Experimental Validation is Always Required for Molten Oxide Electrolysis Laboratory Crucibles

Abstract

Molten oxide electrolysis (MOE) is a promising electrochemical route to de-carbonize primary and secondary metal production. Developing MOE processes starts with laboratory experiments at temperatures > $1000\,^\circ{\rm C}$ lasting ~10 hours and requiring long heating/cooling times to protect furnace hardware. Before investigating MOE processes, crucibles must be selected that tolerate the required temperatures while minimizing chemical interactions with the oxide to control melt contamination and contain the melt. Unfortunately no general procedure guiding MOE crucible selection is documented. Here we focus on laboratory crucibles in air for two MOE melts: titania-sodia at $1100\,^\circ{\rm C}$ and neodymia-boria at $1300\,^\circ{\rm C}$. After shortlisting generic crucible materials using Ashby’s method, thermodynamic predictions were made for all-oxide titania-sodia charges using FactSage and cup test experiments were conducted on (i) all-oxide and carbonate charges for titania-sodia and (ii) neodymia-boria charges. While magnesia was predicted to be the best crucible for the titania-sodia melt, alumina was the best choice for both oxide and carbonate charges. The grain boundary networks of both magnesia and YSZ were infiltrated by the oxide and carbonate charges. Platinum was the best crucible for neodymia-boria melts. We show that compositional control during long, high-temperature MOE experiments requires experimental validation for specific chemistries every time.

Introduction

Electrochemical processes in molten oxides have the potential to reduce environmental impact, cost and social inequities of metal production while increasing supply of critical metals. This is an important pathway toward achieving the United Nations Sustainable Development Goal 12^[1] relating to responsible production. While there are challenges aplenty in developing the electrochemistry as discussed by Allanore,^[2] one of the most pressing hurdles for laboratory experimental work is finding a crucible to contain molten oxide electrolysis (MOE) at high temperatures for times on the order of hours. At elevated temperatures, reaction kinetics are accelerated leading to the adage that “everything reacts with everything.”

Copious literature exists on selecting refractories for industrial scale applications like steelmaking,^[3,4,5,6,7] glassmaking,^[8,9] and nuclear waste treatment,^[10] where theoretical and experimental methods are required. Where comprehensive thermochemical and physical properties are available, e.g., for common slag chemistries^[11,12] or glass compositions,^[8] shortlisting for experimental evaluation is possible. Phase diagrams and phase relations, where available, can be used to predict the interactions of container and charge,^[4,5] but predictions must be validated by experiments.

In contrast to steelmaking literature above, the literature on crucible or refractory reactivity with specific molten oxide or molten salt systems is more limited. White et al.^[13] studied the reactions between molten CaO-SiO₂ and different grades of graphite as refractory material for reducing silica at 1550 °C to 1600 °C under argon. Zhang et al.^[14] studied aluminothermic reduction to form Ti–Si–Al alloys from slags at 1550 °C under argon, analysing the corrosion behavior of carbon, Al₂O₃ and MgO crucibles. Other studies include testing regolith-ceramic interfaces for molten materials containment on the lunar surface by Standish et al.,^[15] evaluating materials compatibility for solar thermal energy systems with molten alkali carbonate salts and solids at 600 °C by Coyle et al.,^[16] characterizing corrosion behavior of stainless steel 316 for thermal energy storage of eutectic molten salts at 450 and 700 °C by Sarvghad et al.^[17] and investigating Ni for containing molten Li-Li₂O-LiCl at 650 °C for electrolytic reduction by Choi and Lee.^[18] The Molten Salt Handbook by Janz^[19] lists containment materials for various molten oxides and molten salts, but it is limited to well-studied or published materials up to 1967.

There are a wealth of books with guidance on suitable crucibles for high temperature experiments from the twentieth century. In the introduction to their 1959 book, Bokris et al.^[20] note that at temperatures above $1600\,^\circ{\rm C}$, reaction between refractories and the system of study is inevitable and requiring slow enough rates of reaction to leave the property of interest unchanged is the relevant demand, and give, in Appendix I, a table of containers used for melts and atmospheres reported in the literature. In Chapter 4, Livey and Murray^[21] caution that while thermodynamics can be used for indicating potential refractories, “it may be necessary to resort to experiment to determine which material is most suitable for a given application.”, and that using free energies of formation to predict reactivity is naive when solid or liquid solutions form. Books, such as 1979’s Kubaschewski and Alcock,^[22] on metallurgical thermochemistry also contain relevant advice on methods for selecting refractories for laboratory experiments. In his 2012 book “High Temperature Experiments in Chemistry and Materials Science”,^[23] Motzfeldt bemoans the futility of surveying crucible—melt compatibility due to the vast number of combinations and writes that “special problems require special solutions.” Svantesson et al.^[24] note that, while it is common practice to use metal crucibles for oxide melts and vice versa, chemical compatibility must also be maintained.

Chemical wear/corrosion is characterized using static (e.g., sessile drop, cup/crucible test, static finger/immersion/dipping test) and dynamic (e.g., rotating figure test, induction furnace test, rotary slag test) methods at service temperatures in the atmosphere of interest depending on the application.^[7] The cup test is commonly used because it is easy.^[25] Standard methods have been developed; e.g. the German CEN/TS 15418:2006 standard^[26] details cup, finger-dip, rotary slag and induction crucible tests for corrosion of dense refractory products by liquids including metals, fluxes and slags; and the ASTM C621 standard^[9] details isothermal, static finger tests for corrosion resistance of refractories to molten glass for glassmaking, where corrosion depth at two locations is reported.

Ashby’s text “Materials Selection in Mechanical Design” describes a solution-neutral method of selecting materials based on properties and physics.^[27] While Ashby’s general methodology for materials selection has been successfully applied in applications such as various aeroturbines^[28] and implemented with databases of property information in the ANSYS Granta software,^[29] the lack of specific chemical reactivity is a severe limitation for crucible selection.

Nonetheless, we employ Ashby’s paradigm here. Ashby’s materials selection methodology follows four stages: translation, screening, ranking and obtaining further information.^[27] The most important stage is translation where the function, objectives, constraints and free variables are identified. We translate the selection of a laboratory crucible for MOE in Table I.

Table I Translation of MOE Crucible Material Selection Using Ashby’s Framework

While the ANSYS software contains an excellent cross-section of materials with tabulated properties, the only reactivity information is a qualitative measure of “resistance to oxidation at $500\,^\circ{\rm C}$”. Selecting crucibles for a specific oxide mixture (charge) requires more specific reactivity data.

Phase diagrams or CALPHAD thermodynamic software, such as FactSage^[30] or Thermo-calc,^[31] provide measures of reactivity from equilibrium predictions.^[4] Rates of reaction must be determined experimentally.^[7] For systems without established phase diagrams or without parameters in the CALPHAD tools, only experimental methods are available.

In this work, we distinguish between the charge, the species added to the crucible corresponding to the target mixed-oxide for MOE, and the product, the contents of the crucible after the charge has been heated to the target MOE temperature but before electrolysis. There are four outcomes of crucible-oxide charge interactions in air that make a crucible unsuitable for a specific mixed oxide, Figure 1.

• 1. Crucible elements can react with the atmosphere to form solid, liquid or vapor.

• 2. Crucible elements can dissolve into the oxide melt leading to contamination of the product and thinning of the crucible, termed coherent/direct/homogeneous dissolution.^[4]

• 3. Elements from the oxide mixture can selectively dissolve into the crucible leading to compositional uncertainty of the product and stresses in the crucible.

• 4. The crucible and charge can react to form new solid or liquid phases leading to compositional uncertainty of the product and stresses in the crucible.

Fig. 1

Possible limiting crucible-charge interactions corresponding to mass changes in the melt and crucible at the temperature of interest. Combinations of these interactions will alter the coordinates of the percentage mass change on the diagram. Each of the interactions in quadrants 2 to 4 alters the composition of the product, and its extent should be minimized for MOE. Some reactions with the vapor were not illustrated to simplify the diagram

The first three mechanisms will change the total mass and composition of the charge. Changes in the composition of the oxide mixture must be avoided or minimized. The mechanical integrity of the crucible can degrade with any of the four reactions, limiting the lifetime of the crucible. If a solid reaction product forms at the crucible-charge interface, termed incoherent/indirect/heterogeneous dissolution,^[4] and has a kinetically-limited thickness, it can act to protect the charge from the crucible.

In this paper, we demonstrate the challenges of selecting crucibles for high temperature laboratory experiments on molten oxide mixtures. Starting from a shortlist of generic crucible materials, we illustrate crucible selection for two specific target oxide mixtures: titania-sodia and neodymia-boria. The first is chosen because FactSage databases allow reactivity predictions with candidate crucibles, and we have validated FactSage predictions for these molten-oxides in various atmospheres in previous work.^[32,33] The second is chosen because it is technologically relevant for the urgent production of critical rare earth elements for the transition to zero-carbon, and it illustrates the challenges of crucible selection without guiding thermodynamic predictions. The compositions and temperatures of interest in the two systems are indicated with red circles on representative phase diagrams,^[34] Figure 2. MOE processes are designed in single phase liquid states, which correspond to $1100\,^\circ{\rm C}$ for 54 mol pct titania-sodia and $1300\,^\circ{\rm C}$ for 62 mol pct neodymia-boria. Cup test^[7] experiments are used to determine crucible-charge interactions and to compare to predictions.

Fig. 2

Phase diagrams corresponding to the systems of interest with (a) calculated sodia-titania phase diagram in oxygen from FactSage and (b) boria-neodymia phase diagram adapted from Ji et al.^[14] Target compositions of interest, 54 mol pct titania-sodia and 62 mol pct neodymia-boria, are marked with red circles (Color figure online)

Methods

Short-Listing Generic Crucibles

Materials selection was implemented in ANSYS Granta software using the Level 3 Aerospace database. Starting from all materials, only classes that were fully-dense were enabled (Ceramics and glasses; Hybrids/Composites; Metals and alloys). Screening de-selected materials that could only be formed into sheet products and did not have excellent resistance to “oxidation at 500C”. Maximum service temperature, ${T}_{{\text{max}}}$, was used for screening at $1100\,^\circ{\rm C}$ and $1300\,^\circ{\rm C}$. Kingery’s thermal shock resistance factor, ${R}^{\prime}=\frac{k{\sigma }_{y}(1-\nu )}{E\alpha }$, was used for ranking, where $k$ is thermal conductivity, ${\sigma }_{y}$ is the elastic limit in tension, $\nu$ is Poisson’s ratio, $E$ is Young’s modulus and $\alpha$ is the linear coefficient of thermal expansion.

Thermodynamic Predictions

The FactSage 8.1 FToxid and FactPS software packages^[30] were used to make equilibrium predictions on the Ti–Na–O charge interaction with Al₂O₃, MgO, ZrO₂ (8 mol pct Y₂O₃), BN and Ni crucibles. Closed system predictions were performed with 60 g TiO₂, 40 g Na₂O, 100 g of the crucible material, 11.6 g of N₂ and 3.5 g of O₂. The N₂ and O₂ correspond to the volume of gas in a 1 m long, 90 mm internal diameter furnace tube for dry air (21 pct O₂, 79 pct N₂) from previous electrolysis experiments.^[32] The temperature and pressure were 1100 °C and 1 bar, respectively, with the “gas-real” and “pure solids” species selected. Pure liquid species were not selected; instead, all Ti and Na species from the available solution phases were selected. The Equilib module was used to predict the equilibrium phases, phase fractions and compositions.

Experimental Reactivity

The target 54 mol pct titania-sodia mixed oxide for cup tests was obtained by two routes from the same oxide-carbonate precursor blend: in situ reaction^[35] of charge in candidate crucibles and pre-reacted/all-oxide charge. The TiO₂ (≥ 99 pct, Sigma Aldrich) and Na₂CO₃ (99.5 pct, May & Baker) powders were dried at 105 °C for 16 hours then blended and crushed with a mortar and pestle.

To obtain an all-oxide charge, 19.2 g of precursor was fired for 10 hours at $1100\,^\circ{\rm C}$ in a nickel crucible in an inert atmosphere in a vertical tube furnace. The crucible was suspended in a secondary graphite crucible in the hot-zone of the furnace using a molybdenum support structure. The furnace was purged using zero grade argon gas (99.9999 pct, BOC) at 500 mL min⁻¹. Once the oxygen content dropped below 500 ppm O₂, measured by a Thermox CG-1000 Oxygen Analyser, the flow rate was reduced to 150 mL min⁻¹. The furnace was heated at a rate of 1.25 °C min⁻¹ with temperature measured by a sheathed Type B thermocouple positioned with the tip alongside the crucible in the furnace. Once pre-fired, the mixture was cooled, crushed using a mortar and pestle, and stored in a desiccator.

Three candidate crucibles of size CC35 (27 mL capacity, 42 mm outer diameter, 33 mm height) supplied by Almath Crucibles Ltd were tested: Al₂O₃ (recrystallized alumina (99.8 pct)), YSZ (ZrO₂-8 mol pct Y₂O₃) and MgO. Charges of either 4.8 g of all-oxide charge or 6.4 g of carbonate precursor blend were fired in each candidate crucible in air inside a Lindburg box furnace with MoSi₂ elements at a rate of 10 °C min⁻¹ to $1100\,^\circ{\rm C}$ and held at temperature for 4 hours. Crucibles were cooled to room temperature at a rate of 10 °C min⁻¹ and desiccated before cutting and analysis.

The target 62 mol pct neodymia-boria mixed-oxide was obtained by mixing oxide powders: Nd₂O₃ (99.9 pct trace metal basis, Sigma Aldrich) and B₂O₃ (99.98 pct trace metal basis, Sigma Aldrich). X-ray diffraction revealed that the B₂O₃ was all boric acid (H₃BO₃) and 73 pct of the Nd₂O₃ was Nd(OH)₃. Further, Ji et al.^[34] advised using an excess of 2 wt pct B₂O₃ to obtain target compositions, due to its volatility. Moisture content and boria volatility were accounted for in the starting mass ratios. Batches of 6 g mixed-oxide charge were fired in one of three candidate crucibles, Pt (Platinum Gold Rhodium (94/5/1), 40 mL capacity, Standard form crucible—GCS40, Cleveland Process Automation), Al₂O₃ (as above) and MgO (as above), in the Lindburg box furnace in air at $1300\,^\circ{\rm C}$ for 4 hours. Heating rates were 10 °C min⁻¹ to $1100\,^\circ{\rm C}$ and 5 °C min⁻¹ to $1300\,^\circ{\rm C}$, and the same schedule was used for cooling to room temperature.

Cross-sections from each experiment were cut using a Buehler IsoMet 11-1180 low-speed diamond saw. Optical images were captured with a Zeiss Axio Zoom. V 16. A JEOL JSM-IT300LV scanning electron microscope (SEM) with LaB6 filament equipped with an X-Maxä 50 silicon drift energy dispersive spectroscopy (EDS) detector was used for SEM imaging and EDS on cross-sections. For neodymia-boria, cooled product material was removed from each crucible and crushed using a mortar and pestle, then desiccated before analysis by x-ray fluorescence (XRF). Cellulose (Sigmacell Type 20, approximately 4 g) was lightly pressed in a 40 mm die before 200 mg of the sample was sprinkled over the center using a 3D-printed distribution funnel. The cellulose with sample was then pressed to approximately 350 bar to make a solid pellet. The sample was analysed with a Rigaku ZSX Primus IV XRF in conjunction with the ZSX guidance software over elemental range B-U.

Results and Discussion

Generic Crucible Selection for Molten Oxide Electrolysis

Implementing materials selection excluding “by excessive reaction with oxide mixture” in the Level 3 Aerospace database, reduces the initial 4181 materials down to 358, where maximum service temperature, ${T}_{{\text{max}}}>1100\,^\circ{\rm C}$ is illustrated with a thick black line in Figure 3(a) and some key material classes are labelled. There are non-technical ceramics, technical ceramics, ceramic matrix composites and alloys shown in family envelopes. Unsurprisingly, the technical ceramics have the best combination of high resistance to thermal shock ${R}^{\prime}$ and high ${T}_{{\text{max}}}$, upper right corner. Diamond, beryllia, thoria, silicon carbide, magnesia (MgO) and boron nitride are good choices. Figure 3(b) shows the materials from (a) on electrical conductivity-price axes. The alloys and ceramic matrix composites are electrical conductors, opening the possibility for the crucible to function as the counter-electrode in the electrochemical cell. The cost of diamond and sapphire is too high for laboratory use in this application. Further, materials containing silicon (SiC) are unsuitable because silicon contamination of molten oxide melts decreases ionic mobilities and silicon can have a lower reduction potential than target metals for MOE, both of which are undesirable.^[36,37] Thoria^[38] and beryllia^[39] are unsuitable due to toxicity. Tungsten and cobalt alloys are unsuitable due to oxidation in air at high temperatures^[40] and high cost.

Fig. 3

Materials selection for generic crucible materials for high temperature MOE after screening out materials with maximum service temperature below $1100\,^\circ{\rm C}$. (a) candidate materials in thermal shock resistance—maximum service temperature space, (b) candidate materials from (a) in electrical conductivity—price space and (c) short list of candidate materials after additional screening (see text) with maximum service temperatures of $1100\,^\circ{\rm C}$ and $1300\,^\circ{\rm C}$ indicated after (a). Images used courtesy of ANSYS, Inc. ^[29]

While stainless steels and commercially pure nickel are commonly supplied for crucibles,^[41] they are deselected in the software due to ${T}_{{\text{max}}}$. Here, maximum service temperature is a nebulous parameter and can be limited by degradation of mechanical properties such as yield strength or Young’s modulus, phase transformations, breakdown of protective oxides, creep deformation or other unenumerated behaviors. Clay-graphite crucibles are common in the laboratory, but would be unsuitable here due to Si in the clay and the oxidation of C in air.

Yttria-stabilized zirconia (YSZ), boron nitride, magnesia, alumina (Al₂O₃) and Ni were selected for further study with ${T}_{max}>1100\,^\circ{\rm C}$. Ni was selected as a representative metallic crucible, in part, because interpreting thermodynamic predictions for a pure metal is easier than for an alloy like 316 stainless steel.

The remaining materials are visualized with maximum service temperature-resistance to thermal shock trade-off in Figure 3(c)). The effect of increasing the maximum service temperature to ${T}_{{\text{max}}}>1300\,^\circ{\rm C}$ is illustrated with the dashed grey line, i.e., it further restricts crucible materials to the technical ceramics and excludes YSZ.

Crucibles for Titania-Sodia Molten Oxide Electrolysis

Thermodynamic prediction of oxide charge: crucible reactivity

Thermodynamic predictions of reactivity for the target 54 mol pct titania-sodia charge in boron nitride, magnesia, alumina, YSZ and nickel are shown with contamination in Figure 4 and with phase fractions in Figure 5. Contamination of both the charge and of the crucible should be avoided and the best candidates should lie in the lower left corner of Figure 4(a)). Contaminants in the crucible material can be Ti, Na, N or O from oxide melt or air. Because Ti and Na can react in any proportion with the crucible, their mole fractions in the molten oxide phase at equilibrium are shown in Figure 4(b)). The line indicates a constant molar ratio of 54:46 titania:sodia as the charge is diluted by contaminants from the crucible or from the air. Compositions lying above the line are sodia-rich and compositions lying below the line are titania-rich compared to the target.

Fig. 4

Predicted reactivity in air at $1100\,^\circ{\rm C}$^[29] for 54 mol pct titania-sodia mixed oxide in candidate crucible materials with (a) contamination of charge and crucible and (b) compositional changes in the charge, where the line indicates fixed 54:46 titania:sodia ratio

Fig. 5

Predicted equilibrium phases and their compositions for 54 mol pct titania-sodia mixed oxide at 1100$\,^\circ{\rm C}$ in air from initial gas, molten oxide and crucible masses shown in (a) for five different crucibles materials: (b) Al₂O₃, (c) YSZ, (d) Mg0, (e) BN and (f) Ni

Alumina (square symbol in Figure 4) is predicted to completely react with sodia to form new solid phases, NaAl₉O₁₄ and Na₂Al₁₂O₁₉, that consume the crucible in totality, Figure 5(b)), corresponding to quadrant 3 in Figure 1, and is shown as 100 pct contamination on Figure 4(a)). The YSZ crucible (diamond symbol) is predicted to lose 40 pct of its mass by dissolving into the charge leaving an yttria-enriched (23 wt pct compared to 14 wt pct Y₂O₃) crucible, Figure 5(c)), corresponding to quadrant 4 in Figure 1. Both Y and Zr contaminate the oxide charge, but the titania:sodia ratio is predicted to be unaffected, Figure 4(b)). Magnesia (triangle) is predicted to be virtually inert with this oxide charge at $1100\,^\circ{\rm C}$, Figures 4(a)) and 5(d)), and corresponds to a coordinate near the origin in Figure 1. Boron nitride (circle) is predicted to consume all the oxygen in the air and to react with Ti in the oxide to form solid TiN, Figure 5(e)), and corresponds to quadrant 3 in Figure 1. Ni (cross) is predicted to form NiO by consuming all oxygen in the air, Figure 5(f)), but to leave the charge virtually unchanged, Figure 4, corresponding to region 1 in Figure 1. In summary, we predict that boron nitride and nickel are poor choices in air at $1100\,^\circ{\rm C}$, that the alumina crucible will react with the charge but may undergo self-limiting heterogeneous dissolution and that the YSZ crucible will undergo homogeneous dissolution. We predict that magnesia is the best crucible for 54 mol pct titania-sodia melts in air at $1100\,^\circ{\rm C}$.

Experimental determination of oxide charge: crucible reactivity

The FactSage predictions are thermodynamic and do not account for the kinetics of reactions between crucibles, atmosphere and charge. Cup tests were performed for 4 hours at $1100\,^\circ{\rm C}$ to evaluate interactions of the oxide candidate crucibles: alumina, YSZ and magnesia. Optical images of the cross-section of the crucible-charge interface near the corner (where the wall meets the base) are shown in Figure 6, with the red box on the left-hand images shown at higher magnification in the corresponding right-hand image. The alumina and magnesia crucibles develop a boundary layer at the crucible-product interface, Figures 6(a)) and 6(c)), respectively, and all along the base. In magnesia, product was only found near the corner of the base and no product remained in the center of the base. Voids formed near the inner surface of the YSZ crucible, Figure 6(b)), but no boundary layer is evident in optical images.

Fig. 6

Crucible-charge interaction regions for 54 mol pct titania-sodia mixed oxide charge fired in air at $1100\,^\circ{\rm C}$ for 4 hours for (a) alumina (b) YSZ and (c) magnesia crucibles with right-hand inset corresponding to box on left-hand image

Corner regions were characterized by SEM and EDS elemental maps and line-scans were collected, Figures 7 through 9. The line scans are reported on a metal basis so that the target 54 mol pct titania-sodia would appear as a 37:63 = Ti:Na metal ratio. The alumina crucible had little contamination, but a boundary layer with uniform metal elemental contrast and 25 mol pct Ti—30 mol pct Al—45 mol pct Na was detected, Figure 7. Either sample preparation or stresses during cooling caused the boundary layer to crack away from the crucible wall. The interface location is marked with a black arrow on Figure 7(c). An average of 35 mol pct Al was found in the first $150 \mu \text{m}$ of the boundary layer/product, while 0.2 mol pct Ti and 0.9 mol pct Na were found in the first $240 \mu \text{m}$ of the crucible. Alumina is widely used as a crucible for glasses, where the main consideration is contamination of the glass causing degradation of optical and structural properties: dos Santos et al.^[42] reported contamination of 2.9 mol pct alumina in heavy metal oxide glasses after 1 hour at $1000\,^\circ{\rm C}$, and Sawangboon et al.^[43] found 10 mol pct alumina in silicophosphate glasses after 12 hours at $1200\,^\circ{\rm C}$. Wiencke et al.^[44] note that in order to use an alumina crucible for iron reduction by MOE, the electrolyte must be saturated in alumina to prevent degradation of the crucible.

Fig. 7

Alumina crucible-charge interaction region for 54 mol pct titania-sodia mixed oxide charge fired in air at $1100\,^\circ{\rm C}$ for 4 hours with (a) secondary electron image indicating line scan in red (b) EDS maps of Al, Na and Ti, and (c) EDS line scan from (a) with black arrow marking edge of crucible (Color figure online)

The YSZ crucible elements Zr and Y were both detected in the product region, with an average of 1.7 mol pct Zr and 0.4 mol pct Y in the first $75 \mu \text{m}$, and Ti and Na were detected around the grain boundary network in YSZ, as evident in their EDS maps, Figure 8. Over the first $235 \mu \text{m}$ of the crucible, an average of 4 mol pct of Ti and 9 mol pct Na were found. Figures 8(b) and (c) show an $\sim 30 \mu \text{m}$ thick Y-enriched and Zr-depleted region developed at the edge of the crucible. The observed mesoscopic Y-enrichement is consistent with our FactSage predictions, Figures 4 and 5. The dissolution of the crucible occurs preferentially at the grain boundaries that have been infiltrated by the molten oxide.

Fig. 8

YSZ crucible-charge interaction region for 54 mol pct titania-sodia mixed oxide charge fired in air at $1100\,^\circ{\rm C}$ for 4 hours with (a) secondary electron image indicating line scan in red (b) EDS maps of Zr, Y, Na and Ti, and (c) EDS line scan from (a) with black arrow marking edge of crucible (Color figure online)

Infiltration, sometimes described as penetration or wetting, of ceramics by oxide liquids/glasses is well documented,^[3,7] even in fully-dense ceramics,^[45] and is a major performance limitation for YSZ environmental barrier coatings in jet engines in siliceous environments.^[46] Rodrigues et al.^[47] reported infiltration of grain boundaries in 8 mol pct YSZ by green soda-lime glass after 2 hours at $1500\,^\circ{\rm C}$. In contrast, to the Y-enrichment at our crucible surface, depletion of cubic ziroconia-stabilizing elements is reported in the literature and causes phase changes and mechanical stresses in ceramics.^[42,47,48] For example, Xu et al.^[48] reported an Y-depletion layer at the outer surfaces of YSZ tubes and radial cracks spanning the bulk of tubes exposed to oxyfluorite flux for 10 hours at $1100\,^\circ{\rm C}$. Their Y-depleted zone contained many voids that they attributed to leaching of Y into the flux and diffusion. A difference in the intrinsic diffusivities of Y and Zr in a chemical potential gradient would lead to Kirkendall voids^[49] that are mostly reported in alloys but have recently been reported by Cao et al.^[50] in alumina-chromia ceramic diffusion couples. In our work, the surface is enriched in Y, which would create the conditions for Kirkendall void formation, but it was out of scope to characterize this behavior.

The Mg from the magnesia crucible contaminated the oxide charge with an average of 2 mol pct Mg detected over the first $115 \mu \text{m}$. Like the YSZ crucibles, the magnesia grain boundary network was infiltrated by Na and Ti, Figure 9. Oxide slag infiltration of magnesia grain boundaries is a failure mechanism for steelmaking refractories: Zhang et al.^[51] showed MgO-CaO-SiO₂ slag chemistry and magnesia grain size could accelerate the rate of infiltration of magnesia by slag for 3 hours at $1600\,^\circ{\rm C}$. Mukai et al.^[52] studied penetration of magnesia refractories with various porosities by steelmaking slag in situ and showed that penetration depth followed parabolic kinetics. On average, 8 mol pct of Ti and 12 mol pct of Na were detected in the first $255 \mu \text{m}$ of the crucible. The boundary layer identified in the optical image, Figure 6(c), likely corresponds to this infiltration near the crucible-charge interface, which is estimated to be $\sim 600 \mu \text{m}$ wide from Figure 6(c). The boundary layer was observed in the center of the base even though no product remained there. In contrast to the FactSage predictions that magnesia would be almost inert under the cup test conditions, infiltration of the grain boundary network was observed. Even with no change in molten oxide chemistry, infiltration would deplete the quantity of charge in the crucible and could cause leaking, both undesirable outcomes.

Fig. 9

Magnesia crucible-charge interaction region for 54 mol pct titania-sodia mixed oxide charge fired in air at $1100\,^\circ{\rm C}$ for 4 hours with (a) secondary electron image indicating line scan in red (b) EDS maps of Mg, Na and Ti, and (c) EDS line scan from (a) with black arrow marking edge of crucible (Color figure online)

This study focusses on selecting commercial crucibles for laboratory scale electrochemical experiments on molten oxides. It was outside the scope of this work to study the observed infiltration kinetics and dependence on specific crucible chemistries or microstructures:^[52,53] porosity, wetting behavior and infiltration metrics were not characterized.

Using the EDS linescans from Figures 7 through 9, the average compositions of the crucible and product regions were determined on an oxide basis. The measured contamination of the crucible and charge and the composition of the charge are summarized in Figures 10(a) and (b), respectively. Contamination of the alumina crucible was minimal, solid square in Figure 10(a), but alumina contaminated the charge (~ 55  mol pct) and pushed its composition to the sodia-rich side of the ideal stoichiometry, Figure 10(b). In contrast, for both the YSZ and magnesia crucibles, solid diamond and solid triangle, respectively, in Figure 10(a), there was contamination of ~ 17 mol pct in the crucibles due to the infiltration of their grain boundary networks by the charge, but limited contamination of the charge itself. In both cases, the charge was pushed to the sodia-rich side of target titania:sodia ratio.

Fig. 10

Comparison of predicted (open symbols) to experimental (solid symbols) reactivity of crucibles with 54 mol pct titania-sodia mixed oxide charges at $1100\,^\circ{\rm C}$ with (a) contamination of crucible and charge and (b) compositional changes in charge. The target composition and fixed 54:46 titania:sodia ratio are indicated on (b)

Comparison of predicted and experimental oxide charge: crucible reactivity

Experimental (solid symbols) contamination and composition of the charge are compared to FactSage equilibrium predictions (open symbols) in Figure 10. Alumina was predicted to preferentially react with Na and shift the charge composition to the titania-rich side, but no experimental evidence for the sodium-titanium oxide phases was found. While sodium was depleted from the charge over the small region that was examined by EDS, the titania content was depleted further to push the titania:sodia ratio to the sodia-rich side. For YSZ and magnesia crucibles, the average crucible contamination was predicted to be minimal, but the charge infiltrated the grain boundary networks so that the average contamination was recorded here as ~ 17 mol pct, Figure 10(a).

The utility of the Factsage predictions is limited because (1) they predict only equilibrium phase relations at the temperature of interest and (2) they were performed for a representative closed system. In contrast, our cup test experiments require slow cooling from the temperature of interest, rather than quenching, to protect furnace components. During slow cooling, further reactions and phase changes can occur, which is why no attempt was made here to identify the phases in the boundary layers or product. The cup tests were performed for 4 hours at $1100\,^\circ{\rm C}$, in line with literature studies,^[6,7] but equilibration times are unknown. In-situ monitoring or a time series of tests to validate phase equilibrium was beyond the scope of this work. The furnace has an uncontrolled air atmosphere so that any volatile elements or compounds may leave the system. Performing thermodynamic predictions with fixed activities rather than fixed amounts of components could address this, but determining relevant reservoirs is a challenge. Because the products were not able to be removed from the crucibles in our work, we report compositions averaged over regions near the crucible-charge interface. Ideally, a bulk (average) chemical analysis would be determined and used for comparison.

Experimental determination of carbonate charge: crucible reactivity

To decrease overall MOE experimental time in our laboratory, we charge the crucible with a mixed oxide-carbonate precursor blend, and calcination proceeds in-situ to form the target all-oxide melt. Thus, we quantified the crucible-melt interactions with a carbonate-bearing charge: TiO₂-Na₂CO₃. It is not possible to predict the oxide-carbonate reactions using FactSage. Optical images of crucible-carbonate charge interface regions for alumina, YSZ and magnesia are shown in Figure 11, with right-hand insets corresponding to the red box in the corresponding left-hand image. A boundary layer was detected on the alumina crucible, and the melt appears foamy, Figure 11(a). (The brown marking is from cutting oil that could not be removed on the top surface.) In YSZ, voids were observed again in the near-charge region of the crucible, but, product was only found around the edge of the base (lower left corner in cross-section) and no product remained in the center of the base. In magnesia, no product remained at all, and a boundary layer was visible along the whole inner crucible wall.

Fig. 11

Crucible-charge interaction regions for 54 mol pct titania-sodia from carbonate charge fired in air at $1100\,^\circ{\rm C}$ for 4 hours for (a) alumina (b) YSZ and (c) magnesia crucibles with right-hand inset corresponding to box on left-hand image

Again, corner regions were characterized by SEM and EDS elemental maps and line-scans were collected, Figures 12 through 15. The alumina crucible contaminated the product and 30 mol pct Al on average was found in the first $100 \mu \text{m}$ from the interface, Figure 12. In the first $110 \mu \text{m}$ of crucible wall, an average of 1 mol pct Ti and 3 mol pct Na was detected, corresponding to the boundary layer observed in Figure 11. This is consistent with literature reports on alumina solubility in molten carbonates. Tathavadkar and Jha^[54] reported on roasting anatase ore with sodium carbonate in alumina crucibles in flowing air for 2 hours at temperatures $800\,^\circ{\rm C} -1000\,^\circ{\rm C} .$ Their product fused to the alumina crucible only at $1000\,^\circ{\rm C}$, which they interpreted as a sign that the charge became fully molten based on a titania-sodia phase diagram. Alumina electrodes are used in molten carbonate fuel cells where they are inert in the sodium carbonate electrolyte,^[55] but $\upbeta$-phase alumina is employed.

Fig. 12

Alumina crucible-charge interaction region for 54 mol pct titania-sodia carbonate charge fired in air at $1100\,^\circ{\rm C}$ for 4 hours with (a) secondary electron image indicating line scan in red (b) EDS maps of Al, Na and Ti, and (c) EDS line scan from (a) (Color figure online)

The YSZ crucible underwent significant changes with Na and Ti infiltrating the grain boundary network, corresponding to 15 mol pct Ti and 20 mol pct Na on average over the first $90 \mu \text{m}$ into the crucible. The melt was so aggressive that YSZ grains were found embedded in the glassy product, Zr and Y elemental maps in Figure 13(b). Again, an Y-enriched region formed at the crucible-product interface. YSZ is being investigated for electrolytes in molten carbonate direct carbon fuel cells,^[56] but there are conflicting reports on its longevity depending on the particular carbonate chemistry and atmosphere. The extent of infiltration of the YSZ crucibles by all-oxide and carbonate charges was compared in the bases, Figure 14. Some product remained in the base and the charge infiltrated approximately $1700 \mu \text{m}$ through the base with all-oxide charge, Figures 14(a) and (b). In contrast, no product remained in the center of the base and the charge infiltrated through the thickness to the outside of the crucible with the carbonate charge, Figures 14(c) and (d). At the wall near the base of the YSZ crucible, external bulging was observed while the upper half of the crucible wall retained its original shape. Further, voids and cracks were found in both crucible bases, with larger voids in the crucible fired with the oxide mixture compared to that fired with the carbonate mixture. EDS maps corresponding to Figure 14 are shown in Figures S-1 and S-2 (refer to electronic supplementary material).

Fig. 13

YSZ crucible-charge interaction region for 54 mol pct titania-sodia carbonate charge fired in air at $1100\,^\circ{\rm C}$ for 4 hours with (a) secondary electron image indicating line scan in red (b) EDS maps of Zr, Y, Na and Ti, and (c) EDS line scan from (a) (Color figure online)

Fig. 14

Infiltration of bases of YSZ crucibles by 54 mol pct titania-sodia charges fired in air at $1100\,^\circ{\rm C}$ for 4 hours with backscatter electron images in (a) and (c) and EDS line scans in (b) and (d). The oxide charge partially infiltrates the base and product remains inside the crucible, (a) and (b). The carbonate charge completely infiltrates the base and no product remains inside the crucible, (c) and (d). Examples of voids (circles) and, where they connect, cracks (arrows) are indicated in (a) and occur in (b) also. See Figures S-1 and S-2 (refer to electronic supplementary material) for corresponding EDS maps

No product phase remained inside the magnesia crucible after 4 hours at $1100\,^\circ{\rm C}$, Figure 15, and it’s clear from the EDS maps, Figure 15(b), that the melt infiltrated the grain boundary network. The line scan represents the boundary layer observed in Figure 11(c) with an average of 15 mol pct Ti and 11 mol pct Na over the $240 \mu \text{m}$ scan. This is consistent with literature reports of magnesia-carbonate composites. Sang et al.^[57] created composites from mixed powders of ternary carbonates and magnesia for thermal energy storage applications. No new phases were detected by XRD after processing, but the carbonate phase was evenly dispersed in the ceramic and no mass loss was detected at $1000\,^\circ{\rm C}$ in thermogravimetric analysis. Similarly, Ye et al.^[58] studied sodium carbonate – magnesia composites made in a similar fashion and demonstrated stability at $900\,^\circ{\rm C}$ by thermogravimetric analysis.

Fig. 15

Magnesia crucible-charge interaction region for 54 mol pct titania-sodia carbonate charge fired in air at $1100\,^\circ{\rm C}$ for 4 hours with (a) secondary electron image indicating line scan in red (b) EDS maps of Mg, Na and Ti, and (c) EDS line scan from (a). No product was left inside the crucible (Color figure online)

Comparison of all-oxide and carbonate charges: crucible reactivity

Experimental contamination and composition of the charge from EDS linescans are compared for all-oxide (solid symbols) and carbonate (solid symbol with black border) charges in Figure 16. Alumina contaminated the product for both oxide and carbonate charges, but neither charge caused significant contamination of the crucible, which would correspond to quadrants 3 or 4 in Figure 1. YSZ and magnesia crucibles were attacked by both oxide and carbonate charges by infiltration of their grain boundary structures, which is not easily captured with the crucible contamination values in Figure 16, and suggests that Figure 1 should include physical reactions like infiltration, not just chemical reactions. Figure 16 shows that all products had decreased titania:sodia ratios, which was surprising because our FactSage predictions are that the vapor pressure of sodium-containing species is several orders of magnitude higher than that of titanium-containing species.

Fig. 16

Experimental reactivity of crucibles with charges in air at $1100\,^\circ{\rm C}$ for 54 mol pct titania-sodia from mixed oxide (solid symbols) and carbonate (solid symbol with black outline) charges with (a) contamination of crucible and charge and (b) compositional changes in charge. The target composition and fixed 54:46 titania:sodia ratio are indicated on (b)

The best choice for containment of the oxide charge and the carbonate charge is alumina because it contaminates the melts the least and the crucible successfully contains the charge, unlike magnesia and YSZ where the charge infiltrates their grain boundary networks. While contamination of the charge is to be avoided, it might be acceptable under some circumstances for MOE experiments. If the charge could be saturated with alumina to create a single-phase ternary oxide liquid, further dissolution of the crucible could be avoided, which is a tactic suggested by Wiencke et al.^[44] and used to minimize chemical wear of refractories.^[4,24] The alumina must be inert or favorable for the MOE conditions in terms of reduction reaction and kinetics. The reduction potential for Al³⁺ should be larger than that required for reduction of the target metal Ti⁴⁺ or Ti³⁺, so as not to cause competitive reduction reactions that reduce MOE process efficiency. Optimum MOE conditions have large ionic transference numbers so that most electrical current is carried by ions.^[2] Large ionic mobility correlates with low viscosity, and aluminum is reported to be a network former which decreases ionic mobility in certain industrial slags.^[59] Because the structure of molten oxides is complicated, the effect of alumina saturation on ionic mobility would need to be experimentally determined for the specific chemistry of interest.

Crucibles for Neodymia-Boria Molten Oxide Electrolysis

No thermodynamic predictions are available for the neodymia-boria system with candidate crucible materials from either published phase diagrams or FactSage. We selected alumina, magnesia and platinum crucibles for cup test experiments with 62 mol pct neodymia-boria charges at $1300\,^\circ{\rm C}$, Figure 2. We included platinum because it was used in the thermodynamic assessments of the neodymia-boria system by Ji et al.^[34] Starting powders were blue-grey while the fired powders were purple-brown after 4 hours in air at $1300\,^\circ{\rm C}$, Figure 17.

Fig. 17

Effect of firing 62 mol pct neodymia-boria mixed oxide charge in air at $1300\,^\circ{\rm C}$ for 4 hours in (a) alumina (b) magnesia and (c) platinum crucibles with pre-fired (left) and post-fired (right) images

Fortunately, and unlike the titania-sodia experiments, all neodymia-boria products were able to be removed and the average compositions were determined by XRF, Figure 18. The raw results (maroon symbols) are boria-rich, which highlights the problem with detecting boron in XRF because it is at the bottom of the detection range. The raw results were corrected (black symbols) assuming that the raw neodymia:crucible element ratio was correct and that the neodymia:boria ratio was the 62:38 target ratio. Due to the correction method, all product compositions lie on the dashed line indicating the target neodymia:boria ratio on the Gibbs triangle. No platinum was detected in the product. In contrast, the charges were contaminated with ~ 50 mol pct magnesia or ~ 20 mol pct alumina after firing in respective crucibles. Even though the product remained in powder form, this experiment demonstrated both the sluggish melting kinetics and the high reactitivity of the charge with the ceramic crucibles. While the composition of the products was not directly measured, it’s clear that platinum is the best choice.

Fig. 18

Composition of neodymia-boria product after firing in each of three crucibles for 4 hours at $1300\,^\circ{\rm C}$ in air. The target composition, 62 mol pct neodymia-boria, is indicated with a red circle. The dashed line corresponds to fixed 62:38 neodymia:boria ratio and contamination increases with distance along the line away from the red circle (Color figure online)

Our study focusses on identifying crucibles for laboratory scale MOE experiments, where precious metals can be used. Clearly platinum is not a viable industrial crucible material. At industrial scale, a frozen crust of the charge (freeze lining) forms at the cooler outer charge/refractory interface and chemically isolates the molten charge from the container, for example, the cryolite crust in aluminum smelting pots^[60] and the proposed Molten Regolith Electrolysis reactors for oxygen production on the Moon.^[61] At laboratory scale, freeze linings are hard to realize due to low charge volumes and difficulty with temperature control.^[23] Optimisition of industrial freeze lining requires systematic thermochemical data:^[62] the recent review by Bellemans et al.^[63] describes the challenges, particularly for new slag chemistries related to secondary feed materials. For industrial scale containment, properties such as modulus of rupture^[64] and thermal conductivities^[65] at operating temperature further change the selection criteria for crucibles.

FactSage is the most widely used CALPHAD tool for high temperature materials processing.^[66] While mobility data and some diffusion-controlled predictions are implemented in Thermo-calc and its companion DICTA,^[31] FactSage does not have kinetic information. Van Ende et al.^[67] introduced the Effective Equilibrium Reaction Zone (EERZ) model for the reaction kinetics between phases, and it has been applied to a range of industrial metallurgical processes with FactSage. But the EERZ model still requires assumptions about kinetics that require experimental data sets. Computational methods such Monte Carlo or Molecular Dynamics could be applied to predict reaction rates and mechanisms, but these require long computational times and apply to specific, usually simplified, chemistries and geometries. See Jiang et al.^[68] for a recent application of ab initio molecular dynamics to develop detailed predictions of molten slag-graphite chemical wear. In the present work, we illustrate that experimental investigations are always required for specific systems of interest. To refine laboratory crucible choice with the least experimental effort, time-series tests should be conducted and detailed phase identification and composition profiles characterized.

Conclusions

The aim of this study was to select crucibles for MOE laboratory experiments at elevated temperatures. Materials selection using Ashby’s method with ANSYS Granta software is limited due to a lack of chemical compatibility data, which must be determined for the specific melt composition or range of compositions, and is beyond the scope of software such as Granta. Further, the generic maximum service temperature in Granta encompasses mechanical properties that are not relevant for laboratory scale crucibles and causes materials, particularly metals and alloys, to be deselected for this application.

Even with thermodynamic data, slow kinetics can limit reactivity so that crucibles that are predicted to be poor choices can actually be good choices. For 54 mol pct titania-sodia at $1100\,^\circ{\rm C}$, FactSage predicted magnesia to be the best choice, but experiments showed that alumina is the best choice for containment of both oxide and carbonate charges. While alumina contaminated the charges, it might be acceptable for MOE if (i) the charge could be saturated with alumina to prevent further damage to the crucible, (ii) the ionic mobilities in the single-phase molten oxide were not decreased by the alumina and (iii) aluminum was less electrochemically active than the target metal Ti. In our application, infiltration of the magnesia and YSZ grain boundary networks by the carbonate is a negative result: neither magnesia nor YSZ is suitable for crucibles for our experiments. On the other hand, our negative results may be of interest for molten carbonate fuel cells or for thermal energy storage, where additional characterization would be required. For the 62 mol pct neodymia-boria melt, contamination was minimal in the platinum crucible after 4 hours at $1300\,^\circ{\rm C}$.

Selecting crucibles for industrial MOE is significantly more complicated due to the additional mechanical requirements and the potential to design freeze linings to isolate the oxide charge from the crucible.