Professional Documents
Culture Documents
PII: S0098-1354(17)30296-X
DOI: http://dx.doi.org/10.1016/j.compchemeng.2017.08.004
Reference: CACE 5871
Please cite this article as: Nair, Sajitha K., Soon, Melvin., & Karimi, I.A., Locating
Exchangers in an EIP-wide Heat Integration Network.Computers and Chemical
Engineering http://dx.doi.org/10.1016/j.compchemeng.2017.08.004
This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
Locating Exchangers in an EIP-wide Heat Integration Network
Corresponding author: Email cheiak@nus.edu.sg, Phone +65 6516-6359, Fax +65 6779-1936.
Highlights
Multi-plant multi-enterprise collaborative heat integration in an eco-industrial park
Simultaneous intra-plant and inter-plant direct heat integration
Heat exchangers distributed at plants and centralized sites
Accounting of ambient heat losses/gains and piping, exchanger, and pumping costs
Abstract
Inter-plant heat integration offers an energy-saving opportunity in EIPs beyond the traditional
intra-plant integration. The economic feasibility of this integration depends critically on the
locations of heat exchangers. In this work, we generalize our previous study on centralized
HENS (Heat Exchanger Network Synthesis) for EIPs to allow exchangers to be located at either
plant or central sites facilitating both intra-plant and inter-plant heat integration in a seamless
manner. We propose a mixed integer non-linear programming (MINLP) model that synthesizes
a maximum-NPV (Net Present Value) EIP-wide HEN, while accounting for all the capital (e.g.
heat exchangers, pumps, and pipelines) and operating (pumping costs, utility savings) cash
flows along with the ambient heat gains/losses during transports. The model is tested on five
examples from the literature and gives better HENs compared to the previous results. This work
highlights and quantifies the impact of heat exchanger locations in an EIP-wide HEN.
The world’s population is growing and projected to increase from 6.51 billion in 2005 to 7.79-
10.76 billion in 2050 (Nehring, 2009). Fossil fuels, the main source of energy in the world, are
depleting and causing global warming. These concerns are encouraging global efforts for
energy efficiency and conservation. Heat integration is a well-established technique for energy
conservation in the chemical industry. The heating/cooling demands of streams in a plant can
be synergized to gain economic and environmental benefits. This strategy can also be applied
to a community or complex of closely located plants, called an Eco-Industrial Park (EIP) (Tudor
et al., 2007), to gain environmental, economic, and societal benefits through collaboration
(Manahan, 2004). The collaboration reduces the net consumption of energy, raw materials,
water, and other resources. Some successful EIPs are Kashima Industrial park in Japan, Jurong
Island in Singapore, London Remade eco-industrial sites in the United Kingdom, Ecosite du
(Chen and Lin, 2012; Nair et al., 2016). Hui and Ahmad (1994) developed a nine-step procedure
based on pinch analysis for inter-plant HEN. Instead of using explicit plant-to-plant distances,
they assigned a fixed investment for all plant-to-plant connections. Bagajewicz and Rodera
(2000) used LP (Linear Programming) and MILP (Mixed Integer Linear Programming) models
to integrate four plants, but ignored the critical reality of plant-to-plant distances, transport
costs, and investment for pumps and pipelines. Laukkanen et al. (2012) developed a model for
both direct and indirect heat integration among processes. They assumed that the fixed costs of
the exchangers included the piping costs, which is rather simplistic. Hiete et al. (2012) used
pinch analysis to heat integrate four plants, but did not report any HEN. Cheng et al. (2014)
proposed a game-theoretic strategy where a plant can “trade” a stream with another plant for
heat integration. However, like Bagajewicz and Rodera (2000), Cheng et al. (2014) also ignored
locations, transports, and investments. Wang et al. (2015) proposed a combined direct / indirect
heat integration for two plants. They considered piping costs, but neglected pumps and pumping
costs. Nemet et al. (2016) considered the key issues of pipeline design, heat losses, and pressure
drops, while designing a direct / indirect heat integration network. However, their models were
either too complex or not designed to handle more than two plants.
None of the above interplant heat integration efforts addressed the issue of exchanger
locations. Heat exchanger locations in an EIP-based network are critical, as they affect the
piping and pumping costs and ambient effects. Nair et al. (2016) proposed the idea of locating
all exchangers at a central location and developed an MINLP model for an EIP-wide HEN.
They accounted for the relevant costs of pipes, pumps, exchangers, and pumping, and presented
a shared and rational approach for enterprise collaboration. However, as we show later, forcing
all exchanging streams to gather at a centralized location has economic drawbacks. Chang et
al. (2017) added intervals to the isothermal stagewise superstructure (Yee and Grossmann,
1990) and accounted for exchanger locations, pumping costs, and piping investments. However,
they seem to have unnecessarily complicated the superstructure, as stage numbers can easily be
used as intervals. Furthermore, the assumption of isothermal mixing leads to inferior HEN.
In this work, we propose the idea of a distributed HEN, in which heat exchangers can be
placed at some selected sites or participating plants. By allowing a heat exchanger to be located
at a plant, we can eliminate the movement of streams that already belong to that plant. In
addition, we account for pressure and ambient heat gains/losses. We present a mixed integer
nonlinear programming (MINLP) model that maximizes the NPV for the EIP-wide HEN.
2. Problem Statement
project that can save utility costs by conserving energy. Each enterprise has one or more plants
with streams that are currently heated or cooled internally by some utilities (e.g. steam, cooling
water, ...). The proposed project aims to match the heating/cooling demands of these streams
across the EIP to reduce the total utility usage in a cost-effective manner. It involves
enable hot process streams (those that need cooling) from various EIP plants to heat cold
process streams (those that need heating). Hence, there is a need to (1) identify the most cost-
effective HEN for the EIP, (2) locate its exchangers at optimal places, and (3) estimate the costs
and benefits for the various enterprises that may participate in the project. It is agreed that an
exchanger may be located at either a pre-designated centralized site or any plant in the EIP.
With this, the problem for synthesizing an EIP-wide HEN can be stated as follows.
Given:
be heated.
2) Their mass flows, initial and target temperatures, and allowable temperature limits (lower
3) Their thermophysical properties such as density, heat capacity, viscosity, film heat
𝑃 plants (𝑝 = 1, 2, … , 𝑃) are the actual plants belonging to various enterprises, and the last plant
(𝑝 = 𝑃 + 1) is a dummy plant representing the centralized site. The dummy plant belongs to
all 𝐸 enterprises.
6) Minimum temperature approach and pressure drops for each potential 2-stream heat
exchanger.
9) Estimated life time (N) for the project and annual interest rate.
Obtain:
1) The EIP-wide HEN with participating streams, their temperatures, and flows.
5) Participating enterprises, their CAPEX and OPEX contributions, and IROR (Internal Rate
of return on investment).
Aiming to:
Assuming:
1) Inter-plant flows of streams cannot be partial. A stream may split into one or more
substreams after entering a plant, but the substreams must merge again before leaving that plant.
This is to minimize the cost of transporting streams. Transporting multiple substreams would
increase piping and transport costs, and may pose operational issues.
3) Existing in-plant heaters/coolers for the streams are retained or retrofitted to provide
4) A stream either gains heat from or loses heat to the ambient during its movement from
one plant to another in the EIP, and its gain/loss behavior remains unchanged through its entire
movement.
5) Existing in-plant movers (pumps or compressors) for the streams are retained or
retrofitted to transport streams internally. New pumps are installed for transporting streams to
other plants. Only one new pump is used for such transport.
6) All matches between hot and cold streams are allowable, if thermodynamically feasible.
9) Inflation rate is zero, so utility and transportation costs remain constant throughout the 𝑁
years.
3. Model Formulation
We assume a superstructure (Huang and Karimi, 2013; Yee and Grossmann, 1990) with 𝐾 + 2
pass through stages 1 through 𝐾 + 1 before exiting the HEN. In contrast, the cold streams pass
through stages 𝐾 through 0 before exiting the HEN. However, a stream may bypass one or more
stages from 1 to 𝐾 entirely, if needed. If a hot (cold) stream enters stage k (1 ≤ 𝑘 ≤ 𝐾), then it
splits into 𝐽 (𝐼) substreams to exchange heat with 𝐽 (𝐼) cold (hot) substreams in 2-stream
heat, the substreams merge back to remake the parent stream before exiting the stage. Each hot
(cold) stream enters stage 𝐾 + 1 (0) to be cooled (heated) with utilities to meet its target
temperature.
streams may not need all the 𝐾 stages. Therefore, we allow a stream to bypass a stage (1 ≤ 𝑘 ≤
Clearly, a stage can exist, only if at least one stream enters that stage. We force the redundant
stages to be the end ones by defining the following 0-1 continuous variable.
1 if stage 𝑘 exists
𝑌𝑘 = { 1≤𝑘≤𝐾
0 otherwise
a part or whole of hot stream 𝑖 heats a part or whole of cold stream 𝑗 in stage 𝑘. We define a 0-
1 if 𝐻𝐸𝑖𝑗𝑘 exists
𝑥𝑖𝑗𝑘 = { 1 ≤ 𝑖 ≤ 𝐼; 1 ≤ 𝑗 ≤ 𝐽; 1 ≤ 𝑘 ≤ 𝐾
0 otherwise
It is possible that two stages have identical stream matches, i.e. the same combinations of hot
and cold substreams exchange heat. In this case, we can merge the two stages into a single stage
by demanding that a stream can exchange heat with the same stream in two consecutive stages
only if it also exchanges heat with another substream in one of the stages.
𝑥𝑖𝑗𝑘 + 𝑥𝑖𝑗(𝑘+1) − ∑𝐽𝑗′ =1,𝑗 ′ ≠𝑗 𝑥𝑖𝑗 ′ 𝑘 − ∑𝐽𝑗′ =1,𝑗 ′ ≠𝑗 𝑥𝑖𝑗 ′ (𝑘+1) − ∑𝐼𝑖′ =1,𝑖 ′ ≠𝑖 𝑥𝑖 ′ 𝑗𝑘 −
respectively be the fractional flows of streams 𝑖 and 𝑗 entering 𝐻𝐸𝑖𝑗𝑘 . For a stream to flow
through an exchanger, the exchanger must exist, and the stream must enter the stage.
Furthermore, if a stream enters the stage, the fractional flows must sum to one. Therefore,
Exchanger Locations
Locating exchangers at right plants in an EIP is crucial, as the key costs of pumping and
piping depend heavily on the distances that the heat-exchanging streams must travel. An
exchanger can only be placed in a plant that accepts to host it. Hence, we define,
Eq. 5 fixes 𝑥𝑖𝑗𝑘 to be binary. Now, we identify the location of each stream at each stage using
1 ≤ 𝑠 ≤ 𝑆; 0 ≤ 𝑘 ≤ 𝐾 + 1; 𝑝 ∈ 𝑃𝑠
Because a stream must begin from and return to its own plant, we set 𝑧𝑠(𝐾+1)𝑝𝑠 = 𝑧𝑠0𝑝𝑠 = 1,
and 𝑧𝑠(𝐾+1)𝑝 = 𝑧𝑠0𝑝 = 0 for 𝑝 ∈ 𝑃𝑠 and 𝑝 ≠ 𝑝𝑠 . Furthermore, we demand that stream 𝑠 must
be at one plant in each stage 𝑘, even if it skips (𝑦𝑠𝑘 = 0) stage 𝑘 or does not even have it.
Furthermore, a stream must also be at the same plant where its exchanger is.
Having located the streams at each stage, we now track their movements through the HEN by:
1 ≤ 𝑖 ≤ 𝐼; 1 ≤ 𝑘 ≤ 𝐾 + 1; 𝑝 ∈ 𝑃𝑖 , 𝑝′ ∈ 𝑃𝑖
1 ≤ 𝑗 ≤ 𝐽; 0 ≤ 𝑘 ≤ 𝐾; 𝑝 ∈ 𝑃𝑗 , 𝑝′ ∈ 𝑃𝑗
We linearize 𝑍𝑠𝑘𝑝′𝑝 as follows,
If a stream does not enter stage 𝑘, then we keep it at its current location to minimize unnecessary
transport. Hence,
Let
𝑘 ≤ 𝐾)
𝑘 ≤ 𝐾)
Let stream 𝑠 enter stage 𝑘 at temperature 𝑇𝑠𝑘 and leave stage 𝑘 at 𝑇𝐸𝑠𝑘 . Thus, its
temperature may change, as it moves from one stage to another due to ambient heat losses/gains.
Assuming that the temperature change is linearly proportional to the distance travelled by 𝑠, we
obtain,
𝑇𝑗𝐾 − 𝑇𝐼𝑁𝑗 = ∆𝑇𝑗 ∑𝑝∈𝑃𝑗 ∑𝑝′ ∈𝑃𝑗 (𝑍𝑗𝐾𝑝′ 𝑝 𝑑𝑑𝑝′ 𝑝 ) 1≤𝑗≤𝐽 (11b)
where, ∆𝑇𝑠 is the average rise/drop in the temperature of stream 𝑠 per unit distance and
Nair et al. (2016) allowed two utility exchangers for each stream at stages 0 and (𝐾 + 1). In
contrast, we allow only one utility exchanger to be placed at the HEN exit. Hence, 𝑄𝑖(𝐾+1)
represents the utility cooling of stream 𝑖, and 𝑄𝑗0 represents the utility heating of stream 𝑗. Also,
1≤𝑖≤𝐼 (14a)
1≤𝑗≤𝐽 (14b)
Then, the heat duty and area of exchanger 𝐻𝐸𝑖𝑗𝑘 are given by,
𝐴𝑈𝑖𝑗 (1 ≤ 𝑖 ≤ 𝐼; 1 ≤ 𝑗 ≤ 𝐽) are the lower and upper bounds of 𝐴𝑖𝑗𝑘 . Appendix A gives the
remaining equations for the temperature changes and approaches from Nair et al. (2016).
We now compute the capital and operating expenses (CAPEX and OPEX) of HEN to obtain its
NPV. 𝐶𝐴𝑃𝐸𝑋 is the total capital cost invested at time zero. It includes the cost of purchasing
and installing the pumps, heat exchangers, and pipelines. As the enterprises retain their existing
heaters/coolers and pumps for each stream, we do not include their costs. For the new pumps
required for transporting streams out of their parent plants, we use the following 0-1 continuous
variable.
1 − 𝑧𝑠𝑘𝑝𝑠 ≤ 𝑁𝑃𝑠 ≤ 𝐾 − ∑𝐾
𝑘 ′ =1 𝑧𝑠𝑘 ′ 𝑝𝑠 1 ≤ 𝑠 ≤ 𝑆; 1 ≤ 𝑘 ≤ 𝐾 (18)
Then, we have,
𝐽
𝐶𝐴𝑃𝐸𝑋 = ∑𝑆𝑠=1[𝐶𝑃𝐿𝑠 𝐿𝑃𝐿𝑠 + 𝑀𝑃𝑠 𝐶𝑃𝑠 ] + ∑𝐾 𝐼
𝑘=1 ∑𝑖=1 ∑𝑗=1 𝑀𝐻𝑖𝑗 𝐻𝐸𝐶𝑖𝑗𝑘 (19a)
𝛽𝑖𝑗 𝛽𝑖𝑗
𝐻𝐸𝐶𝑖𝑗𝑘 = 𝐹𝐶𝐻𝑖𝑗 𝑥𝑖𝑗𝑘 + 𝑉𝐶𝐻𝑖𝑗 (𝐴𝑖𝑗𝑘 ) − 𝑉𝐶𝐻𝑖𝑗 (𝐴𝐿𝑖𝑗 ) (1 − 𝑥𝑖𝑗𝑘 )
1 ≤ 𝑖 ≤ 𝐼; 1 ≤ 𝑗 ≤ 𝐽; 1 ≤ 𝑘 ≤ 𝐾 (19b)
𝐿𝑃𝐿𝑖 = ∑𝐾+1
𝑘=1 ∑𝑝∈𝑃𝑖 ∑𝑝′ ∈𝑃𝑖 (𝑍𝑖𝑘𝑝′ 𝑝 𝑑𝑑𝑝′ 𝑝 ) 1≤𝑖≤𝐼 (19c)
𝐿𝑃𝐿𝑗 = ∑𝐾
𝑘=0 ∑𝑝∈𝑃𝑗 ∑𝑝′ ∈𝑃𝑗(𝑍𝑗𝑘𝑝′ 𝑝 𝑑𝑑𝑝′ 𝑝 ) 1≤𝑗≤𝐽 (19d)
where,
𝐶𝑃𝑠 = Purchase cost of a pump for stream 𝑠 with 𝐹𝐶𝑃𝑠 as the fixed component, 𝑉𝐶𝑃𝑠 as the
𝐻𝐸𝐶𝑖𝑗𝑘 = Purchase cost of heat exchanger 𝐻𝐸𝑖𝑗𝑘 with 𝐹𝐶𝐻𝑖𝑗 as the fixed component, 𝑉𝐶𝐻𝑖𝑗 as
For costing new pumps, we compute the total pump power for each stream as follows.
where, ∆𝑝𝑠𝐻𝐸 is the pressure drop for stream 𝑠 in one heat exchanger, ∆𝑝𝑠𝐿 is the pressure drop
for stream 𝑠 per unit length of its pipeline, 𝑉𝐹𝑠 is the volumetric flow rate (𝑚3 /𝑎) of stream 𝑠,
and 𝜂𝑠 is the pump efficiency. Then, the purchase cost of a new pump for stream 𝑠 is,
𝐶𝑃𝑠 ≥ 𝐹𝐶𝑃𝑠 𝑁𝑃𝑠 + 𝑉𝐶𝑃𝑠 (𝐸𝑠 )𝛼𝑠 − 𝑉𝐶𝑃𝑠 (𝐾 ∙ 𝑉𝐹𝑠 ∙ ∆𝑝𝑠𝐻𝐸 /𝜂𝑠 )𝛼𝑠 (1 − 𝑁𝑃𝑠 )
1≤𝑠≤𝑆 (21)
When stream 𝑠 needs a new pump (𝑁𝑃𝑠 = 1), then eq. 21 will give its cost. If it does not, then
eq. 21 will give a zero cost, as stream 𝑠 will use its own in-plant pump.
The OPEX for the HEN is the annual cost of moving streams, which is 𝐶𝑂𝐸 ∑𝑆𝑠=1 𝐸𝑠 ,
where 𝐶𝑂𝐸 is the cost of electricity ($/kW). The use of the HEN may save utility costs annually
for the EIP. We compute the NPV for these annual OPEX and utility savings by using a factor
𝛾 as done by Nair et al. (2016). Then, the objective for the EIP-wide HEN project is to maximize
NPV.
(1+𝑟)𝑁 −1
𝑀𝑎𝑥 𝑁𝑃𝑉 = {∑𝐽𝑗=1[𝑈𝐶𝑗 (𝑄𝑗𝑈 − 𝑄𝑗0 )] + ∑𝐼𝑖=1[𝑈𝐶𝑖 (𝑄𝑖𝑈 − 𝑄𝑖(𝐾+1) )] −
𝑟(1+𝑟)𝑁
where, 𝑈𝐶𝑠 (1 ≤ 𝑠 ≤ 𝑆) is the unit utility cost ($/kW-a) for stream 𝑠, 𝑟 is the rate of interest,
This completes our MINLP model for the synthesis of EIP-wide HEN. The model
comprises eqs. 1-22 except eq. 19a-b that are eliminated by substitution into eq. 22. We now
5. Examples
We take five examples with only one plant for each enterprise, hence plant and enterprise
are interchangeable labels. We assume that for a plant to host an exchanger, the plant or its
enterprise must own at least one of the two exchanging streams. Table 1 lists all the data; Table
2 gives the cost correlations, pressure drops, stream densities, and pipe sizes; and Table 3 lists
the various model parameters such as variable bounds. For examples with no plant-to-plant
distance data, we assume the distances as in Table 1. For all examples, we used SCIP 3.1
(Achterberg, 2009) in GAMS 24.4.6 (2015) as the MINLP solver. GAMS/SCIP uses CPLEX
as the LP solver, COIN-OR Interior Point Optimizer IPOPT (Wächter and Biegler, 2006) as the
5.1. Example 1
This example is from Nair et al. (2016) who adopted it from Björk and Westerlund (2002). It
has three enterprises or plants: E1, E2, and E3. E1 owns two hot streams (H1 and H2), E2 owns
three cold streams (C1, C2, and C3) and E3 owns one cold stream (C4). We consider three
variations for this example with 𝐾 = 3 and 𝐾 = 4. Case 1 is the original example from Nair et
al. (2016) with no ambient heat effects. Case 2 considers the ambient heat effects. In Case 3,
we move E2 away from E1 and E3 to see the impact on HEN. For all cases except one, we
achieve an optimality gap of 0.1 % for both 𝐾 = 3 and 𝐾 = 4, and compare the results of our
distributed HENs with the centralized HEN reported by Nair et al. (2016) in Table 4. The final
HEN and results are the same for both 𝐾 = 3 and 𝐾 = 4. However, the computation effort is
Case 1
This case ignores the ambient heat gains/losses. Figure 2 shows the best distributed HEN.
Circled numbers denote the plants where streams and exchangers are located in various stages.
Only H1 and H2 flow from E1 to other plants for heat exchange. C1-C4 remain in their parent
plants. H1 bypasses stage 1, moves from E1 to E3 in stage 2, moves to E2 in stage 3, and returns
to E1 to bypass stage 4. H2 moves from E1 to E2 in stage 1, where it splits into two substreams.
The distributed HEN has NPV = $2.39 million and IROR = 161.5 %. In contrast, Nair et
al. (2016) reported NPV = $1.84 million and IROR = 78.8 % for their centralized HEN. Thus,
by optimally locating the exchangers, the distributed HEN increases NPV by 30 % and IROR
by nearly 100 %. Only two streams flow out of their plants in the distributed HEN versus six in
the centralized HEN. The lower transport needs reduce CAPEX by 45.9 %, and increase NPV
and IROR. However, the distributed HEN saves less energy (i.e. utility heating and cooling)
than the centralized HEN. As a result, the savings in the utility costs are slightly lower ($1.35
The centralized HEN involved five exchanges: H1-C2, H1-C3, H2-C1, H2-C2, and H2-
C4. In contrast, the distributed HEN involves five exchanges: H1-C1, H1-C4, H2-C1, H2-C2,
and H2-C3. Clearly, the exchanges have changed due to the economics of transporting streams.
The restricted availability of streams at each plant has limited exchanges and reduced energy
savings. In the centralized HEN, H2 from E1 could heat in parallel both C1 (E2) and C4 (E3)
that belong to different enterprises. In the distributed HEN, this is not economical, as both C1
and C4 must travel from E2 and E3 to E1, increasing piping and pumping costs. The distributed
HEN saved 58 % in piping costs, and found alternate exchanges. While this reduced energy
savings slightly, it increased NPV and IROR drastically. This shows the importance of trading
off CAPEX versus OPEX, and true costs rather that simple energy savings to give a better HEN.
Case 2
(Engineering Toolbox) as the overall heat transfer coefficient of all pipelines. In this case, all
streams lose heat to the ambient. This represents a loss for both hot and cold streams, as hot
streams lose the heating energy, and cold streams need more cooling energy. These ambient
heat losses reduce the energy savings as compared to Case 1. The utility cost savings reduce
from $1.35 million/a to $1.14 million/a. The best HEN (Figure 3) is also very different from
Case 1. NPV is lower ($2.02 million versus $2.3 million), but IROR remains nearly the same.
In contrast to Case 1, H2 and C3 do not need their utility exchangers, but all other streams need
them. In contrast to Case 1, the heat losses in Case 2 reduced the incentive for heat exchange
between H1 and C1. As a result, the best HEN for this case does not send H1 to E2. This reduced
the pipeline for H1, and overall CAPEX to $0.65 million. Also, the exchange duties for H1-C4
and H2-C2 are also reduced because of the ambient heat losses compared to Case 1. Thus, it is
Case 3
In this case, we move E2 away from both E1 and E3, and ignore the ambient heat gains/losses.
Thus, the distances of E2 are now 3 km versus 0.61 km from E1, and 3.06 km versus 0.70 km
from E3. Since E2 owns three of the four cold streams, we expect this change to impact the
distributed HEN. Figure 4 shows the best distributed HEN for this case. Even though E2 is
farther away from E1 and E3, the best HEN still sends H2 to E2, and it does not send H1 to E2
as in Case 1. As a result, the duty of the utility cooler for H1 increases. Since H2 flows to E2,
the heat from H2 is fully utilized for the cold streams of E2, and H2 does not need its own utility
cooler. However, as the energy from H2 is still not sufficient for C1, C2, and C4; they need
their in-plant utility heaters. These differences lower the utility cost savings from $1.35
million/a to $1.27 million/a. Moving H2 to E2 increases piping costs and CAPEX considerably.
CAPEX is 98 % higher than Case 1. This reduces IROR and NPV to 59 % and $1.36 million
respectively.
5.2. Example 2
Consider Example 2b from Nair et al. (2016). It has 3 enterprises with 4 hot and 5 cold streams.
We solve this example for 𝐾 = 3 only, as computational effort is excessive for 𝐾 = 4. Figure
5 shows the best distributed HEN for this example with an optimality gap of 4.8 %. It involves
6 exchanges: H1-C1, H2-C3, H2-C4, H3-C1, H3-C2 and H4-C5. In contrast, the centralized
HEN involved only 3 exchanges; H2-C1, H3-C1, and H3-C2. A major difference between the
two sets of exchanges is the fact that the exchanges in the distributed HEN involve both intra-
plant and inter-plant heat integration. The dotted lines indicate the intra-plant heat integration.
In this example, the availability of both hot and cold streams at each plant allows the possibility
of intra-plant exchanges, which are well exploited by the distributed HEN model to reduce the
piping and pump costs. This illustrates the ability of our model to naturally address both intra-
plant and inter-plant integrations. The centralized HEN of Nair et al. (2016) for this example
did not use H1, H4, C3, C4, and C5 for integration, because transporting them to the central
site was not economical compared to the potential utility savings. In contrast, the distributed
HEN uses all of them for integration. This leads to a greater energy savings for the distributed
HEN and utility cost savings ($0.190 million/a vs $0.158 million/a), The distributed HEN has
49.7 % lower CAPEX, as no streams are transported. IROR increases from 14.2 % to 54.5 %
and NPV increases from $0.09 million to $0.587 million. These results show the significant
advantage of using the distributed versus centralized HEN model. Table 5 compares the full
This example was also addressed using a game-theory based optimization strategy for
inter-plant heat integration by Cheng et al. (2014). However, no meaningful comparison can be
5.3. Example 3
This has four plants, six hot streams, and six cold streams. Again, we could solve this example
for at most 𝐾 = 3. We compare our distributed HEN with the solution reported by Hiete et al.
(2012) using thermal pinch analysis. While we use their plant-to-plant distances and cost
correlations, a fair and direct comparison is not possible due to the following differences:
a) Some streams could use two different utility costs in Hiete et al. (2012), while we fixed
utility cost for each stream. Besides, Hiete et al. (2012) did not justify their utility costs.
b) Their heat exchanger costs differ with exchanger type, while we fixed our heat exchanger
c) Their piping costs are functions of volumetric flowrate; but they provided mass flowrates
without giving stream densities. Therefore, we assumed a density of 900 kg/m3 for all streams.
d) In their model, plants hosted the heat exchangers, but it was not clear how they selected
plant locations. Their solutions did not report the order and locations of exchangers.
e) They gave no data on heat transfer coefficients and minimum temperature approach.
Table 6 summarizes the results for our best distributed HEN with an optimality gap of 5.2
%. Figure 6 shows the plants with both hot and cold streams exploited intra-plant heat
exchanges. Only four streams (C3, C4, C5 and C6) were sent away from their owner plants. C3
stage 3 and returned to E2 in stage 2. C5 was sent from E3 to E1 in stage 3, and returned to E3
in stage 0. Finally, C6 was sent from E4 to E1 in stage 3, and returned to E4 in stage 1. The
distributed HEN involves 10 exchanges: H1-C2, H2-C2, H2-C5, H2-C6, H3-C1, H3-C2, H3-
C4, H5-C3, H5-C4, and H6-C6. Of these, H1-C2, H2-C2, H3-C1, H3-C2, H3-C2, H5-C4, and
H6-C6 are intra-plant exchanges. In other words, our proposed model optimizes both intra-plant
and inter-plant exchanges simultaneously. Zero fixed costs of exchangers in this example leads
to many exchangers in HEN. Despite using lower hot (25.7 MW) and cold (27.1 MW) utilities,
the distributed HEN has a higher utility cost ($3.676 million/a vs $3.406 million/a) than Hiete
et al. (2012). This is due to a fixed utility cost for each stream, as Hiete et al. (2012) allowed
two utility costs for some streams. The amortized capital cost ($0.395 million/a) from the
distributed HEN model is significantly lower than Hiete et al. (2012) ($0.773 million/a). The
missing stream densities could cause this difference; resulting in different volumetric flowrates
and stream piping costs. The distributed HEN has NPV = $8.9 million and IROR = 113 %.
5.4. Example 4
Wang et al. (2015) used this example. It has two plants with E1 having four hot streams and E2
having seven cold streams. While they illustrated their model for direct, indirect, and combined
heat integration, we consider only direct heat integration for the sake of a fair comparison.
However, comparing the two solutions (ours and theirs) was not easy because of the following
differences. First, their model applies to two plants only. Second, they ignored the costs of
pumps and pumping. Third, they reported the pipe costs for their HEN, but did not report the
heat exchanger locations. Fourth, they did not report crucial stream properties such as specific
heat capacities and densities for their direct integration example. Last, their model assumed
isothermal mixing. Therefore, we set pump and pumping costs to zero for this example, and
take the stream properties from their example on indirect heat integration.
The computational effort again limited us to 𝐾 = 3 for this exercise. Table 7 compares
our HEN (optimality gap = 3.6 %) with theirs. Our best HEN in Figure 7 shows four streams
being transported. Wang et al. (2015) also had transported four streams. However, we cannot
know which four streams, as heat exchanger locations are unknown. Our HEN has 10
exchanges: H1-C1, H1-C2, H1-C3, H1-C4, H1-C5, H2-C5, H2-C6, H3-C6, H3-C7, and H4-
C7. Of these, H1-C5, H2-C6, and H3-C7 do not exist in Wang et al. (2015). In both HENs, the
cold streams fully meet hot stream duties. Both HENs achieve the same maximum energy
savings and utility costs ($0.228 million/a). However, our HEN is better, as it needs less heat
exchange area and cost (1667 m2 vs. 2069 m2 and $0.280 million/a vs. $0.334 million/a). The
difference is mainly due to the isothermal mixing assumption. Although both HENs have four
transported streams, our pipe costs are lower ($0.371 vs $0.632 million/a) probably due to the
missing stream properties like heat capacities and densities. Overall, our HEN has a lower TAC
Example 5
Chang et al. (2017) proposed to use Generalized Disjunctive Programming (GDP) for inter-
plant HENS. In addition to stages, they introduced “intervals” to locate streams at different
plants for heat exchange. They minimized TAC rather than maximize NPV, so we do the same
for a fair comparison. While we used their parameters, costs, and pressure drop correlations,
we encountered certain differences. First, eqs. 56 and 57 for pressure drops in Chang et al.
(2017) are incorrect, so we used the equations from the original source of Soltani and Shafiei
(2011). Second, the exchanger fixed cost is stated as $1,100 instead of $11,000. Third, our
calculations for their network give higher piping costs than those reported by them. Fourth, they
included the costs of in-plant utility exchangers as well, so we also did the same for a valid
We solved this example for both 𝐾 = 3 and 𝐾 = 4. Figure 8a and Table 8 show our
optimal HEN with a gap of 1 % for 𝐾 = 3. It involves 7 exchanges: H1-C1, H1-C3, H2-C2,
H2-C3, H2-C5, H4-C1, and H4-C4. Among these, H1-C3, H2-C5 and H4-C1 are inter-plant
heat exchanges. In Chang et al. (2017), exchanges H1-C3 and H4-C1 are not present. Also, H1
and C5 are transported in Chang et al. (2017), whereas H1, H2, and H4 in this work. Thus, our
HEN has a higher piping cost ($0.083 million/a Vs $0.054 million/a). However, allowing non-
isothermal mixing gives 41.9 % lower utility costs ($0.96 million/a vs $1.66 million/a) and 33.4
For 𝐾 = 4 (see Table 8), we obtain the network in Figure 8b. This has 0.5 % lower TAC
than 𝐾 = 3. However, it takes 200,000 s of CPU time (versus 61,186 s for 𝐾 = 3) to reach a
relative optimality gap of 21.9 % (versus 1 % for 𝐾 = 3). The primary reason for the greater
optimality gap seems to be the significantly poorer relaxation resulting in a much worse lower
6. Conclusion
The costs of piping and pumping streams can be a deal-breaker for an EIP-wide HEN, as they
constitute major capital and operating expenses. We presented an MINLP model for
synthesizing an HEN where the heat exchangers can be located at any plant or central sites.
Thus, we generalized our previous work (Nair et al, 2016) on centralized EIP-wide HENS. In
addition, we included the effects of ambient heat gains/losses, pressure drops during stream
transports between locations, and a seamless treatment of both inter-plant and intra-plant heat
exchanges. Our numerical evaluation on five literature examples shows that our distributed
HEN model correctly trades off between energy and transport costs. The idea of distributed
HEN reduces stream movements and piping/pumping costs (both CAPEX and OPEX)
significantly. For the first two examples that allowed a central site, no exchanger was placed at
that site in the best HENs. Our model gives better results for all the five examples. However,
reaching a low optimality gap was a challenge for most examples. Thus, a method for efficient
Acknowledgement
Sajitha K Nair acknowledges financial support under the President’s Graduate Fellowship from
the National University of Singapore. The work was also funded in part by the National
Nair et al. (2016) derived/used the following for temperature changes and approaches. For
𝑄𝑖𝑘
𝑇𝐸𝑖𝑘 = 𝑇𝑖𝑘 − 1 ≤ 𝑖 ≤ 𝐼; 1 ≤ 𝑘 ≤ 𝐾 (A.1)
𝐹𝑖
𝑄𝑗𝑘
𝑇𝐸𝑗𝑘 = 𝑇𝑗𝑘 + 1 ≤ 𝑗 ≤ 𝐽; 1 ≤ 𝑘 ≤ 𝐾 (A.2)
𝐺𝑗
𝑄𝑖(𝐾+1)
𝑇𝑂𝑈𝑇𝑖 = 𝑇𝑖(𝐾+1) − 1≤𝑖≤𝐼 (A.3)
𝐹𝑖
𝑄𝑗0
𝑇𝑂𝑈𝑇𝑗 = 𝑇𝑗0 + 1≤𝑗≤𝐽 (A.4)
𝐺𝑗
1 ≤ 𝑖 ≤ 𝐼; 1 ≤ 𝑗 ≤ 𝐽 (A.5)
𝑈
𝑅𝑖𝑗 = 𝑚𝑖𝑛[𝑇𝑗𝑈 − 𝑇𝐼𝑁𝑗 , 𝑚𝑎𝑥{𝑅𝑗𝐿 , 𝑇𝐼𝑁𝑖 − 𝑇𝐼𝑁𝑗 − 𝑀𝑇𝐴𝑖𝑗 }]
1 ≤ 𝑖 ≤ 𝐼; 1 ≤ 𝑗 ≤ 𝐽 (A.6)
𝑀𝑇𝐴𝑖𝑗 ≤ 𝐻𝑇𝐴𝑖𝑗𝑘 ≤ 𝑇𝑖𝑘 − 𝑇𝑗𝑘 − 𝑅𝑖𝑗𝑘 + (𝑀𝑇𝐴𝑖𝑗 + |𝑇𝑂𝑈𝑇𝑖 − 𝑇𝑂𝑈𝑇𝑗 |)(1 − 𝑥𝑖𝑗𝑘 )
1 ≤ 𝑖 ≤ 𝐼; 1 ≤ 𝑗 ≤ 𝐽; 1 ≤ 𝑘 ≤ 𝐾 (A.11)
𝑀𝑇𝐴𝑖𝑗 ≤ 𝐶𝑇𝐴𝑖𝑗𝑘 ≤ 𝑇𝑖𝑘 − 𝑇𝑗𝑘 − 𝐷𝑖𝑗𝑘 + (𝑀𝑇𝐴𝑖𝑗 + |𝑇𝑂𝑈𝑇𝑖 − 𝑇𝑂𝑈𝑇𝑗 |)(1 − 𝑥𝑖𝑗𝑘 )
1 ≤ 𝑖 ≤ 𝐼; 1 ≤ 𝑗 ≤ 𝐽; 1 ≤ 𝑘 ≤ 𝐾 (A.12)
1 ≤ 𝑖 ≤ 𝐼; 1 ≤ 𝑗 ≤ 𝐽; 1 ≤ 𝑘 ≤ 𝐾 (A.13)
Subscripts
𝑒 Enterprise in an EIP
𝑝 Plant
𝑠 Process stream
Superscripts
𝐿 Lower limit
𝑈 Upper Limit
Parameters
𝐸 Number of enterprises
𝑉𝐶𝐻 Coefficient in the variable component of the purchase cost of a heat exchanger
η Pump efficiency
𝜌 Density (kg/m3)
Continuous Variables
Binary variables
𝑌 Existence of stage
𝑍 Stream movement
Acronyms
HE Heat Exchanger
HEN Heat Exchanger Network
Achterberg, T. (2009). Scip: Solving Constraint Integer Programs. Mathematical Programming Computation, 1, 1-
41.
Bagajewicz, M., & Rodera, H. (2000). Energy Savings in the Total Site Heat Integration across Many Plants.
Computers & chemical engineering, 24, 1237-1242.
Björk, K.-M., & Westerlund, T. (2002). Global Optimization of Heat Exchanger Network Synthesis Problems with
and without the Isothermal Mixing Assumption. Computers & chemical engineering, 26, 1581-1593.
Chang, C., Chen, X., Wang, Y., & Feng, X. (2017). Simultaneous Synthesis of Multi-Plant Heat Exchanger Networks
Using Process Streams across Plants. Computers and Chemical Engineering, 101, 95-109.
Chen, C.-L., & Lin, C.-Y. (2012). Design of Entire Energy System for Chemical Plants. Industrial & Engineering
Chemistry Research, 51, 9980-9996.
Cheng, S.-L., Chang, C.-T., & Jiang, D. (2014). A Game-Theory Based Optimization Strategy to Configure Inter-
Plant Heat Integration Schemes. Chemical Engineering Science, 118, 60-73.
Gams Development Corporation, General Algebraic Modeling System (Gams) Release 24.4.6. (2015).
Washington, DC, USA, .
Gibbs, D., Deutz, P., & Proctor, A. (2005). Industrial Ecology and Eco‐Industrial Development: A Potential
Paradigm for Local and Regional Development? Regional Studies, 39, 171-183.
Hiete, M., Ludwig, J., & Schultmann, F. (2012). Intercompany Energy Integration. Journal of industrial Ecology,
16, 689-698.
Huang, K. F., Al-mutairi, E. M., & Karimi, I. (2012). Heat Exchanger Network Synthesis Using a Stagewise
Superstructure with Non-Isothermal Mixing. Chemical Engineering Science, 73, 30-43.
Huang, K. F., & Karimi, I. A. (2013). Simultaneous Synthesis Approaches for Cost-Effective Heat Exchanger
Networks. Chemical Engineering Science, 98, 231-245.
Hui, C., & Ahmad, S. (1994). Minimum Cost Heat Recovery between Separate Plant Regions. Computers &
chemical engineering, 18, 711-728.
Laukkanen, T., Tveit, T.-M., & Fogelholm, C.-J. (2012). Simultaneous Heat Exchanger Network Synthesis for Direct
and Indirect Heat Transfer inside and between Processes. Chemical Engineering Research and Design,
90, 1129-1140.
Manahan, S. E. (2004). Environmental Chemistry, Eighth Edition. Crc Press, Boca Raton.
Matsuda, K., Hirochi, Y., Tatsumi, H., & Shire, T. (2009). Applying Heat Integration Total Site Based Pinch
Technology to a Large Industrial Area in Japan to Further Improve Performance of Highly Efficient
Process Plants. Energy, 34, 1687-1692.
Nair, S. K., Guo, Y., Mukherjee, U., Karimi, I., & Elkamel, A. (2016). Shared and Practical Approach to Conserve
Utilities in Eco-Industrial Parks. Computers & chemical engineering, 93, 221-233.
Nehring, R. (2009). Traversing the Mountaintop: World Fossil Fuel Production to 2050. Philosophical
Transactions: Biological Sciences, 364, 3067-3079.
Nemet, A., Čuček, L., & Kravanja, Z. (2016). Procedure for the Simultaneous Synthesis of Heat Exchanger
Networks at Process and Total Site Level Chemical Engineering Transactions, 52, 1057-1062.
Overall Heat Transfer Coefficient. Engineering ToolBox. Accessed on 09/04/ 2017,
http://www.engineeringtoolbox.com/overall-heat-transfer-coefficient-d_434.html.
Soltani, H., & Shafiei, S. (2011). Heat Exchanger Networks Retrofit with Considering Pressure Drop by Coupling
Genetic Algorithm with Lp (Linear Programming) and Ilp (Integer Linear Programming) Methods. Energy,
36, 2381-2391.
Tudor, T., Adam, E., & Bates, M. (2007). Drivers and Limitations for the Successful Development and Functioning
of Eips (Eco-Industrial Parks): A Literature Review. Ecological Economics, 61, 199-207.
Wächter, A., & Biegler, L. T. (2006). On the Implementation of an Interior-Point Filter Line-Search Algorithm for
Large-Scale Nonlinear Programming. Mathematical Programming, 106, 25-57.
Wang, Y., Chang, C., & Feng, X. (2015). A Systematic Framework for Multi-Plants Heat Integration Combining
Direct and Indirect Heat Integration Methods. Energy, 90, 56-67.
Yee, T. F., & Grossmann, I. E. (1990). Simultaneous Optimization Models for Heat Integration—Ii. Heat Exchanger
Network Synthesis. Computers & chemical engineering, 14, 1165-1184.
List of Tables
Table 1 Stream temperatures, heat contents, utility costs, and plant-to-plant distances for
Examples 1-5.
Table 2 Stream densities, pressure drops, pipe/pump sizing data, and cost data for Examples 1-
5.
Table 6 Comparison of our HEN with Hiete et al. (2012) for Example 3
Table 7 Comparison of our HEN with Wang et al. (2015) for Example 4
Table 8 Comparison of our HEN with Chang et al. (2017) for Example 5
List of Figures
Table 6. Comparison of our HEN with Hiete et al. (2012) for Example 3
Performance Metric Distributed HEN Hiete et al. (2012)
IROR 112.9 % -
NPV ($) 8,897,517 -
Total Utility Savings ($/a) 1,844,015 -
CAPEX ($) 1,497,138 -
OPEX ($/a) 152,342 -
Utility cost ($/a) 3,676,342 3,406,000
Hot utility demand (kW) 25,678 31,319
Cold utility demand (kW) 27,125 32,766
Amortized capital and operating cost ($/a) 395,993 773,000
Total annual cost ($/a) 4,072,334 4,180,000
Relative Gap (%) 5.2 -
Table 7. Comparison of our HEN with Wang et al. (2015) for Example 4
Performance Metric Distributed HEN Wang et al. (2015)
Hot utility demand (kW) 1,200 1,200
Cold utility demand (kW) 0 0
Energy savings (kW) 22,200 22,200
Additional heat exchangers 10 8
Heat exchanger area (m2) 1,667 2,069
Heat exchanger cost ($/a) 280,005 334, 465
Pipeline cost ($/a) 370,649 632,220
Cost of utilities ($/a) 228,000 228,000
Total annual Cost ($/a) 878,653 1,194,686
Relative Gap (%) 3.6 -
Table 8. Comparison of our HEN with Chang et al. (2017) for Example 5
Performance Metric Distributed Distributed Chang et al.
HEN (𝑲 = 𝟑) HEN (𝑲 = 𝟒) (2017)
Hot utility demand (kW) 2,766 2,760 6,250
Cold utility demand (kW) 53,616 53,610 57,100
Total number of heat exchangers 15 17 14
Total number of inter-plant
3 4 4
matches
Total annualized heat exchanger
377,304 327,162 460,404
cost ($/a)
Pipeline cost ($/a) 83,032 76,660 54,238
Pumping cost ($/a) 60,676 109,803 51,789
Utilities cost ($/a) 962,729 962,137 1,661,000
Total annual Cost ($/a) 1,483,741 1,475,762 2,227,431
Relative Gap (%) 1 21.9 -
CPU time (s) 61,186 200,001
Figure 1. Schematic of HEN superstructure