Modeling the Kinetics of Deactivation of Catalysts during the

Dec 12, 2014 - The fouling of catalysts for the upgrading of bio-oils appears to be very different from the fouling of catalysts for the hydroprocessi...
33 downloads 18 Views 1MB Size
Article pubs.acs.org/EF

Modeling the Kinetics of Deactivation of Catalysts during the Upgrading of Bio-oil Robert S. Weber,* Mariefel V. Olarte, and Huamin Wang Institute for Integrated Catalysis, Pacific Northwest National Laboratory, P. O. Box 999, MS-IN K2-12, Richland, Washington 99352, United States ABSTRACT: The fouling of catalysts for the upgrading of bio-oils appears to be very different from the fouling of catalysts for the hydroprocessing of petroleum-derived streams. There are two reasons for the differences: (a) bio-oil contains polarizable components and phases that can stabilize reaction intermediates exhibiting charge separation and (b) bio-oil components contain functional groups that contain O, notably carbonyls (>CO). Aldol condensation of carbonyls affords very different pathways for the production of oligomeric, refractory deposits than does dehydrogenation/polymerization of petroleum-derived hydrocarbons. Colloquially, we refer to the bio-oil-derived deposits as “gunk” to discriminate them from coke, the carbonaceous deposits encountered in petroleum refining. Classical gelation appears to be a suitable model for the “gunking” reaction. Our work has helped explain the temperature range at which bio-oil should be preprocessed (“stabilized”) to confer longer lifetimes on the catalysts used for more severe processing. Stochastic modeling (kinetic Monte Carlo simulations) appears promising to capture the rates of oligomerization of bio-oil.



literature. Data available only in graphical form were digitized using shareware6 and likely exhibit transcription errors of a few percent. The data analysis (curve fitting) employed standard techniques of data regression available in Microsoft Excel. The kinetic Monte Carlo simulations were run using our own implementation of Gillespie’s original, unaccelerated algorithm7 written in MatLab8 and run on a 64-bit Macintosh computer.

INTRODUCTION The activity of heterogeneous catalysts can degrade in four ways: leaching of the components that comprise the active sites; agglomeration and rearrangement of active sites into less accessible or less active moieties; poisoning of the active sites and fouling of the catalyst surface with a barrier coating that prevents access to the active sites. All four modes occur in the hydrotreating of pyrolytic bio-oil; here we have focused on fouling. The goal is to provide additional insight into “stabilization” of bio-oil1−4preliminary hydrotreating at low temperaturewhich serves to decrease the rate of fouling when the oil is eventually treated under the more severe conditions needed to deoxygenate components, such as hindered phenols and furans. Ultimately, the analysis of the transformation of bio-oiland its effects on catalystsneeds to account for the complexity of the starting material5 (wide ranges of molecular weights of polyfunctional species), the complexity of its processing (uncertain contacting of the viscous oil, hydrogen gas, and a solid catalyst that provide ample opportunity for maldistribution of flow), and thus the complexity and uncertainty of the composition and structure of the product. The analysis here is only a small step toward addressing those complications. The entire exercise will be a “bootstrapping” operation, in which the community will, over time, find a way to incorporate empirical, practicable knowledge into an increasingly detailed framework. Three topics are addressed in the following text: (1) extrapolation of model compound studies to reactions of biooil, (2) evidence for a chemistry that is consistent with the fouling process, and (3) a suggestion for impeding the chemistry of fouling, which involves a support effect that does not appear to have been previously discussed.





RESULTS Model Compounds. To root our analysis in the chemistry of the functional groups that are known to be prevalent in pyrolytic bio-oil,5 we first compared the kinetics of hydrogen uptake of model compounds9 (Table 1) to those of the hydrogenation of a real bio-oil.10 In the latter case, the authors report the amount of hydrogen taken up by their bio-oils over time, from which can be inferred the conversion of the unsaturated or reducible species (their concentration was approximately 2.5 mol/kg of feed, which was the asymptotic consumption of hydrogen reported by De Miguel Mercarder et al.10). The three sets of hydrogen consumption data in Figure 1 were fit simultaneously to regress A and Eact, the preexponential factor and apparent activation energy in an Arrhenius rate constant k(T) = A exp(−Eact/RT), using the integrated form of a normalized, first order rate expression for the conversion, X, as a function of temperature, T and time, t: X = 1 − exp( −k(T )t )

The parity plot (Figure 2) shows that the results are, indeed, consistent with first order kinetics, e.g., zero order in hydrogen and first order in the reactive moieties, with an apparent activation energy of about 42 kJ/mol.

METHODS

Received: November 6, 2014 Revised: December 10, 2014 Published: December 12, 2014

No additional experiments are reported here; rather this work presents further analyses and comparisons of data already published in the © 2014 American Chemical Society

273

dx.doi.org/10.1021/ef502483t | Energy Fuels 2015, 29, 273−277

Energy & Fuels

Article

Table 1. Apparent Activation Energies. Ea, for the Hydrogenation of Compounds Exhibiting the Functionalities Found in Bio-oila

Figure 2. Parity plot for the fit shown in Figure 1, with an apparent activation energy of 42 kJ/mol. a

Data from Grange et al.,2 Yang et al.,9 and He et al.26

Figure 3. Pressure rise during the hydrotreating of unstabilized (catalyst beds A + B1) (red) and stabilized (catalyst beds A + B2) (blue) bio-oil. The solid curves represent fits to the data points assuming Stockmayer gelation kinetics. The dashed lines represent the pressure rise expected for flow through a packed bed by a fluid whose viscosity follows the Mark−Houwink−Sakaruda relation, with simple chain-growth kinetics.

Figure 1. Conversion of the unsaturated species inferred from the consumption of hydrogen during the hydrotreating of a bio-oil, fit with first order kinetics. Data from De Miguel Mercader et al.10

Kinetics of Gunking. The fouling of catalysts during the hydrotreating of bio-oil can be tracked through the increase in pressure across the catalyst bed. Figure 3 shows such results from two different hydrotreating experiments using a continuous-flow, fixed-bed, bench-scale hydrotreater1 with two different bio-oils. The bio-oils were derived from pyrolysis of pine sawdust and had either been “stabilized” with hydrogen at low temperature (∼140°C) over a Ru/C catalyst at ∼140°C or not stabilized. Both bio-oils were then catalytically hydrotreated over a train of two catalyst beds. The first bed in both experiments was a sulfided Ru/C maintained at ∼180°C (bed A). It was followed by a commercial, sulfided, promoted-Mo catalyst (bed B1) for the unstabilized feed and a

mixed bed of PNNL-synthesized and the commercial, sulfided, promoted-Mo catalyst (bed B2) for the stabilized oil, both maintained at ∼400 °C. The curves in Figure 3 represent the time evolution of a scaled pressure head attributed to channel blocking by monomers that oligomerize according to classical gelation kinetics.11 Figure 3 also shows (dashed traces) the scaled pressure heads that would be predicted had the monomers oligomerized linearly instead of gelling. The two dashed traces each assume pressure drops that follow the Ergun equation with fluid viscosities proportional to molecular weight. 274

dx.doi.org/10.1021/ef502483t | Energy Fuels 2015, 29, 273−277

Energy & Fuels

Article

again favoring the ketone). The implication for catalyst deactivation is that, in condensed bio-oil, hydrotreating ineluctably produces intermediates that are susceptible to condensation reactions, for example, aldol condensations13 (Figure 5 top), which result in refractory polymers.1,2,14

An analysis of the kinetics (more precisely, the statistics) of agglomeration of the bio-oil were motivated by observed changes in the molecular weight distribution of a typical, pinederived bio-oil during its hydrotreating.10 The molecular weight distributions had been followed using gel permeation chromatography, which we digitized at a resolution of about 20 Da (Figure 4).

Figure 5. Top, aldol condensation of two phenols; bottom, product of the aldol condensation of three phenols with the aldehyde groups in a fragment of an amylopectin to illustrate how a gel (star polymer) might form.

The levoglucosan produced in the pyrolysis of cellulose15 can thermally polymerize to form branched oligomers.16 We extrapolate that the carboxyl groups in the branches can then undergo aldol condensations to form the sort of star polymers illustrated in the bottom portion of Figure 5. Gelation, the formation of 2-D or 3-D polymers, as opposed to simple, chaingrowth 1-D polymerization is consistent with the sudden rise in bed pressure observed during the hydrotreating of bio-oil (Figure 3). The solid curves in that figure represent a fluid whose molecular weight and viscosity increases according to Stockmayer gelation kinetics.11,17 The dashed curves in that figure represent second order polymerization kinetics for the molecular weight of the bio-oil combined with the Mark− Houwink−Sakurda relation18 between viscosity and molecular weight (with α = 0.8) To further explore the connection between desired activity and catalyst deactivation, we compared (Figure 6) the reported Arrhenius activation energies for the model compounds listed in Table 1 with that estimated for a prototypic aldol condensation, furfural plus acetone.19 The undesired, condensation reaction exhibits an activation energy higher than that of any of the desired, hydrogenation reactions. Therefore, raising the temperature in a hydrotreating reactor will serve to preferentially accelerate the fouling of the catalysts compared to the desired hydrogenations, while lowering the temperature will preclude significant hydrogenation. Therefore, there will be a temperature range useful for achieving some hydrotreating without excessive fouling. Experimentally that temperature was determined to be about 170 °C for the hydrotreating of pinederived bio-oil over a Ru/C catalyst.1 In passing, we note that coking in the refinery upgrading of petroleum-derived feedstocks can exhibit quite low apparent activation energies (∼12 kJ/mol), comparable to activated diffusion.20 The difference in kinetics of formation combined with likely differences in composition have led us to devise the new terms, “gunk” and “gunking” to describe the material that

Figure 4. Molecular weight distributions reported by De Miguel Mercader et al.10 for pine-derived bio-oils either raw (red curve) or hydrotreated (cross-hatched). The solid black curve results from a kinetic Monte Carlo simulation of a second order polymerization of the distribution for the raw bio-oil.



DISCUSSION Utility of Model Compound Studies. If, as suggested by the literature,9 we assume that the overall reaction in the hydrotreating of actual bio-oil is first order in the bio-oil and zero order at high hydrogen pressures, then there is a good fit (Figure 1) of the rate of hydrogen uptake by the bio-oil with a low, apparent activation energy (Figure 2). The low activation energy (ca. 42 kJ/mol) is consistent with the entries in Table 1 that correspond to rapid (low temperature) initial hydrogenations.2 However, it might also represent, in this instance, a suite of reactions whose rates are disguised by mass transport, i.e., at high Thiele number and thus exhibit only half of the intrinsic activation energy. Deactivation Consistent with Aldol Condensation. Full hydrogenation and hydrodeoxygenation of the oxygenates that are abundant in pyrolytic bio-oil produce light alkanes. At low to moderate conversion (i.e., prior to hydrodeoxygenation), however, the reaction will produce alcohols or ketones as intermediate products. Recently,12 we found by modeling the hydrotreating of phenol in vapor and liquid phase that the density and polarizability of the reaction medium affects the selectivity toward the intermediate. In the gas phase the alcohol, cyclohexanol, is preferred; in the liquid phase (modeled as water) the ketone, cyclohexanone. Water stabilizes the charged intermediates in the enol-keto epimerization (and thus decreases the activation energy for the ketonization) and also stabilizes the products (and thus increases the driving force, 275

dx.doi.org/10.1021/ef502483t | Energy Fuels 2015, 29, 273−277

Energy & Fuels

Article

from the entire distribution and an approximately fitted rate constant. The solid black curve represents the results of the KMC, which, encouragingly, appears to capture both the diminution of light species and their combination with heavier species so as to widen the distribution. Further refinements and extensions are underway to include the effects of the other three processes. Impeding the Fouling. Aldol condensation requires that the adding epimer be in its keto form (the enol form can polymerize in other ways). As mentioned earlier, we have found12 that the keto form is stabilized by a polarizable reaction medium, notably by water. Therefore, it is plausible that a catalyst which creates a hydrophilic reaction space near the active site would promote the aldol condensation. In the case of silica−alumina type supports such as zeolites, hydrophilicity corresponds to high Al/Si ratios. However, isolated alumina sites in zeolites are stronger acids than are alumina sites in more concentrated domains.23 Therefore, an optimal silica/aluminum ratio should create the most active catalyst: one likely on the hydrophilic/hydrophobic border (Figure 7) having sites strong

Figure 6. Comparison of reported activation energies for hydrogenation (HYD) and hydrodeoxygenation (HDO) model reactions with that of an aldol condensation (heavy line). Data for the model compound studies come from Yang et al.9 We estimated the apparent activation energy for the aldol condensation (52 kJ/mol), catalyzed by acid zeolites, from data presented in Kikhtyanin et al.19

deposits during the hydrotreating of bio-oil. Unlike the study of hydrogenation discussed earlier, the contribution of model compounds to catalyst deactivation21 may not be representative of the full range of deactivation processes. Deactivation Kinetics. As should be expected from the condensation reactions shown in Figure 5, the deactivation reactions should exhibit a second order dependence on the molecules that carry the polymerization sites (here assumed to be carbonyl groups susceptible to aldol condensation). Those moieties are likely distributed throughout the wide molecular weight distribution (Figure 4) that characterizes pyrolytic biooil, in a pattern that is not yet known. That deficiency hinders the formulation of deterministic kinetics to describe the changes in the molecular weight distributions that arise from applying different process conditions. As a first step toward modeling the chemical processes that alter the molecular weight distributions (MWD), we have tested the utility of stochastic kinetics (kinetic Monte Carlo, KMC). The data in Figure 4 were digitized from the work of De Miguel Mercader et al.,10 who report the changes in molecular weight distributions of pine-derived bio-oils upon hydrotreating them at different temperatures. The red curve represents their starting material, and the cross-hatched curve represents the bio-oil after hydrotreating it at 250 °C, which allowed for at least four classes of reaction: polymerization, thermal cracking, hydrogenation, and hydrodeoxygenation, (both of which likely resulted in light products that no longer contributed to the MWD). Following the suggestion of McDermott and Klein,22 we found it prudent to allow the reaction transition probabilities to depend on the molecular weights of the reactants as a way of accounting for bulk and finite diffusivities of the heavier molecules (i.e., multiplying the probabilities by a Thiele modulus). We only tried, in this preliminary effort, to model the first process: the polymerization. We assumed it proceeded with reaction transition probability that depended on the product of “concentrations” of two reactants randomly selected

Figure 7. Normalized rates of zeolite-catalyzed aldol condensation of furfural and acetone. Reaction kinetics from Kikhtyanin et al.;19 Hydrophilic/hydrophobic divide from Kawai and Tsutsumi.25

and populous enough and a reaction space that stabilizes the critical reaction intermediates. Such an optimum has been found by groups seeking to produce active catalysts for aldol condensation.19,24 But, if the goal is to prevent catalyst fouling, then that is precisely the regime of catalyst composition to avoid. Instead, the preferred catalyst support, all other aspects being equal, would be chosen from the aldol-suppressing wings of the curve shown in Figure 7. Practically, however, the nanoenvironment that suppresses aldol condensations may also interfere with the synthesis or stabilization of the catalytic sites for the desired hydrotreating reactions. A systematic search would require probing the hydrophobicity/hydrophilicity near the catalytic sites under reaction conditions.



CONCLUSION This reanalysis of published data on the reactions related to the upgrading of bio-oil and the concurrent deactivation of the upgrading catalyst has confirmed the likely utility of studies of model compounds and has reinforced the hypothesis that catalyst deactivation can be attributed, in large measure, to 276

dx.doi.org/10.1021/ef502483t | Energy Fuels 2015, 29, 273−277

Energy & Fuels

Article

(23) Barthomeuf, D. In Proceedings of the NATO Advanced Study Institute on Zeolites: Science and Technology; Ribeiro, F. R., Ed.; Nijhoff: The Hague, The Netherlands, 1984; pp 317−346. (24) Climent, M. J. Catal. 2004, 225, 316−326. (25) Kawai, T.; Tsutsumi, K. Colloid Polym. Sci. 1992, 270, 711−715. (26) He, J.; Zhao, C.; Lercher, J. A. J. Catal. 2014, 309, 362−375.

condensation reactions among the partially upgraded components of bio-oil. Stochastic methods appear to be useful for capturing the kinetics of the chemical reactions among the, as yet, poorly characterized reactants, much as they have been proved suitable for describing the depolymerization reactions of lignin.22 We anticipate that we will be able to extend the KMC modeling to include the other processes that likely occur during the hydrotreating of bio-oil and, eventually, to include detailed, speciated composition profiles as well as catalyst deactivation.



AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS



REFERENCES

This work was supported by the U.S. Department of Energy, Office of Energy Efficiency and Renewable Energy, Bioenergy Technologies Office. Pacific Northwest National Laboratory (PNNL) is a multiprogram national laboratory operated for DOE by Battelle.

(1) Elliott, D. C.; Hart, T. R.; Neuenschwander, G. G.; Rotness, L. J.; Olarte, M. V.; Zacher, A. H.; Solantausta, Y. Energy Fuels 2012, 26, 3891−3896. (2) Grange, P.; Laurent, E.; Maggi, R.; Centeno, A.; Delmon, B. Catal. Today 1996, 29, 297−301. (3) Elliott, D. C.; Baker, E. G. Process for Upgrading Biomass Pyrolyzates U.S. Patent 4 795 841, Jan. 3, 1989. (4) Xu, X.; Zhang, C.; Liu, Y.; Zhai, Y.; Zhang, R. Chemosphere 2013, 93, 652−660. (5) Resasco, D. E.; Crossley, S. P. Implementation of concepts derived from model compound studies inthe separation and conversion of bio-oil to fuel. Catal. Today 2014, in press, DOI: http://dx.doi.org/10.1016/j.cattod.2014.06.037. Published Online, 2014. (6) Tummers, B. DataThief III, 2006, http://datathief.org. (7) Gillespie, D. T. J. Phys. Chem. 1977, 81, 2340−2360. (8) MatLab; MathWorks: Natick, MA, USA, 2014. (9) Yang, J.; Williams, C. L.; Ramasubramaniam, A.; Dauenhauer, P. J. Green Chem. 2014, 16, 675−682. (10) De Miguel Mercader, F.; Koehorst, P. J. J.; Heeres, H. J.; Kersten, S. R. A.; Hogendoorn, J. A. AIChE J. 2011, 57, 3160−3170. (11) Stockmayer, W. H. J. Chem. Phys. 1943, 11, 45−55. (12) Yoon, Y.; Rousseau, R.; Weber, R. S.; Mei, D.; Lercher, J. A. J. Am. Chem. Soc. 2014, 136, 10287−10298. (13) Gangadharan, A.; Shen, M.; Sooknoi, T.; Resasco, D. E.; Mallinson, R. G. Appl. Catal., A 2010, 385, 80−91. (14) Elliott, D. C. Energy Fuels 2007, 21, 1792−1815. (15) Lakshmanan, C. M.; Hoelscher, H. E. Ind. Eng. Chem. Prod. Res. Dev. 1970, 9, 57−59. (16) Houminer, Y.; Patai, S. J. Polym. Sci., Part A-1: Polym. Chem. 1969, 7, 3005−3024. (17) Ziff, R. M. J. Stat. Phys. 1980, 23, 241−263. (18) Rai, P.; Rosen, S. L. J. Polym. Sci., Part B: Polym. Phys. 1997, 35, 1985−1987. (19) Kikhtyanin, O.; Kelbichová, V.; Vitvarová, D.; Kubů, M.; Kubička, D. Catal. Today 2014, 227, 154−162. (20) Wolf, E. E.; Alfani, F. Catal. Rev.: Sci. Eng. 1982, 24, 329−371. (21) Gao, D.; Schweitzer, C.; Hwang, H. T.; Varma, A. Ind. Eng. Chem. Res. 2014, 53, 18658−18667. (22) McDermott, J. B.; Klein, M. T. Chem. Eng. Sci. 1986, 41, 1053− 1060. 277

dx.doi.org/10.1021/ef502483t | Energy Fuels 2015, 29, 273−277