Startup
of
an Industrial Adiabatic Tubular
Reactor
J.
W.
Verwijs
and
H.
van den Berg
Process Development
&
Control Dept., Dow Benelux N.V.,
4530
AA
Terneuzen, The Netherlands
K.
R.
Westerterp
Chemical Reaction Engineering Laboratories, Dept.
of
Chemical Engineering, Twente University
of
Technology,
7500
AE
Enschede, The Netherlands
The dynamic behavior of an adiabatic tubular plant reactor during the startup is
demonstrated, together with the impact of
a
feed-pump failure of one
of
the reac-
tants.
A
dynamic model of the reactor system is presented, and the system response
is calculated as
a
function of experimentally-determined, time-dependent, manip-
ulated variables. The values of modelparameters are estimated by using the SimuSolv
(
I991
)
computer program. The data set collected during the reactor start-up is used
for the parameter estimation procedure. An excellent agreement is obtained between
the experimental and the calculated system response. Many continuously-operated
commercial reactors require
a
complete conversion of one
of
the main reactants at
the reactor exit. It is shown that for an industrial tubular reactor
a
much higher
initial reactor temperature is required during the startup, compared to the reactor
inlet temperature during normal steady-state operation, to ensure
a
complete reactant
conversion. Much more research is necessary to determine whether this is
a
generally
valid rule.
Introduction
Higher safety standards, more stringent environmental reg-
ulations, better management
of
energy and raw materials, and
increased product quality requirements have caused the struc-
ture of continuously-operated chemical processes to become
increasingly complex and the process operating limits tight-
ened. Therefore, understanding
of
the physical and chemical
phenomena is required to set up safe, simple and effective
operating procedures, both for steady-state and for dynamic
operations. Process control is critical to running a continuously-
operated chemical plant successfully within the operating lim-
its, but basically the process control system is a translation of
the strategy of “how to run the process” into controllers and
related subjects. Therefore, operating procedures (specifica-
tion of the equipment operation) are key issues for a successful
operation, because they are intermediates between process de-
sign (equipment specification) and process control (equipment
operation).
Correspondence concerning this article should be addressed to
K. R.
Westerterp.
Operating procedures for continuously-operated chemical
plants consist of two parts: the first part describes the pro-
cedures during normal operation (steady-state operation); the
second describes the procedures for start-up, switch-overs, and
shutdown (dynamic operation). Often operating limits during
the normal operation are determined primarily by the process
economics. Hence, during the past decades, the chemical in-
dustry has put much effort on steady-state process optimiza-
tion, resulting in increased process yields, improved product
quality, and a reduced energy consumption and environmental
pollution.
The most complex problems in defining operating proce-
dures
is
to quantify the process dynamics and
to
specify the
operating limits during dynamic operation. Usually in a proc-
ess, a number
of
variables are available which can be adjusted
freely within the operating limits by the plant operator and/
or the process control system. The selection
of
correct variables
for control purposes, along with the specification of the op-
erating windows of these variables, can be a crucial step.
Also
AIChE
Journal
December
1992
Vol.
38.
No.
12
1871
the sequence of putting equipment into
or
out
of
service, to-
gether with the necessary optimum process-related conditions,
and the impact of unexpected process interruptions due to
equipment failures bear the same level of importance.
Most published studies on tubular reactor dynamics are the-
oretical and concerned with the characterization of the phe-
nomenon taking place in fixed-bed reactors. Some studies
available take operating limits into account.
Hahn et al.
(1971)
studied the start-up of
a
jacketed tubular
reactor from uniform initial conditions to predefined steady-
state conditions by manipulating the wall temperature of the
reactor. A distributed maximum principle is used to drive the
system from the initial state toward the final steady state by
minimizing the spatial integral of the weighted sum of the
squared concentration and temperature deviations from the
desired steady-state profiles, integrated over the transient start-
up period of a fixed length. The proposed method can also be
used when operating limits have to be taken into account.
Mann et al.
(1979, 1980)
developeci a dynamic model of a
SO2 fixed-bed oxidation reactor to determine improved start-
up procedures, which are required to reduce
SO2
emissions to
atmosphere during the start-up
of
a sulphuric acid plant. A
number
of
model simulations were performed to determine
the influence of some variables, and significant differences are
found on SO2 emission levels. The suggestion is made to de-
termine the optimal mode for the manipulated variables during
start-up via the calculus of variations and Pontryagins maxi-
mum principle. Laboratory studies were carried out by Mann
et aI.
(1986)
to prove some model assumptions.
It
did not lead
to principle changes in the philosophy to define start-up pro-
cedures compared to the preceeding studies of Mann et al.
(1979, 1980).
In general, the reactor dynamics including reactor start-up
are studied within a certain mathematical framework, such as
impact of different types
of
boundary conditions and step
changes of adjustable variables. Often laboratory experiments
are carried out to prove the theoretical assumptions for heat-
and mass-transfer mechanisms in the catalyst bed and/or par-
ticles,
or
to prove the predicted system response as a function
of the variables studied within the boundaries
of
the indicated
mathematical framework, without a direct link to industrial
practice. This “limited research scope” has to be extended to
operating procedure synthesis and nonsteady-state process
control to reduce environmental pollution and to improve
process safety, because most accidents in chemical plants occur
during dynamic operation (Amundson et al.,
1988).
Haastrup
(1983)
concluded from a study of accident case stories that the
frequency rate, on a time basis, of design related accidents in
continuously-operated chemical plants seems to be at least one
order of magnitude higher during dynamic operation than
during normal running conditions. Therefore, the current
process engineering practice of designing continuously-oper-
ated chemical plants for steady-state running conditions poses
a related research problem for operating procedure synthesis
and nonsteady-state process control (Amundson et al.,
1988).
Two types of knowledge can be used in plant-wide operations
(Stephanopoulos,
1988):
a) Declarative knowledge, based on first principles, char-
b) Procedural knowledge, representing the methodologies
acterizing the behavior
of
processing units
employed by plant operating personnel and process control
systems to operate the process.
Both types of knowledge are required to study operational
aspects of a reactor startup. A dynamic process model, based
on first principles, should answer questions about reactor be-
havior and operating limits.
A
representation
of
the operating
procedures should answer questions about “how” and “when”
to start up a reactor system, because a reactor startup cannot
be isolated from an entire plant startup.
To our best knowledge,
no
results are published about re-
actor dynamics and operating procedures of commercial re-
actor systems. This article describes the startup of an industrial
adiabatic reactor, as well as the impact of
a
failure
of
the feed
pump of one
of
the reactants.
A
dynamic model, characterizing
the behavior
of
the system, is presented, and the values
of
the
model parameters are estimated by using the SimuSolv
(1991)
computer program. Some research results on the synthesis
of
reactor operating procedures for the startup
of
a
continuously-
operated chemical plant will be discussed in a future article.
Plant
Reactor System
The plant reactor system consists
of
a feed mixer,
a
preheater
and a series
of
seven horizontal vessels with baffles. The first
vessel
is
bigger in size than the others, which have all the same
dimensions.
A
sketch
of
the reactor is given
in
Figure
1.
Also
indicated are the locations
of
the relevant flow devices
(fl
and
thermoelements
(T).
The reactor is insulated and located out-
doors in an open structure. The total volume of the mixer and
preheater is less than
1%
of the total reactor volume. Basically
the reaction starts already in the mixer, but the reactant con-
I
m
I
i
T
I
Process
1
43d
+
Figure
1.
Plant reactor.
1872
December
1992
Vol.
38,
No, 12
AIChE
Journal
version in the mixer and preheater will be very small due to
the relatively short residence time and low process tempera-
tures. Hence, the thermoelement in the preheater outlet is
considered as being the location
of
the reactor inlet.
Thermoelements are located at several positions between
reactor inlet
(z
=
0)
and reactor outlet
(z
=
1).
The relative po-
sition
(z)
of
a thermoelement is calculated by taking the ratio
of the volume from the reactor inlet up to the particular ther-
moelement and the total reactor volume, according to the
model assumptions.
The operation of the reactor system is computer-controlled,
including the startup
or
shutdown
of
pumps and opening
or
closing of block valves (not indicated in Figure
1).
Process
data can be logged at regular time intervals from the process
control computer.
For
this study, the whole scheme of reactions carried out in
the plant reactor is simplified to one overall reaction that
describes the consumption
of
the main reactant
B:
A+B-C
This exothermal reaction is carried out in the liquid phase at
a pressure level sufficiently high to avoid boiling. Reactant
A
is fed in excess, because reactant
B
should be totally converted
at the reactor exit. The excess amount
of
reactant
A
is also
used to absorb the heat
of
reaction and limit the adiabatic
temperature rise.
The conversion level of reactant
B
is the most important
operating requirement. Therefore, a minimum average reac-
tion rate and a sufficient residence time in the system are
necessary to ensure the required conversion level of reactant
B
during reactor operation. Other operating limits resulting
from the process economics are beyond the scope of this article.
Reactor Startup
The reactor system is filled up with
a
mixture
of
reactant
A
and final product
C
from storage before startup
to
create
a
mixture in the reactor that can be processed by the downstream
plant section. This is done via the reactor feed system and with
the preheater in service. During this period, the inert gases
present in the reactor system are vented to atmosphere via the
downstream plant section. This step is necessary only when
the system had been emptied for inspection and/or mainte-
nance, and can be skipped from the startup sequence if the
reactor was not emptied beforehand.
First, reactant
A
is fed into the reactor. Reactant
B
is added
into the system, and the preheater is put into service at the
same moment to control the reactor inlet temperature, when
the
flow
of
reactant
A
has reached the required setpoint.
The flow of reactant
A
is controlled in ratio with the flow
of reactant
B,
as soon as the flow of reactant
B
has reached
a certain minimum flow setting. The total reactor feed is in-
creased to minimum plant capacity, if the flow
of
both reac-
tants is at the required initial setpoints and the outlet
temperature of the reactor is above a certain minimum tem-
perature limit.
The flow
of
both reactants as
a
function
of
the dimensionless
time
u
is shown in Figure
2.
The dimensionless variables are
defined in the Notation Section. In this figure the flow is
expressed as
a
percentage
of
the particular flowmeter range.
AIChE
Journal
December
1992
50
Em
$30
5
20
Em
$30
5
20
1°2
0
I/
I
1°4
0
0
1
2
3
4
5
6
CT
(-)
Figure
2.
Observed
flow of
reactant
A
and
B
as a func-
tion
of
time.
The time
u
is scaled by taking the time origin
(a
=
0)
just before
reactant
B
is fed into the system. The flow
of
reactant
A
at
a=O
is already at the required capacity. At
u=0.10,
the feed
pump of reactant
B
is started up. During the period
u=
1.47
until
u=
1.70,
no reactant
B
is fed into the system due to a
pump failure. The reactant
B
feed pump is restarted at
u
=
1.70.
At
u=
3.00,
the reactor feed is ramped up to a minimum plant
capacity which is reached at
a=5.50.
The initial dimensionless temperature profile
O(z,
0)
at
u
=
0
over the entire reactor is shown in Figure
3.
The dots in this
figure are the actual data, and the solid line is an approximation
of the temperature values between the thermoelements, ac-
cording to the Akima
(1970)
method to produce a smooth
curve. These approximated values are used in the model
of
the
reactor system.
The initial temperature profile depends on the following
important conditions:
The way the system is preheated and filled up by the plant
operator
The period between filling
or
reactor stop and the actual
reactor startup in view of the heat loss to the environment
The time span between the start
of
the reactant
A
feed
pump and putting the reactant
B
feed pump into service
The temperature and amount of reactant
A
fed into the
system before starting the reactant
B
feed pump.
8."-
,
0.88-l
,
.
.
0.0
0.2 04 06
08
10
z
(-)
Figure
3.
Temperature profile over the entire reactor
length at
u
=
0.
Vol.
38,
No.
12
1873
1021
I
1.W-
098-
096-
J?
0941..
.
,
.
,
.
. .
I
.
.
,
.
.
,
. .
1
0
1
2
3
4
5
6
u
(-)
Figure
4.
Observed reactor inlet temperature as
a
func-
tion
of
time.
The dimensionless reactor inlet temperature
vg
as a function
of time is shown in Figure 4. The shape of this curve is influ-
enced strongly by the step changes in the total flow through
the system due to the start or stop
of
the reactant
B
feed-pump
and putting the reactor preheater into or out
of service at the
same moment.
The temperature
8
over the entire reactor is shown in Figure
5.
In this figure, the lines parallel to the u-axis represent the
response of the thermoelements at the dimensionless location
z,
and the lines parallel to the z-axis connect the data at the
same moment. The overall error of the data
is
estimated to be
I
5%
for
the flow
of
reactant
A,
I
3%
for the flow of
reactant
B,
and
50.5%
for the temperature.
Reactor
Model
The performance of this plant reactor indicates that a tubular
reactor model can be used to describe the system. The following
assumptions are made to define the model:
1.
The reactor volume used in the model is equal to the total
volume of the vessels and the piping in the system.
2.
To represent the plant reactor as an empty tube reactor,
the inside tube diameter
(d,)
or tube length
(L)
should be
defined. The tube diameters of six equal-sized vessels are used
to calculate the tube length.
3.
Approximately
8%
of the heat produced by reaction is
absorbed by the vessel wall during startup. Hence, the energy
take-up in the reactor vessels is included in the model. The
total amount
of
construction materials of the vessels and the
piping is used to calculate an average outside tube diameter
(d2)
to be used
in
the model. The heat transfer coefficient
(v)
between fluid and tube wall is assumed to be constant. Heat
transported through the tube wall by conduction in the axial
direction is neglected.
4. Heat take-up in the insulation blanket and heat losses to
the surroundings are neglected.
5.
Due to the system geometry the actual flow pattern will
not be plug flow. The deviation
of plug flow will be described
by axial dispersion. There is no reason to distinguish between
axial and radial dispersion in the model, due to the simplifi-
cation of the system geometry. Mass
(Dux)
and heat
(ha)
dis-
persion coefficients are assumed to be constant over the entire
reactor length despite the geometry variations.
6.
The reaction scheme is simplified to one reaction, de-
scribing the consumption
of
reactant
B.
This reaction is as-
sumed to be irreversible and first-order with an Arrhenius-
type rate constant
(k,
=
klo.exp[-El/(R.T)J).
7.
Physical property values normally change during reac-
tion, due to changes in the reaction mixture composition and
temperature. Calculations showed that the fluid density varies
less than
2%
over the entire reactor length under steady-state
conditions. In this particular case, the temperature dependence
of
the fluid density
(p)
is compensated by the composition
change due to reaction. Hence, the fluid density is assumed to
be constant. The Rackett equation of the ASPEN Plus program
(1990)
is used for the fluid density calculations.
8.
The reaction enthalpy value
(
-
AHr)
of the main reaction
is used in the model. The fluid heat capacity value
(C,)
is
calculated from the total temperature rise over the entire re-
actor length under steady-state conditions. Both properties are
assumed to be constant.
9.
The density
(pw)
and heat capacity
(C,,)
values of the
construction materials
of
the reactor are assumed to be con-
stant.
With these assumptions the following equations are ob-
tained:
Component mass balances:
a2cA
ac,
a2c,
(1)
ac.4
ac*
at
ax
acB
at ax
a2cc
acC
ace
at ax
a2
-
u,*-+
Dux.--
a2
kICB
--
-
-
-
u,.-+DD,.--
(2)
a2
kICB
--
(3)
-vt*---+Da.-+klCB
-=
Energy balance for the
fluid:
aT
h
a2T
4.u
-v,.-++.---
(
T- Tw)
_-
-
6
aT
CJ
(-)
at ax
p.cp
a2
p.c,.dl
k,CB
(4)
(
-
AHr)
P.CP
Figure
5.
Observed reactor temperature
vs.
reactor
lo-
--.
cation and time.
December
1992
Vol.
38, No.
12
AIChE
Journal
1874
Energy balance for the reactor vessel:
The fluid velocity
u,
or
in dimensionless notation
4"
is a time-
dependent function, see Figure 7. The reactor wall temperature
T,
at time
t
=
0
is assumed to be equal to the fluid temperature.
Hence,
Initial
conditions:
t=O;
CA(X, t)=C,(x,
0)
T(x,
t)
=
T(x,
0)
TW(x,
t)
=
T,(x,
0)=
T(x,
0)
The boundary conditions used by Sdrensen (1976) are also
used in this study to reduce the computational effort.
Boundary conditions:
trO
and
x=O;
The parameters
u,,
C,
and
To
can be adjusted freely within
the operating limits by the plant operator
or
the process control
system, and can be classified as manipulated
or
adjustable
variables according to the nomenclature used in control theory
(Stephanopoulos, 1984).
In a purely mathematical sense, boundary conditions are
necessary to single out a particular solution
to
the model equa-
tions. From a physical point of view, these conditions must
express the interaction between the system and its surroundings
(Novy et al., 1990).
The boundary conditions used at the reactor inlet and outlet
are approximations
of
the continuity equations at the bound-
aries (Fan and Ahn, 1963). The PCclet number
(Pe= v,.L/Do.J
is estimated to be 175, indicating that plug
flow
is approached
(Pe
>
loo), according to Westerterp et al. (1984~). The Wen
and Fan (1975) method is used to estimate the Ptclet number:
1
3.10' 1.35
+-
Bo-
Re'.' Re"'
and
(9)
For example, the difference between the well-known boundary
conditions of Danckwerts (1953) and the boundary conditions
used in this study will be negligible due to the expected high
Peclet number (Fan and Ahn, 1963) and the inaccuracy
of
the
experimental data, which will have a higher impact on the
values
of
the estimated model parameters than the boundary
conditions.
Since Eqs.
1
and 3 can be solved separately, once the solution
for
C,,
Tand T, is known it is not necessary to further consider
them.
The design and operating conditions
of
the tubular reactor
itself and the reaction system can be characterized with a certain
set of dimensionless groups as shown by Westerterp and Ptas-
insky (1984a), Westerterp et al. (1984b), Westerterp and
Ov-
ertoom (1989, and Westerink and Westerterp (1988).
According
to
this method, the operating and design parameters
are related uniquely with the selectivity and the reaction system
parameters under steady-state conditions.
Similar dimensionless groups are used in this study, and the
method is extended to describe the dynamic behavior of the
reactor. To do this, additional reference values are introduced
for the fluid velocity
(vr)
and the reactant
B
concentration
(CBr),
resulting in the following set of dimensionless manip-
ulated variables:
The values
of
the reference variables
v,
and
C,,
are chosen
as the actual values of
v,
and
C,
at the steady-state condition
of the reactor system at maximum plant capacity. The reference
temperature
T,
is chosen as the reactor inlet temperature To at
the same conditions.
Different symbols are used for the dimensionless manipu-
lated variables
yo,
$B
and
4"
in comparison with the symbols
used for the state variables
I'B
and
0
to
accentuate that the
manipulated variables are the functions that can be adjusted
freely in time. The functions of the manipulated variables
ve,
$B
and
4"
during the observed reactor startup are shown in
Figures
4,
6 and
7,
respectively.
Equations
2, 4 and
5
are made dimensionless by the intro-
duction of the dimensionless quantities defined in the Notation
Section. The transformed equations are:
Mass balance
of
component
B:
Energy balance for the fluid:
Energy balance for
the
reactor vessel:
-
Da,.
U'
.
wh.
(0
-
e,)
30,
au
_-
(13)
1875
AIChE
Journal
December 1992
Vol.
38,
No.
12