Screening NOx Storage Performance—Demonstrating a High Throughput Approach for Evaluating Emission Control Catalysts under Transient Conditions

At hte the high throughput (HT) approach is applied in the field of environmental catalysis on a routine basis. Research programs for automotive applications require validated screening protocols for conditions relevant to engine exhaust as well as experimental measures to ensure quality control using statistical design of experiment. To illustrate the HT approach for a test protocol with dynamic feed switches in a 48-fold reactor, 15 model catalysts for lean NOx traps (LNT) were prepared and screened fresh and after 800 ◦C hydrothermal aging. In the fresh state, highest NOx efficiency was found at 350–450 ◦C. A ranking of BaO > SrO > CaO was found as the most active NOx storage components when used as dopants on alumina. 800 ◦C aging results in a severe performance loss. Using XRD and BET analysis, Pt sintering is identified as most likely cause. These findings agree well with results from conventional tests reported in the literature.


Introduction
To meet current and future emission targets, rather complex after-treatment systems consisting of several catalysts and filters are applied [1]. Each component has specific functionalities and depending on the field of application (e.g., light-duty vs. heavy duty or off-road equipment) it must be able to survive under application-specific aging conditions. When combining these varying technical requirements with overall cost constraints, there is an obvious need for advanced test technologies to support rapid improvement and optimization of existing commercial emission control catalysts. Increased screening capacity is also required for long-term R&D projects searching for future technologies. Using high-throughput (HT) approaches allows preparation as well as evaluation of large sample libraries. A typical scenario is the optimization of washcoat formulations by systematic variation of different parameters (e.g., catalyst support, PGM type, PGM loading, promoter type and quantity, preparation route, calcination temperature, binder, slurry pH, aging condition, etc.) based on statistical design of experiment methods [2,3]. During the last decades high-throughput methods have become an established tool for evaluation of heterogeneous catalysts for different chemical processes [4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23]. hte's approach to HT testing is based on a patent filed on March 3, 1998 which relates to parallel reactor systems for testing the activity measurement of solid catalysts simultaneously exposed to gaseous feed streams [24,25].
In addition to the reactor system, which includes a flexible process automation software at the core of the HT screening unit, several other challenges need to be addressed specifically in the field of automotive applications: (1) Time resolved on-line analytics of all reactants and products, when using small catalyst amounts, i.e., limited gas flow and the increased impact of wall effects. (2) Data acquisition, especially for test protocols involving fast feed switches in an automated mode of operation. (3) Software solutions for robust data processing and a data reduction infra-structure. (4) Besides the testing infrastructure, for a full assessment, new materials have to be evaluated after implementation in relevant washcoat formulations which are close to production and therefore usually proprietary. This requires development of a small-scale slurry processing workflow as well as aging procedures (high temperature aging, catalyst poisoning e.g., with sulfur).
Since 2000 hte has developed laboratory workflows to apply HT technology in the field of catalytic automotive exhaust after-treatment. The power of the rapid testing paradigm has been successfully demonstrated in several R&D programs exploring large formulation matrices. Results of some examples have been described in previous publications [26,27].
This work describes hte's HT screening approach with focus on test protocols involving transient conditions implemented by dynamic feed switches. Lean NO x Trap (LNT) catalysts are an important emission control technology, especially for light passenger cars with small displacement Diesel engines having comparatively low exhaust temperatures. Since there is no known catalyst for direct NO x decomposition (into N 2 and O 2 ) under lean conditions, LNTs must use a different strategy. NO x is stored during an extended phase of lean operation (several minutes). Before the storage capacity is exhausted, a short rich pulse (3-10 s) is applied [1]. In this period, the amount of available reductants (CO, H 2 , unburned HC) exceeds the residual oxygen and NO x is reduced to N 2 -clearing out the NO x storage capacity for another lean cycle. LNT catalysts consist of a NO x storage component as well as precious metal components (responsible NO 2 formation in the lean phase and for the rapid NO x reduction during rich pulses). Both components/functionalities are closely linked since storage and release kinetics of one component need to match the activity of the other. Thus, screening NO x storage and PGM functionality independently have limited prospect for success. Consequently, a test protocol with rapid feed switches has to be implemented in the HT screening equipment for rapid differentiation of NO x storage material/PGM combinations under realistic conditions. In addition to a general description of the experimental setup used at hte for routine screening, some results from LNT testing are provided to exemplify the process automation, data acquisition and data reduction workflow.

Description of the HT Platform
HT evaluation workflows for emission control catalyst screening at hte include sample preparation, aging (hydrothermal and sulfation), characterization, HT lab testing, data management and processing (cf. Figure 1). Main fields of application include: • Fast primary screening of new materials • Optimization of washcoat composition in large parameter spaces • Accelerated catalyst evaluation for multiple applications by automated variation of test conditions (GHSV, T, feed composition, etc.).
Test units with 48 parallel reactors operated at hte have been described in detail in previous publications [33,34]. Generally, the units consist of a reactor block operated at isothermal conditions. The feed is evenly distributed over all positions to allow the catalysts to equilibrate. For catalytic measurement one position at a time is selected and put under active flow control. The exhaust from the selected position is switched to a dedicated line for catalytic measurement. In typical stationary test protocols, the 48 positions are scanned sequentially, allowing~3-5 min equilibration time + 30 s sampling time for each catalyst, for tests with feed switches the measurement times are usually longer because several cycles are required to allow the catalyst to stabilize. For test protocols like the LNT catalysts screening described here, measurement times > 10 min for each reactor position are needed. This makes it even more important, that all operations are fully automated to ensure 24/7 utilization of the test equipment.  The key reactor features comprise: • 48 catalytic reaction positions and 1 by-pass position for measurement of catalyst inlet gas composition (cf., Figure 2)  The key reactor features comprise: • 48 catalytic reaction positions and 1 by-pass position for measurement of catalyst inlet gas composition (cf., Figure 2  Monitoring specific components by MS (mass spectrometer) (m/z 1-512).

•
Rapid switching of feed gas composition with cycle frequencies up to 0.5 Hz (e.g., lean/rich cycles).

•
Flexible process automation ("hteControl4" software) to run complex test protocols in unattended and safe 24/7 operation.

•
Change between several operation modes without hardware reconfiguration (e.g., DOC, TWC, SCR, LNT) • Integration into an automated data management system ("myhte" software) for automated reduction of primary data, allowing easy data export to more sophisticated data analysis solutions. The key reactor features comprise:

Screening Protocols
Testing protocols and on-line analytics are as close as possible to the conventional lab testing of automotive catalysts. Feed composition, space velocities and temperatures are adjusted to mimic the conditions in the exhaust after-treatment system for selected operating points within regulated driving cycles (e.g., different for light-duty and heavy-duty applications). Due to the high thermal mass of the 49-fold reactor block, in this setup only isothermal operation is possible. Instead of dynamic temperature ramps only stationary experiments with discrete temperature setpoints can be performed in the parallel testing units. At each setpoint, all positions are exposed to the same feed all the time, the measured channel is selected by down-stream multiport selection valves [34]. While the catalyst in a selected position is evaluated at a defined space velocity under active flow control, the remaining channels are exposed to a lower flow rate (~1/50-1/25 of the rate selected for the measured sample), at the same gas composition. This methodology has the advantage to keep all catalysts close to their steady state activity for each operating condition, thus allowing for a short equilibration time after a new position is selected and space velocity is increased to the target value. In general, an experiment in the parallel reactor system has the following test sequence, automated in the control software: (1) Set first experimental condition (temperature, feed gas composition) (2) Wait until the whole reactor is equilibrated (3) Switch to position 1. (4) Equilibrate in stationary feed or run dynamic feed switching program (5) Repeat steps 3 and 4 for all 48 reactor positions. Run duplicate tests on selected positions to obtain statistical data. (6) Set next experimental condition (e.g., higher temperature) (7) Continue with steps 2-6 until all conditions are evaluated for all 48 reactors. (8) This sequence has the advantage over many conventional experiments in that detector drifts in the analytical equipment can be decoupled from sample comparison by repeated measurement of inert and control samples within each plate.
The two groups of test protocols: (i) steady-state tests and (ii) dynamic tests with feed switches result in different levels of complexity in process control as well as in data management and data reduction. In all cases, continuous validation and adaption based on feedback from scale-up experiments on cores and full-size parts in associated laboratories is needed.

Dynamic Tests with Feed Switches
Such protocols (see Figure 3) are required for the evaluation of TWC (three-way catalyst) functionality for exhaust after-treatment in gasoline cars. Also, catalyst technologies for NO x abatement have aspects that require dynamic testing. Two examples are lean NO x traps, which work by repeated reductive regeneration, or measurement of the NH 3 storage capacity to characterize an important property or SCR catalysts: • TWC (Three-way catalysis): For optimal catalytic performance, Gasoline engines are operated near an air:fuel ratio of 1. This is ensured by active engine management using λ sensors. However, in dynamic driving conditions deviations from the optimal value difficult to avoid and the catalyst should tolerate excursions from this optimal point. Therefore, tests with λ perturbations are crucial. Several protocols are used to evaluate the different functionalities of fresh and aged catalysts. Results of this test have been previously reported [33,34].
Light-off tests (i.e., multiple temperature set points) with average λ = 1 and high frequency (up to 0.5 Hz) λ perturbations.   For sample preparation, all conventional methods (e.g., impregnation, precipitation, ion-exchange etc.) are explored in small scale (3-5 g samples) including slurry processing methods such as pH adjustment and milling. Usually a DoE approach is used for the design of sample matrixes (cf. Figure 4) with relevant variables for PGM powders (support material, PGM type, PGM loading, promoters, preparation route, shaping procedure, calcination temperature, aging conditions). The type of DoE depends on the objective. For sample preparation, all conventional methods (e.g., impregnation, precipitation, ionexchange etc.) are explored in small scale (3-5 g samples) including slurry processing methods such as pH adjustment and milling. Usually a DoE approach is used for the design of sample matrixes (cf. Figure 4) with relevant variables for PGM powders (support material, PGM type, PGM loading, promoters, preparation route, shaping procedure, calcination temperature, aging conditions). The type of DoE depends on the objective. In studies targeting optimization of washcoat compositions, the variables for slurry processing include slurry additives, binder, milling conditions and pH. To perform statistically robust experiments standard reference samples should be included in every run and preparation and testing of duplicate samples should be used to control the statistical error of the whole procedure (cf. Table  1 for the experimental design of the current study, a detailed description of the sample preparation is given in the Materials and Methods section).   In studies targeting optimization of washcoat compositions, the variables for slurry processing include slurry additives, binder, milling conditions and pH. To perform statistically robust experiments standard reference samples should be included in every run and preparation and testing of duplicate samples should be used to control the statistical error of the whole procedure (cf. Table 1 for the experimental design of the current study, a detailed description of the sample preparation is given in the Materials and Methods section).
To get an acceptable back-pressure in tests with comparatively high space velocity, powders need to be shaped for testing. A particle size fraction of 250-500 µm is a good compromise between back-pressure and simulating diffusion lengths found in coated catalysts. Typically, shaping is performed by formulating the active components into a slurry, milling to a D 50 < 15 µm, drying under agitation and crushing/sieving after calcination. To simulate realistic washcoat loadings, most catalysts are tested as reactor loads with an active mass of 100-300 mg diluted to 1 mL bed volume using corundum (α-Al 2 O 3 ) of same particle size fraction. The exact quantity of catalyst is selected to represent washcoat amount found in 1 mL of coated monolith catalyst. Testing coated samples in a high-throughput reactor involves careful crushing of the monolith catalysts and using a sieve fraction between 500 and 1000 µm and loading a mass that corresponds to 1mL of coated catalyst. Space velocities can then be calculated with reference to 1 mL of catalyst volume facilitating direct comparison with monolith core and full-size tests.  Typically, catalysts are tested both fresh and after hydrothermal oven aging. Usually aged performance is more critical as it allows to predict catalyst performance over the full lifetime. To simulate realistic exhaust conditions during aging, an atmosphere containing 5-15% water is used. Aging temperatures and durations are set to simulate typical conditions and depend on the application. For Diesel exhaust, catalysts are kept for 5-24 h at temperature between 600 and 850 • C in water/air mixture, the higher temperature range is typical for systems with active particulate filter regeneration. In the case of gasoline exhaust for the TWC application, aging temperatures between 850 and 1150 • C are applied with duration of 5-24 h. To simulate the impact of changing oxidizing and reducing environment, "rich/lean" aging is performed with two gas feeds containing 10% water (4% H 2 in N 2 and air) that are switched every 10 min during aging. For quality assurance, furnaces are equipped with time resolved monitoring of temperature and steam dosing. A certain degree of automation ensures that samples are cooled to 300 • C in the presence of water and further to room temperature in air.
A critical aspect of HT testing is the requirement, that all 48 channels are equivalent with respect to temperature and flow rate. This assumption needs to be verified for every new test protocol. In the present study, three identical loads of each LNT catalyst were filled. Figure 5 shows the results of such an experiment. It illustrates the time dependencies of NO x at 450 • C for all fresh samples. On the one hand, the curves for three loads are rather close (i.e., good load-load reproducibility) on the other hand it reflects the typical behaviour for each catalyst composition. Largest differences between loads were found at low temperatures, where also a strong time on stream effect is observed by large cycle-cycle differences (this effect is shown in Figure 6b, at 250 • C for the fresh catalyst). This might be explained by the absence of the Rh component in these model catalysts which is typically included in real LNT formulations [36], which is essential for the effective reduction of the stored NO x especially at low temperatures. Consequently, the storage capacity is not fully cleared out and the effective capacity shows a slow cycle-cycle deterioration until a steady state is reached. Otherwise, load-load variations give no evidence for systematic errors like dosing stability or temperature distribution over the reactor block. Aging at 800°C has a significant negative impact on NOx storage; both aged samples no longer show any activity at 200°C and the performance of Pt/20%BaO-Al2O3 is not significantly better than that of 1%Pt/Al2O3. The detrimental effect of aging on model Pt/BaO/Al2O3 catalysts was also described elsewhere [37].
The essential Pt functionality is oxidation of NO to NO2 which according to the simplified LNT mechanism [38] is then stored on the catalyst. The time dependency of the NO2 concentration in the lean phase reflects the catalysts ability to form and store NO2. Once the storage capacity is fully saturated the NO2 concentration becomes constant over time as for example on fresh Pt/Ba-alumina at 200°C and on both aged samples shown in Figure 6 at T = 250, 350 and 450 °C. In these cases, saturation is reached within less than two minutes of the lean operation. For the simple model system, there are also pronounced NOx spikes during the rich phase due to incomplete reduction of the stored NOx to N2. This is most pronounced at lower temperatures of 200-250 °C. These results are in line with [36] which showed for lean NOx trap technology that temperatures of at least 300 °C are favourable for complete regeneration as well as NOx reduction.
In the present study, the lean phase of the last cycle is used as performance indicator for catalyst comparison (cf. Materials and Methods section and corresponding figures). The average NOx efficiency within 2 min of this phase is automatically calculated for each measured sample. Results at different temperatures for fresh and aged samples are plotted as function of the Ba-content in Figure 7 (results for the three separate loads are shown).
A clear difference in the performance of different samples upon variation of the NSC composition is observed in the fresh state. After aging the NOx efficiencies are rather low for all catalysts but the general trends are preserved. In all cases, replacing BaO by Sr, Ca, Mg and Zn oxides results in lower NOx efficiencies under the test conditions applied. As long as only 1/3 of Ba is replaced by another alkaline earth metal, the detrimental effect is small, as soon as larger amounts are substituted the performance loss becomes substantial. The performance ranking of samples in which the NSC contains less than 1/3 BaO indicates that Sr > Ca are most active replacements for Ba while Zn and Mg are ineffective. Aging at 800 • C has a significant negative impact on NO x storage; both aged samples no longer show any activity at 200 • C and the performance of Pt/20%BaO-Al 2 O 3 is not significantly better than that of 1%Pt/Al 2 O 3 . The detrimental effect of aging on model Pt/BaO/Al 2 O 3 catalysts was also described elsewhere [37].
The essential Pt functionality is oxidation of NO to NO 2 which according to the simplified LNT mechanism [38] is then stored on the catalyst. The time dependency of the NO 2 concentration in the lean phase reflects the catalysts ability to form and store NO 2 . Once the storage capacity is fully saturated the NO 2 concentration becomes constant over time as for example on fresh Pt/Ba-alumina at 200 • C and on both aged samples shown in Figure 6 at T = 250, 350 and 450 • C. In these cases, saturation is reached within less than two minutes of the lean operation. For the simple model system, there are also pronounced NO x spikes during the rich phase due to incomplete reduction of the stored NO x to N 2 . This is most pronounced at lower temperatures of 200-250 • C. These results are in line with [36] which showed for lean NO x trap technology that temperatures of at least 300 • C are favourable for complete regeneration as well as NO x reduction.
In the present study, the lean phase of the last cycle is used as performance indicator for catalyst comparison (cf. Materials and Methods section and corresponding figures). The average NO x efficiency within 2 min of this phase is automatically calculated for each measured sample. Results at different temperatures for fresh and aged samples are plotted as function of the Ba-content in Figure 7 (results for the three separate loads are shown).
A clear difference in the performance of different samples upon variation of the NSC composition is observed in the fresh state. After aging the NO x efficiencies are rather low for all catalysts but the general trends are preserved. In all cases, replacing BaO by Sr, Ca, Mg and Zn oxides results in lower NO x efficiencies under the test conditions applied. As long as only 1/3 of Ba is replaced by another alkaline earth metal, the detrimental effect is small, as soon as larger amounts are substituted the performance loss becomes substantial. The performance ranking of samples in which the NSC contains less than 1/3 BaO indicates that Sr > Ca are most active replacements for Ba while Zn and Mg are ineffective.  Table 2.
Especially Mg on the alumina support results in a storage capacity which is indistinguishable from the undoped alumina (i.e. the baseline sample used in this study). The results also demonstrate that the ranking of dopants does not significantly change over the whole tested temperature range and no synergies (e.g. for extending the effective temperature window) between different dopants are found. Another finding from this study is that at 550 °C neither aging nor alkaline earth metal doping have strong impact on the performance, which is an indication that mainly the alumina support is acting as storage function.
Based on the lean efficiency data, the NOx storage capacities are calculated and the average value from three loads for each catalyst are summarized in Table 3. Due to significant loss of NOx storage capacities after hydrothermal aging at 800 °C BET and XRD characterization of selected catalysts were performed. Results on BET surface areas are summarised in Table 2. As to be expected, none of the samples showed a significantly decreasing BET surface area upon aging. Therefore, loss of surface area does not explain the nearly complete activity loss. However, in the XRD patterns a significant increase in the intensity of Pt peaks is observed; an example for the effect of aging on XRD patterns Figure 7. Correlation of NO x efficiencies at different temperatures with Ba content, grouped by co-dopants (three loads of each composition are shown as separate points, lines are drawn through the averages of three loads with the intention to highlight trends), data points are calculated from the area highlighted in Figure 5. Detailed numerical values are given in Table 2.
Especially Mg on the alumina support results in a storage capacity which is indistinguishable from the undoped alumina (i.e., the baseline sample used in this study). The results also demonstrate that the ranking of dopants does not significantly change over the whole tested temperature range and no synergies (e.g., for extending the effective temperature window) between different dopants are found. Another finding from this study is that at 550 • C neither aging nor alkaline earth metal doping have strong impact on the performance, which is an indication that mainly the alumina support is acting as storage function.
Based on the lean efficiency data, the NO x storage capacities are calculated and the average value from three loads for each catalyst are summarized in Table 3. Due to significant loss of NO x storage capacities after hydrothermal aging at 800 • C BET and XRD characterization of selected catalysts were performed. Results on BET surface areas are summarised in Table 2. As to be expected, none of the samples showed a significantly decreasing BET surface area upon aging. Therefore, loss of surface area does not explain the nearly complete activity loss. However, in the XRD patterns a significant increase in the intensity of Pt peaks is observed; an example for the effect of aging on XRD patterns is given in Figure 8.

Discussion
hte has been running parallel reactors for 19 years. During this time a large amount of operating experience has been acquired. In the field of environmental catalyst screening, the most important lesson learned is that realistic test conditions are required to generate relevant data. Oversimplified test conditions and sample preparation methodologies must be avoided. Typical examples are neglecting well-known inhibiting effects of steam and sulfur in the exhaust. This will then postpone the detection of problems to a later, more expensive screening stage. HT units are used for the fast primary screening of new materials under conditions close to the actual application. Even for screening large sample libraries in the initial stage, in most cases such studies take hydrothermal and S-aging into account to avoid costly false positives.
Another important aspect is close interaction with scale-up and engine testing. There is a constant pressure to simplify experiments to save time and/or costs, however, accepting any simplification requires frequent re-evaluation of the underlying assumptions. With every new finding, it has to be verified that an observed effect is real for the application. Environmental catalysis is a rather mature field and performance of state of the art catalysts is at a high level. Every new development needs to be benchmarked against relevant references under the same reaction conditions. Therefore, HT technology is often used for incremental improvements. Existing catalyst technologies are modified by small changes rather than a fully combinatorial screening using application detached, simplified conditions. If necessary, HT units need to be modified to be able run more relevant test conditions. This is facilitated by close integration of HT powder testing with scale-up experiments.
The sample throughput in environmental catalyst screening is still orders of magnitude smaller than what is considered "high throughput" in pharmaceutical or biochemical fields. Neverthless, HT workflows can significantly increase the screening capacity of an environmental catalyst development lab using conventional testing. By applying proper statistical tools like DoE much larger parameter spaces can be screened in a more reliable way.
For efficient utilization of HT screening capacity, designed experiments, automated data processing and statistical methods for catalyst optimization are of high importance. In addition for fully automated test unit operation, also software for data reduction and data management are crucial to handle the large amount of data and drive a rational approach to catalyst development. Under this aspect, HT experimentation has matured beyond the level of pure primary material screening and has become a valuable enhancement to automotive catalyst development. hte's technology platform enables fast material screening while providing for variation of catalyst properties such as preparation methodology and aging parameters. Analysis of HT data allows for differentiation of intrinsic differences between catalyst formulations and can be reliably used for the development of advanced emission control systems to meet ever stricter emission regulations.

Materials and Methods
For illustration of HT screening approach, a sample matrix consisting of 15 model catalysts was prepared and tested fresh and after hydrothermal aging using an LNT protocol.

•
Design of sample matrix: The LNT model catalysts contained alumina as support material. This is loaded with a NO x storage component (NSC) and Pt for NO oxidation. The variation parameter in the sample matrix for this case study was the NSC composition (cf. was used as support. For each catalyst, 5 g alumina were impregnated with a solution of the corresponding alkaline earth metal nitrate using the incipient wetness technique. After careful mixing samples were dried at 100 • C and calcined for 2 h at 500 • C in air. The resulting powders were then impregnated with a solution of Pt(NH 3 ) 4 (NO 3 ) 2 (CAS: 20634-12-2) using incipient wetness impregnation, dried and calcined for 1 h, 400 • C in air. As the alkaline earth metal content in each sample was normalized on a molar basis, the Pt content was set to 1 wt% based on the weight of the alumina carrier. For shaping, the calcined powders were set to slurry with D.I. water (~30 wt% solid content) and milled for 5 min at 500 rpm in a ball mill (using ZrO 2 beakers and milling balls). For pure alumina this procedure was verified to result in a particle size distribution with D 50 < 15 µm. The slurry was then dried under stirring and calcined for 2 h at 500 • C in air. Afterwards the resulting cake was crushed and sieved to a particle size fraction of 250-500 µm used for testing. A fraction of these shaped particles was aged for 12 h at 800 • C in a muffle oven flowed through with a stream of 10% H 2 O in air. Additional aliquots of selected catalysts were aged and submitted to XRD and BET analysis.

•
Test procedure: For the catalytic test, sample amounts were adjusted to have the same amount of Pt in each reactor. This also ensures that, with exception of the baseline sample w/o NSC, the molar amount of the storage component is constant for the whole sample library. For the pure alumina reference, this corresponded to 200 mg diluted with corundum to simulate 1 mL coated catalyst with a washcoat loading of 3.3 g/in 3 . To control time on stream effects and to achieve better statistical robustness, each catalyst was tested in 3 loads, filling one plate of fresh and one plate of aged catalysts in the 48 fold parallel screening unit. Splitting the samples in this way into two plates aims at achieving maximum resolution of the NSC effect among fresh and aged samples.
In the experiment, each catalyst was tested for 5 lean/rich cycles at temperatures T = 550, 450, 350, 250, 200 • C. The total flow in the measured reactor is set to meet a GHSV of 60,000 h −1 based on 1 mL bed volume. A simulated Diesel exhaust gas was mixed using mass flow controllers. During the lean phase (2 min) the feed consisted of: 200 ppm NO, 1500 pm CO, 10% O 2 , 6% H 2 O, 6% CO 2 , balance N 2. Using fast switching magnetic valves this feed is replaced by a rich gas for 10s with minimal perturbation of the flow. In the rich phase the O 2 concentration is reduced to 1% and λ is adjusted to λ = 0.95 by adding CO/H 2 in a ratio of 1:3 while the concentration of other gases remains at their lean level. The gap between lean and rich flow rate is compensated by additional balance N 2 added to the rich stream. • Data processing: Throughout each experiment, the process values from all sensors (temperature, flow, pressure, gas analysers) are recorded with a frequency of 1 Hz and automatically linked by the control software to the corresponding set-points for that condition, and most importantly to the reactor position that is being tested. An example of the typical LNT raw data output at one temperature for each position is shown in Figure 5. For the whole experiment on the complete LNT matrix in the current study, taking about 4 days on the test unit close to 300,000 data points are collected for each individual sensor. Obviously, raw data are not suitable for direct catalyst comparison and data reduction is required. This data reduction process is developed when a new test protocol is implemented and automated in the control software, which e.g., averages the concentration readings for several lean/rich cycles or over a predefined time interval within a lean rich cycle. Some examples on possible sampling time intervals in LNT tests are shown in Figure 6. For efficient screening, different evaluations should be easy to configure in the data management system. In the current study, the average NO x efficiency in the lean phase of the last cycles has been used as performance indicator. The reduced data sets are then stored in a relational database system ("myhte" data warehouse) from which they can be retrieved for further processing (e.g., R, a language and environment for statistical computing was used for analysis in the current case) [39]. For results stored in the database it is possible to relate individual measurements, such as using the inlet concentration measured for the by-pass line to calculate conversion. For the LNT application, an important step is the calculation of the average NO x efficiencies and product distribution (e.g., NO 2 /NO x ratio) within different time windows or after a certain time of the lean phase. The overall process of calculating relevant parameters for the LNT application has been implemented as an automated data processing workflow. Even for an experiment on 48 different samples (see Figures 9 and 10) evaluation is a routine task which requires only little human interaction. In hte's lab, similar workflows have been established for other test protocols, some of them, such as automated extraction of light-off temperatures have been described elsewhere [33,34]. • DoE evaluation: In most cases, experiments in HT screening involve catalyst libraries that are designed based on principles of statistical design of experiments (DoE) [2,3], rather than a collection of unrelated catalysts. The goal of a DoE is it identify cause-effect relationships between the parameters controlled in the experiment (such as composition, thermal activation, or other treatments like milling or binders) and the observed catalyst performance. If possible, samples of a library are prepared and aged together with a reference of known performance to avoid aliasing of effects by uncontrolled factors and errors. If libraries cannot be fit into a single plate, some care has to be taken to control statistical error by introducing proper blocking factors e.g., using split plot designs. Depending on the amount of prior knowledge, either DoEs for factor screening in an early stage (such as fractional factorial designs) or response surface design methodologies can be applied. In the current case a factor screening (impact of different elements as NSC) was attempted. By screening the concentration at three levels also secondary effects can be resolved by the DoE. As the capacity of the 48-fold parallel reactor naturally limits the number of samples, in the current case the variable "Aging" was used as a splitting factor because we were mainly interested in the effect of composition before and after aging. In less obvious cases, computer-generated optimal designs such as D-optimal designs are required to ensure that split-block design constraints do not introduce uncontrolled statistical bias. However, in the current case the hydrothermal aging had such a dramatic effect that this was not critical. As additional QC measure protect against creeping loss of precision, it is advisable to include at least one standard sample into every experimental plate to control effects caused by sensor aging or contamination of the equipment. Specifically, for parallel reactors it is critical to avoid that factors of the experimental design are aliased with respect to either reactor position or time on stream. If time on stream effects can be expected as in   • DoE evaluation: In most cases, experiments in HT screening involve catalyst libraries that are designed based on principles of statistical design of experiments (DoE) [2,3], rather than a collection of unrelated catalysts. The goal of a DoE is it identify cause-effect relationships between the parameters controlled in the experiment (such as composition, thermal activation, or other treatments like milling or binders) and the observed catalyst performance. If possible, samples of a library are prepared and aged together with a reference of known performance to avoid aliasing of effects by uncontrolled factors and errors. If libraries cannot be fit into a single plate, some care has to be taken to control statistical error by introducing proper blocking factors e.g. using split plot designs. Depending on the amount of prior knowledge, either DoEs for factor screening in an early stage (such as fractional factorial designs) or response surface design methodologies can be applied. In the current case a factor screening (impact of different elements as NSC) was attempted. By screening the concentration at three levels also secondary effects can be resolved by the DoE. As the capacity of the 48-fold parallel reactor naturally limits the number of samples, in the current case the variable "Aging" was used as a splitting factor because we were mainly interested in the effect of composition before and after aging. In less obvious cases, computer-generated optimal designs such as D-optimal designs are required to ensure that splitblock design constraints do not introduce uncontrolled statistical bias. However, in the current case the hydrothermal aging had such a dramatic effect that this was not critical. As additional QC measure protect against creeping loss of precision, it is advisable to include at least one standard sample into every experimental plate to control effects caused by sensor aging or contamination of the equipment. Specifically, for parallel reactors it is critical to avoid that factors of the experimental design are aliased with respect to either reactor position or time on stream. If time on stream effects can be expected as in the current case of storage catalysts, testing multiple loads of at least some samples helps to make experiments more robust.
Author Contributions: M.K., O.G. and A.S. wrote the paper. The authors conceived and designed the experiments, collected and analyzed the data and wrote the paper. A.S. did all work related to the numerical data processing.  Accelerating R&D for biofuels and biochemicals T he use of biofuels and biochemicals as alternatives to petroleum-based products has attracted much attention over the last few years. This attractiveness stems from their classification as sustainable products supported by legislative incentives to promote their market penetration.
The use of biomass as a feedstock for the production of fuels and chemicals will, in theory, decrease the dependence on fossil fuels and petrochemicals and reduce greenhouse gas emissions. It is however clear that the introduction of biofuels and biochemicals, and hence the replacement of petroleum-based products, will only be successful if they are commercially competitive. This will result in superior product characteristics and can be directly used as dropin solutions for established chemical value chains.
Biochemicals and biofuels are typically produced by biotechnological processes (e.g. via fermentation or enzymatic transformation), thermochemical processes (e.g. via homogeneous or heterogeneous catalysis) or a combination of both.
In any case, the development of new processes to convert biofeedstocks into drop-in biofuels or biochemicals with a short time to market requires efficient R&D tools. High throughput experimentation (HTE), i.e. the 'many at once' approach, has proven a valuable tool for accelerating traditional chemical and biochemical R&D 1 .

Field of expertise
hte GmbH (hte), located in Heidelberg, Germany, is a worldwide provider of high throughput experimentation tools and services focusing on industrial catalysis. According to hte, the development of novel processes can be accelerated by at least a factor of three when applying its technology as compared to classical (few at a time) R&D.
hte's business model comprises classical contract research utilising high throughput workflows to speed up and optimise R&D processes allowing to test several catalysts and process conditions in a short time, and the sale of turnkey high throughput systems as customised solutions. Test systems range from micro scale to sub-pilot scale.
The main application at hte is in the area of industrial catalysis, comprising homogeneous, heterogeneous and immobilised catalysis. These include continuous systems, such as trickle-bed reactors for heterogeneous catalysts or plug-flow reactors for continuous homogeneous catalysts, as well as batch or CSTR-type reactor systems and bubble column reactors. Reactor volumes range from sub-millilitre scale to several hundred millilitres. The degree of parallelisation typically lies in the range of 1 to 48-fold systems.
The main areas of application include: • Renewables (biofuels, biochemicals, catalytic upgrading, CO2 utilisation) • Environmental (automotive and stationary air pollution control, e.g. TWC, SCR-DeNOx) • Gas to chemicals and fuels (Fischer-Tropsch, GTL, GTO, higher alcohols, reforming) • Catalysis for chemicals (oxidation, reduction, (de-)hydrogenation, condensation, amination, alkane activation, polymerisation) • Petroleum refining (hydroprocessing, hydrocracking, FCC, reforming, alkylation, isomerisation). When developing a chemical or biochemical process, catalyst characteristics must be combined with kinetic process data and product characteristics in order to obtain structureperformance correlations. The experimental data is gathered using different techniques and equipment. In order to obtain a comprehensive view of the data, it has to be merged in a scientific data warehouse. The complete cycle made up of sample preparation, testing, product analysis and data evaluation is called the workflow.
Each element in the cycle has to be de-bottlenecked in order to obtain a fast response. In high throughput experimentation with many experiments taking place simultaneously, the amount of data increases at least in magnitude when compared to conventional testing.
Manual handling is no longer an option due to complexity and the huge number of process steps. This calls for automation of the workflow cycle. Therefore, hte implemented its own fully integrated software workflow with the hteControl process control software and the myhte scientific data warehouse.

Case studies
Converting biomass to biochemicals and biofuels can be performed by means of biotechnology or chemical catalysis. In many cases, both technologies can be successfully combined. Biofuels are typically produced in high quantities, whereby the margin is comparatively low. Biochemicals are inquired and produced in smaller quantities but reach higher prices. As a consequence, the business model for a biorefinery contains both biofuels production to generate sufficient business volume and side stream valorisation of biochemicals to optimise profits.
Chemical catalysis plays a major role in many biomass It is clear that the introduction of biofuels and biochemicals, and hence the replacement of petroleum-based products, will only be successful if they are commercially competitive Developing new processes to convert bio-feedstocks into drop-in biofuels with a short time to market requires efficient R&D tools RENEWABLES Green CMYK c76 m0 y100 k0 Pantone 362 c rgb r61 h164 b42 Blue CMYK c100 m56 y0 k0 Pantone 293 c rgb r12 g71 b157 Helvetica Black and Helvetica Ultra light biofuels R&D 60 november/december 2014 biofuels international Figure 1: Upgrading of first generation biofuel in a continuous trickle flow high throughput unit at hte. The hydrogenation of rapeseed oil with hydrogen and HDS catalysts at different temperatures leads to products with C-numbers ranging from C1 (gaseous), over C8 (liquid) to C18 (solid). Pictures adapted from [2] conversion routes either as a core technology (e.g. biomonomers for bioplastics) or in product upgrading and downstream processing (e.g. upgrading of bioethanol through fermentation to dropin biofuels and bioethylene for bioplastics). In particular, biofuels can be significantly improved through catalytic upgrading. Biofuels obtained by fermentation or from biooils typically face blend wall limitations when combined with or expected to directly replace traditional petroleumbased fuels. Catalytic upgrading can convert the biofuels into drop-in fuels with characteristics very similar to petroleum-based fuels and that can hence undergo a faster certification process.
Below are two case studies on bio-oil upgrading and waste stream valorisation by using high throughput experimentation to speed up R&D and to enable the testing of several catalysts and conditions in a short time. The upgrading of vegetable oil to hydrotreated vegetable oil (HVO) is referred to as a first generation drop-in biofuel and the valorisation of glycerol as a by-product from biodiesel production.

Hydrotreated vegetable oil
First generation biodiesel directly derived from vegetable oil, such as rapeseed oil, is only of limited use as a transportation fuel due to engine restrictions and storage instabilities. When upgraded by catalytic hydrogenation it is composed of long-chain hydrocarbons and called HVO. As its chemical composition and fuel characteristics are very close to petroleum-based diesel, it can be considered a drop-in fuel fully applicable as a diesel substitute.
Catalytic testing comprises the performance screening of different suitable catalysts with different experimental parameters. Therefore, a 16-fold hte test unit was chosen to obtain a high degree of parallelisation facilitating the testing of many different catalysts and reaction conditions within a single experiment 2 . The challenge is to find the optimal catalyst and conditions for obtaining a well-defined liquid hydrocarbon which can be directly used as a drop-in fuel. Figure 1 shows a typical continuous 16-fold trickle bed unit suitable for such hydroprocessing applications.
The content of the sample glasses shown in Figure 1 indicate a varying product distribution ranging from solid to gaseous products, depending on the type of catalyst and reaction conditions. Low reaction temperatures lead to solid n-alkanes in the C17-C18 range, whereby octadecane is the main product. By increasing the reaction temperature liquid n-alkanes in the range of C7-C13 are formed. At even higher temperatures mainly gaseous products in the C1-C6 range are generated.
In summary, the study shows that it is not only important to do a performance screening of different catalysts but also of suitable reaction conditions, here demonstrated using the example of reaction temperature. If the temperature is too low, only solid products will be obtained. If the temperature is too high, the products become gaseous. Therefore, the optimum temperature, leading to a liquid saturated hydrocarbon, lies within a small window. In this case study, high throughput experimentation is applied successfully for the parameter screening of bio-oil exhibiting properties comparable to fuel. hte technologies can handle first generation biooil and its hydroprocessing products, which are suitable as drop-in fuels.

Valorisation of glycerol for biochemicals
Glycerol represents a fundamental feedstock molecule due to its availability as by-product within first generation biodiesel production (transesterification) and its importance as a platform chemical within the petrochemical value chain. Therefore, the oxidative transformation of glycerol to acrolein and acrylic acids as well as the carbonylation of glycerol to C4 acids has been Catalytic upgrading can convert biofuels to drop-in fuels with characteristics very similar to petroleum-based fuels and can hence undergo a faster certification process RENEWABLES Green CMYK c76 m0 y100 k0 Pantone 362 c rgb r61 h164 b42 Blue CMYK c100 m56 y0 k0 Pantone 293 c rgb r12 g71 b157 Helvetica Black and Helvetica Ultra light R&D biofuels biofuels international november/december 2014 61 chosen to demonstrate the benefits of high throughput experimentation both in catalyst screening and process optimisation 3 . The oxidative transformation of glycerol to acrolein was performed in hte's 48-fold fixed bed unit. This unit is suitable for fast screening in the gasphase to identify interesting lead structures from a large number of possible catalyst candidates. Moreover, the feed composition was investigated since glycerol is available as glycerol/water mixture from biodiesel production.
A fully integrated software workflow is a necessary tool for handling and correlating the large amount of data gathered from the screening (catalyst performance, reaction conditions, feed composition) and the catalysts themselves (physical and chemical properties). For instance, more than 1500 online GC chromatograms are recorded and have to be evaluated per week. The aim of this fast screening was to find the catalyst with the best performance in acrolein production with a high catalyst lifetime. During screening, not only were a group of promising catalyst candidates found, it was also observed that the reaction conditions strongly affect acrolein yield and catalyst deactivation.
A similar study is demonstrated by the carbonylation of glycerol to C4 acids. In this case, many different potential catalyst candidates and varying reaction conditions with different feed mixtures and co-feeds were investigated. The homogeneously catalysed liquid phase reaction was carried out in an 8-fold batch reactor system built by hte 3 .
The liquid products containing C4 acids were analysed by offline gas chromatography and mass spectrometry. Basically, several C4 acids are obtained from the glycerol carbonylation, whereby their composition strongly depends on the reaction parameters. Figure 2 shows how the yields of the C4 acids depend on the reaction parameters as a maximum yield is achieved at a defined temperature and CO partial pressure. By focusing on the yield and the product distribution at different reaction conditions it is possible to fine-tune the variables that can enhance catalyst performance.
With these two case studies the valorisation of a waste stream product, glycerol, was demonstrated in liquid and gas phase. The case studies show that high throughput experimentation is an important tool not only for screening catalyst libraries but also for finding optimal process conditions. The degree of parallelisation has to be adapted to the needs of the individual project. Through fast screening of potential valorisation options for side or waste stream products, hte can directly add value to the profitability of a biorefinery.

Conclusion
High throughput experimentation is a powerful tool for accelerating R&D on novel chemical and biochemical processes by using a high degree of parallelisation and automation.
The advantage with high throughput technology was demonstrated in two case studies: the upgrading of rapeseed oil as a first generation drop-in biofuel and the valorisation of glycerol as a side stream product. In both cases, as well as in general for R&D activities in the field of biofuels and biochemicals, the demand to shorten the time to market is extremely important and this can be significantly reduced by applying high throughput technology developed by hte. l