Angewandte Chemie- International Edition, 38, 2794-2799 (1999) - Designated as HOT PAPER

High-throughput Testing of Heterogeneous Catalyst Libraries Using Array Microreactors and Mass Spectrometry^**

Combinatorial chemistry holds the promise to significantly accelerate the pace of research for the discovery and optimization of heterogeneous catalysts and to contribute to our understanding of catalytic function. However, in order to realize this potential advances must be made in two complementary areas: 1. The rapid generation and processing of a large diversity of structurally and compositionally different solid state materials, and 2. High throughput testing of these libraries for desired catalytic properties.

The preparation of solid state libraries can be accomplished using techniques that can basically be categorized into two major groups: 1. Thin film deposition based methods of synthesis,^[1-4] and 2. Solution based methods of preparation of combinatorial libraries.^[5-7] These library preparation techniques recently were used for the discovery and optimization of superconducting ^[1], magnetoresistive ^[2] and luminescent ^[3-5,8] materials, electrochemical ^[6] and heterogeneous catalysts.^[7,9]

Once solid state catalyst libraries are prepared, they have to be tested for activity, selectivity, resistance to poisoning and operational lifetimes to identify leads that are of practical significance. To date several methods have been proposed to screen heterogeneous catalyst libraries: 1. In-situ optical methods based on infrared (IR) thermography,^[10,11] fluorescence indicator,^[6] and resonance enhanced multiphoton ionization (REMPI) spectroscopy;^[7,12] 2.Probe sampling followed by mass spectrometry (MS).^[9]

Optical methods allow the acquisition of high-speed data on the activities of catalysts in libraries without the need to take samples. However, their applications present some challenges. For example, in the case of IR thermography, no information on selectivity or specific reaction products are generated. The fluorescence screening requires that either the products must be fluorescent or that product specific fluorescent indicators are available or can be developed for each product. Among the optical methods, REMPI represents the most broadly applicable catalyst screening technique. However, in this case the development and availability of suitable REMPI strategies for the desired reaction products are necessary.^[7]

On the other hand, mass spectrometry is a mature and universal detection technology and can readily be used to analyze complex gaseous mixtures. However, its application to combinatorial catalysis requires the development and implementation of new strategies. Recently, Cong et al.^[9] reported a mass spectrometer based system to screen heterogeneous catalyst libraries in a sequential fashion. In that work, the catalyst library consisted of small circular patches of films deposited on a non-porous silica wafer using sputtering with masks. Each catalyst site on the library was sequentially heated to the desired reaction temperature by a CO₂ laser beam. The reactant gases were then blown on to the catalyst site through the annular section of a double concentric probe in a stagnation flow manner. Reaction products emanating from the catalyst site were then withdrawn through the innermost tubing of the probe and analyzed by quadrupole mass spectrometry. After the completion of the screening of a particular site, the library was physically moved to test the next site. The total time to heat and screen one catalyst site was reported to be about 1 minute.

It is important to recognize that the sequential screening system developed by Cong et al.^[9] is slow and is only useful to determine the initial activities and selectivities of catalysts. Consequently, it will be of limited utility to identify leads for practical applications. The limitation arises from the fact that the activities and selectivities of heterogeneous catalysts significantly change (decrease and sometimes increase) with time on stream, which renders short time data practically useless to assess the long time performance of catalysts.

Array microreactors developed earlier in our laboratories are well suited to monitor the activities and selectivities of heterogeneous catalyst libraries over extended periods of time in a parallel fashion.^[7] Recently we also coupled our array microreactors with capillary micro-probe sampling and on-line mass spectrometry for the high throughput testing of the performance of heterogeneous catalyst libraries. In this paper we report the results of time on stream testing of a 66 combination ternary Pt-Pd-In library for the catalytic dehydrogenation of cyclohexane into benzene, a system that was previously studied using REMPI screening as well.^[7]

The catalyst library consisted of 66 ternary combinations of Pt, Pd and In prepared in 0.1% wt increments of each with 1 wt % total metal loading on alumina using the conventional co-impregnation method. Pellets were prepared by compacting 30 mg of high surface area g -Al₂O₃ powder (Alfa Aesar, Ward Hill, MA, 150 m²/gm) into a cylindrical shape (0.4 cm diameter and 0.1 cm height) using a die. These pellet dimensions were chosen to allow for easy characterization and scale up of newly discovered and/or optimized catalysts for commercial applications. Metal precursor stock solutions were prepared from high purity H₂PtCl₆, PdCl₂, InCl₂ and aqueous HCl (Alfa Aesar, Ward Hill MA). The solution library was prepared automatically by mixing predetermined volumes of different catalyst precursor solutions into individually addressable test tubes in an array using a computer controlled x-y-z translation table and a high accuracy liquid delivery system. Alumina pellets were then added into each test tube and were allowed to impregnate under competition with HCl. Subsequently, the solutions were slowly evaporated and the pellets dried, and then calcined at 500^oC for 2 hr.

Catalyst reduction and testing were accomplished in a modified version of the array microreactors described previously.^[7] In the present design, each reactor array consisted of 20 rectangular channels that were micromachined on a flat non-porous silica ceramic slab (7.5 cm x 3.75 cm x 0.63 cm). Channels were 0.1 cm wide, 0.1cm deep and 2 cm long, and possessed a cylindrical well (0.4 cm diameter by 0.2 cm deep) to hold the cylindrical catalyst pellets (0.4 cm diameter and 0.1 cm high). In this configuration, the flat top surface of the catalyst pellet, when placed into the well would be flush with the bottom of the reactor channel. A similar flat ceramic slab was used to cover the microreactor array slab and to form isolated channels. The flat and rigid nature of the ceramic slabs also result in the formation of a good gas seal, thereby isolating each channel from one another under a broad range of temperatures. This design results in unobstructed gas flow rates through the channels with only wall skin friction providing flow resistance. Consequently, the establishment of nearly identical flow rates in each channel, and therefore identical contact times between the reactant gases and the catalyst pellets in each microreactor were feasible. Gas flow rates in the channels were determined to be within 5% of one another using hot-wire anemometry.

In the current arrangement 4 microreactor arrays, each having 20 channels, were stacked and placed inside an aluminum heating block as shown in Figure 1. Thus it was possible to test 80 different catalysts in parallel. Clearly, a larger number of catalysts can easily be tested by stacking more arrays or by increasing the number of channels in each array. As seen in Figure 1, one of the arrays has been pulled out to show internal details. The heating block was precision machined from a single piece of aluminum. As can be seen in Figure 1, each array was surrounded with 1.25 cm aluminum metal walls to ensure temperature uniformity and compact design. Individual microreactor arrays were also fitted with dedicated quartz feed lines to separately adjust their flow rates. However, in the present set of experiments feed gas flow rates were maintained the same for each array. Feed lines pass through a preheater system that was also machined from a solid aluminum block. The reactor and preheat blocks were thermally insulated using a porous ceramic board as shown in Figure 1. It should be noted that during the actual catalyst testing process, a ceramic panel is also attached to the front of the reactor to further minimize heat losses and to maintain temperature uniformity within the array. This panel, however, has been removed to better illustrate the details of the reactor system. The temperatures of the array microreactors were regulated with electrical heating cartridges and thermocouples inserted into the aluminum blocks and by using PID controllers (Omega, Stamford CT). Gas temperatures in the vicinity of the catalyst sites were separately monitored by small thermocouples inserted from the reactor exit to assess their uniformity across the array, and to determine the time periods for heat up and cool down. These studies revealed the presence of a nearly uniform temperature, within 1^oC, across each array and between the arrays. The entire reactor system was mounted on a stand that was part of a high precision, computer controlled x-y-z movement mechanism. Catalyst testing proceeded in the following manner. First, all of the 66 calcined pellets, together with selected duplicates and blanks, were placed into the wells of the microreactors, and the library was heated to 350^oC under argon carrier gas. Upon reaching the set point temperature, the gas flow was switched to pure hydrogen and the pellets were reduced for 2 hrs. After reduction, the gas flows were changed back to argon and the temperature was lowered to the desired test temperature. Once the test temperature was reached, the gas stream was switched to the feed stream that contained about 10% cyclohexane in argon carrier gas. Under the experimental conditions investigated, the nominal contact time between the feed gas and the pellets was about 0.004 s.

The levels of the reactants, products and the inert carrier gas were determined by withdrawing a small stream (about 1 cc/min) from each microreactor channel using a 50m m diameter capillary sampling probe. The probe was inserted 2 mm into the channels to ensure the acquisition of representative samples. Gas analysis was accomplished by quadrupole mass spectrometry (QMS). The sampling probe and the analysis system are also indicated in Figure 1. Each microreactor effluent was sequentially analyzed by moving the heated reactor block relative to the stationary sampling probe with the aid of the x-y-z positioning system. Gas samples upon entering the QMS were immediately subjected to 70 eV energy electrons and analyzed. In order to accelerate the process of data acquisition and to minimize the volume of data acquired, only selected mass ions, determined earlier in the scoping experiments, were monitored during the testing process. Specific masses monitored during the tests were: the reactant cyclohexane (C₆H₁₂, mass 84 Daltons), cyclohexene (C₆H₁₀, 82), C₆H₈ (80), benzene (C₆H₆, 78), and argon (40). It is particularly interesting to note that although different levels of benzene production were observed from different catalysts, partially dehydrogenated cyclohexane products were not detected in any of the reactor effluents under the conditions investigated.

Figure 1. Microreactor catalyst screening system. A: Array microreactor; B: Capillary sampling probe; C: Mass spectrometer; D: Catalyst pellets; E: Aluminum heating block; F: Insulation.

The data were acquired by rapidly inserting the capillary probe into each microreactor channel and sampling for about 5 seconds. This time period was adequate to acquire 2-5 sets of for data, which was deemed sufficient for the present work. The probe subsequently was withdrawn from the channel and positioned into the next channel. By repeating this approach for each microreactor, we were able to screen the entire 80 channel library in about 10 minutes. Further acceleration of the data acquisition process is clearly possible. However, this was not necessary for the present library size considered here. It should be noted that even the conservative screening rate of 80 microreactors per 10 minute achieved with the present system represents a factor of 8 improvement in speed over the scanning mass spectrometer system of Cong et al.^[9]

In Figure 2 the benzene ion current signal determined by the QMS during the screening process are presented as a function of nominal channel number at 300^oC temperature. The peaks correspond to benzene levels measured in each microreactor effluent, while the sharp drops correspond to the movement of the capillary probe from one site to the next one. As evident from Figure 2, the measurement technique developed allows for easy discrimination of the individual sites. The relative activities of the different catalyst sites are readily discernible, illustrating the potential utility of system as a high-throughput screening tool. It is particularly significant to note the stark differences that exist between catalytic and blank sites. The data presented in Figure 2 were also transformed into a triangular bar chart indicating the catalyst composition, and this is shown in Figure 3. As seen from Figure 3, both Pt and Pd individually exhibited catalytic activities, with Pt being more active than Pd. On the other hand In was not catalytic towards benzene production from cyclohexane, and the ternary mixture of 0.8%Pt, 0.1%Pd and 0.1% In exhibited superior benzene productivity compared to all other metal combinations investigated. These results are in complete agreement with those obtained using the REMPI screening and the same catalyst and operating conditions.^[7]

Figure 2. Benzene signal strength, I, as a function of microreactor channel number, n. Total time to acquire this information was less than 10 min.

As noted earlier, an important concern in the development of practical catalysts is time on stream performance. For example, although catalysts can exhibit high initial activities, this may degrade with time as a consequence of a number of events that can include coking, sintering or poisoning. Alternately, selectivities to desired products can also change with time. The array microreactors coupled with QMS can readily address these concerns. In order to demonstrate this capability, we evaluated the performance of our ternary Pt-Pd-In catalyst library over a 24 hour time period. In Figure 4, the levels of benzene in the reactor effluents are presented as a function of time for the higher activity catalysts. Data for other catalysts were omitted to avoid congestion. The legend on the right side of Figure 4 indicate the catalyst compositions. It should also be noted that the number of time on stream data points shown in Figure 4 represent only a subset of the data acquired by QMS to avoid clutter and to clearly illustrate a number of issues. First, as evident from Figure 4, all the catalysts, without any exception, experienced deactivation over the 24 hour period investigated; deactivation was most significant early on in the experiments. Surface analysis of the used catalysts indicated that the mechanism of deactivation was coke formation. This was supported by the fact that it was possible to regain most of the initial activity after calcination in air. Second, it is significant to note that different catalysts deactivated at different rates, with the activity profiles of some catalysts crossing each other over time. These issues clearly demonstrate the inadequacy of the initial activity/selectivity data as a screening criterion for catalyst development.^[9]

Figure 3. Initial benzene formation activity, I, of the 66 combination Pt-Pd-In catalyst library.

Figure 4. Performance of selected catalysts as a function of time on stream.

The trends presented in Figure 4 also provide useful practical and fundamental insights on catalyst formulation. For example, consider the comparative deactivation behavior of 0.9%Pt/0.1%Pd and 0.9%Pt/0.1%In catalysts (solid lines). As evident from Figure 4, both catalysts exhibited virtually the same initial activity. However, the 0.9%Pt/0.1%Pd catalyst deactivated significantly faster than the 0.9%Pt/0.1%In catalyst, suggesting the positive impact of In in slowing rate of coke formation.

In Figure 5 the bar diagrams for benzene levels in the exhausts of all the microreactor channels are presented at the beginning, and after 2, 4 and 24 hrs of operation, respectively. As can be seen from this figure the ternary mixture of 0.8%Pt, 0.1%Pd and 0.1% In, which exhibited highest initial activity maintained its relative superiority over the entire testing period. However, as seen in Figure 5, the activities of all the catalysts decreased significantly during the testing period.

Figure 5. Benzene formation activity (I) profiles for the 66 combination Pt-Pd-In catalyst library: a) initial activity; b) activity after 2 h of operation; c) activity after 4 h of operation; and d) activity after 24 h of operation.

The array microreactor and QMS system developed also was used for the acquisition of information on reaction kinetics from which proper operating conditions can be determined. In order to illustrate this we studied the effects of reaction temperature on benzene formation. This was accomplished by starting with freshly calcined, i.e. coke free, catalyst library. The array microreactor system was first heated to 150^oC and then screened for activity and selectivity. At 150^oC both benzene formation and catalyst deactivation rates were low. After data acquisition, the temperature of the reactor was increased to 200^oC again under inert gas flow. Following the establishment of the steady reaction temperature, the catalyst library was again screened for activity and selectivity. This procedure was repeated at 250^oC and then 300^oC as well. The total time to screen 80 microreactor channels at 4 different temperatures was about 3 hours.

Heterogeneous reactions in porous catalyst particles proceed through a number of sequential events. These events include the transport of reactants to the external surface of the particle, followed by internal (pore) diffusion together with adsorption and then surface reaction processes.^[13] Products then desorb from the surface and are transported back through the pores into the bulk gas stream. Temperature influences each of these steps differently. Generally surface reaction kinetics and desorption rates dominate product formation rates at lower temperatures due to their higher activation energies. However, increasing temperatures increase intrinsic reaction and desorption rates significantly more than diffusion rates, thus the latter begin to influence the overall rate of conversion. These changes typically manifest themselves by decreases in observed (apparent) activation energies with increasing temperature.^[13] In other words, Arrhenius diagrams, i.e. ln[rate] vs. 1/T plots, exhibit non-linear behavior.

In order to explore above issues with the current catalyst system, the benzene signals measured at different temperatures have been analyzed using the following power-law kinetics:

d[I_C6H6]/dt = k_oe(-E/RT) . f(C)

where I_C6H6 is the intensity of benzene signal measured which is proportional to concentration, t the time, k_o the pre-exponential factor, E the apparent (overall) activation energy and f(C) is the concentration dependent part of the rate expression, which can be in any functional form. Assuming differential conversion of cyclohexane, the above equation can be integrated along the pellet and arranged to yield:

ln(I_C6H6 ) = ln(k_o) - E/RT + ln(f(C)) + ln(t )

where t is the total contact time for the pellet, which is a constant for each microchannel reactor. In Figure 6 the benzene data are presented in accordance with the above relationship, again for the higher activity catalysts. As evident from Figure 6, all the catalysts exhibited non-linear Arrhenius behavior, suggesting changes in rate controlling mechanisms with increasing temperature. In particular, the rate of formation of benzene exhibited higher apparent activation energies at lower temperatures than at higher temperatures for the higher activity catalysts, suggestive of the increasing importance of diffusion limitations with increasing temperature. Clearly, it is also possible that catalyst deactivation could have contributed to this behavior. However, due to the shortness of the test period (15 minutes at each temperature) and the fact that lower temperature experiments were conducted first, coke formation should not be a significant factor even at 300^oC. For the highest activity catalyst (0.8%Pt-0.1%Pd-0.1%In), the apparent activation energy was about 7 kcal/mole in the temperature interval 150-250^oC, but decreased to 5 kcal/mole in the temperature range 250-300^oC (see the upper solid line in Figure 6).

Figure 6. Arrhenius diagrams for selected catalysts.

For the pure Pt catalyst (1.0%Pt), the apparent activation energy was about 6 kcal/mole over the 150-200^oC range (lower solid line in Figure 6). This value is smaller than the 9.6 kcal/mole value reported earlier in packed bed experiments where 40-65 mesh pellets of Pt catalysts were used.^[14] However, these results are consistent with the fact that the larger pellets used in our experiments would be more prone to diffusional limitations thus exhibit smaller activation energies.^[13]

It is also interesting to note that the activation energies of lower activity catalysts increased with increasing temperature over the temperature ranged considered (Figure 6). In addition, all of these catalysts were rich in indium, a fundamentally significant result that could be useful for the development practical catalysts.

In summary, the array microreactor system developed, in conjunction with mass spectrometry allows the acquisition of valuable data on activity, selectivity, time on stream behavior and reaction kinetics in a high throughput fashion. As such the system developed should be of considerable utility for the implementation of combinatorial strategies for the discovery and optimization of practical catalysts.

Combinatorial chemistry, high throughput screening, mass spectrometry, heterogeneous catalysis, array microreactors.

[1] X-D. Xiang, X. Sun, G. Briceno, Y. Lou, K-A. Wang, H. Chang, W.G. Wallace-Freedman, S-W. Chen, P.G. Schultz, Science 1995, 268, 1738.

[2] G. Briceno, H. Chang, X-D. Sun, P.G. Schultz, X-D. Xiang, Science 1995, 270, 273.

[3] X-D. Sun, C. Gao, J. Wang, X-D. Xiang, Appl. Phys. Lett. 1997, 70, 3353.

[4] E. Danielson, J.H. Golden, E.W. McFarland, C.M. Reaves, W.H. Weinberg, X.D. Wu, Nature 1997, 389, 944.

[5] X-D. Sun, K-A. Wang, Y. Yoo, W.G. Wallace-Freedman, C. Gao, X-D. Xiang, P.G. Schultz, Adv. Mater. 1997, 9, 1046.

[6] E. Reddington, A. Sapienza, E. Guraou, R. Viswanathan, S. Sarangapani, E.S. Smotkin, T.E. Mallouk, Science 1998, 280,1735.

[7] S.Senkan, S. Ozturk, Angew. Chem.,1999,111, 867; Angew. Chem. Int. Ed.,1999,38, 791.

[8] J. Wang, Y. Yoo, C. Gao, I. Takeuchi, X-D. Sun, H. Chang, X-D. Xiang, P.G.

Schultz, Science 1998, 279,1712.

[9] Cong, P., Doolen, R., Fan, Q., Giaquinta, D.M., Guan, S., McFarland, E., Poojary, D.M., Turner, H.W., Weinberg, W.H., Angew. Chem.,1999,111, 508; Angew. Chem. Int. Ed., 1999, 38, 483.

[10] F.C. Moates, M. Somani, J. Annamalai, J.T. Richardson, D. Luss, R.C. Willson, Ind. Eng. Chem. Res. 1996, 35, 4801.

[11] S.J. Taylor, J.P. Morken, Science 1998, 280, 267.

[12] S.M. Senkan, Nature 1998, 394, 350.

[13] Satterfield,C.N, Mass Transfer in Heterogeneous, MIT Press, 1970.

[14] K. Ahmed, H.M. Chowdhury, Chem. Eng. J. 1992, 50,165.

[*] Professor Dr. S. Senkan, K. Krantz, S. Ozturk, V. Zengin, I. Onal, Department of Chemical Engineering, University of California, Los Angeles, CA 90095, Fax: 310-267-0177.

[**] We would like to thank LCS Inc. and ATG Inc. for their support and use of their facilities.