Process Optimization for Hydrogen Production using

0 downloads 0 Views 518KB Size Report
Keywords: hydrogen, methane, methanol, ethanol, optimization, control ...... Ethanol steam reforming for the production of synthesis gas has been studied ...... quadratic Gaussian and loop transfer recovery method for a fuel processing system.
CONTROL ENGINEERING LABORATORY

Process Optimization for Hydrogen Production using Methane, Methanol or Ethanol Visa Virtanen and Kauko Leiviskä

December 2009

University of Oulu Control Engineering Laboratory

Process Optimization for Hydrogen Production using Methane, Methanol or Ethanol Visa Virtanen and Kauko Leiviskä University of Oulu, Control Engineering Laboratory

Abstract: Latest developments in the field of hydrogen production for fuel cells from ethanol, methanol, and methane are reviewed in this report. Both developments on the use of optimization in sub-processes of hydrogen production including reforming, gas cleaning, fuel cell control, and catalyst development, as well as optimization of whole production chains are examined. The scope of this review is limited on hydrogen production from ethanol, methanol, and methane. Steam reforming is by far the most common and widely used method for hydrogen production. Other methods of hydrogen production are not examined in this review. Catalyst development has been included in this review only concerning the new methodologies that have been presented. Any experimental results have not been included in this review. Research examined in this study is focused on the most recent research of the last few years. This review examines the latest research from the point of view of process optimization. The main emphasis is on intelligent optimization taking advantage of neural networks, genetic algorithms, particle swarm optimizers, support vector machines, etc. Attention has been paid on new methodology in modelling and control which can result in optimization of the hydrogen production process for fuel cells. Keywords: hydrogen, methane, methanol, ethanol, optimization, control

University of Oulu Control Engineering Laboratory P.O. Box 4300 FIN-90014 University of Oulu

Abbreviations

artificial neural networks genetic algorithms holographic research strategy multi-layer perceptron preferential oxidation proton exchange membrane solid oxide fuel cell support vector machines water gas shift

ANN GA HRS MLP PROX PEM SOFC SVM WGS

Table of contents

1 2

INTRODUCTION ................................................................................................... 1 ETHANOL, METHANOL, AND METHANE HYDROGEN PRODUCTION ........ 2 2.1 Hydrogen production from ethanol .................................................................... 2 2.1.1 Manufacturing process ............................................................................... 2 2.1.2 Main reactions ............................................................................................ 2 2.2 Hydrogen production from methanol ................................................................. 3 2.2.1 Manufacturing process ............................................................................... 3 2.2.2 Main reactions ............................................................................................ 4 2.3 Hydrogen production from methane .................................................................. 4 2.3.1 Manufacturing process ............................................................................... 4 2.3.2 Main reactions ............................................................................................ 4 3 REFORMER OPTIMIZATION ............................................................................... 6 3.1 Conventional reactors ........................................................................................ 6 3.1.1 Reformer design ......................................................................................... 7 3.1.2 Prediction of concentration and temperature profiles .................................. 7 3.1.3 Formation of hot spots................................................................................ 8 3.1.4 Kinetic modelling ....................................................................................... 8 3.1.5 Optimization of operation conditions .......................................................... 9 3.2 Membrane reactors ............................................................................................ 9 3.2.1 Comparisons against traditional reformer ................................................. 10 3.2.2 Other studies ............................................................................................ 11 3.3 Other reactors .................................................................................................. 12 3.3.1 Microreactors ........................................................................................... 12 3.3.2 Fluidized bed reactors .............................................................................. 12 4 GAS CLEANING OPTIMIZATION ..................................................................... 14 4.1 Water gas shift ................................................................................................ 14 4.1.1 Rector volume minimization .................................................................... 15 4.1.2 Behaviour prediction ................................................................................ 15 4.1.3 Kinetic studies .......................................................................................... 16 4.1.4 Other studies ............................................................................................ 16 4.2 Preferential oxidation ...................................................................................... 17 4.2.1 Reactor design optimization ..................................................................... 17 4.2.2 Other studies ............................................................................................ 18 5 FUEL CELL CONTROL OPTIMIZATION .......................................................... 19 5.1 Hydrogen proton exchange membrane fuel cells .............................................. 19 5.1.1 Neural network modelling ........................................................................ 20 5.1.2 Mechanistic modeling .............................................................................. 21 5.1.3 Support vector machines modeling ........................................................... 22 5.1.4 Parameter identification............................................................................ 22 5.1.5 Design ...................................................................................................... 23 5.1.6 Control ..................................................................................................... 23 5.2 Solid oxide fuel cells ....................................................................................... 24 5.2.1 Control ..................................................................................................... 25

5.2.2 Mechanistic modeling .............................................................................. 25 5.3 Molten carbonate fuel cells .............................................................................. 26 6 CATALYST OPTIMIZATION ............................................................................. 27 6.1 Combinatorial and high-throughput experimentation methods ......................... 27 6.1.1 Selective combinatorial catalysis .............................................................. 27 6.1.2 Different pre-treatment procedures ........................................................... 28 6.2 Neural network methods .................................................................................. 29 7 OVERALL PROCESS OPTIMIZATION .............................................................. 30 7.1 Modelling ........................................................................................................ 30 7.2 Control ............................................................................................................ 31 7.3 Other studies ................................................................................................... 32 8 CONCLUSIONS ................................................................................................... 34 REFERENCES.............................................................................................................. 35

1 INTRODUCTION Hydrogen economy offers significant advantages in activities towards the sustainable energy economy. Fuel cell technology is the main stream of development in energy production. Hydrogen fuel for fuel cells can be produced with chemical, biological or electrolysis processes. Chemical processes utilise hydrocarbons, water and biomass as starting materials. Nowadays, steam reforming of hydrocarbons, i.e. natural gas, methanol, ethanol, etc., is the most common and generally the most economical method for hydrogen production. The requirement for producing renewable hydrogen cleanly and safely has increased the interest towards using ethanol. Ethanol can also be produced by fermentation of biomass from e.g. energy plants, agroindustrial wastes, forestry residues, and organics in municipal solid waste. In reactor technology, the interest goes to membrane reactors combining separation and catalytic reactions in one system. Catalytic membranes have attracted increased attention because of their compact configuration. Combining hydrogen selective membrane with the reforming equipment makes continuous production and separation of hydrogen possible. On catalysis side, photocatalysis on semiconductor nanoparticles can be carried out at low temperatures. In microreactor technology, miniatyrization, micro channels and microstructures have been the key interest areas. Microreactors offer optimal contact time, excellent heat transfer, and effective mass transfer. Mathematical modelling gives several advantages to technology and process development. It gives possibilities to simulate process behaviour long time before the process has even been built. It offers multiple ways to process phenomenological development and optimization. Concerning process chains, modelling is a good starting point for optimization of production routes and operation conditions. This report is a part of results from the research project “Hydrogen production from biobased raw materials by reforming and photocatalysis (REFORMH2-2)”. The project is based on the COST Action 543 –project ‘Research and Development of Bioethanol Processing for Fuel Cells, BIOETHANOL’ (2006-2010), which studies the production of hydrogen by reforming bioethanol in microreactors using catalytic membranes as catalysts. It is also a continuation for a Tekes funded project “Polttokennoon soveltuvan vedyn tuottaminen bioetanolia reformoimalla, REFORMH2” This report collects some newer literature on the use of optimization in sub-processes of hydrogen production and the whole production chains. The main emphasis is on intelligent optimization taking advantage of artificial neural networks, genetic algorithms, particle swarm optimizers, support vector machines, etc. The report is supporting the further studies on actual optimization and selection of production chains.

2 ETHANOL, METHANOL, AND METHANE HYDROGEN PRODUCTION Hydrogen can be produced for fuel cells by various methods and from various fuels. Steam reforming of natural gas, methanol, and ethanol is however the most common and generally the most economical method for hydrogen production. Production processes of hydrogen from these three fuels are similar with some differences in the type of used reactors. Although there are also some variations in the number and type of used reactors within a used fuel, the typical production chains can be distinguished for each fuel, and are described in the following subsections along with the main reactions occurring in the process.

2.1 Hydrogen production from ethanol Hydrogen production from ethanol has several advantages. Ethanol is less toxic than methanol. It can also be more easily stored and safely handled. It can be produced in large amounts from biomass. It is free from sulphur and metals. It is also CO2 neutral since the CO2 produced by the process is consumed by the biomass growth [Vaidya and Rodriques 2006]. 2.1.1 Manufacturing process In conventional steam reforming of ethanol, the steam reformer is followed by one or two water gas shift (WGS) reactors and preferential oxidation (PROX) units. It is presented in Figure 1. [Biset et al. 2009]:

Figure 1. Ethanol steam reforming process for fuel cells [Biset et al. 2009]. 2.1.2 Main reactions Following reactions occur at the steam reforming process of ethanol [Batista et al. 2006][Ni et al. 2007]: The overall reaction of steam reforming is: C2H5OH + 3H2O → 2CO2 + 6H2

(1)

WGS reaction is: CO + H2O → CO2 + H2

(2)

PROX reaction is: CO + ½O2 → CO2

(3)

2.2 Hydrogen production from methanol Methanol is unique and advantageous substance for hydrogen production in many ways. It has an H/C ratio of 4/1, it is liquid in atmospheric pressure and common environmental temperature and it has a low boiling point of 65º C. In terms of environmental impact, it is readily metabolized by ambient orgasms and it is miscible with water. Methanol can be converted to hydrogen at a temperature of 150-350 ºC which is lower than most other fuels. This is because it contains no carbon-carbon bonds that must be broken and unlike methane it is easily activated at low temperatures. Low-temperature conversion leads to low levels of CO formation even if the catalyst has no special mechanism for selectivity of CO2 over CO [Palo et al. 2007]. Majority of interest in hydrogen production from methanol has focused on steam reforming. This is because of the high hydrogen yield and low CO production. The advantage that steam reforming has over the methanol conversion with low CO production does not translate to other hydrogen fuels because fuels with C-C bonds require different conversion mechanisms [Palo et al. 2007]. 2.2.1 Manufacturing process Conventional steam reforming of methanol comprises of two main elements: Steam reforming and PROX units. Since methanol is an easily converted fuel it can be operated at low temperatures in the reformer. This results in that the secondary conversion in WGS operation is unnecessary [Palo et al. 2007]. The methanol reforming process is presented with a Figure 2. [Lattner and Harold 2005]:

Figure 2. Methanol steam reforming process for fuel cells [Lattner and Harold 2005].

2.2.2 Main reactions Following reactions occur at the steam reforming process of methanol [Lattner and Harold 2005]: The steam reforming reaction is: CH3OH + H2O ↔ CO2 + 3H2

(4)

Methanol decomposition reaction in steam reformer is: CH3OH ↔ CO + 2H2

(5)

PROX reactions are: CO + H2 + O2 → CO2 + H2O CH3OH + O2 → CO2 + H2O + H2

(6) (7)

2.3 Hydrogen production from methane The dominant industrial process to produce hydrogen is the steam reforming of methane. It has been in use for several decades as an effective mean for production of hydrogen and produces nearly all the hydrogen in chemical industry and the supplemental hydrogen in refineries [Barelli et al.2008]. 2.3.1 Manufacturing process Methane steam reforming is a catalytic process that involves a reaction between natural gas or other light hydrocarbons and steam. When the resulting hydrogen is wanted to be particularly pure from carbon monoxide, like generally the case with fuel cells, the WGS reactors can be supported by the PROX reactor. The conventional methane steam reforming process with the potential PROX phase shown with dashed line is presented in Figure 3. [Barelli et al.2008]:

Figure 3 Methane steam reforming process for fuel cells [Barelli et al.2008]. 2.3.2 Main reactions Following reactions occur at the steam reforming process of methane [Barelli et al.2008]: The steam reforming reactions are:

CH4 + H2O ↔ CO + 3H2 CH4 + 2H2O ↔ CO2 + 4H2

(8) (9)

WGS reaction is: CO + H2O ↔ CO2 + H2

(10)

3 REFORMER OPTIMIZATION Hydrogen can be manufactured by a variety of methods. There are thermal, electrochemical and biochemical manufacturing methods. The most common manufacturing method of hydrogen is steam reforming. Arguments for steam reforming for hydrogen production on fuel cells over other methods include thorough research and practical experience over several decades, together with capability of production in smallsized units or in reformers inside fuel cell units. When producing hydrogen by steam reforming the products of the reformer may be used as the fuel to the fuel cells, thus avoiding the troubles related to hydrogen storage and transportation [Ohenoja and Leiviskä 2008]. Reformer is the first unit in the hydrogen production chain for fuel cells. It transforms majority of the fuel to hydrogen making its optimization profoundly significant for the overall efficiency of the fuel cell system. Several issues still complicate the optimal operation of reformers. Problems with non-optimal hydrogen yield, formation of carbon monoxide, catalyst deactivation, coke formation and occurrence of disturbances or hot spots in hydrogen production are still a cause and focus of active research. Reformer reactors can be divided into several categories by their operating principle. In this review a categorization is made by dividing the studied reformers into three broad categories: Conventional reactors, membrane reactors and other reactors. The following subsections describe the research done over the last few years with those three reactor types.

3.1 Conventional reactors Conventional reformers consist of several tubular pipes which have been filled with catalyst pellets. Because the reforming requires high temperatures, the tubes are positioned inside a furnace in which they are heated directly by heat radiation of the furnace flame [Ohenoja and Leiviskä 2008]. A depiction of the basic operation principle of a conventional reformer is presented in figure 4.

Figure 4 Conventional reformer.

3.1.1 Reformer design Several researchers have developed models to aid in design of reformers. [De Jong et al. 2009] developed a model for the conversion process of methane for a single reactor tube including both chemical reaction models and heat transfer models. The model yields data of temperature heat transfer and concentrations of hydrogen, carbon monoxide and natural gas along a reactor tube. Focus was on modelling of limiting effects of, and interaction between heat transfer and chemistry of methane rather than on maximum performance using a membrane. Finally, the model was used to evaluate the performance of the reformer for six design modifications: Tube length, air fraction, hole geometry, hole location, steam fraction and shield thickness. [Hoang and Chan 2004] report on mathematical modelling of catalytic autothermal reforming of methane for hydrogen production. A two-dimensional unsteady reformer model was developed to simulate the conversion behaviour of the reformer. The model can be used to determine the optimum feed conditions and to guide the design of an autothermal reformer. The model covered all aspects of major chemical kinetics and heat and mass transfer phenomena in the reformer. A bioethanol processing system to feed a 200kW solid oxide fuel cell (SOFC) has been simulated and evaluated by [Arteaga et al. 2008]. The general scheme of process was composed of vaporization, heating and bioethanol steam reforming in a fixed bed reactor. The performance of the pseudo-homogenous model of the reactor consisting of the catalytic ethanol steam reforming was developed based on the principles of classical kinetics and thermodynamics through a complex reaction scheme and a LangmuirHinshelwood kinetic pattern. The model has been employed to evaluate effects of several design and operation parameters on the process. In one study, an in-house experimental system was fabricated by [Wang 2008] for autothermal methane reforming to produce hydrogen-rich gas. The temperature profile along the axis of the reformer was measured, and the location of peak temperature from the inlet of the reformer of the methane autothermal reforming process is determined. These results are important for optimization of the design and operation of the autothermal reformer. 3.1.2 Prediction of concentration and temperature profiles Prediction of concentration and temperature profiles has also been investigated by several research groups. Mechanistic kinetic models were formulated by [Aboudheir et al. 2006] for catalytic reforming of concentrated crude ethanol in a packed bed tubular reactor based on coupling of mass and energy balances and finite element method. The developed model is capable to accurately predict concentration profiles of all the chemical species and temperature profiles of the fluid in both axial and radial directions. Another mechanistic kinetic models were formulated by [Akpan et al. 2007] based on Langmuir-Hinshelwood-Hougen-Watson and Eley-Rideal approaches to describe the

kinetics of hydrogen production by the catalytic reforming of concentrated crude ethanol in a packed bed tubular reactor. The mathematical model is able to predict the concentration profiles of the species in both of the radial and axial directions with high accuracy. [Lattner and Harold 2007] report on a bench-scale fixed-bed reactor under near-adiabatic conditions that was constructed for experimental demonstration with axial distribution of air through multiple porous ceramic membranes to limit the peak temperature within the catalyst bed. The experimental design attempted to approach adiabatic operation to provide realistic temperature profiles for full scale reactors. Provisions for measurement of axial temperature and composition profiles were designed into the reactor. Experiments were performed under varied conditions of feed temperature, pressure, steam/carbon ratio and with two different air distributor lengths. The effect of space velocity was studied implicitly via the detailed axial profile data. 3.1.3 Formation of hot spots Performance analysis for the autothermal reforming process of methane in a fixed bed reformer has been presented by [Halabi et al. 2008]. The model accounted for mass and thermal dispersion in the axial direction, axial pressure distribution and interfacial and inter-particle transport. The process performance under dynamic and steady state conditions was analyzed with respect to key operational parameters: Temperatures of gas feed and catalyst bed, oxygen/carbon ratios, gas feed space velocity and feed contaminations. Furthermore the formation of hot spots in the fixed bed reformer as a function of oxygen partial pressure was predicted. A quasi-homogenous one-dimensional model was developed for modeling a lab-scale fixed-bed autothermal reactor by [Barrio et al.2007]. A special attention was paid to the hot spot formation due to exothermal oxidation reactions that can lead to severe catalyst deactivation, and the effects of varying molar feed ratios in the product distribution achieved. Furthermore an oxygen and water feed splitting study was performed. 3.1.4 Kinetic modelling Kinetics is an essential part in modelling of a reformer. Elements of kinetic modelling are present in a number of studies presented during the last years. Papers focusing particularly on kinetic modelling are less common. One such paper dealing with kinetic modelling has been given by [Akande et al. 2006] for catalytic reforming of crude ethanol in a packed bed tubular reactor over a 15%-Ni/Al2 O3 catalyst prepared by the coprecipitation technique. Eley-Rideal assumptions where the surface reaction involved an adsorbed species and a free gaseous species were used to develop the reaction mechanism and four models were proposed based on this mechanism, from which a new kinetic model based on the dissociation of adsorbed crude ethanol as the rate determining step was developed for this novel catalytic process.

Kinetics study of steam reforming of ethanol was done by [Sahoo et al. 2007] using a Co/Al2O3 catalyst to investigate the effect of reaction temperature, contact-time and steam to ethanol molar ratio on hydrogen production. The main reactions that occurred with this catalyst were steam reforming, WGS and ethanol decomposition reactions. The author viewed that it is important to consider all of these three reactions simultaneously while developing the comprehensive model for steam reforming of ethanol process. Mechanistic kinetic model using Langmuir-Hinshelwood approach was developed considering surface reaction mechanisms of steam reforming, water gas shift and ethanol decomposition reactions in fixed-bed tubular reactor. 3.1.5 Optimization of operation conditions An industrial steam reformer in a hydrogen plant was simulated by [Nandasana et al. 2003] under dynamic conditions. Dynamic model incorporated aspects of heat transfer in the furnace and diffusion inside the catalyst pellet and it was used to study effects of disturbances that reduce production of hydrogen in the reformer. The operation of the steam reformer was simulated in the presence of three idealized disturbances in the inlet feed temperature, the inlet feed natural gas and the furnace temperature. Two objective functions were minimized simultaneously: the cumulative deviation of the flow rate of hydrogen and the cumulative deviation of steam flow. An elitist non-dominated sorting genetic algorithm (GA) was used to solve the optimization problems. Catalytic autothermal reforming of methane in a tubular quartz reactor was studied by [Simeone et al. 2008] over a commercial nickel catalyst as a function of feed flow rate, feed composition and oven temperature. The effect of water addition on product composition and catalyst temperature profile was investigated. Temperature distribution in the catalyst was measured and for each operating condition, the faction of catalyst above a limiting value for deactivation and below a limiting value for coke formation was quantified. This information was coupled with data of methane conversion and hydrogen production, to identify a region of favourable reactor operating conditions in the plane AIR/CH4-H2O/CH4.

3.2 Membrane reactors Use of membrane reactors instead of conventional ones has drawn increased attention and research in recent years. In membrane reactors, the fuel conversion can be increased at lower temperatures. A membrane reactor is a device in which a reaction and a selective separation take place simultaneously. Continuous removal of one of the products permits obtaining reaction conversion beyond the thermodynamic equilibrium that is the upper limit considered for conventional reactor. A depiction of the basic operation principle of a membrane reformer is presented in figure 5.

Figure 5 Schematic of a double-jacketed membrane reactor. In the figure 5. it is shown how reactants flow into annular side packed with catalysts and hydrogen permeates across the membrane. The un-reacted effluents flow into the oxidiced zone, mixed with air, where heat supplies the steam reforming reaction. 3.2.1 Comparisons against traditional reformer Several comparisons have been made for membrane reactors against traditional reactors. [Lattner and Harold 2005] simulated methanol steam reforming system using an adiabatic one-dimensional rector model comprising kinetic rate expressions of [Peppley et al. 1999]. Steam reforming, autothermal reforming and autothermal reforming membrane reactor based fuel processors were compared for hydrogen production for proton exchange membrane (PEM) fuel cells. Overall system efficiencies and reactor volumes were compared as a function of fuel processor design. Variables in the design and operation of the fuel cell system included oxygen/carbon ratio, steam/carbon ratio and reformer feed temperature. After optimization of the reformer operating conditions, reaction kinetics was applied consistently to each case to estimate and compare reactor volumes. Ethanol steam reforming for the production of synthesis gas has been studied theoretically by [Gallucci et al. 2008b]. A mathematical model was formulated for a traditional reactor packed with a Co-based catalyst and then applied to a in a dense Pd-Ag membrane reactor in which the hydrogen production is increased by removing the hydrogen produced from the reaction mixture through a highly selective palladium-based membrane. The model was developed using all the thermodynamic values for the species involved in the reaction and the kinetic expressions. Hydrogen removal through a selfsupported membrane having a 23% Ag content was calculated using experimental permeabilities. The performances in terms of ethanol conversion and hydrogen production of both the traditional system and the membrane reactor were compared taking into account the effect of the operating conditions of temperature, pressure, sweep gas mode and the water/ethanol feed molar ratio.

[Fu and Wu 2007] modelled methanol steam reforming by one-dimensional nonisothermal model for double jacketed Pd membrane reactor. Both mass and heat transfer behaviours were evaluated simultaneously in three parts of the reactor: annular side, permeation tube and the oxidation side. The performance of a double-jacketed membrane reactor was compared with an autothermal reactor by judging against methanol conversion, hydrogen recovery yield and production rate. 3.2.2 Other studies [Gallucci et al. 2008a] simulated a theoretical model for ethanol steam reforming in a dense Pd-Ag membrane reactor operated in both co-current and counter-current modes. The model was developed using all the thermodynamic values for the reaction species and the kinetic expressions. Hydrogen removal was calculated using experimental permeabilities. The set of differential equations was solved using the IV Runge-Kutta method and the counter-current mode was solved using the shooting method. The performances in terms of ethanol conversion and hydrogen recovery of both co-current and counter-current configurations were compared taking into account the effect of the various operating conditions of temperature, pressure, sweep gas mode and water/ethanol feed molar ratio. [Cheng et al. 2008] presented multi-objective optimization results for two important catalytic membrane reactors, methanol synthesis and hydrogen generation. For given reactor size, multiple optimal solutions with trade-off relations for triple objective functions, including major feed rate, major product rate ad exergy loss were obtained. The correlations between objective functions and optimal operating variables were analyzed. The effects of membrane thicknesses and areas on the optimal solutions were explored. One-dimensional mathematical models with rigorous consideration on the mass diffusion and reaction kinetics inside catalyst particles were used for the membrane reactors. Elitist nondominated sorting GA was employed for searching for the multiobjective optimal solutions. The ethanol steam reforming reaction carried out in a Pd-based tubular membrane reactor was modelled by [Tosti et al. 2009] via finite element code. The model considered the membrane tube divided into finite volume elements where the mass balances for both lumen and shell sides were carried out accordingly to the reaction and permeation kinetics. According to the author, membrane rectors have so far been modelled by using the catalyst kinetics found for traditional reactors. Differently, in this study the reaction kinetics were determined from experiments carried out with a membrane reformer. Based on the Damkohler-Peclet analysis, the optimization of the membrane reformer was approached.

3.3 Other reactors 3.3.1 Microreactors Microreaction technology is a relatively new concept which offers the possibility of miniaturization of conventional reactors while providing the same throughput. One of the first micro-structured devices was reported in the literature in 1989. With the tools of micro-fabrication, several novel reactor configurations can be fabricated allowing different design concepts not feasible with conventional packed-bed technology. Such microchannel reactors or microreactors typically carry small channels with dimensions in the sub-millimeter range. This also results in a relatively large surface area-to-volume ratio and increased driving forces for heat and mass transport. This translates into short response times. The pressure drop in microchannels is substantially lower compared to packed-bed reactors. The unique flow and heat/mass transfer properties have led to process improvements. These benefits however need to be significant to counter-balance the higher production cost of microreactors compared to fixed-bed reactors [Nikolaidis et al. 2009]. In the field of integrated fuel processing, particularly microchannel technologies may make the whole system integrated, more efficient and compact and thus have the potential for making the integrated fuel processor into practice [Qi et al. 2007]. Modelling of flow of reacting mixture of methanol and steam in a 2D microslot was studied numerically by [Kuznetsov and Kozlov 2008] at activation of the reactions on the channel wall. The modelling was carried out in the framework of Navier-Stokes equations for a laminar flow of multi-component compressible gas. The work was aimed at numerical investigation of correlations between thermal, diffusion, and physicalchemical processes under the conditions of significantly non-isothermal reactions and external heat supply distributed along the channel. [Stutz et al. 2006] investigated the effects of catalyst loading and reactor geometry defined by gas space velocity and catalyst space velocity on the performance of a wallcoated, single channel methane microreactor. Such a reactor, consisting of a tubular flow channel and a thermal conductive channel wall, is a good representation of microfabricated channels and monoliths. The channel wall heat transfer was included in the axisymmetric numerical model of an autothermal tubular reformer, described by an explicit model of the flow channel and a detailed elementary surface reaction mechanism consisting of 38 steps. 3.3.2 Fluidized bed reactors An important decision in the design of fluidized bed reactors is which of several flow regimes to choose. Almost all models for fluidized bed reactors are restricted to a single flow regime, making comparison difficult, especially near the regime boundaries. [Mahecha-Botero et al. 2009] examined the performance of fluidized bed methane reformers with three models – a simple equilibrium model and two kinetic distributed models, based on different assumptions of varying sophistication. Membranes were incorporated to improve reactor performance. Eighteen cases were simulated for different

flow regimes and membrane configurations. Relative merits of different fluidization hydrodynamic regimes – bubbling, turbulent and fast fluidization were investigated. In addition, the influence of operating variables, such as reactor pressure and temperature, and key equipment variables, such as the membrane thickness of palladium alloy coating and membrane surface area per unit volume of reactor were studied.

4 GAS CLEANING OPTIMIZATION A significant amount carbon monoxide exists in the resulting product gas mixture from the reformer. Since fuel cells demand almost carbon monoxide-free hydrogen for their operation, the carbon monoxide needs to be removed from the product gas. Currently the number, size and cost of these carbon monoxide removal reactors tend to be a bottleneck for many practical applications for fuel cells. Therefore research on more efficient processes is still required to minimize the number and size of the gas cleaning reactors. Main processes in this gas cleaning phase are WGS process and PROX process. Following subsections review these processes. For further reading on gas cleaning operations, a review on catalyst development on high temperature WGS, low temperature WGS and removal of carbon monoxide has been given by [Trimm 2005]. Another review has been given by [Kolb et al. 2007] on selective oxidation for micro-structured catalytic reactors. The review is split into two parts. The first part concerns catalytic gas-phase oxidation reactions in microreactors. In the second part, selective oxidation is described as one gas purification step in the framework of fuel processing for fuel cells. It is also called preferential oxidation in this context.

4.1 Water gas shift In a typical operation, the resulting gas from the reformer has a temperature of about 350450 ºC when it is taken through a high temperature water-gas-shift reactor. The carbon monoxide concentration will be reduced at high temperature water-gas-shift reactor to about 3-4 %. The temperature in low temperature water-gas-shift reactor is about 200300ºC and results in carbon monoxide concentrations of about 0.5-1 % [Trimm 2005]. Despite the fact that WGS reaction has been recognized for over a century, there is still discussion about the reaction mechanism. Infra red spectroscopy and temperature programmed desorption studies have suggested that the reaction proceeds via the formation of an adsorbed formate group which decomposes to hydrogen and carbon dioxide as follows [Trimm 2005]: H2O + 2 * → HO * + H * CO + * → CO* CO * + HO * → * -O - CH = O * - O – CH = O → CO2 + H* 2H * → H2 + *

(11) (12) (13) (14) (15)

Several other groups have suggested that a redox mechanism is predominant [Trimm 2005]: CO + * → CO* H2O + MO → H2 + MO2 MO2 + CO → CO2 + MO + *

(16) (17) (18)

Recent density functional theory calculations have suggested that the process involves interaction of two *OH groups [Trimm 2005]: H2O + 2 * → * OH + *H 2 * OH → H2O + O * + * O * + MO → MO2

(19) (20) (21)

In most hydrocarbon processors, the WGS reactor is the biggest and heaviest component because the reaction is relatively slow compared to the other reactions and is inhibited at higher temperatures by thermodynamics. Therefore reducing the size of the WGS reactor is an important issue [Choi and Stenger 2003]. 4.1.1 Rector volume minimization A one-dimensional heterogeneous model was applied to a simulation for a fixed bed WGS reactor by [Giunta et al. 2006]. The catalyst deactivation due to thermal factors (sintering) was taken into account in the model. Simple first order differential equations were solved by the fourth order Runge-Kutta method and non-linear boundary value problems were solved with inverse shooting method. A parametric sensitivity analysis was carried out for some of the process variables with the purpose of finding criteria to minimize the reactor volume. [Francesconi et al. 2007a] investigated WGS reactor as a process component of a fuel processor for applications in PEM- type fuel cells, and showed how the reactor design using mathematical programming techniques allows computation of both reduced volumes and optimal operation conditions. Different reactor configurations, reactor size and relative sizes of reactor components were evaluated and process performance bottlenecks and opportunities for optimization identified and analyzed. 4.1.2 Behaviour prediction Extensive experiments with WGS reaction were conducted by [Ding and Chan 2008] under different operating conditions to understand the WGS reaction better. WGS kinetic rates were extracted from the experimental results and applied in a two-dimensional unsteady kinetic model for performance and behaviour prediction. The diffusion of gas species in the catalyst, non-plug flow gas velocity profile in the reactor and non-uniform catalyst porosity were also included in the model. Parametric studies on WGS reaction were also covered. By converting all the results of the parametric studies with respect to space velocity, the optimum range of WGS performance was determined. A dynamic, heterogeneous, two-dimensional model for packed-bed WGS reactors was presented by [Adams II and Barton 2009] that can be applied to both high and low temperature shifts and at scales ranging from industrial to small applications. The model used a plug-flow reactor configuration, which is currently preferred for industrial coal gasification applications. It is also suitable for any catalyst for which kinetic data is

available and it can be applied for any number of chemical species. The model was used to examine dynamic behaviour of the WGS units. 4.1.3 Kinetic studies [Choi and Stenger 2003] studied WGS reaction with the goal of obtaining highly accurate kinetics expressions and to create a tool for an integrated and optimized simulation of a whole fuel processing system. Kinetics of the WGS reaction was studied to evaluate existing reaction mechanisms, test various rate expressions and simulate the performance in a methanol fuel processor for fuel cell applications. The parameters of five rate expressions were fitted to experimental data using non-linear least squares optimization. Numerical integration of a one-dimensional plug flow reactor model was used for parameter fitting. Kinetics of the WGS reaction over a commercial copper/zinc oxide/alumina catalyst was studied by [Ayastuy et al.2005]. 16 redox and associative models were proposed to obtain rate equations to which the obtained experimental data were fitted. Both the differential reactor approach and integral reactor approach were used in obtaining the experimental data. The discrimination between models was based on F-test. 4.1.4 Other studies [W.-H. Chen 2008] developed chemical kinetic modelling and comparison of low and high temperature shift reactions. Detailed reaction phenomena of both of those reactions were investigated to aid in explaining experimental results and to provide observations on the reactions in catalyst bed. In the developed models, the effects of both the reaction temperature and the steam/CO-ratio were taken into account. Development of integrated reaction and heat exchange approach to microreactor design that enhance reaction yields by allowing the reactant stream to follow optimal reactant temperature profiles was studied by [Kim et al. 2005]. Both one-dimensional and twodimensional models for the integrated WGS reaction and heat exchange design were presented. The models were then applied to a parametric study investigating sensitivities of design parameters for both the parallel and countercurrent flow configurations of the integrated reaction and heat exchange design. A dynamic model of a catalytic reactor converting carbon oxide into hydrogen by WGS was presented by [Bittani et al. 2008]. The model was based on the description of the kinetic-chemical mechanisms of adsorption and desorption in the porous catalyst and on mass, energy and momentum conservation equations. This dynamical model was developed in order to study different control strategies for the gas shift reactor. Based on water catalytic gas shift mechanism on precious metal catalyst a LangmuirHinshelwood kinetics model and a power law kinetics model were derived by [Sun et al. 2005] for the operating conditions of syngas from natural gas reforming at near-ambient pressure. The two kinetic models were integrated in a dynamic distributed reactor model

for design of full-scale WGS reactors for a natural gas fuel processing system. In the work, a procedure for parameter estimation for the Langmuir-Hinshelwood rate equations for WGS was demonstrated and optimal operating conditions, namely the reactor inlet temperatures of high temperature shift and low temperature shift were determined.

4.2 Preferential oxidation Among different hydrogen purification systems, the CO preferential oxidation is preferred for small-scale systems because of its relative simple implementation, lower operating costs and minimal hydrogen loss [Oliva et al. 2008]. Main reactions involved in PROX are: CO + ½ O2 → CO2 H2 + ½ O2 → H2O

(22) (23)

Some researchers point out that depending on the operation conditions and the catalyst used, the inverse WGS reactions and methanation reaction also occur, but to a lower extent: CO + H2O ↔ CO2 + H2 CO + 3H2 ↔ CH4 + H2O

(24) (25)

Selection of a highly selective catalyst, reactor design and determination of operation conditions are critical aspects that have to be addressed in PROX design process. Once the catalyst to be used is defined, temperature, O2/CO-ratio, reactor operation mode and number of stages constitute key decisions, leading to opportunities for process optimization [Oliva et al. 2008]. 4.2.1 Reactor design optimization A two-dimensional model of PROX reactor was developed by [Cipiti et al.2009] in order to investigate the effects of main geometrical parameters on CO preferential oxidation performance. The reactor configuration was a single stage multi-tube geometry with a concurrent air flow in the cooling jacket. The study considered heat and mass transport phenomena associated with two main simultaneous chemical kinetic reactions in the reactor: CO oxidation and H2 oxidation. The model optimized key geometrical parameters by a parametric analysis for different tube pitch, tube diameter, number of tubes and inlet reagent temperature, and evaluates the temperature distribution and gas concentration in the reactor to gain insight into the design and operation of the PROX unit. In most numerical studies reported in the literature simulation of a PROX reactor has been done by neglecting flow details and pressure gradients along the reactor. This is not always a realistic approximation. [Vahabi and Akbari 2009] performed numerical simulation of a reacting flow containing carbon monoxide and hydrogen in a three

dimensional micro-channel by considering the Navier-Stokes equations for the fluid flow. Optimization of a microreactor process stream was done by solving conservation equations of mass, momentum, energy and species for an ideal mixture with three surface reactions. The required reactor length for the removal of CO to less than 10 ppm was found, and the effects of various parameters on the microreactor performance were investigated. Furthermore, by considering the definition of CO yield per unit perimeter, the optimum channel geometry was identified. [Oliva et al. 2008] analyzed the CO- preferential oxidation reactor design as a component of the CO clean-up system of the ethanol processor for H2 production applied to PEM fuel cells. An egg-shell catalyst type of Pt/Al2O3 was under consideration. Onedimensional heterogeneous catalytic reactor model accounting for interfacial gradients was used to optimize the PROX reactor. The optimization problem determined the optimal reactor length, reactor diameter, catalyst particle diameter, inlet reactants temperature and insulating material thickness that minimize the total system volume. The work reflected clearly the advantages of applying mathematical programming techniques to optimize both design and operation conditions of the PROX reactor. To the authors’ knowledge, model-based optimal reactor design and analysis applying mathematical programming techniques have not been yet addressed in this area. 4.2.2 Other studies [Nikolaidis et al. 2009] developed a micro-structured recycle reactor for studying the kinetics of heterogeneously catalysed gas phase reactions. The preferential CO oxidation reaction over Pt-Rh/γ-Al2O3 catalyst in reformate surrogate containing up to 60 vol% H2 was investigated over a wide range of CO concentrations between one and four times O2 excess. Initial concentration of H2O and CO2 were 2.5 vol% and up to 20 vol% respectively and their influence on the reaction rate was investigated. The gas composition was equivalent to the reformate of a hydrocarbon reformer which was purified already by WGS as first clean-up stage. The highly undesired oxidation of hydrogen was also investigated. To the authors’ knowledge, the determination of reaction kinetic for a microreactor operated as a recycle reactor is a novel approach. The results compared well with the same reaction carried out in a fixed bed reactor which is the common choice for this kind of studies. This study investigates the performance of a preferential CO oxidation reactor operating in two modes – single and two-stage. Preferential CO oxidation in excess H2 was carried out by [Srinivas and Gulari 2006] on a Pt/γ-Al2O3 catalyst in a packed-bed microreactor system consisting of two cascaded stages. Oxygen split ratio between the two stages and temperature of two reactor stages was optimized to obtain best performance. System robustness was evaluated by analyzing the system’s response to step changes in inlet conditions. The motivation behind operating the PROX reactor system in a cascaded multi-stage mode is based on the hypothesis that the number of catalytic sites that participate in CO oxidation could be increased compared to those that participate in H2 oxidation.

5 FUEL CELL CONTROL OPTIMIZATION Fuel cells are the last unit in the hydrogen production chain for fuel cells. A fuel cell is a device that can directly convert chemical energy to electric and thermal energy. Among various types of fuel cells, PEM fuel cells, which are also called polymer electrolyte membrane fuel cells, have attracted great amount of research interest in the past decade, especially in stationary and mobile power generators and electric vehicles [Ou and Achenie 2005]. There are primarily two types of PEM fuel cells, namely hydrogen PEM fuel cells and direct methanol fuel cells [Ou and Achenie 2005]. The PEM fuel cell is a nonlinear, multi-variable electro-chemical system that is hard to model. A large number of publications on fuel cell modelling [Biyikoglu 2005] target the complicated internal phenomena at the molecular level. Among them, two-dimensional and more complex three-dimensional, two-phase and non-isothermal models have been presented, which have had very complicated expressions with some key physical parameters that may even be immeasurable [Zhong et al. 2007]. Two main modeling approaches can be found in literature. The first approach includes mechanistic models, which aim at simulating the heat transfer, mass transfer and electrochemical phenomena encountered in fuel cells. The second approach includes models that are based on empirical or semi-empirical equations, which are applied to predict the effect of different input parameters on the voltage-current characteristics of the fuel cell, without examining in depth the physical and electrochemical phenomena in fuel cell operation [Mo et al. 2006]. A thorough review on direct methanol fuel cells has been given by [Biyikoglu 2005]. In the review PEM fuel cell modelling is presented, along with comparison of the modelling to experimental studies. Another review has been given on direct methanol fuel cells by [Oliveira et al. 2007]. In the review the fuel cells are divided into three types of models based on the approach taken: analytical, semi-empirical and mechanistic models.

5.1 Hydrogen proton exchange membrane fuel cells In the figure 6. a two-dimensional schematic of a fuel cell with its subdomains is presented. It shows how hydrogen and oxygen are supplied, respectively to the anode and cathode gas channels. Some of the flow in the channels permeates the porous gas diffusion layers. In the channels it is transported, primarily by diffusion, to the catalyst layers. The gases then dissolve in the electrolyte phase, which “wets” the catalyst, after which they diffuse to the sites where the following half cell reactions take place [Cheddie and Munroe 2007]: H2(aq) → 2H+(aq) + 2 e- (anode) ½ O2(aq) + 2H+(aq) + 2e- → H2O(aq) (cathode)

(26) (27)

Figure 6. Schematic of a PEM fuel cell. There are several issues related to the modelling of fuel cells. Considering the air circuit, a compressor must be placed at the inlet of the fuel cell so as to provide oxygen for the electrochemical reaction and to raise its efficiency by increasing the air pressure. Considering the hydrogen circuit, hydrogen can be stored directly on-board or supplied through a reforming system. Special care must also be taken with the water management system. The PEM must always be maintained in a well-hydrated state so as to ensure the migration of protons H+ from the anode to the cathode. Moreover, in most cases there is a special water coolant circuit within the stack. The aim is to obtain a completely selfsufficient system with respect to water [Jemei et al. 2003]. Among various kinds of fuel cells, the PEM fuel cell, with its low operating temperature, high power density, high efficiency, fast start-up, quick response and zero emission, is most suitable for vehicles and is the subject of priority research in many countries [Ye et al. 2009]. 5.1.1 Neural network modelling Availability of the electrochemical equations or models may not be sufficient to accurately design fuel cell system for optimum performance. In addition, these models may be very complicated. In most control applications, the designer may be interested in relationship between inputs and outputs. Such prediction may be performed by artificial neural networks (ANN) [Saengrung et al. 2007]. ANN have been used in numerous modeling studies. [Jemei et al. 2003] used a multilayer perceptron (MLP) neural network with back-propagation is used to model PEM fuel cell. Cathode and anode gas pressures, stack temperature and current were used as inputs

and stack voltage as the output. As the main objective in the work a model was proposed able to represent the system behaviour as perfectly as possible with the smallest number of sensors on the vehicle. An MLP neural network was also used by [Lee et al. 2004] for modelling PEM fuel cell. The focus of the study was to derive a non-parametric empirical model including process variations to estimate the performance of fuel cells without extensive calculations. In addition to current density, the model takes into account process variations such as gas pressure, temperature, humidity, and utilization to cover processes which are important factors in determining the real performance of fuel cells. A back-propagation MLP neural network and a radial basis function network hybrid models were built and compared in predicting stack voltage and current of commercial PEM system by [Saengrung et al. 2007]. Only mass air flow and stack temperature were used as input variables. In a different kind of study by [Rouss and Charon 2008] the mechanical behaviour of a PEM fuel cell system was modelled. In order to implement fuel cells in transportation systems, the mechanical behaviour of fuel cell systems, and the influences of mechanical loads on their structure needs to be mastered. ANN modelling approach for the mechanical nonlinear behaviour of a PEM fuel cell system was proposed. An experimental set was designed for this purpose: a fuel cell system in operation was subjected to random and swept-sine excitations on a vibrating platform in three axis directions. Its mechanical response was measured with three-dimensional accelerometers. The raw experimental data was used to create multi-input, multi-output model using MLP neural network combined with time regression input vector. 5.1.2 Mechanistic modeling Representing mechanistic modeling approach a two-phase model of intermediate temperature (120-200 ºC) PEM fuel cell was presented by [Cheddie and Munroe 2007]. The model takes into account two phase effects due to gas solubility in the phosphoric acid/polybenzimidazole electrolyte and considers aqueous phase electrochemical reactions. The model also accounts for all polarization and transport phenomena. The dependence of the fuel cell performance on membrane doping level, catalyst activity and transport properties of dissolved gases in the electrolyte medium were also investigated. [Hou et al. 2007] analyzed voltage transient properties of a PEM fuel cell stack. Based on the analysis a simple and accurate transient semi-empirical voltage model was designed, which can be easily applied to simulation of vehicle dynamics. The model took into consideration two aspects during the optimization: First, the shapes of the curves should be similar and also the absolute error between the curves should be at minimum. The model was semi-empirical and its purpose was to take indirectly the influence of both pressure and temperature into account.

The goal of the research by [Ou and Achenie 2005] was to develop a quantitatively good model for PEM fuel cells. A back-propagation feed-forward ANN and a radial basis function network hybrid models consisting of neural component and a physical component were developed and compared with the full blown ANN models. The rationale behind this approach was to combine the part of the model that is well known from the physics of the problem with the part that is poorly known but can be estimated quite efficiently using ANN. Inputs used were current density, concentration and flow rate of the methanol solution. 5.1.3 Support vector machines modeling Another technique used in empirical modeling besides ANN has been the use of support vector machines (SVM). A behavioural model with which one can predict PEM fuel cell behaviour under various operating conditions by using SVM was investigated by [Zhong et al. 2006]. He stated that for a given PEM fuel cell system, the relation between terminal voltage and current density is influenced by many operating parameters: cell temperature, air flow rate, hydrogen flow rate, air pressure, hydrogen pressure, relative humidity, membrane humidity and so on. In the experiment current density, which is decided by the uncontrollable load, and cell temperature were taken as variables and other operating parameters were held constant. Several reasons are stated for singling out cell temperature as one of the variables among all the operating parameters: first, it figures significantly in determining terminal voltage, it is easy to measure, temperature is hard to be held constant in real-world applications, and finally this simplification did does not impair the validity of the study, which is aimed at modeling PEM fuel cell with SVM. In another study by the same group a hybrid model composed of a least squares SVM and a pressure-incremental model was developed by [Zhong et al. 2007] to dispose operation conditions of current, temperature, cathode and anode gas pressures, which have major impacts in the performance of PEM fuel cells. The least squares SVM model was built to incorporate current and temperature and particle swarm optimization algorithm was used to improve its performance. The model can predict a PEM fuel cell voltage under any current, temperature, cathode and anode pressure. This model was stated to be a competitive solution for system level designs such as simulation, real-time control, online optimization and so on. 5.1.4 Parameter identification A study was made by [Ye et al. 2009] in which the physical parameters of a PEM fuel cell were identified with the help of an optimization algorithm from the voltage-current curve obtained by series of experiments. Parameter identification of a PEM fuel cell system was studied using particle swarm optimization. The proposed method did not necessitate initial guesses as close as possible to solutions, only a broad range specified for each of the parameters was required. Parameters of PEM fuel cell model were determined and optimized by [Mo et al. 2006] by means of niche hybrid GA using stack output-voltage, stack demand current, anode

pressure, and cathode pressure as input-output data. In the proposed method, niche techniques were used to maintain the population diversity for preventing premature convergence while Nelder-Mead simplex algorithm was used to improve local search capacities. The proposed algorithm was said to effectively alleviate premature convergence and improve weak exploitation of GA. In a similar work with [Mo et al. 2006] by [Ohenoja and Leiviskä 2009] GA was applied for parameter identification of different fuel cells. In the work two electrochemical models were fitted for three different fuel cells. Same parameters were indentified with [Mo et al. 2006], but the Nelder-Mead simplex method was not included in the search algorithm. The results were promising, even though the test material was too small to draw solid conclusions [Ohenoja and Leiviskä 2009]. 5.1.5 Design [Bao et al. 2006] states that stack and air system are the two most important components in the PEM fuel cell system. A modified, one-dimensional, steady-state, analytical model was used for studying their properties and trade-off between them. Three kinds of air system topologies, the pure screw compressor, serial booster, and exhaust expander were analyzed in the study. Combined with coordinate change preprocessing and analogue technique, ANN was used to treat the map of compressor and turbine air system. Also, a real-code GA was programmed to obtain the global optimum stoichiometric air ratio and the cathode outlet pressure. The work was said to be helpful in design of air system in fuel cells and that the steady-state optimum can also be used in dynamic control. [Mohamed and Jenkins 2004] emphasized that for stand alone power supply systems based on fuel cells to work efficiently, the fuel cell stack has to be configured so that it delivers maximum power output at the load’s operating voltage. In the study, GA was used to optimize the number of stack cells in series, number of stack cells in parallel and cell’s surface area. A mathematical polarization curve model for the PEM fuel cell was also presented. 5.1.6 Control [Hasikos et al. 2009] presented an integrated optimization and control tool for PEM fuel cell system. Using a detailed simulation model, a database was first generated containing steady-state values of the manipulated and controlled variables. In the second step, meta models utilizing radial basis function network (RBFN) were produced using the database. Finally a non-linear programming model was formulated that takes into account the constraints and limitations of the system and minimizes the consumption of hydrogen, for a given value of power demand. In another paper about fuel cell control [Chen Q. et al. 2009] studied control strategy for the hybrid system of power management and oxygen control of a PEM fuel cell system. The control employed a fuzzy clustering based modelling, constrained model predictive control, and adaptive switching between multiple models. In the proposed control

scheme, the characteristics of the hybrid system over its whole operating range were first identified and expressed as multiple linear discrete-time models by employing the fuzzy clustering method. Each model corresponded to a typical operating zone. Then constrained model-based predictive controls were designed for each model. Finally, an upper-layer adaptive switch was designed to determine the most appropriate model and to switch the corresponding model-based predictive control as needed.

5.2 Solid oxide fuel cells A schematic diagram of a direct-SOFC is shown in the figure below. At the anode, the methane-steam gas is reformed inside to generate hydrogen and carbon monoxide for the electrochemical reactions. At the cathode, oxygen decomposes into O2- combining with electrons due to the function of the catalyst. O2- passes through the electrolyte and reacts with hydrogen and carbon monoxide to form water and carbon dioxide respectively. The released electrons are the output to drive the external load, and finally return to cathode. The reforming and electrochemical reactions are given as follows [Li J. et al. 2008b]: Reforming reactions: CH4 +2 H2O ↔ CO2 + 4H2 CH4 + H2O ↔ CO + 3H2 CO + H2O ↔ CO2 + H2

(28) (29) (30)

Electrochemical reactions, anode: H2 + O2- → H´2O + 2 eCO + O2- → CO2 + 2 e-

(31) (32)

Electrochemical reactions, cathode: ½ O2 + 2 e- → O2-

(33)

The separators clamp together the electrodes and electrolyte, and reinforce the cell units. Further, the separators with machined channels determine the kind of flow field of fuel cells and act as polar plates to collect the electrical current [Li J. et al. 2008b]. The SOFC is based on a solid state ion-conducting electrolyte, which functions at high temperature. Due to high efficiency, high reliability, and low levels of noise and pollution, the SOFC has been considered as one of the most promising technologies for electrical energy generation. The high operating temperature of 1000 ºC allows internal reforming and promotes rapid kinetics with non-precious materials. Therefore, the SOFC can be directly fueled with pure hydrogen, natural gas, and other hydrocarbons [Li J. et al. 2008a].

Currently, the challenging tasks associated with SOFC are on reducing cost, achieving high electrical efficiency, and improving the stack operation security and load regulation [Zhang et al. 2008]. 5.2.1 Control Control of SOFCs has been studied in several papers. To facilitate controller design and analysis of systems, a nonlinear black-box multi-input-multi-output model for methanol direct-SOFC stack using wavelet networks was presented by [Li J. et al. 2008a]. To implement the construction and optimization of the wavelet network, a recursive approach, the Gram-Schmidt algorithm, cross-validation algorithm, and immune selection principles were applied. The obtained wavelet network model can be used for developing model based controllers of direct-SOFC. [Zhang et al. 2008] stated that the primary goal of the control system for SOFC is to allow the power plant deliver the desired power output under maximal electrical efficiency, but still within the security margin of operability. Based on that, a closed loop feedback control based on the nonlinear model predictive control for a planar SOFC was presented. The current density, fuel, and air molar rates were chosen as manipulated variables to control the output power, fuel utilization, and temperature. The mole fraction and temperature of the exit gases were set as state variables, which can be estimated by the moving horizon estimation method. In order to design effective temperature control strategies by model based control methods, a dynamic temperature model was presented by [Kang et al. 2008] using least squares SVM. The nonlinear temperature dynamics of the SOFC was represented by a nonlinear autoregressive with exogenous inputs model that was implemented using an least squares SVM regression model. The hyperparameters of the least squares SVM were automatically tuned by GA. [Jurado and Valverde 2005] developed a multi-objective GA in the design of a fuzzy logic control system for a SOFC by using the strategy of inverter flux control method. The GA optimized positions of vertexes of membership functions of each input and output of the fuzzy logic controller. The study showed how the multi-objective GA control can optimize a number of competing design objectives using only a single formulation. 5.2.2 Mechanistic modeling A detailed mathematical model for methanol direct-SOFC was presented by [Li J. et al. 2008b] incorporating simulation of chemical and physical processes in the fuel cell. The model was developed based on reforming and electrochemical reaction mechanisms, mass and energy conservation and heat transfer. A computational fluid dynamics method was used for solving the complicated multiple partial differential equations to obtain numerical applications. Further, the influence between distributions of chemical species concentrations, temperature, and current density during the simulation was illustrated and

discussed. The results showed that the particular characteristics of the direct-SOFC among fuel cells can aid in stack design and control. [Andreadis and Tsiakaras 2006] developed a one-dimension, steady-state, and single phase model describing mass transport and performance of a direct ethanol PEM fuel cell with the purpose of investigating how the operating parameters influence the cell performance. Information in the investigation included a wide range of temperature values, the membrane’s swelling, and the effect of catalyst layer thickness on the ethanol crossover rate.

5.3 Molten carbonate fuel cells In the only paper found dealing with molten carbonate fuel cells, [Sun et al. 2001] developed a novel GA with the whole colon divided into sub-colonies and thereby having two levels of competition both in colony level and sub-colony level. A molten carbonate fuel cell control was studied as an example to obtain high performance and safety, as well as long lifetime for the fuel cells working temperature and gas pressure were controlled. Fuzzy rules are designed based on variable structure control technology, in order to ensure stability and robustness of the control system. The GA was applied to searching the global optimum parameters of fuzzy variable structure control system of molten carbonate fuel cell.

6 CATALYST OPTIMIZATION Catalyst optimization is of fundamental significance in both reforming and gas cleaning units in the hydrogen production chain for fuel cells. A major part, if not majority, among all research done in the area of hydrogen production for fuel cells is about experimental research concerning catalyst optimization. Although this experimental catalyst research as such is not within the scope of this paper, a brief review on the latest methodological developments in the field of catalyst optimization is presented in this section. Catalyst design is a tedious and complex process involving many steps, many variables, and complex interactions among these variables making the experimental studies quite expensive and time consuming. Therefore, effective computational methods such as ANN can be used to interpret the findings of experimental studies, to feed the results to the future experiments, and therefore to increase the efficiency and the effectiveness of the experimental work [Günay and Yildirim 2008].

6.1 Combinatorial and high-throughput experimentation methods Combinatorial and high-speed screening techniques, which have revolutionized the search for new drug molecules have the ability to generate large libraries of samples and to evaluate their performance, and simultaneously reduce time and cost per sample and enable multicomponent parameter spaces to be explored. In catalyst development, the impact of this technology is promising, not only in synthesis and performance evaluation, but also for the optimization of operating parameters [Pescarmona et al. 1999]. 6.1.1 Selective combinatorial catalysis Selective combinatorial catalysis is a method for utilization of previous knowledge and understanding of catalytic systems to reduce experimental parameter space for catalyst screening. It starts with study of the existing literature that is synthesized into a reaction model that is used to select families of materials to be studied. The number of examined materials decreases with as the yield of experimentation increases [Li et al. 2003]. The application of selective combinatorial catalysis as described here is by [Li et al. 2003]. Other research described in this subsection applies similar procedure. The first level of experimentation in selective combinatorial catalysis is the use of in-situ, spatially resolved technique, to obtain an indication of catalytic activity of an array of materials in a single experiment. In the second level of experimentation, a parallel flow reactor is used that can measure the activity of the 10 most active samples found in the first level of experimentation. Finally the most active and selective catalyst found in the parallel reactor is studied in a single flow recycle reactor to get the reaction rate parameters, and is characterized to determine its bulk and surface structure. The knowledge gained from these results is used to formulate a revised model of the surface and the reaction pathway. The revised model is used in a second round of experiments to optimize the catalyst [Li et al. 2003].

Using the method presented above [Li et al. 2003] presented a knowledge-based approach for the selection of catalysts along with an experimental methodology to be used with high throughput and combinatorial catalytic experimentation. As a demonstration of the methodology, new results on the activity and selectivity of different catalysts for the preferential oxidation of CO were presented. The work of [Li et al. 2003] was extended by [Gracia et al. 2003]. A new pre-treatment procedure was used that resulted in a different sequence of activity selectivity among the various catalysts studied. Also new characterization results were presented which showed that dispersion is an important variable in considering the design of the next combinatorial iteration. Again, the method of [Li et al. 2003] was applied to study of the activity and selectivity for partial oxidation of methanol to H2 and CO2 on Zr, Ce promoted catalysts. Infrared thermography was first used as a descriptor of overall catalytic activity. Then activity and selectivity of samples with high infrared signal were measured in a flow reactor and characterized by nitrogen surface are measurement, x-ray diffraction, and x-ray photoelectron spectroscopy studies [Schuyten and Wolf 2006]. 6.1.2 Different pre-treatment procedures In combinatorial approaches catalysts with different compositions are considered as members of a catalyst library. However, in addition to the compositional variables, further elements of diversity can be created, such as structural and process variables. For example, it is well known that the application of different pre-treatment conditions can lead to different types of active sites with different catalytic performance. Consequently, different composition-activity relationships can be obtained as a result of different pretreatment conditions, i.e. the position of the optimum in the multidimensional experimental space can be significantly different [Tompos et al. 2008]. Based on the views stated above, a complex combinatorial approach was applied by [Tompos et al. 2008] for the design of multicomponent Au/MgO catalysts for CO oxidation in the presence of hydrogen. The aim was to demonstrate that supported gold catalysts can be prepared using combinatorial and high-throughput methods and the designed multicomponent catalysts can effectively be used in PROX reaction. Additionally it was demonstrated that the optimum composition is not a “steady concept”, but rather it depends strongly on variables such as pre-treatment conditions, which significantly influence the nanostructure of the multicomponent supported gold catalyst. The process started with real experimental optimization by means of holographic research strategy (HRS). After accumulation of adequate number of catalytic results ANN was trained, and combined with a GA and HRS. In combination with ANN the GA can find virtual optimum, while the combination to HRS lead to visualization of the experimental space in holographic maps allowing analysis of relevant experimental areas.

6.2 Neural network methods Recently, ANN have been increasingly applied to catalyst development through the prediction of catalyst performance such as activity, selectivity, and durability. ANN show high ability to find out the non-linear relationship between the experimentally obtained parameters and catalytic parameters [Omata et al. 2007]. [Omata et al. 2007] investigated preferential oxidation of 0.7-1 vol.% CO using the stoichiometric amount of O2 in excess hydrogen. Cobalt supported on SrCO3 showed high selectivity to preferential oxidation of CO and the new additive to the CO/SrCO3 catalyst was investigated for the high tolerance towards CO2 and H2O. Representative 10 elements (B, K, Sc, Mn, Zn, Nb, Ag, Nd, Re and Tl) were selected to for additives of solid catalyst. A supported cobalt catalyst with one of the above kind was prepared for CO preferential oxidation reaction. The activities at 240ºC and the physicochemical properties of the 10 elements were used as training data for RBFN. After training the RBFN the catalytic performance of the supported catalyst containing various element X as Co-X/SrCO3 was predicted. Activity of Cu-based mixed oxide catalysts for oxidative steam reforming of methanol was investigated by [Umegaki et al.2008]. In order to find binary Cu-X mixed oxide catalyst with high methanol conversion, high H2 selectivity, and low CO selectivity, an ANN was applied to relate the physicochemical properties of additive element X and the catalytic performance of the Cu-X mixed oxide catalyst. Experimental results of 14 Cu-X catalysts were used to train the ANN. The same method of using physiocchemical properties and ANN was applied to find a good third additive for each binary oxide. In the final step, ANN was also applied to improve the performance of the best found catalyst by optimizing the catalyst composition and the preparation conditions. This was reported to be the first report of applying ANN for discovery and development of catalysts for oxidative steam reforming of methanol. In a study by [Günay and Yildirim 2008] the design of Pt-Co-Ce/Al2O3 catalyst for the low temperature CO oxidation in hydrogen streams was modeled using ANN. The effects of five design parameters, namely Pt wt.%, Co wt.%, Ce wt.%, calcinations temperature, and calcinations time on CO conversion were investigated by modelling the experimental data obtained in the laboratory for 30 catalysts. Multiple regression models were also developed and compared with ANN models since the experimental data was produces using response surface method, which is generally used to perform multiple regression analysis.

7 OVERALL PROCESS OPTIMIZATION Usual main components for fuel processing system are reformer, WGS reactors, PROX reactors, and fuel cells. Efficient operation of the fuel process system depends on operating conditions of the reformer and their efficient energetic integration to the rest of the system. Changes in the major decision parameters of water-to-fuel ratio and reforming temperature affect the whole process chain. Optimization of the whole integrated process is preferable to optimization of each process unit separately [Francesconi et al. 2007b]. A review on key technological progress made over last two decades in the field of development of integrated fuel processors for hydrogen generation is given in [Qi et al. 2007].

7.1 Modelling An experimental methanol fuel reformer has been steady-state modelled by [Chuang et al. 2008] for fuel cell applications. The fuel reformer consisted of an auto-thermal reformer followed by an oxygen removal reactor, steam reformer and WGS reactor. The effluent from the WGS reactor was fed to a series of three PROX reactors. The steady state reactor models consisted of coupled nonlinear ordinary differential equations that described the change in gas phase molar flow rate and reactor temperature down the length of the reactor. Design optimization was conducted by using the mathematical model for the purpose of minimizing the combined volume of the steam reformer and WGS reactor. [Markova et al. 2009] presented experimental investigation of bio-ethanol auto-thermal reforming and water-gas shift processes for hydrogen production and regression analysis of the data. The main goal was to obtain regression relations between the most critical variables such as hydrogen, carbon monoxide and methane content in the reformate gas and independent factors such as air-to-fuel ratio, steam-to-carbon ratio, inlet temperature of reactants into reforming process and pressure and temperature in the auto-thermal reactor from the experimental data. Purpose of the regression models was to provide optimum values of the process factors that give the maximum amount of hydrogen. The experimental autothermal reforming system consisted of an evaporator, an autothermal reforming reactor, and a one-stage WGS reactor. The integrated power system consisting of reformer, PROX reactors, fuel cell, and heat management system was taken under consideration by [Ipsakis et al. 2009]. The two reactors were modelled via a system of partial differential equations and the species flow rates and reactor temperature were analyzed along the length of each reactor. The PEM fuel cell voltage-current characteristic was modelled via a non-linear equation depending on mass & energy balances of the concerned species. The main target was to present a complete study of an integrated system under operation with the aim of effective

integration of the developed mathematical models of subsystems. Finally, the heat management system was analyzed for providing insights for future control studies.

7.2 Control A number of studies on control of integrated systems for hydrogen production for fuel cells have been presented. [Wu and Pai 2009b] presented a novel heat-integrated fuel cell stack system with methanol reforming. The fuel processing units included methanol reformer, heat exchangers, a WGS reactor, and a PROX unit. The configuration was composed of fuel processing units, PEM fuel cell stack and heat exchangers. Well mixed methanol and oxygen flows in contact with counter-current flowing water dominate the production of hydrogen at the exit of fuel processing units and influence the stack temperature. The heat exchange conditions enhanced the utilization of energy of fuel processing units. To ensure stable operation, the model-free fuzzy incremental control scheme within the multi-loop feedback control framework was developed. [Pukrushpan J. et al. 2006] developed a control-oriented dynamic model of a catalytic partial oxidation-based fuel processor using physical principles. The components of the fuel processing system were hydro-desulfurizer, catalytic partial oxidation reactor, WGS reactor and PROX unit. The dynamic model for the fuel processing system concentrated on the dynamics associated with two main control objectives of precise and fast followup of hydrogen production to the current load and maintaining temperature of the catalytic partial oxidation reactor at certain point. It was also demonstrated how control theoretic tools can be used for analyzing necessary tradeoffs between the two control objectives and thus guide the controller and system design. Relative gain array analysis was applied to the model to determine control input/output pairs and to identify the interactions between two control loops. Moreover, it was demonstrated how simple linear observability analysis can facilitate decisions on sensor selection. A control system consisting of both feedforward and state-feedback controllers was presented by [Tsai 2008]. The system was designed using a well developed linear quadratic Gaussian and loop transfer recovery method for a fuel processing system. The fuel processing system used natural gas as a fuel and reacts with atmospheric air through a catalytic partial oxidation process. The control objective was focused on regulatory performance of output vector in response to a desired stack current command in face of load variation. First a Kalman filter was designed to provide an optimal estimation of state variables and to shape the target feedback function. Then an optimal two-degree-offreedom controller was designed to a linear quadratic performance index in the loop transfer recovery process. [Pukrushpan and Stefanopolou 2005] presented a model-based control analysis and design for fuel processing system. The system managed natural gas flow and humidified atmospheric air flow in order to regulate the amount of hydrogen in the fuel cell anode and the temperature of the catalytic partial oxidation reactor during transient power demands from the fuel cell. The focus of the control study was on hydrogen generation and thus the model incorporates relatively more details of the catalytic partial oxidation

reactor while using simple models for the WGS and PROX reactors, which function mainly for CO removal. The controller was developed using multivariable and modelbased control design techniques and linear quadratic methodology. Aim of a work by [Biset et al. 2009] was to investigate which would be a good preliminary plantwide control structure for the process of hydrogen production from bioethanol to be used in PEM fuel cell accounting only for steady-state information. The objective was to keep the process under optimal operation point, that is doing energy integration to achieve maximum efficiency. The heat exchange network involved reformer, burner, gas cleaning units and PEM fuel cell system. Applying steady-state simulation techniques and using thermodynamic models the performance of the complete system with two different control structures was evaluated for the most typical perturbations. The authors claim that up to time of the article there was no works considering the overall fuel processing system with PEM fuel cell, using bioethanol, where the main control structure was analyzed through a steady-state model. The analysis considered in the paper was stated to be useful as a basis for further process dynamic simulation working at optimal structure determination and testing whether or not the proposed control structure works well for each time instant. [Tsai et al. 2007] contributed in designing mass flow rates control of ethanol/water and temperature control of a thermal plasma reformer by first performing thermodynamic equilibrium prediction for ethanol steam reforming. Both mass flow rate and temperature controls were implemented in an electric control unit with a PID control.

7.3 Other studies In addition there are several studies published related to the optimization of the overall process of hydrogen production for fuel cells, but not directly related to either modeling or control. In one such paper [Simakov and Sheintuch 2009] demonstrated experimentally a novel concept for hydrogen generation my methane steam reforming in a thermally coupled catalytic fixed bed membrane reformer. The reactor was built from three concentric compartments, indirectly couples the endothermic methane steam reforming with the exothermic methane oxidation, while hydrogen was separated by a perm-selective Pd membrane. The study focused on the determination of the key operation parameters and understanding their influence on the reactor performance. The study also focused on the determination of the key operation parameters and understanding their influence on the reactor performance. To the best of authors knowledge this was the first experimental demonstration of methane steam reforming carried out in a compact membrane packed bed reformer without any external heat supply, when the enthalpy required for the endothermic steam reforming and for the heat losses compensation is provided by methane catalytic oxidation carried out in an adjacent fixed bed.

With the object of developing an integrated process flowsheet for production of hydrogen by bioethanol, three scenarios were simulated by Aspen - HYSYS® software by

[Hernandez and Kafarov 2009] and their efficiencies were calculated to find the best process configuration. In the first scenario, heat integration was made by using the exhausted gas from SOFC to pre-heat the feed streams of air, water and ethanol. In the second scenario, heat and mass integration was made by recycling a fraction of outlet stream from SOFC anode and using the post-combustion gas from SOFC to pre-heat the feed streams of air and ethanol. In third scenario, heat and mass integration was made in the same way as in scenario 2. and additional power recuperation by a micro-gas-turbine was made and fed with post combustion gas from SOFC in a before stage pre-heating the air and bioethanol feed streams. Heat exchanger network and SOFC were simulated the same way as in scenario 2. Also a sensitivity analysis was made with the object of evaluating the influence of temperature and bioethanol/water feed molar ratio in SOFC performance and process efficiency. In a paper by [Rabenstein and Hacker 2008] different options in the process of reforming ethanol to hydrogen-rich feed for fuel cells using steam-reforming, partial-oxidation and the combined processes was explored. Equilibrium compositions of ethanol reforming as a function of steam-to-ethanol ratio, oxygen-to-ethanol ratio and temperature at atmospheric pressure were presented. Thermodynamic equilibrium was studied by Gibbs enthalpy minimization including the possibility of solid coke formation. The species considered are hydrogen, carbon monoxide, carbon dioxide, methane, water, ethanol, ethylene, ethane, acetaldehyde, acetic acid and solid carbon. Performance of hydrogen production via steam methane reforming was evaluated by [Simpson and Lutz 2007] using exergy analysis, with emphasis on exergy flows, destruction, waste and efficiencies. The system comprised of steam reformer, WGS reactor and hydrogen separation phases. The analysis presented in the paper used system component models, a chemical equilibrium reformer model and detailed heat integration to perform exergy analysis. Variations in the operating parameters illustrated the system’s performance sensitivity and provided guidance for where research and development efforts should be concentrated. An exergy analysis of methanol autothermal generating hydrogen system for PEM fuel cell is presented by [Wang. and Wang 2006]. The process combined a catalytic combustion heat exchanger using partial off-gases containing hydrogen as a feedstock with an autothermal reformer, two WGS reactors and four PROX reactors. Energy and exergy of the system were calculated and analyzed. By the comparison and analysis of exergy losses in various subsystems, the unit bottlenecking the optimization of system was found. Based on the thermodynamic equilibrium analysis, the unit was optimized by the selection of the favorable operational conditions. The conclusions can help to optimize methanol autothermal generating hydrogen system for PEM fuel cell.

8 CONCLUSIONS Latest developments in the field of hydrogen production for fuel cells from ethanol, methanol, and methane were reviewed in this report. With an emphasis on intelligent optimization, both developments on the use of optimization in sub-processes of hydrogen production, as well as optimization of whole production chains have been examined. Research has been found to be active on every unit process of the production chain. In reformer optimization, the focus is shifting from conventional reactors to the research of membrane and other new reactor types, particularly microreactor technology. The advantages of higher conversion of membrane reactors and unique flow and heat/mass transfer properties of microreactors have drawn significant amount of interest resulting in accelerated research. The problem of slowness of the WGS reaction, resulting in WGS units to be the biggest and thereby bottleneck units for many fuel cell applications has inspired a renewed interest in gas cleaning reactor design research. The research on volume minimization, different reactor configurations and relative sizes of different reactor components has been a central issue on latest research. Additionally, as with reformers, research on the microreactor technology seems to be increasing. The research on control strategies however, has concentrated on whole process control along with fuel cell control, with very few studies on control of reformers or gas cleaning units. The fuel cell control optimization is the most active area of research of the whole production chain. Two major focus areas of latest research are empirical modelling of fuel cell performance, by ANN etc. and intelligent control systems of fuel cells. Catalysts are of outmost importance in efficient hydrogen production. Catalyst development has traditionally been a slow and expensive process. Interesting new methodology has emerged in catalyst development using either combinatorial techniques or empirical methods like ANN which enable partly automatic screening of potentially effective candidates for generation of more efficient catalysts. There are clear oppotunities related to efficient energy usage and bottlenecking for integrated whole process optimization. A variety of methods have been employed in research of control of the whole process optimization. No single control technique has been found to dominate research on the field of overall process control.

REFERENCES Aboudheir A., Akande A., Idem R. and Dalai A. (2006). Experimental studies and comprehensive reactor modeling of hydrogen production by the catalytic reforming of crude ethanol in a packed bed tubular reactor over a Ni/Al2O3 catalyst. International Journal of Hydrogen Energy, vol. 31, 752-761. Adams II T.A. and Barton P.I. (2009). A dynamic two-dimensional heterogeneous model for water gas shift reactors. International Journal of Hydrogen Energy, vol. 34, 88778891. Akande A., Aboudheir A., Idem R. and Dalai A. (2006). Kinetic modeling of hydrogen production by the catalytic reforming of crude ethanol over a co-precipitated Ni-Al2O3 catalyst in a packed bed tubular reactor. International Journal of Hydrogen Energy, vol. 31, 1707-1715. Akpan E., Akande A., Aboudheir A., Ibrahim H. and Idem R. (2007). Experimental, kinetic and 2-D reactor modeling for simulation of the production of hydrogen by the catalytic reforming of concentrated crude ethanol (CRCCE) over a Ni-based commercial catalyst in a packed-bed tubular reactor. Chemical Engineering Science, vol. 62, 3112-3126. Andreadis G. and Tsiakaras P. (2006). Ethanol crossover and direct ethanol PEM fuel cell performance modeling and experimental validation. Chemical Engineering Science, vol. 61, 7497-7508. Arteaga L.E., Peralta L.M., Kafarov V., Casas Y. and Gonzales E. (2008). Bioethanol steam reforming for ecological syngas and electricity production using a fuel cell SOFC system. Chemical Engineering Journal, vol. 136, 256-266. Ayastuy J.L., Gutierrez-Ortiz M.N., Gonzalez-Marcos J.A., Aranzabal A. and GonzalezVelasco J.R. (2005). Kinetics of the low-temperature WGS reaction over a CuO/ZnO/Al2O3 catalyst. Ind.Eng.Chem.Res., vol. 44, 41-50. Bao C., Ouyang M. and Yi B. (2006). Modeling and optimization of the air system in polymer exchange membrane fuel cell systems. Journal of Power Sources, vol. 156, 232-243. Barelli L., Bidini G., Gallorini F. and Servili S. (2008). Hydrogen production through sorption-enchanced steam methane reforming and membrane technology. Energy, vol. 33, 554-570. Barrio V.L., Schaub G., Rohde M., Rabe S., Vogel F., Cambra J.F., Arias P.L. and Guemez M.B. (2007). Reactor modeling to simulate catalytic partial oxidation and steam reforming of methane. Comparison of temperature profiles and strategies for hot spot minimization. International Journal of Hydrogen Energy, vol. 32, 1421-1428. Batista M.S., Assaf E.M., Assaf J.M. and Ticianelli E.A. (2006). Double bed reactor for the simultaneous steam reforming of ethanol and water gas shift reactions. International journal of hydrogen energy, vol. 31, 1204-1209. Biset S., Nieto Deglioumini L., Basualdo M., Garcia V.M. and Serra M. Analysis of control structures for an integrated ethanol processor for proton exchange membrane fuel systems. Journal of Power Sources, vol. 192, 107-113. Bittani S., Canavese S., De Marco A., Prandoni V. and Serrau D. (2008). Towards CleanCoal Control Technologies: Modeling Conversion of Carbon Oxide into Hydrogen by

Shift Reactor. Proceedings of the 17th World Congress, The International Federation of Automatic Control, Seoul, Korea, July 6-11, 2008, 10963-10970. Biyikoglu A. (2005). Review of proton exchange membrane fuel cell models. International Journal of Hydrogen Energy, vol. 30, 1181-1212. Cheddie D.F. and Munroe D.H. (2007). A two-phase model of an intermediate PEM fuel cell design. International Journal of Hydrogen Energy, vol. 32, 832-841. Chen Q., Gao L., Dougal R. and Quan S. (2009). Multiple model predictive control for a hybrid proton exchange membrane fuel cell system. Journal of Power Sources, vol. 191, 473-482. Chen W.-H., Lin M.-R., Jiang T.L. and Chen M.-H. (2008). Modeling and simulation of hydrogen generation from high-temperature and low-temperature water gas shift reactions. International Journal of Hydrogen Energy, vol. 33, 6644-6656. Cheng S.-H., Chen H.-J., Chang H., Chang C.-K., Chen Y.-M. (2008). Multi-objective optimization for two catalytic membrane reactors – Methanol synthesis and hydrogen production. Chemical Engineering Journal, vol. 63, 1428-1437. Choi Y. and Stenger G. (2003). Water gas shift reaction kinetics and reactor modeling for fuel cell grade hydrogen. Journal of Power Sources, vol. 124, 432-439. Chuang C.-C., Chen Y.-H., Ward J.D., Yu C.-C., Liu Y.-C. and Lee C.-H. (2008). Optimal design of an experimental methanol fuel reformer. International Journal of Hydrogen Energy, vol. 33, 7062-7073. Cipiti F., Pino L., Vita A., Lagana M. and Recupero V. (2009). Model-based analysis of reactor geometrical configuration on CO preferential oxidation performance. International Journal of Hydrogen Energy, vol. 34, 4463-4474. De Jong M., Reinders A.H.M.E., Kok J.B.W. and Westendorp G. (2009). Optimizing a steam-methane reformer for hydrogen production. International Journal of Hydrogen Energy, vol. 34, 285-292. Ding O.L. and Chan S.H. (2008). Water-gas-shift reaction – A 2-D modeling approach. International Journal of Hydrogen Energy, vol. 33, 4325-4336. Dobrego K.V., Gnezdilov N.N., Lee S.H. and Choi H.K. (2008). Partial oxidation of methane in a reverse flow porous media reactor. Water admixing optimization. International Journal of Hydrogen Energy, vol. 33, 5535-5544. Francesconi J.A., Mussati M.C. and Aguirre P.A. (2007a). Analysis of design variables for water-gas-shift reactors by model based optimization. Journal of Power Sources, vol. 173, 467-477. Francesconi J. A., Mussati M. C., Mato R.O. and Aguirre P.A. (2007b). Analysis of the efficiency of an integrated ethanol processor for PEM fuel cell systems. Journal of Power Sources, vol. 167 151-161. Fu C.-H. and Wu J.C.S. (2007). Mathematical simulation of hydrogen production via methanol steam reforming using double-jacketed membrane reactor. International Journal of Hydrogen Energy, vol. 32, 4830-4839. Gallucci F., De Falco M., Tosti S., Marrelli L. and Basile A. (2008a). Co-current and counter-current configurations for ethanol steam reforming in a dense Pd-Ag membrane reactor. International Journal of Hydrogen Energy, vol. 33, 6165 – 6171. Gallucci F., De Falco M., Tosti S., Marrelli L. and Basile A. (2008b). Ethanol steam reforming in a dense Pd-Ag membrane reactor: A modeling work. Comparison with the traditional system. International Journal of Hydrogen Energy, vol. 33, 644-651.

Giunta P., Amadeo N. and Laborde M. (2006). Simulation of a low temperature water gas shift reactor using the heterogenous model/application to a pem fuel cell. Journal of Power Sources, vol. 156, 489-496. Gracia F., Li W. and Wolf E.E. (2003). The preferential oxidation of CO: selective combinatorial activity and infrared studies. Catalysis Letters, vol. 91, 235-242. Günay M.E. and Yildirim R. (2008). Neural network aided design of Pt-Co-Ce/Al2O3 catalyst for selective CO oxidation in hydrogen rich streams. Chemical Engineering Journal, vol. 140, 324-331. Halabi M.H., De Groon M.H.J.M., Van der Schaaf J., Cobden P.D. and Schouten J.C. (2008). Modeling and analysis of autothermal reforming of methane to hydrogen in a fixed bed reformer. Chemical Engineering Journal, vol. 137, 568-578. Hasikos J., Sarimveis H., Zervas P.L. and Markatos N.C. (2009). Operational optimization and real-time control of fuel-cell systems. Journal of Power Sources, vol. 193, 258-268. Hernandez L. and Kafarov V. (2009). Use of bioethanol for sustainable electrical energy production. International Journal of Hydrogen Energy, vol. 34, 7041-7050. Hoang D.L. and Chan S.H. (2004). Modeling of a catalytic autothermal methane reformer for fuel cell applications. Applied Catalysis A, vol. 269, 207-216. Hou Y., Zhuang M. and Wan G. (2007). International Journal of Hydrogen Energy, vol. 32, 857-862. Ipsakis D., Voutetakis S., Seferlis P., Papadopoulou S. and Stoukides M. (2009) Modeling and Analysis of an Integrated Power System Based on Methanol Autothermal Reforming. 17th Mediterranean Conference on Control & Automation, Makedonia Palace, Thessaloniki, Greece, June 24-26, 2009, 1421–1426. Jemei S., Hissel D., Pera M.C. and Kaufmann J.M. (2003). On-board fuel cell power supply modeling on the basis of neural network methodology. Journal of Power Sources, vol. 124, 479-486. Jurado F. and Valverde M. (2005). Enchancing the electrical performance of a solid oxide fuel cell using multiobjective genetic algorithms. Renewable Energy, vol. 30, 881-902. Kang Y.-W., Li J., Cao G.-Y., Tu H.-Y., Li J. and Yang J. (2008). Dynamic modeling of an SOFC using least squares support vector machines. Journal of Power Sources, vol. 179, 683-692. Kim G.-Y., Mayor J.R. and Ni J. (2005). Parametric study of microreactor design for water gas shift reactor using an integrated reaction and heat exchange model. Chemical Engineering Journal, vol. 110, 1-10. Kolb G., Hessel V., Cominos V., Hofmann C., Löwe H., Nikolaidis G., Zapf R., Ziogas A., Delsman E.R., de Croon M.H.J.M., Schouten J.C., de la Iglesia O., Mallada R. and Santamaria J. (2007). Selective oxidations in micro-structured catalytic reactors – For gas-phase reactions and specifically for fuel processing for fuel cells. Catalysis Today, vol. 120, 2-20. Kuznetsov V.V. and Kozlov S.P. (2008). Modeling of methanol-to-hydrogen steam reforming with a heat flux distributed along a microchannel. Thermophysics and aeromechanics, vol. 15, no 3, 509-517. Lattner J.R. and Harold M.P. (2007). Autothermal reforming of methanol: Experiments and modeling. Catalysis Today, vol. 120, 78-89.

Lattner J.R. and Harold M.P. (2005). Comparison of methanol-based fuel processors for PEM fuel cell systems. Applied Catalysis B: Environmental, vol. 56, 149-169. Lee W.-Y., Park G.-G., Yang T.-H., Yoon Y.-G. and Kim C.-S. (2004). Empirical modeling of polymer electrolyte membrane fuel cell performance using artificial neural networks. International Journal of Hydrogen Energy, vol. 29, 961-966. Li J., Kang Y.-W., cao G.-Y., Zhu X.-J., Tu H.-Y. and Li J. (2008a). Nonlinear identification of a DIR-SOFC stack using wavelet networks. Journal of Power Sources, vol. 179, 673-682. Li J., Kang Y.-W., Cao G.-Y., Zhu X.-J., Tu H.-Y. and Li J. (2008b). Numerical simulation of a direct internal reforming solid oxide fuel cell using computational fluid dynamics method. Journal of Zhejiang university Science A, vol. 9, 961-969. Li W., Gracia F.J. and Wolf E.E. (2003) Selective combinatorial catalysis; challenges and opportunities: the preferential oxidation of carbon monoxide. Catalysis Today, vol. 81, 437-447. Mahecha-Botero A., Chen Z., Grace J.R., Elnashaie S.S.E.H., Lim C.J., Rakib M., Yasuda I. and Shirasaki Y. (2009). Comparison of fluidized bed flow regimes for steam methane reforming in membrane reactor: A simulation study. Chemical Engineering Journal, vol. 64, 3598-3613. Markova D., Bazbauers G., Valters K., Alhucema Arias R. Weuffen C. and Rochlitz L. (2009) Optimization of bio-ethanol autothermal reforming and carbon monoxide removal processes. Journal of Power Sources, vol. 193, 9-16. Mo Z.-J., Zhu X.-J., Wei L.-Y. and Cao G.-Y. (2006). Parameter optimization for a PEMFC model with a hybrid genetic algorithm. International Journal of Energy Research, vol. 30, 585-597. Mohamed I. and Jenkins N. (2004). Proton exchange membrane (PEM) fuel cell stack configuration using genetic algorithms. Journal of Power Sources, vol. 131, 142-146. Nandasana A.D., Ray A. K. and Gupta S.K. (2003). Dynamic Model of an Industrial Steam Reformer and Its Use for Multiobjective Optimization. Ind. Eng. Chem. Res., vol. 42, 4028-4042. Ni M., Leung D.Y.C. and Leung M.K.H (2007). A review on reforming bio-ethanol for hydrogen production. International Journal of Hydrogen Energy, vol. 32, 3238-3247. Nikolaidis G., Baier T., Zapf R., Kolb G., Hessel V. and Maier W. (2009). Kinetic study of CO preferential oxidation over Pt-Rh/γ-Al2 O3 catalyst in a micro-structured recycle reactor. Catalysis Today, vol. 145, 90-100. Ohenoja M. and Leiviskä K. (2008). Etanolin reformoinnin mallinnus ja simulointi. Diplomityö, Oulun Yliopisto, Prosessi- ja ympäristötekniikan osasto, Säätötekniikan laboratorio. Ohenoja M. and Leiviskä K. (2009). Identification of electrochemical model parameters in PEM fuel cells. Power Engineering, Energy and Electrical Drives, POWERENG ’09, International Conference on Power Engineering, Energy and Electrical Drives, 363-368. Oliveira V.B., Falcao D.S., Rangel C.M. and Pinto A.M.F.R. (2007). A comparative study of approaches to direct methanol fuel cells modeling. International Journal of Hydrogen Energy, vol. 32, 415-424. Oliva D.G., Francesconi J.A., Mussati M.C. and Aguirre P.A. (2008). CO-PrOx reactor design by model based optimization. Journal of Power Sources, vol. 182, 307-316.

Omata K., Kobayashi Y. and Yamada M. (2007). Artificial neural network aided virtual screening of additives to a Co/SrCO3 catalyst for preferential oxidation of CO in excess hydrogen. Catalysis Communications, vol. 8, 1-5. Ou S. and Achenie E.K. (2005). A hybrid neural network model for PEM fuel cells. Journal of Power Sources, vol. 140, 319-330. Palo D.R., Dagle R.A. and Holladay J.D. (2007). Methanol steam reforming for hydrogen production. Chem.Rev., vol. 107, 3992-4021. Peppley B.A., Amphlett J.C., Kearns L.M. and Mann R.F.(1999). Applied Catalysis A, vol. 179, 31-49. Perna A. (2007). Hydrogen from ethanol: Theoretical optimization of a PEMFC system integrated with a steam reforming processor. International Journal of Hydrogen Energy, vol. 32, 1811-1819. Pescarmona P.P., Van der Waal J.C., Maxwell I.E. and Maschmeyer T. (1999). Combinatorial chemistry, high-speed screening and catalysis. Catalysis Letters, vol. 63, 1-11. Pukrushpan J., Stefanopolou A., Varigonda S., Eborn J. and Haugstetter C. (2006). Control-orientated model of fuel processor for hydrogen generation in fuel cell applications. Control Engineering Practise, vol. 14, 277-293. Pukrushpan J.T. and Stefanopolou A.G. (2005). Control of natural gas catalytic partial oxidation for hydrogen generation in fuel cell applications. IEEE Transactions on Control Systems Technology, vol. 13, no. 1, 3-14. Qi A., Peppley B. and Karan K. (2007). Integrated fuel processors for fuel cell application: A review. Fuel processing technology, vol. 88, 3-22. Quiney A.S., Germani G. and Schuurman Y. (2006). Optimization of a water-gas shift reactor over a Pt/ceria/alumina monolith. Journal of Power Sources, vol. 160, 11631169. Rabenstein G. and Hacker V. (2008). Hydrogen for fuel cells from ethanol by steamreforming, partial oxidation and combined autothermal reforming: A thermodynamical analysis. Journal of Power Sources, vol. 185, 1293-1304. Rouss V. and Charon W. (2008). Multi-input and multi-output neural model of the mechanical nonlinear behavior of a PEM fuel cell system. Journal of Power Sources, vol. 175, 1-17. Saengrung A., Abtahi A. and Zilouchian A. (2007). Neural network model for commercial PEM fuel cell system. Journal of Power Sources, vol. 172, 749-759. Sahoo D.R., Vajpai S., Patel S. and Pant K.K. (2007). Kinetic modeling of steam reforming of ethanol for the production of hydrogen over Co/Al2O3 catalyst. Chemical Engineering Journal, vol. 125, 139-147. Schuyten S. and Wolf E.E. (2006). Selective combinatorial studies on Ce and Zr promoted Cu/Zn/Pd catalysts for hydrogen production via methanol oxidative reforming. Catalysis Letters, vol. 106, No 1-2, 7-14. Simakov D.S.A. and Sheintuch M. (2009). Demonstration of a scaled-down autothermal membrane methane reformer for hydrogen generation. International Journal of Hydrogen Energy, vol. 34, 8866-8876. Simeone M., Salamme L., Scognamiglio D., Allouis C. and Volcipelli G. (2008). Effect of water addition and stoichiometry variations on temperature profiles in an

autothermal methane reforming reactor with Ni catalyst. International Journal of Hydrogen Energy, vol. 33, 1252-1261. Simpson A.P. and Lutz A.E. (2007). Exergy analysis of hydrogen production via steam methane reforming. International Journal of Hydrogen Energy, vol. 32, 4811-4820. Srinivas S. and Gulari E. (2006). Preferential CO oxidation in a two stage packed-bed reactor: Optimization of oxygen split ratio and evaluation of system robustness. Catalysis Communications, vol. 7, 819-826. Stutz M.J., Hotz N. and Poulikakos D. (2006). Optimization of methane reforming in a microreactor – effects of catalyst loading and geometry. Chemical Engineering Science, vol. 61, 4027-4040. Sun J., DesJardins J., Buglass J. and Liu K. (2005). Noble metal water gas shift analysis: Kinetics study and reactor design. International Journal of Hydrogen Energy, vol. 30, 1259-1264. Sun X.-J., Cao G.-Y. and Zhu X.-J. (2001). A novel genetic algorithm and its application in fuzzy variable structure control of fuel cell. Journal of Intelligent and Robotic Systems, vol. 31, 299-316. Tompos A., Hegedüs M., Margitfalvi J.L., Szabó E.G. and Vėgvári L. (2008). Multicomponent Au/MgO catalysts designed for selective oxidation of carbon monoxide. Application of a combinatorial approach. Applied Catalysis A: General, vol. 334, 348-356. Tosti S., Basile A., Borelli R., Borgognoni F., Castelli S., Fabbricino M. Gallucci F. and Licusati C. (2009). Ethanol steam reforming kinetics of a Pd-Ag membrane reactor. International Journal of Hydrogen Energy, vol. 34, 4747-4754. Trimm D.L. (2005). Minimization of carbon monoxide in a hydrogen stream for fuel cell application. Applied Catalysis A, vol. 296, 1-11. Tsai H.-L. (2008). Optimal control design of fuel processing system by linear quadratic Gaussian and loop transfer recovery method. Journal of Chinese Institute of Engineers, vol. 31, no 3, 369-378. Tsai H.-L., Wang C.-S. and Duc P. M. (2007). Control Design of Ethanol Steam Reforming in Thermal Plasma Reformer. 16th IEEE International Conference on Control Applications, Singapore, 1–3 October, 2007, 706–711. Umegaki T., Masuda A., Omata K. and Yamada M. (2008) Development of high performance Cu-based ternary oxide catalyst for oxidative steam reforming of methanol using an artificial neural network. Applied Catalysis A: General, vol. 351, 210-216. Vahabi M. and Akbari M.H. (2009). Three-dimensional simulation and optimization of an isothermal PROX microreactor for fuel cell applications. International Journal of Hydrogen Energy, vol. 34, 1531-1541. Vaidya P.V. and Rodriques A. E. (2006). Insight into steam reforming of ethanol to produce hydrogen for fuel cells. Chemical Engineering Journal, vol. 117, 39-49. Wang H.M. (2008). Experimental studies on hydrogen generation by methane autothermal reforming over nickel-based catalyst. Journal of Power Sources, vol. 506511. Wang S. and Wang S. (2006). Exergy analysis and optimization of methanol generating hydrogen system for PEMFC. International Journal of Hydrogen Energy, vol. 31, 1747-1755.

Wu W. and Pai C. (2009a). Modeling and control of a proton exchange membrane fuel cell system with alternative fuel sources. Ind. Eng. Chem. Res., vol. 48, 8999-9005. Wu W. and Pai C. (2009b). Control of a heat-integrated proton exchange membrane fuel cell system with methanol reforming. Journal of Power Sources, vol. 194, 920-930. Yang J., Mou H.-G. and Jian L. (2009). Predictive control of solid oxide fuel cell based on an improved Takagi-Sugeno fuzzy model. Journal of Power Sources, vol. 193, 699705. Ye M., Wang X. and Xu Y. (2009). Parameter identification for proton membrane fuel cell model using particle swarm optimization. International Journal of Hydrogen Energy, vol. 34, 981-989. Zhang X.W., Chan S.H., Ho H.K., Li G. and Feng Z. (2008). Nonlinear model predictive control based on the moving horizon state estimation for the solid oxide fuel cell. International Journal of Hydrogen Energy, vol. 33, 2355-2366. Zhong Z.-D., Zhu X.-J. and Cao G.-Y. (2006). Modeling a PEMFC by a support vector machine. Journal of Power Sources, vol. 160, 293-298. Zhong Z.-D., Jian X.-J., Cao G.-Y. and Shi J.-H. (2007). A hybrid multi-variable experimental model for a PEMFC. Journal of Power Sources, vol. 164, 746-751.