Documentos de Académico
Documentos de Profesional
Documentos de Cultura
Surface Science Western, University of Western Ontario, London, Ont., Canada N6G 0J3
Center for Applied Energy Research, University of Kentucky, 2540 Research Park Drive, Lexington, KY 40511, USA
a r t i c l e
i n f o
Article history:
Received 13 November 2009
Received in revised form 22 August 2010
Accepted 26 September 2010
Available online 20 October 2010
Keywords:
Free-swelling index (FSI)
Proximate and ultimate analysis
Regression
Articial neural network
a b s t r a c t
The effects of proximate, ultimate and elemental analysis for a wide range of American coal samples on
Free-swelling Index (FSI) have been investigated by multivariable regression and articial neural network
methods (ANN). The stepwise least square mathematical method shows that variables of ultimate analysis are
better predictors than those from proximate analysis. The non linear multivariable regression, correlation
coefcients (R2) from ultimate analysis inputs was 0.71, and for proximate analysis input variables was 0.49.
With the same input sets, feed-forward articial neural network (FANN) procedures improved accuracy of
predicted FSI with R2 = 0.89, and 0.94 for proximate and ultimate analyses, respectively. The ANN based
prediction method, as a rst report, shows FSI is a predictable variable, and ANN can be further employed as a
reliable and accurate method in the free-swelling index prediction.
2010 Elsevier B.V. All rights reserved.
1. Introduction
Coke is an expensive component in the steel-making process [1]
where about 90% of the coke produced from blends of coking coals is
used to maintain the process of iron production in the blast furnace
[2]. The plasticity of coal during heating, measured by rheological tests
of coals in their softened state [3], is the major factor responsible for
coke formation [4].
The uid and swelling properties of coals relate to the ability of the
reactive components in a coal to fuse with the inert material in the
coal, thereby making a strong coke. Impurities present in coke affect
its performance in the blast furnace by decreasing its role as a fuel in
terms of carbon available for direct and indirect reduction roles and its
role as a permeable support. These impurities include moisture,
volatile matter, ash, sulfur, phosphorous, and alkali contents [2].
In spite of the relative smallness in the amount of inorganic mineral
content, the rank and chemistry of the parent coal strongly inuences
the optical textures of cokes [58]. Some mineral matters (for example,
calcium-containing substances) deteriorate the thermoplastic properties of coals, decrease their swelling and simultaneously favor the yield
of solid residue (coke) [9,10].
Because the inorganic components of the coal remain in the coke,
the ash content of the coal can have an adverse affect on the coke
quality. Coke properties decrease in proportion to increasing ash
content of the coal [11,12]. Diez et al. investigated how coke reactivity
is affected by ash components, especially Fe2O3 and K2O [2]. The alkali
Corresponding author. Tel.: + 1 519 702 9356.
E-mail address: Sos4552@gmail.com (S.C. Chelgani).
0378-3820/$ see front matter 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.fuproc.2010.09.027
Table 1
The number of samples for different states.
State
Number of samples
State
Number of samples
Alabama
Illinois
Kansas
Iowa
New Mexico
Oklahoma
Tennessee
Virginia
Wyoming
733
16
21
53
29
29
51
320
16
Colorado
Indiana
Kentucky
Missouri
Ohio
Pennsylvania
Utah
West Virginia
96
97
798
65
581
354
66
366
350
Table 2
The ranges of proximate and ultimate analysis of coal samples (as received).
Variables
Minimum
Maximum
Mean
Std. deviation
Moisture
Volatile matter
Ash
Hydrogen
Carbon
Nitrogen
Oxygen
Pyritic sulfur
Organic sulfur
Hydrogen exclusive
Oxygen exclusive
SiO2
Al2O3
CaO
MgO
MnO
Na2O
K2O
Fe2O3
TiO2
P2O5
SO3
0.90
25.70
1.00
3.60
46.70
0.60
2.00
0.00
0.00
3.08
1.12
0.00
0.00
.0.00
0.20
0.00
0.00
0.00
0.00
0.00
0.00
0.00
11.10
45.30
20.00
6.10
82.20
2.00
19.50
7.26
3.57
5.74
13.38
73.00
39.00
30.00
2.90
0.94
8.40
4.40
69.00
3.20
2.80
19.00
3.89
34.03
11.73
5.06
68.39
1.40
11.72
.8472
0.72
4.63
8.26
45.48
24.71
2.91
0.85
.032
0.50
1.85
14.15
1.27
0.42
2.64
1.55
3.02
4.82
0.34
5.41
0.26
2.55
1.32
0.48
0.37
1.62
13.05
7.61
3.91
0.40
0.07
0.72
0.98
14.66
0.55
0.42
3.01
2. Experimental data
Data used to test the proposed approaches are from U.S. Geological
Survey Coal Quality (COALQUAL) database, open le report 97-134 [17].
The database includes the determined proximate and ultimate analysis,
as well as free-swelling index (FSI) on an as received basis. The samples
with more than 20% ash, and volatile matter (dry ash free) less than 32%,
Fig. 1. Distribution of difference between actual and estimated FSI from Eq. (3) according to coking ranges.
351
higher K2O contents in coal (0.20) results higher FSI. No other elemental
analysis parameters were signicant. The ranges of proximate and
element analysis of coal samples are shown in Table 2. The best
multivariable equations between the proximate and element analysis
and FSI can be presented as the following equations:
3.1. Regression
R2 = 0:45
R2 = 0:47
Fig. 2. Distribution of difference between actual and estimated FSI from Eq. (7) according to coking ranges.
R2 = 0:49
352
R = 0:69
Fig. 3. Predicted FSI by neural network using proximate analysis versus actual measured
FSI in testing and validation process.
R = 0:71
Fig. 4. Predicted FSI by neural network using ultimate analysis versus actual measured
FSI in testing and validation process.
353
Fig. 5. Distribution of difference between actual and FANN predicted FSI in testing phase using proximate analysis according to coking ranges.
Ap meanAps
stdAp
The validation and testing phase results show the capability of the
model. The R2 values for validations were R2 = 0.88 for proximate
analysis input data seta, and 0.94 using ultimate analysis. The results
show the model has an ability to reliably predict FSI values (Figs. 4, 5). In
the testing stage, samples (n = 400) over a wide range of FSI values were
applied to certify the potential of FANN model. Figs 4 and 5 show the
relationship between estimated FSIs by FANN model from testing phase
and the FSI value determined by the traditional ASTM method. The R2
values for the testing sets were 0.89 for proximate analysis input data
seta, and 0.95 using ultimate analysis inputs. Upon comparison with the
regression results, the FANN model was shown greater accuracy to
predict FSI values from the same inputs, and (as a rst report) were
convinced FSI is predictable. According to Figs. 5 and 6, that show the
distribution of differences between FANN calculated FSIs and actual
determined FSIs, it can be seen that ultimate rather than proximate
analysis parameters, provide a much better estimation of the FSI values.
4. Conclusion
Inter correlations between ultimate and proximate analysis with FSI
show that with the increase content of ash, moisture, volatile matter,
354
Fig. 6. Distribution of difference between actual and FANN predicted FSI in testing phase using ultimate analysis according to coking ranges.
355