You are on page 1of 10

Fuel Processing Technology 121 (2014) 104–113

Contents lists available at ScienceDirect

Fuel Processing Technology

journal homepage:

A review of the correlations of coal properties with

elemental composition
Jonathan P. Mathews a,b,1,⁎, Vijayaragavan Krishnamoorthy a,b,1, Enette Louw a,b,1, Aime H.N. Tchapda a,b,1,
Fidel Castro-Marcano a,b,1, Vamsi Karri a,b,1, Dennis A. Alexis a,b,1, Gareth D. Mitchell b,1
John and Willie Leone Family Department of Energy and Mineral Engineering, The Pennsylvania State University, 126 Hosler Building, University Park, PA 16802, USA
The EMS Energy Institute, The Pennsylvania State University, 126 Hosler Building, University Park, PA 16802, USA

a r t i c l e i n f o a b s t r a c t

Article history: The spatial arrangement and abundance of the elements: C, H, N, O, S often correlate or directly influence a
Received 11 December 2012 plethora of coal properties. For N 90 years, attempts have utilized the ultimate (elemental) analysis of coal to
Received in revised form 15 January 2014 predict a wide variety of properties such as: calorific value (higher heating value), volatile matter, vitrinite reflec-
Accepted 16 January 2014
tance (mean maximum), Hardgrove grindability index, helium density, aromaticity, etc. While many relation-
Available online 3 February 2014
ships resulted in graphical plots that have utility even today, numerical values can also be directly calculated
utilizing the correlations. These have the potential to allow rapid predictions and low-cost approaches to coal
Ultimate analysis property determination. Here the many correlations addressing multiple coal properties were reviewed and
Correlation where possible evaluated against a sampling of the Pennsylvania State University Coal Sample Bank and Database
Statistical analyses of coal for vitrinite-rich (N80% by point counting) United States coals. Over 42 correlations were found in the literature.
Coal properties While some correlations, such as calorific value predictions are accurate over a wide range of compositions,
others are restricted in applicability to a select rank range. For many correlations, there are challenges to predict
the property accurately, over a wide range, but may capture the trends.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction The Penn State Coal Sample Bank and Database currently houses
1058 coals of the 1579 samples collected over the past 45 years. While
The structure and resultant property/behavior of coals are often most of these samples are historical in nature, 38 are DECS-series sam-
dependent on the spatial arrangement and abundance of the elements: ples protected from deterioration using foil/polyethylene laminate bags
C, H, N, O, S, also water and cations in low-rank coals, mineral matter under argon gas and another 22 are premium samples contained in
and process conditions: particle size, mass, heating rate, etc. If we ignore glass vials under nitrogen. Significant data have been collected on
the latter components and focus on the elemental composition, it is about 1270 of these coals including chemical, petrographic, mechanical
possible to predict a large number of coal parameters. Even if there is and thermoplastic properties in addition to geologic and geographic in-
no direct cause–effect link between composition and properties, the formation [6,7]. Unless otherwise noted, the predictions are applicable
systematic transformations accompanying coalification often permit only for vitrinite-rich coals. Other maceral-rich coals differ significantly
correlations. Thus, while causation and correlation need to be separated, in composition, structure and behavior [3,8–11]; thus, many of the
there is an opportunity to build on the remarkable extended studies of correlations here may have limited applicability to non-U.S. coals.
coal that have been 70 plus years in preparation. These studies often Predictions using the Penn State Coal Sample Bank and Database were
generated graphical plots which still serve the needs of the coal science restricted to coals of vitrinite composition N80% determined by point
community, for example Seyler's plot [1], van Krevelen diagrams [2], counting approach.
etc. and the many collections of graphical representations [3–5]. Recent-
ly, the trends are often accompanied by predictive equations. Gathering 2. Predicting coal parameters
these relationships will improve their utility and may offer considerable
synergy. This paper attempts to determine if indeed “coal is elementary” 2.1. Calorific value
and examines the predictive capability from a literature review and
where possible tests predictions against the Penn State Coal Sample Calorific value of coal is one of the most important parameters for
Bank Database for a subsample (N80%) of vitrinite-rich coals from the predicting coal behavior, defining rank (for much of the rank range),
United States [6]. for evaluating its potential use in combustion systems [12] and for
evaluating pollution compliance. Calorific value can be determined ex-
⁎ Corresponding author.
perimentally by measuring the enthalpy change of a sample in a bomb
E-mail address: (J.P. Mathews). calorimeter by either isothermal or adiabatic methods [13]. However,
Tel.: +1 814 863 6213. there are numerous correlations for predicting higher heating values

0378-3820/$ – see front matter © 2014 Elsevier B.V. All rights reserved.
J.P. Mathews et al. / Fuel Processing Technology 121 (2014) 104–113 105

(HHV) for various fuels and wastes [12,14–26]. The earlier relationships The correlation was obtained from 450 proximate analyses covering
are well reviewed by Selvig and Gibson [27]. For the well known, and a wide range of fuels: industrial wastes, coals, cokes, municipal
still used Dulong formula, from 1820, (Eq. (1)) the errors are expected wastes, chars, and manufactured fuels. The parameters FC , VM and
to be small for the bituminous and anthracite (b1%) with larger nega- ASH are fixed carbon, volatile matter and ash yield all on a dry
tive values for coals of subbituminous (− 1.4%) and lignite rank basis. When the correlation was tested with 100 data points of fuel
(− 3.1%) [27]. With the Penn State Coal Sample Bank and Database ranging from coal, coke, chars and biomass, along with 450 data points
coals (circa 1980) supplemented by the Bureau of Mines data, Mason used in derivation of the correlation, the average absolute error was
and Ghandi [28] also reported a higher variance for lignite. There, found to be 3.74%.
however, the mean differences were negative for both anthracite and
bituminous coal. HHV ðMJ=kgÞ ¼ 0:3536½ F C  þ 0:1559½V M −0:0078½ASH  ð6Þ

HHV ðBTU=lbÞ ¼ 145:44½C  þ 620:28½H þ 40:5½S–77:54½O ð1Þ Sheng and Azevedo reviewed correlations to predict HHV of biomass
based on its ultimate, proximate analysis, and biochemical composition
The elemental content is in weight percent on a dry basis. (such as cellulose component) [21]. The models based on ultimate
Most of the coal-specific relationships were developed for regional analysis were more accurate than proximate analysis or biochemical
coals. They often follow the approach of Dulong, and later Dryden composition. Their correlation (Eq. (7)) calculates the higher heating
[14], utilizing heat of combustion summation approaches with varia- value in dry basis. C and H are carbon and hydrogen, in dry wt.%. An
tions [29] mostly based on multivariate regression analyses. Neavel additional correlation (not shown) icorrected for ash yield.
et al. developed an empirical correlation based on 66 low mineral mat-
ter vitrinite-rich North American coal samples ranging in rank from HHV ðMJ=kgÞ ¼ −1:3675 þ 0:3137 ½C  þ 0:7009 ½H  þ 0:0318 ½O ð7Þ
lignite to low-volatile bituminous [16]. According to this correlation
(Eq. (2)), HHV (dry basis) is a function of carbon, hydrogen, oxygen, Majumdar et al. [26] developed a correlation based on proximate
total sulfur, and ash yield. In their analysis with a set of 120 coal analysis (A, M, VM , and F C ) of 164 Indian coal samples to predict
samples, ranging from peat to anthracite yielded a mean error of HHV (as received basis) (Eq. (7)). Where A, M, VM, and FC are ash,
86 Btu lb−1. moisture, volatile matter, and fixed carbon in wt.%, respectively.
The model was shown to predict HHV of 79 Indian coal samples
HHV ðBTU=lbÞ ¼ 145:9½C  þ 569:6½H –53:89½O þ 43:08½S–6:3½ASH ð2Þ
with an average error of 13.1%. However, with a different set of 86
Indian coal samples, the average absolute error was significantly re-
Values in square brackets indicate mass basis values.
duced to 1.46%. The authors attributed the poor prediction of 79
Singh and Kakati developed empirical models to predict HHV for
Indian coal samples to experimental errors involved in measuring
Indian coal samples [12]. They obtained four different models by corre-
HHV of too few coal samples.
lating O, O/C, O/H, and O/C and O/H atomic ratios to HHV for 160 Indian
coal samples. Of all the developed correlations, the authors found Eq. (3) HHV ðMJ=kg Þ ¼ −0:03 ½ASH −0:11½M  þ 0:33 ½V M  þ 0:35 ½ F C  ð8Þ
to be most applicable. The HHV is calculated on a dmmf basis. Their
correlation predicted 160 Indian coal samples with an average error of
Selected models, based on elemental composition (and ash yield),
were tested and compared for their predictability on 971 coal samples,
HHV ðMJ=kgÞ ¼ 37:4541−14:2040ðO=C Þ−21:2929ðO=HÞ ð3Þ ranging from lignite to anthracite, obtained from the Penn State Coal
Sample Database. However, it should be noted that the validity of
Mazumdar developed a correlation that relates theoretical oxygen the published equations holds only for the rank-range for which they
requirement for complete combustion of coal (Ord), ash yield, and H/C were derived. With exception of the Sheng and Azevedo model [21],
ratio of coal to HHV (dry basis) (Eq. (3)) [17]. Where HHV is calorific all were coal-derived expressions. The Sheng and Azevedo model was
value in dry basis, H/C (wt.%/wt.%) is the hydrogen to carbon ratio. compared to test the validity of the biomass-based model for coal.
The correlation was found to predict HHV of 31 vitrinite-rich North Hence, the comparison was academic. All the models were corrected
American coal samples from lignite to low-volatile bituminous coals to dry-ash-free basis. Comparison of the models, for various coal
with an average absolute deviation of 0.45% (Eq. (4)). ranks, is shown in Table 1. As expected, the predictability of models
specifically developed for coal performed better. Surprisingly, the
HHV ðMJ=kgÞ ¼ 13:03½Ord  þ ð100−MMÞ Sheng and Azevedo model [21] predicted anthracite with better accura-
 ð0:238  ðH=C Þ−0:0065Þ−0:007 ½ASH  ð4Þ cy than other coal models (Table 1). The Sheng and Azevedo model
however had considerable scatter. The general under prediction with
Channiwala and Parikh went a step further and developed a unified higher carbon content happens to match the slight drop in the HHV of
correlation for estimating HHV of solid, liquid and gaseous fuels from anthracite due to the reduced hydrogen content. The Singh and Kakati
elemental analysis (Eq. (5)) [18]. The correlation was derived using model was derived for Indian coal samples and tends to slightly over
225 elemental compositions and ash yields, and was validated with 50 estimate HHV. This may be attributed to differences in petrography
fuels: coals, cokes, municipal wastes, chars, and various gaseous fuels. (Gondwana inertinite-rich coals vs. U.S. vitrinite-rich coals). The models
The HHV in the correlation is the higher heating value on a dry basis. of: Neavel et al., Channiwala and Parikh, and Mazumdar predict equally
The elements C, H, O, N, S and ash are in dry wt.%. The authors claim well as shown (Fig. 1 and Table 1). The Dulong formula was poor as
that the model was reasonable in estimating HHV of fuel having ele- indicated by the deviation from 45° for the regression line (Fig. 1) and
mental composition (wt.%): C from 0 to 92.25%, H from 0.43 to 25.15%, the deviation (Table 1). However, the wide applicability makes the uni-
O from 0 to 50%, N from 0 to 5.6%, S 0 to 94.8%, and ash ≤71.4%. However, versal correlation of Channiwala and Parikh [18] attractive. Given et al.
the range is wide enough to cover most of the fuels used in industry. also considered calorific value with the Penn State Coal Bank and Data-
base (circa 1986) and also proposed relations with the improvement of
HHV ðMJ=kg Þ ¼ 0:3491½C  þ 1:1783½H including the term related to the enthalpy of decomposition [15]. His
þ 0:1005½S−0:1034½O−0:0151½N−0:0211½ASH  ð5Þ breakdown included a relationship clustered into six subsets to capture
coal rank transitions and regional (Eastern/Interior/Rocky Mountain)
Parikh et al. developed a unified correlation for estimating HHV of influences. Van Krevelen considers the Neavel et al. and the Given
solid, liquid and gaseous fuels from proximate analysis (Eq. (6)) [20]. et al. relationships to be the better ones [4]. Although not covered
106 J.P. Mathews et al. / Fuel Processing Technology 121 (2014) 104–113

Table 1
Comparison of errors for different HHV models for PSU Sample Bank coals.

Coal rank N Average absolute error (AAE) %

Neavel Singh and Kakati Channiwala Sheng and Mazumdar Dulong

et al. correlation and Parikh Azevedo correlation correlation
correlation correlation correlation

Lignite 51 1.63 4.08 1.92 10.71 1.66 15.45

Subbituminous 114 1.54 3.59 1.71 12.43 1.72 12.22
High-vol. bituminous 622 1.03 2.18 0.98 16.80 1.47 14.69
Medium-vol. bituminous 74 0.98 2.68 0.92 10.04 1.65 14.62
Low-vol. bituminous 83 0.95 2.78 0.82 17.50 1.59 16.66
Anthracite 27 1.31 5.23 1.42 8.18 1.30 17.89
Total 971 1.12 2.62 1.11 15.27 1.53 14.70
ðExperimental HHV−Predicted HHV Þ 
Σ x100
Experimental HHV
AAE ð% Þ ¼

here, recent work has utilized a more sophisticated neural network are percentages of vitrinite, “exinite” (liptinite) and inertinite. The sym-
approach [24,25,30]. bols x, y, and z are their volatile matter values for “pure” macerals. Com-
parison of their calculated and experimental data was considered
“statistically insignificant” by those authors.
2.2. Volatile matter

VM ¼ ða  xÞ=100 þ ðb  yÞ=100 þ ðc  zÞ=100 ð9Þ

Volatile matter is an integral part of the ASTM classification system
[31] and is indicative (along with other parameters) of coals' suitability
for combustion and carbonization. The standard method for determin- The Borrego et al. [36] formula (Eqs. 10–13) used a combination
ing volatile matter is the ASTM D3175 [32]; however, the instrumental of maceral composition and vitrinite reflectance. A summary of
proximate approach is now more common (recently (2010) replaced their formula is shown in Table 2. The fractional volatile matter con-
by a macro thermal analysis approach [33]), but there are differences in tent of the residual maceral in coal (φi) was determined according to
yields between the two classic and modern approaches for some ranks the formula. Then, the volatile matter contribution of each maceral is
[34]. Various attempts have been made to estimate its value from other determined by multiplying the fractional content of the residual
coal parameters such as elemental composition [16], maceral composition maceral φi , by the percent content of the residual maceral (X i ) in
[35], and vitrinite reflectance [36]. Chatterjee et al. [35] used maceral coal. The volatile matter yield is the sum of the individual maceral
composition to estimate the volatile matter (Eq. (9)) where a, b, and c volatile matter contributions.

Fig. 1. Models of Neavel et al., Singh and Kakati, Channiwala and Parikh, Sheng and Azevedo, Mazumdar, and Dulong formula.
J.P. Mathews et al. / Fuel Processing Technology 121 (2014) 104–113 107

Table 2 Fig. 2 shows the calculated versus measured volatile matter. The
Equations of Borrego et al. for volatile matter prediction from maceral composition. top left graph shows how Neavel et al. equation performs for all
Maceral Range (Rr) φi VMi ranks. As expected, Neavel's equations fail to predict the volatile
v1 matter content of high-rank coals; their equations were derived for
Vitrinite 0.42–6.08% φv ¼ v v (10) VMv = φv*Xv
Rr 2 −Rr 3 −v4 þv5
lignite to subbituminous rank. Their equation gave negative values
Rr 3
Inertinite 0.42–6.08% φi ¼ i1 exp − i2 (11) VMi = φi*Xi when the volatile matter yield of the coal was low (high-rank
I 1 þRr2
I coals). When applied to lignite and subbituminous coals, their re-
Liptinite 0.42–1.40% φl ¼ I I
(12) VMl = φl*Xl
Rr3 −Rr4 −I 5 þI 6
sults improved. The bottom graphs of Fig. 2 give the results of
1.40–6.08% φv ¼ (13) VMv = φl*Xl
v v
Rr 2 −Rr 3 −v4 þv5 Borrego et al. equations. Borrego's equations that are restricted to
v1, v2, v3, v4, v5, i1, i2, i3, I1, I2, I3, I4, I5, and I6 are constants and Rr is the random vitrinite
the random vitrinite reflectance of 0.46–6.08 give a fairly good pre-
reflectance, predicted VM value is the sum of the maceral contributions. diction for high-rank coals. It is noteworthy that in 1916, there was
work on achieving the reverse, the prediction of elemental data from
proximate analysis [37]. For completeness, Koba [38] also provided
VM predictions for coking coals mostly using variations of the H/C
Neavel et al. [16] produced the simplest equation to estimate the ratio or hydrogen content.
volatile matter, using ultimate analysis data for low-rank coals. Their
equations were segregated into a sulfate-free ash group and an uncor-
rected ash group. 2.3. Vitrinite reflectance

Whole dataset and applying an intercept to the analysis Vitrinite reflectance (mean maximum) (R0) is often used as a coal
rank or basin thermal maturation indicator. The R0 value is determined
VM ¼ 94:160−1:350½C  þ 10:630½H−1:780½N−0:845½Ash ð14Þ
by measuring the amount of incident light reflected from the polished
surface of a statistically significant number of vitrinite particles and
VM ¼ 92:380−1:34½C  þ 10:440½H  þ 0:366½S−0:963½Ash ð15Þ
reporting the mean value. Reflectance values are collected from a micro-
scope system capable of plane polarized, white light illumination
Selected samples of the dataset and applying an intercept to the between ×400 and ×750 magnification, 360° stage rotation and having
analysis a photoelectric system standardized to a known reflectance range that
can record the maximum amount of incident light through a 360°
VM ¼ 94:670−1:345½C  þ 10:590½H−2:290½N−0:825½Ash ð16Þ
rotation. Flueckinger et al. determined correlation tables for mean max-
imum vitrinite reflectance and found a strongly negative correlation
VM ¼ 93:670−1:363½C  þ 10:430½H þ 0:396½S−0:964½Ash ð17Þ
with hydrogen and a strongly positive correlation with carbon as ex-
pected [39]. Koba also developed various R0 correlation parameters for
Selected samples without intercept coking coals [38]. McCartney and Teichmüller [40] conducted a compre-
hensive study of European and U.S. coals with N75% vitrinite content
VM ¼ −0:408½C  þ 11:250½H  þ ½O þ 1:3½S ð18Þ
and H/C atomic ratios ranging from 0.6 to 1.2. The R0 values correlate
well with other commonly used rank parameters such as volatile
VM ¼ −0:417½C  þ 11:520½H  þ 0:912½O þ 1:260½Ash ð19Þ
matter, carbon content, and H/C ratios. They indicated that generally
vitrinite aromaticity increases with increasing rank, raising the refrac-
VM*: Ash uncorrected, otherwise sulfate-free ash tive index and consequently the reflectance. Several correlations have

Fig. 2. Calculated versus measured volatile matter.

108 J.P. Mathews et al. / Fuel Processing Technology 121 (2014) 104–113

been derived for estimation of R0 from elemental composition data vitrinite coals and experimental values for comparison. As seen, Neavel
[40–44]. Sweeny and Burnham developed a simplified Arrhenius reac- [41] and Burnham and Sweeney [42] relationships are able to predict
tion model (EASY %R0) that incorporates a distribution of activation the increase in R0 for %C N 91%. George et al. [43], and Smith and Smith
energies to calculate vitrinite maturation with changing time and tem- [44] correlations showed consistency with experimental trends for
perature [45]. Neavel [41] proposed an exponential fit to the data of carbon content of 75–85%. The fit was poor for Chelgani et al. outside
McCartney and Teichmüller [40] (Eq. (20)). of the bituminous range as expected for a bituminous correlation.
Fig. 4 shows a comparison between calculated and measured values
R0 ¼ 15:64  exp½−3:6  ðH=C Þ ð20Þ for these equations. Neavel and also Burnham and Sweeney exhibit a
reasonable agreement with experimentally determined R0 values,
An improved correlation based on results given by McCartney and whereas George et al. and the Smith and Smith approach had higher de-
Teichmüller [40] was obtained by Burnham and Sweeny [42] recognizing viations for these coals. Recent work used the combination of proximate
that oxygen content also tends to reduce the refractive index (Eq. (21)). and ultimate analyses with a neural network approach to predict
maceral composition for Indian coals with significantly improved
R0 ¼ 12  exp½−3:3  ðH=C Þ−ðO=C Þ ð21Þ results over multivariate regression analyses [46]. Their prediction
(Eq. (25)) for vitrinite had a coefficient of determination of 0.6495 vs.
George et al. [43] used samples from the Permian Greta seam in the 0.9684 for the equivalent neural work determination.
Sydney Basin (Australia) with H/C atomic ratios N0.85 (Eq. (22)).
Vitriniteð% Þ ¼ 65:1470 þ ð0 : 131 ½MÞ þ ð0 : 4832 ½VMÞ–ð0 : 8263½AshÞ
R0 ¼ 2:12–1:71  ðH=C Þ ð22Þ

Smith and Smith [44] developed correlations (Eq. (23)) for Australian ð25Þ
coals with 80–95 %C (dmmf).
Average absolute deviation values obtained from correlations ap-
R0 ¼ −5:035 þ 0:0804½C –0:184½H  ð23Þ
plied to PSU Coal Sample Bank and Database are compared in Table 3.
As expected, these correlations exhibit lower deviations for high- and
Recent regression analysis approach generated Eq. (24) with an R2
mid-rank coals since they were developed using vitrinite reflectance
value of 0.77 for 1012 Kentucky coals by Chelgani et al. [30]. An im-
data from these ranks. Specifically, for anthracite and high-volatile
proved R2 value of 0.84 was obtained with the feed-forward artificial
bituminous rank, Burnham and Sweeney [42] correlation showed the
neural network approach.
lowest deviation. For medium- and low-volatile bituminous rank, the
2 equation proposed by George et al. [43] exhibited the best agreement
R0 ¼ 9:096−0:129½C  þ 0:001½C  −0 : 03½O−0:021½VM−0 : 081Ln½S
with the experimental values, whereas for lignite and subbituminous
−0 : 738½H  þ 0:077½H  ð24Þ rank, the experimental data were best correlated by the Burnham and
Sweeney relationship. The results obtained suggest that the equation
Fig. 3 shows R0 values computed from Eqs. (20 to 24), with respect to developed by Burnham and Sweeney better correlates overall with the
carbon content, utilizing PSU Coal Sample Bank and Database for N80% PSU Coal Sample Bank and Database.

Fig. 3. R0 values computed from Eqs. (20 to 24), with respect to carbon content.
J.P. Mathews et al. / Fuel Processing Technology 121 (2014) 104–113 109

Fig. 4. Comparison between calculated and measured values.

2.4. Aromaticity tars, and gases.

Coal properties can often be predicted by the degree of aromaticity. f a ¼ ð0:0159½C Þ–0:564 ð27Þ
van Krevelen and colleagues determined molar volumes and aromaticity
via a graphical densimetric method via a group contribution approach Aromaticity, determined through single pulse excitation 13C NMR,
[47]. Mazumdar used corrected values (Eq. (26)) with X = 0.115 when correlated the aromaticity to the atomic H/C ratio for coals in the
H/C ≥ 0.7 and 0.125 when H/C is less [48]. The Mc/d value is a molar bituminous rank range Maroto-Valer et al. [52,53]. The correlation
volume per carbon atom (via group addition approach with assignment gave reasonably accurate results for 39 different coals. Mazumdar in a
assumptions) with an estimated helium density prediction to obtain follow-up letter further discussed applicable ranges and locations of
mean values around 11.33. the transformation between linear ranges [54].

f a ¼ 1:22−0:58  ðH=C Þ ð28Þ

f a ¼ ½1–ðH=C Þ þ X ½ðMc =dÞ−5:34 ð26Þ

2.5. Physical parameters

A linear correlation was also generated by Niksa [49] based on
Gerstein et al. [50] data (Eq. (27)). The goal in that work (Flashchain) There is the expectation that “the macroscopic physical properties of
and in the Chemical Percolation Devolatilization model [51] was to a substance depend to some extent on its density because the interac-
utilize various NMR parameters in describing the coal structure and tion between the molecules is a function of the intra-molecular
subsequent breakup upon devolatilization to determine yields of char, distance” [47]. Coal density (daf/dmmf) is an important property

Table 3
Comparison of average errors for different vitrinite reflectance models PSU Coal Sample Bank and Database.

Rank Average error (%)

Neavel (1981) Burnham and Sweeney (1989) George et al. (1994) Smith and Smith (2007)

Lignite −132 36 94 67
Subbituminous −91 39 86 54
High-volatile bituminous −23 9 29 14
Medium-volatile bituminous −20 −9 2 −22
Low-volatile bituminous −21 −12 −3 −37
Anthracite −54 −8 10 −66
110 J.P. Mathews et al. / Fuel Processing Technology 121 (2014) 104–113

as it can provide structural information and is used in porosity fraction of the ‘hydrocarbon moieties’ and the ‘oxygen plus nitrogen
determinations [55]. The helium density is determined through helium moieties’ (Eq. (36)) where, x(C–H), x(O + N) are the mass fractions of
picnometry. While the analysis is not challenging, it can be time con- ‘hydrocarbon moieties’ and ‘oxygen plus nitrogen moieties’ respectively
suming; however, modern analytical devices are now highly automat- and d(C–H), d(O + N) are the real densities of ‘hydrocarbon moieties’ and
ed, and the determination is becoming trivial. X-ray techniques can ‘oxygen plus nitrogen moieties’ respectively.
also be used for a true density determination. Coalification transitions:
loss of oxygen, hydrogen etc. with reorientation and enhanced stack-
1=do ¼ xðC−HÞ =dðC−HÞ þ xðOþNÞ =dðOþNÞ ð36Þ
ing are partially responsible for the general trend of density decreasing
then increasing with rank parameters. The inflection point is around
85% C (daf basis) [47]. Many correlations have been developed to re- Based on the dependency of the specific volume of hydrocarbon
late organic density (also termed true density, or the helium density es- moieties (1/d(C–H)) on different characteristics the linear variation
timate) to the elemental composition. Parkash [55] developed a of specific volume of oxygen plus nitrogen moieties (1/d(O + N))
correlation (Eq. (29)) to estimate true density from the elemental com- with (O + N) content and subsequently calculating regression coeffi-
position by performing a multivariate analysis of the data from 32 Al- cients using measured organic matter density data they were able to de-
berta (Canada) subbituminous coals. termine the density of organic matter (Eq. (37)).

True Density ¼ 3:5742−ð00:0197½C  þ 0:0192½OÞ−0:0691½H  ð29Þ h i

1=do ¼ 0:6625–0:5696 C =ðC þ HÞ–0:4820C þ H þ 1:0494½C  ð37Þ
Neavel et al. [56] developed a similar equation based on the data
obtained by analyzing 66 fresh, vitrinite-rich U.S. coals (Eq. (30)).
Fig. 5 shows the plots of Cwt.% (dmmf) versus density (particle or
True Density ¼ 0:023½C  þ 0:0292½O−0:0261½H  organic density). The general expected trend for the variation of coal
þ 0:0225½S−0:765 ð30Þ density with carbon is to observe a gradual reduction as the carbon
content increases until about 85% then increases rapidly for carbon
Neavel et al. [16] also developed a multivariate correlation to predict values greater than approximately 93% as shown in Fig. 5. There is lim-
density (He pycnometry) from the elemental composition by using ited experimental density data available so Gan et al. [60] and Walker
analytical data from 66 samples starting from lignite to low-volatile et al. Penn State Coal data [58] was plotted for reference. For Neavel
bituminous coals (Eq. (31)). et al.'s [16] correlations, the point of inflexion is around 85% for the den-
sity curve, but it fails to rise sharply enough after 90% C. The Walker et al.
Density ¼ 0:01556½C −0:04117½H þ 0:02247½O þ 0:02049½S [58] correlation captures the expected trend. Mazumdar [57] used a
þ 0:0208½Ash ð31Þ modified H/C ratio to calculate density. The modified (H/C) ratio calcu-
lation used functional group distribution estimates for heteroatoms.
Additionally, they also developed a correlation to estimate the den- Here the standard H/C ratio was used for density predictions as func-
sity of organic matter in coal on a dry-ash-free basis (Eq. (32)) tional group distribution data was not readily available. The density fol-
lows the trend but less effectively in comparison to other correlations.
Densityorganic ¼ 1:534−0:05196½H  þ 0:007375½O–0:02472½N Parkash's [55] trend does not capture the trend initially; predictions
þ 0:003853½S ð32Þ are scattered between 70% and 90% carbon, increasing after Strugala's
[59] correlation for the organic density shows a relatively flat correla-
Mazumdar [57] used the modified H/C atomic ratios to calculate tion for low-ranks with considerable scatter. A steep increase occurs
helium density from the elemental composition. This correlation is after 90% C and is similar to the experimental observations [60,61].
generally applicable to lignite to semianthracite (Eq. (33)).
3. Predicting behavior
20 ’ 2
d 4 ¼ 1:714−0:686½H=C  þ 0:5½ðH=C−0:6Þ ð33Þ
3.1. HGI
Walker et al. [58] conducted experiments on a large number of coal
macerals of different ranks with He, Hg, methanol, water and CO2 to The Hardgrove grindability index (HGI) is an important practical
determine densities, porosities and surface areas. Subsequently they measure for coal handling and utilization, especially for pulverized-
developed a correlation to estimate the densities from elemental com- coal-fired boilers [62]. HGI reflects the characteristics of coal in terms
position (Eq. (34)). of fracture, hardness, and tenacity and is related to coal rank, petrogra-
phy, moisture, and mineral composition [63]. HGI is experimentally de-
ρHe ¼ 0:033½C  þ 0:036½O−0:031½H  þ 0:009½S−1:577 ð34Þ termined and even though the equipment is not expensive, the process
is tedious and therefore there is an interest in predicting its value [64].
Strugala [59] developed correlations to determine the “real” density Attempts have been made to correlate HGI with coal properties such
of hard coals by using ultimate and proximate analysis data of coal as moisture content, ash yield, vitrinite reflectance, etc. [65–68]. By
samples obtained from different seams of Polish coal mines. The “true” means of an empirical modeling tool, Chelgani et al. [64] evaluated the
density (dHe) was expressed as a combination of the density of mineral complex relationship between HGI and elemental analysis (Eq. (38)).
matter (dM) and the density of organic matter (do) where, Md is the per- Data for 600 Kentucky (presumably all bituminous) coals were used
centage of mineral matter in dry coal (Eq. (35)). to determine this correlation, and a fraction of variance (R2) of 0.75
    was achieved [64].
d d
1=dHe ¼ 1−M =100  ð1=do Þ þ M =100  ð1=dM Þ ð35Þ HGI ¼ 77:162 þ 3:994  ln½S−10:920½H 
þ 1:904½M −0:424½ASH−11:765  lnð½O þ N=C Þ ð38Þ

The density of mineral matter (dM) can be calculated if the composi-

tion of different minerals is known by using a simple arithmetic summa- 3.2. Thermal swelling
tion of the product of their respective densities and fractions. See
citation for a detailed discussion [59]. The density of organic matter The free swelling index (FSI) is an empirical value that can be used to
in coal (do) is expressed as the arithmetic sum of the product of the determine the caking characteristics of coals. It is measured by heating
J.P. Mathews et al. / Fuel Processing Technology 121 (2014) 104–113 111

Fig. 5. Plots of Cwt.% (dmmf) versus density (particle or organic density).

particulate coal in a closed crucible to 820 °C and measuring the volume relationship between the sulfur content of coal and the total conversion
of the agglomerated mass after devolatilization. By definition, only was observed, two separate equations were derived; one for low sulfur
bituminous coals possess caking abilities, and therefore the test is coals (Eq. (41)) and one for coals with content higher than 1.6%
often used to evaluate coking coals. Neavel et al. attempted to correlate (Eq. (42)). Both correlations are only valid for coals with a total reactive
the free swelling index to the oxygen content and total non-plastic com- maceral content above 70%.
ponents (Eqs. (39) and (40)) [16]. The non-plastic components in the
coal were defined as the sum of the mineral matter and the total Conversion ð% Þ ¼ 3:03  ½S þ 0:27  ðtotal reactivesÞ–0:82  ½C  þ 109
inertinite and liptinite content. Coals with N 15% oxygen were consid- ð41Þ
ered to have no swelling properties; hence, the correlation was devel-
oped only for coals of oxygen content between 2.5% and 15 %dmmf by Conversion ð% Þ ¼ 3:40  ½S þ 0:31  ðtotal reactivesÞ–0:66  ½C  þ 91:9
mass. In their Eq. (38), NPC is defined as the sum of ash value (weight ð42Þ
%) and the total inertinite content (volume %) and LIP is the total
liptinite content. The behavior however is very sensitive to the condi-
tions used; thus, applicability is limited to these specific conditions
3.4. Others
and the rank range considered.
Using full maceral reflectogram data from an automated coal petrology
FSI ¼ 5 þ 4:68f ðOÞ−0:00395  NPC þ 0:0897  NPC−0:0601  LIP ð39Þ approach, O'Brian et al. related (separately) various compositional data
and volatile matter for Australian coals and for a few coals coke
f ðOÞ ¼ sinð0:2513  ½O þ 0:9505Þ ð40Þ parameters (reactivity index and coke strength after reaction) with
graphical approaches [72]. Quick and Glick determined CO2 predictions
(kg CO2 per net GJ) for coal emissions, from regression analyses, yield-
3.3. Direct liquefaction ing a correlation for coals b69% fixed carbon (Eq. (43)) and for N 69%
(Eq. (44)) mostly based on PSU Coal Sample Bank data [73]. CVg is the
Coal rank and petrographic composition are important factors in gross calorific value (MJ kg− 1) and can be predicted from elemental
direct liquefaction behavior [62] but yields are sensitive to the condi- composition.
tions used, and catalyst, so these correlation attempts are also limited 2
to those conditions. Bright low-rank coals are generally more reactive CO2 ¼ 0:03663  ðCVgÞ —ð2:584  CVg Þ þ 136:5 ð43Þ
during non-catalyzed direct liquefaction [69–71]. Abdel-Baset et al.
[70] investigated the relationship between elemental composition and CO2 ¼ 0:0101  ðFC Þ —ð1 : 29  FC Þ þ 132 : 2 ð44Þ
liquefaction conversion of 68 coals ranging from lignite to high-
volatile bituminous coals. The highest yields were found for coals with Winschel had demonstrated a general reduction of CO2 emissions
80–85% carbon (dmmf) [70]. Through multiple stepwise regression with increasing S content [74]. Thus, for international coals, Quick deter-
analysis, linear equations were developed that relate conversion to sul- mined equations for the CO2 emission estimates [75]. Where Sef is an S
fur, carbon, and the amount of reactive macerals. Since a strong emission factor, CV is the calorific value (m,mmf basis), FC is fixed
112 J.P. Mathews et al. / Fuel Processing Technology 121 (2014) 104–113

carbon (dmmf). Again equations differentiate between low-rank to [22] P. Thipkhunthod, V. Meeyoo, P. Rangsunvigit, B. Kitiyanan, K. Siemanond, T.
Rirksomboon, Predicting the heating value of sewage sludges in Thailand from
high-volatile bituminous coals (Eq. (45)), and for medium-volatile bitu- proximate and ultimate analyses, Fuel 84 (7–8) (2005) 849–857.
minous and higher (Eq. (46)). [23] A. Demirbas, Theoretical heating values and impacts of pure compounds and fuels,
Energy Sources, Part A: Recovery, Utilization, and Environmental Effects 28 (5)
    (2006) 459–467.
−2 2
CO2 ¼ 108:73– 1:647  Sef –ð1:085  ½CV Þ þ 1:45E  ½CV  ð45Þ [24] S.U. Patel, B.J. Kumar, Y.P. Badhe, B.K. Sharma, S. Saha, S. Biswas, A. Chaudhury, S.S.
Tambe, B.D. Kulkarni, Estimation of gross calorific value of coals using artificial neu-
    ral networks, Fuel 86 (3) (2007) 334–344.
CO2 ¼ 144:45– 0:7647  Sef –ð1:6521  ½FC Þ þ 1:228E  ½FC  ð46Þ [25] S. Mesroghli, E. Jorjani, S. Chehreh Chelgani, Estimation of gross calorific value based
on coal analysis using regression and artificial neural networks, International Jour-
nal of Coal Geology 79 (1–2) (2009) 49–54.
[26] A.K. Majumder, R. Jain, P. Banerjee, J.P. Barnwal, Development of a new proximate
analysis based correlation to predict calorific value of coal, Fuel 87 (13–14) (2008)
4. Conclusions 3077–3081.
[27] W.A. Selvig, F.H. Gibson, Calorific value of coal, in: H.H. Lowry (Ed.), Chemistry of
There is a rich history and a wide variety of predictive equations that Coal Utilization, vol. 1, J. Wiley & Sons, New York, 1945, pp. 132–144.
[28] D.M. Mason, K.N. Gandhi, Formulas for calculating the calorific value of coal and coal
relate to elemental composition and coal properties. Specifically, chars: development, tests, and uses, Fuel Processing Technology 7 (1) (1983) 11–22.
there have been attempts to predict the higher heating value, vola- [29] D.W. Van Krevelen, Coal Topology Physics Chemistry Constitution, Third edition,
tile matter yield, vitrinite reflectance, aromaticity, density, the Elsevier, London, 1993.
[30] S.C. Chelgani, S. Mesroghli, J.C. Hower, Simultaneous prediction of coal rank param-
Hardgrove Grindability Index, the free swelling index, pyrolysis eters based on ultimate analysis using regression and artificial neural network,
yields, direct liquefaction yields, and carbon dioxide yields. Greater International Journal of Coal Geology 83 (1) (2010) 31–34.
than 40 relationships were collected and where possible evaluated [31] ASTM International, ASTM D 388–05 Standard Classification of Coals by Rank, ASTM
Standards, 2005.
for their utility broadly with a subset of the Penn State University
[32] ASTM International, ASTM D3175—11 Standard Test Method for Volatile Matter in
Coal Sample Bank and Database. For some relationships, the causa- the Analysis Sample of Coal and Coke, 2011.
tion directly produces good agreement, typically however for a lim- [33] ASTM International, ASTM D7582—10 Standard Test Methods for Proximate Analy-
sis of Coal and Coke by Macro Thermogravimetric Analysis, 2010.
ited rank range that reflects the correlations creation intent. For
[34] J.T. Riley, E.G. Yanes, M. Marsh, D. Lawrenz, L. Eichenbaum, Coal and coke vola-
other correlations, the often-systematic transformations accompa- tile matter determination and reconciliation of differences in yields determined
nying coalification typically enable a reasonable proxy or reasonable by two ASTM methods, Journal of Testing and Evaluation 38 (4) (2010)
trend given the complex nature of coal. 458–466.
[35] N.N. Chatterjee, D. Chandra, T.K. Ghosh, Calculation of chemical composition of com-
posite samples of coal from maceral composition, Economic Geology 63 (1968)
References 80–84.
[36] A.G. Borrego, G. Marban, M.J.G. Alonso, D. Alvarez, R. Menendez, Maceral effects in
[1] C.A. Seyler, The classification of coal, Transactions of the American Institute of the determination of proximate volatiles in coals, Energy & Fuels 14 (1) (2000)
Mining, Metallurgical and Petroleum Engineers 76 (1928) 189–196. 117–126.
[2] D.W. Van Krevelen, Graphical-statistical method for the study of structure and reac- [37] J.P. Calderwood, An empirical method for determining the ultimate from the proxi-
tion processes of coal, Fuel 29 (1950) 269–284. mate analysis of coal, (Thesis (M.S.)) Pennsylvania State College, 1916.
[3] E. Stach, M.-T. Mackowsky, M. Teichmuller, G.H. Taylor, D. Chandra, R. Teichmuller, [38] K. Koba, Statistical-analyses and prediction of coking properties of coal, Fuel 59 (6)
Coal Petrology, Gebrueder Borntraeger, Berlin, Germany, 1982. (1980) 380–388.
[4] D.W. Van Krevelen, Coal Science and Technology 3, Coal: Typology - Chemistry - [39] L.A. Flueckinger, Statistical evaluation of coal compositional data, Journal of Geology
Physics - Constitution, 3rd ed. Elsevier, New York, 1981. 80 (2) (1972) 237–247.
[5] W. Kalkreuth, M. Steller, I. Wieschenkamper, S. Ganz, Petrographic and chemical [40] J.T. McCartney, M. Teichmüller, Classification of coals according to degree of coalifi-
characterization of Canadian and German coals in relation to utilization potential.1. cation by reflectance of vitrinite component, Fuel 51 (1) (1972) 64–68.
Petrographic and chemical characterization of feedcoals, Fuel 70 (6) (1991) [41] R.C. Neavel, Origin, petrography and classification of coal, in: M.A. Elliot (Ed.), Chem-
683–694. istry of Coal Utilization Second Supplementary Volume, John Wiley & Sons, New
[6] The Pennsylvania State University Penn State Coal Sample Bank and Data Base, York, 1981. [42] A.K. Burnham, J.J. Sweeney, A chemical kinetic-model of vitrinite maturation
[7] G.D. Mitchell, J.P. Mathews, Penn State’s Coal Repository: Penn State and Argonne and reflectance, Geochimica et Cosmochimica Acta 53 (10) (1989)
Premium Coal Sample Bank & Database, International Conference on Coal Science 2649–2657.
& Technology, The Pennsylvania State University, PA, USA, 2013, pp. 1122–1132. [43] S. George, J. Smith, D. Jardine, Vitrinite suppression in coal due to marine transgression:
[8] M.F. Kessler, Interpretation of the chemical composition of bituminous coal case study of the organic-geochemistry of the Greta seam, Sydney Basin, The APEA
macerals, Fuel 52 (1973) 191–197. Journal 34 (1984) 241–255.
[9] G.R. Dyrkacz, C.A.A. Bloomquist, L. Ruscic, High-resolution density variations of coal [44] J.R. Smith, J.W. Smith, A relationship between the carbon and hydrogen content of
macerals, Fuel 63 (10) (1984) 1367–1373. coals and their vitrinite reflectance, International Journal of Coal Geology 70 (1–3)
[10] M.A. Wilson, R.J. Pugmire, J. Karas, L.B. Alemany, W.R. Woolfenden, D.M. Grant, P.H. (2007) 79–86.
Given, Carbon distribution in coals and coal macerals by CPMAS 13C NMR spectros- [45] J.J. Sweeney, A.K. Burnham, Evaluation of a simple-model of vitrinite reflectance
copy, Analytical Chemistry 56 (1984) 933–943. based on chemical kinetics, AAPG Bulletin 74 (10) (1990) 1559–1570.
[11] R. Winans, E. Hayatsu, R. Ryoichi, G. Scott, R.L. McBeth, Reactivity and characteriza- [46] M. Khandelwal, T.N. Singh, Prediction of macerals contents of Indian coals from proxi-
tion of coal macerals, Chemistry and Characterization of Coal Macerals, vol. 252, mate and ultimate analyses using artificial neural networks, Fuel 89 (5) (2010)
American Chemical Society, 1984, pp. 137–155. 1101–1109.
[12] P.H. Singh, M.C. Kakati, New models for prediction of specific energy of coal, Fuel 73 [47] D.W. Van Krevelen, The spatial structure of the coal molecules, Coal Science and
(2) (1994) 301–303. Technology 3, Coal Typology - Chemistry - Physics - Constitution, 3rd ed., Elsevier,
[13] ASTM International, ASTM D5865—10 Standard Test Method for Gross Calorific New York, 1981, pp. 313–342.
Value of Coal and Coke, 2011. [48] B.K. Mazumdar, Aromaticity of coal—a reappraisal of the graphical-densimetric ap-
[14] I.G.C. Dryden, M. Griffith, Calorific value and the chemical constitution of coal, Fuel proach, Fuel Processing Technology 19 (2) (1988) 179–202.
34 (1955) S36–S47. [49] S. Niksa, FLASHCHAIN theory for rapid coal devolatilization kinetics. 3. Modeling the
[15] P.H. Given, D. Weldon, J.H. Zoeller, Calculation of calorific values of coals from ulti- behavior of various coals, Energy & Fuels 5 (1991) 637–683.
mate analyses—theoretical basis and geochemical implications, Fuel 65 (6) (1986) [50] B.C. Gerstein, P.D. Murphy, M.L. Ryan, Aromaticity in coal, in: R.A. Meyers (Ed.), Coal
849–854. Structure, Academic Press, New York, 1982.
[16] R.C. Neavel, S.E. Smith, E.J. Hippo, R.N. Miller, Interrelationships between coal com- [51] D. Genetti, T.H. Fletcher, R.J. Pugmire, Development and application of a correlation
positional parameters, Fuel 65 (3) (1986) 312–320. of C-13 NMR chemical structural analyses of coal based on elemental composition
[17] B.K. Mazumdar, Theoretical oxygen requirement for coal combustion: relationship and volatile matter content, Energy & Fuels 13 (1) (1999) 60–68.
with its calorific value, Fuel 79 (11) (2000) 1413–1419. [52] M.M. Maroto-Valer, G.D. Love, C.E. Snape, Relationship between carbon aromatic-
[18] S.A. Channiwala, P.P. Parikh, A unified correlation for estimating HHV of solid, liquid ities and H/C ratios for bituminous coals, Fuel 73 (12) (1994) 1926–1928.
and gaseous fuels, Fuel 81 (8) (2002) 1051–1063. [53] M.M. Maroto-Valer, J.M. Andresen, C.E. Snape, Verification of the linear relationship
[19] B.K. Mazumdar, Konovalov correlation for molar heats of combustion of organic between carbon aromaticities and H/C ratios for bituminous coals, Fuel 77 (7)
compounds: its adoption for coal, Fuel 81 (1) (2002) 119–124. (1998) 783–785.
[20] J. Parikh, S.A. Channiwala, G.K. Ghosal, A correlation for calculating HHV from prox- [54] B.K. Mazumdar, On the relationship between carbon aromaticities and H C ratios for
imate analysis of solid fuels, Fuel 84 (5) (2005) 487–494. bituminous coals, Fuel 78 (10) (1999) 1239–1241.
[21] C.D. Sheng, J.L.T. Azevedo, Estimating the higher heating value of biomass fuels from [55] S. Parkash, True density and elemental composition of subbituminous coals, Fuel 64
basic analysis data, Biomass and Bioenergy 28 (5) (2005) 499–507. (5) (1985) 631–634.
J.P. Mathews et al. / Fuel Processing Technology 121 (2014) 104–113 113

[56] R.C. Neavel, S.E. Smith, E.J. Hippo, R.N. Miller, Coal characterization research: sample [66] H.B. Vuthaluru, R.J. Brooke, D.K. Zhang, H.M. Yan, Effects of moisture and coal blend-
selection, preparation, and analyses, Preprint Papers—American Chemical Society, ing on Hardgrove Grindability Index of western Australian coal, Fuel Processing
Division of Fuel Chemistry 25 (1980) 246–257. Technology 81 (1) (2003) 67–76.
[57] B.K. Mazumdar, Molecular structure and molar volume of organic compounds and [67] E. Jorjani, J.C. Hower, S. Chehreh Chelgani, M.A. Shirazi, S. Mesroghli, Studies of
complexes with special reference to coal, Fuel 78 (1999) 1097–1107. relationship between petrography and elemental analysis with grindability for
[58] P.L. Walker, S.K. Verma, J. Rivera-Utrilla, A. Davis, Densities, porosities and surface Kentucky coals, Fuel 87 (6) (2008) 707–713.
areas of coal macerals as measured by their interaction with gases, vapours and [68] B.V. Rao, S.J. Gopalakrishna, Hardgrove grindability index prediction using support
liquids, Fuel 67 (1988) 1615–1623. vector regression, International Journal of Mineral Processing 91 (1–2) (2009)
[59] A. Strugala, Empirical formulae for calculation of real density and total pore volume 55–59.
of hard coals, Fuel 73 (11) (1994) 1781–1785. [69] W. Gabzdyl, A.K. Varma, Thermogravimetric studies in prediction of coal behavior
[60] H. Gan, S.P. Nandi, P.L. Walker, Nature of the porosity in American Coals, Fuel 51 during coal liquefication, Journal of Thermal Analysis 39 (1) (1993) 31–40.
(1972) 272–277. [70] M.B. Abdel-Baset, R.F. Yarzab, P.H. Given, Dependence of coal liquefaction behavior
[61] J.A. Dulhunty, R.E. Penrose, Some relations between density and rank of coal, Fuel 30 on coal characteristics.3. Statistical correlations of conversion in coal-tetralin inter-
(1951) 109–113. actions, Fuel 57 (2) (1978) 89–94.
[62] I. Suarez-Ruiz, J.C. Crelling, The role of petrology in coal utilization, in: S.-R. Isabel, [71] P.H. Given, D.C. Cronauer, W. Spackman, H.L. Lovell, A. Davis, B. Biswas, Dependence
C.C. John (Eds.), Applied Coal Petrology, Elsevier, Burlington, 2008. of coal liquefaction behavior on coal characteristics.1. Vitrinite-rich samples, Fuel 54
[63] S. Ural, M. Akyildiz, Studies of the relationship between mineral matter and grinding (1) (1975) 34–39.
properties for low-rank coals, International Journal of Coal Geology 60 (1) (2004) [72] G. O'Brien, B. Jenkins, J. Esterle, H. Beath, Coal characterisation by automated coal
81–84. petrography, Fuel 82 (9) (2003) 1067–1073.
[64] S.C. Chelgani, J.C. Hower, E. Jorjani, S. Mesroghli, A.H. Bagherieh, Prediction of coal [73] J.C. Quick, D.C. Glick, Carbon dioxide from coal combustion: variation with rank of
grindability based on petrography, proximate and ultimate analysis using multiple US coal, Fuel 79 (7) (2000) 803–812.
regression and artificial neural network models, Fuel Processing Technology 89 [74] R.A. Winschel, The relationship of carbon-dioxide emissions with coal rank and sulfur-
(1) (2008) 13–20. content, Journal of the Air & Waste Management Association 40 (6) (1990) 861–865.
[65] A.N. Sengupta, An assessment of grindability index of coal, Fuel Processing Technol- [75] J.C. Quick, Carbon dioxide emission factors for U.S. coal by origin and destination,
ogy 76 (1) (2002) 1–10. Environmental Science & Technology 44 (7) (2010) 2709–2714.