Chemical Engineering Science 56 (2001) 173}191

The attainable region and process synthesis: reaction

systems with external cooling and heating
The e!ect of relative cost of reactor volume to heat exchange
area on the optimal process layout
Willie Nicol , Markus Hernier
, Diane Hildebrant *, David Glasser
School of Process and Materials Engineering, University of the Witwatersrand, Private Bag 3, WITS, 2050, South Africa

RWTH Aachen, Institut fuer Verfahrenstechnik, Turmstrasse 46, 52062 Aachen, Germany
Received 21 October 1997; accepted 26 June 2000

Abstract
The attainable region (AR) technique is used to "nd the optimum process designs for an exothermic reversible reaction system
where allowance is made for cooling and heating with a constant temperature utility. The optimum process layout (equipment), which
results directly from solving the problem using the AR method, incorporates both a direct and indirect method of cooling material
undergoing reaction. The results have shown that direct cooling (cold shot) is favoured at high temperature and indirect cooling (heat
exchange) at lower temperature. The values of system parameters were changed and this caused the relative shape and size of the
attainable regions to vary. The inclusion of the cost of heating introduces a new complexity in the process layouts of the AR. In
particular the direct cooling (cold shot) is split into two stages: The hot reaction mixture is "rst cooled by adding preheated feed of
varying temperature and this is followed by the addition of cold feed to the reaction mixture.  2001 Elsevier Science Ltd. All rights
reserved.
Keywords: Attainable region; Reaction systems; Optimization; Process synthesis; Process design

1. Introduction
The attainable region (AR) technique is used in this
paper to "nd the optimum process layouts for an
exothermic reversible reaction system where allowance is
made for cooling and heating with a constant temperature utilities, which we will refer to as external cooling
and heating. Systems with exothermic reversible reactions have always presented a reactor design challenge
because the optimum reactor requires both high temperatures at low conversions to favour high initial reaction
rates and low temperatures at high conversions to
achieve high equilibrium conversions. The previous emphasis of research has been to "nd optimum ways of
achieving the required falling temperature progression.
The optimum temperature pro"les for a range of di!erent

* Corresponding author. Tel.: #27-11-717-7527; fax: #27-11-7177527.

E-mail address: dihil@chemeng.chmt.wits.ac.za (D. Hildebrant).

exothermic kinetic schemes were determined by Denbigh

(1944). Once a theoretical optimum temperature progression is determined the focus shifts to more practical
methods for obtaining similar temperature pro"les.
However, when considering more practical situations,
the cost of cooling starts playing a role and the best
temperature pro"les to follow is not obvious.
In practice two cooling methods are commonly used to
achieve the decreasing temperature progression. These
are direct cooling (cold feed addition) or indirect cooling
(heat exchange). Much research has been done to optimise the cold feed addition to a cold shot converter
(Hellinckx & van Rompay, 1968; Kramers & Westerterp,
1963; Burghardt & Patzek, 1978). The direct method of
cooling has also been examined using the AR approach
(Glasser, Hildebrandt & Glasser, 1992) and an optimum
conversion-temperature path was determined. It was further shown how this path could be implemented in terms
of reactors. Other research has focused on heat exchange
methods of cooling the reaction mixture (Beckett, Evans,
Ryan & Akgerman, 1989; Ravella & de Lasa, 1987) and

0009-2509/01/$ - see front matter  2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 9 - 2 5 0 9 ( 0 0 ) 0 0 1 9 6 - 2

174

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

the practical results of this work can be seen in methanol

and ammonia synthesis reactors. The cooling strategy
often chosen involves a set of adiabatic reacting stages
with inter stage coolers. Aris (1960) used dynamic programming to optimise such a system for a "xed number
of units.
Practical converters for exothermic reversible reactions tend to use either the direct or the indirect method
of cooling and a combination of the two cooling methods
is rarely seen. The work in this paper looks at such
systems and how optimum combinations of the two
cooling methods can be found. The AR technique has
been successfully used for synthesising and optimising
di!erent reactor networks (Glasser, Hildebrandt
& Crowe, 1987, 1990; Godorr, Glasser & Hildebrandt,
1994). Most previous AR work has concentrated on
systems where only the fundamental processes of reaction and mixing were considered. The result of this work
was that the optimum reactor network in terms of di!erent reaction processing units (reactors) and their interconnections (mixing strategies) were determined for
a number of di!erent systems. The AR technique has also
been used for the optimisation of an exothermic reversible reaction system. In their work (Glasser et al., 1992)
only the fundamental processes of mixing and reaction
were considered. In this paper we extend this work by
introducing indirect cooling and heating as extra fundamental processes.

2. Attainable region background

In recent years a geometric technique known as the
attainable region (AR) technique for synthesising and
optimising process speci"cations (which includes both
the process layout and the operating conditions) has been
developed. The initial work focused on reaction systems
(Glasser et al., 1987, 1990) and was based on the idea of
Horn (1961). Horn introduced the concept of the AR and
de"ned it as the set of all possible outputs from all
physically realisable reactors. Glasser and Hildebrandt
(1987, 1990) approached the idea of the AR from a geometric perspective. They considered a reactor as a system
where the only processes occurring are reaction and
mixing. They introduced geometrical interpretations of
these processes and derived a set of necessary conditions
for the boundary of the AR. They have also shown that
once the AR is found the optimisation of the problem is
straightforward, provided that the objective function is
an algebraic function of the system variables only.
Once the AR is known, a path between the feed point
and a point in the AR can be found. This path could be
fairly complex combinations of reaction, mixing and
other fundamental processes. This combination of di!erent processes could in turn be interpreted as a process
layout (equipment) with speci"ed operating conditions.

Hence by "nding the AR, the optimum process speci"cations to achieve points in the AR can also be found.

3. Attainable region theory

The AR is de"ned as the set of values of all output
variables which can be achieved by any possible steadystate processes using a given feed(s). The output variables, also called the state variables, de"ne the space in
which the achievable points need to be found. A point or
co-ordinate in the space is represented by a characteristic
vector c.
We de"ne the characteristic vector c to be an n-dimensional vector, i.e. c"(c , c ,2, c ) which comprises:
 
L
(i) variables needed to de"ne the fundamental processes
such as concentration, temperatures and residence time,
(ii) variables contained in the objective function, including
cost variables such as heat and mass transfer areas.
Thus if we take some material in the system, we could
assign a vector c to the state of the material, for example
the concentration, temperature, average residence time.
This vector would change as the material passes through
the system. Geometrically, one could represent the vector
c as a point in an n-dimensional Cartesian space where
each axis of the system represents one of the components
of c.
It is convenient to choose the state variables in such
a manner that they all obey linear mixing laws. This is
done so that the AR method can more easily handle
mixing. It also results in the AR being convex which is
a powerful mathematical result.
For any system under consideration we need to specify
the fundamental processes of that system. For a chemical
processing system typical processes will be reaction, separation and heat exchange. We can now de"ne the processes in a space where the state variables are the axes of
the space. Firstly, one can de"ne the processes as vector
"elds. A vector "eld is such that for each point c there is
a well-de"ned single vector or in some cases a set of
vectors describing a certain process. As the processes are
functions of only the state variables the vector "elds are
also just a function of these variables.
Glasser et al. determined a set of necessary conditions
for the AR for reaction and mixing. These conditions can
be extended to other processes and used when constructing the AR. These necessary conditions are as follows:
(i) The AR includes the feed point(s).
(ii) The boundary of the AR (*AR) is convex.
(iii) All process vectors on *AR must be tangent, point
into the region or be zero.
(iv) No process vector in the constrained (mass balance,
etc.) complement of the AR when extrapolated backwards can intersect the AR.

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

These conditions are only necessary conditions of *AR.

We have no su$ciency conditions for the AR and therefore an AR which satis"es all the necessary conditions
can only claim to be an candidate AR. We use the term
AR! to describe a candidate AR and the term *AR! for
the candidate AR boundary.
If we have an objective function that we wish to optimise, that is an algebraic function of some or all of the
variables that make up the space (that is of the characteristic vector) then once we have found *AR!, which de"nes AR!, the optimisation is straightforward. Either its
optimal value lies on the boundary or it is an interior
point. In the latter case we can because of convexity
achieve these as a linear combination of boundary points.
An attainable point on *AR! will always have a path
from the feed point associated with it. We can then
interpret these paths in terms of process layouts or equipment which could be built (or at least approximated) in
practice. The optimum layout will thus result from the
optimisation.
We may, however, not at this stage have a speci"c
objective function in mind. An alternative way of using
the AR is to de"ne a general optimal process layout. This
is a layout su$ciently general to enable us to achieve
every point on either *AR! or in the interior of the AR!.
In order to do this we will need to be able to de"ne paths
by which we can get from the feed to any point on the
boundary surface and hence to the interior by a convex
combination of such paths.
Feinberg and Hildebrandt (1997) have proven some of
the properties of *AR for systems with the processes of
mixing and reaction. The results for AR examples with
di!erent fundamental processes seem to have similar
properties and we therefore postulate the following properties of *AR to be of a general nature:
(i) *AR is made up of surfaces comprising single-process trajectories or surfaces comprising mixing lines.
(ii) The process surfaces are joined by intersectors which
can either be smooth or not. Smoothness will imply
that there is an unique tangent plane at every point
on the intersector, while non-smoothness will imply
two or more tangent planes.
(iii) The principle smooth intersectors are termed
generating intersectors and switching intersectors.
(iv) The generating intersectors act as main process pathways to points on *AR.
(v) If two surfaces do not join smoothly then the intersector is itself a union of process trajectories.
(vi) A proposed *AR can always be tested with the necessary conditions.
Some background has to be given on the properties of
intersectors. We use a speci"c notation to describe intersectors. Firstly, we distinguish between smooth intersectors, described by smooth brackets `( )a and non-smooth
intersectors described by angled brackets `1 2a. Second-

175

ly, we want to describe the direction of the process

surfaces intersecting. A process surface will have the
direction indicated by the process vectors tangent to the
surface. In the neighbourhood of the intersector the two
process surfaces that are intersecting always have associated directions. We distinguish between process surfaces pointing towards the intersector and process surfaces
pointing away from the intersector. If the process surface
vectors point towards the intersector we use the superscript `!a which indicates that the process surfaces
moves into the interior of the AR when continued past
the intersector. We can also say that the process surfaces
terminate at the intersector. On the other hand, we use
the superscript `#a if the process surface vectors point
away from the intersector. Such a process surface originates at the intersector and moves along *AR. It can also
be said that the surface is locally generated from the
intersector. Lastly the superscript `a is used to describe
that the intersector is itself a trajectory. The absence of
the dot implies that the intersector is the union of locus
points.
Let's take two arbitrary process vectors p and

p describing two adjoining process surfaces on *AR. If

both process surfaces point away from the intersector the
intersector will be noted: (p>, p>) which is the general
 
notation for a generating intersector. Note that the intersection is smooth, the process surfaces point away from
one another and that the intersector can be followed like
a trajectory. In the case where one process surface terminates at the intersector and another originates (or
generates) from the intersector we use the notation:
(p\, p>) which is the general notation for a switching
 
intersector. Note that the intersection is smooth and that
the intersector cannot be followed like a trajectory (it
consists of a set of locus points).
Geometrically a generating intersector represents the
optimum simultaneous operation of two processes. It can
be interpreted as a processing unit where two processes
are taking place simultaneously. The generating intersector (p>, p>) is "rstly described by the di!erential equa 
tion:
dc
"p (c)#ap (c) (a'0)


dt

(1a)

and the necessary condition for this curve to represent

the optimum simultaneous operation of the two processes is given by (Godorr et al., 1994)
u(p (c), p (c))"(p (c);p (c))




*p (c)
*p (c)
 p (c)!  p (c) "0,


*c
*c 

(1b)

where all the terms are calculated at point c. Curves that

simultaneously satisfy Eqs. (1a) and (1b) may be possible
intersectors on *AR!. However, it may also be that these
curves do not lie on *AR!. Curves that satisfy Eqs. (1a)

176

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

and (1b) simultaneously will be referred to as possible

generating intersectors denoted by (p>, p>).. The sec 
tions (or entire curves) that are found to lie on *AR! will
be called candidate generating intersectors denoted by
(p>,p>)!.
 
The optimum control variable a for a (p>,p>). can be
 
determined from
!
u(p (c), p (c))p (c)



a"

u(p (c), p (c))p (c)




;(!
u(p (c), p (c))p (c)O0),
(1c)



where all the terms are again calculated at point c. The
switching intersector of (p\, p>) gives the locus of points
 
where one process surface terminates (or moves into the
AR) and where another process surface originate to de"ne *AR!. This intersector represents switch over points
between di!erent processing units. The position of the
switching intersector is determined by testing the p vec
tors on the p process surface. If p vector start pointing


out (in terms of *AR!) of the p process surface we know

the switch over point. This corresponds to the necessary
condition that no process vector can point out of *AR!.

4. The system
Omoleye, Adesina and Udegbunam (1989) have shown
that a wide range of industrial exothermic reactions such
as polymerisation, isomerisation, oxidation and esteri"cation reactions have characteristics that are similar to
those having "rst-order reversible kinetics. A "rst-order
exothermic reversible reaction is thus considered, where
the reactant A reacts reversibly to form B (A & B). The
rate of formation of B is de"ned by





r (x, )"A exp

V


!E
 ) (1!x)

!A exp


!E
 ) (x),

(2)

where x (fraction of A that has reacted) is conversion and

the reaction mixture temperature. A and A have


values of 5;10 and 5;10, respectively, while the
values of E and E are 4000 and 8000, respectively. We


will assume as in the example considered by Glasser et al.
(1992) (Feinberg & Hildebrandt, 1997) that the pressure,
density and heat capacity of the system remain constant
as the composition of the reaction mixture changes. Furthermore we assume su$ciently ideal mixing so that the
heat of mixing is negligible in comparison with the reaction enthalpy (*H). Consequently, we can write an energy balance such that the exit temperature of an
adiabatic reactor is a linear function of conversion and is
given by
" # ) x,



(3)

where is the adiabatic temperature increase and is


equal to *H/Cp. was taken as 200 for the example

considered. is the basis temperature from where the

adiabatic reaction is started (corresponding to x"0).
The feed temperature in all the examples is taken as
300 K.
External cooling can be done with a constant temperature utility where the temperature of the utility is ,
A
where was chosen as 300 K for all the examples.
A
Cooling is described by a heat transfer coe$cient (h )

which for simplicity is assumed to be constant. The rate
of cooling of a process stream of temperature per unit
heat exchange area is given by q:
q"h ( ! ).
(4)
 A
The external heating is done in a similar way with a constant temperature utility of 600 K ( ). The preheating
F
constraints di!er for the di!erent examples and will be
discussed separately.

5. Choosing a set of state variables and de5ning the

process vectors
The "rst step of the AR method is to identify the
fundamental processes occurring in the system under
consideration. The fundamental processes for the system
described above are reaction, mixing, cooling and heating. The reaction rate depends on the temperature and
conversion while the cooling and heating rate depends on
the stream temperature only. In view of the assumptions
about the system these two variables obey linear mixing
rules and we therefore choose them as state variables
which de"ne the space in which we work.
For optimisation purposes the cost of the system plays
an important role. For this system one needs an overall
cost function allowing for both cooling/heating and reaction costs. Residence time is normally used as a cost
indicator for a reactor as it gives an indication of the
size (volume) for a "xed #ow capacity. The heat
transfer area is also normally used as the size indicator
for heat exchange processes. To combine these two di!erent costs we assume a linear cost relationship between
the heat transfer area and the reactor residence time. We
then choose the value of the relative cost for a speci"c
system. By using a linear relationship between the area
for heat transfer and the volume for reaction it can be
shown that the resulting cost variable obeys a linear
mixing law.
We relate the cost of heat transfer area to the reactor
volume with the constant k so that the combined reacA
tion and cooling/heating cost is given by q"q #k u
P
A
where q is the reactor residence time and u the area
P
heat exchange per unit #ow through the heat exchanger. When allowing only the cooling process,
the rate of change of temperature with respect to cost is

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

177

given by
d h k
"  A ( ! )"K ( !).
(5)
A
A A
dq
C
N
Note that the variable q is now a relative cost variable.
We thus have three variables, conversion, temperature
and relative cost, that we will use to describe the system.
Thus the characteristic vector c is de"ned as c"[x, , q].
The AR is de"ned as the set of all the possible c that can
be achieved by a steady-state system that uses only the
processes of reaction, mixing, cooling and heating.
The fundamental processes have to be de"ned as vectors in the c space. These vectors represent the instantaneous change of state variables c at a de"ned point in
space when each of these processes occur on their own.
Glasser et al. have shown that for the processes of reaction and mixing in this space, the vectors are given by

 

(6)

(7)

r (x, )
x !x
V

r(c)" ) r (x, ) , v(c, c )" ! ,

 V

1
q !q


where r is the reaction process vector and v the mixing

process vector and c is the arbitrary but attainable point

(i.e. c 3AR) with which one mixes. From Eq. (5) the

cooling process vector k and heating process vector h can
be determined and are given by

 

k(c)" K ( !) , h(c)" K ( !) .
A A
F F
1
1

When only cooling or heating occurs it can be seen that

the conversion does not change as the process stream is
cooled or heated while the temperature and relative cost
variables change according to Eq. (5). The di!erence
between cooling and heating are the utility temperatures
one exchanges with ( and ) and the constant that
A
F
relates the cost of cooling/heating to reactor residence
time (K and K ). If K di!ers from K it would imply
A
F
A
F
that cooling and heating have di!erent associated costs.
The cooling vector points in the direction of temperature
decrease, the heating vector in the direction of temperature increase while the reaction vector points in the
direction of both conversion and temperature increase
(due to it being an exothermic reaction).

Fig. 1. Conversion, temperature relationships.

own cannot cross this line. Some isorate lines (W ) are


also shown and one can see that the rates at low conversion are higher as the temperature increases. The optimum temperature pro"le (W ), that is the maximum

rate at a "xed degree of reaction can easily be determined
from the kinetics and is plotted on the graph. An adiabat
is the straight line that gives the temperature conversion
relationship for any adiabatic reactor unit (see Eq. (3)).
The W line is the adiabatic maximum rate line which is

the locus of maximum rates along adiabats.
At this stage we can start looking at the positions of
the possible generating intersectors that might form part
of *AR. For reaction (r) and mixing (v), Eq. (1b) simpli"es
to
*r (x, )
*r (x, )
u(r, v)" V
) (x !x)# V
) ( ! ). (8a)


*x
*
Eq. (1a) becomes
dc
"r(c)#av(c) (a'0).
dt

(8b)

Hence a curve that satis"es both Eqs. (8a) and (8b)

simultaneously we will refer to as (r>, v>). (possible
generating intersector for reaction and mixing). Similarly
for the case of reaction (r) and cooling (k) Eq. (1b) simpli"es to
u(r, k)" ) [r (x, )]#K )
 V
A

*r (x, )
V
) ( !). (9a)
A
*

A curve that satis"es both Eq. (9a) and

6. Conversion, temperature relationships of the system
In Fig. 1 the typical conversion, temperature relationships for the kinetics of the system are given. The equilibrium line (W ) represents points where the reaction rate

is zero. These points give the equilibrium conversion as
a function of temperature. Note a reaction process on its

dc
"r(c)#bk(c) (b'0),
dt

(9b)

simultaneously we will refer to as (r>, k>)..

Eq. (1b) for mixing (v) and cooling (k), u(v, k) was also
calculated but no meaningful results were obtained from
the condition.

178

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

The two conditions u(r,v) and u(r,k) can be geometrically interpreted as surfaces. The surfaces are conditions only in terms of conversion and temperature and
can therefore be plotted in the x space as pro"les seen
in Fig. 1. It is interesting to note that u(r,k) lies slightly
above W . If cooling could have been done without any

cost (K "J) Eq. (8) would have reduced to
A
*r (x, )
V
"0,
(10)
*
which is the condition for the optimum temperature
pro"le and hence u(r, k) would correspond to W . The

optimum temperature pro"le can therefore be seen as
a possible generating intersector for reaction and cooling
where only reaction residence time adds to the cost (i.e.
cooling has no cost). The optimum combination that will
extend the AR the furthest for a process where the cost of
cooling adds to the total cost u(r, k) will di!er from the
W as can be seen in Fig. 1. It should be noted that the

optimum temperature proxle can still in principle be obtained with this system, it will however not lie on the
boundary of the AR but in the interior.

7. Example 1
7.1. Specixcations and conversion-temperature limits
We start with an example where only the processes of
reaction, mixing, preheating and cooling are considered.
To simplify the problem the cost of preheating is assumed
to be negligible. Before the construction of the AR can
begin one needs to de"ne the ranges of conversion and
temperature that we wish to consider. From Fig. 1 one
can see that the conversion is restricted by the equilibrium
line (W ). The equilibrium line is therefore one of the

boundaries in the x space. We furthermore restrict our
system by allowing preheating only to a temperature of
450 K. From this maximum allowable preheating temperature the temperature can still increase as a result of
adiabatic reaction. We therefore de"ne the adiabat with
"450 K up to equilibrium (AB) as another edge of the

AR in the x space. As no heating process is allowed
(except for the preheating), we can see that this adiabatic
edge cannot be crossed. Furthermore we de"ne the adiabat
with "300 up to equilibrium (CD) as another edge. The

conversion, temperature edges can be seen in Fig. 2. Note
that all x points inside the region ABCD are in principle achievable but we need to "nd out at what cost. For
the "rst example we choose the value of K as 150.
A
7.2. Constructing the AR!
Once the edges of the AR! are de"ned, the construction process can begin. The AR! is de"ned by its

Fig. 2. AR edges for Example 1.

boundary (*AR!). Thus the construction of the AR! only

involves "nding *AR!. For this problem, *AR! will be
the lowest relative cost values for the range of temperature-conversion values de"ned by region ABCD. The
boundary will be a surface in three dimensions consisting
out of di!erent process surfaces. The boundary can be
projected from above onto a plane in x space as outlined by the AR edges.
We will use the conditions of Eqs. (8) and (9) as the
basis of our AR construction. The u(r, v) and u(r, k)
conditions give us necessary conditions for optimum
combinations of processes. Note that for both the u(r, v)
and u(r, k) conditions, the amount of reaction cannot be
controlled while the amounts of mixing and cooling,
respectively, can be controlled. Both u(r, v) and u(r, k)
can be represented in the x space as pro"les. These
pro"les run from the hot adiabatic edge AB (Fig. 2) to the
cold adiabatic edge CD. One thus can get onto these
pro"les by initially operating an adiabatic reactor network on AB. This makes sense in terms of "nding the
lowest possible cost as the reaction rate for unconverted
material is much higher in the high-temperature regions.
The procedure for achieving an initial point on the u(r, v)
and u(r, k) pro"les and operating on them corresponds to
the idea of optimum temperature control for exothermic
reversible reactions, where one starts at a high temperature to utilise the high reaction rates of the unconverted
material but then one cools the reaction mixture down
to favour the higher equilibrium conversions at lower
temperatures. In our example preheating has no cost
associated with it and therefore the range of zero conversion temperatures (300}400 K) all have a zero cost value.
It is clear that one should use the free preheating up to
the maximum allowable temperature so that one can
operate the "rst reactor at a high rate. As one reaches the
u(r, v) and u(r, k) pro"les, clearly the equilibrium limitation starts having an e!ect and cooling needs to be
utilised.

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

Fig. 3. Optimum adiabatic reactor sequence.

What combinations of reaction and mixing do we use

on the right-hand edge (AB) of the AR!? As discussed we
use the free preheating (DA) and then react adiabatically
on the right hand AR! edge. Glasser et al. (1992) have
shown that the *AR! on the adiabat is de"ned by
a CSTR that operates at the maximum adiabatic rate
followed by a PFR up to equilibrium as shown in Fig. 3.
Starting with the cold feed, the CSTR locus (A) can be
compared with the PFR trajectory (B). It can be seen that
the CSTR locus lies below that of the PFR (smaller q for
a given conversion). A point P is found where the straight
line from the origin is tangent to the CSTR locus and is
clearly the lowest line we can achieve. This point can be
shown to be a point on the maximum adiabatic rate locus
(W ). Line OP represent a bypass from the feed around

the CSTR operating at this maximum adiabatic rate.
Reaction in a PFR from point P makes up the rest of
*AR! on the adiabat. The process layouts needed to
achieve all points on *AR for this two-dimensional
example are given in Table 1.

Table 1
Optimum adiabtic reactor sequence

179

The minimum q-path along the adiabat AB is now

known. We can therefore "nd the minimum q-values for
the intersections of the u(r, v) and u(r, k) pro"les with the
adiabat AB. The resultant [x, , q] co-ordinates represent starting points for (r>, v>). and (r>, k>).. Note that
we cannot claim before hand that (r>, v>). and
(r>, k>). will start at these points. We can claim though
that if they start from one of these points they will satisfy
Eqs. (7a) and (7b) or Eqs. (8a) and (8b), respectively.
We assume that (r>, v>). or (r>, k>). will start from
one of the above-mentioned intersections. A generating
intersector on *AR! will continue as long as parts of its
corresponding process surfaces lie on *AR!. When
switching intersectors start appearing on the boundary
surfaces these might be an indication of the end of one
generating intersector and the beginning of a new
generating intersector. Examples of this will be seen in
the work that follows.
Before the construction of *AR! is discussed further it
is useful to have a look at the candidate that have been
found for *AR! as shown in Fig. 4. Fig. 4 is a projection
of *AR! from above so that it is viewed in the conversion,
temperature plane. The di!erent process surfaces with
their corresponding process vectors can be seen. The
main process pathway is given by the arrows numbered
from
to . Further discussion of the construction of
*AR! will be explained in terms of this "gure.
As discussed previously one needs to "nd out whether
either (r>, v>). or (r>, k>). lie on *AR!. Testing the
process surfaces one "nds that the mixing and reaction
surfaces from (r>, v>). initially lie the lowest. We thus
start the cooling process with direct cooling and
in Fig. 4). If (r>, k>). is
(r>, v>)! lies on *AR! (curve
going to lie in *AR! we can see from Fig. 1 that the only
path to get on (r>, k>). will be reaction starting from the
(r>, v>)!. We test this by checking whether the cooling
process extends any of the plug #ow trajectories from the

180

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

extended cooling surfaces ((k\, v>)!). Switching intersector 8 is the switch over between cooling surface E to
mixing surface H and switching intersector 9 is the
switch over between cooling surface G to mixing
surface I.
All the surfaces have to be tested to show they satisfy
the AR necessary conditions. If all the necessary conditions are satis"ed, which was the case for the example
under consideration we can say that the region obtained
is a candidate for an attainable region. The candidate
boundary can now be used to determine the optimum
process layout for the system.
7.3. Process layouts on *AR!

Fig. 4. x-projection of Example 1.

reaction surface C leaving (r>, v>)!. Moving from right

to left on (r>, v>)! testing all the plug #ow trajectories
we found the plug #ow trajectory
where cooling extended the reaction surface at point T. Point T lies on
(r>, k>). and this implies that T is the starting point for
(r>, k>)! (curve ). The reaction surface generated by
(r>, k>)! thus gives rise to lower relative cost values
than a reaction surface from (r>, v>). in the F region.
Continuing to test the PFR trajectories from (r>, v>)!
(beyond point R) we "nd switching intersector 7,
(r\, k>)! where reaction from (r>, v>)! stops and cooling (region E) continues. (r\, k>)! ends in point S where
the mixing vector with the feed point, the reaction vector
and the cooling vector are coplanar. Point S represents
the last point of (r>, v>)! that lies on *AR!.
At this stage we have most of the important information to construct the process surfaces on *AR!. Before we
do that let us "rst summarise the main process pathway
that de"nes the boundary. We start by preheating (line
) only a portion of the feed as we need to keep some
feed for mixing in (r>, v>)! (see process layout). We then
react in a CSTR up to the maximum adiabatic rate (line
) and afterwards react in a PFR up to the intersection
of the u(r, v) pro"le (line ). The (r>, v>)! (curve ) is
now followed up to point R, where a PFR (line ) is used
to get onto (r>, k>)! (curve ). From this main process
pathway all the surfaces on *AR! can be determined.
The only other factors that determine the process surfaces on *AR! are the switching intersectors. All the
surfaces de"ned by the main process pathway must be
tested to see whether any of the other processes extends
them (i.e. process vectors point outwards). It was already
shown how the cooling surface E extended the reaction
surface C at switching intersector 7 ((r\, k>)!). Two other
switching intersectors were found on *AR!. Both of these
switching intersectors were where mixing with the origin

The generating intersectors have certain process

layouts associated with them. (r>, v>)! is interpreted as
a di!erential side stream reactor (DSR). The processing
unit correspond to a plug #ow reactor where material
is di!erentially mixed into the reactor along the length
of the reactor. The amount of di!erential side feeds is
controlled in an optimum manner described by Eqs. (1b)
and (1c). Similarly (r>, k>)! is interpreted as a
di!erential cooling reactor (DCR) where controllable
amounts of cooling is done along the length of a plug
#ow reactor.
The boundary can be interpreted in terms of a process
layout. Each single-process surface can be associated
with a speci"c processing unit. A projection of the threedimensional *AR onto the x-plane can be seen in Fig. 4.
This eases the viewing of the di!erent process surfaces.
The processing unit of each process surface will be discussed and can be seen in Table 2. In Fig. 4 it can be seen
that there is a main pathway, marked }
and drawn
in bold that in a sense de"nes *AR!. This process layout
corresponds to the feed points of many of the singleprocess trajectories that make up *AR!. The only initial
points that are not speci"ed by this pathway are
those that start at switching intersectors (e.g. surfaces E
and H).
We begin with the process surface A. The bypass
(mixing line
) from the CSTR operating at the
adiabatic maximum rate de"nes *AR! on the right-hand
edge. The points on this line are connected with mixing
lines to feed point O to obtain the lowest possible boundary. If a point on surface A must be obtained for some
optimisation objective it would be achieved with the
process layout seen in Table 2. Surface B is similar to
surface A, but this time the mixing lines are to the PFR
. When we get to the (r>, v>)!
that corresponds to
a DSR we know, according to the de"nition, that the two
process surfaces on either side of the generating intersector de"ne the boundary. Therefore the reaction surface
C and the mixing surface D de"ne the boundary
and their corresponding process layouts can be seen
in Table 2.

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

181

Table 2
Process layouts of *AR! for Example 1

On the reaction surface C there is a switching intersector 7 where we switch over to the cooling surface E and
from this surface there is a switching intersector 8 where
we switch over to mixing with the feed H. In Table 2 it
can be seen that the layouts of these regions branch o!
the main pathway layout (as do all the layouts). The PFR
is the process unit on the boundary that takes one
to the DCR . Once the operation
from the DSR
starts on the DCR
the corresponding reaction F and
cooling surface G de"ne *AR! and the process layouts of
these sections can be seen as primary branches of the
main pathway layout. On the cooling surface G there is
a switching intersector 9 where we switch over to mixing
with the feed. Intersector 9 joins smoothly with intersector 8 as surfaces E and G join smoothly.

8. AR's with maximum allowable temperatures and

di4erent cost relationships
The results of the above example are general in nature
and can be used to construct AR!'s for similar systems.
The previous example only allowed for a maximum preheating temperature. This might be due to the temperature of the heating utility which is only capable of
heating up the cold feed to a certain temperature because
the reactor is designed not to exceed a speci"c temperature limit so as not to damage the equipment or catalyst
or to limit undesired reactions. It might thus be useful to
change the constraint regarding maximum temperatures
so that instead an overall maximum temperature is speci"ed. Another important parameter is the reaction to heat

182

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

exchange cost relation. One would like to know how the

optimum layout changes as the value of K changes.
A
In the next two examples we examine how the structure of *AR! changes with di!erent system de"nitions. In these examples an overall maximum temperature of 550 K is speci"ed and the K cost factor is
A
varied. The cost of preheating is still assumed to be
negligible.

9. Example 2
The K value for this example is chosen as 400. This
A
implies that the cooling costs are a lot cheaper than that
of the previous example. The x-projection of *AR! of
this example can be seen in Fig. 5. It will be noted that the
edges di!er from the previous example in that the righthand edge is an isothermal line at 550 K which represents
the maximum reactor operating temperature. The process layouts for the di!erent process surfaces can be seen
in Table 3. The three-dimensional presentation of *AR!
can be seen in Fig. 6.
One notes that the main di!erence between this
example and the previous one is the path that one uses to
the initial points on the generating intersectors. In this
example we need to distinguish between two kinds of
situations in which simultaneous processes occur. The
"rst one is isothermal while the second corresponds to
the generating intersector. The control variable for the
isothermal simultaneous operation is determined by
balancing the two processes such that the temperature
remains constant. The two types of isothermal simultaneous operation are the iso-DSR and the iso-DCR. For
the iso-DSR the di!erential sidestream amounts are controlled so that the temperature increase of reaction is
cancelled by the addition of cold feed. The iso-DCR
works on the same principle so that the reaction heat is
cancelled with utility heat exchange.
The iso-DSR on the maximum temperature line was
found to lie on *AR! along
. Once (r>, v>)! was
reached the main pathway on *AR! was the same as that
of the previous example (Fig. 4). The only di!erence was
that the switching intersector 7 between surfaces C and
D (also switching intersector 7 in the previous example)
was closer to the right-hand edge. Therefore (r>, v>)!
was shorter and this implies a smaller DSR on the
boundary pathway. It thus becomes clear that cooling
the reaction mixture by means of cold shot makes up
a smaller fraction of the total process layout. A smaller
DSR is used when cooling is cheaper and therefore
switching intersector 7 will gradually move to the right
as K increases. The process layouts of these
A
examples can be seen in Table 3. The main di!erences
between the "rst and the second example are the occurrence of the iso-DSR and the sizes of the optimum DSR
and DCR.

Fig. 5. x-projection of Example 2.

10. Example 3
For the third example the same AR edges were kept as
in example 2, but the K value was chosen as 2000 which
A
implies very cheap cooling relative to reactor volume
costs. The x-projection of *AR can be seen in Fig. 7. In
this example no (r>, v>)! was on *AR! as switching
intersector 7 of the previous example has moved out of
the de"ned edges. We do still "nd that the iso-DSR at the
maximum temperature provides the best initial isother). A point P is reached
mal reactor operation (line
where the iso-DCR moves into *AR!. For bigger
K values point P will move downwards on the isotherm
A
and for an in"nite K point P will be on the unreacted
A
feed baseline (x"0). From point P the iso-DCR (line
) is operated up to the u(r, k) pro"le from where
(r>, k>)! (optimum DCR) generates the cooling surface
D and the reaction surface E. Switching intersector 5 is
the switch over from the cooling surface B of the isoDCR to the mixing fan hull C. The optimum process
layouts can be seen in Table 4. These layouts di!ered
signi"cantly from the previous examples and showed
how varying the relative cost of reactor volume and heat
exchange area can change the layouts of *AR!.

11. Including the cost of heating

In the previous examples the only heating process was
the process of preheating. The cost of preheating was
assumed to be negligible. In practice one could "nd that
external heating with a hot utility could be at least as or
even more expensive than cooling with an external utility
(Kramers & Westerterp, 1963). One could for example
consider steam as a heating medium and cooling water
for the cooling utility. Steam has a higher operational
cost and the cost of steam heat exchangers is usually
more than that of cooling water heat exchangers. It is

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

183

Table 3
Process layouts of *AR! for Example 2

therefore of importance to consider all the heating costs

of a system in the preliminary design.
In the last two examples the cost of heating is taken into
consideration. We choose the value of K as 80 and K as
F
A
150 for the next two examples. This implies that heating is
more expensive than cooling. In the fourth example only
the processes of preheating, mixing and reaction are considered. In the "fth example the processes of overall heating, cooling, mixing and reaction are considered.

12. Example 4: preheating, reaction and mixing

Glasser et al. (1992) studied a similar example where
preheating had no cost associated with it. In Glasser's

work a generating intersector of reaction and mixing

with the cold feed ((r>, v>)!) was found to de"ne the AR
boundary. The complexity of this problem suggested that
other mixing patterns might form part of *AR and therefore other possible mixing and reaction generating intersectors were considered.
There are two other generating intersectors de"ned by
reaction and mixing vectors that can exist on *AR: "rstly
a DSR where material at a state di!erent from that of the
cold feed is added along a reactor; and secondly
a generating intersector where the mixing point (cold
shot state) varies along some process trajectory. The
second option can be interpreted as a plug #ow reactor
running co-currently or counter currently parallel to
another process from which material is di!erentially

184

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

The other condition for the (r>, v>). is the tangency

H
condition of the AR (Hopley, Glasser & Hildebrandt,
1996) that states that the reaction vector, mixing vector
and the vector tangent to *AR at c (preheating vector),
K
termed h must all lie in a plane.
(v (c, c );r(c))h"0.
(13)
H
K
Considering all the constraints that a possible (r>, v>).
H
on *AR must satisfy, one realises that the process by
which the varying mixing points are generated are likely
to be via a plug #ow reactor or the preheater. An example
of a (r>, v>)! where the varying mixing points are generH
ated by a plug #ow reactor can be seen in the work of
Hopley et al. (1996). We postulated that if there was an
(r>, v>)! in *AR, that the mixing point c would move
K
H
along the preheater curve (line AB of Fig. 8). Hence we
assumed that v "v(c, c ), where c lies on the preheatH
K
K
ing curve.
Eq. (12) needs to be simpli"ed to a more useful form.
This is done in the following set of equations. We start by
determining dv along the preheating curve (AB on
H
Fig. 8)).

Fig. 6. *AR! of Example 2.

*x
K
*x
*
K
dv "
H
*x
*q
K
*x

*x
K
*
*
K
*
*q
K
*

*x
K
*q
1 0 0
*
K ! 0 1 0 .
*q
0 0 1
*q
K
*q

 

(14a)

If we choose a scalar s that parameterises (r>, v>). as

H
well as q so that it parameterises the preheating trajectK
ory we can show the following:
Fig. 7. x-projection of Example 3.

removed and fed to the plug #ow reactor. This processing

unit is called a DSR with variable mixing (DSRH). The
generating intersector de"ned by r(c) and a mixing vector
with a moving mixing point v (c(t), c (t)) will be denoted
H
K
by (r>, v>)..
H
The mixing vector is now not only a function of the
points along the generating intersector, but also a function of the point that one mixes to (c (t)) . The mixing
K
vector is given by

x (t)!x(t)
K
v (c(t),c (t))" (t)!(t) .
K
H
K
q (t)!q(t)
K

(11)

The "rst condition for the (r>, v>). is that all points
H
along the generating intersector lie on the surface de"ned
by
(v (c, c );r(c))[dr(v (c, c ))!dv (r(c))]"0.
H
K
H
K
H

(12)

*c
*c *q *s (*c /*q ) *q
K,
K" K K " K K
(14b)
*c
*q *s *c
(*c/*s) *s
K
(*c /*q ) and (*c/*s) are the tangents to the preheating
K K
trajectory and (r>, v>)., respectively. These vector exH
pressions can be written for each of the single-vector
elements. By dividing the two resulting tangents we can
show that Eq. (14b) for one of its elements becomes
(*c /*q )x *q
P *q
*x
K" K K
K" V K,
(14c)
*x
(*c/*s)x *s
D *s
V
P and D are the tangents to the preheating and
V
V
(r>, v>). trajectories in the x direction, respectively.
H
Similar expressions can be obtained for the and q directional tangents. dv now simpli"es to
H
P
P
P
V
V
V
D
D
D
V
2
O
*q P
P
P
2
2 !I,
dv " K 2
(14d)
H
*s D
D
D
V
2
O
P
P
P
O
O
O
D
D
D
V
2
O

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

185

Table 4
Process layouts of *AR! for Example 3

dv (r) now becomes

H
P
P
P
V )r # V ) )r # V
 V D
D V D
V
2
O
*q P
P
P
dv (r)" K 2 ) r # 2 ) ) r # 2 !r.
(14e)
H
V D
 V D
*s D
2
O
V
P
P
P
O )r # O ) )r # O
V

V
D
D
D
V
2
O
The vector *c /*q has the same direction as the preheatK K
ing vector h, thus *c /*q "dh. Therefore,
K K
P
V
)r
1
*q r
dv (r)" K V #  V #
P !r
H
2
*s D
D
D
2
O P
V
O
*q r
)r
1
" K V #  V#
d h!r.
(14f )
*s D
D
D
V
2
O
Eq. (12) can be written as

The condition for (r>,v>). now becomes

H

(14g)
(v ;r)(dr(v ))!(v ;r)(dv (r))"0.
H
H
H
H
The second term of this equation falls away if Eq. (14f ) is
substituted in the above equation

For preheating the trajectory can be integrated to

obtain







(v ;r)(dv (r))
H
H
*q r
)r
1
" K V #  V#
v (v ;r)h
H
*s D
D
D
V
O
O
!(v ;r)r"0.
H





(14h)

(v (c, c );r(c))(dr(v (c, c )))"uH(r, v )"0.

H
K
H
K
H

(15)

This equation looks similar to that for (r>, v>). where

we mix to a "xed mixing point, but it implies that v is
H
changing along the intersector in such a way as to satisfy
Eq. (13). The condition (Eq. (15)) must be satis"ed at
every mixing point c . The condition can be rewritten
K
as
*r (x, )
*r (x, )
V
) (x !x)# V
) ( !)"0,
K
K
*x
*

(16)

where x and are variables. Eq. (7) can be rewritten

K
K
as
(x !x)
(x !x)
( !)
K
q"q #  K
! K
!
. (17)
K K ( ! )
r
K ( ! )
F F
K
V
F F
K

1
!
 .
q "
) ln F
K K
!
N
F
K

(18)

Solving Eqs. (16)}(18) simultaneously allows us to determine the position of (r>, v>). in the three-dimensional
H
space. Eq. (12) can be used to determine as a function
K

186

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

Fig. 8. Conversion, temperature relationships.

of x and as x is always zero along the preheating

K
trajectory. The result can be substituted in Eq. (14) to
determine q as a function of x and . Lastly Eq. (13) can
K
be used to determine q as a function of x and . This can
be represented by the following equation:
X(x, , q)"0.

(19)

The (r>, v>). trajectory is a curve that lies on the X surH

face. The corresponding mixing point for a point on
(r>, v>). can be determined from Eqs. (12) and (13) and
H
hence the mixing vector only becomes a function of the
state variables. One needs to describe the control variable a that keeps the (r>, v>). integration in the deterH
mined X surface. The (r>, v>). can be integrated using
H
the following equation:
dc
"r(c)#av (c),
H
dt

(20)

where t is a positive scalar. The control variable a can be

determined from the gradient of Eq. (20):
!
X(c)r(c)
c"
provided
X(c)l (c)O0.
H

X(c)v (c)
H

(21)

12.1. *ARA and its process layouts for preheating, reaction

and mixing
In the work of Glasser et al. (1992), a (r>, v>)! with
mixing to the cold feed was found to de"ne the AR
boundary (*AR!). This was interpreted as a DSR.
A (r>, v>). with mixing to the feed point was tried in
H
this example but we found that the boundary of the
region generated by this intersector did not satisfy the
necessary conditions for the *AR!. We were unable to
"nd a surface that satis"es Eqs. (3), (11) and (13) simultaneously. This suggested that the mixing point did not
vary along an adiabatic plug #ow reactor trajectory. The

next step was to test whether the preheating (r>, v>).

H
forms part of *AR!.
The equations that have been derived can now be used
to construct *AR! for the system where only preheating,
reaction and mixing are considered. For examples 1}3 it
was possible to determine the conditions for candidate
generating intersectors as pro"les in the temperature and
conversion space. This could not be done with (r>, v>).
H
as the generating intersector control variable is also
a function of the relative cost q. We now start by determining the position of (r>, v>). in the three-dimensional
H
space. The X surface can easily be determined for the
speci"c system but one needs to know where to start the
(r>, v>). curve. For (r>, v>). to move from hotter to
H
H
colder temperatures (as is required for exothermic reversible reactions) the varying mixing point has to decrease in
temperature. This is done by moving backwards on the
preheating trajectory.
As a result of calculations it was found that the only
way to get on the X surface, was to operate a CSTR at the
maximum adiabatic rate from any point on the preheating trajectory. Any other reactor combinations could not
attain a point on the X surface. The question now becomes what CSTR would be the best or in other words
what starting point on the X surface will result in the
lowest possible intersector. The answer to this is the
CSTR operating from the maximum allowable preheating temperature. This result applies in general for
di!erent K values and di!erent maximum allowable
N
preheating temperatures. In some of the cases considered
it was found that CSTRs operating from temperatures
lower than the maximum allowable preheating temperature could be the starting points for di!erent (r>, v>).'s
H
that joined smoothly with the optimum (r>, v>). from
H

the maximum preheating (r>, v>) .. These (r>, v>).'s

H
H
though were never better than the maximum preheating

(r>, v>) ..
H
Projecting the (r>, v>). trajectory (RQ) in the xH
plane one sees u (v , k) in Fig. 8. One notices that it
H H
crosses u (v, k) (u (v, k) is the condition for reaction and


mixing to the feed point A). The crossing point of
u (v , k) and u (v, k) is the point where the correspondH H

ing mixing temperature on the preheating trajectory is
300 K, the feed point. From that point one can no longer
operate (r>, v>). as no colder temperature than that of
H
the feed is available. Such a point represents the end of

(r>, v>) . and will always lie on the cold feed u (v, k)
H

pro"le.
The maximum adiabatic rate line (W ) and the equi 
librium edge (W ) are also given in Fig. 8. Di!erent

K values resulted in di!erent u (v , k) projections. The
N
H H
tendency was that smaller K values, implying higher
N
heating costs, resulted in u (v , k) projections that lay
H H
closer to the W
line. One notices that (r>, v>). starts

H
on W
(point Q) and that the u (v , k) pro"le intersects

H H
W
on the cold feed adiabatic edge. This intersection


W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

187

Fig. 10. x-projection of Example 4.

Fig. 9. *AR! of Example 4.

(P) presents a special stagnant point on the (r>, v>).

trajectory. Glasser et al. (1992) have shown that this point
on the (r>, v>). trajectory is the same as the maximum
adiabatic rate CSTR from the cold feed. One cannot
cross point P on the (r>, v>). as the mixing and reaction
vectors become collinear. According to the AR theory
a CSTR between points A and P results as the mixing
and reaction vectors become collinear. The (r>, v>). at
the mixing limit thus becomes a CSTR. The CSTR from
which (r>, v>). starts (point Q) represents a similar
H
stagnant point for a (r>, v>). with mixing to the point of
temperature 450 K. One is however able to move o!
point Q as the mixing point temperature decreases with
the (r>, v>). operation.
H
The construction of *AR! can now begin. Fig. 9 gives
the three-dimensional boundary of *AR!. The boundary
will however be discussed on the projection of *AR! onto
the x-plane, as seen in Fig. 10. The main process pathway that de"nes *AR! ( } ) with the single-process
surfaces (A}D) that branch from the pathway can be
seen in Fig. 10. Preheating is done up to the maximum
preheating temperature on line . Only a portion of the
cold feed is preheated and only a portion of the preheated
material is heated to the maximum preheating temperature as material is di!erentially subtracted from the
preheater and fed the DSRH. The DSRH as previously
described is the processing unit representing (r>, v>)..
H
The CSTR (line
) on the right-hand adiabatic edge
moves one to the DSRH start. The projection of the
(r>, v>)! trajectory can be seen as curve . A represH
entation in terms of process layout of the DSRH can be
seen in Table 5. One can see that heated material is
di!erentially subtracted along the preheater and mixed
into the DSRH. The quantities of the side streams can be
determined from the control variable a and hence the
portions of cold feed distribution can be calculated. The

(r>, v>). generates the mixing surface or hull A and the

H
reaction surface B up to point Q. The corresponding
mixing point for a point on (r>, v>). can be determined
H
from Eqs. (12) and (14).

From point Q a (r>, v>) ! with mixing to the cold feed

) makes up the boundary. The mixing surface
(curve
C is now a fan hull as the mixing point is "xed. The
reaction surface D is generated by (r>, v>)!. The process
layouts for the di!erent boundary regions can be seen in
Table 5. The (r>, v>)! can be interpreted as a "xed cold
feed DSR. The process layouts on the edges of *AR! are
also given. The layout on the right-hand edge is as in the
other examples, a CSTR with bypass followed by a plug
#ow reactor. The left-hand edge up to point R can either
be achieved with a CSTR and bypass or with a mixing
line to (r>, v>)! at point R.
13. Example 5: *AR! including overall heating
and cooling
The above example can now be extended to include the
processes of cooling and heating. Heating can now be
used anywhere on *AR! and not just from the unreacted
feed as in the previous example. The system's AR edges
are the 300 and 500 K adiabat. One is not restricted by
the equilibrium line as the process of heating can allow
one to cross the equilibrium line.
The x-projection of *AR! with the main process
pathway ( } ) and process surfaces (A}H) can be
seen in Fig. 11. The only possible generating intersectors
are (r>, v>)., (r>, v>). and (r>, k>).. There is no feasH
ible generating intersector for heating and reaction. We
start the boundary processes on *AR! in the same manner as in the previous example with preheating. We
preheat ( ), react in a CSTR( ), operate along
(r>, v>)! ( ) (interpreted as a DSRH), followed by
H
a (r>, v>)! with mixing to the cold feed ( ) (interpreted

188

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

Table 5
Process layouts of *AR! for Example 4

Fig. 11. x-projection of Example 5.

as a DSR). The surfaces A}D are exactly the same as in

the preheating example. From point R on (r>, v>)! we
react in a plug #ow reactor
up to point T where we

operate a (r>, k>)! ( ) (This is interpreted as a DCR.)

Switching intersector 7 represents the switch over from
reaction surface D to the external cooling surface
F((r\, k>)!). Switching intersector 8 represents the
switching from cooling surface F to mixing surface
E((k\, v>)!). The (r>, k>)! generates the reaction surface
G and part of the cooling surface F. The results were
found to be similar to those of the "rst example.
The process of heating never extended the boundary
except at switching intersector 9 on the reaction surfaces
((r\, h>)!). Switching intersector 9 lies very close to the
equilibrium line. Heating therefore only extends *AR! on
the reaction surfaces where reaction is close to the
equilibrium limit. This is an expected result as reaction
close to equilibrium is very expensive (see Fig. 9). The
heating surface H forms the rest of *AR! between
the de"ned edges. The boundary was shown to satisfy all
the necessary conditions. All the process layouts corresponding to the di!erent boundary regions can be seen in
Table 6.

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

189

Table 6
Process layouts of *AR! for Example 5

14. Conclusion
The AR technique has been successfully applied to an
exothermic reversible reaction system where allowance
has been made for both external cooling and external
heating with a utility of constant temperature. Using the
AR theory *AR! has been constructed for various systems. The e!ect of di!erent constraints on the process
and the e!ect of varying the cost parameters on *AR! has
been investigated and consequently the optimum process
layout changed.

The results have shown that direct cooling is favoured

initially at high temperature and indirect cooling at lower
temperature. The simultaneous reaction and cooling processes are described in a di!erential form where the
amounts of the cold shots or heat exchange are controlled. The former is clearly not a very practical operation
but can be easily approximated in practice by cold shot
converters with a "nite number of cold shots or adiabatic
reacting units with utility heat exchangers in-between.
The boundary process layouts of the "rst three examples * where the cost of preheating were assumed to be

190

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

negligible * di!ered signi"cantly from one another.

Firstly one could see that the relation between the cost of
heat exchanger area to the cost of reactor volume (K
A
value) had an in#uence on the size of the DSR and DCR
used. In the extreme case (example 3) no DSR was used at
all. Secondly one could see that a new type of processing
unit termed the isothermal DSR and DCR was introduced into the problem when a maximum allowable temperature was speci"ed.
In the fourth and "fth example the cost of preheating
and overall heating was considered. This consideration is
necessary as external heating costs normally exceeds external cooling costs. A new type of processing unit on
*AR!, called the DSRH was found. This processing unit
has similar properties to those of DSR where the processes of mixing and reaction occur simultaneously in an
optimum fashion. The DSRH however has a variable
mixing point which means that di!erential streams of
di!erent states are mixed into the #ow reactor. In the
example studied in this paper the states of the side
streams changed in the preheater. The boundary set of
optimum process layouts thus included a preheater from
which material was di!erentially subtracted and fed to
a #ow reactor called a DSRH. Such a processing unit
could in principle be approximated with a "nite number
of side streams.
It is clear from the results of examples four and "ve
that one only needs to preheat a fraction of the feed to the
maximum allowable temperature. Another fraction of the
feed is only preheated to an intermediate temperature
and then used to be mixed into the DSRH. The remaining
fraction is not preheated at all and mixed into the DSR.
The optimum combinations of cold shot mixing and
external cooling as a means of cooling down the reaction
mixture were determined in the "fth example. The results
of the example are fairly general in nature and similar
results are expected for di!erent cost relation constants.
For certain values of the cost some of the simultaneous
processes will not appear in the optimal layout. An example
of this can be seen in the "rst three examples where heating
had no cost associated with it. In this limiting case of free
preheating it can be shown that the DSRH disappears from
*AR!. Another example is that of cheap external cooling
so that one does not use cold shot cooling.
In general one could see how the complexity of the
process layouts de"ning *AR! change as more fundamental processes are introduced to the problem. The
layouts of the last example where allowance was made
for reaction, mixing, cooling and heating were very complicated compared to the "rst three examples where preheating was done without any cost. We can probably
expect that as we take the cost of more processes into
account that the optimal process layout will become
more and more complex.
The work has shown how the AR technique can be
applied to optimising process layouts. The advantage of

this technique is that the optimum process layout results

directly from the geometry of the AR as can be seen from
the examples. These results could be of use for more
complex optimisation problems where more process
variables are considered. This three-dimensional example
of process design and synthesis therefore only gives preliminary results but also provides valuable insight on
what the optimum process layout should look like and
how these layouts are in#uenced by cost parameters and
system constraints. The method can be considered as
a fundamental method of process design.

15. De5nition of AR terms is given in Appendix A

Appendix A
process layout * the inter linked serial or parallel
combination of processing units.
processing units * a single unit of equipment like
a CSTR, PFR or DSR.
process surface * the family of a single-process trajectories or the family of mixing lines that make up a surface
on the boundary of the AR.
process pathway * the main combination of processing units representing a series of connected lines in the
three-dimensional space from where all the process surfaces originate.
AR edges * the outlines of the xT-region for which the
three-dimensional AR is determined.
*AR * the boundary of the AR.
*AR! * the candidate boundary of the AR.
intersectors * intersections between process surfaces
on *AR.
(p>, p>) * smooth intersector between two arbitrary
 
process surfaces de"ned by p and p . Both process


surfaces generate (originate) from the intersector and the
intersector can be followed like a trajectory (therefore the
`a notation). The term generating intersector is used to
describe these type of intersectors.
(p>, p>). * possible (p>, p>) on *AR!.
 
 
(p>, p>)! * candidate (p>, p>) and thus part of the
 
 
candidate boundary (*AR!).
(p\, p>) * smooth intersector between two arbitrary
 
process surfaces de"ned by p and p . Process surface


p terminates or stop at the intersector while p gener

ates or originates from the intersector. The intersector is
also termed a switching intersector.
generating intersector * see (p>, p>).
 
switching intersector * see (p\, p>).
 
u(p p ) * necessary condition for p>, p>)..
 
 
DSR * di!erential sidestream reactor.
DSRH * di!erential sidestream reactor with variable
mixing.
DCR * di!erential cooling reactor.
iso-DSR * isothermal DSR.

W. Nicol et al. / Chemical Engineering Science 56 (2001) 173}191

iso-DCR * isothermal DCR.

( ) * denotes smooth intersector.
1 2 * denotes non-smooth intersector.

Notation
x
c

r
V
*H
C
N

fraction of initial reactant that has reacted

characteristic vector or state
temperature of reaction mixture, K
rate of formation of B
enthalpy of reaction, kJ/kmol
constant molar heat capacity of reaction mixture, kJ/kmol K

adiabatic temperature increase"*H/C


N

basis temperature for adiabatic reaction


A , A Arhenius coe$cients in rate equation
 
E , E Activation energies used in rate equation
 
q
reactor residence time
P
u
area heat exchange per unit #ow of the heat
exchanger
q
rate of cooling per unit heat exchange area
k
constant relation between reactor residence time
A
and heat exchange area

temperature of the cold utility

A

temperature of the hot utility

F
h
overall heat transfer coe$cient, W/m K

A
area of heat exchange, m
K
h k /C
A
 A N
K
similar to K , just for heating
N
A
W
equilibrium line in x space

W
isorate lines

W
optimum temperature pro"le

W
adiabatic maximum rate line

X
surface in which (r>, v>). must lie
H

temperature of the cold shot

AQ
q
overall relative cost
p ,p
arbitrary process vectors
 
a, b
control variables for generating intersectors
t
arbitrary scalar along a generating intersector
r
reaction vector
v
mixing vector
k
cooling vector
h
heating vector
A}H process surfaces
}
main process pathway
Subscripts
*

varied mixing

191

Mathematical notation
*p/*c
ab
a;b

Jacobian matrix of partial derivatives of vector

p such that element ij is equal to dp /dc
G H
dot or scalar products of vectors a and b
cross or vector product of vectors a and b
vector such that element i is equal to dp /dc
G G

References
Aris, R. (1960). The optimum design of adiabatic reactors with several
beds. Chemical Engineering Science, 12, 243}252.
Beckett, P. C., Evans, T. I., Ryan, W. E., & Akgerman, A. (1989).
Designing a multistage adiabatic reactor for minimum
catalyst volume. Chemical Engineering Communications, 76,
107}124.
Burghardt, A., & Patzek, T. (1978). Constrained optimisation of cold
shot converters. Chemical Engineering Journal, 16, 153}164.
Denbigh, K. (1944). Transactions of Faraday Society, 40, 352.
Feinberg, M., & Hildebrandt, D. (1997). Optimal reactor design from
a geometric viewpoint. Chemical Engineering Science, 52(10),
1637}1665.
Glasser, B., Hildebrandt, D., & Glasser, D. (1992). Optimal mixing for
exothermic reversible reactions. Industrial Engineering Chemistry
Research, 31, 1541}1549.
Glasser, D., Hildebrandt, D., & Crowe, C. (1987). A geometric approach
to steady #ow reactors: The attainable region and optimization in
concentration space. Industrial Engineering Chemistry Research,
26(9), 1803}1810.
Glasser, D., Hildebrandt, D., & Crowe, C. (1990). Geometry of the
attainable region generated by reaction and mixing: With and
without constraints. Industrial Engineering Chemistry Research,
29(1), 49}58.
Godorr, S., Glasser, D., & Hildebrandt, D. (1994). The attainable
region for systems with mixing and multiple-rate processes: Finding
optimal reaction structures. Chemical Engineering Journal, 54,
175}186.
Hellinckx, L. G., & van Rompay, P. V. (1968). Optimum adiabatic bed
reactor with cold shot cooling. Industrial Engineering Chemistry
Process Design and Development, 7, 595}596.
Hopley, F., Glasser, D., & Hildebrandt, D. (1996). Optimal reactor
structures for exothermic reversible reactions with complex kinetics.
Chemical Engineering Science, 51(10), 2399}2407.
Horn, F. J. M. (1961). Proceeding of 3rd European symposium on
chemical reaction engineering (pp. 1}10), New York: Pergamon.
Kramers, H., & Westerterp, K. R. (1963). Elements of Chemical Reaction
Design and Operation, Amsterdam, Netherlands: University Press
(Chapter 6).
Omoleye, J. A., Adesina, A. A., & Udegbunam, E. O. (1989). Optimum
design of nonisothermal reactors: Derivation of equations for the
rate-temperature-conversion and the optimum temperature progression for a general class of reversible reactions. Chemical Engineering Communications, 79, 95}107.
Ravella, A., & de Lasa, H. (1987). Cooling exothermic catalytic "xed
bed reactors: Co-current versus countercurrent operation in a methanol conversion reactor. Canadian Journal of Chemical Engineering,
65, 1021}1025.