Hall-Héroult process produces aluminium by alumina electrolysis (Al2O3) in a molten mixture of cryolite at high temperature (960˚C)  :
Aluminum metal is obtained at the cathode and oxygen ions react with a reactive carbon anode to produce carbon dioxide. The electrolyte is composed of Na3AlF6 - AlF3 - CaF2 and Al2O3 according to a NaF/AlF3 molar ratio of 2.2 (called cryolitic ratio CR).
One of the key challenges for the aluminum industry is to replace CO2 by O2 release on an inert anode. The basic requirements for an inert anode are listed below  :
- High thermo-mechanical properties (shock and thermal gradient resistances).
- High electrical conductivity (>80 S∙cm−1).
- High stability to oxidation at high temperature.
- Consumption rate lower than 1 cm/year for 0.8 A∙cm−2.
To fulfill these conditions, three types of inert anode for aluminum electrolysis have been examined up to now:
· Metals or alloys    , for their good ductility, conductivity and thermal chock resistance. However, metals are highly degraded under oxidation and thus cannot be used as anode (eg: Fe, Ni or Co metals  ).
· Ceramics   , for their excellent oxidation resistance, but are generally brittle and have a low conductivity incompatible with an electrolysis process (eg: spinel NixFe3−xO4).
· Cermets (CERamic-METals)   , made of ceramics (monoxide MO, nickel ferrite NixFe3−xO4) and metal, combining the advantages of both materials. The first patents on cermet anodes have been published by Aluminum Company of America (ALCOA) in the 80’s   .
ALCOA studied cermets composed of copper as metallic phase and a mixture of NiFe2O4 and NiO   . Anodes were tested in cryolite at 0.8 A∙cm−2 at 970˚C and analyzed by SEM-EDS. The 83 wt% (80NiFe2O4 - 20NiO)/17 wt% Cu composition presented the best resistance to polarisation.
Studies by Lai et al.  on cermets composed of 83(90NiFe2O4 - 10NiO)/17Ni concern the degradation mechanism at 960˚C under electrolysis at 0.8 A∙cm−2 in molten cryolite (saturated in Al2O3). The authors observed preferential oxidation of metallic Ni, leading to porosity in the structure, and electrolyte penetration along the grain boundaries. The authors also pointed out a chemical dissolution of the spinel phase and a reaction between NiO and AlF3 to form NiF2 and Al metal. Tian et al.  completed this work by varying the metallic Ni content from 0 to 20 wt% at 970˚C under polarisation at 1 A∙cm−2 and CR = 2.3 (4 wt% Al2O3). According to their results, cermet composed of 5wt% of metallic Ni presented the best resistance to degradation and the best electrical conductivity under polarisation.
Liu et al.  investigated a Cu-Ni mixture as metallic phase in the 83(90NiFe2O4 - 10NiO)/17(Cu - 10Ni) material. Studies were performed in molten cryolite at 960˚C, under electrolysis at 1 A∙cm−2 and CR = 2.3, saturated in Al2O3. A dense NiFe2O4 layer was formed at the anode surface, whose thickness increased during the electrolysis process. A higher oxidation rate of Cu compared to Ni was observed in the cermet. By comparing the microstructure of 90(90NiFe2O4 - 10NiO)/10Cu and 90(90NiFe2O4 - 10NiO)/10(Ni-Cu) cermets after electrolysis, He et al.  suggested, in agreement with Lai et al.  , a preferential oxidation of the metallic phase and an increase of the porosity. A dense layer was formed at the salt/electrode interface by reaction between NiFe2O4 and Al2O3 to form NiAl2O4 and FeAl2O4. The authors suggested that this layer acted as a protective barrier, limiting the cermet oxidation, and concluded that the cermet composition with metallic Ni is the most promising inert anode.
Tests were also performed on (90NiFe2O4 - 10NiO)/(Ni-Cu) at 0.95 A∙cm−2 in Na3AlF6 - K3AlF6 - AlF3 - Al2O3 at 900˚C by Tian et al.  but anodes were still highly deteriorated. A dense layer composed of FeAl2O4, NiAl2O4 and CuAl2O4 was formed (150 µm) at the salt/electrode interface, and is supposed to protect the initial cermet material. More recently, Meyer et al.  observed that FeAl2O4 is rapidly dissolved in molten cryolite.
Despite intensive investigations, no satisfactory inert anode material composition has been found yet and no global degradation mechanism of cermet under polarisation is proposed.
A cermet developed in house (described in §2.d) was tested under galvanostatic electrolysis at 0.8 A∙cm−2 in Na3AlF6 - AlF3 - CaF2 - Al2O3 at 960˚C for different durations under Ar atmosphere. Anodes were analyzed by Scanning Electron Microscopy (SEM), coupled with the energy-dispersive X-ray spectroscopy (EDS), to determine the microstructure compositions. From these analyses, an electrochemical degradation mechanism of the cermet is proposed and validated by thermodynamic calculations using Factsage 7.1 software.
2.1. The Cell
The cell was a vitreous carbon crucible placed in a cylindrical vessel made of refractory steel and closed by a stainless steel lid cooled by circulating water. The inside part of the walls was protected against fluoride vapors by a graphite liner. The experiments were performed under an inert argon atmosphere (4.5 grade), previously dehydrated. The cell was heated using a programmable furnace and the temperatures were measured using a chromel-alumel thermocouple. A schematic diagram of the experimental setup has already been presented in previous article  .
The cryolite melt is composed of Na3AlF6 (Carlo Erba, purity 99.99%), with an addition of 11 wt% excess of AlF3 (ProChem Inc, purity 99.99%) and 5 wt% of CaF2 (Apollo Scientific, purity 99.99%), leading to a cryolitic ratio (CR) of 2.2. Alumina saturation was achieved by addition of 8 wt% of alumina (Alfa Aesar, purity 99%). The salt mixture was heated until 960˚C under argon atmosphere.
A parallelepipedic shaped cermet (1.25 cm2 surface area) was used as working electrode. This relatively small surface area was selected to ensure its full degradation within few hours. The auxiliary electrode was a Ni plate to avoid Al metal in the salt (formation of Al-Ni alloys). The reference electrode is a molybdenum wire (Goodfellow 99.99%, 0.5 mm diameter) enclosed in a boron nitride crucible (BN, AX05 grade), immersed into the bath, in contact with the solution through a 0.3 mm diameter hole.
The electrochemical studies and galvanostatic electrolyses were performed with an Autolab PGSTAT302N potentiostat/galvanostat controlled with the NOVA 2.1 software. The cermets were embedded in cold conductive resin and polished for surface analysis of their cross-sections. Scanning electron microscope (SEM) with a Backscattered Electron (BSE) detector observations coupled with EDS analysis were performed on a PhenomWorldXL.
2.4. The Cermet
Cermet was obtained by a powder metallurgy process: NiFe2O4, NiO, Ni and Cu powder were mixed together with an organic binder, isostatically pressed and sintered at 1350˚C. After sintering, this cermet is composed of spinel (NixFe3−xO4), nickel iron monoxide (Ni1−xFexO) and copper-nickel-iron alloy (CuxNiyFez). Figure 1 shows the stable phases when the partial pressure of oxygen is 10−10 atm. in the ternary diagram of Ni, Cu, Fe system. The back point is the proportion of the nickel, iron and copper elements in the studied cermet; its microstructure is composed of monoxide and metal phases, included in a continuous spinel matrix, and the residual porosity is 2%vol.
3. Results and Discussion
3.1. Thermodynamic Approach
To predict the oxidised phase compositions, thermodynamic evaluation of Gibbs
Figure 1. Ternary diagram of the Ni, Cu, Fe system (pO2 = 10−10 atm). Zone 1: metal and spinel coexistence. Zone 2: metal, spinel and monoxide coexistence; Zone 3: metal and monoxide coexistence.
free energies was realised with FactSage 7.1 software, with FactPS, FTOxid, SGTEa and FTHall databases. Standard states were selected as follows for the different species: the gaseous species (g) obeys the ideal gas equation at standard pressure (P = 1 atm), the solid (s) and liquid (l) compounds are considered as pure substances. Moreover, all the calculations were performed at 960˚C.
From these values, standard potentials were calculated as follows:
· In presence of fluoride ions, F2(g)/F− redox couple was used to express the iron oxidation as:
Gibbs free energy was determined for Equation (4) at 960˚C: .
The FeF2/Fe standard potential is equal to vs. F2(g)/F−.
· In presence of oxide ions, O2(g)/O2− redox couple was selected and the iron oxidation reaction is:
where O2− represents all the fluoroaluminates ions that can be oxidized to produce dioxygen    .
The Gibbs free energy determined by FactSage for Equation (7) at 960˚C is , corresponding to vs. O2(g)/O2−.
To compare the fluoride and the oxide species standard potentials, the same reference potential is needed: AlF3/Al(l) redox couple was chosen; the reference potential change is explained elsewhere  .
The study was realised with the constitutive elements of the anode (Cu, Ni, Fe) considered as pure elements and the results are gathered in Figure 2.
Figure 2. Standard potentials scales determined at 960˚C.
From these scales, metallic iron is the first element to be oxidized into FeO and FeF2, depending on O2− content. Then, nickel is oxidized into NiO or NiF2 and copper into Cu2O, CuF2 and CuO. All these oxide/fluoride formations were confronted to micrographic analyses of the anode after electrolysis. In this work, the potentials will be referenced to AlF3/Al redox couple.
3.2. Experimental Anodic Potential during the Electrolysis Process at 0.8 A∙cm−2
On the cermet anode, oxygen is produced; to characterise the material oxidation, a linear voltammogram was plotted at 50 mV∙s−1 on the cermet and on gold electrode, in a CR 2.2 bath containing 5wt% of CaF2 saturated in alumina at 960˚C (Figure 3).
On the j vs. E curve, two zones are observed, meaning that at least two reactions take place. To identify the O2 evolution, a gold electrode was selected as only O2 is formed  . The linear voltammetry superimposition shows that the second zone (from 2 V vs. ref) corresponds to O2 evolution and the first zone can be attributed to the cermet material oxidation (between 1.55 V vs. ref and 2 V vs. ref).
In our experimental conditions (j = 0.8 A.cm−2), the anode potential during the electrolysis is around 2.8 V vs. ref, meaning that both O2 evolution and material oxidation take place: the next paragraph deals with the investigation of the material electrochemical degradation.
3.3. Determination of the Electrochemical Degradation Mechanism
The approach used to determine the material degradation steps was:
· Electrolyses at 0.8 A.cm−2 for different polarization durations.
· SEM analysis of the anode cross-section to observe the changes of oxide and metallic phases.
· EDS analysis to determine their compositions.
As the cermet degradation state is function of its geometrical dimensions, the
Figure 3. Linear voltammograms (vb = 0.050 V∙s−1) plotted on gold electrode (thin line) and cermet (bold) in cryolitic bath (CR = 2.2, 5 wt% CaF2, saturated in Al2O3, T = 960˚C).
electrolysis time was converted into a degradation progress (λ); for λ = 1, the CERMET is fully oxidised, and corresponds to the anode service life.
The global cermet electrolysis curve E = f(λ), corresponding to the anode end of life (λ = 1), is shown in Figure 4.
This curve is divided into three zones:
· In zone 1 (Z1), the potential is stable and corresponds to the potential previously measured by linear voltammetry (E = 2.8 V vs. AlF3/Al).
· In zone 2 (Z2), a significant increase of the potential is observed.
· In zone 3 (Z3), the potential strongly increases up to 10 V vs. AlF3/Al.
A progressive increase of the anode potential is then observed with the electrolysis duration. To investigate the electrochemical degradation mechanism, electrolyses were performed in the different zones and results are detailed below.
3.3.1. Zone 1: Formation of Monoxide and Fluoride Phases
Figure 5 shows the SEM micrographs of the cermet interface after different degradation progresses λ = 6.25%, 12.5% and 25%.
These micrographs show an important decrease of the metallic phase amount with the electrolysis duration: at the anode/electrolyte interface (100 µm), only 30% of the initial metal remains in the structure for λ = 25%. The metallic phase is then continuously oxidized and leads to:
· FeF2: this new phase is detected in contact with metallic grains up to 120 µm depth for λ = 6.25%, (Figure 5(a)). Its formation can result from the oxidation of iron metal in presence of fluoride ions according to the reaction:
This result is in agreement with the thermodynamic approach (Figure 2), where the iron metal is the first element to be oxidized.
Moreover, at 45 μm from the anode/electrolyte interface (Figure 5(c)), the detected fluoride phases contain both FeF2 and NiF2. NiF2 can be obtained by oxidation of nickel metal with fluoride ions, according to the following reaction:
Figure 4. Variation of anode potential vs the degradation progress for electrolysis at 0.8 A∙cm−2 in a CR = 2.2 bath saturated in alumina with 5%wt of CaF2 at 960˚C. X: tested durations.
Figure 5. SEM micrographs of the cermet interface after electrolysis at 0.8 A∙cm−2 in a CR = 2.2 bath, saturated in Al2O3 with 5%wt of CaF2 at 960˚C: (a) λ = 6.25% (b) λ = 12.5% (c) λ = 25%.
However, at the anode/electrolyte interface, FeF2 and NiF2 are never detected: it might be due to their dissolution in the electrolyte. FeF2 and NiF2 dissolution generates an additional porosity and therefore accelerates the electrolyte penetration into the cermet.
From these observations, oxide phases are then formed next to the salt interface whereas fluoride phase are detected in the bulk. However, micrographs clearly show that their formation front moves inwards with the anode degradation progress, as presented in Table 1 where the degradation front has been measured for different degradation progress.
· Ni0.8Fe0.2O: this monoxide is observed close to the electrolyte.
According to the thermodynamic scales, in presence of oxide ions, iron and nickel can be oxidized into FeO and NiO. However, a mixture iron and nickel monoxide is detected by EDS, so a two-step formation mechanism is proposed:
1) Metallic iron in contact with the electrolyte is oxidized into FeO:
2) FeO is then dissolved into NiO, formed by oxidation of metallic nickel, according to the following reaction:
For λ = 25%, (Figure 5(c)), only one monoxide composition is detected with x = 0.9 nickel amount. This oxide composition is then the most stable and the Equation (9) becomes:
· The NixFe3−xO4 spinel phase is continuously enriched in nickel, up to a limit value of x = 0.90 (Ni0.90Fe2.10O4). This final spinel composition is assumed to be the most stable one in this experimental condition  . To explain the nickel increase in the spinel phase, a ternary phase diagram of the Ni-Fe-O system at 960˚C was plotted in Figure 6.
Table 1. Degradation front into the cermet for different degradation progresses.
Figure 6. Phase diagram of Ni-Fe-O system at 960˚C.
The stability zones of the material phases are represented: NixFe3−xO4 spinel, Ni1−xFexO nickel iron monoxide and metallic alloy. The only part of the diagram where the three phases are in equilibrium is in the part 4 (grey zone).
The initial composition of the spinel is displayed with a circle (zone 4) and the oxidized spinel phase Ni0.90Fe2.10O4 with a cross. The Ni-Fe-O phase diagram at 960˚C confirms the possible increase of Ni into the spinel phase until x = 0.90, as observed for the monoxide phase (Ni0.90Fe0.10O).
The reaction producing the nickel increase can be written as:
This reaction releases Fe2+ which forms FeF2 according to the reaction:
3.3.2. Zone 2: Cu2O Formation
Electrolyses were then carried out in zone 2. Figure 7 represents the SEM micrographs of cermet interface (a) and core (b) at the half of the degradation (λ = 50%).
The core of the anode (Figure 7(b)) is mainly composed of:
· Few grains of metallic phase.
· Ni0.90Fe2.10O4 spinel phase.
· Ni0.90Fe0.10O monoxide phase.
· A mixture of NiF2 and FeF2.
The same phases detected at the electrolyte/anode interface in zone 1 are present at the core at λ = 50% (Figure 7(b)), showing a continuous progression of the degradation front towards the core of the anode. Up to 400 μm depth (Figure 7(a)), the material is only composed of oxides: Ni0.90Fe0.10O, Ni0.90Fe2.10O4
Figure 7. SEM micrographs of the cermet after electrolysis (λ = 50%) at 0.8 A∙cm−2 in a CR = 2.2 bath, saturated in Al2O3 with 5 wt% of CaF2 at 960˚C: (a) interface (b) core.
and Cu2O, formed by oxidation of metallic copper in presence of oxide ions according to the reaction:
These results are in agreement with the potential scale (Figure 2), where copper is more noble than iron and nickel. At this step, metallic iron is not present anymore and the oxide phases have reached a thermodynamically stable composition: nickel is then preferentially oxidized into NiF2, while copper is oxidized into Cu2O.
The electrolyte is also present into the porosity and keeps on contributing to the dissolution of the fluoride phases.
The potential increase in zone 2 is directly correlated to the cermet composition: the metallic phase is converted into oxides and leading to porosity. The global conductivity of the anode decreases then continuously because of low conductivity oxide phase formation at 1000˚C:
- Ni0.90Fe0.10O: 18 S∙cm−1 
- Ni0.90Fe2.10O4: 0.5 S∙cm−1 
- Cu2O: 1 - 2 S∙cm−1 
3.3.3. Zone 3: Formation of (Ni, Cu, Fe)O Monoxide
Figure 8. SEM micrographs of the cermet interface after electrolysis at 0.8 A∙cm−2 in a CR = 2.2 bath, saturated in Al2O3 with 5%wt of CaF2 at 960˚C: (a) λ = 75%; (b) λ = 100%.
At λ = 75%, the anode is only composed of Ni0.90Fe2.10O4, Cu2O and electrolyte inside the material.
At λ = 100%, the monoxide composition is Ni0.90Cu0.08Fe0.02O. Copper(I) from Cu2O is oxidized into copper(II), substituting the iron(II) in the structure. As a consequence, iron(II) is released and then dissolved in the electrolyte:
In these experiments, the cermet is totally degraded:
· The metallic phase is no more detected.
· The porosity is very important, leading to a complete dissolution of fluoride phases.
· The nickel increases until x = 0.90 in the spinel and monoxide phases confirms the formation of the most thermodynamically stable composition (§3.2).
Ni0.90Fe2.10O4 and Ni0.90Cu0.08Fe0.02O phases are not conductive enough to ensure the current flow, meaning that the cermet has then reached its end of life.
Figure 9. SEM micrographs of the core of the cermet after electrolysis at 0.8 A∙cm−2 in a CR = 2.2 bath, saturated in Al2O3 with 5%wt of CaF2 at 960˚C: (a) λ = 75%; (b) λ = 100%.
Cermet composed of spinel, nickel monoxide and copper-nickel-iron alloy was studied as inert anode for aluminum production. Electrolysis with different durations at 0.8 A∙cm−2 in molten mixture of cryolite at 960˚C was undertaken and new phases formation, in addition of dioxygen evolving, was evidenced by SEM coupled with EDS analysis.
- The oxidation of metallic phase, coupling with thermodynamic calculations, showed that iron is the first element to be oxidized, into FeF2 in presence of fluoride F− and into FeO with O2− from the electrolyte:
- NiF2 and Ni0.90Fe0.10O phases are also detected, obtained from the oxidation of nickel metal with fluoride and O2− ions, according to the following reactions:
- Bath penetration into the cermet occurred and led to the dissolution of the first fluorides formed.
- For the spinel phase, a nickel enrichment up to x = 0.90 due to metallic nickel oxidation intervened:
This final spinel and monoxide compositions are the most stable in our experimental conditions.
- Then Cu2O is formed by metallic copper oxidation:
- Fully oxidized, the cermet becomes non-conductive and only two oxide phases with a nickel amount of x = 0.90 are detected:
For the first time, a degradation mechanism of a cermet anode was proposed, in agreement with thermodynamic calculations. When Cu2O phase is formed (zone 2), the measured potential increases meaning that the anode is less conductive. For a long-term use, it would be interesting to increase the zone 1, where the electrolysis potential is stable, by increasing the metal proportion into the cermet: it would prevent the formation of non-conductive phases. Another possible way is the use of a low-temperature electrolyte, to slow down the oxidation phenomenon into the cermet.
The authors express their thanks to the French Research National Agency ANR-13-RMNP-0012 for their financial support for this research.