Deciphering Coke Blends by Log-Probability (Rosin-Rammler) Analyses of Coke Microtextures and Modeling Anisotropy-Quotients (AQ's)

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

AISTech 2019 — Proceedings of the Iron & Steel Technology Conference

6–9 May 2019, Pittsburgh, Pa., USA


DOI 10.1000.377.023

Deciphering Coke Blends by Log-probability (Rosin-Rammler) Analyses of Coke Microtextures


and Modeling Anisotropy-Quotients (AQ’s)

David Pearson1, Richard Pearson1, HeeKyoung (Jackie) Park2, Yuekan Jiao1, and Drazen Gajic3
1
Pearson Coal Petrography Inc.,
1
#1 -740 Discovery Street, Victoria, British Columbia, V8T 1H2, Canada.
Email: dpearson@coalpetrography.com; rpearson@coalpetrography.com; yjiao@coalpetrography.com
2
16070, Vandustrial Lane, South Holland, Illinois, 60473, USA.
Email: jpark@coalpetrography.com
3
DMT GmbH Co. KG
3
Am Technologiepark 1, 45307 Essen, Germany.
Email: drazen.gajic@dmt-group.com

INTRODUCTION
Coke mosaics from single-source cokes, when analyzed with log-probability (Rosin-Rammler) plots, reveal same-slope linear
distributions, which means that types and proportions of Coke Mosaic Assemblages in these graphs can be predicted from the
romax reflectance of parent-coals. Coke blends, revealed by a characteristic step, are bimodal mixtures of two log-normal
populations; the dominant assemblage is Isotropic-Incipient-Circular, the supplementary is Lenticular-Ribbon-Encapsulite.
Proportions of both assemblages are graphically located by the inflection point in log-probability plots, and confirmed by the
second derivative.
AQ reflectance probability distributions, calculated from automated bireflectance mapping of whole cokes, define convex
traces in probability plots. Single-source cokes with dominant Isotropic-Incipient-Circular mosaics are positively skewed;
those enriched in fused Inertinite and supplementary Lenticular-Ribbon-Encapsulite mosaics, are negatively skewed. When
several AQ distributions are arranged side-by-side in order of rank, their traces resemble sections through onions – each skin
representing AQ distribution from a different rank single-source coke. Like onion skins, the AQ distribution lines in general
do not cross each other; by contrast AQ distributions of blended cokes always cross the onionskin fabric of the plots. Single-
source AQ distributions can be replicated and the models used in blend interpretation, but in such situations, components
must be confirmed microscopically otherwise inferred proportions may be ambiguous.
This paper describes petrographic deciphering of lump metallurgical coke by the following methods:
1. Dimensional measurement of microtextures and tallying by point-counting methods, as in, for example, ASTM
Standard D5061. Many have commented on how difficult it can be to accurately measure coke textures, and the
authors have relied specifically on published primary single-source microtextural coke measurements by Diessel
(2013), shown in Table 1. His data covers the rank-range of metallurgical cokes and are modeled here to initiate and
describe a different approach to the understanding of coke, by equating domain size, with reflectance.
2. Automated Bireflectance mapping, with measurement of Maximum and Minimum reflectance at every pixel
location; followed by calculation of Bireflectance (Maximum- Minimum reflectance); and calculation of Anisotropy
Quotient (Bireflectance/Maximum Reflectance * 10), or AQ (Pearson & Pearson 2012).
The objectives of this study were twofold: (1) to relate volume of single-source coke microtextures to the Romax rank of the
parent coals and to determine the contribution of each coal’s microtexture to a coke blend; and (2) to determine the
proportions and ranks of coal components in a coke. Analysis by Rosin-Rammler Log-probability plots aided in the first
objective, and probability plots were used in modeling AQ distributions in part two. Polar diagrams are used to relate results
from the two methodologies.

© 2019 by the Association for Iron & Steel Technology. 225


MICROTEXTURES

Rosin-Rammler distribution of coke microtextures.


Coke carbon forms, or microtextures, (Coin, 1987), are identified by size, and dimensions and arranged into anisotropic
domain size-categories. In conventional coke petrography the volume percent of these domains are tallied by point-counting
techniques, when as few as five (Coin et al), or as many as eleven, (Gray & DeVanney) coke domain sizes are identified and
classified, Gray & DeVanney, (1986), Coin (1987).
In earth science disciplines, log-probability plots are identified with Rosin & Rammler (1933), whose graphical analyses and
empirical formula for coal particle size distribution (PSD’s) was developed in Germany in the 1920’s and is still widely used
today in mineral processing.
Coke microtextures are analogous to sieve-sizes in PSD analyses and can also be interpreted by three-cycle log-
probability graphs, as demonstrated here. The Rosin-Rammler (log/double-log probability) graph is also known as
the Weibull distribution, and gives a good straight-line distribution for data that follows the function.

Distribution pattern of single-source coke mosaics.


Application of log-probability graphs in coke microtexture petrography is demonstrated here by normalized frequency
distributions of anisotropic domain-size data from five Australian single-source cokes reported by Diessel 2013, (Table 6.2),
and shown in Table 1.

Table 1. An inventory (in percent) of the anisotropy domain sizes of the vitrinite-derived fabric elements
of five cokes, and the Romax rank of their parent coals. Modified from Diessel 2013.
Fabric Element Anisotropy Domain Size Romax of parent coals
0.84 1.00 1.22* 1.40* 1.64
Isotropic sub-microscopic 75.4 16
Incipient Anisotropic <0.5 um 23.7 60.5 1.9
Circular Anisotropic Fine 0.5 - 1.0 um 0.9 20 7.2 2.8 1
Circular Anisotropic Medium 1.0 - 1.5 um 2 43.1 24.5 2
Circular Anisotropic Coarse 1.5 - 2.0 um 0.9 35.6 38.6 6.3
Lenticular Anisotropic Fine 2.0 - 5.0 um 0.6 9.5 30.9 45.8
Lenticular Anisotropic Medium 5.0 - 10.0 um 1.5 2 28.9
Lenticular Anisotropic Coarse 10 - 15.0 um 0.8 0.8 10.3
Ribbon Anisotropic Fine 15.0 - 20.0 um 0.4 0.4 3.7
Ribbon Anisotropic Medium 20.0 - 25.0 um 1.3
Ribbon Anisotropic Coarse > 25.0 um 0.7
All Fused Coke Matrix(%) 100 100 100 100 100

*The distribution pattern of these mosaics better fit a mixed-coal pattern.

Three of Diessel’s data sets from Table 1 follow log-normal distributions, with approximately-parallel slopes of y=2.48x to
y=1.93x, and because of the similar gradients the location of the slope relative to the vertical size classes can be used to
forecast mosaic-size sequences for any single-source coke produced from within this vitrinite rank-range (Figure 1).

226 © 2019 by the Association for Iron & Steel Technology.


Figure 1. Cumulative frequency distributions for three cokes plotted in Rosin-Rammler (log/double-log probability) plot.
Data is taken from Diessel 2013.
To appreciate the effect of rank intersections on the mosaic domain size proportions, calculations were made for every 0.05%
change in rank (half V-steps), from soft coking coal of 0.70% romax to 1.75% romax. The spatial distribution of the
anisotropic size domains produced in this rank space is shown in Figure 2, together with the five specific locations of
Diessel’s cokes. The diagram confirms and supersedes two earlier, similar graphs - Patrick (1973) Fig 10, and Loison (1989)
Fig 4.14.
This analysis show that mosaics for each single-source coke cover three domain sizes, forming part of a predictable
Assemblage of Mosaics, as shown by all vertical sections through Figure 2. The graph shows that with an increasing vitrinite
romax reflectance of the source coal, in the matching coke, there is a corresponding increase in the degree of anisotropy (the
coloured AQ lines, from green through red), and also the dimensions of the microtextural carbon domains (Patrick, Reynolds
and Shaw, 1973, Marsh and Cornford, 1976). This is demonstrated by all horizontal sections through Figure 2.
Figure 2 shows that the most extensive coke textural form is circular, derived from a range of coal rank precursors of 0.85%
to 1.70% romax, and which together with isotropic and incipient mosaics, constitutes 70% of all cokes.

© 2019 by the Association for Iron & Steel Technology. 227


Figure 2. Volume percent of anisotropy domain-size fabric-elements of single-source cokes, as a function of Romax rank of
the parent coal.
Among soft coking coals with vitrinite reflectance of <0.70%, the only coke mosaic microtexture is Isotropic, whereas in
coals of 0.80-0.90% (Figures 3, below), it is accompanied by increasing proportions of incipient, or circular, or a combination
of all three textures in the Isotropic-Incipient-Circular Assemblage, as predicted by the Rosin-Rammler graph.

Figure 3. A single image example of the Isotropic-Incipient-Circular Assemblage, from a single-source coke made from a
0.89% Romax coal. Left image. Plane-polarized-light, PPL, with superimposed structural data in red. The bright clear area (at
lower left) with elongated 60um pore is fused Isotropic vitrinite displaying no discernible structure; adjacent grey areas (at
top left) displaying greyscale structural variation with short discontinuous red lines are Incipient texture, the lines indicating
anisotropy development. Areas at right-center, with continuous red lines are fully anisotropic, either Circular, or Lenticular
mosaics. Right image. LUT-coloured AQ image, with grey/teal-green – fused Isotropic Vitrinite; mustard/purple/green -
Incipient; dark-blue/sky-blue - Circular; lime-green/yellow/orange - Lenticular.

228 © 2019 by the Association for Iron & Steel Technology.


Figure 4. Polar diagram correlating the Isotropic-Incipient-Circular Assemblage of the mosaics on the left of Figure 3, with
the AQ’s of pixels on the right in Figure 3.

Distribution patterns of blended coke mosaics.


Tabulations of textural components of coke blends are sensitive, rarely published information. However, three analyses taken
from the literature are listed in Table 2, Gray & DeVanney, (1986) (Tables 4A & 6A); ASTM Standard D5061 (Table 2); and
McPhee et al (2011) (Table 1). Also included are the two aberrant values from Diessel’s data described above.

Table 2. Mosaic compositions for three coke blends.


ASTM Table 2 (Opt A). Gray & DeVanney (Table 6) McPhee et al. (Table 1)
Isotropic 0.2% Isotropic 1.4% Isotropic 3.5%
Incipient 2.4% Incipient 3.8% Incipient 6.2%
Circular 74.8% Circular 66.2% Circular 39.8%
Lenticular 10.2% Lenticular 6.2% Lenticular 27.9%
Ribbon 12.4% Ribbon 22.4% Ribbon 22.6%

Cumulative probabilities of the five blends are plotted in a Rosin-Rammler (log/double-log probability) graph (Figure 5),
where their shapes depart drastically from the linear pattern of single-source cokes. All five are distinctly step-like, with
inflection points, or changes in direction of curvature, in the vicinity of 65 to 80 cumulative percentile, or, in the case of
Diessel’s cokes, 95 to 98 cumulative percentile, all of which are indicative of bimodality.
Interpretation procedure for bimodal distributions requires the location of the infection point, which can be precisely located
within the gently sloping central segment. The stepped configuration is characteristic for coke blends and can be interpreted
using the bimodal distribution model of Sinclair (1981, p 37), by partitioning the log-normal microtextural distribution into
two populations; a dominant assemblage Isotropic-Incipient-Circular, with a Romax of 0.98% and a supplementary
assemblage, Lenticular-Ribbon-Encapsulite, with a Romax of 1.60%.
The departure from linearity, the characteristic step with an inflection point in log-probability plots of coke mosaics indicate
bimodal mixtures of different rank coal blends.

© 2019 by the Association for Iron & Steel Technology. 229


Figure 5. Cumulative frequency distributions for five cokes plotted in Rosin-Rammler log-probability plot. Legend indicates
data source.

ANISOTROPY QUOTIENT - AQ
The Anisotropy Quotient or AQ, is the ratio of bireflectance to maximum reflectance, and when measured on metallurgical
coke, its magnitude depends specifically on (1) initial rank of the fused vitrinite; (2) heating rate; and (3) the final oven
temperature (Coin & Brown, 1986). The median value of the AQ reflectance distribution, or MAQ, is a highly accurate
descriptor of any coke because it exactly separates an upper half from a lower half of the usually skewed data, and when
plotted in a probability plot, with a vertical log-log probability scale, each coke has a unique location along the 50 percentile
line, for comparison with all other cokes. This diagram is the MAQ Probability plot.
AQ is a useful parameter in resolving the thermal history of a metallurgical coke; under-heated (green coke) from aged
industrial ovens, and thermally distinctive (over-heated) test-retort cokes can be demonstrated by deviation from expected
trends (Fig 6).

230 © 2019 by the Association for Iron & Steel Technology.


Figure 6. The relationship between Maximum Reflectance, Bireflectance (Maximum – Minimum Reflectance), and
Anisotropy Quotient, AQ, (Bireflectance/MaximumReflectance * 10), for normal global cokes; for under-heated blends from
aged ovens; and for test samples from over-heated retorts. Each shape represents a median AQ value of over 2 billion
individual readings. Coloured lines are AQ’s.
The median AQ range of cokes shown in Figure 6 is from 1.3 to 6, but in practice, coke is manufactured to meet quality
requirements of a specific size of blast furnace, with larger blast furnaces requiring the greater strength of higher AQ cokes
(Figure 7). Of the twenty-one industrial cokes examined by the authors to date, this range is 2.8 to 5.2.

6000
Blast furnace Internal

5000 y = 1263.3x - 1117.9


R² = 0.98
4000
Volume m3

3000

2000
1000
0
2 3 4 5 6
Median Anisotropy Quotient (Biref/Romax *10)

Figure 7. Plot of Blast furnace internal volume (m3), versus blend MAQ’s. Higher quality coke, identified by elevated
MAQ’s, is required by larger furnace.
As an index, MAQ is important because it identifies a coke quality target for a specific size blast furnace, and also provides
the yardstick to confirm whether target is achieved (Geerdes, M. Toxopeus, H, & van der Vliet, C.). MAQ therefore provides
post-carbonization microscopic confirmation that the coke oven temperature has been met, and verifies that, ranks of
components, textural composition, and cold strength expectations, (M40 & ASTM Stability), are acceptable.

© 2019 by the Association for Iron & Steel Technology. 231


Single-source coke MAQ patterns and modelled replicates.
Bireflectance mapping of coke produces a reflectance distribution of the whole-coke, not just the anisotropic components
(fused vitrinite), but also all infusible- and fusible-Inertinites. And, because the anisotropy quotient reflectance data cover a
small range of values, from 0 to 9, the selected graphical scale is linear, not logarithmic.
AQ reflectance data for eight single-source cokes, ranging in parent-coal Romax-reflectance from 0.70% to 1.65%, are
plotted in grey in a MAQ log-probability graph in Figure 8, where most of the AQ distributions are seen to be either
positively- or negatively-skewed. Nevertheless, the range of median AQ’s of these cokes is about 1 to about 6, viewed at the
50 cumulative percentile. In appearance these graphs resemble up-side-down shallot onions – from a constricted base,
bulging at the center to broad at the top, and in this analogy, adjacent lines that parallel each other, represent AQ distributions
of cokes from lower-or higher-rank parent coals. These distributions are a function of the mosaic makeup of cokes. All cokes
from high volatile rank parentage, up to a threshold of about 1.11% romax, are left-skewed, and comprise the primary I-I-C
mosaic assemblage. Single-source coke from coal rank parentage higher than this threshold are of the supplementary L-R-E
mosaic assemblage.
Figure 8 also includes five modeled virtual coke replicates, and although these share the same Romax parentage, they are not
the actual cokes analyzed by Diessel, (which were not available for analysis). Virtual coke replicates are whole coke AQ
distributions that are interpolated between real known coke curves. Two virtual cokes overlie the AQ distributions of coals #7
and #8, with parent coal romax of 1.40% and 1.65%, and this confirms that real coke AQ distributions can be replicated by
this technique. Thus, the MAQ Probability plot of Figure 8 can be used to identify the parent coal rank of any single-source
coke.

Figure 8. MAQ probability plot, (Onion plot), of eight single-source cokes and five modeled virtual-coke replicates.

AQ patterns of multiple component cokes and modelled replicates.


The realization that virtual cokes, made from modeled cumulative probability distributions can accurately replicate real
single-source cokes, facilitates modeling of blended cokes in terms of ranks of their component coals, and component
proportions, where components are single-source cokes, if available, or virtual cokes, if not. The rationale for using AQ
modelling would be, for example, to confirm a planned coke recipe (comparison of modeled and sample MAQ’s), or to
ascertain the makeup of an anonymous coke sample (make model MAQ the same as sample MAQ), or to determine what
went wrong!
Figure 9 is an MAQ probability plot of a coke made from three coals. Two of the component AQ distributions, 0.70% and
1.40% are real cokes. The AQ distribution for 1.05% is a virtual coke. A measure for goodness-of-fit in this technique has not

232 © 2019 by the Association for Iron & Steel Technology.


yet been developed, but the coincidence of sample MAQ with modelled MAQ data and essentially overlapped to about 95
cumulative percentile is encouraging.

Figure 9. An example of an MAQ probability plot of coke made from three coals.
Figure 10 is an MAQ probability plot of a coke made from four coals; two high volatile coals and two low-volatile coals.
There is coincidence of sample MAQ with modelled MAQ data and the high volatile component is essentially overlapped to
the inflection point at about 75 cumulative percentile, suggesting that the modeling of the low volatile population could be
improved.

© 2019 by the Association for Iron & Steel Technology. 233


Figure 10. This is a MAQ probability plot of an industrial coke manufactured from four known coals.

CONCLUSIONS
1. Whether or not a coke is comprised of one or more coal populations is readily determined by the pattern of domain
size distributions in log-probability plots.
2. The proportions and types of microtextures in single-source metallurgical cokes can be predicted from romax
reflectance of the parent coal using the Rosin–Rammler probability distribution.
3. Proportions of microtextural fabric elements in bimodal two-population coke blends can be determined from the
Rosin–Rammler probability distribution.
4. Parent coal rank of any single source coke can be determined from the MAQ probability plot.

REFERENCES
1. ASTM, Standard Test Method for Microscopical Determination of the Textural Components of Metallurgical Coke,
ASTM Standards D5061-07, 2007
2. Brown, N.A. & Coin C.D.A. 1986. Improved Evaluation of Australian Coking Coals and Blast Furnace Coke.
NERDDC Project 0545. ACARP Report No. C0537
3. Coin CDA, 1987. Coke microtextural description: comparison of nomenclature, classification and methods. FUEL. Vol
66. 702-705.
4. Coin CDA & Gill WW 1983. Microtextural analysis in metallurgical coke studies. Joint Symp. Iron and Steel Inst.
Japan and Australasia Inst Min. Metall., Tokyo, Iron Steel Inst Japan, 41-52.
5. Crelling, JC, Glasspool, IJ, Gibbons JR, and Seitz, M. 2005. “Bireflectance imaging of coal and carbon specimens,”
International Journal of Coal Geology, Vol. 64, 2005, pp. 204-216
6. Diessel, CFK. 2013 Notes from 6th ICCP Organic Petrology Course. Potsdam June 10 – 14, 2014.

234 © 2019 by the Association for Iron & Steel Technology.


7. Diessel, CFK & Diessel LJ 2012. The application of optical microscopy to investigations related to the coal mining
industry, including coke making. ACARP Project C12060
8. Geerdes, M. Toxopeus, H, & van der Vliet, C. 2009. Modern Blast Furnace Ironmaking. IOS Press BV. 161p.
9. Gray, R. J., & DeVanney, K.F. 1986. “Coke Carbon Forms: Microscopic Classification and Industrial Application” Int.
Jour. Coal Geology. Vol 6. pp 277-297
10. Loison, R., Foch, P., and Boyer, A. 1989. Coke: quality and production. Butterworths Press. 553p.
11. Marsh H and Cornford C 1976 Mesophase: the precursor to graphitisable carbon. Deviney ML and O’Grady RM (ed)
Petroleum derived carbons ACS Symp Series 21. Washington 266-281.
12. McPhee, T. Giroux, L, Ng, KW, Todoschuk, T. Conejeros M and Kolijn, C. 2011. A facile approach to the CSR
determination of metallurgical coke. METEC. 2011.
13. Patrick JW, Reynolds MJ and Shaw FH. 1973. Development of optical anisotropy in vitrains during carbonization.
FUEL 52 198-204
14. Patrick JW, Reynolds MJ and Shaw FH. 1979. Optical anisotropy of carbonized coking-coal vitrains. FUEL 58 501-509
15. Pearson DE (1995), Probability analysis of Blended Coking Coals. 1991. International Journal of Coal Geology. Vol
19. 109-119
16. Pearson, D.E., & Pearson, R.A., 2012, “Bireflectance Mapping and Automated Petrography of Metallurgical Cokes”
AIST. Atlanta, GA, 2012, pp 247
17. Rosin, P., & Rammler, E. (1933), The Laws Governing the Fineness of Powdered Coal. J. Inst. of Fuel, 7: 29–36.
18. Sinclair, A. J. Selection of threshold values in geochemical data using probability graphs. J. Geochem. Explor. 3, 129–
149 (1974)
19. Sinclair, A.J., 1981. Applications of Probability Graphs in Mineral Exploration. Assoc. Explor. Geochem. Spec. Vol. 4,
95 pp.

© 2019 by the Association for Iron & Steel Technology. 235

You might also like