Systematic Optimization of an H2 PEM Fuel Cell

Power Generation System with Heat Integration

Cong Xu, Lorenz T. Biegler, and Myung S. Jhon
Dept. of Chemical Engineering, Carnegie Mellon University, Pittsburgh, PA 15213

DOI 10.1002/aic.10873
Published online April 20, 2006 in Wiley InterScience (www.interscience.wiley.com).

An H2 Polymer Electrolyte Membrane (PEM) Fuel Cell power generation system

composed of three subsystems: fuel reforming, fuel cell stack, and post combustion is
examined. The system is simulated and optimized with a fuel cell model integrated within
the process flow sheet. A case study optimization of the entire power generation system is
presented and the influence of heat integration on the optimization is demonstrated. Our
objectives are to maximize the energy and system efficiencies as well as the system profit.
Here, optimization formulations with heat integration are used to realize our objectives,
which explore the synergy between optimization and heat integration. Our results showed
that we can achieve an energy efficiency as high as 58.30% and a system efficiency as high
as 28.35%. For comparison, we provide results obtained from sequential process opti-
mization followed by heat integration. The gain from the synergy is mostly exemplified in
the profit maximization study, where the profit increases by almost 27.15% in the best
case. Finally, we obtain optimization results that include feed vaporization of water to
steam as a comparison for the heat integration. These results demonstrate the importance
of simultaneous optimal heat integration for fuel cell-based processes. © 2006 American
Institute of Chemical Engineers AIChE J, 52: 2496 –2506, 2006
Keywords: fuel cell system, H2 PEM fuel cell, heat integration, power generation,
efficiency, cost optimization, fuel processing

Introduction the optimal process design. Since the modeling work should
Over the last few years, fuel cells have attracted considerable reflect the complexity of the overall process, unravel the interac-
attention as a potential replacement for current power genera- tions, and simplify input/output characteristics of components,2
the National Energy Technology Laboratory (NETL) is currently
tion and automobile engine systems. Fuel cells have the ad-
developing a multi-level model of fuel cells, including micro or
vantage of high power density, high efficiency, and zero or
fundamental modeling, cell modeling, stack modeling, and system
ultra-low emissions, and promise to ease concerns regarding
modeling. Many studies3-8 describe the modeling of fuel cells.
fossil energy and the environment. Government and the indus-
Starting from the early 90s, researchers3,4,6,9 have developed sim-
try supported research programs have concentrated on the
ple 0-D models, some of which focused on separate parts in the
development of operable fuel cells and their commercializa-
fuel cells, such as the electrodes or the gas diffusion layers.9
tion. In a “fuel cell report to Congress,”1 cost and durability are
Recently, Springer et al.5 developed a 1-D H2 PEMFC model that
the primary technical barriers to commercialization.
examines the effect of CO poisoning and hydrogen dilution. Wang
To reduce the cost and improve the durability of fuel cells, we
et al.8 have modeled PEMFCs extensively and have extended the
need to understand the fuel cell at every level in order to provide
model to two and even three spatial dimensions.7 Water and heat
management have also been considered in their models since the
Correspondence concerning this article should be addressed to L. T. Biegler at temperature profiles can be easily obtained with higher dimen-
lb01@andrew.cmu.edu.
sional models by incorporating computational fluid dynamics
© 2006 American Institute of Chemical Engineers (CFD) techniques.

2496 July 2006 Vol. 52, No. 7 AIChE Journal

Figure 1. The ASPEN Plus flow sheet for the fuel cell system with hot and cold streams denoted.
[Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Despite the rapid development of fuel cell modeling, there are Our study also focuses on the optimization of this particular
few studies on fuel cell systems, including fuel reforming pro- heat integrated PEM fuel cell power generation system. In
cesses. To reduce the cost and improve the performance, it will be addition to the work of Godat and Marechal,12 it performs a
desirable to examine fuel cell behavior under the environment of systematic process optimization using an integrated heat inte-
the whole process. We not only need to understand cells and gration formulation along with Sequential Quadratic Program-
stacks, but also the supporting technology relevant to fuel cell ming (SQP), a state-of-the-art nonlinear programming (NLP)
development, such as fuel storage, fuel distribution, fuel reform- algorithm. Using SQP within the Aspen Plus simulator, we
ing, post-combustion, and heat exchanger networks. Moreover, consider the maximization of the fuel cell efficiency, as well as
optimization has been in high demand for fuel cell design and system efficiency and profit. Finally, heat exchanger network
operation. Since the late 90s, Moore and coworkers10 have been (HEN) structures are constructed to provide guidance for the
working on the optimization of the overall energy conversion optimal design and operation of the system.
efficiency and concentrated on performance of DMFC stacks in In the following sections, we first introduce the fuel cell
automotive applications. They aimed to develop an operational system for our simulation. We provide sensitivity studies for
strategy to achieve maximum efficiency by varying the methanol the flowsheet design variables and follow this with the descrip-
feed concentration and flow rate under a wide range of power tion of simultaneous optimization strategy with heat integra-
demands. Similarly, we have recently performed an optimization tion. Then we provide our results about the minimization of the
study on a DMFC, which provides methanol feeding strategies utility and system cost and the maximization of the system
both in the steady and dynamic states to achieve the maximum efficiency by manipulating key decision variables in the pro-
power output under various current densities.11 Godat and Mare- cess flow sheet.
chal12 considered the effect of heat integration of the optimal
performance of power generation systems that incorporate H2
PEM fuel cells. They considered key variables of the fuel cell The Fuel Cell System
process and performed single variable optimizations along with Our system model is based on work by Godat and Mare-
sensitivity analysis of these heat integrated processes. More re- chal12 and builds on a flow sheet using the Aspen Plus14
cently, Marechal and coworkers13 extended this approach with a simulator, as shown in Figure 1. It consists of the hydrogen
genetic algorithm for multi-objective optimization; they were able production process, fuel cell stacks, and a post combustion
to consider structural flowsheet options as well as capital costs. process. Typically, a steam reformer reactor running on natural

AIChE Journal July 2006 Vol. 52, No. 7 Published on behalf of the AIChE DOI 10.1002/aic 2497
Figure 2. Anode H2 inlet flow for different CH4 feeding Figure 3. Case study on system power output under var-
rates versus SMR reactor temperatures. ious CH4 feeding rates and different TSMR.
[Color figure can be viewed in the online issue, which is [Color figure can be viewed in the online issue, which is
available at www.interscience.wiley.com.] available at www.interscience.wiley.com.]

gas is chosen for the hydrogen generation. Since methane is the and H2O. After removing H2O in a flash separator, the O2
main component of natural gas, for simplicity we use pure stream combines with the anode effluent and enters the post-
methane as the inlet. Here, stream #1 is saturated steam feed at combustor where O2 reacts with H2. Finally, the hot outlet
3 bar and stream #2 is a pure methane feed provided at 3 bar stream from the post combustor is depressurized and dis-
and 25°C. The two streams are combined, heated to reactor charged to the atmosphere.
temperature, and fed to the steam methane reforming (SMR) The SMR and WGS reactors are both modeled with ap-
reactor, where methane and steam are reacted on a catalyst proach to equilibrium reactor models (REquil) in ASPEN Plus.
surface to produce hydrogen and carbon monoxide. Generally, Reactions in the PROX and post-combustion (POSTCOMB)
the reformate from the steam reformer reactor is composed of reactors essentially go to completion and are both modeled
75% hydrogen and 10-15% carbon monoxide.15 with the stoichiometric models (RStoic) in ASPEN Plus. All
For the PEM fuel cell application, the carbon monoxide reactors are operated under isothermal conditions, and heat is
content must be 10 ppm or less. To remove the carbon mon- supplied or removed to maintain the temperature in the reactor
oxide, the SMR reformate enters the water gas shift (WGS) from an external hot or cold utility. In order to define appro-
reactor, where carbon monoxide reacts further with steam to priately the energy requirement of the heat exchange, we have
produce hydrogen and carbon dioxide. Depending on the re- assumed that the inlet streams are preheated up to the reaction
formate composition, more than one shift reactor may be temperature. To model them, we place heat exchangers before
needed. In industrial practice, there might be a two-stage WGS each unit to ensure that the inlet stream is at the same temper-
system: a high (400-500°C) temperature WGS reactor followed
by a low temperature (200-250°C) one. These two stages of
shift conversion can reduce the content of carbon monoxide to
a level as low as 2500 ppm. However, this is still not sufficient
for the PEM fuel cell operation. A preferential oxidation
(PROX) reactor is needed to further remove carbon monoxide.
For simplicity, we only add one WGS reactor in the system, as
in Godat and Marechal.12 However, the PROX reactor will
ensure the carbon monoxide content is decreased to the accept-
able level, though some hydrogen may also be oxidized in the
PROX reactor if the carbon monoxide inlet concentration is too
high. In the flowsheet shown in Figure 1, the WGS effluent
enters the PROX reactor along with a make-up system to
oxidize the remaining CO in the stream. This stream consists of
pure oxygen, required for the oxidation of CO. The effluent
stream from the PROX reactor, which is hydrogen-rich with Figure 4. Generalized polarization curve for a fuel cell
trace amounts of CO, is fed to the anode of the fuel cell stacks. showing regions dominated by various types
A compressed oxygen stream (#13) is fed to the cathode of the of losses.
fuel cell stacks. The outlet stream from the anode is a hydro- [Color figure can be viewed in the online issue, which is
gen-lean stream, while the cathode effluent contains both O2 available at www.interscience.wiley.com.]

2498 DOI 10.1002/aic Published on behalf of the AIChE July 2006 Vol. 52, No. 7 AIChE Journal
 Table 1. Energy Efficiency Optimization Results for Cases icell ⫽ Icell/Acell
Where fCH4 Assumes Five Different Lower Bounds,
Icell ⫽ 2Fni,cathode
H2
Respectively
Pcell ⫽ VcellIcell (1)
LB f CH 4* T SMR * T WGS * P cell
(kmol/hr) (kmol/hr) (°C) (°C) (kW) Efficiency
50.00 50.00 780.00 236.44 6505.76 58.30% Here, UNernst is the cell voltage adjusted to the cell temper-
80.00 80.00 679.00 244.81 6714.02 37.60% ature (Tcell), pressure (pcell), and component mole fractions (Y␰)
100.00 100.00 655.48 245.85 6701.60 30.03% of species ␰, Vcell is the cell voltage, icell is the current density,
120.00 120.00 647.87 250.84 6673.95 24.92%
150.00 150.00 627.52 252.27 6664.47 19.91% Icell is the current, Pcell is the power output of the cell, Rcell is
the internal resistance of the cell, and Acell is the cross sectional
*Denotes optimal values. area of the cell, respectively. From Eq. 1, we find that Tcell and
pcell have a significant effect on the value of Vcell. These
ature. Like the construction of pinch curves, the process inte- decision variables should be set at their upper bounds and
gration optimization formulation allows us to avoid detailed practical limits from other studies. The modern PEM fuel cells
construction of the complex heat exchange network within the operate at a 60-100°C range.15 Generally, the higher the tem-
heat integrated system. perature, the better the performance will be, mainly due to
The fuel cell system is composed of three subsystems as cathode overpotential reduction. Since 100°C is the water
follows: vaporization point, close to 100°C cell temperature will raise
(1) The fuel processing (FP) subsystem. significant difficulty to maintain a reasonable humidity level
(2) The PEMFC stack subsystem. inside the PEM. However, Larminie and Dicks15 show it is
(3) The post-combustion subsystem. viable to run the fuel cell at temperatures as high as 90°C
without causing the PEM to dry out. So we set Tcell and pcell to
be 90°C and 3 bar, respectively. Since we operate all the other
Fuel reforming components of the system under constant pressure, it is also
The fuel processing subsystem includes the following three valid to operate the stacks this way.
main reactors:
(1) To produce hydrogen from methane, the SMR proceeds
with two reactions: endothermic reforming of methane and Post combustion
steam, CH4 ⫹ H2O ⬍⫽⫽⬎ 3H2 ⫹ CO, and an exothermic The post combustion subsystem burns the depleted fuel of
water gas shift (WGS) reaction, CO ⫹ H2O ⬍⫽⫽⬎ CO2 ⫹ the PEMFC subsystem and generates heat that is used to
H2. balance the energy requirement of the fuel processing section.
(2) By reacting CO with steam, the WGS reactor produces Complete combustion is assumed using a stoichiometric
additional hydrogen through the water gas shift reaction, but at model. The stream at the cathode outlet is cooled down and the
an operating temperature lower than the SMR reactor. water is separated and possibly recycled into the steam meth-
(3) The PROX reactor converts the remaining CO into CO2, ane reformer.
by the reaction CO ⫹ 1⁄2 O2 ⬍⫽⫽⬎ CO2, to avoid poisoning
the fuel cell membrane. However, the catalyst is nonselective
and some hydrogen in the gas stream is also combusted ac- Heat Integration
cording to H2 ⫹ 1⁄2 O2 ⬍⫽⫽⬎ H2O. The idea of simultaneous optimization and heat integration
Further details of the SMR, WGS, and PROX reactors can be of chemical processes was initially proposed by Duran and
found in Godat and Marechal.12 Grossmann.16 Lang et al.17 and Yee et al.18 improved this
approach and expanded its application for flowsheet simulators
Fuel cell stack model and MINLP process design, respectively. Note that the heat
The fuel cell stack model is equilibrium based and adopted exchange network is not considered directly in the simulation
from Godat and Marechal.12 The model equations are given as model described above. Instead, an optimization formulation
follows: related to pinch technology can be used to model the integrated
heat exchange system without having to impose a heat ex-
change network structure. A typical heat integration optimiza-
U Nernst ⫽ UNernst
0
共Tcell兲 tion problem using pinch technology can be described as fol-

⫹
RTcell
2F
H2
ln共Yanode O2
(Yanode 冒
兲1/ 2 Yanode
H2 O
)⫹
RTcell
4F
ln共 pcell兲
lows:

V cell ⫽ UNernst ⫺ Rcellicell min C ⫽ ␾共x兲 ⫹ cs Qs ⫹ cw Qw

Table 2. System Efficiency Optimization Results With and Without Heat Integration (HI) for LB ⴝ 50
f CH 4* T SMR * T WGS * P cell Qs Qw
Cases (kmol/hr) (°C) (°C) (kW) (kW) (kW) Efficiency
Simultaneous optimization with HI 50.00 780.00 278.00 6497.10 4922.02 6854.21 28.35%
Sequential optimization with HI 50.00 780.00 234.39 6506.18 5030.79 6964.43 28.10%
Optimization without HI 50.00 780.00 234.39 6506.18 10418.95 8852.44 21.38%

AIChE Journal July 2006 Vol. 52, No. 7 Published on behalf of the AIChE DOI 10.1002/aic 2499
 Table 3. Cost Coefficients22 Values for the System temperatures as variables for the optimization and the heat
Cost CS Cw Cp C CH4 integration. The remaining difficulty is the presence of nondif-
Coeff. ($/kWh) ($/kWh) ($/kWh) ($/kmol) ferentiable max operators in Eq. 2. As shown in Eq. 4, a
Value 0.0147 0.0017 0.0921 2.8739 smoothing approximation procedure can be used that avoids
difficulties with the use of NLP solvers.19,20
C s is for the hot utility, that is, natural gas; C w is for the cold water utility; C p
is for the power output; C CH4 is for the CH4 inlet stream.

max兵0, f共x兲其 ⫽ 0.5关 f共x兲2 ⫹ ␧2 兴1/ 2 ⫹ 0.5f共x兲 (4)

s.t.

h共 x兲 ⫽ 0 where ␧ usually assumes a relatively small value (such as 10⫺3

g共 x兲 ⱕ 0 in our study).
The optimization formulation (Eq. 2) is applied to a number
冘 f 关max兵0, t
nc

Qs ⱖ out
⫺ 共T p ⫺ ⌬T min兲其 of fuel-cell optimization cases in the next section. Each of these
j j
j⫽1 optimization problems was solved with the SQP solver in
Aspen Plus. Because SQP is an efficient gradient-based NLP
⫺ max兵0, t jin ⫺ 共T p ⫺ ⌬T min兲其兴 solver, optimizations of all of these cases were obtained quite

冘 F 关max兵0, T
nH efficiently, within an order of magnitude of the effort required
⫺ i
in
i ⫺ T p其 ⫺ max兵0, T iout ⫺ T p其兴, p僆⌸ for a single flowsheet simulation.
i⫽1

Qw ⫽ Qs ⫹ ⍀

冘 F 共T 冘 f 共t
nH nC Results and Discussion
⍀⫽ i
in
i ⫺ T iout兲 ⫺ i
out
j ⫺ t jin兲 (2) Sensitivity analysis on H2 inlet versus CH4 inlet and
i⫽1 j⫽1 TSMR
There are many variables in the system that can be chosen as
where ␾(x) is the cost/profit function other than utility cost, ⍀
the decision variables for an optimization problem. We ran the
is the heat surplus of the system, Qs is the requirement of the
case study to choose the most sensitive ones, such as TSMR,
hot utility, Qw is the requirement of the cold utility, h(x) is the
ASPEN Plus flow sheet model, g(x) represents the inequality TWGS, Tcomb, fCH4, and fH2O, using the sensitivity analysis in
constraints, j is the index for cold streams in the flow sheet, i ASPEN Plus’s model analysis tool. Here, f␰ denotes the feed
is the index for hot streams, tj is the temperature of cold stream flow rate of species ␰.
j, Ti is the temperature of hot stream i, and Fi and fj are the heat The H2O and CH4 inlet streams have a similar effect on the
capacity flowrates for the hot and cold streams, respectively. conversion of H2; either of them can be fixed relative to the
The index set ⌸ defines candidate pinch points from inlet other. For convenience, we set the H2O inlet flow rate at a
temperatures of hot and cold streams; Tp, p␧⌸ is the candidate constant value (200 kmol/hr) and choose fCH4 as the decision
pinch temperature defined as follows: variable.
Generally, if we increase the inlet flow rate of the H2O and

j
再
T iin if candidate p is hot stream i
T P ⫽ tin ⫹ ⌬T if candidate p is cold stream j
min
(3)
CH4 or TSMR, more H2 is produced. When TSMR is at 200°C or
below, almost no H2 is produced because the catalyst for the
SMR reaction is inactive at this relatively low temperature.
⌬Tmin is the minimal temperature interval that is possible for Similarly, we noticed that after TSMR reached 800°C, further
the heat exchange to take place, which is set to be 20°C in our increase of TSMR does not have a significant positive effect on
study. The number of pinch temperature candidates is deter- the production of the H2; this is thermodynamically consistent
mined by the total number of hot and cold streams. The right with the result in Lutz et al.21 Also, Figure 2 provides a rough
hand side of the inequality for Qs in Eq. 2 represents the estimate on the lower and upper bounds that could be set for the
difference between the heat contents of the cold and hot decision variables. For example, H2 production is much more
streams (heat deficit) above any candidate pinch p␧⌸. The sensitive to CH4 inlet flow rate in the range between 50 and 200
problem is then posed to find the smallest value of Qs such that kmol/hr than outside of this range.
all the inequalities hold; Qs, therefore, equals the maximal The WGS temperature shows far less sensitivity to H2 pro-
value of the heat deficit. The temperature Tp where the heat duction, although we include it in the optimization studies
deficit achieves this maximum is the pinch point. below. Finally, we observe that the high post-combustor tem-
Note that the above formulation can treat the flows and the perature Tcomb has no influence on hydrogen conversion.

Table 4. Simultaneous System Profit Optimization Results with Heat Integration

LB f CH 4* T *SMR T WGS * P cell Energy Qs Qw Profit
(kmol/hr) (kmol/hr) (°C) (°C) (kW) Efficiency (kW) (kW) ($/hr)
50.00 50.00 780.00 257.60 6502.00 58.27% 4922.00 6855.00 370.70
100.00 100.00 650.81 256.29 6697.00 30.01% 5551.46 8535.82 232.79
150.00 150.00 618.69 245.57 6667.00 19.91% 6036.52 8488.49 79.31

2500 DOI 10.1002/aic Published on behalf of the AIChE July 2006 Vol. 52, No. 7 AIChE Journal
 Table 5. System Profit Optimization Results Without Heat Integration
LB f CH 4* T SMR * T WGS * Qgen Qsink TotalQ Profit_NoHI
(kmol/hr) (kmol/hr) (°C) (°C) (kW) (kW) (kW) ($/hr)
50.00 50.00 780.00 257.60 10420.00 8488.00 18908.00 286.90
100.00 100.00 650.81 256.29 12274.00 9290.00 21564.00 132.47
150.00 150.00 618.69 245.57 12535.00 10084.00 22619.00 ⫺19.21

Cell performance Pcell Vcell 䡠 icell

␩ energy ⬅ ⬅ (7)
Cell power output is a key measure of the performance of the Pcombustion fCH4 䡠 LHVCH4
power generation system. The cell power equals to the product
of cell voltage and current: where LHVCH4 denotes the lower heating value of methane.
The system efficiency can be defined as:
P cell ⬅ VcellIcell (5)
Vcell 䡠 icell
Figure 3 shows how the two decision variables would affect ␩ sys ⬅ (8)
fCH4 䡠 LHVCH4 ⫹ Q
the performance of the system. We already know that high
TSMR or fCH4 are generally beneficial for H2 production.
On the other hand, the current of the fuel cell stack is where Q is the additional energy consumed to operate the
correlated with the anode H2 inlet by the following equation: system. This quantity consists mostly of the hot and cold
utilities.
I cell ⫽ 2 䡠 F 䡠 f Hreacted
2
(6)
Optimal energy efficiency
Here, F is the Faraday constant and reacted
fHis the amount of H2 The optimization problem can be defined as follows:
2
reacted to produce power in the stack. We assume a constant
Pcell共 fCH4, TSMR, y兲
reacted
conversion ratio for the H2 such that f H is directly propor-
max ␩FCsys ⫽
2
tional to the H2 inlet flow rate. The conversion ratio is also LHVCH4 䡠 fCH4
ln
known as the fuel utilization parameter, which we assume to be 兵 f CH 4
,T SMR其

95% as was commonly used in the literature.12 The character- s.t. h共 f CH4, TSMR, TWGS, y兲 ⫽ 0
ization curve of the fuel cell stack is the polarization curve LB ⱕ fCH4 ⱕ 200 kmol/h
(Figure 4).
420 ⱕ T SMR ⱕ 780⬚C
Typically, if current is zero, cell voltage is at its maximum
value, called OCV (open circuit voltage). As current increases, 200 ⱕ T WGS ⱕ 400⬚C (9)
the cell voltage decreases due to various polarizations inside
the cells. After current reaches a limit current, the cell voltage where the equation h(fCH4, TSMR,TWGS, y) ⫽ 0 represents the
becomes zero. This explains the zero power densities in the ASPEN Plus flow sheet model. The lower and upper bounds
upper right hand corner of Figure 3. This phenomenon also for fCH4 and TSMR were chosen according to the results obtained
confirms that there exist optimal values for the TSMR under from the sensitivity analysis. The range of TSMR is determined
different CH4 inlet conditions to maximize the power output. from the thermodynamic analysis of the hydrogen production
For example, when TSMR equals about 600°C, increasing the by steam reforming.22
CH4 inlet flow rate from 200 kmol/hr leads to decreasing power As is shown in Table 1, optimal energy efficiency can be as
output. high as 58.3% in our study when fCH4 assumes the lower bound
value of 50 kmol/hr and TSMR is at the upper bound, 780°C. As
Efficiency analysis the CH4 inlet rate increases, the energy efficiency decreases
sharply before it reaches 100 kmol/hr and levels off afterwards.
The above case studies provide us with an understanding of The optimal TSMR also decreases accordingly until it reaches a
the effect of the two decision variables in the following opti- lowest value at 627°C.
mization problem formulation. We first focus on the efficiency.
There are two efficiencies to evaluate the system; one is the
energy efficiency and the other is the system efficiency. The System efficiency
energy efficiency only considers the conversion rate of the fuel By considering the heat and cold utility consumption in the
source energy to the cell power, which can be defined as whole system, we get the overall efficiency of the power
follows: generation system. This can be denoted as the cell power over

Table 6. Comparison of the System Profit Optimization Results Between Cases With and Without Heat Integration
LB (kmol/hr) Profit_NoHI ($/hr) Profit_HI ($/hr) Percentage Gain with HI
50.00 286.90 370.70 29.21%
100.00 132.47 232.79 75.73%
150.00 ⫺19.21 79.31 N/A

AIChE Journal July 2006 Vol. 52, No. 7 Published on behalf of the AIChE DOI 10.1002/aic 2501
 decreases by 38% with heat integration. We see that heat
integration improves the system efficiency by about 6.7%,
which is a significant achievement from the simultaneous op-
timization with heat integration.

Sequential optimization and heat integration

To show the importance of simultaneous optimization with
heat integration, we carry out a sequential optimization study.
We first perform the process optimization without heat inte-
gration, then followed by heat integration. The optimization
problem formulation is largely the same as in Eq. 10 except
that Qs and Qw are simply the hot and cold utility requirements
of the system. As expected, we obtain different optimal values
for certain decision variables than those from simultaneous
optimization: methane flow rate and temperature of SMR are
the same while temperature of WGS is different. Next, we
perform heat integration based on the settings from the process
optimization and obtain the system efficiency. From Table 2,
we can see that the optimal efficiency results from simulta-
neous optimization are approximately 0.25% higher than the
sequential optimization results.

Profit optimization
The total system profit equals the revenue from the power
generated by the fuel cell stack minus utility cost and the cost
of the material (methane only). For cost coefficients, refer to
Table 3. The corresponding minimization problem can be rep-
Figure 5. Comparison of the profit for cases with and
resented as follows:
without heat integration.
[Color figure can be viewed in the online issue, which is
available at www.interscience.wiley.com.] min C ⫽ cs Qs ⫹ cw Qw ⫺ cp Pcell ⫹ cCH4fCH4

the total of the fuel and utility energy consumption. Similar to s.t.
the energy efficiency case, we can formulate the system effi-
ciency optimization problem as follows: h共 x兲 ⫽ 0
g共 x兲 ⱕ 0
Pcell共 fCH4, TSMR, y兲
max ␩FCsys ⫽
兵 f ln
LHVCH4 䡠 fCH4 ⫹ Qs ⫹ Qw
CH4, T SMR 其

s.t. h共 f CH4, TSMR, y兲 ⫽ 0

LB ⱕ fCH4 ⱕ 200
420 ⱕ T SMR ⱕ 780
200 ⱕ T WGS ⱕ 400 (10)

Here Qs and Qw result from the heat integration calculation.

Compared to energy efficiency, the optimal value of the system
efficiency is correspondingly lower due to the consideration of
the utility consumption, which is comparable to the power
generated, as is shown in Table 2. The maximal system effi-
ciency is about 28.35%. The corresponding CH4 inlet flow rate,
TSMR, and TWGS are 50 kmol/hr, 780°C and 278°C, respec-
tively. The lower bound can be explained because the fCH4 term
appears in the denominator of the efficiency, which has a
dominant effect on the objective value. TSMR is at the upper Figure 6. The composite curves of the system with a
bound because the SMR reaction is decisive to convert CH4 to steam to carbon ratio of 4:1 and TSMR of 780°C,
H2. Since the SMR reaction is endothermic, higher reaction and TWGS of 258°C.
temperature benefits the conversion rate a lot. Also from Table
(b) is an expansion of (a) near the pinch points. [Color figure
2, we observe a big difference between the utility requirements can be viewed in the online issue, which is available at
of the system with and without heat integration. The utility www.interscience.wiley.com.]

2502 DOI 10.1002/aic Published on behalf of the AIChE July 2006 Vol. 52, No. 7 AIChE Journal
 Table 7. Comparison of the System Profit Optimization Results Between Sequential Optimization (SequOpt) and
Simultaneous Optimization (SimulOpt)
LB f Ch 4* T SMR * T WGS * Profit_SequOpt Profit_SimulOpt $ Gain With
(kmol/hr) (kmol/hr) (°C) (°C) ($/hr) ($/hr) SimulOpt
50.00 50.00 713.00 200.00 358.79 370.70 3.32%
100.00 100.00 615.25 200.00 212.08 232.79 9.77%
150.00 150.00 589.83 200.00 62.37 79.31 27.15%

冘 f 关max兵0, t
nc utes to the reason why the optimal value of fCH4 hit the lower
Qs ⱖ j
out
j ⫺ 共T p ⫺ ⌬T min兲其 ⫺ max兵0, t jin ⫺ 共T P bound. And we observe a sharp decrease in the profit. For the
j⫽1 case that fCH4 assumes the lower bound of 150 kmol/h, the

冘 F 关max兵0, T
nH system loses its profitability without heat integration; with heat
⫺ ⌬T min兲其兴 ⫺ i
in
i ⫺ T p其 ⫺ max兵0, T iout ⫺ T p其兴 integration it is still profitable. From these results, we can
i⫽1 conclude that a steam to carbon ratio of 4:1 is the optimal
Qw ⫽ Qs ⫹ ⍀ operating condition for this system, since we not only achieve
the largest profit, but almost the highest energy efficiency (58%
冘 F 共T 冘 f 共t
nH nC

⍀⫽ in
⫺ T iout兲 ⫺ out
⫺ t jin兲 in this case). For comparison purposes, the system profit with-
i i j j
i⫽1 j⫽1
out heat integration was listed in Table 5. In Table 6 we made
a brief comparison between the results with and without heat
LB ⱕ f CH4 ⱕ 200
integration. With heat integration, we can increase the profit by
420 ⱕ T SMR ⱕ 780 at least 29%. The corresponding curves are shown in Figure 5.
200 ⱕ T WGS ⱕ 400 (11) We observe a significant decrease in profit since utility cost
occupies a big portion of the total cost. This is the key contri-
We initially set the lower bound of fCH4 to be 50.0 kmol/h. The bution from heat integration.
optimal fCH4 hits the lower bound. Then we raise the lower As with system efficiency, we also performed a sequential
bound in the following two cases, that is, 100.0 and 150.0 optimization study of the system profit problem. The results are
kmol/h, respectively. As seen in Table 4, the optimal fCH4 hits shown in Table 7. The optimal values for temperature of WGS
the lower bound in both cases. The methane cost occupies all hit the lower bounds, and the optimal SMR temperatures are
approximately 25% of the revenue when fCH4 is around 50.0 correspondingly lower than the results from simultaneous op-
kmol/h; as we increase the lower bound of fCH4, the cost timization. Simultaneous optimization improves the sequential
percentage for methane increases tremendously. This contrib- optimization results in a range from 3.32% to 27.15%. This

Figure 7. The suggested HEN for the system with a steam to carbon ratio of 4:1 and TSMR of 627°C, and TWGS of 258°C.
Heat duties (kW) are given for each exchanger. [Color figure can be viewed in the online issue, which is available at www.interscience.
wiley.com.]

AIChE Journal July 2006 Vol. 52, No. 7 Published on behalf of the AIChE DOI 10.1002/aic 2503
Figure 8. Part of the flowsheet that converts water feed
(25°C) to steam (134°C) at pressure of 3 bar.
[Color figure can be viewed in the online issue, which is
available at www.interscience.wiley.com.]

again shows the significance in carrying out the simultaneous

optimization study.
The composite curves for the optimal operational conditions
for fCH4 at 50 kmol/h are shown in Figure 6. Here, we can see
that the heat exchange network utilizes most of the heat, except
for the heat generated by the fuel cell stacks to cover the heat
sink from the cold streams, other than the SMR reactor. Ex-
ternal cold utility should be provided mainly for fuel cell
stacks. Similarly, external hot utility should be used for the
SMR reactor. If we expand the curves near the pinch point,
then we can clearly see more detailed heat integration behavior.
Here, the temperature of the WGS (258°C) and the PROX
(200°C) are the two pinch points, which are the bottlenecks in
developing the HEN. The corresponding HEN is shown in
Figure 7.

Case Study with Water Instead of Steam as

Process Feed
Up to this point, we chose steam as the process feed. Con-
sidering the fact that water at 25°C might also be used as the
feed, we provide our study results with water feed as a com- Figure 9. The composite curves of the system with a
parison. In this case, as is shown in Figure 8, we first raise the steam to carbon ratio of 4:1, TSMR of 780°C,
temperature of the water from 25°C to 133°C, which is very and TWGS of 233°C.
close to the boiling point (133.5°C) at 3 bar using heat ex- (b) is an expansion of (a) near the pinch points. [Color figure
can be viewed in the online issue, which is available at
changer HEX9. Then, we realize the state conversion of water www.interscience.wiley.com.]
to steam via HEX10. Hence, according to our previous defini-
tion, we add two cold streams (C5 and C6) to the whole
composite curves in Figure 9. We can see the effect of the
flowsheet that requires heat. That is the only difference be-
water feed on the heat integration clearly from the composite
tween this flowsheet and the previous one. Since the change in
curves. The pinch point is now 90°C. And we can now utilize
the cold streams would affect the results of heat integration, we
the heat generated from the fuel cell to cover part of the heat
expect that the results for the optimal system efficiency and
requirement for converting the water liquid to steam. Due to
profit would change accordingly, as shown below.
the additional two cold streams in the water feed case, the total
heat requirement increases significantly. We can see the opti-
Optimal System Efficiency with Water Feed mal efficiency decreases by 2.72 percent (from 21.61% in the
The problem statement is the same as Eq. 10. We provide steam feed case to 18.89%) without heat integration. However,
our optimization results in Table 8. Compared to steam feed, with heat integration, the optimal efficiency only decreases by
the optimal system efficiency is slightly lower for cases with 1.38 percent (from 28.35% to 26.97%), which is around half of
and without heat integration. The optimal SMR reactor tem- the decrease in the case without heat integration. This reflects
perature still hits the upper bound. However, the optimal WGS the importance of heat integration in adjusting to the change of
reactor temperature is 233°C instead of 278°C in the steam operating conditions by trying to maintain the most efficient
feed case. way to operate the system.
To see the details of the heat integration, we draw the The HEN is shown in Figure 10. The hot and cold steams are

Table 8. System Efficiency Optimization Results With Water Feed for Cases With and Without Heat Integration (HI) With
LB ⴝ 50
f CH 4* T SMR * T WGS * Pcell Qs Qw
Cases (kmol/hr) (°C) (°C) (kW) (kW) (kW) Efficiency
With HI 50 780 232.56 6513.74 7664.49 5284.58 26.97%
Without HI 50 780 232.56 6513.74 10469 12849 18.89%

2504 DOI 10.1002/aic Published on behalf of the AIChE July 2006 Vol. 52, No. 7 AIChE Journal
Figure 10. The suggested HEN for the system with a steam to carbon ratio of 4:1, TSMR of 780°C, and TWGS of 233°C.
Heat duties (kW) are given for each exchanger. [Color figure can be viewed in the online issue, which is available at www.interscience.
wiley.com.]

“more” integrated in the sense that only hot stream H8 is totally as the lower bound for the methane inlet, the optimal WGS
dependent on the external cold utility while the rest of the hot reactor temperature hits the upper bound (400°C).
streams either totally or partially utilized the internal cold
streams. This also explains why the optimal efficiency of the Conclusions
system doesn’t decrease much with two additional cold streams
We perform the simulation and optimization of a typical
in the water feed case.
power generation system using ASPEN Plus. The CH4 inlet
flow rate, the WGS reactor temperature, and the temperature of
Optimal System Profit with Water Feed the SMR reactor are the main decision variables in our study.
The problem statement is the same as Eq. 12. The optimal The optimization is concentrated on efficiency improvement
solution is provided in Table 9. Optimization without heat and profit maximization, which are goals of the current fuel cell
integration and a comparison are in Tables 10 and 11, respec- development. We also perform a sensitivity case study on the
tively. Compared to the steam feed case, the optimal system fuel cell performance characterized by the power output. Since
profit is correspondingly lower. Both the composite curves in the water feed and steam feed are both considered in industrial
Figure 9 and HEN in Figure 10 still apply here in the case with systems, we provide results for both of them as a comparison.
50.00 methane flow rate. In the cases with 100.00 and 150.00 From the results, we can see the importance of simultaneous

Table 9. System Profit Optimization Results With Heat Integration in the Water Feed Case
LB ⫽ f *CH 4 T *wgs P cell Energy Qs Qw Profit
(kmol/hr) T *smr (°C) (°C) (kW) Efficiency (kW) (kW) ($/hr)
50.00 780.00 232.56 6513.74 58.37% 7664.49 5284.58 333.52
100.00 667.24 400.00 6699.99 30.02% 7508.09 7336.77 206.31
150.00 632.61 400.00 6667.78 19.92% 8069.99 7401.37 51.25

Results are listed for cases where the lower bound for f CH4 was set to be 50.00, 100.00, and 150.00, respectively.

Table 10. System Profit Optimization Results Without Heat Integration in the Water Feed Case
LB ⫽ f *CH 4 T WGS * Qgen Qsink TotalQ
(kmol/hr) T *SMR (°C) (°C) (kW) (kW) (kW) Profit_NoHI
50.00 780.00 232.56 10469.00 12849.00 23318.00 279.75
100.00 667.24 400.00 13793.69 13965.00 27758.69 102.31
150.00 632.61 400.00 14099.69 14768.00 28867.69 ⫺50.25

AIChE Journal July 2006 Vol. 52, No. 7 Published on behalf of the AIChE DOI 10.1002/aic 2505
 Table 11. Comparison of the System Profit Optimization Results Between Cases With and Without Heat Integration for
Water Feed
LB ⫽ f CH 4 T *SMR T WGS * Profit_NoHI Profit Percentage
(kmol/hr) (°C) (°C) ($/hr) ($/hr) Gain With HI
50 780.00 232.56 279.75 333.52 19.22%
100 667.24 400 102.31 206.31 101.66%
150 632.6141 400 ⫺50.25 51.25 N/A

optimization with heat integration in balancing the heat re- trochemical reactions in polymer electrolyte fuel cells. J Power
quirement of the system and adjusting to the change of the Sources. 2004;125:40.
8. Wang ZH, Wang CY. Mathematical modeling of liquid-feed direct
operational conditions. The significance of simultaneous opti- methanol fuel cells. J Electrochem Soc. 2003;150:A508.
mization is illustrated by comparing with the results from 9. Bernardi DM, Verbrugge MW. A mathematical model of a gas diffu-
sequential optimization in system efficiency and profit optimi- sion electrode bonded to a polymer electrolyte. AIChE J. 1992;37:
zation with steam feed. 2477.
As future work, we will generalize our model to include 10. Friedman DJ, Moore RM. Optimum operating curves for fuel cell
systems. Record of the 1st Annual Fuel Cell Vehicle Technology
more detailed fuel cell and reactor models. Also, the capital Conference, University of California, Davis, CA, 1998.
costs of the process flowsheet and heat integration network 11. Xu C, Follmann PM, Biegler LT, Jhon MS. Numerical simulation and
were not considered, which might influence our simulation and optimization of a direct methanol fuel cell. Computers Chem Eng.
optimization results. In a more complete study, a mixed integer 2005;29(8):1849-1860.
nonlinear programming (MINLP) formulation will be very 12. Godat J, Marechal F. Optimization of a fuel cell system using process
integration techniques. J Power Sources. 2003;118:411.
helpful to determine the selection of process equipment, the 13. Marechal F, Palazzi F, Godat J, Favrat D. Thermo-economic modeling
HEN, and the utility system. and optimization of fuel cell systems. Fuel Cells. 2004;5, 1:5-24.
14. Aspen Plus Documentation Version 11.1. Cambridge, MA: Aspen
Technology, Inc.; 2001.
Acknowledgments 15. Larminie J, Dicks A. Fuel Cell Systems Explained. New York: John
PITA (Pennsylvania Infrastructure and Technology Alliance) adminis- Wiley; 2000.
tered by the Institute of Complex Engineered Systems (ICES) at Carnegie 16. Duran MA, Grossmann IE. Simultaneous optimization and heat inte-
Mellon University is gratefully acknowledged for the financial support of gration of chemical processes. AICHE J. 1986;32:123.
this project. 17. Lang YD, Biegler LT, Grossmann IE. Simultaneous optimization and
heat integration with process simulators. Computers Chem Eng. 1988;
12:311.
Literature Cited 18. Yee TF, Grossmann IE, Kravanja Z. Simultaneous optimization mod-
1. Garman DK. Fuel Cell Report to the Congress (Esecs Ee-1973), 2003. els for heat integration. III. Optimization of process flowsheets and
2. Gemmen RS, Selman JR. Speeding the progress of fuel cell develop- heat exchanger networks. Computers Chem Eng. 1990;14:1185.
ment. Proceedings of the NETL Workshop on Fuel Cell Modeling, 19. Balakrishna S, Biegler LT. Targeting strategies for the synthesis and
Morgantown, WV, 2000. energy integration of nonisothermal reactor networks. Ind Eng Chem
3. Fuller TF, Newman J. Water and thermal management in solid poly- Res. 1992;31:2152.
mer electrolyte fuel cells. J Electrochem Soc. 1993;140:1218. 20. Biegler LT, Grossmann IE, Westerberg AW. Systematic Methods for
4. Nguyen TV, White RE. A water and heat management model for Chemical Process Design. Upper Saddle River, NJ: Prentice Hall;
proton exchange membrane fuel cells. J Electrochem Soc. 1993;140: 1997.
2178. 21. Lutz AE, Bradshaw RW, Keller JO, Witmer DE. Thermodynamic
5. Springer TE, Rockward T, Zawodzinski TA, Gottesfeld S. Model for analysis of hydrogen production by steam reforming. Int J Hydrogen
polymer electrolyte fuel cell operation on reformate feed: effects of Energy. 2003;28:159.
CO, H2 dilution, high fuel utilization. J Electrochem Soc. 2001;148(1): 22. IES Group. Process Utilities Cost Guide 2002. Available at http://
A11. ed.icheme.org/costutil.html.
6. Springer TE, Zawodzinski TA, Gottesfeld S. Polymer electrolyte fuel
cell model. J Electrochem Soc. 1991;138:2334.
7. Um S, Wang CY. Three-dimensional analysis of transport and elec- Manuscript received Mar. 13, 2005, and revision received Jan. 31, 2006.

2506 DOI 10.1002/aic Published on behalf of the AIChE July 2006 Vol. 52, No. 7 AIChE Journal