You are on page 1of 7

Fuel Processing Technology 70 Ž2001.

1–7
www.elsevier.comrlocaterfuproc

AUniversalB yield models for the steam pyrolysis of


hydrocarbons to olefins
John E. Gwyn )
26602 Willow Lane, Katy, TX, 77494-5418 USA

Received 28 February 2000; received in revised form 31 October 2000; accepted 7 December 2000

Abstract

Hydrocarbon pyrolysis with steam is the miracle process of the 20th century. Almost all
petrochemical base chemicals are produced by pyrolysis using a range of feedstocks from ethane
to gas oils. The reactions are so stiff that radical reactions are in equilibrium while feedstocks
fragment into primary products at relatively fast rates. These react at slower rate into precursors
Že.g. pentene-1. of desired products Že.g. butadiene, iso-butene.. Unfortunately desired products
Žethylene, propylene, etc.. also undergo degrading reactions. High temperature, short contact
conditions favor these products due to their high energies of activation.
AUniversalB models accurately predict yields for the wide variety of hydrocarbon feeds in the
numerous reactors under the range of severities. Indeed, it’s a miracle that a single computer
model of pyrolysis could meet the AuniversalB requirements. Indeed three approaches: free radical
mechanisms; molecular kinetics; and methane severity each have met the requirements. The rigor
and sophistication required is the basis of this work and is considered in detail in the body of this
report. q 2001 Elsevier Science B.V. All rights reserved.

Keywords: Pyrolysis; Hydrocarbons; Olefins

1. Scope

Pyrolysis Žhigh temperature steam cracking. of hydrocarbons to olefins is the largest


and most diverse petrochemical process. Feedstocks include ethane, propane, butane,
light naphthas, light and heavy gas oils, and hydrocracked products. Products include
H 2 , CH 4 , ethylene, propylene, butadiene, benzene, toluene, xylenes, aromatic gasoline

)
Tel.: q1-281-392-6057.

0378-3820r01r$ - see front matter q 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 8 - 3 8 2 0 Ž 0 0 . 0 0 1 4 9 - 1
2 J.E. Gwyn r Fuel Processing Technology 70 (2001) 1–7

Žpygas., aromatic gas oils and residues, and numerous other components. Pyrolysis
Conditions range from: 5908C process gas temperatures Žto 10408C tube temperatures.,
0.1 to 4 s residence times, 1 to 4 atm of pressure, 2 mm to 15 cm reactor tubular
diameters. Many yield models have been developed over the last 6 q decades. Rice w1x
developed the free radical reaction concepts between 1931 and 1943. Zdonik et al. w2x, in
1970, reported many useful concepts including primary products from individual
components, K 5 theta Žintegrated K 5 du along the reactor., heats of reaction, heat
balance and calculation procedures. Sundaram and Froment w3x presented a mechanistic
model in 1977. Dente et al. w4x developed the first model in 1979 that, with minor
exceptions, met the AUniversalB requirements. Froment w5x reviewed the status of
pyrolysis and modeling in 1981. A limited number of references are quoted because
many different schemes are not compatible and some of the intimate detail w4x is
available only through secrecy agreement with model licensees. Since then other
models, using several different approaches, have been developed to better meet these
requirements. AUniversalB models are required to predict yields to specified accuracy
over the full range of the feeds, reactors and conditions. Models may be expanded to
include other processing factors such as coking of pyrolysis tubes, run lengths, optimiza-
tion of conditions and other strategies.

2. Feedstock

Feedstock selection is the single most important factor in product yields. For
example: ethane yields only ethylene and hydrogen at low conversions; for other
paraffins, propylene is also produced in amounts depending upon conditions and the
alkyl branching in the feed; and feeds containing aromatic nuclei will yield a wide range
of aromatic products.
Therefore, in a AUniversalB yield model, feedstocks must first be characterized to
completely identify and quantify the product components’ precursors. The problem is
that a complete description is impossible for some feeds. This is due to the sheer number
of individual components that exist and the inability to separate and identify them by
analyses. Fortunately, higher carbon components have similar compositions and crack-
ing behavior as the next higher or lower carbon number group. It is possible to lump
these components and compositions into a manageable number of pseudo-components.
Composition by individual hydrocarbons is required through C6s. Where they can not
be separated by available analytical techniques, they are lumped Žparticularly in the C6
iso and normal paraffins.. In the C7 through the, say, C12 range, each individual carbon
number is lumped by type, i.e., normal paraffins, isoparaffins, naphthalenes and
aromatics. Higher boiling components require lumping by carbon number. One such
lumping is C13–C15, C16–C20, C21–C25, C26–C30 and C31 q . Each lump is further
subdivided into precursor types. In addition, to general type compositions indicated
above, it includes functional forms including normal and iso-akyl branches on the
aromatics, nuclei structure of the aromatics Žmono, di, and condensed..
Mechanistic pyrolysis models w4x achieve a AUniversalityB by lumping into more than
100 feed and product components or equivalent component. The molecular kinetics
J.E. Gwyn r Fuel Processing Technology 70 (2001) 1–7 3

models and correlation models also require individual component composition through
C5s and some C6s. They have coarser lumping of liquid components into 35–50 feed
and product components.
When feedstock composition is not available from analyses or library data, it may be
satisfactorily estimated by inspection properties and correlations previously derived from
feeds with detailed analyses. The usual inspection properties are density, refractive
index, TBP distribution and elemental analysis ŽH, C, O, S, N.. The correlations are as
follows:
M s f Ž d, T50 . ,
where M s mol. wt.
HrC s f Ž n 20
D , d, M . .

These functions vary slightly among investigators.


Then P, N and A Žparaffin, naphthene and aromatic, respectively. for the total feed
are obtained from the following three relationships:
1rd s Prd P q Nrd N q ArdA Ž volumes are additive. ,
) ) )
H s P H P q N H N q A HA ,
and
1 s P q N q A.
The individual cut fractions, i, are resolved in the same manner for Pi , Ni and A i ,
where,
d P i , d N i , dA i s f Ž M . ,
H pi , H ni , Hai s f Ž M . ,
and the balance is maintained such that,
P s P1 q P2 q . . . qPn ,
N s N1 q N2 q . . . qNn ,
A s A 1 q A 2 q . . . qA n .

3. The yield models

The Free Radical Model handles a kinetic scheme of about 2000 reactions w4x. They
involve: Ž1. Chain initiation Žformation of radicals.; Ž2. H-abstraction; Ž3. Radical
addition; Ž4. Radical decomposition; Ž5. Radical isomerisation; Ž6. Chain termination;
and Ž7. Molecular. Radical reactions are extremely fast and considered to be in Asteady
stateB equilibrium of insignificantly low concentrations. The reactions are reduced to
about 500 by summing coefficients of equivalent reactions into one.
The Molecular Kinetics Model w6,7x assumes that each feed component pyrolyzes by
first order kinetics to a predefined product slate Žalso a function of temperature.. Each
4 J.E. Gwyn r Fuel Processing Technology 70 (2001) 1–7

has it own basic rate constant and energy of activation. Each product component is also
a reactant to some degree and is included in the overall reaction scheme. These may
include bimolecular and condensation reactions. For example, the pyrolysis of the heavy
gas oil feed component goes directly to a light gas oil Ždifferent composition than light
gas oil in the feed., naphtha, gaseous components and a residual, non-reactive, aromatic
heavy gas oil. The yield distribution matrix is sparse for the pure component region and
is dense for the liquid feed components. The number of distribution coefficients is about
50, but each has an activation coefficient different from the base.
The Methane SeÕerityr Yield Correlation Model considers that methane is produced
by the cracking of feed hydrocarbon Žprimary methane. and by the further reaction of
the primary product hydrocarbons to yield secondary methane. The total methane is:
C1 s X 1F Ž 1 y X F . q X 1S Ž X Pf y X Pr . ,
where: X 1F is the fraction of converted feed going to methane; X F is the unconverted
feed remaining, Ž1 y X F . s feed conversion; X 1S is the fraction of further converted
product going to methane; X Pf is the total primary hydrocarbon formed; X Pr is the total
primary hydrocarbon remaining.
Hence, there are two pre-exponential coefficients and two activation energy coeffi-
cients that are correlated with feed composition and properties. Also, the fractions of
methane from converted feed and from converted products are correlated with feed
properties. Heats of reaction and average molecular weight of the unreacted feedrprod-
uct mix at any point in the reactor are correlated with the methane yield at that point and
the feed properties. This allows the calculation of volume flow throughout the reactor.
Product yields are calculated only at the outlet of the reactor by correlation to the
outlet methane yield. For a given feed the these product yields are correlated through
linearized and other invariant functions. For example, the propylene to methane ratio is a
unique linear decreasing value when plotted versus methane yield. Any deviant data
point represents a bad run! Similarly, ethylene yields are nearly proportional to methane
Žwith some minor adjustments.. Each of the other yield components is correlated by
simple functions of ratios of components, or of feed conversion, etc. Material, H and C
balances are forced. Naphtha is usually the AslackB component. The parameters for all
these functions are, in turn, correlated with feedstock properties or stored in a library for
a large number of feeds that have been tested. In this regard, the methane severity model
is valid for all feed stocks, reactors and conditions as required to be AuniversalB!

4. Model formulation and reactor mechanics

The Models have two operating modes. The first is for the model developer in order
to fine tune the internal constants and modify reaction schemes. This involves multiple
case runs, say 20 cases, in which calculated individual component yields are compared
with experimental values. A strong convergence routine corrects the yield coefficients
and the rate constants and repeats the procedure two or three times to get Abest fitB
values. The second mode is for users to run case studies for feedstock selection, effect of
outlet temperature, etc.
J.E. Gwyn r Fuel Processing Technology 70 (2001) 1–7 5

In either mode for any of the models it is necessary to determine the profiles of
temperature, pressure and volume flow throughout the reactor. These change very
rapidly and can only be determined by assigning inlet and outlet conditions; integrating
the rate equations through the reactor; and adjusting unfixed conditions, say inlet
pressure, radiation temperature, or flow friction factor, to give closer values and by
repeating the process three or four times to convergence. The Free Radical and the
Molecular Kinetics models calculate yields at each integration step. These two models
now first use the Methane Yield or another simplified model to converge more rapidly
on conditions. The detailed model is then run for yields and usually converges in a
single pass.
For all the models it is required that the following balances be maintained.
Ž1. Material balances for all components and hydrogen and carbon Žonly molecular
weight for Methane Model..
Ž2. Energy balances include sensible heat, heat of reaction, and radiation flux. This
requires radial calculation for each reactor step element including: radiation transfer
from the flames and furnace walls to the tube outside surface; conduction to the inside
tube wall, conduction transfer across the coke layer, if any, and heat transfer to the
process stream by convection.
Ž3. Mechanical energy balance including friction losses and momentum changes.

5. Model program flow structure

The yield model uses the following scheme: Ž1. Data input includes coil dimensions,
fixed conditions Žusually outlet temperature and pressure., feedstock ID and flow rate,
steam rate, other hardware data as needed; Ž2. Detailed feedstock composition is either
available from files or calculated from inspection properties; Ž3. Preliminary estimates
of uncontrolled variables are obtained from other cases, plant data or correlations; Ž4.
Convergence of conditions is accomplished by integration of a simplified ŽMethane.
model through the reaction, adjusting the unfixed conditions and repeating the procedure
until converged; Ž5. Convergence is repeated with the rigorous model of choice Žusually
one pass is sufficient.. The Methane model skips this step; Ž6. Calculate useful
information including: component yields for the Methane model, selective product
properties such as aromatic content and octane number for pygas, coking rates; etc; Ž7.
Print out case results including yields, profiles, properties, etc.

6. Modelers’ dilemmas

As complex as pyrolysis is, no two separate persons or teams will arrive at exactly
the same model. For example, decisions must be made on integration step size through
the reactor tubes. Some reactions are orders of magnitude faster than others. Small step
sizes are required until the fast reactionŽs. near completion and then the step may be
lengthened for the next fastest reactionŽs. and so forth. This may be automated through
application of integration routines such as the Fehlberg’s Runge–Kutta 4–5 method with
6 J.E. Gwyn r Fuel Processing Technology 70 (2001) 1–7

adaptive step size control for reducing integration error. Also it is pragmatic to have the
modeler in the loop. If profile data are to be saved at certain points along the reactor Žfor
printing out. then logic has to be included so that these points are not overstepped.
When the outlet of the reactor is reached on the first integration, the outlet conditions are
not normally met within specified tolerances. So logic is applied to adjust appropriate
variables to give a better fit. After the second integration some strong convergence
techniques can be applied to converge in one or two more iterations.
When rate constants are being evaluated, then data from numerous runs, say 20, are
required for AuniversalB fits. Calculated yields and conversions are compared with
experimental values and pre-exponential and exponential rate constants are adjusted
through appropriate logic. Conversion is achieved in two or three iterations. Printout
routines are usually included to show calculated versus experimental yields for every
data run case for each component. Deviates from the trend are identified and resolved.
A simple stepwise integration has been applied over, say, 100 steps with a resulting
slight bias in the calculated yields and conditions that results in a corresponding slight
bias in the rate constants. Simple stepwise integration using delta lnŽ C j,Iq1rC j,I . instead
of delta C j s C j,iq1 y C j,i can yield quantitatively accurate solutions w6,7x and does not
allow negative yields. Conversion on rate constants can similarly be speeded and
improved by the least squares best fit of the linearized forms of the pre-exponential and
exponential constants.
A major dilemma is to provide accurate data for model development. Accurate flow,
condition, and material balance measurements are required. Pilot plant data w8x that
simulate commercial plant operation Žpressure and temperature profiles. are valuable.
Pure component, blended feed, and a range of plant feed yield data are required.

7. Model users’ requirements

The user has many other requirements for the model to be useful. The yield model
starts with the vaporized feed at the reactor entrance and ends with the yield structure at
the reactor outlet. Preheating and superheated steam vaporization of the feed has been
developed; also, quenching of the products, without fouling the quench or fractionator,
has been developed. Pyrolysis tube and quench coking models for calculating run
lengths have been included in most yield models. Detailed furnace and convection
section heat transfer models have been included in plant control and feedstock optimiza-
tion models. Simplified severity models have been based upon coil outlet temperature,
temperature profile convexity, PropylenerMethane ratio, ultimate ethylene yield, etc.
The yield models have been applied to coil design, de-bottle necking, and other
technology advances.

8. Conclusions—accuracy and choice of yield models

The different models have reached the point of diminishing returns for accuracy and
useful features. The accuracy for major product components is about 0.05%w basis feed
Žabout 1% of the value.. A few components such as 1,2-butadiene and butene-1 are
J.E. Gwyn r Fuel Processing Technology 70 (2001) 1–7 7

accurate to about 0.1%w. Minor products such as isoprene and pentene-1 are calculated
to within about 0.02%w. Liquid products such as naphtha are less accurately predicted
Ž0.1–0.5%w.. Molecular Kinetics and Methane models are slightly less accurate than the
Free Radical model. Indeed, Free Radical case calculations have been used to generate
Anoise-freeB data for the other models Žespecially for pure component data..
High-speed, high storage capacity computers have allowed the mechanistic model to
be used for almost all applications. The other models may still find use in large
applications that require simple model forms.

References

w1x F.O. Rice, et. al., JACS, numerous from 53,1931 to 63, 1943.
w2x S.B. Zdonik, E.J. Green, L.P. Hallee, Manufacturing Ethylene, Petroleum Pub., Tulsa, OK, 1970.
w3x M. Sundaram, G.F. Fremont, Chem. Eng. Sci. 32 Ž1977. 601.
w4x M. Dente, E. Ranzi, A.G. Goossens, Comput. Chem. Eng. 3 Ž1979. 61–75.
w5x G.F. Froment, Chem. Eng. Sci. 36 Ž1981. 1271–1282.
w6x J.E. Gwyn, APyrolysis Yield Modeling: A Radiant Heat Flux Model Specific to a Bench Scale UnitB,
AIChE San Francisco Mtg. ŽNov. 25–30, 1984..
w7x J.E. Gwyn, Steam cracking of hydrocarbons to olefins,13th Int’l. Symp. on Chemical Reaction Eng.,
Baltimore, MD, September 25–28, 1994.
w8x P.S. Van Damme, G.F. Froment, IEC Process Des. Dev. 20 Ž1981. 366–376.

You might also like