A High-Throughput Screening Approach to Identify New Active and Long-Term Stable Catalysts for Total Oxidation of Methane from Gas-Fueled Lean–Burn Engines

: A unique high-throughput approach to identify new catalysts for total oxidation of methane from the exhaust gas of biogas-operated lean-burn engines is presented. The approach consists of three steps: (1) A primary screening using emission-corrected Infrared Thermography (ecIRT). (2) Validation in a conventional plug ﬂow gas phase reactor using a model exhaust gas containing CH 4 , O 2 , CO, CO 2 , NO, NO 2 , N 2 O, SO 2 , H 2 O. (3) Ageing tests using a simpliﬁed exhaust gas (CH 4 , O 2 , CO 2 , SO 2 , H 2 O). To demonstrate the efﬁciency of this approach, one selected dataset with a sol-gel-based catalysts is presented. Compositions are 3 at.% precious metals (Pt, Rh) combined with different amounts of Al, Mn, and Ce in the form of mixed oxides. To ﬁnd new promising materials for the abatement of methane, about two thousand different compositions were synthesized and ranked using ecIRT, and several hundred were characterized using a plug ﬂow reactor and their ageing behaviour was determined.

The radiation of a real body can be described using a black body as a model system and a coefficient describing the "imperfection" of a real body. The factor connecting both is called emissivity ε. It is possible to correct for the different individual emissivities of each sample by calibration. To make the temperature differences under reaction conditions visible, typically, the difference between an image recorded under an inert atmosphere is subtracted from an image recorded under the reaction atmosphere. Combining this differential image with the temperature calibration mentioned previously resulted in an ecIRT image. A custom color scale can be applied to those images to differentiate between areas. The jet colormap as well as a cubehelix colormap have been used, each having different advantages and drawbacks [62,63].
Our catalysts oxidize methane, which is a highly exothermic reaction (∆ R H − • = −802.3 kJ/mol). The heat created will heat up the catalyst and, due to heat transport, also the surrounding slate library plate. We were using an in-house developed software called IRTestRig to perform calculations from the recorded image files like subtraction or emissivity correction as well as ∆T computation [64]. The computation procedure for ∆T uses an annulus around each sample well to get a baseline for the integration, integrates the sample area and calculates an average value for temperature change ∆T for each sample. The resulting value and the obtained image are used to compare different samples and rank them to select samples that might be suitable for further investigations. This method is, therefore, a truly parallel HTS method, useful for primary screening of catalytic activities based on reaction heat. A scheme displaying this principle as well as three possibilities for classification of different samples is displayed in Figure 1. Low activity samples can be easily identified by a low ∆T value and no temperature increase in the surrounding area. Therefore, those samples will neither be highlighted in the obtained ecIRT image nor due to ∆T value. High activity samples have a high ∆T value and a temperature increase of the slate library plate surrounding the sample can also be observed in the ecIRT image. False positive samples have a high ∆T value, but no temperature increase around the sample can be observed in the ecIRT image. False negative samples in which thermal effects and emissivity change are of opposite sign or compensate each other, resulting in low or even negative ∆T values, are possible. Nevertheless, they should still be visible on ecIRT images due to temperature increase around the sample. Therefore, ranking samples based only on their ∆T value might be misleading, resulting in misleading conclusions.

Design of Experiment
There are some promising material combinations known for total oxidation of methane, but most of them lack long-term stability or performance. Therefore, our approach was to perform a fundamental Catalysts 2020, 10, 159 4 of 31 search for new catalysts based on HTS techniques. We decided to define our compositions based on the following observations: 1. Many industrial catalysts are based on a few substrates-e.g., Al 2 O 3 , TiO 2 or SiO 2 . 2. Doping with additional elements often increases performance or stability. 3. PM or PM based species are the catalytic active species in many oxidation reactions.
Based on these criteria, we chose suitable primary elements, additional elements, and PMs as depicted in Figure 2. Co, As, Os, and Pb are excluded due to toxicity of these elements, e.g., Os forms volatile and hazardous OsO 4 . In preliminary work reported previously, the first generation of samples has been described [42]. A total of 156 PM free compositions (one out of seven primary elements combined with one out of 26 additional elements with equal molar amounts; not all samples could be tested) were synthesized. Adding one out of seven PMs to the 156 combinations that have been successfully synthesized before resulted in 1092 PM containing samples [42]. These materials are referred to as generation one, containing of one precious metal (3 at.%), one primary element and one additional element (each 48.5 at.%). It is known from the literature that Pd-containing catalysts are more active than those containing Pt. Pd catalysts suffer from rapid deactivation from poisoning with sulfur-containing species, whereas Pt catalysts are more stable to sulfur [12]. Therefore, we selected one subsystem of all investigated compositions and lay the focus of this paper on finding a Pt containing catalyst that is more active than previously reported ones.
Some promising materials have been identified using our high-throughput approach and we decided to combine these materials in a second generation. Second generation materials contain one or two PM with total 3 at.%, and two or three elements from the list of primary and additional elements in variable amounts, totaling to 97 at.%. Some material combinations from the first generation, like Al and Mn containing samples, had promising results. Therefore, these materials were combined with Ce, that had also shown promising results and is known in the literature to improve catalytic performance [4,6,14,15,18,20,21,23,27,29,31,[33][34][35]46,50,53]. A composition spread of all three elements has been synthesized to have the highest chance to find a local composition based catalytic performance Catalysts 2020, 10, 159 5 of 31 maximum within the selected system. Therefore, samples selected for a more detailed screening had the general composition PM 3 Al a Mn b Ce c O x with a, b, c ∈ 0 at.% to 97 at.%, 9.7 at.% steps. x depends on preparation conditions and thus cannot be specified. For this reason, samples will be denominated without oxygen in our work. To ease sample denomination further, the constant amount of PM of 3 at.% is not used for calculation of molar ratios for nomenclature, resulting in Pt 3 Al 10 Mn 70 Ce 20 for a material with a molar composition of Pt 3 Al 9.7 Mn 67.9 Ce 19.4 O x . The numbers of Al, Mn and Ce in the short notation represent the ratio between the non-PM elements, the PM always was 3 at.%. The molar amount can be calculated by dividing the numbers from the short notation of the three non-PM elements by 100 and multiply by 97 at.%.
Some of our previous results also indicated that samples containing Pt 1.5 Rh 1.5 might have higher catalytic performance compared to samples based on Pt 3 . Following these two guidelines, we synthesized three library plates, one containing samples for Pt 3 Al a Mn b Ce c , one with contains samples for Pt 1.5 Rh 1.5 Al a Mn b Ce c , and Pt 3 Al a Mn b Ce c , and one which contains samples for Pt 1.5 Rh 1.5 Al a Mn b Ce c .

Emission Corrected Infrared Thermography Results
ecIRT has been used for high-throughput screening for over 20 years, brief details about working principles of this method can be found in Section 1.1. There are several effects that have to be considered before using ecIRT: adsorption effects, side reactions, high temperature effects, shadow effects, and emissivity changes [60].
Adsorption effects might lead to steep temperature changes (a high increase in temperature followed by a rapid temperature decrease), but occur only for a short period of time after the adsorbing component is added to the samples under test. Adsorption effects are repeatable and can, therefore, be differentiated from other effects like a steep temperature change due to high activity and a rapid temperature decrease due to fast ageing. To avoid this effect, we use the image taken after 120 min of reaction, therefore, any adsorption effects that might occur should not affect our data.
Parallel or side reactions might obscure the difference between catalysts, because only total change of temperature can be observed. An oxidation of other typical exhaust gas components like CO might give false positive results. Furthermore, side reactions between reaction components and sample material are known in literature, e.g., sulfur decomposition from mixed metal sulfides [60] or sulfatization of Cu with SO 2 [65]. Images can be taken in different atmospheres to test for this behaviour. For primary screening, we used a simplified gas mixture containing only CH 4 , O 2 , H 2 O, and N 2 to reduce the amount of possible side reactions.
At high temperatures, the IR sensors might overexpose, even with shortest available integration time. Neutral density filters can be used to reduce radiation, allowing ecIRT measurements to be carried out even at high temperatures. Another problem is an increased energy loss due to a high temperature difference between reactor and environment, resulting in temperature gradients in the reactor. This has been addressed with improved insulation of the reactor and temperature calibration prior to experiments.
Due to the radial design, gas velocity as well as gas composition can differ at different sample positions, resulting in different results for one composition depending on sample position on the slate library plate. Different gas velocities at different positions can be addressed by placing samples (more than once) at different places with different distances to the central gas outlet on the slate library plate. If there are several high active catalysts placed next to each other, other samples that are placed further downstream in the gas flow will have a different gas composition, since some or even all of the components are already converted. Our approach to limit the influence of both effects is to place each sample twice in two randomly selected positions (Available sample positions are determined using the number of samples that should be distributed on a library plate, to ensure an even distribution. If 103 samples have to be placed in the 206 available wells, every second well is empty. Samples are then placed randomly in the available sample positions.), and decrease the sample density leaving positions in the slate library plate blank.
When the sample changes under reaction conditions, e.g., due to a reaction or change of phase, a change of emissivtiy occurs. If the change in emissivity is not overcompensated by heat of reaction, it might lead to false positive or negative results.
To illustrate the measurement principle, we decided to use a non-completely randomized pattern, 66 samples are arranged in a triangular shape (no sample can be loaded in positon (9,9), so the corresponding sample has been placed at (3,14) instead. and the same 66 compositions are placed evenly distributed in random order around this triangle. The result of a ecIRT measurement are ∆T values, resulting from the temperature difference of the samples between the stated gas mixture (reaction condition) and the temperature in wetted nitrogen atmosphere (background condition). Results of ecIRT measurement are displayed in Figure 3.  Identifying promising samples is done using the ∆T values to rank the samples as well as using the images from Figure 3 to check for false positive or negative samples. ∆T values for both sample spots containing catalyst of the same composition are averaged and the result is displayed in the corresponding graphical composition spread representation. Due to the fact that only two samples have been used to obtain the average value, no standard deviation is given for ∆T values. Even this averaging can be disputed considering the differences in gas composition over the catalyst library as discussed above. We assume that these differences are so small that they can be neglected. The average value for each sample composition is then displayed in an ternary contour heat map using OriginLabs Origin Pro software suite. See Figure 4 for placement of promising samples on the library, Table 1 for a ranking using ∆T values and Figure 5 for average values of ∆T for all samples.         Areas with potential hits have been synthesized with a higher sample density. At the same time, samples with Pt 1.5 Rh 1.5 instead of Pt 3 have been added to test for possible higher activity. Sample compositions are defined in Figure 6. These samples have been placed in slate library plate 2, the results of ecIRT measurement for this library plate are displayed in Figures 7 and 8, and Table 2. Samples with a high amount of Ce are omitted in slate library 2, because these samples are false positive hits. They show a high ∆T value, the CH 4 conversion is below 20 % (see Figure 5).        To test if there are more interesting compositions in the Pt 1.5 Rh 1.5 system, we synthesized the remaining compositions with a resolution of 10 at.%. Results of ecIRT measurement are displayed in Figures 9, 10 and Table 3. For the two samples with the highest ∆T values, no increase in temperature could be detected for the slate library surrounding these samples. For the other samples in Table 3, an increase in temperature of the surrounding slate library can be detected in many cases. Taking the low observed methane conversion for Ce rich samples from Pt 3 Al a Mn b Ce c system and the results of samples from Pt 1.5 Rh 1.5 Al a Mn b Ce c , see Section 4.3, into consideration, no samples from the composition range from slate library 3 have been selected for further testing.   (b) Jet colormap. Figure 10. Average ∆T for samples from slate library plate 3 Pt 1.5 Rh 1.5 Al a Mn b Ce c . Measured using ecIRT. Axis labels denoting the mole fraction χ of given element as percentage, e.g., Al representing χ Al = n Al n Al +n Mn +n Ce , PMs are excluded from this calculation. Sample compositions are displayed by a dot, values between dots are generated using linear interpolation. Table 3. Compositions of most promising materials from Figure 10, position on slate library, and average ∆T.

Pt Rh Al Mn
Ce Positions x, y ∆T [K]  Figure 11. There is no sudden change of ∆T visible on the edge of the merged areas. The ∆T scale on both measurements were overlapping, their lowest and highest value were in the same order of magnitude. Therefore, displaying data from both measurements in the same figure was possible. In general, data points from different measurements are difficult to compare, because each sample set is an unique ensemble of catalysts where activities of individual samples are depending on each other. Therefore, interpretation should be done with great care. ecIRT should be used preferably to compare samples obtained in one measurement. To increase comparability between different libraries, a standard material could be used.   Comparing Figure 11 to Figure 5, the samples with the highest ∆T values have a composition with an amount of Ce of χ Ce = 80 % to 100 %.

Methane Conversion from Gas Phase Plug-Flow Reactor
Samples displayed in Table 1 have been synthesized in larger quantities and have been characterized in our conventional validation setup, results are displayed in Figure 12. Details about the setup used are given in Section 4.3.
CO conversion for all samples presented in this work is 100 % at all examined temperatures. This is in accordance to literature, since reactivity of CO with O 2 is higher than that of CH 4 with O 2 [4]. Version January 8, 2020 submitted to Catalysts  Other examples of artifacts can be found in literature. [59,60] 269 Therefore, results from a primary screening method that does not measure the target value directly  Table 2 and Table 3, are displayed in Figure 13 and Samples with a high amount of Ce are to be considered false positive results in ecIRT, as they have a high value for ∆T but do not show high methane conversion in plug-flow reactor measurements. Indicator for the classification as false positive is the homogeneity of temperature distribution over the sample well in ecIRT-images, whereas active samples reveal a radial temperature profile over the well with a maximum in the middle of the sample position. Another indication of false positive results is the lack of temperature increase in the slate library surrounding the sample well. A possible explanation for this observation of a false positive result is a change in emissivity due to irreversible adsorption or reaction effects of the gas mixture components with the catalyst surface, a possible reaction might be: 8 CeO 2 + CH 4 4 Ce 2 O 3 + CO 2 + 2 H 2 O. CH 4 and O 2 have been removed from the gas mixture for emissivity correction and temperature calibration, which might lead to a oxygen evolution from some oxygen containing species (e.g., 2 CeO 2 Ce 2 O 3 + O 2 ) or the addition of water to oxygen vacancies resulting in hydroxyl species like Ce 3+ -OH [20].
Usually only one component is removed from the gas mixture for emission correction and temperature calibration. Preliminary experiments showed fewer artifacts when the emission correction image is recorded in N 2 after the reaction in comparison to an image recorded in a mixture of O 2 and N 2 prior to the reaction. Using wet O 2 /N 2 mixture as background resulted in many samples showing high (apparent) temperature increases, but no change of temperature in the surrounding slate plate, i.e., frequently false positive results. These apparent temperature changes would disturb the measurement. Other examples of artifacts can be found in literature [59,60]. Therefore, results from a primary screening method that does not measure the target value directly (here: catalytic activity for methane oxidation) but using a secondary criterion (here: temperature increase using a method that relies on constant emissivity under measurement conditions) always have to be validated.
Results of conventional validation for slate libraries 2 and 3, with sample compositions selected from Tables 2 and 3 Methane conversion for samples selected from Table 2         In Figure 15, the displayed error bars for the sample Pt 1.5 Rh 1.5 Al 0 Mn 90 Ce 10 are larger than those for the sample Pt 3 Al 0 Mn 90 Ce 10 , especially at temperatures higher than 400 • C. This larger error bars result from a decrease in methane conversion over time at the given temperature. The first of three GC measurements at a given temperature always revealed highest methane conversion, whereas the one recorded about 30 min later had already decreased in performance. The decrease of performance at constant temperature over 30 min is usually in the range of 6 to 11 percentage points. This behavior can be observed for six out of seven Pt 1.5 Rh 1.5 containing samples, but only for two out of five Pt 3 -containing samples. Gélin et al. reported a drop in conversion for Pt/Al 2 O 3 samples when measured at constant temperature for 1 h [12], a similar phenomenon might be the reason for the observed behavior of some samples.
Due to vastly different conditions for analyzing CH 4 conversion, comparing our results with the literature is challenging. Besides inherent differences in synthesis method and composition, and, therefore, a limited comparability, the most important differences are: GHSV, gas composition (amount of CH 4 , O 2 , presence of H 2 O, CO 2 , other exhaust gas components like SO 2 , NO x ), temperature program (heating or cooling or both, temperature rate, stationary process at discrete temperatures or continuous process), pretreatment (with O 2 , H 2 , or none). Our aim is to test materials under conditions as close to reality as possible. Therefore, we test our catalysts starting at low temperatures going to higher temperatures (starting the engine), measuring at discrete temperatures after 15 min of equilibrating. We also test with a gas composition getting as close to a real-world exhaust gas as possible. Limitations in our setup are: He as balance instead of N 2 , no particulate matter/soot, and catalysts are tested as powder (not as a monolith/honeycomb structured catalyst).
For comparison between different catalytic systems, usually T 50 and in some cases T 10 and T 90 are used in the literature. T x represents the temperature T, at which a CH 4 conversion of x % is achieved. In Table 4 T 50 values from previously published materials as well as measuring conditions are given. The best performing material from Figure 5 is Pt 3 Al 0 Mn 90 Ce 10 with T 50 = 450 • C. Since our measurement conditions are closer to a real exhaust gas composition, it is hard to compare with other systems. Many T 50 values given in Table 4 are lower than that of our best catalyst, but many of them were recorded with simplified gas composition like CH 4 , O 2 , and a balance gas. Space velocity is also different for many cases. Therefore, differences in T 50 can be as high as some 200 K (5 % Pd@ZrO 2 from [66]), where gas composition lacks components that makes CH 4 oxidation more difficult like H 2 O, SO 2 , and CO 2 . Additionally, space velocity in the referenced work is 50 % of the space velocity we have used. Other catalysts have T 50 in the same order of magnitude (e.g., those from [67,68]), where they also used a gas mixture containing only CH 4 , O 2 and N 2 , and the space velocity (GHSV) used is about four times higher than the space velocity used in this work. Other samples show a lower T 50 than our best performing sample, where gas composition is more complex than CH 4 , O 2 and balance (e.g., [7,11,43]), and space velocity (GHSV) is in the same order of magnitude. The most likely reason for different performance is due to different synthesis methods (impregnation, commercial sample vs. sol-gel-approach), probably resulting in vastly different surface areas, porosity and dispersion of active components. Another reason for higher T 50 of our catalyst might be the additional components in our gas mixture (SO 2 , NO x , CO), which have a detrimental influence on performance. CO molecules are blocking active PM sites reversibly as a parallel reaction, whereas SO 2 often leads to irreversible deactivation [12,23,43,44]. Our sample does not reach 100 % CH 4 conversion even at 475 • C, whereas many samples from literature reach 100 % CH 4 conversion at lower temperatures, but those conversions are almost always measured for exclusive methane combustion with oxygen excess and in many cases, without water and other active site blocking species present.

Ageing Behavior Results
Ageing tests have been done for samples selected by activity from library compositions previously tested by ecIRT and are displayed in Figure 17. Aiming towards a long-term stable catalyst, our ageing method was designed to age for at least one week with an option to age for up to several months.
None of the nine selected samples achieved a methane conversion that is stable for 168 h. Methane conversion dropped about 50 % for all samples within the period of investigation. Samples Pt 3 Al 50 Mn 50 Ce 0 and Pt 3 Al 50 Mn 50 from our first generation showed comparable ageing behaviour, proofing again the reproducibility of our method. Pt 3 Al 0 Mn 90 Ce 10 performed 20 percentage points to 10 percentage points better than our first generation Pt 3 Al 50 Mn 50 or Pt 3 Al 50 Mn 50 Ce 0 after 0 h and 168 h, respectively. Comparing samples that have similar composition supports the observation that the addition of Ce can improve the performance. Sample 4 (S4) performed better than Sample 5 (S5), exchanging 9.7 at.% of Al with Ce increased performance to roughly the same level as Sample 2 (S2) had. Samples 5, 8 and 6 (the samples with 0 at.% or 4.85 at.% of Ce) had the lowest methane conversion from all 9 investigated samples.
Sample 3 (S3) shows better performance compared to the other samples and initial CH 4 conversion was even higher than measured before at the same temperature of 450 • C (see Figure 12). This increased performance and a larger error is due to problems with gas supply. We tested the gas flow through each sample prior to starting the experiment; the gas flow for S3 was about one third of the usual gas flow. Lower flow rate resulted in higher residence time and, therefore, a higher conversion. Larger error bars than usual (based on experience) indicate anomalies during the measurement, where some errors occur in addition to normal statistical spread (outlier). Ageing during the measurement as in the conventional validation could be possible in theory, but a single measurement takes about 2 min 30 s. It is unlikely to affect the results, because the ageing are observed in the range of several hours. Therefore, measurements of S3 are used to illustrate that for large sets of data, error bar size can be used to validate the consistency of obtained data. SO 2 often leads to irreversible deactivation [13,24,43,44]. Our sample does not reach 100 % CH 4 334 conversion even at 475 • C, whereas many samples from literature reach 100 % CH 4 conversion at lower 335 temperatures, but those conversions are almost always measured for exclusive methane combustion 336 with oxygen excess and in many cases without water and other active site blocking species present.  Figure 17. Aiming towards a long-term stable catalyst, our ageing 340 method was designed to age for at least one week with an option to age for up to several months. None of the nine selected samples achieved a methane conversion that is stable for 168 h. Sample 3 (S3) shows better performance compared to the other samples and initial CH 4 conversion was 352 even higher than measured before at the same temperature of 450 • C (see Figure 12). This increased 353 performance and a larger error is due to problems with gas supply. We tested the gas flow through Comparing ageing results with results from conventional validation shows initial methane conversions in the same order of magnitude. Differences are in the range of 6 to 23 percentage points. CH 4 conversion measured in conventional validation setup is usually higher than that from ageing setup. The most likely reason for these differences is a WHSV of 72,000 mL/(g h) in ageing versus 36,000 mL/(g h) in conventional validation. One more reason are differences in gas composition, see Section 4.4 for details. Additionally, CH 4 conversion is measured in the ageing setup at 350 • C prior to the first measurement at 450 • C, which takes about 3 h to 5 h. Conversion data for 350 • C are not shown in Figure 17. This might also decrease overall performance prior to the first displayed data at 450 • C.
In Table 5, conditions for ageing experiments from different previous publications are displayed. Comparing our results with data from different sources is even more challenging than before. Among these differences are ageing conditions and ageing time. Different conditions have been reported which are aiming for different situations, e.g., focusing on only one component like H 2 O or H 2 S or using a more complex approach to simulate an exhaust gas [12,43,44]. Ageing times are in the range of a couple of hours (there are publications with ageing times less than 12 h) to a couple of days. Gas compositions differ from using only CH 4 , O 2 , and a balance gas to a more complex approach using components like H 2 S, H 2 O, CO 2 , and NO x . Because of these differences in measurement conditions, a detailed comparison of results is omitted here. Table 5. Ageing conditions (gas composition, last gas used for balance; temperature, space velocity as WHSV or GHSV) and time for CH 4 abatement catalysts from literature. WHSV has been calculated from given data were applicable and possible. As a conclusion from comparing the results from all the different groups working on methane abatement, it would be beneficial for the scientific community to agree on a set of testing conditions to increase the comparability between publications. We encourage to always test with a complex model exhaust gas containing the same amount of CH 4 , O 2 , CO 2 , and H 2 O, this would be closer to a real-world application. If possible, the addition of further gases (like SO 2 and NO x in our work) should be considered, especially if a component is known to poison catalysts. We further suggest to record the methane conversion starting at low temperatures if the research is focused on a catalytic converter, especially if it might be used in cyclic operation. The increasing temperature while measuring conversion mimics the start up behaviour and tends to give more realistic results. If the conversion is measured starting at high temperatures, hysteresis effects might lead systematically to high methane conversion. We also encourage to include ageing tests into standard procedure, particularly when there might be a real-world application for the catalyst.

Conditions
We are aware that testing with a complex gas mixtures is a technically demanding task. It is also clear that different scenarios for real-world application (e.g., methane abatement from mine gas versus abatement from a gas-operated light-duty vehicle) demand for different test settings. A common basis for different applications would be testing with only methane in air, but this would be misleading. Resulting methane conversions would be higher than in any complex gas mixture.

Discussion
We developed a unique set of tools for high-throughput screening of catalysts. We are able to measure catalyst performance for methane oxidation under close to reality conditions except for particulate matter/soot.
We have selected one data set and presented the results of our high-throughput workflow for this set here. Some hundred samples have been tested with ecIRT, 18 samples have been tested in our conventional plug-flow reactor setup and nine samples have been tested for their ageing behavior. Our best performing catalysts are Pt 3 Al 0 Mn 95 Ce 5 , Pt 3 Al 0 Mn 90 Ce 10 and Pt 1.5 Rh 1.5 Al 0 Mn 90 Ce 10 , but none of them reached 100 % methane conversion in our plug-flow reactor at 475 • C or had a stable methane conversion in ageing tests for 168 h. Comparing the results with the literature is difficult, due to vastly different gas composition as well as different experimental conditions. We are, to the best of our knowledge, the only group that has tested ageing behavior of up to ten different methane abatement catalysts in parallel for 168 h. Our setup was designed to allow for even longer ageing measurements, which might be needed to test long-term stable catalysts and rate their performance.
The results presented here represent only a small section of the compositional parameter space investigated (see Figure 2). Other sol-gel-based samples that are under investigation have outstanding performance and stability.
We are aware that the catalytic systems we have presented are not applicable in an real-world application yet, due to expensive sol-gel synthesis as well as low performance and stability.
Our set of HTS tools can also be applied to other synthesis methods for catalysts, like incipient wetness or other industrial applicable methods. Additionally, the setup is flexible and could also be used to screen catalysts for other applications, for example, oxidation of other hydrocarbons like propane, propene, or to test for other gas phase reactions like selective catalytic reduction of NOx.

Synthesis
All chemicals were used as supplied.  Precursor solutions were prepared following standard laboratory practice. An equal volume mixture of propan-2-ol (IP) and propionic acid (PA) was used as solvent for Al, Mn, and Ce precursors, whereas the Pt precursor was dissolved in methanol (MeOH) and the Rh precursor in a equal volume mixture of MeOH and pentane-2,4-dione (Hacac). The calculated volume of 4-Hydroxy-4-methylpentan-2-one (CA), corresponding to three times the molar amount of precursor, was added to selected solutions. Hacac and CA were used as complexing agents to prevent precipitation of components from the solution. The mixture was stirred until a clear (and for Mn, Pt, and Rh a colored) solution was obtained. Solutions for the selected elements were prepared according to Table 6. The molar amount of all samples prepared for ecIRT was 350 µmol, resulting in enough material to fill two sample wells in a slate library. After sample definition, in-house developed software Plattenbau [64,71] was used to prepare a pipetting list for one of our pipetting robots. For liquid dosing of precursor solutions a pipetting robot of type Lissy made by Zinsser Analytic, Germany, was used. For liquid handling 2 needles and one 500 µL syringe pump for each needle are available, the needles have a liquid level detection. Several racks machined from different steel materials with different well sizes for accommodation of reservoirs of various volumes were employed for storage of the precursor solutions during the pipetting steps. Precursor solutions were placed in one rack in 8 mL glass tubes (Macherey-Nagel, 702096), glass tubes for samples (1.5 mL GC vials, Macherey-Nagel, 702282) were placed in a second rack. After pipetting, the sample rack was placed on a vortexer (IKA KS 260 basic) and the solutions were mixed for 2 h at 300/min. Gelation was achieved in an oven for about two weeks at 40 • C. Calcining is done following the process displayed in Scheme 1.

Samples for Characterization in Conventional Plug-Flow Tube Reactor and Ageing Tests
The molar amount of all samples prepared was 5000 µmol. For each sample, the necessary volumes of each precursor solution was transferred in a 100 mL beaker using Eppendorf pipettes (Research plus, 20 µL to 200 µL, 100 µL to 1000 µL, 0.5 mL to 5 mL, 1 mL to 10 mL). The beaker was covered with Parafilm and stirred for 2 h. The solution was then transferred and split into three glass vials (volume 30 mL, VWR, 548-0624), to reduce the time needed for gelation. Gelation and calcining were done following the same procedures described in Section 4.1.3.

Library Preparation
The libraries for emission-corrected Infrared Thermography (ecIRT) were machined from natural slate. This material behaves nearly as an ideal black body emitter in ecIRT and shows no reflections of radiation. Other materials like steel have to be blackened before usage. The plates had a diameter of 100 mm and were 6 mm thick. A total of 206 wells were drilled into the surface as sample positions. Wells had a diameter of 3 mm and were 2 mm deep. They were arranged in a hexagonal pattern for best use of space and were named after their x and y position as can be seen in Figure 18. Slate libraries had a small mark over (9,2) to identify orientation. Position (9,9) was drilled through the library and operated as gas outlet, diameter 4 mm. Four additional holes (2,4), (2,13), (16,4) and (16,13) were drilled though the library to ease handling while assembly and disassembly of ecIRT setup, their diameter was 4 mm. of space and were named after their x and y position as can be seen in Figure 18. Slate libraries had a 483 small mark over (9,2) to identify orientation. The position (9,9) was drilled through the library and 484 operated as gas outlet, diameter 4 mm. 4 additional holes (2,4), (2,13), (16,4) and (16,13) were drilled 485 though the library to ease handling while assembly and disassembly of ecIRT setup, their diameter 486 was 4 mm. Figure 18. Scheme of wells in a slate library used in ecIRT. Gray highlighted wells are not present due to circular shape or drilled through the library as gas outlet (9,9) or to ease handling (2,4), (2,13), (16,4) and (16,13 Figure 18. Scheme of wells in a slate library used in ecIRT. Gray highlighted wells are not present due to circular shape or drilled through the library as gas outlet (9,9) or to ease handling (2,4), (2,13), (16,4) and (16,13).
Prior to filling a slate library plate with samples, it was heated following the procedure displayed in Scheme 2. Samples were filled in the wells manually. Samples were placed in random order in a pattern that allowed as much distance between individual samples as possible. To increase the accuracy of this method, samples were placed on the same library plate twice. Samples were ground lightly prior to filling using a glass rod (diameter 3 mm). Samples were filled in selected wells with a small spatula to at least 80 % and were compressed with a glass rod (diameter 3 mm) to achieve a flat surface. Spilled material was removed carefully using a small vacuum cleaner setup, consisting of a vacuum pump, a gas wash bottle, and a tube connected to an Eppendorf tip. After filling the library plate, it was inserted into the measurement setup described below.

Experimental Setup for ecIRT
The ecIRT setup consisted of three mass flow controllers (CH 4 /N 2 , N 2 , O 2 ), a heated gas washing bottle, a heated reactor setup with a sapphire glass plate covering the slate library, and a Thermosensorik IR camera with a 640 x 486 PtSi FPA detector. The camera was equipped with a set of neutral density filters made from germanium (Thermosensorik). The filter with a transmission of 1 % was used for results shown in Figures 3, 7 and 9. 100, 10 and 0.1 % transmission filters were also available. With manually operated valves, gasses from the mass flow controllers could bypass the heated gas washing bottle to turn humidification on or off. Humidification was turned on after heating was completed and was turned off prior to cool down, so H 2 O was present at all times during the experiment. A simplified scheme of the setup can be seen in Figure 19. The gas composition for ecIRT experiments was 2 vol.% CH 4 , 8 vol.% O 2 , and 12 vol.% H 2 O. N 2 was used for balance, the total gas flow was 50 mL/min. Prior to conducting experiments, our heating setup was calibrated using a metal cover with six holes, a slate library plate filled with sand, and thermocouples measuring the temperature at the SiO 2 surface at six different positions. The metal cover was placed in three different positions to measure the temperature across the slate library. The average of these temperatures was used to calculate the necessary setpoints for heating to achieve desired temperatures.
To reduce thermal stress for the reactor, but especially for the sapphire window, heating and cooling was done with a slow heating rate with waiting steps in between to achieve thermal equilibrium. Heating procedure is shown in Scheme 3, measurement procedure is shown in Scheme 4, and cooling procedure is shown in Scheme 5. Temperatures were controlled to be in an interval of 1 K around the desired value for 10 min before proceeding with the next step (either set a new or stay at the current temperature) in each of those schemes. For temperature calibration, each library plate was heated under inert conditions (N 2 ) to 6 different temperatures (396, 398, 400, 402, 404 and 406 • C) near the measuring temperature (400 • C) and IR thermography images were recorded at each temperature.
To make the temperature differences under reaction conditions visible (emission correction), typically the difference between an image recorded under N 2 atmosphere is subtracted from an image recorded under reaction atmosphere. We used the images recorded after purging with N 2 for 10 min after the reaction and recorded after 120 min under reaction atmosphere.

Catalytic Testing by Conventional Validation in a Gas Phase Plug-Flow Reactor Setup
100 mg of sample sieved in the range from 100 µm to 200 µm were mixed with 100 mg SiO 2 sieved in the range from 200 µm to 280 µm. The powder was placed inside a silica glass tube (inner diameter 6 mm) on silica glass wool. All samples were placed at the same height within the reactor tube, ensuring a placement in the middle of the temperature constant zone of the reactor. A thermocouple placed within the sample was used to record the sample temperature and control the reactor temperature. A sample made from 100 mg SiO 2 sieved in the range from 100 µm to 200 µm mixed with 100 mg SiO 2 sieved in the range from 200 µm to 280 µm was placed at the same height on silica glass wool constituted the bypass sample. Figure 20 shows a scheme of the setup. We automatized the measurement using an in-house developed LabView program. After a heating period to the first temperature (350 • C) with only He flow, all gases were turned on, resulting in a gas flow of 60.2 mL/min with a composition of: 12.5 vol.% H 2 O, 12.5 vol.% CO 2 , 9.0 vol.% O 2 , 0.25 vol.% CH 4 , 0.055 vol.% CO, 0.012 vol.% NO, 0.004 vol.% NO 2 , 0.001 vol.% SO 2 , balance He. This results in a GHSV of about 26,000/h assuming a sample height of 5 mm, or a WHSV of about 36,000 mL/(h g). Methane conversion was measured at 350, 375, 400, 425, 435, 450 and 475 • C, at each temperature three GC-TCD measurements were performed, starting after 15 min of equilibration at the set temperature. A gaschromatograph from Hewlett Packard, HP GC 6890, equipped with a TCD, and a micropacked column (ShinCarbon ST 19808, length: 2 m, inner diameter: 1 mm) from Restek Corp was used. At the beginning and the end of each measurement cycle three bypass measurements were performed. The average methane signal from these six measurements I CH 4 , bypass was used to calculate the methane conversion X CH 4 using the sample signal I CH 4 ,sample according to Equation (1). I CH 4 ,sample is the averaged methane signal from three measurements at a given temperature. A similar formula was be used to calculate CO conversion.
X CH 4 = I CH 4 , bypass − I CH 4 , sample I CH 4 , bypass Sample standard deviation s of conversion X CH 4 at each temperature was calculated using Equation (2) with n = 3 This standard deviation is displayed as error bars in the corresponding figures.

Parallel Catalytic Ageing Test in Gas-Phase Plug-Flow Reactors
Ageing tests were carried out using a 10-fold parallel tubular plug-flow reactor setup. A total of 20.7 mg of sample sieved in the range from 100 µm to 200 µm were mixed with 20.7 mg SiO 2 sieved in the range from 200 µm to 280 µm and placed in a silica glass tube on silica glass wool. Each tubular glass reactor had a narrowing in the middle to hold both wool plug and catalyst as well as guarantee a reproducible sample position within the reactor. Flow restrictors made from PTFE tubes (inner diameter 0.2 mm, length of about 1.5 m and pressure drop of about 2 bar [72]) were carefully adjusted in length to ensure an equal gas flow through all reactors. They revealed a flow resistance significantly exceeding those of the catalyst beds and thus dominating the overall flow resistance of the reactor and were placed in front of each reactor. One reactor contained SiO 2 sieved in the range of 100 µm to 200 µm as a blank sample. Using a 10/2 multiport valve, one tubular reactor could be selected and the corresponding gas composition was analyzed by GC-TCD, whereas the gas from the other nine tubes was combined in a common outlet. Conversion and standard deviation displayed as error bars were calculated according to Equations (1) and (2). A scheme of the setup is shown in Figure 21. Gas flow for each sample was 25 mL/min, resulting in WHSV ≈ 72,000 mL/(g h). Gas composition was 0.5 vol.% CH 4 , 9.0 vol.% O 2 , 12.5 vol.% CO 2 , 12.5 vol.% H 2 O, 0.001 vol.% SO 2 , balanced with N 2 . Gas composition and flow rate/GHSV differ from the conventional validation to allow for faster ageing. We automatized the measurement using an in-house developed LabView program.
After heating the reactor to 350 • C in dry N 2 , all mass flow controllers except SO 2 were set to their according value, and the valves were set to flow the gas through the heated washing bottle. For each sample, three GC-TCD measurements were performed at 350 • C. Afterwards, the temperature was increased to 450 • C and after 0, 24, 48, 72, 96, 120, 144 and 168 h for each sample, three GC-TCD measurements were performed. SO 2 was turned on after the first measurement at 450 • C. A µ-GC from Varian, CP-4900, equipped with a module with 10 m PPQ column, a heated injector, and backflush (Partnumber 494001430, Agilent) was used. To protect the GC column from increased wear and tear, SO 2 was substituted automatically with N 2 30 min prior to measurements at 450 • C and turned back on afterwards.

Conclusions
In this contribution, we presented our high-throughput approach to identify new catalysts for the total oxidation of methane. We chose one subset of our second generation of catalysts, Pt 3 Al a Mn b Ce c and Pt 1.5 Rh 1.5 Al a Mn b Ce c , and showed the efficiency with which we can search a large parameter space for optimal material compositions. Synthesis was carried out using a sol-gel method for ecIRT in an automated way to be able to synthesize a large number of samples. Our best sample, Pt 3 Al 0 Mn 90 Ce 10 , outperformed the corresponding sample Pt 3 Al 50 Mn 50 from first generation by about 25 percentage points in conventional validation. Maximum methane conversion was (66.7 ± 2.2) % at 475 • C, in comparison to (38.7 ± 1.1) % for Pt 3 Al 50 Mn 50 Ce 0 , which performed similar to Pt 3 Al 50 Mn 50 in conventional validation as well as ageing.
The sample Pt 3 Al 0 Mn 90 Ce 10 showed a methane conversion of (39.4 ± 1.0) % and (19.1 ± 0.4) % after 0 h and 168 h of ageing, respectively. This is about 20 percentage points more methane conversion than the sample Pt 3 Al 50 Mn 50 Ce 0 , which achieved (18.8 ± 0.3) % and (9.0 ± 0.4) after 0 h and 168 h of ageing, respectively. Indications that the addition of Ce increases catalytic performance were found as well, being in agreement with the literature.
Samples with Pt 1.5 Rh 1.5 Al a Mn b Ce c performed better than the tested Pt 3 Al a Mn b Ce c samples, but methane conversion dropped at a given temperature by several percentage points in less than 1 h. Other samples that are still under investigation have outstanding performance and long-term stability.
We highlighted the advantages of our testing methods and challenges that arise when trying to compare our results with other publications.
Author Contributions: A.G. designed and build the experimental setups used, T.L. conducted experiments, analyzed data, and wrote this paper, K.S. supervised the work and reviewed the article, T.S. supervised the work, C.B., R.K. and S.C. provided samples, materials and funding. All authors have read and agree to the published version of the manuscript.
Funding: The publication of this article was funded by Chemnitz University of Technology, which is gratefully acknowledged.