International Journal of Greenhouse Gas Control 62 (2017) 1–12

Contents lists available at ScienceDirect

International Journal of Greenhouse Gas Control

journal homepage: www.elsevier.com/locate/ijggc

Membrane process optimization for carbon capture MARK

a,b b,⁎
Norfamila Che Mat , G. Glenn Lipscomb
a
Department of Chemical Engineering & Energy Sustainability, Faculty of Engineering, Universiti Malaysia Sarawak (UNIMAS), 94300 Kota Samarahan, Sarawak,
Malaysia
b
Department of Chemical Engineering, The University of Toledo, 2801 W. Bancroft, Toledo, OH 43606, United States

A R T I C L E I N F O A B S T R A C T

Keywords: The utilization of membrane technology in post-combustion applications is hindered by low CO2 feed partial
Hollow fiber membrane pressure in the flue gas stream. In order to meet the separation targets of 90% CO2 recovery, 95% or higher CO2
Carbon dioxide purity, and a Levelized Cost of Electricity (LCOE) increase of less than 35%, Membrane Technology and
Post-combustion CO2 capture Research, Inc. (MTR) (Merkel et al., 2010) has proposed the use of the boiler air feed as a sweep stream to
Gas separation membranes
increase the CO2 concentration in the flue gas and partial pressure driving force for permeation without
additional compression or vacuum. Such a design significantly reduces capture cost but leads to a detrimental
reduction in the O2 concentration of the feed air to the boiler. The transport properties and operating pressures
of the membrane stages are optimized in this study. Membrane CO2/N2 selectivity is varied over a broad range
encompassing the values considered by MTR. Membrane CO2 permeability is varied with selectivity according to
the variation anticipated by the upper bound of the Robeson plot for CO2 and N2. Membrane CO2 permeance is
calculated assuming membranes can be fabricated with an effective thickness of 0.1 μm. Additionally, the two
stages may utilize different membrane materials. The feed and permeate pressures also are varied over ranges
encompassing the values proposed by MTR. The optimization space of membrane properties and operating
conditions is scanned globally to determine the process design that minimizes LCOE. The O2 concentration to the
boiler is evaluated during the optimization process and can be used to constrain viable alternatives. The results
indicate a fairly broad range of membrane properties can yield comparable LCOE near the minimum. The
optimal operating pressure range is somewhat narrower. The minimum allowable O2 concentration constrains
viable designs significantly and is critical to process economics.

1. Introduction
In comparison with pre-combustion and oxy-fuel combustion cap-
ture processes, post-combustion carbon dioxide (CO2) capture processes
must handle the lowest CO2 content in the flue gas stream due to the
presence of nitrogen (N2) in air. Because of the low CO2 feed partial
pressure driving force, this technology suffers from a severe energy
efficiency penalty. Therefore, the key to achieve the stringent 90%
recovery and 95% purity targets, without increasing the Levelized Cost
of Electricity (LCOE) by more than 35% (Matuszewski et al., 2012), is to
generate an affordable CO2 feed partial pressure driving force. Mem-
brane processes are one option under development that offer unique
advantages in terms of energy consumption and physical size (Mat
et al., 2014).
Several factors hinder the effective use of membrane processes in
post-combustion applications. The first is the availability of a mem-
brane material that will allow synthesis of a process that satisfies
capture criteria. Increasing CO2 permeability reduces the required
membrane area and associated capital cost while increasing CO2/N2
selectivity reduces the process energy requirements and associated
operating cost. Unfortunately, the literature suggests that material
changes which increase CO2 permeability will lead to a concomitant
decrease in CO2/N2 selectivity. In the context of carbon capture
applications, such a tradeoff also reflects a tradeoff between total
module area and energy requirement. This relationship commonly is
referred to as the Robeson plot (Robeson, 2008) and arises from the
physics of the solution-diffusion process that governs transport in
polymeric materials. Therefore, to move above the upper bound line,
most of the reported CO2 membrane material development work has
sought to improve both CO2 permeability and CO2/N2 selectivity by
using non-polymeric membranes or mixed matrix membranes that are
polymer-inorganic composites (Rezakazemi et al., 2014). Recently,
Roussanaly et al. (2016) used a cost-based comparative numerical
model to identify optimal membrane properties that includes the
impact of membrane technology maturity level and membrane price
sensitivity. While optimal membrane properties may vary with technol-

⁎
Corresponding author.
E-mail address: glenn.lipscomb@utoledo.edu (G.G. Lipscomb).

http://dx.doi.org/10.1016/j.ijggc.2017.04.002
Received 22 November 2016; Received in revised form 4 April 2017; Accepted 6 April 2017
1750-5836/ © 2017 Elsevier Ltd. All rights reserved.
N.C. Mat, G.G. Lipscomb
ogy maturity level, membrane price, and process configurations, the
ability to manufacture high performance, high permeance membranes
is the crucial factor when evaluating the economic competitiveness of
membrane processes with other technologies.
In addition to development of novel membrane materials with
improved separation properties, improvement of existing process de-
signs is essential to minimize LCOE. Because of the low CO2 feed partial
pressure driving force in flue gas, it is impossible for a single stage-
membrane process to achieve the prescribed CO2 recovery and purity
targets even for high CO2/N2 selectivity materials due to pressure ratio
limitations (Merkel et al., 2010). However, the targets can be realized in
multi-stage configurations (Zhao et al., 2008). Thus, implementing
multi-stage designs is required to increase the CO2 feed partial pressure
by creating an internal gas recycle loop. Zhao et al. (2010) describes
two stage configurations with and without permeate vacuum as well as
stage cycle. Results are presented for CO2 recoveries up to 90%. While
the results show the promise of carbon capture with staged membrane
processes, the target of 90% and 95% purity were not achieved for the
process conditions considered. Several other authors have addressed
the analysis of multi-stage membrane configurations and performed
economic feasibility studies for post-combustion applications (Hussain
and Hägg, 2010; Swisher and Bhown, 2014). Most of the reported
multi-stage process configurations for membrane systems have sought
to generate affordable CO2 partial pressure driving forces for permea-
tion by a combination of feed compression, vacuum permeation and
feed-air sweep system (Ho et al., 2008; Merkel et al., 2010;
Ramasubramanian et al., 2012). Combining membranes with other
separation techniques, such as cryogenic flash and absorption, also can
potentially minimize CO2 capture cost (Chowdhury 2011; Belaissaoui
et al., 2012).
Membrane Technology and Research, Inc. (MTR) (Merkel et al.,
2010) proposed a new scheme to increase CO2 partial pressure driving
force in an affordable manner by incorporating an air-feed sweep in a
staged membrane system. The proposed design is a hybrid configura-
tion which utilizes a cryogenic liquefaction and flash step to yield a
liquid CO2 product in the desired purity and recovery (Fig. 1). This
process utilizes MTR’s Polaris membrane which possesses a high CO2
permeance of 1000 GPU and a CO2/N2 selectivity of 50. The high CO2/
N2 selectivity allows production of a CO2 permeate with the purity
Fig. 1. Membrane-cryogenic hybrid configuration proposed by Merkel et al. (2010).
2
International Journal of Greenhouse Gas Control 62 (2017) 1–12
required to meet the final purity target after a cryogenic flash. The high
CO2 permeance reduces the required membrane area to achieve the CO2
recovery target. As shown in Fig. 1, the majority of CO2 re-circulation
arises from the CO2-enriched permeate stream produced by sweeping
the second counter-current stripping stage (module II) with the boiler
feed air. Although this scheme has been successfully demonstrated to
reduce capture cost, it produces an oxygen (O2) deficient feed air
stream to the boiler (the O2 concentration is reduced from 21% to 18%)
that could potentially reduce the boiler adiabatic temperature effi-
ciency (Franz et al., 2013; Scholes et al., 2013). The impact of this
reduction in O2 concentration on LCOE remains unclear.
Recent work on optimization of membrane separation applications
has utilized both stochastic algorithms (Corriou et al., 2008; Yuan et al.,
2014) as well as gradient based methods such as Nonlinear Program-
ming (NLP) and Mixed Integer Non Linear Programming (MINLP) (Qi
and Henson, 2000; Kookos, 2002; Chowdhury, 2011; Scholz et al.,
2015). While past work with gradient based optimization has included
membrane CO2/N2 selectivity and permeability as decision variables
and the effect of multicomponent permeation, the case studies con-
sidered were not representative of utility scale power plants. Moreover,
the past work neglected the initialization strategies in both MINLP and
NLP models. The convergence processes in gradient based methods are
extremely sensitive to the initial points in the model. Thus, a poor
initial guess may lead to suboptimal configurations and process
parameters (Safdarnejad et al., 2015).
Optimization of multi-stage hybrid membrane processes for post
combustion CO2 capture requires systematic evaluation of several
design variables on LCOE. While increasing CO2 membrane perme-
ability results in significant savings in capital cost, it comes with the
tradeoff of lower CO2/N2 selectivity which eventually results in larger
plant parasitic loads. Cost reductions in membrane manufacturing also
may reduce capital cost and reduce LCOE (Roussanaly et al., 2016).
Therefore, to determine the optimum operating conditions and mem-
brane separation properties for post-combustion application, one
should not only consider the balance between required feed to
permeate operating pressure ratio and CO2/N2 selectivity (Huang
et al., 2014), but also the trade-off relationship between CO2/N2
selectivity and CO2 permeability embodied in the Robeson plot.
Here the multistage hybrid membrane-cryogenic air-feed sweep
N.C. Mat, G.G. Lipscomb
configuration in Fig. 1 is optimized according to the Robeson upper
bound. The previously reported operating conditions and membrane
properties are taken as the base design case (Merkel et al., 2010). In this
study, a systematic global search of the design variable space is used to
determine the membrane properties and operating conditions that
minimize LCOE. CO2 permeability is varied with CO2/N2 selectivity
according to the variation dictated by the Robeson plot for CO2 and N2.
The enriching feed and permeate pressure also are varied over ranges
encompassing the values proposed by MTR. Because the O2 reduction in
the boiler feed air that accompanies use of the feed air as a sweep may
reduce boiler efficiency, the O2 concentration to the boiler is evaluated
during the optimization and results are presented for a range of fixed
concentrations. Since the CO2 liquefaction unit operation cost accounts
for ∼10% of overall capture cost (Scholes et al., 2013), cryogenic
process variables are held constant in this work.
2. Methodology
2.1. Base coal-fired power plant descriptions and economics
The optimization study in this work in based on the assumption that
the multi-stage membrane-cryogenic in Fig. 1 is retrofitted in an
existing 550 MW (net) subcritical pulverized coal-fired power plant.
The base plant derating due to inefficiency and internal power
consumption is assumed to be approximately 22 MW. It is assumed
that trace components such as particulate matter and SO2 are removed
from the flue gas through an electrostatic precipitator (ESP) and lime-
based desulfurization (FGD) prior to entering the CO2 capture system.
Thus, the entering flue gas is comprised of four major components: CO2,
N2, O2 and water. The base cost of electricity (COE) prior to carbon
capture is assumed to be $53.96/MWh-net.
The economic metric used in this work is Levelized Cost of
Electricity (LCOE): the price per MWh that utilities must charge to
recoup all expenses plus a desired rate of return (Rubin et al., 2013).
The LCOE calculation used here is based on the guidelines proposed in
the literature (Rubin et al., 2013; Zhai and Rubin, 2013). The plant life
is assumed to be 25 years with a capacity of 85% and interest rate of
10.30%. A unitary membrane module price of $50/m2 is assumed,
where price is independent of membrane properties. The membrane
replacement cost after 5 years is assumed to be $10/m2. All compressor
and expander efficiencies are assumed to be 0.85 while the vacuum
pump efficiency is assumed to be 0.75 with a fixed installation cost of
500/kW/m2. Heat exchangers are assumed installed at a fixed price of
$300/m2. Furthermore, the net electricity produced and sold each year
is assumed constant over the life of the plant as well as constant Fixed
Charge Factor (FCF), fuel and Operating and Maintenance (O & M)
costs. General inflation and real escalation rates also are assumed zero.
All other treated flue gas conditions and base plant data are summar-
ized in Table 1.
2.2. Membrane unit operation model and simulations
Process simulation and optimization was performed using Matlab
R2015a (MathWorks, USA). The Soave-Redlich Kwong (SRK) (Poling
et al., 2001) equation of state was used for thermodynamic calculations.
The solution method for the membrane permeator model is based on a
stage in series approximation to transform the governing differential
mass balances into set of a coupled non-linear algebraic equations
(Coker et al., 1998). Unlike previous work, a direct substitution
algorithm is used to solve the equations instead of the Thomas
algorithm. Input values for the model include the feed flow rate, sweep
flow rate, feed pressure, permeate pressure, sweep pressure, and
membrane properties including fiber dimensions and gas permeances.
The model calculates the retentate and permeate flow rate for both
cross-flow and counter-current contacting.
The multi-component membrane permeator model was validated
3
International Journal of Greenhouse Gas Control 62 (2017) 1–12
Table 1
Base power plant and flue gas data used in this study.
Variable Value
Plant type Subcritical pulverized coal
Coal type Mid-western bituminous coal
Cooling type Wet tower
Gross/Net electrical output (MW) 571.54/550.00
Net plant efficiency (%) 35.01
Operating time (hour/year) 7446
Coal cost ($/MMBtu) 1.80
Main steam pressure (bar) 175
Main steam temperature (°C) 538
Air feed sweep flow rate (MMSCFD) 1300
Flue gas flow rate (MMSCFD) 1514
Flue gas temperature (°C) 51.67
Flue gas pressure (bar) 1.01
Flue gas composition (mole%)
CO2 13.90
N2 73.70
O2 3.10
Water 9.00
with the experimental data reported for CO2 recovery from sour natural
gas (Pan, 1986). Fig. 2 illustrates the recovery of each component in the
permeate stream as a function of stage cut (i.e., ratio of permeate flow
rate to total feed flow rate). The results illustrate the typical tradeoff
between stage cut and purity in membrane enrichment: the concentra-
tion of the most permeable species (the desired product) asymptotically
approaches a maximum at low stage cuts and decreases monotonically
as stage cut increases. As can be seen, the simulation results agree well
with experimental results. The CO2 purity decreases from a maximum
of ∼0.95 with increasing stage cut, so increasing CO2 permeate
concentration comes at the expense of lower CO2 recovery.
Additionally, the base case for the MTR configuration was simulated
and predicted process performance compared to the results reported in
the literature (Merkel et al., 2010). The simulations used identical
membrane properties: CO2 permeance of 1000 GPU, CO2/N2 selectivity
of 50, feed pressure of 2 bar, and permeate pressure of 0.2 bar. Table 2
indicates good agreement between the literature and the present
results. Plant parasitic load is calculated as the ratio of the energy
requirement (i.e., sum of compressor and vacuum energy requirements
less the energy recovered in the expander) to the total plant output of
550 MWh. As found in previous work, the feed compression energy
constitutes the largest portion of the parasitic load due to the large
volume of N2 in the flue gas. Therefore, plant parasitic load is relatively
high in comparison to MEA absorption (Pfaff et al., 2010). Note that the
product produced by the process also affects the energy requirement.
The membrane process considered here produces a cryogenic liquid
while other processes may produce a supercritical fluid.
2.3. Optimization problem formulation
The objective function for optimizing Fig. 1 superstructure is
defined as LCOE and the problem can be stated as follows:
For a given flue gas feed and O2 concentration to the boiler,
determine the process design and operating conditions that minimize
LCOE while fulfilling the capture targets of 90% CO2 recovery and 95%
purity. The cost related parameters are assumed to be known and
membrane separation properties, selectivity and permeability of flue
gas components, are constrained by the Robeson upper-bound empiri-
cal relationship.
The optimization is based on the following assumptions:
1) The flue gas feed streams consist of a multicomponent gas mixture at
known conditions.
2) The pressure drop in both lumen and shell is negligible.
3) The process is conducted at steady and isothermal conditions.
N.C. Mat, G.G. Lipscomb International Journal of Greenhouse Gas Control 62 (2017) 1–12

Fig. 2. Validation of model simulations with experiment. Lines represent model predictions and markers experimental concentration measurements from Pan (1986): diamond – CO2,
square – CH4, triangle – C2H6, and circle – C2H8.

Table 2 PO2
αO2 /N2 =
Comparison of simulation results with previous work (Merkel et al., 2010). Results are for PN2 (4)
a CO2 permeance of 1000 GPU, CO2/N2 selectivity of 50, feed pressure of 2 bar, and
permeate pressure of 0.2 bar. For a specified CO2/N2 selectivity, PCO2 is calculated from Eq. (1)
Variable Merkel et al. (2010) This work
and PN2 from Eq. (2). Combining Eqs. (3) and (4) yields Eq. (5) which
can be used to calculate the O2/N2 selectivity from PN2:
2
Membrane area (MM m ) 1.3 1.32
1
Total power use (MW) 97 97.35 ⎛ 1,396,000 ⎞ 6.6666
Fraction of power plant energy used (%) 16 16.22 αO2 /N2 = ⎜ ⎟
CO2 liquid product concentration (%) 95+ 98.25 ⎝ PN2 ⎠ (5)
O2 concentration (%) in boiler feed air 18 18.03
PO2 then can be calculated from Eq. (3). Assuming membranes can
be fabricated with an effective thickness of 0.1 μm, the permeance for
4) Membrane permeability is calculated using CO2/N2 selectivity each component can be calculated from the permeability. The calcu-
values from the upper bound line of the Robeson plot. O2 perme- lated CO2 permeances are provided in Table 3 for the range of CO2/N2
ability is taken from the upper bound for O2/N2 in the absence of selectivity considered here. Table 3 indicates a CO2/N2 selectivity of 50
another relationship for the permeability. However, any other corresponds to a higher CO2 permeance than that reported for the
relationship could be used. Polaris module. A CO2/N2 selectivity of 75 and CO2 permeance of 1180
5) Water permeance is assumed constant at the value of 10,000 GPU GPU are used as the base case for the comparisons reported here to be
(1 GPU = 1 Gas Permeation Unit = 1 × 10−6 cm3(STP)/s/cm2/cm representative of the Polaris results. Note that the results in Table 2 are
Hg = 7.50 × 10−9 m3(STP)/s/m2/kPa = 3.35 × 10−10 mol/s/ for exactly the same conditions as used previously (Merkel et al., 2010).
m2/Pa). Water permeances are typically so high that permeation is However, the base case for the optimization differs slightly due to the
limited by the pressure ratio and the specific value of permeances restriction of material properties to lie along the upper bound.
does not affect the results. The optimization variables are listed in Table 4. Constraints for this
problem are as follows:
Optimization variables are the enriching and stripping stage module
areas, membrane CO2/N2 selectivity (modules I, II, and III), feed Table 3
compressor outlet pressure and vacuum pump operating pressure. As CO2 permeance values calculated from CO2/N2 selectivity using the
permeability upper bound relationship (Robeson, 2008) and
mentioned previously, cryogenic process variables are held constant in
assuming an effective membrane thickness of 0.1 μm. 1 GPU = 1
this work. Gas Permeation Unit = 1 × 10−6 cm3(STP)/s/cm2/cm
The Robeson upper bound is used to determine the membrane Hg = 7.50 × 10−9 m3(STP)/s/m2/kPa = 3.35 × 10−10 mol/s/
permeability. CO2, O2 and N2 permeability, (PCO2, PO2 and PN2 respec- m2/Pa.
tively) are calculated from the following upper bound relationships:
Selectivity CO2 Permeance (GPU)
PCO2 = 30,967,000(αCO2 /N2 )−2.888 barrer (1) (CO2/N2)

Where αCO2 /N2 is the CO2/N2 selectivity: 150 160

100 519
PCO2 85 830
αCO2 /N2 = 75 1180
PN2 (2) 65 1800
50 3790
PO2 = 1,396,000(αO2 /N2 )−5.666 barrer (3) 45 4840
35 10,700
Where αO2 /N2 is the O2/N2 selectivity:

4
N.C. Mat, G.G. Lipscomb
Table 4
Decision variables used for optimization.
Decision variable Lower and upper limits
Enriching module (I) feed pressure (bar) 2–7.5
(Feed compressor outlet pressure)
Enriching module (I) permeate pressure (bar) 0.2–0.5
(Vacuum pump inlet pressure)
Module I, II and III CO2/N2 selectivity 35–100
Module I, II and III area (m2) 0–10,000,000
Module I, II and III CO2 permeance (GPU) 150–12,000
1) CO2 capture rate from the membrane hybrid system must be ≥90%;
vented CO2 losses must be less than 10%.
2) CO2 purity in the liquefied CO2 must be ≥95%.
Fig. 3 illustrates the optimization procedure used in this study. Prior
to starting the optimization, at the given feed pressure, permeate
pressure and CO2/N2 selectivity, a series of simulations were performed
to determine approximate membrane areas that meet or exceed the CO2
recovery and purity target. These values are used as initial guesses for
determining the area that satisfies the separations target of 90% CO2
recover and 95%+ CO2 purity. Additionally, the optimizations were
conducted for a fixed boiler feed O2 concentration ranging from 17% to
20%. An interior point optimizer is used to determine the membrane
area required to meet the separations target starting from the area
estimates obtained previously. This area is used to determine the LCOE.
The simulation is repeated over the ranges of feed pressure, permeate
pressure and CO2/N2 selectivity considered in this work to determine a
global optimum.
3. Results and discussions
3.1. Impact of enriching and stripping module area variation
As described previously, prior work (Merkel et al., 2010) has
demonstrated that the use of the boiler feed air as a sweep in the CO2
stripping stage can significantly improve process economics. This sweep
increases the CO2 concentration in the feed to the first CO2 enriching
stage by recirculating CO2 through the boiler and increases the partial
pressure driving force for permeation with reduced energy input. The
degree of CO2 recirculation can have a substantial impact on overall
capture cost due to the tradeoff between enriching and stripping stage
Fig. 3. Flowchart for optimization of membrane-cryogenic hybrid process.
5
International Journal of Greenhouse Gas Control 62 (2017) 1–12
areas as well as overall compression power requirements
(Ramasubramanian et al., 2012; Scholes et al., 2013).
The effect of both enriching and stripping stage area on CO2
recovery and purity is determined for the base conditions of feed
pressure = 2 bar, permeate pressure = 0.2 bar, CO2/N2 selectiv-
ity = 75 and CO2 permeance = 1180 GPU which are representative
of past work (Merkel et al., 2010). The results are shown in Fig. 4(a). As
can be seen, the liquid CO2 purity constraint is met over the entire
range of stage areas considered. For a given CO2 recovery, increasing
stripping module area increases the CO2 recirculation rate to the boiler
which increases the driving force for membrane separation and results
in higher CO2 concentration in the enriching stage permeate. Thus,
minimizing enriching module results in slight increases in liquefied CO2
purity.
The effect of enriching and stripping stage area on CO2 recovery,
LCOE and boiler O2 mole fraction is shown in Fig. 4(b). A range of
enriching and stripping stage combinations satisfy the minimum 90%
CO2 capture requirement. Along the 90% capture contour, LCOE and O2
concentration vary dramatically. Maximizing stripping stage area
(minimizing enriching stage area) results in lower LCOE, but the O2
concentration in the boiler feed air decreases undesirably. Conversely,
maximizing enriching stage area gives higher O2 concentrations and
higher LCOE. Reducing the CO2 capture target from 90% would give
higher boiler feed air concentrations for a given LCOE. It is important to
note that LCOE calculated here is based on a fixed compression/vacuum
installation cost of $500/kW. LCOE and may change significantly
depending on actual costs.
The variation of LCOE with boiler feed air O2 concentration along
the 90% recovery contour in Fig. 4(b) is illustrated in Fig. 5. The
decrease in LCOE with O2 concentration is shown for a CO2/N2
selectivity of 75, which was used to generate Fig. 4, as well as other
values. For all values of CO2/N2 selectivity, LCOE decreases as boiler O2
concentration decreases because of an increase in the CO2 concentra-
tion entering the enriching stage. A minimum is not observed as the
boiler concentration decreases. LCOE appears to decrease asymptoti-
cally to a minimum instead. For a given boiler O2 concentration, LCOE
passes through a broad minimum for low CO2/N2 selectivity and
increases dramatically as CO2/N2 selectivity increases above 75. While
increasing CO2/N2 selectivity reduces the compression energy require-
ment (Swisher and Bhown, 2014), it concomitantly leads to an increase
in membrane area because of the lower CO2 permeance dictated by the
Robeson upper bound. The increase in capital costs associated with
N.C. Mat, G.G. Lipscomb
Fig. 4. Impact of enriching and stripping stage area variation for a selectivity of 75, feed pressure = 2 bar, and permeate pressure = 0.2 bar: (a) CO2 recovery (dashed line) and purity
(solid line), (b) LCOE (dash-dot line), O2 concentration in boiler feed air (dotted line), CO2 recovery (dashed line).
membrane area offsets the reduction in operating costs associated with
lower energy requirements.
3.2. Impact of feed and permeate operating pressures
Changing the feed and permeate pressure to 4 and 0.5 bar,
respectively, alters the magnitude of the area requirements for the
enriching and stripping stages. The areas are reduced but the qualita-
tive nature of the variation of CO2 recovery, purity, boiler O2
concentration, and LCOE remain unchanged and the results are similar
to those presented in Fig. 4. Fig. 6 illustrates the relationship between
LCOE and feed boiler O2 concentration along the 90% CO2 recovery
contour for CO2/N2 selectivity of 75 with various operating feed and
permeate pressure combination. The lowest LCOE is found for a feed:
permeate pressure ratio of 4:0.5. Increasing both pressures relative to
base case of 2:0.2 is beneficial as the higher feed pressure reduces the
membrane area requirement and higher permeate pressure does not
have a significant adverse effect on CO2 recovery and purity; gas
6
International Journal of Greenhouse Gas Control 62 (2017) 1–12
liquefaction is robust in its ability to meet the purity requirement from
a broad range of initial concentrations. Note that the maximum boiler
feed air O2 concentration that can be achieved is a function of pressure
ratio as indicated by where the symbols end in Fig. 6. As the feed
pressure increases, the maximum possible O2 concentration also
increases. The highest possible O2 concentration was achieved with a
pressure ratio of 4:0.25, but this is accompanied by an increase in
LCOE. These results suggest that for a given CO2/N2 selectivity, the use
of higher feed and permeate pressures relative to the base case of 2:0.2,
may reduce LCOE while maintaining boiler feed O2 concentrations.
3.3. Influence of feed and permeate pressure variation at fixed CO2/N2
selectivity and boiler O2 mole faction on LCOE
The effect of feed and permeate pressure on LCOE for a CO2/N2
selectivity of 75 is illustrated in Fig. 7(a)–(c) for fixed boiler feed air O2
mole fractions ranging from 0.17 to 0.19. The results in Fig. 7 exactly
satisfy the 90% CO2 recovery target and give liquid CO2 purities of at
N.C. Mat, G.G. Lipscomb International Journal of Greenhouse Gas Control 62 (2017) 1–12

Fig. 5. LCOE variation with boiler O2 mole fraction as a function of CO2/N2 selectivity: square – 100, circle – 75, triangle – 45, and diamond – 35 for 90% CO2 recovery, feed
pressure = 2 bar, and permeate pressure = 0.2 bar. Base electric price without CCS is assumed to be $53.96/MWh.

least 95%. Lines of constant LCOE are indicated numerically and by
contour lines.
Over the range of pressures considered, it was not possible to
determine enriching and stripping stage areas that met the CO2 capture
targets and kept the boiler feed air O2 concentration at the desired
value. Where possible, the capture targets were maintained, otherwise
the O2 concentration was allowed to decrease from the target value and
an artificial cost was added to the LCOE calculation based on the
formula below:
LCOE = LCOEβ + (wβ) (6)
Where LCOEβ is the calculated LCOE for the O2 concentration that
could be achieved, w is a cost adjustment factor, and β is the magnitude
of the difference between the target and achieved O2 concentrations.
The point where this additional cost was added is indicated in the
figure.
As one might expect from the results in Fig. 6, although the
minimum CO2 recovery target of 90% may be met at lower feed
pressure, the desired boiler feed air O2 concentration may not be

Fig. 6. LCOE variation with boiler feed air O2 concentration for a CO2/N2 selectivity of 75 and feed: permeate operating pressures (bar) of: circle – 2:0.2, diamond – 2:0.4, square – 4:0.25
and triangle – 4:0.5.

7
N.C. Mat, G.G. Lipscomb
Fig. 7. LCOE dependence on feed and permeate pressure for a CO2/N2 selectivity of 75 and boiler feed air O2 concentrations of: (a) 17%, (b) 18%, and (c) 19%.
achieved throughout the CO2/N2 selectivity range considered here. This
observation may limit the feasible pressure ranges for the given CO2/N2
selectivity value. Similar to the cases where the O2 concentration could
not be achieved, an artificial cost was added to the LCOE calculation
when recovery was less than 90% based on the formula below:
LCOE = LCOEθ + (wθ)
Where LCOEθ is the calculated LCOE for the recovery that could be
achieved, w is a cost adjustment factor and θ is the magnitude of the
difference between the recovery achieved and the desired value of 90%.
The point where the recovery target could not be satisfied also is
indicated in Fig. 7. The introduction of these artificial costs allowed the
generation of continuous LCOE contours over the entire pressure range
International Journal of Greenhouse Gas Control 62 (2017) 1–12
and increases LCOE sufficiently that the minimum LCOE will corre-
spond only to those conditions that meet CO2 recovery and purity
targets for the given boiler feed air O2 concentration. The latter cost
might be considered a tax on carbon emissions while the former cost
might be the cost of providing auxiliary O2 to the boiler.
A broad minimum in LCOE appears for all boiler feed air O2
(7)
concentrations within the ranges of feed and permeate pressure
examined. As the O2 concentration increases, the minimum LCOE value
increases from less than 70 to ∼71.5. It is interesting to note that the
process proposed by (Merkel et al., 2010) using a feed pressure of 2 bar
and permeate pressure of 0.2 bar lies near the minimum region in
Fig. 7(b) for 18% boiler feed air O2 concentration.
The existence of a minimum in LCOE as feed pressure is varied, for a
8
N.C. Mat, G.G. Lipscomb International Journal of Greenhouse Gas Control 62 (2017) 1–12

fixed permeate pressure, is expected from consideration of operating

and capital costs. For the lowest feed pressure, carbon capture targets
may not be met as seen in Fig. 7. Once the minimum pressure required
to satisfy the CO2 recovery target is reached, the membrane area will be
the largest and energy requirement for compression the smallest. As
pressure continues to increase, the area requirement (capital costs) will
decrease while the energy requirement (operating costs) will increase.
This trade off leads to the emergence of a minimum LCOE as one moves
from left to right along the feed pressure axis for a fixed value on the
permeate pressure axis.
The existence of a minimum in LCOE as permeate pressure is varied,
for a fixed pressure, is less obvious. For low permeate pressures (high
vacuum), the permeate from the enriching stage will be more highly
enriched in CO2 and able to meet the recovery and purity targets with
less CO2 recirculation from the stripping stage through the boiler to the
enriching stage and less stage area. However, the required CO2
recirculation and stage area asymptotically approach limiting values
and the additional energy input into creating a higher vacuum does not
affect performance. As permeate pressure increases (lower vacuum),
the enrichment possible in the enriching stage decreases and more CO2
must be recirculated to satisfy capture targets. Consequently, as
permeate pressure increases, energy costs required to produce the
vacuum decrease but energy costs associated with recirculating CO2
through the boiler (in particular recompression before the enriching
stage) and capital costs associated with stage area increase. The trade-
off between these costs leads to a minimum in LCOE.
Fig. 8(a)–(b) illustrates the variation of LCOE with operating
pressures for a CO2/N2 selectivity of 100 and 45 at 18% boiler feed
air O2 concentrations. Results for O2 concentrations of 17% and 19%
are similar with slightly lower LCOE for 17% and slightly higher LCOE
for 19% as observed in Fig. 7. The boiler O2 concentration is fixed at
18% as this appears to be near the minimum acceptable level without
significant loss of boiler efficiency. As can be seen in Fig. 8(a), a broad
minimum in LCOE is found. The minimum is slightly higher than that
for a CO2/N2 selectivity of 75. The differences are remarkably small
given the assumptions required to perform this analysis. Additionally,
the optimal feed and permeate pressure ranges shift to higher pressures:
from 2.5–4 to 3–5 bar for the feed pressure and from 0.25–0.5+ to
0.3–0.5+ bar. Such shifts are expected to compensate partially for the
lower CO2 permeance of the higher CO2/N2 selectivity membrane and
the higher pressure ratio required to utilize the higher CO2/N2
selectivity effectively.
Results for a CO2/N2 selectivity of 45 are presented in Fig. 8(b). The
contours of constant LCOE are influenced more by an inability to meet
capture and boiler O2 targets than for the higher selectivity materials;
such an observation suggests that lower values of CO2/N2 selectivity
may require additional stages. The minimum values of LCOE are Fig. 8. LCOE dependence on feed and permeate pressure for a CO2/N2 selectivity of 100
slightly smaller than for a CO2/N2 selectivity of 75 as one might expect (a), 45 (b), and boiler feed air O2 concentration of 18%.
from the results in Fig. 5. Moreover, the optimal ranges of pressure are
similar to those for a CO2/N2 selectivity of 75 but slightly narrower, selectivity of 75, using a lower CO2/N2 selectivity membrane for the
especially for the optimal feed pressure. These results indicate the stripping stage reduces LCOE. The optimal operating pressure range
reduction in membrane area costs (capital) afforded by use of a lower also shifts to lower values, especially for the permeate pressure. These
CO2/N2 selectivity membrane are largely offset by an increase in CO2 results are consistent with the expectation of a tradeoff in capital and
recirculation costs (operating) to raise the CO2 concentration in the feed operating costs as membrane CO2/N2 selectivity and CO2 permeability
to the enriching stage – this increase is needed to satisfy CO2 capture are varied along the upper bound.
targets. Fig. 9(c–d) illustrates the changes in LCOE that occur for a fixed
stripping stage CO2/N2 selectivity of 75. Reducing the selectivity to 45
3.4. Influence of feed pressure, permeate pressure, and CO2/N2 selectivity leads to a slightly lower LCOE but a lower limit on the operating
on LCOE for fixed boiler O2 mole fraction pressures that satisfy the capture targets exists. Therefore, as discussed
previously, the use of low CO2/N2 selectivity materials may require
Fig. 9 illustrates the effect of using membranes with different CO2/ additional stages in order to meet separation targets.
N2 selectivities for the enriching and stripping stages. The results Examination of the results presented in Figs. 7–9 indicate the lowest
correspond to 90% CO2 recovery, 95% CO2 purity, and a boiler feed LCOE is found with use of lower CO2/N2 selectivity membranes in the
air O2 concentration of 18%. In all cases, a broad minimum in LCOE is stripping stage and higher CO2/N2 selectivity in the enriching stage.
observed consistent with the results presented in the previous sections. The lowest LCOE ($70/KWh) can be achieved for a feed pressure of
As shown in Fig. 9(a–b), for a fixed enriching stage CO2/N2

9
N.C. Mat, G.G. Lipscomb
Fig. 9. LCOE variation with feed and permeate pressure for a fixed boiler feed air O2 concentration of 18%, and stripping stage CO2/N2 selectivity of 75 (a–b), enriching stage CO2/N2 75
(c–d), and the indicated stripping and enriching stage CO2/N2 selectivity.
∼3 bar and a permeate pressure of 0.3–0.45 bar with any combination
of the 45 and 75 CO2/N2 selectivity materials. This insensitivity
indicates the production of different membrane modules for the
enriching and stripping module stages may not be warranted.
3.5. Sensitivity of LCOE to permeability deviation from Robeson upper
bound
It is important to note that the results presented here are based on
the assumption of using membranes that possess properties along the
Robeson upper bound, i.e., each material possesses an ideal perme-
ability and selectivity combination corresponding to individual points
on the curve. However, real materials may not be found along the
10
International Journal of Greenhouse Gas Control 62 (2017) 1–12
length of the upper bound due to competitive transport or imperfections
introduced during membrane and module manufacture. To address
these concerns, a sensitivity analysis was performed to determine how
LCOE changes with deviation from the upper bound. The analysis was
performed by systematically reducing the CO2 permeability from the
value given by the upper bound for a specified selectivity. LCOE is
determined for each value. Reducing permeability is equivalent to
shifting the entire curve below its current position. The CO2 perme-
ability ratio is defined as the ratio of reduced CO2 permeability to ideal
CO2 permeability and ranges from 0 to 1. The CO2 permeance ratio is
defined in a similar manner and is numerically identical to the
permeability ratio. For each CO2/N2 selectivity, the operating pressure
that minimized LCOE was determined.
N.C. Mat, G.G. Lipscomb International Journal of Greenhouse Gas Control 62 (2017) 1–12

Fig. 10. Sensitivity of LCOE to permeability deviation from upper bound for boiler feed air O2 concentration of 18% and CO2/N2 selectivity of: circle – 45, diamond – 75, square – 100 and
triangle – enriching:75, stripping:45. Results are for identical percentage decreases in CO2, N2, and O2 permeability. CO2 permeance ratio calculated as ratio between actual CO2
permeance and ideal CO2 permeance.

The results of the sensitivity analysis are presented in Fig. 10. For all
values of CO2/N2 selectivity, LCOE increases as the CO2 permeance
ratio decreases. The percentage change in LCOE is less than ∼10% for a
CO2 permeance ratio greater than 0.5 or, equivalently, an actual
permeability greater than one-half of the ideal permeability. However,
increases of ∼50% for a CO2/N2 selectivity of 45 and nearly 400% for a
CO2/N2 selectivity of 100 occur over the range of permeability
reduction considered. While use of high CO2/N2 selectivity results in
a broader optimum operating pressure range, deviation from the upper
bound results in larger LCOE increases. The larger increases arise from
the greater contribution of capital costs (membrane area) to LCOE for
high CO2/N2 selectivity, low CO2 permeability materials.
4. Conclusions
Prior work demonstrates the benefits of CO2 recirculation through
the boiler of a coal fired power plant using a boiler feed air swept
stripping stage in a two-stage membrane based process for CO2 capture.
However, the dependence of process performance on the tradeoff
between CO2/N2 selectivity and CO2 permeability (embodied in the
Robeson upper bound) and process operating pressures is not under-
stood well. In this paper, the dependence of LCOE on membrane
transport properties, constrained along Robeson’s upper bound, and
stage operating pressures is reported and results presented as a function
of the boiler feed air composition. While it is unlikely that real materials
can be identified along the entire length of the upper bound, such
materials would give the best performance within the solution-diffusion
transport controlled class of materials. Identifying whether increasing
selectivity is preferred to increasing permeability (or vice versa) would
help direct material research for carbon capture.
The process design space of membrane CO2/N2 selectivity and
operating pressures was searched globally to determine the process
design parameters that minimize LCOE. Results are presented for a
range of fixed boiler O2 concentrations. A broad LCOE minimum is
found in virtually all cases. The minimum of 69–70$/Kwh meets the
DOE target of less than a 35% increase in electricity cost. This
observation recommends focusing not on improving transport proper-
ties as much as identifying a material that can used to produce
membranes and modules at the lowest cost. However, it is important
to note that the LCOE calculated in this work is based on a set of
economic assumptions that may change if cost estimates are refined
further.
Small reductions in LCOE are achievable by using lower CO2/N2
selectivity materials. These materials possess higher CO2 permeances
which reduce capital costs by reducing the required membrane area.
The capital savings are offset in part by an increase in operating costs
required to achieve capture targets but the net effect is a slight decrease
in LCOE.
The optimal operating pressures are slightly higher than proposed
originally (Merkel et al., 2010). This is especially important for the
permeate vacuum pressure as a value of 0.2 bar pushes the limits of
current vacuum technology. Operating at pressures closer to 0.5 bar
may be desirable and appears feasible based on this work.
Simulations with different materials in the enriching and stripping
stags indicate that LCOE can be minimized with the same material in
both stages. Values of CO2/N2 selectivity from 45 to 75 lead to nearly
identical results. This suggests a manufacturer will not have to produce
different products for the enriching and stripping stages thereby
reducing manufacturing complexity and cost.
The minimum LCOE decreases as boiler feed air O2 concentration
decreases. However, one expects boiler performance to be compro-
mised if the concentration is too low, and the impact on process
economics would need to be included in the optimization. Additional
work is needed to establish this lower limit and its effect on LCOE.
The relative insensitivity of the LCOE to O2 concentration in the
boiler feed air is due in part to the insensitivity of the total required
membrane area to the size of the enriching and stripping stages.
Fig. 5(b) indicates that the total area varies from ∼1.3 to 1.4 million
m2 as the feed O2 concentration varies from 16% to 19%. Additionally,
the optimal operating pressures are nearly identical.
A sensitivity analysis of the results to deviation from the upper
bound indicates the results do not change significantly for permeability
reductions of up to ∼50%. Larger deviations have a greater impact on
LCOE and will limit the viability of membrane processes if they cannot

11
N.C. Mat, G.G. Lipscomb
be avoided.
It is important to note that the results presented here assume fixed
downstream cryogenic process parameters as the literature suggests the
cryogenic unit accounts for only ∼10% of overall capture cost (Scholes
et al., 2013). One might expect the minimum LCOE values to vary by
similar amount if robust optimization of cryogenic operations is
performed
Acknowledgements
The authors would like to acknowledge UNIMAS and Ministry of
Higher Education Malaysia for providing funding through PhD fellow-
ship program.
References
Belaissaoui, B., Le Moullec, Y., Willson, D., Favre, E., 2012. Hybrid membrane cryogenic
process for post-combustion CO2 capture. J. Membr. Sci. 415–416, 424–434.
Chowdhury, M.H.M., 2011. Simulation, Design and Optimization of Membrane Gas
Separation, Chemical Absorption and Hybrid Processes for CO2 Capture. University
of Waterloo.
Coker, D.T., Freeman, B.D., Fleming, G.K., 1998. Modeling multicomponent gas
separation using hollow-fiber membrane contactors. AIChE J. 44 (6), 1289–1302.
Corriou, J.-P., Fonteix, C., Favre, E., 2008. Optimization of a pulsed operation of gas
separation by membrane. AIChE J. 54 (5), 1224–1234.
Franz, J., Schiebahn, S., Zhao, L., Riensche, E., Scherer, V., Stolten, D., 2013.
Investigating the influence of sweep gas on CO2/N2 membranes for post-combustion
capture. Int. J. Greenh. Gas Control 13, 180–190.
Ho, M.T., Allinson, G.W., Wiley, D.E., 2008. Reducing the cost of CO2 capture from flue
gases using membrane technology. Ind. Eng. Chem. Res. 47 (5), 1562–1568.
Huang, Y., Merkel, T.C., Baker, R.W., 2014. Pressure ratio and its impact on membrane
gas separation processes. J. Membr. Sci. 463, 33–40.
Hussain, A., Hägg, M.-B., 2010. A feasibility study of CO2 capture from flue gas by a
facilitated transport membrane. J. Membr. Sci. 359 (1–2), 140–148.
Kookos, I.K., 2002. A targeting approach to the synthesis of membrane networks for gas
separations. J. Membr. Sci. 208 (1–2), 193–202.
Mat, N.C., Lou, Y., Lipscomb, G.G., 2014. Hollow fiber membrane modules. Curr. Opin.
Chem. Eng. 4, 18–24.
Matuszewski, Michael, Ciferno, Jared, Chen, Scott, 2012. Research and Development
Goals for CO2 Capture Technology. NETL Report DOE/NETL-2009/1366. .
Merkel, T.C., Lin, H., Wei, X., Baker, R., 2010. Power plant post-combustion carbon
12
International Journal of Greenhouse Gas Control 62 (2017) 1–12
dioxide capture: an opportunity for membranes. J. Membr. Sci. 359 (1–2), 126–139.
Pan, C.Y., 1986. Gas separation by high-flux, asymmetric hollow-fiber membrane. AIChE
J. 32 (12), 2020–2027.
Pfaff, I., Oexmann, J., Kather, A., 2010. Optimised integration of post-combustion CO2
capture process in greenfield power plants. Energy 35 (10), 4030–4041.
Poling, B.E., Prausnitz, J.M., O’Connell, J.P., 2001. The Properties of Gases and Liquids.
McGraw-Hill.
Qi, R., Henson, M.A., 2000. Membrane system design for multicomponent gas mixtures
via mixed-integer nonlinear programming. Comput. Chem. Eng. 24 (12), 2719–2737.
Ramasubramanian, K., Verweij, H., Winston Ho, W.S., 2012. Membrane processes for
carbon capture from coal-fired power plant flue gas: a modeling and cost study. J.
Membr. Sci. 421–422, 299–310.
Rezakazemi, M., Ebadi Amooghin, A., Montazer-Rahmati, M.M., Ismail, A.F., Matsuura,
T., 2014. State-of-the-art membrane based CO2 separation using mixed matrix
membranes (MMMs): an overview on current status and future directions. Prog.
Polym. Sci. 39 (5), 817–861.
Robeson, L.M., 2008. The upper bound revisited. J. Membr. Sci. 320 (1–2), 390–400.
Roussanaly, S., Anantharaman, R., Lindqvist, K., Zhai, H., Rubin, E., 2016. Membrane
properties required for post-combustion CO2 capture at coal-fired power plants. J.
Membr. Sci. 511, 250–264.
Rubin, E.S., Short, C., Booras, G., Davison, J., Ekstrom, C., Matuszewski, M., McCoy, S.,
2013. A proposed methodology for CO2 capture and storage cost estimates. Int. J.
Greenh. Gas Control 17, 488–503.
Safdarnejad, S.M., Hedengren, J.D., Lewis, N.R., Haseltine, E.L., 2015. Initialization
strategies for optimization of dynamic systems. Comput. Chem. Eng. 78, 39–50.
Scholes, C.A., Ho, M.T., Wiley, D.E., Stevens, G.W., Kentish, S.E., 2013. Cost competitive
membrane—cryogenic post-combustion carbon capture. Int. J. Greenh. Gas Control
17, 341–348.
Scholz, M., Alders, M., Lohaus, T., Wessling, M., 2015. Structural optimization of
membrane-based biogas upgrading processes. J. Membr. Sci. 474, 1–10.
Swisher, J.A., Bhown, A.S., 2014. Analysis and optimal design of membrane-based CO2
capture processes for coal and natural gas-derived flue gas. Energy Procedia 63,
225–234.
Yuan, M., Narakornpijit, K., Haghpanah, R., Wilcox, J., 2014. Consideration of a nitrogen-
selective membrane for postcombustion carbon capture through process modeling
and optimization. J. Membr. Sci. 465, 177–184.
Zhai, H., Rubin, E.S., 2013. Techno-economic assessment of polymer membrane systems
for postcombustion carbon capture at coal-fired power plants. Environ. Sci. Technol.
47 (6), 3006–3014.
Zhao, L., Riensche, E., Menzer, R., Blum, L., Stolten, D., 2008. A parametric study of CO2/
N2 gas separation membrane processes for post-combustion capture. J. Membr. Sci.
325 (1), 284–294.
Zhao, L., Riensche, E., Blum, L., Stolten, D., 2010. Multi-stage gas separation membrane
processes used in post-combustion capture: energetic and economic analyses. J.
Membr. Sci. 359 (1–2), 160–172.