Discriminating hematite and magnetite via Scanning Electron Microscope–Mineral Liberation Analyzer in the −200 mesh size fraction of iron ores

Abstract

Scanning Electron Microscope–Mineral Liberation Analysis (SEM–MLA) can be used to discriminate between hematite and magnetite in iron ores. However, achieving backscattered electron (BSE) segmentation between the two minerals is difficult for particles ≤75 µm using typical preparation and analysis methods for the MLA method based on a tungsten filament SEM (Quanta 400) with 25 kV high voltage. Preparing iron ore sample mounts using a slow-speed polishing method, and conducting the experiment on a field emission gun SEM–MLA (Quanta 650) with the high voltage setting lowered to 15 kV reduces instrument noise and results in very clean BSE images and segmentation. This method requires new X-ray standards for each mineral at 15 kV because of major changes in the spectra at lower kV. However, once these X-ray spectra are added to the mineral reference list, effective segmentation can be achieved and an accurate analysis obtained.

Keywords

Iron ore automated mineralogy mineral liberation analysis field emission gun scanning electron microscopy

Introduction

The Mineral Liberation Analyzer (MLA) is useful in a wide variety of automated mineralogical applications (Gu 2003; Sylvester 2012). One of the recent advances in mineral liberation software has been the automated discrimination of hematite (Fe₂O₃) and magnetite (Fe₃O₄) in polished particle epoxy mounts (Shaffer 2009; Figueroa et al. 2011; Grant et al. 2016). However, these studies focused primarily on particles ≥75 µm. In our experience, particles less than 75 µm (−200 mesh) are more problematic for segmentation of hematite and magnetite. In particular, these fine particles often do not polish as well as larger particles exposing many imperfections and pits (Figure 1). Thus, the MLA will record two different brightness values in backscattered electron (BSE) images. These are the normal brightness associated with the bulk of the particle interior, and a higher brightness associated with particle edges, or polishing artefacts and cracks in the interior, thus potentially misidentifying the ores. The MLA segmentation identifies different brightness as multiple phases, and associates different BSE grey values with them. The problem manifests itself especially with hematite whereby the interior darker BSE associates correctly with hematite, but the brighter BSE edges and cracks incorrectly associate with the brighter mineral magnetite. An electron beam accelerating high voltage of 25 kV exacerbates the problem. High-quality polish does not entirely eliminate the problem very close to particle edges or at severe cracks.

Another problem with this size fraction is that it is not a true size fraction in that there is no specific lower size limit. Rather it is what particles remain after passing a 200-mesh sieve. These particle sizes extend down to less than a micrometre in typical ore concentrates. The magnification is correctly chosen for the larger particles that comprise most of the <75 µm fraction, but this magnification is inappropriate for the smallest particles. The brightness problem described above is also exacerbated for the smallest particles.

Our experiment aimed to analyse the size fraction at 15 kV, and employ the FEI MLA software on a Quanta 650 FEG (field emission gun), as well as a tungsten instrument (a Quanta 400 MLA), in an attempt to include the smallest of particles. From this point on we will refer to these as the Quanta 650 FEG and Quanta 400 for simplicity. The lower limit of grains measured in this study is ∼5 µm, below which grains identified by the software as low counts become much more numerous and the grey levels become uncertain. Although it is possible to simply omit analysis of this size fraction (<75 µm), our experience has been that omission of the fine fraction from some iron ores result in high calculated Fe and low calculated Si assays determined for the bulk sample by mineral liberation analysis. Also, many iron ore concentrates contain <75-µm particles for which liberation of magnetite from hematite is desirable.

Mineral reference standards (a spectral library reference of X-ray standards) must be collected with the same accelerating voltage as the unknowns. A high voltage of 25 kV is commonly used for mineral liberation analysis measurements of the coarser particle sizes of iron ores. However, using 25 kV mineral references for mineral unknowns measured at 15 kV results in very low matching scores of 50% for mineral identifications even for the most easily determined minerals (e.g. magnetite, hematite, quartz). The elemental peaks are in the correct x-axis position, but they vary greatly in intensity. They are enhanced for low energy-dispersive X-ray energies and attenuated for high energies. To overcome this, each 25 kV spectra within a mineral definition must be removed and new 15 kV spectra acquired for these phases. Using 15 kV results in lower sensitivities for the transition metals, including Mn and Fe. Nonetheless, Fe oxides are still easily identified with the use of 15 kV reference spectra.

Methods

Five samples of iron ores were acquired from deposits in the Labrador Trough (Canada). The samples contain various hematite to magnetite to quartz to carbonate ratios. Aliquots from a single riffle (−200 mesh size fraction, <75 µm) for each replicate were produced as a trans-vertical mould (Grant et al. 2016), polished, coated with evaporated carbon and analysed by SEM–MLA. These sections were polished to a high standard, and were analysed on a Quanta 400 at 25 and 15 kV, and on a Quanta 650 FEG at 25 and 15 kV. The full details of the single-step trans-vertical mould preparation and polishing method are listed in Grant et al. (2016). Instrument conditions and parameters for both instruments include 10 nA beam current for the Quanta 650 FEG (13 nA for Quanta 400), 16 µs BSE dwell time, 10 pixel minimum size (800 pixel frame resolution for 1 mm horizontal field width), 60, 000 maximum particle count and 12 ms spectrum dwell time for improving sensitivity of transition metals. Each of these mineral liberation analysis acquisitions was completed using version 3.1.4.683 MLA™ software and took between 1 and 1.5 hours

The quality of sample polish and the instrument high voltage setting both affect the BSE segmentation between hematite and magnetite. A sample with multiple surface defects in the iron oxides analysed on a Quanta 400 with a high voltage of 25 kV illustrate an imperfect histogram caused by multiple artefact peaks, as well as regions affected by 25 kV brightening (Figure 1(a)). However, when the polish quality was improved and the same sample analysed with a Quanta 650 FEG at 15 kV high voltage, the resulting histogram has much lower signal-to-noise, fewer artefacts, and more obvious segmentation between hematite and magnetite (Figure 1(b)). The increase in polish quality was achieved by doubling the amount of time spent on polishing at the 3 and 1 µm stages (our typical method utilises 10 min per stage, but in this case 20 min were spent polishing at each of the 3 and 1 µm stages; Grant et al. 2016), and lowering the accelerating voltage to 15 kV. Note that magnetite is the brighter mineral of the two, with higher BSE brightness.

The spectrum window limits method is used for mineral liberation analysis classification, for both 25 and 15 kV (Figure 2). Spectrum windows are first defined relative to energy-dispersive X-ray energy for each element of concern – in this case, Mg (1.15–1.37 keV), Ca (3.49–3.89 keV), Mn (5.65–6.15 keV) and Fe (6.15–6.65 keV), which are shown in the lighter vertical bands (Figure 2). The percentages shown in this figure represent the total number of X-ray counts in the respective window relative to all counts in the spectrum. The classification is defined relative to these percentages, and the analyst uses the mineral liberation analysis Image Processing software classification dialog to define the rules. When defining a reference for a particular mineral, e.g. ferroan-magnesite (+Mn), a rule may be created stating that Mg, at 25 kV, should fall between ∼13 and 30% and Fe should be between ∼0 and 9%. The confidence criteria for the classification is set under the basic criteria tab (default = 70% confidence). For applications in hematite–magnetite segmentation, we discriminate relative to BSE brightness under the BSE range limits tab, and for the carbonates we created rules under the spectrum window limits tab.

The MLA software provides a method for determining elemental percentages. For a given analysis that has already been classified, each grain can be selected to view its X-ray particle and have the spectral region of interest magnified (e.g. magnesium). This is done by gradually zooming in and approaching the region that encompasses the element with minor background on either side. Once zoomed in the Visible percentage value can be observed (15.9%) to the right of Total Counts – it is this value that is the percentage of magnesium in the total spectrum (Figure 3). The concentration range of Mg for this mineral is then defined. For this relatively Fe-rich magnesite that contains approximately as much Fe as is acceptable at the upper limit for the particular mineral, whereas nearing the lower limit of Mg to allow for a mineral defined as ferroan-magnesite, and a range of 14–30% might be appropriate. Setting these relatively wide concentration limits also helps deal with the extreme statistical variation in low-count spectra (observed more on the tungsten-based instrument at lower voltages). In the same mineral definition, allow for a small range of Ca and Mn (e.g. 0–3%), but the Fe region needs to be scrutinised in the same way as was done for Mg.

Results

It can be seen that the Quanta 650 FEG with a high voltage of 15 kV is more accurate at identifying carbonate content than the other methods (Figure 4). The same voltage setting on the Quanta 400 reveals much less carbonate, and consistently higher proportions of iron oxides and quartz. There are three subdivisions of dolomite, with two being significant enough to be displayed. The subdivisions were Dolomite (<2.5% Fe), Ferroan-Dolomite (> 2.5% Fe) and Ferroan-Dolomite (+Mn) (>2.5% Fe with Mn present as well), with the manganese-bearing variant being the least abundant phase.

The effect of instrument beam voltage on the particle size distribution was observed on these samples. A plot of particle size distributions displays two pairings of lines, with the Quanta 650 FEG producing smaller particle sizes than the Quanta 400 (Figure 5). This is consistent with the FEG having smaller spot size than the tungsten instrument.

Discussion

A major factor improving the segmentation is the use of a FEG instead of a tungsten filament instrument. The Gaussian symmetry is somewhat skewed on a Quanta 400 with high voltage of 25 kV, as compared to when the accelerating voltage is lowered and a FEG instrument is used (Figure 1). The lower accelerating voltage also makes it easier to determine which BSE value is used as the dividing line for segmentation of hematite and magnetite (grey level = 198 in Figure 1(b), but it is less clear in Figure 1(a)).

If one considers a horizontal line measuring brightness across a single frame, using the Quanta 650 FEG, the transition in BSE image (BEI) brightness when the line is near the edge of epoxy (grey level = 10) to an oxide (grey level > 125) will demonstrate very vertical and sharp edges. The same line experiment on a Quanta 400 reveals less steep transitions from dark epoxy to brighter oxide. Thus, particle edges determined on the FEG instrument appear to be sharp and crisp, compared to those on the tungsten-based scanning electron microscope (SEM), which appear fuzzier. This decrease in resolution can lead to an apparent increase in size.

It can be seen that the higher 25 kV results in a curve that is shifted right on the Quanta 650 FEG, and thus implies that the particles are registering with slightly larger areas than observed at 15 kV. This makes sense because of the bigger interaction volume at higher voltages, and thus larger plume and depth penetration into the sample. This trend is not consistent on the Quanta 400, and we suspect this is due to the lower resolution and fuzziness resulting in an inability to distinguish the grain boundaries as clearly.

When comparing the BSE segmentation between hematite and magnetite for sample A via the four different measurement modes, the Quanta 400 instrument produced more asymmetrical histograms than those obtained via the Quanta 650 FEG. Another observation is that the 15 kV mode on the Quanta 650 FEG produced much lower baseline noise, and hence improved segmentation between hematite and magnetite. These trends were consistent across all five samples and illustrate that the 15 kV method on the Quanta 650 FEG is the preferred analytical method.

The MLA software was used to calculate the ratio for the defined surface area (phase specific surface area) of the sample for the 25 kV method on the Quanta 400 versus the 15 kV method on the Quanta 650 FEG for all five samples. For all samples the area reported is 11–13% larger than when using a lower voltage. This can have a considerable effect on an analysis, and lead to operators routinely reporting that the ore minerals are larger than they actually are.

MLA calculated assay values can be reconciled with inductively coupled plasma optical emission spectroscopy (ICP-OES) for elements including Fe, Si, Mg and Ca. This is an important correlation between a bulk solution analytical method (ICP-OES) and the MLA which does not measure the elements themselves. In the case of MLA, the elemental assay values are derived from the Dataview software, which calculates these values from the user-defined mineral formulae for each mineral applied against the modal mineralogy. The best correlations (higher R² values) are found when employing the Quanta 650 FEG at 15 kV (Table 1). For the Quanta 400 at 15 kV, some of the carbonates are small and thus not identified as such, instead being misclassified as various iron oxides or silicates. The worst reconciliation values for Ca, Fe (as well as Mn) are observed on the Quanta 400 at 15 kV.

Table 1

R² correlation values for ICP-OES assay versus MLA assay.

	Quanta 400, 25 kV	Quanta 400, -15 kV	Quanta 650 FEG, 25 kV	Quanta 650 FEG, 15 kV
Fe	0.9746	0.9299	0.9738	0.9897
Si	0.9947	0.9919	0.9965	0.9962
Mg	0.985	0.982	0.985	0.994
Ca	0.994	0.972	0.988	0.993

Conclusions

SEM–MLA can be used to discriminate between hematite and magnetite in iron ores, but achieving accurate BSE segmentation between the two minerals is difficult for particles ≤75 µm using typical preparation and analysis methods on a Quanta 400 at 25 kV high voltage. We have demonstrated that when preparing this fine size fraction with single-step TV moulds, a slow-speed polishing method should be used to improve quality. This enables the MLA to more likely determine the correct BSE value for hematite and magnetite. As well, 25 kV settings do not yield optimum spatial resolution. Decreasing the high voltage setting to 15 kV, and conducting the experiment on a Quanta 650 FEG, reduces instrument noise and results in a very clean BEI and segmentation. This method requires the replacement of 25 kV reference spectra in the spectral library with 15 kV X-ray standard spectra. However, once these X-rays are added to the mineral reference list, very clean segmentation can be achieved and an accurate analysis obtained.

Footnotes

Acknowledgements

We would like to thank the Technical Services department at Memorial University for aiding in development of the TV moulds, as well as polishing and sample adapters. We would also like to thank Chris Finch of the Department of Natural Resources, NL, for his help with ICP-OES analyses.

Disclosure statement

No potential conflict of interest was reported by the authors.

References

Figueroa

, Moeller

, Buhot

, Gloy

, Haberlah

2011. Advanced Discrimination of Hematite and Magnetite by Automated Mineralogy. 10th Int. Cong. for Appl. Mineralogy, Trondheim, Norway, August 1–5; p. 197–204.

Grant

, Goudie

, Shaffer

, Sylvester

2016. A single-step trans-vertical epoxy preparation method for maximising throughput of iron-ore samples via SEM-MLA analysis. App Earth Sci: TIMM B. 125(1):57–62. doi: 10.1080/03717453.2015.1104056

2003. Automated scanning electron microscope based Mineral Liberation Analyzer. An introduction to JKMRC/FEI Mineral Liberation Analyser. J Minerals Mater Characterization Eng. 2:31–41.

Shaffer

2009. Discriminating hematite and magnetite and quantifying their associations using the JKTech mineral liberation Analyzer™. Proceedings of 48th conference of metallurgists, COM2009, Sudbury, p. 055–064.

Sylvester

PJ.

2012. Chapter 1: Use of the Mineral Liberation Analyzer (MLA) for Mineralogical Studies of Sediments and Sedimentary Rocks. Mineralogical Assoc. of Canada Short Course Series. 01/2012; 42:1–16.