Morphological characterisation of explosive powders by X-ray computed tomography: when grain numbers count

Ammonium nitrate (AN) prills are commonly used as an ingredient in industrial explosives and in fertilisers. Conventional techniques (such as BET or mercury intrusion porosimetry) can measure the open porosity and specific surface area of AN prill, but the closed porosity is not obtainable. This work was focused on evaluating X-ray computed tomography (XCT) as a non-destructive technique for the assessment of porosity in AN prills. An advanced data processing workflow was developed so that the segmentation and quantification of the CT data could be performed on the entire 3D volume, yet allowing the measurements ( e.g. ; volume, area, shape factor … ) to be extracted for each individual phase (prill, open porosity, closed porosity) of each individual prill, in order to obtain statistically relevant data. Clear morphological and structural differences were seen and quantified between fertiliser and explosive products. Overall, CT can provide a very wide range of parameters that are not accessible to other techniques, destructive or non-destructive, and thus offers new insights and complementary information.


Introduction
In the mining industry, the term ANFO, for ammonium nitrate / fuel oil, specifically describes a mixture of solid ammonium nitrate prills (see Figure 1a) and diesel fuel (commonly AN 94.5 % / FO 5.5 % in weight [1]), widely used as a bulk industrial explosive.While the worldwide production of AN for fertilizer is around 40,000 tonnes per day, the global ammonium nitrate market is expected to reach USD 6.18 billion by 2025 [2].One of the major performance predictors of the ANFO prills is the fuel oil retention, which is itself governed by the porosity of the AN prills.Presently, the oil retention capacity of ammonium nitrate in prilled and granulated forms is determined by means of a standardised test in the EU [3].However, this method faces technical difficulties, mainly because the porosity from different types of ammonium nitrate prills varies significantly.The porosity connected to the prill surface, open porosity, is available for oil retention.The pores not connected to the prill surface, closed porosity, are not available for oil retention but are however important for the explosive sensitivity.The current test methods cannot account for the closed porosity, which explains some of the technical limitations of such test.One way to investigate and characterise the porosity of AN prills in a more accurate and differentiating way is to use X-ray computed tomography (XCT).We believe that XCT could be an invaluable tool to the explosives community, by providing qualitative and quantitative measurements of both the open and closed porosity, and the total surface area of AN prills.In this paper, a data processing workflow was developed to extract these measurements from the high-resolution scan of a single AN prill.However, to obtain more representative data, the workflow was amended to extract the same data on a lower resolution scan covering around 20 AN prills, i.e. to extract the measurements for each individual prill grain whilst performing the segmentation on the entire 3D volume only once.

Material and method
Two types of AN prills were CT scanned, the type labelled hereafter type E used in the mining industry as a constituent in ANFO mixtures, and the type labelled type F, used as a fertilizer in farming.A first scan was performed on a single prill glued onto a carbon fibre rod (see Figure 1b), to obtain the best voxel size possible (around 2.5 µm) and assess the dimensions of the porosity.A second scan was performed on several prills contained in a polyimide tube of 4.2 mm diameter, so that a good compromise between voxel size (around 5 µm) and field of view (number of grains scanned) was attained.

AN prills
Four different types of AN prills were under investigation in this study.Two were fertilisers (F1 and F2) used in farming, and two were explosives (E1 and E2) used in the mining industry as a constituent in ANFO mixtures; the two types are displayed in Figure 1.Type E1 and E2 are similar, as E2 prills are crushed E1 prills.Therefore, the structural parameters between E1 and E2 should be similar, whilst the morphological parameters are expected to show significant differences.These two materials were selected in order to assess the sensitivity of the data processing.

Laboratory X-ray computed tomography
Two type of scans were performed to assess the AN prills (see Figure 2).A first scan was performed at the highest possible magnification on a single prill glued on top of a carbon fibre rod, so that the features of the AN prills could be evaluated with the highest resolution possible (voxel size around 2.5 µm) and the CT data processing workflow could be developed (see section 2.3).Then, a lower resolution scan was performed so that a more significant number of AN prills could be assessed (voxel size around 5.0 µm).The AN prills were scanned on a GE V|Tome|x L 180/300 system [4] equipped with a 180 kV source, a tungsten transmission target (actual focal spot size below 2 μm as determined with JIMA test pattern RTC02), a diamond window, and a GE 2000 × 2000 pixel DXR-250 detector.The voxel size was calibrated after the scans were performed, by scanning a ball bar consisting of 2 rubis spheres glued onto a carbon fibre rod and separated by a calibrated distance of 2.273 mm +/-0.001mm.The calibrated voxel size was determined by comparing the calibrated distance to the distance between the 2 spheres in the volumetric XCT data using VGstudio MAX version 3.2 [5] (Surface determination then 2 spheres were generated by fitting 25 points to the surface of the rubis, then the distance between the spheres' centres was measured).The calculated value was then employed as voxel size for each XCT scan during the data processing step.The data visualisation, processing and quantification was performed using Amira ZIB Edition version 2019.03[6].

XCT data processing
The data processing workflow presented Figure 3 was developed for the prills scanned individually.The single prill scans served as a starting point in understanding the challenges in segmenting a complex open pore network in contact with the air outside the prills.Taking a single prill allowed optimizing the entire processing workflow by getting rid of the contact points between the several prills, and the presence of the container (polyimide tube).First, the data were imported in Amira ZIB Edition, and a calibrated voxel size was used (value determined from the scanning of a calibrated ball bar).Then the entire volume was filtered using a non-local means filter, so that the noise was removed whist retaining the features of interest.The segmentation step was focused on separating the ammonium nitrate (AN) and the high-density inclusions, when present.The closed porosity was then segmented using a 3D closing operation, whilst the open porosity required a succession of growing and filling operations to be adequately selected.Once all the labels are created (AN, high density inclusion, open porosity, and closed porosity), a label analysis was performed, measuring the following parameters: surface area, volume, shape factor and Euler characteristic χ.The shape factor is defined as  3 (36 •  •  2 ) ⁄ and describes how far from a sphere the shape of a considered label is (shape factor equals 1 for a perfect sphere); whilst the Euler characteristic χ (also called Euler-Poincaré characteristic or Euler number) [7], is an indicator of the connectivity of a 3D complex structure, a measure of how many connections in a structure can be severed before the structure falls into two separate pieces.The Euler characteristic measures what might be called "redundant connectivity", the degree to which parts of an object are multiply connected [8], and is used here as an indicator of the complexity of the topology of the AN prill and associated open porosity.The specific surface values were calculated for each prill, from the ratio between the surface area of the AN label and the corresponding AN label volume, hence the unit in mm 2 /mm 3 (unit of surface area per unit of volume).It is important to mention that it not possible to directly convert this measurement into a more common mm 2 /g unit (unit of surface area per unit of mass) as the density of the bulk material is not well known and cannot be accurately defined from the XCT data for each individual prill.In order to obtain the radial distribution of the different phases, all the labels are selected together, and a distance transform (Euclidean type) is calculated on that selection.By masking the distance field with each of the individual labels, i.e.AN, high density inclusions, open porosity, and closed porosity; the individual radial profiles can be obtained.

Results & discussion
Examples of 2D slices from the 2 prill types presented in Figure 5 qualitatively show the differences in porosity and overall structures.Both fertilizer prills contain high density inclusions, whilst the explosives contain none.The porosity contents also appear to be significantly different.F1 prills have a limited amount of porosity, whilst more extensive porosity contents are visible for F2, E1 and E2.The porosity is quite rounded for F2 whilst for the explosive materials, there is often a large round cavity in the centre of the prills and more elongated, or crack-like pore channels, running radially.These observations are in good agreement with the structural features of the prill porous network described in [10].However, it is impossible to distinguish the open porosity from the closed porosity from the 2D slices.It can also be noticed the significant differences in terms of size of the prills, the explosives materials have much larger prill sizes, whilst the fragmented prills from E2 have much sharper edges, compared to the fully round E1 prills.
• Importing raw data and voxel size calibration Second, the morphological aspects are presented in Figure 6b, where the AN and the open porosity are assessed.The analysis of the AN yields more representative results, but it is also of interest to consider the open porosity itself, as it is a true representation of the volume available for the fuel oil to soak into.There are clear differences between E1 and E2, with E1 shape factor being a factor of 2 higher than that of E2, and E1 Euler characteristic being a factor of 3 higher than that of E2.Those differences reflect the smaller prill fragments from E2 compared to E1 (resulting from crushing), thus having less complex structures per unit of volume (also resulting in greater measurement deviation).
Figure 6c shows the correlation between AN volume and surface area for each prill of the XCT volume.There is a linear relationship for all the materials under investigation.There is a good agreement between E1 and E2 over the entire range of surface areas, with E2 having more data points towards the low AN volumes, corresponding to the prill fragments.The fertilisers have much greater AN volumes, due to the larger prill size, but E1 has associated low surface areas (< 100 mm 2 ), whilst E2 has associated large surface areas (> 150 mm 2 ).To be able to better compare the materials, the specific surface area values, defined as the surface area of AN per unit of AN volume, were calculated.Figure 6d shows the linear relationship between the specific surface values determined by X-ray computed tomography and the oil retention values determined according to the European regulation [3].With this result, we demonstrate that XCT could be used to predict the performance of explosives over a very wide range of porosity content.
Based on a distance transform operator, the radial volume fraction of each phase can be determined.The plots for each material are gathered in Figure 7.For sample F1, the AN content decreases radially when moving inwards the grain, whilst the closed porosity increases, particularly in the innermost 20 % of the grains, to reach a volume fractions as high as 20 %, whilst the average closed porosity content is only 1 %.The high-density inclusions are mostly present in the outermost 10 % and innermost 50 % but are relatively well distributed.Both explosive materials exhibit similar radial distribution profiles for each of their phases.The AN content drops rapidly in the outermost 10 % of the prills, then decreases slowly on the central 10 % to 80 %, and finally drops rapidly over the innermost 20 %.The evolution of the open porosity content is opposite to that of the AN, whilst most of the closed porosity is located in the first 0.05 % of the prills which corresponds roughly the outermost 100 µ m layer.The effect of the crushing can be seen close to the centre of the prills, with E1 having an open porosity content plateauing at 100%, corresponding to the central cavities observed in Figure 4

Conclusion
Overall, the results presented here demonstrate that XCT can be successfully applied to the structural and morphological characterisations of AN prills in a non-destructive manner, as a wide range of morphological parametars can be extracted, in addition to overall volume fraction values.The workflow developed was capable of quantifying the morphological differences between E1 and E2 samples, whith E2 being crushed E1 prills.The differences in shape factors and Euler characteristics are consistent with the morphology of the fragments (E2) versus the spherical prills (E1).Clear differences were also evidenced between the fertilisers and explosive materials.Being able to differenciate between explosive and fertilizer AN prills is extremely important for safety reasons and this is one of the core missions of BAM (Bundesanstalt für Materialforschung und -prüfung).No high-density inclusions were found in the explosive materials.If no open porosity was found in F1, around 20% was found in F2, which is a level similar to some explosive prills.However, when looking at the specific surface values, the F2 value is much smaller that what was found for E1 and E2.A linear correlation was found between the specific surface area values extracted from the XCT data and the oil retention values, similarly to other results [9], demonstrating that XCT can be use to predict the performance of AN prills.
The future work will focus on comparing the present XCT results to those of conventional techniques such as BET and mercury porosimetry, as assess which morphological parameter are most relevant to the mining industry.The workflow developed here can also be applied to a a broad range of small porous parts and porous powders.
a) Single AN fertilizer prill glued onto a C fibre rod b) Multiple AN prills into a polyimide tube Figure 2. Example AN prills (a) and single prill mounted on carbon fibre rod (b).

Figure 3 .
Figure 3. Overview of the data processing workflow for single prills.Regarding the data processing of the prills in the polyimide tube, the entire data processing workflow is fully detailed in[9].A watershed segmentation is used to select the AN, air and polyimide tube materials.The individual materials (AN, open porosity, closed porosity and high-density inclusions) are then separated before the prills are separated.The determination of the structural and morphological parameters of the prills are based on the same metrics presented before.Examples of 3D renderings of the open porosity and the closed porosity are given Figure 4.

•Figure 5 .Figure 6 .
Figure 5. Examples of 2D slices for each material under investigation.The data processing workflow developed here was aimed at extracting the most relevant structural parameters of the AN prills, both on a global and a local scale, and in a quantitative fashion.The structural and morphological results obtained are gathered in Figure6.First, the volume fraction measurements (Figure6a) show that material F1 is significantly different from the other AN prills, with a much greater AN content (above 95 %).This is associated with virtually no open porosity, and 1 % closed porosity and 2 % high density inclusion.Only F2 also contains high density inclusions (around 1 %) but a much greater porosity content, both closed (3 %) and open (18 %).As expected, materials E1 and E2 are very similar, and for each constituent, the variations between E1 and E2 are within the measurement errors.In particular, the closed porosity contents are similar, whilst the open porosity content is lower for the cruched prills E2.From the 2D slices, crushed prills will likely not include the inner cavity, which is consistent with a lower open porosity content.
& 5. F2 is somewhat of a mix between F1 and E materials, with a sharp increase in open porosity in the outer 20 % of the prills but the content then stabilises around 25 % over the remainder of the prills.a) F1 prills b) F2 prills c) E1 prills d) E2 prills Figure 7. Radial profiles of AN prills.