You are on page 1of 29

Chemical Product and Process

Modeling
Volume 4, Issue 1 2009 Article 14

Modelling and Optimisation of an Industrial


Ethylene Oxide Reactor

Shahla Aryana, The University of Sydney


Mitra Ahmadi, The University of Sydney
Vincent G. Gomes, The University of Sydney
Jose A. Romagnoli, The University of Sydney
Kian Ngian, Huntsman Corporation, Australia

Recommended Citation:
Aryana, Shahla; Ahmadi, Mitra; Gomes, Vincent G.; Romagnoli, Jose A.; and Ngian, Kian
(2009) "Modelling and Optimisation of an Industrial Ethylene Oxide Reactor," Chemical
Product and Process Modeling: Vol. 4: Iss. 1, Article 14.
DOI: 10.2202/1934-2659.1231
Available at: http://www.bepress.com/cppm/vol4/iss1/14
©2009 Berkeley Electronic Press. All rights reserved.
Modelling and Optimisation of an Industrial
Ethylene Oxide Reactor
Shahla Aryana, Mitra Ahmadi, Vincent G. Gomes, Jose A. Romagnoli, and Kian
Ngian

Abstract
A dynamic model of an industrial packed-bed multi-tubular reactor was developed to
investigate performance of an industrial ethylene oxide (EO) reactor, conducting epoxidation of
ethylene over a silver-based catalyst. The set of nonlinear kinetic rate equations for the catalytic
oxidation process in the presence of ethylene dichloride (EDC) as a moderator was coupled with
the governing heat and mass transfer equations along the packed bed. Catalyst deactivation was
modelled as a nonlinear function of operating time and the equation was benchmarked against
plant data for the period of operation. Our process model was compared with experimental data
obtained from an industrial EO reactor. The model predictions were found to agree well with the
plant data. The influences of operating parameters such as EDC level, reactant concentrations,
reactor pressure, coolant temperature and the feed temperature on reactor performance were
investigated. The variables having significant impact on work rate and selectivity were identified.
The model was used to optimise the performance of ethylene oxide reactor for maximising work
rate and selectivity.

KEYWORDS: ethylene oxide, oxidation reactor, ethylene dichloride, optimisation, selectivity


Aryana et al.: Modeling and Optimisation: Industrial Ethylene Oxide Reactor

Introduction

Ethylene oxide (EO) is a major chemical, and the global demand for ethylene
oxide continues to increase because of its importance as an intermediate in
producing antifreeze, polyester fibres and other petrochemical products. The
production of EO is a critical process because the reactor can generate eleven
times as much heat in a runaway condition as under normal operation conditions
[Rebsdat 1987; Kishor 2003]. From 0.03 to 1 mole fraction, a mixture of EO in air
is explosive at room temperature. As result, EO is normally stored at 5oC under
4.5 kg/cm2 pressure [Rebsdat, 1987]. Therefore, the safety issues for an EO
reactor are of paramount importance.
Commercial processes for producing EO are based on direct oxidation of
ethylene with pure oxygen. Complete combustion of ethylene, as well as further
oxidation of ethylene oxide to carbon dioxide and water, also occurs during the
reaction. Silver catalyst, packed in tubes with promoters, is the state-of-the-art
catalyst for ethylene oxidation, because of its high activity and selectivity for EO.
Organohalide inhibitors such as 1,2-dichloroethane or ethylene dichloride (EDC)
are added to control the reaction rate and improve the selectivity of the catalyst
and its concentration is usually 1-3 ppm. The complete oxidation of ethylene is
inhibited to a great extent by EDC than partial oxidation; hence the selectivity for
ethylene oxide is promoted. Thus, EDC is termed an “inhibitor” and a “promoter”
due to its promoting effect on selectivity and inhibiting effect on oxidation.
Since the reactions involved are highly exothermic, heat transfer
efficiency is critical. Multi-tubular fixed bed reactors are widely used in industrial
EO production. Heat is removed by a coolant circulating through the shell side.
Failure to remove the heat adequately may lead to thermal runaway. With time,
the silver based catalysts for EO production exhibit a continuous loss of activity
and selectivity, which is accelerated if the reactor is run at higher than necessary
temperatures. Run-away conditions and burn-out of the catalyst and equipment
may transpire at elevated temperatures. These considerations are particularly
important in industrial EO manufacture, where maximising production and
minimising risks under variable conditions are imperative.
Several authors investigated the modelling and simulation of ethylene
epoxidation [Azevedo, 1990; Cornelio, 2006; Zhou & Yuan, 2005] however, the
published information is insufficient. Zhou and Yuan [2005] studied the steady-
state and dynamic optimisation of an ethylene epoxidation process. They
investigated the effect of inlet ethylene, O2 and CO2 concentrations, and operating
pressure on productivity but did not consider the effect of ethylene oxide
oxidation, which has significant consequences. The current work considered all
relevant reactions for analysis and optimization of the EO reactor over the period
of catalyst life time.

Published by Berkeley Electronic Press, 2009 1


Chemical Product and Process Modeling, Vol. 4 [2009], Iss. 1, Art. 14

Reaction Kinetics

The parallel consecutive reaction of ethylene and ethylene oxide is shown in


Figure 1. The first reaction is known as partial oxidation or epoxidation of
ethylene, while the second reaction involves total oxidation or combustion.

(1)
½ O2
C2H4 C2H4O
3O2 5/2 O2
(2) (3)
2CO2 + 2H2O

Figure 1- The parallel consecutive reaction of ethylene and ethylene oxide

The three main reactions in our process are.

H 2 C = CH 2 + 1 / 2O2 → .C 2 H 4 O ΔH 1 = −1.23 × 10 5 kJ .kmol −1 (R1)

H 2 C = CH 2 + 3.O2 → 2CO2 + 2 H 2 O ΔH 1 = −1.30 × 10 6 kJ .kmol −1 (R2)


5
C 2 H 4 O + O2 → .2CO2 + 2 H 2 O
2 ΔH 1 = −1.17 × 10 6 kJ .kmol −1 (R3)

The overall reaction rate is described by:

Ri = f i (T , p j ) × ai (t, T , pk ) × Ci Eq.1

where ai and Ci denote deactivation term and correction factor, respectively.


Park and Gau [1987] presented a Langmuir-Hinshelwood mechanism for
oxidation and epoxidation of ethylene with surface reaction as the rate-
determining step. Petrov et al. [1984] developed a kinetic model of ethylene
epoxidation over supported silver catalyst based on a Rideal-Eley type
mechanism. They concluded that the gaseous or weakly adsorbed ethylene
interacts with chemisorbed oxygen. Westerterp et al. [1984] suggested that with a
large excess of ethylene, the rate equations simplify to first order kinetics in
oxygen concentration. Borman et al. [1992], proposed a rate expression without
any inhibitor. Petrov et al. [1986], considered a rate expression including EDC
effects. We tested a number of published rate equations in conjunction with our
reactor model. The rate expressions used in the current work, including the three

http://www.bepress.com/cppm/vol4/iss1/14 2
DOI: 10.2202/1934-2659.1231
Aryana et al.: Modeling and Optimisation: Industrial Ethylene Oxide Reactor

main reactions with effects of EDC as inhibitor, were adapted from elsewhere
[Petrov et al., 1984, 1986, 1988, Eliyas et al., 1988] as given below:

1 K 1 PEth POxy − K 2 PEth POxy PEDC


k 13

f (1) = − × × ρ cat Eq.2


2 1 + K 5 P0 xy + K 6 PEth

K 3 P Eth POxy − K 4 PEth POxy P EDC


k 14

f (2) = −3 × × ρ cat Eq.3


1 + K 5 P0 xy + K 6 PEth

5 K 7 PEO POxy − K 8 PEth PEO PEDC k 15

f (3) = − × − 0 .5 2
× ρ cat Eq.4
2 (1 + K 9 P0 xy + K 10 P. 0oxy. 5 + K 11 PEO + K 12 PEO P. oxy )

where Eth, Oxy, EO, and EDC stand for ethylene, oxygen, ethylene oxide and
dichloroethane, respectively. The kinetic rate constants k1 –k12 have an Arrhenius
dependency on temperature and can be expressed as:

− E a ,i
K i =1 − 4 , 7 − 8 = K 0 , i e ( )
RT S

E a ,i
K i = 5 , 6 , 9 −12 = K 0 , i e ( )
RT S

The equation parameters are given elsewhere [Petrov et al, 1986]. The
correction factors (Ci) were calculated based on the information provided by the
catalyst vendor. The catalyst density of 1260 kg/m3 is calculated from catalyst
pellet packing data received from plant.

Process Description

The industrial ethylene oxide reactor [Huntsman, 2005], a multi-tubular reactor


(Figure 2) with 7750 tubes, is packed with silver catalyst supported on alumina.
MobilTherm "607" flows in the 3.3 m diameter reactor shell to remove the heat
generated. The operating temperature of the reactor is 230-260ºC and the pressure
varies from 20-25 atm. The oil feed temperature is about 245 ºC and heats the
incoming gas in the first part of the reactor before acting as a coolant. The process
is oxygen-based and operates in the oxygen lean zone to prevent explosion.

Published by Berkeley Electronic Press, 2009 3


Chemical Product and Process Modeling, Vol. 4 [2009], Iss. 1, Art. 14

Figure 2- Schematic representation of the EO reactor (not to scale)

The product gas from the outlet of the reactor preheats the inlet gas in a
separate heat exchanger. The product gas is then stripped off ethylene oxide in 2
scrubbers. The stripped gas is thereafter de-carbonated before being recycled
back to the reactor. The outlet cooling oil is cooled in a kettle boiler to produce
saturated steam.
The raw feed materials including pure oxygen and ethylene mixed with
recycle gas stream (from the scrubbers), are heated and fed to the reactor. The
reactor feed consists of 7 main components including: ethylene, oxygen, ethylene
oxide (usually less than 0.01%), carbon dioxide, water (reaction’s by-products),
argon (built up in the system by adding oxygen) and nitrogen (used as ballast).
Also, a small concentration of 1,2-dichloroethane (EDC) and ethane are
intentionally added to the feed to improve reactor selectivity and to moderate the
oxidation. A typical reactor feed and product composition is shown in Table 1.
Table 2 shows the reactor typical operating conditions. We note that a
difference in oil inlet temperature between start of run (SOR) and end of run
(EOR) exists due to catalyst deactivation.

Table 1- Typical reactor feed and product composition (mol %)


Composition (mol %) Feed Product
Ethylene 17.5 16
Oxygen 7 5
ethylene oxide 0.01 1.7
Carbon dioxide 5.0 6.1
Water 0.3 1.51
Nitrogen + Argon 70.2 69.7
EDC (ppm) 1-2 1-2

http://www.bepress.com/cppm/vol4/iss1/14 4
DOI: 10.2202/1934-2659.1231
Aryana et al.: Modeling and Optimisation: Industrial Ethylene Oxide Reactor

Table 2- Reactor operating conditions


SOR (@t = 0) EOR (@t = 1000 days)
Gas feed flow rate (kg/hr) 163 700 163 700
Pressure (bar) 20.68 20.68
Gas inlet temperature (oC) 204 204
Oil flow (kg/s) 1250 1250
Oil inlet temperature (oC) 235 250

Model Development

In our one-dimensional model, radial variations of concentration and temperature


are not considered. The radial dispersion of concentration and temperature within
the reactor bed is negligible due to the small radius of the tubes relative to the
length (r/L<<1). EO reactor operates below 673 K and a radiation term is
negligible [Khanna and Seinfeld, 1987]. The main assumptions for the model are:
• Radial dispersion is negligible compared to axial dispersion;
• Mass transfer resistance between gas and catalyst is negligible;
• Nitrogen and argon molar flows are constant along the bed;
• Alumina support data are taken from Bradshaw et al. [1970];
• Catalyst-gas heat transfer is included;
• Convective heat transfer dominates heat effects in an industrial reactor;
• Heat transfer coefficient for catalyst pellet is calculated from j-factor
[Bradshaw et al., 1970];
• Heat transfer coefficient on the shell side is given by Bell-Delaware method
[Hewitt et al., 1994];
• Heat transfer coefficient in the gas film adjacent to the tube wall is estimated
from correlations [Drew et al., 1962];
• Pressure drop estimate is based on plant data and literature correlation [Fogler,
1999].

Mass Balance

The species mass balance for the gas phase is given by:

∂F j ∂F j ∂ 2 Fj
= Vgas × ( ∑ν j / ν BASE ( i ) × Ri ) × AC ,T × (1 − e) − Vgas × + Dax × Eq.5
∂t i =1−3 ∂Z ∂Z 2

Published by Berkeley Electronic Press, 2009 5


Chemical Product and Process Modeling, Vol. 4 [2009], Iss. 1, Art. 14

Ethylene species balance

∂FC2 H 4
= V gas × (2 × R1 + 1 / 3 × R2 ) × AC ,T × (1 − e) −
∂t Eq.6
∂FC2 H 4 ∂ 2 FC2 H 4
V gas × + Dax ×
∂Z ∂Z 2

Oxygen species balance:

∂FO2 ∂FO2 ∂ 2 FO2


= V gas × ( R1 + R2 + R3 ) × AC ,T × (1 − e) − V gas × + Dax × Eq.7
∂t ∂Z ∂Z 2

Ethylene oxide species balance:

∂FEO ∂F
= −V gas × (2 × R1 − 2 / 5 × R3 ) × AC ,T × (1 − e) − V gas × EO +
∂t ∂Z
Eq.8
∂ FEO
2
Dax ×
∂Z 2

Carbon dioxide species balance:

∂FCO2 ∂FCO2
= −V gas × (2 / 3 × R2 + 4 / 5 × R3 ) × AC ,T × (1 − e) − V gas × +
∂t ∂Z
Eq.9
∂ 2 FCO2
Dax ×
∂Z 2

Water species balance:

∂FH 2O ∂FH 2O
= −V gas × (2 / 3 × R2 + 4 / 5 × R3 ) × AC ,T × (1 − e) − V gas × +
∂t ∂Z
Eq.10
∂ 2 FH 2O
Dax ×
∂Z 2

http://www.bepress.com/cppm/vol4/iss1/14 6
DOI: 10.2202/1934-2659.1231
Aryana et al.: Modeling and Optimisation: Industrial Ethylene Oxide Reactor

Energy balance on catalyst pellet


∂ Tcat
ρ cat × Cp cat × = α cat × SVP × (Tgas − Tcat )
∂t
Eq.11
1 − eb
+ ( ∑ν j /ν BASE (i ) × Ri × ΔH i ) ×
i =1− 3 eb

Energy balance on gas side


∂ Tgas ∂Tgas
ρ mol , gas × Cp gas × = − ρ mol , gas × Cp gas × V gas ×
∂t ∂Z
+ α T × SVI × (Twall , I − Tgas ) / eb + Eq.12
α Cat × SVP × (Tcat − Tgas ) × (1 − eb ) eb

Energy balance on tube wall


The outer tube wall temperature is dependent on the heat transfer to the tube inner
surface, characterized by α wall , and the heat transfer of the shell side ( α S ).

∂ Twall ,O
ρ wall × Cp wall × = α s × (4 × DO ) ( DO2 − DI2 ) × (Toil − Twall ,O )
∂t Eq.13
− α wall × (4 × DO ) ( DO2 − DI2 ) × (Twall ,O − Twall , I )

The inner sides of the tube wall exchange heat with the outer side of tube wall as
well as the gas, where α wall the tube side heat transfer coefficient is:
∂ Twall , I
ρ wall × Cp wall × = α wall × (4 × DO ) ( DO2 − DI2 ) × (Twall ,O − Twall , I )
∂t Eq.14
− α T × (4 × DI ) ( D − D ) × (Twall , I − Tgas )
2
O
2
I

Energy balance on cooling oil


The oil temperature is dependent on the heat transfer to the outer surface of the
tube wall, which is characterized by α S , and the heat loss from the shell to the
surface.

∂ Toil ∂T
ρ oil × Cp oil × Ac ,c × = −Goil × Cp oil × oil + α s × (π × N T × DO )
∂t ∂Z Eq.15
× (Twall ,O − Toil ) + U loss × (π × DS ) × (Tamb − Toil )

Published by Berkeley Electronic Press, 2009 7


Chemical Product and Process Modeling, Vol. 4 [2009], Iss. 1, Art. 14

Heat capacity, viscosity and thermal conductivity data [Vargaftik, 1975]


were approximated by linear functions of pressure and temperature in the range of
the reactor operating conditions.

Catalyst Deactivation

The main mechanisms of catalyst deactivation can be grouped into six types: (i)
poisoning, (ii) fouling, (iii) thermal degradation, (iv) 1,2-dichloroethane vapor
formation accompanied by transport, (v) vapor–solid and/or solid–solid reactions,
and (vi) attrition/crushing [Bartholomew, 2001].
The catalyst deactivation model used in this study is derived from
Boskovic et al. [2004a] for a commercial Ag/Ag2O3 catalyst. The equation
describes catalyst activity decline as a function of time with a term of steady-state
activity, ass, reached at infinite time [Fuentes, 1985]:

da
− = K d POα2 ( a − a ss ) d
dt Eq.16

Boskovic et al. [2004b] used a power-law giving index of 1 (d=1):

Ea
a = a ss + (1 − a ss ) exp( − K d0 ( − ) POα2 t ) Eq.17
RT

The catalyst deactivation is modeled using Eq.17 which is a nonlinear function of


operating time and catalyst temperature. Data from the catalyst vendor were used
to evaluate the catalyst deactivation parameters (with subscripts ass and kd).
The temporal profiles of a1 and a2, the catalyst deactivation functions,
introduced for the partial and complete oxidation reaction rate equations of
ethylene, are shown in Figure 3 for a period of three years.

http://www.bepress.com/cppm/vol4/iss1/14 8
DOI: 10.2202/1934-2659.1231
Aryana et al.: Modeling and Optimisation: Industrial Ethylene Oxide Reactor

Deactivation 0.9 Deact 1

0.8

0.7
Deact 2
0.6

0.5
0 200 400 600 800 1000
Time (day)

Figure 3- Catalyst deactivation profiles for a period of 3 years operating time

Model Predictions and Industrial Reactor Data

The catalyst temperature has a major effect on the reactor performance, catalyst
aging and safety. Since no plant data are available for the catalyst temperature,
data for the oil and gas outlet temperature were used to validate our model. Figure
4 and Figure 5 show the oil and gas outlet temperatures of the model predictions
against plant data during 3 years of reactor operation. The model predictions are
in reasonable agreement with plant operating data. However, some discrepancies
exist due to error in measurement related to model input data. We observe that the
oil inlet temperature increases with time to compensate for the loss in catalyst
activity and maintain desired production rate and selectivity.

Published by Berkeley Electronic Press, 2009 9


Chemical Product and Process Modeling, Vol. 4 [2009], Iss. 1, Art. 14

270
Plant Data
265
Model Prediction
260
Temp (C) 255

250

245

240

235

230
0 200 400 600 800 1000
Time (day)

Figure 4- Comparison of outlet oil temperature predictions with plant data during
3 years operation

270
Plant Data
265
Model Prediction
260

255
Temp (C)

250

245

240

235

230
0 200 400 600 800 1000
Time (day)

Figure 5- Comparison of outlet gas temperature predictions with plant data during
3 years operation
The other important parameters are selectivity and work-rate. These two
parameters represent the EO reactor efficiency and product rate, respectively. The
selectivity (Sel%) and the work-rate (WR) definitions we used are given as
follows:

EO produced
Sel % = × 100 Eq.18
Total ethylene consumed

http://www.bepress.com/cppm/vol4/iss1/14 10
DOI: 10.2202/1934-2659.1231
Aryana et al.: Modeling and Optimisation: Industrial Ethylene Oxide Reactor

WR = EO produced (T / day ) Eq.19

Comparisons of the selectivity and the work-rate between the model


prediction and the plant data are shown in Figure 6 and Figure 7, respectively. The
selectivity is decreasing over the period of catalyst life time due to loss in catalyst
activity and the work-rate changes are related to the market demand.

90%
Plant Data
Model Prediction
85%
Selectivity (%)

80%

75%

70%
0 200 400 600 800 1000
Time (day)

Figure 6- Comparison of catalyst selectivity predictions with plant data during


3 years operation

Overall, the model predictions for the selectivity and the work-rate against
the plant data are satisfactory. Whenever the plant data for the input variables
were available, the model predictions were very close to plant data. However,
when accurate plant data are unavailable, we noticed disagreements between
model predictions and plant data. For example, EDC concentration has significant
effect on the selectivity and work-rate; however, accurate data on EDC
concentration (ppm levels) were not always available. Despite the uncertainties in
plant data and unknown disturbances, our model prediction show reasonable
agreement with the plant data.

Published by Berkeley Electronic Press, 2009 11


Chemical Product and Process Modeling, Vol. 4 [2009], Iss. 1, Art. 14

160
Plant Data
Model Prediction
Work Rate (T/Day)
140

120

100

80
0 200 400 600 800 1000
Time (Day)

Figure 7- Comparison of work-rate predictions with plant data during


3 years operation

Sensitivity Analysis

Sensitivity analysis was undertaken to check the sensitivity of the reactor


performance, using the validated model, to inlet variables such as coolant and
feed temperature, EDC level, and reactant concentration.

Coolant and feed temperature


The inlet temperature of oil as coolant is one of the important control variables
which could control possible thermal runaways as well as the reactor
performance. There is a limitation on the allowable temperature rise of the coolant
as it passes through the reactor. On one hand, it is advantageous for the
temperature to rise large enough to allow efficient heat transfer in the secondary
heat exchanger for steam generation, but there is also strong motivation to
minimize this temperature rise in order to reduce thermal shocks in the reactor.
Figure 8 shows the effect of step increasing (+5%) the oil inlet temperature (@ t =
0 s) on the temperature profile along the reactor. The disturbance propagates in a
linear fashion and the transition time is less than 100 s due to the high gas velocity
and fast reactions.

http://www.bepress.com/cppm/vol4/iss1/14 12
DOI: 10.2202/1934-2659.1231
Aryana et al.: Modeling and Optimisation: Industrial Ethylene Oxide Reactor

280

270
Time = 100 Sec
260

Gas Temperature (ºC) .


250

240 Time = 0 Sec


230

220

210

200

190
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5
Distance along the reactor (m)

Figure 8-Temperature profile in reactor with +5% step-change in T-oil-inlet

We note that the first and the last half meter of the 7.35 meter long tubes
have no catalyst which means no reaction, so, just heat exchange between the oil
and gas take place. During the first half meter of tubes and the last half meter,
respectively, the model shows a significant rise and small decline in the gas
temperature due to heat exchanger between gas and oil.
Figure 9 shows the effect of step increasing (+5%) the oil inlet temperature
(@ t = 0 s) on the work-rate and selectivity. With increasing oil temperature, the
work-rate increases by about 12% while selectivity decreases by 2.5% as be
expected, since the reaction rates of all the three reactions increase with
temperature and the work-rate increases with increasing oil temperature.
However, a decrease on selectivity is expected due to stronger effect of
temperature on the undesired reaction (the total ethylene and EO oxidation) than
the desired one (partial oxidation of ethylene).
130 81.5

81
125
80.5
Selectivity (%)

Work-rate
Work Rate (Ton/day)

Selectivity 80
120
79.5

79
115
78.5

110 78
0 40 80 120 160 200
Time (s)

Figure 9- Selectivity and work-rate with +5% step-changes in oil inlet temperature

Published by Berkeley Electronic Press, 2009 13


Chemical Product and Process Modeling, Vol. 4 [2009], Iss. 1, Art. 14

For further investigation, the amount of heat transferred between the tube
wall and the coolant oil is shown in Figure 10. This graph shows that in the
beginning, the amount of transferred heat drops dramatically due to the step
increase of shell side temperature (oil inlet temperature) which results in
decreasing the temperature difference between the shell side and the gas phase.
Consequently, the bedside temperature starts to increase enhancing the reactions
thereby increasing the released heat by reactions and the reactor inside
temperature. Thus, heat transfer starts to increase. The dramatic decrease in heat
transfer rate could lead to a thermal runaway condition and catalyst damage.

248 3000

246 2500

Heat Removed (J/s)


244
Temperature (C)

2000
242
Oil Inlet temp 1500
240 Heat Transferred
1000
238
Possible
236 Runaway 500
Condition
234 0
0 10 20 30 40 50 60 70 80 90 100
Time (s)

Figure 10- The oil inlet temperature and transferred heat between oil and tube wall
in response to a +5% step-change in oil inlet temperature

During industrial operation, use of a linear-change is better suited than a


step-change. Figure 11 shows the modification of oil inlet temperature according
to a linear increase and the amount of heat removed from tube side. As seen, the
heat transferred is increasing smoothly and there is no dramatic change.
Figure 12 shows the selectivity and work-rate following a linear-change. In
both cases the final outcomes are similar but the reactor performance is expected
to be better in case of linear increase. So, to avoid any runaway condition, linear
increase is suggested.

http://www.bepress.com/cppm/vol4/iss1/14 14
DOI: 10.2202/1934-2659.1231
Aryana et al.: Modeling and Optimisation: Industrial Ethylene Oxide Reactor

250 2700

2500
246

Heat Removed (J/s)


Temp. (ºC) 2300
242
2100
238
1900
Oil Inlet Temp
234 Heat Transferred 1700

230 1500
0 10 20 30 40 50
Time (min)

Figure 11- Oil inlet temperature and heat transferred between oil and tube wall
with +5% linear change in oil inlet temperature

130 81.5
Work-rate
81
Selectivity
125
80.5 Selectivity (%)
Work Rate (Ton/day)

80
120
79.5

79
115
78.5

110 78
0 10 20 30 40 50
Time (min)

Figure 12- Selectivity and work-rate with +5% linear change


in oil inlet temperature

Published by Berkeley Electronic Press, 2009 15


Chemical Product and Process Modeling, Vol. 4 [2009], Iss. 1, Art. 14

EDC level

The 1,2-dichloroethene (EDC) helps enhance process selectivity by reducing the


availability of active sites on the silver/alumina catalyst. Figure 13 shows the effect
of EDC level on the selectivity and work-rate, the disturbance propagates in a
nonlinear fashion. With the addition of EDC, there is a shift in the rates of
reactions, resulting in a relatively smaller decrease in the amount of EO formed.
Thus, the selectivity for EO increases, although the work-rate decreases.
Figure 14 shows the production rates of ethylene oxide and carbon dioxide
decrease with increasing level of EDC which ethylene oxide decreases to a
smaller extent (about 28%) compare to carbon dioxide decrease of 10%. This
results in a decreased load on the downstream separation units.
125 89
88
Work-rate
87
115 Selectivity 86

Selectivity (%)
Work Rate (Ton/day)

85
84
105
83
82
95 81
80
79
85 78
1 1.4 1.8 2.2 2.6 3
EDC Level (ppm)

Figure 13- The influence of EDC level on work-rate and selectivity

110
100
90
80
Mole Flow (mol/s)

70 EO Mole Flow

60 CO2 Mole Flow

50
40
30
20
10
1 1.4 1.8 2.2 2.6 3
EDC Level (ppm)

Figure 14- The influence of EDC level on product mole flow (EO and CO2)

http://www.bepress.com/cppm/vol4/iss1/14 16
DOI: 10.2202/1934-2659.1231
Aryana et al.: Modeling and Optimisation: Industrial Ethylene Oxide Reactor

Reactant

The influences of reactant concentrations were studied by varying oxygen


concentration (5-8 mol%) with constant ethylene (17.5 mol%), and by varying
ethylene concentration (15-19 mol%) with constant oxygen concentration (6.16
mol%). As shown in Figure 15, with increasing molar concentration of oxygen in
the feed, work-rate increases significantly while selectivity decreases about 4%.
Also, Figure 16 shows that with ethylene addition in the feed, work-rate increases
significantly and selectivity increases slightly.

150 83

82.5
140 Work-rate
Selectivity 82
130 81.5

Selectivity (%)
Work Rate (Ton/day)

81
120
80.5
110
80

100 79.5

79
90
78.5

80 78
5 5.5 6 6.5 7 7.5 8
Oxygen Mole %

Figure 15- The influence of inlet oxygen concentration on work-rate and selectivity
150 81.5

140
81.1
130
Work-rate
Selectivity (%)
Work Rate (Ton/day)

Selectivity 80.7
120

110
80.3

100
79.9
90

80 79.5
15 16 17 18 19
Ethylene Mole %

Figure 16- The influence of inlet ethylene concentration on work-rate and selectivity

Published by Berkeley Electronic Press, 2009 17


Chemical Product and Process Modeling, Vol. 4 [2009], Iss. 1, Art. 14

Optimisation

For the EO reactor, higher temperature leads to excessive carbon dioxide and
water formation which results in loss of selectivity. Lower temperatures result in
lower conversion and loss of productivity. Thus optimisation is to maximise
conversion of ethylene to EO (selectivity) within the flammability limit for safe
operation. The objective function is defined to maximise selectivity for desired
production rate (work rate).
Flammability is a major constraint of the EO reactor. The flammability
limit gives the proportion of combustible gases in a mixture, between which limits
this mixture is flammable. It is known that in actual practice in order to remain
outside the flammability limit of the gas mixture the concentration of oxygen has
to be lowered as the concentration of ethylene is raised. The actual safe operating
ranges depend, along with the gas composition (reactants and balance gases), also
on individual plant conditions such as temperature and pressure. Therefore in each
individual plant a flammability equation is used to determine the concentration of
oxygen which may be used with any given concentration of ethylene. The
flammability equation can be expressed via a flammability curve.
The flammability limit is calculated based on the maximum allowable
oxygen limit using equation provided by industry patent. Based on our sensitivity
analysis, the coolant temperature, reactant concentration and EDC level in the
feed, which significantly affect the reactor performance, can be manipulated for
maximal profit. EO has a significantly negative effect on selectivity and work-
rate. In practice, its concentration is kept as low as possible. The inlet feed
temperature has insignificant effect on productivity and is unsuited as a
manipulated variable. Steady state optimisation was carried out for 2 cases:
¾ Case 1: Low production rate (110 Ton/day EO)
¾ Case 2: High production rate (130 Ton/day EO)
The constraints and the final values for first and second case studies are shown in
Tables 3 and 4, respectively.

Table 3- Constraints for low production rate

Optimum Lower Upper


Constraint
Value Bound Bound
Work-rate (Ton/day) 111 110 111
Oil Outlet Temperature (ºC) 237.6 200 280
Gas Outlet Temperature (ºC) 239.8 200 275
Flammability Limit (%) 0.6 0.3 1

http://www.bepress.com/cppm/vol4/iss1/14 18
DOI: 10.2202/1934-2659.1231
Aryana et al.: Modeling and Optimisation: Industrial Ethylene Oxide Reactor

Table 4- Constraints for high production rate


Optimum Lower Upper
Constraint
Value Bound Bound
Work-rate (Ton/day) 130 129 131
Oil Outlet Temperature (ºC) 242.0 200 280
Gas Outlet Temperature (ºC) 244.0 200 280
Flammability Limit (%) 0.3 0.3 1

The control variables and their optimum values the first and second case
studies are shown in Table 5 and 6, respectively. The EDC level, ethylene and
oxygen concentration in the feed and oil inlet temperature are chosen as control
variables. At the start of the operation, the results show that in both low and high
production cases, ethylene and EDC concentrations act as active bounds.
Furthermore, in high production rate (second case), the oxygen concentration acts
as lower active bound according to the flammability point.

Table 5- Control variables for low production rate


Optimum Lower Upper
Control Variable
Value Bound Bound
Oil Inlet Temperature (ºC) 233.8 230 260
Ethylene Inlet (Mol. Frac. %) 17.5 15 17.5
Oxygen Inlet (Mol. Frac. %) 6.48 5 7.5
EDC Level (ppm) 1.5 1.2 1.5
Optimum
81.7%
(Selectivity %)

Table 6- Control variables for high production rate


Optimum Lower Upper
Control Variable
Value Bound Bound
Oil Inlet Temperature (ºC) 234.1 223 260
Ethylene Inlet (Mol. Frac. %) 19 16 19
Oxygen Inlet (Mol. Frac. %) 7.05 5 7.5
EDC Level (ppm) 1.5 1 1.5
Objective Function Value
81.0%
(Selectivity %)

In industry, usually oxygen concentration is kept constant at 7% to avoid


the flammability limit. In low production case, the required inlet oxygen is less
than 7%. As seen in Figure 15, oxygen concentration has a negative effect on

Published by Berkeley Electronic Press, 2009 19


Chemical Product and Process Modeling, Vol. 4 [2009], Iss. 1, Art. 14

selectivity. Also the results show that the optimum selectivity in the first case
(low work-rate) is higher than the second case (high work-rate). To increase the
work-rate, the oil inlet temperature should be increased to enhance the reactions,
although the selectivity would be decreased. Nevertheless, the result shows that
the selectivity could be improved by keeping the control variable close to the
optimal point.

Dynamic Optimisation

The dynamic optimisation problems can be represented in the following algebraic


form:
Max ( z (t f )) Eq.20
t f ,u ( t ),v ,t∈[ 0 ,t f ]

F ( x& (t ), x(t ), y (t ), u (t ), v) = 0 , t ∈ [0, t f ] Eq.21

I ( x& (0), x(0), y (0), u (0), v) = 0 Eq.22

t min
f ≤ t f ≤ t max
f Eq.23

u min (t ) ≤ u (t ) ≤ u max (t ) , t ∈ [0, t f ] Eq.24

v min ≤ v ≤ v max Eq.25

weimin ≤ w(t f ) ≤ weimax Eq.26

wee (t f ) = weetgt Eq.27

wiimin (t ) ≤ w(t ) ≤ wiimax (t ) , t ∈ [0, t f ] Eq.28

w min (t ) ≤ w(t ) ≤ w max (t ) , t ∈ [0, t f ] Eq.29

Equation 20 indicates that the optimisation (maximisation) is performed


considering the variable z (t f ) as performance measure or objective function.
Without any loss of generality, the objective function is simply the magnitude of a
variable z (t ) evaluated at the end of the optimisation horizon t = t f . In addition,
Equation 20 denotes the fact that the decision variables of the optimisation
problem are the time horizon t f (a scalar) and a subset of variables given by the
vectors u (t ) and v . The former denotes control variables that are allowed to vary

http://www.bepress.com/cppm/vol4/iss1/14 20
DOI: 10.2202/1934-2659.1231
Aryana et al.: Modeling and Optimisation: Industrial Ethylene Oxide Reactor

according to the functionality u (t ) over the span of the time horizon [0, t f ] . The
latter indicates parametric variables that are fixed at a value.
Equation 21 denotes the set of differential-algebraic equations (DAEs)
representing the process model, while equation 22 symbolises a set of additional
equations (initial conditions) that must be satisfied at the beginning of the
optimisation horizon. In these equations x and y denote differential and algebraic
variables respectively, and indicates the time derivatives of x. Equations 23, 24
and 25 denote lower and upper bounds on the decision variables, indicated by the
superscripts min and max, respectively.
There are also other types of constraints and importantly those which
bound the decision variables and in some cases the time horizon of the process.
Further constraints are in the form of interior point constraints, path constraints
and end-point constraint. The end product concentration must be greater than
99%, a purity demand. This constraint can be formulated as an inequality where
the constrained condition must be less than a desired value (Equation 26), greater
than a desired value or within a desired region or may be an equality constraint
where the final value must equal a desired value (Equation 27).
Equation 28 denotes interior-point constraint variables, which are used to
enforce process variables to lie within the defined upper and lower bounds at any
other time except the end of the optimisation horizon. By definition, these are
inequality constraints. Although the inclusion of these constraints is not strictly
essential, research [Vassiliadis et al., 1994] demonstrated that adopting them may
increase the robustness and efficiency of algorithms handling inequality
constraints. Eq. 29 symbolises the inequality path constraints mentioned above.
Two case studies were considered:
¾ Case 1: Optimal coolant temperature profile during 3 years operating
time.
¾ Case 2: The optimal profiles for coolant temperature and oxygen inlet
concentration with transition between low production (110 Ton/day) and
high production rate (127 Ton/day).
Results from steady state optimisation were used as initial conditions for
the dynamic optimisation. Since the catalyst decays continuously, the production
process is dynamic and a dynamic optimisation strategy was used. However, it is
infeasible to simultaneously change all the variables dynamically. Experience
shows that the improvement of the optimisation performance with more than one
variable in dynamic optimisation is minor over that when only one variable is
varied dynamically while others are fixed to constant sub-optimal values [Bonvin,
1998]. The variable that is most sensitive to dynamic behavior is bed temperature,
which is responsible for thermal sintering and catalyst decay.Activity losses are
attributed to a decrease of the catalyst surface area, by poisoning or sintering, and

Published by Berkeley Electronic Press, 2009 21


Chemical Product and Process Modeling, Vol. 4 [2009], Iss. 1, Art. 14

a monotonic decline in selectivity is implicated as the temperature is increased to


obtain a desired total reaction rate (work-rate).

Case 1

Figure 17 shows the oil inlet temperature optimal profile to obtain the desired
work-rate (127 Ton/day) over the 3 year operation which maximises selectivity.
As observed, the oil inlet temperature needs to be increased by 15 ºC to
compensate for catalyst deactivation. By increasing oil inlet temperature, it is
expected that the selectivity will decrease gradually. The loss of selectivity
according to the optimal condition is shown Figure 17.

250 81.2
T_Oil
TOil_In
In
81
247 Selectivity
Selectivity 80.8

Selectivity (%)
80.6
244
80.4
Temp (°C)

80.2
241
80

238 79.8
79.6
235 79.4
0 6 12 18 24 30 36
Time (month)

Figure 17- Optimal oil inlet and selectivity losses profile for 3 years operating time

Case 2

Oil inlet temperature and oxygen inlet concentration were used as control
variables to shift work-rate from 110 (Ton/day) to 127 (Ton/day) and objective
function was to maximise selectivity. The ethylene inlet concentration and the
EDC levels are kept constant at 17.5 mol% and 1.5 ppm, respectively. It was
assumed that the transition operation takes place in 2 hours. To avoid a runaway
condition, linear increase was suggested. Therefore, piecewise-linear control was
used for the oxygen inlet concentration and oil inlet temperature.
Figure 18 shows the optimal transition profiles for the oil inlet temperature
and the optimal transition profiles for the oxygen concentration. The results show
that to increase the work-rate from 110 (Ton/day) to 127 (Ton/day), the oil inlet
temperature and the oxygen concentration require to increase by 4 ºC and 0.5

http://www.bepress.com/cppm/vol4/iss1/14 22
DOI: 10.2202/1934-2659.1231
Aryana et al.: Modeling and Optimisation: Industrial Ethylene Oxide Reactor

mol%, respectively. Also, Figure 18 shows the change in work-rate during the
transition period. By increasing oil inlet temperature, the selectivity decreases by
1.2% due to increasing oil inlet temperature.
7 238

6.8 237
(Ton/day)
(%)

Temp (°C)
6.6 236
Percent

6.4 235
Rate

O2 Mole Percent
O2 Mole Frac %
O2 Mole

T_Oil
Toil-In
In
Work

6.2 234

6 233
0 2000 4000 6000 8000
Time (s)

Figure 18- Optimal transition profiles for oil inlet temperature


and oxygen inlet concentration

Conclusions

A one-dimensional heterogeneous dynamic model of the EO reactor was


developed. The partial and total ethylene oxidations as well as the EO oxidation
are considered with the inhibiting effect of dichloroethane as an inhibitor. Also
the process energy balances were developed and included in one model. The set
of nonlinear kinetic expressions are coupled with heat and mass conservation
equations and variable physical properties along the reactor. The catalyst
deactivation kinetics was modeled as a function of time and oil inlet temperature
using the catalyst vendor data.
Plant data from Huntsman EO reactor were used to validate the model.
The results showed satisfactory agreement between the model predictions and
plant data. No fitting parameters were used for the model. The kinetic parameters
were obtained from external sources. For sensitivity analysis, the effects of inlet
variables such as coolant and feed temperature, EDC level, and reactant were
investigated. Using the validated EO reactor model, the effect of the oil and gas
inlet temperature, the EDC level, the ethylene and oxygen inlet concentration on
the reactor performance were investigated. The results showed that the EDC level
and oil inlet temperature have major effects on both selectivity and work-rate and
the gas inlet temperature has negligible effect on the reactor. Ethylene and oxygen
concentrations in the feed only affect the work-rate and selectivity, respectively.

Published by Berkeley Electronic Press, 2009 23


Chemical Product and Process Modeling, Vol. 4 [2009], Iss. 1, Art. 14

By increasing EO concentration in the recycle gas, the selectivity decreases


significantly, while it has negligible effect on the work-rate.
Steady state optimisation was carried out for two different scenarios: low
product (110 Ton/day EO) and high product (130 Ton EO/day). The EDC level,
the oil inlet temperature and the ethylene oxygen inlet concentration were used as
control variables. The flammability limit, the fundamental constraint in the EO
reactor and other constraints were introduced to the system. The objective
function in all scenarios was to maximise the selectivity.
The results from steady state optimisation study showed that the
ethylene concentration and the EDC level act as active bounds. In high production
rate, the oxygen concentration acts as lower active bound according to the
flammability point. The outcomes of steady state optimisation were used as initial
conditions for the dynamic optimisation. In dynamic optimisation study, firstly,
the oil inlet temperature was used as control variable to compensate for catalyst
activity losses during 3 years operating time to obtain desired work-rate (127
T/day) to maximise selectivity. The results compared to the vendor data indicated
that the selectivity could be improved by more than 2%. Optimal transition
profiles for the oil inlet temperature and the oxygen concentration were obtained
to shift the work-rate from 110 to 127 T/day to maximise the selectivity.

Nomenclature

ai catalyst deactivation term


A area [m2]
AC,T empty tube cross sectional inner area [m2]
ass catalyst deactivation in steady-state
BASE(i) component that the reaction i is based on
CP heat capacity [kJ/kmol °C]
Dax axial dispersion coefficient [m2/s]
DI tube inside diameter [m]
DO tube outside diameter [m]
Ds reactor shell diameter [m]
Ea Activation Energy [kJ/mol]
e catalyst bed voidage
fi reaction rate [mol O2/kg cat s]
Fi mole flow [kmol/hr]
G mass gas flow per cross area [kg/m2]
k1,3,7 kinetic coefficient [mol/gcat.atm2.hr]
k2,4,8 kinetic coefficient [mol/gcat.atm3.hr]
k5,6,9,10,11,12 kinetic coefficient [atm-1]
k do sintering rate constant

http://www.bepress.com/cppm/vol4/iss1/14 24
DOI: 10.2202/1934-2659.1231
Aryana et al.: Modeling and Optimisation: Industrial Ethylene Oxide Reactor

NT number of tubes
Pi partial pressure of component i [in bar unless noted otherwise]
S and SV specific surface area [m2/m3]
Sel Selectivity (%)
Ri overall reaction rate [kmol O2/m3 (cat) s]
t time [s]
T temperature [°C unless noted otherwise]
U heat tranfer (overall unless stated) [kW/m2 °C]
WR workrate (ton/day)
Vgas velocity [m/s]
Z length in tube direction

Greek letters

α heat transfer coefficient [kW/m2 °C]


υ stoichiometric coefficient
ρ density [kg/m3]

Subscripts

0 inlet condition
1 reaction 1 (C2H4 + 1/2O2 ⇒ C2H4O)
2 reaction 2 (C2H4 + 3O2 ⇒ 2CO2 + 2H2O)
3 reaction 3 (C2H40 + 5/2O2 ⇒ 2CO2 + 2H2O)
amb ambient
b catalyst bed
c cross (e.g. area)
cat catalyst
Eth ethylene
EO ethylene oxide
Oxy oxygen
EDC ethylene Dichloride
gas gas phase
H2O water
I inside tube
in inlet
Loss heat loss
O outside tube
oil cooling oil
out outlet
p particle
S shell

Published by Berkeley Electronic Press, 2009 25


Chemical Product and Process Modeling, Vol. 4 [2009], Iss. 1, Art. 14

T tube or total
wall tube wall

References

Azevedo S.F.D., Romero-Ogawa M.A., Wardle A.P. (1990). “Modelling of


Tubular Fixed-bed Catalytic Reactors: A Brief Review”, Trans IChemE, 68
(Part A): 483-502.
Bartholomew C. H. (2001), “Mechanisms of Catalyst Deactivation”, Applied
Catalysis A, 212: 17.
Bonvin D. (1998), “Optimal Operation of Batch Reactors – A Personal View”,
Journal of Process Control, 8: 355.
Boskovic G., Wolf D., Brückner A., Baerns M. (2004a), “Deactivation of a
Commercial Catalyst in the Epoxidation of ethylene to ethylene oxide—
Basis for Accelerated Testing”, Journal of Catalysis, 224: 187.
Boskovic G, Dropka N., Wolf D., Brückner A., Baerns M. (2004b), “Deactivation
Kinetics of Ag/Al2O3 Catalyst for ethylene Epoxidation”, J Catalysis, 226:
334.
Bradshaw A.V., Johnson A., Chiu Y-T., McLachlan N. H. (1970) “Heat Transfer
between Air and Nitrogen and Packed Beds of Non-reacting solids”, Trans.
Instn. Chem. Engrs, 48 p T77-T84.
Cornelio A.A. (2006), “Dynamic Modelling of an Industrial Ethylene Oxide
Reactor”, Indian Chemical Engr. V48 (3) D-6940, 5. edition, P117–135.
Drew T.B., Hoopes J.W., Vermeulen T. (1962), Advances in Chemical
Engineering. Academic Press, NY.
Eliyas A., Petrov L. and Shopov D., (1988), “Ethylene Oxide Oxidation over a
Supported Silver Catalyst. II. Kinetics of Inhibited Oxidation”, Applied
Catalysis, 41, 39-52.
Fogler, H. S. (1999). Elements of Chemical Reaction Engineering. Prentice Hall,
NJ.
Fuentes G.A. (1985), “Catalyst Deactivation and Steady State Activity: A
Generalized Power Law Equation Model”, Applied Catalysis 15: 33.
Hewitt G. F., Shires G. L., Bott T. R. (1994), Process Heat Transfer, 285-292,
CRC Press, Boca Raton.
Huntsman Corporation Australia. (2005), Ethylene Oxide Plant, Operating Data.
Khanna R., Seinfeld, J.H. (1987), “Mathematical Modeling of Packed Bed
Reactors: Numerical Solutions and Control Model Development”,
Advances in Chemical Engineering, 13:113–191.
Kishor G.G., Riggs J.B. (2003). “Bifurcation and Stability Analysis of an Ethylene
Oxide Reactor System”, Ind. Eng. Chem. Res. 42, 3285-3293.

http://www.bepress.com/cppm/vol4/iss1/14 26
DOI: 10.2202/1934-2659.1231
Aryana et al.: Modeling and Optimisation: Industrial Ethylene Oxide Reactor

Park D. W., Gau G. (1987), “Ethylene Epoxidation on a Silver Catalyst: Unsteady


and Steady State Kinetics”, Journal of Catalysis 1(105): 81-94.
Petrov L., Eliyas A., Shopov D., (1984), “A Kinetics of Steady State Ethylene
Epoxidation over a Supported Silver Catalyst”, Applied Catalysis, 18, 87-
103.
Petrov L., Eliyas A., Shopov D., (1986), “Kinetics of Ethylene Oxidation over a
Silver Catalyst in the Presence of Dichloroethane”, Applied Catalysis, 24,
145-161.
Petrov L., Eliyas A., Maximov C., Shopov D., (1988), “Ethylene Oxide Oxidation
over a Supported Silver Catalyst. I. Kinetics of Uninhibited Oxidation”,
Applied Catalysis, 41, 23-38.
Rebsdat S., (2001), “Ethylene Oxide”, Ullmann’s Encyclopedia of Industrial
Chemistry, John Wiley & Sons, NY.
Vargaftik N. B. (1975), “Tables on the Thermophysical Properties of Liquids and
Gases”, 2nd Ed., John Wiley & Sons, NY.
Vassiliadis V.S., Sargent R.W.H., Pantelides C.C. (1994b), “Solution of a Class of
Multistage Dynamic Optimization Problems: 2. Problems with Path
Constraints”, Industrial & Engineering Chemistry Research, 33: 2123.
Westerterp K.R., Ptasinski K. J. (1984). "Safe Design of Cooled Tubular Reactors
for Exothermic, Multiple Reactions; Parallel Reactions - II, the Design and
Operation of an Ethylene Oxide Reactor", Chemical Engineering Science,
39(2):245–252.
Zhou X.G., Yuan W.K. (2004), “Modeling Silver Catalyst Sintering and
Epoxidation Selectivity Evolution in ethylene Oxidation’, Chemical
Engineering Science, 59: 1723-1732.
Zhou X.G., Yuan W.K. (2005), “Optimization of the fixed-bed reactor for
ethylene epoxidation”, Chemical Engineering & Processing, 44: 1098–
1107.

Published by Berkeley Electronic Press, 2009 27

You might also like