As Almeida 20051114 PDF

Designing Reactive Distillation
Processes with Improved Eciency

economy, exergy loss and responsiveness
Designing Reactive Distillation

Processes with Improved Eciency
economy, exergy loss and responsiveness
Proefschrift
ter verkrijging van de graad van doctor

aan de Technische Universiteit Delft,
op gezag van de Rector Magnicus prof. dr. ir. J. T. Fokkema,
voorzitter van het College voor Promoties,
in het openbaar te verdedigen op maandag 14 november 2005 om 13:00 uur
door
Cristhian Pa
ul ALMEIDA-RIVERA
Ingeniero Qumico
(Escuela Politecnica Nacional, Ecuador)
Scheikundig ingenieur
geboren te Quito, Ecuador
Dit proefschrift is goedgekeurd door de promotor:

Prof. ir. J. Grievink
Samenstelling promotiecommissie:
Rector Magnicus
Prof. ir. J. Grievink
Prof. dr. G. Frens
Prof. ir. G. J. Harmsen
Prof. dr. F. Kapteijn
Prof. dr. ir. H. van den Berg
dr. A. C. Dimian
Prof. dr. ir. A. I. Stankiewicz
Prof. dr. ir. P. J. Jansens
Voorzitter
Technische Universiteit Delft, promotor
Technische Universiteit Delft
Technische Universiteit Delft/Shell Chemicals
Technische Universiteit Delft
Twente Universiteit
Universiteit van Amsterdam
Technische Universiteit Delft/DSM
Technische Universiteit Delft (reserve lid)
c 2005 by Cristhian P. Almeida-Rivera, Delft

Copyright
All rights reserved. No part of the material protected by this copyright notice may be reproduced
or utilized in any form or by any means, electronic or mechanical, including photocopying, recording
or by any information storage and retrieval system, without written permission from the author. An
electronic version of this thesis is available at http://www.library.tudelft.nl
Published by Cristhian P. Almeida-Rivera, Delft
ISBN 9-090200-37-1 / 9789090200378
Keywords: process systems engineering, reactive distillation, conceptual process design, multiechelon
design approach, life-span inspired design methodology, residue curve mapping, multilevel approach,
dynamic optimization, singularity theory, dynamic simulation, non-equilibrium thermodynamics, exergy, responsiveness
Printed by PrintPartners Ipskamp in the Netherlands
Dedicated to
my daughter Luca and
my wife Paty
Contents
1 Introduction
1.1
A Changing Environment for the Chemical Process Industry . . . . . .
1.2
Reactive Distillation Potential . . . . . . . . . . . . . . . . . . . . . . . .
1.3
Signicance of Conceptual Design in Process Systems Engineering . . .
1.4
Scope of Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.5
Outline and Scientic Novelty of the Thesis . . . . . . . . . . . . . . . .
11
2 Fundamentals of Reactive Distillation
13
2.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14
2.2
One-stage Level: Physical and Chemical (non-) Equilibrium . . . . . . .
16
2.3
Multi-stage Level: Combined Eect of Phase and Chemical Equilibrium
17
2.4
Multi-stage Level: Reactive Azeotropy . . . . . . . . . . . . . . . . . . .
20
2.5
Non-equilibrium Conditions and Rate Processes . . . . . . . . . . . . . .
23
2.6
Distributed Level: Column Structures . . . . . . . . . . . . . . . . . . .
26
2.7
Distributed Level: Hydrodynamics . . . . . . . . . . . . . . . . . . . . .
29
2.8
Flowsheet Level: Units and Connectivities . . . . . . . . . . . . . . . . .
30
2.9
Flowsheet Level: Steady-State Multiplicities . . . . . . . . . . . . . . . .
31
2.10 Summary of Design Decision Variables . . . . . . . . . . . . . . . . . . .
38
3 Conceptual Design of Reactive Distillation Processes: A Review
41
3.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
42
3.2
Graphical Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
42
3.3
Optimization-Based Methods . . . . . . . . . . . . . . . . . . . . . . . .
61
3.4
Evolutionary/Heuristic Methods . . . . . . . . . . . . . . . . . . . . . .
65
3.5
Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
70
Contents
4 A New Approach in the Conceptual Design of RD Processes
75
4.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
76
4.2
Interactions between Process Development and Process Design . . . . .
77
4.3
Structure of the Design Process . . . . . . . . . . . . . . . . . . . . . . .
79
4.4
Life-Span Performance Criteria . . . . . . . . . . . . . . . . . . . . . . .
82
4.5
Multiechelon Approach: The Framework of the Integrated Design Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
84
Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
87
4.6
5 Feasibility Analysis and Sequencing: A Residue Curve Mapping Approach

89
5.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
90
5.2
Input-Output Information Flow . . . . . . . . . . . . . . . . . . . . . . .
90
5.3
Residue Curve Mapping Technique . . . . . . . . . . . . . . . . . . . . .
91
5.4
Feasibility Analysis: An RCM-Based Approach . . . . . . . . . . . . . .
95
5.5
Case Study: Synthesis of MTBE . . . . . . . . . . . . . . . . . . . . . .
97
5.6
Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6 Spatial and Control Structure Design in Reactive Distillation

6.1
Multilevel Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.2
Simultaneous Optimization of Spatial and Control Structures in Reactive

Distillation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.3
7 Steady and Dynamic Behavioral Analysis
ii
107
127
7.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
7.2
Steady-State Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
7.3
Dynamic Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
7.4
Contents
8 A Design Approach Based on Irreversibility
155
8.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
8.2
Generic Lumped Reactive Distillation Volume Element . . . . . . . . . . 158
8.3
Integration of Volume Elements to a Column Structure . . . . . . . . . . 168
8.4
Application 1. Steady-state Entropy Production Prole in a MTBE Reactive Distillation Column . . . . . . . . . . . . . . . . . . . . . . . . . . 178
8.5
Application 2. Bi-Objective Optimization of a MTBE Reactive Distillation Column . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
8.6
Application 3. Tri-Objective Optimization of a MTBE Reactive Distillation Column: A Sensitivity-Based Approach . . . . . . . . . . . . . . . 185
8.7
Comparison Between Classical and Green Designs . . . . . . . . . . . . 189
8.8
9 Conclusions and Outlook
193
9.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
9.2
Conclusions Regarding Specic Scientic Design Questions . . . . . . . . 194
9.3
Conclusions Regarding Goal-Oriented Questions . . . . . . . . . . . . . 200
9.4
Scientic Novelty of this Work . . . . . . . . . . . . . . . . . . . . . . . 201
9.5
Outlook and Further Research . . . . . . . . . . . . . . . . . . . . . . . 205
A Model Description and D.O.F. Analysis of a RD Unit
209
A.1 Mathematical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209

A.2 Degree of Freedom Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 217
B Synthesis of MTBE: Features of the System
221
B.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

B.2 Description of the System . . . . . . . . . . . . . . . . . . . . . . . . . . 222
B.3 Thermodynamic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
B.4 Physical Properties, Reaction Equilibrium and Kinetics . . . . . . . . . 224
References
230
Summary
247
iii
Contents
Sammenvatting
251
Acknowledgements
257
Publications
261
About the author
263
Index
265
List of Symbols
267
Colophon
277
iv
List of Figures
1.1
Schematic representation of the conventional and highly task-integrated

RD unit for the synthesis of methyl acetate . . . . . . . . . . . . . . . .
Schematic representation of the relevant spatial scales in reactive distillation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14
2.2
Representation of stoichiometric and reactive distillation lines . . . . . .
19
2.3
Graphical determination of reactive azeotropy . . . . . . . . . . . . . . .
21
2.4
Phase diagram for methanol in the synthesis of MTBE expressed in terms

of transformed compositions . . . . . . . . . . . . . . . . . . . . . . . . .
23
2.5
Schematic representation of the Film Model . . . . . . . . . . . . . . . .
26
2.6
Separation train for an homogeneous catalyst . . . . . . . . . . . . . . .
27
2.7
Key design decision variables in RD . . . . . . . . . . . . . . . . . . . .
39
3.1
Method of statics analysis . . . . . . . . . . . . . . . . . . . . . . . . . .
43
3.2
Procedure for the construction of attainable region . . . . . . . . . . . .
48
3.3
Dimension reduction through transformed compositions . . . . . . . . .
50
3.4
Procedure for sketching the McCabe-Thiele diagram for an isomerization

reaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
57
Schematic representation of the phenomena vectors in the composition

space. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
58
3.6
Inuence of feed location on reactant conversion . . . . . . . . . . . . .
67
3.7
Column internals driven design: ideal reactor-separator train . . . . . .
68
3.8
Relation between conversion and reux ratio . . . . . . . . . . . . . . .
68
3.9
Procedure to estimate reactive zone height, reux ratio and column diameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
69
2.1
3.5
List of Figures
4.1
The design problem regarded as the combination of a design program

and a development program. . . . . . . . . . . . . . . . . . . . . . . . .
78
4.2
Overall design problem . . . . . . . . . . . . . . . . . . . . . . . . . . . .
80
4.3
SHEET approach for the denition of life-span performance criteria. . .
83
4.4
Multiechelon design approach in the conceptual design of RD processes:

tools and decisions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
85
Multiechelon design approach in the conceptual design of RD processes:

interstage ow of information. . . . . . . . . . . . . . . . . . . . . . . . .
86
Schematic representation of a simple batch still for the experimental determination of (non-) reactive residue curves . . . . . . . . . . . . . . . .
92
5.2
Construction of bow-tie regions in RCM . . . . . . . . . . . . . . . . . .
97
5.3
Residue curve map for the nonreactive system iC4 -MeOH-MTBE-nC4 at

11105 Pa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
99
4.5
5.1
vi
5.4
Residue curve map for the synthesis of MTBE at 11105 Pa . . . . . . . 100
5.5
Residue curve for the MTBE synthesis at 11105 Pa . . . . . . . . . . . 101
5.6
Quaternary and pseudo-azeotropes in synthesis of MTBE at 11105 Pa . 101
5.7
Schematic representation of distillation boundaries and zones for the synthesis of MTBE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.8
Residue curve map and separation sequence for zone b in the synthesis
of MTBE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.9
Residue curve map and separation sequence for zone a in the synthesis
of MTBE by reactive distillation . . . . . . . . . . . . . . . . . . . . . . 104
6.1
Representation of the overall design structure for a RD structure . . . . 110
6.2
Schematic representation of the generic lumped reactive distillation volume element (GLRDVE) . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.3
Schematic representation of the link between the input-output level and

the task level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.4
Composition proles in the synthesis of MTBE obtained by a multilevel

modeling approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.5
Control structure in the synthesis of MTBE by RD . . . . . . . . . . . . 119
6.6
Time dependence of the disturbances scenario in the dynamic optimization of MTBE synthesis by RD . . . . . . . . . . . . . . . . . . . . . . . 120
List of Figures
6.7
Dynamic behavior of the controllers input (controlled) variables in the

synthesis of MTBE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.8
Time evolution of MTBE molar fraction in the top and bottom streams
and temperature proles for the simultaneous optimization of spatial and
control structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
7.1
Schematic representation of a reactive ash for an isomerization reaction

in the liquid phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
7.2
Bifurcation diagram f-x for a reactive ash undergoing an exothermic

isomerization reaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
7.3
Codimension-1 singular points for a reactive ash . . . . . . . . . . . . . 135
7.4
Qualitatively dierent bifurcation diagrams for a reactive ash . . . . . 136
7.5
Zoomed view of gure 7.3 . . . . . . . . . . . . . . . . . . . . . . . . . . 137
7.6
Phase diagram for the reactive ash model
7.7
Eects of feed condition on feasibility boundaries . . . . . . . . . . . . . 139
7.8
Eects of feed condition on feasibility boundaries at large reaction heat
7.9
Eects of heat of reaction on codimension-1 singular points . . . . . . . 141
. . . . . . . . . . . . . . . . 138
7.10 Eects of feed condition on feasibility boundaries at large reaction heat
140
142
7.11 Combined eects of heat of reaction, activation energy and relative volatility on codimension-1 singular points . . . . . . . . . . . . . . . . . . . . 143
7.12 Schematic representation of a RD column in the synthesis of MTBE . . 147
7.13 Eect of reboiler heat duty on the temperature prole in an MTBE RD
column . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
7.14 Schematic representation of a MTBE RD column with a 44 SISO control structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
7.15 Disturbance scenarios considered for the analysis of the dynamic behavior
of a MTBE RD column . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
7.16 Comparison between steady-state proles obtained in this work and by
Wang et al. (2003) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
7.17 Time variation of MTBE product stream in the presence of deterministic
disturbance scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
8.1
Schematic representation of the generic lumped reactive distillation volume element GLRDVE . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
vii
List of Figures
8.2
Representation of a RD column as the integration of GLRDVEs . . . . . 170
8.3
Schematic representation of an ideal countercurrent heat exchanger . . . 171
8.4
Response time as a function of the thermal driving force for an idealized

heat exchanger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
8.5
Response time as a function of the thermal driving force for an idealized

heat exchanger at dierent hold-up values . . . . . . . . . . . . . . . . . 175
8.6
Utopia Point in multiobjective optimization . . . . . . . . . . . . . . . . 177
8.7
Schematic representation of a RD column in the synthesis of MTBE . . 179
8.8
Entropy production rate prole for a 15-stage RD column for MTBE

synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
8.9
Pareto optimal curve fecon versus fexergy . . . . . . . . . . . . . . . . . . 182
8.10 Normalized catalyst distribution in MTBE synthesis with respect to economic performance and exergy eciency . . . . . . . . . . . . . . . . . . 183
8.11 Entropy production rate prole for an optimal design of a MTBE RD
column based on exergy eciency (X-design) . . . . . . . . . . . . . . . 184
8.12 Driving forces as a function of the MeOH feed owrate . . . . . . . . . . 187
8.13 Response time as a function of the MeOH feed owrate . . . . . . . . . 188
8.14 Time variation of MTBE product stream for the classic and green designs
in the presence of a MeOH feed owrate disturbance . . . . . . . . . . . 190
9.1
Schematic representation of the tools and concepts required at each design echelon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
B.1 Conventional route for MTBE synthesis: two-stage H

uls -MTBE process 223
viii
List of Tables
2.1
Systems instances to be considered for the analysis of physical and chemical processes in a RD unit . . . . . . . . . . . . . . . . . . . . . . . . .
15
3.1
Combination of reactive and nonreactive sections in a RD column . . . .
55
3.2
Qualitative ngerprint of the design methods used in reactive distillation
72
4.1
Design problem statement in reactive distillation . . . . . . . . . . . . .
81
4.2
Categories of information resulting from the design process in reactive

distillation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
82
5.1
Input-output information for the feasibility analysis phase . . . . . . . .
90
5.2
Input-output information for the column sequencing phase . . . . . . . .
91
6.1
Input-output information for the internal spatial structure space . . . . 108
6.2
Nominal values in the MTBE synthesis . . . . . . . . . . . . . . . . . . . 115
6.3
Control loops in a reactive distillation stage column. . . . . . . . . . . . 118
6.4
Optimized steady-state design of a RD column for MTBE synthesis
6.5
Simulation results for the conventionally-used sequential and simultaneous approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
7.1
Input-output information for the behavior analysis space . . . . . . . . . 128
7.2
Set of governing dimensionless expressions for the reactive ash . . . . . 131
7.3
Properties of the reactive ash system . . . . . . . . . . . . . . . . . . . 132
7.4
Optimized design of a RD column for MTBE synthesis as obtained in

chapter 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
7.5
Control loops in a reactive distillation stage column . . . . . . . . . . . 149

ix
. . 121
List of Tables
8.1
Input-output information for the thermodynamic-based evaluation space 157
8.2
Set of governing expressions for an ideal heat exchanger . . . . . . . . . 172
8.3
Properties and operational parameters of the ideal heat exchanger system 173
8.4
Optimized design of a RD column for MTBE synthesis based on economic

performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
8.5
Summary of expressions of all contributions to the entropy production

in a GLRDVE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
8.6
Optimized design of a RD column for MTBE synthesis based on economic

performance and exergy eciency . . . . . . . . . . . . . . . . . . . . . . 186
8.7
Entropy produced in classical and green designs . . . . . . . . . . . . . . 189
9.1
Summary of input-output information ow . . . . . . . . . . . . . . . . 203
A.1 Degree of freedom analysis for the spatial and control design of a RD
unit: relevant variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
A.2 Degree of freedom analysis for the spatial and control design: relevant
expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
A.3 Degree of freedom analysis: results . . . . . . . . . . . . . . . . . . . . . 219
B.1 Typical compositions of C4 streams from FCC . . . . . . . . . . . . . . 222
B.2 Wilson interaction parameters for the system iC4 -MeOH-MTBE-nC4 at
11105 Pa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
B.3 Set of expressions used to predict relevant physical properties . . . . . . 226
B.4 Parameters used for the estimation of physical properties in the synthesis
of MTBE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
B.5 Temperature dependence of equilibrium constant in MTBE synthesis . . 228
B.6 Temperature dependence of kinetic constant in MTBE synthesis . . . . 229
List of Explanatory Notes
2.1
2.2
3.1
3.2
3.3
5.1
8.1
Rate-based mass and heat transfer: the lm model . . . . . . . . . . . . 25

Multiplicity regions in the synthesis of MTBE . . . . . . . . . . . . . . . 35
Fixed points in reactive distillation . . . . . . . . . . . . . . . . . . . . . 49
Reactive cascade dierence points . . . . . . . . . . . . . . . . . . . . . . 60
Mixed-integer dynamic optimization problem formulation . . . . . . . . 65
Denition of stable nodes, unstable nodes and saddles points . . . . . . 96
Utopia point in optimization problems with more than one objective
function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
B.1 Phase equilibrium intermezzo: the thermodynamic model . . . . . 225
xi
With the possible exception of the equator, everything begins

somewhere.
Peter Robert Fleming, writer (1907-1971)
Introduction
The conceptual design of reactive distillation processes is investigated in this PhD

thesis. The motivation for this research came from taking a sustainable life-span
perspective on conceptual design, in which economics and potential losses over the
process life span were taken into consideration. The technological and scientic scenarios used in this research are described in this chapter. First, the drivers for change
in the current dynamic environment of chemical processing industry are identied.
Then the reactive distillation processing is introduced. The generalities of this process
together with its technical challenges in design and operation are addressed. The scientic setting of conceptual design in process systems engineering, with an emphasis
on the key challenges in the design of reactive distillation is addressed. The scope
of this thesis is then introduced, together with a statement of the scientic questions
dealt with in the thesis. The chapter is concluded with a thesis outline and a concise
description of the scientic novelty of this research.
Chapter 1
1.1
A Changing Environment for the Chemical Process Industry
The chemical process industry is subject to a rapidly changing environment, characterized by slim prot margins and erce competitiveness. Rapid changes are not exclusively
found in the demands of society for new, high quality, safe, clean and environmentally
benign products (Herder, 1999), they can be found in the dynamics of business operations, which include global operations, competition and strategic alliances mapping,
among others.
Being able to operate at reduced costs with increasingly shorter time-to-market times is
the common denominator of successful companies, however, attaining this performance
level is not a straightforward or trivial issue. Success is dependant on coping eectively
with dynamic environments and short process development and design times. Taking
into account life span considerations of products and processes is becoming essential for
development and production activities. Special attention needs to be paid the potential
losses of resources over the process life span. Since these resources dier in nature, for
example they can be capital, raw materials, labor, energy. Implementing this life-span
aspect is a challenge for the chemical industry. Moreover, manufacturing excellence
practice needs to be pursued, with a stress on the paramount importance of stretching
prot margins, while maintaining safety procedures. In addition, society is increasingly
demanding sustainable processes and products. It is no longer innovative to say that the
chemical industry needs to take into account biospheres sustainability. Closely related
to sustainable development, risk minimization, another process aspect, must also be
taken into consideration. In todays world, processes and products must be safe for
their complete life span. Major incidents such as Flixborough (1974) with 28 casualties
and Bhopal (1984) with 4000+ casualties may irreversibly aect societys perception of
the chemical industry and should be a thing of the past.
Addressing all these process aspects, given the underlying aim of coping eectively with
the dynamic environment of short process development and design times, has resulted
in a wide set of technical responses. Examples of these responses include advanced
process control strategies and real-time optimization. Special attention is paid to the
synthesis of novel unit operations that can integrate several functions and units to give
substantial increases in process and plant eciency (Grossman and Westerberg, 2000;
Stankiewicz and Moulijn, 2002). These operations are conventionally referred to as
hybrid and intensied units, respectively and are characterized by reduced costs and
process complexity. Reactive distillation is an example of such an operation.
Introduction
1.2
Reactive Distillation Potential
1.2.1 Main Features and Successful Stories

Reactive distillation is a hybrid operation that combines two of the key tasks in chemical engineering, reaction and separation. The rst patents for this processing route
appeared in the 1920s, cf. Backhaus (1921a,b,c), but little was done with it before the
1980s Malone and Doherty (2000); Agreda and Partin (1984) when reactive distillation
gained increasing attention as an alternative process that could be used instead of the
conventional sequence chemical reaction-distillation.
The RD synthesis of methyl acetate by Eastman Chemicals is considered to be the textbook example of a task integration-based process synthesis (Stankiewicz and Moulijn,
2002; Stankiewicz, 2003, 2001; Li and Kraslawski, 2004; Siirola, 1996a) (see gure 1.1).
Using this example one can qualitatively assess the inherent value of this processing
strategy. The process costs are substantially reduced ( 80%) by the elimination of
units and the possibility of heat integration. Using task integration-based synthesis the
conventional process, consisting of 11 dierent steps and involving 28 major pieces of
Acetic acid
MeOH
Catalyst
Methyl
acetate
S02
R01
S03
S04
Methyl
acetate
S01
S06
S05
Acetic acid
rectifying
Catalyst
solvent
enhanced
distillation
S07
V01
S08
MeOH
chemical
reaction
stripping
Water
Solvent
Heavies
S09
Heavies
Water
Figure 1.1. Schematic representation of the conventional process for the synthesis of methyl acetate (left) and the highly task-integrated RD
unit (right). Legend: R01: reactor; S01: splitter; S02: extractive distillation; S03: solvent recovery; S04: MeOH recovery; S05:
extractor; S06: azeotropic column; S07,S09: ash columns; S08:
color column; V01: decanter
Chapter 1
equipment, is eectively replaced by a highly task-integrated RD unit.
The last decades have seen a signicant increase in the number of experimentally research studies dealing with RD applications. For example, Doherty and Malone (2001)
(see table 10.5) state more than 60 RD systems have been studied, with the synthesis of
methyl t -butyl ether (MTBE) and ethyl t -butyl ether (ETBE) gaining considerable attention. Taking an industrial perspective Stankiewicz (2003) lists the following processes
as potential candidates for RD technology: (i) decomposition of ethers to high purity
olens; (ii) dimerization; (iii) alkylation of aromatics and aliphatics (e.g. ethylbenzene
from ethylene and benzene, cumene from propylene and benzene); (iv ) hydroisomerizations; (v ) hydrolyses; (vi) dehydrations of ethers to alcohols; (vii) oxidative dehydrogenations; (viii) carbonylations (e.g. n-butanol from propylene and syngas); and (ix )
C1 chemistry reactions (e.g. methylal from formaldehyde and methanol). Recently, in
the frame of ne chemicals technology Omota et al. (2001, 2003) propose an innovative
RD process for the esterication reaction of fatty acids. The feasibility of this process is
rstly suggested using a smart combination of thermodynamic analysis and computer
simulation (Omota et al., 2003). Secondly, the proposed design methodology is successfully applied to a representative esterication reaction in the kinetic regime (Omota
et al., 2001).
Process development, design and operation of RD processes are highly complex tasks.
The potential benets of this intensied process come with signicant complexity in
process development and design. The nonlinear coupling of reactions, transport phenomena and phase equilibria can give rise to highly system-dependent features, possibly
leading to the presence of reactive azeotropes and/or the occurrence of steady-state multiplicities (cf. section 2.9). Furthermore, the number of design decision variables for
such an integrated unit is much higher than the overall design degrees of freedom of
separate reaction and separation units. As industrial relevance requires that design
issues are not separated from the context of process development and plant operations,
a life-span perspective was adopted for the research presented in this thesis.
1.2.2 Technical Challenges in the Process Design and Operation of Reactive Distillation
A generalized applicability of RD technology is a key challenge for the process-oriented
community. Operational applicability is seen as strategic goal coupled with the development of (conceptual) design methodologies that can be used to support the RD
decision making process. Thus, the process systems engineering community is expected
to provide tools and supporting methods that can be used to faster develop and better
operation of RD processes. Designing chemical process involves the joint consideration
of process unit development and design programs (cf. section 4.5) and these are key
4
Introduction
challenges in RD process design.
A topic that is emerging as a challenge in the RD arena, is that, due to its systemdependency, RD processing is strongly limited by its reduced operation window (P,T ).
This feasibility domain, which is determined by the overlapping area between feasible
reaction and distillation conditions (Schembecker and Tlatlik, 2003), spans a small
region of the P-T space. On top of these two an additional window could be imposed
by the equipment and material feasibility. In this context and within the development
program, the key challenges for the RD community include: (i) the introduction of novel
and more selective catalysts; (ii) the design of more eective and functional packing
structures (e.g. super X-pack (Stankiewicz, 2003)); and (iii) nding new applications.
The rst two challenges are strongly driven by the need to expand the RD operational
window beyond the current bounds for a given application.
1.3
Significance of Conceptual Design in Process Systems

Engineering
1.3.1 Scientic Setting of Conceptual Design in Process Systems Engineering

Since its introduction, process systems engineering (PSE) has been used eectively by
chemical engineers to assist the development of chemical engineering. In tying science to
engineering PSE provides engineers with the systematic design and operation methods,
tools that they require to successfully face the challenges of todays chemical-oriented
industry (Grossman and Westerberg, 2000).
At the highest level of aggregation and regardless of length scale (i.e. from micro-scale
to industrial-scale) the eld of PSE discipline relies strongly on engineers being able
to identify production systems. For the particular case of chemical engineering, a production system is dened as a purposeful sequence of physical, chemical and biological
transformations used to implement a certain function (Marquardt, 2004). A production system is characterized by its function, deliberate delimitation of its boundaries
within the environment, its internal network structure and its physical behavior and
performance. These production systems are used to transform raw materials into product materials characterized by dierent chemical identities, compositions, morphologies
and shapes. From a PSE perspective the most remarkable feature of a system is its
ability to be decomposed or aggregated in a goal-oriented manner to generate smaller
or larger systems (Frass, 2005). Evidently, the level of scrutiny is very much linked to
the trade-o between complexity and transparency.
At a lower level of aggregation a system comprises the above mentioned sequence of
transformations or processes. Thus, a process can be regarded as a realization of a
5
Chapter 1
system and is made up of an interacting set of physical, chemical or biological transformations, that are used to bring about changes in the states of matter. These states
can be chemical and biological composition, thermodynamic phases, a morphological
structure and electrical and magnetic properties.
Going one level down in the aggregation scale gives us the chemical plant. This is no
more than the physical chemical process system. It is a man-made system, a chemical
plant, in which processes are conducted and controlled to produce valuable products
in a sustainable and protable way. The conceptual process design (CPD) is made at
the following level of reduced aggregation. In the remainder of this section particular
attention is given to CPD in the context of PSE.
Since its introduction, CPD has been dened in a wide variety of ways. CPD and PSE
activities are rooted in the concept of unit operations and the various denitions of CPD
are basically process unit-inspired. The denition of CPD given by Douglas (1988) is
regarded as the one which extracts the essence of this activity. Thus, CPD is dened
as the task of nding the best process owsheet, in terms of selecting the process units
and interconnections among these units and estimating the optimum design conditions
(Goel, 2004). The best process is regarded as the one that allows for an economical,
safe and environmental responsible conversion of specic feed stream(s) into specic
product(s).
Although this CDP denition might suggest a straight-forward and viable activity, the
art of process design is complicated by the nontrivial tasks of (Grievink, 2003): (i)
identifying and sequencing the physical and chemical tasks; (ii) selecting feasible types
of unit operations to perform these tasks; (iii) nding ranges of operating conditions
per unit operation; (iv ) establishing connectivity between units with respect to mass
and energy streams; (v ) selecting suitable equipment options and dimensioning; and
(vi) control of process operations.
Moreover, the design activity increases in complexity due to the combinatorial explosion
of options. This combination of many degrees of freedom and the constraints of the
design space has its origin in one or more of the following: (i) there are many ways to
select implementations of physical/chemical/biological/information processing tasks in
unit operations/controllers; (ii) there are many topological options available to connect
the unit operations (i.e. owsheet structure), but every logically conceivable connection
is physically feasible; (iii) there is the freedom to pick the operating conditions over
a physical range, while still remaining within the domain in which the tasks can be
eectively carried out; (iv ) there is a range of conceivable operational policies; and (v )
there is a range of geometric equipment design parameters. The number of possible
combinations can easily run into many thousands.
The CPD task is carried out by specifying the state of the feeds and the targets on
6
Introduction
the output streams of a system (Doherty and Buzad, 1992; Buzad and Doherty, 1995)
and by making complex and emerging decisions. In spite of its inherent complexity,
the development of novel CPD trends has lately gained increasing interest from within
academia and industry. This phenomenon is reected in the number of scientic publications focusing on CPD research issues and its applicability in industrial practice
(Li and Kraslawski, 2004). For instance, the eective application of CPD practices
in industry has lead to large cost savings, up to 60% as reported by Harmsen et al.
(2000) and the development of intensied and multifunctional units (e.g. the welldocumented methyl acetate reactive distillation unit as mentioned by Stankiewicz and
Moulijn (2002); Harmsen and Chewter (1999); Stankiewicz (2003)).
CPD plays an important role under the umbrella of process development and engineering. As stated by Moulijn et al. (2001), process development features a continuous
interaction between experimental and design programs, together with carefully monitored cost and planning studies. The conventional course of process development involves several sequential stages: an exploratory stage, a conceptual process design, a
preliminary plant owsheet, miniplant(s) trials, trials at a pilot plant level and design
of the production plant on an industrial scale. CPD is used to provide the rst and
most inuential decision-making scenario and it is at this stage that approximately 80%
of the combined capital and operational costs of the nal production plant are xed
(Meeuse, 2003). Performing individual economic evaluations for all design alternatives
is commonly hindered by the large number of possible designs. Therefore, systematic
methods, based on process knowledge, expertise and creativity, are required to determine which will be the best design given a pool of thousands of alternatives.
1.3.2 Developments in New Processes and Retrots

From its introduction the development of CPD trends has been responding to the harmonic satisfaction of specic requirements. In the early stages of CPD development
economic considerations were the most predominant issue to be taken into account.
Seventy years on, the issues surrounding CPD methodologies have been extended to
encompass a wide range of issues involving economics, sustainability and process responsiveness (Almeida-Rivera et al., 2004b; Harmsen et al., 2000). Spatial and temporal
aspects must be taken into account when designing a process plant. Additionally, the
time dimension and loss prevention are of paramount importance if the performance
of a chemical plant is to be optimized over its manufacturing life-span. This broad
perspective accounts for the use of multiple resources (e.g. capital, raw materials and
labor) during the design phase and the manufacturing stages. In this context and in
view of the need to support the sustainability of the biosphere and human society, the
design of sustainable, environmentally benign and highly ecient processes becomes a
7
Chapter 1
major challenge for the PSE community. Identifying and monitoring potential losses in
a process, together with their causes are key tasks to be embraced by a CPD approach.
Means need to be put in place to minimize losses of mass, energy, run time availability
and, subsequently, prot. Poor process controllability and lack of plant responsiveness
to market demands are just two issues that need to be considered by CPD engineers as
causes of prot loss.
Li and Kraslawski (2004) have recently presented a detailed overview on the developments in CPD, in which they show that the levels of aggregation in CPD (i.e. micro-,
meso- and macroscales) have gradually been added to the application domain of CPD
methodologies. This refocus on the design problem has lead to a wide variety of success stories at all three scale levels and a coming to scientic maturity of the current
methodologies.
At the mesolevel the research interests have tended towards the synthesis of heat
exchange networks, reaction path kinetics and sequencing of multicomponent separation trains (Li and Kraslawski, 2004). A harmonic compromise between economics,
environmental and societal issues is the driving force at the CPD macrolevel. Under
the framework of multiobjective optimization (Clark and Westerberg, 1983), several
approaches have been derived to balance better the trade-o between protability and
environmental concerns (Almeida-Rivera et al., 2004b; Kim and Smith, 2004; Lim et al.,
1999). Complementary to this activity, an extensive list of environmental indicators
(e.g. environmental performance indicators) has been produced in the last decades
(Lim et al., 1999; Kim and Smith, 2004; Li and Kraslawski, 2004). At the CPD microlevel the motivating force has been the demand for more ecient processes with
respect to equipment volume, energy consumption and waste formation (Stankiewicz
and Moulijn, 2002). In this context a breakthrough strategy has emerged: abstraction
from the historically equipment-inspired design paradigm to a task-oriented process
synthesis. This refocus allows for the possibility of task integration and the design
of novel unit operations or microsystems, which integrate several functions/tasks and
reduce the cost and complexity of process systems (Grossman and Westerberg, 2000).
An intensied unit is normally characterized by drastic improvements, sometimes in
an order of magnitude, in production cost, process safety, controllability, time to the
market and societal acceptance (Stankiewicz, 2003; Stankiewicz and Moulijn, 2002).
Among the proven intensied processes reactive distillation (RD) occupies a place of
preference and it is this which is covered in the course of this thesis, coupled with an
emphasis on RD conceptual design.
Introduction
1.3.3 Key Challenges in the Design of Reactive Distillation

Due to its highly complex nature, the RD design task is still a challenge for the PSE
community. The following intellectually challenging problems need to be considered by
the PSE community (Grossman and Westerberg, 2000): (i) design methodologies for
sustainable and environmentally benign processes; (ii) design methodologies for intensied processes; (iii) tighter integration between design and the control of processes;
(iv ) synthesizing plantwide control systems; (v ) optimal planning and scheduling for
new product discovery; (vi) planning of process networks; (vii) exible modeling environments; (viii) life-cycle modeling; (ix ) advanced large-scale solving methods; and (x )
availability of industrial nonsensitive data. An additional PSE specic challenge is to
dene a widely accepted set of eectiveness criteria that can be used to assess process
performance. These criteria should take into account the economic, sustainability and
responsiveness/controllability performances of the design alternatives.
1.4
Scope of Research
During the course of this PhD thesis we deal with the design of grassroots reactive
distillation processes. At this high level of aggregation, the process design is far more
comprehensive than for most of the industrial activities (e.g. retrot, debottlenecking
and optimal operation of existing equipment) as it involves a wide range of domain
knowledge and augmented (design) degrees of freedom.
Here it becomes necessary to introduce an explanatory caveat regarding the concept
of life span. From a formal standpoint this term includes all stages through which
a production system or activity passes during its lifetime (Schneider and Marquardt,
2002). From a product perspective, for instance, the life span describes one, the life of
the product from cradle to grave (Meeuse, 2003; Korevaar, 2004). Two, each process
step along the cradle-to-grave path is characterized by an inventory of the energy, materials used and wastes released to the environment and an assessment of the potential
environmental impact of those emissions (Jimenez-Gonzalez et al., 2000, 2004). If our
viewpoint is the process rather than the product life span, we can foresee a sequence
of stages including need identication, research and development, process design, plant
operation and retrot/demolition. As covering all these process stages in a unique
methodology is a highly demanding task, we limited our research to the sustainability
aspects that are exclusively under the control of the design and operational phases.
Thus, in our scope of life span we do not consider any sustainability issue related to
feed-stock selection, for example, (re-)use of catalyst, (re-)use of solvent, among others.
A life-span inspired design methodology (LiSp-IDM) is suggested as the rst attempt
towards a design program strategy. Although more rened and detailed approaches
9
Chapter 1
are expected to be introduced in the future, their underlying framework will probably
remain unchanged. In this framework, a life-span perspective is adopted, accounting for
the responsible use of multiple resources in the design and manufacturing stages and a
systems ability to maintain product specications (e.g. compositions and conversion) in
a desired range when disturbances occur. In this thesis and for the rst time, economic,
sustainability and responsiveness/controllability aspects are embedded within a single
design approach. The driving force behind this perspective is derived from regarding
the design activity as a highly aggregated and large-scale task, in which process unit
development and design are jointly contemplated.
All the aforementioned features of the LiSp-IDM are captured in the following denition,
LiSp-IDM is taken to be a systematic approach that can be used to solve
design problems from a life-span perspective. In this context, LiSp-IDM is
supported by dening performance criteria that can be used to account for
the economic performance, sustainability and responsiveness of the process.
Moreover, LiSp-IDM framework allows a designer to combine, in a systematic way, the capabilities and complementary strengths of the available
graphical and optimization-based design methodologies. Additionally, this
design methodology addresses the steady-state behavior of the conceived
unit and a strong emphasis is given to the unit dynamics.
The following set of goal-oriented engineering questions were formulated based in the
above denition,
Question 1. What benets can be gained from having a more integrated design
methodology?
Question 2. What are the practical constraints that need to be considered from
a resources point of view (i.e. time, costs, tools and skill levels), when developing
and applying a design methodology in a work process?
Question 3. What are the essential ingredients for such a design methodology?
Additionally, a set of specic scientic design questions were formulated based on the
steps of a generic design cycle,
Question 4. What is the domain knowledge required and which new building
blocks are needed for process synthesis?
Question 5. What are the (performance) criteria that need to be considered from
a life-span perspective when specifying a reactive distillation design problem?
10
Introduction
Question 6. What (new) methods and tools are needed for reactive distillation
process synthesis, analysis and evaluation?
Question 7. Are there structural dierences and signicant improvements in
designs derived using conventional methodologies from those obtained using an
integrated design methodology?
These questions will be either qualitatively or quantitatively answered during the course
of this thesis.
1.5
Outline and Scientific Novelty of the Thesis
This dissertation is divided into several chapters, covering the conceptual design of grassroots RD processes. The fundamentals, weaknesses and opportunities of RD processing
are addressed in chapter 2. A detailed description of the current design methodologies in
RD forms the subject of chapter 3. Special attention is paid to the identication of the
methodologies strengths and missing opportunities and a combination of methodology
capabilities is used to derive a new multiechelon design approach, which is presented in
detail in chapter 4. The essential elements of this design approach are then addressed
using the synthesis of methyl-tert butyl ether as tutorial example. Chapter 5 deals with
the feasibility analysis of RD processing based on an improved residue curve mapping
technique and column sequencing. In chapter 6, the focus is on the synthesis of internal spatial structures, in particular. A multilevel modeling approach and the dynamic
optimization of spatial and control structures in RD are introduced. The steady-state
and dynamic performance in RD form the subjects of chapter 7. The performance criteria embedded in the proposed RD methodology are covered in chapter 8, in which a
life-span perspective is adopted leading to the denition of performance criteria related
to economics, thermodynamic eciency and responsiveness. The interactions between
those performance criteria in the design of RD are explored in particular. The information generated in the previous chapters is summarized in chapter 9, where a nal
evaluation of the integrated design is presented. The chapter is concluded with remarks
and recommendations for further research in the RD eld.
The scientic novelty of this work is embedded in several areas.
Formulation of an extended design problem. A renewed and more comprehensive
design problem in RD is formulated in the wider context of process development and
engineering. The nature of the extension is found in the identication of the design
The
term echelon refers to one stage, among several under common control, in the
ow of materials and information, at which items are recorded and/or stored (source:
http://www.pnl.com.au/glossary/cid/32/t/glossary visited in August 2005).
11
Chapter 1
decision variables and their grouping into three categories: (i) those related to physical
and operational considerations; (ii) those related to spatial issues; and (iii) those related
to temporal considerations.
Integrated design methodology. An integrated design is presented (i.e. LiSpIDM) based on a detailed analysis of the current design methodologies in RD. As
industrial relevance requires that design issues are not separated from the context of
process development and plant operations, a life-span perspective was adopted. The
framework of this methodology was structured using the multiechelon approach, which
combines in a systematic way the capabilities and complementary strengths of the
available graphical and optimization-based design methodologies. This approach is
supported by a decomposition in a hierarchy of imbedded design spaces of increasing
renement. As a design progresses the level of design resolution can be increased, while
constraints on the physical feasibility of structures and operating conditions derived
from rst principles analysis can be propagated to limit the searches in the expanded
design space.
Improvement in design tools.
The proposed design methodology is supported
by improved design tools. Firstly, the residue curve mapping technique is extended
to the RD case and systematically applied to reactive mixtures outside conventional
composition ranges. This technique is found to be particularly useful for the sequencing
of (non-) reactive separation trains. Secondly, the models of the process synthesis
building blocks are rened leading to the following sub-improvements: (i) a rened
modular representation of the building blocks; (ii) changes/improvements in the models
of the building blocks; and (iii) enhancements in synthesis/analysis tools. Regarding
the last item, a multilevel modeling approach is introduced with the aim of facilitating
the decision-making task in the design of RD spatial structures.
Performance criteria.
To account for the process performance from a life-span
perspective, criteria related to economic, sustainability and responsiveness aspects are
dened and embraced in the proposed design methodology. For the rst time the interactions between economic performance, thermodynamic eciency and responsiveness
in RD process design are explored and possible trade-os are identied. This research
suggests that incorporating a sustainability-related objective in the design problem formulation might lead to promising benets from a life-span perspective. On one hand,
exergy losses are accounted for, aiming at their minimization and on the other hand
the process responsiveness is positively enhanced.
12
The important thing is not to stop questioning. Curiosity

has its own reason for existing.
Albert Einstein, scientist and Nobel prize laureate
(1879-1955)
Fundamentals of Reactive Distillation
The combination of reaction and separation processes in a single unit has been found
to generate several advantages from an economic perspective. However, from a design
and operational point of view this hybrid process is far more complex than the individual and conventional chemical reaction - distillation operation. In this chapter a
general description of the fundamentals of reactive distillation is presented. A sound
understanding and awareness of these issues will enable more intuitive explanations
of some of the particular phenomena featured by a RD unit. The starting point of our
approach is the systematic description of the physical and chemical phenomena that
occur in a RD unit. Relevant combinations of those phenomena are grouped in levels
of dierent aggregation scrutiny (i.e. one-stage, multistage, distributed and owsheet
levels). Each level deals with the particularities of the involved phenomena. Moreover, each level is wrapped-up with the identication of the design decision variables
that are available to inuence or control the phenomena under consideration.
13
Chapter 2
2.1
Introduction
It is a well-known and accepted fact that complicated interactions between chemical

reaction and separation make dicult the design and control of RD columns. These
interactions originate primarily from VLL equilibria, VL mass transfer, intra-catalyst
diusion and chemical kinetics. Moreover, they are considered to have a large inuence
on the design parameters of the unit (e.g. size and location of (non)-reactive sections,
reux ratio, feed location and throughput) and to lead to multiple steady states (Chen
et al., 2002; Jacobs and Krishna, 1993; G
uttinger and Morari, 1999b,a), complex dynamics (Baur et al., 2000; Taylor and Krishna, 2000) and reactive azeotropy (Doherty
and Malone, 2001; Malone and Doherty, 2000).
To provide a clear insight on the physical and chemical phenomena that take place
within a RD unit, a systems approach is proposed according to the following classication features,
spatial scrutiny scale: where the system can be lumped with one-stage, lumped
with multistages or distributed. Note that the spatial scales involved in this
research are approached from bottom to top. A schematic representation of these
scales is given in gure 2.1,
contact with the surroundings: where the system can be either open or closed,
equilibrium between the involved phases: where they can be either in equilibrium
or non-equilibrium, and
transient response: where the non-equilibrium system can be regarded as stationary or dynamic.
P Pcond
to sequence zone b
D2
L lcond
Fcool
F
z
L
n-1
F nfeed,L
FR,L FD,L
MTBE
(<0.1%m)
reactive
zone
interphase
Rx
gas phase
MeOH
F nfeed,V
F nL
9
10
Qreb
C4-stream
15
Fout,Lreb
V
F n+1
film
D1
M1
T Treb
F nside,L
xI
liquid phase
Fout,Lcond
1
F nside,V
yI
C2
C1
F nV
MTBE
(>99%m)
B1
MeOH
B2
lreb
Km
Spatial scale
Figure 2.1. Schematic representation of the relevant spatial scales in reactive

distillation
These four features can be smartly combined leading to 10 dierent system instances,
as listed in table 2.1. Instances 2, 5 and 8 are not further considered in this explanatory
chapter because of their reduced practical interest. As there is no interaction between
14

the system and surroundings, the closed system will gradually reach equilibrium state
due to dynamic relaxation. Furthermore, instances 4, 7 and 10 are closely linked to
their associated steady-state situation and are then embedded within instances 3, 6 and
9, respectively.
This analysis results in instances 1 (lumped one-stage, closed and equilibrium), 3
(lumped one-stage, open and non-equilibrium), 6 (lumped multistage, open and nonequilibrium) and 10 (distributed, open and non-equilibrium) for further consideration
in the course of this chapter. A rst classication of these instances is performed based
on their spatial structure, aiming to cover the whole range of physical and chemical
phenomena that occur in a RD unit. The following levels are then dened,
Level A. One-stage level: where the system is represented by a single lumped stage
[embracing instances I1 and I3 ],
Level B. Multi-stage level: where the system is represented by a set of interconnected
trays [embracing instance I6 ], and
Level C. Distributed level: where the system is dened in terms of an spatial coordinate [embracing instance I10 ].
Recalling the engineering and scientic design questions given in chapter 1, in this
chapter we address question 4, namely,
What is the domain knowledge required and which new building blocks are needed
for process synthesis?
As the nature of this chapter is explanatory, the knowledge presented is borrowed from
various scientic sources. Thus, the novelty of this chapter is exclusively given by the
systems approach adopted to address the phenomena description. Note that the spatial
scales involved in this research are approached from bottom to top.
Table 2.1. Systems instances to be considered for the analysis of physical and
chemical processes in a RD unit
Instance
Spatial level
Contact
Phase behavior
I1
lumped one-stage
closed
equilibrium
I2
I3 /I4
I5
I6 /I7
I8
I9 /I10
lumped one-stage
lumped one-stage
lumped multistage
lumped multistage
distributed
distributed
closed
open
closed
open
closed
open
non-equilibrium
non-equilibrium
non-equilibrium
non-equilibrium
non-equilibrium
non-equilibrium
Time response
dynamic relaxation
steady state/dynamic
dynamic relaxation
dynamic relaxation
15
Chapter 2
2.2
One-stage Level: Physical and Chemical (non-) Equilibrium
Phase equilibrium. For a nc -component mixture phase equilibrium is determined

when the Gibbs free energy G for the overall open system is at a minimum (Biegler
et al., 1997),
Function :
min n G =
(j)
(j)
ni ,i
i=n
c j=n
p
(j)
ni
= ni,0 +
j=1
(j)
ni
(j)
i=1 j=1
j=np
s.t. :
(j)
ni i
k=n
rx
i,k k ,
i Znc
(2.1)
k=1
0,
()
where ni is the total number of moles for component i, ni is the number of moles of
component i in phase (), n is the total number of moles, np is the number of phases
in the system, nrx is the number of chemical reactions, k is the extent of reaction k
()
and i is the chemical potential of component i in phase () and given by,
()
()
= G0i + R T ln fi
i Znc .
(2.2)
()
In the previous expression, the fugacity of component i in phase (), fi , is estimated

in terms of the fugacity coecient and components concentration in phase () (i.e.
ai = i xi ).
The necessary condition for an extremum of the minimization problem is given by the
equality of chemical potentials across phases and the thermal and mechanical equilibrium,
(j)
(k)
j, k Znp ,
(2.3)
(k)
j, k Z ,
(2.4)
P (j) = P (k) ,
j, k Znp .
(2.5)
i
T
(j)
= i ,
=T
np
For the particular case of vapor-liquid equilibrium, the necessary conditions results in
the well-known expression,
yi i P = xi i p0i ,
i Znc .
(2.6)
A detailed description of the thermodynamic model used to derive the VL equilibrium

expression 2.6 is presented in appendix B.
16

Chemical Equilibrium. For a single-phase uid and open system a fundamental
thermodynamic property relation is given by Smith and van Ness (1987),
d(n G) = (n V )dP (n S)dT +
i=1
i dni ,
(2.7)
j=nc
where dni = i d denotes the change in mole numbers of component i, i is the

stoichiometric coecient of component i, is the extent of reaction and S is the molar
entropy of the system.
Knowing that n G is a state function, an expression is obtained for the rate of change
of the total Gibbs energy of the system with the extent of reaction at constant T and
P,

i=1

(n G)
i i =
.
(2.8)
T,P
j=n
c
At equilibrium conditions the Gibbs energy expression 2.8 equals to zero and allows
one to dene the equilibrium state in terms of the measurable temperature-dependent
equilibrium constant Keq ,
R T ln Keq
i=n
c
i G0i ,
(2.9)
i=1
Keq
i=n
c
ai i ,
(2.10)
i=1
where R is the universal gas constant, ai denotes the activity of specie i and the
exponent zero represents the standard state. Note that the liquid-phase activity ai and
fugacity are related according to the expression (Malanowski and Anderko, 1992),
fiL = fi0 ai ,
(2.11)
where fi0 is the reference state fugacity, which is estimated for pure liquid at equilibrium
conditions.
2.3
Multi-stage Level: Combined Effect of Phase and Chemical Equilibrium
For the sake of clarifying the interaction between chemical and phase equilibrium in RD,
several concepts need to be introduced. Note, rstly, that the spatial conguration at
this level is assumed to allow ecient contact between the phases and in an staged-wise
operation.
The concept of stoichiometric lines captures the concentration change of a given specie
due to chemical reaction (Frey and Stichlmair, 1999b).
17
Chapter 2
Let ni0 be the number of moles of specie i at time t0 in a closed system. Chemical
reactions proceed according to their extent of reaction j (j Znrx ). After chemical
equilibrium is reached, the number of moles at anytime is is given by,

i,j j ,
i Znc , j Znrx .
(2.12)
ni = ni0 +
j
Knowing that the total number of moles in the closed system equals to,

total,j j ,
j Znrx ,
n = n0 +
(2.13)
it follows that the stoichiometric line is expresses as,

xi =
xi0 (k total xk ) + i (xk xk0 )

,
k total xk0
i Znc ,
where total is the sum of the stoichiometric coecients (total =

reference component (k Znc ).
(2.14)
i ) and k is a
Any mixture apart from the chemical reaction equilibrium reacts along these lines to
the corresponding equilibrium state (Frey and Stichlmair, 1999a). Thus, a family of
stoichiometric lines results from the variation of initial components concentration, which
1
.
intercept at the pole (gure 2.2), whose concentration is dened as xi = i total
Another term to be introduced is the widespread distillation line concept, commonly
used to depict phase equilibrium in conventional distillation. According to Frey and
Stichlmair (1999b) and Westerberg et al. (2000), distillation lines correspond to the
liquid concentration prole within a column operating at total reux. It follows from this
denition that a distillation line is characterized by sequential steps of phase equilibria
and condensation. Thus, the following sequence is adopted for the case of an equilibrium
stage model,
phase equilibrium
condensation
xi1
(2.15)
xi1 yi1
The starting point of this sequence is the reactive mixture x1 (depicted conventionally on
the chemical equilibrium line in gure 2.2). This liquid mixture is in phase equilibrium
with the vapor y1 , which is totally condensed to x1 .
It is relevant to mention that an alternative concept might be used to depict the phase
equilibrium: residue curve (Fien and Liu, 1994; Westerberg and Wahnschat, 1996). In
contrast to distillation lines, residue curves track the liquid composition in a distillation
unit operated at nite reux. Although for nite columns distillation and residue lines
dier slightly, this dierence is normally not signicant at the rst stages of design.
Graphically, the tangent of a residue curve at liquid composition x intercepts the distillation line at the vapor composition y in phase equilibrium with x (Fien and Liu,
1994). A more detailed application of residue curve is covered in chapter 5.
18

A
3*
Stoichiometric
lines
2*
3
1*
2
A*
10*
1
reactive
azeotrope
11*
10
11
12
13
12*
Figure 2.2. Representation of stoichiometric and reactive distillation lines for

the reactive system A-B-C undergoing the reaction A+B C.
Remark: denotes the pole at which stoichiometric lines coincide, xi = i /j . Legend: dashed line: stoichiometric
line; dotted line: phase-equilibrium line; continuous line: chemical equilibrium line. System feature: TCboil > TBboil > TAboil .
(adapted from Frey and Stichlmair (1999b)).
A third concept is introduced when chemical reaction and phase equilibrium phenomena
are superimposed in the unit: reactive distillation lines. The following sequence of phase
equilibriumchemical reaction steps is adopted for the evolution of reactive distillation
lines,
phase eq.
condensation
reaction
phase eq.
condensation
xi1 xi2 yi2

xi2
xi1 yi1
(2.16)
The starting point of this sequence is the reactive mixture x1 on the chemical equilibrium
line. This liquid mixture is in phase equilibrium with the vapor y1 , which is totally
condensed to x1 . Since this mixture is apart from the chemical equilibrium line, it
reacts along the stoichiometric line to the equilibrium composition x2 . As can be seen
in gure 2.2, the dierence of the slope between the stoichiometric and liquid-vapor
equilibrium lines denes the orientation of the reactive distillation lines. This dierence
in behavior allows one to identify a point, at which both the phase equilibrium and
stoichiometric lines are collinear and where liquid concentration remains unchanged.
This special point (labelled A in gure 2.2) is conventionally referred to as reactive
azeotrope and is surveyed in section 2.4.
19
Chapter 2
2.4
Multi-stage Level: Reactive Azeotropy
Unlike its nonreactive counterpart, reactive azeotropes occur in both ideal and nonideal
mixtures, limiting the products of a reactive distillation process in the same way that
ordinary azeotropy does in a nonreactive distillation operation (Doherty and Buzad,
1992; Harding and Floudas, 2001). In both cases, however, the prediction whether a
given mixture will form (reactive) azeotropes and the calculation of their compositions
are considered essential steps in the process design task (Harding et al., 1997). Due to
the high non-linearity of the thermodynamic models this prediction is not trivial at all
and requires an accurate and detailed knowledge of the phase equilibria (expressed by
residue or distillation lines), accurate reaction equilibria and the development of specialized computational methods. In addition to the diculty of enclosing all azeotropes in a
multicomponent mixture, azeotropy is closely linked to numerous phenomena occurring
in the process (e.g. run-away of nonreactive azeotropes and the vanishing of distillation
boundaries) as mentioned by Barbosa and Doherty (1988a); Song et al. (1997); Frey
and Stichlmair (1999b); Harding and Floudas (2000); Maier et al. (2000); Harding and
Floudas (2001). Accordingly, a solid understanding of the thermodynamic behavior of
azeotropes becomes a relevant issue to address in process development. It allows one to
determine whether a process is favorable and to account for the inuence of operating
conditions on process feasibility.
From a physical point of view, the necessary and sucient condition for reactive azeotropy
is that the change in concentration due to distillation is totally compensated for by the
change in concentration due to reaction (Frey and Stichlmair, 1999a,b). As mentioned
in the previous section, this condition is materialized when a stoichiometric line is colinear with the phase equilibrium line, as depicted in gure 2.2. If the residue curves
are used to represent the phase equilibrium, the reactive azeotropic compositions are
to be found according to the following graphical procedure (gure 2.3): (i) the points
of tangential contact between the residue and stoichiometric lines (marker: ) dene a
curve of potential reactive azeotropes (thick line), which runs always between singular
points; (ii) a reactive azeotrope (RAz, marker: ) then occurs at the intersection point
between the chemical equilibrium line and the curve of potential reactive azeotropes.
Based on the graphical estimation of reactive azeotropes, Frey and Stichlmair (1999a)
establishes the following rules of thumb,
Rule 1. A maximum reactive azeotrope occurs when the line of possible reactive
azeotropes runs between a local temperature maximum and a saddle point and when
there is only one point of intersection with the line of chemical equilibrium,
Rule 2. A minimum reactive azeotrope occurs when the line of possible reactive
azeotropes runs between a local temperature minimum and a saddle point and when
there is only one point of intersection with the line of chemical equilibrium.
20
A
Stoichiometric
lines
Residue
lines
RAz
Loci of reactive
azeotropes
Figure 2.3. Graphical determination of reactive azeotropy. Symbols: ():

reactive azeotrope RAz at the given chemical equilibrium constant
Keq ; : pole at which stoichiometric lines coincide, xi = i /j .
System features: chemical reaction A + B C. (adapted from
Frey and Stichlmair (1999b)).
A practical limitation of this method is imposed by its graphical nature. Thus, this
method has been only applied to systems of three components at the most undergoing
a single reaction. Extending this method to systems with nc > 3 might not be feasible
due to the physical limitation of plotting the full-component composition space.
From a dierent and more rigorous perspective, a reactive azeotrope might be characterized by the satisfaction of the following necessary and sucient conditions for a
system undergoing a single equilibrium chemical reaction (Barbosa and Doherty, 1987b;
Doherty and Buzad, 1992),
yi xi
y1 x1
=
,
1 total x1
i total xi
i [2, nc 1].
(2.17)
It is relevant to point out that the azeotropy expression 2.17 also applies to the last
i=n
i=n
component (nc ) as may be veried by knowing that i=1 c xi = i=1 c yi = 1 (Barbosa
and Doherty, 1987b).
These necessary and sucient conditions for reactive azeotropes have been generalized
and theoretically established for the case of multicomponent mixtures undergoing multiple equilibrium chemical reactions by Ung and Doherty (1995b). The starting point for
their analysis is the introduction of transformed compositions. It is widely recognized
that mole fractions are not the most convenient measures of composition for equilibrium
reactive mixtures, as they might lead to distortions in the equilibrium surfaces (Barbosa
and Doherty, 1988a; Doherty and Buzad, 1992). In order to visualize in a much more
21
Chapter 2
comprehensive manner the presence of reactive azeotropes, a transformed composition
variable has been introduced by Doherty and co-workers (Barbosa and Doherty, 1987a;
Doherty and Buzad, 1992; Ung and Doherty, 1995a,c; Okasinski and Doherty, 1997),
Xi
k xi i xk
,
k total xk
Yi
k yi i yk
,
k total yk
i Znc , i = k,
(2.18)
where k denotes a reference component that satises the following conditions for total :
(i) k is a reactant if total > 0; (ii) k is a product if total < 0, and (iii) k is a product
or reactant if total = 0.
The necessary and sucient conditions for reactive azeotropy for the multicomponent,
multireaction system can be then expressed in terms of the transformed variables (Ung
and Doherty, 1995b),
Xi = Yi ,
i Znc 2 .
(2.19)
For this system, the transformed compositions are given by the following generalized
expressions,
Xi
xi iT (Vref )1 xref
,
T
1 total
(Vref )1 xref
i Znc nrx ,
(2.20)
Yi
yi iT (Vref )1 yref
,
T
1 total
(Vref )1 yref
i Znc nrx ,
(2.21)
(nc nrx +1),nrx
..
,
.
nc ,nrx
(2.22)
where
Vref
dw
(nc nrx +1),1
..
nc ,1
dt
i,r
T
1total
(Vref )1 yref
T
1total
(Vref )1 xref
V /L,
where iT is the row vector of stoichiometric coecients for component i Znc nrx
T
in all the nrx reactions [i,1 i,2
is the row vector of the total mole
i,R ], total
i=nc
i=nc
number change in each reaction

i,1
i=1
i=1 i,nrx , ref denotes the reference
components for the nrx reactions, numbered from nc nrx + 1 to nc and Vref is the
square matrix of stoichiometric coecients for the nrx reference components in the nrx
reactions.
These azeotropy expressions 2.19 state that in the space of transformed composition
variables the bubble-point and dew-point surfaces are tangent at an azeotropic state
(Barbosa and Doherty, 1988a), allowing the azeotropes to be found easily by visual
inspection in the reactive phase diagram for the case of nc nrx 3 (gure 2.4).
For systems beyond this space, a graphical determination of azeotropes might not be
feasible.
22
Y methanol []
1
0.9
0.8
0.7
0.6
0.5
0.4
Reactive azeotrope
0.3
0.2
0.1
0.2
0.4
0.6
0.8
X methanol []
Figure 2.4. Phase diagram for methanol in the synthesis of MTBE expressed
in terms of transformed compositions (equation 2.18). Remarks:
the location of the reactive azeotrope is tracked down at the intersection of the residue curve and the line X = Y ; reacting mixtures
of various compositions are depicted. System features: operating pressure is 11105 Pa; inert nC4 is present in the mixture.
Note that the azeotropy expression 2.19 holds also for pure components and nonreactive
azeotropes surviving reaction. Moreover, the necessary and sucient condition for
reactive azeotropy resembles largely the conditions for nonreactive azeotropy in terms
of molar fractions,
xi = yi ,
2.5
i Znc .
(2.23)
Non-equilibrium Conditions and Rate Processes
Phase non-equilibrium. If a chemical potential gradient exists between two adjacent

and homogenous phases, the system is not in local phase equilibrium and a rate-limited
mass transfer phenomenon occurs. A momentum balance of specie i results in the
following expression for the driving force exerted on i as a function on the friction
23
Chapter 2
forces due to diusive motion of all other species,
fi =
i=n
c
i,j xj (udi
udi
i
j ),
i, j Znc ,
(2.24)
j=i
where f is the frictional force, udi

is the diusive velocity of specie i and i,j is the
i
friction coecient between species i and j.
This expression is the well-known Maxwell-Stefan equation (MSE) and states that the
driving force on a specie i in a mixture equals the sum of the friction forces between i
and the other species j. MSE applied to multicomponent mixtures oers obviously a
detailed description of the transfer phenomena, but on the other hand the model complexity and parametric uncertainty are considerably increased. The rate-based approach
is of particular importance in (reactive) distillation, as the use of conventional stage eciencies might be resulting in absurd variations ( to + as acknowledged by Kenig
et al. (2000); Wesselingh (1997)). If MSE 2.24 is expressed in terms of Maxwell-Stefan
diusivities D, species ux N, average concentration Ctotal and chemical potential ,
the following MSE is obtained in a multidimensional space (Wesselingh and Krishna,
2000),
j=n
c yiNVj yj NVi
yi
V
=
, i Znc ,
T,P
V
V
i
RT
Ctotal
DV
ij
j=1;j=i
(2.25)
j=n
c xiNLj xjNLi
xi
L
nc
, iZ .
RT L T,P i =
C L DL
j=1;j=i
total
ij
A fully description of the non-equilibrium mass transfer phenomenon embraces simultaneously a Maxwell-Stefan diusion model and a ow model. The most commonly used
approach is the lm model , which is addressed in more detail in the explanatory note
2.1.
Rate-limited chemical reaction. According to the universally applicable law of
mass action, the rate of chemical reaction is directly proportional to the active masses
of each of the reactants. For the particular case of a system not in chemical equilibrium,
i=n
the term i=1 c ai i is commonly called reaction quotient. At the limiting case, the nonequilibrium system tends to equilibrium state, which results in the reaction quotient
tending to Keq . The degree of separation between current system state and chemical
equilibrium might be estimated by computing the Damk
ohler number,
Da =
tprocess
.
treaction
(2.32)
According to this denition, Damk

ohler number gives the ratio of the characteristic
process time to the characteristic reaction time (Doherty and Malone, 2001) and captures eectively the major dependence of conversion, production rate and product purities on the feed owrate, catalyst level and holdup (Chen et al., 2002). For an open
24
Film model is undoubtedly the most widespread approach for the rate-based mass and heat transfer
through an interface (Wesselingh and Krishna, 2000). It eectively combines species diusion
and uid ows and is based on the assumption that the resistance to mass and heat transfer is
exclusively concentrated in a thin lm where steady state diusion and mass and heat convection
take place (gure 2.5).
Thus, the interfacial mass and heat transfer uxes are composed of a laminar, steady state
diusion term in the stagnant lm and a convection term from the complete turbulent bulk
phase (Taylor and Krishna, 1993),
vapor
V
V
NV
i = Ji + Ntotal yi ,
liquid
i=nc
I
V
= hV
HX (T T ) +
L
L
L
Ni = Ji + Ntotal xi ,
L
I
EL = hL
HX (T T ) +
EV
i=1
i Znc ,
NV
i
i=nc
i=1
V
Hi
.
i Znc ,
(2.26)
NL
i Hi ,
with
()
Ntotal
i=n
c
()
Ni
() {liquid, vapor},
(2.27)
i=1
()
()
where Ji is the diusion ux of component i in phase (), H i is the partial molar enthalpy
of component i in phase () and hHX is the heat transfer coecient.
With the assumption of perfectly mixed bulk phases and considering the ux direction (gure
2.5), the diusion ux may be re-written in terms of the concentration gradient between the
bulk and the interface,
V
V (y I y ),
JiV = Ctotal
ki,m
i
i
L
L
L
Ji = Ctotal ki,m (xi xIi ),
with
()
Ctotal =
where
()
ki,m
()
Ci
i Znc ,
(2.28)
(2.29)
is the mass transfer coecient of component i in the mixture m in phase ().
The lm model assumes additionally, no accumulation in the interface and equilibrium state at
the interface,
V
NL
i Znc ,
i = Ni ,
(2.30)
EL = EV ,
and
yiI = fk (xIi , T ),
i Znc ,
(2.31)
where fk is a nonlinear function of the distribution coecients.
Note 2.1. Rate-based mass and heat transfer: the lm model
25
Chapter 2
yI
x
NL
liquid phase
interphase
xI
gas phase
film
Figure 2.5. Schematic representation of the Film Model. The mass and heat
transfer uxes are assumed to be positive in the direction liquid
vapor.
system, for instance, with phase holdup M , feed owrate F and undergoing a chemical
reaction, the Damk
ohler number might be given as follows,
homogenously catalyzed
Da
heterogeneously catalyzed Da
hom
Mkf,0
,
F
het
Wcatkf,0
,
F
(2.33)
where kf,0 is the homogenously catalyzed forward rate constant evaluated at a reference
temperature and Wcat is the amount of catalyst in the system.
The denition of Damk
ohler number implies that no reaction takes place if Da << 1
and chemical equilibrium limit is reached if Da >> 1.
2.6
Distributed Level: Column Structures
Hardware design in RD inuences largely the performance of the unit, particularly in

the formation of reactive azeotropes (Frey and Stichlmair, 1999a) and the prediction of
multiple steady states (Baur et al., 2000, 2001d ; Baur and Krishna, 2002). Hardware
congurations in RD (e.g. type of packing and tray geometry) are closely related to
mass and heat transfer between phases and are ruled by considerations dierent to those
for conventional distillation. Moreover, they are inuenced by the state of aggregation
of the involved catalyst (Bessling et al., 1997).
Homogeneous reactive distillation takes place in bubble-cap or sieve trays columns,
where the residence time and liquid holdup needed for the chemical reaction are de26
cat.
A,B,D
A,B,C,cat
A,B,C
cat.
cat.
C
purge
Figure 2.6. Separation train for an homogeneous catalyst. Remark: the

dark sections of the columns represent reactive zones. System
features: chemical reaction A + B C; D is inert component.
(adapted from Bessling et al. (1997))
pendent on the weir height and play a mayor role in the selectivity and conversion of
the process (Baur et al., 2000). In homogeneous RD, the catalyst is the highest-boiling
component in the system and for fast reactions its recycling might involve more than a
single piece of equipment (gure 2.6). Depending on the kinetic regime for mass transfer
and chemical reaction, dierent situations might take place: (i) phase equilibrium exists
and chemical reaction is rate-limited; (ii) phases are not in equilibrium and chemical
reaction is equilibrium controlled; (iii) neither phase nor chemical reaction equilibria
hold but phases are in thermal equilibrium; and (iv ) no equilibrium occurs.
The solutions of each situation involve the simultaneous formulation of convenient chemical reaction and mass transfer models. For the case of phase equilibrium the widely
spread equilibrium-based model (Seader, 1985) should be considered. In the case of
rate limited mass transfer non-equilibrium models (Taylor and Krishna, 2000, 1993;
Wesselingh, 1997) are to be used. The eect of Damk
ohler number on the unit performance should be taken into account if the chemical reaction is rate-limited. A detailed
overview of the aforementioned situations is given in section 2.5.
Heterogeneous reactive distillation operation -sometimes referred to as catalytic
distillation- is carried out in a catalyst-containing packed column. The chemical reaction takes place (mostly) within the catalyst particle, whereas the packing allows for
simultaneous reaction and mass transfer between vapor and liquid phases (Yuxiang and
Xien, 1992b,a; Subawalla and Fair, 1999). As stated by Yuxiang and Xien (1992b), ve
mass transfer/reaction processes may occur simultaneously in a reactive bed of catalytic
27
Chapter 2
distillation: (i) mass transfer from vapor to liquid; (ii) mass transfer from liquid to catalyst granules; (iii) diusion into the pores of the catalyst granules; (iv ) adsorption of
the reactive agents on the surface of the catalyst, and (v ) chemical reaction.
Sundmacher and Homann (1999) suggest that models can be distinguished with respect
to mass transfer between the bulk phases, internal catalyst mass transport phenomena
and chemical reaction. For the cases of chemical reaction and phase (non-)equilibrium,
the models mentioned in homogeneous RD might be also used. As for the internal catalyst transport phenomena, two models are distinguished. Thus, heterogenous V/L/S
models include mass and heat transfer limitations inside the solid catalyst, whereas
quasi-homogeneous models assume that liquid bulk phase composition is present at each
catalytically active site. This last simplication allows one to eectively regard heterogenous RD as a homogenous V/L case. Heterogeneous models are computationally more
expensive than homogeneous models, requiring much more catalyst-related parameters.
In some industrial applications, however, simplied approaches (e.g. quasi-homogenous,
phase equilibrium models) are found to be well applicable, without compromising the
output accuracy and reliability (Sundmacher and Homann, 1999).
The design requirements of commercial RD packings dier to some extent from those in
conventional distillation. For RD packing design two main features are to be taken into
account (Subawalla and Fair, 1999): (i) the reaction eciency should be maximized by
providing intimate contact between vapor, liquid and solid catalyst, and (ii) the pressure
drop through the catalyst bed should be minimized. These two desired characteristics
should be reected in the design variables in packings (i.e. hold-up, pressure drop,
interfacial area and phase ratio). These key variables are detailed in section 2.7.
To improve the interphase mass transfer process a considerable variety of novel supports
have been developed. In general terms, RD units for heterogeneous operation might
be lled with traditional porous catalytic packing (e.g. Raschig rings, tea bag-, or
sandwich-type) or with structured sheet metal packing with catalytic active surface.
Industrially, the bale packing and the structured sheet-based packing are the most
commonly used RD catalytic internals. In the rst case, the particles of catalyst are
enclosed in cloth wrapped in the form of bales, which are rolled with alternating layers
of steel mesh to form a cylinder. In the second case, catalyst particles (size range:
1-3 mm (Tuchlenski et al., 2001)) are enveloped in wire gauzes, which are then bound
together in cubes. In both cases the spent catalyst is replaced with packing containing
fresh catalyst. Due to the crossed ow path followed by the liquid in the sandwich-type,
its radial distribution is about an order of magnitude higher than that reported for bale
packing. This relevant advantage enhances the interest in designing novel variations of
the sandwich-type structured packing.
In recent years considerable attention has been drawn to the use of multiphase mono28

lith structures in various applications (e.g. gas phase processing, VL reaction systems
and biotechnology (Heiszwolf et al., 2001)). Reactive distillation is not the exemption
(Schildhauer et al., 2005). Monolith structures oer attractive advantages over conventional packings regarding low-pressure drop, high-geometric surface, enhanced mass
transfer rates and ecient catalyst utilization.
2.7
Distributed Level: Hydrodynamics
As mentioned by Malone and Doherty (2000), special attention should be paid to hydrodynamics and its interactions with the chemical reaction rate (e.g. good mixing
and sucient holdup for the intended extent of reaction). Unlike conventional columns,
both the liquid and vapor ows have been found to determine the diameter of the RD
column. Moreover, the liquid ow largely determines the residence time and ow patterns inside the unit (Tuchlenski et al., 2001). For columns operating under diusion
limited conditions, a clear inuence of catalyst thickness on steady state conversion is
reported by Higler et al. (2000). All these facts lead to the necessity of deriving overall
models, capable of predicting with acceptable reliability the performance of the column
internals during RD operation.
2.7.1 Hydrodynamic Models

The development of an hydrodynamic model involves the prediction of pressure drop,
hold-up, contact areas between phases, phases ratios and residence time distributions for
a given column internal. Although various predictive models are available in the open
domain (see e.g. Kooijman and Taylor (2000)), the model developed by The University
of Texas Separations Research Program (SRP model) (Rocha et al., 2004, 1996; Fair
et al., 2000) enjoys a widespread preference within packed columns internals. Lately,
the Delft model has been introduced (Fair et al., 2000) and validated for the case of
zig-zag triangular ow channels. Note that the following descriptions are referred to a
VL system.
Holdup models are extensively reported in the open literature and validated for specic types of packing and under specic operating conditions. Generally, a holdup
expression is a function of the internals geometry, ow regime and physical properties
of the involved phases,
M = fM (geometry, F R (L, V ), P P (L, V )),
(2.34)
where F R (L, V ) is a set of functions for the liquid and vapor ow regimes and P P (L, V )
is a set of physical properties functions for the liquid and vapor phases.
29
Chapter 2
A reliable pressure drop model should contain all the hydrodynamic interactions
that occur between the liquid-gas-internal (Fair et al., 2000). Thus, the corresponding
generic expression is given by,
P = fP (geometry, , F R (V ), P P (V )),
(2.35)
where is the total friction factor, which is composed of all independent contributions
due to V/V, V/L and direction change friction.
The interfacial area is directly related to the liquid holdup in the internal (Rocha
et al., 2004). Thus, the generic correlation for the contact area between phases is,
a = fa (M ).
(2.36)
For a three-phase system special attention should be paid to the areas between all
phases (i.e. aLV , aLS and aV S ) in particular when incomplete wetting occurs.
The phases ratio in the internal is clearly dependent on the geometry and liquid
holdup, namely,
(2.37)
p = f (geometry, M ).
For the case of a monolith structure, the hydrodynamic models are slightly dierent
(Heiszwolf et al., 2001),
holdup : M = fM (geometry, F R (L, V ), P P (L, V )),
(2.38)
pressure drop : P = fP (geometry, ff r , F R (V ), P P (V )),
(2.39)
interfacial area : a = fa (H, geometry, F R (L, V )),

phases ratio :
p = f (geometry, M ),
(2.40)
(2.41)
where ff r is the two-phase friction factor.

Attaining control of the unit hydrodynamics involves the denition of a set of key
decision variables: packing parameters (geometry and functionality) and ow regime of
the involved phases.
2.8
Flowsheet Level: Units and Connectivities
Going one level up in aggregation we consider the structures of complete columns.

Conventionally, RD columns are classied in hybrid and non-hybrid units, depending
on the distribution of reactive trays inside the column. Thus, hybrid columns are
distillation columns with a reactive core and nonreactive sections (rectifying and/or
stripping sections). Non-hybrid RD columns denote columns where all trays including
condenser and reboiler are reactive (G
uttinger, 1998). Hybrid columns are commonly
30

designed for solid catalyst, whereas homogeneous catalyst would lead to non-hybrid
column designs.
An additional classication of RD units is possible among the hybrid columns. If
the reactive core is placed in the reactive section of the unit, the column is denoted as
reactive rectifying unit. Accordingly, reactive stripping units are those with reactive core
in the stripping section. The location of the reactive core involves a good compromise
between the physical properties of the reactive species and the kinetic regime of the
chemical reaction(s). In this respect, Schoenmakers and Bessling (2003a) present a
comprehensive set of principles for equipment choice in homogeneous and heterogeneous
RD. For fast chemical reactions, for instance, if the relative volatility of the component
to be removed is low, a complete column with catalytic internals is required. On the
other hand, for slow chemical reactions and high relative volatility, an evaporator and
reactor are sucient. In this case, all the energy required to evaporate the corresponding
component is provided by a single stage evaporator followed by a reactor.
The recycle structures in RD may also vary depending on the physical properties of the
reactive system and the kinetic regime of chemical reaction. For example, side stream
reactors are in particular required for the case of rate-limited chemical reactions in
heterogenous RD, whereas stirred vessels are needed in homogeneous RD.
The key design decision variables at this level of aggregation include: functionality of
unit sections, number of trays or length of sections, section diameters, feeding and withdrawal points, energy exchange, column owsheet conguration with recycle structures
and heat integration.
2.9
Flowsheet Level: Steady-State Multiplicities
The phenomenon of multiple steady states (MSS) has been known and experimentally
detected for already some decades, even for the simple case of binary distillation with
ideal vapor-liquid equilibrium (Kienle et al., 1995; Pellegrini and Possio, 1996; Bildea
and Dimian, 1999; Gr
uner and Kienle, 2004). In the framework of RD, understanding
the sources of multiplicity and their prediction within a RD unit have been topics of
challenging interest for the CACE-community (see e.g. Jacobs and Krishna (1993); Ciric
and Miao (1994); Hauan et al. (1995a,b); Gehrke and Marquardt (1997); G
uttinger and
Morari (1997); G
uttinger (1998); Higler et al. (1999b); G
uttinger and Morari (1999b,a);
Chen et al. (2002); Baur et al. (2003); Rodriguez et al. (2001, 2004)). In this section,
we address steady state multiplicity in an entire RD column structure.
31
Chapter 2
2.9.1 Generalities of Multiple Steady States

Among many others, G
uttinger (1998) and Sneesby et al. (1998a) study the occurrence
of multiple steady states in columns with reactive core (i.e. hybrid RD column). Their
research work suggests that two types of multiplicities may occur in RD,
Input multiplicities: two or more sets of input variables give the same set of
output variables. If a column of a given design and given feed (F, xF , TF ) exhibits
input multiplicities, the same product compositions xD and xB can be achieved
for dierent values of operating parameters (e.g. Da and reboiler duty). This
type of multiplicity is frequent in azeotropic and RD and has mainly economical
importance.
Output multiplicities: a column of a given design exhibits dierent column
proles and product compositions at steady state for the same set of inputs and the
same values of operating parameters. This type of multiplicity will be addressed
during the course of this section.
2.9.1.1
Sources of multiplicity
The actual causes of MSS in RD have been under discussion by the CACE-community
for the last decades. Mathematically, MSS might be originated basically from the
nonlinearity of the transformation between mass and molar ow rates and from the
interaction between ows and composition due to energy balance (Bildea and Dimian,
1999). Jacobs and Krishna (1993) suggest that MSS occur in RD due the simultaneous physical and chemical equilibrium, which in turn is reected in the topology of
the residue maps. Thus, column proles may develop along dierent paths, leading to
distinctive stable nodes. Together with this hypothesis, Hauan et al. (1995b) provide a
thermodynamic-based reasoning on the sources of MSS in the synthesis of MTBE. Based
on a lift-dilution eect, high MTBE conversions are attained when the lower section of
the column contains sucient amount of MTBE to lift all the methanol and when the
reactive mixture at the lower part of the reactive zone is highly diluted to avoid MTBE
decomposition (Hauan et al., 1995b; Stichlmair and Frey, 2001). Morari and co-workers
(G
uttinger, 1998; G
uttinger and Morari, 1999b,a) study extensively the occurrence of
multiplicities for the case of simultaneous reaction and phase equilibrium in RD systems. They present a combined approach of the // analysis (i.e. innite column
length, innite reux ow rate and innite fast reactions) with a singularity analysis of
the product composition paths to predict the existence of MSS (G
uttinger and Morari,
1997). Their study suggests that all non-hybrid RD columns (i.e. RD columns without
nonreactive sections) operating with a feed mixture inside a certain region of the transformed composition space and using molar specications will exhibit MSS caused by the
32

reactive VL equilibrium. Furthermore, for hybrid columns and for molar specications
of bottoms ow rate, the MSS are caused by the interactions between the reactive and
the nonreactive column sections. Ciric and Miao (1994) attribute the origin of MSS
to the combination of a bilinear reaction rate and a sharp volatility dierence between
the reactant feeds. The last source is also reported in the case of a two-phase reactor operating under boiling conditions (Waschler et al., 2003). Gehrke and Marquardt
(1997) apply a singularity approach to a one-stage RD column, with special emphasis
to the eect of chemical reaction stoichiometry on MSS occurrence. For esterications
it has been reported that multiplicity occurs due to the same reason as multiplicity in
conventional distillation; i.e. due to the dierent values of the heat of evaporation for
pure components. Furthermore, the interaction between the temperature dependency
of the reaction rates and the boiling temperature in the system has been suggested as
one essential physical source for MSS. Mohl et al. (1997) identify two sources of MSS
for esterication reactions: the interaction between separation and reaction (i) at simultaneous phase and reaction equilibria; and (ii) in the kinetic regime. According to
Mohl et al. (2001) output multiplicity in RD appears to be caused essentially by the
same phenomenon as in conventional distillation.
From a mathematical point of view, the sucient and necessary conditions for output
multiplicity are given by the following inequalities,
Dv
> 0,
Lv
Bv
> 0,
Qreb
(2.42)
where Dv is the volumetric distillate ow rate, Lv is the volumetric reux ow rate, Bv

is the volumetric bottoms ow rate and Qreb is the reboiler duty.
Mohl et al. (1999) mention that multiplicity in the synthesis of TAME is caused by
kinetic instabilities, clearly contrasting with the statements given by G
uttinger and
Morari (1997); Mohl et al. (1997); Jacobs and Krishna (1993). According to Rehnger and Homann (1990), MSS in both a RD unit and a conventional CSTR might
be originated by the same sources. In a later contribution, Mohl et al. (2001) suggest
the kinetic instability as a source of MSS in the synthesis of MTBE. Special attention
is paid to the inuence of transport limitations on the kinetic instabilities in heterogeneously catalyzed RD processes. The inuence of kinetics on multiplicity is investigated
by Chen et al. (2002), resulting in nding non-reported isola multiplicity behavior. Recently, Rodriguez et al. (2001, 2004) study the sources of multiplicity in a reactive ash.
It is derived from those contributions that the presence of VL equilibrium combined
with a reaction can remove or create MSS, in both one-phase and two-phase regions.
An extended classication of the steady state behavior in a reactive ash is presented
in section 7.2.1, in the context of the design frame developed in this thesis. In this
extended contribution a total of 25 qualitatively dierent bifurcation diagrams are identied. Moreover, the eect of system properties on multiplicity regions is covered.
33
Chapter 2
2.9.1.2
Implications of Multiplicity
In the previous subsection multiplicity in RD has been described and possible causes
have been discussed. This subsection addresses the consequences of multiplicity from
an industrial point of view. Undesired situations can be encountered in RD processing
due to multiplicity, namely,
Start-up operation: if more than one stable steady state is possible for a certain
set of operating conditions, an undesired steady state can be reached if the initial
conditions are unsuitable. Furthermore, a given start-up policy can send the
column to a completely inappropriate steady state (Scenna et al., 1998; Mohl
et al., 1999; Bisowarno and Tade, 2000).
Perturbations leading to steady state transition: perturbations in the feed
stream, for instance, can potentially shift the operating point to an undesired
steady state (Hauan et al., 1995a; Schrans et al., 1996; Sneesby et al., 1998b;
Mohl et al., 1999; Baur et al., 2001a, 2003).
Oscillatory behavior: steady state multiplicity can trigger oscillatory behavior,
as has been found by dynamic simulation (Hauan et al., 1995a; Mohl et al., 2001).
Dicult process control: the strong nonlinear behavior of a RD column can
seriously hinder the controllability of the process, as reported by Al-Arfaj and
Luyben (2002b,a).
Implications for dynamic behavior. The strong nonlinearity and multiplicity of a
RD column can cause undesirable behavior from a dynamic point of view, even leading
to the occurrence of steady state transitions. According to Schrans et al. (1996), if
a disturbance is large enough for a certain time, the system can jump to a dierent
steady state. Sneesby et al. (1998b) perform rigorous dynamic simulations of a MTBE
RD column that exhibits multiplicity. Certain perturbations in the feed composition
cause transitions between parallel steady states. Steady state transitions are found
only when no product ow rate is used as column input. Furthermore, the possibility
of forcing a RD column from an undesired operating point to a desired operating point
is studied, without coming up with a practical control mechanism. Mohl et al. (1999)
determine experimentally that temporary disturbances in the operating conditions (e.g.
feed ow rate) may cause a transition to an undesired steady state of the process. Using
a non-equilibrium stage model, Baur et al. (2001a) show that when operating at a
high-conversion steady state, small perturbations in the feed lead to a transition to a
low-conversion branch. Additionally, it is suggested that the type and duration of a
feed perturbation causing a transition from one to another steady state depends on the
34
Multiplicity region. The size and robustness of multiplicity regions play a key role in the design and
operation of RD units. Mohl et al. (1997) show that the multiplicity region in the parameter
space of the operating conditions occurs for high yields of desired MTBE in the bottoms but
is fairly small with respect to the bottoms ow rate. Therefore, a design can easily be moved
to the unique-steady state region by slightly adjusting the bottoms ow rate. Eldarsi and
Douglas (1998) conclude that MSS can easily be eliminated by adjusting either the methanolto-butylenes molar feed ratio, the reux ratio, or installing a cooler on the last reactive stage.
Higler et al. (1999b) nd that for a decreasing iC4 feed ow rate, the multiple steady states
tend to disappear. On the other hand, Mohl et al. (1999) state that the multiplicity region for
the column conguration used by Jacobs and Krishna (1993); Hauan et al. (1995a) and Hauan
et al. (1995b) is relatively small to be veried experimentally. Furthermore, the multiplicity
region is found to occur only at high reux ratios. According to Mohl et al. (2001) operating
in the kinetic regime can lead to a large number of dierent steady state solutions for a given
set of operating conditions. However, kinetic multiplicities can be avoided by an increase of the
Damk
ohler number, which is obtained by increasing the number of catalytically active sites, or
by decreasing the feed rate. Moreover, MSS vanish if the column pressure is increased and the
multiplicity regions for equilibrium controlled chemical reactions are much smaller than in the
kinetic case. The eect of the kinetic regime on the multiplicity region is studied additionally by
Chen et al. (2002). The solution multiplicities are predicted to disappear below a critical value
of Damk
ohler, there being a Da-range at which high conversions are expected without solution
multiplicities. These results imply that operation in the kinetic limited regime (low Damk
ohler)
without MSS is possible. Qi et al. (2002) state that the values of reux and reboil ratios where
multiplicity regions occurs, are relatively near the values where the highest iC4 conversion could
be obtained.
Note 2.2. Multiplicity regions in the synthesis of MTBE

model description, on the column conguration and on the residence time distribution
along the column height. In a later contribution, Baur et al. (2003) nd that transitions
from the low-conversion steady state to the high-conversion steady state are possible for
sucient large and long feed perturbations. During its start-up procedure, a RD column
might move to an undesired steady state, specially if the operating conditions are within
a multiplicity region (cf. explanatory note 2.2). Using dynamic simulations Scenna
et al. (1998) and Bisowarno and Tade (2000) study the eect of start-up policies on
the overall process performance. A start-up procedure is recommended (Bisowarno and
Tade, 2000) to avoid low-conversion operation. The strategy is based on stabilization
of the RD column at the optimum operating conditions before the reactants are fed.
For the case of ETBE synthesis, an optimum operation is attained (i.e. high product
purity, short start-up time and reduced energy requirement) by stabilizing exclusively
the separation-controlled region.
Implications for process control. Avoiding multiplicity is certainly an important
consideration for process control and its interactions with design (Chen et al., 2002). An
appropriate control conguration is an eective tool to avoid multiplicity. G
uttinger and
Morari (1997) show, for the MTBE column studied by Jacobs and Krishna (1993), that
35
Chapter 2
multiplicities can easily be avoided by selecting the appropriate control conguration.
Additionally, Sneesby et al. (1998a) show that an appropriate closed-loop control is able
to prevent steady state transitions. In practice, reasonably tight control of almost any
temperature within the column should be sucient to prevent a steady state transition.
Wang et al. (2003) study the control strategy of a RD column for MTBE synthesis. For
output multiplicity, oscillation between steady states can occur even without observable
disturbances. Input multiplicity results in the change in the sign of process gain around
the nominal operating condition leading to control instability. The authors conclude
that although these phenomena are present, a satisfactorily linear control is still possible
to be designed. Monroy-Loperena et al. (2000) study the control problem of an ethylene
glycol RD column. Despite the strong nonlinearity of the model equations and the
uncertainties in the dynamics, the authors are able to design a robust PI controller.
The impact of multiplicity and high nonlinearity on the control of RD synthesis of
methyl acetate is discussed by Al-Arfaj and Luyben (2002a). It is suggested that for
output multiplicity the control of open-loop unstable operating points is more dicult
than the control of open-loop stable operating points.
2.9.2 Multiple Steady States in Industrial Applications

Multiplicity has been reported in various RD applications, with special emphasis in
fuel ethers. In the case of MTBE synthesis, two conversion branches have been reported, corresponding to a low and high conversion of reactants (Jacobs and Krishna,
1993; Hauan et al., 1995a,b; Schrans et al., 1996; G
uttinger and Morari, 1997, 1999a,b;
Mohl et al., 1997; G
uttinger, 1998; Sundmacher and Homann, 1999; Mohl et al., 1999;
Baur et al., 2003). Studies on multiplicity in the production of ethylene glycol by RD
have reported an impressive amount of nine dierent steady states (Ciric and Miao,
1994; Monroy-Loperena et al., 2000). The production of methyl acetate by RD has
been studied by Al-Arfaj and Luyben (2002a), identifying three steady states: a stable
high-conversion, a stable low-conversion and an unstable intermediate-conversion steady
state. MSS have been conrmed for TAME synthesis in pilot-scale columns operating
in the kinetic regime (Mohl et al., 1999; Sundmacher and Homann, 1999; Baur et al.,
2003). Furthermore, multiplicity has been reported in the RD synthesis of ethyl t -butyl
ether by Bisowarno and Tade (2000).
2.9.3 Additional Modeling Implications for Multiple Steady States

Equilibrium (EQ) and non-equilibrium (NEQ) models are generally used to represent the governing phenomena in RD. Lately, the eect of the modeling approach on
the occurrence of MSS has become a hot topic within the CACE community. This trend
is easily evident by listing the recent publications on EQ and NEQ models and their im36

plications to MSS (e.g. Krishnamurthy and Taylor (1985); Yuxiang and Xien (1992a);
Ciric and Gu (1994); Ciric and Miao (1994); Yu et al. (1997); Wesselingh (1997); Podrebarac et al. (1998); Higler et al. (1998, 1999a,b, 2000); Higler (1999); Sundmacher
and Homann (1996); Taylor and Krishna (2000); Baur et al. (2000, 2001a,b,c, 2003);
Wolf-Maciel et al. (2001); Peng et al. (2003); Taylor et al. (2003); Zheng et al. (2004)).
The contributions of Baur et al. (2000, 2001a, 2003) and Higler et al. (2000, 1999b)
are of particular relevancy as equilibrium and non-equilibrium stage modeling are compared in terms of the occurrence of MSS. The synthesis of fuel ethers and ethylene glycol
are used as illustrative case studies. In the synthesis of MTBE, for instance, a higher
conversion is attained by introducing mass transfer resistance. This counterintuitive situation is explained by the fact that in the low conversion branch the reaction proceeds
in the reverse direction. Thus, any increment in mass transfer resistance promotes the
desired direction (Baur et al., 2000). Furthermore, the range in which MSS occur is
highly inuenced by the nature of the model. Thus, the region of multiplicity in MTBE
synthesis is narrower for NEQ modeling than the exaggerated window encountered for
EQ models (Higler et al., 2000; Baur et al., 2000). It should be mentioned that these
modeling considerations are relevant for the representation of a part of the physical
reality. In design, the implication of NEQ modeling on multiplicity could be that the
phase contacting and hydrodynamics should be made such that NEQ conditions are
realized.
Hardware congurations play a determinant role in the occurrence of MSS, as reported by Baur et al. (2000, 2001d ); Baur and Krishna (2002) for NEQ models and by
Pilavachi et al. (1997) for EQ models. Hardware congurations refer basically to the
type of packing in packed columns (e.g. catalytic bales or active Raschig rings (Baur
and Krishna, 2002)) and tray sizes and mixing pattern in trayed columns (Baur et al.,
2000). It is suggested that ooding and weeping conditions exclude unfeasible congurations when interface transfer is considered, whereas for EQ models hardware features
may exclude some of the multiple solutions.
Presence of inerts is a key issue in the design and operation of RD processes (Espinosa
et al., 1996; Giessler et al., 1998). Product composition regions can drastically change,
inuencing the eciency and operability of the RD unit. Moreover, inerts in the system
increase the occurrence of MSS since an additional azeotrope might be introduced, as
is the case in the MTBE synthesis (Hauan et al., 1995b). In that system and for highly
diluted reacting mixtures (e.g. xnC4 > 72%), the methanol activity coecient increases,
leading to a decrease in the systems temperature; this results in an increment of the
reaction equilibrium constant and the achievement of a high-purity MTBE product.
On the other hand, high conversion proles may be obtained for inert-free mixtures
provided a sucient number of reactive stages (Espinosa et al., 1996). According to
those authors, an optimal column design requires a bottom specication with a ratio
37
Chapter 2
xinert
xlimiting reagent
> 1.
The physical and chemical behavior of the system requires the denition of key design
decision variables. For one-stage and multistage levels those decision parameters and
operational variables include operational pressure and temperature, chemical reaction
kinetics (i.e. equilibrium or rate-limited chemical reaction), mass/heat transfer regime,
phase velocity, residence time of the reactive phase and temporal operation (i.e. batch,
fed batch and continuous).
2.10
Summary of Design Decision Variables
In the case of RD, design issues are far more complex than those involved in conventional distillation and may dier signicantly from case to case. For instance, the design
variables for conventional distillation include the number of trays, the feed tray location and the reux ratio for a constant molar overow system with single feed tray.
In contrast, in a RD column the liquid hold-up is the major design parameter that
determines the extent of reaction, the feed might be distributed within the unit and
constant molar overow (CMO) can not be assumed unless the reaction is thermally
neutral and stoichiometrically balanced (Ciric and Gu, 1994). Additionally, RD design
is far more complicated than conventional distillation because of: (i) the presence of
additional variables (e.g. distribution of reactive core); (ii) the nonlinear interaction
between phases and reactions; and (iii) stronger coupling between heat and mass ow
throughout a RD unit.
All the key design decision variable mentioned previously in the levels of aggregation
are summarized in this section and schematically presented in gure 2.7. In terms
of priority of decision, a rst class of key design decision variables involves temporal
considerations. The variables of interest are related to the mode of operation of the
system, namely batch, fed batch or continuous operation. The second class of design
variables includes the physical and operational conditions of the system. Those aim
to the favorable creation of enough reaction and separation driving forces and mutual
synergy. In this class design variables such as as operation pressure, temperature, reux
ratio and phase velocity are of utmost importance. Moreover, chemical reaction regime
and mass/heat transfer regime should be specied. The third and last class of design
variables embraces all key spatial parameters, ranging from establishing section with
dierent functionalities, packing parameters, number of trays or length of sections,
feeding and withdrawal points, energy exchange, section diameters, column owsheet
conguration with recycle structures and heat integration.
Note that appropriate choices for those design decision variables are closely linked to the
challenges of process design and operation of reactive distillation (cf. section 1.2.2).
Thus, for instance, the development of novel catalyst should take into account both
38
Class 1: temporal
batch
fed-batch
continuous
Class 3: spatial
section functionality
packing parameters
number of trays/length of sections
feed-, withdrawal-points
energy exchange
section diameters
column flowsheet configuration with recycle structures
heat integration
Class 2: physical+
operational
P
T
reflux ratio
phase velocities
Figure 2.7. Key design decision variables in reactive distillation resulting from
a systematic phenomenological analysis.
temporal considerations (e.g. operation mode and catalyst deactivation and spatial
considerations (e.g. scale-up factors).
Mind journey to the scientific questions.

If we take a mind journey back to the engineering/scientic design question given at the beginning of this chapter, the following
remark can be made. Regarding Question 4, the domain knowledge required for process synthesis in the case of RD is essentially
composed of three classes of design decision variables (or building
blocks): temporal, physical and operational and spatial. Each
class embraces key design variables, which are identied by approaching the physical and chemical phenomena from a systems
perspective.
39
The systems approach is a response to a pressing need for

synthesizing and analyzing complexity.
Herb Simon, Nobel prize laureate (1916-2001)
Conceptual Design of
Reactive Distillation Processes: A Review
In this chapter the design methods for reactive distillation processing developed over
the last decades are classied and described. A ngerprint of the most representative work in three categories (graphical, optimization-based and heuristic-based) is
presented and missing opportunities are identied. Special attention is paid to the
methods description, outputs, assumptions, advantages and limitations. The result
of this analysis, presented as a ngerprint chart, points forward to the development
of an integrated conceptual design approach for reactive distillation processes. This
multiechelon approach is covered in detail in the coming chapters and is a constitutive part of a highly-aggregated design methodology, which addresses the RD design
problem from a life-span inspired perspective.
Parts
of this chapter have appeared in Almeida-Rivera et al. (2004a) and are based on publications
until mid 2003
41
Chapter 3
3.1
Introduction
This chapter provides an exhaustive overview on current design methodologies with

application to reactive distillation processing. By no means the presented material
should be regarded as the ultimate overview, specially due to the dynamic developments
lately experienced by RD research. The overview ends with a list of scientic literature
published in 2003
The available approaches for the design of RD may be categorized into three main
groups: those based on graphical/topologycal considerations, those based on optimization techniques and those derived from heuristic/evolutionary considerations. In the
course of this chapter, representative methodologies of each category will be discussed
focussing in particular on their description, outputs, assumptions, advantages and limitations.
A qualitative analysis of the features that play relevant roles in RD design result in
the denition of an integrated strategy. This approach -termed multiechelon approach
and extensively described in chapter 4- combines systematically the capabilities and
complementary strengths of available graphical and optimization-based methods.
What is the domain knowledge required and which new building blocks are needed
for process synthesis?
3.2
Graphical Methods
The graphical methods are called so, because the decisions are made on the basis of
graphical information. This information is often generated using models. These approaches rely on the behavior of residue curves or distillation lines and may be classied
into two well-dened trends: (i) methods based on thermodynamic-topological analysis
of distillation lines; and (ii), methods based on composition transformations.
3.2.1 Statics Analysis

The design method based on statics analysis in RD units involves nding the complete
set of steady state modes and the set of corresponding operating parameters (Seramov et al., 1999). Since separation is the most costly task, the prime purpose of statics
analysis method is the synthesis of RD processes with reduced separation costs and with
improved reaction parameters (Seramov et al., 1999). For this purpose, several approaches are proposed, which are in general terms based on thermodynamic-topological
analysis of distillation diagrams (Giessler et al., 1999, 1998).
42
Conceptual Design of Reactive Distillation Processes: A Review

A
distillate
line of chemical
interaction
line of mass
balance
XF
X*
B
for
wa
rd
rea
ctio
nr
eg
ion
X*
reverse reaction region
chemical equilibrium
manifold
C
distillation
line
bottoms
(a) Schematic representation of a RD

sequence, where the feed stream is rstly
converted to a pseudoinitial mixture X .
(b) Distillation lines for a system undergoing

the reaction A + B C.
Figure 3.1. Method of statics analysis (adapted from Giessler et al. (1999)).
Assumptions: (i) the vapor and liquid ow rates in the column are innitely large;
(ii) the capacity of the reaction part in the column is large enough to carry out a given
conversion rate; (iii) the plant is operated at steady state and theoretical stages are
chosen; and (iv ) one reversible equilibrium reaction is considered.
These assumptions allow one to estimate the liquid composition on a tray as the composition of vapor one tray below (i.e. xi = yi+1 ) and the proles may be estimated by
using usual distillation lines.
Description: this method considers that composition change due to reaction is negligible on each stage and that RD is a succession of reaction and distillation operations
(Giessler et al., 1998), where the feed is rstly converted to a pseudoinitial mixture
-with composition X assuming a certain extent of reaction - which is then separated
in a distillation column (gure 3.1a). For that pseudo-initial mixture feasible steady
states occur when: (i) the composition X and the product compositions satisfy the
mass balance and (ii) a part of the distillation line lies inside the forward reaction rate
region.
The number of theoretical and reactive stages is determined from the distillation line
and from the intersection of the distillation line and chemical equilibrium manifold
(CEM) and represents the boundary of the forward and backward reactions) (Giessler
et al., 1999). Since there are multiple pairs of X and product composition that satisfy
the mass balance, the method sets one of the product composition as reference point
and solves for the other two (for a 3-component system) by using material balance
expressions. Thus, two of the components compositions and X lie on the same line
of mass balance (LMB) in the diagram and allow the estimation of the ratio D/B at
a certain reboil ratio only by exploring the ratio of the line segments (gure 3.1b).
43
Chapter 3
After xing the reference point, all the compositions of LMB are explored to estimate
the degree of conversion required in the reaction step. The highest conversion value
is found by iteration among the feasible degrees of conversion and denes the limiting
steady state (LSS) (i.e. condition of maximum conversion for a certain reactant (Giessler
et al., 1999)). At this condition, the location and the length of the reactive zone in
the RD column are estimated by counting all reactive stages present in the forward
reaction section. Attaining the LSS can proceed by several limiting paths, depending
on the nature of the stability of the xed point from which the distillation lines start
or end. According to Giessler et al. (1998), if LSS is not convenient, the composition
of the initial mixture, the structure of the phase equilibrium diagram and the reactions
conditions (P, T ) have to be changed and the iteration loop is performed again. The
proposed method stresses the importance of identifying all feed composition regions,
since they lead to the complete set of column congurations. Thus, the entire feed
region is classied into several sub-regions, each of which has the same characteristics
with respect to the product congurations (Giessler et al., 2001).
Advantages: this method (i) gives a straight forward solution to the question whether
or not a reaction should be performed in a RD column; (ii) allows one to synthesize
owsheets qualitatively (e.g. maximum degree of conversion, reactive zone location and
length and column sequencing) (iii) for its application this method requires little data
(i.e. feed composition, phase equilibrium model parameters, chemical equilibrium model
parameters and reaction stoichiometry); this method (iv ) provides an eective tool for
the study of strongly nonideal mixtures with multiple chemical reactions and number of
components; (v ) allows the selection of the appropriate steady states from their complete
set; (vi) simplies troublesome calculations (including expensive full-scale experiments)
reducing dramatically computational time (up to 5 times); (vii) represents a general
method that does not depend on the concrete column structure; (viii) does not require
that product compositions, extent of reaction and number of stages are xed a priori;
and (ix ) allows one to determine the stability of various product regimes and derive the
possible product compositions and the column structure for one xed feed composition
and the entire feed region.
Limitations: (i) this method assumes innite separation eciency; and (ii) matching
the operating lines with the assumed product composition can be sometimes troublesome.
3.2.2 Residue Curve Mapping Technique

The residue curve mapping (RCM) technique has traditionally been considered a powerful tool for owsheet development and preliminary design of conventional multicomponent separation processes. It represents a good approximation of actual equilibrium
44

behavior and allows one to perform feasibility analysis of separation processes where
nonideal and azeotropic mixtures are involved (Siirola, 1996b,a; Malone and Doherty,
2000). Furthermore, residue curves have been used to predict the liquid-composition
trajectories in continuous distillation units in the / case (Barbosa and Doherty,
1988a; Espinosa et al., 1996). Although for nite columns those proles dier slightly
compared to the residue curves under same isobaric conditions (Fien and Liu, 1994),
this dierence is normally considered negligible at the rst stages of design. A feasibility
criterion may be derived from RCM considerations, where feasible composition proles
should be enclosed by the total reux proles (i.e. residue curves) and the reversible
proles (i.e. proles at the minimum reux ratio) (Espinosa et al., 1999). Conceptually, residue curve mapping is a phase equilibrium representation for systems involving
azeotropes (Siirola, 1996a,b) and qualitatively studies the behavior of a (non-) reactive
mixture in a heated still. Analytically, RCM are constructed based on physical properties of the system (i.e. vapor-liquid equilibrium, liquid-liquid equilibrium and solubility
data), wherein the composition of a non-reacting liquid remaining in the system may be
determined by performing overall and component material balances (Westerberg et al.,
2000; Jimenez et al., 2000).
For multicomponent systems undergoing nrx equilibrium reactions, RCM are dened
in terms of transformed molar compositions (Ung and Doherty, 1995a; Doherty and
Buzad, 1992),
dXi
= Xi Yi ,
dw
with Xi (0) = Xi,0 ,

with
dw = dt
Xi =
T
1 total
(Vref )1 yref
T
1 total (Vref )1 xref
xi iT (Vref )1 xref
,
T
1total
(Vref )1 xref
Vref
Yi =
i Znc nrx 1 ,
(3.1)
V /M L ,
(3.2)
yi iT (Vref )1 yref
,
T
1total
(Vref )1 yref
(nc nrx +1),1

..
.
nc ,1
i,r
i Znc nrx ,
(nc nrx +1),nrx
..
,
.
nc ,nrx
(3.3)
(3.4)
where Xi and Yi are the transformed composition variables for the component i in the
liquid and vapor phases respectively (i Znc nrx ), w is a warped time, xref is the column vector composed of the liquid molar compositions of the nrx reference components,
Vref is the square matrix of stoichiometric coecients for the nrx reference components
in the nrx chemical reactions. Note that non-reference components [1, nc nrx ] and
reference components [nc nrx + 1, nc ].
This new set of variables denes a sub-region of lower dimension in the composition
space composed of the feasible product compositions for equilibrium controlled reactions. The RCM problem is solved by specifying the relation between the transformed
45
Chapter 3
compositions of the vapor and liquid phases and by using relevant properties of the
transformed compositions.
Description: (i) specify the compositions of the mixture charged to the pot (i.e.
Xi,0 = Xie ); (ii) calculate composition of the equilibrium vapor; and (iii) calculate Yi
and integrate expression 3.1 until a singular point is reached (i.e. Xi = Yi ).
For kinetically controlled reactions, an autonomous residue curve expression can be
k
M L
derived in terms of the Damk
ohler number (Da = f,refV
) and for the heating policy
L
L
V /V0 = M /M0 ,
dxi
kf
= xi yi + Da (i total xi )
r,
dw
kf,ref
with xi (0) = xi,0 ,
i Znc ,
(3.5)
where M0L is the initial mass of liquid in the pot, V0 is the initial vapor ow, kf is the
rate constant for the forward reaction, r is the composition-dependent reaction rate, ref
denotes the reference conditions of the system and dw = dt V (M L )1 .
The inuence of Damk
ohler number on the trajectory of the residue curves is evident.
For Da=0 the residue curve lines will end in the highest boiling stable node (i.e. pure
component or nonreactive azeotrope), whereas for high Da the lines will also end in
a stable node, corresponding to a pure component, chemical equilibrium point or a
reactive azeotrope.
Advantages: (i) this method provides a reliable tool to perform feasibility analysis
in RD; and (ii) for its application this method requires little data (e.g. feed compositions, phase equilibrium model parameters, chemical equilibrium model parameters
and reaction stoichiometry).
Limitations: (i) although relevant improvements have been made towards multiple
component (nc > 3) mixtures (Malone and Doherty, 2000), the RCM technique is
limited by its intrinsic graphical nature; and (ii) accurate thermodynamic data are
required to correctly describe the RD process (Barbosa and Doherty, 1988b).
RCM techniques have been used as feasibility tools to develop a generalized and systematic design approach for reactive feeds outside conventional composition ranges
(Almeida-Rivera and Grievink, 2002, 2004). For the synthesis of MTBE two separation sequences have been found that depend on the location of the reacting feed in
the composition space.
3.2.3 Attainable Region Technique

The concept of attainable region (AR) has been extensively employed in the synthesis
of reactor structures (Feinberg, 1999) and successfully extended to processes combining
simultaneous reaction, mixing and separation (Nisoli et al., 1997). In combination with
46

graphical methods, an activity based reaction-separation vector is dened, which satises the same geometric properties as the reaction vector does. Therefore, it allows one
to construct the attainable region using the existing procedure for reaction-mixing systems. In general terms and for a given system of reactions with given reaction kinetics,
the AR may be dened as the portion of concentration space that can be achieved from
a given feed composition by any combination of reaction and mixing (Nisoli et al., 1997;
Gadewar et al., 2003). Thus, this technique aims to identify feasible reactor networks
-albeit not the optimum- based on graphical properties of simple models of CSTR and
PFR.
Assumptions: [Given: c1 and c2 are two attainable product compositions]: (i) every
composition that lies on the segment line c1 c2 belongs to the AR; (ii) on the AR
boundary the reaction vector points inward, is tangent or is zero; and (iii) in the
complement of AR no reaction vector intersects the AR.
Description: for the case of CSTR a component mass balance has been expressed in
terms of Damkohler number and the vapor fraction,
M W (xF )
M W (xF ) r(x)
(x
xi xF
=
y
)
+
Da
(i total xi ),
m
i
i
i
M W (y)
M W (x) kf,min

i Znc ,
RS(xi ,yi )Da
(3.6)
where m ( Vm /Fm ) is the mass-based vapor fraction, Vm is the mass-based vapor
ow, Fm is the mass-based feed ow, M W is the average molecular weight, Da is dened
as Da = kf,min m /Fm , kf,min is the forward rate constant evaluated at the lowest
temperature on the boiling surface, m is the mass-based molar density, is the total
liquid volume, r(x) is the reaction vector, subscript m denotes mass-based values and
RS(xi , yi ) is the reaction-separation term.
For a PFR, the material balance is given by,
dxi
d
m M W (xF )
M W (xF ) r(x)
(xi yi ) +
(i total xi ),
F
Da M W (y )
M W (xF ) kf,min

c
i Zn(3.7)
,
RS(xi ,yi )
xi (0) = xi,0 ,
d
iZ ,
nc
M W (xF ) kf,min m
A(z) dz,
M W (xF )
Lm (z)
(3.8)
(3.9)
where A(z) is the cross sectional area of the PFR at the distance z [0, lPFR ], lPFR is
the length of the PFR and is a scaled time.
Expressions 3.6 and 3.7 are identical to those that describe the behavior of a CSTR and
PFR and therefore, allow one to follow the same procedure for the construction of the
AR (gure3.2).
47
Chapter 3
feed point
Da: parameter
construct
PFR and CSTR
trajectories
construct
convex hull
RS
pointing out
extend hull by
using a PFR
N
attainable
region
Figure 3.2. Procedure for the construction of attainable region. Nomenclature: RS: reaction-separation term.
Advantages: (i) this method allows one to point out promising owsheets; (ii) the
reaction-separation vector has the same geometric properties as simple reactor models,
allowing the application of the theory derived for PFRs and CSTRs; and (iii) this
method can be easily extended to multiphase systems.
Limitations: (i) economic considerations are not taken into account to rank the design
alternatives.
3.2.4 Fixed-Point Algorithm

A sound understanding on xed points (cf. explanatory note 3.1) and the prediction of
their behavior as a function of design parameters have been lately used as a design tool
for RD columns.
Assumptions: this technique is based on several assumptions, which depend on the
nature of the contacting structure. Staged columns (Buzad and Doherty, 1994): (i)
three component systems with a chemical reaction given by A A + B B C C; (ii)
ideal vapor-liquid equilibrium; (iii) iso-molar reaction; (iv ) negligible heat of reaction;
(v ) negligible heat of mixing; (vi) equal latent heats; and (vii) constant liquid hold-up
on the reactive stages. Okasinski and Doherty (1998) removed these assumptions in the
development of their more generalized design method. Packed columns (Mahajani and
Kolah, 1996): (i) applicability of the lm theory of diusion; (ii) uniform liquid composition in the bulk; (iii) negligible axial dispersion/diusion; (iv ) CMO; (v ) saturated
48
Fixed points At total reux conditions, certain points in the concentration space are reached where
the composition on successive stages is constant (Buzad and Doherty, 1994, 1995) and in the
vicinity of which the columns proles behave in a peculiar manner (Mahajani and Kolah, 1996;
Mahajani, 1999a). These xed points occur at the pure component and azeotropic compositions
(Buzad and Doherty, 1994; Okasinski and Doherty, 1998) and move to new positions as the
reux ratio (or external reboil ratio or Damk
ohler number) is changed. Fixed points can be
nodes (stable or unstable) or saddles and have a signicant inuence on the column composition
proles, minimum reux, etc. (Okasinski and Doherty, 1998). Unlike stable and unstable
nodes, the location of xed points is dependent on process parameters and represents chemical
and physical equilibrium conditions. According to Espinosa et al. (1996), xed points can be
obtained by nding tangents to residue curves that pass through the distillate or bottom streams.
In order to solve for the xed-point compositions in terms of the Damk
ohler number -particularly
in terms of the critical value Dac - continuation methods are commonly used. These methods
involve solving the design equations for a range of Damk
ohler numbers, where the lower bound
corresponds to zero and the upper bound to the critical Damk
ohler number. The criterion for
xed-point location establishes that for low Da the end of the prole will be located in the
negative reaction space (i.e. where the backward reaction is predominant), whereas for Da
above the critical value the end of the prole will be located in the positive reaction space. If
the balance expressions along the column are written as explicit nite dierence equations (i.e.
xn+1 = f (xn , p)) the stability of the xed points is determined by the eigenvalues J of the
Jacobian of f evaluated at the xed points (Okasinski and Doherty, 1998).
Note 3.1. Fixed points in reactive distillation

liquid feed; (vi) negligible heat losses to the walls, heat of reaction and heat of mixing;
(vii) total condenser; (viii) reboiler modelled as equilibrium stage with reaction; and
(ix ) constant mass transfer coecients, interfacial area and liquid hold-up.
Description: most of the pioneering work in the eld of RD design for equilibrium
controlled chemical reactions is based on transformations of the molar compositions by
Doherty and co-workers (Barbosa and Doherty, 1987a,b, 1988a,b; Doherty and Buzad,
1992; Ung and Doherty, 1995a,c). They devoted considerable eort in dening a design
strategy for equilibrium chemical reactions in staged columns. The resulting approach
is based on the boundary value design method for nonreactive columns. It introduces
a novel composition coordinate system, expressions 3.3, to transform the RD problem
into a form completely analogous to a non-RD problem, which may be solved using
conventional algorithms.
Similarly to the molar fractions, molar ow rates are transformed as suggested by
G
uttinger (1998),

()
1
T
(3.10)
Vref
xref ,
= 1 total
where denotes any of the streams present in the RD column (i.e. distillate, bottoms,
feed, liquid and vapor ows) and represents the corresponding transformed value. As
mentioned by Frey and Stichlmair (1999b), the transformed compositions can be interpreted as a projection of dierent chemical equilibrium states along their stoichiometric
49
Chapter 3
i
A
F'
B
R'
(a) Non-transformed composition space.
reactive azeotrope
(b) Transformed composition space.
Figure 3.3. Dimension reduction through transformed compositions. Systems features: chemical reaction A + B + I C + I; where I is
an inert component (adapted from Frey and Stichlmair (1999b)).
lines (gure 2.2).

These denitions present convenient and simplifying properties: (i) the dimensions
of the system are reduced, simplifying the depiction of equilibrium (gure 3.3) (Frey
and Stichlmair, 1999b; Barbosa and Doherty, 1987a); (ii) they have the same numerical
values before and after reaction; (iii) they sum up to unity; (iv ) they clearly indicate the
presence of reactive azeotropes when Xi = Yi ; (v ) the nonreactive limits are well dened;
(vi) the number of linear independent transformed composition variables coincides with
the number of independent variables that describe the chemical equilibrium problem;
and (vii) the lever rule is valid as the chemical reaction no longer impacts the material
balance.
However, the introduction of transformed variables reduces the understanding of the
phase equilibrium as a whole, hiding eventually some nonreactive azeotropes (Frey and
Stichlmair, 1999b).
For kinetically controlled reactions, the approach involves specifying all the degrees
of freedom of the system and performing expanding envelope balances (material balances around the rectifying, stripping sections and the entire column) from the ends
of the column towards the middle (Buzad and Doherty, 1994; Huss et al., 1999). The
methodology is supported by the RCM technique (cf. chapter 5) and allows both a
fast screening for reactive azeotropes and feasible designs and the computation of the
number of trays and the minimum reux ratio (Ciric and Gu, 1994). If the target product compositions are feasible, the trajectories (for the rectifying and stripping sections)
intersect. Then, for a given value of Damkohler there will be a single value of the reux
50

ratio that satises the design equations (Doherty and Buzad, 1992). Intersection of
proles is a necessary but not sucient condition. Thus, the accumulated extent of
reaction in the rectifying section plus that in the stripping section must be equal to
the initially assumed overall extent of reaction in the column (Okasinski and Doherty,
1998),
j=nrect
t
B
feed
D xD
i + B xi F xi
i F
(Da
rect
MjL kf,r
rj )rect
M0L k0
j=1
j=nstrip
t
+(Da
strip

j=1
(3.11)
MjL kf,r
rj )strip .
M0L k0
Fixed-point algorithm is not recommended by Ciric and Gu (1994) and Espinosa et al.
(1996) for systems where the reaction kinetics, heat of reaction, liquid hold-up and
residence times are the determining design variables. However, Okasinski and Doherty
(1998) successfully use it to reproduce the ethylene glycol synthesis example of Ciric
and Gu (1994).
Mahajani and Kolah (1996) successfully extend the approach of Doherty and co-workers
to packed columns, where liquid phase backmixing is totally absent. This approach relies
on the performance of mass balances over dierential elements of length dz in both
rectifying and stripping sections, which can be re-written as functions of dimensionless
numbers (e.g. Da, HTU, dimensionless ux i and ratio of characteristic times in the
V
V
L
L
km
/Ctotal
km
))).
vapor and liquid sides ( Ctotal
Rectifying section (i {1, 3}):
dyi
dz
= HTUi V ,
yi (0) = yi,0 , i Znc ,

HTUi V + Da VF g(x, T ) ,
= RR+1
RR
xi (0) = xi,0 , i Znc .
dxi
dz
Stripping section (i {1, 3}):

dyi
dz
= HTUi V ,
yi (0) = yi,0 ,
dxi
dz
i Znc ,
(3.12)

RS+1
i
F
+
Da
g(x,
T
)
,
RS
HT U V
L+F
xi (0) = xi,0 ,
i Znc .
(3.13)
V-L equilibrium in the interface (i {1, 3}):

yi i = f (i + xi ),
(3.14)
where z is the dimensionless height of the column and RR and RS are the reux and
reboil ratios, respectively.
The extent of reaction is given by,
zrect

ri dz +
i =
0
z strip
ri dz,
i {1, 3},
(3.15)
51
Chapter 3
where z rect and z strip are the lengths of the rectifying and stripping sections, respectively.
The solution procedure reported by Mahajani and Kolah (1996) for packed columns is
quite similar to that described by Buzad and Doherty (1994) for staged columns: (i)
expressions 3.12-3.14 are solved simultaneously for various values of Da, determining
the range where the stripping and rectifying proles intersect; (ii) the satisfaction of
the second necessary condition is then evaluated; and (iii) the sizing of the RD column
is performed.
Remarkable from Mahajanis contributions is the parametric analysis of the design, in
terms of identifying the dependence of HT U V with Dac . It was concluded that for
kinetically controlled reactions accompanied by slow mass transfer, the dimensionless
number (HT U V Dac ) plays a mayor role in the feasibility of the design. Thus, proles
intersect if and only if (HT U V Dac ) falls in a specic range, whose width depends
on the relative position of the xed points (Mahajani, 1999b).
For RD columns where equilibrium controlled reactions are involved, the design methodology follows the same guidelines developed by Barbosa and Doherty (1988a,b) for
staged columns. Thus, transformed variables can be introduced, reducing the dimensionality of the reactive system and simplifying the set of design equations.
It is revealed from this approach that the liquid hold-up has no direct inuence on the
performance of the column, highly resembling the nonreactive distillation case. Furthermore, the reux ratio is found to determine exclusively the feasibility of the design.
Thus, for low reux ratio values stable nodes are reached before the concentration
proles intersect (i.e. non-feasible design) whereas at high reux ratios the proles
intersect and the design turns to be feasible. The value at which the proles just touch
corresponds to the minimum reux ratio. Although the parameters HT U V and do
not play a role in the intersection of the proles, they determine the column height and
the extent of reaction.
Advantages: (i) this method is highly exible in generating alternative designs at
various design parameters (e.g. Damkohler number, reboil ratio or liquid hold-up in
reactive zones) allowing one to put realistic bounds on those parameters (Okasinski and
Doherty, 1998); and (ii) the mass and energy balances can be decoupled, allowing the
determination of the column prole only using mass balance and equilibrium equations
(Espinosa et al., 1996).
Limitations: this algorithm is limited by its inherent graphical nature, that imposes
practical limitations and complicates its extension to systems where the number of
components minus the number of independent stoichiometric reactions exceeds a value
of 3.
Fortunately, several modications and extensions of the xed-point method for staged
and packed columns are available in the open literature, aiming to relax part of the
52

assumptions on which the original algorithm was based. These relaxations include: (i)
double feed columns (Okasinski, 1998); (ii) kinetically controlled reactions, which account for the liquid hold-up in the column design and economics (Barbosa and Doherty,
1988b, 1987a,b; Buzad and Doherty, 1994, 1995; Doherty and Buzad, 1992; Mahajani,
1999a); (iii) multiple reactions (Ung and Doherty, 1995c); (iv ) dierent hold-ups per
section (Melles et al., 2000; Okasinski and Doherty, 1998); (v ) heat eects (Melles et al.,

2000; Okasinski and Doherty, 1998); and (vi) reactions where
vi = 0 (Melles et al.,
2000; Okasinski, 1998).
3.2.5 Reactive Cascades

Chadda et al. (2003) and Gadewar et al. (2003) use the concept of reactive cascades to
screen the feasibility of hybrid RD systems. This approach has been successfully applied
to the preliminary design of single-feed and double-feed RD columns. The governing
expressions for the stripping (nstrip
stages) and rectifying (nrect
stages) cascades are,
t
t
rectifying: yi,j1 = R,j yi,j + (1 R,j ) xi,j i
nc 1
iZ
,j
iZ
,j
kf
kf,ref
D
1D

r(xj )
[2, nrect
].
t
stripping: xi,j1 = R,j yi,j + (1 R,j ) xi,j i

nc 1
[2, nstrip
],
t
kf
kf,ref
D
1D
(3.16)

r(xj )
(3.17)
j
is the fraction of feed L vaporized in the j th unit; and D is a
where R,j = Lj1
Da
).
normalized Damkohler number (D = 1+Da
Description: (i) for a given operating pressure, the feasibility diagram is drawn, which
is obtained by solving the component balances for the liquid/vapor molar compositions
in the stripping/rectifying sections. The independent variable R,j is assigned a convenient value, whereas D is used as parameter to move from equilibrium to kinetically
controlled regimes; (ii) for a given D-value the column conguration (reactive or hybrid)
is selected based on the satisfaction of the product specications; (iii) the rectifying and
stripping cascades expressions are solved at the given D-value until the xed points are
reached; (iv ) depending on the column conguration, the stripping and/or rectifying
cascade expressions are solved for D=0; (v ) the transformed compositions in the liquid/vapor phases are estimated according to expressions 3.3 for the stripping/rectifying
sections; and (vi) in the transformed composition space the lever rule is applied for
a desired product composition. The design is deemed to be feasible if the lever rule
intersects both the rectifying and stripping cascade proles.
Advantages: (i) a global feasibility analysis can be performed as a function of the
production rate, catalyst concentration and liquid hold-up; (ii) this method does not
53
Chapter 3
have any restriction in the number of components or reactions; and (iii) this method is
claimed to be easily implemented.
Limitations: (i) the Damk
ohler number is assumed to be invariant on all the reactive
stages; (ii) the eect of structural aspects of the unit (e.g. feed ratio type, location of
multiple feeds, reux and reboil ratios) are not considered; and (iii) the energy balances
are left out from the design methodology.
3.2.6 Thermodynamics-Based Approach

Stichlmair and co-workers (Frey and Stichlmair, 1999b; Stichlmair and Frey, 1999) study
the RD design from a thermodynamic-driven point of view. Their approach is based on
the existence of reactive distillation lines and potential reactive azeotropes (cf. 2.4).
Description: (i) the possible products (or feasible separation regions) are determined
from the knowledge of the existing concentration proles within the column, which
in turn are represented by reactive distillation lines; (ii) the column mass balance is
included; and (iii) possible separation borders are identied.
A thorough analysis of the behavior of reactive distillation lines with respect to the
volatility values of the involved components and the feed staging suggests the following
(Bessling et al., 1997, 1998),
RD is a feasible alternative for systems where the desired products are the nodes
in a reactive distillation line diagram and their boiling points dier considerably,
in the event that one of the desired products behaves as a saddle point, the
conversion may be increased by adding a second feed stream in the column (e.g.
synthesis of methyl acetate),
the conversion of a two feed column reaches a maximum at nite reux in the
case of products that behave as saddle points,
RD is a feasible and cost-eective alternative if one of the desired products behaves
as a saddle point and,
high conversion is required at stoichiometric reactants ratio,
separation between desired and undesired products is dicult,
the boiling point dierence between desired products is large,
there is no distillation boundary line, and
chemical equilibrium is on the side of the desired products (e.g. synthesis of
methyl acetate).
54
Table 3.1. Combination of reactive and nonreactive sections (nrs) in a RD

column. (source: Bessling et al. (1997))
Reaction
AB
AB
AB
AB
AB
A+B
A+B
A+B
A+B
A+B
A+B
C
C
C
C
C+D
C+D
Inerts
C
C
C, D
C, D
D
D
Desired products
nrs
A or B
B/C
A/C
BC/D;BD/C;B/CD
AC/D;AD/C;A/CD
C
A/B
C/D
AD/B;A/BD
A/B
C/D
yes
yes
yes
yes
yes
yes
no
yes
no
no
no
all components that need a reactant for the forward or backward reaction(s) are
located in the low-dimensional reaction space,
all inert components are located in the reaction space, and,
if it is aimed to separate a product that is not present in the reaction space, a
nonreactive distillation section is required. Table 3.1 presents a general overview
for the combination of reactive and nonreactive sections for several stoichiometries.
Advantages: (i) this approach can be applied to both fast and slow chemical equilibrium reactions; and (ii) feasible products can be determined within a RD column.
Limitations: (i) nding the reactive azeotropes might be sometimes troublesome; and
(ii) detailed knowledge of phase equilibrium, reaction kinetics and residence time within
the column is required.
3.2.7 Conventional Graphical Techniques

These techniques allow the distribution of reaction zones within a column using the
McCabe-Thiele and Ponchon-Savarit methods for a given reaction conversion by adjusting the catalyst hold-up on each stage. Although the methods are currently limited
to binary reactive mixtures, they provide fundamental insights towards multiple component systems. In general terms, it is suggested that if the reaction has a heavy reactant
and a light product, the reaction zone should be placed in the rectifying section, whereas
if the reaction has a light reactant and a heavy product, the reaction zone should be
located in the stripping section (Lee et al., 2000b,c,d ).
Assumptions: (i) binary system with a single chemical reaction; (ii) CMO; and (iii)
vapor-liquid equilibrium.
55
Chapter 3
Ponchon-Savarit method. This method is traditionally considered a powerful visualization tool for the synthesis of distillation columns for binary mixtures. Thus,
it allows the designer to account graphically for the enthalpy eects that cause
a varying molar overow when stepping o stages (Lee et al., 2000b). The main
issue of this approach is the denition of the reactive cascade dierence point ,
which for an exothermic isomerization reaction occurring in the rectifying zone
moves towards the reactant and downward as the calculation goes down the column. The following nomenclature is adopted for the diagram coordinates (Lee
and Westerberg, 2000),
rect
R,n
hrect
R,n
qc
D xD cP n + cR n
,
D
hD + qc + r H n
,
D
Qcond
,
D
(3.18)
(3.19)
(3.20)
rect
where R,n
and hrect
R,n are the composition and enthalpy coordinates in the PonchonSavarit diagram, D is the distillate ow, cP is the product coecient vector, cR is
the reactant coecient vector, n is the sum of the extent of reactions occurring in
stages n and above, h is the liquid enthalpy, Qcond is the condenser duty and r H
is the heat of reaction. It is concluded from this study that the number of reactive
stages and the location of the feed stage are highly dependent on the extent of
reaction in the RD column. Thus, the determination of the optimal feed stage is
possible by connecting the bottom dierence point and the nal dierence point
of the rectifying section, which in turn depend on the assumed extent of reaction
on each stage (Lee et al., 2000b).
McCabe-Thiele method. This method is used to assess qualitatively the impact of

changing a design variable (Lee et al., 2000c). For the RD case, two main features
are tracked to sketch the diagram for a binary mixture of reactants undergoing
an isomerization reaction and under CMO assumption,
the intersection point of the operating line with the y = x line denes the
reactive cascade dierence point, which moves proportionally as the ratio
molar extent of reaction to product owrate (i.e. xn = xD n /D); where
n denotes the stage number,
the heat supplied by the reaction decreases the slope of the operating line
provided that the location of the reactive cascade dierence point is not
changed.
A simplied procedure for sketching the McCabe-Thiele diagram is depicted in
gure 3.4 for an isomerization reaction.
56
assume total
extent of reaction
by material balance
choose a bottom purity

above the purity of the
original feed
based upon feed

composition and quality
determine
D and B
assume top and

bottom product
purities
construct
q-line
assume an extent
of reaction
for stage n
switch between operating

lines leading to the
fewest stages
decide the optimal

feed tray location
n=n+1
move difference point
difference point does not

move anymore
Y
passing through the new
difference point
construct
operating line
more reactive
stages
intersect operaring line

with the stripping line
on the q-line
Figure 3.4. Procedure for sketching the McCabe-Thiele diagram. Systems

features: isomerization reaction of the type A 2B (adapted
from Lee et al. (2000c)).
Advantages: (i) these methods allow a transparent visualization of tray-by-tray calculations; (ii) they might be fundamental tools for the conceptual design of RD column;
and (iii) they allow one to visualize the impact of the location of the reactive sections
and the feed location within the column.
Limitations:(i) these methods are exclusively applied to isomerization reactions; and
(ii) they are limited by their inherent graphical nature.
3.2.8 Phenomena-Based Approach

According to this method -proposed by Hauan (1998)- three independent phenomena
take place simultaneously in a RD column: mixing, separation and reaction (Hauan and
Lien, 1996; Westerberg et al., 2000). These phenomena may be represented by vectors
(gure 3.5) and allow the description of the total process prole of the system as a
57
Chapter 3
separation
mixing
reaction
Figure 3.5. Schematic representation of the phenomena vectors in the composition space.
trajectory in the composition space for a control volume moving through a complete
sequence of process units.
Thus, each step along the prole may be represented by the vectorial addition of the
phenomena vector mixture (M ), separation ( S ) and reaction ( R ),

dx
= M + S + R,
dt
with x(0) = x0 ,
where x is the phase composition and the phenomena vectors are given by,
x01
xfeed
1
..
,
M = M
.
feed
0
xnc xnc
x1 y1
..
S = S
.
xnc ync
1 x01 i
..
,
R = R
.

0
nc xnc i
(3.21)
(3.22)
is the
where superscripts feed and 0 denote feed and initial values respectively, M
length of the mixing vector and is determined by the relative amount of material in
the streams being mixed, S is a matrix reecting average mass transfer rates and R
is a scalar calculated from catalyst activity, hold-up, temperature and pressure. Each
vector has a unique local direction (i.e. process feasibility) and vector length (i.e.
process eciency). Vector directions are calculated by physical and chemical data only,
whereas vector lengths depend on the operating parameters and design variables.
58

Description: any composition change in the system due to chemical reaction may
be represented by a straight line from the singular point, termed dierence point (i.e.

xs = / ), through the local composition x0 (Hauan et al., 2000b),

,
(3.23)
x = x0 + (xs x0 ) 0
n +
where n is the total number of moles.
As the local compositions may be those of feeds and chemical reaction products, the
composition in the desired point may be expressed as a normalized linear combination
of the reaction dierence points and distillate-, bottoms- or feed-composition vectors.
From expression 3.21, a kinetic xed point occurs when the vectorial contributions of
the nonzero phenomena balance each other,
dx

= M + S + R = 0,
dt
with x(0) = x0 .
(3.24)
Fixed points, which become evident by the presence of at concentration proles for
large portions of the column, are by no means desired in RD and indicate a weak
design or operation. Therefore, several ways to move away from a xed point in the
composition space are suggested,
altering the mixing vector: by changing process variables (e.g. reux and reboil
policies, external feeds and hold-ups),
altering the reaction vector length: by manipulating process variables (e.g. catalyst density, hold-up) or by introducing nonreactive stages in the reactive zone,
incorporating pumparound reactors: in order to avoid a counter-productive separation vector.
Advantages: (i) only physical and chemical data are required to estimate the phenomena vectors, which are independent of the structural design of the unit; and (ii) this
approach allows the designer to assess independently the eect of equipment structure
and operating policies on the composition change in RD.
Limitations: (i) although this method is driven by fundamental considerations, its
applicability requires further studies.
3.2.9 Scalar/Vectorial Dierence Points Technique

The concepts of reactive cascade and scalar/vectorial dierence points in reactive cascades (cf. explanatory note 3.2) have been successfully applied by Hauan et al. (2000a);
Lee et al. (2000a) and Lee and Westerberg (2000) in the design of RD processes.
59
Chapter 3
Reactive cascade dierence points are dened as the combined points of stoichiometric coecients vectors and product compositions. For the isomolar reaction 2R P1 + P2 , the dierence
points in the rectifying and stripping sections are given by,
DxD +2n (cRcP)
,
D+2n 2n
strip
BxB +2n (cRcP)
,
R,n =
B+2n 2n
rect =
R,n
(3.25)
where the material balances are estimated as follows,

rectifying:
stripping:
rect ,
Vn+1 yn+1 = xn Ln + (D + 2n 2n )R,n
strip
Vs ys = xs1 Ls1 (D + 2s 2s )R,s
(3.26)
where subscripts s and n denote the number of stages in the stripping and rectifying sections,
respectively. The dierence points move parallel to the vector (cRcP), in the direction dened
by the dominant chemical reaction.
Note 3.2. Reactive cascade dierence points

Assumptions: (i) known top and bottom compositions; (ii) known reux ratio; (iii)
CMO assumption; and (iv ) chemical reaction of the type 2R P1 + P2 .
Based on rigorous graphical considerations, the number and location of the reactive
and nonreactive stages is obtained. For the system studied by Lee and Westerberg
(2000), the reaction conversion of the feed stage is found to be always larger than that
of any single stage in the rectifying and stripping sections. This situation allows one
to guarantee the conversion on the feed stage at any reux ratio. The location of the
bubble point curve with respect to the chemical equilibrium curve plays a key role in the
feasibility of the process within the stripping and rectifying sections. Thus, if the bubble
point curve lies to the right of the reaction equilibrium curve, the rectifying section shows
a better performance towards reactants conversion than that of the stripping section.
In this case, the forward reaction always takes place and enhances the separation task.
This approach is tested in the gas-phase disproportionation of toluene and in the dehydratation of methanol into water and dimethyl ether. In the rst case study the
most competitive conguration is reported to be the reaction zone in the stripping and
feed sections. The eect of the reaction equilibrium constant is analyzed in the second
example, resulting in feasible congurations with a single reactive stage.
Description [for a given phase and reaction equilibria in a stage RD column]: (i)
for a liquid phase chemical reaction at the operating pressure P draw the reaction
equilibrium and the bubble point curves (for gas phase chemical reactions, the dew
point curve should be drawn and an equivalent procedure should be followed (Lee and
Westerberg, 2000)); (ii) specify the top and bottom compositions, reboil and reux
ratios; (iii) step up/down from the feed stage to the top/bottom stage; (iv ) from
graphical considerations (according to expressions 3.25, 3.26 and phase equilibrium
relationship), the accumulated extents of reactions for stripping/rectifying sections are
determined; (iv ) the liquid composition on the tray is calculated; and (v ) tray-by-tray
60

calculations are performed upwards/downwards until the calculated liquid composition
equalizes the top/bottom product composition.
3.3
Optimization-Based Methods
This second group of design methods is composed of those methodologies, which are
supported by the power of computational subroutines and take into account important
elements of an overall design strategy (e.g. types of phase equilibrium, number of
components, occurrence or not of chemical reaction on a given stage) (Giessler et al.,
1998). In this category, the work of Ciric and Miao (1994); Ciric and Gu (1994);
Pekkanen (1995); Cardoso et al. (2000); Bessling et al. (1997); Seferlis and Grievink
(2001); Bansal et al. (2000); Bansal (2000); Bedenik et al. (2001); Poth et al. (2001);
Frey and Stichlmair (2001); Stichlmair and Frey (2001); Gumus and Ciric (1997) and
Georgiadis et al. (2002) compile the fundamentals of this design approach. This section
gives an individual description of the methods and concludes with the advantages and
limitations of this design method category.
3.3.1 Mixed-Integer Nonlinear Programming (MINLP)

This method is particularly helpful to nd fully or partially optimized solutions for RD
design variables (Malone and Doherty, 2000). The objective function for the RD problem is commonly composed of two basic terms: annual operating cost (e.g. consumption
of raw materials, steam and cooling water) and the annualized investment (i.e. column,
internals, reboiler and condenser). The constraints are formed from the MESH equations on each tray, material balances at the top and bottom of the column, kinetic and
thermodynamic relationships and logical relationships between process variables and
the number of trays.
Assumptions: (i) the vapor and liquid phases are in equilibrium on each tray; (ii) no
reaction takes place in the vapor phase; (iii) the liquid phase is always homogeneous;
(iv ) the enthalpy of liquid streams is negligible; (v ) the heat of evaporation is constant;
(vi) temperature dependence of the reaction rates can be expressed in Arrhenius form;
and (vii) the cost of separating products downstream is given by an analytical function.
The solution to the RD problems results in the optimum number of trays, the optimal
feed tray location, reux ratio, condenser and reboiler duties and liquid hold-ups on each
tray. Since the model contains both continuous (e.g. temperature and composition)
and discrete (i.e. number of trays) design variables, it should be solved by MINLP
optimization technique.
Description: (i) a master sub-problem decomposition approach selects integer variables (i.e. number of trays) in a master program; and (ii) an optimal column design is
61
Chapter 3
obtained with a xed number of trays in the primal problem.
Due to their combinatorial nature, MINLP problems have been lately tackled using
evolutionary algorithms (e.g. genetic algorithm, simulated annealing and evolution
strategies), which require only information regarding the objective function (Costa and
Oliveira, 2001). However, these novel approaches demand huge computational time and
lack robustness when solving highly constrained problems.
Ciric and Gu (1994) present a MINLP-based approach for the design of RD columns
for systems where multiple reactions take place and/or where reactive equilibrium or
thermal neutrality cannot be assured. This method is based on the combination of a
rigorous tray-by-tray model and kinetic-rate-based expressions to give basic constraints
of an optimization model that minimizes the total annual cost. The major variables
are the number of trays in the column, the feed tray location, the temperature and
composition proles within the column, the reux ratio, the internal ows within the
column and the column diameter.
Lately Poth et al. (2001) apply MINLP technique to the synthesis of MTBE and compare the optimal processes for the assumption of chemical equilibrium and kinetically
limitations. Their results show that the total cost of the kinetic column is higher than
that of the equilibrium case. Cardoso et al. (2000) apply a simulated annealing-based
algorithm to solve a non-equilibrium RD column in the synthesis of ethylene glycol.
The results obtained from this approach coincide satisfactorily with those reported by
Ciric and Gu (1994) using commercial optimization packages. Pekkanen (1995) applies
a local optimization algorithm to take advantage of the large number of degrees of
freedom ( 3 + nc ) in the design of RD columns. In that study, the stage hold-ups
and tray eciencies are taken as optimization variables and the local optimum is found
by tracking the directions of the concentration vectors in successive stages. Frey and
Stichlmair (2001) optimize the RD column for the synthesis of methyl acetate in the
kinetically controlled regime with respect to total annualized cost. Multiple feeding
of reactants is allowed in the optimization routine leading to a decrease in the energy
demand. Stichlmair and Frey (2001) design RD columns for the synthesis of MTBE
and methyl acetate. Their starting point is the denition of the process superstructure,
which in turn is derived from thermodynamic fundamentals (e.g. stoichiometric lines
and lines of possible reactive azeotropes). A generalized modular representation for
process synthesis is presented and extended by Papalexandri and Pistikopoulos (1996)
and Ismail et al. (2001), respectively. That approach is supported by fundamental mass
and heat transfer principles, which allow the denition of a multipurpose mass/heat
transfer module. The optimized combination of building blocks denes the process superstructure. The eectiveness of this steady state approach is tested in homogeneous
and two phase-reactions, absorption/extraction and RD processes. Ismail et al. (2001)
include Gibbs free energy-based driving forces as constraints to guarantee the feasibility
62

of the reaction/separation/RD processes. The design of RD units with VLL equilibrium
is studied by Gumus and Ciric (1997). Their optimization problem is formulated as a
MINL-bilevel exercise, whose underlying feature involves the development of a singlelevel optimization problem. The solution of the single-level problem is a lower bound for
the solution of the bilevel problem. The objective function is the Gibbs free energy on
each tray subjected to the mass, energy balances, constitutive expressions and integer
variables dening the number of trays at the minimum total annualized cost.
3.3.2 Orthogonal Collocation on Finite Elements (OCFE)

MINLP has been found to be troublesome for units with large number of trays and
for the simultaneous design of more than one distillation unit due to the considerable
computational time required for attaining the solution (Seferlis and Grievink, 2001).
Furthermore, special attention must be paid in the MINLP model formulation and initial
guesses outside the feasibility domain may lead into convergence diculties. OCFE
techniques is suggested by Seferlis and Grievink (2001) for the design and optimization
of staged RD columns. This approach transforms the discrete number of stages in
the column into a continuous variable and treats the composition and temperature as
functions of the position.
Assumptions: (i) complete mixing of each phase at each stage; (ii) constant liquid
hold-up and no vapor hold-up at each stage; (iii) no liquid entrainment from stage to
stage; (iv ) adiabatic stages; and (v ) thermal equilibrium between the liquid and vapor
streams leaving each stage.
Description: (i) the column is separated into sections; (ii) each column section is divided into smaller sub-domains (i.e. ne elements); (iii) for each ne element a number
of collocation points is specied, where the mass and energy balances are exclusively
satised; (iv ) the collocation points are chosen as the roots of the discrete Hahn family
of orthogonal polynomials; and (v ) Lagrange interpolation polynomials are used within
each nite element to approximate the liquid- and vapor- component ow rates, the
total stream ow rates and the liquid and vapor stream enthalpies.
In the study by Seferlis and Grievink (2001), the oscillatory behavior in the composition space of highly-diluted components is circumvented by transforming the molar
fractions of those components. Additionally, a second alternative is proposed, where
inert components with low concentrations are removed from the energy and mass balances. Both approaches are found to increase the stability and robustness of the solution
at the same level of accuracy. This technique is successfully applied to the design of
a RD column for the synthesis of ethyl acetate (Seferlis and Grievink, 2001), using a
single element with four collocation points in the rectifying section, three elements with
three collocation points per element in the reactive section and two elements with two
63
Chapter 3
collocation points per element in the stripping section. The study concludes that the
production of ethyl acetate is strongly favored by large liquid-stage hold-ups in the reactive and stripping sections because of the relatively low rate of reactions. Furthermore,
a double-column production owsheet is selected, which allows the production of high
purity ethyl acetate.
3.3.3 Mixed Integer Dynamic Optimization (MIDO)

In general a chemical process that shows poor ability of meeting control objectives gives
rise to missing opportunities with respect to economic performance and satisfaction of
product specications. This fact may imply the existence of a trade-o between those
processes designed with respect to control performance considerations and those driven
by economic issues. In order to circumvent this duality, relevant improvements are recently done towards the integration (Russel et al., 2000) and simultaneous optimization
(Georgiadis et al., 2002; Bansal, 2000; Bansal et al., 2000) of the design and control
tasks. The general design and control problem (cf. explanatory note 3.3) involves nding parameters (continuous and/or integers) that dene the RD column and control
tuning parameters at the minimum economic cost and with an acceptable dynamic performance in the presence of disturbances. The continuous parameters include column
diameter, reboiler and condenser areas and tuning parameters (gain K, set-point SP
y - involve the existence or not of catalyst
and reset time c ), while integer variables -
k
on a given stage (frx {0, 1}). According to Georgiadis et al. (2002), the simultaneous approach takes advantage of the interactions between design and control and
provides a fully operational process design that is cheaper and more controllable than
that generated by the sequential approach.
Description: (Bansal et al., 2000; Bansal, 2000; Georgiadis et al., 2002) (i) set the
termination tolerance, ; initialize the iteration counter, k = 1; lower bound, LB = ;
and upper bound, U B = ; (ii) for xed values of the integer variables, y = yk ,
solve the k th primal problem to obtain a solution, J k . Omit the search variables and
constraints that are superuous due to the current choice of integer variables. Set
U B = min(U B, J k ) and store the continuous and integer variables corresponding to
the best primal solution; (iii) solve the primal problem at the solution found in step
(ii) with the full set of search variables and constraints included. Convergence will be
achieved in one iteration. Obtain the Lagrange multipliers needed to construct the k th
master problem; (iv ) solve the k th master problem to obtain a solution k . Update the
lower bound, LB = k ; and (v ) if U B LB , or the master problem is infeasible,
then stop. The optimal solution corresponds to the values stored in step (ii). Otherwise
set k = k + 1, yk+1 equal to the integer solution of the k th master problem and return
to step (ii).
64
Mixed-Integer Dynamic Optimization Problem

Obj.:
s.t.
min J(x(tf ), u(tf ), d(tf ), y(tf ), pp (tf ), pc (tf ), z p , z c , y, tf ),
z p ,z c ,y
0 = f (x(t),
x(t), u(t), d(t), y(t), pp (t), pc (t), z p , z c , y),

0 = g(xd (t), xa (t), u(t), d(t), y(t), pp (t), pc (t), z p , z c , y),
0 h(x(t), u(t), d(t), y(t), pp (t), pc (t), z p , z c , y),
0 i(x(0), u(0), d(0), y(0), pp (0), pc (0), z p , z c , y),
z p Z p , z c Z c , y [0, 1]
p
c
pp (t) P p Rp , pc (t) P c Rp
x
x
x(t) X R , x(0) X R ,
t [t0 , tf ]
t [t0 , tf ]
t [t0 , tf ]
t [t0 , tf ],
(3.27)
t [t0 , tf ]
where the objective function J() represents the overall cost of the RD processes (i.e. capital and
operational costs), the equalities f () and g() comprise the dierential and algebraic expressions,
respectively; h() represents the path constraints that must be satised at all times during the
system operation, i() is the set of initial conditions, x denotes the set of state variables, u is the
set of input variables, y is the set of output variables, d is the set of disturbance variables, pp
and pc are the sets of time-varying variables for the process and control scheme, respectively; z p
and z c are the sets of time-invariant parameters for the process and control scheme, respectively;
and y is the set of integer variables.
Note 3.3. Mixed-integer dynamic optimization problem formulation (Bansal

et al., 2000; Bansal, 2000).
Advantages: (i) optimization-based approaches are applicable for multiple reactions

and where reactive equilibrium or thermal neutrality cannot be assured; and (ii) it
allows the generation of optimum designs with respect to economics and controllability.
Limitations: (i) special care should be paid to the model formulation and the physical
signicance of the solution; scaling of optimization variables might be required to reduce
computational eort; (ii) the optimization problem might be dicult to solve; (iii)
huge computational time can be required for multiple alternative optimizations; (iv )
it is dicult to perform sensitivity analysis of the optimum solution; and (v ) local
limitations can be found when the initial guess lies outside the feasibility region.
3.4
Evolutionary/Heuristic Methods
A third trend of conceptual design approaches may be identied from the evolution of
RD as a derivation from conventional distillation processes. As stated by Subawalla and
Fair (1999), residue curve map techniques and xed-point algorithms have been traditionally very useful tools for preliminary screening and design of RD systems. However,
they cannot be used for detailed design, since they rely on several limiting assumptions
and do not account for the special nature of reactive column internals. Those authors
propose a post -design algorithm to estimate parameters such as column pressure, reactive zone location, catalytic mass, reactant feed location, reux ratio, column diameter,
number of equilibrium stages and packed height. Moreover, the feasibility of up-stream
65
Chapter 3
pre-reactor is taken into account. For equilibrium-limited reactions, the following issues
are addressed (Subawalla and Fair, 1999),
use of a pre-reactor: incorporating a pre-reactor could protably enhance the
conversion of equilibrium-limited reactions, especially when it handles a substantial part of the reaction duty.
Rule of thumb 1: use a pre-reactor when reaction rate at 80% conversion is
more than half the initial rate.
operating pressure: the column pressure ranges from the condenser pressure
(determined by the condenser coolant temperature) to the reboiler pressure (xed
by the heating medium temperature and the column pressure drop). In general
terms and within this range, the operating pressure depends on the reaction temperatures, relative volatility and the eect of pressure on azeotropes. An upper
pressure limit exists beyond which rates decrease because of reaction depletion due
to relative volatility and/or azeotropic eects and due to departure from chemical
equilibrium.
reactive zone location: this parameter represents the best trade-o between
conversion and product purity. For instance, when the limiting reactant is the
most volatile component and the product is the heaviest component, the reactive
zone is located towards the top of the column. For high purity requirements,
stripping or rectifying sections must be incorporated either below or above the
reaction zone.
Rule of thumb 2: a reactive zone section should be located where the concentration of at least one reactant (preferably the limiting reactant) is the maximum.
feed location: a right choice of feed location guarantees high concentrations of
reactants in the reactive zone, as shown in gure 3.6 for the synthesis of methyl
acetate. The following rules of thumb might be suggested:
Rule of thumb 3: if the reactants are the most volatile components, then
the feed should be introduced at the bottom of the reactive zone,
Rule of thumb 4: if the volatilities of the reactants are very dierent, an
additional feed location may be required to ensure stoichiometric reactant quantities
Rule of thumb 5: a pre-reacted product-containing feed enters the column at
some distance from the reactive zone, ensuring thereby that separation between
reactants and products takes place on stages between the feed point and the
reactive zone.
66
MeOH 6 HAC 19
MeOH 13; HAC 13
MeOH 12; HAC 14
MeOH 8; HAC 17
100
Conversion of HAC [%]
90
80
70
60
50
10
15
20
25
30
Reflux ratio [-]
Figure 3.6. Inuence of feed location on reactant conversion. Systems features: synthesis of methyl acetate from methanol and acetic acid.
Legend: the numbers in the data series denote the feed tray location from the top of the column. Conversion reaches a maximum
when the feed ows are located directly below and above the reactive section of the column. (adapted from Bessling et al. (1998)).
reactant ratio: reactants requirements can be calculated from product specications, desired conversion and azeotropic compositions at dierent operating
pressures.
catalyst mass: the minimum catalyst requirements are determined by simulating
a series of isothermal PFRs and ideal separators in series. As depicted in gure
3.7, the feed to the rst PFR contains the necessary quantity of unconverted reactants; the partially reacted euent is fed to an ideal separator that completely
separates reactants from products. This procedure is repeated for dierent temperatures until the desired conversion is reached, where the operating temperature
denes the minimum amount of catalyst. Since there is a relevant inuence of operating temperature on catalyst life, chemical equilibrium and column pressure,
the adopted operation temperature value diers from the highest possible value.
In order to account for the non-isothermal operation and for the imperfect separation on each stage, the mass of catalyst should be 20-30% greater than the
estimated value. Additionally, inadequate catalyst volume reduces residence time
and gives poor conversion.
number of theoretical stages: the number of stages may be estimated to
67
Chapter 3
...
Figure 3.7. Column internals driven design: ideal reactor-separator train for
the estimation of minimum catalyst requirements. The partially
reacted euent is fed to an ideal separator that completely separates reactants from products.
100
Conversion of A [%]
90
80
70
60
50
40
0
10
Reflux ratio
Figure 3.8. Relation between conversion and reux ratio for a RD column.
Systems features: chemical reaction A + B C + D; components volatilities: A/B/C/D = 4/1/8/2. (adapted from Bessling
et al. (1997)).
a certain extend by short-cut methods (e.g. Fenske Underwood), especially for

the nonreactive zones (stripping and rectifying). Note that there is an optimum
number of reactive stages, which allows one to obtain the highest reaction extent
for given vapor ow and column height (Solokhin and Blagov, 1996).
reactive zone height and column diameter: the diameter of the column is
determined by the ooding velocity, vapor-liquid trac, pressure-drop and packing catalyst density; the reactive zone height, on the other hand, depends on the
catalyst mass, packing catalyst density and column diameter.
reux ratio: in RD columns, the rate and ratio of reux play an important
role in reaction and separation phenomena. For equilibrium-controlled systems,
a high reux rate increases the separation of products from reactants, thereby
increasing the reaction driving force. As noted in gure 3.8, excessive reux leads
68
Vcat =
mcat
pack
; hcat =
vvapor , vliquid
RR min
x feed , x B, x D
Vcat
4
2
Dcol
; N cat =
D col
hcat
Vcat , hcat, Ncat
HETPcat
Simulation
Y
x= x desired
mcat mcat . f (f<1)
RR 1.1 . RR
Y
x changes
N
vvapour , v liquid,Dcol
RR
RR 0.9 . RR
N
x decreases
0.9 . RR
Figure 3.9. Procedure to estimate reactive zone height, reux ratio and column diameter for a catalytic distillation column (adapted from
Subawalla and Fair (1999)).
to operating problems and insucient reaction hold-up, leading to incomplete

conversion.
Rule of thumb 6: use a factor of 1.2-1.4 the minimum ratio.
Short-cut methods derived from conventional distillation have been found to be
inappropriate for the estimation of reux ratio for reactive case and new approaches have been adopted. Methods derived from the boundary-value approach
(cf. 3.2.4) allow the determination of the minimum and real reux ratio, which
are key factors in the determination of vapor and liquid velocities. The reux
ratio can be calculated using an iterative procedure, described in Subawalla and
Fair (1999) and summarized in gure 3.9.
reboil ratio: the feasibility of RD processes has been closely related to the
minimum reboil ratio (RSmin ), at which the composition of the rectifying xed
point equals the composition of the stripping xed point (Okasinski and Doherty,
1998). The reboil ratio normally ranges from 3 to 8 and reaches a minimum when
plotted versus the total number of stages.
actual trays/packed height: For the stripping and rectifying sections experimental tray eciency or packing HETP should be used, whereas for the reactive
section mass-transfer coecients should be calculated from experimental data at
the operating conditions of the system. In the event that thermo-physical data,
69
Chapter 3
transport phenomena data and hydraulic models are available, non-equilibrium
design approaches are found to provide a better agreement between experimentation and simulation.
The proposed approach may be extended by incorporating hydraulic models which account for the column capacity in terms of pressure drop and for mass-transfer eciency
estimations in terms of liquid hold-up, mass-transfer coecients and interfacial area.
Advantages: this method estimates easily several equipment and/or operational variables.
Limitations: this method is basically a post-design algorithm and as such it requires
an already dened process structure.
3.5
Concluding Remarks
Both the graphical and optimization-based design methods have already proven their
potential in RD process design but have also shown some signicant limitations. Graphical methods are fairly exible in generating alternative designs at various design parameters, allowing the designer to dene realistic bounds on them. Furthermore, the
graphical nature of the methods claries the understanding of those fundamental issues
in RD, which might be disguised by other approaches (e.g. reactive azeotropes). On
the other hand, graphical methods are strongly limited by their graphical nature (i.e.
(nc nrx ) 3). Optimization-based methods overcome the last limitation and have
successfully solved design problems, where multicomponent mixtures, multiple chemical
reactions and multiple units are involved. However, additional caution is required in the
problem formulation and inappropriate initial guesses of the optimization variables may
lead into convergence diculties. The evolutionary method proposed by Subawalla and
Fair (1999) covers in more detail some physical aspects of the design of RD columns.
However, this approach requires a pre-dened structure and accordingly the optimum
design is not guaranteed.
Based on their assumptions, limitations and outputs a ngerprint chart was produced for
the methods described in this chapter. From this qualitative plot (table 3.2) the following conclusions can be drawn: (i) most of the graphical methods provided fundamental
insights on the feasibility of the RD process and dene the initial process structure; (ii)
economic considerations are taken into account exclusively by optimization-based methods; (iii) the availability of the process has not been included as design output by any
method, since this temporal feature is normally dealt at the next level of engineering
design; (iv ) the heuristic-based methodology is based on an already designed process
and is therefore a post-design tool; (v ) most of the design methods have been developed
for steady state operation; (vi) the initial eort required for optimization-based meth70

ods is far more signicant than that for graphical approaches, but the output generated
by the programming-based design is more comprehensive.
Due to the size and conceptual/computational complexity of the RD design problem, a
single comprehensive design strategy might not be available for years to come. AlmeidaRivera and Grievink (2001); Almeida-Rivera et al. (2004a), however, propose an integrated design approach -termed multiechelon design approach. This strategy allows the
designer to combine in a systematic way the capabilities and complementary strengths
of the available graphical and optimization-based methods. As discussed in detail in
the course of the coming chapters, this approach is supported by a decomposition in a
hierarchy of imbedded design spaces of increasing renement. Moreover, it constitutes
the framework of a highly aggregated design methodology, which focuses on the RD
design problem from a life-span perspective.

remark can be made. An overview and analysis of representative design methodologies are provided in this chapter. Based on
identied opportunities, an integrated design approach is briey
mentioned, which oers the benet of combining the capabilities
and complementary strengths of available methodologies. The
domain knowledge in RD design is addressed. Therefore, the material presented in this chapter contributes to the answering of
Question 4.
71
Chapter 3
Table 3.2. Qualitative ngerprint of the design methods used in reactive
distillation.
G3 :
Nomenclature.
Attainable region G4 :
G1 :
Statics analysis; G2 :
Fixed-points; G5 :
RCM;
Reactive cascade;
G6 : Thermodynamic-based; G7 : Conventional graphic techniques; G8 :

Phenomena-based; G9 : Dierence points; M1 : MINLP; M2 : OCFE;
M3 : MIDO; H1 : Heuristics.
Key.
: applicable; : nonapplica-
ble/nonmentioned; : original assumption relaxed by later contributions
G1 G2 G3 G4 G5 G6 G7 G8 G9 M1 M2 M3 H1
Input
phase equilibrium model
chemical reaction equilibrium
chemical reaction kinetics
feed compositions
compositions of product streams
reboil/reux ratios
initial spatial superstructure
type packing/hydrodynamic features
temporal disturbance scenario
Knowledge required
Model assumptions
phase equilibrium
negligible heat eects
equilibrium controlled regime
kinetically controlled regime
steady state operation
multicomponent systems ( 3)
single reaction
constant molar overow
constant liquid holdup
isomolar stoichiometry
negligible axial dispersion
constant volatilities
innite vapor/liquid ows
lm theory validity
constant mass transfer coecients
Rules
connectivity, units selection
operating conditions
Output
feasibility of the RD process
denition process superstructure
72

Continue in next page

Table 3.2. Qualitative ngerprint of the design methods used in reactive distillation (continuation). Nomenclature. G1 : Statics analysis; G2 :
RCM; G3 : Attainable region G4 : Fixed-points; G5 : Reactive cascade;
G6 : Thermodynamic-based; G7 : Conventional graphic techniques; G8 :
Phenomena-based; G9 : Dierence points; M1 : MINLP; M2 : OCFE;
M3 : MIDO; H1 : Heuristics.
Key.
: applicable; : nonapplica-
ble/nonmentioned; : original assumption relaxed by later contributions
operating conditions (reux, reboil)

number of (non-) reactive stages
location feed stages
location (non-) reactive azeotropes
determination MSS
Life-span perspective
economic considerations
exergy eciency considerations
controllability considerations
Nature of output
structure
performance (economics)
physical behavior (operability)
physical behavior (availability)
Eort
model complexity
implementation time
tools complexity
G1

G2

G3

G4

G5

G6

G7

G8

G9

M1

M2

M3

H1

Coming from previous page
73
A problem clearly stated is a problem half solved.

Dorothea Brande, Becoming A Writer (1934)
A New Approach in the Conceptual Design

of Reactive Distillation Processes
The potential benets of applying RD processes are taxed by signicant complexities

in process development and design. Several topics regarding these two activities are
addressed in this chapter. The interactions between process development and design
activities are considered rst. The RD design problem is then formulated in the
wider context of process development and engineering. The identication of requested
output information from current methodologies results in the introduction of a set
of performance criteria SHEET (safety, health, environment, economics, technology).
The focus is on the design problem from a life-span perspective. A multiechelon
approach is subsequently introduced, in which a hierarchy of embedded design spaces
of increasing renement is used and the strengths of current design methods are
conjugated.
Parts
of this chapter have appeared in Almeida-Rivera et al. (2004a) and Almeida-Rivera and
Grievink (2001)
75
Chapter 4
4.1
Introduction
When chemical reactions and physical separations have some overlapping operating
conditions the combining of these tasks into a single process unit can oer signicant
benets. These benets may involve: avoidance of reaction equilibrium restrictions,
higher conversion, selectivity and yield, removal of side reactions and recycling streams,
circumvention of nonreactive azeotropes and a reduction in number of units (i.e. investment cost) and energy demands (i.e. heat integration) (Schembecker and Tlatlik,
2003; Subawalla and Fair, 1999; Stichlmair and Frey, 1999; Jimenez et al., 2001; Malone
and Doherty, 2000). However, there is a price to pay in terms of increased complexity
in design and operations. The nonlinear coupling of reactions, transport phenomena
and phase equilibria can give rise to highly system-dependent features, possibly leading
to the presence of reactive azeotropes and/or the occurrence of steady-state multiplicities. The number of design decision variables for such an integrated unit will be high.
Industrial relevance requires that design issues are not separated from the context of
process development and plant operations. Hence, the design problem is addressed in
this chapter from a process life-span perspective.
The starting point of this approach is the qualitative analysis of current design methodologies in RD presented in chapter 3. Table 3.2 clearly shows the complementary aspects of the graphical, programming-based and heuristic-based methods. The graphical
methods are capable of generating feasible process structure options under somewhat
idealized chemical and physical conditions. Optimization-based methods, once given
a superstructure, can generate more detailed and complete designs from an engineering perspective and can even be used to strive to account for controllability aspects.
Heuristic-based methodologies provide only rough estimates of the design variables and
require a pre-dened process structure.
The complexity and full range of design variables in process integrated RD units are, as
yet, not considered in the scope of the current RD design methods. Process integrated
designs should take into account the economic and exergy-related interactions of the
RD units with the process surroundings and comply with the desired dynamics and
control of the overall process supply chain.
In view of the size and of the conceptual and computational complexity of a RD design
problem, a single comprehensive design technique is unlikely to become available for
some years to come. An integrated design strategy (Almeida-Rivera and Grievink,
2001; Almeida-Rivera et al., 2004a) is introduced in the forthcoming sections. This
design strategy allows the designer to combine, in a systematic way, the capabilities and
complementary strengths of the available graphical and optimization-based methods.
This integrated approach is embedded within a highly-aggregated design methodology,
which is characterized by a set of performance criteria. In clear agreement with a life76
A New Approach in the Conceptual Design of RD Processes

span perspective (cf. explanatory caveat in section 1.4) these criteria account for the
economic, exergy-related and temporal performance of a design focussed on prevention
of loss of resources.
It is worth mentioning here that the present contribution focuses on the denition of
the RD process structure, paying less explicit attention to a detailed analysis of the
actual dynamics of the unit, which might be described using non-equilibrium models.
An extended review on this last topic can be found in the publications of Taylor and
Krishna (2000); Krishnamurthy and Taylor (1985); Taylor and Krishna (1993, 2003).
Furthermore, the temporal features of the process unit(s) (e.g. operability and availability) are not fully considered in the proposed methodology, as availability is foreseen
to be taken into account at the next level of engineering detail.
What are the (performance) criteria that need to be considered from a life-span
perspective when specifying a reactive distillation design problem?
4.2
Interactions between Process Development and Process

Design
The aim of chemical process design has, conventionally, been considered as the nding
of equipment sizes, congurations and operating conditions that will allow for the economical, safe and environmental responsible conversion of specic feed stream(s) into
specic product(s). This task is in principle carried out only by specifying the state
of the feeds and the targets on the output streams of a system (Doherty and Buzad,
1992; Buzad and Doherty, 1995). Frequently, however, the design activity is found to be
severely complicated by the diculty of identifying feasible equipment congurations
and ranges of suitable operating conditions. This situation can be explained by the fact
that the design of chemical process units is only a constitutive part of a larger scale and,
therefore, more complex problem: the joint consideration of process unit development
and design. At this higher level of aggregation, the objective of the design problem
includes sizing the unit and establishing the requirements of the design, screening the
process feasibility and, at the top, identifying the process opportunities (gure 4.1).
From this point of view, two parallel and inter-related activities can be identied: the
design program and the development program. For a given level of design detail all the
design tasks should be considered from top (i.e. higher aggregation/process level) to
bottom (i.e. low aggregation/unit level). The information leaving the design program
is compared with the desired outputs. If an acceptable agreement exists the level of
detail is increased and the same strategy is adopted. For non-acceptable outputs the
77
Chapter 4
desired output
DESIGN PROGRAM
input data
assumptions
models
strategies
sin
gd
eta
il le
vel
Identify opportunities
Screen feasibility
Design requirements
decreasing aggregation level
rea
i nc
agreement ?
refined
design level
PROCESS DESIGN
Design of units
redefinition
improvement
reformulation
design/perform experiment
method development
Improved knowledge
DEVELOPMENT PROGRAM
Figure 4.1. The design problem regarded as the combination of a design program and a development program.
loop is performed at the same level of detail after redening/reformulating appropriate

assumptions/models/generalities in the design program. This is attained by designing
and performing an appropriate development program, which allows the designer to
extend/improve/correct the existing knowledge and methodology (e.g. removing the
uncertainties introduced by estimation of parameters).
Within the design program, the following design tasks are dened.
Identication of opportunities. This task permits one to select qualitatively the set
of competing process technologies/equipment congurations, which might be used
to convert the given feeds to the desired products. Schoenmakers and Bessling
(2003b), for instance, use the physical/chemical characteristics of the system (e.g.
reaction velocity and relative volatility) to choose the equipment conguration.
Proven technologies together with novel congurations might arise at this stage
(e.g. dividing wall columns, pump-around reactors, side-reactors, etc.) (Baur and
Krishna, 2003; Krishna, 2003; Mutalib and Smith, 1998a,b).
Screening of feasibility. The feasibility of the selected opportunities is then screened
with respect to the attainable products of a given set of feed streams. In this stage
the concepts of attainable region (Feinberg, 1999; Nisoli et al., 1997) and /
approach (Westerberg and Wahnschat, 1996; Fien and Liu, 1994; G
uttinger and
Morari, 1997; G
uttinger, 1998; G
uttinger and Morari, 1999a,b) might be used.
Design requirements. The set of functional requirements for the design can be deemed
the basis of the design. This information involves the boundary conditions, the
process structure, its physical behavior (i.e. operability and availability) and
minimum performance (e.g. safety and economics).
78

Design of units. The nal stage in the overall design problem involves dening the
spatial and temporal variables of the unit(s), together with its (their) operation
window(s) and mode(s) of operation.
4.3
Structure of the Design Process
Following the denitions given above it is noticeable that the rst three tasks in the
design program are related to the scope of design, whereas the last task involves essentially the design of the units. If the general design paradigm (Siirola, 1996a,b; Biegler
et al., 1997) is used as the framework for the design task, the overall design problem can
be represented as a multistage goal-directed process, with dynamic ow of information
between the design tasks, method development and knowledge development (gure 4.2).
In the case of RD, the design issues are far more complex than those involved in conventional distillation and may dier signicantly from case to case. For instance, the design
variables for conventional distillation include the number of trays, feed tray location and
reux ratio for a constant molar overow system with single feed tray. In contrast, in
a RD column the liquid hold-up is the major design parameter that determines the
extent of reaction, the feed might be distributed within the unit and constant molar
overow can not be assumed unless the reaction is thermally neutral and stoichiometrically balanced (Ciric and Gu, 1994). All these facts have led to a smaller number of RD
applications in the chemical industry (Gorissen, 2003), than might be expected given
the economic benets that this combined process oers (Doherty and Buzad, 1992;
Taylor and Krishna, 2000; Georgiadis et al., 2002). The endeavors of the RD scientic community have resulted in the development of design algorithms mostly based on
equilibrium models applied to systems with simplied thermodynamics and reactions
kinetics (Malone and Doherty, 2000; Okasinski and Doherty, 1998, 1997; Ung and Doherty, 1995a,c; Smejkal and Soos, 2002; Venkataraman et al., 1990). Lately, however,
non-equilibrium (NEQ) models have been developed to use for the accurate modeling
of RD units, accounting for mass transfer limitations and the interaction between mass
transfer and chemical reaction (Baur et al., 2000, 2001b,a,c; Baur and Krishna, 2002,
2003; Higler et al., 1998, 1999a,b; Higler, 1999; Higler et al., 2000; Taylor and Krishna,
1993, 2000; Krishna, 2003; Krishnamurthy and Taylor, 1985; Taylor and Krishna, 2003;
Wesselingh, 1997). Although NEQ-based models are highly recommended for a more
realistic description of the process, especially in advanced stages of the design cycle,
their implementation still relies on the designers expertise. Furthermore, inappropriate parameters estimators may introduce an increased number of uncertainties in the
design problem.
Three core questions arise when reviewing the currently available design methods for
RD processes,
79
Chapter 4
Abstract description
STAGE n-1
DEVELOPMENT OF METHOD
DESIGN
DEVELOPMENT OF KNOWLEDGE
INPUTS
definition/
specification
needs
objectives
definition/
specification
models
method
development
STEPS
FORMULATION
DEVELOPMENT EFFORT
experimental
theoretical
BUILDING
BLOCKS
tactics
error correction
strategies
reformulation
representation
design methods
SYNTHESIS
structure
parameters
STAGE n
method
development
analysis
methods
evaluation
methods
PHYSICAL
BEHAVIOUR
EVALUATION
satisfaction
abstraction
learning
DESIGN
DESCRIPTION
ANALYSIS
actual
behaviour
method
development
OUTPUT
not good
enough
PERFORMANCE
SELECTION
good
enough
PREFERED
ALTERNATIVE
STAGE n+1
time span
Refined description
Figure 4.2. Overall design problem. Some ideas have been borrowed from
Siirola (1996b), Biegler et al. (1997) and Bermingham (2003).
What is the output information requested from the design program?

To what extent are the current methods able to provide this output?
What resources (i.e. time wise, tools, costs) are required for the design
program?
In order to answer the rst question, an extended list of output design variables (gure
2.7) is dened. The output variables together with the design problem statement in
RD are summarized in table 4.1. It can be noticed that the desired outputs from the
design program might be grouped into two main categories: those related to the process
structure and performance (i.e. minimum economic performance) and those related to
temporal features of the unit(s) (i.e. operability and availability). For the sake of
clarity, the denitions adopted for each category are included in table 4.2.
80
Table 4.1. Design problem statement in reactive distillation

Designing a reactive distillation processes can be seen as the creative task of structuring
and sizing the RD process providing a set of constraints, windows of feed specications
and ranges of product targets. The design problem might be formulated as follows,
Given,
a set of components and chemical reactions
the characteristics of the feed streams (P , T , x)
a set of dierent modes of operation for the process
a set of product specications and safety and operational constraints for each
mode of operation
data regarding the process economic targets based on prices of the products and
the reactants as well as correlations for the investment and utility costs
ranges of possible values for the major model parameters and ranges of variation
for key process variables
disturbances scenarios
targets of availability of the unit(s)
Determine,
the number/type of units and their tasks (i.e. reaction, distillation, reactivedistillation)
the connectivity of the units (i.e. owsheet)
the type of internal contacting structure (i.e. continuous or structured packing,
discontinuous or stages and dividing wall) and hydrodynamic features (i.e. P ,
RTD, contacting area and volumetric hold-up)
the total number of stages in the column (i.e. reactive/nonreactive)
the location of the feed, withdraw and turn-around stages
the hold-up of the liquid phase for homogeneous reactions or the catalyst load
for heterogeneous reactions
ow regimes, column diameter
control structure, controllers type and tuning parameters, placing of actuators
and sensors
operating conditions (e.g. operating pressure)
reboil and reux ratios
reactant ratio
economic, safety and environmental performance
Regarding the second question, the output space of the design methodologies under
consideration is mapped in table 3.2. It should be noticed that most of the methodologies focus on the estimation of spatial variables, paying less explicit attention to the
temporal response of the design. Moreover, no method up to now has addressed exergy
considerations for the case of RD.
81
Chapter 4
Table 4.2. Categories of information resulting from the design process in reactive distillation.
Structure
includes the complete spatial structure of the process, together

with the control loops and operating conditions (i.e. ranges of
temperature, pressure and reboil/reux ratios).
Performance involves the estimation of criteria related to: safety, health, environment and economics. Technological aspects are addressed
in the next category. This group of criteria are referred to as
SHEET.
Operability involves the ability to operate the unit(s) at preferred conditions
in spite of disturbances and changes in operational policies.
Availability is rarely considered and is a measure of the degree to which the
unit(s) is in an operable state at any time.
The amount of resources required for each methodology is directly dependent on their
complexity (cf. table 3.2). Thus, as optimization-based methods are more sophisticated
than graphical methods, they require far more signicant eort, especially in modeling.
Based on the above a set of performance criteria is dened in the next section. These
criteria allow one to focus on the design problem from a life-span perspective, embracing
all the output categories of interest. An extended analysis of sustainability issues in
conceptual design is given in Korevaar (2004).
4.4
Life-Span Performance Criteria
Aiming at a design methodology that addresses the problem from a life-span perspective, Grievink (2005) suggests a set of performance criteria. The motivation behind
this perspective relies on the relevance of bearing in mind economic performance and
potential losses over the process life span. Specically, exergy-eciency and responsiveness aspects in the operational phase need to be considered. As stated in section 1.4
we only consider sustainability aspects that are exclusively under the control of the
process design and aim at the eective utilization of resources in the operational phase.
The SHEET set of criteria embraces issues related to safety, health, environment,
economics and Technology. Although safety and health aspects are of utmost relevance for the chemical industry, we consider them to be implicitly outside the scope
of our proposed methodology. We acknowledge that safety issues are key drivers for
the innovation and technological breakthroughs that characterize intensied processes.
Aiming at cheaper, cleaner, safer and simpler manufacturing of products, process intensication faces various challenges, such as the minimization of chemicals inventory and
82

the simplication of equipment complexity. Disasters such as, for example, the forty
metric tons of methyl isocyanate released in Bhopal, 1984 should never have happened
if a conscious safety-oriented process design had been carried out.
The starting point of our approach is the assumption that we (society) have learnt
from past mistakes. Therefore, provided that safety and health issues are inherent in
any responsible design approach, the remaining issues dene a novel performance space.
Thus, in the course of this dissertation we pay explicit attention to the EET performance
space. Environmental considerations are directly linked to the minimization of mass,
energy and exergy losses. Economic performance is translated quantitatively to the capital and operational costs of the design. The total annualized cost (TAC) is a convenient
index for this. Technical considerations are grouped into two sub-categories: availability and operability. Availability embraces both reliability and maintainability, whereas
operability includes process design attributes such as exibility (steady-state), stability
(multiplicity), controllability (dynamic) and switchability (start-up/shut-down). Controllability and switchability are lumped conveniently into a responsiveness criteria,
which accounts for the dynamic response of the design in the presence of disturbances.
A schematic representation of the performance criteria set is given in gure 4.3.
It is worth mentioning the link between our environment-related index and commonly
used approaches such as life cycle assessment. According to our scope of research (see
explanatory caveat in section 1.4) we limit our contribution to the design and operational space in the process life span. Thus, our focus is exclusively on the sustainability
aspects that are under the control of the design and operational phases. In this regard,
we aim at an operational eciency, which means basically an eective utilization of
resources in the process operation. Similar to life cycle assessment, we quantify the
OPERABILITY
flexibility
stability
controllability
switchability
EXERGY EFFICIENCY
TAC
AVAILABILITY
SHEET
reliability
maintainability
Safety
Technology
Health
Environment
Economics
RESPONSIVENESS
Figure 4.3. SHEET approach for the denition of life-span performance criteria.
83
Chapter 4
materials, energy and exergy used and assess the environmental impact of those ows,
via loss minimization. However, we explicitly do not address any sustainability issue
related to feedstock selection, (re-)use of catalyst, (re-)use of solvents, among others.
The thrust in our rather liberal approach is that these steps (i.e. design and operation) in the overall process life span (i.e. upper aggregation level from cradle-to-grave)
contribute largely to the overall life-span goals.
As these EET proposed performance criteria are of dierent natures, their formulation into an objective function involves the assignment of appropriate weighing factors.
Assigning these values, however, is not a trivial task as it is an ultimate decision of
the designer and made according to the requirements of the design problem (Clark and
Westerberg, 1983). Pareto curve mapping (cf. explanatory note 8.1) is a convenient way
to depict the eects of the weighing factors in a multi-objective optimization problem.
The design approach is introduced in the next section and represents the framework of
the integrated design methodology.
4.5
Multiechelon Approach: The Framework of the Integrated Design Methodology
Due to the size and conceptual/computational complexity of the RD design problem,

a single comprehensive design strategy may not be available for years to come. As a
rst step towards such an integrated design approach, Almeida-Rivera and Grievink
(2001); Almeida-Rivera et al. (2004a) propose a multiechelon design approach, which
allows the designer to combine, in a systematic way, the capabilities and complementary
strengths of the available graphical and optimization-based methods. In general terms,
the methodology is supported by a decomposition in a hierarchy of imbedded design
spaces of increasing renement. As a design progresses the level of design resolution can
be increased, while constraints on the physical feasibility of structures and operating
conditions derived from rst principles analysis can be propagated to limit the searches
in the expanded design space.
The general design paradigm of Siirola (Siirola, 1996a,b) is taken as the framework
for the approach. As depicted in gures 4.4 and 4.5 dierent decisions are made at
the dierent levels, supported by specic design tools. Furthermore, all the levels are
inter-related by the ow of information coming from the lower levels. As one steps
up the design pyramid, the scope of design decreases whereas the information ow is
incremented.
The following design spaces are derived for this approach.
Space 1. Formulation: sustainability issues are considered, the basis of design (e.g.
product and feedstock purities ranges, production modes, operating window
84
Opt
constraints
rigorous dynamic modelling

non-equilibrium models
optimum ?
SHE
behaviour
analysis
controllability
economic assessment
internal spatial
structure
mathematical programming
economic assessment
tools
hydrodynamics
graphical methods
component balances
generation of
separation sequences
operational skill,
rules of thumb
continuous/discontinuous
packing
batch/continuous
temporal structure
decisions
optimisation tools
operational skills
graphical methods
overall balances
feasibility analysis
thermodynamics
kinetics
feasibility,
economic assessment,
SHE
definition of the design problem
Figure 4.4. Multiechelon design approach in the conceptual design of RD

processes: tools and decisions
and economics) are dened and the building model blocks are developed (e.g.
thermodynamic, kinetic, hydrodynamic).
this corresponds to levels Formulation and Building Blocks in gure
4.2.
Space 2. Feasibility analysis: using graphical methods the feasible product compositions are determined for a given set of input variables (e.g. feed composition
and Damk
ohler number), incorporating all occurring phenomena (e.g. distillation boundaries and (non)-reactive azeotropy). If the process occurs to
be feasible, economically attractive and satises SHE issues, proceed to level
3. Otherwise, stop the design and explore other processes alternatives.
this corresponds to level Synthesis in gure 4.2.
Space 3. Temporal structure: the mode of operation (e.g. batch, continuous or
semibatch) is selected based on operational skills, production requirements
and supply chain dynamics.
Space 4. Generation of separation sequences: rigorous graphical methods allow
the identication and organization of feasible column sequences. Their type
of task is estimated (e.g. fully reactive, hybrid and nonreactive columns),
together with the number of units and their connectivity. In this regard,
each column might be considered as a compartment performing a given task.
A decision has to be made related to the type of internal (i.e. continuous or
85
Chapter 4
Opt
constraints
multiple state states

heat duty, contacting area, r,
s, diameter, heat exchange
area, flow
controllability indices,
operating conditions
behaviour
analysis
packing size, staging, RTD

catalyst concentration,
hold-up, pressure drop
internal spatial
structure
types of task (separation/reaction),

number of columns,
connectivity
hydrodynamics
generation of
mode of operation
temporal structure
distillation boundaries
azeotropy
operational skills
basis of design
(feed state, product targets
and operational boundaries)
feasibility analysis
thermodynamics
kinetics
sustainability issues
definition of the design problem
Figure 4.5. Multiechelon design approach in the conceptual design of RD

processes: interstage ow of information.
structured packing and discontinuous or stages).

Space 5. Internal spatial structure: the process structure obtained from the previous level is used as the starting point for this stage. Using optimization-based
methods continuous variables (e.g. column diameter, heat exchange areas,
contacting area, catalyst concentration, reboil and reux ratios, RTD, pressure drop, hold-up, ooding and wetting conditions) and discrete variables
(e.g. feed staging, reactive and nonreactive stages, withdrawal staging) are
obtained. It is relevant to mention that at more detailed design level, rigorous simulation methods are foreseen as key tools (e.g. FEM and CFD). If
the process structure options are economically attractive, proceed to the next
level. Otherwise, relax, if applicable, some of the constraining assumptions
made for level 1 and resume level 2.
Space 6. Behavior analysis: steady state simulations are carried out to explore the
kinetics, gas/liquid mass transfer, hydrodynamics and multiplicities inside
the units. Dynamic simulations are performed to check the robustness of the
design, in terms of the ability to maintain product purities and conversion in
a desired range when disturbances occur. If the process is controllable and
economically attractive at the estimated operating conditions, go to the next
86

level. Otherwise, relax, if applicable, some of the constraining assumptions
made for level 1 and resume level 2.
this corresponds to level Analysis in gure 4.2.
Space 7. Evaluation: the design options are checked with respect to their performance. Within the wide range of SHEET criteria specic choices are made
for economic performance, sustainability and technology. For economics, the
capital and operational costs of the process during the time horizon (i.e.
TAC) are calculated. Exergy eciency is used as a sustainability metric,
while a systems response time is used to dene the responsiveness of the
processes. Thus, sustainability aspects like proper choice of feedstocks and
products supply chain are not considered. From a scientic perspective, we
choose exergy eciency and responsiveness to exploit their underlying link:
the rate of entropy production. If the designed process satises all three constraining criteria the design task is nished. Otherwise, a trade-o is sought.
this corresponds to level Evaluation in gure 4.2.
The successful application of the integrated design approach relies on considering the
output of the graphical methods as initial structure for the optimization-driven design
approaches. In this way, the optimized design is likely to be embedded within the
feasible space and the computational eort becomes less demanding due to the reduced
number of design variables. This approach is intended to reduce the complexity of the
design problem and it should not be considered as a mechanistic recipe.
4.6
Concluding Remarks
A multiechelon approach was introduced aimed at an integrated and generalized design methodology for RD. An attempt is made with this design strategy to conjugate
the strengths of graphical and optimization-based methods. Economic performance,
sustainability considerations and process responsiveness are included as evaluation criteria under the umbrella of the SHEET set. A decomposition of the design problem is
proposed in this approach to give a hierarchy of design spaces of increasing renement.
The challenges in the design of RD are numerous and exciting. The drive for process
intensication and more sustainable plant operations can lead to more intricate geometric structures (e.g. packings, dividing walls, distributed heat exchange) and to more
rened models, which couple mass and energy transfer in a more realistic way. Finally,
incorporating supply chain motivated temporal features within the design task (e.g.
operation and switching modes) may increase tremendously the degrees of freedom and
the problem complexity.
87
Chapter 4
If we take a mind journey back to the engineering/scientic design
question presented at the beginning of this chapter, the following
remark can be made. The analysis of strengths and weaknesses of
current design methodologies is addressed in this chapter. Practical constraints regarding their development and application are
mentioned. The multiechelon approach is introduced and its
building blocks are described. With respect to Question 5, a
set of performance criteria are suggested for a life-span inspired
design approach.
88
What saves a man is to take a step; then another step.

Antoine De Saint-Exupery, writer and pilot (1900-1944)
Feasibility Analysis and Sequencing:

A Residue Curve Mapping Approach
The second (feasibility analysis) and fourth (column sequencing) spaces of the multiechelon approach (cf. chapter 4) in RD design are studied. Applications of residue
curve mapping to RD processes have recently been reported, but a generalized and
systematic approach is still missing for the case of reactive feeds outside the conventional composition ranges (i.e. stoichiometric ratio of reactants for the case of
MTBE synthesis). This chapter is addressing these design aspects. The reaction invariant space, dened in terms of transformed composition variables, is divided into
sub-regions characterized by separation boundaries. A feasibility analysis of the RD
process is performed based on the location of the reacting mixture and initial separation sequences are generated according to the feed transformed-composition. The
practical conclusion of this chapter is that by using RCM technique one can virtually
generate RD trains over the entire possible feed composition space. The synthesis of
MTBE is used as case study.
Parts
of this chapter have appeared in Almeida-Rivera and Grievink (2002) and Almeida-Rivera
and Grievink (2004)
89
Chapter 5
5.1
Introduction
The art of process design involves nding process congurations, operating conditions
and size of equipment that will allow an economical, safe and environmental responsible
operation, only by specifying the state of the feeds and the targets on the output
streams of a system (Almeida-Rivera et al., 2004a; Doherty and Buzad, 1992; Buzad
and Doherty, 1995). This task is not trivial and requires the development of specialized
design tools (e.g. computational methods, algorithms and procedures). At early stages
of the design activity (e.g. conceptual phase) the chemical process designer faces a
largely inuencing question: is the processing route feasible? The impact of preliminary
design and feasibility analysis on the fate of the nal plant is enormous. Surprisingly, it
is a common practice to allocate a relative small fraction (< 2-3%) of total budget for
the conceptual phase. Due to the coarse level of detail, the design tools in the conceptual
phase should be powerful enough to screen feasible design alternatives within a huge
design space.
What (new) methods and tools are needed for reactive distillation process synthesis,
analysis and evaluation?
5.2
Input-Output Information Flow
Each level of the proposed multiechelon approach (cf. 4.5) is characterized by ows of
information needed as input and obtained as result from that design stage. Moreover,
set of tools are required to assist the design task and design decisions are to be made.
In the case of the feasibility analysis stage (Space 2 in section 4.5) the required input
information involves the process basis of design. In other words, feedstock and product
purities and operational boundaries should be dened. Additionally, sustainability and
Table 5.1. Input-output information for the feasibility analysis phase. The
multiechelon approach is explained in section 4.5.
Design specications
Design/operational variables Domain knowledge
Process:
- set of components
- set of chemical reactions
- feed composition
- operational pressure
Product:
- composition target(s)
- SHE constraints
Discrete:
- distillation boundaries
- (non-) reactive azeotropes
Continuous:
- product feasible compositions
90
- thermodynamics
- kinetics
- overall balances
Feasibility Analysis and Sequencing: A Residue Curve Mapping

Approach
Table 5.2. Input-output information for the column sequencing phase. The
Design specications
Process:
- mode of operation
Discrete:
- type of task
(separation and/or reactive)
- number of columns
- columns connectivity
Continuous:
- feed ratio of reactant streams
- operational skill
- component balances
safety constraints should be included as inputs. The output information comprises

the determination of distillation boundaries, (non-) reactive azeotropic mixtures and
feasible product compositions. The domain knowledge at this design space belongs to
the eld of thermodynamics, kinetics and overall mass balances. An extended overview
of input and output information at this design level is given in table 5.1.
The generation of separation sequences (Space 4 in section 4.5) is composed of process
requirements related to the mode of operation, which is in turn determined using operational skills. The temporal mode of operation mode is chosen based on operational
expertise in RD processing. For the sake of analysis simplication and without restricting the validity of the approach to other temporal modes, a continuous operation is
chosen. The output information at this design space includes the type of task required
for the design specications (i.e. separation and/or reaction), the type of column internal (i.e. tray or packing), the number of units and their connectivity. The relevant
domain knowledge at this design space corresponds to operational thumb-ruled skills
and basic component balances. An extended overview of input and output information
at this design level is given in table 5.2.
Making each stage of the design approach operational requires the use of (novel) design
tools. The Residue Curve Mapping technique (RCM) is found to be of particular
eectiveness in the case of RD feasibility analysis and generation of column sequencing.
The next sections apply this concept in the corresponding design spaces 2 and 4.
5.3
Residue Curve Mapping Technique
The RCM technique has been considered a powerful tool for the owsheet development
and preliminary design of conventional multicomponent, nonideal separation processes
(Fien and Liu, 1994). It represents a good approximation to actual equilibrium behavior
and allows the designer to perform feasibility analysis of separation processes where
nonideal and azeotropic mixtures are involved. Traditionally, nonreactive residue curves
have been used to predict the liquid-composition trajectories in continuous distillation
91
Chapter 5
units in the / case (i.e. innite number of stages and innite reux). Although
for nite columns -either batch or continuous- those proles dier slightly compared to
the residue curves under same isobaric conditions (Fien and Liu, 1994), the dierence
might be considered negligible at the rst stages of design. It is relevant to mention
that the applicability of RCM technique is fundamentally valid for both continuous and
batch separation units.
Analytically, RCMs are constructed based on physical properties of the system (i.e. VL
equilibrium, LL equilibrium and solubility data), wherein the composition of a nonreacting liquid remaining in the system (gure 5.1) may be determined by performing
an overall and component material balances in the system (Westerberg et al., 2000),
dM L
= V,
dt
with M L (0) = M0L ,
d(xi M L )
dxi
dM L
= ML
+ xi
= yi V,
dt
dt
dt
(5.1)
i Znc ,
(5.2)
where M L is the mass of liquid in the system, V is the vapor ow, xi is the molar composition of component i in the liquid phase, yi is the molar composition of component
i in the vapor phase and nc is the number of components present in the system. Apart
from liquid hold-up no equipment related data are required.
Replacing expression 5.1 in 5.2 and introducing a warped time (dw = dt V /M L )
vapor:
V(t)
y(t)
liquid:
ML(t)
x(t)
heat
Figure 5.1. Schematic representation of a simple batch still for the experimental determination of (non-) reactive residue curves
92

Approach
allow one to rewrite the RCM expression in vectorial notation,
dx
= x y,
dw
with x(0) = x0 ,
(5.3)
T
where x is the composition vector in the liquid phase [x1 x2 x3 xnc ] and y is the
T
composition vector in the vapor phase [y1 y2 y3 ync ] .
Combining the previous [n-1 ] independent expressions 5.3 with and appropriate vaporliquid equilibrium model enables to dene the set of simultaneous equations that describe the residue curves,
dxi
dw
yi
xi yi ,
f (T, P, xi ),
i=n
c
1 =
xi =
i=1
i=n
c
i Znc 1 ,
i Znc ,
(5.4)
(5.5)
yi .
(5.6)
i=1
When a mixture of nc -components undergoes nrx simultaneous equilibrium chemical

reactions, the RCM expression may be described in terms of transformed molar compositions and a reaction-warped time (Ung and Doherty, 1995a),
dXi
= Xi Yi ,
dw
with Xi (0) = Xi,0 ,
i Znc nrx 1 ,
(5.7)
with,
Xi
xi iT (Vref )1 xref
,
T
1 total
(Vref )1 xref
i Znc nrx ,
(5.8)
Yi
yi iT (Vref )1 yref
,
T
1 total
(Vref )1 yref
i Znc nrx ,
(5.9)
where
Vref
(nc nrx +1),1
..
.
nc ,1
dw
dt
i,r
T
1total
(Vref )1 yref
T
1total
(Vref )1 xref
(nc nrx +1),nrx
..
,
.
nc ,nrx

(5.10)
V /L,
where Xi and Yi are the transformed composition variables for the component i
Znc nrx in the liquid and vapor phases respectively, iT is the row vector of stoichiometric coecients for component i Znc nrx in all the nrx reactions [i,1 i,2 i,R
],
i=nc
T
total is the row vector of the total mole number change in each reaction
i=1 i,k ,
k
k Znrx , ref denotes the reference components for the nrx reactions, numbered from
nc nrx + 1 to nc and Vref is the square matrix of stoichiometric coecients for the nrx
reference components in the nrx reactions.
From the above-mentioned expressions it can be inferred that,
93
Chapter 5
the new set of variables denes a sub-region of lower dimension in the composition
space, reaction invariant space, composed of the feasible product compositions for
equilibrium controlled reactions,
the transformed compositions keep the same value for a liquid mixture before and
after reaction equilibrium is attained (i.e. Xi0 = Xie ) and are normalized (i.e.
nc nrx
nc nrx
Xi = i=1
Yi = 1),
i=1
in the new composition space the reactive curves may be described by a simplied
expression, which has exactly the same functional form as the distillation equation
for non-reacting mixtures in terms of mole fractions,
ref can be a product or a reactant, depending on the value of total < 0, and
the reaction warped time increases monotonically with time
w |t0 = 0; w |tL=0 = .
(5.11)
The reactive problem is completed by calculating the chemical equilibrium constants

for the multicomponent vapor-liquid mixture,
Keq =
i=n
c
(i xi )i = f (xi , i (xi , T )),
(5.12)
i=1
where Keq is the chemical equilibrium constant for reaction r Znrx .

The equations involved in the reactive problem include [nc nrx 1] RCM expressions,
[2nc 2nrx] transformed composition denitions, [nc ] phase equilibrium conditions, [nrx ]
chemical equilibrium constant and [2] normalization expressions, resulting in [4nc
2nrx + 1] equations. On the other hand, the involved variables are xiZnc , yiZnc ,
XiZnc nrx , YiZnc nrx , P and T , resulting in [4nc 2nrx + 2] variables. Since the
problem has only one degree of freedom, all the unknown variables may be found by
specifying a single variable (e.g. P ) and solving simultaneously the following set of
dierential and algebraic expressions,
dXi
= Xi Yi ,
i Znc nrx 1 ,
dw
Xi (0) = Xi,0 ,
i Znc nrx 1 ,
Xi
Yi
yi
Keq
xi iT (Vref )1 xref
,
T
1 total
(Vref )1 xref
yi iT (Vref )1 yref
,
T
1 total
(Vref )1 yref
f (T, P, xi ),
i Znc ,
i=n
c
i=1
94
(i xi )i ,
(5.13)
(5.14)
i Znc nrx ,
(5.15)
i Znc nrx ,
(5.16)
r Znrx .
(5.17)
(5.18)

Approach
The integration of dierential expression 5.13 stops when either pure components or
reactive azeotropes are met. These singular points are given by the following conditions:
Pure component:
Reactive azeotrope:
Xi = Yi = 1, {Xj , Yj } = 0,
0 < Xi = Yi < 1,
i = j, i, j Znc
(5.19)
i Znc .
(5.20)
Note, however, that the reference composition vectors xref and yref are back calculated
after each time integration of X using the corresponding denition expression. Assigning
the operation pressure is a task that should be carefully analyzed since bifurcation
pressures have been reported (Fien and Liu, 1994; Okasinski and Doherty, 1997). At
those system-dependent P values the thermodynamic and topological behavior of the
reacting mixture changes considerably.
5.4
Feasibility Analysis: An RCM-Based Approach
Residue curve maps provide a powerful tool to represent relevant properties of the
system, particularly those aiming to predict feasible design sequences. In addition,
analytical material balances may be represented in a RCM, resulting in constraints
to feasible product compositions and convenient operating strategies (e.g. direct or
indirect distillation).
The presence of (non-)reactive singular points (i.e. pure components and (non-)reactive
azeotropes) in RCMs allows the division of the (transformed) composition diagram into
separate (reactive) distillation regions by introducing (reactive) distillation separatrices,
which connect two singular points in the composition space. The explanatory note 5.1
gives a brief overview of the concepts of nodes and saddle points, as features of singular
points.
If a separatrice ends at a saddle point, it is commonly referred to as boundary curve.
Crossing distillation boundaries has been extensively reported for industrial cases and
especially for those with curved distillation boundaries. For those systems, the boundaries can be crossed by simple recycle methods provided an appropriate entrainer. However, conventional design strategies require that the feed and products remain in the
same distillation region (Fien and Liu, 1994).
Since sharp separations are not possible in the industrial practice, short-cut methods
have been developed to assist in the design of distillation sequences. A transparent
example of those procedures is the so-called bow-tie region method (gure 5.2), which
intends to nd for a specic feed a set of feasible top and product compositions for
columns at nite reux or when distillation-boundaries are crossed. The following
procedure for the construction of bow-tie regions (cf. gure 5.2) is suggested by Fien
and Liu (1994): (i) the distillation boundaries are identied and drawn in a composition
95
Chapter 5
Saddle point is dened in the stability theory of ODE as the point of a function or surface, which is
a stationary point but not an extremum. Let
f = f (x1 , x2 , x3 , ...xn )
(5.21)
(xs1 , xs2 , xs3 ....xsn)
where the
be a scalar function of n variables. Then a saddle point is any point
gradient of f evaluated at xs is zero, but which is not a local maximum or minimum because
the Hessian matrix is indenite, namely,
f (xs )
a, b x, a = b
(5.22)
T
a H(x )a < 0 b H(x )b > 0
(5.23)
In RCM technique saddle points are considered to be those intermediate-boiling vertices (pure
components or (reactive) azeotropes), at which no residue curve has neither its initial nor nal
point (Fien and Liu, 1994).
Stable node is a xed point for which the gradient of f evaluated at xn is zero and the Hessian
matrix at xn is negative denite (Fien and Liu, 1994), namely,
f (xn )
xT H(xn )x
<
0,
(5.24)
for all nonzero x Rdim(H)
(5.25)
Unstable node is a xed point for which the gradient of f evaluated at xn is zero and the Hessian
matrix at xn is positive denite (Leon, 1998), namely,
f (xn )
T
x H(x )x
=
>
0
0,
(5.26)
for all nonzero x R
dim(H)
(5.27)
In RCM technique stable nodes are considered to be those high-boiling vertices (pure components
or (reactive) azeotropes), to which a residue curve is converging. Alternatively, unstable nodes
are those low-boiling vertices, from which a residue curve is diverging.
Note 5.1. Denition of stable nodes, unstable nodes and saddles points
space diagram; (ii) mass-balance lines are drawn for both direct and indirect splits
considering the constraints imposed by the distillation boundaries; (iii) a mass-balance
line is drawn for an infeasible intermediate split; (iv ) the regions that are not crossed
by this split are shaded; and (v ) the nal bow-tie area is dened by considering that
each feasible residue curve must go through both shaded areas. Further information on
the construction of bow-tie areas is given by Fien and Liu (1994).
For separations that involve entrainers, the selection of them is without question the
most determining task. For reactive distillation, the entrainment agent should react
selectively with one of the azeotropic components allowing to cross distillation boundaries.
96

Approach
low-boiling
direct split
boundary
indirect split
feed
residue curve
infeasible split
high-boiling
Figure 5.2. Construction of bow-tie regions in RCM. Remark: feasible

residue curves must go through both shaded areas, so that top
and bottom compositions line on a material balance line passing
through the feed.
5.5
Case Study: Synthesis of MTBE
The esterication of iC4 with methanol in the presence of inert nC4 and an acidic ionexchange resin is used as case study. A detailed description of the systems features,
together with relevant physical-chemical data are available in appendix B. For the
feasibility analysis the following parameters and design specications are of interest,
number of components, nc = 4,
number of independent reactions, nrx = 1,
desired product and purity: MTBE (>97%),
feed composition: outside conventional MeOH/iC4 stoichiometric ratios,
continuous mode of operation.
The novelty of the approach relies on the fact that it addresses feed mixtures outside
conventional composition ranges. With this, an extended picture of the feasibility analysis and column sequencing in RD processing is attained. To the best of our knowledge,
such exceptional feeds have not been reported yet in the open knowledge domain.
5.5.1 Simulation Problem

The simulation problem encompasses the solution of the RCM expressions for two cases:
nonreactive and reactive mixture of iC4 -MeOH-MTBE-nC4 system. According to the
97
Chapter 5
degree of freedom analysis of the previous sections, only the total pressure is specied
(P = 11105 Pa). The criterion for choosing this value is determined by the solubility
of iC4 in the liquid phase and enables to carry out the reaction to a desirable extent
(Zhang and Datta, 1995).
c
c
/gOPT
programming software (Process Systems Enterprise
In this study gPROMS
Ltd.) is used to solve the set of simultaneous dierential and nonlinear algebraic expressions. The solving algorithm is composed of an DAE integrator based on variable
step-size/variable order backward dierentiation formulae. Initial molar compositions
in the nonreactive case and transformed variables in the reactive case are specied to
generate the residue lines, aiming -by visual inspection- to completely cover the composition space. In this contribution, the initial points are uniformly distributed in the
corresponding composition space. It is possible, however, to eciently localize them
by considering that ner grids (i.e. more initial points) are required around singular
points.
The simulations are performed until singular solutions are found, which correspond to
pure component or (non-) reactive azeotropes (i.e. x=y and X=Y).
5.5.2 Nonreactive System: Feasibility Results

Three minimum boiling point azeotropes are found for the nonreactive system at the
operating pressure of 11105 Pa,
binary azeotrope iC4 -MeOH occurring at xiC4 =0.917; [label (a) in gure 5.3]
binary azeotrope MeOH-nC4 ocurring at xnC4 =0.8794; [label (b) in gure 5.3]
binary azeotrope MeOH-MTBE occurring at xMTBE =0.577; [label (c) in gure
5.3].
These three azeotropic mixtures represent the corners of a nonreactive distillation
boundary plane and divides the composition simplex into a MTBE-rich and a MeOHrich regions (G
uttinger, 1998). As depicted in gure 5.3, all the residue curves originate
at the iC4 -MeOH azeotrope and end either at the MeOH or MTBE pure component.
5.5.3 Reactive System: Feasibility Analysis and Generation of Reactive

Distillation Sequences
When chemical reaction takes place all the reactive residue curves start at either pure
iC4 corner or at the MeOH-nC4 azeotrope (MNAz) and end predominantly at the pure
MeOH corner (gure 5.4). The remaining low-boiling azeotropes did not survive the
98

Approach
MTBE 1
0.8
0.6
x
(c)
0.4
0.2
MeOH
0.8
0.6
0.4
0.2
(b)
nC4
0.2
0.4
0.6
0.8
(a)
iC
boundary plane
Figure 5.3. Residue curve map for the nonreactive system iC4 -MeOH-MTBEnC4 at an operating pressure of 11105 Pa.
reaction, whereas an additional saddle point (quaternary azeotrope QAz) has appeared
close to the pure nC4 corner. For highly inert-containing mixtures, the residue curves
are found to end at the nC4 pure corner.
A clearer representation of the reactive system is possible when the conventional composition axis are substituted by the Dohertys composition transformations (gure 5.5).
This composition space allows to easily visualize the saddle behavior of the quaternary azeotrope (gure 5.6a) and, furthermore, to identify a pseudo -reactive azeotrope,
PRAz, (gure 5.6b) located in the hypotenuse of the transformed composition simplex
and corresponding to an iC4 -MeOH-MTBE mixture at high conversion.
The PRAz is characterized by the fact that the residue lines curve toward it but never
converge to it (Ung and Doherty, 1995c) and for practical purposes it imposes the limit
beyond which distillation can not proceed (Ung and Doherty, 1995c). This reactive
mixture is located halfway up the hypothenuse and behaves as a severe pinch point.
As depicted in gure 5.7, the singular points of the reacting system dene ve reactive distillation boundaries: (QAz:MNAz), (QAz:iC4 ), (QAz:MeOH), (QAz:nC4 ) and
(QAz:PRAz).
The composition space is divided into the following ve sub-regions,
The pseudo terminology has been introduced by Ung and Doherty (1995c,a) due to the dual
topological behavior of this point, as it acts as a saddle on one side and as a stable node on the other
side of the map.
99
Chapter 5
MTBE
0.8
0.6
x
0.4
0.2
MeOH
0.8
0.6
0.4
0.2
MnAz
0.2
nC
QAz
0.6
0.4
0.8
iC
Figure 5.4. Residue curve map for the synthesis of MTBE at an operating
pressure of 11105 Pa. Remark: reference component is MTBE
region I:(QAz:MNAz:nC4 ),
region II: (QAz:iC4 :nC4 ),
region III: (QAz:MeOH:MNAz),
region IV: (QAz:PRAz:iC4 ),
region V: (QAz:PRAz:MeOH).
Due to their relative small size, the regions I ( 0.1%), II ( 0.25%) and III ( 0.8%)
might be neglected at the rst stage of design without incurring in measurable errors.
This assumption simplies considerably the feasibility of the process, since all but one
of the boundaries have been removed. The composition simplex is then divided into
two well dened zones,
zone a: where residue curves start at MNAz azeotrope and end at pure MeOH
corner with a saddle point at the PRAz (i.e. region IV),
zone b: where residue curves start at pure iC4 and MNAz and end at the PRAz
(i.e. region V).
Thus, the separation sequences may dier depending on the location of the feed in the
composition simplex. Zone b embraces mainly feed streams at reactants ratios close
100

Approach
MeOH1
0.8
0.6
PRAz
X
0.4
0.2
MNAz
QAz
0.2
nC4
0.4
0.6
0.8
iC
Figure 5.5. Residue curve for the MTBE synthesis at an operating pressure
of 11105 Pa in terms of the transformed compositions as dened
in equation 5.8.
MeOH
MeOH
0.75
0.0125
0.7
0.01
0.65
0.6
0.0075
PRAz
0.55
0.005
0.5
QAz
0.0025
0.45
0.0025
0.005
0.0075
0.01
0.0125
XiC
0.25
0.3
0.35
0.4
0.45
0.5
0.55
(a) Quaternary azeotrope
XiC
(b) Pseudo azeotrope
Figure 5.6. Quaternary and pseudo-azeotropes in synthesis of MTBE at an

operating pressure of 11105 Pa. Remark: transformed compositions (equation 5.8) are used.
101
Chapter 5
MeOH
PRAz
region III
X
MNAz
region IV
QAz
region V
region I
region II
Figure 5.7. Schematic representation of distillation boundaries and zones for

MTBE synthesis. Remarks: transformed compositions are used;
operating pressure is set at 11105 Pa.
to unity and, therefore, are conventionally treated as normal feed compositions (i.e.
xiC4 /xMeOH 1.0). Zone a, on the other hand, is composed of those feed compositions
outside conventional reactant ratios (i.e. 0 xiC4 /xMeOH 0.5).
According to the sequencing analysis zone b sequence is composed of a single column
as depicted in gure 5.8. In this case the fresh feed stream (iC4 , MeOH and nC4 ) is
mixed with a recycle stream (enriched in iC4 after purging most of the existing inert
nC4 ) before entering the column, while the product stream (i.e. PRAz) is continuously
removed at the bottom in an indirect separation strategy.
When the feed to the column is embedded within zone a the sequence synthesis is not
straightforward and requires some additional insight originated from the residue curve
map (gure 5.9a). Thus, a distillation sequence is generated for the MTBE synthesis
using MeOH as an entrainer, according to the approach presented by Fien and Liu
(1994) for nonreactive mixtures. The sequence involves the following operations (gure
5.9b),
feed F (MeOH:iC4 :nC4 ) may be mixed with stream B2 to obtain feed mixture
M1,
M1 is fed to column C1 and separated into distillate D1 (reactive mixture close
to separation boundary) and bottoms B1 (pure MeOH),
In
this case, the term sequence might be loosing its meaning, as only one compact device is
considered.
102

Approach
MeOH
D
inerts purge
PRAz
zone a
QAz
n
ou
da
ry
F
D
zone b
(a) Residue curve map
high purity MTBE

(pseudo azeotrope)
(b) Separation train.
Figure 5.8. Residue curve map and separation sequence for zone b in the synthesis of MTBE by reactive distillation. Remark: high purity
MTBE (pseudo azeotrope) is recovered at the bottom of the column in an indirect separation; operating pressure is set at 11105
Pa.
D1 enters column C2 and is divided into stream D2 (high purity nC4 containing
stream) and stream B2 (high purity MTBE containing stream). It is of utmost
importance acknowledging that the distillation balance line is able to cross the
boundary because of the concavity of the distillation boundary,
stream D2 may be then eventually separated following the sequence designed for
zone b depending on its composition. Moreover, stream B1 may be recycled back
to the rst column,
stream B2 may be recycled back to the rst column, allowing to shift the feed
mixture towards the distillation boundary. Alternatively, stream B2 may be withdrawn from the system when its composition is similar to that of the PRAz.
It can be noticed that the PRAz is an intermediate boiling reacting mixture and behaves
like a tangent pinch that prohibits designing the separation sequence towards pure
MTBE.
5.6
Concluding Remarks
Residue curve maps have shown to provide valuable insights and design assistance for
nonideal systems, particularly for reactive distillation. Transforming the composition
variables according to Dohertys approach allows to dene a reaction invariant space of
103
Chapter 5
lower dimension, formed by attainable product compositions and where the conventional
concepts for residue curves can be applied.
For the MTBE synthesis two main subregions were dened in the modied composition
simplex, each of both characterized by singular points (nodes and saddles) and delimited by separation boundaries. Two dierent separation sequences were generated,
covering all the possible reacting mixture compositions. In the generated sequences,
high-purity MTBE was obtained as a product, due to the appearance of a pseudo reactive azeotrope, which imposes limitation to the separation task. As all geometrical
methods, the RCM technique is limited by its intrinsic graphical nature and requires
accurate thermodynamic data to correctly describe reactive distillation processes. Furthermore, RCM technique is suitable only for equilibrium-based processes, which are
relatively common in industrial practice (Fien and Liu, 1994).
The combinatorial number of possible separation sequences for multicomponent mixtures is considerably large and calls for the denition of heuristics and thermodynamic
guidelines based on RCMs to nd convenient initial designs. Furthermore, operational
issues should be addressed in terms of, for instance, predicting the controllability of the
sequence and dening feasible operational parameters (e.g. reux ratios).
MeOH
B1
to sequence zone b
D2
PRAz
X
F
B2
M1
D1
QAz
C2
C1
zone a
n
ou
da
D1
M1
ry
zone b
D2
X
(a) Residue curve map
B1
MeOH
B2
(b) Separation train.
Figure 5.9. Residue curve map and separation sequence for zone a in the
synthesis of MTBE by reactive distillation. Remark: operating
pressure is set at 11105 Pa; stream D1 may contain all species
and be close to the distillation boundary; stream D2 may be an
inert nC4 - stream; stream B2 may have a composition similar to
the PRAz.
104

Approach
remark can be made. Two design aspects of the multiechelon design approach are considered in this chapter: feasibility analysis
and column sequencing. The RCM technique is identied as an
eective easy-to-use tool to assist the designer in those tasks, particularly for the case of reactive feeds outside conventional composition ranges. Regarding Question 6, a new (at least more
rened) tool is developed to assist the design of RD processes.
105
If people never did silly things, nothing intelligent would ever

get done.
Ludwig Wittgenstein, philosopher (1889-1951)
Spatial and Control Structure Design in

Reactive Distillation
In this chapter the fth space (internal spatial structure) of the multiechelon approach
(cf. chapter 4) in RD design is studied. This design space is addressed from two
perspectives. In the rst section, a multilevel modeling approach is proposed as a
tool for the decision-making task in the design of spatial structures of RD processes.
The proposed strategy involves formulating the simulation problem at dierent levels
of increasing degree of complexity, in which mass, energy and momentum balances are
conveniently applied to represent the behavior of the process. In the second section
of this chapter, the interactions between spatial and control design are discussed.
Both spatial-related and control-related design variables are optimized with respect
to economic and dynamic performance in the presence of time-varying disturbances.
The synthesis of MTBE is used as tutorial example in the course of this chapter.
Parts
of this chapter have appeared in Almeida-Rivera et al. (2003)
107
Chapter 6
6.1
Multilevel Modeling
6.1.1 Introduction
As already discussed in previous chapters, model development and design methodologies
of RD processes are far more complex than those involved in conventional reactors and
separation technologies. Therefore, novel and systematic approaches are required to
support the decision-making tasks for planning, design and operation of this integrated
process. In this section an approach is suggested as a decision-making tool in the
conceptual design of RD processes. This approach allows the designer to obtain rst
estimates of the spatial design variables at an external level or envelop and subsequently
determine their real values at an internal level using more sophisticated and detailed
models.
6.1.1.1
Input-Output Information Flow
This space of the multiechelon approach (cf. section 4.5) involves the ow of information listed in table 6.1. This information is related basically to the unit and internals
sizing. Moreover, operational parameters such as reux ratio and heat duty are considered. It is worth mentioning that although a staged-column conguration is considered
in this subsection, the approach is not restricted to stages geometries. Figure 6.1 depicts a generic design structure for a RD column conguration. It is composed of
several building blocks: condenser, reux drum, reboiler and (non-) reactive compartments. The compartments allow mass and heat ows to enter and leave the unit at
any location. A detailed description of the compartments characteristics is covered
in chapter 8. All building blocks are connected according to generic rules detailed in
chapter 8.
Table 6.1. Input-output information for the internal spatial structure space.
The multiechelon approach is explained in section 4.5.
Design specications
Process:
- type of task
- number of columns
- feed ratio of reactants
Continuous:
- column diameter
- heat exchange/contacting areas
- catalyst concentration
- reboil and reux ratios
- controllers parameters
Discrete:
- feed staging
- reactive and nonreactive stages
- withdrawal staging
- control conguration
108
- overall balances
- dynamic behavior
Spatial and Control Structure Design in Reactive Distillation

For this spatial conguration the multilevel approach involves the derivation of two
levels of dierent complexity and degree of aggregation. These levels are characterized
by inherent assumptions, particular constraints and modeling objectives. All these
features are derived from mass, energy and momentum balances and thermodynamic
constitutive expressions.
The outer input-output level (steady-state) and inner task level (dynamic) are conveniently linked by the extent of reaction (), which allows one to establish whether or not
the business-related decision variables cope with the constraints imposed by the problem
formulation. Moreover, from this approach it can be elucidated whether or not the corresponding models are well specied and mutually consistent. A successful realization
at the outer level (column-level) would suggest relaxing constraining assumptions and
stepping up to the more rigorous modeling framework (task-level). Furthermore, this
approach allows the designer to determine whether or not the business-related decision
variables cope with the operational constraints imposed by the problem formulation.
6.1.2 Level 1: Input-Output Level

The envelop dened at this level of aggregation is delimited by the overall superstructure
(see gure 6.1). The streams of mass and heat are given by: (i) the feed streams entering
the structure; (ii) the product and side streams leaving the structure; and (iii) all heat
transferred to the structure.
At this input-output level the following assumptions are considered: (i) steady-state
modeling; (ii) multiple feeds (nfeed Z+ ), side draw-o streams (no Z+ ), heat
streams (nq Z+ ), reactions (nrx Z+ ) and components (nc Z+ ); and (iii) single
bottom and distillate streams. Component and energy balances are performed over the
structure. The component molar balances (MC) include the distillate, bottom, feed
and draw-o streams and the production (or consumption) of species due to chemical
reaction,
MC i
j=n
feed
Fjfeed zfeed
+ ViT +
j
(6.1)
j=1
B
D xD
i B xi
k=n
off
Fkside pk = 0,
i Znc ,
k=1
109
Chapter 6
input-output level
Qcond
1
Ffeed2
Ffeed3
Ffeedn-2
Q2
Q3
n-3
Qn-3
Fside2
task level
n-2
Ffeedn-1
Fside3
Fsiden-3
Qn-2
Fsiden-2
n-1
Qreb
n
Figure 6.1. Schematic representation of the overall design structure for a RD

structure. Nomenclature: F feed : feed owrate; B : bottom
owrate; D: distillate owrate; F side : side stream; Q: heat duty.
Remarks: mass and heat ows are allowed to enter and/or leave
the unit at any point; compartments are numbered from top to
bottom; dashed lines dene the envelopes for input-output and
task levels.
where F feed , B, D, F side are the molar rates of feed, bottom, distillate and side drawo streams, respectively; zfeed , xD , xB , p are the molar fractions vector for the feed,
distillate, bottom and side draw-o streams, respectively; Vi is the vector (nrx 1) of
stoichiometric factors of component i in the nrx reactions and is the vector (nrx 1)
of the overall extents of reaction.
The overall energy balance (E) is given by the heat requirements of the reboiler and condenser, heat of reaction, draw-o heat duties and the enthalpy of all involved streams.
Thus,
E
i=n
feed
Fifeed (hfeed
) + (r H)T B hB +
i
i=1
D h
D
j=n
off
j=1
110
i=nq
Fjside
hP
j
k=1
Qk Qcond + Qreb = 0,
(6.2)

where h is the molar enthalpy without heat of formation, Qcond is the condenser duty,
Qreb is the reboiler duty, Qk is the draw-o heat rate at stage k and r H is the row
vector of heats of chemical reaction.
Normalization expressions for molar fractions are additionally included in the problem
formulation. The set of balance expressions 6.1 and 6.2 turns to be well-posed by specifying appropriate decision variables according to a degree of freedom analysis. Thus,
for a given set [nc , nfeed, nq , no , Hrx , V] and an extent of reaction , the modeling
problem is consolidated by removing the degrees of freedom, grouped in vector A,
(nc 1)(nfeed)zfeed
A = (no )T P
Qcond
(nc 1)(no )p
TD
(nfeed )Ffeed
(nc 1)xD
(nq )Q
(no )Fside
(nfeed )T F
.
Qreb
B
(6.3)
Special attention should be paid in the denition of A. The specication of its variables
should take into account the physical feasibility of the problem solution. Thus, vector
A includes variables such as compositions of all incoming streams, compositions of all
outgoing streams (side, distillate and bottom streams), temperatures of all incoming
and outgoing streams, heat duties of reboiler and condenser, heat ows transferred to
each compartment and the molar ows of all feed streams, side streams and bottom
stream.
The column-level simulation problem is then stated as follows,
Given
Solve for
s.t.
:
:
:
and A,
Xvariables ,
|Ytotal
| = 0,
X
feed
F , D, B, Fside R+ ,
Qreb, Qcond , Q R,
zfeed, xD , xB , p, [0, 1],
(6.4)
where Xvariables is the set of all involved variables (2(nfeed + no ) + nc (nfeed + no + 2) +

6 + nq ), Ytotal is the set of algebraic equations obtained from mass (expression 6.1) and
energy balances (expression 6.2) and normalization expressions (nc + nfeed + no + 3).
Relevant information is extracted from this level with respect to process ow/block
diagram possibilities and estimation of boundary conditions. The practical gain of
this approach is the assessment of the eect of the extent of reaction vector on the
product stream(s). More specically, at a given set of input design variables, ow
and composition might be computed having the extent of reaction vector as an input
parameter. Moreover, the output of this aggregation level denes good design targets
for the task-oriented level.
111
Chapter 6
L
F n-1
F nV
F nfeed,L
Q
F nside,V
Rx
F nside,L
F nfeed,V
F nL
V
F n+1
Figure 6.2. Schematic representation of the generic lumped reactive distilfeed,()
lation volume element (GLRDVE). Nomenclature: Fi

:
side,()
: owrate of side
feed owrate of phase () at position i; Fi
stream corresponding to phase () at position i; m : interfacial
mass transfer ux; e : interfacial energy transfer ux; Q: heat
duty.
6.1.3 Level 2: Task-Level

At this modeling level the internal working of the column is taken into account (i.e. reaction rates, mass and heat transfer). A generic lumped RD volume element (GLRDVE)
is conceived to represent any element within the RD column along with its occurring
tasks and around the steady-state dened in level 1. As depicted in gure 6.2 the
GLRDVE allows one to account for the presence of incoming and outgoing streams
(feeds/draw-os), the change of chemical identity and the removal/addition of heat.
The novelty of this approach stems from the derivation of four independent sub-modules
within the GLRDVE. Those sub-modules are characterized by the dynamic behavior
of unique tasks: reaction/separation, separation without reaction, mixing and heat
exchange module. A detailed derivation of the rst principles occurring within the
GLRDVE is addressed in section 8.2.
Along the same line of reasoning Papalexandri and Pistikopoulos (1996) present a generalized modular approach for process synthesis, which includes among other reactiveseparation building blocks. Relevant is the denition of thermodynamic-based connectivity rules among the dierent blocks and the optimization-driven approach used. Recent contributions by Proios and Pistikopoulos (2005) extend the work of Papalexandri
with a particular focus on separation trains. All dynamic sub-models imply characteristic assumptions and the presence of species mass and energy balances in the volume
element and the inclusion of hydraulic, constitutive and normalization expressions. In
112

general terms, the key sets of equations include: (i) conservation equation of species
mass, energy and electrical charge; (ii) phase and reaction equilibrium equations; (iii)
thermodynamic constitutive equation per phase, involving density, specic heat and
fugacity coecients; (iv ) approximations to the momentum balance involving pressure
drop, phase ratio and phase contact area; and (v ) kinetic equation for chemical reactions, mass and heat transfer between lumped phases.
Thus, for any GLRDVE the set of involved expressions is given with respect to mathematical structure by,
Given
Solve for
: f (x(t),
x(t), u(t), y(t), z) = 0,

g(x(t), u(t), y(t), z) = 0,
h(x(t), u(t), y(t), z) 0,
i(x(0), u(0), y(0), z) = 0,
: x(t) X Rx , u(t) U Ru , y(t) Y Ry ,
z Z Rz ,
t [t0 , tf ],
t [t0 , tf ],
t [t0 , tf ],
(6.5)
t [t0 , tf ],
where f (), g() and h() are sets of dierential, algebraic equality and inequality equations, respectively, i() is the set of initial conditions, x is the set of state variables, u is
the set of input variables, y is the set of output variables and z denotes time-invariant
parameters.
During the model development for the GLRDVE the following assumptions are made:
(i) the liquid and vapor streams leaving a compartment are in thermodynamic equilibrium; (ii) the liquid and vapor in each compartment are perfectly-well mixed (i.e. no
radial gradient in temperature and composition); (iii) the reactions are taking place in
the liquid phase and enhanced by a solid catalyst; (iv ) the reactions occur only in the
compartments containing catalyst; and (v ) vapor and liquid entrainment is neglected.
The assumed extents of reaction at level 1 (input-output) dene the targeted conversion
for the various species. At the inner level (task level) the cumulative eect of all local
conversions should match the targets at the outer level, whenever physically feasible.
This constraining information might be expressed as follows (Melles et al., 2000),
s.t.
: |input-output level task level | 0.
(6.6)
Figure 6.3 visualizes the link between both levels and the streams shared by them. Note
that the actual input variables to both levels are given by formulation 6.4.
6.1.4 Case Study: Synthesis of MTBE

The multilevel modeling approach is applied to the synthesis of MTBE at 11105 Pa in a
staged RD column loaded with solid catalyst. A detailed description of the system can
113
Chapter 6
feed streams: flows, compositions, temperatures
heat duty
outgoing streams: flows,
compositions, temperatures
I/O level
task level
modules
connectivity
Figure 6.3. Schematic representation of the link between the input-output

level and the task level. Remark: the actual input variables to
the levels are given by the formulation 6.4.
be found in appendix B. The reaction kinetics is estimated according to the pseudohomogenous expression reported by Mohl et al. (1999); Sundmacher and Homann
(1999) and Schrans et al. (1996),

gj (x, T ) =
aiC4
aMTBE
aMeOH
Keq a2MeOH

,
(6.7)
where g() is the chemical reaction driving force based on activities and Keq is the
temperature-dependent chemical equilibrium constant.
The spatial structure of the reactive distillation column is similar to that reported by
Hauan et al. (1995a) and Jacobs and Krishna (1993) and detailed in the upper section
of table 6.2.
At the input-output level an initial decision is made according to formulation 6.4. Bottom compositions are specied, from which the extent of reaction is computed using
input-output model. A desired value of 150 mols1 is imposed to the overall extent
of reaction and the set Ytotal
is successfully solved for a wide range of A values. As
X
feasible solutions are shown to exist, the decision of solving the simulation problem
according to the task model formulation 6.5 is made. The task-level model is developed
c
c
/gOPT
(PSEnterprise, Ltd.) equation-oriented environment. The set
in gPROMS
of expressions f (), g() and h() are simultaneously solved provided an appropriate set
of initial conditions i(). Moreover, the sizing design variables (column diameter and
reboiler and condenser areas, listed at the lower section of table 6.2) are conveniently
chosen by the algorithm to avoid any violation of the extent of reaction constraint 6.6
and hydraulic considerations. As depicted in gure 6.4, the components compositions
show an expected behavior at a simulation time that guarantees steady-state condition.
A highly pure MTBE stream is drawn at the bottom of the unit; whereas the top stream
114
Table 6.2. Nominal values in the MTBE synthesis. Remark: operational

conditions are those reported in Hauan et al. (1995a); Jacobs and
Krishna (1993); Seader and Henley (1998).
Value
Number of trays
Feed 1: location
Feed 1: owrate
Feed 1: composition
Feed 1: temperature
Feed 2: location
Feed 2: owrate
Feed 2: composition
Feed 2: temperature
Operating pressure
Reux ratio
Mass of catalyst/tray
Extent of reaction
Column diameter
Condenser area
Reboiler area
15
iC4 + nC4 : 10th tray (g)
455 mols1
iC4 : 0.37; nC4 : 0.63
350K
MeOH: 9th tray (l)
168 mols1
MeOH: 1.00
320K
11105 Pa (top)
7
204.1 kg=1000 eq[H + ]
150 mols1
3.573 m
4.460101 m2
2.456102 m2
[2:rectifying;8:reactive; 5:stripping]
is predominantly composed of inert nC4 . Furthermore, the problem formulations 6.5,

6.4 and 6.6 are solved without violating any constraint at both levels of detail.
6.2
Simultaneous Optimization of Spatial and Control Structures in Reactive Distillation
The interactions between spatial and control design have been classical subjects of academic research during the last decade. It is widely accepted -in both academia and
industry- that controllability and operability issues play an essential role in the design
of chemical processes. The traditionally-used sequential approach (i.e. spatial design
followed by control structure design) leads frequently to units/process with poor controllability and responsiveness properties and inoperative in the presence of disturbances
and/or uncertainties in design parameters. Relevant developments in this eld have
been reported (e.g. Skogestad and Postlethwaite (1999) and recently by Meeuse and
Tousain (2002)) with the aim to consider operability and controllability aspects in the
early stage of the design cycle. However, most of such contributions rely on linearization
techniques around operational points, steady state models or nonapplicable/complicate
controllability indices.
115
Chapter 6
Molar fraction []
0.9
0.8
CH3OH
0.7
0.6
0.5
0.4
MTBE
0.3
0.2
0.1
iC4
2
10
12
14
16
Tray location []
Figure 6.4. Composition proles in the synthesis of MTBE obtained by a

multilevel modeling approach. Remarks: the molar fraction
of nC4 is not plotted; a target value of 150 mols1 is imposed
to the overall extent of reaction; the spatial conguration of the
unit is given in table 6.2; time horizon is 14400 s.
This section addresses the application of a dynamic optimization-based design approach

to RD. The liquid-phase esterication reaction of iC4 and methanol in the presence of
inert nC4 in a staged RD column is used as tutorial example. Similar to the study
on binary distillation (Bansal et al., 2000; Bansal, 2000) and on the synthesis of ethyl
acetate by RD (Georgiadis et al., 2002), both spatial-related (e.g. column diameter
and heat exchanger areas) and control-related (e.g. gain, set-point and reset time)
design variables are optimized with respect to economic and dynamic performance in
the presence of time-varying disturbances.
6.2.1 Description of the Design and Control Problem in Reactive Distillation

Generally speaking, a chemical process that shows poor ability of meeting control objectives gives rise to missing opportunities with respect to economic performance and
satisfaction of product specications. This fact may imply the existence of a trade-o
between those processes designed with respect to control performance considerations
116

and those driven by economic issues. In order to circumvent this inconvenient duality,
a novel design approach has been proposed (Bansal et al., 2000; Bansal, 2000; Georgiadis et al., 2002) where both economic and control oriented issues are conveniently
incorporated in the formulation of the optimization problem and simultaneously solved.
In the case of RD design, the general design and control problem involves nding parameters (continuous and/or integers), which dene the RD column and control tuning
parameters at the minimum economic cost and with an acceptable dynamic performance
in the presence of disturbances. The continuous parameters include column diameter,
reboiler and condenser areas and tuning parameters (gain Kc , set-point SP and reset
time c ), while integer variables involve the existence or not of catalyst on a given stage
(
y = 1 for reactive stage and y = 0 for nonreactive stage). Alternatively, the presence
of catalyst on a tray can be expressed using the Kronecker delta function. For the sake
of comparison with literature data, the amount of catalyst is assumed to be identical
on all reactive trays.
6.2.2 Description of the Dynamic Optimization Problem in Reactive Distillation

All involved expressions are combined in the formulation of an optimization problem.
The set of expressions includes: (i) conservation equation of species mass, energy and
electrical charge; (ii) phase and reaction equilibrium equations; (iii) thermodynamic
constitutive equation per phase, involving density, specic heat and fugacity coecients; (iv ) approximations to the momentum balance involving pressure drop, phase
ratio and phase contact area; and (v ) kinetic equation for chemical reactions, mass and
heat transfer between lumped phases. Additionally, algebraic expressions are included
to account for constitutive relations and to estimate physical properties of the components, plate hydraulics, column sizing and economic performance indices. A detailed
description of the model can be found in appendix A. Thus, the general optimization
problem is formulated as follows (Bansal et al., 2000; Bansal, 2000),
Obj. :
s.t.
min
p c
"tf
z ,z ,
y0
J(xd (tf ), xa (tf ), u(tf ), d(tf ), y(tf ), pp (tf ), pc (tf ), z p , z c , y, tf )dt,
: f (x(t),
x(t), u(t), d(t), y(t), pp (t), pc (t), z p , z c , y) = 0,

t [t0 , tf ],
p
c
p c
t [t0 , tf ],
g(x(t), u(t), d(t), y(t), p (t), p (t), z , z , y) = 0,
t [t0 , tf ],
h(x(t), u(t), d(t), y(t), pp (t), pc (t), z p , z c , y) 0,
i(x(0), u(0), d(0), y(0), pp (0), pc (0), z p , z c, y) 0,
t [t0 , tf ],
x(t) X Rx , u(t) U Ru , d(t) D Rd , y(t) Y Ry , t [t0 , tf ],
p
c
z p Z p Rz , z c Z c Rz ,
p
c
t [t0 , tf ],
pp (t) P p Rp , pc (t) P c Rp ,
y [0, 1],
(6.8)
117
Chapter 6
where the objective function is the cumulative value of J () over the time horizon
under consideration, the equalities f () and g() comprise the dierential and algebraic
expressions respectively, h() represents the path constraints that must be satised at
all times during the system operation (Bansal et al., 2000; PSE, 1999), i() represents
the set of initial conditions, x denotes the set of state variables, u is the set of input
variables, y is the set of output variables, d is the set of disturbance variables, pp and
pc are the sets of time-varying variables for the process and control scheme respectively,
z p and z c are the sets of time-invariant parameters for the process and control scheme
respectively and y is the set of integer variables. Additionally, bounds are specied for
the time-invariant and time-varying variables for both process and control scheme.
6.2.3 Spatial and Control Design Variables

Due to the high non-linearity of the physical phenomena, even small variations in the
input variables might lead to highly sensitive conditions. This fact implies that a
feasible operation of a RD column requires several control loops to mitigate the eects
of disturbances. To identify the variables to be controlled and manipulated and then
fully formulate the design problem a rigorous degree of freedom analysis is performed
(tables A.1-A.3 in appendix A). This analysis suggests that for a given set {RR, nt
Dcol , Acond , Areb } a 44 SISO control structure could be implemented to cope with
the disturbances eects. The pairing of controlled-manipulated variables aims to the
satisfaction of product specications and to the protection of the unit. The proposed
control structure is depicted in gure 6.5 and detailed in table 6.3. Note that PI
controllers are a-priori chosen.
Table 6.3. Control loops in a RD stage column. Remark: the schematic
representation of the unit is given in gure 6.5.
118
Unit Controlled variable
Manipulated variable
P1
Outgoing owrate in condenser,

out,L
Fcond
Flowrate of cold utility in condenser, F cool
Outgoing owrate in reboiler,
out,L
Freb
Heat duty of hot utility in reboiler, Qreb
P2
Liquid level in condenser,

levelcond
Pressure in condenser, Pcond
P3
Liquid level in reboiler, levelreb
P4
Temperature in reboiler, Treb

P
Pcond
Fcool
levelcond
Fout,Lcond
1
R,L
FD,L
MTBE
(<0.1%m)
reactive
zone
T Treb
MeOH
9
10
Qreb
C4-stream
15
Fout,Lreb
MTBE
(>99%m)
level reb
Figure 6.5. Control structure in the synthesis of MTBE by RD. Remark:

the manipulated and controlled variables for each control loop
are listed in table 6.3.
6.2.3.1
Case Study: Synthesis of MTBE
A detailed description of this system is available in appendix B. The spatial structure

of the RDC is similar to that reported by Hauan et al. (1995a) and Jacobs and Krishna
(1993) and detailed in the upper section of table 6.2. Two sinusoidal disturbances (gure
6.6) are included in the problem description. They account for both the variation in the
feed composition of the C4 -stream and the change in the inlet temperature of the cold
utility T in,cool during the process operation. As suggested by Georgiadis et al. (2002);
Bansal et al. (2000), the feed composition disturbance is modelled as a fast moving
function and represented by the following general expression,
zfeed
feed
+ f (t)
ziC
4 ,0
feed
zMeOH,0
,
=
feed
z
(t)
f
nC4 ,0
feed
zMT BE,0
with
f (t) := 0.05 sin(
2
t).
3600
(6.9)
The composition disturbance frequency mimic an upstream time response of 1 hour,

which is commonly occurring in industrial practice.
The temperature disturbance tries to represent the variation of inlet temperature of the
119
Chapter 6
in,cool
[K]
294
Molar fraction [-]
nC4
0.65
292
290
0.60
0.55
288
0.50
286
0.45
iC4
0.40
284
0.35
282
0
2000
4000
6000
8000
10000 12000 14000
2000
4000
6000
8000
10000 12000 14000
Time [s]
Time [s]
(a) Fast moving disturbance in the feed

compositions of iC4 and nC4 , zfeed .
(b) Slow disturbance in the inlet

temperature of the cold utility, T in,cool .
Figure 6.6. Time dependence of the disturbances scenario in the dynamic

optimization of MTBE synthesis by reactive distillation. Time
horizon=14400 s.
cooling medium during the time horizon. Thus,

T in,cool = 298 + s (t),
with s (t) := 5 sin(
2
t).
14400
(6.10)
In this case, the temperature disturbance has a time response of 4 hours.

6.2.3.2
Nominal Spatial/Control Design
Nominal design values of the spatial structure of the column for the synthesis of MTBE
(i.e. column diameter and reboiler and condenser heat exchange areas) are rstly estimated. They are obtained by solving a steady-state optimization problem, which
minimizes the total annualized cost of the RD unit in the absence of disturbances. The
following constraints are included to the problem formulation: (i) the column diameter
is bounded by ooding conditions; (ii) the heat exchange areas of the condenser and
reboiler are estimated by bounded values of outlet temperatures of the hot and cold
utilities, and (iii) the molar fraction of MTBE at the top and bottom stream is constrained to values lower than 0.1% and 99% respectively. Thus, the statics optimization
problem results in,
120
Function:
s.t. :
min
Dcol ,Areb ,Acond
J = TAC,
g() = 0,
h() 0,
min
max
Dcol Dcol
,
Dcol
min
max
Areb Areb Areb ,
max
Amin
cond Acond Acond ,
D
xMTBE 0.1% and xB
MTBE 99%,
(6.11)
where TAC is the total annualized cost of the RD unit and includes operational and
investment costs. A detailed specication of the terms involved in the denition of TAC
is presented in the appendix A.1.
The relevant results of the optimized steady-state nominal design together with the
nominal design inputs are presented in table 6.4. According to the sequential approach,
the control structure is chosen and controllers parameters are estimated only after
the spatial structure is known and a disturbance scenario is dened. For the sake of
simplicity and according to the degree of freedom analysis in appendix A, the control
structure depicted in gure 6.5 is implemented in the RD unit.
Table 6.4. Optimized steady-state design of a RD column for MTBE synthesis. The parameter xMTBE accounts for the error between
computed and set-point values for the molar fractions of MTBE in
the top and bottom streams.
Value
Number of trays
Feed 1: location
Feed 1: owrate
Feed 1: composition
Feed 1: temperature
Feed 2: location
Feed 2: owrate
Feed 2: composition
Feed 2: temperature
Operating pressure
Reux ratio
Bottom owrate
xMTBE
Column diameter
Condenser area
Reboiler area
15
455 mols1
iC4 : 0.37; nC4 : 0.63
350K
MeOH: 9th tray (l)
168 mols1
MeOH: 1.00
320K
11105 Pa (top)
7
204.1 kg=1000 eq[H + ]
197 mols1
2.2 105
7.132 m
1.059103 m2
6.93102 m2
121
Chapter 6
Tuning the controllers is initially carried out based on engineering judgement in combination with the rules of thumb given by Luyben (2002). Hence biases are set to the
corresponding steady state values, gains are such that a variation of approximately 50%
of the controlled-variable value is achieved by a change of 1% in the value of the manipulated variable and reset times are set around 1-2 minutes. To better estimate the
values of the controllers gains a sensitivity analysis is performed considering a simplied
disturbance scenario given by the variation (+10%) of the owrate of methanol. For the
sake of exemplication, gures 6.7a-d depict the dynamic behavior of the controllers
input variables in this simplied disturbance scenario. It can be noticed that an acceptable dynamic behavior of the dierent condenser and reboiler blocks is obtained
(see bold curves in gures 6.7a-d ). For practical purposes these controllers parameters are considered only as initial guesses for the optimization formulation addressed
later on. The problem of estimating the controllers parameters is now formulated as
an optimization problem in the time domain, aiming at a minimization of the integral
absolute error of the control loops. This criteria is in particular chosen for the sake of
comparison with the work of Georgiadis et al. (2002).
The two sinusoidal disturbances introduced previously (expressions 6.9 and 6.10) are
included in the optimization problem, resulting in the generic formulation,
Function:
s.t.:
min
Ki ,Biasi ,c,i ,SPi
J=
"tf

(xb xb )2 + (xd xd )2 dt,
0
f () = 0
g() = 0
h() 0
d() = 0
min
max
Treb Treb
Treb
out,min
out,max
out
Twater Twater Twater
out,min
out,max
out
Tsteam Tsteam
Tsteam
t [t0 , tf ],
t [t0 , tf ],
t [t0 , tf ],
t [t0 , tf ],
t [t0 , tf ],
t [t0 , tf ],
t [t0 , tf ],
(6.12)
where the objective function J accounts for deviations from product and top specications (
x), the constraint paths set bounds for the temperature in the reboiler and the
outlet temperatures of the cold and hot utilities and the set of disturbances (Equations
6.9 and 6.10) are grouped in the vector d().
Solving problem 6.12 leads to the set of controllers parameters listed in the second
column of table 6.5.
6.2.3.3
Simultaneous Optimization of Spatial and Control Structures
In this subsection the sequential approach adopted previously is eectively replaced

by a formulation in which both spatial and control-related constraints are included in
the optimization problem and solved for the most economic operation. Besides the
122

already mentioned constraint paths, the following spatial-related constraints have been
included,
lcond [m]
P cond [bar]
0.503
11.106
0.502
K=-100
K=-2.0 10
11.104
0.501
11.102
0.500
K=-2.0 10
11.100
K=-1000
0.499
11.098
0.498
11.096
10000
20000
30000
40000
50000
60000
10000
20000
30000
Time [s]
40000
50000
60000
Time [s]
(a) Level in the condenser
(b) Pressure in the condenser
lreb [m]
Treb [K]
0.8
429.2
K=-10
429.1
0.7
429.0
K=4.0 10
428.9
K=-100
0.6
428.8
428.7
0.5
K=4.0 10
428.6
428.5
0.4
K=4.0 10
428.4
428.3
0.3
428.2
0.2
428.1
428.0
10000
20000
30000
40000
50000
Time [s]
(c) Level in the reboiler
60000
10000
20000
30000
40000
50000
60000
Time [s]
(d) Temperature in the reboiler
Figure 6.7. Dynamic behavior of the controllers input (controlled) variables

in the synthesis of MTBE. Remarks: the disturbance scenario
involves the time dependent variation of the owrate of methanol:
SS
SS
+ (t)FMeOH , with (t)FMeOH = 0.1 FMeOH
FMeOH = FMeOH
if 10000s < t < 40000s and (t)FMeOH = 0 if 10000s > t or
t > 40000s; controllers parameters are only considered as initial
estimates.
123
Chapter 6
Function:
s.t.:
min
Ki ,Biasi ,c,i ,SPi ,Dcol ,Areb ,Acond
f () = 0
g() = 0
h() 0
d() = 0
min
max
Treb Treb
Treb
out,min
out,max
out
Tcool
Tcool Tcool
out,min
out,max
out
Theat
Theat
Theat
min
max
Dcol
Dcol Dcol
min
max
Areb Areb Areb
max
Amin
cond Acond Acond
J=
"tf
0
[TAC]dt
t [t0 , tf ]
t [t0 , tf ]
t [t0 , tf ]
t [t0 , tf ]
t [t0 , tf ]
t [t0 , tf ]
t [t0 , tf ]
(6.13)
The optimized parameters of problem 6.13 are listed in the third column of table 6.5.
By comparing the TAC of both approaches, it can easily been inferred that the simultaneous optimization of both spatial and control structures leads to a cheaper and
more controllable process than that obtained via the conventional sequential strategy.
Although the relative dierence between both economic indexes is fairly small, the accumulative cost saving sums up to 0.2 million euros per year. This cost reduction is
predominantly originated from the reduced column diameter and smaller reboiler.
The time evolution of the product compositions and temperature prole at the endpoint are depicted in gure 6.8 for the simultaneous approach. It should be noticed
that the variations of MTBE composition are negligible from a practical stand-point.
Moreover, the temperature along the unit presents a steep increase only at the stripping
section. Heat eects due to exothermicity are relatively mild.
6.3
Concluding Remarks
The conclusions derived from this chapter are grouped in accordance to the abovementioned sections. Firstly, it is worth mentioning that the approach presented in this
chapter has been applied to staged units, without it being restricted to that geometry.
Moreover, variables pairing for the control conguration has not been considered. A
MINLP approach could be used to address this last issue, as applied by Bansal et al.
(2000, 2003) for binary ideal distillation.
Multilevel modeling. The proposed multilevel modeling approach is able to provide
an easy-to-handle tool in the early stages of design. Two model levels of dierent degree
of complexity were derived to model the processes taking place in a RD column. Rigorous and reliable degree of freedom analysis were performed at each level, where appropriate business-related decision variables were included. The outer (input-output) and
124
Table 6.5. Simulation results for the conventionally-used sequential and simultaneous approaches. Remarks: inputs of the design problem
are given in table 6.4; the notation used in the control loops corresponds to the description of table 6.3 and gure 6.5; the convergence tolerance of the design variables is 1 108 ; catalyst hold-up
is considered constant (relaxation of this assumption is addressed
in chapter 8).
Variable
P1
P2
P3
P4
K1
Bias
c
SP
K2
Bias
c
SP
K3
Bias
c
SP
K4
Bias
c
SP
Dcol
Acond
Areb
Ctotal
Unit
Sequential
Simultaneous
mols1
s
m
mols1
s
bar
mols
s
m
Js
s
K
m
m2
m2
euro/y
1.000 10
2.556 103
1.000 100
5.000 101
2.000 104
4.973 104
1.200 102
1.100 101
5.000 100
1.517 102
1.200 102
5.000 101
4.000 105
4.350 107
1.000 102
4.289 102
7.132 100
1.059 103
6.933 102
2.112 108
4.999 102
2.556 103
5.047 101
5.000 101
9.999 104
4.973 104
5.000 101
1.100 101
2.500 101
1.517 102
1.500 101
5.000 101
2.000 106
4.350 107
4.761 100
4.289 102
6.924 100
1.250 103
4.500 102
2.108 108
inner (task) levels were linked by the extent of reaction vector, which allows establishing
whether or not the models are well specied and mutually consistent. Additionally, the
approach allows one to determine whether or not the business-related decision variables
cope with the operational constraints imposed by the problem formulation (e.g. SHE
issues, reactants availability and market demand).
Simultaneous dynamic optimization. In order to validate this approach, the synthesis of MTBE is used as case study. Three spatial variables and sixteen control-related
parameters were optimized during a time horizon of 14400 s. A tradeo between control
and economic performances exists in the design of RD processes, as shown in table 6.5.
The design optimized sequentially with respect to dynamic behavior led to a RD process
125
Chapter 6
T [K]
xMTBE [-]
430
1.0000
420
0.9995
0.9990
410
bottom stream
0.9985
400
0.9980
390
0.9975
380
0.00025
0.00020
370
0.00015
top stream
0.00010
360
350
0.00005
0.00000
0
2000
4000
6000
8000
10000 12000 14000
Time [s]
(a) Molar fraction of MTBE. Time

horizon: 14400 s.
10
12
14
16
Tray location [-]
(b) Temperature prole along the RD

column.
Figure 6.8. Time evolution of MTBE molar fraction in the top and bottom
streams (left) and temperature proles at the end-point (right) for
the simultaneous optimization of spatial and control structures.
Remarks: the trays are numbered from top to bottom (1st : condenser; 17th : reboiler); the sizing and control parameters are
listed in table 6.5.
with a total annualized cost higher than that obtained using simultaneous optimization
of spatial and control structures. This multiobjective optimization problem was solved
by incorporating appropriate time-invariant parameters (e.g. column diameter, heat
transfer areas and controllers parameters), in the frame of a dynamic optimization
in the presence of deterministic disturbances. A rigorous degree of freedom analysis
was performed to the involved building blocks and an adequate control structure was
implemented.

remark can be made. The complexities of RD process design are
handled in this chapter by the denition of an easy-to-handle tool
to assist the modeling task. Furthermore, a more design advanced
approach is addressed, which involves the simultaneous dynamic
optimization of spatial and control structures in the presence of
a disturbance scenario. Therefore, the material presented in this
chapter contributes to the answering of Question 6.
126
Engineering is the art or science of making practical.

Samuel C. Florman, writer and engineer
Steady and Dynamic Behavior Analysis
The steady-state and dynamic behavior in RD are studied in line with the proposed
design procedure. Two levels of aggregation are considered: a single stage reactive
ash and a full column. The rst conguration is used to address the steady-state
behavior, whereas the full column is used as case study for the process dynamics. By
selecting a reactive ash we focus exclusively on the phenomena interaction rather
than on equipment eects. In this regard, twenty-ve bifurcation diagrams are presented, exhibiting a maximum of three steady states and ve feasibility boundaries.
Singularity theory is used to divide the parametric space into regions with qualitatively dierent behavior. Furthermore, parametric dependence of MSS regions is
studied. Additionally, the dynamic behavior of a complete MTBE RD column is covered. Two deterministic disturbances are included to analyze the unit dynamics. The
control structure is a conventional SISO and PID based one, aiming at disturbance
rejection. This structure is dened based on a combination of rigorous degree of
freedom analysis and engineering judgement. The variations encountered in product
compositions reveal that the unit operates eectively in the presence of disturbances.
The output of this analysis denes the reference case for the new developments to be
introduced in chapter 8.
Section
7.2.1 on multiplicities in a reactive ash has appeared as Lakerveld et al. (2005).
127
Chapter 7
7.1
Introduction
The sixth design space of the multiechelon approach (cf. chapter 4) in reactive distillation is the focus of this chapter. After having dened the spatial and control structures
the process behavior in steady-state and dynamic domains are now addressed. The
input and output information ow at this design space is detailed on table 7.1.
Table 7.1. Input-output information for the behavior analysis space. The
Design specications
Process:
- unit structure
- control structure
Continuous:
- controllability index
- operating conditions
- steady-state behavior
- dynamic behavior
The analysis of steady-state behavior allows the designer to explore the eect of kinetics,
gas/liquid mass transfer and hydrodynamics and the steady-states of the unit. On the
other hand, dynamic simulations are carried out to check the robustness of the design,
in terms of the ability of maintain product purities and conversion in desired range
when disturbances are occurring.
This chapter addresses those two aspects of process behavior: steady-state and dynamic
operation. In the rst case, a reactive ash with a rather ideal exothermic isomerization
reaction is considered. This simplied system is selected to account exclusively on the
eect of phenomena interaction on the occurrence of input and output multiplicity.
Therefore, any eect related to unit conguration is considered outside the scope of the
analysis.
The second part of this chapter deals with the dynamic behavior of a full RD column,
with special focus on the synthesis of MTBE. Although the design of the control structure is rather conventional (SISO and PI controllers), it is regarded as reference case
for the new developments to be introduced in chapter 8.
128
Steady and Dynamic Behavioral Analysis
7.2
Steady-State Behavior
7.2.1 Introduction
Section 2.9 has addressed the generalities and sources of steady-state multiplicity in
RD processing. Furthermore, special attention has been paid to the implications of
MSS for the design and operation of RD. In this section the steady-state behavior of a
simplied RD unit (i.e. reactive ash) is covered for the case of an exothermic isomerization reaction with rst-order kinetics and light-boiling reactant. Several approaches
are employed to analyze the steady-state behavior of a process model. For instance,
regions of MSS can be identied by tracing the dependence of the steady state on one
varying parameter (Mohl et al., 1999; Hauan et al., 1995a; Ciric and Miao, 1994; Baur
et al., 2001b; Sneesby et al., 1998b,a; Chen et al., 2002). The starting point of our
approach is provided by the study on MSS in a reactive ash by Rodriguez et al. (2001,
2004). They show that multiplicity is caused by the interaction between separation
and reaction in systems in which the activation energy and the gradient of boiling temperature with composition are suciently large. Because VL equilibrium can create
multiple steady states, state multiplicity becomes possible in endothermic or slightly
exothermic systems, which typically do not display multiplicity in a one-phase CSTR.
Several examples were considered and parametric studies were performed. In addition,
for an isomerization reaction, necessary conditions for multiplicity were derived. These
conditions are somehow dicult to apply, because they are expressed in terms of system
properties (e.g. heat of vaporization or heat of reaction), but also include, however, the
unknown liquid and vapor phase compositions.
Although Rodriguezs work is restricted to a single stage ash, it clearly shows that the
combination of reaction and separation is a key cause for multiplicity in RD.
In this section, we approach the reactive ash from a dierent perspective. We employ
a dimensionless model and distinguish the state variables (e.g. liquid and vapor compositions), from system physical parameters (e.g. heat of reaction or activation energy)
and control parameters (e.g. feed rate or heat duty). Aiming to uncover the possible
behavior patterns, we present a considerable number of bifurcation diagrams that are
qualitatively dierent. We demonstrate the importance of the feasibility boundaries
corresponding to no-liquid or no-vapor products, illustrated in bifurcation diagrams exhibiting three and even ve such points. Finally, we show that high heat of reaction,
activation energy and reactant volatility enlarge the range of operating parameters for
which multiple states exist. To the best of our knowledge, such a detailed analysis has
not been provided yet for the case of RD. Section 7.2.1 is based on the contribution
entitled Exothermic isomerization reaction in reactive ash: steady state behavior by
Lakerveld et al. (2005).
129
Chapter 7
7.2.2 Systems Features and Model Development

The system under consideration intends to be as simplied and ideal as possible, so
that we focus exclusively on the interaction between occurring phenomena. Thus, the
properties of the system include: (i) binary system undergoing a kinetically controlled
isomerization reaction A B, (ii) the RD ash is operated at isobaric conditions, (iii)
the chemical reaction takes place exclusively in the liquid phase and is exothermic, (iv )
the reactant A is the lighter component in the system, (v ) both phases are assumed to
be well mixed, (vi) gas and liquid phases behave ideally, (vii) vapor holdup is negligible
compared to the liquid holdup, (viii) vapor pressures are described by Antoines correlation and (ix ) both components have equal physical properties (i.e. heat of vaporization,
heat capacity and density).
TC
Figure 7.1. Schematic representation of a reactive ash for an isomerization

reaction in the liquid phase.
According to the scheme of gure 7.1, pure reactant A is fed to the unit with a ow
rate F, pre-heated to temperature T0 and with a vapor fraction . The ash is equipped
with a level controller and a pressure controller to maintain the liquid holdup M L and
operating pressure P at desired values. The liquid ow rate L and vapor ow rate V are
used as manipulated variables, respectively. Heat is allowed to be added or removed at a
rate Q in the liquid phase. For the sake of model solvability, a dimensionless model has
been derived for the conguration depicted in gure 7.1. The set of governing equations
and physical properties of the system are given in tables 7.2 and 7.3, respectively.
The reference values for temperature, ow rate and pressure are: the boiling point of
component A (i.e. Tb,A ) and the nominal values of feed ow rate (i.e. Fref ) and pressure
(i.e. Pref ) for an existing design, respectively.
130
Table 7.2. Set of governing dimensionless expressions for the reactive ash as
depicted in gure 7.1.
Balance
Dimensionless expression
Overall mass
= l+vf
Component
BP condition
L-V equilibrium
Energy
dx
dt
1
= Da
(f z l x v y) x e 1+
= x p01 + (1 x) p02 p
= y p x p01
= f (0 ) + Da B x e 1+ v + q
: liquid feed, 0 < 0
0
=

: liquid-vapor feed, 0 = 0
CpV
+ Cp 0 : vapor feed, 0 > 0

Ea

M
RT
ref
k(T
)
ref
Da
Fref
r H
B CpTref
F
f
Fref
v H
CpTref

z
z
=
, S a1 =
A1
T0 Tref

0
Tref

b
1
1
P
p
Pref
a
A
2
2
Q
q
A1B2
FrefCpTref
b2
A B
0
0
0
Feed temperature 0
x
l
y
v
Remark: the validity of the model is limited by the boundaries v = 0 and l = 0.

Nomenclature: X are unknown variables, O are operation parameters, S are system parameters; z, x and y are the molar fractions in the feed, liquid and vapor phases, respectively; f , l and v are the dimensionless feed, liquid and vapor ow rates, respectively
ref ),
(f = F/Fref , l = L/Fref , v V /Fref ), is the dimensionless temperature ( = T T
Tref
is the vapor fraction in the feed ( [0, 1]), the component saturation pressure is given
1bj +
by: p0j = exp(aj 1+
), i Znc , Cp is the heat capacity and the feed temperature is
estimated as,
: T0 < Tb,A
T0
v
T0 = Tb,A + CpH
: T0 = Tb,A
CpV
v H
Tb,A + Cp + Cp (T0 Tb,A ) : T0 > Tb,A
7.2.3 Classication of Steady States

From the model parameters the dimensionless feed, f, is chosen as distinguished continuation parameter and used to depict the dependence of liquid composition, x, for a VL
feed (=0.13, T0 =280K) at dierent values of dimensionless heat duty, q (gure 7.2).
The relevance of the parameter choice is grounded by the fact that feed ow rate determines the process throughput and is frequently an operational parameter subjected
to input disturbances (e.g. market demands and supply uctuations). The parametric
space considered in this study is given by Damk
ohler number and dimensionless heat
duty.
131
Chapter 7
Table 7.3. Properties of the reactive ash system as depicted in gure 7.1.
Remarks: data are taken from Rodriguez et al. (2001, 2004); accordingly, the reactant is the light-boiling component.
Property
Dimensional variable
Dimensionless variable
Pre-exponential factor k0 =5.88 10 s

Antoine coecients
A1 = 9.784
A2 = 7.704
B1 = 2428
B2 = 2428
Heat of vaporization v H = 4807 kcalkmol1
Heat capacity
Cp = 20.0 kcal(kmolK)1
Heat of reaction
r H= 500 kcalkmol1
Activation energy
Ea = 14454 kcalkmol1
10
a1 = 9.784
a2 = 7.704
b1 = 1
b2 = 1.27
=1
B = 0.1
= 30
Three qualitatively dierent bifurcation diagrams can be observed if the heat input to
the system is varied. For adiabatic operation (q=0) no feasible solution exists for small
f . As f increases, the number of steady states changes from 0-2-3-1. The lower branch
of the curve starts at a feasibility boundary (v=0, marker: ) and exhibits two turning
points (marker: ), also known as limit or fold . If the heat duty is further increased
(q=1), the bifurcation diagram presents the pattern 0-1-3-1. For even larger values
of q, the bifurcation curves start at the l -boundary (i.e. l = 0 feasibility boundary,
marker: ) and a single steady state occurs. It can be noted that the type of diagram
depends on the number and relative location of turning points and feasibility boundaries.
These points are normally termed codimension-1 bifurcation or singular points, as they
correspond to certain values of one single parameter (e.g. f ).
Based on the multiplicity behavior of the RD model, a singularity theory approach
(Bildea and Dimian, 1999; Gehrke and Marquardt, 1997; Balakotaiah and Luss, 1984)
is used to divide the space of model parameters, excluding the distinguished one, into
regions of qualitatively dierent bifurcation pattern. All nonlinear equations are solved
with a Newton-Raphson solver routine combined with a local parametrization algorithm
(Seydel and Hlavacek, 1987; Razon and Schmitz, 1987; Wayburn and Seader, 1987;
Vadapalli and Seader, 2001; Dhooge et al., 2003) to track possible turning points. To
deal with the large number of parameters involved, we adopt the following strategy.
First, we achieve the classication in the Da-q space, for a system characterized by
small heat of reaction, moderate activation energy and high relative volatility. For a
VL feed, we identify the main types of behavior occurring in large regions and the
The following notation is used for the bifurcation diagrams: BD(p, x; p , p ), where p denotes the
1 2
distinguished parameter, x is the state variable and p1 and p2 are systems parameters. A BD can be
eectively dened by specifying parameter p1 or p2 in addition to the distinguished parameter p.
132

x []
0.9
q=5
0.8
q=2.5
q=1
0.7
q=0
0.6
0.5
0.4
0.3
0.2
0.1
0
1
10
10
10
f []
Figure 7.2. Bifurcation diagram f-x at dierent values of dimensionless heat

duty for a reactive ash undergoing an exothermic isomerization.
Symbols: (): v = 0 feasibility boundary; (): l = 0 feasibility
boundary; (): limit point. Systems features: Da=0.10; z=1;
p=1; 0 =0.13; physical properties of the system are listed in table
7.3; governing expressions are given in table 7.2.
important varieties bounding these regions. Afterwards, the eect of the feed condition
0 , the remaining control parameter, is investigated. Finally, we look at the eect of
larger heat of reaction, larger activation energy and lower relative volatility.
Singularity theory states that the qualitative features of the bifurcation diagram may
change only when the parameter set crosses the hysteresis, isola, or double-limit varieties
(Golubitsky and Schaeer, 1985). Only the hysteresis variety is found to exist for the
reactive ash model. When the hysteresis variety is crossed, the number of possible
steady states changes by two, as two limit points appear or disappear. When feasibility
boundaries exist, the bifurcation diagram may also change at special sets of parameters
(Balakotaiah and Luss, 1984). The following varieties are considered in this study,
Variety 1. boundary-limit set (BL): a limit point occurs at a feasibility boundary,
Variety 2. boundary - tangent set (BT ): the bifurcation diagram is tangent to the
feasibility boundary. After crossing this set (e.g. moving from high to low
dimensionless heat duty q in gure 7.5) the bifurcation diagram changes
by appearance / disappearance of two feasibility boundaries,
133
Chapter 7
Variety 3. cross-and-limit set (CL): the position of one limit point changes relative to
one solution located at the feasibility boundary,
Variety 4. corner set (C ): the bifurcation parameter and the state variable are, simultaneously, at their boundaries.
Because the dening conditions can be solved for the state variables and two parameters (e.g. f and q), the above mentioned varieties are said to be of codimension-2 .
The dynamic model of the reactive ash contains several algebraic, but only one dierential equation, when the holdup and pressure are xed and the phase equilibrium is
instantaneous. Such one-dimensional systems cannot exhibit Hopf bifurcations leading
to oscillatory behavior. Therefore, dynamic classication is not necessary.
A better understanding of the steady state classication is carried out by depicting
the loci of codimension-1 singular points in a f -q plot (gure 7.3). As the BD(f, x, q)
bifurcation diagrams are obtained for a given value of q, the number and relative location
of singular points can be obtained by the intersection of q=constant (dotted horizontal
lines in gure 7.3) with the loci of codimension-1 singular points. It can be extracted
from the gure that at large q, there is only one intersection with the l-boundary,
corresponding to the bifurcation diagram I +III in gure 7.4. As q decreases, a narrow
region with three l-boundaries is encountered. This region is bounded by the boundarytangent points BT1 and BT2 in the q-f plot (zoomed view in gure 7.5a). Diagram
II in gure 7.4 represents the behavior in this region. As q is further decreased, two
turning points are met below the cusp H. State multiplicity becomes possible, as shown
in diagram IV of gure 7.4. Note that the loci of cusp points is called hysteresis.
Many qualitative changes occur within a small elbow region around q=0, which is
zoomed in Figs. 7.5b-d. If we decrease the value of q, the boundary-tangent BT3
is rstly crossed to region V (gure 7.5c) leading to two v- boundaries. Immediately,
the relative locations of a turning point and a v-boundary change at a cross-and-limit
point CL, leading to diagram VI (gure 7.4). By further decreasing q to region VII
(gure 7.5b), the v- and l-boundaries move towards f =0 (left section in gure 7.3) and
disappear at a corner point (not shown). One turning point leaves the feasible domain
at the boundary limit BL1 and the maximum number of steady states becomes two, as
in diagram VIII (gure 7.4). Finally, the turning point left becomes unfeasible and the
state multiplicity disappears as depicted in gure 7.4-diagram XI. Furthermore, a very
narrow range of q values exists (gure 7.5d ) where three v-boundaries occur (diagrams
IX, X in gure 7.4).
134
7.2.4 Eect of Damkohler Number, Da

The steady state classication is achieved by tracing the codimension-2 singular points
when an additional parameter (e.g. Da) is changed. In this way, the Da-q space is
divided into regions where qualitatively dierent f -x bifurcation diagrams exist (gure
7.6). For each region, the corresponding diagram is presented in gure 7.4. It should
be noticed that regions II, V, IX and X, exhibit three feasibility boundaries and are
very small.
7.2.5 Eect of Feed Condition

In the previous section a VL feed (0 < 0 < 1) has been considered. Among the
bifurcation diagrams of gure 7.4, those presenting two disconnected branches and
three feasibility boundaries should be remarked. Such diagrams are obtained when BT
sets are crossed (see gures 7.5a-d ), but are occurring exclusively in narrow regions of
the Da q space (gure 7.6). This subsection discusses the eect of the feed condition
(i.e. sub-cooled liquid: 0 < 0 and over-heated vapor: 0 > 1) on the regions with
qualitatively dierent bifurcation pattern.
q []
l=0
3.5
2.5
unfeasible
region
I
III
H
1.5
feasible
region
0.5
1
2
10
0.5
MSS
region
v=0
unfeasible
region
IV
XI
1
10
f []
10
10
Figure 7.3. Codimension-1 singular points for a reactive ash undergoing an

exothermic isomerization. Nomenclature: H : hysteresis variety; T: turning point loci. Systems features: Da=0.10; z=1;
p=1; 0 =0.13; physical properties of the system are listed in table
135
Chapter 7
I+III
0-1
VI
0-1-0-2-3-1
II
IV
0-1-0-1
0-1-3-1
VII
0-2-3-1
VI
V
0-1-3-2-3-1
VIII
IX
0-2-1
0-2-1-2-1
VII
XI
0-1-2-1
0-1
Figure 7.4. Qualitatively dierent bifurcation diagrams for a reactive ash

undergoing an isomerization reaction. Symbols: (): if v = 0;
(): if l = 0; Systems features: Da=0.10; z=1; p=1; 0 =0.13;
physical properties of the system are listed in table 7.3; governing
expressions are given in table 7.2.
In the f q space (gure 7.7), the qualitative bifurcation pattern of the v- and lboundaries change at values of 0 = 0 and 0 = = 1, respectively. By comparing
gure 7.5b-d with gure 7.7a, it can be noticed that the boundary tangent sets BT3
and BT5 disappear when the vapor fraction increases (e.g. 1) and when the feed
is a sub-cooled liquid (i.e. 0 < 0), respectively. However, the distance between two
boundary-tangent sets becomes larger, which is reected by larger regions where the
bifurcation diagram consists of two disconnected branches separated by v = 0 points.
Moreover, for the l-boundary (gure 7.7b), the set BT2 disappears at 0 = = 1.
It is noted, additionally, that for vapor-only (0 > 1) or liquid-only (0 < 0) feeds the
f x bifurcation diagram exhibits at least one v-boundary and one l-boundary. Finally,
for nominal feed conditions (i.e. 0 < 0 < 1) at least one feasibility boundary exists,
depending on the value of q.
As a vapor product from a liquid feed (0 < 0) is possible only if heat is supplied
(q > 0), the feasible region in gure 7.7a is located at the upper plane. Similarly, as
a liquid product from a vapor feed (0 > 1) is attained exclusively by heat removal
(q < 0), the feasible region in gure 7.7b occupies the lower plane. Thus, when the feed
is sub-cooled liquid or over-heated vapor, the reactive ash model has feasible steady
states only over a nite range of heat duty.
Phase diagrams for these three 0 -regimes (i.e. 0 < 0; 0 < 0 < 1 and 0 > 1) exhibit
dierent qualitative topology (see gure 7 in Lakerveld et al. (2005)). That gure
136

2.35
l=0
0.04
CL
0.02
2.3
BT
BT
VII
0.02
II
2.25
BT2
0.04
III
2.2
BL1
VIII
BL
0.06
BT4
BT
0.08
2.15
0.31
0.34
10
0.37
10
0.4
10
10
(a) Zoomed view at inection point
0.1
0.1
10
10
(b) Zoomed view at elbow location

0.05
0.02
0.06
0.018
BT3
0.016
0.07
V
VI
IX
X
5
v=0
0.2
BT4
0.08
0.012
10
BL2
BT
CL
0.014
0.01 0.3
10
0.1 0.3
10
0.1
10
10
0.09
0.1
10
(c) Zoomed view at upper elbow branch
0.1 0.3
10
0.2
10
0.1
10
10
(d) Zoomed view at lower elbow branch
Figure 7.5. Zoomed view of gure 7.3. Codimension-1 bifurcation points in

the f q plane. Remark: the parametric space is f (x-axis)q (y-axis). Nomenclature: BT : boundary tangent; CL: crossand-limit; BL: boundary limit. Systems features: Da=0.10;
z=1; p=1; 0 =0.13; physical properties are listed in table 7.3;
governing expressions are given in table 7.2.
suggests that increasing Damkohler number might move an operating point into a region
with multiplicity if 0 < 1. Furthermore, regardless the 0 -value, lower heat input q
might move an operating point into a multiplicity region. Additionally, increasing 0
(with 0 < 1) reduces the multiplicity region as the hysteresis variety is shifted to the
right of the plane. At 0 1 the hysteresis variety moves to negative values of q and
a qualitatively dierent topology is obtained .
The eect of the systems properties on the steady state behavior is conveyed in the
next subsection. As the level of detail might be considerably increased, three physical
properties are considered: heat of reaction, activation energy and relative volatility.
The ultimate goal of this analysis involves the denition of trends in the steady state
behavior, rather than a rigorous quantitative classication. For the sake of comparison,
the system introduced in table 7.3 is labelled as reference case.
The work of Lakerveld et al. (2005) presents 14 additional bifurcation diagrams, obtained by
detailed analysis of the inuence of feed condition and systems physical properties on the codimension-1
singular points.
137
Chapter 7
Figure 7.6. Phase diagram for the reactive ash model. Remark: numbers
correspond to f x bifurcation diagrams in gure 7.4. Nomenclature: H : hysteresis variety; C: corner set; BT: boundary tangent set; CL: cross-and-limit set; BL: boundary limit set. Systems features: z=1; p=1; 0 =0.13; physical properties of the
system are listed in table 7.3; governing expressions are given in
table 7.2.
7.2.6 Eect of Heat of Reaction

Increasing the adiabatic temperature rise B (from B = 0.1 (reference case) to B = 1.0)
leads to the relocation of the hysteresis variety to lower values of Da and higher values
of q. This situation results in a larger area of state multiplicity. However, the MSS
region does not vanish when the heat of reaction is reduced to zero. Figure 7.8 depicts
the feasibility boundaries at large heat of reaction and dierent feed condition 0 .
Compared to the reference case (cf. gure 7.7), the feasibility boundaries exhibit different shapes and are located at lower values of the heat input q. This last situation
is justied by the fact that as more heat is generated by the reaction, less heat has to
be added to the system to vaporize the entire feed (l=0), or to generate vapors from
a given feed (v=0). Furthermore, the above-mentioned limiting cases (i.e. 0 = 0 and
0 = ) are recognized.
In gure 7.9 the loci of codimension-1 points are depicted in the f q space. As the
heat of reaction is increased, the cusp point H crosses the l-boundary at a codimension3 point and moves into the unfeasible region. Although the area between the loci of
138

q []6
* =1.0
0
feasible region
*0=0.13
* =0
0
* =0.13
unfeasible region
*
0=1.0
6
2
10
10
10
10
f []
(a) v = 0 feasibility boundaries

q []
10
* =1.0
0
*
0=0.5
*
0=0
4
* =0.13
0
*0=0.5
* =1.0
unfeasible region
0
feasible region
1
10
10
f []
(b) l = 0 feasibility boundaries
Figure 7.7. Eects of feed condition on feasibility boundaries. Systems features: Da=0.10; z=1; p=1; physical properties are listed in table
the turning points is partly in the unfeasible region, not all branches of the bifurcation
diagram are unfeasible. These changes result in new types of bifurcation diagrams, as
has been reported by Lakerveld et al. (2005) (cf. gure 13 in Lakerveld et al. (2005)).
139
Chapter 7
q [] 35
30
*
0=0.5
25
20
*
0=0.2
15
10
* =0.0
* =0.1
0
10
15
3
10
10
10
10
10
10
f []
(a) v = 0 feasibility boundaries

q [] 4
2
*
0=0.95
0
* =1.0
* =1.05
10
12
2
10
10
10
10
10
f []
(b) l = 0 feasibility boundaries
Figure 7.8. Eects of feed condition on feasibility boundaries at large heat

of reaction. Systems features: B=1.0; Da=0.10; z=1; p=1;
physical properties are listed in table 7.3; governing expressions
are given in table 7.2.
7.2.7 Eect of Activation Energy

As shown in gure 7.10a, increasing the activation energy (from = 30 (reference)
to = 45) leads to a larger multiplicity area. The cusp point becomes physically
unfeasible, as it occurs at l < 0. The elbow region in the l-boundary becomes more
pronounced, resulting in larger regions with bifurcation diagrams exhibiting three fea140

sibility boundaries.
7.2.8 Eect of Relative Volatility

In the reference case, the relative volatility (=p01 /p02 ) has a large value ( 8). As
reported in the case of synthesis of ethylene glycol by RD (Ciric and Miao, 1994),
high volatility contributes to state multiplicity. The eect of reducing the relative
volatility is addressed in this subsection. A value of =4.4
is considered, keeping all

other parameters at their reference value. As depicted in gure 7.10b, the extent of
the multiplicity region becomes smaller than that in the reference case (cf. gure 7.3).
Moreover, the feasibility boundaries are less relevant, as the elbow region on the lboundary disappears and the region between the turning points on the v=0 curve is
quite small.
The eect of increasing reaction heat and activation energy and decreasing relative
volatility is illustrated in gure 7.11. The multiplicity area remains small with respect
to the reference case, but the feasibility boundaries become more signicant. Lakerveld
et al. (2005) report bifurcation diagrams that exhibit up to ve feasibility boundaries.
q []
l=0
3
unfeasible
region
feasible
region
T
MSS
region
T
v=0
unfeasible
region
2
2
10
10
10
10
f []
Figure 7.9. Eects of heat of reaction on codimension-1 singular points.

Nomenclature: H : hysteresis variety (cusp loci); T : turning
point loci. Systems features: B=1.0; Da=0.10; z=1; p=1;
0 =0.13; physical properties are listed in table 7.3; governing expressions are given in table 7.2.
141
Chapter 7
q []50
l=0
H
40
30
unfeasible
region
20
feasible
region
T
T
10
MSS
region
0
v=0
unfeasible
region
10
10
10
10
10
f []
(a) Large activation energy, =45
q []
l=0
0.8
0.6
unfeasible
region
feasible
region
0.4
T
0.2
T
MSS
region
v=0
unfeasible
region
0.2
1
10
10
10
f []
(b) Smaller relative volatility, =4.4
Figure 7.10. Eects of activation energy and relative volatility on

codimension-1 singular points. Nomenclature: H : hysteresis
variety (cusp loci); T : turning point loci. Systems features:
Da=0.10; z=1; p=1; 0 =0.13; physical properties are listed in
table 7.3; governing expressions are given in table 7.2.
7.2.9 Signicance of Results and Extension Possibilities

The results of this section show that a considerable number of bifurcation diagrams can
be identied even for a simplied RD application. By addressing a single reactive tray
we aim to explore exclusively the interaction between phase equilibrium and chemical
reaction. The possibilities to extend this approach to a more realistic and full-scale RD
142

system are numerous.
q []
60
H
l=0
40
T
20
unfeasible
region
MSS
region
20
unfeasible
region
feasible
region
v=0
40
60
2
10
10
10
10
10
f []
Figure 7.11. Combined eects of heat of reaction, activation energy and relative volatility on codimension-1 singular points. Nomenclature: H : hysteresis variety; T : turning point loci. Systems
features: B=5; =50;
=4.4;
Da=0.10; z=1; p=1; 0 =0.13;

physical properties are listed in table 7.3; governing expressions
are given in table 7.2.
For instance, the idealized physical properties of the system could be replaced by ratelimited mass/heat transfer models, as suggested by Baur et al. (2001c); Higler et al.
(1999b). Multiplicity behavior is reported in both models for the case of MTBE synthesis and only diering in the relative size of region of occurrence. Additionally, if
we extend our reactive tray to a RD column with reactive and nonreactive sections,
the eect of both regions on MSS occurrence should be borne in mind. As mentioned
by (G
uttinger and Morari, 1997), these interactions could be largely responsible for
multiplicity. In both cases, singularity theory approach could be useful to investigate
the main behavior patterns. The approach presented in this section is not restricted
to our case study but could be equally applicable to more complex reactive distillation
systems, including multicomponent, multireaction and nonideal systems.
143
Chapter 7
7.3
Dynamic Behavior
7.3.1 Introduction
The nonlinear coupling of reactions, transport phenomena, phase equilibria and hardware hydraulics can give rise to highly complicated dynamic behavior of the system. In
the previous section, we showed that even for a simplied reactive ash and an idealized
system a large number of bifurcation diagrams could be identied. The main source of
multiplicity is then suggested to be the strong nonlinear coupling of reactions, transport
phenomena and phase equilibria.
The strong parametric sensitivity resulting from this coupling largely determines the
start-up procedure, steady-state and transient operations (Ruiz et al., 1995). Moreover,
even small variations in the input variables might lead to unstable conditions (Alejski
and Duprat, 1996). This fact implies that a feasible operation of a RD column requires
several control loops to mitigate the eects of disturbances.
In this section, we explore the process dynamics of a full RD unit. The scope of this
analysis is restricted to a limited version of control system design, as a SISO structure
and PID controllers are a priori selected. The starting point of the analysis involves the
development of a reliable mathematical model, capable of predicting with acceptable accuracy the unit behavior. Moreover, the model should not be computational expensive
at this conceptual design level. To identify the variables to be controlled and manipulated and then fully analyze the dynamic behavior of the system a rigorous degree of
freedom analysis is performed. Moreover, the controllers have to be eectively tuned
and a realistic disturbance scenario has to be dened. The analysis of the unit involves
assessing its capability of coping with the disturbance scenario without violating the
specication constraints. It is worth mentioning that this approach does not address
any interaction between column design and control design.
The content of this rather classical control approach sets the stage for a next novel
contribution to be introduced in chapter 8, involving the trade-o between design,
responsiveness and sustainability aspects. Thus, the outcome of this section serves as
a reference case to be compared with that of the life-span inspired design methodology
(cf. section 1.4).
As in previous chapters, the synthesis of MTBE is used as case study. As the focus
of this section is the dynamic behavior of the unit around a desired steady state, we
choose appropriate processing conditions and design parameters to avoid multiplicity as
much as possible. The steady state of interest is the high-purity one as given elsewhere
Hauan et al. (1995a,b); Jacobs and Krishna (1993); Schrans et al. (1996).
144
7.3.2 Model Description, Control Structure and Disturbance Scenarios

7.3.2.1
Model Description
A RD column is considered for modeling and analysis purposes. During the model
development of this unit the following set of assumptions are made: (i) the liquid and
vapor streams leaving a trays are in thermodynamic equilibrium; (ii) both liquid and
vapor hold-ups are fully considered; (iii) the liquid and vapor on each tray are perfectlywell mixed (i.e. no radial gradient in temperature and composition and the liquid on
the tray has a concentration equal to that of the liquid leaving the tray); (iv ) the
reactions are taking place in the liquid phase and enhanced by a solid catalyst; (v ) the
reactions occur only on the trays containing catalyst; (vi) vapor and liquid entrainment
is neglected; (vii) the column is equipped with a condenser, reux drum and reboiler;
(viii) the location(s) of the feed stream(s) and product stream(s) are known a priori; (ix )
the sizing parameters of the unit and operational variables are obtained from previous
design stages (cf. section 6.2); (x ) the liquid hold-up on a tray is related to the
liquid outow according to Francis weir formula; (xi) the ow of vapor leaving a tray is
determined by the pressure drop between two consecutive trays; (xii) the tray pressure
drop includes contributions of dry pressure drop, liquid height (weir height plus height
over the weir) and residual head; (xiii) the physical properties of the involved streams
are computed as a function of temperature and/or phase composition; (xiv ) the heat
of reaction is omitted in the energy balance, because the heat of formation at standard
conditions (1.033 105 Pa/298 K) is used as reference in the enthalpy computations; (xv )
the reboiler, condenser and reux drum are nonreactive units; (xvi) the tray sizing is
estimated by rules-of-thumb criteria (Coulson and Richardson, 1991); (xvii) the column
diameter is larger than the minimum ooding value.
The RD model consists of sets of algebraic and dierential equations, which are obtained
from the mass, energy and momentum balances performed on each tray, reboiler, condenser, reux drum and PI controller instances. Additionally, algebraic expressions are
included to account for constitutive relations and to estimate physical properties of the
components, plate hydraulics and column sizing. Moreover, initial values are included
for each state variable. A detailed description of the mathematical model can be found
c
c
/gOPT
and solved using for
in appendix A. The model is implemented in gPROMS
the DAE a variable time step/variable order Backward Dierentiation Formulae (BDF).
A rigorous degree of freedom analysis is performed to asses the consistency of the model
and identify the set of controlled and manipulated variables. According to this analysis
(tables A.1-A.3 in appendix A), the following set of input variables should be xed,
the reux ratio, RR
the RD column diameter at each location, nt Dcol ,
145
Chapter 7
Table 7.4. Optimized design of a RD column for MTBE synthesis as obtained

in chapter 6.
Value
Number of trays
Feed 1: location
Feed 1: owrate
Feed 1: composition
Feed 1: temperature
Feed 2: location
Feed 2: owrate
Feed 2: composition
Feed 2: temperature
Operating pressure
Reux ratio
Bottom owrate
Column diameter
Condenser area
Reboiler area
MTBE bottom fraction
15 [2:rectifying;8:reactive; 5:stripping]
455 mols1
iC4 : 0.37; nC4 : 0.63
350K
MeOH: 9th tray (l)
168 mols1
MeOH: 1.00
320K
11105 Pa (top)
7
204.1 kg=1000 eq[H + ]
197 mols1
7.132 m
1.059103 m2
6.93102 m2
0.990-0.998
the condenser area, Acond , and

the reboiler area Areb .
The synthesis of MTBE from a C4 -stream and methanol in the presence of an acidic
catalyst is used as tutorial application. A detailed description of this system is available
in appendix B. The spatial structure of the RDC is that obtained by optimization in
chapter 6 and summarized again in table 7.4. Discrete variables are arbitrarily xed
for the sake of simplicity. Thus, the structural conguration (i.e. number of trays
and streams staging) used in this example corresponds to the one used by Jacobs and
Krishna (1993); Hauan et al. (1995a); Seader and Henley (1998). The values of these
variables are also included in table 7.4.
7.3.2.2
Control Structure
To mitigate any eect caused by a disturbance scenario a control structure is proposed

for the RD unit depicted in gure 7.12. The following assumptions are considered: (i)
four control valves are associated with the RD unit; (ii) the unit is operated at constant
reux ratio RR (=F R,L /F D,L =7); and (iii) the throughput of the unit is set.
146
Fcool
;;;
;;;
;;;
1
Fout,Lcond
FR,L FD,L
MTBE
(<0.1%m)
reactive
zone
MeOH
9
10
Qreb
C4-stream
15
Fout,Lreb
MTBE
(>99%m)
Figure 7.12. Schematic representation of a RD column in the synthesis of

MTBE. Remark: numbers represent tray location.
In agreement with the degree of freedom analysis in table A.3 (appendix A) a 44

SISO control structure is suggested. The controlled variables are chosen aiming at
the satisfaction of product specications and protection of the unit. The pairing of
controlled-manipulated variables, on the other hand, tries to minimize cross interactions
between manipulations of input variables on the controlled output variables and looks
for a dominant 1-1 correspondence between input and output variables. Thus, two
control valves are used to control the liquid levels in the reux drum and column bottom.
As the reux ratio is larger than a value of 4 (RR > 4), the reux drum level is controlled
with the liquid reux or drum liquid outow (Richardsons rule (Luyben et al., 1999)).
The level at the column bottom might be by convention controlled with the reboiler
liquid outow. The third control valve is used to manipulate the owrate of cooling
utility to control the column pressure. The remaining control degree of freedom is used
to control the purity of the MTBE product by manipulating the reboiler heat duty. The
composition of MTBE at the bottom stream is conveniently inferred from temperature
measurements at a given tray. Steady-state open loop simulations are carried out to
assess the eect of varying manipulated variable (i.e. reboiler duty Qreb ). Figure
7.13 shows that the largest temperature changes are occurring either at the rectifying
section, the top section of the reactive section or at the reboiler. By selecting the reboiler
temperature to infer the MTBE composition in the bottom stream, any hydraulic lag
is eectively avoided.
147
Chapter 7
T reb [K]
440
430
base case
420
410
400
390
380
+ 0.5% Q
370
reb
- 0.5% Q
360
reb
350
2
10
12
14
16
Tray location [-]
Figure 7.13. Eect of reboiler heat duty on the temperature prole in an

MTBE RD column.
(condenser: 1st ) to
stream is withdrawn
cations are listed in
Remarks: trays are numbered from top

bottom (reboiler: 1st ); MTBE product
from the bottom location; column specitable 7.4 .
This single-end point composition control structure is originated from the assumption
of invariant reux ratio. As mentioned by Luyben et al. (1999), this control structure is
advantageous from both operational (i.e. easy tuning) and economical (i.e. comparable
energy consumption) point of view compared to a dual composition control. The overall
control structure is labelled [RR] Qreb , meaning that the reux ratio is x and the
reboiler heat duty is used to control the bottom stream purity. Figure 7.14 depicts
the proposed control structure and table 7.5 details each control loop. The controllers
are tuned according to the guidelines presented in section 6.2.3.2. Table 6.5 lists all
controllers parameters obtained using optimization and adopting a sequential approach
(i.e. optimization of sizing parameters followed by optimization of control parameters).
The controllers are assumed to be correctly tuned. Alternative control congurations
are proposed for the MTBE synthesis by Wang et al. (2003); Sneesby et al. (1998b)
addressing switch-over to dierent steady states. As the focus of this section is the
dynamic behavior around a known steady state, no steady-state transition is considered
by constraining the disturbances variability.
148

P
Pcond
levelcond
Fcool
Fout,Lcond
1
FR,L FD,L
MTBE
(<0.1%m)
reactive
zone
T Treb
MeOH
9
10
Qreb
C4-stream
15
Fout,Lreb
MTBE
(>99%m)
levelreb
Figure 7.14. Schematic representation of a MTBE RD column with a 44

SISO control structure. Remarks: the coupling between controlled and manipulated variables for each control loop are listed
in table 7.5; numbers represent tray location.
Table 7.5. Control loops in a reactive distillation stage column. The schematic
representation of the unit is given in gure 7.14.
Unit Controlled variable
Manipulated variable
P1
Outgoing owrate in condenser,

out,L
Fcond
Flowrate of cold utility in condenser, F cool
Outgoing owrate in reboiler,
out,L
Freb
Heat duty of hot utility in reboiler, Qreb
P2
Liquid level in condenser,

levelcond
Pressure in condenser, Pcond
P3
Liquid level in reboiler, levelreb
P4
Temperature in reboiler, Treb
149
Chapter 7
in,cool
[K]
294
F MeOH [mol/s]
292
186
184
290
182
180
288
178
176
286
174
172
284
170
168
282
0
20000
40000
60000
80000
100000
166
0
10000
Time [s]
20000
30000
40000
50000
60000
Time [s]
(a) Disturbance in the inlet temperature

of the cold utility, T in,cool , as dened in
equation 7.1. Time horizon=86400 s.
(b) Disturbance in the feed owrate of

methanol, FMeOH , as dened in equation
7.2. Time horizon=60000 s.
Figure 7.15. Disturbance scenarios considered for the analysis of the dynamic
behavior of a MTBE RD column. Remark: column specications are listed in table 7.4.
7.3.2.3
Disturbance Scenarios
In addition to the disturbance scenarios already dened in section 6.2, two disturbances
are considered for this behavioral analysis (gure 7.15),
a sinusoidal variation of coolant inlet temperature during 24 h (=86400 s) and
dened as,
T in,cool = 298 + s (t),
with s (t) := 5 sin(
2
t).
8.64 104
(7.1)
a step variation in the owrate of methanol during the time horizon of 6104 s and
dened as,
F SS
if t < 1 104 s or t > 4 104 s,
MeOH
(7.2)
FMeOH =
1.1 F SS
if 4 104 s > t > 1 104 s.
MeOH
It is relevant to mention that the disturbance scenarios are dened in such a way
that multiplicities are avoided. Since a detailed MSS analysis in RD has been already
addressed in section 7.2.1, this section focuses exclusively on the dynamic behavior of
the RD unit around an already known steady state. Thus, the control structure is
dened so that no steady-state transitions are occurring. The eect of large methanol
feed disturbances (FMeOH > 0.1) on the operational steady state has been covered by
Schrans et al. (1996); Sneesby et al. (1998a,b).
150
x MTBE [-]
Temperature [K]
1.0
430
420
0.8
410
400
0.6
390
380
0.4
this work
370
Wang et al. (2003)

0.2
this work
360
Wang et al. (2003)
350
0.0
340
10
12
14
Tray location [-]
(a) MTBE composition prole.
16
10
12
14
16
Tray location [-]
(b) Temperature prole.
Figure 7.16. Comparison between steady-state proles obtained in this work

and by Wang et al. (2003). Remarks: trays are numbered
from top (condenser: 1st ) to bottom (reboiler: 17th ); MTBE
product stream is withdrawn from the bottom location; column
specications are listed in table 7.4.
7.3.3 Dynamic Response in the Presence of Disturbances

A preliminary analysis of the RD model is performed in the absence of any disturbance
scenario. For the sake of model validation, the proles of MTBE molar fraction and
stage temperature are compared with those reported by Wang et al. (2003).
As noticed from gure 7.16 a relatively good agreement exists between both simulations.
In the temperature prole, however, a slight dierence is reported in the rectifying and
reactive sections of the unit. The origin of this mismatch could be attributed to the
dierent inert component considered in both cases. Wang et al. (2003) assume that the
C4 -stream contains n-butene as inert component, whereas in the course of this thesis
work inert n-butane is used. The dierent physical properties of these components
might be leading to slightly altered energy balances and subsequently to dierent tray
temperature. As the temperature mismatch starts to play a role on the reactive trays,
it could be suggested that dierences exist between the estimations of reaction heat.
Moreover, the kinetic and thermodynamic models used in both cases do not fully coincide. On the other hand, the degree of agreement between MTBE molar fractions
obtained by Wang et al. (2003) and this work is quite satisfactory.
At this point the dynamic behavior of the RD unit is analyzed with respect to disturbance rejection. Figure 7.17 depicts the evolution of MTBE bottoms composition
in presence of the deterministic disturbances. Oscillations around the set-point value
are noticed, but of extremely small magnitude , specially for the disturbance in the
151
Chapter 7
x MTBE [-]
x MTBE [-]
1.000
0.99780
0.99778
0.999
0.99776
0.99774
0.998
0.99772
0.99770
0.997
0.99768
0.99766
0.996
0.99764
0.99762
0.995
0.99760
0.08
0.06
0.04
0.02
0.00
0.08
0.06
0.04
0.02
0.00
0
20000
40000
60000
80000
100000
(a) T in,cool disturbance scenario. Time

horizon: 8.64104 s.
10000
20000
30000
40000
50000
60000
Time [s]
Time [s]
(b) FMeOH disturbance scenario. Time

horizon: 6104 s.
Figure 7.17. Time variation of MTBE product stream in the presence of deterministic disturbance scenarios. Remarks: MTBE product
stream is withdrawn from the bottom location; the column specications are listed in table 7.4.
coolant inlet temperature,

|| < 0.002% for the T in,cool disturbance
|| < 0.2% for the FMeOH disturbance
After the disturbances the RD column returns to its initial state, without any steadystate transition. From these results it can be suggested that the unit operates coping
with product specication in spite of the presence of the dened disturbances. Obviously, more aggressive disturbance scenarios might lead to unit instability or even
steady-state transitions. Schrans et al. (1996); Sneesby et al. (1998a,b) show that instabilities are expected to occur at slightly higher methanol ow disturbances.
7.4
Concluding Remarks
The conclusions derived from this chapter are grouped in accordance to the presented
sections.
Steady-state behavior.
In this rst section, a reactive ash with a rather ideal
system (mildly exothermic isomerization reaction and light-boiling reactant) was considered. This simplied system was selected to account exclusively for the eect of
phenomena interaction on the occurrence of input and output multiplicity. Therefore,
any eect related to unit conguration is considered outside the scope of the analysis.
152

For this particular case, the number and relative location of the turning points and
feasibility boundaries corresponding to no-liquid or no-vapor products determined the
qualitative nature of the bifurcation diagram. State multiplicity was present in a large
area of the parameter space, for the case under consideration. A large number (25)
of qualitatively dierent bifurcation diagrams could be identied for the system. The
feasibility boundaries played an important role in the classication of the reactive ash.
Up to ve feasibility boundaries and a maximum of 3 steady states were found in a
single bifurcation diagram. Bifurcation diagrams exhibiting more then two feasibility
boundaries are common for liquid or vapor feed. Increasing the heat of reaction, activation energy or relative volatility enlarges the area of multiplicity. Furthermore, a
large heat of reaction can create regions in parameter space with 5 feasibility boundaries in a single bifurcation diagram. The singularity theory approach could be useful
to investigate the main behavior patterns in more complex reactive distillation systems,
including multicomponent, multireaction and nonideal systems. Moreover, the eect of
mass transfer would be worthwhile to examine, having as initial step the contributions
of Baur et al. (2001c); Higler et al. (1999b).
Dynamic behavior.
The second part of this chapter deals with the dynamic behavior of a full RD column, with special focus on the synthesis of MTBE. The content
of this rather classical control approach (SISO and PI controller) sets the stage for a
next novel contribution to be introduced in chapter 8, involving the trade-o between
design, responsiveness and sustainability aspects. A generic dynamic model was developed for a complete RD column. The model was composed of instances of (non-)
reactive trays, condenser, reux drum, reboiler and PI controllers. Based on a rigorous
degree of freedom analysis and engineering judgement a control structure was dened.
The synthesis of MTBE was used as tutorial example for this section. The sizing of
the unit corresponded to the optimized values obtained in previous chapters (cf. chapter 6). A set of deterministic disturbances were included, accounting for a sinusoidal
variation of coolant inlet temperature and a methanol feed owrate pulse. Moreover,
the disturbances were dened in such a way that steady-state transitions were avoided.
The controllers were tuned based on the optimization of the integral absolute error
(cf. section 6.2.3.2). To rst validate the model, a comparison was made between the
proles obtained in this work and those reported in literature. A satisfactory degree of
agreement was noticed. When the unit was dynamically simulated in the presence of
the disturbances scenarios, minor variations in the MTBE bottom fraction were found.
As the magnitude of these set-point deviations was very small, it is suggested that the
unit operates under product specications in the presence of the dened disturbance
scenarios. Thus, the outcome of this section serves as a reference case to be compared
with that of the life-span inspired design methodology (cf. section 1.4).
153
Chapter 7
The steady-state and dynamic behavior in RD are addressed in
this chapter. Fundamental understanding of multiplicity in a reactive ash is provided. Moreover, the domain knowledge is extended by dynamic analysis, whose tasks include model development, control structure selection, controller tuning and simulation
of closed loop dynamics. The output of this section serves as reference case to be compared with that of the life-span inspired
design methodology (cf. section 1.4). The material presented in
this chapter answers partially Question 6.
154
It is not the strongest of the species that survive, nor the

most intelligent, but the one most responsive to change.
Charles Darwin, naturist (1809-1882)
A Design Approach Based on Irreversibility
The interactions between economic performance, thermodynamic eciency and responsiveness in RD process design are explored in this chapter. This motivation is derived from taking a sustainable life-span perspective, where economics and avoidance
of potential losses of resources (i.e. mass, energy and exergy) in process operation
over the process life span are taken into consideration. The approach to the column
design involves dening a building block, generic lumped RD volume element and
the development of rigorous dynamic models for the behavior of the element. As an
extension of conventional existing models, this model, which is used as a standard
building block, includes entropy and exergy computations. Three objective functions
are formulated which account for the process performance regarding economy, exergy
loss and responsiveness. A fundamental understanding of the strengths and shortcomings of this approach is developed by addressing a case study: (i) steady- state
simulation of a classical MTBE design based on economics only, useful as a reference case; the multiobjective optimization of a MTBE RD column with respect to
(ii) economics and thermodynamic eciency; and (iii) economics, thermodynamic
eciency and responsiveness. Finally, structural dierences and dynamic response
are identied between the optimized and classical designs, to stress the importance
of considering exergy and responsiveness criteria.
Parts
of this chapter have appeared in Almeida-Rivera et al. (2004b)
155
Chapter 8
8.1
Introduction
Advances in computing and information technology allow chemical engineers to solve

complex design problems arising from a need to improve sustainability of the biosphere
and human society. In such a context the performance of a chemical plant needs to
be optimized over its manufacturing life span, while accounting for the use of multiple
resources in the design and manufacturing stages. Since such resources are of a dierent
nature (e.g. capital, raw materials and labor) with dierent degrees of depletion and
rates of replenishment the performance of a chemical plant is characterized by multiple
objectives with trade-os. An integrated unit operation like RD can oer signicant
benets for resource utilization, when it is accompanied with the avoidance of chemical
equilibrium limitations, enhancement of conversion, selectivity and yield and the reduction of both operational and capital costs. The nonlinear coupling of chemical reaction,
phase equilibria and transport phenomena can however, leads to undesired, systemdependent features, such as multiple steady states and reactive azeotropes (Malone and
Doherty, 2000). The results of a case study on the interactions between economics,
exergy eciency and process responsiveness in the conceptual design of RD processes
are discussed in this chapter.
In the wider context of conceptual process design, sustainability issues like feedstock
selection, (re-)use of catalyst and (re-)use of solvents are of key importance. In our
perspective of life span (cf. explanatory caveat in section 1.4), we do not cover those
aspects as we focus exclusively on the minimization of loss of resources in the operational
phase.
Currently, conceptual design in chemical engineering is predominantly driven by (steady
state) economic considerations, postponing to latter engineering stages any issue related
to exergy analysis, responsiveness and control performance. The importance of incorporating exergy eciency in the early stages of the design cycle is stressed by the increasing
depletion of nonrenewable energy resources and the relatively low second law eciencies
(5-20%) for the majority of chemical processes. In spite of these facts, exergy analysis is
exclusively performed as a follow-up task to the process synthesis (Koeijer et al., 2002).
Controllability and process responsiveness are important to maintain product quality
and to avoid excessive use of utilities for control actions and reprocessing of o-spec
material.
Regarding the multiechelon approach in RD design (cf. chapter 4) in this chapter the
last design space is addressed. The evaluation of the design obtained so far is checked
with respect to economic performance, exergy eciency and responsiveness. The focus
of this chapter is the importance of including interactions between economics, exergy
eciency and process responsiveness in RD conceptual design. Aiming at these goals,
a generic compartment model is dened, which accounts for mass, energy and momen156
Table 8.1. Input-output information for the thermodynamic-based evaluation

space. The multiechelon approach is explained in section 4.5.
Design specications
Process:
- spatial conguration
- control structure and settings
- operational parameters
- disturbance scenario
Continuous:
- optimal design variables
- performance criteria
- non-equilibrium
thermodynamics
- steady-state simulation
- dynamic simulation
- multiobjective optimization
tum balances and includes entropy and exergy computations. This behavioral model
is integrated to give a column structure. The design is cast into a nonlinear optimization problem, where degrees of freedom in the design model are xed by optimization
of multiple objectives and applying constraints, reecting heuristics and engineering
knowledge. At this column level three performance criteria are dened for the optimization of the unit design. These embrace the unit performance regarding economics,
thermodynamic eciency and process responsiveness. The economic performance index is given by the total annualized cost (TAC), whereas the thermodynamic eciency
index accounts for the process irreversibilities and is expressed in terms of the entropy
production rate by chemical reaction, heat and mass transfer. The responsiveness index, in turn, is similar to the response time dened by Luyben et al. (1999). The
three objective functions are solved simultaneously within a single multiobjective optimization problem. At this level of understanding, it is believed that a multiobjective
optimization gives more insight into trade-os between objectives than combining the
three objectives into a single monetary objective function using ad-hoc weighting factors. Comparing the design based on the three-fold performance criteria with a reference
economic-driven design allows us to identify structural dierences. The response of the
optimized design in the presence of a deterministic disturbance scenario is validated
using dynamic simulation.
The input-output ow of information is given in table 8.1. It is worth mentioning that
the thrust of this chapter is the optimization-based design approach. Multiple objective
functions are considered and embedded in a large scale optimization problem.
Are there structural dierences and signicant improvements in designs derived
using conventional methodologies from those obtained using an integrated design
methodology?
157
Chapter 8
8.2
Generic Lumped Reactive Distillation Volume Element
The starting point for a fundamental understanding of the strengths and shortcomings of the proposed multiobjective design approach is to dene a generic lumped RD
volume element (GLRDVE) . This element, or compartment, was chosen as building
block in the design approach (cf. section 6.1.3) and selected to represent the governing phenomena occurring in RD processing. Thus, this behavioral model accounts for:
(i) mass, energy and momentum balances, together with entropy and exergy computations; (ii) thermodynamic equations of state for the specic heat and specic density
and for the chemical potential per species; (iii) thermodynamic phase equilibrium conditions at the contact surface between phases; (iv ) the constitutive rate equations for
chemical reactions; (v ) the rate equations for interphase transport of species mass and
energy; (vi) the rate equation for heat transfer with an external solid surface; (vii) the
hydrodynamics-based rate equations for convective mass and energy ows with the external world and the pressure drop; and (viii) the equations for the interphase contact
surfaces and for the void fraction of the vapor phase in the liquid phase.
The following attributes are introduced: (i) two internally (well mixed) homogeneous
uid phases; and (ii) chemical reactions in the liquid phase only. Thus, mass transfer between phases and heat exchange are allowed in the compartment and chemical reaction
is triggered by the presence of an appropriate catalyst (gure 8.1). Each GLRDVE has
its own complexities due to the many interacting phenomena. However, this does not
account for the interactions between the stages in a real multistage column. The latter
class of interactions could give rise to nonlinear dynamic behavior and poor controllability if an unfortunate combination of reux ratio and number of (non) reactive trays
is chosen. Issues related to connecting compartments and superstructure description
are addressed in section 8.3.
8.2.1 Mass/Energy/Entropy/Momentum balances

Component, energy and entropy balances are developed in the GLRDVE. The entropy
and, in particular, the entropy production rate are of paramount importance to determine exergy losses and the responsiveness of the process. The balances lead to the
following set of DAEs,
158
F nV
L
F n-1
F nfeed,L
F nside,V
Rx
Q
F nside,L
F nfeed,V
V
F n+1
F nL
Figure 8.1. Schematic representation of the generic lumped reactive distillation volume element (GLRDVE). Remark: feed and side drawo streams and heat exchange are allowed. Nomenclature:
feed,()
: feed owrate of phase (); F side,() : owrate of side
Fi
stream corresponding to phase (); m : interfacial mass transfer
ux; e : interfacial energy transfer ux; Q: heat duty.

()
i V ()
t
j=n
st
j,()
Fi
i Znc , (8.1)
j=1
U()
t
j=n
st i=n
c
j=1
j,()
Fi
j,()
hi
+ e a +
i=1
j=n
st i=n
c
j=1
()
L
i a + () ri VRX
,
j,()
Fi
j,()
gi
+ () Q() ,
(8.2)
i=1
j,()
where ci is the molar density of component i (i Znc ) in phase () Znp , Fi

is the convective ow of component i in phase () in the incoming or outgoing stream
j (j Znst ) for the exchange with the external world, i is the interfacial molar ux
L
of component i, a is the mass and heat transfer area, VRX
is the reaction volume in
the liquid phase, rj is the net cumulative rate of reaction of specie j per unit, U is
j,()
the internal energy of the compartment, hi
is the molar enthalpy of component i in
j,()
is the external force exerted on component i per molar
stream j and phase (), gi
unit (including the contribution from electrical eld) in stream j and phase (), e is
the energy ux associated with the interface mass and heat transfer, Q() is the external
heat exchange rate in phase (), which is assumed to only occur in the liquid phase
(i.e. Q() = QL ), nc is the number of components, nst is the number of incoming and
159
Chapter 8
outgoing streams, np is the number of phases and () is the Kronecker delta function
dened as,
0, if () = L
(8.3)
() :=
1, if () = L.
The energy ux associated with the interface mass and heat transfer e is composed
of a measurable heat ux and the partial molar enthalpies brought by the component
uxes. namely,
i=n
c

()
i hi + q ( ) ,
(8.4)
e =
i=1
where q ( ) is the heat ux from phase ( ) to phase ().

The entropy balance in the transient state at a macroscopic level is given by the net
amount of entropy transferred into the volume element per unit time and the amount of
entropy used/produced in the control volume per unit time. Thus, the entropy balance
for a given phase () can be stated as,
i=n
st j=n
c i,()
S()
i,()
()
=
Fj
sj
+ V ,
t
i=1 j=1
i,()
where S() is the entropy of the compartment in phase (), Sj
(8.5)
is the molar entropy
()
V
of component j in stream i and phase () and

is the entropy production rate
in phase (), which includes exclusively interfacial diusional and chemical reaction
contributions.
Alternatively, the entropy balance can be expressed in terms of the entropy density and
the uxes of entropy density (Kjelstrup and Bedeaux, 2001; Marquardt, 2004),
s()
()
= Js() + V ,
t
(8.6)
()
()
where V is the entropy production rate per unit of volume in phase () and Js
the entropy ux in phase ().
is
i,()
The ux of entropy of stream i in phase () is related to the total ux of energy Je

i,()
and mass uxes Jm according to the expression (Kjelstrup and Bedeaux, 2001),
Jsi,() =
i,()
where j
().
160
j=n
c i,()
1
j
i,()
Jei,()
Jm,j ,
T
T
j=1
i Znst ,
(8.7)
denotes chemical potential of component j (j Znc ) in stream i and phase

The uxes of mass and energy required in expression 8.7 can be obtained from the
equation of continuity per volume unit and energy balance, respectively, for an open
unsteady state multicomponent system. Let () be a given phase, then the fundamental equation of continuity is given by convective, diusive and chemical reaction
contributions,
()
()
()
()
= (i v() ) (ji ) +ri ,
t i

()
({i
()
v() +ji
i Znc ,
(8.8)
()
})Jm,i
and
()
()
()
Jm,i = i v() + ji ,
i Znc ,
(8.9)
()
where v() is the uid velocity in phase () and the molar diusion ux vector ji
involves terms originated by mechanical and thermal driving forces (Bird et al., 2003).
()
k,()
It is worth mentioning that the ji vector is composed of the elements ji
with
k Znst .
i,()
()
(Je in vectorial notation) can be obtained by the

Similarly, the energy ux Je
energy equation derived for the volume element in an open multicomponent dynamic
system,
i=n
c () ()
()
u = ( u() v() ) (q() ) ( () : v() ) +
Jm,i gi + E j, (8.10)

t
i=1
()
({u() v() +q+[ () v() ]})Je
where u is the internal energy density of the volume element, () is the pressure
tensor in phase (), which accounts for the increase of the rate of internal energy by
reversible pressure forces and by irreversible viscous forces (Bird et al., 2003), q() is
the multicomponent energy ux through the element interphase relative to the mass
average velocity in phase (), E is the electric eld and j the electric current density.
If it is assumed that the energy ux q() includes mainly conductive and interdiusive
energy uxes (i.e. q() = qconductive + qdiusive ) and that the medium is isotropic, the
()
total energy ux Je with respect to stationary coordinates is given, after eliminating
negligible terms, by,
i=n
c () ()
()
J()
=
J
+
Ji hi ,
(8.11)
e
q
i=1
where Jq
()
is the measurable heat ux of phase ().

()
The entropy production rate at macroscopic level, V (equation 8.5) can be obtained
after integration of the local entropy production rate density () over the volume element in phase (). For the sake of simplication the following assumptions are adopted:
(i) entropy contributions of all involving phases in the GLRDVE are combined in a sin()
()
gle variable (V = f (V )); (ii) entropy is produced due to mass and heat diusion in
161
Chapter 8
the liquid-vapor interphase, chemical reaction in the bulk of the liquid and heat transfer
from external heat sources/sinks and the surrounding environment to the liquid phase
only; and (iii) the eects of electric eld, external forces (g = 0) and diusive momentum uxes are neglected. The following expression for the integral entropy production
term can be then derived (Seider et al., 1999; Bedeaux and Kjelstrup, 2004; Kjelstrup
and Bedeaux, 2001),

i=n
1 rx L
()
L
L
() dV =V
=
A L
(8.12)
i VRX + Jq a Xq +
T L i=1 i
V
(
' i=n
i=n
c
c
+
Ji a hi Xe +
Ji a Xm,i +
i=1
i=1
+ Q X
HX
nrx
) in the liquid phase, L
where AL
i is the anity of chemical reaction i (i Z
i is the
L
degree of advancement of chemical reaction i in the liquid phase, Jq is the measurable
heat ux into the liquid phase, a is the mass/heat transfer area, hi is the molar
enthalpy of component i, Xq , Xe , Xm,j and X HX are the conjugated driving forces for
heat exchange between phases, interfacial heat diusion, interfacial mass diusion of
component j and heat exchange with the external world, respectively (Koeijer et al.,
2002), Q is the heat ow to the liquid phase, T is the absolute temperature of the
volume element and nrx is the number of chemical reactions.
The entropy contribution due to chemical reaction contains the anity of chemical
reaction and its degree of advancement, which are dened respectively as,
AL
j
i=n
c
L
i,j
L
i ,
j Znrx ,
(8.13)
i=1
riL
j=n
rx
L
ij
L
j,
i Znc ,
(8.14)
j=1
where riL is the net change of the amount of component i in the liquid phase due to all
reactions.
In order to estimate the driving forces and uxes occurring in the volume element
additional assumptions are introduced: (i) the GLRDVE is composed of incoming and
outgoing streams of liquid and vapor phases; (ii) the vapor phase behaves ideally; (iii)
the ux of mass and heat through the interface is positive from the liquid side to the
gas side; (iv ) dynamic changes in gas phase hold-ups of species mass and energy are
ignored; (v ) Gibbs free energies at liquid and vapor sides are equal; and (vi) there is a
L 0
negligible eect of pressure on Gibbs free energies, so that GL
i (P, T ) = Gi (pi , T ).
The underlying reasoning behind the inclusion of these assumptions is a more transparent derivation of the forces and uxes expressions, but without compromising the
162

validity of the approach.
The driving force for chemical reaction j, Xrx,j , is a function of the chemical anity
and temperature, according to the expression,
Xrx,j =
AL
j
T
(8.15)
The ow associated to this driving force is given by the degree of advancement i .

The driving force for mass transfer Xm,i is related to the chemical potential gradient
between the incoming and outgoing vapor streams (Koeijer, 2002; Kjelstrup, 2005).
Thus, according to gure 8.1 the conjugated driving force Xm,i is,
Xm,i
1
= i,T ,
T
y in P in
,
= R ln i
yi P
i Znc ,
(8.16)
i Znc .
(8.17)
The ux of component i (i Znc ) associated to Xm,i through the interface can be

calculated by mass balance over the vapor phase (Koeijer, 2002),
Ji a =
j=n
st
F j,V yij ,
i Znc ,
(8.18)
j=1
where yij denotes the vapor molar fraction of component i (i Znc ) in stream j (j
Znst ) and F j,V is the total ow of stream j in the vapor phase and equals to F j,V =
i=nc j,V
.
i=1 Fi
Note that the previous ux expression is consistent with the component balance given
by expression 8.1 if i equals Ji . Moreover, the total stream ow is related to the
component ow according to,
F j,V yij = Fij,V ,
i Znc .
(8.19)
The driving force for heat transfer, Xe , can be estimated as suggested by Koeijer and
Rivero (2003),

1
1
1
1
Xe =
in,V .
=
(8.20)
TV
2 T out,V
T
The measurable heat ux through the interphase can be derived from the energy balance
in the vapor side (Koeijer, 2002),
Je a =
j=n
st
F j,V hj,V .
(8.21)
j=1
In this case also the energy expression 8.21 is consistent with the general energy balance
given in expression 8.2 if e equals Je .
163
Chapter 8
According to Koeijer et al. (2002); Koeijer (2002) the average thermal driving force for
external heat transfer X HX (e.g. when a heat exchanger is present) is given by the
expression,
2
1
X HX = u,in
,
(8.22)
T
T
+ T u,out
where T u denotes temperature of the utility present in the GLRDVE.
The momentum balance in the GLRDVE involves, in its general form, contributions
due to convection of the uid bulk, pressure, viscous transfer and gravitational force in
each phase (Bird et al., 2003),
() v()
= [ () v() v() ] P () [ ] + c() g() ,
(8.23)
t
i=n
where = i=1 c i , [ v() v() ] is the rate of momentum gain by convection of
the uid per unit volume and is the rate of momentum gain by viscous transfer
per unit volume.
If viscous eects are considered unimportant the following simplied expression might
be obtained,
() v()
= [ () v() v() ] P () + () g() .
t
(8.24)
In the development of the GLRDVE a reduced form of the momentum balance is considered. We assume a pseudo steady state and negligible viscous contribution. Moreover,
the gradient of pressure is assumed to be much larger than the rate of momentum gain
by convention of the uid. Thus, the resulting momentum balance is of the form,
P () = () g() .
(8.25)
8.2.2 Availability and Availability Function

Two additional variable of key importance for this thermodynamic-based approach are
the availability A and the availability function B. The availability is regarded as a
measure of the maximum amount of useful energy that can be extracted when the
system is brought to equilibrium with the dead state surroundings (Seider et al., 1999).
Thus, the availability of the system in phase () and state (T1 , P1 ) is given by the
expression,
A() = (h() T0 S () )T1 ,P1 (h() T0 S () )T0 ,P0 ,
(8.26)
where the dead state conditions are those at which T0 = 298.15K and P0 = 1 atm.
The availability function at state given by (T, P ) is computed according to the following
equality,
()
(8.27)
BT,P = (h() T0 S () )T,P .
164

Contrary to the availability the availability function of the system at a given state
(T1 , P1 ) can be referenced to any other state (T2 , P2 ),
B() = (h() T0 S () )T1 ,P1 (h() T0 S () )T2 ,P2
(8.28)
From the denitions of availability and availability function it should be noticed that
both variables increase with increasing enthalpy and decrease with increasing entropy
and the numerical values of their variation coincide (i.e.B() = A() ).
Based on the combination of energy and entropy balances and according to the denition
of availability an exergy balance can be obtained for a given phase () (appendix A in
Luyben et al. (1999)),
B() M ()
t
j=n
st
j=1

T0
F j,() Bj,() + 1
Q() T0 V , (8.29)
TQ
where M () is the molar hold-up of material in phase () and TQ is the temperature of

the heat source/sink. Note that in the previous expression the external force term has
been omitted.
To solve the set of balances mentioned in this section it is necessary to specify the initial
conditions for all DAEs. Thus, the following variables per phase should be known,
()
[i (0),
U () (0),
S () (0),
() (0) v() (0),
B() (0) M () (0)].
(8.30)
A degree of freedom analysis should be then performed to indicate how many design
decision variables need to be picked per volume element. Appendix A gives a detailed
overview of a degree of freedom analysis for the case of a complete RD column.
8.2.3 Design Decision Variables

According to a degree of freedom analysis performed in the GLRDVE the following
design variables are to be specied for a given operating conditions and feed specications,
()
reaction volume VRX ,
V ()
()
VRX =
mcat 1
m,cat
for liquid phase homogeneously catalyzed reaction(s)

for heterogeneously catalyzed reactions(s)
external and heating/cooling owrate Q,

the interfacial area for heat and mass transfer a .
165
Chapter 8
Note that operational variables such as reux ratio and pressure are xed at appropriate
j,()
are assumed
reference value and the connecting streams and associated owrates Fi
to be given.
The perspective adopted for selecting the design decision variables is task-oriented
rather than the conventionally used equipment-oriented design approach. Firstly, the
task-based concept would allow us to derive the phase volume V () from the reaction
time (i.e. residence time in element) to meet a specied degree of advancement of
reaction ; secondly, the amount of heat to be added or removed allows one to compute the heat exchange surface and coolant ows; and thirdly, the interfacial areas for
species mass and heat transfer may be translated into the mechanical design of a tray
and its prevailing hydrodynamic regime such that the required areas can be met within
constraints.
The main benet of this functional, task-oriented perspective involves the process design unconstrained by equipment geometry. Thus, task-oriented design prevails against
equipment-oriented design.
8.2.4 Economics/Exergy Eciency and Responsiveness Objective Functions

A set of three performance criteria indices are dened for the GLRDVE: the economic
performance index fecon , the sustainability performance index fexergy and the responsiveness performance index fresponse .
The economic performance indicator fecon represents the total annualized cost TAC of
the GLRDVE, including capital investment and operational costs of the process,
fecon TAC =
j=n
st i=n
c
j=1
j,()
Fi
ci + Fcu ccu + Fhu chu + cunit + catalyst ,
Znp , (8.31)
i=1
where cj is the cost of component j or utility per molar unit, cunit+catalyst is the investment cost of the unit and catalyst and cu and hu denote cool and hot utilities
respectively.
The exergy eciency, fexergy , is accounting for the process irreversibilities and is expressed in terms of the 2nd law thermodynamic eciency, II ,
fexergy II =
Wideal
,
Wideal T0 V
(8.32)
where Wideal is the minimum work required by the process (Wideal < 0), as given in
Seider et al. (1999). Note that from a sustainable perspective the desired goal is the
From
a more practical perspective it is easier to vary residence time than work with several extents
of reactions in parallel.
166

minimization of fexergy, which might be explicitly translated into the minimization of
V . In this contribution we adopt this approach and compute T0 V .
The responsiveness index fresponse is approximated to the settling time dened by Luyben et al. (1999),
B(x)
,
(8.33)
fresponse = max
x T0 V (x)
where T0 represents the temperature of a innitely large sink, which approximates to
the surrounding conditions.
At steady state conditions B(x) is at its minimum value and increases when the system
is perturbed by a disturbance. Since a similar situation holds for V (x), the response
time measures how fast a small variation B(x) would dissipate due to an increased
V (x) (Luyben et al., 1999).
In the previous denition of fresponse Luyben et al. (1999) assume that both the availability and entropy production rate are exclusively functions of the vector of state
variables x. Although this assumption is fundamentally correct, it somehow disguises
the way of estimating the eect of a small disturbance or a change in an input on the
system. Therefore, we propose a subtle modication of the arguments of B and V ,
by which a second argument is included, representing the systemic view on a dynamic
system. Thus, the availability and entropy production rate are regarded as functions
of an internal state vector x and a vector of disturbances and/or manipulated external
excitations u,
exergy : B(x, u),
entropy production rate : V (x, u).
(8.34)
(8.35)
Combining denitions 8.33 and 8.34 leads to a more generalized expression for fresponse,
fresponse =
B(x, u)
.
T0 V (x, u)
(8.36)
The previous expression could be further elaborated by dening the values of state
variables x and disturbances u at each state. Hence,
B(x, u)
V (x, u)
= B(x, u) B(x0 , u0 ),
(8.37)
= V (x, u) V (x0 , u0 ),
(8.38)
u =
u0 + u,
(8.39)
x =
x(u),
(8.40)
V (x0 , u0 ).
(8.41)
with
V (x, u)
167
Chapter 8
An additional feature of Luybens original denition of fresponse , not shown in expression
8.36, is related to the maximization for the time constant over all possible steady states
surrounding the reference steady state. In this way one looks for the worst case over
all possible small scale disturbances and/or input change scenarios. This optimizationdriven approach justies Luybens approach of including a single argument for B and
V .
In our approach such an maximization approach is not considered due to its inherent
high computational eort. However, a rather formal denition of fresponse is adopted
(equation 8.36) together with a single input excitation. Note that by not searching all
possible state trajectories, the generality of the results might be questionable and the
external excitation is a fairly conservative one. However, engineering judgement can be
exercised in selecting the disturbance variable u and the magnitude of the disturbance
u.
8.3
Integration of Volume Elements to a Column Structure
Having addressed the generalities of the GLRDVE the next step is the integration of
the volume elements to a column structure. This is addressed in this section together
with the re-denition of the performance criteria for the case of a RD column structure.
Additionally, fundamental understanding of the eect of design decision variables on
process responsiveness is discussed, using a simplied heat exchanger as the case study.
Finally, this section includes the optimization approach to tackle the multiobjective
problem.
8.3.1 Creating a Column Structure from the Generic Volume Element

The integration of compartments to give a column structure introduces additional degrees of freedom, which should be conveniently specied (e.g. reux ratio). In the more
general context of process synthesis, the issue of coupling compartments to a process
unit has been frequently addressed and is not, therefore, explicitly considered in this
research. Briey speaking, the compartment integration is carried out by introducing
a superstructure and nding the optimal conguration using a MINLP optimization
over the superstructure. The description of the superstructure, the rules for connecting
compartments with each other and with the outer world are embedded in the structure
model . Extensive information on structure models can be found in the research work
of e.g. Ismail et al. (2001); Papalexandri and Pistikopoulos (1996). In our particular
case, the compartments are integrated to a column structure in such a way that the
superstructure is comparable to the unit specied in table 6.2 of section 6.1.
The rules of connectivity used in the RD superstructure state that: (i) the outgoing
168

liquid at a given compartment enters the compartment located below; (ii) the outgoing
gas at a given compartment enters the compartment located above; (iii) the liquid
stream leaving the condenser is partially recycled back to the top compartment and
withdrawn from the unit as top product; (iv ) the liquid leaving the bottom compartment
enters the reboiler; (v ) the streams leaving the reboiler are a vapor stream, which enters
the bottom compartment and a liquid stream, which is withdrawn from the unit as a
bottom product; and (vi) side feed streams and side withdrawal streams may occur.
The connectivity rules, however, imposed some conguration limitations: (i) heat integration over volume elements is not considered (i.e. carrying heat from high temperature to low temperature elements); (ii) side reactors are not present; and (iii) no
parallel structures are considered.
Note that the compartments representing the behavior of the condenser and reboiler
are, in principle, identical to the GLRDVE, provided the following characteristics,
Condenser drum: no chemical reaction takes place de to the absence of catalyst;
a vapor phase stream enters the compartment and a liquid stream leaves it; heat
exchange takes place in both phases.
Reboiler: no chemical reaction takes place due to the absence of a catalyst; a
liquid stream enters the compartment and liquid and vapor streams leave it; heat
exchange takes place in both phases.
In addition to the specication of the reux ratio, the compartment integration to superstructure involves the following features of the structure model: (i) all compartments
can have a dierent catalyst load, but as starting point the catalyst load distribution
suggested by Hauan et al. (1995a); Jacobs and Krishna (1993); Seader and Henley
(1998) is used; (ii) the column sections can have dierent diameters, but initially a
common value is chosen (cf. table 6.2 of section 6.1). The structure to be further
considered during the course of this dissertation is depicted schematically in gure 8.2.
8.3.2 Development of Three Criteria for Optimization of Column Design

Having dened the superstructure to be considered for further analysis the optimization
criteria have to be adapted to the new structure. Stepping out from the GLRDVE case
to the integrated RD unit is intuitive for fecon and fexergy. The variables B(x, u) and
V (x, u) need to be globally evaluated over all the structure for the responsiveness
criterion fresponse. The resulting performance criteria are given by the following set of
169
Chapter 8
expressions,
fecon
k=n
t j=n
st i=n
c
k=1
j=1
k
k
[Fij,k ci + Fcu
ccu + Fhu
chu + ckunit + catalyst ], (8.42)
i=1
fexergy
(Wideal )total
,
(Wideal )total T0 (V )total
(8.43)
fresponse
(B(x, u))total
,
T0 (V (x, u))total
(8.44)
where Fij,k is the externally connecting molar ow of component i (i Znc ) in stream

j (j Znst ) on compartment k (k Znt ), nt is the number of compartments in the
integrated structure, (V (x, u))total is the total rate of entropy production over all
superstructure and (B(x, u))total is the total availability over all the superstructure.
The validity of an additivity property (i.e. the objective functions per compartment
cant be added up to a corresponding objective function over the entire structure) is
implicitly assumed in these denitions.
8.3.3 Responsiveness Explained Using a Heat Exchanger Example

The implications of process responsiveness for design are not totally revealed by the
denition of fresponse (equation 8.44). In this subsection the aim is to provide further
foundations for the understanding and applicability of the responsiveness criterion and
;;
;;
;;
Fcool
Fout,Lcond
1
R,L
FD,L
reactive
zone
F2,L
F1,V
9
10
Qreb
15
Fout,Lreb
FB,L
Figure 8.2. Schematic representation of a RD column as the integration of

GLRDVEs. Remark: numbers represent tray/compartment location.
170

its implications for design. An ideal heat exchanger is considered in this context as the
case study. The starting point of this analysis is directly derived from the denition of
fresponse . According to expression 8.44 a short responsiveness time is desired and can be
attained by a large change in the entropy production rate for a given change in stored
exergy.
8.3.3.1
Systems Features and Model Development
The following properties of the system are considered: (i) two liquid-phase streams
enter the heat exchange in a countercurrent set-up; (ii) the heat exchanger is operated
at isobaric conditions; (iii) both liquid streams are assumed behave ideally, (iv ) both
streams have equal physical properties, excluding heat capacity, which are temperature independent; (v ) heat transfer due to conduction is the only source of entropy
production; and (vi) the driving force for heat transfer is computed between outlet
conditions.
The last assumption is driven by two practical motivations. Firstly, it allows a simplied solution of the problem as no computations over all the units length are required.
Secondly, the situation under consideration is the milder one in terms of entropy produced. At this condition the smallest thermal driving force is expected. If any eect on
fresponse is registered, it may occur also with a magnied intensity at any other location.
Figure 8.3. Schematic representation of an ideal countercurrent heat exchanger to assess the responsiveness index.
According to the schematic representation in gure 8.3, both liquid-phase streams enter
the heat exchanger from opposite sides. Heat is transferred through the solid wall
between the phases. The liquid holdup of each liquid is kept constant during the
operation so that inlet and outlet ows are equal. The set of governing equations and
physical properties of the system are given in tables 8.2 and 8.3, respectively. A degree
of freedom analysis is now performed to estimate the number of variables to be specied
to consolidate the model. All variables of this system are grouped in a column vector
X, which contains 13 elements,
T

Fi Ti hi Q Tlm A .
(8.45)
X Tiin hin
i
171
Chapter 8
The number of governing expressions as listed in table 8.2 equals 8. Out of the ve
remaining variables to be specied, three are considered xed inputs and given by the
vector I,
I
F1
Tiin
T
(8.46)
Within the two remaining degrees of freedom, one is xed by the design target (T1 ),
leading to a single variable to be specied to consolidate the model. This variable could
be either the outlet temperature of the cold utility or the heat transfer area. In the
course of this subsection and to study the eect of varying thermal driving forces on the
responsiveness index, the area A is xed. For the sake of comparison among dierent
units, results for additional heat transfer areas are included.
8.3.3.2
Entropy Production Rate and Stored Exergy
The entropy of the system is assumed to be exclusively generated by the heat conduction from the hot to the cold uids. The entropy production rate, at microscopic level,
can be estimated as the product of thermal driving force and heat ux. From a macroscopic stand-point the measurable heat ow is used for this computation. A better
approximation can be obtained by introducing phenomenological coecients (Hasse,
1969; Koeijer, 2002; Meeuse, 2003). For our analysis, however, we adopt an alternative approach. The overall steady state entropy equation of change is applied and the
production term is related to the net change of entropy.
Therefore, for this ideal system the entropy production rate is given by the overall
contributions of both streams,
Si = Cpi ln
Ti
,
Tiin
i [1, 2],
Table 8.2. Set of governing expressions for the ideal heat exchanger as depicted in gure 8.3. Remarks: stream 1 corresponds to the hot
utility, whereas stream 2 is the cold utility.
Balance
Energy balance
Enthalpy denition
Heat transfer
Governing expression
Q = F1 (hin
1 h1 )
Q = F2 (h2 hin
2 )
in
hin
=
Cp
(T
T0 )
i
i
i
hi = Cpi (Ti T0 )
Q = U HX A Tlm
Temperature dierence Tlm
172
in
T1in T2 (T1 T
2 )
log
in T
T1
2
in
T1 T2
(8.47)
Table 8.3. Properties and operational parameters of the ideal heat exchanger
system. Remarks: most of the operational parameters are taken
from Nummedal (2001); Nummedal and Kjelstrup (2001); the nominal case results in a heat duty of Q=60 kW and a owrate of cold
utility F2 =0.286 kg s1 .
Parameter
Label Value
Flowrate hot utility

F1
Heat capacity hot utility
Cp1
Heat capacity cold utility
Cp2
Inlet temperature hot utility
T1in
Inlet temperature cold utility
T2in
Outlet temperature hot utility (target) T1
Heat transfer area
A
Heat transfer coecient
U HX
1 kgs1
2 kJ(kgK)1
4.2 kJ(kgK)1
400 K
300 K
370 K
3 m2
340 W(m2 K)1
where Cpi is the heat capacity of stream i evaluated in the temperature range [T, T in].
Summing up all the contributions leads to the overall entropy production rate of the
system,
i=2

V =
Fi Si .
(8.48)
i=1
Exergy is similarly estimated by summing up the exergy individual contributions of

both liquids,
Bi = hi T0 Si ,
i [1, 2],
(8.49)
which result in the overall exergy of the system,

B = h T0 V ,
(8.50)
where Bi is the molar exergy of stream i, hi is the change of molar enthalpy of stream
i and equals to hi hin
i , B is the change of exergy of the system.
The stored exergy function is then obtained by accounting for the holdup of liquids
within the boundaries of the system,
B =
i=2

(hi T0 Si ) Mi ,
(8.51)
i=1
where Mi denotes the liquid holdup of utility i. Note that initially the molar hold-up is
not considered as design variable. This assumption is relaxed at the end of this section
to study the eect of M on the responsiveness criterion.
173
Chapter 8
A disturbance scenario is considered to assess the response time function , where the
temperature target T1 changes +/- 10K around the nominal value,
T1l < T1 (t) < T1u ,
(8.52)
where the values of T1l and T1u are 360K and 380K, respectively.
The driving force for heat conduction from the hot to the cold utility X HX is given by
the simplied expression,
1
1
1
X HX = =
.
(8.53)
T
T2
T1
It is worth mentioning again that thermal driving force is computed between outlet
conditions. The motivation behind this assumption is given in section 8.3.3.1.
Response
time [s]
2.8
290
280
Q=60 kW
270
A=3.0 m
260
250
3.2
240
230
0
8 10 12 14 16 18 20 22 24 26 28 30
Thermal driving force [1/K*exp(-6)]
Figure 8.4. Response time as a function of the thermal driving force for an
idealized heat exchanger. Remarks: alternative designs with
dierent heat transfer areas are plotted and compared to the case
where Q=60 103 W (marker: ).
The values of response time as a function of the thermal driving force X HX are
presented in gure 8.4 and for designs of dierent heat transfer area A. Note that from
a responsiveness stand point small driving forces are desired for a given heat exchanger
design and the heat exchanger with the larger transfer area copes more eciently with
the disturbance scenario. Summarizing, better process responsiveness is attained at
smaller driving forces (i.e. larger temperature dierence and larger heat exchangers).
174
Response
time [s]
550
M=200 mol
500
450
400
350
300
M=100 mol
250
200
M=50 mol
150
100
5
0
0
10
12
14
16
18
20
22
Thermal driving force [1/K*exp(-6)]
Figure 8.5. Response time as a function of the thermal driving force for an idealized heat exchanger at dierent hold-up values. Remarks: heat
transfer area: A=3 m2 ; heat duty of reference case: Q=60103 W
(marker: ).
The eect of hold-up M on the responsiveness criterion is now analyzed. As suggested

by equation 8.51, smaller hold-up decreases the exergy storage and so helps to decrease
the time constant. This fact is depicted in gure 8.5, where hold-up values of 100 mol
(reference value), 200 mol and 50 mol are used.
Based on the previous analysis the following practical heuristic can be suggested,
Heuristic: maximize the exchange surface per unit of mass and keep driving force
small at constant transfer duty
8.3.4 Approach to Design Optimization and References Cases

At this point we have dened the superstructure of interest, developed three performance criteria for the optimization of the superstructure and provided some fundamental insight into the impact of design variables on the responsiveness criterion. In
this section the approach to solve the multiobjective optimization problem is presented
together with the reference design case to be compared with the optimized designs.
Two solving approaches are adopted to solve the optimization problem involving more
175
Chapter 8
Utopia point U in a R -space is given by the -objective functions values that solve independently
the scalar optimization problems (Lim et al., 1999). For instance, in an optimization problem
involving two objective functions the mathematical formulation of the utopia point U is,
min f1 (x) = f1l (x ),
and
f2 (x ) = f2u (x ),
(8.54)
min f2 (x) = f2l (x ),
and
f1 (x ) = f1u (x ),
(8.55)
x
x
where fil and fiu denote the lower and upper limits of the objective function fi , respectively;
is the set of constraints, which dene the space of design variables x.
The coordinates of the utopia point are then given by (f1l , f2l ). The meaning and optimization
potentiality of the utopia point is stressed using the following simple example. Lets consider
the following multiobjective optimization problem,
Function:
s.t.:
min(f1 (x1 , x2 ), f2 (x1 , x2 )),
(8.56)
g(x1 , x2 ) = 0,
(8.57)
where
f1 (x1 , x2 )
x1 ,
(8.58)
f2 (x1 , x2 )
x2 ,
(8.59)
g(x1 , x2 )
x1 x22 1 = 0.
(8.60)
The solution of the optimization problem is depicted in a 2D plot of the involved objectives
(gure 8.6). Each non-inferior attainable optimal solution is estimated at a given combination of
objectives including constraint g(x1 , x2 ). All these points dene a curve of non-inferior solutions,
normally referred to as Pareto curve. For the sake of transparency a linear relation between
objectives is considered (i.e. F = wi fi ), where wi is the weighting factor of the objective
function fi and wi = 1. As expected, the utopia point is given by the coordinates
(f1l , f2l ) = (0, 50),
(8.61)
which are the solutions of the scalar optimization problems.
Note 8.1. Utopia point in optimization problems with more than one objective
function
than one objective function,

each objective function is used independently, disregarding the other two. This
approach localizes the coordinates of the utopia point and subsequently, reduces
the computational eort by limiting the search space. For a formal denition of
the utopia point, the reader is referred to the explanatory note in table 8.1,
the second approach accounts for the interactions and trade-os among the objectives. Thus, a multiobjective problem is formulated by using a weighted pnorm
for the vector of objective functions, as given by Clark and Westerberg (1983),
min F
D
176
)

i
* p1
(wi fi )
, p [1, ),
(8.62)

f2 []
fl1
10
Region of
attainable objectives
20
30
fl
Pareto frontier
40
50
Utopia point
60
500
1000
1500
2000
2500
f []
1
Figure 8.6. The role of utopia point in multiobjective optimization. Remarks: the optimization problem is formulated according to expression 8.56 given in the explanatory note of table 8.1; same note
provides the mathematical denition of the utopia point.
where F is the multiobjective function, w is a vector of weighting factors, f is the

vector of objective functions, D is the set of decision variables and constrained by
the set . According to Clark and Westerberg (1983) a norm of p 2 is used to
localize non-inferior points in the non-convex regions.
This theory will now be applied to the multiobjective design cases for the MTBE column. The reference design used for the sake of comparison is the structure previously
introduced in chapter 7. This structure corresponds to a reactive distillation column for
the synthesis of MTBE. The spatial design variables were obtained by steady-state optimization of the units economic performance. For background information, the reader
is referred to section 6.2. The relevant parameters and schematic representation of the
optimized design are given in table 8.4 and gure 8.7, respectively.
177
Chapter 8
8.4
Application 1. Steady-state Entropy Production Profile in a MTBE Reactive Distillation Column
The esterication of iC4 and methanol is used as the case study in this application. A
detailed description of the system is presented in appendix B. The chemical reaction is
represented by a pseudo-homogeneous kinetics. The spatial conguration of the unit is
that obtained previously in section 6.2.
To prepare for the optimization task and to reduce the problem complexity we identify
the dominant contributions to entropy production rate over the MTBE column. The
expressions to compute all contributions to the entropy production in the GLRDVE
are summarized in table 8.5. In the case of a RD column structure these expressions
are applicable for all reactive and nonreactive compartments. Note that the reboiler
and condenser models assume that the entropy contribution by external heat transfer
is dominant.
Table 8.4. Optimized design of a RD column for MTBE synthesis based on
economic performance. Remarks: information on this design
problem is given in section 6.2; bottom section lists the operational constraints and product specications.
Value
Number of trays
Feed 1: location
Feed 1: owrate
Feed 1: composition
Feed 1: temperature
Feed 2: location
Feed 2: owrate
Feed 2: composition
Feed 2: temperature
Operating pressure
Reux ratio
Bottom owrate
Column diameter
Condenser area
Reboiler area
Operational constraints
Product specications
178
15
455 mols1
iC4 : 0.37; nC4 : 0.63
350K
MeOH: 9th tray (l)
168 mols1
MeOH: 1.00
320K
11105 Pa (top)
7
204.1 kg=1000 eq[H + ]
197 mols1
7.132 m
1.059103 m2
6.93102 m2
P Pcond
Treb < Tc,iC4
xD
MTBE 0.1%
xB
MTBE 99%
Fcool
;;;
;;;
;;;
1
Fout,Lcond
FR,L FD,L
MTBE
(<0.1%m)
reactive
zone
MeOH
9
10
Qreb
C4-stream
15
Fout,Lreb
MTBE
(>99%m)
Figure 8.7. Schematic representation of a RD column in the synthesis of

MTBE. Remarks: spatial design variables are listed in table
8.4; numbers represent tray location.
Table 8.5. Summary of expressions of all contributions to the entropy production in a GLRDVE
Expression
mass transfer
i=n
c
Ji a Xm,i
i=1
interface energy transfer

chemical reaction
external heat exchange
JqL a Xq +
1
TL
i=n
rx
i=n
c
Ji a hi Xe
i=1
L
L
AL
i i VRX
i=1
HX
Q X
At this point the entropy proles along the MTBE RD column due to the dierent
contributions are estimated. These proles are presented in gure 8.8 and correspond
to the reference design as specied in table 8.4. The following results are of interest,
the entropy proles show that energy transfer contributes largely to the overall
entropy production and particularly in the stripping section of the column, followed by mass diusion. In the reactive zone, however, the dominant source of
entropy production is chemical reaction. In the rectifying section both mass and
179
Chapter 8
Entropy rate [W/K]

C 4-stream
MeOH stream
10000
Se
8000
Reactive zone
Sm
6000
4000
2000
Srx
0
st
th
10
th
15
th
Tray location [-]
Figure 8.8. Entropy production rate prole for a 15 stage RD column for
MTBE synthesis. Legend: Sm : entropy produced by interfacial mass diusion; Se : entropy produced by interfacial energy
transfer; Srx : entropy produced by chemical reaction; Remarks:
stages are numbered from top to bottom; reboiler and condenser
drum are not depicted; MeOH feed stream is fed at 9th tray and
iC4 stream at 10th tray. The operational and design variables are
energy transfer contributions are equally important,

the entropy contribution originated from chemical reaction shows negative values
at the bottom of the reactive section. This fact may be explained by possible
inconsistency between the reaction kinetics model and the anity expression.
Although MTBE decomposition takes place at that column location, due to high
MTBE concentration, the reaction anity is excepted also to change sign exactly
at the same point in the (P, T, x) space. These facts result always in a nonnegative
entropy production contribution,
at the feed stages (9th -10th ) the conjugated driving forces are considerably larger
than elsewhere due to the dierence of chemical potential and the temperature
between the entering streams and the mixture coexisting on the tray,
external heat transfer plays a dominant role in the reboiler and condenser, exclusively.
180

In terms of entropy production, summarizing, the separation function dominates over
the reaction function for this MTBE RD column conguration.
8.5
Application 2. Bi-Objective Optimization of a MTBE Reactive Distillation Column
A multiobjective optimization problem is formulated for the MTBE RD column with

respect to economic performance and exergy eciency. The formulation includes the
balance equations 8.1-8.5 and 8.29, the criteria denitions 8.42 and the optimization formulation 8.62. Constraints imposed by operating conditions and product specications
are included. For the sake of controlled built-up of optimization complexity the rst
approximation of this approach omits the response time constant as objective function.
The resulting optimization problem is formulated as follows and is implemented and
c
c
/gOPT
,
solved in gPROMS
min
x
s.t.
J(fecon (x), fexergy (x), w),
(8.63)
f () = 0,
h() 0,
where x denotes the set of design variables ([mcat Q a ]) , f () is the set of algebraic expressions that describe the system and h() is the set of inequalities dened by operating
conditions and product specications (cf. bottom section of table 8.4).
The planar view of the two normalized objective functions at dierent weighting factors
is depicted in gure 8.9. The normalization of the objectives requires a priori knowledge
of the minimum (f min ) and maximum (f max ) values for each objective. These values
are found by solving the optimization problem at dierent weighting factors. Each
objective value is then scaled based on the corresponding utopia coordinate and for
a given weighting factor Thus, the normalized objective functions (f ) are computed
according to the expression,
fi fimin
,
(8.64)
fi =
fimax fimin
where the objective function fi is scaled with respect to the corresponding coordinates
of the utopia point,

p
fi
f i = wi U
,
i Zno ,
(8.65)
fi
where no denotes the number of objective functions.
Operational
variables such as reux ratio and condenser pressure have been xed to appropriate
reference values. Table 8.4 lists the actual values.
181
Chapter 8
f'econ
1.0
X[0,1]
0.8
0.6
0.4
Q[0.5,0.5]
0.2
E[1,0]
0.0
0.0
0.2
0.4
0.6
0.8
1.0
f'exergy
Figure 8.9. Pareto optimal curve fecon versus fexergy . Legend: E: economicbased optimal design; X: exergy-based optimal design; Q: optimal
design based equally on economic performance and exergy eciency; Remarks: each point of the curve represents an optimal
solution in terms of the design variables [mcat , Q, a ] and weighting factors w; axes are normalized so that utopia point coordinates are: U=(0,0); original coordinates of U are: U0 =(5.247104
J(sK)1 ,2.082 108 euroy1 ).
This curve is termed Pareto curve (cf. explanatory note 8.1) and clearly shows that one
objective function can only be improved at the expense of the other objective function.
This trade-o leads to signicant deviations between optimal designs and the utopia

fexergy
. The decision maker should have in mind
point (up to 60%) for the pair fecon
the importance of compromising both objectives at early stages of the design cycle.
Signicant structural dierences are noticed between the economic-driven (E; w =
[1 0]T ) and exergy-driven (X; w = [0 1]T ) designs.
As depicted in gure 8.10, for instance, the catalyst load along the reactive trays diers
among the proposed designs when compared to the reference case (R ). The catalyst
load of the reference case is used to normalize the plot. When exergy is fully considered
It
is worth mentioning that, as expected, the optimal design based on economic performance (E)
coincides with the reference case (R).
182
m' cat [-]

6
Q
5
E
1
R
0
3
10
Tray location [-]
Figure 8.10. Normalized catalyst distribution in the synthesis of MTBE with

respect to economic performance and exergy eciency. Legend: R: reference design, whose catalyst distribution is used
for normalization; E: optimal design based only on economic
performance; X: optimal design based only on exergy eciency;
Q: optimal design based equally on economic performance and
exergy eciency.
as an objective function the last reactive tray vanishes together with the sixth reactive
tray. Furthermore, the catalyst load of the seventh reactive tray is considerably higher
than that of the nominal cases. This output suggest an alternative design, where a
side reactor with high catalyst load is placed at the 7th tray. The catalyst load on the
remaining reactive trays decreases when exergy is taken into account. The design with
equally weighted objectives (Q; w = [0.5 0.5]T ) presents an interesting catalyst load
prole, specially at the 9th tray. The rather large catalyst load might be explained by
the reduced catalyst load on the last reactive tray and the constraint of targeting the
product specications. The reduced catalyst load on the 9th tray diminishes the MTBE
decomposition, at high MTBE composition and the product synthesis, at low MTBE
composition. As a constraint in the product specication is included in the problem
formulation, the required extent of reaction is accommodated only by having the 9th tray
fully loaded with catalyst. Since MeOH is fed at the 9th tray, larger holdup of reactant
is present and only converted into products by an increased amount of catalyst.
183
Chapter 8
Entropy rate [W/K]

Se
8000
7000
Sm
Reactive zone
6000
5000
Srx
4000
3000
2000
1000
0
st
th
10
th
15
th
Tray location [-]
Figure 8.11. Entropy production rate prole for an optimal design of a MTBE
RD column based on exergy eciency (X-design). Legend:
Sm : entropy produced by interfacial mass diusion; Se : entropy
produced by interfacial energy transfer; Srx : entropy produced
by chemical reaction; Remarks: stages are numbered from top
to bottom; MeOH feed stream is fed at 9th tray and iC4 stream
at 10th tray.
When the entropy proles of the X-design (cf. gure 8.11) are compared with those of
the E-design, it can be realized that the mayor contribution of entropy is still coming
from the heat transfer process in the stripping section. However, mass diusion becomes
more relevant than in the E-design. In the reactive zone and in contrast to the results
of gure 8.8, the exergy by chemical reaction is unevenly produced along the reactive
trays. Thus, the 9th tray largely contributes to the overall exergy produced within the
reactive section. A fully catalyst load on that tray intuitively explains this behavior.
According to these last results, it seems more attractive to reach high conversions on
one or a few trays and low conversion on surrounding trays. Apparently, this nding
seems to violate the equipartition principle of driving forces, under idealized conditions.
Note that here we focus on the performance of RD process designs, paying less explicit
attention to the industrial conventional processing route (see gure B.1). Therefore,
a comparison between RD and non-RD processes with regard to entropy production
and the design implications for the principle of equipartition of forces are recommended
topics to be further researched.
184
8.6
Application 3. Tri-Objective Optimization of a MTBE

Reactive Distillation Column: A Sensitivity-Based Approach
Based on the insight gathered in the previous applications, we are now ready to assess
the optimal design of a MTBE RD column with respect to fecon, fexergy and fresponse.
For the sake of simplication, the following strategy is adopted: (i) an optimized design
based in economics and exergy eciency is considered; (ii) a disturbance scenario is
dened; and (iii) the responsiveness index of the optimized design is estimated at the
given disturbance scenario.
Optimized design. According to applications 1 (section 8.4) and 2 (section 8.5),
the optimized designs based on fecon and fexergy present an important tradeo between
both objectives. This tradeo depends on the vector of weighting factors, w. The
ultimate decision on w is made by the designer according to the requirements of the
design problem. In our case and only for tutorial purposes, we choose the vector w =
[0.3 0.7]T over all available designs. The optimized design variables for this case are
Disturbance scenario. The selection of a disturbance scenario is carried out by
analyzing qualitatively the entropy proles of gures 8.8 (E-design) and (X-design)
8.11 and by focussing on process disturbances occurring in industrial practice. An
upstream variation of the MeOH feed owrate up to -10% is chosen as disturbance
scenario.
SS
FMeOH = FMeOH
(1 (t)),
0 (t) 0.1,
t [0, tf ].
(8.66)
Note that the disturbance scenario is chosen so that any steady-state transition is
avoided. Further supporting arguments are given in section 7.3.2.
Responsiveness index. For each MeOH feed rate the variables B(x, u) and V (x, u)
are estimated in the corresponding steady state and the responsiveness index is computed. Note that the responsiveness index is treated as a function of the magnitude of
the MeOH feed perturbation, fresponse (FMeOH ).
Specically, this MeOH owrate disturbance results in a variation of the mass transfer
driving forces for each component along the unit. From this point on, the analysis
is focused on the top tray (1st ), the 9th tray and the bottom tray (15th ) of the RD
column specied in table 8.6. Considering these tray locations, we provide a generalized
overview on the units responsiveness in the rectifying, reactive and stripping sections.
The 9th tray is chosen because it is on that particular feeding tray where MeOH is fed to
the column and, therefore, where a more pronounced eect of feed owrate disturbance
is expected. The corresponding mass transfer and chemical reaction driving forces are
185
Chapter 8
Table 8.6. Optimized design of a RD column for MTBE synthesis based

on economic performance and exergy eciency. Remark: w =
[0.3 0.7]T .
Value
Number of trays
Feed 1: location
Feed 1: owrate
Feed 1: composition
Feed 1: temperature
Feed 2: location
Feed 2: owrate
Feed 2: composition
Feed 2: temperature
Operating pressure
Reux ratio
Mass of catalyst [eq[H + ] ]
Bottom owrate
Column diameter
Condenser area
Reboiler area
15
455 mols1
iC4 : 0.37; nC4 : 0.63
350K
MeOH: 9th tray (l)
168 mols1
MeOH: 1.00
320K
11105 Pa (top)
7
m1:2 =0; m3 =771.9
m4 =826.122; m5 =762.405
m6 =688.412; m7 =652.292
m8 =785.991; m9 =1359.42
m10 =982.448; m11:15 =0
197 mols1
rectifying: 6.95 m
reactive: 6.95 m
stripping: 6.33 m
1.059103 m2
6.93102 m2
plotted for trays 1, 15 and 9 in gure 8.12. It can be seen from these plots that both
MeOH mass transfer driving forces increase with larger disturbances, whereas chemical
reaction driving force decreases.
The response time of the MTBE RD column as a function of Xi,MeOH (1st and 15th tray)
and Xrx (9th tray) is depicted in gure 8.13. The variation of B(x, u) and V (x, u)
are computed between the initial reference state and between consecutive states. Note
that the response time becomes larger at small Xi,MeOH -values in the rectifying and
stripping sections, whereas small driving forces lead to short response times in the
reactive section. From a responsiveness perspective, therefore, small driving forces
are desirable for reactive sections and large driving forces are desirable for separation
(nonreactive) sections.
The rst statement suggests that in the case of reactive sections driving forces should
be manipulated to control the unit while ux is specied (i.e. a certain conversion rate).
186
MeOH feed flowrate [mol/s]

168
164.6
161.2
157.8
154.4
151
147.6
168
164.6
161.2
157.8
154.4
151
147.6
10
12
8.5
MeOH mass transfer driving force [J/mol/K]
11.91
8.0
7.5
reference steady
state
11.90
11.89
11.88
11.87
11.86
reference steady
state
11.85
11.84
10
12
Variation of MeOH feed flowrate [%]
(a) Rectifying section, 1st tray.
(b) Stripping section, 15th tray.

168
164.6
161.2
157.8
154.4
151
147.6
10
12
64.0
Reaction driving force [J/mol/K]
reference steady
state
63.5
63.0
62.5
62.0
(c) Reactive section, 9th tray.
Figure 8.12. Driving forces as a function of the MeOH feed owrate. Remarks: MeOH feed owrate varies from the nominal value (168
mol s1 ) to 150 mol s1 ; the optimal design variables and
sizing parameters corresponds to a fecon fexergy design, with
w = [0.3 0.7]T .
Explaining this requires the introduction of an approximated relation between uxes

and forces (Luyben et al., 1999; Hasse, 1969),
Jk
T0 V
Xk ,

2 T0
dV
Xk Xk ,
V
(8.67)
(8.68)
where is a design factor, called thermodynamic design factor by Meeuse (2003).

At small gradient X the design factor should be large enough to obtain the required
ux. In the reactive section small reaction driving forces requires a large reaction volume
on the tray. Small variations in the MeOH feed owrate produce large changes in the
anity and subsequently large entropy production rate values. As the stored exergy
187
Chapter 8
Response
time [s]
Response
time [s]
10000
10000
difference between
reference state
difference between
reference state
1000
1000
difference between
consecutive states
difference between
consecutive states
100
100
7.8
7.9
8.0
8.1
8.2
8.3
8.4
8.5
8.6
11.870
8.7
11.875
11.880
11.885
11.890
11.895
11.900
11.905
11.910
(a) Rectifying section, 1st tray.
(b) Stripping section, 15th tray.
Response
time [s]
10000
difference between
reference state
1000
difference between
consecutive states
100
61.8
62.0
62.2
62.4
62.6
62.8
63.0
63.2
Reaction driving force [J/mol/K]
(c) Reactive section, 9th tray.
Figure 8.13. Response time as a function of the MeOH feed owrate. Remarks: MeOH feed owrate varies from the nominal value (168
mol s1 ) to 150 mol s1 ; the optimal design variables and
sizing parameters corresponds to a fecon fexergy design, with
w = [0.3 0.7]T .
changes only modestly with a small ow disturbance, the overall eect enhances the
process responsiveness (i.e. small response times).
The second statement can be explained by following the same reasoning. In a distillation
(separation) process the driving force (top-bottom) is specied and the ux remains to
inuence. Thus, in nonreactive sections ux should be manipulated to control the unit.
The results of these observations fully agrees with the analysis given by Luyben et al.
(1999) and Meeuse (2003) for a reactor and distillation column.
Meaningfulness and validity of results. Note that in the current approach a
continuous range of disturbances for one input variable is considered for the computation
of the responsiveness time constant. In comparison with the generalized denition of
188
Table 8.7. Entropy produced in classical and green designs. Remarks: classical design corresponds to w = [1
to w = [0.3 0.7]T .
w
[1.0 0.0]T
[0.3 0.7]T
[0.0 1.0]T
0]T ; green design corresponds
fecon [euroy1] fexergy [J(sK)1 ]

2.082108
2.085108
2.086108
5.604104
5.305104
5.251104
fresponse by Luyben et al. (1999), this is a very specic case. Their (Luybens et al.)
denition demands an optimization of the states x to nd the worst case situation over
a domain of states. Therefore, there is a latent risk that the presented results are not
optimized over the feasible domain of states for normal operation. Our approach might
be locally optimized for the given operational window, but might have a low degree of
validity to other cases.
8.7
Comparison Between Classical and Green Designs
In the course of this chapter the economic-driven design (classical design) was implicitly
compared with a new bi-optimized design (green design) for the case of a MTBE RD
column. A set of indicators can be used to compare both designs: (i) exergy loss; and
(ii) closed loop control performance.
The exergy loss (or entropy produced) can be extracted from the Pareto curve presented
in gure 8.9. The relevant data for the classical and green designs and for the fully
exergy-driven design (i.e. w = [0 1]T ) are listed in table 8.7.
As it was expected, the green design produces less entropy than the classical one,
however, economically it is less attractive than the conventional design, as it presents a
higher fecon value. It is worth mentioning that the increase in costs is rather marginal
(2%) and within the uncertainty range of the correlation to predict investments costs
for units and catalyst. Nevertheless, a tendency towards increased cost can be suggested,
but is not signicant for this particular case.
A more signicant change is observed for fexergy (8%). A more revealing concept to
asses the impact of making a process greener is the shadow price p . It is a cost penalty
function for changing entropy production and given by,
p =
fecon
.
fexergy
(8.69)
The shadow prices can be computed directly out from the data in table 8.7 or from
the Pareto curve in gure 8.9. It was found that, when compared to the classical case,
189
Chapter 8
x MTBE [-]
Economic-driven design
(classical)
0.9985
0.9980
0.9975
Bi-optimized design
(green)
0.9970
0.9965
0.9960
0
10000
20000
30000
40000
50000
60000
Time [s]
Figure 8.14. Time variation of MTBE product stream for the classic and
green designs in the presence of a MeOH feed owrate disturbance. Remarks: green design corresponds to w = [0.3 0.7]T ;
column specications for classic and green designs are given in
tables 8.4 and 8.6, respectively.
the green design presents a cost penalty of -2.02, whereas the X-design a value of
-4.37 [euro K MJ1 ]. These values conrm the trade-o between both objectives
and suggest that the greener the design the larger the involved cost penalty.
The closed loop control performance of both designs is now analyzed. The disturbance
scenario under consideration is the one used previously in section 7.3. Thus, a step
variation in the owrate of methanol during the time horizon of 6104 s is given by,
F SS
if t < 1 104 s or t > 4 104 s,
MeOH
(8.70)
FMeOH =
1.1 F SS
if 4 104 s > t > 1 104 s.
MeOH
The dynamic response of the classic and green designs are depicted in gure 8.14. In this
gure the MTBE composition at the bottom stream is presented as a function of time
in the presence of the MeOH feed disturbance. Note that the bi-optimized design (i.e.
green) has an improved closed loop performance compared with the classical design.
The deviation from xMTBE set-point are less pronounced in the green design, which
smoothly returns back to the initial steady state when the disturbance disappears.
Note that the set-point of the green design is slightly lower (0.05%) than the one of
the classic design. This negligible variation in set-points does not aect the quality of
190

the analysis results. It is suggested from our results that incorporating both economic
and exergy-related objectives in the unit design results in a process with better closed
loop performance, provided that control structure and tuning are kept invariant. It is
worth clarifying that a well-tuned control can compensate for any change in the process
operability feature.
8.8
Concluding Remarks
The starting point of this chapter was the denition of a generic building block for
process synthesis (GLRDVE). The sources and the distribution of the entropy production rate in a complete MTBE column were then determined and analyzed. It is believed that insight into such a distribution is helpful for assessing which column related
design decision variables (e.g. feed tray location, reux ratio, heat load distribution
using diabatic operation, catalyst load and mass/heat transfer area) are key to optimize the column performance. The perspective adopted for selecting the design decision
variables was task-oriented rather than the conventionally used equipment-oriented design approach. The chosen design decision variables included the reaction volume, the
amount of heat to be added or removed and the interfacial areas for species mass and
heat transfer. Operational variables such as reux ratio and condenser pressure were
xed at appropriate reference values. Analyzing the steady state entropy production
proles in a MTBE RD column showed that, in terms of entropy production, the separation function dominates over the reaction function. A bi-objective function was then
studied regarding economic and exergy performance. It is suggested that designs based
exclusively on economic or exergy performances dier signicantly in their structure,
showing an existing trade-o. The Pareto curve provides sensitivity information on the
cost eects of improving the exergy eciency. In addition to economics and exergy
objectives, this optimization-based approach covers responsiveness aspects. To dene
a responsiveness criteria, the availability and entropy production rate are regarded as
functions of an internal state vector x and a vector of disturbances and/or manipulated
external excitations u. In our approach the maximization approach suggested by Luyben et al. (1999) is not considered due to its inherent need for high computational eort.
However, a rather formal denition of fresponse was adopted together with a single input
excitation. Note that by not searching over all possible state trajectories, the generality
of the results might be questionable and the external excitation is a fairly conservative
one. Analyzing the responsiveness criteria of the bi-optimized design it reconrmed
that small driving forces are desirable for reactive sections and large driving forces are
desirable for separation (nonreactive) sections (Luyben et al., 1999; Meeuse, 2003). The
rst statement suggests that, in the case of reactive sections, driving forces should be
manipulated to control the unit. The second statement suggests that in nonreactive
191
Chapter 8
sections ux should be manipulated to control the unit. Finally, the economic-driven
design (classical design) was compared with the new bi-optimized design (green design)
for the case of the MTBE RD column. The green design produces less entropy than the
classical one, while being economically less attractive (marginal). The bi-optimized design has an improved closed loop performance compared to the classical design, keeping
the same control structure and settings. It is concluded that incorporating both economic and exergy-related objectives in the unit design results in a process with better
closed loop performance and reduced exergy loss.

In this chapter three performance criteria were introduced in the
RD design problem derived from a life-span perspective. The incorporation of those performance criteria resulted in designs with
structural dierences and with balanced economic-performance,
thermodynamic eciency and that are capable of coping with a
realistic disturbance scenario. Thus, the material presented in
this chapter eectively answers Question 7.
192
It is through science that we prove, but through intuition

that we discover.
Henri Poincare, mathematician (1854-1912)
Conclusions and Outlook
This dissertation covers the design of grassroots reactive distillation processes. In this
concluding chapter we attempt to highlight the key contributions of this research to
the scientic knowledge domain of reactive distillation process design. Here we pay
special attention to the answers of the goal-oriented and engineering questions posted
at the end of chapter 1. Additionally, we discuss various paths for further research
and give some recommendations for future endeavors.
193
Chapter 9
9.1
Introduction
The design and operation of reactive distillation processes has received increasing attention within academia and industrial spheres over the last decades (cf. gure 1 in Malone
and Doherty (2000)). From a process intensied perspective and because of its inherent
complexity, addressing RD is a challenging task. In the course of this dissertation we
have present the relevant results and approach that can be used to produce a life-span
inspired design for reactive distillation processes. In this regard, after having covered
the fundamentals of RD and current design approaches (chapters 2 and 3), we suggest
a novel methodology for RD process design (chapter 4), which then becomes the focus
of chapters 5-8. Each chapter covers a specic aspect of the LiSp-IDM, i.e. life-span
inspired design methodology) and answers to the engineering questions (questions 4-7)
posted in chapter 1 are provided. Taking a broader perspective these engineering answers will now be used to address the goal-oriented questions (questions 1-3), either
qualitatively or quantitatively.
This chapter is organized as follows: in section 9.2 we recall the specic design questions
and provide the supporting scientic argumentation for the answers. In section 9.3, we
infer the answers to the goal-oriented engineering questions. We adopted this ordering
based on the assumption that the material presented in the previous chapters remains
in the readers memory, or at least it is easily retrieved. We address the novelty aspects
of this research in section 9.4. The open opportunities identied in the course of this
work are discussed along with recommendations for further research in section 9.5.
9.2
Conclusions Regarding Specific Scientific Design Questions
Recalling the specic scientic design questions formulated at the beginning of this thesis, we can draw the following conclusions.
Question 4. What is the domain knowledge required and which new echelons
are needed for process synthesis?
Answer. Temporal domain knowledge, physical and operational knowledge and
spatial-related knowledge; a multiechelon approach: formulation, feasibility analysis,
temporal structure, generation of separation sequences, internal spatial structure, behavior analysis, evaluation.
In the case of RD, the design issues are far more complex than those relating to conventional distillation and they may dier signicantly from case to case. Thus, we suggest
that the domain knowledge required for any (integrated) design methodology should
be classied into three key areas: (i) temporal considerations, (ii) physical and opera194

tional conditions and (iii) spatial parameters. In terms of priority of decision the rst
class of key design decision variables involves temporal considerations. The variables of
interest relate to the mode of operation of the system, namely batch, fed batch or continuous operation, together with the time horizon. The second class of design variables
includes the physical and operational conditions of the system. These are aimed at the
favorable creation of enough reaction and separation driving forces and mutual synergy.
In this class design variables such as operation pressure, temperature, reux ratio and
phase velocity are of utmost importance. Moreover, the chemical reaction regime and
mass/heat transfer regime should be specied. The third and last class of decision class
embraces all key spatial parameters, ranging from establishing section with dierent
functionalities, packing parameters, number of trays or length of sections, feeding and
withdrawal points, energy exchange, section diameters, column owsheet conguration
with recycle structures and heat integration.
Both graphical and optimization-based design methods have already shown proven potential in RD process design, but some signicant limitations have also been found.
Graphical methods are fairly exible for generating alternative designs at various design parameters, allowing the designer to dene realistic bounds. Furthermore, the
graphical nature of the methods helps us to clarify our understanding of these fundamental issues in RD, which might be disguised by other approaches. However, graphical
methods are strongly limited by their graphical nature.
Optimization-based methods overcome this last limitation and can be used to solve
successfully design problems that involve multicomponent mixtures, multiple chemical
reactions and multiple units. However, additional caution is required during problem
formulation and inappropriate initial guesses made for the optimization variables may
lead to convergence diculties. Evolutionary methods are used to cover, in more detail,
some of the physical aspects of the design of RD columns. However, this approach
requires a pre-dened structure and accordingly, an optimum design is not guaranteed.
Due to the size and conceptual/computational complexity of the RD design problem, a
single comprehensive design strategy might not be available for some years to come. In
this research, however, we propose an integrated design approach, termed multiechelon
design approach. This strategy allows the designer to combine, in a systematic way, the
capabilities and complementary strengths of the available graphical and optimizationbased methods. Decomposition to give a hierarchy of imbedded design spaces of increasing renement is used to support this approach. Moreover, it constitutes the framework
of a highly aggregated design methodology, which focuses on the RD design problem
from a life-span perspective (LiSp-IDM in chapter 4). The following building blocks
constitute the body of the multiechelon design approach:
Formulation. Sustainability issue are considered and the basis of design are dened,
195
Chapter 9
i.e. product and feedstock purities ranges, production modes, operating window and
economics, and appropriate building model blocks are developed (e.g. thermodynamic,
kinetic and hydrodynamic).
Feasibility analysis. Using graphical methods the feasible product compositions for
a given set of input variables are determined (e.g. feed composition and Damkohler
number) incorporating all occurring phenomena (e.g. distillation boundaries, (non)reactive azeotropy).
Temporal structure. The operation mode is selected, i.e. batch, continuous or semibatch, based on operational skills, production requirements and supply chain dynamics.
Generation of separation sequences. The use of rigorous graphical methods allows
us to identify and organize feasible column sequences. It is necessary to estimate their
type of task, i.e. fully reactive, hybrid and nonreactive columns, together with the
number of units and their connectivity.
Internal spatial structure. The process structure obtained from the previous level
is used as starting point for this stage. Using optimization-based methods we can
obtain continuous variables (e.g. column diameter, heat exchange areas, contacting
area, catalyst concentration, reboil and reux ratios) and discrete variables (e.g. feed
staging, reactive and nonreactive stages, withdrawal staging).
Behavior analysis. Steady-state simulations are used to explore the kinetics, G/L
mass transfer, hydrodynamics and multiplicities inside the units. Additionally, dynamic
simulations are performed to check the robustness of the design, in terms of the ability
to maintain product purities and conversion within a desired range when disturbances
occur.
Evaluation. The design options are checked with respect to their performance. Taking
the wide range of criteria in the safety, health, economics, environment and technology
(SHEET) aspects specic choices are made for economic performance, sustainability and
technology. With respect to economics, we calculate the capital and operational costs
of a process for its time horizon. We use exergy eciency as a sustainability metric,
while a systems response time is used to dene the responsiveness of the process.
Question 5. What are the (performance) criteria that need to be considered
from a life-span perspective when specifying a reactive distillation design
problem?
Answer. Safety, health, environment, economics and technology (SHEET)-related
metrics; in particular TAC, entropy production rate and response time.
Aiming at a design methodology that addresses the problem from a life-span perspective,
we suggested a set of performance criteria. The motivation behind our perspective relies
196

on the relevance of economic performance and potential losses over the process life span.
Specically, we took into consideration exergy-eciency and responsiveness issues.
We considered a set of criteria embracing issues related to safety, health, environment,
economics and technology (SHEET). Provided that safety and health issues are inherent
in a responsible design approach, the remaining issues can be used to dene a novel performance space. Environmental considerations are directly linked to the minimization
of mass, energy and exergy losses. Economic performance is translated quantitatively
to the capital and operational costs of the design. We used the total annualized cost
(TAC) as a convenient index. Additionally, we considered exergy eciency to be an
environment-based metric, more specically the entropy production rate. We completed
the performance criteria set by appending a technology-based index. In this work we
suggested that the responsiveness of the process can be used as an adequate metric and
more specically the response time as dened in Luyben et al. (1999). Using this metric
we get a measure for prevention of loss of material and energy.
Question 6. What (new) methods and tools are needed for reactive distillation process synthesis, analysis and evaluation?
Answer. Improved residue curve mapping technique, multilevel modeling approach,
dynamic optimization of spatial and control structures, steady-state and dynamic behavior analysis, generic lumped reactive distillation volume element, multiobjective optimization criteria.
In the course of this thesis we have covered novel tools and methods that can be used
to assist the design of reactive distillation processes (see chapters 5 to 7). A brief
description of the highlights of the chapters is given below.
We showed that residue curve maps provide valuable insights and design assistance
for nonideal systems, particularly for RD. Transforming the composition variables according to Dohertys approach allows us to dene a reaction invariant space of lower
dimension, formed by attainable product compositions and where the conventional concepts for residue curves can be applied. For the MTBE synthesis we dened two main
subregions in the modied composition simplex, each of which were characterized by
singular points and delimited by separation boundaries. We generated two dierent
separation sequences, covering all the possible reacting mixture compositions.
We proposed a multilevel modeling approach, regarded as an easy-to-handle tool in the
early stages of design. We derived two model levels of dierent degree of complexity to
model the processes taking place in a RD column. Additionally, we performed rigorous
and reliable degree of freedom analysis at each level, where appropriate business-related
decision variables are included. These levels involve an outer (input-output) view and
an inner (physical task) view. These models are linked by the extent of reaction vector,
197
Chapter 9
which allows us to establish whether or not the models are well specied and mutually
consistent. Additionally, the approach allows one to determine whether or not the
business-related decision variables cope with the operational constraints imposed by the
problem formulation (e.g. the SHE issues, reactants availability and market demand).
Formulating design as a dynamic optimization problem, we found that for the synthesis
of MTBE, a tradeo between control and economic performances exists. We solved
this multiobjective optimization problem by incorporating appropriate time-invariant
parameters (e.g. column diameter, heat transfer areas and controllers parameters) in
the frame of a dynamic optimization problem in the presence of deterministic disturbances. The design optimized sequentially with respect to dynamic behavior leads to a
RD process with a total annualized cost higher than that obtained using simultaneous
optimization of spatial and control structures.
The nonlinear coupling of reactions, transport phenomena, phase equilibria and hardware hydraulics arises as a result of the highly complicated dynamic behavior of the
system. We showed that, even for a simplied reactive ash and an idealized system,
a large number of bifurcation diagrams could be identied. We suggest that the main
source of multiplicity is the strong nonlinear coupling of reactions, transport phenomena
and phase equilibria. We used singularity to study the steady-state behavior in RD,
with a particular focus on a reactive ash with a rather ideal system undergoing a mildly
exothermic isomerization reaction using a light-boiling reactant. For this case the number and relative location of the turning points and feasibility boundaries corresponding
to no-liquid or no-vapor products determined the qualitative nature of the bifurcation
diagram. State multiplicity is present in a large area of the parameter space, for the
case under consideration. We identied a large number (25) of qualitatively dierent
bifurcation diagrams for the system. The feasibility boundaries played an important
role in the classication of the reactive ash. Up to ve feasibility boundaries and a
maximum of 3 steady states were found in a single bifurcation diagram. Bifurcation diagrams exhibiting more then two feasibility boundaries are common for liquid or vapor
feed. Increasing the heat of reaction, activation energy or relative volatility enlarges
the area of multiplicity. Furthermore, a large heat of reaction can create regions in
parameter space with 5 feasibility boundaries in a single bifurcation diagram.
Moreover, we explored the process dynamics of a full RD unit using a rather conventional
approach. The scope of this analysis was restricted to a limited version of control system
design, as a single input-single output structure and proportional-integral controllers
were selected a priori selected. The outcome of this approach was used to set a reference
performance for the next novel contribution involving the trade-o between design,
responsiveness and sustainability aspects. We developed a generic dynamic model for
a complete RD column. This model is composed of instances of (non-) reactive trays,
condenser, reux drum, reboiler and controllers. Based on a rigorous degree of freedom
198

analysis and some engineering judgment a control structure was dened. Using the
synthesis of MTBE as our case study, we simulated dynamically the unit behavior in
the presence of realistic disturbances scenarios. Using comparison with published work,
we found a satisfactory degree of agreement between our unit output and literature
data. Moreover, our proposed control structure led to minor variations in the MTBE
bottom fraction in the presence of disturbances.
Question 7. Are there structural dierences and signicant improvements in
designs derived using conventional methodologies from those obtained using
an integrated design methodology?
Answer. Reduction in the number of reactive trays, more uneven distribution of
catalyst and identication of suitable spots for side reactors.
We dened a generic building block for process synthesis: the generic lumped reactive
distillation volume element. We then determined and analyzed the sources and distribution of the entropy production rate in a complete MTBE column. We strongly believe
that insight into such a distribution is helpful for assessing which column related design
decision variables are key to optimize column performance. The perspective adopted for
selecting the design decision variables was task-oriented rather than the conventionally
used equipment-oriented design approach. Analyzing the steady state entropy production proles in a MTBE RD column we found that, in terms of entropy production,
the separation function dominates over the reaction function. Afterwards, we studied
a bi-objective function with respect to economic and exergy performance. We found
that designs based on economic or exergy performances diered signicantly in their
structure, showing an existing trade-o. For instance, the catalyst load along the reactive trays diered among bi-optimized designs (green) when compared to the reference
conventional case. Hence, the number of reactive trays and catalyst load diminishes
when exergy is optimized. Trays with high catalyst loading could be candidate spots for
placing a side reactor. In addition to the economics and exergy objectives, we analyzed
responsiveness aspects in this optimization-based approach. To dene a responsiveness
criterion we adopted a rather specic denition of response time, it being the change in
availability over the change in entropy production rate due to a single input excitation.
Analyzing the responsiveness criteria of the bi-optimized design we reconrmed that
small driving forces are desirable for reactive sections and large driving forces are desirable for separation (nonreactive) sections. Finally, we compared the economic-driven
design (classical design) with the green design for the case of the MTBE RD column.
We found that the green design produced less entropy than the classical one, while being
marginally less attractive from an economics standpoint. Moreover, the green design
had an improved closed loop performance compared to the classical design, keeping
the same control structure and settings. We concluded that incorporating both eco199
Chapter 9
nomic and exergy-related objectives in the unit design results in a process with better
closed loop performance and reduced exergy loss for this specic case study. These facts
provided the motivation for our life-span inspired perspective.
9.3
Conclusions Regarding Goal-Oriented Questions
After having addressed the engineering specic questions we propose the following answers to the goal-oriented questions.
Question 1. What benets can be gained from having a more integrated
design methodology?
In this thesis we proposed and addressed, at a certain degree of detail, a life-span inspired design methodology (LiSp-IDM, see chapter 4). The framework of our approach
was derived from an extensive analysis of current design methodologies in RD. Thus,
the suggested multiechelon approach combines, in a systematic way, the capabilities
and complementary strengths of the available graphical and optimization-based design
methodologies. The multiechelon approach is also supported by decomposition to give
a hierarchy of imbedded design spaces of increasing renement. As a design progresses
the level of design resolution can be increased, while constraints on the physical feasibility of structures and operating conditions derived from rst principles analysis can
be propagated to limit the searches in the expanded design space.
The benets of adopting such an integrated methodology are of dierent nature, ranging from responsible use of resources (via minimization of entropy production rate),
economical attractiveness (via minimization of total annualized costs) and improved
process responsiveness (via minimization of response time metric). Using gradual renement of the design spaces we introduced a benecial understanding of the ongoing
physical and chemical phenomena and guaranteed a more ecient conceptual design
with respect to the SHEET criteria.
The successful application of the integrated design methodology relies on considering the output of the graphical methods as an initial structure on which to base the
optimization-driven design approaches. In this way, the optimized design is likely to
be embedded within the feasible space and the computational eort becomes less demanding due to the reduced number of design variables. This approach is intended
as a means to decompose the complexity of the design problem and it should not be
considered as a mechanistic recipe.
200

Question 2. What are the practical constraints that need to be considered
from a resources point of view, i.e. time, costs, tools and skill levels, when
developing and applying a design methodology in a work process?
The design task is often regarded as a dicult to master activity, due to the high
number of design decision variables and combinatorial complexity. For the RD case this
aspect is magnied due to the large number of additional design variables that have
to be considered and the presence of more interacting phenomena, when compared to
conventional nonreactive separation.
This research has shown that the integrated design methodology, more specically the
multiechelon design approach, while it is a way to beat design complexity by decomposition, requires the mastery of many tools (e.g. RCM, MINLP and rigorous dynamic
modeling) and many concepts (e.g. bifurcation, singulary theory and dynamic optimization). A schematic representation of the tools and concepts required for each design
echelon is given in gure 9.1. This amazing complexity means that the required skill
level is even higher than was initially assumed.
Question 3. What are the essential ingredients for such a design methodology?
Our integrated design methodology is composed of various design spaces, as described
in chapter 4. These design spaces are the essential ingredients for a life-span approach
and are characterized by the ow of information and knowledge domain. In the course
of this thesis, we have addressed each space and, moreover, we have discussed the
input-output information ow. These information aspects and the knowledge domains
involved in each design space are summarized in table 9.1. Each design space (or
echelon) is supported by a tool-kit, which includes techniques and concepts such as
RCM, MINLP, dynamic modeling, dynamic optimization, irreversible thermodynamicsbased and singularity theory.
9.4
Scientific Novelty of this Work
In this dissertation we addressed the fundamentals of reactive distillation process design

and attempt to generate scientic novelty. These contributions to the knowledge domain
are grouped in the following categories.
Formulation of an extended design problem. We formulated a refreshed and
more comprehensive RD design problem in the wider context of process development
and engineering and in a more relevant way regarding sustainability. The nature of
the extension is given by the identication of the design decision variables and their
grouping into three categories: (i) those related to temporal considerations; (ii) those
201
Chapter 9
DESIGN SPACE (echelon)
Feasibility analysis
TOOLS
RCM
Generation of
RCM
Internal spatial
structure
dynamic optimization,
rigorous modeling,
MINLP
Behavior analysis
singularity theory,
dynamic optimization
Evaluation
irreversible
thermodynamics,
multiobjective
optimization
Figure 9.1. Schematic representation of the tools and concepts required at

each design echelon
202
Table 9.1. Summary of input-output information ow

Design specications

Feasibility analysis
Process:
Discrete:
- thermodynamics
- set of components
- set of chemical reactions
- feed composition
- operational pressure
Product:
- composition target(s)
- SHE constraints
- distillation boundaries
- (non-) reactive azeotropes
Continuous:
- product feasible compositions
- kinetics
- overall balances
Column sequencing
Process:
Discrete:
- operational skill
- mode of operation
- type of task,
separation and/or reactive
- number of columns
Continuous:
- feed ratio of reactant streams
Spatial structure
Process:
Continuous:
- overall balances
- column diameter
- heat exchange/contacting areas
- catalyst concentration
- reboil and reux ratios
- controllers parameters
Discrete:
- feed staging/withdrawal
- reactive and nonreactive stages
- control conguration
- dynamic behavior
type of task
number of columns
columns connectivity
feed ratio of reactants
Behavior analysis
Process:
Continuous:
- steady-state behavior
- unit structure
- control structure
- controllability index
- operating conditions
- dynamic behavior
Irreversible thermodynamics-based evaluation

minimization of life-span oriented loss prevention
Process:
Continuous:
- NEQ thermodynamics
- optimal design variables

- performance criteria
- steady-state simulation
- dynamic simulation
- multiobjective
optimization
spatial conguration
control structure and settings
operational parameters
disturbance scenario
203
Chapter 9
related to physical and operational considerations and (iii) those related to spatial
issues.
Integrated design methodology. Based on a detailed analysis of the current design
methodologies in RD we presented an integrated design, i.e. LiSp-IDM. As industrial
relevance requires design issues not to be separated from the context of process development and plant operations, we adopted a life-span inspired perspective. The framework
of this methodology was structured using the multiechelon approach, which combines in
a systematic way the capabilities and complementary strengths of the available graphical and optimization-based design methodologies. A decomposition into a hierarchy
of imbedded design spaces of increasing renement supports this approach. As a design progresses the level of design resolution can be increased, while constraints on the
physical feasibility of structures and operating conditions derived from rst principles
analysis can be propagated to limit the searches in the expanded design space. Moreover, we considered a multilevel approach for classifying the phenomena occurring in a
RD element. We approached the levels top to bottom: from process to unit and from
unit to compartment.
Improvement in design tools. We supported the proposed design methodology
using improved design tools. Firstly, we extended the residue curve mapping technique
to the RD case and systematically applied it to reactive mixtures outside conventional
composition ranges. We found this technique to be particularly useful for the sequencing of (non-) reactive separation trains. Secondly, we rened the models of the process
synthesis building blocks leading to the following sub-improvements: (i) a rened modular representation of the building blocks; (ii) changes/improvements in the models of
the building blocks; and (iii) enhancements in synthesis/analysis tools. Regarding the
last item, we introduced a multilevel modeling approach with the aim of facilitating
the decision-making task in the design of RD spatial structures. Moreover, we formulated our rigorous behavioral model as a dynamic optimization problem and used this
technique as a design tool.
Performance criteria.
To account for the process performance from a life-span
inspired perspective, we dened criteria related to economic, sustainability and responsiveness aspects and embraced them in the proposed design methodology. For the rst
time, the interactions between economic performance, thermodynamic eciency and responsiveness in RD process design were explored and possible trade-os were identied.
This research suggests that incorporating a sustainability-related objective in the design
problem formulation might lead to promising benets from a life-span inspired perspective. On one hand, we accounted for the exergy losses, aiming at their minimization
and, on the other hand, the process responsiveness was positively enhanced.
204
9.5
Outlook and Further Research
In the course of this thesis we mentioned various opportunities for further research. In
this section we briey mention these possible paths for future work.
Sustainability and supply chain considerations in design problems. We used
loss prevention concepts like entropy production rate as a metric for sustainability. We
disregarded any other sustainability aspects like proper choice of feedstocks and products supply chain. We strongly believe that these aspects should be explored in future
work. It is important to realize, however, that incorporating supply chain motivated
temporal features within the design task (e.g. operation and switching modes) might
tremendously increase the degrees of freedom and the problem complexity.
Extended design methodology. In our integrated design methodology we considered exclusively buildings blocks with behavioral models and paid less explicit attention
to structural models. Thus, in our description of the GLRDVE building block we dened a set of connectivity rules, which could be further elaborated. Our connectivity
rules imposed conguration limitations, such as: (i) heat integration over volume elements was not considered, i.e. carrying heat from high temperature to low temperature
elements; (ii) side reactors were not present; and (iii) no parallel structures were considered. Our design approach could be eectively complemented by building blocks with
structural models, allowing for superstructure generation. The work of Pistikopoulos
and co-workers (Papalexandri and Pistikopoulos, 1996; Proios and Pistikopoulos, 2005)
can be used to provide the theoretical background for this task.
More rened simulation models. The challenges in the design of RD are numerous
and exciting. The drive for process intensication and more sustainable plant operations can lead to more intricate geometric structures (e.g. packings, dividing walls
and distributed heat exchange). These structures require new generation of behavioral
and structural models, which couple, in a more realistic way, mass and energy transfer
phenomena.
Steady-state and dynamic behavioral analysis over extended systems. Addressing a single reactive tray we aimed to explore exclusively the interaction between
phase equilibrium and chemical reaction. Therefore, we consider any eect related to
unit conguration outside the scope of the analysis. The possibilities for extending
the steady-state analysis to a more realistic and full-scale RD system are numerous.
Further research should focus on the eect of dierent model assumptions (e.g. a more
rened behavioral model that includes rate limitations) and dierent systems (e.g. a
full column with control structure). Thus, the eect of mass transfer would be interesting to explore, together with the behavior of systems with inert components and side
reactions. Finally, the model of the reactive ash could be extended with nonreactive
trays including a condenser and reboiler. Regarding dynamic behavioral analysis, we
205
Chapter 9
did not address any interaction between column design and control design, this aspect
is worth researching. Moreover, in this research we did not address operational aspects
like exibility. As these operational aspects could become more important than the cost
aspect, we recommend that they are explored in the future.
More rigorous responsiveness index. The responsiveness index, as dened by
Luyben et al. (1999), is a truly system feature with optimization over all possible states.
In our approach we did not consider the maximization approach suggested by Luyben
et al. (1999), due to its inherent computational eort. It is worth mentioning, therefore,
that having not searched over all possible state trajectories, the generality of the results
might be questionable and the external excitation given is a fairly conservative. In
designing a process from a control perspective we want a process that is fast with
respect to responding to changes of manipulated variables towards controlled outputs.
Yet we aim for low sensitivity and slow response to disturbances acting on the process.
The Luyben et al. responsiveness criterion does not make this kind of distinction and
might require a more detailed description.
Non-equilibrium thermodynamics-based analysis. Our analysis based on multiobjective optimization of economics and exergy objective functions revealed that a
benecial design, from an economic and sustainability perspective, is obtained when
high conversion rates occur in a few reaction compartments. Apparently, this nding
seems to violate the equipartition principle of driving forces, under idealized conditions
and for a single driving force. A detailed analysis of this aspect is a relevant topic for
further exploration. Moreover, from a modeling perspective, more accurate approaches
are required when it comes to predicting thermodynamic variables. In particular, estimation of the gradient in chemical potential for multiphase systems in turbulent mixed
systems needs to be explored further.
Testing and validation of the proposed design methodology. We suggest as
further activity, that our LiSp-IDM should be tested and validated. The design method
is similar to a scientic hypothesis that must be tested with regard its truth using some
experiments. The outcome of the experiments, when properly conducted, will give us
the necessary information as which to extent the hypothesis can be maintained or be
rejected. The approach presented by Ajah et al. (2004) provides an adequate theoretical
background for achieving this validation. They suggest that a reliable approach for
the testing and validation of a design framework (including our LiSp-IDM) might be
composed of a sequence of activities: (i) specication of design performance criteria; (ii)
selection of design teams and cases; (iii) testing and monitoring of design process and
(iv ) evaluation of design outputs. Each activity could be constrained by the quantity
and quality of the resources allocated, in terms of time, costs, tools and skill levels.
Thus, we should not fail to realize that, in the industrial sphere, it may rather dicult to
endorse tentative design frameworks on any real scenario (Ajah et al., 2004). Moreover,
206

a high-pressure work environment reduces the availability of experienced practicing
designers who are able to adopt novel design methodologies. Nevertheless, we should
not forget that historical breakthroughs in technology have all originated from human
curiosity rather than from an intimate marriage with society.
207
A
Model Description and Degree of Freedom
Analysis of a Reactive Distillation Unit
A.1
Mathematical Models
A.1.1 (Non)Reactive Tray

Component molar balance:
dMi
dt
in,V
F feed,L zifeed,L + F in,L xin
yiin
i +F
out,L
xi F
out,V
yi +
j=n
rx
j=1
where M in mol, F in mols1 , r in mols1 .

209
frx,j ri,j ,
i Znc
(A.1)
Appendix A
Energy balance :
dU
dt
F feed hfeed,L + F in,L hin,L + F in,V hin,V

F out,L hout,L F out,V hout,V + Q
j=n
rx
frx,j
i=n
c
j=1
ri,j r Hj
(A.2)
i=1
where U in J, h in Jmol1 , Q in W, r H in Jmol1 .

Molar holdup:
Mi = M L xi + M V yi ,
i Znc
(A.3)
U = M L hL + M V hV P Vtray
(A.4)
Internal energy:
where V in m3 , P in Pa.
Heat duty:
Q=0
(A.5)
xi = yi = 1
(A.6)
Normalization expressions:
i=n
c
i=1
Phase equilibrium:
L (T, P, x)i xi = V (T, P, y)i yi ,
i Znc
(A.7)
Chemical reaction driving force:

gj = fg (x, T, P ),
j Znrx
(A.8)
Chemical reaction constant:

Ea
kf,j = k0,j e RT ,
j Znrx
(A.9)
where kf in mol(seq[H + ] )1 , Ea in Jmol1 , R is 8.31 J(molK)1 , T in K.

Chemical equilibrium constant:
ln(Keq,j ) = c1 +
The
c2
+ c3 ln(T ) + c4 T + c5 T 2 + c6 T 3 ,
T
j Znrx (A.10)
heat of reaction in the energy balance can be omitted if the standard heat of formation is used
as reference value for the enthalpy computations.
210
Model Description and D.O.F. Analysis of a RD Unit

Chemical reaction rate:
ri,j = i,j kf,j gj mcat ,
i Znc , j Znrx
(A.11)
where mcat in eq[H + ] .

Tray reactivity:
frx,j
1
=
0
j Znrx
if reactive tray,
if nonreactive tray,
(A.12)
j Znrx
Liquid outow (Francis weir formula)

F out,L = 1.54 leveldownc max((leveltray w htweir ), 0)1.5 L
m,
w [0, 1] (A.13)
where leveldownc , leveltray , htweir in m, m in kgm3 .

Vapor inow:
F
in,V
v
V
max(, 0)
L
m
Vm
(A.14)
P
12.5 1 103
3
level
10
tray
3
L
m L
m g 1 10
c0 Aholes
51
(A.15)
(A.16)
where P in Pa, in molm3 , leveltray in mm, Aholes in m2 , c0 0.8.

Pressure drop:
P in,V P = (P )
(A.17)
Geometry constraints:
Vtray
leveltray
ML
MV
+
L
V
L
M
L
Atray
(A.18)
(A.19)
where leveltray in m, Atray in m2 .
A.1.2 Block Model: PI Controller

Controller input:
InS = SP e
Controller output:
(A.20)
"t
OuS = Bias + K(e +
e dt
)
c
(A.21)
211
Appendix A
A.1.3 Block Model: Condenser

dMi
= F in,V yiin (F D,L + F R,L ) xi ,
dt
i Znc
(A.22)
Energy balance:
dU
= F in,V hin,V (F D,L hD,L + F R,L hR,L ) Qcond
dt
(A.23)
Molar holdup:
i Znc
(A.24)
U = M L hL + M V hV P Vcond
(A.25)
Internal energy:
Reux ratio:
RR =
F R,L
F D,L
(A.26)
Heat duty:
Qcond = F cool Cpcool M W cool (T out,cool T in,cool )
(A.27)
where Cp in J(gK)1 , M W in gmol1.

Heat transfer:
HX
Acond Tlm,cond
Qcond = Ucond
(A.28)
where U HX in W(m2K)1 , A in m2 , Tlm in K.

Logarithmic mean temperature dierence:
Tlm,cond =
(T in T out,cool ) (T T in,cool )
T
ln T T T
in,cool
in
out,cool
(A.29)
i=n
c
xi = yi = 1
(A.30)
i=1
Phase equilibrium:
212
i Znc
(A.31)

Pressure drop:
(P ) =
ff r M W feed (F in,V )2
feed A2total-tube,cond
(A.32)
P in
P + (P )
(A.33)
ff r
8 fpd (lengthtube /di ) + 2.5
(A.34)
where Atotal-tube,cond equals ntubes (d2i 41 ), di in m, lengthtube in m, fpd 3.5103.

Vcond =
ML
MV
+
L
V
levelcond =
(A.35)
ML
L Acond
(A.36)
A.1.4 Block Model: Reboiler

dMi
out,L
= F in,L xin
xi + F out,V yi ),
i (F
dt
i Znc
(A.37)
Energy balance:
dU
= F in,L hin,L (F out,L hL + F out,V hV ) + Qreb
dt
(A.38)
Molar holdup:
i Znc
(A.39)
U = M L hL + M V hV P Vreb
(A.40)
Qreb = F heat (hin,heat hheat )
(A.41)
HX
Qreb = Ureb
Areb Tlm,reb
(A.42)
Internal energy:
Heat duty:
Heat transfer:
Logarithmic mean temperature dierence:

Tlm,reb =
T in,heat T heat
|T
T |
ln |T heat
T |+T
in,heat
(A.43)
lm
where Tlm in K.
213
Appendix A
i=n
c
xi = yi = 1
(A.44)
i=1
Phase equilibrium:
i Znc
(A.45)
Vreb =
ML
MV
+ V
L
levelreb =
(A.46)
ML
L Areb
(A.47)
A.1.5 Block Model: Physical Properties

A detailed description of the calculation of physical properties and thermodynamic
model is given in tables B.3 and B.1 in appendix B.
Molar enthalpy:
h() = fh (T, P, z() )
(A.48)
Thermodynamic model:
() = f (T, P, z() ),
i Znc
(A.49)
Molar density:
() = f (T, P, z() )
(A.50)
m () = fm (T, P, z() )
(A.51)
M W () = fMW (z() )
(A.52)
Mass density:
Molecular weight:
Component vapor pressure:

p0i = fpi (T ),
where p0i in Pa.
214
i Znc
(A.53)
A.1.6 Block Connectivities

Trays connections:
(F, x, P, T, h)out,L
j
(F, y, P, T, h)out,V
j1
(F, z, P, T, h)feed
j
(F, x, P, T, h)in,L
j+1 ,
j [1, nt 1]
(A.54)
(F, y, P, T, h)in,V
j
in
j [2, nt ]
(A.55)
(F, z, P, T, h) ,
jZ
nt
(A.56)
(A.57)
Internal connections:
nt
= (F, x, P, T, h)in,L
reb
(A.58)
reb
out,V
(F, y, P, T, h)reb
(F, y, P, T, h)out,V
1
(F, x, P, T, h)D,L
cond
(F, x, P, T, h)R,L
cond
B,L
= (F, x, P, T, h)
(A.59)
= (F, y, P, T, h)in,V
nt
(A.60)
(F, y, P, T, h)in,V
cond
D,L
(A.61)
= (F, x, P, T, h)
(A.62)
= (F, x, P, T, h)in,L
1
(A.63)
A.1.7 Design Block

Column area:
2
4 Acol = Dcol
(A.64)
Aactive = 0.9 Acol
(A.65)
Aholes = 0.1 Aactive
(A.66)
leveldownc = 0.6 Dcol
(A.67)
where Dcol in m.
Active area:
Holes area:
Downcomer length:
Flooding velocity:
Kood
uood
0.171 ht2space + 0.27 htspace 0.047

+
L V
Kood
V
(A.68)
(A.69)
where htspace in m, u in m s1 .
215
Appendix A
Minimum column diameter:
+
Dcol =
4 FV
1
uood food V
0.9
(A.70)
where food 0.8 0.85.
A.1.8 Economics Block

Capital costs:
cshell
ctrays
ccond
creb
MSI 101.9
(Dcol fm2f t )1.066 (htcol fm2f t )0.802
280 POTshell
(218 + Fc ) Ff ac
MSI 4.7
=
(Dcol fm2f t )1.55 (htstack fm2f t )
280 POTtrays
Fc Fac
MSI 101.3
2
0.65
=
(Acond fm2f
Fc Ff ac
t)
280 POTcond
MSI 101.3
2
0.65
=
(Areb fm2f
Fc Ff ac
t)
280 POTreb
=
(A.71)
(A.72)
(A.73)
(A.74)
where c in eurosy1, POT in years, ht in m.

Utility costs:
cheat
= F heat P Uheat fs2y fmol2kg
(A.75)
ccool
= F
P Ucool fs2y fmol2kg
mcat P Ucat
=
POTcat
= F feed zifeed P Uraw-materials,i fs2y fmol2kg
(A.76)
ccat
craw-materials,i
cool
(A.77)
(A.78)
where P U in eurosunit1 , i Znc .

Total cost:
ccapital
coperational
= cshell + ctrays + ccond + creb

= cheat + ccool + ccat +
i=n
c
craw-materials,i
(A.79)
(A.80)
i=1
ctotal
216
= TAC = ccapital + coperational
(A.81)
A.2
Degree of Freedom Analysis

Table A.1. Degree of freedom analysis for the spatial and control design of a
RD unit: relevant variables.
Variable Tray
Condenser Reboiler Column
F feed
nt
zifeed
P feed
T feed
hfeed
F in,()
xin
i
yiin
T in,()
P in,()
hin,()
Mi
M ()
U
F out,()
xi
yi
T
P
h()
P
ri,j
gj
keq,j
kf,j
level
Dcol
RR
A
F utility
LM T D
T out,utility
Q
ccapital
coperationa
ctotal
nt nc
nt
nt
nt
2 nt
nt nc
nt nc
2 nt
2 nt
2 nt
nt nc
2 nt
nt
2 nt
nt nc
nt nc
nt
nt
2 nt
nt
nt nc nrx
nrx nt
nrx nt
nrx nt
nt
nt
nt
-
1
nc
1
1
1
nc
2
1
1
nc
nc
1
1
2
1
1
1
1
1
1
1
1
-
1
nc
1
1
1
nc
2
1
2
nc
nc
1
1
2
1
1
1
1
1
1
-
1
1
1
TOTAL
A = nt (25 + 3nrx + 6nc + nc nrx )

B = 4nc + 20
C = 4nc + 19
D=3
217
Appendix A
Table A.2. Degree of freedom analysis for the spatial and control design: relevant expressions. Remarks: gures in square brackets refer to
the corresponding expression listed in the mathematical model (cf.
section A.1); the number of phases () is set to a value of 2.
Expression
Tray
Condenser Reboiler Column
Component molar balance nt nc [A.1]
nc [A.22]
nc [A.37] -
Energy balance
Molar holdup
Internal energy
Heat duty
Heat transfer
LMTD
Normalization
Phase equilibrium
Enthalpy
Chem. driving force
Chem. rx. constant
Chem. equi. constant
Chem. rx. rate
Liquid outow
Vapor inow
Pressure drop
Geometry: volume
Geometry: level
Tray connectivity
Internal connectivity
Costs
nt [A.2]
nt nc [A.3]
nt [A.4]
nt [A.5]
2 nt [A.6]
nt nc [A.7]
2 nt [A.48]
nrx nt [A.8]
nrx nt [A.9]
nrx nt [A.10]
nt nc nrx [A.11]
nt [A.13]
nt [A.14]
nt [A.17]
nt [A.18]
nt [A.19]
2(nt 1)(nc + 4)[A.54]
-
1 [A.23]
nc [A.24]
1 [A.25]
1 [A.27]
1 [A.28]
1 [A.29]
2 [A.30]
nc [A.31]
2 [A.48]
1[A.32]
1 [A.32]
1 [A.35]
1 [A.36]
-
1 [A.38]
nc [A.39]
1 [A.40]
1 [A.41]
1 [A.42]
1 [A.43]
2 [A.44]
nc [A.45]
2 [A.48]
1 [A.46]
1 [A.47]
-
4(nc + 4)[A.58]
3[A.79]
TOTAL
F = nt (20 + 5nc ) + 3nrx nt 2nc + nrx nc nt 8

G = 3nc + 13
H = 3nc + 11
I = 4nc + 19
218
Table A.3. Degree of freedom analysis: results

Parameter
Number of variables or equations
Relevant variables
nt (25 + 6nc + 3nrx + nrx nc ) + 8nc + 42
X = A+B+C+D
Model equations
nt (20 + 5nc + 3nrx + +nrx nc ) + 8nc + 35
Y =F +G +H+I
Degrees of freedom
5nt + nt nc + 7
DF = X Y
Input variables
nt F feed,L
nt P feed
nt T feed
nt hfeed,L
nt nc zifeed,L
nt + 7
Degrees of freedom left

Remaining variables to fix
Spatial variables
nt Dcol
Acond
Areb
Operational variable
RR
out,L
Manipulated variables
Fcond
out,L
Freb
F cool
F heat
219
Synthesis of Methyl t-butyl ether

Features of the System
B.1
Motivation
Although methyl t -butyl ether (MTBE) was rst synthesized in 1904, it was only during
the World War II that studies revealed its outstanding qualities as a high-octane fuel
enhancer (Peters et al., 2000). Hence in the early 1990s MTBE was considered one of the
most promising clean burning octane enhancers, being expected to be the second largest
organic chemical produced in United States by the beginning of the second millennium
(Kirk-Othmer, 1994). Furthermore, its interested was promoted even further by the
increased demand for premium-grade fuels and the elimination of lead gasolines.
Lately, however, MTBE harmlessness has been questioned as it has been found in drinking water supplies in several American states. In the state of California, for instance,
MTBE addition in reformulated gasolines has been already banned since the late 90s.
Although similar phase-out scenarios are expected in the USA during the years to come,
a growth in the outside-USA markets could oset lost market demand. This fact is evident by recent joint ventures and capacity acquisitions. Roughly speaking, nowadays
more than 140 MTBE plants are operating, representing a total installed capacity of
20106 t/y (Peters et al., 2000).
221
Appendix B
Table B.1. Typical compositions of C4 streams from FCC (Peters et al., 2000)
Weight fraction
Isobutane
nC4
iC4
1-butene
cis-2-butene
trans-2-butene
1,3-butadiene
balance
0.36
0.13
0.15
0.12
0.09
0.14
0.003
0.007
From a research point of view, MTBE synthesis via reactive distillation still represents
a major development in the eld and an open door to novel and innovative processes.
B.2
Description of the System
MTBE is produced from methanol (MeOH) and isobutene (iC4 ) in an equilibriumlimited liquid-phase reaction (Jimenez et al., 2001) catalyzed heterogeneously by bentonites, zeolites or strong acidic macroporous ion-exchange resins (e.g. Amberlyst 15),
or homogeneously by sulfuric acid according to the following stoichiometry,
C4 H8
isobutene
CH3 OH C5 H12 O
methanol
MTBE
(B.1)
The iC4 stream can be originated from several sources (Peters et al., 2000): (i) as
coproduct of butadiene production from steam cracker C4 fractions, (ii) as product of
selective hydrogenation of butadiene in mixed C4 fractions from steam crackers (iii)
as iC4 in the C4 fraction of FCC units (iv ) as coproduct of the dehydrogenation of
isobutane, and (v ) as coproduct of dehydration of t -butanol.
In this research, the FCC source has been considered as feedstock for the MTBE production. This stream, which represents 29% of the MTBE production worldwide, contains
a complete range of butanes and butenes, as can be seen in table B.1. For the sake of
simplifying the study and provided that the reaction is highly selective for iC4 all the
uttinger,
organic components in the iC4 stream are lumped together as inert nC4 (G
1998).
Conventional process. The conventional route in the synthesis of MTBE is described in detail in Kirk-Othmer (1994) and Peters et al. (2000). For instance, the
H
uls-MTBE process (gure B.1) operates with a given molar excess of methanol on
222
HX03
M03
METHANOL
C4-FRESH
D02
13
M01
M02
14
IN-REAC
HX01
15
18
R01
R03
12
30
HX02
16
R02
11
RECOVERY
SECTION
S01
17
S02
D01
MTBE
RAFF-2
Synthesis of MTBE: Features of the System
Figure B.1. Conventional route for the synthesis of MTBE: two-stage Huls
-MTBE process. Legend: R01: tubular reactor; R02-R03: adiabatic reactors; HX: heat exchanger; M: mixer; D: divider; S01S02: distillation towers (adapted from Peters et al. (2000)).
223
Appendix B
a macroporous acidic ion exchanger (50-90 C, 1-1.5105 Pa). The reaction section is
equipped with adiabatic reactors for the chosen FCC-C4 feedstock. The iC4 conversion reaches 85%. By simultaneous C4 evaporation a nal iC4 conversion of 97% is
achieved. The separation train can involve RD units, where the partly reacted mixture
from the reaction section enters the RD column below the catalyst packing zone to
ensure the separation of the high-boiling component MTBE from the feed stream. The
catalyst packing is installed in the upper mid portion of the column with nonreactive
separation sections above and below. The separation of the C4 stream from the MTBE
is aected by the azeotropy between MeOH and nC4 (cf. section 2.4). An extraction train with water wash is required to completely remove the methanol from the C4
stream. The recovered methanol is recycled to the reaction section.
B.3
Thermodynamic Model
In this study the formulation (see explanatory note B.1) is used, in which the
fugacities of each component in the vapor phase equal those in the liquid phase. This
results is the following expression,
yi i P = xi i p0i ,
i [1, n].
(B.2)
The vapor phase is assumed to behave ideally (i = 1), whereas the liquid-phase is
modelled using Wilson activity coecient equation. This thermodynamic method is
particularly accurate for multicomponent mixtures that do not form two liquids (Kooijman and Taylor, 2000) and is given by the following set of algebraic equations,
ln(ij ) =
SW,i
aij + bij /T,

j=n
c
(B.9)
xj ij ,
(B.10)
j=1
ln(i ) =
1 ln(SW,i )
k=n
c
k=1
xk ki
,
SW,k
(B.11)
where ij is the binary interaction parameter for the pair i, j Znc and aij and bij are
model parameters (table B.2).
B.4
Physical Properties, Reaction Equilibrium and Kinetics
The models that predict the physical properties of the iC4 -MeOH-MTBE-inert system
are given in tables B.3 and B.4. The reaction equilibrium and kinetics of MTBE synthesis have been subject of intensive research during the last decade (Colombo et al., 1983;
224
thermodynamic model. This method uses the denition of the fugacity coecient () and
activity coecient () to describe the properties of the gas and liquid phase(s), respectively
(Malanowski and Anderko, 1992). The computation of VL-equilibria is based on the condition
of equality of intensive properties in the coexisting phases. In terms of fugacities, the equality
is given by,
fiL = fiV ,
(B.3)
()
where fi
is the fugacity of component i in phase ().
The vapor fugacity can be written in terms of the fugacity coecient phi,
fiV = P yi i .
(B.4)
Taking as reference state the pure liquid i at equilibrium temperature T and pressure P , the
liquid phase fugacity fiL can be related to the activity coecient i by,
fiL = fi0 (P ) xi i ,
(B.5)
where fi0 is the fugacity of the pure substance at pressure P , which can be expressed in terms
of the fugacity coecients of pure substances, 0i and the Poynting correction factor,
fi0 (P ) = p0i 0i e
"
vi
P0 RT
dP
p
i
(B.6)
Replacing B.4 and B.5 into B.6 leads to the general VL-equilibrium condition,
P yi i = xi i p0i ,
(B.7)
where the correction factor i accounts for the vapor phase nonideality and the dierence
between the equilibrium pressure and the pure component vapor pressure, p0i ,
i =
i
e
0i
"P
vi
RT
p0
i
dP
(B.8)
Note B.1. Phase equilibrium intermezzo: the thermodynamic model

Table B.2. Wilson interaction parameters for the system iC4 -MeOH-MTBEuttinger, 1998)
nC4 at 11105 Pa (Ung and Doherty, 1995c; G
iC4
iC4
MeOH
MTBE
nC4
iC4
MeOH
MTBE
nC4
MeOH
aij [-]
0.00000 -0.74200
0.74200 0.00000
-0.2413 -0.98330
0.00000 -0.81492
MTBE
0.24130
0.98330
0.0000
0.00000
nC4
0.00000
0.81492
0.00000
0.00000
bij [K]
0.0000 -85.5447 30.2477
0.0000
-1296.719
0.0000 -746.3971 -1149.280
-136.6574 204.5029
0.0000
0.0000
0.0000 -192.4019
0.0000
0.0000
225
Appendix B
Rehnger and Homann, 1990; Zhang and Datta, 1995; Izquierdo et al., 1992; Thiel
et al., 2002). A brief description of the reported models for equilibrium and kinetics
constants is given in tables B.5 and B.6, respectively.
Table B.3. Set of expressions used to predict relevant physical properties
Property
Molecular weight
Expression and nomenclature

i=n
()
M W () = i=1 c M W i zi
M W is in gmol1 and z () denotes molar fraction in the
corresponding phase ()
Vapour pressure
ln(p0i ) = Ai + Bi /(T + Ci )
p0i is in Pa and Ai , Bi and Ci are model parameters of comuttinger, 1998) and listed in
ponent i Znc , taken from (G
table B.4
Component enthalpy
v Hi = R Tc,i (7.08 (1 TTc,i )0.354 +

+ 10.95 i (1 TTc,i )0.456 )
hVi = hVf,i + cpai (T Tref ) +
c
2
2
+ pbi
2 (T Tref ) +
cpci
3
3
) +
+ 3 (T Tref
cpdi
4
4
+ 4 (T Tref
)
V
v
=
h
H
hL
i
i
i
h is in Jmol1 , Tc is the critical temperature, is the acentric factor and cpa , cpb , cpc and cpd are parameters listed in
table B.4
Mixture enthalpy
h() =
i=nc
i=1
()
hi
()
zi
h is given in Jmol1 and z () denotes the corresponding

molar fraction in phase ()
1+(1 TT )0.2857
Component molar density L

i Tc,i zri
P
V
R = i T
c,i
Pc,i
R
is given in molm3
Mixture molar density
Mixture mass density
i=n
V = i=1 c Vi yi
i=nc xi
1
L
i=1
L =
i=ni
feed = i=1 c feed
zifeed
i
()
()
1000 m
= M W ()
m is in kgm3 and () denotes the corresponding phase
226
Table B.4. Parameters used for the estimation of physical properties in the
synthesis of MTBE
Parameter
MW
Pc
Tc
cpa
cpb
cpc
cpd
hVf
zr
A
B
C
Unit
iC4
MeOH
nC4
MTBE
gmol
56.11
32.01
58.12
88.15
5
5
5
Pa
4010
80.910
3810
33.7105
K
417.9
513.15
425.18
497.14
0.194
0.556
0.199
0.266059
Jmol1K1
16.05
21.15
9.487
2.534
1
2
Jmol K
0.2804
0.07092
0.3313
0.5136
Jmol1K3 -1.091104 2.587105 -1.1408104 -2.596104
Jmol1K4 9.098109 -2.852108 -2.822109 4.303108
Jmol1 -1.691104 -2.013105 -1.262105 -2.931105
0.2728
001224
0.2730
0.26721
20.6556
23.49989
20.57070
20.71616
K -2125.74886 -3643.31362 -2154.8973 -2571.58460
K -33.16000 -33.43400
-34.42000 -48.40600
227
Appendix B
Table B.5. Temperature dependence of equilibrium constant in MTBE synthesis
ln(Keq )
+ ln(T )
T
10.0982
4254.0500
0.2667
(B.12)
Source:
G
uttinger (1998); Colombo et al. (1983)
Validity: homogeneous catalyst (e.g. sulfuric acid); UNIFAC model
ln Keq
ln Keq,0 + (
1
1
T
) + ln( ) + (T T0 ) +
T
T0
T0
+ (T 2 T02 ) + (T 3 T03 ) + (T 4 T04 )
1.49277 103
7.74002 101
5.07563 101
9.12739 104
1.10649 106
6.27996 1010
T0
298.15K
Keq,0
284
(B.13)
Source:
Rehnger and Homann (1990); Thiel et al. (2002)

Validity: heterogeneous catalyst (Amberlyst 15)
ln Keq
ln keq,0 +
1
+ ln(T ) + T +
T
T2 + T3
Source:
14634
233
1.066
1.077 103
5.306 107
ln Keq,0
1144
Izquierdo et al. (1992)

Validity: heterogeneous catalyst (ion exchange resin Lewatit SPC 118 BG); low temperature (313K<T<353K)
228
(B.14)
Table B.6. Temperature dependence of kinetic constant in MTBE synthesis
kf (T ) =
kf (T ) =
Ea
kf (T ) e R ( T T )
1
(B.15)
243.8 103 mol (s eq[H+] )1
Ea
92400 J mol1
363K
Source:
Rehnger and Homann (1990)

kf (T ) =
kf (T ) =
Ea
kf (T ) e R ( T T )
1
(B.16)
15.5 103 mol (s eq[H+] )1
Ea
92400 J mol1
333.15K
Source:
Thiel et al. (2002)

kf (T ) =
Ea
kf (T ) e RT
kf,0
3.5714 1012 mol (s eq[H+] )1
Ea
85400 J mol1
(B.17)
Source:
Izquierdo et al. (1992)

Validity: heterogeneous catalyst (ion exchange resin Lewatit SPC 118 BG);
low temperature (313K< T <353K)
229
References
Agreda V. and Partin L. (1984). Reactive distillation process for the production of methyl acetate
(US4435595). Patent. 1.2.1
Ajah A., Swinkels P.L.J. and Grievink J. (2004). Towards testing and validation of novel conceptual
process design frameworks. AIChE Symposium Series 2004, 231235. 9.5, B.4, B.4
Al-Arfaj M. and Luyben W. (2002a). Comparative control study of ideal and methyl acetate reactive
distillation. Chemical Engineering Science 57, 50395050. 2.9.1.2, 2.9.1.2, 2.9.2
Al-Arfaj M. and Luyben W. (2002b). Design and control of an olen metathesis reactive distillation
column. Chemical Engineering Science 57 (2002), 715733. 2.9.1.2
Alejski K. and Duprat F. (1996). Dynamic simulation of the multicomponent reactive distillation.
Chemical Engineering Science 51 (18), 42374252. 7.3.1
Almeida-Rivera C.P. and Grievink J. (2001). Reactive distillation: On the development of an integrated
design methodology. Chemie Ingenieur Technik 73 (6), 777777. 3.5, , 4.1, 4.5
Almeida-Rivera C.P. and Grievink J. (2002). Feasibility of equilibrium-controlled reactive distillation
process: application of residue curve mapping. Computer-Aided Chemical Engineering 10, 151156.
3.2.2,
Almeida-Rivera C.P. and Grievink J. (2004). Feasibility of equilibrium-controlled reactive distillation
process: application of residue curve mapping. Computers and Chemical Engineering 28 (1-2), 17
25. 3.2.2,
Almeida-Rivera C.P., Swinkels P.L.J. and Grievink J. (2003). Multilevel modelling of spatial structures
in reactive distillation units. Computer-Aided Chemical Engineering 15, 696701.
Almeida-Rivera C.P., Swinkels P.L.J. and Grievink J. (2004a). Designing reactive distillation processes:
present and future. Computers and Chemical Engineering 28 (10), 19972020. , 3.5, , 4.1, 4.5, 5.1
Almeida-Rivera C.P., Swinkels P.L.J. and Grievink J. (2004b). Economics and exergy eciency in the
conceptual design of reactive distillation processes. AIChE Symposium Series 2004, 237240. 1.3.2,
Backhaus A.A. (1921a). Apparatus for the manufacture of esters (US1.400.849.20.12). Patent. 1.2.1
Backhaus A.A. (1921b). Apparatus for the manufacture of esters (US1.400.850.20.12). Patent. 1.2.1
Backhaus A.A. (1921c). Apparatus for the manufacture of esters (US1.400.851.20.12). Patent. 1.2.1
231
References
Balakotaiah V. and Luss D. (1984). Global analysis of the multiplicity features of multi-reaction
lumped-parameter systems. Chemical Engineering Science 39, 865881. 7.2.3
Bansal V. (2000). Analysis, design and control optimization of process systems under uncertainty.
Thesis/dissertation, University of London, Imperial College of Science, Technology and Medicine.
3.3, 3.3.3, 3.3.3, 3.3, 6.2, 6.2.1, 6.2.2
Bansal V., Perkins J., Pistikopoulos E., Ross R. and van Schijndel J. (2000). Simultaneous design and
control optimisation under uncertainty. Computers and Chemical Engineering 24, 261266. 3.3,
3.3.3, 3.3.3, 3.3, 6.2, 6.2.1, 6.2.2, 6.2.2, 6.2.3.1, 6.3
Bansal V., Sakizlis V., Ross R., Perkins J. and Pistikopoulos E. (2003). New algorithms for mixedinteger dynamic optimization. Computers and Chemical Engineering 27, 647668. 6.3
Barbosa D. and Doherty M. (1988a). Design and minimum-reux calculations for single-feed multicomponent reactive distillation columns. Chemical Engineering Science 43 (7), 15231537. 2.4, 2.4,
2.4, 3.2.2, 3.2.4, 3.2.4
Barbosa D. and Doherty M.F. (1987a). A new set of composition variables for the representation of
reactive-phase diagrams. Proceedings of the Royal Society of London Series A Mathematical Physical
and Engineering Sciences 413, 459464. 2.4, 3.2.4, 3.2.4, 3.2.4
Barbosa D. and Doherty M.F. (1987b). Theory of phase diagrams and azeotropic conditions for twophase reactive systems. Proceedings of the Royal Society of London Series A Mathematical Physical
and Engineering Sciences 413, 443458. 2.4, 2.4, 3.2.4, 3.2.4
Barbosa D. and Doherty M.F. (1988b). The simple distillation of homogeneous reactive mixtures.
Chemical Engineering Science 43 (3), 541550. 3.2.2, 3.2.4, 3.2.4
Baur R., Higler A., Taylor R. and Krishna R. (2000). Comparison of equilibrium stage and nonequilibrium stage models for reactive distillation. Chemical Engineering Journal 76, 3347. 2.1, 2.6, 2.9.3,
4.3
Baur R. and Krishna R. (2002). Hardware selection and design aspects for reactive distillation columns.
A case study on synthesis of TAME. Chemical Engineering and Processing 41, 445462. 2.6, 2.9.3,
4.3
Baur R. and Krishna R. (2003). Distillation column with reactive pump arounds: an alternative to
reactive distillation. Catalysis Today 79-80, 113123. 4.2, 4.3
Baur R., Taylor R. and Krishna R. (2001a). Dynamic behaviour of reactive distillation columns described by a nonequilibrium stage model. Chemical Engineering Science 56, 20852102. 2.9.1.2,
2.9.1.2, 2.9.3, 4.3
Baur R., Taylor R. and Krishna R. (2001b). Dynamic behaviour of reactive distillation tray columns
described with a nonequilibrium cell model. Chemical Engineering Science 56, 17211729. 2.9.3,
4.3, 7.2.1
Baur R., Taylor R. and Krishna R. (2001c). Dynamics of a reactive distillation column for TAME
synthesis described by a nonequilibrium stage model. Computer-Aided Chemical Engineering 9,
9398. 2.9.3, 4.3, 7.2.9, 7.4
Baur R., Taylor R. and Krishna R. (2001d). Inuence of column hardware on the performance of
reactive distillation columns. Catalysis Today 66, 225232. 2.6, 2.9.3
232
References
Baur R., Taylor R. and Krishna R. (2003). Bifurcation analysis for TAME synthesis in a reactive
distillation column: Comparison of pseudo-homogeneous and heterogeneous reaction kinetics models.
Chemical Engineering and Processing 42, 211221. 2.9, 2.9.1.2, 2.9.1.2, 2.9.2, 2.9.3
Bedeaux D. and Kjelstrup S. (2004). Irreversible thermodynamics - a tool to describe phase transitions
far from global equilibrium. Chemical Engineering Science 59, 109118. 8.2.1
Bedenik M.I., Pahor B. and Kravanja Z. (2001). Synthesis of reactor/separator networks by the combined MINLP/analysis approach. Computer-Aided Chemical Engineering 9, 5964. 3.3
Bermingham S. (2003). A design procedure and predictive models for solution crystallisation processes.
Thesis/dissertation, Delft University of Technology. 4.2
Bessling B., Loning J.M., Ohligschlager A., Schembecker G. and Sundmacher K. (1998). Investigations on the synthesis of methyl acetate in a heterogeneous reactive distillation process. Chemical
Engineering and Technology 21 (5), 393400. 3.2.6, 3.6
Bessling B., Schembecker G. and Simmrock K.H. (1997). Design of processes with reactive distillation
line diagrams. Industrial and Engineering Chemistry Research 36, 30323042. 2.6, 2.6, 3.2.6, 3.1,
3.3, 3.8
Biegler L., Grossman I. and Westerberg A. (1997). Systematic methods of chemical process design.
Prentice Hall, New York. 2.2, 4.3, 4.2
Bildea C.S. and Dimian A. (1999). Singularity theory approach to ideal binary distillation. AIChE
Journal 45 (12), 26622666. 2.9, 2.9.1.1, 7.2.3
Bird B., Stewart W. and Lightfoot E. (2003). Transport phenomena. John Wiley and Sons, Inc., New
York. 8.2.1, 8.2.1, 8.2.1
Bisowarno B.H. and Tade M.O. (2000). Dynamic simulation of startup in ethyl tert-butyl ether reactive
distillation with input multiplicity. Process Design and Control 39, 19501954. 2.9.1.2, 2.9.1.2, 2.9.2
Buzad G. and Doherty M. (1994). Design of three-component kinetically controlled reactive distillation
columns using xed-point methods. Chemical Engineering Science 49 (12), 19471963. 3.2.4, 3.2.4,
3.2.4, 3.2.4
Buzad G. and Doherty M. (1995). New tools for the design of kinetically controlled reactive distillation
columns for ternary mixtures. Computers and Chemical Engineering 19 (4), 395408. 1.3.1, 3.2.4,
3.2.4, 4.2, 5.1
Cardoso M.F., Salcedo R.L., de Azevedo S.F. and Barbosa D. (2000). Optimization of reactive distillation processes with simulated annealing. Chemical Engineering Science 55, 50595078. 3.3,
3.3.1
Chadda N., Malone M.F. and Doherty M. (2003). Feasibility and synthesis of hybrid reactive distillation
systems. AIChE Journal 48 (12), 27542768. 3.2.5
Chen F., Huss R.S., Doherty M. and Malone M.F. (2002). Multiple steady states in reactive distillation:
kinetic eects. Computers and Chemical Engineering 26, 8193. 2.1, 2.5, 2.9, 2.9.1.1, 2.9.1.2, 2.9.1.2,
7.2.1
Ciric A. and Gu D. (1994). Synthesis of nonequilibrium reactive distillation processes by MINLP
optimization. AIChE Journal 40 (9), 14791487. 2.9.3, 2.10, 3.2.4, 3.2.4, 3.3, 3.3.1, 4.3
233
References
Ciric A. and Miao P. (1994). Steady state multiplicites in a ethylene glycol reactive distillation column.
Industrial and Engineering Chemistry Research 33, 27382748. 2.9, 2.9.1.1, 2.9.2, 2.9.3, 3.3, 7.2.1,
7.2.8
Clark P. and Westerberg A. (1983). Optimization for design problems having more than one objective.
Computers and Chemical Engineering 7 (4), 259278. 1.3.2, 4.4, 8.3.4, 8.3.4
Colombo F., Cori L., Dalioro L. and Delogu P. (1983). Equilibrium constant for the methyl tert-butyl
ether liquid phase synthesis by use of UNIFAC. Industrial Engineering Chemistry Fundamentals
22 (2), 219223. B.4, B.5
Costa L. and Oliveira P. (2001). Evolutionary algorithms apporach to the solution of mixed integer
non-linear programming problems. Computers and Chemical Engineering 25, 257266. 3.3.1
Coulson J.M. and Richardson J.F. (1991). Chemical engineering. Volume 6. Design. BPCC Wheatons
Ltd., London. 7.3.2.1
Dhooge A., Govaerts W., Kuznetsov Y., Mestrom W. and Riet A.M. (2003). cl-matcont: a continuation
toolbox in Matlab. In: Proceedings of the 2003 ACM symposium on Applied computing, (161166),
ACM Press. 7.2.3
Doherty M. and Malone M.F. (2001). Conceptual design of distillation systems. McGraw-Hill, USA.
1.2.1, 2.1, 2.5
Doherty M.F. and Buzad G. (1992). Reactive distillation by design. Trans. Institution of Chemical
Engineers - Part A 70, 448458. 1.3.1, 2.4, 2.4, 2.4, 3.2.2, 3.2.4, 3.2.4, 3.2.4, 4.2, 4.3, 5.1
Douglas J. (1988). Conceptual design of chemical process. McGraw-Hill, USA. 1.3.1
Eldarsi H.S. and Douglas P.L. (1998). Methyl-tert butyl ether catalytic distillation column. Part I:
multiple steady states. Chemical Engineering Research and Design 76, 509516. 2.9.1.2
Espinosa J., Aguirre P., Frey T. and Stichlmair J. (1999). Analysis of nishing reactive distillation
columns. Industrial and Engineering Chemistry Research 38 (1), 187196. 3.2.2
Espinosa J., Aguirre P. and Perez G. (1996). Some aspects in the design of multicumponent reactive
distillation columns with a reacting core: mixtures containing inerts. Industrial and Engineering
Chemistry Research 35 (12), 45374549. 2.9.3, 3.2.2, 3.2.4, 3.2.4, 3.2.4
Fair J., Seibert F., Behrens M., Saraber P. and Olujic Z. (2000). Structured packing performance experimental evaluation of two predictive models. Industrial and Engineering Chemistry Research
39, 17881796. 2.7.1, 2.7.1
Feinberg M. (1999). Recent results in optimal reactor synthesis via attainable theory. Chemical Engineering Science 54, 25352543. 3.2.3, 4.2
Fien G. and Liu Y. (1994). Heuristic synthesis and shortcut design of separation processes using residue
curve maps: a review. Industrial and Engineering Chemistry Research 33, 25052522. 2.3, 3.2.2,
4.2, 5.3, 5.3, 5.3, 5.4, 5.4, 5.5.3, 5.6
Frass M. (2005). Begrie der DIN 19226 - Regelung und Steuerung. Data File. 1.3.1
Frey T. and Stichlmair J. (1999a). Reactive azeotropes in kinetically controlled reactive distillation.
Trans. Institution of Chemical Engineers - Part A 77 (7), 613618. 2.3, 2.4, 2.4, 2.6
234
References
Frey T. and Stichlmair J. (1999b). Thermodynamic fundamentals of reactive distillation. Chemical
Engineering and Technology 22 (1), 1118. 2.3, 2.3, 2.2, 2.4, 2.3, 3.2.4, 3.3, 3.2.6
Frey T. and Stichlmair J. (2001). MINLP Optimization of reactive distillation columns. ComputerAided Chemical Engineering 8, 115120. 3.3, 3.3.1
Gadewar S.B., Chadda N., Malone M.F. and Doherty M. (2003). Feasibility and process alternatives
for reactive distillation. In: Reactive distillation. Status and future directions (K. Sundmacher and
A. Kienle, eds.), book chapter 6, (145168), Wiley-VCH. 3.2.3, 3.2.5
Gehrke V. and Marquardt W. (1997). A singularity theory approach to the study of reactive distillation.
Computers and Chemical Engineering 21 (SS), S1001S1006. 2.9, 2.9.1.1, 7.2.3
Georgiadis M., Schenk M., Pistikopoulos E. and Gani R. (2002). The interactions of design, control
and operability in reactive distillation systems. Computers and Chemical Engineering 26, 735746.
3.3, 3.3.3, 3.3.3, 4.3, 6.2, 6.2.1, 6.2.3.1, 6.2.3.2
Giessler S., Danilov R.Y., Pisarenko R.Y., Seramov L.A., Hasebe S. and Hashimoto I. (1998). Feasibility study of reactive distillation using the analysis of the statics. Industrial and Engineering
Chemistry Research 37, 43754382. 2.9.3, 3.2.1, 3.2.1, 3.3
Giessler S., Danilov R.Y., Pisarenko R.Y., Seramov L.A., Hasebe S. and Hashimoto I. (2001). Systematic structure generation for reactive distillation processes. Computers and Chemical Engineering
25, 4960. 3.2.1
Giessler S., Danilov R.Y., Seramov L.A., Hasebe S. and Hashimoto I. (1999). Feasible separation
modes for various reactive distillation systems. Industrial and Engineering Chemistry Research 38,
40604067. 3.2.1, 3.1, 3.2.1
Goel H. (2004). Integrating Reliability, Availability and Maintainability (RAM) in conceptual process
design. Thesis/dissertation, Delft University of Technology. 1.3.1
Golubitsky M. and Schaeer D.G. (1985). Singularities and groups in bifurcation theory. SpringerVerslag, New York. 7.2.3
Gorissen H.J. (2003). A general approach for the conceptual design of counter-current reactive separations. Chemical Engineering Science 58, 809814. 4.3
Grievink J. (2003). Lecture notes on Process Systems Design. Unpublished Work. 1.3.1
Grievink J. (2005). SHEET performance criteria. Personal Communication. 4.4
Grossman I. and Westerberg A. (2000). Research challenges in Process Systems Engineering. AIChE
Journal 46 (9), 17001703. 1.1, 1.3.1, 1.3.2, 1.3.3
Gr
uner S. and Kienle A. (2004). Equilibrium theory and nonlinear waves for reactive distillation
columns and chromatographic reactors. Chemical Engineering Science 59, 901918. 2.9
Gumus Z. and Ciric A. (1997). Reactive distillation column design with vapour/liquid/liquid equilibria.
Computers and Chemical Engineering 21 (SS), S983S988. 3.3, 3.3.1
G
uttinger T.E. (1998). Multiple steady states in azeotropic and reactive distillation. Thesis/dissertation, Swiss Federal Institute of Technology. 2.8, 2.9, 2.9.1, 2.9.1.1, 2.9.2, 3.2.4, 4.2,
5.5.2, B.2, B.2, B.3, B.5
235
References
G
uttinger T.E. and Morari M. (1997). Predicting multiple steady states in distillation: singularity
analysis and reactive systems. Computers and Chemical Engineering 21 (SS), S995S1000. 2.9,
2.9.1.1, 2.9.1.1, 2.9.1.2, 2.9.2, 4.2, 7.2.9
G
uttinger T.E. and Morari M. (1999a). Predicting multiple steady states in equilibrium reactive distillation. 1. Analysis of non-hybrid systems. Industrial and Engineering Chemistry Research 38 (4),
16331648. 2.1, 2.9, 2.9.1.1, 2.9.2, 4.2
G
uttinger T.E. and Morari M. (1999b). Predicting multiple steady states in equilibrium reactive distillation. 2. Analysis of hybrid systems. Industrial and Engineering Chemistry Research 38 (4),
16491665. 2.1, 2.9, 2.9.1.1, 2.9.2, 4.2
Harding S.T. and Floudas C.A. (2000). Locaring all heterogeneous and reactive azeotropes in multicomponent mixtures. Industrial and Engineering Chemistry Research 39, 15761595. 2.4
Harding S.T. and Floudas C.A. (2001). EQUISTAR: Reliable software for design of nonideal and
reactive systems. Computer-Aided Chemical Engineering 9, 153158. 2.4
Harding S.T., Maranas C.D., McDonald C.M. and Floudas C.A. (1997). Locating all homogeneous
azeotropes in multicomponent mixtures. Industrial and Engineering Chemistry Research 36, 160
178. 2.4
Harmsen G.J. and Chewter L.A. (1999). Industrial applications of multi-functional, multi-phase reactors. Chemical Engineering Science 54, 15411545. 1.3.1
Harmsen G.J., Hinderink P., Sijben J., Gottschalk A. and Schembecker G. (2000). Industrially applied
process synthesis method creates synergy between economy and sustainability. AIChE Symposium
Series 96 (323), 364366. 1.3.1, 1.3.2
Hasse R. (1969). Thermodynamics of irreversible processes. Addison-Wesley Publishing Company,
Darmstadt. 8.3.3.2, 8.6
Hauan S. (1998). On the behaviour of reactive distillation systems. Thesis/dissertation, Norwegian
University of Science and Technology. 3.2.8
Hauan S., Ciric A., Westerberg A. and Lien K. (2000a). Dierence points in extractive and reactive
cascades. I. Basic properties and analysis. Chemical Engineering Science 55, 31453159. 3.2.9
Hauan S., Hertzberg T. and Lien K. (1995a). Multiplicity in reactive distillation of MTBE. Computers
and Chemical Engineering 19 (SS), S327S332. 2.9, 2.9.1.2, 2.9.1.2, 2.9.2, 6.1.4, 6.2, 6.2.3.1, 7.2.1,
7.3.1, 7.3.2.1, 8.3.1
Hauan S., Hertzberg T. and Lien K. (1995b). Why methyl ter-butyl ether production by reactive
distillation may yield multiple solutions. Industrial and Engineering Chemistry Research 34, 987
991. 2.9, 2.9.1.1, 2.9.1.2, 2.9.2, 2.9.3, 7.3.1
Hauan S., Lee J.W., Westerberg A. and Lien K.M. (2000b). Properties of sectional proles in reactive
separation cascades. Foundations of Computer-Aided Process Design 96 (323), 397401. 3.2.8
Hauan S. and Lien K. (1996). Geometric visualisation of reactive xed points. Computers and Chemical
Engineering 20 (SS), S133S138. 3.2.8
Heiszwolf J.J., Engelvaart L.B., van den Eijnden M.G., Kreutzer M.T., Kapteijn F. and Moulijn J.
(2001). Hydrodynamic aspects of monolith loop reactor. Chemical Engineering Science 56 (2001),
805812. 2.6, 2.7.1
236
References
Herder P. (1999). Process Design in a changing environment. Thesis/dissertation, Delft University of
Technology. 1.1
Higler A. (1999). A non-equilibrium model for reactive distillation. Thesis/dissertation, Clarkson University. 2.9.3, 4.3
Higler A., Krishna R. and Taylor R. (1999a). Nonequilibrium cell model for packed distillation columns
- the inuence of maldistribution. Industrial and Engineering Chemistry Research 38, 39883999.
2.9.3, 4.3
Higler A., Krishna R. and Taylor R. (2000). Nonequilibrium modelling of reactive distillation: a dusty
uid model for heterogenously catalyzed processes. Industrial and Engineering Chemistry Research
39 (6), 15961607. 2.7, 2.9.3, 4.3
Higler A., Taylor R. and Krishna R. (1998). Modelling of a reactive separation process using a nonequilibrium stage model. Computers and Chemical Engineering 22 (SS), S111S118. 2.9.3, 4.3
Higler A., Taylor R. and Krishna R. (1999b). Nonequilibrium modelling of reactive distillation: mutiple
steady states in MTBE synthesis. Chemical Engineering Science 54, 13891395. 2.9, 2.9.1.2, 2.9.3,
4.3, 7.2.9, 7.4
Huss R.S., Chen F., Malone M.F. and Doherty M. (1999). Computer-aided tools for the design of
reactive distillation systems. Computers and Chemical Engineering 23 (SS), S955S962. 3.2.4
Ismail S., Proios P. and Pistikopoulos E. (2001). Modular synthesis framework for combined separation/reaction systems. AIChE Journal 47 (3), 629649. 3.3.1, 8.3.1
Izquierdo J., Cunill F., Vila M., Tejero J. and Iborra M. (1992). Equilibrium constants for methyl tertbutyl ether liquid-phase synthesis. Journal of Chemical and Engineering Data 37, 339343. B.4,
B.5, B.6
Jacobs R. and Krishna R. (1993). Multiple solutions in reactive distillation for methyl tert-butyl ether
synthesis. Industrial and Engineering Chemistry Research 32 (8), 17061709. 2.1, 2.9, 2.9.1.1,
2.9.1.1, 2.9.1.2, 2.9.1.2, 2.9.2, 6.1.4, 6.2, 6.2.3.1, 7.3.1, 7.3.2.1, 8.3.1
Jimenez L., Wanhschat O. and Julka V. (2000). Synthesis of reactive and extractive distillation units
using distillation line diagrams. Computer-Aided Chemical Engineering 8, 985990. 3.2.2
Jimenez L., Wanhschat O. and Julka V. (2001). Analysis of residue curve maps of reactive and
extractive distillation units. Computers and Chemical Engineering 25, 635642. 4.1, B.2
Jimenez-Gonzalez C., Curzons A.D., Constable D.J.C. and Cunningham V. (2004). Cradle-to-Gate life
cycle inventory and assessment of pharmaceutical compounds. International Journal of Life Cycle
Assessment 9 (2), 114121. 1.4
Jimenez-Gonzalez C., Kim S. and Overcash M.R. (2000). Methodology for developing gate-to-gate life
cycle inventory information. International Journal of Life Cycle Assessment 5 (3), 153159. 1.4
Kenig E., Butzmann F., Kucka L. and G
orak A. (2000). Comparison of numerical and analytical
solutions of a multicomponent reaction-mass-transfer problem in terms of the lm model. Chemical
Engineering Science 55, 14831496. 2.5
Kienle A., Groebel M. and Gilles E.D. (1995). Multiple steady states in binary distillation - theoretical
and experimental results. Chemical Engineering Science 50 (17), 26912703. 2.9
237
References
Kim K.J. and Smith R.L. (2004). Parallel multiobjective evolutionary algorithms for waste solvent
recycling. Industrial and Engineering Chemistry Research 43 (11), 26692679. 1.3.2
Kirk-Othmer (1994). Methyl Tert-Butyl Ether. In: Encyclopedia of Chemical Technology, book chapter 9, (871875), John Wiley and Sons, Inc. B.1, B.2
Kjelstrup S. (2005). Personal Communication. 8.2.1
Kjelstrup S. and Bedeaux D. (2001). Elements of Irreversible Thermodynamics. International Centre
for Applied Thermodynamics, Istanbul. 8.2.1, 8.2.1, 8.2.1
Koeijer G. (2002). Energy ecient operation of distillation columns and a reactor applying irreversible
thermodynamics. Thesis/dissertation, Norwegian University of Science and Technology. 8.2.1, 8.2.1,
8.2.1, 8.2.1, 8.3.3.2
Koeijer G., Kjelstrup S., vander Kooi H.J., Gros B., Knoche K.F. and Andersen T.R. (2002). Positioning heat exchangers in binary tray distillation using isoforce operation. Energy Conversion and
Management 43, 15711581. 8.1, 8.2.1, 8.2.1
Koeijer G. and Rivero R. (2003). Entropy production and exergy loss in experimetal distillation
columns. Chemical Engineering Science 58, 15871597. 8.2.1
Kooijman H. and Taylor R. (2000). The ChemSep book . Libri books on demand, USA. 2.7.1, B.3
Korevaar G. (2004). Sustainable chemical processes and products. Thesis/dissertation, Delft University
of Technology. 1.4, 4.3
Krishna R. (2003). Hardware selection and design aspects for reactive distillation columns. In: Reactive
distillation. Status and future directions (K. Sundmacher and A. Kienle, eds.), book chapter 7, (169
189), Wiley-VCH. 4.2, 4.3
Krishnamurthy R. and Taylor R. (1985). Nonequilibrium stage model of multicomponent separation
processes. AIChE Journal 31 (3), 449456. 2.9.3, 4.1, 4.3
Lakerveld R., Bildea C.S. and Almeida-Rivera C.P. (2005). Exothermic isomerization reaction in reactive ash: steady state behaviour. Industrial and Engineering Chemistry Research 44 (10), 3815
3822. , 7.2.1, 7.2.5, , 7.2.6, 7.2.8
Lee J.W., Hauan S., Lien K. and Westerberg A. (2000a). Dierence points in extractive and reactive cascades. II. Generating design alternatives by the lever rule for reactive systems. Chemical
Engineering Science 55, 31613174. 3.2.9
Lee J.W., Hauan S., Lien K.M. and Westerberg A. (2000b). A graphical method for designing reactive
distillation columns I. The Ponchon-Savarit method. Proceedings of the Royal Society of London
Series A Mathematical Physical and Engineering Sciences 456, 19531964. 3.2.7, 3.2.7
Lee J.W., Hauan S., Lien K.M. and Westerberg A. (2000c). A graphical method for designing reactive
distillation columns II. The McCabe-Thiele method. Proceedings of the Royal Society of London
Series A Mathematical Physical and Engineering Sciences 456, 19651978. 3.2.7, 3.2.7, 3.4
Lee J.W., Hauan S. and Westerberg A. (2000d). Graphical methods for reaction distribution in a
reactive distillation column. AIChE Journal 46 (6), 12181233. 3.2.7
Lee J.W. and Westerberg A. (2000). Visualization of stage calculation in ternary reacting mixtures.
Computers and Chemical Engineering 24, 639644. 3.2.7, 3.2.9, 3.2.9
238
References
Leon S.J. (1998). Linear algebra with applications. Prentice-Hall, New Jersey. 5.4
Li X. and Kraslawski A. (2004). Conceptual design synthesis: past and current trends. Chemical
Engineering and Processing 43, 589600. 1.2.1, 1.3.1, 1.3.2
Lim Y.I., Floquer P., Joulia X. and Kim S.D. (1999). Multiobjective optimization in terms of economics
and potential environment impact for process design and analysis in a chemical process simulator.
Industrial and Engineering Chemistry Research 38 (12), 47294741. 1.3.2, 8.3.4
Luyben W. (2002). Plantwide dynamic simulators in chemical processing and control. Marcel Dekker
Inc., USA. 6.2.3.2
Luyben W.L., Tyreus B.D. and Luyben M.L. (1999). Plantwide process control. McGraw-Hill, New
York. 7.3.2.2, 7.3.2.2, 8.1, 8.2.2, 8.2.4, 8.2.4, 8.6, 8.6, 8.8, 9.2, 9.5, B.4, B.4
Mahajani S. (1999a). Design of reactive distillation columns for multicomponent kinetically controlled
reactive systems. Chemical Engineering Science 54 (10), 14251430. 3.2.4, 3.2.4
Mahajani S. (1999b). Kinetic azeotropy and design of reactive distillation columns. Industrial and
Engineering Chemistry Research 38 (1), 177186. 3.2.4
Mahajani S. and Kolah A. (1996). Some design aspects of reactive distillation columns (RDC). Industrial and Engineering Chemistry Research 35, 45874596. 3.2.4, 3.2.4, 3.2.4, 3.2.4
Maier R., Brennecke J. and Stadtherr M. (2000). Reliable computation of reactive azeotropes. Computers and Chemical Engineering 24, 18511858. 2.4
Malanowski S. and Anderko A. (1992). Modelling phase equilibria - thermodynamic background and
practical tools. John Wiley and Sons, Inc., New York. 2.2, B.3
Malone M.F. and Doherty M.F. (2000). Reactive Distillation. Industrial and Engineering Chemistry
Research 39, 39533957. 1.2.1, 2.1, 2.7, 3.2.2, 3.2.2, 3.3.1, 4.1, 4.3, 8.1, 9.1, B.4, B.4
Marquardt W. (2004). Lectures on Modeling and Analysis of Chemical Process Systems - Delft
University of Technology. Audiovisual Material. 1.3.1, 8.2.1
Meeuse M. (2003). On the design of chemical processes with improved controllability characteristics.
Thesis/dissertation, Delft University of Technology. 1.3.1, 1.4, 8.3.3.2, 8.6, 8.8
Meeuse M. and Tousain R. (2002). Close-loop controllability analysis of process designs: Application
to distillation column design. Computers and Chemical Engineering 26, 641647. 6.2
Melles S., Grievink J. and Schrans S.M. (2000). Optimization of the conceptual design of reactive
distillation columns. Chemical Engineering Science 55, 20892097. 3.2.4, 6.1.3
Mohl K.D., Kienle A., Gilles E.D., Rapmund P., Sundmacher K. and Homann U. (1997). Nonlinear
dynamics of reactive distillation process for the production of fuel ethers. Computers and Chemical
Engineering 21 (SS), S989S994. 2.9.1.1, 2.9.1.1, 2.9.1.2, 2.9.2
Mohl K.D., Kienle A., Gilles E.D., Rapmund P., Sundmacher K. and Homann U. (1999). Steady-state
multiplicities in reactive distillation columns for the production of fuel ethers MTBE and TAME:
Theoretical analysis and experimental validation. Chemical Engineering Science 54, 10291043.
2.9.1.1, 2.9.1.2, 2.9.1.2, 2.9.2, 6.1.4, 7.2.1
239
References
Mohl K.D., Kienle A., Sundmacher K. and Gilles E.D. (2001). A theoretical study of kinetic instabilities
in calatylic distillation processes: inuence of transport limitations inside the catalyst. Chemical
Engineering Science 56, 52395254. 2.9.1.1, 2.9.1.1, 2.9.1.2, 2.9.1.2
Monroy-Loperena R., Perez-Cisneros E. and Alvarez-Ramirez J. (2000). A robust PI control conguration for a high-purity ethylene glycol reactive distillation column. Chemical Engineering Science
55, 49254937. 2.9.1.2, 2.9.2
Moulijn J., Makkee M. and van A.D. (2001). Chemical Process Technology. John Wiley and Sons, Inc.,
London. 1.3.1
Mutalib A. and Smith R. (1998a). Operation and control of dividing wall distillation columns. Part I:
Degrees of freedom and dynamic simulation. Trans. Institution of Chemical Engineers - Part A 76,
308318. 4.2
Mutalib A. and Smith R. (1998b). Operation and control of diving wall distillation columns. Part
II: Simulation and pilot plant studies using temperature control. Trans. Institution of Chemical
Engineers - Part A 76, 319334. 4.2
Nisoli A., Malone M.F. and Doherty M. (1997). Attainable regions for reaction with separation. AIChE
Journal 43 (2), 374387. 3.2.3, 4.2
Nummedal L. (2001). Entropy production minimization of chemical reactors and heat exchangers.
Thesis/dissertation, Norwegian University of Science and Technology. 8.3
Nummedal L. and Kjelstrup S. (2001). Equipartition of forces as a lower bound on the entropy production in heat exchange. International Journal of Heat and Mass Transfer 44, 28272833. 8.3
Okasinski M. (1998). Design methods for kinetically controlled, stagewise reactive distillation columns.
Thesis/dissertation, University of Massachusetts. 3.2.4
Okasinski M. and Doherty M. (1998). Design method for kinetically controlled, staged reactive distillation columns. Industrial and Engineering Chemistry Research 37 (7), 28212834. 3.2.4, 3.2.4, 3.2.4,
3.2.4, 3.2.4, 3.4, 4.3
Okasinski M. and Doherty M.F. (1997). Thermodynamic behavior of reactive azeotropes. AIChE Journal 43 (9), 22272238. 2.4, 4.3, 5.3
Omota F., Dimian A. and Bliek A. (2001). Design of reactive distillation process for fatty acid esterication. Computer-Aided Chemical Engineering 9, 463468. 1.2.1
Omota F., Dimian A. and Bliek A. (2003). Fatty acid esterication by reactive distillation. Part 1:
Equilibrium-based design. Chemical Engineering Science 58, 31593174. 1.2.1
Papalexandri K. and Pistikopoulos E. (1996). Generalized modular representation framework for
process synthesis. AIChE Journal 42 (4), 10101032. 3.3.1, 6.1.3, 8.3.1, 9.5
Pekkanen M. (1995). A local optimization method for the design of reactive distillation. Computers
and Chemical Engineering 19 (SS), S235S240. 3.3, 3.3.1
Pellegrini L. and Possio C.T. (1996). A non-ideal CSTR: a hight codimension bifurcation analysis.
Chemical Engineering Science 51 (11), 31513156. 2.9
Peng J., Edgar T. and Eldridge R.B. (2003). Dynamic rate-based and equilibrium models for a packed
reactive distillation column. Chemical Engineering Science 58, 26712678. 2.9.3
240
References
Peters U., Nierlich F., Schulte-K
orne E. and Sakuth M. (2000). Methyl tert-butyl ether. In: Ullmanns
Encyclopedia of Industrial Chemistry, John Wiley and Sons, Inc. B.1, B.1, B.2, B.2, B.1
Pilavachi P.A., Schenk M., Perez-Cisneros E. and Gani R. (1997). Modeling and simulation of reactive
distillation operations. Industrial and Engineering Chemistry Research 36, 31883197. 2.9.3
Podrebarac G.G., Ng F.T.T. and Rempel G.L. (1998). The production of diacetone alcohol with catalytic distillation. Part II: a rate-based catalytic distillation model for the reaction zone. Chemical
Engineering Science 53 (5), 10771088. 2.9.3
Poth N., Frey T. and Stichlmair J. (2001). MINLP optimization of kinetically controlled reactive
distillation processes. Computer-Aided Chemical Engineering 9, 7984. 3.3, 3.3.1
Proios P. and Pistikopoulos E. (2005). Generalized modular framework for the representation and
synthesis of complex distillation column sequences. Industrial and Engineering Chemistry Research
44 (13), 46564675. 6.1.3, 9.5
PSE (1999). gPROMS advanced user guide. Process Systems Enterprise, London. 6.2.2
Qi Z., Sundmacher K., Stein E., Kienle A. and Kolah A. (2002). Reactive separation of isobutene from
C4 crack fractions by catalytic distillation processes. Separation and Purication Technology 26,
147163. 2.9.1.2
Razon L.F. and Schmitz R.A. (1987). Multiplicities and instabilities in chemically reacting systems - a
review. Chemical Engineering Science 42 (5), 10051047. 7.2.3
Rehnger A. and Homann U. (1990). Kinetics of methyl tertiary butyl ether liquid phase synthesis
catalyzed by ion exchange resin - I. Intrinsic rate expression in liquid phase activities. Chemical
Engineering Science 45 (6), 16051617. 2.9.1.1, B.4, B.5, B.6
Rocha J.A., Bravo J.L. and Fair J. (1996). Distillation columns containing structured packings: a
comprehensive model for their performance. 2. Mass-transfer model. Industrial and Engineering
Chemistry Research 35, 19601967. 2.7.1
Rocha J.A., Bravo J.L. and Fair J. (2004). Distillation columns containing structured packings: A comprehensive model for their performance. 1. Hydraulic models. Industrial and Engineering Chemistry
Research 32, 641651. 2.7.1, 2.7.1
Rodriguez I.E., Zheng A. and Malone M.F. (2001). The stability of a reactive ash. Chemical Engineering Science 56, 47374745. 2.9, 2.9.1.1, 7.2.1, 7.3
Rodriguez I.E., Zheng A. and Malone M.F. (2004). Parametric dependence of solution multiplicity in
reactive ashes. Chemical Engineering Science 59, 15891600. 2.9, 2.9.1.1, 7.2.1, 7.3
Ruiz C.A., Basualdo M.S. and Scenna N.J. (1995). Reactive distillation dynamic simulation. Trans.
Institution of Chemical Engineers - Part A 73 (5), 363378. 7.3.1
Russel B.M., Henriksen J.P., Jorgensen S. and Gani R. (2000). Integration of design and control through
model analysis. Computers and Chemical Engineering 24, 967973. 3.3.3
Scenna N.J., Ruiz C.A. and Benz S.J. (1998). Dynamic simulation of start-up procedures of reactive
distillation columns. Computers and Chemical Engineering 22 (SS), S719S722. 2.9.1.2, 2.9.1.2
Schembecker G. and Tlatlik S. (2003). Process synthesis for reactive separations. Chemical Engineering
and Processing 42, 179189. 1.2.2, 4.1
241
References
Schildhauer T., Kapteijn F. and Moulijn J. (2005). Reactive stripping in pilot plant scale monolith
reactors - application to esterication. Chemical Engineering and Processing 44 (2005), 695699.
2.6
Schneider R. and Marquardt W. (2002). Information technology in the chemical process design life
cycle. Chemical Engineering Science 57, 17631792. 1.4
Schoenmakers H. and Bessling B. (2003a). Reactive and catalytic distillation from an industrial perspective. Chemical Engineering and Processing 42, 145155. 2.8
Schoenmakers H. and Bessling B. (2003b). Reactive distillation process development in the chemical process industries. In: Reactive Distillation. Status and future directions (K. Sundmacher and
A. Kienle, eds.), book chapter 2, (3048), Wiley-VCH. 4.2
Schrans S.M., Wolf S. and Baur R. (1996). Dynamic simulation of reactive distillation: an MTBE case
study. Computers and Chemical Engineering 20 (SS), S1619S1624. 2.9.1.2, 2.9.1.2, 2.9.2, 6.1.4,
7.3.1, 7.3.2.3, 7.3.3
Seader J.D. (1985). The B.C. (Before Computers) and A.D. of equilibrium-stage operations. Chemical
Engineering Education 19, 88103. 2.6
Seader J.D. and Henley E.J. (1998). Separation Process Principles. John Wiley and Sons, Inc., New
York. 6.2, 7.3.2.1, 8.3.1
Seferlis P. and Grievink J. (2001). Optimal design and sensitivity analysis of reactive distillation units
using collocation models. Industrial and Engineering Chemistry Research 40, 16731685. 3.3, 3.3.2
Seider W., Seader J.D. and Lewin D. (1999). Process design principles. John Wiley and Sons, Inc.,
New York. 8.2.1, 8.2.2, 8.2.4
Seramov L.A., Pisarenko Y.A. and Kardona K.A. (1999). Optimization of reactive distillation
processes. Theoretical Foundations of Chemical Engineering 33 (5), 455463. 3.2.1
Seydel R. and Hlavacek V. (1987). Role of continuation in engineering analysis. Chemical Engineering
Science 42 (6), 12811295. 7.2.3
Siirola J.J. (1996a). Industrial applications of chemical process synthesis. In: Advances in Chemical
Engineering. Process Synthesis (J. Anderson, ed.), book chapter 23, (161), Academic Press. 1.2.1,
3.2.2, 4.3, 4.5
Siirola J.J. (1996b). Strategic process synthesis: advances in the hierarchical approach. Computers and
Chemical Engineering 20 (SS), S1637S1643. 3.2.2, 4.3, 4.2, 4.5
Skogestad S. and Postlethwaite I. (1999). Multivariable feedback control: analysis and design. John
Wiley and Sons, Inc., New York. 6.2
Smejkal Q. and Soos M. (2002). Comparison of computer simulation of reactive distillation usign
ASPEN PLUS and HYSYS software. Chemical Engineering and Technology 41, 413418. 4.3
Smith J.M. and van Ness H.C. (1987). Introduction to chemical engineering thermodynamics. McGrawHill, New York. 2.2
Sneesby M.G., Tade M.O. and R. S. (1998a). Multiplicity and pseudo-multiplicity in MTBE and ETBE
reactive distillation. Chemical Engineering Research and Design 76 (4), 524531. 2.9.1, 2.9.1.2, 7.2.1,
7.3.2.3, 7.3.3
242
References
Sneesby M.G., Tade M.O. and Smith T.N. (1998b). Steady-state transitions in the reactive distillation
of MTBE. Computers and Chemical Engineering 22, 879892. 2.9.1.2, 2.9.1.2, 7.2.1, 7.3.2.2, 7.3.2.3,
7.3.3
Solokhin A.V. and Blagov S.A. (1996). Reactive distillation is an advanced technique of reaction process
operation. Chemical Engineering Science 51 (11), 25592564. 3.4
Song W., Huss R.S., Doherty M. and Malone M.F. (1997). Discovery of reactive azeotropes. Nature
388 (7), 561563. 2.4
Stankiewicz A. (2001). Between the chip and the blast furnace. NPT Processtechnologie 8 (2), 1822.
1.2.1
Stankiewicz A. (2003). Reactive separations for process intensication: an industrial perspective. Chemical Engineering and Processing 42, 137144. 1.2.1, 1.2.1, 1.2.2, 1.3.1, 1.3.2
Stankiewicz A. and Moulijn J. (2002). Process intensication. Industrial and Engineering Chemistry
Research 41 (8), 19201924. 1.1, 1.2.1, 1.3.1, 1.3.2
Stichlmair J. and Frey T. (1999). Review: Reactive distillation process. Chemical Engineering and
Technology 22 (2), 95103. 3.2.6, 4.1
Stichlmair J. and Frey T. (2001). Mixed-Integer Nonlinear Programming optimization of reactive distillation processes. Industrial and Engineering Chemistry Research 40 (25), 59785982. 2.9.1.1, 3.3,
3.3.1
Subawalla H. and Fair J. (1999). Design guidelines for solid-catalyzed reactive distillation systems.
Industrial and Engineering Chemistry Research 38 (10), 36963709. 2.6, 3.4, 3.9, 3.4, 3.5, 4.1
Sundmacher K. and Homann U. (1996). Development of a new catalytic distillation process for fuel
ethers via a detailed nonequilibrium model. Chemical Engineering Science 51 (10), 23592368. 2.9.3
Sundmacher K. and Homann U. (1999). Multiple reactions in catalytic distillation processes for the
production of the fuel oxygenates MTBE and TAME: Analysis by rigourous model and experimental
validation. Chemical Engineering Science 54, 28392847. 2.6, 2.9.2, 6.1.4
Taylor R. and Krishna R. (1993). Multicomponent mass transfer . John Wiley and Sons, Inc., New
York. 2.5, 2.6, 4.1, 4.3
Taylor R. and Krishna R. (2000). Review: modelling reactive distillation. Chemical Engineering Science
55, 51835229. 2.1, 2.6, 2.9.3, 4.1, 4.3
Taylor R. and Krishna R. (2003). Modelling of homogeneous and heterogeneous reactive distillation
processes. In: Reactive distillation. status and future directions (K. Sundmacher and A. Kienle,
eds.), book chapter 9, (217240), Wiley-VCH. 4.1, 4.3
Taylor R., Krishna R. and Kooijman H. (2003). Real-world modeling of distillation. CEP Magazine
(7), 2839. 2.9.3
Thiel C., Sundmacher K. and Homann U. (2002). Residue curve maps for heterogeneously catalysed
reactive distillation of fuel ethers MTBE and TAME. Chemical Engineering Science 52 (6), 993
1005. B.4, B.5, B.6
Tuchlenski A., Beckmann A., Reusch D., Dussel R., Weidlich U. and Janowsky R. (2001). Reactive
distillation-industrial applications, process design and scale-up. Chemical Engineering Science 56,
387394. 2.6, 2.7
243
References
Ung S. and Doherty M. (1995a). Calculation of residue curve maps for mixture with multiple equilibrium chemical reactions. Industrial and Engineering Chemistry Research 34, 31953202. 2.4, 3.2.2,
3.2.4, 4.3, 5.3,
Ung S. and Doherty M. (1995b). Necessary and sucient conditions for reactive azeotropes in multireaction mixtures. AIChE Journal 41 (11), 23832392. 2.4, 2.4
Ung S. and Doherty M. (1995c). Vapor-liquid phase equilibrium in systems with multiple chemical
reactors. Chemical Engineering Science 50 (1), 2348. 2.4, 3.2.4, 3.2.4, 4.3, 5.5.3, , B.2
Vadapalli A. and Seader J.D. (2001). A generalized framework for computing bifurcation diagrams
using process simulation programs. Computers and Chemical Engineering 25, 445464. 7.2.3
Venkataraman S., Chan W. and Boston J. (1990). Reactive distillation using ASPEN PLUS. Chemical
Engineering Progress (4554). 4.3
Wang S.J., Wong D.S.H. and Lee E.K. (2003). Eect of interaction multiplicity on control system design
for a MTBE reactive distillation column. Journal of Process Control 13, 503515. (document),
2.9.1.2, 7.3.2.2, 7.16, 7.3.3, 7.3.3
Waschler R., Pushpavanam S. and Kienle A. (2003). Multiple steady states in two-phase reactors under
boiling conditions. Chemical Engineering Science 58, 22032214. 2.9.1.1
Wayburn T.L. and Seader J.D. (1987). Homotopy continuation methods for computer-aided process
design. Computers and Chemical Engineering 11 (1), 725. 7.2.3
Wesselingh J.A. (1997). Non-equilibrium modelling of distillation. Transactions Institution of Chemical
Engineers - Part A 75, 529538. 2.5, 2.6, 2.9.3, 4.3
Wesselingh J.A. and Krishna R. (2000). Mass transfer in multicomponent mixtures. Delft University
Press, Delft. 2.5, 2.5
Westerberg A., Lee J.W. and Hauan S. (2000). Synthesis of distillation-based processes for non-ideal
mixtures. Computers and Chemical Engineering 24, 20432054. 2.3, 3.2.2, 3.2.8, 5.3
Westerberg A. and Wahnschat O. (1996). Synthesis of distillation-based separation systems. In: Advances in Chemical Engineering. Process synthesis (J. Anderson, ed.), book chapter 23, (63170),
Academic Press. 2.3, 4.2
Wolf-Maciel M.R., Soares C. and Barros A.A.C. (2001). Validations of the nonequilibrium stage model
and of a new eciency correlation for non ideal distillation process through simulated and experimental data. Computer-Aided Chemical Engineering 9, 321326. 2.9.3
Yu S., Zhou A. and Tan Q. (1997). Simulation of multistage catalytic stripping with a nonequilibrium
stage model. Computers and Chemical Engineering 21 (4), 409415. 2.9.3
Yuxiang Z. and Xien X. (1992a). Study on catalytic distillation proceses. Part II. Simulation of catalytic
distillation processes - Quasi-homogeneous and rate-based model. Trans. Institution of Chemical
Engineers - Part 70 (A), 465470. 2.6, 2.9.3
Yuxiang Z. and Xien X. (1992b). Study on catalytic distillation processes. Part I. Mass transfer characteristics in catalyst bed within the column. Trans. Institution of Chemical Engineers - Part A 70,
459464. 2.6
244
References
Zhang T. and Datta R. (1995). Integral analysis of methyl tert-butyl ether synthesis kinetics. Industrial
and Engineering Chemistry Research 34 (3), 730740. 5.5.1, B.4
Zheng Y., Ng F.T.T. and Rempel G.L. (2004). A comparison of a pseudo-homogeneous non-equilibrium
model and a three-phase non-equilibrium model for catalytic distillation. Chemical Engineering
Journal 100, 119127. 2.9.3
245
Summary
Designing Reactive Distillation Process with Improved Efficiency economy, exergy loss and responsiveness
Cristhian P. Almeida-Rivera
ISBN 9-090200-37-1 / 9789090200378
November, 2005
The design and operation of reactive distillation processes have become increasingly important within academia and industrial spheres over the last few decades (cf. gure 1
in Malone and Doherty (2000)). Addressing reactive distillation design is a challenging
task both from a process-intensied perspective and because of its inherent complexity. Reactive distillation design is far more complicated than conventional distillation
because of: (i) the presence of additional design decision variables (e.g. distribution of
catalyst in the reactive section); (ii) the nonlinear interaction between thermodynamic
phases and chemical reactions; and because of (iii) strong coupling between heat and
mass ow throughout a reactive distillation unit.
In this dissertation a life-span inspired perspective is taken on the conceptual design of
grassroots reactive distillation processes. Attention was paid to the economic performance of the process and to potential losses of valuable resources over the process life
span.
The research was cast in a set of goal-oriented engineering and specic scientic design
questions. The scientic novelty of this work is based around four key aspects of reactive
distillation process design: (i) the formulation of an extended design problem in reactive
distillation achieved by refreshing it in the wider context of process development and
engineering and in a more relevant way regarding sustainability; (ii) the denition of an
integrated design methodology achieved by analyzing current design methodologies and
bridging the gaps between them; while we suggest this methodology as a way to beat the
design complexity by decomposition, it requires the mastery of many tools and many
concepts; (iii) the improvement of design tools achieved by exploring and extending
current techniques and systematically applying them to the reactive distillation case;
247
Summary
(iv ) the denition of performance criteria that can be used to account for the process
performance from a life-span inspired perspective, as well as applications of them.
After having covered the fundamentals of reactive distillation and current design approaches in the initial chapters of this dissertation, we suggest a novel methodology for
reactive distillation process design: life-span inspired design methodology (LiSp-IDM).
The framework of this design method is the multiechelon design approach, in which a
hierarchy of embedded design spaces of increasing renement is used and the strengths
of current design methods are conjugated. A fundamental understanding of the various echelons of the proposed methodology is obtained through direct application to
an industrial case study (i.e. the catalytic synthesis of MTBE from isobutylene and
methanol in the presence of inert components).
We propose the following design echelons:
Formulation, in which sustainability issues are considered and the basis of design
(i.e. product and feedstock purities ranges, production capacity, operating window and
economics life-span accounting for sustainability aspects) are dened.
Feasibility analysis, in which, using graphical methods, the feasible product compositions for a given set of input variables incorporating all occurring phenomena are
determined.
Temporal structure, in which the mode of operation based on operational skills,
production requirements and supply chain dynamics is selected.
Generation of separation sequences, in which the feasible column sequences are
identied and organized; and their type of task, together with the number of units and
their connectivity are estimated.
Internal spatial structure, in which the process structure obtained from the previous
design echelons is used as the starting point for this design space. Using a multilevel
modeling approach, developing appropriate building model blocks and optimizationbased methods both continuous variables and discrete variables for the process and its
control structure are obtained.
Behavior analysis, in which steady-state simulations are carried out to analyze the
kinetics, gas/liquid mass transfer, hydrodynamics and steady-state multiplicities inside
the units. Additionally, dynamic simulations are performed to check the robustness
of the design, in terms of the ability to maintain product purities and conversion in a
desired range when disturbances occur. At this design echelon the amazing complexity
of the reactive distillation design problem is acknowledged. Even for a simplied and
idealized system (i.e. reactive ash), a large number of bifurcation diagrams, exhibiting
up to three steady states are reported.
248
Summary
Evaluation, in which the design options are checked with respect to their performance. Within the wide range of criteria in safety, health, economics, environment and
technology aspects specic choices are made for economic performance, sustainability
and technology. The capital and operational costs of the process during the time horizon are calculated and the exergy eciency is used as sustainability metric, while the
responsiveness of the process acts as a metric for its capability to deal with disturbances.
A short response time reduces losses from o-spec products.
The specics of each design echelon are described as follows. In the course of this
thesis the feasibility of reactive distillation processes was explored using an improved
residue curve mapping technique. The analysis of the behavior of singular points in
a transformed composition space resulted in dierent reactive separation sequences
according to the feed compositions. The spatial and control design of a chosen superstructure conguration was addressed using a multilevel approach and simultaneous
dynamic optimization of spatial and control structures, respectively. The multilevel
strategy is proposed as being an easy-to-handle tool that can be used to assist one in
the early stages of the design task. Additionally, this approach allows the designer to
determine whether or not the business-related decision variables cope with the operational constraints imposed by the problem formulation. Aspects of steady-state and
dynamic behavior in reactive distillation are also covered in this thesis. We found
that, even for a simplied case study, a large number of bifurcation diagrams is possible, exhibiting a maximum of three steady states. A rather conventional control design
was considered, with the intention to dene a reference design for new developments
later in this research. In this regard, a thermodynamic-based design approach was explored aiming at incorporating life-span considerations in the design problem. Thus,
for evaluation and optimization purposes, a set of performance criteria accounting
for the economic performance, process sustainability and responsiveness was dened.
The entropy production rate was chosen as metric for process sustainability, whereas
response time, as dened by Luyben et al. (1999), was used as a responsiveness index. In fact we accept a simplied version of the response time, considering only one
disturbance rather the worst case disturbance. In particular, process responsiveness
was expressed in the framework of thermodynamics of irreversible processes. Gradually incorporating each performance criterion the design problem was reformulated as
a multiobjective optimization and a suggestion is made that a trade-o exists between
economic and sustainability-based aspects. Designs were compared with respect to
exergy loss and closed loop control performance. The rst design (conventional) was
obtained by optimization of economic performance only; the second design (green) included both economic and sustainability aspects in the formulation of the optimization
problem. Substantial structural dierences were observed between the designs, such as
a reduction in the number of reactive trays, more uneven distribution of catalyst along
249
Summary
the units and the identication of suitable spots for side reactors. Having analyzed
the response times of the bi-optimized designs a suggestion is made that small driving
forces are desirable for reactive sections, while large driving forces are desirable for separation (nonreactive) sections. It is acknowledged that the design comparison might be
of limited generality as only one case study was considered.
The complexity of reactive distillation process design is amazing and even higher than
initially envisaged. Although the LiSp-IDM design approach is our rst attempt towards a systematic design strategy, it requires the mastery of many tools and advanced
concepts. Therefore, there are numerous opportunities to extend and validate this work,
these are briey mentioned below.
Sustainability aspects like proper choice of feedstocks and a products supply chain
should be included as metrics together with a loss prevention concept. The extension
of the proposed design methodology should include more rened behavioral models
and structural models, to allow for superstructure generation. Further study is recommended on the use of the current methodology for reaction systems with more than
four species and multiple reactions. Moreover, it is worth researching on how to link
the outcome of graphical methods for problems of low dimensionality to optimization
design methods with high dimensionality. Regarding steady-state analysis, further
research should focus on the eect of dierent model assumptions (e.g. rate limitations) and more extended systems (e.g. full column with control structure). Regarding
the dynamic behavioral analysis, it is recommended to study further operational
aspects like exibility, as they could become more important than cost aspects. One
topic for further research would be the denition of a more descriptive responsiveness criterion. This new index should be capable of making the distinction between
response on changes of manipulated variables towards controlled outputs and response
to disturbances acting on the process. From a non-equilibirum thermodynamics
angle, more accurate approaches are required when it comes to the prediction of certain
thermodynamic variables. In particular, the estimation of the gradient in chemical potential for multiphase systems needs to be further explored in turbulent mixed systems.
Finally, it is suggested that the LiSp-IDM presented here should be tested in regard
of its practical applicability. The approach presented by Ajah et al. (2004) would
provide a convenient theoretical background for this work.
250
Sammenvatting
Designing Reactive Distillation Process with Improved Efficiency economy, exergy loss and responsiveness
Cristhian P. Almeida-Rivera
ISBN 9-090200-37-1 / 9789090200378
November, 2005
Ontwerp en uitvoering van reactieve distillatie (RD) processen zijn belangrijker geworden in de procesindustrie over de laatste decennia (cf. guur 1 in Malone and Doherty
(2000)). Het ontwerpen van een RD proces is een uitdagende taak, niet alleen vanwege
het bereiken van proces-intensicatie, maar ook vanwege zijn inherente complexiteit.
Bijvoorbeeld, het ontwerpen van RD is veel ingewikkelder dan dat van conventionele
distillatie wegens: (i) het voorkomen van extra ontwerpvariabelen (b.v. distributie van
katalysator in de reactieve sectie); (ii) de niet-lineaire interactie tussen thermodynamische fasen en chemische reacties en (iii) sterke koppeling tussen energie- en massastromen
door een RD eenheid.
Dit proefschrift presenteert resultaten van onderzoek aan het conceptueel ontwerpen
van nieuwe reactieve distillatie processen. Het onderzoek houdt rekening met bepaalde
aspecten van duurzaamheid over de levensduur van het proces. D.w.z., we geven niet
alleen aandacht aan de economische prestaties van het proces, maar ook aan preventie
van potentiele verliezen van andere waardevolle middelen over de levensduur.
Wij geven dit onderzoek vorm door middel van een reeks doelgerichte engineering vragen
en vragen betreende specieke, wetenschappelijke aspecten van ontwerpen. De wetenschappelijke vernieuwing in dit onderzoek betreft vier sleutelaspecten van RD procesontwerp: (i) formulering van een uitgebreid ontwerpprobleem in RD, nieuw geplaatst in
een bredere context van procesontwikkeling en -engineering en met meer relevantie voor
duurzaamheid; (ii) presentatie van een gentegreerde ontwerpmethodologie, verkregen
door analyse van huidige ontwerpmethoden en overbrugging van de hiaten daartussen.
Deze methodologie die we zien als een manier om de complexiteit van ontwerp te verslaan door decompositie, vereist de beheersing van vele concepten en hulpmiddelen die
alle relevant zijn; (iii) verbetering van ontwerphulpmiddelen door onderzoek aan en
251
Sammenvatting
uitbreiding van huidige technieken met een systematische toepassing op een RD case
studie; (iv ) denitie van criteria voor prestaties van een proces, die sommige levensduuraspecten in rekening brengen, alsmede toepassing daarvan.
De grondbeginselen van RD en huidige ontwerpbenaderingen worden behandeld in de
eerste hoofdstukken van dit proefschrift. Vervolgens stellen wij een nieuwe ontwerpmethodologie voor RD processen voor, genaamd LiSp-IDM (life-span inspired design
methodology). De ruggraat van deze ontwerpmethode is de zogenaamde multiechelon ontwerpbenadering, die gebruik maakt van een hierarchie van ingebedde ontwerpruimten van toenemende verjning en tevens de sterke punten van huidige ontwerpmethodes benut. Het fundamentele begrip van de diverse echelons in de voorgestelde
methodologie wordt opgebouwd door directe toepassing op een industriele case studie,
namelijk de katalytische synthese van MTBE uit isobuteen en methanol in aanwezigheid
van inerte componenten.
Wij benoemen de volgende ontwerpechelons:
Probleemformulering, waar wij de basis van het ontwerp vaststellen (systeemgrenzen, zuiverheidbereik van product en grondstoen, productie capaciteit, venster van
toelaatbare procescondities en proceseconomie, gewenste levensduur) rekening houdend
met duurzaamheidaspecten.
Feasibility analyse, waar wij door middel van grasche methodes de reikwijdte
aan productcomposities bepalen die mogelijk is voor een gegeven voedingsamenstelling,
wanneer alle optredende fysische verschijselen worden meegenomen.
Temporele structuur, waar de wijze van procesvoering wordt gekozen, gebaseerd op
operationele vaardigheden, productievereisten en de externe dynamica van de productieketen waarin het proces is opgenomen.
Volgorde van scheidingen, waar wij uitvoerbare schakelingen van processtappen
identiceren en structureren. Voorts bepalen we de fysische taken in de processtappen
samen met het aantal eenheden en de wijze van verbinding.
Interne ruimtelijke kolomstructuur, waar wij de processtructuur, die uit vorige
ontwerpechelons is verkregen, als uitgangspunt gebruiken in deze ontwerpruimte. Door
middel van een modelleringbenadering waarbij op verschillende aggregatieniveaus modellen worden ontwikkeld van de bouwstenen voor de synthese verkrijgen wij zowel continue als discrete beslissingsvariabelen voor een proceseenheid en zijn regeling. Door
inbedding van de synthesemodellen in een optimalisatie raamwerk worden deze variabelen vastgelegd.
Analyse van fysisch gedrag, waar we stationaire simulaties uitvoeren om kinetiek
van reacties, gas-vloeistof massaoverdracht en hydrodynamica alsmede multipliciteiten
van stationaire toestanden binnen de eenheden te analyseren. Bovendien, voeren we
252
Sammenvatting
dynamische simulaties uit om de robuustheid van het ontwerp te controleren, in termen van het vermogen productzuiverheid en omzetting binnen de gewenste marge te
handhaven wanneer zich storingen voordoen. Bij dit ontwerpechelon ondervinden we
de verbazingwekkende ingewikkeldheid van het ontwerpprobleem in RD. Zelfs bij een
vereenvoudigd en gedealiseerd systeem als een reactieve ash zien we een groot aantal
gevarieerde bifurcaties, waarbij tot drie stationaire toestanden kunnen optreden.
Evaluatie, waar de uitgewerkte ontwerpopties met betrekking tot hun prestaties worden bekeken. Binnen de brede waaier van criteria voor safety, health, economics,
environment & technology aspecten maken we een specieke keuze voor economische
prestatie, duurzaamheid en technologie. Bij de economie berekenen wij kapitaal en operationele kosten van het proces over de levensduur. Voorts gebruiken wij de exergy
eciency als maat voor duurzaamheid, terwijl de response tijd van het proces een
maat is voor de snelheid waarmee het proces zich kan herstellen van de gevolgen van
storingen. Een korte response tijd reduceert verliezen door o-spec producten.
De resultaten op elk ontwerpechelon worden aangeduid. In dit proefschrift bepalen
wij de haalbaarheid van RD processen door middel van een verbeterde residue curve
map. De analyse van het gedrag van singuliere punten in een getransformeerde compositieruimte resulteert in verschillende sequenties van reactieve scheidingen afhankelijk
van de voedingssamenstelling. Het ontwerpen van de ruimtelijke indeling en de
regeling van een RD eenheid (kolom) gebeurt binnen een gekozen superstructuur door
middel van een modelbeschrijving op verschillende niveaus van detail en door een simultane dynamische optimalisering van de ruimtelijke structuur en de regeling. Wij zien
deze strategie van modelbeschrijving op verschillende niveaus als een gemakkelijk te
hanteren aanpak voor de beginfase van het ontwerp. Bovendien, staat deze benadering
de ontwerper toe te bepalen of de beslissingsvariabelen die gerelateerd zijn aan de bedrijfseconomie voldoen aan de operationele beperkingen zoals vastgelegd in de formulering
van het ontwerpprobleem. De aspecten van stationair en dynamisch gedrag in RD
zijn ook behandeld in dit proefschrift. Wij rapporteren dat zelfs voor een sterk vereenvoudigd geval een groot aantal bifurcaties mogelijk is, waarbij maximaal drie stationaire
toestanden kunnen voorkomen. Bij de procesregeling is gekozen voor een conventioneel
ontwerp, louter met de bedoeling een referentieontwerp te hebben voor vergelijk met
nieuw te ontwikkelen procesontwerpen. Aangaande het maken van nieuwe ontwerpen
is gekozen voor het uitdrukken van duurzaamheid en robuustheid in het ontwerpprobleem in fysische grootheden uit de niet-evenwicht thermodynamica, zoals exergie en
entropie productie. Aldus, voor evaluatie en optimalisering doeleinden maken we
gebruik van een drietal prestatiecriteria die rekenschap geven van de economie, de duurzaamheid van procesvoering en de robuustheid. Wij kiezen exergie- ecientie als een
maat voor de procesduurzaamheid, terwijl wij een responsetijd - zoals bepaald door
Luyben et al. (1999) - als robuustheidindex beschouwen. In feite nemen we genoe253
Sammenvatting
gen met een vereenvoudigde versie van het begrip responsetijd, waarbij we slechts een
enkele dominante verstoring meenemen in plaats van de meest slechte verstoring. Door
stapsgewijs deze prestatiecriteria te introduceren wordt het ontwerpprobleem vertaald
in een multi-objective optimalisatie probleem, waarbij een Pareto afweging kan worden
gemaakt tussen economische en duurzaamheidgebaseerde prestaties. We vergelijken
procesontwerpen met betrekking tot exergy verlies en de robuustheidprestatie van
het geregelde proces. Het eerste, conventionele ontwerp wordt verkregen door alleen
maar te optimaliseren naar de economische prestatie, terwijl het tweede, groene ontwerp
zowel economische als duurzaamheidaspecten in de formulering van het optimaliseringprobleem omvat. We vinden wezenlijke, structurele verschillen tussen deze ontwerpen,
zoals de vermindering van het aantal reactieve schotels, meer ongelijke verdeling van de
katalysator over de eenheden en de andere geschikte plekken voor zijreactoren. Door
de responsetijden van de bi-geoptimaliseerde ontwerpen te analyseren concluderen wij
dat kleine drijvende thermodynamische krachten in de reactieve secties wenselijk zijn,
terwijl grote drijvende krachten voor scheidingsecties wenselijk zijn. We tekenen hierbij
aan dat de uitkomst van de vergelijking van deze ontwerpen een beperkte geldigheid
kan hebben omdat we slechts n geval hebben onderzocht.
De ingewikkeldheid van RD procesontwerp is verbazend hoog; veel hoger dan aanvankelijk ingeschat. Hoewel onze LiSp-IDM ontwerpbenadering onze eerste poging is tot een
systematische ontwerpstrategie, vereist het reeds de beheersing van vele hulpmiddelen
en geavanceerde concepten. Daarom zijn de kansen om dit werk uit te breiden en te
valideren talrijk. Ze worden kort vermeld.
Wij geloven dat duurzaamheidaspecten breder kunnen worden behandeld zoals de
keuze van geschikte grondstoen en de plaatsing van het RD proces in de gehele productieketen, tezamen met het concept van preventie van verliezen in het proces zelf.
De voorgestelde ontwerpmethodologie zou bij uitbreiding niet alleen meer verjnde
gedragsmodellen moeten bevatten, maar ook de formele uitwerking van structuur modellen, passend bij een superstructuur voor het ontwerp. Verder wordt aanbevolen de
toepasbaarheid van de huidige methodologie op reactiesystemen met meer dan vier componenten en meerdere reacties te bestuderen. In verband hiermee is het de moeite waard
na te gaan hoe het resultaat van grasche methodes voor problemen van lage dimensionaliteit kan worden verbonden met optimalisatie gebaseerde ontwerptechnieken voor
problemen met hoge dimensionaliteit. Betreende analyse van stationair gedrag zou
het verdere onderzoek zich kunnen concentreren op het eect van verschillende modelveronderstellingen (b.v. beperkingen in overdrachtssnelheden) en op uitbreidingen
van het systeem (b.v. volledige kolom met regelstructuur). Betreende analyse van
dynamisch gedrag adviseren wij om operationele aspecten zoals exibiliteit verder
te bestuderen, aangezien zij in belang kunnen toenemen ten opzichte van de kosten
aspecten. Langs die lijn past ook verder onderzoek aan een verjning van de denitie
254
Sammenvatting
van response tijd. Deze nieuwe index zou onderscheid moeten kunnen maken tussen
reacties op veranderingen van gemanipuleerde variabelen naar geregelde output variabelen en reacties op storingen die inwerken op het proces. Binnen het kader van het
gebruik van de niet-evenwicht thermodynamica zijn nauwkeurigere benaderingen
vereist voor de voorspelling van bepaalde thermodynamische variabelen. In het bijzonder behoeft de schatting van de gradient in de chemisch potentiaal in een turbulent
gemengd meerfasen systemen nader onderzoek.
Tot slot stellen wij voor onze ontwerpmethode, LiSp-IDM, verder te testen op praktische bruikbaarheid. De benadering die door Ajah et al. (2004) werd voorgesteld verschaft
een geschikt theoretische raamwerk om zulke testen uit te voeren.
255
Acknowledgments
This research thesis compiles the upmost results of a 4-year period of intense and
challenging work. Independently on the nature of their contribution (intellectual or/and
emotional), many people have pushed toward the completion of this work. To all of
them I am deeply and sincerely indebted.
My rst words of thanks go, sincerely, to my promotor, Prof. Johan Grievink. I thank
you Johan for your continuous guidance and scientic insight during these 4+ years
and, on top of this, for your admirable kindness and politeness. I fully enjoyed the long
hours we spent together, discussing a wide range of topics and always with the pleasant
companion of sucient quantities of loose leaves tea. Furthermore, I highly appreciate
the scientic freedom you provided me, which enabled me to explore numerous topics
within the technical knowledge domain. Despite of this enjoyable scientic freedom,
you kindly put me back on track when my compass was functioning wrongly or when
my writing was turning to be baroque and fuzzy. Your helicopter view was very much
appreciated and valuable for the completion of this dissertation. Without all your
comments and suggestions this thesis probably would have remained as a bunch of
unconnected manuscripts. Additionally, I am very much thankful for allocating time
in your tight agenda to discuss and repeatedly and repletely correct all the dissertation
chapters.
I am in particular indebted to Peter Verheijen for always being aware of my well-being
before, during and after this research. Many thanks Peter for your suggestions while I
was in the midst of a PhD quest. Actually, it was in your oce where all this scientic
journey started. My thanks are extended to Pieter Swinkels and C. Sorin Bildea for their
contributions to this research. Your scientic insight, practical expertise and fruitful
discussions were successfully materialized in various publications.
I would like to express my thanks to all my PSE-colleagues for their companion, coeebreak discussions, scientic criticism and common quest for knowledge. Special thanks
go to those fellows, with whom I shared the AIO-kamer at the end of the corridor:
Austin Ajah, Gijsbert Korevaar, Marco van Goethem, Michiel Meeuse, Michiel van
Wissen, Pieter Schmal and my reactive-corner fellow Sachin Singare. These thanks
257
Acknowledgments
are extended to my partly-PSE colleagues: Abilash Menon and Harish Goel. I am in
particular indebted to Harish for the last-minute tips regarding the nal polishing of
the manuscript.
This dissertation would not have been materialized without the eort and (in-)direct
contribution of several graduate students. Many thanks to Arjan van Benthem for
setting-up the basis of a RD column model and for enabling my rst exposure to
c
c
gPROMS
/gOPT
, to Vincent van Hijkoop for the discussions on reactive azeotropy,
to Tymon Vermeulen for his extremely hard work in the dynamic simulation of the
MTBE RDC and professional commitment to achieve the assignments ultimate goal, to
Gilles Camaret for his enthusiasm despite a non-optimal scenario, to Richard Lakerveld
for successfully mastering one of the complexities in RD (MSS) and nicely conveying the
project output into a scientic article, to all the graduate students of the CPD groups
3268, 3285, 3296 and 3310 for their technical and creativity discussions.
Many thanks go to Mart Valentijn for the technical discussions we had regarding owsheeting and continuation methods and to Caroline Monna for her assistance and keeping our agendas in good shape.
Many people provided valuable comments to this research and contributed (in-)directly
towards a broader scientic added value. The following researchers are sincerely acknowledged: Prof. Michael Doherty (UCSB, USA), Prof. Jan Harmsen (TUDelft and
Shell Chemicals, The Netherlands), Dr. Eugene Kenig (University of Dortmund, Germany), Prof. Signe Kjelstrup (NTNU Trondheim, Norway), Prof. Jae Woo Lee (UCNYC, USA), Prof. Wolfgang Marquardt (RWTH-Aachen, Germany), Prof. Stratos
Pistikopoulos (Imperial College, UK) for his hospitality during my stay at the CPSEImperial College, Dr. Tylman Schildhauer (ex TUDelft, The Netherlands), Prof. Hartmut Schoenmakers (BASF, Germany), Dr. Panos Seferlis (CPERI, Greece) and Dr.
Je Siriola (Eastman Chemicals, USA).
Many thanks to my Ecuadorian fellows in the Netherlands for allowing the much-needed
emotional balance for one in a foreign land. Thanks to you all: Fam. Guerra-Torres,
Fam. LeMarie-Yepez and Maria Luisa Martinod. I shall not fail thanking all the people
that -in one way or another- contributed in making life outside university boundaries so
much fun. In particular, many thanks to Ana Maria Garavito, Marc de Reuver, Fabian
Zhindon, Yenory Morales, Pilar Garate and Nacho Guerra, among others. Special
thanks go to Fr. Ben Engelbertink for the spiritual guidance, the dot-painting and
those Thursdays of webdesign.
Furthermore, I am indebted to Ann OBrien for her unconditional and continuous support for the last 6+ years and to Miranda Aldham-Breary, Grace and Eric Eccarius for
their help in improving the technical writing of this thesis.
Moreover, I would like to thank my colleagues of the Process and Supply Chain Design
258
Acknowledgments
group in Unilever Food and Health Research Institute (Vlaardingen, The Netherlands)
for the challenging and stimulating working environment. Special thanks go to Clive
Marshman, Ardjan Krijgsman and Peter Bongers for their understanding and support
during the last and determining steps in this promotie-process.
Deep in my heart I thank my parents, brothers and sister for their everyday support
and trust in me. Papi and Mami, I owe you the most valuable lesson one can receive: be
responsible and work hard. But on top of this, I am fully indebted for all your prayers,
blessings and tender care during the last 3+ decades. A Thank You goes to my siblings
Fernando, Doris and Hern
an for all the joyful moments that we have shared, those that
in a way or another have contributed to a balanced emotional quotient. Furthermore, I
would like to thank my parents- and siblings-in-law for their trust and understanding.
Visiting periodically my family allowed undoubtedly a synergic enhancement of my
scientic performance.
Last and not least I would like to express my most sincere acknowledgement to my wife
Paty and our beloved daughter Lucia. Thanks to you, Paty, for your understanding
and unconditional support during the moments of deep uncertainty, for your words of
condence when the light at the end of the tunnel seemed to vanish, for the quality of
your love, for your criticism and reection towards my sometimes awkward thoughts,
for always believing in me despite my human failures, for reminding me every now and
then what life is all about and for sharing with me the most precious gift of life: our
daughter. My beloved Luca, I thank you for all the joyful and cheering-up moments
we spend together, for you laughing, jumping and singing in my dreams, for your rst
words, steps and kisses, for always receiving me with a big hug and calling me papito.
But, on top of these, thank you my dearest Luca for allowing me to experience the
endless love that a father can give to his child.
Finally, I thank my God almighty for His blessings and providing me with the strength
and rmness to succeed.
CPAR/October 25, 2005
259
Selected publications
Almeida-Rivera C.P. and Grievink J. (2001). Reactive distillation: on the

development of an integrated design methodology. Chemie Ingenieur
Technik 73(6), 777.
Almeida-Rivera C.P. and Grievink J. (2002). Feasibility of equilibrium controlled reactive distillation processes: application of residue curve
mapping. Computer-Aided Chemical Engineering 10, 151-156.
Almeida-Rivera C.P., Swinkels P.L.J. and Grievink J. (2003). Multilevel modeling of spatial structures in reactive distillation units. ComputerAided Chemical Engineering 15 (Part A & B), 696-701.
Almeida-Rivera C.P. and Grievink J. (2004). Feasibility of equilibrium controlled reactive distillation process: application of residue curve
mapping. Computers and Chemical Engineering, 28(1-2), 17-25.
Almeida-Rivera C.P., Swinkels P.L.J. and Grievink J. (2004). Economics and
exergy efficiency in the conceptual design of reactive distillation
processes. AIChE Symposium Series 2004, 237-240.
Almeida-Rivera C.P., Swinkels P., Grievink J. (2004). Designing reactive distillation processes: present and future. Computers and Chemical Engineering, 28(10), 1997-2020.
Lakerveld R., Bildea C.S. and Almeida-Rivera C.P. (2005). Exothermic isomerization reaction in a reactive flash: steady state behavior. Industrial Engineering Chemistry Research, 44(10), 3815-3822.
Top
ranked (2nd in March 2005) in the list of TOP25 Hottest articles in Computers and Chemical
Engineering (http://top25.sciencedirect.com); listed since March 2005 until now (November 2005)
261
About the author
Cristhian P. Almeida-Rivera was born in 1972 in Quito-Ecuador. In 1990 he completed

his high-school studies and graduated from the German School (Quito) majoring in
Physics and Mathematics, with distinction and best grade average. In the same year
he started his university education in the Escuela Politecnica Nacional (Quito) and in
1997 he received a degree in chemical engineering with honors (cum laude) and best
grade average. In 1998 he joined the Master of Science International Programme at
Delft University of Technology (The Netherlands). Two years later he was awarded a
degree with honors (cum laude) majoring in Particle Technology and Chemical Risk
Management. Moreover, Cristhian was valedictorian of his class. During his MSc
studies he spent a 3-month internship at the Process Engineering Department of the
Eidgen
ossiche Anstalt f
ur Wasserversorgung, Abwasserreinigung und Gewasserschutz
(EAWAG) in Z
urich-Switzerland. In 2000 Cristhian was appointed as PhD student for
the Process Systems Engineering group at DelftChemTech, Faculty of Applied Sciences,
Delft University of Technology under the supervision of Prof. Johan Grievink. In
2002 Cristhian spent a 6-week visit at the Centre for Process Systems Engineering
of Imperial College of Science and Technology in London-UK under the supervision
of Prof. Stratos Pistikopoulos. The results of Cristhians PhD research have been
presented locally and internationally in CACE-symposia and congresses (e.g. NPS24, NL-GUTS, IROP, DCSIP, ISMR-2, ESCAPE12, PSE2003 and FOCAPD2004) and
published in refereed scientic journals. For the last four years, he was additionally
involved in the supervision of undergraduate students during their graduation projects
(4), 3-month Conceptual Process Design assignments (20), research assignment (1) and
literature review assignments (2).
Since January 2005 Cristhian works as process/research engineer at Unilever Food
and Health Research Institute in Vlaardingen (The Netherlands) in the department of
Process and Supply Chain Design. Cristhian was in the deans list (Escuela Politecnica
Nacional, 1992-1997), was a UFP fellowship holder (1998-2000), has been student representative (1995-1997), is a soccer fan and dot-painter, has been married with Patricia
since 1998 and in March 2003 became the father of Luca.
October 25, 2005
263
Index
A
AR, attainable region, 46
autonomous reside curve expression, 46
B
bifurcation diagram, 132
bifurcation pressures, 95
boundary curve, 95
bow-tie region method, 95
C
catalytic distillation, see reactive distillation
catalytic packings, 28
chemical process design, 77
CMO, constant molar overow, 38
codimension-1 bifurcation, 132
codimension-2 bifurcation, 134
column structures, 26
continuation methods, 49
CPD, conceptual process design, 6
cradle to grave, 9
D
Damk
ohler number, 24
dierence point, 58
distillation line, 18
dynamic behavior, 112
E
elbow region, 134
equilibrium controlled chemical reactions, 49
F
feasibility boundary, 132
lm model, 24
xed points, 49
fold, 132
H
hydrodynamic model, 29
hysteresis, 134
K
kinetically controlled reactions, 46
L
law of mass action, 24
life span, 9
limit point, 132
LiSp-IDM, life-span inspired design methodology,
9
LMB, line of mass balance, 43
LSS, limiting steady state, 44
M
McCabe-Thiele method, 55
MSE, Maxwell-Stefan equation, 24
MSS
lift-dilution eect, 32
MSS, multiple steady states, 31
multiechelon design approach, 12, 71
Multiple steady state
reactive ash, 129
multipurpose mass/heat transfer module, 62
O
objective function, 61
optimization constraints, 61
P
packings, see catalytic packings
Pareto curve, 176
phase equilibrium, 16
phenomena vector, 58
Ponchon-Savarit method, 55
potential reactive azeotrope, 20
PRAz, pseudo-reactive azeotrope, 99
PSE, process systems engineering, 5
R
rate-limited mass transfer phenomenon, 24
RCM, residue curve mapping, 44
RD, reactive distillation, 8
reaction invariant space, 94
265
Index
reactive azeotrope, 20
reactive cascade dierence point, 56
reactive distillation
heterogeneous, 27
homogeneous, 26
line, 19
regions, 95
separatrices, 95
reactive rectifying, 31
reactive stripping, 31
reside curve, 18
responsiveness criteria, 83
S
SHEET, safety, health, environment, economics,
technology, 75
singular points, 132
stoichiometric line, 17
structure model, 168
T
thermodynamic design factor, 187
transformed composition, 21
turning point, 132
U
utopia point, 176
266
List of Symbols
Acronyms
AR
attainable region
BD
bifurcation diagram
CME
chemical equilibrium manifold
CMO
constant molar overow
CPD
conceptual process design
GLRDVE
generic lumped reactive distillation volume element
LiSp-IDM
life-span inspired design methodology
LMB
line of mass balance
LSS
limiting steady state
MESH
mass balance, equilibrium, summation, energy balance
MIDO
mixed integer dynamic optimization
MINLP
mixed integer nonlinear programming
MNAz
methanol-nC4 azeotrope
MSE
Maxwell-Stefan equation
MSI
Marshall and Swift index
MSS
multiple steady state
MTBE
methyl t-butyl ether
NEQ
non-equilibrium
POT
pay-out time
PRAz
pseudo reactive azeotrope
PSE
process systems engineering
PU
price per unit
QAz
quaternary azeotrope
RCM
residue curve mapping
RD
reactive distillation
267
List of Symbols
RTD
residence time distribution
SHEET
safety, health, environment, economics, technology criteria
SISO
single input single output
SP
set point
SS
steady state
TAC
total annualized cost
Universal constants
R
universal gas constant
gravitational acceleration
Greek symbols
thermodynamic design factor
owrate disturbance
constant for vapor inow
ratio of characteristic times in the vapor and liquid sides
geometry parameter of weir
pressure drop
Tlm
log-mean temperature dierence
dierence operator
rect
R,n
composition coordinate in the Ponchon-Savarit diagram for rectifying

section
r H
heat of chemical reaction
v H
heat of vaporization
relative volatility
dimensionless activation energy
numerical tolerance
scaled time
II
second law eciency
activity coecient
dimensionless heat capacity
shadow price
ij
binary interaction parameter for the pair i j
Vref
matrix of stoichiometric coecients for reference components
()
i
chemical potential of component i in phase ()
rate of momentum gain by viscous transfer per unit volume
268
List of Symbols
total
sum of the stoichiometric coecients
stoichiometric coecient of component i
acentric factor
correction factor accounting for system nonideality
fugacity coecient of component i
mass-based vapor fraction
ratio between phases
R,j
fraction of feed vaporized in the unit j in a reactive cascade
()
pressure tensor in phase ()
energy ux associated with the interface mass and heat transfer
F R
functions for the liquid and vapor ow regimes
P P
physical properties functions for the liquid and vapor phases
molar density of component i
m,i
mass density of component i
fast moving disturbance
slow moving disturbance
()
V
entropy production rate in phase ()
()
entropy production rate per unit of volume in phase ()
response time
controller reset time
warped time
dimensionless temperature
vapor fraction in reactive ash
extent of reaction k; degree of advancement of chemical reaction k
dimensionless ux
total friction factor
dimensionless height of the column
i,j
friction coecient between species i and j
total
vector of total mole number change in each reaction
vector of stoichiometric coecients for component i
vector of stoichiometric coecients
Roman symbols
x
deviation from specications
269
List of Symbols
length of the mixing vector
scalar
matrix of average mass transfer rates
frictional force
vector of unknowns
Maxwell-Stefan diusivity
species ux
Xvariables
set of all involved variables
Ytotal
set of algebraic equations
vector of external forces per molar unit
Jq ()
measurable heat ux of phase ()
()
ji
molar diusion ux vector
row vector of draw-o heat rates
q()
multicomponent energy ux through the element interphase relative to

the mass average velocity in phase ()
RS(xi , yi )
reaction-separation term
v()
uid velocity vector in phase ()
vector of liquid molar compositions
zfeed ,
molar fractions vector for feed, side draw-o streams, respectively
z()
vector of molar fractions in phase ()
AL
i
anity of chemical reaction i
normalized Damk
ohler number
overall energy balance
MC
component molar balances
availability
availability function
objective function scaled with respect to utopia point
()
Hi
partial molar enthalpy of component i in phase ()
interfacial molar ux of component i
utopia point
()
total entropy in phase ()
total internal energy
integer variables
F j,V
total ow of stream j in the vapor phase
270
List of Symbols
A
heat exchange area
A(z)
cross sectional area at distance z
interfacial area
Aactive
active area
Aholes
holes area
ai
activity of specie i
Ai , Bi , Ci
constants for component i in Antoine expression
aij , bij
constants for the pair i j in Wilson model
adiabatic temperature rise
Bv
volumetric bottoms owrate
Bias
controller bias
monetary cost
c0 c6
constants for hydrodynamic and chemical equilibrium
Ctotal
average concentration
cpai cpdi
coecients for heat capacity of component i
Cp
heat capacity
distillate ow
d()
set of disturbance variables
di
tube internal diameter
Dv
volumetric distillate owrate
Da
Damk
ohler number
dni
change in mole numbers of component i
electric eld
controller error
Ea
activation energy
molar owrate
dimensionless feed
f
normalized objective function
f (), g()
dierential and algebraic expressions respectively
F feed ,
molar rates of feed and side draw-o streams, respectively
F side
fecon
economic objective function
fexergy
sustainability objective function
food
hydrodynamic constant
271
List of Symbols
fresponse
responsiveness objective function
Fc
factor related to material of construction
fi0
reference state fugacity of component i
()
fi
fugacity of component i in phase ()
fk
nonlinear function of the distribution coecients
Fm
mass-based feed ow
Ff ac
scale-up factor
ff r
two-phase friction factor
fm2f t , fmol2kg , fs2y
conversion factors
fpd
friction constant for pressure drop
Gibbs free energy
j,()
gi
external force exerted on component i per molar unit in stream j and

phase ()
gj ()
driving force of chemical reaction j based on activities
hrect
R,n
enthalpy coordinate in the Ponchon-Savarit diagram for rectifying section
()
hHX
heat transfer coecient in phase ()
hV
f,i
enthalpy of formation of component i in the vapor state
ht
height
htspace
space between stages/compartments
InS
controller input
electric current density
i,()
Je
total ux of energy
()
Ji
diusion ux of component i in phase ()
i,()
Jm
mass ux
JqL
measurable heat ux into the liquid phase
()
Js
entropy ux in phase ()
controller gain
Keq
temperature-dependent chemical equilibrium constant
kf
forward reaction rate cosntant
kf,0
forward reaction rate constant evaluated at a reference temperature T0
ki,m
mass transfer coecient of component i in mixture m
dimensionless liquid owrate
lPFR
length of PFR
272
List of Symbols
Lv
volumetric reux owrate
M ()
holdup of phase ()
M0L
initial mass of liquid in the pot
MW
molecular weight; averaged molecular weight
total number of moles
nfeed
number of feed streams
no
number of side draw-o streams
nc
number of components
()
ni
number of moles of component i in phase ()
no
number of objective functions
np
number of phases in the system
nq
number of heat streams
nrx
number of chemical reactions
nst
number of streams
absolute pressure
P
pp ,
pc
sets of time-varying variables
Pc,i
critical pressure of component i
dimensionless heat duty

q ( )
heat ux from phase ( ) to phase ()
Qcond
condenser duty
Qreb
reboiler duty
composition-dependent reaction rate
RR,RS
reux and reboil ratios
molar entropy
i,()
Sj
molar entropy of component j in stream i and phase ()
SW
auxiliary variable for phase-equilibrium model
temperature
Tb
boiling temperature
TQ
temperature of the heat source/sink
Tc,i
critical temperature of component i
molar internal energy
internal energy density
u()
set of input variables
273
List of Symbols
U HX
global heat transfer coecient
uood
ooding velocity
udi
i
diusive velocity of specie i
dimensionless vapor owrate
V0
initial vapor ow
vi
molar volume of component i
Vm
mass-based vapor ow
work
vector of weighting factors
Wcat
amount of catalyst
liquid phase composition
x()
set of state variables
Xi ,Yi
transformed composition variables for component i in the liquid and

vapor phases, respectively
Xrx,j , Xq , Xe , Xm,j , X HX
conjugated driving forces for chemical reaction, heat exchange between

phases, interfacial heat diusion, interfacial mass diusion of component j and heat exchange with the external world, respectively
vapor phase composition
y()
set of output variables

length of section
z
zp,
zc
sets of time-invariant parameters
zri
constant of component i
molar enthalpy without heat of formation
nC4
n-butane
iC4
isobutylene
cP
product coecient vector
cR
reactant coecient vector
f (), g(), h()
sets of dierential, algebraic equality and inequality equations, respectively
i()
set of initial conditions
J ()
objective function
level
level
MeOH
methanol
Subscripts and superscripts

0
274
standard state
List of Symbols
cat
catalyst
col
column
ref
reference component; reference conditions
shell
shell
stack
tray stack
trays
trays
liquid phase
mass-based values
out
outlet
reux stream
entropy-related value
vapor phase
in
inlet
cond
condenser
cool
cooling utility
cu
cool utility
downc
downcomer
feed
feed stream
heat
heating utility
hu
hot utility
reb
reboiler
rect
rectifying section
strip
stripping section
tubes
condenser tubes
weir
weir
275
Colophon
This dissertation was typeset by CPAR on a Pentium IV workstation using the latest version of LATEX and the compiler e-TEX (http://www.MIKTEX.org). The input
les were typed and edited using WinEdt (http://www.WinEdt.com) and translated
into PostScript using dvips. Figures were generated with Microcal Origin and Matlab.
Drawings and plots (.eps) were produced with Macromedia Flash and Visio. The postscript document was converted into portable digital le (.pdf) using Adobe Distiller
with a resolution of 2400 dpi. Times package (times.sty) was used to produce the family
of roman fonts. The bibliography entries were produced with Reference Manager and
its bibtex.os output style. The natbib.sty package was used to produce the BibTEX bibliography le in combination with an authors customized bibliography style le (.bst).
The interface makebst.tex was particularly helpful in the generation of the .bst style le.
The index was generated from LATEX .idx le using MakeIndex.
The design of the cover represents the owsheet of a reactive distillation process for the
synthesis of MTBE.
Finally, this dissertation was printed by PrintPartners Ipskamp (http://www.ppi.nl)
in Enschede-The Netherlands with a production volume of 200 copies.
277

As Almeida 20051114 PDF

Uploaded by

Copyright:

Available Formats

As Almeida 20051114 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

As Almeida 20051114 PDF

Uploaded by

Copyright:

Available Formats

Designing Reactive Distillation

Processes with Improved Eciency

Designing Reactive Distillation

ter verkrijging van de graad van doctor

Dit proefschrift is goedgekeurd door de promotor:

c 2005 by Cristhian P. Almeida-Rivera, Delft

Printed by PrintPartners Ipskamp in the Netherlands

A Changing Environment for the Chemical Process Industry . . . . . .

Reactive Distillation Potential . . . . . . . . . . . . . . . . . . . . . . . .

Signicance of Conceptual Design in Process Systems Engineering . . .

Outline and Scientic Novelty of the Thesis . . . . . . . . . . . . . . . .

2 Fundamentals of Reactive Distillation

One-stage Level: Physical and Chemical (non-) Equilibrium . . . . . . .

Multi-stage Level: Combined Eect of Phase and Chemical Equilibrium

Multi-stage Level: Reactive Azeotropy . . . . . . . . . . . . . . . . . . .

Non-equilibrium Conditions and Rate Processes . . . . . . . . . . . . . .

Distributed Level: Column Structures . . . . . . . . . . . . . . . . . . .

Distributed Level: Hydrodynamics . . . . . . . . . . . . . . . . . . . . .

Flowsheet Level: Units and Connectivities . . . . . . . . . . . . . . . . .

Flowsheet Level: Steady-State Multiplicities . . . . . . . . . . . . . . . .

2.10 Summary of Design Decision Variables . . . . . . . . . . . . . . . . . . .

3 Conceptual Design of Reactive Distillation Processes: A Review

Interactions between Process Development and Process Design . . . . .

Structure of the Design Process . . . . . . . . . . . . . . . . . . . . . . .

Life-Span Performance Criteria . . . . . . . . . . . . . . . . . . . . . . .

Multiechelon Approach: The Framework of the Integrated Design Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5 Feasibility Analysis and Sequencing: A Residue Curve Mapping Approach

Input-Output Information Flow . . . . . . . . . . . . . . . . . . . . . . .

Residue Curve Mapping Technique . . . . . . . . . . . . . . . . . . . . .

Feasibility Analysis: An RCM-Based Approach . . . . . . . . . . . . . .

Case Study: Synthesis of MTBE . . . . . . . . . . . . . . . . . . . . . .

Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

6 Spatial and Control Structure Design in Reactive Distillation

Multilevel Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

Simultaneous Optimization of Spatial and Control Structures in Reactive

Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

7 Steady and Dynamic Behavioral Analysis

Steady-State Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

Dynamic Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

Generic Lumped Reactive Distillation Volume Element . . . . . . . . . . 158

Integration of Volume Elements to a Column Structure . . . . . . . . . . 168

Application 2. Bi-Objective Optimization of a MTBE Reactive Distillation Column . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

Comparison Between Classical and Green Designs . . . . . . . . . . . . 189

Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191

9 Conclusions and Outlook

Conclusions Regarding Specic Scientic Design Questions . . . . . . . . 194

Conclusions Regarding Goal-Oriented Questions . . . . . . . . . . . . . 200

Scientic Novelty of this Work . . . . . . . . . . . . . . . . . . . . . . . 201

Outlook and Further Research . . . . . . . . . . . . . . . . . . . . . . . 205

A Model Description and D.O.F. Analysis of a RD Unit

A.1 Mathematical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209

B.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

About the author

Schematic representation of the conventional and highly task-integrated

Schematic representation of the relevant spatial scales in reactive distillation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Representation of stoichiometric and reactive distillation lines . . . . . .

Graphical determination of reactive azeotropy . . . . . . . . . . . . . . .

Phase diagram for methanol in the synthesis of MTBE expressed in terms