Kinetic Modeling of Autothermal Reforming of Dimethyl Ether

Sep 28, 2010 - A global kinetic model was developed for the autothermal reforming of dimethyl ether (DME) over a Pd−Zn/Al2O3 catalyst on a cordierit...
3 downloads 9 Views 2MB Size
Download PDF
9712

Ind. Eng. Chem. Res. 2010, 49, 9712–9719

Kinetic Modeling of Autothermal Reforming of Dimethyl Ether Derek Creaser,*,† Marita Nilsson,‡ Lars J. Pettersson,‡ and Jazaer Dawody§ Chemical Reaction Engineering, Chalmers UniVersity of Technology, SE-412 96 Go¨teborg, Sweden, Department of Chemical Engineering and Technology, Royal Institute of Technology, SE-100 44 Stockholm, Sweden, and PowerCell Sweden AB, RuskVa¨dersgatan 12, SE-418 34 Go¨teborg, Sweden

A global kinetic model was developed for the autothermal reforming of dimethyl ether (DME) over a Pd-Zn/ Al2O3 catalyst on a cordierite monolith. A kinetic model consisting of five key overall reactions was found to capture the main features of experimental data. The modeling also accounted for heat transport effects in the reactor that are of importance when coupling the exothermic oxidation reactions with endothermic steam reforming reactions. The modeling confirmed that oxidation reactions dominate near the inlet of the reactor, generating a local hot spot. The heat from oxidation reactions accelerates the reforming reactions. Water adsorption was found to have a weak detrimental influence on the activity. On the basis of the model, the influence of the reactor scale and oxygen supply by air feed on the performance of the reactor was examined. 1. Introduction Dimethyl ether (DME) is considered an attractive alternative diesel fuel. A DME fueled diesel engine has lower emissions of particulates,1 which permits operation of the engine to be better tuned for lower NOx emissions. Engine noise emissions are also reduced.2 In addition, there is reportedly a strong potential for low CO2 emissions for DME produced from renewable feedstocks according to well-to-wheel studies.3 Idling of trucks and other heavy-duty vehicles for on board power generation is an enormously inefficient use of power from a diesel engine. As a result, there have been ongoing efforts to develop fuel cell auxiliary power (APU) units that produce electricity efficiently with low emissions.4 Experimental studies concerning the production of hydrogen via reforming of diesel or surrogate diesel fuels have started to appear in the literature.5-8 Yet also production of hydrogen from DME either by steam reforming9,10 or by autothermal reforming11-13 has garnered some attention. Hydrogen can be produced by reforming of a fuel by three general processes, where the two most well established in industry are partial oxidation (POX) and steam reforming (SR). POX is an exothermic process that offers advantages such as fast start-up and compact design, whereas SR is an endothermic process that offers higher efficiency for hydrogen production and lower byproduct formation. The third reforming process is autothermal reforming or oxidative reforming that can be considered somewhat of a hybrid because it combines both oxidation and steam reforming reactions to become a nearly thermally neutral process. An autothermal reforming process ideally merges the advantages of fast-response and compact design of POX with efficiency for hydrogen production close to that of SR.14 Although experimental studies of DME reforming have been undertaken, kinetic modeling studies are sparse. Feng et al. have reported on a kinetic study of steam reforming of DME over a physical mixture of ZSM-5 intended to catalyze DME hydrolysis and a CuO-ZnO-based catalyst for methanol SR.15 Here, we report on a kinetic modeling study of autothermal reforming of * To whom correspondence should be addressed. E-mail: creaser@ chalmers.se. † Chalmers University of Technology. ‡ Royal Institute of Technology. § PowerCell Sweden AB.

DME over a PdZn/γ-Al2O3 catalyst. The catalyst is in the form of a washcoated monolith similar to that likely to be applied in an actual APU process. Because autothermal reforming combines exothermic and endothermic reactions, special attention is given to properly describing heat transport processes in the reactor. The experimental data forming the basis of this modeling study are taken from a part of a previous publication.11 In summary, the unique contributions of this work to the literature include both a kinetic study of DME reforming on a noble metal-based catalyst but also a modeling study of an autothermal reforming process. 2. Experimental Methods Because the experimental results are taken from a previous publication,11 only a brief review of the experimental methods will be given here. The catalyst was PdZn/γ-Al2O3. The weight ratio of Pd:Zn was 1:9, and the loading of Pd was 3 wt %. The catalyst was prepared by coimpregnation of the γ-Al2O3 with the aqueous metal nitrate solutions using the incipient wetness method. The catalyst was supported on a cordierite monolith sample (400 cpsi) as a washcoat with a loading of 20 wt %. The monolith had a diameter of 20 mm and a length of 35 mm. It was placed in a metallic sample holder that allowed reactant gases to flow through an open inlet area with an 18 mm diameter. The sample holder was subsequently placed in the reactor vessel. The reactor walls and sample holder were maintained at the reactant feed temperature with an external heating coil. The reactant gases were separately preheated with a capillary heater. The product gas temperature was measured with a thermocouple positioned directly downstream the monolith. The reactant gases were fed with separate mass flow controllers. All experiments were performed with a constant gas hourly space velocity of 15 000 h-1. The feed mole fraction of DME was also constant at 6.7 mol %. The feed molar ratio of O2: DME was varied between 0.4 to 0.9, and the feed molar ratio of H2O:DME was varied between 1 to 3. The feed rate of N2 was varied to maintain the space velocity at 15 000 h-1. At these conditions, the O2:N2 feed molar ratio varied from 0.036 to 0.086 and thus was lower than that for air feed (0.41). The feed gas temperature was varied from 350 to 450 °C.

10.1021/ie100834v  2010 American Chemical Society Published on Web 09/28/2010

Ind. Eng. Chem. Res., Vol. 49, No. 20, 2010

The concentrations of CO, CO2, and H2 were analyzed in the product gas with a Maihak modular system S710 by nondispersive infrared and thermal conductivity. In addition, the concentrations of DME, CO2, CO, CH3OH, and CH4 were analyzed by an FTIR instrument (Gasmet Cr-2000).

9713

Table 1. Boundary Conditions for the 2D Monolithic Reactor Washcoat Mass Balance and Solid Heat Balance location

boundary condition

x ) 0 (washcoat surface) x ) twc (washcoat/ cordierite boundary)

yic ) yis dyic )0 dx

3. Modeling Methods Two different transport models were used with different levels of detail. A more rigorous 2D model accounted for mass and heat transport between the gas and solid as well as heat transport both radially and axially in the monolith. Convergence of this model to a solution could not be easily obtained for all ranges of kinetic parameters and the various kinetic models during the parameter optimization studies. Instead, for parameter optimization purposes, a more robust and less computationally demanding 1D model was used. This model lumped the radial heat transport into an overall heat transport term that was determined from parallel simulations with the 2D model. The two transport models are described below in detail. 3.1. 2D Monolith Reactor Model. The 2D model is pseudocontinuous, assuming axial symmetry, and is similar to a model presented earlier by other authors.16 The following mass balance accounted for mass transport between the bulk gas and the catalyst washcoat surface of each component: mtot dwi AcSvkcPtot ) (yis - yig) Mi dz RT

∑νr

i javg

)

SvkcPtot (yis - yig) RT

In addition, diffusion and reaction of each component in the washcoat was accounted for by the following mass balance: d2yic 2

) -(

dx

FcRTs totDeff

∑ ν r )P i j

(3)

where the washcoat is represented as a layer of constant average thickness. The boundary conditions are shown in Table 1. The average washcoat thickness was estimated to be 34.8 µm, and the effective diffusivities were estimated by the random pore model.17 The average reaction rates (ravg) in the washcoat surface mass balance (eq 2) were obtained by their integration over the washcoat layer:

rjavg )

∫

twc

0

rj dx

twc

(4)

Similar to the gas phase mass balance of eq 1, axial and radial dispersions were neglected in the gas-phase heat balance: dTg AcSVh(Ts - Tg) ) dz wi mtot c Mi Pi i

∑

(5)

dTs ) εmσ(Ts4 - Tfeed4 ) g dz

dTs )0 dz

r ) 0 (monolith center)

dTs )0 dr

r ) R (monolith outer edge)

dTs (Tw - Ts) ) dr x

The heat balance for the solid materials accounted for axial and radial heat conduction through the cordierite and washcoat as well as heat transport to the gas phase:

( )

(1)

(2)

(1 - ε)λs

z ) L (outlet)

(1 - ε)λs

Axial dispersion in the gas phase was neglected because it was found to be negligible at the high space velocity used. Radial diffusion is of course also neglected because the monolith channel walls segregate the gas flow. The following mass balance was used for the washcoat surface: FB

z ) 0 (inlet)

d2Ts 2

dz

(

+ (1 - ε)λs F B(

∑

d2Ts 2

+

)

1 dTs + r dr

dr rjavg(-∆HRj)) ) SVh(Ts - Tg)

(6)

Table 1 shows the boundary conditions used for the solid heat balance. They account for radiation heat losses from the monolith inlet as well as the heat transport resistance of the outer 1 mm edge of the monolith blocked from reactant flow by the sample holder. The local heat transport resistance in the washcoat layer was neglected as confirmed by solution of the analogous heat balance of eq 3 for some of the reaction conditions. The mass and heat transport correlations in the equations above were estimated from a correlation where the Sherwood and Nusselt numbers decreased along the channel axes and approached an asymptotic value of 2.98.18 The thermal conductivity of the cordierite monolith was set to 0.5 W m-1 K-1.19 The system of algebraic and partial differential equations comprising the 2D model was solved using finite difference methods in gPROMS Modelbuilder 3.1.3. 3.2. 1D Monolithic Reactor Model. In the 1D monolith reactor model, the gas-phase mass balance (eq 1) was expressed in terms of the molar flow rates instead of local component mass fractions: dFi AcSvkcPtot (yis - yig) ) dz RT

(7)

Many of the other mass and heat balances, including eqs 2-5, were the same as for the 2D model. However, the solid material heat balance was simplified to the following form: FB(

∑r

javg(-∆HRj))

) SVh(Ts - Tg) + exp(R(Ts - Tw)) - 1 UA exp(R) - 1

(8)

where the radial heat transport is lumped into a term containing an overall heat transport coefficient (UA) times a driving force. A linear driving force (e.g., Ts - Tw) was of course unsuitable because the radial temperature gradient was nonlinear. The driving force used in eq 8 amplified the linear driving force at

9714

Ind. Eng. Chem. Res., Vol. 49, No. 20, 2010

larger radial temperature gradients. This expression was found to be appropriate to allow the 1D reactor model to reproduce the axial temperature profile predicted by the 2D reactor model over the range of conditions in this study. A comparison of the 1D and 2D reactor model simulations is further discussed below. The system of algebraic and first-order differential equations comprising the 1D model was numerically solved using a variable order finite difference method in the ode15s function in MATLAB 7.1 with a defined mass matrix. The second-order differential eq 3 relating transport and reaction in the catalyst washcoat was solved using a collocation method in the bvp4c function in MATLAB. 3.3. Kinetic Model Parameter Fitting. The kinetic model is presented in section 4.2. An objective of the modeling was to estimate the kinetic parameters based on experimental results. To avoid correlation between the pre-exponential factors and activation energies of the Arrhenius equation, the rate constant (kmj) at an average temperature (Tm) of 300 °C and the activation energy were used as fitting parameters. Thus, for parameter fitting purposes, the Arrhenius equation was reformulated as:

( (

kj ) kmj exp

EAj 1 1 R Tm T

))

(9)

where kj is the rate constant at temperature T. Similarly, the adsorption equilibrium constants (Kj) were expressed as a function of temperature by:

( (

Kj ) Kmj exp

∆Hadsj 1 1 R Tm T

))

(10)

where the adsorption equilibrium constant at the average temperature (Kmj) and the enthalpy of adsorption (∆Hadsj) were used as fitting parameters. Parameter fitting was performed using a gradient search method with the objective to minimize the residual sums of squares for the measured outlet temperature as well as the concentrations of DME, CO, CO2, H2O, H2, CH4, and methanol. The residuals were weighted by either the inverse of the measured values or the average measured values. The function lsqnonlin in MATLAB 7.1 was used with upper and lower bounds for the scaled parameter values. 4. Results and Discussion 4.1. Observations from Experimental Data. The experimental results were reported in a previous publication11 and analyzed with respect to the general trends in DME conversion, CO2 selectivity, and hydrogen yield. In this section, some further details regarding the experimental data, of interest for modeling purposes, are provided. The outlet oxygen concentration was not measured; however, it could be estimated, although with poor precision, by mass balances. It was found that for experiments at the lowest oxygen feed concentration (O2:DME ) 0.4), the oxygen conversion was at or nearly 100%. However, it was clear that at higher oxygen feed concentrations the oxygen conversion was not complete and may have been as low as 40% under some conditions. For all of the experimental conditions, the oxygen feed concentrations were sufficiently high so that the overall process would be exothermic upon complete conversion of oxygen and DME. The outlet gas temperatures were measured, and they were in most cases greater than the feed temperature. The average temperature increase over the reactor was 12 °C, but it was as high as 30-35 °C under some conditions. Overall heat

balances indicated that the reactor could not be regarded as operating adiabatically. The calculated heat losses ranged from 1 to 17 W. Both methane and methanol were minor products. Their outlet concentrations ranged from 450 to 1300 ppm for methane and from 20 to 1400 ppm for methanol. The methanol yield, although small, tended to be higher at reaction conditions with a lower conversion of DME. These observations support the belief that methanol is an intermediate product, first produced by the hydrolysis of DME and then subsequently consumed by steam reforming. No clear correlation between the methane yield and the experimental conditions could be observed. Experiments were also carried out with the cordierite monolith free of the catalyst washcoat at O2:DME ) 0.4 and H2O:DME ) 2.5 over a temperature range of 240-400 °C. Under these conditions, no conversion of reactants was detected, suggesting that gas-phase initiated reactions are of little importance. 4.2. Kinetic Model. It has been discussed that at least eight overall reactions could conceivably play a role during the combined partial oxidation and steam reforming of DME.12 Including all possible reactions in a global kinetic model is often not practical unless one has a priori knowledge of estimated rates of at least some reactions. It is also inevitable that only a limited number of phenomena can be elucidated from a set of experimental data, and as a result kinetic parameters for minor or competing reactions will be poorly estimated and have low statistical significance. The approach adopted with this modeling effort was to limit the kinetic model to a few key overall reactions that could reasonably capture the experimental observations and yield well-defined kinetic parameters. Table 2 shows the kinetic model that was found to be appropriate for the experimental data. Oxygen reacts only via the total oxidation of DME. The partial oxidation of DME was excluded because it has been recently observed from various spatially resolved measurement techniques that partial oxidation, although of methane, in fact occurs initially via total combustion that produces water allowing reforming reactions to follow.20,21 The difference here with autothermal steam reforming is that it is possible for the oxidation and steam reforming reactions to occur from the start simultaneously and compete for consumption of DME. Also, if total oxidation, steam reforming, and the water-gas shift (WGS) occur simultaneously, they can of course produce partial oxidation reaction products. The reforming reactions are initiated by the hydrolysis of DME, forming methanol. The experimental methanol yield tended to be higher at lower conversion of DME, suggesting that methanol is an intermediate product. Because methanol is only present in low concentrations, methanol oxidation reactions were excluded because they should only have a minor contribution as compared to DME oxidation. The reforming of methanol proceeds first by its decomposition followed by the water-gas shift (WGS) reaction. The steam reforming of methanol could have been expressed in a single reaction producing CO2, but that would be equivalent to the sum of methanol decomposition and WGS. Also, the use of these two reactions (methanol decomposition and WGS) was found to better describe the CO/CO2 product mole ratios. Table 2 also shows the Langmuir-Hinshelwood-type rate expressions for each of the reactions. It was assumed that all reactions occur on a single site, probably Pd atoms but perhaps also in contact with Zn. Certainly the hydrolysis of DME may occur on acid sites because it has been shown that DME hydrolysis and steam reforming activity are related to acidity of solid acid catalysts.9,10 However, no direct relation between

Ind. Eng. Chem. Res., Vol. 49, No. 20, 2010

9715

Table 2. Kinetic Model Overall Reactions and Rate Expression reaction

formula

rate expression

CH3OCH3 + 3O2 f 2CO2 + 3H2O

DME total oxidation

CH3OCH3 + H2O f 2CH3OH

DME hydrolysis

methanol decomposition

CH3OH f CO + 2H2

water-gas shift

CO + H2O T CO2 + H2

CH3OCH3 f H2 + CO + CH4

DME decomposition

r1 )

r2 )

r3 )

r4 )

r5 )

k1yDMEyO2 (1 + KH2OyH2O)2 k2yDMEyH2O (1 + KH2OyH2O)2 k3yCH3OH (1 + KH2OyH2O) k4yCOyH2O (1 + KH2OyH2O)

2

(

1-

yCO2yH2 yCOyH2OKeq

)

k5yDME (1 + KH2OyH2O)

Table 3. Fitted Kinetic Parameter Values and 95% Confidence Intervals reaction/adsorbed species parameter DME total oxidation DME hydrolysis methanol decomposition water-gas shift DME decomposition H2O adsorption

km1 EA1 km2 EA2 km3 EA3 km4 EA4 km5 EA5 Km ∆Hads

value

confidence interval

53.07 103.7 59.00 173.9 133.4 107.3 354.1 178.8 9.980 × 10-2 114.8 3.871 -31.71

(6.47 (4.5 (13.82 (23.8 (29.75 (44.5 (180.4 (29.3 (2.764 × 10-2 (30.2 1.113 0.74

units mol kg-1 kJ mol-1 mol kg-1 kJ mol-1 mol kg-1 kJ mol-1 mol kg-1 kJ mol-1 mol kg-1 kJ mol-1

s-1 s-1 s-1 s-1 s-1

kJ mol-1

the activity of Pd-Zn-based catalysts for autothermal DME reforming and their acid content has been observed.12 The adsorption of various species was included in the rate expressions; however, only water adsorption was found to be of any importance to describe the experimental data at hand. 4.3. Modeling Results. Table 3 shows the fitted kinetic parameter values as well as the 95% confidence intervals. In all cases, the confidence intervals are smaller than the actual parameter values. Figure 1 compares the experimental and model predicted outlet DME and major product compositions for a series of experiments with varying feed oxygen concentration and feed temperature. With higher feed temperature and oxygen feed concentration, the conversion of DME and outlet concentrations of major products (CO2, CO, H2, and H2O) increases, both according to the experiments and model predictions. It is expected that a higher feed concentration of oxygen promotes the DME oxidation reaction, which in turn generates more heat that sustains the DME hydrolysis and methanol decomposition reactions. Promotion of DME hydrolysis and the water-gas shift reactions with increased oxygen feed is also indicated by the drop in the outlet water concentration. This behavior was confirmed by the model and can be illustrated by the model predictions of the axial temperature profiles in the monolith. Figure 2A shows the 2D model mixed-cup gas-phase temperature profiles for a range of O2:DME feed ratios. Clearly, an increased O2:DME feed ratio caused a higher and more distinct temperature peak to be reached closer to the monolith inlet. According to the model of course, the DME oxidation reaction dominates near the monolith inlet, causing initially the temperature to increase, which activated the endothermic DME hydrolysis and methanol decomposition reactions that began to

Figure 1. Outlet concentrations of DME, H2O, and major products as a function of O2:DME molar feed ratio and feed temperature. Symbols are experimental results, and curves are simulated.

9716

Ind. Eng. Chem. Res., Vol. 49, No. 20, 2010

Figure 2. Temperature profiles in reactor: (A) Axial temperature profile predicted by 2D and 1D models as a function of the O2:DME molar feed ratio and compared to experimental measurements of outlet temperature; (B) 2D temperature profile for O2:DME molar feed of 0.9. For all conditions, the feed temperature is 400 °C, and the H2O:DME molar feed ratio is 2.5.

increasingly dominate, thus causing the temperature to later decrease. For the model to predict these features, it was necessary for the activation energies of particularly the DME hydrolysis reaction to be large as compared to that of the DME oxidation reaction, as is the case in Table 3. It can also be seen in Figure 2A that the predicted outlet temperatures agreed rather well with the measured values. It is also shown in Figure 1 that both the experiment and the model predicted CO2 selectivities increase with the oxygen feed concentration. This is indicated by the relatively greater increase in the CO2 outlet concentrations as compared to the CO outlet concentrations. In Figure 1, the model outlet CO concentrations are compared to the WGS reaction equilibrium CO concentrations at the model outlet conditions. It is evident that at higher feed temperature and higher feed oxygen concentration, the outlet CO concentration is increasingly close to the equilibrium values. Apparently, also the WGS reaction is activated by the higher peak temperatures reached with high oxygen feed concentrations. This explains the rather larger value of the activation energy of the WGS reaction in Table 3. Also, the rather large confidence interval for WGS reaction rate constant (km4) is due to the fact that for parts of the experimental results, the WGS reaction is at or close to equilibrium. Certainly the model predictions in Figure 1 capture the main trends of the experimental results, but they only moderately well reproduce the experimental measurements. Various modifications in the kinetic model were tested to improve the model fit.

Figure 3. Outlet concentrations of DME, H2O, and major products as a function of H2O:DME molar feed ratio and feed temperature. Symbols are experimental results, and curves are simulated.

For example, the model tends to poorly predict the low DME conversion observed particularly at the lowest O2:DME feed ratio. An oxygen adsorption term, like that for water, was included in the model, but it was found to be insignificant as indicated by a low value for the fitted adsorption equilibrium constant and a large confidence interval. Instead, it was found that using a power slightly greater than unity on the oxygen concentration in the DME oxidation rate expression could improve the fit of the predicted DME conversion over the whole range of oxygen feed concentrations. However, this was found to cause large deviations between the predicted and measured outlet temperatures. The oxygen conversions and outlet temperatures became low particularly at low oxygen feed concentrations. As a result, the best compromise to satisfy both the material and the heat balances was to maintain a value of unity for the power on the oxygen concentration in the DME oxidation rate expression. Figure 3 compares the experimental and model predicted outlet DME and major product compositions for a series of experiments with varying feedwater concentrations and feed temperature. There was a slightly lower conversion of DME at

Ind. Eng. Chem. Res., Vol. 49, No. 20, 2010

Figure 4. Parity plots comparing the simulated and predicted outlet concentrations of the minor products: (A) methanol; (B) methane.

higher feed concentrations of water, which is particularly evident for the experiments with the lowest feed temperature (350 °C). The water adsorption term in the denominator of the rate expressions (see Table 2) was necessary to allow the model to also predict lower conversions with higher feedwater concentration. A DME adsorption term was also included but was found to be insignificant; however, the design of the experiments did not allow for a precise assessment of the influence of DME concentration on the reaction rates. Also in Figure 3 it can be seen that both the experiments and the model indicate a higher CO2 selectivity at higher feedwater concentration. The model predicts this behavior because the higher water concentration promotes the WGS reaction. In addition, although it is not apparent in Figure 3, the predicted temperatures in the reactor were lower with higher water feed concentration, and this also shifts the equilibrium of the WGS reaction to favor higher CO2 selectivity. From the experiments, it was evident the outlet CO2 concentrations first increased with the water concentration, but then decreased at the highest water feed concentration. The eventual decrease in the CO2 concentration could only be reproduced by the model at the lower feed temperatures. It results, according to the model, due to the lower conversion of DME at higher water feed concentration and particularly at lower feed temperatures. It should also be noted from rate expressions in Table 2 that even a higher water concentration is detrimental to the rate of the WGS reaction. Figure 4 shows the parity plots for the outlet concentrations of the minor products methanol and methane for all the experiments. For methanol, it can be seen that there is some reasonable agreement between model predictions and experiments. Because methanol is an intermediate product in the kinetic model, its outlet concentration was predicted to be higher at conditions with a lower conversion of DME, similar to what

9717

was observed in the experiments. Also, because methanol was only observed in low concentrations, the rate of the methanol decomposition reaction in the kinetic model is very close to that of DME hydrolysis. On the other hand, the agreement between model and experiments for the outlet methane concentrations in Figure 4 is generally poor. It was found that the DME decomposition reaction as used in the model (see Table 2) was slightly superior to predict the methane concentration than an alternative model with the CO methanation reaction (CO + 3H2 f CH4 + H2O). From these experiments, it can thus not be clearly determined which of these reactions may be most responsible for methane formation. Because methane formation is a minor byproduct, the rate of the methane formation reaction has at least negligible influence on the rates or values of fitted parameters for the other reactions producing the major products. In separate experiments, it was found that there was no detectable conversion of reactants for the maximum feed temperature (400 °C) in the absence of the catalyst material. However, due to the oxidation reactions, the actual temperature in the reactor can be as much as 70 °C higher under some conditions. Thus, it is also possible that there is some initiation of homogeneous gas-phase reactions at these local peak temperatures, which are not accounted for by the model and may explain some of the discrepancies in methane and methanol predictions. 4.4. Comparison of Rigorous and Lumped Models for Heat Transport. Figure 2 shows examples of the axial temperature profiles in the monolith at various conditions as well as the 2D temperature profile for one set of conditions. In Figure 2A are shown the mixed-cup temperatures predicted by the 2D model. From Figure 2B it can be seen that the local temperatures at the center of the monolith were highest, whereas the temperatures at the outer edge of the monolith were closer to the reactor wall temperature, which for all experiments was the feed temperature (400 °C in this case). For example, for the O2:DME feed ratio of 0.9, the maximum local temperature was 451 °C at the center of the monolith and 6 mm from the inlet. Also, in Figure 2A the 2D mixed-cup temperature is compared to the axial temperature profile for the 1D model for one of the experiments. The 1D model, being more robust and less computationally demanding, was used for kinetic model parameter fitting. Note that the radial heat transport term in the 1D model was tuned to reproduce the 2D mixed-cup temperature. The example of deviations between the 1D and 2D model temperatures shown in Figure 2A was typical for all experiments and was found to be never more than (5 °C. These local temperature deviations were found not to cause deviations in the predicted outlet concentrations. The exclusion of axial heat transport from the 1D model was justifiable because it was found that a version of the 1D model including axial heat transport (a variation of solid heat balances, i.e., eqs 6 and 8) predicted an axial temperature profile essentially identical to the 1D model that excluded axial heat transport. 4.5. Reactor Scale-Up Considerations. The reactor and kinetic model can be used to estimate the behavior of a reactor with a scale and operating conditions similar to that in a vehicle APU system. In the experimental studies used to develop the kinetic model, the feed concentration of nitrogen was always higher than the case if air was used to supply oxygen. From a simulation of the experimental reactor operating with air feed, a feed temperature of 400 °C, an O2:DME ratio of 0.7, and a H2O:DME ratio of 2.5, it was found that the predicted

9718

Ind. Eng. Chem. Res., Vol. 49, No. 20, 2010

K-1) could be about 40 times greater than that for the cordierite monolith (0.5 W m-1 K-1). Figure 5 shows the predicted temperature profile for the 5 kW scale reactor operating with a metallic monolith. The peak temperature is significantly reduced as compared to that for the monolith reactor, mainly due to the improved radial transport of heat out of the reactor. The DME conversion for the metallic monolith reactor is reduced to 96.5%, due to the lower temperatures in the reactor. The advantages of the metallic monolith are not immediately evident from the simulations carried out here because heat transported out of the reactor is lost. However, in an actual APU system, if the reactor is designed as a heat exchanger and used to preheat the reactor feed, the improved heat exchange would serve to even out operating temperatures and reduce hot spots in the reactor system. 5. Conclusions Figure 5. Axial temperature profile in various scales of reactor with air feed as predicted by 2D model. For all simulations, the feed conditions are 400 °C, O2:DME ) 0.7, H2O:DME ) 2.5. (a) Experimental scaled reactor, (b) reactor scaled to produce 5 kW electricity, (c) reactor scaled to produce 5 kW electricity operating adiabatically, and (d) reactor scaled to produce 5 kW electricity but with metallic catalyst support.

DME conversion was 98.7% and the outlet concentration of H2 was 43.6 mol %. Under these conditions, the gas hourly space velocity is reduced to 6900 h-1. The axial temperature profile for the reactor operated under these conditions is shown in Figure 5. With air feed, the DME conversion is increased from the experimental result of 91.3%, due to the lower space velocity. Also, by comparing the corresponding temperature profiles in Figures 5 and 2, it can be see that the predicted peak temperature in the reactor with air feed is only slightly higher. Although the reactant concentrations are higher and space velocity lower with air feed, the rates of both the exothermic and the endothermic reactions, which always occur to varying degrees simultaneously, are higher, resulting in little change in the temperature profile. On the basis of the yield of hydrogen from the experimental reactor with air feed, it was estimated that the reactor size would have to be scaled up by a factor of 43 to produce sufficient hydrogen to power a 5 kW fuel cell.22 Simulations were carried out to examine the performance of a 5 kW scale reactor with the same feed conditions and the length to diameter ratio of the experimental monolith reactor. The wall temperature of the reactor was set to the feed temperature of 400 °C, as in the experimental study; however, in an APU system, this temperature may be lower because the reactor could be designed as a heat exchanger to use excess heat from the autothermal reforming to preheat the feed. The predicted axial temperature profile for the 5 kW scale reactor is shown in Figure 5, and the DME conversion was 99.5%. The DME conversion is even higher with this scaled-up reactor because the operating temperature is higher (see Figure 5). The 5 kW scale reactor in fact operates nearly adiabatically as can be seen by comparing its temperature profile to that for the identical adiabatic reactor in Figure 5. This is due to the fact that the external area to volume ratio of the monolith reactor is inversely proportional to its diameter. Measures could be considered to improve the heat transport properties of the monolith reactor, such as using a metallic catalyst support instead of cordierite as used in the experimental study. For an FeCrAlloy support, it is estimated that the effective thermal conductivity of the washcoated monolith (ca. 20 W m-1

A kinetic model consisting of overall reactions for DME total oxidation, DME hydrolysis, methanol decomposition, and the water-gas shift reaction was found to describe the key features of experimental data for DME autothermal reforming. No single overall reaction, neither DME decomposition nor methanation reactions, could satisfactorily describe the low formation of methane byproduct observed in the experiments. The kinetic rate equations included a water adsorption term, because higher water concentration appeared to reduce DME conversion. Because DME autothermal reforming couples both exothermic and endothermic reactions, special attention had to be paid to properly describing heat transport in the monolith reactor. The most rigorous model used here was a 2D model that included heat transport between gas and solid phases as well as axial and radial heat transport in the solid. However, for the development of the kinetic model, a reduced 1D model was used that neglected axial heat transport and included a lumped term to express the radial heat transport. The models predicted the temperature profile in the reactor as a function of operating conditions. Generally, it was found that the oxidation reactions dominated close to the reactor inlet, creating a temperature peak. The peak temperature reached or amount of heat generated by the oxidation reactions, which is controlled by the feed temperature and feed concentration of oxygen, determined the extent of activation of the following endothermic reactions including DME hydrolysis and methanol decomposition. The WGS reaction was found to be close to equilibrium for conditions at higher operating temperature. On the basis of the model, the performance of the monolith reactor was examined if it were operated with air feed to supply oxygen and scaled-up to produce sufficient hydrogen to power a 5 kW fuel cell. It was found that at operating conditions giving a high conversion of DME (>95%), the outlet H2 concentration can be greater than 40 mol %. According to the model, the reactor can be operated with varying total feed concentrations of the reactants with only small changes in the temperature profile; however, the reactor scale strongly influences the heat transport behavior of the reactor and as a result the temperature profile. A catalyst support material with higher thermal conductivity may be advantageously used in a larger scale reactor to improve the heat transport behavior and moderate operating temperatures in an APU system. Acknowledgment Financial support from the Foundation for Strategic Research (SSF) and the Foundation for Strategic Environmental Research (Mistra) is gratefully acknowledged.

Ind. Eng. Chem. Res., Vol. 49, No. 20, 2010

Nomenclature Ac ) monolith cross-sectional area (m ) cP ) heat capacity (J mol-1 K-1) Deff ) effective diffusivity in washcoat (m2 s-1) EA ) activation energy (J mol-1) F ) molar flow rate (mol s-1) h ) heat transport coefficient (W m-2 K-1) ∆HR ) enthalpy of reaction (J mol-1) k ) reaction rate constant (mol kg-1 s-1) kc ) mass transport coefficient (m s-1) K ) adsorption equilibrium constant L ) total monolith length (m) mtot ) total mass flow rate (kg s-1) M ) molecular weight (kg mol-1) Ptot ) total pressure (Pa) r ) reaction rate (mol kg-1 s-1) r ) monolith radial distance (m) R ) molar gas constant (8.314 Pa m3 mol-1 K-1) R ) total monolith radius (m) Sv ) washcoat surface area per monolith volume (m2 m-3) twc ) washcoat total thickness (m) T ) temperature (K) UA ) overall heat transport coefficient for radial heat transport (W K-1) w ) mass fraction x ) washcoat axial distance (m) y ) mole fraction z ) monolith axial distance (m) R ) radial heat transport amplification factor ε ) monolith void fraction εm ) emissivity σ ) Stefan-Boltzmann constant (5.67 × 10-8 W m-2 K-4) λs ) monolith effective thermal conductivity (W m-1 K-1) FB ) washcoat mass per monolith volume (kg m-3) Fc ) washcoat density (kg m-3) ν ) species stoichiometric coefficient 2

Subscripts avg ) washcoat volume average (for reaction rates) c ) washcoat interior g ) gas phase i ) species index j ) reaction index m ) property at mean temperature 400 °C s ) catalyst washcoat surface w ) reactor wall

Literature Cited (1) Salsing, H.; Denbratt, I. Performance of a Heavy Duty DME Engine - An Experimental Study. SAE International Technical Paper, 2007-014167, 2007. (2) Goto, S.; Oguma, M.; Suzuki, S. Research and Development of a Medium Duty DME Truck. SAE International Technical Paper, 2005-012194, 2005.

9719

(3) Edwards, R.; Larive´, J.-F.; Mahieu, V.; Rouveirolles, P. Well-toWheels Report, Version 2c, March 2007, 2007; http://ies.jrc.ec.europa.eu/ WTW. (4) Mengel, C.; Konrad, M.; Wruck, R.; Lucka, K.; Ko¨hne, H. Diesel Steam Reforming of PEM Fuel Cells. J. Fuel Cell Sci. Technol. 2008, 5, 021005-1. (5) Kaila, R. K.; Gutie´rrez, A.; Krause, O. I. Autothermal Reforming of Simulated and Commercial Diesel: The Performance of ZirconiaSupported RhPt Catalyst in the Presence of Sulfur. Appl. Catal., B 2008, 84, 324. (6) Thormann, J.; Pfeifer, P.; Kunz, U.; Schubert, K. Reforming of Diesel Fuel in a Micro Reactor. Int. J. Chem. React. Eng. 2008, 6, P1. (7) Alvarez-Galvan, M. C.; Navarro, R. M.; Rosa, F.; Bricenˇo, Y.; Ridao, M. A.; Fierro, J. L. G. Hydrogen Production for Fuel Cell by Oxidative Reforming of diesel Surrogate: Influence of Ceria and/or Lanthana over the Activity of Pt/Al2O3 Catalysts. Fuel 2008, 87, 2502. (8) Karatzas, X.; Nilsson, M.; Dawoody, J.; Lindstro¨m, B.; Pettersson, L. J. Characterization and Optimization of an Autothermal Diesel and Jet Fuel Reformer for 5 kW Mobile Fuel Cell Applications. Chem. Eng. J. 2010, 156, 366. (9) Faungnawakij, K.; Kikuchi, R.; Matsui, T.; Fukunaga, T.; Eguchi, K. A Comparative Study of Solid Acids in Hydrolysis and Steam Reforming of Dimethyl Ether. Appl. Catal., A 2007, 333, 114. (10) Fukunaga, T.; Ryumon, N.; Shimazu, S. The Influence of Metals and Acidic Oxide Species on the Steam Reforming of Dimethyl Ether (DME). Appl. Catal., A 2008, 348, 193. (11) Nilsson, M.; Jozsa, P.; Pettersson, L. J. Evaluation of Pd-based Catalysts and the Influence of Operating Conditions for Autothermal Reforming of Dimethyl Ether. Appl. Catal., B 2007, 76, 42. (12) Nilsson, M.; Jansson, K.; Jozsa, P.; Pettersson, L. J. Catalytic Properties of Pd supported on ZnO/ZnAl2O4/Al2O3 Mixtures in Dimethyl Ether Autothermal Reforming. Appl. Catal., B 2009, 86, 18. (13) Zhang, Q.; Du, F.; He, X.; Liu, Z.-T.; Liu, Z.-W.; Zhou, Y. Hydrogen Production via Partial Oxidation and Reforming of Dimethyl Ether. Catal. Today 2009, 146, 50. (14) Holladay, J. D.; Hu, J.; King, D. L.; Wang, Y. An Overview of Hydrogen Production Technologies. Catal. Today 2009, 139, 244. (15) Feng, D.; Wang, Y.; Wang, D.; Wang, J. Steam Reforming of Dimethyl Ether over CuO-ZnO-Al2O3-ZrO2 + ZSM-5: A Kinetic Study. Chem. Eng. J. 2009, 146, 485. (16) Groppi, G.; Tronconi, E. Simulation of Structured Catalytic Reactors with Enhanced Thermal Conductivity for Selective Oxidation Reactions. Catal. Today 2001, 69, 63. (17) Froment, G. F.; Bischoff, K. B. Chemical Reactor Analysis and Design, 2nd ed.; John Wiley & Sons: New York, 1990. (18) Hawthorn, R. D. Afterburner Catalysts - Effects of Heat and Mass Transfer between Gas and Catalyst Surface. AIChE Symp. Ser. 1974, 70, 428. (19) Garcia, E.; Osendi, M. I.; Miranzo, P. Thermal Diffusivity of Porous Cordierite Ceramic Burners. J. Appl. Phys. 2002, 92, 2346. (20) Horn, R.; Williams, K. A.; Degenstein, N. J.; Schmidt, L. D. Syngas by Catalytic Partial Oxidation of Methane on Rhodium: Mechanistic Conclusions from Spatially Resolved Measurements and Numerical Simulations. J. Catal. 2006, 242, 92. (21) Kimmerle, B.; Haider, P.; Grunwaldt, J.-D.; Baiker, A.; Boye, P.; Schroer, C. G. High Throughput Cell for X-ray Adsorption Spectroscopy Applied to Study the Effect of Au on Rh-catalyzed Partial Oxidation of Methane. Appl. Catal., A 2009, 353, 36. (22) Larminie, J.; Dicks, A. Fuel Cell Systems Explained, 2nd ed.; John Wiley & Sons: New York, 2003.

ReceiVed for reView April 7, 2010 ReVised manuscript receiVed August 23, 2010 Accepted September 15, 2010 IE100834V