Version of Record: https://www.sciencedirect.

com/science/article/pii/S0098135421000132
Manuscript_bc86cd2b9180c95c6b6d4d80cae2fa5d

A SIMPLE AND PRACTICAL PROCESS MODELING

METHODOLOGY FOR PRESSURE SWING ADSORPTION

Michael D. Sees, Toni Kirkes and Chau-Chyun Chen†

Department of Chemical Engineering, Texas Tech University, Lubbock, TX, 79409, USA.
† Corresponding author; Email: chauchyun.chen@ttu.edu

ORCID: Chau-Chyun Chen: 0000-0003-0026-9176

Abstract

Although many dynamic models exist for the design and simulation of pressure swing

adsorption (PSA) processes, these models involve the solution of a complex system of

coupled partial differential equations. Process engineers need a simple, practical, and yet

robust short-cut model that helps decide whether to implement a PSA system in a process

flowsheet. This work presents a “virtual” moving bed modeling methodology that considers

only mass and energy balances and adsorption isotherms to describe the cyclic steady state

behavior of PSA systems. Similar to tray efficiencies in distillation calculations, adsorption

efficiencies are further introduced to account for system “non-ideality.” A lab-scale air

separation system is used to illustrate the application of this modeling methodology.

Topical Heading: Process Systems Engineering

Keywords: Adsorption; Pressure Swing Adsorption; Process Intensification; Steady State

Process Simulation

© 2021 published by Elsevier. This manuscript is made available under the Elsevier user license
https://www.elsevier.com/open-access/userlicense/1.0/
 1

1.0 Introduction

Pressure-swing adsorption (PSA) is potentially the separation technology of choice for

separation of gas mixtures at moderate throughputs with high purity requirements. (Ruthven

et al., 1994). PSA has been extensively used for air drying, air separation, and hydrogen

purification (Basmadjian, 1996; Crittenden and Thomas, 1998; Ruthven, 1984; Ruthven et al.,

1994), ethanol dewatering (Teo and Ruthven, 1986; Simo et al., 2008; Simo, 2013), and more

recently for direct air and post-combustion CO2 capture (Kikkinides et al., 1993; Lin et al.,

2012; Schell et al., 2013; Wawrzyńczak et al., 2019), natural gas treating (Ackley et al., 1992;

Effendy et al., 2017; Effendy et al., 2018; Tagliabue et al., 2009; Tao et al., 2019) and

landfill/biogas recovery (Cavenati et al., 2005; Grande and Rodrigues, 2007; Knaebel and

Reinhold, 2003; Ryckebosch et al., 2011). However, despite the tremendous potential of PSA

or other adsorptive separations, their broader applications are hampered by the lack of easy-

to-use and reliable process modeling tools for process engineers.

Indeed, one major barrier to the wider application of PSA is the challenge in

understanding and modeling mixed-gas adsorption: both the physics of adsorption itself and

the unit operation (Basmadjian, 1996; Crittenden and Thomas, 1998; Ruthven, 1984; Ruthven

et al., 1994; Sircar, 1991; Sircar, 2002). As a unit operation, PSA has a strong dynamic

character due to its cyclic nature; the coupled partial differential equations (PDEs) which

describe the mass and energy transport in the system must be solved in tandem with the

differential algebraic expressions that describe the inter- and intraparticle transport and

adsorption (Biegler et al., 2005; Crittenden and Thomas, 1998; Ruthven, 1984; Ruthven et al.,

1994; Tien, 1994). Solvers for these stiff and highly non-linear equations exist, and dynamic
 2

PSA models have been implemented in commercial simulators with some success. However,

use of these dynamic PSA models requires specialized expertise, knowledge, and extensive

process and operation input that most practicing engineers do not possess, which in turn

hinders the wide application of PSA in the industry. The practicing engineer needs a simple

modeling methodology which represents the performance at the cyclical steady state (CSS) to

ascertain whether PSA fits the needs of their application.

Assuming instantaneous heat and mass transfer such that “local equilibrium” would be

achieved at all points in the bed, equilibrium PSA models like that proposed by Knaebel and

coworkers (Knaebel and Hill, 1985; Kayser and Knaebel, 1989; Ruthven et al., 1994) are

conceptually and mathematically simple, and can give analytical solutions for the CSS under

certain assumptions. However, these equilibrium PSA models are limited in scope in that they

are only applicable for equilibrium-based separations at their idealized conditions. They are

not versatile enough to truly represent the applicability of PSA to industrial separations:

separations that take advantage of diffusion kinetics (Ackley et al., 1992; Grande and

Rodrigues, 2007), leverage coupled pressure and temperature swings (Gales et al., 2003; Lin

et al., 2012), separate multiple components (He et al., 2012; Lopez et al., 2011; Ribeiro et al.,

2008), or involve very short cycle times (Sircar and Hanley, 1995). Therefore, current

equilibrium-based models cannot serve the needs of engineering community to ascertain the

applicability of PSA in their studies, although they make good teaching tools to explain the

fundamentals of PSA operation.

Some attention has been given to continuous countercurrent flow (CCF) models for

adsorptive separations, which has provided insight into adsorptive fractionation (Tondeur,
 3

1981) and has formed the basis for analyzing simulated moving bed chromatography (SMB)

(Ruthven and Ching, 1989). These models typically involve contacting moving gas and solid

phases in a countercurrent fashion; a very traditional concept in the study of mass transfer

equipment that can sometimes lend itself to simple McCabe-Thiele analysis. Solutions to CCF

models still require the user to find solutions to a system of PDEs to relate stream information

to bed design but avoid the need for dynamic simulation. However, CCF models have yet to

be adopted for PSA in part because CCF ensures that the bed will always be undersized since

CCF maximizes the driving force for mass transfer, but also because they have difficulty

accurately representing pressure-changing steps (e.g, blowdown) which may be important to

the overall unit performance; users have often resorted to the so-called “frozen solid”

approximation to simplify calculations (which has also been common practice in dynamic

simulation).

A CCF model specifically for PSA was proposed by Suzuki (Suzuki, 1984) in which

the solid adsorbent was treated essentially as a “membrane” separating a high-pressure and

low-pressure compartment. Suzuki’s original CCF-PSA model was extended to bulk

separations (Farooq and Ruthven, 1990) and to cycles with non-negligible pressure changing

steps (Leinekugel-le-Cocq et al., 2006), but still only works for systems which operate far

removed from equilibrium; the concentration profile in the adsorbent cannot be in equilibrium

with both the high- and low-pressure gases simultaneously, so the applicability is limited to

isothermal systems with significant mass transfer resistance and/or short cycle times (Ruthven

et al., 1994). In a sense, the previously discussed equilibrium models and the CCF-PSA

model represent two extremes in modeling PSA processes. However, they both are limited in
 4

their applicability and there are no models which span the divide between the two besides

rigorous dynamic simulation.

A solution to this problem may lie in a simplified dynamic PSA model which was

proposed by Oliver Smith IV and Arthur Westerberg targeted at identifying the ideal

operating cycle for a specific separation (Smith IV and Westerberg, 1990, 1991, 1992). In

their dynamic model, the authors relaxed the rigorous PSA equations and introduced two

adjustable parameters to simulate the actual adsorption behavior of the bed. The concentration

profile of the bed at the CSS was averaged, with the actual hold-up of the bed only reaching a

fraction of the equilibrium value and using a portion of the available adsorbent bed. The first

adjustable parameter, the adsorption efficiency ( ), describes the departure from equilibrium

while the second parameter, the bed utilization factor ( ), describes the fraction of the bed

covered by adsorbate. The definition of these parameters can be inferred from Fig. 1a, which

is recreated from the original text (Smith IV and Westerberg, 1991, 1992). The authors noted

that even though the profiles are dissimilar, the area-under-the-curve of the profiles are the

same and therefore the total number of moles adsorbed is also the same. Although the model

showed promise, the concept was only applied to the adsorption of a trace component in an

inert stream and still requires the user to simulate the cyclical batch process in time until the

CSS is approached. However, the concept of using efficiencies to average the profiles in the

bed to match experimental data might have some use.

This work presents a PSA modeling methodology based on combining Smith IV and

Westerberg’s adsorption efficiency idea with features from equilibrium and CCF models, with

the aim to allow practicing engineers to use steady state process simulators to simulate PSA
 5

unit operations. The modeling methodology is not intended to rigorously simulate the exact

movement of temperature and composition fronts, but to support engineering efforts to

simulate and evaluate applicability of PSA units for separations. We first present a general

adsorption unit operation model which treats the unit as a moving bed separator before

extending the adsorption model to a full PSA unit. Simple alternatives to the standard PSA

flowsheet are also discussed, and a case study on an air separation unit (ASU) is presented to

illustrate the methodology.

2.0 Methodology

2.1 The Virtual Moving Bed Model for Adsorption

To explain the workings of the so-called “virtual moving bed model”, we will initially

limit ourselves to separations which are based solely on a difference in adsorption equilibrium

selectivity. A single column in an adsorption unit can be treated mathematically as a black-

box which separates a fluid mixture feed into a product stream enriched in one or more of the

mixture components, while the remaining material is retained in the bed divided between the

bulk and adsorbate phases. In commercial process simulators, the black-box separator is

typically referred to as a component splitter. A generic component splitter is shown in Fig. 2;

feed enters the process at a molar flow rate of . The feed mixture is then separated into two

streams: 1) product stream with flow rate enriched with the light key (LK) components

and 2) retentate stream with flow rate enriched with the heavy key (HK) components. In

theory, a component splitter can have any number of feed streams, so a secondary feed is

shown in Fig. 2, , although has a particular definition within the context of the moving
 6

bed model and will be discussed later. The total mole balance and the mole balance for

component i around the splitter are given by Eqn. (1) and (2).

0 (1)

, 0 (2)

In Eqn. (1) and (2), is the molar flow rate of stream k and , is the mole fraction of

component i in stream k. In a typical adsorption system, the feed streams would be fully

specified (T, P, , , and ), but the outlet stream conditions and flow rates need to be

calculated by specifying split values for each component passed into the splitter. The split

values, , , relate the flow rate of component i in a specific outlet stream j to the total influx

of component i, as given in Eqn. (3).

,
,
∑
(3)
,

To obtain a unique solution, only one , for each component needs to be specified and ,

ranges from zero, representing no transfer, to unity, representing full transfer. Obviously, the

values of , which are attainable are dictated by the physics of the adsorption process,

namely the bed loading due to the adsorption kinetics and the pure component and mixture

isotherms.

When a fluid mixture is fed into a closed column which contains solid adsorbent and

allowed to equilibrate at a specified temperature and pressure, the bulk phase will become
 7

enriched in one or more of the less adsorbed components. The bulk phase fluid can be drawn

off as more feed is introduced at constant temperature and pressure until eventually the solid

adsorbent is saturated and adsorption ceases. The bulk phase fluid has the same composition

as the feed and contains , ⁄ moles of adsorbate i per unit mass of adsorbent, where is

the void fraction of the bed, is the bulk density of the adsorbent, and , is the bulk fluid

phase concentration of the adsorbative. The solid has adsorbed some number of moles of
∗
adsorbate, , according to the adsorption isotherm which can be obtained by a variety of

techniques. (Bartholdy et al., 2013; Crittenden and Thomas, 1998; Meyers and Prausnitz,

1965; Ruthven, 1984; Ruthven et al., 1994; Shapiro and Stenby, 1998; Tien, 1994) The mass-

specific total molar hold-up of component i when the bed is fully saturated with feed, " # ,

can be given by Eqn. (4).

#
" ∗
$%, , &+ , ⁄ (4)

If fresh or pre-loaded solid could be constantly introduced into the bed at an

appropriate mass flow rate (( ) ) and the saturated solid and entrained fluid removed, the

unit would approach a steady state operation. Operations involving the co- or countercurrent

flow of solids in a fluid stream are referred to as “moving bed” units, hence the title of the

model. Moving bed models themselves are not necessarily novel but have focused on

analytical solutions for simple isotherms with limited deviations from equilibrium conditions

and were most applicable to chromatography processes. (Ruthven, 1984) The moving bed

likely does not exist in the system, although the dated Hypersorption process did involve

solids conveying (Crittenden and Thomas, 1998; Ruthven, 1984), so the moving bed here is a
 8

“virtual moving bed”. At steady state, the flow rate of the saturated solid and entrained fluid

in the retentate stream could be described by Eqn. (5).

#
, " ( ) (5)

Equations (3) and (5) can be combined to give a relationship between operating

conditions and the split value for the retentate stream .

#
" ( )
,
∑
(6)
,

Equation (6) exhibits many features which agree with general trends in adsorption operations.

A good separation would involve large differences between , for the components involved.

Changes in the denominator of Eqn. (6) would involve either changing the feed conditions or

adding additional material to the bed, such as feeding an adsorbent which is pre-loaded with

one of the components ( in Fig. 2). Changing the feed conditions could either increase or

decrease , , but secondary feeds will always decrease , . Such behavior is consistent with

the observation that adsorption beds pre-loaded with the HK component breakthrough before

beds which are initially clean. On the other hand, changes to the numerator are dominated by

changes in operating conditions and adsorbent properties. The composition of the outlet

retentate stream is fixed for an adsorbent operating at a given temperature and pressure, so

changing ( ) will only adjust the flow rate of the outlets. Therefore, large differences in ,

are most easily achieved by proper adsorbent selection to change the hold-up term of Eqn. (6).

2.2 Adjustment for Non-Ideal Operation

 9

Clearly, Eqn. (6) represents a limiting case; non-idealities can affect the adsorption

process, such as mass transfer resistance, the presence of multiple transitions in the bed,

and/or displacement by a multicomponent feed (Basmadjian, 1996; Crittenden and Thomas,

1998; Ruthven, 1984; Ruthven et al., 1994). Similar to Smith IV and Westerberg (Smith IV

and Westerberg, 1991, 1992), the adsorption efficiency ( * ) and bed utilization factor ( * ) can

be added to Eqn. (4) to account for non-ideality. There are, however, some differences in the

definitions used in this work and those in the work of Smith IV and Westerberg. Primarily,

Smith IV and Westerberg treat the adsorbent bed in a cyclical batch manner, defining * and

* based on the CSS profile in the bed for a trace adsorbative in an inert stream. Therefore, *

is understood as being the fraction of the adsorbent in the bed which is in equilibrium with the

feed gas while the rest of the bed is in equilibrium with pure product. However, both

parameters were defined based on the concentration profile of the adsorbate for the HK

component, which is difficult to measure directly in the bed in practice. The bulk phase

composition is much easier to measure since it can be calculated from temperature and

pressure measurements. Therefore, * in the moving bed model, is defined as the fraction of

the equilibrium void space hold-up that is achieved. To avoid confusion, bed utilization in the

moving bed model is referred to as the bed-state efficiency and given the symbol + . A visual

comparison of the two models is given in Fig. 1.

Applying these ideas, the “two-efficiency” model (TEM) for the moving bed can be

described by Eqn. (7).

+, ∗
$%, , &+ , ⁄ -( )
,
∑
(7)
,
 10

In Eqn. (7), * is the adsorption efficiency and + is the bed-state efficiency for component i.

Equation (7) is the most general form of the moving bed model, and one can infer the

relationship between the void space hold-up, adsorbate hold-up, + , and * from Fig. 1.

Effectors which cause the void space hold-up to be non-homogeneous through-out the bed

will affect the value of + (e.g. axial/radial dispersion, pressure drop, presence of multiple

transitions, etc.) while effectors which cause the adsorbate phase to be out of equilibrium with

the bulk (i.e. mass/heat transfer resistance) are reflected in the value of *.

For simple cases, * can be taken to be unity and only + will vary. The resulting split

values would be given by the “one-efficiency” model (OEM), Eqn. (8).

#
+" ( )
,
∑
(8)
,

The OEM approach would only be appropriate in situations where local equilibrium is

attained nearly instantaneously (i.e. little to no mass transfer resistance) because the changes

in the adsorbate hold-up must be dictated entirely by changes in the bulk phase concentration.

Since both * and + are related to the departure from the equilibrium hold-up, nearly ideal

equilibrium operations should be described by OEM and have values of + which are close to

unity. The converse is not necessarily true; + = 1 only means that the void spaces of the bed

have been completely filled with feed and the adsorbate phase is in equilibrium. The bed

could be well past breakthrough and still have + 1, which would result in an off-spec

product stream. The TEM has the additional flexibility introduced by * to account for

potentially complex rate behavior. Therefore, kinetic separations, which are based on

differences in diffusivities of the adsorbatives, are best modeled with the TEM approach.
 11

At this time, there is no method to estimate the efficiencies for an adsorption unit a

priori, and such predictions are out of the scope of the current work. The efficiencies are

model parameters which can be adjusted to match plant or experimental data. As an analogy

to tray efficiencies in multi-stage distillation calculations, an initial estimate of unity should

be used in a simulation to assess the performance of PSA units at idealized conditions while

non-unity values should be used to allow for system non-ideality and design safety margins.

For simplicity, users should default to OEM when data is limited. TEM requires the user to

specify two efficiency values for each component, which is difficult to do without some

knowledge of the hold-up of each component in the bed, whereas OEM only requires a single

value per component. To determine the efficiencies for TEM, measurements of either the gas

phase concentration or bed loading at different points in the bed would be required. Although

bed loading is more directly related to *, gas phase measurements are much easier to perform

and give + directly: see Fig. 1b. So, + can be calculated from measurements and then * can

be adjusted to match performance data. OEM efficiencies, on the other hand, can be identified

by using only measurement of the product flow rate and composition at minimum, which

should be readily available. Packed bed correlations for mass/heat transfer resistance,

dispersion, and pressure drop are available (Ruthven, 1984; Yang 1997) which may be a

useful tool for the user to determine which model suits their needs.

2.3 The Energy Balance in the Moving Bed Model

Adsorption involves both mass and energy exchange between the adsorbate and the

adsorbent, and the temperature change due to the heat of adsorption is typically non-

negligible for gas adsorption (Basmadjian, 1996; Crittenden and Thomas, 1998; Ruthven,
 12

1984; Ruthven et al., 1994) while liquid adsorption units typically operate nearly

isothermally. (Basmadjian, 1996; Crittenden and Thomas, 1998; Ruthven, 1984) The energy

balance for the adsorption unit is given by Eqn. (9), assuming the adsorbate phase has the

same heat capacity as the bulk fluid.

:;

/ , 0∆2 3% , 4 5 + 6 ∆2 )
3% , 4 , ∗ 57

89

(9)
< , 0∆2 3% , 4 5 + 6 ∆2 )
3% , 4 , ∗ 57

89

+( ) => ) 3%? < % 5

In Eqn. (9), => ) is the heat capacity of the solid adsorbent, Tr and Ts are the outlet and inlet

temperatures of the moving bed, ∆2 is the molar enthalpy of component i in the stream, and

∆2 )
is the heat of adsorption for component i. The molar heat of adsorption, ∆2 )
, is a

function of temperature and pressure of the adsorbate stream as well as the @ ABCBDEBA(

loading. [1 – 3]
The parameter 6* is the adsorbate fraction of the stream, similar to a liquid or

vapor fraction. For fluid streams, 6* is zero, but for streams containing adsorbate, 6* has a

value less than unity. Adiabatic operation is common in many adsorption units (Basmadjian,

1996; Crittenden and Thomas, 1998; Ruthven, 1984; Ruthven et al., 1994), so there is a need

to solve Eqn. (9) simultaneously with Eqn. (1) and (2) by iteration to properly evaluate the

equilibrium conditions in Eqn. (7) and (8), where the equilibrium condition is now evaluated

at %? . Since * and + adjust the adsorbate hold-up by averaging the composition profile, the

calculated temperature is also an average value for the entire bed and may differ from plant or
 13

experimental data. Non-isothermal conditions could also be considered by modifying Eqn. (9)

to include heat transfer between the bed, wall, and surroundings.

2.4 The Moving Bed Model for PSA

For simplicity, we now limit the discussion of the adsorption model to a PSA unit

separating a binary ideal gas mixture by a basic Skarstrom cycle which can be modeled

appropriately with the OEM approach. Full PSA units are modeled by connecting component

splitters together by their virtual moving bed inlets and outlets. Each component splitter

represents the state of the bed at the termination of a specific combination of pressure,

temperature, and feed conditions: they are therefore referred to as operating states. Operating

states are the true fundamental building block of the virtual moving bed model and can

represent a single step in the cycle schedule or a combination of steps when appropriate, such

as combined pressure changing and constant pressure steps (e.g. pressurization and feed, or

blowdown and purging). As before, the split values are written for the retentate streams which

represent the total molar hold-up of the bed in a given operating state.

A PSA unit typically uses multiple beds which are the same size, so the mass of

adsorbent in each bed is the same. However, the time scales for the adsorption and desorption

operations are not the same. To represent the unit at the CSS, the flow rates involved in the

flowsheet need to be normalized, and the total cycle time, FGHG , provides a convenient

normalization factor since it dictates the total time for the adsorbent to return to its initial

state. Therefore, the flow rates involved are cycle-average flow rates and are given an

overbar designation to distinguish them from the actual average flow rates; the two are related
 14

by I $JF ⁄FGHG &, where M is the number of times the particular state occurs and is

typically the same as the number of beds in the system, N. A similar method is used in

Suzuki’s CCF-PSA model (Suzuki, 1984). The mass flow rate of adsorbent in the moving bed

is then related to the total mass of adsorbent in the unit and is given by (
K ) L( ) ⁄FGHG .

The practicing engineer should keep in mind that the actual flow rates from plant or

experimental data may differ from the flowsheet values since the simulation flow rates are

scaled to FGHG .

The flowsheet for a PSA-Skarstrom cycle is given in Fig. 3. In a Skarstrom cycle, the

bed operates between two states: a high-pressure adsorbing state at a pressure 4M and a low-

pressure desorbing state at a pressure 4N . Therefore, the PSA flowsheet consists of two

operating states connected by their respective retentate streams IM and I N , which make up

the virtual moving bed, and various streams representing the recycling of material in the

Skarstrom cycle. The splitters are arranged from the highest pressure on the left to lowest

pressure on the right. Stream I M is representative of the retentate after the pressurization and

feed steps, while I N is representative of the retentate after the blowdown and purge steps.

The light key-enriched product, or high-pressure product (HPP), is given by I M while the

heavy key-enriched product, or low-pressure product (LPP), is given by I N . The LPP

stream combines the blowdown and purge products although, in theory, the blowdown

product could be collected independently from the purge product. The feed gas required to re-

pressurize the bed to 4M is given by I and the sweep gas required to purge the bed is given

by I O.
 15

Besides specifying the split values (i.e. efficiencies) and the feed gas conditions, the

flowsheet requires two additional specifications: I O for the purge step and I for the

pressurization step. The sweep gas flow rate can be conveniently specified using a sweep-to-

feed ratio defined by PQ/S I O⁄ I . If plant data is available, then PQ/S can be fixed; if

the value is unknown, then the sweep flow rate must be treated as an adjustable parameter. A

lower bound for the sweep gas requirement can be estimated by I TO UMV 3% O 54N (
K ) ,

where UMV is Henry’s constant for the most strongly adsorbed component at the temperature

of the sweep gas, % O . (Basmadjian, 1996) Similarly, the pressurization flow rate can be

given by a pressurization-to-feed ratio, defined as 4⁄S I ⁄ I . The pressurization flow

rate also has a lower bound given by I T 34M < 4N 5(

K ) ⁄ W%, which is the amount of

gas needed to pressurize the void space of the bed in the absence of adsorption. This value

would be realized for small beds undergoing very fast pressurization such that adsorption does

not have time to occur. If pressurization was allowed to proceed sufficiently slowly, the

pressurization flow rate would approach an upper bound given by I T X

IT + ∆ ∗(
K ) ,

where ∆ ∗
is the total change in adsorbate hold-up during pressurization. Clearly, if the unit is

being pressurized with feed the upper limit would completely saturate the bed and would be

avoided in practical operation.

From the moving bed model, the split values for the high-pressure and low-pressure

states are given by Eqn. (10) and (11). The mole fraction of component i in a gas stream is

denoted by Y , while the total mole fraction is still used for adsorbate-containing streams.

Both states have their own sets of + since composition waves travel differently during

adsorption and desorption and the flow conditions may differ. The bed-state efficiency for the
 16

high-pressure splitter is denoted by +M and the low-pressure by +N . The equilibrium

conditions for the two operating states are also different; the high-pressure state should be in

equilibrium with the feed gas while the low-pressure state should be in equilibrium with the

sweep gas.

+M , ∗ $Y , , 4M , %M & + Y , 4M ⁄ W%M -(
K )
,M
31 + 4/S5Y , I + ,N I N
(10)

+N , ∗ $Y , O , 4N , %N & + Y , O , 4N ⁄ W%N -(
K )
,N
31 + PQ ⁄S 5Y , O I O + Y ,M I M
(11)

Eqn. (10) and (11) can be further simplified by denoting the total molar hold-up under the

high-pressure equilibrium conditions by " M and the low-pressure molar hold-up by " N . It is

convenient to write the 31 + 4/S5Y , term as Y Z, . The sweep gas term can be treated is a

similar way, rewriting 3PQ ⁄S 5Y , O as Y Z, O, yielding Eqn. (12) and (13).

+M "M (
K )
,M
Y , I + ,N IN
Z (12)

+N " N (
K )
,N
Y , OI O +
Z
IM
(13)
,M

Conceivably, a PSA modeling problem may involve the determination of + from

plant or experimental data, but , is the actual input for the model in process simulators. It is

convenient to recognize that ,N IN +N "N (

K ) and ,M IM +M "M (
K ) . Therefore,

Eqn. (12) and (13) can be recast in a more useful form as Eqn. (14) and (15), which allow the

user to directly determine + from the process simulator output.

 17

,M I
+M 0Y Z, + ,N Y Z, O7 [ \[ M \
1< " ( K
(14)
,M ,N )

,N I
+N 0 ,M Y Z, + Y Z, O7 [ \[ N \
1< " (K
(15)
,M ,N )

Most PSA units are operated adiabatically, so if the adiabatic temperature rise or fall is

known then the non-isothermal behavior could be captured by setting the high (%M ) and low

(%N ) temperatures in the high and low-pressure beds, as written in Eqn. (10) and (11) and

adjusting the product gas stream temperatures. The values of %M and %N could be found from

plant or experimental data by the same method as +M or +N : averaging the temperature

profile of the bed. With %M and %N specified, the gas product temperatures would be calculated

by Eqn. (8). If the temperature change is not known, then additional energy balances can be

written around each splitter using Eqn. (8) which would allow for estimation of the final bed

temperatures.

2.5 Model Flowsheet for Alternate Cycles

The general nature of the moving bed model lends itself to use in cycles involving

scheduling different from the Skarstrom cycle. So long as the beds in the PSA unit are only

operating between two pressure levels, representation of the new cycle steps just involves the

addition of new streams to the flowsheet. Clearly, these streams create more internal recycles

for the flowsheet and will require some degree of specification. Two simple extensions of the

virtual moving bed model are shown in Fig. 4.

 18

2.5.1 Pressurization by Product (Backfill)

Pressurization with feed gas pre-loads the PSA column with the more strongly

adsorbed component which reduces the overall capacity of the bed and the efficiency of the

unit. (Kayser and Knaebel, 1989; Knaebel and Hill, 1985; Ruthven et al., 1994) To avoid

introducing undesired composition fronts into the column, especially in equilibrium

separations, some PSA units pressurize using a portion of the HPP, referred to as backfill. The

flowsheet is shown in Fig. 4a. The equations relating + to , are very similar to Eqn. (14)

and (15), with a minor modification.

,M I
+M 0Y ZZ, + ,N Y Z, O7 [ \[ M \
1< " ( K
(16)
,M ,N )

,N I
+N 0 ,M Y ZZ, + Y Z, O7 [ \[ N \
1< " (K
(17)
,M ,N )

In Eqn. (16) and (17), Y Z, O remains unchanged, but Y ZZ, takes the place of Y Z, to account for

the fact that the pressurizing gas now has a different composition than the feed and is defined

as Y ZZ, 0Y , + 34/S5Y ,M 7. This substitution can be made in any flowsheet where

pressurization with product is desired over pressurization with feed.

2.5.2 Desorption by Vacuum Purging

For PSA systems which involve desorption at sub-atmospheric conditions, commonly

referred to as vacuum swing adsorption (VSA) systems (Basmadjian, 1996; Crittenden and

Thomas, 1998; Ruthven, 1984; Ruthven et al., 1994), the flowsheet is the same as the

traditional Skarstrom cycle and the split values are described by Eqn. (14) and (15). However,
 19

the use of sweep gas is no longer required at sufficiently low pressures, and purging can be

accomplished solely by vacuum. Removal of the sweep gas stream makes the VSA flowsheet

the simplest to analyze, as shown in Fig. 4b.

,M I
+M Y Z, [ \[ M \
1< " ( K
(18)
,M ,N )

,M ,N I
+N Y Z, [ \[ N \
1< " (K
(19)
,M ,N )

Equations (18) and (19) greatly simplify the relationship between + and , due to

the removal of the internal recycle introduced by traditional purging. However, the

equilibrium conditions for the low-pressure separator are less clear since the high-pressure

retentate is the only feed to the operating state. In a rigorous sense, the user could split the

moving bed into a “solid phase” stream and a “void phase” stream, and an iterative flash

calculation could be performed to determine the equilibrium conditions of the bed.

Alternatively, since the pressure is lowered deep into the Henry’s law regime, the user could

assume that the equilibrium conditions are pure gases at 4N and %N . The assumption is justified

since, at sufficiently low pressures, components of a gas mixture tend to adsorb independent

of one another. (Basmadjian, 1996; Crittenden and Thomas, 1998; Ruthven, 1984; Ruthven et

al., 1994)

The flowsheet for vacuum purging also works, in theory, for self-purging kinetic

separations, where desorption of the light key component from the bed sweeps the heavy key

from the adsorbent. However, self-purging cycles may not operate at low enough pressure to

justify the pure component assumption, so the more rigorous approach is potentially needed.

Alternatively, the user could choose to use the frozen adsorbate assumption to simplify
 20

calculations; if the adsorbate composition is assumed to be “frozen” during the blowdown and

subsequent desorption steps (Chou and Huang, 1994; Ruthven et al., 1994), then the

composition of the bulk gas also remains constant through the desorption. Therefore, the

equilibrium conditions can be taken as Y , at 4N and %N .

3.0 Calculation

3.1 Case Study: Air Separation on Zeolite 5A

Consider the bulk separation of oxygen from dry air over zeolite 5A in an air

separation unit (ASU) as described by Kayser and Knaebel. (Kayser, 1985; Kayser and

Knaebel, 1986) In the equilibrium separation, oxygen (YNV, 0.2174) is the LK while

nitrogen is the HK; argon adsorbs similarly to oxygen and therefore is included in the “pure”

oxygen product. Six air separation cases with the pressure ratio varying from ~6 to ~800 were

reported at an average bed temperature of 45 °C with a cycle time of 210 s. Each of the two

beds contained 252.4 g of a Union Carbide zeolite 5A ( = 0.480, = 810 kg/m3). According

to their work, the ASU should operate ideally in that there is negligible mass transfer

resistance, dispersion, pressure drop, heat effects, and dead volume. Therefore, the system

should be appropriately simulated by the PSA-Skarstrom flowsheet with backfill (Fig. 4a)

using the OEM approach with the values of + set to unity and then adjusted to examine the

effects of system non-ideality. Also, TEM could be applied if one wishes to examine the

system non-ideality related to the hold-up at the conclusion of the operating and purging

steps.

Inputs to the moving bed model are given in Table 1. Given that the feed rate, HPP

rate, and HPP purity were reported, one could identify +M from the measurements and
 21

validate the assumption of unity +M . To identify the values for +N , measurements would be

needed of the total hold-up of the bed after the purge step was complete, which could be

determined by overall mass balances if the number of moles used to pressurize and sweep the

bed was known. Therefore, the low-pressure state is assumed to operate ideally (+N 1).

The number of moles used to pressurize and sweep the bed were not reported, so the

ASU was assumed to use the minimum amount of sweep gas ( I TO 5 and the maximum

amount of pressurization gas ( I T X 5, which is consistent with the equilibrium assumptions

inherent to Kayser’s work. Doing so fixes the values of PQ/S and 4W/S for each case. For

Cases 5 and 6, the ASU was purged by vacuum alone (see vacuum purging flowsheet, Fig.

4b), so the low-pressure equilibrium conditions were taken to be pure component gas at the

system temperature and pressure.

Lastly, the unit was modeled as an isothermal unit since the authors did not publish

enough information on the thermal behavior of the ASU. For each case, the authors report the

highest and lowest temperatures recorded during the operation of the unit. These could be

used as estimates for %M and %N , but the measurements are ambiguous as there is no indication

of when the measurement occurs during the cycle or if the value represents an accurate

average temperature of the whole bed. It is also difficult to model the non-isothermal behavior

of the ASU since the adiabatic bed assumption is a poor assumption for lab-scale PSA units.

(Ruthven et al., 1994)

For the representation of the isotherm data, the pure component pressure vs.

equilibrium loading for both oxygen and nitrogen (Kayser, 1985) were fit simultaneously to a

temperature-dependent Langmuir expression given by Eqn. (18 - 19), which also allows the
 22

calculation of Henry’s constant analytically by Eqn. (20), which is needed for estimating the

sweep gas requirement.

4TD
]
∗
1+D 4
(18)

c 1 1
D D] exp ab d b ] < de
(19)
W % %

U TD
] (20)

In Eqn. (18 – 20), ]

T is the saturation capacity, D] is the Langmuir affinity constant at the

reference temperature % ] , and 3c ⁄W 5 is the reduced isosteric heat of adsorption. Since

pressures are relatively low, the ratio of the loading to the pressure was fit instead of the

loading data itself, akin to fitting the spreading pressure. (Richter et al., 1989)

∗ ∗ f
X>, ,
[ < \ min
4 4
(21)
89

The fit of the data is shown in Fig. 5 and the regressed parameters are given in Table 2. The

data for nitrogen at 333 K was omitted from the fit because the resulting heat of adsorption for

oxygen was higher than nitrogen, which is known to not be the case. However, as can be seen

in Fig. 5, the loading prediction for nitrogen matches the data satisfactorily.

For gas mixtures, the equilibrium loading was calculated using the extended Langmuir

expression given by Eqn. (22). Since oxygen and nitrogen are of similar molecular size and

have similar chemical properties, the extended Langmuir expression is expected to provide

reasonable representation of the mixture equilibrium.

 23

TD4
]
∗
1 + ∑D 4
(22)

4.0 Results and Discussion

To examine the validity of assuming unity efficiencies, the values of +M were

identified from the data of Kayser and Knaebel. (Kayser, 1985; Kayser and Knaebel, 1986).

The results of the regression are shown in Fig. 6. The values of +M all fall around unity,

supporting the feasibility of the proposed modeling methodology. Since the unit is expected to

be ideal, the conclusion is that the ASU does operate nearly ideally. As mentioned previously,

+ is designed to capture system non-idealities which affect the performance of the unit.

Kayser, in his thesis (Kayser, 1985), lists several potential sources of non-ideality:

temperature changes (± 5 K), valve interaction (leading to a slight pressure rise during

operation), non-negligible pressure drop, and difficulty operating at the actual cycling limit

(i.e. +N ≠ 1). It is also possible that the Langmuir expression for pure component adsorption

and the extended Langmuir expression for mixed-gas adsorption do not accurately describe

the adsorption equilibria (Bartholdy et al., 2013), although they are likely reasonable for this

specific case given the chemical and physical similarity of the adsorbing species, and choice

of T and P. Beyond the scope of this study, detailed investigation of the performance of the

unit would be required to understand the relative contribution of each effector to +M and

would likely require dynamic simulation, or additional studies on the original ASU or a

reconstruction thereof. None the less, the moving bed model with unity +M or properly
 24

identified values of +M should provide practicing engineers a simple, practical, and robust

short-cut tool to simulate and evaluate applicability of PSA units for separations.

With the regression done, the assumption of unity efficiencies can be evaluated futher.

Fig. 7 shows the OEM results for the “ideal” case where +M +N 1 and the “regressed”

case with the values of +M shown in Fig. 6. Clearly, the resulting oxygen recovery and

product purity align well with the experimental results when the efficiency parameters are

used to capture the non-ideal operation of the system. The “ideal” case is not in quantitative

agreement with the experimental data, with the exception of Case 5 where nitrogen is

predicted to be completely removed. It is reasonable to assume the system studied is highly

sensitive to the value of +M given the sharpness of the separation. The “ideal” case does show

qualitative agreement with the data and also with expected trends, namely that as recovery

increases between two cases the product purity decreases. Therefore, unity efficiencies are

appropriate as a first guess for a particular simulation but could be useful for setting bounds

on process variables. Here, unity efficiencies would imply (

K ) needs to be increased to

achieve the expected performance, which can be done by increasing the mass of adsorbent in

the system (or alternatively lowering the cycle time). This would set an upper bound limit on

the mass of adsorbent in the system required for the separation.

The choice of equilibrium conditions makes a difference for the two highest-pressure

ratio cases since they use vacuum purging (Fig. 4b). The results in Fig. 6 and 7 were obtained

by assuming the equilibrium conditions corresponded to pure gases adsorbing at 4N . Since this

assumption is only justified at very low pressures, the calculations were repeated using the

composition of the void space hold-up from the high-pressure operating state at 4N as the
 25

equilibrium condition. This is equivalent to assuming that only the total number of moles of

adsorbate change, but not the composition of the adsorbate itself: the frozen solid

approximation. The resulting values of +M were within less than 1 % of the original values,

indicating that the pure gas assumption is reasonable for this case.

5.0 Conclusions

A general model for PSA was proposed which should enable the practicing engineer to

use process simulation software to model the performance of PSA units. Introducing

efficiencies to account for system “non-ideality”, this “virtual” moving bed modeling

approach uses only simple mass and energy balances combined with adsorption fundamentals

to explain the behavior of PSA units simply and concisely. The model is most applicable to

simple cycle schedules; each operating state requiring a minimum of 2N efficiency

specifications. Cycles which involve a large number of operating states, like polybed

processes where beds routinely exchange material outside of purging, or beds which have

multiple layers of adsorbent, each layer requiring a separate splitter, could become somewhat

unwieldy.

Operating states which represent pressure changes were not explicitly considered in

this work but play an important role in the process as a whole, particularly in co- or

countercurrent depressurization and pressure equalization steps. Since pressure changing steps

are particularly sensitive to the interplay between pressure, composition, and spatial position,

they present a difficult modeling challenge for steady state PSA models. In the context of the

virtual moving bed model, the composition profile is averaged over the length of the bed, so

the variation is gas purity during the pressure change is difficult to predict reliably. Some
 26

future work should be dedicated to investigating representation of explicit pressure changing

states in the model, especially states which share material directly between one another as in

pressure equalization.

Additional future work should be focused on understanding the impact of different

effectors on the efficiency parameters and + . Knowledge of attainable values of and +

is a necessity for both the simulation and design of PSA units. Ideally, engineering

correlations could be developed which relate the efficiencies to the various design and

operational factors which affect the unit’s performance, such as superficial fluid velocity and

fluid physical properties. However, in the least, recommended ranges for different separations

could be proposed which allow the engineer to represent the unit appropriately. Regression of

appropriate efficiency values requires a concerted effort to collect and analyze the data of real

PSA units operating with different adsorbent-adsorbate pairs over a range of operating

conditions. Dynamic simulations of PSA units can also help fill the data gaps in PSA

literature and special attention should be paid to developing dynamic simulation tools which

are simpler to use and understand.

Acknowledgments

Funding support is provided by the U. S. Department of Energy under the grant DE-

EE0007888. The authors gratefully acknowledge the financial support of the Jack Maddox

Distinguished Engineering Chair Professorship in Sustainable Energy sponsored by the J.F

Maddox Foundation.
 27

Disclaimer

This report was prepared as an account of work sponsored by an agency of the United States

Government. Neither the United States Government nor any agency thereof, nor any of their

employees, makes any warranty, express or implied, or assumes any legal liability or

responsibility for the accuracy, completeness, or usefulness of any information, apparatus,

product, or process disclosed, or represents that its use would not infringe privately owned

rights. Reference herein to any specific commercial product, process, or service by trade

name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its

endorsement, recommendation, or favoring by the United States Government or any agency

thereof. The views and opinions of authors expressed herein do not necessarily state or reflect

those of the United States Government or any agency thereof.

 28

7.0 Works Cited

1. Ackley, M.W., Giese, R.F, Yang, R.T., 1992. Zeolites 12, 780 – 788.

2. Bartholdy, S., Bjorner, M.G, Solbraa, E., Shapiro, A., Kontogeorgis, G.M., 2013. Ind.
Eng. Chem. Res. 52, 11552 – 11563.

3. Basmadjian, D., 1996. The Little Adsorption Book: A Practical Guide for Engineers and
Scientists, CRC Press, Florida.

4. Biegler, L.T., Jiang, L., Fox, V.G, 2005. Sep. Pur. Rev. 33, 1 – 39.

5. Cavenati, S., Grande, C.A., Rodrigues, A.E., 2005. Energy and Fuels 19, 2545 – 2555.

6. Chou, C.-T., Huang, W.-C., 1994. Chem. Eng. Sci. 49, 75 – 84.

7. Crittenden, B., Thomas, W.J., 1998. Adsorption Technology & Design, Butterworth-
Heinemann, Oxford.

8. Effendy, S., Xu, C., Farooq, S., 2017. Ind. Eng. Chem. Res. 56, 5417 – 5431.

9. Effendy, S., Xu, C., Farooq, S., 2018. Ind. Eng. Chem. Res. 57, 1763 - 1766.

10. Farooq, S. and Ruthven, D.M., 1990. AIChE J. 36(2), 310 – 314.

11. Gales, L., Mendes, A., Costa, C., 2003. Chem. Eng. Sci. 58, 5279- 5289.

12. Grande, C.A., Rodrigues, A.E., 2007. Ind. Eng. Chem. Res. 46, 4595 – 4605.

13. He, Y., Krishna, R., Chen, B., 2012. Energy Env. Sci. 5, 9107 – 9120.

14. Kayser, J.C., 1985. Master’s thesis, Ohio State University, Ohio.

15. Kayser, J.C., Knaebel, K.S., 1986. Chem. Eng. Sci. 41, 2931 - 2938.

16. Kayser, J.C., Knaebel, K.S, 1989. Chem. Eng. Sci. 44, 1 – 8.

17. Kikkinides, E. S., Yang, R.T., Cho, S.-H., 1993. Ind. Eng. Chem. Res. 32, 2714 – 2720.

18. Knaebel, K.S., Hill, F.B, 1985. Chem. Eng. Sci. 40, 2351 – 2360.

19. Knaebel, K.S., Reinhold, H.E., 2003. Adsorption 9, 87 – 94.

 29

20. Leinekugel-le-Cocq, D., Tayakout-Fayolle, M., Jallut, C., Jolimaitre, E., 2006. Chem.
Eng. Sci. 61, 633 – 651.

21. Lin, L.-C., Berger, A.H., Martin, R.L., Kim, J., Swisher, J.A., Jariwala, K., Rycroft, C.H.,
Bhown, A.S, Deem, M.W., Haranczyk, M., Smit, B., 2012. Nat. Mat. 11, 633 – 641.

22. Lopez, F.V.S., Grande, C.A., Rodrigues, A.E., 2011. Chem. Eng. Sci. 66, 303 – 317.

23. Myers, A. L., Prausnitz, J.M., 1965. AIChE J. 11, 121 – 127.

24. Ribeiro, R.P., Sauer, T.P., Lopes, F.V., Moreira, R.F., Grande, C.A., Rodrigues, A.E.,
2008. J. Chem. Eng. Data 53, 2311 – 2317.

25. Richter, E., Schütz, W., Meyers, A.L., 1989. Chem. Eng. Sci, 44(8), 1609 – 1616.

26. Ruthven, D.M., 1984. Principles of Adsorption and Adsorption Processes, John Wiley &
Sons, New Jersey.

27. Ruthven, D.M., Ching, C.B., 1989. Chem. Eng. Sci., 44(5), 1011 – 1038.

28. Ruthven, D.M., Farooq, S., Knaebel, K.S., 1994. Pressure Swing Adsorption. John Wiley
& Sons, New Jersey.

29. Ryckebosch, E., Drouillon, M., Vervaeren, H., 2011. Biomass and Bioenergy 35, 1633 –
1645.

30. Schell, J., Casas, N., Marx, D., Mazzotti, M., 2013. Ind. Eng. Chem. Res. 52, 8311 –
8322.

31. Shapiro, A.A., Stenby, E.H., 1998. J. Colloid Interface. Sci. 201, 146 – 157.

32. Simo, M, Brown, C.J., Hlavacek, V., 2008. Comp. Chem. Eng. 32, 1635 – 1649.

33. Simo, M, 2013. Dehydration of Ethanol using Pressure Swing Adsorption, in:
Ramaswamy, S., Huang, H.-J., Ramarao, B.V. (Eds.), Separation and Purification
Technologies in Biorefineries. John Wiley & Sons, New Jersey, pp. 503 – 511.

34. Sircar, S., 1991. Pressure Swing Adsorption-- Research Needs in Industry. In
Fundamentals of Adsorption, Proceedings of Engineering Foundation Conference,
 30

Sonthofen, Germany; Mersmann, A. B. et al., Eds.; Engineering Foundation: New York,

815.

35. Sircar, S., Hanley, B.F., 1995. Adsorption 1, 313 – 320.

36. Sircar, S., 2002. Ind. Eng. Chem. Res. 41, 1389 – 1392.

37. Smith IV, O.J., Westerberg, A.W., 1990. Chem. Eng. Sci. 45, 2833 – 2842.

38. Smith IV, O.J., Westerberg, A.W., 1991. Chem. Eng. Sci. 46, 2967 – 2976.

39. Smith IV, O.J., Westerberg, A.W., 1992. Chem. Eng. Sci. 47, 4213 – 4217.

40. Tagliabue, M., Farrusseng, D., Valencia, S., Aguado, S., Ravon, U., Rizzo, C., Corma, A.,
Mirodatos, C., 2009. Chem. Eng. J. 155, 553 – 566.

41. Tao, L., Xiao, P., Qader, A., Webley, P.A., 2019. Int. J. Greenh. Gas Control. 83, 1 – 10.

42. Teo, W.K., Ruthven, D.M., 1986. Ind. Eng. Chem. Pro. Des. Dev., 25, 431 – 440.

43. Tien, C., 1994. Adsorption Calculations and Modeling, Butterworth-Heinemann, Oxford.

44. Tondeur, D., 1981.In Percolation Processes: Theory and Applications, NATO ASI No. 33,
Rodrigues, A.E. and Tondeur, D, Eds.; Sijthoff and Noordhoff, Alphen van Rijn Holland,
571.

45. Wawrzyńczak, D., Majchrzak-Kucęba, I., Srokosz, K., Kozak, M., Nowak, W., Zdeb, J.,
Smółka, W., Zajchowski, A., 2019. Sep. Pur. Tech. 209, 560 – 570.

46. Yang, R.T., 1997. Gas Separation by Adsorption Processes, vol. 1. World Scientific.
 31

I I kl M
%M 4M I TO IT X

Ykl M
Cas Pressur (SLPM) (L/min)†
(°C (psi) (mol/min (mol/min
e e Ratio (mol/min (mol/min
) (bar) -cycle) -cycle)
-cycle) -cycle)
44.75
0.997 2.342 0.0570
1 45 3 6.476 0.0159 0.0493
2 0.1046 0.0023
3.086
45.46
0.997 2.342 0.1025
2 45 4 7.695 0.0136 0.0516
4 0.1046 0.0042
3.135
45.29
0.997 2.333 0.1606
3 45 7 12.363 0.0084 0.0543
5 0.1042 0.0065
3.123
43.94
0.998 2.314 0.1822
4 45 6 23.538 0.0043 0.0553
0 0.1034 0.0074
3.030
43.85
0.992 2.344 0.2278
5* 45 0 803.305 0.0000 0.0613
7 0.1047 0.0093
3.023
46.60
0.993 1.851 0.1917
6* 60 0 839.889 0.0000 0.0473
4 0.0826 0.0078
3.213
*Bed purged by vacuum alone. Pure component gas used as the low-pressure equilibrium
hold-up reference state.

† Measured at 747 mmHg and 22 °C

 32

Table 1. Inputs to the moving bed model for each ASU case reported by Kayser and Knaebel.
(Kayser, 1985; Kayser and Knaebel, 1986) The HPP purity, PH, and the pressure ratio were
taken directly from (Kayser, 1985). The feed and product flow rates were calculated based on
the reported measurements and are shown in their original and converted units. The sweep
and backfill flow rates were calculated from equilibrium considerations ( I TO and I T X ).
 33

c
]
T D] b d %]
W
AEE
Species RMSE
(mmol/g) (cm3/mmol) (K) (%)
(K)
Oxygen 0.6140 -1450 0.0025 0.34
5.0623 318.15
Nitrogen 1.2080 -1822 0.0073 1.23

Table 2. Temperature-dependent Langmuir isotherm parameters resulting from fitting the

pure component data for oxygen and nitrogen on zeolite 5A. (Kayser, 1985)
 34

a)
 35

b)

Figure 1. Visual comparison of the definitions of the adsorption efficiency ( ), bed utilization
factor ( ), and bed-state efficiency (+) between (a) the work of Smith and Westerberg (Smith
and Westerberg, 1991, 1992) and (b) the current work. The shaded areas have the same total
area as the composition profile, so the number of moles involved is the same.
 36

% %
4 4
9 9
f f
⋮ ⋮

% %
4 4
9 9
f f
⋮ ⋮

Figure 2. Block-flow diagram of a component splitter. Fluid streams are given by solid
arrows while adsorbate-containing streams are denoted by double arrows.
 37

%M
4M
IM YNV,M
YMV,M I O

High Low
Pressure Pressure
Adsorption IN Desorption
%N 4N
NV,N MV,N

IM
%M 4M
NV,M MV,M

% %N
4M 4N
I I IN
YNV, YNV,N
YMV, YMV,N

Figure 3. Moving bed model for separation of a binary mixture using a Skarstrom cycle. The
retentate streams, which represent the material retained in the bed between operational states,
are shown as double arrows.
 38

%M
4M
IM YNV,M
YMV,M I O

I
High Low
Pressure Pressure
Adsorption IN Desorption
%N 4N
NV,N
MV,N

IM
%M 4M
NV,M MV,M

% %N
4M 4N
I YNV, IN YNV,N
YMV, YMV,N

(a)
 39

%M
4M
IM YNV,M
YMV,M

High Low
Pressure Pressure
Adsorption IN Desorption
%N 4N
NV,N MV,N

IM
%M 4M
NV,M MV,M

% %N
4M 4N
I I IN
YNV, YNV,N
YMV, YMV,N

(b)

Figure 4. Flowsheets for simple cycle alternatives to the Skarstom cycle represented by the
moving bed model.

a) Pressurization of the high-pressure splitter by backfill.

b) Purging of the low-pressure splitter by vacuum desorption. Also the flowsheet for self-
purging cycles in kinetic separations.
 40

0.700

0.600
Equilibrium Loading (mmol/g)

0.500

0.400

0.300

0.200

0.100

0.000
0.000 1.000 2.000 3.000 4.000 5.000
Pressure (bar)

(a)
 41

1.400

1.200
Equlibrium Loading (mmol/g)

1.000

0.800

0.600

0.400

0.200

0.000
0.000 1.000 2.000 3.000 4.000 5.000
Pressure (bar)

(b)

Figure 5. Equilibrium loading data for oxygen (a) and nitrogen (b) on zeolite 5A from
(Kayser, 1985). The lines represent the fit of the isotherms to the temperature-dependent
Langmuir expression.

a) Adsorption of oxygen as a function of temperature: T = 30 °C (●), 45 °C (■), and

60 °C (n ).
b) Adsorption of nitrogen as a function of temperature: T = 30 °C (●), 45 °C (■), and
60 °C (n ).
 42

120%

115%
Bed-State Efficiency (%)

110%

105%

100%

95%

90%

85%

80%
6.476 7.695 12.36 23.54 803.3* 839.9*
Pressure Ratio
Oxygen Nitrogen

*Bed purged by vacuum alone. Pure component gas used as the low-pressure equilibrium
hold-up reference state.

Figure 6. Regressed high-pressure bed-state efficiency values for the ASU case study using
the OEM model. Error bars show a 95 % confidence interval. The unit is clearly affected by
some inefficiencies, but since + for both components is relativly close to unity the effects are
not very strong.
 43

100% 100%

80% 80%

Product Purity (%)

LK Recovery (%)

60% 60%

40% 40%

20% 20%

0% 0%
6.476 7.695 12.36 23.54 803.3* 839.9*
Pressure Ratio
Experiment "Ideal" Regressed

*Bed purged by vacuum alone. Pure component gas used as the low-pressure equilibrium
hold-up reference state.

Figure 7. Comparison between the experimental recovery (bars) and product purity (squares)
versus OEM model results using

a) The “ideal” case where +M 1 (open circles, dashed line)

b) The regressed values of +M shown in Fig. 6 (closed circles, dotted line)

The “ideal” case captures the the experimental data qualitatvely, while the data can be
matched exactly if +M ’s are adjusted. Lines serve only as a guide-to-the-eye.