You are on page 1of 30

Subscriber access provided by UNIV OF CAMBRIDGE

Article
Simultaneous Utility and Heat Exchanger Area Targeting
for Integrated Process Synthesis and Heat Integration
Lingxun Kong, Yaqing Wu, and Christos T. Maravelias
Ind. Eng. Chem. Res., Just Accepted Manuscript • DOI: 10.1021/acs.iecr.7b01689 • Publication Date (Web): 19 Sep 2017
Downloaded from http://pubs.acs.org on September 20, 2017

Just Accepted

“Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted
online prior to technical editing, formatting for publication and author proofing. The American Chemical
Society provides “Just Accepted” as a free service to the research community to expedite the
dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts
appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been
fully peer reviewed, but should not be considered the official version of record. They are accessible to all
readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered
to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published
in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just
Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor
changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers
and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors
or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Industrial & Engineering Chemistry Research is published by the American Chemical


Society. 1155 Sixteenth Street N.W., Washington, DC 20036
Published by American Chemical Society. Copyright © American Chemical Society.
However, no copyright claim is made to original U.S. Government works, or works
produced by employees of any Commonwealth realm Crown government in the course
of their duties.
Page 1 of 29 Industrial & Engineering Chemistry Research

1
2
3
4 Simultaneous Utility and Heat Exchanger Area Targeting
5
6
for Integrated Process Synthesis and Heat Integration
7
8 Lingxun Kong, Yaqing Wu and Christos T. Maravelias*
9 Department of Chemical and Biological Engineering, University of Wisconsin-Madison
10
11 1415 Engineering Dr., Madison, WI 53706, USA
12
13
14
15
Abstract
16 We propose a mixed-integer nonlinear programming (MINLP) model for
17
18 simultaneous utility and heat exchanger area targeting with variable stream
19 conditions. The model represents the composite-curve-based area targeting method
20
by constructing the hot and cold composite curves mathematically. We introduce a
21
22 “dynamic” enthalpy grid onto which the stream inlet/outlet temperatures and
23 enthalpies are mapped. By calculating the temperatures at each grid point and the
24
25 stream heat duties at each interval, the utility consumption and heat exchanger
26 areas are simultaneously optimized using an economic criterion. We discuss
27 preprocessing methods tailored to aid the solution of the proposed MINLP model.
28
29 The model is applied to two illustrative examples as well as an example where it is
30 integrated with a process synthesis model.
31
32
33 1. Introduction
34
35 Heat integration has drawn a lot of attention in the field of process systems engineering over the
36 past forty years. Heat integration between process streams not only brings economic benefits by
37
38 reducing energy consumption 1, but also leads to environmental benefits such as a reduction in CO2
39 emissions. Since the concept of heat integration was first introduced by Linnhoff and Flower2, the
40
field has flourished and many approaches have been proposed. These approaches can be divided
41
42 into pinch-analysis-based and mathematical-programming-based methods.3 The former utilize the
43 concept of temperature intervals and a heat recovery approach temperature (HRAT) to identify the
44
45
location of the energy saving bottlenecks (i.e. pinch points). The latter are based on an optimization
46 model that accounts, in some form, for energy balances between streams.4 Heat integration has
47 recently seen renewed interest because it can play a critical role in the design of processes based on
48
49 renewable energy technologies (e.g., biofuels).
50
51
In general, a heat integration problem can be decomposed into three sub-problems.5 The sub-
52 problems can be solved sequentially in three steps and part of the solution to the previous step is
53 used in the next. First, given a heat recovery approach temperature, the minimum utility
54
55 consumption is determined. In pinch analysis, this step involves constructing the hot and cold
56 composite curves and obtaining the minimum utilities graphically. Alternatively, utility targeting
57 can be formulated and solved as a linear programming (LP) problem.6-10 Once the utility target is
58
59
60 1
ACS Paragon Plus Environment
Industrial & Engineering Chemistry Research Page 2 of 29

1
2
3
4 obtained, the second step is to estimate the minimum number of heat exchangers. Typically, this is
5 accomplished by minimizing the number of matches between hot and cold streams. For example,
6 the expanded transshipment model proposed by Papoulias and Grossmann7 predicts the minimum
7
8 number of matches; Galli and Cerda11 proposed an alternative network-based formulation, and
9 Gundersen and coworkers12-14 proposed a vertical heat transfer model-based formulation for
10
determining matches. The last step is to estimate the total area and total cost of the heat exchanger
11
12 network. Floudas et. al.15 introduced a nonlinear programming (NLP) model to minimize the capital
13 cost by generating an actual heat exchanger network. Linnhoff and Ahmad 16 introduced a
14
15 procedure, called “supertargeting”, which considers simultaneously utility and capital costs.
16 Colberg and Morari17 proposed a transshipment-based NLP model for area and capital cost
17 targeting in heat exchange network synthesis (HENS). Later, Zhu18 proposed a MILP model based
18
19 on block decomposition and heuristic rules. Recently, Jezowski et al.19 proposed a transportation-
20 based LP model for area targeting. In the end, the three-step process is repeated with different
21
values of HRAT until the cost of the heat exchanger network (HEN) is minimized.
22
23 The main drawback of the aforementioned sequential approaches is that the tradeoffs among
24
25 energy consumption, number of units, and heat exchanger area are not simultaneously taken into
26 account. Since the decisions made in a step constrain the options in the next steps, the HEN design
27
using a sequential approach does not typically yield the minimum annual cost. To overcome this
28
29 limitation, researchers proposed simultaneous HENS methods in which the optimal network is
30 obtained without decomposition. Floudas et al.15 proposed a MILP model to automatically generate
31
32 the exchanger network that minimizes utility cost and the number of units. Floudas and
33 Grossmann20-21 proposed an NLP formulation that is based on a superstructure representation of
34 all possible HENs. Yuan et al.22 proposed a superstructure-based formulation. Floudas and Ciric23
35
36 proposed the simultaneous match-network HENS model for optimizing the number of heat
37 exchanger and capital investment. Yee and Grossmann24 proposed a stage-wise superstructure-
38 based mixed-integer nonlinear programming (MINLP) model that allows simultaneous HENS,
39
40 which has been widely used as a basis in subsequent work.25-31 Alternatively, transportation-based
41 MILP formulations have been proposed for simultaneous HENS.32-33
42
43 The methods introduced so far assume that stream conditions are fixed, that is, the process is given,
44
and the stream temperatures and flowrates are used as parameters for the HEN synthesis. It is
45
46 known that the main limitation of these approaches is that they neglect the interaction and
47 tradeoffs between the process and the heat integration, which can lead to globally (process + HEN)
48
49
suboptimal solutions. Accordingly, researchers have proposed approaches for simultaneous
50 process synthesis and heat integration. Specifically, methods considering simultaneously process
51 synthesis and utility targeting (but not area targeting) have been proposed. Using the “pinch
52
53 location method”, Duran and Grossmann34 proposed an NLP model for utility targeting under
54 variable stream conditions. The model involves the non-differentiable “max” operator, which is
55 addressed by a smooth approximation.35 This model has been frequently used as the heat
56
57 integration module in simultaneous process synthesis and utility targeting approaches.36-41
58
59
60 2
ACS Paragon Plus Environment
Page 3 of 29 Industrial & Engineering Chemistry Research

1
2
3
4 Grossmann, Yeomans, and Kravanja42 proposed a disjunctive model, which is reformulated into an
5 MINLP, for isothermal and non-isothermal streams. Recently, Navarro-Amoros et al.43 proposed an
6 alternative disjunctive model for simultaneous process design and utility targeting. A common
7
8 assumption in all these methods is that a set of hot streams and a set of cold streams are given
9 before solving the optimization problem. To account for the possibility that some streams cannot be
10
classified as hot or cold a priori, Kong et al.44 proposed an MINLP formulation for simultaneous
11
12 process synthesis and utility targeting with unclassified process streams.
13
14 Despite the work in the area, all previously proposed approaches to simultaneous process synthesis
15 and heat integration do not consider utility and area targeting. Accordingly, in this paper we
16
propose a mathematical-programming-based model for simultaneous utility and heat exchanger
17
18 area targeting with variable stream conditions, which can be integrated with process synthesis (see
19 Figure 1). The model is built upon the composite-curve-based area targeting method by Townsend
20
21 and Linnhoff.45 We introduce “dynamic” enthalpy intervals to account for variable stream
22 temperatures and flowrates. Without obtaining the heat exchanger network, we estimate the heat
23 exchanger areas at each interval46, which are then related to the capital investment. The remaining
24
25 of the paper is structured as follows: In the next section, we present some background on
26 composite-curve-based area targeting. In Section 3, we present the MINLP formulation, and in
27
Section 4, we discuss several extensions. In Section 5, we introduce solution methods. Finally, in
28
29 Section 6 we provide three examples to demonstrate the applicability of the proposed models.
30
31
Process design
32
33 Utility Process Synthesis +
utility target
34
35
36
Exchanger
37
number
38
39
40 Area
41
42 This work
43
Simultaneous
44
HENS
45
46 Figure 1. A classification of work in the field of heat integration.
47
48
49 2. Composite-curve-based Area Targeting
50
In the composite-curve-based area targeting method with fixed stream conditions, the composite
51
52 curves that include all the process streams and utilities are first obtained.45 Since utilities are
53 included, two composite curves are aligned vertically on both ends (i.e. there are no “uncovered”
54
55 sections). The curves are then divided vertically into different sections, namely enthalpy intervals
56 (see Figure 2). The boundaries of these enthalpy intervals are associated with the stream inlet or
57 outlet temperatures. This approach assumes completely countercurrent stream matching (i.e.
58
59
60 3
ACS Paragon Plus Environment
Industrial & Engineering Chemistry Research Page 4 of 29

1
2
3
4 “vertical” matching), that is, heat can only be exchanged between hot and cold streams in the same
5 enthalpy interval. Therefore, a true area target can be achieved only if all the streams have equal
6 film heat transfer coefficients. On the other hand, if the heat transfer coefficients are different, the
7
8 area target can be overestimated. Nevertheless, the composite-curve-based approach provides a
9 reasonable approximation of heat exchanger areas when the differences among stream heat
10
transfer coefficients are within one order of magnitude.47 Further, even when the error is large
11
12 (typically less than 15%), the error in the total HEN cost is expected to be ~5%, which in turn leads
13 to <2% error in the total process cost. This level of error is negligible when trying to screen among
14
15 structural process alternatives using optimization –based methods.
16
The temperature profiles of all the hot and cold streams in each interval follow the temperature
17
18 profile of the corresponding composite curve. Therefore, log-mean temperature differences (Δ )
19 are the same for all stream pairs at interval . The resulting heat exchanger network (HEN) at each
20
21 interval will involve stream splitting and heat exchangers are placed in parallel at the branches of
22 stream splits. Upon splitting, the flowrates and exchanger heat load at each branch are chosen so
23 that isothermal mixing is achieved. Therefore, the area of each individual exchanger at each interval
24
25 can be calculated using Δ and the total heat exchanger area (  ) is the summation of all the
26 individual exchanger areas from all intervals. As shown in Eq. (1),   can be alternatively
27 
estimated from the stream heat duties at each interval (
, and
, ) without generating the
28
29 exchanger network,
30  
,
 ,
31   = ∑   (∑ + ∑ ) (1)
  
32
33 where ! and ! are the heat transfer coefficients of hot and cold streams, respectively.
34
35
36
37
38
39 T
40
41
42
43
44
45
46
47 H
48
49 Figure 2. Enthalpy intervals in the composite-curve-based area targeting. Double lines represent utilities and
50 black arrows represent heat transfer at each interval.
51
52 When the area target is combined with utility target and simultaneous process synthesis, the
53 tradeoffs within the HEN and the interactions between the HEN and the process can be accounted
54
55 for using an economic metric.
56
57
58
59
60 4
ACS Paragon Plus Environment
Page 5 of 29 Industrial & Engineering Chemistry Research

1
2
3
4 3. Proposed Targeting Model
5
6
The proposed targeting model reproduces the composite-curve-based method using a
7 mathematical representation to allow variable stream temperatures and flowrates. For now, we
8 assume that each stream has constant heat capacity. We also assume one shell per heat exchanger and
9
10 do not consider shell targeting. We use " to denote the set of hot streams, which includes hot process
11 streams (" # ) and hot utilities (" $ ); and % to denote the set of cold streams, including cold process
12 streams (% # ) and cold utilities (% $ ).
13
14
15 3.1 Dynamic Enthalpy Grid
16
Under the assumption of constant heat capacity, the composite curves are represented by line
17
18 segments in the temperature-enthalpy graph. Therefore, to construct the composite curves we only
19 need to find the temperature and enthalpy levels at the points where the composite curves change
20
21 slope (i.e. the stream inlets or outlets). When the stream conditions are all fixed, the hot and cold
22 composite curves can be readily constructed from the stream temperatures and enthalpies.
23 However, in our model the variable temperature and flowrate lead to unknown ordering of stream
24
25 inlet and outlet temperatures, making the definition of enthalpy intervals nontrivial.
26
27 Accordingly, to reproduce the enthalpy intervals in composite-curve based targeting, we introduce
28 a “dynamic” enthalpy grid onto which stream temperatures and enthalpies are mapped, in the
29 correct order. Each grid point corresponds to one interval boundary and thus is defined by one
30
31 stream inlet or outlet. Since each stream has one inlet and one outlet, the number of grid points is,
32 at most, twice the total number of streams, including utilities. However, we can reduce the number
33
of grid points to 2(|"| + |%|) − 2 due to the overlap at the first and last grid point (see Figure 3). In
34
35 other words, we allow two stream inlets or outlets at the first and last grid point while all other grid
36 points correspond to exactly one stream inlet or outlet. Here we use ) to denote a set of grid points
37
38 and ) * = )\{0} to denote the set of enthalpy intervals. Following this convention, interval  is in
39 between grid points  − 1 and .
40
41 The mapping is based on a set of binary variables that represents the mapping of a stream
42 inlet/outlet and onto a grid point:
43

44 (1) 0 , /0, equal to one if inlet of stream 2/3 is at grid point 
45

46 (2) 4 , /4, equal to one if outlet of stream 2/3 is at grid point , and
47

48 (3) 5 , / 5, equal to one if stream 2/3 spans enthalpy interval 
49
50 If the inlet of a hot stream is at grid point 1 and its outlet is at 2 (1 > 2), it will span intervals
51 from 2 + 1 to 1; and if a cold stream inlet is at grid point 3 and its outlet is at 4 (4 > 3), it
52
will span intervals from 3 + 1 to 4. This is enforced by the following constraints,
53
54
5 , = 5 ,9 − 0 ,9 + 4 ,9 2 ∈ ",  ∈ ) * (2)
55
   
56 5, = 5,9 + 0,9 − 4,9 3 ∈ %,  ∈ ) * (3)
57
58
59
60 5
ACS Paragon Plus Environment
Industrial & Engineering Chemistry Research Page 6 of 29

1
2
3
4 Further, each stream inlet or outlet should be assigned to exactly one grid point,
5
6 0 1 2 3 4 5
7
8
9
10
11 5
3 4
12
13 T 2
14
15
1
16
17
18
19 H
20
21 Figure 3. Composite curves and enthalpy grid for two hot streams and two cold streams (including non-
22 isothermal utilities). At most 6 grid points and 5 intervals are required.
23

24 ∑ 0 , =1 2∈" (4)
25

26 ∑ 4 , =1 2∈" (5)
27 
28 ∑ 0, =1 3∈% (6)
29 
30 ∑ 4, =1 3∈% (7)
31
32 Finally, to aid the solution process we constrain the number of stream inlets and outlets mapped to
33 each grid point. In general, if we define the number of grid points as 2(|"| + |%|) − 2, then the first
34 and last grid points will each be assigned two stream inlet/outlet while all the other grid points are
35
36 assigned one,
37  
38
∑ (0 , + 4 , ) + ∑(0, + 4, ) = ; ∈) (8)
39
where ; = 1 for  ∈ )\{0, |)|} and ; = 2 for  = {0, |)|} due to the overlap.
40
41
42 3.2 Temperatures and Heat Duties
43
44 First, we introduce a pair of “grid” temperature variables ( and  ) to represent the
45 temperatures of the hot and cold composite curves at grid point . They are ordered as follows,
46
47  ≥ 9

∈) (9)
48
49  ≥ 9

∈) (10)
50
51 To satisfy the laws of thermodynamics, the hot and cold composite curves should not “cross”. Under
52 the constant heat capacity assumption, the pinch locations can only be at the grid points. Therefore,
53 we only need to introduce a non-negative minimum difference (=) between  and  at each grid
54
55 point to prevent the hot and cold composite curves from crossing,
56
57  −  ≥ = ∈) (11)
58
59
60 6
ACS Paragon Plus Environment
Page 7 of 29 Industrial & Engineering Chemistry Research

1
2
3
4 The parameter = can be any sufficiently small positive number (e.g. 1K) since the approach
5 temperature (the minimum of  −  for all  ∈ )) will be determined by optimization.
6
7 Second, using the binary variables and logic introduced in Section 3.1 we propose a mixed-integer
8 formulation to map the stream inlet and outlet temperatures to the grid. Specifically, each
9
10 inequality below becomes equality when the binary variable is one, while it is relaxed when the
11 binary is zero,
12
13 >? − @1 , (1 − 0 ,

) ≤  ≤ >? + @2 , (1 − 0 ,

) 2 ∈ ",  ∈ ) (12)
14
15 BC − D1 , (1 − 4 ,

) ≤  ≤ BC + D2 , (1 − 4 ,

) 2 ∈ ",  ∈ ) (13)
16
17 >? − @1, (1 − 0,

) ≤  ≤ >? + @2, (1 − 0,

) 3 ∈ %,  ∈ ) (14)
18
19
BC − D1, (1 − 4,

) ≤  ≤ BC + D2, (1 − 4,

) 3 ∈ %,  ∈ ) (15)
20 where @1/@2 and D1/D2 are nonnegative parameters for the big-M constraints. These parameters
21
22 represent the largest difference between a stream temperature and a grid temperature. They are
23 determined from the bounds on each pair of stream temperature and grid temperature, and are set
24
25
to zero in the case that the temperature difference is always negative. For example, @1 , can be

26 HHHHH −  ); while D2, = max(0,  − BC ). Note that throughout this work,
chosen as max(0, >?
27 we use overbars/underbars to represent the upper/lower bounds on the corresponding variables.
28
29 If  is always larger than >?I (i.e.  ≥ >?I ), then @1I, is set to zero and the constraints for
30 this pair become: >?I ≤  ≤ >? + @2I, (1 − 0I,
). Also note that for each pair of stream and
31 grid point, @1/D1 can be very different from @2/D2 since they represent relative temperature
32
33 differences. The mapping is illustrated using an example in Figure 4.
34
35 Through Eqs. (2) – (15), stream temperatures are assigned to the appropriate grid temperatures
36 when the corresponding binary variables 0 or Y are equal to one. Note that each grid point  has
37 two corresponding temperatures,  and  , one for the hot and one for the cold composite curve.
38
39 The mapping is performed for only one of the two temperatures at each grid point.
40 Enthalpy grid
41
Eq. (14)
42
43
44 Eq. (12)
45
46
47
48 Eq. (15)

49
50 Eq. (13)

51
52 Figure 4. An example of mapping one hot stream and one cold stream onto an enthalpy grid with 4 grid
53 points. Stream temperatures are connected to the grid temperatures via Eqs. (12) – (15). The grid
54 temperatures that correspond to the hollow points are not mapped by these constraints.
55
56 If we define 2(|"| + |%|) − 2 grid points and enforce Eq. (8), then at each grid point one of the two
57 points on the composite curves does not correspond to any stream inlet or outlet (shown as hollow
58
59
60 7
ACS Paragon Plus Environment
Industrial & Engineering Chemistry Research Page 8 of 29

1
2
3
4 points in Figure 3 and Figure 4). It is of great importance to detect these “hollow points” in order to
5 calculate all the temperatures in the grid. Accordingly, we introduce binary variables C /C that
6 are equal to one when there is no hot/cold stream inlet or outlet locates at grid point .
7

8 C = 1 − ∑ (0 , + 4 , ) ∈) (16)
9
 
10 C = 1 − ∑(0, + 4, ) ∈) (17)
11
12 If C = 1, the slope of the hot composite curve at grid point  should not change; while if C = 1,
13 the slope of the cold composite curve remains unchanged at this point. Mathematically, it implies
14
15 that the total heat capacity flowrates (J ,  and J,  ) of the two intervals adjacent to grid point
16  are equal (Figure 5). Accordingly, we formulate the following mixed-integer constraints in which
17
the heat capacity flowrate equality is enforced if and only if the binary variable U is equal to one,
18
19 J ,  + J ,K = JL
, 
+ J ,#  ∈ ) * \{|)|} (18)
20
21 J ,K ≤ M (1 − C )  ∈ )* (19)
22
23 J ,# ≤ M (1 − C )  ∈ )* (20)
24
25 J,  + J,K = JL
, 
+ J,#  ∈ ) * \{|)|} (21)
26
27 J,K ≤ M (1 − C )  ∈ )* (22)
28
29 J,# ≤ M (1 − C )  ∈ )* (23)
30
31
32
33
34
35 T
36
37
38
39
40
41
42 H
43
44 Figure 5. Heat capacity flowrate equality enforced at the adjacent intervals to the hollow points where no
45 stream enters or finishes at the composite curve.
46
47
where J ,K , J ,# , J,K , and J,# are nonnegative slack variables that are (de)activated by U, and
48
49 M /M are upper bounds on total heat capacity flowrates at interval . For instance, when C
= 1,
,K
50 no hot stream inlet or outlet locates at grid point 1. Thus, Eqs. (19) and (20) enforce J =
51 ,# ,  , 
52
J = 0, and Eq. (18) enforces total-heat-capacity-flowrate equality (i.e. J = JL ). On the
,K ,#
53 other hand, when C = 0, J and J can take positive values so that the equality is relaxed.
54
55 When both stream temperatures and heat capacity flowrates are variables, the calculation of total
56 heat capacity flowrate at each interval becomes nontrivial. Recall that we introduced binary
57 ,#
58
variables Z to detect if a stream spans a given interval or not. Here we introduce variables J , and
59
60 8
ACS Paragon Plus Environment
Page 9 of 29 Industrial & Engineering Chemistry Research

1
2
3 ,#
4 J, to represent the heat capacity flowrate of streams 2/3 in interval . If a hot stream 2 spans
,#
5 interval  (i.e. 5 , = 1), then J , will be equal to the actual stream heat capacity flowrate (J );
,#
6 while if the stream does not span interval , J , will be forced to zero and J will be equal to a
7 ,K ,#
8 slack variable (J , ) instead. Finally, the summation of J , for all the hot streams in interval 
, 
9 yields the total heat capacity flowrate J ,
10
,# ,K
11 J , + J , = J 2 ∈ ",  ∈ ) * (24)
12
,K
13 J , ≤ N , (1 − 5 , ) 2 ∈ ",  ∈ ) * (25)
14 ,#
15 J , ≤ N , 5 , 2 ∈ ",  ∈ ) * (26)
16
17 J ,  = ∑ J ,
,#
 ∈ )* (27)
18 ,#
19
where N , are upper bounds on J , . Similar constraints are formulated for the cold streams:
,# ,K
20 J, + J, = J 3 ∈ %,  ∈ ) * (28)
21
,K  
22 J, ≤ N, (1 − 5, ) 3 ∈ %,  ∈ ) * (29)
23
,#  
24 J, ≤ N, 5, 3 ∈ %,  ∈ ) * (30)
25
26 J,  = ∑ J,
,#
 ∈ )* (31)
27
 ,#
28 where N, are upper bounds on J, .
29

30 Finally, the heat duty of each individual stream (
, or
, ) can be calculated from the
31 temperature differences and the heat capacity flowrates at the interval (Figure 6). Note that the
32
33 enthalpy level at each grid point (O ) can be simultaneously calculated by accumulating the total
34 stream heat duties at each interval, assuming OP = 0 and O increases with .
35 ,#

36 O − O9 = ∑
, = ∑ J , Q − 9

R  ∈ )* (32)
37  ,#
38 O − O9 = ∑
, = ∑ J, ( − 9

)  ∈ )* (33)
39
40 In fact, it is not necessary to calculate O in order to estimate the areas. As shown in Eq. (1), if all
41 the heat transfer coefficients are given, we only need to calculate stream heat duties at each interval
42 and the log-mean temperature differences. Therefore, Eqs. (32) – (33) can be equivalently written
43
44 as a single constraint that enforces energy balance at each interval:
45 ,# ,#
46 ∑ J , Q − 9

R = ∑ J, ( − 9

)  ∈ )* (34)
47
For now we assume that all streams, including utilities, are non-isothermal. In Section 4, we will
48
49 discuss how isothermal streams can be handled. Please also note that, unlike the approach of Duran
50 and Grossmann 34, the use of dynamic enthalpy intervals means we have to introduce binary
51
52 variables and formulate the model as an MINLP.
53
54
55
56
57
58
59
60 9
ACS Paragon Plus Environment
Industrial & Engineering Chemistry Research Page 10 of 29

1
2
3
4
5
6
7
8
9 k
10
11
12
13
14
15
16
17 H
18
19 Figure 6. Heat duty calculations at interval .
20
21 3.3 Areas and Objective
22
23 Recall that in order to estimate the exchanger area at each interval a log-mean temperature
24 difference (Δ ) must be calculated (Eq. (1)). To avoid the numerical difficulties when the
25
26 temperature differences on two boundaries of an interval are equal, here we use an approximation
27 of the log-mean temperature difference 48:
28
I 
29 Δ = TΔ Δ9 + (Δ + Δ9 )  ∈ )* (35)
S U
30
31 where Δ =  −  for all  ∈ ).
32
33 Even with the approximation, temperature difference calculations in Eq. (35) are still nonlinear. To
34 further reduce computational complexity, arithmetic mean temperature differences, Δ =
35
36 (Δ + Δ9 )/2, can be used instead.
37 
38 Since
, and
, are already calculated via Eqs. (32) and (33) and the film heat transfer
39 coefficients are assumed to be constants, exchanger areas (VWX and   ) can be calculated
40 through Eq. (1).
41
42 Finally, the objective is to minimize the total cost, which includes the hot and cold utility costs, and
43
44 the area cost,
45
min ∑ ∈"\ [
,$ + ∑∈%\ [
,$ + ](VWX) (36)
46
47 where [ and [ are prices of hot and cold utilities, respectively,
,$ = ∑
,

and
,$ = ∑
,

48
49 are the heat duties of hot/cold utilities, and ](VWX) represents a cost function of heat exchanger
50 areas:
51
52 • ](VWX) = ^(  )_ , where the area cost is calculated from the power-law function of the
53 total area,
54
55 • ](VWX) = ∑[^(VWX )_ ], where the area cost is a summation of area cost from all
56 intervals, or
57
• ](VWX) = ∑  bc , + ^  , which is a linear approximation of the nonlinear area cost,
58
59
60 10
ACS Paragon Plus Environment
Page 11 of 29 Industrial & Engineering Chemistry Research

1
2
3
4 where parameters ^, d, and b represent the cost prefactor, exponent of the power-law function, and
5 the fixed cost for stream matching, respectively; and c , is a binary variable which is one if the
6 ef
matching between streams 2 and j exists. It is activated via heat exchange (
,, ) between streams 2
7
8 and j at interval ,
9 ef
10
,, ≤ g , c , 2 ∈ ", 3 ∈ %,  ∈ ) * (37)
11
which is related to the stream heat duties,
12
13 ef

, = ∑
,, 2 ∈ ",  ∈ ) * (38)
14
 ef
15
, = ∑
,, 3 ∈ %,  ∈ ) * (39)
16
17 Since the exchanger network configuration is not determined, the economy of scale is represented
18
19
by a power law function of total exchanger area (option 1), or the exchanger area at each interval
20 (option 2). Alternatively, in the last option the economy of scale is accounted for by adding a fixed
21 cost for each pair of streams 2 and 3 on top of a linear cost of total area. The last approach is linear,
22
23 but requires additional variables and constraints.
24
25
26 4. Extensions
27
28 4.1 Isothermal Streams and Multiple Utilities
29
30 As presented in Eqs. (32) – (34), the stream heat duties are calculated as the product of heat
31 capacity flowrates and the interval temperature differences. If isothermal streams (e.g. steam) are
32
present, using the same inlet and outlet temperatures leads to difficulties when constructing the
33
34 enthalpy grid and when calculating heat duties.
35
36 Accordingly, we introduce a small temperature difference (here assumed 1K) between the inlet and
37 outlet of an isothermal stream, which means that, for isothermal streams, J /J represent the
38
latent heat flowrates instead of heat capacity flowrates. The temperature difference, in principle,
39
40 can be chosen differently, e.g., define a 0.1K temperature difference and let J /J be 10 times the
41 latent heat flowrates. In general, using a smaller temperature difference can lead to slightly more
42
43 accurate results, but the computation can become harder because the resulting J /J can be very
44 large, which leads to poor scaling.
45
46 As mentioned previously, utilities are included in the set " and %. Therefore, the heat duties and
47 exchanger areas for multiple utilities are simultaneously calculated. Note that it is necessary to
48 ,# ,#
49 define J /J for utilities for the calculation of J , /J, , which are then used to obtain heat duties
50 and areas in each interval these utilities span. Thus, isothermal and non-isothermal utilities will be
51 handled like isothermal and non-isothermal process streams, respectively, when calculating heat
52
53 duties and exchanger areas.
54
55
56
57
58
59
60 11
ACS Paragon Plus Environment
Industrial & Engineering Chemistry Research Page 12 of 29

1
2
3
4
4.2 Unclassified Streams
5 So far, we have assumed that any process stream in the model either belongs to " # or % # . However,
6
7 it is possible that some streams cannot be classified as hot or cold prior to solving the optimization
8 problem. This can happen, for example, in the superstructure-based process optimization when
9
alternative units operate at different temperatures.
10
11 Following the approach by Kong et. al.44, we extend the model to consider unclassified process
12
13 streams. First, we use h to denote the set of unclassifed process streams. Variables previously

14 defined over 2 ∈ " or 3 ∈ % are now also defined over i ∈ h (e.g. 0j, ).
15
16 In addition, for each unclassified stream we introduce a pair of classification binaries (kj and kj )
17 to represent the hot/cold identities. The classification binaries, kj /kj , will be equal to one if the
18
19 stream is hot/cold, while they will be set to zero otherwise. Using these classification binaries, we
20 determine the stream identities by comparing the inlet and outlet temperatures:
21
22 kj + kj = 1 i∈h (40)
23 >?j − BCj = jL − j9 i∈h (41)
24
25 jL ≤ lj kj i∈h (42)
26
27 j9 ≤ lj kj i∈h (43)
28
29
where jL and j9 are nonnegative slack variables, and lj is an upper bound on the difference
30 between inlet and outlet temperatures.
31
32 Using classification binaries, we modify constraints in Section 3. Since each stream should be
33 assigned to only one grid point and each stream is either classified as hot or cold, Eqs. (4) – (7) for
34
35
unclassified streams become:
36
∑ 0j, = kj i∈h (44)
37

38 ∑ 4j, = kj i∈h (45)
39

40 ∑ 0j, = kj i∈h (46)
41

42 ∑ 4j, = kj i∈h (47)
43
44 For all the remaining constraints in Section 3, we need to replace subscripts “2" and “3” with “s”, and
,#
45 replace sets “"" and “%” with “h”. For example, Eq. (26) becomes: Jj, ≤ Nj, 5j, , i ∈ h,  ∈ ) * .
46
47 Through these constraints, if i′ is a hot stream (kj n = 1 and kjn = 0), then Eqs. (46) and (47) force
48 all the 0jn , and 4jn , to zero, which further set 5jn , and continuous variables with superscript “C”
49
(e.g. Jj,#  *
n , and
j n , ) to zero for all  ∈ ) ; while if i′ is a cold stream, Eqs. (44) and (45) deactivate
50
51 the “hot” counterpart of these variables with superscript “H”. Please see appendix for a complete
52 reformulation for unclassified streams.
53
54 Since the number of grid points required is a function of the total number of streams, following the
55
rules in Section 3 we introduce a grid with 2(|h| + |"| + |%|) − 2 grid points. Therefore, regardless of
56
57 the number of hot and cold streams in the final solution, the enthalpy grid can be constructed by
58
59
60 12
ACS Paragon Plus Environment
Page 13 of 29 Industrial & Engineering Chemistry Research

1
2
3
4 linking binary variables X and Y with classification binaries, and the areas and heat duties at each
5 interval can be calculated from constraints in Section 3.
6
7 4.3 Integration with Process Synthesis
8
9 The heat integration model can be solved in conjunction with a process synthesis model. In this
10
case, the stream temperatures and heat capacity flowrates are the linking variables that couple the
11
12 heat integration and process synthesis modules (Figure 7). The sets of hot and cold process
13 streams (" # and % # ) in the heat integration module include a subset of streams in the process and a
14
15 set of streams that represents the heat duties of processing units.44 For additional details on how to
16 integrate a process model with a heat integration model, the reader is referred to our previous
17 works.44, 49 Given the heat transfer coefficients and utility prices, the integrated model finds the
18
19 optimal stream temperature and heat capacity flowrate that optimize the overall process economy.
20 The utility consumption and exchanger area are obtained from the heat integration module, while
21
the annualized cost of the processing units, material costs, and revenue are calculated from the
22
23 process synthesis model.
24
25 Process
26
synthesis
27
28
29
30
31
32
33 Heat transfer
Utility and area Utility prices
34 coefficients
35 targeting
36
37
38
39 Utilities and Areas
40
41 Figure 7. Integration of process synthesis and heat integration.
42
43
44
5. Solution methods
45
46 5.1 Number of grid points
47
48 As mentioned before, Eq. (8) improves the solution efficiency by introducing a one-to-one
49 correspondence between stream inlet/outlet and grid points for all but the first and last grid points.
50 However, it can be modified to further reduce solution time. For example, if two or more streams
51
52 have the same inlet or outlet temperatures, then Eq. (8) will result in two or more grid points
53 having the same temperatures. In this case, a more effective approach is to define fewer grid points
54
and allow streams with same temperatures to “share” one grid point.
55
56
57
58
59
60 13
ACS Paragon Plus Environment
Industrial & Engineering Chemistry Research Page 14 of 29

1
2
3
4 However, by using fewer than 2(|"| + |%|) − 2 grid points, some of the constraints must be modified
5 accordingly. First, Eq. (8) must be updated since more than one stream inlet/outlet temperatures
6 can be assigned to the same grid point,
7
 
8 1 ≤ ∑ (0 , + 4 , ) + ∑(0, + 4, )≤;  ∈ )\{0, |)|} (8b)
9
10 while for the first and last grid points the equation remains unchanged. The parameter ; represents
11
the maximum number of streams with the same temperatures.
12
13 Other necessary modifications are to remove Eqs. (16) and (17), and rewrite Eqs. (19) – (20) and
14
15 (22) – (23) as follows:
16
17 J ,K ≤ M ∑ (0 ,

+ 4 , )  ∈ )* (19b)
18
19
J ,# ≤ M ∑ (0 ,

+ 4 , )  ∈ )* (20b)
20 J,K ≤ M ∑(0,
 
+ 4, )  ∈ )* (22b)
21
22 J,# ≤ M ∑(0,
 
+ 4, )  ∈ )* (23b)
23
24 The idea is illustrated further through an example in Section 6.
25
26
5.2 Utility Targeting and Preprocessing
27
28 The proposed model can be modified to only minimize utility consumption, by simply removing the
29
30 cost function of exchanger area, ](VWX), from the objective. In this way, the exchanger areas are
31 still calculated, although they are not included in the objective. Alternatively, Eq. (35) can be
32
removed from the optimization model and the area calculation can be performed after the utility-
33
34 targeting problem is solved.
35
36 Compared to the utility and area targeting model (P1), the utility targeting model (P0) is less
37 complex and can be solved significantly faster. By removing constraint (35) and ](VWX) from the
38
objective function, we reduce the number of variables and constraints, especially nonlinear
39
40 constraints. The nonlinear terms in both models are summarized in Table 1.
41
42 Table 1. Comparisons of nonlinearities between P0 and P1.
43 Types of nonlinearities
44 P0 (Utility targeting) Bilinear: Eqs. (32) – (34)
45 Bilinear: Eqs. (1), (32) – (34)
46 P1 (Utility + area targeting)
Power: Eqs. (35) and (36)
47
48 We can trivially show that the optimal solution of the utility targeting model (P0) is a feasible but
49
50 suboptimal solution of the utility and area targeting model (P1), and since P0 is in general easier to
51 solve than P1, for some large instances we could start by solving P0, then use the solutions from P0
52 to aid the solution of P1.
53
54 Here we use (o∗ )qP to denote the globally optimal solution of P0 with a given minimum value of the
55
56 approach temperature (=). Thus, the utility consumption and cost in (o∗ )qP provide lower bounds
57 on the utility consumption and cost in P1. If multiple utilities are present, (o∗ )qP cannot provide
58
59
60 14
ACS Paragon Plus Environment
Page 15 of 29 Industrial & Engineering Chemistry Research

1
2
3
4 lower bounds on each individual utility stream because the individual utility usages that lead to
5 minimum utility cost do not necessary lead to minimum utility consumption. Nevertheless, based
6 on (o∗ )qP we can calculate valid lower bounds on utility costs.
7
8 Further, we can use the total area calculated by (o∗ )qP as an upper bound on the total area in P1.
9
10 Note that although (  )∗qP does not necessary represent the largest total area in any feasible
11 design in P1, it is used as a cutoff to reduce the feasible design space while preserving the optimal
12 solution. This is because (o∗ )qP is a feasible (but suboptimal) solution for P1, and any solution that
13
14 is better must have higher utility cost and lower area cost. In addition, we can provide an upper
15 bound on the objective of P1 as follows:
16
17 (5 ∗ )q ≤ (5 ∗ )qP + ](VWX)∗qP (48)
18
19 where (5 ∗ )qP and ](VWX)∗qP are the optimum objective and area cost in P0, respectively.
20
21 In summary, the preprocessing procedure includes solving a smaller model (P0) to obtain a
22 suboptimal design (for P1), and then based on the optimal solution of P0 reduce the search space by
23
generating variable bounds and eliminating designs that are not as good as (o∗ )qP. As shown in the
24
25 examples in Section 6, implementing this procedure can lead to reduction in solution time.
26
27 5.3 Variable Bounds
28
29 The proposed model is a nonconvex MINLP, which means global optimization solvers should be
30 used to find the globally optimal solution. The generation of tight variable bounds results in tighter
31
32 convex relaxations of nonconvex constraints, leading to potential improvements when using global
33 optimization solvers 50. Further, the parameters (@, D, M, and N) in the variable upper bound
34
constraints in Section 3 are related to the variable upper bounds and thus smaller parameters can
35
36 lead to tighter relaxations. Accordingly, we introduce the following methods to calculate bounds
37 and some important parameters.
38
39 As discussed in Section 3.2, parameters @ and D are calculated from the upper and lower bounds on
40
the stream inlet/outlet and grid temperatures which are inputs to the model. If our model is used in
41
42 conjunction with a model for process design, bounds can be inferred from design and unit
43 operation specifications (e.g. the reaction temperature in the reactor). Second, in terms of bounds
44
45 on grid temperatures, since  and  are ordered stream inlet or outlet temperatures, a valid
46 choice is H = max (>? HHHHHHHH ),  = min (>? , BC ), H = max (>?
HHHHH , BC HHHHHHHH ), and
HHHHH , BC
47  = min (>? , BC ) for all  ∈ ). However, in some cases we can obtain tighter bounds by
48
49 analyzing bounds on stream inlet and outlet temperatures. For example, if we know that cooling
50 water has a lower outlet temperature than any other stream, then  and I will be equal to the
51
cooling water inlet and outlet temperatures, respectively. Therefore, H / and HI /I can be
52
53 updated to be equal to the bounds on cooling water inlet and outlet temperatures, respectively.
54 Similar procedure can be applied whenever we know a priori the matching between a grid
55
56 temperature and a stream temperature, and in such cases we propagate bounds on stream
57 temperatures to bounds on grid temperatures.
58
59
60 15
ACS Paragon Plus Environment
Industrial & Engineering Chemistry Research Page 16 of 29

1
2
3  ,# ,#
4 Parameters N , and N, , which are the upper bounds on J , and J, , respectively, are the same as

5 the upper bounds on the corresponding J and J . For process streams (" # and % # ), upper bounds
6 on stream heat capacity flowrates are given as inputs to the model; while for utilities, bounds on J
7
8 and J are related to the utility usage (
,$ and
,$ ) since the temperature difference between a

9 utility stream inlet and outlet is usually known. Therefore, in order to calculate all N , and N, , we
first need to estimate upper bounds on
and
 . The maximum utility usages (
and
,$ )
,$ ,$ ,$
10 H H
11
12 can be obtained by assuming all process stream heat duties are satisfied by utilities:
13
14
H ,$ = ∑∈%r JH (BC
HHHHHHHH − >? ) 2 ∈ " $ (49)

15
16
H,$ = ∑ ∈"r JH (>?
HHHHH − BC ) 3 ∈ % $ (50)

17
18 where
H ,$ /
H,$ are calculated from bounds on stream temperatures and heat capacity flowrates;
19 % # is the set of cold streams whose inlet temperatures are lower than the inlet temperature of hot
20
21 utility 2; and "# is the set of hot streams whose outlet temperatures are higher than the outlet
22 temperature of cold utility 3. In other words, "# and % # are the streams that can potentially exchange
23
heat with a given utility. If the hot and cold utilities are at the highest and lowest temperatures,
24
25 respectively, then % # = % # and "# = " # . Once
H ,$ and
H,$ are determined, N ,
, 2 ∈ " $ and

26 N, , 3 ∈ % $ can be back calculated:
27
28
N , =
H ,$ /(>? − BC ) 2 ∈ " st ,  ∈ ) * (51)
29
30 
N, =
H,$ /(BC − >? ) 3 ∈ % ut ,  ∈ ) * (52)
31
32 where the inlet and outlet temperatures of utilities are assumed fixed.
33
34 Finally, the upper bounds on total heat capacity flowrates at interval , M /M , must be determined.
35 One valid choice is M = ∑ ∈" JH and M = ∑∈% JH for all ∈ ) * , where " and % are the sets of hot
36
and cold streams that could potentially span interval , respectively.
37
38 In general, the bounds obtained using the approaches in this section are not the tightest, but when
39
40 combined with the bounds calculated in Section 5.3 lead to tighter models and improve
41 computational performance. A preprocessing algorithm that includes the procedures described in
42
Section 5.2 and 5.3 is shown in the supporting information.
43
44
45 6. Examples
46
47 We present three examples with variable stream temperatures and flowrates. The first example
48
49 includes 6 non-isothermal streams, and one hot and one cold utility; the second example has an
50 isothermal process stream and an additional intermediate hot utility; and the third example
51
involves simultaneous process synthesis and heat integration with utility and exchanger area
52
53 targeting. The models for examples are formulated in GAMS (24.7.4) and solved using BARON50 on a
54 desktop with 3.60 GHz Intel i7 processors.
55
56
57
58
59
60 16
ACS Paragon Plus Environment
Page 17 of 29 Industrial & Engineering Chemistry Research

1
2
3
4
6.1 Example 1
5 The specifications on stream temperatures are shown in Figure 8, while heat capacity flowrates
6
7 (thereafter referred to as “flows”), and heat transfer coefficients are given in the supporting
8 information. An isothermal hot utility is assumed to be available at the highest temperature (330°C),
9
while a non-isothermal cold utility enters at 20°C and exits at 30°C.
10
11
12 T( ) T( )
13 350 350
Hot utility
14
15 300 300
16
17 250 250
18 h3
200 200
19 h2
20
21 150 h1 150
22 c3
100 100
23
24 c2
50 c1 50
25
26 0 0
Cold utility
27
28 Figure 8. Stream optimal temperatures in Example 1. If the inlet/outlet temperature of a stream is allowed to
29 vary, the bound is shown on the immediate left.
30
31 Since there are 8 streams in total (including utilities), we define an enthalpy grid with 14 grid
32
33 points. We use Eq. (35) to calculate the temperature difference at each interval. The minimum value
34 of approach temperature (=) is assumed to be 5°C. The area cost function is chosen as:
35
](VWX) = $40000/yr ∙ m9.U (  )P.{, which approximates the area cost as a power-law function
36
37 of the total exchanger area.
38
39 Before solving the simultaneous utility and area targeting model (P1), we solve the utility targeting
40 model (P0) to obtain variable bounds. P0 comprises of 719 variables (328 binary) and 1132
41
constraints (312 nonlinear entries). A globally optimal solution is obtained in 5 seconds. The hot
42
43 and cold utility consumptions and costs, and the objective value of P0 are then used to lower-bound
44 the corresponding variables in P1. Using the temperatures and heat duties from (o∗ )qP, we
45
46 calculate the areas and area costs and use them as upper bounds in P1.
47
Next, we solve P1 to simultaneously obtain the utility and area target. P1 includes 747 variables
48
49 (328 binary) and 1160 constraints (391 nonlinear entries). The globally optimal solution is
50 obtained in 140 seconds. On the other hand, solving P1 directly without generating bounds using
51
52
(o∗ )qP, requires about 22% longer solution time (180 seconds).
53
In this example, several streams have the same inlet or outlet temperature, which leads to
54
55 symmetric solutions. To address this, we introduce fewer grid points and allow streams with same

56 temperatures to share one grid point. Specifically, >?| , BC|I , and BC|S can be mapped to
57   
58 one grid point; and BC} , BC}I , and >?}S can share one grid point. Therefore, the total
59
60 17
ACS Paragon Plus Environment
Industrial & Engineering Chemistry Research Page 18 of 29

1
2
3
4 number of grid points required can be reduced to 10. With the modifications introduced in Section
5 5.1, we solve the example again using 10 grid points. The model now includes only 828 constraints
6 and 531 variables (232 binary variables). The solution time is reduced to 26 seconds.
7
8
350
9
10 300

11 250 Pinch point


12 200
T (C)

13
14 150

15 100 Hot streams


16 50 Cold streams
17
0
18
0 50 100 150 200
19 H (MW)
20
21 Figure 9. Hot and cold composite curves from the optimal solution of example 1. The pinch location and the
22 approach temperature (17°C) are also determined.
23
24 Optimal stream temperatures are shown in Figure 8, while other information such as optimal
25 flowrates and utility consumptions can be found in the supporting information. The total area is
26
27 6606 m2, resulting in a $4.5MM/yr area cost. The results can be alternatively shown as hot and cold
28 composite curves (see Figure 9), from which the locations of the pinch is determined. The approach
29 temperature in this example is found to be 17°C, which is larger than the pre-specified = of 5°C.
30
31
32 6.2 Example 2
33
The second example features multiple utilities and isothermal streams. An isothermal intermediate
34
35 utility is introduced with a constant temperature of 180°C and a unit cost of $50/kW-yr. Seven
36 process streams are included, of which one stream (Stream #h3) is a variable temperature
37
38 isothermal hot stream (>?S − BCS = 1°C).
39
40 T( ) T( )
41 350 Hot utility 350
42
43 300 300
44
45 250 250
Interm. h2
46
200 hot utility 200
47 h1
48
150 150
49 h3
50 100 100
c2 c3
51
52 50 c1 50
53 c4
54 0 0
Cold utility
55
56 Figure 10. Stream optimal temperatures in Example 2. If the inlet/outlet temperature of a stream is allowed
57 to vary, the bound is shown on the immediate left.
58
59
60 18
ACS Paragon Plus Environment
Page 19 of 29 Industrial & Engineering Chemistry Research

1
2
3
4 The example thus has 10 streams, which leads to 18 grid points and 17 intervals. Again, the
5 Paterson48 approximation is used for the log-mean temperature differences and = = 5°C is assumed.
6 The area cost function is the same as that in the previous example. We apply the preprocessing
7
8 procedures from Section 5.2 and 5.3 to calculate variable bounds and parameters.
9
10 The simultaneous utility and area targeting model (P1) consists of 1587 constraints and 1609
11 variables, of which 566 are binary variables. The problem (including preprocessing) is solved to
12 optimality in 2000 seconds. For comparison, without the preprocessing methods discussed in
13
14 Section 5.2, the solution time increases to 2800 seconds. The optimal stream temperatures are
15 shown in Figure 10. The total area is found to be 15961 m2, with a cost of $9.2MM/yr. As shown in
16
Figure 11, the composite curve pinches at the 7th grid point, which corresponds to the inlet of
17
18 stream #3, with an approach temperature of 10.5°C.
19
20 350
21 300
22
250
23
Pinch point
24 200
T (C)

25 150
26
100 Hot streams
27
50 Cold streams
28
29 0
30 0 100 200 300 400
31 H (MW)
32 Figure 11. Hot and cold composite curves from the optimal solution of example 2. The pinch location and the
33 approach temperature (10.5°C) are also determined.
34
35 6.3 Simultaneous Process Synthesis and Heat Integration
36
37 6.3.1 Problem Statement and Assumptions
38
39 In this example, product  is synthesized from reactants  and ‚ via intermediate ƒ. The first
40 reaction (RXN1) produces ƒ from  and ‚ in a continuous stir-tank reactor (CSTR1), which
41
42
operates isothermally at 400K. The heat of reaction in RXN1 is insignificant and thus not considered
43 in the heat integration. The effluent stream of CSTR1 is sent to a flash tank (SEP1) to separate
44 unreacted  and ‚ from intermediate ƒ. The separation is assumed to be sharp (i.e. all the light
45
46 components come out from the top and all the heavy components exit at the bottom). The top of the
47 flash containing unreacted  and ‚ is recycled, while intermediate ƒ from the bottom of the flash is
48
sent to another isothermal continuous stir-tank reactor (CSTR2) to produce final product D via
49
50 reaction RXN2, which is assumed to reach equilibrium. The equilibrium constant is a function of
51 temperature (see supporting information for details). RXN2 is an exothermic reaction and the heat
52
53
of reaction is significant, so an isothermal hot stream (r1) is introduced here to represent the heat
54 duty of CSTR2. Finally, the effluent stream of CSTR2 is sent to a series of two separators (SEP2 and
55 SEP3) for final product purification. While the separation in SEP3 is assumed to be sharp, the split
56
57 fraction of SEP2 is 0.6 for ƒ and 0 for . The unreacted ƒ from the top of the separator is recycled.
58
59
60 19
ACS Paragon Plus Environment
Industrial & Engineering Chemistry Research Page 20 of 29

1
2
3
4 The annualized cost of each unit is approximated by a power-law function of the total inlet molar
5 flow. It is assumed that the feed stream (stream 1) component molar flowrates are 2 kmol/s of 
6 and 1 kmol/s of ‚, and the final product D is sold at a price of $1.65/kmol. The objective is to
7
8 maximize profit, which takes into account the revenue, cost of materials, annualized capital cost,
9 utility cost, and heat exchanger area cost.
10
11 There are five streams that require heating or cooling ({2,3,8,9, V1} in Figure 12), of which three
12 (streams 3, 8, and r1) are hot and two (streams 2 and 9) are cold. With one hot and one cold utility,
13
14 we define an enthalpy grid with 12 grid points.
15
16 5
17 A,B C 15 14
18 6
19 380K 4
20 A,B
21 334K 2 3
A,B,C 10 13
300K 350K
22 SEP1 C C
1 8 7.38MW 9
23 17.8MW 420K C,D 11 12
24 CSTR1 4.68MW SEP3
7 SEP2
25 C C,D D
21.2MW CSTR2 24.6MW 430K 430K
26
27
28 Figure 12. Optimal process flowsheet in the example. Streams numbered in circles are process streams that
29 require heating/cooling.
30
31 6.3.2 Results
32
33 The resulting model consists of 663 variables (245 binaries) and 979 constraints. It was solved to
34 global optimality in 8300 seconds with an objective of -$10.7 MM/yr. Temperature of CSTR2 is
35 optimized to be 350K, leading to an equilibrium constant of 0.25 and a 0.20 single-pass conversion
36
37 of ƒ in RXN2. The final product  is produced at a rate of 0.923 kmol/s, which leads to an overall
38 yield of 92.3% with respect to limiting reactant ‚. Optimal stream conditions are shown in Figure
39
12 and in the supporting information. After heat integration, the process requires 25.6 MW external
40
41 heating and 16.4 MW cooling, leading to a total utility cost of $2.4 MM/yr. The total heat exchange
42 area is estimated to be 1059 m2 and the area cost is $3.14 MM/yr.
43
44 Note that the temperature of the second reactor is at its upper bound (350K), which is unfavorable
45
in terms of a single-pass conversion because the reaction is exothermic. While high overall
46
47 conversion can be achieved through recycle (i.e. a larger flowrate in stream 15), this would in turn
48 increase the sizes/costs of CSTR2, SEP2, and SEP3. Moreover, the heat duty in CSTR2 must be
49
50 satisfied by cold utility because no cold process stream has low enough temperature to receive the
51 heat. This penalizes a larger recycle flowrate and thus makes high reactor temperature unfavorable.
52
53 However, the advantage of a high reactor temperature is that it results in a smaller heat exchanger
54 area. Since CSTR2 exchanges heat only with cold utility, a higher temperature implies greater
55
56 driving force and therefore smaller exchanger area. In this example, the reduction in area cost
57
58
59
60 20
ACS Paragon Plus Environment
Page 21 of 29 Industrial & Engineering Chemistry Research

1
2
3
4 offsets the aforementioned disadvantages of high temperature, and thus †‡ˆ‰I = 350Š is globally
5 optimal.
6
7 Interestingly, if exchanger costs were not considered, †‡ˆ‰I would instead be at its lower bound.
8 This is confirmed by solving the utility targeting model (P0) for this example. Previous approaches
9
10 such as the ones by Duran and Grossmann 34 and Kong et al.44 will also set the reactor temperature
11 to its lower bound. This further demonstrates the advantage of the proposed method.
12
13
14 7. Conclusions
15
16 We proposed an MINLP model for simultaneous utility and exchanger area targeting. The model
17 accounts for variable stream temperatures and flowrates, allowing it to be used for simultaneous
18
19 process synthesis and heat integration. The model represents the composite-curve-based area
20 targeting method. Stream heat duties and exchanger areas are calculated in each interval. We
21 present three alternatives to account for the cost of exchanger area. One possible future work
22
23 direction is to improve the current area calculation to obtain more accurate area targets. The model
24 can be extended to handle isothermal streams and multiple utilities, as well as streams that cannot
25
be classified as hot/cold a priori. Furthermore, we discussed several solution techniques, including
26
27 a preprocessing algorithm for the calculation of tight variable bounds. Finally, we showed how the
28 proposed model can be integrated with a process models and used for simultaneous process
29
30
synthesis and heat integration.
31
32
33
Supporting Information
34 Preprocessing algorithm; stream specifications and optimal solutions of Example 1 and Example 2;
35
36 process model in §6.3; and additional information for example in §6.3.
37
38 This information is available free of charge via the internet at http://pubs.acs.org/.
39
40
41 Author Information
42 Corresponding author
43
44 *E-mail address: maravelias@wisc.edu
45
46
47 Acknowledgements
48
49 This work was funded by the DOE Great Lakes Bioenergy Research Center (DOE BER Office of
50 Science DE-FC02-07ER64494).
51
52
53
54
55
56
57
58
59
60 21
ACS Paragon Plus Environment
Industrial & Engineering Chemistry Research Page 22 of 29

1
2
3
4 Appendix A. Alternative mapping constraints
5
We present an alternative convex-hull-based reformulation of constraints (12) – (15),
6
,*Œ ,*Œ
7  + ‹ , = >? +  , 2 ∈ ",  ∈ ) (A1)
8
,*Œ
9 ‹ ,
≤ @1 , (1 − 0 , ) 2 ∈ ",  ∈ ) (A2)
10
,*Œ
11  ,
≤ @2 , (1 − 0 , ) 2 ∈ ",  ∈ ) (A3)
12
, $ , $
13  + ‹ , = BC +  , 2 ∈ ",  ∈ ) (A4)
14
, $
15 ‹ ,
≤ D1 , (1 − 4 , ) 2 ∈ ",  ∈ ) (A5)
16 , $
17  ,
≤ D2 , (1 − 4 , ) 2 ∈ ",  ∈ ) (A6)
18 ,*Œ ,*Œ
19  + ‹, = >? + , 3 ∈ %,  ∈ ) (A7)
20 ,*Œ
21 ‹, 
≤ @1, (1 − 0, ) 3 ∈ %,  ∈ ) (A8)
22 ,*Œ
, 
≤ @2, (1 − 0, ) 3 ∈ %,  ∈ ) (A9)
23
, $ , $
24  + ‹, = BC + , 3 ∈ %,  ∈ ) (A10)
25
, $
26 ‹, 
≤ D1, (1 − 4, ) 3 ∈ %,  ∈ ) (A11)
27
, $
28 , 
≤ D2, (1 − 4, ) 3 ∈ %,  ∈ ) (A12)
29
,*Œ  ,*Œ ‹ , $  , $ ‹ ,*Œ  ,*Œ ‹ , $ , $
30 where ‹ , ,  , ,  , ,  , , , , , , , , and , are nonnegative slack variables
31 that are deactivated by the corresponding binary variables through variable-upper-bound
32
33 constraints.
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 22
ACS Paragon Plus Environment
Page 23 of 29 Industrial & Engineering Chemistry Research

1
2
3
4 Appendix B. Modeling of classified and unclassified streams
5
6
If unclassified process streams (i ∈ h) are present, we first introduce the following constraints:
7
5j, = 5j,9 − 0j,9 + 4j,9 i ∈ h,  ∈ ) * (B1)
8
   
9 5j, = 5j,9 + 0j,9 − 4j,9 i ∈ h,  ∈ ) * (B2)
10

11 >?j − @1j, (1 − 0j, ) ≤  ≤ >?j + @2j, (1 − 0j,

) i ∈ h,  ∈ ) (B3)
12

13 BCj − D1j, (1 − 4j, ) ≤  ≤ BCj + D2j, (1 − 4j,

) i ∈ h,  ∈ ) (B4)
14

15 >?j − @1j, (1 − 0, ) ≤  ≤ >?j + @2j, (1 − 0j,

) i ∈ h,  ∈ ) (B5)
16

17 BCj − D1j, (1 − 4, ) ≤  ≤ BCj + D2j, (1 − 4j,

) i ∈ h,  ∈ ) (B6)
18 ,# ,K
19 Jj, + Jj, = Jj i ∈ h,  ∈ ) * (B7)
20 ,K
21 Jj, ≤ Nj, (1 − 5j, ) i ∈ h,  ∈ ) * (B8)
22 ,#
Jj, ≤ Nj, 5j, i ∈ h,  ∈ ) * (B9)
23
24 ,# ,K
Jj, + Jj, = Jj i ∈ h,  ∈ ) * (B10)
25
,K  
26 Jj, ≤ Nj, (1 − 5j, ) i ∈ h,  ∈ ) * (B11)
27
,#  
28 Jj, ≤ Nj, 5j, i ∈ h,  ∈ ) * (B12)
29
,#
30 ∑j
j, = ∑j Jj, Q − 9

R  ∈ )* (B13)
31
 ,#
32 ∑j
j, = ∑j Jj, ( − 9

)  ∈ )* (B14)
33
34 Then, we replace Eqs. (1), (8), (16) – (17), (27), (31), and (34) with the following constraints:
35    
 , , , L,
36   = ∑  (∑ + ∑ + ∑j ) (B15)
    Ž
37
38    
∑ (0 , + 4 , ) + ∑(0, + 4, ) + ∑j(0j, + 4j, + 0, + 4, ) = ; ∈) (B16)
39

40 C = 1 − ∑ (0 , + 4 ,
) − ∑j(0j,
+ 4j, ) ∈) (B17)
41
 
42 C = 1 − ∑(0, + 4, 
) − ∑j(0j, 
+ 4j, ) ∈) (B18)
43
44 ,#
J ,  = ∑ J , ,#
+ ∑j Jj,  ∈ )* (B19)
45
46 J,  = ∑ J,
,# ,#
+ ∑j Jj,  ∈ )* (B20)
47
,# ,# ,# ,#
48 Q − 9

RQ∑ J , + ∑j Jj, R = Q − 9

R(∑ J, + ∑j Jj, )  ∈ )* (B21)
49
50 All the constraints in this section, together with those from Section 3, complete the modeling of
51 classified and unclassified streams.
52
53
54
55
56
57
58
59
60 23
ACS Paragon Plus Environment
Industrial & Engineering Chemistry Research Page 24 of 29

1
2
3
4 Notation
5
Sets
6
7 " hot streams
8
9 % cold streams
10
11 ) enthalpy interval boundaries/enthalpy grid points
12
13 h unclassified streams
14
15 t processing units
16
17 Subsets
18
19 " $ hot utility streams
20
21 " set of hot streams that potentially span interval 
22
"# hot process streams
23
24 % $ cold utility streams
25
26 % set of cold streams that potentially span interval 
27
28 %# cold process streams
29
30 )* enthalpy intervals
31
32 Parameters
33
34 @ /@ upper bound on hot/cold stream inlet temperature
35
36 D /D upper bound on hot/cold stream outlet temperature
37
38
lj upper bound on the inlet and outlet temperature difference for unclassified
39 stream s
40
41 ; maximum number of streams at point 
42
43 = minimum approach temperature
44
45 M /M upper bound on total heat capacity flowrate at interval 
46
! /! heat transfer coefficient of hot/cold process stream 2/3
47
48 
N , /N, upper bound on heat capacity flowrate of stream 2/3 at interval 
49
50 ^ cost prefactor for heat exchanger area
51
52  cost prefactor of unit ‘
53
54 [ /[ unit cost for hot/cold utilities
55
56 g , upper bound on heat exchanger between stream 2 and 3
57
58
59
60 24
ACS Paragon Plus Environment
Page 25 of 29 Industrial & Engineering Chemistry Research

1
2
3
4 Binary variables
5 
0 , /0, 1 if the inlet of hot/cold stream 2/3 is mapped onto grid point 
6
7 
4 , /4, 1 if the outlet of hot/cold stream 2/3 is mapped onto grid point 
8

9 5 , /5, 1 if hot/cold stream 2/3 spans enthalpy interval 
10
11 C /C 1 if no hot/cold stream inlet or outlet is mapped onto grid point 
12
13 c , 1 if there exists a match between streams 2 and 3
14
15 kj /kj 1 if stream i is a hot/cold stream
16
17 Non-negative continuous variables
18
19 VWX heat exchanger area at interval k
20
21   total heat exchanger area
22
J /J stream heat capacity flowrate
23
24 ,# ,#
J , /J, stream heat capacity flowrate at interval 
25
26 J ,  /J,  total heat capacity flowrate at interval 
27
28 J ,K /J,K dummy variable for total heat capacity flowrate at interval k
29
30 J ,# /J,# dummy variable for total heat capacity flowrate at interval k
31
,K ,K
32 J , /J, dummy heat capacity flowrate
33
34 O cumulative enthalpy at grid point k
35 ef
36
,, heat exchanged between stream 2 and 3 at interval 
37 
38

, /
, heat duty of stream 2/3 at interval k
39

,$ /
,$ hot/cold utility consumption
40
41  / hot/cold temperature at grid point k
42
43 Δ temperature difference at grid point k
44
45  log-mean temperature difference at enthalpy interval k
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 25
ACS Paragon Plus Environment
Industrial & Engineering Chemistry Research Page 26 of 29

1
2
3
4 References
5 1. Linnhoff, B.; Hindmarsh, E., The pinch design method for heat exchanger networks. Chemical
6 Engineering Science 1983, 38 (5), 745-763.
7
8 2. Linnhoff, B.; Flower, J. R., Synthesis of heat exchanger networks: I. Systematic generation of
9 energy optimal networks. Aiche J 1978, 24 (4), 633-642.
10
11 3. Morar, M.; Agachi, P. S., Review: Important contributions in development and improvement
12 of the heat integration techniques. Computers & Chemical Engineering 2010, 34 (8), 1171-1179.
13
4. Klemeš, J. J.; Kravanja, Z., Forty years of Heat Integration: Pinch Analysis (PA) and
14
15 Mathematical Programming (MP). Current Opinion in Chemical Engineering 2013, 2 (4), 461-474.
16 5. Furman, K. C.; Sahinidis, N. V., A critical review and annotated bibliography for heat
17 exchanger network synthesis in the 20th century. Industrial & Engineering Chemistry Research
18 2002, 41 (10), 2335-2370.
19
20 6. Cerda, J.; Westerberg, A. W.; Mason, D.; Linnhoff, B., Minimum utility usage in heat
21 exchanger network synthesis A transportation problem. Chemical Engineering Science 1983, 38 (3),
22 373-387.
23
24 7. Papoulias, S. A.; Grossmann, I. E., A structural optimization approach in process synthesis—
25 II. Computers & Chemical Engineering 1983, 7 (6), 707-721.
26
27 8. Jez̆ owski, J.; Friedler, F., A simple approach for maximum heat recovery calculations.
28 Chemical engineering science 1992, 47 (6), 1481-1494.
29 9. El-Halwagi, M. M.; Manousiouthakis, V., Synthesis of mass exchange networks. Aiche J 1989,
30 35 (8), 1233-1244.
31
32 10. Roxenby, S.; Manousiouthakis, V., Non-Isothermal Separable Mass Exchange Networks:
33 Minimum Utility Cost through the State Space Approach. Report, LUTKDH=(TKKA-5004) 1994, 1-
34 133.
35
36 11. Galli, M. a. R.; Cerdá, J., Synthesis of structural-constrained heat exchanger networks—I.
37 Series networks. Computers & chemical engineering 1998, 22 (7), 819-839.
38
39 12. Gundersen, T.; Grossmann, I. E., Improved optimization strategies for automated heat
40 exchanger network synthesis through physical insights. Computers & chemical engineering 1990, 14
41 (9), 925-944.
42 13. Gundersen, T.; Duvold, S.; Hashemi-Ahmady, A., An extended vertical MILP model for heat
43
44
exchanger network synthesis. Computers & chemical engineering 1996, 20, S97-S102.
45 14. Gundersen, T.; Traedal, P.; Hashemi-Ahmady, A., Improved sequential strategy for the
46 synthesis of near-optimal heat exchanger networks. Computers & chemical engineering 1997, 21,
47 S59-S64.
48
49 15. Floudas, C. A.; Ciric, A. R.; Grossmann, I. E., Automatic synthesis of optimum heat exchanger
50 network configurations. Aiche J 1986, 32 (2), 276-290.
51
52 16. Linnhoff, B.; Ahmad, S., Supertargeting: Optimum synthesis of energy management systems.
53 Journal of energy resources technology 1989, 111 (3), 121-130.
54
17. Colberg, R.; Morari, M., Area and capital cost targets for heat exchanger network synthesis
55
56
with constrained matches and unequal heat transfer coefficients. Computers & chemical engineering
57 1990, 14 (1), 1-22.
58
59
60 26
ACS Paragon Plus Environment
Page 27 of 29 Industrial & Engineering Chemistry Research

1
2
3 18. Zhu, X. X., Automated design method for heat exchanger network using block decomposition
4
and heuristic rules. Computers & chemical engineering 1997, 21 (10), 1095-1104.
5
6 19. Jeżowski, J. M.; Shethna, H. K.; Castillo, F. J., Area target for heat exchanger networks using
7 linear programming. Industrial & engineering chemistry research 2003, 42 (8), 1723-1730.
8
9 20. Floudas, C. A.; Grossmann, I. E., Synthesis of flexible heat exchanger networks for
10 multiperiod operation. Computers & chemical engineering 1986, 10 (2), 153-168.
11
12 21. Floudas, C. A.; Grossmann, I. E., Synthesis of flexible heat exchanger networks with
13 uncertain flowrates and temperatures. Computers & chemical engineering 1987, 11 (4), 319-336.
14 22. Yuan, X.; Pibouleau, L.; Domenech, S., Experiments in process synthesis via mixed-integer
15 programming. Chemical Engineering and Processing: Process Intensification 1989, 25 (2), 99-116.
16
17 23. Floudas, C.; Ciric, A., Strategies for overcoming uncertainties in heat exchanger network
18 synthesis. Computers & Chemical Engineering 1989, 13 (10), 1133-1152.
19
20 24. Yee, T. F.; Grossmann, I. E., Simultaneous optimization models for heat integration--II. Heat
21 exchanger network synthesis. Computers & Chemical Engineering 1990, 14 (10), 1165-1184.
22
25. Ravagnani, M.; Caballero, J. A., Optimal heat exchanger network synthesis with the detailed
23
24 heat transfer equipment design. Computers & Chemical Engineering 2007, 31 (11), 1432-1448.
25 26. Isafiade, A.; Fraser, D., Interval-based MINLP superstructure synthesis of heat exchange
26 networks. chemical engineering research and design 2008, 86 (3), 245-257.
27
28 27. Ponce-Ortega, J. M.; Jiménez-Gutiérrez, A.; Grossmann, I. E., Optimal synthesis of heat
29 exchanger networks involving isothermal process streams. Computers & Chemical Engineering
30 2008, 32 (8), 1918-1942.
31
32 28. Frausto-Hernández, S.; Rico-Ramırez, V.; Jiménez-Gutiérrez, A.; Hernández-Castro, S., MINLP
33 synthesis of heat exchanger networks considering pressure drop effects. Computers & chemical
34 engineering 2003, 27 (8), 1143-1152.
35
36 29. Soršak, A.; Kravanja, Z., Simultaneous MINLP synthesis of heat exchanger networks
37 comprising different exchanger types. Computers & Chemical Engineering 2002, 26 (4), 599-615.
38 30. Bagajewicz, M. J.; Pham, R.; Manousiouthakis, V., On the state space approach to mass/heat
39
exchanger network design. Chemical Engineering Science 1998, 53 (14), 2595-2621.
40
41 31. Huang, K. F.; Al-mutairi, E. M.; Karimi, I., Heat exchanger network synthesis using a
42 stagewise superstructure with non-isothermal mixing. Chemical engineering science 2012, 73, 30-
43 43.
44
45 32. Nguyen, D. Q.; Barbaro, A.; Vipanurat, N.; Bagajewicz, M. J., All-at-once and step-wise
46 detailed retrofit of heat exchanger networks using an MILP model. Industrial & Engineering
47 Chemistry Research 2010, 49 (13), 6080-6103.
48
49 33. Barbaro, A.; Bagajewicz, M. J., New rigorous one-step MILP formulation for heat exchanger
50 network synthesis. Computers & chemical engineering 2005, 29 (9), 1945-1976.
51
34. Duran, M. A.; Grossmann, I. E., Simultaneous optimization and heat integration of chemical
52
53
processes. Aiche J 1986, 32 (1), 123-138.
54 35. Balakrishna, S.; Biegler, L. T., Targeting strategies for the synthesis and energy integration of
55 nonisothermal reactor networks. Industrial & Engineering Chemistry Research 1992, 31 (9), 2152-
56 2164.
57
58
59
60 27
ACS Paragon Plus Environment
Industrial & Engineering Chemistry Research Page 28 of 29

1
2
3 36. Baliban, R. C.; Elia, J. A.; Floudas, C. A., Simultaneous process synthesis, heat, power, and
4
water integration of thermochemical hybrid biomass, coal, and natural gas facilities. Computers &
5
6 Chemical Engineering 2012, 37, 297-327.
7 37. Pattison, R. C.; Baldea, M., Multistream heat exchangers: Equation‐oriented modeling and
8
flowsheet optimization. Aiche J 2015, 61 (6), 1856-1866.
9
10 38. Dowling, A. W.; Biegler, L. T., A framework for efficient large scale equation-oriented
11 flowsheet optimization. Computers & Chemical Engineering 2015, 72, 3-20.
12
13 39. Kamath, R. S.; Biegler, L. T.; Grossmann, I. E., Modeling multistream heat exchangers with
14 and without phase changes for simultaneous optimization and heat integration. Aiche J 2012, 58 (1),
15 190-204.
16
17 40. Srinivas, B.; El-Halwagi, M., Synthesis of combined heat and reactive mass-exchange
18 networks. Chemical Engineering Science 1994, 49 (13), 2059-2074.
19 41. El‐Halwagi, M. M.; Manousiouthakis, V., Simultaneous synthesis of mass‐exchange and
20
21 regeneration networks. Aiche J 1990, 36 (8), 1209-1219.
22 42. Grossmann, I. E.; Yeomans, H.; Kravanja, Z., A rigorous disjunctive optimization model for
23 simultaneous flowsheet optimization and heat integration. Computers & Chemical Engineering 1998,
24 22, Supplement 1, S157-S164.
25
26 43. Navarro-Amorós, M. A.; Caballero, J. A.; Ruiz-Femenia, R.; Grossmann, I. E., An alternative
27 disjunctive optimization model for heat integration with variable temperatures. Computers &
28 Chemical Engineering 2013, 56, 12-26.
29
30 44. Kong, L.; Avadiappan, V.; Huang, K.; Maravelias, C. T., Simultaneous chemical process
31 synthesis and heat integration with unclassified hot/cold process streams. Computers & Chemical
32 Engineering 2017, 101, 210-225.
33
34 45. Townsend, D. W.; Linnhoff, B., Surface area targets for heat exchanger networks. In IChemE
35 Annl Res. Mtg, 1984.
36 46. Linnhoff, B.; Ahmad, S., Cost optimum heat exchanger networks—1. Minimum energy and
37
38
capital using simple models for capital cost. Computers & Chemical Engineering 1990, 14 (7), 729-
39 750.
40 47. Kemp, I. C., Pinch analysis and process integration: a user guide on process integration for the
41 efficient use of energy. Butterworth-Heinemann: 2011.
42
43 48. Paterson, W. R., A replacement for the logarithmic mean. Chemical Engineering Science 1984,
44 39 (11), 1635-1636.
45
46 49. Kong, L.; Sen, S. M.; Henao, C. A.; Dumesic, J. A.; Maravelias, C. T., A superstructure-based
47 framework for simultaneous process synthesis, heat integration, and utility plant design. Computers
48 & Chemical Engineering 2016, 91, 68-84.
49
50
50. Tawarmalani, M.; Sahinidis, N. V., A polyhedral branch-and-cut approach to global
51 optimization. Math Program 2005, 103 (2), 225-249.
52
53
54
55
56
57
58
59
60 28
ACS Paragon Plus Environment
Page 29 of 29 Industrial & Engineering Chemistry Research

1
2
3
4
5
6
7 Abstract Graphic
8
9 Design goals, constraints Heat transfer Utility
10 & specifications coefficients prices
11
12
13
14 Process Utility and area
15 synthesis targeting
16
17
18
19 Process Operating
Utility usage Area
20 structure conditions
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 29
ACS Paragon Plus Environment

You might also like