3.1. Evaluation of Chemical Classification Reliability
In order to verify the CCSEM-EDX routine, it was first used to characterize the six known dust samples previously described above. A total of 500 particles were analyzed in each sample, and the chemistry classification results are presented in Table 3
Results for the kaolinite and calcite analysis confirmed that respirable-sized particles from these samples are reliably classified by the CCSEM-EDX routine. For the calcite sample, of the 500 particles analyzed, 99% were classified as carbonates and the other 1% were categorized as carbonaceous. This may indicate that a small number of particles were either misclassified—perhaps due to filter background interference for very thin particles—or that very minor contamination occurred during dust collection. For the kaolinite sample, in which all particles were expected to be alumino-silicates, 100% of the particles were indeed classified as alumino-silicates (It should be noted that, per an earlier classification scheme that still included a “mixed carbonaceous” category, 89% of particles in this sample were classified as alumino-silicate and 11% were mixed carbonaceous. Consistent with prior work [27
], this provides further evidence that the mixed carbonaceous particles are actually thin and/or small alumino-silicate particles in many instances.).
For the quartz sample, 92% of the particles were classified as quartz with minor fractions of carbonaceous (4%) and alumino-silicate (4%). Unlike the kaolinite and calcite materials, which required no size reduction prior to collection of respirable particles, the quartz material had to be pulverized. While care was taken to thoroughly clean the pulverizer prior to the introduction of each new material, it is possible that some contamination of the quartz sample occurred due to carry over of dust particles from other samples previously prepared in the apparatus (e.g., coal or shale). It is also quite possible that the source material itself contained impurities.
As mentioned above, the coal, shale, and rock dust product samples were all generated from raw materials, and thus were expected to contain some impurities. For the coal sample, 92% of particles were classified as carbonaceous, with minor fractions (3% or less) showing up in all other defined compositional categories. For the shale sample, 88% of the particles were classified as alumino-silicates, with another 6% as coal, 1% as carbonates, 3% as quartz, and 2% as heavy minerals. The presence of carbonaceous and carbonate constituents in this material is consistent with separate thermogravimetric analysis (TGA) of several samples. The TGA results showed roughly 5% coal and 2% carbonate by mass [52
]. For the rock dust sample, 88% of particles were classified as carbonates, and another 7% were classified as alumino-silicates. These results are also in good agreement with separate analyses. XRD results on this material (donated by the mine partner that supplied it) showed 91% carbonate minerals and 5% alumino-silicate minerals by mass [52
3.2. Evaluation of Reproducibility
Following verification that the CCSEM routine can accurately classify particles into the chemical composition categories of interest, reproducibility of results was also evaluated. For this, the automated routine was run three separate times on each of the ten field samples shown in Table 1
. In each case, the sample orientation within the SEM instrument was different during each scan (i.e., the sample was either removed from the SEM or rotated between scans). This ensured that different areas of the sample were analyzed during each scan, and thus agreement of results between scans should indicate that enough particles across enough of the sample area were analyzed to adequately represent the sample characteristics. For all samples and all scans, a total of 500 particles were analyzed, and most scans were completed in less than 20 min, which equates to an analysis rate nearly 25× faster than the manual dust characterization method (i.e., about 25 vs. 1 particle per minute).
Results were compared with respect to particle cross-sectional diameter, aspect ratio, and chemistry number distributions (Figure 4
, Figure 5
and Figure 6
). For most of the samples, very little difference is seen between diameter (Figure 4
) and aspect ratio (Figure 5
) results from the three different scans. The chemistry distribution results (Figure 6
) exhibit some slight variability (e.g., percentage of carbonaceous particles varies somewhat in Sample 6).
shows the estimated mass (as opposed to number) distribution of particles in each chemistry category. For this, the spherical diameter, and then volume, of each particle was calculated using an assumed short to intermediate dimension ratio (S:I) value for each respective category. (The short dimension is theoretically the length of the particle in the direction perpendicular to the plane in which the long and intermediate dimensions are observed, with different mineral types tending to have characteristic S:I values.) Then, using the particles spherical volume and an assumed specific gravity (SG) value for its category, the mass was calculated. The basis for these calculations, along with assumed values for each chemistry category, is provided in Sellaro et al. [28
]; For the total alumino-silicate category, the SG and S:I values given for alumino-silicates in [28
] were used; For particles in the HM category, SG values of 7.9, 2.5 and 4.5 were used for particles that were observed to be dominated by iron, aluminum and titanium, respectively. Figure 7
illustrates that carbonaceous particles make up much less of the total dust by mass than by number, which is due to the relatively low specific gravity of coal compared to other primary dust constituents. On the other hand, the abundances of quartz and carbonate, and often alumino-silicate, particles by mass appear relatively higher than by number. The quartz and carbonate categories have relatively high specific gravity and S:I values, whereas the specific gravity is high but the S:I value is low for the total alumino-silicate category.
To determine whether the data collected during different scans of the same sample were statistically different, the Freeman-Halton test was performed on each pairwise data set for a given sample with respect to particle diameter, aspect ratio, and chemistry number distribution results. This test is a two-sided exact test of independence, which outputs a p
-value representing the likelihood of the dependence of two data sets [53
]. Resulting p
-values can be used to determine if, for example, the particle diameter distribution found for Sample 1 by Scan 1 agrees with that of Scan 3. The null hypothesis of each test is that the pair in question agrees (i.e., is statistically similar), and the alternative is that the pair disagrees. At a 95% confidence level, p
-values > 0.05 indicate pair agreement, while p
-values < 0.05 indicate disagreement.
The results of all Freeman-Halton tests and chemistry are displayed in Table 4
. All comparisons of particle diameter and aspect ratio distributions from different scans on the same sample were found to be in agreement, and all except one comparison of particle chemistry distributions (Sample 6, Scan 1 vs. Scan 3) were found to be in agreement. This suggests that the automated routine generally analyzes a sufficient number of particles across a sufficient sample area to yield reproducible results, meaning that they are representative of the entire filter area prepared for analysis. This is consistent with observations in other studies of CCSEM-EDX (e.g., see [37
]), which have shown very good reproducibility of particle chemistry classification results—particularly for major category classes—when a relatively large number of particles across a relatively large number of fields are analyzed. While repeatability of the automated routine (i.e., the ability to consistently reproduce results for the same particles, or a large number of particles in the same fields) was not investigated here, others using CCSEM-EDX for fine dust analysis have evaluated this and generally found that repeatability is quite good (i.e., given consistent operating conditions and classification criteria) [37
Beyond verifying that the CCSEM-EDX routine produces reproducible results on respirable coal mine dust samples, Figure 4
, Figure 5
, Figure 6
and Figure 7
underscore the usefulness of such results by illustrating how dust collected in different mine environments can vary. For example, Samples 7 and 9 exhibited relatively high percentages of quartz particles (on a number- and mass-basis), and these samples were both collected in a mine with predominantly sandstone roof rock, which is known to frequently have relatively high respirable quartz concentrations (by mass). Samples 3 and 6–9 all had relatively high percentages of alumino-silicate particles, and these were taken near cutting (i.e., at the continuous miner machine or just downstream in the return entry) or drilling (i.e., roof bolting) activities, where significant dust from rock strata can be generated. Schatzel, using bulk analytical techniques including X-ray fluorescence (XRF), XRD, and Fourier transform infrared spectroscopy (FTIR), has previously reported that roof and floor rock can be major sources of respirable silica and alumino-silicates [55
]. Samples 6-8 also had slightly higher aspect ratios than other samples, which is generally consistent with expectations for platy minerals, but the observed differences here are fairly subtle.
Further, Samples 1, 2 and 10 were found to have relatively high percentages of carbonates. These samples were taken in areas where significant rock dusting activities were occurring. While little information is currently available on the relative contribution of rock dust products to the total airborne respirable dust concentration in underground coal mines, Colinet and Listak have shown that the respirable fraction of particles in these products can be more than 30% (by volume) [57
]. Assuming that rock dust is the primary source of carbonate particles in the mines represented in the current study, the results presented here indicate that enough respirable rock dust particles may become airborne to contribute significantly to the total respirable dust concentration. Although rock dust products are not generally considered to pose health hazards, from an operational perspective, this issue is of increasing importance given that the total respirable dust concentration limit in US coal mines has just been reduced [58
]. Moreover, recent research has suggested that finer rock dust products, which presumably contain even higher respirable fractions, should be more effective in reducing explosibility hazards [59
Notably, most samples were not observed to be dominated by carbonaceous (i.e., coal dust) particles. This may seem somewhat counterintuitive, though historical studies have shown that the “ash” or total mineral (i.e., non-coal) fraction of dust in coal mines can be relatively high (i.e., 30% or more by mass) [2
]. Understanding the actual composition of the non-coal respirable dust fraction may well be important for advancing the understanding of health outcomes associated with mine dust exposures. For instance, recent work by Cohen et al. has identified both silica and silicate particles in the lung tissues of severely diseased coal miners [60
], but casual links between specific exposure characteristics and such disease pathology has yet to be established.
The fact that relatively small particles (i.e., less than 3 µm in cross-sectional diameter) appeared to dominate the field samples analyzed here is not surprising necessarily. The Dorr Oliver cyclones used for sample collection are designed to discard particles with aerodynamic diameters greater than about 10 µm; and they have a penetration efficiency of about 50% around 4 µm at the 1.7 L/min sampling flow rate used. Though cross-sectional and aerodynamic diameters are not equivalent for most particles, this does mean that the results in Figure 4
likely underestimate the relative proportions of large particles (i.e., greater than 3 µm in cross-sectional diameter). Nonetheless, the difference between the abundance of particles in the 0.95–2.0 µm and 2.0–3.0 µm size classes shown in Figure 4
should provide a good comparison of particles in these narrow ranges.
In the context of health implications, a better understanding of the abundance and characteristics of very small particles, including those below the effective analysis limit set by the 1000× magnification utilized here, would be highly valuable. Adaptation of the CCSEM-EDX routine outlined in the current work to analyze sub-micron particles is the focus of ongoing work by the authors. This will require analysis at higher magnification (reducing data acquisition rate), and further modification of the chemistry classification criteria. While the practical limit for particle sizing by SEM is about 0.1 µm, chemical analysis may be limited well above this due to interference by the sample (i.e., filter) background. For samples on PC filters, discrimination between high-carbon particles (i.e., those in the carbonaceous and carbonate categories) is anticipated to be particularly challenging. The challenges of SEM-EDX work (including CC work) on sub-micron particles has been reviewed in detail by the EPA [37