STATISTICAL ANALYSYS OF THE SCFE OF A BRAZILAN MINERAL COAL

STATISTICAL ANALYSYS OF THE SCFE OF A BRAZILAN MINERAL COAL¹

Cláudio DARIVA², José Vladimir OLIVEIRA², ^*, José C. PINTO², Maria G. VALE³, Elina B. CARAMÃO³

SUMMARY

The influence of some process variables on the productivity of the fractions (liquid yield times fraction percent) obtained from SCFE of a Brazilian mineral coal using isopropanol and ethanol as primary solvents is analyzed using statistical techniques. A full factorial 2³ experimental design was adopted to investigate the effects of process variables (temperature, pressure and cosolvent concentration) on the extraction products. The extracts were analyzed by the Preparative Liquid Chromatography-8 fractions method (PLC-8), a reliable, non destructive solvent fractionation method, especially developed for coal-derived liquids. Empirical statistical modeling was carried out in order to reproduce the experimental data. Correlations obtained were always greater than 0.98. Four specific process criteria were used to allow process optimization. Results obtained show that it is not possible to maximize both extract productivity and purity (through the minimization of heavy fraction content) simultaneously by manipulating the mentioned process variables.

Keywords: Statistical analysis, SCFE, productivity, PLC-8 characterization

RESUMO

Esta trabalho apresenta uma análise estatística da influência das variáveis de processo na produtividade das frações (rendimento líquido vezes percentagem da fração) obtidas através da extração supercrítica de carvão mineral utilizando como solventes primários etanol e isopropanol. Um projeto fatorial completo com 2³ experimentos (com triplicata no ponto central) foi adotado objetivando-se a investigação dos efeitos das variáveis de processo (temperatura, pressão e concentração de co-solventes) nos produtos da extração. Os extratos foam analisados pela técnica de Cromatografia Líquida Preparativa – 8 Frações, uma técnica confiável, não-destrutiva, especialmente desenvolvida para derivados líquidos do carvão. A modelagem estatística, empregando-se modelos empíricos foi efetuada, visando a reprodução dos dados experimentais apresentados neste trabalho. As correlações obtidas foram sempre superiores a 98%. Os resultados obtidos pela utilização de quatro critérios de otimização mostraram que não é possível maximizar produtividade e pureza simultaneamente manipulando-se as variáveis de processo acima citadas e, portanto, uma análise econômica dos produtos resultantes deve ser considerada.

Palavras-chave: Análise estatística, extração supercrítica, carvão mineral, método PLC-8

1 — INTRODUCTION

The SCFE technique as applied to mineral coals employs compressed solvents near their critical temperature to extract/solubilize the low molecular weight, decomposed, and entrapped volatiles as they evolve at temperatures where such compounds normally would not distill [8]. The liquid products obtained may be used as an alternate source of energy to crude oil or as a substitute for petroleum-based feedstocks for the chemical industry [9]. The influence of process variables on the characteristics of the extracts are usually evaluated based solely on the distribution of the fractions obtained. However, from a practical viewpoint this analysis should also take the productivity of the fractions (liquid yield times fraction percent) into account, which means that a compromise between liquid yield and fractions distribution should be sought.

The statistical design of experiments is a rational and efficient way to assess the influence of input variables in a process. In spite of that, such statistical tools have been seldomly used to study the SCFE of coal. Available studies [2,8] have employed the factorial experimental design to evaluate the influence of process variables on the extraction yield and conversion.

The main objective of this work is to analyze the influence of some process variables (temperature, pressure and addition of cosolvent) on the productivity of the fractions obtained from SCFE of a Brazilian mineral coal using ethanol and isopropanol as primary solvents. Empirical models are built to reproduce experimental data and different optimization criteria are used to investigate whether it is possible to optimize the process by manipulating the variables studied.

2 — MATERIALS AND METHODS

2.1 – Coal characterization

A high-ash sub-bituminous coal from Butiá-Brazil was used in the experiments. Its proximate and ultimate analysis are given in Table 1.

TABLE 1
. Proximate and ultimate analysis of Butiá-Brazil coal

Proximate analysis (wt%)
Moisture	1.1
Ashdb	33.4
Volatile matter (VM)db	27.3
Fixed carbon (FC)db	39.3
Ultimate analysis (wt%)
C daf	78.1
H daf	5.1
N daf	1.5
(O + S)daf,a	15.3

^db

-dry-basis;

^daf-dry, ash-free,

^a-calculated by differenc

2.2 – Apparatus and procedure

The experimental apparatus is shown in Fig. 1. It consists of a 97.5 ml extraction vessel with an electrical heater provided with a PID temperature controller with a precision of ± 0.1 K. The pressure is monitored by an absolute pressure transducer with a precision of ± 0.012MPa and the data acquisition is made by a portable programmer.

After the desired extraction temperature is reached, about 40g of a high-ash sub-bituminous coal, Butiá - Brazil, with a particle size of 16-20 mesh, dried overnight at 383 K, is charged into the extraction vessel, supported by two 260 mesh wire disks. The system is immediately closed and the solvent is continuously fed into the vessel by a high pressure pump. The product recovery system is formed by a glass collector at atmospheric pressure, which collects the solvent and the extract after depressurization by a micrometering valve, and a cold trap to avoid solvent losses.

FIGURE 1.
Schematic diagram of the SCFE apparatus. A-solvent reservoir; B-high pressure pump; C-pre-heating coil; D-extraction vessel; E-absolute pressure transducer; F-glass collector; G-trap.

The experiments were accomplished in 90 min., isothermally, at constant pressure and using a ratio solvent:coal of 6.3 + 0.1. After the extraction is completed, the solvent delivery system is turned off and the system is brought to atmospheric pressure.

The process variables investigated are: temperature (T) (598-698 and 598-673K for ethanol and isopropanol, respectively), pressure (P) (7.5-12.5 and 5.5-9.5MPa for ethanol and isopropanol, respectively) and cosolvent concentration(Cos) (0-10 mol% of water).

2.3 – Extract analysis

The extracts were analyzed by the Preparative Liquid Chromatography-8 fractions method (PLC-8), which is an extension of the SARA method [5], combining solubility and chromatographic fractionation. This method provides eight distinct chemical classes: five non-polar fractions (saturated hydrocarbons, monoaromatics, diaromatics, triaromatics and polycyclic aromatics), one fraction of intermediate polarity (resins) and two high molecular weight polar fractions (asphaltenes and asphaltols). A detailed discussion about this subject is presented by Karam et al. [4] and hence only a brief description is provided here. The samples (300 mg) were dissolved in a small amount of tetrahydrofuran (THF) and the solution was stirred with silica gel (2g). After the solvent evaporation, the sample-coated silica gel was placed on a 70 cm x 11 mm id glass column with teflon stopcock, previously wet-packed with silica gel (18g) in hexane. The solvent from each fraction was removed under vacuum by rotatory evaporation (323 K). This method has been applied successfully to characterize Brazilian coals [3,6]].

The liquid yield and productivity of the fractions are here defined by:

(1)

(2)

3 — RESULTS AND DISCUSSION

A full factorial 2³ experimental design was adopted to evaluate the effects of temperature, pressure and addition of water to both ethanol and isopropanol on the productivities of the extracts obtained from SCFE of a high-ash Brazilian mineral coal. Triplicate runs were performed for the central point, showing that output variability is below 3% for all variables. The extract fractions and their productivities were grouped into three classes: F1-F5 (light compounds), F6 (intermediate) and F7-F8 (heavy compounds). Tables 2 and ⁴ present a compilation of the experimental results obtained from SCFE of a Brazilian mineral coal using ethanol and isopropanol, respectively.

It can be observed from these tables that the increase of either temperature or pressure causes the increase in the liquid yield for both pure alcohols and alcohol-water mixtures. At the same temperature and pressure, however, the addition of cosolvent leads to a lower liquid yield when compared to the yields obtained with the pure alcohols. Concerning the fractionation of the extracts by the PLC-8 method, Tables 2 and ⁴ show that a temperature increase leads to an increase of the lighter compounds (F1-F5) at the expense of the higher-molecular-weight compounds (F7-F8) due to the more severe thermolysis of the coal structure. On the other hand, when pressure increases, the light fraction decreases, as the extraction of the heaviest compounds is favored because of the enhancement of the solvent power (solvent density). It should also be noted the importance of the PLC-8 fractionation method considering that it allows the analysis of the individual fractions. For instance, the addition of water to ethanol (Table 2) causes a reduction in the F2 fraction under all experimental conditions, whereas such behavior is not verified for the whole F1-F5 group fraction.

TABLE 2.
The effect of temperature, pressure and addition of water on the distribution and on the productivities of the fractions of the extracts from SCFE with ethanol of Butiá-Brazil coal

T (K)	598				635			673
P (MPa)	7.0		12.5		9.8			7.0		12.5
Cos(mol%)	0	10	0	10	5			0	10	0	10
Yield	2.71	2.22	5.65	4.65	3.49	3.61	3.55	3.77	3.66	8.52	7.71
F1	2.3	2.4	0.9	0.5	1.3	1.2	1.3	2.1	1.9	0.3	2.4
F2	0.8	0.4	1.1	0.1	0.3	0.2	0.3	0.7	0.6	1.4	0.1
F3	1.1	2.9	0.6	0.1	0.2	0.3	0.2	1.2	0.3	0.6	0.2
F4	2.9	2.7	1.5	1.0	2.2	2.2	2.2	3.3	5.1	1.7	1.5
F5	6.5	6.4	3.3	3.2	5.2	5.4	5.2	7.9	8.1	4.4	4.5
F6	61.4	58.4	63.0	65.3	70.1	70.2	70.0	71.0	65.4	69.7	73.6
F7	17.1	11.9	16.8	12.5	11.1	11.1	11.2	5.6	7.9	15.3	9.6
F8	7.8	15.0	12.7	17.5	9.6	9.4	9.6	8.1	10.7	6.6	8.4
F1-F5	13.7	14.7	7.5	4.7	9.2	9.2	9.2	15.3	16.0	8.4	8.4
F6	61.4	58.4	63.0	65.3	70.1	70.1	70.0	71.0	65.4	69.7	73.6
F7-F8	24.9	26.9	29.5	30.0	20.7	20.7	20.8	13.7	18.6	21.9	18.0

The independent variables were normalized in the range of -1 to +1, so as to allow a direct comparison of each variable effect. The "-1" level represents the inferior limit, while the "+1" level represents the superior limit of each variable. Tables 3 and ⁵ present the productivities of each grouped fraction obtained from SCFE of a Brazilian mineral coal using ethanol and isopropanol, respectively. The productivity has the dimension of gram of grouped fraction extracted per gram of coal in daf basis. In these tables T, P and Cos mean the normalized values of temperature, pressure and cosolvent concentration.

TABLE 3.
Productivity of each grouped fraction in funciton of the normalized variables, using ethanol as primary solvent.

			h
T	P	Cos	F1-F5	F6	F7-F8
-1	-1	-1	0.00371	0.01664	0.00677
-1	-1	1	0.00326	0.01297	0.00597
-1	1	-1	0.00424	0.03560	0.01667
-1	1	1	0.00223	0.03036	0.01395
0	0	0	0.00321	0.02446	0.00722
0	0	0	0.00332	0.02531	0.00747
0	0	0	0.00327	0.02485	0.00738
1	-1	-1	0.00577	0.02677	0.00520
1	-1	1	0.00586	0.02394	0.00681
1	1	-1	0.00716	0.05938	0.01866
1	1	1	0.00648	0.05675	0.01388

T(K) 598 648 698 P (MPa) 5.5 9.5 7.5 5.5 9.5 Cos(mol%) 0 10 0 10 5 0 10 0 10 Yield 1.89

1.72

4.31

4.07

4.15

4.05

3.91

4.81

4.49

7.32

6.46

F1

2.5

2.9

1.3

0.9

1.9

2.0

1.9

2.9

2.9

2.6

1.9

F2

0.8

3.1

0.5

0.5

0.4

0.3

0.4

1.3

2.0

1.1

2.1

F3

1.5

1.5

1.1

0.7

1.1

1.1

1.1

2.7

2.0

0.9

0.9

F4

5.2

3.6

3.1

1.5

3.7

3.7

3.7

6.2

4.6

3.5

3.5

F5

8.8

8.4

4.9

4.8

7.1

7.1

7.1

9.0

9.6

8.3

5.9

F6

54.0

61.6

71.6

65.4

63.7

63.7

63.6

65.4

59.0

63.4

69.9

F7

7.3

8.0

11.2

9.9

7.7

7.6

7.7

4.9

4.5

13.6

5.5

F8

19.9

10.9

6.3

16.3

14.5

14.5

14.5

7.6

15.4

6.6

10.3

F1-F5

18.8

19.5

10.9

8.4

14.1

14.2

14.2

22.1

21.1

16.4

14.3

F6

54.0

61.6

71.6

65.4

63.7

63.7

63.6

65.4

59.0

63.4

69.9

F7-F8

27.2

18.9

17.5

26.2

22.2

22.1

22.2

12.5

19.9

20.2

15.8

A statistical modeling technique was used in order to build empirical models able to reproduce the experimental data. Empirical models were built by assuming that all variable interactions were significant, estimating the parameters related to each variable interaction and main variable effects. The t-test was used to discard the meaningless parameters considering a confidence level of 95%. The parameters were estimated by minimizing a least square objective function based on the experimental and calculated values of h, using a commercial software [7].

T

P

Cos

F1-F5

F6

F7-F8

-1

-1

-1

0.00509

0.01463

0.00737

-1

-1

1

0.00335

0.01060

0.00325

-1

1

-1

0.00470

0.03086

0.00754

-1

1

1

0.00342

0.02621

0.01066

0

0

0

0.00585

0.02644

0.00921

0

0

0

0.00575

0.02580

0.00895

0

0

0

0.00555

0.02487

0.00868

1

-1

-1

0.01063

0.03146

0.00601

1

-1

1

0.00947

0.02649

0.00894

1

1

-1

0.01200

0.04641

0.01479

1

1

1

0.00924

0.04516

0.01021

Tables 6 to 8 present the empirical models able to reproduce the productivity of each grouped fraction, the fitted parameters (a_i) (significant) and their standard deviations. From these tables it can be observed that the correlation coefficient is greater than 0.98 for all cases, indicating that the models are suitable to reproduce the experimental data. In these tables are also shown the experimental variance s² and the variance of the estimation procedure (s²), defined as:

(3)

where Nexp is the number of experimental points, Npar is the number of parameter of the model and the superscripts exp and mod mean experimental and calculated, respectively. The value of P(F) (percent within the F-distribution) presented in Tables 6 to 8 were calculated using a aproximation given by Abramovitz and Stegun (1965). The inclusion of the quadratic terms in some models were significant and have improved the model performance. The factor that has shown the most pronounced effect was chosen to be the quadratic term.

Observing the empirical models shown in Table 6 one can conclude that the temperature is the most important variable to describe the light fraction productivity variations, i.e., it has the most significant effect among all variables. As we move towards the heavier compounds, the importance of pressure increases. For the intermediate compounds, both temperature and pressure can be considered to be equally important, and a combination of these two variables is relevant to describe the productivity for both primary solvents (Table 7). For the heaviest fraction, pressure may be regarded as the most important variable (Table 8). These facts just corroborate our previous discussion concerning the experimental results and thus, the empirical models seem to be suitable to represent the experimental data, providing a correct interpretation of the effects of the process variables on the productivities of the grouped extract fractions.

TABLE 6.
Regression results for the F1-F5 grouped fraction .

Ethanol Isopropanol
h = a0+a1T+a2T2+a3Cos+a4TP+a5TCos+a6PCos h = a0+a1T+a2T2+a3Cos
Correlation: 0.9930 Correlation: 0.9932
s 2 = 3.047e-9 s2 = 6.535e-7 s 2 = 2.320e-8 s2 = 9.737e-7
P(F) = 0.979 P(F) = 0.913
Parameter		Standard Deviation	Parameter		Standard Deviation
a0	0.00327	0.00017	a0	0.00572	0.00024
a1	0.00148	0.00011	a1	0.00311	0.00015
a2	0.00157	0.00020	a2	0.00152	0.00028
a3	-0.00038	0.00011	a3	-0.00088	0.00015
a4	0.00032	0.00011
a5	0.00023	0.00011
a6	-0.00029	0.00011

TABLE 7.
Regression results for the F6 fraction.

Ethanol Isopropanol
h = a0+a1T+a2P+a3P2+a4TP h = a0+a1T+a2P+a3Cos+a4T2
Correlation: 0.9936 Correlation: 0.9974
s 2 = 1.773e-7 s2 = 3.867e-5 s 2 = 6.218e-7 s2 = 7.565e-6
P(F) = 0.975 P(F) = 0.791
Parameter		Standard Deviation	Parameter		Standard Deviation
a0	0.02487	0.00125	a0	0.02570	0.00057
a1	0.00891	0.00077	a1	0.00840	0.00035
a2	0.01272	0.00077	a2	0.00818	0.00035
a3	0.00793	0.00147	a3	-0.00186	0.00035
a4	0.00363	0.00077	a4	0.00328	0.00067

TABLE 8.
Regression results for the F7-F8 grouped fraction

Ethanol Isopropanol
h = a0+a1P+a2P2+a3PCos h = a0+a1T+a2P+a3Cos+a4TP+a5TPCos
Correlation: 0.9799 Correlation: 0.9972
s 2 = 1.585e-8 s2 = 2.026e-5 s 2 = 7.097e-8 s2 = 5.673e-7
P(F) = 0.988 P(F) = 0.746
Parameter		Standard Deviation	Parameter		Standard Deviation
a0	0.00736	0.00066	a0	0.00869	0.00009
a1	0.00481	0.00041	a1	0.00139	0.00011
a2	0.00363	0.00078	a2	0.00220	0.00011
a3	-0.00105	0.00041	a3	-0.00033	0.00011
			a4	0.00031	0.00011
			a5	-0.00180	0.00010

It is also worth noticing additional aspects. First, when ethanol is considered to be the primary solvent, interaction effects among the process variables studied are always present. When isopropanol is the primary solvent, interaction effects occur only for the heaviest fraction. Second, the addition of water presents a negative effect whenever its effect may be regarded as significant.

After identifying the empirical models, process variables are manipulated in order to allow the optimization of certain criteria. Four specific process criteria are analyzed: 1-maximizing light fraction productivity, 2-maximizing light and intermediate fractions productivity, 3-minimizing heavy fraction productivity and 4-maximizing light and intermediate fractions and minimizing the heavy fraction productivity simultaneously. Table 9 shows the results obtained.

In this work the purity of the extracts was defined in terms of the content of the light and intermediate compounds. From Table 9, one can observe that maximizing only the light fraction and the light and intermediate fractions productivities simultaneously lead to the same results. In these cases, the purity of the extract was low because of the high content of heavy compounds. When the purity of the extracts was taken into account (criteria 3 and 4), the productivities of the light and intermediate fractions were smaller.

TABLE 9.
Optimization results.

Solvent: Ethanol		Solvent: Isopropanol
Criterion: 1 Maximization of light fraction productivity
T:1.00 P:1.00 Cos:-1.00		T:1.00 P:-1.00 Cos:-1.00
Grouped fraction	Productivity (h)	Grouped fraction	Productivity (h)
F1-F5	0.007065	F1-F5	0.012213
F6	0.058060	F6	0.031182
F7-F8	0.016828	F7-F8	0.006216
Criterion: 2 Maximization of light and intermediate fractions productivity
T:1.00 P:1.00 Cos:-1.00		T:1.00 P:1.00 Cos:-1.00
Grouped fraction	Productivity (h)	Grouped fraction	Productivity (h)
F1-F5	0.007065	F1-F5	0.012213
F6	0.058060	F6	0.047379
F7-F8	0.016828	F7-F8	0.014676
Criterion: 3 Minimization of heavy fraction productivity
T:-1.00 P:-0.80 Cos:-1.00		T:-1.00 P:-1.00 Cos:1.00
Grouped fraction	Productivity (h)	Grouped fraction	Productivity (h)
F1-F5	0.003945	F1-F5	0.003253
F6	0.013884	F6	0.010738
F7-F8	0.005011	F7-F8	0.003410
Criterion: 4 Maximization of light and intermediate fraction productivity and minimization of heavy productivity
T:1.00 P:-0.70 Cos:-1.00		T:1.00 P:-1.00 Cos:-1.00
Grouped fraction	Productivity (h)	Grouped fraction	Productivity (h)
F1-F5	0.006047	F1-F5	0.012213
F6	0.026216	F6	0.031182
F7-F8	0.005051	F7-F8	0.006216

Table 9 also shows that the results obtained using ethanol or isopropanol as primary solvents are similar. In both cases it was not possible to maximize productivity and purity of the extracts manipulating only the mentioned process variables in the ranges studied. From these tables one can observe that small gains in extract purity lead to large losses of light and intermediate fraction productivities indicating that liquid yield and purity of the extracts follow opposite trends. Therefore, process operating conditions must be designed on an economical basis, depending on the market values of the fractions and on the separation costs induced by the presence of heavy fractions in the extracts.

4 — CONCLUSIONS

Empirical models were built to represent experimental data of productivity of the extracts obtained from SCFE of a high-ash Brazilian mineral coal. These models were used to find the experimental conditions that would maximize the productivity and purity of the extracts simultaneously. The results obtained, however, show that it is not possible to reach this goal by manipulating the process variables used here in the ranges studied. In this sense, an economic criterion should also be considered in order to optimize the process considering the value (in terms of purity and productivity) of each fraction.

5 — REFERENCES

[1] ABRAMOVITZ, M. and STEGUN, I.A., Handbook of Mathematical Functions, Dover Publications Inc., 1965.

[2] BARNAKOV, C.N., VOLGIN, A.A. and AKSYENOVA, O.P., Fuel, 70, p.1005, 1991.

[3] DARIVA, C., OLIVEIRA, J.V., VALE, M.G.R. and CARAMÃO, E.B., Fuel, 76, p. 585, 1997.

[4] KARAM, H.S., MCNAIR, H.M. and LANÇAS, F.M., Lc-Gc, 5, p.41, 1987.

[5] LANÇAS, F.M., VILEGAS, J.H.Y. and MARTINS, S., J. High Resolut. Chromatogr., 17, p.237, 1994.

[6] ROCHA, S.R.P, Oliveira, J.V., d’Ávila, S.G., Pereira, D.M. and Lanças, F.M., Fuel, 76, p.93, 1997.

[7] STATSOFT^TM, Copyright^© Statsoft, 1991.

[8] SUNOL, K.A. and Beyer, G.H., Ind. Eng. Chem. Res., 29, p.842, 1990.

[9] WILLIAMS, D., Chem, Engng. Science, 36, p.1769, 1981.

6 — ACKNOWLEDGMENTS

This work was supported by FUNCITEC (Santa Catarina State - Brazil), grant no. 004/94. The authors would like to thank CNPq (Conselho Nacional de Desenvolvimento Científico e Tecnológico - Brazil) and FAPERGS (Rio Grande do Sul State - Brazil) for financial support.

Programa de Engenharia Química / COPPE / Universidade Federal do Rio de Janeiro — Cidade Universitária — Cx. Postal: 68502 — Rio de Janeiro —- 21945-970 —- Brazil

³

Instituto de Química / UFGRS - Porto Alegre-RS Brazil

* To whom correspondence should be addressed.

TABLE 4. The effect of temperature, pressure and addition of water on the distribution and on the productivities of the fractions of the extracts from SCFE with isopropanol of Butiá-Brazil coal

TABLE 5. Productivity of each grouped fraction in funciton of the normalized variables, using isopropanol as primary solvent.

¹ Received fo publication in 5/8/97. Acepted for publication in 12/1297.

²

Publication Dates

• Publication in this collection
  17 Dec 1998
• Date of issue
  Dec 1997

History

• Accepted
  12 Dec 1997
• Received
  05 Aug 1997