Steel Slag Characterisation—Beneﬁt of Coupling Chemical, Mineralogical and Magnetic Techniques

: Steel-making slag is largely used today in road construction and other applications, but signiﬁcant volumes are landﬁlled and cannot be recycled for excessive contents in hazardous metals, such as chromium or vanadium. The long-term behaviour of this material is still little known, and the characterisation of large volume slag dumps remains an environmental challenge. In this study various analytical techniques are used to characterise Electric Arc Furnace (EAF) slag landﬁlled for several decades and exposed to chemical weathering and erosion. Coupling chemical, mineralogical and magnetic techniques helps to understand the relations between hazardous metals and mineral phases. A special interest is given to Fe-bearing minerals microstructure so as to link the magnetic properties of the material to its mineralogical composition. The studied slag presents high amounts of chromium (between 1 and 3 wt. %) and very high magnetic susceptibility values (near 60 × 10 − 6 m 3 / kg), explained by the presence of magnetite and a spinel solid solution. Some correlations are found between magnetic susceptibility and potentially hazardous metals, providing new perspectives for future environmental investigations.


Introduction
More than 400 million tonnes of iron and steel slag are produced each year in the world [1,2]. In Europe, about 18 million tonnes of steel slag were produced each year from 2000 to 2016 [3]. An important part of this slag can be recycled as aggregates for road construction, as additives in cement and concrete production, as fertilizers and various others uses, but between 1 and 4 million tonnes per year are landfilled. This part remains unusable because of the presence of potentially hazardous metals. Electric Arc Furnace (EAF) slag, which present high contents in chromium (Cr), vanadium (V) and molybdenum (Mo), often fall into that category. They result from recycled steel scrap melting and represent between 30% and 40% of the total steel-making slag production. The European directive 1999/31/CE [4] about waste landfilling encourages the companies to change their process in order to produce heavy metal-free EAF slag for recycling. However, the significant volumes of former complete the mineralogical information about magnetic phases in slag and to extend the already existing magnetic data on industrial wastes.

Sampling
The material studied comes from the Industeel France Arcelor Mittal site of Châteauneuf (Loire, France) where a very large volume of slag was accumulated over six decades and spread over a surface of approximately 4 ha, with a thickness varying between 2 and 9 m. The site is located in an urban area, surrounded by a highway and a railway. The slag lies on the alluvial deposits of the Gier River. This study focuses on the surface layer which, in contact with rainwater and the atmosphere, should reveal the most significant transformations. Thirty samples of approximately 300 g of EAF slag (named slag-1 to slag-30) were collected on the slag dump surface, up to 10 cm deep, along two 150 m profiles, with a sampling step of 10 m (Figure 1).

Sampling
The material studied comes from the Industeel France Arcelor Mittal site of Châteauneuf (Loire, France) where a very large volume of slag was accumulated over six decades and spread over a surface of approximately 4 ha, with a thickness varying between 2 and 9 m. The site is located in an urban area, surrounded by a highway and a railway. The slag lies on the alluvial deposits of the Gier River. This study focuses on the surface layer which, in contact with rainwater and the atmosphere, should reveal the most significant transformations. Thirty samples of approximately 300 g of EAF slag (named slag-1 to slag-30) were collected on the slag dump surface, up to 10 cm deep, along two 150 m profiles, with a sampling step of 10 m (Figure 1). All of the material collected was dried at 60 °C in an oven. Then different treatments (subsampling, grinding, sieving) were applied depending on the type of analysis.
-Two sub-samples of 8 cm 3 (approximately 15 g) were taken with a beaker from each of the 30 initial samples for magnetic and chemical analyses. This method allows to evaluate the variability within the same sample and provides a sufficient amount of data for a statistical analysis (see section 2.2.4.). These samples are hereafter referred to as "slag-1A", "slag-1B", "slag-2A", "slag-2B" and so forth. -The sample slag-13, representative of the average chemical composition of the 30 EAF slag samples (Table 2), was selected to be sieved so as to explore the main properties of the slag as a function of grain size. Eight grain size fractions were examined: <0.25 mm, 0.25-0.5 mm, 0.5-1.0 mm, 1.0-1.6 mm, 1.6-2.5 mm, 2.5-5 mm, 5-10 mm and >10 mm. -A last sample called Mix 1 was prepared by mixing equal volumes from each of the 30 initial samples (16 cm 3 taken with a beaker). It was then roughly crushed with a jaw crusher, to a size smaller than 4 mm, with the aim to get a representative set of grains the size of which is convenient for micro-analyses (EMPA and Raman spectroscopy) and thermo-magnetic measurements (0.5-1.0 mm). In order to evaluate the potential for magnetic measurements to discriminate slag from other materials, rock samples were taken in the vicinity of the site. Surrounding rock formations consist of gneisses, micachists and coal sediments. One sample of gneiss, one sample of micaschist and one sample of coarse sandstone were collected from rocky outcrops within a 1-km radius of the site. All of the material collected was dried at 60 • C in an oven. Then different treatments (sub-sampling, grinding, sieving) were applied depending on the type of analysis.
-Two sub-samples of 8 cm 3 (approximately 15 g) were taken with a beaker from each of the 30 initial samples for magnetic and chemical analyses. This method allows to evaluate the variability within the same sample and provides a sufficient amount of data for a statistical analysis (see Section 2.2.4). These samples are hereafter referred to as "slag-1A", "slag-1B", "slag-2A", "slag-2B" and so forth. - The sample slag-13, representative of the average chemical composition of the 30 EAF slag samples (Table 2), was selected to be sieved so as to explore the main properties of the slag as a function of grain size. Eight grain size fractions were examined: <0.25 mm, 0.25-0.5 mm, 0.5-1.0 mm, 1.0-1.6 mm, 1.6-2.5 mm, 2.5-5 mm, 5-10 mm and >10 mm. -A last sample called Mix 1 was prepared by mixing equal volumes from each of the 30 initial samples (16 cm 3 taken with a beaker). It was then roughly crushed with a jaw crusher, to a size smaller than 4 mm, with the aim to get a representative set of grains the size of which is convenient for micro-analyses (EMPA and Raman spectroscopy) and thermo-magnetic measurements (0.5-1.0 mm).
In order to evaluate the potential for magnetic measurements to discriminate slag from other materials, rock samples were taken in the vicinity of the site. Surrounding rock formations consist of gneisses, micachists and coal sediments. One sample of gneiss, one sample of micaschist and one sample of coarse sandstone were collected from rocky outcrops within a 1-km radius of the site.

Bulk Chemical Analyses
Bulk chemical analyses were performed on the 60 sub-samples "slag1A", slag1B" and so forth, at Laboratoire Magmas et Volcans, Saint-Etienne, France, with an energy-dispersive X-ray fluorescence (ED-XRF) analyser (Olympus Vanta serie M). The samples were finely ground with a ring mill (median grain size ≈ 20 µm) and compacted in capsules.
The different particle size fractions from the sample slag-13 were solubilized by alkali fusion in order to perform bulk chemical analyses by ICP-AES (induced coupled plasma-atomic emission spectrometry). Each sample was finely ground and 100 mg of the powder was mixed with 300 mg of LiBO 2 in a porcelain dish. It was then transferred to a graphite crucible and fused for 5 min at about 1100 • C in an induction furnace (2 kW). The bead obtained was dissolved in an acidic solution of HNO 3 (1 M) and, after complete dissolution the solution was filtered to remove graphite particles. ICP-AES measurements were done at Laboratoire Magmas et Volcans, Clermont-Ferrand, France using ULTIMA C spectrometer (Horiba scientific, Jobin-Yvon), with an incident power of 1.1 kW and a reflected power <15 W. Reference materials GH (for Si, Na, K) and BR (for the other elements), both from Centre de Recherches Pétrographiques et Géochimiques (CRPG), Nancy, France, provided high points of the calibration lines, while a pure LiBO 2 solution (300 mg in 200 mL of 1M HNO 3 ) was used as the zero.
Double chemical analyses were performed on a subset of 17 slag samples by alkali fusion/ICP-AES and ED-XRF. The results showed good correlations between the two techniques and assessed the reliability of ED-XRF. This latter technique is much faster and was favoured when many samples had to be analysed.
In addition to chemical analyses, loss on ignition (LOI) was measured on the different particle size fractions of sample slag 13. Each sample was weighted before and after heating at 110 • C and 1000 • C. The mass lost at 110 • C gives information on the amount of adsorbed water, while the mass lost at 1000 • C shows the presence of hydroxides and carbonates [27]. In addition, total carbon contents were measured on the same samples at Laboratoire Magmas et Volcans in Saint-Etienne using an elemental analyser-Thermo Isolink.

Mineralogical Characterisation
X-ray diffraction (XRD) analyses were performed at Mines de Saint-Etienne by a Bruker D8 A25 diffractometer equipped with Cu radiation (WL = 1.5406 Å) on the eight particle size fractions. Each one was previously ground in the same conditions as for ED-XRF analyses (see Section 2.2.1).
Electron Microprobe and Raman analyses were done on 0.5-1 mm grains of slag. A total of 112 grains was selected from Mix 1 and moulded in epoxy resin. The five blocks obtained were lapped with a silicon carbide powder and ethanol and polished with diamond paste. The polished sections were then metallized with carbon for electron microprobe analyses (EMPA). Subsequently, the grains studied are designated by the number of the block to which they belong and by their number within the block, for example, grain 25, block 1.
Two types of EMPA analyses were performed-Point quantitative analyses and quantified chemical maps, both using the Cameca SxFiveTactis microprobe at Laboratoire Magmas et Volcans in Clermont-Ferrand.
For point quantitative analyses, the acceleration voltage was 15 kV and the current was 15 nA; the wavelength dispersive spectrometer mode was used for all the elements. Kα emissions were measured with the help of the following crystals-TAP (thallium acid phthalate) on spectrometer 1 for Si, Al, and Mg, LPET (pentaerythritol, line L) on spectrometer 2 for P and V, LLIF (lithium fluoride, line L) on spectrometer 3 for Fe, Mn and Cr and PET (pentaerythritol) on spectrometer 4 for Ca and Ti. The instrument was calibrated with the following standards-wollastonite for Ca and Si, TiMnO 3  Quantified EMPA chemical maps were generated with an acceleration voltage of 15 kV and a current of 30 nA. The step size was 0.4 µm and the dwell time was 30 ms. Mg, Al and Mn were measured in energy dispersive spectroscopy (EDS) mode and Si, Cr, Fe and Ca in wavelength dispersive spectroscopy (WDS) mode, Si with TAP on spectrometer 1, Cr with LPET on spectrometer 2, Fe with LLIF on spectrometer 3 and Ca with PET on spectrometer 4. Under these conditions, the resolution of the images is finer than the spatial resolution of the chemical analyses. This means that in the case of very fine intergrowths (<1 µm), the analyse also takes into account adjacent phases. Raman spectroscopic measurements were performed at Mines Saint-Etienne (France) using a confocal XploRA Raman microscope (Horiba Jobin-Yvon) equipped with a laser operating at 532 nm. Analyses were acquired with a laser power of 10 mW after calibration of the spectrometer by Si line at 520 ± 1 cm −1 . The spatial and spectral resolutions of the measurements were 2 µm and 1 cm −1 , respectively. Raman spectra were carried out between 250 and 1750 cm −1 , depending on the samples, to characterise the internal and lattice modes of mineral phases. Data were collected for 30 to 40 s and averaged over three accumulation cycles.

Magnetic Measurements
Mass specific magnetic susceptibility was measured using an MFK1-FA susceptibility bridge (AGICO) at Laboratoire Magmas et Volcans in Saint Etienne, on the samples slag-1A, slag-1B, . . . slag-30B, under weak magnetic field (200 A/m) and low frequency (976 Hz). Magnetic susceptibility of surrounding rocks samples was also measured with the same equipment.
Thermomagnetic analyses of low field magnetic susceptibility κ(T) were performed on the MFK1-FA susceptibility bridge with CS-4 and CS-L attachments (AGICO) at Laboratoire Magmas et Volcans in Saint Etienne. Thermomagnetic curves were carried out on 0.5-0.8 mm grains from Mix-1. For high temperature measurements, the furnace CS-4 gradually heated the samples from 20 to 700 • C under inert atmosphere (argon gas). For low temperature measurements, the samples were cooled down to −192 • C with liquid nitrogen and then gradually heated to room temperature (approximately 20 • C) with the cryostat CS-L.

Principal Component Analysis
A Principal Component Analysis (PCA) was carried out on the set of samples slag-1A, slag-1B . . . slag-30B, using 10 variables-Magnetic susceptibility previously measured with MFK1-FA Kappabridge and FeO, CaO, Al 2 O 3 , SiO 2 , MnO, Cr 2 O 3 , MgO, V 2 O 3 and MoO 3 contents obtained with ED-XRF analyses. This approach aims to relate chemical components to magnetic susceptibility. The PCA was performed with R thanks to the libraries "FactoMineR" and "factoextra", and the function "PCA". It is carried out in two steps-Normalization of the data set, followed by Eigen values decomposition of the data covariance matrix.

Bulk Chemical Composition
The average chemical composition of the 60 samples slag-1A to slag-30B is given in Table 1. Detailed results obtained for each sample are given in Supplementary Material, Table S1. Major elements are Ca, Fe and Si, with a CaO/SiO 2 mass ratio of 3.15. In addition, Table 2 shows the variation in bulk chemical composition as a function of grain size. Coarser fractions are richer in Fe, Si, Mn, P, Ti and V while finer fractions are richer in Mg and Al. They are also richer in volatile elements such as C, H and O according to total carbon analyses and LOI at 1000 • C. Mo could not be measured by ICP-AES.  Table 2. Bulk chemical composition (wt. %) of the different particle size fractions of slag-13 (obtained by ICP-AES analyses, carbon contents measured by elemental analyser) and loss on ignition mass percentage (LOI) at 110 • C and 1000 • C. N.B.: bulk chemical composition obtained by ED-XRF analysis for the initial sample slag-13 (before sieving) is given as a reference.

Mineralogical Composition
Eight mineral phases are detected by X-ray diffraction-Wustite (FeO), a spinel solid solution ((Fe,Mg)(Fe,Mn,Cr) 2 O 4 ), chromite (Fe(Cr,Al) 2 O 4 ), brownmillerite(Ca 2 (Fe,Al,Cr) 2 O 5 ), calcium chromite (CaCr 2 O 4 ), larnite (Ca 2 SiO 4 ), calcite (CaCO 3 ) and quartz (SiO 2 ). The XRD pattern of the finest (<0.25 mm) and the coarsest (>10 mm) fractions are presented in Figure 2. Table 3 presents the relative abundance of these minerals in each fraction estimated by comparing the different XRD patterns. Wustite, brownmillerite and larnite abundances decrease with particle size, while quartz, calcium chromite, spinel solid solution and calcite follow the opposite trend. These two latter minerals are known as possible secondary minerals in slag [15,28], so their presence in fine fractions, which are more prone to weathering, makes sense. On the contrary, wustite brownmillerite and larnite, which are preferentially found in coarse fractions, reflect probably the initial composition of the slag. Quartz which cannot be formed at room temperature and pressure is supposed to come from other materials stored with slag on the site (sand, refractory bricks). This distribution of minerals among the different fractions must also be related to the disintegration phenomenon that takes place during cooling. It is known that at a certain temperature, slag pieces crumble into a fine powder due to the transformation of dicalcium silicate (larnite) from the polymorph βCa 2 SiO 4 to γCa 2 SiO 4 . This transformation is accompanied by an increase in volume of about 12% and causes the slag pieces to burst [29]. The consequence in the present study is that the larnite particles that undergo this disintegration are found in the fine fractions, and are very quickly transformed into calcite on contact with air and moisture.
Minerals 2020, 10, x FOR PEER REVIEW 7 of 19 of dicalcium silicate (larnite) from the polymorph βCa2SiO4 to γCa2SiO4. This transformation is accompanied by an increase in volume of about 12% and causes the slag pieces to burst [29]. The consequence in the present study is that the larnite particles that undergo this disintegration are found in the fine fractions, and are very quickly transformed into calcite on contact with air and moisture.  Electron Microprobe analyses coupled with Raman spectroscopy gives further information about mineral composition. Most of the phases detected by XRD are observed, except for quartz and calcite. Quartz has been considered to be a contamination coming from other materials, thus, it is not observed in individual slag grains. Calcite is mainly present in the fine fractions (<0.5 mm) that have  Electron Microprobe analyses coupled with Raman spectroscopy gives further information about mineral composition. Most of the phases detected by XRD are observed, except for quartz and calcite. Quartz has been considered to be a contamination coming from other materials, thus, it is not observed in individual slag grains. Calcite is mainly present in the fine fractions (<0.5 mm) that have not been incorporated into the epoxy blocks. In addition, as a secondary mineral, calcite is expected to be found at the rims of the grains, and so was probably removed by polishing. Tricalcium silicate (Ca 3 SiO 5 ) is detected in addition of larnite (Ca 2 SiO 4 ) according to EMPA (Table 4). It can also be larnite with lime micro-inclusions resulting from the decomposition of tricalcium silicate at 1250 • C during cooling process, as highlighted by Waligora et al. and Engström et al. [14,30].
The most common structures found in the investigated slags are presented in Figure 3. They are almost entirely crystallized in accordance with slow cooling conditions (air-cooling). Some grains present a dendritic texture, indicating a slightly higher cooling rate as in Figure 4a. Calcium silicate particles are generally subhedral with an ovoid shape, while brownmillerite appears as elongated lamellae (Figure 3a,b). Calcium chromite appears as needle shaped minerals, sometimes surrounded by brownmillerite (Figure 3b). The spinel solid solution must be differentiated from chromite which also makes part of the spinel family. Indeed, in the studied samples, chromite crystals are euhedral, and so easily recognizable with their cubic shape, while for the spinel solid solution (Fe,Mg)(Fe,Mn,Cr) 2 O 4 , the grains have a rounded shape (Figure 3a,b) which is reminiscent of the shape of wustite in many studies of EAF slag [7,13]. In addition, a significant number of these grains have a heterogeneous appearance (Figure 4b,c and Figure 5). These observations suggest that this spinel solid solution is actually secondary, and results from an oxidation of a primary wustite solid solution (Fe,Mn,Mg)O. Indeed wustite is known to be metastable [31] and several occurrence of spinel exsolution in wustite are reported in the literature [32,33]. EMPA and Raman analyses below help to understand the heterogeneous nature of the spinel solid solution.
Minerals 2020, 10, x FOR PEER REVIEW 8 of 19 not been incorporated into the epoxy blocks. In addition, as a secondary mineral, calcite is expected to be found at the rims of the grains, and so was probably removed by polishing. Tricalcium silicate (Ca3SiO5) is detected in addition of larnite (Ca2SiO4) according to EMPA (Table 4). It can also be larnite with lime micro-inclusions resulting from the decomposition of tricalcium silicate at 1250 °C during cooling process, as highlighted by Waligora et al. and Engström et al. [14,30]. The most common structures found in the investigated slags are presented in Figure 3. They are almost entirely crystallized in accordance with slow cooling conditions (air-cooling). Some grains present a dendritic texture, indicating a slightly higher cooling rate as in Figure 4a. Calcium silicate particles are generally subhedral with an ovoid shape, while brownmillerite appears as elongated lamellae (Figure 3a,b). Calcium chromite appears as needle shaped minerals, sometimes surrounded by brownmillerite (Figure 3b). The spinel solid solution must be differentiated from chromite which also makes part of the spinel family. Indeed, in the studied samples, chromite crystals are euhedral, and so easily recognizable with their cubic shape, while for the spinel solid solution (Fe,Mg)(Fe,Mn,Cr)2O4, the grains have a rounded shape (Figure 3a,b) which is reminiscent of the shape of wustite in many studies of EAF slag [7,13]. In addition, a significant number of these grains have a heterogeneous appearance (Figure 4b,c and Figure 5). These observations suggest that this spinel solid solution is actually secondary, and results from an oxidation of a primary wustite solid solution (Fe,Mn,Mg)O. Indeed wustite is known to be metastable [31] and several occurrence of spinel exsolution in wustite are reported in the literature [32,33]. EMPA and Raman analyses below help to understand the heterogeneous nature of the spinel solid solution. The average chemical composition of each mineral phase obtained by EMPA is given in Table 4.  The average chemical composition of each mineral phase obtained by EMPA is given in  The diversity of the spinel solid solution particles is illustrated in Figures 4 and 5. Four typical forms are presented:

•
Particles with patchy zonations. On Figure 4b, three different phases are actually distinguishable within the same particle: Fe rich jacobsite (points 5 and 6), jacobsite (points 7 and 8) and harmunite, CaFe 2 O 4 (points 9 and 10) based on the work of Galuskina et al. [34]. Some hematite peaks visible on Raman spectra for point 8 suggest that the corresponding phase is partially oxidized (Figure 4d). However, this oxidation may be caused by Raman laser heating [35,36].  (Table 5). Grey lamellae are identified as exsolved harmunite thanks to EMPA (Table 5), however it could not be confirmed with Raman analyses since the laser beam was to large. Thin light borders around the lamellae may be pure magnetite or wustite.  (Figure 5b,e,f).
In conclusion, various chemical zonations, harmunite exsolutions and maghemite are highlighted thanks to EMPA and Raman analyses explaining most of the heterogeneity observed. Wustite is not clearly identified whereas it is detected by XRD. This primary phase can however reside in very fine intergrowths for which it is very difficult to carry out a single-phase analysis without contamination from the adjacent phases. In conclusion, various chemical zonations, harmunite exsolutions and maghemite are highlighted thanks to EMPA and Raman analyses explaining most of the heterogeneity observed. Wustite is not clearly identified whereas it is detected by XRD. This primary phase can however reside in very fine intergrowths for which it is very difficult to carry out a single-phase analysis without contamination from the adjacent phases.

Potential Hazardous Metal Location
Three potential hazardous metals are identified-Chromium (Cr), molybdenum (Mo) and vanadium (V). EMPA results presented in Table 4 give the concentration of these metals in each mineral phase. It shows that the main chromium bearing minerals are chromite with an average of

Potential Hazardous Metal Location
Three potential hazardous metals are identified-Chromium (Cr), molybdenum (Mo) and vanadium (V). EMPA results presented in Table 4 give the concentration of these metals in each mineral phase. It shows that the main chromium bearing minerals are chromite with an average of 24.88 at. % and calcium chromite with an average of 24.43 at. %. Cr is also present in brownmillerite (3.39 at. %) and in a lesser extend in the spinel solid solution (Mg,Fe)(Fe,Mn) 2 O 4 (0.94 at. %). V is present in small amounts in brownmillerite (up to 0.98 at. %) and in chromite (up to 0.13 at. %). Mo contents measured by EMPA are below detection limit, so this element is not clearly localized.
The evolution of magnetic susceptibility as a function of temperature for the sample Mix-1 is shown in Figure 7. This thermomagnetic curve shows a peak of κ around −155 °C which corresponds to the Verwey transition and a drop around 575 °C, the Curie temperature both indicating the presence of magnetite (Fe3O4). This means that the spinel solid solution (Mg,Fe)(Fe,Mn)2O4 takes the form of rather pure magnetite in some cases. Indeed, the Verwey transition is detectable only for magnetite, and the Curie temperature is very sensitive to stoichiometry, so the other forms of magnetic spinel will manifest themselves at different temperatures. The Curie temperatures of the  [17,18,[21][22][23][24]38,[40][41][42][43][44][45][46][47][48][49][50][51][52]. Magnetic susceptibility measured for some surrounding rocks (micaschiste, gneiss, sandstone) are included. Hematite and magnetite usual susceptibilities are given as points of reference: the wide range of values is explained by the fact that magnetic susceptibility depends on the size and shape of the crystals, and on the presence of trace elements [38,52].
The evolution of magnetic susceptibility as a function of temperature for the sample Mix-1 is shown in Figure 7. This thermomagnetic curve shows a peak of κ around −155 • C which corresponds to the Verwey transition and a drop around 575 • C, the Curie temperature both indicating the presence of magnetite (Fe 3 O 4 ). This means that the spinel solid solution (Mg,Fe)(Fe,Mn) 2 O 4 takes the form of rather pure magnetite in some cases. Indeed, the Verwey transition is detectable only for magnetite, and the Curie temperature is very sensitive to stoichiometry, so the other forms of magnetic spinel will manifest themselves at different temperatures. The Curie temperatures of the end members jacobsite MnFe 2 O 4 and magnesio-ferrite MgFe 2 O 4 are respectively 310 • C and 440 • C [53]. The progressive decrease of κ between 280 • C and 557 • C in Figure 7 shows that a wide range of the intermediary phases between jacobsite, magnesio-ferrite and magnetite are present in the studied slag. The loss of approximately 70% of the initial magnetic susceptibility between 280 • C and 557 • C indicates that the jacobsite-magnesio-ferrite solid solution is mostly responsible for the magnetic behaviour of the slag. Pure magnetite was not clearly observed by electron microprobe and should therefore be present in only very small quantities. The irreversibility of the curve must be pointed out. The production of thermomagnetic curves (heating and cooling) from room temperature to 200 • C and from room temperature to 400 • C shows that reversibility is lost from 300 • C onwards, indicating a phase transformation near this temperature that could be caused by dehydration. from room temperature to 400 °C shows that reversibility is lost from 300 °C onwards, indicating a phase transformation near this temperature that could be caused by dehydration. Figure 7. Variation of the magnetic susceptibility (κ) during heating-cooling cycle (red and blue line respectively) from −192 °C to room temperature and from room temperature to 700 °C.

Principal Component Analysis
The correlation circle resulting from PCA is shown on Figure 8. It gives a projection of the eight initial variables as a function of the first two principal components, dimension 1 and dimension 2, which bear respectively 39.30% and 18.77% of the total variance. Magnetic susceptibility is obviously correlated with FeO content. Another strong correlation appears between MnO, Cr2O3 and V2O3 contents. More broadly, FeO, MnO, Cr2O3 and V2O3 contents seem to behave in the same way. Thus, magnetic susceptibility seems to be an indicator for high contents in MnO, Cr2O3 and V2O3, in addition to FeO. The contents in CaO and MgO are also correlated, while CaO is anticorrelated with SiO2. Al2O3 is anticorrelated with FeO and Cr2O3. No obvious correlation appears for the contents in MoO3. Figure 7. Variation of the magnetic susceptibility (κ) during heating-cooling cycle (red and blue line respectively) from −192 • C to room temperature and from room temperature to 700 • C.

Principal Component Analysis
The correlation circle resulting from PCA is shown on Figure 8. It gives a projection of the eight initial variables as a function of the first two principal components, dimension 1 and dimension 2, which bear respectively 39 correlated with FeO content. Another strong correlation appears between MnO, Cr2O3 and V2O3 contents. More broadly, FeO, MnO, Cr2O3 and V2O3 contents seem to behave in the same way. Thus, magnetic susceptibility seems to be an indicator for high contents in MnO, Cr2O3 and V2O3, in addition to FeO. The contents in CaO and MgO are also correlated, while CaO is anticorrelated with SiO2. Al2O3 is anticorrelated with FeO and Cr2O3. No obvious correlation appears for the contents in MoO3.

Comparison with Other EAF Slag
The studied slags have quite a high CaO/SiO 2 ratio, with an average of 3.15, while in previous studies focusing on EAF slag, it typically ranged from 0.9 to 2.9 [5,7,8,15,[54][55][56][57][58][59]. Consequently, the little amount of Si present is only found in calcium silicates, the other silicate minerals usually found in slag such as gehlenite, fayalite or kirschsteinite are absent. The remaining Ca is associated with Fe, Al and Cr to form brownmillerite and in a lesser extent calcium chromite and calcium ferrite (harmunite). The presence of these two latter minerals is not common in EAF slag and appears as a specificity of very basic slag. Calcium chromite occurrences were mentioned by Albertsson et al., Cabrera-Real et al. and Mombelli et al. [6,7,60] and was shown to be favoured in highly alkaline slag. The distribution of the mineral phases between the different grain size fractions gives an idea on how the material breaks up over time. It appears that ferrous minerals (wustite, brownmillerite, chromite), which are more resistant, are concentrated in the coarsest fractions, while calcium silicates tend to disintegrate and form calcite. Some studies outlined that calcite thus formed on the surface of slag particle can have a protective effect and limit the leaching of hazardous metals [15,61]. This applies only if the material does not undergo further fragmentation due to natural alteration or any manipulation.

Hazardous Metals Location and Potential Mobility
This study focuses on three potential hazardous metals-Cr, Mo and V. Cr is usually present as Cr (III) and Cr (VI) in the environment. Cr (III) is poorly soluble unlike Cr (VI) which has a high mobility and is known to be highly toxic [62,63]. Mo and V play an essential role in biological systems, but have a negative effect on organisms above a certain concentration [64][65][66]. V interfere with the function of P-containing biomolecules [65], and Mo can be adsorbed in soils and accumulated in plants, causing ruminants disease such as molybdenosis [67,68]. The contents of these two metals being very low in the studied slag, the main concerns are about Cr which has been found in chromite, calcium chromite, brownmillerite and in a lesser extent in the spinel solid solution (Mg,Fe)(Fe,Mn) 2 O 4 . In these phases, it is expected to be present in its less toxic form Cr (III), but it can react on cooling with free CaO and form Cr (VI) as described by Lee and Nassaralla [69]. Albertsson et al., Cabrera-Real et al. and Mombelli et al. [6,7,60] showed that this element is hardly leachable when dissolved in spinel minerals but could be released by calcium chromite and brownmillerite. This latter mineral also contains up to 0.98 at. % of V, and could release this metal at the same time. De Windt et al. [28] mention an incongruent dissolution of brownmillerite in Basic Oxygen Furnace slag-Al and V are leached while Ca, Fe and Cr remain forming dicalcium ferrite.

Hazardous Metals and Magnetic Susceptibility
The global correlation between magnetic susceptibility, Fe, Cr, V, and Mn obtained with the PCA is consistent since these elements are present in significant amounts in chromite and in the spinel solid solution, except for V which is mainly found in brownmillerite. Magnetic susceptibility is mainly due to the spinel solid solution which has been identified as ferromagnetic. Whereas the amount of Fe in chromite is too low to give ferromagnetic properties to chromite. Indeed, Ziemniak and Castelli [70] showed that the solid solution Fe(Fe 1−n Cr n ) 2 O 4 is paramagnetic at room temperature if n ≥ 0.65. According to EMPA analysis, the studied chromite formula would be close to (Fe 0.3 ,Mg 0.5 ,Mn 0.2 )(Al 0.1 ,Cr 0.9 ) 2 O 4 , so with n = 0.9, the magnetic susceptibility should be much lower than for magnetite and for the spinel solid solution (Mg,Fe)(Fe,Mn) 2 O 4 . Thus, magnetic susceptibility appears as an indicator for high contents in Fe, Mn, Cr and V in the studied slag, because chromite and the spinel solid solution seem to be associated. Mo has not been detected in a particular phase and is supposed to be disseminated in all minerals, which would be in accordance with the PCA showing no correlation with any other variable. Some precautions must be taken about V and Mo since the contents are very low the significance of the correlations can be questioned for these elements.

Magnetic Susceptibility as a Field Tool
The very high susceptibility of the studied slag makes this material easily distinguishable from soils, rocks or other wastes. Magnetic measurements are quick and can be easily carried out on the field scale. Magnetic susceptibility could be used as a tool to detect the dispersion of slag particles in soils or in sediments, or to delineate historical slagheaps areas, especially when the history of the site is little known. The use of magnetic susceptibility as a tracer for heavy metals (Cr, V) seems possible according to the correlations obtained by PCA, but it needs to be tested at a larger scale. It could help to identify the highest concentrations spots before a more detailed study.

Conclusions
In this work, partially weathered EAF slag samples are studied, with a multi-analytical approach combining chemical, mineralogical and magnetic techniques. The material is characterised by a high CaO/SiO 2 ratio compared to other EAF slags. It contains several potentially hazardous metals-Cr, Mo and V. Mineralogical analyses reveal that Cr is mainly present in two minerals-chromite, which is rather stable and Ca-chromite, which is much more soluble. Mo and V are present in very low quantities but their leachability should be controlled.
Magnetic measurements show a very high magnetic susceptibility compared to other industrial wastes, it is explained by the presence of the spinel solid solution (Mg,Fe)(Fe,Mn) 2 O 4 and some magnetite. Magnetic techniques are easily transposable to the field scale and must be considered when estimating the dispersal of slag particles in the environment.
Coupling chemical and magnetic analyses is interesting but does not exclude a detailed study of the mineralogy. It is important to keep in mind that the hazardous levels of these metals depends not only on their concentrations, but also on their speciation and the solubility of their host minerals. In this study, a Principal Component Analysis shows a correlation between magnetic susceptibility and Fe with the metals Cr, Mn and V. It means that samples richer in ferromagnetic minerals are also richer in Cr and V bearing minerals. It must be assessed at the field scale in future studies.