Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

16th European Symposium on Computer Aided Process Engineering

and 9th International Symposium on Process Systems Engineering


W. Marquardt, C. Pantelides (Editors) 725
© 2006 Published by Elsevier B.V.

Developments in the Sequential Framework for


Heat Exchanger Network Synthesis of industrial
size problems
Rahul Anantharaman and Truls Gundersen
Dept. of Energy & Process Engineering, Norwegian University of Science and
Technology, Kolbjorn Hejes vei 1B, NO-7491, Trondheim

Abstract
A Sequential Framework for Heat Exchanger Network Synthesis (HENS) is presented
and the philosophy of this iterative methodology is explained. There are two main
advantages of the proposed methodology. First, the design procedure is, to a large
extent, automated while keeping significant user interaction. Second, the subtasks of the
framework (MILP and NLP problems) are much easier to solve numerically than the
MINLP models that have been suggested for HENS. The limiting factors of the
methodology are the NLP and MILP models where enhanced convex estimators are
required to reach global optimum in the former while significant improvements are
required to prevent combinatorial explosion in the latter. This paper makes an attempt to
address a few of the limiting elements of the framework.

Keywords: Heat Exchanger Network Synthesis, Sequential Framework, Optimization

1. Introduction
The Heat Exchanger Network Synthesis (HENS) problem involves solving a three-way
trade-off between energy (E), heat transfer area (A), and how this total area is
distributed into a number of heat transfer units (U). For details about the subject, see
exhaustive reviews [1] and [2].
Optimization methods have been routinely applied in an effort to solve the complex and
multiple trade-offs that are inherent to the HENS problem. Simultaneous MINLP
models (for example [3]) can, in theory, address and solve the trade-offs in the HENS
problem. These models have demonstrated severe numerical problems related to the
non-linear (non-convex) and discrete (combinatorial) nature of the HENS problem.
Even with the rapid advancements in computing power and optimization technology,
the size of the problems solved with these methods does not meet industrial needs.
The HENS problem has been proved to be NP-hard in the strong sense by [4] and has
prompted a renewed interest in synthesis methods for HENS that utilize the strategy of
dividing the HENS problem into a series of sub-problems to reduce the computational
complexity of obtaining a network design. This paper presents developments in a
Sequential Framework that combines and takes maximum benefit from thermodynamic
insight, logic, heuristics and efficient optimization techniques.

2. The Sequential Framework


As a compromise between Pinch Analysis and simultaneous MINLP models, a
sequential and iterative framework has been in development in our group with the main
objective of finding near optimal heat exchanger networks for industrial size problems.
726 R. Anantharaman and T. Gundersen

The subtasks of the process are solved sequentially using Mathematical Programming.
Briefly, these steps involve: establishing the minimum energy consumption (LP),
determining the minimum number of units (MILP), finding sets of matches and
corresponding heat load distributions (HLDs) for minimum or a given number of units
(MILP), and network generation and optimization (NLP) as shown in Figure 1.
The Sequential Framework is based on the recognition that the selection of HLDs
impacts both the quantitative (network cost) and the qualitative aspects such as network
complexity, operability and controllability. The Vertical MILP model for selection of
matches and the subsequent NLP model for generating and optimizing the network form
the core engine of the framework.
Significant user interaction is built into the framework in the form of iterative loops to
enable the designer to explore and evaluate the most promising networks with respect to
Total Annual Cost (TAC), network complexity (number of units, splits, etc.),
operability and controllability.
2.1. Rationale for the loops in the framework
The loops in the framework simulate the three-way trade-off indicated in the
introduction. Loops 1 and 2 can be thought of as the area loops, loop 3 as the unit loop
and finally loop 4 as the energy loop.
2.1.1. Initialization
The level of heat recovery (represented by HRAT, the Heat Recovery Approach
Temperature) is initialized by a pre-optimization procedure such as SuperTargeting
(ST). The number of units (U) is initialized, for the corresponding HRAT, to be the
minimum number of units (Umin) using a simple MILP Transhipment model allowing
the Exchanger Minimum Approach Temperature (EMAT) to be zero. Using an EMAT in
addition to HRAT, where EMAT = HRAT, allows heat exchange across pinch points and
hence more feasible solutions. The EMAT for the Vertical Transportation Model in the
core of the framework is initialized to a small value (ex. HRAT/8).
2.1.2. Loop Sequence
The logical sequence of actions is indicated in Figure 1 as the following nested loops:
1. Derive networks for the second or the third best HLDs, keeping U, EMAT and HRAT
unchanged: Experience has shown that the Vertical MILP Transportation Model
identifies an almost perfectly ranked sequence of HLDs that leads to networks with
increasing cost. The HLD loop is mainly relevant for the qualitative aspects of the
network, as described earlier.
2. Adjust the value of EMAT slightly above the earlier value: Choosing EMAT is not
straightforward in the Vertical MILP Transportation model as the value of EMAT is
used to create additional enthalpy intervals and has to be balanced. If it is set too low,
the non-vertical values of ΔTLM,mn (see Eq 1) become very small and HLDs with such
non-vertical heat transfer will face large penalties. On the other hand, if EMAT is set
too high, potential HLDs will be excluded from the feasible set of solutions. EMAT
can be varied in two or three steps between HRAT/8 and HRAT/2. It is worth noting
that EMAT performs a similar function to the +X/-X rule when optimizing networks
in the Pinch Design Method.
3. Increase the number of units by one: For a given value of HRAT, the best solution is
one where U is close to the corresponding Umin. Hence starting at Umin ensures that
the number of loops the designer has to run through to obtain the best solution is
minimal. Also, in the first run of the framework, with U = Umin, EMAT does not
affect the HLDs obtained - thus loop 2 can be ignored. This loop is terminated when
increasing the value of U does not lead to a decrease in the TAC.
Developments in the Sequential Framework for HENS 727

Figure 1: A Sequential Framework for HENS with the Vertical Transportation model

4. Adjust the value of HRAT.

From the above discussion it is evident that, though there are a number of loops in the
framework, the best solution is arrived at early in the synthesis process.

3. Temperature Intervals (TIs) in the Vertical MILP Transportation model


The original Vertical model [5] was developed to include area considerations in the
selection of HLDs. This transhipment model was modified [6] to a transportation model
to include effects due to the differences in film heat transfer coefficients. This Vertical
Transportation model was further enhanced [7] to reduce the size of the model.
The Vertical MILP Transportation model minimizes pseudo-area as given below:

⎡ Qim, jn ⎤ (1)
min ∑∑∑∑ ⎢ ⎥
i ⎣U ij ⋅ ΔTLM ,mn ⎥⎦
j m n ⎢

where i and j are the hot and cold streams respectively and, m and n are the hot and cold
temperature intervals.
The model gives a ranked sequence of increasing network area when the pseudo-area
of Eq 1 replicates the actual area of heat exchangers in the network. This is possible
when the sizes of the TIs are small (or a large number of TIs). This can be visualized as:
creating more intervals allows matching corresponding to the spaghetti structure – and
thus minimum area.
The transportation model is a polynomial time algorithm [4]. Hence the number of TIs
must be limited to reduce computational time while ensuring that the model predicts the
accurate ranked sequence. A heuristic approach for creating TIs, described in [8] for an
area targeting model, was tested but failed to meet our requirements. The procedure
below describes how to generate Temperature Intervals based on Enthalpy Intervals
(EIs) of the balanced composite curves – a Vertical model.
Step 1. Establish the balanced composite curves, using HRAT, stream and utility data.
Step 2. Supply and target temperatures of all streams, including utility streams, are set
to be the Primary Hot/Cold Temperatures.
728 R. Anantharam and T. Gundersen

Figure 2: The Modal Trimming method

Step 3. For all cold supply and target temperatures, find adjacent hot temperatures
placed vertically above the kinks of the cold composite curve. These are the
Secondary Hot Temperatures. Similarly, for all hot supply and target
temperatures, find adjacent cold temperatures placed vertically below the hot
composite curve. These are the Secondary Cold Temperatures.
Step 4. For all cold supply temperatures, find the corresponding Tertiary Hot
Temperatures by adding EMAT. Disregard any hot temperature that is colder
than the coldest hot target temperature. Similarly, for all hot supply
temperatures, find the corresponding Tertiary Cold Temperature by subtracting
EMAT. Disregard any cold temperature that is hotter than the hottest cold target
temperature.
Step 5. Quaternary Hot/Cold Temperatures are calculated by adding/subtracting EMAT
to the Secondary Cold/Hot Temperatures.
Step 6. The hot/cold temperatures from Steps 2 to 5 are merged. They are then sorted
and duplicate temperatures removed to give the corresponding hot and cold TIs.
Note: The number of hot TIs need not equal the number of cold TIs.

4. GLOBAL OPTIMA FOR NETWORK GENERATING NLP MODEL


In the Sequential Framework, network generation and optimization is performed by an
NLP formulation, where the actual network topologies are extracted from the stream
superstructure given in [9]. All possible network structures for a given set of HLDs are
included in this superstructure. The non-convexities associated with the formulation and
convex estimators to overcome these are discussed in [10]. In this work, global
optimization of the NLP formulation by the Modal Trimming method is explored.
4.1. The Modal Trimming Method
The Modal Trimming Method [11] is based on the Tunneling Algorithm [12] that has
evolved into a robust method for global optimization with many applications. The
modal trimming method consists of the following two phases: 1) finding a local optimal
solution to obtain a global quasi-optimal one, and 2) finding a feasible solution with the
Developments in the Sequential Framework for HENS 729

value of the objective function equal to that of the global quasi-optimal one to obtain a
starting point for finding a better local optimal solution. These steps are repeated until a
feasible solution is not found. The current global quasi-optimal is taken to be the global
optimal solution. Figure 2 shows the functioning of the modal trimming method.
The second step, the search for feasible solutions, is the most important step in this
method. An extended Newton Raphson method is applied to search for a feasible
solution where the value of the objective function equals the current global quasi-
optimum. The Moore-Penrose generalized inverse of the Jacobian is utilized in the
normal Newton-Raphson method to improve the Newton-Raphson method for this case.
GAMS/CONOPT is used to find the local optima (Step 1), while Visual Basic with
Matlab-Excel Link is used to search for feasible solutions (Step 2) when implementing
the Modal Trimming method. The four stream, six exchanger example presented in [10]
was used as a test case. The method proved to be inefficient, taking over two hours to
attain the global optimum. The method and its implementation will need further testing
to identify cause(s) of the inefficiency. It is also unclear, at this stage, if the method is
suitable for this NLP formulation.

5. EXAMPLE
In this section, a medium sized problem is solved to illustrate the use of the Sequential
Framework. SeqHENS, an Excel Add-in, for the Sequential Framework, was used to
generate networks for the given stream data. For comparison purposes, the operating
cost of the solution presented in [13], 1,014,323 $yr-1, was unchanged. This corresponds
to a HRAT of 20.35 °C.
The fourth step of the framework generates the best solution with a TAC of 1,532,148
$yr-1. This compares well with the solution for base case given in [13] with a TAC of
1,530,063 $yr-1. The network generated is shown in Figure 3. It is also worth noting that
the Vertical MILP transportation model for selecting HLDs in the Sequential
Framework allows only one match between streams. This “simplification” of allowing
just one match between streams in [13] gives a TAC of 1,568,745 $yr-1.
Yet another example with more details on computations and the search for the ‘best’
network can be found in [7].

6. CONCLUSION
A Sequential Framework for Heat Exchanger Network Synthesis and the rationale
behind the method and the loops in the framework has been presented. The importance
of TIs in the Vertical MILP transportation model was discussed and a procedure for
establishing a set of optimum TIs is detailed. Extensive tests were performed to ensure
that the Vertical Model gives the correct ranked sequence of HLDs with these TIs.
The Modal Trimming method was applied to the NLP formulation to achieve global
minimum. This system was found to be inefficient and improvement in orders of
magnitude is required. Further testing must also be done to check the suitability of this
method to solve the NLP formulation in the Sequential Framework.
This framework has been applied successfully to a test problem of 15 streams and gave
comparable solutions to the previously published results.
Significant improvements are required to solve the minimum number of units MILP
model and the Vertical MILP Transportation model (in addition to the solution methods
for the NLP model) for industrial cases. These are areas of ongoing research in our
group.
730 R. Anantharaman and T. Gundersen

Figure 3: Best network for example problem

References
1. T. Gundersen and L. Naess, 1988, The synthesis of cost optimal heat exchanger networks,
Comp. & Chem. Eng., 12, 6, 503-530.
2. K.C. Furman and N.V. Sahinidis, 2002, A critical review and annotated bibliography for heat
exchanger network synthesis in the 20th century, Ind. Eng. Chem. Res., 41, 2335-2370.
3. T.F. Yee and I.E. Grossmann, 1990, Simultaneous optimization models for heat integration –
II. Heat exchanger network synthesis, Comp. & Chem. Eng., 14, 10, 1165-1184.
4. K.C. Furman and N.V. Sahinidis, 2001, Computational complexity of heat exchanger network
synthesis, Comp. & Chem. Eng., 25, 1371-1390.
5. T. Gundersen and I.E. Grossmann, 1990, Improved optimization strategies for automated heat
exchanger network synthesis through physical insights, Comp. & Chem. Eng., 14, 9, 924-944.
6. T. Gundersen, P. Traedal and A. Hashemi-Ahmady, 1997, Improved sequential strategy for the
synthesis of near-optimal heat exchanger networks, Comp. & Chem. Eng., 21, Suppl., 59-64.
7. R. Anantharaman and T. Gundersen, 2005, Revisiting the sequential framework for near-
optimal heat exchanger network synthesis, In Proceedings from PRES 2005, Vol 1, 67-72.
8. J.M. Jeżowski, H.K. Shethna and F.J.L. Castillo, 2003, Area targets for heat exchanger
networks using linear programming, Ind. Eng. Chem. Res., 42, 1723-1730.
9. C.A. Floudas, A.R. Ciric and I.E. Grossmann, 1986, Automatic synthesis of optimum heat
exchanger network configurations, AIChE Journal, 32, 2, 276-290.
10. A. Hashemi-Ahmady, J.M. Zamora and T. Gundersen, 1999, A sequential frameowrk for
optimal synthesis of industrial size heat exchanger networks, In Proceedings from PRES ’99,
329-334.
11. R. Yokoyama, M. Inui, K. Ito, 2005, Prediction of energy demands using neural networks by
a global optimization method, In Proceedings of ECOS ’05, Vol. 2, 609-616.
12. A.V. Levy and A. Montalvo, 1985, The tunneling algorithm for the global optimization of
functions, SIAM Journal of Scientific and Statistical Computing, 6, 1, 15-29.
13. K. Björk and R. Nordman, 2005, Solving large-scale retrofit heat exchanger network synthesis
problems with mathematical optimization methods, Chem. Eng. Process, 44, 869-876.

You might also like