www.elsevier.com/locate/ijhydene
Dynamic modeling and control structure design of an experimental
fuel processor
Shi-Tin Lin
a, Yih-Hang Chen
b, Cheng-Ching Yu
a,∗, Yen-Chun Liu
c, Chiou-Hwang Lee
caDepartment of Chemical Engineering, National Taiwan University, Taipei 106-17, Taiwan
bAdvanced Energy Technology Laboratory, Industrial Technology Research Institute, Hsinchu 300, Taiwan
cUnion Chemical Laboratory, Industrial Technology Research Institute, Hsinchu 300, Taiwan
Received 7 January 2005; received in revised form 13 June 2005; accepted 20 June 2005 Available online 15 August 2005
Abstract
In this work, a dynamic model is developed to describe an experimental methane fuel processor which is intended to provide hydrogen for a proton exchange membrane fuel cell (PEMFC) for power generation (2–3 kWe). First-principle reactor models were constructed to describe dynamic behavior for a series of reactions, starting from reforming (SR/ATR), to high- and low-temperature water gas shift reactions (HTS/LTS), and then to preferential oxidation (PROX) reactions. A systematic procedure is proposed to identify dynamic-relevant model parameters, and reasonable behavior description can be obtained. Finally, two plantwide control structures, on-demand structure and on-supply structure are designed and the performance of these two control structures is evaluated for load disturbance rejection. The results indicate that the on-demand control structure gives a rapid transition to different power demands.
䉷 2005 International Association for Hydrogen Energy. Published by Elsevier Ltd. All rights reserved.
Keywords: Proton exchange membrane fuel cell; Fuel processor; Process dynamics; Plantwide control; Autothermal reforming; Steam reforming
1. Introduction
Fuel cell systems offer high potential for efficiency and
reduced emissions in power generation[1]. The proton
ex-change membrane fuel cell (PEMFC) is one of the most popular fuel cell systems in which fuels such as methanol or methane are converted to hydrogen-rich syn-gas in a re-former and which is subsequently used in the fuel cell stack. In addition to the reformer, a series of CO reducing steps, water gas shift reactions and preferential oxidation reac-tions were taken to keep CO concentrareac-tions below 100 ppm before the syn-gas enters the cell stack. This combination
∗Corresponding author. Tel.: +886 2 33 65 1759; fax: +886 2 33 66 3037.
E-mail address:[email protected](C.-C. Yu).
0360-3199/$30.00䉷2005 International Association for Hydrogen Energy. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.ijhydene.2005.06.027
constitutes the entire fuel processor[2]. A dynamic model is essential for the fuel processor operation for the following reasons: (1) discriminating control system design for im-proved load rejection, and (2) evaluating start-up strategies for fast start-up.
Extensive literature has examined various aspects of fuel processor systems for hydrogen-rich syn-gas production, which include overviews of the fuel processing technology [3–6], in which the reforming technology of hydrocarbon fuels is still the major focus. Steady-state simulations are often performed for sensitivity analyses in the design and
operation phases of the fuel processor[7–10]. Studies on
dynamic behavior of the fuel processor have received some attention lately, and typically the relationships between feed conditions and dynamic responses were explored in [11,12]. The start-up dynamics was explored in [13] in order to devise a more efficient start-up strategy. Literature
Nomenclature
CP heat capacity of the gas(kJ mol−1K−1) CP,S heat capacity of the carrier(kJ g−1K−1) CP w heat capacity of the metal reforming reactor
wall(kJ g−1K−1)
DI inner diameter of the reactor (cm)
Do outer diameter of the reforming reactor (cm) F total molar flow rate(mol min−1)
kcond thermal conductivity of the metal reactor wall (kJ min−1cm−1K−1)
mw weight of the reforming reactor (g) MF molar holdup of the burner (mol)
P pressure (atm)
QF heat input for preheating(kJ min−1) r rate of reaction
T reaction temperature (K) Tw reactor wall temperature (K) VR volume of the gas (ml)
TA surrounding temperature (K) Tf temperature of the feed (K)
Tin inlet temperature of the reformer (K) TH1 inlet temperature of the HTS1 (K) TH2 inlet temperature of the HTS2 (K) TL inlet temperature of the LTS (K) TP inlet temperature of the PROX (K)
U heat transfer coefficient (kJ min−1cm−2 K−1)
WS carrier weight (g) Wcat catalyst weight (g)
y mole fraction
HR heat of reaction of reaction(kJ mol−1) density of the carrier(ml g−1)
av average density of the gas in the reforming reactor(mol ml−1)
stoichiometric coefficient of the reaction
for dynamics and control of fuel processor is scattered in conference proceedings with only a handful of journal pa-pers.
The objective of this work is to construct a dynamic model for a methane fuel processor and different control structures can be evaluated based on the disturbance rejection capa-bility. The remainder of this paper is organized as follows. Section 2 describes the process and dynamic modeling of the fuel processor. Sensitivities of operating parameters are explored in Section 3, followed by control structure design and evaluation in Section 4. The conclusion is drawn in Section 5.
2. Process studied 2.1. Reaction kinetics
A fuel processor consists of several reactors, heat ex-changers and cooling devices (direct waters injection). It can be viewed as a small chemical plant with a series of reac-tors for reforming and gas cleaning.Fig. 1shows the exper-imental setup of the fuel processor in the facility of Union Chemical Laboratory (UCL) of the Industrial Technology Research Institute (ITRI); the corresponding dimensions are also given. The experimental fuel processor is simplified to a reformer, a burner, three water gas shift reactors and a preferential oxidation reactor for the modeling purpose (Fig. 2). Methane, air and water were fed into the reformer to carry out autothermal reforming (ATR). The reformer was integrated with a burner which had the function of preheat-ing the feed and supplypreheat-ing heat needed for the reactions. Table 1shows the reactions that occurred in the reformer
[7], wherer1is an endothermic reaction whiler2andr3are
exothermic reactions.
Fig. 1. Experimental fuel processor and corresponding dimensions.
The effluent of the reformer is passed through a feed-effluent heat exchanger, followed by a liquid water injection to cool the temperature down to the desired HTS1 inlet temperature. In fact, the hydrogen-rich syn-gas goes through a series of reactors to perform the water gas shift reaction (HTS1, HTS2 and LTS) in which CO was removed to meet the specification. Because of the monotonically decreasing arrangement of the temperature profile for the water gas
Reformer HTS1 HTS2 LTS PrOx O2 CH4 H2O H 2 H2O TF Fuel(CH4) Exhaust gas H2O H2O H2O O2 H2O
Fig. 2. Process configuration of the experimental fuel processor.
Table 1
Reaction rate expressions and parameter values for the fuel processor system
Reactor Reaction Kinetics HR(kJ mol−1)
ATR CH4+ H2O⇔ CO + 3H2 r1= k1PCH4PH2O−k1PCOPH23 P2.5 H2(1)2 206 CO+ H2O⇔ CO2+ H2 r2= k2PCOPH2O−k2PCO2PH2 PH2(1)2 −41.2 CH4+ 2O2⇒ CO2+ 2H r3= k3aPCH4PO2 (2)2 + k3bPCH4PO2 (2) 810 HTS1 CO+ H2O⇔ CO2+ H2 rHTS= kHPCOPH2O− kHPH2PCO2 −41.2 HTS2 LTS CO+ H2O⇔ CO2+ H2 rLTS= kLPCOPH2O− kLPH2PCO2 −41.2
PROX CO+12O2⇒ CO2 rPROX1= kCOPCO −283
H2+12O2⇒ H2O rPROX2= 1.5rCO −243 1=(1+KCOPCO+KH2PH2+KCH4PCH4+KH2OPH2O/PH2),2=(1+K OX CH4PCH4+K OX O2PO2), with KCH4=6.65×10−4e(4607/Tn), KCO= 8.23 × 10−5e(8504/Tn), KH2= 6.12 × 10−9e(9971/Tn), KH2O= 1.77 × 10 5e(−10669/Tn), KOX CH4 = 1.26 × 10−1e(3284/Tn) and KOX O2 = 7.87 × 10−7e(11162/Tn).
shift reactors, liquid water injection devices were installed between reactors. The reaction that takes place in the wa-ter gas shift reactors is the same as r2 except that a dif-ferent type of catalyst is used (shown in Table 1). In this work, we use the rate expression of Choi and Stenger[9] for kinetics expression. Corresponding rate constants ob-tained from the regression of the steady-state data are shown inTable 2.
Generally, the CO concentration out of the LTS was still too high, so the preferential oxidation reaction (PROX) was
performed. An oxygen(O2) injection device was installed
at the inlet of PROX, and then CO was further oxidized to
CO2, while, simultaneously, H2was oxidized to H2O. Note
that a H2O stream is injected right before the PROX to bring
the temperature from 241◦C (LTS outlet) down to 150◦C
(PROX inlet). Both reactions in the PROX are exothermic
reactions. The rate expressions are given in Table 1 and
corresponding parameter values are shown in Table 2for
the entire fuel processor.
Table 2
Regression reaction kinetic data for experimental fuel processor Reactor Reaction Pre-exponential Activation
factora0 energy (kJ mol−1) ATR r(1), forwardk1 6.32 × 1016 240.1 r(1), reversek1 1.759 × 103 17.0 r(2), forwardk2 2.77 × 106 67.1 r(3), forwardk3a 1.56 × 108 86.0 r(3), forwardk3b 1.31 × 108 86.0 HTS rHTS, forwardkH 9.886 × 105 47.4 rHTS, reversekH 1.32 × 10−2 38.1 LTS rLTS, forwardkL 1.285 × 106 47.4 rLTS, reversekL 1.32 × 10−2 38.1 PROX rPROX, forwardkPROX 1.34 × 104 8.3
2.2. Dynamic modeling
Homogeneous reactor models were set up to describe the dynamic behavior of the experimental fuel processor. The assumptions and simplifications made for the system are: (1) Constant pressure in the fuel processor (1 bar). (2) Ideal gas behavior for each component.
(3) Temperature of the vapor and solid phase being the same.
(4) Thermal capacitance of the gas in the reactor negligible as compared to that of the solid catalyst and carrier. (5) Negligible heat loss for the HTS1, HTS2, LTS and
PROX reactors.
Partial differential equations describing the energy and mass balances are lumped intoN sections using the hon-eycomb carrier weight (WS) as the independent variable (Note that this also applies to the catalyst weightWcat, be-cause it is assumed that the catalyst is distributed uniformly throughout the carrier). Thus, we haveN ordinary differen-tial equations instead of one pardifferen-tial differendifferen-tial equation for each component balance[15]. Consider the nth section in the axial direction.
The energy balance of the reformer can be expressed as
CP,SWS,n dTn dt = Fn−1CP,n−1Tn−1− FnCP,nTn − Wcat,n j H0 R,jrn,j − 4U(Tn− Tw,n)WS,n/(SDI). (1) HereCP,Sis the heat capacity of the carrier,WS,ndenotes the weight of the carrier, andn represents the nth lump. Tn andTw,nrepresent the reaction temperature and reactor wall temperature in the nth lump, respectively. Fn is the total molar flow rate at the nth lump CP,n is the heat capacity of the gas in thenth lump and Wcat,nis the weight of the catalyst in thenth lump. rn,j is the reaction rate of thejth reaction at thenth section, HR,jis the heat of reaction for the jth reaction, U is the overall heat transfer coefficient, DIis the inner diameter of the reactor andSis the density of the carrier. The component material balance for the ith composition becomes avVR,n dyn,i dt = Fn−1yn−1,i− Fnyn,i + Wcat,n j ijrn,j, (2) whereavis the averaged density of the gas in the reforming reactor.VR,n is the volume of the gas in thenth lump of the reformer,i,j is the stoichiometric coefficient of theith component underjth reaction and yn,i is the mole fraction of theith component in the nth lump.
The reactor metal wall temperature is also lumped as follows: mw,nCP w,n dTw,n dt = kcondA(Tw,n−1− Tw,n) + 4U(Tn−Tw,n)WS,n/(DI) + 4U(TA−Tw,n)WS,n/(Do), (3) wheremw,n is the weight of the metal reactor wall in the nth lump, CP w,n is the heat capacity of the metal reactor wall,kcond is the thermal conductivity of the reactor wall andDois the outer diameter of the reforming reactor.TA represents the ambient temperature.
The inlet of the reformer is heated by a burner and the temperature can be expressed as
CP,avgMF
dTin
dt = QF+ F
yiCP,i(Tf− Tin), (4)
whereCP is the heat capacity of the feed andMF is the molar holdup of the burner.QFis the heat needed for pre-heating andTf is the temperature of the fresh feed, which is assumed to be 25◦C.
The relationship between the reactor inlet temperatureTin (Tois the lumped notation) and the reactor wall temperature at the inlet (Tw,ois the lumped notation) is established from a regression model of the formTw,o= 1.65 To− 864. The energy balance equations describing the burner provide the inlet conditions for the reformer gas and metal wall tem-peratures. The composition and temperature profiles can be evaluated by solving these ordinary differential equations.
Similarly, the equations describing water gas shift reactor and PROX can be derived. The energy balance equation becomes CP,SWS,n dT dt = Fn−1CP,n−1Tn−1− FnCP,nTn − Wcat,n j H0 R,jijrn,j, (5)
and the component material balance equation is avVR,n dyn,i dt = Fn−1yn−1,i− Fnyn,i + Wcat,n j ijrn,j. (6) The modeling equations of the HTS1, HTS2, LTS and PROX were assumed to be adiabatic and they were simpler than the modeling equation of the reformer.
The rate expressions of the reactions(rj) that take place in the fuel processor were obtained form the regression of the experimental data as shown inTable 1and the parameter
values are given inTable 2.Table 3summarizes the
steady-state operating condition with a H2O/CH4 feed ratio of
1.45, O2/CH4feed ratio of 0.45 while the reformer inlet
temperature was set to 717◦C (high temperature constraint
Table 3
Steady-state operating condition(mol min−1) for fuel processor system
T CH4 H2O CO CO2 O2 H2 N2 (◦C) ATR Inlet 717 0.480 0.696 0 0 0.216 0 0.812 Outlet 650 0.015 0.413 0.274 0.191 0.03 1.21 0.812 Cooling (ATR) — — 0.486 — — — — — HTS1 Inlet 350 0.015 0.898 0.274 0.191 0.03 1.210 0.812 Outlet 402 0.015 0.710 0.081 0.390 0.03 1.410 0.812 Cooling (HTS1) — — 0.200 — — — — — HTS2 Inlet 317 0.015 0.910 0.081 0.390 0.03 1.410 0.812 Outlet 322 0.015 0.880 0.059 0.410 0.03 1.427 0.812 Cooling (HTS2) — — 0.220 — — — — — LTS Inlet 237 0.015 1.110 0.059 0.410 0.03 1.427 0.812 Outlet 241 0.015 1.090 0.043 0.420 0.03 1.440 0.812 Cooling (LTS) — — 0.270 — — — — — PROX Inlet 150 0.015 1.360 0.043 0.420 0.03 1.440 0.812 Outlet 320 0.015 1.360 1.3 × 10−4 0.470 0.011 1.380 0.812
Table 4
Catalyst weight, heat capacities and dimensions for the reactors
Wcat WS CP Length Diameter Carrier volume
(g) (g) (J g−1K−1) (cm) (cm) (ml) ATR 80 3033.6 0.221 12.0 10.2 700 HTS1 190 1555.4 0.767 9.0 12.0 981 HTS2 106 1571.0 0.847 7.5 12.0 820 LTS 250 3000.0 0.419 10.0 12.0 600 PROX 300 2057.1 0.814 16.5 12.0 1801
Fig. 4. Start-up dynamics for the SR pathway form dynamic simulation showing all flow rates and compositions at the outlet stream of the reformer.
corresponds to the fuel processor with a maximum efficiency of 68.4%[14].
2.3. Parameter fitting
In order to describe the temperature and composition dy-namics, it is necessary to modify some of the model param-eters in the dynamic model (Eqs. (1–6)). Because the hon-eycomb catalyst is quite complex and enormous, instead of modeling the honeycomb details, the heat capacity of the solid carrier (Eq. (1)) was adjusted to meet the dynamic tra-jectory of the reformer outlet temperature. Similarly, for the composition profiles, the gas holdup in the reformer was also adjusted to meet the dynamic trajectory of the compo-sition of each component. From the process dynamics point
of view, this is similar to adjusting the time constant of a transfer function which should be effective to obtain dy-namical behavior description. Thus, the dynamic modeling of the fuel processor consists of the following steps:
1. Obtain feed condition and the heat input (QF) to the burner.
2. AdjustCP S,WS,n(Eq. (1)) to meet the dynamic trajec-tory of the reformer inlet temperature.
3. AdjustavVR,n to meet the dynamic trajectory of the conversion of CH4 and CO concentration out of the reformer.
4. Adjust themw,nCP w,nof the metal reactor wall to find the dynamic trajectory reactor wall temperature.
Fig. 3shows the ATR pathway start-up trajectories of the reformer inlet temperature, molar flow rate of each
compo-nent, CO concentration and the conversion of CH4predicted
by the modified dynamic model. The simulation results give good behavior description of the experimental data given by UCL of ITRI. The reformer inlet temperature and out-let flow rate of each component predicted by the dynamic model were practically the same as that of the experiment. The steady-state offset in the reformer outlet temperature comes from the mismatches between temperature and
com-position profiles[14], which can be the results of the heat
loss which was unaccounted for.Table 4gives the heat
ca-pacities, solid and gas holdup in the fuel processor from the regression. In the experiment, the reformer switches
0 10 20 30 40 8 10 12 QF x10 4 (J/min) 0 10 20 30 40 500 1000 TReformer,in ( °C) 0 10 20 30 40 20 40 60 YCO,PROX ppm 0 10 20 30 40 0 2 4 H2 O/C ratio 0 10 20 30 40 1 1.5 2 FH 2 ,PROX (mol/min) 0 10 20 30 40 0 0.05 0.1 FCH 4 ,PROX (mol/min) time (min) 0 10 20 30 40 0 1 2 O2 /C ratio 0 10 20 30 40 0 0.5 1 FCH 4 ,Feed (mol/min) time (min)
Fig. 5. Open-loop dynamic responses for±20% changes of CH4 feed flow rate (solid line for +20% increase and dashed line for−20% change).
to the SR pathway after the ATR start-up, so no start-up experimental data were available for the SR pathway. How-ever, the dynamic model enables us to study the dynamics with the SR start-up. Fig. 4 reveals that, as expected, the dynamic responses for the SR pathway are slower than that of the ATR pathway because of higher reac-tion temperature and endothermic nature of the reacreac-tion mechanism.
Due to the lack of complete information for the dynamic behavior of the gas cleaning unit, we scale the dynamic of each reactor proportional to the size of the reactors. So the dynamic models of HTS1, HTS2, LTS and PROX were also set up to explore dynamics and control of the fuel processor.
0 10 20 30 40 8 10 12 0 10 20 30 40 600 700 800 TReformer,in (° C) 0 10 20 30 40 40 45 50 YCO,PROX (ppm) 0 10 20 30 40 1 1.5 2 0 10 20 30 40 1.3 1.4 1.5 FH 2 ,PROX (mol/min) 0 10 20 30 40 0 0.02 0.04 FCH 4 ,PROX (mol/min) time (min) 0 10 20 30 40 0 0.5 1 time (min) 0 10 20 30 40 0 0.5 1 QF x10 4 (J/min) H2 O/C ratio O2 /C ratio FCH 4 ,Feed (mol/min)
Fig. 6. Open-loop dynamic responses for±20% changes of H2O/CH4feed ratio (solid line for +20% increase and dashed line for−20% change).
3. Steady-state and dynamical sensitivities
The effects of process variables to the operation of the fuel processor are explored here. Important process vari-ables include CH4 feed flow rate, H2O/CH4 feed ratio, O2/CH4feed ratio and the reformer inlet temperatureTin. The H2O/CH4feed ratio and O2/CH4feed ratio used here denote the ratio of two corresponding feed streams into the reformer (Fig. 1). Here, we are interested in the steady state as well as dynamic aspects of these sensitivities. Note that the simulations were carried out for the entire fuel processor system, including reformer, HTS1, HTS2, LTS and PROX
reactors (Fig. 2).
Open-loop step changes were made to explore the effects
of these manipulated variables.Fig. 5presents dynamic
re-sponses for±20% changes in CH4feed flow rate att =8 m,
when the H2O/CH4feed ratio O2/CH4feed ratio and the heat input to the burner(QF) were kept constant.Fig. 5also
shows that when the CH4feed flow rate increases by 20%,
the H2production rate goes through a fast increase and then
reaches a steady state with a 10% increase in the H2yield.
The reformer inlet temperature(Tin) also decreases due to
the fact that the heat inputQFremains unchanged. The CO
concentration also increases by a factor which is similar to
that of the H2production rate. The opposite behavior was
observed for a negative change in the CH4 feed flow rate
and nonlinear responses were observed.
Fig. 6 presents the open-loop responses for ±20%
changes in H2O/CH4 feed ratio, while keeping the CH4
feed flow rate, O2/CH4 feed ratio and QF constant. As
the H2O/CH4feed ratio increases, the H2production rate
How-0 10 20 30 40 8 10 12 QF x10 4 (J/min) 0 10 20 30 40 600 700 800 TReformer,in (° C) 0 10 20 30 40 40 45 50 YCO,PROX (ppm) 0 10 20 30 40 0 2 4 H2 O/C ratio 0 10 20 30 40 1.2 1.4 1.6 FH 2 ,PROX (mol/min) 0 10 20 30 40 0 0.02 0.04 FCH 4 ,PROX (mol/min) time (min) 0 10 20 30 40 0.2 0.4 0.6 O2 /C ratio time (min) 0 10 20 30 40 0 0.5 1 FCH 2 ,Feed (mol/min)
Fig. 7. Open-loop dynamic responses for±20% changes of O2/CH4feed ratio (solid line for +20% increase and dashed line for−20% change).
ever, the H2production rate is relatively insensitive to the H2O/CH4ratio change (cf.Figs. 5and6).Fig. 6also gives
the dynamic response of the CO concentration. In
compari-son, the dynamic responses of H2O/CH4feed ratio change
were faster than that of the CH4 feed flow rate, because
only feed water flow rate was changed for the former case,
while both CH4feed flow and water feed flow were varied
in the later one.
Fig. 7shows the open-loop responses for±20% change
in O2/CH4feed ratio. Similar to the case of CH4feed rate
change, the H2production rate shows an increase, but only
by 5%. The CO concentration also shows a small increase
for an O2/CH4feed ratio increase. The speed of response
is similar to that of H2O/CH4feed ratio change.
The ongoing analysis indicates that the CH4flow rate is an important manipulated variable to handle H2production rate variation as compared to H2O/CH4and O2/CH4feed ratios. Either H2O/CH4or O2/CH4can be used to main-tain a safe CO concentration while showing little impact on H2production rate. Finally, instead of fixing theQF, the re-former inlet temperature should be controlled at a constant value to prevent temperature constraint violation.
4. Control structure design
In order to accommodate the load changes in a PEMFC, the hydrogen flow rate from the fuel processor should be adjusted to satisfy the power demand. The control objective
Fig. 8. On-supply control structure (CS1).
Fig. 9. On-demand control structure (CS2).
of the fuel processor is to provide smooth H2production rate changes while keeping the CO concentration at a safe level such that the CO concentration will not poison the catalyst of PEMFC. We would like to achieve the objective with the
simplest possible control structure. Two control structures were studied. The first control structure uses the methane feed flow rate as the throughput manipulator (TPM) which is denoted as the on-supply control structure (CS1)[16–19]. The other control structure uses the hydrogen production rate as the throughput manipulator, denoted as the on-demand control structure (CS2) hereafter.
4.1. On-supply control structure (CS1)
The open-loop tests clearly indicate that the CH4 feed
flow rate is effective in handling H2production rate
varia-tion. Thus, the CH4feed flow is a good candidate for the
TPM. However, it is less obvious that which one of two
ra-tios, H2O/CH4or O2/CH4, should be used to maintain the
CO concentration. In this work, the H2O/CH4ratio is
se-lected. The CS1 control structure has the following loops: 1. Use the CH4 feed flow rate as the throughput
manipulator.
2. Maintain the fuel processor outlet pressure by changing the CH4feed flow rate.
3. Use the H2O/CH4feed ratio to control the CO concen-tration.
4. Keep the reformer inlet temperature constant by adjusting the energy supply.
5. Control the HTS1, HTS2, LTS and PROX inlet temper-atures by changing the H2O addition.
6. Fix the O2/CH4feed ratio via a ratio control.
Fig. 8shows the CS1 with one pressure loop, one tem-perature loop, one composition loop, two ratio controllers, three flow controllers and four temperature controllers for the water addition. In the dynamic simulation, a third-order lag with a time constant of 0.1 min is assumed for the
composition analyzer. The relay feedback test[20]is used to
find the ultimate gain(Ku) and ultimate period (Pu). Then,
the Tyreus and Luyben[21]tuning rule is employed to find
the controller gain and the reset time for PI controllers.
4.2. On-demand control structure (CS2)
The on-demand control structure differs from CS1 in that
the H2 production rate is adjusted directly by the
down-stream demand. A change in the H2production rate leads to
a variation in the system pressure and, subsequently, the CH4
feed flow is changed. Therefore, the CS2 control structure consists of the following loops:
1. H2production rate is under flow control.
2. CH4feed flow rate is adjusted by the reformer pressure controller.
3. H2O flow rate is ratio to the CH4flow rate via RC1. 4. O2flow rate is ratio to the CH4flow rate via RC2. 5. H2O/CH4 feed ratio is adjusted by the CO2
0 10 20 30 40 0 10 20 QF x10 4 (J/min) 0 10 20 30 40 700 720 740 TReformer,in ( °C) 0 10 20 30 40 0 50 100 ppm CO,PROX 0 10 20 30 40 0 2 4 H2 O/C ratio 0 10 20 30 40 1 1.5 2 FH 2 ,PROX (mol/min) 0 10 20 30 40 0 0.02 0.04 FCH 4 ,PROX (mol/min) time (min) 0 10 20 30 40 0.45 O2 /C ratio time (min) 0 10 20 30 40 0 0.5 1 FCH 4 ,Feed (mol/min)
Fig. 10. Closed-loop responses for±20% H2production rate changes using CS1 with CO composition control (solid line for +20% increase and dashed line for−20% change).
6. Reformer inlet temperature is maintained by changing the fuel flow rate.
Fig. 9show the CS2 control with one temperature loop, one composition loop, two ratio loops, four flow loops and
three temperature loops for H2O addition. The controller
design follows the same steps as that of the CS1.
4.3. Results
Consider the case of±20% changes in the H2
produc-tion rate.Fig. 10shows the CH4feed rate goes through step
changes while the H2 production rate (FH2,PROX) shows
a first-order type of response and it takes approximately 10 min. to reach the new steady state. However, the entire composition profiles do not settle until 35 min. for the
pro-duction rate increase.Fig. 10also reveals that the H2O/CH4
ratio is adjusted to bring the CO concentration back to the
set point, and this is the result of the composition control. Because of the low sensitivity between the H2O/CH4 ra-tio, and CO concentration, it is likely to eliminate the com-position controller without violating the CO concentration constraint.
Fig. 11shows the closed-loop responses without the
com-position controller, i.e. the H2O/CH4 ratio is fixed at the
nominal value. Results indicate that we have quite similar
dynamics in the H2production rate and the CO
concentra-tion reaches 60 ppm for a 20% feed rate increase. Further-more, all the dynamics settle in 10 min, which is faster than
the previous case (Fig. 10). From the control complexity,
fast transition and constraint violation points of view, it is clear that the CO composition control cannot be necessary and it is eliminated for the subsequent development.
The on-demand control structure (CS2 in Fig. 9) is
ex-plored next. Fig. 12 shows that a very fast transition in
con-0 10 20 30 0 10 20 QF x10 4 (J/min) 0 10 20 30 700 720 740 TReformer,in ( °C) 0 10 20 30 0 50 100 YCO,PROX ppm 0 10 20 30 0 2 4 H2 O/C ratio 0 10 20 30 1 1.5 2 FH 2 ,PROX (mol/min) 0 10 20 30 0.01 0.02 0.03 FCH 4 ,PROX (mol/min) time (min) 0 10 20 30 0 0.5 1 O2 /C ratio time (min) 0 10 20 30 0 0.5 1 FCH 4 ,Feed (mol/min)
Fig. 11. Closed-loop responses for±20% H2 production rate changes using CS1 without CO composition control (solid line for +20% increase and dashed line for−20% change).
stant approximately equal to 2 min) without CO constraint violation. Tight reformer inlet temperature control can be achieved. But for a 20% decrease in the H2production rate, the temperature exceeds 720◦C monoterary and then back to the set point. ComparingFig. 12toFig. 11, it is obvious that the CS2 is a much better control structure to handle pro-duction rate variation, which is one of the most important disturbances for the fuel processor.
5. Conclusions
A systematical approach is proposed to model the dy-namic responses of an experimental fuel process. Rea-sonable behavior description can be obtained by adjusting
model parameters. Then, the control issue was addressed. The control objective of a fuel processing system is quite clear: provide responsiveness to the changes in hydrogen demand while keeping the carbon monoxide concentra-tion below 100 ppm. Two control structures are proposed. One uses the fuel feed flow rate as the throughput manip-ulator (TPM), which was called the on-supply structure (CS1), and the other uses the reactor outlet flow as the TPM, which was called the on-demand structure (CS2). In both control structures, reasonable control can be obtained while maintaining the CO at allowable level. Moreover, the composition control can be eliminated without possible constraint violation. Judging on the speed of response, the on-demand control structure (CS2) is an ideal candidate to provide rapid transition for load changes.
0 10 20 30 5 10 15 QF x10 4 (J/mol) 0 10 20 30 700 720 740 TReformer,in ( °C) 0 10 20 30 0 50 100 YCO,PROX ppm 0 10 20 30 0 2 4 H2 O/C ratio 0 10 20 30 1 1.5 2 FH 2 (mol/min) 0 10 20 30 0.01 0.02 0.03 FCH 4 ,PROX (mol/min) time (min) 0 10 20 30 0 0.5 1 time (min) 0 10 20 30 2 4 6 FT,PROX (mol/min) FCH 4 ,in (mol/min)
Fig. 12. Closed-loop responses for±20% H2 production rate changes using CS2 without CO composition control (solid line for +20% increase and dashed line for−20% change).
References
[1]Ahmed S, Krumpelt M. Hydrogen from hydrocarbon fuels for fuel cells. Int J Hydrogen Energy 2001;26:291–301.
[2]Fuel cell handbook, EG&G Services Parsons, Inc. Science Applications International Corporation; 2000.
[3]Larminie J, Dicks AL. Fuel cell systems explained. New York: Wiley; 2000.
[4]Hirschenhofer JH, Stauffer DB, Engleman RR, Klett MG. Fuel cell handbook, 4th ed., DOE/FETC-99/1076, Morgantown, WV: US Department of Energy, Federal Energy Technology Center, 1998.
[5]Song C. Fuel processing for low-temperature and high-temperature fuel cells. challenges, and opportunities for sustainable development in the 21st century. Catal Today 2002;77:17–49.
[6]Farrauto RJ, Heck RM. Environmental catalysis into the 21st century. Catal Today 2000;55:179–87.
[7]de Smet CRH, de Croon MHJ, Berger MRJ, Marin GB, Schouten JC. Design of adiabatic fixed-bed reactors for the partial oxidation of methane to synthesis gas. Application to production of methanol and hydrogen-for-fuel-cells. Chem Eng Sci 2001;56:4849–61.
[8]Doss ED, Kumar R, Ahluwalia RK, Krumpelt M. Fuel processors for automotive fuel cell systems: a parametric analysis. J Power Sources 2001;102:1–15.
[9]Choi Y, Stenger HG. Kinetics, simulation and insights for CO selective oxidation in fuel cell applications. J Power Sources 2004;129:246–54.
[10]Lattner JR, Harold MP. Comparison of conventional and membrane reactor fuel processors for hydrocarbon-based PEM fuel cell systems. Int J Hydrogen Energy 2004;29:393–417.
[11]Beckhaus P, Heinzel A, Mathiak J, Roes J. Dynamics of H2 production by steam reforming. J Power Sources 2004;127:294–9.
[12]Sommer M, Lamm A, Docter A, Agar D. Modelling and dynamic simulation of a fuel cell system with an autothermal gasoline reformer. J Power Sources 2004;127:313–8.
[13]Springmann S, Bohnet M, Docter A, Lammd A, Eigenberger G. Cold start simulations of a gasoline based fuel processor for mobile fuel cell applications. J Power Sources 2004;128: 13–24.
[14]Lin ST, Chen YH, Yu CC, Liu YC, Lee CH. Modeling of an experimental fuel processor. J Power Sources, in press.
[15]Reyes F, Luyben WL. Steady-state and dynamic effects of design alternatives in heat-exchanger/furnace/reactor processes. Ind Eng Chem Res 2000;39:3335–46.
[16]Luyben WL, Tyreus BD, Luyben ML. Plantwide process control. New York: McGraw-Hill; 1999.
[17]Chen YH, Yu CC. Design and control of heat-integrated reactors. Ind Eng Chem Res 2003;42:2791–808.
[18]Cheng YC, Wu KL, Yu CC. Arrangement of throughput manipulator and inventory control in plantwide control. J Chin Inst Chem Eng 2002;33:283.
[19]Cheng YC, Yu CC. Optimal region for design and control of ternary systems. AIChE J 2003;49:682–705.
[20]Yu CC. Autotuning of PID controllers. London: Springer; 1999.
[21]Tyreus BD, Luyben WL. Tuning PI controllers for integrator-dead time processes. Ind Eng Chem Res 1992;35:3480.
