&Chemical
Engineering
Computers and Chemical Engineering 24 (2000) 225-229
www.elsevier.com/locate/compchemeng
Abstract
This paper develops a systematic procedure for constructing an attainable region (AR). The approach uses two dimensional
ARs constructed in orthogonal subspaces to construct higher dimensional ARs. Our technique relies on previous algorithms that
provide a practical assurance of the completeness of ARs in two dimensions, using only PFR and CSTR reactors and mixing.
Here we build on a modification of this property by constructing 2D projections and their intersections that provide upper and
lower bounds of the AR. These bounds are then improved by applying AR constructions sequentially to candidate regions in
orthogonal subspaces. The approach is demonstrated on a well-known AR problem in three dimensions. 0 2000 Elsevier Science
Ltd. All rights reserved.
0098-1354/00/F% - see front matter 0 2000 Elsevier Science Ltd. All rights reserved.
PII: SOO98-1354(00)00518-4
226 W.C. Rooney et al. /Computers and Chemical Engineering 24 (2000) 225-229
tial sidestream reactors DSRs (a PFR with additional component B can be produced from (l), only a 2D AR
feed distributed along its length) and PFRs that are analysis is needed. Fig. 2 shows the 2D AR in AB
projected down into the 2D subspaces. The approach is space, which is a subspace of the full 3D AR in ABD
tested on a common kinetic scheme, frequently used in space. Note the boundary of this 2D AR is composed
reactor network synthesis problems. of a CSTR with bypass, followed by a PFR
with bypass. In addition, this 2D projection is a func-
tion of only the state variables A and B since the rate
2. Motivating example functions
1
To motivate this work, we start by presenting a
- klfA + k,,B - k,A2
candidate AR for the reversible Van de Vusse kinetics
byA - B&r + k,)
(1).
kLfB2C
A, (1)
kzl kv are completely described by the state variables. Now
D assume that information is needed on the 2D AR in BC
The 3D AR in Fig. 1 was constructed using the guideli- composition space. The rate vectors in this space are
nes in McGregor’s (1998) thesis until this region sa-
tisfied all the AR necessary conditions. The boundary
of this AR is composed of a CSTR operating point,
PFR trajectories, DSR trajectories, and mixing lines.
If a designer is only interested in how much of
Besides the variables of interest, these rate expressions
also involve component A. Here, the value for A is free
and can be chosen to extend the BC AR the furthest.
Thus, it functions as a control variable in BC space.
Moreover, the choice of A must be attainable. The
addition of the control variable A in the BC space
makes the construction of the 2D BC AR more difficult
than the AR in AB space.
We transform the nD AR problem into a series of Here, CE%I”is the vector of compositions, l3 is a non-
2D problems by working in a series of 2D AR sub- negative scalar that controls how much mixing takes
spaces. This is accomplished by working in each one of place relative to reaction, c* is an attainable mixing
point, and r is an independent variable of integration.
the *2 possible combinations of state variables. We Eq. (3) is used to maintain a constant control
0 composition.
define a candidate 2D AR in the subspace of larger Two instances of CSTRs appear in the subspaces. If
dimensional ARs (particularly 3D in this paper) as an points are found in the subspaces where ac/& = 0 (in
orthogonal subspace. In composition space, this re- Eq. (3)) for all components c, a CSTR with residence
quires the remaining n - 2 compositions be (advanta- time /3 can be used to achieve that point. For example,
geously) available for mixing to remain in the subspace this point could be obtained by solving the CSTR
of interest. For example, one might work in the BD design equation from the feed and looking where its
subspace, OBDIA, which is orthogonal to component A locus intersects each subspace.
(component C is found by mass balance and stoi- However, another instance of CSTR points in the
chiometry). In BD space, component A must be held subspaces occurs as a result of projecting points from
fixed to any desired value by mixing with another nD into the 2D subspaces. Let ci c c and c, c c be the
attainable point outside of the subspace considered. To control variables and the subspace variables, respec-
create an entire family of 2D ARs in BD subspace tively. At the fixed CSTR point c, in a subspace,
requires a variety of values for component A. Thus, A a~,/& = 0, otherwise the current point c, could move
is the control variable when looking at 2D ARs in the outside of the subspace via some combination of reac-
228 W.C. Rooney et al. /Computers and Chemical Engineering 24 (2000) 225-229
tion and mixing. No such requirement is imposed on 3.4. Additional intersections and extensions in 20
the control variables cj since these components of the subspaces
reaction and mixing vectors are not restricted to lie in
the plane of the subspace. In addition, the current point The data generated from the extensions of all the
c, must result in feasible, non-negative values of p. subspaces is recombined into the nD set Sk. Now, we
Feasibility can be checked from: consider new intersection points in the subspaces that
are generated from the data in Sk. For instance, the
family of subspaces OBDIA contains information about
all components in the system. Points obtained from this
subspace can be projected into the other orthogonal
which follows from (3).
subspaces. These new intersections will either lie within
Points satisfying the above conditions are produced
the current candidate subspace or they will lie outside
from iso-compositional CSTRs (iCSTRs), which we of it. If the former occurs, no extensions can be consid-
define as a 2D CSTR that operates at a point where
ered from that intersection since that point is already
&Jar = 0 for only the variables in the subspace. iCSTR attainable (i.e. any interior point can be obtained by the
points can be achieved from either a DSR (/I > 0) or appropriate mixing of boundary attainable points).
mixing to a PFR (p = 0) in higher dimensions. However, intersection points lying outside of the cur-
The analysis of PFRs must also change in the 2D rent candidate AR may be candidates for extending the
subspaces. Like the CSTR, a PFR cannot be operated subspares by simple mixing and possibly by additional
in the traditional manner in the subspaces. To achieve reaction. The process repeats until no further intersec-
constant composition of the control variables, material tions can be created in any of the subspaces from the
must be mixed in along the length of a PFR. Thus, set Sk.
iso-compositional PFRS (iPFRs) are needed to do the
2D AR analysis in the subspaces, not actual PFRs. The
mixing policy in these iPFRs, /?, is chosen to maintain 4. Algorithm to construct nD ARs using 2D subspaces
the control variables at their fixed value in the sub-
space. Eq. (3) shows the iPFR design equation. It Here, we summarize the algorithm outlined above.
differs from the PFR design equation by the addition of 1. Initialization
the mixing term, /?(c* - c). It should be noted that p is 1.1. Initialize counter, k = 0
not the mixing policy calculated by Feinberg (2000) 1.2. Initialize set So with:
that ensures a DSR rides along the boundary of the _ PFR trajectories from feed.
AR. Thus, iPFRS in the subspaces act in a similar - CSTR loci from the feed.
fashion to the isothermal DSR found by Nicol, Hilde- _ Critical DSR trajectories (Feinberg, 2000) (if
brandt and Glasser (1997) where the control policy is they exist).
chosen to maintain constant temperature, whereas here - Mixing vectors along each trajectory and loci.
we will be fixing one or more composition values. 2. Subspace initialization
Mixing is also considered in the 2D subspaces. The 2.1. Determine initial range of each variable Ack
process of mixing ensures that each subspace will be using Eq. (2)
convex as any nonconvexities in the region can be filled 2.2. Initialize subspaces, 0,” j = 1,. . . , NP using inter-
in by mixing. Mixing in the subspaces follows the sections between the data in Sk at the values of
normal lever rule (4). Composition c in the subspace Ack
can be achieved by the linear combination of any two 2.3. Convexify points in 0: j = l,..., NP forming
attainable points in the same subspace, c1 and c2, initial candidate ARs in the subspaces
weighted by the positive scalar 1. 3. Subspace extension
3.1. Extend subspaces O;j= l,..., NP using:
c = /Ic, + (1 - A)c, _ iPFRs
_ iCSTRs
01111 (4)
3.2. Convexify extensions in subspaces using mixing
Now, the initial candidate ARs in the subspaces can be 4. Subsnace recombination
extended using the properties outlined in the preceding 4.1. *Increment iteration counter k = k + 1
paragraphs. Points from the ARs are sought where the 4.2. Add subspace extensions to Sk
fundamental processes of reaction and mixing using 4.3. Project points from Sk into the subspaces
iCSTRs and iPFRs point out from the current AR in 4.4. If any points from Sk provide feasible extensions
the subspace. These extensions can always be made to the subspaces, GOT0 (2)
convex since mixing is considered in all of the 4.5. If no points from Sk extend the subspaces,
subspaces. STOP.
W.C. Rooney et al. /Computers and Chemical Engineering 24 (2000) 225-229 229