Purification strategy and effect of impurities on corrosivity of dehydrated carnallite for thermal solar applications

This paper presents a purification method for dehydrated carnallite (DC)—a commercial ternary MgCl2–KCl–NaCl salt—for concentrating solar power (CSP) applications based on a thermal and chemical treatment using the reduction power of Mg. The purification is effective at reducing MgOH+ by an order of magnitude—from around 5 wt% in non-treated salt to less than 0.5 wt% in post-purification salt. The corresponding decrease in the measured corrosion rate of Haynes 230 at 800 °C from >3200 μm per year to around 40 μm per year indicates that soluble MgOH+ is indeed correlated to corrosion. The addition of elemental Mg serves as both a scavenger of impurities and corrosion potential control, which are considered the primary mechanisms for corrosion mitigation.


Introduction
To enable the use of a supercritical carbon dioxide (sCO 2 ) Brayton power cycle, next-generation (Gen3) concentrating solar power (CSP) requires a heat-transfer uid (HTF) and thermal energy storage (TES) medium that can operate in the temperature range of 500-750 C. This operational temperature differs from the current Gen2's molten nitrate HTF and TES medium, which is stable up to $565 C. Gen3 CSP's higher operating temperature range demands new chemistry with high thermal stability and low corrosion on metallic materials used in the receiver, primary heat exchangers, piping, and TES tanks.
Dehydrated carnallite (DC)-a commercial ternary MgCl 2 -KCl-NaCl chloride-salt-has been proposed given its thermal stability up to $800 C, low cost, and availability. The liquidus temperature of the DC salt is also around 440-450 C, which also falls into the operating temperature range of the sCO 2 Brayton cycle. However, the corrosion behavior of the ternary salt is less known. The hygroscopic nature of the MgCl 2 component complicates the investigation because dehydration of the MgCl 2 -containing salt is not as simple as heating up the mixture to different temperatures to remove different forms of MgCl 2 hydrates. The hygroscopic nature of MgCl 2 has been known for a long time in the literature. [1][2][3][4][5][6][7][8][9] Stepwise dehydration of MgCl 2 $6H 2 O to form lower hydrates can be achieved by heating MgCl 2 $6H 2 O to different temperatures as shown by eqn (1a)-(1d). 4 However, hydrolysis reactions during dehydration to form MgOHCl and HCl, as shown in eqn (2a) and (2b), occur simultaneously in a similar temperature range (of >240 C) as dehydration: [3][4][5]8,10,11 MgCl 2 $2H 2 O ¼ MgOHCl + HCl(g) + H 2 O (2a) It is very challenging to control the formation of MgOHCl during dehydration in a normal atmosphere. Dry HCl gas with partial pressure exceeding values calculated by Kipouros and Sadoway 4 can inhibit the hydrolysis reactions based on Le Chatelier's principle. During production of the chloride salt electrolyte for Mg production at Israel Chemicals Ltd (ICL), Cl 2 gas is bubbled through the molten DC in a process used by ICL for controlled dehydration. At high temperatures of 533-555 C, MgOHCl thermally decomposes to form MgO and HCl gas as shown in eqn (3). 4,12 From eqn (2) and (3), both the formation and thermal decomposition of MgOHCl lead to formation of highly corrosive HCl gas. Combined with the H 2 O released during dehydration, HCl can be detrimental to most alloy components. In addition, dissolved MgOHCl in molten chlorides, in the form of MgOH + , is also known to corrode alloys, as given by eqn (4a) and (4b), 13 where M can be a pure metal or a metal component in an alloy such as Fe, Cr, or Mn: Therefore, an effective method is needed to remove MgOHCl from the chloride salt before performing corrosion.
By leveraging the experience from the magnesium production industry, which has been using similar MgCl 2 -containing salt for decades, and from past research in the literature, 4,14 we designed a thermal and chemical purication procedure that is effective at removing MgOHCl and other cationic impurities in the commercial salt. 100 hour corrosion evaluation at 800 C was performed on Haynes 230, which is of interest to Gen3 CSP technology to serve as the solar-receiver material because of its high temperature stability and chemical resistance. DC salts subjected to different purications were used for corrosion evaluation while their pre-corrosion MgOHCl content was tracked by an analytical titration technique. 15 The objectives of our study are to (1) understand the corrosiveness of the ternary chloride salt, (2) investigate the effectiveness of each purication process, (3) verify that MgOHCl is indeed correlated to corrosion, and (4) design future corrosion mitigation strategies.

Materials
The salt used for corrosion was "dehydrated" carnallite supplied by ICL. The primary chemical constituents of DC are KCl and MgCl 2 in a chemical formula of KMgCl 3 , whose mineralogical name is carnallite. There is an additional 5-7 wt% of NaCl in the DC based on the mineralogical resource from the Dead Sea where ICL is located. Table 1 shows the elemental composition of DC provided by ICL. Note that DC salt contains roughly 5 wt% water (in the form of chemical hydrate and/or physically absorbed moisture). Once received, the DC salt was stored in an MBraun glove box under nitrogen with <0.5 ppm H 2 O and <0.5 ppm O 2 .

Salt purication
The dehydrated carnallite salt was puried to remove H 2 O and impurities before being used for corrosion with Haynes 230 coupons at 800 AE 10 C for 100 hours. Salt purication consists of two separate parts: a thermal purication and a chemical purication. The thermal purication follows the principles of Kipouros and Sadoway, 4 which provides the guideline for a stepwise dehydration process at 117 C for 8 hours, 180 C for 8 hours, 240 C for 2 hours, 400 C for 1 hour, and 600 C for 1 hour, with a heating rate of 5 C min À1 between the isothermal steps. The chemical purication follows the principles of using an active metal such as Mg or Zr to remove MgOHCl impurity as given by the German Aerospace Center (DLR) and Savannah River National Laboratory (SRNL). 13,16,17 Following SRNL's suggestion based on their investigation, we used 1.7 wt% of elemental Mg chips (99.98% trace metals basis, 6-35 mesh, Sigma Aldrich). 1.7 wt% of Mg was a conservative estimation to ensure Mg is in excess. Investigation of a more accurate amount of Mg addition was determined in a follow-up study. Aer chemical purication, excess Mg, in the form of droplets at the bottom of the solidied salt, was removed.

Corrosion setup
Haynes 230 supplied by Haynes International was cut by a waterjet cutter into coupons with dimensions shown in Fig. 1. The coupon surfaces were lightly polished with 120-grit sandpaper in water prior to corrosion tests. A 250 mL Ni crucible (Sigma-Aldrich Z246581) and a Ni crucible cover (Sigma-Aldrich Z245700) were used as the corrosion vessel. The crucible and cover were used as received without special treatment. Three Haynes 230 coupons were used in each corrosion test. A 4 mm-diameter hole was drilled at the center of the Ni crucible cover to form a mechanism to hang the Haynes coupons on a Ni wire (99.98%, 1 mm diameter, Goodfellow NI005171) into the molten chloride salt. The schematic in Fig. 2 shows the details of the corrosion vessel setup. We used a combination of alumina tube (3.08 mm outer diameter, 1.53 mm inner diameter, CoorsTek), alumina plate (2 mm thickness, CoorsTek), and quartz disc (25.4 mm diameter with a 3.2 mm-diameter center hole, AdValue Technology FQ-D-1N-N1/16) to prevent galvanic coupling between Haynes 230 coupons and the Ni crucible/cover. The spacing between Haynes 230 coupons was about 1 cm to allow ample space to minimize mass-transfer limitation during corrosion testing.
About 200 g of puried DC salt (aer removal of excess Mg droplets) were placed into the Ni crucible. This amount of salt ensured full submersion of the Haynes 230 coupons in the molten salt during corrosion. The corrosion setup was then assembled according to Fig. 2. The assembly was next placed into a stainless- The entire assembly was completed in the glove box under ultra-highpurity (UHP) nitrogen atmosphere and then transferred into the furnace test vessel within a few minutes. The furnace test vessel was sealed from ambient atmosphere with a grafoil gasket. The furnace test vessel was pumped down to À0.08 MPa and relled with nitrogen gas (UHP, Airgas) at least three times to remove oxygen and moisture, aer which UHP nitrogen gas was owed continuously at 150 sccm during corrosion. The heating schedule shown in Table 2 was used for the corrosion. Post-corrosion sample preparation: aer corrosion testing, the assembly was transferred to the glove box and the corroded Haynes 230 coupons were carefully retrieved aer breaking the solidied salt. Next, the Haynes 230 coupons were taken out of the glove box, immersed in deionized (DI) water, and ultrasonicated for 15 min to remove residual salt. Aer ultrasonication, the coupons were rinsed with ethanol, dried, and stored in a desiccator. The coupons were cut by a diamond-blade saw about 6-7 mm from the short edge. The cut surface was mounted in phenolic resin and polished to a mirror nish for metallographic characterization with scanning electron microscopy (SEM) and energy-dispersive spectroscopy (EDS).

Corrosion rate calculation
The corrosion rate C 1 dened as the loss of metal surface thickness per unit time (e.g., mm per year) is given by eqn (5) based on the initial metal-coupon surface area (A), the weight change of the metal coupons during corrosion test (DW), corrosion time (t), and metal density (r) assuming that uniform corrosion occurred on the Haynes 230 coupons and that there was no change of corrosion mechanism(s).

MgOHCl content measurement
An analytical titration technique was developed at NREL. 15 This titration method used the different solubilities and chemical reactions of chloride salt components, MgOHCl and MgO, in DI water and methanol to have successful physical separation of each species, followed by an ethylenediaminetetraacetic acid (EDTA) titration to determine Mg 2+ content in each species. The method was able to detect MgOHCl content down to at least 0.1 wt% with standard deviation on the order of 0.01 wt% as reported by NREL. 15  180 C, 240 C, and 400 C), followed by a sharp drop at the last isotherm of the thermal purication (i.e., 600 C). The MgOHCl content further drops during the chemical purication (i.e., with 1.7 wt% of Mg). It should be noted that multiple measurements were performed on non-treated DC salt, DC salt aer heat treatment at 117 C for 8 hours, and thermally + chemically puried salt (i.e., 1.7 wt% of Mg). Our major objective is to detect the decrease of MgOHCl content by the thermal + chemical purication, so we focused our titration effort on the thermally/chemically puried salt sample (1.7 wt% of Mg in Fig. 3). We did not perform multiple titrations on every salt sample because the existing data suggested that the standard deviation of the titration measurements was at a fairly small level of less than 0.5 wt%. Nevertheless, we acknowledge that more measurements should have been performed to have standard deviation information. Hence, only qualitative (instead of quantitative) analysis of the data is given in the following Discussion section.

MgOHCl content
Corrosion rates Fig. 4 shows the corrosion rates of Haynes 230 coupons in molten DC salts at 800 AE 10 C for 100 hours aer different purications were performed on DC salt following the procedures outlined in the Experimental section. The standard deviation reported in Fig. 4 was based on at least three coupons for each salt treatment (i.e., non-treated, thermal, and thermal + chemical). Positive corrosion rate is dened as mass loss. Experimental corrosion tests were attempted on non-treated DC, but no results were obtained successfully because the extremely high corrosivity of the non-treated DC caused substantial damage of the experimental setup, e.g., leakage of the furnace vessel. Hence, we did not believe that the results could be used. Instead, the corrosion rate in non-treated DC salt could only be estimated based on the fact that the maximum amount of MgOHCl that the non-treated DC can generate (i.e., 8-9 wt% aer 400 C in Fig. 3) is more than 3 times that for the thermally puried DC (i.e., $2.5 wt% aer 600 C in Fig. 3). If we assume that corrosion is roughly linearly correlated to the amount of MgOHCl, then, for rst approximation, it is estimated that non-treated DC will produce a corrosion rate of 3200 mm per year or above. It should be noted that corrosion rates based on conversion from mass loss instead of direct measurement of corrosion thickness is used here because (1) intergranular corrosion is commonly seen in metals corroded by molten chlorides which makes it difficult to dene such thickness, and (2) the high corrosion rates of "Thermal" in Fig. 4 should not support a build-up layer of corrosion products based on an unpublished study on similar molten chlorides by the same authors. However, for low corrosion rates such as the one reported for "Thermal + Chemical", it should be cautious to solely rely on mass losses to determine corrosion rates. We acknowledge that there can be errors associated with the growth of corrosion products at the corrosion interface. Our best estimated error for the "Thermal + Chemical" is up to 40 mm per year based on the most likely type of corrosion product (e.g., MgO) and its thickness (e.g., less than 5 mm).

Salt chemistry analysis
Inductively coupled plasma (ICP)-mass spectroscopy (MS)/ optical emission spectroscopy (OES) was used to measure the elemental composition of the cationic impurities present in the non-treated, thermally puried, and chemically puried DC salts (Table 3). Fe, Cr, Mn, and Ni were found to be the major cationic species, in addition to the major salt components of Mg, K, and Na (see Table 1). The increase of Fe, Cr, Mn, and Ni (i.e., major metallic components of stainless steel) aer thermal purication is attributed to contamination from the stainlesssteel furnace test vessel because corrosion products between salt vapor/HCl vapor and furnace test vessel-especially those formed on the test vessel cap directly above the purication crucible-could drop back into the salt, thus causing contamination. To clarify, salt purication did not use the same setup as the corrosion setup (shown in Fig. 2). An open crucible without a cover was used. Thus, contamination could have occurred and contributed to the increase of Fe, Cr, Mn, and Ni.

Discussion
The decrease of MgOHCl content from $5 wt% in non-treated salt to $2.5 wt% in thermally puried salt and to <0.5 wt% in both thermally and chemically puried salt, as shown in Fig. 3, agrees with the decrease of corrosion rate in the corresponding salt, as shown in Fig. 4. This suggests that MgOHCl can be used as a good indicator of the corrosiveness of the molten salt. It has Fig. 4 Corrosion rates of Haynes 230 coupons in molten DC salts at 800 AE 10 C for 100 hours following different purification treatments. Corrosion rate in non-treated DC salt was estimated at 3200 mm per year or above. Given the similarity in salt condition and nal corrosion rate, it is expected that intergranular corrosion has also been effectively controlled in this work. Therefore, for Gen3 CSP that aims to use MgCl 2 -containing molten chlorides as HTF and TES to reach a desired 30 year lifespan, it is critical to purify the molten chlorides to minimize MgOHCl content as well as the detrimental intergranular corrosion.

Effect of thermal purication
Stepwise heating of DC salt at the specied isothermal temperatures is proven to be effective at purifying the salt given the drop of MgOHCl content from a maximum of 8-9 wt% in non-treated DC to $2.5 wt% in thermally puried DC (i.e., aer 600 C), as shown in Fig. 3. (Note that the initial MgOHCl was $5 wt% in non-treated DC; but the increase of MgOHCl due to hydrolysis of MgCl 2 during heat treatment must also be taken into account.) This 3-to 4-fold drop in MgOHCl content leads our hypothesis that a signicant reduction in corrosion should occur-by comparing the estimated corrosion of non-treated salt to that of the thermally puried salt.
The variation of MgOHCl content during thermal purication also qualitatively corroborates the literature understanding of the chemical changes occurring in hydrated MgCl 2 system. Hydrolysis of anhydrous MgCl 2 or MgCl 2 $2H 2 O to form MgOHCl can occur in the temperature range of 210-445 C. 8,10 Our measured MgOHCl content indeed shows two increases at the isothermal temperature of 240 C and 400 C, which suggest that some MgCl 2 molecules in DC salt are in the form of MgCl 2 $2H 2 O and MgCl 2 $H 2 O because 240 C and 400 C are the corresponding dehydration temperatures, respectively. The variation among non-treated DC at 117 C and 180 C may be attributed to inherent inconsistency of salt samples considering the magnitude of the available error bars. The decrease of MgOHCl at 600 C is the result of MgOHCl thermal decomposition, which is shown to occur at a temperature as low as 415 C. 8,10 The thermodynamically predicted onset temperature for MgOHCl thermal decomposition is 555 C by Kipouros and Sadoway 4 and 568 C as calculated by FactSage. 22

Effect of metallic Mg during chemical purication
Addition of active metals such as Mg or Zr can reduce MgOHCl, as proposed by SRNL. The Ellingham diagram in chloride systems 23 (Fig. 5) gives the Gibbs free energy of formation DG f for different cations, which predicts the following: to remove the major metallic impurities such as Fe cations in the DC salt, one needs to use a metal that forms a more stable cation in the chloride system (i.e., with a more negative DG f ). At the same time, the metal cation cannot be more stable than the major constituents of the salt (i.e., Na + , K + , and Mg 2+ ); otherwise, we will lose these salt components (i.e., metallic Ca and Li should not be used). Fig. 5 then predicts that Zn, Zr, and Mg are good candidates for this purpose from the thermodynamic perspective. Mg was selected over Zn and Zr because it does not introduce other elements into the ternary MgCl 2 -KCl-NaCl salt system. As shown in Table 3, the concentrations of Fe, Cr, Mn, and Ni decrease signicantly in the thermally/chemically puri-ed salt compared to thermally puried salt. Note that Fe, Cr, Mn, and Ni cations are all less thermodynamically stable than Mg cation, as shown in Fig. 5. Although the absolute magnitude of Fe, Cr, Mn, and Ni aer thermal/chemical purication was not as low as expected (primarily due to contamination, as explained in the Experimental section), these substantial decreases indeed prove that elemental Mg is able to remove these cations from the salt. A follow-up study (publication in preparation) showed that the concentrations of Fe, Cr, Mn, and Ni can be reduced to less than 10 ppm. In a separate purication experiment using an alternative chloride salt with Ca and Li impurities, chemical purication with Mg is not able to reduce their content, which also agrees with thermodynamic prediction.

Metallic Mg as corrosion potential control
Metallic Mg is known to provide cathodic protection to metal alloys in molten chlorides where Mg serves as a sacricial anode. 16,17,19 It is generally accepted that Cr leaching is the major corrosion mechanism of Cr-containing alloys in molten chlorides. 16,17,19,20,24 Past work from SRNL 16,17 showed that without Mg, the corrosion potential of a Fe-Ni-Cr alloy is higher than that of CrCl 2 such that Cr in the alloy will leach and form CrCl 2 . Addition of Mg shis the corrosion potential of the alloy by more than 0.5 V to be lower than that of both CrCl 2 and CrCl 3 -meaning that Cr in the alloy is thermodynamically stabilized. Also, adding elemental Mg into molten chlorides during corrosion testing also signicantly reduces the corrosion current-by two orders of magnitude-of the reaction Cr + 2Cl À / CrCl 2 + 2e À (6) which represents the major mechanism of Cr corrosion in Crcontaining alloys. In this case, the oxidation reaction of metallic magnesium Mg / Mg 2+ + 2e À provides the majority of the anodic current (rather than that provided by the reaction given by eqn (6)), which balances the cathodic current provided by the following reaction: The anodic current of reaction in eqn (6), which is directly related to the corrosion of Cr in metal alloys, is lowered; so, addition of elemental Mg should reduce Cr leaching.

Metallic Mg to facilitate oxygen scavenging
In addition to its use as potential control of corrosion, elemental Mg has a second use to promote oxygen scavenging. Based on eqn (4b), elemental Mg can reduce MgOH + and form MgO and H 2 : The fundamental cause of MgOH + (or MgOHCl) formation is the reaction of MgCl 2 with water (and oxygen) in the chloride salt as shown by eqn (2a) and (2b). These soluble MgOH + ions then carry the oxygen atoms from water and become reactive with components in the alloy such as Cr, Mn, and Fe (eqn (4b)). Aer purication, MgO carries over the oxygen from MgOH + as one of the purication products. The low solubility of MgO in molten chloride (<0.2 wt% 25,26 ) suggests that these oxygen atoms, once carried by MgO molecules, should be a minor source of corrosiveness. Eqn (9) then suggests that elemental Mg promotes the reaction of oxygen scavenging during the initial salt purication stage by transforming the most detrimental oxygen-containing species in the molten salt, i.e., soluble MgOH + , to MgO. It also suggests that the presence of Mg in the molten salt is important for continued protection of the metal alloys against corrosion as oxygen and moisture ingress (e.g., through anges and valves) can form more MgOH + , which is expected during normal CSP plant operation.
The photo in Fig. 6 shows a dark sludge phase formed at the bottom of the chloride salt aer chemical purication with Mg. The sludge was dissolved in DI water to remove residual chloride salt and ltered through a 450 nm lter paper. The collected particles were analyzed by X-ray diffraction. Fig. 7 shows that the collected particles in the sludge phase were identied as predominantly MgO [27][28][29] ). Only two peaks around 2q of 43-45 and one minor peak around 2q of 51 cannot be attributed to MgO. The XRD result conrms the effectiveness of Mg to promote oxygen scavenging as shown in eqn (9) because reduction of MgOHCl by Mg is the major chemical pathway to form a signicant amount of MgO particles. Thermal decomposition of MgOHCl to form MgO (eqn (3)) alone is not able to produce such a large quantity of MgO because the amount of sludge produced aer thermal purication was found to be smaller. MgO formation also indicates that Mg is a consumable during the purication process. Therefore, in the future, we need to investigate how much and how frequently Mg should be replenished if oxygen and moisture ingress in the CSP plants is inevitable.
Amount of Mg added during chemical purication 1.7 wt% of metallic Mg used during chemical purication (the specic amount was provided by SRNL to NREL) is  a conservative amount to ensure that Mg is in absolute excess during purication and that most MgOHCl can be reduced to minimize the corrosiveness of the salt during corrosion testing. No optimization of the Mg amount has been performed. The analytical titration method measured that the MgOHCl content was around 2-3 wt% aer thermal purication and before chemical purication (see Fig. 3). The stoichiometry in eqn (9) predicts that for 2-3 wt% of MgOHCl, roughly 0.31-0.47 wt% of Mg should be used to ensure complete reduction of MgOHCl. Therefore, an excess of Mg is expected if using 1.7 wt% of Mg. Fig. 6 also proves that 1.7 wt% of Mg is in excess because droplets of Mg are clearly observed in the sludge phase. Therefore, Mg addition needs to be optimized in terms of amount, form factor, and temperature at which addition should occur. This optimization will minimize materials and processing costs because Mg is signicantly more expensive than the salt at >$4000/MT. For CSP to be cost competitive, it is essential to minimize the amount of Mg used during initial purication and/or plant operation.

Conclusions
This work conrmed previous literature understanding of the complex chemistry of MgCl 2 -containing salts and its dehydration strategy. We leveraged such knowledge and demonstrated experimentally that the corrosion rates of Haynes 230 in a MgCl 2 -KCl-NaCl ternary chloride salt can be signicantly reduced from >3200 mm per year (estimated) to about 40 mm per year by (1) a thermal purication process as well as (2) a chemical purication process with Mg as corrosion potential control and oxygen scavenger. As shown by the correlation between MgOHCl content in the salt and corrosion rate at different stages of purication, MgOHCl, or soluble MgOH + , is primarily responsible for corrosion. Therefore, the corrosion mitigation method using the thermal and chemical purication process is effective because it reduces the MgOHCl content by an order of magnitude to prevent corrosion due to formation of HCl gas and soluble MgOH + ions.
In the future, we plan to optimize the purication process, which aims at (1) reducing the total amount of elemental Mg added to lower the cost at industrial scale and (2) combining the thermal and chemical purication processes to reduce overall energy use.

Conflicts of interest
There are no conicts of interest to declare.