- Email: [email protected]
Sintering of A Ni-Based Catalyst During Co Hydrogenation: Kinetics and Modeling
C.H. Bartholomew and J.B. Butt (Editors ), C*ata/ystDeactiuation 1991 0 1991 Elsevier Science Publishers B.V., Amsterdam
605
SINTERING OF A NI-BASED CATALYST DURING CO HYDROGENATION: KINETICS AND MODELING
M. AGNELLI. M. KOLB*. c. NICOT and c. MIRODATOS. Institut d e Recherches sur la catalyse. 2 Av. A. Einstein, 69626 Villeurbanne Cedex, France.
* Ecole Normale SupCrieure de Lyon. 46 Allee d’ltalie, 69364 Lyon Cedex 07. France. SUMMARY The aging of a Ni/SiO2 catalyst during low temperature CO-H7 methanation is shown to be primarily related to N i sintering. It was observed that the initial honomodal distribution of N i articles was readil transformed into a bimodal distribution (small spherical particles and yarge faceted crystaiites), the nickel transfer from article to article being ensured via nickel carbonyl species. No single theoretical model has Keen f o u n f t o apply to the particle growth kinetics, attesting the complexity of the overall sintering mechanism. INTRODUCTION Sintering of supported metal particles is of general concern in catalyst aging (refs. I , 2); however it is mostly related to high temperature processes involving a complex combination of metal migration such as particle motion/coalescence and metal evaporation/condensation (ref. 3 ) . In a previous study (ref. 4). we have shown that metal sintering might occur at low temperatures on Ni/SiO2 during CO hydrogenation into methane, most likely via nickel carbonylation i.e. a sintering process involving only a molecular transport of nickel. This was confirmed later by Van Stiphout for similar conditions (ref. 5 ) . However no detailed mechanism was proposed for this specific case due to insufficient data on morphological changes during methanation. Accordingly, this system has been chosen for a combined experimental and modeling approach of sintering in relation with catalyst deactivation, provided that no other form of deactivation such as carbon deposition was significantly interfering with the process. Precise particle size distributions as a function of time on stream have been obtained both from magnetic measurements and electron microscopy. Several basic models can be considered for describing the evolution of metal particles far from the equilibrium. One important class of models describes the dynamics of phase separation (one distinguishes between nucleation/growth and spinodal decomposition) (ref.6). In this process growth is induced by the thermodynamic interaction between macroscopically coexisting phases. Each of the phases is assumed to be in local equilibrium. I n its simplest form domains of high density (solid) coexist with a low density phase (gas) and the domains grow or disappear via an evaporation/condensation mechanism. A generalisation includes particle diffusion/coagulation processes (ref.7). A different type of growth is irreversible aggregation (diffusion- or reaction limited) (ref.8). In these mechanisms no relaxation is possible and hence one never reaches an equilibrium state. They primarily describe disordered processes leading to ramified structures.
606
The applicability of such models to the present case of nickel sintering at low temperature has been tested aiming at identifying the pertaining mechanism(s).
EXPERIMENTAL Catalyst preparation. A 15 wt% Ni/SiO2 catalyst was prepared by impregnating silica (Aerosil Degussa 200 m2g-l) with [Ni(NH3)6]+2 according to ref. 4. The dried precursor was reduced "in situ" in flowing H2 at 650°C for 15h, the temperature being raised at 2OC mn-l. C O hydrogenation. Methanation reaction (flowing CO + 3 H2 mixture) was carried out at 230-C (with additioaal runs at 45OoC for checking temperature effects) in a fixed bed flow reactor allowing in situ magnetic measurements. Catalytic runs were always started with fresh catalyst; after given times on stream, the reactor was flushed with He and cooled down to room temperature and the used samples were analyzed by electron microscopy and magnetic measurements. Transmission electron microscopy. In order to get images suitable for accurate particle size distribution, the catalyst samples were embedded in Epon resin, then cut in a section thinner than 50 nm. Highly contrasted micrographs were then obtained on a JEOL JEM lOOCX instrument. Each histogram was worked out from a statistical analysis of around 1000 particles. Magnetic measurements Magnetic measurements based on the Weiss extraction method (ref.9) were performed in an electromagnet providing fieids up to 21 kOe. From the saturation magnetization, the degree of nickel reduction was determined, allowing also to check any possible loss of nickel during the reaction. For nickel particles remaining in the superparamagnetic domain (diameter <15nm), two average diameters of particle were determined: D1 at high field, corresponding to small particles and D2 at low field, corresponding to large particles (ref.10).
RESULTS Magnetic measurements revealed that the nickel reduction was completed with an average size of particle of ca. 4nm after the initial reduction. Fig. 1 reports the changes in particle diameters as a function of time on stream for methanation at 230 and 450°C. At 230°C, a fraction of particles (curve D2) grow in size with time on stream while another fraction of small particles grow only slightly (curve D 1). However, after methanation at 450°C, the size of nickel particles remains unchanged (curves D1' and D2'). Magnetization data for the used samples (after regeneration of the metallic phase by hydrogenation of carbonaceous or oxygenated species formed during the catalytic run) indicate no loss of nickel during reaction, at any temperature.
607
I_.
E, r 0)
12 lo',
21
0
8
5
10 time [h]
- -- D 2 d - -
15
20
Fig. I . Average diameters determined from magnetic measurements as a function of time on stream at 2300C (D1 and D2) and at 45OoC (D1' and D2'). Dotted lines indicate samples in which particle size exceeds the superparamagnetic domain (diam. larger than 15 nm). Figs. 2a. 2b and 2c report the micrographs of the catalyst after reduction, after l h and after 16h on methanation stream a t 230°C, respectively. Initially (Fig. 2a) the particles are small, uniformly distributed with a clearly spherical shape. After l h on stream (Fig. 2b), the particles are less contrasted, \lightly larger, the largest ones having lost their sphericity. After 16h on stream (Fig. 2c), large faceted crystallites are now present together with the previous small particles. Particle size distributions obtained from micrographs after different times on stream are presented in Fig. 3. In agreement with the previous statements, the initially narrow distribution of nickel particles is broadened and the total number of particles (Table 1 ) is decreased after l h contact with the reacting mixture. After 3h on stream, a second distribution of larger particles develops while the initial distribution still broadens \lowly. After 16h on stream, the first distribution of this bimodal system tends to stabilize, although larger faceted particles are still appearing in the second distribution. A magnetic analysis after 96h on stream confirmed that the large particles were still developing at that time. Table 1 allowr to check that the total number of particles decreases with time on stream in agreement with the process of growth. Evaluating the amount of nickel pertaining to both distributions shows that the second distribution concerns only a limited fraction of the nickel phase (ca. 9% after 16h).
DISCUSSION Effect of nickel sintering on catalyst aging. Before analyzing and modeling the process of nickel sintering in the methanation conditions, it seems worthwhile to evaluate its effect on catalyst aging. After 16h on stream, mol/h/g), without significant the intrinsic activity dropped to 47% of its initial value (4.7 change in selectivity; meanwhile, the metallic dispersion evaluated from the difference in magnetization between evacuated and rehydrogenated samples decreased from 0.33 to 0.19. representing a loss of 4276, which compares well with the loss of activity. Moreover, the
608
FIG.2. T.E.M. micrographs (a) after reduction, (b) after 1 h on stream methanation ,(c) after 16 h on stream methanation .
609
TABLE 1. Average number of particles [number of part. /p$] after different time on stream [h]
I -E
0
a
3
5
10
"
16 2
3
4
5
6
7
8
diameter [nm]
9
(0
I number
11
;::K 10
6
diameter [nm] 30
-s. F
1
6
3
3
4
2
diameter [nm]
5
4
1
FJ
20
I
10
0
10
2
3
4
5
6
7
8
9
10
11
2
3
4
5
6
7
B
9
10
11
diameter [nm]
diameter [nm]
of
75000 54000 43000 42000 38000
1
20
. c
(number of
time
n
6
1
6
3
6
1
6
3
0
4
2
5
4
6
6
0
4
2
5
4
6
6
diameter [nm]
diameter [nm]
FIG.3. Particle size distributions a t different time on stream.
610
kinetics of deactivation and of metallic surface ioss paralIeled satisfactorily (ref. 8). As no toxic carbon building was detected by temperature programmed hydrogenation of used samples (beside the formation of the active Ni3C carbidic layer) (refs. 4,5,11), it may be inferred that for these conditions the aging process is essentially due to the loss of active wrface by Ni sintering. Origin of the low temperature sintering. The absence of any sintering phenomenon at 450°C enables ruling out any thermal process such as particle migration or metal vaporization at 230°C. A chemical process for nickel sintering at low temperature has therefore to be accounted for. It was shown in ref. 4 that sintering is CO pressure dependent but insensitive to other gases such as H 2 0 or CH4; it was then inferred that nickel is likely to be transported via carbonyl or subcarbonyl species migrating from particle to particle. This hypothesis has been recently strengthened by the two following observations: (i) EELS measurements have shown that a carbon species appears on the silica surface (even in areas free from Ni particles) in samples in contact with a CO/H2 mixture at 230°C. This species is not detected on samples reacted at 450°C, i.e. when sintering doesn't occur (ref. 11). (ii) When starting from a nickel-copper alloy (Ni/Cu=0.13), the large particles formed in CO/H2 at 24OOC are found to be nearly pure nickel from STEM analysis, in agreement with a mechanism of selective chemical extraction and transport of nickel via carbonylation (ref. 1I). According to these observations, the state of nickel in the reaction conditions may be therefore described as a bimodal distribution of small and large metal particles surrounded by an atomically dispersed phase allowing the inter-particle transfer. Based on the observed changes with time on stream, modeling of this system may be undertaken. Modeling Theoretical investigations of sintering models show that some features are universal; they do not depend on details such as the precise shape of the interaction potential. For example, the particle size grows with time as a power law, with a universal exponent. A very useful concept is the scaling representation for functions such as the particle size distribution h(x,t). The scaling hypothesis postulates that the time dependence of h(x,t) only enters via a single relevant length x(t) (i.e. the average particle size). Then h(x,t) can be represented by a scaling function Q(z) which depends on the variable z=x/x. In order to test whether the experimental distribution satisfies this scaling hypothesis one determines cD(x/x) =xa+ Ih(x,t). The' factor xa+l in the definition of @(z) assures identical normalizations for @ and h, Zxxczh(x,t)=Z,,za@(z)=constant. a=O is appropriate if the total number of particles is conserved;a=3 for conservation of the total mass. If the function @(z) determined from the experiments is time independent. the size distribution is scale invariant. The scale invariance is a prerequisite for a growth process to be describable in terms of a specific mechanism. From a theoretical point of view the scaling hypothesis is valid in the long time limit. In the present experiments a monomodal distribution of particles is observed initially. This distribution results from the sintering process during H2 activation at high temperature. A different type of sintering process is induced by the reaction. Because the sintering is now induced chemically, the coexisting equilibrium phases have a different structure. At early
611
stages of the experiment, the initial particles restructure and a new equilibrium is reached in less than lh. The next step in the evolution is the nucleation and growth of large crystalline particles with well defined faceted surfaces. These crystals grow at the expense of the smaller ones. A bimodal distribution appears. With the development of the second distribution the first one diminishes. Apparently the true equilibrium phase is the crystalline phase of the large particles. After several hours the growth of the large particles slows down. This may be due to structural limitation of the support (for instance, it would be difficult to transfer Ni from a silica grain to another one). Alternatively this slowing down may be caused by a lack of atomic Ni from the small particles (whose surface has fully restructured). But even without such restrictions the sintering process is expected to slow down with time. It may be first noted that from a theoretical point of view different structures may coexist during phase separation. In particular, Monte Carlo calculations have shown that surface diffusion tends to favor the formation of crystalline particles (ref. 12). A second point is to determine if only one or several mechanisms are dominating the observed growth of particles, whether small or large. We have seen previously that the elementary growth processes exhibit a scaling behaviour (refs. 12,13). Accordingly, the experimental particle size distributions have been represented in scaled form, to test if all the data collapse on a single, universal master curve or not. The distributions of the small and of the large faceted particles were treated separately. Depending on the process, either the total number of particles or the total mass of nickel related to a given distribution should be conserved. The experimental data are therefore plotted using both assumptions, i.e. normalizing with an exponent a = 0 and a = 3 respectively. No distribution is found to collapse exactly on a unique curve; however, a reasonable fit is obtained for the large particles using the particle conservation model (a= O), as shown in Fig. 4.
21 31
0
0
20
40
diameter
60
80
0
i
1
2
scaled diameter
3
Fig. 4. Scaling procedure for the large particle d'stribution (a=O).On the left, the unscaled distribution i.e the number of particles per pm versus diameter in nm. On the right, the scaled distribution (dimensionless units). 0 3honstream + Shonstream + 16h on stream.
612
These findings support the idea that the observed growth is in a transitory regime between the initial high temperature sintering and the more stable reaction induced sintering at low temperature. In summary, the available information on the metallic particles suggests that the observed bimodal distribution is a consequence of a change of the equilibrium state at low temperature. This change is activated chemically. Equilibration proceeds in several steps, each on a larger time scale: 1) restructuring of the original particles, 2) nucleation/growth of well crystallized equilibrium particles, 3) slow evolution of the large particles following a model which conserves the total number of particles. CONCLUSION In the present study, the different states of nickel during methanation at low temperature and their role in the sintering process have been clearly identified. In a preliminary theoretical approach, no unique model has been found to apply to the particle growth kinetics, ruling out the hypothesis of a single mechanism. Further work is in progress for a more complete modeling, for instance accounting for the morphological transition between spherical particles and faceted ones. It may be anticipated however that the sintering process should eventually lead to a very poor Ni dispersion. Solutions aiming at slowing down the process such as hindering the carbonyl formation and diffusion onto the support (e.g. by modifying its basicity) have therefore to be considered. ACKNOWLEDGMENT The french-argentine program P.I.C.S. is fully acknowledged for supporting part of this work and granting a fellowship to M.A. REFERENCES 1 2 3 4 5 6
7
8 9 10 11 12 13
B. Pulvermacher and E. Ruckenstein, J.Catal., 35 (1974) 115-139. P.C. Flynn and S.E. Wanke, J. Catal. 37 (1975) 432-448. D.B. Dadvburior. in Catalvst Deactivation 1987, B. Delmon and G.F. Froment (Eds.) Elsevier,
605
SINTERING OF A NI-BASED CATALYST DURING CO HYDROGENATION: KINETICS AND MODELING
M. AGNELLI. M. KOLB*. c. NICOT and c. MIRODATOS. Institut d e Recherches sur la catalyse. 2 Av. A. Einstein, 69626 Villeurbanne Cedex, France.
* Ecole Normale SupCrieure de Lyon. 46 Allee d’ltalie, 69364 Lyon Cedex 07. France. SUMMARY The aging of a Ni/SiO2 catalyst during low temperature CO-H7 methanation is shown to be primarily related to N i sintering. It was observed that the initial honomodal distribution of N i articles was readil transformed into a bimodal distribution (small spherical particles and yarge faceted crystaiites), the nickel transfer from article to article being ensured via nickel carbonyl species. No single theoretical model has Keen f o u n f t o apply to the particle growth kinetics, attesting the complexity of the overall sintering mechanism. INTRODUCTION Sintering of supported metal particles is of general concern in catalyst aging (refs. I , 2); however it is mostly related to high temperature processes involving a complex combination of metal migration such as particle motion/coalescence and metal evaporation/condensation (ref. 3 ) . In a previous study (ref. 4). we have shown that metal sintering might occur at low temperatures on Ni/SiO2 during CO hydrogenation into methane, most likely via nickel carbonylation i.e. a sintering process involving only a molecular transport of nickel. This was confirmed later by Van Stiphout for similar conditions (ref. 5 ) . However no detailed mechanism was proposed for this specific case due to insufficient data on morphological changes during methanation. Accordingly, this system has been chosen for a combined experimental and modeling approach of sintering in relation with catalyst deactivation, provided that no other form of deactivation such as carbon deposition was significantly interfering with the process. Precise particle size distributions as a function of time on stream have been obtained both from magnetic measurements and electron microscopy. Several basic models can be considered for describing the evolution of metal particles far from the equilibrium. One important class of models describes the dynamics of phase separation (one distinguishes between nucleation/growth and spinodal decomposition) (ref.6). In this process growth is induced by the thermodynamic interaction between macroscopically coexisting phases. Each of the phases is assumed to be in local equilibrium. I n its simplest form domains of high density (solid) coexist with a low density phase (gas) and the domains grow or disappear via an evaporation/condensation mechanism. A generalisation includes particle diffusion/coagulation processes (ref.7). A different type of growth is irreversible aggregation (diffusion- or reaction limited) (ref.8). In these mechanisms no relaxation is possible and hence one never reaches an equilibrium state. They primarily describe disordered processes leading to ramified structures.
606
The applicability of such models to the present case of nickel sintering at low temperature has been tested aiming at identifying the pertaining mechanism(s).
EXPERIMENTAL Catalyst preparation. A 15 wt% Ni/SiO2 catalyst was prepared by impregnating silica (Aerosil Degussa 200 m2g-l) with [Ni(NH3)6]+2 according to ref. 4. The dried precursor was reduced "in situ" in flowing H2 at 650°C for 15h, the temperature being raised at 2OC mn-l. C O hydrogenation. Methanation reaction (flowing CO + 3 H2 mixture) was carried out at 230-C (with additioaal runs at 45OoC for checking temperature effects) in a fixed bed flow reactor allowing in situ magnetic measurements. Catalytic runs were always started with fresh catalyst; after given times on stream, the reactor was flushed with He and cooled down to room temperature and the used samples were analyzed by electron microscopy and magnetic measurements. Transmission electron microscopy. In order to get images suitable for accurate particle size distribution, the catalyst samples were embedded in Epon resin, then cut in a section thinner than 50 nm. Highly contrasted micrographs were then obtained on a JEOL JEM lOOCX instrument. Each histogram was worked out from a statistical analysis of around 1000 particles. Magnetic measurements Magnetic measurements based on the Weiss extraction method (ref.9) were performed in an electromagnet providing fieids up to 21 kOe. From the saturation magnetization, the degree of nickel reduction was determined, allowing also to check any possible loss of nickel during the reaction. For nickel particles remaining in the superparamagnetic domain (diameter <15nm), two average diameters of particle were determined: D1 at high field, corresponding to small particles and D2 at low field, corresponding to large particles (ref.10).
RESULTS Magnetic measurements revealed that the nickel reduction was completed with an average size of particle of ca. 4nm after the initial reduction. Fig. 1 reports the changes in particle diameters as a function of time on stream for methanation at 230 and 450°C. At 230°C, a fraction of particles (curve D2) grow in size with time on stream while another fraction of small particles grow only slightly (curve D 1). However, after methanation at 450°C, the size of nickel particles remains unchanged (curves D1' and D2'). Magnetization data for the used samples (after regeneration of the metallic phase by hydrogenation of carbonaceous or oxygenated species formed during the catalytic run) indicate no loss of nickel during reaction, at any temperature.
607
I_.
E, r 0)
12 lo',
21
0
8
5
10 time [h]
- -- D 2 d - -
15
20
Fig. I . Average diameters determined from magnetic measurements as a function of time on stream at 2300C (D1 and D2) and at 45OoC (D1' and D2'). Dotted lines indicate samples in which particle size exceeds the superparamagnetic domain (diam. larger than 15 nm). Figs. 2a. 2b and 2c report the micrographs of the catalyst after reduction, after l h and after 16h on methanation stream a t 230°C, respectively. Initially (Fig. 2a) the particles are small, uniformly distributed with a clearly spherical shape. After l h on stream (Fig. 2b), the particles are less contrasted, \lightly larger, the largest ones having lost their sphericity. After 16h on stream (Fig. 2c), large faceted crystallites are now present together with the previous small particles. Particle size distributions obtained from micrographs after different times on stream are presented in Fig. 3. In agreement with the previous statements, the initially narrow distribution of nickel particles is broadened and the total number of particles (Table 1 ) is decreased after l h contact with the reacting mixture. After 3h on stream, a second distribution of larger particles develops while the initial distribution still broadens \lowly. After 16h on stream, the first distribution of this bimodal system tends to stabilize, although larger faceted particles are still appearing in the second distribution. A magnetic analysis after 96h on stream confirmed that the large particles were still developing at that time. Table 1 allowr to check that the total number of particles decreases with time on stream in agreement with the process of growth. Evaluating the amount of nickel pertaining to both distributions shows that the second distribution concerns only a limited fraction of the nickel phase (ca. 9% after 16h).
DISCUSSION Effect of nickel sintering on catalyst aging. Before analyzing and modeling the process of nickel sintering in the methanation conditions, it seems worthwhile to evaluate its effect on catalyst aging. After 16h on stream, mol/h/g), without significant the intrinsic activity dropped to 47% of its initial value (4.7 change in selectivity; meanwhile, the metallic dispersion evaluated from the difference in magnetization between evacuated and rehydrogenated samples decreased from 0.33 to 0.19. representing a loss of 4276, which compares well with the loss of activity. Moreover, the
608
FIG.2. T.E.M. micrographs (a) after reduction, (b) after 1 h on stream methanation ,(c) after 16 h on stream methanation .
609
TABLE 1. Average number of particles [number of part. /p$] after different time on stream [h]
I -E
0
a
3
5
10
"
16 2
3
4
5
6
7
8
diameter [nm]
9
(0
I number
11
;::K 10
6
diameter [nm] 30
-s. F
1
6
3
3
4
2
diameter [nm]
5
4
1
FJ
20
I
10
0
10
2
3
4
5
6
7
8
9
10
11
2
3
4
5
6
7
B
9
10
11
diameter [nm]
diameter [nm]
of
75000 54000 43000 42000 38000
1
20
. c
(number of
time
n
6
1
6
3
6
1
6
3
0
4
2
5
4
6
6
0
4
2
5
4
6
6
diameter [nm]
diameter [nm]
FIG.3. Particle size distributions a t different time on stream.
610
kinetics of deactivation and of metallic surface ioss paralIeled satisfactorily (ref. 8). As no toxic carbon building was detected by temperature programmed hydrogenation of used samples (beside the formation of the active Ni3C carbidic layer) (refs. 4,5,11), it may be inferred that for these conditions the aging process is essentially due to the loss of active wrface by Ni sintering. Origin of the low temperature sintering. The absence of any sintering phenomenon at 450°C enables ruling out any thermal process such as particle migration or metal vaporization at 230°C. A chemical process for nickel sintering at low temperature has therefore to be accounted for. It was shown in ref. 4 that sintering is CO pressure dependent but insensitive to other gases such as H 2 0 or CH4; it was then inferred that nickel is likely to be transported via carbonyl or subcarbonyl species migrating from particle to particle. This hypothesis has been recently strengthened by the two following observations: (i) EELS measurements have shown that a carbon species appears on the silica surface (even in areas free from Ni particles) in samples in contact with a CO/H2 mixture at 230°C. This species is not detected on samples reacted at 450°C, i.e. when sintering doesn't occur (ref. 11). (ii) When starting from a nickel-copper alloy (Ni/Cu=0.13), the large particles formed in CO/H2 at 24OOC are found to be nearly pure nickel from STEM analysis, in agreement with a mechanism of selective chemical extraction and transport of nickel via carbonylation (ref. 1I). According to these observations, the state of nickel in the reaction conditions may be therefore described as a bimodal distribution of small and large metal particles surrounded by an atomically dispersed phase allowing the inter-particle transfer. Based on the observed changes with time on stream, modeling of this system may be undertaken. Modeling Theoretical investigations of sintering models show that some features are universal; they do not depend on details such as the precise shape of the interaction potential. For example, the particle size grows with time as a power law, with a universal exponent. A very useful concept is the scaling representation for functions such as the particle size distribution h(x,t). The scaling hypothesis postulates that the time dependence of h(x,t) only enters via a single relevant length x(t) (i.e. the average particle size). Then h(x,t) can be represented by a scaling function Q(z) which depends on the variable z=x/x. In order to test whether the experimental distribution satisfies this scaling hypothesis one determines cD(x/x) =xa+ Ih(x,t). The' factor xa+l in the definition of @(z) assures identical normalizations for @ and h, Zxxczh(x,t)=Z,,za@(z)=constant. a=O is appropriate if the total number of particles is conserved;a=3 for conservation of the total mass. If the function @(z) determined from the experiments is time independent. the size distribution is scale invariant. The scale invariance is a prerequisite for a growth process to be describable in terms of a specific mechanism. From a theoretical point of view the scaling hypothesis is valid in the long time limit. In the present experiments a monomodal distribution of particles is observed initially. This distribution results from the sintering process during H2 activation at high temperature. A different type of sintering process is induced by the reaction. Because the sintering is now induced chemically, the coexisting equilibrium phases have a different structure. At early
611
stages of the experiment, the initial particles restructure and a new equilibrium is reached in less than lh. The next step in the evolution is the nucleation and growth of large crystalline particles with well defined faceted surfaces. These crystals grow at the expense of the smaller ones. A bimodal distribution appears. With the development of the second distribution the first one diminishes. Apparently the true equilibrium phase is the crystalline phase of the large particles. After several hours the growth of the large particles slows down. This may be due to structural limitation of the support (for instance, it would be difficult to transfer Ni from a silica grain to another one). Alternatively this slowing down may be caused by a lack of atomic Ni from the small particles (whose surface has fully restructured). But even without such restrictions the sintering process is expected to slow down with time. It may be first noted that from a theoretical point of view different structures may coexist during phase separation. In particular, Monte Carlo calculations have shown that surface diffusion tends to favor the formation of crystalline particles (ref. 12). A second point is to determine if only one or several mechanisms are dominating the observed growth of particles, whether small or large. We have seen previously that the elementary growth processes exhibit a scaling behaviour (refs. 12,13). Accordingly, the experimental particle size distributions have been represented in scaled form, to test if all the data collapse on a single, universal master curve or not. The distributions of the small and of the large faceted particles were treated separately. Depending on the process, either the total number of particles or the total mass of nickel related to a given distribution should be conserved. The experimental data are therefore plotted using both assumptions, i.e. normalizing with an exponent a = 0 and a = 3 respectively. No distribution is found to collapse exactly on a unique curve; however, a reasonable fit is obtained for the large particles using the particle conservation model (a= O), as shown in Fig. 4.
21 31
0
0
20
40
diameter
60
80
0
i
1
2
scaled diameter
3
Fig. 4. Scaling procedure for the large particle d'stribution (a=O).On the left, the unscaled distribution i.e the number of particles per pm versus diameter in nm. On the right, the scaled distribution (dimensionless units). 0 3honstream + Shonstream + 16h on stream.
612
These findings support the idea that the observed growth is in a transitory regime between the initial high temperature sintering and the more stable reaction induced sintering at low temperature. In summary, the available information on the metallic particles suggests that the observed bimodal distribution is a consequence of a change of the equilibrium state at low temperature. This change is activated chemically. Equilibration proceeds in several steps, each on a larger time scale: 1) restructuring of the original particles, 2) nucleation/growth of well crystallized equilibrium particles, 3) slow evolution of the large particles following a model which conserves the total number of particles. CONCLUSION In the present study, the different states of nickel during methanation at low temperature and their role in the sintering process have been clearly identified. In a preliminary theoretical approach, no unique model has been found to apply to the particle growth kinetics, ruling out the hypothesis of a single mechanism. Further work is in progress for a more complete modeling, for instance accounting for the morphological transition between spherical particles and faceted ones. It may be anticipated however that the sintering process should eventually lead to a very poor Ni dispersion. Solutions aiming at slowing down the process such as hindering the carbonyl formation and diffusion onto the support (e.g. by modifying its basicity) have therefore to be considered. ACKNOWLEDGMENT The french-argentine program P.I.C.S. is fully acknowledged for supporting part of this work and granting a fellowship to M.A. REFERENCES 1 2 3 4 5 6
7
8 9 10 11 12 13
B. Pulvermacher and E. Ruckenstein, J.Catal., 35 (1974) 115-139. P.C. Flynn and S.E. Wanke, J. Catal. 37 (1975) 432-448. D.B. Dadvburior. in Catalvst Deactivation 1987, B. Delmon and G.F. Froment (Eds.) Elsevier,