Professional Documents
Culture Documents
Structural and Textural Characterization of Coke With Optical Image Analysis Software
CSIRO
PO Box 883, Kenmore QLD 4069, Australia Phone:
Email: Eugene.Donskoi@csiro.au
INTRODUCTION
Metallurgical coke is heavily used in ironmaking and its qualities such as strength, abrasion resistance and reactivity are
highly important for stable blast furnace operation. It has been widely shown that coke structure and texture determine coke
quality [1-7]. An understanding of the connection between coke structure, texture, strength and reactivity and the parent coal
blend gives an insight into the factors affecting coke quality and the possible means for its improvement/optimization during
parent coal blending, sizing, coking and other processing involved in coke production.
Metallurgical coke is highly porous, with its solid part consisting of two major components: Inert Maceral Derived
Components (IMDC) and Reactive Maceral Derived Components (RMDC). In this paper by coke “structure” we mean the
spatial distribution of porosity and the coke matrix, which can be further subdivided into different phases such as IMDC and
RMDC, or alternatively into walls and nodes segmented by structural identification. By the term “texture” we instead
describe the spatial distribution of different isotropic and anisotropic carbon types within the coke, identified using
bireflectance optical imaging [8-10].
To properly characterize coke structure, innovative image acquisition and analysis facilities are necessary. Image acquisition
can use different techniques such as optical, Raman or scanning electron microscopy. To segment carbon types with different
anisotropy, that is, to perform textural characterization, a specialized optical system capable of proper treatment of polarized
light is needed [8-10]. In such systems the orientation of the polarizer in the incident light path can be changed and images
are taken during polarizer step rotation from 0 to 180 degrees. Two measurements are required to uniquely characterize and
classify the varieties of different carbon types: maximum reflectance (Rmax) and bireflectance ([Rmax – Rmin], the
difference between maximum and minimum reflectance.
CSIRO has developed the unique optical image analysis software package Mineral4/Recognition4 [11-13] which is used for
the comprehensive characterization of all major ironmaking commodities: iron ore fines and lumps, sinter, pellets and coke.
Its unique structural characterization of coke [5-7] includes: automated segmentation of the coke matrix into two types of
IMDC (non-fused and partially reacted) and RMDC; segmentation into nodes and walls including their comprehensive
characterization; comprehensive characterization of porosity, including characterization of “Separated pores” and
identification/characterization of “Weak Areas”; and the identification/characterization of IMDC boundaries and their
connection with RMDC.
In order to perform textural characterization of coke, CSIRO Mineral Resources developed a rotating polarizer and the
software to control it within the Mineral4/Recognition4 platform (Figure 1). This allowed for the automated collection of
images from large coke areas under different polarizer angles. Using automated image analysis, Mineral4 creates RMax and
bireflectance maps of the coke studied and, based on them, segments five different major carbon types within coke: isotropic
and fused inerts, and also three types of fused vitrinite (low, medium and high bireflectance). These texturally identified
phases and their relationship with the separately conducted structural identification are fully characterized.
Correlation Coefficients between M40 and Cumulative
amount of IMDC down to certain size fraction
IMDC ECD (µm)
‐0.3
154 217 307 435 615 869 1229 1738 2458
‐0.4
Correlation Coefficient
‐0.5
‐0.6
‐0.7
‐0.8
‐0.9
‐1
Figure 3. Dependence of the correlation coefficient between M40 (AS1038) and the cumulative amount of IMDC down to a
specific IMDC size, expressed in equivalent circle diameter for that size (reproduced from [6]).
Porosity is a very important part of the coke structure and its effect on coke strength has been studied by many authors [1, 3,
4, 15, 16]. Mineral4 enables the user to create a coke porosity map and characterize each pore separately, as well as
separating distinct pockets of the same pore (“Pore Separation”) or, oppositely, removing very fine/weak walls revealing the
so called “Weak Areas”, which helps delineate trends in coke matrix cracking (see Figure 4).
(d) (e)
Figure 4. Porosity Separation procedure: (a) photomicrograph of coke with pore separation lines; (b) unprocessed porosity
color map (in all maps individual pores are marked with different random colors); (c) porosity color map after separation, and
Weak Area processing: (d) walls with a thickness of 4 pixels (~4.25 µm) and less removed; (e) walls with a thickness of 8
pixels (~8.51 µm) and less removed (reproduced from [6]).
Kubota et al. [2] showed that connected pores can cause coke fractures and reported a high negative correlation between the
abundance of pores with low roundness and coke strength index DI150/6. Mineral4 calculates the roundness of each pore and
measures the abundance of pores with different roundness (see Figure 5).
(a) (b)
Figure 5. Identification of porosity with different roundness: a) original fragment of a coke image; b) the corresponding color
map of porosity (red – porosity with roundness less than 0.1, cyan – roundness between 0.1 and 0.2, blue – roundness
between 0.2 and 0.3, yellow – roundness higher than 0.3).
(a) (b)
Figure 7. Schematic presentation of two structures: (a) large nodes, thin walls; b) small nodes, thick walls (reproduced from
[6]).
Such a comprehensive description of coke structure can help to explain unexpected observations. For example, for a set of
cokes made from coals with Ro-max between 0.96% to 1.56% [6] strong positive significant correlations were obtained
between porosity and a set of coke strength indices: M40 (AS1038) (0.922), I40 (0.827), M40 (BS1016) (0.809), DI3015
(0.703) and MCS (0.948) (Figure 8). The dependence was counterintuitive as it meant that higher coke porosity apparently
corresponds to increased coke strength.
82.5
60
80.0
M40 (AS1038)
77.5
MCS
55
75.0
72.5 50
70.0
45
45.0 47.5 50.0 52.5 55.0 57.5 35 40 45 50 55
Porosity Porosity
Figure 8. Dependences between porosity and coke strength indices M40 (AS1038) and MCS.
In order to explain this correlation it is important to note that porosity (measured optically) positively correlated with the
amount of vitrinite in the parent coal (see Figure 9a). Also, the roundness of pores increased with the increased amount of
vitrinite in the parent coal, especially in the presence of pore separation processing (the correlation coefficient increased from
0.558 to 0.828 after the separation procedure was applied, see Figure 9b). Donskoi et al. [6, 7] and Kubota [1] demonstrated
that coke strength increased with increasing pore roundness. Parameters such as the ratio of walls to nodes area, “Modified
Wall Thickness” and “Specific Neck Thickness” increased with increasing vitrinite, which apparently resulted in increasing
matrix strength (see Figure 9c, d). Finally, increasing vitrinite means less IMDC, the presence of which negatively affects
coke strength [2, 6]. So it can be concluded that even though the coke porosity increased with increasing vitrinite in the
parent coal, several other related factors, namely increasing pore roundness, increasing wall thickness and quantity, and less
IMDC, all combined to have a positive effect on the coke strength, which overcompensated for the presumed detrimental
effects of increased coke porosity.
0.38
55
Separated Pore Roundness
0.36
50
Porosity
0.34
45
0.32
40
0.30
40 50 60 70 80
40 50 60 70 80
Vitrinite
Vitrinite
(a) (b)
Scatterplot of Wall Area/Node Area vs Vitrinite Scatterplot of Specific Neck Thickness vs Vitrinite
1 .2 0.0024
0.0022
1 .0
Specific Neck Thickness
Wall Area/Node Area
0.0020
0.8
0.001 8
0.6
0.001 6
0.4
0.001 4
0.2 0.001 2
0.0 0.001 0
40 50 60 70 80 40 50 60 70 80
Vitrinite Vitrinite
(c) (d)
Figure 9. Dependences of coke structure parameters on the amount of vitrinite in parent coal blend: a) porosity; b) pore
roundness; c) ratio of wall to node area; d) “Specific Neck Thickness”.
(a) (b)
(c) (d)
Figure 10. Images created by Mineral4 during textural analysis for the coke area shown in Figure 2a: a) RMax image; b)
RMin image; c) bireflectance image; d) the resulting textural map: magenta – isotropic inerts; blue – fused inerts; green,
yellow, cyan – fused vitrinite with high, medium and low bireflectance correspondingly.
Figure 10 shows RMin, RMax, Bireflectance (RMax - RMin) and textural maps corresponding to the image shown in Figure
2a. The caption corresponding to the textural map describes five different phases, however only two are immediately obvious
70
60
Y-Data
50
40
30
20
0 5 10 15 20 25 30
coke number
Figure 11 Scatterplot of vitrinite abundance in the initial coal blend, fused vitrinite calculated from textural identification and
RMDC calculated from structural characterisation.
Figure 12 shows structural and textural identification of different phases for another coke with a high amount of fused inerts,
and Figure 13 shows a closer view of a small part of the image in Figure 12 (upper-left from the image center). The higher
magnification of Figure 13 enables the presence of fused vitrinite with high and medium bireflectance to be clearly seen,
even though Figure 12 may give the impression that only low bireflectance fused vitrinite (cyan) is present.
(c) (d)
Figure 12. Images and calculated maps for coke with a high presence of fused inerts: a) image obtained under polarized light
(0°); b) bireflectance image; c) structural identification; d) textural identification (refer to Figure 2 and Figure 10 for color
coding).
(c) (d)
Figure 13. A closer view of a part of the image from Figure 12 (upper-left from the image center): a) image obtained under
polarized light (0o); b) RMax image; c) bireflectance image; d) textural map (refer to Figure 10 for color coding).
For the majority of coke indices their correlations with texturally-identified inert or reactive phase abundances are higher than
for their correlations with structurally-identified phase abundances. For example, for a set consisting of 22 cokes made from
coals with RoMax between 1.13 and 1.56, correlation coefficients of the total abundance of texturally-identified inerts with
I40, DI 30/15, CRI were (-0.921), (-0.828) and (0.573) respectively (Figure 14; note that some indices were not available for
each coke). Correlations coefficients of the total abundance of structurally-identified inerts were (-0.862), (-0.714) and
(0.451) respectively, all of them statistically significant at the 95% level. If isotropic and fused inerts were considered
separately, the correlation coefficients of their abundance with CRI, for example, were (0.740) and (-0.535) respectively. The
explanation is that coke reactivity index has high positive correlation with the abundance of isotropic inert because isotropic
carbon reacts faster than anisotropic carbon represented by fused inerts, which in turn gives a negative correlation with CRI.
Overall inerts abundance gave a positive correlation coefficient (0.573) with CRI due to the fact that the abundance of
isotropic inerts in the majority of cokes was significantly higher than that of fused inerts.
For reference, the research technique has revealed many statistically significant dependences between parent coal
composition, coke properties, and the textural and structural characteristics of the coke matrix, which can help with deepen
the understanding of coke properties and coke structure formation. Readers are encouraged to refer to Donskoi et al. [7, 17]
for full details.
60.0 95.0
57.5
94.5
DI 30/15
55.0
I40
52.5 94.0
50.0
93.5
47.5
93.0
45.0
20 25 30 35 40 45 50 20 30 40 50 60
Total_Inerts Total_Inerts
(a) (b)
Figure 14 Scatterplot of the dependence of a) I40, and b) DI 30/15 with the total abundance of inerts (both isotropic and
fused) determined texturally.
Such a comprehensive characterization of coke structure and texture also allow for the reliable modelling of coke properties
with high predictability. Figure 15 shows a scatterplot of I40 modelled vs I40 measured.
62.5
60.0
57.5
I40 Measured
55.0
52.5
50.0
47.5
45.0
45.0 47.5 50.0 52.5 55.0 57.5 60.0 62.5
I40 Modelled
Analysis of Variance
Model Summary
Coefficients
Figure 16. Iron ore mineral segmentation using Mineral4/Recognition4 (magenta – magnetite, blue – hematite, light green –
vitreous goethite, olive – ochreous goethite, dark green – kaolinite, cyan – quartz, yellow – porosity).
Identification of different morphologies of the same mineral is extremely important in sinter characterization [13, 14]. Some
of the major mineral characteristics of iron ore sinter are the abundances of primary (unreacted) and secondary (derived from
melt) hematite. Standard optical image analysis, which is based on RGB thresholding, is not able to distinguish these phases,
as they consist of the same mineral and so have the same reflectivity. However these two types of hematite in iron ore sinter
have different morphologies recognizable by trained mineralogists, and the Mineral4/Recognition4 software is also capable
of distinguishing them as shown in Figure 17. The situation is similar with segmentation of different types of SFCA (Silico-
Ferrite of Calcium and Aluminum), where the most important are SFCA-I (or fibrous SFCA), columnar SFCA which
displays the texture of relatively large elongated grains, and dense SFCA which has very low porosity and lacks any specific
texture. Segmentation of other sinter minerals such as magnetite, larnite, glass, unreacted fluxes and remnant aluminosilicates
can also be challenging, however Mineral4/Recognition4 successfully addresses all those issues.
(b)
Figure 17. a) A typical image of crushed sinter represented by a MosaiX of 3x3 elementary images; and b) the corresponding
mineral map obtained during automated image analysis (primary hematite – light blue, secondary hematite – dark blue,
magnetite – magenta, Fibrous SFCA-I – light green, columnar + dense SFCA – olive, glass – dark green, soluble
silicates/larnite - cyan, porosity and epoxy within particles – yellow).
In ironmaking very often there is a necessity to characterize relatively large objects with very high magnification, so their
mineralogy, porosity and fine texture are represented in adequate detail, and at the same time the objects can be analyzed as a
whole [20]. Pellets and large lumps of ore and sinter are the typical examples. For that purpose Mineral4/Recognition4
combines the elementary images obtained from a single microscope camera frame on two different levels. Initially, to
average out local irregularities, individual frames obtained at high resolution can be stitched during imaging into larger
MosaiX images. Subsequently, Mineral4/Recognition4 combines these MosaiX images into a single Panorama image of the
whole object. This approach allows for high resolution and comprehensive mineral identification at any point of the object,
(a) (b)
(c) (d)
Figure 20. a) MosaiX image (2x2) from the center of a high basicity pellet; b) the corresponding mineral map (magenta –
magnetite, blue – hematite, green – SFCA, dark brown – gangue, cyan – glass, yellow – porosity); c) a magnified part of
image (a) marked by the white square; d) mineral map corresponding to (c).
CONCLUSIONS
The novel hardware and software package Mineral4/Recognition4 developed by CSIRO allows for the comprehensive
characterization of coke structure and texture. Such characterization enables a more thorough integrated understanding of
dependences between parent coal blend composition, coke structure/texture and coke quality. Such an understanding may
REFERENCES
1. Kubota Y., Nomura S., Arima, T., Kato K., Quantitative Evaluation of Relationship between Coke Strength and Pore
Structure, ISIJ International, 2011, 51(11), pp. 1800–1808.
2. Kubota Y., Nomura S., Arima T., Kato K., Effects of coal inertinite size on coke strength, ISIJ International, 2008,
48(5), pp. 563–571.
3. Grant M. J. K., Chaklader A. C. D., Price J. T., Factors Affecting the Strength of Blast Furnace Coke, Fuel, 1991,
70(2), pp.181–188.
4. Sato H., Patrick J. W., Walker A., Effect of coal properties and porous structure on tensile strength of metallurgical
coke, Fuel, 1998, 77(11), pp. 1203–1208.
5. Donskoi E., Poliakov A., Lu L., Advances in Optical Image Analysis of Coke Structure, 10th Australian Coal Science
Conference, 2013, p. 17.
6. Donskoi, E., Poliakov, A., Mahoney, M. R., Scholes, O., Novel Optical Image Analysis Coke Characterisation and its
Application to Study of the Relationships between Coke Structure, Coke Strength and Parent Coal Composition, Fuel,
2017, 208, pp. 281–295.
7. Donskoi, E., Poliakov, A., Mahoney M., Optical image analysis of coke structure and the effects of structural
parameters on coke strength, 2015, ACARP project C23051, CSIRO report EP153398, pp. 1–118.
8. Eilertsen, J. L., Rrvik, S., Foosns, T., Oye, H. A., An automatic image analysis of coke texture, Carbon, 1996, 34(3),
pp. 375–385.
9. Crelling, J., Glasspool, I., Gibbins, J., Seitz, M., Bireflectance imaging of coal and carbon specimens, Coal Geology,
2005, 64, pp. 204–216.
10. Pearson, D., Pearson, R., Mahoney, M., Scholes, O., Application of automated petrography to characterize coke quality
distribution in slot-oven & non-recovery oven coke, 6th International Congress on the Science and Technology of
Ironmaking – ICSTI, 2012, pp. 1020–1030.
11. Donskoi E., Poliakov A., Manuel J. R., Raynlyn T. D., Advances in optical image analysis and textural classification of
iron ore fines, XXV International Mineral Processing Congress – IMPC, 2010, pp. 2823–2836.
12. Donskoi, E., Poliakov, A., Manuel, J., Peterson, M., Hapugoda, S., Novel developments in optical image analysis for
iron ore, sinter and coke characterisation, Applied Earth Science (Trans. Inst. Min. Metall. B), 2015, 124 (4), pp. 227–
244.
13. Donskoi, E., Poliakov, A., Manuel, J. R., Automated Optical Image Analysis of Natural and Sintered Iron Ore, in Iron
ore: mineralogy, processing and environmental sustainability, 2015, ed. – L.Lu, Elsevier, Cambridge, UK, pp. 101–
159.
14. Donskoi E., Hapugoda S., Lu L., Poliakov A., Peterson M., Haileslassie A., Advances in Optical Image Analysis of
Iron Ore Sinter, Iron Ore 2015, pp. 543–548.