Multi-Analytical Characterization of Slags to Determine the Chromium Concentration for a Possible Re-Extraction

The CHROMIC project (effiCient mineral processing and Hydrometallurgical RecOvery of by-product Metals from low-grade metal contaIning seCondary raw materials) aims to recover chromium from steelmaking and ferrochrome slags to regain valuable resources and simultaneously reduce potential environmental impacts. To develop the recovery flowsheets and reliably calculate metal recovery, an accurate assessment of chromium concentration and distribution is essential. Therefore, model streams were thoroughly characterized using a combination of analytical techniques. In all materials, chromium is present in distinct but often small spinel phases, intergrown with other minerals and showing a considerable zonation in Cr content with higher amounts in the cores. The small size of the Cr-rich particles makes recovery by mineral processing challenging. Measured chromium content was found to differ largely based on the chemical dissolution method applied. The analysis of insoluble residues and comparison with a standard reference material evidenced that standard acid dissolution procedures based on HCl/HNO3/HBF4 and HNO3/HF/H2O2 are insufficient to fully dissolve spinel structures, leading to severe underestimations of chromium content. A sodium peroxide treatment is required for a full dissolution of spinels. This is noteworthy, since most of the legislation for the reuse of slags is currently based on acid dissolution methods.


Introduction
Slags are an integral part of steel and ferrochrome (FeCr) production, as they capture impurities so that high-quality metals can be produced. Steelmaking slags are commonly divided into basic oxygen steelmaking (BOS or BOF), electric arc furnace (EAF), and ladle slags based on the method of production, as detailed in [1,2]. In Europe, 18.4 Mt of steelmaking slags were produced in 2016, of which 57% are BOS, 7.3% are EAF stainless steel slags (SS), and 25% are EAF carbon steel slags (CS) [3]. For ferrochrome, slag production varies according to the type of FeCr produced. High-carbon ferrochrome (HC FeCr) produces 1.1-1.6 tons slag/ton of metal [4,5] with an annual global production of 13 Mt HC FeCr in 2018 [6]. Slag production is relatively higher for low-carbon ferrochrome (LC FeCr) (2.4-2.5 tons of slag/ton metal), but the global annual production was only 0.9 Mt in 2018 [6]. microscopy (SEM; e.g., [19]), X-ray fluorescence (XRF; e.g., [20][21][22]), optical emission spectroscopy (ICP-OES; e.g., [23,24]), as well as other methods aiming at certain material characteristics, such as thermal analysis, infrared spectroscopy, CNS (carbon, nitrogen, sulfur) analysis, and others. In this article, we compare and evaluate different techniques (SEM, mineral liberation analysis (MLA), XRD, XRF, wet chemical analysis) for their suitability to identify and quantify Cr and Cr-rich phases. This multi-analytical approach is aimed at gaining in-depth knowledge about the way in which Cr is bound into the slags, which is indispensable to design a successful recovery process.

Materials
The following input materials were received from industrial partners: • CS: Carbon steel EAF slags crushed to <5 mm • SS: Fine fraction of stainless steel slags (<0.5 mm) (SS) resulting from a demetallization process • LC FeCr: low-carbon ferrochrome slags (LC FeCr), obtained from a single slag pot, crushed, and sieved to a fraction 4-9 mm for homogenization • HC FeCr: high-carbon ferrochrome slags crushed and sieved to 0-4 mm The samples were homogenized, and representative subsamples for the analyses were obtained by quartering until the proper sample size was reached. Where necessary, additional milling was performed using a disk and ball mill (see below).

Methods for Chemical Composition
X-ray fluorescence methods: The samples were also analyzed using a high-performance energy-dispersive XRF (EDXRF) spectrometer with polarized X-ray excitation geometry (XEPOS HE, Spectro Analytical Systems, Kleve, Germany) under He atmosphere. The instrument was equipped with a 50-W tungsten end window tube (maximum 60 kV, 2 mA) and a Silicon Drift Detector. The dried and finely ground (<125 µm) samples were analyzed both as loose powder and as fused beads. For analysis as loose powder, an XRF sampling cup, provided with a 4-µm prolene foil (Chemplex, Palm City, FL, USA), was filled with the sample and subsequently placed in the autosampler of the EDXRF system. The quantification was performed using a precalibrated software package (Spectro X-LAB Pro Version 5.1, Spectro Analytical Instruments, Kleve, Germany) for (semi)-quantitative analysis of geological materials. For bead analysis, all samples were ashed at 1000 • C for 4 h in a muffle furnace (Thermolyne, Dubeque, IA, USA). About 0.9 g of ashed sample was subsequently mixed in a platinum crucible with a 100% lithiumtetraborate flux (XRF Scientific, Brussels, Belgium) in a flux:sample ratio of 10:1. The fusion was performed with an automatic fusion system (XrFuse 2, XRF Scientific, Brussels, Belgium), in which the sample was fused at 1250 • C for 11 min. An ammonium iodide tablet (XRF Scientific, Brussels, Belgium) was added as a release agent. The fused sample was removed from the furnace and poured into the pre-heated mold. During cooling, fan-forced air was applied to accelerate the solidification of the beads. For the XRF measurement, the fused bead sample was placed in the autosampler of the EDXRF system. The determination of the content of the bead samples was performed using an in-house developed calibration program.
HCl/HNO 3 /HBF 4 dissolution: The metal content in the sample was determined with inductively coupled plasma atomic emission spectrometry (ICP-OES) after acid digestion using a microwave system. About 0.5 g of sample (<125 µm) was weighed into the digestion vessel. Then, the following acids were separately added: 6 mL of HCl, 2 mL of HNO 3 , and 4 mL of HBF 4 . The digestion vessel was placed into the microwave unit (six positions, Milestone), and the following digestion process was applied: 2 min at 250 W, 2 min at 0 W, 5 min at 250 W, 5 min at 400 W, and 5 min at 500 W. At the end of the program, the vessels were cooled down to room temperature and transferred to a volumetric flask and filled up to 100 mL with ultrapure water. The digested solution was measured with an ICP-OES (Agilent 5100, Agilent Technologies Belgium, S.A., Diegem, Belgium) with an analytical error of 3% (ICP-OES EN ISO 11885, 2009-09). The insoluble residue (IR) remaining in the digested solution was recovered by centrifugation and drying at 85 • C for 24 h. A reference material (SARM 77 Ferrochrome slag, certified content of 12.5% Cr 2 O 3 (8.55% Cr) and 5.31% Fe) was analyzed for comparison. Aqua regia composition (content excluding spinel phases): 250 mg of sample was digested for 20 min in a microwave using aqua regia (HNO 3 + 3HCl). The eluates were analyzed with ICP-OES, with analytical error of 3%.
Na-peroxide dissolution: About 0.1 g of sample was mixed in a Zr crucible together with 1.5 g of Na 2 O 2 and 0.25 g of Na 2 CO 3 . The crucible was heated for 4 min in a muffle furnace at 560 • C (heating rate 8 • C/min). After cooling, the crucible was first rinsed with 40 • C milli-Q water and then carefully reacted with 12.5 mL of concentrated HNO 3 (suprapur). The resulting solution was made up to 100 mL with milli-Q water and measured with an ICP-OES (Agilent, 5100, with an analytical error of 3% (ICP-OES EN ISO 11885, 2009-09)). A reference material (SARM 77 Ferrochrome slag, certified content of 12.5% Cr 2 O 3 (8.55% Cr) and 5.31% Fe) was analyzed for comparison.

Mineralogy and Mineral Liberation
Samples for XRD were wet (ethanol) milled down to approximately 4 µm by means of a McCrone Mill (Retsch, Haan, Germany) using Y-stabilized ZrO 2 grinding media. After drying, the samples were homogenized and prepared in a 27-mm sample holder. Measurements were conducted in a PANalytical Empyrean diffractometer (Malvern Panalytical, Kassel, Germany) equipped with a PIXcel 3Dmedipix area detector, an automatic divergence slit to provide constant irradiated area (10 × 12 mm 2 ), and a Co tube. Quantification was performed via the Rietveld method using the BGMN/Profex software package (v 3.12.0) [25]. X-ray powder diffraction on the IR after dissolution was carried out with a PANalytical Empyrean system operated at 40 kV and 45 mA, with Co tube (fine focus, λ = 1.7903 Å). Continuous scans with a step-size rate of 0.013 • /49.725 s were performed within a 2θ-range of 5 • -120 • (2D detector). The obtained diffractograms were qualitatively analyzed with the aid of HighScore Plus software (version 4.6a, Malvern Panalytical B.V., Almelo, the Netherlands).
For SEM-EDX (Scanning Electron Microscopy-Energy Dispersive X-ray Spectroscopy) analysis, the samples were embedded in a low-viscosity epoxy resin and gradually polished down to 1 4 -µm diamond powder grit size. The samples were coated with Pt-Pd prior to microscopic analysis.
Mineral liberation analysis (MLA) samples include two grain mounts of each slag sample, each prepared in epoxy resin, one cut after solidification perpendicular to the mount and prepared again in 25-mm mounts. This procedure compensates for gravimetric effects and therefore distortion [26]. Samples were carbon coated. An FEI Quanta 650F MLA-FEG system (FEI MLA Software Suite 3.14; FEI, Hilsboro, OR, USA) was used for the subsequent investigation. Table 1 shows the main components of the materials, as an average of the different XRF measurements methods. Overall, for the major elements, the difference between measurements is limited (standard deviation is low). The results show that the chemical composition of LC FeCr and SS is quite similar, with Ca, Mg, and Si as the main elements. CS has a much higher Fe 2 O 3 content and lower CaO and SiO 2 content, whereas for HC FeCr, the main elements are Al, Mg, and Si. Chromium concentration is also significantly higher in HC FeCr than in the other materials. Table 1. Main components (average ± standard deviation, in wt %, as measured by X-ray fluorescence (XRF) analyses).

Mineralogy
Results of the XRD investigation for CS, LC FeCr, HC FeCr, and SS are compiled in Table 2. Depending on the specific slag, the modal mineralogy and the contents differ quite a lot. Especially the spinel contents-as well as their chemical composition-differs from slag to slag. The dominating spinel in the CS slag is a Mg-Fe-containing spinel (magnesioferrite, MgFe 2 O 4 ) with 17 wt % content, followed by magnetite (Fe 3 O 4 ) (~5 wt %) and minor contents of magnesiochromite (MgCr 2 O 4 ) (~3 wt %). In contrast, the spinels of the LC slag contain no or very little Fe, and the main spinel phase is an Mg-rich chromite (MgCr 2 O 4~7 wt %). The same appears for the SS slag with MgCr 2 O 4 being the main spinel phase (~6 wt %). In the HC FeCr slag, both a dominant chromium-rich ((Fe,Mg)Cr 2 O 4 , 35 wt %) and a magnesiospinel (Mg,Fe)(Al,Cr) 2 O 4 can be found. The high MgO values (see Table 1) of the HC FeCr slag can be explained by the high percentage of Mg silicates (forsterite and enstatite).
Back-calculated chemical information shows good correlation to the chemical compositions of at least the major elements, proving the consistency of the data. Only the stainless steel slag reveals a slight underestimation of Al-rich phases (sum 2.6 wt % Al 2 O 3 ) and a corresponding overestimation of CaO (sum 49.0 wt %). The most likely reason is that phases with varying composition (spinels, gehlenite-akermanite) lead to a peak broadening that is compensated by different physical effects (line broadening due to stress), but does not reflect the actual chemical diversity (leading to overlapping peaks, with similar resulting peak shape). This could be overcome by using a respective number of correct structure models-which in turn needs the correct chemical composition for each single one of the individual minerals. Since the quantification of mineral (groups) led to comparable results and other main components (SiO 2 30.0 wt % and MgO 10.0 wt % from XRD, respectively, cf. Table 1) fit quite well in our case, we considered these mineralogical data as close to perfect, even without the corresponding chemical accordance for Al and Ca.

SEM and MLA
While XRD yielded valuable information about the mineralogical composition, it is not possible to obtain a detailed view on Cr distribution, as the Cr-containing phases are only a minority among the different minerals present, and are therefore difficult to isolate. SEM-based techniques were applied to gain a deeper insight into the Cr distribution in the phases (mainly in the spinels) and the characteristics of the grains themselves.
SEM analysis confirms the findings of XRD that the main phases of the CS slags are Ca(Al) silicates (larnite, gehlenite) and Fe oxides, including Fe-based spinel phases (magnetite, Fe-Mg spinel), which are difficult to distinguish from each other. The minerals are intergrown at small scale (<100 µm), with the Cr-rich spinels being 10-100 µm in size (Figure 1). Figure 2 shows that the chromium is unevenly distributed within the spinels, with clear enrichment in distinct zones.    As shown in Figure 4, HC FeCr metal particles are well liberated and typically larger than in the LC FeCr sample (up to 500 µm), although smaller droplets are also present. Besides the Fe-Cr metal particles, chromium is also present in different spinel forms, as confirmed by XRD. The chromium content differs between regions in the metal ( Figure 4) and spinel ( Figure 5) particles.  SEM-EDX analysis of the SS samples shows that the matrix consists of CaMg silicates (merwinite, bredigite), Ca silicates (larnite, cuspidine), and CaMgAl silicates (akermanite) (Figure 6 combined with Table 2). Overall, the material contains many very fine particles, but also some composite grains. Small metallic stainless steel particles are present, which are mostly <5 µm in size. Chromium is present within these metallic particles, and also in distinct magnesiochromite spinel phases (10-30 µm). MLA (mineral liberation analysis) gives the opportunity to determine particle size distributions of both whole sample and prechosen phases (grains), and additional grain characteristics such as liberation, association, etc. Figure 7 illustrates the assignment of a backscattered electron (BSE) image to the false-colored phase-distinguishing image of mineral liberation analysis, derived from the combination of image analysis (grey values) and EDX spectra. Chromium-rich phases in the particles can be clearly determined. Figure 8 gives the summarized grain size distribution of Cr-rich phases. The Cr-rich phases (mainly Cr-rich spinel phases) in the different samples reveal quite similar and small grain sizes, with the majority of Cr particles below 100 µm. The only exception is the HC FeCr, which has significantly larger particle sizes. In general, the results from SEM analyses were confirmed, and the grain size distribution for all Cr-containing phases were achieved from the MLA data (see Figure 7).   Table 3 compares the Cr content of the materials as determined by the different wet chemical analyses. Although the individual methods show good reproducibility (low standard deviation), there is clearly a significant difference in the amount of Cr that is liberated. The Na-peroxide dissolution method, commonly applied in the ferrochrome industry for the analysis of Cr ores and FeCr slags, gives the highest recovery, with 91% of Cr recovered from the SARM 77 reference material. This method is not regularly applied in analytical laboratories outside the ferrochrome industry, and is not well known for the analysis of secondary resources.

Chromium Quantification
Moreover, legislation often refers to other, more traditional methods of destruction. For example, the HCl/HNO 3 /HBF 4 dissolution method is the standard destruction method foreseen in the "Compendium for Sampling and Analysis" that must be used in the Flemish Environmental Legislation.
Therefore, it is highly relevant to evaluate their performance. It is clear that both these methods (HCl/HNO 3 /HBF 4 and HNO 3 /HF/H 2 O 2 /H 3 BO 3 ) only liberate a portion of the Cr that is present, although both are considered to represent the "total content".
Aqua regia dissolution is known to liberate only metals not included in spinels [27]. Based on the aqua regia results, it is apparent that only a minor fraction of the Cr (10-20%) is present in minerals outside spinels, part of which may be the metallic fraction.
To better understand the underestimation by the acid dissolution methods, the insoluble residues (IR) that remained after the HCl/HNO 3 /HBF 4 method (about 20 wt % for CS, LC FeCr, and HC FeCr, 30 wt % for SARM 77 and only 1 wt % for SS) were recovered and analyzed with XRD.
The XRD analyses showed that the IRs consist of spinels ( Figure 9). From the different peak positions of spinels, as illustrated in Figure 9, it is clear that the spinel composition differs between the materials, in accordance with the spinels identified in the original bulk samples. For the HC FeCr, two types of spinels are present, the most predominant being a Cr-free spinel (spinel, (Al 2 MgO 4 )) with a smaller fraction of chromite (Cr 2 FeO 4 ). The peak pattern of the reference material SARM 77 is quite identical, as expected, since this is the same type of material. For the LC FeCr, broad peaks are present, indicating that both spinel types occur (but with lower contribution of the Cr-free spinel) as well as different compositions in between. For the SS slags, magnesiochromite is identified, whereas in the CS, an intermediate composition of (Al,Cr) 2 (Fe, Mg)O 4 is found.

Discussion
The detailed multi-analysis characterization has given valuable insight into the speciation of Cr inside the materials, with implications for the potential recovery. Table 4 summarizes the advantages and disadvantages of each method, and the main characteristics that can be deduced from the single method for the slags. Wet chemistry methods give the total amount of Cr that is present, and thereby the inherent potential for recovery. However, they do not give any insight in the way that this Cr is distributed, and how accessible it is. A first step to gain deeper insight is XRD, in order to define the mineralogical phases that are present. From this analysis, it was clear that chromium is present in all materials in spinel phases, but the exact composition and quantification is complicated by overlapping peaks in the XRD patterns. Furthermore, XRD is a bulk method, and does not give any information on spatial distribution or particle size, and requires phases to be present in rather high amounts to be identifiable in the diffractogram. The SEM-EDX analysis gave additional insights, such as the occurrence of metallic inclusions and considerable zonation in Cr content with higher amounts in the spinel core. It also demonstrated the size of the relevant particles, but only on a spot-wise basis. MLA was used to define the grain size distribution for the full sample, divided into different classes. Apart from providing a different level of information, results from one method can also be used to clarify results from the other. For example, EDX spectra combined with XRD results and expertise on mineral morphology were used to identify the mineral phases observed in SEM. Backcalculation of the chemical composition (from wet chemistry) was needed to verify the XRD results. The uneven distribution of Cr within the Cr-rich phases (see e.g., Figure 2), observed by SEM, explains the peak shift and overlapping characteristics of the respective phases in the XRD patterns. This gradually changing chemical composition within single phases also hampers the correct quantification of the Cr content in MLA, as it would require the inclusion of a large number of phases to the mineral list (each for one possible composition in the solid solution). A verification of the MLA data of slags with those restrictions falls out of the scope of this study.
The small size of the Cr-rich particles makes recovery by mineral processing challenging, as a high degree of comminution is needed to reach full mineral liberation. Often, wet separation techniques need to be employed, incurring additional costs for water treatment. Only for the HC FeCr slags is the size of the metallic particles sufficient for dry separation.
A reliable quantification of Cr in the materials before and after recovery is indispensable to calculate recovery efficiency, as well as determine the potential applications of the cleaned-up residue. However, the results of the wet chemical analysis show that many dissolution methods that are commonly used and required by legislation e.g., in Flanders and Germany, are not able to recover all Cr. The main reason for this is that Cr is mostly present in spinels, which are highly stable phases that are very resistant to attack. The XRD analyses of the insoluble residues after HCl/HNO 3 /HBF 4 dissolution confirmed the presence of spinels, thereby supporting the hypothesis that Cr content is underestimated by incompletely dissolving the spinel phases. The degree of spinel attack can differ strongly based on the dissolution method and spinel composition, as evidenced by the remarkably better performance of the HNO 3 /HF/H 2 O 2 /H 3 BO 3 method for CS slags than for the other materials. A possible explanation is that the addition of H 2 O 2 as oxidant manages to attack the Fe spinels present in these slags, but not the Mg spinels, which are the predominant species in the other slags.
The spinels' high resistance to attack has major implications for the recovery of Cr from slags, since it is clear that aggressive methods of attack are needed to liberate Cr. Furthermore, the question can be asked of whether methods of analysis based on acid dissolution are relevant to determine Cr content in slags. This issue, especially for chromites, has been discussed for a long time [28][29][30]. Already in 1972, Rodgers [30] compared different methods for chrome-rich spinels, stating that their refractory nature and high concentration of Fe and Cr makes it difficult to obtain reliable analytical results. He concluded that peroxide fusion is the best dissolution method for chromite, while the widespread perchloric acid digestions are deficient. The peroxide fusion is commonly applied in the ferrochrome industry, with companies such as PerkinElmer providing dedicated equipment. However, for the reuse of slags, analysis is commonly done according to the standard methods prescribed in environmental legislation, which rely on acid digestion (in the case of Flanders and Germany, even without perchloric acid). Therefore, it is important to be aware that the chromium content obtained with these methods may be an underestimation. Nevertheless, even with the preferred peroxide fusion, it was not possible to recover more than 91% Cr of the SARM 77 reference material.
Overall, it can be concluded that a combination of analytical techniques is needed to gain a complete view on the material's characteristics, such as the chemical composition, liberation, and grain-size distribution, and is crucial to further develop processing schemes. Good insight is needed to avoid wrong conclusions based on unsuited analytical techniques. It is indispensable to cross-check and validate the used methods in order gain maximum information for any further treatment and increase its efficiency.