Beruflich Dokumente
Kultur Dokumente
00
q Institution of Chemical Engineers
Trans IChemE, Vol 76, Part A, March 1998
INTRODUCTION
260
SHENOY et al.
single point-value for the globally optimum D Tm i n corresponding to the minimum TAC. It is important to examine
the TAC curve as a function of the process D Tm i n .
The TAC vs. D Tm i n curve is oftentimes found to be fairly
at in the neighbourhood of the minimum TAC. Consequently, it is better to de ne the optimum D Tm in in terms of a
range rather than a single value.
There are often constraints on capital investment. These
may be partially incorporated if targeting procedures de ne
an optimum D Tm i n range based on the TAC curve and
choose the upper or lower end of this range depending on
whether an investment ceiling exists or not.
The global optimum may demand the use of too many
utilities. In this case, it may be practically advantageous to
incur a small TAC penalty by eliminating utilities used in
little quantities and redistribute this load amongst the other
utilities.
Of the above four observations, the rst two have been also
emphasized by Kemp6 . This paper describes a targeting
methodology in the case of multiple utilities for generating
the TAC curve and de ning a meaningful D Tm i n target based
on the above techno-economic considerations.
MULTIPLE UTILITIES LOAD OPTIMIZATION AT
A FIXED PROCESS D Tm i n
If (n + 1) is the number of utilities (with temperature
levels speci ed), then the HEN problem typically features n
pinches [1 process pinch and (n - 1) utility pinches]. As
there is an approach temperature (say, D Ti with i = 1, ..., n)
for each pinch that is to be optimized, the optimization
problem is an n-dimensional one. However, it is possible7 , 8
to decompose this n-dimensional problem into a sequence of
n one-dimensional optimizations, each involving only two
utilities. The idea of decomposition is used in the
methodology below, but the sequence in which the onedimensional optimizations are performed is different from
that previously used5 ,7 .
The optimization sequence is demonstrated in Figure 1
Figure 1. Sequence of utilities load optimizations at a xed process D Tmin starting with maximum energy cost and minimum capital cost. (a) Initialization
with hottest hot utility (HP steam) and coldest cold utility (CW). (b) Optimization of the MP utility pinch by introduction of MP steam in place of HP steam.
(c) Optimization of the LP utility pinch by introduction of LP steam in place of MP steam.
261
Figure 2. Sequence of utilities load optimizations at a xed process D Tmin starting with minimum energy cost and maximum capital cost. (a) Initialization
with all utility pinches at zero approach temperature. (b) Optimization of the LP utility pinch by decreasing LP steam and correspondingly increasing MP
steam. (c) Optimization of the MP utility pinch by decreasing MP steam and correspondingly increasing HP steam.
Outlet temperature
(8 C)
H1
H2
C3
105
185
25
25
35
185
10
5
7.5
0.5
0.5
0.5
HP Steam
MP Steam
LP Steam
CW
210
160
130
5
209
159
129
6
Stream
=
=
=
=
5.0
5.0
5.0
2.6
Cost
(/kW/yr)
160
110
50
10
262
SHENOY et al.
Table 2. Optimum utility load targets for Example 1 for D Tmin = 308 C.
Number of
hot utilities
1
2
3
CW Load
(kW)
HP Load
(kW)
MP Load
(kW)
LP Load
(kW)
Area Target
(m2)
Energy Cost
(/yr)
775
775
775
425
203
203
NIL
222
53
NIL
NIL
169
137
158
178
75750
64650
54510
Capital Cost
(/yr)
32692
37710
42482
TAC
(/yr)
108442
102360
96992
Figure 3. Optimum load distribution (OLD) plots for multiple utilities as a function of process D Tmin. (a) Use of single hot utility (HP steam). (b) Use of two
hot utilities (HP and MP steam). (c) Use of three hot utilities (HP, MP and LP steam).
+ CCW QCW + Af CC
(1)
263
(2)
(3)
=C
T, MP
or
C HP + Af CC/ QHP
= C + A CC/ Q
MP
MP
(4)
=0
or
264
SHENOY et al.
The salient features of OLD plots and the CUP may be
summarized as follows.
1. When the process D Tm in is increased, the additional utility
is always satis ed by the utility with the lowest CT in order
to minimize TAC.
2. Over a certain range of D Tm in , the load on only one utility
(namely, the lowest CT utility) will change whereas the
loads on all the other utilities will remain constant.
3. The load on a utility will not increase with D Tm i n when its
CT becomes equal to that of another utility (equation 4); in
this case, a transition occurs to the other new utility in terms
of an increase in its load and a new D Tm i n range on the OLD
plot.
4. If CT for a particular utility goes to zero (equation 5), then
the global optimum will lie in that D Tm i n range where the
load for this utility is changing on the OLD plot.
5. The OLD plot being cumulative leads to the region under
the piecewise-linear total utility curve being divided into
sub-regions associated with the individual utilities: the subregions are vertically stacked starting with the one
corresponding to the most expensive utility at the bottom.
6. The OLD plots exhibit typical characteristics (as in
Figure 4a) with the maximum number of D Tm i n ranges equal
to the number of utilities; however, some of the ranges may
lie over infeasible (negative) D Tm in values and therefore
often a vertical window of the entire OLD plot (i.e., a
portion of Figure 4a) will appear depending on the cost
coef cients.
Network Design at the Global Optimum
Figure 4 . The optimum load distribution plot for multiple utilities with the
associated TAC curve as a function of process D Tmin. (a) OLD plot for three
hot utilities highlightingthe three D Tmin ranges. (b) TAC curve showing the
global optimum.
Figure 5. Network designed at the global optimum target uses all three hot
utilities (HP, MP and LP steam). Temperature in 8 C and load in kW.
Figure 6. The total annual cost (TAC) curves as a function of process D Tmin
for different combinations of utilities.
265
Figure 7. Network that is very close to the global optimum target with the
practical advantage of having only seven units and using only two hot
utilities (HP and LP steam). Temperature in 8 C and load in kW.
Table 3. Practical utility load, area and cost targets for Example 1.
D Tmin
(8 C)
CW Load
(kW)
HP Load
(kW)
MP Load
(kW)
LP Load
(kW)
Area Target
(m2)
Energy Cost
(/yr)
Capital Investment
()
TAC
(/yr)
23.4
23.4
17
32
15.2
725.5
725.5
677.5
790
664
203
240
240
240
314
53
NIL
NIL
NIL
NIL
119.5
135.5
87.5
200
NIL
188
186
202
174
185
51540
52430
49550
56300
56880
150418
148867
161348
139010
147779
96412
96839
97683
97769
100965
266
SHENOY et al.
only one hot utility and ve units are partially offset by the
increased TAC (which shows a penalty of about 5% over the
global optimum as per the targets given in the last row of
Table 3).
Thus, optimal networks can be readily generated which
use one, two or three hot utilities for the above example. The
capital investment for each of these networks may be varied
to match the capital fund availability for the project without
signi cantly affecting the TAC. The globally optimal
network (Figure 5) is not promising in terms of practical
implementation due to its complexity (too many units and
three steam levels). Whether to implement the network with
one steam level (only HP steam) or that with two steam
levels (HP and LP steams as in Figure 7) is in general a
techno-economic decision that needs to consider the total
annual cost, capital investment, energy bill, as well as the
steam and power balance of the total site. Network
complexity and its effect on operability need to be also
considered. The important point is that several promising
options may be rapidly identi ed by pre-design optimization through three useful representations: TAC curves for
different utility combinations (Figure 6), corresponding
OLD plots (Figure 3), and a results summary (Table 3)
showing utility, area and cost implications. Targeting using
TAC curves and OLD plots considerably reduces the tedium
associated with the synthesis, evolution and optimization of
a large number of networks. It thus provides a very effective
screening and scoping tool for multiple utility designs as has
been observed earlier with regard to the supertargeting
methodology3 in the case of a single hot/cold utility.
EFFECT OF CAPITAL COST COEFFICIENTS
During the discussions so far on the CUP and the OLD
plots, a simpli ed exchanger capital cost law (equation (2)
with a = 0 and c = 1) was assumed. Now, a capital cost
equation of the form CC = Nu [b (A/Nu )c ] is considered, and
the effect on the optimal utility load distributions is studied
when c is not unity (typically, less than 1). Figure 8 shows
OLD plots for Example 1 with capital cost coef cients
Figure 8. Optimum load distribution plots for multiple utilities for capital cost coef cient c = 0.83. (a) Use of two hot utilities (HP and MP steam). (b) Use of
three hot utilities (HP, MP and LP steam). Dotted lines highlight the assumption of constant optimal utility load as per the CUP.
267
268
SHENOY et al.
Example 2
The data for the example are given in Table 4. The results
of the six steps followed in the multiple utilities targeting for
this example are brie y outlined below.
1. The total HU varies linearly from 4700 kW (at D Tm in =
Outlet temperature
(8 C)
H1
H2
C3
C4
C5
155
230
115
50
60
85
40
210
180
175
150
85
140
55
60
0.5
0.5
0.5
0.5
0.5
HP Steam
MP Steam
LP Steam
CW
AC
255
205
150
30
40
254
204
149
40
65
Stream
=
=
=
=
0.5
0.5
0.5
0.5
0.5
Cost
($/kW/yr)
70
50
20
10
5
269
Figure 11. The driving force plot (DFP) at a process D Tmin of 168 C for
Example 2 shows that the temperature driving force at the utility pinch
caused by MP steam is not tight.
Figure 9. The optimum load distribution plots for Example 2. (a) OLD
plot for three hot utilities. (b) OLD plot for two cold utilities.
Figure 10. The total annual cost (TAC) curves as a function of process
D Tmin for different combinations of utilities for Example 2.
Figure 12. Network designed with a stream split based on the global
optimum target for Example 2. Temperature in 8 C and load in kW.
270
SHENOY et al.
Figure 13. Network designed without a stream split based on the global
optimum target for Example 2. Temperature in 8 C and load in kW.
Figure 14. Evolved network for Example 2 with a stream split that has the
advantage of having only nine units. Temperature in 8 C and load in kW.
Figure 15. Evolved network for Example 2 that has the advantage of having
only nine units and no stream split. Temperature in 8 C and load in kW.
271
t
TAC
D T1
D Ti
D Tmin
D TR1, D TR2
Subscripts
CU
CW
cold utility
cooling water
A
Af
AC
b
c
C
CT
CC
CU
CUP
CW
DFP
HEN
HP
HU
i
LP
MCp
MP
n
Nu
NLP
OLD
Q
r
R
R1, R2, R3
272
HP
HU
LP
MP
opt
SHENOY et al.
high pressure steam
hot utility
low pressure steam
medium pressure steam
optimum
12.
13.
14.
REFERENCES
1. Linnhoff, B., Townsend, D. W., Boland, D., Hewitt, G. F., Thomas, B.
E. A., Guy, A. R. and Marsland, R. H., 1982, User Guide on Process
Integration for the Ef cient Use of Energy (The Institution of Chemical
Engineers, Rugby, UK).
2. Linnhoff, B. and Hindmarsh, E., 1983, The pinch design method for
heat exchanger networks, Chem Eng Sci, 38(5): 745-763.
3. Linnhoff, B. and Ahmad, S., 1989, Supertargeting: Optimum synthesis
of energy management systems, ASME J Energy Resources Tech,
111(3): 121-130.
4. Shenoy, U. V., 1995, Heat Exchanger Network Synthesis: Process
Optimization by Energy and Resource Analysis (Gulf Publishing Co.,
Houston, TX, USA).
5. Hall, S. G., Parker, S. J. and Linnhoff, B., 1992, Process integration of
utility systems. IEA Workshop on Process Integration, Gothenburg,
Sweden.
6. Kemp, I. C., 1991, Some aspects of the practical application of pinch
technology methods, Trans IChemE, 69(11): 471-479.
7. Parker, S. J., 1989, Supertargeting for multiple utilities, PhD Thesis
(UMIST, UK).
8. Hui, C. W. and Ahmad, S., 1994, Total site heat integration using the
utility system, Comput Chem Eng, 18(8): 729-742.
9. Townsend, D. W. and Linnhoff, B., 1984, Surface area targets for heat
exchanger networks. IChemE Annual Research Meeting, Bath, UK.
10. Colberg, R. D. and Morari, M., 1990, Area and capital cost targets
for heat exchanger network synthesis with constrained matches
and unequal heat transfer coef cients, Comput Chem Eng, 14(1):
1-22.
11. Linnhoff, B. and Flower, J. R., 1978, Synthesis of heat exchanger
15.
16.
17.
18.
ACKNOWLEDGEMENT
The rst author would like to thank Mr Rajdeep Aggarwal (Research &
Development Centre, Engineers India Limited) for his preliminary
observations on this work.
ADDRESS
Correspondence concerning this paper should be addressed to Dr Uday
V. Shenoy, Department of Chemical Engineering, Indian Institute of
Technology, Powai, Bombay 400 076, India. e-mail: uvs@che.iitb.ernet.in.
The manuscript was received 7 August 1997 and accepted for publication
after revision 19 January 1998.