Recognition of coal from other minerals in powder form using terahertz spectroscopy

Currently a significant fraction of the world energy is still produced from the combustion of mineral coal. The extraction of coal from mines is a relatively complex and dangerous activity that still requires the intervention of human miners, and therefore in order to minimize risks, automation of the coal mining process is desirable. An aspect that is still under investigation is potential techniques that can recognize on-line if the mineral being extracted from the mine is coal or if it is the surrounding rock. In this contribution we present the proof of concept of a method that has potential for recognition of the extraction debris from mining based on their terahertz transmission. © 2020 Optical Society of America under the terms of the OSA Open Access Publishing Agreement


Introduction
While there is an important effort to transit from fossil fuels to renewable energy sources, the current energetic requirement in several parts of the world will still depend on the use of oil and coal for several years. [1,2] Countries such as China, use considerable amounts of coal in their energy production, and this implies that mining coal is an important activity that still employs many workers that are subject to uncomfortable and risky work conditions. Therefore, there is an interest to automatize, as much as possible all the steps of the coal-mining process in order to minimize the requirement for human intervention, and hence, the risk associated with this activity. [3][4][5] One particular aspect that requires appropriate technological solutions, is the recognition of coal from the surrounding rock sections while extracting it from the mines. Various techniques have been proposed in the past for this. For instance, the recognition of the coal-rock interface including methods based on the hardness difference between the coal bed and the rock layer, [6,7] which result in changes in the shearer operation parameters, such as applied voltage, current, motor speed, etc. [8] Additionally, the idea to monitor the temperature of the cutting tool has been explored. [9,10]. Methods based on the acoustic signals produced at the point of contact of the tool with either coal or rocks have been explored, [11] among several other techniques. [12][13][14] However, these methods are not always suitable for coal mines since the contrast in hardness of some types of rock is not large enough from that of coal for appropriate recognition.
The terahertz band of the electromagnetic spectrum, located between the microwave and the infrared regions, was only accessible to scientists about three decades ago. The introduction of the technique called terahertz time-domain spectroscopy (THz-TDS) developed in the late 1980s opened the possibility of performing spectroscopic measurements in this band of the electromagnetic spectrum [15]. The number of applications that terahertz has found since then is enormous [16], and ranges from study of materials [17][18][19][20] to non-destructive testing in biology [21][22][23], chemistry [24,25], industry [26][27][28] and art [29][30][31]. Terahertz technology was first proposed to recognize coal-rock interface by Wang and co-workers [32]. They presented the classification of compressed powder pellets of rock and coal using the absorption or the refractive index spectra from 0.2 THz to 1.6 THz for the classification model, demonstrating the potential of the technique in coal-rock interface recognition. Yet, the complexity and time required to fabricate the compressed tablets cannot meet the needs of on-line coal-rock interface identification, and the cost of a broad-band terahertz time-domain spectrometer is too high to be an appropriate solution with potential of broad real-world application.
In this article we present a method based on terahertz spectroscopy of powders that allows to distinguish several types of coal from various rocks. The method is based on quantifying the losses in the powders caused by scattering, which are related to both the grain size, and the refractive index of the powder, which in turn is a characteristic of each type of mineral.

Results
Terahertz waves have frequencies between 100 GHz and 10 THz correspond to wavelengths between 3 mm and 30 µm. Powders with grain sizes in that range are expected to exhibit significant scattering in this band. When the wavelength and the scattering center dimensions are comparable, the appropriate formalism to treat their interaction is what is called the Mie theory.
[33] While we will not go into the details of the derivation and assumptions of this theory, we will explore the possibility of material recognition in powder form using the transmission through the mineral powders.

Terahertz dielectric properties of bulk minerals
Before we get started with the modeling, we need to know the refractive indices of the bulk materials involved. Three types of coal: Anthracite (CA), Lignite (CL) and Fat coal (CF) as well as five types of rock usually found in coal mines Carbonaceous mudstone (RCM), Mudstone (RM), Conglomerate (RC), Limestone (RL), Siltstone (RS) were chosen. Samples of all of these minerals were obtained from the China Coal Science and Technology Museum and Dr. Weining Xie from the China University of Mining and Technology. The refractive index of all these minerals in "solid" form were measured as described in the Methods section over the band between 75 GHz and 2 THz. As seen in Fig. 1 the refractive indices show very little optical dispersion and the imaginary part of the refractive index is negligible. For the purposes of this initial modelling it is enough to know that all values fall between 1.9 and 2.9.

Theory of scattering in powders at THz frequencies
The Mie theory of scattering [34] assumes a spherical particle of radius a and refractive index n, and predicts that the extinction cross-section is given by where ω is the angular frequency, c is the speed of light, Here and y i (z) are the spherical Bessel and Newmann functions, not to be confused with the Bessel functions of the first and second kind usually denoted by J i (z) and Y i (z), and denotes the derivative with respect to the corresponding independent variable (x or y). Notice that i is used for the index, while i is used for the imaginary unit. In practice the series shown in Eq. (1) can not be calculated with an infinite number of terms. We calculated the number of terms M to truncate the sum by using the empirical rule M = min{n ∈ Z|n ≥ x + 4.3x 1/3 + 1}. From the cross-section it is possible to calculate the transmittance where N = a −3 is the density of particles and d is the thickness of the powder layer. Since in the experiments that we will present in the following sections, the particles could be separated by size intervals, the transmittance was calculated and averaged for ten particle sizes within the experimental intervals and d=3.6 mm. The transmittance for three different size intervals is shown in Fig. 2 for refractive indices between 1.9 (light color) and 2.9 (dark color). It is worth mentioning that this calculation was performed assuming that the material was non-absorbing and non-dispersive, which is a reasonable assumption based on the refractive indices presented in Fig. 1. The immediate conclusion we can draw from these plots is that it might be possible to distinguish between the various rocks and coals, since the transmission of their powders is dependent on their refractive indices which are, in general, different.

Recognition of mineral powders
Powders of the three coal and five rock types were prepared as described in the Methods section. Sieves with different hole sizes were used to separate the powders into eleven different ranges, resulting in (3 + 5) × 11 = 88 different powder samples. The powders were placed in cuvettes consisting of flat-faced polyethylene windows with a cavity of 3.6 mm separation between walls (see Fig. 3(i)). The transmission of the powders was measured resulting in the plots shown in Fig. 3. We can see that, at least qualitatively, the trend as function of the grain size is consistent with the theoretical scattering calculations presented earlier. Although at first glance the spectra of all the coal and stone samples seem to be almost identical, it is possible to see that subtle changes emerge between them. The maximum amplitudes of the spectra are different for the various minerals, and the drop in transmission at higher frequencies follows a different shape. Additionally, the variation of the curves as function of the grain size is also different, for instance, Anthracite (CA) and Limestone (RL) show greater changes in their spectral behaviour with particle size in comparison to Lignite (CL). In order to identify the rock and coal powders based on their terahertz transmission, we propose to construct parameters of the form where f i and f j are frequencies chosen for each grain size, such that the separation of p for the different minerals is maximized. By plotting two parameters for a different choice of f i we are able to generate a 2-dimensional parametric plot. Such plots are shown in Fig. 4 where the frequencies used are shown in THz on the axes (for visual clarity without units). As seen in the different plots, significant dispersion of the various minerals is seen. Coals are shown in "warm" colors (red-yellow) and with open symbols (*,×,+), while rocks are shown in "cold" colors (green-blue) and with closed symbols ( ,•, , , ) for easy identification. In addition, rectangles are shown around the regions where the points that correspond to coals are provided as a guide-to-the-eye. In some cases, such as the particle range between 101 µm and 238 µm, shown in Fig. 4(h), perfect separation of all minerals is possible, indicating that it could be possible to distinguish between these materials by simply using tree monochromatic THz sources and detectors on appropriately chosen grain sizes. Yet, in such a technology additional aspects such as standing waves will have to be taken into account [35].  Fig. 3, "warm" colors (red-yellow) are used for coals, while "cold" colors (green-blue) are used for rocks. As a "guide-to-the-eye" we also included rectangles that indicate the regions where the parameters that correspond to the three different coals are found.

Discussion and conclusions
We performed a characterization of the terahertz dielectric properties of a series of coals and rocks that are commonly found around them in a mine. Firstly, we found that all minerals have relatively small extinction coefficients (0.01<κ<0.06) and can, to a certain extent be regarded as reasonably transparent dielectrics with relatively little optical dispersion. In addition we found that rocks have slightly higher (2.01<n<2.88) refractive indices than coals (1.94<n<2.07). With additional experiments we were able to determine that powders of those minerals have a strong dependence on the grain size as well as moderate dependence on the refractive index of the material. This ties up very well with Mie scattering theory that explains the dependence on both variables. Furthermore, using two parameters based on transmittance at three specific frequencies, it was possible to distinguish between different minerals. After further development this technique could represent the working principle of an online mineral discrimination system if the appropriate machinery to produce powders reliably and swiftly is incorporated. Furthermore, this could be achieved with only three terahertz monochromatic sources and detectors, which are more economical and less sensitive to the operation conditions than a full THz-TDS system. This could be used for feedback of automated mining machinery in order to optimize the coal extraction process without the intervention of personnel in the mines. This technology, when further developed, has the potential to reduce the risks that mining personnel are currently subjected to when operating machinery in the mine, since the inspection of the mineral by a worker present in a mine would be avoided.

Sample selection and preparation
We selected Anthracite, Fat coal and Lignite as coal samples for this study since they cover broadly different qualities of coal. Five types of rock: mudstone, carbonaceous mudstone, siltstone, conglomerate and limestone, common in coal-bearing formations were also selected for the experiment. Mudstone is a sedimentary rock of complex composition, mainly composed of clay minerals, a hydrous silicate ore containing aluminum and magnesium, followed by clastic minerals, epigenetic minerals, and ferromanganese and organic matter. Carbonaceous mudstone is also a sedimentary rock with an organic carbon content of 6%-40%. Its main component is clay minerals followed by quartz, muscovite and a small amount of feldspar. Limestone is a carbonate rock with calcite as its main component. Siltstone is mainly composed of quartz. Finally, conglomerate is an heterogeneous metamorphic rock.
In order to get the refractive indices of the rock and coal samples selected, flat slabs of each material were made using a power saw. The faces were subsequently polished using sand paper in order to obtain flat and parallel faces for the slabs. The final thickness of each slab was measured by using a digital vernier.
Powders of each mineral were prepared by electric grinder and an agate mortar and pestle. Then the particulates were separated by eleven sieves with different hole sizes. The size graded particles were dried at 120 • C for ∼25 min in a furnace. The particle size of all eleven size intervals were measured by the LEICA microscope. In order to obtain quantify the particle size range. Particles chosen randomly from each mineral and sieve micrograph, such as the ones shown in Fig. 5 were measured. The interval range reported is the resultant of the average plus minus 1 standard deviation of the sizes found for each sieve. As the shape of the particles is irregular, the particle size is determined by the longest line on the two-dimensional image. The resulting particle size intervals are shown explicitly in the different panels of Fig. 4.
The powders were placed in a sealed "cuvette" with polyethylene windows, which is highly transparent across the entire terahertz range, separated by a polymethyl methacrylate 3.6 mm-thick spacer with a 1 cm hole which formed the powder cavity.

Terahertz time-domain spectroscopy
The measurements were made using terahertz time-domain spectrometer based on a 1550 nm ∼60 fs Er:fiber mode locked laser. The pulses were split into two. One part was sent to a delay variable line and guided by an optical fiber onto a photoconductive emitter, the other half was guided to a photoconductive detector. The terahertz transients produced in the emitter were collimated and focused onto the samples by high-density polyethylene lenses. Subsequently the terahertz radiation transmitted was also collected and focused onto the detector using high density polyethylene lenses. Further details on the terahertz time-domain technique and setup can be found in [36,37].
The measurements were taken in a nitrogen atmosphere in order to prevent atmospheric water vapour absorption. The refractive indices were obtained according to the procedure described in [38]. The transmittance of the powder samples is given by whereẼ sample andẼ reference are the Fourier transforms of the waveforms acquired for each sample and a reference recorded in the absence of sample. Each powder sample was measured in three different positions three times at each position, also the three references were acquired in between each position reference in order to account for possible slow drifts of the spectrometer, meaning that a total of 1056 spectra were acquired for the various powder samples, and their respective references.