1 Introduction

X-ray computed tomography (XCT) is a non-destructive analytical technique which has experienced an increase in popularity in the field of geosciences in recent years. In part, this is due to improvements through faster computers and graphics cards, better storage media, and improved X-ray transmission detectors. Application areas for XCT and micro-XCT, include sample characterization in terms of the in-situ gold distribution [1,2,3], mineral identification and liberation [4], extending 2D mineral information to 3D [5, 6], as well as high spatial resolution information on 3D rock textures, structures and density [7, 8].

Whereas XCT can provide a unique 3D view, highlighting the interior of a rock sample, other techniques can provide complementary information directly related to concentrations of elements based on surface measurements. X-ray fluorescence (XRF) offers a direct detection of chemical elements, although when applied as a non-destructive technique, this method faces challenges related to the nugget effect, grain size, and matrix effects, typically limiting the results to semi-quantitative use.

Recently, a dedicated XCT-XRF drill core scanner, the GeoCore X10®, has been put on the market [9]. Since the launch, GeoCore X10s have scanned tens of thousands of meters of drill core through increased adoption by researchers, miners, and explorers [e.g., 10, 11, 8, 12].

X-ray transmission measurements have proven to be valuable for tungsten ore sorting [13, 14]. It has previously been suggested that XCT can replace traditional analysis methods in some types of tungsten deposits where scheelite is the dominant mineral at high attenuation and where the grain size is resolved by the XCT analysis [15]. However, existing XCT studies of tungsten ore are only based on minor scanned volumes [15, 16]. In this paper, we present XCT-XRF scans and analysis of tungsten concentrations of 57 drill core samples from the Mittersill tungsten mine weighing 8.2 kg. All tungsten grades have been compared with lab assays. As the employed scanning speed is 13 min per meter drill core, a practical solution for rapid on-site analysis in certain tungsten mines is tested on a small scale.

2 Geological Setting of the Mittersill W-Deposit

The Mittersill (Felbertal) W-deposit (\(47.22598^{\circ }\) North, \(12.48871^{\circ }\) East) is located 8 km south of Mittersill town in the Tauern Window of the Eastern Alps in southwest Austria. The host rock is dominated by basement rocks of the Early Paleozoic Metabasite Complex, hosting the mineralization, and the overlying volcanosedimentary units of the Habach Complex comprising the Upper Magmatic Series and the Habach Phyllites [17, 18].

The Early Paleozoic Metabasite Complex is represented by an ophiolite and arc sequence metamorphosed during the Variscan orogeny characterized by fine and coarse-grained amphibolites, hornblendite, schists, and various types of layer parallel granitic and felsic volcanic gneisses [18, 19]. These lithologies are intruded by 340–270 Ma Variscan and post-Variscan granitoids of the Zentalgneis, ranging in composition from early high-K and calc-alkaline I-type granites to younger garnet-bearing S-type Granatspitz granites [20] subsequently deformed and metamorphosed under greenschist to amphibolite facies conditions at c. 30 Ma during the Alpine orogeny [21]. The W-deposit is associated with the metaluminous scheelite-bearing, two-mica metagranite suggested being an evolved member of the same magma as the 340 Ma Felbertal augengneiss [18, 22].

The Mittersill W-deposit occurs as eight distinct ore bodies (K1-K8) separated into eastern and western ore fields [23]. Recently, the scheelite-dotted gneiss (SD gneiss) was classified as a ninth ore body [24, 25]. Scheelite is present in all lithologies, but the economic grade is mainly found in quartz-rich zones either as stockwork to vein-type mineralizations, breccias, or in overlying, foliated quartz-scheelite bodies. The eastern ore field was mined in an open pit between 1975 and 1986 with a total production of 2.5 Mt @ 0.6 wt.% WO3. It consisted of WNW, slope parallel, bodies plunging 25–55° to the westsouthwest [24]. The western ore field is still in operation and mined underground [e.g., 22], and has an ore grade of c. 0.3 wt.% WO3 [24]. The ore plunges toward WNW and in crossection mineralized volume covers an area of about 300 × 500 m [24]. The most important ore bodies are K1-K3 [22, 24] that are structurally controlled by shear zones established at the margin of the K1-K3 metagranite, in which a major mineralized quartz lens is present [24].

Four stages of mineralization have been identified [26]. Scheelite formed during the first and second stage is Mo-rich (0.3 - 1.7 wt.%), of which scheelite 1 crystallized at 335 Ma [27]. It is fine-grained, zoned, U-rich, has a white to yellowish white fluorescence, and occurs in both stockwork and in the quartz-scheelite bodies [24]. Scheelite formed during stage 2 is common in the western ore field. This generation is present as fine to coarse-grained porphyroblasts with a yellow fluorescence and is associated with bluish beryl. Scheelite 1 and 2 may have overgrowths or fractures filled with stage 3 scheelite that is additionally present as disseminations along the foliation planes. Stage 3 scheelite is Mo-poor with a blue fluorescence and is often associated with fine-grained molybdenite [26]. Scheelite 4 also has white to pale blue fluorescence and is present as porphyroblasts in Alpine quartz veins [28].

Geologists at the mine classify the scheelite grain size into three categories; fine (< 0.5 mm), intermediate (0.5–5 mm), and coarse (> 5 mm). All three categories are common in the deposit. The largest scheelite grains observed at the mine are greater than 1 dm, but due to their rarity, such grains are very unlikely to show up in the drill core.

In the Mittersill W-deposit, Bi concentrations are not very well understood, although they seem to be patchy, varying from area to area. A few assays of W concentrate have been made by the mine, indicating Bi concentrations of up to around 10% of the W concentrations.

3 Material

The material used in this study is a 3 m core intercept of low-grade ore from the Mittersill W-deposit. It consists of amphibolite schist with minor quartz veining in addition to the scheelite mineralization and it is associated with the SD gneiss ore body. The diameter of the drill core is 36 mm. XCT-XRF scanning is a non-destructive technique. However, to enable separate assays of subsections in this study, the drill core was divided into \(\sim \)5 cm pieces using natural core breaks and rock saw, yielding 57 samples with a total weight of 8.2 kg.

4 Methods

4.1 XCT-XRF Scans

XCT data from the GeoCore X10, hereafter the drill core scanner, comes in the form of a grid of 3D pixels (a.k.a. voxels) with reconstructed X-ray attenuation values. In the standard scan scenario, which is used in this study, each voxel is cubic and has a size of 200 × 200 × 200 µm. This voxel size is expected to be sufficient for the vast majority of scheelite found in the deposit as only some of the fine-grained scheelite would remain unresolved. In the literature, X-ray attenuation values are often referred to as XCT grayscale values, and they are measures of how much of the X-rays coming from the source are absorbed or scattered by the material. In addition, the reconstructed attenuation values are to some degree affected by measurement noise and XCT artifacts [see 29].

X-ray attenuation values, hereafter referred to simply as attenuation values, for materials depend primarily on the atomic numbers of the atoms, but also, to a lesser degree, on mass density. In geosciences, attenuation values are therefore tightly linked to mineralogy, even though some minerals span a range in chemical composition.

In this work, the X-ray source operated at a tension of 120 kVp and a current of 300 µA.

Usually, the drill core scanner scans consecutive drill core pieces of up to 1 m per sample tube, but in this case, the \(\sim \)5 cm samples are stacked in tubes with spacers in between to avoid any potential mix of data from different samples in the assay comparison. The machine was pre-configured with a list of minerals which are expected to be found at Mittersill (actinolite, albite, anorthite, biotite, celsian, ferro-hornblende, magnesio-hornblende, microcline, muscovite, powellite, quartz, and scheelite). This prior is used to improve the XRF matrix correction used to convert XCT-XRF data to surface concentrations [see 8].

A simple and straightforward way to measure the volume of certain minerals, or mineral phases, in a sample is to employ global thresholds in attenuation, where volumetric concentrations are obtained by dividing the number of voxels in a certain attenuation range by the total number of rock-containing voxels. More advanced methods, like fitting surfaces to grains in XCT data could be employed [15], but that goes beyond the scope of this paper.

The precision in the XCT data can be addressed through repeatability measurements. In this work, the samples were scanned 20 times each, to extract the standard deviations in the measured volumetric concentrations.

4.2 Comparative Assays

Post scanning, the pieces were crushed to 70% passing 2 mm, and the entire samples were pulverized to more than 85% passing 75 microns. In between each sample, the pulverizers were cleaned with a barren material. Sample preparation in terms of crushing and pulverizing as well as analysis in the form of lithium borate fusion XRF (W), aqua regia digestion followed by ICP-MS (Bi) and whole rock ICP-AES (all other reported elements) were carried out by ALS. Hereafter, the lithium borate fusion XRF analysis results are referred to as assays.

Six samples made of certified reference materials purchased from OREAS were submitted to ALS along with the rock samples. Tungsten concentrations for these, as measured with the method labeled ME-XRF15b (fusion followed by XRF), all fell within the 95% confidence limits of the certified values.

5 Results

5.1 XCT Data

An attenuation histogram can give insights into the dominating minerals or mineral phases in the scanned material. Such a histogram is shown in Fig. 1. Main peaks can be seen at around 0.94 and 1.30, where the latter is a broad peak with a potential sub-peak at around 1.47. The histogram has a tail toward very high values.

Fig. 1
figure 1

The voxel count versus XCT grayscale values for a scan of all drill core samples in this study. The peak corresponding to quartz is marked in cyan and the peak corresponding to mainly amphiboles is marked in gray, whereas high attenuation values mainly corresponding to scheelite are marked in red

Full size image

The peak at 0.94 corresponds to quartz whereas the peak at 1.30 mainly corresponds to the amphiboles. The part above around 10 in attenuation mainly corresponds to scheelite, although some bismuth mineral is also contributing to these values for a couple of samples.

By mapping the attenuation ranges of the two peaks and the high (> 10) attenuation tail to specific colors and opacities, and projecting the mapped voxel values on 2D using alpha compositing, the XCT scans can be visualized as shown in Fig. 2. The selected colors are also shown in Fig. 1.

Fig. 2
figure 2

2D images of the entire core volumes rendered from XCT data of all 57 drill core pieces analyzed in this study. Attenuation values corresponding to quartz (cyan), mainly amphiboles (gray) and scheelite (red) are highlighted. Most attenuation values are selected to be mainly transparent, whereas high attenuation values are rendered as being predominantly opaque to bring out the ore mineral. Two samples are rendered at a higher resolution on the top to illustrate the details available

Full size image
Fig. 3
figure 3

Concentration statistics from destructive, comparative assays for all 57 samples. Medians (orange lines), first and third quartiles (boxes), as well as minimum and maximum values (whiskers) are shown. Bi exceeded the upper limit of detection of 250 ppm for one sample

Full size image

Volumes, through a total rock voxel count, is another dataset retrieved by the drill core scanner. Moreover, as the machine weighs each loaded sample tube, measured density can be provided. However, densities in the current work were determined by manually weighing each sample, combined with the volume estimate based on XCT voxel counting. Densities of the \(\sim \)5 cm samples span a range from 2.8 to 3.6 g/cm\(^{3}\), with a mean of 3.1 g/cm\(^{3}\).

5.2 Concentrations of Elements

The drill core scanner’s standard XCT-XRF routines report average elemental concentrations of 25% Si, 8% Ca, 6% Fe, 4% Al, and 1.0% K for the scanned material, which is comparable to the average fusion composition ICP-AES values of 23% Si, 7% Ca, 7% Fe, 8% Al, and 0.6% K. Figure 3 shows some concentration statistics for elements present based on the comparative assays.

Elements with atomic numbers greater than 56 are found in low concentrations of less than 100 ppm per element and sample, based on the lithogeochemistry whole rock analysis, except for one instance of Pb (1660 ppm) and two instances of Bi (> 250 ppm and 142.5 ppm). Corresponding concentration values for elements in the atomic number range from 37 to 56 are less than 600 ppm, except for one instance of Mo (1440 ppm).

Bi is detected in all 20 scans of the high Bi sample with an average of 5500 ppm and a relative standard deviation of 35.3%.

Fig. 4
figure 4

Lab assay tungsten concentrations versus volumetric concentrations of scheelite-like XCT attenuation values (above 10.0) for 57 drill core samples with lengths of about 5 cm. Error bars in the x-direction show standard deviations from 20 repeated scans, whereas error bars in the y-direction indicate the precision ALS states. The blue dotted line and blue text show the outcome of a fit to the blue data points, whereas the green dashed arrow shows how one sample moved due to the bismuth compensation (see Sect. 5.3). The average ore grade reported in [13] is shown as a loosely dotted line

Full size image

5.3 Tungsten Concentrations

The most important results from this paper are shown in Fig. 4 where the 57 tungsten concentrations from assays are plotted against volumetric concentrations of XCT attenuation values above 10.0. A very strong correlation is present between the two parameters, with an R\(^{2}\) value of 0.99 if the high Bi sample is excluded.

The only outlier has a high concentration of Bi, which apparently also contributes to the high attenuation values. 34 bismuth containing minerals found at the Mittersill W-deposit (https://www.mindat.org/loc-191189.html) have an average density of \(\rho _{\mathrm {Bi\_minerals}} = 7.1\) g/cm\(^3\) and an average bismuth concentration, \(c_{\mathrm {Bi\_minerals}}\), of 59%. For a rock density, \(\rho \), and a bismuth concentration, \(c_{\textrm{Bi}}\), we therefore expect this element to contribute to the following volumetric concentration of high (> 10) attenuation values, \(c_{\textrm{10}}\):

$$\begin{aligned} c_{\textrm{10}} = \frac{c_{\textrm{Bi}} \rho }{c_{\mathrm {Bi\_minerals}} \rho _{\mathrm {Bi\_minerals}}} \end{aligned}$$
(1)

If the presence of Bi is compensated for by moving the outlier data point in Fig. 4 along the x-axis, with an amount corresponding to \(c_{\textrm{10}}\) in Eq. 1 based on scan data on bismuth concentrations and densities, then that data point also falls close to the linear relation.

The repeated measurements allow us to compute the standard deviation in volumetric concentrations in the high attenuation phase for each sample. The outcome is presented as error bars in Fig. 4 whereas the error bars in the assay concentrations indicate the precision ALS states for the particular method. At volumetric tungsten concentrations above or equal to 100 ppm, the average relative error is 12.7%, but it increases to 51.0% for the samples below this limit.

In the first round of scans, high attenuation values were absent in 12 samples and those samples all had tungsten assay concentrations of less than 10 ppm. Moreover, an additional 8 samples do not have any high attenuation values on the sample surface, despite having 29 ppm of tungsten on average according to the assays.

With the XRF system, tungsten is detected in 7 of the 37 samples with confirmed non-zero tungsten concentrations based on the assays, indicating that the tungsten content in the outermost layers is in many cases too low for this particular XRF system and scan time.

5.4 Energy Consumption

With the typical 350 W electrical power consumption of the XCT-XRF drill core scanner, about 0.1 kWh is consumed per analyzed meter of drill core.

6 Discussion

Matrix amphiboles and quartz are very distinct features in the XCT data and the associated attenuation histogram. On the other hand, the relative lack of prominent peaks above 1.6 in attenuation might be explained by minerals occurring in low concentrations in combination with the so-called partial volume effect [30], i.e., a mixing of materials within a voxel, although tomographic artifacts could also play a role.

So has the right decision been made regarding the attenuation limits of scheelite? In Fig. 1, there is a change in the derivatives slightly above 10 in attenuation, which could correspond to the tip of a broad peak. With a theoretical scheelite density, \(\rho _{\textrm{scheelite}}\), of 6.117 g/cm\(^{3}\), a theoretical tungsten concentration in scheelite, \(c_{\textrm{W, scheelite}}\), of 63.85% and an average sample density, \(\rho _{\textrm{avg}}\), of 3.1 g/cm\(^{3}\), we would expect the linear relation in Fig. 4 to have a slope, s, of:

$$\begin{aligned} s = \frac{\rho _{\textrm{scheelite}} c_{\textrm{W, scheelite}}}{\rho _{\textrm{avg}}} = 1.26 \end{aligned}$$
(2)

The theoretical slope is rather close to the observed slope of 1.41, so our selected limit should be reasonably good without any further manual tuning, a process which could potentially lead to overfitting.

The precision in volumetric concentrations of attenuation values above 10 is somewhat similar to tungsten assay errors (see Fig. 4). Note that each \(\sim \)5 cm sample was only scanned for about 20 s (not counting the repeatability analysis) and that smaller errors would be expected for 1 m samples. XCT errors and artifacts are present to some extent, but they do not seem to limit analytical capabilities judging from the repeatability results and the correlation with tungsten assay values.

With the aim to identify all significant contributions to high attenuation values, the XRF of the drill core scanner performs well in terms of the identification of samples with high concentrations of elements with atomic numbers greater than 36 in this geological setting.

Volumetric concentrations of XCT high attenuation values are a more suitable method than the XRF in the drill core scanner for the quantification of tungsten in this study. A strong correlation with lab assay data is seen, but in contrast, tungsten is only detected by the XRF in a fraction of the samples with confirmed non-zero tungsten concentrations. One reason is the lack of scheelite at the drill core surfaces in contrast to the total volume, i.e., a manifestation of the so-called nugget effect, which is further emphasized by 8 samples without high attenuation values at the surface. However, the XRF is still needed to catch other elements with high atomic numbers like Bi (83).

Bi is significantly less abundant than W in the Mittersill W-deposit. Nevertheless, Bi is expected to cause some minor interference on the XCT-based W grades, mainly through increased noise levels. This can be seen in Fig. 4, although we expect less extreme Bi concentrations when averaging over larger drill core segments instead of the 5 cm sections used in this study.

Figure 4 suggests that the drill core scanner could be utilized for on-site analysis in geological settings similar to those in the area studied here, i.e., where A) scheelite is the main ore mineral B) the scheelite grain size is predominantly at least about 200 µm, like in Mittersill ore C) the host rock has relatively low X-ray attenuation like this form of amphibolite or quartz, and D) where elements with high atomic numbers (\(\gtrsim \)36), except for W, occur only at moderate or low concentrations.

It is worth noting that exploration is another potential application area where the presence, lack, or abundance of high attenuation values could steer the explorer in the right direction toward an ore body.

With its non-destructive nature and energy consumption of only \(\sim \)0.1 kWh/m of analyzed drill core, an on-site deployment can be very energy efficient as compared to standard assay methods which include sample preparation in terms of both crushing and grinding and may include transportation of core to a distant laboratory. However, the Mittersill tungsten mine has an on-site assay laboratory, so transportation is negligible in this particular case.

7 Conclusion

In the right geological settings, the XCT-XRF technology may be competitive with respect to standard lab assays for certain elements. The statement is, in this study, backed up by measurements of tungsten concentrations, relative errors, and lab assays on Mittersill scheelite ore.

XCT-XRF is a method suitable for rapid on-site analysis, which, at least in some scheelite deposits, seems to be fit for purpose. Future trials could confirm whether or not the geological requirements are met in other zones of the deposit and at different sites.

Irrespective of grade quality, the technology has several advantages. Extremely high spatial resolution, as compared to standard lab assay sampling, could enable higher precision ore body modeling. Very short turnaround times, at least a few days faster than an on-site laboratory, might allow for dynamic drill plans and, at some sites, the whole core could be analyzed instead of half cores due to the non-destructive nature of the technology. Using whole core would give significantly better sample support, yielding more representative grades. Used on-site, the method has a low energy consumption of only \(\sim \)0.1 kWh per meter of drill core.