Effect of Porosity of Carbogenic Molecular Sieve Catalysts on

The initial yield of styrene was over 70% and slowly fell to 55% with time on stream. For Carbosieve G under the same conditions, the initial yield ro...
0 downloads 0 Views 241KB Size
Ind. Eng. Chem. Res. 1996, 35, 3319-3331

3319

Effect of Porosity of Carbogenic Molecular Sieve Catalysts on Ethylbenzene Oxidative Dehydrogenation Michael S. Kane, Lien C. Kao, Ravindra K. Mariwala, David F. Hilscher, and Henry C. Foley* Center for Catalytic Science and Technology, Department of Chemical Engineering, University of Delaware, Newark, Delaware 19716

The conversion of ethylbenzene to styrene by oxidative dehydrogenation is compared over several carbogenic molecular sieves. At 300 °C, Carbosieve G deactivated rapidly due to its nanoporous structure. During deactivation, the apparent activation energy dropped nearly a factor of 2 with a corresponding pore volume decrease from 87 to 15 mg/g. The carbogenic Ambersorb adsorbents have meso- and macroporosity in addition to nanoporosity. With 1% oxygen in the feed and at 400°C, the rates of coke deposition on Ambersorb 563 were below measurable levels. The initial yield of styrene was over 70% and slowly fell to 55% with time on stream. For Carbosieve G under the same conditions, the initial yield rose rapidly to 70% but then fell to approximately 10%. The differences in these materials can be attributed to the effects of differences in pore structure and their influence on the coupled reaction and diffusion phenomena. These results indicate that carbogenic catalysts can produce commercially relevant levels of styrene with minor deactivation, providing the pore structure includes substantial amounts of transport porosity and the oxidizing potential of the gas phase is mediated properly. Introduction In industrial practice the conversion of ethylbenzene to styrene is currently conducted as a dehydrogenation process (Lewis et al., 1983; Satterfield, 1991). Because of the endothermicity of this reaction (∆H° ) 121 kJ/ mol), the reaction is operated at high temperatures (Trxn ) 520 °F) and with excess steam (15:1 steam/ethylbenzene). An alternative approach is oxidative dehydrogenation or oxydehydrogenation. By co-feeding oxygen with ethylbenzene, water is produced rather than hydrogen, resulting in a reaction that is markedly exothermic (∆H° ) -134 kJ/mol). The advantage of this route is that the reaction can be operated at lower temperatures, without a steam co-feed, and the secondary cracking, thermal and acid, that leads to toluene, benzene and coke in the dehydrogenation process can be avoided. Because of these potential advantages, oxidative dehydrogenation of ethylbenzene has been the object of considerable research (Emig and Hofmann, 1983; Batist et al., 1966; Murakami et al., 1981, 1982). Many of the oxide catalysts studied are mildly acidic and tend to form carbon-coke. A considerable body of evidence now points to this carbon-coke as the actual active phase for the catalytic reaction (Cadus et al., 1988, 1990; Alkhazov and Lisovski, 1976; Fiederow et al., 1978; Roth and Schaefer, 1966, 1969; Berger and Roth, 1968). The well-known quinone-hydroquinone redox shuttle is invoked as an analogy to this carbon-coke surface process for polynaphthoquinone (Iwasawa et al., 1972, 1973). In this regard, the use of unsupported carbonsscarbogenic molecular sieves and activated carbonssshould provide high surface area, porosity control, and elimination of any environmental hazard associated with the disposal of used metal oxide catalysts. Mannassen and Wallach (1965) found that pyrolyzed polyacrylonitrile

was an active dehydrogenation catalyst in the presence of air. Alkhazov (1978) also showed that activated carbon was catalytic. Later, Lee (1987) compared the oxydehydrogenation for styrene production with coconut charcoal versus the carbogenic molecular sieves (CMS), Carbosieve G and MSC-V. The CMS catalysts provided a styrene yield that was 10 times higher than that of the charcoal (30% versus 3% yield). Grunewald and Drago (1990) reported that at 350 °C AX-21 produced styrene in 80% yield, for 20 h on stream with no evidence of deactivation or coking. More recent work (Drago and Jurczyk, 1994) indicates that the carbon catalysts, including Ambersorb adsorbents, Carbosieve G and AX-21 were actually subject to coking with time on stream. Now Ambersorb adsorbents provided 50-56% and the AX-21 27% styrene yields, respectively. With CBr4 in the feed, coking could be kept to near zero levels with styrene yields ranging from 40% to 72%. Guerrero-Ruiz and Rodriguez-Ramos (1994) examined graphites, activated carbons, and alumina. Significantly, they concluded that selectivity to styrene was a function of the capacity for labile hydrogen of the carbon surface. This capacity, a function of the chemical nature of the surface coke, is in line with the proposal of a quinone-hydroquinone redox shuttle doing the catalysis. Because of the potential importance of carbogenic catalysts and the apparent effects of coking in this reaction, we sought (1) to examine the behavior of several different CMS materials to determine if the behavior was general, (2) to study one of these materials, Carbosieve G, in detail during deactivation, and (3) to compare the time-factored behavior of the exclusively nanoporous Carbosieve G with that of CMS having both meso- and macroporosity, namely, Ambersorb 563 and Ambersorb 572. Experimental Section

* Author to whom correspondence is addressed. Phone: (302) 831-6856. Fax: (302) 831-2085. E-mail: foley@ che.udel.edu.

S0888-5885(95)00762-7 CCC: $12.00

Carbon Materials. The catalysts used in this study were obtained from commercial sources, Carbosieve G © 1996 American Chemical Society

3320 Ind. Eng. Chem. Res., Vol. 35, No. 10, 1996 Table 1. Surface Areas and Pore Volumes of Carbosieve G and Ambersorb Adsorbents pore volume (cm3/g) sample

surface area (m2/g)

micro

meso

macro

Carbosieve G Ambersorb 563 Ambersorb 572

1100 680 1100

0.36 0.23 0.41

0 0.14 0.19

0 0.23 0.24

was purchased from Supelco, and Ambersorb 563 and 572 were obtained from Rohm and Haas Co. These materials were utilized as obtained without further modification. The physical properties of these materials are briefly summarized in Table 1. Calgon BPL carbon was obtained from Calgon Corp. and was used as supplied. Pyrolyzed poly(vinylidene chloride) (PPVDC) was prepared from Saran beads by heating in flowing nitrogen at 900 °C for 10 h. Pyrolyzed poly(furfuryl alcohol) (PPFA) was prepared following published procedures (Mariwala and Foley, 1994a). The final pyrolysis temperature was 600 °C, and the sample was treated at this temperature for “zero hours”, meaning that the temperature was ramped to 600 °C and then the CMS was allowed to cool as the furnace temperature dropped back to room temperature. Reactor Studies. The time on-stream studies of the oxidative dehydrogenation reaction were conducted in a microflow reactor fashioned from a quartz tube (1.25 cm o.d. × 50.8 cm), heated in a close-fitting tube furnace, and provisioned to feed ethylbenzene and oxidant to 0.5 g of the catalyst particles held between two plugs of quartz wool. The ethylbenzene was fed to the reactor in the vapor phase by flowing air or 1% O2 in helium through a sparger held at room temperature. The vapor pressure of ethylbenzene at 25 °C is 7.1 Torr, and the feed was taken to be saturated with ethylbenzene. Spot checks on the feed by gas chromatography confirmed the accuracy of this assumption. Product gases were passed through a cold trap prior to being sampled with a gas-tight syringe. The products were analyzed on a Gow-Mac 750P gas chromatograph with GP 5% SP1200/1.75% Bentone 34 on a 100/120 Supelcoport column (6 ft × 1/8 in. stainless steel (SS)) and a Hewlett Packard (HP) 5700A gas chromatograph fitted with Porapak R (20 ft × 1/8 in. SS) and 60/80 Carboxen 1000 (15 ft × 1/8 in. SS) columns. The products isolated in the cold trap were identified using an HP 5890/5970 GC/ MS spectrometer and an HP-1 capillary column (50 m × 0.2 mm) by injecting authentic samples of the species to be identified. Some samples also were characterized by infrared spectroscopy on a Nicolet 510M FT-IR spectrometer. To ensure safe operation, the reactor was shut down at the end of each day by cooling in flowing air. The reactor was disconnected from the feed and exit lines and weighed. The next day the reactor was placed in the unit and reheated to the experimental temperature with air flowing through the bed. This step typically required 1 h. Once the temperature set point was reached, the feed was switched to ethylbenzene and air or ethylbenzene and either 1% oxygen in helium or 10% oxygen in helium. For the activation energy measurements a second reactor was utilized. A 0.5 g sample of Carbosieve G was positioned inside a 6.25 mm quartz tube and held in place by glass wool on either end. Heat was supplied to the reactor resistively. Pretreatment of this sample involved purging with an atmosphere containing Ar/O2 (60 sccm; 2.8% O2) for 30 min at room temperature, followed by heating to 400 °C for 30 min, and then cooling to 300 °C. The gaseous mixture was diverted

through a sparger filled with ethylbenzene to produce a mixture of argon, oxygen, and ethylbenzene near the saturation equilibrium point at 20 °C (7.1 Torr EB or ∼1% in Ar). The oxygen/ethylbenzene ratio was approximately 2.8:1. The gases exiting the reactor were monitored with an on-line Balzers QMG 511 mass spectrometer. Pore Structure Analysis. Analyses of the porosity of new and used samples of Carbosieve G were made with methyl chloride adsorption at room temperature following previously published procedures (Mariwala and Foley, 1994b). Results Comparison of Carbogenic Catalysts. Reactor Studies with Air Co-feed. In the study of different carbogenic catalysts, the feed consisted of air and ethylbenzene with oxygen in nearly 15-fold excess over the ethylbenzene (30-fold stoichiometric excess). The temperature was 300 °C, and the flowrate was held at 10 sccm (weight hourly space velocity of 0.05 h-1). Catalysts tested included Carbosieve G, Calgon BPL, PPVDC, and PPFA (Figure 1). Each provided complete consumption of ethylbenzene, and no products were eluted from the reactor at early times. When gas-phase species were detected, styrene was detected first, followed by ethylbenzene some time later. As the yield of styrene increased with time on stream, the conversion of ethylbenzene monotonically decreased from its apparent 100% level. Detailed examination of the data showed that the materials were quite different. PPFA deactivated the most rapidly with time on stream as the ethylbenzene conversion and styrene yield merged to the same value, indicating that the selectivity to styrene was greater than 99% after 220 min on stream. In contrast, for the other three catalysts the conversion and yield curves never merged during the course of these experiments, evidence that products other than gas-phase styrene were produced but not detected. The Calgon BPL and PPVDC catalysts behaved similarly, and the styrene yield rose to 20-25% and then dropped slowly to a final level just under 20%. The Carbosieve G sample deactivated the most slowly, based on the change in ethylbenzene conversion with time on stream, and the styrene yield rose to nearly 45% before dropping to ∼40% just prior to completion of the experiment. For these reasons Carbosieve G was chosen for further study. Detailed Examination of Carbosieve G. Reactor Results with Long Time on Stream. The conversion of ethylbenzene and the yield of styrene are displayed in Figure 2. The most important feature of the results is the wide gap between the conversion of ethylbenzene and the yield of styrene. At early times on stream (95%) plus minor components was produced. In the next experiment the bed size was reduced to its original size (0.5 g) and the flow set at 20 sccm to give a WHSV of 0.1 h-1 (Figure 8b). During this run the bed mass increased by 100 mg, and nearly 100 mg of (primarily) benzoic acid was produced. But in this case the time-factored behavior is quite different since the conversion of ethylbenzene drops more rapidly, with breakthrough occurring at less than 80 min on stream. After elution of styrene at just over 30 min on stream, the yield of styrene rose to a plateau at 45-50%. Notably, the region between the yield and conversion

3324 Ind. Eng. Chem. Res., Vol. 35, No. 10, 1996

Figure 8. Effects of space velocity on ethylbenzene conversion and styrene yield over Ambersorb 563 (T ) 300 °C, PEB ) 7.1 mmHg, co-feed air, (a) WHSV ) 0.05 h-1, (b) 0.1 h-1, (c) 0.2 h-1, (d) 0.4 h-1).

curve decreased in area at long times, but the two curves never merged at these conditions. Next, the flow rate was raised to 40 sccm, corresponding to a WHSV of 0.2 h-1 (Figure 8c). Here the breakthrough times of styrene and ethylbenzene were quite short, occurring at less than 20 min on stream. The conversion of ethylbenzene fell rapidly as the yield of styrene rose to nearly 50% and then dropped slowly with time to 40%. The catalyst bed increased in mass by 94 mg, and 131 mg of solid products was isolated which, once again, was primarily benzoic acid. However, the solid did include a mixture of more highly oxidized products in low concentrations. Finally, the flow rate was raised to 80 sccm to produce a WHSV of 0.4 h-1, which corresponds closely to the space velocity used industrially. The behavior was quite different than in any of the previous experiments (Figure 8d). The conversion of ethylbenzene dropped precipitously from 100% to less than 20% in just over 1 h; the styrene yield rose to approximately 20%, and then it too dropped precipitously. Examination of the catalyst bed upon completion of the run indicated that it had been consumed under these forcing conditions. The net mass of the catalyst lost after the reaction was 224 mg or 50% of its original mass. Comparisons of Ambersorb and Carbosieve G Using a 2-Fold Excess of Oxygen in Helium Feed. Based on the results of the experiments conducted with excess oxygen in the feed, it was of interest to raise the flow rate through the catalyst, to raise the styrene yield, and to approach space velocities used in the industrial process (a WHSV of approximately 0.4 h-1). However, since the catalyst was consumed at this space velocity

with a co-feed of air, the oxygen content was lowered to 1% in the feedstock while keeping the ethylbenzene fixed. At 1% oxygen and with 7.1 torr of ethylbenzene in the feed, the mole ratio is 1:1 which is a 2-fold stoichiometric excess of oxygen. Interestingly, the conversion of oxygen under these conditions is always well below the level it should be (at 70% ethylbenzene conversion, 20% experimental oxygen conversion versus 35% theoretical). This suggests the reaction may be due partially to oxydehydrogenation and partly to dehydrogenation occurring in parallel. With 1% oxygen in the feed and a WHSV of 0.4 h-1, the temperature needed to be raised to 400 °C in order to achieve conversions and yields comparable to those obtained with excess (20%) oxygen at 300 °C (Figure 9). With Ambersorb 563 under these conditions, the time-factored behavior of the reaction was quite dissimilar to that obtained in the previous experiments (Figure 10). First, for all five periods of testing the conversion of ethylbenzene and the yield of styrene were equal within experimental error. Second, the amount of mass gained by the bed was negligible to slightly negative and the mass of secondary product, mostly benzoic acid, produced was quite small. Finally, the yield of styrene remained at 50% or higher throughout the duration of the experiment. In the first period the bed mass decreased by 3 mg and the mass of benzoic acid plus minor components was 21 mg. In the second, third, and fourth periods the catalyst mass increased by 6, 9, and 7 mg, and the masses of solid benzoic acid plus minor components produced were 11, 7, and 9 mg. The same experiment was run with Ambersorb 572

Ind. Eng. Chem. Res., Vol. 35, No. 10, 1996 3325

despite the standard air treatment prior to each subsequent test period, the activity did not recover; styrene yield dropped from 20% to under 10% during the second period and then stabilized between 5 and 10% during the last two periods of testing. Discussion

Figure 9. Effect of temperature on ethylbenzene conversion and styrene yield over Ambersorb 563 (1% O2 in He, q ) 80 sccm, WHSV ) 0.4 h-1, (a) T ) 300 °C, (b) T ) 350 °C, (c) T ) 400 °C).

for 360 min on stream. The yield of styrene and conversion of ethylbenzene were nearly the same as those obtained in the first period with Ambersorb 563 (Figure 11). The deposit on the catalyst weighed 34 mg, while the benzoic acid plus minor components amounted to 8 mg during this period. Finally, Carbosieve G was examined under these new reaction and reactor conditions to determine if its behavior would be affected in the same way as that of the two Ambersorb catalysts. The results of an extended experiment were very different compared to those observed earlier or to those obtained with the Ambersorb catalysts (Figure 12). During the first period of testing the activity and yield increased rapidly from just above 10% to nearly 70% after 40 min on stream. However, after this the catalyst deactivated rapidly, dropping the yield to under 20% at the end of the period (460 min). During the next three periods of testing,

The behavior of the carbogenic molecular sieve catalysts was quite different depending upon their origin. PPFA deactivated the most rapidly, while Carbosieve G deactivated somewhat more slowly (Figure 1). Calgon BPL and PPVDC were more intermediate in their behavior. Earlier work with methyl chloride adsorption on Carbosieve G and PPFA (Mariwala and Foley, 1994a,b) has shown that the total porosity, the mode of the micropore size distribution, and the breadth of the micropore distribution are each larger for Carbosieve G than for PPFA. Both materials are completely microporous, with no meso- or macroporosity. The rate of deactivation should be expected to be higher on PPFA than on Carbosieve G, if that deactivation process arises from coke deposition in the pore structure. Calgon BPL and PPVDC behaved so similarly to Carbosieve G that it is logical to surmise that their micropore structures are similar. This is not surprising since PPVDC prepared in our laboratory and Carbosieve G are derived from the same polymer blend. However, no attempt was made in preparing the sample to duplicate the conditions for Carbosieve G production, since these are not known. The time-factored behavior of the Carbosieve G catalyst with long times on stream bolsters the argument for the important role of catalyst porosity in this reaction. When tested for long times on stream with intermediate reactivation, the trend was toward lower activity, higher selectivity, and thus a higher yield of

Figure 10. Ethylbenzene conversion and styrene yield over Ambersorb 563 (1% O2 in He, q ) 80 sccm, WHSV ) 0.4 h-1, T ) 400 °C).

3326 Ind. Eng. Chem. Res., Vol. 35, No. 10, 1996

Figure 11. Ethylbenzene conversion and styrene yield over Ambersorb 572 (1% O2 in He, q ) 80 sccm, WHSV ) 0.4 h-1, T ) 400 °C).

Figure 12. Ethylbenzene conversion and styrene yield over Carbosieve G (1% O2 in He, q ) 80 sccm, WHSV ) 0.4 h-1, T ) 400 °C).

styrene. The activation energy was observed to fall from 15.6 kcal/mol after 6 h on stream to 7.9 kcal/mol after 24 h on stream. Thus, both the activation energy and the activity fell simultaneously; furthermore the drop by a factor of 2 in the activation energy and to such a low value was significant, because it points to a transition with time on stream from a chemically controlled

regime to one dominated by internal or pore-diffusion control (Satterfield, 1991). The fall in activation energy and activity and the accumulation of mass on the catalyst indicates that, as the reaction proceeded in time, the catalyst porosity was lost due to coke accumulation. Thus, transport rates fell due to pore narrowing and blockage.

Ind. Eng. Chem. Res., Vol. 35, No. 10, 1996 3327

η)

3 φ(coth φ) - 1 φ φ

(2)

As φ approaches infinity, eq 2 simplifies to:

η ) 3/φ

(3)

Assuming that the reaction of ethylbenzene to styrene follows first-order kinetics, we can calculate first-order rate constants from our conversion versus time data using:

kτ ) -ln(1 - χ)

Figure 13. Schematic of coke deposition in Carbosieve G as a function of time on stream during ethylbenzene oxydehydrogenation.

Methyl chloride adsorption on the fresh and used samples of Carbosieve G proved the case for coking and its effect on this microporous material. Referring to the uptake of methyl chloride as a function of exposure time (Figure 5), it is evident that the catalyst used for the activation energy measurement experiments had lost significant porosity and that the diffusivity of methyl chloride in the remaining porosity had dropped significantly. This is consistent with the proposal that the pores actually have been reduced in size as well as blocked. Schematically, our interpretation of the change in the Carbosieve G catalyst as a function of time on stream is shown in Figure 13. This motivates an analysis of the reaction and diffusion phenomena in these catalysts. Coking-Induced Internal Pore-Diffusion Limitations: Effectiveness Factor Changes with Time on Stream for Carbosieve G versus Ambersorb 563. The results for Carbosieve G for long times on stream were characterized by 100% conversion of ethylbenzene at short times followed by a nearly linear decrease in conversion to a final level of 22% at long times on stream (Figure 2). The deactivation of this entirely microporous catalyst was presumably due to coking, which causes narrowing of the micropores and an increase in the diffusion limitations experienced by the ethylbenzene reactant. To further explore the impact of internal mass-transfer limitations on the observed long-time behavior, a more detailed analysis of the coupled reaction and diffusion phenomena was needed. Coupled reaction and diffusion is most appropriately analyzed using the formalism of the Thiele modulus (φ) and effectiveness factor (η). We can assume that the deactivation due to coking manifests itself entirely through changes in the effective diffusivity of the reactant within the pores of Carbosieve G, and we use the Thiele model to test this assumption. For a firstorder reaction the Thiele modulus is 1/2

φ ) L(kint/D)

(1)

where L is the particle diameter, kint is the intrinsic first-order rate constant, and D is the effective diffusivity of the reactant. The effectiveness factor is experimentally determined as (for a spherical particle geometry):

(4)

where k is the observed first-order rate constant, τ is the residence time, and χ is the fractional conversion. The Thiele model requires kint, the rate constant observed in the absence of any diffusion limitations, which can be estimated by extrapolating back to zero time on stream. This is the lower bound on kint since, if η ) 1 at t ) 0, then kt)0 ) kint and, if η < 1 at t ) 0, then kt)0 < kint. Using a residence time of 3.1 s, the observed rate constant can be calculated for each time on stream. The deactivation is characteristic of an exponential decrease in the first-order rate constant with time on stream. From a semilog plot of k versus t, a k0 of 0.73 s-1 was calculated by linear regression (Figure 14). To get a true measure of the deactivation phenomena, we have excluded the first few data points at the beginning of each reaction test period. At the start of each successive period of reaction the conversion of ethylbenzene was initially higher than that at the end of the previous day and dropped rapidly to a value close to the end point of the previous day. We attribute this initial increase and sharp decrease in conversion to the fact that the reactor was cooled and then reheated in air between tests, which may have partially reoxidized the surface of the catalyst and provided an oxygen-rich surface during the initial stages of reaction in the subsequent test period. As this surface oxygen was consumed, the rate dropped dramatically, followed by the slower decrease in rate which is attributed to coking. Recall that methyl chloride adsorption has been used to characterize changes in the porosity of Carbosieve G that occur during reaction. The diffusivities calculated for methyl chloride before and after reaction can be used as approximate values for the diffusivities of ethylbenzene before and after the reaction (D0 ) 6.3 × 10-8 cm2/s and D∞ ) 1.2 × 10-9 cm2/s). This does not assume that diffusivities for ethylbenzene and methyl chloride are identical but rather that the ratios of the diffusivities before and after reaction are similar for the two species. These estimates for the kinetic and diffusive parameters in the system allow comparison of the Thiele model to the observed deactivation behavior. The Thiele modulus at t ) 0 and t ) ∞ can be calculated using the regressed value for k0 and the diffusivities measured for methyl chloride:

x

x

x

x

φ0 ) L

φ∞ ) L

k0 ) 0.04 cm D0

0.73 s-1 ) 135 6.38 × 10-8 cm2/s (5)

k0 ) 0.04 cm D∞

0.73 s-1 ) 987 1.20 × 10-9 cm2/s (6)

As expected, the Thiele modulus drops considerably with time on stream, also note that φ . 1, so that the

3328 Ind. Eng. Chem. Res., Vol. 35, No. 10, 1996

Figure 14. Observed first-order rate constant for ethylbenzene conversion versus time on stream for Carbosieve G (T ) 300 °C, q ) 10 sccm, WHSV ) 0.05 h-1, co-feed air). Table 2. Conversions and First-Order Rate Constants for Ethylbenzene Conversion on Carbosieve G at Short and Long Times on Stream t (min) 0 ∞

χ

k (s-1)

≈1.0 0.22

0.73 ln(1.0-0.22)/3.1 ) 0.08

effectiveness factor can be estimated by eq 3. The decrease in rate predicted by the Thiele model is calculated to be:

η∞ 3/φ∞ φ0 135 ) 0.137 ) ) ) η0 3/φ0 φ∞ 987

(7)

The Thiele model predicts that with long time on stream the rate will drop to 14% of its initial value if catalyst deactivation is assumed to reduce only the effective diffusivity of the reactant. This can be compared to what is obtained experimentally by calculating the observed rate constant at long time on stream (Table 2). The experimental decrease in rate can be obtained from the rate constants at t ) 0 and t ) ∞:

k∞ 0.08 s-1 ) ) 0.110 k0 0.73 s-1

(8)

This decrease in rate to 11% of its initial value agrees well with the value of 14% predicted by the Thiele model. The Thiele analysis, coupled with reasonable values for the kinetic and diffusional parameters, supports our contention that catalyst deactivation primarily occurred through coking and narrowing of the micropores, which led to a decrease in the effective diffusivity of the reactant and a corresponding decrease

in the observed rate of reaction. For Carbosieve G the decrease in rate is consistent with internal masstransfer limitations induced by coking of the micropores. The results obtained with Ambersorb 563 were similar to Carbosieve G in that they are characterized by 100% conversion of ethylbenzene at short times, followed by a nearly linear decrease in conversion (Figure 6). Two important differences are that the conversion for Ambersorb 563 only decreased to 80% of its initial value (as compared to 22% for Carbosieve G) and the yield and conversion curves did not converge as they did for Carbosieve G. In contrast to Carbosieve G, Ambersorb 563 contains meso- and macroporosity in addition to microporosity, so one would expect to see smaller changes in the effective diffusivity with time on stream. The overall effective diffusivity will be dominated by the macropore diffusivity term, which is several orders of magnitude larger than the largest micropore diffusivity and it is relativity insensitive to changes in pore size. The Thiele modulus is inversely proportional to the effective diffusivity, so the larger diffusivity for Ambersorb 563 will lead to a smaller Thiele modulus. As the Thiele modulus becomes small, the effectiveness factor approaches unity, pore diffusion has little or no effect on the overall observed rate of reaction, and the observed decrease in rate should be attributed to something other than coke-induced diffusional limitations. Another possible explanation for the decrease in activity with time on stream seen with Ambersorb 563 is that the macroporosity allows the micro- and mesopores to become completely coked. This would render the surface area associated with the micro- and mesopores inaccessible to the ethylbenzene reactant, so that all of the catalysis would be occurring in the macropores

Ind. Eng. Chem. Res., Vol. 35, No. 10, 1996 3329

Figure 15. Observed first-order rate constant for ethylbenzene conversion versus time on stream for Ambersorb 563 (T ) 300°C, q ) 10 sccm, WHSV ) 0.05 h-1, co-feed air).

at long times on stream. The validity of this hypothesis can be tested by comparing the observed decrease to that predicted by a complete loss of micro- and mesoporosity. First-order rate constants can be calculated from the conversion versus time on stream data. From a semilog plot of k versus t, a k0 of 1.23 s-1 was calculated by linear regression (Figure 15). The first few data points at the beginning of each reaction test period were neglected for reasons stated previously. The observed rate constant at long time on stream can be calculated to be 0.54 s-1. The experimental decrease in rate is the ratio of these, or 44%. We can assume that the concentration of active sites in the micro-, meso-, and macropores is proportional to the pore volume for each pore size. Surface area may be more appropriate in this case, but a microporous surface area is difficult to conceptualize. The proportion of macropores in Ambersorb 563 can be calculated to be:

macro )

0.23 ) 0.38 0.23 + 0.14 + 0.23

(9)

This suggests that the activity would drop to 38% of its initial value if the micro- and mesoporosity were completely coked and rendered inaccessible, which is close to the experimental value of 44%. Although this does not prove this accessibility argument, it suggests that complete coking of the micro- and mesopores could be responsible for the observed deactivation behavior. This also is consistent with the fact that the conversion/ selectivity curves do not merge at long times on stream for Ambersorb 563 as they do for Carbosieve G. The

larger macropores in the Ambersorb 563 give room for reactions other than dehydrogenation to occur. Perspective Finally, it is worth pausing for a moment to place these results in context with earlier work. Over Carbosieve G the yield of styrene did reach 30% with very high selectivity (>95%) as Lee (1987) observed, but only after many hours on stream under our conditions. The severity of the transient behavior of the catalysts due to catalyst coke accumulation and the formation of benzoic acid led to significant yield losses. Although earlier reports (Lee, 1987; Grunewald and Drago, 1990) did not mention coking and byproduct formation per se, the later report by Drago and Jurczyk (1994) did include results that indicated they also detected high levels of coking over their catalysts at their conditions. Yet, no mention was made of the production of benzoic acid as a major byproduct of the reaction along with other more complex oxyaromatics in minute quantities. This could have arisen from differences in conditions between the experiments, or, since the product condenses in the reactor just downstream of the catalyst bed, it could have gone undetected. Since it was our goal to test the Ambersorb catalysts at weight hourly space velocities that were comparable to industrial conditions (∼0.4 h-1), it was necessary to lower the oxygen fugacity in an effort to lower the rate of catalyst oxidation. In the experiments with air as the co-feed, oxygen was in great excess (15-fold) over the ethylbenzene. By lowering the oxygen content of the feed to 1 mol %, the ratio of oxygen to ethylbenzene was 1:1, which is still a 2-fold excess based on the

3330 Ind. Eng. Chem. Res., Vol. 35, No. 10, 1996

stoichiometric requirement of the reaction. Starting at 300 °C, the rate of reaction was much lower, but the rate of catalyst oxidation was negligible. The same was true at 350 and 400 °C, but since the conversions and selectivities at 400 °C approached those observed previously at 300 °C and 0.05 h-1, it was decided to operate at this condition. The long-term test results for the Ambersorb 563 are remarkable for several reasons. First, at this weight hourly space velocity, oxygen fugacity, and temperature, the conversion of ethylbenzene and yield of styrene were nearly identical, thus no coking occurred over the extended test period. Results similar to these were reported by Drago over AX21 (Grunewald and Drago, 1992), but over Ambersorb 563 and 572 at their conditions, the addition of CBr4 to the feed was required to prevent coking. So, these results are significant because they closely approach the previously reported activity, selectivity, and styrene yield reported for AX-21 (the best results reported to date for this reaction over a carbon catalyst) but here the halogenated additive was not required. Second, the early time yield of styrene was above 70%, while the later time yield was above 50%. Since the weight hourly space velocities were comparable to those of the industrial process, the space time yield levels are of commercial relevance. Third, not only was the rate of coking drastically suppressed, but the production rate of overoxidized product, namely, benzoic acid, also was suppressed to very low levels. Hence, the two primary reaction manifolds for yield loss were significantly diminished. Using air as the co-feed with a large excess of oxygen led to results similar to those of the Ambersorb adsorbents, but the deactivation phenomena were much slower. Under these conditions the ethylbenzene conversion and styrene yield never merge during the course of the long-time experiment. The Ambersorb catalysts are endowed with macro- and mesoporosity in addition to microporosity. Furthermore, the micropore size distributions of the Ambersorbs are quite similar to those of Carbosieve G. Thus, it is logical to infer that the “transport” porosity and added pore volume of the Ambersorbs significantly enhanced their lifetimes under the same conditions, since they could carry much more coke per gram before the transport rates of reactants and products were suppressed severely. The observed effect of the variation in weight hourly space velocity is consistent with the role of porosity in the behavior of the Ambersorb catalysts. As the velocity was raised from 0.05 to 0.1 to 0.2 h-1, the rate of deactivation rose. The higher space velocities simply accelerated the rate of deactivation, and so for the same time on stream the conversion and yield numbers converged more closely at the higher velocities. However, at 0.4 h-1 the behavior changed drastically. Although the fugacities of oxygen and ethylbenzene were the same and the temperature in the catalyst bed was the same, the catalyst was oxidized and consumed along with the ethylbenzene. This suggests that the rate of oxidation of the catalyst, a process that runs in parallel with the oxidation of the ethylbenzene, is externally mass(or heat)-transfer-limited at the lower space velocities, but at the higher space velocity it is not. At the higher space velocity the transport-limited rate of solid carbon oxidation began to run at a rate comparable to or higher than that of the hydrocarbon oxidation. Finally, when we compare the results of the Car-

bosieve G catalyst with those of the Ambersorb catalysts under the same conditions, the effect of transport porosity is immediately evident. An alternative factor that could also contribute is the presence of sulfur in the two Ambersorb samples (9 and 1 wt %, Ambersorb 563 and Ambersorb 572, respectively). However, the similar catalytic behavior of these catalysts despite their marked differences in sulfur content is indicative of little or no effect of sulfur. Furthermore, even though sulfur is well-known as a dehydrogenation reagent in organic synthesis (House, 1992), its behavior is stoichiometric and not catalytic. The results reported here are truly catalytic, so for sulfur to underlie the catalysis an unprecedented, new reaction mechanism would need to be responsible. Conclusions The conversion of ethylbenzene to styrene is an important industrial process technology. The use of carbon catalysts offers a potential new approach to oxidative dehydrogenation and the opportunity to utilize a catalyst which does not contain heavy metals. These results point to the real potential for this kind of catalyst system. Contrary to expectation, the carbogenic catalysts did not oxidize under the range of conditions examined; in fact, in most cases the catalysts tended to gain mass due to coke deposition. A number of conclusions can be drawn from the information reported here. All the nanoporous CMS catalysts deactivated rapidly with time on stream and a large excess of oxygen. The results for Carbosieve G indicate that it loses porosity due to filling of the pore structure with carbon residue, or coke. By utilizing a CMS which incorporates transport porosity, such as the Ambersorb, the effect of deactivation is considerably lower. The coupled reaction and diffusion processes that give rise to these differences in behavior can be quantified using the Thiele model. This analysis indicates that, for the microporous Carbosieve G, deactivation can be attributed to internal mass-transfer limitations caused by coking and narrowing of the micropores. Meso- and macroporosity in the Ambersorb adsorbents allow the reactants to have high effective diffusivities even after considerable coking of the pore structure and thus a deactivation behavior that is quite different. The most significant result was associated with switching from an air co-feed to 1% oxygen in inert. Under these conditions it is significant that the catalysts with transport porosity demonstrated low rates of deactivation, high selectivities, and high space time yields. The space time yields were high enough to approach the range of commercial relevance and are as high as the highest reported to date for carbogenic catalysts. In contrast, the Carbosieve G catalyst deactivated rapidly even under these more mildly oxidizing conditions. This work also points out the potential for operating the reaction in a thermally neutral regime, 1 mol of styrene produced via oxidation and 1 mol produced via dehydrogenation. The fact that we observe below the stoichiometric conversion of oxygen at 400 °C over the Ambersorb catalysts points to this as a real possibility well worth further investigation and verification. Acknowledgment H.C.F. acknowledges the National Science Foundation for partial support of this work along with grants from Rohm and Haas Co. and the Mobil Foundation.

Ind. Eng. Chem. Res., Vol. 35, No. 10, 1996 3331

Literature Cited Alkhazov, T. G.; Lisovskii, A. E. The Role of Condensation Products in the Oxidative Dehydrogenation Process of Ethylbenzene on Aluminum Oxide Catalyst. Kinet. Katal. 1976, 17, 434. Alkhazov, T. G.; Lisovskii, E. A.; Ismailov, Yu. A.; Kozharov, A. I. Oxidative Dehydrogenation of Ethylbenzene on Activated Carbons. I. General Characteristics of the Process. Kinet. Katal. 1978, 19, 611. Batist, P. H. A.; Lippens, B. C.; Schuit, G. C. A. The Catalytic Oxidation of 1-Butene Over Bismuth Molybdate Catalysts. J. Catal. 1966, 5, 55. Berger, P. A.; Roth, J. F. Electron Spin Resonance Studies of Carbon Dispersed on Alumina. J. Phys. Chem. 1968, 72, 3186. Cadus, L. E.; Arrua, L. A.; Gorriz, O. F.; Rivarola, J. B. Action of Activated Coke as a Catalyst: Oxydehydrogenation of Ethylbenzene to Styrene. Ind. Eng. Chem. Res. 1988, 27, 2241. Cadus, L. E.; Gorriz, O. F.; Rivarola, J. B. Nature of Active Coke in the Oxydehydrogenation of Ethylbenzene to Styrene. Ind. Eng. Chem. Res. 1990, 29, 1143. Clearfield, A.; Thakur, D. S. Zirconium and Titanium Phosphates as Catalysts: A Review. Appl. Catal. 1986, 26, 1. Drago, R. S.; Jurczyk, K. Oxidative Dehydrogenation of Ethylbenzene to Styrene over Carbonaceous Catalysts. Appl. Catal. A 1994, 112, 117. Emig, G.; Hofmann, H. Action of Zirconium Phosphate as a Catalyst for the Oxydehydrogenation of Ethylbenzene to Styrene. J. Catal. 1983, 84, 15. Emig, G.; Hofmann, H.; Schraut, A. Active Coke as an Organic Catalyst in the Oxydehydrogenation of Ethylbenzene. In Recent Trends in Chemical Reaction Engineering, Kulkarni, B. D., Mashlekar, R. A., Sharma, M. M., Eds.; Wiley:New York, 1987. Fiedorow, R.; Kania, W.; Nowinska, K.; Sopa, M.; Wojciechowska, M. Activity of Alumina Promoted by Inorganic Acids in the Process of Oxidative Dehydrogenation of Ethylbenzene. Bull. Acad. Pol. Sci., Ser. Sci. Chim. 1978, 26, 641. Fiedorow, R.; Przystajko, W.; Sopa, M. The Nature and Catalytic Influence of Coke Formed on Alumina: Oxidative Dehydrogenation of Ethylbenzene. J. Catal. 1981, 68, 33. Foley, H. C. In Perspectives in Molecular Sieve Science; Flank, W. H., Whyte, T. E., Jr., Eds.; American Chemical Society: Washington, DC, 1988. Foley, H. C.; Lafyatis, D. S.; Mariwala, R. K.; Sonnichsenn, G. C.; Brake, L. D. Shape Selective Methylamines Synthesis: Reaction and Diffusion in a CMS-SiO2-Al2O3 Composite Catalyst. Chem. Eng. Sci. 1995, in press. Grunewald, J.; Drago, R. S. Oxidative Dehydrogenation of Ethylbenzene to Styrene over Carbon-Based Catalysts. J. Mol. Catal. 1990, 58, 227. Guerrero-Ruiz, A.; Rodriguez-Ramos, I. Oxydehydrogenation of Ethylbenzene to Styrene Catalyzed by Graphites and Activated Carbons. Carbon 1994, 32, 23. Hong, A.; Mariwala, R. K.; Kane, M. S.; Foley, H. C. Adsorbate Shape Selectivity: Separation of the HF/134a Azeotrope over Carbogenic Molecular Sieve. Ind. Eng. Chem. Res. 1995, 34 in press. House, H. O. Modern Synthetic Reactions, 2nd ed.; Benjamin: Menlo Park, CA, 1972; p35. Iwasawa, Y.; Soma, M.; Onishi, T.; Tamaru, K. Catalytic Activities of Polynaphthoquinone Containing Various Metal Halides. J. Chem. Soc., Faraday Trans. 1 1972, 68. Iwasawa, Y.; Nobe, H.; Ogasawara, S. Reaction Mechanism for Styrene Synthesis over Polynaphthoquinone. J. Catal. 1973, 31, 444. Juntgen, H.; Knoblauch, K.; Harder, K. Carbon Molecular Sieves: Production from Coal and Application in Gas Separation. Fuel 1981, 60, 817.

Lafyatis, D. S.; Foley, H. C. Molecular Modeling of Shape Selectivity for the Fischer Tropsch Reaction Using a Trifunctional Carbon Molecular Sieving Catalyst. Chem. Eng. Sci. 1990, 45, 2567. Lafyatis, D. S.; Tung, J.; Foley, H. C. Poly(Furfuryl alcohol)Derived Carbon Molecular Sieves: Dependence of Adsorptive Properties on Carbonization Temperature, Time and Poly(ethylene glycol) Additives. Ind. Eng. Chem. Res. 1991, 30, 865. Lafyatis, D. S.; Mariwala, R. K.; Lowenthal, E. E.; Foley, H. C. Design and Synthesis of Carbon Molecular Sieves for Separation and Catalysis. In Synthesis of Microporous Materials: Expanded Clays and Other Microporous Solids; Occelli, M. L., Robson, H., Eds.; Van Nostrand: New York, Vol. II, p 318. Lee, C. S. (Mobil Oil). U.S. Patent No. 4 652 690, 1987. Lewis, P. J.; Hagopian, C.; Koch, P. Styrene. Kirk-Othmer Encyclopedia of Chemical Technology, 3rd ed.; Wiley: New York, 1983; Vol. 21, pp 770-801. Manassen, J.; Wallach, J. Organic Polymers. Correlation between their Structure and Catalytic Activity in Heterogeneous Systems. I. Pyrolyzed Polyacrylonitrile and Poly(cyanoacetylene). J. Am. Chem. Soc. 1965, 87, 2671. Manassen, J.; Khalif, S. H. Organic Polymers: Correlation between Their Structure and Catalytic Activity in Heterogeneous Systems. II. Oxidative Dehydrogenation: A Comparison between the Catalytic Activity of an Organic Polymer and that of Some Molybdate Catalysts. J. Catal. 1969, 13, 290. Mariwala, R. K.; Foley, H. C. Calculation of Micropore Sizes in Carbogenic Materials from the Methyl Chloride Adsorption Isotherm. Ind. Eng. Chem. Res. 1994a, 33, 2314. Mariwala, R. K.; Foley, H. C. Evolution of Ultramicroporous Adsorptive Structure in Poly(furfuryl alcohol)-Derived Carbogenic Molecular Sieves. Ind. Eng. Chem. Res. 1994b, 33, 607. Murakami, Y.; Iwayama, K.; Uchida, H.; Hattori, T.; Tagawa, T. Study of the Oxidative Dehydrogenation of Ethylbenzene. I. Catalytic Behavior of SnO2-P2O5. J. Catal. 1981, 71, 257. Murakami, Y.; Iwayama, K.; Uchida, H.; Hattori, T.; Tagawa, T. Screening of Catalysts for the Oxidative Dehydrogenation of Ethylbenzene. Appl. Catal. 1982, 2, 67. Roth, J. F.; Schaefer, A. R. Belgian Patent 682 863, 1966. Roth, J. F.; Schaefer, A. R. U.S. Patent 3 446 865, 1969. Roth, J. F.; Abell, J. B.; Fannin, L. W.; Schaeffer, A. R. Catalytic Dehydrogenation of Higher Normal Paraffins to Linear Olefins. Symposium on Refining Petroleum for Chemicals, Presented before the Division of Petroleum Chemistry and the Division of Industrial and Engineering Chemistry, American Chemical Society, New York City Meeting, 1969; Paper No. D 146. Satterfield, C. N. Heterogeneous Catalysis in Industrial Practice, 2nd ed.; McGraw Hill: New York, 1991; p 479. Schraut, A.; Emig, G.; Sockel, H.-G. Composition and Structure of Active Coke in the Oxydehydrogenation of Ethylbenzene. Appl. Catal. 1987, 29, 311. Vrieland, G. E. Oxydehydrogenation of Ethylbenzene to Styrene over Metal Pyrophosphates. 1. Catalyst Composition and Reaction Variables. J. Catal. 1988a, 111, 1. Vrieland, G. E. Oxydehydrogenation of Ethylbenzene to Styrene over Metal Pyrophosphates. 2. Microbalance Studies of Carbon Deposition and Burnoff. J. Catal. 1988b, 111, 14.

Received for review December 15, 1995 Accepted May 23, 1996X IE950762G

X Abstract published in Advance ACS Abstracts, July 15, 1996.