Professional Documents
Culture Documents
New Approach To The Characterisation of Petroleum Mixtures Used in The Modelling of Separation Processes
New Approach To The Characterisation of Petroleum Mixtures Used in The Modelling of Separation Processes
Abstract
Characterisation of complex mixtures is a common tool especially in oil processing industry. Characterisation procedures result in experimentally
gained characterisation curves, but for the simulation of industrial processes the definition of a substitute mixture is required. Traditionally, a system
of pseudocomponents is derived from the True Boiling Point (TBP) characterisation curve, but there are a number of disadvantages, e.g. the physical
properties of pseudocomponents must be estimated by unreliable empirical methods. The new approach to the characterisation of complex mixtures
is based on representing the original mixture by a system of real components. Such substitute mixture is fully defined, it has a chemical character,
and physical properties can be simply retrieved from databases. Utilisation of a substitute mixture of real components in the simulation of crude
oil processing proved that the new approach could replace the traditional one in normal boiling temperature ranges where real components are
available. Both approaches could be also easily combined.
© 2005 Elsevier Ltd. All rights reserved.
Keywords: Refinery processes; Oil processing; Characterisation procedures; Complex mixtures; Pseudocomponents
0098-1354/$ – see front matter © 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compchemeng.2005.10.005
344 E. Eckert, T. Vaněk / Computers and Chemical Engineering 30 (2005) 343–356
i ) + Tb (Φi )
Tb (ΦR L
2. Characterisation of mixtures using Tbi = , i = 1, . . . , I (1)
2
pseudocomponents
ΦR
1 i
Normally, two steps are needed to define a system of pseu- Tbi = R Tb (Φ) dΦ, i = 1, . . . , I (2)
Φi − Φi ΦLi
L
docomponents for a complex mixture. First, some standard
characterisation curve must be experimentally obtained, prefer- As also illustrated by Fig. 1, the interval of fraction distilled
i − Φi ), i = 1, . . ., I determines at the same time the relative
(ΦR
ably the True Boiling Points (TBP) curve, which is directly L
used in the next step. The TBP curve represents the depen- contribution of each pseudocomponent to the mixture, which is
dence of temperature measured in the head of a laboratory used to define its composition.
batch column on mass or volume fraction distilled. The col- For the consequent usage of the substitute mixture of pseu-
umn should have a high number of stages and perform at large docomponents it is necessary to supply the same structure of
reflux ratio (API, 1992). There are other possibilities, e.g. the physical properties as for real components. Usually, it is not
346 E. Eckert, T. Vaněk / Computers and Chemical Engineering 30 (2005) 343–356
difficult to measure besides the TBP curve also the density and/or where M is the molecular weight, ρ the liquid density (even-
viscosity of individual fractions collected in the laboratory col- tually specific or API gravity) and η is the liquid viscosity.
umn during the distillation test. Moreover, for each fraction the The liquid density curve can be also obtained approximately
mean molecular weight can be estimated. Therefore, if we use using the Watson factor as mentioned above. Instead of using
pseudocomponents to represent these fractions then two to four Eqs. (5)–(7) directly, it is better to convert them into “phase
data items are available for each pseudocomponent: the mean portraits” by eliminating the mass or volume fraction distilled
normal boiling point and one to three other properties, which and establishing direct relation between these properties and
can be used to trigger a series of empirical estimation procedures the boiling point temperature:
delivering the remaining set of properties—critical properties,
acentric factor, etc. (see, e.g. Riazi, 2005; Whitson & Brulé, M = M(Tb ) (8)
2000). Unfortunately, we should be aware of low reliability of
ρ = ρ(Tb ) (9)
such methods as it was mentioned in the introductory part of this
contribution. η = η(Tb ) (10)
It is relatively simple to measure global parameters of a
complex mixture, i.e. bulk properties as the density, molecular c. The overall temperature range is divided (cut) into a sys-
weight, refraction index or viscosity, and this is often delivered tem of non-overlapping temperature intervals continuously
with other measurements. Unfortunately, their direct use for the covering the range. There are several possible approaches
derivation of a substitute or model mixture is known only for the similarly to the definition of pseudocomponents – an equidis-
density where it is possible to construct under certain conditions tant grid, adaptive grid taking into account the shape of the
an approximation of the characterisation curve from the Watson curve or a grid based on boiling points of homological series
factor (Wauquier, 1995). When the global density of the mixture of alkane compounds – see, e.g. Table 5.2 in Whitson and
is known and the Watson factor Brulé (2000).
d. A sufficiently large database of chemical components and
1/3
(1.8Tb ) their physical properties is available. Detailed requirements
KW = (3)
S will be discussed further.
can be assumed to be constant then it is possible to get an approx-
3.1. Selection of components
imation of the density characterisation curve (Wauquier, 1995).
This could be an alternative to density data measured for each
The basic assumption is that a suitable database is available.
fraction during the characterisation experiment.
The “quality” of such database can be expressed not only by the
reliability of data included but also by the extent of the database.
3. Characterisation of mixtures using real components Here, we take primarily into account the number of components
but on its own this is not sufficient. It is important to have satis-
The main disadvantages of the approach based on pseudo- factory representation for important families of chemical com-
components can be overcome if we employ real components pounds, in our case especially hydrocarbons or detailed groups
to form the substitute mixture. Of course, the selection of suit- as, e.g. paraffines or aromatics. At the same time, the range of
able real components and the derivation of the substitute mixture normal boiling points for these compounds must be wide and
must follow certain criteria and an appropriate algorithm must be the normal boiling points should cover the range uniformly and
defined. We shall use two consecutive phases, i.e. a set of compo- densely. This is, of course, an ideal situation, which is more or
nents to be present in the substitute mixture must be first selected less fulfilled by databases from various sources. In “open” lit-
and then the composition of their mixture will be adjusted. The erature we can find, for example, the API database (API, 1992)
prerequisites for the algorithm are as follows: with 478 hydrocarbons up to C30 , normal boiling points up to
720 K and molecular weights up to 420 kg/kmol. Generally, nor-
a. The TBP curve measured for the original complex mixture mal boiling points of higher hydrocarbons are very difficult to
is available obtain as the thermal decomposition takes place (Kopsch, 1995).
Accordingly, data for higher hydrocarbons (C18+ ) in databases
Tb = Tb (Φ) (4) are frequently estimated or their origin is ambiguous. Neverthe-
less, their long-term usage in many chemical engineering cal-
where Tb is the measured “True Boiling Point” temperature
culations giving realistic results gives certain guaranty that also
and Φ is the mass or volume fraction distilled.
the substitute mixture of real components selected from such
b. Some other characterisation curve is available, e.g. all or
databases can sufficiently model the original complex mixture.
some of dependencies
The internal database of the HYSYS simulation program
M = M(Φ) (5) (Hyprotech, 2003), which we used for test calculations, incor-
porates 521 hydrocarbons and covers approximately the same
ρ = ρ(Φ) (6) range of normal boiling points and molecular weights as the
API database. The source of data is not explicitly specified and
η = η(Φ) (7) probably the same limitations can be expected as for API data.
E. Eckert, T. Vaněk / Computers and Chemical Engineering 30 (2005) 343–356 347
Fig. 4. Situation after the selection of a real component in the ith primary tem-
perature interval (Ti , Ti+1 ). Fig. 5. Construction of intervals of fraction distilled in the second phase of the
algorithm attached to real components selected in the first phase.
viscosity and density in fact represent properties of discrete mix-
tures (fractions), which can be hardly derived from properties of model for varying composition of the mixture until a satisfactory
contained pure components according to non-trivial and unreli- match with the experimental characterisation curve is reached.
able mixing rules. This kind of an optimisation can be easily employed, for exam-
ple, for the EFV curve (Eckert, 1999).
Sometimes no other curves than the TBP curve are available.
When the TBP curve is available then the procedure sug-
Step 2 of the algorithm can be then replaced by choosing a com-
gested by Ba et al. (2003) can be used. Fig. 5 illustrates the
ponent with its normal boiling point being closest to the mean
principles of the procedure, which at first glance could be found
temperature of the appropriate temperature interval. This is an
similar to the method used to derive the composition of a mixture
emergency solution but it can be expected that the selected real
of pseudocomponents. The presumption that each component
component would be close to a pseudocomponent defined tradi-
should represent an interval of fraction distilled, where its nor-
tionally while all properties of the real component are directly
mal boiling point is the mean temperature according to Eq. (1)
available from the database.
or (2), is preserved. The difference is that we cannot expect
The result of this phase of the algorithm is simply a list of
that the intervals could be generally chosen to cover the whole
real components together with their normal boiling points and
range without gaps and/or overlapping. This is automatically
other properties. If it is desired for some reason to include other
fulfilled for pseudocomponents where the procedure starts from
components into the substitute mixture, it can be done now. Such
the other end—the mean temperatures are calculated for temper-
an obvious reason is, for example, the confirmed presence of a
ature intervals initially chosen to continuously cover the entire
particular component or component type in the original mixture.
temperature range without overlapping. Nevertheless, in the case
Especially polar compounds are in the focus since they strongly
of real components the idea is to distribute the intervals to make
affect the phase equilibrium in multiphase systems. A compound
them cover the range as much as possible with minimum gaps
can be added to the substitute mixture with or without informa-
and overlapping. This is an optimisation task where the objective
tion about its amount in the original mixture. Both possibilities
function can be defined as follows:
are aided in the second phase of the algorithm.
I+1
2
F (ΦL ) = i−1 − Φi ) → min
(ΦR L
(12)
3.2. Determination of the composition of a substitute
i=LE+1
mixture
For each interval the starting and end fractions distilled are
The problem is that the normal boiling point of each selected denoted by the superscript ‘L’ (=left) and ‘R’ (=right). Gener-
component can fall anywhere in the primarily defined temper- ally, we can utilise the information about light-end components,
ature interval (Ti , Ti+1 ) (see Fig. 4), and no longer it is a mean which can be usually identified in the original mixture together
temperature of this interval. If a model of the experimental char- with their composition, and incorporate them into the substitute
acterisation procedure were available then we could abandon mixture. The index of the last light-end component is there-
the principle of mean temperatures and this step would consist fore denoted by ‘LE’ and the corresponding fraction distilled
in a repeated evaluation of the characterisation curve from the by ΦR LE . On the other end of the range, I + 1 can be denoted by
E. Eckert, T. Vaněk / Computers and Chemical Engineering 30 (2005) 343–356 349
elements of the vector ΦL = {ΦLLE+1 , . . . , ΦLI } are considered No. Component Liquid volume (%)
as independent optimisation variables in (12) is that Eqs. (1) 1 i-Butane 0.19
or (2) can be treated as implicit equations for ΦR i , i = LE + 1, 2 n-Butane 0.11
. . ., I, i.e. ΦRi = Φ R (ΦL ). Alternatively, the reversed relation
i i 3 i-Pentane 0.37
ΦLi = ΦLi (ΦR i ) could have been used with Φi as the set of opti-
R 4 n-Pentane 0.46
misation variables. It is reasonable to define bounds for the
values of ΦLi in order to avoid getting some unreal solutions
and the following set of constraints was considered: (Eckert & Vaněk, 2003) we used the concept of a “theoretical
TBP curve” as curve of a staircase shape reflecting the distribu-
ΦLi ≤ ΦLi+1 , i = LE + 1, . . . , I tion of normal boiling points of individual components from the
(13) mixture. Such curve we could get in a batch distillation column
Φi−1 ≤ ΦLi ≤ Φi+1 , i = LE + 1, . . . , I
with the number of stages approaching infinity and very high
Optimisation problem (12) together with constraints (13) can be reflux ratio. In its discrete version the theoretical TBP curve is
j=i−1
solved using some standard package. composed of points with co-ordinates ( j=1 (xj + xi )/2; Tbi )
In order to obtain a consistent composition of the substitute for i = 1, . . ., I, taking into account for simplicity the arithmetic
mixture it is necessary to convert the intervals (ΦLi , ΦR
i ) resulting mean values defined by Eq. (1). Analogical theoretical curves
from the optimisation to mass or volume fractions. In ideal case, can be constructed for molecular weight and density, eventually
the intervals would exactly cover the entire range, the value of the API gravity or some other properties.
object function (12) would be zero and the following condition Theoretical TBP curves were constructed for examples of
would be fulfilled: separations in a distillation and absorption columns (Eckert &
I
Vaněk, 2003) and the match between results obtained by both
i − Φi ) = ΦHE − ΦLE
(ΦR L L R approaches was excellent, thus showing that substitute mixtures
(14)
i=LE+1 of real components can replace mixtures of pseudocomponents
with all the extra benefits, which this approach brings. For these
Then, the length of each interval can be directly taken as the examples only the measured TBP curves were available as input
appropriate mass or volume fraction. Nevertheless, this is not information.
usually true according to possible overlapping of the intervals The example presented further is based on the “Refining
and existence of areas not covered with any interval. The sim- Tutorial” case delivered with the simulation program HYSYS
plest way how to get a consistent composition derived from the (Version 3.2) and at the same time this program has been used
lengths of intervals (ΦLi , ΦR
i ) is to “normalise” the vector of for all calculations. The advantage is that HYSYS incorporates a
mass or volume fractions: relatively extensive database of hydrocarbons and it also enables
j − Φj )(ΦHE − ΦLE )
(ΦR L L R to define pseudocomponents in a built-in tool called “Oil Envi-
xj = I , j = LE + 1, . . . , I (15) ronment”.
i=LE+1 (Φi − Φi )
R L
composition is known. The main flowsheet for the process in in the HYSYS program further represented by an embedded
consideration is shown in Fig. 6. Preheated crude is initially fed flowsheet, i.e. subflowsheet in HYSYS terminology. The appro-
to a preflash drum to separate vapours from the liquids, which priate subflowsheet for our example is shown in Fig. 7. The
are further heated in a furnace. The preflash vapours are recom- Peng–Robinson property package was used in HYSYS for the
bined again with the hot crude from the furnace and the resulting calculation of physical properties. In the “Oil Environment” the
stream is then fed to the atmospheric crude column for frac- standard setting of estimation methods was used, i.e. Twu (1984)
tionation. Detailed description and specifications are presented critical property correlation for molecular weight, constant Wat-
in Appendix A. The icon of the distillation column in Fig. 6 son factor method for specific gravity and Lee–Kesler method for
represents in fact a complex separation process, which can be critical properties, the acentric factor and ideal enthalpy (Kesler
& Lee, 1976).
It was intended to solve the simulation case using both meth-
Table 3
ods for the characterisation of the process feed (Preheat Crude
API gravity assay
stream). Particularly, Table 4 globally summarises how the sub-
Liquid volume (%) API gravity stitute mixtures were assembled. Temperature ranges defined in
13 63.28 the first column of the table were for pseudocomponents divided
33 54.86 uniformly to intervals of the same width. For real components
57 45.91 this was also done but, as we know from the description of the
74 38.21
algorithm above, it serves only for the initial selection of subsets
91 26.01
of real components as candidates for representing these intervals.
E. Eckert, T. Vaněk / Computers and Chemical Engineering 30 (2005) 343–356 351
Table 5
Resulting selection of real components (without the known light-end)
No. Temperature interval [K] Component Tb [K] M [kg/kmol] API gravity Number of candidate
components in interval
Fig. 8. Phase portrait for the molecular weight (—) compared with values Fig. 10. Measured TBP curve (䊉) compared with theoretical TBP curve for the
retrieved from the database for real components (). substitute mixture gained by the new approach ().
Fig. 9. Phase portrait for the API gravity (—) compared with values retrieved Fig. 12. Measured characterisation curve for the API gravity (䊉) compared with
from the database for real components (). the theoretical curve for substitute mixture gained by the new approach ().
E. Eckert, T. Vaněk / Computers and Chemical Engineering 30 (2005) 343–356 353
Table 6
Resulting substitute mixtures
No. Substitute mixture of pseudocomponents Substitute mixture of real components and pseudocomponents
Real component/pseudocomponent Liquid volume (%) Real component/pseudocomponent Liquid volume (%)
ures, it can be also observed that in the region of higher boiling ber of unit operations within a relatively complex technology
temperatures program HYSYS is not so successful in the choice with several recycle streams. Also for other parameters of the
of pseudocomponents and a considerable deviation between the product streams, as flowrates, temperatures and enthalpies, the
experimental and estimated values of the molecular weight as results of simulation proved very nice match. It can be noted
well as API gravity is apparent. that these results were reached despite the fact that the database
The simulation calculations of the entire crude oil fractiona- used is relatively small compared to other data sources. More-
tion process employing both approaches to the characterisation over, the content of some of the real components selected into
of complex mixtures run without any problems. The excel- the substitute mixture resulting from the second phase of the
lent match between both approaches can be demonstrated by algorithm is practically negligible (see Table 6—components
Fig. 13, where the comparison is provided for all the prod- numbers 6, 14, 18, 28 and 30). Results presented in Table 7
ucts of fractionation, which have distinct mean boiling points. are restricted to the calculated values of temperatures and mass
The plotted theoretical TBP curves were recalculated in order flowrates of product streams and to calculated duties, i.e. energy
to eliminate water. In fact, the TBP curves reflect the match streams in the process. Presentation of complete set of cal-
in the composition of product streams, as the most important culated parameters is here impossible according to its large
parameter, after processing of a single feed stream in a num- extent.
354 E. Eckert, T. Vaněk / Computers and Chemical Engineering 30 (2005) 343–356
Table A.1
Specifications of the parameters of material streams
Stream Temperature (◦ C) Pressure (kPa) Flowrate (kg/h) Flowrate (m3 /h) Composition
Preheated crude 232.2 517.1 662.4 (i.e. 100,000 bbl/day) See substitute mixtures (Table 6)
Bottom steam 190.6 1034 3402 H2 O only
Diesel steam 148.9 344.7 1361 H2 O only
AGO steam 148.9 344.7 1134 H2 O only
Hot crude 343.3
Naphtha 132.5
Offgas 0
Kerosene 86.12
Diesel 112.6
AGO 33.12
PA 1 Draw 331.2
PA 2 Draw 198.7
PA 3 Draw 198.7
Hyprotech, a subsidiary of Aspen Technology, Inc. (2003). HYSYS.Plant 3.2 cal Conference and Exhibition, paper SPE 13119, Houston, 16–19 Sept,
documentation. Cambridge, MA: Aspen Technology, Inc. 1984.
Katz, D., & Brown, G. (1933). Vapor pressure and vaporization of Murtagh, B. A., & Saunders, M. A. (1995). MINOS 5.4 user’s guide. Stan-
petroleum fractions. Industrial and Engineering Chemistry, 25(12), ford, CA: TR SOL 83-20R, Department of Operations Research, Stanford
1373. University.
Kesler, M. G., & Lee, B. I. (1976). Improved prediction of enthalpy of Petrov, A. A. (1987). Petroleum hydrocarbons. Berlin/Heidelberg: Springer-
fractions. Hydrocarbon Processing, 55, 153. Verlag.
Kopsch, H. (1995). Thermal methods in petroleum analysis. Weinheim: VCH. Rätzsch, M., & Kehlen, H. (1983). Continuous thermodynamics of complex
Krambeck, F. (1991). Continuous mixtures in fluid catalytic cracking and mixtures. Fluid Phase Equilibria, 14, 225.
extensions. In A. Sapre & F. Krambeck (Eds.), Chemical reactions in Riazi, M. R. (2005). Characterization and properties of petroleum fractions.
complex mixtures (pp. 42–59). New York: Van Nostrand Reinhold. Barr Harbor: ASTM.
Lindqvist, P., Markkanen, V., & Happonen, V. M. (1994). Simulation of a Twu, C. H. (1984). An internally consistent correlation for predicting the
heavy residue vacuum column. In ASPENWORLD’94. critical properties and molecular weights of petroleum and coal-tar liquids.
MDL Information Systems. (2005). Crossfire Beilstein database. San Fluid Phase Equilibria, 16(2), 137.
Leandro, CA: Elsevier MDL (http://www.mdl.com/products/knowledge/ Wauquier, J.-P. (Ed.). (1995). Petroleum refining, Vol. 1: Crude oil. Petroleum
crossfire beilstein/). products. Process flowsheets. Paris: Éditions Technip.
Montel, F., Gouel, P. L. (1984). A new lumping scheme of analytical Whitson, C. H., & Brulé, M. R. (2000). Phase behavior. SPE monograph
data for compositional studies. Presented at the 59th Annual Techni- series. Richardson: Society of Petroleum Engineers, Inc.