Professional Documents
Culture Documents
Keywords: Production routes from renewable resources, such as biohydrogen fermentation, are potentially process options
Membrane process for a sustainable hydrogen supply. In such fermentation-based processes, CO2 is produced as a byproduct and
Pressure swing adsorption must be removed before hydrogen utilization. Membrane and pressure swing adsorption (PSA) processes are
Hybrid process viable separation methods with high energy efficiencies. Membrane processes are particularly advantageous for
Process optimization
bulk separation but tend to be less cost-effective as a stand-alone separation method if high product purities are
Biohydrogen separation
required. In comparison, PSA processes can achieve high purities but tend to suffer from low recoveries for bulk
separation. Combining and integrating these technologies allows benefiting from both their advantages.
However, the design and integrated optimization of such hybrid processes are challenging. Thus, we present an
optimization methodology based on mixed-integer-nonlinear programming (MINLP) to identify the optimal
process design and to exploit the synergies of these separation methods fully. The process model comprises a
short-cut model for the PSA unit, a discretized model for the membrane unit as well as economic correlations.
The biohydrogen purification costs are in the range of 0.60 €/kg and 1.86 €/kg for a H2 feed fraction of 75 mol-%
and 25 mol-%, respectively. Furthermore, we show that the optimal process design, such as the process structure,
the optimal operation pressure and the membrane area, depends significantly on the feed composition. The
methodology can easily be adapted to other hybrid membrane-PSA processes.
1. Introduction promising energy carrier for the future [4] and is also an important
feedstock for the chemical industry [5], e.g. for the ammonia and me-
A sustainable energy and feedstock supply remains one of the major thanol production. Currently, it is mainly produced from fossil re-
challenges of today’s industrial society [1–3]. Hydrogen (H2) is a very sources such as coal and natural gas [6]. For a more sustainable future,
⁎
Corresponding author.
E-mail addresses: Manuscripts.CVT@avt.rwth-aachen.de, matthias.wessling@avt.rwth-aachen.de (M. Wessling).
https://doi.org/10.1016/j.cej.2019.01.136
Received 8 October 2018; Received in revised form 21 January 2019; Accepted 22 January 2019
Available online 24 January 2019
1385-8947/ © 2019 Published by Elsevier B.V.
B. Ohs et al. Chemical Engineering Journal 364 (2019) 452–461
alternative production routes must be developed [7]. Many options, hybrid PSA-processes [55] which optimizes the recovery of the process.
such as electrochemical [8] or photocatalytic water splitting [9,10], are In contrast, hydrogen from biogenic origin is available at ambient
currently heavily investigated. Alternatively, H2 can also be produced pressure and temperature only. Thus, the compression costs for the
by fermentation processes in which renewable feedstocks, such as waste separation are critical and require the pressure levels to be included in
streams, can be used [11–14]. This route allows combining waste the optimization.
treatment and energy production. Unfortunately, the H2-fermentation is To address these challenges, we develop a semi-physical, but precise
inhibited by the produced H2 [15–17]. Thus, methods to reduce the H2 short-cut-model for the pressure swing adsorption unit. With this short-
partial pressure such as vacuum operation or nitrogen sparging have cut model at hand, optimization based on mixed-integer-nonlinear
been investigated [18–21]. However, these methods decrease the en- programming (MINLP) carries the potential to identify the optimal
ergy efficiency of the biohydrogen production significantly because the design of a hybrid membrane pressure swing adsorption process de-
H2 purification becomes challenging at high degrees of dilution [22]. pending on the pressure conditions chosen. To showcase the new
Furthermore, in such a process CO2 is produced as well, which must be methodology, we consider the separation of H2 and CO2 from biogenic
removed before H2 utilization. Processes based on polymeric mem- origin. Three different feedstocks with varying H2 contents are con-
branes [23–27], membrane contactors [28,29] as well as electro- sidered (H2 feed fraction of 25, 50 and 75 mol-%). In contrast to fossil
chemical methods [30] have been investigated for H2 upgrading. As the fuel-based hydrogen sources and their related hybrid membrane-PSA
separation of the product gas is energy consuming and costly, it is processes for H2/CO2 separation, the biogenic feed stream is present at
crucial to develop an efficient process to make biohydrogen production ambient pressure and temperature. This makes membrane processes
economically attractive. more attractive for the bulk separation. Furthermore, the process design
Membrane processes have already been commercialized for many varies from conventional hydrogen separation processes as pressuriza-
gas separation applications, such as natural gas treatment, air separa- tion of the feed stream is required, which causes additional costs. So far,
tion [31] and biogas upgrading [32]. In comparison to conventional research on biohydrogen separation mainly focuses on hydrogen en-
separation tasks, they show many benefits such as low energy con- richment up to 90% [23,27,56,57]. Yet, a design study for the pro-
sumption [33], small footprints [34] as well as the possibility of flexible duction of highly-pure hydrogen from biogenic origin and, in parti-
operation [35]. Unfortunately, multi-stage processes are required to cular, hybrid membrane-PSA processes has not been presented yet.
achieve both high purities and recoveries [36,37]. In such processes,
recompression is one of the primary cost drivers and optimizing 2. Optimization approach
membrane processes is crucial to balance compression costs, membrane
costs and recovery [31]. Due to the large variety of possible process As discussed above, developing hybrid membrane PSA processes is
configurations, developing the optimal process design represents a challenging. To meet this challenge, we present an optimization fra-
challenging task. Recently, advanced optimization methods have been mework for hybrid membrane-PSA processes using mixed-integer-non-
developed to find optimal process configurations and conditions for linear programming. The optimization model consists of [39]:
multi-stage membrane processes [38–42].
Hydrogen utilization often requires high purities [43,44], which • A process structure with connect-options to allow for different
usually results in high costs using stand-alone membrane processes process configurations, e.g. recycle points [58],
[45]. In comparison, pressure-swing adsorption (PSA) processes are • Physical equations describing the various process units, i.e. com-
well known and widely used for the production of highly pure hydrogen pressors, membranes and PSA,
[46,47], but only exhibit low recoveries when bulk streams are sepa- • Cost correlations and economic calculations, e.g. investment costs
rated [48]. Consequently, hybrid separation processes have been in- and present value calculations,
vestigated to combine the advantages of both separation methods [49]. • Technical boundary conditions, e.g. minimal required recovery and
It has been shown that hybrid processes outperform both stand-alone purity,
PSA as well as stand-alone membrane processes [48,50–52]. However, • Process and cost parameters, e.g. feed flow rate and H price.2
the design of such hybrid processes is intricate and challenging as the
PSA unit gives additional degrees of freedom. Two methods to develop With this framework, we can evaluate a large number of potential
hybrid membrane-PSA separation processes are commonly applied: process configurations and identify the best design resulting in the
lowest costs. So far, many optimization studies often focus on identi-
• Black-box approach: The PSA unit is modeled as a simple splitter fying the process design with the lowest energy demand. However, we
[48,53,54]. Particularly, this approach is used for intricate produc- think that identifying the economic optimum is of higher practical re-
tion systems combining many operations steps. levance as the energy demand, which can often be reduced by pro-
• Iterative approach by subsequent simulation of the two process viding more membrane area and/or adsorber size.
steps: In a first step, a membrane process is simulated. Subsequently,
the PSA process is calculated using the outflow of the membrane 2.1. Process model
unit as the inflow. Often, the tail gas of the PSA process is recycled
to the membrane process and then the membrane process must be Fig. 1 provides a schematic of the generic flow sheet. It represents
simulated anew. The subsequent simulation of the two process steps all potential process configurations considered in this study. For the
is repeated until convergence is achieved [50]. bulk separation, we consider a two-stage membrane process using CO2-
selective membranes (see stage 1 and stage 2 in Fig. 1). This type of
As membrane and PSA processes are complementary separation membranes typically exhibits higher selectivities and permeabilities
steps each with its advantages and disadvantages, it is essential to ex- compared to H2 selective membranes. Additionally, for CO2 selective
ploit their synergies fully. For the black-box approach presented above, membranes, the hydrogen remains on the high-pressure side which
detailed information about the PSA process is missing. For instance, the avoids repressurization of the hydrogen-rich stream for the PSA unit,
performance of the PSA process depends on the pressure level of the which produces highly-pure hydrogen. Furthermore, recycling the tail
membrane process and the particular feed composition. Thus, finding gas of the PSA unit may increase the overall recovery of the process
the optimal process design with a simple splitter function is impossible. [50]. The recycle stream can also pass a third membrane unit. The
In comparison, the second approach is very time and calculation in- optimization model then identifies the best process configuration, i.e.
tensive since it is challenging to investigate all potential process con- the process configuration resulting in the lowest cost, and answers the
figuration. Nikolić et al. presented an optimization framework for following questions:
453
B. Ohs et al. Chemical Engineering Journal 364 (2019) 452–461
Feed
Pressure
PSA
Product
permeance of 1000 GPU and 85 GPU, respectively, are used [59]. The
Area
Volume Waste permeance and selectivity of the membrane may be influenced by a
Stage 1
high CO2 partial pressure as well as the presence of minor components
Waste
Pressure [59]. However, in this design study these effects have not been taken
Area into account, but should be investigated in subsequent studies.
Stage 2 Waste
Pressure
2.2.2. Semi-physical short-cut model for pressure swing adsorption unit
Area Due to their dynamic character, it is very challenging to simulate
Stage 3 Waste
PSA processes, which is even more difficult for hybrid processes as the
complexity further increases. To circumvent this problem, we devel-
Fig. 1. Schematic representation of the process model. The optimization model oped a semi-physical short-cut model for the PSA unit, which we de-
identifies the process structure, the optimal membrane area, the optimal ad- rived from a rigorous PSA process simulation at cyclic steady state
sorber size, recycle locations and system pressure. conducted in Aspen Adsorption™. The rigorous PSA process uses a state-
of-the-art eight step adsorption cycle (see Fig. 3). The longer cycle steps
• To which degree does the separation take place in the membrane (adsorption & supply, purge pressurization, regeneration, purge) have a
duration of 100 s. In comparison, the shorter cycle steps (adsorption &
and in the PSA unit, respectively?
• Which membrane unit is required and what is its optimal membrane feed pressurization (AD & FP), blowdown (BD), pressure equalization
(PE)) are only 30 s long.
area?
• What is the optimal system pressure? For the rigorous simulation of the adsorption process conducted in
• Where are the optimal recycle locations and what are the recycle Aspen Adsorption™, we assumed the following:
rates?
• What is the optimal size of the adsorption bed? • the gas phase obeys the ideal gas law,
• Where is the waste stream purged out of the system? • no radial variation in gas concentration, temperature and pressure
exists,
2.2. Unit operations • the axial pressure drop in the column is described by the Ergun
equation,
In this work, the hybrid separation process comprises three process • the porosities of both the adsorption bed and the adsorbent particles
entities: compressors, membrane units and a PSA unit. We present the are uniform along the bed,
governing equations in the following. • the inter-phase mass transfer coefficient is described by the diffusion
pore model,
2.2.1. Membrane module • the adsorption isotherm is described by the extended Langmuir
In this study, we implemented a discretized model, which is ap- model
plicable for global optimization, to describe the gas separation within a
membrane module [39–41]. The following equation describes the local The governing equations for the cyclic adsorption are given in [61].
flux along a counter-current hollow fiber membrane module: As discussed above, the simulation of PSA is challenging and in-
corporating dynamic PSA models into integrated process simulation is
ni, j = Qi ·(pR, i, j pP, i, j ) (1) almost impossible. Thus, Chung et al. proposed a short-cut model for
with ni, j as the local flux of component i through the membrane within adsorption-based CO2 separation [62]. However, the short-cut model
the discretization element j, Qi as the permeance of the membrane for overestimates the PSA performance and was designed for a process in
component i and pR, i, j as well as pP , i, j as the partial pressure of com- which the adsorbate represents the product [62]. Based on the work of
ponent i within the discretization element j on the retentate and the Chung et al., we developed a novel short-cut model for the cyclic steady
permeate side, respectively. The molar balance for both the retentate state that is specifically designed to describe processes in which the
and permeate is then described by (see also Fig. 2) [39–41]: non-adsorbed component represents the product. Furthermore, the
model also takes non-ideal effects such as mass-transfer zones into ac-
nR, i, j = n i, j · W · z (2) count. A schematic principle of the novel model is given in Fig. 4. The
and model divides the PSA cycles into an adsorption and a desorption step
denoted by the subscripts ‘ad‘ and ‘de‘, respectively. For both steps, the
nP, i, j = n i, j · W · z (3) equilibrium between the gas and the solid phase is assumed. At the
where nR, i, j and nR, i, j denote the flow rate at the discretization element j beginning of the adsorption step, the packed-bed is in the state of
on the retentate and permeate side, respectively, W the total width of desorption, wherein ng, de is the amount of gas present in the gas phase
the membrane module, i.e. the sum of the circumference of all fibers, of the adsorption column and ns, de represents the amount of gas ad-
and z the length of the discretization element. In this work, ten dis- sorbed on the solid phase. During the adsorption step, the feed nf is fed
cretization elements and polymeric membranes with a CO2 and H2 into the column. The amount of gas molecules present in the gas phase
at the end of the adsorption step is denoted by ng, ad and the amount of
adsorbed molecules by ns, ad . The amount nout , which is withdrawn
j j+1
during the adsorption step, is split into the amount used for purging
Feed Retentate npurge and the amount representing the product nprod . The overall mass
balance for the adsorption step is thus given by:
Membrane
nf + ng , de + ns, de = ng, ad + ns, ad + nout (4)
Permeate and for component i:
nf , i + ng , de, i + ns, de, i = ng, ad, i + ns, ad, i + nout , i for i: N 1 (5)
z
with N as the number of components in the system. Equivalently, at the
beginning of the desorption step the amount of molecules in the gas
Fig. 2. Schematic representation of the discretized membrane unit. phase is ng, ad and the amount of gas molecules adsorbed is ns, ad . nres is
454
B. Ohs et al. Chemical Engineering Journal 364 (2019) 452–461
Fig. 3. Pressure trajectories during the eight steps of the adsorption cycle. Here AD stands for adsorption, FP for feed pressurization, DP for depressurization, BD for
blowdown and PE for pressure equalization. The scheme was adapted from [60].
the amount of gas leaving the column as tail gas. Additionally, npurge is yg, de, i = yfeed, i + (yprod, i yfeed, i )· de (11)
fed into the column during desorption to improve regeneration of the
with yg, de, i as the molar fraction of component i in the gas phase at the
column. This leads to the following equation for the overall mass bal-
end of the desorption step. The fitting parameters were obtained by
ance for the desorption step:
simulating the PSA unit rigorously for a wide range of feed composi-
npurge + ng, ad + ns, ad = ng , de + ns, de + nres (6) tions (70, 80 and 90 mol-%) and adsorber volumes using Aspen Ad-
sorption™. For all simulations, the product flow was fixed and the re-
and for component i:
quired feed adsorption pressure, the feed flow needed and the resulting
npurge, i + ng, ad, i + ns, ad, i = ng, de, i + ns, de, i + nres, i for i: Ni 1 (7) product recovery were calculated. Based on the sensitivity study with
the rigorous PSA model, a regression analysis was performed to de-
The equilibrium of the gas and adsorbent phase is described by the termine the fitting parameters. The resulting fitting parameters are
Langmuir equation (it is assumed that only CO2 is adsorbed): utilizable for a wide range of adsorption pressures and feed composi-
bCO2 ·qmax, CO2 · pCO2 tions. Fig. 5 depicts the method to develop the short-cut model.
qCO2 = Fig. 6(a) and (b) compare the adsorber volume per molar flow of
1 + bCO2 ·pCO2 (8) product as well hydrogen recovery for the rigorous and the short-cut
with qCO2 as the CO2 loading of the adsorbent. The two parameters bCO2 model for a feed fraction of 80%. The comparison underlines that the
and qmax, CO2 represent the isothermal constants of the Langmuir model. short-cut model with the fitting parameters (see Table 1) can describe
Delgado et al. discussed adsorption materials for the separation of many the adsorption process precisely.
gas mixtures and showed that BPL 4X10 activated carbon is attractive
for H2/CO2-separation [63]. Thus, we used this material in this work. 2.2.3. Compressor
The two parameters bCO2 and qmax, CO2 are 0.4 bar 1 and 7.07 mol/kg, For the hybrid separation process, compressors are used to generate
respectively. To account for non-ideal effects, such as limitations by the driving force for the membrane units and the adsorption pressure.
kinetics, three fitting parameters ( V , ad , ad ) are incorporated in the To reduce the calculation effort, we assumed multi-stage compressors
455
B. Ohs et al. Chemical Engineering Journal 364 (2019) 452–461
Table 1
Rigorous
model
Fitting parameters for the PSA short-cut model.
Fitting parameter Value
ad 0.18
Sensitivity study
de −2.12
V 1.10
Parameter fitting
flow rate, R the ideal gas constant, T the temperature of the gas stream
and p0 as well as pN the inlet and outlet pressure of the compressor,
Short-cut respectively.
model
2.3. Economics
Fig. 5. Method for developing the short-cut model.
In this study, we systematically investigated the optimal design of
15.0 hybrid membrane-PSA processes for H2 purification. For these pro-
cesses, it is challenging to balance the multiple trade-offs, e.g. plant size
versus energy costs for compression. In principle, two different optima
Adsorption pressure [bar]
12.5 Dynamic model can be distinguished: (i) the economic and (ii) the energetic optimum.
However, the economic optimum is of higher practical relevance as the
energy demand can often be reduced by providing more membrane
10.0 area and/or a higher adsorber volume. Since only the separation step of
the biohydrogen production is considered, it is more appropriate to
7.5 reduce the separation costs instead of maximizing the profit of the
process. By taking the hydrogen loss into account, the benefit of the
process is indirectly optimized as well. Thus, we minimized the specific
Short-Cut model purification costs to balance the multiple trade-offs of the separation
5.0
process, e.g. plant size versus energy costs for compression. Therefore,
the investment costs, i.e. costs for membrane, adsorption bed and
2.5 compressor, were annualized. The operation costs entail the energy
0.0 0.1 0.2 0.3 0.4 costs for the compression, the membrane replacement, the adsorbent
Adsorber volume [m3/mol/s] replacement as well as opportunity costs for product losses in the waste
streams. The equivalent annual costs EAC are defined as follows:
ITot
95 EAC = C +
AtPlant, r (13)
Short-Cut model with C as the annual operation costs, ITot as the total investment costs,
Hydrogen recovery [%]
tPlant as the plant lifetime, r as the interest rate and A tPlant, r as the annuity
90 factor:
1
Dynamic model 1 (1 + r )tPlant
AtPlant , r =
r (14)
The specific costs cSpec are then calculated using the product mass
85
stream mProd and the annual operation time tOpe :
EAC
cSpec =
mProd ·tOpe (15)
1 pN with AMem, tot as the total membrane area to be installed and pMem as the
PCom = · n· R·T · ln specific membrane price. In this work, the PSA unit process consists of
Com p0 (12)
four pressure vessels operating in parallel, but with shifted cycle
where Com is the overall efficiency of the compressor, n is the molar schedules. The investment costs of the PSA unit IPSA are calculated by
456
B. Ohs et al. Chemical Engineering Journal 364 (2019) 452–461
adding the costs for the pressure vessels IVessel to the costs of the ad- range or product purity. x is the vector of the optimization variables,
sorption material IAdorbent : e.g. the system pressure, the membrane area for each stage or the ad-
sorber volume, and y the vector of integer variables which describe the
IPSA = (IVessel + IAdsorbent )· nPSA (19)
connections between the operation units. We implemented the opti-
with nPSA as the number of adsorber beds. The costs for the pressure mization model including the models for the unit operations as well as
vessels are calculated by the Guthrie method [64]: the processes interconnections in the General Algebraic Modeling
IVessel = UF · BC·(MPF + MF 1)·CF (20) System (GAMS). The Antigone solver (Algorithms for coNTinuous/
Integer Global Optimization of Nonlinear Equations) was used for
with UF as the update factor, BC the base costs, MPF as the material global optimization.
pressure correction and MF as the module factor and CF as the cost
factor to account for the valves in connections of the PSA unit. The base
costs are calculated as follows: 3. Case study
w e
L D Various feedstocks and production paths can be used for biogenic
BC = C0· ·
L0 D0 (21) hydrogen production [12,65]. Thus, the gas composition can vary sig-
Here C0, L0 and D0 are the base costs as well as the reference length nificantly. Furthermore, the H2 yield of the fermentation process can be
increased by nitrogen purging because a high H2 partial pressure in-
and diameter according to the Guthrie method [64]. w and e denote the
length and diameter exponent, respectively. The investment costs of the hibits the biohydrogen production. However, nitrogen sparging in-
creases the costs and the difficulty to purify H2. Thus, we consider a
adsorbent material are calculated as follows:
process without nitrogen sparging in this work. Besides H2 and CO2,
IAdsorbent = VAdsorber ·(1 Bed )· Sorbent ·CAds (22) biohydrogen streams also contain smaller amounts of H2S, CH4, CO, N2
with VAdsorber as the volume of the adsorption columns, Bed as the bed and water [66]. H2S and water permeate faster through the membrane
porosity and Sorbent as the adsorbent density. The sum of the energy than H2 and thus concentrate in the CO2 waste stream [67]. Depending
costs for compression CCom , the membrane replacement costs CMem , the on further utilization of the CO2 stream, subsequent removal of H2S and
adsorbent service costs CPSA as well as lost revenues due to incomplete water might be necessary [67]. Alternatively, a dehumidifier can re-
recovery CH2, loss determines the operating costs: duce the water content of the feed stream. Furthermore, CH4, CO and
N2 more favorable adsorb on activated carbon [68], which allows to
C = CCom + CMem + CPSA + CH2, loss (23) reach the targeted product purity. Thus, we only consider a binary gas
The energy costs are the product of the compressor power and elec- mixture for simplification. In subsequent studies, the model can be
tricity price pEnergy : easily extended to account for additional components. However, an
increased computational effort for the global optimization has to be
CCom = PCom· pEnergy (24) taken into account. To represent different biohydrogen feedstocks, we
For simplification, the membrane replacement costs are calculated by considered a wide range of H2 feed fractions (25%, 50% and 75%).
multiplying the membrane investment costs IMem by the ratio of the Depending on the utilization, different hydrogen purities are re-
plant lifetime tPlant and the lifetime of the membrane tMem : quired. For example, heat treatment of metals, power generator cooling
and hydrogen distribution systems require low grade hydrogen [69,70].
tPlant
CCom = IMem· Thus, in this work, we target a product purity of 99% [70]. This pre-
tMem (25) sents a product purity which can be reached by both membrane and
Equivalently, the costs for the adsorbent replacement is calculated as PSA processes and is thus particular interesting as sophisticated process
follows: design methods are necessary for an efficient process design. We used
150 Nm3/h as the feed stream volume, representing a feasible size of
tPlant
CPSA = IAdsorbent · biogenic gas production processes [32]. To investigate the influence of
tAds (26)
the system pressure on the process design and the operation costs, we
with tAds as the lifetime of the sorbent. The costs due to product losses conducted a sensitivity study for different pressure levels, i.e. that for
per year are calculated as follows: each pressure (depending on the feed concentration between 2 and
16 bar) the optimal process design as well as the related costs were
CH2, loss = mFeed · p H2 ·(1 )· tOpe (27)
identified. For each configuration, the desorption takes place at 1 bar.
with mFeed , p H2 as the price of hydrogen and the recovery . All important process parameters are summarized in Table 2.
The main economic parameters used in this case study are sum-
2.4. Mathematical problem definition marized in Table 3. A membrane price of 55 €/m2 and a lifetime of four
years were assumed. For the adsorbent materials, life times of more
As mentioned before, identifying the optimal process design was
formulated as an optimization problem using MINLP: Table 2
Process parameters for biohydrogen upgrading.
min f (x , y ) (28)
Case study parameter Description Value Unit
subject to:
pDes Desorption pressure 1 bar
h (x , y ) = 0 (29)
pFeed Process feed gas pressure 1 bar
CO2 permeance of the membrane 1000 GPU
g (x , y ) 0 (30) QCO2
Q H2 H2 permeance of the membrane 85 GPU
x X (31) TFeed Process feed gas temperature 293.15 °K
VFeed Feed rate 150 Nm3/h
y {0;1} (32) yFeed H2 feed fraction 25/50/75 mol-%
yProduct Purity 99 mol-%
where f (x , y ) is the objective function, i.e. the specific operation costs,
Overall compressor efficiency 0.72 –
h (x , y ) are the model equations presented above such as the equations
Com
Sorbent Density of sorbent particle 916 kg/m3
describing the adsorber and membrane units, g (x , y ) is the set of con- Sorbent Adsorption bed porosity 0.5 kg/m3
straints that define the feasible space of operation such as the pressure
457
B. Ohs et al. Chemical Engineering Journal 364 (2019) 452–461
Recovery [%]
C0 Vessel base costs 1 k€
CF Cost factor to account for valves 3 –
0.6 Costs 90
D0 Vessel reference diameter 0.91 m
e Vessel diameter exponent 1.05 –
r Interest rate 9 %
L0 Vessel reference length 1.22 m 0.5 85
MF Vessel module factor 4.23 –
MPF Vessel pressure and materials correction factor 1.45 –
pCom Compressor investment price 6.5 k€/kW 0.4 80
pEnergy Electricity price 0.08 €/kWh 6 8 10 12 14 16
Adsorption pressure [bar]
p H2 H2 price 5 €/kg
pMem Membrane price 55 €/m2
pSorbent Adsorber material costs 3 €/kg
tMem Membrane lifetime 4 a 111.68 m³/h
p = 9.8 bar
tOpe Annual operation time 8000 h/a 150 m³/h 156.5 m³/h PSA 99 % H2
75 % H2 A = 3.05 m² 77.5 % H2
tPlant Plant lifetime 15 a V = 0.9 m³
1 bar
tSorbent Adsorber material lifetime 4 a
UF Update factor (1968/2016) 4.9 –
p = 9.8 bar 44.9 m³/h
w Vessel length exponent 0.81 –
A = 0.75 m² 24.0 % H2
10.3 m³/h
5.2 % H2 p = 9.8 bar
than ten years have been reported [71]. However, minor impurities
A = 2.00 m²
might be present in the feed stream and reduce the sorbent’s life time. 27.9 m³/h
Thus, a life time of four years, as a conservative estimation, was as- 4.9 % H2
sumed as well. The price of the sorbent material is 3 €/kg. The costs for
the pressure vessels in the adsorption unit were calculated using
Fig. 7. (a) Minimal specific purification costs of the hybrid membrane PSA
Guthrie’s cost method [64,72]. Additionally, a cost factor CF of three
process for a H2 feed fraction of 75 mol-% at different system pressures. For
has been applied to account for the valves and piping between the
each pressure, the optimal process was identified. (b) Optimal process design
pressure vessels. The cost factor is based on the mutual communication for a H2 feed fraction of 75 mol-%. The volume of the adsorber column re-
with a manufacturer of PSA plants. The investment costs for the com- presents the volume for one of the four columns.
pressor were considered with 6.5 k€/kW. Besides the investment and
operation expenditures, the costs for incomplete recovery and thus
necessary, which would result in the loss of mechanical energy, re-
hydrogen loss must be taken into account to optimize the benefit of the
spectively. Feed conditions of the separation process can vary sig-
process. Unfortunately, only little information of the hydrogen price in
nificantly with feedstock and operating conditions of the biohydrogen
dependence of the hydrogen purity can be found in the literature. The
production process. To provide optimal designs for a wide range of
market prices depend on the costs of the production route and feed-
potential feed compositions, we optimized the hybrid membrane-PSA
stocks, which vary both with location and time. For example, the U.S.
systems for H2 feed fractions of 25, 50 and 75 mol-%.
Department of Energy targets a hydrogen production costs for 2020 of
9.20/kg for photobiological biohydrogen production [73]. In this work,
we assumed a H2 price of 5 €/kg, which is a conservative assumption 4.1. Process optimization for a H2 feed fraction of 75 mol-%
for the hydrogen price from biological origin. Even if this causes some
uncertainty, we think that identifying the economic optimum is of The specific separation costs for different system pressures are
higher practical relevance as the energy demand, which is used in most presented in Fig. 7(a). The lowest costs of 0.60 €/kg are obtained at a
optimization studies, can often be reduced by providing more mem- system pressure of 9.8 bar. The resulting H2 recovery at this pressure is
brane area and/or adsorber size. The electricity price is 0.08 €/kWh 98.3%. The optimal process consists of three membrane stages and a
and the plant operates 8000 h/a. A plant lifetime of 15 years and an PSA unit for the final purification (see Fig. 7(b)). In the first membrane
interest rate r of 9% was assumed. For the optimal process design, the unit, the hydrogen is concentrated in the retentate stream and then fed
trade-offs between investment costs and operation costs, e.g. energy to the PSA unit. The permeate of the first membrane unit is pressurized
consumption and product recovery, must be balanced. Although the before entering a second membrane for enhanced recovery. The tail gas
ecomomic parameters may vary with time and location, which influ- of the PSA unit is repressurized as well and fed to a third membrane
ences the optimal solution of the mathematical problem, we think that unit, which retentate stream is recycled to the first membrane unit. A
the costs are a reasonable objective function as it comprises the most process with a system pressure of 6 bar and 16 bar results in almost 10%
important criteria of the integrated process. and 5% higher costs, respectively. For commodity products, such as
hydrogen, this can be a significant cost difference having a significant
economic impact on manufacturers. While the costs increase, it still
4. Results might be beneficial to use and operate at higher pressures depending on
the downstream utilization of the purified H2.
Using MINLP, we optimized the hybrid membrane-PSA processes.
Therefore, we systematically varied the system pressure level and op- 4.2. Process optimization for a H2 feed fraction of 50 mol-%
timized the process configuration and parameters, such as membrane
area and adsorbent bed size. A single system pressure was set by the For a hydrogen feed fraction of 50 mol-%, a system pressure of
boundary condition of the process model as it allows recycling the 7.7 bar resulted in the lowest costs of 1.02 €/kg (see Fig. 8(a)). Com-
process streams without any further compression, which would require pared to a feed fraction of 75 mol-%, the dependency on the system
an additional compressor and thus increase the costs. Furthermore, it pressure is more pronounced. In comparison to a feed content of 75%,
guarantees that any reduction of the pressure of single streams is the recovery is slightly lower (94.9% for 4 bar and 97.1% for 14 bar).
458
B. Ohs et al. Chemical Engineering Journal 364 (2019) 452–461
Recovery [%]
2.2
Recovery [%]
1.1 90
Costs 2.1 90
1.0 85 2.0
85
1.9 Costs
0.9 80
4 6 8 10 12 14
System pressure [bar]
1.8 80
2 4 6 8 10 12
System pressure [bar]
Fig. 9. (a) Minimal specific purification costs of the hybrid membrane PSA
process for a H2 feed fraction of 25 mol-% at different system pressures. For
Fig. 8. (a) Minimal specific purification costs of the hybrid membrane PSA
each pressure, the optimal process was identified. (b) Optimal process design
process for a H2 feed fraction of 50 mol-% at different system pressures. For
for a H2 feed fraction of 25 mol-%.
each pressure, the optimal process was identified. (b) Optimal process design
for a H2 feed fraction of 50 mol-%. The volume of the adsorber column re-
presents the volume for one of the four columns. 100
2.0
Costs 95
For the feed stream with a H2 feed fraction of 50 mol-%, Fig. 8(b) de-
459
B. Ohs et al. Chemical Engineering Journal 364 (2019) 452–461
develop biohydrogen production paths which allow high hydrogen gas Humboldt Foundation.
phase concentrations.
While the conclusions drawn are specific for the process conditions References
simulated, some general conclusions for hybrid membrane-PSA pro-
cesses can be derived nonetheless: [1] H. Lund, Renewable energy strategies for sustainable development, Energy 32
(2007) 912–919.
• At low system pressures, the required size for the membrane (and
sustainable development scenario, Int. J. Hydrogen Energy 28 (2003) 267–284.
[5] D. Stolten, Hydrogen Science and Engineering: Materials, Processes, Systems and
PSA) units increases and the recovery decreases sharply (compare Technology 2 Volume Set John Wiley & Sons, 2016.
Fig. 6). Thus, a high degree of pre-purification is more beneficial. [6] P.N.R. Vennestrøm, C.M. Osmundsen, C.H. Christensen, E. Taarning, Beyond pet-
460
B. Ohs et al. Chemical Engineering Journal 364 (2019) 452–461
B.D. Freeman, Energy-efficient polymeric gas separation membranes for a sustain- separation processes, Adsorption 21 (2015) 283–305.
able future: a review, Polymer 54 (2013) 4729–4761. [56] K. Bélafi-Bakó, D. Búcsú, Z. Pientka, B. Bálint, Z. Herbel, K. Kovács, M. Wessling,
[34] P. Bernardo, E. Drioli, G. Golemme, Membrane gas separation: a review/state of the Integration of biohydrogen fermentation and gas separation processes to recover
art, Ind. Eng. Chem. Res. 48 (2009) 4638–4663. and enrich hydrogen, Int. J. Hydrogen Energy 31 (2006) 1490–1495.
[35] M. Scholz, M. Alders, J. Lölsberg, M. Wessling, Dynamic process simulation and [57] P. Bakonyi, G. Kumar, N. Nemestóthy, C.Y. Lin, K. Bélafi-Bakó, Biohydrogen pur-
process control of biogas permeation processes, J. Membr. Sci. 484 (2015) 107–118. ification using a commercial polyimide membrane module: studying the effects of
[36] R. Baker, J. Wijmans, J. Kaschemekat, The design of membrane vapor-gas separa- some process variables, Int. J. Hydrogen Energy 38 (2013) 15092–15099.
tion systems, J. Membr. Sci. 151 (1998) 55–62. [58] M. Skiborowski, A. Mhamdi, K. Kraemer, W. Marquardt, Model-based structural
[37] Y. Huang, T.C. Merkel, R.W. Baker, Pressure ratio and its impact on membrane gas optimization of seawater desalination plants, Desalination 292 (2012) 30–44.
separation processes, J. Membr. Sci. 463 (2014) 33–40. [59] H. Lin, Z. He, Z. Sun, J. Vu, A. Ng, M. Mohammed, J. Kniep, T.C. Merkel, T. Wu,
[38] R. Qi, M. Henson, Optimization-based design of spiral-wound membrane systems R.C. Lambrecht, CO2-selective membranes for hydrogen production and CO2 cap-
for CO2/CH4 separations, Sep. Purif. Technol. 13 (1998) 209–225. ture – part i: Membrane development, J. Membr. Sci. 457 (2014) 149–161.
[39] B. Ohs, J. Lohaus, M. Wessling, Optimization of membrane based nitrogen removal [60] Linde Engineering AG, Hydrogen recovery by pressure swing adsorption, 2016.
from natural gas, J. Membr. Sci. 498 (2016) 291–301. URL: http://www.linde-engineering.de/internet.le.le.deu/en/images/HA_H_1_1_e_
[40] S. Alsayegh, J. Johnson, B. Ohs, J. Lohaus, M. Wessling, Systematic optimization of 09_150dpi_NB543_6130.pdf?v=3.0, (accessed July 2017).
H2 recovery from water splitting process using membranes and N2 diluent, Int. J. [61] D. Li, Y. Zhou, Y. Shen, W. Sun, Q. Fu, H. Yan, D. Zhang, Experiment and simulation
Hydrogen Energy 42 (2017) 6000–6011. for separating CO2/N2 by dual-reflux pressure swing adsorption process, Chem.
[41] M. Scholz, M. Alders, T. Lohaus, M. Wessling, Structural optimization of membrane- Eng. J. 297 (2016) 315–324.
based biogas upgrading processes, J. Membr. Sci. 474 (2015) 1–10. [62] Y. Chung, B.-K. Na, H.K. Song, Short-cut evaluation of pressure swing adsorption
[42] S. Alsayegh, J. Johnson, X. Wei, B. Ohs, J. Lohaus, M. Wessling, CO2 aided H2 systems, Comput. Chem. Eng. 22 (1998) S637–S640.
recovery from water splitting processes, Int. J. Hydrogen Energy 42 (2017) [63] J.A. Delgado, V.I. Gueda, M.A. Uguina, J.L. Sotelo, P. Brea, C.A. Grande, Adsorption
21793–21805. and diffusion of H2, CO, CH4, and CO2 in bpl activated carbon and 13X zeolite:
[43] D.G. Loffler, K. Taylor, D. Mason, A light hydrocarbon fuel processor producing evaluation of performance in pressure swing adsorption hydrogen purification by
high-purity hydrogen, J. Power Sources 117 (2003) 84–91. simulation, Ind. Eng. Chem. Res. 53 (2014) 15414–15426.
[44] J.R. Hufton, S. Mayorga, S. Sircar, Sorption-enhanced reaction process for hydrogen [64] L.T. Biegler, I.E. Grossmann, A.W. Westerberg, Systematic Methods for Chemical
production, AIChE J. 45 (1999) 248–256. Process Design, Prentice Hall, Old Tappan, NJ (United States), 1997.
[45] A. Brunetti, F. Scura, G. Barbieri, E. Drioli, Membrane technologies for CO2 se- [65] B. Bharathiraja, T. Sudharsanaa, A. Bharghavi, J. Jayamuthunagai,
paration, J. Membr. Sci. 359 (2010) 115–125. R. Praveenkumar, Biohydrogen and biogas – an overview on feedstocks and en-
[46] S. Sircar, T.C. Golden, Purification of hydrogen by pressure swing adsorption, Sep. hancement process, Fuel 185 (2016) 810–828.
Sci. Technol. 35 (2000) 667–687. [66] M. Shalygin, S. Abramov, A. Netrusov, V. Teplyakov, Membrane recovery of hy-
[47] B. Ohs, J. Lohaus, D. Marten, R. Hannemann-Tamas, A. Krieger, M. Wessling, drogen from gaseous mixtures of biogenic and technogenic origin, Int. J. Hydrogen
Optimized hollow fiber sorbents and pressure swing adsorption process for h2 re- Energy 40 (2015) 3438–3451.
covery, Ind. Eng. Chem. Res. 57 (2018) 5093–5105. [67] H. Lin, Z. He, Z. Sun, J. Kniep, A. Ng, R.W. Baker, T.C. Merkel, CO2-selective
[48] N.W. Ockwig, T.M. Nenoff, Membranes for hydrogen separation, Chem. Rev. 107 membranes for hydrogen production and CO_2 capture – part ii: Techno-economic
(2007) 4078–4110. analysis, J. Membr. Sci. 493 (2015) 794–806.
[49] B. Li, G. He, X. Jiang, Y. Dai, X. Ruan, Pressure swing adsorption/membrane hybrid [68] F.V. Lopes, C.A. Grande, A.M. Ribeiro, J.M. Loureiro, O. Evaggelos, V. Nikolakis,
processes for hydrogen purification with a high recovery, Front. Chem. Sci. Eng. 10 A.E. Rodrigues, Adsorption of H2, CO2, CH4, CO, N2 and H2O in activated carbon
(2016) 255–264. and zeolite for hydrogen production, Sep. Sci. Technol. 44 (2009) 1045–1073.
[50] S. Sircar, W. Waldron, M. Rao, M. Anand, Hydrogen production by hybrid SMR- [69] D.B. Levin, R. Chahine, Challenges for renewable hydrogen production from bio-
PSA-SSF membrane system, Sep. Purif. Technol. 17 (1999) 11–20. mass, Int. J. Hydrogen Energy 35 (2010) 4962–4969.
[51] I. Esteves, J. Mota, Simulation of a new hybrid membrane/pressure swing adsorp- [70] Z. Liao, J. Wang, Y. Yang, G. Rong, Integrating purifiers in refinery hydrogen net-
tion process for gas separation, Desalination 148 (2002) 275–280. works: a retrofit case study, J. Clean. Prod. 18 (2010) 233–241.
[52] I.A.A.C. Esteves, J.P.B. Mota, Gas separation by a novel hybrid membrane/pressure [71] F.G. Wiessner, Basics and industrial applications of pressure swing adsorption
swing adsorption process, Ind. Eng. Chem. Res. 46 (2007) 5723–5733. (PSA), the modern way to separate gas, Gas Sep. Purif. 2 (1988) 115–119.
[53] E. Ochoa-Fernández, G. Haugen, T. Zhao, M. Rønning, I. Aartun, B. Børresen, [72] K.M. Guthrie, Data and techniques for preliminary capital cost estimating, Chem.
E. Rytter, M. Rønnekleiv, D. Chen, Process design simulation of H2 production by Eng. 76 (1969) 114.
sorption enhanced steam methane reforming: evaluation of potential CO2 acceptors, [73] U.S. Department of Energy, DOE Technical Targets for Photobiological Hydrogen ,
Green Chem. 9 (2007) 654–662. 2015. URL: https://www.energy.gov/eere/fuelcells/doe-technical-targets-
[54] A. Carrara, A. Perdichizzi, G. Barigozzi, Simulation of an hydrogen production photobiological-hydrogen-production.
steam reforming industrial plant for energetic performance prediction, Int. J. [74] N. Aryal, T. Kvist, F. Ammam, D. Pant, L.D. Ottosen, An overview of microbial
Hydrogen Energy 35 (2010) 3499–3508. biogas enrichment, Bioresour. Technol. 264 (2018) 359–369.
[55] D.D. Nikolić, E.S. Kikkinides, Modelling and optimization of hybrid PSA/membrane
461