Demonstrating on-demand production of bio-ethylene oxide in a two-step dehydration-epoxidation process with chemical looping operations

Ethylene oxide (EO) is a key chemical intermediate produced almost exclusively from petrochemically derived ethylene. Currently


Introduction
Ethylene oxide (EO) is a versatile chemical intermediate with an annual global production of 34.5 Mt (2017) [1].The majority of EO feeds to the manufacture of ethylene glycol or other specialty chemicals [2], with a small fraction directly used in sterilising medical equipment [3,4].The incumbent EO synthesis has a large carbon footprint, with CO 2 emissions of 0.88 t CO2,eq per tonne of EO [5], making the process one of the most carbon-intensive in the industrial production of bulk chemicals [6].Further emissions associated with EO come from the petrochemical-based ethylene, with 1.2 t CO2,eq per tonne of ethylene from steam-cracking installations [5].Current EO plants can only tolerate small deviations (±10 %) in the production outputs, thus, flexible operations are difficult, making the current EO production unsuitable for direct integration with renewable and low CO 2 resources [7].
The incumbent method for producing EO proceeds via direct epoxidation, where ethylene reacts with O 2 , either selectively to EO or unselectively to CO 2 and H 2 O.The process is carried out at 230-270 • C in multi-tubular reactors over Ag/α-Al 2 O 3 catalysts [2].The theoretical limit on selectivity is ~50 %, but catalyst modifications with metal dopants (e.g.Cs, Re and Na) and Cl-based reaction moderators, lead to improvements ranging ~70-90 % for selectivity to EO [8,9].Simultaneously, the conversion of ethylene is limited to prevent flammable mixtures from forming [2].Additionally, the extreme flammability of EO requires that large volumes of inert ballast gas are introduced and the EO product at the reactor outlet limited to 1-3 vol%.The hazards and downstream processing associated with direct epoxidation make the process only feasible at a large scale.
Chemical looping epoxidation (CLE) replaces the Ag/α-Al 2 O 3 catalyst with Ag/SrFeO 3 [10], using SrFeO 3 as a source of oxygen to the epoxidation reaction, thus the solid oxide is commonly termed oxygen carrier.Compared to direct epoxidation, the use of an oxygen carrier eliminates the need for O 2 , preventing any EO-O 2 mixing, thus, eliminating the biggest process risk associated with highly explosive EO mixtures.Once depleted of oxygen, SrFeO 3 is regenerated in a separate step in air and the process operates cyclically.Chemical looping epoxidation has several advantages over direct epoxidation: 1) process safety is improved; 2) ballast gases are no longer required, which decreases the need for downstream processing; 3) the costly separation of oxygen from air is eliminated as CLE uses ambient air for the regeneration step.Thus, chemical looping epoxidation is a promising alternative to direct epoxidation, offering a safer and potentially less-costly route to EO.
The production of EO can be uncoupled from petrochemistry by replacing ethylene, with widely available and renewable bioethanol as a feedstock.Ethanol readily loses water over acidic catalysts (e.g.γ-Al 2 O 3 and HZSM-5 zeolite) [11], producing bio-based ethylene.Hence, by coupling the dehydration of bioethanol and CLE, EO can be produced from a bio-derived feedstock and at a small scale.Here, we present a novel route to EO, whereby EO is formed in a combined dehydrationepoxidation process, following the chemical looping approach.The presented results demonstrate that the novel method operates flexibly and on-demand, both using laboratory-grade or highly denatured ethanol as a feedstock.Different reactor arrangements for the dehydration and CLE reactions were explored to identify that a single-reactor dehydration-epoxidation process results in a promising and simple process configuration for producing green EO from a widely available bio-feedstock.
Particles of SrFeO 3-δ impregnated with Ag catalyst were used for the epoxidation step.Strontium ferrite perovskite was prepared by a solidstate synthesis method described in a previous study [13].Briefly, stoichiometric amounts of Fe 2 O 3 (Honeywell, >98 %) and SrCO 3 (Sigma Aldrich, >99.8 %) were added to a stainless-steel grinding bowl, with ethanol (Alfa Aesar, 99.95 vol%) as a binding agent.The mixture was ball-milled at 600 rpm for a total of 30 min before drying at 50 C for 24 h and sieving to 180-355 µm.The resulting particles were calcined four times in static air for 3 h at 1000 C, using a ramp rate of 5 C min − 1 , then, again sieved to 180-355 µm.The obtained SrFeO 3-δ has a stoichiometry of SrFeO 2.82 at room temperature [14], but will be denoted at SrFeO 3 .
Silver was deposited onto the surface of SrFeO 3 to a target loading of 15 wt% via incipient wetness impregnation.Silver nitrate (Alfa Aesar, >99.9 %) was dissolved in deionised water and the solution added dropwise to 5.00 g of SrFeO 3 under constant manual mixing with a spatula.The volume of liquid matched the accessible pore volume of the SrFeO 3 , determined empirically beforehand.The sample was dried at 120 C for 12 h, followed by calcination at 650 C for 5 h, both in static air.
In some experiments, water removal from the dehydration step was considered.In such cases, A3 molecular sieve (A3 MS, Sigma Aldrich) was used as a drying agent.Beforehand, the A3 MS was crushed and sieved to 180-355 µm and dried at 400 C under N 2 (99.9995 vol% N 2 , BOC) for 1 h.

Material characterisation
Thermogravimetric analysis (TGA) with a Mettler Toledo TGA/DSC 1 was performed on the fresh and spent samples of Ag/SrFeO 3 to assess the extent of surface impurities following experiments.About 40 mg of sample was placed in a 70 µL alumina crucible and positioned on the balance arm of the TGA.Each experiment consisted of two cycles of heating from 50 to 900 C at 10 C min − 1 , followed by cooling down to 50 C at the same rate.A stream of blended air (21 ± 0.5 vol% O 2 / N 2 , BOC) was passed over the sample at 50 mL min − 1 (NTP) through a capillary tube positioned above the crucible.Additionally, the TGA balance required purging with 100 mL min − 1 N 2 , thus, the effective oxygen concentration in the TGA chamber was ~7 vol%.Mass changes recorded in the experiments were corrected for buoyancy effects by extracting a blank experiment with 40 mg of silica sand.To discern the mass of impurities from the mass of SrFeO 3 , the obtained results were normalised against the sample mass recorded at the end of the first cycle in air, with 100 wt% corresponding to the re-oxidised sample, free of surface impurities.
Bright-field (BF) and high-angle annular dark field (HAADF) Scanning Transmission Electron Microscopy (STEM) images and Energy Dispersive X-ray Analysis (EDS) maps were acquired using a Thermo Scientific (FEI) Talos F200X G2 operating at 200 kV.The STEM images were collected at a camera distance of 98 mm and EDS maps were collected using the Super-X EDS detector.The samples were prepared grinding Ag/SrFeO 3 particles in isopropyl (IPA) alcohol and pipetting μL of the suspension onto holey carbon film on 300 mesh Cu/Ni grids (EM Resolutions).

Experiments in packed-bed reactors
The conversion of ethanol to EO was performed in two experimental arrangements: (1) two separate packed-beds for the dehydration and epoxidation reactions, (2) a single packed-bed reactor.
The two-reactor setup is presented in Fig. 1a.The rig consisted of two identical stainless-steel tubes (15 mm i.d.) each placed within a tubular furnace.The bed materials were introduced from the top of the tubes in three layers: (1) a bottom layer of 2 g SiC (Alfa Aesar, 305-356 µm); (2) 2 g of active material; (3) a top layer of 4 g SiC for preheating the feed gas.The active material in the 1st reactor was either HZSM-5 or γ-Al 2 O 3 , while Ag/SrFeO 3 was used in the 2nd reactor.The active layers were positioned in the isothermal region of the furnace, with K-type thermocouples in the middle of each bed to control the heating of the furnace.
The feed to the dehydration reactor was prepared by passing 190 mL min − 1 N 2 through a bubbler containing ethanol, resulting in a ~5.5 vol % ethanol (Alfa Aesar, 99.95 vol%).In certain experiments, the ethanol feedstock was denatured with 7.1 mol% methanol (Acros Organics), and 3.8 mol% IPA (VWR).Ethanol/N 2 stream was fed into the dehydration reactor, operating at 200-350 C. The outlet gas was passed through a bed of A3 MS, held at ambient temperature, to remove water formed during dehydration.The resulting dry stream of ethylene was then either passed to vent or fed to the second reactor.
The second reactor was used to selectively oxidise ethylene to EO in a CL manner.The feed to the reactor thus changed with the chemical looping steps, which included of: 1) 1.5 min reduction in ethylene; 2) min purge in N 2 ; 3) 15 min regeneration in air; and 4) 2 min purge in N 2 .The reactor operated at 270 C with feed-gas delivered at 200 mL min − 1 , giving GHSV of 9600 h − 1 , with the bulk density of Ag/SrFeO 3 measured as 1590 kg m − 3 .The second reactor was either operated with ethylene feed produced by the dehydration reactor, or as a stand-alone unit, drawing ethylene feed from a cylinder (5.4 vol% C 2 H 4 /N 2 , BOC).
The combined two-reactor rig was configured such that the second reactor could be bypassed, so that outlet streams from either reactor could be sent to an online Fourier Transform Infrared (FTIR) analyser (MKS Instruments, Multigas 2030) with a liquid N 2 cooled HgCdTe detector.Before each experiment, the detector was cooled for 2 h.During that time, if the epoxidation section was operated, the epoxidation reactor was heated to 270 • C and kept at this temperature under a flow of air.A single FTIR scan was performed for 0.97 s, measuring the signal in the range of 100 -5000 cm − 1 at a resolution of 0.5 cm − 1 .Intermittently, gas samples were manually collected with a syringe and introduced into an Agilent 6850 gas-chromatograph (GC), equipped with a CP7595i5 column (Agilent) and flame ionisation detector (FID) detector.
The setup with a single dehydration-epoxidation reactor (Fig. 1b) consisted of a stainless-steel tube (9 mm i.d.) with thermocouples (TCs) positioned along the tube.The bed material was loaded in 6 layers: 1) a bottom layer of 2 g Ag/SrFeO 3 aligned so that TC 9 was at the bottom of the layer; 2) a layer of α-Al 2 O 3 ; 3) 4 g of A3 MS aligned so that TC 5 was in the centre of the layer; 4) a layer of α-Al 2 O 3 ; 5) 2 g of HZSM-5 positioned such that TC 1 was at the bottom of the layer; and; 6) a layer of α-Al 2 O 3 .The layer of A3 MS was replaced with more α-Al 2 O 3 for experiments without intermediate drying.The gas feed to the reactor was identical to the operation in the two-reactor setup, i.e. 5-5.5 vol% C 2 H 5 OH in N 2 .
In the single dehydration-epoxidation reactor, two heating tapes were used with thermocouples TC 1 and 9 controlling the heating power.The TCs position aligned with the bottom sections of HZSM-5 and Ag/ SrFeO 3 layers, respectively.The heating tapes were wrapped such that they centred around the middle of the corresponding active materials.The HZSM-5 layer was heated to 280 • C, and Ag/SrFeO 3 to 270 • C.
The analysis of the packed-bed experiments is described in Section S1 in the SI.

Results
The dehydration of 5.5 vol% ethanol over γ-Al 2 O 3 was investigated over 250-400 C, with the average product distribution obtained at steady state shown in Fig. 2a.Dehydration at 250 C, the lowest applied temperature, resulted in the incomplete conversion of ethanol, producing diethyl ether rather than ethylene, as expected from earlier reports [17].A complete conversion of ethanol to ethylene and water was achieved by increasing the operating temperature to 300-400 C.After changing the catalyst from γ-Al 2 O 3 to HSM-5 zeolite, dehydration was also successful, but, besides producing ethylene, led to coke formation and the production of C3 and C3 + products from ethylene oligomerisation (results in Fig. S2 in the SI).For later experiments aimed at EO synthesis, where we put together dehydration and epoxidation reactors, only dehydration over γ-Al 2 O 3 at 350 C was used, to ensure high-purity ethylene and complete conversion of the ethanol feedstock.
The results from operating just the epoxidation reactor are provided in Fig. 2b, with the required ethylene provided from a cylinder (5.4 vol% C 2 H 4 in N 2 ).The reactor was operated in a chemical looping (CL) mode, with cycles comprising two steps: epoxidation and material regeneration.Consequently, EO and CO 2 products in Fig. 2b were detected only during the epoxidation step.Consistent with previous CLE studies [10,13], the concentrations of CO 2 and EO were highest during the 1st CLE cycle, decreasing in the next cycles, and stabilising from the 4th cycle onwards.
Fig. 2c and d provide the results for the production of EO from ethanol using combined dehydration and CLE rigs, thus, demonstrating the proposed new concept for on-demand EO delivery.The amount of EO produced per CLE cycle was ~20 % higher than when using ethylene from a cylinder (Fig. 2c vs 2b), i.e. when running CLE alone, possibly because of a higher concentration of ethylene (5.5 vol%) achieved in the dehydration step than provided from a cylinder (5.4 vol%).The selectivity to EO remained similar in both configurations, at 55 %, but the conversion of ethylene improved, reaching 15 % in the combined dehydration-epoxidation rig.
The combined dehydration-epoxidation process produced EO at a high selectivity, consistently remaining at ~55 % between the 2nd to 10th cycle (Fig. 2d).The only start-up procedure required was a 2 h preheat in air, making the proposed technical arrangement suitable for the 'on-demand' production of EO.To assess whether the combined dehydration-epoxidation process could produce EO with similarly consistent results over a longer timescale, the rig was operated once a week for 3 weeks (Fig. 3a).Between experiments, the active materials, Ag/SrFeO 3 and γ-Al 2 O 3 , were kept at ambient temperature and under static air, without any regeneration steps.Over the 3 weeks of sparse experiments, the obtained selectivity to EO and conversion of ethylene, shown in Fig. 3a, were very stable and repeatable, with a small and unexplained drop in selectivity in the 2nd week.The active materials did not degrade during operation or the prolonged inactive time.Results in Fig. 3a demonstrate that the combined dehydration-epoxidation process could be operated periodically, only producing EO when required, without any additional storage requirements between use.
Ethanol-water mixtures below the azeotrope (<95 vol% ethanol) and ethanol mixed with denaturants (e.g.methanol, IPA) are substantially cheaper than high-purity ethanol (>99 vol%) and are often more readily available.The high-purity ethanol was thus changed to denatured ethanol containing 7.1 mol% methanol, and 3.8 mol% IPA.The liquid was evaporated, obtaining a stream of 5.2 vol% ethanol, 0.4 vol% methanol and 0.2 mol% IPA in N 2 , which was fed to the combined dehydration-epoxidation processresults are presented in Fig. 3b.Within the dehydration reactor, IPA almost completely dehydrated to propylene, regardless whether HZSM-5 or γ-Al 2 O 3 were used for the dehydration step, whilst the methanol partially dehydrated to dimethyl ether (DME).The fate of both components was assessed with FTIR and GC, and results are given in Figs.S4-S5 in the SI.
The results in Fig. 3b demonstrate that our catalyst-oxygen carrier composite, the Ag/SrFeO 3 , tolerated the presence of denaturants, and their dehydration products, without obvious deactivation or degradation, evidenced by the consistent performance between CLE cycles.The presence of denaturants did result in a decrease in selectivity to EO compared to the results using a pure ethanol feed (Fig. 2), most likely because of the lower concentration of ethylene in the feed.The high activity of the material resulted in the overall higher conversion, assessed for all dehydration products entering the CLE reactor.
The GC results (see Fig. S6 in the SI) revealed that negligible amounts of methanol, DME, IPA, and propylene were detected at the CLE outlet.Therefore, most of the denaturants, introduced at the % level, must have oxidised to CO 2 over Ag/SrFeO 3 .To account for the oxidation of the denaturants into account, the fraction of CO 2 produced from their combustion was extracted from the calculations of selectivity to EO and conversion of ethylene, giving results at ~55 % and ~17 %, respectively very similar to those obtained with high-purity ethanol (Fig. 3a).
An alternative configuration for the dehydration-epoxidation process was investigated, whereby both the dehydration and epoxidation catalysts were sequentially layered within a single packed-bed reactor.With ethanol fed from the top of the reactor, the first catalytic layer was HZSM-5 for dehydrating ethanol to ethylene, followed by a layer of Ag/ SrFeO 3 for oxidising ethylene to EO.The entire reactor was operated in CL cycles, i.e. exposing the multi-layered bed for 1.5 min to ethanol in N 2 , 2 min N 2 , 15 min air, 2 min N 2 .Besides the two catalysts, an intermediate layer of A3 molecular sieve (MS) was added in some experiments to remove the water produced by the dehydration reaction, but the drying layer was omitted in another set of experiments to investigate the effect of water on epoxidation.The results from a single-reactor configuration with and without intermediate drying are shown in Fig. 4. In both cases, the selectivity to EO and conversion of ethylene were not consistent between cycles, thus, the experiment was conducted over multiple days until the performance stabilised.The results in Fig. 4a and b, without a drying layer and with it, both clearly demonstrate that EO can be produced from ethanol within a single reactor vessel operated in a CL mode.In the case without an intermediate layer of A3 MS, the selectivity to EO was lower than in a reference CLE experiment when sourcing ethylene from a cylinder (Fig. S7 in the SI).In the single reactor with multi-layer configurations, the selectivity to EO improved over subsequent cycles, at the cost of dropping ethylene conversion, and both parameters stabilised with cycling, each day achieving approximately constant yield of EO per cycle.Such behaviour indicates that the unselective overoxidation of ethylene diminished as the CL cycles proceeded, confirmed by the product distribution at the rig outlet -given in Fig. S8 in the SI.Looking at Fig. 4a -since these experiments did not involve the drying step, the CLE reaction was operated in the presence of H 2 O (~5 vol%), which did not degrade the activity of Ag/SrFeO 3 .
The inclusion of an intermediate layer of A3 MS (Fig. 4b) drastically improved both the selectivity to EO and amount of EO produced per cycle with the achieved conversion comparable to the case without the drying material (Fig. 4a).Again, the selectivity to EO increased upon cycling, whilst the conversion decreased, and the amount of EO produced per day was roughly constant each day.Following the 16th cycle, the selectivity to EO exceeded that observed in the reference experiment feeding ethylene/N 2 mixture (Fig. S7 in the SI).Comparing the configurations with and without an intermediate layer of A3 MS (Fig. 4a vs Fig. 4b), the presence of water at ~5.5 vol% drastically decreased the production of EO, but did not affect conversion, demonstrating that H 2 O enhanced complete combustion either of ethylene or EO.
The amount of oxygen released to EO and CO 2 during each reduction step in Fig. 4a and b is presented in Fig. 4c.The setup without a drying layer consistently released more oxygen compared to the setup with a layer of A3 MS.The unselective combustion of ethylene requires 6 times the oxygen as the epoxidation reaction.Thus, the decrease in oxygen release in the setup with a layer of A3 MS was a result of the improved selectivity to EO, since the conversion of ethylene was consistent between both experiments.
Thermogravimetric analysis was used to assess whether impurities accumulated on the surface of the Ag/SrFeO 3 during the experiments within the single reactor vessel.Spent Ag/SrFeO 3 samples collected from the packed-bed reactor were heated from 50 to 900 • C and cooled back to 50 • C under air flow in the TGA, with the heating-cooling cycle performed twice.The observed mass changes and 1st derivatives of each heating step are shown in Fig. 5.The change in mass over the second cycle was a result of thermally induced reduction and oxidation of Ag/ SrFeO 3 , and agreed with previously reported results [18].Thus, differences in mass changes between the 1st and 2nd cycles in Fig. 5 correspond to impurities and other changes to Ag/SrFeO 3 that occurred in the packed-bed experiments.
In Fig. 5, a mass increase over 300-400 • C can be ascribed to the regeneration of SrFeO 3 in air, indicating that the retrieved samples were partially reduced despite each experiment ending with an oxidation step.Only the sample used without a drying layer exhibited a substantial mass loss over 50-300 • C. Given the sample was exposed to ~5 vol% water from the dehydration reactions, the mass loss can be linked to the removal of adsorbed water [19] or the dehydration Sr(OH) 2 ⋅xH 2 O [20], a reported impurity on SrFeO 3 following exposure to water [21].For both retrieved samples, the 1st derivatives of the mass loss differ between the 1st and 2nd cycles at 700-800 • C, which is attributable to SrCO 3 decomposition [18,22].Potential coke removal would have occurred at lower temperatures [18].Thus, Fig. 5 reveals that the main impurity on the spent samples was SrCO 3 , which accumulated regardless of the presence of H 2 O in the experiments.Surface Sr(OH) 2 has been found to decompose to SrO at 300-400 • C [23], and thus cannot be ruled out as an impurity since its removal overlaps with the observed changes and might be masked by the mass gains from SrFeO 3 reoxidation.The XRD (see Fig. S10 in the SI) detected only Ag and SrFeO 3 in both retrieved samples, indicating that the impurities were likely limited to the surface of the Ag/SrFeO 3 composite, or were amorphous in nature.
Scanning transmission electron microscopy (STEM) images and EDS maps (Fig. 6) were taken of fresh Ag/SrFeO 3 and spent samples of Ag/ SrFeO 3 , retrieved from the experiments within a single reactor.Images of spent samples after experiments with and without the wateradsorbing layer both contained regions (circled in red or white) rich in Sr and devoid of Fe.Based on lattice spacings identified with high resolution TEM, the Sr-segregation is likely connected to the presence of SrCO 3 or Sr(OH) 2 -see Fig. S11-S12 in the SI.No Sr-rich regions were found when imaging the fresh sample or spent Ag/SrFeO 3 from experiments when only the CLE reaction was carried out.Besides the Sr separation, the fresh and spent Ag/SrFeO 3 samples were otherwise similar in morphology.The size of the Ag nanoparticles was determined from SEM and images and is shown in Fig. S13 in the SI.The Ag particle size remain the same, regardless of experiments, with mean diameters of ~110 nmconsistent with the Ag crystallite size determined by refinement of the X-ray diffraction patterns.

Discussion
The combined dehydration-epoxidation process using two reactors (Fig. 2c) successfully produced EO from ethanol.The performance of the combined process was almost identical to CLE operated with ethylene from a cylinder.In fact, CLE over Ag/SrFeO 3 was not only able to tolerate the presence of dehydration side products at low concentrations (Fig. S2 in the SI), but also percentage-levels of contaminants without deactivation, as revealed in Fig. 3b.We thus demonstrate that ethylene from ethanol dehydration is a suitable feed to CLE to produce EO.In the dehydration-epoxidation process operated in two separate reactors (Fig. 2), the production of EO was >0.45 vol% across multiple cycles of CLE with a selectivity to EO oscillating ~55 % and conversion of ethylene of ~15 % -these results surpass the outputs from conventional epoxidation on unpromoted Ag/α-Al 2 O 3 [24].A comparison of performance parameters from CL-epoxidation with results for direct epoxidation from published studies is provided in Fig. S14 in the SI.Industrially, direct epoxidation reactors are typically operated with gas hourly space velocities (GHSVs) of 2000-7000 h − 1 , an outlet concentration of EO of 1-4 vol%, and with a selectivity to EO of 80-90 % [2,25].However, the silver catalysts used in industry are heavily modified with  b) The average yield of EO during the same CLE cycles.The range of GHSVs was achieved by varying the feed flow rate from 200 to 800 mL min − 1 , and the mass of Ag/SrFeO 3 from 2 to 8 g.The density of Ag/SrFeO 3 is 1590 kg m − 3 .Typical selectivities to EO, conversions of ethylene, and GHSVs, at which EO is industrially produced, are also indicated -specifically for processes which use purified oxygen rather than air as a feedstock [2,25].
surface promotors to elevate the selectivity to EO from 50 to ~90 % [8], whereas the Ag catalyst used here was not.Conceivably, modification of the Ag/SrFeO 3 with surface promotors could result in similar improvements in CLE.Reaching the industrial outlet concentrations of 1-4 vol% would then be readily achievable by increasing the partial pressure of ethylene and decreasing the space velocity.The conversion of ethylene and selectivity to EO over a range of GHSVs are presented in Fig. 7.The yield of EO improved with lower GHSV, a result of the conversion of ethylene increasing.However, operating at a low GHSV was heavily penalised by a drop in selectivity to EO, indicating the need for surface promoters that supress combustion [9,26].Nevertheless, the combined dehydration-epoxidation process compares favourably with other experimental epoxidation results over unpromoted Ag/SrFeO 3 [24], and could foreseeably match the selectivity to EO and concentrations of EO seen in industrial epoxidation reactions.
The operation of the dehydration-epoxidation process offers several advantages over the incumbent method for EO production.One such advantage is a drastic reduction in the time required for start-up.The start-up of the incumbent method for EO production takes hours to days [27].The initial period of operation is accompanied by excessive oxidation of ethylene, giving low selectivity to EO and large heat release, which can be challenging to control.One method used to overcome the excessive overoxidation in start-ups is to saturate the silver catalyst with Cl-based reaction moderators in a catalyst pre-processing step, adding 16-200 h to the overhead time [27].Another approach is to follow a startup strategy, where the ethylene and O 2 flowrates are gradually increased as the selectivity to EO improves.In contrast to both, the combined dehydration-CLE-epoxidation concept proposed here produced EO at a high selectivity immediately after the feed was contacted with the chemical looping catalysts.Additionally, as shown in Fig. 3a, EO can be produced periodically without compromising selectivity to EO or conversion of ethylene, even with frequent and multiple-day pauses in operation.These results are especially meaningful, because many renewable sources of energy are non-dispatchable [28,29], thus, chemical manufacturing that can rapidly start, cease or adjust production can be easily matched to operate when surplus renewable electricity is available [30].Here, electrical consumption would mainly result from process heating.The discussed reactions are both endo-and exothermic (dehydration of C 2 H 5 OH is endothermic, similarly to the selective oxidation of C 2 H 4, while oxidation of SrFeO 3 and combustion of EO and ethylene over Ag/SrFeO 3 are exothermic, see Section S7 of the SI), so the net-effect will depend on the process parameters (selectivity, conversion) and the timings of CL-steps.Nevertheless, the final design will most likely require preheating.Electrification of the process industry is a substantial challenge, thus, even process heating with low-carbon electricity can offer notable improvements over fuel-based heating methods [31].For large-scale industry, electrification with carbon-free electricity means new capital costs and additional operating expenses, Fig. 8. Proposed scheme for the pseudo-continuous production of EO via parallel operation of CLE reactors, with the ability to adjust EO production by switching individual reactors on or off.The ratio of reactors undergoing epoxidation to reactors undergoing regeneration shown here is arbitrary; in reality the number of required reactors will be defined by the ratio of the reduction time to oxidation time in CLE [13].
while the variability of renewable electricity hinders any direct adoption [32].In contrast, our small-scale, on-demand dehydration-epoxidation process is well suited to adopt electrical heating given its ability to operate on demand and suitability to small-scale implementation.
The proposed CLE process offers benefits typically associated with small-scale and modular systems, such as: 1) improved flexibility in production capacity and ability to match market demand; 2) smaller capital investment, andconsequentlyreduced financial risk; and 3) elimination of high EO transportation and storage costs by allowing ondemand production [4].Thus, a careful techno-economic assessment, accounting for storage, transport and flexibility of operation, would be required to compare CLE with incumbent petrochemistry-based EO production, which, in contrast, offers benefits associated with economies of scale (multi-process integration, inexpensive feedstock).
While here, demonstrated only in a lab-scale reactor and giving cyclical EO outputs, the production of EO via CLE could be readily changed into pseudo-continuous, through the addition of multiple CLE reactors operating in parallel [13].We present such concept in Fig. 8.Such a modular system could adjust EO production by switching individual reactors on or off, offering multifaceted flexibility.The new ethanol to EO route proposed here can be used for periodic and 'ondemand' production were EO is only required intermittentlyalso in contrast to the incumbent petrochemical methods, where production continues even if the need for products ceases [33].On-demand capability could be useful in lean manufacturing within the pharmaceutical industry for the synthesis of polyethylene glycol (PEG), and its functionalised versions.Polyethylene glycol is used for the PEGylation of therapeutics [5], usually carried out as a batch process, requiring PEG feedstock only intermittently.The use of EO for fumigation or sterilisation of medical devices [6] gives further application where EO is only needed temporarily.Using CLE for producing EO on-demand for those applications eliminates the need to store and transport the highly hazardous EO.
One aspect of switching from petrochemicals to bio-ethanol is the increased cost of the feedstock [34].Fermentation of biomass to produce ethanol is accompanied by side products (e.g.methanol, aldehydes, and carboxylic acids [35]), and their removal translates to higher prices of ethanol.Additionally, denaturants such as methanol, IPA and petroleum ethers can be deliberately added to ethanol to prevent the incursion of many taxes and duties that are applied to ethanol that is fit for human consumption.Common impurities in commercially available bioethanol are given in Table S1 in the SI.Through the experiments presented in Fig. 3b, we show that our combined dehydration-epoxidation process was insensitive to the presence of percentage levels of methanol and IPA as denaturants.The dehydration catalysts still completely dehydrated the ethanol feed to ethylene, at the same time also dehydrating the methanol and IPA.Similarly, the Ag/SrFeO 3 produced EO without any signs of degradation of performance.Taking denatured ethanol as a proxy for unrefined bioethanol, we conclude that bioethanol directly from fermentation could potentially be used a feedstock.Thus, the new ethanol to EO route offers an attractive opportunity to use low-grade and low-cost ethanol, avoiding competition for highpurity grade bio-ethanol.
In the goal of improving the process attractiveness, we contained the ethanol to EO route within a single reactor vesselwith results shown in Fig. 4.Such arrangement offers several advantages over the configuration with two packed beds, namely fewer process units and control systems, decreased capital costs, and simpler point-of-use operation.However, the single reactor vessel has an extended period of altering selectivity to EO, and in the case without the removal of water, the selectivity to EO is significantly reduced.
The presence of water at ~5 mol% had little effect on the conversion of ethylene, which was unchanged from the experiment with a layer of A3 MS dryer.At the same time, water clearly enhanced the unselective overoxidation reactions.Water can adsorb dissociatively onto both Ag surfaces at 30-330 • C [19] and SrFeO 3 [36], potentially affecting the final selectivity to EO because: 1) water ions reduce the availability of oxygen species on the Ag surface [37] and promote the overoxidation of ethylene.Such mechanisms require further investigations, because, in the referenced work, the oxygen species were also created by dissociation (from O 2 ), so the two gases competed for the Ag surface.2) Alternatively, the dissociatively adsorbed water altered the acid/base characteristics of SrFeO 3 , with the adsorbed OH -potentially providing Bronsted acidity [38] and promoting the oligomerisation of EO to acetaldehyde and further oxidation to CO 2 [39].The insensitivity of conversion to the presence of water suggests that modifications to the acid/base properties is the more likely mechanism, as oxidising already formed EO would not affect the overall conversion of ethylene.
The extended period of altering selectivity to EO observed in the experiments within a single reactor (Fig. 4) did not occur when the dehydration-epoxidation process was carried out in two reactor vessels (Fig. 2).Thus, the cause must lie in the difference between the two processes, namely operating the dehydration reaction in a CL manner.The product distribution from dehydration of ethanol over HZSM-5 was the same, regardless of steady-state or CL-type cyclic operations (Fig. S3 in the SI).Thus, the improving selectivity in the epoxidation step was unlikely a result of changes in the ethanol dehydration process.However, the HZSM-5 catalyst accumulated coke and other impurities during dehydration (Fig. S1 in the SI), a portion of which oxidised during the CL regeneration periods, and, consequently, a mixture of air, water, and CO 2 gets passed over the Ag/SrFeO 3 during regenerationwhich could explain the Sr(OH) 2 or SrCO 3 regions (Fig. 6) in the spent Ag/SrFeO 3 .Since these regions are not observed in the samples from the two-reactor setup, either i) the Sr-rich regions form as a result of the CO 2 and H 2 O being present during regeneration, or ii) the Sr-rich regions form during epoxidation but are subsequently removed during regeneration in CO 2free and dry air (21 vol% O 2 /N 2 ).Indeed, layers of Sr(OH) 2 and SrCO 3 have been reported to form on and passivate the surfaces of Srcontaining perovskites [40].The SrFeO 3 support is not active for epoxidation [10], but does combust ethylene and EO [41].The increasing passivation of the SrFeO 3 surface should lead to a gradual improvement to selectivity to EO, decrease in the conversion of ethylene, and reduction in the amount of oxygen releasedin agreement with the experimental results in Fig. 4. The implication of these results to CLE is that passivation of the SrFeO 3 surface, either through Srimpurities or via controlled modification, could be another effective strategy to minimise complete combustion and improve selectivity to EO.

Conclusions
A novel method was successfully demonstrated to produce EO from ethanol by coupling the dehydration of ethanol and chemical looping epoxidation of ethylene.The combined dehydration-epoxidation process is well suited for small-scale and flexible operation, ideal for coupling with renewable power and for producing EO on demand.
The most promising performance from the two-reactor configuration gave EO at a selectivity of 57 % and ethylene conversion of 15 %, exceeding what is achievable over unpromoted Ag/α-Al 2 O 3 in a traditional epoxidation approach, exceeding the selectivity achieved over unpromoted Ag/α-Al 2 O 3 , while matching the yield of EO to that reported for modified Ag-based catalysts.
We also demonstrated that the Ag/SrFeO 3 catalyst robustly tolerates percentage levels of water and impurities in the ethanol, revealing that denatured or unrefined bioethanol would be a viable feedstock for lowcost EO production.High resolution TEM and STEM-EDS revealed that Sr-rich regions, likely SrCO 3 or Sr(OH) 2 , appeared on the SrFeO 3 surface during CLE epoxidation experiments carried out in the presence of H 2 O; however, their formation appeared to be to some degree beneficial, inhibiting the unwanted over-oxidation of EO, suggesting surface passivation of SrFeO 3 as a strategy to enhance selectivity to EO.
Since ethanol can be produced from biomass, the new route developed here, opens up an opportunity for the sustainable, low-cost, and on-demand production of bio-ethylene oxide with performance competitive to the incumbent method of EO production.

Fig. 1 .
Fig. 1.Schematic of two experimental arrangements used in this study.(a) The two-reactor setup, with separate reactors for the dehydration and epoxidation reactions.(b) The single dehydration-epoxidation reactor using a multi-layered approach.

Fig. 2 .
Fig. 2. In the middle: schematics of two reactors (Reactor A and B) working separately or in tandem.(a) Reactor A: dehydration of 5.5 vol% ethanol over γ-Al 2 O 3 ; (b) Reactor B: concentrations of EO and CO 2 during CLE of 5.4 vol% ethylene provided from a cylinder, (c) Reactor A and B operated in series with the outlet of A providing ethylene to B, (d) comparison of the selectivity to EO and conversion of ethylene between operating Reactor B alone and operating Reactor A and B in series.

Fig. 3 .
Fig. 3.The selectivity to EO, and conversion of ethylene during experiments with the combined dehydration-epoxidation process: (a) operated once a week for 3 weeks with the Ag/SrFeO 3 and γ-Al 2 O 3 materials kept in the reactors at ambient temperature and under static air between experiments.(b) with a feed containing 89.1 mol% ethanol, 7.1 mol% methanol, and 3.8 mol% isopropyl alcohol in N 2 , using either HZSM-5 or γ-Al 2 O 3 as the dehydration catalyst, and Ag/SrFeO 3 as the epoxidation catalyst.

Fig. 4 .
Fig. 4. Selectivity to EO (left axis), conversion of ethylene (left axis) and amount of EO produced (right axis) during each CLE cycle in a single dehydrationepoxidation reactor: (a) without the inclusion of an A3 drying layer, and (b) with the inclusion of an A3 drying layer, (c) comparison of oxygen released from SrFeO 3 in experiments from (a) and (b).

Fig. 5 .
Fig. 5. Relative mass change of samples of spent Ag/SrFeO 3 that were used in single-reactor experiments a) without an intermediate drying layer, and b) with an intermediate layer of A3 MS.

Fig. 6 .Fig. 7 .
Fig. 6.HAADF and EDS maps of samples of a) fresh Ag/SrFeO 3 , b) and c) spent Ag/SrFeO 3 from experiments in a single reactor: b) with a layer of A3 MS, and c) without a layer of A3 MS.Circles indicate Sr-rich and Fe-deficient areas of the samples.The samples were cleaned at 270 • C in air for 5 h prior to imaging.