Modeling and Simulation of Steam Crackers

Tivadar Gál (Tisza Chemical Works Co. Ltd., Tiszaújváros, Hungary)*

Béla G. Lakatos (Department of Process Engineering, University of Veszprém,
Veszprém, Hungary)

Abstract
Steam cracking of hydrocarbons (pyrolysis) is the main source of principal feedstocks
in petrochemical industry. Because of the relatively high price of the so called
“traditional” European cracker feeds (mainly naphta and gas oil), as well as the fact that
a demand-increment could be realized in last decade for monomers, made it necessary
to search for new feedstocks. The economics associated with different feeds and
effluents of the crackers are very large and continuously changing, which need a careful
choice of feeds and operating conditions. The mathematical models and simulation can
help in choosing the feeds and optimize operations. One of the existing furnaces at TVK
Co. Ltd. /Tiszai Vegyi Kombinát RT., Hungary/ processes the recycled and
hydrogenated C4/C5-cut. The aim of this work was to investigate kinetics and product
distribution in cracking of a C4/C5-mixture under conditions representative for
industrial scale. The results show a good coincidence with the laboratory analysis.

Keywords: hydrocarbons, pyrolysis, C4/C5 – cut, modeling, simulation

1. Introduction
The main objective of a mathematical model is to accurately simulate the system.
Namely, to predict the process behavior both inside and outside the range of
experimental data. Increasing flexibilities of the reactors are claimed both for the
effluent distribution and for the feedstock utilization. Mathematical models and
simulation can help in solving these problems and optimize operations. Costly and time-
consuming experimental programs can often be minimized or even avoided.
Completeness and complexity of the models proposed recently is related to the
evolution of computation facilities that permit handling of much larger numerical
problems. On the basis of these models, new types of cracking reactors /furnaces/, as
well as highly developed process control systems were developed in the last decade,
which make possible a reliable prediction of effluent distribution.

2. Modeling on Industrial Scale

Mathematical modeling of a process is, in general, based on at least three
considerations:

*
Author to whom correspondence should be addressed: [email protected]
 1.) phenomenological description of the process system, which has to be as
accurate and complete as possible
2.) translation of the chemical and physical steps into mathematical terms
3.) numerical methods to solve the set of equations and conditions
In our examined case three streams of C4 and C5 cuts are collected, mixed and
hydrogenated then re-pyrolysed in the Olefin Unit of TVK. Mixing of purchased n-
butane and C5-condensates into the furnace feed is also a practice. In conclusion, the
feedstock composition may vary with time significantly, so it is very important to study
the operating conditions in order that the process can be controlled successfully.
Furthermore, an overall kinetic equation is not sufficient for the purpose since the
prediction of product distribution may be carried out merely based on a detailed set of
kinetic equations of reactions, yielding the products of interest.
This furnace has been selected for the following reasons: the number of hydrocarbons
that can be found in the feed is relatively small, the number of species and elementary
reactions can be identified relatively easily, co-cracking of purchased n-butane and/or
C5-condensates is a practice and examining the possibilities of propane and/or ethane
co-cracking.
The elementary reaction schemes have been derived from the observed input and output
compositions, radical mechanisms and thermodynamic principles. A number of
parameters were taken from the published literature data whilst those remaining were
estimated by using a non-linear regression procedure based on the steady-state
equations of the model. The form of the model was based on the material- and energy-
balances formulated in terms of the plug flow reactor model [1,2]:
∂ci ( x, t ) N r ∂c ( x, t )
= ∑ α ki r k (c, T ) − v i , i = 1 → N c , k = 1 →N r (1)
∂t k =1 ∂x
 Nc  ∂T ( x, t ) N r  Nc  ∂T ( x, t )
 ∑ C pi ci  = ∑ (−∆H k )r k (c, T ) − v ∑ C pi ci  + Q ( x, t )
 i =1  ∂t k =1  i =1  ∂x
(2)
subject to the following initial
ci ( x,0) = ci 0 ( x) i = 1 → N c , T ( x,0) = T0 (0) (3)
and boundary conditions:
ci (0, t ) = ciin (t ) i = 1 → N c , T (0, t ) = Tin (t ) (4)
(ci – concentration; rk – reaction rate; v – velocity; Cp – heat capacity; T – temperature;
∆Hk – reaction heat; Q – heat supply)
A fragment of the elementary reaction scheme is presented in Table 1.
Most reactions involve radicals, but some purely molecular reactions play a significant
role. It was observed [2] that the exclusion of molecular reactions that occur
simultaneously with radical reactions has been responsible for some misleading
conclusions of rate parameters, particularly for olefins and diolefins. Decompositions of
radicals are performed according to the so-called β-rule, namely, the second C-C bond
is broken counted from the end of the hydrocarbon chain. [6] In fact, once the values for
the frequency factor and the activation energy for the reference reaction have been
obtained experimentally and found to be in accordance with the theoretical predictions
of thermo-chemical kinetics, analogous reactions can be predicted with little or no
experimental investigations. It was stated [4] that the yields of propylene and higher
olefins show a maximum at certain conversion values. The supposed reason is that
propylene and higher olefins slow down the further reactions because the chain-
forwarder radicals react with them by addition. [5] This fenomenon is also named auto-
inhibition.
The compositions of feeds and cracked gases were measured by laboratory- and process
chromatographs. The furnace outlet temperatures were ranged between 835-845 oC,
while the dilution steam ratios were between 0.39-0.41 kg steam/kg hydrocarbon. The
furnace outlet pressures were of 0.5-0.7 bar, while the pressure drop along the radiant
coil was ranged between 0.8-1.1 bar.
The basic parameters of the examined furnace can be seen in Table 2.

Table 1: A part of the elementary reaction scheme with the kinetic data
Nr. Reaction Ao (1/sec) E (kcal/mol) Source
16
1. C2H6 > 2CH3* 4.0 x 10 87.5 2
16
2. C3H8 > 2C2H5* + CH3* 2.0 x 10 84.5 2
16
3. n-C4H10 > 1-C3H7* + CH3* 1.5 x 10 82.1 2
…
8
12. C2H4 + H* > C2H3* + H2 8.0 x 10 4.0 2
…
16
101. n-C5H12 > C3H7* + C2H5* 1.1 x 10 80 estimated
…
14
142. 2-C6H12 > 2-C5H9* + CH3* 9.0 x 10 62 estimated
…

Table 2: Operating parameters of the simulated furnace

Parameter SOR EOR Unit
Hydrocarbon feed 19400 19400 kg/h
Inlet pressure 3.98 4.02 bar(g)
Dilution steam 7800 7800 kg/h
Inlet temperature 106 106 degC
COT 835 840 degC
Cracked gas temperature after TLE 354 450 degC
BFW quantity 20800 20600 kg/h
Flue gas cross-over temperature 1160 1250 degC
HP steam temperature 487 502 degC
3. Simulation with Chemcad – Comparison of Simulated Results
with Lab Analysis
During theoretical examinations 224 elementary reactions and 53 species were
identified. As the aim of the steady-state and dynamic simulation tests it was to be
presented which parameters can the desired optimum yields be reached by, from the
feedstocks available. On basis of its sufficiently large thermochemical and kinetic
database, the ChemCAD software was selected for the computer-aided simulation.
Applicability of the ChemCAD simulator was also checked by re-simulating the
pyrolysis of n-butane, a model published by Sundaram and Froment [3]. The ethylene
yields, for example, are practically identical at each conversion for both the published
and ChemCAD-simulated models. Though there are minor differences in the yields of
other products, the simulator was considered to be suitable for the purpose.
For simulation of the existing pyrolysis furnace, it was divided into zones. A different
zone started when there was a modification in the cross-sectional area of the reactor.
Since the temperatures entering and leaving the radiant section are well known, a
temperature profile was to be built up along the radiant coil. Though this profile can
also be created by the simulator itself, it was verified by the thermal-balance equations
as well. Using the standard kinetics, the published [3] and estimated kinetic parameters,
the temperature profile and coil geometry were dedicated as input data for simulation.
The results are shown in Fig.1 and 2 in comparison with the laboratory analysis.

40

30

Yie lds, w t % 20

10

0
1 2 3 4
Sample number

propylene simulated yields propylene measured yields

ethylene simulated yields ethylene measured yields
residual n-butane simulated residual n-butane measured

Figure 1: Comparison of simulated and measured yields at different feed compositions

 Ethylene yields at different parameters

35
Ethylene yield, wt%

30 simulated
measured
25

20
T=835C T=840C T=845C p=1.2bar p=1.5 bar p=1.8 bar

Temperatures and pressures

Figure 2: Comparison of ethylene yields at different operating parameters

Although the simulated yields of basic products /ethylene and propylene/ show a good
coincidence with the experimental data /cracked gas analysis by the lab/ the sensitivity
analysis of the model is still in progress.
Since a large number of kinetic data were to have been estimated /namely, for 90
elementary reactions/, a system of 180 parameters are to be mathematically determined
in order that their values can be specified as accurately as possible. Of course, many
simplifications were also possible due to the similarities existing between the reactions.
On basis of these ideas, the values of those parameters /activation energies and pre-
exponential factors, respectively/, which could not be found in published literature,
were estimated by a “try-and-error” procedure.

4. Summary
The good match between the experimental and simulated results indicates large
opportunities of applying the simulation program in cracking of gaseous feeds and their
mixtures. After a longer control period the model will also be applied for online
simulation as a tool for parameter optimization. In order that the furnace in question and
its feed-stock system can be treated as a whole, building up the model of hydrogenation
reactor is also intended.
As far as the further examinations are concerned, the different co-cracking possibilities,
opportunities for parameter-optimization, as well as the integration of the model into the
existing APC /Advanced Process Control System/ are still to be investigated.
References
Froment, G.F. ,B. Bischoff: Chemical Reactor Analysis and Design; Wiley & Sons,
1990.
Dente, M.E., M. Ranzi: Mathematical Modeling of Hydrocarbon Pyrolysis Reactions.
Academic Press 1983.
Sundaram, K.M., F. Froment, Ind. Eng. Chem. 1978.
Rice, F.O., O.L. Polly: Chem. Phys. 6, 273 1938.
Zdonik, S.B., E.J. Green, L.P. Hallee: Oil and Gas Journal, 26,96 1967.
Rice, F.O., K.F. Herzfeld: J. Am.Chem. Soc. 74, 324 1954.