sensors

Article Fast 2D -Induced Mapping of Rare Earth Elements in Rock Samples

Peter Seidel 1,*, Sandra Lorenz 1, Thomas Heinig 1, Robert Zimmermann 1 , René Booysen 1,3, Jan Beyer 2 , Johannes Heitmann 2 and Richard Gloaguen 1

1 Helmholtz-Zentrum Dresden-Rossendorf, Helmholtz Institute Freiberg for Resource Technology, Chemnitzer Str. 40, 09599 Freiberg, Germany; [email protected] (S.L.); [email protected] (T.H.); [email protected] (R.Z.); [email protected] (R.B.); [email protected] (R.G.) 2 Technische Universität Bergakademie Freiberg, Institute of Applied Physics, Leipziger Straße 23, 09599 Freiberg, Germany; [email protected] (J.B.); [email protected] (J.H.) 3 School of Geosciences, University of the Witwatersrand, Johannesburg 2000, South Africa * Correspondence: [email protected]; Tel.: +49-351-260-4492

 Received: 23 April 2019; Accepted: 9 May 2019; Published: 14 May 2019 

Abstract: Due to the rapidly increasing use of energy-efficient technologies, the need for complex materials containing rare earth elements (REEs) is steadily growing. The high demand for REEs requires the exploration of new mineral deposits of these valuable elements, as recovery by recycling is still very low. Easy-to-deploy sensor technologies featuring high sensitivity to REEs are required to overcome limitations by traditional techniques, such as X-ray fluorescence. We demonstrate the ability of laser-induced fluorescence (LIF) to detect REEs rapidly in relevant geological samples. Weintroduce two-dimensional LIF mapping to scan rock samples from two Namibian REE deposits and cross-validate the obtained results by employing mineral liberation analysis (MLA) and hyperspectral imaging (HSI). Technique-specific parameters, such as acquisition speed, spatial resolution, and detection limits, are discussed and compared to established analysis methods. We also focus on the attribution of REE occurrences to mineralogical features, which may be helpful for the further geological interpretation of a deposit. This study sets the basis for the development of a combined mapping sensor for HSI and 2D LIF measurements, which could be used for drill-core logging in REE exploration, as well as in recovery plants.

Keywords: laser-induced fluorescence; rare earth elements; imaging sensor; optical spectroscopy; reflectance spectroscopy

1. Introduction The rise of modern high-technology industries, such as semiconductor manufacturing or automotive engineering, has been accompanied by an increased complexity of products, which now include a great variety of elements. Many of the new material components designed during the last decades, such as light-emitting diodes, permanent magnets, and catalysts, contain large amounts of rare-earth elements (REE), a group of 17 metals comprising the lanthanoid group, scandium, and yttrium. The increasing demand for REEs, reaching a global production of 130,000 tons in 2017 [1], led to extended exploration and mining activities over the last decade. Despite their name, REEs are not “rare” in absolute number, but are rarely concentrated and mostly occur homogeneously and are usually finely disseminated in their host rocks. The low relative abundance, combined with a heterogeneous distribution of potential ore bodies on a global scale, makes exploration for REEs challenging. For larger exploration campaigns, many kilometers of drill cores are extracted and selected

Sensors 2019, 19, 2219; doi:10.3390/s19102219 www.mdpi.com/journal/sensors Sensors 2019, 19, 2219 2 of 14 rock pieces are analyzed by geochemical methods, such as mass spectrometry or electron microscopy. These techniques are time- and cost-consuming, destructive, and can only be performed for small segments of the entire drill core. To overcome these obstacles, several non-destructive alternative techniques, which are fast and can be used for in-line analysis, have been tested for core logging. These techniques include X-ray fluorescence [2], laser-induced breakdown spectroscopy [3], and optical reflectance spectroscopy [4]. The latter, also known as hyperspectral imaging (HSI), has proven capable both for indirect remote sensing of REEs by identifying their host rocks using space and airborne detectors [5], and for the direct identification of REEs in drill cores when the distance between sample and detector shrinks to cm-m scale [6]. The good sensitivity in sub-% range for several REEs such as neodymium (Nd) and dysprosium (Dy) [7], together with high acquisition speed and robustness, promoted the introduction of commercially available core logging tools based on reflectance spectroscopy. However, another opportunity to identify REEs selectively and sensitively is laser-induced fluorescence spectroscopy (LIF). This technique has been employed extensively in the last two decades for REE detection in synthetic crystals [8] and various minerals (see overview in [9]). We recently started to investigate the potential benefits of combining hyperspectral imaging and laser-induced fluorescence to profit from the advantages of both techniques, such as their easy scalability, high acquisition speed, and high sensitivity for LIF [10,11]. Based on the idea of developing a new sensor, which is capable of identifying REEs in complex rocks, we pursue a systematic approach to extend our knowledge on the applicability of the LIF-HSI combination. In a first fundamental study, the signatures of synthetic REE phosphates and fluorides for reflectance and laser-induced fluorescence spectroscopy were identified and correlated to physical processes within these model substances, extending the spectral database for further studies [9]. In a second step, we investigated the spectra of REE-bearing mineral grains from different deposits all over the world [10], building up a database for both HSI and LIF spectra for the most abundant REE minerals. The increasing complexity of the materials (e.g., mixed REE salts with variable stoichiometry) led to the appearance and extinction of several absorption and luminescence peaks, which complicated the interpretation of obtained spectra. However, a combination of the HSI and LIF techniques proved to be beneficial for the identification of rare earth elements in these materials. These studies focused mainly on spatially homogeneous, small samples with high amounts of REEs (up to 60 wt%). In this article, we aim for the next step towards a potential sensor for REE detection in drill cores, investigating common rock samples from current exploration areas with a combination of reflectance and laser-induced fluorescence spectroscopy. The emphasis of this study is placed on the development of surface mapping by LIF. When using rocks, several new constraints have to be taken into account: the relative amount of REE-containing phases in the low- or even sub-% range, the complex host rock matrices, and the high spatial heterogeneity, in particular for finely grained samples. Thus, a careful interpretation of obtained spectra with respect to signal position and shape is needed, based on the findings from the previous studies on REE salts and minerals. In addition, scanning electron microcopy-based X-ray mapping (known as mineral liberation analysis, or MLA) is employed to validate the findings from the HSI and LIF maps and to correlate the detected REE areas to the mineral phases.

2. Materials and Methods We collected several rock specimens from one South African and three Namibian deposits, which were prepared in our labs to ensure a fitting sample size for our spectroscopy experiments. For several samples, the surfaces were especially smoothed to compare them to unprepared surfaces. The non-invasive techniques (HSI and LIF) were performed before the MLA measurements, due to the latter changing the state of the surface irreversibly. Sensors 2019, 19, 2219 3 of 14

2.1. Rock Sampling and Preparation From a collection of 13 rock samples, we used two rocks from two Namibian deposits, Lofdal Farm and Epembe, for the feasibility tests of 2D LIF mapping for REE detection in rocks. The Lofdal Alkaline Carbonatite Complex in central Namibia is a well-studied REE deposit [12–14], exhibitingSensors 2019 a,high 19, x FOR enrichment PEER REVIEW of heavy rare earth elements (HREE, Eu–Lu) compared to light3 of rare15 earth elements (LREE, La–Sm). Thus, rock samples from this deposit show high concentrations in a broad earth elements (LREE, La–Sm). Thus, rock samples from this deposit show high concentrations in a range of lanthanides, making it ideal to test 2D LIF mapping. In the Epembe deposit, critical elements broad range of lanthanides, making it ideal to test 2D LIF mapping. In the Epembe deposit, critical (Nb, Ta, P, LREE) are hosted within two high-grade zones of hydrothermal origin [15,16]. The REE elements (Nb, Ta, P, LREE) are hosted within two high-grade zones of hydrothermal origin [15,16]. patternThe REE is dominated pattern is dominated by LREE by and LREE mineralization and mineralization is hosted is hosted by (in by order (in order of importance) of importance) REE-rich apatite,REE-rich monazite, apatite, andmonazite, secondary and secondary REE-fluorocarbonates. REE-fluorocarbonates. All rock All samplesrock samples were were collected collected during a fieldduring campaign, a field campaign, where most where promising most promising sampling sampling spots spots had beenhad been identified identified by by preliminary preliminary remote sensingremote observations sensing observations of the area, of the combined area, combin withed awith detailed a detailed survey survey on the on ground.the ground. These These samples weresamples analyzed were by analyzed conventional by conventional whole-rock whole-rock techniques, techniques, such as such X-ray as di X-ffraction,ray diffraction, X-ray fluorescence,X-ray andfluorescence, inductively-coupled and inductively-coup plasma massled plasma spectrometry mass spectrometry in a laboratory in a laboratory environment. environment. Thesetups The and parameterssetups and used parameters for these used methods for these are methods described are described in more detailin more in detail [17]. in [17]. To obtain a planar surface, the collected rock specimens were cut and then sliced into 10 mm To obtain a planar surface, the collected rock specimens were cut and then sliced into 10 mm thick rock pieces, which were subsequently gritted to obtain a low surface roughness of thick rock pieces, which were subsequently gritted to obtain a low surface roughness of approximately approximately 15 µm (Figure 1). For high-resolution MLA measurements, the rock pieces were µ 15 furtherm (Figure gritted,1). Forfixed high-resolution onto a glass substrate, MLA measurements, and finally lapped the rock and pieces polished were until further a surface gritted, fixed ontoroughness a glass substrate, of about 0.25 and µm finally was lapped reached. and Importan polishedtly, until the thickness a surface of roughness the rock piece of about was 0.25 keptµ m was reached.higher than Importantly, 1 mm. Otherwise, the thickness some parts of the of rockthe sa piecemple became was kept partially higher transparent than 1 mm. to visible Otherwise, light, some partswhich of the distorts sample HSI became and LIF partiallyanalyses. transparent to visible light, which distorts HSI and LIF analyses.

(a) (b)

2 FigureFigure 1. RGB1. RGB images images of of the the surface surface from from the the 40 40 20× 20 mm mm2 large large rock rock pieces pieces from from the the rare rare earth earth element element (REE) deposits: (a) Epembe (NA-RZ-05); ×(b) Lofdal Farm (NA-RB-02). (REE) deposits: (a) Epembe (NA-RZ-05); (b) Lofdal Farm (NA-RB-02). 2.2.2.2. Laser Laser System System and and Data Data ProcessingProcessing ForFor the the LIF LIF experiments, experiments, we used used a asetting setting in ina da arkroom darkroom laboratory, laboratory, as described as described previously previously [11]. [11]. TheThe samples samples were were exposed exposed toto two different different laser laser ex excitationcitation : wavelengths: 325 nm 325 (UV) nm and (UV) 442 and nm 442 nm (). The UV and blue lines were generated with a dual- Kimmon HeCd and (blue). The UV and blue lines were generated with a dual-wavelength Kimmon HeCd gas laser their beam powers were adjusted to 5 mW and 20 mW, respectively, by a neutral density filter. For andseveral their beamexcitation powers tests, werean Nd:YAG adjusted laser to with 5 mW an excitation and 20 mW, wavelength respectively, of 532 nm by (green) a neutral was density used. filter. ForDue several to acquisition excitation time tests, limits an (see Nd:YAG in Section laser 3.2), with we used an excitation an unfocused wavelength beam on the of 532sample nm for (green) the was used.mapping Due to experiments acquisition (Figure time limits 2). The (see laser in Section beam wa 3.2s), directly we used guided an unfocused towards the beam sample on the stage, sample for thepassing mapping two experiments tilted flat mirrors. (Figure For2 several). The laserindividual beam point was spectr directlya, the guided beam towardswas guided the across sample an stage, passingaperture two and tilted focused flat mirrors.by a parabolic For severalmirror. The individual Gaussian point spot diameters, spectra, the i.e., beam where was the guidedintensity across is an aperture>1/e2 of and the focused maximum by aintensity parabolic formirror. the focused The Gaussianand unfocused spot diameters, beams and i.e., the where corresponding the intensity is >1/excitatione2 of the maximumpower densities, intensity are summarized for the focused in Table and 1. unfocused beams and the corresponding excitation power densities, are summarized in Table1.

Sensors 2019, 19, 2219 4 of 14

SensorsTable 2019 1. ,List 19, x ofFOR Gaussian PEER REVIEW spot diameters and resulting average excitation power densities for4 of the 15 employed laser configurations. Table 1. List of Gaussian spot diameters and resulting average excitation power densities for the 2 Excitationemployed laser Wavelength configurations. Condition Spot Diameter/nm Power Density/(W/cm ) focused 0.16 24.9 Excitation325 nm–UV Wavelength Condition Spot Diameter/nm Power Density/(W/cm2) unfocused 0.71 1.3 focused 0.16 24.9 325 nm–UV focused 0.18 78.6 442 nm–blue unfocused 0.71 1.3 unfocused 0.79 4.1 focused 0.18 78.6 442 nm–blue unfocused 0.79 4.1 In addition to the previous studies [10,11], a remotely controlled, x-y moving stage was employed to generateIn addition a LIF map to bythe scanning previous the studies sample [10,11], according a remotely to a pre-programmed controlled, x-y raster.moving The stage micrometer was screwsemployed of the to stage generate are capablea LIF map of by precise scanning movements the sample in according 1-µm steps. to a pre-programmed Commonly used raster. full-sample The scansmicrometer had a step screws (pixel) of the size stage of 0.5 are capable0.5 mm of2 toprecise 1 1movements mm2, resulting in 1-µm in steps. the acquisition Commonly ofused 800 to full-sample scans had a step (pixel)× size of 0.5 × 0.5 mm× 2 to 1 × 1 mm2, resulting in the acquisition of 3200 individual luminescence spectra per scan. The sample was fixed on top of the x-y stage, aligned 800 to 3200 individual luminescence spectra per scan. The sample was fixed on top of the x-y stage, parallel to the moving directions to acquire the full sample area. The outermost millimeter of the rock aligned parallel to the moving directions to acquire the full sample area. The outermost millimeter of piecethe was rock not piece scanned was not to scanned avoid sample to avoid edge sample effects. edge effects.

FigureFigure 2. 2.Sketch Sketch of of setup setupused used forfor laser-inducedlaser-induced fluorescence fluorescence (LIF) (LIF) spectroscopy spectroscopy experiments, experiments, including laser excitation source, sample on an x-y-stage, monochromator, and a charge-coupled including laser excitation source, sample on an x-y-stage, monochromator, and a charge-coupled device (CCD) detector. device (CCD) detector. Afterwards, the luminescence signal of the sample was collected by a parabolic mirror and Afterwards, the luminescence signal of the sample was collected by a parabolic mirror and guided across two different long pass filters towards an Acton SP2560 triple grating monochromator guided across two different long pass filters towards an Acton SP2560 triple grating monochromator (300 gr/mm grating, blazed at 750 nm). The wavelength-dispersed photons were finally recorded by (300a gr Princeton/mm grating, Instruments blazed SPEC-10: at 750 nm).100BR_eXcelon The wavelength-dispersed CCD camera, enablin photonsg a detection were finally of photons recorded with by a Princetonwavelengths Instruments from 340 SPEC-10:100BR_eXcelon to 1080 nm. Since the spectr CCDometer camera, images enabling a wavelength a detection range ofof photons170 nm at with wavelengthsonce, four from (blue 340 excitation) to 1080 nm. to five Since (UV the excitati spectrometeron) maps images with adifferent wavelength wavelength range of ranges 170 nm were at once, fouracquired (blue excitation) and assembled to five (UVafterwards excitation) by post-proce maps withssing diff erentsoftware wavelength developed ranges in-house. were Data acquired from and assembledwavelengths afterwards below by360 post-processing nm and above 1070 software nm were developed excluded in-house. from further Data analysis from wavelengths because of the below 360 nmlow andcamera above sensitivity 1070 nm at werethe edges excluded of its wavelength from further range. analysis because of the low camera sensitivity at the edgesThe obtained of its wavelength raw luminescence range. data was corrected for spectral response of the setup and stored asThe a 3D obtained data cube, raw in luminescence analogy to the datahypercube was corrected created by for the spectral acquisition response of hyperspectral of the setup absorption and stored as a 3D data cube, in analogy to the hypercube created by the acquisition of hyperspectral absorption data.

Sensors 2019, 19, 2219 5 of 14

For 2D visualization, the LIF images showed the intensity distribution in the two spatial dimensions for a given spectral band (grey scale) or in a three-band combination (RGB false color).

2.3. Further Experimental Methods for Cross-Validation Two spectroscopic methods were employed for cross-validation of the 2D LIF mapping results, hyperspectral absorption spectroscopy (HSI) and mineral liberation analysis (MLA). A detailed comparison of different HS imagers for REE detection was given in a previous study [11]. For the experiments presented in this publication, we used an FX 10 push broom scanner from SPECIM, mounted in a SiSuRock drill core scanner framework. The scanner acquires the data line-by-line, while the sample is moving at constant speed. This sensor provides a good spatial resolution, which resulted in our case in a pixel size of 0.5 0.5 mm2. One spectrum covers a wavelength range from × 400 to 1000 nm, equally divided into 224 bands. The spectral resolution of this camera, expressed by the full width at half maximum, is about 5.5 nm. Radiance values were converted to reflectance using a pre-calibrated poly(tetrafluoroethylene) panel (Spectralon SRS-99) with >99% reflectance in the visible-near- range of the electromagnetic spectrum. The MLA experiments were performed by employing a combination of an FEI Quanta 650F scanning electron microscope with two Bruker Quantax X-Flash 5030 energy-dispersive X-ray spectrometers and the MLA 3.1.4 software package for semi-automated data acquisition [18]. Further, the software was used for data processing and evaluation. The polished and unpolished thin sections were analyzed with a grain-based X-ray mapping (GXMAP) mode. Detailed information about MLA and the offline data processing can be found in [19,20]. For the GXMAP measurements, the electron beam was accelerated by a voltage of 25 kV, resulting in a 10 nA probe current. The mapping was done frame by frame, with each frame having a size of 2000 2000 µm2 and a frame resolution × of 500 500 pixels leading to a pixel size of 4 4 µm per pixel. The distance between each X-ray × × measurement was set to 10 pixels and the measurement time to 7 ms.

3. Results and Discussion The sampling of the rock pieces was guided by prior knowledge of the investigated REE deposits obtained by remote sensing techniques, such as satellite and airborne multispectral imaging, and from the knowledge of experienced geologists during the ground survey. The presented samples stem from iron-rich carbonatite trenches (for Lofdal) and calcitic carbonatite trenches (for Epembe). The overall mineralogy of the samples was described by the prospectors as carbonatite-goethite mixtures (for Lofdal) and apatite grains in a calcitic host rock (for Epembe). Geochemical analysis of the whole-rock assays revealed a comparatively high amount of REEs (~0.5% of total rare earth oxide) for both samples.

3.1. LIF Spectroscopy As we draw our focus towards complex rocks in the LIF experiments, we observe an expected high variability of both shape of the spectra and position of the REE-related signals across the sample (Figure3). For example, at an excitation wavelength of 325 nm (UV), distinct sharp features, which are indicators of the presence of REE3+ ions, can be seen in the spectra of the spots A and B in Figure3. The luminescence of REE3+ ions exhibit unique features, which are related to their characteristic electronic configuration [21,22]. The sharpness of the emission lines is a result of the screened 4f-4f intraconfigurational transitions, which remain comparably unaffected by the chemical environment [23]. Thus, almost no losses due to multiple phonon emissions occur during the excited state of the REE3+ ion, leading to a well-defined emission energy. These sharp lines appear for spots A and B at 575 nm, 750 nm, 800 nm, 978 nm, and in the range of 865–925 nm. Based on the extended research by Gaft et al. [9], in combination with results from our previous studies on REE salts and minerals [10,11], we can attribute these signals to individual REE3+ ions, such as Dy3+, Eu3+, Nd3+, and Er3+, taking the emission peak wavelength and the shape of the spectra into account. Sensors 2019, 19, 2219 6 of 14 Sensors 2019, 19, x FOR PEER REVIEW 6 of 15

FigureFigure 3. (3.a )(a RGB) RGB image image of of Lofdal Lofdal sample sample marked marked with with three three points points (spots (spots A, A, B, B, and and C). C). (b )(b Focused) Focused LIFLIF spectra spectra acquired acquired at at the the three three marked marked spots. spots. They They exhibit exhibit di ffdifferenterent shapes shapes and and varying varying relative relative intensitiesintensities of individualof individual peaks peaks (right), (right), which canwhich be attributedcan be attributed to individual to REEindividual3+ ions. REE The3+ excitation ions. The wavelengthexcitation of wavelength the laser was of 325the nmlaser (UV). was Note325 nm the (UV). logarithmic Note the ordinate logarithmic and the ordinate offset for and the individualthe offset for curvesthe individual for clarity. curves for clarity.

InIn contrast contrast to to transition transition metals metals ions,ions, such as as Fe Fe3+3 +oror Ti Ti4+,4 the+, the crystal crystal field field splitting, splitting, induced induced by the bypresence the presence of different of diff elecerenttrostatic electrostatic environments environments for a rare forearth a rareion, is earth extraordinarily ion, is extraordinarily low (~100–200 low cm-1) 1 (~100–200[23,24]. A cm result− )[23 of,24 this]. Aeffect result is ofthe this variable effect isrelative the variable intensities relative of the intensities split emission of the lines, split emission only if the lines,rare only earth if ion the is rare hosted earth in ion different is hosted matrices. in different Most matrices. of the sharp Most signals of the exhibit sharp signalsonly one exhibit peak onlywithin onethe peak resolution within of the the resolution obtained ofspectra. the obtained In contrast spectra., the characteristic In contrast, thepattern characteristic between 865 pattern nm and between 925 nm 865is nmmost and likely 925 nmsolely is most related likely to solelyNd3+. It related shows to a Nd multiplet3+. It shows of emission a multiplet lines, of emissionbecause of lines, the becauserelatively 4 ofhigh the relatively crystal field-induced high crystal field-induced splitting of its splitting 4F3/2 level, of its fromF3/ 2whichlevel, the from radiation which the is radiationemitted. is emitted. EachEach spectrum spectrum in Figurein Figure3 exhibits 3 exhibits several several broad features broad difeaturesfferent different from the REEfrom signals, the REE spanning signals, a wavelengthspanning a ofwavelength a few hundred of a few nanometers. hundred Thesenanomete peaksrs. are These more peaks pronounced are more in pronounced the rock spectra in the than rock inspectra the REE than mineral in the or REE saltmineral experiments. or REE salt We expe attributeriments. these We multiple, attribute overlapping these multiple, emissions overlapping to the luminescenceemissions to of the defects luminescence in the host of rock defects matrix. in Probably,the host therock excitation matrix. Probably, with high-energy the excitation light (UV) with activateshigh-energy several light transitions (UV) activates of states several within transitions the rock-forming of states crystals. within For the therock-forming detection of crystals. REEs in For rock the samples,detection these of spectralREEs in features rock samples, are less utile, these even spectral preventing features an identificationare less utile, of even REE luminescencepreventing an signals.identification Several possibleof REE matrix luminescence luminescence signals. centers Several are reported possible in thematrix literature luminescence and are summarized centers are inreported an extensive in the review literature [25]. Anand explicit are summarized interpretation in an and extensive attribution review of these [25]. signalsAn explicit in finely-grained interpretation rockand samples attribution remains of these uncertain. signals The in originfinely-grained of this luminescence rock samples was remains not investigated uncertain. further,The origin but of will this beluminescence part of an additional was not study. investigated further, but will be part of an additional study. TheThe comparison comparison of of the the three three di ffdifferenterent spectra spectra hints hint tos to the the high high compositional compositional variability variability of of the the rockrock sample. sample. While While the the spectra spectra of of spots spots A A and and B exhibitB exhibit REE-related REE-related luminescence, luminescence, none none of of these these featuresfeatures can can be seenbe seen in the in spectrumthe spectrum of spot of C,spot where C, where a broader a broader signal fromsignal 570 from nm to570 700 nm nm to is700 visible. nm is Accordingvisible. According to previous to studies, previous this studies, peak can this be peak attributed can be to attributed the presence to the of Fepresence3+ or other of Fe transition3+ or other metaltransition ions, such metal as ions, Mn2+ such[26,27 as]. Mn The2+ brown-rusty [26,27]. The colorbrown-rusty of the grains color in of the the RGB grains image in the supports RGB image this assumption.supports this Although assumption. the color Although of the grainsthe color in spotsof the A grains and C in and spots the A shapes and C of and their the spectra shapes appear of their ratherspectra similar, appear small rather differences similar, in small the ratio differences of the several in the REE ratio features of the hintseveral to a REE distinct features composition. hint to a Comparingdistinct composition. the relative intensitiesComparing of the most relative of the REEintensities peaks suggestsof most of a higherthe REE total peaks REE suggests content ina higher spot A (notetotal REE the logarithmic content in plotting).spot A (note Considering the logarith themic ratio plotting). of the Er 3Considering+-related doublet the ratio peaks of atthe 978 Er nm3+-related and 4 4 3+ 983doublet nm (corresponding peaks at 978 nm to the andI 983 nm I(correspondingtransition [28 to]) the with 4I11/2 respect → 4I15/2 to transition the ratio of[28]) the with other respect REE to 11/2 → 15/2 ionthe signals, ratio aof selective the other enrichment REE3+ ion of signals, erbium ina selective area B is assumed,enrichment though, of erbium a direct in quantification area B is assumed, from though, a direct quantification from the emission spectrum is often very challenging and can be erroneous. The absolute luminescence signal from a given solid material also depends on its surface

Sensors 2019, 19, 2219 7 of 14

Sensors 2019, 19, x FOR PEER REVIEW 7 of 15 the emission spectrum is often very challenging and can be erroneous. The absolute luminescence signalroughness, from a givenoptical solid density, material and alsograin depends size distribution on its surface [29]. roughness,Non-radiating optical relaxa density,tion andof excited grain sizestates distributionand defects [29]. Non-radiatingcan quench relaxationthe emissions of excited from states luminescence-active and defects can quench materials. the emissions Thus, from even luminescence-activenon-luminescent areas materials. can Thus,still contain even non-luminescent REEs. To relate areas the canLIF stillmapping contain results REEs. Toto relatethe sample the LIFchemistry, mapping resultsfurther toprobing the sample techniques chemistry, were further used (see probing Section techniques 3.2). were used (see Section 3.2). AsAs demonstrated demonstrated in in earlier earlier studies, studies, e.g. e.g. [30 [30],], the the shape shape of of the the spectrum spectrum and and the the variety variety of of identifiableidentifiable REEs REEs change change under under excitation excitation with with di ffdifferenterent wavelengths wavelengths (Figure (Figure4). 4). In general,In general, the the matrix-relatedmatrix-related luminescence luminescence in in the the emission emission wavelength wavelength range range of of 350 350 nm nm to to 650 650 nm nm is significantlyis significantly strongerstronger under under UV UV excitation excitation conditions, conditions, overlapping overlapping several several REE REE features. features. Furthermore, Furthermore, the the overall overall luminescenceluminescence intensity intensity is is clearly clearly reduced reduced using using an an excitation excitation source source at at 442 442 nm, nm, even even though though the the powerpower density density of of the the incident incident light light at at the the sample sample is is enhanced enhanced by by a factora factor of of three three (see (see Table Table1). 1). Nevertheless,Nevertheless, an excitationan excitation with with 442 nm442 gives nm gives rise to rise a new to a REE-related new REE-related peak, exhibiting peak, exhibiting a unique a shape. unique Forshape. example, For aexample, multiplet a multiplet peak between peak 580between nm and 580 630 nm nm and appears, 630 nm which appears, is attributed which is toattributed various to 3+ 3+ 3+ 3+ luminescencevarious luminescence centers from centers different from REE differentions (DyREE3+, Smions ,(Dy and3+ Eu, Sm)[3+,31 and,32]. Eu A3+ clear) [31,32]. correlation A clear betweencorrelation peak between position peak and the position corresponding and the correspon REE cannotding be REE drawn cannot in this be wavelengthdrawn in this range, wavelength since therange, three REEssince arethe often three intermixed REEs are in often natural intermixed samples. Inin addition, natural thesamples. crystal In field-induced addition, the splitting crystal andfield-induced peak shifting splitting of several and nanometers peak shifting complicate of several aclear nanometers identification complicate of the a luminescence clear identification origin. of Employingthe luminescence time-resolved origin. luminescenceEmploying time-resolved experiments luminescence could help distinguish experiments the could signals, help sincedistinguish the threethe luminescencessignals, since the show three diff luminescenceserent decay times show [33 diffe]. Inrent general, decay a combinationtimes [33]. In of general, two (or a even combination three) diffoferent two excitation(or even three) wavelengths different show excitation the best wavelengths results for show the detectability the best results of a for broad the rangedetectability of REEs. of a 3+ 3+ Whilebroad UV range excitation of REEs. is better While suited UV forexcitation identification is better of Ersuitedand for Nd identification, a blue laser of isEr more3+ and appropriate Nd3+, a blue 3+ 3+ 3+ forlaser the detectionis more appropriate of Dy , Sm for, the and detection Eu . of Dy3+, Sm3+, and Eu3+.

FigureFigure 4. Comparison4. Comparison of LIFof LIF spectra spectra from from spot spot A onA on the the Lofdal Lofdal sample sample at anat an excitation excitation wavelength wavelength of of 325325 nm nm (bottom) (bottom) and and 442 442 nm nm (top). (top). Various Various REE-related REE-relate signalsd signals are are marked marked by by dashed dashed lines. lines. Note Note the the artificialartificial enhancement enhancement of the of spectralthe spectral intensity intensity for the fo 442r the nm 442 excitation nm excitation by a factor by a of factor 20 and of the20 oandffset the foroffset clarity. for clarity.

ByBy using using a controllable,a controllable, micrometer-precise micrometer-precise positioning positioning system system in in x-y x-y directions, directions, we we are are able able to to scanscan the the sample sample in in an an automated automated way. way. Afterwards, Afterwards, the the rectangular rectangular rasterraster ofof pointpoint measurementsmeasurements is is combinedcombined toto aa spatiallyspatially resolvedresolved two-dimensional two-dimensional map map of of the the sample sample surface surface by by data data processing processing software.software. Each Each pixel pixel contains contains a full a emission full emission spectrum spec oftrum the specific of the point,specific resulting point, inresulting a data structure in a data similarstructure to the similar one of to the the hyperspectral one of the hyperspectral imagery. To visualize imagery. the To acquiredvisualize 3D the data acquired in two 3D dimensions, data in two false-colordimensions, RGB false-color maps are used, RGB which maps areare created used, which by the combinationare created by of twothe combination or three different of two channels, or three different channels, i.e. spectral bands. For example, a false-color LIF map for two REE-related luminescence signals (876 nm [Nd3+] and 983 nm [Er3+]) after excitation with a 325 nm laser is composed by normalizing the intensities of individual bands to their global maximum value and

Sensors 2019, 19, 2219 8 of 14

i.e.Sensors spectral 2019, 19 bands., x FOR PEER For example,REVIEW a false-color LIF map for two REE-related luminescence signals8 of 15 (876 nm [Nd3+] and 983 nm [Er3+]) after excitation with a 325 nm laser is composed by normalizing theattributing intensities the of band individual to a certain bands tocolor their in globalthe RGB maximum color space value (Figure and attributing 5). There, the blue band pixels to a signify certain a colorhigher in relative the RGB intensity color space of the (Figure luminescence5). There, at blue 983 pixelsnm, which signify hints a higher to higher relative enrichment intensity of oferbium the luminescencein these regions. at 983 The nm, brighter which the hints color, to higher the higher enrichment is this ofintensity. erbium inSimilarly, these regions. yellow The (mixed brighter color thefrom color, red theand higher green) ispixels this intensity.represent areas Similarly, with yellowhigher (mixedintensity color of the from 876 rednm andluminescence, green) pixels i.e., representhinting at areas higher with occurrences higher intensity of neodymium. of the 876 nmDark luminescence, pixels stand i.e.,for hintingspots where at higher no luminescence occurrences of in neodymium.the defined channels Dark pixels was stand received. for spots where no luminescence in the defined channels was received.

FigureFigure 5. 5.2D 2D LIF LIF maps maps from from the the Lofdal Lofdal sample sample after after excitation excitation with with a 325 a 325 nm nm laser laser at a at pixel a pixel step step size size of 1of mm. 1 (amm.,b): The(a,b grayscale): The grayscale images show images the relative show intensities the relati ofve a certain intensities emission of wavelengtha certain /spectralemission band,wavelength/spectral which is related band, to the occurrencewhich is related of a specific to the REE occurrence3+ ion. (c )of The a normalizedspecific REE combination3+ ion. (c) The of bothnormalized maps results combination in a false-color of both RGB maps map results of the in sample, a false-color where RGB the “red” map andof the the sample, “green” where channels the are“red” combined. and the Thus, “green” yellow channels pixels representare combined. a higher Thus, Nd 3yellow+-related pixels signal represent and dark a bluehigher pixels Nd a3+-related higher Ersignal3+-related and dark signal. blue Dark pixels areas a higher do not Er show3+-related any luminescence signal. Dark inareas the do given not wavelengthshow any luminescence ranges. in the given wavelength ranges. Although the attribution of REE-rich areas to regions with high REE luminescence is challenging (see discussionAlthough before),the attribution we assume of severalREE-rich REE-bearing areas to regions domains with on thehigh Lofdal REE sample luminescence based on is thechallenging REE luminescence (see discussion distribution: before), an we Nd assume3+-enriched several zone REE-bearing at the upper domains part and on anthe ErLofdal3+-enriched sample zonebased at theon the center REE and luminescence at the bottom distribution: of the rock piece.an Nd In3+-enriched between, therezone areat the regions upper with part very and low an luminescenceEr3+-enriched inzone the centerat the center and at and a vein at inthe the bottom upper of half. the Besidesrock piece. the sensitiveIn between, detection there are of several regions rarewith earth very elements, low luminescence the LIF mapping in the at center 325 nm and excitation at a vein contains in the the upper additional half. benefitBesides of the localization sensitive ofdetection the different of several REE sub-groups. rare earth Since elements, most of the the LI REEsF mapping occur intermingled at 325 nm within excitation the same contains mineral the phase,additional single benefit REEs canof localization be regarded of as the proxies differen fort theREE REE sub-groups. sub-groups: Since neodymium most of the for REEs light occur rare earthintermingled elements (LREE)within the and same erbium mineral for heavy phase, rare sing earthle elementsREEs can (HREE). be regarded A LIF as map proxies of the for Nd 3the+- andREE Ersub-groups:3+-related luminescence neodymium enablesfor light an rare extraction earth elem of enrichmentents (LREE) zones and alongerbium mineralogical for heavy rare features, earth suchelements as veins (HREE). and textures, A LIF map which of is the very Nd valuable3+- and Er for3+ exploration-related luminescence of potential enables REE deposits an extraction [34]. of enrichmentThe ability zones of localizing along mineralogical luminescence features, signals such in a sampleas veins depends and textures, on the which size of is the very individual valuable pixelfor exploration and the resolution of potential of the REE employed deposits [34]. sensor. For applications in the mining industry, such as drill coreThe scanning,ability of localizing the required luminescence acquisition signals time hasin a tosample be taken depends into account.on the size We of usedthe individual pixel sizes pixel of 0.5–1and mmthe resolution2, although of the the resolution employed of sensor. the detection For applications setup and in the the laser mining spot industry, would have such allowed as drill forcore 50scanning,µm-wide the increments. required acquisition Still, the acquisitiontime has to ofbe high-resolutiontaken into account. LIF We images used from pixel a sizes small of rock-piece 0.5–1 mm2, withalthough a pixel the size resolution of 0.5 of0.5 the mm detection2 (as seen setup in and Figure the 6laser) takes spot 30–40 would min, have which allowed is caused for 50 µm-wide by the × increments. Still, the acquisition of high-resolution LIF images from a small rock-piece with a pixel size of 0.5 × 0.5 mm2 (as seen in Figure 6) takes 30–40 min, which is caused by the comparably low luminescence signal emitted from the samples. For the realization of an in-line drill core scanning sensor,

Sensors 2019, 19, x FOR PEER REVIEW 9 of 15

further advances in the sensitivity of the detection system and in the motion control of the positioning system have to be implemented.

3.2. Comparison of HSI and MLA To compare the LIF maps with the results from other 2D imaging techniques, all rock pieces were investigated by HSI as well as MLA. Reflectance spectroscopy in the visible and near-infrared range is already well established for the detection of REE in geological samples [35–37]. Especially for REE-rich minerals, such as monazite or bastnaesite, pronounced absorption features, which can be used for the non-invasive detection of several REEs even in sub-% concentrations, have been reported [11,37]. On the other hand, features mainly related to Nd3+ and Dy3+ dominate the spectra, whereas other REE3+ ions exhibit no significant signal. Recently, the superior sensitivity of laser-induced fluorescence spectroscopy for the characterization of REEs in minerals has been reported [11]. Applying LIF spectroscopy to the inhomogeneous Namibian rock samples composed of mainly non-REE-bearing minerals (calcite, goethite), we could confirm their higher REE sensitivity (Figure 6). Sensors 2019, 19, 2219 9 of 14 The HS image is expressed by means of band ratios from Nd3+-related absorption features, which is a commonly used method to better visualize absorption signals [37]. The band ratios are calculated by comparablydividing the low intensity luminescence of the absorption signal emitted minimum from theby samples.the intensity For theof band realization on a shoulder of an in-line next drillto it, corewhich scanning is not affected sensor, furtherby the absorption. advances in The the green sensitivity color ofrepresents the detection a local system enrichment and in of the Fe3+ motion ions, controlwhich ofdoes the not positioning show any system luminescence, have to be thus implemented. appearing black in the LIF map. In contrast, magenta pixels are related to weak Nd3+ absorption features. Although the so-created false-color HS image for 3.2.three Comparison spectral bands of HSI exhibits and MLA local structures similar to the LIF map, the individual REE signals are moreTo pronounced compare the in LIF the maps latter with image, the resultswhich fromis also other visualized 2D imaging by the techniques, comparison all rockof individual pieces were HS investigatedand LIF spectra by HSI (middle). as well Whereas as MLA. no Reflectance sharp absorption spectroscopy features in the are visible visible and in the near-infrared range from range 400 nm is alreadyto 1000 well nm, established clear REE-related for the detection fluorescence of REE insignals geological are samplesobserved. [35 Corresponding–37]. Especially forgeochemical REE-rich minerals,analysis from such the as monaziterock, from or which bastnaesite, the piece pronounced was cut, gave absorption Dy3+ amounts features, of 0.01–0.2 which can wt%, be which usedfor are thebelow non-invasive the detection detection limit of severalHSI [11]. REEs Still, even these in small sub-% quantities concentrations, of REEs have could been be reported detected [ 11by,37 the]. OnLIF the measurements, other hand, featureswhich is mainly the main related advantage to Nd of3+ usingand Dy LIF.3+ dominateMagenta pixels the spectra, (mixed whereas color from other red REEand3 +blue)ions in exhibit the LIF no map significant visualize signal. the occurrence Recently, the of superior Nd3+-related sensitivity luminescence, of laser-induced whereas fluorescence green pixels spectroscopyshow a higher for relative the characterization intensity of matrix of REEs emissions. in minerals has been reported [11].

Figure 6. Comparison of false-color maps from the Lofdal sample. (a) Hyperspectral imaging (HSI) map Figure 6. Comparison of false-color maps from the Lofdal sample. (a) Hyperspectral imaging (HSI) (b) LIF map under 442 nm excitation. For the HS image visualization, band ratios of Fe3+-related (green) map (b) LIF map under 442 nm excitation. For the HS image visualization, band ratios of Fe3+-related and Nd3+-related (magenta) absorption features are used. The ratios are calculated by the division of (green) and Nd3+-related (magenta) absorption features are used. The ratios are calculated by the the intensities from the absorption band and from a band that is not affected by this absorption signal. division of the intensities from the absorption band and from a band that is not affected by this Magenta coloring in the LIF map corresponds to higher relative occurrences of Nd3+ ions, while green areas are correlated to the luminescence from the calcitic host rock. (c) The comparison of LIF and HSI spectra from one pixel (white squares in (a,b)) with REE occurrence reveals the differing sensitivity of each method. Dashed lines mark the position of potential REE-related absorption features in the HSI spectrum based on previous studies [36,38].

Applying LIF spectroscopy to the inhomogeneous Namibian rock samples composed of mainly non-REE-bearing minerals (calcite, goethite), we could confirm their higher REE sensitivity (Figure6). The HS image is expressed by means of band ratios from Nd3+-related absorption features, which is a commonly used method to better visualize absorption signals [37]. The band ratios are calculated by dividing the intensity of the absorption minimum by the intensity of band on a shoulder next to it, which is not affected by the absorption. The green color represents a local enrichment of Fe3+ ions, which does not show any luminescence, thus appearing black in the LIF map. In contrast, magenta pixels are related to weak Nd3+ absorption features. Although the so-created false-color HS image for three spectral bands exhibits local structures similar to the LIF map, the individual REE signals are more pronounced in the latter image, which is also visualized by the comparison of individual HS and LIF spectra (middle). Whereas no sharp absorption features are visible in the range from SensorsSensors2019 2019, 19, 19, 2219, x FOR PEER REVIEW 1010 of of 14 15

absorption signal. Magenta coloring in the LIF map corresponds to higher relative occurrences of 400 nm to 1000 nm, clear REE-related fluorescence signals are observed. Corresponding geochemical Nd3+ ions, while green areas are correlated to the luminescence from the calcitic host rock. (c) The analysis from the rock, from which the piece was cut, gave Dy3+ amounts of 0.01–0.2 wt%, which are comparison of LIF and HSI spectra from one pixel (white squares in (a) and (b)) with REE occurrence below the detection limit of HSI [11]. Still, these small quantities of REEs could be detected by the reveals the differing sensitivity of each method. Dashed lines mark the position of potential LIF measurements,REE-related absorption which is features the main in the advantage HSI spectrum of using based LIF. on Magentaprevious studies pixels [36,38]. (mixed color from red and blue) in the LIF map visualize the occurrence of Nd3+-related luminescence, whereas green pixels show aTo higher validate relative the obtained intensity results of matrix from emissions. LIF mapping and correlate the distribution of REE-related signalsTo validate to mineral the obtainedphases, we results employed from LIF MLA mapping on all androckcorrelate pieces after the distributionthe LIF measurements of REE-related took signalsplace. toSamples mineral from phases, two deposits we employed with differing MLA on lit allhologies rock pieces were aftertested the to LIFexamine measurements the influence took of place.the host Samples rock matrix from two on depositsthe detectability with di ffofering REEs lithologies in geological were settings tested to(Figure examine 7). The the influenceshown MLA of themaps host have rock been matrix resampled on the detectability to a pixel size of REEs of 0.5 in mm geological to meet settings with the (Figure pixel7 size). The of shownthe LIF MLA maps. mapsMagenta have MLA been pixels resampled stand tofor a spots pixel where size of the 0.5 minera mm tol apatite meet with is present. the pixel Green size pixels of the represent LIF maps. the Magentaoccurrence MLA of main pixels host stand rock for minerals, spots where such theas goethite, mineral apatiteFeO(OH), is present.for the Lofdal Green sample pixels and represent calcite, theCaCO occurrence3, for the of Epembe main host sample. rock minerals, To avoid such an asovershadowing goethite, FeO(OH), of the fornon-calcitic the Lofdal phases sample by and the calcite,predominant CaCO3, calcite for the phase Epembe in sample.the Lofdal To sample, avoid an the overshadowing calcite is not ofshown the non-calcitic there. The phasesdistribution by the of predominantREE luminescence calcite phasematches in thevery Lofdal well sample,to the occurrence the calcite of is notapatite, shown suggesting there. The an distribution enrichment of of REE rare luminescenceearth elements matches in the very apatite well to grains. the occurrence This effect of apatite, is well suggesting known anfrom enrichment former ofmineralogical rare earth elementsinvestigations, in the apatitewhere apatite, grains. ThisCa5[(F,Cl,OH)|(PO effect is well known4)3], was from found former to incorporate mineralogical up to investigations, 1 wt% of total whereREEs apatite,in its crystal Ca5[(F,Cl,OH) lattice [39,40].|(PO4)3 ],Moreover, was found no to luminescence incorporate up was to 1obtained wt% of total from REEs the region in its crystal where latticethe MLA [39,40 experiments]. Moreover, norevealed luminescence a goethite was vein obtained without from any the regionREE-bearing where theminerals. MLA experiments However, a revealedselective a goethiteenhancement vein without of the anyREE REE-bearing luminescence minerals. by the However,apatite host a selective and the enhancement quenching ofof the the REEREE-related luminescence luminescence by the apatite by host different and the minerals quenching could of the REE-relatedbe alternative luminescence explanations by di ffforerent the mineralsluminescence could beintensity alternative differences explanations in the forvarious the luminescence phases. Luminescence intensity diquenchingfferences inin theminerals various by 2+/3+ phases.Fe2+/3+ species Luminescence has been quenching particularly in minerals well reported by Fe in literaturespecies has[41,42]. been Nevertheless, particularly well complementary reported in literatureMLA experiments [41,42]. Nevertheless, on the same complementary Fe-rich areas showed MLA experiments no REE occurrences on the same there, Fe-rich suggesting areas showed that the noquenching REE occurrences effect there,is not suggesting present for that our the quenchingsamples. The effect good is not match present between for our samples.both distributions The good matchdemonstrates between boththe ability distributions of the demonstratesLIF technique the to ability detect of such the LIF low technique REE amounts to detect in suchcomplex low REE rock amountssamples in without complex damaging rock samples the investigated without damaging sample. the investigated sample.

FigureFigure 7. 7.(a )(a 2D) 2D LIF LIF map map from from the Lofdalthe Lofdal sample sample after excitationafter excitation with a 442with nm a laser442 nm at a laser pixel at distance a pixel ofdistance 0.5 mm. of (c) 0.5 2D LIFmm. map (c) under2D LIF the map same under conditions the same for theconditions Epembe sample.for the Epembe (b,d) For sample. comparison, (b,d) the For correspondingcomparison, the resampled corresponding mineral resampled liberation analysismineral (MLA)liberation maps analysis are shown (MLA) next maps to the are LIF shown maps, next the to depictedthe LIF mineralmaps, the phases depicted visualize mineral the occurrencephases visualiz of thee mainthe occurrence minerals (calcite, of the main goethite, minerals and apatite). (calcite, Magentagoethite, pixels and representapatite). Magenta the occurrence pixels ofrepresent REE emissions the occurrence in the LIF of images REE emissions and the related in theoccurrence LIF images ofand apatite the related as main occurrence mineral phase of apatite in the as MLA main maps. mineral phase in the MLA maps.

Sensors 2019, 19, 2219 11 of 14

Furthermore, the lithology variations in both deposits are successfully reproduced by the LIF map, when the correlation between apatite and REE luminescence is assumed. Whereas in the Lofdal sample, the few-micrometer large apatite grains are agglomerated into larger structures and intermingled with the calcite host rock, the apatite is ordered in few-millimeter large mineral aggregations within the calcite for the Epembe deposit. These findings are in good agreement with previous studies on the geology and mineralogy of the Lofdal [14] and the Epembe [16] deposits. A comparison between HSI, MLA, and LIF, where several metrological parameters for each technique are summarized, is given in Table2. It should be taken into account that both MLA and LIF are spot-scanning methods, whereas HSI is performed by a line-scanner. In general, MLA proved to be a versatile tool for the validation of LIF imaging results and concatenation of luminescence and local chemistry. Its smaller pixel size allows for a more detailed and direct identification of minerals not only limited to REE-bearing phases. However, 2D LIF spectroscopy is suited for faster and non-invasive detection of REEs and allows for a good differentiation between individual rare earth elements, whereas distinguishing single REEs by the X-ray emission lines used for MLA assignment is challenging. In addition, both the purchase and the maintenance costs are lower for LIF sensors compared to MLA sensors. HSI is the fastest method, but in our setup had the lowest spatial resolution and poorest detection limit for REEs. Both HSI and LIF are in favor of not damaging the sample, in contrast, MLA experiments require small (few cm2) samples with flat, polished surfaces, which are coated with graphite and need to be transferred to high vacuum conditions for measurement.

Table 2. Comparison of employed 2D imaging techniques for geological samples. The acquisition time is given for the measurements of 4 2 cm2 rock pieces in steps of 500 µm (HSI + LIF) and 4 µm (MLA), × respectively. Detection limits for HSI were taken from [36,43], for MLA calculated from the quantitative analysis, and for LIF from complementary electron microprobe experiments.

Parameter HSI MLA LIF Acquisition time ~1 s ~10,000 s ~1500 s Spatial resolution >300 µm >2 µm >50 µm REE detection Limit >0.03 wt% >0.01 wt% >0.01 wt% Sample preparation flat surface preferred flat surface + C-coating + transfer to high vacuum flat surface

3.3. Implications for Mineralogy and Exploration of REEs As shown in the previous section, 2D mapping by laser-induced fluorescence spectroscopy is a promising technique for identifying individual REEs and revealing their distribution in geological samples. Contrary to classical geochemical methods, such as mass spectrometry and X-ray fluorescence analysis, the samples can be measured faster and in the future possibly in-line, i.e., on a drill core scanner. Furthermore, only limited time is required for the preparation of a flat surface and this surface is not damaged by the laser beam, if a certain (high) limit of excitation power density is not exceeded. On the other hand, only information from the upper micrometers of an opaque sample are collected by LIF spectroscopy, whereas a three-dimensional image of the whole rock piece cannot be extracted. Thus, a careful sample selection and statistical interpolation of the gathered results to the whole rock body are necessary. Further advances in integrating an LIF sensor with other 3D-imaging sensors, such as dual-channel X-ray tomography, could lead to a versatile tool for minimal-invasive detection of REEs in a material stream, regardless of drill-core scanning for exploration or characterization of extracted rock materials in a mine operation. However, for a satisfactory analysis of various REEs, more than one excitation wavelength is recommended. From our research, we assume an excitation light with a wavelength of 442 nm as suitable to detect sharp REE luminescence while suppressing overlapping signals from the host rock matrix. By the addition of an excitation with 325 nm, we can clearly examine the occurrence of Nd3+ and Er3+, which serve as proxies for LREE and HREE, respectively. A LIF map with the relative abundance of LREE and HREE adds a high value to the interpretation of geological structures in the Sensors 2019, 19, 2219 12 of 14 rock samples/drill cores, aiding in the assessment of the REEs’ origin in the host rock (e.g., tracing of magmatic processes or hydrothermal fluids), and thus a better understanding of the whole deposit. In the case of the discussed Namibian deposits, LIF mapping helps us to understand their lithology and their REE enrichment zones. In the example of Lofdal, the sample shows a distinct layering, supporting the theory of a repeated overprint by hydrothermal fluids, which carried the rare earth elements. HREE are locally enriched around iron-rich veins, a state which is in agreement with a previous report on the geochemical dynamics at this deposit [14]. In the case of Epembe, it confirmed apatite as being the major host for LREE enrichment. Another challenge in the interpretation of REE LIF maps is the correlation of the luminescence intensity to the elemental concentration of the individual REE. Further studies on this specific topic have to be conducted, including comparing LIF spectroscopy with different other techniques, for example for structural characterization of the REE-bearing phase and investigating many samples from various kinds of REE deposits (e.g., ion-adsorption clays, carbonatites, and alkali-pegmatites). We have already performed several experiments for testing the quantification of REEs by LIF, but their results will be presented in a separate publication, due to the extent and complexity of this topic.

4. Conclusions We employed two-dimensional laser-induced fluorescence mapping for the identification of REEs in rock samples. After excitation with blue (442 nm) and UV (325 nm) , we could assign sharp luminescence signals to individual REE3+ ions based on previous reports. Using a remote-controlled x-y translation stage, we were able to gather LIF maps of rock pieces from two REE deposits in Namibia, enabling a precise localization of the areas with elevated REE concentrations. The effects of the inhomogeneous rock matrix, complex lithology, and signal dependency on the excitation wavelength have been examined. LIF mapping as a non-invasive imaging technique proved to be a versatile tool for REE characterization, detecting REEs even in a low concentration of 0.02 wt%. Moreover, a comparison with other widely used surface imaging methods, HSI and MLA, revealed the strengths and limitations of LIF spectroscopy. Its superior sensitivity for REEs, comparably high acquisition speed, and low need for sample preparation qualify it for usage as a sensor in the modern mining and exploration industries. The combination of LIF and MLA was especially beneficial, demonstrating, for example, the association of REEs with the apatite grains in Lofdal and Epembe and showing a significant enrichment of HREEs around iron-rich veins for the Lofdal samples, which is in agreement with previous geological studies. This outcome proves that 2D LIF mapping can help geoscientists with the interpretation of geological structures and the dynamics of an REE deposit. Further investigation using different sensors, such as electron microprobes, could help in establishing a more direct relation between REE concentration and luminescence signals and is subject of another study currently in preparation. Moreover, mechanical and electronic assembling of the main parts, used for our experiments, into a single combined LIF + HSI framework will be the next step for the implementation of an LIF sensor unit, which would benefit the REE exploration and mining industries considerably.

Author Contributions: Conceptualization, P.S. and R.G.; methodology, P.S., S.L. and J.B.; software, S.L.; validation, P.S., T.H. and R.Z.; formal analysis, P.S., S.L. and T.H.; investigation, P.S. and S.L.; resources, R.Z., R.B. and J.B.; writing—original draft preparation, P.S.; writing—review and editing, S.L., T.H., R.Z., R.B., J.B., J.H. and R.G.; visualization, P.S.; supervision, J.H. and R.G. Funding: This research was supported by EIT RawMaterials GmbH within the KAVA up-scaling project “inSPECtor”. Acknowledgments: The authors want to thank the preparation laboratory at Helmholtz-Institute Freiberg for the careful preparation of the rock specimen and Suchinder Sharma and Margret Fuchs for fruitful discussions about the origin of the REE luminescences and data processing. We thank as well the central funds of the Helmholtz-Institute Freiberg for Resource Technology for the payment of the publication costs. Conflicts of Interest: The authors declare no conflict of interest. Sensors 2019, 19, 2219 13 of 14

References

1. Gambogi, J. Rare Earths, Mineral Commodity Summaries 2017; United States Geological Survey: Reston, VG, USA, 2017. 2. Qing, S.; XinLei, Z.; Yan, Z.; Wenbao, J.; Yongsheng, L.; DaQian, H.; ShengNan, C. Development of an Online X-Ray Fluorescence Analysis System for Heavy Metals Measurement in Cement Raw Meal. Spectrosc. Lett. 2015, 49, 188–193. [CrossRef] 3. Khajehzadeh, N.; Kauppinen, T.K. Fast mineral identification using elemental LIBS technique. IFAC-PapersOnLine 2015, 48, 119–124. [CrossRef] 4. Kruse, F.; Weatherbee, O.; Peppin, W.; Bedell, R.; Calvin, W.; Taranik, J.V. HSI Mineral Mapping from Airborne, Outcrop, and Drill-Core Perspectives. Proc. SPIE 2010, 7687.[CrossRef] 5. Zimmermann, R.; Brandmeier, M.; Andreani, L.; Mhopjeni, K.; Gloaguen, R. Remote Sensing Exploration of Nb-Ta-LREE-Enriched Carbonatite (Epembe/Namibia). Remote Sens. 2016, 8, 620. [CrossRef] 6. Turner, D.; Rivard, B.; Groat, L.A. Visible to shortwave infrared reflectance spectroscopy of rare earth element minerals. Br. Columbia Geol. Surv. Pap. 2015, 3, 219–229. 7. Neave, D.A.; Black, M.; Riley, T.R.; Gibson, S.A.; Ferrier, G.; Wall, F.; Broom-Fendley, S. On the feasibility of imaging carbonatite-hosted rare earth element deposits using remote sensing. Econ. Geol. 2016, 111, 641–665. [CrossRef] 8. Lenz, C.; Nasdala, L.; Talla, D.; Hauzenberger, C.; Seitz, R.; Kolitsch, U. Laser-induced REE3+ photoluminescence of selected accessory minerals—An “advantageous artefact” in Raman spectroscopy. Chem. Geol. 2015, 415, 1–16. [CrossRef] 9. Gaft, M.; Reisfeld, R.; Panczer, G. Modern Luminescence Spectroscopy of Minerals and Materials, 2nd ed.; Springer: Cham, Switzerland, 2015. 10. Fuchs, M.C.; Gloaguen, R.; Beyer, J.; Jakob, S.; Heitmann, J. Emission spectroscopy for the identification of rare earth elements using laser-induced photoluminescence. In Proceedings of the 2016 8th Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS), Los Angeles, CA, USA, 21–24 August 2016; pp. 1–5. 11. Lorenz, S.; Beyer, J.; Fuchs, M.; Seidel, P.; Turner, D.; Heitmann, J.; Gloaguen, R. The Potential of Reflectance and Laser Induced Luminescence Spectroscopy for Near-Field Rare Earth Element Detection in Mineral Exploration. Remote Sens. 2019, 11, 21. [CrossRef] 12. Wall, F.; Niku-Paavola, V.N.; Storey, C.; Müller, A.; Jeffries, T. Xenotime-(Y) from carbonatite dykes at Lofdal, Namibia: Unusually low LREE:HREE ratio in carbonatite, and the first dating of xenotime overgrowths on zircon. Can. Mineral. 2008, 46, 861–877. [CrossRef] 13. Williams-Jones, A.E.; Wollenberg, R.; Bodeving, S. Hydrothermal Fractionation of the Rare Earth Elements and the Genesis of the Lofdal REE Deposit. In Proceedings of the Symposium on Strategic and Critical Materials 2015, Victoria, BC, Canada, 13–14 November 2015; pp. 125–130. 14. Bodeving, S.; Williams-Jones, A.E.; Swinden, S. Carbonate–silicate melt immiscibility, REE mineralising fluids, and the evolution of the Lofdal Intrusive Suite, Namibia. Lithos 2017, 268, 383–398. [CrossRef] 15. Broom-Fendley, S.; Styles, M.T.; Appleton, J.D.; Gunn, G.; Wall, F. Evidence for dissolution-reprecipitation of apatite and preferential LREE mobility in carbonatite-derived late-stage hydrothermal processes. Am. Mineral. 2016, 101, 596–611. [CrossRef] 16. Unger, G.; Zimmermann, R.; Gloaguen, R. 3D Modeling of the Epembe (Namibia) Nb-Ta-P-(LREE) Carbonatite Deposit: New Insights into Geometry Related to Rare Metal Enrichment. Minerals 2018, 8, 600. [CrossRef] 17. Rahfeld, A.; Wiehl, N.; Dressler, S.; Möckel, R.; Gutzmer, J. Major and trace element geochemistry of the European Kupferschiefer—An evaluation of analytical techniques. Geochem. Explor. Env. A 2018, 18, 132–141. [CrossRef] 18. Gu, Y. Automated scanning electron microscope based mineral liberation analysis an introduction to JKMRC/FEI mineral liberation analyser. J. Miner. Mater. Char. Eng. 2003, 2, 33–41. [CrossRef] 19. Fandrich, R.; Gu, Y.; Burrows, D.; Moeller, K. Modem SEM-based mineral liberation analysis. Int. J. Miner. Process. 2007, 84, 310–320. [CrossRef] 20. Bachmann, K.; Frenzel, M.; Krause, J.; Gutzmer, J. Advanced Identification and Quantification of In-Bearing Minerals by Scanning Electron Microscope-Based Image Analysis. Microsc. Microanal. 2017, 23, 527–537. [CrossRef] Sensors 2019, 19, 2219 14 of 14

21. Reisfeld, R. Radiative and non-radiative transitions of rare-earth ions in glasses. In Rare Earths. Structure and Bonding; Springer: Berlin/Heidelberg, Germany, 1975; Volume 22, ISBN 978-038-707-268-3. 22. Hull, R.; Parisi, J.; OsgoodJr, R.M.; Warlimont, H.; Liu, G.; Jacquier, B. Spectroscopic Properties of Rare Earths in Optical Materials; Springer: Berlin/Heidelberg, Germany, 2005; ISBN 978-3-540-28209-9. 23. Lüthi, B. Crystal field effects in rare earth systems. J. Magn. Magn. Mater. 1980, 15, 1–8. [CrossRef] 24. Elliott, R.J. Crystal Field Theory in the Rare Earths. Rev. Mod. Phys. 1953, 25, 167. [CrossRef] 25. Marfunin, S.A. Spectroscopy, Luminescence and Radiation Centers in Minerals; Springer: Berlin/Heidelberg, Germany, 1979; ISBN 978-3-642-67112-8. 26. Habermann, D.; Niklas, J.R.; Meijer, J.; Stephan, A.; Götte, T. Structural point defects in “Iceland spar” calcite. Nucl. Instrum. Methods Phys. Res. B 2001, 181, 563–569. [CrossRef] 27. Cazenave, S.; Chapoulie, R.; Villeneuve, G. Cathodoluminescence of synthetic and natural calcite: The effects of manganese and iron on orange emission. Miner. Petrol. 2003, 78, 243–253. [CrossRef] 28. Czaja, M.; Bodył-Gajowska, S.; Lisiecki, R.; Meijerink, A.; Mazurak, Z. The luminescence properties of rare-earth ions in natural fluorite. Phys. Chem. Minerals 2012, 39, 639–648. [CrossRef] 29. Seferis, I.E.; Zeler, J.; Michail, C.; David, S.; Valais, I.; Fountos, G.; Kalyvas, N.; Bakas, A.; Kandarakis, I.; Zych, E.; et al. Grains size and shape dependence of luminescence efficiency of Lu2O3:Eu thin screens. Results Phys. 2017, 7, 980–981. [CrossRef] 30. Burruss, R.C.; Ging, T.G.; Eppinger, R.G.; Samson, M. Laser-excited fluorescence of rare earth elements in fluorite: Initial observations with a laser Raman microprobe. Geochim. Cosmochim. Acta 1992, 56, 2713–2723. [CrossRef] 31. Gaft, M.; Reisfeld, R.; Panczer, G.; Boulon, G.; Shoval, S.; Champagnon, B. Accommodation of rare-earths and manganese by apatite. Opt. Mat. 1997, 8, 149–156. [CrossRef] 3+ 32. Lu, Z.; Chen, L.; Tang, Y.; Li, Y. Preparation and luminescence properties of Eu -doped MSnO3 (M = Ca, Sr and Ba) perovskite materials. J. Alloys Compd. 2005, 387, L1–L4. [CrossRef] 33. Reisfeld, R.; Gaft, M.; Boulon, G.; Panczer, C.; Jørgensen, C.K. Laser-induced luminescence of rare-earth elements in natural fluor-apatites. J. Lumin. 1996, 69, 343–353. [CrossRef] 34. Foley, N.; De Vivo, B.; Salminen, R. Rare Earth Elements: The role of geology, exploration, and analytical geochemistry in ensuring diverse sources of supply and a globally sustainable resource. J. Geochem. Explor. 2013, 133, 1–5. [CrossRef] 35. Adams, J.W. The visible region absorption spectra of rare earth minerals. Am. Mineral. 1965, 50, 356–366. 36. Rowan, L.C.; Kingston, M.J.; Crowley, J.K. Spectral reflectance of carbonatites and related alkalic igneous rocks; selected samples from four North American localities. Econ. Geol. 1986, 81, 857–871. [CrossRef] 37. Turner, D.J.; Rivard, B.; Groat, L.A. Visible and short-wave infrared reflectance spectroscopy of REE fluorocarbonates. Am. Mineral. 2014, 99, 1335–1346. [CrossRef] 38. Turner, D.; Rivard, B.; Groat, L. Rare earth element ore grade estimation of mineralized drill core from hyperspectral imaging spectroscopy. In Proceedings of the 2014 IEEE Geoscience and Remote Sensing Symposium, Quebec City, QC, Canada, 13–18 July 2014. 39. Nagasawa, H. Rare earth concentrations in zircons and apatites and their host dacites and granites. Earth Planet. Sci. Lett. 1970, 9, 359–364. [CrossRef] 40. Frietsch, R.; Perdahl, J.-A. Rare earth elements in apatite and magnetite in Kiruna-type iron ores and some other iron ore types. Ore Geol. Rev. 1995, 9, 489–510. [CrossRef] 41. Filippelli, G.M.; Delaney, M.L. The effects of Manganese(II) and Iron(II) on the cathodoluminescence signal in synthetic apatite. J. Sedimen. Res. 1993, 63, 167–173. [CrossRef] 42. Pierson, B.J. The control of cathodoluminescence in dolomite by iron and manganese. Sedimentology 1981, 28, 601–610. [CrossRef] 43. Boesche, N.; Rogass, C.; Lubitz, C.; Brell, M.; Herrmann, S.; Mielke, C.; Tonn, S.; Appelt, O.; Altenberger, U.; Kaufmann, H. Hyperspectral REE (Rare Earth Element) Mapping of Outcrops—Applications for Neodymium Detection. Remote Sens. 2015, 7, 5160–5186. [CrossRef]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).