2626
Ind. Eng. Chem. Res. 1993,32, 2626-2632
Deactivation of a Silica-Alumina Catalyst by Coke Deposition F6lix Garcia-Ochoa’ and Aurora Santos Departamento de Ingenierh Quimica, Facultad de CC. Quimicas, Universidad Complutense, Madrid, Spain
Deactivation by fouling of a silica-alumina catalyst when it is used for cyclohexanol dehydration has been studied in a fixed bed laboratory reactor between 548 and 573 K. A kinetic model of the main reaction has been determined, corresponding to a mechanism in which the surface reaction in two active sites is the controlling step of the process rate. The deactivation rate has been determined from activity-time data, calculated from outlet conversion-time data. Coke precursor formation has been determined that occurs by reaction of three adsorbed molecules of reactant (cyclohexanol) or product (cyclohexene). A greater contribution (bigger parallel contribution) of coke formation from the reactant has been found. Also, the variation of different physical and chemical catalyst properties, such as surface, pore volume, acidity, and coke composition, have been measured a t different coke contents, with the result that relationships between coke content and activity and between activity and acidity support the hypothesis of site coverage deactivation, a t least for the coke level range achieved in this study (0-4.1% (w/w), corresponding t o an activity decay from 1 to 0.13); hence, the monolayer coke formation over the catalyst surface can be assumed. These results are confirmed by the nonvariation of physical properties.
Introduction In most of the catalytic heterogeneous processes carried out using a solid as catalyst, the activity of such solid decays with time. There are hundreds of papers dealing with research works on this subject and several reviews where the main progress is described, such as that from Butt (1972, 1980, 1982), Butt and Billimoria (1978), Hegedus and McCabe (1981), Wolf and Alfani (1982), Forzatti et al. (1984a,b), Hughes (1984), Butt and Petersen (19881, Froment and Bischoff (1990),and Krishna (1991). Fouling by coking is the deactivation type more widely studied. Coke formation can be produced by aromatic compounds (Appleby et al., 1962),olefines (Wojciechowskiet al., 1974), paraffin and naphtenic compounds, depending on the catalyst support and the temperature (Rollmann, 1977; Walsh and Rollmann, 1977 and 1979). Coke is a polymer of a high condensation grade, with a high molecular weight, whose general composition is CH,, where n decreases with increasing coke content (Stiegel et al., 1985;Bilbao et al., 1985). This polymer is produced by a precursor, with a H/C ratio higher than that in the final form of coke, which can be formed by reactants (parallel deactivation), by products (series deactivation), or by both. The precursor after a sequence of steps involving polymerization by addition and/or dehydrogenation yields the coke, which can produce active-site coverage or pore blockage, and thus the catalytic activity decay. Usually the catalyst state has been described by the variable “activity”, a,which can be included in the reaction kinetic model as a new chemical compound, in the following way (Levenspiel, 1972;Forzatti et al., 1984a; Froment and Bischoff, 1990): r = r0a (1) Usually the reaction rate with fresh catalyst, ro, is determined, without diffusional limitations, by a kinetic equation
ro = f ( T , P ) (2) while to describe the activity evolution with time, a kinetic equation such as eq. (3) is usually employed (Jodra et al., 1976). rd = -da/dt = $( T,P)ad
(3)
0888-5885/93/2632-2626$04.00IO
A t low coke content a linear relationship between precursor and coke concentration is found, the coke is deposited in a monolayer, producing active-site coverage but not pore blockage, and the physical properties of the solid are not affected (specific surface, pore volume, and pore size remain constants). For silica-alumina catalysts this seems to happen for a coke concentration under 6 % (w/w) (Haldeman and Botty, 1959;Wolf and Alfani, 1982; Dean and Dadyburjor, 1986; Acharya et al., 1990). According to the development followed by several authors (Chu, 1968;Dumez and Froment, 1976;Jodra et al., 1976) the coke content-activity relationship can be expressed as
c, = (S&,/h)(l-
a””)
(4)
where h is the number of active sites involved on the precursor formation reaction and m is the number of active sites involved in the controlling step of the main reaction. For acid catalysts, as silica-alumina, activity decay is related to decreasing acidity (Aguayoet al., 1990),and the activity-acidity relationship can be written as (Santos, 1992): a = b,A”
(5) There are many works in the literature dealing with some of the different aspects quoted, but there are not many relating to these different items. Thus, the kinetic and structural aspects for coke deposition in a silicaaluminacatalyst have been studied by Bilbao et al. (19851, and the coke effect on the activity, acidity, and surface area has been analyzed by Dean and Dadyburjor (1989) on a composite acid catalyst. The aim of this work is to determine the kinetic equation, for a fresh catalyst and for a partially deactivated catalyst, and the relationships between activity and coke content and between activity and catalyst acidity. At the same time, the influence of coke content on physical properties, such as specificsurface and pore volume, will also be taken into account. Dehydration of cyclohexanol on a commercial silica-alumina catalyst has been the chemical system utilized. Dehydration of alcohols on acid Catalystshas been widely studied (Kabel and Johanson, 1962; Carrti et al., 1966;Figueras et al., 1971; Ravindran and Muthy, 1981; Miller and Kirk, 1982) for chemical reaction mechanism and kinetic pa0 1993 American Chemical Society
Ind. Eng. Chem. Res., Vol. 32, No. 11, 1993 2627 Table 11. Experimental Results from Runs in Differential Fixed Bed Reactor: Group A of Experiments
- FeQd Llquld - Vaporizator - Prcheatcr - Reactor C - Condensator
F V P R
Vessel
11
-
T Sample collector GC-Gas Chromatograph A - Valve D Flame Ionization Detector
-
Figure 1. Schematic diagram of experimental setup. Table I. Fresh Catalyst Properties of KA-3 from SBdchemie (Chemical Composition Alz0~4SiOrH20) property
V,, (crn8.g') pP (g.cm-9
(total) tM (P
value 0.587 0.912 0.535 0.219 0.316
property
n~(nm) r, (nm) S, (m2.g1) S 8 (m2.g1) ~ S , (m2.g1)
run 1 2 3 4 5 6 7 8 9 10
value 333 4.5 126 10 116
rameter calculation method studies (Butler et al., 1970; Figueras et al., 1971). Cyclohexanol (OL)dehydration catalyzed by silica-alumina is a good example of the above mentioned, where only the dehydration reaction needs to be taken into account (Carrl et al., 1966), producing cyclohexene (E) and water (W).
Experimental Section Experimental Setup and Analysis Method. A laboratory fixed bed reactor was used, measuring the change of the outside gas composition with time. A scheme of the installation is shown in Figure 1; the evaporator, preheater, and reactor were of Pyrex glass and the reactor tube was 1 cm in diameter. Isothermicity was checked with axial and radial position and with time. Analysis was performed by gas-liquid chromatography, using a Hewlett-Packard 5840A with FI detector. A stainlesg steel column of 1/8 in. in diameter with 7 f t in length with 25 % poly(ethy1eneglycol) on Chromosorb W 80/100 was used. Analysis conditions were an oven temperature of 135 "C and a gas flow (nitrogen) of 25 cm3/min. Cyclohexanone was used as internal standard. Materials and Catalyst. Cyclohexanol (Panreac) of 98% purity, cyclohexene (Panreac) and water of 99% purity, and nitrogen L48 (Liquid Carbonic) as inert were used. For catalyst acidity determination N-butylamine (Carlo Erba) 99 % was employed diluted in benzene (dry, from MercK). A commercial silica-alumina catalyst from Sudchemie (KA-3) was utilized. Table I shows the main properties of the fresh catalyst. Specific surface was determined by the BET method (Micromeritics Accusorb 2000 Sorptometer). Pore size distribution, macroporosity, and microporosity, were found using a mercury porosimeter Carlo Erba Milestone 2000 and a Coulter Omnisorb 100 Sorptometer. Coke analysis was performed by elemental analysis with a Leco HCM 600. Experimental Results and Interpretation Preliminary runs for the determination of the experimental intervals of the operational conditions and a study
12 13
T (K) WIFou, (g.s/mol) 533 533 533 533 533 533 573 573 573 573 573 573 573
504 490 508 478 497 465 437 382 386 416 405 385 391
Pou, (atm) 0.274 0.386 0.554 0.709 0.749 0.839 0.346 0.473 0.654 0.682 0.686 0.790 0.836
XOL(t = 0) 0.038 0.048 0.073 0.066 0.079 0.071 0.080 0.103 0.104 0.137 0.120 0.142 0.133
to eliminate diffusional resistances were carried out. Interand intraphase composition and temperature differences are negligible a t linear gas velocities greater than 6 cm/s and for catalyst particle diameter smaller than 0.1 cm, which was experimentally determined at the maximum temperature studied, 300 "C. In the following experimentation, 8 cm/s as linear velocityand a particle diameter of 0.066 cm were employed. Later on, runs were performed with changing temperature (260, 275, and 300 "C) and inlet partial pressure of cyclohexanol (between 0.25 and 0.85 atm) using nitrogen always as inert, feeding cyclohexene and/or water in some runs, and using different space times (between 380 and 7000 g.s/mol). Analyses of the outlet composition were carried out at time intervals between 5 and 20 min, depending on the run and the time. Three types of runs were carried out: (A) For the kinetic equation determination with fresh catalyst, the fixed bed reactor was used in a differential way (that is, usually XOLI 0.10 and always XOLI 0.15). Experimental results at zero time were obtained by extrapolation of results collected in the first hour. (B) For the kinetic equation determination both with fresh and partially deactivated catalyst, the experimental reactor was used in an integral way (XOLbetween 0.07 and 0.98). The outlet gas composition was analyzed during a time between 2 and 3 h. (C) For the influence of coke on different catalyst properties, one integral run was carried out. It was stopped at different times on stream, and the catalyst was removed. Some properties were measured, such as coke content and composition, specific surface, pore volume, and catalyst acidity (changes in activity were available from data obtained previously). Table I1shows the experimental results in a differential reactor (groupA). Table I11showsthe results in an integral reactor (group B), extrapolated at zero time; some of the results obtained with time can be seen in Figure 6 (together with the fitting achieved with the model finally selected). Table IV and Figure 2 show the results obtained in the catalyst properties determination. Kinetic Model for Main Reaction. Six different mechanisms have been taken into account, which are shown in Table V, from which 21 different kinetic models can be deduced, depending on the rate-controlling step (adsorption, surface reaction, or desorption). Initial rates (Figure 3) were analyzed (Yangand Hougen, 1950; Kittrell, 1970) by employing the data from experiments of group A (Table 11). Kinetic models with adsorption and desorption as the controlling step can be rejected, and surface reaction is considered as the controlling step of the process rate. Kinetic models for the six mechanisms with surface reaction as the controlling step are shown in Table V. Using data from experiments
2628 Ind. Eng. Chem. Res., Vol. 32, No. 11, 1993 Table 111. Experimental Results from Runs in Integral Fixed Bed Reactor: Group B of Experiments. T=533K T = 548 K T = 573 K
run 14 15 16 17 18 19 20 21 22
WIFOU,(g.s/mol) 454 803 1010
1617 2604 3379 4881 5786 6961
XOL(t = 0) 0.068 0.102 0.127 0.192 0.345 0.435 0.621 0.703 0.815
run 23 24 25 26 27 28 29 30 31
XOL(t = 0) 0.082 0.170 0.248 0.361 0.449
WIFOU,(g.s/mol) 477 869 1547 2133 2564 3523 4616 5784 7360
run 32 33 34 35 36 37 38 39 40
0.611
0.746 0.843 0.958
41 a
Inlet Conditions:
YOM
= 0.76; ywo = 0.1; y m = 0; PT = 0.93 atm.
Table IV. Chemical and Physical Properties of the Catalyst for Different Coke Content
run 42a 42a 42b 42c 42d
t (min) 0 25 200 360 720
XOL a 0.75 1 0.67 0.85 0.32 0.45 0.19 0.27 0.08 0.13
XOL(t = 0) 0.121 0.160 0.247 0.400 0.503 0.584 0.683 0.804 0.938 0.980
WIFOU,(g.s/mol) 381 561 812 1581 1982 2444 2953 3855 5124 6458
A (m2/g) (cmYg) (mequiv/g) 9.5 126 0.587 8.5 131 0.572 0.561 6.5 121 4.7 128 0.570 2.7 125 0.559
s,
VP
11
I
42-c 62-d
0.6
of group B, carried out in the fixed bed in an integral way, the reaction rate has been calculated according to
rob = dXobld( WIFOb)
(6)
those values being extrapolated to zero time. Kinetic models were linearized, and a multiple linear regression was applied at each temperature. Only model 1fulfills all the criteria applied (Garcia-Ochoa et al., 1989,1990), but this was not considered as a definitive choice, because of the statistical lack introduced by the linearization technique (Hofmann, 1972; Kittrell, 1970). After this, nonlinear regression was applied to the same data (Marquardt, 1963), first to each of the temperatures, according to eq 7. Models 2, 3,4, and 6 can be rejected because some of
XoL, = J,w’Fob ~OL,(TQOLQE,PW) d(WIFob)
(7)
the parameters at several temperatures show negative values of some parameters. Finally, nonlinear regression (Marquardt, 1963) was applied, taking into account only models 1and 5 including the temperature as variable in eq 7, with the parameters K , KOL,K E ,and KWaccording to an Arrhenius function, using the reparametrization technique quoted by Kittrell (1970). Activation energy calculated for model 5 gave a very low value. Boudart’s criteria were then applied (Boudart et al., 1967) according to eq 8. (a)
ASja< 0
I
I
T 3 5L8 K W / F o ~ ~ : L 3 7 6g . s / m o l POLO: 0 L a t m
% c, (g/g)
total C H 0 0 0 0.9 0.6 0.3 2.1 1.6 0.5 2.7 2.1 0.6 4.1 3.4 0.7
I
0
I 200
A I
I
I
LOO
600
A
J
800
-t (min) Figure 2. Cyclohexanol conversion vs time for run 42 stopped at different times.
Table V. Mechanisms and Kinetic Models for Cyclohexanol Dehydration
mechanism OL + s += OLS OLs + s + Es + Ws Es*E+s ws=w+s
model with surface reaction as controlling step
OL + 2s ?= 2OL1/2S 2OLlps + ES+ WS Es+E+s ws=w+s OL + 8 + OLS 20Ls + E25 WzS Ezs+2E+s WzS 2w + 9 OL + 9 += OLS OLs+E+Ws ws=w+s OL + 8 =OLs E + 2Wl/zS OLS + 2WIfZS + w + 2s
(c)
10 I -ASj, I 12.2 - 0.0014AHje
(8)
Model 5 does not fulfill criterion b, and it was finally rejected. Therefore, model 1 has been finally selected. Similar models have been chosen by other authors for alcohol dehydration (Kabel and Johanson, 1962; Carrl et al., 1966;Athappan and Srivastava, 1980). Model 1, with the parameter values, is given by eq 9, where the confidence intervals of the parameters are for 95%. Figure 4 shows a comparison between calculated and experimental data.
OL 2s + 20LlpS 2OLlps + E + WS + s ws=w+s
model 6
As can be seen, the model selected fits the experimental data very accurately. Kinetic Modelingof Deactivation. Deactivation rate can be written as eq 3. For \L(T,P)and d determination, the data XOLvs time from experiments of group B have
Ind. Eng. Chem. Res., Vol. 32, No. 11, 1993 2629 I
I
0.4
0.6
I
I
k = 11.422 f 0.75 exp[(-5295 f 15)/Tl mol-g-'-s-' KOL = 1.891 X
f 0.16 X
exp(3914 f 33/T) atm-'
KE =
/ O
1.225 X lo4 f 0.295
X
lo4 exp(3833 f 91/T) atm-'
Kw = 8.673 X 10'' f 2.72 X lo4 exp(3093 f 72/T) atm-' (9) been analyzed (see Table 111). For each run, the activity at the reactor outlet conditions has been calculated according to eq 1, following the procedure applied by Bharati and Bhatia (1987). Figure 5 shows some a values calculated at different time and operational conditions. The deactivation rate, r d , has been calculated by numerical derivation of these curves a vs time. For discrimination among different values of order d and different models for the function J/(Ts>,a mechanism for coke formation has been assumed, according to the following scheme:
0
0.2
-
0.8 1.0 POL,latm)
Figure 3. Initial reaction rates obtained in different runs vs inlet cyclohexanol partial pressure. A
0
X
t
0.9
0.7 E + W OL'
I
0.5 0.3 0.1
where a series plus parallel deactivation is considered to yield the coke precursor, PSh. Thus, the following kinetic model can be proposed for the deactivation rate: Figure 4. Comparison between calculated and experimental cyclohexanol conversion at zero time from runs in integral reactor, employing model 1.
combining eq 3 and eq 11,the following can be deduced: m+h-1 m In eq 11, different parameter values of h and n have been considered, the first taking the values 1,2, and 3 and the second taking the values 0, 1, and 2. In a first discrimination step the value of order d was calculated. For this, a linear regression of activity decay rate vs different d power of activity was carried out, d taking values of 1,3/2, and 2 according to eq 12. Only the data for XOL I0.08 were used, because for this range J/(T,P)can be considered constant, and according toeq 3the relationship between activity decay, -da/dt, and the activity power term must be linear. Thus, a best value of d = 2 was deduced, which means a value of h = 3, according to eq 12. Therefore, only three models must be considered in the next discrimination step, those corresponding to eq 11 with values of n = 0, 1, or 2. In a second step, a nonlinear regression (Marquardt, 1963) at each temperature was applied to eq 11, with h, m, KOL,KE, and Kw fixed at the values previously calculated for each temperature. In all cases, KdoL gave an anomalous variation with temperature, but the model with a value of n = 0 yields the smaller residuals.
d=
Finally, nonlinear regression to all the data, including the temperature as a variable, was performed. All three models considered, those with n = 0, 1, or 2, gave good results, with m e d n g in the parameter values and with confidence intervals not including zero value. Nevertheless, the model with n = 0 was chosen because the residuals were smaller and the parameter confidence intervals were narrower than with the other two models (n = 1and n = 2). This model (h = 3, n = 0, d = 2) is given in eq 13
kdoL = exp 3.524 f 1.2 - 3123:
(
628) min-l
kdE = exp 31.25 f 4.1 - 13780 f 2010) min-l (13) T
(
(adsorption constants values are given in eq lo), together with the confidence intervals of the parameters for 95 % probability. As can be seen in Figure 6, this model fits the experimental results very well, and the residuals do not present a trend with any variable.
2630 Ind. Eng. Chem. Res., Vol. 32, No. 11, 1993
0.8 0.6 0.4
0.2
0
1
I
1
40
80
120
-
I 40
I
160 t (min)
(a) Figure 5. Activity vs time. (a) 533 K; (b) 548 K (c) 573 K.
-
I
I
80
120
I 80
1 120
-
I
160 t (min)
(b)
(b)
t (min)
1 40
I
160 t (min)
(C)
-t(min)
-t(min)
(a) Figure 6. Comparison between calculated and experimental cyclohexanol conversion at different times for runs in integral reactor, employing eq 17. (a) 533 K;(b) 548 K;(c) 573 K.
reaction. At this point, m values of 1 and 2 have been considered, with the value m = 2 yielding a better fitting of experimental results, as can be seen in Figure 7,where coke content, C,, vs a function of activity, (1 - a l / m ) , is represented. On the other hand, the value m = 2 is also the best to fit the data of activity, a,vs acidity, A, as shown in Figure 8. The value m = 2 was also the one previously found by fitting the experimental results of XoL vs time. Therefore,the relationships between activity, acidity and coke content can be written as
C, = 0.163 X 0
0.2
0.4
0.6
-
0.8
+ 5.872 X
(1 -
(14)
1
1 -a112
Figure 7. Catalyst coke content vs a functfon of catalyst activity, (1 - c i ' / 2 ) .
Coke Content Effect on Chemical and Physical Catalyst Properties. As mentioned in the Experimental Section, one of the runs was stopped at different times, and some properties of the catalyhst with different coke content were analyzed. Figure 2 shows that the XOL variation with time was very similar to that found in the runs carried out without the several interruptions. The experimentalresults for physical properties S, and V , and acidity, A, are shown in Table IV, together with coke content, in percent (w/w), the coke composition (C, H), and the activity values calculated according to eqs 9 and 13, with the data from Figure 2. When the activity decreases to 0.13,the coke content is hardly 5 % . At such small coke content, nonsensitive variation of physical properties (S,and V,) is achieved. Therefore, it seems reasonable to suppose that the deactivation is produced by active-site coverage, in such a way that the activitycoke content relationship of eq 4 can be applied, where m is again the number of active sites taking part in the surface
a = 0.0108A2
(15)
Conclusions Cyclohexanol dehydration, between 533 and 573 K,on a silica-alumina catalyst is very accurately described by the kinetic model of eq 9, with very narrow confidence intervals of the parameter values, yielding a very good fitting of the experimental data obtained in an integral fiied bed reactor extrapolatedat zero time. As the catalyst activity decays with time, the reaction rate changes with time. Equation 13 is a kinetic model for the deactivation rate, which together with eq 9 describes accurately the experimental data. Both equations combined give
roL =
~KOLPOL X (1 + KoLPoL + KEpE + KwPw12 1
(16) + kd&E9pE3 A+ J 0 ( 1 + KOLPOL + KEpE + KWPW)' uc The kinetic model a t zero time, that is, for a fresh catalyst, can be deduced from a Langmuir-Hinshelwood 1 A L
I
rt
kdOLKO