1. Introduction
The molten salt reactor (MSR) is a next-generation reactor design in which the coolant is a molten salt [
1,
2]. CEA Saclay is currently studying corrosion in molten chlorides to develop an MSR that would use NaCl-MgCl
2-PuCl
3/Am as a fuel. The salt is then the coolant and the fuel as it contains PuCl
3 and possibly UCl
3 [
3,
4]. This design enables the fuel to be renewed without shutting down the plant by regularly using new batches of salt. It also makes the plant safer than a pressurised water reactor since the reactivity coefficient is negative. In the event of excess reactivity, the temperature rises and the salt expands, increasing the distance between the plutonium atoms and reducing reactivity [
5]. The reactor is then self-regulating. Before studying the ternary salt, the binary salt NaCl-MgCl
2 is used in corrosion tests because it is the fuel solvent.
However, NaCl-MgCl
2 is a very corrosive medium due to the presence of dissolved impurities such as oxide ions, water, oxygen and chlorine in the salt [
6]. These impurities are mainly due to their reactivity with oxygen and moisture, as MgCl
2 is very hygroscopic [
7,
8]. In order to have comparable and reproducible data, it is very important to quantify the impurities in the molten salt before any immersion [
9]. Various methods have been described in the literature, like cyclic voltammetry, square-wave voltammetry or the acid–base titration [
10,
11,
12]. A purification protocol must be implemented to reach a state where the salt is free of impurities. To reach this objective, the dehydration and melting of NaCl-MgCl
2 salts must be studied.
In molten salts, two types of alloys can potentially be used: chromia-forming alloys and alumina-forming alloys. Chromia-forming alloys have been widely studied in molten chlorides. They are not suitable for MSR application due to chromia solubility in molten chlorides [
13]. Iron-based alloys undergo intragranular corrosion that can be superior from 20 to 30 µm deep after 100 h of immersion [
14,
15,
16,
17]. The dissolution of the matrix is often observed with the apparition of intergranular corrosion. MgO has been observed at the surface of the corroded alloys, and Mg–Cr–O oxides have also been observed by different authors [
14,
18,
19]. The same corrosion facies are observed in the corrosion of nickel-based alloys like Inconel 625 [
6,
18,
20]. If the matrix does not dissolve, iron and chromium will preferentially dissolve from the matrix, as reported by Mortavazi et al. [
6]. This author also showed that the level of purity of the salt in terms of moisture and oxide ions plays a major role in the behaviour of the immersed alloy: a cleaner salt will be less corrosive.
There are few studies on alumina-forming alloys [
21,
22,
23]. Even if it is difficult to compare different studies that do not provide information on the purity level of their salt, alumina-forming alloys seem to be a better alternative, as shown by Gomez-Vidal et al. [
21]. They proved the usefulness of preoxidation to enhance the corrosion resistance of alumina-forming alloys in molten chloride salts.
The aim of this study is twofold: (i) to design and investigate a purification protocol for NaCl-MgCl2 molten salts that will ensure reproducible and comparable corrosion results between different alloys; (ii) to assess whether alumina-forming alloys are attractive candidates as structural materials for MSRs. To this end, for economic reasons, alumina-forming steels were considered in this study (Ni-based alloys will be evaluated in another study).
OC1 and OC4 steels were developed by Oak Ridge National Laboratory (ORNL) in collaboration with Carpenter Technologies [
24]. They were designed to be used as micro turbine recuperator components; thus, they were designed to withstand oxidation and creep, after which the alloys were named (OC: Oxidation Creep). Their low nickel content allows them to be economical, and their high Nb content makes them resistant to creep due to the formation of niobium carbides. Their high aluminium content makes them resistant to high-temperature corrosion (up to 200 °C higher than conventional stainless steels) due to the formation of an alumina scale at the surface of the material. They were tested up to 6000 h in engines and showed good performance overall, but a 2–3 µm-deep aluminium depletion was observed under the surface. According to their designers, these alloys are very promising and cost-effective for micro turbine applications but also for all industrial applications up to 700–750 °C. Above 750 °C, OC1 and OC4 are subject to internal aluminium oxidation, which makes them unsuitable for such applications. ORNL is currently studying the effect of minor elements on the ability of alloys to form alumina in order to increase their operating temperature range. A corrosion period of 500 h was chosen because previous studies of the corrosion of alumina-forming alloys have demonstrated the good corrosion resistance of In-702 alloy immersed up to 185 h in MgCl
2-KCl salt in cycling temperature from 500 °C to 700 °C [
25]. This test should enable OC1 and/or OC4 to be qualified for other longer-term tests, notably on corrosion loops. The present work also aims to understand the behaviour of alumina during prolonged immersion in molten chloride, given that Al
2O
3 has been reported in the literature to be capable of immersion of 185 h. This study should provide a better understanding of corrosion mechanisms in molten chloride salts in order to select or design an alloy that performs better than OC1 and OC4 in molten NaCl-MgCl
2.
2. Materials and Methods
Molten salt was prepared from NaCl supplied by Sigma-Aldrich with 99.9% purity and MgCl2 supplied by Acros International with 99% purity. The salts were stored in a glovebox under pure argon. They were mixed to form approximately 180 g of the eutectic 0.57 mol NaCl—0.43 mol MgCl2. FeCl2 and CrCl2 were supplied by ThermoScientific (Geel, Belgium) with a purity of 99.9%. Both components were anhydrous.
The experimental set-up is described in
Figure 1a,b. The reactor consists of a lower section and a lid through which the electrodes and sample holder pass. The lid and lower section are sealed with a Teflon
® gasket (DWK, Meiningen, Germany). High-purity argon (Ar BIP) sweeps the volume above the liquid salt. The working and counter electrodes used are 99% purity, 1 mm diameter tungsten wires supplied by Goodfellow (Lille, France), and the reference electrode is Ag/AgCl. The electrodes consist of a Pyrex tube through which a tungsten wire passes. A silicone paste seals the top of the tube. The reference electrode is custom-made using a silver wire supplied by Thermoscientific (Geel, Belgium) immersed in the 0.55 mol NaCl—0.40 mol MgCl
2—0.6 mol AgCl salt contained in a Pyrex pocket.
Cyclic voltammetry (CV) was mainly used to analyse the molten salt after experiments. A preliminary parametric study aiming at defining the optimal potential sweep rate was carried out between 20 mV/s and 200 mV/s on the tungsten/molten chloride salts system and led to the choice of 100 mV/s. This study allowed us to choose a sweep rate that gives a quick and stable response as well as a good definition of peaks.
The working electrode surface was 0.33 cm2. All electrochemical experiments were performed using an SP-200 potentiostat (Biologic, Seysinnet-Pariset, France) controlled by an EC-lab software v11.43. The voltammograms were traced 5 times, and the 4th cycle was used for representation. The CV graphs were then traced in a way to use Cl2/Cl– equilibrium potential as a reference: a tangent to the oxidation signal of Cl– is traced in the part where the current density is superior to 100 mA/cm2. Then, the voltammogram is displaced in a way such that the tangent intersects the abscises in 0. This process allows for the comparison of CV despite possible instabilities of the Ag/AgCl reference. Moreover, it allows us to compare the CV curves from different studies that use other reference electrodes. The non-coincidence of Mg/Mg2+ peaks between different CV measurements (or cycle) can be a consequence of salt composition variations as the experiment progresses.
The materials used in this study are two austenitic steels provided by ORNL with a 4 wt% of aluminium. Their chemical compositions communicated by the provider are given in
Table 1. OC1 was preoxidised under dry air for 24 h at 700 °C, and OC4 was preoxidised under dry air for 24 h at 800 °C [
26]. The alloys were preoxided on a silica rack placed in the oven. They were hanged by an alumina stem. They were inserted when the oven was at preoxidation temperature. The samples were cooled in the oven. They were immersed, as shown in
Figure 1b.
The outlet gas was analysed during some experiments using a Pfeiffer Vaccum (Annecy, France) Omnistar GSD 301C mass spectrometer with a 950 V coil voltage.
Raman spectroscopy was performed on preoxidised alloys using a HORIBA (Palaiseau, France) LabRam HR800 with 532 nm laser.
Thermo-gravimetry analysis (TGA), using a Setaram (Caluire, France) TAG 24 under sweeping argon with a heating rate of 2 °C/min, coupled to mass spectrometry, was performed on the salt in order to identify the salt dehydration temperatures and the species formed during this process.
The coupons were observed using a Zeiss (Rueil-Malmaison, France) Gemini 2 Crossbeam 550 FIB-SEM and a Zeiss Gemini Ultra 55 SEM coupled with a Brucker (Palaiseau, France) Nano EDX detector. After these preoxidations, OC1 and OC4 samples were immersed in molten NaCl-MgCl2 at 600 °C for 500 h. They were half-immersed in the salt and hanged with a silver wire. Each alloy was immersed in its own reactor to prevent any interaction between the different coupons. The potential of each coupon was measured during the corrosion test.
4. Discussion
During the preparation of the salt, it was observed in
Figure 4 that only CO
2 was produced. Therefore, the anodic reaction could be the oxidation of carbon reacting with oxide ions according to reaction (4).
The cathodic reaction would be:
The global reaction during electrolysis would be (4) + 2(5):
As observed on the cyclic voltammetry before and after the electrolysis in
Figure 3, the reduction peak between E = −1.6 V vs. E(Cl
2/Cl
–) and E = −2.5 V vs. E(Cl
2/Cl
–) that Skar et al. associated with hydroxychloride reduction decreased after the electrolysis [
29]. The electrolysis purified the salt faster by accelerating the disappearance of MgOHCl at the beginning of the process. At that time, all the MgOH
+ is not yet transformed into MgO and can react during the electrolysis on the glassy carbon anode while magnesium is deposited at the tungsten cathode.
The anodic reaction proposed is then:
The cathodic reaction is reaction (5). The global reaction is (7) + (5).
In this case, H2 production should be observed by the mass spectrometer. Until now, this mass has not been monitored during the electrolysis. Further work will attempt to observe H2 production to conclude the electrolysis reactions.
During the electrolysis, magnesium dissolution from the cathode was observed when the current was not high enough [
31]. This can decrease the potential of the salt as Mg/Mg
2+ acts as a buffer. This influences the corrosion behaviour of the materials in the molten salt. The electrolysis current density has to be chosen wisely. The current density used in this work was 54 mA/cm
2 and was efficient. If it is too high, Cl
2 or COCl
2 can be produced, which is very dangerous.
The electrolysis reduces the amount of MgO present at the end of the purification as MgO is a reaction product from the hydrolysis of MgOHCl as shown in
Figure 4 and reported by Kipouros et al. and Kirsch et al. [
7,
8]. The MgO already present does not react during the electrolysis. However, the remaining MgO present at the bottom of the crucible (identified by XRD analysis not shown here) could react with spots of the metal surface directly exposed to molten salts. To eliminate as much MgO as possible, the best solution is, after natural precipitation by argon sweeping (or electrolysis), to solidify the purified salt and cut out the bottom of the crucible where all the MgO is located.
According to Delpech et al. [
32], the salt potential is controlled by Mg/MgCl
2 and Cl
2/Cl
– redox couples. In that case, the theoretical value for the salt potential is −1.33 V vs. E(Cl
2/Cl
–) at 600 °C, which corresponds to the measure made on the W electrode. As shown by
Figure 8, the potential of OC4 and OC1 are inferior to that value. Therefore, corrosion is expected. Moreover, the results show that the salt potential can be controlled by influencing the activity of chloride ions or the activity of magnesium chloride. To mitigate corrosion in NaCl-MgCl
2, the salt potential must be lowered by diminishing the activity of MgCl
2 or by increasing the activity of Cl
–. Diminishing the activity of MgCl
2 would reduce the proportion of MgCl
2 in the salt.
To understand the behaviour of the immersed alloys, E-p(a
MgO) diagrams were traced. They are comparable to Pourbaix’s diagram: they describe the stability domain of metallic species as a function of the potential and the dissolved oxide ion activity. In this case, the dissolved oxide ion activity is represented by the MgO activity as oxide ions are not implemented in the thermodynamic databases (HSC, Factsage, Thermocalc). The diagrams are calculated with Cl
– activity equal to 0.5 and MgCl
2 activity equal to 0.45, as calculated by Delpech et al. [
32]. The activity of alloying elements is equal to the value calculated by Thermocalc in OC4 alloys at 600 °C: 0.26 for Ni, 0.69 for Fe, 0.9 for Cr, 6.3 × 10
−7 for Al and 2 × 10
−4 for Nb. The activities of dissolved species are considered equal to their concentration measured by ICP. As ICP values for the concentration of Cr, Fe, Ni and Al are between <8 × 10
−5 and 2 × 10
−3 mol/L, an activity equal to 10
−4 is used for all dissolved Cr, Al, Ni, Fe chlorides as they are not observed in the OC4 salt. The value of 10
−4 mol/L corresponds to the limit concentration given by the 10 mA/cm
2 on the E-j curves that correspond to the background noise. All other activities are considered equal to 1 for the calculation of the predominance diagram. The Gibbs energies were calculated using the HSC database. It must be mentioned that thermodynamic data are inconsistent concerning the molten chloride environment, resulting in discontinuities in some diagrams. However, the E-p(a
MgO) can provide a qualitative idea of the species that can form and is useful as a first approach when compared to the potential of the salt and of the alloys during the corrosion test. Due to their similar composition, OC1 and OC4 diagrams are very similar. That is why only OC4 diagrams are presented. The salt chemical conditions, in terms of E
salt-pa
MgO, are represented by the red cross in
Figure 21,
Figure 22,
Figure 23,
Figure 24 and
Figure 25 that corresponds to E = E
salt and to a saturated salt in MgO (pa
MgO = 0), as solid MgO is observed at the bottom of the crucible. The aluminium diagram shows a large stability domain for corrosion products (Al
2O
3 and MgAl
2O
4) at the salt potential. The alumina scale should be stable and protective if the p(a
MgO) conditions are convenient, i.e., for activity higher than 10
−5. This shows the potential benefit of alumina-forming alloy in a molten salt environment. According to
Figure 21, for a low oxide ion activity, the acid reaction occurs, dissolving Al
2O
3 into Al
3+. In the present case, where MgO activity equals 1 and according to
Figure 21, the basic reaction transforms Al
2O
3 into MgAl
2O
4. This is exactly what can be seen in
Figure 14,
Figure 16 and
Figure 19, where alumina seems to be formed in the deepest part of the alloy before transforming into MgAl
2O
4. For instance, the circled zone in
Figure 19 corresponds to a zone where alumina is present, while the zone just above it has already been transformed into spinel. Consequently, MgO activity is equal to 1 at the oxidised alloy/salt interface and decreases with depth in the alloy.
At the salt potential,
Figure 22 shows that Ni is in its immunity domain. NiO can be formed for higher potential high-oxide activity conditions.
According to the iron stability diagram (
Figure 21), iron should dissolve in Fe
2+ form as the potential-oxoacidity of the salt is in the FeCl
2 stability domain.
Chromium (
Figure 24) and niobium (
Figure 25) have similar behaviours as iron, but they can form oxides in higher potential/high MgO activity conditions.
According to their E-p(aMgO) diagrams, Nb is supposed to form NbO2 and Cr to form Cr2+.
These predictions are in accordance with what is observed in the case of OC1: chromium and iron dissolution are observed as expected (ICP-AES analysis of salt after OC1 corrosion test,
Table 5). In the case of OC4, no iron nor chromium dissolution is observed. It is probably due to the thicker and then more protective alumina layer on the surface, as shown in
Figure 7. In both cases, Nb does not oxidise. As it is present in the form of carbides, it would be necessary to study the thermodynamic stability of Nb carbides in the presence of salt.
Finally, the corrosion potentials of the OC1 and OC4 alloys can also be explained by the predominance diagrams, which show that they are close to the Cr/Cr
2+ and Al/Al
3+ thermodynamic equilibrium potentials (
Figure 21 and
Figure 24).
The two alloys have a complex microstructure containing several types of precipitates. The alloys are out of equilibrium, as suggested by the fact that many phases predicted by ThermoCalc were not observed. Predictions were made both at 600 °C and at the preoxidation temperature for both alloys (700 °C for OC1 and 800 °C for OC4), and the results did not correspond to the observations (see
Table 3 and
Table 4).
In general, OC1 and OC4 performed better than other iron-based alloys as the corrosion depth is less than 12 µm after 500 h of corrosion in liquid molten NaCl-MgCl
2. Iron-based alloys and sometimes nickel-based alloys usually show corrosion that is several decades of micrometres deep [
6,
15,
17]. The salt purification process also played a role in this result. Indeed, without any salt purification and if the immersion begins too soon, MgOHCl concentration cannot be neglected, leading to chlorhydric acid production by hydrolysis, as shown in
Figure 4b. One could suppose it would lead to greater corrosion as H
+ is an oxidising ion. The preoxidation and the presence of the alumina scale seem to improve the alloy’s resistance to corrosion because the alumina acts as a protective layer. It reacts with dissolved MgO, delaying the onset of the reaction between the salt and the alloy itself. Longer-term experiments would be needed to assess the protective role of the alumina scale.
In both alloys, two types of corrosion processes are observed during immersion. The first type is intragranular and is 2 µm deep on average. This intragranular corrosion is uniform for both alloys but is less pronounced for OC4. (See
Figure 13 for OC1 and
Figure 18 for OC4). As shown by the CV (
Figure 9) and ICP analysis (
Table 5) carried out after the immersion, the dissolution of iron and chromium are measured in the case of OC1 but not in the case of OC4.
No dissolved metals were measured on the CV (
Figure 9) and ICP (
Table 5) of the salt of OC4 after immersion. After pre-oxidation, the Raman signal (
Figure 7) associated with α-alumina is much more intense in the case of OC4. The matrix of OC4 could not dissolve due to a thicker α-alumina scale, as observed on the Raman spectra. For both alloys, the alumina scale formed during the pre-oxidation (
Figure 7) reacts with dissolved MgO to give a MgAl
2O
4 spinel (
Figure 14 and
Figure 19), as predicted by the E-pO
2– diagram of aluminium (
Figure 21). The pre-oxidation enhanced the corrosion resistance of OC4 alloy, contrarily to the prediction of the chromium and iron stability diagram (
Figure 23 and
Figure 24); there is no depletion of these two elements in the corroded material, as
Figure 17 shows. This can be attributed to the presence of the alumina scale that later transforms into MgAl
2O
4 and that acts as a protecting barrier, preventing the Fe and Cr from dissolution.
In both alloys, the deepest corrosion attacks are intergranular and are 10 µm deep on average. In the literature, intergranular corrosion is often observed because diffusivities of species are often higher in grain boundaries than in the alloy bulk. If the grain boundaries of the alloy contain precipitates (carbides, Ni
3Nb, etc.), two processes may occur: the precipitate at the grain boundary corrodes if it is less noble than the matrix; otherwise, the matrix corrodes around the nobler precipitate. As shown in
Figure 13 and
Figure 18, for both alloys, in the grain boundaries, the matrix is oxidised around the Nb-rich precipitates. This is coherent with the fact that niobium has a high nobility in molten salts and should form NbO
2, according to
Figure 25 [
33]. This oxide could form around the precipitate, and the less noble matrix corrodes around it. The greyish precipitates in
Figure 6d, whose composition is Ni–Fe–Al rich (Spec. 10 in
Table 2), could be the one that corrodes in the grain boundaries. In that case, Al oxidises into Al
2O
3, as predicted by the stability diagram (
Figure 21), and Fe and Ni diffuse in the matrix.
As shown on the EDX mapping in
Figure 19, Al
2O
3 is observed at the tip of the MgAl
2O
4 precipitation. Mg is not detected at this tip, whereas Cl is detected. It means that Al
2O
3 forms first by faster diffusion of O and Cl and then transforms into MgAl
2O
4 with Mg diffusion. These observations lead to the hypothesis that the corrosion rate in those two alloys is controlled by oxygen and Cl diffusion. In the case of O diffusion, the corrosion kinetics would be parabolic, following Wagner’s law. The depth of corrosion can be expressed as:
where
X is the corrosion depth in cm, t is the time of immersion in seconds;
is the oxide molar ratio, which equals 1.5 in the case of Al2O3;
NAl is the molar fraction of aluminium in the alloy;
NoDo is the oxygen permeability in cm2/s.
According to Prilleux [
34], work on oxygen permeability in iron-based alloys, the oxygen permeability at 600 °C in a Fe–20 Ni (%wt) alloy is 1.67 × 10
–15 cm
2/s for the intragranular zones. Under the hypothesis that the value is the same for OC1 and OC4 alloys that contain, respectively, 20%wt and 25%wt of nickel, the corrosion depth should be 2.6 µm, which is in good agreement with the observations of internal intragranular oxidation. For the intergranular zones, Sanviemvongsak [
35] showed that oxygen permeability is 10 times higher in grain boundaries than in the matrix for the IN718 alloy. Assuming that the same ratio exists in OC1 and OC4 alloys, the intergranular corrosion should be 8 µm deep, which is close to what is observed. From these observations and calculations, it can be assumed that the internal oxidation could be controlled by oxygen diffusion.
The corrosion mechanism can be described as follows.
MgO, dissolved in the salt, reacts with the outer alumina scale to form MgAl2O4. Meanwhile, the iron and the chromium contained in the matrix dissolve into the salt (mainly for OC1), and the intragranular internal oxidation is controlled by O diffusion in the alloy. In parallel, O, Cl and Mg diffuse in the grain boundaries. O and Cl diffuse faster than Mg, and Al2O3 forms first at the tip of the intergranular oxidation. Then, Al2O3 transforms into MgAl2O4 when Mg reaches those regions. The Nb-rich precipitates do not react with the salt as they are protected by the nobility of Nb. TEM analysis at the matrix Nb precipitate interface would determine if the precipitates are protected by Nb oxide.
In the gas phase, a small layer of salt deposits on the coupon. The salt deposited must be liquid or nearly liquid as the temperature above the molten salt is likely superior to 500 °C. The deposited salt absorbs oxygen and moisture from the sweeping argon, and then MgO deposits on the coupon. The corrosion mechanism is similar to the liquid phase in the case of OC1, with a faster diffusion of oxygen that controls the corrosion mechanism. Inter and intragranular corrosion occurs with the formation of Al2O3 that later transforms into MgAl2O4.
Under the gas phase, OC4 seems very resistant, with no intergranular corrosion and no metal depletion. This performance seems to be a consequence of its better preoxidation, as a Mg–Al–O oxide that could be protective is observed.