Documente Academic
Documente Profesional
Documente Cultură
Krzysztof Gosiewski
Institute of Inorganic Chemistry, ul. Sowiriskiego 11, 44-101 Gliwice (Poland)
Abstract
The paper describes a comparison of some results computed using a mathematical dynamic model of a single reactor
bed with real temperature unsteady state transients, measured in a large industrial scale SO, oxidation reactor.
Problems of the dynamic modelling of complete oxidation plants, including a heat exchange section and a start-up
installation are discussed in the next section of the paper. As a basis for comparison some start-up dynamic
transients were analysed. The analysis shows that the model (two-phase model for heat transfer and homogeneous
model for mass transfer) is sufficient for qualitative analysis of the problem. The problems of obtaining reliable
parameters for simulation as well as parameter sensitivity of the model are also described. The plant is modelled
using the dynamic flowsheeting simulator DYNAM, whose algorithm is presented against a background of similar
simulators known from the literature.
The major part of the paper presents the results of the identification of the model based on an actual industrial
plant, together with the start-up analysis concerning the SO, oxidation plant. The analysis has been aimed at
minimizing both the emission of SO* during start-up and its duration. The results are given for the various modes
of the start-up operation.
Introduction
Jk
Inlet
The dynamic behaviour of SO2 oxidation reactors
can be readily described by the models developed in a
number of papers.
The two-phase heterogeneous model, well described
in the sixties and seventies, seems to be appropriate for
the solid bed reactors packed with vanadium catalyst
pellets. A good classification of such models was given
by Luebeck [l]. It would be beyond the scope of this
study to cite all the articles dealing with the models
similar to the one discussed here. We may quote only
some of them (Mann et al. [2, 31, Venkatchalam et al.
Outlet
[4], Eigenberger [5,6], Weng et al. [7], Van Doesburg
and De Jong [8] and Hansen and Jorgensen [9]). Fig. 1. SO, oxidation reactor with interpass cooling.
In spite of numerous publications, only a few of them
contain a comparison with experimental results; addi- experiments with laboratory tubes of diameter of sev-
tionally, it is very difficult to find a reliable comparison eral centimetres and height comparable with the indus-
of the computed results with the industrial plant un- trial scale, as was the case in the papers cited, must be
steady state transients rather than with the laboratory treated with caution if one wants to apply them to the
experiments. This paper deals with such a comparison industrial scale units. In particular, an analysis of the
for an SO, oxidation large-scale industrial reactor. influence of the reactor body on the dynamic transients
At present, such a reactor can have four to five should not be extended from the small to the large
passes (beds), as shown schematically in Fig. 1. Each scale. Figure 2 shows the bed which is the subject of the
bed is 0.5 to 1 m high with diameter from 5 to about numerical identification in this paper, compared with
12 m. It seems obvious that any conclusions based on the experimental reactor described by Mann et al. [3]. It
by random, non-uniform distribution of the flow rather n,C,[T,(L, t) - Tr’] =(1 -c)u[Tr’- T,(L, t)] (13)
than by the radial dispersion effects.
To simplify eqns. ( 1) -( 4) the following assumptions
aTk& 4
(1 - &)CI[Tk(L, t) - .yy = ny
were made: ax
For the mass balances: Elimination of TP’ from eqns. (13) and (14) gives
- homogeneous reactant concentration, that is, the the following form:
lack of mass transport resistance at the surface and
inside the catalyst particles (c = c*), ACP aTk(G t) = ngCg(1 - 4a
c
ax ( n
1 _[T,G 0 - T,(L, 01
E)u + c
O1k
ac
---0 (154
ax
y,-++rp,=O
ax The model eqns. (6), (7) and (8) can be presented in
The boundary and initial conditions for this sim- the following form, by introducing the usual dimension-
plified form will be as follows: less numbers:
jbr t =0: The heat balances:
Tk(x, 0) = T,(x, 0) = f(x) (9)
2 + St(SL)(T, - Tk) = 0 (gas phase) (16)
where f(x) is some arbitrary given initial temperature
distribution. This initial condition is of great impor- a”T,
--&$$+Bi $ (SL)(T,-Tk)+Da’=O
tance for the model adequacy. It will be discussed later ax2 0
in more detail.
(solid phase) (17)
for x = 0:
and molar reactant balance:
From the heat flux balance at the inlet cross-section
a(
surface for x = 0, a condition similar to that given in
ref. 1 for the two-phase model z =Da” (18)
n,C,[TF - T,(O, t)] = (1 - E)E[~; - T,(O, t)] (10) Results and discussion for a single reactor bed
From the practical point of view the start-up is an
aTk = (1 - &)CIITk(O,t) - TF]
1, - interesting case of an unsteady state, due to the exces-
(11)
ax sive pollution observed at that stage. This period, often
c(0, t) = c’“(t) (12) occurring in the existing full-scale industrial plants,
allows one to compare the computed results with the
for x = L:
real transients observed in the industrial reactor, thus
Usually the Danckwerts condition is taken for x = L. providing a good opportunity to identify the model
It should be emphasized, however, that this ‘classic’ experimentally.
condition concerns the case when for x > L, fluid flows Generally, there are two types of start-ups observed
to the space where Ae = 0. Since for the case discussed it in the industrial SO2 oxidation plants.
flows further through the ,ceramic pellets with Azp # 0, (1) The ‘hot start-up’ - that is, the start-up of the
this condition is not exactly fulfilled. Then the following reactor after some pause in the reactor operation, after
condition was taken as a first approach: which the reactor is only partially cooled down. This
115
means that there are some regions along the reactor bed modelling the dynamics of the bed cooling. Some qual-
with the temperature enabling the immediate re-start of itative study of such a cooling process was carried out.
the reactor by supplying the hot gas with an appropri- A strict mathematical model of the cooling of the
ate SO, content. reactor bed would be very complicated, due to the
(2) The ‘cold start-up’ - that is, the start-up of the complicated ways of the heat outflow of the bed, when
reactor from its initial state when it is completely it is being cooled down. Additionally, one should de-
cooled down, first by using hot air from some external velop an individual model for any geometry of the
preheater and then by supplying the reactant when the reactor. Moreover, it seems impossible to provide reli-
reactor (or at least a part of it) has already been able quantitative parameters for such a model.
preheated to the appropriate reaction temperature For the analysis of the dynamic behaviour of the bed
(above 400 “C for the SOP oxidation case). during the ‘hot start-up’, the profile was estimated by a
The case of the ‘hot start-up’ was chosen as a subject trial and error method beginning with an initial steady
of investigation and identification of a single bed, as a state profile. The temperature of the bed was measured
first approach to the dynamic modelling problem. at -two points, namely for X Y 0.2 and X = 1, where the
During the ‘hot start-ups’ some interesting ‘oscilla- thermocouples were installed. The initial temperature
tions’ are observed. The simulation was expected to profile should fit these measurements at these points.
explain the nature of this phenomenon. For the com- The remaining part of the profile should have been
plete oxidation plant, both ‘cold’ and ‘hot’ start-ups estimated in such a way that an agreement could be
were investigated. obtained between the transients calculated with those
measured (at these two points) during the start-up
Identification of the model parameters for a single period. However, one should be aware of the fact that
reactor bed the assumption of the initial profile is the weakest point
To identify the model one should have not only the of the analysis. The initial profile taken for further
adequate mathematical equations, but also reliable val- simulation is shown in Fig. 3 as curve 1.
ues of the parameters used in the numerical calcula- Curve 2, also shown in this figure (dots), presents the
tions. Only few of the parameter values can be initial profile intentionally deformed, to test the sensitiv-
estimated with confidence. Others, like u or A,, have to ity of the transients to the initial profile shape; this will
be evaluated based on the relations taken from the be discussed further. As may be seen, the simulated tran-
literature. One can obtain different values of the sients are sensitive to the profile assumed, but this does
parameters, depending on a particular reference taken not change the time scale of the observed oscillations.
into account. Thus, an analysis of the sensitivity of the
results with respect .to these ‘uncertain’ parameters Problem of the heat capacity of the bed
seemed to be of particular importance. The parameter C, appearing in eqn. (2) should not
For the SO, oxidation reactor case, there are two be regarded simply as a physical specific heat of the
problems which should be treated with special care. catalyst.
These are: the initial temperature profile and the heat When the catalyst is heated up, the following pro-
capacity of the bed. cesses are observed:
_ melting of the active compounds of the catalyst (see TABLE 1. The set of parameters assumed for the ‘hot start-up’
Tandy [ 151 or Boreskov [ 16]), tests
- desorption of the gas compounds already adsorbed
Parameter Symbol Value
on the porous surface of the catalyst and adsorption of
the new compounds, Catalyst eff. spec. heat 5.5 kJ kg-’ K-’
_ in some cases evaporation of the moisture from the {Physical spec. heat) {-l.O5kJkg-‘K-‘}
catalyst can also occur to some extent. Heat of reaction 8.89 x lo4 kJ kmol-’
The above physico-chemical processes can signifi- Heat transfer coeff. 151 W m-2K-’
Eff. heat conductivity 0.465 W m-’ K-’
cantly increase the effective heat capacity of the bed Inlet molar flux 47.65 kmol mm2 hm’
over the simple physical specific heat effect of the solid Inlet gas composition:
bulk of the catalyst. Moreover, the heat capacity de- SO, lOvol%
pends on the range of temperature and composition of 0, 11 vol %
the gas supplied into the system. N, 79 vol %
Bed length L 0.48 m
Specifically, if we substitute some equivalent effective Reactor diameter D 8.6 m
value Ckeff for C,, this value will cause some additional Spec. catalyst surface s 568 m* m-3
non-linearity of the model, very difficult to allow for a Porosity of the bed 0.5
priori. To simplify the problem some rough simplifica- Bulk density of catal. P. 620 kg mm3
tion of the problem was assumed taking a constant
value of CkeWfor a wide range of temperatures and gas
compositions. The value of C, is virtually the only
uncertain parameter of the model influencing directly Results of simulation
the time scale of the process. It can be estimated simply Some real industrial catalyst temperature transients
by ‘adjusting’ the time scale of the computed transients of the ‘hot start-up’ run measured in an industrial
to get an agreement with the real ones, measured in the reactor are compared with the computed results in Fig.
industrial reactor. 4. The solid lines ‘a’ in the figure present the results
The identification of the Ckeff values was made for computed by solving eqns. (6), (7) and (8), while the
various reactors with different vanadium catalysts. It dots show the measured transients, recorded by the
turned out that the ratio Ckeff/Ck (where C, denotes industrial strip-chart recorder. The transients were
physical specific heat of the catalyst) can be ascribed to shown for the two points of the bed, 2 = 0.2 and R = 1,
one of the three groups of transients: corresponding to the points in the reactor where the
CkeR/Ck g 1, for heating of the catalyst with hot dry air thermocouples were installed. One can notice that the
up to the temperature of 400 “C, transients are of similar character with regard to the
C,,/C, g 2, for the start-up of the reaction after the ‘oscillations’. This ‘oscillatory’ transients can be ex-
bed has been preheated with dry hot air (i.e. during the plained when the movement of the temperature profile
‘cold start-up’), along the bed is observed (see the profiles for various
C,,/C, s 3-5, for the start-up of the reaction during
the ‘hot start-up’.
The differences in the ratio CkaK/Ck described above
are obvious if we remember that ‘cold start-up’ takes
place after a long break in the reactor operation. Before
such a pause the reactor is carefully purged with clean
and dry hot air, which significantly reduces the influ-
ence of the surface processes.
from the literature. These are the gas-solid heat trans- Fig. 6. Computed temperature transients. Influence of 1.
fer coefficient a and effective heat conductivity of the
bed 1,. The values given in Table 1 were calculated
as the ‘mean’ values of the parameters.
The influence of c( was studied by assuming the
values of u’ = 2a and IX”= 0.5a in the model.
To examine the influence of 1, the following values
were assumed:
0.65
The temperature transients thus calculated are shown
s
._
in Figs. 5 and 6, while the influence of the two tests
E
on the computed conversion is shown in Fig. 7.
9
There is a clear dependence of the results on the 0.60
Sensitivity of the model to the initial temperature Model of the complete oxidation plant
proJile
The most questionable point of the simulation is the Mathematical simulation of the complete oxidation
assumption of the initial temperature profile along the plant needs special numerical tools to model the reactor
bed, at the beginning of the start-up. It seemed worth- passes together with an auxiliary installation and heat
while to test how sensitive the results are when the exchange system in order to include the mutual inter-
profile is deformed. Line 1 in Fig. 3 shows the profile connections of the process. This tool is described briefly
which gives the best agreement with the measured tran- in the following section.
sients. As may be seen, this initial profile is tangential
to the steady-state profile (line for t = 150 min in Numerical problems in the dynamic flowsheeting
Fig. 3), which is understandable from the practical simulators
viewpoint, since the most intense cooling occurs at both The need for dynamic flowsheeting simulators arose
ends of the bed. Line 2 shows the profile arbitrarily as a natural consequence, when static simulators be-
deformed, although agreeing with the measured temper- came available as commercial software products. In
atures at the points X z 0.2 and X = 1. The temperature the seventies numerous papers appeared concerning
and conversion transients for the two initial profiles the modelling of the dynamics of the complex chemical
(i.e. lines for profiles 1 and 2) are shown in Figs. 8 and plants. Although the review published by HlavaEek in
9. It may be seen that the computed transients are 1977 [17] quotes 130 articles and 5 books, only a few
clearly dependent on the initial temperature profile. of them contained information on the complete dy-
namic flowsheeting software systems. Probably, the
most general system dating back to this period is that
developed by Bobrow et al. [ 181 (DYNSYS); the other
papers dealt usually with numerical methods for
solving large numbers of ordinary and partial differ-
ential equations. While these methods are essential
in simulating the complex dynamic systems, they do
not provide any comprehensive treatment of the
problem.
It was only in the eighties that the rapid develop-
ment began of the complete software systems for simu-
lating complex chemical installations (e.g. DYNSYL,
PRODYC, DYSCO, REMUS and DYFLOW cited in
ref. 19, or the MUNICHEPS system described in ref.
20). The most recent review of Marquardt [21] presents
the state of the art and future challenges of dynamic
Fig. 8. Computed temperature transients. Influence of initial tem- process simulation in chemical engineering. It transpires
perature profile. from ref. 21 and other publications that, at present, the
dynamic modelling of large-scale systems is still being
C-l developed and rarely takes the form of fully commer-
cialized and generally applicable software, not requiring
a good computer background, similar to the systems of
the static flowsheeting simulation.
Mathematically, the problem of dynamic simulation
of a complex system consists of solving the set of n
equations with n unknowns; these equations may be as
.-s
follows:
F - partial differential equations (dynamic elements with
8 0.60 parameters distributed in space),
- ordinary differential equations (dynamic elements
with lumped parameters),
- algebraic equations.
If the set contains only algebraic equations or differ-
ential equations with spatial derivatives, the problem
Fig. 9. Computed conversion transients. Influence of initial tem- reduces to that of static modelling. Therefore the condi-
perature profile. tion for a problem to be dynamic rather than static is
119
the existence of the time derivatives in at least part of The literature survey reveals that it is customary to
the equations. Generally, the problem of the dynamic employ methods based on the modules of already exist-
simulation may thus be formulated as follows: ing software systems used in the dynamic simulation of
large installations.
aY According to ref. 17, these methods may be further
at =Fd_vxx~yx~y>
0
divided into the following groups:
0= Fsi_~Y Y) _ simultaneous modular
XX? x9 (20)
- sequential modular.
The vector eqn. (19) describes the dynamic part of
In the former group, the equations are lumped to-
the system, whereas eqn. (20) describes its static com-
gether to form the modules usually corresponding to
ponent. The partial derivatives of the state variable y
individual apparatus or unit operations, and then inte-
with respect to the coordinate x may appear in the
grated simultaneously with respect to time employing a
functions Fd and F,. Usually, the number of apparatus single integration procedure. In the latter group the
and the complexity of the dynamic model for a large individual modules are integrated one after another
chemical plant result in a large dimension of the vectors according to an assumed sequence, often using different
in eqns. (19) and (20). integration procedures; this method is therefore an imi-
Similarly as for the static problems, the methods of tation of that employed in a number of static simula-
solution may be divided into two groups [22]: tors. However, while the sequential modular method
- equation-based methods, offers certain advantages in the case of static simula-
_ modular-based methods. tions, it has serious drawbacks when dealing with dy-
The equation-based methods seem to offer the sim- namic problems (iterative complications), which may be
plest way of tackling the problem, which reduces to a avoided by using a different method of solution (this
coherent set of equations of the type ( 19) and (20) and method, discussed later, is applied in modelling the SOa
then, upon employing an appropriate numerical proce- oxidation plant). The possibility of employing different
dure, to a set of algebraic equations which are solved integration methods for different modules seems to be
using any of the sparse matrix techniques. This appar- the only advantage of the sequential methods.
ently simple method of solution is, however, not always Equations of the type (19) may be solved using
employed in practice. The complexity of the solution various numerical procedures which, with respect to the
procedures for the sets of many differential equations, time derivative (a/&), may be divided into explicit
and especially the difficulties with solving the so-called methods and implicit ones. The theory of the numerical
stiff equations (see e.g. [20, 22-261) make it necessary methods shows that those based on an implicit form of
to carry out judicious decomposition of the whole the difference notation are absolutely stable and lead to
problem into smaller sets, which are then tackled using higher accuracy of the numerical solutions at larger
different numerical approaches. For instance, the linear values of the integration step. The implicit methods are
equations may be solved separately from the non-linear particularly useful in solving the stiff differential equa-
ones, or the stiff part may be separated from the tions. They have, however, a serious defect when deal-
non-stiff part. ing with the structures of complex topology. As follows
from ref. 25, only extremely simple structures, usually
described by a linear differential equation, may be
Note: The notion of low stiffness of the differential equations is
solved by means of non-iterative procedures. In the
known in the theory of the numerical methods used for solving majority of cases, especially when the installation con-
these equations. tains recycles of flows or information recycles, the so
If for eqn. (19) Jacobian is defined called ‘stream tearing’ has to be carried out, leading to
problems with the convergence of the iteration loops.
and if eigenvalues of the Jacobian contain both a real and an
Even a cursory examination of the problem suggests
imaginary part that for the non-linear equations which are usually
dealt with when describing chemical appliances, the
1, = ( l/si) + joi
necessity of employing an iteration loop makes the
the stiffness of the equations may be expressed using the so-called nature of the methods implicit with respect to the time
stiffness coefficient S which is given by the ratio of the largest to
derivative. This is clearly seen in simple numerical
the smallest value of ‘I~:
procedures formulated for the two methods (to simplify
S=Z the discussion the derivatives with respect to the spatial
variable x have been omitted from eqn. ( 19)):
In practice, this is the ratio of the largest to the smallest value - explicit procedure
of the time constant of the system. Numerical problems appear
for high values of S. Y(t + 4) =u(O + h,Fd{Y(4) (21)
120
300
Ident$cation of the model based on the simulation of
the ‘cold’ start-up 250
The dynamic model has been identified by comparing
the numerical results with the actual start-up data s 200
H
recorded by the strip-chart recorders. The decision
$150
parameters for identification included all the parame-
I-0
ters difficult to determine a priori and having an essen- 100
tial effect on the dynamic transients, namely the activity
coefficients for the catalyst, heat-transfer coefficients for 50
Fig. 13. Predicted SO, concentration transients at the reactor Fig. 14. Estimated temperature profiles along the reactor passes
outlet during Phase 2 W~SUS values measured using chemical after the 4-h break.
analyses (dots).
reactor pass and several values measured during the 0.0 1.0 2.0 3.0 4.0- i?a Cho”$O
start-up operation. The agreement between the two Fig. 15. Predicted temperature transients during a ‘hot’ start-up
concentrations is quite satisfactory. It should be men- following the 4-h break.
tioned that the SO, emission in the plant studied is
particularly high. This results from a very low capacity mated for all the reactor passes after a break in the
of the start-up heater, which is unable to ensure the plant operation lasting for about 4 h; Fig. 15 shows the
heating of the installation to above 400 “C (673 K) temperature transients during the ‘hot’ start-up after the
during Phase 1. Consequently, the objective of the same break. Figure 16 presents a copy of the chart of a
simulation analysis has been not only to find an appro- conventional temperature recorder, revealing the actual
priaie start-up algorithm, but also to determine a cor- transients for the ‘hot’ start-up following a 4-h break.
rect heat efficiency of the start-up heater. Figure 17 gives the calculated SO, concentration tran-
sient after the fifth pass of the reactor related to the
Analysis of the ‘hot’ start-ups steady-state value. The transient recorded by an indus-
The model parameters obtained during the ‘cold’ trial SO2 analyser at the outlet from the plant has the
start-up identification described earlier were mostly em- shape analogous to that shown in Fig. 17. The values of
ployed in modelling the ‘hot’ start-ups. Only the de- concentration may, however, be much higher than those
grees of opening of bypasses in the heat exchangers and calculated if, due to poor thermal insulation of the re-
the effective specific heat of the catalyst had to be actor, the oxidation of SO2 does not occur at all for some
identified once again due to change in the operating time in the part of the gas flowing close to the cold wall.
conditions of the plant. As has been mentioned before,
in the modelling of the start-ups an additional problem SeIection qfsuitable start-up procedures for the ‘cold’
of estimating the initial temperature profile at the mo- and ‘hot’ start-ups
ment when the sulphur burner is switched on again In the heuristic ‘optimization’ of the start-ups slightly
arises. Figure 14 presents the temperature profiles esti- more degrees of freedom (decision variables) have been
124
Fig. 16. Copy of chart of an industrial temperature recorder Procedure proposed for the ‘cold’ start-up
showing the temperatures of the ‘hot’ start-ups after the 4-h
Upon testing several intuitive start-up procedures,
break.
the following strategy for heating of the contact reactor
has been proposed.
(a) Carry out the heating with hot air in about 50%
of the normal flow (i.e. 15 000 standard cubic metres
per hour (SCMH) in the case discussed), gradually
increasing the amount of fuel to reach maximum heater
capacity (without exceeding the admissible rate of tem-
perature increase). The whole air flow should be di-
rected straight onto pass no. 1.
(b) Upon reaching the temperature of 350 “C
(623 K) at the outlet of pass no. 1, decrease the air flow
by about one third (down to 10 000 SCMH).
(c) When the outlet temperature from pass no. 2
has reached 420 “C (693 K), introduce 70% of the air
directly onto pass no. 5.
(d) When the outlet temperature from pass no. 5
has reached 410 “C (683 K), disconnect the start-up
Fig. 17. Predicted SO, concentration transients at the reactor installation swiftly, connect the furnace and the boiler,
outlet (relative to the steady state value) following the 4-h break. and then light the sulphur burner at the flows of the
sulphur and air equal to 50% of the normal values.
assumed to enable a more flexible approach to the (e) Some 3.5-4 h after the sulphur burner was
algorithm. Thus, in fact, the start-up strategy consisted started, switch to normal parameters.
of a larger number of stages defined as follows. The simulation results are shown in Figs. 18 and 19.
Figure 18 presents the temperature transients at the
Piruse I Heating by means of a start-up heater. outlet from the passes, whereas Fig. 19 gives the outlet
SO* concentration relative to the steady-state value.
Stage 1.1. Heating of the reactor pass no. 1 using a The heating of the installation takes about 7.3 h,
high flow of hot air at a lower temperature. whereupon the sulphur is ignited at a 50% load and
then, after 11 h, the load is increased up to its normal
Stage 1.2. Heating of pass no. 1 using a lower flow level. In practice, the necessity to meet all the require-
of hot air at a higher temperature. ments concerning the rate of the temperature increase
125
200
Fig. 18. Temperature transients at the pass outlets predicted for Fig. 20. Temperature transients at the pass outlets predicted for
the algorithm proposed for the ‘cold’ start-ups. the algorithm proposed for the ‘hot’ start-ups following the 8-h
break.
C-l
in the lining and in the catalyst prolongs the heating 5.0
TABLE 2. The SO* emission (kg h-‘) expected for the start-up start-up of an industrial SO2 oxidation plant. The accu-
procedures selected using the computer simulation racy of the complex model of the plant seems to be
sufficient to yield strategies which could then serve as a
Parameter ‘Cold’ ‘Hot’ start-ups after the break lasting for:
start-up first approximation of the start-up experiments in an
2h 4h 8h 16h 24h 32h 40h actual installation. The algorithms proposed in the
present work could not be fully implemented in the
Average 13 32 16 17 22 25 33 27 plant studied because of the very low heat efficiency of
emission
the existing start-up section (it is to be modified in the
Maximum 27 45 36 45 59 63 173* 137*
emission near future). The verification has, however, been carried
out for the simplified algorithm more suitable for the
*NOTE: The admissible emission level is 100 kg h-’ existing (inadequate) heat efficiency, leading to an ex-
cellent agreement between the experiment and the nu-
merical predictions.
start-up algorithm, following an 8-h break. The concen- The results obtained using the computer modelling of
tration transient is compared with that occurring when the transient behaviour of the plant provide an impor-
a ‘jump’ start-up, that is, immediate switching to the tant guideline in dealing with actual start-ups. It follows
full yield, is applied. from the simulation performed that during the ‘cold’
Table 2 presents average and minimum emissions for start-ups and ‘hot’ start-ups after breaks shorter than
both the ‘cold’ and ‘hot’ start-up procedures after the 24 h, it should be possible to curb emission below the
various breaks in the plant operation. admissible levels. After breaks exceeding 24 h the emis-
It should be stressed that the emissions given in Table sions has to be controlled by means of the supplemen-
2 do not increase monotonically with the duration of tary heating of the reactor using an external heat source
the break, as for longer times, more complex start-up (start-up heater).
strategies are employed.
The emission values shown in Table 2 do not take
into account the increased emission resulting from the Acknowledgements
enhanced cooling of the catalyst near the reactor wall.
The author is grateful to all his collaborators for
their help in completing the study. Special thanks are
Conclusions due to Dr Witold Klaudel, who worked out the DY-
NAM flowsheeting simulator software, and to Mr
The model described here gives temperature and con- Marek Moszczynski, who tested tens of the start-up
version transients similar to those measured in large strategies.
industrial reactors, if the physical specific heat of the
catalyst is replaced by an effective value, incorporating
all heat-consuming effects occurring in the catalyst. Nomenclature
The agreement of the model is not very good, since
there is a visible influence of some large ‘side’ heat a, b inlet SO, and 0, concentra-
capacity effect. This effect is stronger for the upper tion respectively, kmol
layers, closer to the gas inlet. kmol-’
The model with the parameter values appropriate for Bi =aD/L,, Biot number
the large scale industrial reactor is sensitive to the a and C reactant (SO*) concentra-
1, values, but the influence of these parameters is not tion, kmol kmol-’
very significant when start-ups of the plant are consid- C* reactant concentration at
ered. Much larger is the influence of an (generally the catalyst surface, kmol
unknown) initial temperature profile on the final simu- kmol-’
lation results. c, gas specific heat, kJ
The model can be used for a qualitative analysis of kmol-’ K-’
the start-up of an SO, oxidation plant. G catalyst specific heat, kJ
Despite the fact that the model does not provide a very kg-’ K-’
precise description of the process simulated it seemed C heat radiation constant,
sufficient for finding the best start-up procedure for the kJ h-’ K-4
complete system at the stage of qualitative analysis. D reactor diameter, m
The dynamic flowsheeting simulation makes it possi- Dar = Hrp,L2/(& r,,), 1st Dam-
ble to analyse the control algorithms concerning the kiihler number
127
12 K. Gosiewski and M. Moszczynski, Dynamic simulation of heat transport from the bed to the body was employed.
the start-up of SO, oxidation unit in sulph. acid plants (in The model assumes that the only mechanism through
Polish), Przem. Chem., 70 (1991) 422-425.
which the heat may be exchanged between the catalyst
13 J. C. Mishra and M. T. Sebastian, Mathematical modelling of
unsteady state process in SO, converter, Proc. Inr. Co& bed and the body of the reactor is radiation from the
Unsteady State Processes Catal., Novosibirsk, USSR, 1990, upper and lower surfaces of the catalyst bed. Such an
659-664. assumption results from the qualitative analysis of
14 G. K. Boreskov, M. G. Slink0 and V. S. Beskov, Volume of other mechanisms of the heat exchange. These mecha-
catalyst, stability and parametric sensitivity in sulfuric dioxide
nisms have a much smaller share in the overall amount
oxidation equipment, Khim. Prom., 3 (1968) 13- 18.
15 G. H. Tandy, The role of alkali sulphates in vanadium cata- of heat exchanged than radiation.
lysts for sulphur dioxide oxidation, J. Appl. Chem., (1956) The incorporation of radiation into the model
68-14. modifies only the boundary conditions:
16 G. K. Boreskov, W. W. Illaronov, R. P. Ozierov and E. W.
Kildisheva, Chemical interaction in the vanadium pentoxide- for x = 0:
potassium sulphate and vanadium pentoxide-potassium pyro-
sulphate systems, Zh. Obshch. Khim., 34 (1964) 23-29. & 2 - 2;; [TZ(O, t) - r&l = (1 -E)CI[T&JO, t) - TF]
17 V. HlavaEek, Analysis of a complex plant steady-state and
transient behavior, Comput. Chem. Eng., 1 (1977) 75-100. (A-l)
18 S. Bobrow, J. W. Ponton and A. I. Johnson, Simulation of the
transient behavior of complex chemical plants using a modu- a7-
lar approach, Cue. J. Chum. Eng., 49 (1971) 391-397.
C[TZ(O, f) - Ti”] = IV,“& at (A-2)
19 G. K. Patterson and R. B. Rozsa, DYNSYL: A general-pur-
pose dynamic simulator for them. processes, Comput. Chem. for x = L:
Eng., 4 ( 1980) l-20.
20 J. C. Fagley and B. Camahan. The sequential clustered &. 2 = (1 _&)CI[Tk(‘L, t) - TF,] -s [Ti(L* t) - T4,J
method for dynamic chemical plant simulation, Comput.
Chem. Eng., 14 (1990) 161-177. (A-3)
21 W. Marquardt, Dynamic process simulation - recent progress
and future challenges, in W. H. Ray and Y. Arkun (eds.), 3T..vr_
Proc. Chem. Process Control, 1991, 17-22.
CP-;Z(l, t) - ~4,,1 = MwLCwT$- (A-4)
22 M. Hillestad and T. Hertzberg, Dynamic simulation of chemi-
cal eng. systems by the sequential modular approach, Comput. Solving eqns. (6) (7), (8) subject to the above condi-
Chem. Eng., 10 (1986) 377-388. tions (A-l) -(A-4) we get the temperature transients
23 H. S. Kheshgi, Transient reactor analysis by approximate given in Fig. 4 as lines ‘b’.
Jacobian factorization, Comput. Chem. Eng., 14 (1990) 149-
156.
24 Y. Ch. Liu and C. B. Brosilow, Simulation of large scale
Appendix B
dynamic systems, Comput. Chem. Eng., II (1987) 241-253.
2.5 J. W. Ponton, Dynamic process simulation using flowsheet
structure, Comput. Chem. Eng., 7 (1983) 13-17 D_vnamic model of a counter current gas gas heat
26 J. W. Ponton and V. Vasek, A two-level approach to chemical exchanger (with dominating thermal capacity of the
plant and process simulation, Comput. Chem. Eng., JO (1986) W&l)
277-286.
The model has been reduced to that of an equivalent
27 J. C. Heydweiller, R. F. Sinovec and L. T. Fan, Dynamic
simulation of them. processes described by distributed and shell-and-tube exchanger by assuming that the condi-
lumped parameter models, Comput. Chem. Eng., 1 (1977) tions in all tubes are uniform. It is assumed that the
125-131. accumulation of heat in the shell and the tube them-
28 R. S. H. Mah, S. Michaelson and R. W. H. Sargent, Dynamic selves is high enough to justify the omission of the
behaviour of multi-component multi-stage systems. Numerical
accumulation in two gaseous media. The axial conduc-
methods for solution, Chem. Eng. Sci., 17 (1962) 619-639.
29 J. C. Heydweiller and R. F. Sinovec, The influence of A-stabil- tion of heat along the exchanger is also neglected.
ity on the efficiency of stiff integrators, Comput. Chem. Eng., 6 Thus, the balance equations may be written as fol-
(1982) 185-188. lows:
30 S. Kuru and A. W. Westerberg, A Newton-Raphson based
strategy for exploiting latency in dynamic simulation, Comput. v, c, _ + (T, - T&d, ~1,= 0 (B-1)
Chem: Eng., 9-(1985) 175-182. 3X
subject to the following initial conditions (t = 0) m unit mass of shell or tube, kg m-’
temperature, K
TdO9 4 =“A(42 Ts(O, 4 =f,G) (B-5) f
gas flow rate, kmol h-’
and the boundary conditions u heat transfer coefficient, kJ m-* h-’ K-’
for x = 0 TI (& 0) = CPI(t) (B-6) cp temperature transients imposed at inlets, K
for x = L T,(t, L) = 92(t) (B-7)
Subscripts
where: 1 inside of tube
C specific heat, kJ mol-’ K-’ for gas; kJ kg-’ K-’ for 2 inside of shell
material of shell or tube 3 outside surface of shell
d diameter, m t tube
f initial temperature profile assumed, K S shell