Production of Acetaldehyde via Oxidative Dehydrogenation of Ethanol in a Chemical Looping Setup

A novel chemical looping (CL) process was demonstrated to produce acetaldehyde (AA) via oxidative dehydrogenation (ODH) of ethanol. Here, the ODH of ethanol takes place in the absence of a gaseous oxygen stream; instead, oxygen is supplied from a metal oxide, an active support for an ODH catalyst. The support material reduces as the reaction takes place and needs to be regenerated in air in a separate step, resulting in a CL process. Here, strontium ferrite perovskite (SrFeO3−δ) was used as the active support, with both silver and copper as the ODH catalysts. The performance of Ag/SrFeO3−δ and Cu/SrFeO3−δ was investigated in a packed bed reactor, operated at temperatures from 200 to 270 °C and a gas hourly space velocity of 9600 h–1. The CL capability to produce AA was then compared to the performance of bare SrFeO3−δ (no catalysts) and materials comprising a catalyst on an inert support, Cu or Ag on Al2O3. The Ag/Al2O3 catalyst was completely inactive in the absence of air, confirming that oxygen supplied from the support is required to oxidize ethanol to AA and water, while Cu/Al2O3 gradually got covered in coke, indicating cracking of ethanol. The bare SrFeO3−δ achieved a similar selectivity to AA as Ag/SrFeO3−δ but at a greatly reduced activity. For the best performing catalyst, Ag/SrFeO3−δ, the obtained selectivity to AA reached 92–98% at yields of up to 70%, comparable to the incumbent Veba-Chemie process for ethanol ODH, but at around 250 °C lower temperature. The CL-ODH setup was operated at high effective production times (i.e., the time spent producing AA to the time spent regenerating SrFeO3−δ). In the investigated configuration with 2 g of the CLC catalyst and 200 mL/min feed flowrate ∼5.8 vol % ethanol, only three reactors would be required for the pseudo-continuous production of AA via CL-ODH.


GC analysis of the gas products from the reduction step
An extended reduction step was performed over CuO/Al2O3. Samples of the outlet gas were collected manually at 50 s and 10 min into the reduction step and analysed with the GC; results are shown in Fig. S1. Consistent with the CuO/Al2O3 chemical looping (CL) profiles shown in Fig. 2 of the main manuscript, no acetaldehyde (AA) (4.4 min) or ethanol (6.8 min) was observed in the FID signal for a sample taken 50 s into the reduction step. Very small amounts of ethylene (1.1 min) and diethyl ether (14 min) were found instead, indicating that a small fraction of ethanol dehydrated over the α-Al2O3 support, which was unexpected given the low operating temperature (250°C) and α-Al2O3 being largely inactive for dehydration 1 . At the same time, low concentrations of CO2 (4.3 min) and H2 (1.2 min) were detected by the TCD.
The absence of AA when H2 was detected confirms that ethanol cracking dominated early in the reduction step.
The GC samples taken 10 min into the reduction step yielded considerably more dehydrogenation products. Acetaldehyde was present at 2.1vol%, H2 at 7.4vol%, whilst unreacted ethanol was also detected, with no CO2 produced. Again, dehydration products, ethylene and diethyl ether, were also detected, albeit at low concentrations. The prolonged reduction with CuO/Al2O3 demonstrates that this Cu-sample becomes selective towards AA only after the first event of prominent coking and carbonation (also see Fig. S10 for the XRD results of the samples after 1.5 or 60 min of reduction). Figure S2: The FID results from GC measurements taken during the reduction step of the 3 rd CL cycle over SrFeO3, Ag/SrFeO3, and Cu/SrFeO3.
Typical FID results of GC measurements taken during the reduction step are shown in Fig. S2. Acetaldehyde and ethanol are present at the 2 to 4vol% range. Whilst other hydrocarbons can be detected, none exhibited an FID response greater than 3 pA, which corresponds to concentrations in the 10s and 100s of ppm range. The surface-weighted mean diameter (Sauter diameter), d3,2, was calculated by Eq. S1 using particle diameters, di, that were manually measured from STEM images and EDS maps of fresh and spent Ag/SrFeO3 and Cu/SrFeO3. At least three different sections of the samples were imaged and no less than 150 particles were measured.

Size distribution of Cu and Ag particles in spent and fresh samples
3,2 = ∑ 3 ∑ 2 Eq. S1     The surface elemental composition was determined via energy-dispersive X-ray spectroscopy (EDS) analysis performed using an Oxford Instruments Aztec Energy X-maxN system with an accelerating voltage of 15 kV and a working distance of 15 mm. A SEM-EDS image of the fresh CuO/SrFeO3 sample is given in Fig. S8.

XRD results of spent and fresh SrFeO3-based materials
The following figures (Figs. S9 -S11) present the same XRD results as in Fig. 9 of the main manuscript. Here, plotted over a smaller range of signal intensity to show minor peaks.

Reduction of SrFeO3-δ
Inspection of the reactor bed containing Ag/SrFeO3 following 1.5 min of reduction, seen in Fig. S12, showed a clear visible change in the top portion of the bed to SrFeO2.5, which corresponds to the section of the bed that was sampled for XRD analysis.

Additional TGA results of SrCO3 under an air 'reactive' gas and SrFeO3-based materials under a CO2 'reactive' gas
The decomposition of SrCO3 (Sigma Aldrich, ≥ 98%) over a temperature cycle from 50-900°C with a temperature ramp rate of 10 °C min -1 with ai as th ' activ ' gas was obs v d in th TGA, seen in Fig. S14. Under a CO2-free atmosphere, SrCO3 is expected to be thermodynamically unfavorable, however, the decomposition of SrCO3 to SrO is limited by slow kinetics 3 . The SrCO3 started decomposing at ~700°C and continued to decompose until the TGA had cooled to 800°C. Two temperature cycles were performed in CO2 with the spent SrFeO3-based materials, shown in Fig. S15, as CO2 is expected not to re-oxidise samples beyond the SrFeO2.5 phase 2 , nor remove any SrCO3; whilst coke removal is expected at temperatures above 750°C 4 . Immediately after the cycles in CO2, two temperature cycles in air were performed, also seen in Fig. S15. All samples gained mass over 600-800°C, a result of carbonate formation, also observed with fresh CuO/SrFeO3, presented in Fig. S16. The Cu/SrFeO3 sample gained significantly more mass than either SrFeO3 or Ag/SrFeO3. The mass change between the start and end of the first cycle was greatest again with Cu/SrFeO3, indicating the largest removal of impurities, likely coke in this instancecorroborating the observations in Fig. 10 in the main manuscript.
The first cycle in air for both SrFeO3 and Ag/SrFeO3 showed very additional mass loss compared to the second cycle with most impurities being removed during the cycles in CO2. Little to no carbonates remained following the cycles in CO2, as SrCO3 only begins to decompose at >750°C, which was not observed in the first cycle in air. Thus, SrFeO3 and Ag/SrFeO3 were resistant to carbonate reformation in a CO2 atmosphere (0.33 bar CO 2 ). The Cu/SrFeO3 sample showed significant additional mass loss at ~700°C compared to the second air cycle. Thus Cu/SrFeO3 appears to be the most prone to carbonate formation in a CO2 atmosphere. The presence of water in a CO2 atmosphere has been found to enhance carbonate formation, thus resistance to carbonate formation in a water-free CO2 atmosphere may not be a reliable indicator for carbonate resistance under dehydrogenation conditions 5 Figure S15: Relative mass changes of spent (a) SrFeO3, (b) Ag/SrFeO3, and (c) Cu/SrFeO3 during 2 cycles in CO2 followed by 2 cycles in air with temperature cycling between 50 -900°C and heating rate of 10°C min -1 . Figure S16: Relative mass change of a sample of fully oxidised and impurity free CuO/SrFeO3 during a temperature-programmed cycle from 50-900°C with a temperature ramp rate of 10°C min -1 performed with either N2 or CO2 as the reactive gas.

Raman analysis of spent and fresh samples
The fresh and spent samples of SrFeO3, CuO/SrFeO3 and Ag/SrFeO3 were characterised using Raman spectroscopy to detect the formation of coke on the surface of the samples. The Raman spectrometer (Horiba Jobin Yvon), equipped with an Olympus BX41 microscope (×50 objective) was used to acquire the spectra of the samples. Raman spectra were excited by a Nd:TAG laser (532.8 nm, 5 mW) in the range of 100 -1700 cm -1 . An exposure time of 30 s was used. To improve the signal-to-noise ratio, the acquisition was repeated five times. Prior to the measurement, the instrument was calibrated using the 520.5 cm -1 line of silicon. All spectra were normalised against the maximum intensity of the signal recorded during each scan for a given sample and the results are presented in Fig. S17.
The acquired spectra were compared with the spectra from the literature for SrFeO3-δ 6,7 , SrCO3 8 and Ag 9 . For amorphous carbon, Raman responses at ~1350 cm -1 (D band) and ~1580 cm -1 (G band) were expected 10 . In the analysed samples, the carbon D-band overlapped with the broad spectra of SrFeO3; hence, the presence of coke could not be inferred from the D-band. From Fig. S17 (a), SrCO3 was detected in the fresh sample of SrFeO3 but not in the spent samples, after reduction. However, SrCO3 was present in both the fresh and spent samples of CuO/SrFeO3 and Ag/SrFeO3. The intensity of the broad peak at ~ 1370 cm -1 for all the spent samples was greater than for the fresh sample, owing to the lower value of oxygen nonstoichiometry, 3 -δ, consistent with the findings in the literature 7 . For the spent sample of Ag/SrFeO3-δ, the measurement of Raman spectra was repeated at 50 different locations. Two types of spectra with different relative intensities at various Raman shifts were obtained, presented as Spent#1 and Spent#2. Raman spectra of Spent#1 were slightly more probable, acquired at ~ 30 different locations. The most distinctive difference in Spent#1 and Spent#2 is the intensity at wavenumber of ~680 cm -1 , which was attributed to the Ag particles.
In summary, Raman spectra, presented in Fig. S17, showed only very weak evidence of the presence of coke on the surface of samples, SrFeO3 and Cu/ SrFeO3. The inability of Raman spectroscopy to detect coke was not surprising considering the low amount of coke on the samples as shown in the main manuscript, Fig. 10.