Você está na página 1de 10

Computers and Chemical Engineering 35 (2011) 28762885

Contents lists available at ScienceDirect


Computers and Chemical Engineering
j our nal homepage: www. el sevi er . com/ l ocat e/ compchemeng
Modelling and dynamic optimization of thermal cracking of propane for
ethylene manufacturing
Mehdi Berreni, Meihong Wang

Process Systems Engineering Group, School of Engineering, Craneld University, MK43 0AL, UK
a r t i c l e i n f o
Article history:
Received 8 September 2010
Received in revised form21 April 2011
Accepted 13 May 2011
Available online 20 May 2011
Keywords:
Mathematical modelling
Dynamic optimization
Tubular reactor
Case study
Ethylene
Thermal cracking
a b s t r a c t
In tubular reactors inside a cracking furnace, heat transfer, thermal cracking reactions and coke buildup
take place and closely interact with each other. It is important to understand the process and optimize
its operation. A 1-dimensional (1D) pseudo-dynamic model was developed based on rst principle and
implemented in gPROMS

. Coke buildup inside the tube wall was also accounted for. The model was
validated dynamically. The impact of process gas temperature prole, and constant tube outer wall tem-
perature prole on product yields and coking rate are assessed. Finally, dynamic optimization was applied
to the operation of this tubular reactor. The effects of coking on reduction of production time and the
decoking cost have been considered. The tube outer wall temperature prole and steam to propane ratio
in the feed were used as optimization variables. Dynamic optimization investigation indicates that it can
improve operating prot by 13.1%.
2011 Elsevier Ltd. All rights reserved.
1. Introduction
Ethylene is one of the major building blocks in the petro-
chemical industry because it is a very reactive intermediate to
produce plastics, resins and bers. Thermal cracking of naphtha,
ethane or propane is the most widely accepted technology to pro-
duce ethylene. Thermal cracking of propane in tubular reactor will
be the topic of this paper. The main advantage of using propane as
feedstock is that the costs for the separation before and after the
cracking process are lower compared to thermal cracking of naph-
tha, moreover the total yield for ethylene and propylene is higher
compared to thermal cracking of ethane (Sundaram, Shreehan, and
Olszewski, 2001).
The reaction from propane to ethylene is endothermic. There-
fore it needs lots of external energy. The furnace is made up of
two main sections. The mixture of steamand propane is preheated
in the convection section. It is then introduced into the radiation
sectionwhichconsists of longtubular reactors for thecrackingreac-
tions to take place. Floor and/or side burners apply a high heat ux
to the tube outer wall so that a suitable process gas temperature
prole can be achieved. During operation, coke is formed inside the
tube wall and the reactor must be put ofine regularly to remove
the coke buildup. In order to reduce the coking rate, steamis mixed
with the propane as diluents (Sundaramet al., 2001).

Corresponding author. Tel.: +44 1234 754655; fax: +44 1234 754685.
E-mail addresses: meihong.wang@craneld.ac.uk,
wang 2003 uk@yahoo.co.uk (M. Wang).
1.1. Literature review
This section is to critically assess research progress in different
aspects of thermal cracking for ethylene production.
1.1.1. Thermal cracking reaction mechanism
Thermal cracking of propane proceeds via free-radical reactions.
The reaction mechanism has been studied and improved several
times by different researchers. Zou, Lou, and Liu (1988) proposed a
reduced free-radicals reaction scheme for the pyrolysis of propane
which includes ten reactions. Three main reaction types namely
initiation, propagation and termination were included in this free-
radical reaction mechanism. Due to numerical reasons, molecular
reaction schemes have been widely used in simulations to approx-
imate the behaviour of the reactions. Based on experiments from
Van Damme, Narayanan, and Froment (1975) and Sundaram and
Froment (1977) developed a complete molecular scheme including
9 reactions and 9 species. The same authors also investigated the
kinetics of coking rate (Sundaram and Froment, 1979). They con-
sidered propylene as the only precursor and two reactions were
added to their previous reaction scheme to account for the coke
formation.
1.1.2. Steady state modelling and steady state optimization
Shahrokhi and Nejati (2002) developed a 1Dsteady state model
of thermal cracking of propane. Their aim was to determine the
optimal process gas temperature prole based on the molecular
reaction scheme of Sundaram and Froment (1979). They consid-
ered both ethylene and propylene production using an objective
0098-1354/$ see front matter 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compchemeng.2011.05.010
M. Berreni, M. Wang / Computers and Chemical Engineering 35 (2011) 28762885 2877
Nomenclature
Bend angle [

]
H
r,j
(z) Reaction heat [kJ/kmol]
(z) Bend factor [-]

i,j
Stoichiometry coefcient [-]
(z) Coke thickness [m]

c
Coke density [kgm
3
]

c
Conduction heat transfer coefcient of coke

t
(z) Conduction heat transfer coefcient of tube wall
[kJ/(s mK)]
A
c
Pre-exponential factor for coking reaction
[kgm
3
/(kmol m
2
s)]
A
j
Pre-exponential factor (kmol
1
m
3
s
1
for a 2nd
order reaction or s
1
for a 1st order reaction)
C
C
3
H
6
(z) Molar concentration of propylene [mol/m
3
]
C
i
(z) Mass concentration [kgm
3
]
Cpm
i
(z) Specic heat capacity [kJ/(kgK)]
Di Internal diameter of the pipe [m]
Dt Outer diameter of the pipe [m]
Ec Activation energy of coking reaction [kJ/kmol]
Ea
j
Activation energy [kJ/kmol]
F
0
Outlet mass owrate of propane [kg/s]
F
i
Outlet mass owrate of desirable product i [kg/s]
F
i
(z) Mass owrate [kg/s]
F
r
(z) Friction factor [m
1
]
F
s
Intlet mass owrate of steam[kg/s]
NC Number of components [-]
NR Number of reaction [-]
n
ij
Reaction order [-]
N
steam
Molar owrate of steam[kmol/s]
P(z) Pressure [Pa]

Q(z) Heat transfer rate [kJ/s]

Q
t
Total heat transfer rate [kJ/s]
R Ideal gas constant [kJ/(Km)]
R
1
(z) Thermal resistivity relative to gas convection
[K/kW]
R
2
(z) Thermal resistivity relative to coke conductivity
[K/kW]
R
3
(z) Thermal resistivity relative to tube conductivity
[K/kW]
Rb Tube bend radius [m]
Re(z) Reynolds number [K]
r
c
(z) Coking rate [kg/(m
2
s)]
r
j
(z) Reaction rate [kmol/(s m
3
)]
T(z) Process gas temperature [K]
T
c
(z) Coke surface temperature [K]
T
we
(z) Tube outer wall temperature [K]
T
wi
Tube inner wall temperature [K]
function that calculates operating prot. Masoumi, Sadrameli,
Towghi, and Niaei (2006) developed a 1D steady-state model for
tubular reactors for thermal cracking of naphtha. A free-radical
reaction scheme including 90 species and 543 reactions was used.
A steady-state optimization study was carried out in an effort to
maximise operating prot.
Gao, Wang, Ramshaw, Li, and Yeung (2009) applied steady-state
optimization for naphtha cracking based on steady state model in
Aspen HYSYS. The results were very encouraging since the opti-
mization improved the operating prot signicantly. Gao, Wang,
Ramshaw, et al. (2009) assumed the process to be under steady-
state conditions, but in reality, the coke buildup takes place with
time andaffects performance. Moreover, they neglectedthe impact
of the coke thickness on heat transfer coefcient.
Ethylene furnace was investigated with 3D model using com-
putational uid dynamics (CFD) technique by Detemmerman and
Froment (1998) and Lan, Gao, Xu, and Zhang (2007) respectively.
The authors investigated the inuence of the position of the gas
burners as well as the ue gas patterns in the rebox. The studies
were carried out for process design purposes.
1.1.3. Dynamic modelling and dynamic optimization
Shahrokhi and Nejati (2002) also developed 1D dynamic model
for the PFR. By adding the heat transfer from furnace wall to the
PFR in the dynamic model, they evaluated the performance of a PID
controller for furnace wall temperature control in thermal cracking
of propane. Gao, Wang, Pantelides, Li, and Yeung (2009) developed
adetailed1Ddynamic model ingPROMS

for naphthacracking. The


model includes dynamics inmass balance, energy balance and coke
buildup, then steady-state optimization was applied for optimal
operation.
GhashghaeeandKarimzadeh(2007) studieddynamic modelling
of thermal cracking furnaces which is capable of describing and
predicting the unsteady-state behaviour of crackers during start-
up. The mathematical model includes four sub-models: convection
model, tubular coil, rebox model and tube skin model. Based on
these sub-models, the authors developed a two-dimensional (2D)
zone model to predict fuel consumption, heat transfer rates and
cracking temperatures during start-up.
Dynamic optimization is dened as the establishment of the
best procedure for correcting the uctuations in a process (Babu &
Knovel, 2004). Sowers and Reed (2001) studied the dynamic opti-
mization of ethylene furnaces cracking propane using the software
SPYRO
TM
. Most particularly they focused on the optimal conver-
sion prole and owrate as a function of time in order to maximize
operating prot. The model they used was very simple in com-
parison to the one that is developed in this paper. They observed
12% improvement in cumulative prot compared with steady-
state optimization.
1.2. Scope of this study
Only the radiant section of the cracking furnace will be consid-
ered since this is where the reactions take place. No downstream
separation and quenching are taken into account. This paper pro-
vides detailed 1D mathematical modelling of the tubular reactors
based on rst principles. The molecular reaction scheme for ther-
mal cracking of propane and the coke formation developed by
Sundaram and Froment (1979) is used. The model also takes into
account the impact of coking on the product yield and on the run
length(i.e. thetimebetweentwoconsecutivedecokingoperations).
Themodel canpredict accuratelytheproduct yields, cokethickness,
runlength, pressure dropandheat transfer rate. The impact of vary-
ing process gas temperature on the furnace operation is assessed.
This is followed by investigating the impact of applying a constant
tube outer wall temperature prole. Finally dynamic optimization
was carried out with operating prot as objective function. Tube
outer wall temperatureproleandsteam-to-propaneratioareused
as optimizationvariables. The overall aimof this paper is to provide
insights for optimal operations in the ethylene industry.
1.3. Novelties
The novel contributions of this paper include (a) 1D pseudo-
dynamic modelling of the tubular reactor and model validation
from different aspects such as product yields, pressure drop, tube
outer wall temperature prole and their changes with produc-
tion time (i.e. dynamic validation); (b) dynamic optimization based
on the model developed. The use of tube outer wall temperature
2878 M. Berreni, M. Wang / Computers and Chemical Engineering 35 (2011) 28762885
prole as one of the optimizationvariables makes the optimal oper-
ation study more realistic.
1.4. Outline of the paper
Mathematical modelling is presented in Section 2. Then an
appropriate benchmark is introduced in Section 3, and the model is
validated dynamically by comparison to industrial data after 100,
300 and 700h of operation. Two case studies are performed in Sec-
tion 4. Section 5 focuses on dynamic optimization of the process.
The optimization results have also been discussed in great detail.
Conclusions are drawn in the end.
2. Mathematic modelling
2.1. Assumptions
Inorder todevelopthe1Dpseudo-dynamic model, thefollowing
assumptions were made:

Negligible radial gradient (i.e. variation only along Z-axis)

Inertness of steamin the furnace

Only vapour or gas inside the PFR (i.e. no liquid)

All cracking reactions take place in the tubular reactors only (i.e.
cracking reactions start at the entrance of the reactor and stop at
the exit)

Nohydrodynamic or thermal entrance regioneffects (i.e. noinu-


ence fromthe environment)

Only the coke buildupis considereddynamically (i.e. heat or mass


accumulation inside the reactor is ignored) and coke thickness
keeps constant within each hour

Ideal gas behaviour is assumed only when component concen-


trations are calculated (i.e. relevant with Eq. (2) only)
2.2. Material balance
The amount of coke accumulated is tiny compared to the large
mass owrate of propane and/or products. The residence time for
the PFR is also very short. Consequently, the change of concentra-
tion with time can be neglected. Steady-state material balance is
written as follows:
F
i
(z)
z
=
NR

j=1

i,j
r
j
(z) M
wi

(Di 2(z))
2
4
(1)
The molar concentration is calculated based on ideal gas law.
C
i
(z) =
_
F
i
(z)/M
wi
N
steam
+

NC
i=1
(F
i
(z)/M
wi
)
_

_
P(z)
RT(z)
_
(2)
The reactionscheme usedinthis paper (developedby Sundaram
and Froment, 1977) assumes that the reactions are elementary.
Therefore, thereactionrates aredeterminedaccordingtoArrhenius
law.
r
j
(z) = A
j
e
Ea
j
/RT(z)

NC

i=1
C
i
nij
(z) (3)
Coking rate is calculated based on Arrhenius law as well
(Sundaramand Froment, 1979).
r
c
(z) = A
c
e
Ec/RTc(z)
.C
C
3
H
6
(z) (4)
As for the coke thickness (Fig. 1), it is determined as following
(Sundaramand Froment, 1979).
d(z)
dt
=
r
c
(z)

c
(5)
2.3. Energy balance
The steady-state energy balance on the tubular reactor with
respect to differential length can be written as:
NC

i=1
F
i
(z)Cpm
i
(z)
T(z)
z
=

Q(z)
L
+
(Di 2(z))
2
4

NR

j=1
r
j
(z) (H
r,j
(z)) (6)
Similar to an electrical system, the gaseous phase, the coke
thickness and the tube thickness can be assimilated to thermal
resistances R
1
, R
2
and R
3
respectively as shown in Fig. 1. The overall
heat transfer can be written as:

Q(z) =
1
R
1
(z) +R
2
(z) +R
3
(z)
(T
we
(z) T(z)) (7)
2.4. Momentumbalance
The steady-state momentumbalance is determinedas following
based on Shahrokhi and Nejati (2002), Masoumi et al. (2006), and
Sundaramand Froment (1979).
P(z)
z
=

_
1
Mm(z)
_
)/z +
1
Mm(z)

_
1
T(z)

T(z)
z
+Fr(z)
_
1
Mm(z)P(z)

P(z)
G(z)
2
RT(z)
(8)
The friction factor is calculated in Eq. (9). For straight section,
the bend coefcient is 0, and for bended section, it is calculated
with Eq. (10). Both equations were extracted from Shahrokhi and
Nejati (2002) and Masoumi et al. (2006).
Fr(z) =
0.092 Re(z)
0.2
Di 2(z)
+
(z)
Rb
(9)
(z) =
_
0.7 +
0.35
90
_
_
0.051 +0.19
Di 2(z)
Rb
_
(10)
2.5. Physical properties
Physical properties such as dynamic viscosity, mass density and
enthalpies were calculated with Multiash

using PengRobinson
equation of state. gPROMS

and Multiash

can communicate
through external property package interfaced as a Foreign Object.
With this software, hypothetical component C
6
could be cre-
ated to account for all the organic chemicals with 5 or more
carbon atoms. The molecular weight was xed at 76gmol
1
based on conservation of mass principle and the boiling point
was set at 353.25K which is the boiling point of benzene.
Indeed, the fraction C
6
is mainly benzene in thermal cracking of
propane.
3. Model validation
3.1. Industrial data
In order to validate the model, industrial data from Sundaram
and Froment (1979) is used. This industrial reactor was used
several times in the literature. Most recently it was used by
Shahrokhi and Nejati (2002). The technical specications for the
reactor that is simulated are presented in Table 1 and the tem-
perature prole of the base case are presented in Fig. 2. The
reaction scheme used here was developed by Sundaram and
Froment (1977), who added later two reactions in Sundaram and
M. Berreni, M. Wang / Computers and Chemical Engineering 35 (2011) 28762885 2879
Fig. 1. 1D pseudo-dynamic model in a plug owreactor.
Froment (1979). Those two reactions involved C
6
, a hypotheti-
cal component, which represents the fraction of all the organic
chemicals with 5 or more carbon atoms. In the thermal crack-
ing of propane, it mainly consists of benzene, that is why it
is referred as C
6
. The whole reaction scheme is described in
Table 2.
Table 1
Industrial reactor (Sundaramand Froment, 1979).
Paramaters/variable Value
Length of the coil in the radiant section 95 [m]
Length of the straight portion of the coil 8.85 [m]
Length of the bends 0.554 [m]
Radius of the bends 0.178 [m]
Tube internal diameter 0.108 [m]
Wall thickness 0.008 [m]
Total feed per coil 0.7635 [kg/s]
Steamdilution rate 0.4 [kg steam/kg propane]
Inlet pressure 3 [bar]
Outlet pressure 2 [bar]
Inlet temperature 873.15 [K]
Outlet temperature 1111.15[K]
Cokes density 1600 [kg/m
3
]
Thermal conductivity of the coke 0.00154 [kcal/(ms K)]
850
900
950
1000
1050
1100
1150
100 80 60 40 20 0
P
r
o
c
e
s
s

g
a
s

t
e
m
p
e
r
a

r
e

[
K
]
Tube length [m]
Fig. 2. Process gas temperature prole of the base case (Sundaram and Froment,
1979).
3.2. Simulated results vs industrial data
All the results presented in this section were obtained by sim-
ulating the pseudo-dynamic model for PFR under the conditions
given in Table 1 for the reactor and the process gas temperature
prole given in Fig. 2. Table 3 summarizes the simulated product
yields at the outlet of the reactor vs the industrial data. It can be
Table 2
Reaction scheme for propane cracking (Sundaramand Froment, 1977) and for coking reaction (Sundaramand Froment, 1979).
Reactions A(s
1
or *kmol
1
m
3
s
1
or **kgmol
1
m
1
s
1
) E (kJ mol
1
)
1 C
3
H
8
C
2
H
4
+CH
4
4.69210
10
211.7
2 C
3
H
8
C
3
H
6
+H
2
5.88810
10
9.04 x10
5
214.695.3
3 C
3
H
8
+C
2
H
4
C
2
H
6
+C
3
H
6
*2.53610
13
247.1
4 2C
3
H
6
3C
2
H
4
1.51410
11
233.5
5 2C
2
H
6
0.5C
6
+3CH
4
1.42310
9
190.4
6 C
3
H
6
C
2
H
2
+CH
4
3.79410
11
2.3210
7
248.5123.7
7 C
3
H
6
+C
2
H6C
4
H
8
+CH
4
*5.55310
14
251.1
8 C
2
H
6
C
2
H
4
+H
2
4.65210
13
9.9710
8
272.8138.0
9 C
2
H
4
+C
2
H
2
C
4
H
6
*1.02610
12
172.6
10 C
4
H
8
C
6
**6.9610
7
143.6
11 C
3
H
6
Coke **5.8210
14
308.0
Table 3
Comparison of main product yields between pilot plant data and simulated data.
gPROMS simulation Industrial data Absolute difference (wt%) Relative difference (%)
Propane conversion wt% 90.19 90.60 0.41 0.45
CH
4
wt% 25.65 24.00 1.65 6.43
C
2
H
4
wt% 35.08 34.50 0.58 1.65
C
3
H
6
wt% 14.44 14.70 0.26 1.77
C
3
H
8
wt% 9.81 9.30 0.51 5.20
H
2
wt% 1.68 1.20 0.48 28.57
2880 M. Berreni, M. Wang / Computers and Chemical Engineering 35 (2011) 28762885
84
85
86
87
88
89
90
91
800 700 600 500 400 300 200 100 0
P
r
o
p
a
n
e

c
o
n
v
e
r
s
i
o
n

[
%
]
Producon me [h]
Simulated data
Industrial data
Fig. 3. Propane conversion with production time.
2
2.2
2.4
2.6
2.8
3
3.2
100 90 80 70 60 50 40 30 20 10 0
P
r
e
s
s
u
r
e

[
b
a
r
]
tube length [%]
Simulated data
Industrial data
Fig. 4. Pressure prole at production time 0h (i.e. for clean tube).
seen that the result for the two most valuable products (ethylene
and propylene) are in good agreement with the industrial data.
The decrease of the propane conversion with the production
time calculated by the simulation is very close to the one observed
in a real life reactor as can be seen fromFig. 3. As the coke builds up
with production time, inner diameter is reduced and propane has
less time to react inside the reactor (i.e. shorter residence time).
This results in a decreasing propane conversion with production
time.
The prole of the pressure along the reactor (for clean tube, i.e.
production time at 0h) is drawn in Fig. 4. Each step corresponds
to a U-bend where the pressure loss is bigger because of an added
term for the component loss as seen in Eq. (10). The tube length
was actually divided into 300 sections to achieve such an accurate
prole.
FromFig. 5, the pressure drop across the reactor increases with
the production time due to the coke buildup. In order to get the
same process gas temperature prole, the tube outer wall temper-
ature is also increased over the production time as shown in Fig. 5.
This is to compensate for the higher thermal resistance caused by
thicker coke layer.
Sundaram and Froment (1979) also provided the whole tube
outer wall temperature proles for clean tube (i.e. production time
at 0h) and after operations of 700h. The industrial data are com-
1000
1050
1100
1150
1200
1250
0.7
0.9
1.1
1.3
1.5
1.7
1.9
600 400 200 0
M
a
x
i
m
u
m
t
u
b
e
o
u
t
e
r
w
a
l
l
t
e
m
p
e
r
a
t
u
r
e
[
K
]
P
r
e
s
s
u
r
e

d
r
o
p

[
b
a
r
]
Producon me [h]
Simulated data
T
we
max
P
Industrial data
Fig. 5. Pressure drop and maximumtube skin temperature as a function of produc-
tion time.
Fig. 6. Tube outer wall temperature prole at t =0h and t =700h.
pared to the simulated results in Fig. 6 and are again in good
agreement.
The validation is ended with investigating the diameter reduc-
tion after 100h, 300h and 700h in Fig. 7. The change of coke
thickness is important sinceit dramaticallyaffects theheat transfer,
subsequently the process operation.
In summary, Figs. 3, 57 present results simulated under the
specic design/operating conditions (Table 1) and temperature
prole (Fig. 2). They indicate how coke thickness, pressure drop,
propane conversion and tube outer wall temperature change with
production time. These form dynamic validation of the pseudo-
dynamic model.
When the pressure drop or the reduction of diameter reaches a
pre-determined value, the PFR has to be shutdown for de-coking.
Pressure drop and the reduction of diameter are closely related.
However diameter reduction will be used in this paper as indicator
becauseit is moreconvenient for simulationpurposes. Inthis paper,
it was assumed that the shutdown of the process is carried out
when the reduction of diameter reaches 25% at the outlet. For the
case simulated in this section, the run length of the reactor is 39.8
days.
4. Case studies
4.1. Impact of process gas temperature prole
The aimof this case study is to look at the impact of the process
gas temperature prole on the process performances. The process
gas temperature at the outlet (i.e. coil outlet temperature COT)
will be used as an input variable and output variables will be prod-
uct yields, propane conversion, run length and coking rate. COT
will vary from 973.15K up to 1143.15K. This range has been cho-
sen because this is the range of temperature for which Van Damme
et al. (1975) have carried out their experiments, later reconsidered
by Sundaram and Froment (1979) in order to derive the reaction
scheme selected for this simulation. The steam-to-propane ratio
will remain constant at 0.4kg/kg. The pressure at the inlet of the
reactor is xed at 3 bars and Eqs. (8)(10) were used to calculate
the pressure drop along the reactor.
0
2
4
6
8
10
12
14
16
18
20
100 80
700 hrs
300 hrs
100 hrs
60 40 20 0
D
i
a
m
e
t
e
r

r
e
d
u
c

o
n

[
%
]
Reactor length [%]
Simulated data
Industrial data
Fig. 7. Diameter reduction after various production time.
M. Berreni, M. Wang / Computers and Chemical Engineering 35 (2011) 28762885 2881
850
900
950
1000
1050
1100
1150
100 80 60 40 20 0
P
r
o
c
e
s
s

g
a
s

t
e
m
p
e
r
a
t
u
r
e

[
K
]
Tube length [%]
COT=1143.15 K
COT=1111.15 K
COT=1073.15 K
COT=1043.15 K
COT=1013.15 K
COT=973.15 K
Fig. 8. Process gas temperature proles.
0
5
10
15
20
25
30
35
40
45
1150
C
2
H
4
C
3
H
6
CH
4
1100 1050 1000 950
Y
i
e
l
d

[
w
t
%
]
Coil Outlet Temperature (K)
Fig. 9. Mainproduct yields under cleantube condition(i.e. no coking) under various
COTs.
Basedonthe PFRinlet temperature (xedat 873.15K) andoutlet
temperature (varying between 973.18K and 1143.15K), the strat-
egy adopted enables to derive process gas temperature proles
with similar shape as the process gas temperature prole provided
by Sundaramand Froment (1979).
In order to get a prole that will achieve for example 1043.15K
at the outlet, the ratio between the original and the new prole
is calculated i.e. r =(1043.15/1108.15) =0.94. This ratio is weighted
with a variable factor along the tube length so that the temperature
of the base case is multiplied by 1 at the inlet and 0.94 at the outlet.
Thus, the temperature remains at 873.15K at the inlet, and it is
thengradually decreasedcomparedto the base case until the outlet
where the temperature of the base case is multiplied by 0.94. A
slowly changing factor enables us to generate smooth proles as
can be seen in Fig. 8. The ratio is weighted with Eq (11) and the
temperature at any position x is then determined with Eq. (12).
Weight factor (x)
=
Temperature (x) Temperature (l = 0)
Temperature (l = 95) Temperature (l = 0)
(11)
T(x) = Weight factor (x)
T(l = 95)
desired
1108.15
(12)
Fig. 9 shows the propylene, ethylene and methane yields at the
outlet of the reactor for clean tube conditions (i.e. operating hours
at 0h) under various COTs. It can be observed that all the yields
are very lowat 973K, 6.8wt% for ethylene, 8.12wt% for propylene
and 4.11wt% for methane. Ethylene and methane yields increase
steadily with the increase of COT, respectively up to 40.7wt% and
30.2wt% at 1143K. As for propylene yield, it reaches its maximum
value of 19.5wt% around 1073K. Afterwards it decreases down to
6.7wt% at 1143K.
In Fig. 10, the impact of the COT on the propane conversion
has been investigated. It is observed that the propane conversion
increases quite steadily from20% up to nearly complete conversion
(97.5%) for COTs from973.15K and 1143.15K. It should be empha-
sizedthat basedonFig. 10, we caninfer that COTs lower than1050K
0
20
40
60
80
100
1150 1100 1050 1000 950
P
o
r
p
a
n
e

c
o
n
v
e
r
s
i
o
n

(
%
)
Coil Outlet Temperature [K]
Fig. 10. Propane conversion under clean tube condition (i.e. no coking) under vari-
ous COTs.
0
0.005
0.01
0.015
0.02
0.025
100
1000
10000
100000
1000000
1200 1150 1100 1050 1000 950
C
o
k
i
n
g

r
a
t
e
r
a
t
e
[
k
g
/
(
m

.
h
)
]
R
u
n

l
e
n
g
t
h

[
h
]
Coil Oulet Temperature [K]
Run length
Coking rate
Fig. 11. Coking rate under clean tube condition (i.e. no coking) and run length.
should be avoided because they result in propane conversion lower
than 50%.
The effect of COT on coking rate can be seen in Fig. 11. As
could be expected, as the COT increases, so does the coking rate
at the outlet but it also starts decreasing for the highest tempera-
ture. At 973K, the coking rate is very small 1.110
4
kg/(m
2
h).
Then, it increases faster and faster and reaches its maximum
value of 2.310
2
kg/(m
2
h) at 1123K before decreasing to
2.210
2
kg/(m
2
h) at 1143K. This behaviour is due to the fact that
the coking rate depends uponnot onlythe process gas temperature,
but also the molar concentration of the coke precursor propylene.
As we have seen from Fig. 9, the yields of propylene and conse-
quently its concentration, starts decreasing gradually at 1073K. At
the beginning, the decrease of concentrationis compensated by the
increase of temperature. But at 1123K, the increase in temperature
is not sufcient any more and the coking rate decreases. Fig. 11 also
shows the run length determined dynamically for various COTs. It
can be seen that, the faster the coking rate, the shorter the run
length. At 1043K, the run length is around 5526h (230.3 days), but
it then goes down to 986h (31.1 days) at 1143K. For COT below
1043K, run length is very long, but the cracking reactions are also
very slow.
In summary, process gas temperature affects the product yields
signicantly. Low temperatures result in low ethylene yield. High
temperatures increase the total yield of ethylene and propylene,
and also favour coke formation. A compromise in determining pro-
cess gas temperature prole is necessary.
4.2. Impact of constant tube outer wall prole
Here the tube outer wall temperature prole along the length of
thereactor is xed, theprocess gas temperatureproleis calculated
from the pseudo-dynamic model. The impact of xed tube outer
wall temperature prole on the process will be studied over the
whole run length. The conditions applied are a steam-to-propane
ratio of 0.4kg/kg and a tube outer wall temperature prole shown
in Fig. 12 that generates a COT of 1111.15K under clean tube con-
ditions.
2882 M. Berreni, M. Wang / Computers and Chemical Engineering 35 (2011) 28762885
1060
1080
1100
1120
1140
1160
1180
100 80 60 40 20 0
T
u
b
e

w
a
l
l

t
e
m
p
e
r
a
t
u
e

[
K
]
Reactor length [%]
Fig. 12. Tube outer wall temperature prole.
Since the overall heat transfer coefcient will decrease with
higher coke thickness, the process gas temperature prole will
decrease with production time when the tube wall temperature
prole remains constant. Because of the reduced process gas tem-
perature, therunlengthis 1681h(70days). Theimpact of aconstant
tube outer wall temperature prole on the yields of ethylene and
propylene with time is shown in Fig. 13. On one hand, the ethy-
lene yields decreases rapidly from 35wt% at the beginning down
to 25.1wt% after 1681h. On the other hand, the propylene yield
increases from 14.34wt% to 18.7wt% in the rst 700h and then it
keeps increasing slowly up to 19.5wt% at the end of the run length.
Propane conversion with production time is shown in Fig. 14. The
propane conversion decreases during the production time. Right
after decoking operation, the propane conversion is 90%, but it
decreases to 70.5% after 1681h (i.e. 70 days) of operation.
Fig. 15 shows that the coil outlet temperature (COT) of the pro-
cess gas decreases signicantly with the production time. Initially
the process gas temperature at the outlet was 1111.15K. But after
1000h of operation, it is only 1080.9K. Finally the process gas tem-
perature at the outlet drops to 1072.2Kat the end of the run length.
This means that the reactor is going to operate under conditions far
fromthe one initially set by the operator. Consequently, the prod-
uct yields are dramatically affected as shown in Fig. 13. In Fig. 15,
it can also be seen that the coking rate at the outlet decreases with
10
15
20
25
30
35
40
1600 1400 1200 1000 800 600 400 200 0
Y
i
e
l
d
s

[
w
t
%
]
Producon me [h]
Fig. 13. Product yields.
70
75
80
85
90
95
1600 1400 1200 1000 800 600 400 200 0
P
r
o
p
a
n
e

c
o
n
v
e
r
s
i
o
n

[
%
]
Producon me [h]
Fig. 14. Propane conversion.
0.005
0.01
0.015
0.02
0.025
1070
1080
1090
1100
1110
1120
1600 1400 1200 1000 800 600 400 200 0
C
o
k
i
n
g

r
a
t
e

[
k
g
/
(
m
.
s
)
]
P
r
o
c
e
s
s

g
a
s

t
e
m
p
e
r
a
t
u
r
e

a
t
t
h
e

o
u
t
l
e
t

[
K
]
Producon me [h]
Process gas temperature at
the outlet
Coking rate
Fig. 15. Process gas temperature at the outlet and coking rate with production time.
0
10000
20000
30000
40000
50000
60000
70000
80000
1600 1400 1200 1000 800 600 400 200 0
P
r
o
d
u
c

o
n

i
n

1
0
0

h
o
u
r
s

[
k
g
]
Producon me [h]
Ethylene
Propylene
Fig. 16. Amount of ethylene and propylene produced within every 100h with pro-
duction time.
time rapidly from 0.022kg/(m
2
s) down to 0.0077kg/(m
2
s) at the
end of the run length. This can be explained by the trend of the
process gas temperature, and the low coking rate is the reason for
this unusually high run length.
Fig. 16plots the productionat a giventime minus the production
100h earlier for different production time. In this way, it is pos-
sible to analyse the amount of ethylene and propylene produced
every 100h. At the very beginning of the operation, the furnace
produced 67,400kg of ethylene in the rst 100h, but at the end of
the run length, the reactor produces only 49,900kg in 100h. On the
contrary, the furnace produces 29,400kg of propylene in the rst
100h and 38,100kg at the end of the run length within the same
period of time (100h). The production of propylene increases, but
the production of ethylene decreases. Overall, the amount of valu-
able products (ethylene plus propylene) produced during a given
period decreases signicantly with the production time.
In conclusion, this case study shows the importance of tuning
the outer wall temperature prole dynamically. Although xing a
constant value is much easier from a control point of view, the
losses of production are signicant, and may not be acceptable.
5. Process operation optimization
Steam-to-propane ratio and process gas temperature prole are
major operatingvariables andsome trade-offs inselectingthemare
necessary. However in actual plants, operators adjust the fuel burn
rates in side or bottomburners, thus changing the tube outer wall
temperature prole, which will subsequently affect the process gas
temperature prole. Therefore, tube outer wall temperature prole
and steam-to-propane ratio are used as optimization variables in
this section.
5.1. Methodology
Operating prot over a period of 1 year is taken as objective
function to maximise. In this dynamic optimization, the following
assumptions were made: (a) No downstream product separation
costs were considered; (b) Coke thickness is updated every hour;
M. Berreni, M. Wang / Computers and Chemical Engineering 35 (2011) 28762885 2883
Table 4
Optimization parameters.
Parameters Physical meaning Values Units
tp Yearly production time 8160 h
t
d
Decoking time per cycle 48 h
C
0
Propane price factor 0.596 $/kg
C
1
Ethylene price factor 1.356 $/kg
C
2
Propylene price factor 1.576 $/kg
C
S
Steamprice factor 0.0129 $/kg
C
Q
Heat price factor 1.2610
5
$/kJ
DCC Decoking cost per cycle 66,000 $
c Density 1600 kg/m
3
max Maximumcoke thickness 0.0135 m
(c) Only ethylene and propylene are assumed to be sold. Methane
was not included because of its relative lowprice.
Constraints were set for the optimization variables. Since
Sundaram and Froment (1979) developed their reaction scheme
for a certain operating range, the same limits were applied:

973K<coil outlet temperature for process gas <1143K

0.2kg/kg<steam-to-propane ratio<1.0kg/kg
Moreover, a constraint was applied for the maximum temper-
ature that can reach the tube metal. The material considered is
HK40 and this limit (for tube outer wall temperature) was set at
1338.15K(Albright, Crynes, &Corcoran, 1983). Table 4 summarizes
the parameters assumed for the optimization.
Price factors, decoking cost, steam price factor, decoking time
per cycle and yearly production time were taken fromGao, Wang,
Ramshaw, et al. (2009). As for chemicals price factors, raw mate-
rial would be propane and the two most desirable products are
propylene and ethylene. Spot prices fromMay 2010 have been con-
sidered (Chemical Weekly, 2010). Finally, coke density was taken
fromSundaramand Froment (1979). Shutdown is assumed to take
place when coke thickness reaches 25% tube diameter reduction at
the outlet. It should be emphasized that the diameter reduction is
therefore much lower for the rest of the tube.
A newobjective function was developed in this paper based on
Gao, Wang, Ramshaw, et al. (2009). The principle is to calculate
the prot that is made every hour, and to accumulate those prots
over the whole duration of a cycle. A cycle includes the time of
operation of the reactor plus the time of decoking operation. Thus,
the decoking frequency is also taken into account. The objective
function is as follows:
Prot
t
= Income Cost (13)
Income =
t
c
t
c
+t
d

F
i
C
i
(14)
Cost =
t
c
t
c
+t
d

_
F
0
C
0
+F
s
C
s
+

Q
t
C
Q
+
DCC
t
c
_
(15)
More particularly, the run length (i.e. the time between two
consecutive decoking operations) t
c
can be determined as follows:
t
c
=

max

c
r
c
(z = 95)
(16)
0.33
0.34
0.35
0.36
0.37
0.38
0.39
0.4
0.41
1200 1000 800 600 400 200 0
S
t
e
a
m
-
t
o
-
p
r
o
p
a
n
e

r
a

o

[
k
g
/
k
g
]
Producon me [h]
Base case
Opmal case
Fig. 17. Steam-to-propane ratio with production time.
5.2. Results
Dynamic optimization is performed in gPROMS

with a control
interval of 50h. The results are compared to the base case used
for the model validation. Table 5 shows a signicant increase of
13.1% of the yearly prot ($851,840 for the base case and $963,279
for the optimal case). The run length is increased to 55.6 days com-
pared to 39.8 days for the base case. In Table 5, two values are given
for pressure drop, heat transfer rate and propane conversion, they
represent values at the start and end of the run length respectively.
The prole of the steam-to-propane ratio is investigated in
Fig. 17. The optimized ratio increases from 0.34kg/kg up to
0.38kg/kg during the rst 400h, and then decreases slowly down
to 0.35kg/kg at the end of the cycle. The ratio is however always
lower than in the base case. Fig. 18 shows the optimal tube outer
wall temperature at the outlet changing with the production time.
It can be seen that in the optimal case, the temperature is always
between 25 and 30K lower than in the base case. The process gas
temperature at the outlet is also given in Fig. 18. In the base case,
the process gas temperature at the outlet is maintained constant.
While in the optimal case, it increases with production time. The
overall trend is an increase from1087K at the beginning to 1100K
at the end of the run length. A cycle takes place, due to an increase
of the tube outer wall temperature, roughly every 2 days. Fig. 19
shows the tube outer wall temperature prole at the beginning of
the run length and at the end. Figs. 20 and 21 show the ethylene
yield and propylene yield in both optimal and base cases. In the
1080
1100
1120
1140
1160
1180
1200
1220
0 200 400 600 800 1000 1200
T
e
m
p
e
r
a
t
u
r
e

a
t

t
h
e

o
u
t
l
e
t

[
K
]
Producon me [h]
Opmal case
Base case
Tube outer wall temperature
Process gas temperature
Fig. 18. Tube outer wall temperature andprocess gas temperature at the outlet with
production time.
Table 5
Optimization results.
Parameter/variable Unit Dynamic optimization Base case
Pressure drop Pa 95,612183,255 95,758181,700
Radiant heat transfer rate kJ/h 1.6510
6
1.7410
6
1.7710
6
1.7210
6
Propane conversion % 90.286 82.381.6
Objective function $/year 963,279 851,840
Run length days 55.6 39.8
2884 M. Berreni, M. Wang / Computers and Chemical Engineering 35 (2011) 28762885
1060
1080
1100
1120
1140
1160
1180
1200
1220
1240
100 80 60 40 20 0
T
u
b
e

o
u
t
e
r

w
a
l
l

t
e
m
p
e
r
a
t
u
r
e

[
K
]
Tube length [%]
Base case
Opmal
case
Fig. 19. Tube outer wall temperature prole at the beginning and at the end of the
run length.
28
29
30
31
32
33
34
35
36
1200 1000 800 600 400 200 0
E
t
h
y
l
e
n
e

y
i
e
l
d

[
w
t
%
]
Producon me [h]
Base case
Opmal case
Fig. 20. Ethylene yield with production time.
optimal case, ethylene and propylene yields are maintained con-
stant respectively around 30.5wt% and 18wt%. On the contrary, in
the base case, the propylene yield increases and the ethylene yield
decreases considerably.
5.3. Discussion
This section is to explain the results presented in Section 5.2.
Here the product prices and the raw material price are xed
(Table 4). In order to maximize the objective function (i.e. Eq.
(13)) for higher yearly prot, we can either increase the amount
of desired products (i.e. ethylene and propylene) manufactured or
increase the run length.
Even though the ethylene product yield for the optimal case is
always lower thanfor the base case (Fig. 20), the propylene yieldfor
the optimal case is always higher than for the base case (Fig. 21).
Ethylene and propylene are all desired products and their prices
are close to each other (Table 4). Adding the two product yields
together, the total for the optimal case is slightly lower than for
the base case. However, the run length for the optimal case (55.6
days) is much longer than for the base case (39.8 days), as pre-
12
13
14
15
16
17
18
19
20
1200 1000 800 600 400 200 0
P
r
o
p
y
l
e
n
e

y
i
e
l
d

[
w
t
%
]
Producon me [h]
Base case
Opmal case
Fig. 21. Propylene yield with production time.
sented in Table 5. This can explain why the yearly prot for the
optimal case ($963,279) is much higher than for the base case
($851,840).
Thesteam-to-propaneratiohas impact ontheamount of desired
products manufactured and the run length. The optimization study
was carried out when the molar ow rate in the feed to PFR was
xed at 87.75kmol/h. This is to consider the maximum capacity
in the reactor in terms of volumetric ow rate. Under this con-
dition, the increase of steam-to-propane ratio means more steam
is added, then the propane in the feed is reduced. This indicates
lower amount of products will be manufactured. Secondly, increase
of steam-to-propane ratio can reduce coking rate. This has been
studied in detail in Berreni (2010).
Process gas temperature also has impact on the amount of
desired products manufactured and the run length (Berreni, 2010).
On one hand, higher temperature means faster cracking reactions
(i.e. more desired products manufactured at unit time). On the
other hand, this higher temperature speeds up coking, which sub-
sequently makes run length shorter.
Following the discussions on the impacts of steam-to-propane
ratio in the feed and process gas temperature prole on the desired
products manufactured and the run length, the explanations for
the optimal steam-to-propane ratio prole (Fig. 17) and process
gas temperature prole (Fig. 18) will be easier to understand.
At the start (i.e. production time 0h in Fig. 17), the steam-to-
propane ratio for the optimal case is 0.34 compared with 0.40 for
the base case. The process gas temperature (Fig. 18) is still quite low
(1087K). At this temperature, the coking rate is quite slow, using
lower steam-to-propane ratio would get more benets since more
desired products will be manufactured.
From 0h to 400h (Figs. 17 and 18), the process gas temper-
ature increases all the time, the cracking reactions speed up, so
does the coking reaction. Generally the reaction rate doubles every
10K increase of the temperature. Steam-to-propane ratio has been
increased to compensate. So the increase of the coking rate would
not be detrimental to the increase of yearly prot.
At around 400h (Figs. 17 and 18), the process gas temperature
has reached a point, at which the economic benet brought by
continuous increase of steam-to-propane is not higher than that
brought by keeping or even reducing steam-to-propane ratio. This
is because lower steam-to-propane ratio means more propane is
fed to the PFR to get more products. That is why the steam-to-
propane ratio reaches its maximum.
It should be noted that the results presented in Table 5 and from
Figs. 17 to 21 were obtained assuming xed product prices and raw
material price (Table 4). In real life, these prices uctuate all the
time. Therefore the methodology developed in this paper should
beveryhelpful for determiningmost protableoperatingstrategies
under different market conditions.
6. Conclusions
In this paper, modelling, simulation and optimization of ther-
mal cracking of propane in a tubular reactor were implemented
successfully in gPROMS

. A 1D pseudo-dynamic model consider-


ing the radiant section alone was developed and the model was
validated. The simulated results were in good agreement with the
industrial data. Case studies were then carried out to investigate
the behaviour of the process under different process gas tempera-
ture proles and under a xed tube outer wall temperature prole.
Finally dynamic optimization was performed with steam-to-
propane ratio and tube outer wall temperature prole as optimiza-
tion variables. Operating prot was dened as objective function.
The impact of the coke buildup on the run length and on the prod-
ucts yields was taken into account. The cumulative operating prot
M. Berreni, M. Wang / Computers and Chemical Engineering 35 (2011) 28762885 2885
over a periodof 1year was increasedby 13.1%comparedto the base
case, proving the efciency of the method developed in this paper.
Acknowledgements
The authors would like to thank Dr Praveen Lawrence fromPSE
Ltd for help to use optimization tool in gPROMS

. Meihong Wang
would like to acknowledge the helpful discussions with Professor
Costas C. Pantelides from Centre for Process Systems Engineering
(CPSE), Department of Chemical Engineering, Imperial College
London.
References
Albright, L. F., Crynes, B. L., & Corcoran, W. H. (1983). Pyrolysis Theory and Industrial
Practice. NewYork, USA: Academic Press Inc.
Babu, B. V., & Knovel. (2004). Process plant simulation. New Delhi, India/New York:
Oxford University Press.
Berreni, M. (2010). Modelling, simulationanddynamic optimizationof thermal cracking
of propane for ethylene industry. MSc Thesis, Craneld University, UK.
Chemical Weekly (May 2009May 2010). CW PRICE REPORT, available at
http://web.ebscohost.com/ehost/pdfviewer/pdfviewer?vid=4&hid=11&sid=123
b8038-b568-4a5c-8f37-d95f56e765c6%40sessionmgr14 (last accessed in July
2010).
Detemmerman, T., & Froment, F. (1998). Three dimensional coupled simulation of
furnaces and reactor tubes for the thermal cracking of hydrocarbons. Oil & Gas
Science and Technology Review. IFP, 2(2), 181194.
Gao, G.-Y., Wang, M., Pantelides, C. C., Li, X.-G., & Yeung, H. (2009). Mathemati-
cal modeling and optimal operation of industrial tubular reactor for naphtha
cracking. Computer Aided Chemical Engineering, 27, 501506.
Gao, G.-Y., Wang, M., Ramshaw, C., Li, X.-G., & Yeung, H. (2009). Optimal operation
of tubular reactors for naphtha cracking by numerical simulation. Asian Pacic
Journal of Chemical Engineering, 4(6), 885892.
Ghashghaee, M., & Karimzadeh, R. (2007). Dynamic modelling and simulation of
steamcracking furnaces. Chemical Engineering and Technology, 30(7), 835843.
Lan, X., Gao, J., Xu, C., & Zhang, H. (2007). Numerical simulation of transfer and reac-
tion processes in ethylene furnaces. Chemical Engineering Research and Design,
12(85), 15651579.
Masoumi, M. E., Sadrameli, S. M., Towghi, J., & Niaei, A. (2006). Simulation, opti-
mization and control of a thermal cracking furnace. Energy, 31(4), 516527.
Shahrokhi, M., &Nejati, A. (2002). Optimal temperature control of a propane thermal
crackingreactor. Industrial &EngineeringChemistryResearch, 41(25), 65726578.
Sowers, G., & Reed, C. (2001). Dynamic optimization of ethylene furnaces cracking
propane. In Proceedings of AIChE Ethylene Producers Conference Houston, TX, (pp.
366405).
Sundaram, K. M., & Froment, G. F. (1977). Modelling of thermal cracking kinetics-I.
Thermal cracking of ethane, propane and their mixture. Chemical Engineering
Science, 32(6), 601608.
Sundaram, K. M., & Froment, G. F. (1979). Kinetics of coke deposition in the thermal
cracking of propane. Chemical Engineering Science, 34(5), 635644.
Sundaram, K. M., Shreehan, M. M., & Olszewski, E. F. (2001). Ethylene. In J. I.
Kroschwitz, A. Seidel, & R. E. Kirk, et al. 5th (edtors), Kirk-Othmer Encyclopedia
of Chemical Technology. Hoboken, NJ: Wiley-Interscience.
VanDamme, P. S., Narayanan, S., &Froment, G. F. (1975). Thermal crackingof propane
and propane-propylene mixtures: Pilot plant versus industrial data. AIChE Jour-
nal, 21(6), 10651073.
Zou, R., Lou, Q., & Liu, Z. (1988). Study of kinetic models for the pyrolysis reaction of
propane. Journal of Analytical and Applied Pyrolysis, 13(3), 183190.

Você também pode gostar