A solvent-resid phase diagram for tracking resid conversion

This solvent-resid phase diagram enables one to track the path of chemical changes that do not convert a molecule completely from one solubility class...
0 downloads 0 Views 972KB Size
I n d . E n g . C h e m . R e s . 1992, 31, 530-536

530

PROCESS ENGINEERING AND DESIGN A Solvent-Resid Phase Diagram For Tracking Resid Conversion I r w i n A. Wiehe Corporate Research Laboratories, Exxon Research and Engineering Company, Route 22 East, Clinton Township, Annandale, New Jersey 08801 -0998

A plot of molecular weight versus hydrogen content is capable of representing solvent-separated pseudocomponents of petroleum resids and their thermal conversion products by unique areas that are independent of the resid. This solvent-resid phase diagram enables one to track the path of chemical changes that do not convert a molecule completely from one solubility class to others. In this study petroleum resids and the thermal conversion products of the resids and of resid fractions were separated by solubility and clay adsorption into five pseudocomponents: saturates, aromatics, resins, asphaltenes, and coke. The combination of this solvent fractionation and the solvent-resid phase diagram shows how the solubility classes of decreasing solubility evolve during thermal conversion either by increasing aromaticity or by molecular weight growth of low-volatility fragments. Since petroleum resids are complex mixtures of thousands of compounds of low volatility, it is difficult to deduce the chemical changes that occur during their processing. A common approach has been to use solvents to separate petroleum resids and their reaction products into pseudocomponents and then track the pathway for chemical changes by the conversion of each pseudocomponent into others. One of the most common of such pseudocomponents is the asphaltenes that are soluble in aromatic solvents (benzene or toluene) and insoluble in paraffinic solvents (n-pentane or n-heptane). However, Bunger and Cogswell (1981) have raised valid concerns about this approach because no single chemical feature distinguishes species found in the asphaltene fraction from those in the mdtenes (paraffinic solvent soluble) fraction and because pseudocomponents can undergo chemical changes without being completely converted to other pseudocomponents. Here, it is shown that a molecular weight-hydrogen content “phase diagram” or heavy oil “map” in combination with pseudocomponent separation overcomes both of these concerns. Not only is it capable of distinguishing one pseudocomponent from another, but it can be used to track chemical changes that do not move a molecule completely from one solubility class to others. Separation Scheme In this section, the procedure is described for separating petroleum resids and their reaction products that was used in this study. Unfortunately, such separations are sensitive to changes in experimental procedures (Speight et al., 1984). The definition of “soluble” or “insoluble” is dependent on the amount of solvent, the pore size of the filter, the filter material, and the time (when less than 4 h) between when the solvent and resid are mixed and when the mixture is filtered. However, if the procedure is kept constant, the yield of a fraction from a given resid and its chemical and physical properties can be quite reproducible. A schematic of the separation procedure is shown in Figure 1. Often vacuum distillation was used to remove volatile products from resid conversion products as a first step. When this was not done, light volatile products 0888-5885/92/2631-0530$03.00/0

evaporated during solvent removal and were determined by mass loss. In the second step the resid nonvolatile product was mixed in a flask with 15 parts toluene and let to sit for at least 16 h at room temperature. This mixture was poured through a fine (4-5.5-pm pores) fritted-glass filter. The solids on the filter were washed with at least 25 parts additional solvent, and washing was continued until the solvent passed the filter without color. The toluene-insoluble solids were vacuum dried on the filter at 100 OC for at least 16 h and are called “coke”, following the convention of Speight (19701, Schucker and Keweshan (19801, and Savage et al. (1985). The toluene was removed from the toluene solubles by rotary evaporation at 50 “C followed by vacuum drying for 16 h at 50 “C. Since the unreacted resids in this study all had prior vacuum distillation and were completely toluene soluble, the first two steps were only applied to resid thermal conversion products. The n-heptane separation in the third step followed the same procedure as for the toluene separation, except the n-heptane was not removed from the heptane solubles. The solid, heptane-insoluble product is called “asphaltenes”. In the fourth step the heptane solution is mixed in a flask with Attapulgus clay, 30 times the weight of heptane solubles, for a minimum of 16 h. The Attapulgus clay was National Bureau of Standards qualified for ASTM D2007-86, a clay-gel chromatographic separation test. This clay contains a controlled amount of water and meets azobenzene activity specifications of 30-35 equiv. This assured reproducible adsorption, while only with water on the clay can one completely desorb the hydrocarbons off the clay. This mixture of clay, heptane, and oil was filtered with a fine, fritted-glass filter. The clay on the filter was washed with at least 200 parts of heptane per part of solubles, and washing was continued until the heptane ran through the filter without color. The solvent was removed from the solution so produced by rotary evaporation and by vacuum drying at 50 OC. The fraction remaining on the clay was desorbed in the fifth step by mixing with a mixture of 50% toluene and 50% acetone on the filter. At least 100 parts of this 0 1992 American Chemical Society

Ind. Eng. Chem. Res., Vol. 31, No. 2, 1992 631 Table I. ExamDle Analysis of Fractions Separated from an Unreacted Resid yield/ fraction saturates aromatics resins asphaltenes total starting resid

wt%

wt%

H/C

wt%

(wt %)

C

H

18 17 25 100

84.54 81.87 82.08 81.93 82.45

12.31 10.00 9.50 7.94 9.70

(atomic) 1.73 1.46 1.38 1.15 1.40

S 2.74 5.56 6.09 7.50 5.75

100

82.95

9.92

1.42

5.73

40

wt%

wt%

ppm

ppm

total elements/

0“

V

NI

wt%

1.72 1.55 1.08

N 0.03 0.12 0.77 1.15 0.62

0 0 260 740

0 0 110

289

0.77

0.77

300

0.0 0.0

% wt%

230 102

99.6 97.6 100.2 100.2 99.6

VPOb MW 920 613 986 2980 1040

100

100.2

1183

24.0

MCR 1.8 8.7 24.6 49.6 24.2

Oxygen determined by neutron activation. VPO number average molecular weight measured in o-dichlorobenzene at 130 O C . residue technique for measuring Conradson carbon residue (ASTM Test D4530). dAromatic carbon determined by NMR. VOLATILES t

RESID

DISTILLATION

AROMATICS SOLUBLE

INSOLUBLE COKE

INSOLUBLE ASPHALTENES

f

f

TOLUENE FILTRATION

HEPTANE FILTRATION

-(TI !F{iz$$N d 1

4

ATTAPULGUS

t

RESINS

Figure 1. Resid separation scheme.

mixture per part heavy oil was used to wash the clay. This was followed by washing with at least the same amount of 90% toluene and 10% methanol, which was continued until the solvent mixture passed through the clay without color. The solvent was removed from the resulting solution by rotary evaporation followed by vaccum drying at 50 OC. The resulting fraction was reported as “resins”. In the sixth step, the heptane-soluble oil that was not adsorbed on clay was mixed with 30 parts methyl ethyl ketone (MEK)and cooled to -78 “C in a dry ice-isopropyl alcohol bath. After a 4 h waiting period, this cold mixture was filtered through a fine, fritted-glass filter that was kept at -78 “C by adding dry ice directly on the filter. The filtration was done under a flow of dry nitrogen to prevent water from the air condensing and adding to the MEK solution. The waxy solids remaining on the filter were dissolved off with room-temperature heptane, and the heptane was removed by rotary evaporation and vaccum drying at 50 “C. This fraction was called “saturates”. The solvent was removed from the solution that passed through the fiiter by rotary evaporation and vaccum drying at 50 “C, leaving the oil fraction that was called “aromatics”. Example Unreacted Resid Fractions A 1050 O F + resid was separated by the above separations scheme with yields and analyses of the fractions shown in Table I. %ically, the mass and elemental balances are good, except, because of air oxidation, the fractions total higher oxygen and lower hydrogen than the starting resid. This does not seem to affect the coke-forming tendency as measured by Microcarbon Residue (MCR) because the sum of the fractions equal the MCR of the starting resid, a result reported previously (Long and Speight, 1989b; Roberts, 1989). The terms saturates and aromatics have only relative meaning and should not be taken on an absolute basis. Thus,the saturates are more saturated than the other fractions but still contain aromatic carbon and sulfur. However, this fraction contains little or no oxygen,

arom C4 15 37 45 50 40 41 Microcarbon

nitrogen, vanadium, and nickel and exhibits little cokeforming tendency. The aromatics fraction (37% aromatic carbon) is more aromatic than the saturates fraction but also is relatively free of heteroatoms, other than sulfur. It is also the lowest molecular weight fraction. The resins fraction is sometimes called “polar aromatics” because of ita high oxygen and nitrogen content relative to the saturates and aromatics fractions. However, the average resin molecule has only one oxygen or one nitrogen atom out of a molecular weight of 986. Thus, it is not as polar as many common low molecular weight compounds that contain oxygen and nitrogen, such as acetone or pyridine. As will be discussed later, the present evidence indicates that resins adsorb on clay more because they contain large-ring aromatics than because of oxygen and nitrogen functionality. Nevertheless, there is also a sharp increase in vanadium, nickel, aromatic carbon, and MCR in going from aromatics to resins. On the other hand, the asphaltenes, as is well-known (Speight, 1980), have still higher concentration of all the undesirable components and properties: sulfur, nitrogen, vanadium, nickel, molecular weight, MCR, and aromatic carbon. I t is no wonder that problems in converting resids usually focus on asphaltenes. However, since the yield of resins is much higher than the yield of asphaltenes, as a percentage of resid the resins have comparable amounta of undesirable components and properties as the asphaltenes. One of the odd features of the separation of resids is that the fractions have nearly constant carbon content. For this separation all the carbon contents were within 83 f 1.5 wt %. The lower hydrogen content in the more aromatic fractions is compensated by the higher heteroatom content, keeping the carbon content nearly constant. As we shall see, this feature also carries over to resid thermal conversion products. Thus, the hydrogen to carbon atomic ratio that tends to measure aromaticity (lower value for more aromatic) is really equivalent to a constant times the hydrogen content for these fractions. Example Fractions of Thermal Conversion Products The vacuum resid with properties shown in Table I was reacted for 60 min at 400 “C in small, batch-tubing bomb reactors under 7 MPa of nitrogen. After the gases were vented, the resid product was separated according to the separation scheme. The yields and analysis of each fraction are shown in Table 11. The gases and the fraction that volatilized during solvent removal were not analyzed, and the molecular weight of the coke could not be measured because of insolubility in o-dichlorobenzene. Again the carbon content is relatively constant, with the carbon content of saturates actually being slightly higher than the carbon content of coke. The average molecular weight of each of the fractions (except coke) is decreased by thermal conversion. While the hydrogen contents of

532 Ind. Eng. Chem. Res., Vol. 31, No. 2, 1992 Table 11. Analvsis of Resid Reacted at 400 OC for 60 min fraction yield/(wt %) wt % C wt % H gases 1.5 light volatiles 12.3 saturates 12.9 85.25 12.53 aromatics 25.8 83.77 10.10 resins 20.5 81.72 8.27 asphaltenes 18.3 83.16 6.34 coke 8.7 82.19 5.54

H/C (atomic) 1.75 1.44 1.21 0.91 0.80

Table 111. Thermal Conversion of Resid Fractions (60 min at 400 "C) reactant product yield/(& %) wt % C wt % H 29.5 saturates + aromatics volatiles 32.1 85.59 12.50 saturates 38.4 10.59 83.95 aromatics 19.4 resins 83.06 9.01 10.8 resins volatiles 5.7 84.36 12.03 saturates 30.8 aromatics 81.60 9.99 30.6 resins 82.01 8.44 22.1 asphaltenes 83.17 6.66 10.4 volatiles asphaltenes 2.6 83.87 saturates 12.63 14.2 aromatics 10.36 81.79 12.4 resins 81.60 8.05 21.0 asphaltenes 82.00 6.18 39.4 coke 82.42 5.50

saturates and aromatics are not significantly changed by thermal conversion, the hydrogen contents of resins and asphaltenes are significantly decreased. This demonstrates one of the objections of Bunger and Cogswell (1981), that asphaltenes change properties during conversion as well as convert to other fractions. The conversion is more of a continuum than a discrete transformation. Meanwhile, toluene-insolublecoke, not present in the resid, forms and exceeds the asphaltenes in concentrations of aromatics (measured by H/C), sulfur, and nitrogen. A more direct way to track resid thermal conversion is to react the resid fractions separately and then to separate the products according to the separation scheme, recognizing that reactions and intermolecular interactions among the fractions may alter the chemistry when reacted together in the resid. This was again performed for 60 min at 400 "C in tubing bomb reactors. A summary of the result is shown in Table 111. In order to have sufficient quantity for reaction, the saturates and aromatics were reacted together in the proportion found in the resid. In each case the thermal conversion of a fraction formed the next more aromatic and higher molecular weight fraction and the whole series of less aromatic and lower molecular weight fractions. Thus, saturates plus aromatics formed the more aromatic and higher molecular weight resins with a byproduct of gas and light volatiles that evaporate with the solvents. The resins formed more aromatic and higher molecular weight asphaltenes, with a byproduct of lower molecular weight and less aromatic saturates and aromatics. The asphaltenes formed more aromatic and higher molecular weight coke with byproducts of lower molecular weight and lower aromatic resins, aromatics, and saturates. Reactants of increasing aromaticity and molecular weight produce an increasing yield of higher molecular weight-more aromatic byproduct at the same thermal reaction conditions. As was seen with the whole resid, the saturates and aromatics fractions maintain similar aromaticity (H/C atomic ratio) during thermal conversion but decrease in molecular weight. This is even the case when saturates and aromatics are formed from resins and asphaltenes. On the other hand, the resins and asphaltenes become more aromatic and decrease in molecular weight prior to being

wt %

S

wt%N

VPO MW

0.08 0.11 1.15 1.50 1.78

690 470 899 2009

2.18 5.04 5.26 7.23 7.63

H/C (atomic)

wt % S

wt % N

VPO MW

1.74 1.50 1.29

1.79 4.17 5.32

0.0 0.0 0.0

694 345 839

1.70 1.46 1.23 0.95

2.77 4.78 4.60 6.76

0.17 0.12 2.18 1.98

670 442 804 1841

1.79 1.51 1.18 0.90 0.80

2.54 4.92 6.68 7.96 7.92

0.0 0.0 0.90 2.18 1.73

422 622 1557 7525

converted to another fraction. Again, the situation is no different if the resin or asphaltene is the reactant, a product of another fraction, or a product of the entire resid. One difference is that resins formed from saturates and aromatics contain below the detectable nitrogen level. Thus, resins need not contain nitrogen to adsorb on Attapulgus clay, showing that the separation is made on the basis of aromaticity or the size of aromatic rings. This is the first time that the elemental analyses and molecular weights of five fractions of the thermal conversion products of a resid and its fractions has been published. Much has been published on the thermal conversion of asphaltenes with emphasis on the volatile products (for example, Ritche et al. (1979) and Speight and Pancirov (1984)). Speight (1987) has characterized the benzene-soluble, benzene-insoluble, and benzene-insolublepyridine-soluble products of asphaltene thermal conversion. While Speight observed a lower average molecular weight (VPO measured in pyridine) for the benzene-insoluble-pyridine-soluble product than that of the starting asphaltene, pyridine did not dissolve all of the benzene insolubles. In this study o-dichlorobenzene, a superior solvent for carbonaceous materials, was used to dissolve the entire toluene-insoluble product of asphaltene thermolysis and show that, on average, the molecular weight is higher for this product than the starting asphaltenes. Actually, the combination of these two studies shows that both routes to coke formation, higher molecular weight and lower molecular weight but more aromatic, happen simultaneously. The path of saturates and aromatics to resins to asphaltenes has been produced by oxidation reactions (Speight and Moschopedis, 1981) and postulated in a hydroconversion kinetic model (Takatauka et al., 1989a). However, this is the first time that direct evidence of thermal reactions following this path has been published, let alone determining that each step involves molecular weight growth. A previous study of resin thermolysis (Pakash et al., 1980) emphasized the volatile products. Although Speight (1970) characterized the products of thermal conversion of bitumen, asphaltenes, and maltenea, he carried the conversion too far (2-570 nonvolatile benzene solubles) to reach conclusions as to how the in-

Ind. Eng. Chem. Res., Vol. 31, No. 2,1992 533 termediate products evolved.

Concept of a Compositional Map In agreement with Bunger and Cogswell (19811, this author has found that no one property uniquely differentiates one petroleum fraction from another. However, Long (1981) has suggested that the composition of heavy oils could be displayed in terms of a map of molecular weight vesus "polarity". This concept has been more recently extended by Long and Speight (1989a) into a map of molecular weight versus solubility parameter. They used desorption off Attapulgus clay with various solvents to separate Cold Lake Crude and Arab Heavy vacuum resid into eight fractions. Using the solubility parameter of the fraction and the 95 and the 5% points in the molecular weight distribution determined by gel permeation chromatography (GPC), they found that these two different feeds gave similar molecular weight ranges for fractions eluted by the same solvent. The molecular weight range increases sharply above a solubility parameter of 9 hildebrands ( ~ a l / c m ~ ) ' /However, ~. the relative amounts of the fractions are different for the two feeds. Possibly more important than the exact selection of GPC molecular weight and solubility parameters as the independent variables is the concept of Long and Speight that petroleum fractions may be distinguished with combinations of two properties: one that measures molecular attraction and one that measures molecular size. The use of solubility parameter as the measure of molecular attraction is awkward due to its indirect measurement. While measuring molecular weight distribution is preferred, GPC suffers from adsorption effects with petroleum fractions that become more severe after partial conversion. Thus, to meet the objective of a convenient method to track resid conversion, different independent variables for a compositional map were further investigated. Measurement of Molecular Attraction The feature that causes petroleum fractions to adsorb strongly on Attapulgus clay and to be insoluble in paraffinic liquids is their aromaticity. Since the more aromatic fractions have higher oxygen and nitrogen content, it is not easy to discount completely the influence of these heteroatoms. However, as already pointed out, resins formed from saturates and aromatics had no nitrogen, showing that it was not a necessary part of resins. In addition, the average resin of molecular weight 600-1000 and the average asphaltene of molecular weight 1500-3000 have one or fewer oxygen or nitrogen atoms, yet the resins have at least 45% aromatic carbons and asphaltenes have a t least 50% aromatic carbons. Thus, while oxygen and nitrogen functionality may play a small role, aromaticity is the dominant cause of molecular attraction in petroleum fractions that cause adsorption on solids and insolubility in paraffinic liquids. This conclusion contradicts the current view (Speight, 1980). However, if one wanted to extend this approach to coal liquefaction products then a measure of molecular attraction involving oxygen functionality would be required in addition to aromaticity and molecular weight. This was the conclusion of Snape and Bartle (1984, 1985) who found that three independent variables, number average molecular weight, the proportion of internal aromatic carbon to told carbon, and the percent of acidic OH, were needed to distinguish among benzene insolubles, asphaltenes, and n-pentane solubles for coaland petroleum-derived liquids. Thus, their "phase diagram" is three dimensional. The simplest measure of aromaticity is the hydrogen to carbon atomic ratio (H/C). Ouchi (1985) has shown that

i o

0 130 DEGREES C PO ao 40 60 eo 70 80 so CONCENTRATION, DRAMS SOLUTE I KQRAMS SOLVENT

10

io0

Figure 2. Asphaltene association broken by dilution is also broken by temperature.

H/C is a linear function of the fraction of aromatic carbons as measured by 13CNMR. He even proposes that benzene insolubles, asphaltenes, and n-pentane solubles each follow a different linear function of either H/C or aromaticity versus molecular weight as long as the fraction does not contain large amounts of oxygen (>12 wt %). However, if such a relationship exists, one could distinguish among the fractions with only one property, which one cannot. Nevertheless, the use of H/C to measure aromaticity rather than 13CNMR meets our need for convenience. However, since we have shown that the carbon content of petroleum fractions from saturates to coke have nearly the same carbon content (81.6-86.5 wt %), without much loss in accuracy, the hydrogen content can be used to measure aromaticity and, thus, molecular attraction.

Measurement of Molecular Weight The measurement of the molecular weight of petroleum asphaltenes is not an exact science (Speight et al., 1985), as the value depends on the technique (Dickie and Yen, 1967), the solvent, and the temperature (Moschopedis et al., 1976). The problem is that asphaltenes fall in that gray area between being in solution and being a colloid because of their tendency to self-associate and to form aggregates that can be deteded by small-angle X-ray (Kim and Long, 1979) and neutron (Overfield et al., 1989) scattering. As a result, techniques that measure molecular weight in solution tend to give values that are high. On the other hand, the low volatility of asphaltenes interfere with mass spectrometry techniques and produce molecular weight measurements that tend to be low. Also, the strong tendency of asphaltenes to adsorb causes GPC techniques to produce low molecular weight measurements because the higher molecular weight, more adsorbing asphaltenes are supposed to pass through the gel column first. The most encouraging news is that neutron scattering (Overfield et al., 1989) shows that asphaltenes dissociate rapidly with increasing temperature. This author has obtained the most consistent results in measuring the number average molecular weight of unreacted and converted petroleum fractions by using vapor pressure osmometry near the maximum instrument temperature of 130 "C and using one of the best solvents, o-dichlorobenzene. An example result is shown in Figure 2, where the apparent molecular weight of an asphaltene from a different resid than the one in Table I is plotted against concentration. When the measurement is done at 70 "C, the apparent molecular weight decreases linearly with decreasing concentration. However, at 130 "C there is little change in apparent molecular weight with concentration, and the average molecular weight of 3400 at

534 Ind. Eng. Chem. Res., Vol. 31, No. 2, 1992 ,o.*J

~

Legend

*oo*]

0

0

z

ASPHALTENES 7000-

0 RESINS

;

rnWCNE

-I

/\ROHATICS

I

0 SATURATES

I

V

1

COKE

1

o

I

P

s

4

I

I

7

I

8

~

1

i

i

~

1

a

WEIGHT PERCENT HYDROQEN

Figure 3. Solvent-resid phase diegram.

130 "C agrees with the value of 3380 at 70 "C, extrapolated back to zero concentration. This is in contrast with vapor pressure osmometry measurements on the same asphaltene in toluene at 50 "C that gives an extrapolated molecular weight of 4900. Thus, it is concluded that either the better solvent or the higher temperature causes asphaltenes to dissociate down to a consistent molecular size. However, even this evidence does not ensure that the asphaltenes are completely dissociated. Asphaltenes (and coke) that are partially converted associate even more strongly than unconverted asphaltenes. For these, both higher temperature and the better solvent are needed to maximize the degree of dissociation. Moschopedis et al. (1976) have proposed nitrobenzene and pyridine as superior solvents for vapor pressure osmometry measurements on petroleum asphaltenes because they yield low molecular weight measurements. However, observation of mixtures of these two liquids with asphaltenes between glass slides with an optical microscope at 600X reveals that they do not completely dissolve the asphaltenes. Thus, the low molecular weight measurements are a result of the higher molecular weight fraction not being in solution. In this study, all mixtures of asphaltenes and coke with o-dichlorobenzene were checked with an optical microscope to ensure that they were in solution to the limit of resolution (0.5 pm).

Solvent-Resid Phase Diagram In Figure 3, the number average molecular weight versus hydrogen content is plotted for fractions of eight different resids and their thermal reaction products. Each of the five pseudocomponents occupy areas of the graph without overlap with points representing other pseudocomponents. Thus, it is concluded that the combination of number average molecular weight and hydrogen content is capable of differentiating between petroleum fractions, whether or not the fraction has been partially converted. For convenicence, solid curves are drawn to show that all species with molecular weight and hydrogen content that lie to the right of one curve are heptane soluble and to the right of the other curve are toluene soluble. This, plus the unique area for each of the petroleum fractions, suggests that the diagram is similar to a phase diagram but here the different "phases" are due to their solubility behavior in particular solvents or due to their adsorption from solution. This is why it is called a "solvent-resid phase diagram". The dashed curve in Figure 3 encloses all the points representing fractions from unconverted resid. The area representing convertad resids is much larger than the area occupied by unconverted resids. Actually, the aromatics and saturates occupy nearly the same area for converted

~

and unconverted resids. However, the resins and the asphaltenea upon conversion tend to move to lower hydrogen content (higher aromaticity) and equal or lower molecular weight. Since a pure compound would be a point on this diagram, the smaller the area of a pseudocomponent the better is its approximation as a pure component. The saturates, aromatics, and resins occupy about the same area, but the area for asphaltenes is much larger. The area for coke is probably even larger, but one can only represent coke that is soluble enough in o-dichlorobenzene at 130 "C to measure its molecular weight. The solvent-resid phase diagram also provides insight on the transformation of one pseudocomponent to another. Resins can be formed from aromatics by either molecular i 6 weight growth or by decreasing hydrogen content by cracking off more saturated fragments or by aromatization of naphthenoaromatics. On average, both mechanisms occur as was found with the resins formed from saturates and aromatics. Likewise, asphaltenes form from resins either by molecular weight growth or by decreasing hydrogen content. However, here it clearly does not occur only by molecular weight growth, as the only asphaltenes of equal hydrogen content as resins are from uncoverted resids. On the other hand, coke probably does occur from molecular weight growth of partially converted asphaltenes, as well as by decreasing hydrogen content. Of course, as we saw by reacting the pseudocomponents, each of the thermal reactions produces the more desirable range of lower molecular weight-higher hydrogen content byproducts.

Implications and Discussion The solvent-resid phase diagram points out some of the advantages and overcomes the greatest disadvantage in using solubility and adsorption fractions for tracking resid chemical changes. It shows that pseudocomponents characterized by solubility and adsorption can be distinguished from each other by areas on a plot of molecular weight and hydrogen content. By being areas rather than points, it also clearly shows that each pseudocomponent represents a collection of a large number of different molecules and that chemical changes can occur without changing one pseudocomponent into mothers. However, even when this occurs, by measuring the hydrogen content and the molecular weight of the pseudocomponent, one can continue to track the path of the chemical change using the solvent-resid phase diagram. This is most easily seen for the asphaltenes in which all the asphaltenes from unconverted resids lie within the dashed curve while all those from converted resids lie outside these boundaries. This justifies the asphaltene reaction pathway presented by Schucker and Keweshan (1980) and used by Savage et al. (1985), in which asphaltene cores are considered to be separate pseudocomponents from unconverted asphaltenes. The similarity of the pseudocomponents after thermal reaction, no matter which pseudocomponent is the reactant, has often been assumed but not previously experimentally observed. In balance, solubility and adsorption fractionation is a logical experimental approach for tracking resid conversion as long as one understands the limitations. While the pseudocomponents represent large distributions of molecules, the molecules have something in common (range of molecular weights and hydrogen contents) and can be distinguished from each other. It one wishes to better define a pseudocomponent, he can experimentally fractionate the pseudocomponent into two or more fractions. However, the separation scheme presented here is already quite labor intensive, requiring considerable motivation

Ind. Eng. Chem. Res., Vol. 31, No. 2, 1992 535

V = VOLATILES R

w

s

+

v

S=SATURATES

-

AR = AROMATICS

A

R

+mAR ’ ”

RESINS

A=ASPHALTENES

+

C = COKE C *R+

AR

+

S

+

V

*

+ MORE AROMATIC AND/OR HIGHER MOLECULAR WEIGHT

LESS AROMATIC AN0 LOWER MOLECULAR WEIGHT

Figure 4. Thermal conversion reaction path.

for additional steps. Another approach is to keep the number of separation steps to a minimum but model the chemical changes with a large number of molecular species, such as the stochastic model of Savage and Klein (1989). However, one still needs to predict from the chemical structure of a molecule to which experimental pseudocomponent it belongs. Savage and Klein were visionary enough to use criteria for their model based upon molecular weight and H/C, albeit in detail their boundary assignments were inaccurate. For instance, they defined asphaltenes as having molecular weights above 300 and H/C greater than 1.0 (hydrogen content greater than 7.0 wt %). Figure 3 shows that such criteria would include all resins and some saturates and aromatics while excluding most of the asphaltenes from converted resids. Nevertheless, now with the solvent-resid phase diagram, the assignments of a molecule in such a model to an experimental pseudocomponent can be done with much greater accuracy. In addition, this same diagram enables modeling chemical changes that do not completely convert pseudocomponents. A mechanism that can rationalize much of the data presented in this paper is that the same aromatic moieties that impart insolubility in solvents, adsorption on clay, and low volatility also impart a thermal reaction limit to the resid. As is represented in Figure 4,the saturates can nearly be completely converted thermally to volatile products. However, the resins and asphaltenes tend to leave low-volatilityfragments that are more aromatic and lower in molecular weight after cracking off fragments that are less aromatic and lower in molecular weight. Eventually, these more aromatic fragments are converted to a leas soluble pseudocomponent either by combining to form a higher molecular weight species and/or by sufficient increase in aromaticity. On the other hand, aromatics form higher molecular weight and more aromatic resins by direct conversion without passing through a more aromatic intermediate. Naturally, as the aromaticity of a pseudocomponent is increased, it produces a greater fraction of more aromatic products (toward the left in Figure 4). Previously, the path to coke formation has either been classified completely as a polymerization process (Takatsuka et al., 1989b) or completely as a molecular weight reduction and aromaticity increase process (Savage and Klein, 1989). Our evidence is that it is a combination of these two paths, except that the molecular weight growth is more of an oligomerization process involving the combination of two to five molecules. While the examples of resid conversion in this paper have all been thermal conversion, the principles should also apply to catalytic hydroconversion. The solvent-resid phase diagram is process independent. However, the ad-

dition of hydrogen reduces both the aromaticity by hydrogenation and the molecular weight growth by radical capping. Thus, an even greater percentage of the hydroconversion products from each pseudocomponent are directed to the right in Figure 4 toward less aromatic and lower molecular weight products. Acknowledgment I am grateful to Exxon Research and Engineering Co. for permission to publish this paper and to J. L. Machusak, W. R. Gerald, and K. Greany for careful experimental and computer expertise. Literature Cited Bunger, J. W.; Cogswell, D. E. Characteristics of Tar Sand Bitumen Asphaltenes as Studied by Conversion of Bitumen by Hydropyrolysis. In Advances in Chemistry Series 295; Bunger, J. W., Li, N. C., Eds.; American Chemical Society: Washington, DC, 1981;pp 219-236. Dickie, J. P.; Yen, T. F. Macrostructures of the Asphaltic Fractions by Various Instrumental Methods. Anal. Chem. 1967, 39, 1867-1857. Kim, H.; Long, R. B. Characterization of Heavy Residuum by a Small Angle X-ray Scattering Technique. Znd. Eng. Chem. Fundam. 1979,18,60-63. Long, R. B. The Concept of Asphaltenes. In Advances in Chemistry Series 195; Bunger, J. W., Li, N. C., Eds; American Chemical Society: Washington, DC, 1981;pp 17-27. Long, R. B.; Speight, J. G. Studies in Petroleum Composition. Development of a Compositonal ”Map” for Various Feedstocks. Rev. Znst. Fr. Pet. 1989a,44,205-217. Long, R. B.; Speight, J. G. Studies in Petroleum Composition. 2. Scale-up studies for Separating Heavy Feedstocks by Adsorption. Znd. Eng. Chem. Res. 1989b,28, 1053-1057. Moschopedis, S. E.; Fryer, J. F.; Speight, J. G. Investigation of Asphaltene Molecular Weights. Fuel 1976,55,227-232. Ouchi, K. Correlation of Aromaticity and Molecular Weight of Oil, Asphaltene and Preasphaltene. Fuel 1985,64,426-427. Overfield,R. E.; Sheu, E. Y.;Sinha, S. L.; Liang, K. S. SANS Study of Asphaltene Aggregation. Fuel Sci. Technol. Znt. 1989, 7, 611-624. Pakash, S.;Moschopedis, S. E.; Speight, J. G. Thermal Decomposition of Resins. Fuel 1980,59, 64-66. Ritche, G. R.; Roche, R. S.; Steedman, W. Pyrolysis of Athabasca Tar Sands: Analysis of the Condensible Products From Asphaltene. Fuel 1979,58, 523-530. Roberts, I. The Chemical Significance of Carbon Residue Data. Prepr.-Am. Chem. SOC.,Diu. Pet. Chem. 1989,34,251-254. Savage, P. E.; Klein, M. T. Asphaltene Reaction Pathways. V. Chemical and Mathematical Modeling. Chem. Erg. Sci. 1989,44, 393-404. Savage, P. E.; Klein, M. T.; Kukes, S. G. Asphaltene Reaction Pathways. 1. Thermolysis. Ind. Eng. Chem. Process Des. Dev. 1985,24,1169-1174. Schucker, R. C.; Keweshan, C. F. The Reactivity of Cold Lake Asphaltenes. Prepr. Pap.-Am. Chern. SOC.,Diu. Fuel Chem. 1980, 25, 155-164. Snape, C. E.;Bartle, K. D. Definition of Fossil Fuel Derived Asphaltenes in Terms of Average Structural Properties. Fuel 1984, 63,883-887. Snape, C. E.;Bartle, K. D. Further Information on Defining Asphaltenes in Terms of Average Structural Properties. Fuel 1985, 64,427-429. Speight, J. G. Thermal Cracking of Athabasca Bitumen, Athabasca Asphaltenes and Athabasca Deasphalted Heavy Oil. Fuel 1970, 49,134-135. Speight, J. G. The Chemistry and Technology of Petroleum; Marcel Dekker, Inc.: New York, 1980;pp 192-243. Speight, J. G. Initial Reactions in the Coking of Residua. Prepr.Am. Chem. SOC.,Diu. Pet. Chern. 1987,32,413-418. Speight, J. G.; Moschopedis, S. E. On the Molecular Nature of Petroleum Asphaltenes. In Advances in Chemistry Series 195; Bunger, J. W., Li, N. C., Eds.; American Chemical Society: Washington, DC, 1981;pp 1-15. Speight, J. G.; Pancirov, R. J. Structure Types in Petroleum Asphaltenes as Deduced from Pyrolysis/Gas Chromatography/Mase Spectrometry. Liq. Fuels Technol. 1984,2,287-305.

Ind. Eng. Chem. Res. 1992, 31, 536-543

536

Speight, J. G.; Long, R. B.; Trowbridge, T. D. Factors Influencing the Separation of Asphaltenes from Heavy Petroleum Feedstocks. Fuel 1984,63,616-620. Speight, J. G.; Wernick, D.L.; Gould, K. A.; Overfield, R. E.; Rao, B. M. L.; Savage, D. W. Molecular Weight and Association of Asphaltenes: A Critical Review Rev. Znst. Fr. Pet. 1985,40,51-61. Takatsuka, T.; Wada, Y.; Hirohama, S.; Fukui, Y. A Prediction Model for Dry Sludge Formation in Residue Hydroconversion. J .

Chem. Eng. Jpn. 1989a,22, 298-303. Takatsuka, T.; Kajiyama, R.; Hashimoto, H. A Practical Model of Thermal Cracking of Residual Oil. J. Chem. Eng. Jpn. 1989b,22, 304-310. Received for review April 15,1991 Revised manuscript received September 3, 1991 Accepted September 14,1991

Multivariable Controller for Distillation Columns in the Presence of Strong Directionality and Model Errors Alessandro Brambilla* Dipartimento d i Ingegneria Chimica, Uniuersitci di Pisa, Via Diotisalvi, 2, 56100 Pisa, Italy

Luigi D’Elia C O N T A S Process Control s.r.l., Via U. Visconti, 24/b, 56122 Pisa, Italy

An easy-tuning multivariable controller useful for the control of the two products of distillation columns is presented. The controller is based on the singular value decomposition (SVD) of the model transfer function matrix and requires only one tuning parameter, with values in the range 0-1, by which a trade-off between the nominal performance and its sensitivity to the model/real process mismatch is accomplished. The development of this control algorithm does not require any particular software, and the criteria for the tuning procedure are reported in the paper. The controller can be easily implemented and on-line tuned on the modern control systems. The effectiveness of the controller and the transparent meaning of its parameter are shown by the simulation of a liquefied petroleum gas (LPG)splitter and a high purity distillation column. Introduction Most of the industrial chemical processes have the characteristic of multiinput/multioutput system (MIMO). Among these processes, the distillation units are the ones which might show to a large extent the peculiarities of the MIMO systems: the interactions between the control loops and the process directionality. In this paper, after a brief recall of these characteristics and their effect on the performances of a control scheme, a multivariable controller based on the singular value decomposition (SVD)is presented. The main advantage of the controller presented rests on the possibility of its making a trade-off between performance and robustness to model/process mismatches, by means of only one parameter which makes the tuning simple and transparent. The application of the controller to distillation columns characterized by strong directionality in the presence of model errors is presented. Loop Interactions. When the control structure is decentralized, i.e., each product composition is controlled by the action of one manipulated variable as for single-input/single-output systems (SISO),the presence of the interactions between the two control loops may lead to a severe degradation of the control system performances. The classic approach with multivariable controllers allows us to solve this problem at least from a theoretical point of view (Luyben, 1970; Ray, 1981). Unfortunately, it is not always possible to attain improvements through the application of these multivariable techniques to real processes. In fact, conventional multivariable controllers with the ability to eliminate or reduce the interactions are based on the process model, or more precisely on its inverse (“inverse-based”controllers), and their performances are in some conditions very sensitive to the mismatch between the model and the real process (model errors) (Skogestad and Morari, 1988). These model errors are generally OSSS-5SS5/92/2631-0536$03.00/0

present because of the following reasons: necessary simplifications adopted in modeling the process; lack of knowledge of the process parameters; use of linear model to describe nonlinear processes; errors in assigning the exact value of the manipulated variables to the process. Lau et al. (1985) proposed a method to design a structural compensator able to minimize control loop interactions based on the singular value decomposition. The effectiveness of such compensator in distillation has been verified by Bequette and Edgar (1989). The relative gain array (RGA) (Bristol, 1966) is an adequate tool for the interaction analysis although based on the steady-state model of the process. Shinskey (1984) and Mc Avoy (1983), have used the RGA for the selection of the control system structure for distillation columns. The ij element of the RGA is defined as X,J =

16Yi/6mjlmK,K#j

(1)

16Yi/6mjlYK,K#i

where y and m are respectively the controlled and manipulated variables. For a 2 X 2 system with the process transfer function at the steady-state G(s=O) = Go, the application of eq 1 gives 1--

g11g22

where gij are the gains of the matrix Go. According to the definition, the interaction degree is measured as the ratio between the process gain with all the loops open and the gain with all the other loops closed. Thus, the interactions are small when X i j is close to 1. Although the definition of eq 1is limited to the steady state, Mc Avoy (1983) has extended the use of the RGA in the frequency domain. 0 1992 American Chemical Society