293 K ..85
Figure 4 Process flow diagram of the TR-PSA model during FE/PU step in the PL-A
configuration. Solid and open valves indicate close or open positions, respectively ......85
XVIII
Figure 5 Comparison of experimental and simulated CH4 molar fraction in light and
heavy products .................................................................................................................87
Figure 6 Light (a) and heavy (b) product compositions using the conventional DR-PSA
cycle with constant light reflux ratio RL/F = 2.5, H/F = 0.2, tPR/BD = 45 s and ZF = 0.5
.........................................................................................................................................87
Figure 7 Concentration profiles of gas phase inside the adsorption column during FE (a)
and PU (b) step for a conventional DR-PSA cycle with following operating parameters:
RL/F = 2.5, H/F = 0.2, tFE/PU = 30 s, tPR/BD = 45 s and ZF = 0.5 .......................................88
Figure 8 The impact of tFE/PU on the purity of CH4, He and N2 in the corresponding
product streams, feed compositions are: a) 3% He, 10% CH4 and 87% N2; b) 5% He, 15%
CH4 and 80% N2; c) 10% He, 20% CH4 and 70% N2. Operating parameters: ZF = 0.5,
ZM,out = ZM,in = 0.9, RL/F = 1 and RM /F = 2 .....................................................................89
Figure 9 Axial composition profiles in the column at the end of FE (a), PR (b), PU(c) and
BD (d) steps. The solid line indicates tFE/PU = 30 s case, and the dashed line indicates tFE/PU
= 90 s case. Other operating parameters: ZF = 0.5, ZM,out = ZM,in = 0.9, RL/F = 1 and RM /F
= 2. Annotation of gas streams: 1) feed flow in; 2) mid reflux flow in; 3) light reflux flow
in; 4) heavy product flow out; 5) heavy reflux flow in; 6) mid product flow out; 7) light
product flow out and red/blue indicates gas streams entering/leaving the column .........90
Figure 10 Temperature profile over a cycle at CSS, feed compositions are 10% He, 20%
CH4 and 70% N2. Operating parameters are as follows: ZF = 0.5, ZM,out = ZM,in = 0.9, tFE/PU
=30 s, RL/F = 1 and RM /F = 2 .........................................................................................91
Figure 11 Purities of CH4, He and N2 in the corresponding product streams under varying
ZF. Operating parameters: ZM,out = 0.9, ZM,in = 0.9, tFE/PU =30 s, H/F =0.2, L/F =0.1, RL/F
= 1 and RM /F = 2 .............................................................................................................91
XIX
ZM,out. Other operating parameters are set as: ZF = 0.5, ZM,in = 0.9, tFE/PU =30 s, H/F =0.2,
L/F =0.1, RL/F = 1 and RM /F = 2 ....................................................................................92
ZM,in. Other operating parameters are set as flows: ZF = 0.5, ZM,out = 0.9, tFE/PU =30 s, H/F
=0.2, L/F =0.1, RL/F = 1 and RM /F = 2 ...........................................................................92
Figure 14 The product purity (a) and recovery (b) of CH4, He and N2 as a function of H/F
ratio. The arrows indicate the direction in which the H/F ratio should be varied to bias the
separation performance favouring the CH4 purity or the CH4 recovery. Operating
parameters: ZF = 0.5, ZM,in = ZM,out = 0.9, tFE/PU =30 s, L/F = 0.1, RL/F = 1 and RM /F = 2
.........................................................................................................................................93
Figure 15 The product purity (a) and recovery (b) of CH4, He and N2 as a function of L/F
ratio. The arrows indicate the direction in which the L/F ratio should be varied to bias the
separation performance favouring the He purity or the He recovery. Operating
parameters: ZF = 0.5, ZM,in = ZM,out = 0.9, tFE/PU =30 s, H/F = 0.2, RL/F = 1 and RM /F = 2
.........................................................................................................................................93
Figure 16 The impact of RL/F on the product puritiy (a) and recovery (b) of CH4, He and
N2. Operating parameters: ZF = 0.5, ZM,in = ZM,out = 0.9, H/F = 0.3, L/F = 0.2 and RM/F =
2 .......................................................................................................................................94
Figure 17 The product purity (a) and recovery (b) of CH4, He and N2 as a function of
RM/F ratio. Operating parameters: ZF = 0.5, ZM,in =ZM,out = 0.9, tFE/PU =30 s, H/F = 0.3,
L/F = 0.2 and RL/F = 0.5..................................................................................................95
XX
Figure 18 Specific energy consumption for per mole captured CH4 and He as a function
of RM/F ratio. Constant operating parameters: ZF = 0.5, ZM,out = 0.9, tFE/PU =30 s, H/F =
0.3, L/F = 0.2 and RL/F = 0.5...........................................................................................95
Figure 19 Schematic diagram of the double-stage DR-PSA process (two PL-A
configurations) .................................................................................................................96
Figure 20 Comparison of optimal separation performance and specific work of TR-PSA
and 2-stage DR-PSA processes. Operating parameters for the TR-PSA cycle: ZF = 0.5,
ZM,out = ZM,in = 0.9, tFE/PU =30 s, H/F = 0.3, L/F = 0.2, RL/F = 0.5 and RM/F = 1.0. Operating
conditions for the 2-stage DR-PSA cycle: feed positions are fixed at exact middle of the
adsorption bed for both stages, light reflux flowrate is 1 SLPM for first stage and 0.4
SLPM for the second stage, tFE/PU =30 s, methane product flowrate is 0.375 SLPM (i.e.,
H/F = 0.3) and helium product flowrate is 0.25 SLPM (i.e., L/F = 0.2) .........................96
Figure S.1 Process flow diagram of the complete TR-PSA cycle. Valves are shaded
black/white to indicate whether they are closed/opened during each step. The arrow
indicates the flow direction of each stream ...................................................................102
Figure S.2 The impact of column nodes number on the purity of CH4, He and N2 in the
corresponding product streams. The feed gas contains 10% helium, 20% methane and 70%
nitrogen. Operating parameters are specified as follows: ZF = 0.5, ZM,out = ZM,in = 0.9,
tFE/PU =30 s, L/F=0.1, H/F=0.2, RL/F = 1 and RM /F = 2 ...............................................108
Chapter 5
Figure 1 Schematic diagram of DR-PSA configurations: feed to high- (PH) or low-
pressure (PL) column and pressure inversion with the heavy (i.e., the more adsorbable,
A) or light (i.e., the less adsorbable, B) component ......................................................118
XXI
Figure 2 Process diagram of the DR-VSA model (during FE/PU step). Black color
indicates the valve is closed; white color indicates the valve is open. Vacuum pump is
used in this study to replace the function of the compressor used in previously reported
DR-PSA processes .........................................................................................................119
Figure 3 Flowchart of the optimization approach, T1 (0.0002) is the threshold value to
determine whether the optimization can end the iteration loop and move to the next
process ...........................................................................................................................121
Figure 4 Impacts of theoretical heavy product purity (yH′) on the separation performance
of A-type configuration using ILZ (full symbols) and AC (hollow symbols) under various
feed CH4 concentrations at tF = 45 s and RL = 1 SLPM: a) product purity, b) CH4 recovery
.......................................................................................................................................121
Figure 5 Impacts of theoretical heavy product purity (yH′) on the separation performance
of B-type configuration using ILZ (full symbols) and AC (hollow symbols) under various
feed CH4 concentrations at tF = 45 s and RL = 1 SLPM: a) product purity, b) CH4 recovery
.......................................................................................................................................122
Figure 6 CH4 recovery versus heavy product purity (yH) caused by varying yH′ from 0.3
to 2.0 under feed CH4 concentration of 5%, tF = 45 s and RL = 1 SLPM for A-type (solid
line) and B-type (dash line) configurations using ILZ (full symbols) and AC (hollow
symbols) ........................................................................................................................122
Figure 7 Effects of feed step time (tF) on the separation performance of A-type
configuration using ILZ (full symbols) and AC (hollow symbols) under various feed CH4
concentrations at yH′=0.8 and RL=1 SLPM: a) heavy product purity, b) CH4 recovery, c)
productivity, and d) specific energy consumption.........................................................123
XXII
Figure 8 Effects of feed step time (tF) on the separation performance of B-type
concentrations at yH′ = 0.8 and RL = 1 SLPM: a) heavy product purity, b) CH4 recovery,
c) productivity, and d) specific energy consumption ....................................................124
Figure 9 The influence of light reflux flowrate (RL) on the performance of A-type
concentrations at tF = 30 s and yH′ = 0.8: a) product purity, b) CH4 recovery, c) specific
energy consumption.......................................................................................................125
Figure 10 The influence of light reflux flowrate (RL) on the performance of B-type
concentrations at tF = 30 s and yH′ = 0.8: a) product purity, b) CH4 recovery, c) specific
energy consumption.......................................................................................................126
Figure 11 Determination of optimized separation results of 5% CH4 feed gas via A-type
cycle using ILZ ..............................................................................................................126
XXIII
Chapter 1
Chapter 1: Introduction
1.1 Concept and mechanisms of adsorption
Separation is an essential branch of chemistry as well as related scientific fields. It is
challenging to complete a separation process as it is an entropy reduction process that
defies the second law of thermodynamics[1]. Currently, gas separation is principally
achieved by four kinds of approaches, 1) cryogenic process, based on physical property;
2) membrane separation, based on both physical property and chemical reaction; 3)
absorption, primarily based on chemical reaction; and 4) adsorption methods, both
physisorption and chemisorption exist.
Adsorption is a spontaneous phenomenon that gas molecules will form interaction with
the solid surface and get “captured” when the solid is exposed to a gas [2-5]. The solid
and gas phases are named adsorbent and adsorbate, respectively. Since there are different
interactions between the adsorbent and different adsorbates, the adsorbent will preferably
capture specific molecules from the mixture gas, which is eventually realizable for gas
separation. The mechanisms of adsorptive separation can be classified into three kinds:
1) steric, which is based on a particular property of zeolite, in which case, only small-
enough and proper-shape molecules can transmit into the pores of the zeolite, whereas
other kinds of molecules are thoroughly rejected by the molecular sieves;2) kinetic, based
on the different molecular diffusion rates in the porous material; 3) equilibrium effect,
which is the most widely applied, based on the equilibrium adsorption capacity between
the adsorbent and gas mixtures. There is a crucial parameter for selecting adsorbent and
assessing the adsorbent performance for gas separation called the selectivity, which can
be calculated by Eq. 1 and Eq. 2 for equilibrium- and kinetic-based separations,
respectively.
1
Chapter 1
Q1 p2
α=
eq ⋅ Eq. 1
Q2 p1
K1 D1
α kin = Eq. 2
K2 D2
Where Q1 and Q2 are the molar loadings of components 1 and 2 under partial pressure p1
and p2, respectively; while K1 and K2 are the Henry’s law constants and D1 and D2 are the
diffusivities of components 1 and 2, respectively.
At a given temperature, the relationship between the adsorbed amount and pressure at the
same temperature is generally described by the adsorption isotherms. Figure 1 indicates
an example of two adsorption isotherms at two different temperatures. A larger adsorption
capacity could be achieved by lower temperature or higher pressure and is beneficial to
the adsorption process. As shown in Figure 1, the adsorption capacity will decrease when
raising the temperature or reducing the pressure. Accordingly, either “temperature-
swing” or “pressure-swing” can be applied to conduct the adsorption-desorption cycle
and be named temperature swing adsorption (TSA) and Pressure swing adsorption (PSA),
respectively.
In the TSA cycles, the adsorbent is primarily regenerated by heating, and the total
temperature "swings" between a low temperature (T1) during the adsorption step and a
high temperature (T2) during the desorption step. The difference in adsorption amount at
two temperatures Δq is the actual working capacity of this adsorbent under the given
operating conditions. The TSA process is usually time-consuming as each heating-
cooling cycle usually requires a few hours to over a day. Some novel concepts have been
reported to elevate heating efficiency by using electricity, magnetic or microwave to
rapidly increase the temperature, which is usually referred to as electricity swing
2
Chapter 1
adsorption (ESA) [6-8], magnetic inducted heating adsorption (MISA) [9] and microwave
swing adsorption (MSA) [10], respectively.
PSA has been a rapidly growing process since its conception. The adsorbents in a PSA
unit get regenerated by decreasing the pressure of the adsorption column, which only
requires a few seconds and minutes. Since there is no significant temperature wave, the
PSA process used to be called a heatless adsorption process. The concept of the PSA
system is proposed in 1932, and the first reference to a PSA cycle can be backtracked to
1942 in Germany, which is published as a patent named as a “sorbogen” process [11].
Further developments were concurrently achieved by Skarstrom[12] for the basic 2-bed
4-step PSA cyclic process in the U.S. and Guerin and Domine for the Vacuum Swing
Adsorption (VSA) process in France [13]. Until the late 1970s, the PSA process started
to be commercially accepted[14]. In recent years, The PSA technology has been
intensively investigated for a large variety of gas separation occasions, such as hydrogen
purification[15-21], air separation[22-27], noble gases purification[28-31], separation of
hydrogen isotopes[32-34], CO2 capture[35-40] and CH4 upgrading[41-50]. The PSA
process is considered to be low energy consumption and high automation when compared
to other separation technologies[51, 52].
3
Chapter 1
Figure 1 Representation of gas adsorption isotherms and pathways for PSA and TSA (T2
> T1 and p2 > p1, ΔQ indicates the working capacity).
The performance of a PSA system is primarily influenced by two aspects: 1) which
adsorbent is selected to be packed; 2) the cycle design and operation of the PSA unit.
Therefore, studies of the PSA process could be classified into three key areas: the
discovery and synthesis of new kinds of adsorbents (materials), rapid and energy-efficient
methods for adsorbents regeneration (engineering) and optimizing the design and process
of PSA system (engineering)[53].
1.2 Fundamental PSA cycle
The fundamental characteristic of the PSA cycle is the cyclic alternation of pressure
between adsorption and desorption steps. The first patent application of a continuous and
cyclic PSA system is presented by Skarstrom and is designed to enrich oxygen from
air[12]. The two-bed 4-step PSA scheme described in Skarstrom’s patent and its valve
sequencing in a complete cycle are shown in Figure 2. Each of the two adsorption
columns undergoes four steps in sequence, where the opening and closing of valves and
the gas flow direction would be altered to execute the entire PSA process cyclically. This
system is usually unsteady during initial operation and eventually reaches the cyclic
steady state (CSS) after several cycles of operation.
The four fundamental steps of a Skarstrom cycle are illustrated in Figure 2: 1) feed
process (adsorption); 2) blowdown process (desorption or depressurization); 3) purge
process (reflux) and 4) pressurization process. This type of PSA cycle, also referred to as
the stripping PSA process, specializes in stripping heavy components (preferred adsorbed
adsorbate, generally marked as adsorbate A) from the gas mixture to obtain high-purity
light (component, which is relatively inert adsorbate, generally referred to as adsorbate
B) product gas. This process aims to produce oxygen from air using zeolite 5A as the
4
Chapter 1
adsorbent, which prefers to capture nitrogen rather than oxygen. During the feed step, the
air is compressed to feed into Column 1 (C1) at high pressure, and oxygen enrichment
gas could be obtained as the product stream from the top side. The compressed air is
diverted into column 2 (C2) when the nitrogen front arrives at the top end of C1,
completing the pressurization step in C2. Meantime, the countercurrent (gas flows in the
opposite direction of the feed stream) blowdown process is carried out in C1 by venting
to the atmosphere for the regeneration of adsorbents. Column 1 is primarily dominated
by the desorbed nitrogen at the end of blowdown step, which will significantly limit the
product purity in the following cycle without purging.
Figure 2 Scheme of the first 2-beds PSA unit and the four basic steps.
The purge step (also known as the reflux step) is introduced to replace the mentioned
nitrogen residual in C1 to further sweep the column. The purge gas is provided by using
partial of the pure oxygen gas produced by C2, driven by the pressure difference between
two columns. The purge step could be characterized as the reflux step as in distillation
process, a part of the pure product with high pressure (produced in step 1) is "refluxed"
countercurrently to the column which at low pressure. The ratio of the purge (reflux)
5
Chapter 1
stream flowrate to the light product stream flowrate is the reflux ratio. Generally, the
higher the reflux ratio is, the purer light product could be obtained (the adsorbents get
regenerated more thoroughly). However, as the purge gas is a part of the light product,
increasing the reflux ratio will consequently lower the productivity and raise the cost of
the system. The reflux ratio has a theoretical minimum value below which high-purity
oxygen may be impossible to obtain, which refers to distillation, is that when the volume
of reflux gas is equivalent to feed gas. Since there is the difference of pressure between
two columns, the molar reflux ratio could be expressed as an equation as RL=PL/PH. The
recommend volumetric purge to feed ratio is from 1.1 to 1.5 [54]. A significant difference
between the distillation process and PSA process is that in a PSA system, the purge is
carried out after the expansion of the gas phase, which allows the PSA process to use a
sizeable volumetric reflux ratio at the expense of little light gas product. It is unable to
realize in the distillation process due to that expansion is not possible for liquid reflux,
thereby much larger molar reflux ratio is required. When the adsorbent is almost entirely
desorbed and is prepared to capture nitrogen, the valve V1 should be opened, and V2
should be closed to increase the pressure. After all these four steps were finished, a full
cycle was completed.
The chief contribution of Skarstrom cycle is allowing a marvellous advance in the
application of adsorbents: the cycle duration of PSA is much shorter than that of TSA,
thereby, can realize a higher usage rate of adsorbents per unit time. Another outstanding
work in Skarstrom’s application is the mentioned concern about heat waves and
accumulation. For instance, a slight temperature increase can be seen during the feed step
in the column due to the adsorption heat (adsorption is a spontaneous exothermic reaction)
[3, 4], reducing the adsorbent capacity to some degree. While in the blowdown steps, the
opposite phenomenon happens (temperature slightly decreases), which is detrimental to
the desorption. The heat effects are usually ignored by researchers when they are
6
Chapter 1
designing the laboratory-scale experiments and considering these rigs as isothermal,
owing to the high heat transfer coefficient between the column and the environment.
However, it is significantly crucial to take this into account when designing a large-scale
PSA system[55] due to the adiabatic property of the real wall used to build the adsorbent
columns. Thus, in recently reported research, the temperature history profile plays a vital
role in performance evaluation that can be found in both simulation and experimental
works[47, 56-60].
1.3 Dual-reflux PSA cycle
The Skarstrom cycle (a typical stripping PSA cycle) is generally used to removing the
heavy component from the binary mixture gas and is suitable to produce pure light
product from gas mixtures. In the stripping PSA process, it is challenging to collect high-
purity heavy product due to the restriction of the so-called enriching limit, as indicated in
Figure 3 a). In contrast, the rectifying PSA[61, 62] (also known as the enriching reflux
PSA[63]) cycle is developed to obtain pure heavy product. Enriching PSA reverses the
pressure ratio of the adsorption step and the purge step of the stripping cycle. Thus, the
mixture gas is fed into the adsorption column at a lower pressure (PL), whereas purge is
accomplished at a higher pressure (PH) aiding by a compressor. Contrary to the stripping
PSA, the enriching PSA is good at enriching the heavy component, but it is challenging
and energy-consuming to elevate the light product purity. The separation limitations of
both the stripping PSA case and enriching PSA case are depicted in Figure 3. For the
stripping PSA process, there is an enriching limitation which restricts the heavy
component enrichment ratio between the heavy product (yH) and the feed gas (yF). This
enrichment ratio is limited to be lower than the pressure ratio (PH/PL). However, the
concentration of heavy component in the light product stream (yL) can theoretically reach
zero. In contrast to the stripping PSA, even dilute heavy component can still be enriched
7
Chapter 1
up to a very high concentration by the enriching PSA. Nevertheless, the stripping ratio
(yF/yL) in the lean gas stream cannot be lower than the pressure ratio owing to the stripping
restriction.
Figure 3 Schematic diagram of separation limits in a) stripping and b) enriching PSA
configurations.
The state-of-the-art dual-reflux PSA (DR-PSA) which combines the stripping and
enriching PSA units using a lateral feed inlet is first reported by Diagne et al. [61] to
experimentally investigate the simultaneous enrichment and removal of carbon dioxide
(CO2) from air-CO2 mixtures. The DR-PSA process employs an intermediate feed
position along the low-pressure adsorption column and uses a dual reflux policy to
overcome the mentioned separation limitations of conventional PSA cycles and can
obtain both enriched CO2 gas and CO2-lean gas at different ends with a low operating
pressure ratio. The conceptual diagram of the DR-PSA unit is presented in Figure 4,
which consists of two adsorption columns and four fundamental steps: feed (FE, step 1),
pressurization (PR, step 2), purge (PU, step 3) and blowdown (BD, step 4). Each identical
column is segmented into two column sections, by analogy to distillation, named as
enriching section and stripping section, respectively. Diagne et al. [61, 64, 65] determined
8
Chapter 1
some key operating parameters which can significantly affect the cycle’s separation
performance, such as feed inlet position, reflux ratio, pressure ratio and feed concentration
and investigated their practical effects by experiments and simulations. The first high-
pressure feed DR-PSA configuration is also reported by Diagne et al. [66], in which work
the feed gas is admitted to the high-pressure adsorption column and the combined effects
of various pairs of operating parameters on final separation results of the DR-PSA cycles
are discussed.
Figure 4 Schematic diagram of the 2-bed DR-PSA process with four basic steps: 1) feed
(FE), 2) pressurization (PR), 3) purge (PU) and 4) blowdown (BD).
Another crucial contribution made by Diagne et al. is pointing out the common mixing
problem [66] of DR-PSA cycles, which is also discussed by McIntyre et al.[67] in an
experimental work for enriching dilute ethane by DR-PSA. The root cause of this mixing
problem is due to the lateral feed inlet, where the gas concentrations in upstream reflux
and feed gas streams are different, resulting in a sharp concentration change as a
“perturbation” at the feed position. Therefore, in analogy to the optimal feed tray of
distillation columns, the optimal feed inlet of DR-PSA should be located at a position
9
Chapter 1
where the inlet and reflux concentrations are similar [67]. In the open literature, several
studies [65, 68-70] have investigated the optimization of feed position to minimize the
impact of mixing problem on overall separation performance. However, the column
concentration is continuously changing during the feed step, leading to variable optimum
feed positions. The mixing problem is unavoidable when adopting a fixed feed position
method.
The first concept using DR-PSA process to separate ternary gas mixtures is reported by
Dong et al. [71] in a simulation study which takes the enrichment of CO2-CH4-N2 on
activated carbon as an example. They demonstrate a Petlyuk PSA process, analogous to
Petlyuk distillation (a three-product integrated distillation process), which consists of four
identical adsorption columns. The Petlyuk PSA cycle is compared with a two-stage
combined DR-PSA process in terms of concentrations in three products but without
detailed parametric study and energy analysis. Dong et al. also present a compacted
double-layer adsorption column using variable orifices to modify reflux ratio at top or
bottom end of inner column, however, practical operational suggestion is not provided.
The implementation of this Petlyuk PSA cycle remains challenging in view of its complex
column design and operational control.
1.4 Thesis objectives and outline
The primary objective of this research is to develop new PSA processes to overcome the
aforementioned limitations and achieve better separation performance as well as specific
separation targets. Following this introduction, we review the published literature in
Chapter 2 to presents key adsorbent properties, milestones in the development of PSA
and DR-PSA processes and describe the initial numerical model for this research. In
Chapter 3, we demonstrate a dynamic-feed DR-PSA cycle for the capture of dilute CH4
from N2 using activated carbon and numerically conduct a series of parametric studies.
10
Chapter 1
The results indicate that the dynamic-feed strategy can mitigate the concentration
perturbation and obtain higher CH4 product purity as well as recovery when compared
with fixed-feed DR-PSA cycle under same operating conditions. Chapter 4 presents a
novel triple-reflux PSA (TR-PSA) process which can achieve one-step separation of
ternary gas mixtures using activated carbon. The performance of this process in terms of
product purity, recovery and its energy consumption is compared with a two-stage DR-
PSA process. The separation performance of the TR-PSA cycle is competitive with the
two-stage DR-PSA, and the power and capital costs of TR-PSA are significant lower.
The enrichment of low-grade CH4 from CH4 and N2 mixtures is numerically investigated
in Chapter 5 using two different adsorbents: activated carbon and ionic liquidic zeolites.
This research is the first application of dual-reflux policy at sub-atmospheric pressure
conditions to avoid the use of compressor. Chapter 5 also presents a new optimization
method integrating a double convergence iteration algorithm with a multiplicative
evaluation score which can effectively determine the optimal operating parameters to
obtain both high CH4 purity and recovery. Chapter 6 summarizes key findings of this PhD
thesis together with recommendations for future work.
11
Chapter 1
Reference
1. R. T. Yang, Gas separation by adsorption processes. Vol. 1. 1997: World Scientific.
2. I. Langmuir, The Adsorption of Gases on Plane Surfaces of Glass, Mica and
Platinum. Journal of the American Chemical Society, 1918. 40(9): p. 1361-1403.
3. D. M. Ruthven, Principles of adsorption and adsorption processes. 1984: John
Wiley & Sons.
4. R. T. Yang, Adsorbents: fundamentals and applications. 2003: John Wiley & Sons.
5. P. C. Wankat, Separation process engineering. 2006: Pearson Education.
6. C. Grande and A. Rodrigues, Electric Swing Adsorption for CO2 removal from
flue gases. International Journal of Greenhouse Gas Control, 2007. 2(2): p. 194-
202.
7. M. J. Regufe, A. F. P. Ferreira, J. M. Loureiro, A. Rodrigues, and A. M. Ribeiro,
Development of Hybrid Materials with Activated Carbon and Zeolite 13X for CO2
Capture from Flue Gases by Electric Swing Adsorption. Industrial & Engineering
Chemistry Research, 2020. 59(26): p. 12197–12211.
8. Q. Zhao, F. Wu, Y. Men, X. Fang, J. Zhao, P. Xiao, P. A. Webley, and C. A.
Grande, CO2 capture using a novel hybrid monolith (H-ZSM5/activated carbon)
as adsorbent by combined vacuum and electric swing adsorption (VESA).
Chemical Engineering Journal, 2019. 358: p. 707-717.
9. M. M. Sadiq, H. Li, A. J. Hill, P. Falcaro, M. R. Hill, and K. Suzuki, Magnetic
Induction Swing Adsorption: An Energy Efficient Route to Porous Adsorbent
Regeneration. Chemistry of Materials, 2016. 28(17): p. 6219-6226.
10. Y. Gomez-Rueda, B. Verougstraete, C. Ranga, E. Perez-Botella, F. Reniers, and
J. F. M. Denayer, Rapid temperature swing adsorption using microwave
regeneration for carbon capture. Chemical Engineering Journal, 2022. 446: p.
137345-137357.
12
Chapter 1
11. H. Kahle, Luftreinigung durch Adsorption nach dem Sorbogen I-Verfahren. 1954.
26(2): p. 75-81.
12. C. W. Skarstrom, Method and apparatus for fractionating gaseous mixtures by
adsorption, 1960, U.S. Patent No. 2,944,627.
13. P. Guerin De Montgareuil and D. Domine, Process for separating a binary gaseous
mixture by adsorption, 1964, U.S. Patent No. 3,155,468.
14. D. M. Ruthven and F. Pressure, Pressure Swing Adsorption. Vol. 1. 1994, New
York: VCH Publishers. 235.
15. S. Sircar and T. C. Golden, Purification of Hydrogen by Pressure Swing
Adsorption. Separation Science and Technology, 2000. 35(5): p. 667-687.
16. S.-I. Yang, D.-Y. Choi, S.-C. Jang, S.-H. Kim, and D.-K. Choi, Hydrogen
separation by multi-bed pressure swing adsorption of synthesis gas. Adsorption,
2008. 14(4-5): p. 583-590.
17. F. V. S. Lopes, C. A. Grande, and A. E. Rodrigues, Activated carbon for hydrogen
purification by pressure swing adsorption: Multicomponent breakthrough curves
and PSA performance. Chemical Engineering Science, 2011. 66(3): p. 303-317.
18. J. Xiao, L. Fang, P. Bénard, and R. Chahine, Parametric study of pressure swing
adsorption cycle for hydrogen purification using Cu-BTC. International Journal
of Hydrogen Energy, 2018. 43(30): p. 13962-13974.
19. F. Ye, S. Ma, L. Tong, J. Xiao, P. Bénard, and R. Chahine, Artificial neural
network based optimization for hydrogen purification performance of pressure
swing adsorption. International Journal of Hydrogen Energy, 2019. 44(11): p.
5334-5344.
20. S. Ahn, Y.-W. You, D.-G. Lee, K.-H. Kim, M. Oh, and C.-H. Lee, Layered two-
and four-bed PSA processes for H2 recovery from coal gas. Chemical Engineering
Science, 2012. 68(1): p. 413-423.
13
Chapter 1
21. B. Silva, I. Solomon, A. M. Ribeiro, U. H. Lee, Y. K. Hwang, J.-S. Chang, J. M.
Loureiro, and A. E. Rodrigues, H2 purification by pressure swing adsorption using
CuBTC. Separation and Purification Technology, 2013. 118: p. 744-756.
22. M. M. Hassan, D. M. Ruthven, and N. S. Raghavan, Air separation by pressure
swing adsorption on a carbon molecular sieve. Chemical Engineering Science,
1986. 41(5): p. 1333-1343.
23. D. M. Ruthven, N. S. Raghavan, and M. M. Hassan, Adsorption and diffusion of
nitrogen and oxygen in a carbon molecular sieve. Chemical engineering science,
1986. 41(5): p. 1325-1332.
24. J.-G. Jee, J.-S. Lee, and C.-H. Lee, Air separation by a small-scale two-bed
medical O2 pressure swing adsorption. Industrial & Engineering Chemistry
Research, 2001. 40(16): p. 3647-3658.
25. L. Jiang, L. T. Biegler, and V. G. Fox, Simulation and optimization of pressure‐
swing adsorption systems for air separation. AIChE Journal, 2003. 49(5): p. 1140-
1157.
26. J. Santos, A. Portugal, F. Magalhaes, and A. Mendes, Optimization of medical
PSA units for oxygen production. Industrial & Engineering Chemistry Research,
2006. 45(3): p. 1085-1096.
27. M. Xu, H.-C. Wu, Y. S. Lin, and S. Deng, Simulation and optimization of pressure
swing adsorption process for high-temperature air separation by perovskite
sorbents. Chemical Engineering Journal, 2018. 354: p. 62-74.
28. S. U. Rege and R. T. Yang, Kinetic separation of oxygen and argon using
molecular sieve carbon. Adsorption, 2000. 6(1): p. 15-22.
29. X. Jin, A. Malek, and S. Farooq, Production of Argon from an Oxygen−Argon
Mixture by Pressure Swing Adsorption. Industrial & Engineering Chemistry
Research, 2006. 45(16): p. 5775-5787.
14
Chapter 1
30. N. K. Das, H. Chaudhuri, R. K. Bhandari, D. Ghose, P. Sen, and B. Sinha,
Purification of helium from natural gas by pressure swing adsorption. Current
Science, 2008. 95(12): p. 1684-1687.
31. R. Weh, G. Xiao, E. Sadeghi Pouya, and E. F. May, Helium recovery and
purification by dual reflux pressure swing adsorption. Separation and Purification
Technology, 2022. 288.
32. K. Weaver and C. E. Hamrin Jr, Separation of hydrogen isotopes by heatless
adsorption. Chemical Engineering Science, 1974. 29(9): p. 1873-1882.
33. Y. Si, X. He, J. Jiang, Z. Duan, W. Wang, and D. Yuan, Highly effective H2/D2
separation in a stable Cu-based metal-organic framework. Nano Research, 2019.
34. K. Kotoh, M. Tanaka, Y. Nakamura, T. Sakamoto, Y. Asakura, T. Uda, and T.
Sugiyama, Experimental Verification of Hydrogen Isotope Separation by Pressure
Swing Adsorption. Fusion Science and Technology, 2008. 54(2): p. 411-414.
35. K. T. Chue, J. N. Kim, Y. J. Yoo, S. H. Cho, and R. T. Yang, Comparison of
Activated Carbon and Zeolite 13X for CO2 Recovery from Flue-Gas by Pressure
Swing Adsorption. Industrial & Engineering Chemistry Research, 1995. 34(2): p.
591-598.
36. A. L. Chaffee, G. P. Knowles, Z. Liang, J. Zhang, P. Xiao, and P. A. Webley, CO2
capture by adsorption: materials and process development. International journal
of greenhouse gas control, 2007. 1(1): p. 11-18.
37. R. Haghpanah, R. Nilam, A. Rajendran, S. Farooq, and I. A. Karimi, Cycle
Synthesis and Optimization of a VSA Process for Postcombustion CO2 Capture.
AICHE Journal, 2013. 59(12): p. 4735-4748.
38. J. Ling, A. Ntiamoah, P. Xiao, P. A. Webley, and Y. Zhai, Effects of feed gas
concentration, temperature and process parameters on vacuum swing adsorption
15
Chapter 1
performance for CO2 capture. Chemical Engineering Journal, 2015. 265: p. 47-
57.
39. L. Riboldi and O. Bolland, Evaluating Pressure Swing Adsorption as a CO2
separation technique in coal-fired power plants. International Journal of
Greenhouse Gas Control, 2015. 39: p. 1-16.
40. M. Xu, S. Chen, D.-K. Seo, and S. Deng, Evaluation and optimization of VPSA
processes with nanostructured zeolite NaX for post-combustion CO2 capture.
Chemical Engineering Journal, 2019. 371: p. 693-705.
41. S. Cavenati, C. A. Grande, and A. E. Rodrigues, Separation of CH4/CO2/N2
mixtures by layered pressure swing adsorption for upgrade of natural gas.
Chemical Engineering Science, 2006. 61(12): p. 3893-3906.
42. C. A. Grande and R. Blom, Utilization of dual-PSA technology for natural gas
upgrading and integrated CO2 capture. Energy Procedia, 2012. 26: p. 2-14.
43. M. Kacem, M. Pellerano, and A. Delebarre, Pressure swing adsorption for CO2/N2
and CO2/CH4 separation: Comparison between activated carbons and zeolites
performances. Fuel Processing Technology, 2015. 138: p. 271-283.
44. Y. J. Kim, Y. S. Nam, and Y. T. Kang, Study on a numerical model and PSA
(pressure swing adsorption) process experiment for CH4/CO2 separation from
biogas. Energy, 2015. 91: p. 732-741.
45. A. Abdullah, I. Idris, I. K. Shamsudin, and M. R. Othman, Methane enrichment
from high carbon dioxide content natural gas by pressure swing adsorption.
Journal of Natural Gas Science and Engineering, 2019. 69.
46. R. L. Canevesi, K. A. Andreassen, E. A. Silva, C. E. Borba, and C. A. Grande,
Evaluation of simplified pressure swing adsorption cycles for bio-methane
production. Adsorption, 2019. 25(4): p. 783-793.
16
Chapter 1
47. L. Tao, P. Xiao, A. Qader, and P. A. Webley, CO2 capture from high concentration
CO2 natural gas by pressure swing adsorption at the CO2CRC Otway site,
Australia. International Journal of Greenhouse Gas Control, 2019. 83: p. 1-10.
48. Y.-F. Chen, P.-W. Lin, W.-H. Chen, F.-Y. Yen, H.-S. Yang, and C.-T. Chou,
Biogas Upgrading by Pressure Swing Adsorption with Design of Experiments.
Processes, 2021. 9(8): p. 1325.
49. G. Hu, Q. Zhao, L. Tao, P. Xiao, P. A. Webley, and K. G. Li, Enrichment of low
grade CH4 from N2/CH4 mixtures using vacuum swing adsorption with activated
carbon. Chemical Engineering Science, 2021. 229.
50. G. Hu, G. Xiao, Y. Guo, M. Manning, L. Chen, L. Yu, K. G. Li, and E. F. May,
Separation of methane and nitrogen using ionic liquidic zeolites by pressure
vacuum swing adsorption. AIChE Journal, 2022. 68(7).
51. A. Mersmann, B. Fill, R. Hartmann, and S. Maurer, The Potential of Energy
Saving by Gas-Phase Adsorption Processes. Chemical Engineering & Technology,
2000. 23(11): p. 937-944.
52. C. Voss, Applications of pressure swing adsorption technology. Adsorption 2005.
11(1): p. 527-529.
53. C. A. Grande, Advances in Pressure Swing Adsorption for Gas Separation. ISRN
Chemical Engineering, 2012. 2012: p. 1-13.
54. D. Tondeur and P. C. Wankat, Gas Purification by Pressure Swing Adsorption.
Separation and Purification Methods, 1985. 14(2): p. 157-212.
55. R. Yang and P. Cen, Improved pressure swing adsorption processes for gas
separation: by heat exchange between adsorbers and by using high-heat-capacity
inert additives. Industrial & Engineering Chemistry Process Design and
Development, 1986. 25(1): p. 54-59.
17
Chapter 1
56. Z. Liu, C. A. Grande, P. Li, J. Yu, and A. E. Rodrigues, Multi-bed Vacuum
Pressure Swing Adsorption for carbon dioxide capture from flue gas. Separation
and Purification Technology, 2011. 81(3): p. 307-317.
57. C. Tian, Q. Fu, Z. Ding, Z. Han, and D. Zhang, Experiment and simulation study
of a dual-reflux pressure swing adsorption process for separating N2/O2.
Separation & Purification Technology, 2017. 189: p. 54-65.
58. R. R. Vemula and S. Sircar, Comparative performance of an adiabatic and a
nonadiabatic PSA process for bulk gas separation—a numerical simulation.
AIChE Journal, 2017. 63(9): p. 4066-4078.
59. A. Abdeljaoued, F. Relvas, A. Mendes, and M. H. Chahbani, Simulation and
experimental results of a PSA process for production of hydrogen used in fuel
cells. Journal of Environmental Chemical Engineering, 2018. 6(1): p. 338-355.
60. P. Brea, J. Delgado, V. I. Águeda, and M. A. Uguina, Comparison between MOF
UTSA-16 and BPL activated carbon in hydrogen purification by PSA. Chemical
Engineering Journal, 2019. 355: p. 279-289.
61. D. Diagne, M. Goto, and T. Hirose, New Psa Process with Intermediate Feed Inlet
Position Operated with Dual Refluxes - Application to Carbon-Dioxide Removal
and Enrichment. Journal of Chemical Engineering of Japan, 1994. 27(1): p. 85-
89.
62. A. D. Ebner and J. A. Ritter, Equilibrium theory analysis of rectifying PSA for
heavy component production. AIChE Journal, 2002. 48(8): p. 1679-1691.
63. M. Yoshida, J. A. Ritter, A. Kodama, M. Goto, and T. Hirose, Enriching reflux
and parallel equalization PSA process for concentrating trace components in air.
Industrial & Engineering Chemistry Research, 2003. 42(8): p. 1795-1803.
18
Chapter 1
64. D. Diagne, M. Goto, and T. Hirose, Experimental-Study of Simultaneous
Removal and Concentration of CO2 by an Improved Pressure Swing Adsorption
Process. Energy Conversion and Management, 1995. 36(6-9): p. 431-434.
65. D. Diagne, M. Goto, and T. Hirose, Numerical analysis of a dual refluxed PSA
process during simultaneous removal and concentration of carbon dioxide dilute
gas from air. Journal of Chemical Technology and Biotechnology, 1996. 65(1): p.
29-38.
66. D. Diagne, M. Goto, and T. Hirose, Parametric Studies on CO2 Separation and
Recovery by a Dual Reflux Psa Process Consisting of Both Rectifying and
Stripping Sections. Industrial & Engineering Chemistry Research, 1995. 34(9): p.
3083-3089.
67. J. A. McIntyre, A. D. Ebner, and J. A. Ritter, Experimental Study of a Dual Reflux
Enriching Pressure Swing Adsorption Process for Concentrating Dilute Feed
Streams. Industrial & Engineering Chemistry Research, 2010. 49(4): p. 1848-
1858.
68. W. Lu, G. Xiao, R. Weh, and E. F. May, Rejecting N2 from natural gas by dual
reflux pressure swing adsorption with activated carbon. Journal of Natural Gas
Science and Engineering, 2020. 81.
69. T. S. Bhatt, G. Storti, and R. Rota, Optimal design of dual-reflux pressure swing
adsorption units via equilibrium theory. Chemical Engineering Science, 2013. 102:
p. 42-55.
70. T. S. Bhatt, G. Storti, and R. Rota, Detailed simulation of dual-reflux pressure
swing adsorption process. Chemical Engineering Science, 2015. 122: p. 34-52.
71. F. Dong, H. Lou, M. Goto, and T. Hirose, A new PSA process as an extension of
the Petlyuk distillation concept. Separation and Purification Technology, 1999.
15(1): p. 31-40.
19
Chapter 2
Chapter 2: Literature review and DR-PSA model
In this chapter, pivotal properties of adsorbents will be discussed to gain a picture of
adsorbent screening and how to construct an accurate model to describe the adsorption
process. After the adsorbent properties are detailed, key milestones in PSA cycle design
and DR-PSA process development in recent decades will be listed and discussed. Finally,
the fundamental DR-PSA model used in this PhD research will be demonstrated and
compared with the reported experimental results.
2.1 Key adsorbent properties
The adsorption can be divided into two types: physisorption and chemisorption, based on
different interactions between the adsorbent and the adsorbate. Physisorption is caused
by dispersion force, dipole interaction force, electrostatic attraction, or ammonia bond,
which all belong to the van der Waals force, so physisorption is also called van der Waals
adsorption. Chemisorption is an adsorption phenomenon which accompanies the charges
movement interaction or the formation of chemical bonds. In physisorption, there is no
overlap between electron orbits of the adsorbed molecule and the adsorbent surface;
however, in chemisorption, the electron orbits of the adsorbate molecule and the
adsorbent molecule would overlap or exchange electrons. Therefore, physisorption is
often regarded as weak adsorption and adsorbent can be easily regenerated. In contrast,
chemisorption is strong adsorption which commonly requires high temperature to break
the bond. The adsorption capacity, selectivity and adsorption kinetics are three main
factors to assess an adsorbent.
2.1.1 Equilibrium adsorption capacity and selectivity
Adsorption capacity is an important property which describes the maximum quantity of
adsorbate that the adsorbent can uptake at given temperature and pressure. For any given
condition, the gas-phase and adsorbed-phase molecules will finally reach the adsorption
20
Chapter 2
equilibrium, i.e., the adsorption rate is equal to the desorption rate. The adsorption
isotherms can describe this equilibrium relationship of the adsorbate between the gas- and
adsorbed-phases. S. Brunauer, L. S. Deming, W. E. Deming and E. Teller classify various
adsorption isotherms into five types, named as BDDT classification [1], together with the
VI-type isotherm introduced later, forms the IUPAC (International Union of Pure and
Applied Chemistry) adsorption isotherm, as shown in Fig. 1. The I-type isotherm is an
upward convex Langmuir curve, indicates that the adsorbent will still have a relatively
high adsorption capacity even when the partial pressure of the adsorbate is very low,
generally used to describe the single-layer adsorption or microporous adsorption. The
types II, III and VI isotherms depict the weak adsorption between the adsorbate and
nonporous and macroporous adsorbents. The VI-type isotherm is a multi-step curve,
indicating a representative multi-layer adsorption on a highly uniform surface. The
adsorption loop in IV-type and V-type isotherms corresponds to mesoporous
condensation of adsorbate at moderate and high pressure.
Fig. 1 The IUPAC classification of adsorption isotherms: red arrows indicate the
adsorption pathway and blue arrows indicate the desorption pathway. Reproduced with
permission from Ref. [2]. Copyright (2019) Royal Society of Chemistry.
21
Chapter 2
Langmuir isotherm is the first non-linear isotherm which can accurately plot the
relationship between the adsorbed amount Q and pressure p, described as Eq. 1[3].
b⋅ p
Q Qmax ⋅
= Eq.1
1+ b ⋅ p
Where Qmax is the saturated adsorption capacity and b is the Langmuir equilibrium
parameter. The b can be expressed as Eq. 2.
 −∆H 
b= b0 ⋅ exp   Eq. 2
 ℜ ⋅T 
b0 is the Langmuir equilibrium constant, -ΔH is the adsorption heat (indicates that
adsorption is an exothermal process) and is a temperature-dependent parameter, ℜ is the
universal gas constant, and T is the temperature of the gas.
The values of three parameters: Qmax, b0 and -ΔH could be determined by fitting to the
adsorbed amount over a range of temperature and pressure. It should be noted that, by
this approach, the adsorption heat would be treated as a constant, ignoring the correlation
between the -ΔH with temperature and adsorbent coverage. According to work done by
Stadie et al. and Czerny et al. [4, 5], the adsorption heat would gain a slight alteration
when the temperature was changed over a wide range. Since the temperature swings
slightly for most PSA process, the -ΔH can be considered as a constant in PSA process
simulation. Overall, the Langmuir isotherm equation has been widely used in numerical
models of various PSA cycles and has proven to be reliable.
The gas mixtures contain multiple components in the PSA process, making the
competitive adsorption is non-negligible. To describe the competitive adsorption between
different components, Eq. 1 is extended from a pure gas adsorption isotherm to a multi-
component adsorption isotherm by adding terms to the denominator, i.e., the extended
Langmuir isotherm, as shown in Eq.3.
22
Chapter 2
Qmax,i ⋅ pi ⋅ bi
Qi = Eq .3
1 + ∑ j =0 p j ⋅ b j
n
Where the subscripts i and j refers to different components in gas phase and n is the
number of components in the mixture gas. The extended Langmuir isotherm equation
directly uses regression isotherm parameters calculated from the pure gas adsorption
isotherm of each component.
The selectivity denotes the efficiency of selecting one gas component from the others.
The equilibrium-selectivity can be described by combining its definition described in
Chapter 1 and the extended Langmuir isotherm equation, as shown in Eq .4.
Qi ⋅ p j Qmax,i ⋅ bi
α ij
= = Eq .4
Q j ⋅ pi Qmax, j ⋅ b j
2.1.2 Adsorption kinetics
The adsorption kinetics is one of the key properties of the adsorbent for designing an
adsorption cycle and is generally described by an adsorption rate model which could be
established according to the mass transfer mechanism and rule of gas adsorbed in the
solid adsorbent. Three main models have been proposed to depict the adsorption kinetics.
1) Pore diffusion model
The pore diffusion model assumes that the adsorption rate is mainly limited by the
diffusion rate of adsorbates through the pores of the adsorbent. The distribution of the
adsorbate mass on the adsorbent particles and the adsorption rate can be calculated from
the diffusion equation. However, the solving of pre diffusion equation requires intensive
computational efforts as it is a second-order partial differential equation. The complete
pore diffusion equation in spherical coordinates is shown as Eq. 5:
23
Chapter 2
 ∂ 2C p 2 ∂C p  ∂Q ∂C p
ε p D p  2 +  =ρ +εp Eq. 5
 ∂r r ∂r  ∂t ∂t
Initial conditions: Q ( 0, r ) = 0, CP ( 0, r ) = 0 ;
∂CP ∂C p
Boundary conditions: = 0, ε p D p = kg (C − C p ) .
∂r r =0 ∂r r = rp
Where the C is the adsorbate concentration in the gas phase, Cp is the concentration of
adsorbate in the pore of the adsorbent, εp is the particle porosity, Dp is the pore diffusion
coefficient, kg is the mass-transfer coefficient from the bulk to the particle surface, r is
the spherical coordinate, rp is the radius of adsorbent particles and t is the time. The
relationship between Q and CP can be connected by adsorption isotherms.
2) Solid-phase diffusion model
The diffusion of adsorbate in adsorbent particles can be described by solid-phase
diffusion partial differential equation, the equation in spherical coordinates is shown as
Eq. 6:
∂Q Ds ∂  2 ∂Q 
= r  Eq. 6
∂t r 2 ∂r  ∂r 
Initial conditions: Q ( 0, r ) = 0 ;
∂Q
( )
Boundary conditions: Q t , rp = Q∗ ,
∂r r =0
=0.
Here Ds is the solid-phase diffusion coefficient, Q* is the concentration of the adsorbed
phase in equilibrium with the concentration of the fluid phase.
3) Linear driving force (LDF) model
24
Chapter 2
Some approximate expressions of the adsorption rate model have been proposed to
simplify the adsorption kinetics model, among which the LDF model, proposed by
Glueckauf [6], is most frequently used in PSA simulations. This model assumes that the
adsorption rate is proportional to the concentration difference, as shown in Eq. 7.
dQ
= k Q −Q
dt
( ) Eq. 7
Where k is the lumped mass transfer co-efficient, Q is the equilibrium adsorbent loading
amount, Q is the adsorbent loading amount. The LDF approximation has been widely
accepted and used in the research and design of various adsorption process, mainly owing
to its simplicity and analyzability. Liaw et al., Li and Yang both derive the LDF model
by assuming different concentration distribution models within the particles, and the
results are completely consistent with the model proposed by Glueckauf [7, 8]. These
assumptions provide the mathematical and physical basis for the LDF model.
2.2 Development of the PSA cycle
PSA has seen rapid development since the first patent was demonstrated. Several
innovative steps and concepts have been proposed to further improve the performance of
PSA cycles with engineering methods.
2.2.1 Vacuum swing adsorption (VSA)
The vacuum swing adsorption (VSA) is a concurrent patent with the Skarstrom PSA
cycle, which is demonstrated by Guerin and Domine [9]. In Guerin-Domine’s reported
work, numerous VSA cycles with different bed numbers and interconnection methods
have been demonstrated. Here, the most straightforward 3-step VSA cycle is illustrated
in Figure 9. A single column can accomplish this VSA cycle because there is no
interaction requirement with another column. There are two main differences between the
25
Chapter 2
Guerin-Domine and Skarstrom cycles: 1) operating pressure window; 2) the absence of
purge step. In the 3-step VSA cycle, firstly, the feed gas flows into adsorption column at
the atmospheric pressure without the necessity of compressor and light product gas can
be collected at the top outlet; secondly, the evacuation step is carried out using a vacuum
pump to countercurrently depressurize the adsorption column to a pressure lower than the
atmospheric pressure for adsorbent regeneration and heavy product collection at the
bottom of column; lastly, feed gas is injected to pressurize the column to the atmospheric
pressure.
Light gas
Feed Evacuation Pressurization
Feed gas Heavy gas Feed gas

Pressure
pH pH
pL
Cycle time
Fig. 2 Schematic diagram of the 3-step VSA cycle.
The regeneration of adsorbents in VSA cycles is also achieved by reducing pressure.
Thereby, VSA process is also classified as pressure-swing technology. The demonstration
of VSA cycles diversify the operating pressure windows of PSA processes. High pressure
ratio can be achieved by combining PSA and VSA as the pressure vacuum swing
adsorption (PVSA) to carry out adsorption at a pressure higher than the atmospheric
26
Chapter 2
pressure while operate the desorption at a vacuum pressure. The pressure drop along
adsorption column is one of the main causes which limits the application of PSA cycles
in industrial scale. Numerous studies have been reported using structured adsorbents,
such as honeycomb monolith [10, 11], foam structures [12, 13], laminate adsorbents [14],
fabric adsorbents [15] and 3-D printed adsorbents [16], to elevate the pressure swing
efficiency, reduce the pressure drop along the column and enhance the mass transfer
kinetics.
2.2.2 Pressure equalization step
The pressure equalization (PE) step, quickly connecting the high-pressure and low-
pressure columns, is first introduced by ESSO Research group [17]. The primary merits
of introducing the PE step are improving the recovery of the light component and
reducing the power duty of the pump or compressor [18-21]. The main drawback of
introducing the PE step is the negative impact on the process continuity. The two columns
cannot process feed gas during the PE step, requiring at least one more adsorption column
to be deployed to continuously provide gas production. The use of multiple adsorption
columns allows the incorporation of multiple PE steps while keeping a continuous feed.
Numerous PSA processes with multiple columns and complicated step designs have been
reported in the open literature [21-27]. Fig. 3 demonstrates the operating sequences of
one adsorption column during the entire 12-step cycle. In this PVSA system, there are
three pairs of PE steps. The feed gas is processed during steps 1 and 2 while light gas can
be collected cocurrently. The heavy gas can be collected countercurrently during steps 7
and 8. The gas production steps take one-sixth of the total steps, hence, at least six
adsorption columns are required to ensure continuous gas production of the system. Hu
et al. [25] and Dehdari et al. [26] separately investigated the impact of number of PE steps
on separation performance in terms of product purity, recovery and productivity. In the
work reported by Hu et al., the target product gas is heavy component, introducing more
27
Chapter 2
PE steps is proven to be beneficial to the heavy product purity at the expense of recovery
and productivity. Dehdari et al. use the PSA to achieve the purification of light component
and find that increasing the number of PE steps leads to the growth in product recovery
and the reduction in purity and productivity. Overall, results in these two studies show
the same conclusion, i.e., introducing the PE step can enhance the heavy product purity
and light component recovery while decreasing the heavy component recovery and light
product purity as well as the productivity.
AD PED Idle EVA PER LPR
Step 1 2 3 4 5 6 7 8 9 10 11 12
Fig. 3 Operating sequences of a 6-column 12-step PVSA cycle. AD: adsorption; PED:
pressure equalization (depressurization); Idle: no operation, totally isolated; EVA:
evacuation; PER: pressure equalization (pressurization) and LPR: light product
pressurization.
2.2.3 Eliminating the impact of adsorption heat
The adsorption process is an exothermal process and the heat generated during the
adsorption step is accumulated in column, leading to the temperature increase and the
reduction of adsorbent loading capacity. Conversely, the column temperature decreases
during the desorption step, which is unfavourable for the regeneration of adsorbents. A
study reported by Yang and Cen [28] focuses on eliminating or mitigating the temperature
fluctuations in the PSA cycle. Yang and Cen design the 2-bed PSA as a shell-and-tube
28
Chapter 2
type heat exchanger to allow heat transfer between the adsorption and desorption
columns. They also mix non-adsorptive and high specific heat capacity materials with
adsorbents to be packed in the adsorption column as an inert additive. To the best of the
author's knowledge, this work is the only reported work in the open literature which
proposed an approach to reduce the thermal wave in PSA. The thermal accumulation in
industrial-scale PSA systems is more significant than in laboratory-scale units owing to
the low thermal conductivity of the thick column wall. Therefore, the relevant studies are
still meaningful from the perspectives of industrial applications.
2.2.4 Heavy purge (reflux) step
The concept of light purge step has been introduced in Chapter 1 when discussing the
Skarstrom cycle. To replace the light gas in adsorption column, a heavy purge (also
known as rinse step or heavy reflux) step is proposed [29] by Basmadjian and Pogorski.
In contrast to the light purge step, the heavy purge step injects partial of the heavy product
gas from the heavy end of the column co-currently to the feed direction. Sircar reports a
5-bed 5-step VSA process with the heavy purge step to separate CH4 and CO2 gas mixture
using a proprietary zeolite as adsorbents. The CO2-rinse step is set after the adsorption
step to ensure that the adsorption column is primarily saturated with CO2. Then pure CO2
product gas can be collected in the following desorption step. The reported cycle can
simultaneously obtain both high-purity CO2 (~ 99%) and CH4 (~ 99%) products as well
as high recover (~99%) for two components. Reynolds et al. [30] evaluate the separation
performance of a 5-bed 5-step heavy reflux PSA cycle for capturing CO2 from the flue
gas (CO2 concentration is 15%) and can obtain 98.7% product purity and 98.7% CO2
recovery in the best cycle. Hu et al. [25] compare the separation performance of a 4-bed
VSA system with and without the introduction of a heavy purge step and find that
introducing the heavy purge step can significantly increase the heavy product purity by
sacrificing the recovery of heavy component as well as the productivity.
29
Chapter 2
2.2.5 Displacement step
The displacement step uses a more strongly adsorbed component, which is not included
in the feed composition, to displace the adsorbed heavy component. The CO2-
displacement step is usually introduced in separation of CH4 and N2 binary mixtures to
increase the CH4 product purity and recovery [31-34]. Liu et al. [31]reports a 4-step PSA
cycle with CO2-displacement step to separate CH4 and N2 using activated carbon. This
cycle consists of pressurization, adsorption, displacement, and regeneration steps.
Compared with the CO2, CH4 can be regarded as a relatively lighter component.
Following the adsorption step, the CH4-loaded column will be rinsed by CO2 gas and CH4
can be collected cocurrently during the displacement step. The authors also compare two
operation modes of the CO2 displacement step based on operating pressure. The
adsorption pressure is 0.4 MPa in the reported experiment. The first mode is operating
the displacement step at 0.4 MPa while the second one is releasing pressure to the ambient
and then followed with the displacement step at atmospheric pressure. Results show that
the displacement at ambient pressure is more efficient than high-pressure displacement
and the process can enrich 17.62% CH4 up to 90% with a recovery of 98%.
2.3 Advances in dual-reflux PSA process
Since the seminal experimental apparatus demonstrated by Diagne et al. [35-38],
extensive studies of DR-PSA processes have been reported numerically and
experimentally.
Kearns and Webley identify the basic nomenclature of four basic DR-PSA configurations
based on which gas component is used to complete the pressure inversion and which
column the feed gas flows into, as shown in Fig. 4. The PL/PH indicates whether the feed
gas is admitted into the low-pressure or high-pressure column. The A or B show that the
pressure reversal (pressurization and blowdown) steps are accomplished by transferring
30
Chapter 2
the heavy or light gas through the compressor. They also demonstrate the four basic steps
of an entire DR-PSA cycle, consisting of feed/purge and pressurization/blowdown steps.
During the feed/purge step, the feed gas can be admitted to either low- or high-pressure
column at an intermediate position along the adsorption column while part of the light
gas can be diverted to the low-pressure column through a control valve, i.e., the light
reflux flow and partial heavy product extracted from the low-pressure column can be
compressed and injected into the high-pressure column as the heavy reflux flow. During
the pressurization/blowdown step, high-pressure gas can be transferred to the low-
pressure column to complete the pressure reversal either through the compressor at the
light or heavy end of the column.
Light gas
B
RL L, yL
(z=1)
PH PL
Feed Feed
PH PL
F, yF F, yF PL PH
(z=0)
Heavy gas RH
A
H, yH
1.Feed (FE) 2.Pressurization (PR)
Light gas
L, yL R B
L
(z=1)
PL PH
Feed Feed
PL PH
F, yF F, yF PH PL
(z=0)
RH Heavy gas
A
H, yH
3.Purge (PU) 4.Blowdown (BD)
Fig. 4 Schematic of four basic DR-PSA configurations: DR-PL-A, DR-PH-A, DR-PL-B
and DR-PH-B. The whole DR-PSA cycle includes four steps: feed, pressurization, purge
and blowdown. Reprinted with permission from Ref [39]. Copyright (2022) John Wiley
and Sons.
31
Chapter 2
DR-PSA has been reported as experimental studies for separating gas mixtures in
different scenarios, including hydrocarbons enrichment (capture ethane or methane from
nitrogen), air separation (separating oxygen and nitrogen), VOC removal from air and
CO2 capture, etc. Diagne et al. demonstrate the first bench-scale DR-PSA experimental
apparatus to separate CO2 and air mixture at different CO2 concentrations (0–20%) using
the zeolite 13X as adsorbents [35]. This cycle is a PL-type configuration as the feed gas
is injected into the low-pressure column and purge in the high-pressure column. But the
procedure of pressure reversal steps is not detailed in this research. In this research, the
DR-PSA can simultaneously achieve high enrichment and stripping ratio between 6–7,
which are both higher than the pressure ratio (pH/pL) = 2. Diagne et al. then adopt the
high-pressure feed (PH) configuration in their following research with various pressure
ratios (between 1.64–12.50) [37]. They investigate the combined impacts of feed position
and reflux ratio and determine that the optimum dimensionless feed position is between
0.4–0.6 and the optimum reflux ratio is between 0.3–0.6. The feed gas with 20% CO2 is
enriched to a 95% CO2 product gas with 95% fractional recovery.
McIntyre et al. [40] enrich the dilute ethane (C2H6) in N2 with a lab-scale DR-PL-A
system using activated carbon as the adsorbent. The pressure window of this PSA system
is between 27–211 kPa, which can enrich a dilute feed gas (containing 0.75 vol% C2H6)
to a 68.4 vol% product gas with a recovery of 99.6%, while removing the C2H6 content
to only 30 ppm in the light product. In this research, the enrichment ratio and stripping
ratio are ~91.2 and ~75.2, respectively with the pressure ratio of ~7.8. In their later
research, McIntyre et al. [41] point out that the two reflux streams of the internal
circulation are coupled to each other, i.e., any change in flowrate of either light reflux or
heavy reflux will cause the other to change. Another conclusion is about the optimum
feed position which is suggested to be located at a position where feed and column
32
Chapter 2
concentrations are similar, with reference to the optimal feed position in the distillation
process.
Wakasugi et al. [42] assemble a laboratory-scale twin bed DR-PH-A unit to strip the low-
concentration VOC from air stream using activated carbon. The VOC vapor in feed gas
can be enriched to a concentration which is high enough to be condensed while achieving
a stripping ratio higher than 100 with a pressure ratio of 2. They also run a series of
experiments to compare the results of DR-PSA process with other reported PSA processes
and find that DR-PSA delivers the best separation performance.
The separation of CH4 and N2 is inherently challenging due to their similar physical and
chemical properties. Bhatt et al. demonstrate a 2-bed lab-scale DR-PSA unit to separate
a 5% CH4 and 95% N2 gaseous mixture using activated carbon. This DR-PSA system is
deployed in DR-PH-A configuration and can obtain a heavy product with 55% CH4 molar
fraction and a light gas with 1.5% CH4 molar fraction. Saleman et al. [43] report a bench-
scale DR-PL-A apparatus to enrich low-grade CH4 (concentration is from 2.4 mol.% to
49.6 mol.%) in N2 and investigate the impacts of light reflux fowrate and heavy product
flowrate on CH4 purity and recovery. The feed gas containing 2.4 mol% CH4 can be
enriched 21 times, resulting in a product with 51.3 mol.% purity. For case of 49.6% CH4
feed gas, the DR-PSA requires 20 kJ energy to process per mole feed gas and produce a
product of 72.5% purity with 98% CH4 recovery. Saleman et al. also discuss the function
of two reflux streams in this research and indicates a threshold reflux fowrate to complete
the thorough swipe in the low-pressure column by reflux streams and adequate rinse in
the high-pressure column by heavy reflux streams. May et al. [44] experimentally
demonstrate the four basic DR-PSA configurations [45] to separate 89.6% N2 and 10.4%
CH4 mixture using activated carbon and compare their respective separation performance
in terms of CH4 purity, recovery, productivity and power duty. They reveal the correlation
33
Chapter 2
between theoretical bond of CH4 concentration in heavy/light product and the heavy
product to feed flowrate (H/F) ratio and determine that increasing or decreasing the H/F
ratio can shift the separation performance towards obtaining either high recovery or high
purity.
CO2 capture has been of interest due to the serious global warming phenomenon, and the
DR-PSA process has been investigated for removing CO2 from flue gas. Li et al. [46] use
a 2-bed 6-step DR-PH-A system packed with silica gel adsorbent to enrich the CO2 from
15 mol.% in the mimical dry flue gas to 99.18 mol.% in the product gas with the recovery
rate, productivity and specific energy consumption of 99.62%, 0.542 m3kg-1hr-1 and 1.937
MJ/kg CO2 captured, respectively. Wawrzyńczak et al. [47] demonstrate an onsite pilot-
scale DR-PVSA process to capture CO2 from the real flue gas stream. The flue gas is pre-
treated by a processing container to remove dust, SOx and NOx and to conduct the gas
cooling, gas drying and gas compression. This DR-PVSA system consists of four
adsorption beds packed with two kinds of activated carbon and nine steps which can
achieve continuous treatment of flue gas. The pilot test shows separation results: 85.7%
purity, 44.6% CO2 recovery and productivity of 0.274 t/(d∙m3 adsorbent). The specific
energy consumption of this plant is 3.52 GJ/tonne CO2 captured.
Another application of DR-PSA is air separation to produce O2 and has been investigated
by Tian et al. [48]. In this research, the DR-PH-A configuration is demonstrated using
LiLSX zeolite as the adsorbent. At the optimum operating conditions, the air (21% O2
and 79% N2) can be separated to two high-purity products: 99.1% O2 product with 98%
recovery and 99.0% N2 product with a recovery of 99.3%.
Helium recovery from vent gas stream of nitrogen rejection units in liquified natural gas
plants by cascade DR-PSA system using binder free zeolite 13X has been reported by
Weh et al. [49]. In this research, a series of experiments have been operated using four
34
Chapter 2
kinds of DR-PSA configurations to process feed gases with (1–50 mol.%) He, among
which PH-A cycle shows the best separation performance. The dilute feed gas with 0.8
mol.% He can be concentrated to 31.2 mol.% with the recovery of 89.7% which can be
used as the first-stage DR-PSA to produce crude helium. The second-stage DR-PSA is
for helium purification to further enrich the crude helium, for instance, 35.5% He can be
enriched to 99.1%. This cascade DR-PSA system is further numerically studied by a
numerical model and can achieve high-purity He product (99.999 mol.%) at 95%
recovery with energy consumption of 1.5 MJ/(mol He produced).
The simulation studies of DR-PSA range from simplified analytical models to rigorous
numerical models. The analytical models are based on equilibrium theory with lots of
assumptions to reduce the computational efforts of simulation, leading to inaccurate
predictions and very limited application. Ebner et al. [50] build a equilibrium-based
model to simulate a DR-PL-A system. In this study, there are several assumptions: 1)
isothermal model; 2) linear isotherms; 3) local equilibrium; 4) no pressure drop along the
axis 5) complete separation (purity of two products are 100%). The authors find that in
the idealize simulation study, the DR-PSA process can achieve perfect performance over
a wide range of operation conditions, even when the pressure ratio is as low as 1.1.
Kearns and Webley [45, 51] develop the mathematical models to examine the power and
productivity tradeoffs of four DR-PSA configurations. It is found that the concentration
of heavy component in feed gas has the most significant effect on the selection of DR-
PSA configuration. PH-A cycle can deliver best performance when processing dilute
heavy component feed gas while PL-B cycle is recommended to separate feed gas with
high-concentration heavy component. The power-productivity tradeoff curves between
four configurations are quite different when using adsorbents with high selectivity.
35
Chapter 2
Bhatt et al. [52] report the optimal design of a DR-PH-A configuration based on the
equilibrium theory. The authors propose the concept “Triangular Operating Zone (TOZ)”
to determine under which operating parameters the complete separation can be achieved.
Bhatt et al. [53] also develop an isothermal numerical model on Aspen Adsorption
platform to numerically study the performance of their experimental rig. This model uses
a so-called “interaction” module provided by Aspen Adsorption, and three individual
interactions are used to storage and output the information of heavy reflux, light reflux
and pressure reversal, respectively. This model can successfully predict the product
compositions with an average product purity error between -2–8%, which might be cause
by the assumption of isothermal operation.
Zhang et al. [54] develop a non-isothermal numerical model of DR-PSA for methane
upgrading with rigorous energy, momentum and material balance. The accuracy of this
model is validated by comparing the pressure history profiles, temperature history profiles
and stream flowrate profiles as well as the composition in both light and heavy products
with the experimental data reported by Saleman et al. [43]. The material balance error of
this model is high at 8.6%, which is mainly caused by the infinitively high axial dispersion
when gas velocity approaches to zero as explained by Weh et al. [55]. The root mean
square deviations (RMSD) of the CH4 concentrations are only 0.3 and 2.4 mol.% for the
light and heavy products, respectively.
Zou et al. [56] develop an advanced simulation model using two PID controllers to
separately control the inlet volumetric flowrate of the compressor and the molar flowrate
of the light product stream. The standard deviation of the simulated results is 0.3 mol.%
in the C2H6 product gas compared with the experimental results from McIntyre et al. [41]
and 4 mol.% in the CH4 product stream benchmarked against results reported by Saleman
et al. [43].
36
Chapter 2
In recent years, there is a promising kind of interdisciplinary researches between PSA
technology and computer science, like using machine learning [57-59]or artificial neural
network [60, 61]to optimize the PSA system. It is foreseeable that the rapid development
of computer science and process engineering will drive the advances of PSA technology.
2.4. Simulation methods
2.4.1 Mathematical model
We first established an initial DR-PSA model and benchmarked its simulation results
with experimental data reported in the open literature [43]. This DR-PSA model was then
extended to numerically investigate the separation performance of newly developed PSA
cycles. The detailed non-isothermal numerical model was built on the Aspen Adsorption
v10.0 platform. Aspen Adsorption is a commercial software package which can simulate
the operation of dynamic adsorption cycles with rigorous material, momentum and energy
balances. The governing equations and assumptions used to depict the adsorption model
are given as follows:
Equilibrium adsorption isotherm model: using the Extended Langmuir isotherm (Eq. 4)
to describe the competitive adsorption for gas mixtures, expressed as a function of gas-
phase partial pressure. The regression adsorption isotherm parameters are summarized in
Table 1.
Table 1 Isotherm parameters of CH4 and N2 on Norit RB 3 activated carbon
Parameter CH4 N2
Qmax [mol/kg] 5.53 4.17
b0 [bar-1] 1.19 x 10-4 1.75 x 10-4
-ΔH [kJ/mol] 17.27 14.32
The momentum balance and pressure drop are calculated by Ergun equation (Eq. 8).
37
Chapter 2
∂p  150 ×10−5 µ g (1 − ε i ) 2 1.75 ×10−5 M W ρ g (1 − ε i ) 2 
= ± vg + vg  Eq. 8
∂z  (2rpψ ) 2 ε i3 2rpψε i3 
 
Where µg is the viscosity of gas mixture, z is the axial distance coordinate, εi is the bed
voidage, vg is the superficial gas velocity, ψ is the particle shape factor, MW is the
molecular weight of gas mixture and ρg is the gas density.
The adsorption column is regarded as a one-dimensional which neglect the radial
diffusion of temperature, pressure and concentration in the gas-solid phase. The material
balance in the gas phase is described by convection with axial dispersion, as shown in Eq.
9.
∂ (vg ci ) ∂ci ∂Q ∂ 2c
+ [ε i + (1 − ε i )ε p ] + ρb i − DL 2i =0 Eq. 9
∂z ∂t ∂t ∂z
Where, ci is the molar fraction of component i, εp is the particle voidage, ρb is the bulk
density of adsorbent, Qi is the adsorbed loading of component i, t is the time and DL is the
constant dispersion coefficient.
The gas mixtures obey the ideal gas law, as shown in Eq. 10.
pV = nℜT Eq. 10
Where, V is the gas volume and n is the amount of substance.
The adsorption kinetics is described using LDF model with constant mass transfer
coefficients for each component, as shown in Eq. 7.
The energy balance was assumed as non-isothermal. Hence both gas and solid phase
conduction were considered and the heat transfer to the environment was rigorous. The
governing equation of energy balance can be classified into three parts: solid phase, gas
phase and column wall, as described in Eqs. 11–13, respectively. The specifications of
38
Chapter 2
energy assumptions are listed as follows: 1) constant adsorbed-phase heat capacities; 2)
constant adsorption heat of each component; 3) constant column heat transfer coefficient;
4) constant gas-phase thermal conductivity; 5) rigorous heat transfer coefficient to the
environment; 6) estimated gas-wall heat transfer coefficient; 7) constant solid-phase heat
capacity.
Solid phase:
∂ 2Ts ∂T ∂T
−ks 2 + C ps ρb s + ρb ∑ i =1 ( C pai Qi ) s
n
Eq. 11
∂z ∂t ∂t
n  ∂Q 
+ ρb ∑ i =1  ∆H i i  − hgs a p (T − Ts ) =
0
 ∂t 
Gas phase:
∂ 2T ∂T ∂T ∂vg
−k g ε i 2
+ Cvg vg ρ g + [ε i + (1 − ε i )ε p ]Cvg ρ g +P Eq. 12
∂z ∂z ∂t ∂z
4hw
+ hgs a p (T − TS ) + (T − TW ) = 0
DB
Column wall:
∂ 2Tw ∂T 4 DB
−k w 2 + ρ wC pw w − hw (T − Tw ) Eq. 13
( DB + Wr ) − DB2
2
∂z ∂t
4 ( DB + Wr )
2
+ hb (Tw − Tenv ) =
0
( DB + Wr )
2
− DB2
Where ks is the heat conductivity of adsorbent, Ts is the temperature of adsorbent, Cps is
the specific heat capacity of adsorbent, Cpai is the specific heat capacity of the adsorbed
gas, hgs is the heat transfer coefficient between gas and solid phase, αp is the specific
particle surface area per unit length of bed, kg is the heat conductivity of gas phase, Cvg is
the specific gas phase heat capacity at constant volume, hw is the heat transfer coefficient
between gas and wall, DB is the internal diameter of column, Tw is the column wall
39
Chapter 2
temperature, kw is the heat conductivity of wall, ρw is the column wall density, Cpw is the
specific heat capacity of the wall, Wr is the wall thickness, hb is the heat transfer
coefficient between column and ambient and Tenv is the environmental temperature.
The adsorption column and operational parameters used in the numerical model are same
to the parameters reported in experimental work by Saleman et al. [43], summarized in
Table 2. The axial dispersion coefficient and mass transfer coefficients of CH4 and N2 are
same to values reported by Zhang et al. [54].
Table 2 Adsorption column and operational parameters
Parameter Value Unit

Column details
Height (z) 0.98 m
Internal diameter (DB) 0.035 m
Wall thickness (Wr) 0.0016 m
Wall specific heat capacity (Cpw) 0.5 kJ·kg-1·K-1
Wall-ambient heat transfer coefficient (hb) 10 W·m-2·K-1
Wall thermal conductivity (kw) 16 W·m-1·K-1
Axial dispersion coefficient 3.5 x 10-5 m2·s-1
Operational parameters
Cycle low pressure (PL) 1.4 bar
Cycle high pressure (PH) 5 bar
Environmental temperature (Tenv) 293 K
Feed flowrate (F) 1.25 SLPM
FE/PU step time (tFE/PU) 120 s
PR/BD step time (tPR/BD) 90 s
Adsorbent parameters
Inter-particle voidage (εp) 0.42 m3/m3
Intra-particle voidage (εi) 0.65 m3/m3
Bulk density (ρb) 435 kg·m3
Particle radius (rp) 0.0015 m
Specific heat capacity (Cps) 1.00 kJ·kg-1·K-1
Thermal conductivity (ks) 0.8 W·m-1·K-1
CH4 mass transfer coefficient(kCH4) 1 s-1
N2 mass transfer coefficient (kN2) 3 s-1
40
Chapter 2
The partial differential equations (PDEs) are discretized using the first-order upwind
scheme (UDS1, based on the first-order Taylor expansion), with 100 nodes distributed
along an entire adsorption column. The Implicit Euler method is selected as the primary
integration method. The detailed integrator configuration used to solve the PDEs is listed
in Table 3.
Table 3: Integrator configuration deployed in the model
An algorithm which performs numerical integration by

Integration evaluating the state of the system at the current time and
Implicit Euler
method uses this in conjunction with time derivatives to evaluate
the state of the system at the next time step.
Variable step size Selected to ensure fast calculation speed while
Step size
from 10-5 s to 5s maintaining calculation accuracy.
10 for absolute and
-5
Used to determine convergence by comparing the
Error
relative error maximum change in a variable, or the numerical error
tolerance
tolerances in an equation, with the specified tolerances.
Gaussian elimination was applied to solve a sparse,
Linear
MA48 asymmetric system of m linear equations in n
solver
unknowns.
The Newton method was used for initialization and
Nonlinear steady-state steps, and the Fast Newton method for
Mixed Newton
solver dynamic steps offering a balance between solution
speed and robustness.
2.4.2 The flowsheet of DR-PSA model
The flowsheet of the non-isothermal DR-PSA model is demonstrated in Fig. 5. The
intermediate position of feed inlet is modelled by segmenting the entire adsorption bed
into two separate column sections, acting as enriching section and stripping section,
respectively. The height of each column section is determined by the entire column’s
length and the dimensionless feed inlet point, for instance, is 0.49 m for each section in
this DR-PSA model. While in the Chapter 2, multiple feed positions are introduced along
the adsorption column, which requires that the height of each section should be adjusted
41
Chapter 2
accordingly. The model initialization is accomplished by assigning the concentration at
all points in the pressure-flow network to be equal to that of the feed gas. The temperature
throughout the network is initially set to 20℃, i.e., the environmental temperature
specified in reference [43], and the initial pressures were set as 1.4 bar at the low-pressure
side and as 5 bar at the high-pressure side, as shown in Fig. 5.
Low pressure
BHR1 BLR1
VH1
Enriching Stripping
PID1 VHR1 section VF1 section VLR1 PID2
VH VF Feed
BC BH BF BL
Heavy VL Light
Product Product
VHR2 VF2 VLR2
VH2
BHR2 BLR2
High pressure
Adsorption column Void tank PID controller Valve Gas stream Isentropic Compressor
Fig. 5 Flowsheet of the DR-PL-A model. Valves are coloured black or white to indicate
whether the valves are closed or opened during the feed/purge (FE/PU) step.
The basic modules of the DR-PSA model can be classified as follows:
1) Boundary streams: there are three gas streams forming the external circulation,
flowing into or out of the whole DR-PSA system, marked as “Feed”, “Heavy product”
and “Light product” in Fig. 5. These three streams are specified as non-reversible flow
with fixed pressure (1.4 bar for feed stream while 5 bar for two product streams).
During the pressurization/blowdown (PR/BD) step, the flow rates of three boundary
streams are set to zero, namely no gas is being processed in this step. During the
42
Chapter 2
FE/PU step, the molar flowrates of feed gas is fixed at 1.25 SLPM (standard litres per
minute) while the heavy product flowrate is specified with a constant value calculated
from the heavy to feed (H/F) ratio and the light product flowrate is automatically
adjusted by the proportional-integral-derivative (PID) controller 2 according to the
real-time pressure in tank BL, as illustrated in Fig. 5. The feed gas concentration is
also specified as a constant whist the concentrations of heavy and light product
streams are specified as free variables and are predicted by the simulation.
2) Two PID controllers are used to stabilize the low pressure (1.4 bar) and high pressure
(5 bar) of the system. PID 1 altering the volumetric flowrate through the compressor
to guarantee a constant pressure in tank Bc during the FE/PU step and maintain a
constant volumetric flowrate to complete the pressure inversion during the PR/BD
step. PID 2 ensure the fixed high pressure in tank BL by controlling the molar flowrate
through control valve VL during the FE/PU step. The performance of a PID controller
can be well tuned by altering the triple key parameters: proportional, integral, and
derivative terms of a PID controller.
3) Void tanks: the void tanks used in this model represents two different spaces existed
in the real experiments: one is the gas tank and buffer tank which used to store the gas
and avoid the pressure or concentration shock, with a high volume (0.1–1 L); the other
one acts as a pressure node to match the so-called P-F-P connection requirement in
Aspen Adsorption and represents the piping manifolds or dead volume in the system,
with a low volume (5 mL). In Saleman’s experiments, there is no heat exchanger
connecting to the compressor; thus, the cooling down of outlet gas from the
compressor (temperature can rise as high as 200℃) is mainly achieved by the heat
transfer from the physical piping manifolds to the environment. To simulate this heat
transfer, the void tank after the compressor (BH) serves as a “cooler” by specifying it
with a large heat transfer coefficient (10000W·m-2·K-1).
43
Chapter 2
4) Valves: valves are used to alter the flow direction and update steps of the DR-PSA
cycle. Aspen Adsorption platform provide four valve states: 0, fully closed, which
blocks all flow; 1, fully open, which allows unrestricted flow with no pressure drop;
2, constant CV, where the flowrate is impacted by the pressure drop between the
upstream and the downstream; 3, constant molar flowrate, the molar flowrate is
constant, independent of pressure behaviour. In this model, the valves are specified
according to the flow status of different streams during the cycle, for instance, the
states of VLR1 and VLR2 are switched between 0, 2 and 3 at different steps.
2.4.3 Benchmark results and discussion
The numerical model is benchmarked by using the experimental results from the
publication of Saleman et al. [43] as the baseline to compare pressure and temperature
profiles, internal and external flows and product gas concentrations between experimental
measurements and simulated results. These comparisons are performed under the cyclic
steady state (CSS). The criterion for judging whether the simulation has achieved the CSS
is based on the error of the solid loading amount and temperature between two successive
cycles. The tolerance error is 10-5 in this model, under which the simulation can be
considered as arriving at the CSS.
The accurate pressure and temperature profiles are crucial to ensure the correct prediction
of separation results by simulation. The simulated and experimental pressure history
profiles are shown in Fig. 6. The dotted line represents the pressure variation at the first
cycle in the experiment while the solid line represents the pressure curve at the final cycle
when the system arrives the CSS in the experiment. The dashed line is the predicted
pressure swing by simulation at the CSS. As shown in Fig. 6, the simulated pressure curve
shows good agreement with the experimental measurements, which indicates that the
44
Chapter 2
pressure history profiles obtained by simulation are precise enough to predict the real
pressure variation in the DR-PSA unit.
600
I II III IV
500
Bed Pressure / kPa
400
300
200
100
0 60 120 180 240 300 360 420
Time / s
Fig. 6 Comparison of pressure history profiles in simulation and experiment reported by
Saleman et al. [43]: dotted line and solid line indicate the pressure variation during the
first cycle and last cycle in the experiment, respectively while dashed line represents the
predicted pressure curve in a whole cycle at the CSS by simulation. Adapted with
permission from Ref. [43]. Copyright (2015) Elsevier.
Fig. 7 illustrates the comparison of bed temperature trajectories between the experiment
and simulation. The solid lines trace the experimental data, and the dotted lines represent
the simulated data. T01 (exp), T06 (exp) and T11 (exp) represent the thermal probes in
the experimental rig, which are used to monitor the temperature variations at the top,
middle and bottom position of the adsorption bed. T01 (sim), T06 (sim) and T11 (sim)
are simulated temperature curves at the same locations along the adsorption column. The
environmental temperature in the numerical model is set to 20 ℃, the same as the
experimental condition. T01 records the temperature swing at the stripping end within a
range from 18 ℃ to 22 ℃, indicating that adsorption and desorption rarely happen at this
location. T06 shows the temperature variation at the middle of the column, which is more
45
Chapter 2
significant than T01 and between 16–26 ℃, indicating that both the adsorption and
desorption front can pass through this position during the feed and purge steps. T11 views
a similar temperature changing trend as T06, however the average temperature at this
position is lower than that at the middle, which is chiefly due to the adsorption heat effect
in the bed. For instance, during the PR and PU steps when adsorption happens in the
column, the heavy gas flows from the bottom end (T11) to the middle (T06) and gas
temperature increases gradually due to the adsorption heat released, causing a higher
temperature of T06 than T11, additionally, during the BD and FE steps (desorption), the
light reflux stream flows reversely from T06 to T11, resulting in a comparatively lower
temperature at the bottom. The deviations between simulation results and experimental
data are considered reasonable and are limited within 2 ℃, which are mainly caused by
the introduction of a constant heat transfer coefficient.
28
T01 (exp)
T06 (exp)
26
T11 (exp)
T01 (sim)
24 T06 (sim)
T11 (sim)
Temperature (
22
20
℃
18
)
Environment temperature
16
14
0 100 200 300 400
Time (s)
Fig. 7 Comparison of temperature history profiles in simulation and experiment reported
by Saleman et al. [43]: T01 (blue), T06 (red) and T11 (green) indicate temperature
variations at top, middle and bottom positions of the adsorption column, respectively. The
solid lines represent experimental measurements while the dotted lines represent the
simulated results. Adapted with permission from Ref. [43]. Copyright (2015) Elsevier.
46
Chapter 2
The flowrates of internal and external gas streams are crucial operating parameters which
have significant impacts on the performance of DR-PSA cycles. The comparisons of
internal and external gas flowrates are carried out to further validate the model’s
simulation accuracy.
Fig. 8 compares the flowrates of internal circulation, i.e., the light (RL) and heavy (RH)
reflux streams, between simulated (dotted lines) and experimental results (solid lines).
The light reflux molar flowrate is specified as constant (1.992 SLPM) in both experiments
and simulations, hence, there is no fluctuation of light reflux flowrate during the FE and
PU steps. During the stable pressure steps (FE and PU steps), the average value of the
simulated heavy reflux flowrates is in good agreement with the average value of the
experimental results. The main difference is found during the pressure-reverse steps (PR
and BD steps) where the calculated heavy reflux flowrates are significantly higher than
the measured ones, especially in the first few seconds of these steps. The cause of this
phenomenon is explained by Saleman et al. that the flowrate of heavy reflux gas exceeds
the full-range scale of the flow meter used in the experiment, resulting in the absence of
actual flowrate recording in the first few seconds of the PR/BD step. In the experiment,
there is a small rise following the continuous decrease of heavy reflux flowrates at the
end of PR/BD step, which signifies an automatic switch of the compressor from the low-
to the high-power operating mode.
47
Chapter 2
RH (exp)
RL (exp)
12
RH (sim)
RL (sim)
Flowrate (slpm) 8
0
0 60 120 180 240 300 360 420
Time (s)
Fig. 8 Comparison of heavy (blue) and light (green) reflux flowrate profiles in the
experiment reported by Saleman et al. [43] and the simulation under the CSS. Adapted
with permission from Ref. [43]. Copyright (2015) Elsevier.
The external flowrate profiles of experiment and simulation are shown in Fig. 9. The
flowrate of feed gas (F) is fixed at 1.25 SLPM in all experiment and simulation cases.
The heavy product gas is specified as a constant value according to the heavy to feed ratio
(referred to as H/F ratio) of each case. Hence, the trajectories of feed and heavy product
flowrates are smooth in both experiment and simulation. The flowrate of light (L) product
stream is controlled by a back-pressure regulator in the experiment which is challenging
to switch the start/stop operation of this stream, which leads to the light product only
flowing in a short period during the FE and PU steps. In the simulation, the light product
flowrate is regulated by a PID controller, which makes the light product flowrates more
uniform during the FE/PU step. Overall, the agreement between calculated and measured
flow profiles is reasonable.
48
Chapter 2
3.5
L (exp) H (exp) F (exp)
L (sim) H (sim) F (sim)
3.0
2.5
Flowrate (slpm)
2.0
1.5
1.0
0.5
0.0
0 60 120 180 240 300 360 420
Time (s)
Fig. 9 Comparison of flowrate profiles of boundary gas streams: feed (red), heavy (blue)
and light (green) gases in the experiment Saleman et al. [43] and simulation at the CSS.
Adapted with permission from Ref. [43]. Copyright (2015) Elsevier.
Following the identification of reasonable similarities in pressure, temperature and flow
history profiles between simulation and experiment, the ensuing step is to verify the
accuracy of the predicted separation results. The simulated and experimental cycle-
averaged separation results at the CSS are summarized in Table 4.
The operating parameters in simulations are specified according to the corresponding
values used in experiments, including the flowrates of feed gas (F), heavy gas (H) and
light reflux (RL) as well as the feed CH4 concentration (yF). The purity of N2 and CH4
product gases, light gas flowrate (L) and heavy reflux flowrate (RH) are selected for the
comparison. The root mean squared deviation (RMSD) is used to evaluate the average
error between the experimental results and simulated results, can be calculated with Eq.
15.
∑ ( x − xˆ )
N 2
i i
RMSD = i =1
Eq.15
N
49
Chapter 2
Where i denotes the case No., N denotes the number of total cases, xi denotes the
experimental results of case i and xˆi denotes the simulated results of case i.
Table 4 Summarization of simulated and experimental results
Simulated results Experimental results (Ref. [43])

Heavy Light Light Heavy Heavy Light Light Heavy
Case
reflux product gas gas reflux product gas gas
No.
(SLPM) (SLPM) N2 (%) CH4 (%) (SLPM) (SLPM) N2 (%) CH4 (%)
1 4.284 0.354 99.1 68.6 3.78 0.32 99.2 65.9
2 5.361 0.423 85.6 68.0 5.102 0.394 93.3 68.3
3 4.375 0.397 98.5 71.4 4.017 0.387 97.2 72.5
4 4.239 0.988 99.6 47.0 4.03 0.965 99.3 45.7
5 4.059 0.991 99.0 44.8 3.826 0.955 99.3 46.2
6 2.939 0.979 98.5 40.8 3.099 1.003 98.9 44.8
7 5.156 1.005 98.9 47.2 4.529 0.967 99.2 48.4
8 2.223 1.013 96.8 39.2 2.558 1.028 96.3 40.2
9 4.348 1.033 98.4 50.5 4.013 1.03 98.4 52.1
10 4.588 1.065 97.7 54.8 4.322 1.057 98 57.8
11 4.816 1.118 95.9 60.4 4.527 1.099 96.1 64.6
12 4.440 1.083 97.0 55.9 4.293 1.044 97.6 61.5
13 3.289 1.090 96.1 51.8 3.43 1.046 96.7 58.0
14 5.407 1.079 97.1 55.7 4.84 1.044 97.1 55.8
15 5.108 1.021 98.6 49.2 4.341 0.951 99.1 42.4
16 3.108 0.815 99.5 28.0 3.095 0.767 99.7 25.8
17 3.505 0.912 99.3 35.5 3.287 0.854 99.5 32.4
18 4.754 1.166 93.9 63.1 4.451 1.101 94.3 66.0
19 3.314 0.998 99.8 18.1 3.423 0.959 99.9 19.2
20 2.534 1.023 99.6 19.0 2.837 1.001 99.8 21.2
21 4.214 1.172 99.7 32.6 3.971 1.153 99.7 35.7
22 2.200 1.196 99.1 26.8 2.501 1.205 98.7 27.9
23 6.095 1.191 99.2 32.2 5.265 1.077 99.3 32.5
24 4.720 1.222 98.8 45.8 4.353 1.165 98.4 51.3
50
Chapter 2
Fig. 10 shows the deviation of the yH and yL results between the experiment and
simulation. The R.M.S.D values are 0.021 for yH and 0.007 for yL, which indicates that
this model can accurately predict the performance of DR-PSA and can be used to analyze
and optimize the newly developed cycles in this research.
8
yH
6
yL
4
2
ysim-yexp (%)
-2
-4
-6
-8
0 4 8 12 16 20 24
Case No.
Fig. 10 Deviation of the CH4 concentration in heavy (yH, marked as black) and light (yL,
marked as red) product gases between experiments and simulations.
51
Chapter 2
Reference
1. S. Brunauer, L. S. Deming, W. E. Deming, and E. Teller, On a Theory of the van
der Waals Adsorption of Gases. Journal of the American Chemical Society, 1940.
62(7): p. 1723-1732.
2. K. V. Kumar, S. Gadipelli, B. Wood, K. A. Ramisetty, A. A. Stewart, C. A.
Howard, D. J. L. Brett, and F. Rodriguez-Reinoso, Characterization of the
adsorption site energies and heterogeneous surfaces of porous materials. Journal
of Materials Chemistry A, 2019. 7(17): p. 10104-10137.
3. I. Langmuir, The Adsorption of Gases on Plane Surfaces of Glass, Mica and
Platinum. Journal of the American Chemical Society, 1918. 40(9): p. 1361-1403.
4. A. M. Czerny, P. Benard, and R. Chahine, Adsorption of nitrogen on granular
activated carbon: experiment and modeling. Langmuir, 2005. 21(7): p. 2871-5.
5. N. P. Stadie, M. Murialdo, C. C. Ahn, and B. Fultz, Anomalous isosteric enthalpy
of adsorption of methane on zeolite-templated carbon. J Am Chem Soc, 2013.
135(3): p. 990-3.
6. E. Glueckauf, Theory of chromatography. Part 10.—Formulæ for diffusion into
spheres and their application to chromatography. Trans. Faraday Soc., 1955.
51(0): p. 1540-1551.
7. C. H. Liaw, J. S. P. Wang, R. A. Greenkorn, and K. C. Chao, Kinetics of fixed-
bed adsorption: A new solution. AIChE Journal, 1979. 25(2): p. 376-381.
8. Z. Li and R. T. Yang, Concentration profile for linear driving force model for
diffusion in a particle. AIChE Journal, 1999. 45(1): p. 196-200.
9. P. Guerin De Montgareuil and D. Domine, Process for separating a binary gaseous
mixture by adsorption, 1964, U.S. Patent No. 3,155,468.
10. W. A. H. Cutler, Lin.; Sorensen, Charles M.; Owens, Jr.Lorraine K. , Structured
adsorbents for desulfurizing fuels, 2004, WO2004058372A2.

52
Chapter 2
11. K. P. Gadkaree, Carbon honeycomb structures for adsorption applications.
Carbon, 1998. 36(7-8): p. 981-989.
12. L. Bonaccorsi, A. Freni, E. Proverbio, G. Restuccia, and F. Russo, Zeolite coated
copper foams for heat pumping applications. Microporous and Mesoporous
Materials, 2006. 91(1-3): p. 7-14.
13. J. T. Richardson, D. Remue, and J. K. Hung, Properties of ceramic foam catalyst
supports: mass and heat transfer. Applied Catalysis A: General, 2003. 250(2): p.
319-329.
14. B. G. C. Keefer, Alain.; Sellars, Brian.; Shaw, Ian.; Larisch, Belinda. , Adsorbent
laminate structures, 2004, US6692626B2.
15. A. M. Bailey, Frederick A. P. , Filter assemblies with layers of activated carbon
fibrous cloth, 1980, US4234326A.
16. H. Thakkar, S. Lawson, A. A. Rownaghi, and F. Rezaei, Development of 3D-
printed polymer-zeolite composite monoliths for gas separation. Chemical
Engineering Journal, 2018. 348: p. 109-116.
17. W. D. Marsh, F. S. Pramuk, R. C. Hoke, and C. W. Skarstrom, Pressure
equalization depressuring in heatless adsorption, 1964, U.S. Patent No. 3,142,547.
18. M. M. Hassan, N. S. Raghavan, and D. M. Ruthven, Pressure swing air separation
on a carbon molecular sieve—II. Investigation of a modified cycle with pressure
equalization and no purge. Chemical Engineering Science, 1987. 42(8): p. 2037-
2043.
19. K. Warmuzinski, Effect of pressure equalization on power requirements in PSA
systems. Chemical Engineering Science, 2002. 57(8): p. 1475-1478.
20. J. A. Delgado and A. Rodrigues, Analysis of the boundary conditions for the
simulation of the pressure equalization step in PSA cycles. Chemical Engineering
Science, 2008. 63(18): p. 4452-4463.
53
Chapter 2
21. A. Ntiamoah, J. Ling, P. Xiao, P. A. Webley, and Y. Zhai, CO2 capture by vacuum
swing adsorption: role of multiple pressure equalization steps. Adsorption, 2015.
21(6-7): p. 509-522.
22. B.-K. Na, K.-K. Koo, H.-M. Eum, H. Lee, and H. K. Song, CO2 recovery from
flue gas by PSA process using activated carbon. Korean Journal of Chemical
Engineering, 2001. 18(2): p. 220-227.
23. A. Malek and S. Farooq, Study of a six-bed pressure swing adsorption process.
AIChE Journal, 1997. 43(10): p. 2509-2523.
24. G. Hu, Q. Zhao, M. Manning, L. Chen, L. Yu, E. F. May, and K. G. Li, Pilot scale
assessment of methane capture from low concentration sources to town gas
specification by pressure vacuum swing adsorption (PVSA). Chemical
Engineering Journal, 2022. 427: p. 130810.
25. G. Hu, Q. Zhao, L. Tao, P. Xiao, P. A. Webley, and K. G. Li, Enrichment of low
grade CH4 from N2/CH4 mixtures using vacuum swing adsorption with activated
carbon. Chemical Engineering Science, 2021. 229.
26. L. Dehdari, P. Xiao, K. Gang Li, R. Singh, and P. A. Webley, Separation of
hydrogen from methane by vacuum swing adsorption. Chemical Engineering
Journal, 2022. 450.
27. G. Hu, G. Xiao, Y. Guo, M. Manning, L. Chen, L. Yu, K. G. Li, and E. F. May,
Separation of methane and nitrogen using ionic liquidic zeolites by pressure
vacuum swing adsorption. AIChE Journal, 2022. 68(7).
28. R. Yang and P. Cen, Improved pressure swing adsorption processes for gas
separation: by heat exchange between adsorbers and by using high-heat-capacity
inert additives. Industrial & Engineering Chemistry Process Design and
Development, 1986. 25(1): p. 54-59.
54
Chapter 2
29. D. Basmadjian and L. A. Pogorski, Process for the separation of gases by
adsorption, 1966, U.S. Patent No. 3,279,153.
30. S. P. Reynolds, A. Mehrotra, A. D. Ebner, and J. A. Ritter, Heavy reflux PSA
cycles for CO2 recovery from flue gas: Part I. Performance evaluation. Adsorption,
2008. 14(2-3): p. 399-413.
31. C. Liu, Y. Zhou, Y. Sun, W. Su, and L. Zhou, Enrichment of coal-bed methane
by PSA complemented with CO2 displacement. AIChE Journal, 2011. 57(3): p.
645-654.
32. Y. Yang, Y. Wu, H. Liu, A. M. Ribeiro, P. Li, J. Yu, and A. E. Rodrigues,
Enrichment of ventilation air methane by adsorption with displacement
chromatography technology: Experiment and numerical simulation. Chemical
Engineering Science, 2016. 149: p. 215-228.
33. S. Zhang, Y. Huo, J. Ye, S. Tang, and D. Ma, Displacement desorption test of
coalbed methane and its mechanism exploring. Chinese Science Bulletin, 2005.
50(S1): p. 158-162.
34. D. Qu, Y. Yang, Z. Qian, P. Li, J. Yu, A. M. Ribeiro, and A. E. Rodrigues,
Enrichment of low-grade methane gas from nitrogen mixture by VPSA with CO2
displacement process: Modeling and experiment. Chemical Engineering Journal,
2020. 380.
35. D. Diagne, M. Goto, and T. Hirose, New Psa Process with Intermediate Feed Inlet
Position Operated with Dual Refluxes - Application to Carbon-Dioxide Removal
and Enrichment. Journal of Chemical Engineering of Japan, 1994. 27(1): p. 85-
89.
36. D. Diagne, M. Goto, and T. Hirose, Experimental-Study of Simultaneous
Removal and Concentration of CO2 by an Improved Pressure Swing Adsorption
Process. Energy Conversion and Management, 1995. 36(6-9): p. 431-434.
55
Chapter 2
37. D. Diagne, M. Goto, and T. Hirose, Parametric Studies on CO2 Separation and
Recovery by a Dual Reflux Psa Process Consisting of Both Rectifying and
Stripping Sections. Industrial & Engineering Chemistry Research, 1995. 34(9): p.
3083-3089.
38. D. Diagne, M. Goto, and T. Hirose, Numerical analysis of a dual refluxed PSA
process during simultaneous removal and concentration of carbon dioxide dilute
gas from air. Journal of Chemical Technology and Biotechnology, 1996. 65(1): p.
29-38.
39. Y. Guo, G. Hu, K. Chen, J. Guo, P. A. Webley, and G. K. Li, Capture of dilute
methane with a novel dynamic‐feed dual‐reflux pressure swing adsorption process.
AIChE Journal, 2022. 68(1): p. e17390.
40. J. A. Mc Intyre, C. E. Holland, and J. A. Ritter, High Enrichment and Recovery
of Dilute Hydrocarbons by Dual-Reflux Pressure-Swing Adsorption. Industrial &
Engineering Chemistry Research, 2002. 41(14): p. 3499-3504.
41. J. A. McIntyre, A. D. Ebner, and J. A. Ritter, Experimental Study of a Dual Reflux
Streams. Industrial & Engineering Chemistry Research, 2010. 49(4): p. 1848-
1858.
42. R. Wakasugi, A. Kodama, M. Goto, and T. Hirose, Dual Reflux PSA Process
Applied to VOC Recovery as Liquid Condensate. Adsorption, 2005. 11(S1): p.
561-566.
43. T. L. Saleman, G. Li, T. E. Rufford, P. L. Stanwix, K. I. Chan, S. H. Huang, and
E. F. May, Capture of low grade methane from nitrogen gas using dual-reflux
pressure swing adsorption. Chemical Engineering Journal, 2015. 281: p. 739-748.
44. E. F. May, Y. Zhang, T. L. H. Saleman, G. Xiao, G. Li, and B. R. Young,
Demonstration and optimisation of the four Dual-Reflux Pressure Swing
56
Chapter 2
Adsorption configurations. Separation and Purification Technology, 2017. 177:
p. 161-175.
45. D. T. Kearns and P. A. Webley, Modelling and evaluation of dual-reflux pressure
swing adsorption cycles: Part I. Mathematical models. Chemical Engineering
Science, 2006. 61(22): p. 7223-7233.
46. D. Li, Y. Zhou, Y. Shen, W. Sun, Q. Fu, H. Yan, and D. Zhang, Experiment and
simulation for separating CO2/N2 by dual-reflux pressure swing adsorption
process. Chemical Engineering Journal, 2016. 297: p. 315-324.
47. D. Wawrzyńczak, I. Majchrzak-Kucęba, K. Srokosz, M. Kozak, W. Nowak, J.
Zdeb, W. Smółka, and A. Zajchowski, The pilot dual-reflux vacuum pressure
swing adsorption unit for CO2 capture from flue gas. Separation and Purification
Technology, 2019. 209: p. 560-570.
48. C. Tian, Q. Fu, Z. Ding, Z. Han, and D. Zhang, Experiment and simulation study
of a dual-reflux pressure swing adsorption process for separating N2/O2.
Separation and Purification Technology, 2017. 189: p. 54-65.
49. R. Weh, G. Xiao, E. Sadeghi Pouya, and E. F. May, Helium recovery and
purification by dual reflux pressure swing adsorption. Separation and Purification
Technology, 2022. 288.
50. A. D. Ebner and J. A. Ritter, Equilibrium theory analysis of dual reflux PSA for
separation of a binary mixture. AIChE Journal, 2004. 50(10): p. 2418-2429.
51. D. T. Kearns and P. A. Webley, Modelling and evaluation of dual reflux pressure
swing adsorption cycles: Part II. Productivity and energy consumption. Chemical
Engineering Science, 2006. 61(22): p. 7234-7239.
52. T. S. Bhatt, G. Storti, and R. Rota, Optimal design of dual-reflux pressure swing
adsorption units via equilibrium theory. Chemical Engineering Science, 2013. 102:
p. 42-55.
57
Chapter 2
53. T. S. Bhatt, A. Sliepcevich, G. Storti, and R. Rota, Experimental and Modeling
Analysis of Dual-Reflux Pressure Swing Adsorption Process. Industrial &
54. Y. Zhang, T. L. H. Saleman, G. Li, G. Xiao, B. R. Young, and E. F. May, Non-
isothermal numerical simulations of dual reflux pressure swing adsorption cycles
for separating N2 + CH4. Chemical Engineering Journal, 2016. 292: p. 366-381.
55. R. Weh, G. Xiao, M. A. Islam, and E. F. May, Nitrogen Rejection by Dual Reflux
Pressure Swing Adsorption Using Engelhard Titanosilicate Type 4. Industrial &
56. Y. Zou, G. Xiao, G. Li, W. Lu, and E. F. May, Advanced non-isothermal dynamic
simulations of dual reflux pressure swing adsorption cycles. Chemical
Engineering Research and Design, 2017. 126: p. 76-88.
57. H. R. S. Anna, A. G. Barreto Jr, F. W. Tavares, and M. B. de Souza Jr, Machine
learning model and optimization of a PSA unit for methane-nitrogen separation.
Computers and Chemical Engineering Journal, 2017. 104: p. 377-391.
58. S. G. Subraveti, Z. Li, V. Prasad, and A. Rajendran, Machine Learning-Based
Multiobjective Optimization of Pressure Swing Adsorption. Industrial &
59. K. N. Pai, V. Prasad, A. J. S. Rajendran, and P. Technology, Experimentally
validated machine learning frameworks for accelerated prediction of cyclic steady
state and optimization of pressure swing adsorption processes. 2020: p. 116651.
60. K. T. Leperi, D. Yancy-Caballero, R. Q. Snurr, and F. You, 110th Anniversary:
Surrogate Models Based on Artificial Neural Networks To Simulate and Optimize
Pressure Swing Adsorption Cycles for CO2 Capture. Industrial & Engineering
Chemistry Research, 2019. 58(39): p. 18241-18252.
58
Chapter 2
61. F. Ye, S. Ma, L. Tong, J. Xiao, P. Bénard, and R. Chahine, Artificial neural
network based optimization for hydrogen purification performance of pressure
swing adsorption. International Journal of Hydrogen Energy, 2019. 44(11): p.
5334-5344.
59
Chapter 3
Chapter 3: Capture of dilute methane with a novel

dynamic-feed dual reflux pressure swing adsorption
process
3.1 Chapter perspectives
Mixing problem is an unavoidable problem of fixed-feed DR-PSA processes and is detrimental
to the separation performance. The feed gas will keep diluting the column gas when adsorption
wavefront goes past the feed position, limiting the product purity and recovery. This chapter
focuses on developing a dynamic-feed strategy by switching the feed inlet according to the
column concentration profiles to solve the mixing problem. The dynamic-feed DR-PSA shows
better results when using the uniform space method to determine feed inlet location of each
sub-step. The column concentration profiles of the dynamic-feed DR-PSA cycle is significantly
smoother than that of the fixed-feed DR-PSA cycle. Results show that both increase in purity
(from 47.5% to 53.5%) and recovery (from 72.2% to 81.1%) can be seen when adopting the
dynamic-feed method.
This chapter was published as a paper in AIChE Journal.
60
Received: 15 April 2021 Revised: 6 July 2021 Accepted: 29 July 2021
DOI: 10.1002/aic.17390
SEPARATIONS: MATERIALS, DEVICES AND PROCESSES
Capture of dilute methane with a novel dynamic-feed

dual-reflux pressure swing adsorption process
Yalou Guo | Guoping Hu | Kaifei Chen | Jining Guo | Paul A. Webley |

Gang Kevin Li
Department of Chemical Engineering, The

University of Melbourne, Parkville, Victoria, Abstract
Australia Dual-reflux pressure swing adsorption (DR-PSA) is a state-of-art adsorption technol-
Correspondence ogy claiming to achieve both high product purity and recovery. However, its perfor-
Gang Kevin Li, Department of Chemical mance is inevitably limited by the gas mixing problem at the feed inlet position in
Engineering, The University of Melbourne,
Parkville, VIC 3010, Australia. adsorption columns, where a discontinuity in the columns' axial concentration pro-
Email: li.g@unimelb.edu.au files disturbs the separation. Here, we developed a dynamic-feed DR-PSA (DF-DR-
Funding information PSA) process to solve the mixing problem and obtain better separation performance
Australia Research Council, Grant/Award by introducing multiple feed inlets along the column. The DF-DR-PSA process was
Number: DP190100983
demonstrated by enrichment methane (CH4) from a binary gas mixture containing
2.4 mol.% CH4 in nitrogen (N2) using activated carbon material. Non-isothermal
dynamic models were constructed to investigate the effects of operating parameters
on the separation performance. The results showed the DF-DR-PSA process
achieved both higher methane purity (53.5% vs. 47.5%) and recovery (81.1%
vs. 72.2%) over the conventional DR-PSA process at the same energy consumption.
KEYWORDS
binary mixture, gas separation, methane upgrading, numerical modeling, pressure swing
adsorption
1 | I N T RO DU CT I O N gas) simultaneously by combining two conventional PSA units using

an intermediate feed position and a double-reflux (i.e., light and
Pressure swing adsorption (PSA) has been commonly investigated for heavy reflux) strategy, hence the name of DR-PSA. The schematic
gas separations, such as hydrogen purification,1 air separation,2 noble diagram of a DR-PSA process is presented in Figure 1. This process
gases purification,3 separation of hydrogen isotopes,4 carbon dioxide consists of four fundamental steps: feed (FE), pressurization (PR),
5 6
(CO2) capture, and methane (CH4) upgrading. Conventional PSA purge (PU), and blowdown (BD). Since the feed gas can be injected
processes can be deployed in either stripping or enriching into the high-pressure (PH) or the low-pressure (PL) adsorption col-
configurations,7 and they often produce only one high-purity product umn, the pressure reversal can be accomplished by transferring gas
by sacrificing its recovery due to their thermodynamic restrictions.8 between the enriching ends of both columns, which are rich in the
However, many gas separation processes often require both the heavy component (i.e., the more adsorbable, A) or the stripping
enriched and lean gas streams to be relatively pure, that is, targeting ends that are rich in the light component (i.e., the less adsorbable,
at high recoveries and purities simultaneously. B). DR-PSA cycles have four basic configurations: DR-PL-A, DR-
Dual-reflux PSA (DR-PSA) is a relatively new adsorption pro- PH-A, DR-PL-B, and DR-PH-B.7
9 10
cess proposed by Hirose and Leavitt in the 1990s. This process The DR-PSA process has been widely investigated and success-
aims at overcoming the process limitations of conventional PSA fully demonstrated to achieve both high-purity light and heavy prod-
and achieving two high-purity gas products (i.e., enriched and lean ucts simultaneously for various applications. The seminal experimental
1 of 11 © 2021 American Institute of Chemical Engineers. wileyonlinelibrary.com/journal/aic AIChE J. 2022;68:e17390.

https://doi.org/10.1002/aic.17390
GUO ET AL. 2 of 11
deviations in the simulated CH4 mole fractions ranging from 0.02 to

0.08 for the heavy product. Zhang et al.27 developed a non-isothermal
numerical model in the software package Aspen Adsorption and com-
pared their results with experiments by Saleman et al.22 In this work,
they used an ideal compressor during the constant pressure (FE and
PU) steps, resulting in a high average material balance error of 8.6%;
hence, the flow rate of light product and the concentrations of both
product streams needs to be corrected manually. Zou et al.28 reported
an advanced non-isothermal simulation model of the DR-PSA process
and compared the predicted results with the experimental data from
the study by McIntyre et al.14 and Saleman et al.22 In this study, the
isentropic compressor module and the proportional–integral–
derivative (PID) controller module were applied to achieve the desired
pressure profile, the standard deviation of the simulated predictions
was 0.003 mol fraction in the ethane (C2H6) product stream compared
with the experimental results from the study by McIntyre et al. and
0.04 in the CH4 product stream compared with the experimental data
from the study by Saleman et al. May and coworkers29–32 conducted
F I G U R E 1 Schematic diagram of a complete dual-reflux pressure
a series of simulation-based studies of upgrading subquality natural
swing adsorption cycle in four basic configurations: feed to high-
pressure or low-pressure adsorption column, and heavy (A) or light gas by the DR-PSA cycle, in which activated carbon, ionic liquidic
(B) gases for pressure inversion zeolite,34 and ETS-4 were used as adsorbents successively.
The mixing problem that manifests as a discontinuity in the col-
umns' axial concentration profiles has been reported by Diagne12
8
DR-PSA studies were conducted by Diagne et al., in which zeolite and McIntyre,16 shown in Figure 2A,B. As illustrated in Figure 2A,
MS 13X was used to separate CO2 from the air with a PH-A configu- the CO2 concentration profiles are smooth on the upstream side of
ration. The results showed that a feed gas stream with 20 mol.% CO2 the feed inlet position. For instance, in the case of ZF/L = 0.6, the
was enriched to 90 mol.% in the heavy product stream and stripped CO2 molar fraction in the reflux flow is roughly 85%, dropping dra-
to 2.5 mol.% in the lean gas stream. Since then, several research11–20 matically to around 20% after mixing with the feed gas (consisting
have proven that DR-PSA processes showed high separation perfor- of 20% CO2). Figure 2B indicates a similar dip of ethane concentra-
mance for various applications, such as CO2 capture, hydrocarbon tion (from approximately 57% to about 32%) just after the feed
21
upgrading, and air separation. Bhatt et al. performed a series of position. Overall, this mixing problem particularly occurs at the feed
experiments for concentrating 5 mol.% CH4 in N2 using the DR-PH-A inlet position in adsorption columns, caused by the concentration
process and Norit RB1 activated carbon adsorbents. The CH4 fraction mismatch between the feed flow and upstream reflux flow, and can
in the enriched gas stream was 55.0 mol.%, and the N2 purity in the hamper the separation performance of the DR-PSA process. A
22
lean gas stream was 98.5 mol.%. Saleman et al. reported the separa- range of researchers have focused on optimizing the feed position
tion of CH4 and N2 mixtures with 2.4–49.6 mol.% CH4 by a by matching the concentration of the feed gas and upstream reflux
laboratory-scale DR-PL-A apparatus. The methane enrichment ratio flow to minimize the effects of such concentration discontinuity
could reach 21, upgrading CH4 2.4 mol.% to 51.0 mol.%; however, the problem.12,31,35 Nevertheless, as the axial concentration profiles in
methane recovery was only 34.0%. the adsorption columns are time-dependent and the optimum feed
There have been several simulation investigations of DR-PSA position is also dynamic, this mixing problem is unavoidable for the
processes, ranging from equilibrium-based analytical models7,23–26 to DR-PSA process with a fixed feed location and feed concentration.
rigorous non-isothermal numerical models.20,27–32 Ebner and Ritter23 In this work, we proposed a dynamic-feed (DF) DR-PSA process by
established the first mathematical model based on equilibrium theory introducing multiple feed positions along with the adsorption columns to
with a few simplified isotherm assumptions. They found that the DR- a conventional DR-PSA process. This process is then demonstrated and
PSA process is feasible over a wide range of operation conditions, compared with conventional DR-PSA processes using non-isothermal
even with a pressure ratio as low as 1.1. Kearns and Webley7 further numerical simulations in Aspen Adsorption. A binary gas mixture of
advanced the equilibrium theory model and identified the nomencla- 2.4 mol.% CH4 + 97.6 mol.% N2 was used as the feed gas, a common
ture of four basic DR-PSA configurations. Bhatt et al.21,33 demon- vent stream in industrial liquefied natural gas facilities. The robustness
strated an isothermal model of the DR-PSA process to describe their and reliability of the model were validated using the experimental data
experiments. Two separate interactions (i.e., heavy reflux interaction reported by Saleman et al.22 A sensitivity study was also conducted to
and pressure reverse interaction) were adopted to build the model on investigate the impacts of the number of feed positions and the schedule
the Aspen Adsorption platform. This model successfully predicted the of multiple feed steps. The outcome of this study will advance the
product compositions observed in their reported experiments with knowledge of PSA and foster high-performance gas separations.
3 of 11 GUO ET AL.
F I G U R E 2 Mixing problem reported by Diagne (A) and McIntyre (B). The red line indicates the sharp concentration increase, and the blue line
exhibits a significant concentration decrease. (A) CO2 concentration profiles at various values of the feed inlet position, adapted with permission
from Ref. 12. Copyright (1995) American Chemical Society. (B) Ethane concentration profiles in the column at the end of the heavy reflux step,
adapted with permission from Ref. 16. Copyright (2010) American Chemical Society
2 | SIMULATION DETAILS simulation model is achieved by assigning the concentration at all

points in the PF network to be equal to that of the feed gas mixture.
2.1 | Model equations and description The temperature throughout the system was initially set to 293.15 K,
the same as the environment temperature specified in simulations
The non-isothermal models of DR-PSA and DF-DR-PSA are con- reported by Zhang et al.27 The initial pressures were set as 1.4 bar at
structed in Aspen Adsorption v10.0, a commercial software that can the low-pressure side and 5 bar at the high-pressure side.
simulate the operation of dynamic adsorption cycles with rigorous The model is configured to use the Norit RB3, a kind of activated
pressure-flow (PF) networks and mass/energy balances. The carbon used in the experiments of Saleman et al.22 Operational and
governing equations and assumptions used in this model are summa- bed model parameters, including the physical properties of the adsor-
rized in Table S1. bent used, are provided in Table 1.
The partial differential equations used for the bed model are dis- The basic modules of the PF network are classified as follows:
cretized using the Upwind Differencing Scheme 1, with 100 nodes
distributed across a column; the Implicit Euler method is selected as 1. Boundary streams: three boundary streams are named “Feed,” “Heavy
the primary integration method; and the integration error tests only product,” and “Light product” (as identified in Figure 3), specified as
included states; and the integrator configuration used in this simula- nonreversible with constant pressure and temperature. During the
tion work is detailed in Table S2. PU/BD steps, the flowrates of boundary streams are set as zero. Dur-
ing the FE/PU steps, the molar flowrates of feed gas and heavy prod-
uct streams are specified with a constant value, while the flowrate of
2.2 | The flowsheet of DF-DR-PSA model the light product stream is controlled by a PID controller.
2. The PID controllers are applied to set the low and high pressures
The flowsheet of the DF-DR-PSA process used to construct the non- of the cycle. PID1 is introduced to determine the system's low
isothermal model in Aspen Adsorption is shown in Figure 3. The multi- pressure by controlling the volumetric flowrate through the com-
ple positions of the feed inlet are modeled by segmenting the entire pressor during the FE/PU steps and used to maintain a constant
column into several column sections based on the simulated axial con- volumetric flowrate (5 105 m3/s) during the PR/BD steps. PID2
centration profiles of the conventional DR-PSA process. The height of is used to set the high pressure of the cycle by controlling the
each column is specified in line with the entire column's length (Hb) molar flowrate of the light product stream during the FE/PU steps.
and dimensionless feed inlet points (zF). The minimum length of each 3. Void tanks: two kinds of void tanks are used in this model. One is used
column section is set at 0.049 m (i.e., five grid nodes) to provide suffi- to represent the practical gas tank and acted as a buffer tank to avoid
cient space between the inlet locations. The initialization of the the pressure shock, set with a high volume value; and the other one is
GUO ET AL. 4 of 11
F I G U R E 3 Process flowsheet diagram for the numerical model to simulate the conventional dual-reflux pressure swing adsorption process
and the dynamic-feed dual-reflux pressure swing adsorption process. The number of adsorption columns (N + 1) is determined by the number of
feed positions (n). PID1, PID controller for the compressor; PID2, PID controller for the Light product valve; VF, system feed valve; VF1-n, column
1 feed valve at feed point n; VF2-n, column 2 feed valve at feed point n; VH, system heavy product valve; VH1, column 1 heavy product valve; VH2,
column 2 heavy product valve; VHR1, column 1 heavy reflux valve; VHR2, column 2 heavy reflux valve; VL, system light product valve; VLR1, column
1 light reflux valve; VLR2, column 2 light reflux valve
served as a pressure node to match the so-called “P-F-P” connection Hence, this model accurately predicts the performance of the DR-PSA
requirement in Aspen Adsorption, specified with a little volume value. apparatus and can be exploited for further study.
4. Valves: valves are used to alter the flow direction and update the
steps of the DR-PSA cycle. In Aspen Adsorption, valves could be
specified as four statuses: 0, fully closed; 1, fully open; 2, constant 3 | RESULTS AND DISCUSSION
CV; and 3, constant molar flowrate. In this model, the valves are
specified according to the flow status of different streams during This rigorous non-isothermal model was used to numerically study the
the cycle; for instance, VLR1 and VLR2 would be switched between separation performance of the DF-DR-PSA process for concentrating
0, 2, and 3 at different steps. dilute methane gas mixture (containing 2.4 mol.% methane) under var-
ious operating conditions.
2.3 | Model validation

3.1 | Determination of dynamic-feed range
This non-isothermal numerical model of DR-PSA was validated against
the experimental results from Saleman et al.,22 consisting of 24 runs The concentration discontinuity in the axial profiles for conventional
of various feed compositions (2.4–49.6 mol.% CH4 balanced with N2) DR-PSA processes, considered as the “mixing problem” within the
and operating parameters (H/F ratio and RL, i.e., the heavy to feed column, is due to the mixing of the feed stream and the upstream
flow ratio and the light reflux flowrate). The comparison of the simu- reflux flow,12 as shown in Figure 5A. Such discontinuity and sharp
lated and experimental cycle-averaged concentration of methane in concentration changes cause significant perturbation to the column
both light (yL) and heavy product (yH) streams at the cyclic steady and hamper the separation performance. Previous studies12,31,35 opti-
state (CSS) are shown in Figure 4. In this simulation, the criterion for mized the feed position via matching the concentration between the
judging whether the simulation had achieved the CSS was built on the upstream reflux flow and the feed mixture. However, as illustrated in
error of the total loading and solid-phase temperature between Figure 5A, the column's adsorption front is time-dependent during the
the two successive cycles. The simulation will be considered at the feed step, moving from the stripping end to the enriching end, which
5
CSS when the error is lower than the preset tolerance, which is 10 means that there are varying optimum feed inlet positions at different
in this model. At CSS, the average overall material balance error is times. Hence, the DF strategy was exploited to solve the so-called
0.3%, whereas the average CH4 and N2 component material balance “perturbation” problem. In Figure 5, the black dashed line indicates a
error is 3.5% and 0.7%, respectively, comparable in magnitude to methane fraction of 2.4 mol.%, that is, the feed methane concentra-
those of other DR-PSA models reported in the open litera- tion. The DF range is determined by the two interactions of the
ture.27,29,32,33 The root-mean-square deviation (RMSD) evaluates the dashed line and axial concentration distribution curves at the start
average error between the experimental and simulated results. (0 s) and the end (120 s) of the feed step, along which the DF strategy
The RMSD values for yH and yL are 0.021 and 0.007, respectively. could be adopted, as indicated in Figure 5B.
5 of 11 GUO ET AL.
TABLE 1 Operational and bed model parameters used in the simulation
Parameter Value Unit Source

Column parameters
Height (Hb) 0.98 m 29
Column internal diameter 0.035 m 29
Wall thickness (Wt) 0.0016 m 29
1 1
Wall specific heat capacity (Cpw) 0.5 kJ kg K 29
Wall-ambient heat transfer coefficient 10 W m2 K1 29
1 1
Wall thermal conductivity 16 Wm K 29
Void tank heat transfer coefficient 1000 W m2 K1 27
5 2 1
Axial dispersion coefficient 3.5 x 10 m s 29
Cycle low pressure 1.4 bar 22
Cycle high pressure 5 bar 22
Ambient temperature 293.15 K 22
Feed flowrate 1.25 SLPM 22
FE/PU step time 120 s 22
PR/BD step time 90 s 22
Inter-particle voidage 0.42 m3/m3 29
Intra-particle voidage 0.65 m3/m3 29
Bulk density 435 kg/m3 29
Particle radius 0.0015 m 29
Particle shape factor 0.752 — 29
1 1
Specific heat capacity 1.0 kJ kg K 29
Thermal conductivity 0.8 W m1 K1 29
Adsorbate parameters
N2 CH4
Mass transfer coefficient 3 1 s1 29
Heat capacity 29 35 kJ kmol1 K1 29
Isotherm parameters
IP1 7.30 x 107 6.58 x 107 kmol kg1 bar1 29
IP2 1722 2077 K 29
4 4 1
IP3 1.75 x 10 1.19 x 10 bar 29
IP4 1722 2077 K 29
1
Adsorption heat (ΔH) 14.322 17.27 kJ mol 22
Gas-phase heat conductivity 0.16 0.16 W m1 K1 29
3.2 | Sensitivity study of the number of feed 2. Dividing the DF range equally into multiple column sections with
points identified feed locations and finding the time points at which the
upstream reflux concentration is equal to the feed concentration,
The switch policy of feed position was determined based on the simulated that is, the uniform space method.
axial concentration profiles of the DR-PSA configuration. Two methods
were implemented to optimize the DF policy for the DF-DR-PSA process: The impact of feed inlet points numbers (n) on the separation per-
formance was investigated, as shown in Figure 6. The results indicate
1. Dividing the feed step averagely into multiple sub-steps and locat- that the DF-DR-PSA process, introducing either of the mentioned
ing the optimum feed inlet position for each sub-step, that is, the approaches, can achieve better separation performance than the con-
uniform time method. ventional DR-PSA process. As the feed inlet positions increase from
GUO ET AL. 6 of 11
FIGURE 4 Deviations of ysim predicted in the numerical simulation from the corresponding values yexp measured in experiments22
F I G U R E 5 (A) Axial concentration distributions in the conventional dual-reflux pressure swing adsorption column at the time point of 0, 20,
40, 60, 80, 100, and 120 s during the feed step (operating parameters are specified as RL = 2.302, H/F ≈ 0.042, and tFE/PU = 120 s). z is the
dimensionless axial position along the column, z = 0 is the enriching end, z = 1 is the stripping end, and feed position is zF = 0.7. (B) Enlarged
diagram of the DF range
3 to 8, the methane purity (yH) and recovery increased by 2.7 mol.% perturbation at each feed point, indicating that the dynamic-feed
and 5.8%, respectively. The cycles generally perform better when strategy is capable of solving the mixing problem. It is noteworthy that
exploiting the uniform space method in terms of product purity, the concentration profile using the uniform space method is generally
whereas for the methane recovery, there is no significant difference smoother than the one using the uniform time method, which explains
between these two methods. why the uniform space method performs better. Therefore, the uni-
Figure 7A,B, respectively, shows the methane axial concentration form space method was adopted to achieve the DF target in the fol-
profiles of the two methods: case (a) uses the uniform time strategy, lowing study due to its better separation performance. The
whereas case (b) uses the uniform space method; both runs had eight independence of computed results on the mesh division of adsorption
feed inlet positions. As illustrated in Figure 7, compared with the frac- column sections was investigated to avoid the performance improve-
tion profiles in Figure 5, the axial concentration distribution curves of ments driven by different meshing between the DR-PSA and DF-DR-
the DF-DR-PSA process are smoother and have no significant PSA configurations, as detailed in Table S3 and Figure S1.
7 of 11 GUO ET AL.
F I G U R E 6 The purity (A) and recovery (B) of methane by dynamic-feed dual-reflux pressure swing adsorption cycles as a function of n (the
number of feed inlet points). The operating parameters are RL = 2.302, H/F ≈ 0.042, and tFE/PU = 120 s. The dashed line represents the result
using the conventional dual-reflux pressure swing adsorption (DR-PSA) process under the same operating conditions
F I G U R E 7 Axial concentration distributions in the dynamic-feed dual-reflux pressure swing adsorption column at the time point of 0, 20,
40, 60, 80, 100, and 120 during the feed step by the uniform time method (A) and the uniform space method (B). The operating parameters are
RL = 2.302, H/F ≈ 0.042, and tFE/PU = 120 s
3.3 | Parametric study of the DF-DR-PSA process 3.3.1 | Effect of the H/F ratio
The purities and recoveries of the conventional DR-PSA used in this The theoretical boundary values of y H and y L are a function of the
section are selected at the optimum feed position. The complete ratio of the heavy product flowrate to feed flowrate (H/F) and
results of purity and recovery at different feed positions under varying the feed gas concentration (y F ) and shown in Equations (1)
operating conditions obtained by the conventional DR-PSA process and (2).
are plotted as Figures S2–S4. The parameters H/F ratio, light reflux
8
flowrate (RL), and feed step time (tFE/PU), which have been proven to > ≤ 1, if H=F ≤ yF
<
y
significantly impact the separation performance of DR-PSA cycles, yH ≤ F , if H=F ≥ yF ð1Þ
>
: H
were chosen to conduct the parametric study in this work. F
GUO ET AL. 8 of 11
3.3.2 | Effect of light reflux flowrate

8
>
> H
>
> y
< ≥ F F , if H=F ≤ y
F Reflux flowrate is another crucial operating parameter, which has sig-
yL H ð2Þ
>
> 1
>
> F nificant impacts on separation performance. Since the heavy reflux
:
≥ 0, if H=F ≥ yF flowrate (RH) is coupled to the light reflux flowrate (RL), in this study,
the light reflux flowrate was specified as flow-controlled and the
As the feed flowrate is constant (F = 1.25 SLPM) in this study, heavy reflux flowrate was set as pressure-controlled. The H/F ratio
the H/F ratio can be adjusted by changing the heavy product flow- was set constant at 0.042, and the feed step duration is 120 s. The
rate. The separation performance of conventional DR-PSA and simulated purity and recovery are plotted in Figure 10 as a function of
DF-DR-PSA processes is shown in Figure 8 for relevant H/F ratios. RL. As shown in Figure 10, for the conventional DR-PSA process,
According to Equations (1) and (2), the perfect separation perfor- product purity yH first increases when the light reflux flow raises from
mance can only be obtained when H/F is equal to yF for an ideal
DR-PSA process, which is impractical in reality. The separation
performance can be biased towards higher purity or recovery by
varying the H/F ratio for an actual cycle. Figure 8 shows an
increase in purity and decrease in recovery with a decrease in the
H/F ratio, which applies to both processes. The numerical simula-
tion results in the figure show that the performance of the DF-DR-
PSA cycle, in terms of both purity and recovery, is generally better
than that of the conventional DR-PSA process. The DF-DR-PSA
process can obtain both higher purity (53.5% vs. 47.5%) and recov-
ery (81.1% vs. 72.2%) compared with the conventional DR-PSA
process.
A Pareto plot of methane recovery versus product purity is
shown in Figure 9 to illustrate their competitive relationship. The
arrow indicates the direction of decreasing H/F ratios. The purity
is increased at the expense of methane recovery, revealing the
usual trade-off between product purity and recovery. The DF-DR-
PSA shows distinct superiority over the conventional DR-PSA, F I G U R E 9 The correlation between methane recovery and heavy
especially when H/F is around 0.024, which is the theoretically product purity. DF-DR-PSA, dynamic-feed dual-reflux pressure swing
ideal H/F value as discussed. adsorption; DR-PSA, dual-reflux pressure swing adsorption
F I G U R E 8 The effect of H/F ratios on the purity (A) and recovery (B) of methane at constant tFE/PU = 120 s and RL = 2.302 SLPM. The
arrows indicate the direction in which the H/F ratio should be varied to bias the separation performance favoring the purity or the recovery. DF-
DR-PSA, dynamic-feed dual-reflux pressure swing adsorption; DR-PSA, dual-reflux pressure swing adsorption
9 of 11 GUO ET AL.
F I G U R E 1 0 The purity (A) and recovery (B) of methane by two kinds of cycles as a function of the RL. The results of the conventional DR-
PSA process are picked from the optimum fixed feed position. DF-DR-PSA, dynamic-feed dual-reflux pressure swing adsorption; DR-PSA, dual-
reflux pressure swing adsorption
F I G U R E 1 1 The purity (A) and recovery (B) of methane by two processes as a function of the tFE/PU at constant RL = 2.302 SLPM and H/
F = 0.042. DF-DR-PSA, dynamic-feed dual-reflux pressure swing adsorption; DR-PSA, dual-reflux pressure swing adsorption
0.8 to 2.0 SLPM, reaching a maximum value when the RL is around necessary to ensure the adequate desorption of CH4 from the low-
2 SLPM. After that, the product purity declines gradually with a fur- pressure column and guarantee a sufficiently wide dynamic-feed
ther increase in RL. The optimum RL value is between 2 and 2.5 SLPM range for the DF-DR-PSA cycle.
to obtain the best recovery. A similar trend can be observed when
altering RL for the DF-DR-PSA process. The optimum RL is approxi-
mately 2.8 SLPM, corresponding to the best purity and about 2.3 3.3.3 | Effect of feed step duration
SLPM for the best recovery. It is worthy to note that there is a mini-
mum RL value below which the purity of the DF-DR-PSA process is Two steps are involved in the DR-PSA and DF-DR-PSA processes:
even lower than that of conventional DR-PSA configuration. When
the RL value is small, there is insufficient light purge gas to clean the 1. FE/PU step for adsorption in the high-pressure column and
adsorbent. In addition, the DF range is too narrow to set enough feed desorption in the low-pressure column;
positions. Hence, a light reflux flowrate above the threshold value is 2. PR/BD step duration for pressure inversion.
GUO ET AL. 10 of 11
FE/PU step duration is crucial for optimizing the adsorption front product purity, revealing the general trade-off between purity and
position to avoid the breakthrough and maximize CH4 productivity. A productivity.
series of strep durations including 80 s, 100 s, 120 s, 140 s, and 160 s
are investigated to optimize the step duration, as illustrated in
Figure 11. Figure 11A shows that the DF-DR-PSA cycle reaches the 4 | CONC LU SIONS
maximum yH value of 44.5 mol.% at 100 s. The deviations between
the two processes reduce with the decreasing feed step time, which is In this work, we proposed and demonstrated a novel DF DR-PSA pro-
due to the reduction in tFE/PU will shorten the DF range. Figure 11B cess for the enrichment of dilute CH4 (2.4 mol.%) from its mixture with
shows that both the methane purity and recovery using the DF-DR- N2. A performance comparison of the conventional DR-PSA and DF-DR-
PSA process are superior to the conventional DR-PSA process. For PSA processes was conducted with numerical simulation using Norit
instance, when the tFE/PU is 100 s, the purity and recovery are RB3 activated carbon adsorbents. The effects of heavy product/feed
improved by roughly 2% and 5%, respectively. flowrate (H/F) ratio, light reflux flowrate (RL), and step durations (tFE/PU)
Bed productivity is another indicator to evaluate the performance on the process performance of both two processes were investigated.
of a process, which is defined by the amount of product per unit The sensitivity study illustrated that both higher purity and recovery
amount of adsorbent per unit time, as illustrated in Equation (3). could be achieved when more feed inlet positions were set along with
the columns. The simulated results showed that the DF-DR-PSA cycle
yH H tFE=PU performed better than the conventional DR-PSA cycle in terms of prod-
ProdCH4 ¼ ð3Þ
tFE=PU þ tPR=BD mads uct purity, recovery, and productivity. The comparison of methane con-
centration profiles in the adsorption column during the feed step
Here, H is the flowrate of the heavy product stream, and mads suggested that the mixing problem of conventional DR-PSA processes
denotes the mass of the adsorbent. could be practically solved by introducing the DF strategy. The results
In Figure 12, bed productivity is shown at varying feed step dura- showed the DF-DR-PSA process achieved both higher methane purity
tions tFE/PU with a constant light reflux flowrate RL and H/F ratio. For (53.5% vs. 47.5%) and recovery (81.1% vs. 72.2%) over the conventional
the DF-DR-PSA process, productivity ranges from (0.036–0.046) mol DR-PSA process at the same energy consumption.
CH4 kg1 hr1, which is higher than that of the conventional DR-PSA
cycle, whose productivity ranges from 0.34 to 0.043 mol CH4 ACKNOWLEDG MENTS
1 1
kg hr . The DF-DR-PSA cycle achieves higher productivity due to This project is funded by the Australia Research Council
its better product purity results than the conventional DR-PSA config- (DP190100983). Y. Guo would like to thank J. Yu for providing
uration. The productivity reduces with the decreasing values of tFE/PU research facilities to run part of simulations in this work.
as the percentage of gas production time to the total cycle time
decreases. However, such lower tFE/PU values also lead to higher AUTHOR CONTRIBU TIONS
Yalou Guo: Conceptualization (equal); data curation (lead); formal
analysis (lead); investigation (lead); methodology (lead); software
(lead); visualization (lead). Guoping Hu: Software (supporting); writing-
review & editing (equal). Kaifei Chen: Formal analysis (supporting);
visualization (supporting). Jining Guo: Formal analysis (supporting);
visualization (supporting). Paul Webley: Funding acquisition (equal);
supervision (equal). Kevin Gang Li: Conceptualization (equal); funding
acquisition (equal); supervision (lead); writing-review & editing (equal).
DATA AVAILABILITY STAT EMEN T

The data that support the findings of this study are available from the
corresponding author upon reasonable request.
GLOSSARY
BD blowdown step
DR-PSA dual-reflux pressure swing adsorption
DF-DR-PSA dynamic-feed dual-reflux pressure swing adsorption
FE feed step (desorption)
F I G U R E 1 2 The productivity of methane by two kinds of cycles
as a function of the tFE/PU at constant RL = 2.302 SLPM and H/ PID1 PID controller for the compressor
F = 0.042. DF-DR-PSA, dynamic-feed dual-reflux pressure swing PID2 PID controller for the light product valve
adsorption; DR-PSA, dual-reflux pressure swing adsorption PR pressurization step
11 of 11 GUO ET AL.
PU purge step (adsorption) 18. May EF, Zhang Y, Saleman TLH, Xiao G, Li G, Young BR. Demonstra-
VLR1 column 1 light reflux valve tion and optimisation of the four Dual-Reflux Pressure Swing Adsorp-
tion configurations. Sep Purif Technol. 2017;177:161-175.
VLR2 column 2 light reflux valve
19. Shen Y, Zhou Y, Li D, Fu Q, Zhang D, Na P. Dual-reflux pressure
swing adsorption process for carbon dioxide capture from dry flue
ORCID gas. Int J Greenh Gas Control. 2017;65:55-64.
Yalou Guo https://orcid.org/0000-0003-4052-9681 20. Tian C, Fu Q, Ding Z, Han Z, Zhang D. Experiment and simulation
study of a dual-reflux pressure swing adsorption process for separat-
Guoping Hu https://orcid.org/0000-0002-7480-8411
ing N2/O2. Sep Purif Technol. 2017;189:54-65.
Paul A. Webley https://orcid.org/0000-0003-3598-3767 21. Bhatt TS, Sliepcevich A, Storti G, Rota R. Experimental and modeling
Gang Kevin Li https://orcid.org/0000-0003-1700-297X analysis of dual-reflux pressure swing adsorption process. Ind Eng
Chem Res. 2014;53(34):13448-13458.
22. Saleman TL, Li G, Rufford TE, et al. Capture of low grade methane
RE FE R ENC E S
from nitrogen gas using dual-reflux pressure swing adsorption. Chem
1. Silva B, Solomon I, Ribeiro AM, et al. H2 purification by pressure Eng J. 2015;281:739-748.
swing adsorption using CuBTC. Sep Purif Technol. 2013;118:744-756. 23. Ebner AD, Ritter JA. Equilibrium theory analysis of dual reflux PSA for
2. Xu M, Wu H-C, Lin YS, Deng S. Simulation and optimization of pres- separation of a binary mixture. AIChE J. 2004;50(10):2418-2429.
sure swing adsorption process for high-temperature air separation by 24. Bhatt TS, Storti G, Rota R. Optimal design of dual-reflux pressure swing
perovskite sorbents. Chem Eng J. 2018;354:62-74. adsorption units via equilibrium theory. Chem Eng Sci. 2013;102:42-55.
3. Jin X, Malek A, Farooq S. Production of argon from an oxygenargon 25. Rossi E, Storti G, Rota R. Influence of the main operating parameters
mixture by pressure swing adsorption. Ind Eng Chem Res. 2006;45 on the DRPSA process design based on the equilibrium theory.
(16):5775-5787. Adsorption. 2021;27:27-39.
4. Si Y, He X, Jiang J, Duan Z, Wang W, Yuan D. Highly effective H2/D2 26. Bhatt TS, Storti G, Denayer JFM, Rota R. Equilibrium theory-based
separation in a stable Cu-based metal–organic framework. Nano Res. assessment of dual-reflux pressure swing adsorption cycles that utilize
2021;14:518-525. light gas for Pressure Swing. Ind Eng Chem Res. 2018;58(1):350-365.
5. Xu M, Chen S, Seo D-K, Deng S. Evaluation and optimization of VPSA 27. Zhang Y, Saleman TLH, Li G, Xiao G, Young BR, May EF. Non-isothermal
processes with nanostructured zeolite NaX for post-combustion CO2 numerical simulations of dual reflux pressure swing adsorption cycles for
capture. Chem Eng J. 2019;371:693-705. separating N2 + CH4. Chem Eng J. 2016;292:366-381.
6. Hu G, Zhao Q, Tao L, Xiao P, Webley PA, Li KG. Enrichment of low 28. Zou Y, Xiao G, Li G, Lu W, May EF. Advanced non-isothermal
grade CH4 from N2/CH4 mixtures using vacuum swing adsorption dynamic simulations of dual reflux pressure swing adsorption cycles.
with activated carbon. Chem Eng Sci. 2021;229:116152. Chem Eng Res Des. 2017;126:76-88.
7. Kearns DT, Webley PA. Modelling and evaluation of dual-reflux pres- 29. Weh R, Xiao G, Islam MA, May EF. Upgrading sub-quality natural gas
sure swing adsorption cycles: part I. Mathematical models. Chem Eng by dual reflux-pressure swing adsorption using activated carbon and
Sci. 2006;61(22):7223-7233. ionic liquidic zeolite. Chem Eng J. 2020;392:123753-123761.
8. Diagne D, Goto M, Hirose T. New PSA process with intermediate 30. Xiao G, Saleman TL, Zou Y, Li G, May EF. Nitrogen rejection from
feed inlet position operated with dual refluxes - application to methane using dual-reflux pressure swing adsorption with a
carbon-dioxide removal and enrichment. J Chem Eng Japan. 1994;27 kinetically-selective adsorbent. Chem Eng J. 2019;372:1038-1046.
(1):85-89. 31. Lu W, Xiao G, Weh R, May EF. Rejecting N2 from natural gas by dual
9. Hirose T. A simple design method of a new PSA process consisting of reflux pressure swing adsorption with activated carbon. J Nat Gas Sci
both rectifying and stripping sections. Paper presented at: Proceed- Eng. 2020;81:103457.
ings of the 2nd China-Japan-USA Symposium on Adsorption 1991. 32. Weh R, Xiao G, Islam MA, May EF. Nitrogen rejection from natural
10. Leavitt FW. Duplex adsorption process: US Patent 5,085,674; 1992. gas by dual reflux-pressure swing adsorption using activated carbon
11. Diagne D, Goto M, Hirose T. Experimental-study of simultaneous and ionic liquidic zeolite. Sep Purif Technol. 2020;235:116215.
removal and concentration of CO 2 by an improved pressure 33. Bhatt TS, Storti G, Rota R. Detailed simulation of dual-reflux pressure
swing adsorption process. Energ Conver Manage. 1995;36(6–9): swing adsorption process. Chem Eng Sci. 2015;122:34-52.
431-434. 34. Li G, May EF, Webley PA, Huang SH-W, Chan KI. Method for gas
12. Diagne D, Goto M, Hirose T. Parametric studies on CO2 separa- separation. USA 2017.
tion and recovery by a dual reflux PSA process consisting of both 35. Diagne D, Goto M, Hirose T. Numerical analysis of a dual refluxed PSA
rectifying and stripping sections. Ind Eng Chem Res. 1995;34(9): process during simultaneous removal and concentration of carbon diox-
3083-3089. ide dilute gas from air. J Chem Technol Biotechnol. 1996;65(1):29-38.
13. Dong F, Lou H, Goto M, Hirose T. A new PSA process as an extension
of the Petlyuk distillation concept. Sep Purif Technol. 1999;15(1):31-40.
SUPPORTING INF ORMATION
14. McIntyre JA, Holland CE, Ritter JA. High enrichment and recovery of
dilute hydrocarbons by dual-reflux pressure-swing adsorption. Ind Additional supporting information may be found online in the
Eng Chem Res. 2002;41(14):3499-3504. Supporting Information section at the end of this article.
15. Wakasugi R, Kodama A, Goto M, Hirose T. Dual reflux PSA process
applied to VOC recovery as liquid condensate. Adsorption. 2005;11
(S1):561-566. How to cite this article: Guo Y, Hu G, Chen K, Guo J,
16. McIntyre JA, Ebner AD, Ritter JA. Experimental study of a dual reflux
Webley PA, Li GK. Capture of dilute methane with a novel
enriching pressure swing adsorption process for concentrating dilute
dynamic-feed dual-reflux pressure swing adsorption process.
feed streams. Ind Eng Chem Res. 2010;49(4):1848-1858.
17. Li D, Zhou Y, Shen Y, et al. Experiment and simulation for separating AIChE J. 2022;68(1):e17390. https://doi.org/10.1002/aic.
CO2/N2 by dual-reflux pressure swing adsorption process. Chem Eng 17390
J. 2016;297:315-324.
Chapter 3
Supplementary information
Capture of Dilute Methane with a Novel Dynamic-Feed Dual-
Reflux Pressure Swing Adsorption Process
Yalou Guo, Guoping Hu, Jining Guo, Kaifei Chen, Paul A. Webley, Gang
(Kevin) Li
Department of Chemical Engineering, The University of Melbourne, Parkville, VIC 3010,
Australia
Corresponding authors: li.g@unimelb.edu.au
72
Chapter 3
Model details:
The following assumptions are used to simplify the calculations:
• Material and momentum balance
a. All gases are treated as ideal gases;
b. The momentum balance and pressure drop are described using the Ergun equation;
c. The gas accumulation in the inter-particle void space and adsorption flux to the
adsorbent surface are estimated using gas-phase convection with axial dispersion;
constant axial dispersion is described using a fixed axial dispersion coefficient (3.5
x 10-5 m2·s-1).
• Kinetic model
The flux between adsorbed molecules is calculated using the linear driving force (LDF)
model with constant mass transfer coefficients.
• Equilibrium capacity model
Extended Langmuir II isotherms are used to describe competitive adsorption for the
binary gas mixture.
• Energy balance
Both heat convection in the gas phase and thermal conduction in the solid phase are
considered with the specifications listed as follows:
i. Constant heat capacity of the adsorbed phase;
ii. Constant heat of adsorption;
iii. Constant heat transfer coefficients;
iv. Constant gas thermal conductivity;
v. Rigorous heat transfer between the column and environment;
vi. Estimated gas-wall heat transfer coefficient;
vii. Constant solid-phase heat capacity.
73
Chapter 3
Table S1 Governing equations used to describe the material, momentum and energy balance in
the adsorption column in this model.
Description Equation Eq No.

Momentum ∂P  150 × 10−5 µ g (1 − ε i ) 2 1.75 × 10−5 M W ρ g (1 − ε i ) 2  Eq. (S.1)
Ergun equation = ± vg + vg 
∂z  (2rpψ ) ε i
2 3
2rpψε i3 
balance  
Plug flow with ∂ (vg ci ) ∂ci ∂w ∂ 2c

Material balance + [ε i + (1 − ε i )ε p ] + ρb i − ε i DL 2i =0 Eq. (S.2)
axial dispersion ∂z ∂t ∂t ∂z
Fixed dispersion  20 1
Axial dispersion =DL 2vi rp  +  Eq. (S.3)
coefficient  Re ⋅ Sc 2
LDF model with ∂wi

Kinetic model = ki ( wi* − wi ) Eq. (S.4)
constant MTC ∂t
Extended  IP 
IP1,i ⋅ exp  2,i  pi
wi* =  T 
Isotherm model Langmuir Eq. (S.5)
  IP  
1 + ∑ i  IP3,i ⋅ exp  4,i  pi 
isotherm   T  
∂ 2T ∂T ∂T ∂vg
−k gε i + Cvg vg ρ g + [ε i + (1 − ε i )ε p ]Cvg ρ g +P
∂z 2 ∂z ∂t ∂z
Gas phase Eq. (S.6)
4hw
+ hgs a p (T − TS ) + (T − TW ) = 0
DB
∂ 2Ts ∂T ∂T
+ C ps ρb s + ρb ∑ i =1 ( C pai wi ) s
n
−ks 2
∂z ∂t ∂t
Energy balance Solid phase Eq. (S.7)
n  ∂wi 
+ ρb ∑ i =1  ∆H i  − hgs a p (T − Ts ) =
0
 ∂t 
∂ 2Tw ∂T 4 DB
−kw + ρ wC pw w − hw (T − Tw )
( DB + Wr ) − DB2
2
∂z 2 ∂t
Column wall Eq. (S.8)
4 ( DB + Wr )
2
0
( DB + Wr )
2 2
− DB
Equation of state Ideal gas

Pi = ciℜT Eq. (S.9)
equation
74
Chapter 3
Table S2: The detailed descriptions of the integrator configuration in the model
An algorithm that performs numerical integration by

evaluating the state of the system at the current time
Integration method Implicit Euler
and uses this in conjunction with time derivatives to
assess the system's state at the next time step.
Variable step size from Selected to ensure fast calculation speed while
Step size
10 s to 5 s
-5
maintaining calculation accuracy.
Used to determine convergence by comparing the
10-5 for absolute and
Error tolerance maximum change in a variable, or the numerical error
relative error tolerances
in an equation, with the specified tolerances.
Gaussian elimination was applied to solve a sparse,
Linear solver MA48 asymmetric system of m linear equations in n
unknowns.
The Newton method was used for initialization and
steady-state steps, and the Fast Newton method for
Nonlinear solver Mixed Newton
dynamic steps offering a balance between solution
speed and robustness.
75
Chapter 3
Independence study:
The entire adsorption column was divided into 10 column sections. The feed positions,
determined based on the axial concentration profiles of DR-PSA cycles, decide each section's
length and the corresponding nodes number (as 1 node always represents 0.0098 m). Therefore,
the independence of the simulated results on the number of nodes of each column section was
investigated. 5 cases were conducted in which only the length (i.e., the nodes number) of each
column section was different. The operating parameters were the same as those specified in
run 21 of Saleman's experiments. The feed position was fixed at ZF=0.5. Table S3 exhibits the
nodes number of each section in 5 cases.
Table S3: The mesh division of each column section
Column section
Number of nodes
No.
1 10 5 5 5 5
2 10 5 5 5 5
3 10 10 5 5 5
4 10 10 5 5 5
5 10 10 10 5 5
6 10 10 10 5 5
7 10 10 10 10 5
8 10 10 10 10 5
9 10 10 10 10 10
10 10 20 30 40 50
Case No. 1 2 3 4 5
76
Chapter 3
The simulated results of yH and yL are illustrated in Fig. S1. Results show that the impact of
mesh division on calculated results of CH4 molar fraction in two product streams is minimal.
The variation of yH and yL is within 0.24 mol.% and 0.01 mol.%, which can be ignored.
Fig. S1 The independence study of the simulated yH and yL results with the mesh division of
the adsorption column.
77
Chapter 3
The simulation results of the DR-PSA configuration at different feed positions under varying
operating parameters are illustrated in Fig. S2 – S4. The results at the optimum feed position
were selected to compare with the separation performance achieved by the DF-DR-PSA
process.
a) 0.44 tFE/PU=80s
tFE/PU=100s
0.42
tFE/PU=120s
0.40 tFE/PU=140s
0.38 tFE/PU=160s
yH (mol/mol)
0.36
0.34
0.32
0.30
0.28
0.26
0.0 0.2 0.4 0.6 0.8 1.0
ZF
90
b) tFE/PU=80s
tFE/PU=100s
85 tFE/PU=120s
tFE/PU=140s
80 tFE/PU=160s
Recovery (%)
75
70
65
0.0 0.2 0.4 0.6 0.8 1.0

ZF
Fig. S2 The purity (a) and recovery (b) of methane in the heavy product at different feed
positions with varying feed step durations.
78
Chapter 3
a) 0.60
H/F=0.016
H/F=0.0224
0.55 H/F=0.032
H/F=0.0424
0.50 H/F=0.048
H/F=0.056
yH (mol/mol)
0.45
increase H/F ratio
0.40
0.35
0.30
0.25
0.0 0.2 0.4 0.6 0.8 1.0
ZF
b)
90
80
Recovery (%)
70
60 increase H/F ratio
50
40
0.0 0.2 0.4 0.6 0.8 1.0

ZF
positions with varying H/F ratios.
79
Chapter 3
a) 0.45
0.40
0.35
0.30
yH (mol/mol)
RL=0.811
0.25 RL=1.237
RL=1.600
0.20 RL=2.003
RL=2.302
0.15
RL=2.796
0.10 RL=3.300
RL=3.775
0.05
0.0 0.2 0.4 0.6 0.8 1.0
ZF
b)
80
70
Recovery (%)
RL=0.811
60
RL=1.237
RL=1.600
50 RL=2.003
RL=2.302
RL=2.796
40
RL=3.300
RL=3.775
30
0.0 0.2 0.4 0.6 0.8 1.0
ZF
positions with varying RL.
80
Chapter 3
Notation
specific heat capacity of the adsorbed

Cpai r radial coordinate of the adsorbent
gas
specific heat capacity of the
Cps rp particle radius
adsorbent
Cpw specific heat capacity of the wall Sc Schmidt number
specific gas phase heat capacity at
Cvg T gas temperature
constant volume
ci molar fraction of component i Tenv environmental temperature
DB internal diameter of column Ts adsorbent temperature
DL dispersion coefficient Tw column wall temperature
heat transfer coefficient between
hb vi interstitial gas velocity
column and ambient
heat transfer coefficient between gas
hgs vg superficial gas velocity
and solid
hw Wr wall thickness
and wall
isotherm parameter X for component adsorbent loading of component i per
IPXi wi
i unit mass of adsorbent
adsorbent loading of component i per
constant mass transfer coefficient of
ki wi * unit mass of adsorbent in equilibrium
component i
with its partial pressure in gas phase
kg heat conductivity of gas z axial distance coordinate
specific particle surface area per unit
ks heat conductivity of adsorbent αp
length of bed
enthalpy of adsorption for component
kw heat conductivity of wall ΔHi
i
MW molecular weight of gas mixture εi bed voidage
P gas pressure εp particle voidage
PH high pressure µg viscosity of gas mixture
PL low pressure ρb bulk density of adsorbent
pi partial pressure of component i ρg gas molar density
ℜ universal gas constant ρw column wall density
Re Reynolds number ψ particle shape factor
81
Chapter 4
Chapter 4: Separation of He/N2/CH4 ternary

mixtures by a triple-reflux pressure swing adsorption
process
Although DR-PSA process shows excellent performance for separating binary gas mixtures, it
is challenging to achieve the separation of ternary gases mixture via a one-step DR-PSA
system. Multi-stage process usually requires higher capital and operational cost when
compared with one-step process. In this chapter, a triple-reflux PSA (TR-PSA) cycle is
demonstrated for separating CH4/He/N2 mixture using the activated carbon as adsorbents. The
TR-PSA is designed keeping key features of a DR-PSA cycle while adding an intermediate
outlet to side draw the middle product (N2) gas and an intermediate inlet for middle reflux
stream. The parametric studies are detailed by investigate the impacts of several operating
parameters. The TR-PSA can simultaneously achieve enrichment of three components: helium
is enriched from 10% to 45.3%, CH4 is enriched from 20% to 60% and N2 is enriched from 70%
to 95.8%. TR-PSA shows comparable separation performance when compared with a two-
stage DR-PSA cycle but requires lower capital and power cost.
This chapter was published as a paper in AIChE Journal.
82
Received: 16 September 2021 Revised: 16 November 2021 Accepted: 28 December 2021
DOI: 10.1002/aic.17569
SEPARATIONS: MATERIALS, DEVICES AND PROCESSES
Separation of He/N2/CH4 ternary mixtures by a triple-reflux

Yalou Guo1 | Vincent Jusko2 | Gongkui Xiao2 | Jurgen Hanekom2 |

1,2 1 2 1
Guoping Hu | Paul A. Webley | Eric F. May | Gang Kevin Li
1
Department of Chemical Engineering, The
University of Melbourne, Parkville, Victoria, Abstract
Australia Conventional pressure swing adsorption (PSA) processes can only produce one high
2
Fluid Science and Resources Research
purity product in a single stage, whereas the state-of-art dual-reflux PSA (DR-PSA)
Division, Department of Chemical Engineering,
The University of Western Australia, Crawley, can produce two high purity products simultaneously. However, multicomponent gas
Western Australia, Australia
separation is often required in the industry, targeting at recovering several valued
Correspondence products at the same time. In this study, we propose a novel adsorption process,
Gang Kevin Li, Department of Chemical
namely triple-reflux PSA (TR-PSA), to separate three components simultaneously. A
Engineering, The University of Melbourne,
Parkville, VIC 3010, Australia. middle product outlet and a middle reflux stream were introduced to the adsorption
Email: li.g@unimelb.edu.au
columns of a conventional DR-PSA process to separate ternary mixtures of nitrogen,
Funding information methane, and helium. Nonisothermal dynamic models were built to investigate the
Australia Research Council, Grant/Award
impacts of operating parameters particularly the location of the middle reflux/
Number: DP190100983
product stream and the middle reflux flow rates. Results showed that the TR-PSA
process successfully separated ternary mixtures obtaining three enriched products
simultaneously in a single stage, yielding a separation performance comparable to
that of the double-stage DR-PSA with significantly lower capital and energy cost.
KEYWORDS
gas separation, helium, natural gas, pressure swing adsorption, ternary mixture
1 | I N T RO DU CT I O N rejection unit (NRU).6 The LNG process produces three main streams:
the LNG product, the end-flash gas, and the N2 vent gas. Both the
Helium (He) is a nonrenewable noble gas, which has a wide range of end-flash gas and N2 vent gas, consisting primarily of nitrogen and a
applications in the space industry, scientific research, medical facilities, few percent of methane and helium, can be used as the feed gas and
and many other critical areas with increasing market demand.1,2 injected into a helium extraction unit (HeXU), generally integrated
Hence, helium has been admitted as an indispensable strategic with the NRU, to obtain the crude helium (molar fraction is commonly
resource.2,3 As the concentration of He in the air is only 5.2 ppm (part ≥50%) product.2,7 Such HeXU is typically operated by cryogenic pro-
per million), helium production by air separation is extremely energy cesses, which requires even lower operating temperatures (below
intensive.4 The helium reserve on the Earth is also quite limited, and 93 K) and more restrictions of impurity quantity than LNG processes.
most commercially viable helium sources are from helium bearing nat- Therefore, the use of additional HeXU will significantly increase the
ural gas (NG) fields, which contains up to 1–8 vol.% helium.5 Cur- capital and operating cost of the LNG production facilities.
rently, concentrated helium is primarily obtained during the Pressure swing adsorption (PSA) has been widely used in various
production of liquefied NG (LNG) as a by-product. gas separation processes due to its small footprint, high separation
LNG production facilities are commonly operated at cryogenic efficiency, and flexibility. Various process innovations have been
conditions, containing a liquefaction unit (with an operating tempera- reported to improve the separation performance of PSA over the
ture around 112 K), a heavy hydrocarbon removal unit and a nitrogen years, such as layered PSA process using multiple adsorbents in
AIChE J. 2022;e17569. wileyonlinelibrary.com/journal/aic © 2022 American Institute of Chemical Engineers. 1 of 17

https://doi.org/10.1002/aic.17569
2 of 17 GUO ET AL.
superimposed layers in the columns,8,9 multicolumn PSA/vacuum numerical model is reliable for predicting the CH4/N2 separation per-
swing adsorption with pressure equalization step,10 PSA process formance of DR-PSA.
11
involving a displacement step by a more adhesive gas, pressure tem- Despite the success of DR-PSA for separating binary mixtures, a
perature swing adsorption,12 integrated adsorption and absorption single-stage PSA process that can separate ternary mixtures and produce
process,13 and the dual-reflux PSA (DR-PSA).14 three product gases is yet to be developed. In this study, we proposed a
The DR-PSA with a lateral feed and two reflux streams was first novel triple-reflux PSA (TR-PSA) process, which is accomplished by intro-
15 16
proposed by Hirose and Leavitt to overcome the process limita- ducing a middle product stream and a middle reflux stream along the
tions (stripping or enriching limitation14) and achieve both high purity adsorption columns of a conventional DR-PSA process, as shown in
and recovery. The four standard DR-PSA configurations were identi- Figure 2. The TR-PSA process is then demonstrated using nonisothermal
fied by Kearns and Webley,17 which are schematically shown in numerical simulations in Aspen Adsorption. Ternary mixtures containing
Figure 1, that is, PH-A, PL-A, PH-B, and PL-B. Following the seminal different concentrations of CH4, N2, and He, representing various indus-
DR-PSA apparatus demonstrated by Diagne et al.,14,18,19 extensive trially important gases, were used as feedstocks to investigate the char-
experimental results have indicated the potential and efficacy of DR- acteristics of the TR-PSA cycles. A ternary mixture consisting of
PSA processes at the separation of different binary mixtures, such as 70 mol.% N2, 20 mol.% CH4, and 10 mol.% was utilized to conduct para-
20,21 22–28
air separation, hydrocarbon concentration, and CO2 metric studies to gain further insight into the impacts of operating
capture.29–32 parameters on the separation performance of the TR-PSA process. The
Several simulation studies on DR-PSA have been reported, rang- findings of TR-PSA may be extended to the design of multiple reflux
ing from simplified analytical models17,33–38 to rigorous nonisothermal PSA with more than one side-draw products under the same principle.
26–28,39–44 45
numerical models. Based on the equilibrium theory, sim-
plified analytical models revealed critical insights into the operating
parameters and separation performance of DR-PSA. However, the 2 | SIMULATION METHOD
simplified assumptions of isothermal behavior, no axial pressure drop,
complete separation, and local equilibrium lead to inaccurate predic- 2.1 | Adsorption isotherm parameters
tions and very limited application. The first nonisothermal numerical
model of the DR-PSA process was constructed by Zhang et al.40 to The Langmuir isotherm equation (Equation (1)) was used to describe
simulate the separation of CH4 and N2 using Norit RB3 activated car- the adsorption isotherms of pure CH4 and N2 on Norit RB3, with iso-
bon as adsorbents. The accuracy of this model was validated by therm parameters calculated by Zhang et al.40 Due to its highly inert
24
experiments, and the root mean square deviations (RMSDs) of the nature, the adsorption capacity of helium at ambient temperature on
CH4 concentrations were only 0.3 and 2.4 mol.% for the light and carbonaceous adsorbents is negligible. Here, the isotherm parameters
heavy products, respectively. Therefore, the rigorous nonisothermal of He were determined by fitting the reported experimental data46
with the Langmuir isotherm equation (Equation (1)). Single-component
F I G U R E 1 Schematic diagram of DR-PSA configurations: feed to

high- (PH) or low-pressure (PL) column and pressure reversal with the F I G U R E 2 Schematic diagram of the TR-PSA process. BD,
heavy (i.e., the more adsorbable, A) or light (i.e., the less adsorbable, B) blowdown; FE, feed; PR, pressurization; PU, purge. Valves in solid
adsorbate represent closed
GUO ET AL. 3 of 17
T A B L E 1 Isotherm parameters of
CH4 He N2
CH4, He, and N2
1 1 7 7
IP1 (kmol kg bar ) 6.58 10 7.13 10 7.30 107
IP2 (K) 2077 172.0 1722
1 4 5
IP3 (bar ) 1.19 10 6.99 10 1.75 104
IP4 (K) 2077 172.0 1722
F I G U R E 3 Isotherms of pure CH4, He, and N2 on Norit RB3

activated carbon at 293 K
isotherm parameters of CH4, He, and N2 on this adsorbent are sum-

marized in Table 1. These parameters were used in the extended
Langmuir model (Equation (2)) to describe the competitive adsorption
of CH4, He, and N2 in gas mixtures.
IP1 expðIP2 =T Þ p
Q ¼ : ð1Þ
1 þ IP3 expðIP=T Þ p
IP expðIP2,i =T Þ pi
Qj ¼ P1,i : ð2Þ
1þ i ðIP3,i expðIP4,i =T Þ pi Þ F I G U R E 4 Process flow diagram of the TR-PSA model during
FE/PU step in the PL-A configuration. Solid and hollow valves
indicate close or open positions, respectively
Figure 3 shows the isotherms of CH4, He, and N2 on activated carbon
at 293 K, calculated using the isotherm parameters. The adsorption of
CH4 on this adsorbent is much stronger than that of He and N2. The and middle reflux inlet position. There are four external gas streams:
adsorption capacity of helium on Norit RB3 is close to zero, which is “feed,” “heavy product,” “light product,” and “middle product,” which
tally with the actual situation. are specified as nonreversible flow with constant pressure. At cyclic
steady state (CSS), the flow rate of feed gas is equal to the sum of
flow rates of heavy, light, and middle products due to the constraining
2.2 | Pressure flow network of overall material balance. The flow rates of feed gas, heavy product,
and light product are specified directly, whereas the flow rate of mid-
The process flow diagram applied for the model construction and dle product is automatically controlled by a proportional-integral-
numerical simulation is depicted in Figure 4. Each of the two identical derivative (PID) controller (C1). The internal material cycle contains
columns is segmented into several sections in this model to represent three reflux streams: the light, middle, and heavy reflux streams. As
side feed and side draws, the length of each section is determined for the internal material balance, the heavy reflux flow rate is con-
according to the feed position, middle product extraction location, strained by the light and middle reflux stream, which is ensured by
4 of 17 GUO ET AL.
T A B L E 2 Operational and adsorption

Parameter Value Unit Reference
bed parameters
Column parameters
Height 0.98 m 40
Internal diameter 0.035 m 40
Wall thickness 0.0016 m 40
1 1
Wall specific heat capacity 0.5 kJ kg K 40
Wall-ambient heat-transfer coefficient 10 W m2 K1 40
1 1
Wall thermal conductivity 16 Wm K 40
Cycle low pressure 1 bar
Cycle high pressure 5 bar
Ambient temperature 293 K 24
Feed flow rate 1.25 SLPM 24
Inter-particle voidage 0.42 m3 m3 40
3 3
Intra-particle voidage 0.65 m m 40
Bulk density 435 kg m3 40
Particle radius 0.0015 m 40
Particle shape factor 0.752 40
Specific heat capacity 1.0 kJ kg1 K1 40
Thermal conductivity 0.8 W m1 K1 40
Mass-transfer coefficient
CH4 mass-transfer coefficient 1 s1 40
1
He mass-transfer coefficient 0 s
N2 mass-transfer coefficient 3 s1 40
introducing the PID controller (C2). C2 is used to satisfy the internal equations were discretized using the upwind differencing scheme 1, a
material balance by altering the volumetric flow rate through the com- first-order differential scheme based on the first-order Taylor expan-
pressor during the FE/PU step and to maintain a constant flow rate of sion with 100 grid nodes per adsorption column, which can ensure
around 1.5 104 m3 s1 (varied based on the pressure profiles of the accuracy of the calculated results, as can be inferred from
each simulation) for pressure reversal during the PR/BD step. C3 is Figure S2. The implicit Euler integration method was adopted to solve
only adopted when running the cases of DR-PSA process. The operat- the differential algebraic equation group with the MA48 linear solver
ing mode of each valve was varied at different steps of the TR-PSA and the mixed Newton nonlinear solver.47 The error tolerance, used
cycle as listed in Table S1. Three product tanks (i.e., tank 2, 4 and 8 in to determine convergence, is set as 105 for absolute and relative
Figure 4) are used as surge tanks to buffer pressure and composition error tolerance. The CSS criterion is the relative variation in the total
fluctuations. Other void tanks are added due to the “pressure setter– loading and total solid temperature between two consecutive cycles
flow setter–pressure setter” connection requirement in Aspen lower than the CSS test tolerance,47 set as 105. The TR-PSA process
47
Adsorption and represent the piping manifolds. Volumes for each of simulation usually reached CSS after around 300 cycles (45,000 s
void tanks used in this model are summarized in Table S2. integrator time), taking approximately 10 h for computation of real
The simulation parameters are listed in Table 2. As the adsorption time with 6 simulations running in parallel on an AMD 3900X 12-core
of helium is negligible, the impact of its adsorption kinetics is insignifi- processor. At CSS, the overall material balance error j100%
cant. The mass-transfer coefficient of helium is set at 0 s1. The simu- (in out)/inj was roughly 0.2% for all simulations, whereas the com-
lation was initialized by setting the gaseous and adsorbed phase ponent material balance errors j100% (in out)/inj of CH4, He, and
compositions everywhere within the PF network to be equal to the N2 were around 0.5%, 0.1%, and 0.2%, respectively.
feed gas. The initial temperature throughout the network was speci-
fied as the ambient temperature (293 K). The initial pressures were
set to be equal to either that of the low- or high-pressure column. 2.3 | Model verification
The governing equations of the adsorption column, model
assumptions, and equations for calculating component recoveries (Ri) The TR-PSA numerical model was first simplified to a standard DR-
are summarized in the Supporting information. These governing PSA simulation by closing all valves associated with the middle reflux/
GUO ET AL. 5 of 17
F I G U R E 5 Comparison of experimental and simulated CH4 molar

fraction in light and heavy products
product stream, and utilized to simulate DR-PSA experiments

reported by Saleman et al.24 for the separation of CH4/N2 mixtures.
Figure 5 shows the comparison of simulated and experimental prod-
uct compositions at CSS across 15 runs (run 4–18, where the CH4
molar fraction of feed gas was 10.4 mol.%). The RMSDs between the
experimental and predicted values of CH4 molar fraction in the heavy
product (yH,CH4 ) and light product (yL,CH4 ) were as low as 0.018 and
0.004, respectively, indicating the high accuracy of our model.
3 | RESULTS AND DISCUSSION
3.1 | Conventional DR-PSA cycle
The separation of the CH4/N2/He mixture was investigated with a

conventional DR-PSA process using the method described in
F I G U R E 6 Light (A) and heavy (B) product compositions using the
Section 2.3. Seven simulations were conducted with varying FE/PU conventional DR-PSA cycle with constant light reflux ratio RL/F = 2.5,
step durations (tFE/PU) from 30 to 120 s to provide a reference for H/F = 0.2, tPR/BD = 45 s, and ZF = 0.5
evaluating the new TR-PSA cycles. As shown in Figure 6A, the molar
fraction of He in the light product (yL,He) remains relatively constant,
achieving an enrichment ratio of 1.1. The yL,CH4 increases gradually Further insights into this process can be found in Figure 7. For
with the longer tFE/PU due to the gradual breakthrough of CH4. each component, profiles are given for 0, 15, and 30 s elapsed. As
Figure 6B shows that increasing tFE/PU from 30 to 120 s leads to the illustrated in Figure 7A, the axial concentration profiles show high
reduction of yH,CH4 from 0.833 to 0.546 and the increase of both CH4 concentration at the heavy end and high N2 concentration at
yH,N2 and yH,He from 0.252 to 0.421 and 0.058 to 0.084, respectively. the light end. The feed gas enters the adsorption column during
These results suggest that the conventional single-stage DR-PSA pro- the FE step, leading to a discontinuity of composition at Z = 0.5,
cess is challenging to separate ternary mixtures because the medium where the compositions of the feed gas and upstream reflux are
component (nitrogen in this study) can significantly contaminate the different.
light product (i.e., helium) as they are collected together in the same Figure 7B clearly shows the spatial difference for the front of the
product tank, which can be avoided by adding another product stream concentration profiles during the PU step. It can be seen that helium
and tank to separately collect the N2 and He products or be solved by accumulates at the top of the high-pressure column at the start of the
introducing a second stage of DR-PSA process to further separate the PU step, and the concentration decreases sharply as the PU step pro-
N2/He mixtures. ceeds. The CH4 adsorption front reaches the light end of the
6 of 17 GUO ET AL.
F I G U R E 7 Concentration
profiles of gas phase inside the
adsorption column during FE (A)
and PU (B) step for a
conventional DR-PSA cycle with
following operating parameters:
RL/F = 2.5, H/F = 0.2,
tFE/PU = 30 s, tPR/BD = 45 s,
and ZF = 0.5
adsorption column at t = 30 s, and hence, further extension of FE/PU purge at the same location, hence the name triple-reflux (TR) PSA.
step will lead to the breakthrough of CH4. As shown in Figure 7B, the The side-draw outlet was located at a position in the upper part of the
maximum concentration positions of N2 and helium are quite differ- adsorption column to collect a middle product (N2 in this case) from
ent, Z of 0.8 and 0.9, respectively, at time 0 s. Therefore, we speculate the high-pressure column during the FE/PU step, and part of the mid-
that helium and nitrogen could be collected separately if an extra gas dle product was injected into the low-pressure column as the middle
product line is added to the side of the column near the light end. It is reflux stream, that is, the triple-reflux strategy. Parametric studies
also believed that accordingly the purity of N2 and He can be signifi- were conducted to investigate the effects of operating parameters
cantly increased. such as feed step duration (tFE/PU), heavy product to feed ratio (H/F),
light product to feed ratio (L/F), the locations of feed inlet (ZF), middle
product outlet (ZM,out) and middle reflux inlet (ZM,in), the light reflux
3.2 | TR-PSA cycle flow rate (RL), and the middle reflux flow rate (RM) on separation per-
formance. In all cases, the feed flow rate was fixed at 1.25 SLPM
A middle product stream was introduced to the side of the conven- (standard liter per minute) and the PR/BD step time was con-
tional DR-PSA columns to side draw the N2-rich gas and allow for side stant (45 s).
GUO ET AL. 7 of 17
F I G U R E 8 The impact of tFE/PU on the purity of CH4, He, and N2 in the corresponding product streams, feed compositions are: (A) 3% He,
10% CH4, and 87% N2; (B) 5% He, 15% CH4, and 80% N2; (C) 10% He, 20% CH4, and 70% N2. Operating parameters: ZF = 0.5,
ZM,out = ZM,in = 0.9, RL/F = 1, and RM/F = 2
3.2.1 | Effect of feed step duration 42.9%, and from 80% to 91.5%, respectively, for feed gas (b); and,
from 20% to 70.3%, from 10% to 47.9%, and from 70% to 85.9%,
Three ternary mixtures with different concentrations were used as respectively, for feed gas (c).
the feed gas, containing the following compositions: (a) 3% He, 10% Figure 9 shows the axial composition profiles in the adsorption
CH4, and 87% N2; (b) 5% He, 15% CH4, and 80% N2, and (c) 10% He, column at the end of each stage for tFE/PU = 30 s case and 90 s case
20% CH4, and 70% N2. The effect of tFE/PU was studied by varying with the feed compositions being 10% He, 20% CH4, and 70% N2. As
the FE/PU step duration between 30 and 90 s. In all cases, the H/F shown in Figure 9A, the molar fraction of CH4 at Z = 0 (the heavy
ratio and L/F ratio were set to the same values as the molar fraction end) in the tFE/PU = 30 s case is much higher than that in the
of methane and helium in the feed gas stream, respectively. The tFE/PU = 90 s case, leading to the higher CH4 purity in the heavy prod-
recovery of each component is equal to the purity in corresponding uct (70.3% vs. 53.5%). The sharp increase of CH4 concentration and
product streams, as described in the Supporting information. dramatic decrease of He concentration at Z = 0.9 were caused by the
The separation performance of the three groups of mixtures injection of the middle reflux stream due to the mixing. At the end of
shows a similar trend, the purities of CH4, He, and N2 in the the PR step, as illustrated in Figure 9B, the He accumulates at Z = 1,
corresponding product streams decrease with the increasing tFE/PU, as that is, the light end, and higher He concentration can be obtained in
illustrated in Figure 8. For three different feed gases, the best purity tFE/PU = 30 s case. The peak of N2 concentration is located at the
results were all achieved at tFE/PU = 30 s, the concentrations of CH4 upper part of the adsorption column, moving from Z = 0.7 to Z = 0.9
in the heavy, He in the light, and N2 in the middle product were during the PU step. The accumulated He is extracted from the high-
enriched from 10% to 62.9%, from 3% to 39.0% and from 87% to pressure column during the following PU step and collected as the
94.6%, respectively, for feed gas (a); from 15% to 67.9%, from 5% to light product. The N2 concentration peaks at the Z values of 0.7 and
8 of 17 GUO ET AL.
F I G U R E 9 Axial composition profiles in the column at the end of FE (A), PR (B), PU (C) and BD (D) steps. The solid line indicates tFE/PU = 30 s
case, and the dashed line indicates tFE/PU = 90 s case. Other operating parameters: ZF = 0.5, ZM,out = ZM,in = 0.9, RL/F = 1, and RM/F = 2.
Annotation of gas streams: (1) feed flow in; (2) mid reflux flow in; (3) light reflux flow in; (4) heavy product flow out; (5) heavy reflux flow in;
(6) mid product flow out; (7) light product flow out and red/blue indicates gas streams entering/leaving the column
0.6 in the 30 and 90 s cases, respectively. Figure 9C indicates the consistent with the reported experimental results.24 The highest tem-
breakthrough of CH4 in the high-pressure column. The adsorption perature variation can be found at the middle of the column (T2),
fronts of CH4 in both cases exceed the middle product outlet position, roughly 9 K and cannot be ignored. The high temperature during the
resulting in the severe loss of CH4. During the BD step, the pressure PU step is due to the methane adsorption front passing through the
in the column decreases, allowing the desorption of CH4. Therefore, middle of the bed, which agrees with the axial concentration profiles
at the end of the BD step, the CH4 concentration in the lower part of illustrated in Figure 9. The decrease in temperature during the FE step
the column is high, especially at the heavy end, as presented in is mainly caused by the desorption of methane. The temperature
Figure 9D. trend at the bottom of the column (T3) is similar to that of T2 but with
Figure 10 indicates the bed temperature profile over a cycle at generally lower values, caused by the alteration of internal flow direc-
CSS calculated at three different nodes, T1 located at the top of the tion at different steps and the thermal effect of adsorption/desorption
adsorption column (next to the void tank 6 in Figure 4), T2 is at the process.24 During the desorption steps (BD and FE, the endothermic
middle of the column, and T3 is at the bottom of the column (next to steps), T3 is at the downstream position and the cold stream flows
void tank 10 in Figure 4). The simulated temperature swing trend is downward; in contrast, during the adsorption steps (PR and PU, the
GUO ET AL. 9 of 17
F I G U R E 1 1 Purities of CH4, He, and N2 in the corresponding

F I G U R E 1 0 Temperature profile over a cycle at CSS, feed
product streams under varying ZF. Operating parameters: ZM,out = 0.9,
compositions are 10% He, 20% CH4, and 70% N2. Operating
ZM,in = 0.9, tFE/PU = 30 s, H/F = 0.2, L/F = 0.1, RL/F = 1, and RM/F = 2
parameters are as follows: ZF = 0.5, ZM,out = ZM,in = 0.9, tFE/PU = 30 s,
RL/F = 1, and RM/F = 2
exothermic steps), T3 is at the upstream location and the hot stream injection of middle reflux stream dominates the effect on axial con-
flows upward, as the flow direction shown in Figure 9. The tempera- centration distribution, making the impact of feed inlet position can
ture at the top of the column (T1) varies to a much less extent than be ignored. However, the ZF determines the left boundary of the mid-
T2 and T3 because this part of column is mainly occupied by nitrogen dle reflux inlet position (ZM,in). Since the N2 concentration of middle
and helium, which have lower enthalpies of adsorption. reflux is always higher than that of the feed flow, the middle reflux
To further enhance the separation performance of the TR-PSA inlet position should be closer to the light end of the column and
process, the aforementioned CH4 breakthrough phenomenon must be above the feed location, as depicted in Figure 2.
avoided and a further parametric study is required. The feed gas con-
taining 10% He, 20% CH4, and 70% N2 was selected to conduct the
following parametric study. 3.2.3 | Effect of middle outlet/inlet position
The impact of the middle product outlet position (ZM,out) on purities of

3.2.2 | Effect of feed inlet position three product streams is displayed in Figure 12. Results indicate that
the ZM,out has a significant impact on the separation performance of
For the TR-PSA and DR-PSA cycles, the feed gas is injected into the the TR-PSA process. When the ZM,out value is decreasing (i.e., the mid-
column at a lateral intermediate position. The concentration of com- dle outlet position moving from the light end to heavy end), the He
ponents in the feed gas stream is generally different from that in the purity of light product increases sharply from 13.2% to a plateau of
upstream reflux flow, resulting in the so-called “mixing problem,” roughly 76.8%. Meanwhile, the CH4 purity of heavy product reduces
which manifests as the discontinuity of axial concentration curves at steadily from 79.2% to 30.3%. The maximum N2 purity (87.1%) is
the injection position. The mixing problem has been reported and obtained at ZM,out = 0.8. This can be explained by analyzing the axial
investigated in the open literature23,42,44,48 and has been confirmed concentration profiles, as shown in Figure 9B, methane occupies the
to have negative impacts on the separation performance of the pro- bottom half part of the column, whereas helium accumulates at the
cess. The effect of feed position (ZF) was investigated by changing the top of the column. As indicated in Figure 9B (at the end of the PR
ZF between 0.4 and 0.9. As shown in Figure 11, when the feed posi- step, namely, the start of the PU step) and Figure 9C (at the end of
tion moves towards the light end, the CH4 product purity rises first, PU step), the peak of nitrogen concentration is between the range of
reaching a maximum of 70.6% at ZF = 0.6, followed by a slight decline. Z = 0.7–0.9 during the PU step (N2 product is extracted from the
The purity of He and N2 product shows a similar trend, except that high-pressure column). Hence, the maximum N2 product purity was
the optimum purity is 48.2% at ZF = 0.6 for helium product and obtained when ZM,out is around at 0.8, the middle position of the range
86.2% at ZF = 0.7 for nitrogen product, respectively. The impact of 0.7–0.9. When the ZM,out moves forward to the heavy end, methane
the ZF on product purity is minimal, within ±0.6% for CH4 purity in will gradually dominate the gas composition, leading to more methane
the heavy product, ±0.2% for He purity in the light product, and escaping with the middle product and ultimately reducing the purity
±0.2% for N2 purity in the middle product. As shown in Figure 9A, the of the heavy product gas due to the constraint of component material
perturbation at Z = 0.9 (the middle reflux inlet position) is apparently balance. Conversely, more helium is lost when the ZM,out approaches
more severe than at 0.5 (the feed inlet position). Therefore, the lateral closer to the light side, revealing a tradeoff between the recovery of
10 of 17 GUO ET AL.
“mixing problem.” When reducing the ZM,in, the middle reflux stream will
be admitted to the column at a position near the heavy end, the impact
of perturbation on heavy product will be predictably more significant and
leading to the decrease in CH4 product purity, in contrast, the increase of
ZM,in will have a more apparent impact on the He product purity. It is wor-
thy to note that the case of ZM,in = 0.5 and 1.0 are two particular condi-
tions, where the middle reflux stream is mixed with feed gas and light
reflux stream in the void tank, respectively, and then flows into the
adsorption column. The results under these operating conditions indicate
the primary factors that affect the separation results are the concentra-
tion and inlet position of the lateral inflow gas, rather than the number of
inflowing gas streams.
F I G U R E 1 2 Purities of CH4, He, and N2 in the corresponding 3.2.4 | Effect of heavy/light product to feed flow
product streams under varying ZM,out. Other operating parameters are ratio
set as: ZF = 0.5, ZM,in = 0.9, tFE/PU = 30 s, H/F = 0.2, L/F = 0.1,
The ratios of the heavy product to feed (H/F) and light product to feed
(L/F) are crucial operating parameters, which determine the theoreti-
cal boundary of heavy and light product purity, described by
Equations S15 and S16 in the Supporting information. The effects of
H/F and L/F on the separation performance were investigated sepa-
rately by varying the molar flow rate of the heavy or light product.
The overall material balance was ensured by modifying the molar flow
rate of middle product stream, which was automatically controlled by
the PID controller. Since the heavy/light product to feed flow ratio
will differ from the molar fraction of CH4/N2 in the feed gas, the
recovery of each component cannot be directly represented by its
purity in the corresponding product stream (as described by
Equations S9–S11). Therefore, the additional recovery results will be
provided in the following discussions.
The arrows in Figure 14 indicate to which target the TR-PSA pro-
cess is adjusted by varying the H/F ratio. In this part of simulation, the
flow rate of light product was constant at 0.125 SLPM
F I G U R E 1 3 Purities of CH4, He, and N2 in the corresponding
product streams under varying ZM,in. Other operating parameters are (i.e., H/F = 0.1). As shown in Figure 14A, raising the H/F ratio from
set as flows: ZF = 0.5, ZM,out = 0.9, tFE/PU = 30 s, H/F = 0.2, L/ 0.1 to 0.4 causes a decrease in the purity of CH4 product from 69.0%
F = 0.1, RL/F = 1, and RM/F = 2 to 48.5%, which is limited by the component material balance of CH4.
The N2 purity in middle product gas rises from 78.7% to 93.0% with
methane and helium. It should be noted that the case of ZM,out = 1.0 the increasing H/F ratio. The H/F ratio has minimal impacts on the
is equivalent to collecting light and middle products together since the helium product purity, leading to only a decrease of 2% in the light
N2 and He are already mixed in the tank (void tank 6 or 7 in Figure 4) product purity when the H/F ratio ascends from 0.1 to 0.4.
before being split into two product streams, hence, the results of this Figure 14B describes that the recovery of CH4 increases dramatically
case are equivalent to those of the conventional DR-PSA cycle. The from 36.9% to 95.5% with the growth in H/F ratio. Increasing the H/F
notable difference between results at ZM,out = 0.9 and 1.0 demon- ratio reduces the N2 recovery from 90.3% to 63.9%, due to the rising
strates the importance of collecting N2 and He separately to enhance N2 concentration in the heavy product gas and the higher heavy prod-
the separation performance of TR-PSA process. uct flow rate, which results in a massive loss of N2. Since the flow rate
Figure 13 shows the purities of CH4, He, and N2 in its product of helium product is fixed here, the variation of recovery should be
streams with different middle reflux inlet positions (ZM,in). When the ZM,in consistent with the purity (refer to Equation S10 in the Supporting
increases (the middle inlet position shifts up to the light end) from 0.5 to information), reducing roughly by 2%. The H/F = 0.3 can be regarded
1.0, the CH4 product purity increases by 6.2%. The He product purity as the optimum condition that both high CH4 purity and recovery can
reduces gradually with an increasing ZM,in. The N2 product purity first be obtained.
ascends slowly with the increasing ZM,in, peaks at ZM,in = 0.8 with a value Figure 15 reflects the impact of L/F ratio on the separation per-
of 86.5%, and then decreases slightly. The impact of middle reflux inlet formance of the TR-PSA process. Here, heavy reflux flow rate was set
position on the separation performance is due to the previous mentioned constant at 0.25 SLPM (i.e., H/F = 0.2). As shown in Figure 15,
GUO ET AL. 11 of 17
F I G U R E 1 4 The product purity (A) and recovery (B) of CH4, He,

and N2 as a function of H/F ratio. The arrows indicate the direction in
F I G U R E 1 5 The product purity (A) and recovery (B) of CH4, He,
which the H/F ratio should be varied to bias the separation
and N2 as a function of L/F ratio. The arrows indicate the direction in
performance favoring the CH4 purity or the CH4 recovery. Operating
which the L/F ratio should be varied to bias the separation
parameters: ZF = 0.5, ZM,in = ZM,out = 0.9, tFE/PU = 30 s, L/F = 0.1, RL/
performance favoring the He purity or the He recovery. Operating
F = 1, and RM/F = 2
parameters: ZF = 0.5, ZM,in = ZM,out = 0.9, tFE/PU = 30 s, H/F = 0.2,
decreasing the L/F ratio contributes to the elevation of He purity in light
product (increase proportionally from 28.6% to 58.5%) at the sacrifice
of He recovery (decrease by 71.2%). When increasing the L/F ratio, the
purity of N2 product ascends from 82.8% to 88.0% and then remains the appropriate H/F and L/F ratios, which can yield relatively high product
stable, whereas the N2 recovery decreases gradually due to the reduc- purity and recovery of both CH4 and He. Additionally, identifying the
tion of middle product flow rate. The influence of L/F ratio on CH4 proper light and middle reflux flow rates, which can concurrently satisfy
purity and recovery can be ignored with a variation of only 2%. the regeneration requirement of the adsorption column and can avoid the
When the L/F ratio increases, more helium is extracted from the severe breakthrough of He at the heavy end, is another effective method.
system to the light product stream, hence, we can observe the slight
decrease of helium concentration in the heavy stream, as listed in
Table 3. However, even for the case of L/F = 0.3, the concentration 3.2.5 | Effect of light and middle reflux to feed flow
of He in the heavy product stream is still a bit high, accounting for ratio
47.7% of the total helium loss. The massive He escaping from the
heavy end is probably due to the relatively high flow rate of reflux The reflux ratio is a crucial operating parameter for the separation
stream from the high-pressure column (light reflux flow rate is 1 SLPM performance of gases using the TR-PSA process. Saleman et al. 24
and middle reflux flow rate is 2.5 SLPM), making the breakthrough of suggested that the main role of the light reflux flow for the DR-
helium be unavoidable. PSA process is to assist the desorption of loaded heavy compo-
Since the adsorption of He in the column is negligible, it is challenging nent from the low-pressure bed, meanwhile, an equivalent amount
to prevent the escape of helium from the heavy end during the FE step of heavy reflux flows into the high-pressure bed due to the inter-
(i.e., the low-pressure column). To minimize the loss of He, we need to find nal flow balance leading to an increase in the usage rate of
12 of 17 GUO ET AL.
T A B L E 3 Product compositions
Heavy product (%) Middle product (%) Light product (%)
obtained during the parametric study of
L/F CH4 He N2 CH4 He N2 CH4 He N2 L/F ratio
0.02 69.9 6.9 23.2 7.6 9.6 82.8 0.1 58.5 41.4
0.05 70.0 6.5 23.6 7.7 8.0 84.3 0.2 54.5 45.3
0.10 70.3 5.6 24.1 8.1 5.8 86.1 0.2 48.3 51.5
0.15 70.8 4.9 24.3 8.8 4.1 87.1 0.3 42.5 57.1
0.20 71.3 4.2 24.5 9.2 2.8 88.0 0.4 37.4 62.2
0.25 71.6 3.8 24.7 9.8 2.0 88.2 0.5 32.7 66.8
0.30 71.9 3.4 24.7 10.5 1.5 88.0 0.6 28.6 70.7
adsorption beds.25 There is generally an optimum light reflux flow

rate ensuring adequate desorption without causing the break-
through in each bed. In the TR-PSA process, the internal flow
cycle contains three streams, that is, the light reflux, middle reflux,
and heavy reflux. The heavy reflux flow rate is highly constrained
by flow rates of light and middle reflux and adjusted directly by a
PID controller. Therefore, here we only separately investigate the
influence of light reflux and middle reflux to process performance.
The light reflux stream is introduced from the light product tank
with the high concentration of He, and the impact of light reflux
ratio (R L/F) is investigated by changing the flow rate specification
of light reflux stream. Figure 16 shows that the helium product purity
rises to the peak (45.2%) at RL/F = 0.2, followed by the gradual decline
when raises the RL/F ratio. The maximum N2 purity is 95.9% at RL/
F = 0.5. The purity of CH4 in the heavy product goes up steadily with
the increasing RL/F ratio. The variation trend of recovery as a function
of the light reflux to feed ratio should be the same as the purity. The
ratio of results between recovery and purity is constant, 1.5 for CH4,
2 for He and roughly 0.714 for N2 (with neglectable fluctuations caused
by the minimal material balance error of simulations), which can be
deduced from Equations S9–S11.
The compositions of three product streams are summarized
in Table 4. The light reflux stream assists the regeneration of the
N2-loaded adsorbent, thus facilitating the separation of N2 and He in
the PU step (in the high-pressure bed), explaining why the growth of
light reflux flow rate leads to the increase in helium purity when the
RL/F is lower than 0.2. However, further raising the RL/F ratio leads to
an elevated helium concentration in the heavy product (from 1.8% to
4.3%), exacerbating the helium fugitive from the heavy end. Although F I G U R E 1 6 The impact of RL/F on the product puritiy (A) and
recovery (B) of CH4, He, and N2. Operating parameters: ZF = 0.5,
increasing the RL/F helps to improve the CH4 purity, its negative
ZM,in = ZM,out = 0.9, H/F = 0.3, L/F = 0.2, and RM/F = 2
impact on light product should also be considered as helium is also an
important target gas in this study. Hence, RL/F = 0.5 (i.e., RL = 0.625
SLPM) is regarded as the optimum value to obtain both relatively high
CH4 purity and He purity and will be used in the subsequent investi- respectively. Increasing the RM/F ratio leads to a gradual reduction in
gation to optimize the middle reflux flow rate. He product purity (from 46.9% to 42.6%) and recovery (from 95.1%
The effect of middle reflux to feed ratio was studied by varying to 85.3%), suggesting that an appropriate RM/F ratio is crucial for
the molar flow rate of middle reflux (RM). Figure 17 illustrates that ensuring both high recovery of methane and helium. RM/F = 0 is a
both the highest product purity and recovery of CH4 (60.6% and particular operating condition, which means that the middle reflux
90.8%, respectively) can be obtained at RM/F ratio of 2.0. The N2 flow rate is zero, that is, a special DR-PSA process with N2 side-drawn
purity and recovery peak at 95.9% and 68.2% when RM/F = 1.5, product stream but without the middle reflux stream. According to the
GUO ET AL. 13 of 17
T A B L E 4 Product compositions
Heavy product (%) Middle product (%) Light product (%)
obtained during the parametric study of
RL/F ratio RL/F CH4 He N2 CH4 He N2 CH4 He N2
0.05 58.9 1.9 39.2 4.3 1.5 94.3 0.8 43.5 55.7
0.20 59.4 1.8 38.8 4.1 0.8 95.1 0.3 45.2 54.5
0.35 60.2 2.2 37.6 3.8 0.8 95.5 0.2 44.9 54.9
0.50 61.4 2.6 36.0 3.2 0.8 95.9 0.1 44.0 55.9
0.75 61.8 3.4 34.7 2.8 1.5 95.7 0.1 41.1 58.8
1.00 62.0 4.3 33.7 2.6 2.7 94.7 0.1 36.9 63.0
F I G U R E 1 8 Specific energy consumption for per mole captured

CH4 and He as a function of RM/F ratio. Constant operating
parameters: ZF = 0.5, ZM,out = 0.9, tFE/PU = 30 s, H/F = 0.3, L/F = 0.2,
and RL/F = 0.5
column. However, in the TR-PSA process, heavy reflux flow rate is

constrained by flow rates of light reflux and middle reflux, there-
fore, the middle reflux stream can buffer the conflict between the
high heavy reflux flow rate demand and the low light reflux flow
rate requirement, balancing between the enrichment of CH4 and
capture of He better than the process without the middle reflux
stream.
The reflux flow rates also have a significant impact on the energy
F I G U R E 1 7 The product purity (A) and recovery (B) of CH4, He, consumption of the compressor (estimated as a single-stage isentropic
and N2 as a function of RM/F ratio. Operating parameters: ZF = 0.5, compressor in this research) in per cycle, as described in the
ZM,in = ZM,out = 0.9, tFE/PU = 30 s, H/F = 0.3, L/F = 0.2, and
Supporting information (Equation S17). The specific energy consump-
RL/F = 0.5
tion for per mole captured CH4 and He shows approximately as a lin-
ear function of the RM/F ratio, as plotted in Figure 18. When the RM/F
preliminary simulations results (Table S4), it is challenging for this process ratio increases from 0.0 to 3.0, both specific energy consumptions are
to balance the capture between CH4 and He due to that the regenera- nearly doubled. When RM/F = 0, there is still a certain amount of com-
tion of adsorbents in low-pressure (FE) column and pushing partial CH4 pressor duty, mainly caused by the energy consumption for the pres-
product gas flowing into the high-pressure (PU) column as the heavy sure reversal between two columns and pressurizing the heavy reflux
reflux stream can only be accomplished by the light reflux. Since the gas (due to the presence of light reflux flow) from low pressure (1 bar)
heavy reflux flow rate is only constrained by the light reflux flow rate, a to high pressure (5 bar). Here, we take the separation performance at
high light reflux flow rate is generally required to satisfy the threshold RM/F = 1.0 as the optimum results, where a heavy product containing
heavy reflux flow rate for providing adequate CH4 to be adsorbed 60.0% CH4 with a recovery of 90.4% and specific energy consumption
in the PU column to elevate the degree of CH4 enrichment, which of 0.274 MJ mol1 captured CH4, a light product containing 45.3%
simultaneously leads to the severe breakthrough of He in the FE He with a recovery of 91.3% and specific energy consumption of
14 of 17 GUO ET AL.
0.543 MJ mol1 captured He and a middle product containing 95.8% adsorption columns were the same as the TR-PSA unit, as summa-
N2 can be obtained. rized in Table 2. Figure 19 displays the schematic diagram of
2-stage DR-PSA process, the light product stream obtained from
the first-stage DR-PSA unit was used as the feed gas in the second-
3.2.6 | Comparison between TR-PSA and 2-stage stage DR-PSA unit, where the helium and nitrogen mixture were
DR-PSA processes separated into the He dominant product and N2 dominant product.
The feed flow rate and step duration of 2-stage DR-PSA cycle was
The simulation of the 2-stage DR-PSA process was conducted by set to be the same as the TR-PSA cycle, that is, feed flow rate is
combining two DR-PSA units in series. The parameters of 1.25 SLPM, FE/PU step time is 30 s, and PR/BD step time is 45 s.
As discussed previously, the product to feed flow ratio will deter-
mine the upper limit of product purity, hence, in the simulation of
2-stage DR-PSA process, the flow rates of CH4-dominant and He-
dominant product streams were adjusted to be the same as in the
TR-PSA cycle. Therefore, the main optimization of 2-stage DR-PSA
process was tuning the reflux flow rate. For the first stage, an opti-
mized light reflux flow rate was identified to obtain both high CH4
purity and low He concentration in the heavy product to minimize
the flow of CH4 into the second stage and to avoid excessive He
lost in the first stage. Meanwhile, an optimal reflux flow rate can be
determined to obtain the highest helium purity for the second
stage. Here, only the optimum result of 2-stage DR-PSA cycle was
selected for comparison with the TR-PSA cycle, and all simulation
results of 2-stage DR-PSA process were provided in the Supporting
information with detailed operating conditions. At CSS, the overall
material balance error of 2-stage DR-PSA simulations is around
0.2%, whereas component material balance error is 0.8%, 0.6%, and
0.4% for CH4, He, and N2, respectively.
F I G U R E 1 9 Schematic diagram of the double-stage DR-PSA Figure 20 highlights the achieved performance and specific work
process (two PL-A configurations) of both TR-PSA and 2-stage DR-PSA processes. The 2-stage DR-PSA
F I G U R E 2 0 Comparison of
optimal separation performance
and specific work of TR-PSA and
2-stage DR-PSA processes.
Operating parameters for the TR-
PSA cycle: ZF = 0.5, ZM,out = ZM,
in = 0.9, tFE/PU = 30 s, H/F = 0.3,
L/F = 0.2, RL/F = 0.5, and RM/
F = 1.0. Operating conditions for
the 2-stage DR-PSA cycle: feed
positions are fixed at exact
middle of the adsorption bed for
both stages, light reflux flow rate
is 1 SLPM for first stage and
0.4 SLPM for the second stage,
tFE/PU = 30 s, methane product
flow rate is 0.375 SLPM (i.e., H/
F = 0.3), and helium product flow
rate is 0.25 SLPM (i.e., L/F = 0.2)
GUO ET AL. 15 of 17
process achieved a bit higher CH4 product purity (61.4% vs. 60.0%) Conceptualization (equal); formal analysis (supporting); methodology
and CH4 recovery (92.1% vs. 90.4%). While the TR-PSA process could (supporting); validation (supporting). Gongkui Xiao: Conceptualization
obtain slightly better product purity and recovery of He (45.3% purity (supporting); funding acquisition (equal); investigation (supporting);
with a recovery of 91.3%) than the 2-stage DR-PSA process (purity is software (supporting). Jurgen Hanekom: Methodology (supporting);
44.8% and recovery is 89.7%). Overall, the separation performance of software (supporting); validation (supporting). Guoping Hu: Software
TR-PSA is comparable to that of 2-stage DR-PSA process in terms of (supporting); visualization (supporting); writing – review and editing
the capture of CH4 and He. As for the comparison of compressor (supporting). Paul A. Webley: Supervision (supporting); writing –
duty, TR-PSA process only requires a cycle work of 49.6 kJ mol1 review and editing (supporting). Eric F. May: Conceptualization
(feed), which is significantly (30%) lower than the specific work of the (supporting); funding acquisition (equal); software (supporting). Gang
2-stage DR-PSA process (64.3 kJ mol1 (feed)) due to the use of one Kevin Li: Conceptualization (equal); funding acquisition (equal); super-
compressor in TR-PSA vs. two in the 2-stage DR-PSA system. There- vision (lead); writing – review and editing (lead).
fore, for the conditions considered in this work, TR-PSA shows
remarkable advantages compared to 2-stage DR-PSA in terms of capi- DATA AVAILABILITY STAT EMEN T
tal investment and energy consumption and guarantees a comparable The data that support the findings of this study are available from the
separation performance. corresponding author upon reasonable request.
NOTATIONS
4 | C O N CL U S I O N Cpai specific heat capacity of the adsorbed gas
Cps specific heat capacity of the adsorbent
In this work, we have developed and demonstrated a novel TR-PSA Cpw specific heat capacity of the wall
process to obtain three high-purity products from ternary mixtures Cvg specific gas phase heat capacity at constant volume
within a single stage. A nonisothermal, nonadiabatic numerical model ci molar fraction of component i
of the TR-PSA cycle was developed using Aspen Adsorption to inves- DB internal diameter of column
tigate its performance for separating the mixtures of He, CH4, and N2. F standard volumetric flow rate of feed
Parametric studies were performed to investigate the effect of con- H standard volumetric flow rate of heavy product
centration of the feed gas, feed step duration (tFE/PU), feed inlet posi- H/F heavy product to feed flow rate ratio
tion (ZF), middle inlet/outlet position (ZM,out and ZM,in), heavy/light hb heat-transfer coefficient between column and ambient
product to feed ratio (H/F and L/F), and the reflux flow rate to feed hgs heat-transfer coefficient between gas and solid
ratio (RL/F and RM/F) on the separation performance. Results hw heat-transfer coefficient between gas and wall
showed that the TR-PSA process could effectively separate ternary IPXi isotherm parameter X for component i
mixtures into three high-purity products by a single-stage PSA pro- ki constant mass-transfer coefficient of component i
cess. For the separation of a ternary mixture containing 10% He, kg heat conductivity of gas
20% CH4, and 70% N2, we can obtain a heavy product with 60.0% ks heat conductivity of adsorbent
purity and 90.4% recovery of CH4, a light product containing 45.3% kw heat conductivity of wall
He with a recovery of 91.3%, and a middle product of 95.8% N2 L standard volumetric flow rate of light product
with a recovery of 68.9%, with the specific work of 49.6 kJ mol1 L/F light product to feed flow rate ratio
(feed). The TR-PSA process can deliver comparable separation per- M standard volumetric flow rate of middle product
formance of the 2-stage DR-PSA process with significantly lower mads the mass of the adsorbent
capital and energy costs. Based on the concept of this study, the MW molecular weight of gas mixture
design of “multireflux” might be further extended to accomplish ni molar of component i
the separation of gas mixtures with more than three components P gas pressure
as long as the concentration front of each component is sharp PH high pressure
enough. PL low pressure
pi partial pressure of component i
ACKNOWLEDGMENTS Qi adsorbent loading of component i per unit mass of adsorbent
This project is funded by the Australia Research Council Qi adsorbent loading of component i per unit mass of adsor-
(DP190100983). Y. Guo would like to thank Juncheng Yu for provid- bent in equilibrium with its partial pressure in gas phase
ing research facilities to run part of simulations in this study. R universal gas constant
r radial coordinate of the adsorbent
AUTHOR CONTRIBUTIONS RL standard volumetric flow rate of light reflux
Yalou Guo: Conceptualization (equal); formal analysis (lead); investiga- RM standard volumetric flow rate of middle reflux
tion (lead); methodology (lead); visualization (lead); writing – original rp particle radius
draft (lead); writing – review and editing (equal). Vincent Jusko: tFE/PU feed/purge step duration
16 of 17 GUO ET AL.
tPR/BD pressurization/blowdown step duration 7. Häussinger P, Glatthaar R, Rhode W, et al. Noble gases. Ullmann's
T gas temperature Encyclopedia of Industrial Chemistry. Wiley-VCH Weinheim; 2001.
8. Cavenati S, Grande CA, Rodrigues AE. Separation of CH4/CO2/N2
Tenv environmental temperature
mixtures by layered pressure swing adsorption for upgrade of natural
Ts adsorbent temperature gas. Chem Eng Sci. 2006;61(12):3893-3906.
Tw column wall temperature 9. Cavenati S, Grande CA, Rodrigues AE. Layered pressure swing
Vi the volume of component i adsorption for methane recovery from CH4/CO2/N2 streams. Adsorp-
tion. 2005;11(1):549-554.
vg superficial gas velocity
10. Hu G, Zhao Q, Tao L, Xiao P, Webley PA, Li KG. Enrichment of low
Wr wall thickness grade CH4 from N2/CH4 mixtures using vacuum swing adsorption
x axial distance coordinate with activated carbon. Chem Eng Sci. 2021;229:116152.
yF,i the concentration of component i in the feed gas 11. Qu D, Yang Y, Qian Z, et al. Enrichment of low-grade methane gas
from nitrogen mixture by VPSA with CO2 displacement process:
yH,i the concentration of component i in the heavy product
modeling and experiment. Chem Eng J. 2020;380:122509.
yM,i the concentration of component i in the middle product 12. Ishibashi M, Ota H, Akutsu N, et al. Technology for removing carbon
yL,i the concentration of component i in the light product dioxide from power plant flue gas by the physical adsorption method.
Z dimensionless axial position along the bed Energy Convers Manage. 1996;37(6–8):929-933.
13. Xiao G, Xiao P, Hoadley A, Webley P. Integrated adsorption and
ZF feed inlet position
absorption process for post-combustion CO2 capture. Front Chem Sci
ZM,in middle reflux inlet position
Eng. 2020;15:483-492.
ZM,out middle product outlet position 14. Diagne D, Goto M, Hirose T. New PSA process with intermediate
αp specific particle surface area per unit length of bed feed inlet position operated with dual refluxes – application to
ΔHi enthalpy of adsorption for component i carbon-dioxide removal and enrichment. J Chem Eng Jpn. 1994;27(1):
85-89.
ΔM overall material balance error
15. Hirose T. A simple design method of a new PSA process consisting of
ΔMi material balance error for component i both rectifying and stripping sections. Paper presented at: Proceed-
εi bed voidage ings of the 2nd China–Japan–USA Symposium on Adsorption; 1991.
εp particle voidage 16. Leavitt FW. Duplex adsorption process: US patent 5,085,674. 1992.
17. Kearns DT, Webley PA. Modelling and evaluation of dual-reflux pres-
μg viscosity of gas mixture
sure swing adsorption cycles: part I. Mathematical models. Chem Eng
ρb bulk density of adsorbent Sci. 2006;61(22):7223-7233.
ρg gas molar density 18. Diagne D, Goto M, Hirose T. Experimental study of simultaneous
ρw column wall density removal and concentration of CO2 by an improved pressure swing
adsorption process. Energy Convers Manage. 1995;36(6–9):431-434.
ψ particle shape factor
19. Diagne D, Goto M, Hirose T. Parametric studies on CO2 separation
and recovery by a dual reflux PSA process consisting of both rectify-
ORCID ing and stripping sections. Ind Eng Chem Res. 1995;34(9):3083-3089.
Yalou Guo https://orcid.org/0000-0003-4052-9681 20. Tian C, Fu Q, Ding Z, Han Z, Zhang D. Experiment and simulation
study of a dual-reflux pressure swing adsorption process for separat-
Vincent Jusko https://orcid.org/0000-0002-1193-6694
ing N2/O2. Sep Purif Technol. 2017;189:54-65.
Gongkui Xiao https://orcid.org/0000-0001-9659-6969 21. Wang Y, An Y, Ding Z, Shen Y, Tang Z, Zhang D. Integrated VPSA
Jurgen Hanekom https://orcid.org/0000-0002-1219-3183 processes for air separation based on dual reflux configuration. Ind
Guoping Hu https://orcid.org/0000-0002-7480-8411 Eng Chem Res. 2019;58(16):6562-6575.
22. Mc Intyre JA, Holland CE, Ritter JA. High enrichment and recovery of
Paul A. Webley https://orcid.org/0000-0003-3598-3767
dilute hydrocarbons by dual-reflux pressure-swing adsorption. Ind
Eric F. May https://orcid.org/0000-0001-5472-6921
Eng Chem Res. 2002;41(14):3499-3504.
Gang Kevin Li https://orcid.org/0000-0003-1700-297X 23. McIntyre JA, Ebner AD, Ritter JA. Experimental study of a dual reflux
enriching pressure swing adsorption process for concentrating dilute
RE FE R ENC E S feed streams. Ind Eng Chem Res. 2010;49(4):1848-1858.
24. Saleman TL, Li G, Rufford TE, et al. Capture of low grade methane
1. Xiong L, Peng N, Liu L, Gong L. Helium extraction and nitrogen removal
from nitrogen gas using dual-reflux pressure swing adsorption. Chem
from LNG boil-off gas. IOP Conf Ser Mater Sci Eng. 2017;171:012003.
Eng J. 2015;281:739-748.
2. Rufford TE, Chan KI, Huang SH, May EF. A review of conventional 25. May EF, Zhang Y, Saleman TLH, Xiao G, Li G, Young BR. Demonstra-
and emerging process technologies for the recovery of helium from tion and optimisation of the four dual-reflux pressure swing adsorp-
natural gas. Adsorpt Sci Technol. 2014;32(1):49-72. tion configurations. Sep Purif Technol. 2017;177:161-175.
3. Boreham CJ, Edwards DS, Poreda RJ, et al. Helium in the Australian 26. Weh R, Xiao G, Ashraful Islam M, May EF. Upgrading sub-quality natural
liquefied natural gas economy. APPEA J. 2018;58(1):209. gas by dual reflux-pressure swing adsorption using activated carbon and
4. Oliver BM, Bradley JG, Farrar H. Helium concentration in the Earth's ionic liquidic zeolite. Chem Eng J. 2019;392:123753.
lower atmosphere. Geochim Cosmochim Acta. 1984;48(9):1759-1767. 27. Xiao G, Saleman TL, Zou Y, Li G, May EF. Nitrogen rejection from
5. Mukhopadhyay M. Helium sources and recovery processes – with methane using dual-reflux pressure swing adsorption with a
special reference to India. Cryogenics. 1980;20(5):244-246. kinetically-selective adsorbent. Chem Eng J. 2019;372:1038-1046.
6. Rufford TE, Smart S, Watson GCY, et al. The removal of CO2 and N2 28. Bhatt TS, Sliepcevich A, Storti G, Rota R. Experimental and modeling
from natural gas: a review of conventional and emerging process analysis of dual-reflux pressure swing adsorption process. Ind Eng
technologies. J Petrol Sci Eng. 2012;94–95:123-154. Chem Res. 2014;53(34):13448-13458.
GUO ET AL. 17 of 17
29. Li D, Zhou Y, Shen Y, et al. Experiment and simulation for separating 41. Zou Y, Xiao G, Li G, Lu W, May EF. Advanced non-isothermal
CO2/N2 by dual-reflux pressure swing adsorption process. Chem Eng dynamic simulations of dual reflux pressure swing adsorption cycles.
J. 2016;297:315-324. Chem Eng Res Des. 2017;126:76-88.
30. Shen Y, Zhou Y, Li D, Fu Q, Zhang D, Na P. Dual-reflux pressure 42. Lu W, Xiao G, Weh R, May EF. Rejecting N2 from natural gas by dual
swing adsorption process for carbon dioxide capture from dry flue reflux pressure swing adsorption with activated carbon. J Nat Gas Sci
gas. Int J Greenhouse Gas Control. 2017;65:55-64. Eng. 2020;81:103457.
31. Wawrzyn czak D, Majchrzak-Kucęba I, Srokosz K, et al. The pilot dual- 43. Weh R, Xiao G, Islam MA, May EF. Nitrogen rejection from natural
reflux vacuum pressure swing adsorption unit for CO2 capture from gas by dual reflux-pressure swing adsorption using activated carbon
flue gas. Sep Purif Technol. 2019;209:560-570. and ionic liquidic zeolite. Sep Purif Technol. 2020;235:116215.
32. Takamura Y. Evaluation of dual-bed pressure swing adsorption for 44. Guo Y, Hu G, Chen K, Guo J, Webley PA, Li GK. Capture of dilute
CO2 recovery from boiler exhaust gas. Sep Purif Technol. 2001;24(3): methane with a novel dynamic-feed dual-reflux pressure swing
519-528. adsorption process. AIChE J. 2022;68(1):e17390.
33. Ebner AD, Ritter JA. Equilibrium theory analysis of dual reflux PSA for 45. Knaebel KS, Hill FB. Pressure swing adsorption: development of an
separation of a binary mixture. AIChE J. 2004;50(10):2418-2429. equilibrium theory for gas separations. Chem Eng Sci. 1985;40(12):
34. Kearns DT, Webley PA. Modelling and evaluation of dual reflux pres- 2351-2360.
sure swing adsorption cycles: part II. Productivity and energy con- 46. Malbrunot P, Vidal D, Vermesse J, Chahine R, Bose TK. Adsorbent
sumption. Chem Eng Sci. 2006;61(22):7234-7239. helium density measurement and its effect on adsorption isotherms
35. Bhatt TS, Storti G, Rota R. Optimal design of dual-reflux pressure at high pressure. Langmuir. 1997;13(3):539-544.
swing adsorption units via equilibrium theory. Chem Eng Sci. 2013; 47. Aspen Tech. Aspen Adsorption V10 Help; 2017.
102:42-55. 48. Diagne D, Goto M, Hirose T. Numerical analysis of a dual refluxed PSA
36. Bhatt TS, Storti G, Denayer JFM, Rota R. Equilibrium theory-based process during simultaneous removal and concentration of carbon diox-
assessment of dual-reflux pressure swing adsorption cycles that uti- ide dilute gas from air. J Chem Technol Biotechnol. 1996;65(1):29-38.
lize light gas for pressure swing. Ind Eng Chem Res. 2018;58(1):
350-365.
37. Rossi E, Storti G, Rota R. Optimal design procedure of dual-reflux
SUPPORTING INF ORMATION
pressure swing adsorption units for nonlinear separations. Ind Eng
Chem Res. 2019;58(16):6644-6652.
Additional supporting information may be found in the online version
38. Bhatt TS, Storti G, Denayer JFM, Rota R. Optimal design of dual- of the article at the publisher's website.
reflux pressure swing adsorption units via equilibrium theory: process
configurations employing heavy gas for pressure swing. Chem Eng J.
2017;311:385-406. How to cite this article: Guo Y, Jusko V, Xiao G, et al.
39. Bhatt TS, Storti G, Rota R. Detailed simulation of dual-reflux pressure Separation of He/N2/CH4 ternary mixtures by a triple-reflux
swing adsorption process. Chem Eng Sci. 2015;122:34-52.
pressure swing adsorption process. AIChE J. 2022;e17569.
40. Zhang Y, Saleman TLH, Li G, Xiao G, Young BR, May EF. Non-
isothermal numerical simulations of dual reflux pressure swing adsorp- doi:10.1002/aic.17569
tion cycles for separating N2 + CH4. Chem Eng J. 2016;292:366-381.
Chapter 4
Separation of He/N2/CH4 ternary mixtures by a triple-reflux
Yalou Guo,a Vincent Jusko,b Gongkui Xiao,b Jurgen Hanekom,b Guoping
Hu,a Paul A. Webley,a Eric F. May,b Gang (Kevin) Li*a

a
Department of Chemical Engineering, The University of Melbourne, Parkville, VIC 3010,
Australia
b
Fluid Science and Resources research division, Department of Chemical Engineering, The
University of Western Australia, Crawley, WA 6009, Australia
*Corresponding author: li.g@unimelb.edu.au
100
Chapter 4
Valves control
Valves were used to control the internal- (reflux streams) and external flow (product and feed
streams) of the TR-PSA cycle. In Aspen Adsorption valves could be specified as one of the
four idealized statuses based on the specific operation of each step as follows: 0, fully closed,
which blocks all flow; 1, fully open, which allows unrestricted flow with no pressure drop; 2,
constant CV, where the flowrate is determined by the pressure drop between the upstream flow
and downstream flow; 3, constant molar flowrate, which is used to specify a constant molar
flowrate independent of pressure behaviour. The detailed valve logic for the TR-PSA cycle is
indicated in Fig. S.1 and Table S.1.
101
Chapter 4
102
Chapter 4
Fig. S.1 Process flow diagram of the complete TR-PSA cycle. Valves are shaded black/white
to indicate whether they are closed/opened during each step. The arrow indicates the flow
direction of each stream.
103
Chapter 4
Table S. 1: Valve logic for the TR-PSA cycle. P indicates that a PID controller directly
controls the valve.
Step VF VF1 VF2 VH VH1 VH2 VHR1 VHR2 VL
FE 3 1 0 3 1 0 0 2 3
PR 0 0 0 0 0 1 2 0 0
PU 3 0 1 3 0 1 2 0 3
BD 0 0 0 0 1 0 0 2 0
Step VL1 VL2 VLP VLR VLR1 VLR2 VM VM1 VM2
FE 0 1 2 3 1 0 P 3 1
PR 0 0 0 0 0 0 0 0 0
PU 1 0 2 3 0 1 P 1 3
BD 0 0 0 0 0 0 0 0 0
104
Chapter 4
Void tanks
Two kinds of adsorbent-free tanks were used to represent the volume of piping manifolds and
product tanks, respectively, as listed in Table S. 2.
Table S. 2: Volumes of tanks used in the simulation.
Tank No. Volume (cc)
1 10
2 200
3 10
4 200
5 5
6 5
7 5
8 200
9 10
10 10
11 10
105
Chapter 4
Adsorption bed model
The rigorous non-isothermal mathematical model was constructed on the commercial software
platform, i.e., Aspen Adsorption, for detailed simulation of the separation performance of TR-
PSA processes under various operating conditions. The model assumptions can be summarised
as follows:
a. Ideal gas behaviour.
b. Momentum balance and pressure drop are estimated via the Ergun equation.
c. Material balance in the gas phase is described as convection only (plug flow without
considering axial dispersion).
d. The linear driving force (LDF) model with constant mass transfer coefficients (MTC)
describes the flux between adsorbed and gaseous molecules.
e. Extended Langmuir II isotherms were used to describe the competitive adsorption as
determined by the partial pressure of each component in the gas phase.
f. Non-isothermal operation.
The governing equations of the adsorption column are listed in Table S.3.
106
Chapter 4
Table S. 3: Governing equations for the adsorption column in this model.
Momentum ∂P  150 × 10−5 µ g (1 − ε i ) 2 1.75 × 10−5 M W ρ g (1 − ε i ) 2 

Ergun equation = ± v + vg  Eq. (S.1)
∂z  (2rpψ ) 2 ε i3
g
2rpψε i3 
balance  
∂ (vg ci ) ∂ci ∂w
Material balance Plug flow + [ε i + (1 − ε i )ε p ] + ρb i =0 Eq. (S.2)
∂x ∂t ∂t
LDF model with ∂Qi

Kinetic model = ki (Qi* − Qi ) Eq. (S.3)
∂t
constant MTC
Extended  IP 
IP1,i ⋅ exp  2,i  pi
*
Q =  T 
Isotherm model Langmuir II i Eq. (S.4)
  IP  
1 + ∑ i  IP3,i ⋅ exp  4,i  pi 
isotherm   T  
∂ 2T ∂T ∂T ∂vg
−k gε i 2
+ Cvg vg ρ g + [ε i + (1 − ε i )ε p ]Cvg ρ g +P
∂x ∂x ∂t ∂x
Gas phase Eq. (S.5)
4hw
+ hgs a p (T − TS ) + (T − TW ) = 0
DB
∂ 2Ts ∂T ∂T
+ C ps ρb s + ρb ∑ i =1 ( C pai wi ) s
n
−ks 2
∂x ∂t ∂t
Energy balance Solid phase Eq. (S.6)
n  ∂wi 
+ ρb ∑ i =1  ∆H i  − hgs a p (T − Ts ) =
0
 ∂t 
∂ 2Tw ∂T 4 DB
( DB + Wr ) − DB2
2
∂x 2 ∂t
Column wall Eq. (S.7)
4 ( DB + Wr )
2
0
( DB + Wr )
2 2
− DB
Ideal gas law PVi = ni RT Eq. (S.8)
107
Chapter 4
Nodes number of column
The impact of column nodes number has an apparent effect on the purity of heavy product,
whilst the influence on product purity of helium and methane is minimal, consistent with
the reported results by Zhang et al.1. The heavy product purity rises slightly with
increasing discretization, reaching an asymptotic value starting from 80 nodes. Here, 100
nodes were selected to ensure the calculation accuracy.
Fig. S.2 The impact of column nodes number on the purity of CH4, He and N2 in the
corresponding product streams. The feed gas contains 10% helium, 20% methane and
70% nitrogen. Operating parameters are specified as follows: ZF = 0.5, ZM,out = ZM,in = 0.9,
tFE/PU =30 s, L/F=0.1, H/F=0.2, RL/F = 1 and RM /F = 2.
108
Chapter 4
Recovery
In this study, the recovery of each component was calculated with the following equations:
RCH4 = (
H ⋅ yH ,CH4 ) (F ⋅ y F ,CH 4 ) Eq. (S.9)
( L ⋅ yL,He ) ( F ⋅ yF ,He )
RHe = Eq. (S.10)
RN2 = (
M ⋅ yM ,N2 ) (F ⋅ y )F ,N 2 Eq. (S.11)
According to Eq. (S.9 – 11), in the TR-PSA process, when the product to feed flow ratio
is set the same as the value of its concentration in feed mixture (i.e., H/F=yF,CH4 and
L/F=yF,He), the recovery of each gas component will be directly equal to the purity in the
corresponding product stream.
109
Chapter 4
Product purity boundary
The methane and helium product purity bound can be deduced based on the material
balance relations, as follows:
F = H +L+M Eq. (S.12)
F ⋅ yF ,CH4 = H ⋅ yH ,CH4 + L ⋅ yL ,CH4 + M ⋅ yM ,CH4 Eq. (S.13)
F ⋅ yF ,He = H ⋅ yH ,He + L ⋅ yL ,He + M ⋅ yM ,He Eq. (S.14)
Here, the theoretical upper purity limits of heavy and light products can be determined by
assuming that all of the methane and helium in the feed gas flows into the heavy and light
product streams, respectively. Hence, the heavy/light product purity bound will be
directly adjusted by the heavy/light product to feed flow ratio, as described in Eq.15 and
16.
 H 
≤ 1, if F ≤ yF ,CH4 
yH ,CH4   Eq. (S.15)
≤ F ⋅ yF ,CH4 , if H ≥ y 
 H F
F ,CH 4

 L 
≤ 1, if F ≤ yF ,He 
yL ,He   Eq. (S.16)
≤ F ⋅ yF ,He , if H ≥ y 
 
F ,He
L F
110
Chapter 4
Energy consumption
γ
 
γ  P  γ −1
E ( RTin )   out  − 1 Eq. (S.17)
η ( γ − 1)   pin  
 
In this equation, E is the compressor’s energy consumption per cycle, γ is the specific
heat ratio of the gas, Tin is the inlet gas temperature, η is the compressor efficiency (set to
0.75 in this simulation), Pin and Pout are the transient pressures in the tank before and after
the compressor, respectively.
111
Chapter 4
N2 side-drawn DR-PSA process
Results of the process with N2 lateral extraction stream but without middle reflux stream
under varying light reflux to feed flow ratios are listed in Table S.4.
Table S.4 Purity and recovery of CH4, He and N2 obtained by nitrogen side-drawn DR-
PSA process under varying RL/F ratios.
RL/F Purity (%) Recovery (%)
CH4 He N2 CH4 He N2
0.20 50.7 47.7 89.9 76.5 95.4 61.6
0.50 55.7 46.9 93.9 83.6 93.8 67.5
0.75 58.8 45.2 95.1 88.2 90.4 67.6
1.00 60.1 41.1 94.8 90.1 82.2 67.3
112
Chapter 4
2-stage DR-PSA process
The compositions of heavy and light products and compressor duty for the first stage DR-
PSA process are summarized in Table S.5. For the first stage, the heavy product is the
eventual methane product obtained from the 2-stage DR-PSA process, whilst the light
product represents the compositions of feed gas for the second stage. Results indicate
that increasing the light reflux flowrate raise the CH4 purity in heavy product at the
expense of losing more He. Here, RL=1 SLPM was regarded as the optimum light reflux
flowrate, balancing between the CH4 purity and He purity with a relatively lower energy
requirement. In addition, this light gas was used as feed gas and injected into the
adsorption column of the second stage DR-PSA unit to further achieve the separation of
helium and nitrogen.
Table S.5 Products obtained with varying RL from the first stage of the 2-stage DR-PSA
cycle. Operating parameters: ZF = 0.5, tFE/PU =30 s, F =1.250 SLPM, H =0.375 SLPM.
RL Heavy Product [%] Light Product [%] E1 [kJ mol-1
(SLPM) CH4 He N2 CH4 He N2 (feed)]
0.50 55.8 1.4 42.9 4.3 13.7 81.9 35.6
0.75 59.3 1.5 39.1 2.8 13.7 83.5 38.4
1.00 61.4 1.7 36.9 1.9 13.6 84.5 41.3
1.25 62.8 2.0 35.1 1.3 13.5 85.2 44.1
1.50 63.7 2.3 34.0 1.1 13.4 85.6 46.2
2.00 64.0 2.6 33.4 0.9 13.2 85.9 48.9
2.25 64.1 3.0 32.9 0.8 13.0 86.2 51.5
2.50 64.0 3.3 32.7 0.7 12.9 86.4 53.8
113
Chapter 4
The separation performance of helium and nitrogen by the second stage DR-PSA unit are
listed below in Table S.6. The heavy product obtained from this stage is equivalent to the
middle product from the TR-PSA cycle. The optimum results can be determined at RL=0.2
SLPM, where He product purity is 44.8% with a He recovery of 89.7% (recovery is
calculated based on the amount of helium flowing into the whole system). Since two
compressors were working in parallel for the 2-stage DR-PSA process, an addition
column was added in Table S.5 to calculate the sum of work duty of two compressors.
The results at RL=0.2 SLPM was selected to compare with the TR-PSA cycle.
Table S.6 Helium and nitrogen products obtained under the varying RL from the second
stage of the 2-stage DR-PSA cycle. Operating parameters: ZF = 0.5, tFE/PU =30 s, F=0.875
SLPM, H=0.625 SLPM.
E2 [kJ mol-1 Esum [kJ mol-1

RL Heavy Product [%] Light Product [%]
(feed)] (feed)]
(SLPM)
CH4 He N2 CH4 He N2
0.20 2.6 1.2 96.2 0.0 44.8 55.2 23.0 64.3
0.40 2.5 1.3 96.1 0.0 44.6 55.4 25.3 66.5
0.60 2.7 1.6 95.8 0.0 43.8 56.2 27.5 68.8
0.80 2.9 2.0 95.1 0.0 43.2 56.8 29.8 71.1
1.00 2.6 2.4 94.9 0.0 41.7 58.3 31.6 72.9
114
Chapter 4
Reference
1. Zhang Y, Saleman TLH, Li G, Xiao G, Young BR, May EF. Non-isothermal numerical
simulations of dual reflux pressure swing adsorption cycles for separating N2+CH4.
Chemical Engineering Journal. 2016; 292: 366-381.
115
Chapter 5
Chapter 5: Enrichment of low-grade methane by

dual reflux vacuum swing adsorption
The fugitive CH4 emission from low-grade CH4 sources, such as low-concentration
coalbed methane, landfill gas and biogas, is one of key contributors to the global warming.
However, it is challenging to upgrade low-grade CH4 due to the strict safety regulation of
transportation, enrichment and application of CH4 gas falling into the concentration range
of 5–30%. In this chapter, the dual reflux (DR) strategy is combined with vacuum swing
adsorption (VSA) as DR-VSA cycles to enrich 5–20% CH4 gas using two adsorbents:
activated carbon (AC) and ionic liquidic zeolites (ILZ). A new optimization method
integrating a dual-convergency iteration algorithm with a multiplicative assessment score
is also demonstrated in this chapter to effectively optimize the operating parameters. At
the optimal conditions, the DR-VSA cycle packed with ILZ adsorbents, which meets the
town gas requirement. The CH4 recovery and specific energy consumption is 95.5% and
180.8 kJ/(mol CH4 captured), respectively.
This chapter was published as a paper in Separation and Purification Technology.
116
Separation and Purification Technology 301 (2022) 121907
Contents lists available at ScienceDirect
Separation and Purification Technology

journal homepage: www.elsevier.com/locate/seppur
Enrichment of low-grade methane by dual reflux vacuum swing adsorption

Yalou Guo a, Xuewei Gu a, Guoping Hu b, *, Paul A. Webley c, Gang Kevin Li a, *
a
Department of Chemical Engineering, the University of Melbourne, Parkville, Victoria 3010, Australia
b
Ganjiang Innovation Academy, Chinese Academy of Sciences, Ganzhou 341119, China
c
Department of Chemical Engineering, Monash University, Clayton, Victoria 3800, Australia
A R T I C L E I N F O A B S T R A C T
Keywords: Methane (CH4) has been recognized as a clean energy resource and crucial chemical feedstock. However, low-
Methane enrichment grade CH4 gases (<30%) such as coal mining gas are often released directly to the atmosphere due to limited
Adsorption economic values and technical challenges in enrichment, transportation and utilization, resulting in significant
Activated carbon
greenhouse gas (GHG) emissions and energy waste. In this work, a non-isothermal model of the dual reflux
Ionic liquidic zeolites
Low-grade methane
vacuum swing adsorption (DR-VSA) was developed on the Aspen Adsorption platform to numerically study the
enrichment of low-grade CH4 from N2 gas by activated carbon (AC) and ionic liquidic zeolites (ILZ). Since the
low-grade methane source was often at atmospheric pressure, feed gas was introduced to a high-pressure column
(1 bar) and evacuation was operated in a low-pressure column (0.2 bar) by a vacuum pump, avoiding the uti
lization of compressors. Two cycle configurations with pressure reversal at the heavy end (type A) and light end
(type B) were demonstrated and investigated. The parametric study of DR-VSA cycles was achieved by varying
the operating variables such as the light reflux flow rate, theoretical heavy product purity and step duration. A
new optimization approach was proposed and demonstrated using a dual convergence algorithm to iteratively
vary operating conditions and a multiplicative score function to evaluate separation performances. Results
indicated that the A-type cycle achieved better separation performance than the B-type cycle, and ILZ showed a
significant superiority to the AC. In the case of 20% CH4 feed gas, the A-type process using ILZ adsorbents
achieved a CH4 purity, recovery, and energy consumption of 80.2 mol.%, 95.5%, and 180.8 kJ/(mol CH4
captured), respectively. The enrichment of low-grade CH4 via the DR-VSA cycle is energy-sustainable and can
generate high-quality fuel gas and GHG mitigation benefits.
1. Introduction Coal bed methane (CBM) is a typical unconventional natural gas,

mainly stored in subsurface coal seams. CBM is often pre-extracted to
Natural gas is a cleaner fossil fuel with 50% less CO2 and no par ensure the safe operation of coal mines. Medium and high concentration
ticulate emissions compared with petroleum and coal. It has been widely CBM (CH4 concentration ≥30 vol%) can be utilized for power genera
used for industrial applications and power generation, with a record tion or heating. However, the utilization of low concentration CBM
high share of 24.2% in the primary energy supply market in 2019 [1]. falling into the methane explosive limit (5–15 vol%) is challenging and
However, methane (CH4), the primary component in natural gas, has often forbidden by safety regulations. The typical minimum safety
been proven as the second largest anthropological greenhouse gas requirement of CH4 concentration for transportation and utilization is
(GHG) with 28 times higher global warming potential than carbon di often twice of the upper explosive limit (i.e., 30 vol%) [6]. Gas sources
oxide (CO2) on a centennial time scale [2]. CH4 currently accounts for with a CH4 concentration lower than this limit are often released directly
approximately 16–20% of GHG emissions [2–4]. There has been an to the atmosphere, causing significant GHG emissions and energy loss.
increasing interest in recovering unconventional natural gas with low The coal mining industry contributes to around 8.1% of the global non-
CH4 concentrations from various scenarios, such as biogas, landfill gas, CO2 emissions in 2015, estimated by the United States environmental
coal bed methane and nitrogen-rich vent gas from liquefied natural gas protection agency (U.S. EPA) [7]. Kholod et al. [8] reported a new model
(LNG) facilities to reduce GHG emissions and gain extra energy benefits to estimate the global methane emissions from the coal mining process
simultaneously [5]. and found that the fugitive CH4 emissions estimated in their study were
* Corresponding authors at: Ganjiang Innovation Academy, Chinese Academy of Sciences, Ganzhou 341119, China (G. Hu).
E-mail addresses: gphu@gia.cas.cn (G. Hu), li.g@unimelb.edu.au (G.K. Li).
https://doi.org/10.1016/j.seppur.2022.121907
Received 5 July 2022; Received in revised form 3 August 2022; Accepted 7 August 2022
Available online 19 August 2022
1383-5866/© 2022 Elsevier B.V. All rights reserved.
Y. Guo et al. Separation and Purification Technology 301 (2022) 121907
Fig. 1. Schematic diagram of DR-PSA configurations: feed to high- (PH) or low-pressure (PL) column and pressure inversion with the heavy (i.e., the more
adsorbable, A) or light (i.e., the less adsorbable, B) component.
73% higher than the data from U.S. EPA (103 billion cubic meters (bcm) DR-PSA apparatus is demonstrated by Diagne [23,24] for concentrating
versus 59.5 bcm in 2010), revealing the underestimation of fugitive CH4 CO2 from the air, extensive experimental studies have been reported for
emissions. Therefore, curbing CH4 emissions from the CBM is critical to various binary mixtures separations, such as air separation [25], carbon
achieving carbon neutrality and stabilizing the climate. dioxide capture [26,27], hydrocarbon enrichment [19,28–30] and vol
Nitrogen is a major constituent of fugitive CH4 gases. The separation atile organic compounds (VOCs) removal [31], indicating its potential to
of CH4 and N2 is particularly challenging due to their similar physical produce two streams of high purity products simultaneously. Simulation
and chemical properties, such as kinetic diameters (3.8 Å for CH4 and research, ranging from simplified analytical models to rigorous numer
3.64 Å for N2) [9]. Cryogenic distillation is only economically viable in ical models, provides deep insights into the DR-PSA process. Analytical
large-scale processing due to its high capital investment requirement models are based on equilibrium theories with assumptions of
and energy intensity [10]. Membranes show unsatisfactory separation isothermal conditions, no mass transport resistance, complete separa
performance in this application, owing to the pressure limitation and tion and no pressure drop [32–35], limiting their application and ac
similar kinetic diameters of CH4 and N2. Pressure swing adsorption curacy. Bhatt et al. [30] reported an isothermal numerical model to
(PSA) has been widely investigated as an alternative technique for N2 predict the separation of CH4 and N2 with the DR-PSA process. The
rejection from high concentration CH4 mixtures [11–14]. However, the predicted and experimental results showed similar trends under
N2 concentration in feed mixtures is limited to less than 50%, which is different operating parameters with an average percentage error be
significantly lower than the N2 fraction in low concentration CBM gases. tween − 2–8%. Zhang et al. [36] developed a non-isothermal DR-PSA
Enrichment of dilute CH4 (molar fraction ≤20%) from N2 mixture by model for CH4/N2 mixtures separation, which was benchmarked against
PSA technologies with various novel concepts has been reported, such as experiments reported by Saleman et al. [19]. The root mean square
the vacuum pressure swing adsorption (VPSA) process using molecular deviations (RMSD) between the simulated and measured CH4 concen
sieves as adsorbents to enrich 10% CH4 up to 80 mol.%[15]; the vacuum trations were 0.3% and 2.4% for the light and heavy products, respec
swing adsorption (VSA) process with multiple pressure equalization tively, only 3 times larger than the uncertainty of experiments. May et al.
(PE) steps to concentrate 17% CH4 to 42 mol.% with a recovery of [37] further demonstrated the non-isothermal model of four basic DR-
roughly 80% using activated carbon as adsorbents [16]; the VPSA cycle PSA configurations and validated it using experimental results. The
with CO2 displacement process to enrich 10% CH4 up to 75 mol.% with simulated CH4 concentrations in both light and heavy products agreed
89% recovery using granular activated carbon [17]; a mobile skid- well with experimental values. Some innovative configurations were
mounted prototype VPSA system designed for site trials to enrich low- also proposed based on the DR-PSA process with the aim of obtaining
grade CH4 from 16 to 30% to 43.3–70.7 mol.% using TUTJ-1 and acti better separation performance or specific separation targets, such as
vated carbon [18]; the dual reflux pressure swing adsorption (DR-PSA) modified DR-PSA [38–40], dynamic feed DR-PSA [41], triple reflux PSA
process to enrich 10% CH4 to 52 mol.% using activated carbon as ad [42] and 2-stage DR-PSA cycles [43,44]. However, the reported DR-PSA
sorbents, with a recovery of 87% and specific energy consumption of configurations are not capable of low-grade CBM enrichment due to
191 kJ/(mol CH4 captured) [19]. safety concerns. Ning et al. reported a proportional dependence between
The DR-PSA (also known as duplex PSA) process, proposed by Leavitt the CBM explosion limit range of CH4 and pressure [45], suggesting that
and Hirose [20,21], has a lateral feed injection and two reflux streams compressing low-grade CBM gas could aggravate the explosion risk.
and can be deployed to four basic configurations [22], namely PH-A, PL- The optimization of PSA processes is inherently complicated and
A, PH-B and PL-B, as schematically shown in Fig. 1. Since the seminal computationally intensive due to different cycle designs and integrated
2
Table 1 concentration CH4 at atmospheric feed gas pressures and avoid the
Langmuir isotherm parameters of AC and ILZ. utilization of compressors. The feed gas (at atmospheric pressure, i.e.,
Parameter Component AC ILZ 1.01 bar) is introduced into the high-pressure column (1 bar) while
evacuating is operated in the low-pressure column by a vacuum pump
Qmax [mol/kg] CH4 5.53 3.34
N2 4.17 2.85 (0.2 bar). This study is conducted using a non-isothermal numerical
b0 [bar − 1
] CH4 1.19 × 10-4 3.44 × 10-5 model implemented on the Aspen Adsorption platform with 5–20%
N2 1.75 × 10-4 4.45 × 10-5 methane in a simulated CBM feed. We evaluate the separation perfor
-ΔH [kJ/mol] CH4 17.29 21.03 mance of the DR-VSA processes using two types of adsorbents: 1) acti
N2 14.30 16.54
vated carbon (AC), and 2) ionic liquidic zeolites (ILZ) [50]. Parametric
studies are conducted to investigate the impacts of step duration, heavy
tuning work of a series of operating parameters. The trade-offs in product to feed flow ratio and light reflux flowrate on the separation
adsorption performance, especially between purity and recovery, make performance and energy consumption. A new optimization method
the optimization of PSA cycles more complex [46]. Previous studies using a dual convergence iteration algorithm is proposed and demon
[19,29,37,47–49] report some key operating parameters which can strated to effectively reduce the computation efforts and quickly
significantly affect the separation performance of DR-PSA processes, e. accomplish the optimization of DR-VSA cycles. The multiplicative score
g., external flows (flowrates of feed and product gases), internal flows evaluation approach is introduced to visualize the process optimization
(reflux flowrate), step time, feed concentration, pressure ratio, etc. and determine the optimal operating conditions to achieve the best
Nevertheless, how to effectively iterate these decision variables to ach balance of trade-off between purity and recovery.
ieve quick and precise optimization and appropriately balance the trade-
offs of separation performance has not been discussed yet.
Here, we combine the VSA process with the dual reflux configura
tion, namely, the DR-VSA process, to carry out the enrichment of low-
Fig. 2. Process diagram of the DR-VSA model (during FE/PU step). Black color indicates the valve is closed; white color indicates the valve is open. Vacuum pump is
used in this study to replace the function of the compressor used in previously reported DR-PSA processes.
3
Table 2 ( )
− ΔH
Adsorption bed and operational parameters. Qmax,CH4 ⋅b0,CH4 ⋅exp R⋅TCH4
Selectivity = ( ) (4)
Parameter Value Unit − ΔH
Qmax,N2 ⋅b0,N2 ⋅exp R⋅TN2
Column details
Height 1 m
Internal diameter 0.035 m 2.2. Model layout and details
Wall thickness 0.0016 m
Wall specific heat capacity 0.5 kJ⋅kg− 1⋅K− 1
The numerical model was developed on Aspen Adsorption, as
Wall-ambient heat transfer coefficient 10 W⋅m− 2⋅K− 1
Wall thermal conductivity 16 W⋅m− 1⋅K− 1 depicted in Fig. 2. Each of the two adsorption beds was segmented
Operational parameters equally into two sections to satisfy the intermediate feed inlet position of
Cycle low pressure 0.2 bar DR-VSA processes. Three boundary streams were specified as “non-
Cycle high pressure 1 bar
reversible flow” with fixed pressure, representing the feed gas stream
Ambient temperature 293 K
Feed flowrate 1.25 SLPM
(1.01 bar), heavy product stream (1.00 bar) and light product stream
FE/PU step time 30–90 s (1.00 bar), respectively. Four void tanks were introduced to buffer the
PR/BD step time 60/45 s pressure and composition variations. Tanks A and B were also used as
Adsorbent parameters AC ILZ product tanks to store CH4-rich and N2-rich gas products, respectively.
Inter-particle voidage 0.42 0.62 m3/m3
Two proportional-integral-derivative controllers (i.e., PID 1 and 2 in
Intra-particle voidage 0.65 0.37 m3/m3
Bulk density 435 700 kg⋅m3 Fig. 2) were applied to maintain the high and low pressure of the system
Particle radius 0.0015 0.0015 m by tuning the molar flowrate of the light product stream and the inlet
Specific heat capacity 1.00 0.96 kJ⋅kg− 1⋅K− 1 flowrate of the bottom vacuum pump according to pressure variables in
Thermal conductivity 0.8 0.8 W⋅m− 1⋅K− 1
tank B and C, respectively, during the FE/PU steps. PID 1 ensured a
Mass transfer coefficient AC ILZ
CH4 mass transfer coefficient 1 10 s− 1 constant volumetric flowrate of the bottom pump (solid line) for pres
N2 mass transfer coefficient 3 5 s− 1 sure reversal by transferring CH4-rich gas at the heavy end during PR/
BD step (A-type cycle). The top pump (dashed line) was only applied to
accomplish the pressure inversion (PR/BD step) using N2-rich gas in the
2. Simulation method B-type process. The bottom pump was used to evacuate the low-pressure
column under the control of PID 1 with a variable volumetric flowrate
2.1. Adsorption isotherm parameters during the FE/PU steps and was set to a constant volumetric flowrate
during the PR/BD steps in the A-type cycles. During the FE/PU steps, gas
The adsorption isotherms of pure methane and nitrogen on both at the outflow side of bottom pump was pressurized to 1 bar and divided
activated carbon (AC) and ionic liquidic zeolites (ILZ) are described by into two streams: the heavy product stream (collected as CH4 product
the Langmuir isotherm equation (Eq.1) with the same parameters re gas) and heavy reflux stream which was injected into the high-pressure
ported by Zhang et al. [36] and Li et al. [50], respectively. column to accomplish the heavy purge. Partial outflow gas from the top
( ΔH )
Qmax ⋅b0 ⋅p⋅exp −R⋅T side of high-pressure column was depressurized by the control valve and
Q= ( − ΔH) (1) was admitted into the low-pressure column as light reflux stream while
1 + b0 ⋅p⋅exp R⋅T
the other part of gas was collected as N2 product gas.
Here, Q is the adsorption capacity, Qmax is the saturated adsorption The assumptions and governing equations of the rigorous non-
amount, b0 is the pre-exponential factor, ΔH is the adsorption enthalpy, isothermal numerical model were the same as our previous work
p is the pressure, R is the universal gas constant and T is the [41,42] and are summarized below. Details were also provided in the
temperature. supplementary information. The governing equations were solved using
The regression parameters Qmax, b0 and ΔH of each component are the Quadratic Upwind Scheme (QDS) discretization approach, a high-
applied in the Extended Langmuir isotherm equation (Eq. (2)) to order differential scheme based on a three-point upstream weighted
describe the competitive adsorption of components in binary mixtures, quadratic interpolation, with 50 grid nodes per section (half adsorption
summarized in Table 1. bed). The cyclic steady state (CSS) was deemed to be achieved when the
( ΔHi ) relative deviation in the total loading and temperature between two
Qmax,i ⋅b0,i ⋅pi ⋅exp − R⋅T
Qi = ∑ ( ( )) (2) sequential cycles was lower than the CSS test tolerance (10− 5). The DR-
VSA process simulation usually reached CSS after around 300 cycles
− ΔH
1 + j b0,j ⋅pj ⋅exp R⋅T j
(54,000 s integrator time), taking approximately 4 h for computation in
Here, Qi is the equilibrium adsorption capacity of component i in the real time with 8 simulations running in parallel on an AMD 3900X 12-
gas mixtures, Qmax,i is the saturated adsorption amount of component i; core processor. The acceleration factor, defined as the ratio of inte
b0,i is the pre-exponential factor of component i; pi is the partial pressure grator time and real time, of this model was roughly 4.
of component i in the gas mixture, ΔHi is the adsorption enthalpy of The overall (ΔM) and component (ΔmCH4) relative material errors
component i. can be calculated with Eq. (5) and Eq. (6), respectively.
The adsorption capacity of CH4 and N2 on AC is higher than on ILZ.
The CH4/N2 selectivity can be described as Eq. (3), where yN2 and yCH4 ΔM = (F − H − L)/F (5)
are molar fractions of N2 and CH4 in the gas mixtures; QCH4 and QN2 are
equilibrium adsorption quantity of CH4 and N2 calculated using the ΔmCH4 = (yF ⋅F − yH ⋅H − yL ⋅L)/(yF ⋅F) (6)
Extended Langmuir isotherm equation. Combing Eqs. (2) and (3) can Here, F, H and L are the molar flowrates of the feed, heavy product
simplify the equation as Eq. (4). At 20 ◦ C, the calculated CH4/N2 and light product gas, respectively; yF, yH and yL are methane concen
selectivity of ILZ is 1.8 times higher than that of AC (5.7 versus 3.1). trations in the feed, heavy product and light product flow. At CSS, the
QCH4 yN average ΔM observed in this simulation for AC and ILZ are 0.5% and
Selectivity = × 2 (3) 0.5%, whereas the average ΔmCH4 are 0.5% and 0.2%, respectively.
Q N2 yCH4
The simulation parameters are summarized in Table 2. The initial
composition throughout the pressure-flow network is set to be equal to
the feed gas. The initial temperature is specified as the ambient tem
perature (293 K). The initial pressure is set to be equal to either that of
4
Fig. 3. Flowchart of the optimization approach, T1 (0.0002) is the threshold value to determine whether the optimization can end the iteration loop and move to the
next process.
Fig. 4. Impacts of theoretical heavy product purity (yH′ ) on the separation performance of A-type configuration using ILZ (full symbols) and AC (hollow symbols)
under various feed CH4 concentrations at tF = 45 s and RL = 1 SLPM: a) product purity, b) CH4 recovery.
the low- (0.2 bar) or high-pressure (1 bar) column. The PR/BD step time 2.3. Optimization method
is 60 s and 45 s for A-type and B-type cycles, respectively.
The target of this work is enriching low-grade CH4 (5–20%) to meet
the general national statutory minimum concentration limits for civil
application and avoid massive CH4 fugitive emissions. Therefore, both
5
Fig. 5. Impacts of theoretical heavy product purity (yH′ ) on the separation performance of B-type configuration using ILZ (full symbols) and AC (hollow symbols)
under various feed CH4 concentrations at tF = 45 s and RL = 1 SLPM: a) product purity, b) CH4 recovery.
light product flows and the internal material cycle including light reflux
and heavy reflux streams. During the feed/purge step, for external cir
culation, the feed gas (inlet flow) flowrate is equal to the sum of flow
rates of heavy and light products (outlet flow) due to the material
balance constraints, so that the freedom of outlet flow is 1, i.e., only one
of the heavy or light flowrate can be chosen to be adjustable for para
metric studies and subsequent optimization. Analogous to the external
circulation, the flowrates of heavy reflux and light reflux streams are
coupled, so the degree of freedom of internal material cycle is also 1. The
feed flowrate is usually coupled with the feed step time, indicating the
amount of gas flowing into the DR-VSA system per cycle. Therefore, the
throughput of DR-VSA cycle can be varied by altering either feed
flowrate or feed step time. Herein, we choose feed step time (tF), theo
retical heavy product purity (yH′ , based on heavy to feed product ratio)
and light reflux flowrate (RL) as optimization parameters that have been
proven to significantly impact the separation performance of DR-PSA
processes, as detailed below:
1) yH′ variation: the theoretical heavy product purity can be calculated

using yH′ = F•yF/H. yH′ determines the upper limit of heavy product
purity according to the material balance when its value is ≤1. It
Fig. 6. CH4 recovery versus heavy product purity (yH) caused by varying yH′
should be noted that when yH′ > 1, it has no practical meaning but
from 0.3 to 2.0 under feed CH4 concentration of 5%, tF = 45 s and RL = 1 SLPM
can still be regarded as a crucial indicator, suggesting whether the
for A-type (solid line) and B-type (dash line) configurations using ILZ (full
symbols) and AC (hollow symbols). separation performance is biased towards obtaining either higher
CH4 purity in the heavy product (increasing yH′ ) or higher CH4 re
covery (reducing yH′ ).
product CH4 purity and recovery are critical criteria in determining the
2) tF variation: there is commonly an optimum tF value to achieve
optimal results and beneficial attributes, i.e., more is better.
maximum product purity and recovery. Generally, increasing the tF
The separation results are evaluated via a multiplicative score
can provide higher productivity and lower specific energy con
function where the beneficial criteria (more is better, marked as i) are
sumption. The lowest feed time is set as 30 s to allow sufficient time
multiplied together and divided by the cost criteria (more is worse,
for the pressure and flowrate fluctuations (commonly occur after the
marked as j), as shown in Eq. (7). Here, a is the performance value of
pressure reversal step) to be buffered.
each criterion, w denotes the relative weight of each criterion and σ is
3) RL variation: RL impacts the separation performance. Optimal light
the overall assessment score from a set of indicators. The purity and
reflux flowrate can satisfy adequate desorption of CH4 in the low-
recovery of CH4, both as beneficial criteria, are selected as decision-
pressure column and high utilization rate of the adsorption column
making indicators for the process optimization with the same relative
while avoiding breakthrough. The increase of RL usually leads to a
weight (50%/50%). The optimized separation performance of DR-VSA
higher evacuation energy consumption.
cycles is obtained at a maximum σ value where the results of purity
and recovery are both at high values.
The flowchart of optimization approach is demonstrated in Fig. 3. An
n n
σ = Π awi i / Π aj j
w
(7) initial run is conducted at the start of the optimization process to obtain
i=1 j=1 initial separation results and σ0 for reference. Each variation contains
There are two flow circulations in the DR-VSA system, namely the three values, in the case of feed/purge time (tF), including the initial
external material balance consisting of feed, heavy product as well as input value (tF,0) and tF,0 ± ΔtF. The iteration of three parameter
6
Fig. 7. Effects of feed step time (tF) on the separation performance of A-type configuration using ILZ (full symbols) and AC (hollow symbols) under various feed CH4
concentrations at yH′ =0.8 and RL = 1 SLPM: a) heavy product purity, b) CH4 recovery, c) productivity, and d) specific energy consumption.
variations (3 × 3 × 3) generates 27 operating conditions for simulation 2.4. Recovery, productivity and energy consumption
to obtain separation results, which are evaluated via the multiplicative
score function. The optimal separation results among these 27 runs and The CH4 recovery is the percentage of CH4 being collected in the
their corresponding operating parameters are determined when the heavy product stream, calculated from the CH4 amount in feed gas and
maximum value of σ is achieved, marked as σ max. These optimal results that collected in light gas, as shown in Eq. (8), The detailed analysis
are then used to overwrite the initial values for the next round of opti about why using light gas flow instead of heavy stream was provided in
mization. This loop can only be ended when σ max – σ 0 = 0 (namely, the the Supplementary Information.
optimized operating conditions under the current step size) to start the
F⋅yF − L⋅yL
second optimization loop. In the second loop, the step size is halved to Recovery(R) = (8)
F⋅yF
obtain more precise iterations. The optimization process is regarded as
converged once σmax – σ0 ≤ T1 (arbitrarily set to 0.0002 in this work). The productivity is a crucial parameter for assessing the capital and
The first loop is for rough optimization, which aims to reduce compu operating costs of the DR-PSA process as it indicates the outputs per unit
tational efforts, while the second loop is for precise optimization. The volume of the adsorption column, which can be determined with Eq. (9),
optimal separation results are collected at the end of this optimization where Vbed represents the volume of an identical adsorption bed, and tP
process. This optimization method can also be used for optimizing other is the PR/BD step time.
PSA processes by choosing different decision-making indicators and
yH ⋅H⋅tF
setting relative weights according to specific requirements of the sepa pro.CH4 = (9)
2⋅Vbed ⋅(tF + tP )
ration performance. The assessment score can be utilized for economic
and energy optimization when including other beneficial criteria such as The major energy consumption of the DR-VSA process is the vacuum
productivity, etc. and more cost criteria, e.g., energy consumption, pump duty, which is calculated using a single-stage isentropic vacuum
adsorbent cost and so on. pump integrated in the model (Eq. (10)) [51], where E denotes the
pump’s energy consumption per cycle, N is the throughput of the system
in each cycle, γ is the specific heat ratio of the gas, Tin is the inlet gas
temperature, Pin and Pout are the transient pressures in the tank before
7
Fig. 8. Effects of feed step time (tF) on the separation performance of B-type configuration using ILZ (full symbols) and AC (hollow symbols) under various feed CH4
concentrations at yH′ = 0.8 and RL = 1 SLPM: a) heavy product purity, b) CH4 recovery, c) productivity, and d) specific energy consumption.
and after the compressor, respectively. The specific energy consumption maximal with increasing yH′ . The yH obtained using ILZ is higher than
can be given with Eq. (11). that with AC, and the difference enlarges as yH′ increases. The product
(( )γ ) purity of ILZ and AC is 81.5 mol.% and 63.7 mol.% for the 20% CH4 feed
E = N⋅
γ
(RTin )
Pout γ− 1
− 1 (10) gas when yH′ = 1.0, while for the 5% CH4 feed gas, the difference of
γ− 1 Pin purity results between ILZ and AC is larger (72.5 mol.% versus 41.7 mol.
%).
Sp.E =
E
(11) The recovery ascends gradually with an increasing yH′ for both ad
H⋅yH ⋅2⋅tF sorbents, as displayed in Fig. 4 (b). The difference of recovery results is
more distinguishable between two adsorbents when feed CH4 concen
3. Results and discussion tration is lower. For the 5% CH4 feed case, the recovery value achieved
by ILZ and AC is 72.4% and 42.0% at yH′ = 1.0, respectively.
3.1. Effects of theoretical heavy product purity The superior performance of ILZ over AC is due to the higher CH4/N2
selectivity of ILZ over AC and the higher packing density of ILZ over AC,
The theoretical heavy product purity (yH′ ) significantly determines contributing to more packing material in the adsorption column. When
the maximum heavy product purity that can be achieved according to the yH′ is lower than 0.9, the recovery of CH4 is higher than 95% and
the material balance of CH4. The yH bound, product purity and recovery relatively stable, while decreasing yH′ from 0.9 to 0.8 only leads to a
as a function of yH′ with various feed concentrations are shown in Fig. 4 decrease of 3 mol.% in CH4 purity. Therefore, 0.8–0.9 can be regarded as
for A-type cycles and Fig. 5 for B-type cycles. the most effective range for optimizing parameter yH′ when aiming to
Fig. 4 (a) indicates that the CH4 purity gradually increases to a obtain both high purity and recovery.
plateau with the increase of yH′ for both adsorbents, which is similar to Fig. 5 displays the impacts of yH′ on the separation performance of B-
the yH bound trend. According to the definition of yH′ , the upper purity type cycles. Results indicate that it is challenging for the B-type
boundary for CH4 in heavy product gas equals to the value of yH′ for yH′ configuration to enrich the low-grade CH4 to high-purity product while
≤ 1, whereas, when yH′ > 1, the CH4 purity bound slowly reaches its ensuring a high recovery by using ILZ or AC adsorbents. In general, ILZ
8
Fig. 9. The influence of light reflux flowrate (RL) on the performance of A-type configuration using ILZ (full symbols) and AC (hollow symbols) under various feed
CH4 concentrations at tF = 30 s and yH′ = 0.8: a) product purity, b) CH4 recovery, c) specific energy consumption.
still performs better than AC, with a difference of approximately 8 mol. results can be achieved. By introducing the overall assessment score, σ,
% in purity and about 10% in recovery. the evaluation of separation performance is more distinct. For instance,
Results show that the A-type cycles can obtain better separation AILZ configuration can achieve the optimal results of 64.1 mol.% purity
performance than B-type cycles while using ILZ or AC as adsorbents. For and 91.8% recovery at σ = 0.767 under the investigated conditions.
the B-type configuration, the gas at light end (containing more N2) is
used to complete the pressure inversion between two adsorption col 3.2. Effects of feed step time
umns. Hence, nitrogen occupies the majority part of adsorption columns
and reduces the overall CH4 concentration along the column compared The impacts of feed step time (tF) on the separation performance,
with A-type cycles. Weh et al. [48] demonstrated the axial composition such as product purity (yH), recovery, productivity and specific energy
profiles of A- and B-type cycles during each step and found that using consumption (Sp.E) of DR-VSA processes are illustrated in Fig. 7 for A-
light gas to accomplish pressure reversion in B-cycle could not push back type configuration and Fig. 8 for B-type configuration.
enough CH4 from BD-column to PR-column, resulting in less steep CH4 As shown in Fig. 7 (a) and (b), yH and CH4 recovery display a steadily
adsorption front in the column compared to A-cycle. Therefore, the CH4 rising trend when reducing the tF for two adsorbents. Fig. 7 (c) and (d)
mole fraction in the B-cycle adsorption column is usually lower at the indicate that increasing tF leads to stable growth in productivity and a
heavy end and higher at the light end, leading to the less desirable gradual reduction in the specific energy consumption. For A-type cycles,
separation performance. ILZ still shows significant superiority over AC. Taking the yF = 20% feed
The Pareto plot of CH4 recovery versus product purity is drawn to gas as an instance, the energy required is 149.2 kJ per mole CH4
further analyze the separation performances using 5% feed gas as an captured to achieve a purity of 77.5 mol.% with 97.1% CH4 recovery and
example, as shown in Fig. 6. The separation results obtained from the A- the productivity of 27.7 mol/(m3•hr) by ILZ at tF = 30 s; whilst for AC,
cycle using ILZ adsorbents (referred to as AILZ configuration) are the energy consumption is 162.0 kJ per mole CH4 captured to obtain a
apparently better than other configurations in terms of both purity and purity of 62.3 mol.%, a recovery of 77.9% and a productivity of 22.2
recovery. The Pareto plot shows a trade-off relationship between purity mol/(m3•hr) at tF = 30 s.
and recovery when varying the yH′ . Thus, optimizing yH′ is quite chal The impacts of feed time on the separation performance of B-type
lenging among three selected operating parameters, since there is no cycles are illustrated in Fig. 8. Results demonstrate an optimum feed step
intuitive optimal yH′ value where both maximum purity and recovery time (tF = 45 s) to obtain the maximum purity and recovery results when
9
Fig. 10. The influence of light reflux flowrate (RL) on the performance of B-type configuration using ILZ (full symbols) and AC (hollow symbols) under various feed
CH4 concentrations at tF = 30 s and yH′ = 0.8: a) product purity, b) CH4 recovery, c) specific energy consumption.
the feed CH4 concentration is 5% and 10% for the ILZ, whereas, when
the yF = 15% or 20%, the purity and recovery gradually decrease with
the increase of tF. For AC, a decreasing trend in yH and recovery is seen as
tF rises. As exhibited in Fig. 8 (c) and (d), increasing tF also leads to
higher productivity and lower Sp.E values.
As displayed in Fig. 7 and Fig. 8, higher feed concentration leads to
higher product purity, CH4 recovery and productivity, and lower specific
energy consumption. The reason is that increasing the feed CH4 con
centration results in the elevation in CH4 partial pressure and CH4
throughput of the system. As for the recovery, under the same yH′ , higher
CH4 concentration in feed gas results in a larger H/F ratio according to
the determination of parameter yH′ , leading to more gas flowing into the
heavy product tank and less gas collected in the light product tank.
Hence, according to Eq. (8), the recovery rises with increasing feed CH4
concentrations. Since the calculation of productivity and specific energy
consumption are associated with the FE/PU step time, as shown in Eqs.
(9) and (11), increasing tF is beneficial to productivity and Sp. E, namely
leading to the growth in productivity and reduction in Sp. E.
3.3. Effects of light reflux flowrate

Fig. 11. Determination of optimized separation results of 5% CH4 feed gas via
A-type cycle using ILZ. The internal material cycling has a significant impact on the sepa
ration results, especially energy duty. The freedom of internal circula
tion is 1. The RL is regarded as an independent variable and chosen to
conduct the parametric study while RH is dependent variable and
10
Table 3
Optimized operating parameters and separation results.
yF (%) Cycle yH′ RL (SLPM) tF (s) yH (mol.%) R (%) Sp.E (kJ/mol CH4 captured) Pro.CH4 (mol•m− 3•hr-1) σmax
5 AILZ 0.76 1.50 30 70.3 92.8 657.3 6.6 0.808
AAC 0.55 1.75 30 39.9 73.2 710.4 5.1 0.540
BILZ 0.90 1.50 30 64.0 71.2 871.0 6.1 0.675
BAC 1.18 1.25 30 55.3 46.6 1737.2 4.0 0.508
10 AILZ 0.81 1.25 30 75.2 93.2 329.5 13.4 0.837

AAC 0.65 1.75 30 50.5 78.3 361.8 11.0 0.629
BILZ 0.86 1.75 30 64.1 75.1 456.4 12.8 0.694
BAC 1.24 1.25 30 62.1 49.7 830.3 8.6 0.555
15 AILZ 0.86 1.25 30 79.5 92.7 240.2 19.9 0.858

AAC 0.64 1.50 30 54.4 85.4 215.7 18.2 0.682
BILZ 1.02 1.25 30 72.7 71.4 282.3 18.4 0.721
BAC 1.24 1.25 30 65.6 52.3 531.8 13.6 0.586
20 AILZ 0.84 1.25 30 80.2 95.5 180.8 27.6 0.875

AAC 0.68 1.50 30 59.4 87.6 165.4 24.9 0.721
BILZ 1.03 1.25 30 75.1 72.8 212.1 25.1 0.740
BAC 1.24 1.25 30 68.7 54.7 392.1 19.0 0.613
adjusted automatically. The product purity, recovery and specific en where maximum purity and recovery are obtained.
ergy consumption as a function of RL are displayed in Fig. 9 for A-type
cycles and Fig. 10 for B-type cycles.
Fig. 9 shows that the optimum RL can be found for each case to obtain 3.4. Optimized results
maximum purity or recovery. The optimum RL for ILZ (between 1.25 and
1.50 SLPM) is slightly lower than that for the AC (between 1.50 and 1.75 The detailed optimization route of AILZ cycle for upgrading 5% CH4 is
SLPM) due to the lower CH4 adsorption capacity of ILZ compared to AC. displayed in Fig. 11 using the optimization approach shown in Fig. 3.
Purity and recovery obtained by ILZ are higher than that by AC. Fig. 9 (c) Separation results of each optimizing iteration are distributed within
displays a rising trend of Sp.E with RL increasing due to the growth in different rhomboid frames. The optimization of DR-VSA cycles is a
energy duty of the pump to drive more gas transferring from low- process of tuning the frame to be close to the right top corner. The
pressure column to high-pressure column. However, when RL in starting point is corresponding to operating parameters of yH′ = 0.7, tF =
creases from 0.5 to 0.75 SLPM, there is a slight dip of Sp.E for the 45 s and RL = 1 SLPM with an overall assessment score (σ0) of 0.767. The
enrichment of 5% feed gas. According to Eq. (11), the Sp.E is determined parameter variations for the first optimizing iteration can be determined
as a function of pump duty divided by the amount of captured CH4. This by the operating conditions at the start point and the step sizes, i.e., yH′
slight drop of Sp.E is due to the increasing rate of CH4 purity being more variations are 0.6, 0.7 and 0.8; tF variations are 30, 45 and 60 s; RL
significant than the growth rate of pump duty. The energy required is variations are 0.5, 1 and 1.5 SLPM. These parameter variations deter
595, 327, 224 and 171 kJ per mol CH4 captured for upgrading the feed mine 27 simulation cases and always include the starting point to avoid
gas of 5, 10, 15 and 20% to 70.5, 73.9, 75.9 and 77.3 mol.% respectively, the possibility of σ max < σ0. The maximum σ value obtained in the first
at RL = 1.25 SLPM for ILZ, which are lower than the heating value of optimizing iteration is marked as σ1 with a result of 0.782. The operating
CH4. Therefore, the enrichment of low-grade methane is of energy parameters at σ = 0.782 are then set as the new input for determining
interest. the second optimizing iteration cases. After three iterations, solely
As indicated in Fig. 10 (a) and (b), a higher RL value (approximately relocating the optimization frame cannot achieve higher σ values with
between 1.75 and 2.00 SLPM) is commonly required for B-type cycles to the current step size. The frame is then downsized by halving the step
obtain the maximum purity and recovery. The performance of B-type size to obtain more precise iteration and better separation performance.
cycles is inferior to that of A-type cycles, with a reduction of approxi Eventually, the difference of σ values between two subsequent iterations
mately 20 mol.% in the purity and 25% in the recovery for upgrading the (i.e., σ5 and σmax) is lower than the threshold value (0.0002). The dif
feed gas with 20% CH4. The impacts of feed concentration and adsor ference in separation results between the fifth (purity is 70.8 mol.% and
bents on the separation performance of the B-type configuration are less recovery is 92.2%) and final iteration (purity is 70.3 mol.% and recovery
distinguishable when compared with the A-type configuration. Fig. 10 is 92.8%) is minimal. Therefore, the iteration is seen as converged.
(c) shows that the Sp.E values of B-type processes are higher than that of The optimal operating parameters and separation results of A- and B-
A-type processes-. For the enrichment of 5% feed gas, the Sp.E values of type DR-VSA cycles for upgrading low-grade CH4 (with feed concen
B-type cycles are even roughly twice as high as the heating value of CH4, tration between 5 and 20%) from N2 mixtures using ILZ and AC are
losing the energy benefit of enriching the low-grade CH4. summarized in Table 3. The overall separation performance of ILZ is
The internal reflux circulation allows light and heavy reflux streams superior to AC and A-cycles perform better than B-cycles with both
to effectively clean the adsorbent in the low-pressure column and drive higher CH4 product purity and recovery. AILZ is the most promising
the CH4 adsorption front to propagate further in the high-pressure col configuration with the highest assessment score. It is challenging for B-
umn, respectively. When the RL value is low, there is insufficient internal type configurations to energy-sustainably capture 5% CH4 from the N2
gas reflux to regenerate the CH4-occupied adsorbents and facilitate the streams using either the ILZ or AC under the optimal operating condi
CH4 adsorption front to move further towards the light end, resulting in tions. In addition, when packing with same adsorbent materials, the B-
low utilization of the adsorption column. In contrast, high RL values can type setup usually requires higher Sp.E than the A-type setup, mainly
lead to excessive internal flow, thereby causing the breakthrough at due to the relatively inferior recovery results of B-cycles. Decreasing
light and heavy end, which then results in energy and material wastes. feed concentration also elevates the Sp.E values as more gases are
Therefore, an optimal RL should be determined for each configuration evacuated from the columns per cycle which leads to energy
inefficiency.
11
4. Conclusions [4] M. Saunois, R.B. Jackson, P. Bousquet, B. Poulter, J.G. Canadell, The growing role
of methane in anthropogenic climate change, Environ. Res. Lett. 11 (2016),
120207.
This study demonstrated the process design of DR-VSA by combining [5] G. Hu, E.F. May, K.G. Li, We commercialized a methane capture technology in ten
dual reflux policy and VSA process to enrich low-grade CBM (5–20% years—here’s how, Nature 604 (2022) 242–245.
CH4 mixed with N2) at atmospheric feed pressure, which can avoid the [6] E.E. SERIES, Best Practice Guidance for Effective Methane Drainage and Use in
Coal Mines-Second edition, 2016.
possible explosive hazards caused by compressor in DR-PSA configura [7] U.S. EPA, Global Non-CO2 Greenhouse Gas Emission Projections & Mitigation
tions. A new process optimization strategy with a dual convergence al Potential: 2015–2050, United States Environmental Protection Agency,
gorithm to iterate three key operating parameters was provided and the Washington, DC, 2019.
[8] N. Kholod, M. Evans, R.C. Pilcher, V. Roshchanka, F. Ruiz, M. Coté, R. Collings,
multiplicative score σ was introduced for intuitively determining the Global methane emissions from coal mining to continue growing even with
optimal separation performance. Two types of configurations (i.e., A- declining coal production, J. Cleaner Prod. 256 (2020), 120489.
and B- type) with two adsorbents (Norit RB3 activated carbon and ionic [9] M. Tagliabue, D. Farrusseng, S. Valencia, S. Aguado, U. Ravon, C. Rizzo, A. Corma,
C. Mirodatos, Natural gas treating by selective adsorption: Material science and
liquidic zeolites) were numerically investigated via a non-isothermal chemical engineering interplay, Chem. Eng. J. 155 (2009) 553–566.
model constructed on the Aspen Adsorption platform. Parametric [10] A. Jayaraman, A.J. Hernandez-Maldonado, R.T. Yang, D. Chinn, C.L. Munson, D.
studies were conducted to investigate the impact of operating parame H. Mohr, Clinoptilolites for nitrogen/methane separation, Chem. Eng. Sci. 59
(2004) 2407–2417.
ters, such as feed step time, theoretical heavy product purity and light [11] S. Effendy, C. Xu, S. Farooq, Optimization of a Pressure Swing Adsorption Process
reflux flowrate on the separation performance. Ionic liquidic zeolites for Nitrogen Rejection from Natural Gas, Ind. Eng. Chem. Res. 56 (18) (2017)
showed superiority to the activated carbon in terms of purity, recovery, 5417–5431.
[12] A.I. Fatehi, K.F. Loughlin, M.M. Hassan, Separation of methane—nitrogen mixtures
productivity and specific energy consumption, especially while using A-
by pressure swing adsorption using a carbon molecular sieve, Gas Sep. Purif. 9 (3)
type configurations. At the optimized conditions, A-type cycles with ILZ (1995) 199–204.
were found to present the best separation performance, which can [13] J.A. Delgado, M.A. Uguina, J.L. Sotelo, B. Ruíz, Modelling of the fixed-bed
enrich the feed gas of 5, 10, 15 and 20% methane up to 70.3, 75.2, 79.5 adsorption of methane/nitrogen mixtures on silicalite pellets, Sep. Purif. Technol.
50 (2) (2006) 192–203.
and 80.2% with a recovery of 92.8, 93.2, 92.7 and 95.5% by consuming [14] J.A. Delgado, M.A. Uguina, J.L. Sotelo, V.I. Águeda, P. Gómez, Numerical
657.3, 329.5, 240.2 and 180.8 kJ energy for per mole captured CH4, simulation of a three-bed PSA cycle for the methane/nitrogen separation with
respectively. The energy analysis suggested that the recovery of low- silicalite, Sep. Purif. Technol. 77 (1) (2011) 7–17.
[15] B.o. Yang, E.L. Xu, M. Li, Purification of coal mine methane on carbon molecular
grade methane by the DR-VSA process using ILZ is energy-sustainable sieve by vacuum pressure swing adsorption, Sep. Sci. Technol. 51 (6) (2016)
with the benefits of GHG emissions reduction. 909–916.
[16] G. Hu, Q. Zhao, L. Tao, P. Xiao, P.A. Webley, K.G. Li, Enrichment of low grade CH4
from N2/CH4 mixtures using vacuum swing adsorption with activated carbon,
CRediT authorship contribution statement Chem. Eng. Sci. 229 (2021), 116152.
[17] D. Qu, Y. Yang, Z. Qian, P. Li, J. Yu, A.M. Ribeiro, A.E. Rodrigues, Enrichment of
low-grade methane gas from nitrogen mixture by VPSA with CO2 displacement
Yalou Guo: Conceptualization, Methodology, Validation, Formal process: Modeling and experiment, Chem. Eng. J. 380 (2020) 122509.
analysis, Writing – original draft, Investigation, Data curation. Xuewei [18] H. Shang, H. Bai, X. Li, J. Li, J. Yang, Site trials of methane capture from low-
concentration coalbed methane drainage wells using a mobile skid-mounted
Gu: Data curation, Software, Visualization. Guoping Hu: Formal anal vacuum pressure swing adsorption system, Sep. Purif. Technol. 295 (2022),
ysis, Visualization, Writing – review & editing. Paul A. Webley: Writing 121271.
– review & editing, Supervision, Conceptualization. Gang Kevin Li: [19] T.L. Saleman, G. Li, T.E. Rufford, P.L. Stanwix, K.I. Chan, S.H. Huang, E.F. May,
Capture of low grade methane from nitrogen gas using dual-reflux pressure swing
Conceptualization, Supervision, Funding acquisition, Project adminis
adsorption, Chem. Eng. J. 281 (2015) 739–748.
tration, Writing – review & editing. [20] T. Hirose, A simple design method of a new PSA process consisting of both
rectifying and stripping sections, Proceedings of the 2nd China-Japan-USA
Symposium on Adsorption, vol. 123, 1991.
Declaration of Competing Interest [21] F.W. Leavitt, Duplex adsorption process, US Patent 5,085,674, 1992.
[22] D.T. Kearns, P.A. Webley, Modelling and evaluation of dual-reflux pressure swing
adsorption cycles: Part I. Mathematical models, Chem. Eng. Sci. 61 (2006)
The authors declare that they have no known competing financial 7223–7233.
interests or personal relationships that could have appeared to influence [23] D. Diagne, M. Goto, T. Hirose, New PSA Process with Intermediate Feed Inlet
the work reported in this paper. Position Operated with Dual Refluxes - Application to Carbon-Dioxide Removal
and Enrichment, J. Chem. Eng. Jpn. 27 (1994) 85–89.
[24] D. Diagne, M. Goto, T. Hirose, Experimental-Study of Simultaneous Removal and
Data availability Concentration of CO2 by an Improved Pressure Swing Adsorption Process, Energ
Convers, Manage 36 (1995) 431–434.
[25] C. Tian, Q. Fu, Z. Ding, Z. Han, D. Zhang, Experiment and simulation study of a
Data will be made available on request. dual-reflux pressure swing adsorption process for separating N2/O2, Sep. Purif.
Technol. 189 (2017) 54–65.
[26] D. Li, Y. Zhou, Y. Shen, W. Sun, Q. Fu, H. Yan, D. Zhang, Experiment and
Acknowledgements
simulation for separating CO2/N2 by dual-reflux pressure swing adsorption
process, Chem. Eng. J. 297 (2016) 315–324.
This project is funded by the Australia Research Council [27] Y. Shen, Y. Zhou, D. Li, Q. Fu, D. Zhang, P. Na, Dual-reflux pressure swing
(DP190100983). Y. Guo would like to thank Juncheng Yu for offering adsorption process for carbon dioxide capture from dry flue gas, Int. J. Greenhouse
Gas Control 65 (2017) 55–64.
facilities to run part of simulations in this study. [28] J.A. Mc Intyre, C.E. Holland, J.A. Ritter, High Enrichment and Recovery of Dilute
Hydrocarbons by Dual-Reflux Pressure-Swing Adsorption, Ind. Eng. Chem. Res. 41
(2002) 3499–3504.
Appendix A. Supplementary data
[29] J.A. McIntyre, A.D. Ebner, J.A. Ritter, Experimental Study of a Dual Reflux
Supplementary data to this article can be found online at https://doi. Streams, Ind. Eng. Chem. Res. 49 (2010) 1848–1858.
[30] T.S. Bhatt, A. Sliepcevich, G. Storti, R. Rota, Experimental and Modeling Analysis
org/10.1016/j.seppur.2022.121907.
of Dual-Reflux Pressure Swing Adsorption Process, Ind. Eng. Chem. Res. 53 (2014)
13448–13458.
References [31] R. Wakasugi, A. Kodama, M. Goto, T. Hirose, Dual Reflux PSA Process Applied to
VOC Recovery as Liquid Condensate, Adsorption 11 (2005) 561–566.
[32] A.D. Ebner, J.A. Ritter, Equilibrium theory analysis of dual reflux PSA for
[1] BP, BP Statistical Review of World Energy,69th edition, 2020.
separation of a binary mixture, AIChE J. 50 (2004) 2418–2429.
[2] R.K. Pachauri, L.A. Meyer, Climate Change 2014: Synthesis Report. Contribution of
[33] T.S. Bhatt, G. Storti, J.F.M. Denayer, R. Rota, Optimal design of dual-reflux
Working Groups I, II and III to the Fifth Assessment Report of the
pressure swing adsorption units via equilibrium theory: Process configurations
Intergovernmental Panel on Climate Change, IPCC, Geneva, Switzerland, 2014.
employing heavy gas for pressure swing, Chem. Eng. J. 311 (2017) 385–406.
[3] U.S. EPA, Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990-2019,
United States Environmental Protection Agency, Washington, DC, 2021.
12
[34] T.S. Bhatt, G. Storti, J.F.M. Denayer, R. Rota, Equilibrium Theory-Based [43] F. Dong, H. Lou, M. Goto, T. Hirose, A new PSA process as an extension of the
Assessment of Dual-Reflux Pressure Swing Adsorption Cycles That Utilize Light Gas Petlyuk distillation concept, Sep. Purif. Technol. 15 (1999) 31–40.
for Pressure Swing, Ind. Eng. Chem. Res. 58 (2018) 350–365. [44] R. Weh, G. Xiao, E. Sadeghi Pouya, E.F. May, Helium recovery and purification by
[35] E. Rossi, G. Storti, R. Rota, Optimal Design Procedure of Dual-Reflux Pressure dual reflux pressure swing adsorption, Sep. Purif. Technol. 288 (2022), 120603.
Swing Adsorption Units for Nonlinear Separations, Ind. Eng. Chem. Res. 58 (2019) [45] J.G. Ning, C. Wang, J. Lu, Explosion characteristics of coal gas under various initial
6644–6652. temperatures and pressures, Shock Waves 15 (2006) 461–472.
[36] Y. Zhang, T.L.H. Saleman, G. Li, G. Xiao, B.R. Young, E.F. May, Non-isothermal [46] S.G. Subraveti, Z. Li, V. Prasad, A. Rajendran, Machine Learning-Based
numerical simulations of dual reflux pressure swing adsorption cycles for Multiobjective Optimization of Pressure Swing Adsorption, Ind. Eng. Chem. Res. 58
separating N2 + CH4, Chem. Eng. J. 292 (2016) 366–381. (44) (2019) 20412–20422.
[37] E.F. May, Y. Zhang, T.L.H. Saleman, G. Xiao, G. Li, B.R. Young, Demonstration and [47] D. Diagne, M. Goto, T. Hirose, Parametric Studies on CO2 Separation and Recovery
optimisation of the four Dual-Reflux Pressure Swing Adsorption configurations, by a Dual Reflux Psa Process Consisting of Both Rectifying and Stripping Sections,
Sep. Purif. Technol. 177 (2017) 161–175. Ind. Eng. Chem. Res. 34 (9) (1995) 3083–3089.
[38] S.V. Sivakumar, D.P. Rao, Modified Duplex PSA. 2. Sharp Separation and Process [48] R. Weh, G. Xiao, M.d. Ashraful Islam, E.F. May, Upgrading sub-quality natural gas
Intensification for N2-O2-5A Zeolite System, Ind. Eng. Chem. Res. 50 (2011) by dual reflux-pressure swing adsorption using activated carbon and ionic liquidic
3437–3445. zeolite, Chem. Eng. J. 392 (2020) 123753.
[39] S.V. Sivakumar, D.P. Rao, Modified Duplex PSA. 1. Sharp Separation and Process [49] R. Weh, G. Xiao, M.A. Islam, E.F. May, Nitrogen rejection from natural gas by dual
Intensification for CO2− N2− 13X Zeolite System, Ind. Eng. Chem. Res. 50 (2011) reflux-pressure swing adsorption using activated carbon and ionic liquidic zeolite,
3426–3436. Sep. Purif. Technol. 235 (2020), 116215.
[40] R.S. Thakur, N. Kaistha, D.P. Rao, Process intensification in duplex pressure swing [50] G. Li, E.F. May, P.A. Webley, S.H. Huang, K.I. Chan, Method for gas separation, US
adsorption, Comput. Chem. Eng. 35 (2011) 973–983. Patent US20170348670A1, 2015.
[41] Y. Guo, G. Hu, K. Chen, J. Guo, P.A. Webley, G.K. Li, Capture of dilute methane [51] D.T. Kearns, P.A. Webley, Modelling and evaluation of dual reflux pressure swing
with a novel dynamic-feed dual-reflux pressure swing adsorption process, AIChE J. adsorption cycles: Part II. Productivity and energy consumption, Chem. Eng. Sci.
68 (2022) e17390. 61 (2006) 7234–7239.
[42] Y. Guo, V. Jusko, G. Xiao, J. Hanekom, G. Hu, P.A. Webley, E.F. May, G.K. Li,
Separation of He/N2/CH4 ternary mixtures by a triple-reflux pressure swing
adsorption process, AIChE J. 68 (2022) e17569.
13
Chapter 5
Enrichment of low-grade methane by dual reflux vacuum
swing adsorption
Yalou Guo1, Xuewei Gu1, Guoping Hu1,2, Paul A. Webley3, Gang Kevin
Li*1
1
Department of Chemical Engineering, the University of Melbourne, Parkville, Victoria
3010, Australia
2
Ganjiang Innovation Academy, Chinese Academy of Sciences, Ganzhou 341119, China
3
Department of Chemical Engineering, Monash University, Clayton, Victoria 3800,
Australia
* Corresponding author: li.g@unimelb.edu.au
130
Chapter 5
Model assumptions
The main assumptions of DR-VSA model are summarized as follows:
a. Properties of gas-phase components are calculated using the Peng-Robinson
equation.
b. Momentum balance and pressure drop are described via the Ergun equation.
c. Material balance in the gas phase is described by convection only.
d. The flux between adsorbed and gaseous molecules is described by the linear
driving force (LDF) model with constant mass transfer coefficients (MTC).
e. Extended Langmuir II isotherms were used to describe the competitive
adsorption, determined by the partial pressure of each component in the gas phase.
f. Non-isothermal operation.
131
Chapter 5
Governing equations
The governing equations of the adsorption bed are summarized in Table A. 1.
Table A. 1: Governing equations for the adsorption column in this model

 150 × 10−5 µ g (1 − ε i ) 2 
 vg 
Momentum ∂P  (2rpψ ) ε i
2 3
 Eq. (A.1)
Ergun equation = ± 
balance ∂z  + 1.75 × 10 M W ρ g (1 − ε i ) v 2 
−5
 2rpψε i 3 g 
 
Material ∂ (vg ci ) ∂ci ∂Q
Plug flow + [ε i + (1 − ε i )ε p ] + ρb i =0 Eq. (A.2)
balance ∂z ∂t ∂t
LDF model with ∂Qi
Kinetic model = ki (Qi* − Qi ) Eq. (A.3)
constant MTC ∂t
∂ 2T ∂T ∂T
−k gε i + Cvg vg ρ g + [ε i + (1 − ε i )ε p ]Cvg ρ g
∂z 2 ∂z ∂t
Gas phase ∂vg
Eq. (A.4)
4hw
+P + hgs a p (T − TS ) + (T − TW ) = 0
∂z DB
∂ 2Ts ∂T ∂T
+ C ps ρb s + ρb ∑ i =1 ( C paiQi ) s
n
−ks
∂z 2 ∂t ∂t
Energy balance Solid phase Eq. (A.5)
n  ∂Qi 
+ ρb ∑ i =1  ∆H i  − hgs a p (T − Ts ) =
0
 ∂t 
∂ 2Tw ∂T 4 DB
( DB + Wr ) − DB2
2 2
∂z ∂t
Column wall Eq. (A.6)
4 ( DB + Wr )
2
0
( DB + Wr )
2
− DB2
Equation of Peng-Robinson ℜT a
=P − Eq. (A.7)
state equation Vi − b Vi (Vi + b ) + b (Vi − b )
132
Chapter 5
Analysis of recovery calculation method
Material relative balance error is a common problem of simulation work, manifesting as
the difference between inflow mass and outflow mass when the system reaches at the
cyclic steady state. The overall and component relative balance error were calculated by
Eqs. 5 and 6 in this study. Positive error value means that the mass flowing into the system
is more than that flowing out of the configuration, i.e., there are some materials
“disappeared”. In contrast, negative error value indicates that some materials are
“created”. Hence, even under the same operating conditions, different material balance
errors will also lead to different concentrations in two product streams, namely positive
error value results in high CH4 purity result while negative error value results in low
purity. If the calculation of recovery is accomplished using the feed gas and heavy flow,
the recovery will suffer similar effects by the material balance error while if the recovery
is calculated using the feed gas and light flow, as shown in Eq. 8, evaluating the separation
performance by the overall assessment score can mitigate the impact of material balance
error.
133
Chapter 5
Notation
adsorbent loading of component

coefficient for Peng-Robinson i per unit mass of adsorbent in
a Qi*
equation equilibrium with its partial
pressure in gas phase
coefficient for Peng-Robinson
b ℜ universal gas constant
equation
specific heat capacity of the adsorbed
Cpai rp particle radius
gas
specific heat capacity of the
Cps T gas temperature
adsorbent
Cpw specific heat capacity of the wall Tenv environmental temperature
specific gas phase heat capacity at
Cvg Ts adsorbent temperature
constant volume
ci molar fraction of component i Tw column wall temperature
DB internal diameter of column Vi the volume of component i
heat transfer coefficient between
hb vg superficial gas velocity
column and ambient
hgs Wr wall thickness
and solid
hw z axial distance coordinate
and wall
constant mass transfer coefficient of specific particle surface area per
ki αp
component i unit length of bed
enthalpy of adsorption for
kg heat conductivity of gas ΔHi
component i
ks heat conductivity of adsorbent εp particle voidage
kw heat conductivity of wall µg viscosity of gas mixture
MW molecular weight of gas mixture ρg gas density
P gas pressure ρw column wall density
adsorbent loading of component i per
Qi ψ particle shape factor
unit mass of adsorbent
134
Chapter 6
Chapter 6: Conclusions and general discussions

6.1 Conclusions
This research develops three new PSA processes to overcome some limitations of the
conventional PSA cycles. The improvement of PSA cycles in this thesis is primarily
achieved by engineering methods, i.e., proposing or combining innovative concepts to
achieve better separation performance or specific separation targets.
Chapter 3: In this chapter, we demonstrated a dynamic-feed to solve the mixing problem
of DR-PSA. The newly developed dynamic-feed DR-PSA cycle is investigated to enrich
dilute CH4 (2.4 mol.%) from N2. The sensitivity study shows that introducing more feed
inlet positions along the columns can lead to both higher CH4 purity and recovery. Results
show that the dynamic-feed DR-PSA cycle perform smoother axial concentration profiles
and better separation performance than the fixed-feed DR-PSA cycle. DR-PSA with the
dynamic-feed policy can achieve both higher methane purity (53.5% versus 47.5%) and
recovery (81.1% versus 72.2%) over the fixed-feed DR-PSA under the same operating
conditions.
Chapter 4: In this chapter, a novel triple-reflux pressure swing adsorption (TR-PSA)
process is developed to simultaneously obtain three high purity products from He-CH4-
N2 ternary mixtures within a single stage. Parametric studies were performed to
investigate the effect of some crucial operating parameters on the separation performance.
Results show that the TR-PSA process can effectively separate ternary mixtures into three
high-purity products by a single-stage PSA process. For the separation of a ternary
mixture consisting of 10% He, 20% CH4 and 70% N2, TR-PSA can obtain 60.0 % CH4
purity and 90.4% recovery, a He product with 45.3 % purity and a recovery of 91.3% as
well as a middle product of 95.8 % N2 with 68.9% recovery. The TR-PSA process shows
135
Chapter 6
comparable separation performance compared with the 2-stage DR-PSA process but
requires apparently lower capital and energy costs. The concept of this study can inspire
the design of “multi-reflux” PSA process for the separation of multi-component gas
mixtures.
Chapter 5: This chapter demonstrates the DR-VSA by combining dual reflux policy and
VSA process for the enrichment of low-grade CBM (5–20% CH4 mixed with N2) under
atmospheric pressure to avoid the use of compressor. A new process optimization method
with a dual convergence iteration algorithm and the multiplicative score is proposed to
effectively determine the optimal operating conditions. Two types of configurations (i.e.,
A- and B-type) with two adsorbents (Norit RB3 activated carbon and ionic liquidic
zeolites) are numerically investigated via a non-isothermal model. At the optimized
conditions, configuration AILZ is found to present the best separation results, which
enriches the feed gas of 5, 10, 15 and 20% methane up to 70.3, 75.2, 79.5 and 80.2% with
a recovery of 92.8, 93.2, 92.7 and 95.5% by consuming 657.3, 329.5, 240.2 and 180.8 kJ
energy for per mole captured CH4, respectively. The energy analysis suggests that the
recovery of low-grade methane by the AILZ cycle is energy-sustainable and can generate
extra benefits of GHG emissions reduction.
6.2 Future perspective
Although in this research, the proposed PSA cycles have been numerically investigated
and show great separation performance, further experimental validation is still necessary.
In Chapter 3, we demonstrated a dynamic-feed DR-PSA cycle, however, this dynamic-
feed strategy was not realized by a real-time monitoring and control system but by
continuously switching the feed positions according to measured axial concentration
profiles. Thereby, it is challenging to adopt this method when processing gases with
136
Chapter 6
variable concentrations. Additionally, the dynamic-feed strategy was only studied based
on a DR-PL-A configuration, and more work should be done to explore its performance
when deployed as the other three DR-PSA configurations.
In Chapter 4, we proposed a triple-reflux PSA cycle for one-step separation of ternary
mixtures. The adsorption column can be regarded as three main sections: each section is
for enrichment of one specific component. This cycle can be further improved by
combining with multi-layer adsorption column. The top section of column can be packed
with adsorbents with high N2 adsorption capacity while the bottom section can use
adsorbents with high CH4/N2 selectivity when taking the CH4-N2-He mixture as an
example.
In Chapter 5, the multiplicative score was based on dual-objective optimization and
assessment. However, more aspects need to be considered when evaluating a process,
such as capital investment, energy cost and operational cost, etc. More studies should be
done to extend this dual-objective optimization method as a multiple-objective
optimization method. Moreover, the DR-VSA process only considered the separation of
CH4 and N2. In real scenarios, low-grade CBM usually contains oxygen (O2), and more
rigorous safety analysis needs to be performed when considering the presence of O2.
Furthermore, most currently reported studies are based on lab-scale apparatus. The
scaling up of lab-scale PSA process to the industrial scale is a challenging task. In the
industrial scale, the temperature wave caused by adsorption heat cannot be ignored
anymore and the ambient conditions also have significant impacts on the separation
performance. Additionally, axial pressure drop can be neglect for a lab-scale rig, however,
for industry applications, pressure drop is more significant and is hard to avoid. The
137
Chapter 6
design and development of PSA processes should consider how to implement them in
industry and minimize their operational complexity.
Finally, the developed PSA cycles can be further explored in the future work for
separation of other gas mixtures, such as carbon capture, separation of hydrocarbons,
upgrade of biogas, etc. Further experimental and numerical investigations should be
conducted to study the application prospects of these cycles in other scenarios.
138

Thesis Doctor of Philosophy GUO Yalou

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Thesis Doctor of Philosophy GUO Yalou

Uploaded by

Copyright:

Available Formats

Development of new pressure

swing adsorption (PSA) cycles

Student ID: 1063461

ORCID ID: 0000-0003-4052-9681

Department of Chemical Engineering

The dissertation is submitted for the degree of

Doctor of Philosophy in Chemical Engineering under the guidance and supervision of

100,000 words in total, exclusive of tables, pictures, bibliographies, and appendix.

obtaining better separation performance.

Dual reflux pressure swing adsorption (DR-PSA), as a state-of-the-art process, uses a

swing adsorption (TR-PSA) for separating methane-nitrogen-helium ternary gas

of a new dual-objective optimization method.

81.1% vs. 72.2% for recovery, respectively.

In the second section, a triple-reflux (TR)-PSA process is demonstrated to separate a

two in the 2-stage DR-PSA system.

multiplicative score function to evaluate the separation performance of each case.

and Conversion, Carbon Capture Technologies, which is currently accepted and in

for publication, and all co-authors have authorized this publication to be

incorporated into this thesis.

iii. Chapter 4 is a paper “Separation of He/N2/CH4 ternary mixtures by a triple-reflux

pressure swing adsorption process” AIChE Journal, 2022, 68(5), e17569.

for publication, and all co-authors have authorized this publication to be

incorporated into this thesis.

adsorption” Separation and Purification Technology, 2022, 301, 121907.

for publication, and all co-authors have authorized this publication to be

incorporated into this thesis.

methane by dual reflux vacuum swing adsorption. Separation and Purification

Technology, 301, p.121907.

G.K., 2022. Separation of He/N2/CH4 ternary mixtures by a triple‐reflux pressure

swing adsorption process. AIChE Journal, 68(5), e17569.

methane with a novel dynamic-feed dual-reflux pressure swing adsorption process.

AIChE Journal, 68(1), e17390.

vacuum swing adsorption. AIChE Journal, e17668.

Framework MIL-53. ACS Applied Materials & Interfaces, 14(11), pp.13904-13913.

Hydrogen production from the air. Nature Communications, 13, p.5046.

acknowledge my committee chair, A/Prof. Gabriel da Silva, who provided me with

up of my experimental work, supported me in my simulation work, and helped me a lot

in writing and reviewing my papers.

thesis is dedicated to you.

List of publications ......................................................................................................... VI

Acknowledgements ....................................................................................................... VII

List of Tables ................................................................................................................. XII

List of Figures .............................................................................................................. XIV

1.1 Concept and mechanisms of adsorption ................................................................. 1

1.2 Fundamental PSA cycle .......................................................................................... 4

1.3 Dual-reflux PSA cycle ............................................................................................ 7

1.4 Thesis objectives and outline ................................................................................ 10

Chapter 2: Literature review and DR-PSA model

2.1 Key adsorbent properties ...................................................................................... 20

2.1.1 Equilibrium adsorption capacity and selectivity ............................................ 20

2.1.2 Adsorption kinetics ........................................................................................ 23

2.2 Development of the PSA cycle ............................................................................. 25

2.2.1 Vacuum swing adsorption (VSA) .................................................................. 25

2.2.2 Pressure equalization step .............................................................................. 27

2.2.4 Heavy purge (reflux) step............................................................................... 29

2.2.5 Displacement step .......................................................................................... 30

2.3 Advances in dual-reflux PSA process................................................................... 30

2.4. Simulation methods ............................................................................................. 37

2.4.1 Mathematical model ....................................................................................... 37

2.4.2 The flowsheet of DR-PSA model................................................................... 41

2.4.3 Benchmark results and discussion.................................................................. 44

Chapter 3: Capture of dilute methane with a novel dynamic-feed dual reflux

pressure swing adsorption process