PDF

Free ebooks ==> www.Ebook777.
com
www.Ebook777.com
Free ebooks ==> www.Ebook777.com
Understanding
Batch Chemical
Processes
Modelling and Case Studies
www.Ebook777.com
www.Ebook777.com
Understanding
Batch Chemical
Processes
Modelling and Case Studies
Thokozani Majozi
Esmael R. Seid
Jui-Yuan Lee
www.Ebook777.com
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2017 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works
Printed on acid-free paper

Version Date: 20160901
International Standard Book Number-13: 978-1-4987-7317-1 (Hardback)
This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have
been made to publish reliable data and information, but the author and publisher cannot assume responsibility
for the validity of all materials or the consequences of their use. The authors and publishers have attempted to
trace the copyright holders of all material reproduced in this publication and apologize to copyright holders if
permission to publish in this form has not been obtained. If any copyright material has not been acknowledged
please write and let us know so we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmit-
ted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented,
including photocopying, microfilming, and recording, or in any information storage or retrieval system, with-
out written permission from the publishers.
For permission to photocopy or use material electronically from this work, please access www.copyright.com
(http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive,
Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registration
for a variety of users. For organizations that have been granted a photocopy license by the CCC, a separate
system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used
only for identification and explanation without intent to infringe.
Library of Congress Cataloging-in-Publication Data
Names: Majozi, Thokozani, author. | Seid, Esmael Reshid, author. | Lee, Jui-Yuan,
author.
Title: Understanding batch chemical processes : modelling and case studies / Thokozani
Majozi, E R Seid, and J-Y Lee.
Description: Boca Raton : Taylor & Francis, a CRC title, part of the Taylor & Francis
imprint, a member of the Taylor & Francis Group, the academic division of T&F
Informa, plc, [2017]
Identifiers: LCCN 2016034450 (print) | LCCN 2016053916 (ebook) | ISBN
9781498773171 (acid-free paper) | ISBN 9781498773188 (eBook)
Subjects: LCSH: Chemical processes.
Classification: LCC TP155.7 .M33 2017 (print) | LCC TP155.7 (ebook) | DDC
660/.2812--dc23
LC record available at https://lccn.loc.gov/2016034450
Visit the Taylor & Francis Web site at

http://www.taylorandfrancis.com
and the CRC Press Web site at

http://www.crcpress.com
Contents
Foreword................................................................................................................. xiii
Acknowledgements................................................................................................... xv
Authors....................................................................................................................xvii
Chapter 1 Introduction to Batch Processes............................................................1

1.1 Introduction................................................................................1
1.2 Batch Process Representation....................................................2
1.3 Different Storage Operational Policies in Batch Plant............... 4
1.4 Multipurpose and Multiproduct Batch Plants............................ 4
1.5 Scheduling Techniques for Batch Processes..............................6
1.5.1 Graphical Techniques....................................................6
1.5.2 Mathematical Techniques.............................................. 8
1.5.2.1 Models Based on Precedence Sequencing.... 9
1.5.2.2 Models Based on Slot Representation......... 10
1.6 Scheduling Techniques for Continuous and
Semi-Continuous Multistage Multiproduct Plants................... 12
1.7 Scheduling Techniques for Multipurpose Batch Plants........... 13
1.7.1 Models Based on Discrete-Time Representation........ 13
1.7.2 Models Based on Global Event Point
Representations........................................................... 14
1.7.3 Models Based on Slot Time Representation............... 15
1.7.4 Models Based on Unit-Specific Event-Point
Representation............................................................. 16
1.8 Conclusions...............................................................................20
References........................................................................................... 22
Chapter 2 Modelling for Effective Solutions: Reduction of Binary Variables.... 27

2.1 Introduction.............................................................................. 27
2.2 Necessary Background............................................................. 27
2.3 Problem Statement.................................................................... 29
2.4 Model Formulation................................................................... 29
2.4.1 Allocation Constraints................................................ 29
2.4.2 Capacity Constraints................................................... 30
2.4.3 Material Balance around a Storage Unit..................... 30
2.4.4 Duration Constraints................................................... 31
2.4.5 Sequence Constraints for Processing Units................ 31
2.4.6 Sequence Constraints for Different Tasks in
Different Units............................................................. 32
v
vi Contents
2.4.6.1 Sequence Constraints for Different

Tasks in Different Units for Stable
State That Can Wait after It Is Produced..... 32
2.4.6.2 Sequence Constraints for Different
Tasks in Different Units for Unstable
States with Zero Wait.................................. 33
2.4.6.3 Sequence Constraint That Links the
Starting Time of the Consuming Units
with the Finishing Time of Producing
Units............................................................. 33
2.4.7 Sequence Constraints for FIS Policy........................... 33
2.4.8 Time Horizon Constraints...........................................34
2.4.9 Objective Function......................................................34
2.5 Illustrative Example.................................................................. 35
2.5.1 Results and Discussion................................................ 36
2.6 Case Studies............................................................................. 39
2.6.1 Case Study I................................................................ 39
2.6.1.1 Results and Discussions............................... 39
2.6.2 Case Study II............................................................... 41
2.6.2.1 Results and Discussion................................ 41
2.7 Conclusions...............................................................................44
Nomenclature......................................................................................44
References........................................................................................... 45
Chapter 3 Methods to Reduce Computational Time: Prediction of Time Points..... 49

3.1 Introduction.............................................................................. 49
3.2 Motivation................................................................................. 49
3.4 Mathematical Model Using State Sequence Network.............. 50
3.4.1 Allocation Constraints................................................ 51
3.4.2 Capacity Constraints................................................... 51
3.4.3 Material Balance for Storage...................................... 51
3.4.4 Duration Constraints................................................... 51
3.4.5 Sequence Constraints.................................................. 51
3.4.5.1 Same Task in Same Unit.............................. 51
3.4.5.2 Different Tasks in Same Unit...................... 51
3.4.6 Different Tasks in Different Units............................... 52
3.4.6.1 If an Intermediate State Is Produced
from One Unit.............................................. 52
from More than One Unit............................ 52
3.4.7 Sequence Constraint for Fixed Intermediate
Storage Policy.............................................................. 52
3.4.8 Storage Constraints..................................................... 52
3.4.9 Time Horizon Constraints........................................... 53
Contents vii
3.4.10 Calculation of the Total Duration a Unit Is

Active within the Time Horizon................................. 53
3.4.10.1 Duration Constraints Are Fixed.................. 53
3.4.10.2 Duration Constraints Dependent on
Batch Size....................................................54
3.4.11 Objective Function...................................................... 54
3.4.12 Prediction of Optimal Number of Time Points........... 55
3.5 Case Studies............................................................................. 57
3.5.1 Case Study I: Simple Batch Plant Involving
Three Tasks..................................................................57
3.5.2 Case Study II: Batch Plant with a Recycle Stream.....64
3.5.2.1 Results and Discussions...............................64
3.5.3 Case Study III: Multipurpose Batch Plant with
Six Units and Eight Tasks........................................... 67
3.5.4 Case Study IV: Industrial Case Study......................... 75
3.6 Conclusions...............................................................................80
Notation...............................................................................................80
References........................................................................................... 82
Chapter 4 Integration of Scheduling and Heat Integration:

Minimization of Energy Requirements.............................................. 83
4.1 Introduction.............................................................................. 83
4.2 Necessary Background............................................................. 83
4.3 Motivating Example.................................................................84
4.5 Model Formulation................................................................... 86
4.5.1 Mass and Energy Balances for Processing Units........ 86
4.5.2 Mass and Energy Balances for Storage....................... 89
4.5.3 Heating and Cooling in Processing Units...................90
4.5.4 Heat Integration Constraints....................................... 93
4.5.5 Objective Function...................................................... 95
4.6 Illustrative Case Study..............................................................96
4.7 Summary................................................................................ 103
4A Appendix................................................................................ 103
Nomenclature.................................................................................... 106
References......................................................................................... 108
Chapter 5 Heat Integration in Multipurpose Batch Plants................................. 109

5.1 Introduction............................................................................ 109
5.2 Necessary Background........................................................... 109
5.2.1 Developments in the Twentieth Century................... 110
5.2.2 Developments in the Twenty-First Century.............. 111
viii Contents
5.2.2.1 Models Developed for Direct Heat

Integration.................................................. 111
5.2.2.2 Models Developed for Indirect Heat
Integration.................................................. 112
5.2.2.3 Models Applied for Real Industrial
Case Studies............................................... 112
5.3 Problem Statement and Objectives......................................... 114
5.4 Mathematical Formulation..................................................... 114
5.4.1 Model Constraints..................................................... 114
5.4.2 Model Objective Functions....................................... 123
5.5 Case Studies........................................................................... 124
5.5.1 Case Study I.............................................................. 124
5.5.1.1 Results and Discussion.............................. 126
5.5.2 Case Study II............................................................. 130
5.5.3 Case Study III............................................................ 135
5.5.3.1 Results and Discussions............................. 135
5.6 Conclusions............................................................................. 141
Nomenclature.................................................................................... 142
References......................................................................................... 143
Chapter 6 Design and Synthesis of Heat-Integrated Batch Plants Using

an Effective Technique...................................................................... 147
6.1 Introduction............................................................................ 147
6.2.1 Basic Grass Root Design for Multipurpose
Batch Plants............................................................... 148
6.2.2 Extended Design....................................................... 149
6.3 Problem Statement.................................................................. 150
6.4 Model Formulation................................................................. 151
6.4.1 Equipment Existence Constraints............................. 151
6.4.2 Unit Size Constraints................................................. 151
6.4.3 Capacity Constraints................................................. 151
6.4.4 Material Balance for Storage.................................... 152
6.4.5 Material Balance around the Processing Unit.......... 152
6.4.6 Existence Constraints for Piping............................... 153
6.4.7 Duration Constraints (Batch Time as a Function
of Batch Size)............................................................ 153
6.4.8 Sequence Constraints................................................ 153
6.4.8.1 Same Task in Same Unit............................ 154
6.4.8.2 Different Tasks in Same Unit.................... 154
6.4.9 Sequence Constraints for Different Tasks in
Different Units........................................................... 154
6.4.9.1 If an Intermediate State s Is Produced
from One Unit............................................ 154
Contents ix

from More than One Unit.......................... 155
6.4.9.3 Sequence Constraints for Completion
of Previous Tasks....................................... 155
6.4.10 Sequence Constraints for Finite Intermediate
Storage (FIS) Policy.................................................. 155
6.4.11 Time Horizon Constraints......................................... 156
6.4.12 Energy Integration Constraints................................. 156
6.4.13 Objective Function.................................................... 158
6.5 Case Study.............................................................................. 159
6.6 Results and Discussion........................................................... 159
6.7 Conclusions............................................................................. 163
Nomenclature.................................................................................... 163
References......................................................................................... 166
Chapter 7 Simultaneous Scheduling and Water Optimization:

Reduction of Effluent in Batch Facilities.......................................... 169
7.1 Introduction............................................................................ 169
7.2 Motivation for the Study......................................................... 169
7.4 Problem Superstructure.......................................................... 171
7.5.1 Sets............................................................................ 172
7.5.2 Variables.................................................................... 173
7.5.3 Parameters................................................................. 174
7.5.4 Mass Balance Constraints......................................... 174
7.5.4.1 Mass Balance around a Unit j.................... 174
7.5.4.2 Mass Balance around Central Storage.......176
7.5.4.3 Mass Balance around the Regenerator...... 177
7.5.5 Scheduling Constraints............................................. 178
7.5.5.1 Task Scheduling......................................... 178
7.5.5.2 Recycle/Reuse Scheduling......................... 179
7.5.5.3 Central Storage Scheduling....................... 180
7.5.5.4 Regenerator Scheduling............................. 182
7.5.6 Feasibility and Time Horizon Constraints................ 183
7.5.7 Objective Function.................................................... 184
7.6 Case Studies........................................................................... 185
7.6.1 Cast Study I............................................................... 185
7.6.2 Case Study II............................................................. 188
7.7 Conclusions............................................................................. 191
7A Appendix................................................................................ 191
References......................................................................................... 195
x Contents
Chapter 8 Optimization of Energy and Water Use in Multipurpose Batch

Plants Using an Improved Mathematical Formulation..................... 197
8.1 Introduction............................................................................ 197
8.2 Necessary Background...........................................................197
8.2.1 Scheduling of Batch Plants....................................... 198
8.2.2 Energy Integration in Batch Plants............................198
8.2.3 Wastewater Minimization in Batch Plants................ 200
8.3 Problem Statement..................................................................202
8.4.1 Heat Integration Model............................................. 203
8.4.2 Wastewater Minimization Model..............................208
8.5 Case Studies............................................................................ 211
8.5.1 Case Study I.............................................................. 211
8.5.2 Case Study II............................................................. 218
8.6 Conclusions............................................................................ 225
Nomenclature....................................................................................225
References......................................................................................... 228
Chapter 9 Targeting for Long-Term Time Horizons: Water Optimization........ 233

9.1 Introduction............................................................................ 233
9.3.1 Concepts of Cyclic Scheduling................................. 234
9.3.2 Nomenclature............................................................ 235
9.3.2.1 Sets............................................................. 235
9.3.2.2 Variables Associated with Wastewater
Minimization�� 235
9.3.2.3 Variables Associated with Production
Scheduling�� 236
9.3.2.4 Parameters Associated with
Wastewater Minimization�� 236
9.3.2.5 Parameters Associated with Production
Scheduling�� 237
9.3.3 Mathematical Model................................................. 237
9.3.3.1 Mass Balance Constraints......................... 237
9.3.3.2 Sequencing and Scheduling Constraints..... 242
9.3.3.3 Production Scheduling Constraints........... 245
9.4 Illustrative Examples.............................................................. 249
9.4.1 Example 1.................................................................. 249
9.4.2 Example 2.................................................................. 258
9.4.2.2 Remarks..................................................... 265
Contents xi
9.5 Industrial Case Study.............................................................266

9.5.1 Results and Discussions............................................ 267
9.5.1.1 Wastewater Minimization without
Central Storage for 168 h Time Horizon�� 268
9.5.1.2 Wastewater Minimization with Central
Storage for 168 h Time Horizon..............� 269
9.6 Conclusion.............................................................................. 270
9A Appendix................................................................................ 271
References......................................................................................... 273
Chapter 10 Long-Term Heat Integration in Multipurpose Batch Plants

Using Heat Storage............................................................................ 275
10.1 Introduction............................................................................ 275
10.3 Cyclic Scheduling Concepts................................................... 277
10.4 Problem Statement and Objectives......................................... 279
10.4.1 Scheduling Data........................................................ 279
10.4.2 Heat Integration Data................................................ 279
10.5 Mathematical Model............................................................... 279
10.5.1 Sets............................................................................ 279
10.5.2 Continuous Variables................................................280
10.5.3 Binary Variables........................................................ 281
10.5.4 Parameters................................................................. 281
10.5.5 Mass Balance between Two Cycles........................... 288
10.5.6 Sequence Constraints between Cycles:
Completion of Previous Tasks�� 288
10.5.7 Tightening Constraints.............................................. 289
10.6 Multipurpose Example........................................................... 292
10.6.1 Cyclic Portion............................................................ 295
10.6.2 Initial Period.............................................................. 295
10.6.3 Final Period............................................................... 297
10.6.4 Comparison with Direct Solution.............................. 297
10.7 Industrial Case Study............................................................. 298
10.7.1 Cyclic Portion............................................................ 298
10.7.2 Initial Period..............................................................302
10.7.3 Final Period............................................................... 303
10.7.4 Comparison with Existing Literature Model............304
10.8 Conclusions.............................................................................306
References.........................................................................................306
Index��309
Foreword
Batch chemical processes are always ideal in the manufacture of small-volume,
high-value-added products, as traditionally encountered in pharmaceutical, agro-
chemical and food and beverage industries. The main distinction between these
processes and their continuous counterparts pertains to the handling of time.
Continuous operations can attain steady state, wherein time is completely overrid-
den, whereas batch operations are never at steady state. This simply implies that
time can never be ignored in the analysis of batch processes. Consequently, the
synthesis, design and optimization of these operations should embed time as a criti-
cal dimension.
An additional complexity arises from the presence of binary variables, which
are necessary for indicating the activity of a particular task in a particular unit.
Unfortunately, these variables also entail significant mathematical challenges,
since the computational time of concomitant mathematical models strongly
depends on the number of binary variables – the more the binary variables, the
longer the CPU time. Consequently, the major part of this book concerns the
reduction of the number of binary variables without compromising the accuracy
of the final solution.
This book covers most of the key aspects of batch plants, particularly those that
are characterized by complex recipes, the so-called multipurpose batch plants. The
authors start by giving a broad description of these processes to bring the reader
closer to the subject. This is followed by detailed mathematical modelling ranging
from scheduling to design and synthesis with hot water optimization considerations.
The authors have also provided the necessary background for most of the complex
aspects and further substantiate the performance of the presented models with illus-
trative examples and case studies. The aims of the book are summarized as follows:
1. Introduce the reader to the key or critical aspects in the mathematical mod-
elling of batch processes.
2. Present techniques that overcome the computational complexity in order to
yield models that are solvable in near real time.
3. Demonstrate how batch processes could be analyzed, synthesized and
designed optimally using proven mathematical formulations.
4. Demonstrate how water and energy aspects could be incorporated within
the scheduling framework that seeks to capture the essence of time.
5. Present real-life case studies where the mathematical modelling of batch
plants have been successfully applied.
It is my view that the authors have succeeded in achieving this goal. The book is
likely to close a lingering knowledge gap in this area of chemical engineering.
xiii
xiv Foreword
Worthy of mention is the fact that the authors are recognized experts in this area
of research, having published a similar treatise on the subject in 2010 and recently
via an edited book in 2015.
Professor Dominic C. Y. Foo, PEng, CEng, FHEA, MIEM

University of Nottingham, Malaysia Campus
Selangor, Malaysia
Acknowledgements
This work would not have been accomplished without the support from various
funding bodies and inspiration from some of the leading figures in the field of pro-
cess integration. Foremost among these are
• The National Research Foundation (NRF) and the Department of Science

and Technology (DST) through the NRF/DST Chair in Sustainable Process
Integration at Wits University, South Africa.
• Water Research Commission (WRC), South Africa.
• Professor Mahmoud El-Halwagi at Texas A&M, USA, who has always been
a reliable sounding board on various ideas that are presented in this book.
• The late Professor Toshko Zhelev who was a friend, a mentor and one of the
pioneers of batch chemical process integration. His influence is everlasting.
May his soul rest in peace!
xv
Authors
Thokozani Majozi is a full professor in the School of Chemical and Metallurgical
Engineering at Wits University, Johannesburg, where he also holds the NRF/DST
chair in sustainable process engineering. His main research interest is batch chemical
process integration, where he has made significant scientific contributions that have
earned him international recognition. Some of these contributions have been adopted
by the industry. Prior to joining Wits, he spent almost 10 years at the University of
Pretoria, South Africa, initially as an associate professor and then as a full professor
of chemical engineering. He was also an associate professor in computer science at
the University of Pannonia in Hungary from 2005 to 2009. Professor Majozi earned
a PhD in process integration at the University of Manchester Institute of Science and
Technology in the United Kingdom. He is a member of the Academy of Sciences of
South Africa and a fellow of the Academy of Engineering of South Africa. He has
received numerous awards for his research, including the Burianec Memorial Award
(Italy), S2A3 British Association Medal (Silver) and the South African Institution of
Chemical Engineers Bill Neal-May Gold Medal. He was also twice a recipient of the
NSTF Award and twice a recipient of the NRF President’s Award. Professor Majozi
is the author or coauthor of more than 150 scientific publications, including two
books in batch chemical process integration published by Springer and CRC Press in
2010 and 2015, respectively.
Esmael Seid earned a BSc in chemical engineering at Bahir Dar University, Ethiopia.
He then worked in the process industry for three years before joining the University
of Pretoria, Pretoria, South Africa, in 2009, where he earned an MSc Eng and a PhD
in chemical engineering. Dr. Seid has several publications in international refereed
journals on design, synthesis, scheduling and resource conservation, with particular
emphasis on water and energy for multipurpose batch plants. He is currently a visiting
scholar at the Chemical Engineering Department, Texas A&M University, working
on sustainable process design through process integration.
Jui-Yuan Lee is an assistant professor at the Department of Chemical Engineering

and Biotechnology, National Taipei University of Technology, Taiwan. He earned
a PhD in chemical engineering at National Taiwan University (NTU) in 2011 and
conducted postdoctoral research at NTU (2011–2013) and at the University of the
Witwatersrand, Johannesburg, South Africa (2013–2014). His research centres on
process integration for energy savings and waste reduction, and his interests include
low-carbon energy system planning and hybrid power system design. He has pub-
lished more than 30 journal papers and 40 conference papers and coedited a book
on batch process integration. Dr. Lee works closely with several collaborators in
Southeast Asia and Africa.
xvii
1 Introduction to
Batch Processes
1.1 INTRODUCTION
Batch processes differ from continuous processes in several ways. The main differ-
ence is that time is inherent in batch processes. In batch processes, every task has
a definite duration with starting and finishing times, whereas in continuous pro-
cesses, time is important during non-steady-state operation. As a result, scheduling
of batch processes is vital to the operation of any batch facility. Furthermore, in
batch plants, detailed requirements for the various products may be specified on a
day-to-day basis. A production schedule must indicate the sequence and manner in
which the products are to be produced and specify the times at which the process
operations are to be carried out. It is clear that the overall productivity and eco-
nomic effectiveness of batch plants depend critically on the production schedule as
it harmonizes the entire plant operation to attain production goals. While flexibility
of batch plants improves productivity, it also makes plant scheduling a challenging
task. Much research has focused on developing optimization techniques for sched-
uling batch plants with the aim of reducing the CPU time required to attain the
optimal objective value.
This chapter provides a detailed literature review on various scheduling tech-
niques. The review covers work which has been done in the field of mathematical
models used for scheduling of batch plants. Papers presented in the last two decades
are considered in the review, with focus on models based on unit-specific event-point
continuous-time representation. The chapter is systematically divided into six major
sections. Section 1.1 discusses recipe, State Task Network (STN) and State Sequence
Network (SSN) representations for batch plants. One of the major characteristics that
differentiates batch plants from continuous plants is the existence of intermediate
storage in batch plants in order to separate operations in multipurpose equipment
and to free the equipment for subsequent processes. The different intermediate stor-
age operational philosophies that exist in batch plants are discussed in Section 1.2.
Section 1.3 discusses the different types of batch plants under the main division of
multiproduct and multipurpose batch plants.
Section 1.4 discusses the different types of models developed for scheduling of
multiproduct batch plants. In this section, the models are categorized under the main
division of graphical technique and mathematical technique. In the mathematical
section, the different models are grouped into models based on sequence precedence
and slot time representations. A brief of scheduling techniques for continuous and
semi-continuous plants is given in Section 1.5. Section 1.6 details scheduling tech-
niques developed for multipurpose batch plants. Finally, conclusions and limitations
of the current scheduling techniques are briefly discussed in Section 1.7.
1
2 Understanding Batch Chemical Processes
1.2 BATCH PROCESS REPRESENTATION

Batch plant processes were first represented in terms of ‘recipe networks’, as used
by Reklaitis (1991). This is analogous to the flow sheet representation of continuous
plants, but proposed to describe the process itself rather than a specific plant. Each
node on a recipe network corresponds to a task, with directed arcs between nodes
representing task precedence. Although recipe networks are certainly adequate for
several processing structures, they often involve ambiguities when applied to more
complex ones. Kondili et al. (1993) showed that a recipe network was not enough
to represent batch processes without vagueness, as illustrated in Figure 1.1. It is not
clear from this representation whether task 1 produces two different products later
used as inputs of tasks 2 and 3, respectively, or whether it produces one type of prod-
uct which is then shared between unit operations 2 and 3. Similarly, it is also impos-
sible to determine from Figure 1.2 whether task 4 requires two different types of
feedstock, respectively produced by tasks 2 and 5, or whether it only needs one type
of feedstock which can be produced by either 2 or 5. Both interpretations are equally
plausible. The former could be the case if, say, task 4 is a catalytic reaction requiring
a main feedstock produced by task 2 and a catalyst which is then recovered from the
reaction products by the separation task 5. The latter case could arise if task 4 were
an ordinary reaction task with a single feedstock produced by 2, with task 5 separat-
ing the product from the unreacted material which is then recycled to 4.
In order to remove these ambiguities in a systematic fashion, Kondili et al. (1993)
proposed a new representation for chemical processes called State Task Network
representation. The distinctive characteristic of the STN was that it had two types of
nodes: namely, the state nodes, representing the feeds, intermediate and final prod-
ucts; and the task nodes, representing the processing operations which transform
materials from one or more input states to one or more output states. State and task
nodes were denoted by circles and rectangles, respectively. State Task Networks are
free from the ambiguities associated with recipe networks. Figure 1.2 shows two
different STN representations, both of which correspond to the recipe network of
Figure 1.1. The STN representation is equally suitable for networks of all types of
processing tasks, such as continuous, semi-continuous and batch.
Majozi and Zhu (2001) introduced a new concept called the State Sequence
Network representation for scheduling of batch plants. The SSN was a graphical
network representation of all the states that exist in batch plants, and the network was
2 4 5
FIGURE 1.1 Recipe representation of chemical processes.

Introduction to Batch Processes 3
2 4 5
3
(a)
2 4 5
1
3
(b)
FIGURE 1.2 STN representations for recipe representation.
formulated based on the production recipe. A state changes from one state to another
state when it undergoes a unit operation such as mixing, separation or reaction. The
building blocks of the SSN are shown in Figures 1.3 through 1.5. From these build-
ing blocks, it is easy to construct a SSN for any process recipe. The SSN representa-
tion was developed by realizing that (1) the capacity of a unit in which a particular
state is used sets an upper limit on the amount of state used or produced by the
s s΄
s΄
FIGURE 1.3 Simple unit operation.
s1
sN
s1
s΄
sN
s΄
FIGURE 1.4 Unit operation with mixing/reaction.

s˝
s΄
s s
s˝
s΄
FIGURE 1.5 Unit operation with splitting.
corresponding task, (2) the presence of a particular state in an operation corresponds

to the existence of a corresponding task and (3) the usage of state s corresponds to
the production of another state s′.
1.3 DIFFERENT STORAGE OPERATIONAL

POLICIES IN BATCH PLANT
According to Reklaitis (1982), the multistage nature of a batch processing network
allows several different storage and waiting options:
• Unlimited intermediate storage (UIS)

• Finite intermediate storage (FIS)
• No intermediate storage (NIS)
• Central intermediate storage (CIS)
• Unlimited wait (UW)
• Finite wait (FW)
• Zero wait (ZW)
The ZW or FW policy is used where unstable intermediate material must be pro-

cessed immediately or within a short time after the previous step has been completed.
In the NIS policy, there is no intermediate storage between process units, but a prod-
uct can be stored in a process unit before moving to the next available processing
unit. A more practical operational storage policy is that of FIS, where there is limited
storage capacity between process units. The UIS policy is more of a theoretical one.
If the intermediate products are compatible, the CIS policy is recommended. After
materials are processed in a unit, they may be held in the unit temporarily under the
UIS, FIS, UW, FW or NIS modes, but they must be transferred to the downstream
unit immediately in the ZW mode. Some batch plants have a mixed intermediate
storage (MIS) policy with a combination of the FIS, NIS and ZW modes.
1.4 MULTIPURPOSE AND MULTIPRODUCT BATCH PLANTS

Chemical batch plants can be divided into multipurpose and multiproduct. In a multi-
product batch plant, different products are processed in a sequence of single product
campaigns. Each product only follows a single route through the plant. They are
serial multistage (single unit in each stage) plants, single-stage plants with parallel
units and multistage plants with parallel units (Liu 2006), as shown in Figures 1.6
through 1.8, respectively. In a multipurpose batch plant, different batch sizes of
products are produced through different production paths, as shown in Figure 1.9.
A product in this plant is produced in campaigns where each campaign involves one
or more batches.
J1 J2 J3 JM
Stage 1 Stage 2 Stage 3 Stage M
FIGURE 1.6 Schematic of a single-stage process with parallel units.
J1
J2
Raw material Products
JM
FIGURE 1.7 Schematic of a serial multistage process.
Stage 1 Stage 2 Stage 3 Stage 4 Stage 5 Finished

Orders orders
Units Orders
FIGURE 1.8 Schematic of a multistage multiproduct batch process with parallel units.
10%
90%
Reaction 1 Separation
S3 S4
Heat
S1 S2
40%
70% 60%
Reaction 2 Reaction 3
S5 S7
30%
S6
FIGURE 1.9 Schematic of a multipurpose process.
1.5 SCHEDULING TECHNIQUES FOR BATCH PROCESSES

The scheduling techniques for batch processes are broadly categorized into graphi-
cal and mathematical techniques as detailed here.
1.5.1 Graphical Techniques
Sanmarti et al. (1998, 2002) proposed a novel graph representation that takes into
consideration the specific features of chemical processes in scheduling. The advan-
tage of this work was its capability to exploit the problem structure directly, with
drastic reduction in computational intensity. Once all the processing tasks had been
represented in the recipe, an optimal schedule could be generated using the S-graph
framework. The features of S-graph approach are it does not have binary variables
which increase the computational complexity of a problem, and it does not require
the discretization of the time horizon so it can truly be called continuous.
The S-graph was constructed with nodes representing the production tasks and
arcs representing the precedence relationships among tasks. The processing orders of
two consecutive tasks were given by a weighted arc (recipe-arc) established between
the nodes of tasks. Furthermore, additional recipe-arcs were established from the
nodes of tasks generating products to the corresponding product node. The weight of
a recipe-arc was specified by the processing time of the task. Figure 1.10 illustrates
the conventional, while Figure 1.11 shows a graphical representation of the recipe of
an example with three products.
Depending on the operational philosophy (UIS or NIS), the precedence arcs were
set. For UIS, the arcs started at the node of the same unit and terminated at the same
unit completing the next task. However, for the NIS policy the arcs started at the fol-
lowing node and terminated as before. Figure 1.12 gives an example of an S-graph,
from the many possible S-graphs of master recipes with NIS policy.
PT:8 PT:9 PT:10 PT:4

Unit:1 Unit:3 Unit:4 Unit:6
PT:7
Unit:2
PT:13 PT:8 PT:15 PT:7

Unit:2 Unit:3 Unit:1 Unit:5
PT:4 PT:2 PT:3

Unit:2 Unit:4 Unit:6
PT:6
Unit:1
PT:9 PT:10
Unit:3 Unit:5
FIGURE 1.10 Conventional representations of the master recipes for three products.
1 8 3 9 4 10 5 4
E1 E3 E4 E6
2
E2
13 8 15 7
1 2 3 4
E2 E3 E1 E5
2 4 4 2 6 3
6 E2 E4 E6
1
E2
10
6
3 9 5
E3 E5
FIGURE 1.11 Graphical representation of the recipe.

1 8 3 9 4 10 5 4
E1 E3 E4 E6
2
E2
1 13 2 8 3 15 4 7
E2 E3 E1 E5
2 4 4 2 6 3
6 E2 E4 E6
1
E2
10
6 9
3 5
E3 E5
FIGURE 1.12 S-graph for NIS policy for minimization of makespan.
Sanmarti et al. (1998, 2002) introduced the longest path algorithm that used the
schedule graph to calculate the makespan of a schedule. Since the resulting formu-
lation of this algorithm was polynomial, it could be applied effectively in solving
scheduling problems. The longest path algorithm traced the graph backwards start-
ing from the end node and terminating at the source. The path with the maximal
length expressed the makespan of the schedule. The start time of each task in the
schedule could be determined by the longest path of this task. Thus, the longest path
algorithm also provided the timing of the whole schedule.
Further advances on the S-graph approach for scheduling have been made by
Majozi and Friedler (2006). Their formulation extended the work of Sanmarti et al.
(2002), where instead of minimizing makespan they maximized throughput over
a given time horizon of interest. In their formulation, a new concept, which was
a novel node-cutting algorithm which improved the computational efficiency, was
introduced. Although this method seems good compared to the mathematical mod-
els, the generation of S-graphs is complex and is suited to the case where tasks such
as storage are not definite.
1.5.2 Mathematical Techniques
The different continuous time models proposed in the literature can be broadly clas-
sified into three distinct categories: slot-based, event-based and precedence-based
(sequence-based) time representation (Méndez et al., 2006; Shaik et al., 2006;

Shaik and Floudas, 2007, 2008).
1.5.2.1 Models Based on Precedence Sequencing

Hui and Gupta (2000) presented a continuous-time mixed integer linear program-
ming (MILP) model for short-term scheduling of multistage multiproduct batch
plants based on immediate precedence sequencing. The model determined the opti-
mal sequencing and the allocation of customer orders to non-identical processing
units by minimizing the earliness and tardiness of order completion. The previous
approach to this scheduling problem relied on the application of tetra-index binary
variables (i.e., order, order, stage and unit) to represent all the combinations of order
sequences and assignments to units in the various stages. This generated a huge num-
ber of binary variables and, as a consequence, much time was required for solutions.
The developed formulation by Hui and Gupta (2000) replaced the tetra-index binary
variables by one set of tri-index binary variables (order, order and stage) without
losing the model’s generality. Consequently, the computational time was reduced.
Méndez and Cerdá (2000) proposed a new continuous-time MILP framework for
the short-term scheduling of resource constrained multistage batch plants supplying
intermediates to end product facilities at specified time intervals. The model was
based on immediate precedence sequencing. The proposed approach was applied to
the scheduling of a real-world two-stage batch plant involving seven parallel units
and six storage tanks manufacturing eight intermediate products. The proposed
MILP model gave much smaller size than previous approaches.
Méndez et al. (2001) proposed a new MILP mathematical formulation for the
resource-constrained short-term scheduling of flow shop batch facilities with a
known topology and limited supplies of discrete resources. They presented new gen-
eral sequencing constraints. If a pair of orders (i, i′) Î I has been assigned to the same
equipment j Î Jii′s at stage s Î S, then order i Î I can be processed either before or after
order i′ Î I. Order i may be a direct predecessor (if it is manufactured immediately
before order i′) or a non-direct predecessor of order i′. Therefore, an order i′ Î I is not
constrained to have a single predecessor, but it may feature several depending on its
position in the processing sequence. The notion of a predecessor of order i′ adopted
in their work is more general than that used by Cerdá et al. (1997) and Méndez et al.
(2000) since it does not only apply to the order previously processed but to any one
produced before in the unit assigned to order i. Their formulation achieved a better
objective value and lower computational effort when compared to the model of Pinto
and Grossmann (1995).
Lin et al. (2002) proposed medium-term production scheduling for a multiproduct
batch facility. The methodology was based on the decomposition of the whole sched-
uling period into successive short time horizons. A two-level decomposition model
was proposed to determine the current time horizon and identify those products to
be included. Then a continuous-time formulation for short-term scheduling of batch
processes with multiple intermediate due dates was introduced. An iterative proce-
dure was used until the whole time horizon was completed.
Gupta and Karimi (2003a) developed a two-step MILP approach based on
immediate precedence sequencing for multistage facilities in non-continuous
chemical industries. They incorporated in the model the real and complex features,
such as limited shelf lives of intermediate products, batch splitting at the storage, a
batch filling multiple orders and general product specifications. The proposed for-
mulation was industrially more realistic and computationally more efficient when
compared to previous models in literature. The objective of the formulation was
to minimize tardiness subject to common operational considerations such as both
sequence- and unit-dependent setup times, the initial plant state and order-unit
release times. The developed model was able to solve 90 batches, which made the
model an important and significant contribution towards solving industrially impor-
tant scheduling problems.
Gupta and Karimi (2003b) presented an improved MILP formulation for schedul-
ing multiproduct multistage batch plants. The formulation allowed both sequence-
dependent and unit-dependent setup times and common operational considerations
such as initial plant state and order/unit release times. In contrast to the previous
work, the model minimized tardiness as the scheduling objective because of reduc-
ing equipment ideal times in the beginning of the scheduling period and due to
the computational superiority when compared to minimizing earliness as the objec-
tive function. The proposed model was tested for the multiproduct batch plant with
five stages, 25 units and up to 10 orders. When compared to the previous models,
their formulation required roughly 30% fewer constraints, which resulted in superior
schedules and reduced the computational time roughly by 60%.
Liu and Karimi (2007a) contributed by developing a novel continuous-time for-
mulation for scheduling multistage batch plants with identical parallel units based on
immediate precedence sequencing. In contrast to the existing work, they increased
the solution efficiency by considering each stage as a block of multiple identical
units, thereby eliminating numerous binary variables for assigning batches to spe-
cific units. They demonstrated that a multistage process with identical parallel units
could be scheduled much more efficiently than a process with non-identical parallel
units using a MILP formulation that avoided assigning batches to individual units.
Ferrer-Nadal et al. (2008) proposed a scheduling technique for multistage multi-
product batch plants based on general precedence sequencing. In their formulation,
they assumed non-zero transfer time between processing stages. Most of the existing
mixed-integer linear programming optimization approaches have traditionally dealt
with the batch scheduling problem assuming zero transfer times, and consequently
no synchronization between consecutive processing stages. Synchronization implies
that during the execution of the transfer task, one unit will be supplying the material
whereas the other one will be receiving it, and consequently, no other task can be
simultaneously performed in both units. They demonstrated in their work that ignor-
ing the important role of transfer times may seriously compromise the feasibility of
the scheduling whenever shared units and storage tanks, material recycles, or bidirec-
tional flows of products are to be considered, as it is depicted in Figures 1.13 and 1.14.
1.5.2.2 Models Based on Slot Representation

Liu and Karimi (2007b) presented a scheduling technique based on slot representa-
tion where the task starts and ends at the same slot. They compared and contrasted the
different scheduling models for scheduling of multiproduct multistage batch plants.
U1 A B
U2 B A
Time
U1 U2
FIGURE 1.13 Infeasible schedule resulting from ignoring transfer time between units.
U1 B A
U2 B A
Time
FIGURE 1.14 Feasible schedule by considering transfer time between units.
The formulations were based on unit-slots, stage-slots, process-slots, a variety of slot

arrangements and sequence-modelling techniques and four-index and three-index
binary variables. The authors demonstrated that their models give better initial
RMIP values and were generally faster than the two existing models by Castro and
Grossmann (2005) and Gupta and Karimi (2003a,b). However, more importantly,
they found that no single model was a clear winner in all test problems, and the idea
of using a few competitively best models simultaneously was advised. They also sug-
gested that having several novel formulations and using them simultaneously was an
excellent idea, if somehow all of them could be stopped, once one of them achieved
the desired solution.
Liu and Karimi (2008) developed scheduling models for multistage batch plants
with parallel units and no interstage storage. Little work existed on the scheduling of
multistage multiproduct batch plants with no interstage storage in contrast to plants
with unlimited storage. Furthermore, most of this work had assumed non-identical
parallel units and single (uniform) interstage storage configuration. Many real plants
avoid interstage storage, employ mixes of interstage wait policies such as unlimited
and zero wait, and possess mixes of stages with identical and non-identical parallel
units. Their paper made an important contribution to the batch plantscheduling lit-
erature. Furthermore, it has also addressed several important scheduling objectives
such as makespan, weighted tardiness, weighted earliness and weighted just-in-time.
Their models employed an appropriate and justified mix of slot-based and sequence-
based modelling approaches. This was in contrast to most batch plant scheduling
models that employed monolithic modelling approaches. In particular, it was found
that a four-index slot-based approach worked best for sequencing batches on stages
with non-identical parallel units, while a sequence-based approach was the best for
stages with identical parallel units.
Their work also supported a frequent observation that models with fewer binary
variables or those with tighter MILP relaxations do not always perform better. The
performances of models also differ significantly for different objectives. In particu-
lar, they found four-index models to be better than three-index models for makespan
minimization, but the latter seemed better than the former for weighted just-in-time
scheduling.
Erdirik-Dogan and Grossmann (2008) proposed short-term scheduling for
multiproduct batch plant for identical/non-identical parallel units. The scheduling
technique allowed a task to continue processing on multiple slots. Their motiva-
tion was a real case study at the Dow chemical company. Their formulation catered
for sequence-dependent changeover, mixed intermediate storage and batch split-
ting. In order to handle this problem, a MILP model was presented based on asyn-
chronous time slots. Since the formulation needed prepostulation of slots, it was
computationally expensive for large problems. Therefore, they developed a bi-level
decomposition algorithm where the original problem was decomposed into upper
level sequencing and lower level scheduling and sequencing models. The upper-level
model was a recent planning model where mass balances were aggregated over time
periods and detailed timing constraints were dropped. Solving the upper-level model
gave an upper limit for the objective value and a good prediction for the number of
slots required. The lower-level model was solved using the predicted number of slots
and a subset of products to get the lower bound. The problem was solved iteratively
until the bounds converged.
1.6 SCHEDULING TECHNIQUES FOR CONTINUOUS AND

SEMI-CONTINUOUS MULTISTAGE MULTIPRODUCT PLANTS
Sahinidis and Grossmann (1991) presented cyclic scheduling for multiproduct con-
tinuous parallel line plants. The proposed model resulted in a large-scale mixed
integer nonlinear program (MINLP). A reformulation technique was applied to lin-
earize it in the space of the integer variables. The authors applied their formulation
to the real-world case study of a polymer production plant. In contrast to previous
works, transitions were treated as being sequence dependent, processing times were
discretized and nonlinear costs and inventory considerations were accounted for in
the formulation.
Pinto and Grossmann (1994) presented optimal cyclic scheduling of multistage
continuous multiproduct plants. The method consists of an MINLP sub-problem in
which cycle times and inventory levels are optimized for a fixed sequence, and an
MILP master problem that determines the optimal sequence of production.
McDonald and Karimi (1997) proposed production planning for semi-continuous
parallel line processes. The models were formulated as MILP problems. The objec-
tive function in the model was to minimize the production, inventory and transition
costs. They tested their model for a plant that had two facilities and two units pro-
ducing 34 products. Karimi and McDonald (1997) proposed short-term scheduling
for semi-continuous parallel production lines. The developed models were used for
a single stage multiproduct facility. The two developed models were represented as
M1 and M2. The main differences between the two models were how the slots were
represented in the time horizon. For model (M1), there was no a priori assignment of
slots to time periods. A slot could be of any length and could continue over multiple
time periods. For model (M2), each time period had a pre-designated number of
slots, and each slot was therefore fixed to a particular time period. Through the dif-
ferent case studies it was shown that the model (M2) outperformed the model (M1).
Lamba and Karimi (2002a) developed short-term scheduling of semi-continuous
plants with a composite objective of minimizing transitions and maximizing produc-
tivity. The model catered for resource constraints, such as the structure and capac-
ity of upstream and/or downstream material handling facilities and the availability
of common resources. They considered synchronized slots across all processing
units in the formulation. The formulation resulted in a complex combinatorial prob-
lem with a moderate integrality gap. The solution time drastically increased as the
problem size increased. Lamba and Karimi (2002b) proposed a branch and bound
technique to improve the computational performance of the model they presented
(2002a). It was shown in the paper that the developed algorithm gave an optimal
solution in reasonable CPU time when it was applied to an industrial case study of
a detergent plant.
Lim and Karimi (2003) presented scheduling model for semi-continuous plants
with parallel multistage production line. They considered in their models sequence-
dependent transitions, minimum campaign lengths, inventory costs, safety stock
penalties and backorder penalties which have not been addressed in the previous
literatures. The models are based on the continuous-time domains of parallel lines
using slots that are asynchronous across lines and have variable lengths. It was dem-
onstrated that the use of asynchronous slots can reduce the number of binary vari-
ables significantly and improve the performance tremendously.
Reddy et al. (2004) presented a new continuous-time representation for schedul-
ing of crude oil operations. The formulation resulted in a MILP problem with fewer
constraints and variables compared to the discrete-time representation. They tested
their models for a weekly schedule with a refinery that had eight tanks, three distil-
lation units and two classes of crude oil. It was shown that the formulation gave a
better objective value and better computational performance compared to the previ-
ous models used to schedule refinery operations.
1.7 SCHEDULING TECHNIQUES FOR MULTIPURPOSE
BATCH PLANTS
This section presents a brief overview of the time representation methods that have
been developed and presented in literature in recent years.
1.7.1 Models Based on Discrete-Time Representation

In early formulations, the time horizon in the mathematical models was divided
into a number of time intervals of uniform duration. Literature models based on
1 if task i takes place in unit j at time point p

y(i, j, p) =
0 otherwise
0 H
Time horizon
FIGURE 1.15 Discrete-time representation.
1 if task i takes place in unit j at time point p

y(i, j, p) =
0 otherwise
0 H
Time horizon
FIGURE 1.16 Continuous-time representation.
this approach include those presented by Kondili et al. (1993), Shah et al. (1993),
Pantelides (1993), Dedopoulos and Shah (1995) and Pekny and Zentner (1993).
Events such as the beginning and ending of tasks are associated with the boundaries
of these time intervals. To attain appropriate approximation of the original prob-
lem, it was necessary to use a time interval that was sufficiently small, for example,
the greatest common factor (GCF) of the processing times. This generally leads
to very large combinatorial problems of intractable size, especially for real-world
problems, and hence limits its applications. A detailed comparison of the discrete-
and continuous-time representation was presented by Floudas and Lin (2004). They
showed that discrete approximation of the time horizon leads to suboptimal solutions
and unnecessary increase of the overall size of the resulting mathematical formula-
tion due to the introduction of a large number of binary variables associated with
each discrete time interval. The basic concept of the discrete-time approach is illus-
trated in Figure 1.15.
The main benefit of the discrete-time representation is that it provides a reference
grid of time for all operations competing for shared resources, such as equipment
items. This provides the chance of formulating the various constraints in the sched-
uling problem in a relatively easy and simple manner (Floudas and Lin, 2004). The
continuous-time representation is given in Figure 1.16.
1.7.2 Models Based on Global Event Point Representations

Due to the drawbacks of the discrete-time approach, researchers started developing
continuous-time models in the past decade. In these models, events are potentially
allowed to take place at any point in the continuous domain of time. The different
continuous-time models proposed in the literature for scheduling of multipurpose
batch plants can be broadly classified into three distinct categories: slot-based, global
event–based and unit-specific event-based time representation (Shaik et al., 2006;
Shaik and Floudas, 2007, 2008). In the global event point–time representation, the
starting time associated to tasks at the same event point are similar.
Zhang and Sargent (1996) presented a scheduling formulation based on the
Resource Task Network (RTN) representation. In the RTN, the resources include
feeds, intermediates, utilities, equipment, storage facility, manpower, transportation
facility and cleaning. The resource may be temporally engaged on a task and gener-
ated at the end, or permanently consumed and generated as in the case of feeds and
products. The formulation gave a large MINLP problem later linearized to give a
MILP model. They proposed a technique to linearize the nonlinear terms, which
resulted in a model difficult to solve for complex scheduling problems.
Pinto and Grossman (1997) formulated a scheduling technique that considered
resource constraints such as manpower, electricity and utilities. In order to cater for
the resource constraints, binary variables were assigned and as a result gave a large
subset of big-M constraints, which was difficult to solve simultaneously as a MILP.
To circumvent this problem, they proposed a combined LP-based branch and bound
procedure with disjunctive programming. For a smaller problem of a plastic com-
pounding plant, their formulation reduced the number of enumerated nodes by up to
two orders of magnitude when compared to the direct MILP formulation.
Maravelias and Grossmann (2003) proposed a MILP formulation for scheduling
multipurpose batch plants. In the formulation, the utilities were also considered as
resource constraints. Their model was general and addressed changeover time and
costs, utility limitations and storage used for multiple states. A new class of tighten-
ing inequalities was proposed for tightening the relaxed LP solutions. Castro et al.
(2004) presented a scheduling technique for batch and semi-continuous processes.
The formulation was based on the RTN representation. They considered two case
studies, a continuous and a batch plant. For the case of finite intermediate storage
for a semi-continuous plant, they found a new optimal objective value. The work
extended the work of Castro et al. (2001). The formulation performed better than the
previous models in literature.
1.7.3 Models Based on Slot Time Representation

Schilling and Pantelides (1996) proposed a MILP formulation for short-term sched-
uling of multipurpose batch plants based on the RTN representation. They proposed
a novel branch and bound algorithm that branched both on discrete and continuous
variables to reduce the integrality gap of the formulation.
Sundaramoorthy and Karimi (2005) proposed a MILP formulation for schedul-
ing of multipurpose batch plants. In their formulation, they used a continuous-time
representation whereby the time horizon was divided into unknown slot lengths.
Their motivation was to show that some attempts (Ierapetritou and Floudas, 1998,
1999; Giannelos and Georgiadis, 2002; Maravelias and Grossmann, 2003) at sched-
uling multipurpose batch plants to decouple the task and unit binary variables from
the three-index binary variable (task on unit at event point or slot) in order to reduce
the binary variables required in the formulations did not lead to the intended reduc-
tion in binary variables. Although reducing the number of binary variables in a for-
mulation is generally a desirable modelling objective, it is well known that this does
not guarantee improved solution times. Their model performed better than previous
models in literature. They also commented on the importance of different solvers,
software versions and the types of computers used in evaluating the performance of
the different models in the literature.
Susarla et al. (2010) presented models that used unit specific slots that allowed
tasks to span over multiple slots. Their models also allowed non-simultaneous trans-
fer of material into a unit to get a better schedule. They compared their models with
the different models in literature. They demonstrated that it was difficult to com-
pare the computational time required by their model with that of Shaik and Floudas
(2009) since the model by Shaik and Floudas (2009) needed to specify different
values for the maximum number of time points a task needs to span over to get the
global optimal solution. The computational time required increased with the increas-
ing value of the maximum number of time points a task needs to span over. Their
model performed better than the previous models.
1.7.4 Models Based on Unit-Specific Event-Point Representation

In the unit-specific models, the time horizon is divided into unknown lengths,
where the starting times of tasks at the same event point can take different values.
The sequencing of different tasks is achieved through special big-M constraints.
Ierapetritou and Fluodas (1998) were the first to introduce the concept of the unit-
specific event-point time representation. In the paper, the three-index binary vari-
able y(i, j, n) was separated into unit yv(j, n) and task wv(i, n) binary variables.
Although the model originally claimed its superiority due to both decoupling of
tasks and unit events and non-uniform time grid, later it was shown that the model
efficiency was due to the reduction of events as a result of the non-uniform time
grid representation. The model resulted in a MILP problem which was simpler to
solve than the previous models.
Majozi and Zhu (2001) introduced a new batch plant representation called the
State Sequence Network representation. The SSN was a graphical network repre-
sentation of all the states that exist in the batch plant, and a network was formulated
based on the production recipe. Only states were considered in this network, thereby
eliminating the need for task and unit binary variables as required by the STN for-
mulation. This made it possible to define a single unit binary variable y(s, p) through-
out the formulation. This reduction of binary variable was a great achievement in
solving industrial scale problems in reasonable computational time.
Lin and Floudas (2001) presented simultaneous design, synthesis and scheduling
of multipurpose batch plants. The model considered the trade-offs between capital
costs, revenues and operational flexibility. The formulation resulted in both MILP
and MINLP problems, where a globally optimal solution was achieved for the non-
convex MINLP problems based on a key property that arose due to the special struc-
ture of the resulting models. In the previous models, design of batch plants did not
incorporate scheduling strategies very well, which resulted in over-design or under-

design. In the design phase of the plant, for resources that are involved in the design
are to be used as efficiently as possible, detailed considerations of the plant sched-
uling must be examined. Therefore, it was necessary to consider design, synthesis
and scheduling simultaneously. They considered two case studies to demonstrate the
performance of their model in terms of design, synthesis and scheduling with that of
Xia and Macchietto (1997). A better objective value and CPU time was obtained as
compared to the previous models.
Janak et al. (2004) presented a short-term scheduling technique for multipurpose
batch plant that considered resource constraints and different storage policies. The
formulation was based on the enhancement of the unit-specific model of Ierapetritou
and Floudas (1998). Tasks were allowed to continue over several consecutive event
points in order to monitor resource and storage accurately between the specified lim-
its. Two sets of binary variables were used to cater for the situation of tasks spanning
over multiple time points, one that indicated whether a task starts at each event point
ws(i, n) and another that indicated whether a task ends at each event point wf(i, n).
A continuous variable w(i, n) with 0 and 1 bounds was also employed to indicate
if a task was active at each event point. In the formulation, states and utilization of
resources were considered as tasks, where the sequence and timing of these new
tasks were related to the processing tasks in order to maintain the timing change in
resource level and amount of states. Their model performed well when compared to
the global event model of Maravelias and Grossmann (2003). It was also shown in
the paper that the unit-specific event-driven model required fewer event points when
compared to the global event–point models.
Shaik et al. (2006) presented a comparative study on continuous-time models for
short-term scheduling of multipurpose batch plants. They categorized the various
continuous-time models into slot-based, global event–based and unit-specific event-
based formulations. In the paper, several literature examples were taken to compare
the performance of the various models with the two different objective functions,
maximization of profit and minimization of makespan. The different continuous-
time representations are depicted in Figure 1.17. In the slot-based continuous-time
representation of Figure 1.17a the time horizon is divided into time intervals of
unequal and unknown lengths, and typically tasks need to start and finish at an
event. In the global event–based continuous-time representation, only the start time
of the tasks needs to be at an event point and the events considered are common
across all units (Figure 1.17b). In the unit-specific event-based time representation,
only the start time of each task in a unit has to be an event point, and the time associ-
ated to an event can be different across different units (Figure 1.17c).
For the specific instance of the four tasks on the three units as depicted in
Figure 1.17, the slot-based representation required five slots (or six events), the global
event–based representation required four events, while the unit-specific event-based
representation required only two events. They concluded that the unit-specific event-
based models resulted in smaller problem sizes compared to both slot-based and
global event–based models and were computationally superior.
Janak et al. (2007) developed a new robust optimization approach for scheduling
of batch plants under uncertainty in processing time, demand and price of a product.
U1
U2
UN
0 12 3 4 5 H–1 H
(a)
Four events
U1
U2
UH
0 1 2 3 4 H–1 H
(b)
n Two events
U1
n n+1
U2
n
UH
0 1 2 3 4 5 H–1 H
(c)
FIGURE 1.17 Different continuous-time representations. (a) Slot-based continuous time.

(b) Globalevent–based continuous-time. (c) Unit-specific event-based continuous-time.
The uncertainty was considered by a known probability distribution function. The

work extended the robust optimization approach by Lin et al. (2004) for schedul-
ing under bounded uncertainty and bounded and symmetric uncertainty to consider
uncertainty by a known probability distribution function. As it was shown by Lin
et al. (2004), the optimal solution of an MILP program may be severely infeasible,
that is, one or more constraints are violated substantially if the nominal data are
slightly perturbed. This makes the nominal optimal solution questionable.
Different probability distribution functions including uniform distribution,
normal distribution, difference of normal distributions, general discrete distribu-
tion, binomial distribution and Poisson distribution were considered. The general
discrete distribution was less conservative, which gave the best objective value.
The binomial and Poisson distribution always gave the most conservative results
with the worst objective function values. The uniform distribution was less con-
servative than the bounded and normal distribution. Their formulation results
showed that the approach provided an effective way to address scheduling prob-
lems under uncertainty, giving a reliable schedule.
Janak and Floudas (2008) presented an improved unit-specific event-based
continuous-time approach that addressed closing the integrality gap. A new set of
rigorous constraints was developed to reduce the complexity and highly combinato-
rial nature of scheduling problems. The method was based on four distinct steps.
First, the state task network was used to examine the problem under consideration to
determine practical limitations, as well as relationships inherent to the problem. For
instance, tasks that were continued at the first or the last event point could be fixed
and tasks that could not take place at these event points could be excluded. Also,
constraints could be added in the model to represent one or more of a group of tasks
that must occur at the beginning or end of the time horizon. Tasks that were inher-
ent in the process could be fixed, such as tasks that must occur at consecutive event
points due to storage limitations.
In the second step, in order to tighten the relaxed LP solution, the following tight-
ening constraints were developed: (1) The summation of the processing times of the
tasks assigned to a specific unit should be less than or equal to the time horizon.
(2) Constraints that are expressed in (1) were further tightened by defining relevant
minimum and maximum time horizons for each processing unit. (3) The summation
of processing times of all tasks starting in a unit at event point (n) or greater must be
less than or equal to the amount of time remaining. (4) The summation of the pro-
cessing times of all tasks finishing in a unit before event point (n) must be less than
or equal to the finishing time of the unit at event point (n) − 1.
At the third step, the following bounding constraints were developed: (1) In every
unit, the sum of the binary variables over all event points must be between some
lower and upper bounds. (2) The sum of durations of processing times for each task in
each unit over the entire time horizon is placed in between lower and upper bounds.
(3) The amount of material of a state s used in the entire time horizon is also placed in
between lower and upper bounds. In order to get rigorous bounds, supporting sched-
ule problems were solved. These problems sought to minimize and maximize the
sum of the variables represented in the bounding constraints alternatively, subjected
to bounds on the scheduling objective functions. In order to solve these support-
ing problems, a valid lower and upper bound for the scheduling problems objective
function needed to be determined. For the case of profit maximization, a valid lower
bound could be found from any feasible solution. Once these bounds were obtained,
the original scheduling problem, without any timing and sequence constraints, was
solved subjected to lower and upper bounds on the scheduling objective function,
where the new objective function was changed in each supporting problem to either
minimization or maximization of a different sum of variables.
At the last step, new constraints were generated using a reformulation linearization
technique (RLT) developed by Sherali and Adams (1994). The technique could be
used to generate a tight LP representation and strong valid inequalities for the MILP.
The computational results obtained showed that the proposed formulation was more
effective and efficient than the model proposed by Wang and Guignard (2002).
Shaik and Floudas (2008) presented a unit-specific continuous-time approach
for short-term scheduling of batch plants using the RTN representation. The model
addressed the FIS storage policies without considering storage as a separate task. For
unit-specific event-based models, the RTN had not been explored in the literature.
In their paper, they used the RTN representation for short-term scheduling based on
an improved version of the model of Ierapetritou and Floudas (1998). The perfor-
mance of the model was evaluated along with the continuous models of Ierapetritou
and Floudas (1998), Giannelos and Georgiadis (2002), Castro et al. (2001, 2004)
and Sundaramoorthy and Karimi (2005). Benchmark examples using two different
objective functions, maximization of profit and minimization of makespan, were
considered. It was observed that the slot-based and global event–based models always
required the same number of event points while the unit-specific event-based model
required fewer event points to solve the problem. Their model required fewer event
points and as a result gave smaller problem sizes and was computationally supe-
rior. The unit-specific event-based models by Shaik and Floudas (2008) sometimes
required more event points for finite storage case compared to the unlimited storage
case, but the number of events required was still fewer compared to the slot-based
and global event–based models. Their model handles the FIS operational philosophy
accurately, which is overlooked by the pervious unit-specific event-based models.
Shaik and Floudas (2009) developed a novel unified modelling approach for
short-term scheduling of multipurpose batch plants. In the paper, the necessity of
allowing tasks to take place over multiple event points to get a better schedule was
shown. In order to understand the necessity of tasks spanning over multiple time
points, the sequence constraints for different tasks in different units by Ierapetritou
and Floudas (1998) was examined.
( )
tu ( i, j, n + 1) ³ tf ( i¢, j¢, n ) - H 1 - w ( i¢, n )
This constraint states that the consuming task at the current event should start after
the end time of the consuming task at the current event point that processes the
same state, which need not be true if there, is sufficient material for the consuming
task to start production. The model by Janak et al. (2004) also allowed tasks to span
over multiple time points but it resulted in poor LP relaxations and required a large
number of constraints, non-zeros and CPU time. The reason is that the model was
originally developed for resource constraints; it did not reduce well in terms of prob-
lem statistics to the case of no resources. The model by Shaik and Floudas (2009)
had the following features: (1) It could handle problems with resource constraints by
allowing task to take place over multiple event points and (2) merged both schedul-
ing problems with and without resource constraints into a common framework and
effectively reduced to the simple case of no resources.
1.8 CONCLUSIONS
From this literature survey, it is clear that there are two broad categories of schedul-
ing techniques, graphical and mathematical programming. Graphical methods are
used to schedule batch plants running only on the NIS or UIS operational poli-
cies. Although this method seems to be good compared to the mathematical models,
the generation of an S-graph is complex and is suited to the case where storage is

not definite and only one batch of a product is produced in the given time horizon.
The computational time of mathematical techniques depends on how the time is
represented in the model. In early formulation, the time horizon was divided into a
number of time intervals of uniform duration, which led to very large combinatorial
problems of intractable size when modelling practical problems.
In this literature review, the performance of different continuous-time models
proposed in the literature were compared. It is observed that both the slot-based and
global event–based models always require the same number of event points, while
the unit-specific event-based models require fewer event points to solve a problem
to global optimality. Thus, the unit-specific event-based models result in smaller
problem sizes compared to both slot-based and global event–based models and are
computationally superior. The most recent and advanced scheduling technique based
on unit-specific event-based continuous-time representation is that of Shaik and
Floudas (2009). They showed a need to allow tasks to take place over multiple events
in order to find the global optimal solutions. Their model is also general and handles
the different operational philosophies.
The models by Ierapetritou and Floudas (1998), Majozi and Zhu (2001) and Shaik
and Floudas (2007) are capable of solving scheduling problems for batch plants with-
out utility resources and for the unlimited intermediate storage case. The RTN-based
model of Shaik and Floudas (2008) is suitable for batch plants without resource and
with unlimited and dedicated finite storage policies. However, both these models do
not allow task to take place over multiple events, which has been demonstrated to
be necessary for obtaining a global optimal solution, by Shaik and Floudas (2009).
The model of Janak et al. (2004, 2008) is relevant for batch plants with resource
constraints. The model allows tasks to span over multiple time points, but compu-
tationally it does not perform well for problems without resource constraints. The
proposed model by Shaik and Floudas (2009) bridges this gap in literature and uni-
fies both problems with and without resources into a generic common framework
and efficiently reduces to the simple case of no resources. The formulation uses a
three-index binary variable and has fewer big-M terms when compared to Janak
et al. (2004, 2008) hence results to improved LP relaxations.
The recent model developed by Susarla et al. (2010) is based on unit-specific
slots where special sequence constraints are involved in the models in order to take
care of timings when the material balance around a unit and storage is performed,
which makes the model similar to unit-specific events. The model does not require
specifying the maximum number of event points a task should span over to get
the global optimal objective value, unlike the model of Shaik and Floudas (2009).
Consequently, it avoids unnecessary iterations. The added feature of their model is in
handling non-simultaneous transfer of materials into a unit in order to get a globally
optimal solution.
The approach of allowing a task to continue processing on multiple event points
results in an increase of the number of binary variables and event points required. As
a result, the models require excessive computational time and it was difficult to solve
scheduling problems for a long time horizon. All the mathematical models used for
scheduling of multipurpose batch plants require pre-specification of the number of
time points required and the number of time points related to the optimal solution is
found after a number of iterations. This results in an increase in the computational
time, limiting the practical application, especially for a long time horizon. These
drawbacks necessitate the development of a new scheduling technique and introduc-
tion of a new method for the prediction of the optimal number of time points for
scheduling of multipurpose batch plants.
REFERENCES
Castro, P., Barbosa-Póvoa, A.P.F.D., Matos, H., 2001. An improved RTN continuous-time
formulation for the short-term scheduling of multipurpose batch plants. Industrial and
Engineering Chemistry Research. 40, 2059–2068.
Castro, P.M., Barbosa-Povóa, A.P., Matos, H.A., Novais, A.Q., 2004. Simple continuous-time
formulation for short-term scheduling of batch and continuous processes. Industrial
and Engineering Chemistry Research. 43, 105–118.
Castro, P.M., Grossmann, I.E., 2005. New continuous-time MILP model for the short-term
scheduling of multistage batch plants. Industrial and Engineering Chemistry Research.
44, 9175–9190.
Cerdá, J., Henning, G.P., Grossmann, I.E., 1997. A mixed-integer linear programming model
for short-term scheduling of single-stage multiproduct batch plants with parallel lines.
Industrial and Engineering Chemistry Research. 36, 1695–1707.
Dedopoulos, I.T., Shah, N., 1995. Optimal short-term scheduling of maintenance and pro-
duction for multipurpose plants. Industrial and Engineering Chemistry Research. 34,
192–201.
Erdirik-Dogan, M., Grossmann, I.E., 2008. Slot-based formulation for the short-term sched-
uling of multistage, multiproduct batch plants with sequence-dependent changeovers.
Ferrer-Nadal, S., Capón-Garćia, E., Méndez, C.A., Puigjaner, L., 2008. Material transfer
operations in batch scheduling. A critical modeling issue. Industrial and Engineering
Chemistry Research. 47, 7721–7732.
Floudas, C.A., Lin, X., 2004. Continuous-time versus discrete-time approaches for scheduling
of chemical processes: A review. Computers and Chemical Engineering. 28, 2109–2129.
Giannelos, N.F., Georgiadis, M.C., 2002. A simple new continuous-time formulation for
short-term scheduling of multipurpose batch processes. Industrial and Engineering
Gupta, S., Karimi, I.A., 2003a. Scheduling a two-stage multiproduct process with limited
product shelf life in intermediate storage. Industrial and Engineering Chemistry
Research. 42, 490–508.
Gupta, S., Karimi, I.A., 2003b. An improved MILP formulation for scheduling multiprod-
uct, multistage batch plants. Industrial and Engineering Chemistry Research. 42,
2365–2380.
Hui, C.W., Gupta, A., 2000. A novel MILP formulation for short-term scheduling of multi-
stage multi-product batch plants. Computers and Chemical Engineering. 24, 2705–2717.
Ierapetritou, M.G., Floudas, C.A., 1998. Effective continuous-time formulation for short-term
scheduling: 1. Multipurpose batch processes. Industrial and Engineering Chemistry
Research. 37, 4341–4359.
Ierapetritou, M.G., Floudas, C.A., 1999. Effective continuous time formulation for short-term
scheduling. 1. Multiple intermediate due dates. Industrial and Engineering Chemistry
Research. 38, 3446–3461.
Janak, S.L., Floudas, C.A., 2008. Improving unit-specific event based continuous time
approaches for batch processes: Integrality gap and task splitting. Computers and
Chemical Engineering. 32, 913–955.
Janak, S.L., Lin, X., Floudas, C.A., 2004. Enhanced continuous-time unit-specific event-based
formulation for short-term scheduling of multipurpose batch processes: Resource con-
straints and mixed storage policies. Industrial and Engineering Chemistry Research.
43, 2516–2533.
Janak, S.L., Lin, X., Floudas, C.A., 2007. A new robust optimization approach for scheduling
under uncertainty: II. Uncertainty with known probability distribution. Computers and
Karimi, I.A., McDonald, C.M., 1997. Planning and scheduling of parallel semicontinuous
processes. II. Short-term scheduling. Industrial and Engineering Chemistry Research.
36, 2701–2714.
Kondili, E., Pantelides, C.C., Sargent, R.W.H., 1993. A general algorithm for short-term sched-
uling of batch operations. I. MILP formulation. Computers and Chemical Engineering.
17, 211–227.
Lamba, N., Karimi, I.A., 2002a. Scheduling parallel production lines with resource con-
straints. I. Model formulation. Industrial and Engineering Chemistry Research. 41,
779–789.
Lamba, N., Karimi, I.A., 2002b. Scheduling parallel production lines with resource con-
straints. II. Decomposition algorithm. Industrial and Engineering Chemistry Research.
41, 790–800.
Lim, M.F., Karimi, I.A., 2003. Resource-constrained scheduling of parallel production
lines using asynchronous slots. Industrial and Engineering Chemistry Research. 42,
6832–6842.
Lin, X., Floudas, C.A., 2001. Design, synthesis and scheduling of multipurpose batch plants
via an effective continuous-time formulation. Computers and Chemical Engineering.
25, 665–674.
Lin, X., Floudas, C.A., Modi, S., Juhasz, N.M., 2002. Continuous-time optimization approach
for medium-range production scheduling of a multiproduct batch plant. Industrial and
Lin, X., Janak, S.L., Floudas, C.A., 2004. A new robust optimization approach for scheduling
under uncertainty: I. Bounded uncertainty. Computers and Chemical Engineering. 28,
1069–1085.
Liu, Y., 2006. Scheduling of multi-stage multi-product batch plants with parallel units. PhD
thesis. Singapore: Department of Chemical and Biomolecular Engineering, National
University of Singapore.
Liu, Y., Karimi, I.A., 2007a. Scheduling multistage, multiproduct batch plants with noniden-
tical parallel units and unlimited intermediate storage. Chemical Engineering Science.
62, 1549–1566.
Liu, Y., Karimi, I.A., 2007b. Novel continuous-time formulations for scheduling multi-stage
batch plants with identical parallel units. Computers and Chemical Engineering. 31,
1671–1693.
Liu, Y., Karimi, I.A., 2008. Scheduling multistage batch plants with parallel units and no
interstage storage. Computers and Chemical Engineering. 32, 671–693.
Majozi, T., Friedler, F., 2006. Maximization of throughput in a multipurpose batch plant
under a fixed time horizon: S-graph approach. Industrial and Engineering Chemistry
Research. 45, 6713–6720.
Majozi, T., Zhu, X.X., 2001. A novel continuous-time MILP Formulation for multipurpose
batch plants. Industrial and Engineering Chemistry Research. 40, 5935–5949.
Maravelias, C.T., Grossmann, I.E., 2003. New general continuous-time state-task network
formulation for short-term scheduling of multipurpose batch plants. Industrial and
McDonald, C.M., Karimi, I.A., 1997. Planning and scheduling of parallel semicontinuous
processes. I. Production planning. Industrial and Engineering Chemistry Research.
36, 2691–2700.
Méndez, C.A., Cerdá, J., 2000. Optima scheduling of a resource-constrained multiproduct
batch plant supplying intermediates to nearby end-product facilities. Computers and
Méndez, C.A., Cerdá, J., Grossmann, I.E., Harjunkoski, I., Fahl, M., 2006. State of-the-art
review of optimization methods for short-term scheduling of batch processes.
Computers and Chemical Engineering. 30, 913–946.
Méndez, C.A., Henning, G.P., Cerdá, J., 2000. Optimal scheduling of batch plants satisfying
multiple product orders with different due-dates. Computers and Chemical Engineering.
24, 2223–2245.
Méndez, C.A., Henning, G.P., Cerdá, J., 2001. An MILP continuous-time approach to
short-term scheduling of resource-constrained multistage flowshop batch facilities.
Pantelides, C.C., 1993. Unified frameworks for optimal process planning and scheduling.
In D.W.T. Rip-pin, J.C. Hale, J. Davis (Eds.), Proceedings of the Second International
Conference on Foundations of Computer-Aided Process Operations (pp. 253–274).
Crested Butte, CO.
Pekny, J.F., Zentner, M.G., 1993. Learning to solve process scheduling problems: The role
of rigorous knowledge acquisition frameworks. In D.W.T. Rippin, J.C. Hale, J. Davis
(Eds.), Proceedings of the Second International Conference on Foundations of
Computer-Aided Process Operations (pp. 275–309). Crested Butte, CO.
Pinto, J., Grossmann, I.E., 1997. A logic based approach to scheduling problems with resource
constraints. Computers and Chemical Engineering. 21, 801–818.
Pinto, J.M., Grossmann, I.E., 1994. Optimal cyclic scheduling of multistage continuous mul-
tiproduct plants. Computers and Chemical Engineering. 18, 797–816.
Pinto, J.M., Grossmann, I.E., 1995. A continuous-time mixed-integer linear programming
model for short-term scheduling of multistage batch plants. Industrial and Engineering
Reddy, P.C.P., Karimi, I.A., Srinivasan R., 2004. A new continuous-time formulation for
scheduling crude oil operations. Chemical Engineering Science. 59, 1325–1341.
Reklaitis, G.V., 1991. Perspectives of scheduling and planning of process operations. Process
Systems and Engineering PSE’91. Montebello, Quebec, Canada.
Reklaitis, G.V., 1982. Review of scheduling of process operations. AIChE Symposium. 78,
119–133.
Sahinidis, N.V., Grossmann, I.E., 1991. MINLP model for cyclic multiproduct scheduling on
continuous parallel lines. Computers and Chemical Engineering. 15, 85–103.
Sanmarti, E., Friedler, F., Puigjaner, L., 1998. Combinatorial technique for short term sched-
uling of multipurpose batch plants based on schedule-graph representation. Computers
and Chemical Engineering. 22, 847–850.
Sanmarti, E., Holczinger, T., Puigjaner, L., Friedler, F., 2002. Combinatorial framework for
effective scheduling of multipurpose batch plants. AIChE Journal. 48, 2557–2570.
Schilling, G., Pantelides, C., 1996. A simple continuous-time process scheduling formulation
and a novel solution algorithm. Computers and Chemical Engineering. 20, 1221–1226.
Shah, N., Pantelides, C.C., Sargent, R.W.H., 1993. A general algorithm for short-term schedul-
ing of batch operations II. Computational issues. Computers and Chemical Engineering.
17, 229–244.
Shaik, M.A., Floudas, C.A., 2007. Improved unit-specific event-based model continuous-time
model for short-term scheduling of continuous processes: Rigorous treatment of stor-
age requirements. Industrial and Engineering Chemistry Research. 46, 1764–1779.
Shaik, M.A., Floudas, C.A., 2008. Unit-specific event-based continuous time approach for
short-term scheduling of batch plants using RTN framework. Computers and Chemical
Engineering. 32, 260–274.
Shaik, M.A., Floudas, C.A., 2009. Novel unified modeling approach for short term schedul-
ing. Industrial and Engineering Chemistry Research. 48, 2947–2964.
Shaik, M.A., Janak, S.L., Floudas, C.A., 2006. Continuous-time models for short-term sched-
uling of multipurpose batch plants: A comparative study. Industrial and Engineering
Sherali, H.D., Adams, W.P., 1994. A hierarchy of relaxations and convexhull character-
izations for mixed-integer zero–one programming problems. Discrete and Applied
Mathematics. 52, 83–106.
Sundaramoorthy, A., Karimi, I.A., 2005. A simpler better slot-based continuous-time formu-
lation for short-term scheduling in multipurpose batch plants. Chemical Engineering
Science. 60, 2679–2702.
Susarla, N., Li, J., Karimi, I.A., 2010. A novel approach to scheduling of multipurpose batch
plants using unit slots. AICHE Journal. 56, 1859–1879.
Wang, S., Guignard, M., 2002. Redefining event variables for efficient modelling of continu-
ous-time batch processing. Annals of the Operation Research. 116, 113–126.
Xia, Q.S., Macchietto, S., 1997. Design and synthesis of batch plants MINLP solution based
on a stochastic method. Computers and Chemical Engineering. 21, S697–S702.
Zhang, X., Sargent, R.W.H., 1996. The optimal operation of mixed production facilities-a
general formulation and some approaches for the solution. Computers and Chemical
2 Modelling for
Effective Solutions
Reduction of Binary
Variables
2.1 INTRODUCTION
Scheduling of multipurpose batch plants has gotten considerable attention in the
last two decades where considerable scheduling models have been proposed. The
research direction has been to improve the existing scheduling models in terms of
model size, computational time and optimal objective value. This chapter also has
the same objective of presenting a better scheduling model that is compact and com-
putationally efficient. The model is rigorous in handling the different intermediate
storage policies and unit wait time. The binary variables, continuous variables, con-
straints and computational time required are reduced by almost 41%, 15%, 43% and
42%, respectively, when compared to the recent rigorous unit-specific event-based
model for short-term scheduling of batch plants.
2.2 NECESSARY BACKGROUND
The use of optimization techniques for scheduling of multipurpose batch plants has
been receiving considerable attention by researchers from academics and industries
due to the significant economic benefit they bring to industries. The features of multi-
purpose batch plants allow flexibility in the production of small-volume, high value–
added products by sharing common resources like processing equipment, storage
tank, utilities, manpower, etc. However, the same flexibility of multipurpose batch
plants result production scheduling to be a challenging task, which leads to the exis-
tence of different scheduling models. All these developed models can be generally
grouped into three major categories (Sundaramoorthy and Maravelias, 2011). In the
first category, every product follows a sequential production pattern where every task
consumes and produces one material and batch mixing and splitting are not permis-
sible. Most of the models considered a simplified approach of allocating batches to
processing units to optimize different objectives (minimizing makespan, earliness
and tardiness of due date, inventory, etc.) without considering storage and utility con-
straints. The sequence-based or precedence-based representation uses either direct
precedence or indirect precedence sequencing of pairs of tasks on units. In the direct
precedence models (Pinto and Grossmann, 1997; Méndez and Cerdá, 2000; Gupta
and Karimi, 2003; Liu and Karimi, 2007), batch i′ has to follow immediately after
batch i is processed in the same unit. The number of binary variables required in
27
these models increases quadratically with the number of batches. Having this draw-
back, indirect precedence models (Méndez et al., 2001; Méndez and Cerdá, 2004;
Ferrer-Nadal et al., 2008) have been developed to reduce the binary variables required
by allowing batch i′ to be processed not necessarily immediately after batch i iharjun
the same unit. However, it is not trivial to identify subsequent tasks, sequence-depen-
dent costs and prevent certain processing sequences in indirect precedence models.
Kopanos et al. (2009, 2010) have overcome this issue efficiently by combining direct
and indirect precedence decision variables in the same model. Models developed in
the second category are based on network processing approach where batch mixing
and splitting is allowed and storage and utility constraints are also catered for. This
chapter focuses and gives a comprehensive review of models developed under this
category, since it is the interest and motivation of this work. Lastly, it is the hybrid
processing environment which constitutes both sequential and network processing
system (Sundaramoorthy and Maravelias, 2011; Velez and Maravelias, 2013, 2014).
A considerable amount of research work exists for addressing scheduling prob-
lem for network-based batch processing system. In order to capture the essence of
time in batch processing, researchers adopted discrete- and continuous-time rep-
resentation. In the discrete-time representation, the time horizon is divided into
equal intervals, and the starting and finishing time of a task should be aligned at the
interval (Kondili et al., 1993; Shah et al., 1993; Dedopoulos and Shah, 1995). This
time representation allows easy and straightforward modelling of due dates, equip-
ment unavailability, time-dependent utility pricing and holding and backlog costs.
However, models based on discrete-time representation have the following draw-
backs: (1) model accuracy; (2) large discrete-time models due to sequence-dependent
changeover or non-instantaneous material transfer times between units and (3) can-
not handle variable processing times. Continuous-time models are accurate, handle
different magnitude of task processing times without increasing the model size and
can easily be integrated with lower-level control layer. The major drawback is that
they turn to be more complex when handling sequencing of tasks than their discrete-
time counterparts. The different continuous-time models proposed in the literature
for scheduling of multipurpose batch plants can be broadly classified into two dis-
tinct categories: single-grid and multi-grid time representations (Floudas and Lin,
2004; Shaik et al., 2006; Shaik and Floudas, 2008, Harjunkoskia et al., 2014). The
term time-grid designates all time representations in scheduling models that employ
time slots/periods/points/events. The starting and finishing time of tasks are mapped
onto one or more time reference grids (Harjunkoskia et al., 2014).
In the single-grid-based models, the starting time associated with different
tasks at the same grid point is identical (Schilling and Pantelides, 1996; Zhang
and Sargent, 1996; Pinto and Grossmann, 1997; Maravelias and Grossmann, 2003;
Sundaramoorthy and Karimi, 2005). On the other hand, in multiple-grid time repre-
sentation the time horizon is divided into unknown lengths where the starting times of
tasks in different units at the same grid point can take different values. The sequenc-
ing of different tasks is achieved through special big-M constraints (Ierapetritou and
Floudas, 1998; Majozi and Zhu, 2001; Janak et al., 2004, 2007; Shaik et al. 2006;
Janak and Floudas, 2008; Shaik and Floudas, 2009; Susarla et al., 2010; Vooradi
and Shaik, 2012). These models perform better in terms of objective value and
Modelling for Effective Solutions 29
computational time due to the non-uniform time grid representation as a result of

reducing the number of grid points required for solving the scheduling problem.
In all the previous models that are based on multiple-grid time representation, the
sequence constraints that are core in the formulation result in unconditional sequenc-
ing, which implies that different tasks in different units are always aligned without
the actual material transfer between these units. The drawback was first addressed
by Seid and Majozi (2012) by formulating sequence constraints between different
tasks in different units to be conditional, thereby leading to improved objective value
and reduced event points to solve the scheduling problem compared to the previously
developed models. Recently, Vooradi and Shaik (2013) identified that the model by
Seid and Majozi (2012) is actually partially conditional, improved the concept fur-
ther and developed a rigorous conditional sequencing that results in better optimal
objective value and reduced event points required in some of the case studies they
investigated. However, the model size and complexity increased considerably, hence
this contribution. The reader can get a more comprehensive and detailed review on
production scheduling in processing industry from Harjunkoskia et al. (2014).
The main objective of this chapter is to present a rigorous conditional sequenc-
ing formulation based on STN representation with significantly reduced size and
complexity compared to published work in literature. The chapter is organized as
follows. In Section 2.3, the general problem statement for scheduling of network-
based batch facilities is given. The developed mathematical model is presented in
Section 2.4 followed by case studies from literature in Section 2.6 in order to com-
pare the efficiency of the developed model in this chapter. Finally, in Section 2.7,
conclusions are drawn from this work.
2.3 PROBLEM STATEMENT
A typical scheduling problem of multipurpose batch plants can be stated as follows.
Given: (1) the production recipe that shows the steps and sequences the raw materi-
als should follow to become products, (2) the size of the units and their suitability to
perform the different tasks, (3) storage capacity and their suitability to store different
material and (4) the time horizon, using these data it is required to determine (1) the
maximum achievable profit of the plant or the minimum makespan if throughput is
fixed a priori and (2) a production schedule related to the optimal resource utilization.
2.4 MODEL FORMULATION
The mathematical formulation presented in this work is based on STN representa-
tion and contains the following constraints.
2.4.1 Allocation Constraints
Constraint (2.1) allocates one task at any given event point p in a unit. This constraint
is used when a unit is multipurpose and capable of performing more than one task.
åy (i, j, p ) £ 1, "i Î I , j Î J, pÎP (2.1)

iÎI j
2.4.2 Capacity Constraints
Constraint (2.2) enforces the amount of material inside the unit for process should be
between the lower and upper capacity of the unit.
Vi,Lj y ( i, j, p ) £ mu ( i, j, p ) £ ViU, j y ( i, j, p ) , "i Î I , j Î J, p Î P (2.2)
2.4.3 Material Balance around a Storage Unit

Constraint (2.3) determines the amount of intermediate material in the storage at
event point p.
q ( s, p ) = q ( s, p - 1) + ååmu (i, j, p - 1) - ååmu (i¢, j¢, p ) ,

iÎI sp jÎJ i i ¢ÎI sc j ¢ÎJ i¢
"s Î S i , p Î P, p ³1 (2.3)
Constraint (2.4) is used for material balance around the intermediate storage at
the first event point, that is, p = 1.
q ( s, p ) = Q0 ( s ) - ååmu (i¢, j¢, p ) ,

j ¢ÎJ i¢
"s Î S i , p Î P, p =1 (2.4)
i ¢ÎI sc
Constraint (2.5) states that the amount of product material stored at each event
point p is the amount stored at the previous event point p − 1 and product state s pro-
duced by tasks at event point p.
q ( s, p ) = q ( s, p - 1) + ååmu (i, j, p ) ,
iÎI sp jÎJ i
s Î S p, p Î P, p ³1 (2.5)

Constraint (2.6) is used for the amount of product material stored at event point
p = 1.
q ( s, p ) = Q0 ( s ) + ååmu (i, j, p ) , s Î S p, p Î P, p =1 (2.6)

iÎI sp jÎJ i
Constraint (2.7) states that the amount of raw material in storage at each event
point p is the difference between the amount stored at the previous event point p − 1
and material state s consumed by the consuming tasks at event point p.
q ( s, p ) = q ( s, p - 1) - ååmu (i¢, j¢, p ) ,

j ¢ÎJ i¢
s Î Sr , p Î P, p ³1 (2.7)
i ¢ÎI sc
Constraint (2.8) is used for the amount of raw material stored at event point p = 1.
q ( s, p ) = QO ( s ) - ååmu (i, j, p ) , s Î S r, p Î P, p =1 (2.8)

iÎI sc jÎJ i
2.4.4 Duration Constraints
Constraints (2.9) and (2.10) describe the duration constraints. In this mathemati-
cal formulation, the duration constraint is based on processing unit which allows
for the reduction of variables and constraints, hence the compactness of the model.
Constraints (2.9) and (2.10) are only used for tasks that produce stable intermediate
states, where the states can wait for a while before they can be consumed by the
subsequent tasks.
tu ( j, p ) ³ tp ( j, p ) + åa (i, j ) y (i, j, p ) + åb (i, j ) mu (i, j, p ) , "j Î J , pÎP

(2.9)
iÎI j iÎI j
tu ( j, p ) £ tp ( j, p ) + åa (i, j ) y (i, j, p ) + åb (i, j ) mu (i, j, p )

iÎI j iÎI j
+ åUW (i, j ) , "j Î J , pÎP (2.10)

iÎI j
For tasks that produce unstable states, which require states to be consumed imme-
diately by subsequent tasks, as encountered in zero-wait condition, Constraints (2.9)
and (2.11) are used.
æ ö
tp ( j, p ) £ tu ( j, p ) + å a ( i, j ) y ( i, j, p ) + å å
b ( i, j ) mu ( i, j, p ) + H ç 1 - y ( i, j, p ) ÷ ,
ç ÷
iÎI j iÎI j è iÎI j ø
"i Î I ZW , j Î J, pÎP (2.11)
2.4.5 Sequence Constraints for Processing Units

Constraint (2.12) states that the starting time of the processing unit at event point
p is after the completion time of the same unit at previous event point p − 1. Basing
duration and sequence constraints on units rather than tasks allows the avoidance
of sequencing constraints for different tasks in different units, as encountered in
other formulations for unit-specific event representation. For example, in the pre-
vious literature models based on unit-specific event-point time representation if
the unit performs four different tasks, we would need 8 variables for defining the
starting and finishing times for the tasks and 16 constraints relating to sequence
constraints for the same task, as well as different tasks, in the same unit. However,
in the proposed model we need only two variables relating to the starting and fin-
ishing time for a unit and one constraint to sequence this unit, thereby leading to
a compact model.
tu ( j, p ) ³ tp ( j, p - 1) , "j Î J , pÎP (2.12)

2.4.6 Sequence Constraints for Different Tasks in Different Units

The constraints under this section are used to conditionally sequence the consuming
and producing units. In the previous unit-specific event-point-based formulations,
the producing and consuming units are aligned even if there is no material transfer
and this shortcoming was first observed and improved to be conditional by Seid and
Majozi (2012) and later improved to make it rigorously conditional by Vooradi and
Shaik (2013). However, the method proposed by Vooradi and Shaik (2013) results in
increased number of binary variables and big-M constraints. The model proposed in
this manuscript is significantly reduced the number of binary variables and big-M
constraints.
2.4.6.1 Sequence Constraints for Different Tasks in Different Units

for Stable State That Can Wait after It Is Produced
Constraints (2.13) through (2.16) are used to conditionally align the consuming
units with the producing units for stable intermediate states. They trace whether the
amount of material that is consumed in a unit comes from storage or producing units
or from both producing units and storage.
ååq ( s, i¢) mu (i¢, j¢, p ) £ q ( s, p - 1) + ååmex ( s, j, j¢, p - 1) , "s Î S

i ¢ÎI sc j ¢ÎJ i¢ j ¢ÎJ sc jÎJ sp
cw
, pÎP

(2.13)
Constraint (2.14) ensures that the amount of material that is consumed in the unit
is greater than the amount of same material coming from all the producing units,
since some of the material could be taken from storage.
q ( s, i¢ ) mu ( i, j¢, p ) ³ åmes ( s, j, j¢, p ) ,

jÎJ sp
"s Î S cw , j¢ Î J sc , pÎP (2.14)

Constraint (2.15) ensures that the amount of state s sent from producing unit to
consuming units is limited by the amount of state produced by the producing unit.
d ( s, i ) mu ( i, j, p ) ³ åmes ( s, j, j¢, p ) , "s Î S cw , j Î J sp , pÎP (2.15)

j ¢ÎJ sc
Constraint (2.16) is used to trigger the binary variable to take a value of one if
indeed material, that is, state s is transferred between the consuming and producing
units, otherwise they take a value of zero.

åmes ( s, j, j¢, p ) £ V
s
U
j t ( j, j¢, p ) , "s Î S cw , j Î J sp , j¢ Î J sc , p Î P (2.16)

2.4.6.2 Sequence Constraints for Different Tasks in Different

Units for Unstable States with Zero Wait
Constraints (2.17) through (2.19) have the same concept as Constraints (2.14) through
(2.16), but apply to zero-wait states.
q ( s, i¢ ) mu ( i¢, j¢, p ) = åmes ( s, j, j¢, p ) , "s Î S zw , j¢ Î J sc , p Î P (2.17)

jÎJ sp
d ( s, i ) mu ( i, j, p ) = åmes ( s, j, j¢, p ) , "s Î S zw , j Î J sp , pÎP (2.18)

j ¢ÎJ sc

åmes ( s, j, j¢, p ) £ V
s
U
j t ( j, j¢, p ) , "s Î S zw , j Î J sp , j¢ Î J sc , p Î P (2.19)

2.4.6.3 Sequence Constraint That Links the Starting Time of the

Consuming Units with the Finishing Time of Producing Units
Constraint (2.20) ensures that the starting time of the consuming unit is later than
the finishing time of the producing unit if material transfer occurs between the units.
( )
tu ( j¢, p ) ³ tp ( j, p - 1) - H 1 - t ( j, j¢, p - 1) , "s Î S cw , j Î J sp , j¢ Î J sc , p Î P
(2.20)
Constraint (2.21) is used to prevent real-time storage violation if state s is used by

consuming units from storage after it was produced and stored.
æ ö
tu ( j¢, p ) ³ tp ( j, p - 2 ) - H ç 1 -
ç å
y ( i, j, p - 2 ) ÷ ,
÷
è iÎI j ø
"s Î S , j Î J s , j¢ Î J s , p Î P
cw p c
(2.21)

Constraints (2.21) and (2.22) are used for zero-wait condition which forces the
consuming unit j to start immediately after the producing unit j′, if consuming unit j
receives materials from producing unit j′. Otherwise, these two constraints are relaxed.
æ ö
tu ( j¢, p ) £ tp ( j, p - 1) - H ç 2 -
ç
è i
å
y ( i, j, p ) - t ( j, j¢, p - 1) ÷ ,
÷
ø
¢
"s Î S , j Î J s , j Î J s , p Î P
zw c p
(2.22)

2.4.7 Sequence Constraints for FIS Policy

Constraint (2.23) states that if the amount of material produced from the producing
units exceeds the capacity of the storage, then the variable ma ( s, j, p ) should take a
value indicating that there is not enough storage for the material produced from the
producing unit j. Consequently, Constraint (2.24) enforces the binary variable to take
a value of one. However, if there is enough storage capacity to store material that is
produced from the producing unit j, then ma ( s, j, p ) takes a value of zero, which
ultimately forces the binary variable x( j, p) to take the value of zero.
ååd ( s, i ) mu (i, j, p ) + q ( s, p ) £ Q
iÎI sp jÎJ sp
U
( s ) + åma ( s, j, p )
jÎJ sp
(2.23)

åma ( s, j, p ) £ V
s
U
j x ( j, p ) , "s Î S cw , j Î J sp , pÎP

(2.24)
Constraint (2.25) is used to ensure that the consuming unit finishes earlier com-
pared to the finishing time of the producing unit so that the producing unit can send
its material immediately to the available consuming unit if this is necessary at a
particular event point for an optimum overall schedule. In this instance, both the
transfer binary variable, t ( j, j¢, p ), and the binary variable signifying absence of
storage, x ( j, p ), take a value of one. Otherwise, the constraint is relaxed.
¢ ( ¢ )
tp ( j , p ) £ tp ( j, p ) + H 2 - t ( j, j , p ) - x ( j, p ) , "s Î S , j Î J s , j Î J s
cw
¢ c p (2.25)
2.4.8 Time Horizon Constraints

Constraints (2.26) and (2.27) ensure that the starting and finishing times of the pro-
cessing units are within the time horizon of interest.
tu ( j, p ) £ H , "p Î P, j ÎJ (2.26)
tp ( j, p ) £ H , "p Î P, j ÎJ (2.27)
2.4.9 Objective Function
Equation 2.28 is the objective function for maximization of profit.

maximize pro = å price ( s ) q ( s, p ) ,
sÎS p
"p = P

(2.28)
In the event that the demand for product is stipulated a priori, Constraint (2.29) is
necessary and the objective function becomes minimization of makespan as shown
in Constraint (2.30). Worthy of note is the fact that for makespan minimization prob-
lem, H is a variable instead of a parameter.

åq ( s, p ) ³ demand ( s ) ,
sÎS p
"p = P

(2.29)
minimize makespan = H (2.30)

2.5 ILLUSTRATIVE EXAMPLE
This literature example is taken from Shaik and Floudas (2009). Two products
are produced from four raw materials. The facility has a heater, three reactors, a
separator and two mixers. The heater and reactors are suitable for more than one
task. Figure 2.1 shows a unit performing multiple tasks, multiple units suitable for
a task, dedicated units for specific tasks and recycle of material which are common
characteristics of multipurpose batch plants. The corresponding scheduling data for
this example are shown in Tables 2.1 and 2.2. The price for product 1 and 2 is $5/m.u.
Feed 3
S8
Feed 1 0.75 Product 2

H1 {il} H1 {i2} R3 {i5} 0.4
S1 S9 S13
{HR} {HR} {RR1, RR2}
Int 6
0.25 0.6
S3 Int 1 S6 Int 4 Int 7 S10
0.5 0.4 0.4
Int 5 Product 1
R2 {i4} S {i6} 0.5 M {i7}
S5 S7 S12
{RR1, RR2} {SR} 0.4 {MR1, MR2}
Int 3
0.5 0.1 0.2
S4
Int 2 S11
Feed 2 Feed 4
R1 {i3}
S2
{RR1, RR2}
FIGURE 2.1 STN representation for the Example.
TABLE 2.1
Batch Size and Processing Time Data
Task Lable i Unit Lable j α(i, j) β(i, j) Vi,Lj - Vi,Uj

Example 1
Heating-1 i1 Heater HR 0.667 0.00667 0–100
Heating-2 i2 Heater HR 1 0.01 0–100
Reaction-1 i3 Reactor-1 RR-1 1.333 0.01333 0–100
i3 Reactor-2 RR-2 1.333 0.00889 0–150
i4 Reactor-2 RR-2 0.667 0.00445 0–150
i5 Reactor-2 RR-2 1.333 0.00889 0–150
Separation i6 Separator SR 2 0.00667 0–300
Mixing i7 Mixer-1 MR-1 1.333 0.00667 20–200
i7 Mixer-2 MR-2 1.333 0.00667 20–200
TABLE 2.2
Initial Inventory and Storage Size Data
States Storage Capacity (m.u.) Initial Inventory (m.u.)
1 UL AA
2 UL AA
3 100 AA
4 100 0
5 300 0
6 150 50
7 150 50
8 UL 0
9 150 0
10 150 0
11 UL AA
12 UL 0
13 UL 0
2.5.1 Results and Discussion

The computational statistics for the different models are given in Table 2.3. In
maximization of profit for the time horizons of 8 and 12 h, all models get the same
objective value. However, for the time horizons of 10 and 16 h the proposed model
and the model by V.S (2013) give better objective values compared to other mod-
els. The improvement in objective values by the proposed model and V.S (2013)
is attributed to the fact that these models rigorously align sequence constraints
between consuming and producing tasks. So far, the recent model by V.S (2013)
provides better objective values compared to other previous literature models
based on continuous-time representation. However, the model size in terms of
binary variables, continuous variables and constraints has increased consider-
ably in order to rigorously align sequence constraints. V.S (2013) has pointed out
that future work should address reducing the model size so that it can be easy to
solve. With this motivation, this work reduced the model size without compromis-
ing the objective values. For example, for the time horizon of 16 h the proposed
model reduced the binary variables required by 49.4%, continuous variables by
16.9% and constraints by 41.8% when compared to V.S (2013). It is difficult to
compare the model statistics of this work with models by SLK2 (2010), V.S (2012)
and S&M (2012) since these models do not rigorously align sequence constraints
which lead to suboptimal results. As a consequence, for the rest of the examples
in this paper the results obtained by this work are compared to the recent model
by V.S (2013). The Gantt chart for the time horizon of 16 h for this example is
given in Figure 2.2.
Modelling for Effective Solutions
TABLE 2.3
Computational Results for Example 1
Model Events Binary Variables Continuous Variables Constraints MILP CPU Time (s)
H = 8
SLK2 (2010) 7 102 655 1017 1583.44 —
V.S (2012) 6 41 256 608 1583.44 —
S&M (2012) 5 107 418 889 1583.44 —
V.S (2013) 5 234 408 1209 1583.44 4
This work 5 125 354 726 1583.44 1.9
H = 10
SLK2 (2010) 9 136 853 1428 2337.36 —
V.S (2012) 9 137 457 1403 2345.3 —
S&M (2012) 6 131 516 1093 2345.3 —
V.S (2013) 7 358 576 1785 2358.2 1211.7
This work 7 185 486 1052 2358.2 894.9
H = 12
SLK2 (2010) 8 119 754 1249 3041.26 —
V.S (2012) 7 52 302 302 3041.26 —
S&M (2012) 7 155 604 1297 3041.26 —
V.S (2013) 7 358 576 1785 3041.26 0.9
This work 7 185 486 1052 3041.26 0.8
H = 16
SLK2 (2010) 11 170 1051 1786 4241.5 —
V.S (2012) 11 181 571 1755 4262.8 —
S&M (2012) 10 227 868 1909 4261.9 —
V.S (2013) 10 544 828 2649 4262.8 104.1
This work 10 275 684 1541 4262.8 95.9
37
(Continued)
38
TABLE 2.3 (Continued)
Computational Results for Example 1
d(s12) = 100, d(s13) = 200 m.u. and M = 50
SLK2 (2010) 9 136 859 1448 13.366 —
V.S (2012) 9 74 394 977 13.366 —
S&M (2012) 7 155 605 1299 13.366 —
Understanding Batch Chemical Processes

V.S (2013) 7 358 577 1644 13.366 0.8
This work 7 185 487 976 13.366 0.6
d(s12) = 250, d(s13) = 250 m.u. and M = 50
SLK2 (2010) 11 820 170 1057 17.025 —
V.S (2012) 10 159 514 1591 17.025 —
V.S (2013) 10 544 829 2442 17.025 1.8
This work 10 275 685 1429 17.025 0.9
Note: SLK2, Susarla et al.; V.S, Vooradi and Shaik; S&M, Seid and Majozi.

Units
J6 150 60.27 117.26 65.98 38.04

i7 i7 i7 i7 i7
J5 122.47 65.98 38.04

i7 i7 i7
J4 184.4 158.8 50 33.24

i6 i6 i6 i6
J3 50.59 101.2 100 150 75 61.5 87.98 30.4 28.54

i3 i4 i3 i4 i5 i5 i5 i5
J2 46 92 100 50 25 38.5 33.24 20.32 19.07

i3 i4 i5 i4 i5 i5 i4 i5 i5
J1 96.6 100 100 100 100 88 16.6 50.71 50.71

i1 i2 i1 i2 i2 i2 i1 i2 i2
2 4 6 8 10 12 14 16
Time (h)
FIGURE 2.2 Gantt chart for the time horizon of 16 h.
2.6 CASE STUDIES
This section discusses case studies on scheduling and prediction of time points for
multipurpose batch plants for the readers to have an in-depth understanding of the
application of scheduling models.
2.6.1 Case Study I
This case was first studied by Kondili et al. (1993), and it is one of the most com-
mon examples that appear in literature. The STN representation which illustrates
the steps and sequences the raw materials should follow to produce two products is
given in Figure 2.3. The scheduling data required to solve this problem is given in
Tables 2.4 and 2.5. The price of products 1 and 2 is $10/m.u.
2.6.1.1 Results and Discussions

The computational results for this case study are given in Table 2.6. The proposed
model considerably reduced the binary variables required by 38.9%–44% when it
is compared to the literature model by V.S (2013) for the different scenarios consid-
ered. The continuous variables and constraints required by this work are also lesser
Product 1
s7
40%
40% 60%
s1 Heating s5 Reaction 1 s8
10%
60%
s9 Separation
s6
90%
80%
50%
s2 Reaction 2 Reaction 3 s10
50% 20% Product 2

(s3) (s4)
Feed c
FIGURE 2.3 STN representation for Case Study I.
TABLE 2.4
Batch Size and Processing Time Data for Case Studies I and II
Task Lable i Unit Lable j α(i, j) β(i, j) ViL, j - ViU, j
Case Study I
Heating i1 Heater HR 0.667 0.00667 0–100
i2 Reactor-2 RR-2 1.334 0.01665 0–80
i3 Reactor-2 RR-2 1.334 0.01665 0–80
i4 Reactor-2 RR-2 0.667 0.00833 0–80
Separation i5 Separator SR 1.3342 0.00666 0–200
Case Study II
Task-1 i1 Unit-1 U-1 2 0 0–260
Task-2 i2 Unit-2 U-2 3 0 0–140
i2 Unit-3 U-3 2 0 0–120
Task-3 i3 Unit-4 U-4 3 0 0–120
i3 Unit-5 U-5 2 0 0–140
Task-4 i4 Unit-1 U-1 1 0 0–1
i4 Unit-2 U-2 3 0 0–1
TABLE 2.5
Initial Inventory and Storage Size Data for Case Studies I and II
Case Study I Case Study II
Storage Initial Storage Initial
States Capacity (m.u.) Inventory (m.u.) Capacity (m.u.) Inventory (m.u.)
1 UL AA UL AA
2 UL AA 10 0
3 UL AA 10 0
4 100 0 UL 0
5 200 0 UL 0
6 150 0 — —
7 200 0 — —
8 UL 0 — —
9 UL 0 — —
10 — — — —
11 — — — —
12 — — — —
13 — — — —
by 12%–17% and 35%–70%, respectively, when compared to the literature model

of V.S (2013) for the different scenarios considered. The proposed model solved the
case study in much shorter CPU time. For the last scenario, the model is solved by
allowing the processing unit to store its produced material in the next time points
when the unit is idle. By doing so, a better optimal value is obtained.
2.6.2 Case Study II
This example is taken from Vooradi and Shaik (2013) and presented in this work
for comparison. The STN representation is depicted in Figure 2.4. Two products are
produced from one raw material using different production paths. The necessary
scheduling data for this example are given in Tables 2.1 and 2.2. The price of prod-
ucts 1 and 2 is $1/m.u.
2.6.2.1 Results and Discussion

The computational results for Case Study II are shown in Table 2.7. For the time hori-
zon of 9 and 16 h, both models obtained the same optimal objective value. Even for
this small problem considerable reduction in model size is obtained by the proposed
model. For instance, for 16 h time horizon the binary variables, continuous variables
and constraints required are reduced by 32.4%, 15.6%, 26.5%, respectively, using the
proposed formulation, when compared to the literature model. The computation time
required for both instances is lower when using the proposed model. Gantt chart that
shows the type of task performed in each unit, the starting and finishing time for a
task and batch size for the time horizon of 16 h, is depicted in Figure 2.5. It is worth
42
TABLE 2.6
Computational Results for Case Study I
H = 8
V.S (2013) 4 108 231 615 1498.57 0.23
This work 4 66 203 374 1498.58 0.15
H = 10
V.S (2013) 5 146 292 806 1962.69 0.44
This work 5 86 251 485 1962.69 0.3
H = 12
V.S (2013) 6 184 353 997 2658.52 1.3

This work 6 106 299 596 2658.53 0.5
H = 16
V.S (2013) 8 260 475 1379 3738.38 42.1
This work 8 146 395 822 3738.39 21.8
d(s7) = d(s10) = 200 m.u. and M = 50
V.S (2013) 9 298 537 1436 19.34 5.6
This work 9 166 444 931 19.34 1.4
d(s7) = 500, d(s10) = 400 m.u. and M = 100
V.S (2013)
(Δn = 1) 21 876 1387 4459 47.687 40,000
This work 21 490 1356 2493 47.6835 40,000
Product 1
S1 Task 1 S2 Task 2 S3 Task 3 S4
Product 2
Task 4 S5
FIGURE 2.4 STN representation for Case Study II.
TABLE 2.7
Computational Results for Case Study II
Binary Continuous
Model Events Variables Variables Constraints MILP CPU Time (s)
H = 9
V.S (2013) 7 147 265 691 386 14.4
This work 7 102 228 514 386 5.7
H = 16
V.S (2013) 10 222 379 1018 1115 292
This work 10 150 321 784 1115 211
Units
J5
120 130 120 130 120 130
i3 i3 i3 i3 i3 i3
J4 120 120 120

i3 i3 i3
J3 120 120 120 120 120 120

i2 i2 i2 i2 i2 i2
J2 130 130 130 1

i2 i2 i2 i4
J1 260 110 250 120 250 120 1 1 1 1

i1 i1 i1 i1 i1 i1 i4 i4 i4 i4
2 4 6 8 10 12 14 16
Time (h)
FIGURE 2.5 Gantt chart for the time horizon of 16 h for Case Study II.
to note that industrial scale scheduling problems require large variables and con-
straints; the reduction of the model size by the proposed technique will significantly
improve the computational efficiency.
2.7 CONCLUSIONS
Compact MILP scheduling formulation for scheduling of multipurpose batch plants
based on multi-grid time-point representation is presented in this work. The model
is based on rigorous conditional sequencing between consuming and producing
tasks while achieving smaller model size. Through illustrative examples, this work
demonstrated that the proposed model leads to considerable reduction in terms of
binary, continuous variables and constraints required, when compared to the recent
rigorous conditional sequencing model that is based on unit-specific event points as
published in literature.
NOMENCLATURE
Set
I Tasks
I zw Tasks with zero wait
I j Tasks that can be performed in unit j
I sp All tasks which produce material state s
I sc All tasks which consume material state s
S Any material state s
Si Intermediate state s
S p Product state s
S r Raw material state s
cw
S Any state s which can wait in a unit after it is produced
S zw Zero wait state s
P Event points
J Processing units
Ji Units suitable to execute task i
J sc State s consuming units
J sp State s producing units
Parameters
Vi,Lj Minimum batch size for task i in unit j

U
V
i, j Maximum batch size for task i in unit j
U
V
j Maximum batch size capacity for unit j
H Time horizon
Q0 ( s ) Initial amount available for material state s
QU ( s ) Maximum storage capacity for material state s

a ( i, j ) Coefficient for fixed processing time for task i in unit j
b ( i, j ) Coefficient for variable processing time for task i in unit j
UW ( i, j ) Waiting time for task i in unit j after it is produced
q ( s, i ) Stoichiometric coefficient of reactant
d ( s, i ) Stoichiometric coefficient of product
price ( s ) Selling price for product state s
demand ( s ) Market demand for product state s
Binary Variables
y ( i, j, p ) Binary variable to denote task i performed in unit j at event point p
t ( j, j¢, p ) Binary variable to denote if material transfer between the consuming
unit jʹ and producing unit j occurs
x ( j, p ) Binary variable to denote the amount of material produced by unit j is
beyond the storage capacity and needs to be consumed immediately by
the consuming units
Positive Continuous Variables

mu ( i, j, p ) Amount of batch size for task i performed in unit j at event point p
q ( s, p ) Amount of material state s stored at event point p
tu ( j, p ) Starting time of unit j at time event p
tp ( j, p ) Finishing time of unit j at time event p
ma ( s, j, p ) The amount of material state s produced by unit j is beyond the
storage capacity and needs to be consumed by the consuming
units immediately
mex ( s, j, j ¢, p ) Amount of material state s sent from the producing unit j′ to the
consuming unit j
Profit Profit obtained form selling products
mekspan The total length of the schedule, i.e. when all the jobs have fin-
ished processing
REFERENCES
Dedopoulos, I.T., Shah, N., 1995. Optimal short-term scheduling of maintenance and pro-
duction for multipurpose plants. Industrial Engineering and Chemical Research. 34,
192–201.
operations in batch scheduling. A critical modelling issue. Industrial Engineering and
Chemical Research. 47, 7721–7732.
of chemical processes: A review. Computer and Chemical Engineering. 28, 2109–2129.
Gupta, S., Karimi, I.A., 2003. Scheduling a two-stage multiproduct process with limited prod-
uct shelf life in intermediate storage. Industrial Engineering and Chemical Research.
42, 490–508.
Harjunkoskia, I., Maravelias, C.T., Bongers, P., Castro, P.M., Engell, S., Grossmann, I.E.,
Hooker, J., 2014. Scope for industrial applications of production scheduling models and
solution methods. Computer and Chemical Engineering. 62, 161–193.
scheduling: 1 Multipurpose batch processes. Industrial Engineering and Chemical
Research. 37, 4341–4359.
Janak, S.L., Floudas, C.A., 2008. Improving unit-specific event based continuous-time
approaches for batch processes: Integrality gap and task splitting. Computer and
Janak, S.L., Lin, X., Floudas, C.A., 2004. Enhanced continuous-time unit-specific event-based
formulation for short-term scheduling of multipurpose batch processes: Resource con-
straints and mixed storage policies. Industrial Engineering and Chemical Research.
43, 2516–2533.
Janak, S.L., Lin, X., Floudas, C.A., 2007. A new robust optimization approach for scheduling
under uncertainty: II Uncertainty with known probability distribution. Computer and
Kondili, E., Pantelides, C.C., Sargent, R.W.H., 1993. A general algorithm for short- term sched-
uling of batch operations I. MILP formulation. Computer and Chemical Engineering.
17, 211–227.
Kopanos, G.M., Laínez, J.M., Puigjaner, L., 2009. An efficient mixed-integer linear program-
ming scheduling framework for addressing sequence-dependent setup issues in batch
plants. Industrial Engineering and Chemical Research. 48, 6346–6357.
Kopanos, G.M., Méndez, C.A., Puigjaner, L., 2010. MIP-based decomposition strategies for
large-scale scheduling problems in multiproduct multistage batch plants: A benchmark
scheduling problem of the pharmaceutical industry. European Journal of Operation
Research. 207, 644–655.
Liu, Y., Karimi, I.A., 2007. Novel continuous-time formulations for scheduling multi-stage
batch plants with identical parallel units. Computer and Chemical Engineering. 31,
1671–1693.
bach plants. Industrial Engineering and Chemical Research. 40, 5935–5949.
Maravelias, C.T., Grossmann, I.E., 2003. New general continuous-time state-task net-
work formulation for short-term scheduling of multipurpose batch plants. Industrial
Engineering and Chemical Research. 42, 3056–3074.
Méndez, C.A., Cerdá, J., 2000. Optimal scheduling of a resource-constrained multiproduct
batch plant supplying intermediates to nearby end product facilities. Computer and
Méndez, C.A., Cerdá, J., 2004. Short-term scheduling of multistage batch processes subject
to limited finite resources. Computer Aided Chemical Engineering. 15B, 984–989.
Computer and Chemical Engineering. 25, 701–711.
Pinto, J., Grossmann, I.E., 1997. A logic based approach to scheduling problems with resource
constraints. Computer and Chemical Engineering. 21, 801–818.
and a novel solution algorithm. Computer and Chemical Engineering. 20, 1221–1226.
Seid, R., Majozi, T., 2012. A robust mathematical formulation for multipurpose batch plants.
Chemical Engineering Science. 68, 36–53.
Shah, N., Pantelides, C.C., Sargent, R.W.H., 1993. A general algorithm for short-term schedul-
ing of batch operations II Computational issues. Computer and Chemical Engineering.
17, 229–244.
Shaik, M.A., Floudas, C.A., 2008. Unit-specific event-based continuous-time approach for
short-term scheduling of batch plants using RTN framework. Computer and Chemical
Shaik, M.A., Floudas, C.A., 2009. Novel unified modeling approach for short term schedul-
ing. Industrial Engineering and Chemical Research. 48, 2947–2964.
uling of multipurpose batch plants: A comparative study. Industrial Engineering and
Chemical Research. 45, 6190–4209.
Science. 60, 2679–2702.
Sundaramoorthy, A., Maravelias, C.T., 2011. A general framework for process scheduling.
AICHE Journal. 57, 695–710.
Susarla, N., Li, J., Karimi, I., 2010. A novel approach to scheduling of multipurpose batch
Velez, S., Maravelias, C.T., 2013. Mixed-integer programming model and tightening methods
for scheduling in general chemical production environments. Industrial Engineering
and Chemical Research. 52, 3407–3423.
Velez, S., Maravelias, C.T., 2014. Advances in mxed-integer programming methods for
chemical production scheduling. Annual Review of Chemical and Biomolecular
Vooradi, R., Shaik, M.A., 2012. Improved three-index unit-specific event-based model for
short-term scheduling of batch plants. Computer and Chemical Engineering. 43, 148.
Vooradi, R., Shaik, M.A., 2013. Rigorous unit-specific event-based model for short-term
scheduling of batch plants using conditional sequencing and unit-wait times. Industrial
Zhang, X., Sargent, R.W.H., 1996. The optimal operation of mixed production facilities-a
general formulation and some approaches for the solution. Computer and Chemical
3 Methods to Reduce
Computational Time
Prediction of Time Points
3.1 INTRODUCTION
This chapter presents a mathematical technique for the prediction of the optimal
number of time points in short-term scheduling of multipurpose batch plants. The
mathematical formulation is based on the state sequence network (SSN) representa-
tion. The developed method is based on the principle that the optimal number of time
points depends on how frequently the critical unit is used throughout the time horizon.
In the context of this work, a critical unit refers to a unit that is most frequently used
and it is active for most of the time points when it is compared to other units. A linear
model is used to predict how many times the critical unit is used. In conjunction
with knowledge of recipe, this information is used to determine the optimal number
of time points. The statistical R-squared value obtained between the predicted and
actual number of optimal time points in all the problems considered was 0.998, which
suggests that the developed method is accurate in determining the optimal number
of time points. Consequently this avoids costly computational times due to iterations.
In the model by Majozi and Zhu (2001), the sequence constraint that pertains to tasks
that consume and produce the same state, the starting time of the consuming task at
time point p must be later than the finishing time of the producing task at the previ-
ous time point p − 1. This constraint is relaxed by the proposed models if the state is
not used at the current time point p. This relaxation gives a better objective value as
compared to previous models. An added feature of the proposed models is their abil-
ity to exactly handle fixed intermediate storage (FIS) operational philosophy, which
has proven to be a subtle drawback in published scheduling techniques.
3.2 MOTIVATION
In the continuous-time representation, the optimal number of time points which gives
the optimal objective value is found through iteration. This is done by increasing the
number of time points at each iteration by one until the objective value converges.
The objective value may not change with an increment of one additional time point,
but may change with an increment of two or more. For example in Case Study I in this
chapter, where duration constraints depend on batch size, for a time horizon of 36 h,
the objective value for both time points 11 and 12 is 445.5. The iteration is stopped
at this point giving an optimal objective value of 445.5. However, if the time points
are increased by one, a better objective value of 447 is obtained, so that the optimal
49
number of time points is not 11 but 13. This indicates that the iteration method of
obtaining the optimal number of time points, where the criterion is to stop when the
solution does not improve by adding one time point to the previous one, is subject
to a suboptimal solutions and needs to be verified by further addition of time points.
Again, for Case Study I where duration is fixed, which is not dependent on batch
size, for a time horizon of 168 h the objective value is 3,525 at time points 73 and
requires more than 40,000 s of CPU time. At time points 74, the objective value is
3,550 and requires a CPU time of more than 40,000 s. At time points 75, the objective
value is 3550. Increasing the number of time points beyond 74 does not improve the
objective value. As a result, the optimal number of time points is 74. Consequently,
a time horizon of 168 h needs 74 time points. In the iteration method of getting the
optimal objective value, the CPU time required is the sum of the CPU times of each
iteration. This becomes computationally costly as the time horizon increases. For
complicated problems where each iteration takes a day, a number of days will be
required to obtain the objective value which is not desirable for batch plants where
it is usually a norm to schedule on a daily or weekly basis. Moreover, process shifts
might necessitate a schedule revisit in the order of hours, thereby militating against
this iterative procedure.
In the scheduling of multipurpose batch plants, the following are given: (1) the pro-
duction recipe that indicates the sequence of unit processes whereby the raw materi-
als are changed into products, (2) the capacity of a unit and the type of tasks the unit
can perform, (3) the maximum storage capacity for each material and (4) the time
horizon of interest.
Using the given data, it is required to determine (1) the maximum achievable
profit of the plant, (2) the minimum makespan if throughput is given and (3) a pro-
duction schedule related to the optimal resource utilization.
3.4 MATHEMATICAL MODEL USING STATE

SEQUENCE NETWORK
It is important to explain the effective state in the SSN representation because
it renders the opportunity to reduce the number of binary variables. If a process
requires multiple raw materials to make a particular product, then it is a fact that
if one of the raw materials is fed, then all the other feed materials also exist to
make the product. By noting this, it is easy to see that only one of the states needs
to be defined as an effective state. The choice of effective state is not unique; how-
ever, once the choice of effective state has been made, it should remain consistent
throughout the formulation. Effective states are considered in defining the binary
variables. The mathematical model presented in this chapter entails the follow-
ing constraints. For more explanation on effective states and SSN representation,
the reader can refer to the paper by Majozi and Zhu (2001). Presented here is the
condensed mathematical formulation. Detailed formulation on the subject can be
found in Chapter 2.
Methods to Reduce Computational Time 51
3.4.1 Allocation Constraints
å *
y ( sin, j , p ) £ 1, "j Î J , pÎP (3.1)
sin , j ÎSin ,j
Vsin, j y ( sin, j , p ) £ mu ( sin, j , p ) £ Vsin, j y ( sin, j , p ) , "p Î P, j Î J , sin, j Î Sin, J

L U
(3.2)
3.4.3 Material Balance for Storage
qs ( s, p ) = qs ( s, p - 1) - år sc
sin , j ÎSin
sc
sin , j mu ( sin, j , p ) + å sp
sin , j Î Sin
sin , j mu ( sin, j , p - 1) ,
rsp
,J ,J
"p Î P, s Î S (3.3)
( ) (
qs s p , p = qs s p , p - 1 + ) å sin , j mu ( sin, j , p ) , "p Î P, s Î S
rsp p p
(3.4)
spp
sin , j Îsin ,J
3.4.4 Duration Constraints
t p ( sin, j ¢ , p ) ³ tu ( sin, j ¢ , p ) + t ( sin, j ¢ ) y ( sin, j ¢ , p ) ,
"j Î J , p Î P, sin, j Î Sinsp, J , sin, j ¢ Î Sinsc, J (3.5)
3.4.5 Sequence Constraints
The two subsections address the proper allocation of tasks in a given unit that ensures
the starting time of a new task to be later than the finishing time of the previous task.
3.4.5.1 Same Task in Same Unit
tu ( sin, j , p ) ³ t p ( sin, j , p - 1) , "j Î J , p Î P, sin, j Î Sin*, j (3.6)
3.4.5.2 Different Tasks in Same Unit
( )
tu ( sin, j , p ) ³ t p sin¢ , j , p - 1 , "j Î J , p Î P, sin, j ¹ sin¢ , j , sin, j , sin¢ , j ,Î Sin*, j

(3.7)
( ) ( ) ( (
t p sinusp, j , p - 1 ³ tu sinusc, j, p - H 1 - y sinusp, j , p - 1 , ))
"j Î J , p Î P, sinusc, j Î Sinusc, j , sinusp, j Î Sinusp, j (3.8)
3.4.6 Different Tasks in Different Units

3.4.6.1 If an Intermediate State Is Produced from One Unit
( )
r sinsp, j mu ( sin, j , p - 1) £ qs ( s, p ) + V jU t ( j, p ) "j Î J , p Î P, sin, j Î Sispn, j

(3.9)
((
tu ( sin, j ¢ , p ) ³ tp ( sin, j , p ) - H 2 - y ( sin, j , p - 1) - t ( j, p ) , ))
"j Î J , p Î P, sin, j Î S sp
in, J , sin, j ¢ Î Ssc
in, J (3.10)
3.4.6.2 If an Intermediate State Is Produced from More than One Unit
år sc
sin , j ÎSin
sc
sin , j mu ( sin, j , p )
,J
£ qs ( s, p - 1) + år sp
sin , j ÎSin
sp
sin , j mu ( sin, j , p - 1) t ( j, p ) , "j Î J , pÎP (3.11)
,J
(
tu ( sin, j ¢ , p ) ³ t p ( sin, j , p - 2 ) - H 1 - y ( sin, j , p - 2 ) , )
"j Î J , p Î P, sin, j Î Sinsp, J , sin, j ¢ Î Sinsc, J (3.12)

3.4.7 Sequence Constraint for Fixed Intermediate Storage Policy
år sp
sin , j ÎSin
ps
sin , j mu ( sin, j , p - 1) + qs ( s, p - 1) £ QS U + åV (1 - x ( s, p ))
jÎJ s
U
j
,J
"j Î J , p Î P, s Î S (3.13)
(
tp ( sin, j ¢ , p - 1) £ tp ( sin, j , p - 1) + H 2 - y ( sin, j ¢ , p ) - y ( sin, j , p - 1) + H x ( s, p ) ) ( )
in, J , sin, j ¢ Î S sc
in, J

(3.14)
3.4.8 Storage Constraints
qs ( s, p ) £ QS U + å u (s sp
sin , j Îsin
in, j , p ) "s Î S, p Î P, jÎJ (3.15)
,J
u ( sin, j , p ) £ rsp
sin , j mu ( sin, j , p - 1) + u ( sin, j , p - 1) "p Î P, jÎJ (3.16)

æ ö
u ( sin, j , p ) £ V jU ç 1 -
çç å y ( sin, j , p ) ÷ "p Î P,
÷÷
jÎJ (3.17)
è sin, j ÎSin, j ø
*

tu ( sin, j , p ) £ H , "sin, j Î Sin, J , p Î P, j Î J (3.18)
t p ( sin, j , p ) £ H , "sin, j Î Sin, J , p Î P, jÎJ (3.19)

3.4.10 Calculation of the Total Duration a Unit

Is Active within the Time Horizon
The knowledge of this value, that is the total duration a unit is active over a time
horizon, is fundamental in the prediction of the optimal number of time points. The
following two sections show the calculation of this value in the case of fixed and
variable duration constraints, respectively.
3.4.10.1 Duration Constraints Are Fixed

Constraint (3.20) is an important constraint which is used to predict optimal num-
ber of time points. The equation is used in the model when the model is solved as a
linear programming maximization (LP MAX) and a linear programming minimi-
zation (LP MIN). In solving LP MAX and LP MIN, the binary variable y(sin,j, p) is
considered as a continuous variable with a value between 0 and 1. Constraint (3.20)
expresses that the total duration a unit is active is limited by the time horizon, the
sum of mean processing times of previous tasks and the sum of mean processing
times of all subsequent tasks leading to a product as prescribed by the recipe.
ååt ( s
p sin , j
in, j ) y ( sin, j , p ) £ H - å PT
sin , j ¢ ÎSin
t ( sin, j ¢ ) - å ST
sin , j ¢ ÎSin
t ( sin, j ¢ ) ,
,J ,J
"j Î J , p Î P, Sin, j ¹ sin, j ¢ (3.20)
In the beginning, the parameter å PT

sin , j¢ ÎSin ,J
t(sin, j ¢ ) caters for the situation where
the unit j must wait for all the previous tasks to finish before it can start processing.
If there is an intermediate at the beginning of the time horizon so that unit j can start
its operation, then the parameter becomes 0. The parameter ST
t(sin, j ¢ ) states å sin , j¢ ÎSin ,J
that at the end of its operation unit j gives its intermediate to the subsequent tasks
which lead to a product (the shortest path of subsequent tasks lead to a product), so
that the unit is not active at this time. If the unit is the one producing a product,
then the term becomes 0; this assumption is used not to underestimate time points
because of underutilization of a unit. Constraint (3.20) tightens the model better than
one simply stating that the sum of the durations of all tasks suitable in each unit
should be less than the scheduling time horizon without loss of any generality.
3.4.10.2 Duration Constraints Dependent on Batch Size

Constraint (3.21) is similar to (3.20), but for the situation where duration is a function
of batch size.
ååt ( s
p sin , j
in, j ) y ( sin, j , p ) + b ( sin, j ) mu ( sin, j , p ) £ H - å
sin , j ¢ ÎSiPT
t ( sin, j ¢ ) - å ST
sin , j ² ÎSin
t ( sin, j ¢ ) ,
n ,J ,J

"j Î J , p Î P, sin, j ¹ sin, j ¢ (3.21)
Constraints (3.1) through (3.8), (3.10), (3.13) through (3.15), (3.18) through (3.21)
together with the objective function given in Constraint (3.22) constitute LP MAX,
and LP MIN comprises of Constraints (3.1) through (3.8), (3.13) through (3.15), (3.18)
through (3.21), (3.20) through (3.24). Sequence Constraint (3.10) and (3.15) for LP
MAX and LP MIN can be expressed as
(
tu ( sin, j ¢ , p ) ³ tp ( sin, j , p - 1) - H 1 - y ( sin, j , p - 1) )
"j Î J , p Î P, sin, j Î Sinsp, J , sin, j ¢ Î Sinsc, J (3.10ʹ)
qs ( s, p ) £ QS max "s Î S, pÎP (3.15ʹ)
Model ML2 constitutes Constraints (3.1) through (3.22). For the model ML2,
Constraints (3.20) and (3.21) are expressed as
å å t(s *
p sin , j ÎSin
in, j ) y ( sin, j , p ) £ H , "j Î J , pÎP (3.20ʹ)
,j
å å t(s *
p sin , j ÎSin
in, j ) y ( sin, j , p ) + b ( sin, j ) mu ( sin, j , p ) £ H , "j Î J , pÎP (3.21ʹ)
,j
The objective of the scheduling problem is to maximize the product throughput

or to minimize the makespan, as given in Constraint (3.22).

maximize å price ( s ) qs ( s , p ) ,
s
p p
"p = P, s p Î S p

or
minimize H (3.22)
éHù
Worthy of mention is that the number of time points is set at ê L ú, where tL is the
êt ú
minimum constant term of processing time among the given tasks, when solving LP
MAX and LP MIN. Also, the choice of either Constraint (3.20) or Constraint (3.21)
is dependent on whether task durations are fixed or allowed to vary with batch size.
Nevertheless, to emphasize, this formulation will entail a large number of variables
due to the excessive number of time points than required when solving the model as
mixed integer linear programming (MILP). However, this need not place any com-
putational burden on the solution since the formulation is LP.
3.4.12 Prediction of Optimal Number of Time Points

Figure 3.1 shows a procedure used to solve the sequence of scheduling problems
necessary for the prediction of time points. The first step is to solve the model as a
éHù
LP MAX problem using the maximum number of time points calculated with ê L ú.
êt ú
In this step, the maximum achievable throughput that is very close to the dual objec-
tive value at the optimal number of time points is determined. The number of time
points each unit can be active in the time horizon is also determined. Equation 3.23
is used to determine the total time points required using the frequency of utilization
of unit j obtained from the LP MAX problem. The unit which gives the maximum
number of time points is the unit most frequently used in the time horizon. This
becomes the critical unit; therefore, the optimal number of time points is dependent
on the frequency of utilization of this unit. In essence, this particular observation is
the cornerstone of this contribution.
Step 1: Solve LP MAX of a model problem using the maximum time points to obtain the
number of time points each unit can be active throughout the time horizon.
Calculate the total time points required by each unit. The unit that give the maximum time
points becomes the critical unit (Equation 3.23).
Step 2: Solve LP MIN of a model problem with the objective of minimizing time points
required by the critical unit using the objective value obtained from LP MAX problem.
Calculate the predicted optimal mumber of time points using the critical unit (Equation 3.25).
Step 3: Solve MILP of scheduling problem using a branch and bound technique based
on the predicted optimal number of time points as a root node.
FIGURE 3.1 Procedure to solve scheduling problem.

The first term on the right of Equation 3.23 å *

sin , j ÎSin ,j
å p
y ( sin, j , p ) is used to
determine how many times a unit j starts processing throughout the time horizon of
interest.
é
pj = ê å åy ( s in, j , p)+ åi + å iúù , j Î J, pÎP (3.23)
ê sin, j ÎSin, j ú
* p iÎI PT iÎI ST

The term å iÎI PT

i is to cater for those time points not shared by a unit because at
the beginning, a unit must wait until all the previous tasks have been processed. If
there is an intermediate initially in storage to start a task at the beginning of the time
horizon, then å iÎI PT
i = 0. Unit j at the end of its operation passes its intermediate
to the subsequent units (that lead to a product). The time points associated with the
subsequent tasks can be calculated with ST å
i; these time points are not shared
å
iÎI
by unit j. If the unit j is a unit producing a product, then the term i becomes 0;
iÎI ST
this assumption prevents the overestimation of the predicted optimal number of time
points.
The second step is to minimize the number of time points the critical unit is active,
as given in Constraint (3.24). This is done by fixing the objective value and number
of time points for each unit other than the critical unit to the values obtained from LP
MAX. The model used in this step, that is, LP MIN, comprises of Constraints (3.1)
through (3.8), (3.13) through (3.15), (3.18) through (3.24). The term jc represents the
critical unit. The predicted optimal number of time points which is very close to the
actual optimal number of time points is then given in Equation 3.25. Equation 3.23
for LP MIN can be expressed as
é
zj = ê å å y (s in, j , p )ù ,
ú p Î P, jÎJ (3.23ʹ)
ê sin, j ÎSin, j ú
* p

Minimize åå y ( s
S p
in, j c
,p ) (3.24)
in, jc
é
p pre = ê åå y(s in, j c
, p) + å i + å i ùú (3.25)
ê sin, jc p iÎI PT
iÎI ST
ú
The third step is to solve the exact MILP problem using the branch and bound
strategy around the predicted optimal number of time points. Solving the MILP
problem starts at the root node ppre, the left node is a node that is smaller than ppre
and the right node is a node that is larger than ppre. After solving the MILP problem
at the predicted optimal number of time points, the iteration starts at the immediate
node ppre – 1. If the optimal objective value at ppre – 1 is worse than that of ppre, it
is not necessary to continue any time point lower than ppre – 1. Otherwise, continue
solving until the objective value becomes worse. If the objective value at ppre – 1 is
worse than at ppre, then start solving the right node at ppre + 1. If the objective value at
ppre + 1 is better continue solving ppre + 2… until the solution converges. Otherwise,
stop the iteration.
3.5 CASE STUDIES
In order to illustrate this method for predicting the optimal number of time points,
four case studies found in published literature are presented. The results are given
under each case study and discussed. All the results were obtained using GAMS
CPLEX 9.1.2 in a 2.4 GHz, 4GB of RAM, Acer TravelMate 5740G computer. The
prediction method is applied to the model ML2 which gives the lowest time points
and better objective values when compared to the other model.
3.5.1 Case Study I: Simple Batch Plant Involving Three Tasks

This case study was conducted by Ierapetritou and Floudas (1998a) and Majozi
and Zhu (2001). The flowsheet of this well-published literature example is given in
Figure 3.2. The processed material passes through mixing, reaction and purifica-
tion as it is converted to product. The STN and SSN representations are shown in
Figure 3.3. The parameters for Case Study I are given in Table 3.1.

The computational statistics that show the optimal objective value, number of con-
straints required and the total number of continuous and binary variables required by
the model for different time horizons are given in Table 3.2. The duration constraints
in the model are fixed. In the iteration method, a lower time point is chosen first and
then increased by one until no change in the objective value is observed. For a time
horizon of 12 h, the iteration was started at time points 2. The objective value was 0.
The time points were increased by one to time points 3 and a better objective value
of 50 was recorded. This process of increasing the number of time points by one was
continued and the objective value of 100 did not change after time points 4. Hence,
the optimal objective value was 100. The CPU time required to solve the problem
Feed
Mixer Reactor Product

Purificator
FIGURE 3.2 Flowsheet for Case Study I.

S1 Mixer S2 Reaction S3
Purification S4
(a)
S1 S2 S3 S4
(b)
FIGURE 3.3 STN (a) and SSN (b) representations of Case Study I.
TABLE 3.1
Given Parameters for Case Study I
Unit Label j Capacity Suitability t ( sin, j ) b ( sin, j )
Heater 100 Mixing 3 for batch dependent 0.03

4.5 for fixed duration
Reactor 75 Reaction 2 for batch dependent 0.0266
3 for fixed duration
Separator 50 Purification 1 for batch dependent 0.02
1.5 for fixed duration
State Storage Capacity Initial Amount Price
1 Unlimited Unlimited 0
2 100 0 0
3 100 0 0
4 Unlimited 0 1
consists not only of the CPU time required at optimal number of time points 4, but is
the sum of the CPU times of each iteration from time points 2 to 5, which is (0.07 +
0.036 + 0.086 + 0.051) s.
For a time horizon of 24 h, a similar procedure was followed until the solu-
tion converged. The optimal objective value of 350 was obtained at time points 9
(after which there was no improvement in value). For the time horizon of 24 h, the
model consisted of 351 constraints, 215 continuous variables and 61 binary vari-
ables at the optimal number of time points. The model was solved to 0% relative
Methods to Reduce Computational Time
TABLE 3.2
Results Found for Case Study I under Maximization of Profit (Fixed Duration)
Model p CPU Time (s) Nodes RMILP MILP B.V C.V Constraints Non-Zeros Relative Gap (%)
(H = 12)
ML2 2 0.07 0 0 0 12 47 78 163
ML2 3 0.036 0 50 50 19 71 117 267 —
ML2 4 0.086 2 100 100 26 95 156 365 —
ML2 5 0.051 10 150 100 33 119 195 467
(H = 24)
ML2 8 0.01 0 300 300 54 191 312 787 —
ML2 9 0.08 10 350 350 61 215 351 875 —
ML2 10 0.5 1301 400 350 68 239 390 977
(H = 36)
ML2 14 0.12 25 600 600 96 335 546 1385 —
ML2 15 42.5 92,785 650 625 103 359 585 1487 —
ML2 16 246.7 454,229 700 625 110 383 624 1589 —
(H = 168)
ML2 73 40,000 2,054,460 3550 3525 509 1751 2847 7403 0.7
ML2 74 40,000 2,436,601 3600 3550 516 1775 2886 7607 1.4
ML2 75 40,000 1,903,820 3650 3550 523 1799 2925 7705 2.8
59
gap and required a CPU time of 0.08 s. The model was also solved for a time
horizon of 168 h to see how the computational time increased with this time hori-
zon. The problem required 75 iterations to solve. The optimal number of time
points was 74 with an objective value of 3550. The relative gap does not close in
the specified CPU time of 40,000 s. It can be observed, as is expected, that the
CPU time required in solving the problem increases exponentially as the number
of time points increases. This is due to an increasing number of binary variables.
Evidently, the iteration method for obtaining the optimal objective value is com-
putationally costly, which renders it impractical as the time horizon for scheduling
problems increases. This necessitates prediction of the optimal number of time
points beforehand.
The results for this literature example obtained from the prediction method are
explained here in detail. For the first scenario, which is a time horizon of 12 h, the
first step is to solve a LP MAX problem to maximize the throughput. The time points
for the LP model can be calculated with éê L ùú (12 h/1.5 h) to be 8. From Table 3.3,
H
êt ú
it can be seen that the LP MAX problem for the time horizon of 12 h resulted in an
objective value of 150. The value of 1.5 for heater implies that the number of times
the heater can be utilized is two (rounded up to the nearest integer). The number
of times the reactor can be used is two and the number of times the separator can
be used is three. Equation 3.23 was used to determine the time points required by
using the frequency of utilization of each unit obtained from the LP MAX problem.
For the mixer it becomes p1 = é
ê åp
y ( s1, p ) +å iÎI PT
i+ å
iÎI ST
i = 1.5 + 0 + 2 ù = 4.
ú
The parameter å PT
i in this equation is zero because there is no task before the
å
iÎI
mixer that is producing an intermediate. The parameter i in this equation is
iÎI ST
two because at the end of the process mixing, the mixer passes its intermediate to
the reactor and separator to be converted into product. There are two time points
associated with the reactor and separator after the end of the process by the mixer.
Applying the same thought as that of p1, the time points required by the reactor and
separator were calculated.
TABLE 3.3
Computational Results for LP MAX and LP MIN for Case Study I
under Maximization of Profit (Duration Constraints Are Fixed)
CPU Separator CPU
Obj. Time (s) for after LP Time for
H Value Heater Reactor Separator LP MAX MIN LP MIN Ppre
12 150 1.5 2 3 0.015 3 0.015 5
24 433.3 4.33 5.8 8.7 0.015 8.7 0.031 11
36 700 7 9.33 14 0.031 14 0.046 16
168 3633.3 36.3 48.4 72.7 0.33 72.7 0.34 75
é
p2 = ê
ê
å y(s , p) + å i + å i = 2 + 1 + 1ùúú = 4
p
2
iÎI PT iÎI ST
and
é
p3 = ê å y(s , p) + å i + å i = 3 + 2 + 0ùú = 5
3
ê p iÎI PT iÎI ST ú
The maximum number of time points obtained from the LP MAX problem is,
therefore, 5. Consequently, the separator becomes the critical unit.
The second step is to solve the LP MIN problem by setting the objective value to
150, å y(s1, p) = 1.5,
p å
y(s2 , p) = 2.0 and minimizing
p å
y(s3 , p), to obtain a
p
tighter solution for the number of time points the critical unit is used throughout the
time horizon. From the minimization problem, it was found that the critical unit is
used three times. Equation 3.25 was used to predict the optimal number of time points.
p pre = å y ( s , p ) + å i + å i + 1 = 3 + 2 + 0 = 5,
3
p iÎI PT iÎI ST
which is close to the actual optimal number of time points, that is, 4.
For the time horizon of 36 h, the first step of the LP MAX problem gave an objec-
tive value of 700. Using Equation 3.23, y ( s1, p ) = 7 ,å p å
y ( s2 , p ) = 9.33 and
p
å p
y ( s3 , p ) = 14 , the value of p1 was calculated to be 9 (7 + 0 + 2), p2 was calculated
to be 12 = é9.33 + 1 + 1ù and p3 was calculated to be 16 (14 + 2 + 0). From these val-
ues, the maximum of p j (highest integer value) is 16; hence, the separator is still the
critical unit.
The second step is to solve the LP MIN using the objective value (700) and the
number of time points each unit can be active, obtained from the LP MAX problem to
minimize å
y(s3 , p).
p å
y(s3 , p) was found to be 14 implying that the critical
p
unit is used 14 times. Using Equation 3.25, the predicted optimal number of time p pre
was 16 (14 + 2 + 0), which is very close to the actual optimal number of time points, 15.
For the time horizon of 168 h, the first step of LP MAX problem gave an objec-
tive value of 3633.3. Using Equation 3.23, å
y ( s1, p ) = 36.3, å
y ( s2 , p ) = 48.4 and
å
p p
y ( s3 , p ) = 72.7, the value of p1 was calculated to be 39 éë36.3 + 0 + 2 ùû , p2 was
p
calculated to be 51 é 48.4 + 1 + 1ù and p3 was calculated to be 75 é72.7 + 2 + 0 ù. From
these values, the maximum of p j (highest integer value) is 75; hence the separator is
the critical unit.
The second step is to solve the LP MIN using the objective value (3633.3) and the
number of time points each unit can be active, obtained from the LP MAX problem
to minimize å y(s3 , p).
p å
y(s3 , p) was found to be 72.7 implying that the critical
p
unit is used 73 times (rounded up to the nearest integer). Using Equation 3.25, the
predicted optimal number of time points p pre was é72.7 + 2 + 0 ù = 75, which is very
Processing units
I3 I3 I3
Separator (50) (50) (50) I3 (50) I3 (50) I3 (50) I3 (50)
7.5 10.5 12 13.5 15 17 19.5 21 22.5 24
Reactor I2 (70) I2 (30) I2 (70) I2 (70) I2 (60) I2 (50)
4.5 7.5 10.5 13.5 16.5 19.5 22.5
Heater I1 (100) I1 (100) I1 (100) I1 (100)
4.5 9 13.5 15 19.5
2 4 6 8 10 12 14 16 18 20 22 24
Time (h)
FIGURE 3.4 Gantt chart for time horizon of 24 h.
close to the actual optimal number of time points, 74. It is also interesting to see that
the CPU time in solving both LP MAX and LP MIN was 0.67 (0.33 + 0.34), which
is less than 1 s. This implies that the method is also accurate in predicting over long
time horizon in few seconds of processing time, which makes the prediction method
more valuable in avoiding computational costly iterations, especially as the time
horizon increases. The Gantt chart which shows the amount of material used at each
time by a task as well as the starting and finishing time of a task for a time horizon
of 24 h is given in Figure 3.4.
The computational results obtained for Case Study I by modelling duration con-
straints as a function of batch size are given in Table 3.4. The problem was solved for
time horizons between 12 and 168 h. As shown in Table 3.4, for a 12 h time horizon,
an optimal objective value of 71.52 was obtained at time points 4. For the time hori-
zon of 36 h, the iteration was started at time points 11. It is interesting to see that if
the iteration principle is followed, the iteration for this time horizon was stopped at
time points 12, since the objective value at time points 11 and 12 is 445.5. However,
for this time horizon, the optimal objective value was found at time points 13 with an
optimal objective value of 447. This indicates that the criteria of stopping the itera-
tion when the objective value are the same as the previous time points is subjected to
suboptimal results. For the time horizon of 168 h, the model gave an optimal objec-
tive value of 2645 using optimal number of time points 56. For this time horizon, the
problem is solved to 2% relative gap in a specified CPU time of 40,000 s.
The proposed model used to predict the optimal number of time points was also
tested for this case study. Table 3.5 shows the computational results for both the LP
MAX and LP MIN problems for different time horizons. For example, for a time
horizon of 36 h, the model reveals that the critical unit was the separator and is used
11 times throughout the time horizon, requiring 11 time points. At the beginning
of the time horizon, the separator accepts its intermediate after it is processed from
TABLE 3.4
Computational Results for Case Study I under Maximization of Profit (Duration Constraints Dependent on Batch Size)
Model p CPU Time (s) Nodes RMILP MILP B.V C.V Constraints Non-Zeros Relative Gap (%)
(H = 12)
ML2 2 0 0 0 0 12 47 78 173 —
ML2 3 0.062 0 50 50 19 71 117 281 —
ML2 4 0.093 10 100 71.52 26 95 156 389 —
ML2 5 0.23 283 150 71.52 33 119 195 497 —
(H = 24)
ML2 7 0.328 252 250 245.5 47 167 273 713
ML2 8 1.5 1,932 300 250.2 54 191 312 821 —
ML2 9 33.7 48,438 350 250.2 61 215 351 929 —
(H = 36)
ML2 11 15 13,938 450 445.5 75 263 429 1145 —
ML2 12 303 303,937 500 445.5 82 287 468 1253 —
ML2 13 1,549 1,366,014 550 447 89 311 507 1361 —
ML2 14 40,000 3,145,705 600 447 96 335 546 1469 —
(H = 168)
ML2 56 40,000 6,764,693 2700 2645.5 390 1343 2184 6005 2
ML2 57 40,000 5,637,699 2750 2645.5 397 1367 2223 6113 3.9
63
TABLE 3.5
Computational Results for LP MAX and LP MIN for Case Study I (Duration
Constraints Dependent on Batch Size)
CPU CPU
Time (s) Separator Time (s)
Obj. for LP after LP for LP
H Value Heater Reactor Separator MAX MIN MIN ppre
12 150 1.5 2 3 0.093 3 0.015 5
24 350 3.5 4.7 7 0.098 7 0.14 9
36 550 5.5 7 11 0.12 11 0.14 13
168 2750 27.5 36.7 55 0.72 55 0.73 57
the mixer and reactor, which require two time points not shared by the separator.
The predicted optimal number of time points then becomes 13, which is the same
as the actual value 13. For the time horizon of 168 h, the model predicts 57 time
points, which is again close to the actual optimal number of time points 56 in a CPU
time of 1.45 s. The proposed method predicts the optimal number of time points,
which is almost the same as the actual optimal number of time points, even for long
time horizon, proving that it is very useful in avoiding a large number of iterations,
which is encountered as the time horizon increases. It is difficult to compare the
CPU time required by the proposed method with that of Li and Floudas (2010) since
they employed iteration method to determine the maximum number of event points.
3.5.2 Case Study II: Batch Plant with a Recycle Stream

At first this case study was conducted by Kondili et al. (1993), and it becomes one
of the most common examples that appeared in published literature. The flowsheet
of this well-published literature example is given in Figure 3.5. This batch plant
produces two different products sharing the same processing units. The unit opera-
tions consist of preheating, three different reactions and separation. The STN and
SSN representations of the flowsheet are shown in Figure 3.6. The given parameters
for this case study are given in Table 3.6.

Table 3.7 shows the results for Case Study II. For a time horizon of 8 h, the problem
was solved with a 0% relative gap, giving an optimal objective value of 1498.6 with the
corresponding optimal number of time points 4. The proposed method also predicts
the same time points, which is 4. The time required to solve the problem was less
than 1 s for each iteration. As the time horizon increased, the binary variables also
increased resulting in an increase in CPU time. For the time horizon of 16 h, the
model requires 8 time points to give an optimal objective value of 3738.4. As it is
seen from Table 3.7, a time horizon 16 h requires CPU time of 79.7 s, which is high
when compared to time horizon of 8 h with a CPU time of 0.156 s. This indicates that
Raw 3
Product 1
Raw 2
Reactor 1
Heater
Raw 1
Product 2
Still
Reactor 2
FIGURE 3.5 Flowsheet for Case Study II.
as the time horizon increases, CPU time required increases drastically, even worse
if iteration method is followed.
Table 3.8 shows the computational results of the method used to predict the opti-
mal number of time points. For a time horizon of 8 h, the first step was to solve the
LP MAX problem using the determined maximum number of time points, which was
é 8 ù
êê 0.667 úú = 12. The problem was solved in less than 1 s giving an objective value of
2252.6. From Table 3.8, the number of times reactor 1 can be used was 4 throughout
the 8 h time horizon of interest. Similarly, the number of times reactor 2 can be
used was 4. Using Equation 3.23, the time points required using reactor 2 was 4. The
termå iÎIPT
i , in Equation 3.23, for reactor 2 becomes 0 since there is no task before
reactor 2. Since reactor 2 is a unit producing a product and also a unit producing an
intermediate used by the separator, the term åiÎI ST
i becomes 0; this assumption is
used not to overestimate the number of time points required even if there is a subse-
quent task separator.
Product 1
s7
40%
40% 60%
10%
60%
s9 Separation
s6
80% 90%
50% s10
s2 Reaction 2 Reaction 3
50% 20%
Product 2
(s3) (s4)
(a) Feed c
s1
s2 s5
40% Product 1
40%
50% s7
60%
s6 60% 10%
(s3) 50% 80%

s8 s9
Feed c
20%
90%
(s4)
s10
(b)
Product 2
FIGURE 3.6 (a) STN and (b) SSN representations of Case Study II.
The second step of LP MIN also gave the same result for the number of times
the critical unit (R2) was used as that of LP MAX, which is 4. Using Equation 3.25,
the predicted optimal number of time points then becomes 4, which is the same
as the actual optimal number of time points, i.e. 4. For the time horizon of 10 and
12 h, the model predicts 5 and 6 time points, respectively, which is close to the actual
optimal number of time points 6 for 10 h and 7 for 12 h. From Table 3.8, the results
obtained using the proposed method to predict the optimal number of time points,
the predicted optimal number of time points obtained almost matched the actual
optimal number of time points, indicating that the method is rigorous. Figure 3.7
shows the resultant Gantt chart for a time horizon of 16 h.
TABLE 3.6
Given Data for Case Study II
Task VsLin , j VsUin , j Unit t ( sin , j ) b ( sin , j )
Heating (i1) 0–100 Heater (HR) 0.667 0.00667
Reaction 1 (i2) 0–50 Reactor 1 (RR-1) 1.334 0.02664
Separation (i5) 0–200 Separator (SR) 1.3342 0.00666
State Storage Capacity Initial Amount Price
s2 Unlimited Unlimited 0
s3,s4 Unlimited Unlimited 0
s5 100 0 0
s8 200 0 0
s6 150 0 0
s9 200 0 0
s7 Unlimited 0 10
s10 Unlimited 0 10
3.5.3 Case Study III: Multipurpose Batch Plant

with Six Units and Eight Tasks
At first Sundaramoorthy and Karimi (2005) conducted this case study, which is rela-
tively a more complex problem, and later often used in literature to check the effi-
ciency of models in terms of optimal objective value and CPU time required to get
optimal objective value. The plant accommodates many common features of multi-
purpose batch plants such as unit performing multiple tasks, multiple units suitable
for a task and dedicated unit for specific tasks. The STN and SSN representations of
this case study are depicted in Figure 3.8. The data required for the case study are
given in Table 3.9. The problem is solved for 8, 10, 12 and 16 h.

The proposed model for the time horizon of 8 h requires 5 time points to get the opti-
mal objective value of 1583.4 in a CPU time of 1.25 s, as shown in Table 3.10. The
proposed method used to predict optimal number of time points predicted 5 time
points which is exactly the same as the actual optimal number of time points. For
the time horizon of 10 h requires 6 time points to get the optimal objective value of
2345.3 in a CPU time of 8 s. For this case also the prediction method predicted the
same number of time points as that of the actual number of time points 6. For the
time horizon of 16 h, an optimal objective value of 4261.9 was obtained in a specified
CPU time of 10,000 s using the optimal number of time points 10. The prediction
TABLE 3.7
Computational Results of Case Study II under Maximization of Profit
CPU
Model P RMIP MIP Constraints C.V B.V Non-Zeros Time (s)
(H = 8)
ML2 3 975 866.67 382 206 48 1075 0.093
ML2 4 1730.6 1498.6 528 275 64 1518 0.156
ML2 5 2123.3 1498.6 676 344 80 1991 0.608
(H = 10)
ML2 5 2436.7 1917 674 344 80 1961 0.327
ML2 6 2730.7 1962.7 908 403 96 2548 11.3
ML2 7 2780.2 1962.7 1054 470 112 2985 816
(H = 12)
ML2 6 3076.6 2646.8 908 403 96 2548 1.36
ML2 7 3301 2658.5 1054 470 112 2985 35.4
ML2 8 3350.5 2658.5 1200 537 128 3422 1255.3
(H = 16)
ML2 7 3799.6 3638.7 1054 470 112 2985 0.84
ML2 8 4291.7 3738.38 1200 537 128 3422 79.7
ML2 9 4439 3738.38 1258 620 144 3733 887
TABLE 3.8
Computational Results of LP MAX and LP MIN for Case Study II
CPU CPU
Time (s) Time (s) Reactor
Obj. Reactor 1 Reactor 2 Separator Heater for LP for LP 2 after
H Value (RR-1) (RR-2) (SR) (HR) MAX MIN LP MIN Ppre
8 2252.6 3.52 3.8 0.82 0.76 0.031 0.05 3.8 4
10 2815.8 4.4 4.8 1.03 0.96 0.031 0.05 4.8 5
12 3379 5.33 5.64 1.23 1.15 0.062 0.08 5.64 6
16 4505.4 7.03 7.42 1.64 2.5 0.11 0.09 7.42 8
method only requires a CPU time of 0.12 s to predict exactly the same number of
time points as that of the actual number of time points 10. This indicates that the
method is powerful in predicting optimal number of time points in low CPU time.
The method used to predict optimal number of time points is explained here for
the time horizon of 16 h. The computational results of LP MAX and LP MIN are
given in Table 3.11.
Processing units
SR I5 (80) I5 (80) I5 (91.63)
6.606 8.473 10.605 12.919 14.056
I4
RR-2 I2 (76.48) I3 (80) I4 (80) I3 (80) I4 (80) I4 (58.84) I3 (36.67)
(56.38)
2.607 5.273 6.606 9.272 10.605 12.919 14.056
I4
RR-2 I2 (47.8) I3 (46.78) I2 (46.78) I2 (50) I3 (50)
(35.24)
I3 (22.91)
2.607 5.07 7.587 10.253 12.919 14.056
I1 I1
HR (100) (47.4)
1.334 2.317
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Time (h)
FIGURE 3.7 Gantt chart for a time horizon of 16 h.
The first step was to solve the LP MAX problem using the determined maximum
é 16 ù
number of time points, which was ê = 24 ú . The problem was solved in less than
ê 0.667 ú
1 s giving an objective value of 5201.9. From Table 3.11, the number of times reac-
tor 1 can be used was 4 (rounded up to the nearest integer) throughout the 16 h time
horizon of interest. For reactor 2, also implying that the number of times that reactor
2 can be used was 7. The number of times the heater, separator, mixer 1 and mixer 2
can be used throughout the 16 h time horizon of interest are 9, 2, 3 and 2, respectively.
Equation 3.23 was used to determine the time points required by using the frequency
of utilization of each unit obtained from the LP MAX problem. For reactor 1, it
becomes é åå
ê p *
sin , j ÎSin ,j
å å
y ( sin, j , p ) +
iÎI PT
i+
iÎI ST
i = 3.5 + 0 + 0 ù = 4, for reac-
ú
tor 2 it becomes é
êåå p å å
*
sin , j ÎSin ,j
y ( sin, j , p ) +
iÎI PT
i+
iÎI ST
i = 6.72 + 0 + 0 ù = 7.
ú
The term å i for reactor 1 and reactor 2 equals to zero because both reactors
iÎI PT
can start at the beginning of the time horizon by processing reaction 1. The term
å i for both reactors equals to zero since most of the time the units produce
iÎI ST
a product at the last time point rather than producing an intermediate and make
the unit ideal. This assumption can be justified when looking at the Gantt chart for
Feed 3
S8

H1 {i1} H2 {i5} R3 {i7} 0.4
S1 S9 S13
{HR} {HR} {RR1,RR2}
Int 6 0.6
0.25
S3 Int 1 S6 Int 4 Int 7 S10
0.5 0.4 0.4
Int 5 Product 1
R2 {i3} S {i4} 0.5 M {i6}
S5 S7 S12
{RR1,RR2} {SR} 0.4 {MR1,MR2}
Int 3
0.5 0.1 0.2
S4
Int 2 S11
Feed 2 Feed 4
R1 {i2}
S2
{RR1,RR2}
(a)
Feed 3
s8
Feed 1 Product 2
0.75
0.4
s1 s9 s13
0.25 0.6
Int 6
s3 Int 1 Int 7 s10
s6 Int 4
Int 5 0.4 Product 1
0.5 0.4 0.5
s5 s7 s12
0.5 0.4
0.1
Int 3 0.2
S4
Int 2 s11
Feed 2
s2 Feed 4
(b)
FIGURE 3.8 (a) STN and (b) SSN representations for Case Study III.
the time horizon of 10 h in Figure 3.9; the reactors produce a product at the last
time point. Consequently, the assumption helps to predict a fair number of time
points instead of overestimating the time points which result to excessive computa-
tional time.
For the heater, Equation 3.23 gives éå å p *
sin , j ÎSin ,j
y ( sin, j , p ) + å iÎI PT
i+
å iÎI ST ù
i = 9 + 0 + 1 = 10 . The term å iÎI PT
i for the heater becomes zero
TABLE 3.9
Data for Case Study III
Task Unit t ( sin , j ) b ( sin , j ) VinL , j - VinU, j
Heating 1 (i1) Heater (HR) 0.667 0.00667 0–100
Heating 2 (i5) Heater (HR) 1 0.01 0–100
Reaction 1 (i2) Reactor 1 (RR-1) 1.333 0.01333 0–100
Separation (i4) Separator (SR) 2 0.00667 0–300
Mixing (i6) Mixer 1 (MR-1) 1.333 0.00667 20–200
Mixing (i6) Mixer 2 (MR-2) 1.333 0.0067 20–200
State Initial Amount Storage Capacity Price
s1 Available as required Unlimited 0
s3 0 100 0
s4 0 100 0
s5 0 300 0
s6 50 150 0
s7 50 150 0
s9 0 150 0
s10 0 150 0
s12 0 Unlimited 5
s13 0 Unlimited 5
since the heater can start processing at the beginning of the time horizon.
The term å iÎI ST
i for the heater becomes one since the heater at the last time
point gives its intermediate to the reactors to be converted into product. This
time point is not shared by the heater. For the separator, Equation 3.23 gives
é
êåå p *
sin , j ÎSin ,j
å
y ( sin, j , p ) + å
iÎI PT
i+
iÎI ST
i = 1.9 + 2 + 1ù = 5. The term
ú iÎI PTåi
for the separator becomes 2 because at the beginning the separator gets its interme-
diate after reaction 1 and reaction 2 are processed which requires two time points
associated to each task not shared by the separator. The term å ST
iÎI
i for the separa-
tor becomes 1 since at the end of its operation the separator passes its intermediate to
the subsequent task, mixing (the shortest path to form product 1) instead of heating
(the longest path which is the path to form product 2). This assumption is also a fair
72
TABLE 3.10
Computational Results for Case Study III under Maximization of Profit
Time Non- Relative
Model Horizon (H) P CPU Time (s) Nodes RMILP MILP B.V C.V Constraints Zeros Gap (%)

ML2 8 5 1.25 3,662 2100 1583.4 107 428 889 2680 —
ML2 10 6 8 9,250 2871.9 2345.3 131 516 1093 3308 —
ML2 12 7 128 103,886 3465.6 3041.3 155 604 1297 3936 —
ML2 16 10 10,000 4,455,151 4653.1 4262.8 227 868 1909 5820 2
TABLE 3.11
Computational Results of LP MAX and LP MIN for Case Study III
CPU Time (s) CPU Time (s) HR after
H Obj. Value RR-1 RR-2 SR HR MR-1 MR-2 for LP max for LP min LP min ppre
8 2488 1.95 3 0.71 3.7 1 1 0.02 0.02 3.7 5
10 3166.4 1.47 4.7 1 4.8 0 2.5 0.026 0.026 4.8 6
12 3844.9 1 5.7 1.3 6 0.73 2.3 0.032 0.032 6 7
16 5201.9 3.5 6.72 1.9 8.3 3 1.1 0.06 0.06 8.3 10
73
Processing units Time point
p4 p6
MR-2 The type of task the unit I6 (150) I6 (60.062)
processes
5.48 7.82 8.27 10
p5
Amount of material
MR-1 processed I6 (137.9)
7.75 10
p3 p4
SR I4 (64.24) I4 (114.12)
2.56 4.99 7.75
p2 p3 p5 p6
I3
RR-2 I2 (23.504) I7 (100) I7 (39.6) I7 (42.45)
(78.36)
1.54 2.56 3.26 5.48 6.58 8.27 8.29 10
p2 p3 p4 p6
RR-1 I2 (15.67) I2 (50) I3 (100) I7 (92.38) I7 (28.34)
1.54 3.54 4.99 5.19 7.75 8.29 10

p2 p3 p4 p5
HR I1 (89.2) I5 (100) I5 (92.38) I5 (100) I5 (100)
1.26 3.26 5.19 6.58 8.29
1 2 3 4 5 6 7 8 9 10
Time (h)
FIGURE 3.9 Gantt chart for the time horizon of 10 h.
assumption which maximizes the utilization of the separator and prevents overesti-
mating the time points. If the separator is a critical unit, the last intermediates either
state s6 is consumed by the heating followed by reaction 3 or state s7 is consumed
by the mixing task. Now let us examine each case, finally produced state s6 will take
a minimum of 1 h for heating and 1.333 h for reaction 3, which forces the separator
unit to finish 2.333 h earlier than the time horizon. For the case of state s7, it will take
a minimum of 1.333 h which will force the separator to finish 1.333 h earlier than the
time horizon. Using the principle that the critical unit is the unit that is frequently
used and active most of the time horizon, the shortest path to form product 2 allows
this principle for the separator to be active for longer than the longest path to form
product 1.
For mixer 1, Equation 3.23 gives éå å p *
sin , j ÎSin ,j
y ( sin, j , p ) + å iÎI PT
i+
å iÎI ST ù
i = 3 + 2 + 0 = 4 . The term å iÎI PT
i , for mixer 1, becomes 2 since at the
beginning the mixer can only start its operation after heating two and reaction three
are processed, which requires two time points not shared by the mixer. The term
å iÎI ST
i becomes zero since there is no subsequent task after mixing operation, and
it is the unit producing a product and can be active until the last time point.
For mixer 2, Equation 3.23 gives éå å p *
sin , j ÎSin ,j
y ( sin, j , p ) +
å iÎI PT
i+
å iÎI ST ù
i = 1.1 + 2 + 0 = 4 . The unit that gives the highest time points is the heater;
as a result it becomes the critical unit.
The second step of LP MIN also gave the same result for the number of times
the critical unit heater was used as that of LP MAX, that is, 9. Equation 3.25 is
used to predict the optimal number of time points. The term å å
y(sin, j c , p)
s
in , j c
p
in Equation 3.25 for this case becomes 9. The term å

iÎI PT
i for this case is zero
because the heater can start processing at the beginning of the time horizon. The
parameter åST
i for this case is one because at the end of the process heating, the
iÎI
heater passes its intermediate to the reactor to be converted into product. There is
one time point associated with the reactor after the end of the process by the heater.
The predicted optimal number of time points then becomes 10, which is the same
as the actual optimal number of time points. The predicted optimal number of time
points obtained matched the actual optimal number of time points, indicating that
the method is rigorous.
3.5.4 Case Study IV: Industrial Case Study

This case study was taken from the book by Majozi and Zhu (2001). Figure 3.10 is
the flowsheet for the case study. The STN and SSN representations are shown in
Figure 3.11. The process consists of five consecutive steps. The first step involves
Raw 1
Raw 2
Raw 3
SE 1
EV 1 Effluent
Raw 4 R3
R1
SE 2
Waste
SE 3 EV 2
R2 R4
Storage
Product
FIGURE 3.10 Flowsheet for the industrial case study.

s11
s10
s1 Reaction 1 s2 Reaction 2 s3 Reaction 3
s9
s4 Settling s5 Evaporation s6
s7
s8
(a)
S9
S8
S6
S1 S2 S3 S4
S5
S10
S7
S11
(b)
FIGURE 3.11 (a) STN and (b) SSN representations for the industrial case study.
a reaction which forms an arsenate salt. This reaction requires two raw materials,
raw 3 and raw 4, and can be conducted in either reactor R1 or R2. The arsenate salt
from the first step is then transferred to either reactor R3 or R4, wherein two con-
secutive reactions take place. The first of these reactions is aimed at converting the
arsenate salt to a disodium salt using raw material 1 (raw 1). The disodium salt is
then reacted further to form the monosodium salt using raw material 2 (raw 2). The
monosodium salt solution is then transferred to the settling step in order to remove
the solid by-product. Settling can be conducted in any of the three settlers, that is
SE1, SE2 or SE3. The solid by-product is dispensed with as waste and the remain-
ing monosodium salt solution is transferred to the final step. This step consists of
two evaporators, EV1 and EV2, which remove the excess amount of water from the
monosodium solution. Evaporated water is removed as effluent and monosodium

salt (product) is taken to storage. States s1 and s9 in the SSN represent raw 3 and
raw 4, respectively. States s10 and s2 represent raw 1 and the arsenate salt, while
states s11 and s3 represent raw 2 and the disodium salt, respectively. State s4 is
the monosodium solution that is transferred to the settlers to form states s8 (solid
by-product) and s5 (remaining monosodium solution). State s5 is separated into
states s7 (water) and s6 (product). Table 3.12 shows the data for the case study.
The stoichiometric data are included in order to perform material balances in each
TABLE 3.12
Data for Industrial Case Study
Unit v sLin , j - v Usin , j Task t ( sin , j )
RR1 0–10 Reaction 1 (R1) 1.6
RR2 0–10 Reaction 1 (R1) 1.6
RR3 0–10 Reaction 2 (R2) 2.4
RR3 0–10 Reaction 3 (R3) 0.8
RR4 0–10 Reaction 2 (R2) 2.4
RR4 0–10 Reaction 3 (R3) 0.8
SE1 0–10 Settling (SR) 0.8
EV1 0–10 Evaporation (EV) 2.4
EV2 0–10 Evaporation (EV) 2.4
State Storage Capacity Initial Amount
s1 Unlimited Unlimited
s2 100 0
s3 100 0
s4 100 0
s5 100 0
s6 100 0
s7 100 0
s8 100 0
Stoichiometric Data
State Ton/Ton Output Ton/Ton Product
s1 0.2
s9 0.25
s10 0.35
s11 0.2
s7 0.7
s8 1
unit operation. The second column of the stoichiometric data shows the amount
of raw material required (tons) per unit mass (tons) of the overall output, that is,
s6 + s7 + s8. The third column shows the ratio of each by-product (s7 and s8)
to product (s6) in ton/ton product. The objective function is the maximization of
product (s6) output.

This case study is solved using fixed processing time where duration of a task is
independent of batch size. The computational results for this industrial case study
are given in Table 3.13. The model was solved for a time horizon of 8, 10, 12 and
16 h. For a time horizon of 12 h, an optimal objective value of 18.5 was found using
8 time points. For a time horizon of 16 h, an optimal objective value was obtained
using 12 time points. The formulation resulted in 10,515 constraints, 9,397 con-
tinuous variables and 264 binary variables. The optimal objective value of 30.6 is
obtained in a specified CPU time of 10,000 s.
The proposed method for predicting the optimal number of time points was
also tested for this case study. Table 3.14 shows the computational results obtained
from the LP MAX and LP MIN of the problem. For a time horizon of 16 h, using
Equation 3.25, the critical unit is RR4 and it was used nine times. Considering RR4
TABLE 3.13
Computational Results for Industrial Case Study under Maximization
of Profit
Cons CPU Non-
Model H P Nodes RMILP MILP B.V C.V traints Time (s) Zeros
ML2 8 6 494 12.34 9.26 132 4699 5,211 0.7 7,156
ML2 10 8 3,852,196 16.5 13.9 176 6265 6,979 5,271 7,156
ML2 12 8 900 20.6 18.5 176 6265 6,979 2.5 9,678
ML2 16 12 4,330,203 37 30.6 264 9397 10,515 10,000a 14,722
a Relative gap, 17%.
TABLE 3.14
Computational Results Obtained from LP MAX and LP MIN for Industrial
Case Study
H Obj. Value RR1 RR2 RR3 RR4 SE1 SE2 SE3 EV1 EV2 Ppre
8 8.7 0.74 0.31 1.57 2.7 1 1 0.35 1 0.48 6
10 14.16 0.5 1.23 2.93 3.96 3.14 0 0.69 1.7 0.57 7
12 19.6 0 2.38 4.74 4.78 3.27 0 2 2.7 0.7 8
16 30.5 0.15 2.2 6.12 8.7 3 2.3 2.3 0.85 4.3 12
as the critical unit and using Equation 3.25, the predicted optimal number of time
point becomes 12. The term å
iÎI PT
i for this case is 1 since there is one task before
reaction two (R2), which is reaction 1 (R1). At the beginning of the time horizon,
RR4 starts processing after reaction R1 is conducted and this task (reaction 1) has
one time point associated with it. This time point is not shared by RR4. The term
å iÎI PT
i becomes 2 since RR4, at the end of its operation, passes its intermediate to
the next processes which are settling and evaporation with one time point associated
with each of these tasks. As a result, the predicted optimal number of time points
becomes 12, which is the same as the actual optimal number of time points 12. As
shown in Table 3.14, for all time horizons the proposed model predicts the optimal
number of time points close to the actual optimal number of time points.
Table 3.15 summarizes the results obtained in all the cases considered in terms of
actual and predicting optimal number of time points.
Figure 3.12 shows the comparison between the actual optimal number of
time points and the predicted optimal number of time points using the data from
Table 3.15. The R2 value is 0.998, which suggests that the proposed model gives
TABLE 3.15
Results of Predicted and Actual Optimal Number of Time Points
for Each Case Study
Actual Optimal Predicted Optimal
H Case Studies Number of Time Points Number of Time Points
12 I fixed duration 4 5
12 I duration batch dependent 4 5
8 II 4 4
10 II 6 5
12 II 7 6
16 II 8 8
8 III 5 5
10 III 6 6
12 III 7 7
16 III 10 10
8 IV 6 6
10 IV 8 7
12 IV 8 8
16 IV 12 12
80
R2 = 0.9984
Predicted optimal number of time points
70
60
50
40
30
20
10
0
0 10 20 30 40 50 60 70 80
Actual optimal number of time points
FIGURE 3.12 Comparison between predicted and actual optimal number of time points.
accurate predictions. Application of this proposed method is vital to avoid unneces-

sary costly iteration, especially as the time horizon increases.
3.6 CONCLUSIONS
A method for predicting the optimal number of time points has been presented in
this chapter. Linear programming maximization and minimization models were
used to identify the critical unit and the number of times this critical unit was used.
Based on the principle that the optimal number of time points depends on how many
times this critical unit is used, an equation was developed to predict the optimal
number of time points. The model was tested for a number of case studies taken from
published literature. A variance of 0.998 was obtained when comparing predicted
optimal number of time points with the actual optimal number of time points. This
indicates that the method is accurate. Application of the proposed method will avoid
costly computational iteration as the time horizon increases.
NOTATION
Sets
I PT {Previous tasks that are performed before task i in the recipe}
S {s | s any state s}
I {i | i is a task}
I ST {Subsequent tasks that are performed after task i in the recipe}
SinPT,J {sin, j | sin, j previous task before task sin, j}
SinST,J {sin, j | sin, j subsequent task before task sin, j}
J { j | j is a unit}
Sinc , j {sinc , j | sinc , j task which consume state s}
Sin* , j {sin* , j | sin* , j state used by a task}
Sin, j {sin, j | sin, j effective state}
Sinuc, j {sinuc, j | sinuc, j task which consume unstable state s}
Sinp , j {sinp , j | sinp , j task which produce state s}
Sinup, j {sinup, j | sinup, j task which produce unstable state s}
S p {s p | s p a state which is a product}
P {p | p is a time point}
J s { js | js is a unit producing state s}
Variables
t p ( sin, j , p ) Time at which task ends at time point p, sin, j Î Sin, j
tu ( sin, j , p ) Time at which task starts at time point p, sin, j Î Sin, j
mu ( sin, j , p ) Amount of material processed by a task at time point p
qs ( s, p ) Amount of state s stored at time point p
y ( sin, j , p ) Binary variable for assignment of task at time point p
t ( j, p ) Binary variable associated with usage of state produced by unit j at
time point p
t ( j, s, p ) Binary variable associated with usage of state s produced by unit j at
time point p if the unit produces more than one intermediate at time
point p, s Î Sinp , j
x ( s, p ) Binary variable associated with availability of storage for state s at
time point p
u ( sin, j , p ) Amount of material stored in unit j at time point p
Z j The number of time points a unit is active obtained from LP MAX
problem
Parameters
VsU Maximum capacity of unit j to process a particular task
in , j
VsLin, j Minimum capacity of unit j to process a particular task

H Time horizon of interest
QS o Initial amount of state s stored
QS U Maximum capacity of storage to store a state s
t ( sin, j ) Coefficient of constant term of processing time of state s, s Î Sin, j
b ( sin, j ) Coefficient of variable term of processing time of state s, s Î Sin, j
( )
r sinsp, j Portion of state s produced by a task
r ( s )
sc
in, j Portion of state s consumed by a task
REFERENCES
Ierapetritou, M.G., Floudas, C.A., 1998a. Effective continuous-time formulation for short-term
scheduling: 1. Multipurpose batch processes. Industrial and Engineering Chemistry
Research. 37, 4341–4359.
uling of batch operations. I. MILP formulation. Computer and Chemical Engineering.
17, 211–227.
Li, J., Floudas, C.A., 2010. Optimal event point determination for short-term scheduling of
multipurpose batch plants via unit-specific event-based continuous-time approaches.
bach plants. Industrial and Engineering Chemistry Research. 40, 5935–5949.
Science. 60, 2679–2702.
4 Integration of Scheduling
and Heat Integration
Minimization of Energy
Requirements
4.1 INTRODUCTION
Presented in this chapter is a mathematical technique for simultaneous heat integra-
tion and process scheduling in multipurpose batch plants. Taking advantage of the
intermittent continuous behaviour of process streams during transfer from one pro-
cessing unit to another, as determined by the recipe, the presented formulation aims
to maximize the coincidence of availability of hot and cold stream pairs with fea-
sible temperature driving forces, while taking into consideration process scheduling
constraints. Distinct from similar contributions in the published literature, time is
treated as one of the key optimization variables instead of a parameter fixed a priori.
Heat integration during stream transfer has the added benefit of shortened residence
time in processing units, which invariably improves the throughput, as more batches
are likely to be processed within a given time horizon, compared to conventional
cases with heating and cooling in situ.
The trend towards batch processing calls for the development of effective techniques
for production planning and scheduling. Extensive research has been conducted on
developing mathematical models to improve batch plant efficiency (Harjunkoski et al.,
2014), with recent advances in computer technology allowing large-scale and more
complex problems to be handled by using optimization techniques. In addition to pro-
cess scheduling, heat integration may be considered for batch plants to reduce external
utility (e.g. steam and cooling water) requirements for tasks involving heating or cool-
ing, such as endothermic and exothermic reactions. This is driven by not only cost
considerations, but also increased environmental awareness and the pursuit of energy
efficiency. Consequently, the past two decades or so have been characterized by a sig-
nificant body of research on heat integration of batch plants (Fernández et al., 2012).
Heat integration in batch processes can be carried out directly or indirectly
(Majozi, 2010). Direct integration requires the hot and cold tasks to coincide in order
for process–process heat exchange to take place, whereas indirect integration uses
thermal storage enabling heat exchange between tasks performed in different time
intervals, thus providing further and more flexible heat recovery. The methodologies
83
developed for batch heat integration can be categorized into pinch-based and math-
ematical techniques. Early studies for minimizing energy consumption in batch
plants were mostly based on pinch analysis and conducted by extending methods
developed for continuous processes. Mathematical optimization, on the other hand,
has been increasingly adopted for cases with multiple design criteria and/or practical
constraints. More importantly, the use of mathematical techniques allows time to
be treated as a variable, in which case the production schedule is allowed to change.
This implies the incorporation of heat integration into batch process scheduling and
a simultaneous solution for more optimal results.
4.3 MOTIVATING EXAMPLE
While there have been many works addressing scheduling and heat integration of
batch plants, most only consider direct and/or indirect integration between processing
units/vessels. The opportunity for energy savings through heat exchange between pro-
cess streams during material transfer was overlooked and is exploited in this chapter.
The motivation and rationale behind the contribution is illustrated in Figures 4.1
and 4.2. As shown in Figure 4.1, the material transferred from processing unit 1 to 4
Unit 1 Unit 2
Unit 3 Unit 4
Cooling
Steam
water
Heating Cooling Processing Discharge
R4
R3
Δt
Time
0 t1 t2 H
FIGURE 4.1 Impact of overlooking process–process heat integration on time.

Integration of Scheduling and Heat Integration 85
Unit 1 Unit 2
Unit 3 Unit 4
Cooling
Steam
water
Heating Cooling Processing Discharge
R4
R3
Δt Time
0 t΄1 t΄2 t1 t2 H
FIGURE 4.2 Impact of exploiting process–process heat integration on time.
requires cooling prior to the commencement of a subsequent processing task in

unit 4. On the other hand, the material transferred from processing units 2 to 3
requires heating before the next processing task in unit 3. The times required for
both heating and cooling in units 3 and 4 are denoted as Δt for illustration purposes,
given that these need not necessarily be of equal duration.
It is evident that if the thermal driving forces permit and the transfer times from
units 1 and 2 to units 4 and 3 coincide, an opportunity for stream–stream heat inte-
gration could arise and be exploited with significant benefits. Figure 4.2 summarizes
this particular observation. Heat integration of streams from units 1 to 2 is effected
through an adequately sized heat exchanger, during the transfer of materials to sub-
sequent processing units (i.e. units 3 and 4). Consequently, the need for external hot
and cold utilities, as well as dedicated time for heating and cooling inside the pro-
cessing units (in situ) prior to the commencement of processing tasks, is eliminated,
with potentially substantial positive cost implications. In particular, more batches are
likely to be produced in units 3 and 4 with the elimination of dedicated in situ heat-
ing and cooling times. The implication of this time benefit is depicted in Figure 4.2,
where it is clearly shown that t1 − t′1 = t2 − t′2 = Δt.
The overall idea, therefore, is to maximize the coincidence of availability of hot
and cold stream pairs with feasible thermal driving forces within a comprehensive
process scheduling framework. This would require a mathematical formulation to
adequately address simultaneous heat integration and process scheduling where time
is a key optimization variable. This is the essence of the contribution.
The problem addressed in this chapter can be briefly stated as follows. Given:
(1) production scheduling data, including the recipe for each product, task durations,
equipment capacities, the time horizon of interest, raw material costs and product
selling prices; (2) operating temperatures for processing tasks, supply temperatures
of raw materials and storage temperatures for final products; (3) specific heat capaci-
ties of states and (4) costs of external hot and cold utilities, determine an optimal
production schedule that maximizes the profit, which is defined as product revenue
minus raw material and utility costs. For simplicity, the following assumptions are
made in the problem:
• Heat capacities are constant.

• Heat losses for temporary storage are negligible.
• Countercurrent heat exchangers are used for heat integration.
Figure 4.3 shows the superstructure representations on which the mathematical mod-
elling is based. It can be seen that each processing unit can receive material from
and send material to other units or storage. The input material may need to be heated
or cooled in the processing unit before the task commences in order to meet the
operating temperature; similarly, the output material may be heated or cooled in situ
before leaving for further processing or storage. To reduce the use of external hot and
cold utilities, heat recovery between output (i.e. intermediate and product) streams
(Figure 4.3a) may be considered, as well as heat exchange for raw material pre-
heating (Figure 4.3b). Note that output streams are assumed to have heat exchange
before entering downstream processing units or storage, while raw material streams
are, after leaving storage. If there is more than one state produced in a processing
unit, these states are assumed to exit the unit at the same temperature and treated
as separate streams for heat integration. It is also assumed that external utilities can
only be used to unit jackets – no utility heaters or coolers are available for use. In
addition to the necessary scheduling constraints (see Appendix), the mathematical
model consists mainly of mass and energy balances and heat integration constraints,
as presented here. Notation used is given in the Nomenclature.
4.5.1 Mass and Energy Balances for Processing Units

Constraints (4.1) and (4.2) describe the inlet mass balance for unit j at any time point p.
The input state, either a raw material or an intermediate, may come from storage or
upstream units (j′) where it is produced. Note that there is nothing transferred from
producing units at the first time point (p = 1) as no intermediates have been produced
yet. The only possible input at this point in time is raw material from storage. It is
From other From other

units/storage units/storage
Tin(sin,j, p) Tin(sin,j΄, p)
Top(sin,j) j j΄ Top(sin,j΄)
Tout(sin,j, p) Tout(sin,j΄, p)
Ti(sout,j΄, p)
Ti(sout,j, p)
Tf (sout,j, p)
Tf (sout,j΄, p)
To other To other
(a) units/storage units/storage
From other
units/storage
Tin(sin,j, p – 1)
j Feed
Top(sin,j)
storage
Tout(sin,j, p – 1)
Ts0(s)
Ti(sout,j, p – 1)
Tf (sout,j, p – 1)
Tf (srin,j΄, p)
Tin(sin,j΄, p)
To other
units/storage
Top(sin,j΄) j΄
Tout(sin,j΄, p)
To other
(b) units/storage
FIGURE 4.3 Superstructure for heat integration between (a) outlet streams and (b) outlet
and inlet (raw material) streams.
also worth noting that this set of constraints also cater for stoichiometric require-
ments of inputs into a unit, hence the coefficient rcs ( sin, j ).
rcs ( sin, j ) mu ( sin, j , p ) = msout ( s, j, p ) + å mt ( s, j¢, j, p - 1)

j ¢ÎJ sp
"j Î J , c
s p Î P, p > 1, s Î S, sin, j Î Sineff, j (4.1)

rs ( sin, j ) mu ( sin, j , p ) = msout ( s, j, p ) "j Î J s , p Î P, p = 1, s Î S, sin, j Î Sineff, j

c c
(4.2)
Constraint (4.3) describes the outlet mass balance for unit j at time point p. The
output state, an intermediate or a final product, may be sent to storage or downstream
consuming units j′. Again, this constraint takes into account the implications of stoi-
chiometry through rsp ( sin, j ).
rsp ( sin, j ) mu ( sin, j , p ) = msin ( s, j, p ) + å mt ( s, j, j¢, p )

j ¢ÎJ sc
"j Î J sp , p Î P, s Î S , sin, j Î Sineff, j (4.3)
Constraint (4.4) stipulates that the total amount of states consumed in the task
at time point p cannot exceed the capacity of the processing unit. This constraint is
only active if a particular task is active in unit j. In the absence of a task in a unit,
this constraint ensures that no material is processed. It is also worth noting that the
minimum amount of material to be processed for a particular task in a unit need not
be zero. These last two conditions explain the presence of a binary variable in the
constraint.
V jL y ( sin, j , p ) £ mu ( sin, j , p ) £ V jU y ( sin, j , p ) "j Î J , p Î P, sin, j Î Sineff, j (4.4)
Constraints (4.5) and (4.6) describe the inlet energy balance for processing units.
It is important to note that if unit j uses a raw material (s ∈ Sr), the correspond-
ing Tsout (s, p) term on the right-hand side of both constraints will be replaced by
( )
T f sinr , j , p or Ts0(s) for p = 1, since the raw material stream may have heat exchange
before entering unit j from the storage. On the other hand, the outlet energy balance
need not be performed because all the output states are assumed to exit the unit at
the same temperature, Tout (sin,j, p), as mentioned earlier.
mu ( sin, j , p ) Cpin ( sin, j ) Tin ( sin, j , p ) = å ms

sÎS cj
out ( s, j, p ) Cp ( s ) Tsout ( s, p )
+ ååmt ( s, j¢, j, p - 1) Cp ( s ) T ( s
sÎS cj j ¢ÎJ sp
f out , j ¢ , p - 1)
"j Î J , p Î P, p > 1, sin, j Î Sineff, j , sout , j ¢ Î Sout , j (4.5)

mu ( sin, j , p ) Cpin ( sin, j ) Tin ( sin, j , p ) = å ms

sÎS cj
out ( s, j, p ) Cp ( s ) Tsout ( s, p )
"j Î J , p Î P, p = 1, sin, j Î Sineff, j (4.6)
where

Cpin ( sin, j ) = år (s
sÎS
c
s in, j ) Cp ( s ) .

4.5.2 Mass and Energy Balances for Storage

Constraints (4.7) through (4.12) describe the mass balances for the storage of raw
material, intermediate and product states. All these constraints are formulated such
that the amount of material stored at time point p is the amount stored at the previous
time point (p − 1) adjusted by the amounts consumed and/or produced by the tasks.
Note that these constraints assume different forms depending on whether they apply
to a feed, intermediate or final product stream/state. The main distinction between
intermediates and products is that the latter are only stored after production, whereas
the former may be stored or consumed after production.
qs ( s, p ) = qs ( s, p - 1) - å ms out ( s, j, p ) "p Î P, p > 1, s Î S r (4.7)

jÎJ sc
qs ( s, p ) = Q0 ( s ) - å ms out ( s, j, p ) "p Î P, p = 1, s Î S r (4.8)

jÎJ sc
qs ( s, p ) = qs ( s, p - 1) + å ms in ( s, j, p - 1)
jÎJ sp
- åms
jÎJ sc
out ( s, j, p ) "p Î P, p > 1, s Î S i (4.9)

qs ( s, p ) = Q0 ( s ) - å ms out ( s, j, p ) "p Î P, p = 1, s Î S i (4.10)

jÎJ sc
qs ( s, p ) = qs ( s, p - 1) + å ms
jÎJ sp
in ( s, j, p ) "p Î P, p > 1, s Î S p (4.11)

qs ( s, p ) = Q0 ( s ) + å msjÎJ sp
in ( s, j, p ) "p Î P, p = 1, s Î S p (4.12)

Constraints (4.13) and (4.14) state that the amount of state s stored at any time
point cannot exceed the maximum capacity of the storage. It is imperative that
constraint (4.14) takes into account the amount of material produced at time
point p.
qs ( s, p ) £ QsU "p Î P, s Î S r È S p (4.13)
qs ( s, p ) + å ms
jÎJ sp
in ( s, j, p ) £ QsU "p Î P, s Î S i (4.14)

Constraints (4.15) and (4.16) are analogous to previous Constraints (4.5) and
(4.6), but only apply to intermediate states in storage. Similar constraints for raw
material and product storage are not needed, assuming that raw materials are only
sent out for processing at their supply temperatures and final products only sent in at
their storage temperatures.
qs ( s, p ) Tsout ( s, p ) = qs ( s, p - 1) Tsout ( s, p - 1) + å ms in ( s, j, p - 1) T f ( sout , j , p - 1)

jÎJ sp

- å ms
jÎJ sc
out ( s, j, p ) Tsout ( s, p ) "p Î P, p > 1, s Î S i (4.15)

qs ( s, p ) Tsout ( s, p ) = Q0 ( s ) Ts0 ( s ) - å ms
jÎJ sc
out ( s, j, p ) Tsout ( s, p )
"p Î P, p = 1, s Î S i (4.16)
4.5.3 Heating and Cooling in Processing Units

As mentioned earlier, the input and output materials may be heated or cooled in
processing units before and after tasks. Such potential heating and cooling require-
ments can be identified from the temperature data of the production. For example, if
a processing unit (j = 1) operating at a temperature of T1 produces an intermediate to
be further processed in a downstream unit (j = 2) where the operating temperature
is higher (T2 > T1), the intermediate produced at T1 needs to be heated to T2. This
heating can take place either in the first or the second unit using external hot utility,
or during material transfer between the units through heat exchange.
For processing units that may have heating for their input materials, Constraint
(4.17) is used to calculate the heat load, which can be zero for an active task, indicat-
ing that the heating has been finished elsewhere. In this case, the temperature of the
input material is the same as the operating temperature.
(
qinh ( sin, j , p ) = mu ( sin, j , p ) Cpin ( sin, j ) Top ( sin, j ) - Tin ( sin, j , p ) )
"j Î J inh , p Î P, sin, j Î Sineff, j (4.17)
For units that may need cooling for their input materials, Constraint (4.18) is used:
(
qinc ( sin, j , p ) = mu ( sin, j , p ) Cpin ( sin, j ) Tin ( sin, j , p ) - Top ( sin, j ) )
"j Î J inc , p Î P, sin, j Î Sineff, j (4.18)
If either heating or cooling would be needed, for example, in the case of mul-
tiple input states with varying temperatures, constraints (4.19) through (4.23) can
be used. Constraint (4.19) states that unit j may have either heating or cooling, but
not both, for its input material, whenever the task is active (y(sin,j, p) = 1). When
( )
heating is to take place yinh ( sin, j , p ) = 1; yinc ( sin, j , p ) = 0 , Constraint (4.20) quantifies
the heat load, while Constraints (4.21) and (4.22) become redundant and Constraint
(4.23) forces the cooling load to be zero. Overall, this situation reduces Constraint
(4.20) to Constraint (4.17). Similarly, when cooling is to take place (yinh ( sin, j , p ) = 0;
yinc ( sin, j , p ) = 1), Constraints (4.20) and (4.23) become redundant, Constraint (4.21)
forces the heating load to be zero and Constraint (4.22) reduces to Constraint (4.18).
yinh ( sin, j , p ) + yinc ( sin, j , p ) = y ( sin, j , p ) "j Î J in*, p Î P, sin, j Î Sieff

n, j (4.19)
( ) (
QU yinh ( sin, j , p ) - 1 £ qinh ( sin, j , p ) - mu ( sin, j , p ) Cpin ( sin, j ) Top ( sin, j ) - Tin ( sin, j , p ) )
(
£ QU 1 - yinh ( sin, j , p ) ) "j Î J in*, p Î P, sin, j Î Sineff, j (4.20)

qinh ( sin, j , p ) £ QU yinh ( sin, j , p ) "j Î J in*, p Î P, sin, j Î Sineff, j (4.21)
( ) (
QU yinc ( sin, j , p ) - 1 £ qinc ( sin, j , p ) - mu ( sin, j , p ) Cpin ( sin, j ) Tin ( sin, j , p ) - Top ( sin, j ) )

£Q U
(1 - y ( s
c
in in, j , p) ) "j Î J in*, p Î P, sin, j Î S eff
in, j (4.22)

qinc ( sin, j , p ) £ QU yinc ( sin, j , p ) "j Î J in* , p Î P, sin, j Î Sineff, j (4.23)
Constraints (4.24) and (4.25) are used for processing units that may have heating
and those that may have cooling for their output materials, respectively.
h
qout (
( sin, j , p ) = mu ( sin, j , p ) Cpout ( sin, j ) Tout ( sin, j , p ) - Top ( sin, j ) )
"j Î J , h
out p Î P, sin, j Î S eff
in, j (4.24)
c
qout (
( sin, j , p ) = mu ( sin, j , p ) Cpout ( sin, j ) Top ( sin, j ) - Tout ( sin, j , p ) )
"j Î J , c
out p Î P, sin, j Î S eff
in, j (4.25)
where Cpout ( sin, j ) =

sÎS å
rsp ( sin, j ) Cp ( s ). If either heating or cooling would be
needed, for example, in the case of multiple output states proceeding to different
destinations with varying target temperatures, Constraints (4.26) through (4.30) can
be used. Worth the emphasis is the fact that Constraints (4.26) through (4.30) are
directly analogous to Constraints (4.19) through (4.23), with the latter applicable to
input, and the former to output states.

h
yout ( sin, j , p ) + yout
c
( sin, j , p ) £ y ( sin, j , p ) "j Î J out
*, p Î P, sin, j Î Sineff, j (4.26)
QU yout
h
(
( sin, j , p ) - 1 £ qout
h
) (
( sin, j , p ) - mu ( sin, j , p ) Cpout ( sin, j ) Tout ( sin, j , p ) - Top ( sin, j ) )
(
£ Q U 1 - yout
h
( sin, j , p ) ) *,
"j Î J out p Î P, sin, j Î Sineff, j (4.27)

h
qout ( sin, j , p ) £ QU yout
h
( sin, j , p ) "j Î J out* , p Î P, sin, j Î Sineff, j (4.28)
QU yout
c
(
( sin, j , p ) - 1 £ qout
c
) (
( sin, j , p ) - mu ( sin, j , p ) Cpout ( sin, j ) Top ( sin, j ) - Tout ( sin, j , p ) )
(
£ QU 1 - yout
c
( sin, j ,p ) ) *,
"j Î J out p Î P, sin, j Î Sineff, j (4.29)

c
qout ( sin, j , p ) £ QU yout
c
( sin, j , p ) "j Î J out* , p Î P, sin, j Î Sineff, j (4.30)
Constraint (4.31) gives the time required for heating, while Constraint (4.32)
gives the time required for cooling of input materials in processing units dedicated
to either heating or cooling. Constraint (4.33), on the other hand, applies to those
units that are capable of both heating and cooling, depending on the nature of input
materials.
(
trin ( sin, j , p ) = qinh ( sin, j , p ) M st l st ) "j Î J inh , p Î P, sin, j Î Sineff, j

(4.31)

trin ( sin, j , p ) = qinc ( sin, j , p ) ( M cw
Cpcw Tout
cw
(
- Tincw )) "j Î J inc , p Î P, sin, j Î Sineff, j

(4.32)
( (
trin ( sin, j , p ) = qinh ( sin, j , p ) M st l st )) + ( q ( s c
in in, j , p) ( M cw
(
Cpcw Tout
cw
- Tincw )))
"j Î J in*, p Î P, sin, j Î S eff
in, j

(4.33)
Constraints (4.34) through (4.36) are directly analogous to Constraints (4.31)

through (4.33), but apply to output states.

trout ( sin, j , p ) = qout
h
(
( sin, j , p ) M st l st ) "j Î J out
h
, p Î P, sin, j Î Sieff
n, j

(4.34)

c
( sin, j , p ) ( M cw
Cpcw Tout
cw
(
- Tincw )) "j Î J ocut , p Î P, sin, j Î Sineff, j

(4.35)

h
(
( sin, j , p ) M st l st ( )) + ( q ( s c
out in, j , p) ( M cw
Cpcw Tout
cw
(- Tincw )))
*,
"j Î J out p Î P, sin, j Î S eff
(4.36)
in, j
Taking into account the heating and cooling times in processing units, the dura-
tion constraint is formulated as in Constraint (4.37). Note that the processing time is
assumed to be constant and independent of the batch size. The inequality caters for
further residence time in the processing unit in the case of unavailable intermediate
storage. The latter could be due to deliberate omission of intermediate storage in
design or unavailability due to finite intermediate storage.
t p ( sin, j , p ) ³ tu ( sin, j , p ) + trin ( sin, j , p ) + t ( sin, j ) y ( sin, j , p ) + trout ( sin, j , p )
"j Î J , p Î P, sin, j Î Sineff, j (4.37)

4.5.4 Heat Integration Constraints

In order to reduce utility consumption, heat exchange among output and raw mate-
rial streams is considered. As in the case of heating and cooling in processing units,
the potential hot and cold streams can be identified from the temperature data.
For temperature assignment, Constraint (4.38) states that the initial temperature
of an outlet (intermediate or product) stream for heat exchange is the temperature at
which it leaves the processing unit.
Ti ( sout , j , p ) = Tout ( sin, j , p ) "j Î J , p Î P, sin, j Î Sineff, j , sout , j Î Sout , j (4.38)
Constraints (4.39) through (4.41) state that an outlet stream can be heat integrated
with another outlet stream or a raw material stream after leaving the producing unit,
and that a raw material stream may exchange heat with an outlet stream for preheat-
ing before entering the consuming unit. Additionally, there will be no heat exchanged
if the corresponding unit is not active (y(sin,j, p) = 0). It should be noted that raw mate-
rial pre-cooling is not considered in this work, not common in practice. Thus, raw
material streams will only act as cold streams in the context of heat integration. Note
also that the one-to-one heat integration arrangement imposed by these constraints is
designed to simplify process operation based on practical considerations.
å x (s
sout ,j¢ ÎS c*
out , j , sout , j ¢ , p ) +
r
å x (s c
out , j )
, sinr , j ¢ , p £ y ( sin, j , p )
sin , j¢ ÎS *
"j Î J , p Î P, sin, j Î Sineff, j , sout , j Î S h (4.39)

*
å x (s
sout , j¢ ÎS h*
out , j ¢ , sout , j , p ) £ y ( sin, j , p ) "j Î J , p Î P, sin, j Î Sineff, j , sout , j Î S c (4.40)
*

å x (s
sout , j¢ ÎS*h
out , j ¢ )
, sinr , j , p - 1 £ y ( sin, j , p ) "j Î J , p Î P, p > 1,, sin, j Î Sineff, j , sinr , j Î S*c

(4.41)
The amounts of heat exchanged between hot and cold streams are given by
Constraints (4.42) through (4.44). For non-existing matches, Constraint (4.45)
ensures zero heat loads.
å q (s
s c ÎS*c
ex out , j ) (
, s*c , p = rsp ( sin, j ) mu ( sin, j , p ) Cp ( s ) Ti ( sout , j , p ) - T f ( sout , j , p ) )
*
"j Î J sp , p Î P, s Î S, sin, j Î Sineff, j , sout , j Î S h (4.42)

*
åq
sout , j¢ ÎS h
ex ( sout , j¢ , sout , j , p ) = rsp ( sin, j ) mu ( sin, j , p ) Cp ( s ) (T f ( sout , j , p ) - Ti ( sout , j , p ) )
*
"j Î J sp, p Î P, s Î S, sin, j Î Sineff, j , sout , j Î S c (4.43)

*
å q (s ex out , j ¢ ) ( ( )
, sinr , j , p - 1 = rcs ( sin, j ) mu ( sin, j , p ) Cp ( s ) T f sinr , j , p - Ts0 ( s ) )
sout , j¢ ÎS h
*
"j Î J sc , p Î P, p > 1, s Î S r , sin, j Î Sineff, j , sinr , j Î S*c (4.44)

* *( )
qex s h , s c , p £ QU x s h , s c , p
* * ( ) "p Î P, s h Î S h , s c Î S c
* * * *
(4.45)
where s*h only represents sout,j, while s*c can be sout,j or sinr , j. Constraints (4.46) through
(4.49) ensure feasible temperature driving forces for heat exchange between a pair of
outlet streams and between an outlet and a raw material stream, respectively.
(
Ti ( sout , j , p ) - T f ( sout , j ¢ , p ) ³ DTmin - G 1 - x ( sout , j , sout , j ¢ , p ) )
"j, j¢ Î J , p Î P, sout , j Î S*, sout , j ¢ Î S*
h c
(4.46)
(
T f ( sout , j , p ) - Ti ( sout , j ¢ , p ) ³ DTmin - G 1 - x ( sout , j , sout , j ¢ , p ) )
"j, j¢ Î J , p Î P, sout , j Î S*, sout , j ¢ Î S*
h c
(4.47)
( ) ( (
Ti ( sout , j , p - 1) - T f sinr , j ¢ , p ³ DTmin - G 1 - x sout , j 0, sinr , j ¢ , p - 1 ))
"j, j¢ Î J , p Î P, p > 1, sinr , j ¢ Î S*c , sout , j Î S*h (4.48)

( (
T f ( sout , j , p ) - Ts0 ( s ) ³ DTmin - G 1 - x sout , j , sinr , j ¢ , p ))
"j Î J , j¢ Î J sc , p Î P, s Î S r , sinr , j ¢ Î S*c , sout , j Î S*h (4.49)

It is assumed that the temperatures of available hot and cold utilities are adequate
to achieve the desired heating and cooling. While the temperature of the input mate-
rial is bounded by the operating temperature, the highest and lowest achievable tem-
peratures of output materials are limited by utility temperatures. This is given in
Constraints (4.50) and (4.51).

st
Tsat - Tout ( sin, j , p ) ³ DTmin "j Î J out
h
, p Î P, sin, j Î Sineff, j (4.50)
out ( in, j ) in (
T s , p - T cw ³ DTmin - G 1 - y ( sin, j , p ) ) "j Î J out
c
, p Î P, sin, j Î Sineff, j (4.51)

To ensure that the product achieves its storage temperature, Constraints (4.52) and
(4.53) are used for product streams to have and not to have heat exchange, respectively:
T f ( sout , j , p ) £ Tstor ( s ) "j Î J sp , p Î P, s Î S p , sout , j Î Sout , j (4.52)
Tout ( sin, j , p ) £ Tstor ( s ) "j Î J sp , p Î P, s Î S p , sin, j Î Sineff, j (4.53)
The following constraints are necessary for heat integration ensuring that the two
process streams appear at the same time if there is a heat exchange between them.
Constraints (4.54) and (4.55) are applicable to outlet stream matches and Constraints
(4.56) and (4.57) to output/raw material stream matches.
(
t p ( sin, j , p ) ³ t p ( sin, j ¢ , p ) - H 1 - x ( sout , j , sout , j ¢ , p ) )
"j, j¢ Î J , p Î P, sin, j , sin, j ¢ Î S eff
in, j , sout , j Î S*h , sout , j ¢ Î S*
c
(4.54)
(
t p ( sin, j , p ) £ t p ( sin, j ¢ , p ) + H 1 - x ( sout , j , sout , j ¢ , p ) )
"j, j¢ Î J , p Î P, sin, j , sin, j ¢ Î S eff
in, j , sout , j Î S*h , sout , j ¢ Î S*
c
(4.55)
( (
t p ( sin, j , p - 1) ³ tu ( sin, j ¢ , p ) - tt - H 1 - x sout , j , sinr , j ¢ , p - 1 ))
"j, j¢ Î J , p Î P, p > 1, sin, j , sin, j ¢ Î S eff
in, j , sout , j Î S*h , sinr , j ¢ Î S*c (4.56)

( (
t p ( sin, j , p - 1) £ tu ( sin, j ¢ , p ) - tt + H 1 - x sout , j , sinr , j ¢ , p - 1 ))
"j, j¢ Î J , p Î P, p > 1, sin, j , sin, j ¢ Î S eff
in, j , sout , j Î S*h , sinr , j ¢ Î S*c (4.57)

For profit maximization, the objective function may be formulated as the difference
between the product revenue (PR) and the costs of raw materials (RC) and external
utilities (UC), as shown in Equation 4.58. It is evident that Equation 4.58 is aimed at
reducing the utility costs, which is concomitant with maximizing process–process

heat integration.
max Profit = PR - RC - UC (4.58)
where

PR = åPrice ( s ) qs ( s, p )
sÎS p
p= P

(4.59)
RC = åååCost ( s ) ms out ( s, j, p ) (4.60)

sÎS r jÎJ sc pÎP
æ ö
UC =
pÎP
å
CHU çç
å(
ç sin, j ÎSineff, j
H H
)
qin ( sin, j , p ) + qout ( sin, j , p ) ÷÷
÷
è ø
æ ö
+
pÎP
å
CCU çç å(
ç sin, j ÎSineff, j
qinC ( sin, j , p ) + qout
C
( in, j ) ÷÷
s , p
÷
) (4.61)
è ø
Due to the presence of bilinear terms in Constraints (4.5), (4.6), (4.15) through
(4.18), (4.20), (4.22), (4.24), (4.25), (4.27), (4.29) and (4.42) through (4.44) and the
use of binary variables, the overall model is a mixed integer nonlinear program
(MINLP), for which global optimality may not be guaranteed. In the next section,
the application of the presented model is demonstrated through a modified literature
case study. The model is implemented in the GAMS environment (Rosenthal, 2015)
on a Core i7–4790, 3.60 GHz processor with BARON (Tawarmalani and Sahinidis,
2005) as the MINLP solver.
4.6 ILLUSTRATIVE CASE STUDY

A multipurpose batch plant is to be scheduled to produce three products (s9–s11)
from three raw materials (s1, s2 and s7) over a 16-h time horizon. The STN (state-
task network) process representation is shown in Figure 4.4, where states s3–s6 and
s8 are all intermediates. This case study is adapted from Pinto et al. (2008), with
added heating and cooling requirements of materials to explore heat integration
opportunities. Tables 4.1 and 4.2 present the pertinent data for the case study. For
simplicity, it is assumed that each state has a dedicated storage vessel and each pro-
cessing task is performed in a dedicated reactor. Note that the supply temperature in
Table 4.2 refers to the initial temperature of a state in the storage vessel. Therefore,
the supply temperatures of intermediates and products are taken as zero.
It can be seen in Figure 4.4 that unit R1 is likely to have heating for its input
s1 and cooling for its output s4, as the supply temperature of s1 is lower than the
60°C s7
0.5
Task I Task IV 0.5 Task V

80°C s1 s4 s6 s11 80°C
R1 (180°C) 0.6 R4 (120°C) R5 (160°C) 0.3
0.4 0.7
s5 s8
Task II Task III Task VI

40°C s2 s3 s9 40°C
R2 (40°C) R3 (80°C) R6 (100°C) 0.5
0.5
s10 40°C
FIGURE 4.4 STN representation for the case study.
TABLE 4.1
Processing Data for the Case Study
Operating
Task Unit Duration (h) Temperature (°C) Capacity Limits (kg)
I R1 2 180 30–120
II R2 2 40 20–80
III R3 4 80 20–80
IV R4 2 120 50–200
V R5 1 160 50–200
VI R6 1 100 35–140
operating temperature of R1, and the latter is higher than the operating temperature
of R4. Similarly, potential heating requirements for the inputs of R3 and R5 and the
outputs of R2–R4 can be identified from Figure 4.4, together with potential cooling
requirements for the outputs of R5 and R6. However, some uncertainties remain for
R4 and R6. The input of R4 consists of s4 and s5, which are produced at different
temperatures and can give a mixture temperature higher or lower than its operating
temperature. In the case of R6, if the product storage temperature of s11 (80°C)
is achieved by utility cooling inside R5, then s8 will leave R5 at 80°C, which is
lower than the operating temperature of R6 (100°C). However, if product cooling
is achieved by heat exchange during the transfer, the final temperature of s8 may be
higher than the operating temperature of R6. Therefore, constraints (4.19) through
(4.23) are used for both units R4 and R6, while constraints (4.26) through (4.30) are
not needed in this case study.
The hot and cold streams for heat exchange can also be identified from Figure 4.4.
There are three potential hot streams (s4outR1, s8outR5 and s11outR5) and five
TABLE 4.2
Material and Utility Data for the Case Study
Product
Storage Initial Supply Storage Heat
Capacity Inventory Temperature Temperature Capacity
State (kg) (kg) (°C) (°C) (kJ/kg °C) Price ($/kg)
s1 UL AA 80 — 2.5 0.5
s2 UL AA 40 — 3.4 0.5
s3 160 0 0 — 3.3 —
s4 240 0 0 — 3.7 —
s5 160 0 0 — 4.0 —
s6 400 0 0 — 3.6 —
s7 UL AA 60 — 3.5 0.5
s8 280 0 0 — 4.0 —
s9 UL 0 0 40 3.2 10
s10 UL 0 0 40 3.3 5
s11 UL 0 0 80 3.7 10
Material transfer time 10 min
Minimum temperature difference for heat transfer 10°C
Cooling water cost $0.6/MJ
Cooling water inlet/outlet temperature 20/30°C
Heat capacity of cooling water 4.2 kJ/kg °C
Mass flow rate of cooling water 1800 kg/h
Steam cost $3/MJ
Saturated steam temperature 200°C
Latent heat of steam 1940 kJ/kg
Mass flow rate of steam 40 kg/h
Note: UL, unlimited; AA, available as/when required.
potential cold streams (s1inR1, s3outR2, s5outR3, s6outR4 and s7inR5). Note that
the feed stream to R2 (s2inR2) and the product streams from R6 (s9outR6 and
s10outR6) are left out; the former has no need for heating or cooling, while storage
temperatures of the latter can only be achieved with cooling water. The objective is
to maximize the profit given by the difference between product revenue and operat-
ing cost associated with raw materials and external utilities.
Batch operations are normally carried out with fixed batch sizes in the process
industries in order to minimize changes. Therefore, batch sizes of all processing
tasks are fixed at the maximum capacities given in Table 4.1. Prior to exploring heat
integration opportunities, the batch plant produces 140 kg of s9, 140 kg of s10 and
120 kg of s11 within the 16-h time horizon, while consuming 161.76 MJ of steam
and 193.56 MJ of cooling water. This gives a profit of $2498.58, obtained from the
presented model by setting all the binary variables associated with heat exchange to
zero. Figure 4.5 shows the Gantt chart of production without heat integration. Note
that the heating and cooling required in R5 increase the material residence time to
R6 140 140
R5 200 200
R4 200
R3 80
Heating Cooling
R2 80
Processing/holding Transfer
R1 120
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Time (h)
FIGURE 4.5 Gantt chart of production without heat integration.
at least 2.47 h, which is much longer than the task duration of 1 h. Thus, there is not
enough time to process one more batch in R5.
When heat exchange between outlet streams (three hot, s4outR1, s8outR5 and
s11outR5; three cold, s3outR2, s5outR3 and s6outR4) is considered, the plant can
produce 140 kg of s9, 140 kg of s10 and one batch more of s11 (i.e. 180 kg) within
the same time horizon. However, a higher utility cost of $661.97 for 188.16 MJ of
steam and 162.48 MJ of cooling water is involved with increased production. The
profit in this case is $2808.03, corresponding to a 12.38% improvement compared
to the case without heat integration. Figure 4.6 shows the optimal schedule with
Heating Cooling
R6 140 140
R5 200 200 200
R4 200 200
R3 80 80
R2 80 80 80 80
R1 120 120
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Time (h)
FIGURE 4.6 Optimal schedule with heat exchange between outlet streams.
heat exchange between outlet streams. It can be seen that almost double the number
of batches are processed (increased from 8 to 15), but some are only produced for
heat exchange, namely the last two batches in R2. These surplus intermediates may
be used later over an extended time horizon. The detailed heat exchange results are
shown in Tables 4.3 and 4.4. There are six matches of heat exchange to reduce the
heating and cooling required, especially in R1 and R5 where the cooling needs can
even be eliminated.
When raw material streams (two additional cold, s1inR1 and s7inR5) are also
considered for heat exchange, the product revenue can be increased to $4500, with
the production of another batch of s11, and the utility cost reduced to $525.24,
giving a profit of $3534.76. This corresponds to a further 29.09% improvement
compared to the previous cases. Figure 4.7 shows the optimal schedule with
heat exchange among raw material and outlet streams, with Tables 4.5 and 4.6
TABLE 4.3
Heat Exchange between Outlet Streams
Time Point
Stream match
(Hot-cold) p = 2 p = 3 p = 4 p = 5
s4outR1-s3outR2 10,560
s11outR5-s3outR2 17,760 17,760
Note: All values given in kJ.
TABLE 4.4
Initial-Final Temperatures of Outlet Streams for Heat Exchange
Time Point
Stream p = 1 p = 2 p = 3 p = 4 p = 5 p = 6
s3outR2 40–40 40–80 40–107.27 40–107.27
s4outR1 172.61–148.83 180–130
s5outR3 80–80 80–106.13
s6outR4 120–150.83 120–143.33
s8outR5 160–145.07 160–130 80–80
s11outR5 160–80 160–80 80–80
Note: All values given in °C.

Heating Cooling
R6 140 140
R5 200 200 200 200
R4 200 200
R3 80 80
R2 80 80 80
R1 120 120
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Time (h)
FIGURE 4.7 Optimal schedule with heat exchange among raw material and outlet
streams.
TABLE 4.5
Heat Exchange among Raw Material and Outlet Streams
Time Point
Stream Match
(Hot-cold) p = 2 p = 3 p = 4 p = 5 p = 6 p = 7
s4outR1-s1inR1 13,840
s4outR1-s7inR5 36,240
s11outR5-s7inR5 17,760 17,760 17,760
Note: All values given in kJ.
presenting the detailed heat exchange results. Note that raw material streams take
priority over the other cold streams as the former involve larger enthalpy changes.
Consequently, streams s3outR2 and s6outR4 are less involved in heat integra-
tion, and there are fewer batches produced in R2. The computational results
for this case study, including model size and solution time, are summarized in
Table 4.7 An overall comparison shows that heat integration of material transfer
streams enables the batch plant to achieve higher production with lower utility
consumption.
102
TABLE 4.6
Initial-Final Temperatures of Raw Material and Outlet Streams for Heat Exchange
Time Point
Stream p = 1 p = 2 p = 3 p = 4 p = 5 p = 6 p = 7
s1inR1 80–80 80–126.13
s3outR2 80–80 80–80 40–107.27

s4outR1 180–148.83 180–98.38
s5outR3 80–80 80–150
s6outR4 120–120 120–143.33
s7inR5 60–163.54 60–110.74 60–110.74 60–110.74
s8outR5 160–120 160–130 160–160 160–160
s11outR5 160–80 160–80 160–80 160–80
Note: All values given in °C.

TABLE 4.7
Computational Results for the Case Study
Heat Integration of
Without Heat Raw Material/Outlet
Integration Outlet Streams Streams
Profit ($) 2498.58 2808.03 3534.76
Improvement — 12.38% 41.47%
Product revenue ($) 3300 3900 4500
s9 production (kg) 140 140 140
Raw material cost ($) 200 430 440
Utility cost ($) 601.42 661.97 525.24
Steam (MJ) 161.76 188.16 158.56
Cooling water (MJ) 193.56 162.48 82.6
Model statistics
No. of time points 6 6 7
No. of constraints 1492 1492 1966
No. of variables 1029 1029 1296
No. of binaries 150 204 274
Solution time (CPU s) 62 166 2311
4.7 SUMMARY
A superstructure-based mathematical model for simultaneous process scheduling
and heat integration of multipurpose batch plants has been presented in this chap-
ter. Based on a robust scheduling framework adapted from Seid and Majozi (2012),
the MINLP formulation ensures proper sequencing of tasks over the time hori-
zon of interest, with the aim of synchronizing the material transfer times in order
to maximize heat recovery. An illustrative case study was solved to demonstrate
the application of the proposed model. The results indicate that heat integration
of process streams allows batch plants to achieve considerably higher production
(+36.36%) at a much lower utility cost (−12.67%). It should be noted that the pre-
sented formulation assumes intermediate streams to exchange heat only before
storage, and external utilities to be used only to the unit jacket. These simplifying
assumptions can be relaxed for even better results. On the other hand, direct and
indirect heat integration between processing units may be worth considering in
future investigation.
4A APPENDIX
The necessary assignment, sequence and time horizon constraints for scheduling are
adapted from Seid and Majozi (2012) and presented here.
4A.1 Assignment Constraint
Constraint (4A.1) states that at any time point, only one task can be performed in
a unit. This constraint is only needed if more than one task can be carried out in a
given unit.
å eff
sin , j ÎSin
y ( sin, j , p ) £ 1 "j Î J , pÎP (4A.1)
,j
4A.2 Sequence Constraint for the Same Task in the same unit

Constraint (4A.2) states that a new task can only start in a processing unit after the
previous task is complete and the product material removed from the unit.
(
tu ( sin, j , p ) ³ t p ( sin, j , p - 1) + tt y ( sin, j , p - 1) ) "j Î J , p Î P, p > 1, sin, j Î Sineff, j

(4A.2)
4A.3 Sequence Constraints for Storage

Constraints (4A.3) through (4A.6) ensure that the times at which a state is transferred
to and from the storage at a given time point are later than those at the previous time
point.
tmin ( s, p ) ³ tmin ( s, p - 1) "p Î P, p > 1, s Ï S r (4A.3)
tmin ( s, p ) ³ tmout ( s, p - 1) "p Î P, p > 1, s Î S i (4A.4)
tmout ( s, p ) ³ tmin ( s, p - 1) "p Î P, p > 1, s Î S i (4A.5)
tmout ( s, p ) ³ tmout ( s, p - 1) "p Î P, p > 1, s Ï S p (4A.6)
4A.4 Sequence Constraints for Processing Units and Storage

The upper bound for the amount of material sent to storage is given by Constraint
(4A.7). When a state is transferred from its producing unit to the storage, Constraints
(4A.8) and (4A.9) ensure that the material transfer takes place upon completion of
the producing task.
msin ( s, j, p ) £ V jU win ( s, j, p ) "j Î J sp , p Î P, s Ï S r (4A.7)
( ) (
t p ( sin, j , p ) ³ tmin ( s, p ) - tt win ( s, j, p ) - H 1 - win ( s, j, p ) )
"j Î J , s
p
p Î P, s Ï S , sin, j Î S r eff
in, j (4A.8)
( ) (
t p ( sin, j , p ) £ tmin ( s, p ) - tt win ( s, j, p ) + H 1 - win ( s, j, p ) )
"j Î J , s
p
p Î P, s Ï S , sin, j Î S r eff
in, j (4A.9)
The upper bound for the amount of material sent from storage is given by
Constraint (4A.10). When a state is transferred from the storage to the consuming
unit, Constraints (4A.11) and (4A.12) ensure that the consuming task starts right after
the material transfer.
msout ( s, j, p ) £ V jU wout ( s, j, p ) "j Î J sc , p Î P, s Ï S p (4A.10)
( ) (
tu ( sin, j , p ) ³ tmout ( s, p ) + tt wout ( s, j, p ) - H 1 - wout ( s, j, p ) )
"j Î J , c
s p Î P, s Ï S , sin, j Î S
p eff
in, j (4A.11)
( ) (
tu ( sin, j , p ) £ tmout ( s, p ) + tt wout ( s, j, p ) + H 1 - wout ( s, j, p ) )
"j Î J , c
s p Î P, s Ï S , sin, j Î S
p eff
in, j (4A.12)
4A.5 Sequence Constraints for Producing and Consuming Units

The upper bound for the amount of material sent between processing units is given
by Constraint (4A.13). When a state is transferred from its producing unit to its con-
suming unit, Constraints (4A.14) and (4A.15) ensure that the material transfer is car-
ried out between the end of the producing task and the start of the consuming task.
mt ( s, j, j¢, p ) £ V jU z ( j, j¢, p ) "j Î J sp , j¢ Î J sc , p Î P, s Î S i (4A.13)
( ) (
tu ( sin, j ¢ , p ) ³ t p ( sin, j , p - 1) + tt z ( j, j¢, p - 1) - H 1 - z ( j, j¢, p - 1) )
"j Î J , s
p
j¢ Î J sc , p Î P, p > 1, sin, j , sin, j ¢ Î S eff
in, j (4A.14)
( ) (
tu ( sin, j ¢ , p ) £ t p ( sin, j , p - 1) + tt z ( j, j¢, p - 1) + H 1 - z ( j, j¢, p - 1) )
"j Î J , s
p
j¢ Î J ,c
s p Î P, p > 1, sin, j , sin, j ¢ Î S eff
in, j (4A.15)
4A.6 Time Horizon Constraints

The use, production and transfer of states should all be finished within the time
horizon of interest, as given in Constraints (4A.16) through (4A.19). Note that the
material transfer time is taken into account.
tu ( sin, j , p ) £ H - tt "j Î J , p Î P, sin, j Î Sin, j (4A.16)
t p ( sin, j , p ) £ H - tt "j Î J , p Î P, sin, j Î Sin, j (4A.17)
tmin ( s, p ) £ H "p Î P, s Ï S r (4A.18)
tmout ( s, p ) £ H - tt "p Î P, s Ï S p (4A.19)

NOMENCLATURE
Indices and Sets
jÎJ Processing units
j Î J in
c
Processing units to have cooling for input material
j Î J out
c
Processing units to have cooling for output material
j Î Js c
Processing units consuming state s
j Î J in
h
Processing units to have heating for input material
j Î J out
h
Processing units to have heating for output material
j Î J sp Processing units producing state s
*
j Î J in Processing units to have heating or cooling for input material
*
j Î J out Processing units to have heating or cooling for output
material
p Î P Time points
s Î S States
s Î S cj States consumed in unit j
s Î S i Intermediate states
s Î S p Product states
sÎS r
Raw material states
sin, j Î Sin, j Inlet streams
sin, j Î Sineff, j Effective states representing tasks
sout , j Î Sout , j Outlet streams
sinr , j Î Sinr , j Raw material streams
s Î S Ì Sout , j È S Cold streams
c
*
c
*
r
in, j
s*h Î S*h Ì Sout , j Hot streams
Parameters
CCW Cooling water cost
Cp ( s ) Specific heat capacity of state s
Cpin ( sin, j ) Specific heat capacity of the input material to the task
Cpout ( sin, j ) Specific heat capacity of the output material from the task
Cpcw Specific heat capacity of cooling water
CST Steam cost
M cw Constant cooling water flow rate through the jacket
M st Constant steam flow rate through the jacket
Q0 ( s ) Initial amount of state s in the storage
QU Upper bound for heat loads
QsU Maximum storage capacity for state s

Top ( sin, j ) Operating temperature for the task
Tstor ( s ) Storage temperature of product state s
Tincw Inlet temperature of cooling water
cw
Tout Outlet temperature of cooling water
st
Tsat Saturated steam temperature
Ts0 ( s ) Initial temperature of state s in the storage
tt Material transfer time
V jL Minimum capacity of unit j
V jU Maximum capacity of unit j
DTmin Minimum temperature difference
G A large enough positive value
l st Steam latent heat
rcs ( sin, j ) Fraction of state s in the input consumed by the task
r ( sin, j )
p
s Fraction of state s in the output produced by the task
t ( sin, j ) Constant duration of the task
Variables
msin ( s, j, p ) Amount of state s sent to storage from unit j at time point p
msout ( s, j, p ) Amount of state s sent from storage to unit j at time point p
mt ( s, j, j¢, p ) Amount of state s sent from unit j to unit j′ at time point p
mu ( sin, j , p ) Amount of material used for the task at time point p
( )
qex s*h , s*c , p Heat exchanged between the hot and cold streams at time point p
qinc ( sin, j , p ) Amount of cooling for the input material at time point p
q c
out ( sin, j , p ) Amount of cooling for the output material at time point p
q ( sin, j , p )
h
in Amount of heating for the input material at time point p
h
qout ( sin, j , p ) Amount of heating for the output material at time point p
T f ( sout , j , p ) Final temperature of the outlet stream at time point p
(
T f sinr , j , p ) Final temperature of the raw material stream at time point p
Ti ( sout , j , p ) Initial temperature of the outlet stream at time point p
Tin ( sin, j , p ) Temperature of the input material at time point p
Tout ( sin, j , p ) Temperature of the output material at time point p
t p ( sin, j , p ) End time of the task at time point p
tu ( sin, j , p ) Start time of the task at time point p
tinr ( sin, j , p ) Time required for heating or cooling for the input material at time
point p
r
tout ( sin, j , p ) Time required for heating or cooling for the output material at time
point p
tmin ( s, p ) Time at which state s is sent to storage at time point p
tmout ( s, p ) Time at which state s is sent from storage at time point p
Tsout ( s, p ) Outlet temperature of state s from storage at time point p
win ( s, j, p ) Binary variable indicating if state s is sent to storage from unit j at
time point p
wout ( s, j, p ) Binary variable indicating if state s is sent from storage to unit j at
time point p
( )
x s*h, s*c , p Binary variable indicating if there is heat exchange between the hot
and cold streams at time point p
y ( sin, j , p ) Binary variable indicating if the task is active at time point p
yinc ( sin, j , p ) Binary variable indicating if there is cooling for the input material at
time point p
c
yout ( sin, j , p ) Binary variable indicating if there is cooling for the output material
at time point p
yinc ( sin, j , p ) Binary variable indicating if there is heating for the input material at
time point p
h
yout ( sin, j , p ) Binary variable indicating if there is heating for the output material
at time point p
z ( j, j¢, p ) Binary variable indicating if there is material sent from unit j to
unit j′ at time point p
REFERENCES
Fernández, I., Renedo, C.J., Pérez, S.F., Ortiz, A., Mañana, M., 2012. A review: Energy recov-
ery in batch processes. Renewable and Sustainable Energy Reviews. 16, 2260–2277.
Harjunkoski, I., Maravelias, C.T., Bongers, P., Castro, P.M., Engell, S., Grossmann, I.E.,
Hooker, J., Méndez, C., Sand, G., Wassick, J., 2014. Scope for industrial applications
of production scheduling models and solution methods. Computers and Chemical
Majozi, T., 2010. Batch Chemical Process Integration: Analysis, Synthesis and Optimization.
Dordrecht, the Netherlands: Springer.
Pinto, T., Barbósa-Póvoa, A.P.F.D., Novais, A.Q., 2008. Design of multipurpose batch plants:
a comparative analysis between the STN, m-STN, and RTN representations and formu-
lations. Industrial and Engineering Chemical Research. 47, 6025–6044.
Rosenthal, R.E., 2015. GAMS—A User’s Guide. Washington, DC: GAMS Development
Corporation.
Tawarmalani, M., Sahinidis, N.V., 2005. A polyhedral branch-and-cut approach to global
optimization. Mathematical Programmes. 103, 225–249.
5 Heat Integration
in Multipurpose
Batch Plants
5.1 INTRODUCTION
Energy saving is becoming increasingly important in batch processing facilities.
Multipurpose batch plants have become more popular than ever in the processing
environment due to their inherent flexibility and adaptability to market conditions,
even though the same flexibility may lead to complexities, such as the need to sched-
ule process tasks. These are important features to producing high value–added prod-
ucts such as agrochemicals, pharmaceuticals, polymers, food and specialty chemicals
where the demand has grown in recent decades. Many current heat integration meth-
ods for multipurpose batch plants use a sequential methodology where the schedule
is solved first, followed by heat integration. This can lead to suboptimal results. In
this chapter, the heat integration model is built upon a robust scheduling framework.
This scheduling formulation has proven to lead to better results in terms of better
objective values, fewer required time points and reduced computational time. This is
important as inclusion of heat integration into a scheduling model invariably compli-
cates the solution process. The improved scheduling model allows the consideration
of industrial-sized problems to simultaneously optimize both the process schedule
and energy usage. Both direct and indirect heat integration are considered, as well as
fixed and variable batch sizes.
Batch processing is commonly used, when products are required in small quantities
or when the processes are complex or specialized, to manufacture high value–added
products. Examples include food, pharmaceuticals, fine chemicals, biochemicals
and agrochemicals. Batch operations, even though they are becoming increasingly
popular, are generally run on a smaller scale compared to continuous operations,
and utility requirements are therefore considered less significant. However, utility
requirements for the food industry, breweries, dairies, meat processing facilities,
biochemical plants and agrochemical facilities contribute largely to their overall cost
(Stamp and Majozi, 2011). Energy savings have often been neglected in batch pro-
cesses in the past and hence significant savings are possible. The literature review is
organized into two major sections and includes methods developed in the twentieth
and twenty-first centuries.
109
5.2.1 Developments in the Twentieth Century

Early work that aimed at reducing energy consumption in batch plants was by
adopting methods that had been developed for continuous processing plants. One
of the very first contributions in this regard for energy integration in batch plants
was through applying the principle of time average model (TAM) (Clayton, 1986).
However, this approach does not consider the process schedule and assumes that the
hot and cold streams exist simultaneously as in continuous processes. The identified
energy targets could not be achieved when only direct heat integration was applied
since the time schedule was not considered. The TAM formulation was further
extended to incorporate heat storage to achieve the maximum energy saving targets
(Stolze et al., 1995). Vaselenak et al. (1986) explored the possibility of heat recovery
between a number of tanks which required heating and cooling through the consid-
eration of co-current, countercurrent and a combination of the two. The authors did
not account for the time schedule; they assumed that all tanks are available at the
same time. In order to address the limitations of the TAM, various works considered
the time schedule in the heat integration analysis. The work on the design of batch
processes based on pinch technology showed that the cost structure should consider
the interaction between capital, energy, scheduling and yields in order to address
the wider scope (Obeng and Ashton, 1988). The cascade analysis for maximum heat
recovery based on time–temperature cascade tables that consider both direct and
indirect heat exchange between streams was developed (Kemp and Deakin, 1989).
A strategy for creating ‘schedules of maximum power of heat integration’ which
ensures, on the one hand, production targets and, on the other hand, optimal con-
dition for heat integration was demonstrated (Ivanov and Bancheva, 1994), and a
methodology was presented to address the important aspect of rescheduling for
maximum energy recovery (Corominas, 1993). Different theoretical aspects of opti-
mal energy integration in batch chemical plants have been considered in the previ-
ous literature, with little experience on real case studies. This gap was addressed
by applying optimal energy integration formulation for an existing antibiotics plant
where implementation of the formulation gave an overall energy cost reduction of
39% (Boyadjiev et al., 1996). Most of the research work mentioned here is based on
fixed production schedules that already achieve all the plant objectives and minimize
the energy requirement afterward. However, in general, even optimal production
schedules tend to be quite degenerate, in the sense that there often exists a large num-
ber of different schedules, all of which can achieve a given set of production require-
ments. Nevertheless, the potential for heat integration could vary significantly from
one such schedule to another. Consequently, heat integration should be considered
as an integral part of the problem of scheduling the production in a given plant.
The cost of utilities should be incorporated within the overall economic objective
of maximizing the net value of the production over a given time horizon and solved
simultaneously (Papageorgiou et al., 1994). There were other studies that resolved
heat integration in batch processes by establishing an objective that was to mini-
mize the combined operating and annualized capital costs of the heat-exchanger net-
works (HENs) for a class of multipurpose batch plants. A mathematical formulation
that selects the production campaigns and designs HENs simultaneously has been
Heat Integration in Multipurpose Batch Plants 111
presented (Bancheva et al., 1996a). However, the formulation limits a heat exchanger
to be used for a specific pair of processing units. Design of HENs for a system of
batch vessels that exploits heat integration potential with minimum cost has also
been developed (Pozna et al., 1998).
5.2.2 Developments in the Twenty-First Century

The research conducted in this century has been concentrated on improving the
models and methodologies that had been proposed previously as well as the optimi-
zation of heat integration in batch processes through the development of new tools
such as genetic algorithms and network evolution techniques. Moreover, there has
been a significant increase in the utilization of thermal storage to improve heat inte-
gration in a discontinuous process.
5.2.2.1 Models Developed for Direct Heat Integration

The developed models in this category require the hot and cold task should be oper-
ated in the same time interval for the heat exchange to occur. Uhlenbruck et al.
(2000) improved OMNIUM, which is a tool developed for heat exchanger network
synthesis by Hellwig and Thone (1994). The improved OMNIUM tool increased
energy recovery by 20%. A single step, interactive computer program (BatcHEN)
used for the determination of campaigns (i.e. the set of products which can be pro-
duced simultaneously), the heat exchange areas of all possible heat exchangers in
the campaigns and the heat exchanger network were presented (Bozan et al., 2001).
This work addressed the limitation of the graph theory method (Bancheva et al.,
1996b) for the determination of the campaign where it is very complex for han-
dling large numbers of products and process units. A heat integration model based
on S-graph scheduling framework approaches was developed (Adonyi et al., 2003).
The results of this work showed how utility usage can be reduced considerably with
just a slight increase in production makespan. Morrison et al. (2007) developed a
user-friendly software package known as Optimal Batch Integration (OBI). Chen
and Chang (2009) integrated the task scheduling and heat recovery problems into a
unified framework for multipurpose batch processes. The batch scheduling formula-
tion is extended from the continuous Resource Task Network (RTN) formulation,
which was originally proposed by Castro et al. (2004). Halim and Srinivasan (2009)
discussed a sequential method using direct heat integration. A number of optimal
schedules with minimum makespan were found and heat integration analysis was
performed on each. The schedule with minimum utility requirement was chosen as
the best. Later, Halim and Srinivasan (2011) extended their technique to carry out
water reuse network synthesis simultaneously. One key feature of this method is its
ability to find the heat integration and water reuse solution without much sacrificing
the quality of the scheduling solution as compared to other sequential techniques for
heat integration for multipurpose batch plants.
There are other works that considered energy saving and the capital cost associ-
ated to heat exchanger to achieve the minimum utility target. The technique for
design and synthesis of batch plant (Barbosa-Póvoa et al., 2001) was later extended
to incorporate economic savings in utility requirements, while considering the cost

of both auxiliary structures (i.e. heat exchangers through their transfer area) and
the design of the utility circuits and associated piping costs (Pinto et al., 2003). Liu
et al. (2011) formulated a batch heat exchanger network that results in nonlinear
programming (NLP). The application of the formulation has demonstrated that it
can effectively reduce the annual capital cost and annual total cost of HEN with
the employment of common heat exchangers. Maiti et al. (2011) developed a novel
heat integrated batch distillation column in order to improve thermal efficiency and
reduce the total annual cost. The potential energy integration leads to achieving
approximately 56.10% energy savings and 40.53% savings in total annual cost.
5.2.2.2 Models Developed for Indirect Heat Integration

Models developed in this category used heat storage for a more heat recovery and
flexible schedule compared to direct heat integration methods. The use of heat stor-
age allows the exchange of heat from hot task to cold task to takes place in different
time interval. A systematic procedure based on pinch analysis, backed with a graphi-
cal representation, allows the determination of the minimum number of heat storage
units and their range of feasible operation as a function of the amount of heat recov-
ery was presented (Krummenacher and Favrat, 2001). Chen and Ciou (2008) for-
mulated a method to design and optimize indirect energy storage systems for batch
processes. Their work aimed at simultaneously solving the problem of indirect heat
exchange network synthesis and its associated thermal storage policy for recirculated
hot/cold heat storage medium (HEN). Most of the previous work solved this sequen-
tially. The BatchHeat software that aimed to highlight the energy inefficiencies in
the process thereby enabling the scope for possible heat recovery to be established
through direct heat exchange or storage through implementation of cascade analysis
was developed (Pires et al., 2005).
5.2.2.3 Models Applied for Real Industrial Case Studies

This section presents methodologies developed for energy recovery taking existing
batch processing plans. The application of cascade analysis proposed by Kemp and
Macdonald (1987) to reduce utility requirement was applied to a case study about an
industry that produces oleic acid from palm olein using immobilized lipase (Chew
et al., 2005). The result obtained showed savings of 71.4% and 62.5% for hot and cold
utilities, respectively. Application of process integration to investigate the potential
to decrease energy usage in the slaughtering and meat processing industry was stud-
ied. Above ambient temperatures, heating of water with different target temperatures
is a large heat demand in a plant, while at sub-ambient temperatures, the refrigera-
tion plant needs almost all of the shaftwork used at the site. Interaction between, on
one hand, energy demands above ambient temperature and, on the other hand, cool-
ing needs below ambient temperature can take place with freezing compressors or
heat pumps. The result obtained illustrates that 30% of the external heat demand and
more than 10% of the shaftwork used can be saved (Fritzson and Berntsson, 2006).
Majozi (2006) presented a heat integration model for multipurpose batch plants
based on the continuous-time scheduling frame work (Majozi and Zhu, 2001). The
formulation results in smaller problems compared to the discrete-time formulation,
which renders it applicable to large-scale problems. Application of the formulation to

an agro-chemical industrial case study showed an 18.5% improvement in profit. The
direct heat integration model (Majozi, 2006) was extended to incorporate heat stor-
age for more flexible schedules and utility savings in a later work by Majozi (2009).
However, storage size is a parameter in his formulation which is addressed later by
Stamp and Majozi (2011), where the storage size is determined by an optimization
exercise. Foo et al. (2008) extended the minimum units targeting and network evolu-
tion techniques that were developed for batch mass exchange network (MEN) into
batch HEN. They applied the technique for energy integration of oleic acid produc-
tion from palm olein using immobilized lipase. Atkins et al. (2010) applied indirect
heat integration using heat storage for a milk powder plant in New Zealand. The
traditional composite curves have been used to estimate the maximum heat recovery
and to determine the optimal temperatures of the stratified tank. Tokos et al. (2010)
applied a batch heat integration technique to a large beverage plant. The opportuni-
ties of heat integration between batch operations were analysed by a mixed integer
linear programming (MILP) model, which was slightly modified by considering
specific industrial circumstances. Muster-Slawitsch et al. (2011) came up with the
Green Brewery concept to demonstrate the potential for reducing thermal energy
consumption in breweries. Three detailed case studies have been performed. The
‘Green Brewery’ concept has shown a saving potential of more than 5000 t/y fossil
CO2 emissions from thermal energy supply for the three breweries that were closely
considered. Becker et al. (2012) applied time average energy integration approach
to a real case study of a cheese factory with non-simultaneous process operations.
Their work addressed appropriate heat pump integration. A cost saving of more than
40% was reported. Integration of solar thermal energy in a batch fish tinning process
was investigated. The work demonstrated the most favourable heat integration option
for the thermosolar and heat pump (José et al., 2013). Recently, a design model for
heat recovery using heat storage and integration of industrial solar for a case study of
dairy processing was developed. Application of the model gave 37% heat recovery
(Timothy et al., 2014). The reader can get a more comprehensive and detailed review
on energy recovery for batch processes in the chapter by Fernández et al. (2012).
Many heat integration techniques are applied to predefined schedules, which may
lead to suboptimal results. For a more optimal solution, scheduling and heat integra-
tion should be combined into an overall problem and solved simultaneously. This
work aims to improve the efficiency of energy integration techniques by developing
a single framework that contains scheduling and heat integration models for multi-
purpose batch plants to be solved simultaneously for an optimal solution. A recent
robust scheduling formulation by Seid and Majozi (2012) is used as a platform since
it has proven to require fewer time points and reduced computational time compared
to other models and may also lead to an improved objective value. Compared to other
models based on simultaneous approach for heat integration for multipurpose batch
plants, the developed model allows a task to be heat integrated with other tasks in
more than one time interval during its starting and finishing times for better heat
recovery. Additionally, this work generalizes the heat integration problem with the
considerations of temperature change during processing of tasks and heat integration
to occur in any interval between the starting and finishing time of a task, which is
a limitation of the recent work by Stamp and Majozi (2011). It also caters for the
availability of heat storage for indirect heat integration which is a limitation of the
more advanced sequential methodology for heat integration by Halim and Srinivasan
(2009). Most literature has addressed these problems independently.
The subsequent sections are organized as follows. The problem statement and
objectives are given in Section 5.3. The developed mathematical model is then dis-
cussed in Section 5.4. The model is then applied to three literature examples and the
results are compared to recent literature models in Section 5.5. Conclusions are then
drawn to highlight the value of the contribution in Section 5.6.
5.3 PROBLEM STATEMENT AND OBJECTIVES

The problem addressed in this work can be stated as follows:
Given:
1. Production scheduling data, including equipment capacities, durations of

tasks, time horizon of interest, product recipes, cost of starting materials
and selling price of final products
2. Hot duties for tasks requiring heating and cold duties for tasks that require
cooling
3. Costs of hot and cold utilities
4. Operating temperatures of tasks requiring heating and cooling
5. Minimum allowable temperature differences
6. Design capacity limits on heat storage
Determine:
1. An optimal production schedule where the objective is to maximize profit,

defined as the difference between revenue and the cost of hot and cold
utilities
2. The size of heat storage as well as the initial temperature of heat storage
5.4 MATHEMATICAL FORMULATION
5.4.1 Model Constraints
The scheduling model by Seid and Majozi (2012) is adopted since it has proven to
result in fewer binary variables, reduced CPU time and a better optimal objective
value compared to other scheduling models.
The mathematical model is based on the superstructure in Figure 5.1. Each task
may operate using either direct or indirect heat integration. Tasks may also operate
in standalone mode using only external utilities. This may be required for control
reasons or when thermal driving forces or time do not allow for heat integration. If
either direct or indirect heat integration is not sufficient to satisfy the required duty,
external utilities may be used to make up the deficit.
External jh u jc External
cooling heating
Heat
Hot unit Cold unit
storage
FIGURE 5.1 Superstructure for mathematical model.
Constraints (5.1) and (5.2) are active simultaneously and ensure that one hot unit
can only be integrated with one cold unit when direct heat integration takes place in
order to avoid operational complexity of the process. However, it is also possible for
one unit to integrate with more than one unit at a given time point simultaneously
when the summation notation in the equations are not used. Also, if two units are to
be heat integrated at a given time point, they must both be active at that time point.
åx ( s injc , sinjh , p, pp ) £ y ( sinjh , p ) , "p, pp Î P, sinjh Î SinJh , sinjc Î SinJc (5.1)

sinjc
åx ( s injc , sinjh , p, pp ) £ y ( sinjc , p ) , "p, pp Î P, sinjh Î SinJh , sinjc Î SinJc (5.2)

sinjh
For better understanding, the difference between time point p and extended time
point pp. is explained using Figure 5.2. If a unit j that is active at time point p is
integrated with more than one unit in different temperature and time intervals, an
extended time point pp. must be defined. Unit j1 active at time point p can be inte-
grated with units j2 and j3 in different time and temperature intervals. At the begin-
ning, unit j1 is integrated with unit j2 at time point p and the extended time point pp.
is the same as time point p. Later, j1 is integrated with unit j3 in another time interval
where extended time point pp. equals to p + 1. pp. is equal to or greater than time
point p and less than or equal to n + p, where n is a parameter which is greater than
or equal to zero. If n equals 2, then a unit that is active at time point p can be inte-
grated in three different time intervals. The model should be solved starting from n
equals zero and adding one at a time until no better objective value is achieved. This
concept is the novelty of this work which addresses the limitation of models based
on simultaneous approach for heat integration for multipurpose batch plants where a
J3
J2
J1
tuu(j1, p, pp = p) tpp(j1, p, pp = p) tpp(j1, p, pp = p + 1) Time (h)

tuu(j2, p, pp = p) tpp(j2, p, pp = p) tpp(j3, p, pp = p + 1)
x(j1, j2, p, pp = p) tuu(j1, p, pp = p + 1)
tuu(j3, p, pp = p + 1)
x(j1, j3, p, pp = p + 1)
FIGURE 5.2 Differentiating time point p and extended time point pp.
task is allowed to be heat integrated with other tasks only for one time interval dur-
ing the starting and finishing time of the task.
Constraints (5.3) and (5.4) ensure that a unit cannot undergo direct and indi-
rect heat integration simultaneously. This condition simplifies the operation of the
process.
åx ( s injc ( )
, sinjh , p, pp ) + z sinj c , u, p, pp £ 1,
sinjh
"p, pp Î P, sinjh Î SinJh , sinjc Î SinJc u ÎU (5.3)
åx ( s
sinjc
injc , sinjh , p, pp ) + z ( sinjh , u, p, pp ) £ 1,
"p, pp Î P, sinjh Î SinJh , sinjc Î SinJc u ÎU (5.4)
Constraint (5.5) describes the amount of cooling load required by the hot unit to
reach from its initial temperature to its target temperature. On an occasion where
the temperature in the reactor unit is to be fixed during exothermic reaction, the
heat load becomes the product of the amount of mass that undergoes reaction and
the heat of reaction.
(
cl ( sinjh , p ) = mu ( sinjh , p ) cp ( sinjh ) Tsininjh - Tsout
injh )
, "p Î P, sinjh Î SinJh

(5.5)
Constraint (5.6) describes the heating load required by the cold unit to reach from
its initial temperature to its target temperature. On an occasion where the tempera-
ture in the reactor unit is to be fixed during endothermic reaction, the heat load
becomes the product of the amount of mass that undergoes reaction and the heat of
reaction.
(
hl ( sinjc , p ) = mu ( sinjc , p ) cp ( sinjc ) Tsout
injc )
- Tsininjc , "p Î P, sinjc Î SinJc

(5.6)
Constraints (5.7) and (5.8) describe the average heat flow for the hot and cold units,
respectively, during the processing time, which is the same as TAM, to address the
energy balance during heat integration properly.
( )
cl ( sinjh , p ) = avcl ( sinjh , p ) tp ( sinjh , p ) - tu ( sinjh , p ) , "p Î P, sinjh Î SinJh

(5.7)
( )
hl ( sinjc , p ) = avhl ( sinjc , p ) tp ( sinjc , p ) - tu ( sinjc , p ) , "p Î P, sinjc Î SinJc

(5.8)
Constraints (5.9) and (5.10) define the heat load at time point p and extended time
point pp. for the cold and hot unit.
(
hlp ( sinjc , p, pp ) = avhl ( sinjc , p ) tpp ( sinjc , p, pp ) - tuu ( sinjc , p, pp ) ,)
"p, pp Î P, sinjc Î SinJc (5.9)
(
clp ( sinjh , p, pp ) = avcl ( sinjh , p ) tpp ( sinjh , p, pp ) - tuu ( sinjh , p, pp ) ,)
"p, pp Î P, sinjh Î SinJh (5.10)
Constraints (5.11) and (5.12) quantify the amount of heat received from
and transferred to the heat storage unit, respectively. There will be no heat
received or transferred if the binary variable signifying use of the heat storage vessel,
z ( sinj , u, p, pp ), is zero.
( )
Q ( sinjc , u, p, pp ) = W ( u ) cp ( u ) T0 ( u, p, pp ) - T f ( u, p, pp ) z ( sinjc , u, p, pp ) ,
"p Î P, sinjc Î SinJc , u Î U (5.11)

( )
Q ( sinjh , u, p, pp ) = W ( u ) cp ( u ) T f ( u, p, pp ) - T0 ( u, p, pp ) z ( sinjh , u, p, pp ) ,
"p Î P, sinjh Î SinJh , u Î U (5.12)

Constraints (5.13) and (5.14) are used to calculate the temperature of the hot and
cold units at the intervals
(
clp ( sinjh , p, pp ) = mu ( sinjh , p ) cp ( sinjh ) T in ( sinjh , p, pp ) - T out ( sinjh , p, pp ) , )

( )
hlp ( sinjc , p, pp ) = mu ( sinjc , p ) cp ( sinjc ) T out ( sinjc , p, pp ) - T in ( sinjc , p, pp ) ,

Constraint (5.15) states that the amount of heat exchanged between the cold unit
and the heat storage should be less than the heat load required by the cold unit dur-
ing the interval.
Q ( sinjc , u, p, pp ) £ hlp ( sinjc , p, pp ) , "p, pp Î P, sinjc Î SinJc , u Î U (5.15)
Constraint (5.16) states that the amount of heat exchanged between the hot
unit and the heat storage should be less than the cooling load required by the hot unit
during the interval.
Q ( sinjh , u, p, pp ) £ clp ( sinjh , p, pp ) , "p, pp Î P, sinjh Î SinJh , u Î U (5.16)
Constraint (5.17) states that the amount of heat exchanged between the hot and
the cold unit should be less than the cooling load required by the hot unit during the
interval.
åQe ( s injh , sinjc , p, pp ) £ clp ( sinjh , p, pp ) , "p, pp Î P, sinjh Î SinJh , sinjc Î SinJc
sinjc (5.17)
Constraint (5.18) states that the amount of heat exchanged between the cold unit
and the hot unit should be less than the heat load required by the cold unit during
the interval.
åQe ( s injh , sinjc , p, pp ) £ hlp ( sinjc , p, pp ) , "p, pp Î P, sinjh Î SinJh , sinjc Î SinJc
sinjh
(5.18)
Constraint (5.19) ensures that if heat integration occurs, the heat load should have
a value that is less than the maximum amount of heat exchangeable. When the binary
variable associated with heat integration takes a value of zero, no heat integration
occurs and the associated heat load is zero.
Qe ( sinjh , sinjc , p, pp ) £ Q x ( sinjc , sinjh , p, pp ) , "p, pp Î P, sinjh Î SinJh , sinjc Î SinJc

U
(5.19)
Constraints (5.20) and (5.21) ensure that if heat integration between the cold and
hot units takes place with the heat storage, then the heat load takes a positive value.
This happens only when the binary variable associated with the integration of cold
and hot units with the heat storage unit takes the value of one.
Q ( sinjh , u, p, pp ) £ Q z ( sinjh , u, p, pp ) , "p, pp Î P, sinjh Î SinJh , u Î U (5.20)

U
Q ( sinjc , u, p, pp ) £ Q z ( sinjc , u, p, pp ) , "p, pp Î P, sinjc Î SinJc , u Î U

U
(5.21)
Constraints (5.22) and (5.23) state that the temperature of the task at the current
time interval should be equal to the temperature at the end of the previous time
interval.
T in ( sinjh , p, pp ) = T out ( sinjh , p, pp - 1) , "p, pp Î P, sinjh Î SinJh (5.22)
T in ( sinjc , p, pp ) = T out ( sinjc , p, pp - 1) , "p, pp Î P, sinjc Î SinJc (5.23)
Constraint (5.24) states that the temperature of the heat storage unit at the cur-
rent time interval should be equal to the temperature at the end of the previous time
interval.
To ( u, p, pp ) = T f ( u, p, pp - 1) , "p, pp Î P, u Î U (5.24)
Constraint (5.25) states that the initial temperature of the heat storage unit at the
current time point p should be equal to the final temperature at the previous time
point p – 1.
To ( u, p, pp = p ) = T f ( u, p - 1, pp = p - 1 + n ) , "p, pp Î P (5.25)
Constraints (5.26) and (5.27) state that the temperature at the start of the first time
interval, which is time point p, and also pp, should be equal to the initial temperature
of the task.
T in ( sinjh , p, pp ) = Tsininjh , "p, pp Î P, sinjh Î SinJh (5.26)
T in ( sinjc , p, pp ) = Tsininjc , "p, pp Î P, sinjc Î SinJc (5.27)
Constraints (5.28) and (5.29) ensure that the minimum thermal driving forces are
obeyed when there is direct heat integration between a hot and a cold unit.
(
T in ( sinjh , p, pp ) - T out ( sinjc , p, pp ) ³ DT - DT U 1 - x ( sinjc , sinjh , p, pp ) , )
"p, pp Î P, sinjh Î SinJh , sinjc Î SinJc (5.28)

( )
T out ( sinjh , p, pp ) - T in ( sinjc , p, pp ) ³ DT - DT U 1 - x ( sinjc , sinjh , p, pp ) ,
obeyed when there is direct heat integration between a hot task and a heat storage unit.
(
T in ( sinjh , p, pp ) - T f ( u, p, pp ) ³ DT - DT U 1 - z ( sinjh , u, p, pp ) , )
"p, pp Î P, sinjh Î SinJh , u Î U (5.30)
(
T out ( sinjh , p, pp ) - To ( u, p, pp ) ³ DT - DT U 1 - z ( sinjh , u, p, pp ) , )
obeyed when there is direct heat integration between a cold task and a heat storage unit.
(
To ( u, p, pp ) - T out ( sinjc , p, pp ) ³ DT - DT U 1 - z ( sinjc , u, p, pp ) , )
"p, pp Î P, sinjc Î SinJc , u Î U (5.32)
(
T f ( u, p, pp ) - T in ( sinjc , p, pp ) ³ DT - DT U 1 - z ( sinjc , u, p, pp ) , )
Constraints (5.34) and (5.35) state that temperatures change in the heating and
cooling unit when the binary variables associated with heating and cooling are active.
T ( sinjh , p, pp ) - T ( sinjh , p, pp ) £ DT v ( sinjh , p, pp ) , "p, pp Î P, sinjh Î SinJh

in out U
(5.34)
T ( sinjc , p, pp ) - T ( sinjc , p, pp ) £ DT v ( sinjc , p, pp ) , "p, pp Î P, sinjc Î SinJc

out in U
(5.35)
Constraints (5.36) and (5.37) state that temperatures change in the heat storage
unit when the binary variables associated with heating and cooling with the heat
storage unit are active.
æ ö
ç å
T out ( u, p, pp ) - T in ( u, p, pp ) £ DT U ç z ( sinjc , u, p, pp ) + å z ( sinjh , u, p, pp ) ÷ ,
÷
è sinjc sinjh ø
"p, pp Î P, sinjh Î SinJh , sinjc Î SinJc , u Î U (5.36)
æ ö
ç å
T out ( u, p, pp ) - T in ( u, p, pp ) ³ -DT U ç z ( sinjc , u, p, pp ) + å z ( sinjh , u, p, pp ) ÷ ,
÷
è sinjc sinjh ø
"p, pp Î P, sinjh Î SinJh , sinjc Î SinJc , u Î U (5.37)
Constraints (5.38) through (5.41) ensure that the times at which units are active are
synchronized when direct heat integration takes place.
( )
tuu ( sinjh , p, pp ) ³ tuu ( sinjc , p, pp ) - M 1 - x ( sinjc , sinjh , p, pp ) ,

( )
tuu ( sinjh , p, pp ) £ tuu ( sinjc , p, pp ) + M 1 - x ( sinjc , sinjh , p, pp ) ,
( )
tpp ( sinjh , p, pp ) ³ tpp ( sinjc , p, pp ) - M 1 - x ( sinjc , sinjh , p, pp ) ,
( )
tpp ( sinjh , p, pp ) £ tpp ( sinjc , p, pp ) + M 1 - x ( sinjc , sinjh , p, pp ) ,
Constraints (5.42) through (5.49) ensure that the times at which the cold and hot
units are synchronized with the time of the heat storage unit are when heat integra-
tion takes place.
(
tuu ( sinjh , p, pp ) ³ tuu ( u, p, pp ) - M 2 - v ( sinjh , p, pp ) - z ( sinjh , u, p, pp ) , )

(
tuu ( sinjh , p, pp ) £ tuu ( u, p, pp ) + M 2 - v ( sinjh , p, pp ) - z ( sinjh , u, p, pp ) , )

(
tpp ( sinjh , p, pp ) ³ tpp ( u, p, pp ) - M 2 - v ( sinjh , p, pp ) - z ( sinjh , u, p, pp ) , )

(
tpp ( sinjh , p, pp ) £ tpp ( u, p, pp ) + M 2 - v ( sinjh , p, pp ) - z ( sinjh , u, p, pp ) , )

(
tuu ( sinjc , p, pp ) ³ tuu ( u, p, pp ) - M 2 - v ( sinjc , p, pp ) - z ( sinjc , u, p, pp ) , )

(
tuu ( sinjc , p, pp ) £ tuu ( u, p, pp ) + M 2 - v ( sinjc , p, pp ) - z ( sinjc , u, p, pp ) , )

(
tpp ( sinjc , p, pp ) ³ tpp ( u, p, pp ) - M 2 - v ( sinjc , p, pp ) - z ( sinjc , u, p, pp ) , )

(
tpp ( sinjc , p, pp ) £ tpp ( u, p, pp ) + M 2 - v ( sinjc , p, pp ) - z ( sinjc , u, p, pp ) ,)

Constraints (5.50) and (5.51) state that the starting time of the heating load
required for the cold unit and the cooling load required for the hot unit at the first
time interval should be equal to the starting time of the hot and cold unit.
tuu ( sinjh , p, pp = p ) = tu ( sinjh , p ) , "p, pp Î P, sinjh Î SinJh (5.50)
tuu ( sinjc , p, pp = p ) = tu ( sinjc , p ) , "p, pp Î P, sinjc Î SinJc (5.51)
Constraints (5.52) and (5.53) state that the starting time of heating and cooling in
the current time interval should be equal to the finishing time at the previous time
interval.
tuu ( sinjh , p, pp ) = tpp ( sinjh , p, pp - 1) , "p, pp Î P, sinjh Î SinJh (5.52)
tuu ( sinjc , p, pp ) = tpp ( sinjc , p, pp - 1) , "p, pp Î P, sinjc Î SinJc (5.53)
Constraint (5.54) states that the starting time of the heat storage unit at the cur-
rent time interval should be equal to the finishing time at the previous time interval.
tuu ( u, p, pp ) = tpp ( u, p, pp - 1) , "p, pp Î P, u ÎU (5.54)
Constraint (5.55) states that the finishing time of a heat storage unit in a time
interval should be equal to or greater than the starting time of the same time interval.
tpp ( u, p, pp ) ³ tuu ( u, p, pp ) , "p, pp Î P, u Î U (5.55)
Constraint (5.56) states that the starting time of the heat storage unit at time
point p should be greater than or equal to the finishing time at the previous time
point p − 1.
tuu ( u, p, pp = p ) = tpp ( u, p - 1, pp = p - 1 + n ) , "p, pp Î P, u Î U (5.56)
Constraints (5.57) and (5.58) state that if the binary variable associated with heat
integration is active, then the binary variable associated with heating and cooling
must be active.
x ( sinjc , sinjh , p, pp ) £ v ( sinjh , p, pp ) , "p, pp Î P, sinjh Î SinJh , sinjc Î SinJc (5.57)
x ( sinjc , sinjh , p, pp ) £ v ( sinjc , p, pp ) , "p, pp Î P, sinjh Î SinJh , sinjc Î SinJc (5.58)

Constraints (5.59) and (5.60) state that the heating and cooling loads take on a
value for a certain duration when the binary variables associated with heating and
cooling are active.
( )
tpp ( sinjh , p, pp ) - tuu ( sinjh , p, pp ) £ H * v s injh , p, pp , "p, pp Î P, sinjh Î SinJh

(5.59)
( )
tpp ( sinjc , p, pp ) - tuu ( sinjc , p, pp ) £ H * v s injc , p, pp , "p, pp Î P, sinjc Î SinJc

(5.60)
Constraint (5.61) states that the cooling of a hot unit will be satisfied by direct heat
integration, indirect heat integration or external cooling utility, if required.
pp = p + n pp = p + n
cl ( sinjh , p ) = cw ( sinjh , p ) + åå
sinjc pp = p
Qe ( sinjh , sinjc , p, pp ) + å Q (s
pp = p
inj h )
, u, p, pp ,

Constraint (5.62) states that the heating of a cold unit will be satisfied by direct
heat integration, indirect heat integration or external heating utility, if required.
pp = p + n pp = p + n
hl ( sinjc , p ) = st ( sinjc , p ) + åå
sinjh pp = p
Qe ( sinjh , sinjc , p, pp ) + å Q (s
pp = p
injc , u, p, pp ) ,

5.4.2 Model Objective Functions

Equation 5.63 is the objective function in terms of profit maximization, with profit
defined as the difference between revenue from product and cost of utility.
æ ö
max ç
ç p å ( ) ( ) åå
price s p qs s p - costcw * cw ( sinjh , p ) - åå coostst * st ( sinjc , p ) ÷ ,
÷
è s p sinjh p sinjc ø
"p, pp Î P, sinjh Î SinJh , sinjc Î SinJc , sinj Î SinJ (5.63)
Equation 5.64 defines the minimization of utility if the product demand is known.
æ ö
min ç
ç åå costcw * cw ( sinjh , p ) + åå
costst * st ( sinjc , p ) ÷ ,
÷
è p sinjh p sinjc ø
5.5 CASE STUDIES
In order to demonstrate the application of the proposed model, three literature case
studies are presented. The results in all the case studies for the proposed model were
obtained using a Pentium 4 with 3.2 GHz processor and 512 MB RAM. CPLEX
and CONOPT 2 in GAMS 22.0 were used to solve the MILP and NLP problems,
respectively. DICOPT 2 was used as the interface for solving the MINLP problem.
The computational results of this work are compared with results from the literature.
5.5.1 Case Study I
This case study, obtained from Kondili et al. (1993), has become one of the most
commonly used examples in literature. However, this case study has been adapted
by Halim and Srinivasan (2009) to include energy integration. The batch plant pro-
duces two different products sharing the same processing units, where Figure 5.3
shows the plant flowsheet. The unit operations consist of preheating, three differ-
ent reactions and separation. The plant accommodates many common features of
multipurpose batch plants, such as units performing multiple tasks, multiple units
Feed C
Product 1
Feed B
Reactor 1
Heater
Feed A
Product 2
Still
Reactor 2
FIGURE 5.3 Flowsheet for Case Study I.

Product 1
s7
HotA 40% IntAB

40% 60%
10%
Feed A 60%
s9 Separation
IntBC s6
Impure E 90%
80%
50%
Feed B 50% 20% Product 2

(s3) (s4)
(a) Feed C
Feed A
s1
Feed B
s2 s5
HotA 40% Product1
40%
50% s7
60%
60% 10%
s6
(s3) 50% IntBC 80%
s8 s9
Feed C IntAB 20% Impure E
(s4) 90%
s10
(b) Product 2
FIGURE 5.4 STN (a) and (b) SSN representations for Case Study I.
suitable for a task and dedicated units for specific tasks. The STN and state sequence
network (SSN) representations of the flowsheet are shown in Figure 5.4. The produc-
tion recipe is as follows:
1. Raw material Feed A is heated from 50°C to 70°C to form HotA used in
reaction (5.2).
2. Reactant materials 50% Feed B and 50% Feed C are used in reaction (5.1)
to produce IntBC. During the reaction, the material has to be cooled from
100°C to 70°C.
3. 60% of the intermediate material IntBC and 40% of HotA are used in reac-
tion (5.2) to produce product 1 and IntAB. The process needs to be heated
from 70°C to 100°C during its operation.
4. 20% Feed C and 80% of the intermediate IntAB from reaction (5.2) are
used in reaction (5.3) to produce Impure E. The reaction needs its tempera-
ture to be raised from 100°C to 130°C during its operation.
5. The separation process produces 90% product 2 and 10% IntAB from
Impure E. Cooling water is used to lower its temperature from 130°C to
100°C.
The processing time of a task i in unit j is assumed to be linearly dependent, ai + bi B,

on its batch size B, where ai is a constant term of the processing time of task i and bi
is a coefficient of variable processing time of task i. The batch-dependent processing
time makes this case study more complex. Table 5.1 gives the relevant data on coef-
ficients of processing times, the capacity of the processing units, initial inventory of
raw materials, storage capacity and relevant costs. The production demand is given
as 200 kg for both Prod1 and Prod2. Table 5.2 gives data pertaining to heat integra-
tion. The objective here is optimization with respect to makespan and energy.

The computational results obtained using the proposed model as well as those of
Halim and Srinivasan (2009) are presented in Table 5.3. The model of Stamp and
Majozi (2011) is not included in the comparison since the model does not cater for
temperature variation during processing of tasks. The minimum temperature driving
force for heat exchange is assumed 10°C. For makespan minimization, an objective
value of 19.5 h was obtained using the proposed model, which is much better than
19.96 h obtained by Halim and Srinivasan (2009) as a result of using the recent
robust scheduling model. Using the makespan obtained, the case study was solved
to minimize the energy demand by setting customer requirement for Product 1 and
Product 2. The total energy required for the standalone operation was 125.5 MJ.
By using energy integration, the total energy requirement was reduced from
125.5 MJ in standalone operation to 51.4 MJ (38.7 MJ hot utility and 12.8 MJ cold
utility) resulting in a 59% energy saving. The performance of the proposed model
was also compared to the technique by Halim and Srinivasan (2009). They solved
this case study based on a three-step sequential framework. First, the schedule is
optimized to meet the economic objective of makespan minimization. Next, alter-
nate schedules are generated through a stochastic search-based integer cut proce-
dure. Finally, the heat integration model is solved for each of the resulting schedules
to establish the minimum utility targets. In generating alternate optimal solutions,
they used integer cut variables to 6, that is, in each iteration, six tasks were preas-
signed to different units at different slots. After 1000 MILP iteration of stochastic
search, three sets of solutions at three different makespans (20.03, 20.02, 19.96) were
obtained. Based on this set of schedules, the heat integration model was solved. The
makespan that gave the minimum utility target was 19.96, with the corresponding
utility requirement of 40.8 MJ hot utility and 15.6 MJ cold utility. Figure 5.5 shows
the optimization results after 1000 iteration of the stochastic search.
The developed model gives a better makespan and a further utility saving of 8.8%
compared to the results of Halim and Srinivasan (2009). An additional feature of the
proposed model is its efficient solution technique. The optimal solution was found
Heat Integration in Multipurpose Batch Plants
TABLE 5.1
Scheduling Data for Case Study I
Max Batch Revenue or Cost
Task (i) Unit (j ) Size (kg) a ( sinj ) b ( sinj ) Material State (s) Initial Inventory Max Storage (kg) ($/kg or $/MJ)
Heating (H) HR 100 0.667 0.007 Feed A 1000 1000 0
Reaction-1 (R1) RR1 50 1.334 0.027 Feed B 1000 1000 0
RR2 80 1.334 0.017 Feed C 1000 1000 0
Reaction-2 (R2) RR1 50 1.334 0.027 HotA 0 100 0
RR2 80 1.334 0.017 IntAB 0 200 0
Reaction-3 (R3) RR1 50 0.667 0.013 IntBC 0 150 0
RR2 80 0.667 0.008 ImpureE 0 200 0
Separation (S) SR 200 1.334 0.007 Prod1 0 1000 20
Prod2 0 1000 20
Cooling water 0.02
Steam 1
127
TABLE 5.2
Data Required for Energy Integration for Case Study I
Task (i) Tsininj ( °C ) inj (

Tsout °C ) Unit ( j) C p ( kJ / kg°C )
Heating (H) 50 70 HR 2.5
Reaction-1 100 70 RR1 3.5
RR2 3.5
Reaction-2 70 100 RR1 3.2
RR2 3.2
Reaction-3 100 130 RR1 2.6
RR2 2.6
Separation 130 100 SR 2.8
Cooling water 20 30
Steam 170 160
TABLE 5.3
Proposed Halim and
Formulation Proposed Srinivasan
without Formulation (2009) with
Energy with Energy Energy
Integration Integration Integration
Objective ($) 125.5 51.4 56.4
Steam (MJ) 75.2 38.7 40.8
Cooling water (MJ) 50.2 13.5 15.6
Revenue from product ($) 4000 4000 4000
Number of time points/ 8 8 N/A
slots
Number of binary 92 500 N/A
variables
CPU time (h) 0.39 2 N/A
after solving only one MINLP model, which was solved in a reasonable specified
CPU time of 2 h. The authors Halim and Srinivasan (2009) did not report the CPU
time required to solve this problem. We believe that it is computationally expensive
because of the need to generate alternate optimal schedules.
Figure 5.6 details the amount of heat exchanged between the cold and hot units
and the time intervals during which energy integration occurs.
The energy requirements of units RR1, RR2 and HR during the interval 4.976 to
7.660 h are explained to highlight the advantage of the proposed model. The heat-
ing load of unit RR2 between 4.976 and 6.660 h is 4.658 MJ. This is partly satisfied
through energy integration with unit RR1 in the same time interval resulting in an
75 Makespan = 19.96 h
20.02 h
20.03 h
70
Utility demand (MJ)

65
60
55
Best optima (19.96 h, 56.4 MJ)
50
19.9 19.95 20 20.05
Makespan (h)
FIGURE 5.5 Optimization results using stochastic search-based integer cut procedure by
Halim and Srinivasan (2009).
0.0 h 4.417 MJ 2.470 h 1.375 MJ 2.205 MJ 4.976 h 1.364 MJ 2.575 MJ

R1, RR2 cw R2, RR2 st st R2, RR2 st st
2.470 h 7.591 h
4.976 h
H, HR
3.039 MJ 3.294 MJ
3.330 MJ
4.976 h
st R1, RR1 cw R1, RR1
1.173 h 4.141 h 1.520 MJ 4.976 h 7.660 h
0.281 MJ
st = 1.299 MJ
0.0 h 8.965 h 1.6 MJ 1.956 MJ
R1, RR1 cw R3, RR1 st 6.660 h 7.660 h
3.685 MJ 2.470 10.282 h
H, HR st
0.423 MJ
1 MJ 10.439 h
S, SR S, SR
4.964 MJ
7.660 h 8.965 h 0.679 MJ
R3, RR1 st 9.843 h 0.828 MJ
3.829 MJ st
R3, RR2 R2, RR2 st
7.660 h 7.275 MJ 10.282 h 12.966 h
0.539 MJ 1.832 MJ 12.966 h

15.509 h 6.240 MJ cw
S, SR cw R3, RR2 st R1, RR1
16.816 h 10.282 h 1.956 MJ 1.462 MJ
1.141 MJ
12.219 h
2.868 MJ H, HR st
R2, RR2 st 15.588 h 16.816 h 11.219 h
R3, RR1 st 0.423 MJ
15.509 h
3.364 MJ st = 1.386 MJ
cw = 0.283 MJ R2, RR2
2.816 MJ 6.237 MJ 19.5 h
R1, RR1
12.966 h 15.588 h cw = 0.178 MJ

1.910 MJ S, SR
H, HR 19.5 h
14.44 h 15.44 h 3.927 MJ
st 19.5 h
R2, RR1 st
0.468 MJ 16.816 h 0.873 MJ
19.012 h
FIGURE 5.6 Heat exchange network for Case Study I using the proposed model.
Processing units
20 59.095 20 123.127
SR
8.965 11.219 12.966 14.44 16.816 19.5
7.66
RR2 R1 (66.811) R2 (68.962) R2 (75.340) R2 (75.779) R2 (79.412) R2 (71.095) R3 (80) R2 (79.412)
2.47 4.976 7.591 10.282 12.966 15.509 16.816 19.5
R3 R3
R1 (50)
RR1 R1 (42.066) R1 (43.421) (49.095) R3 (50) R1 (50) R1 (47.702) (43.127) R2 (50)
2.47 4.976 7.66 8.965 10.282 12.966 15.588 16.816 19.5
HR 72.226 47.57 47.57 47.57
1.173 6.66 7.66 11.219 12.219 14.44 15.44
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Time (h)
FIGURE 5.7 Gantt chart for Case Study I using the proposed model.
external heating requirement of 1.364 MJ rather than 4.658 MJ if it operated in stand-

alone mode. The heating requirement for RR2 between 6.660 and 7.591 h is fully
satisfied with external steam (2.575 MJ) since it is not in the heat integrated mode.
The cooling requirement of RR1 between 4.976 and 6.660 h is 3.294 MJ. This heat-
ing requirement is satisfied by heat integration with unit RR2. For the time interval
between 6.660 and 7.660 h, the cooling requirement for RR1 is 1.956 MJ that can
be fully satisfied with matching to unit HR. This demonstrates the advantage of the
developed model where it allows unit RR1 to be heat integrated with multiple units
at different time intervals for better heat recovery. Literature models based on simul-
taneous approach allow a unit to be heat integrated with other units only for one time
interval during its operation. Figure 5.7 shows the Gantt chart related to the optimal
usage of resources. It also indicates the types of tasks performed in each piece of
equipment and the starting and finishing times of the processing tasks.
5.5.2 Case Study II
This example, which was first examined by Sundaramoorthy and Karimi (2005), is
studied extensively in literature. It is a relatively complex problem and is often used
in literature to check the efficiency of models in terms of optimal objective value and
CPU time required. The plant has many common features of a multipurpose batch
plant, with the following features: units performing multiple tasks, multiple units
suitable for a task, states shared by multiple tasks and different products produced
following different production paths. The STN and SSN representations for this case
study are shown in Figure 5.8. The scheduling data are modified to incorporate heat
integration opportunities and are presented in Tables 5.4 and 5.5. Data necessary for
heat integration are presented in Table 5.6.
Feed3
S8

H1 {i1} H2 {i5} R3 { i7} 0.4
S1 S9 S13
{HR} {HR} {RR1,RR2}
Int 6
0.25 0.6
S3 Int1 S6 Int 4 Int 7 S10
0.5 0.4 0.4
Int 5 Product 1
R2 {i3} S {i4} 0.5 M {i6}
S5 S7 S12
{RR1,RR2} {SR} 0.4 {MR1,MR2}
Int 3
0.5 0.1 0.2
S4
Int 2 S11
Feed 2 Feed 4
R1 {i2}
S2
{RR1,RR2}
(a)
Feed 3
S8
Feed 1 Product 2
0.75 0.4
S1 S9 S13
0.25 0.6
Int 6
S3 Int 1 Int 4 S6 Int 7 S10
0.4 Int 5 0.4 Product 1

0.5 0.5
S5 S7 S12
0.4
0.5 Int 3 0.1 0.2
S4
Int 2
Feed 2 S11
S2 Feed 4
(b)
FIGURE 5.8 (a) STN representation and (b) SSN representation for Case Study II.

The computational results obtained from the different models for this case study
are presented in Table 5.7. This case study evaluates the efficiency of the proposed
model for indirect heat integration. The recent, more advanced sequential method-
ology for energy integration by Halim and Srinivasan (2009) is not included in the
comparison since their model does not consider the use of heat storage for indirect
heat integration.
For indirect heat integration, using a time horizon of 10 h, the proposed model
and the model by Stamp and Majozi (2011) give the same optimal objective value
of 224,000, while the proposed model gives a better CPU time. It is interesting to
TABLE 5.4
Scheduling Data for Case Study II
Unit Capacity Suitability Mean Processing Time (h)
Heater 100 H1, H2 1, 1.5
Reactor 1 100 RX1, RX2, RX3 2, 1, 2
Reactor 2 150 RX1, RX2, RX3 2, 1, 2
Separator 300 Separation 3
Mixer 1 200 Mixing 2
Mixer 2 200 Mixing 2
TABLE 5.5
Storage Initial Amount Revenue
State Description Capacity (Ton) (Ton) (c.u/Ton)
s1 Feed 1 Unlimited Unlimited 0
s3 Int 1 100 0 0
s4 Int 2 100 0 0
s5 Int 3 300 0 0
s6 Int 4 150 50 0
s7 Int 5 150 50 0
s9 Int 6 150 0 0
s10 Int 7 150 0 0
s12 Product 1 Unlimited 0 1000
s13 Product 2 Unlimited 0 1000
Parameters Values
Specific heat capacity for heat storage 4.2

cp(kJ/kg °C)
Steam cost (c.u/kWh) 10
Cooling water cost (c.u/kWh) 2
ΔTmin (°C) 10
T L (°C) 20
T (°C)
U 180
W L (ton) 1
W U (ton) 3
TABLE 5.6
Heating/Cooling Requirements for Case Study II
Heating/Cooling Operating
Reaction Type Requirement (kWh) Temperature (°C)
Fixed batch size
RX1 Exothermic 60 (cooling) 100
RX2 Endothermic 80 (heating) 60
TABLE 5.7
Computational Results for Case Study II, Fixed Batch Size with
Units Operating at 80% Capacity
Standalone Indirect Heat
Operation, Standalone Integration, Indirect Heat
Stamp and Operation, Stamp and Integration,
Majozi (2011) Proposed Majozi (2011) Proposed
H = 10
Performance index (cost units) 222,000 222,000 224,000 224,000
External cold duty (kWh) 200 200 0 0
External hot duty (kWh) 160 160 0 0
Heat storage capacity (ton) 1.905 1.905
Initial heat storage 82.5 82.5
temperature (°C)
CPU time (s) 5.3 1 68 7.8
Binary variables 66 101 156 209
Time points 7 6 7 6
H = 12
Performance index (cost units) 285,860 285,860 287,640 287,965
External cold duty (kWh) 270 270 130 17.5
External hot duty (kWh) 160 160 10 0
Heat storage capacity (ton) 5 1.905
Initial heat storage 87.143 82.5
temperature (°C)
CPU time (s) 7.7 1.9 238,896 35.2
Binary variables 99 155 206 246
Time points 9 7 11 7
note that by having heat storage, the utility requirement is reduced to 0.0 signi-
fying that operating a heat integrated batch plant with heat storage provides great
chance for heat recovery. For a time horizon of 12 h, the model by Stamp and Majozi
(Obeng and Ashton, 1988) required more than two days to solve, while the proposed
model significantly reduced the required CPU time (35.2 s for the proposed model
vs. 238,640 s for Stamp and Majozi’s (Obeng and Ashton, 1988)). The reduction in
CPU time results from reduced number of time points required (7 for the proposed
model vs. 11 for Stamp and Majozi’s (2011)). However, the proposed model required
a higher number of binary variables. This is due to catering for proper sequenc-
ing of tasks for the FIS policy which was inadvertently violated by the scheduling
model used by Stamp and Majozi (2011) and is explained well in the next case study.
A better optimal objective value of 287,965 was obtained by the proposed model
as compared to 287,640 by Stamp and Majozi (2011). This indicates that the use of
efficient scheduling techniques as a platform for a heat integration model improves
the computational efficiency, both in terms of CPU time and optimal objective value,
for heat integration in multipurpose batch plants.
The amount of material processed, the starting and finishing times, the amount
of heat exchanged between units and heat storage and the utility requirements of the
units for a time horizon of 12 h for indirect heat integration is given in Figure 5.9.
The numbers in the boxes represent the amount of batch processed in the unit.
Processing units
Heat HR HR HR HR HR HR
storage
MR-1
70 kWh (160)
60 kWh 80 kWh 80 kWh
52.5 kWh
70 kWh
(240)
SR
RX2 RX2
RR-2 RX1 (120) RX3 (120) RX3 (120)
(120) (120)
RR-1 RX3 (80)
HR H1 H1 H2 (80) H2 (80) H2 (80) H2 (80)

(80) (80)
1 2 3 4 5 6 7 8 9 10 11 12
Time (h)
FIGURE 5.9 Gantt chart for indirect heat integration, proposed model.
5.5.3 Case Study III

This case study was taken from the petrochemical plant by Kallrath (2002) and
used as a benchmark problem in the scheduling environment for multipurpose batch
plants. We adapted this case study to incorporate energy integration. The recipe
representation for the plant is presented in Figure 5.10. It is a very complex problem
where 10 reactors, 15 storage units and 18 states are considered. There are reactors
dedicated to a specific task, while there are also reactors conducting multiple tasks,
some capable of executing up to 5 different tasks. The mixed intermediate storage
(MIS) policy is assumed in this case study. Scheduling data is given in Table 5.8.
The plant is required to satisfy a demand requirement of 50 ton for S14, S17 and S18;
100 for S15 and 400 for S16 within a time horizon of 120 h. The heating and cooling
requirements for each task are given in Table 5.9.

The computational statistics obtained for this case study are presented in Table 5.10.
For the standalone operation using the technique of Stamp and Majozi (2011) based
on the scheduling techniques of Majozi and Zhu (2001), the model gives an objective
value of 6.74 × 106 c.u. The model required 17 time points and 425 binary variables
and was solved in a specified CPU time of 5000 s. By implementing the proposed
model on the same computer, an objective value of 6.5899 × 106 c.u. was obtained.
This model required 550 binary variables and was solved in a specified CPU time
of 5000 s.
It would be expected that the proposed model should give the same or a better
objective value since the model uses a more robust scheduling technique as a plat-
form. However, the slightly inferior result obtained is due to the need to catering
for proper sequencing for the FIS policy. This is better described by the Gantt chart
depicted in Figure 5.11. The maximum storage capacity for state S1 is 100 ton, but
this limit is violated a number of times during the time horizon using the model of
Stamp and Majozi (2011). The numbers inside the boxes in Figure 5.11 represent
the amount of batch processed. It is clear from this figure that the state S1 pro-
duced from reactor R1 is consumed in reactor R2 in consecutive time points. For
example, the state S1 consumed in reactor 1 at time points P1, P2 and P3 is con-
sumed in reactor 2 at time points P2, P3 and P4, respectively. These indicate that
the amount of material produced is consumed immediately in the next time point
so that the storage constraint is not violated. However, in real time, the storage
capacity is violated because of inadequate sequencing constraints that inproperly
match the consuming task, the producing task and storage. The same explanation
holds for state S5 where there is no storage available to store state S5. However,
from Figure 5.11, it can be seen that this constraint is violated in real time and
50 ton storage is actually required. This inadvertent violation of FIS storage policy
is resolved using the proposed model, and the right Gantt chart for this case study
is presented in Figure 5.12.
Using the proposed model for direct heat integration, an objective value of
7.9107 × 106 c.u. was obtained, which is much better than the standalone operation,
S0
T1 (R1)
4h
S1
0.6 T2 (R2) 0.4

8h
S2
S3
(S2b) (S2a)
0.31
T8 (R10) T8 (R5) T4 (R4) T5 (R4) T3 (R3)
10 h 12 h 8h 8h 4h
0.69
S9 S5 S6 S4
(S4a) (S4b)
T13 (R8) T13 (R9) T10 (R6) T10 (R7) T9 (R5) T6 (R4) T7 (R4)
8h 12 h 8h 10 h 12 h 8h 8h
S11 S10 S7 S8
S14 0.5
Product 1 T15 (R8) T14 (R8) T14 (R9) T11 (R6) T11 (R7) T12 (R6) T12 (R7)
8h 0.5 8h 12 h 10 h 12 h 12 h 12 h
S16 S15 S12 S13
Product 2 Product 3
T16 (R8) T16 (R9) T17 (R8) T17 (R9)

12 h 12 h 12 h 12 h
S17 S18
Product 4 Product 5
FIGURE 5.10 Recipe representation for Case Study III.
corresponding to a 20% improvement in profit. The utility reduction is also com-

pared for the energy integrated mode 528.7 MJ of utility is required which is much
better than the utility requirement of 3053 MJ for standalone operation. This
corresponds to an 82.7% utility reduction when the tasks are operated in energy
integrated mode. The model required 16 time points, n = 2 and 1315 binary vari-
ables and was solved in less than 2 h using specified CPU time of 2000 s for the
MILP subproblem. The energy integration network for the time horizon of 120 h is
depicted in Figure 5.13.
The energy integration in the interval between 12 and 20 h is elaborated for bet-
ter understanding. The cooling water requirement for Task 2 is 10.625 MW (heat
load/duration of the task, 85 MJ/8 h ). During the time interval of 12 to 16 h,
Task 2 is integrated with the cold task 5 and exchanges a heat load of 33.75 MJ.
The cooling requirement for Task 2 in this time interval is 42.5 MJ obtained by
TABLE 5.8
Scheduling Data Required for Case Study III
Unit Suitability Capacity (Ton) Revenue (103c.u./Ton)
R1 Task T1 0:100 —
R2 Task T2 0:100 —
R3 Task T3 0:100 —
R4 Task T4, T5, T6, T7 0:100 —
R5 Task T8, T9 0:150 —
R6 Task T10, T11, T12 0:50 —
R7 Task T10, T11, T12 0:50 —
R8 Task T13, T14, T15, T16, T17 0:100 —
R9 Task T13, T14, T16, T17 0:100 —
R10 Task T8 0:100 —
V0 Store S0 0:1000 0
V1 Store S1 0:100 0
V2 Store S2 0:100 0
V3 Store S3 0:100 0
V4 Store S4 0:100 0
V6 Store S6 0:100 0
V7 Store S7 0:100 0
V8 Store S8 0:100 0
V11 Store S11 0:100 0
V13 Store S13 0:100 0
V5, V9, V10 , V12 V5 for S5, V9 for S9, V10 for 0:0 0
S10, V12 for S12
V14 Store S14 0:1000 5
V15 Store S15 0:1000 5
V16 Store S16 0:1000 10
V17 Store S17 0:1000 7.5
V18 Store S18 0:1000 7.5
TABLE 5.9
Heating/Cooling Requirement for Case Study III
Heating/Cooling Operating
Task Requirement (MWh) Temperature (°C) Cost (c.u/MWh)
T1 40 + 0.2β heating 150 —
T2 70 + 0.15β cooling 200 —
T3 50 + 0.3β cooling 230 —
T4 50 + 0.3β heating 120 —
T5 55 + 0.25β heating 120 —
T6 60 + 0.6β heating 120 —
T7 45 + 0.45β heating 120 —
Steam 270 1
Cooling water 30 0.2
TABLE 5.10
Computational Results for Case Study III
Standalone Operation Standalone Proposed Model
Using Stamp and Operation Using with Direct Energy
Majozi (2011) Proposed Model Integration
Objective value (103 c.u.) 6740 6589.9 7910.7
Product 1 (ton) 100 100 90
Product 2 (ton) 600 600 600
Product 3 (ton) 103.6 165.6 120.8
Product 4 (ton) 100 50 55.2
Product 5 (ton) 100 88 100
Cooling water (MJ) 1447 1600 276.75
Steam (MJ) 1488.53 1453 251.95
Binary variable 425 550 1315
CPU time (s) 5000 5000 6008
Units
R10 70
R9 T14/15 T13/70 T16/100 T14/88.6
T15/100 T15/100 T15/100 T15/100 T17/100 T13/30 T15/100 T15/100

R8
R7 T10/50 T10/5 T10/50 T10/50 T12/50
T10/45 T12/50 T11/50 T10/50 T10/5

R6
R5 50 T9/15 T8/30 T9/88.6
V5 0.0 50 0.0
T5/100 T4/50 T4/50 T5/100 T4/50 T7/100 T4/50 T6/100 T5/100 T4/50 T4/50
R4
80 40 100 100 80 40
R3
p2 p3 p4 p6 p7 p8 p9 p10 p11 p12 p13 p14
R2 100 100 100 100 100 100 100 100 100 100 100 66.7
V1 100 200 324.8 379.9 479.9 379.9 410.9 310.9 210.9 241.9 141.9 41.9 66.7 0.0 12.4
100 100 100 100 100 100 100 42.7 100 100 100
R1
p1 p2 p3 p5 p6 p7 p8 p9 p10 p11 p12
8 16 24 32 40 48 56 64 72 80 88 96 104 112 120

Time (h)
FIGURE 5.11 Gantt chart for Case Study III using Stamp and Majozi (2011) for standalone
operation.
Units
R10 35 50
R9 T13/15 T13/35 T13/50 T16/50 T14/84.6 T14/81.02
R8 T15/100 T15/100 T15/100 T15/100 T17/88 T15/100 T15/100
R7 T10/50 T12/38 T11/50 T10/50
R6 T10/45 T10/50 T10/50 T12/50 T10/50
R5 T8/15 T9/84.6 T9/81.02
R4 T5/100 T6/50 T5/100 T4/50 T7/88 T4/50 T4/50 T4/100 T5/100 T4/50 T4/50
R3 40 40 20 100 22.6 100 77.4 40 10
R2 100 100 100 100 100 100 100 100 100 100 100 66.7
R1 100 100 100 100 100 100 100 100 100 100 42.7
8 16 24 32 40 48 56 64 72 80 88 96 104 112 120

Time (h)
FIGURE 5.12 Gantt chart for Case Study III using the proposed model for standalone
operation.
multiplying the energy requirement per hour with the duration of the heat exchange
(10.625 MW * 4 = 42.5). The deficit cooling requirement for Task 2, which is
(42.5 MJ− 33.75 MJ = 8.75 MJ), is satisfied with external cooling. The heating
requirement for Task 5 in this interval is fully satisfied with heat integration of
Task 2. The cooling requirement for Task 2 during the interval between 16 and
20 h is fully satisfied with energy integration of Task 1. The heating requirement
for Task 1 during this interval is 60 MJ and is satisfied partially with the energy
integration of Task 2 and the rest which is 17.5 MJ from external heating. The heat-
ing requirement for Task 5, which is 33.5 MJ during the interval between 16 and
20 h, is satisfied fully with external heating since it is not in the energy integrated
mode. The same principle is applied to calculate and satisfy the energy requirement
for each hot and cold task for the entire time horizon. The Gantt chart that shows
the amount of batch processed, the type of task performed in a unit and the start-
ing and finishing times for the energy integrated mode for this case is presented in
Figure 5.14.
42.5 MJ 8.12 MJ 36 h 37.5 MJ

T2 cw T5 st T5 T6 st T5 st
12 h 33.75 MJ 72.5 MJ 85 MJ 36 h 37.5 MJ 44 h
42.5 MJ 33.75 MJ
T2 cw T2 T3 cw
12..5 MJ 28 h 26.5 MJ
42.5 MJ
60 MJ 17.5 MJ 12 h 16 h 20 h 14 MJ 24 h 28 h 32 h
T2 cw
T1 st st T2 cw T3 cw T3
20 h 42.5 MJ
0.0 h 4 h 8h 8.75 MJ 42.5 MJ 60 MJ 60 MJ
T1 st
T1 st T1 st T1 7.58 MJ
17.5 MJ 17.5 MJ 40 h
25 MJ 5 MJ
T4 st T5 T4 T4 T7 st
65 MJ 80 MJ 65 MJ 65 MJ 85 MJ
T2 cw T2 cw T2 cw T2 cw T2
44 h 20 MJ 52 h 52 h 5 MJ 60 h 60 h 20 MJ 68 h 68 h 20 MJ 76 h 76 h 84 h
2 MJ
6 MJ
T3 cw T3 T3 cw
T3
60 MJ 60 MJ 60 MJ 60 MJ
T1 T1 T1 T1
48 h 56 h 64 h 72 h
32.5 MJ 32.5 MJ
T4 st T4 st
32.5 MJ 65 MJ 32.5 MJ 112 h
T2 cw T2 cw
10 MJ 96 h 12.5 MJ
14 MJ
20.5 MJ
T3 cw
T3 cw
60 MJ 42.5 MJ
104 h 108 h
T1
84 h 88 h 92 h 96 h
FIGURE 5.13 Heat exchange network for Case Study III for the time horizon of 120 h.
Units
R10
R9 T16/45.2 T14/28.8 T16/10 T17/100 T14/92
R8 T15/100 T15/100 T15/100 T13/90 T15/100 T15/100 T15/100
R7 T10/5 T10/50 T10/50 T10/50 T12/50
R6 T10/45 T11/45.2 T11/10 T10/50 T10/50
R5 T9/28 T8/90 T9/92
R4 T5/50 T5/70 T6/55.2 T5/80 T4/50 T5/100 T4/50 T4/50 T7/100 T4/50 T4/50 T4/50
R3 80 33.3 46.7 40 33.3 33.3 53.3 80 10
R2 100 100 100 100 100 100 100 100 100 100 100 50
R1 100 100 100 100 100 50.4 100 100 100 100 100 12.5
8 16 24 32 40 48 56 64 72 80 88 96 104 112 120

Time (h)
FIGURE 5.14 Gantt chart for Case Study III using direct heat integration.
5.6 CONCLUSIONS
An efficient continuous-time mathematical model for direct and indirect heat inte-
gration is presented. Most heat integration models rely on a predefined schedule,
which leads to suboptimal results. This chapter incorporates heat integration into the
scheduling framework and solved simultaneously. The model is capable of solving
for both direct and indirect heat integration. By using a heat storage vessel, a consid-
erable reduction in utility consumption is achieved. When this work is compared to
recent existing work, the formulation performs better in terms of both optimal objec-
tive value and CPU time. Future comunication will address the developed model to
inororate capital cost of heat exchanger.
NOMENCLATURE
Sets
SinJh {sinjh |sinjh task which needs cooling}
SinJc {sinjc |sinjc task which needs heating}
SinJ {sinj |sinj any task }
P { p| p time point}
U {u|u is a heat storage unit}
Parameters
cp ( sinjh ) Specific heat capacity for the heating task
cp ( sinjc ) Specific heat capacity for the cooling task
cp ( u ) Specific heat capacity for the heat storage
in
T
sinjh Inlet temperature of the heating task
out
T
sinjh Outlet temperature of the heating task
in
T
sinjc Inlet temperature of the cooling task
out
T
sinjc Outlet temperature of the cooling task
DT U
Maximum thermal driving force
DT Minimum thermal driving force
M Big-M mostly equivalent to the time horizon
QU Maximum heat requirement from the heating and cooling task
( )
price s p Price of a product
costst Cost of steam
costcw Cost of cooling water
H time horizon of interest
Variables
x ( sinjc , sinjh , p, pp )
Binary variable signifying whether heat integration occurs
between the hot and cold unit
z ( sinj , u, p, pp ) Binary variable signifying whether heat integration occurs
between a task and heat storage
y ( sinjh , p ) Binary variable associated to whether the hot state is active at
time point p or not
y ( sinjc , p ) Binary variable associated to whether the cold state is active at
time point p or not
v ( sinj , p, pp ) Binary variable associated to whether the hot and cold states are
active at time point p and extended time point pp
cl ( sinjh , p ) Cooling load required by the hot task at time point p
hl ( sinjc , p ) Heating load required by the cold task at time point p
avcl ( sinjh , p ) Average cooling load required by the hot task at time point p
using time average model
avhl ( sinjc , p ) Average heating load required by the cold task at time point p
mu ( sinjh , p ) Amount of material processed by the hot task
mu ( sinjc , p ) Amount of material processed by the cold task

( )
qs s p Amount of product produced at the end of the time horizon
tp ( sinj , p ) Finishing time of a task
tu ( sinj , p ) Starting time of a task
clp ( sinjh , p, pp ) Cooling load required by the hot task active at time point p
and extended time point pp
hlp ( sinjc , p, pp ) Heating load required by the cold task active at time point p
and extended time point pp
tuu ( sinj , p, pp ) Starting time of a task at time point p and extended time
point pp
tpp ( sinj , p, pp ) Finishing time of a task at time point p and extended time
point pp
tuu ( u, p, pp ) Starting time of a heat storage at time point p and extended
time point pp
tpp ( u, p, pp ) Finishing time of a heat storage at time point p and extended
time point pp
T in ( sinj , p, pp ) Inlet temperature of a task active at time point p and extended
time point pp
T out ( sinj , p, pp ) Outlet temperature of a task active at time point p and
extended time point pp
To ( u, p, pp ) Inlet temperature of a heat storage active at time point p and
T f ( u, p, pp ) Outlet temperature of a heat storage active at time point p and
Qe ( sinjh , sinjc , p, pp ) Amount of heat load exchanged by the hot and cold unit active
at time point p and extended time point pp
Q ( sinj , u, p, pp ) Amount of heat load exchanged between a task and heat
storage
cw ( sinjh , p ) External cooling water used by the hot task
st ( sinjc , p ) External heating used by the cold task
W ( u ) Capacity of heat storage unit
REFERENCES
Adonyi, R., Romero, J., Puigjaner, L., Friedler, F., 2003. Incorporating heat integration in
batch process scheduling. Applied Thermal Engineering. 23, 1743–1762.
Atkins, M.J., Walmsley, M.R.W., Neale, J.R., 2010. The challenge of integrating
non-continuous processes – Milk powder plant case study. Journal of Cleaner
Production. 18, 927–934.
Bancheva, N., Ivanov, B., Shah, N., Pantelides, C.C., 1996a. Heat exchanger network design
for multipurpose batch plants. Computers and Chemical Engineering. 20, 989–1001.
Bancheva, N., Ivanov, B., Shah, N., Pantelides, C.C., 1996b. Heat exchanger network design
Barbosa-Póvoa, A.P.F.D., Pinto, T., Novais, A.Q., 2001. Optimal design of heat-integrated
multipurpose batch facilities: A mixed-integer mathematical formulation. Computers
and Chemical Engineering. 25, 547–559.
Becker, H., Vuillermoz, A., Maréchal, F., 2012. Heat pump integration in a cheese factory.
Applied Thermal Engineering. 43, 118–127.
Boyadjiev, C.H.R., Ivanov, B., Vaklieva-Bancheva, N., Pantelides, C.C., Shah, N., 1996.
Optimal energy integration in batch antibiotics manufacture. Computers and Chemical
Engineering. 20, S31–S36.
Bozan, M., Borak, F., Or, I., 2001. A computerized and integrated approach for heat exchanger
network design in multipurpose batch plants. Chemical Engineering Process. 40,
511–524.
Chen, C.L., Chang, C.Y., 2009. A resource-task network approach for optimal short-term/
periodic scheduling and heat integration in multipurpose batch plants. Applied Thermal
Chen, C.L., Ciou, Y.J., 2008. Design and optimization of indirect energy storage sys-
tems for batch process plants. Industrial and Engineering Chemical Research. 47,
4817–4829.
Chew, Y.H., Lee, C.T., Foo, C.Y., 2005. Evaluating heat integration scheme for batch produc-
tion of oleic acid. Malaysian Science and Technology Congress (MSTC). 18–20.
Clayton, R.W., 1986. Cost reductions on an edible oil refinery identified by a process
integration study at Van den Berghs and Jurgens Ltd., Report Nr RD14/14. UK:
Energy Efficiency Office R&D, Energy Technology Support Unit (ETSU), Harwell
Laboratory.
Corominas, J., Espuña, A., Puigjaner, L., 1993. A new look at energy integration in multi-
product batch processes. Computers and Chemical Engineering. 17, 15–20.
Fernández, I., Renedo, C.J., Pérez, S.F., Ortiz, A., Mañana, M., 2012. A review: Energy
recovery in batch processes. Renewable and Sustainable Energy Review. 16,
2260–2277.
Foo, D.C.Y., Chew, Y.H., Lee, C.T., 2008. Minimum units targeting and network evolution for
batch heat exchanger network. Applied Thermal Engineering. 28, 2089–2099.
Fritzson, A., Berntsson, T., 2006. Efficient energy use in a slaughter and meat process-
ing plant – opportunities for process integration. Journal of Food Engineering. 76,
594–604.
Halim, I., Srinivasan, R., 2009. Sequential methodology for scheduling of heat-integrated
batch plants. Industrial Engineering and Chemical Research. 48(18), 8551–8565.
Halim, I., Srinivasana, R., 2011. Sequential methodology for integrated optimization of
energy and water use during batch process scheduling. Computers and Chemical
Hellwig, T., Thöne, E., 1994. Omnium: A method for optimization of waste heat utilization
in Germany eni verfahern zur optimierung der abwarmenutzunga. BWK (Brennstoff,
Warme, Kraft). 46, 393–397 [in German].
Ivanov, B., Bancheva, N., 1994. Optimal reconstruction of batch chemical plants with
regard to maximum heat recuperation. Computers and Chemical Engineering. 18,
313–317.
José, A.Q., Araceli, G., María, G.A., Jalel, L., 2013. Heat integration options based on pinch
and exergy analyses of a thermosolar and heat pump in a fish tinning industrial process.
Energy. 55, 23–37.
Kallrath, J., 2002. Planning and scheduling in the process industry. OR Spectrum. 24(3),
219–250.
Kemp, I.C., Deakin, A.W., 1989. The cascade analysis for energy process integration of batch
processes. Part 1. Calculation of energy targets. Chemical Engineering and Research
Design. 67, 495–509.
Kemp, I.C., Macdonald, E.K., 1987. Energy and process integration in continuous and batch
processes. Innovation in process energy utilization. IChemE Symposium Series. 105,
185–200.
17, 211–227.
Krummenacher, P., Favrat, D., 2001. Indirect and mixed direct–indirect heat integra-
tion of batch processes based on Pinch Analysis. International Journal of Applied
Thermodynamics. 4, 135–143.
Liu, L., Du, J., Xiao, F., Chen, L., Yao, P., 2011. Direct heat exchanger network synthesis for
batch process with cost targets. Applied Thermal Engineering. 31, 2665–2675.
Maiti, D., Jana, A.K., Samanta, A.N., 2011. A novel heat integrated batch distillation scheme.
Applied Energy. 88, 5221–5225.
Majozi, T., 2006. Heat integration of multipurpose batch plants using a continuous-time
framework. Applied Thermal Engineering. 26, 1369–1377.
Majozi, T., 2009. Minimization of energy use in multipurpose batch plants using heat storage:
An aspect of cleaner production. Journal of Cleaner Production. 17, 945–950.
Majozi, T., Zhu, X.X., 2001. A novel continuous-time MILP formulation for multipurpose
batch plants. 1. Short-term scheduling. Industrial Engineering and Chemical Research.
40(25), 5935–5949.
Morrison, A.S., Walmsley, M.R.W., Neale, J.R., Burrell, C.P., Kamp, P.J.J., 2007. Non-
continuous and variable rate processes: Optimization for energy use. Asia-Pacific
Journal of Chemical Engineering. 2(5), 380–387.
Muster-Slawitsch, B., Weiss, W., Schnitzer, H., Brunner, C., 2011. The green brewery con-
cept Energy efficiency and the use of renewable energy sources in breweries. Applied
Thermal Engineering. 31, 2123–2134.
Obeng, E.D.A., Ashton, G.J., 1988. On pinch technology based procedures for the design of
batch processes. Chemical Engineering and Research Design. 66, 255–259.
Papageorgiou, L.G., Shah, N., Pantelides, C.C., 1994. Optimal scheduling of heat-integrated
multipurpose plants. Industrial and Engineering Chemical Research. 33(12),
3168–3186.
Pinto, T., Novais, A.Q., Barbosa-Póvoa, A.P.F.D., 2003. Optimal design of heat-integrated
multipurpose batch facilities with economic savings in utilities: A mixed integer math-
ematical formulation. Annals of the Operation Research. 120, 201–230.
Pires, A.C., Fernandes, C.M., Nunes, C.P., 2005. An energy integration tool for batch
process, sustainable development of energy, water and environment systems. In
Proceedings of the Third Dubrovnik Conference. pp.175–185.
Pozna, A., Ivanov, B., Bancheva, N., 1998. Design of a heat exchanger network for a system of
batch vessels. Hungarian Journal of Industry and Chemistry. 26, 203–211.
Stamp, J.D., Majozi, T., 2011. Optimum heat storage design for heat integrated multipurpose
batch plants. Energy. 36(8), 5119–5131.
Stolze, S., Mikkelsen, J., Lorentzen, B., Petersen, P.M., Qvale, B., 1995. Waste-heat recovery
in batch processes using heat storage. Journal of Energy Resources and Technology.
117, 142–149.
Science. 60, 2679–2702.
Timothy, G, Walmsley, M.R.W., Martin, J., Atkins, M.J., Neale, J.R., 2014. Integration of
industrial solar and gaseous waste heat into heat recovery loops using constant and
variable temperature storage. Energy. 75, 53–67.
Tokos, H., Pintarič, Z.N., Glavič, P., 2010. Energy saving opportunities in heat integrated
beverage plant retrofit. Applied Thermal Engineering. 30, 36–44.
Uhlenbruck, S., Vogel, R., Lucas, K., 2000. Heat integration of batch processes. Chemical
Engineering and Technology. 23(3), 226–229.
Vaselenak, J.A., Grossmann, I.E., Westerberg, A.W., 1986. Heat integration in batch pro-
cessing. Industrial and Engineering Chemical Process Design Development. 25(2),
357–366.
6 Design and Synthesis
of Heat-Integrated
Batch Plants Using an
Effective Technique
6.1 INTRODUCTION
The aim of this chapter is to address design, synthesis and scheduling simultane-
ously with the consideration of economic savings in utility requirements, while
considering the cost of both the auxiliary structures. The recent design and syn-
thesis model by Seid and Majozi (2012) is extended to incorporate the design of the
associated utility facility, since it is proven to result in better design objectives and
computational efficiencies. An additional feature of the proposed model is the deter-
mination of optimal pipe connection between processing equipment. The model is
implemented in a case study in order to demonstrate its application. From the case
study, the profit is increased by 20% and the total utility requirement is reduced by
41.1% for the design and synthesis of energy-integrated batch plant compared to the
basic design.
There is a great deal of interest for manufacturing of fine chemicals, pharmaceuti-
cal products, polymers and food and beverage using batch operations because of
the advantage of producing low-volume, high-quality products, flexibility to adopt
complex operations due to fast market change and suitability to manufacture differ-
ent products using the same facility. In these batch facilities, the different products
compete for the available resources like equipment, utilities, manpower and storage,
which makes the design and operation of this plant a challenging task. The concep-
tual design problem must determine the number and capacity of the major processing
equipment items, utilities and storage tanks so as to meet these design and produc-
tion objectives at the lowest possible capital and operating cost. The modelling and
solution of multipurpose batch processes has received considerable attention in the
last two decades. The literature review which follows covers published work under
two major headings: basic design and extended design.
147
6.2.1 Basic Grass Root Design for Multipurpose Batch Plants

Literatures under this category considered simple choice of equipment and associ-
ated scheduling. The formulation of Suhami and Mah (1982) was based on the work
of Grossmann and Sargent (1979) for multi-product batch plants. A two-level prob-
lem is derived where the upper level is characterized by the definition of production
campaigns using a heuristic procedure, and the lower level includes the design prob-
lem with additional constraints resulting in a mixed integer nonlinear programming
(MINLP) problem deciding allocation of tasks to units during the time of planning.
Klossner and Rippin (1984) looked into the unique assignment case and enumerated
all possible product configurations by solving a set partitioning problem followed by
the solution of a MINLP for each configuration. The set partitioning problem was
further studied by Imai and Nishida (1984), who presented an improved heuristic
procedure to solve it. However, no comparative results were presented.
Vaselenak et al. (1987) avoided the two-stage nature of these approaches by pro-
posing a superstructure embedding all possible product configurations. However, a
major difficulty appears to be associated with the derivation of the limiting set of
production time constraints, the so-called horizon constraint. A more complete for-
mulation based on a theory of linear inequalities was presented by Faqir and Karimi
(1989) where for every case involving a unique production route the equivalent hori-
zon constraints are identified. This work was later extended (Faqir and Karimi, 1990)
to multiple production routes assuming unique task-unit allocations with discrete
batch size equipment. Cerda et al. (1989) proposed a more general problem than the
ones mentioned earlier. Here, the restrictive assumption that all units within a stage
must be allocated to a task if the stage itself is allocated to the same task was relaxed.
Papageorgaki and Reklaitis (1990) proposed a mixed integer linear programming
(MILP) formulation which accommodates equipment used in and out of phase, units
available with different sizes in a processing stage and tasks that can be processed in
different units. Most of the previous published works were restricted to the following
key assumptions: (1) There is a pre-specified assignment of equipment items to prod-
uct tasks. (2) Parallel production is only allowed for products that have no common
equipment requirements. (3) All units of a given type are identical and can be used
in the out-of-phase mode. (4) All units of a given type are devoted to the production
of only one product at a time. Shah and Pantelides (1991) presented a formulation
incorporating long-term campaign planning. Previous work was limited to all the
processing steps involved in the manufacturing of a product taking place within
any campaign producing it. This is relaxed with the use of intermediate storage to
decouple the manufacturing of each product into several stages each of which can be
run independently in campaign mode.
Voudouris and Grossmann (1996) presented a formulation for a special class of
sequential multipurpose batch plants where not all the products use the same pro-
cessing stages. A cyclic schedule was implemented that has the effect of aggregat-
ing the number of batches for each product in order to allow the consideration of
problems of practical size. It is shown that the no-wait characteristics of substrains
can be exploited with a reduction scheme that has the effect of greatly decreasing
the dimensionality of the problem. This reduction scheme can be complemented
Design and Synthesis of Heat-Integrated Batch Plants 149
with a tight formulation of the underlying disjunctions in the MILP problem to

reduce the computational expense.
Xia and Macchietto (1997) presented a general formulation considering linear/
nonlinear models for capital cost of processing units/storage vessels and process-
ing task models with fixed, linear and nonlinear processing time in a general form.
The formulation resulted in a MINLP solved by a stochastic MINLP optimizer.
A short-term scheduling model was implemented based on discrete time representa-
tion which allowed batches of a product to be processed in any sequence.
Barbosa-Povoa and Pantelides (1997) developed a model for single campaign
structure with fixed product slate is assumed within a non-periodic operation.
A resource-task network (RTN) scheduling framework by Pantelides (1994) was
used as a platform. The model allows a detailed consideration of the design problem
taking into account the trade-offs between capital costs, revenues and operational
flexibility. The optimal solution involves the selection of the required processing and
storage equipment items. Lin and Floudas (2001) presented a formulation based on
the scheduling formulation of Ierapetritou and Floudas (1998) using a unit-specific
event-point and continuous-time representation. The formulation addressed a single
campaign production approach where multiple batches of a product can be produced
in the given time horizon. A comparison with earlier formulations was given. Heo
et al. (2003) developed a three-step procedure for the design of multipurpose batch
plants using cyclic scheduling. The first MILP model gives the minimum number of
equipment units required to produce the products; this configuration is used as an
initial plant configuration. The second MILP model determines the minimum cycle
time and the third determines the equipment size and scheduling that minimizes the
cost. The method gave a better solution compared to the separable programming
and the evolutionary design method proposed by Fuchino et al. (1994). Castro et al.
(2005) presented a general mathematical formulation for the simultaneous design
and scheduling of multipurpose plants. The formulation is based on the RTN process
representation and can handle both short-term and periodic problems. The perfor-
mance of the formulation is illustrated through the solution of two periodic example
problems that have been examined in the literature, where the selection and design of
the main equipment items and their connecting pipes is considered. They also inves-
tigated two example problems by changing demand rates, with the results showing
that this set of parameters, which is usually subject to a high degree of uncertainty,
affects not only the optimal structure and cost of the plant but also the optimal
cycle time. This is a clear indication that both the design and scheduling aspects
of the problem should be considered simultaneously. A comparison with an earlier
approach of Lin and Floudas (2001) is also presented.
6.2.2 Extended Design
Literature under this category besides the basic design important features, such as
plant topology, layout and rational use of utility, has been covered. Barbosa-Povoa
and Macchietto (1994) presented for the first time the design of multipurpose batch
plants considering plant topology. The formulation is based on the maximal state
task network (mSTN) which describes both the recipes and plant possible super-
structure. The model optimizes the structural aspect of the plant and the associated
production schedule accounting for capital cost of equipment, pipework, operating
costs and revenues. Both short-term and cyclic scheduling was studied. Penteado and
Ciric (1996) and Barbosa-Povoa et al. (2001a) addressed layout aspects in the design
problems. A 2D small case study was analysed.
Georgiadis et al. (1997) used a space discretization technique to consider the
allocation of equipment items to floor as well as the block layout of each floor.
The main drawback of this formulation is a suboptimal solution may result due to
the discretization of the available space. Barbosa-Povoa et al. (2002) proposed a
general model where both 2D and 3D space was considered. Irregular shapes for
each piece of equipment could also be accounted for in the formulation. Patziatsis
et al. (2005) formulated simultaneous layout, design and planning of pipeless batch
plants. Barbosa-Povoa (2007) reviewed the design and retrofit of batch plants. The
authors included published literature for the last two decades. From the conclusion
of the authors, there is still a gap in developing efficient techniques addressing large-
scale problems, detail process operation (cleaning in place, change over and utility
requirements) and plant layout and topology. Pinto et al. (2008) presented a compara-
tive analysis between the state task network (STN), m-STN and RTN representations
for the design of multipurpose batch plants. A number of problems are solved and
compared with the formulations based on the different representations.
The increasing importance of utilities rationalization within the design of multi-
purpose batch plants was first addressed by Barbosa-Povoa et al. (2001b), who pre-
sented a mathematical formulation for the detailed design of multipurpose batch
process facilities. This work was later extended by Pinto et al. (2003) where eco-
nomic savings in utility requirements were obtained while considering both the cost
of the auxiliary structures (i.e. heat exchanger through their transfer area) and the
design of the utility circuits and associated piping costs.
In this work, an efficient formulation for the design of multipurpose batch plants
that considers plant topology and heat integration for sustainable design is devel-
oped. The formulation is posed as a MILP in which the binary variables are the
structural choice variables. The proposed model is able to accommodate equipment
used in and out of phase, units available in two or more sizes within a processing
stage, multiple choices of equipment types for each product task, unit dependent
processing time and processing time dependent on batch.
The optimal plant synthesis and design for multipurpose batch plants can be achieved
by developing a model that can solve the following problem:
Given:
1. The product recipes (STNs) describing the production of one or more prod-
ucts over a single campaign structure
2. The plant flowsheet with all possible equipment units to be installed and the
involved connectivity
3. The equipment units’ suitability to perform the process/storage tasks

4. The connections’ suitability to transfer materials
5. The operating and capital cost data involved in the plant
6. The time horizon of planning
7. The production requirements over the time horizon
8. Scheduling data
9. Heat integration data such as operating temperature of a task, heat load of a
task, cooling and heating cost
Determine:
1. The optimal plant configuration (i.e. number and type of equipment and the
optimal design size of the equipment)
2. Minimum cooling and heating requirement
3. Detail of the heat exchanger network
4. A process schedule that allows the selected resources to achieve the required
production (i.e. the starting and finishing times of all tasks, storage policies,
batch sizes, amounts transferred and allocation of tasks to the equipment),
so as to optimize the economic performance of the plant, measured in terms
of the capital expenditure and the operating costs and revenues
In this work, the mathematical modelling is developed and has the following constraints.
6.4.1 Equipment Existence Constraints

Constraint (6.1) states that in order to execute a task in the unit, the unit must be
selected first.
å * ,j
y ( sin, j , p ) £ e ( j ) , "j Î J , pÎP (6.1)
sin , j ÎSin
6.4.2 Unit Size Constraints

Constraint (6.2) implies that if the unit is selected, the design capacity should be
between the minimum and maximum design capacity.
V jl e ( j ) £ ss ( j ) £ V jue ( j ) , "j Î J (6.2)
Constraint (6.3) implies that the total amount of all the states consumed at time
point p is limited by the capacity of the unit which consumes the states and repre-
sents the lower and upper bounds in the capacity of a given unit that processes the
effective state.
( )
g sLin, j ss ( j ) - V jL 1 - y ( sin, j , p ) £ mu ( sin, j , p ) £ gUsin, j ss ( j ) ,
"p Î P, j Î J , sin, j Î Sin, J (6.3)
Constraints (6.4) and (6.5) ensure the amount of material stored at any time point
p is limited by the capacity of the storage.
qs ( s, p ) £ s ( v ) , "s Î S, p Î P, v Î V (6.4)
VvL eu ( v ) £ s ( v ) £ VvU eu ( v ) , "s Î S, p Î P, v Î V (6.5)
6.4.4 Material Balance for Storage

Constraint (6.6) states that the amount of material stored at each time point p is the
amount stored at the previous time point adjusted by some amount resulting from the
difference between state s produced by tasks at the previous time point (p − 1) and used
by tasks at the current time point p. This constraint is used for an intermediate state rather
than a product, since the latter is not consumed, but only produced within the process.
qs ( s, p ) = qs ( s, p - 1) - åmuu ( s, j, p ) + åmuu ( s, j, p - 1) ,
jÎJ ssc jÎJ ssp
"p Î P, p ³ 1, s Î S (6.6)

Constraint (6.7) is used for the material balance around storage at the first time
point.
qs ( s, p ) = QO ( s ) - åmuu ( s, j, p ) , "p Î P, p = 1, s Î S (6.7)

jÎJ ssc
Constraint (6.8) states that the amount of product stored at time point p is the
amount stored at the previous time point and the amount of product produced at time
point p.
( ) (
qs s p, p = qs s p, p - 1 + ) år sp
sin , j mu ( sin, j , p ) , "p Î P, s p Î S p (6.8)
sin , j Îsinp ,J

6.4.5 Material Balance around the Processing Unit

Constraint (6.9) is used to cater for the material processed in the unit and equals
the amount of material directly coming from the unit producing it and from the
storage.
sin , j mu ( sin, j , p ) = muu ( s, j, p ) +

rsc å mux ( s, j, j¢, p ) ,
j ¢Î jssp
"p Î P, j, j¢ Î J , sin, j Î Sin, J , s Î S (6.9)

Constraint (6.10) is used to define the amount of material produced at time

point p which is sent to storage at the same time point p and to units that consume it
at time point p + 1.
sin , j mu ( sin, j , p ) = muu ( s, j, p ) +

rsp å mux ( s, j¢, j, p + 1) ,
j ¢Î jssc
"p Î P, j, j¢ Î J , sin, j Î Sin, J , s Î S (6.10)
6.4.6 Existence Constraints for Piping

Constraints (6.11) and (6.12) are applicable to ensure whether pipe connections exist
between the processing units and storage as well as to ensure the amount of material
transferred through the connection is limited by the design capacity of the connec-
tion. The material transfer time is assumed fixed.
V jL,v z ( j, v ) £ pip ( j, v ) £ V jU,v z ( j, v ) , "j Î J , v Î V (6.11)
muu ( s, j, p ) £ pip ( j, v ) , "j Î J , v Î V , p Î P, s Î S (6.12)
Constraints (6.13) and (6.14) are similar to Constraints (6.11) and (6.12) and are
used when the connection is between processing units.
V jL, j ¢ w ( j, j¢ ) £ pipj ( j, j¢ ) £ V jU, j ¢ w ( j, j¢ ) , "j Î J (6.13)
mux ( s, j, j¢, p ) £ pipj ( j, j¢ ) , "j, j¢ Î J , p Î P, s Î S (6.14)
6.4.7 Duration Constraints (Batch Time as a Function of Batch Size)

Constraint (6.15) describes the duration constraint modelled as a function of batch
size where the processing time is a linear function of the batch size. For zero-wait
(ZW), only the equality sign is used.
t p ( sin, j , p ) ³ tu ( sin, j , p ) + t ( sin, j ) y ( sin, j , p ) + b ( sin, j ) mu ( sin, j , p ) ,
"j Î J , p Î P, sin, j Î Sin, J (6.15)
6.4.8 Sequence Constraints
The following two subsections address the proper allocation of tasks in a given unit
to ensure the starting time of a new task is later than the finishing time of the previ-
ous task.
6.4.8.1 Same Task in Same Unit

Constraint (6.16) states that a state can only be used in a unit, at any time point, after
all the previous tasks are complete.
tu ( sin, j , p ) ³ t p ( sin, j , p - 1) , "j Î J , p Î P, sin, j Î Sin*, j (6.16)
6.4.8.2 Different Tasks in Same Unit

Constraint (6.17) states that a task can only start in the unit after the completion of
all the previous tasks that can be performed in the unit.
( )
tu ( sin, j , p ) ³ t p s¢in, j , p - 1 , "j Î J , p Î P, sin, j ¹ s¢in, j , sin, j , s¢in, j Î Sin*, j

(6.17)
If the state is consumed and produced in the same unit, where the produced state
is unstable then in addition to Constraints (6.17) through (6.19) are used.
( ) ( ) ( (
t p sinusp, j , p - 1 ³ tu sinusc, j , p - H 1 - y sinusp, j p - 1 , ))
"j Î J , p Î P, s usc
in, j ÎS usc
in, j , susp
in, j ÎS usp
in, j (6.18)
( ) ( )
t p sinusp, j , p ³ t p sinusp, j , p - 1 , "j Î J , p Î P, sinusp, j Î Sinusp, j

(6.19)
It should be noted that in this particular situation tu sinusc, j , p = tu ( sin, j , p ) and ( )

( )
t p sinusc, j , p = t p ( sin, j , p ) . Consequently, Constraints (6.17) and (6.18) enforce
tu (susc
in, j , p) = t (s
p
usc
sin, j , p -1 . )
6.4.9 Sequence Constraints for Different Tasks in Different Units
These constraints state that for different tasks that consume and produce the same state,
the starting time of the consuming task at time point p must be later than the finishing
time of any task at the previous time point p − 1 provided that the state is used.
6.4.9.1 If an Intermediate State s Is Produced from One Unit

Constraints (6.20) and (6.21) work together in the following manner:
( )
r sinsp, j mu ( sin, j , p - 1) £ qs ( s, p ) + V jU t ( j, p ) , "j Î J , p Î P, sin, j Î Sispn, J

(6.20)
((
tu ( sin, j ¢ , p ) ³ t p ( sin, j , p - 1) - H 2 - y ( sin, j , p - 1) - t ( j, p ) , ))
in, J (6.21)
Constraint (6.20) states that if the state s is produced from unit j at time point
p − 1 but is not consumed at time point p by another unit j′, that is, t(j, p) = 0, then the
amount produced cannot exceed allowed storage, that is, qs(s, p). On the other hand,
if state s produced from unit j at time point p − 1 is used by another unit j′, then the
amount of state s stored at time point p, i.e. qs(s,p), is less than the amount of state s
produced at time point p − 1. The outcome is that the binary variable t(j, p) becomes
1 in order for Constraint (6.20) to hold. Constraint (6.21) states that the starting time
of a task consuming state s at time point p must be later than the finishing time of
a task that produces state s at the previous time point p − 1, provided that state s is
used. Otherwise, the sequence constraint is relaxed.
6.4.9.2 If an Intermediate State Is Produced from More than One Unit

Constraint (6.22) states that the amount of state s used at time point p can either
come from storage or from other units that produce the same state, depending on
the binary variable t(j, p). If the binary variable t(j, p) is 0, which means that state s
produced from unit j at time point p − 1 is not used at time point p, then Constraint
(6.21) is relaxed. If t(j, p) is 1, state s produced from unit j at time point p − 1 is used;
as a result Constraint (6.21) holds. Although Constraint (6.22) is nonlinear, it can be
linearized exactly using the Glover transformation31.
å sc
sin , j ÎSin
sin , j mu ( sin, j , p ) £ qs( s, p - 1) +
rsc å sp
sin , j ÎSin
sin , j mu ( sin, j , p - 1) t ( j, p )
rsp
,J ,J
"j Î J , pÎP (6.22)
6.4.9.3 Sequence Constraints for Completion of Previous Tasks

Constraint (6.23) states that a consuming task can start after the completion of the
previous task. Constraint (6.23) takes care of proper sequencing time when a unit
uses material which is previously stored, that is, when the producing task is active at
time point p − 2 and later produces and transfers the material to the storage at time
point p − 1. This available material in the storage at time point p − 1 is then used by
the consuming task in the next time points. This necessitates that the starting time of
the consuming task must be later than the finishing time of the producing task at time
point p − 2. Consequently, Constraint (6.23) together with Constraint (6.21) results
in a feasible sequencing time when the consuming task uses material, which is pre-
viously stored or/and material which is currently produced by the producing units.
(
tu ( sin, j¢ , p ) ³ t p ( sin, j , p - 2 ) - H 1 - y ( sin, j , p - 2 ) , )
in,J , sin, j¢ Î S sc
in, j¢ (6.23)
6.4.10 Sequence Constraints for Finite Intermediate Storage (FIS) Policy

According to Constraints (6.24) and (6.21), the starting time of a task that consumes
state s at time point p must be equal to the finishing time of a task that produces state
s at time point p − 1, if both consuming and producing tasks are active at time point
p and time point p − 1, respectively.
(
tu ( sin, j ¢ , p ) £ t p ( sin, j , p - 1) + H 2 - y ( sin, j ¢ , p ) - y ( sin, j , p - 1) )
in, J (6.24)

The usage and the production of states should be within the time horizon of interest.
These conditions are expressed in Constraints (6.25) and (6.26).
tu ( sin, j , p ) £ H , "sin, j Î Sin, J , p Î P, j Î J (6.25)
t p ( sin, j , p ) £ H , "sin, j Î Sin, J , p Î P, j Î J (6.26)
6.4.12 Energy Integration Constraints

will be integrated with one cold unit when direct heat integration takes place.

åx (i , i , p ) £ y (i , p ),
ic
h c h "p Î P, ih Î I h

(6.27)

åx (i , i , p ) £ y (i , p ),
ih
h c c "p Î P, ic Î I c

(6.28)
Constraint (6.29) ensures that minimum thermal driving forces are obeyed.
(
T ( ih ) - T ( ic ) ³ DT min - M 1 - x ( ih , ic , p ) , "p Î P, ic , ih Î I )
(6.29)
Constraint (6.30) states that the cooling of a heat source will be satisfied by direct
heat integration as well as external utility if required.
( E (i ) y (i , p ) + x ( i ) mu ( i , p )) ( tp ( i , p ) - tu (i , p ))
h h h h h h

= cw( ih , p ) + åxx(i , i , p),
ic
h c "p Î P, ih Î I h

(6.30)
Constraint (6.31) ensures that the heating of a heat sink will be satisfied by direct
heat integration as well as external utility if required.
( E (i ) y (i , p ) + x (i )) ( tp (i , p ) - tu(i , p))
c c c c c

= st ( ic , p ) + åxx(i , i , p),
ih
h c "p Î P, ic Î I c

(6.31)
cold unit is limited by the total duration of the cold unit.
( )(
w ( ih , ic , p ) £ E ( ih ) y ( ih , p ) + x ( ih ) mu ( ih , p ) tp ( ic , p ) - tu ( ic , p ) , )
"p Î P, ic Î I c , ih Î I h (6.32)
Constraint (6.33) ensures that the amount of heat transferred from the hot unit to
the cold unit is limited by the duration of the hot unit.
( )(
w ( ih , ic , p ) £ E ( ic ) y ( ic , p ) + x ( ic ) mu ( ic , p ) tp ( ih , p ) - tu ( ih , p ) , )
"p Î P, ic Î I c , ih Î I h (6.33)
cold unit is determined by the value of the binary variable x ( ih , ic , p ).
w ( ih , ic , p ) £ Mx ( ih , ic , p ) , "p Î P, ic Î I c , ih Î I h (6.34)
Constraints (6.35) and (6.36) are used to determine the heat exchanger area
required.
( )( )
UA T ( ih ) - T ( ic ) tp ( ih , p ) - tu ( ih , p ) ³ w ( ih , ic , p ) , "ih Î I h , ic Î I c , p Î P
(6.35)
( )( )
UA T ( ih ) - T ( ic ) tp ( ic , p ) - tu ( ic , p ) ³ w ( ih , ic , p ) , "ih Î I h , ic Î I c , p Î P
(6.36)
Constraints (6.37) and (6.38) are used to synchronize units when heat integration
takes place.
( )
tp ( ih , p ) ³ tp ( ic , p ) - M 1 - x ( ih , ic , p ) , "p Î P, ic Î I c , ih Î I h (6.37)
( )
tp ( ih , p ) £ tp ( ic , p ) + M 1 - x ( ih , ic , p ) , "p Î P, ic Î I c , ih Î I h (6.38)
Constraints (6.39) and (6.40) are used to cater for external utilities.

uc = ååcw (i , p ) ,
p ic
c "p Î P, ic Î I c

(6.39)

uh = ååst (i , p ) ,
p ih
h "p Î P, ih Î I h

(6.40)
Constraint (6.41) is the objective function expressed as maximization of profit. This
is obtained from revenue from the sale of products less operating costs for tasks, raw
material costs and capital costs from piping and equipment.
æ
ç åprice ( s ) q ( s , p ) -å å mu ( s
p
s
p
inj , p) - ö
÷
maximize ç ÷* ( AWH /H )
sp p sinj ÎSrm
ç ÷
çç åå( )
FOC * y ( sinj , p ) + VOC *mu ( sinj , p ) - Ccuc - Chuh ÷
÷
è p sinj ÎSinJ ø
- åå (CNC * z ( j, v ) + VCN * pip ( j, v ))
jÎJ vÎV
- åå ( FCNC * w ( j, j¢) + VCNC * pipj ( j, j¢))

j ¢ÎJ jÎJ
- å ( FEC * e ( j ) + VEQ * ss ( j ))
jÎJ

- å(
vÎV
)
FECS * eu ( v ) + VEQS * s ( v ) , "p = P, s p Î S p (6.41)
Constraint (6.42) is the objective function expressed as minimization of capital

and operating cost if the demand for the products is known beforehand within the
specified time horizon.

æ
ç åå mu ( sinj , p ) + ö
÷
minimize ç ÷ *( AWH/H )
p sinj ÎSrm
ç ÷
çç åå( )
FOC * y ( sinj , p ) + VOC * mu ( sinj , p ) + Ccuc + Chuh÷
÷
è p sinj ÎSinJ ø
æ+
ç åå ( CNC * z ( j, v ) + VCN * pip ( j, v ) ) ö
÷
ç j ÎJ v Î V ÷
ç ÷
ç +
ç åå (
j ¢ÎJ jÎJ
FCNC * w ( j, j¢ ) + VCNC * pipj ( j, j¢ ) ÷
÷
)
ç ÷ * CCF ,
ç +
ç
jÎJ
å( FEC * e ( j ) + VEQ * ss ( j ) ) ÷
÷
ç ÷
ç ÷
ç +
ç
è vÎV
å( FECS * eu ( v ) + VEQS * s ( v ) ) ÷
÷
ø
"p = P, s Î S p p
(6.42)
Product 1
s7
40% Int BC
40% 60%
10%
Feed A Hot A 60%
s9 Separation
s6
Impure E
Int AB 80% 90%
50%
50% 20% Product 2
Feed B
(s3) (s4)
Feed c
FIGURE 6.1 STN representation for the case study.
6.5 CASE STUDY
This example, which was first examined by Kondili et al. (1993), is studied exten-
sively in literature. It is a relatively complex problem and is often used in literature
to check the efficiency of models in terms of optimal objective value and CPU time
required. The plant has many common features of a multipurpose batch plant, with
the following features: units performing multiple tasks, multiple units suitable for
a task, states shared by multiple tasks and different products produced following
different production paths. The STN for this case study is depicted in Figure 6.1.
Full connectivity between equipment is assumed. The plant superstructure is
shown in Figure 6.2. The scheduling and design data are presented in Table 6.1.
The raw material costs for state A, B and C are 1, 0.5 and 0.2 c.u. respectively.
The selling price for product 1 and 2 are 4 and 7 respectively. Reaction 1 oper-
ates at 200°C and it is exothermic which requires 600 kWh/m.u. of cooling load.
Reaction 2 operates at 70°C and it is endothermic which requires 800 kWh/m.u.
of heating load. Reaction 3 operates at 160°C and it is exothermic which requires
700 kWh/m.u. of cooling load. Steam cost and cooling water cost are 0.7 and
0.15 c.u/MJ respectively.
6.6 RESULTS AND DISCUSSION

The computational results found are depicted in Tables 6.2 and 6.3. An overall
profit of 2.6 × 108 was obtained using the model that catered for energy integra-
tion which is much better than the objective value of 2.08 × 108 obtained by the
model without considering energy integration. Consequently, a 20% increase in
profit was achieved by allowing heat integration between the hot and cold tasks.
From Table 6.3, a significant saving in hot utility can be attained by integrating the
V1 H V4 V8
R1
V2
V5
St V9
V3
R2
V6
V7
FIGURE 6.2 Superstructure for the case study.
TABLE 6.1
Equipment and Scheduling Data for the Case Study
Processing
Unit Capacity (ton/m2) Suitability Time Cost Model (k c.u.)
Heater 5.0–30.0 Heating 1 200.0 + 10 s
Reactor 1 5.0–50.0 Reaction 1 2 5000.0 + 15 s
Reaction 2 2
Reaction 3 1
Reactor 2 5.0–50 Reaction 1 2 5000.0 + 10s
Reaction 2 2
Reaction 3 1
Still 5.0–20.0 Separation 2.5 150.0 + 5 s
Vessel 4 5.0–50.0 (Hot A) 50.0 + 3 s
Vessel 5 5.0–50.0 (Int BC) 70.0 + 3 s
Vessel 6 5.0–50.0 (Int AB) 60.0 + 4 s
Vessel 7 5.0–50.0 (Impure E) 90.0 + 5 s
Heat exchanger 1 0.0–200.0 RX1–RX2 60.0 + 5 s
Heat exchanger 2 0.0–200.0 RX2–Rx3 60.0 + 5 s
TABLE 6.2
Equipment Design Capacity for the Case Study
Design Capacity with Design Capacity
Equipment Capacity Consideration of without Consideration
Unit Range Energy Integration of Energy Integration
Heater 5–30 19 13.3
Reactor 1 5–50 47.7 20
Reactor 2 5–50 50 20
Still 5–20 20 33.3
Vessel 4 5–50 19 13.3
Vessel 5 5–50 30.6 28.7
Vessel 6 5–50 32.1 14
Vessel 7 5–50 33.3 0
Heat exchanger 1 0–200 25.6 —
Heat exchanger 2 0–200 43.2 —
hot and cold tasks leading to supplanting most of the external utility requirement
(160 GJ hot utility required by energy-integrated batch plant vs. 271.9 GJ hot utility
required by the basic design without energy integration). Consequently, a saving of
41.1% in hot utility requirement is achieved. This indicates that solving a design and
synthesis problem for batch plants by considering scheduling and heat integration
results not only in efficient use of equipment resources for maximum production of
products but also rational use of energy for sustainable operation. It is worth men-
tioning that this comprehensive model also determines the optimal heat exchanger
area required in order to allow the heat transfer. An optimal area of 25.6 m2 for heat
exchanger 1 and 43.2 m2 for heat exchanger 2 was obtained. Figure 6.2 details the
possible amount of energy integration between the cold and hot units and the time
intervals during which energy integration occurred. The energy requirements of
reactor 1 and reactor 2 during the interval 4–6 h are highlighted to elaborate on the
application of the proposed model. The cooling load of reactor 2 between 5 and 6 h
was 25 GJ. This cooling requirement is fully satisfied through energy integration
with reactor 1 in the same time interval, resulting in supplanting an external cool-
ing requirement if operated in standalone mode. At the beginning of the operation
of reactor 1 from 4 to 5 h, the heating requirement was 25.8 GJ. This value was
obtained using the time average model by multiplying the duration (5 – 4 h) and the
energy demand per hour (51.6 GJ/2 h (total duration of the task) = 25.8 GJ), where
the heating requirement is fully satisfied by external heating. For the rest of its
operation between 5 and 6 h, the heating requirement was 25.8 GJ, satisfied partly
with energy integration (25 GJ) and the difference by external heating with energy
requirement of 0.8 GJ. The amount of material each unit is processing, the type of
task each unit is conducting at a specific time point and the starting and finishing
times for each task are also presented in Figure 6.3.
162
TABLE 6.3
Computational Results for the Case Study
Time Continuous
Method Point Objective Value Binary Variables Variables Constraints CPU Time (s) Cold Utility Hot Utility

Design without 11 2.08 * 10 8 176 793 2079 19 220.4 GJ 271.9 GJ
energy integration
Design with energy 11 2.6 * 108 222 887 2367 198 34.4 GJ 160 GJ
integration
Units
20 20 20 20 20 20 20
SR (20)
RX1 RX3 RX3 RX2 RX3 RX3 RX3 RX3 RX3
50 35.8 24.2 39.9 43.3 28.5 50 26.2 50
R2 (50)
60 GJ 47.8 GJ 45.7 GJ 35 GJ
25 GJ 16.9 GJ 30.3 GJ 31.5 GJ 35 GJ
R1 (47.7) 28.6 47.7 32.3 21.2 29.9 37.9 38 43.7 19.7 43.7
RX1 RX2 RX2 RX2 RX2 RX2 RX1 RX2 RX2 RX2
HR (19) 19 13 8.5 19 8.1 19 11.4 17.5 6.4
2 4 6 8 10 12 14 16 18 20 22 24
Time (h)
FIGURE 6.3 Gantt chart for the case study.
6.7 CONCLUSIONS
An efficient method for the design and synthesis of batch plants that incorporate
energy integration was presented. The developed model simultaneously solved
scheduling and energy integration within the same design framework leading to
optimal design of equipment and rational use of energy. From the case study, the
profit maximization was increased by 20% in the energy-integrated batch plant
compared to the basic design where utilities were the only option to satisfy the
heating and cooling load requirements demanded by the tasks. The total hot utility
requirement was reduced by 41.1% for the energy-integrated batch plant compared
to the basic design. From this, it can be concluded that for efficient design of batch
plants, energy integration must be incorporated in the design problem.
NOMENCLATURE
Sets
J { j| j is a piece of equipment including heat exchanger}
J ssc { jssc| jssc is a piece of equipment that consumes state s}
J ssp { jssp | jssp is a piece of equipment that produces state s}
V {v|v is a storage}
P {p|p is a time point}
Sinsc, J {sinsc, j|sinsc, j is a task which consumes state s}
Sin*, j {s*in, j|s*in, j is a task performed in unit j}
Sin, J {sin, j|sin, j is an effective state representing a task}
usc usc
Sinusc, J {sin, j|sin, j is a task which consumes unstable state s}
sp sp
Sinsp, J {sin, j|sin, j is a task which produces state s other than a product}
usp usp
Sinusp, J {sin, j|sin, j is a task which produces unstable state s}
S {s|s is any state}
S p {s|s is any state which is a product}
Srm {s|s is any state which is a raw material}
p p p
Sin, J {sin, j|sin, j task which produce state s which is a product}
I c {ic|ic is a task requires heating}, I h = {ih|ih is a task requires cooling}
Continuous Variables
tu ( sin, j , p ) Time at which a task in unit j starts
t p ( sin, j , p ) Time at which a task in unit j finishes
ss ( j ) Design capacity of unit j
s ( v ) Design capacity of storage unit v
mu ( sin, j , p ) Amount of material processed by a task
muu ( s, j, p ) Amount of state s transferred between unit j and storage unit at
time point p
mux ( s, j, j¢, p ) Amount of state s transferred between unit j and another unit j′ at
time point p
qs ( s, p ) Amount of state s stored at time point p in storage unit
pip ( j, v ) Design capacity of a pipe that connects unit j and storage unit v
pipj ( j, j¢ ) Design capacity of a pipe that connects unit j and another unit j'
cw ( ih , p ) External cooling required by task ih at time point p
st ( ic , p ) External heating required by task ic at time point p
w ( ih , ic , p ) Amount of heat exchanged between the cold and hot task
A Optimal heat exchanger area required
mu ( ih , p ) Amount of material processed by the hot task
mu ( ic , p ) Amount of material processed by the cold task
Binary Variables
y ( sin, j , p ) 1 if state s is used in unit j at time point p; 0 otherwise
e ( j ) 1 if unit j is selected; 0 otherwise
eu ( v ) 1 if storage unit v is selected; 0 otherwise
z ( j , v ) 1 if pipe that connects unit j and storage unit v is selected; 0 otherwise
w ( j, j¢ ) 1 if pipe that connects unit j and j' is selected; 0 otherwise
t ( j, p ) 1 if the state produced by unit j at time point p is consumed; 0 otherwise
x ( ih , ic , p ) 1 if hot task ih is integrated with the cold task ic at time point p;
0 otherwise
Parameters
M Any large number
t ( sin, j ) Duration of a task conducted in unit j
b ( sin, j ) Coefficient of variable term of processing time of a task
V jL Lower bound for unit j
V jU Upper bound for unit j
VvL Lower bound for storage unit v
VvU Upper bound for storage unit v
g sL Minimum percentage equipment utilization for a task
in , j
gUs Maximum percentage equipment utilization for a task

in , j
QO ( s ) Initial amount of state s stored in unit
( )
r sinsp, j Portion of state s produced by a task
r ( s ) sc
Portion of state s consumed by a task
in, j
L
V Lower bound for the capacity of the pipe connecting unit j and storage
j ,v
unit v
V jU,v Upper bound for the capacity of the pipe connecting unit j and storage
unit v
V jL, j ¢ Lower bound for the capacity of the pipe connecting unit j and another
unit jʹ
V jU, j ¢ Upper bound for the capacity of the pipe connecting unit j and another
unit jʹ
( )
price s p Selling price for a product
AWH Annual working hour
FOC Fixed operating cost for a task
VOC Variable operating cost for a task
CCF Capital charge factor
FEC Fixed capital cost of equipment
VEQ Variable capital cost of equipment
FECS Fixed capital cost of storage
VEQS Variable capital cost of storage
FCNC Fixed capital cost for pipe connection between processing units
VCNC Variable capital cost for pipe connection between processing units
CNC Fixed capital cost for pipe connection between processing unit and
storage
VCN Variable capital cost for pipe connection between processing unit and
storage
C p Specific heat capacity of fluid
E ( i ) Constant coefficient of amount of heat required or removed correspond-
ing to task i
x ( i ) Variable coefficient of amount of heat required or removed correspond-

ing to task i
T ( i ) Operating temperature for a task
DT min
Minimum allowable thermal driving force
Cc Cost unit for cooling
Ch Cost unit for steam
U Overall heat transfer coefficient
REFERENCES
Barbosa-Povoa, A.P., Macchietto, S., 1994. Detailed design of multipurpose batch plants.
Computers and Chemical Engineering. 18(11/12), 1013–1042.
Barbosa-Povoa, A.P., Mateus, R., Novais, A.Q., 2001a. Optimal 2D design layout of industrial
facilities. International Journal of Production Research. 39(12), 2567–2593.
Barbosa-Povoa, A.P., Mateus, R., Novais, A.Q., 2002. Optimal design and layout of indus-
trial facilities: An application to multipurpose batch plants. Industrial and Engineering
Chemistry Research. 41(15), 3610–3620.
Barbosa-Povoa, A.P., Pantelides, C.C., 1997. Design of multipurpose plants using the
resource-task network unified framework. Computers and Chemical Engineering. 21,
S703–S708.
Barbosa-Povoa, A.P., Pinto, T., Novais, A.Q., 2001b. Optimal design of heat-integrated mul-
tipurpose batch facilities: A mixed integer mathematical formulation, Computers and
Barbosa-Povoa, A.P.F.D., 2007. A critical review on the design and retrofit of batch plants.
Computers and Chemical Engineering. 31(7), 833–855.
Castro, P.M., Barbosa-Povoa, A.P., Novais, A.Q., 2005. Simultaneous design and scheduling
of multipurpose plants using resource task network based continuous-time formula-
tions. Industrial and Engineering Chemistry Research. 44(2), 343–357.
Cerda, J., Vicente, M., Gutierrez, J.M., Esplugas, S., Mata, J., 1989. A new methodology for
the optimal-design and production schedule of multipurpose batch plants. Industrial
and Engineering Chemistry Research. 28(7), 988–998.
Faqir, N.M., Karimi, I.A., 1989a. Optimal-design of batch plants with single production
routes. Industrial and Engineering Chemistry Research. 1989, 28(8), 1191–1202.
Faqir, N.M., Karimi, L.A., 1989b. Design of multipurpose batch plants with multiple pro-
duction routes. Conference on Foundations of Computer Aided Process Design,
Snowmass, CO, 451–468.
Fuchino, T., Muraki, M., Hayakawa, T., 1994. Scheduling method in design of multipurpose
batch plants with constrained resources. Journal of Chemical Engineering of Japan.
27(3), 363–368.
Georgiadis, M.C., Rotstein, G.E., Macchietto, S., 1997. Optimal layout design in multipur-
pose batch plants. Industrial and Engineering Chemistry Research. 36(11), 4852–4863.
Grossman, I.E., Sargent, R.W.H., 1979. Optimum design of multipurpose chemical plants.
Industrial and Engineering Chemistry Process Design and Development. 18(2),
343–348.
Heo, S.K., Lee, K.H., Lee, H.K., Lee, I.B., Park, J.H., 2003. A new algorithm for cyclic sched-
uling and design of multipurpose batch plants. Industrial and Engineering Chemistry
Research. 42(4), 836–846.
scheduling: 1 Multipurpose batch processes. Industrial and Engineering Chemistry
Research. 37, 4341–4359.
Imai, M., Nishida, N., 1984. New procedure generating suboptimal configurations to the
optimal-design of multipurpose batch plants. Industrial and Engineering Chemistry
Process Design and Development. 23(4), 845–847.
Klossner, J., Rippin, D.W.T., 1984. Combinatorial problems in the design of multiproduct
batch plants—Extension to multiplant and partly parallel operations. In AIChE Annual
Meeting, San Francisco, CA.
17, 211–227.
Lin, X., Floudas, C.A., 2001. Design, synthesis and scheduling of multipurpose batch plants
via an effective continuous-time formulation. Computers and Chemical Engineering.
25665–25674.
Pantelides, C.C., 1994. Unified frameworks for the optimal process planning and schedul-
ing. Foundations on Computer Aided Process Operations, CACHE, Austin, TX,
pp. 253–274.
Papageorgaki, S., Reklaitis, G.V., 1990. Optimal design of multipurpose batch plants. 1.
Problem formulation. Industrial and Engineering Chemistry Research. 29(10),
2054–2062.
Patziatsis, D.I., Xu, G., Papageorgiou, L.G., 2005. Layout aspects of pipeless batch plants.
Industrial and Engineering Chemistry Research. 44(15), 5672–5679.
Penteado, F.D., Ciric, A.R., 1996. An MINLP approach for safe process plant layout.
Pinto, T., Barbosa-Povoa, A.P., Novais, A.Q., 2003. Optimal design of heat-integrated mul-
tipurpose batch facilities: A mixed integer mathematical formulation. Computers and
Pinto, T., Barbosa-Povoa, A.P., Novais, A.Q., 2008. Design of multipurpose batch plants:
A comparative analysis between the STN, m-STN, and RTN representations and
formulations. Industrial and Engineering Chemistry Research. 47, 6025–6044.
Seid, E.R., Majozi, T., 2012. A robust mathematical formulation for multipurpose batch
plants. Chemical Engineering Science. 68, 36–53.
Shah, N., Pantelides, C.C., 1991. Optimal long-term campaign planning and design of batch-
operations. Industrial and Engineering Chemistry Research. 30(10), 2308–2321.
Suhami, I., Mah, R.S.H., 1982. Optimal design of multipurpose batch plants. Industrial and
Engineering Chemistry Process Design and Development. 21(1), 94–100.
Vaselenak, J.A., Grossmann, I.E., Westerberg, A.W., 1987. An embedding formulation
for the optimal scheduling and design of multipurpose batch plants. Industrial and
Engineering Chemistry Research. 26(1), 139–148.
Voudouris, V.T., Grossmann, I.E., 1996. MILP model for scheduling and design of a spe-
cial class of multipurpose batch plants. Computers and Chemical Engineering. 20(11),
1335–1360.
Xia, Q.S., Macchietto, S., 1997. Design and synthesis of batch plants MINLP solution based
on a stochastic method. Computers and Chemical Engineering. 21, S697–S702.
7 Simultaneous Scheduling
and Water Optimization
Reduction of Effluent
in Batch Facilities
7.1 INTRODUCTION
Wastewater minimization can be achieved by employing water reuse opportunities.
This chapter presents a methodology to address the problem of wastewater minimi-
zation by extending the concept of water reuse to include a wastewater regenerator.
The regenerator purifies wastewater to such a quality that it can be reused in other
operations. This further increases water reuse opportunities in the plant, thereby
significantly reducing freshwater demand and effluent generation. The mathematical
model determines the optimum batch production schedule that achieves the mini-
mum wastewater generation within the same framework. The model was applied
to two case studies involving multiple contaminants, and wastewater reductions of
19.2% and 26% were achieved.
7.2 MOTIVATION FOR THE STUDY

The study carried out in this chapter was motivated by the following reasons. Firstly,
earlier work by Majozi and Gouws (2009) did not consider the effect of regeneration
while addressing wastewater minimization in the presence of multiple contaminants.
This has been identified as a limitation as regeneration will improve water reuse
opportunities in a plant.
Secondly, most published works in literature directly or indirectly avoid treating
time as a variable. The results obtained from such formulations could very well be
suboptimal. An example is the second case study investigated by Liu et al. (2009)
which considered wastewater minimization in a single contaminant problem by
incorporating regeneration. The case study was based on a given fixed schedule.
Figure 7.1a and b highlights the impact of treating time as a variable during the first
cycle of operation.
From Figure 7.1, when time is treated as a variable, the freshwater usage is 64.1
tons compared to 68.594 tons obtained by Liu et al. (2009) using a predefined
schedule. This amounts to water savings of 6.6%. This improvement was brought
about by the flexibility of the schedule. Apart from process A which occurs within
the same time interval in both figures, the order of the other processes differs.
169
Tank T
11.875 tons/h
50 tons R
11.875 tons/h
Tank S
50 tons 24.781 tons

A
15.659 tons B
9.122 tons
10 tons
4.312 tons 0.349 tons
2.935 tons C
2.753 tons
15.156 tons 21.094 tons
5.938 tons D
16.667 tons E 16.316 tons

68.594 tons
Effluent
Freshwater
0 1 2 3 4 5 6 7 8 9
(a) Time (h)
Central storage tank

19.5 tons 30 tons
49.5 tons R
19.5 tons 30 tons
Freg = 11.875 tons/h
50 tons A
B
0.5 tons 7.5 tons

C
12 tons
D
14.07 tons 64.07 tons
64.07 tons E
Effluent
Freshwater
0 1 2 3 4 5 6 7 8 9
(b) Time (h)
FIGURE 7.1 (a) Schedule obtained from the case study by Liu et al. (2009). (b) Schedule
obtained when time is treated as a variable.
For example, process E occurs at 2 h in Figure 7.1b as opposed to 6 h in Figure 7.1a,

thus providing an opportunity for the direct reuse of water from process A. In addi-
tion, process B occurs at 7.5 h which allows for the use of only regenerated water as
opposed to 3 h in Figure 7.1a whereby an insufficient amount of regenerated water
is available. It is important to note, however, that in certain situations the order of
Simultaneous Scheduling and Water Optimization 171
processes is specified beforehand due to the given recipe. In such situations, a fixed/
predefined schedule is unavoidable.
The formulation presented in this chapter incorporates regeneration such that the
batch production schedule and water network that correspond to minimum water
requirement are determined simultaneously within the same optimization frame-
work as opposed to a two-step procedure which will inherently rely on a given sched-
ule for wastewater minimization. The proposed method is especially significant in
multipurpose batch facilities where units must be washed before subsequent usage.
Without proper integration between production scheduling and water management,
useful water may be discharged as effluent instead of being reused.
The problem addressed in this chapter can be stated as follows:
Given:
1. The production recipe for each product, including mean processing times in
each unit operation
2. The available units and their capacities
3. The necessary costs and stoichiometric data
4. The contaminant mass load of each contaminant
5. Water requirement and the cleaning duration for each unit to achieve the
required cleanliness
6. Maximum inlet and outlet concentrations of each contaminant
7. The maximum storage available for water reuse
8. The performance of the regenerator
9. The time horizon of interest
Determine:
The production schedule that achieves the minimum amount of wastewater
generated by exploring recycle and reuse opportunities in the presence of a
central storage vessel and a wastewater regenerator.
7.4 PROBLEM SUPERSTRUCTURE
The problem superstructure on which the mathematical formulation is based is
depicted in Figure 7.2. In the figure, only the water using operations which are part
of a complete batch process are depicted. Unit j represents a water using operation
in which the water used consists of freshwater, regenerated water, stored water or
recycle/reuse water. Water from unit j can be recycled into the same unit, reused
elsewhere or sent to storage. The storage vessel can send water to a unit or to the
wastewater regenerator. The philosophy of operation of the regenerator is as follows:
1. Water is sent to the regenerator provided there exists an available unit requir-
ing that water. Whenever the regenerator is active, it operates in a continuous
mode with steady inlet and outlet streams until the sink process is satisfied.
mwf (sout,j, p) mwe (sout,j, p)

Freshwater Effluent
mwr (sout,j΄, sout,j, p) Water in Water out
Reuse sout,j΄ mwr (sout,j, sout,j΄, p)
mwin (sout,j, p) mwout (sout,j, p) sout,j΄
j Reuse
Regenerated mregout (sout,j, p) c (s , c, p) cout (sout,j, c, p)
in out,j
water
msout (sout,j, p) msin (sout,j, p)
From storage To storage
(a)
Stored water
Freshwater
Effluent
Solid waste
j Storage Regenerator
Reuse
j΄
Reuse
Regenerated water
(b)
FIGURE 7.2 Superstructure for the mathematical formulation for processing unit (a) and
storage (b).
2. Transfer of water to regenerator occurs to purify contaminated water

because the contaminant concentration of water in storage is too high to be
utilized in a unit.
3. The regenerator operates intermittently in the sense that it is only active if
there is a unit that requires water; otherwise it remains in the passive mode.
4. The regenerator performance is given in terms of a constant removal ratio
of contaminants.
The mathematical formulation involves the following sets, variables and parameters.
7.5.1 Sets
P {p|p = time point}
J { j| j = unit}
C {c|c = contaminant}
Sin {sin |sin = input state into any unit}
Sout {sout |sout = output state from any unit}
S*in, j {s*in, j|s*in, j = input state into unit} Í Sin
Sout, j {sout, j |sout, j = output state from unit j} Í Sout
7.5.2 Variables
mwin ( sout , j , p ) Mass of water into unit j for cleaning state sout at time point p
mwout ( sout , j , p ) Mass of water produced at time point p from unit j
mw f ( sout , j , p ) Mass of freshwater into unit j at time point p
mwe ( sout , j , p ) Mass of effluent water from unit j at time point p
mwr ( sout , j , sout , j ¢ , p ) Mass of water recycled to unit j¢ from j at time point p
msin ( sout , j , p ) Mass of water transferred from unit j to storage at time point p
msout ( sout , j , p ) Mass of water transferred from storage to unit j at time point p
mregin ( sout , j , p ) Mass of water transferred from storage to regenerator at time
point p
mregout ( sout , j , p ) Mass of water transferred from regenerator to unit j at time
point p
mdirt ( c, p ) Mass of contaminant c removed from water by regenerator at
time point p amount of state delivered to customers at time
point p
cin ( sout , j , c, p ) Inlet concentration of contaminant c, to unit j at time point p
cout ( sout , j , c, p ) Outlet concentration of contaminant c, from unit j at time
point p
csin ( c, p ) Inlet concentration of contaminant c, to storage at time point p
csout ( c, p ) Outlet concentration of contaminant c, from storage at time
point p
crout ( c, p ) Outlet concentration of contaminant c, from regenerator at
time point p
qws ( p ) Amount of water stored in storage at time point p
tout ( sout , j , p ) Time at which a state is produced from unit j at time point p
twin ( sout , j , p ) Time that the water is used at time point p in unit j
twout ( sout , j , p ) Time at which water is produced at time point p from unit j
twr ( sout , j , sout , j ¢, p ) Time at which water is recycled from unit j to unit j¢ at time
point p
tsin ( sout , j , p ) Time at which water is transferred from unit j to storage at
time point p
tsout ( sout , j , p ) Time at which water is transferred from storage to unit j at
time point p
tregin ( sout , j , p ) Time at which water is transferred from storage to regenerator
at time point p
tregout ( sout , j , p ) Time at which water is transferred from regenerator to unit j at
time point p
( *
)
y s in, j , p

Binary variable associated with usage of state s at time point p
ywr ( sout , j , sout , j ¢ , p ) Binary variable showing usage of recycle from unit j to unit j¢
at time point p
yw ( sout , j , p ) Binary variable showing usage of water in unit j at time point p
ysin ( sout , j , p ) Binary variable showing transfer of water from unit j to stor-
age at time point p
ysout ( sout , j , p ) Binary variable showing transfer of water from storage to unit
j at time point p
yregin ( sout , j , p ) Binary variable showing transfer of water from storage to
regenerator at time point p
yregout ( sout , j , p ) Binary variable showing transfer of water from regenerator to
unit j at time point p
7.5.3 Parameters
CE Cost of effluent water treatment (c.u./kg water)
CF Cost of freshwater (c.u./kg water)
CP ( s ) Selling price of product s, s = product
M ( sout , j , c ) Mass load of contaminant c added from unit j to the water stream
MwU ( sout , j ) Maximum inlet water mass of unit j
C U
( sout , j , c )
in Maximum inlet concentration of contaminant c in unit j
C ( sout , j , c )
U
out Maximum outlet concentration of contaminant c from unit j
U
CSout ( c ) Maximum outlet concentration of contaminant c from storage
CRout ( c )
U
Maximum outlet concentration of contaminant c from regenerator
o
CSout ( c ) Initial concentration of contaminant in storage
RR ( c ) Removal ratio of contaminant c from regenerator
Qwso Initial amount of water in storage
QwUs Maximum capacity of storage
tw ( sout , j ) Mean processing time of unit j
freg Regenerator flowrate
In the mathematical formulation, wastewater minimization is embedded in the pro-

duction scheduling framework. The production scheduling framework adopted here,
which is based on the state sequence network (SSN) and the uneven discretization of
the time horizon, has been explained in detail in Majozi and Gouws (2009). It will
not be discussed in this chapter. However, critical constraints have been provided
again to facilitate understanding.
7.5.4 Mass Balance Constraints

It is necessary to define the mass balances around each processing unit: the central
storage vessel and the regenerator.
7.5.4.1 Mass Balance around a Unit j

Constraint (7.1) is the water balance over the inlet to a unit. Water entering the unit
is a combination of reuse/recycle streams from other units j¢ , freshwater, water from
storage and water from the regenerator. Constraint (7.2) states that the water leaving
a unit could be recycled/reused, discarded as effluent or sent to storage. Constraint
(7.3) states that the amount of water exiting a unit must equal the amount of water
entering the unit at the previous time point. This constraint captures the fact that
water is not produced in the unit during the washing operation.
mwin ( sout , j , p ) = åmwr ( s

sout , j¢
out , j ¢ , sout , j , p ) + mw f ( sout , j , p ) + msout ( sout , j , p )
+ mregout ( sout , j , p ) "j, j¢ Î J , sout, j Î Sout , j , pÎP

(7.1)
mwout ( sout , j , p ) = åmwr ( s

sout , j¢
out , j , sout , j ¢ , p ) + mwe ( sout , j , p ) + msin ( sout , j , p )
"j, j¢ Î J , sout , j Î Sout , j , pÎP (7.2)

mwin ( sout , j , p - 1) = mwout ( sout , j , p ) "j Î J , sout , j Î Sout , j , p Î P, p > p1 (7.3)

Constraint (7.4) defines the inlet contaminant mass balance. The contaminant mass
load in the inlet stream consists of contaminant mass load in recycle/reuse water,
the contaminant mass load in water from storage and the contaminant mass load in
water from the regenerator. Constraint (7.5) defines the outlet contaminant mass as
the mass of contaminant that entered the unit at the previous time point and the mass
load of contaminant picked up in the unit during its operation.
mwin ( sout , j , p ) cin ( sout , j , c, p ) = åmw ( s

sout , j¢
r out , j ¢ , sout , j , p ) cout ( sout , j , c, p )
+ msout ( sout , j , p ) csout ( c, p )
+ mregout ( sout , j , p ) crout ( c, p )
"j, j¢ Î J , sout , j Î Sout , j , p Î P, c Î C (7.4)
mwout ( sout , j , p ) cout ( sout , j , c, p ) = M ( sout, j , c ) yw ( sout , j , p - 1)
+ mwin ( sout , j , p - 1) cin ( sout , j , c, p - 1)
"j Î J , sout , j Î Sout , j , p Î P, p > p1, c Î C

(7.5)
Constraints (7.6) and (7.7) ensure that the inlet and outlet contaminant concentra-
tions do not exceed the allowed maximum. Similarly, the maximum allowable water
in a unit must not be exceeded. This is governed by Constraint (7.8). Constraints (7.9)
through (7.11) restrict the mass of water entering the unit from the various sources to
the maximum allowable for the unit. Constraint (7.9) is with respect to recycle/reuse,
Constraint (7.10) is with respect to water from storage, while Constraint (7.11) is with
respect to regenerated water.
cin ( sout , j , c, p ) £ CinU ( sout , j , c ) yw ( sout , j , p ) "j Î J , sout , j Î Sout , j , p Î P, c Î C

(7.6)
cout ( sout , j , c, p ) £ Cout
U
( sout , j , c ) yw ( sout , j , p - 1)
"j Î J , sout , j Î Sout , j , p Î P, p > p1, c Î C (7.7)

mwin ( sout , j , p ) £ MwU ( sout, j ) yw ( sout , j , p ) "j Î J , sout , j Î Sout , j , pÎP (7.8)

mwr ( sout , j ¢ , sout , j , p ) £ MwU ( sout , j ) ywr ( sout , j ¢ , sout , j , p )
msout ( sout , j , p ) £ MwU ( sout , j ) ysout ( sout , j , p ) "j Î J , sout , j Î Sout , j , p Î P (7.10)

mregout ( sout , j , p ) £ MwU ( sout , j ) yregout ( sout , j , p ) "j Î J , sout , j Î Sout , j , pÎP

(7.11)
The maximum quantity of water into a unit is represented by Equation 7.12. It is

important to note that for multi-contaminant wastewater the outlet concentration
of the individual components cannot all be set to the maximum, since the contami-
nants are not limiting simultaneously. The limiting contaminant(s) will always be at
the maximum outlet concentration and the non-limiting contaminants will be below
their respective maximum outlet concentrations.
ìï M ( sout , j , c ) üï
MwU ( sout , j ) = max í U ý , "j Î J , sout , j Î Sout , j (7.12)
îï Cout ( sout , j , c ) - Cin ( sout , j , c ) þï
cÎC U

7.5.4.2 Mass Balance around Central Storage

Constraint (7.13) is the water mass balance around the storage tank. The amount of
water stored in the storage tank consists of water stored at the previous time point
and the difference between water entering the storage tank from processing unit and
water leaving the storage tank either to processing unit or regenerator. Constraint
(7.14) defines the initial amount of water in the tank.
qws ( p ) = qws ( p - 1) + å ms
sout , j
in ( sout , j , p ) - å msout ( sout , j , p )
sout , j
- å mreg
sout , j
in ( sout , j , p ) "j Î J , sout , j Î Sout , j , p Î P, p > p1 (7.13)
qws ( p1 ) = Qwso - å ms
sout , j
out ( sout , j , p1 ) - å mregin ( sout , j , p1 )
sout , j
"j Î J , sout , j Î Sout , j (7.14)

The definition of the inlet contaminant concentration to the storage tank is given in
Constraint (7.15). The concentration of water exiting the storage tank is assumed to
be equal to the concentration of water in the tank as given in Constraint (7.16). This
condition is true in the case of perfect mixing. The initial concentration in the stor-
age tank is given in Constraint (7.17). Constraints (7.18) and (7.19) ensure that the
maximum capacity of the tank is not exceeded.
å ( ms in ( sout , j , p ) cout ( sout , j , c p ) )

csin ( c, p ) =
sout , j
"j Î J , sout , j Î Sout , j , p Î P c ÎC
å msin ( sout , j , p )
sout , j
(7.15)
æ ö
qws ( p - 1) csout ( c, p - 1) + ç
çs å
msin ( sout , j , p ) ÷ csin ( c, p )
÷
csout ( c, p ) = è out , j ø
qws ( p - 1) + msin ( sout , j , p )å
sout , j
csout ( c, p1 ) = CSout
o
( c ) "c Î C (7.17)
qws ( p ) £ QwUs "p Î P (7.18)
msin ( sout , j , p ) £ QwUs s ysin ( sout , j , p ) "j Î J , sout , j Î Sout , j , pÎP (7.19)

Constraint (7.20) ensures that no water is stored in the storage vessel at the end of the
time horizon in order to give a true optimum. Otherwise, the minimum target could
be misleading.
qws ( p ) = 0 "p = P (7.20)
7.5.4.3 Mass Balance around the Regenerator

Constraint (7.21) states that the total amount of outlet water from the regenerator
at time point p consists of the total amount of inlet water to the regenerator at the
previous time point. This constraint captures the assumption that during the opera-
tion of the regenerator, no water is produced or lost.
å mreg
sout , j
in ( sout , j , p - 1) = å mregout ( sout , j , p )
sout , j
"j Î J , sout , j Î Sout , j , p Î P, p > p1 (7.21)
The performance of the regenerator is given in terms of the removal ratio of

contaminant, which specifies a constant ratio between the inlet and outlet mass of
contaminant. This is stated in Constraint (7.22) provided the flowrate of the regen-
erator is constant.

crout ( c, p ) = csout ( c, p - 1) 1 - RR ( c ) ( ) "c Î C , p Î P, p > p1

(7.22)
The contaminant mass balance around the regenerator is expressed in Constraint

(7.23). The contaminant mass load exiting the regenerator to a unit is the difference
between the contaminant mass load entering the regenerator and the contaminant
mass removed from the water during regeneration, mdirt(c, p).
csout ( c, p - 1) å mreg
sout , j
in ( sout , j , p - 1) = crout ( c, p ) å mregout ( sout , j , p ) + mdirt ( c, p )
sout , j

7.5.5 Scheduling Constraints
The scheduling constraints take care of the time dimension associated with batch
processes. The scheduling aspects of the model are divided into four groups as
follows:
1. Those related to task scheduling

2. Those that cater for recycle/reuse
3. Those that account for the central storage vessel
4. Those that account for the regenerator
7.5.5.1 Task Scheduling
These constraints ensure that each water-using operation is integrated with produc-
tion scheduling. Constraints (7.24) and (7.25) together ensure that unit j is washed
immediately after a task that produced sout,j. Constraints assuming the form of
Constraints (7.24) or (7.25) are called big-M constraints. If water is used in the unit,
yw ( sout , j , p ) has a value of 1 causing Constraints (7.24) and (7.25) to become active,
and the start time of washing is forced to coincide with the end time of production.
Otherwise, when water is not used in the unit, that is yw ( sout , j , p ) has a value of zero,
the two constraints become relaxed. Constraint (7.26) defines the duration of the
washing operation performed in unit j. Constraint (7.27) stipulates that the washing
operation can only commence at time point p if the task producing sout,j was active at
the previous time point.

(
twin ( sout , j , p ) ³ tout ( sout , j , p ) - H 1 - yw ( sout , j , p ) ) "j Î J , sout , j Î Sout , j , pÎP

(7.24)

(
twin ( sout , j , p ) £ tout ( sout , j , p ) + H 1 - yw ( sout , j , p ) ) "j Î J , sout , j Î Sout , j , pÎP

(7.25)
twout ( sout , j , p ) = twin ( sout , j , p - 1) + tw ( sout , j ) yw ( sout , j , p - 1)
(
yw ( sout , j , p ) = y s*in, j , p - 1 )
"j Î J , sout , j Î Sout , j , s*in, j Î Sin*, j ® sout , j , p Î P, p > p1 (7.27)
7.5.5.2 Recycle/Reuse Scheduling
Wastewater can only be directly recycled/reused if the unit producing wastewater
and the unit receiving wastewater finish operating and begin operating at the same
time, respectively. Constraint (7.28) describes the relationship between usage of
water in a unit and the opportunity for recycle and reuse. The constraint states
that for a unit j to transfer water to unit j¢ , unit j¢ should require water at that time
point. It does not, however, mean that unit j¢ must use water from unit j; it could
still obtain water from other sources. Constraints (7.29) and (7.30) state that the
time at which water recycle/reuse takes place coincides with the time at which
the water is produced. Constraints (7.31) and (7.32) ensure that the time at which
water recycle/reuse takes place coincides with the starting time of the unit receiv-
ing the water.
ywr ( sout , j , sout , j ¢ , p ) £ yw ( sout , j ¢ , p ) "j, j¢ Î J , sout , j Î Sout , j , pÎP (7.28)

(
twr ( sout , j , sout , j ¢ , p ) £ twout ( sout , j , p ) - H 1 - ywr ( sout , j , sout , j ¢ , p ) )
(
twr ( sout , j , sout , j ¢ , p ) ³ twout ( sout , j , p ) - H 1 - ywr ( sout , j , sout , j ¢ , p ) )
(
twr ( sout , j , sout , j ¢ , p ) £ twin ( sout , j , p ) + H 1 - ywr ( sout , j , sout , j ¢ , p ) )
(
twr ( sout , j , sout , j ¢ , p ) ³ twin ( sout , j , p ) - H 1 - ywr ( sout , j , sout , j ¢ , p ) )
7.5.5.3 Central Storage Scheduling

Constraint (7.33) relates water usage in a unit and water transfer from storage. It
states that water can only be transferred to a unit if it uses water at the same time
point. However, it is not a prerequisite for the unit to use stored water, the water could
be provided from other sources. Constraints (7.34) and (7.35) ensure that the time at
which water is sent from storage to a unit coincides with the start time of washing
of the same unit.
ysout ( sout , j , p ) £ yw ( sout , j , p ) "j Î J , sout , j Î Sout , j , pÎP (7.33)

(
tsout ( sout , j , p ) ³ twin ( sout , j , p ) - H 2 - ysout ( sout , j , p ) - yw ( sout , j , p ) )
"j Î J , sout , j Î Sout , j , pÎP (7.34)
(
tsout ( sout , j , p ) £ twin ( sout , j , p ) + H 2 - ysout ( sout , j , p ) - yw ( sout , j , p ) )
Constraint (7.36) relates water usage in a unit and water transfer to storage. It states
that water can only be transferred from a unit to storage if the unit used water at
the previous time point. However, washing can take place in the unit without dis-
charging water to the storage tank. The water could be discharged to other sinks.
Constraints (7.37) and (7.38) ensure that the time at which water is sent to storage
from a unit must coincide with the finishing time of washing of the same unit.
ysin ( sout , j , p ) £ yw ( sout , j , p - 1) "j Î J , sout , j Î Sout , j , p Î P, p > p1 (7.36)

(
tsin ( sout , j , p ) ³ twout ( sout , j , p ) - H 2 - ysout ( sout , j , p ) - yw ( sout , j , p - 1) )
(
tsin ( sout , j , p ) £ twout ( sout , j , p ) + H 2 - ysout ( sout , j , p ) - y ( sout , j , p - 1) )
If water is transferred from storage to a unit at a later time point, the time at which
this happens must correspond to a later time in the time horizon. This is specified
in Constraint (7.39). Constraint (7.40) ensures that if water is transferred from a unit
to storage at a later time point, the time at which this happens corresponds to a later
time in the time horizon.
(
tsout ( sout , j , p ) ³ tsout ( sout , j ¢ , p¢ ) - H 2 - ysout ( sout , j , p ) - ysout ( sout , j ¢ , p¢ ) )
"j, j¢ Î J , sout , j Î Sout , j , p, p¢ Î P, p ³ p¢ (7.39)
(
tsin ( sout , j , p ) ³ tsin ( sout , j ¢ , p¢ ) - H 2 - ysin ( sout , j , p ) - ysin ( sout , j ¢ , p¢ ) )
Constraints (7.41) and (7.42) state that if water is transferred to storage from more
than one unit at the same time point, the time at which they do so must coincide.
Constraints (7.43) and (7.44) state that if water is discharged from storage to more
than one unit at the same time point, the time at which the water is discharged must
coincide.
(
tsin ( sout , j , p ) ³ tsin ( sout , j ¢ , p ) - H 2 - ysin ( sout , j , p ) - ysin ( sout , j ¢ , p ) )
(
tsin ( sout , j , p ) £ tsin ( sout , j ¢ , p ) + H 2 - ysin ( sout , j , p ) - ysin ( sout , j ¢ , p ) )
(
tsout ( sout , j, p ) ³ tsout ( sout , j ¢, p ) - H 2 - ysout ( sout , j, p ) - ysout ( sout , j ¢, p ) )
(
tsout ( sout , j, p ) £ tsout ( sout , j ¢ , p ) + H 2 - ysout ( sout , j , p ) - ysout ( sout , j ¢, p ) )
If water is simultaneously being transferred to and discharged from storage, the time
at which this happens should coincide. This is given in Constraints (7.45) and (7.46)
(
tsin ( sout , j, p ) ³ tsout ( sout , j ¢, p ) - H 2 - ysin ( sout , j , p ) - ysout ( sout , j ¢, p ) )

(
tsin ( sout , j , p ) £ tsout ( sout , j ¢ , p ) - H 2 - ysin ( sout , j , p ) - ysout ( sout , j ¢ , p ) )
Constraint (7.47) ensures that if water leaves storage at a later time point compared
to water entering the storage, the time at which water leaves the storage must cor-
respond to a later time in the time horizon.
(
tsout ( sout , j , p ) ³ tsin ( sout , j ¢ , p¢ ) - H 2 - ysout ( sout , j , p ) - ysin ( sout , j ¢ , p¢ ) )
Worthy of mention is the fact that most of the constraints in this chapter were
presented in Majozi and Gouws (2009) but are given again here to facilitate
understanding.
7.5.5.4 Regenerator Scheduling
Constraint (7.48) defines the relationship between the transfer of water to the regen-
erator and the discharge of water from the regenerator. It should be re-emphasized
at this point that water is only transferred to the regenerator provided that there is a
unit available downstream requiring that water and concentration in the storage unit
must exceed the maximum allowed inlet concentration in the receiving unit. Hence,
the water transferred to the regenerator at any time is in terms of an available unit.
Constraints (7.49) and (7.50) ensure that the time at which water is discharged from
the regenerator to a unit coincides with the time at which the unit starts to use water.
Constraint (7.51) gives the duration of regeneration.
yregout ( sout , j , p ) = yregin ( sout , j , p - 1) "j Î J , sout , j Î Sout , j , p Î P, p > p1

(7.48)
(
tregout ( sout , j , p ) £ twin ( sout , j , p ) + H 2 - yregout ( sout , j , p ) - yw ( sout , j , p ) )
(
tregout ( sout , j , p ) ³ twin ( sout , j , p ) - H 2 - yregout ( sout , j , p ) - yw ( sout , j , p ) )
é
ê å mregin ( sout , j , p - 1) ù
ú
tregout ( sout, j , p ) = tregin ( sout, j , p - 1) + ê sout , j
ú yregin ( sout, j , p - 1)
êë freg úû

Constraint (7.52) ensures that storage does not supply a unit with water through the
regenerator and also supply water directly to the unit at the same time point. If the
unit is directly supplied with water from storage, the need to regenerate water no
longer exists. Constraint (7.53) is included to ensure that the full potential for waste-
water regeneration is realized. It states that a unit does not receive water from both
the storage tank and the regenerator at the same time point. This is to avoid unneces-
sary mixing.
yregin ( sout , j , p ) + ysout ( sout , j , p ) £ 1 "j Î J , sout , j Î Sout , j , pÎP (7.52)

yregout ( sout , j , p ) + ysout ( sout , j , p ) £ 1 "j Î J , sout , j Î Sout , j , pÎP (7.53)

Constraint (7.54) together with Constraint (7.48) defines the relationship between the
usage of water in a unit and the regeneration of water for the unit. Constraint (7.54)
states that for water to be charged to the regenerator from storage, the unit requir-
ing the water should begin operation at the next time point. This, however, does not
mean that the unit cannot begin operation without regenerated water, since water
required could be provided from other sources.
yw ( sout , j , p ) ³ yregin ( sout , j , p - 1) "j Î J , sout , j Î Sout , j , p Î P, p > p1 (7.54)

Constraint (7.55) ensures that if water is charged to the regenerator at a later time
point, the time at which it does so is at a later time in the time horizon. Constraint
(7.56) is similar, but it applies to the time at which water is discharged from the
regenerator.
(
tregin ( sout , j , p ) ³ tregin ( sout , j ¢ , p¢ ) + H 2 - yregin ( sout , j , p ) - yregin ( sout , j ¢ , p¢ ) )

(
tregout ( sout , j , p ) £ tregout ( sout , j ¢ , p¢ ) + H 2 - yregout ( sout , j , p ) - yregout ( sout , j ¢ , p¢ ) )

7.5.6 Feasibility and Time Horizon Constraints

Constraint (7.57) ensures that if a processing unit j is reusing water from unit j¢ at
time point p, then unit j¢ cannot reuse water from unit j at the same time point.
Constraint (7.58) ensures that at a certain time point both the charging of the regen-
erator and the discharging from the regenerator on behalf of a particular unit cannot
take place at the same time point.
ywr ( sout , j , sout , j ¢, p ) + ywr ( sout , j ¢, sout , j , p ) £ 1 "j, j¢ Î J , sout , j Î Sout , j , pÎP

(7.57)
yregin ( sout , j , p ) + yregout ( sout , j , p ) £ 1 "j Î J , sout , j Î Sout , j , p Î P (7.58)

Constraints (7.59) through (7.65) ensure that each event occurs within the time hori-
zon of time of interest.
twin ( sout , j , p ) £ H "j Î J , sout , j Î Sout , j , pÎP (7.59)

twout ( sout , j , p ) £ H "j Î J , sout , j Î Sout , j , pÎP (7.60)

twr ( sout , j , sout , j ¢ , p ) £ H "j, j¢ Î J , sout , j Î Sout , j , pÎP (7.61)

tsin ( sout , j , p ) £ H "j Î J , sout , j Î Sout , j , pÎP (7.62)

tsout ( sout , j , p ) £ H "j Î J , sout , j Î Sout , j , pÎP (7.63)

tregin ( sout , j , p ) £ H "j Î J , sout , j Î Sout, j , pÎP (7.64)

tregout ( sout , j , p ) £ H "j Î J , sout , j Î Sout , j , pÎP (7.65)

The objective of the formulation depends on whether the production is given or not.
In the situation where production is not given, the objective function is expressed as
Constraint (7.66). This objective here is the maximization of profit while taking into
account freshwater and effluent treatment costs.
max
ååCP ( s ) d ( s, p ) - CF ååmw ( s
s p sout , j p
f , p ) - CE
out , j ååmw ( s
sout , j p
e out , j , p ) (7.66)
Constraint (7.67) is the objective function when the production is given. It is the
minimization of effluent.
æ ö
min åå ç mwe ( sout , j , p ) ÷
çs j ÷
(7.67)
p è out , ø
Due to Constraints (7.4), (7.5), (7.15), (7.16) and (7.23) which involve bilinear terms,
the presented mathematical formulation is a mixed integer nonlinear program,
(MINLP). The method proposed by Majozi and Gouws (2009) was to solve the
MINLP using a two-step procedure in which the MINLP problem was linearized
and solved as a mixed integer linear program (MILP) to provide a starting point
for the exact MINLP problem. The method of linearization is discussed in the
Appendix.
7.6 CASE STUDIES
This section presents the application of this method to two problems: one from litera-
ture and the other from a practical production facility.
7.6.1 Case Study I
This case study is commonly known as BATCH 1 in literature. The SSN representa-
tion of the problem is illustrated in Figure 7.3. The BATCH 1 facility consists of a
heater which is used to heat Feed A before reaction. Three different chemical reac-
tions can be performed in two common reactors. A separator exists to purify product
from reaction 3. The data required for the production aspects are given in Table 7.1.
Similar to the work by Majozi and Gouws (2009), the philosophy is that after a
reaction has taken place in a reactor, the reactor must be washed before another reac-
tion can take place. This ensures that any product residue is removed to retain the
integrity of subsequent product. Data pertaining to the washing operations are given
in Table 7.2. In addition, the cost of freshwater is given as 2 c.u./kg water, while the
effluent treatment cost is 3 c.u./kg.
A central storage tank for wastewater is available with a capacity of 200 kg. The
flowrate of the regenerator is given to be 100 kg/h. The removal ratios of the various
contaminants are provided in Table 7.3. The objective function is the maximization
of profit as given in Constraint (7.66). The time horizon of interest is 10 h.
(s1)
Feed A
(s2)
Feed B Hot A (s5)
50%
(s6) 40% (s7)
60% 40%
Int BC Product 1
50%
(s3) (s9)
80% 10%
60%
Feed C Int AB Impure E
(s8) 90%
(s4) 20%
Product 2
(s10)
FIGURE 7.3 SSN for literature example.

TABLE 7.1
Production Data for Literature Example
Units Capacity Suitability Mean Processing Time (τ)
Heater 100 Heating 1
Reactor 1 50 Reaction 1,2,3 2,2,1
Still 200 Separation 1 for product 2, 2 for IntAB
States Storage Capacity Initial Amount Price
Feed A Unlimited Unlimited 0
Feed B Unlimited Unlimited 0
Feed C Unlimited Unlimited 0
Hot A 100 0 0
IntAB 200 0 0
IntBC 150 0 0
Impure E 200 0 0
Product 1 Unlimited 0 100

The resulting model for Case Study I was formulated in GAMS 22.0 and solved
using the DICOPT algorithm, with CPLEX 9.1.2 as the MIP solver and CONOPT3
as the NLP solver. The model had 532 binary variables. The optimum number of
time points was 10. The value of the objective function, the maximization of profit in
Constraint (7.67) was 21,129 c.u.
The optimum amount of freshwater used was 401.7 kg. If only recycle/reuse of
water and the use of a central storage tank had been considered, the objective func-
tion would have been 20,180 c.u. with an optimum freshwater usage of 479 kg. This
amounts to a 4.7% increase in profit and a 19.24% decrease in effluent p roduction.
The solution time was 46.3 s using a Pentium 4, 3.4 GHz processor. The schedule cor-
responding to the MILP was infeasible. Hence, the resulting solution of 21,129 c.u.
is a local optimum.
The schedule obtained from the optimization is shown in Figure 7.4. The clear
blocks represent the production operation in the unit, while the dark blocks represent
the washing operation in the unit. The numbers above the clear blocks represent the
amount of material used in the unit for production, the numbers in brackets represent
freshwater. The sum of all these amounts to 401.7 kg, hence the target. Water transfer to
and from storage as well as water transfer from regenerator have been clearly labelled.
From Figure 7.4, it can be seen that reaction 1 occurs twice and reactions 2 and 3
occur thrice. The water used for washing Reactors 1 and 2 (R1 and R2) consists of
freshwater only. 150 kg of water produced in R2 after washing is sent to storage at
2.25 h. This provides opportunity for indirect water reuse or wastewater regenera-
tion. The storage tank discharges 145.5 kg of water to the regenerator with respect
to R1 at 2.80 h. The regenerated water is discharged to R1 at 4.25 h for washing.
TABLE 7.2
Wastewater Minimization Data for Literature Example
Maximum Concentration (g Contaminant/kg Water)
Contaminant 1 Contaminant 2 Contaminant 3
Reaction 1 (Reactor 1) Max. inlet 0.5 0.5 2.3
Max. outlet 1 0.9 3
Max. outlet 0.2 0.1 1.2
Max. outlet 0.3 1 1.2
Max. outlet 0.1 1 12
Max. outlet 0.075 0.2 1
Max. outlet 2 1.5 2.5
Mass Load (g)
Reaction 1 Reactor 1 4 80 10
Reactor 2 15 24 358
Reaction 2 Reactor 1 28.5 7.5 135
Reactor 2 9 2 16
Reactor 2 22.5 45 36.5
Duration of Washing (h)
Reaction 1 Reaction 2 Reaction 3

Reactor 1 0.25 0.5 0.25
Reactor 2 0.3 0.25 0.25
TABLE 7.3
Removal Ratio for Contaminants
Contaminant Removal Ratio
Contaminant 1 0.98
Contaminant 2 0.97
Contaminant 3 0.96
The storage tank later discharges 7.5 kg of water to the regenerator with respect to
R2 at 4.43 h. The regenerated water is discharged to R2 at 4.5 h for washing. A simi-
lar scenario is encountered with the water sent to storage from R2 and R1 at 4.75 h.
The regenerated water is utilized in the later washing operations that follow. It can
be observed from this solution that direct water reuse between units did not occur.
Units
50 113.75
Separator
[To storage: 150] [To storage: 120]
[From regenerator: 7.5] [From regenerator: 82.42] [From regenerator: 30]
(150) (112.5) (40.3)
40 80 80 63.75
Reactor 2 Reaction 1 Reaction 2 Reaction 2 Reaction 3
[To storage:145.5]
[From regenerator: 145.5] [From regenerator: 83.08] [From regenerator: 70]
(88.9)
25 50 50 50
52 32
Heater
0 1 2 3 4 5 6 7 8 9 10
2.25 4.25 4.75 5.75 6.75 8.25
4.5 Time (h)
FIGURE 7.4 Resulting production schedule for literature example.
7.6.2 Case Study II
This case study is based on a section of a pharmaceuticals production plant which
produces four types of products, namely shampoos, deodorants, lotions and creams.
Four mixers for product mixing were available in this section. Each product was
produced by a specific mixer. Mixer 1 was dedicated to the mixing of shampoos,
mixer 2 was dedicated to the mixing of deodorants, mixer 3 was dedicated to the
mixing of lotions and mixer 4 was dedicated to the mixing of creams. The general
production procedure is as follows. Raw material is charged to a mixer. The raw
material is then mixed until the required physical characteristics are obtained. Once
a product is mixed, it is removed and stored. The mixers are then washed. There is
sufficient storage available for product. The typical production requirement over a
24-h time horizon is given in Table 7.4. In Table 7.4, the average production time for
each product is also given.
The data pertaining to wastewater minimization are given in Table 7.5 together
with the residue mass left in each mixer after product has been removed. These are
average values taken from actual process data. The maximum outlet concentrations
from each mixer are also given. It is important to note that there is maximum outlet
concentration defined for only one component from each mixer. This is because each
mixer will at any point only contain residue of a specific product. Also given is the
maximum contaminant load that can enter with the water for a cleaning operation.
TABLE 7.4
Production Data for Case Study II
Mixer Product Number of Batches Duration (h)
1 Shampoo 2 7
2 Deodorant 3 5.5
3 Lotion 1 11
4 Cream 2 11
TABLE 7.5
Wastewater Minimization Data for Case Study II
Mixer Contaminant Residue Mass (kg) Limiting Water (kg) Cmax Outlet (kg/kg)
1 Shampoos 15 576.9 0.04
2 Deodorants 15 361.4 0.045
3 Lotions 30 697.6 0.05
4 Creams 70 1238.9 0.06
In this problem, the residue left from the specific product in a mixer accounts for
the contaminant mass added to water. This then makes the limiting component in
each mixer the component that leaves residue in the mixer. For example, mixer 1 has
shampoo as the limiting contaminant, since this is the only component which leaves
a residue in the mixer. The limiting amount of water for each mixer is calculated
using Equation 7.12 and is given for each mixer in Table 7.5.
A central storage vessel with a capacity of 10 tons is available for storing waste-
water used for washing. The washing of each mixer takes 30 min.
The maximum inlet concentration of each mixer is given in Table 7.6. In this
case, the maximum inlet concentration of the deodorant in mixers 1, 3 and 4 is set to
zero. This is because the reuse of wastewater containing deodorant in a mixer with
TABLE 7.6
Maximum Inlet Concentrations for Cleaning Operation in the
Case Study II
Shampoo Deodorant Lotion Creams
(kg Product/kg (kg Product/kg (kg Product/kg (kg Product/kg
Mixer Water) Water) Water) Water)
1 0.014 0 0.007 0.0035
2 0.014 0.0035 0.007 0.007
3 0.014 0 0.007 0.0035
4 0.014 0 0.007 0.0035
TABLE 7.7
Removal Ratio for Contaminants
Contaminant Removal Ratio
Shampoo 0.95
Deodorant 0.99
Lotion 0.96
Cream 0.98
any other residue is undesirable. However, wastewater containing the deodorant can
be reused in a cleaning operation in mixer 2, containing another deodorant residue.
A regenerator with a flowrate of 466 kg/h is available. The removal ratio of the
various contaminants is given in Table 7.7.

Case Study II was formulated using the presented methodology in GAMS 22.0 and
solved using the DICOPT2 algorithm, with CPLEX 9.1.2 as the MIP and CONOPT3
as the NLP solvers. The resulting model had 576 binary variables. The optimum
number of time points was 16. The value of the objective function, which was
the minimization of effluent as given in Constraint (7.67), was 2653 kg of water.
If regeneration was not considered, the objective function was 3587 kg of water. The
incorporation of the regenerator resulted in a 26% decrease in effluent production.
The solution time was 464 s using a Pentium 4, 3.4 GHz processor. The schedule
corresponding to the MILP was infeasible. Hence, the resulting solution of 2653 kg
is a local optimum.
The resulting schedule of the solution is given in Figure 7.5. The clear block repre-
sents the production operation in the unit, while the dark blocks represent the wash-
ing operation in the unit. The numbers in normal brackets represent freshwater. The
sum of these amounts to 2653 kg, hence the target. The numbers in square brackets
represent direct reuse water. Water transfer to and from storage as well as water
transfer from regenerator have been clearly labelled.
From the Gantt chart in Figure 7.5, it can be observed that the schedule meets the
production requirement, that is, two batches of shampoo, three batches of deodorant,
one batch of lotion and two batches of cream. Mixer 3 uses freshwater in addition to
direct reuse water from mixer 1. Mixer 3 sends 600 kg of water to storage at 11.5 h.
Of this water, 201.7 kg is indirectly reused in mixer 4 at 11.5 h. At 20.2 h, the storage
tank discharges 1166.7 kg of water to the regenerator on behalf of mixer 4. At 23 h,
the regenerated water is discharged to mixer 4. The rest of the production schedule
can be observed from Figure 7.5. As mentioned earlier, it is not acceptable to use
water from mixer 2 containing deodorant. This was ensured in the model by setting
the binary variable associated with water transfer from mixer 2 to any other mixer
to zero and also by setting the binary variable associated with mixer 2 transferring
water to the storage tank to zero. These conditions were met, as can be seen from
the Gantt chart.
Units [To storage: 834.5] [From regenerator: 1166.7]

[From storage:
201.7] (965)
Mixer 4
[To storage: 600]
(225)
Mixer 3
[375] [From storage: 66.2] [114.4]
(333.3) (267.2) (218.9)
Mixer 2
[105.9]
(375) (269.1)
Mixer 1
00.5 2 3.5
4 6 8 10 11 12 14 16 18 20 22 24
5.5 6.5 10.5 16.5
11.5 12.5 23 23.5
Time (h)
FIGURE 7.5 Production schedule for the Case Study II.
7.7 CONCLUSIONS
The regeneration methodology presented in this chapter deals with wastewater mini-
mization where wastewater can contain multiple contaminants. The methodology
does not require the production schedule to be given beforehand. The methodol-
ogy is able to determine the minimum wastewater generated and the corresponding
schedule that achieves this minimum. The methodology was applied to two case
studies. In Case Study I where the production was not given, a 19.24% reduction in
wastewater was achieved. In Case Study II where the production requirement was
given, a 26% reduction in wastewater was achieved.
The model has a considerable number of bilinear terms. Upon linearization, addi-
tional variables and constraints are introduced in the model, which results in the
model being very large. The amount of wastewater minimized depends on the given
removal ratios of contaminants in the regenerator.
7A APPENDIX
The mathematical formulation is nonlinear due to the bilinear terms in Constraints
(7.4), (7.5), (7.15), (7.16) and (7.23). Consider Constraint (7.4) containing the follow-
ing bilinear terms:
1. mwin ( sout , j , p ) cin ( sout , j , c, p )
2. mwr ( sout , j ¢ , sout , j , p ) cout ( sout , j , c, p )

3. msout ( sout , j , p ) csout ( c, p )

4. mregout ( sout , j , p ) crout ( c, p )
These terms are linearized using the method by Sherali and Alameddine (1992), as
discussed by Quesada and Grossmann (1995).
Let
mwin ( sout , j , p ) cin ( sout , j , c, p ) = G1 ( sout , j , c, p )

mwr ( sout , j ¢ , sout , j , p ) cout ( sout , j , c, p ) = G 2 ( sout, j ¢, sout, j , c, p )
msout ( sout , j , p ) csout ( c, p ) = G3 ( sout , j , c, p )

mregout ( sout , j , p ) crout ( c, p ) = G 4 ( sout , j , c, p )

with each variable having the following bounds:
0 £ mwin ( sout , j , p ) £ MwU ( sout , j )

0 £ cin ( sout , j , c, p ) £ CinU ( sout , j , c )

0 £ mwr ( sout , j ¢ , sout , j , p ) £ MwU ( sout , j )

0 £ cout ( sout , j , c, p ) £ Cout

U
( sout , j , c )

0 £ msout ( sout , j , p ) £ MwU ( sout , j )

0 £ csout ( c, p ) £ CSout
U
(c)

0 £ mregout ( sout , j , p ) £ MwU ( sout , j )

0 £ crout ( c, p ) £ CRout
U
(c)

The following are true for G1:
G1 ( sout , j , c, p ) ³ 0, "j Î J , sout , j Î Sout , j , c Î C , pÎP (7A.1)

G1 ( sout , j , c, p ) ³ MwU ( sout , j ) cin ( sout , j , c, p ) + mwin ( sout , j , p ) CiUn ( sout , j , c )
- MwU ( sout , j ) CinU ( sout , j , c ) , "j Î J , sout , j Î Sout , j , c Î C , pÎP (7A.2)
G1 £ MwU ( sout , j ) cin ( sout , j , c, p ) , "j Î J , sout , j Î Sout , j , c Î C , p Î P (7A.3)

G1 £ mwin ( sout , j , p ) CinU ( sout , j , c ) , "j Î J , sout , j Î Sout , j , c Î C , pÎP (7A.4)

The following are true for G 2:
G 2 ( sout , j ¢ , sout , j , c, p ) ³ 0, "j, j¢ Î J , sout , j Î Sout , j , c Î C , pÎP (7A.5)

G 2 ( sout , j ¢ , sout , j , c, p ) ³ MwU ( sout , j ) cout ( sout , j ¢ , c, p ) + mwr ( sout , j ¢ , sout , j , p ) Cout
U
( sout , j¢ , c )
- Mw ( sout , j ) Cout ( sout , j ¢ , c ) , "j, j¢ Î J , sout , j Î Sout , j , c Î C , pÎP
U U
(7A.6)
G 2 ( sout , j ¢ , sout , j , c, p ) £ MwU ( sout , j ) cout ( sout , j ¢ , c, p )
"j, j¢ Î J , sout , j Î Sout , j , c Î C , pÎP (7A.7)

G 2 ( sout , j ¢ , sout , j , c, p ) £ mwr ( sout , j ¢ , sout , j , p ) Cout

U
( sout , j¢ , c ) ,
"j, j¢ Î J , sout , j Î Sout , j , c Î C , pÎP (7A.8)
The following are true for G3:
G3 ( sout , j , c, p ) ³ 0, "j Î J , sout , j Î Sout , j , c Î C , pÎP (7A.9)

G3 ( sout , j , c, p ) ³ MwU ( sout , j ) csout ( c, p ) + msout ( sout , j , p ) CSout

U
(c)
- MwU ( sout , j ) CSout
U
( c ) , "j Î J , sout , j Î Sout , j , c Î C, pÎP (7A.10)
G3 ( sout , j , c, p ) £ MwU ( sout , j ) csout ( c, p ) , "j Î J , sout , j Î Sout , j , c Î C , pÎP

(7A.11)
G3 ( sout , j , c, p ) £ msout ( sout , j , p ) CSout

U
( c ) , "j Î J , sout , j Î Sout , j , c Î C, pÎP

(7A.12)
The following are true for G 4 :
G 4 ( sout , j , c, p ) ³ 0, "j Î J , sout , j Î Sout , j , c Î C , pÎP (7A.13)

G 4 ( sout , j , c, p ) ³ MwU ( sout , j ) crout ( c, p ) + mregout ( sout , j , p ) CRou

U
t (c)
- MwU ( sout , j ) CRout

U
( c ) , "j Î J , sout , j Î Sout , j , c Î C, p Î P (7A.14)
G 4 ( sout , j , c, p ) £ MwU ( sout , j ) crout ( c, p ) , "j Î J , sout , j Î Sout , j , c Î C , pÎP

(7A.15)
G 4 ( sout , j , c, p ) £ mregout ( sout , j , p ) CRout

U
( c ) , "j Î J , sout , j Î Sout , j , c Î C, pÎP

(7A.16)
Substituting these linearized variables into Constraint (7.4) gives Constraint (7A.17):
G1 ( sout , j , c, p ) = å G (s
sout , j¢
2 out , j ¢ , sout , j , c, p ) + G3 ( sout , j , c, p ) + G 4 ( sout , j , c, p )

"j, j¢ Î J , sout , j Î Sout , j , p Î P, c Î C (7A.17)
The bilinear terms in Constraints (7.5), (7.15), (7.16) and (7.23) are linearized in a
similar manner and substituted in the original constraints to yield new constraints
similar to Constraint (7A.17).
7A.1 Solution Procedure
The relaxed MILP problem consists of Constraints (7.1) through (7.65), (7.66) or
(7.67) with the new substituted constraints instead of Constraints (7.4), (7.5), (7.15),
(7.16) and (7.23). Also included are the respective forms of Constraints (7A.1) through
(7A.4) for each linearized variable. The relaxed MILP problem is solved to provide
a starting point for the exact MINLP problem. The solution procedure is illustrated
in Figure 7A.1.
Exact MINLP is linearized using reformulation-linearization

(Sherali and Alameddine 1992).
Resulting MILP from linearization is solved.
Solution from MILP is used as starting solution for exact MINLP.
MILP objective = MINLP objective MILP objective ≠ MINLP objective

Globally optimal solution Locally optimal solution
FIGURE 7A.1 Solution procedure.

REFERENCES
Liu, Y., Li, G., Wang, L., Zhang, J., Shams, K., 2009. Optimal design of an integrated discon-
tinuous water-using network coordinating with a central regeneration unit. Industrial
Majozi, T., Gouws, J.F., 2009. A mathematical optimisation approach for wastewater minimi-
sation in multipurpose batch plants: Multiple contaminants. Computers and Chemical
Quesada, I., Grossmann, I.E., 1995. Global optimization of bilinear process networks with
multicomponent flows. Computers and Chemical Engineering. 19(12), 1219–1242.
Sherali, H.D., Alameddine, A., 1992. A new reformulation–linearization technique for bilin-
ear programming problems. Journal of Global Optimum. 2, 379–410.
8 Optimization of
Energy and Water
Use in Multipurpose
Batch Plants Using an
Improved Mathematical
Formulation
8.1 INTRODUCTION
Presented in this chapter is a formulation that addresses optimization of both water
and energy, while simultaneously optimizing the batch process schedule. The sched-
uling framework used in this study is based on a recent and efficient formulation.
This formulation has been shown to result in a significant reduction of computational
time and an improvement of the objective function and lead to fewer time points. The
objective is to improve the profitability of the plant by minimizing wastewater gen-
eration and utility usage. From a case study, it was found that through applying only
water integration the cost is reduced by 11.6%, by applying only energy integration
the cost is reduced by 29.1% and by applying both energy and water integration the
cost is reduced by 34.6%. This indicates that optimizing water and energy integra-
tion in the same scheduling framework will reduce the operating cost and environ-
mental impact significantly.
In recent years, batch processes have been getting more attention due to their suit-
ability for the production of small volume, high value–added products. The flex-
ibility of batch plants allows the production of different products within the same
facility. Batch manufacturing is typically used in the pharmaceutical, polymer, food
and specialty chemical industries as demand for such products are highly seasonal
and are influenced by changing markets. A common feature of many batch plants
is that they utilize fossil fuels as the energy source and use water for process equip-
ment cleaning, due to inherent sharing of equipment by different tasks. Despite the
advantage of batch plants being flexible, they also pose a challenging task to operate
in a sustainable way. In the past, batch industries could tolerate high inefficiencies in
energy and water consumption due to the high value of final products which outstrips
197
the production costs. However, greater public awareness of the impact of industrial
pollution, more stringent environmental regulations and escalating raw materials,
energy, and waste treatment costs have now motivated energy and water saving mea-
sures for more sustainable operations (Halim and Srinivasan, 2011). Since schedul-
ing, energy and wastewater minimization for multipurpose batch plants go hand in
hand, published works in those areas are reviewed.
8.2.1 Scheduling of Batch Plants

Much research has been done on developing mathematical models to improve
batch plant efficiency. The substantial advancement in modern computers allows
the possibility of handling large and more complex problems by using optimization
techniques. Excellent reviews of current scheduling techniques based on different
time representations and associated challenges have been conducted (Floudas and
Lin, 2004; Méndez et al., 2006; Shaik et al., 2006). In the reviews, with regard to time
representation, the models are classified as slot based, event based and precedence
based (sequence based). In slot-based models (Pinto and Grossmann, 1994; Lim and
Karimi, 2003; Liu and Karimi, 2008), the time horizon is divided into ‘non-uniform
unknown slots’ and tasks start and finish in the same slot. On the other hand, slot
models exist that use non-uniform unknown slots where tasks are allowed to continue
to the next slots (Schilling and Pantelides, 1996; Karimi and McDonald, 1997; Reddy
et al., 2004; Sundaramoorthy and Karimi, 2005; Erdirik-Dogan and Grossmann,
2008; Susarla et al., 2010). The event-based models can also be categorized into those
that use uniform unknown events, where the time associated with the events is com-
mon across all units (Maravelias and Grossmann, 2003; Castro et al., 2004), and those
that use unit-specific events where the time associated with the events can be different
across the units (Ierapetritou and Floudas, 1998; Majozi and Zhu, 2001; Shaik et al.,
2006; Janak and Floudas, 2008; Shaik and Floudas, 2009; Li et al., 2010). The het-
erogeneous location of events across the units gives fewer event points as compared
to both the global event–based and slot-based models. As a result, unit-specific event-
based models are computationally superior. The sequence-based or precedence-based
representation uses either direct precedence (Hui and Gupta, 2000; Liu and Karimi,
2007) or indirect precedence sequencing of pairs of tasks in units (Méndez and
Cerdá, 2000, 2003; Méndez et al., 2001; Ferrer-Nadal et al., 2008). The models do not
require pre-postulation of events and slots. Seid and Majozi (2012) presented a mixed
integer linear programming (MILP) formulation based on the state sequence network
(SSN) and unit-specific time points, which can handle proper sequencing of tasks and
fixed intermediate storage (FIS) policy. The model results in a reduction of event or
time points required and as a result gives better performance in terms of objective
value and CPU time required when compared to previous literature models.
8.2.2 Energy Integration in Batch Plants

Many heat integration techniques are applied to predefined schedules which are
inherently suboptimal. Vaklieva-Bancheva et al. (1996) considered direct heat
integration with the objective of minimizing total costs. The resulting overall
Optimization of Energy and Water Use in Multipurpose Batch Plants 199
formulation was an MILP problem, solved to global optimality, although only spe-
cific pairs of units were allowed to undergo heat integration. Uhlenbruck et al. (2000)
improved OMNIUM, which is a tool developed for heat exchanger network synthesis
by Hellwig and Thöne (1994). The improved OMNIUM tool increased the energy
recovery by 20%. Bozan et al. (2001) developed a single-step, interactive computer
program (BatcHEN) used for the determination of the campaigns, that is the set of
products which can be produced simultaneously, the heat exchange areas of all pos-
sible heat exchangers in the campaigns and the heat exchanger network. This work
addressed the limitation of the graph theory method for the determination of the
campaign by Bancheva et al. (1996). Krummenacher and Favrat (2001) proposed a
new systematic procedure, supported by graphics, which made it possible to deter-
mine the minimum number of heat storage units. Chew et al. (2005) applied cascade
analysis proposed by Kemp and Macdonald (1987) to reduce the utility requirement
for the production of oleic acid from palm olein using immobilized lipase. The result
obtained showed savings of 71.4% and 62.5% for hot and cold utilities respectively.
Pires et al. (2005) developed the BatchHeat software, whose aim was to highlight the
energy inefficiencies in the process and thereby enabling the scope for possible heat
recovery to be established through direct heat exchange or storage through imple-
mentation of cascade analysis.
Boer et al. (2006) evaluated the technical and economic feasibility of an indus-
trial heat storage system for an existing production facility of organic surfactants.
Fritzson and Berntsson (2006) applied process integration methods to investigate
the potential to decrease the energy usage in the slaughtering and meat processing
industry. The result obtained illustrates that 30% of the external heat demand and
more than 10% of the shaftwork used can be saved. Morrison et al. (2007) devel-
oped a user-friendly software package known as optimal batch integration (OBI).
Chen and Ciou (2008) formulated a method to design an optimization of indirect
energy storage systems for batch process. Their work aimed at simultaneously
solving the problem of indirect heat exchange network synthesis and its associ-
ated thermal storage policy for recirculated hot/cold heat storage medium (HEN).
Most of the previous works solved this sequentially. Foo et al. (2008) extended
the minimum units targeting and network evolution techniques that were devel-
oped for batch mass exchange network (MEN) into batch HEN. They applied
the technique for energy integration of oleic acid production from palm olein
using immobilized lipase. Halim and Srinivasan (2009) discussed a sequential
method using direct heat integration. A number of optimal schedules with mini-
mum makespan were found, and heat integration analysis was performed on each.
The schedule with minimum utility requirement was chosen as the best. Later,
Halim and Srinivasan (2011) extended their technique to carry out water reuse
network synthesis simultaneously. One key feature of this method is its ability to
find the heat integration and water reuse solution without sacrificing the quality
of the scheduling solution.
Atkins et al. (2010) applied indirect heat integration using heat storage for a milk
powder plant in New Zealand. The traditional composite curves have been used to
estimate the maximum heat recovery and to determine the optimal temperatures of
the stratified tank. Tokos et al. (2010) applied a batch heat integration technique to
a large beverage plant. The opportunities of heat integration between batch opera-
tions were analyzed by a MILP model, which was slightly modified by considering
specific industrial circumstances. Muster-Slawitsch et al. (2011) came up with the
Green Brewery concept to demonstrate the potential for reducing thermal energy
consumption in breweries. Three detailed case studies were investigated. The
‘Green Brewery’ concept has shown a saving potential of more than 5000 t/year
fossil CO2 emissions from thermal energy supply for the three breweries that were
closely considered. Becker et al. (2012) applied time average energy integration
approach to a real case study of a cheese factory with non-simultaneous process
operations. Their work addressed appropriate heat pump integration. A cost saving
of more than 40% was reported.
For a more optimal solution, scheduling and heat integration may be combined
into an overall problem. Papageorgiou et al. (1994) embedded a heat integration
model within the scheduling formulation of Kondili et al. (1993). Opportunities
for both direct and indirect heat integration were considered as well as possi-
ble heat losses from a heat storage tank. The operating policy, in terms of heat
integrated or standalone, was predefined for tasks. Adonyi et al. (2003) used the
‘S-Graph’ scheduling approach and incorporated one to one direct heat integra-
tion. Barbosa-Póvoa et al. (2001) presented a mathematical formulation for the
detailed design of multipurpose batch process facilities with heat integration.
Pinto et al. (2003) extended the work of Barbosa-Póvoa et al. (2001) with the
consideration of the economic savings in utility requirements, while consider-
ing the cost of both the auxiliary structures that is heatexchanger through their
transfer area and the design of the utility circuits and associated piping costs.
Majozi (2006) presented a direct heat integration formulation based on the state
sequence network of Majozi and Zhu (2001) which uses an unevenly discretized
time horizon. The direct heat integration model developed by Majozi (2006) was
extended to incorporate heat storage for more flexible schedules and utility sav-
ings in the later work by Majozi (2009). However, the storage size is a parameter
in his formulation which is addressed later by Stamp and Majozi (2011), where the
storage size is determined by the optimization exercise. Chen and Chang (2009)
extended the work of Majozi (2006) to periodic scheduling, based on the resource
task network (RTN) scheduling framework. The reader can get a more compre-
hensive and detailed review on energy recovery for batch processes in the chapter
by Fernández et al. (2012).
8.2.3 Wastewater Minimization in Batch Plants

Wastewater is generated in batch plants during the cleaning of multipurpose equip-
ment and when water is used as a solvent. Tight environmental regulations and
increased public awareness demand that batch plants consider rational use of water
during their operation. Many researchers have developed methodologies for the effi-
cient use of water through direct reuse, indirect reuse and regeneration of wastewa-
ter. Direct reuse consists of recycle and reuse. Recycle refers to the reuse of an outlet
wastewater stream from a processing unit in the same unit, while reuse refers to the
use of an outlet wastewater stream from a processing unit in another processing unit.
Indirect reuse is when wastewater is temporarily stored in a storage vessel and later
reused in a processing unit requiring water.
Based on the analogy of heat and mass transfer, several methodologies for synthe-
sizing water reuse network in batch processes have also been developed. Gouws et al.
(2010) reviewed these techniques based on graphical-based pinch analysis and math-
ematical optimization approach. The seminal work on pinch analysis application to
batch water network was reported by Wang and Smith (1994). Foo et al. (2005) pro-
posed a time-dependent water cascade analysis to obtain minimum required water
flows in a process. While these graphical-based techniques are useful, they share a
common drawback that their application is limited to single contaminant cases. The
mathematical optimization-based techniques, which are capable of solving multiple
contaminant problems, can be differentiated into two groups, namely, those based
on the schedule being known a priori, that is sequential approach and those that
simultaneously determine the process schedule and minimize the freshwater usage.
Almató et al. (1997) addressed the problem of water reuse through storage tank
allocation based on the optimal schedule being known a priori. Kim and Smith
(2004) proposed a more generalized method for optimal design of discontinuous
water reuse network. In their approach, a production schedule was fixed and direct
reuse of water between operations within the same time interval was allowed with-
out passing through storage tanks. Most of the mathematically based models are
based on a superstructure approach. Majozi and Gouws (2009) proposed a continuous-
time scheduling framework to simultaneously optimize the schedule and water reuse
while addressing both single and multiple contaminants. Cheng and Chang (2007)
considered the optimization of the batch production schedule, water reuse schedule
and wastewater treatment schedule in a single problem based on discrete time sched-
uling framework. At the end of optimization, the production schedule, the num-
ber and sizes of buffer tanks and the physical configuration of the pipeline network
were obtained. Adekola and Majozi (2011) extended the work of Majozi and Gouws
(2009) by incorporating wastewater regenerator for further improvement of water
utilization.
From the review, it can be seen that wastewater minimization and heat inte-
gration in batch plants are addressed separately. To the knowledge of the authors,
only the work presented by Halim and Srinivasan (2011) and Adekola et al. (2013)
addressed this literature gap. In the work of Halim and Srinivasan (2011), the over-
all problem is decomposed into three parts, namely scheduling, heat integration and
water reuse optimization and solved sequentially. Batch scheduling is solved first
to meet an economic objective function. Next, alternate schedules are generated
through a stochastic search-based integer cut procedure. For each resulting sched-
ule, minimum energy and water reuse targets are established and networks identi-
fied which might lead to suboptimal results. Adekola et al. (2013) also addressed
this problem by developing a model that simultaneously optimize energy, water
and production throughput. They demonstrated that the unified approach where
all resources are optimized simultaneously gives a better economic performance
compared to the common sequential techniques for wastewater and energy integra-
tion techniques developed for multipurpose batch plants. However, the model has
two basic limitations. The first drawback is the model is not based on time average
model (TAM) and treats the temperature driving force based on initial and target
temperatures of cold and hot streams. This assumption makes the model impossible
to apply for a case where the starting and finishing time of the heat integrated units
to be anywhere between the starting and finishing time of the processing tasks since
it is required to calculate the intermediate temperatures to ensure for the minimum
thermal driving force. The second limitation is it forces the heat integrating units to
start simultaneously which results suboptimal because of restricting the flexibility
of the schedule.
In this chapter, a section is made to close the literature gap by simultaneously solv-
ing energy integration and wastewater minimization problem in the same scheduling
framework. The model is based on TAM and time slice model (TSM) where the time
slice is a variable determined by optimization to keep the flexibility of the schedule as
compared to previous models based on fixed schedule and fixed time slice for heat inte-
gration. The model also addressed the two basic limitations as discussed in the model of
Adekola et al. (2013). Additionally, the proposed model used the resent robust scheduling
framework of Seid and Majozi (2012) as a platform since the model gave better objective
value as compared to previous literature models. The rest of the chapter is organized as
follows. Section 8.3 defines the problem statement. Section 8.4 describes the detail math-
ematical formulation. Section 8.5 describes the application of the mathematical model to
literature problems. Finally, conclusions are drawn from this work in Section 8.6.
Given:
1. The production recipe (STN or SSN representation)
2. The capacity of units and the type of tasks each unit can perform
3. The maximum storage capacity for each material
4. The task processing times
5. Hot duties for tasks require heating and cold duties for tasks that require
cooling
6. Operating temperatures of heat sources and heat sinks
7. Minimum allowable temperature differences
8. The material heat capacities
9. The units’ washing time
10. The mass load of each contaminant
11. The concentration limits of each contaminant
12. The costs of raw materials, products and utilities
13. The scheduling time horizon (for profit maximization problem)
14. Production demand (for makespan minimization problem)
Determine:
1. The optimum production schedule, that is allocation of tasks to units, tim-
ing of all tasks, and batch sizes
2. Optimum energy requirement and associated heat exchange configuration
3. Optimum water requirement and associated water-reuse network.
The scheduling model by Seid and Majozi (2012) was adopted as a scheduling plat-
form since it has proven to result in better CPU time and optimal objective value
compared to other scheduling models. Uneven discretization of the time horizon
so-called continuous time was used.
8.4.1 Heat Integration Model

The mathematical model is based on the superstructure in Figure 8.1. Each task may
operate using either direct or standalone mode by using only external utilities. If
direct integration is not sufficient to satisfy the required duty, external utilities may
make up for any deficit.
will be integrated with one cold unit when direct heat integration takes place in
order to simplify operation of the process. It is worth noting that mathematically
it is also possible for one unit to integrate with more than one unit at a given time
point when the summation notation is not used. However, this is practically very dif-
ficult to implement. Also, if two units are to be heat integrated at a given time point,
they must both be active at that time point. For better understanding, the difference
between time point p and extended time point pp. is explained using Figure 8.2. If a
unit j that is active at time point p is integrated with more than one unit in different
temperature and time intervals, an extended time point pp. must be defined. Unit j1
active at time point p can be integrated with units j2 and j3 in different time and tem-
perature intervals. At the beginning, unit j1 is integrated with unit j2 at time point
p and the extended time point pp. is the same as time point p. Later, j1 is integrated
with unit j3 in another time interval where extended time point pp. equals to p + 1.
pp. is equal to or greater than time point p and less than or equal to n + p, where n
is a parameter which is greater than or equal to zero. If n equals 2, then a unit that is
active at time point p can be integrated in three different intervals. The model should
Vessel B (colt)
Vessel A (hot) External

heating
External
cooling
FIGURE 8.1 Superstructure for the energy integration.

J3
J2
J1
Time (h)
x(j1, j2, p, pp = p) tuu(j1, p, pp = p + 1)
tuu(j3, p, pp = p + 1)
x(j1, j3, p, pp = p + 1)
FIGURE 8.2 Differentiating time point p and extended time point pp.
be solved starting from n equals zero and adding one at a time until no better objec-
tive value is achieved.
åx ( s injc, sinjh, p, pp ) £ y ( sinjh , p ) , "p, pp Î P, sinjh Î SinJh , sinjc Î SinJc (8.1)

sinjc
åx ( s
sinj h
injc , sinjh, p, pp ) £ y ( sinjc , p ) , "p, pp Î P, sinjh Î SinJh , sinjc Î SinJc (8.2)

Constraint (8.3) describes the amount of cooling load required by the hot unit
from its initial temperature to its target temperature. In a situation where the temper-
ature in the reactor unit is fixed during exothermic reaction, the heat load becomes
the product of the amount of mass that undergoes reaction and the heat of reaction.
( )
cl ( sinjh , p ) = mu ( sinjh , p ) cp ( sinjh ) Tsininjh - Tsout
inj h
, "p Î P, sinjh Î SinJh

(8.3)
Constraint (8.4) describes the amount of heating load required by the cold unit
from its initial temperature to its target temperature. In a situation where the temper-
ature in the reactor unit is fixed during endothermic reaction, the heat load becomes
the product of the amount of mass that undergoes reaction and the heat of reaction.
(
hl ( sinjc , p ) = mu ( sinjc , p ) cp ( sinjc ) Tsout
injc )
- Tsininjc , "p Î P, sinjc Î SinJc

(8.4)
Constraints (8.5) and (8.6) describe the average heat flow for the hot and cold unit,
respectively, during the processing time which is the same as TAM model to address
the energy balance during heat integration properly.
( )
cl ( sinjh , p ) = avcl ( sinjh , p ) tp ( sinjh , p ) - tu ( sinjh , p ) , "p Î P, sinjh Î SinJh

(8.5)
( )
hl ( sinjc , p ) = avhl ( sinjc , p ) tp ( sinjc , p ) - tu ( sinjc , p ) , "p Î P, sinjc Î SinJc

(8.6)
Constraints (8.7) and (8.8) define the heat load at time point p and extended time
point pp. for the hot and cold unit.
(
hlp ( sinjc , p, pp ) = avhl ( sinjc , p ) tpp ( sinjc , p, pp ) - tuu ( sinjc , p, pp ) , )
(
clp ( sinjh , p, pp ) = avcl ( sinjh , p ) tpp ( sinjh , p, pp ) - tuu ( sinjh , p, pp ) , )
Constraints (8.9) and (8.10) are used to calculate the temperature of the hot and
cold unit at the intervals.
(
clp ( sinjh , p, pp ) = mu ( sinjh , p ) cp ( sinjh ) T in ( sinjh , p, pp ) - T out ( sinjh , p, pp ) , )

(
hlp ( sinjc , p, pp ) = mu ( sinjc , p ) cp ( sinjc ) T out ( sinjc , p, pp ) - T in ( sinjc , p, pp ) , )
Constraint (8.11) states that the amount of heat exchanged by the hot unit with the
cold units should be less than the cooling load required by the hot unit during the
interval.

åqe ( s
sinjc
injc , sinjh , p, pp ) £ clp ( sinjh , p, pp ) , "p, pp Î P, sinjh Î SinJh , sinjc Î SinJc

(8.11)
Constraint (8.12) states that the amount of heat exchanged by the cold unit with
the hot units should be less than the heat load required by the cold unit during the
interval.
åqe ( s
sinjh
injc , sinjh , p, pp ) £ hlp ( sinjc , p, pp ) , "p, pp Î P, sinjh Î SinJh , sinjc Î SinJc

(8.12)
Constraints (8.13) and (8.14) state that the temperature of the unit at the start of
an interval should be equal to the temperature at the end of the previous interval.
T in ( sinjh , p, pp ) = T out ( sinjh , p, pp - 1) , "p, pp Î P, sinjh Î SinJh (8.13)
T in ( sinjc , p, pp ) = T out ( sinjc , p, pp - 1) , "p, pp Î P, sinjc Î SinJc (8.14)
Constraints (8.15) and (8.16) state that the temperature at the start of the first
interval, which is time point p, which is also pp, should be equal to the initial tem-
perature of the task.
T in ( sinjh , p, pp ) = Tsininjh , "p, pp Î P, p = pp, sinjh Î SinJh (8.15)
T in ( sinjc , p, pp ) = Tsininjc , "p, pp Î P, p = pp, sinjc Î SinJc (8.16)
obeyed when there is direct heat integration between a hot and a cold unit.
(
T in ( sinjh , p, pp ) - T out ( sinjc , p, pp ) ³ DT - DT U 1 - x ( sinjc , sinjh , p, pp ) , )
(
T out ( sinjh , p, pp ) - T in ( sinjc , p, pp ) ³ DT - DT U 1 - x ( sinjc , sinjh , p, pp ) , )
Constraints (8.19) through (8.22) ensure that the times at which units are active
are synchronized when direct heat integration takes place.
( )
tuu ( sinjh , p, pp ) ³ tuu ( sinjc , p, pp ) - M 1 - x ( sinjc , sinjh , p, pp ) ,
( )
tuu ( sinjh , p, pp ) £ tuu ( sinjc , p, pp ) + M 1 - x ( sinjc , sinjh , p, pp ) ,
( )
tpp ( sinjh , p, pp ) ³ tpp ( sinjc , p, pp ) - M 1 - x ( sinjc , sinjh , p, pp ) ,
( )
tpp ( sinjh , p, pp ) £ tpp ( sinjc , p, pp ) + M 1 - x ( sinjc , sinjh , p, pp ) ,

Constraints (8.23) and (8.24) stipulate that the starting time of the heating load
required for the cold unit and cooling load required for the hot unit at the first inter-
val should be equal to the starting time of the hot and cold unit.
tuu ( sinjh , p, pp ) = tu ( sinjh , p ) , "p, pp Î P, p = pp, sinjh Î SinJh (8.23)
tuu ( sinjc , p, pp ) = tu ( sinjc , p ) , "p, pp Î P, p = pp, sinjc Î SinJc (8.24)
Constraints (8.25) and (8.26) state that the starting time of heating and cooling in
an interval should be equal to the finishing time at the previous interval.
tuu ( sinjh , p, pp ) = tpp ( sinjh , p, pp - 1) , "p, pp Î P, sinjh Î SinJh (8.25)
tuu ( sinjc , p, pp ) = tpp ( sinjc , p, pp - 1) , "p, pp Î P, sinjc Î SinJc (8.26)
Constraint (8.27) ensures that if heat integration occurs, the heat load should have
a value that is less than the maximum amount of heat exchangeable. When the binary
variable associated with heat integration takes a value of zero, no heat integration
occurs and the associated heat load is zero.
qe ( sinjc , sinjh , p, pp ) £ Q x ( sinjc , sinjh , p, pp ) , "p, pp Î P, sinjh Î SinJh , sinjc Î SinJc

U
(8.27)
Constraints (8.28) and (8.29) state that if the binary variable associated with heat
integration is active, then the binary variable associated with heating and cooling
must be active at that interval.
x ( sinjc , sinjh , p, pp ) £ yint ( sinjh , p, pp ) , "p, pp Î P, sinjh Î SinJh , sinjc Î SinJc

(8.28)
x ( sinjc , sinjh , p, pp ) £ yint ( sinjc , p, pp ) , "p, pp Î P, sinjh Î SinJh , sinjc Î SinJc

(8.29)
Constraints (8.30) and (8.31) state that the heating and cooling loads take on a
value for a certain duration when the binary variables associated with heating and
cooling are active.
( )
tpp ( sinjh , p, pp ) - tuu ( sinjh , p, pp ) £ Hyint s injh , p, pp , "p, pp Î P, sinjh Î SinJh

(8.30)
( )
tpp ( sinjc , p, pp ) - tuu ( sinjc , p, pp ) £ Hyint s injc , p, pp , "p, pp Î P, sinjc Î SinJc

(8.31)
Constraints (8.32) and (8.33) state that temperatures change in the heating and
cooling unit when the binary variables associated with heating and cooling are active.
T in ( sinjh , p, pp ) - T out ( sinjh , p, pp ) £ DT U ( sinjh ) yint ( sinjh , p, pp ) ,
"p, pp Î P, sinjh Î SinJ h (8.32)
T out ( sinjc , p, pp ) - T in ( sinjc , p, pp ) £ DT U ( sinjc ) yint ( sinjc , p, pp ) ,
Constraint (8.34) states that the cooling of a hot unit will be satisfied by direct
heat integration and external cooling utility if required.
cl ( sinjh , p ) = cw ( sinjh , p ) + åqe ( s

sinjc
injc , sinjh , p, pp ) ,
Constraint (8.35) states that the heating of a cold unit will be satisfied by direct
heat integration and external heating utility if required.
hl ( sinjc , p ) = st ( sinjh , p ) + åqe ( s

sinjh
injc , sinjh , p, pp ) ,
8.4.2 Wastewater Minimization Model

The superstructure on which the wastewater minimization model is based is depicted
in Figure 8.3. Only the water-using operations which are part of a complete batch
process are depicted. Unit j represents a water-using operation in which the water
mew(sinj, p)
mfw(sinj, p)
mwout(sinj, p)
Unit j
mwin(sinj, p)
mwr(sinj΄, sinj, p)
mwr(sinj, sinj΄, p)
mwout(sinj΄, p)
Unit j΄
mwin(sinj΄, p)
mwr(sinj΄, p)
mew(sinj΄, p)
FIGURE 8.3 Superstructure for water usage.

used can consist of freshwater, reuse water or reuse and freshwater. Water from unit
j can be reused elsewhere or sent to effluent treatment.
Constraint (8.36) defines the amount of water entering the unit as the sum of
freshwater and reuse water from other units.
mwin ( sinj , p ) = mfw ( sinj , p ) + åmrw ( s inj ¢ , sinj , p ) , "p Î P, sinj , sinj ¢ Î SinJ (8.36)
sinj ¢
Constraint (8.37) states that the amount of water leaving the unit is equal to the
sum of reuse water sent to other units and water sent to effluent treatment.
mwout ( sinj , p ) = åmrw ( s inj , sinj ¢ , p ) + mew ( sinj , p ) , "p Î P, sinj , sinj ¢ Î SinJ (8.37)
sinj ¢
Constraint (8.38) is the water balance around the unit and states that the amount
of water entering the unit equals the amount of water leaving the unit.
mwin ( sinj , p ) = mwout ( sinj , p ) , "p Î P, sinj Î SinJ (8.38)
Constraint (8.39) defines the inlet contaminant load as the mass of contaminant,
entering with reuse water.
cin ( sinj , c, p ) mwin ( sinj , p ) = åcout ( s

sinj ¢
inj ¢ , c, p ) mrw ( sinj ¢ , sinj , p ) ,
"p Î P, sinj , sinj ¢ Î SinJ , c Î C (8.39)
Constraint (8.40) states that the amount of contaminant leaving the unit equals
the sum of the contaminant entering into the unit and the contaminant removed from
the process.
( )
mwout s inj , p cout ( sinj , c, p ) = SMC ( sinj ) mu ( sinj , p ) + cin ( sinj , c, p ) mwin ( sinj , p ) ,

"p Î P, sinj Î SinJ , c Î C (8.40)
Constraint (8.41) ensures that the amount of reused water from unit j to other
units does not exceed the maximum allowable water in the receiving units. It also
indicates whether water from unit j is reused or not.
mrw ( sinj , sinj ¢ , p ) £ WinU ( sinj ¢ ) yre ( sinj , sinj ¢ , p ) , "p Î P, sinj , sinj ¢ Î SinJ (8.41)
Constraint (8.42) ensures that the reuse of water from unit j in other units can
occur only if the units are active.
yre ( sinj , sinj ¢ , p ) £ y ( sinj ¢ , p ) , "p Î P, sinj , sinj ¢ Î SinJ (8.42)

Constraint (8.43) gives the upper bound on the water entering into unit j. It also
ensures that water enters into the unit only if it is active.
mwin ( sinj , p ) £ WinU ( sinj ) y ( sinj , p ) , "p Î P, sinj Î SinJ (8.43)
In Constraints (8.44) and (8.45), wastewater can only be directly reused if the
finishing time of the unit producing wastewater and the starting time of the unit
receiving wastewater coincide.
tuw ( sinj , p ) ³ tpw ( sinj ¢ , p ) - M * yre ( sinj , sinj ¢ , p ) , "p Î P, sinj , sinj ¢ Î SinJ (8.44)
tuw ( sinj , p ) £ tpw ( sinj ¢ , p ) + M * yre ( sinj , sinj ¢ , p ) , "p Î P, sinj , sinj ¢ Î SinJ (8.45)
Constraint (8.46) defines the finishing time of the washing operation as the start-
ing time of the washing operation added to the duration of washing.
tpw ( sinj , p ) ³ tuw ( sinj , p ) + tw ( sinj ) y ( sinj , p ) , "p Î P, sinj Î SinJ (8.46)
Constraint (8.47) ensures that the starting time of a task in a unit is greater than
the finishing time of the washing operations.
(
tu ( sinj , p ) ³ tpw sinj )
¢ , p - 1 , "p Î P, s , s ¢ Î S , S*
inj inj inJ inj

(8.47)
Constraint (8.48) stipulates that the starting time of the washing operation in a
unit occurs after the completion of the task in the unit.
tuw ( sinj , p ) ³ tp ( sinj , p ) , "p Î P, sinj Î SinJ (8.48)
Constraints (8.49) and (8.50) ensure that the inlet and outlet concentrations do not
exceed the maximum allowable concentration.
cin ( sinj , c, p ) £ cinU ( sinj , c ) , "p Î P, sinj Î SinJ , c Î C (8.49)
cout ( sinj , c, p ) £ cout U ( sinj , c ) , "p Î P, sinj Î SinJ , c Î C (8.50)
Constraint (8.51) is the objective function in terms of profit maximization, with

profit defined as the difference between revenue from product, cost of utility, raw
material cost, freshwater cost and effluent treatment cost.
æ
ç p å ( ) ( ) åå
price s p d s p - costcw * cw ( sinjh , p ) ö
÷
ç s p sinjh ÷
ç ÷
max ç
ç
- åå
p sinjc
costst * st ( sinjc , p ) - åå
p sinj
costfw * mfw ( sinj , p ) ÷
÷
,
ç ÷
ç -
ç åå costew * mew ( sinj , p ) ÷
÷
è p sinj ø
"p,Î P, sinjh Î SinJh , sinjc Î SinJc , sinj Î SinJ (8.51)
Constraint (8.52) defines minimization of energy and wastewater if the product

demand is known.
æ
ç åå åå
costcw * cw ( sinjh , p ) + costst * st ( sinjc , p ) ö
÷
min ç ÷,
p sinjh p sinjc
ç ÷
çç + åå costfw * mfw ( sinj , p ) + åå costew * mew ( sinj , p ) ÷
÷
è p sinj p sinj ø
8.5 CASE STUDIES
Case studies from published literature were selected to demonstrate the application
of the proposed model. The results from the proposed models were obtained using
CPLEX 9 as MILP solver and CONOPT 3 as NLP solver in DICOPT interface of
GAMS 22.0 and were solved using a 2.4 GHz, 4 GB of RAM, Acer TravelMate
5740G computer.
8.5.1 Case Study I
This case study has been investigated extensively in published literature (Halim and
Srinivasan, 2011). It is a simple batch plant requiring only one raw material to yield
a product as depicted in the state task network (STN) representation in Figure 8.4.
The plant comprises of five units and two intermediate storage units. The conversion
of the raw material into product is achieved through three sequential processes. The
first task can be performed in two units (j1 and j2), the second task can be performed
only in unit j3 and the third task can be performed in units j4 and j5. Tasks 1 and 2 require
cooling during their operation, while task 3 requires heating. The cooling and heat-
ing demands are satisfied by external utilities and heat integration. The operational
philosophy requires that the units are cleaned before the next batch is processed.
Both freshwater and reuse water can be used as cleaning agents. Table 8.1 gives the
capacities of the units, durations of processing and washing tasks, initial availability
of states, storage capacities and selling prices and costs for the states. Table 8.2 gives
data pertaining to initial and target temperatures for the tasks, specific heat capaci-
ties for the states, maximum inlet and outlet contaminant concentrations which are
unit dependent and the specific contaminant loads.
Cooling Washing Cooling Heating Washing
140°C 60°C 40°C 80°C

Task 1 Task 2 Task 3
(i = 1) (i = 2) (i = 3)
{Unit 1, Unit 2} {Unit 3} {Unit 4, Unit 5}
A B C D
(m = 1) (m = 2) (m = 3) (m = 4)
FIGURE 8.4 STN representation of a simple batch plant producing one product.
212
TABLE 8.1
Scheduling Data for Case Study I
Max Batch Total Operation Material Initial Revenue or Cost
Task (i) Unit (j) Size (kg) Time (h) Washing Time (h) State (m) Inventory (kg) Max Storage (kg) ($/kg or $/MJ)
Task 1 Unit 1 100 1.5 0.25 A 1000 1000 0
Unit 2 150 2 0.3 B 0 200 0
Task2 Unit 3 200 1.5 0 C 0 250 0

Task3 Unit 4 100 1 0.25 D 0 1000 5
Unit 5 150 1.5 0.3 Wash water 0.1
Wastewater 0.05
Cooling water 0.02
Steam 1
Note: Total operation time includes processing time and washing time.
Optimization of Energy and Water Use in Multipurpose Batch Plants
TABLE 8.2
Energy and Cleaning Requirements for Case Study I
Contaminant Loading
Max Inlet Max Outlet (g Contaminant/kg
Task (i) Tin(°C) Tout(°C) Unit ( j) Cp(kJ/kg°C) Concentration (ppm) Concentration (ppm) Batch)
Task 1 140 60 Unit 1 4 500 1000 0.2
Unit 2 4 50 100 0.2
Task 2 60 40 Unit 3 3.5 — — 0.2
Task 3 40 80 Unit 4 3 150 300 0.2
Unit 5 3 300 2000 0.2
Cooling water 20 30
Steam 170 160
213

The computational results for Case Study I using the proposed model for the differ-
ent scenarios and results obtained from literature are presented in Table 8.3. For the
scenario without energy and water integration, the total cost of utilities was $293.5.
Applying only water integration, the total cost obtained was $259.5, which is an
11.6% reduction, compared to the standalone operation without energy and water
integration. For the scenario with energy integration only, a total cost of $208 was
obtained, which is a 29.1% reduction compared to the standalone operation. The
fifth column shows the results obtained with combined energy and water integration
solved simultaneously giving a total cost of $191.8, which is a 34.6% saving com-
pared to the standalone operation. These results show that in order to achieve the
best economic performance, the scheduling problem has to be solved simultaneously
considering both water and energy integration.
The performance of the proposed model was compared to the sequential optimi-
zation technique by Halim and Srinivasan (2011) which resulted in an overall cost
of $239.5, which is an 18.4% saving, much less than the 34.6% saving obtained by
the proposed model. This work also gives much better result compared to the recent
simultaneous optimization technique of Adekola et al. (2013) with a cost saving
of 23.4% compared to 34.6% saving obtained by the proposed model. The subop-
timality results of the method by Adekola et al. (2013) are attributed to two basic
drawbacks. The first drawback is due to restricting the flexibility of the schedule by
forcing the heat integrated units to start at the same time. The second drawback is
the model is not based on TAM and the possibility of heat integration between pairs
of tasks as well as possible ΔT violations was investigated for each pair of hot and
cold tasks before optimization using the initial and target temperatures of the heat
integrated tasks. This limits the chance of a unit to be integrated in multiple inter-
vals with different intermediate temperatures with other units. Using the proposed
model, we keep the schedule flexibility by allowing the heat integrated units to start
anywhere between the starting and finishing time of the heat integrated tasks. This
benefit can be demonstrated in Figure 8.5, for example, Unit 2 during processing a
task from 3.2 to 4.9 h is integrated to exchange heat with Unit 4 during processing a
task from 4.25 to 5 h. These two units are integrated from 4.25 to 4.9 h to exchange
heat which is not possible by the method of Adekola et al. (2013). Consequently, this
work reduces the steam requirement by 40.9% compared to technique by Adekola
et al. (2013). The efficiency of the proposed model can be attributed to solving
the scheduling problem while incorporating water and energy integration in the
same framework and also using the recent efficient scheduling technique by Seid
and Majozi (2012). Figure 8.5 details the possible amount of energy integration
between the cold and hot units and the time intervals during which energy integra-
tion occurred.
The energy requirements of unit j2 and unit j4 during the interval 3.2–5 h are
emphasized to elaborate on the application of the proposed model. The cooling load
of unit j2 between 3.2 and 4.9 h was 32 MJ. This is partly satisfied through energy
integration with unit j4 in the same time interval, resulting in an external cooling
requirement of 26.8 MJ rather than 32 MJ if operated in standalone mode. At the
TABLE 8.3
Proposed Halim and
Formulation Proposed Proposed Proposed Formulation Srinivasan (2011) Adekola et al. (2013)
without Water and Formulation with Formulation with with Water and with Water and with Water and
Energy Integration Water Integration Energy Integration Energy Integration Energy Integration Energy Integration
Profit ($) 4706.5 4740.5 4791.5 4808.2 4764.1 4777.3
Steam (MJ) 120 120 36.63 39 43.9 66
Cooling 390 390 281.2 309 313.9 336
water (MJ)
Total 1105 878.2 1105 977.7 1238.4 1013.3
freshwater (kg)
Revenue from 5000 5000 5000 5000 5000 5000
product ($)
Cost of steam ($) 120 120 36.63 39 43.9 66
Cost of cooling 7.8 7.8 5.623 6.2 6.3 6.72
water ($)
Cost of 110.5 87.8 110.5 97.7 123.8 101.3
freshwater ($)
Cost of 55.25 43.9 55.2 48.9 61.9 50.7
wastewater ($)
Total cost ($) 293.5 259.5 208 191.8 235.9 224.72
CPU time (s) 2.3 5000 5000 5000 Not reported 28797
215
7.03 MJ 2.44 MJ 3.67 MJ

j2 C C C C C C
19.76 MJ 15.2 MJ 2.7 MJ
5.2 MJ 12 MJ
0.8 MJ
j4 S
12 MJ
12.72 MJ 1.28 MJ 1.28 MJ 7.2 MJ 12.8 MJ

j1 C C C C C C
12.72 MJ
18 MJ 12 MJ
18 MJ
1.8 MJ 13.5 MJ
j5 S S
4.5 MJ
10.64 MJ
j3 C
3.2 7.25
4.9 5.2 5.7 6 6.9 7.7 8.45 9 9.3 10.2 10.5
4.25 6.15 7.2 7.5 8.95
4.5 5 5.75 8.7
7.3 Time (h)
FIGURE 8.5 Possible energy integration within the time horizon of 12 h for Case Study I.
beginning of the operation of unit j2 from 3.2 to 4.25 h, the cooling requirement was
19.76 MJ. This value was obtained using the time average model by multiplying the
duration (4.25–3.2 h) and the energy demand per hour (32 MJ/1.7 h (total duration of
the task) = 18.823 MJ) where the cooling requirement is fully satisfied by external
cooling. For the rest of its operation between 4.25 and 4.9 h, the cooling requirement
was 12.24 MJ, satisfied partly with energy integration (5.2 MJ) and the difference by
external cooling. The heating requirement of unit j4 when it is operated during the
interval 4.25–5 h was 6 MJ. From 4.25 to 4.9 h, the steam requirement was 5.2 MJ
obtained from the time average model. This heating requirement was fully satisfied
mcr = 20 g
mew = 150 kg
J2
mfw = 200 kg mrw = 50 kg
mew = 32.35 kg
J4
mrw = 17.65 kg
mcr = 10 g
mew = 17.65 kg
J5
mcr = 30 g
4.9 5.2 5.7 6.0
mcr = 30 g
mew = 15.79 kg
J5
mcr = 17 g mrw = 15.79 kg

mew = 54.209 kg
J2
mfw = 170 kg mrw = 100 kg
mew = 71.43 kg
J4
mcr = 20 g
mrw = 28.57 kg mew = 28.57 kg
J1
mcr = 20 g
6.9 7.2 7.45 7.5 7.7
mcr = 20 g
mew = 71.43 kg
J4
mrw = 100 kg mrw = 28.57 kg
mcr = 13 g
mew = 28.57 kg
J2 J1
mfw = 130 kg mew = 30 kg
mcr = 20 g
9 9.3 9.55 9.8
FIGURE 8.6 Water network with water integration within the time horizon of 12 h for Case
Study I. Note: mfw, freshwater; mcr, contaminant removed; mrw, recycled water; mew, water
sent to effluent.
during the interval, by integrating with the hot unit j2. The rest of the heating, 0.8 MJ,
required during its operation between 4.9 and 5 h was satisfied by external steam.
Figure 8.6 shows the amount of contaminant removed, freshwater usage, amount
of reused water and wastewater produced from washing the necessary units. The
washing operation of unit j2 between 4.9 and 5.2 h required 200 kg of freshwater to
remove a contaminant load of 20 g, producing water with a contaminant concentra-
tion of 100 ppm. Part of this water produced from unit j2, 50 kg, was used for clean-
ing unit j4 to remove a contaminant load of 10 g. This was possible because the outlet
concentration from unit j2 (100 ppm) was lower than the maximum inlet contaminant
Unit 5 150 150 100 150 150

(17.65) (28.57) (17.65)
Unit 4 50 100 100 10 40
Unit 3 50 200 200 200 160 190
(50) (100) (100)

Unit 2 150 100 85 65
(28.57)
Unit 1 100 100 100 100 100 100
1 2 3 4 5 6 7 8 9 10 11 12
Time (h)
FIGURE 8.7 Gantt chart for the time horizon of 12 h incorporating energy and water
integration for Case Study I.
concentration (150 ppm) for unit j4. From Figure 8.6, the total amount of reused
water was 358.23 kg, thereby reducing the water usage from 1105 kg (without water
integration) to 977.7 kg (with water integration). This resulted in a saving of 11.5%
freshwater usage and wastewater produced.
The amount of material produced, the starting and finishing times of the pro-
cesses and washing tasks are shown in Figure 8.7 in the form of a Gantt chart.
8.5.2 Case Study II
This case study obtained from Kondili et al. (1993) has become one of the most
commonly used examples in literature. However, this case study has been adapted by
Halim and Srinivasan (2011) to include energy and water integration. The batch plant
produces two different products sharing the same processing units, where Figure 8.8
shows the plant flowsheet. The unit operations consist of preheating, three different
reactions and separation. The plant accommodates many common features of multi-
purpose batch plants such as units performing multiple tasks, multiple units suitable
for a task and dedicated units for specific tasks. The STN and SSN representations
of the flowsheet are shown in Figure 8.9. Tables 8.4 and 8.5 give the required data to
solve the scheduling problem. The production recipe is as follows:
1. Raw material, Feed A, is heated from 50°C to 70°C to form HotA used in
reaction 2.
2. Reactant materials, 50% Feed B and 50% Feed C are used in reaction 1
to produce IntBC. During the reaction, the material has to be cooled from
100°C to 70°C.
Raw 3
Product 1
Raw 2
Reactor 1
Heater
Raw 1
Product 2
Still
Reactor 2
FIGURE 8.8 Flowsheet for Case Study II.
3. 60% of the intermediate material, IntBC, and 40% of HotA are used in
reaction 2 to produce product 1 and IntAB. The process needs to be heated
from 70°C to 100°C during its operation.
4. 20% of the reactant, Feed C, and 80% of intermediate, IntAB, from reac-
tion 2 are used in reaction 3 to produce ImpureE. The reaction needs its
temperature to be raised from 100°C to 130°C during its operation.
5. The separation process produces 90% product 2 and 10% IntAB from
Impure E. Cooling water is used to lower its temperature from 130°C to
100°C.
The processing time of a task i in unit j is assumed to be linearly dependent on its

batch size B, that is, ai + bi B. Where ai is a constant term of the processing time of
task i and bi is a coefficient of variable processing time of task i. The batch-dependent
processing time makes this case study more complex. Table 8.4 gives the relevant
data on coefficients of processing times, the capacity of the processing units, dura-
tion of washing, initial inventory of raw materials, storage capacity and relevant
costs. Four contaminants are considered in the case study. The maximum inlet and
outlet concentrations are given in Table 8.5. The production demand is given as
200 kg for both Prod1 and Prod2. The objective here is to optimize with respect to
makespan, energy and water consumption.
Product 1
s7
40% IntAB
40% 60%
10%
Feed A 60%
s9 Separation
IntBC s6
Impure E 90%
80%
50%
Feed B 50% 20%

Product 2
(s3) (s4)
(a) Feed C
Feed A
s1
Feed B
s2 s5
40% Product 1
40%
50% s7
60%
60% 10%
s6
(s3) 50% 80%
IntBC s8 s9
Feed C IntAB 20% Impure E
(s4) 90%
s10
(b) Product 2
FIGURE 8.9 STN (a) and (b) SSN representations for Case Study II.

The computational statistics for this case study using the proposed model and results
obtained from literature are presented in Table 8.6. For makespan minimization, an
objective value of 19.5 h was obtained using the proposed model, which is better
than 19.96 h obtained by Halim and Srinivasan (2011) and 19.93 obtained by Adekola
et al. (2013). Using the makespan obtained, the case study was solved using the dif-
ferent scenarios for water minimization, energy minimization and the simultaneous
minimization of energy and water by setting customer requirement for Product 1
and Product 2. The total energy and freshwater required for the standalone operation
were 125.5 MJ and 357.94 kg, respectively.
TABLE 8.4
Max
Batch Washing Initial Revenue or Cost
Task (i) Unit ( j) Size (kg) α(Sinj) β(Sinj) Time Material State (s) Inventory Max Storage (kg) ($/kg or $/MJ)
Heating (H) HR 100 0.667 0.007 0 Feed A 1000 1000 0
Reaction-1 (Rl) RR1 50 1.334 0.027 0.25 Feed B 1000 1000 0
RR2 80 1.334 0.017 0.3 Feed C 1000 1000 0
Reaction-2 (R2) RR1 50 1.334 0.027 0.25 HotA 0 100 0
RR2 80 1.334 0.017 0.3 IntAB 0 200 0
Reaction-3 (R3) RR1 50 0.667 0.013 0.25 IntBC 0 150 0
RR2 80 0.667 0.008 0.3 ImpureE 0 200 0
Separation (S) SR 200 1.334 0.007 0 Prod 1 0 1000 20
Prod2 0 1000 20
Wash water 0.1
Wastewater 0.05
Cooling water 0.02
Steam 1
221
222
TABLE 8.5
Data Required for Energy and Water Integration
Max Inlet Concentration Max Outlet Concentration
Contaminants (ar, br, cp
(ppm) (ppm)
and dw) Loading
Task (i) TSininj ( C) TSininj ( C) Unit ( j) Cp(kJ/kg °C) ar br cp dw ar br cp dw (g Contaminant/kg Batch)
Heating (H) 50 70 HR 2.5
Reaction-1 100 70 RR1 3.5 300 500 800 400 700 800 1200 900 0.2
RR2 3.5 300 500 800 400 700 800 1200 900 0.2

Reaction-2 70 100 RRl 3.2 700 600 300 400 1200 1000 600 800 0.2
RR2 3.2 700 600 300 400 1200 1000 600 800 0.2
Reaction-3 100 130 RRl 2.6 500 200 400 300 800 500 700 900 0.2
RR2 2.6 500 200 400 300 800 500 700 900 0.2
Separation 130 100 SR 2.8
Cooling 20 30
water
Steam 170 160
TABLE 8.6
Computational Results for Case Study II
Proposed Proposed Proposed Proposed Halim and Adekola et al.
Formulation Formulation with Formulation with Formulation with Srinivasan (2011) (2013) with Water
without Water and Water Energy Water and Energy with Water and and Energy
Energy Integration Integration Integration Integration Energy Integration Integration
Makespan (h) 19.5 19.5 19.5 19.5 19.96 19.93
Objective ($) 127.5 112 96.4 94.3 103.3 96.5
Steam (MJ) 75.3 75.3 44.9 43.3 61.4 44.88
Cooling water (MJ) 50.2 50.2 19.7 18.1 35.4 19.72
Total freshwater (kg) 357.94 238.1 341.3 337.7 275.1 341.2
Revenue from product ($) 8000 8000 8000 8000 8000 8,000
Cost of steam ($) 75.3 75.3 44.9 43.3 61.4 44.88
Cost of cooling water ($) 1 1 0.4 0.36 0.7 0.3994
Cost of freshwater ($) 35.8 23.8 34.1 33.8 27.5 34.12
Cost of wastewater ($) 17.9 11.9 17.1 16.9 13.8 17.06
Number of time points/ 11 11 11 11 N/A 17
slots
CPU time (s) 5000 5000 5000 6074 Not reported 24,532
223
224
Processing units 10.594
SR 37.477 39.996 144.749

8.998 13.585 3.359 15.199 16.816 7.530 19.163
MJ MJ
12.805
2.11 7.498
4.804 7.198 9.892 12.748 15.199 16.506 19.5
RR2 R1 (45.621) R2 (80) R2 (80) R2 (52.349) R1 (71.93) R2 (80) R3 (80) R2 (80)
9.21 3.148 10.192 12.448 13.585 15.499 16.806 4.629 19.2
1.810 2.903 4.504 2.903
MJ 1.2 MJ
MJ 12.5 kg MJ 12.5 kg
MJ
11.682
8.998 15.283 16.556 19.5
RR1 R1 (37.984) R1 (50) R1 (50) R2 (50) R3 (50) R1 (44.465) R3 (47.4) R2 (50)

R3 (44.846)
2.167 2.11 3.706 5.054 6.4 7.498 8.748 11.432 12.998 19.25
MJ 2.360 15.533
4.794 7.748 12.748 16.816
4.2 MJ
HR 68.429 47.571 84
1.146 2.146 10.192 11.447
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Time (h)
FIGURE 8.10 Resulting production schedule for Case Study II with direct heat integration and direct water reuse.
For the scenario of water integration only allowing the use of reuse water, the total
cost was $112, resulting in 12.2% saving when compared to the standalone operation
which had a total cost of $127.52. By using only energy integration, the total energy
requirement was reduced from 125.5 MJ in standalone operation to 64.56 MJ, result-
ing in a 48.6% energy saving and a total cost saving of 24.4%. For the case of simul-
taneous optimization of energy and water, a significant total cost saving was obtained
compared to energy integration alone and water integration alone. A total cost saving
of 29.4% was obtained, compared to the standalone operation. The performance of
the proposed model was also compared to the technique by Halim and Srinivasan
(2011), a total cost of $103 was found using their technique which is significantly
higher than $94.3 obtained using the proposed model. Furthermore, the proposed
technique is very easy to adopt as opposed to their approach which required to solve
3500 MILP scheduling problem to find the best schedule compared to only 3 MILP
major iterations of the MINLP problem. Each MILP problem is solved in a speci-
fied CPU time of 2000 s. This complex case study was solved in a reasonable CPU
time of 6074 s, which is less than 2 h, using the proposed model. When this work is
compared to the model by Adekola et al. (2013), the number of event points required
reduced considerably from 17 to 11 which have a direct effect on reducing CPU time
required. Additionally, the usage of hot and cold utilities, freshwater and wastewater
is also improved.
Figure 8.10 shows the Gantt chart related to the optimal usage of energy and
water. It also indicates the types of tasks performed in each equipment, the starting
and finishing times of the processes and washing tasks and the amount of material
processed in each batch.
8.6 CONCLUSIONS
In the presented method, wastewater minimization and heat integration are both
embedded within the scheduling framework and solved simultaneously, thus leading
to a truly flexible process schedule. Results from case studies show that addressing
profit maximization together with heat integration and wastewater minimization gives
much better overall economic performance. From the case studies, a better objective
value was achieved using the proposed model compared to previous literature mod-
els. Forthcoming communications will address the usage of heat storage, wastewater
storage and wastewater regenerator with the consideration of capital investment to
investigate further improvement in energy and water usage. Although this invariably
complicates the model formulation. Additionally, this work only addressed short-
term scheduling problem. Extending this work to medium-term scheduling problem
using a cyclic approach will be reported in future communication.
NOMENCLATURE
Sets
SinJh {sinjh |sinjh task which needs cooling}
SinJc {sinjc |sinjc task which needs heating}
SinJ {sinj |sinj any task}

P {p|p time point}
SinJ w
{sinjw |sinjw task which needs washing afterwards}
C {c|c contaminant}
Parameters
cp ( sinjh ) Specific heat capacity for the heating task
cp ( sinjc ) Specific heat capacity for the cooling task
Tsininjh Inlet temperature of the heating task
Tsout
injh
Outlet temperature of the heating task
in
Tsinjc Inlet temperature of the cooling task
Tsout
injc
Outlet temperature of the cooling task
D T Maximum thermal driving force
U
DT U ( sinj ) Maximum temperature change for a task

D T Minimum thermal driving force
M Big-M mostly equivalent to the time horizon
QU Maximum heat requirement from the heating and cooling task
SMC ( sinj ) Specific contaminant load produced by a task
WinU ( sinj ) Maximum water inlet to a processing task
tw ( sinj ) Minimum duration required for a washing task
cinU ( sinj , c ) Maximum inlet contaminant concentration allowed for contami-
nant c
cout U ( sinj , c ) Maximum outlet contaminant concentration allowed for contami-
nant c
( )
price s p Price of a product
( )
d s p Amount of product produced at the end of the time horizon
costfw Cost of freshwater
costew Cost of effluent water
costst Cost of steam
costcw Cost of cooling water
Variables
x ( sinjc , sinjh , p, pp )
Binary variable signifying whether heat integration occurs
between the hot and cold unit
y ( sinjh , p ) Binary variable associated to whether the hot state is active at
time point p or not
y ( sinjc , p ) Binary variable associated to whether the cold state is active at
time point p or not
yint ( sinj , p, pp ) Binary variable associated to whether the hot and cold states are
active at time point p and extended time point pp
yre ( sinj , sinj ¢ , p ) Binary variable associated with reuse of water from unit j to jʹ at
time point p
cl ( sinjh , p ) Cooling load required by the hot task at time point p
hl ( sinjc , p ) Heating load required by the cold task at time point p
avcl ( sinjh , p ) Average cooling load required by the hot task at time point p
avhl ( sinjc , p ) Average heating load required by the cold task at time point p
mu ( sinjh , p ) Amount of material processed by the hot task
mu ( sinjc , p ) Amount of material processed by the cold task
tp ( sinj , p ) End time of a heat flow for a task
tu ( sinj , p ) Starting time of a heat flow for a task
clp ( sinjh , p, pp ) Cooling load required by the hot task active at time point p and
hlp ( sinjc , p, pp ) Heating load required by the cold task active at time point p and
tuu ( sinj , p, pp ) Starting time of a heat flow for a task active at time point p and
tpp ( sinj , p, pp ) Finishing time of a heat flow for a task active at time point p and
T in ( sinj , p, pp ) Inlet temperature of a task active at time point p and extended
time point pp
T out ( sinj , p, pp ) Outlet temperature of a task active at time point p and extended
time point pp
qe ( sinjc , sinjh , p, pp ) Amount of heat load exchanged by the hot and cold unit active
at time point p and extended time point pp
cw ( sinjh , p ) External cooling water used by the hot task
st ( sinjc , p ) External heating used by the cold task
mwin ( sinj , p ) Mass of water entering to wash a unit after a task is performed
mwout ( sinj , p ) Mass of water leaving after washing
mfw ( sinj , p ) Mass of freshwater entering to a unit
mrw ( sinj , sinj ¢, p ) Mass of water recycled from unit j to another unit j'
mew ( sinj , p ) Mass of water entering to effluent treatment produced from
washing
cin ( sinj , c, p ) Inlet contaminant concentration at time point p
cout ( sinj , c, p ) Outlet contaminant concentration at time point p
tuw ( sinj , p ) Starting time of washing operation for unit j
tpw ( sinj , p ) Finishing time of washing operation for unit j
REFERENCES
Adekola, O., Majozi, T., 2011. Wastewater minimization in multipurpose batch plants with a
regeneration unit: Multiple contaminants. Computers and Chemical Engineering. 35,
2824–2836.
Adekola, O., Stamp, J., Majozi, T., Garg, A., Bandyopadhyay, S., 2013. Unified approach
for the optimization of energy and water in multipurpose batch plants using a flex-
ible scheduling framework. Industrial and Engineering Chemistry Research. 52,
8488–8506.
Adonyi, R., Romero, J., Puigjaner, L., Friedler, F., 2003. Incorporating heat integration in
batch process scheduling. Applied Thermal Engineering. 23, 1743–1762.
Almató, M., Sanmartĺ, E., Espuña, A., Puigjaner, L., 1997. Rationalizing the water use in the
batch process industry. Computers and Chemical Engineering. 21, S971–S976.
Atkins, M.J., Walmsley, M.R.W., Neale, J.R., 2010. The challenge of integrating non-
continuous processes—Milk powder plant case study. Journal of Cleaner Production.
34, 9276–9296.
Bancheva, N., Ivanov, B., Shah, N., Pantelides, C.C., 1996. Heat exchanger network design
Barbosa-Póvoa, A.P.F.D., Pinto, T., Novais, A.Q., 2001. Optimal design of heat-integrated mul-
tipurpose batch facilities: A mixed-integer mathematical formulation. Computers and
Becker, H., Vuillermoz, A., Maréchal, F., 2012. Heat pump integration in a cheese factory.
Applied Thermal Engineering. 43, 118–127.
Boer, R., Smeding, S.F., Bach, P.W., 2006. Heat storage systems for use in an industrial batch
process (Results of) a case study. In 10th International Conference on Thermal Energy
Storage ECOSTOCK, Stockton, CA.
Bozan, M., Borak, F., Or, I., 2001. A computerized and integrated approach for heat exchanger
network design in multipurpose batch plants. Chemical Engineering Process. 40,
511–524.
and Engineering Chemistry Research. 43, 105–118.
Chen, C.L., Ciou, Y.J., 2008. Design and optimization of indirect energy storage systems for
batch process plants. Industrial and Engineering Chemistry Research. 47, 4817–4829.
Cheng, K.F., Chang, C.T., 2007. Integrated water network designs for batch processes.
Chew, Y.H., Lee, C.T., Foo, C.Y., 2005. Evaluating heat integration scheme for batch pro-
duction of oleic acid. In Malaysian Science and Technology Congress (MSTC),
(pp. 18–20).
Erdirik-Dogan, M., Grossmann, I.E., 2008. Slot-based formulation for the short-term sched-
uling of multistage, multiproduct batch plants with sequence-dependent changeovers.
Fernández, I., Renedo, C.J., Pérez, S.F., Ortiz, A., Mañana, M., 2012. A review: Energy recov-
ery in batch processes. Renewable and Sustainable Energy Reviews. 16, 2260–2277.
operations in batch scheduling A critical modeling issue. Industrial and Engineering
of chemical processes: A review. Computers and Chemical Engineering. 28, 2109–2129.
Foo, D.C.Y., Manan, Z.A., Tan, Y.L., 2005. Synthesis of maximum water recovery network
for batch process systems. Journal of Cleaner Production. 13, 1381–1394.
Fritzson, A., Berntsson, T., 2006. Efficient energy use in a slaughter and meat process-
ing plant—Opportunities for process integration. Journal of Food Engineering. 76,
594–604.
Gouws, J.F., Majozi, T., Foo., D.C.Y., Chen., C.L., Lee, J Y., 2010. Water minimization tech-
niques for batch processes. Industrial and Engineering Chemistry Research. 48(19),
8877–8893.
batch plants. Industrial and Engineering Chemistry Research. 48, 8551–8565.
Halim, I., Srinivasan, R., 2011. Sequential methodology for integrated optimization of energy
and water use during batch process scheduling. Computers and Chemical Engineering.
35, 1575–1597.
Hellwig, T., Thöne, E., 1994. Omnium: ein verfahren zur optimierung der abwarmenutzung.
BWK (Brennstoff, Warme, Kraft). 46, 393–397 [in German].
Hui, C.W., Gupta, A., 2000. A novel MILP formulation for short-term scheduling of
multi-stage multi-product batch plants. Computers and Chemical Engineering. 24,
2705–2717.
scheduling: 1 Multipurpose batch processes. Industrial and Engineering Chemistry
Research. 37, 4341–4359.
Janak, S.L., Floudas, C.A., 2008. Improving unit-specific event based continuous time
approaches for batch processes: Integrality gap and task splitting. Computers and
Karimi, I.A., McDonald, C.M., 1997. Planning and scheduling of parallel semicontinuous
processes II. Short-term scheduling. Industrial and Engineering Chemistry Research.
36, 2701–2714.
Kemp, I.C., Macdonald, E.K., 1987. Energy and process integration in continuous and batch
processes Innovation in process energy utilization. IChemE Symposium Series. 105,
185–200.
Kim, J.K., Smith, R., 2004. Automated design of discontinuous water systems. Processing
Safety Environment. 82, 238–248.
uling of batch operations I. MILP formulation. Computers and Chemical Engineering.
17, 211–227.
Krummenacher, P., Favrat, D., 2001. Indirect and mixed direct–indirect heat integra-
tion of batch processes based on Pinch Analysis. International Journal of Applied
Thermodynamics. 4, 135–143.
Li, J., Susarla, N., Karimi, I.A., Shaik, M., Floudas, C.A., 2010. An analysis of some unit-
specific event-based models for the short-term scheduling of non-continuous pro-
cesses II. Short-term scheduling. Industrial and Engineering Chemistry Research.
49, 633–647.
Lim, M.F., Karimi, I.A., 2003. Resource-constrained scheduling of parallel production
lines using asynchronous slots. Industrial and Engineering Chemistry Research. 42,
6832–6842.
Liu, Y., Karimi, I.A., 2007. Scheduling multistage, multiproduct batch plants with non identi-
cal parallel units and unlimited intermediate storage. Chemical Engineering Science.
62, 1549–1566.
Liu, Y., Karimi, I.A., 2008. Scheduling multistage batch plants with parallel units and no
interstage storage. Computers and Chemical Engineering. 32, 671–693.
An aspect of cleaner production. Journal of Cleaner Production. 17, 945–950.
Majozi, T., Gouws, J.F., 2009. A mathematical optimisation approach for wastewater minimi-
zation in multipurpose batch plants: Multiple contaminants. Computers and Chemical
bach plants. Industrial and Engineering Chemistry Research. 40, 5935–5949.
Maravelias, C.T., Grossmann, I.E., 2003. New general continuous-time state-task network
formulation for short-term scheduling of multipurpose batch plants. Industrial and
Méndez, C.A., Cerdá, J., 2000. Optimal scheduling of a resource-constrained multiproduct
batch plant supplying intermediates to nearby end product facilities. Computers and
Méndez, C.A., Cerdá, J., 2003. An MILP continuous-time framework for short-term sched-
uling of multipurpose batch processes under different operation strategies. Optical
Méndez, C.A., Cerdá, J., Grossmann, I.E., Harjunkoski, I., Fahl, M., 2006. State of-the-
art review of optimization methods for short-term scheduling of batch processes.
Morrison, A.S., Walmsley, M.R.W., Neale, J.R., Burrell, C.P., Kamp, P.J.J., 2007. Non-
continuous and variable rate processes: Optimisation for energy use. Asia-Pacific
Journal of Chemical Engineering. 5, 380–387.
Muster-Slawitsch, B., Weiss, W., Schnitzer, H., Brunner, C., 2011. The green brewery con-
cept Energy efficiency and the use of renewable energy sources in breweries. Applied
Thermal Engineering. 31, 2123–2134.
Papageorgiou, L.G., Shah, N., Pantelides, C.C., 1994. Optimal scheduling of heat-integrated
multipurpose plants. Industrial and Engineering Chemistry Research. 33, 3168–3186.
Pinto, J.M., Grossmann, I.E., 1994. Optimal cyclic scheduling of multistage continuous mul-
tiproduct plants. Computers and Chemical Engineering. 1994, 18, 797–816.
ematical formulation. Annals of Operation Research. 120, 201–230.
Pires, A.C., Fernandes, C.M., Nunes, C.P., 2005. An energy integration tool for batch process,
sustainable development of energy, water and environment systems. In Proceedings of
the third Dubrovnik Conference, pp. 5–10.
Reddy, P.C.P., Karimi, I.A., Srinivasan, R., 2004. A new continuous-time formulation for
scheduling crude oil operations. Chemical Engineering Science. 59, 1325–1341.
and a novel solution algorithm. Computers and Chemical Engineering. 20, 1221–1226.
Shaik, M., Floudas, C., 2009. Novel unified modeling approach for short term scheduling.
uling of multipurpose batch plants: A comparative study. Industrial and Engineering
Stamp, J., Majozi, T., 2011. Optimal heat storage design for heat integrated multipurpose
batch plants. Energy. 36(8), 1–13.
Science. 60, 2679–2702.
Susarla, N., Li, J., Karimi, I.A., 2010. A novel approach to scheduling of multipurpose batch
Tokos, H., Pintarič, Z.N., Glavič, P., 2010. Energy saving opportunities in heat integrated
plant retrofit. Applied Thermal Engineering. 30, 36–44.
Uhlenbruck, S., Vogel, R., Lucas, K., 2000. Heat integration of batch processes. Chemical
Engineering and Technology. 23, 226–229.
Vaklieva-Bancheva, N., Ivanov, B.B., Shah, N., Pantelides, C.C., 1996. Heat exchanger net-
work design for multipurpose batch plants. Computers and Chemical Engineering. 20,
989–1001.
Wang, Y.P., Smith, R., 1994. Wastewater minimization. Chemical Engineering Science.
49(7), 981–1006.
9 Targeting for Long-Term
Time Horizons
Water Optimization
9.1 INTRODUCTION
Most of the methodologies published in literature on wastewater minimization for
batch processes are based on short-term scheduling techniques. When these methods
are applied to longer time horizons, the computational time becomes intractable,
hence the focus of this chapter. This chapter presents a methodology for simultane-
ous optimization of production schedule and wastewater minimization in a multipur-
pose batch facility. The key feature of the presented methodology is the adaption of
cyclic scheduling concepts to wastewater minimization. The methodology is devel-
oped based on continuous-time formulation and the state sequence network (SSN)
representation. The methodology is successfully applied to two common literature
examples and an industrial case study to demonstrate its effectiveness. None of the
currently published wastewater minimization techniques could solve the case study
for a time horizon of 168 h. However, through the application of the presented meth-
odology, a time horizon of 168 h for the case study was reduced to eight cycles with
the cycle length of 23 h, for which the CPU time for the optimum cycle is 64.53 s.
The problem considered in this chapter can be formally stated as follows:
For each water-using operation
Given:
1. The production recipe for each product
2. The available units and their capacities
3. The mass load, maximum inlet and outlet concentration for each
contaminant
4. Water requirement and the cleaning duration for each unit to achieve the
required cleanliness
5. The maximum storage available for water reuse
6. Time horizon of interest
Determine:
The optimal production schedule which will generate the minimum amount
of wastewater through reuse and recycle opportunities. Reuse refers to
233
the use of an outlet water stream from operation j in another operation

j′, whereas recycle refers to the use of an outlet water stream from oper-
ation j in the same operation j. It is worthy of note that the minimization
of wastewater generation is concomitant with reduction in freshwater
consumption. The absolute minimum wastewater generation can be
achieved through a flexible rather than fixed schedule of the operations.
Thus, time is treated as a variable instead of a parameter in this chapter.
The concept of cyclic scheduling and the proposed mathematical formulation are
presented in this section.
9.3.1 Concepts of Cyclic Scheduling

The concept of cyclic scheduling is based on the following axiom, which was origi-
nally proposed by Shah et al. (1993):
Consider a case where the time horizon (H) is much longer when compared to the
duration of the individual tasks, a sub-schedule exists with a much smaller time hori-
zon (T), periodic execution of which achieves production very close to the optimal
production of the original longer time horizon (H).
The axiom is depicted in Figure 9.1, the optimal cycle length being T on the
diagram.
From Figure 9.1, it is clear that some of the tasks cross the boundary of the opti-
mal cycle length T. A task that has such an effect can be viewed as a task extended
past the cycle of interest notionally wrapping around to the beginning of the cycle.
This notion was introduced in the work by Shah et al. (1995) and it is depicted in
Figure 9.2 by the tasks in units 2 and 3.
The key feature of the presented methodology is to adapt these concepts of cyclic
scheduling into wastewater minimization for batch processes.
T
Units
Unit 4
Unit 3
Unit 2
Unit 1
Time horizon H >>T Time (h)
FIGURE 9.1 The original time horizon H is much greater than the optimal cycle length T.
Targeting for Long-Term Time Horizons 235
T T
Units
Unit 4
Unit 3
Unit 2
Unit 1
(a) (b)
Time horizon H >>T Time (h)
FIGURE 9.2 (a) Task 2 in unit 2 overlaps over the cycle. (b) Task 2 is wrapped around the
beginning of the cycle.
9.3.2 Nomenclature
All the sets, variables and parameters present in the formulation are listed here.
9.3.2.1 Sets
J { j| j = unit}
C {c|c = contaminant}
Sout {sout |sout = output state from any unit}
S {s|s = any state} = Sin È Sout
Sin, j {sin, j|sin, j = input state into unit j} Í Sin
Sin*, j {s*in, j|s*in, j = input state into unit j} Í Sin
Sout , j {sout , j |sout , j = output state from unit j} Í Sout
9.3.2.2 Variables Associated with Wastewater Minimization

mwin ( sout , j , p ) Mass of water into unit j for cleaning state sout at time point p
mwout ( sout , j , p ) Mass of water produced at time point p from unit j
mw f ( sout , j , p ) Mass of freshwater into unit j at time point p
mwe ( sout , j , p ) Mass of effluent water from unit j at time point p
mwr ( sout , j , sout , j , p ) Mass of water recycled to unit j¢ from j at time point p
msin ( sout , j , p ) Mass of water transferred from unit j to storage at time point p
msout ( sout , j , p ) Mass of water transferred from storage to unit j at time point p
cin ( sout , j , c, p ) Inlet concentration of contaminant c, to unit j at time point p
cout ( sout , j , c, p ) Outlet concentration of contaminant c, from unit j at time point p
csin ( sout , j , c, p ) Inlet concentration of contaminant c, to storage at time point p
csout ( sout , j , c, p ) Outlet concentration of contaminant c, from storage at time

point p
qws ( p ) Amount of water stored in storage at time point p
tsin ( sout , j , p ) Time at which water is transferred from unit j to storage at
time point p
tsout ( sout , j , p ) Time at which water is transferred from storage to unit j at
time point p
twin ( sout , j , p ) Time that the water is used at time point p in unit j
twout ( sout , j , p ) Time at which water is produced at time point p from unit j
twr ( sout , j , sout , j ¢ , p ) Time at which water is recycled from unit j to unit j¢ at time
point p
ywr ( sout , j , sout , j ¢ , p ) Binary variable showing usage of recycle from unit j to unit j¢
at time point p
yw ( sout , j , p ) Binary variable showing the usage of unit j at time point p
ysin ( sout , j , p ) Binary variable showing transfer of water from unit j to stor-
age at time point p
ysout ( sout , j , p ) Binary variable showing transfer of water from storage to unit
j at time point p
9.3.2.3 Variables Associated with Production Scheduling
tout ( sout , j , p ) Time at which a state is produced from unit j at time point p
tin ( sin, j , p ) Time at which a state is used in or enters unit j at time point p
mout ( sout , j , p ) Amount of state produced from unit j at time point p
min ( sin, j , p ) Amount of state used in or enters unit j at time point p
( )
y s*in, j , p Binary variable associated with usage of state s at time point p
d ( sout , p ) Amount of state delivered to customers at time point p
H Time horizon for a single cycle
9.3.2.4 Parameters Associated with Wastewater Minimization

CE Cost of effluent water treatment (c.u./kg water)
CF Cost of freshwater (c.u./kg water)
M ( sout , j , c ) Mass load of contaminant c added from unit j to the water stream
MwU ( sout , j ) Maximum inlet water mass of unit j
C U
( sout , j , c )
in Maximum inlet concentration of contaminant c in unit j
C ( sout , j , c )
U
out Maximum outlet concentration of contaminant c from unit j
tw ( sout , j ) Mean processing time of unit j
Qws0 Initial amount of water in storage
QwUs Maximum capacity of storage
9.3.2.5 Parameters Associated with Production Scheduling

V jU Maximum design capacity of a particular unit j
L
V
j Minimum design capacity of a particular unit j
u
H Upper bound of the cycle time length
( )
t s*in, j Mean processing time for a state
Qs0 ( s ) Initial amount of state s stored
QsU ( s ) Maximum amount of state s stored within the time horizon
CF Interest selling price of product s,s = product
9.3.3 Mathematical Model
As mentioned earlier, the presented mathematical formulation is, in essence, an
extension of the published work by Majozi and Gouws (2009). The constraints con-
sidered in the mathematical formulation are divided into three modules. The first
module deals with the water mass balance constraints, for the case where a central
storage vessel for reusable water is available and a case where it is absent. The sec-
ond module deals with the sequencing and scheduling constraints, that is time, for
direct and indirect reuse/recycle of water. The third module, which is presented in
detail in the work by Majozi and Zhu (2001), deals with the necessary scheduling of
operations for production purposes. More emphasis will be on the new constraints
added to cater for cyclic scheduling.
9.3.3.1 Mass Balance Constraints

In this sub-section, all the essential water and contaminant mass balances are for a
washing operation presented. It is assumed throughout the formulation that none of
the water using operations produces or consumes water, that is, water is conserved
during a washing operation.
9.3.3.1.1 Mass Balance Constraints without Storage

The mathematical formulation for multiple contaminant system without storage is
based on the superstructure given in Figure 9.3, adopted from the work by Majozi
and Gouws (2009). For each water-using operation, water into the unit is a combi-
nation of freshwater and water recycled from other units to that unit. Water leav-
ing the unit can either be discarded as effluent or be recycled/reused in other units.
It is highly imperative to realize that the water-use and reuse variables shown in
Figure 9.3 correspond to the washing operation that follows immediately after the
production of a particular state in unit j, that is, sout,j. Consequently, these variables
are task-specific, which allows scheduling to be readily embedded within the water
reuse and recycle framework.
The first constraints considered are the mass balance constraints around unit j.
Constraint (9.1) is a water mass balance over the inlet of unit j. The total water into
a unit is the sum of all the recycles to the unit and the freshwater into the unit j at
time point p. As mentioned earlier, it is assumed that any operation that takes place
in unit j does not produce water. Consequently, water is conserved as captured in
(Freshwater)
mwf (sout,j, p – 1) (Reused downstream)
mwr (sout,j, sout,j΄, p)
sout,j΄
(Water in) (Water out)

mwin (sout,j, p – 1) mwout (sout,j, p)
j
cin (sout,j, c, p – 1) cout (sout,j, c, p)
(Recycled/reused water) (Effluent)

mwr (sout,j΄, sout,j, p – 1) mwe (sout,j, p)
sout,j΄
(a)
(Freshwater)
j–1 j j+1 j
(Recycle/reuse) (Effluent)
(b)
FIGURE 9.3 Superstructure for the mathematical formulation with no reusable water
storage (Majozi and Gouws, 2009).
Constraint (9.2). Constraint (9.3) is the outlet water balance from unit j. Here, the
total water out of a unit is the sum of all the recycle streams to other units and the
water discarded as effluent. Constraint (9.4) is a contaminant balance over unit j.
The mass of the contaminants out of the unit j is the sum of the contaminant into the
unit j and the mass of the contaminant added during the washing operation of
the unit. As there are more than one contaminant present in the system, the balance
has to be done for each contaminant c. Constraint (9.5) is the definition of the inlet
concentration of contaminant c to unit j.
mwin ( sout , j , p ) = å mw ( s
sout , j
r out , j ¢ , sout , j , p ) + mw f ( sout , j , p ) ,
mwin ( sout , j , p - 1) = mwout ( sout , j , p ) , "j Î J , sout , j Î Sout , j , p Î P, p > p1

(9.2)
mwout ( sout , j , p ) = å mw ( s
sout , j
r out , j , sout , j , p ) + mwe ( sout , j , p ) ,

mwout ( sout , j , p ) cout ( sout , j , c, p ) = mwin ( sout , j , p - 1) cin ( sout , j , c, p - 1)

+ M ( sout , c ) yw ( sout , j , p - 1) "j Î J , sout , j Î Sout , j , p Î P, p > p1, c Î C (9.4)
cin ( sout , j , c, p ) =
å sout , j
mwr ( sout , j ¢ , sout , j , p ) cout ( sout , j , c, p )
,
mwin ( sout , j , p )
"j, j¢ Î J , sout , j Î Sout , j , p Î P, c Î C (9.5)
The outlet concentration of each contaminant c in unit j cannot exceed its maxi-
mum limit as stated in Constraint (9.6). Constraint (9.7) ensures that the total water
into a unit j does not exceed the maximum allowable water for the operation in unit j.
Constraint (9.8) restricts mass of water recycled into the unit j to the maximum
allowable water for operation in unit j. Constraint (9.9) stipulates that the inlet con-
centration for contaminant c into unit j cannot exceed its upper limit.
cout ( sout , j , c, p ) £ Cout

U
( sout , j , c ) yw ( sout , j , p - 1) ,
mwin ( sout , j , p ) £ Mw ( sout , j ) yw ( sout , j , p ) , "j Î J , sout , j Î Sout , j , p Î P

U
(9.7)
mwr ( sout , j , sout , j , p ) £ MwU ( sout , j ) ywr ( sout , j , sout , j , p ) ,
cin ( sout , j , c, p ) £ CinU ( sout , j , c ) yw ( sout , j , p ) ,
The maximum water quantity into a unit is represented by Equation 9.10. It is

important to note that for multi-contaminant wastewater the outlet concentration
of the individual components cannot all be set to the maximum, since the contami-
nants are not limiting simultaneously. The limiting contaminant(s) will always be at
the maximum outlet concentration and the non-limiting contaminants will be below
their respective maximum outlet concentrations.
ïì M ( sout , j , c ) üï
MwU ( sout , j ) = max í U ý , "j Î J , sout , j Î Sout , j (9.10)
ïî Cout ( sout , j , c ) - Cin ( sout , j , c ) ïþ
cÎC U

9.3.3.1.2 Linearization
Constraints (9.4) and (9.5) contain bilinear terms. This makes the model thus far
nonlinear. These two constraints are not the only source of nonlinearity in the model,
as this will be more apparent as the rest of the formulation unfolds. These nonlinear
terms can be linearized according to the linearization of bilinear terms proposed by

Quesada and Grossmann (1995). The detailed linearization procedure of these terms
is detailed in the work by Majozi and Gouws (2009) and appears in the Appendix of
this chapter.
9.3.3.1.3 Mass Balance Constraints with Storage

The mathematical formulation for the case where there is reusable water storage
available is based on the superstructure given in Figure 9.4. This superstructure is
similar to the previous one. However, there is a central storage vessel available. The
total water flowing into the unit in this case is the sum of the freshwater, the water
recycled/reused from the other unit and water from the central storage. The water
leaving the unit is the sum of the water going to effluent, water going to storage and
the water being recycled/reused.
Constraints (9.1), (9.3) and (9.5) need to be modified to cater for the water from
storage, thus yielding (9.11), (9.12) and (9.13). The difference in the new constraints
is that water into and out of the unit includes water from and to storage as well.
Moreover, the inlet concentration is not only affected by the concentration from the
previous recycle as in the previous case, but also the concentration in the water from
the storage tank as stated in Constraints (9.13). Constraint (9.13) contains additional
(Freshwater)
mwf (sout,j, p – 1) (Reused downstream)
mwr (sout,j, sout,j΄, p)
sout,j΄
(Water in) (Water out)

mwout (sout,j, p) (Effluent)
mwin (sout,j, p – 1)
j mwe (sout,j, p)
cin (sout,j, c, p – 1) cout (sout,j, c, p)
msin (sout,j, p)
msin (sout,j΄, p)
mwr (sout,j΄, sout,j, p) sout,j΄ csin (c, p)
sout,j΄
(Recycled/reused water)
msout (sout,j, p) csout (c, p)
Σ msin (sout,j΄, p)
sout,j΄
(a)
(Fresh water)
j–1 j j+1 J
(Recycle/reuse)
(Effluent)
(b)
FIGURE 9.4 Superstructure for the mathematical formulation with reusable water storage
(Majozi and Gouws, 2009).
nonlinear terms due to storage. These additional nonlinear terms can be linearized
in the same manner as mentioned.
mwin ( sout , j , p ) = å mw ( s
sout , j
r out , j , sout , j , p ) + mw f ( sout , j , p ) + msout ( sout , j , p ) ,
mwout ( sout , j , p ) = å mw ( s
sout , j
r out , j , sout , j ¢ , p ) + mwe ( sout , j , p ) + msin ( sout , j , p ) ,
cin ( sout , j , c, p ) mwin ( sout , j , p ) = åmw ( s

sout , j ¢
r out , j , sout , j , p ) cout ( sout , j , c, p )
+ csout ( c, p ) msout ( sout , j , p ) , "j, j¢ Î J , sout , j Î Sout , j , p Î P, c Î C (9.13)

Besides the mass balance over a water-using operation, there also has to be mass
balance over the storage unit. Constraint (9.14) is the mass balance over the stor-
age unit. The water in the storage vessel at a certain time point is the difference
between the water flowing into and from the storage vessel and the water possibly
stored from the previous time point. Constraint (9.15) is a water mass balance at the
beginning of the time horizon. Constraints (9.16) and (9.17) are the definition of the
inlet and outlet concentration of the storage vessel, respectively. It should be noted
that Constraint (9.17) is based on the assumption that the contaminant concentration
inside the storage is the same as the concentration of the exit stream. Constraint
(9.18) is the initial contaminant concentration of the water in the storage at the begin-
ning of the time horizon, provided there is an initial amount of water in the storage
tank. Constraint (9.19) ensures that the water in the storage vessel does not exceed
the storage vessel’s capacity. Constraint (9.20) ensures that water flowing from stor-
age to unit j does not exceed unit j capacity.
qws ( p ) = qws ( p - 1) + å ms
sout , j
in ( sout , j , p ) - åmsout ( sout , j , p )
sout , j
qws ( p1 ) = Qwso - å ms out ( sout , j , p1 ) "j Î J , sout , j Î Sout , j (9.15)

sout , j
csin ( c, p ) =
å ( ms ( s , p ) c
sout , j
in out , j out ( sout , j , c, p ) )
å ms ( s sout , j
in out , j , p)
"j Î J , sout , j Î Sout , j , p Î P c ÎC (9.16)

csout ( c, p ) =
æ
qws ( p - 1) csout ( c, p - 1) + ç
è å sout , j
ö
msin ( sout , j , p ) ÷ csin ( c, p )
ø ,
qws ( p - 1) + å sout , j
msin ( sout , j , p )
csout ( c, p1 ) = CSout
o
( c ) "c Î C (9.18)
qws ( p ) £ QwsU "p Î P (9.19)
msin ( sout , j , p ) £ QwUs s ysin ( sout , j , p ) "j Î J , sout , j Î Sout , j , p Î P (9.20)
As one would notice, there are nonlinear terms in Equations 9.16 and 9.17. One of
the nonlinear terms can be eliminated by substituting Equations 9.16 and 9.17.
The nonlinear terms in the resulting equation are then linearized in a similar fashion
to that previously discussed. Due to the discontinuous nature of batch processes, the
mass balances are not enough to fully describe the system. Sequencing constraints
need to capture the discontinuous nature of the operation.
9.3.3.2 Sequencing and Scheduling Constraints

The sequencing and scheduling constraints pertaining to direct and indirect reuse/
recycle are presented in this section. The constraints are presented for both cases
where there is a central storage vessel and where there is none. The sequencing and
scheduling constraints can be divided into three groups. The first group comprises of
constraints pertaining to sequencing and scheduling of the washing operations. The
second group comprises of constraints pertaining to direct recycle/reuse of reusable
water, whilst the third group comprises of the constraints necessary for the schedul-
ing of storage vessel. These groups are discussed here.
9.3.3.2.1 Scheduling Constraints Associated with the Washing Operation

Each water-using operation in the time horizon has to be scheduled accordingly,
within the overall framework of operation scheduling. This is captured by Constraints
(9.21) through (9.25). Constraints (9.21) and (9.22) ensure that unit j is washed imme-
diately after a task that produced sout,j. Constraint (9.23)) is the duration constraint,
which defines the starting and the ending times of a washing operation. Constraint
(9.24) stipulates that the washing operation can only commence at time point p if the
task producing sout,j was active at the previous time point. Constraint (9.25) stipulates
that only one washing operation can take place in a unit at any given time point.
in ( out , j ) out ( out , j ) ( )

tw s , p ³ t s , p - H U 1 - yw ( sout , j , p ) , "j Î J , sout , j Î Sout , j , pÎP

(9.21)
in out , j ( )
tw ( s , p ) £ tout ( sout , j , p ) + H U 1 - yw ( sout , j , p ) , "j Î J , sout , j Î Sout , j , pÎP

(9.22)
twout ( sout , j , p ) = twin ( sout , j , p - 1) + tw ( sout , j ) yw ( sout , j , p - 1) ,
( )
yw ( sout , j , p ) = y s*in, j , p - 1 , "j Î J , sout , j Î Sout , j ,
s*in, j Î Sin*, j , s*in, j ® sout , j , p Î P, p > p1 (9.24)
å yw ( s out , j , p ) £ 1, "j Î J , sout , j Î Sout , j , p Î P1 (9.25)

sout , j
9.3.3.2.2 Water Recycle/Reuse Sequencing Constraints

Scheduling of the recycle/reuse streams is important because of the discontinuous
nature of the streams. Reusable water can only be directly recycled/reused if the unit
producing water and the unit receiving water finish operating and begin operating at
the same time, respectively. Constraint (9.26) stipulates that recycle/reuse between
units can only take place when the unit receiving the reusable water start operating
at that time point. The unit receiving the reusable water, however, does not neces-
sarily need the recycled water to operate, that is, it can operate independently of the
recycled water. Constraints (9.27) and (9.28) ensure that the time at which recycle/
reuse of water occurs coincides with the time that the water is produced. Constraints
(9.29) and (9.30) ensure that the starting time of the unit receiving water coincides
with the time at which that water is recycled/reused.
ywr ( sout , j , sout , j ¢ , p ) £ yw ( sout , j ¢ , p ) , "j, j¢ Î J , sout , j Î Sout , j , p Î P, (9.26)
(
twr ( sout , j , sout , j ¢ , p ) £ twout ( sout , j , p ) + H U 1 - ywr ( sout , j , sout , j ¢ , p ) , )
"j, j¢ Î J , sout , j Î Sout , j , p Î P, (9.27)

(
twr ( sout , j , sout , j ¢ , p ) ³ twout ( sout , j , p ) - H U 1 - ywr ( sout , j , sout , j ¢ , p ) , )
(
twr ( sout , j , sout , j ¢ , p ) £ twin ( sout , j ¢ , p ) + H U 1 - ywr ( sout , j , sout , j ¢ , p ) ,)
(
twr ( sout , j , sout , j ¢ , p ) ³ twin ( sout , j ¢ , p ) - H U 1 - ywr ( sout , j , sout , j ¢ , p ) ,)
9.3.3.2.3 Sequencing and Scheduling Constraints Associated with Storage

Constraints (9.31) and (9.32) ensure that the time at which reusable water goes to
storage coincides with the time at which it is produced. Furthermore, reusable water
can only be sent to storage at time point p if unit j has conducted a washing opera-
tion in the previous time point. This is captured by Constraint (9.33). However, the
fact that unit j has conducted a washing operation in the previous time point does not
mean that the resulting contaminated water needs to be sent to storage.
(
tsin ( sout , j , p ) ³ twout ( sout , j , p ) - H U 2 - ysout ( sout , j , p ) - yw ( sout , j , p ) )
(
tsin ( sout , j , p ) £ twout ( sout , j , p ) + H U 2 - ysout ( sout , j , p ) - yw ( sout , j , p ) )
ysin ( sout , j , p ) £ yw ( sout , j , p - 1) , "j Î J , sout , j Î Sout , j , p Î P, p > p1 (9.33)
Constraints (9.34) and (9.35) ensure that the time at which water is used in unit j
coincides with the time at which water is transferred from the storage to the unit.
Constraint (9.36) ensures that unit j is indeed active when the unit is using recycled
water from storage. The unit does not necessarily have to use the water from storage
when it operates. A unit will not use the water in storage if there is a violation of
the inlet concentration of the water into the unit or if there is simply no water in the
storage vessel.
(
tsout ( sout , j , p ) ³ twin ( sout , j , p ) - H U 2 - ysout ( sout , j , p ) - yw ( sout , j , p ) )
(
tsout ( sout , j , p ) £ twin ( sout , j , p ) + H U 2 - ysout ( sout , j , p ) - yw ( sout , j , p ) )
ysout ( sout , j , p ) £ yw ( sout , j , p ) "j Î J , sout , j Î Sout , j , p Î P (9.36)
Constraint (9.37) ensures that when water is transferred to a unit at time point p,
the time at which this happens is later in the time horizon than any previous time
water was transferred to other units at previous time points. Constraints (9.38) and
(9.39) ensure that the reusable water leaving the storage to different units at time
point p leaves at the same time.
(
tsout ( sout , j , p ) ³ tsout ( sout , j ¢ , p¢ ) - H U 2 - ysout ( sout , j , p ) - ysout ( sout , j ¢ , p¢ ) )

(
tsout ( sout , j , p ) ³ tsout ( sout , j ¢ , p ) - H U 2 - ysout ( sout , j , p ) - ysout ( sout , j ¢ , p ) )
(
tsout ( sout , j , p ) £ tsout ( sout , j ¢ , p ) + H U 2 - ysout ( sout , j , p ) - ysout ( sout , j ¢ , p ) )
Constraint (9.40) is similar to Constraint (9.37), but applies to inlet streams to

the storage tank. Constraints (9.41) and (9.42) ensure that the time at which reusable
water going to storage from unit j and unit j′ corresponds to the same time at the
same time point.
(
tsin ( sout , j , p ) ³ tsin ( sout , j ¢ , p¢ ) - H U 2 - ysin ( sout , j , p ) - ysin ( sout , j ¢ , p¢ ) )
(
tsin ( sout , j , p ) ³ tsin ( sout , j ¢ , p ) - H U 2 - ysin ( sout , j , p ) - ysin ( sout , j ¢ , p ) )
(
tsin ( sout , j , p ) £ tsin ( sout , j ¢ , p ) + H U 2 - ysin ( sout , j , p ) - ysin ( sout , j ¢ , p ) )
Constraint (9.43) ensures that the outlet time of water from storage tank at a time
point occurs later than the inlet time at the previous time point. Constraints (9.44)
and (9.45) ensure that at time point p the time at which reusable water is moved to
storage is the same as the time at which water leaves the storage.
(
tsout ( sout , j , p ) ³ tsin ( sout , j ¢ , p¢ ) - H U 2 - ysout ( sout , j , p ) - ysin ( sout , j ¢ , p¢ ) )
(
tsin ( sout , j , p ) ³ tsout ( sout , j ¢ , p ) - H U 2 - ysin ( sout , j , p ) - ysout ( sout , j ¢ , p ) )
(
tsin ( sout , j , p ) £ tsout ( sout , j ¢ , p ) + H U 2 - ysin ( sout , j ¢ , p ) - ysout ( sout , j ¢ , p ) )
9.3.3.3 Production Scheduling Constraints

As mentioned earlier, the intricate details of this sub-section are presented in another
publication Majozi and Zhu (2001), but it is briefly presented here for continuity
purposes. Emphasis will be on the new constraints added to cater for the concept of
cyclic scheduling.
9.3.3.3.1 Capacity Constraint
The capacity constraint, given by Constraint (9.46), ensures that the amount of mate-
rial processed in unit j at any time point p does not exceed the capacity of the unit.
( ) åm ( s
V jL y s*in, j , p £
sin , j
in in, j ( )
, p ) £ V jU y s*in, j , p ,
"j Î J , sin, j Î Sin, j , s*in, j Î Sin*, j , pÎP (9.46)
9.3.3.3.2 Material Balance Constraints

Constraints (9.47) through (9.53) are the mass balance constraints. These constraints
ensure that the conservation of mass around each unit and the mass of each state
involved in the production is maintained.
åm
sin , j
in ( sin, j , p - 1) = å mout ( sout , j , p )
sout , j
"j Î J , sin, j Î Sin, j , sout , j Î Sout , j , p Î P, p > p1 (9.47)
qs ( s, p1 ) = Qs0 ( s ) - min ( s, p1 ) , s ¹ product , "s Î S (9.48)
qs ( s, p ) = qs ( s, p - 1) - min ( s, p ) , s = feed , "s Î S, p Î P, p > p1 (9.49)
qs ( s, p ) = qs ( s, p - 1) + mout ( s, p ) - min ( s, p ) , s ¹ product , feed,
"s Î S, p Î P, p > p1 (9.50)
qs ( s, p1 ) = Qs0 ( s ) - d ( s, p1 ) , s = product , "s Î S (9.51)
qs ( s, p ) = qs ( s, p - 1) + mout ( s, p ) - d ( s, p ) , s = product, byproduct ,
"s Î S, p Î P, p > p1 (9.52)
qs ( s, p ) £ QU , "s Î S, p Î P (9.53)
9.3.3.3.3 Duration Constraints
The duration constraint is one of the most crucial constraints as it addresses the
intrinsic aspects of time in batch plants. Constraint (9.54) simply states that the time
at which a particular state is produced is dependent on the duration of the task that
produces the same state.
( ) ( )(
tout ( sout , j , p ) = tin s*in, j , p - 1 + t s*in, j y s*in, j , p - 1 , )
"j Î J , s*in, j Î Sin*, j , sout , j Î Sout , j , p Î P, p > p1 (9.54)
9.3.3.3.4 Feasibility and Time Horizon Constraints

A washing operation starts immediately after a task, that is, if there is a task taking
place in a unit at time point p, a washing operation will take place at the next time
point. This is captured by Constraint (9.24). Constraint (9.55) ensures that a washing
operation and a processing task do not coincide at the same time point. Constraint
(9.56) ensures that only one task takes place in a given unit at a given time point.
( )
yw ( sout , j , p ) + y s*in, j , p £ 1, "j Î J , sout , j Î Sout , j , s*in, j Î Sin*, j , p Î P (9.55)

å y ( s* , p ) £ 1,
*
in, j "j Î J , s*in, j Î Sin*, j , pÎP (9.56)
sin
,j

For cyclic scheduling, a task is allowed to cross the boundary of the cycle, as
mentioned, but its duration must not be more than the length of two cycles (2H).
Constraints (9.57) through (9.63) ensure that this is the case.
tin ( sin, j , p ) £ 2 H , "j Î J , sin, j Î Sin, j , p Î P (9.57)
tout ( sout , j , p ) £ 2 H , "j Î J , sout , j Î Sout , j , p Î P (9.58)
twin ( sout , j , p ) £ 2 H , "j Î J , sout , j Î Sout , j , p Î P (9.59)
twout ( sout , j , p ) £ 2 H , "j Î J , sout , j Î Sout , j , p Î P (9.60)
twr ( sout , j , sout , j ¢ , p ) £ 2 H , "j, j¢ Î J , sout , j Î Sout , j , p Î P (9.61)
tsin ( sout , j , p ) £ 2 H , "j Î J , sout , j Î Sout , j , p Î P (9.62)
tsout ( sout , j , p ) £ 2 H , "j Î J , sout , j Î Sout , j , p Î P (9.63)
9.3.3.3.5 Sequence Constraint within the Cycle

Before using a unit, all the tasks must be complete in the unit including their cor-
responding washing operations. Constraints (9.64) and (9.65) ensure that this is the
case. For a unit that is suitable to conduct more than one task, Constraint (9.66) is
necessary in the formulation.
( ) è
( )
tin s*in, j , p ³ twout ( sout , j , p¢ ) - H U æç 2 - y s*in, j , p - yw ( sout , j , p¢ - 1) ö÷ ,
ø
"j Î J , sin, j Î Sin, j , sout , j Î Sout , j , p, p¢ Î p, p¢ > p1, p ³ p¢ (9.64)
* *

( )
tin ( sin, j , p ) ³ tout ( sout , j , p¢ ) - H U çæ 2 - y ( sin, j , p ) - y s*in, j , p¢ - 1 ö÷ ,
è ø
" j Î J , s*
in, j Î S *
in, j , sout , j Î S out , j , p, p¢ Î p, p ¢ > p1 , p ³ p¢, s*in, j ® sout , j
(9.65)
å íîæçè t ( s* ) + tw ( s
ì
tin ( sin, j , p ) ³ in, j out , j ) ö÷ yw ( sout , j , p¢ - 1)üý , "j Î J , sin, j Î Sin, j ,
ø þ
sout , j
s*in, j Î Sin*, j sout , j Î Sout , j , p, p¢ Î P, p ³ p¢, p¢ ³ 2, s*in, j ® sout , j (9.66)
9.3.3.3.6 Mass Balance between Two Cycles

Constraint (9.67) ensures that the amount of intermediate material needed to start the
current cycle is stored at the last time point of the previous cycle in order to maintain
smooth operation without any accumulation or shortage of material between cycles.
This is the key constraint for cyclic scheduling.
qs ( s, p ) = Qs0 ( s ) , s ¹ product , feed, "s Î S, p Î P, p= P (9.67)
9.3.3.3.7 Cycle Length Constraint

Constraint (9.68) stipulates that the sum of all the durations of the tasks and their
corresponding washing operations in a processing unit must be less than the cycle
length.
å íîæçè t ( s* ) + tw ( s
ì
in, j out , j ) ö÷ yw ( sout , j , p - 1)üý £ H ,
sout , j
ø þ
"j Î J , s*in, j Î Sin*, j , p Î P, p > p1, s*in, j ® sout , j (9.68)

9.3.3.3.8 Sequence Constraints between Cycles

Constraints (9.69) and (9.70) express the relationship between the last task of the
previous cycle and the first task of the current cycle in the same unit j to maintain
continuity between cycles. Constraint (9.69) stipulates that the first task in unit j of
the new cycle must start after the completion of the last task and its corresponding
washing operation in the same unit j of the previous cycle. Constraint (9.70) holds for
tasks which do not need washing after its completion.
( ) è
( )
tin s*in, j , p1 ³ twout ( sout , j , p ) - H U æç 2 - y s*in, j , p1 - yw ( sout , j , p - 1) ö÷ - H ,
ø
* *
"j Î J , sin, j Î Sin, j , sout , j Î Sout , j , p = P (9.69)

(
tin ( sin, j , p1 ) ³ tout ( sout , j , p ) - H U æç 2 - y ( sin, j , p1 ) - y s*in, j , p - 1 ö÷ - H ,
è ø
)
"j Î J , s*in, j Î Sin*, j , sout , j Î Sout , j , p = P , s*in, j ® sout , j (9.70)
9.3.3.3.9 Objective Function
The objective function takes the form of Equation 9.71 for cyclic scheduling, that is,
performance index per unit time. The performance index can either be the maximi-
zation of profit or the minimization of effluent. This is dependent on the nature of the
given data for the problem.
Performance index
Z= (9.71)
H
A typical objective function for the maximization of profit can take a form of
Constraint (9.72), which is maximization of profit while taking into account freshwa-
ter and the effluent treatment cost.
Max Z
=
å å CP ( s ) d ( s, p ) - CFå å mw ( s
s p sout , j p
f out , j , p ) -CE å å mw ( s
sout , j p
e out , j , p)
H
(9.72)
The resulting mathematical formulation is an MINLP problem due to the pres-
ence of bilinear terms and a fractional term in the objective function. The linearization
technique presented in Appendix 9A cannot be applied to the objective function. It is
important to mention that for the presented model, global optimality cannot be guaran-
teed through the application of the technique presented in Appendix 9A. However, the
technique can be used to provide a feasible starting point prior to solving the model.
9.3.3.3.10 Solution Procedure
The solution procedure is adopted from Wu and Ierapetritou (2004). The optimum
cycle length is first determined from the methodology presented in this section with the
objective function taking the form of Constraint (9.72). The results dictate the amount
of intermediate products required to start the cyclic scheduling. With this data, the mini-
mum duration of the initial period is determined through makespan minimization prob-
lem. This is done to ensure the existence of a feasible schedule to provide the required
intermediates to start the cyclic scheduling period. The same problem is then solved
with the objective of profit maximization with the time horizon obtained from the solu-
tion of the makespan minimization problem. The intermediate products from the main
period, cyclic period, are consumed in the final period with the objective function being
profit maximization. With the initial, the cyclic and the final period known, a wastewa-
ter minimization problem with a longer time horizon can be solved. Two illustrative
examples are provided to demonstrate the effectiveness of the presented methodology.
9.4 ILLUSTRATIVE EXAMPLES
This section contains two literature examples to illustrate the applicability of the
presented methodology.
9.4.1 Example 1 (Kondili et al., 1993)

The illustrative example considered is a well-published multipurpose facility which
is commonly known as BATCH 1 in literature. It mainly consists of three chemical
reactions which take place in two common reactors, Reactor 1 and 2, as depicted in
Figure 9.5. In addition to the two common reactors, the flowsheet also entails the heater
and the separator, before and after the reactors, respectively, as shown in Figure 9.5.
Product 1
Raw 2 Reactor 1
Waste
Heater
Still
Raw 1
Product 2
Reactor 2
FIGURE 9.5 Flowsheet for BATCH 1 multipurpose plant.
The STN and SSN representations of the flowsheet are given in Figure 9.6a and b,

respectively. The data for this example are given in Table 9.1.
This example was adapted to wastewater minimization in the work by Majozi
and Gouws (2009). The philosophy is that the reactors need to be cleaned after each
reaction in order to remove contaminants that are formed as by-products, so as to
ensure product integrity. Data pertaining to cleaning tasks is shown in Table 9.2. It is
important to note that the streams are characterized by multiple contaminants, which
is a common occurrence in industry. The variation in performance in the reactors
could be ascribed to the different designs, which is indeed a common encounter in
practice. In addition to this data, it is known that the cost of freshwater is 2 c.u./kg
water (c.u. – cost units), whilst the effluent treatment cost is 3 c.u./kg.
The method by Majozi and Gouws (2009) in the absence of a central storage ves-
sel was applied to the illustrative example for different time horizons to demonstrate
the common drawback of techniques currently available in literature and the results
are given in Table 9.3. As mentioned earlier, when these techniques are applied to
problems with a longer time horizons they present longer computational times. The
model was formulated as an MINLP. The model was solved using GAMS 22.0, with
CONOPT and CPLEX being the selected solvers for NLP and the MIP problems,
respectively, in a DICOPT platform. All the results presented in this chapter were
obtained using a Pentium 4, 3.2 GHz processor with a 512 MB RAM.
The results in Table 9.3 indicate that for a time horizon of 48 h, the solution
could not be obtained. From these results, it is apparent that a methodology to solve
scheduling problems for wastewater minimization in batch processes for longer time
horizon is crucial, hence the proposed methodology.
The proposed methodology was applied to the illustrative example. From the
solution procedure presented in Section 9.3.3.3.10, the first step is to determine the
Product 1 (s7)
40% Int AB (s8)

40% 60%
Heating Reaction 2
Feed A Hot A 10%

(s1) (s5) 60% Impure E (s9)
Int BC 80% Separation
(s6)
90%
50%
Reaction 1 Reaction 3
Feed B 50% (s3) (s4) 20% Product 2

(s2) (s10)
(a) Feed C
(s1)
Feed A
(s2)
Feed B
Hot A (s5)
50% (s6) 40% (s7)
50% 60% 40%
Int BC Product 1
60%
(s3) (s9) 10%
80%
Feed C Int AB Feed 1
(s8) 90%
(s4) 20%
Product 2
(b) (s10)
FIGURE 9.6 (a) STN and (b) SSN for the literature example.
optimal cycle length. The strategy used by Wu and Ierapetritou (2004) to determine
the optimum cycle length was adopted. The time horizon range for determining the
cycle length was 3–15 h. Instead of considering the whole time horizon range, the time
horizon was subdivided into smaller intervals as indicated in Table 9.4. The objective
function for the illustrative example is the maximization of profit per unit time, as
given by Constraint (9.72), repeated for convenience.
MaxZ
=
åå s p
CP ( s ) d ( s, p ) - CF å å
sout , j p
mw f ( sout , j , p ) -CE
sout , j p å å
mwe ( sout , j , p )
H
The objective function comprises the product revenue, the freshwater and the waste-
water treatment cost. For the objective function presented to be maximized, fresh-
water and effluent treatment cost must be minimized by minimizing the amount of
freshwater used and wastewater generated per unit time.
TABLE 9.1
Scheduling Data for the Illustrative Example
Units Capacity Suitability Mean Processing Time (τ)
Still 200 Separation 1 for product 2, 2 for IntAB
States Storage Capacity Initial Amount Price
Feed A Unlimited Unlimited 0
Feed B Unlimited Unlimited 0
Feed C Unlimited Unlimited 0
Hot A 100 0 0
IntAB 200 0 0
IntBC 150 0 0
Impure E 200 0 0

9.4.1.1.1 Determination of the Optimum Cycle Length without Central Storage
The results for the determination of the optimum cycle length are presented in
Table 9.4. The results presented in Table 9.4 are for a case where there is no central
storage vessel for reusable water.
As shown in Table 9.4, the optimum cycle length was determined to be 5.75 h
corresponding to a profit per unit time of 2785.507 c.u. per cycle. The schedule for
the determined optimum cycle length is given in Figure 9.7. The values on top of the
blocks represent the amount of material processed in the unit for production pur-
poses. A washing operation starts after every material processing task, represented
as the shaded blocks on the diagram. The values in the curly brackets represent the
amount of freshwater supplied to the unit for the washing operation and the value
in the square brackets represents the amount of reused water from one operation to
another. In the absence of possible reuse and recycle, the amount of freshwater con-
sumption is 527.78 kg per cycle. Applying reuse and recycle concept, the amount of
effluent generated was reduced by 33.68% per cycle.
9.4.1.1.2 Determination of the Optimum Cycle Length with Central Storage

The cycle length for the illustrative example in the case where there is a central stor-
age vessel for reusable water was also determined. It is important to note that to truly
minimize wastewater generation in the presence of a storage vessel, the storage ves-
sel must be empty at the end of every cycle. To ensure that this is the case, Constraint
(9.73) was added to the model. The results for this case are given in Table 9.5.
qws ( p ) = 0 "p = P (9.73)

TABLE 9.2
Wastewater Minimization Data for the Illustrative Example
Maximum Concentration (g contaminant/kg water)
Contaminants
1 2 3
Reaction 1 (Reaction 1) Max. inlet 0.5 0.5 2.3
Max. outlet 1 0.9 3
Max. outlet 0.2 0.1 1.2
Max. outlet 0.075 0.2 1
Mass Load (g)
Contaminants
1 2 3
Reactor 2 15 24 358
Reactor 2 9 2 16
Reactor 2 22.5 45 36.5

Reactor 1 0.25 0.5 0.25
Reactor 2 0.3 0.25 0.25
TABLE 9.3
Results for BATCH 1 Using Method by Majozi and Gouws
(2009)
Time Horizon (h) Objective Function (c.u.) CPU Time (s)
10 11,537.5 0.31
13 19,587.5 2,044.44 (~0.57 h)
15 26,830.556 56,736.67 (~15.76 h)
48 — Intractable
TABLE 9.4
Result for BATCH 1 Using the Presented Methodology without Storage
Cycle Time Number of Objective Function Optimal Cycle
Range (h) Time Points Value (c.u.) Time (h) CPU Time (s)
3–6 7 2785.507 5.75 73.792
6–9 8 2669.444 6 106.634
9–12 10 2774.517 9.25 2056.567
12–15 11 2154.762 12 1353.127
Units
130
Separator
(120) (30) (120)
71.67 80 80
Reactor 2 Reaction 2 Reaction 3 Reaction 2
(80)
[120] [90.57]
48 50 43
32 28.67
Heater
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6

Time (h)
FIGURE 9.7 Schedule for the optimum cycle length without storage.
TABLE 9.5
Result for BATCH 1 Using the Developed Methodology with Storage
3–6 7 2907.246 5.75 274.233
6–9 9 2669.444 8.05 345.485
9–12 11 2847.954 9.25 1125.784
12–15 12 2579.407 12 2157.469
As shown in Table 9.5, the optimum cycle length for this case is also 5.75 h cor-
responding to an average profit of 2907.246 c.u. per cycle. The schedule for the deter-
mined cycle length in the presence of a central storage vessel is given in Figure 9.8.
The values on the diagram are as described in Figure 9.7. In the absence of pos-
sible indirect and direct water reuse/recycle, the amount of freshwater consumed is
518.89 kg per cycle. Applying reuse and recycle in the presence of a central storage
vessel, the amount of effluent generated was reduced by 59.53% per cycle.
[120]
Units Central storage [120]
130
Separator
80 80 71.67
[111.11] [40] [120]
41 50 50
60.67
Heater
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6

Time (h)
FIGURE 9.8 Schedule for the optimum cycle length without storage.
It is important to note the margin of difference in reduction/savings for the two

presented cases with and without central storage vessel is quite big. This is due to
the flexibility in the recycling and reusing time between the units, introduced by the
presence of the storage vessel. In the case where the storage vessel is absent, only
direct reuse is possible which decreases the degree of flexibility in the recycling and
reusing opportunities.
9.4.1.1.3 Wastewater Minimization without Central

Storage for 48 h Time Horizon
The illustrative example could not be solved for a time horizon of 48 h when the
methodology by Majozi and Gouws (2009) was used, as demonstrated in Table 9.3.
Therefore, the proposed methodology was applied for 48 h time horizon. The time
horizon was divided into three periods, as mentioned in the solution procedure in
Section 9.3.3.3.10. The optimum cycle length in the absence of central storage is as
determined in Table 9.4, 5.75 h, with the average profit of 2785.507 c.u. per cycle.
Six cycles were determined to provide enough time for the initial period, to pro-
duce the necessary intermediates, and the final period to wrap up the intermediates.
The CPU time for the initial period problem was 321.276 s. The duration of the
initial period was determined to be 10.35 h with an objective value of 6201.786 c.u.
using 11 time points. The schedule for the initial period is shown in Figure 9.9. The
freshwater usage without exploiting reuse and recycle opportunities is 1103.889 kg.
Exploiting these opportunities, the freshwater consumption and wastewater gener-
ated was reduced by 9.185%.
The duration of the final period was determined to be 3.15 h. This duration was
determined from the duration of the total time horizon and the duration of the
other two periods. The CPU time for the final period problem was 65.73 s. The
objective value for this period is 18237.5 c.u., which was obtained using 5 time
points. The schedule for the final period is shown in Figure 9.10. The freshwater
Units
5.71
Separator
(150) (120) (150) (30) (120)
80 48 80 35.71 48
Reactor 2 Reaction 1 Reaction 2 Reaction 1 Reaction 3 Reaction 2
[108.11] (130) (80) (142.5) (80)
[20]
39 50 50 50 50
32 20 20 60.67
Heater
0 1 2 3 4 5 6 7 8 9 10 11
Time (h)
FIGURE 9.9 Initial period without a central storage vessel.
Units
190
Separator
(30)
60
Reactor 2 Reaction 3
(142.5)
50
Heater
0 1 2 3
Time (h)
FIGURE 9.10 Final period without storage for a time horizon of 48 h.
consumption for this period was determined to be 172.5 kg. There were no reuse and
recycle opportunities for this period as shown in Figure 9.10.
The overall objective value representing the total profit, including the profit of the
initial and the final period, for the time horizon of 48 h is 41,152.33 c.u. Disregarding
the principle of reuse and recycle, the total freshwater consumption is 4443.069 kg.
Exploiting all the possible reuse and recycle opportunities, the total freshwater
consumption was reduced by 26.287%.
9.4.1.1.4 Wastewater Minimization with Central

The same problem of 48 h was solved in the presence of a central storage vessel.
The optimum cycle length in the presence of the storage vessel was determined to
be 5.75 h with an average profit of 2907.246 c.u. per cycle, as mentioned in Table 9.5.
The procedure used to determine the initial and the final period for the case without
central storage vessel was also used for the case with central storage vessel and the
results are presented in Table 9.6.
The corresponding Gantt charts for the initial and the final periods are given
Figures 9.11 and 9.12, respectively.
The overall objective value representing the total profit for the time horizon of
48 h in the presence of a central storage vessel is 41,990.7 c.u. The total freshwater
consumption is 4089.579 kg in the absence of both direct and indirect water reuse/
recycle. Exploiting all the available direct and indirect water reuse/recycle opportu-
nities, the total freshwater consumption was reduced by 47.088%.
To facilitate understanding of the presented results, the freshwater savings for the
time horizon of 48 h in the presence and absence of a central storage vessel are given
in Table 9.7. The savings for the cyclic period are given per cycle and six operating
cycles were obtained for both cases.
TABLE 9.6
Results for Initial and Final Period with Storage for the Time Horizon of 48 h
Objective Water Usage
Function without Storage Freshwater
Period Duration (h) Value (c.u.) (kg) CPU (s) Reduction
Initial 9.5 7,197.222 866.239 254.675 8.35%
Final 4 17,350 110 93.756 0%
Central storage [30.20]

Units
[30.20]
[150]
Separator 80
[150]
(150) (120) (30) (120) (30.20)
61 80 78 76.67 80
(88.89) (142.5) (142.5)
50 50 50 50
32 20 30.67 20
Heater
0 1 2 3 4 5 6 7 8 9 10
Time (h)
FIGURE 9.11 Initial period with storage.

Units
130 70
Separator
(30)
53.75
(80)
16.25
Heater
0 1 2 3 4
Time (h)
FIGURE 9.12 Final period with storage.
TABLE 9.7
Summary of Freshwater Savings for Example 1
With Storage Without Storage
Period Duration (h) Savings (%) Duration (h) Savings (%)
Initial 9.5 8.35 10.35 9.185
Cyclic 5.75 59.53 5.75 33.68
Final 4 0 3.15 0
Overall 48 47.008 48 26.287
9.4.2 Example 2 (Sundaramoorthy and K arimi, 2005)

This second illustrative example was presented in the work by Sundaramoorthy and
Karimi (2005). It is similar to the first example but considerably more complex. In
this example, two products are produced through six processing units and seven
processing tasks. It consists of three reactions which take place in two common reac-
tors and two heating tasks which take place in one common heater. In addition, it
involves one separation task and two mixing tasks which take place in two different
mixers. The STN and SSN representations of the example are given in Figures 9.13
and 9.14, respectively. The scheduling data for this example is given in Table 9.8.
This example is adapted to wastewater minimization in this chapter to illustrate the
applicability of the presented methodology to a more complex multipurpose problem.
The philosophy is that the reactors and the heater must be washed after each production
operation. This is to remove contaminants that are formed as by-products in the units,
so as to ensure product integrity. Data pertaining to cleaning tasks is given in Table 9.9.
The method by Majozi and Gouws (2009) could not solve the illustrative exam-
ple when a time horizon of 168 h was considered. The proposed methodology was
applied to the illustrative example with the same objective function considered for
Feed 3
Product 2
(s8)
75% (s13)
40%
Heating 1 Heating 1 Reaction 3
Feed 1 25% Int 6 60%
(s1) Int 1 Int 4 (s9) Int 7
(s3) Int 3 (s6) (s10) Product 1
50% (s5) 40% 40% (s12)
50% 40%
Reaction 2 Separation Mixing 1&2
50% 10% Int 5 20%
Int 2 (s7)
(s4)
Feed 4
(s11)
Reaction 1
Feed 2
(s2)
FIGURE 9.13 STN representation of the second illustrative example.
(s8) Feed 3
75%
40%
Feed 1 Int 6 Product 2
(s1) 25% (s9) 60% (s13)
(s3) Int 1 (s6) Int 4 Int 7 (s10)

50%
(s5) 40% 40%
50% 40%
Int 3 Int 5 Product 1
50% 10% (s7) (s12)
20%
(s4) Int 2 Feed 4

(s11)
Feed 2
(s2)
FIGURE 9.14 SSN representation of the second illustrative example.
the first illustrative example for a time horizon of 168 h. The results are presented in
the following sub-section.

9.4.2.1.1 Wastewater Minimization without Central
The procedure used in the first illustrative example for determining the optimal
cycle is used in this example. The time horizon range for determining the opti-
mal cycle length and the results are given in Table 9.10. The results presented in
Table 9.10 are for case where a central storage vessel is absent.
TABLE 9.8
Scheduling Data for the Second Illustrative Example
Capacity
Units Min Max Suitability Mean Processing Time
Heater 0 100 Heating 1,2 1, 1.5
Reactor 1 0 100 Reaction 1,2,3 2,1,2
Reactor 2 0 150 Reaction 1,2,4 2,1,2
Still 0 300 Separation 3
Mixer 1 20 200 Mixing 2
Mixer 2 20 200 Mixing 2
State Price Initial Amount Storage Capacity

Feed 1 0 10,000 Unlimited
Int1 0 0 100
Int2 0 0 100
Int3 0 0 300
Int4 0 50 150
Int5 0 50 150
Int6 0 0 150
Int7 0 0 150
Product 1 100 0 Unlimited
Product 2 100 0 Unlimited
The optimum cycle length was determined to be 6.45 h with an average profit
of 5167.557 c.u. per cycle. The schedule for the optimum cycle length for this case
is given in Figure 9.15. In the absence of possible reuse and recycle, the amount of
freshwater consumed is 501.675 kg per cycle. Applying the concept of reuse and
recycle, the amount of freshwater consumption for the determined cycle length was
reduced by 12.53%.
Considering the scheduling problem for the time horizon of 168 h, the proposed
methodology determines 24 cycles of operation. This leaves enough time for the ini-
tial and the final period. The results for the initial and the final schedules are given
in Table 9.11.
The corresponding schedules for the initial and final period are given in Figures
9.16 and 9.17, respectively.
The objective value corresponding to the total profit for the time horizon of
168 h is 179,386.596 c.u. The total freshwater consumption for this time horizon is
12,902.914 kg in the absence of water reuse and recycle. Exploiting all the avail-
able reuse and recycle opportunities, the total freshwater consumption is reduced
by 16.363%.
TABLE 9.9
Wastewater Minimization Data for the Second Illustrative Example
Maximum Concentration
(g Contaminant/kg Water)
Contaminant
1 2 3
Heating 1 (Heater) Max. inlet 0.5 0.35 1
Max. outlet 1.15 0.65 1.5
Heating 2 (Heater) Max. inlet 0.65 0.2 1.35
Max. outlet 1.5 0.7 2.5
Max. outlet 1 0.9 3
Max. outlet 0.2 0.1 1.2
Max. outlet 0.075 0.2 1
Mass Load (g)
Contaminant
1 2 3
Heating 1 Heater 7.5 13 20
Heating 2 Heater 11 15 25
Reactor 2 15 24 358
Reactor 2 9 2 16
Reactor 2 22.5 45 36.5
Heating 1 Heating 2
Heater 0.2 0.25
Reactor 1 0.25 0.5 0.25
Reactor 2 0.3 0.25 0.25
TABLE 9.10
Result for the Example without a Central Storage Vessel
3–6 8 4720.253 4.7 2,391.904
6–9 9 5167.557 6.45 5,042.156
9–12 10 3880.959 9.2 3,845.292
12–15 12 3008.294 12 29,239.092
Units
200 34.38
Mixer 2
Mixer 1
187.5
Separator
(18.57) (120)
150 150 87.5
(142.5) (88.89)
[48.03]
100 100
Reactor 1 [21.43] Reaction 2 Reaction 1
(21.43) [32.5]
(21.43) (26)
100 100 100 100
Heater Heating 2 Heating 2 Heating 1 Heating 2
0 1 2 3 4 5 6
Time (h)
FIGURE 9.15 Schedule for the optimum cycle length without central storage.
TABLE 9.11
Results for Initial and Final Period for 168 h Time Horizon
Objective
Function Water Usage Freshwater
Period Duration (h) Value (c.u.) without Storage (kg) CPU Time (s) Reduction
Initial 6.7 12,897.371 559.857 2923.03 8.00%
Final 6.5 42,467.857 302.857 166.439 7.08%
9.4.2.1.2 Wastewater Minimization with Central Storage for 168 h Time Horizon

The illustrative example for the time horizon of 168 h in the presence of central
storage for reusable water was also considered. The optimum cycle length for the
example for case where a central storage was presented was also determined and the
results are shown in Table 9.12.
The optimum cycle length for this case was also determined to be 6.45 h, corre-
sponding to the average production of 5326.345 c.u. per cycle as shown in Table 9.12.
Units
Mixer 2
118.21
Mixer 1
187.5
Separator
(136.92) (38.67)
82.07 91.30
Reactor 2 Reaction 2 Reaction 3
(142.5) [38.67] (88.89)
[28.26]
91.03 100 31.79
(26) [30] (38.67) (21.53)
91.03 100 (88.89) 50 91.30
0 1 2 3 4 5 6
Time (h)
FIGURE 9.16 Initial period without storage for a time horizon of 168 h.
Units
200
Mixer 2
118.75
Mixer 1
100
Separator
(120) (30) (30)
100 150 91.30
(80)
50
[37.5]
(21.43)
100 100
Heater Heating 2 Heating 2
0 1 2 3 4 5 6
Time (h)
FIGURE 9.17 Final period without storage for a time horizon of 168 h.
The schedule for the optimum cycle length for this case is given in Figure 9.18. The
freshwater consumption for the cycle with exploiting the concept of reuse and recycle
is 372.22 kg per cycle. Considering this concept, the freshwater consumption for the
cycle was reduced by 41.83% per cycle.
Considering the illustrative example for the time horizon of 168 h in the presence
of a central storage, the proposed methodology determines 24 cycles of operation
TABLE 9.12
Result for Example with a Central Storage Vessel Present
3–6 7 4885.904 4.7 2,019.64
6–9 9 5326.345 6.45 14,963.811
9–12 10 4006.984 9 8,401.495
12–15 11 3025.014 12 22,937.718
Central storage
Units
34.38
Mixer 2
[142.5]
200 [3.12]
Mixer 1 [139.38]
187.50
Separator
(30) (26.98)
150 150 87.5
[30] [23.88]
100 75 [120]
(2.05)
(21.43) (21.43) (2.12) [19.38]
100 100 93.75 100
0 1 2 3 4 5 6
Time (h)
FIGURE 9.18 Schedule for the optimum cycle length for the case with central storage.
which leaves enough time for the initial and the final period. The results for the ini-
tial and the final period are given in Table 9.13.
The corresponding Gantt charts for the initial and final period are given in Figures
9.19 and 9.20, respectively.
The objective value representing the total profit for the time horizon of 168 h is
184,731.05 c.u. The total freshwater consumption corresponding for this time horizon
in the absence of direct and indirect water reuse/recycle is 9799.845 kg. Exploiting
all possible direct and indirect water reuse/recycle, the total freshwater consumption
is reduced by 39.498%.
The freshwater savings results for the time horizon of 168 h in the presence and
absence of a central storage vessel are given in Table 9.14. The savings for the cyclic
period are given per cycle and 24 operating cycles were obtained for both cases.
TABLE 9.13
Result for the Initial and the Final Period in the Presence of a Central
Storage
Objective Water Usage
Function without Freshwater
Period Duration (h) Value (c.u.) Storage (kg) CPU Time (s) Reduction
Initial 8.85 14,111.27 592.635 2728.143 19.39%
Final 4.35 42,787.5 273.93 98.235 6.94%
Units
Mixer 2
125
Mixer 1
187.5
Separator
(120) (120)
150 125
(88.89) [92.05] (80)
95 93.75 100
(21.43) (26) (21.43)
[27.95]
100 87.5 100 100
0 1 2 3 4 5 6 7 8 9
Time (h)
FIGURE 9.19 Initial period with storage.
9.4.2.2 Remarks
From the results presented, it is evident that the developed methodology can be used
to reduce problems with longer time horizon to a smaller problem which can be
solved with ease. The first illustrative example could not be solved for 48 h using
the usual technique for wastewater minimization for short-term time horizon. With
the application of the proposed methodology, the 48 h time horizon in the absence of
central storage was reduced to six cycles with a time horizon of 5.75 h, initial period
with a duration of 10.35 h and a final period with a duration of 3.15 h. The second
illustrative example with the time horizon 168 h was reduced 24 h cycles, each cycle
with the length of 6.45 h, initial period of 6.7 h and final period of 6.5 h in the
absence of central storage vessel. The global optimality of the presented results can-
not be guaranteed since the developed methodology is nonlinear. The n onlinearity
of the objective function cannot be linearized with the technique presented in
Units
125 20
Mixer 2
195.63
Mixer 1
100
Separator
(30) [42.18]
100
(142.5) (61)
142.5 93.75
Central storage
(21.43)
[23.75]
100 [21.43]
Heater Heating 2
0 1 2 3 4
Time (h)
FIGURE 9.20 Final period with central storage present.
TABLE 9.14
Summary of Example 2 for 168 h Time Horizon
With Storage Without Storage
Period Duration (h) Savings (%) Duration (h) Savings (%)
Initial 8.85 19.39 6.7 8
Cyclic 6.45 41.83 6.45 16.985
Final 4.35 6.94 6.5 7.08
Overall 168 39.498 168 16.363
the Appendix. The developed technique is further applied to an industrial case study

in the following section to demonstrate the practical application.
9.5 INDUSTRIAL CASE STUDY

In this section, the developed methodology is applied to an industrial case study
to demonstrate practical application. The case study is taken from a pharmaceuti-
cals facility that produces four types of products, namely, shampoos, deodorants,
lotions and creams. All the products involve mixing and there are four mixing units
dedicated to each product, that is, one unit is suitable for mixing of one product. It
is mandatory to wash the unit after every mixing task to ensure product integrity.
TABLE 9.15
Data for the Case Study
Max. Outlet
Residue Conc (kg/kg
Mixer Product Mass (kg) Water) Mass Water (kg) Duration (h)
1 Shampoos 15 0.040 576.9 7
2 Deodorants 15 0.045 361.4 5.5
3 Lotions 30 0.050 697.6 11
4 Creams 70 0.060 1238.9 11
TABLE 9.16
Maximum Inlet Concentrations for Cleaning Operation for the Case Study
Shampoos Deodorants Lotions Creams
Mixer (kg/kg Water) (kg/kg Water) (kg/kg Water) (kg/kg Water)
1 0.014 0 0.007 0.0035
2 0.014 0.0035 0.007 0.007
3 0.014 0 0.007 0.0035
4 0.014 0 0.007 0.0035
The residue mass left in the unit after a mixing operation is given in Table 9.15. Also
given in Table 9.15 is the necessary information to conduct wastewater minimization
and the duration of each mixing operation. Because of different designs of the stir-
rers in the mixing vessels, mixing durations vary according to the vessel used. All
the mixers have a capacity of 2 tons.
The maximum inlet concentration of each mixer is given in Table 9.16. It is
important to note that the inlet concentration for the deodorant is zero in mixers 1, 3
and 4 (due to incompatibility of this residue with other residues). Thus, the reuse of
contaminated water resulting from unit 2 to any other unit is forbidden. The duration
of the washing operation is 30 min. A 10 ton central storage vessel for water reuse
is available.
In addition to the given data, the cost of freshwater is 0.2 c.u./kg of water whilst
the effluent treatment cost is 0.3 c.u./kg.
9.5.1 Results and Discussions

The methodology by Majozi and Gouws (2009) was applied to the industrial case
study to demonstrate the computational problem short-term scheduling techniques
for wastewater minimization have when they are applied to industrial scale prob-
lems. The same objective function used for the illustrative example was used for the
industrial case study, that is, maximization of profit per unit time. The results are
TABLE 9.17
Demonstration of Computational Problems
Time Horizon (h) Objective Function CPU (s)
24 41,758.062 10.625
48 105,167.217 5,724.550
72 229,163.475 50,786.576
168 — —
shown Table 9.17. The methodology encountered a computational difficulty for the
time horizon of 168 h, and the computational time for the time horizon of 72 h is
quite long.
9.5.1.1 Wastewater Minimization without Central Storage

for 168 h Time Horizon
The developed methodology was applied to the presented industrial case study. The
formulated problem was an MINLP for both cases, with and without central stor-
age tank. It is important to note that for the case study the raw materials needed for
mixing are readily available, hence the initial and the final period in this case are
not required. The same procedure followed for the illustrative example to determine
the optimal cycle length is followed for the case study. The time horizon range for
determining the optimal cycle length for the case where there is no central storage
vessel was chosen to be between 12 and 72 h, as shown in Table 9.18. The results for
industrial case study without storage tank are presented in Table 9.18.
As shown in Table 9.18, the optimum cycle length was determined to be 23 h
with a profit per unit time of 1740.910 c.u. per cycle. The Gantt chart corresponding
to the optimum cycle length showing both the production operations and the wash-
ing operations is depicted in Figure 9.21. The values in the curly brackets represent
the amount of freshwater supplied to the unit and the value in the square brackets
represents the amount of water reused from one operation to another. In the absence
of possible reuse and recycle, the amount of freshwater used is 5658.33 kg per cycle
which corresponds to a profit per unit time of 1718.623 c.u. per cycle. Applying the
TABLE 9.18
Results for the Case Study without Central Storage
Cycle Time Number of Time Objective Function Optimal Cycle
Range (h) Points Value (c.u.) Time (h) CPU Time (s)
12–24 7 1740.910 23 55.202
24–36 9 1730.886 24 71.538
36–48 11 1718.852 37.5 984.048
48–60 12 1663.541 48 771.453
60–72 14 1595.436 64 2036.576
Units
(744.82)
Mixer 4
[173.33] [173.33]
(600) (600)
Mixer 3
[248.52]
(156.29) [50.56] (206.85) (206.85)
Mixer 2
[126.48]
(375)
[126.48] (375) [126.48] (375)
Mixer 1
0 2 4 6 8 10 12 14 16 18 20 22 24 26
Time (h)
FIGURE 9.21 Optimum schedule for the case study without a storage vessel.
concept of reuse and recycle, the amount of effluent generated was reduced by 18.12%
per cycle. Consequently, the average profit per cycle is increased by 1.28%. The over-
all objective value which represents the total profit for the time horizon of 168 h is
12,716.21 c.u. This corresponds to a total freshwater consumption of 41,330.41 kg in
the absence of reuse and recycle. The reduction in the overall freshwater consump-
tion is the same as the reduction per cycle for the case study since the initial and the
final period are not present, as mentioned.
9.5.1.2 Wastewater Minimization with Central Storage

for 168 h Time Horizon
In a case where a central storage was present, the cycle length was also obtained to
be 23 h with the objective function of 1787.278 c.u. per cycle, as shown in Table 9.19.
In the absence of possible reuse and recycle, the amount of freshwater required is
5658.33 kg per cycle which corresponds to a profit per unit time of 1726.775 c.u. per
cycle. Applying the concept of reuse and recycle, the amount of effluent generated was
reduced by 45.40% per cycle and the schedule for these results is given in Figure 9.22.
Consequently, the average profit per cycle is increased by 3.39%. The objective value
TABLE 9.19
Results for the Case Study with Central Storage Present
Objective
Cycle Time Number of Function Value Optimal Cycle
Range (h) Time Points (c.u.) Time (h) CPU Time (s)
12–24 8 1787.278 23 64.53
24–36 9 1750.281 24 87.435
36–48 11 1734.563 37.5 882.361
48–60 13 1683.342 51.5 561.453
60–72 14 1695.436 60 1045.583
[To storage: 1238.9] [From storage: [To storage:

Units 1166.61] 1166.61]
(1238.9)
Mixer 4
[From storage: [To storage: From storage: 600
697.6] 697.6]
Mixer 3
[From storage: 333.33]
(333.33) (333.33)
Mixer 2 [From storage:
305.57]
(375) (375) (69.43)
Mixer 1
0 2 4 6 8 10 12 14 16 18 20 22 24 26
Time (h)
FIGURE 9.22 Optimum schedule with the presence of a central storage vessel.
for the time horizon of 168 h in the presence of a central storage vessel is 13,054.9 c.u.
This corresponds to a total freshwater consumption of 41,330.41 kg in the absence of
direct and indirect water reuse/recycle. The reduction in the overall freshwater con-
sumption is the same as the reduction per cycle for the case study since the initial and
the final period are not present, as mentioned earlier.
For the case of 168 h time horizon, the short-term scheduling methodology could
not find a solution. Applying the developed methodology, the time horizon of 168 h
can be reduced to 7.304 (≈7 cycles) with the length of 23 h. The CPU time for the
optimum cycle length is 64.53 s. The reduction in freshwater usage and wastewater
generation for the determined cycle will be 18.12% per cycle in the absence of central
storage vessel and 45.40% per cycle in the presence of the central storage vessel.
From the presented results of the case study, the developed methodology proves to
be capable of reducing an industrial scale problem to a problem with a smaller time
horizon. As mentioned for the illustrative example, the global optimality of the pre-
sented results cannot be proven due to the nonlinearities present in the formulation
of the methodology, particularly in the objective function.
9.6 CONCLUSION
A long-term scheduling methodology for wastewater minimization in multipurpose
batch facilities is presented in this chapter. The main advantage of the presented
methodology is the ability to reduce industrial scale problems to a problem with
a smaller time horizon, which can be solved within reasonable CPU time. This is
achieved through the exploitation of cyclic scheduling concepts. The concept of
water reuse and recycle is used to minimize wastewater generation. The proposed
methodology optimises both the production schedule and wastewater generation
simultaneously. It is applicable to operations with streams characterized by mul-
tiple contaminants, which is more prevalent in industry. The application of the
methodology to an industrial case study has proven its effectiveness. A time horizon
of 168 h of the case study was reduced to seven cycles with the length of 23 h. The
total freshwater consumption for the time horizon of 168 h was reduced by 18.12%
in the absence of central storage vessel and 45.40% in the presence of the central
storage vessel.
9A APPENDIX
9A.1 Linearization
To demonstrate the linearization technique presented in the work by Majozi and
Gouws (2009), Constraints (9.4) and (9.5) in the presented model will be used.
The two constraints contain bilinear terms. These nonlinear terms can be linearized
according to the linearization of bilinear terms proposed by Quesada and Grossmann
(1995). The linearization is given below.
Let
cin ( sout , j , c, p ) mwin ( sout , j , p ) = G1 ( sout , j , c, p )
cout ( sout , j , c, p ) mwout ( sout , j , p ) = G 2 ( sout , j , c, p )
cout ( sout , j ¢ , c, p ) mwr ( sout , j ¢ , sout , j , p ) = G3 ( sout , j ¢ , sout , j , c, p )
with each variable having the following bounds
0 £ cin ( sout , j , c, p ) £ CinU ( sout , j , c )
0 £ mwin ( sout , j , p ) £ MwU ( sout , j )
0 £ cout ( sout , j , c, p ) £ Cout

U
( sout , j , c )
0 £ mwout ( sout , j , p ) £ MwU ( sout , j )
0 £ mwr ( sout , j ¢ , sout , j , p ) £ MwU ( sout , j )
then the following constraints are true for G1:
G1 ( sout , j , c, p ) ³ 0, "j Î J , sout , j Î Sout , j , p Î P, c Î C (9A.1)
G1 ( sout , j , c, p ) ³ MwU ( sout , j ) cin ( sout , j , c, p ) + CinU ( sout , j , c ) mwin ( sout , j , p )
- Mw ( sout , j ) cin ( sout , j , c, p ) , "j Î J , sout , j Î Sout , j , p Î P, c Î C

U
(9A.2)
G1 ( sout , j , c, p ) ³ Mw ( sout , j ) cin ( sout , j , c, p ) , "j Î J , sout , j Î Sout , j , p Î P, c Î C

U
(9A.3)
G1 ( sout , j , c, p ) £ mwin ( sout , j , p ) Cin ( sout , j , c ) , "j Î J , sout , j Î Sout , j , p Î P, c Î C

U
(9A.4)
and the following constraints are true for G 2:
G 2 ( sout , j , c, p ) ³ 0, "j Î J , sout , j Î Sout , j , p Î P, c Î C (9A.5)
G 2 ( sout , j , c, p ) ³ MwU ( sout , j ) cout ( sout , j , c, p ) + Cout

U
( sout , j , c ) mwin ( sout , j , p )
- MwU ( sout , j ) Cout
U
( sout , j , c ) , "j Î J , sout , j Î Sout , j , p Î P, c Î C (9A.6)

G 2 ( sout , j , c, p ) £ mwout ( sout , j , p ) C U
out ( sout , j , c ) , "j Î J , sout , j Î S out , j , p Î P , c Î C
(9A.7)
and the following constraints are true for G3:
G3 ( sout , j ¢ , sout , j , c, p ) ³ 0, "j Î J , sout , j Î Sout , j , p Î P, c Î C (9A.8)
G3 ( sout , j ¢ , sout , j , c, p ) ³ MwU ( sout , j ) cout ( sout , j ¢ , c, p ) + Cout

U
( sout , j¢ , c ) mwr ( sout , j¢ , sout , j , p )
- MwU ( sout , j ) Cout
U
( sout , j , c ) , "j, j¢ Î J , sout , j Î Sout , j , p Î P, c Î C (9A.9)
G3 ( sout , j , sout , j , c, p ) £ MwU ( sout , j ) cout ( sout , j ¢ , c, p ) ,

G3 ( sout , j ¢ , sout , j , c, p ) £ mwr ( sout , j , p ) Cout

U
( sout , j¢ , c ) ,
Substituting these linearized variables into Constraints (9.4) and (9.5) gives
Constraints (9A.12) and (9A.13):
G 2 ( sout , j , c, p ) = G1 ( sout , j , c, p - 1) + M ( sout , j , c ) yw ( sout , j , p -11) ,

"j Î J , sout , j Î Sout , j , p Î P, c Î C (9A.12)
G1 ( sout , j , c, p ) = å G3 ( sout , j ¢ , sout , j , c, p ) , "j, j¢ Î J , sout , j Î Sout , j , p Î P, c Î C

sout , j¢
(9A.13)
Exact MINLP is relaxed using the relaxation-linearization

technique of Quesada and Grossman (1995)
Resulting MILP is them solved
Solution from MILP used as starting

solution for exact MINLP
MILP objective = MINLP objective MILP objective ≠ MINLP objective

Globally optimal Locally optimal
FIGURE 9A.1 Solution procedure from Majozi and Gouws (2009).
The same linearization procedure can be followed to linearize any source of non-
linearity in the model with the exception of the nonlinearity present in the objective
function.
The solution procedure presented in Figure 9A.1 is followed when the nonlinear
terms in the model are linearized. It is important to mention that for the presented
model, global optimality cannot be guaranteed through the application of this solu-
tion procedure. However, this procedure can be used to provide a feasible starting
point prior to solving the exact model.
REFERENCES
uling of batch operations I. MILP formulation. Computers and Chemical Engineering.
17(2), 211–227.
Majozi, T., Gouws, J., 2009. A mathematical optimization approach for wastewater minimi-
zation in multiple contaminant batch plants. Computers and Chemical Engineering.
33, 1826–1840.
Majozi, T., Zhu, X.X., 2001. A novel continuous time MILP formulation for multipurpose
batch plants. 1. Short-term scheduling. Industrial and Engineering Chemical Research.
40(25), 5935–5949.
multi-component flows. Computers and Chemical Engineering. 19, 1219.
Shah, N., Pantelides, C.C., Sargent, R.W.H., 1993. Optimal periodic scheduling of multipur-
pose batch plants. Annals of the Operation Research. 42(1), 193–228.
Sundaramoorthy, A., Karimi, I.A., 2005. A simple better slot-based continuous-time formu-
Science. 60, 2679–2702.
Wu, D., Ierapetritou, M., 2004. Cyclic short-term scheduling of multiproduct batch plants
using continuous-time representation. Computers and Chemical Engineering. 28,
2271–2286.
10 Long-Term Heat
Integration in
Multipurpose Batch
Plants Using Heat Storage
10.1 INTRODUCTION
Most scheduling methods are limited to the short-term scheduling case, and solution
of problems over long time horizons may prove challenging or impossible with these
methods. Inclusion of additional considerations such as heat integration further com-
plicates the problem. A model for the simultaneous optimization of the schedule and
energy usage in heat-integrated multipurpose batch plants operated over long time
horizons is presented in this chapter. The method uses a cyclic scheduling solution
procedure. The proposed model includes indirect heat integration via heat storage,
rather than just direct heat integration. This has largely not been considered in long-
term heat integration models in current literature. Both the heat storage size and
initial heat storage temperature are also optimized. The solution obtained over 24 h
using the proposed cyclic scheduling model with direct heat integration for a multi-
purpose example is compared to the result obtained from the direct solution and an
error of less than 6% is achieved.
In a batch process, discrete tasks follow a specific sequence or recipe, whereby
raw materials are transformed to final products. The recipe includes the amounts
of materials to be processed as well as the processing times of the various tasks
(Majozi, 2010). Batch processes are commonly used for the manufacture of products
required in small quantities or for specialty and complex products of high value.
Approximately half of all production facilities make use of batch processes (Stoltze
et al., 1995). Batch plants are also popular due to their flexible and adaptable nature,
which is particularly important in volatile markets.
Heating and cooling are required in most processing facilities. The objective of
heat integration is to optimize the use of energy. This becomes a possibility if a
process includes both heat generating and heat consuming operations. Heat integra-
tion may be accomplished in two ways in a batch process. If the operating schedule
allows an overlap in time of hot and cold units, direct heat integration may be used if
both units are active. However, due to the time-dependent nature of batch processes
it may be necessary to store heat from a hot unit using an intermediate heat storage
275
fluid and reuse this heat at a later time when it is required, resulting in indirect heat
integration. The inclusion of heat storage instead of only direct heat integration leads
to more flexibility in the process and therefore improved energy usage.
Heat integration in batch plants has in the past been largely disregarded as utility
requirements were considered less significant due to the smaller scale of batch opera-
tions compared to continuous plants. However, utility requirements in some batch
plants, such as in the food and drink industries, dairies, meat processing facilities,
biochemical plants and agrochemical facilities, contribute largely to their overall
costs. Knopf et al. (1982) analysed a non-continuous cottage cheese process and
concluded that capital costs were far outweighed by the energy costs in the plant.
Boyadjiev et al. (1996) applied a sequential analysis for direct heat integration in an
existing antibiotics plant. The effluent cooling water was used as makeup for the hot
water, which reduced the freshwater consumption and wastewater generated. The
overall energy costs of the plant decreased by 39%. Rašković et al. (2010) identified
an opportunity for significant waste heat recovery in a yeast and ethanol produc-
tion plant using pinch analysis. Majozi (2009) combined both direct and indirect
heat integration with scheduling and applied the model to an agrochemical facility.
Savings of more than 75% in external steam consumption were achieved.
Optimal scheduling and equipment use and decreased energy requirements can
have a significant effect on the efficiency and revenue of a batch plant. Minimizing
energy usage is also influenced by the need to comply with stricter environmental
regulations, reduce the effects of higher energy prices and conserve scarce environ-
mental resources.
Early techniques for heat integration in batch processes were based on pinch anal-
ysis, originally developed for heat integration in continuous plants at steady state.
Variations of the technique for batch processes still appear in literature (Kemp, 1990;
Wang and Smith, 1995; Foo et al., 2008; Wang et al., 2014). However, their reliance
on a predefined schedule and averaging energy requirements over time intervals may
lead to suboptimal results.
For many mathematical heat integration techniques presented in published lit-
erature, the processing schedule also tends to be predefined, leading to suboptimal
results. Some methods may include heuristic approaches which also cannot guar-
antee optimality (Vaselenak et al., 1986; Vaklieva-Bancheva et al., 1996; Chen and
Ciou, 2008; Halim and Srinivasan, 2009).
Methods specifically applicable to batch plants are required as they capture the
essence of time. Scheduling and heat integration may be combined into a single
problem for a more optimal solution (Papageorgiou et al., 1994; Pinto et al., 2003;
Majozi, 2006, 2009; Chen and Chang, 2009; Stamp and Majozi, 2011; Seid and
Majozi, 2014; Lee et al., 2015). However, models may then need to be simplified in
order to avoid excessive solution times.
Most heat integration methods are limited to the short-term case and solution
of problems over long time horizons may prove challenging or impossible with
these current methods. The model of Stamp and Majozi (2011) has been extended in
this chapter for the simultaneous optimization of the schedule and energy usage in
multipurpose batch plants operated over long time horizons. The proposed model
Long-Term Heat Integration in Multipurpose Batch Plants 277
uses the cyclic scheduling concepts and solution procedure presented by Wu and
Ierapetritou (2004). These concepts were also used in a similar technique for the
solution of long-term wastewater minimization problems in multipurpose batch
plants by Nonyane and Majozi (2012). Rather than just considering direct heat inte-
gration, the proposed method also includes the concept of indirect heat integra-
tion via heat storage, which has not been considered in long-term heat integration
models in current literature. The initial heat storage temperature and heat storage
capacity are also optimized.
A brief discussion on the concepts used in cyclic scheduling is given in the next
section. The problem statement and objectives are then presented. The mathematical
model and solution procedure are then discussed. The model is applied to a literature
example and an industrial case study, and conclusions are then drawn.
10.3 CYCLIC SCHEDULING CONCEPTS

Shah et al. (1993) proposed an axiom which stated that, in the case where the time
horizon of interest is much longer than the individual task durations, there exists a
sub-schedule of much shorter duration which, when repeated, can achieve production
close to the optimal production achievable for the original long time horizon. This
was the basis for the development of cyclic scheduling techniques. It involves obtain-
ing an optimal cyclic schedule which is then repeated and this therefore reduces the
problem size. Although cyclic scheduling sacrifices some accuracy when compared
to the direct solution of the scheduling problem, this is balanced by the easier and
quicker solution of a less complex scheduling problem. Cyclic scheduling methods
may be preferred in order to obtain a feasible solution where a solution with short-
term methods might not be possible or practical otherwise.
In the cyclic scheduling approach, the length of the cycle is an optimization vari-
able. The unit schedule is then also optimized. For the unit schedule to be repeat-
able, a certain amount of each intermediate state must be available to start the cyclic
period. It is also required that the intermediates be produced and stored at the end of
the period to be available for the next period.
In cyclic scheduling, each unit has an individual cycle with cycle time equal to
the unit period cycle duration. The units therefore do not need to share the same
starting and ending times. This leads to the concept of the “wrapping around” of
tasks as proposed by Shah et al. (1993) which allows tasks to cross a unit schedule
boundary for improved unit utilization. This concept is demonstrated in Figures 10.1
and 10.2. The cyclic scheduling model of Wu and Ierapetritou (2004) was based on
the previous discrete time model of Shah et al. (1993) and uses the continuous time
scheduling constraints of Ierapetritou and Floudas (1998) based on the state task
network (STN) representation.
The overall long time horizon is divided into three periods (Wu and Ierapetritou,
2004). The optimal unit schedule with optimal cycle length is solved for first, where
the objective is to maximize the average profit over the cycle. This is the main period
which will be repeated. Sub-ranges for the cycle time may be used rather than con-
sidering the entire cycle time at once. As the cycle time is a variable, this makes it
Units
Time
FIGURE 10.1 Representation of a cyclic schedule. (From Wu, D. and Ierapetritou, M.,
Comp. Chem. Eng., 28, 2271, 2004.)
Units Units
(a) Time (b) Time
FIGURE 10.2 (a) Unit schedule. (b) With tasks wrapping around. (From Wu, D. and
Ierapetritou, M., Comp. Chem. Eng., 28, 2271, 2004.)
easier to determine the required number of time points. These sub-problems may
also be solved in parallel and generate alternative schedules with different cycle
lengths which may be useful depending on the application.
The initial period, the start-up phase, is then solved. The results from this period
ensure that the intermediates required to start the cyclic period are produced. The
first objective is to minimize the makespan which ensures there is a feasible solu-
tion and the required intermediates are produced in the shortest possible time. The
problem is then solved to maximize profit over the time horizon obtained from the
makespan minimization problem.
A profit maximization problem is then solved for the final period, which is the
remainder of the long time horizon. The final period, shut-down phase, uses up all
the remaining intermediates to form products. The initial period and final period
profit maximization problems are both mixed integer linear programming (MILP)
short-term scheduling problems with known time horizons. The combined lengths of
the three different periods add up to the overall long time horizon.
This method can easily be applied to any long time horizon as the initial period
is not affected by the overall length of the time horizon considered. The number
of cycles can be increased or decreased and the final period solved for again if the
overall time horizon changes. The solution procedure and computational complexity
are also not affected by a change in the overall time horizon. The method provides
another option for the solution of long-term scheduling problems rather than solving
them directly with short-term scheduling methods and its usefulness will therefore
depend on the application.
10.4 PROBLEM STATEMENT AND OBJECTIVES

For the problem addressed in this chapter, the following is given.
10.4.1 Scheduling Data
1. Production recipe for each product.
2. Available units and their capacities.
3. Maximum storage capacity for each material.
4. Task durations.
5. Time horizon of interest.
6. Costs of raw materials.
7. Selling price of final products.
10.4.2 Heat Integration Data

1. Hot duties for tasks require heating and cold duties for tasks that require
cooling.
2. Operating temperatures of heat sources and heat sinks.
3. Minimum allowable temperature differences.
4. Heat capacities of materials.
5. Costs of hot and cold utilities.
6. Design limits on heat storage.
The objectives are then to determine the optimum cycle time and cyclic operating
schedule as well as the schedules for the initial and final periods in order to optimize
the profit over the given long time horizon.
10.5 MATHEMATICAL MODEL
The mathematical formulation comprises the following sets, continuous variables,
binary variables and parameters.
10.5.1 Sets
J { j| j = processing unit}
J c { jc| jc = processing unit which may conduct tasks requiring heating} Í J
J h { jh| jh = processing unit which may conduct tasks requiring cooling} Í J
S {s|s = any state}
Sin, j {sin, j|sin, j = input state to a processing unit} Í S
Sin*, j {s*in, j s*in, j = effective state into a processing unit} Í Sin, j

|

Sinsc, j {sinsc, j|sinsc, j = task consuming state s} Í Sin, j
Sinsp, j {sinsp, j|sinsp, j = task which produces state s, other than a product} Í Sin, j
S p {s p|s p = a state which is a final product}
U {u|u = heat storage unit}
10.5.2 Continuous Variables
B ( sin, j , p ) Batch size, either fixed or variable
CL ( sin, jh , p ) Cooling load for hot state
•
CL ( sin, jh , p ) Cooling load per time, for hot state
cw( sin, jh , p ) External cooling required by unit jh conducting the task corre-
sponding to state sin, jh at time point p
dur ( sin, j , p ) Duration of task, dependent on batch size
extra _ cw ( u ) Additional cooling required in heat storage vessel
extra _ st ( u ) Additional heating required in heat storage vessel
Gcw Glover transformation variable
Gst Glover transformation variable
H Single cycle length for cyclic scheduling problem
HL ( sin, jc , p ) Heating load for cold state
•
HL ( sin, jc , p ) Heating load per time, for cold state
K cw Reformulation–linearization variable
K st Reformulation–linearization variable
mu ( sin, j , p ) Amount of material processed in a unit at time point p
Q ( sin, j , u, p ) Heat exchanged with heat storage unit u at time point p
q ( sin, jh , sin, jc , p ) Amount of heat exchanged during direct heat integration
Qs0 ( s ) Initial amount of intermediate state s stored (cyclic scheduling)
st ( sin, jc , p ) External heating required by unit jc conducting the task corre-
sponding to state sin, jc at time point p
T0 ( u, p ) Initial temperature in heat storage unit u at time point p
T f ( u, p ) Final temperature in heat storage unit u at time point p
DTcw Temperature change required in heat storage vessel to return to
the starting temperature, when additional cooling is required
DTst Temperature change required in heat storage vessel to return to
the starting temperature, when additional heating is required
Tstart Temperature of heat storage at beginning of cycle
Tend Temperature of heat storage at end of cycle
t0 ( sin, j , u, p ) Time at which heat storage unit u commences activity
t f ( sin, j , u, p ) Time at which heat storage unit u ends activity
tu ( sin, j , p ) Time at which a task starts or state is used in unit j
t p ( sin, j , p ) Time at which a task ends in unit j

W (u ) Capacity of heat storage unit u
10.5.3 Binary Variables
tt ( j, p ) Binary variable associated with usage of state produced by unit j
at time point p
x ( sin, jc , sin, jh , p ) Binary variable associated with heat integration between unit jc
conducting the task corresponding to state sin, jc and unit jh con-
ducting the task corresponding to state sin, jh , at time point p
xcw Binary variable signifying that extra cooling is required in the
heat storage vessel
xst Binary variable signifying that extra heating is required in the
heat storage vessel
y ( sin, j , p ) Binary variable associated with usage of state s in unit j at time
point p
z ( sin, j , u, p ) Binary variable associated with heat integration between unit j
conducting the task corresponding to state sin, j with heat storage
unit u at time point p
10.5.4 Parameters
b Variable coefficient of processing time
CL ( sin, jh ) Fixed cooling load for hot state
cp fluid Specific heat capacity of heat storage fluid
cpstate ( sin, j ) Specific heat capacity of state
Cost _ cw Cost of cooling water
Cost _ st Cost of steam
CP(s) Selling price of product s, s = product
HL ( sin, jc ) Fixed heating load for cold state
H U Upper bound for cycle length
MM Any large number
Q max ( sin, j ) Maximum possible heating load or cooling load for a cold state or hot
state, respectively
T ( sin, j ) Operating temperature for processing unit j conducting the task
corresponding to state sin, j, for constant temperature processes
Tin ( sin, j ) Inlet temperature for state sin, j
Tout ( sin, j ) Outlet temperature for state sin, j
TL Lower bound for heat storage temperature
TU Upper bound for heat storage temperature
DT L Lower bound for temperature difference in heat storage vessel
DT U Upper bound for temperature difference in heat storage vessel

DT min Minimum allowable thermal driving force
t ( sin, j ) Duration of the task corresponding to state sin, j conducted in unit j
WL Lower bound for heat storage capacity
WU Upper bound for heat storage capacity
The mathematical formulation consists of the scheduling constraints, with modifica-

tions for cyclic scheduling, as well as the heat integration constraints of Stamp and
Majozi (2011). The model uses the scheduling framework of Seid and Majozi (2012),
which is based on the state sequence network (SSN) recipe representation and an
uneven discretization of the time horizon (Majozi and Zhu, 2001). This scheduling
framework has proven to result in fewer time points required compared to models
based on other representations, with improvements in computational time as well as
objective function in short-term scheduling problems. It also provides for the correct
sequencing of tasks for the fixed intermediate storage (FIS) policy.
The mathematical model is based on the superstructure in Figure 10.3 and
includes opportunities for both direct and indirect heat integration. Direct heat inte-
gration refers to the use of heat generated from a processing unit to supply a process-
ing unit requiring heat without the use of heat storage. Indirect heat integration refers
to the use of heat previously stored in a heat storage vessel to supply a processing
unit requiring heat. Processing units may also operate in standalone mode, using
only external utilities. This may be required for control reasons or when thermal
driving forces or time do not allow for heat integration. If either direct or indirect
heat integration is not sufficient to satisfy the required duty, external utilities may
make up for any deficit.
Constraints (10.1) and (10.2) are active simultaneously and ensure that one hot
unit will be integrated with one cold unit when direct heat integration takes place in
order to simplify operation of the process. Also, if two units are to be heat integrated
at a given time point, they must both be active at that time point. However, if a unit
is active, it may operate in either integrated or standalone mode.
Cold utility jh
cw(sin,jh, p)
Heat integration
Q(sin,jh, sin,jc, p)
Heat jc
Hot utility
To storage storage From storage st(sin,j , p)
Q(sin,jh, u, p) Q(sin,jc, u, p) c
FIGURE 10.3 Superstructure for mathematical formulation when units perform heating/
cooling tasks.
åx ( s in, jc , sin, jh , p ) £ y ( sin, jh , p ) , "p Î P, sin, jh Î Sin, j (10.1)

sin , jc
åx ( s in, jc , sin, jh , p ) £ y ( sin, jc , p ) , "p Î P, sin, jc Î Sin, j (10.2)

sin , jh
Constraints (10.3) and (10.4) ensure that heat integration between a unit and heat
storage may occur only if the unit is active at that time point. However, if a unit is
active, it will not necessarily integrate with heat storage.
z ( sin, jc , u, p ) £ y ( sin, jc , p ) , "p Î P, sin, jc Î Sin, j , u Î U (10.3)
z ( sin, jh , u, p ) £ y ( sin, jh , p ) , "p Î P, sin, jh Î Sin, j , u Î U (10.4)
Constraint (10.5) ensures that heat storage is heat integrated with either one hot
unit or one cold unit at any point in time. This is to simplify and improve operational
efficiency in the plant.
åz ( s in, jc , u, p ) + åz ( s in, jh , u, p ) £ 1, "p Î P, u Î U (10.5)

sin , jc sin , jh
Constraints (10.6) and (10.7) ensure that a unit cannot simultaneously undergo
direct and indirect heat integration. This condition simplifies the operation of the
process.
åx ( s in, jc , sin, jh , p ) + z ( sin, jc , u, p ) £ 1, "p Î P, sin, jc Î Sin, j , u Î U (10.6)

sin , jh
åx ( s in, jc , sin, jh , p ) + z ( sin, jh , u, p ) £ 1, "p Î P, sin, jh Î Sin, j , u Î U (10.7)

sin , jc
Constraints (10.8) and (10.9) quantify the amount of heat received from or
transferred to the heat storage unit, respectively. There will be no heat received
or transferred if the binary variable signifying usage of the heat storage vessel,
z ( sin, j , u, p ), is zero. These constraints are active over the entire time horizon, where
p is the current time point and p -1 is the previous time point.
( )
Q ( sin, jc , u, p - 1) = W ( u )cp fluid T0 ( u, p - 1) - T f ( u, p ) z ( sin, jc , u, p - 1) ,
"p Î P, p > p0, sin, jc Î Sin, j , u Î U (10.8)
( )
Q ( sin, jh , u, p - 1) = W ( u )cp fluid T f ( u, p ) - T0 ( u, p - 1) z ( sin, jh , u, p - 1) ,
"p Î P, p > p0, sin, jk Î Sin, j , u Î U (10.9)

Constraint (10.10) quantifies the heat transferred to the heat storage vessel at
the beginning of the time horizon. The initial temperature of the heat storage fluid
is T0 ( u, p0 ) .
( )
Q ( sin, jh , u, p0 ) = W ( u )cp fluid T f ( u, p1) - T0 ( u, p0 ) z ( sin, jh , u, p0 ) ,
"sin, jh Î Sin, j , u Î U (10.10)

Constraint (10.11) ensures that the final temperature of the heat storage fluid at
any time point becomes the initial temperature of the heat storage fluid at the next
time point. This condition will hold regardless of whether or not there was heat inte-
gration at the previous time point.
T0 ( u, p ) = T f ( u, p - 1) , "p Î P, u Î U (10.11)
Constraints (10.12) and (10.13) ensure that temperature of heat storage does not
change if there is no heat integration with the heat storage unit, unless there is heat
loss from the heat storage unit. MM is any large number, thereby resulting in an
overall “Big M” formulation. If either z ( sin, jc , u, p - 1) or z ( sin, jh , u, p - 1) is equal to
one, Constraints (10.12) and (10.13) will be redundant. However, if these two binary
variables are both zero, the initial temperature at the previous time point will be
equal to the final temperature at the current time point.
æ ö
ç å
T0 ( u, p - 1) £ T f ( u, p ) + MM ç z ( sin, jc , u, p - 1) + å z ( sin, jh , u, p - 1) ÷ ,
÷
è sin, jc sin , jh ø
"p Î P, p > p0, u Î U (10.12)

æ ö
ç å
T0 ( u, p - 1) ³ T f ( u, p ) - MM ç z ( sin, jc , u, p - 1) + å z ( sin, jh , u, p - 1) ÷ ,
÷
è sin, jc sin , jh ø
"p Î P, p > p0, u Î U (10.13)

Constraint (10.14) ensures that minimum thermal driving forces are obeyed when
there is direct heat integration between a hot and a cold unit. This constraint holds
when both hot and cold units operate at constant temperature, which is commonly
encountered in practice. An example is when there is heat integration between an
exothermic and an endothermic reaction.
( )
T ( sin, jh ) - T ( sin, jc ) ³ DT min - MM 1 - x ( sin, jc , sin, jh , p - 1) ,
"p Î P, p > p0, sin, jc , sin, jh Î Sin, j (10.14)

Constraints (10.15) and (10.16) ensure that minimum thermal driving forces are
obeyed when there is heat integration with the heat storage unit. Constraint (10.15)
applies for heat integration between heat storage and a heat sink, while constraint
(10.16) applies for heat integration between heat storage and a heat source. In
Constraints (10.15) and (10.16), the units operate at fixed temperatures. For units not
operating at fixed temperatures, both inlet and outlet minimal thermal driving forces
between the two integrated tasks need also to be enforced.
(
T f ( u, p ) - T ( sin, jc ) ³ DT min - MM 1 - z ( sin, jc , u, p - 1) , )

(
T ( sin, jh ) - T f ( u, p ) ³ DT min - MM 1 - z ( sin, jh , u, p - 1) , )

Constraints (10.17) and (10.18) give the heating load for a cold state and cooling
load for a hot state, respectively, for variable batch size and changing temperature.
( in, jc ) ( )
HL s , p = B ( sin, jc , p )cpstate ( sin, jc ) Tout ( sin, jc ) - Tin ( sin, jc ) , "p Î P, sin, jc Î Sin, j

(10.17)
( in, jh ) ( )
CL s , p = B ( sin, jh , p )cpstate ( sin, jh ) Tin ( sin, jh ) - Tout ( sin, jh ) , "p Î P, sin, jh Î Sin, j

(10.18)
Constraint (10.19) ensures that the heating of a cold state will be satisfied by either
direct or indirect heat integration as well as external utility if required.
HL ( sin, jc , p ) = Q ( sin, jc , u, p ) + st ( sin, jc , p ) + åq ( s

sin , jh
in, jc , sin, jh , p ) ,
"p Î P, sin, jc Î Sin, j , u Î U (10.19)
Constraint (10.20) states that the cooling of a hot state will be satisfied by either
direct or indirect heat integration as well as external utility if required.
CL ( sin, jh , p ) = Q ( sin, jh , u, p ) + cw ( sin, jh , p ) + åq ( s

sin , jc
in, jc , sin, jh , p ) ,
"p Î P, sin, jk Î Sin, j , u Î U (10.20)
The upper bounds of the heating load of a cold state, the cooling load of a hot
state and the amount of heat exchanged during direct heat integration are given in
Constraints (10.21) through (10.23).
HL ( sin, jc , p ) £ y ( sin, jc , p ) Q max ( sin, jc ) , "p Î P, sin, jc Î Sin, j (10.21)
CL ( sin, jh , p ) £ y ( sin, jh , p ) Q max ( sin, jh ) , "p Î P, sin, jk Î Sin, j (10.22)

{
q ( sin, jc , sin, jh , p ) £ x ( sin, jc , sin, jh , p ) min Q max ( sin, jc ) , Q max ( sin, jh ) }
"p Î P, sin, jc , sin, jh Î Sin, j (10.23)
For the specific case where the heating and cooling loads are fixed, Constraints
(10.24) and (10.25) are used instead of Constraints (10.19) and (10.20).
HL ( sin, jc ) y ( sin, jc , p ) = Q ( sin, jc , u, p ) + st ( sin, jc , p ) + x ( sin, jc , sin, jh , p )
´ å
sin , jh
min
sin , jc , sin , jh
{HL ( s ) ,CL ( s )} ,
in, jc in, jh
"p Î P, sin, jc Î Sin, j , u Î U (10.24)
CL ( sin, jh ) y ( sin, jh , p ) = Q ( sin, jh , u, p ) + cw ( sin, jh , p ) + x ( sin, jc , sin, jh , p )
´ å
sin , jc
min
sin , jc , sin , jh
{HL ( s ) ,CL ( s )} ,
in, jc in, jh
"p Î P, sin, jh Î Sin, j , u Î U (10.25)
The amount of heat transferred through direct heat integration will be limited by
the smaller heating or cooling requirement of the heat integrated tasks. Constraints
(10.26) and (10.27) express this. Constraint (10.26) calculates the heat load of the
cold task, while Constraint (10.27) calculates the cooling load of the hot task.
( )
q ( sin, jc , sin, jh , p ) £ B ( sin, jc , p )cpstate ( sin, jc ) Tout ( sin, jc ) - Tin ( sin, jc ) x ( sin, jc , sin, jh , p ) ,

( )
q ( sin, jc , sin, jh , p ) £ B ( sin, jh , p )cpstate ( sin, jh ) Tin ( sin, jh ) - Tout ( sin, jh ) x ( sin, jc , sin, jh , p ) ,

Furthermore, it is possible that a given pair of tasks cannot be heat integrated or

that a possible DT min violation may occur. The possibility of heat integration between
pairs of tasks as well as possible DT min violations should be investigated for each
pair of hot and cold tasks beforehand. If DT min violations occur, the temperatures in
Constraints (10.26) and (10.27) should be adjusted for this.
The amount of heat transferred through direct heat integration can also be limited
by the duration of the shorter task if the tasks have different durations. Constraints
(10.28) through (10.33) capture this. Constraints (10.28) and (10.29) calculate the heat-
ing load per time and cooling load per time, of the cold task and hot task, respectively.
• HL ( sin, jc , p )
HL ( sin, jc , p ) = , "p Î P, sin, jc Î Sin, j (10.28)
dur ( sin, jc , p )
i CL ( sin, jh , p )
CL ( sin, jh , p ) = , "p Î P, sin, jh Î Sin, j (10.29)
dur ( sin, jh , p )
Constraint (10.30) calculates the heat load of the cold task based on the duration
of the same cold task. Constraint (10.31) calculates the heat load of the cold task
based on the duration of the hot task. Constraint (10.32) calculates the cooling load of
the hot task based on the duration of the same hot task. Constraint (10.33) calculates
the cooling load of the hot task based on the duration of the cold task. The amount
of heat integrated directly will effectively be the minimum of these four quantities.
i
q ( sin, jc , sin, jh , p ) £ HL ( sin, jc , p ) dur ( sin, jc , p ) , "p Î P, sin, jc , sin, jh Î Sin, j (10.30)
q ( sin, jc , sin, jh , p ) £ HL ( sin, jc , p ) dur ( sin, jh , p ) , "p Î P, sin, jc , sin, jh Î Sin, j (10.31)
i
q ( sin, jc , sin, jh , p ) £ CL ( sin, jh , p ) dur ( sin, jh , p ) , "p Î P, sin, jc , sin, jh Î Sin, j (10.32)
i
q ( sin, jc , sin, jh , p ) £ CL ( sin, jh , p ) dur ( sin, jc , p ) , "p Î P, sin, jc , sin, jh Î Sin, j (10.33)
In Constraints (10.28) through (10.33), the duration is a function of batch size. If

the duration is fixed, t ( sin, j ) is used and these constraints are then linear.
Constraints (10.34) and (10.35) ensure that the times at which units are active are
synchronized when direct heat integration takes place. Starting times for the tasks
in the integrated units are the same. This constraint may be relaxed for operations
requiring preheating or precooling and is dependent on the process.
(
tu ( sin, jh , p ) ³ tu ( sin, jc , p ) - MM 1 - x ( sin, jc , sin, jh , p ) ) "p Î P, sin, jc , sin, jh Î Sin, j

(10.34)
(
tu ( sin, jh , p ) £ tu ( sin, jc , p ) + MM 1 - x ( sin, jc , sin, jh , p ) ) "p Î P, sin, jc , sin, jh Î Sin, j

(10.35)
Constraints (10.36) and (10.37) ensure that if indirect heat integration takes place,
the time at which a heat storage unit starts either to transfer or receive heat will be
equal to the time a unit is active.
(
tu ( sin, j , p ) ³ t0 ( sin, j , u, p ) - MM y ( sin, j , p ) - z ( sin, j , u, p ) )
"p Î P, u Î U , sin, j Î Sin, j (10.36)
(
tu ( sin, j , p ) £ t0 ( sin, j , u, p ) + MM y ( sin, j , p ) - z ( sin, j , u, p ) )
"p Î P, u Î U , sin, j Î Sin, j (10.37)
Constraints (10.38) and (10.39) state that the time when heat transfer to or from
a heat storage unit is finished will coincide with the time the task transferring or
receiving heat has finished processing.
tu ( sin, j , p - 1) + t ( sin, j ) y ( sin, j , p - 1) ³ t f ( sin, j , u, p )
(
- MM y ( sin, j , p - 1) - z ( sin, j , u, p - 1) ) " Î P, p > p0, u Î U , sin, j Î Sin, j

(10.38)
tu ( sin, j , p - 1) + t ( sin, j ) y ( sin, j , p - 1) £ t f ( sin, j , u, p )
(
+ MM y ( sin, j , p - 1) - z ( sin, j , u, p - 1) ) "p Î P, p > p0, u Î U , sin, j Î Sin, j

(10.39)
The necessary constraints for determining the optimal cyclic schedule are now
described. The scheduling model of Seid and Majozi (2012) was used. Some con-
straints were modified to accommodate cyclic scheduling and are shown here. The
value for “Big M” in the cyclic scheduling constraints and heat integration con-
straints is the value of the upper bound of the cycle length, HU.
10.5.5 Mass Balance between Two Cycles

Constraint (10.40) ensures that the amount of intermediate state s stored at the end of
a cycle is the same as that stored at the beginning of the cycle. There is therefore no
accumulation or shortage of intermediate states when the cycle is repeated.
qs ( s, p ) = Qs0 ( s ) , "s Î S, s ¹ product , feed , p= P (10.40)
10.5.6 Sequence Constraints between Cycles:

Completion of Previous Tasks
Constraint (10.41) defines the relationship between the last task of the previous cycle
and the first task of the current cycle when tasks occur in different units. This main-
tains continuity of tasks between cycles.
( )
tu ( sin, j ¢ , p0 ) ³ t p ( sin, j , p ) - H U 1 - y ( sin, j , p - 2 ) - H ,
"j Î J , p = P , sin, j Î Sinsp, j , sin, j ¢ Î Sinsc, j ¢ (10.41)

Constraint (10.42) is similar to Constraint (10.41), but is applicable when interme-

diate state s is produced from one unit.
( )
tu ( sin, j ¢ , p0 ) ³ t p ( sin, j , p ) - H U 2 - y ( sin, j , p - 1) - tt ( j, p ) - H ,
"j Î J , p = P , sin, j Î Sinsp, j , sin, j ¢ Î Sinsc, j (10.42)

Constraint (10.43) ensures that units are available when the same task is per-
formed in the same unit (the same state is used in the same unit). A state can only be
used in a unit after all preceding tasks in the unit have been completed.
tu ( sin, j , p0 ) ³ t p ( sin, j , p ) - H , "j Î J , p = P , sin, j Î Sin*, j (10.43)
Constraint (10.44) is similar to Constraint (10.43), but pertains to different tasks

being performed in the same unit (different states used in the same unit). A task can
only start in a unit after all previous tasks in the unit are complete.
( )
tu ( sin, j , p0 ) ³ t p s¢in, j , p - H , "j Î J , p = P , sin, j ¹ s ¢in, j , sin, j , s¢in, j Î Sin*, j

(10.44)
10.5.7 Tightening Constraints
This is the same constraint as in the scheduling model, but the time horizon is now
the cycle length. Constraint (10.45) is used to tighten the model. The sum of the dura-
tions of all tasks in a unit must be within one cycle length.
å å(t(s* p
in, j ) y ( sin, j , p ) + b ( sin, j ) mu ( sin, j , p ) ) £ H , "p Î P, jÎJ (10.45)
sin , j ÎSin ,j
Constraints (10.46) and (10.47) ensure processing tasks take place within two
cycles.
tu ( sin, j , p ) £ 2 H , "sin, j Î Sin, j , p Î P, jÎJ (10.46)
t p ( sin, j , p ) £ 2 H , "sin, j Î Sin, j , p Î P, j Î J (10.47)
Constraints (10.48) and (10.49) ensure that the times the heat storage unit is active
are within two cycles.
t0 ( sin, j , u, p ) £ 2 H , "sin, j Î Sin, j , p Î P, j Î J , u Î U (10.48)
t f ( sin, j , u, p ) £ 2 H , "sin, j Î Sin, j , p Î P, j Î J, u ÎU (10.49)
Constraints (10.50) through (10.73) cater for the wrapping of the heat storage tem-
perature and ensure the temperature in heat storage is the same at the beginning and
end of the cycle in order for the cycle to be repeated.
Constraints (10.50) and (10.51) are defined, where Tstart is the initial heat storage
temperature at the beginning of the cycle and Tend is the heat storage temperature at
the end of the cycle.
DTcw = Tend - Tstart (10.50)
DTst = Tstart - Tend (10.51)

If the heat storage temperature at the end of the cycle is higher than the tem-
perature at the beginning of the cycle, cooling will be required to bring the storage
temperature back to the starting temperature—essential for repetition of the cycle.
ΔTcw will then be positive. If the heat storage temperature at the end of the cycle is
lower than the temperature at the beginning of the cycle, heating will be required
and ΔTst will be positive.
Constraint (10.52) is used if the heat storage temperature at the end of the cycle is
too high and cooling is required to bring it back to the temperature at the beginning
of the cycle. Constraint (10.53) is similar, but will be used if extra heating is required
at the end of the cycle to bring the heat storage temperature back to the starting
temperature.
extra _ cw ( u ) = W ( u ) cp fluid DTcw xcw (10.52)
extra _ st ( u ) = W ( u ) cp fluid DTst xst (10.53)
Both Constraints (10.52) and (10.53) contain trilinear terms, where a binary
variable and two continuous variables are multiplied. This results in a non-convex
mixed integer nonlinear programming (MINLP) formulation. The bilinearity result-
ing from the multiplication of a continuous variable with a binary variable may be
handled effectively with the Glover transformation (Glover, 1975). This is an exact
linearization technique and as such will not compromise the accuracy of the model.
Constraints (10.54) and (10.55) are then obtained.
extra _ cw ( u ) = W ( u ) cp fluid Gcw (10.54)
extra _ st ( u ) = W ( u ) cp fluid Gst (10.55)
Constraints (10.54) and (10.55) still contain bilinear terms where two continuous
variables are multiplied. A method to handle this is a reformulation–linearization
technique (Sherali and Alameddine, 1992) as discussed by Quesada and Grossmann
(1995). This is an inexact linearization technique and increases the size of the model
by an additional type of continuous variable and four types of continuous constraints.
These constraints correspond to the convex and concave envelopes of the bilinear
terms over the given bounds. Constraint (10.54) is linearized using Constraints
(10.56) through (10.62) and Constraint (10.55) is linearized using Constraints (10.63)
through (10.69). Constraints (10.70) and (10.71) are then obtained.
W ( u )Gcw = K cw (10.56)
W L £ W (u ) £ W U (10.57)
DT L £ Gcw £ DT U (10.58)
K cw £ W U Gcw + DT LW ( u ) - W U DT L (10.59)
K cw £ W L Gcw + DT U W ( u ) - W L DT U (10.60)
K cw ³ W L Gcw + DT LW ( u ) - W L DT L (10.61)
K cw ³ W U Gcw + DT U W ( u ) - W U DT U (10.62)
W ( u )Gst = K st (10.63)
W L £ W (u ) £ W U (10.64)
DT L £ Gst £ DT U (10.65)
K st £ W U Gst + DT LW ( u ) - W U DT L (10.66)
K st £ W L Gst + DT U W ( u ) - W L DT U (10.67)
K st ³ W L Gst + DT LW ( u ) - W L DT L (10.68)
K st ³ W U Gst + DT U W ( u ) - W U DT U (10.69)
extra _ cw ( u ) = cp fluid K cw (10.70)
extra _ st ( u ) = cp fluid K st (10.71)
Constraints (10.8) through (10.10) also have trilinear terms where a binary vari-
able and two continuous variables are multiplied, while Constraints (10.28) through
(10.33) have bilinear terms where two continuous variables are multiplied. These
constraints are linearized similarly.
Constraint (10.72) is also necessary, as either extra cooling or extra heating will be
required for the heat storage vessel at the end of the cycle, or neither will be required.
xcw + xst £ 1 (10.72)
If neither extra heating nor extra cooling are required to bring the heat storage
temperature back to its initial temperature, Tstart will be equal to Tend and the values
for extra_cw(u) and extra_st(u) will both be zero.
In order for the heat storage temperature to return to its starting temperature, the
total energy into the heat storage vessel must be equal to the energy out of the heat
storage vessel within the cycle. Constraint (10.73) is used to ensure this.
ååQ ( s in, jh , u, p ) + extra _ st ( u ) = ååQ ( s in, jc , u, p ) + extra _ cw ( u ) (10.73)

sin , jh p sin , jc p
The profit is given by Equation 10.74.
Profit = åCP ( s ) q ( s , p ) -Cost_ cwååcw ( s

s
p
s
p
sin , jh p
in, jh , p ) - Cost _ cw * extra _ cw ( u )
- Cost _ st ååst ( s
sin , jc p
in, jc , p ) - Cost _ st * extra _ st ( u ) ,
"p = P, sin, jh Î Sin, j , sin, jc Î Sin, j , s p Î S p (10.74)
The objective for the cyclic period is then to maximize the average profit per
cycle, as given by Constraint (10.75).
Profit
max (10.75)
H
The same solution procedure as used by Stamp and Majozi (2011) is applied. The
overall MINLP model is linearized and solved as a MILP. The solution obtained is
then used as a starting point for the exact MINLP model. If the solutions from the
two models are equal, the solution is globally optimal, as global optimality can be
proven for MILP problems. If the solutions differ, the MINLP solution is locally
optimal. The possibility also exists that no feasible starting point is found.
The objective function for the cyclic portion of the problem, Constraint (10.75) is,
however, nonlinear and the MINLP can therefore not be linearized completely to a
MILP due to the nonlinear objective function. All of the nonlinear constraints apart
from the objective function are therefore linearized to produce a relaxed MINLP
problem, the solution of which provides a starting point for the exact MINLP prob-
lem. The resulting solution to the exact MINLP cannot then be guaranteed to be a
global optimum.
Constraints (10.40) through (10.51) and Constraints (10.56) through (10.75) are
used in the linearized model, while Constraints (10.40) through (10.51), Constraints
(10.54) and (10.55) and Constraints (10.72) through (10.75) are used in the exact
model. This is in addition to the scheduling and heat integration constraints and
applies to the determination of the optimal cyclic schedule. Once the optimal cycle
length has been determined, the initial and final periods can be solved. The overall
objective is the maximization of profit over the entire long time horizon.
10.6 MULTIPURPOSE EXAMPLE
Figure 10.4 shows the STN of a multipurpose batch plant, while Figure 10.5 shows the
SSN(Seid and Majozi, 2012). This example is commonly referred to as “BATCH1” in
literature and has been modified to include heating and cooling requirements for cer-
tain tasks to allow for the possibility of heat integration. For this example, the batch
sizes were not fixed and the energy requirements varied linearly with batch size.
The required scheduling data may be found in Tables 10.1 and 10.2, while the
necessary heat integration data and heating and cooling requirements may be found
in Tables 10.3 and 10.4, respectively.
Product 1 (s7)
40% Int AB (s8)

Heating 40% Reaction 2 60%
Feed A Hot A
60% 10%
(s1) (s5) Impure E (s9)
Int BC Separation
80%
(s6)
90%
50% Reaction 3
Reaction 1
Feed B 50% (s3) 20% Product 2
(s2) (s4) (s10)
Feed C
FIGURE 10.4 STN for multipurpose example.
(s1)
Feed A
(s2)
Feed B Hot A (s5)
50%
40%
(s6) (s7)
60% 40%
Int BC Product 1
50%
(s3)
(s9) 10%
60% 80%
Feed C Int AB Impure E
(s4) (s8) 90%
20%
Product 2
(s10)
FIGURE 10.5 SSN for multipurpose example.
TABLE 10.1
Scheduling Data for Multipurpose Example
Reactor 1 50 RX1, RX2, RX3 2, 2, 1
Reactor 2 80 RX1, RX2, RX3 2, 2, 1
Still 200 Separation 1 (Product 2)
2 (Int AB)
TABLE 10.2
Scheduling Data for Multipurpose Example
State Storage Capacity (ton) Initial Amount (ton) Revenue (c.u./ton)
Feed A (s1) Unlimited Unlimited 0
Feed B (s2) Unlimited Unlimited 0
Feed C (s3/s4) Unlimited Unlimited 0
Hot A (s5) 100 0 0
Int AB (s8) 200 0 0
Int BC (s6) 150 0 0
Impure E (s9) 200 0 0
TABLE 10.3
Heat Integration Data for Multipurpose
Example
Parameter Value
Specific heat capacity, cp fluid (kJ/kg °C) 4.2
Product selling price (c.u./ton) 100
Steam cost (c.u./kWh) 10
Cooling water cost (c.u./kWh) 2
DT min (°C) 10
L
T (°C) 20
U
T (°C) 250
DT L (°C) 0
DT (°C)
U
250
W L (ton) 0.1
U
W (ton) 4
TABLE 10.4
Heating/Cooling Requirements for Multipurpose Example
Max Heating/Cooling
Task Type Requirement (kWh) Operating Temperature (ºC)
RX2 Endothermic 80 (heating) 60
TABLE 10.5
Results of Cyclic Portion for Multipurpose Example
Optimal Cycle
Cycle Range (h) Time (h) Objective (c.u./h) Time Points CPU Time (s)
3–6 5 3581.920 6 204.55
6–9 6 3478.086 7 1,781.50
9–12 12 3560.000 9 280,048.74
TABLE 10.6
Results of Optimal Cyclic Length for Multipurpose Example
Starting and ending amount of intermediate state (ton)
s5 0
s6 0
s8 48
s9 150
Starting and ending storage temperature (°C) 70
Heat storage size (ton) 0.357
Cooling water (kWh) 95.2
Steam (kWh) 0
Extra cooling water (kWh) 0
Extra steam (kWh) 0
10.6.1 Cyclic Portion
As the model used to solve the cyclic portion contains nonlinear terms, the solu-
tion procedure as described earlier this chapter is used. The objective function will,
however, still contain a nonlinearity which cannot be linearized and as such both the
linearized and exact models will be MINLP problems. The problem was solved with
GAMS 24.1.3 using DICOPT for the MINLP with CPLEX as the MIP solver and
MINOS5 as the NLP solver. All models were solved using a computer with an Intel
Core i3, 3.1 GHz processor with 2.0 GB RAM.
The results obtained for the cyclic portion can be seen in Table 10.5. The optimal
cycle length is 5 h and an average profit of 3581.920 c.u./h is achieved. The CPU times
in the table are the sum of the CPU times for both the linearized and the exact models.
The results obtained for the optimal schedule for the optimal cycle length of 5 h
are shown in Table 10.6. The optimal schedule obtained is shown in Figure 10.6.
10.6.2 Initial Period
In order to solve for the initial period of the time horizon, the amounts of interme-
diates at the end of the initial period are fixed to the values required to start the
Units
Still 80 50
<80> <70> <80>

RX2 RX3 RX2
Reactor 2 80 80 80
<30> <60>
RX1 [30] RX3[10] RX1 [55.2] (Heat integration kWh)
Reactor 1 50 50 46 {Steam kWh}
[Cooling water kWh]
Material produced ton
<Heat storage kWh>
Heater 32 32
0 1 2 3 4 5 6 7 8 9
Time (h)
FIGURE 10.6 Gantt chart for optimal cycle length (5 h) for multipurpose example.
(Heat integration kWh)

{Steam kWh}
[Cooling water kWh]
Units Material produced ton
<Heat storage kWh>
Still 8.571
<80> <80> <80>

RX1 [60] RX2 RX2 RX3 [30] RX2
Reactor 2 80 80 80 80 80
(40)
RX1 [52.8] RX1 [60] RX3 [56] RX2 {40} RX3 [54]
Reactor 1 44 50 40 50 38.571
Heater 32 32 20 32
0 1 2 3 4 5 6 7 8 9
Time (h)
FIGURE 10.7 Gantt chart for maximization of profit over initial period (9 h) for multipur-
pose example.
cyclic period. These values were obtained from the solution of the cyclic period.
This ensures there will be sufficient of the intermediate states produced in the initial
period to be available to start the cyclic scheduling period. First, a makespan mini-
mization problem is solved to determine a feasible time horizon for the initial period.
A profit maximization problem is then solved in order to maximize the profit in the
TABLE 10.7
Results for Maximization of Profit over Initial
Period for Multipurpose Example
Time period (h) 9
Objective (c.u.) 11,345.829
Time points 6
Steam (kWh) 40
Initial heat storage temperature (°C) 230.064
CPU time (s) 0.983
initial period. The optimum heat storage size of 0.357 ton which was solved for in
the cyclic portion is now fixed in the initial period. Since this is the case, both the
makespan minimization and profit maximization problems will be MILP problems.
The final temperature for the heat storage vessel is set at 70°C.
From the makespan minimization problem, a feasible time horizon of 9 h was
determined for the initial period, in a CPU time of 0.593 s and required 6 time
points. The Gantt chart for the profit maximization problem over 9 h is shown in
Figure 10.7 and the results are shown in Table 10.7.
10.6.3 Final Period
A total time horizon of 24 h was chosen. Repeating the cyclic portion once and
accounting for the initial period of 9 h, a final period of 10 h remains. A profit
maximization problem was solved for the final period and an objective value of
46,605.928 c.u. was determined in a CPU time of 1.794 s and required 6 time points.
The Gantt chart for the maximization of profit over the final period is shown in
10.6.4 Comparison with Direct Solution

The problem was solved directly over a time horizon of 24 h for the case where only
direct heat integration was used as well as the case where only utilities were avail-
able for heating and cooling. In order to have a long enough time horizon for a fair
comparison, the problem becomes too complex to solve directly for the heat storage
case. A fairly long final period is required in the cyclic scheduling model as most
of the products are formed during this time and this makes the overall time horizon
lengthy. However, the results for the direct solution of the problem for the other
two cases are shown in Table 10.9 as well as a summary of the results for the cyclic
scheduling heat integration model.
As seen from Table 10.9, an error of less than 6% is achieved when the solution
obtained using the proposed cyclic scheduling model with direct heat integration is
compared to the result obtained from the direct solution over 24 h.
(Heat integration kWh)

{Steam kWh}
[Cooling water kWh]
Units <Heat storage kWh>
Still 80 200 122.5
<59.976>
<80> <70> <80>
RX1 [60] RX2 {20} RX3 [10.024] RX2 RX3 RX2
Reactor 2 80 80 80 80 80 80
<29.988> (60)
RX1 [30.012] RX1 RX3 [70] RX2 {80} RX3 [59.5] RX2 {16}
Reactor 1 50 50 50 50 42.5 10
Heater 32 52 36
0 1 2 3 4 5 6 7 8 9 10
Time (h)
FIGURE 10.8 Gantt chart for maximization of profit over final period (10 h) for multipur-
pose example.
TABLE 10.8
Results for Maximization of Profit over Final
Period for Multipurpose Example
Time period (h) 10
Time points 6
Steam (kWh) 116
CPU time (s) 1.794
10.7 INDUSTRIAL CASE STUDY

Figure 10.9 shows the STN of an industrial case study, while Figure 10.10 shows the
SSN (Chen and Chang, 2009).
The scheduling data for the problem may be found in Tables 10.10 and 10.11 with
the stoichiometric data available in Table 10.12. The heat integration data and heat-
ing and cooling requirements may be found in Tables 10.13 and 10.14, respectively.
10.7.1 Cyclic Portion
The results obtained for the cyclic portion can be seen in Table 10.15. The optimal cycle
length is 9 h and an average profit of 20,528.395 c.u./h is achieved. The CPU times in
the table are the sum of the CPU times for both the linearized and the exact models.
TABLE 10.9
Comparison between Direct Solution and Cyclic Scheduling Solution
over 24 h
% Error from
Period Duration (h) Profit (c.u.) Direct Solution (c.u.) Exact Solution
Heat integration with storage
Initial 9 11,345.829
Cyclic 5 17,909.600
Final 10 46,605.928
Overall 24 75,861.357 — —
Direct heat integration
Initial 6 5,547.2
Cyclic 5 (×2) 34,744.0
Final 8 31,840.0
Overall 24 72,131.2 76,580 5.809
Utilities only
Initial 11 10,312.8
Cyclic 5 15,989.6
Final 8 40,840.0
Overall 24 67,142.4 70,790 5.153
s11
s10
s1 Reaction 1 s2 Reaction 2 s3 Reaction 3
s9
s4 Settling s5 Evaporation s6
Product
s7
s8 Water
Waste
FIGURE 10.9 STN for industrial case study.

s11
s6
s10
s5
s1 s2 s3 s4 s7
s8
s9
FIGURE 10.10 SSN for industrial case study.
TABLE 10.10
Scheduling Data for Industrial Case Study
R1 10 RX1 2
R2 10 RX1 2
R3 10 RX2, RX3 3, 1
R4 10 RX2, RX3 3, 1
SE1 10 Settling 1
SE2 10 Settling 1
SE3 10 Settling 1
EV1 10 Evaporation 3
EV2 10 Evaporation 3
TABLE 10.11
Scheduling Data for Industrial Case Study
State Storage Capacity (ton) Initial Amount (ton) Revenue (c.u./ton)
s2 100 0 0
s3 100 0 0
s4 100 0 0
s5 100 0 0
s6 100 0 10,000
s7 100 0 0
s8 100 0 0
TABLE 10.12
Stoichiometric Data for Industrial Case Study
State Ton/Ton Output Ton/Ton Product
s1 0.20
s9 0.25
s10 0.35
s11 0.20
s7 0.7
s8 1
TABLE 10.13
Heat Integration Data for Industrial Case Study
Parameter Value
Specific heat capacity, cp fluid (kJ/kg °C) 4.2

Product selling price (c.u./ton) 10,000
Steam cost (c.u./kWh) 20
Cooling water cost (c.u./kWh) 8
DT min (°C) 5
T L (°C) 20
T U (°C) 180
DT L (°C) 0
DT (°C)
U
180
W L (ton) 0.1
U
W (ton) 1.5
TABLE 10.14
Heating/Cooling Requirements for Industrial Case Study
Max Heating/Cooling Operating
Task Type Requirement (kWh) Temperature (ºC)
Evaporation Endothermic 110 (heating) 90
TABLE 10.15
Results of Cyclic Portion for Industrial Case Study
Optimal Cycle
Cycle Range (h) Time (h) Objective (c.u./h) Time Points CPU Time (s)
3–6 4 18,475.556 3 1.079
6–9 9 20,528.395 6 123.734
9–12 9 20,528.395 6 42.137
TABLE 10.16
Results of Optimal Cyclic Length for Industrial Case Study
Starting and ending amount of intermediate state (ton)
s2 5.625
s3 30
s4 10
s5 4.882
Starting and ending storage temperature (°C) 95
Heat storage size (ton) 0.22
Steam (kWh) 0
Extra cooling water (kWh) 0
Extra steam (kWh) 0
The results obtained for the optimal schedule for the optimal cycle length of 9 h
are shown in Table 10.16. The optimal schedule obtained is shown in Figure 10.11.
10.7.2 Initial Period
The amounts of intermediate states at the end of the initial period are fixed to the
values required to start the cyclic period to ensure there will be sufficient of the
intermediate states produced in the initial period available to start the cyclic sched-
uling period. The optimum heat storage size of 0.22 ton, which was solved for in the
cyclic period, is now fixed in the initial period. The final temperature for the heat
storage vessel is set at 95°C.
From the makespan minimization problem, a feasible time horizon of 11 h
was determined for the initial period, in a CPU time of 0.73 s and required 6 time
points. The Gantt chart for the profit maximization problem over 11 h is shown in
Units
<46.3>
Evaporator 2 9.091
Evaporator 1 9.091 9.091 4.209
Settler 3
Settler 2 10 10 10 (100) (100) (100)
Settler 1 <46.3>
10 10
RX3 RX3 RX3 RX2 RX2
Reactor 4 10 10 10 10 10
RX2 [53.7] RX3 RX3 RX2 (Heat integration kWh)
Reactor 3 10 10 10 10 {Steam kWh}
[Cooling water kWh]
Reactor 2 10 Material produced ton
<Heat storage kWh>
Reactor 1 6.875 5.625
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Time (h)
FIGURE 10.11 Gantt chart for optimal cycle length (9 h) for industrial case study.
Units
Evaporator 2
Evaporator 1 4.562
Settler 3 10
Settler 2 10
(50.187)
Settler 1
RX2 [100] RX2 [100] RX3 RX3 RX3 (Heat integration kWh)
Reactor 4 10 10 10 5 10 {Steam kWh}
RX2 [30.7] RX2 [100] RX2 [49.813] [Cooling water kWh]
<69.3> Material produced ton
Reactor 3 10 10 10
<Heat storage kWh>
Reactor 2 10
Reactor 1 3.75 10 10
0 1 2 3 4 5 6 7 8 9 10 11
Time (h)
FIGURE 10.12 Gantt chart for maximization of profit over initial period (11 h) for indus-
trial case study.
10.7.3 Final Period
A total time horizon of 48 h was chosen. Repeating the cyclic portion three times and
accounting for the initial period of 11 h, a final period of 10 h remains. A profit maxi-
mization problem was solved for the final period and an objective value of 338,884.348
c.u. was determined in a CPU time of 1,515.2 s and required 9 time points.
TABLE 10.17
Results for Maximization of Profit over Initial Period
for Industrial Case Study
Time period (h) 11
Time points 7
Steam (kWh) 0
Initial heat storage temperature (°C) 20
CPU time (s) 81.89
Units {92.406} {10} {110}

Evaporator 2 8.401 10 10
{10} {10}
Evaporator 1 10 10 10 <46.3>
Settler 3 10 10 2.647 5
(100)
Settler 2 (100) 7.353 10 (100) 10
Settler 1 10 10 10
RX3 RX3 RX3 RX3 RX2 RX3
Reactor 4 10 10 10 10 10 10
RX2 RX3 RX2 RX3 RX3
Reactor 3 10 10 10 5 10 (Heat integration kWh)
{Steam kWh}
Reactor 2 5.625 5.625 [Cooling water kWh]
Reactor 1 <Heat storage kWh>
0 1 2 3 4 5 6 7 8 9 10
Time (h)
FIGURE 10.13 Gantt chart for maximisation of profit over final period (10 h) for industrial
case study.
The Gantt chart for the profit maximization over the final period is shown in
The problem was also solved for the case with direct heat integration only as well
as the case where only utilities are available. A summary of results for each case is
given in Table 10.19. In each case, the cyclic period was repeated three times.
10.7.4 Comparison with Existing Literature Model

This same example was considered by Chen and Chang (2009). In solving for the
optimal cycle length, a series of fixed values were used and the cycle time giving
the best objective, which is the average profit per time, was considered optimal.
TABLE 10.18
Results for Maximization of Profit over Final Period
for Industrial Case Study
Time period (h) 10
Objective (c.u.) 338,884.348
Time points 9
Cooling water (kWh) 0
Steam (kWh) 232.406
CPU time (s) 1,515.2
TABLE 10.19
Results of Cyclic Scheduling for Different Cases over
48 h for Industrial Case Study
Period Duration (h) Profit (c.u.)
Heat integration with storage
Initial 11 23,793.804
Cyclic 9 (×3) 554,266.665
Final 10 338,884.348
Overall 48 916,944.817
Direct heat integration
Initial 13 11,625.325
Cyclic 9 (×3) 554,266.668
Final 8 234,494.118
Overall 48 800,386.111
Utilities only
Initial 14 6,457.583
Cyclic 9 (×3) 525,177.777
Final 7 226,494.118
Overall 48 758,129.478
The model proposed in this chapter, however, includes the optimal cycle length as an
optimization variable which is solved for between a given upper and lower bound.
Also, the initial and final periods of the time horizon were not considered in the
model of Chen and Chang (2009).
To compare the models, the heat integration and cost data as used by Chen and
Chang (2009) were used. These values may be obtained from Table 10.20. In the
example, batch sizes were not fixed and the maximum batch sizes were given as
8 ton for all units. Only direct heat integration was considered.
The model proposed in this chapter, with direct heat integration only, was used to
solve for the optimal cycle time between 6 and 9 h and the same optimal objective
value of $161.05 was obtained. The proposed model was also applied over an interval
TABLE 10.20
Data for Comparison of Industrial Case
Study
Product Selling price ($/ton)
s6 100
Utility Unit cost ($/ton)
Cooling water 8
Steam 15
Task Cooling/ Heating duty (ton)
Reaction 2 5
Evaporation 4
of 9–12 h. The optimal cycle time was found to be 11 h and gave a better objective
value of $164.71 compared to $161.05 for the reported optimal cycle time of 9 h.
10.8 CONCLUSIONS
A model for the cyclic scheduling of multipurpose batch plants using direct and indi-
rect heat integration has been presented. Cyclic scheduling constraints were incor-
porated into the heat integration model of Stamp and Majozi (2011). The presented
cyclic scheduling model can be used for any chosen long time horizon. Once the
cyclic and initial portions have been solved, the flexibility of the model allows
the solution over any long time horizon, as only the profit maximization problem
of the final period must be solved for again. The time horizon therefore does not
affect the computational complexity required to solve the model.
The model may give better results when compared to an existing method as the
cycle time is an optimization variable rather than a fixed value and both the initial
and final periods are also accounted for.
The model may be preferred for the solution of problems over long time hori-
zons as direct solution using short-term methods may not be feasible. The solution
obtained over 24 h using the proposed cyclic scheduling model with direct heat inte-
gration for a multipurpose example was compared to the result obtained from the
direct solution and an error of less than 6% was achieved.
REFERENCES
Boyadjiev, C.H.R., Ivanov, B., Vaklieva-Bancheva, N., Pantelides, C.C., Shah, N., 1996.
Optimal energy integration in batch antibiotics manufacture. Computers and Chemical
Engineering. 20, S31–S36.
Chen, C.L., Ciou, Y.J., 2008. Design and optimization of indirect energy storage systems
for batch process plants. Industrial and Engineering Chemical Research. 47(14),
4817–4829.
Glover, F., 1975. Improved linear integer programming formulations of nonlinear integer
problems. Management Science. 22(4), 455–460.
batch plants. Industrial and Engineering Chemical Research. 48(18), 8551–8565.
scheduling. 1. Multipurpose batch processes. Industrial and Engineering Chemical
Research. 37(11), 4341–4359.
Kemp, I.C., 1990. Applications of the time-dependent cascade analysis in process integration.
Heat Recovery System and CHP. 10(4), 423–435.
Knopf, F.C., Okos, M.R., Reklaitis, G.V., 1982. Optimal design of batch/semicontinuous pro-
cesses. Industrial and Engineering Chemical Process Design and Development. 21(1),
79–86.
Lee, J.Y., Seid, E.R., Majozi, T., 2015. Heat integration of intermittently available continu-
ous streams in multipurpose batch plants. Computers and Chemical Engineering. 74,
100–114.
An aspect of cleaner production. Journal of Cleaner Products. 17, 945–950.
Majozi, T., 2010. Batch Chemical Process Integration—Analysis, Synthesis and Optimization,
Springer, New York.
Majozi, T., Zhu, X.X., 2001. A novel continuous-time MILP formulation for multipurpose
batch plants. 1. Short-term scheduling. Industrial and Engineering Chemical Research.
40(25), 5935–5949.
Nonyane, D.R., Majozi, T., 2012. Long term scheduling technique for wastewater minimisa-
tion in multipurpose batch processes. Applied Mathematical Model. 36, 2142–2168.
Papageorgiou, L.G., Shah, N., Pantelides, C.C., 1994. Optimal scheduling of heat-inte-
grated multipurpose plants. Industrial and Engineering Chemical Research. 33(12),
3168–3186.
ematical formulation. Annals of the Operating Research. 120, 201–230.
multicomponent flows. Computers and Chemical Engineering. 19(12), 1219–1242.
Rašković, P., Anastasovski, A., Markovska, L.J., Meško, V., 2010. Process integration in bio-
process indystry: Waste heat recovery in yeast and ethyl alcohol plant. Energy. 35,
704–717.
Seid, E.R., Majozi, T., 2014. Heat integration in multipurpose batch plants using a robust
scheduling framework. Energy. 71, 302–320.
Shah, N., Pantelides, C.C., Sargent, R.W.H., 1993. Optimal periodic scheduling of multipur-
pose batch plants. Annals of the Operating Research. 42(1), 193–228.
Sherali, H.D., Alameddine, A., 1992. A new reformulation–linearization technique for bilin-
ear programming problems. Journal of Global Optimum. 2(4), 379–410.
Stamp, J.D., Majozi, T., 2011. Optimum heat storage design for heat integrated multipurpose
batch plants. Energy. 36, 5119–5131.
Stoltze, S., Mikkelsen, J., Lorentzen, B., Petersen, P.M., Qvale, B., 1995. Waste-heat recovery
in batch processes using heat storage. Journal of Energy Resources and Technology.
117, 142–149.
Vaklieva-Bancheva, N., Ivanov, B.B., Shah, N., Pantelides, C.C., 1996. Heat exchanger net-
work design for multipurpose batch plants. Computers and Chemical Engineering.
20(8), 989–1001.
Vaselenak, J.A., Grossmann, I.E., Westerberg, A.W., 1986. Heat integration in batch process-
ing. Industrial and Engineering Chemical Process Design and Development. 25(2),
357–366.
Wang, Y., Wei, Y., Feng, X., Chu, K.H., 2014. Synthesis of heat exchanger networks featuring
batch streams. Applied Energy. 114, 30–44.
Wang, Y.P., Smith, R., 1995. Time pinch analysis. Trans IChemE. 73(A), 905–914.
Wu, D., Ierapetritou, M., 2004. Cyclic short-term scheduling of multiproduct batch plants
using continuous-time representation. Computers and Chemical Engineering. 28,
2271–2286.
Index
A sub-ranges, 277–278
time horizon, 278
Allocation constraints, 29, 51 unit schedule, 277–278
Assignment constraint, 104 wrapping around, 278
B D
Big-M constraints, 178 Direct heat integration, 83, 203, 296–297, 299
Binary variables Discrete-time representation, 13–14, 28
allocation constraints, 29
capacity constraints, 30
discrete-time representation, 28 E
duration constraints, 31 Energy and water use
features, 27 energy integration, 198–200
material balance, 30 mathematical model
objective function, 34 Halim and Srinivasan study, 211–218
precedence-based representation, 27 heat integration model, 203–208
problem statement, 29 Kondili study, 218–225
sequence constraints wastewater minimization model, 200–202,
consuming unit, 33 208–211
FIS policy, 33–34 problem statement, 202
processing unit, 31 scheduling, 198
production pattern, 27
stable state, 32
unstable state, 33 F
single-grid-based models, 28–29 Finite intermediate storage (FIS), 155–156
STN Finite wait (FW), 4
batch size and processing time data, 35 Fixed intermediate storage (FIS), 33–34
characteristics, 35
computational results, 36–39
initial inventory and storage size data, 35–36 G
Kondili study, 39–41 Gantt chart
Vooradi and Shaik study, 41–44 heat-integrated batch plants, 161, 163
time horizon, 34 optimal cycle length, 296
profit maximization problem, 296–297
C 16 h time horizon, 36, 39, 41, 43
Stamp and Majozi study, 138–139
Capacity constraints, 30, 51, 151–152 STN, 98–99
Central intermediate storage (CIS), 4 storage, 257–258
Continuous multistage multiproduct plants, 4–6, 10 h time horizon, 74
12–13 24 h time horizon, 62
Continuous time Global event point representations, 14–15
Halim and Srinivasan study, 211–218 Glover transformation, 155
heat integration model, 203–208 Graphical techniques, 6–8
Kondili study, 218–225
wastewater minimization model, 208–211
H
Cyclic scheduling
maximization problem, 278 Heat-exchanger networks (HENs), 110–113
minimization problem, 278 Heat-integrated batch plants
optimal production, 277 assignment constraint, 104
representation of, 277–278 basic grass root design, 148–149
309
310 Index
BATCH1 tightening constraints, 289–292

cooling requirements, 292, 294 time horizon, 156
cyclic portion, 295–296 unit size, 151
direct heat integration, 83, 296–297, 299 motivation and rationale
final period, 296, 298 energy savings, 84
initial period, 295–297 exploiting process, 85
scheduling data, 159–162, 292–294 hot and cold stream pairs, 85–86
SSN, 292–293 overlooking process, 84–85
STN, 292–293 optimal scheduling, 276
cyclic scheduling plant superstructure, 159
maximization problem, 278 problem statement, 86, 150–151, 279
minimization problem, 278 requirements, 276
optimal production, 277 sequence constraint, 104–105
representation of, 277–278 STN
sub-ranges, 277–278 batch sizes, 97–98
time horizon, 105, 278 computational results, 101, 103
unit schedule, 277–278 data processing, 96–97
wrapping around, 278 hot and cold streams, 97–98
equipment, 159–162, 276 material and utility data, 96, 98
extended design, 149–150 outlet streams, 99–102
Gantt chart, 161, 163 process representation, 96–97
indirect integration, 83 raw material, 100–102
industrial case study representation, 159
batch sizes, 305–306 utility cost, 99
cooling requirements, 298, 301 superstructure representations, 86–87
cyclic portion, 298, 302–303 time-dependent nature, 275–276
final period, 303–305 Horizon constraint, 148
initial period, 302–304
scheduling data, 298, 300 I
SSN, 298, 300
STN, 298–299 Indirect integration, 83–84
stoichiometric data, 298, 301
Kondili study, 159 M
mass and energy balances
heating and cooling, 90–93 Mass balance constraints
heat integration constraints, 93–95 contaminant mass balance, 175–176
objective function, 95–96 regenerator operation, 177–178
processing units, 86, 88–89 reuse/recycle streams, 174–175
for storage, 89–90 storage tank, 176–177
mathematical formulation Mass exchange network (MEN), 113
batch size, 153 Maximal state task network (mSTN), 150
binary variables, 281 Mixed integer linear programming (MILP) model,
capacity constraints, 151–152 9–10, 113, 194
continuous variables, 280–281 Mixed intermediate storage (MIS) policy, 4, 134
energy integration, 156–157 Multipurpose batch plants, 4–6
existence constraints, 151 developments
FIS, 155–156 twentieth century, 110–111
intermediate state, 154–155 twenty-first century, 111–114
mass balance, 288 discrete-time representation, 13–14
material storage, 152 examples, 109
objective function, 158 features, 124–125
parameters, 281–288 global event point representations, 14–15
piping constraints, 153 Kallrath study, 134–141
processing unit, 152–153 mathematical model
sequence constraints, 153–154, constraints, 114–123
288–289 objective function, 123
sets, 279–280 problem statement, 114
Index 311
product 1 and product 2 duration, 246

computational results, 126, 128–130 feasibility and time horizon, 247
data pertaining, 126, 128 mass balance, 248
plant flowsheet, 124 material balance, 246
relevant data, 126–127 objective function, 248–249
STN and SSN, 125–126 solution procedure, 249
unit operations, 124
slot time representation, 15–16 R
Sundaramoorthy and Karimi study, 130–135
unit-specific models, 16–20 Reformulation–linearization technique, 290–291
Resource-task network (RTN), 149
N
S
No intermediate storage (NIS), 4, 6, 8, 20
Scheduling constraints
O central storage, 180–182
recycle/reuse, 179–180
Optimal batch integration (OBI), 111 regenerator scheduling, 182–183
task scheduling, 178–179
P wastewater minimization
storage, 243–245
Petrochemical plant, 134–141 washing operation, 242–243
Precedence-based representation, 27 Semi-continuous multistage multiproduct plants,
Precedence sequencing, 8–10 4–6, 12–13
Process scheduling Sequence constraints
assignment constraint, 104 binary variables
direct integration, 83 consuming unit, 33
indirect integration, 83 FIS policy, 33–34
mass and energy balances processing unit, 31
heating and cooling, 90–93 production pattern, 27
heat integration constraints, 93–95 stable state, 32
objective function, 95–96 unstable state, 33
processing units, 86, 88–89 wastewater minimization
for storage, 89–90 recycle/reuse streams, 243–244
motivation and rationale storage, 243–245
energy savings, 84 Short-term scheduling model, 149
exploiting process, 85 Single-grid-based models, 28–29
hot and cold stream pairs, 85–86 Slot time representation, 10–12, 15–16
overlooking process, 84–85 Space discretization technique, 150
problem statement, 86 State sequence network (SSN)
sequence constraint, 104–105 allocation constraints, 51
STN BATCH1, 292–293
batch sizes, 97–98 capacity constraints, 51
computational results, 101, 103 duration constraints, 51
data processing, 96–97 effective state, 50
Gantt chart, 98–99 intermediate state, 52
hot and cold streams, 97–98 material balance, 51
material and utility data, 96, 98 objective function, 54–55
outlet streams, 99–102 optimal number, 55–57
process representation, 96–97 recipe representation, 2–3
raw material, 100–102 sequence constraints, 51–52
utility cost, 99 storage constraints, 52–53
superstructure representations, 86–87 time horizon constraints, 53–54
time horizon, 105 State-task network (STN)
Production scheduling constraints BATCH1, 292–293
capacity, 246 batch sizes, 35, 97–98
cycle length, 247–248 characteristics, 35
312 Index
computational results, 36–39, 101, 103 optimum cycle length, 252, 254–255
data processing, 96–97 reactor 1 and 2, 249–250
Gantt chart, 98–99 industrial case study
hot and cold streams, 97–98 computational problem short-term
initial inventory and storage size data, 35–36 scheduling techniques, 267–270
Kondili study, 39–41 maximum inlet concentration, 267
material and utility data, 96, 98 product types, 266
outlet streams, 99–102 residue mass, 267
processing time data, 35 mass balance constraints
process representation, 96–97 linearization, 239–240, 271–273
raw material, 100–102 without storage, 237–239
recipe representation, 3–4 with storage, 240–242
utility cost, 99 mathematical formulation
Vooradi and Shaik study, 41–44 cyclic scheduling, 234–235
parameters, 236–237
T sets, 235
variables, 235–236
Time average model (TAM), 110, 202, 205, 214 model, 200–202, 208–211
Time horizon constraints problem statement, 233–234
binary variables, 34 production scheduling constraints
cyclic scheduling, 105, 278 capacity, 246
feasibility, 183–184 cycle length, 247–248
production scheduling, 247 duration, 246
Time points feasibility and time horizon, 247
Ierapetritou and Floudas study, 57–64 mass balance, 248
industrial case study, 75–80 material balance, 246
Kondili study, 64–69 objective function, 248–249
Majozi and Zhu study, 57–64 solution procedure, 249
motivation, 49–50 scheduling constraints
problem statement, 50 storage, 243–245
SSN washing operation, 242–243
allocation constraints, 51 sequencing constraints
capacity constraints, 51 recycle/reuse streams, 243–244
duration constraints, 51 storage, 243–245
effective state, 50 Sundaramoorthy and Karimi work, 258–266
intermediate state, 52 Wastewater regenerator
material balance, 51 mathematical formulation, 191–192
objective function, 54–55 mass balance constraints, 174–178
optimal number, 55–57 objective function, 184–185
sequence constraints, 51–52 parameters, 174
storage constraints, 52–53 scheduling constraints, 178–183
time horizon constraints, 53–54 sets, 172
Sundaramoorthy and Karimi study, 67–75 time horizon constraints, 183–184
Time slice model (TSM), 202 variables, 173–174
MILP problem, 194
U motivation, 169–171
pharmaceuticals production plant, 188–191
Unit-specific models, 16–20 problem statement, 171
Unlimited intermediate storage (UIS), 4, 6, 20 production result, 186–188
Unlimited wait (UW), 4 Sherali and Alameddine method, 192–194
SSN representation, 185–186
W superstructure, 171–172
Wastewater minimization
BATCH 1 Z
data pertaining, 249–253
48 h time horizon, 253–258 Zero wait (ZW), 4

PDF

Uploaded by

Copyright:

Available Formats

PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PDF

Uploaded by

Copyright:

Available Formats

Free ebooks ==> www.Ebook777.

© 2017 by Taylor & Francis Group, LLC

No claim to original U.S. Government works

Printed on acid-free paper

International Standard Book Number-13: 978-1-4987-7317-1 (Hardback)

Library of Congress Cataloging-in-Publication Data

Visit the Taylor & Francis Web site at

and the CRC Press Web site at

Chapter 1 Introduction to Batch Processes............................................................1

Chapter 2 Modelling for Effective Solutions: Reduction of Binary Variables.... 27

2.4.6.1 Sequence Constraints for Different

Chapter 3 Methods to Reduce Computational Time: Prediction of Time Points..... 49

3.4.10 Calculation of the Total Duration a Unit Is

Chapter 4 Integration of Scheduling and Heat Integration:

Chapter 5 Heat Integration in Multipurpose Batch Plants................................. 109

5.2.2.1 Models Developed for Direct Heat

Chapter 6 Design and Synthesis of Heat-Integrated Batch Plants Using

6.4.9.2 If an Intermediate State Is Produced

Chapter 7 Simultaneous Scheduling and Water Optimization:

Chapter 8 Optimization of Energy and Water Use in Multipurpose Batch

Chapter 9 Targeting for Long-Term Time Horizons: Water Optimization........ 233

9.5 Industrial Case Study.............................................................266

Chapter 10 Long-Term Heat Integration in Multipurpose Batch Plants

Professor Dominic C. Y. Foo, PEng, CEng, FHEA, MIEM

• The National Research Foundation (NRF) and the Department of Science

Jui-Yuan Lee is an assistant professor at the Department of Chemical Engineering

1.2 BATCH PROCESS REPRESENTATION

FIGURE 1.1 Recipe representation of chemical processes.

FIGURE 1.2 STN representations for recipe representation.

FIGURE 1.3 Simple unit operation.

FIGURE 1.4 Unit operation with mixing/reaction.

FIGURE 1.5 Unit operation with splitting.

corresponding task, (2) the presence of a particular state in an operation corresponds

1.3 DIFFERENT STORAGE OPERATIONAL

• Unlimited intermediate storage (UIS)

The ZW or FW policy is used where unstable intermediate material must be pro-

1.4 MULTIPURPOSE AND MULTIPRODUCT BATCH PLANTS

Stage 1 Stage 2 Stage 3 Stage M

FIGURE 1.6 Schematic of a single-stage process with parallel units.

FIGURE 1.7 Schematic of a serial multistage process.

Stage 1 Stage 2 Stage 3 Stage 4 Stage 5 Finished

FIGURE 1.9 Schematic of a multipurpose process.

1.5 SCHEDULING TECHNIQUES FOR BATCH PROCESSES

PT:8 PT:9 PT:10 PT:4

PT:13 PT:8 PT:15 PT:7

PT:4 PT:2 PT:3

FIGURE 1.11 Graphical representation of the recipe.

FIGURE 1.12 S-graph for NIS policy for minimization of makespan.

(sequence-based) time representation (Méndez et al., 2006; Shaik et al., 2006;

1.5.2.1 Models Based on Precedence Sequencing

1.5.2.2 Models Based on Slot Representation

FIGURE 1.14 Feasible schedule by considering transfer time between units.

The formulations were based on unit-slots, stage-slots, process-slots, a variety of slot

1.6 SCHEDULING TECHNIQUES FOR CONTINUOUS AND

1.7.1 Models Based on Discrete-Time Representation

1 if task i takes place in unit j at time point p

FIGURE 1.15 Discrete-time representation.

1 if task i takes place in unit j at time point p

FIGURE 1.16 Continuous-time representation.

9.5 Industrial Case Study.............................................................266

1.2 BATCH PROCESS REPRESENTATION

1.3 DIFFERENT STORAGE OPERATIONAL

1.4 MULTIPURPOSE AND MULTIPRODUCT BATCH PLANTS

1.5 SCHEDULING TECHNIQUES FOR BATCH PROCESSES

1.5.2.1 Models Based on Precedence Sequencing

1.5.2.2 Models Based on Slot Representation

1.6 SCHEDULING TECHNIQUES FOR CONTINUOUS AND

1.7.1 Models Based on Discrete-Time Representation

1.7.2 Models Based on Global Event Point Representations

1.7.3 Models Based on Slot Time Representation

1.7.4 Models Based on Unit-Specific Event-Point Representation

2.4.3 Material Balance around a Storage Unit

2.4.5 Sequence Constraints for Processing Units

2.4.6 Sequence Constraints for Different Tasks in Different Units

2.4.6.1 Sequence Constraints for Different Tasks in Different Units

2.4.6.2 Sequence Constraints for Different Tasks in Different

2.4.6.3 Sequence Constraint That Links the Starting Time of the

2.4.7 Sequence Constraints for FIS Policy

2.4.8 Time Horizon Constraints

2.5.1 Results and Discussion

2.6.1.1 Results and Discussions

2.6.2.1 Results and Discussion

3.4 MATHEMATICAL MODEL USING STATE

3.4.3 Material Balance for Storage

3.4.5.1 Same Task in Same Unit

3.4.5.2 Different Tasks in Same Unit

3.4.6 Different Tasks in Different Units