On the use of mechanistic CO oxidation models with a platinum monolith catalyst

Applied Catalysis B: Environmental 70 (2007) 305–313 www.elsevier.com/locate/apcatb

On the use of mechanistic CO oxidation models with a platinum monolith catalyst S. Salomons a,b, R.E. Hayes b,*, M. Votsmeier c, A. Drochner a, H. Vogel a, S. Malmberg c, J. Gieshoff c a

Technische Universita¨t Darmstadt, Ernst-Berl Institut fu¨r Technische und Makromolekulare Chemie, Petersenstrasse 20, D-64287 Darmstadt, Germany b Department of Chemical and Materials Engineering, University of Alberta, Edmonton, Alberta, Canada T6G 2G6 c Umicore AG & Co. KG, Automotive Catalysis Division, Research and Development, Hanau, Germany Available online 27 October 2006

Abstract This paper presents experiments and model predictions for the oxidation of CO over a platinum catalyst in a monolith reactor. Experimental behavior is broadly consistent with previously reported work on CO oxidation. Ignition–extinction (light-off) curves demonstrated the presence of multiple steady states with a hysteresis effect. The two branches corresponding to the two states of predominantly CO covered or oxygen covered. Admission of CO pulses to an oxygen covered surface results in reaction, indicating the occurrence of adsorption of CO on an oxygen covered surface. A model based on adsorption and surface reaction using the classical Langmuir–Hinshelwood (LH) approach qualitatively was able to reproduce the light-off behavior. The best model assumes dissociative chemisorption of oxygen on two surface sites. It was superior to a model proposing molecular adsorption of oxygen followed by rapid dissociation. Addition of steps allowing CO adsorption on an oxygen filled surface via an ‘‘oxygen compression’’ mechanism enable the qualitative description of the reactor response to step inputs of CO to an oxygen rich feed stream. Some parameter adjustment remains to allow a better fit between experiment and model predictions. # 2006 Elsevier B.V. All rights reserved. Keywords: Light-off; Ignition; Catalytic converter; Kinetic modeling; CO oxidation; Platinum; Monolith

1. Introduction Environmental regulations governing the engine exhaust emission composition from automobiles have resulted in the widespread use of catalytic converters. These include the three way catalyst (TWC) for gasoline (petrol) engines, which oxidizes hydrocarbons and carbon monoxide, and reduces nitrogen oxides. For diesel engines, an oxidizing converter is used. In both cases, the converter is usually a monolith honeycomb, with a thin catalytic washcoat on the walls [1–3]. The design of converter systems can be greatly enhanced by using computer-aided design; however, this methodology requires realistic mathematical models that can be executed in a reasonable amount of time. Such simulators must have

* Corresponding author. E-mail address: [email protected] (R.E. Hayes). 0926-3373/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.apcatb.2006.01.022

realistic kinetic models for the reactions. The chemistry of the multiple oxidation-reduction reactions is very complex, and modeling is complicated by the fact that the converter operates in a transient mode. Most kinetic models used for catalytic converters use only gas phase species concentrations. Such models may be either empirical or mechanistic, but in neither case do the models account for the elementary steps in the reaction mechanism. These models are usually referred to as global models. This approach is often used because the underlying mechanism is not known, or it is not possible to determine the kinetic parameters in each individual step. In catalytic reactions, a classical approach for developing rate expressions is the Langmuir–Hinshelwood–Hougen–Watson method (LHHW), and most converter simulation uses this type of kinetic model. Indeed, the majority of converter simulations reported use a model based on the LHHW type proposed by Voltz et al. [5], who studied supported platinum catalysts [6–9]. Although

306

S. Salomons et al. / Applied Catalysis B: Environmental 70 (2007) 305–313

some success at predicting converter performance has been reported with such models, especially for light-off over a narrow range of operating conditions, LHHW expressions are best suited for steady state operation, and are not well suited for transients. Another drawback is that they do not capture the change in the adsorbed species on the surface with time, which is significant at some conditions. Such schemes are also less suited for extrapolation and engineering design, owing to the limited range of conditions over which they are typically valid. The catalytic reactions depend on the interaction between the active sites and the reactants, which manifest themselves as a series of chemisorption–desorption and surface reaction steps. Recent research work has emphasized the use of transient reaction kinetics, and as a result some reaction mechanisms for the catalyzed reactions that occur in three way converters have been proposed [4,10–17]. These types of models can be called elementary or pseudo-elementary, because they attempt to incorporate the molecular steps in the reaction. These models still retain an element of averaging, since they usually do not distinguish between all of the different steps that can occur on, for example, different crystal faces of the catalyst. It has been widely discussed in literature [18] that the adsorption and reaction behavior is influenced by the crystal face. Although to model the reaction at this level in a catalytic converter would still prove daunting, it should perhaps go without saying that a mechanistic, or even global, model should be as consistent as possible with the evidence supplied by the fundamental surface science studies, especially if the objective is to build a design simulator that is to be extrapolated. Mechanistic models for the catalytic converter reactions contain many parameters, and to fit all of them simultaneously for real exhaust gases is a challenging optimization problem. We propose a methodology to build stepwise for each reaction and then to determine, if the results can be combined without further major modification. In this paper, we examine mechanistic models for the oxidation of CO and compare them against experimental transient data obtained over a platinum oxidation catalyst. Although CO oxidation has been widely studied, it is a necessary first step in the development of a complete scheme. Further, the majority of work previously performed on CO oxidation has been done at low pressure; often under high-vacuum conditions. Studies at atmospheric pressure are fewer. There are also few attempts at using mechanistic models for CO oxidation in catalytic converter models. One major effort [12] used primarily real exhaust mixtures and simultaneously fit all of the reaction parameters. Another [4,10,11,13–17] has used transient analysis on powders with transient cycling. These works probably represent the current state of the art of CO oxidation modeling in the context of automotive catalytic converters. In this work, we used a different pattern of unsteady experiments with a monolith reactor. To date, there are no literature publications that can explain satisfactorily the oxidation of CO under the typical conditions of a catalytic converter. The contribution of this paper is to examine critically the state of the art in the

converter literature, and suggest possible improvements. We also note that we are interested in the CO oxidation under lean light-off conditions. The behavior over a three way catalyst might be quite different, owing to the presence of, for example, the oxygen storage components. 2. Experimental An apparatus was constructed to allow for transient experiments on a monolith catalyst using a widely variable combination of feed gases. The apparatus consisted of three main parts: the gas mixing unit, the gas pre-heater and reactor, and the analyzers. The gas mixing unit was designed to control the input concentrations of all gases fed to the reactor. Gas mixtures were prepared from gas bottles of pure gases (Messer Griesheim GmbH). Nitrogen was used as a diluent (99.9999% pure), and was provided from a large tank of liquid nitrogen, supplied by Linde AG. Mass flow controllers were used to control and monitor the flow rate of each gas entering the reactor, and were supplied by either Bronkhorst High-Tech B.V. or Brooks Instrument B.V. (Emerson Process Management). The catalyst was platinum contained within a washcoat coated onto a ceramic 400 CPSI monolith obtained from Corning Inc. The reactor was 2.54 cm in diameter and 7.62 cm long. Axial temperature profiles in the reactor were measured using five K type thermocouples inserted into the monolith channels. Three were placed along the centerline and two more near the wall. Additional thermocouples were placed immediately before and after the monolith to measure inlet and outlet temperatures, respectively. A pressure sensor (Wagner Mess-und Regeltechnik GmbH) was used to measure the pressure before the catalyst, which was typically between 1.07 and 1.10 bar. Static mixers (provided by Herbert Ott Vertriebsgesellschaft MbH. + Co. KG) were included in the oven, to enhance the mixing of the gas and to ensure that the gas composition and temperature was homogeneous. Gases were pre-heated to the desired inlet temperature using an oven. The reactor effluent was fed to the analysis system, where CO and CO2 concentrations were measured using IR spectroscopy, and O2 concentration was measured using a paramagnetic method (NGA 2000 MLT, supplied by Emerson Process Management & Co. OHG). A computer control and data logging system was built using National Instruments LabVIEW 7.0. Electronic input/output devices were supplied by Elektro Beckhoff GmbH. Communication between the controlling computer, the Beckhoff device, and with the oven regulator, was via a PROFIBUS connection. Data analysis was performed with MATLAB 7.0. 2.1. Reactor model The experiments were simulated using a mathematical model of a single monolith channel. This model assumes that the reactor is adiabatic and that the flow distribution is uniform; thus, a single channel is representative of the entire monolith reactor. The complete description of this model can be found in literature [19]. The model solves the governing partial

S. Salomons et al. / Applied Catalysis B: Environmental 70 (2007) 305–313

differential equations for the conservation of mass and energy using a finite volume technique, in which the system of discretized equations was solved using a Newton–Krylov method. The model is a discrete model, that is, it models the fluid and solid phases separately, with coupling between the two phases occurring at the gas–solid interface. 2.2. Models for CO oxidation The oxidation of CO has been studied for many years, and as a result there are many publications in the open literature, see for example the reviews in [18,20,21]. Although an apparently simple reaction, it offers many challenges in understanding, and new publications regularly appear. In spite of its apparent simplicity, simple global rate expressions are not well suited to modeling the reaction over a wide range of operating conditions. Certainly, the reaction phenomena observed provide a rich mine of bi-stable and oscillatory behavior for academic studies. Platinum is arguably the most used catalyst, with palladium also being common, certainly in the works pertaining to automotive pollution control. Both Eley–Rideal (with CO reacting with adsorbed oxygen) and Langmuir– Hinshelwood mechanisms have been used, with early work favoring the Eley–Rideal mechanism. Currently the LH type is favored. We mention here for completeness the main observations that are currently accepted in literature in regard to the reaction mechanism. CO adsorbs linearly through the carbon atom. The adsorption phenomenon changes in the region of 50% coverage. Oxygen adsorbs and dissociates in a largely irreversible manner. No molecularly adsorbed oxygen exists at temperatures present in catalytic converters. No significant associative desorption of oxygen is detected below about 700 K. Adsorbed oxygen can exist both as chemisorbed oxygen and as oxides. The reaction on the surface occurs between adsorbed oxygen atoms and molecularly adsorbed CO. This mechanism can be summarized by the following steps: COðgÞ þ uV Ð uCO

(1)

O2ðgÞ þ 2uV ! 2uO

(2)

uO þ uCO ! CO2ðgÞ þ 2uV

(3)

uO þ COðgÞ ! CO2ðgÞ þ uV

(4)

Steps (1)–(3) represent an LH mechanism, and Step (4) is an Eley–Rideal step. As a basic route for the reaction, this scheme has been validated using many techniques [22–31]. Lynch et al. [32–36] showed that this basic mechanism, when modeled as a series of elementary steps, could, with some modification, account for many of the observed phenomena in CO oxidation. It has been pointed out in many works that this simple mechanism is insufficient to explain all of the complex behavior observed in CO oxidation, for example oscillations. Much of this discrepancy has been revealed by transient testing. For example, oscillatory behavior has been shown by many investigators. Oscillatory behavior has been shown to depend

307

on the crystal face of platinum and it has been suggested to be caused by different factors at different partial pressures of CO. The most widely used model in literature to explain oscillations is the oxidation-reduction model involving subsurface oxygen [37]. The Eley–Rideal step has been controversial in literature, and although an early favorite, current thinking is to discount it. Surface science [18] and modeling studies [33] have suggested that this step is not plausible. However, it is well known that admission of CO to an oxygen covered surface will result in reaction. In the preceding mechanism, with irreversible adsorption of oxygen, there is no route for reaction in this case. Indeed, according to the model, without the Eley–Rideal step, exposure of a surface to both CO and oxygen with no reaction will, over time, result in a completely oxygen covered surface. Thus, all sites will be oxygen filled, and therefore it is impossible for any CO to adsorb. It has been shown [38] that admission of CO to an oxygen covered surface causes a fast formation of CO2, whereas oxygen introduced to a CO covered surface does not. In [38], the investigators used a combination of Eley–Rideal and LH models. Nievergeld [6] also showed that CO adsorption is not inhibited by oxygen coverage, but that oxygen adsorption is inhibited by adsorbed CO. Other investigators used surface exclusion models to account for the adsorption of CO on an oxygen covered surface [31–33]. In these models, oxygen coverage is limited to less than a monolayer, however, CO is still able to adsorb on the unfilled sites. Nievergeld, however, [4] reported that both CO and oxygen coverage can become unity. They proposed adding the Eley–Rideal mechanism to account for the observed phenomena, and cited work by Barshad et al. [39] who suggested that CO can chemisorb on sites already covered by oxygen. 2.3. Modeling of CO oxidation in catalytic converters As noted earlier, although there are a plethora of papers dealing with CO oxidation on platinum, there are few attempts to use mechanistic models in the simulation of automotive catalytic converters. Chatterjee et al. [12] used a small variation of the LH model; the variation being that all steps were considered to be reversible. Thus, the steps were: COðgÞ þ uV Ð uCO

(5)

O2ðgÞ þ 2uV Ð 2uO

(6)

uO þ uCO Ð uCO2 þ uV

(7)

uCO2 Ð CO2ðgÞ þ uV

(8)

This model allows CO adsorption on an oxygen covered surface by virtue of the reversibility of oxygen adsorption. If this step were to be irreversible, then CO adsorption on an oxygen covered surface would not be possible. For reference, we call this model M1. Another model [4,10–17] retained the Eley– Rideal term, and also wrote the dissociative adsorption of oxygen as two separate steps. The proposed mechanism for

308

S. Salomons et al. / Applied Catalysis B: Environmental 70 (2007) 305–313

CO oxidation on platinum sites was:

2.5. Compressed oxygen model

COðgÞ þ uV Ð uCO

(9)

O2ðgÞ þ uV ! uO2

(10)

uO2 þ uV ! 2uO

(11)

uO þ uCO ! CO2ðgÞ þ 2uV

(12)

COðgÞ þ uO Ð uOCO

(13)

uOCO ! CO2ðgÞ þ uV

(14)

Step (11), the dissociation of adsorbed oxygen molecules, was assumed to be instantaneous. The proposed rate determining step during light-off was the adsorption of oxygen. For reference, we shall refer to this model as M2. In the following, we will discuss the validity of these models in light of experimental and numerical evidence.

Exclusion models proposed by Herz and Marin [31] were also based on reactions (1)–(3). They cited evidence [40–43] that maximum coverage of CO and oxygen were 1.0 and 0.5, respectively, and proposed two models. The rate expressions for their first model are:

Desorption of CO :

Adsorption of O :

r ads;CO ¼ F CO SCO ð1  uCO  uO Þ (15)   ðE2  buCO Þ r des;CO ¼ k2 exp uCO RT (16) r ads;O ¼ 2F O2 SO ð1  f uCO  NuO Þ2 (17) 

Surface reaction

COðgÞ þ uV Ð uCO

(21)

O2ðgÞ þ 2uV ! 2uO

(22)

uO þ uCO ! CO2ðgÞ þ 2uV

(23)

COðgÞ þ 2uO ! uCO þ uOO

(24)

uCO þ uOO ! CO2ðgÞ þ uO þ uV

(25)

uOO ! 2uO

(26)

We shall refer to this model as M4.

2.4. Exclusion model

Adsorption of CO :

As noted earlier, (see literature [18] for a complete description) an oxygen covered surface can adsorb significant amounts of CO. Adsorption of CO is reported to compress the adsorbed oxygen to levels not achievable by adsorption of oxygen alone [18]. We can add terms to the mechanism to represent this compression effect. Thus, we would have a mechanism written as:

r sur;CO ¼ k4 exp



E4 uCO uO RT

(18)

in the above equations, F is the frequency of collision of gas phase molecules with metal surface atoms, S sticking coefficient, b theconstant that depends on the CO-metal bond strength (coverage dependence of the heat of adsorption), N was a constant which depends on the maximum surface coverage of each species (to which they assigned a value of 2), and k are rate constants in s1. The factor f is given by:   1  NuO f ¼ (19) 1  uO

3. Experimental results In automotive catalytic converter studies, the most common type of result presented is the light-off curve, also known as ignition–extinction curves. We present a series of complete ignition–extinction curves for different inlet CO concentrations. In these experiments, the reactor and feed gas had an initial temperature near 350 K. The temperature of the feed gas was then increased at approximately 0.133 K s1 from 350 to 623 K, where it was held constant for about 30 min, after which all temperatures in the monolith were higher than 573 K. The feed temperature was then lowered by approximately 0.038 K s1. The cooling rate was governed by the cooling rate of the furnace used to heat the feed gas. In all cases the feed contained 6 vol.% oxygen, with either 500, 1000, 1500, or 2000 ppm of CO. In each case, the start-up procedure was first to flush with nitrogen for 1 min, followed by addition of oxygen, which was allowed to flow for a further minute, and finally addition of CO. This condition was held for 8 min before commencing the temperature ramp. Table 1 gives the ignition and extinction points, defined as the inlet gas temperature corresponding to 50% conversion, with temperatures in Kelvin. The complete light-off curves are shown in Fig. 1. The GHSV for these experiments was about 25,000 h1. It is seen that increasing concentration of CO causes a higher light-off temperature. The width of the hysteresis is smallest for the 500 ppm experiment, and

In their second model, they proposed a different equation for the adsorption of oxygen:

Table 1 Experimental ignition and extinction points for the light-off of CO in 6 vol.% oxygen at a GHSV of 25,000 h1

Adsorption of O :

ppm CO

Ignition T/(K)

Extinction T/(K)

Delta T/(K)

500 1000 1500 2000

403 426 436 444

373 387 397 404

30 39 39 40

(20)

r ads;O ¼ 2F O2 SO ð1  f uCO  NuO Þð1  NuO Þ We shall refer to this model as M3. Lynch [33] also used a model similar to this second form, with N 2 [1,2].

S. Salomons et al. / Applied Catalysis B: Environmental 70 (2007) 305–313

309

Fig. 2. Experimental light-off curves for inlet concentrations of CO of 2000 ppm with and without water at a GHSV of 25,000 h1.

3.1. Concentration ramps and steps

Fig. 1. Experimental light-off curves for different inlet concentrations of CO at a GHSV of 25,000 h1.

similar for higher concentrations. We observe that the reaction starts at the end of the reactor and then the reaction front propagates backwards towards the reactor entrance. This may be deduced by observing the temperature profiles along the reactor with time. Overall, this type of ignition and extinction behavior is consistent with that observed by other researchers in the field. In the automotive context, the inlet gas typically contains of the order of 10 vol.% water vapor. The effect of the addition of 10 vol.% water to the feed is to alter the light-off curves dramatically. The new ignition and extinction values are given in Table 2. A comparison of the curves with and without water is given in Fig. 2 for the case of 2000 ppm CO in the feed. We see that the light-off is significantly retarded and the hysteresis becomes wider (larger delta T between ignition and extinction points at 50% conversion).

The region between the ignition and extinction branches of the hysteresis curves is a region of two stable steady states, respectively corresponding to a predominantly oxygen covered or a CO covered surface. We consider in this section several experiments in which a step change of 1000 ppm CO was introduced into a 6 vol.% oxygen in nitrogen mixture. Thus, we admit a CO pulse to a surface that is predominantly covered with oxygen. These experiments were performed at different temperatures in the vicinity of the extinction temperature. The lowest temperature, 368 K, corresponds to a single steady state with a CO covered surface and negligible reaction. The middle temperature, 391 K corresponds to a point midway along the extinction arm, and the highest temperature, 395 K is near the top of the extinction arm. Fig. 3 shows the outlet concentration of CO2 at these three temperatures. For all three cases, we observe an immediate and rapid production of CO2. At 368 K, the production rate rapidly falls to zero, as the catalyst switches from oxygen covered to CO covered. At the two higher temperatures, the concentration of CO2 initially rises and then falls to a steady state value.

Table 2 Experimental ignition and extinction points for the light-off of CO in 6 vol.% oxygen at a GHSV of 25,000 h1 with 10 vol.% water in the feed ppm CO

Ignition T/(K)

Extinction T/(K)

Delta T/(K)

1000 1500 2000

442 456 465

368 381 389

74 75 76

Fig. 3. Carbon dioxide production at the reactor outlet after a step change in CO concentration to 1000 ppm at three temperatures. GHSV is 25,000 h1.

310

S. Salomons et al. / Applied Catalysis B: Environmental 70 (2007) 305–313

Fig. 4 shows the response of the system to a steady increase and decrease in the CO concentration. Depending on the temperature, the CO initially reacts until a critical inlet concentration is reached, at which point the reaction decreases sharply. This condition remains until the concentration declines below a second (different) critical value, at which point the reaction ignites and the outlet CO concentration falls to zero. The first point, at which the reaction stops, corresponds to the extinction point, and the second point, where the reaction resumes, corresponds to the ignition point. Effectively, temperature and concentration ramps yield similar information about the ignition and extinction points. Note that the spike in the CO concentration at the inlet is the result of the valve that controls the CO flow rate opening. 3.2. Modeling the Light-off curves In this section, we compare model predictions to the experimental light-off curves. The first test was of literature models M1 and M2. In this comparison, the parameters given in the original publications were used, and the precious metal loading (total number of active sites) was adjusted to give an agreement with the ignition point for one experiment (1500 ppm). Table 3 gives the ignition and extinction points for model M1 based on the inlet gas temperature starting with a surface covered by CO. The light-off curves are shown in Fig. 5. A quite different behavior was observed with model M2. Fig. 6 gives a typical light-off curve obtained with 1500 ppm CO. It is seen that the light-off curve is very broad and that the hysteresis is small. Light-off corresponds to a switch between a CO covered surface to a predominantly oxygen covered surface. Model predictions indicate that, as the temperature increases, the reaction initially ignites at the exit of the reactor. This type of behavior is predicted by the models M1 and M2. Clearly, model M1 does a superior job of representing the experimental light-off curves compared to model M2. The fit is perhaps surprisingly good, considering that we have not attempted to optimize any of the parameters in the model, other

Fig. 4. Experimental output from a continuous linear ramp of CO concentration at constant feed temperature of 394 and 407 K. The GHSV is 25,000 h1.

Table 3 Comparison of experimental ignition and extinction points for the light-off of CO in 6 vol.% oxygen at a GHSV of 25,000 h1 with simulation values, model M1 ppm CO

Ignition T/(K)

Extinction T/(K)

Delta T/(K)

500 1000 1500 2000

412 427 436a 444

380 386 388 388

32 41 48 56

a

Ignition temperature fit exactly to match experiment.

than the platinum loading. The drawback of model M1, however, is that the assumption of reversible oxygen adsorption is contrary to that observed in the surface science literature. However, we should point out that the reversibility of oxygen does not apparently play a significant role in the extinction curve. When the same model simulations were run with the assumption of irreversible oxygen adsorption, no effect was observed on the ignition or extinction behavior. Model M2 clearly does a poor job of reproducing the light-off curves. The curves are very broad and the hysteresis is much smaller than observed experimentally. The broadness of the light-off is a consequence of the oxygen adsorption step proceeding via an adsorbed molecular intermediate, even though the dissociation on the surface is assumed to be rapid. The key factor becomes the order of reaction of the oxygen adsorption. Indeed, disregarding the Eley–Rideal step, which is not important for light-off, the order of the oxygen adsorption reaction is the only difference between M1 and M2. The main conclusion to be drawn at this point is that with this LH type model, it is necessary to model the oxygen adsorption as dissociative chemisorption involving two surface sites to achieve the necessary steepness of the light-off curve, although it remains possible that one can achieve the effect by tuning the rate parameters. It was discussed earlier that the CO oxidation mechanism must contain a route for the adsorption of CO on an oxygen covered surface. Model M1 achieves this by reversible oxygen adsorption, and model M2 uses an Eley–Rideal step. In model M4, we add steps allowing for CO to adsorb on the oxygen covered surface by ‘‘compressing’’ the already adsorbed oxygen onto fewer sites. Model M4 thus retains the features and constants of model M1, albeit with the addition of the compression steps and the deletion of the desorption of oxygen. The light-off portion of the curve is thus the same as for M1, because the surface starts with complete coverage by CO. The extinction curve, however, will adjust depending on the parameters chosen for the compressed oxygen steps. As observed above, the basic LH model gives rise to an extinction based on the competitive adsorption of CO and oxygen. Because, with the parameters currently chosen, the predicted extinction temperatures are higher than those observed experimentally, the parameters chosen for the compressed oxygen sub-model must be such that the rate of compressed oxygen formation is relatively small. We chose values for the adsorption rate to be the maximum possible without increasing the extinction temperature. Therefore, the ignition–extinction

S. Salomons et al. / Applied Catalysis B: Environmental 70 (2007) 305–313

311

Fig. 5. Comparison of experimental light-off curves with model M1. Values of rate constants from literature [14].

curves predicted by model M4 are identical to those predicted by model M1.

In this section, we model the concentration step experiments, starting with model M1. The model results do not predict the experimentally observed behavior. In the experiments, the CO may react initially before ultimately poisoning

the surface (at the correct temperature, see Fig. 4), however, model M1 shows the opposite type of behavior. After the admission of the step, the conversion of CO remains at zero for an induction period, and then switches to 100%. This behavior is shown in Fig. 7, where time zero corresponds to the start of the CO injection. The length of this induction period increases with decreasing temperature. This result is expected, and can be explained in terms of surface coverage. The surface begins as oxygen covered, but the reversible oxygen adsorption will

Fig. 6. Comparison of experimental light-off curve with model M2 for 1500 ppm CO. Values of rate constants from literature [6].

Fig. 7. Production of CO2 from the reactor following a step increase in CO concentration from 0 to 1000 ppm. The step occurred at time zero. Model M1, classical LH with reversible oxygen adsorption. GHSV is 25,000 h1.

3.3. Modeling the concentration steps and ramps

312

S. Salomons et al. / Applied Catalysis B: Environmental 70 (2007) 305–313

Fig. 8. Production of CO2 from the reactor following a step increase in CO concentration from 0 to 1000 ppm. The step occurred at time zero. Model M4, classical LH with compression of oxygen by CO adsorption. GHSV is 25,000 h1.

Fig. 10. Simulations for the output of CO with a continuous linear ramp of CO inlet concentration at three different temperatures. The compressed oxygen model is used. The experimental inlet concentrations were used. The GHSV is 25,000 h1.

allow a gradual displacement of some oxygen by CO. After some induction period, sufficient CO is present on the surface to initiate reaction. The surface then becomes rapidly cleaned of oxygen, and the reaction switches to 100% conversion. We note that, if we make the adsorption of oxygen irreversible (as is usually assumed to be the case below 700 K) then we cannot achieve any conversion of CO at all for the entire pulse duration. Conceptually, model M1 cannot explain the experimental behavior. Although the results are not shown here, modeling the constant ramp experiments gave the same opposite behavior (as expected). That is, initially the conversion is zero, and then it increases to one. The compressed oxygen model, M4, on the other hand, is able to model the trend observed experimentally. For example, a set of simulations performed around the extinction temperature gave the results shown in Fig. 8. At lower temperature the catalyst allows for an initial conversion of CO, and a pulse of CO2 is observed. At higher temperatures the conversion of CO becomes sustained, and essentially complete

conversion is achieved. For the lower temperature runs, the width of the CO2 pulse in the effluent is much narrower than that observed experimentally. This result indicates that the model predicts much fewer adsorbing CO molecules react, before the surface becomes poisoned. This result is consistent with the sharp extinction curve, which is the result of the very rapid switch between oxygen covered and CO covered. Experimentally, both the shallow extinction curve and the relatively large amount of CO2 produced during a step show a much slower transition between these two states. The difference between the model and experiments could indicate an incorrect ratio of the rates of adsorption of CO and oxygen. Finally, the exotherm predicted by the model for the latter case is shown in Fig. 9. At time zero there is an increase in the temperature corresponding to the reaction between CO and adsorbed oxygen. At 600 s, where the CO is switched off, a further jump occurs as the adsorbed CO is reacted away quickly by the incoming oxygen. The small temperature rise in between indicates a small conversion. The compressed oxygen model is also able to predict qualitatively the results observed for the linear ramps of inlet CO concentration. Fig. 10 shows the response of model M4 for the linear ramp from 0 to 2400 ppm CO and back to zero at three different temperatures. The experimental input CO ramp was used, and the small spike at the beginning is the result of the valve opening. In all cases the behavior is similar. Initially, there is a reaction as the CO reacts with the adsorbed oxygen. When the concentration reaches the critical value, the surface is poisoned and the reaction stops. It resumes when the concentration drops below the corresponding light-off point. The classical LH mechanism cannot predict this performance. 4. Conclusions

Fig. 9. Temperature increase in the reactor during a pulse injection of CO at 1000 ppm for 600 s at 411 K. There is temperature jump at the beginning and end of the step. The small temperature rise in the middle indicates some small conversion.

In this paper, we considered some elementary schemes for the oxidation of CO over a platinum catalyst. The ignition– extinction curves were modeled best by an LH mechanism with a dissociative chemisorption step for oxygen requiring two

S. Salomons et al. / Applied Catalysis B: Environmental 70 (2007) 305–313

surface sites. The predicted extinction curve found using literature values for the reaction coefficients is much sharper than the one observed experimentally. Possibly the agreement in this region could be improved by adjusting the parameters used in the model, especially the relative adsorption rates of CO and oxygen. A key requirement in the mechanism is to have a step to allow for the reaction of CO with an oxygen covered surface. The proposal to let oxygen coverage be reversible will allow for this effect, however, the predicted behavior for this assumption is opposite to that observed experimentally under concentration step changes of CO with an oxygen covered surface. Further, this mechanism is not consistent with the surface science. The best mechanism tested to date allows for CO to adsorb on an oxygen covered surface by compressing the oxygen already adsorbed. This mechanism is consistent with surface science observations. This last mechanism is able to predict qualitatively the response of the system to step changes and ramps in the CO concentration. Model M4 is clearly an improvement over both of models M1 and M2, however, additional steps or parameter adjustment must be considered, before a perfect agreement can be obtained. We also note that the model deficiency in the prediction of the extinction curve is consistent with the small amount of CO2 produced in the step experiments. References [1] J. Wei, Adv. Catal. 24 (1975) 57. [2] R.H. Heck, R.J. Farrauto, Catalytic Air Pollution Control—Commercial Technology, first ed., Van Nostrand Reinhold, 115 fifth avenue, New York, NY, 1995. [3] G.C. Koltsakis, A.M. Stamatelos, Prog. Energy Combust. Sci. 43 (1997) 1. [4] A. Nievergeld, Automotive Exhaust Gas Conversion: Reaction Kinetics, Reactor Modeling and Control. Ph.D. Thesis, Technische Universiteit Eindhoven, 1998. [5] S.E. Voltz, C. Morgan, D. Liederman, S.M. Jacob, Ind. Eng. Chem. Prod. Res. Dev. 12 (1973) 294. [6] C. Dubien, D. Schweich, G. Mabilon, B. Martin, M. Prigent, Chem. Eng. Sci. 53 (1998) 471. [7] J.P. Leclerc, D. Schweich, in: H.J. de Lasa, et al. (Eds.), Chemical Reactor Technology for Environmentally Safe Reactors and Products, 547, Kluwer Academic Publishers, 1993.

313

[8] S.H. Oh, J.C. Cavendish, Ind. Eng. Chem. Prod. Res. Dev. 21 (1982) 29. [9] S.H. Oh, J.C. Cavendish, Ind. Eng. Chem. Prod. Res. Dev. 22 (1983) 509. [10] R. Nibbelke, Steady state, transient kinetics in automotive exhaust gas catalysis. Ph.D. Thesis, Technische Universiteit Eindhoven, 1998. [11] R.H. Nibbelke, A.J.L. Nievergeld, J.H.B.J. Hoebink, G.B. Marin, Appl. Catal. B: Environ. 498 (1998) 1. [12] D. Chatterjee, O. Deutschmann, J. Warnatz, Faraday Discuss. 119 (2001) 371. [13] J.M.A. Harmsen, J.H.B.J. Hoebink, J.C. Schouten, Catal. Lett. 71 (1–2) (2001) 81. [14] J.M.A. Harmsen, J.H.B.J. Hoebink, J.C. Schouten, Chem. Eng. Sci. 56 (2001) 2019. [15] J.M.A. Harmsen, J.H.B.J. Hoebink, J.C. Schouten, Ind. Eng. Chem. Res. 39 (2000) 599. [16] R.H. Nibbelke, M.A.J. Campman, J.H.B.J. Hoebink, G.B. Marin, J. Catal. 171 (1997) 358. [17] M.A.J. Campman, Kinetics of Carbon Monoxide Oxidation over Supported Platinum Catalysts. The Role of Steam in the Presence of Ceria, Ph.D. Dissertation, Eindhoven University of Technology, 1996. [18] T. Engel, G. Ertl, Adv. Catal. 28 (1979) 1. [19] L.S. Mukadi, R.E. Hayes, Comput. Chem. Eng. 26 (2002) 439. [20] B.W. Wojciechowski, S.P. Aspey, Appl. Catal. A: Gen. 190 (2000) 1. [21] L.F. Razon, R.A. Schmitz, Catal. Rev. Sci. Eng. 28 (1986) 89. [22] V.P. Zhdanov, B. Kasemo, Surf. Sci. Rep. 20 (1994) 111. [23] M. Ba¨r, Ch. Zu¨licke, M. Eiswirth, G. Ertl, J. Chem. Phys. 96 (1992) 8595. [24] M. Berdau, G.G. Yelenin, A. Karpowicz, M. Ehsasi, K. Christmann, J.H. Block, J. Chem. Phys. 110 (1999) 11551. [25] S. Wehner, F. Baumann, M. Ruckdeschel, J. Ku¨ppers, J. Chem. Phys. 119 (2003) 6823. [26] A. Bourane, D. Bianchi, J. Catal. 222 (2004) 499. [27] A. Bourane, D. Bianchi, J. Catal. 220 (2003) 3. [28] A. Bourane, D. Bianchi, J. Catal. 209 (2002) 114. [29] A. Bourane, D. Bianchi, J. Catal. 209 (2002) 126. [30] A. Bourane, D. Bianchi, J. Catal. 202 (2001) 34. [31] R.K. Herz, S.P. Marin, J. Catal. 65 (1980) 281. [32] D.T. Lynch, Can. J. Chem. Eng. 61 (1983) 183. [33] D.T. Lynch, Can. J. Chem. Eng. 62 (1984) 691. [34] D.T. Lynch, Can. J. Chem. Eng. 64 (1986) 1035. [35] W.R.C. Graham, D.T. Lynch, AICHE J. 33 (1987) 792. [36] W.R.C. Graham, D.T. Lynch, AICHE J. 36 (1990) 1796. [37] B.C. Sales, J.E. Turner, M.B. Maple, Surf. Sci. 114 (1982) 381. [38] M.P. Harold, M.E. Garske, J. Catal. 127 (1991) 524. [39] Y. Barshad, X. Zhou, E. Gulari, J. Catal. 94 (1985) 128. [40] G.R. Wilson, W.K. Hall, J. Catal. 17 (1970) 190. [41] G.R. Wilson, W.K. Hall, J. Catal. 24 (1972) 306. [42] H. Conrad, G. Ertl, J. Kuppers, Surf. Sci. 76 (1978) 323. [43] R.W. McCabe, L.D. Schmidt, Surf. Sci. 66 (1977) 101.