γ-Al2O3 Bitumen Hydroprocessing

3940

Ind. Eng. Chem. Res. 1996, 35, 3940-3950

Initial Coke Deposition on a NiMo/γ-Al2O3 Bitumen Hydroprocessing Catalyst Susan M. Richardson,† Hiroshi Nagaishi,‡ and Murray R. Gray* Department of Chemical Engineering, University of Alberta, Edmonton, Alberta, Canada T6G 2G6

Athabasca bitumen was hydrocracked over a commercial NiMo/γ-Al2O3 catalyst in two reactors, a microbatch reactor and a 1-L continuous stirred tank reactor (CSTR). Coke deposition on catalyst was measured as a function of hydrogen pressure, time on stream, and liquid composition by measuring the carbon content of the cleaned spent catalyst. The carbon content ranged from 11.3% to 17.6% over the pressure range 6.9-15.2 MPa in CSTR experiments. Batch and CSTR experiments showed a rapid approach to a constant coke content with increasing oil/catalyst ratio. Coke deposition was independent of product composition for residue concentrations ranging from 8% to 32% by weight. Removal of the coke by tetralin at reaction conditions suggested reversible adsorption of residue components on the catalyst surface. A physical model based on clearance of coke by hydrogen in the vicinity of metal crystallites is presented for the coke deposition behavior during the first several hours of hydrocracking use. This model gives good agreement with experimental data, including the effect of reaction time, the ratio of total feed weight to catalyst weight, hydrogen pressure, and feed composition, and it agrees with general observations from industrial usage. The model implies that except at the highest coke levels, the active surfaces of the metal crystallites remain exposed. Severe mass-transfer limitations are caused by the overall narrowing of the pore structure, which in γ-Al2O3 would give very low effective diffusivity for residuum molecules in micropores. Introduction Primary upgrading of bitumen and heavy oils has historically been accomplished through two types of technology: thermal coking and hydroprocessing. In recent years, hydroprocessing has become the technology of choice due to higher yields and greater heteroatom removal (Beaton and Bertolacini, 1991). In hydroprocessing, high temperatures cause thermal cracking reactions to occur, while parent molecules and cracked products are subject to catalytic hydrogenation and heteroatom removal. The use of this technology has been hindered by high catalyst costs, incurred due to rapid deactivation of the catalyst. Two major deactivation pathways are recognized: initial coking and deposition of heavy metals (Thakur and Thomas, 1985). Hydroprocessing and hydrotreating catalysts typically accumulate carbonaceous deposits over the first hours or days of service. Coke content, therefore, goes through a rapid initial increase, which then asymptotically approaches a stable value (Oballa et al., 1994). This initial coke deposition is often credited with the deactivation that occurs during the same period (Oballa et al., 1994; Hannerup and Jacobsen, 1983). Given this possible link, understanding the phenomenon of coke deposition on hydrocracking catalysts is important. In general, an increase in hydrogen pressure reduces coke deposition on the catalyst (Thakur and Thomas, 1985). Aitken et al. (1964) developed an empirical relation between the deactivation rate and hydrogen pressure in the hydrocracking of bitumen, which showed that the deactivation rate decreased with increasing hydrogen pressure. The deactivation rate varied pro* Author to whom correspondence should be addressed. † Current address: Syncrude Canada Ltd. Research Centre, 9421-17 Ave., Edmonton, Alberta, Canada T6N 1H4. ‡ Current address: Macromolecular Science Section, Resources and Energy Division, Hokkaido National Industrial Research Institute, 2-17-2-1 Tsukisamu-Higashi, Toyohira-Ku, Sapporo 062, Japan.

S0888-5885(95)00761-5 CCC: $12.00

portionally with the hydrogen pressure raised to the -3 power. The authors did not measure the carbon content of the catalyst; therefore, the link between hydrogen pressure and coke content was unknown. Inoguchi et al. (1972) considered the effect of hydrogen flow rate on coke deposition in a packed bed reactor. They found that coke deposition decreased with increasing hydrogen flow rate. Although they did not measure coke deposition as an explicit function of hydrogen pressure, it was suggested that the data could best be explained by the increase in hydrogen partial pressure with increasing gas flow rates, although concomitant changes in liquid holdup were not considered. Similarly, Zeuthen et al. (1995) attributed higher end-of-run coke contents in the third stage of a three-stage pilot plant to lower hydrogen partial pressure in that stage. Douwes et al. (1979) suggested that lower hydrogen partial pressure will result in an increase in deposited coke. A decrease in coke laydown with increasing hydrogen partial pressure has been widely observed; however, no mechanistic or empirical relation has been proposed for this phenomenon. Literature available on catalyst deactivation and coke deposition in other petroleumrefining processes, such as fluid catalytic cracking, cannot be transferred to hydroprocessing of residue, because of the high hydrogen pressure, the presence of a liquid phase, and different catalyst composition. One of the barriers to accounting for the role of hydrogen is a lack of consensus on the chemical nature of coke and its physical distribution in the catalyst. Two modes of coke deposition have been suggested within catalyst pellets: pore mouth plugging and uniform surface deposition. Muegge and Massoth (1991) studied effective diffusivity in catalysts which had been artificially coked with anthracene/n-heptadecane feeds. At extremely low coke deposition, on the order of 2-3 wt % carbon, the effective diffusivity was significantly reduced. While their observations were consistent with plugging of pore mouths, their experiments may not give the coke deposition one might expect from residue feeds. © 1996 American Chemical Society

Ind. Eng. Chem. Res., Vol. 35, No. 11, 1996 3941

Absi-Halabi et al. (1991) found that the initial coke deposition was most significant in the smallest pores. Coal liquefaction with a catalyst under comparable conditions gave almost complete loss of pore volume in the 2-5-nm pore radius range with no loss in the 5-50nm pore radius range (Thakur and Thomas, 1984). Since the ratio of surface area to pore volume is highest in the smallest pores, uniform deposition would give the largest deposition of coke in these pores. Ternan et al. (1979) studied coke deposition on alumina-supported molybdenum catalysts promoted with various cations. They measured constant values of coke deposition per unit surface area for three aluminas of different surface area. No significant effect of promoter cation was observed. These results indicated uniform coverage. Using the properties of petroleum coke from coking processes, they determined that the observed coke content corresponded to 0.5 monolayer for gas oil hydroprocessing and 1.1 monolayers for Athabasca bitumen hydroprocessing. They observed 96% sulfur conversion after the deposition of 1.1 monolayers of coke and concluded this level of conversion was not consistent with complete loss of hydrogenation activity of the catalyst. This conclusion is supported by the results from thermal experiments (Ayasse, 1994). Ternan et al. (1979) suggested that the coke deposited on the surface may be continuously exchanged with the bulk, in order to account for the observed activity of the catalyst in the apparent presence of a monolayer. Diez et al. (1990) proposed that coke forms in a uniform layer on the γ-alumina support but is thinner near the NiMo metal crystallites due to their hydrogenation activity. As the amount of coke increases, however, the coke layer may occlude the active side surfaces of the metal crystallite, ultimately leading to a catastrophic loss of activity. Melo Faus et al. (1984) used model compounds on artificially coked catalyst to show that isomerization decreased more rapidly than hydrogenation as a function of coke deposition, indicating a differential effect on active sites. In coal liquefaction studies by Nishijima et al. (1987), hydrocracking activity was reduced more significantly than hydrogenation with increasing coke deposition. If coke was deposited primarily at the pore mouth, the deactivation rates of all catalyst functions would be expected to be equal. Conflicting reports on the probable location of coke deposits in γ-Al2O3 hydroprocessing catalysts suggest that the deposition process may depend on the compositions of feed and catalyst and the nature of the porous network. Several authors suggest that coking of the catalyst surface may be linked to catalyst acidity. Gray et al. (1992) concluded that γ-Al2O3 without supported metal promoted severe coking under hydrocracking conditions due to the acidic functionalities of the alumina. AbsiHalabi et al. (1991) emphasized the importance of acidbase interactions in coking of hydrocracking catalysts, while Korre et al. (1995) determined that adsorption of hydrocarbon on a CoMo/γ-Al2O3 catalyst surface was correlated to the gas-phase basicity of the component. This study considers three main questions. What is the dependence of coke deposition on hydrogen pressure? How does the coke distribution on the catalyst surface change with hydrogen pressure? What are the mass-transfer implications that arise from the deposition of coke on the catalyst? In order to answer these questions, experiments were performed using Athabasca bitumen and a commercial NiMo/γ-Al2O3 catalyst in batch and CSTR reactors. These experiments were supplemented by model compound studies and characterization of the catalyst before and after reaction. The

Table 1. Characteristics of the Feed and Selected Products product C, % H, % S, % N, % naphtha (IBP-195 °C) mid distill (195-343 °C) gas oil (343-524 °C) residue (524+, °C) density, g/mL MCR, %

bitumen

1

2

3

4

82.7 10.2 4.9 0.45 0 7.0 38.0 55.0 1.008 13.7

83.6 10.9 2.6 0.44 9.9 12.0 37.9 32.2 0.959 9.9

86.0 11.4 1.6 0.38 11.8 30.2 36.8 21.3 0.927 7.6

86.8 11.8 0.59 0.29 20.2 42.4 28.9 8.5 0.888 2.3

86.8 12.5 0.12 0.10 28.3 58.8 12.2 0.74 0.847 0.1

data were used to construct a physical model of the coke deposition occurring on hydroprocessing catalyst during the first few hours of operation. Experimental Section Materials. Athabasca bitumen feed was provided by Syncrude Canada Ltd., with the properties listed in Table 1. Coker gas oil (CGO) was supplied by Syncrude, with properties described in Yui and Sanford (1991). Tetralin was supplied by Fluka, carbon disulfide and acridine were supplied by Aldrich, anthracene was supplied by Sigma, and dibenzofuran was supplied by Eastman Kodak. The dibenzofuran was purified by dissolving the chemical in dichloromethane, mixing with activated carbon, filtering the solution, and recrystallizing the dibenzofuran. The catalyst was a commercial NiMo/γ-Al2O3 catalyst with 12.5% MoO3 and 3.5% NiO. The catalyst had a surface area of 317 m2/g (by nitrogen BET measurement) and a pore volume of 0.57 mL/g. The catalysts were used in the oxide form, except for one run in which the catalyst was presulfided using CS2. For the batch experiments, the catalyst was crushed and sieved to +350- to -500-µm size. In the CSTR experiments, the catalyst was used in the 1-mm cylindrical extrudate form in which it was delivered. Analysis Techniques. Feed and liquid product characterization was performed at Syncrude Research, Edmonton, Alberta, Canada. The distillation was performed using a atmospheric spinning band distillation to obtain the naphtha (initial boiling point - 177 °C) and middle distillate (177-343 °C) fractions. The remaining oil was then separated into gas oil (343-524 °C) and residue (524+ °C) fractions using the ASTM D1160 procedure (Anonymous, 1996). The overall sulfur content was determined by combustion followed by fluorescence detection. Carbon and hydrogen analysis of the petroleum liquids used a Leco 600 analyzer. Nitrogen analysis used combustion followed by chemiluminescence detection. Microcarbon residue (MCR) content was determined using an Alcor MCR analyzer. The average molecular weight was determined by the Micro Analytical Laboratory at the University of Alberta by vapor pressure osmometry using p-dichlorobenzene as solvent. Prior to carbon determination, catalysts were Soxhlet extracted for 6 h in dichloromethane and then vacuum dried at 65 °C and 10 kPa for 2 h. In order to ensure representative sampling, a 50-mg portion of the catalyst was crushed to a powder and a 2-mg sample used for analysis. The concentration of carbon in the catalyst was measured at the Micro Analytical Laboratory, University of Alberta. This analysis, performed on a Carlo Erba Stumentazione Elemental Analyzer 1108, used catalytic combustion of the sample, followed by gas

3942 Ind. Eng. Chem. Res., Vol. 35, No. 11, 1996

chromatographic separation of the combustion productions and analysis by thermal conductivity detection. The elemental carbon content on a spent catalyst basis is used to represent the coke content of the catalyst. Measurement of the distribution of the strength of acid sites on the surface of the catalyst was performed at Guelph Chemical Laboratories by stepwise thermal desorption. The catalyst was packed into a stainless steel column which was attached to a Hewlett-Packard gas chromatograph (5830A) equipped with a flame ionization detector. The catalyst was saturated with pyridine, and the excess was desorbed in flowing nitrogen at 373 K. The pyridine which was irreversibly adsorbed at 373 K was subjected to a stepwise increasing temperature. The temperature was raised at 15 K min-1 until the desired temperature was reached (see following temperature ranges); then, the temperature was maintained until the desorption of pyridine stopped. This process was repeated for each temperature step. Four categories of strengths of acid sites were reported based on the following desorption temperature steps: (i) very weak, 373 K < Td e 423 K; (ii) weak, 423 K < Td e 523 K; (iii) strong, 523 K < Td e 623 K; (iv) very strong 623 K < Td. Category iv was the material irreversibly adsorbed at 623 K, determined by difference. The overall temperature range from 100 to 350 °C was selected to range from a convenient initial temperature, below the boiling point of pyridine, to an upper limit set to avoid reaction of pyridine and irreversible changes in the catalyst structure. The divisions into temperature ranges are based on even increments of temperature, rather than an absolute measure of the acidity of surface sites. The value of the method is to compare between samples, rather than to obtain absolute data on the distribution of the strength of acid sites (Andersen and Pratt, 1985). The surface area, pore volume, and pore size distribution of the catalyst were determined in the Department of Chemical Engineering, University of Alberta, on an Omnisorb 360 (Coulter Scientific Instruments, Hialeah, FL). Nitrogen adsorption measurements were used to calculate surface areas, using the BET equation (Gregg and Sing, 1967). Pore size distributions and pore volume were determined using nitrogen desorption data and the Kelvin equation (Gregg and Sing, 1967). Reactors and Reaction Procedures. Two reactors were used for this study. Fifteen-milliliter microbatch reactors were constructed from stainless steel tubing and Swagelok tube fittings and replaced as required. CSTR experiments were performed in a 1-L stirred reactor with continuous gas and oil feed which has been described in detail previously (Gray et al., 1995). No noncatalytic runs were performed in either the batch reactor or the CSTR. Microbatch bitumen experiments were carried out in multistep sequences. The sequences were begun with fresh catalyst and bitumen. After reaction, the catalyst was reclaimed by extraction and was used with fresh bitumen for the next step. The sequences consisted of two to six repeated steps of extraction and reaction. For each reaction, approximately 0.75 g of catalyst and 3 g of bitumen were loaded into the reactor. The reactor was sealed, purged with nitrogen, and then pressurized with hydrogen. The hydrogen pressure at room temperature was adjusted to allow the pressure to reach 13.8 MPa at the operating temperature of 430 °C. The reactor was immersed in a fluidized sand bath heater and allowed to remain at temperature for the desired time, typically 1 h. After the reaction was completed, the reactor was cooled quickly in water, and the off

Table 2. Experiments with Model Compound Feeds in Batch Reactor with 1 h of Reaction Time at 430 °C and 13.8 MPa

catalyst

feed

fresh spent spent fresh fresh fresh fresh

tetralin (blank) tetralin tetralin 2.5% anthracene in tetralin 3.2% dibenzofuran in tetralin 3.3% acridine in tetralin 6.5% acridine in methylnaphthalene Athabasca maltenes coker gas oil

fresh spent

no. of reaction steps

final C content, %

5 1 6 6 6 6 2

2.6 11.1 6.6 1.3 1.8 1.1 2.1

4 2

3.8 11

gases were vented. A liquid product sample was decanted if necessary, and the remaining oil and catalyst were washed out of the reactor with dichloromethane. The catalyst was Soxhlet extracted in dichloromethane for 6 h or until the extract ran clear, whichever was longer. The catalyst was vacuum dried for 2 h at 65 °C and 0.1 atm. The catalyst was then recycled for another reaction or sampled for analysis. One sequence of experiments was performed with deasphalted oil. Athabasca bitumen was mixed with toluene in the ratio 1 mL of toluene/g of sample and then diluted with 50 vol of pentane. The mixture was shielded from light and stirred for 16 h. The precipitate was removed using a 0.22-µm Millipore filter. The liquid portion was dried to constant weight, using a rotary evaporator and vacuum drying at the conditions listed above. The model feeds were prepared by mixing the desired amounts of the model compound with tetralin. The concentrations are given in Table 2. These concentrations were chosen based on the expected coke deposition and number of planned reaction steps. In the case of anthracene, a suspension resulted. A sample of this suspension was brought to the reaction temperature of 430 °C and then cooled. Inspection of the liquid showed that the suspension had become extremely finely dispersed, indicating that a solution had formed at operating conditions then reprecipitated upon cooling. One sequence was performed with acridine in 1-methylnaphthalene to explore possible solvent effects. The model feed experiments used 1 g of liquid. CS2 was added to maintain the catalyst in a sulfided form. Four percent CS2 by weight was used in experiments using the fresh oxide form of the catalyst, while the concentration was reduced to 2% for subsequent experiments using the same catalyst. Two experiments used previously coked catalyst with tetralin as a solvent to study the reversibility of coke deposition. Coker gas oil was also used with previously coked catalyst to explore this phenomenon. Two groups of CSTR experiments were performed. Runs 1 through 7 (see Table 3) were single-stage experiments at 430 °C. In these experiments, 78 g of fresh catalyst was loaded into the reactor, the reactor was heated, and then the hydrogen and oil inlets were opened. The hydrogen flow rate was 4.75 std L/min, and the bitumen flow rate was 400 g/h. The reactor was brought to steady state and then run at the desired conditions for the desired time. The reaction times listed in Table 3 represent the total reaction time, including the unsteady-state period. Runs 8 through 12 were experiments with changing feed and feed rate, given in Table 3, but the other operating procedures remained as described above. In this group, the feed

Ind. Eng. Chem. Res., Vol. 35, No. 11, 1996 3943 Table 3. Summary of Reaction Conditions and Carbon Content for CSTR Experimentsa run

feed

feed rate, g/h

pressure, MPa

feed ratio, g oil/g cat.

C, %

time on stream, h

1 2 3 4 5 6 7 8 9 10 11 12 13 14

bitumen bitumen bitumen bitumen bitumen bitumen bitumen bitumen product oil 1 product oil 2 product oil 2 product oil 3 bitumen bitumen

400 250 400 400 400 400 400 1210 960 370 370 176 400 435

13.8 13.8 6.9 10.3 8.3 11.7 15.2 13.8 13.8 13.8 13.8 13.8 13.8 13.8

8.1 49.7 17.9 30.7 30.8 30.8 30.8 117.5 107.4 41.5 39.6 30.2 35.9 66.6

9.4 12.8 17.5 16.7 16.4 14.8 11.3 12.9 14.8 11.2 14.1 6.9 14.6 15.2

1.5 15.5 3.5 6 6 6 6 7.4 9.1 8.1 7.7 11.9 7.0 13.0

a

All experiments were at 430 °C.

Figure 1. Carbon content of the catalyst as a function of cumulative feed oil-to-catalyst ratio. All experiments were at 13.8 MPa, 430 °C. Data for batch reactor are for repeated contacts of catalyst with fresh feed; data for CSTR are from experiments 1, 2, 8, 13, and 14. The model curve is from eq 10.

Figure 2. Carbon content of the catalyst as a function of residue concentration in the reactor. All experiments were at 13.8 MPa, 430 °C. The solid line represents the average carbon concentration observed at intermediate residue concentrations (data from runs 8-12).

to run 9 was the total liquid product from the steadystate portion of run 8. Similarly, the product from run 9 was used as feed for runs 10 and 11, and the liquid product from these runs was used as feed for run 12. This mode of operation simulated sequential reactors with hydrogen gas makeup between reactors.

Conversion of the residue to distillate fractions decreased the concentration of residue in the feed. The composition of the residue which was not sufficiently cracked to be converted to distillate also changed, becoming more aromatic (Nagaishi et al., 1996). The data for coke content of the catalysts showed insensitivity to residue concentration at concentrations above 8% (Figure 2). The concentration of carbon on the catalyst was plotted against product composition since the reactor was well stirred. Despite the number of compositional variables that were varied by successive cracking of the residue, the carbon content of the catalyst was relatively constant at 13 ( 1% over the range of residue concentrations of 8-32 %. Table 1 also shows the MCR content of the liquid products from the multistage hydrocracking experiments. Although MCR data are proportional to coke yields in cokers, catalysts exposed to reactor liquids containing 2-10 wt % MCR gave relatively constant coke contents (Table 1). The MCR content of the product, therefore, did not correlate with coking of the catalyst. The hydrogen-to-carbon atomic ratio of the coke was found to range between 1.0 and 1.2, with one outlying value at 1.78 (see Table 4). The hydrogen-to-carbon atomic ratio of the residue fraction of the product ranged from 1.0 to 1.3. No systematic trend was observed between the residue content of the reactor liquid and the hydrogen-to-carbon atomic ratio of the carbonaceous deposit on the catalyst. The analytical accuracy of the hydrogen analysis was (0.2% absolute. Analysis of the coke deposit was based on total catalyst weight, and hydrogen contents were less than 2% in all cases, which

Results and Discussion Deposition of Coke with Time and Feed/Catalyst Ratio. A two-step reaction protocol was used to measure the dependence on reaction time. Two cycles of reaction and extraction were performed (as described in the Experimental Section), with variable reaction time. The total times varied from 1.5 to 5 h, but coke deposition was relatively constant based on a carbon content of 11.7 ( 0.95 wt %. The data displayed in Figure 1 show the carbon content as a function of feed oil-to-catalyst ratio. These data suggest a rapid buildup of coke on the catalyst, followed by constant coke content. The batch reactor gave a catalyst with a constant coke content between 15 and 20 g/g, while the catalyst in the CSTR reached a constant value between 10 and 30 g/g oil-to-catalyst ratio. This interpretation assumes a constant proportionality between coke content and the measured concentration of carbon. Dependence of Coking on Residue Concentration. The CSTR reactor was operated to simulate multiple reactors in series with hydrogen makeup, producing products of varying residue contents, in the range 0.7-32% (Table 1). This repeated hydroprocessing gave two changes in liquid-phase composition.

3944 Ind. Eng. Chem. Res., Vol. 35, No. 11, 1996 Table 4. Product Composition and Hydrogen-to-Carbon Ratios of Selected CSTR Catalystsa product residue sample

pressure, MPa

product residue, wt %

H/C ratio

cat. H/C ratio

8 9 10 12 15

13.8 13.8 13.8 13.8 13.8

32.2 21.3 8.5 0.7 20.4

1.33 1.22 1.06 1.18 n/a

1.21 1.01 1.1 1.78 n/a

a Sample numbers refer to Table 3, which includes operating conditions.

Figure 3. Carbon content of the catalyst as a function of hydrogen pressure for CSTR experiments. All experiments were at 430 °C. The curve is from eq 6, with Cmax ) 17.5 and k ) 0.15.

gave an uncertainty in the hydrogen-to-carbon atomic ratio of approximately 0.2 for the coke deposit. The variation in the H/C ratio of the deposit was, therefore, within the range of analytical error. The outlying datum may have been due to unusually high levels of water adsorption, since the catalysts were not dried immediately before analysis. The hydrogen-to-carbon atomic ratios of the coke deposits were approximately equal to those of the residue fraction of the product liquids, given the analytical accuracy. These data showed that the carbon content was a useful indicator of total coke content. Dependence of Coke Deposition on Hydrogen Pressure. The hydrogen pressure in the CSTR was varied by changing the total pressure, keeping the gas flow rate (measured in standard liters) constant. The data for the coke content of the catalyst showed a small but statistically significant increase with decreasing hydrogen pressure (Figure 3). Two points are included in this data set which have oil-to-catalyst ratios different from the standard. The experiment performed at 13.8 MPa (run 2) had a greater oil-to-catalyst ratio than the standard conditions, but given the constant coke content observed at high feed-to-catalyst ratios observed in Figure 1, the inclusion of this point is justified. The 6.9-MPa experiment (run 3) had a lower oil-to-catalyst ratio than the standard. It is possible that this catalyst was still experiencing transient coke deposition and therefore underrepresents the steady-state coke content at this pressure; however, the data of Figure 1 suggest that an oil/catalyst ratio of 17.9 g/g was sufficient to approach constant concentration, so the point was included in the figure and calculations. The increase in coke with decreasing hydrogen pressure was sufficient to explain the difference in coke deposition between the batch reactor and the CSTR, illustrated in

Figure 1. The data of Figure 3 suggest that the batch reactor operated at an effective pressure of 10.3 MPa of H2, consistent with expected hydrogen consumption from the initial pressure of 13.8 MPa. The data for coke deposition in the two reactors were consistent, therefore, if the changes in hydrogen pressure in the batch reactor are taken into account. During the CSTR experiment at 6.9 MPa of hydrogen pressure, particulate coke was formed in the liquid phase, plugging the outlet lines from the reactor, unlike experiments at other conditions in which no such plugging problems were observed. When the reactor was opened, no agglomerated coke deposits were observed and the carbon content on the catalyst was consistent with the results at higher hydrogen pressure (Figure 3). Given that the coke deposition on the catalyst was a smooth monotonic function of pressure, while the liquid-phase coking showed a step change, these phenomena appear to be independent. The coking mechanism in the liquid phase, probably linked to thermal cracking, was probably completely distinct from the coking mechanism on the catalyst surface. If precipitation of asphaltenes were responsible for coking of catalyst, as suggested by Absi-Halabi et al. (1991), then it is unreasonable to expect that the coke content of the catalyst would not reflect a step change in liquidphase coking. Kinetic Model of Coking on Catalyst. Based on the data described above, a model must incorporate three observations: coke deposition on the catalyst that approaches a constant value with increasing feed oilto-catalyst ratio, the decrease in coke deposition with increasing hydrogen pressure, and the insensitivity of the coke concentration to composition in the reactor. The maximal deposition of coke on the catalyst surface occurred at the lowest hydrogen pressure, as illustrated in Figure 3. Could the maximum amount of coke correspond to monolayer coverage by residue molecules? This hypothesis can be tested by a calculation based on the assumption that the substance adsorbed on the surface is similar in molecular weight, bulk density, and carbon content to the residue fraction of the oil. The resid fraction of cracked Athabasca bitumen had a measured molecular weight of 625, a carbon content of 82 wt %, and a typical bulk liquid density of 1050 kg/ m3 (Gray, 1994). Given these values, the radius for a typical resid molecule would be given by

rM )

(

)

3MW NAF4π

1/3

(1)

where rM is the radius of a residue molecule (m), MW is the molecular weight, NA is Avogadro’s number (6.02 × 1026 molecules/kmol), and F is the density of oil (kg/ m3). Since molecules typically found in the residue consist of aromatic cores linked by aliphatic bridges, they can be assumed to be deformable (Strausz, 1989). This assumption is extended, for mathematical simplicity, to the assumption that they are sufficiently deformable to allow 100% surface coverage. This assumption results in a carbon content at monolayer coverage of

Mmax )

MWxC NAπrm2

(2)

where Mmax is the maximum carbon loading (g of carbon/ m2 of surface) and xC is the mass fraction of carbon in the residue. Substituting the properties of Athabasca residue into eqs 1 and 2 gives a predicted loading of 7.1

Ind. Eng. Chem. Res., Vol. 35, No. 11, 1996 3945

Figure 4. Schematic representation of catalyst surface, showing a cross section through metal crystallite.

× 10-4 g of carbon/m2. The maximum concentration observed in experiments was 18% carbon, corresponding to a loading of 6.9 × 10-4 g of carbon/m2, consistent with the hypothesis of monolayer coverage of the catalyst surface by residue molecules. The sensitivity of the predicted carbon content on the catalyst due to monolayer coverage of residue was determined as a function of residue properties. Changing the residue density by 10% resulted in a 10% change in the monolayer carbon content. Similarly, changing the assumed carbon content of the adsorbed molecules from the base case of 82% to values of 80% and 85% resulted in only a 3% change in the carbon content of the monolayer. Increasing the assumed molecular mass from 625 (measured) to 1000 Da caused an increase of 17% in the predicted carbon content of a monolayer. Despite these large changes in residue properties, outside the bounds of normal petroleum chemistry, changes in the predicted carbon content were relatively small. The development of the model which follows assumes that the coke deposits on the surface in a uniform manner, similar to the model discussed by Diez et al. (1990). The maximum coke which can be accumulated on the surface is equal to monolayer coverage, an estimate of which is presented above. Other authors have proposed pore mouth plugging models to describe coking of similar catalysts. These models will be compared in later sections. The data show a weak but definite downward trend in coke content with increasing hydrogen pressure, which indicates that the surface adsorption was not simply a function of residue concentration. Diez et al. (1990) hypothesized that a zone near the metal sulfide crystallites would have a reduced carbon content due to the hydrogenating activity of the crystallites. Extending their model, we propose that the active hydrogen species formed by metal crystallites will clear an annular zone on the alumina surface, as illustrated in Figure 4. Outside of the cleared zone, the carbonaceous layer is of uniform thickness due to monolayer adsorption. Inside the cleared zone, no coke remains on the surface. This assumption cannot be proven by direct observation but will be discussed in a later section. The distance to the boundary of the cleared zone is determined by the surface flux of hydrogen species so that the alumina will remain clear if the surface flux is greater than a critical value. From the geometry illustrated in Figure 4, the actual loading on the surface as a function of the size of the cleared zone will be given by

M ) Mmax(1 - Nπ(r2 - rc2))

(3)

where M is the actual loading (g of carbon/m2), rc is the

radius of the metal crystallite (m), N is the number of crystallites (crystallites/m2), and r is the radius of the cleared zone (m). Generation of surface hydrogen at the crystallite is assumed to be proportional to hydrogen pressure. Other relationships have been evaluated but found to be less satisfactory than the simple first-order dependence of pressure (Richardson, 1996). The rate of hydrogen leaving the cleared region by surface diffusion into the adsorbed monolayer will be the product of the critical flux of hydrogen and the circumference of the cleared zone. The rate of reaction of hydrogen within the annular zone will be equal to the rate of deposition of hydrocarbon times an unknown stoichiometric ratio for the reaction of hydrocarbon and hydrogen species times the area of the zone. The mass balance on hydrogen for the annular zone, therefore, is given as

kHPH2 - 2πrFcrit - π(r2 - rc2)σkdCp ) 0

(4)

where kH is rate constant for production of surface hydrogen (mol MPa-1 s-1), PH2 is the pressure of hydrogen (MPa), Fcrit is the critical surface flux of hydrogen (mol m-1 s-1), σ is the stoichiometric coefficient for reaction of surface hydrocarbon with hydrogen, kd is the rate constant for deposition of hydrocarbon (m-1 s-1), and Cp is the concentration of coke precursor (mol/m2). Solving eq 3 for r and then substituting into (4) and simplifying gives a function of the form

(

PH2 ) a 1 -

)

M +b Mmax

1/2

(

+c 1-

M Mmax

)

(5)

where a, b, and c are functions of the parameters from eqs 3 and 4:

a)

2π1/2Fcrit kHN1/2

b ) Nπrc2 c)

σkdCp kHN

In this form, the first term of the equation includes the production of hydrogen at the crystallite and the hydrogen leaving by surface diffusion. The second term gives the consumption of hydrogen necessary to maintain the cleared surface, i.e., the pressure of hydrogen to remove coke which deposits in the cleared region. Equation 5 was fitted to the data from runs 2 through 7 using the least-squares method. The equation gave a good fit to the data, but parameter c was negative, indicating a negative reaction rate. It was not possible to estimate actual values for c, given the number of unknown kinetic parameters included in its definition of eq 5. We assumed, therefore, that the rate of consumption of active hydrogen by reaction was negligible compared to the magnitude of the flux terms; i.e., c , a. This assumption results in eq 6, below, which

(

PH2 ) a 1 -

)

M +b Mmax

1/2

(6)

does not include the term due to reaction within the zone. Despite removal of one of three adjustable parameters, the sum of the squared residuals had

3946 Ind. Eng. Chem. Res., Vol. 35, No. 11, 1996

increased by less than 10%; therefore, the model equation (6) is also consistent with the data. The values of the best fit parameters for this equation were a ) 22.8 MPa (coefficient of variation ) 6.4%) and b ) 0.093 m2 of metal/m2 of surface (coefficient of variation 29%, covariance ) 0.76). These values were used to calculate the solid line in Figure 3. The value of parameter b in eq 6 is given by

b ) πNrc2

(7)

Data for the number and size of crystallites from the work of Madeley and Wanke (1986), for a similar sulfided NiMo catalyst, gave b ) 0.036 m2 of metal/m2 of surface, compared to 0.09 m2 of metal/m2 of surface from the regression of the model to the experimental data. The value for parameter b from the regression was, therefore, physically reasonable. Variations between the experimental value and the literature value were probably the result of variations in catalyst preparation and sulfidation conditions, both of which are known to affect metal dispersion, which the coefficient b reflects (Madeley and Wanke, 1986). Parameter a contains the variables k and Fcrit, neither of which can be evaluated independently; therefore, the value of a cannot be verified. The assumption that hydrogen diffuses across the alumina surface is consistent with the literature on hydrogen spillover from metal crystallites (Paal and Menon, 1988; Kapoor et al., 1989). Diffusion of hydrogen away from the catalyst surface into the liquid in the pores would result in an additional sink term in the mass balance. The diffusion into the bulk, however, would be expected to be proportional to the production of activated hydrogen at the crystallite, which in turn is proportional to the hydrogen pressure. Therefore, the mass balance term kHPH2 represents the net production of surface hydrogen. Hydrogen which was left on the catalyst surface is not expected to participate in coke removal; therefore, the amount of active hydrogen lost into the bulk liquid does not need to be quantified, and eq 4 represents the surface mass balance correctly. If residue adsorbs to form a monolayer, then a maximum level for deposition will exist for given conditions. This maximum amount can be represented by a hypothetical number of adsorption sites. The rate of adsorption therefore will depend on surface vacancies as follows:

Rdep ) -

dθv ) kθv dω

(8)

where Rdep is the rate of coke deposition (g of residue/g of catalyst), θv is the fraction of possible adsorption “sites” currently vacant, ω is the cumulative feed-tocatalyst ratio (g of feed/g of catalyst), a pseudotime coordinate, and k is an adsorption rate constant (g of catalyst/g of feed). The actual amount of carbon on the catalyst (C, wt %) would be related to the maximum carbon deposition (Cmax, wt %) and θv as

C ) 1 - θv Cmax

(9)

Substitution of eq 9 into eq 8 and solving indicates that the buildup of carbon will follow first-order kinetics:

C ) Cmax(1 - e-kω)

(10)

The solid line in Figure 1 was calculated using eq 10, which indicates good agreement between the data and the model. Below approximately 15 g of oil/g of catalyst, the full monolayer has not developed and capacity remains for further adsorption. Beyond this value of feed oil-to-catalyst ratio, there may be reversible exchange of surface hydrocarbon with bulk hydrocarbon, but since the layer is filled, no net change in coke content is observed. Other models for transient behavior may predict similar functions; however, this simple extension of the steady-state model is consistent with the data observed. The surface deposition of coke was consistent with the model predictions for the effect of product composition. Equation 6, which satisfactorily describes the experimental data, includes no explicit dependence of surface coverage on the concentration of coke precursors in the liquid. Experimentally, the coke content was very insensitive to composition. Only when the residue concentration fell below 1% total liquid product did the coke concentration on the catalyst decrease significantly. This observation may have been due to limitations on the amount of adsorbate available. In run 12, the catalyst was only exposed to enough residue in the reactor liquids during the entire experiment to form a monolayer; therefore, unless the capture of residue by adsorption was 100% efficient, a monolayer could not have been achieved. This model is consistent with the data reported by Zeuthen et al. (1995) from a three-stage pilot plant, wherein they observed higher coke contents on the third-stage catalyst at the end of run. Since the series of reactors was operated without hydrogen makeup, the third stage would contain a lower hydrogen partial pressure and would be expected to accumulate higher coke contents based on this model. The authors also observed a phenomenon similar to breakthrough, in which during the early portion of the run, the third bed accumulated coke more slowly than the first bed. During accumulation of the coke on the first bed, the supply of coke precursors to the third bed would be reduced, causing a slower rate of accumulation. Validation of the Kinetic Model. The model presented above satisfactorily explains the behavior observed in the experiments with bitumen as the feed. The model proposes that the adsorbed coke is in a form which resembles the source oil, rather than a more highly condensed aromatic material. This assumption is consistent with the similarity of hydrogen-to-carbon atomic ratios measured for the residue products and coke deposits shown in Table 4. In order to determine the reactivity of the coke, heavily coked catalyst was exposed to tetralin at standard reaction conditions. A similar experiment was performed with coker gas oil as the feed. In the reactions using tetralin, significant coke removal was observed (see Table 2), with approximately one-third of the adsorbed coke removed during the first 1 h of reaction and an additional onethird removed after a total of six 1-h reactions. Somewhat less desorption was observed with coker gas oil, with approximately one-third of the coke removed after two 1-h reactions. A control experiment was not performed to determine the coke content a fresh catalyst would have reached after repeated reactions with coker gas oil, so no information was obtained about the desorption rate and total possible desorption in gas oil. This experiment does, however, generalize the results observed with tetralin, to include much heavier, less aromatic solvents. The degree of reversibility of adsorption under reactive conditions indicated that the surface

Ind. Eng. Chem. Res., Vol. 35, No. 11, 1996 3947

Figure 5. Distribution of the strengths of acid sites on the catalyst, as determined by stepwise thermal desorption of pyridine. The concentration of sites is shown on the ordinate, while the abscissa values are the temperature ranges for desorption of pyridine.

deposit was not highly condensed and unreactive. The reactivity of the coke is also consistent with the densities of coke deposits reported by Zeuthen et al. (1995), which averaged 750 kg/m3. This density was lower than expected on the basis of highly aromatic model compounds and supported our assumption that the initial deposit was similar to the residue. Zeuthen et al. (1995) demonstrated that the coke ages and becomes more aromatic with time in the reactor, which is beyond the scope of our study. Direct experimental verification of the annular zone model was not possible. Study of the external pellet surface would be complicated by artifacts not found on internal surfaces, such as deposition of mineral particles. Fracturing the pellet to expose internal surface area also exposes alumina surfaces which were not accessible to reactants, thus confounding the results. Surface polishing required prior to some analytical techniques will destroy the fine detail predicted by the model. The natural distribution of strengths of acid sites found on the alumina support provided a method for probing the surface. If annular zones of alumina are cleared in the vicinity of crystallites, as illustrated in Figure 4, then a greater concentration of acid sites of all strengths would be expected as the zone was enlarged, i.e., at higher hydrogen pressure. The distributions of the strengths of acid sites in fresh, moderately coked, and highly coked catalysts are shown in Figure 5. These data from stepwise thermal desorption indicated the concentration of acid sites on the surface. Since the least strongly adsorbed pyridine molecules will desorb in the lowest temperature range, this range defines the very weak acid sites, while the highest temperature range defines very strong acid sites. The two intermediate ranges define weak and strong acid sites of intermediate strength, although the division of the temperature ranges was arbitrary. As coke content decreased due to higher hydrogen pressure, a higher concentration of very strong acid sites was exposed on the surface. Given the concentration of nitrogenous and high boiling components found in these feeds, increased exposure of strong acid sites can only be explained by systematic clearing of the surface near metal crystallites. If removal of the coke from the surface was random and not associated with crystallites, the strongly acidic sites would be expected to preferentially adsorb the available residue molecules. The apparent increase in weak acidity with increased coke content was an artifact of the analytical method, caused by interactions

between the coke and pyridine. In this procedure, weak acid sites are defined as those sites from which pyridine desorbed between 373 and 423 K; however, a sample of coke derived from Athabasca bitumen under more severe reaction conditions retained approximately 1.1 mmol/g of pyridine at 373 K, which would correspond to 0.2 mmol/g of catalyst. This coke-pyridine interaction would tend to give more pyridine release at 373423 K with increasing coke content, as observed in Figure 5. The exact nature of the interaction between the retained coke and the surface is unknown. The literature suggests an acid-base interaction between the molecules and the surface, due to both basic nitrogen compounds and large aromatic clusters which exhibit basicity in the gas phase (Absi-Halabi et al., 1991; Korre et al., 1995). This hypothesis was tested through two experiments. In the first, adsorption of acridine was compared to dibenzofuran and anthracene. In these experiments, the model compound was dissolved or suspended in tetralin and then exposed to catalyst at standard reaction conditions (430 °C, 13.8 MPa of hydrogen, 1 h of reaction). Although anthracene could not be dissolved in tetralin at the desired concentration at room conditions, a solution was obtained at reaction conditions. No significant difference in coke deposition on the catalyst was observed between the anthracene solution, the acridine solution, and the tetralin blank (see Table 2). Solvent effects due to tetralin were ruled out when a similar coke content was obtained using acridine in 1-methylnaphthalene. The lack of dependence of coke deposition on basicity indicated that adsorption of basic nitrogen compounds in the residue was not the primary chemical interaction. In the second experiment, a sample of Athabasca maltenes was used as the feed for a multistep reaction series. Removal of the asphaltenes from this feed reduced the concentration of polar and high-molecularweight species from the residue. Coke deposition on the catalyst was marginally higher than was observed for the solvent and model compounds but only one-third of that observed for the whole bitumen (Table 2). These results suggested that the coke precursors were mainly in the asphaltene fraction of the bitumen. Basicity alone did not give deposition on the catalyst, but acidbase interactions may be important in adsorption of high-molecular-weight compounds. Implications of the Model. The surface clearing model predicts that except under conditions severe enough to form a complete monolayer on the surface, the hydrogenation sites on the metal crystallites will remain exposed. This prediction prompts the following question: If the metal sites are not fouled, why is a significant portion of the activity of hydroprocessing catalysts lost very early in the catalyst lifetime? In order to answer this question, one must examine the implications of the deposition of a coke layer on the alumina for diffusive mass transfer within a catalyst pellet. Given the distribution of pore sizes in the catalyst and molecular sizes in the feed, a rigorous determination of the effect of coking on effective diffusivity is not possible. Two techniques were used to estimate the possible magnitude of overall reduction in diffusivity of the residue fraction due to coking. First, hindered diffusion theory was used to determine the change in diffusivity of residue molecules as a function of pore size. Second, the overall effective diffusivity was estimated using the random pore model of Wakao and Smith (1964). Lee et al. (1991a,b) and Baltus and Anderson (1983)

3948 Ind. Eng. Chem. Res., Vol. 35, No. 11, 1996

the subscript m represents macropore, µ represents micropore, and e refers to effective diffusivity. In this model, the distinction between micropores and macropores can be somewhat arbitrary but the typical approach is to define pores smaller than 10 nm as micropores (Froment and Bischoff, 1990). In Wakao and Smith’s original work, the diffusivity in each range of pore sizes was determined using equations for molecular and Knudsen diffusion. Since the residue molecules were diffusing through a liquid medium within narrow pores, appropriate expressions for hindered diffusion must be used in eq 14. Rewriting the hindered diffusion relationship of eq 11 for a single pore gives Figure 6. Pore size distribution (PSD), based on surface area, and the reduction in effective diffusivity caused by coke deposition as a function of pore radius. Total surface area ) ∫A(r) dr.

studied the hindered diffusion of asphaltene molecules and showed that the effective (hindered) diffusivity in catalyst pellets (De, m/s2) was related to the bulk diffusivity (Db, m/s2) as follows:

De P -Bλ ) e Db τ

(11)

where P is the pellet porosity, τ is the tortuosity, B is an empirical coefficient, λ is the ratio of molecule size to pore size, λ ) rM/rp, rM is the radius of the diffusing molecule (m), and rp is the pore radius (m). The pore size distribution of the NiMo/γ-Al2O3 catalyst is given in Figure 6. For any pore size range, the contribution to pellet porosity of that range will be related to the void volume. If one assumes that the coke is a monolayer with a thickness defined by eq 1, then the reduction in pore radius and void volume can be determined. The porosity of the coked catalyst was related to the fresh porosity by

(rp - 2rm)2

c ) f*

rp2

(12)

where c is the coked porosity and f is the fresh porosity. Rewriting eq 11 as the ratio of diffusivities for coked and fresh catalyst gives

De,c c -(λc-λf) ) e De,f f

(13)

This ratio is shown in Figure 6 for rM ) 0.6 nm. From this figure, it is obvious that a significant reduction in diffusivity is predicted for all pore sizes. The predicted ratio of the diffusivity of the coked catalyst compared to the diffusivity of the fresh catalyst decreases rapidly below a pore radius of approximately 8 nm, reaching 0.06 at 2 nm. The random pore model of Wakao and Smith (1964) defines the contribution of diffusion in micro- and macropores to the overall effective diffusivity, De. Their equation for the overall diffusivity in a catalyst pellet was based on a model for the characteristics of the pores in the catalyst to give

De ) m2Dm +

µ2(1 + 3m) Dµ 1 - m

(14)

where  is the porosity, D is the diffusivity (m/s2), and

Dp,i ) Dbe-Bλi

(15)

where the subscript i refers to micropores or macropores and B is an empirical parameter. The reported value of the parameter ranges from 3.9 to 4.6 for comparable hydrocarbons (Lee et al., 1991b; Baltus and Anderson, 1983). Values of λ for the micro- and macropores of the fresh catalyst were determined from the pore size distribution. The maximum of the pore size distribution, 3 nm, was chosen for rµ, and 15 nm was chosen for rm, based on the 10-nm cutoff for micropores and the maximum pore size detected. The radius of the diffusing molecule was rm ) 0.6 nm. Figure 6 shows the pore size distribution before and after coking. Before coking, the total surface area was 317 m2/g and the pore volume was 0.57 mL/g. A catalyst with 17.5% carbon had a surface area of 157 m2/g and a pore volume of 0.25 mL/g. If a uniform layer of coke is deposited in the pore structure, then the surface area attributed to each pore will be reduced in proportion to the square of the ratio of the fresh and coked pore radii. As well, the maximum of the pore size distribution is expected to shift to a smaller radius due to the deposition of a uniform layer of coke. The data clearly show the expected reduction in total surface area, with an accompanying shift in the maximum of the distribution. It is also interesting that the pore size distribution for the coked catalyst shows a peak at approximately 2 nm, the same radius at which the ratio of the effective diffusivities of the coked and fresh catalysts reaches 0.06. These data suggest that residue molecules may be excluded from pores with radius less than 2 nm. Two approaches were used to estimate the pore radii for the coked catalysts. In the first approach, the maximum from the pore size distribution, 2.5 nm, was used for rµ and 15 nm was used for rm. In the second approach, a uniform layer of coke is assumed to form in the pore structure, giving a value of the micropore radius of rµ ) 1.8 nm and a macropore radius of rm ) 13.8. These two approaches were used to estimate the bounds on the reduction in diffusivity in the coked catalyst relative to a fresh catalyst, using eqs 14 and 15. Allowing for the uncertainty in the exact value of parameter B, we bounded the change in diffusivity of residue due to coke deposition by using 3.9 e B e 5.1 to obtain the estimate

0.068 e

De,c e 0.105 De,f

(16)

Given the uncertainty in the radius of coked pores and parameter B for hindered diffusion, this calculation suggests that the effective diffusivity of the coked catalyst will be reduced by approximately 90% relative to the clean catalyst.

Ind. Eng. Chem. Res., Vol. 35, No. 11, 1996 3949

A number of studies have shown that a decrease in pore diameter reduces the effective diffusivity for hydroprocessing reactions, as one would expect from eq 11. Although a change in pore diameter is not the same as depositing a layer of coke, studies by Ammus and Androutsopoulos (1987), Li et al. (1995), and Seo and Massoth (1985) all showed that a reduction of De due to coke deposition could range from 10% to 60%. The work of Lee et al. (1991b) on fresh and coked catalysts suggests that a reduction in De of as much as 65% could result from only 2-3% carbon deposition. Such a drastic change, however, indicated plugging of the pore network, which could not be extrapolated to the present study. Uniform Deposition of Coke vs Pore Mouth Plugging. Observations of drastic reductions in diffusivity due to small amounts of coke, e.g., Lee et al. (1991b), indicate that in some cases, the pore network in the catalyst can be blocked off at pore mouths. The model presented in this paper suggests uniform deposition of coke. One cannot conclusively prove which model more closely reflects reality. Optical microscopy is not useful since both the coke deposits and the supported metal sulfides are black. Electron microscopy suffers from the limitations on sample preparation discussed earlier. Diffusivity measurements may provide some clue as to the structure, but these were not attempted with the catalysts used in this study. The alternative pore mouth plugging model is inconsistent with several of the experimental observations. If one assumes that the deposit has properties similar to the residue fraction of the oil, the volume of the deposited oil accounts for approximately one-half the total pore volume, consistent with the reduction in pore volume observed experimentally. If micropores were blocked at the pore mouth, this level of deposition would not be possible. Furthermore, mesopores (10-20 nm) would be almost unaffected, with virtually no surface area loss in the mesopore range. Our results show a 48% reduction in mesopore volume, inconsistent with pore mouth plugging (Figure 6). The material deposited in pore mouths is usually assumed to be a highly aromatic, unreactive material. Our results clearly indicated that during the initial stages of the reaction, the coke material is reversibly adsorbed and/or reactive. While it is clear from the literature that pore mouth plugging may occur on hydrocracking and hydrotreating catalysts during operation under some conditions, our experimental results for coke content and pore size distribution in spent catalysts from residue processing support a uniform deposition model. More study is required to determine the factors that determine the mode of coke deposition. Conclusions 1. The initial deposition of coke on hydroprocessing catalyst was insensitive to time and residue concentration and weakly dependent on hydrogen pressure. 2. Deposits on the surface were removed by contacting catalyst with tetralin at reaction conditions, consistent with reversible adsorption or high reactivity of the deposit. 3. A physical model for coke deposition was developed to incorporate the experimental observations. The model assumes that active hydrogen clears an annular zone around each crystallite, which remains free of coke. This model was satisfactorily fitted to experimental hydrogen pressure data and was consistent with the

observed effects of feed composition, reaction time, and cumulative oil-to-catalyst ratio. Acknowledgment We acknowledge the support of the following groups: The Alberta Oilsands Technology and Research Authority, the Natural Sciences and Engineering Research Council, and Syncrude Canada Ltd. Nomenclature a ) empirical parameter (MPa) B ) empirical coefficient b ) empirical parameter (m2 of metal/m2 of surface) C ) actual concentration of carbon in the catalyst (wt % C, on a spent catalyst basis) c ) empirical parameter (MPa) Cp ) surface concentration of coke precursor (mol/m2) Db ) diffusivity of residue molecules in bulk solutions (m/s2) De ) effective diffusivity in a catalyst pellet (m/s2) Fcrit ) critical surface flux of hydrogen (mol m-1 s-1) kH ) rate constant for production of surface hydrogen (mol MPa-1 s-1) k ) adsorption rate constant (g of catalyst/g of feed) kd ) deposition rate constant for hydrocarbon (m-1 s-1) M ) surface concentration of carbon (g of carbon/m2) MW ) molecular weight of residue N ) number of metal crystallites (crystallites/m2) NA ) Avogadro’s number (6.02 × 1026 molecules/kmol) PH2 ) partial pressure of hydrogen (MPa) r ) radius (m) rc ) radius of the metal crystallite (m) Rdep ) rate of carbon deposition on the catalyst (g of residue/g of catalyst) Td ) pyridine desorption temperature (K) xC ) mass fraction of carbon in the residue Greek Symbols  ) porosity p ) porosity of the catalyst pellet λ ) ratio of molecule size to pore size: λ ) rr/rp θv ) fraction of possible adsorption sites currently vacant F ) density (kg/m3) σ ) stoichiometric coefficient for reaction of surface hydrocarbon with hydrogen τ ) catalyst tortuosity ω ) cumulative feed-to-catalyst ratio, a pseudotime coordinate (g of feed/g of catalyst) Subscripts c ) coked f ) fresh i ) either macropore (m) or micropore (µ) m ) macropore M ) molecular max ) maximum value p ) pore µ ) micropore

Literature Cited Absi-Halabi, M.; Stanislaus, A.; Trimm, D. L. Coke Formation on Catalysts During the Hydroprocessing of Heavy Oils. Appl. Catal. 1991, 72, 193-215. Aitken, A. R.; Merrill, W. H.; Pleet, M. P. Hydrogenation of a Coker Distillate Derived from Athabasca Bitumen. Can. J. Chem. Eng. 1964, 234-238. Ammus, J. M.; Androutsopoulos, G. P. HDS Kinetic Studies on a Greek Oil Residue in a Spinning Basket Reactor. Ind. Eng. Chem. Res. 1987, 26, 494-501. Andersen, J. R.; Pratt, K. C. Introduction to Characterization and Testing of Catalysis; Academic: Sydney, 1985.

3950 Ind. Eng. Chem. Res., Vol. 35, No. 11, 1996 Anonymous. Standard Test Method for Distillation of Petroleum Products at Reduced Pressure. 1996 Annual Book of ASTM Standards: Section 5, Petroleum Products, Lubricants and Fossil Fuels; ASTM: Philadelphia, PA, 1996; Section 05.01, pp 389406. Ayasse, A. Hydrocracking of Athabasca Bitumen. MSc Thesis, University of Alberta, 1994. Baltus, R. E.; Anderson, J. L.; Hindered Diffusion of Asphaltenes Through Microporous Membranes. Chem. Eng. Sci. 1983, 38, 1959-1969. Beaton, W. I.; Bertolacini, R. J. Resid Hydroprocessing at Amoco. Catal. Rev. Sci. Eng. 1991, 33, 281-317. Diez, F.; Gates, B. C.; Miller, J. T.; Sajkowski, D. J.; Kukes, S. G. Deactivation of a Ni-Mo/γ-Al2O3 Catalyst: Influence of Coke on the Hydroprocessing Activity. Ind. Eng. Chem. Res. 1990, 29, 1999-2004. Douwes, C. T.; van Klinken, J.; Wiffles, J. B.; van Zijll Langhout, W. C. Developments in Hydroconversion Processes for Residues. Proc. 10th World Petrol. Congr., Bucharest 1979, 4, 175-183. Froment, G. F.; Bischoff, K. B. Chemical Reactor Analysis and Design, 2nd ed.; John Wiley and Sons, New York, 1990. Gray, M. R. Upgrading Petroleum Residues and Heavy Oils; Marcel Dekker: New York, 1994. Gray, M. R.; Khorasheh, F.; Wanke, S. E.; Achia, U.; Krzywicki, A.; Sanford, E. C.; Sy, O. K. Y.; Ternan, M. Role of catalyst in Hydrocracking of Residues from Alberta Bitumens. Energy Fuels 1992, 6, 478-485. Gray, M. R.; Ayasse, A. R.; Chan, E. W.; Veljkovic, M. Kinetics of Hydrodesulfurization of Thiophenic and Sulfide Sulfur in Athabasca Bitumen. Energy Fuels 1995, 9, 500-506. Gregg, S. J.; King, K. S. Adsorption, Surface Area and Porosity; Academic: London, 1967. Hannerup, P. N.; Jacobsen, A. C. A Model for the Deactivation of Residual Hydrodesulfurization Catalysts. Prepr. Pap.sAm. Chem. Soc. Div. Pet. Chem. 1983, 28, 576-599. Inoguchi, M.; Kaneko, Y.; Satomi, Y.; Inaba, K.; Kagaya, H.; Tate, K.; Mizutori, T.; Nishiyama, R.; Ota, T.; Niume, K. Studies in the Hydrodesulfurization Catalyst of Residual Fuels (Part 6). Bull. Jpn. Petrol. Inst. 1972, 14, 7-17. Kapoor, A.; Yang, R. T.; Wong, C. Surface Diffusion. Catal. Rev. Sci. Eng. 1989, 31, 129-214. Korre, S. C.; Klein, M. T.; Quann, R. J. Polynuclear Aromatic Hydrocarbon Hydrogenation. 1. Experimental Reaction Pathways and Kinetics. Ind. Eng. Chem. Res. 1995, 34, 101-117. Lee, S. Y.; Seader, J. D.; Tsai, C. H.; Massoth, F. E. Restrictive Liquid Phase Diffusion and Reaction in Bidispersed Catalysts. Ind. Eng. Chem. Res. 1991a, 30, 1683-1693. Lee, S. Y.; Seader, J. D.; Tsai, C. H.; Massoth, F. E. Restrictive Diffusion Under Catalytic Hydroprocessing Conditions. Ind. Eng. Chem. Res. 1991b, 30, 29-38. Li, C.; Chen, Y. W.; Tsai, M. C. Highly Restrictive Diffusion Under Hydrotreating Reactions of Heavy Residual Oils. Ind. Eng. Chem. Res. 1995, 34, 898-905. Madeley, R.; Wanke, S. E. Characterization of (Ni-Mo)/Al2O3 Catalysts by X-ray Diffraction. Prepr. Tenth Can. Symp. Catal., Kingston, Ontario (June 15 - 18, 1986) 1986, 134-143. Melo Faus, F.; Grange, P.; Delmon, B. Influence of Asphaltene Deposition on Catalytic Activity of Cobalt Molybdenum on Alumina Catalysts. Appl. Catal. 1984, 11, 281-293.

Muegge, B. D.; Massoth, F. E. Basic Studies of Deactivation of Hydrotreating Catalysts with Anthracene. Fuel Process. Technol. 1991, 29, 19-30. Nagaishi, H.; Chan, E. W.; Sanford, E. C.; Gray, M. R. Kinetics of High-Conversion Hydrocracking of Bitumen. Energy Fuels 1996, submitted. Nishijima, A.; Shirada, L. T.; Yoshimura, Y.; Sato, T.; Matsubayashi, N. Deactivation of Molybdenum Catalysts by Metal and Carbonaceous Deposits During the Hydrotreating of CoalDerived Liquids and Heavy Petroleums. Catalyst Deactivation; Elsevier: Amsterdam, 1987; pp 39-43. Oballa, M. C.; Wong, C.; Krzywicki, A. Catalyst Deactivation in Residue Hydrocracking. In Catalytic Hydroprocessing of Petroleum and Distillates; Oballa, M. C., Shih, S. S., Eds.; Marcel Dekker: New York, 1994; pp 33-54. Paal, Z.; Menon, P. G. Hydrogen Effects in Catalysis, Fundamentals, and Practical Applications; Marcel Dekker: New York, 1988. Richardson, S. M. Initial Coking of a NiMo/γ-Al2O3 Bitumen Hydroprocessing Catalyst. Ph.D. Thesis, University of Alberta, 1996. Seo, G.; Massoth, F. E. Effect of Pressure and Temperature on Restrictive Diffusion of Solutes in Aluminas. AIChE J. 1985, 31, 494-496. Strausz, O. P. Structural Features of Athabasca Bitumen Related to Upgrading Performance. Symp. Correlation Between Resid Characterization Processability. Prepr. Pap.sAm. Chem. Soc., Div. Pet. Chem. 1989, 34, 395-400. Ternan, M.; Furimsky, E.; Parsons, B. I. Coke Formation of Hydrodesulfurization Catalysts. Fuel Process. Technol. 1979, 2, 45-55. Thakur, D. S.; Thomas, M. G. Catalyst Deactivation During Direct Coal Liquefaction: A Review. Ind. Eng. Chem. Prod. Res. Dev. 1984, 23, 349-360. Thakur, D. S.; Thomas, M. G. Catalyst Deactivation in Heavy Petroleum and Synthetic Crude Processing: A Review. Appl. Catal. 1985, 15, 197-225. Wakao, N.; Smith, J. M. Diffusion and Reaction in Porous Catalysts. Ind. Eng. Chem. Fundam. 1964, 3, 123-127. Yui, S. M.; Sanford, E. C. Kinetics of Aromatics Hydrogenation of Bitumen-Derived Gas Oils. Can. J. Chem. Eng. 1991, 69, 1087-1095. Zeuthen, P.; Cooper, B. H.; Clark, F. T.; Arters, D. Characterization and Deactivation Studies of Spent Resid Catalysts from Ebullating Bed Service. Ind. Eng. Chem. Res. 1995, 34, 755-762.

Received for review December 15, 1995 Revised manuscript received July 11, 1996 Accepted July 25, 1996X IE950761O

X Abstract published in Advance ACS Abstracts, October 1, 1996.