136
Energy & Fuels 1989, 3, 136-143
bottoms derived from conventional crude and heavy oils were used to examine the effect of bridging liquids. It was shown that both types of vacuum bottoms are suitable as a bridging liquid. A processing time of 7-10 min was sufficient to achieve greater than 90% combustibles re-
covery and more than 55% mineral-matter rejection. The present process is ideally suited for mineral-matter reduction of coprocessing feeds because it can use a part of the coprocessing feed oil for agglomeration before preparation of the slurry feed to the reactor.
Kinetics of Heavy Oil/Coal Coprocessingt A. J. Szladow* and R. K. Chan Lobbe Technologies Ltd., Regina, Saskatchewan, Canada S4P 3L7
S . Fouda and J. F. Kelly C A N M E T , Energy, Mines and Resources, Canada, Ottawa, Ontario, Canada K I A OGl Received August 2, 1988. Revised Manuscript Received December 19, 1988
Reaction engineering models were developed for coprocessing of heavy oil and coal. The properties of the coprocessing products, lumped by their solubility and distillation points, were found to be independent of the severity of the process. The effects of temperature and space velocity on coprocessing showed the product yield structure to be determined primarily by temperature. Two types of reaction networks, a sequential network and a parallel network, were tested for correlating coprocessing data. The criteria used for evaluation of the developed model were the percent of variance explained by the model and the accuracy of the model. The final, selected model explained 90% of the target, 92.5% variance. Testing of the developed models to the assumptions made showed that the choice of model is determined primarily by the yield structure of the products and not by the assumptions made about the gas holdup or the amount of product flashed a t the reactor temperature and pressure.
Introduction A number of studies have been reported on coprocessing of coal and oil sand bitumen, petroleum residues, and distillate fractions in catalytic and noncatalytic processe ~ . l - The ~ above studies describe the effects of process chemistry on feedstock characteristics and operating variables; however, very few kinetic data were reported on the processes being currently developed or licensed. This paper presents the results of modeling the kinetics of coprocessing for the CANMET process. Since 1979, CANMET has been conducting research and process development work on coprocessing of Canadian heavy oillbitumen and coal. The program included a factorial design of coprocessing experiments to study the kinetics and mechanisms of coprocessing. As a continuation of that program, in 1986, CANMET and Lobbe Technologies undertook a project on mathematical modeling of coprocessing kinetics with emphasis on the development of reaction engineering models for improved process performance and operation. The results of that project were reported in much detail elsewhere.6 This state of the art review of coal liquefaction and heavy oil upgrading kineticss concluded that both liquefaction and heavy oil upgrading have a number of common features and that principles of reaction engineering which apply in modeling coal liquefaction also apply in modeling kit Presented a t the Symposium on Coal-Derived FuelsCoprocessing, 195th National Meeting of the American Chemical Society and 3rd Chemical Congress of North America, Toronto, Ontario, Canada, June 5-10, 1988.
0887-0624/89/2503-Ol36$01.50/0
Table I. Properties of the Sample Coal Proximate Analysis (As Received) % moisture 19.11 % ash 7.67 % volatile matter 35.57 % fixed carbon 37.65 % sulfur 0.41 calorific value, MJ/kg 19.88 Ultimate Analysis (Daf) % carbon % hydrogen
% nitrogen % sulfur % oxygen (by difference)
71.65 4.38 1.82 0.56 21.59
netics of heavy oil upgrading and vice versa. The review concluded that the choice of evaluation methodology reflects, in general, the type of products obtained or the severity of the process rather than fundamental differences (1) Moschopedis, S.; Ozum, B. Liq. Fuels Technol. 1984,2, 177, 193. (2) Ignasiak, B.; Lewkowicz, L.; Kovacik, G.; Ohuchi, T.; Du Pleassis,
M. P. Catalysis on the Energy Scene; Elsevier: Amsterdam, 19U, p 595. (3) Kelly, J. F.; Fouda, S. A.; Rachimi, P. M.; Ikura, M. Paper presented at the CANMET Coal Conversion Contractors Meeting, Calgary, Canada, November, 1984. (4) MacAurthur, J.; Boehm, F.; Liron, A.; Shannon, R. H. Presented at loth Annual EPRI Contractors' Conference on Clean Liquid and Solid Fuel, Palo Alto, CA, April 23, 1985. ( 5 ) Curtis, C.; Guin, J. A.; Pass,M.; Tsai, K. J. Presented at the 188th National Meeting of the American Chemical Society, Philadelphia, PA, Aug 26, 1984. (6) Szladow, A. J.; Chan, R. K.; Tscheng, J.; Fouda, S. A.; Kelly, J. F. "Mathematical Modelling of Coprocessing Kinetics"; Final Report, Sept 9, 1987. 0 1989 American Chemical Society
PRODUCT GAS
/
-
GAS CHROMATOGRAPHY
H z CONSUMPTION
-
Hz S
cc
+
COC
C1 - c 4 W4TER
I
OILS
PREASPHALTENES THFI
S e n t o n e sohdble
thf
so1
toluene thf
nscl
#nscI
Figure 1. Product definition.
in process chemistry or dynamics. A number of methods that have been used successfully in coal liquefaction or heavy oil upgrading may be applied to coprocessing.
Experimental Section Equipment. The experiments were conducted in a continuous coprocessing hydroliquefaction unit. A detailed description of this unit was given e l s e ~ h e r e .The ~ unit included a stirred-tank reactor that had a n internal diameter of 3 in. and a length of 9 in. The reador had one adjustable impeller and one fmed impeller, one thermowell, and a dip tube mounted in the bottom closure for removal of coprocessing products. The total reactor volume was 1079.58 mL with a gas-phase volume of 786.62 mL and a liquid phase volume of 292.46 mL. The reactor hydrodynamics of gas holdup were investigated in exploratory coproceasing runs. The tests suggested that impeller speed above 500 rpm provided back-mixing equivalent to a perfect stirred-tank reactor. On this basis the reactor equations developed were those for CSTR. A literature search showed no information on gas holdup for processing of heavy oil in stirred tanks. On the basis of general behavior of stirred-tank reactors, it was concluded that gas holdup should be very small unless high impeller speeds would result in a vortex formation around the impeller.' Feedstocks and Experimental Conditions. Forestburg, Alberta, Canada, subbituminous C coal and Cold Lake +454 "C cut vacuum bottoms were used in the coprocessing experiments (Tables I and 11). The coal was ground to -200 mesh and stored under dry ice (COP) atmosphere. Feed slurries were prepared by mixing about 30% of coal and 70% of heavy oil (daf slurry basis), under atmospheric pressure in a stirred feed tank. A dkposable iron sulfide catalyst was added at the amount of approximately 5% on a weight by weight daf slurry basis. A description of the catalyst preparation method was given elsewhere.s The coprocessing runs were conducted by using factorial design (Latin square d e ~ i g n at ) ~three levels of temperature and space velocity. The range of process conditions included the following: temperature, 400-455 "C; nominal slurry space velocity, 0.50-1.54 kg/L/h; reactor pressure, 2000 psig; run duration, 80-180 min; total feed processed, 402-600 g; coal concentration, 29.71-33.38 maf wt % feed. The coprocessing product gas was analyzed for H2S, CO and COP, and C1-CI hydrocarbons. The coprocessing product slurry was analyzed for water, distillate hydrocarbon fractious (naphtha, light gas oil, heavy gas oil), and residue (+525 "C). The residue was analyzed for oils, asphaltenes, preasphaltenes, and tetrahydrofuran-insoluble fractions (THFI) based on solubility in pentane, toluene, and tetrahydrofuran. The distillate products were also analyzed for elemental composition and specific gravity. The residue was analyzed for elemental composition and ash (7) Kang, D.; Ying, D. H. S.; Givens, E. N. "Impact of Hydrodynamics on Liquefaction"; Final Report, DOE/OR/03054-63, 1983. (8) Fouda, S. A.; Kelly, J. F. Coprocessing of High-Volatile Bituminous Coals. In Proceedings of the 1987 International Conference on Coal Science; Elsevier: Amsterdam, 1987. (9) Neter, J.; Wasserman, W.; Kutner, M. H. Applied Linear Statistical Models; Richard D. Irwin, Inc.: Homeweed, IL, 1985.
Table 11. Properties of Vacuum Bottoms specific gravity a t 15/15 "C, 1.038 pentane insolubles, wt % 23.8 pitch content (+525 "C), wt % 83.0 Conradson carbon, wt % 17.1 ultimate analysis, % C 83.07 H 9.85 N 0.55 S 5.50 0 1.80 H/C atomic ratio 1.42 metals, ppm V 235 Ni 93 Fe 18 aromaticity 33 viscosity at 80 "C, P 249 content. A summary of the product definitions and separations procedure is depicted in Figure 1. The percent recovery of the input feed was between 96.40 and 105.02% with a mean of 100.78% and a standard deviation of 2.84%. This accurate mass balance for the products was essential in obtaining high correlation and predictive capabilities of the developed models.
Results A conventional way of depicting the effect of process variables of product yield such as a plot of the yield vs space time, a t fixed temperatures, is too difficult to interpret for a complex reacting mixture like coprocessing. It was concluded, therefore, that a multiple classification analysisg may be more helpful in the development of an understanding of the overall effects of temperature and space velocities on the yield and selectivity of coprocessing products. The multiple classification analysis indicates the deviation of the dependent variables from their mean for each factor (Figure 2). For example, Figure 2 depicts the effect of temperature and space velocity on the oil's yield. The effect of temperature is strong while the effect of space velocity is weak, suggesting that there is at least one formation and one disappearance path for the oil's group. In another example (Figure 2), the yield of asphaltenes first increases with the decreasing space velocity, then goes through a maximum, and finally decreases at longer space times, suggesting that asphaltenes may form an intermediate product. Figure 2 depicts another interesting example for generation of the CI-C4 group. The observed delay in generation of the C1-C4 group until longer space times (smaller space velocity) suggests that it may be a terminating group, thus requiring first a buildup of the intermediate products. In other examples, yields of CO, and heavy gas oil 2 and hydrogen consumption do not change with temperature, suggesting that the formation and disappearance paths
Szladow et al.
138 Energy & Fuels, Vol. 3, No. 2, 1989
425
S w c e Velocity
Spoce Velocity
Tenperoture
4M
05
450
I0
THFl
PASP
-100
'
-100
+IO0
'
*IO0
-100
r
ti00
'
-100
1.5
IO
05
1.5
1 i
-100
,
OILS
-100
'
-100
i
-100
t
+IO0 r ASP
-100
'
425
400
450
-100
L
t
cox
-100
HGO! +loo
-100
t
Spoce V e l o n t y
Temperature
renperature
400
-100
-100
-100
-100
-100
LGO
-100
425
OI_jjji
-100
05
450
-100
IO
15
I
-100 O~
i -100
i
-100
-100
Figure 2. Effects of temperature and space velocity on coprocessirig products (approximate percentage of the mean).
may have small activation energies. The above discussion shows how a multiple classification analysis can assist in formulating reaction paths for coprocessing. It must be remembered, however, that the depicted diagrams (Figure 3) represent complex reaction networks and that a simple interpretation, as above, may indicate only some of many possible networks. Nonetheless, the method is, very helpful, and the suggested networks in Figure 3 are further supported by the results of the analysis of variance for coprocessing products (Table 111).
The ANOVA tableg (Table 111) shows that almost all coprocessing products are strongly affected by temperature, with the exception of preasphaltene, carbon oxides, and heavy gas oil 1 yields and hydrogen consumption, which exhibit strong space velocity effects and also large effects for two-way interactions. It was also surprising to see a weak effect of temperature on hydrogen consumption and a weak correlation between the distillate yield and hydrogen uptake. The calculated correlation coefficients between the hydrogen consumed and C1-CI and distillate yields were 0.406 and 0.398. The correlation coefficients
Heavy OillCoal Coprocessing
Energy & Fuels, Vol. 3, No. 2, 1989 139 THFI
T H F I -A
OIL-
) THFI-
D I S T +C l - C 4
-
-
-
-P A -
-
I
OIL-
F H T
A>
-LIO
DIST
\ THFI-
t - - - - - - L IO
PA
\li -Cl-C4
OIL-DIST
PA
1:=--+
DIST
THFI-
PA
OIL
HG02
c c NAPHT
I
ClkC4
PA
LGO + HGOl
Cl-C4
THFI
___t
OIL
LGO + HGOl
THFI
HG02
OIL -HG02
LGO + HGOl
NAPHT-Cl-C4
-
PA
1
\ NAPHT
/I
DIST
__t
\
PA
/
\
O ) l
+
HGOl
Figure 3. Example of selected kinetic structures for coprocessing. Table 111. Summary of Anova Analysis % of variance explained by variable name temp SV two-way 1 pitch 86 13 coal 5 66 29 oils 1 97 2 asphaltenes 10 64 26 preasphaltenes 28 33 39 THFI 75 25 hydrogen consumption 2 96 2 hydrogen sulfide 69 19 12 13 81 6 carbon oxides 86 4 water 10 1 85 14 total distillates heavy gas oil 1 6 33 61 heavy gas oil 2 82 15 3 1 light gas oil 94 5 12 3 85 naphtha 78 16 6 total C&
became larger in the temperature range 425-450 O C (0.690 and 0.7451, but still were not very strong. The strong temperature effect on the yield of coprocessing products clearly demonstrates that the yield of most products is primarily determined by thermal cracking reactions. This explains why a number of studies reported that at high hydrogen pressures and a temperature range of 400-450 "C, coal conversions are simiiar under coprocessing and coal liquefaction conditions. The principal rate-determining factor, under the above conditions, is not the availability or consumption of hydrogen but the extent of cracking reactions which are determined by the tem-
perature. The role of hydrogen is limited to stabilizing generated smaller molecular fragments and to preventing coke formation. It should be noted the above mechanistic interpretation may not be true at lower hydrogen pressures or at temperatures above 450 "C, where coke formation may be important.
Kinetic Models Background. A number,of studies had been published documenting the pitfalls of using lumped kinetic models for reacting mixtures, in general'OJ' and for coal liquefaction in particular.12J3 The most common difficulties identified were the loss of information about the kinetics of individual reactions, different rate expressions for the grouped species as opposed to individual species rate equations, little theoretical significance underlying the overall (lumped) activation energies, and frequent discrepancy between the order of the rate expressions for lumps and for reacting species. Several investigators showed how these difficulties may be overcome for complex reacting mixture~.'*-'~ These methods were later extended to coal liq~efaction.'~J~ (10) Liu, Y. A.; Lapidus, L. AIChE J. 1973,19, 467. (11) Luss, D.; Hutchinson, Y. Chem. Eng. J. 1971, 2, 172. (12) Szladow, A.; Given, P. 2nd. Eng. Chem. Process. Des. Deu. 1981,
20,27. (13) Prasad, G. N.; Witmann, C. V.; Agnew, J. B.; Sridhar, T. AIChE
J. 1986, 32, 1277, 1288.
(14) Golikeri, S. V.; Luss, D. Chem. Eng. Sci. 1974, 29, 845. (15) Lee, H. H. AIChE J. 1978,24, 116. (16) Bailey, J. E. Chem. Eng. J. 1972, 3, 52.
140 Energy & Fuels, Vol. 3, No. 2, 1989
Table IV. Analysis of Variance (H/C vs T and SV) signifisum of mean cance of F F source of variation squares DF square 4 0.002 2.000 3.637 main effects 0.007 2 0.004 temp 0.007 sv 0.000 2 0.000 4 0.000 two-way 0.000 interactions 0.000 4 0.000 temp and SV explained 0.008 8 0.001 0 0.000 0.000 residual 0.008 8 0.001 total n :
Figure 4. Plot of In t , vs 1/T for THFI fraction.
In our approach to modeling coprocessing kinetics, we first reviewed CANMET's lumping criteria for the coprocessing products. The results of ANOVA analysis for carbon to hydrogen ratio of the coprocessing products indicated that the product properties (H/C ratio, etc.) and, hence, product definitions are independent of the severity of the reactions for the conditions studied (Table IV). Second, the reaction paths for conversion of THFI were evaluated by using methods described by Szladow et a1.I2 In the method, an apparent activation energy is derived from the conversion-time-temperature data directly without any assumptions for the rate equations or kinetic structure. The derived apparent activation energy is a weighted average of activation energies of individual reacting species with the weighing coefficients equivalent to rates of the individual species. Figure 4 shows a plot of In t , (t, = time required to obtain conversion x = a reciprocal of temperature) vs 1/T for two conversion levels ( X = 0.5 and X = 0.7) of THFI materials. Since the slope of E'does not change significantly with conversion, THFI may be described by one initial lump only. Model Equations and Their Solution. Generalized lumped kinetic model equations were derived for an ncomponent reacting mixture in a stirred-tank reactor.
where = rate of total feed input in kg/h (daf), 4 = mass pp = average fraction of product flashed a t reactor T,P, density of products in kg/L, VR = reactor volume (to the Xi = dip tube) in L3, t = gas holdup, Xi,, = Fiin/Pn, FioUt/Pn, r{ = rate of formation (disappearance) for comin h-l. ponent 3" in terms of fractional conversion, Xi, Equation 1 and the least-squares method were used to regress for the model rate constants. The developed computer algorithms first solved the set of normal equa(17) Szladow, A,;Given, P. Chem. Eng. Commun. 1982,19, 115.
Szladow et al. tions for the rate parameters, kij, and later solved for the rate constant coefficient Aij and Eij. This approach provided information on how the calculated rate constants, k , , change with the alternative reaction networks. The approach also permitted better understanding of the physical meaning underlying the regressed rate constants and the postulated product yield structures for coprocessing. Testing the Models. Two different types of models were formulated for predicting the yield structure of coprocessing products: (1) models that had a sequential characteristic; (2) parallel models with coprocessing product generated from initial lumps. There is a fundamental difference between the two types of models in terms of process optimization. For sequential models, the implication is that distillate products reach a maximum beyond which addition of more hydrogen results in generation of C1-CI hydrocarbon gases. In parallel models, the yield of distillate products continuously increaseswithin the range of coprocessing conditions-with the selectivity of products determined by their relative rate of formation. Three criteria were used in selection of a suitable parallel or sequential model: 1. How much of the total variance was explained by the model. 2. How much of the grouped product's variance was explained by the model. This criterion recognized that in our objectives we may be more interested in how well the model predicts specific product fractions, for example, naphtha yield instead of the overall structure of coprocessing products. 3. How well the model predicts the yields of coprocessing products or how accurate is the model. In the approach followed, criterion 1was first; later, as specific difficulties were addressed, criteria 2 and 3 were reviewed. This approach allowed a step-by-step assessment of the models and a detailed interpretation of the rate constants and reaction paths. Over 20 different reaction networks were tested for correlating the yield structure of coprocessing data. Most of these models were not able to accurately predict the CANMET data or to meet the physical constraints on rate constants coefficients,Le., to yield positive preexponential factors and activation energies. Special difficulties were also experienced with modeling preasphaltenes. Figure 5 depicts the effect of temperature levels on the yield of preasphaltenes. As the temperature increased from 400 to 425 "C, the yield increased; however, as the temperature further increased from 425 to 450 "C, the yield decreased, indicating a change in the sign of the temperature coefficient (activation energy). This led us to believe that two mechanisms are responsible for this apparent anomaly, one being essentially a chemical reaction mechanism with the rates increasing with temperature and the second mechanism representing an adduct formation for which the rate decreases with temperature. Such an adduct formation may be possible between preasphaltenes and heavy oil. Cronauer et a1.18 and Satolgreported similar effects for coal liquefaction. They observed a net negative yield of oil a t low space time, suggesting an adduct formation between the solvent and primary liquefaction products. Our analysis of the CANMET experiment revealed that other coals also exhibit similar "reverse" temperature effects (18) Cronauer, D.C.; Ruberto, R. G. EPRI Report AF913, 1979. (19) Sato, S.Prepr. Pup.-Am. Chem. Soc., Diu. Fuel Chem. 1976,24, 270.
Energy &Fuels, Vol. 3, No. 2, 1989 141
Heavy OillCoal Coprocessing
Table V. Rate Constant Coefficients for Models 52, 61, and 53 model 61 model 53 model 52 rate A: E: constant A; E; A; E; 1.24310" 8.31323 1.84312 1.72312 4.97317 4.06319 7.90306 2.28314
1 2 3 4 5 6 7 8 9 10
33020 76180 39030 40000 58910 63 400 23 510 47470
5.33313 5.23303 8.33308 9.48318 5.56309 7.90306 6.53302 1.50306
4404012030 29390 61030 31 990 23 310 13 860 55000
1.24310 9.11323 1.84312 1.51312 4.17317 4.06319 7.90306 2.28314 3.51319 4.52316
33020 77730 39030 40000 59610 63 400 23510 47 470 66400 59900
"In this notation 1.24310 = 1.24 X 1O'O etc. i5C
425
400
T E M P E R A T IJR E
-;----
Figure 5. Effect of temperature on preasphaltenes yield.
k81 x OIL
.
LGO
, ,
"/
/
/ C1-C4
PA
ThF1 A
7 PA A
Table VI. Percent of Variance Explained by Models 52, 61, and 53 a(i) F(i)" model 52 model 61 model 53 15.5 92.1 THFI 91.5 90.4 92.8 PA 53.5 91.7 93.1 96.3