You are on page 1of 337

METODOLOGA PARA EL DISEO

RIGUROSO DE PROCESOS QUMICOS


MODELOS HBRIDOS:
SIMULACIN-OPTIMIZACIN

Miguel ngel Navarro Amors


METODOLOGA PARA EL DISEO
RIGUROSO DE PROCESOS QUMICOS
MODELOS HBRIDOS:
SIMULACIN-OPTIMIZACIN

Miguel ngel Navarro Amors


ESCUELA POLITE CNICA SUPERIOR

INSTITUTO DE INGENIERIA DE LOS PROCESOS QUIMICOS

PHD THESIS:

METODOLOGIA PARA EL DISENO


RIGUROSO DE PROCESOS

QUIMICOS. MODELOS HIBRIDOS: SIMULACION-OPTIMIZACI
ON


Tesis presentada por Miguel Angel para el grado
Navarro Amoros
de Doctor en Ingeniera Qumica en la Universidad de Alicante

Dirigido por:
Antonio Caballero Suarez
Jose
Ruben
Jose Ruz Femenia
D. JOSE ANTONIO CABALLERO SUAREZ,

Catedratico de Ingenieria Qumica de la
Universidad de Alicante y

D. JOSE RUBE N RUIZ FEMENIA, Ayudante Doctor de Ingeniera Qumica de la Univer-


sidad de Alicante

CERTIFICAMOS:


Que D. MIGUEL ANGEL Ingeniera Qumica, ha realizado bajo
NAVARRO AMOROS,
en el Instituto de Ingeniera de Procesos Qumicos, de la Universidad
nuestra direccion,
de Alicante, el trabajo que con el titulo METODOLOGIA PARA EL DISENO
RIGUROSO

DE PROCESOS QUIMICOS. MODELOS HIBRIDOS: SIMULACION-OPTIMIZACI
ON constitu-
ye sus memoria para aspirar al grado de Doctor en Ingeniera Qumica

Y para que conste a los efectos oportunos firmamos el presente Certificado en


Alicante, 2 de Septiembre de 2013

Antonio Caballero Suarez


Fdo. Dr. Jose Ruben
Fdo. Dr. Jose Ruiz Femenia
La diferencia entre lo que hacemos
y somos capaces de hacer,
resolvera la mayora de los
problemas del mundo

Mahatma Gandhi
Agradecimientos

iba a decir hace apenas un a


Al final todo llega. Quien no que esto acabara,

en ese corto espacio de tiempo he concluido mi carrera, un master y ahora, por
fin, mi doctorado. Una nueva etapa de mi vida concluye, una etapa ilusionante y
maravillosa, que no habra sido posible sin la ayuda de numerosas personas a las que

quiero agradecer todo el apoyo, animos y ayuda recibida.

En primer lugar, me gustara dar las gracias especialmente a mis directores de tesis,
Antonio Caballero y Ruben
Jose Ruiz. Nunca podre
agradecerles todo el apoyo reci-
a lo largo de todo este periodo. Sin ellos, este trabajo
bido, su ayuda y comprension
no habra sido posible. Muchsimas Gracias.

me gustara agradecer a todos los compa


Tambien neros del departamento y del
instituto. Durante estos 3 a
nos, he tenido la maravillosa oportunidad de conocer a
numerosas personas. Personas como Blanca, Gloria, Nacho, Juan Javaloyes, Agustn,
Nando, etc. . . (y muchsimas otras que no
Raquel, M Dolores, Dani, Noelia, M Jose,
nombro porque no acabara), que han hecho que estos a
nos hayan pasado sin dar-

me cuenta. No me quiero olvidar, de las chicas del sotano de Ciencias (Luca, Mila,
Isa, Desi, Erika y Vero), por permitirme, sin poner pega alguna, desayunar todas las
ma
nanas con ellas. Pero en especial, quiero agradecer a Adriana Pineda, mi eterna
compa tiempo he pasado durante
nera del zulo del CTQ, aquella con la que mas
estos a
nos, gracias por tu alegra y tu ayuda durante todos estos a
nos.

No quiero ni puedo olvidarme de agradecer y dedicar este doctorado a toda mi


importantes
familia, en especial mis padres, Angel y Rosa, dos de las personas mas
en mi vida. Muchsimas gracias por todo, por haberme ense
nado que con trabajo y
todo se puede conseguir. Sin su apoyo, sus sacrificios y su amor desintere-
dedicacion
en deuda con
sado esto sera solo una fantasia. Muchsimas gracias, siempre estare
vosotros.

Y finalmente, quiero agradecer y dedicar especialmente este trabajo a la perso-


importante de mi vida. La persona que hace que me despierte y me acues-
na mas
sencillo.
te cada da con una sonrisa, aquella que hace que todo sea mucho mas

VII
cari
Muchsimas gracias por todo tu apoyo, ayuda, comprension,
no, animo, etc. . . ,
las palabras se quedan cortas para agradecerte tanto. Te quiero muchsimo Vicky.

Muchisimas gracias a tod@s

VIII
INDICE GENERAL

1. Resumen 1

2. Estructura de la Memoria y Objetivos 11


3. Introduccion 15

3.1. Energa en el Mundo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

energetica
3.1.1. Situacion actual . . . . . . . . . . . . . . . . . . . . . . . 18


3.1.2. Futuro de la Energa vs. Cambio Climatico . . . . . . . . . . . . . . 22


3.1.3. Principales Mejoras Energeticas . . . . . . . . . . . . . . . . . . . . . 26

3.2. Simuladores de Procesos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

Historica
3.2.1. Revision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

3.2.2. Arquitectura de los Simuladores Qumicos . . . . . . . . . . . . . . . 34

. . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.2.1. Clasificacion

3.2.3. Simulador Modular-Secuencial . . . . . . . . . . . . . . . . . . . . . . 38

3.2.4. Ventajas e inconvenientes de un simulador secuencial modular . . 47

3.3. Sntesis y Dise


no de procesos . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

IX
3.3.1. Metodologas en la Sntesis de Procesos . . . . . . . . . . . . . . . . 49


3.3.1.1. Metodos basados en reglas heursticas . . . . . . . . . . . . 50


3.3.1.2. Metodos matematica
basados en programacion . . . . . 53


3.3.1.3. Metodos hbridos . . . . . . . . . . . . . . . . . . . . . . . . 56

3.3.2. Principales Avances en la Sntesis de Procesos Qumicos . . . . . . . 57

3.3.3. Prospectivas de la Sntesis de Procesos . . . . . . . . . . . . . . . . . 60

matematica
3.4. Programacion en la sntesis de procesos . . . . . . . . . . . . 69

Lineal . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.4.1. Programacion

No Lineal
3.4.2. Programacion . . . . . . . . . . . . . . . . . . . . . . . . . 71

Lineal Mixta con Variables Enteras y Continuas . . . 75


3.4.3. Programacion

No Lineal Mixta con Variables Enteras y Continuas . 77


3.4.4. Programacion

General Disyuntiva . . . . . . . . . . . . . . . . . . . . 80
3.4.5. Programacion


3.4.5.1. Metodos para Problemas GDP . . . . . . . . 82
de Resolucion

3.4.6. Algoritmos computacionales . . . . . . . . . . . . . . . . . . . . . . . 94

4. Resultados 105

I: Strategies for the robust simulation of thermally coupled


4.1. Publicacion
distillation sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

4.1.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

4.1.2. Application of the proposed simulation strategy: acyclic system


simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

4.1.3. Examples and results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

4.1.3.1. Separations of three component mixtures . . . . . . . . . . 120

4.1.3.2. Difficult separations of three component mixtures . . . . . . 125

X
4.1.3.3. Separation of 4 and 5 component mixtures . . . . . . . . . 130

4.1.4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

II: A new technique for recovering energy in thermally coupled


4.2. Publicacion
distillation using vapor recompression cycles . . . . . . . . . . . . . . . . . 137

4.2.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

4.2.2. Inefficiency in thermally coupled distillation systems . . . . . . . . . 144

4.2.3. Vapor recompression and reverse vapor recompression cycles in


thermally coupled distillation . . . . . . . . . . . . . . . . . . . . . . . 147

4.2.4. Examples and implementation details . . . . . . . . . . . . . . . . . 153

4.2.4.1. Example 1. VRC . . . . . . . . . . . . . . . . . . . . . . . . . 154

4.2.4.2. Example 2: RVRC . . . . . . . . . . . . . . . . . . . . . . . . . 159

4.2.5. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

III: An alternative disjunctive optimization model for heat in-


4.3. Publicacion
tegration with variable temperatures . . . . . . . . . . . . . . . . . . . . . . 173

4.3.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176

4.3.2. Problem statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

4.3.3. Simultaneous heat and process optimization: model formulation . . 181

4.3.3.1. Implicit ordering of temperatures . . . . . . . . . . . . . . . 183

4.3.3.2. Heat transfer within each temperature interval . . . . . . . 187

4.3.3.3. Transshipment model . . . . . . . . . . . . . . . . . . . . . . 193

4.3.3.4. Logical relationships relating reordering and heat exchan-


ge binary variables . . . . . . . . . . . . . . . . . . . . . . . 195

4.3.3.5. Logical relationships involving heat exchange intervals . . 198

4.3.3.6. Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . 199

XI
4.3.3.7. Final remarks, extensions and performance of the disjun-
ctive model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201

4.3.4. Examples and results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203

4.3.4.1. Process with fixed streams conditions (MILP) . . . . . . . . . 204

4.3.4.2. Process with variable stream conditions (MILP) . . . . . . . 205

4.3.4.3. Heat integration of distillation columns (MINLP) . . . . . . . 209

4.3.4.4. Hybrid simulation-optimization process (MINLP) . . . . . . . 211

4.3.5. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214

4.4. Anexo I: Integration of different models in the design of chemical proces-


ses: Application to the design of a Power Plant . . . . . . . . . . . . . . . . 217

4.4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219

4.4.2. Modeling framework . . . . . . . . . . . . . . . . . . . . . . . . . . . 226

4.4.3. Logic Based Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . 230

4.4.3.1. Initial and primal subproblems . . . . . . . . . . . . . . . . . 232

4.4.3.2. Master subproblem . . . . . . . . . . . . . . . . . . . . . . . 233

4.4.4. Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234

4.4.5. Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251

4.4.6. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259

4.5. Anexo II: Integration of modular process simulators under the Generalized
Disjunctive Programming framework for the structural flowsheet optimiza-
tion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265

4.5.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267

4.5.2. Problem statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271

4.5.3. Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271

XII
4.5.3.1. Logic Based Outer Approximation algorithm with an em-
bedded process simulator . . . . . . . . . . . . . . . . . . . 272

4.5.3.2. Connection between Matlab and Aspen Hysys . . . . . . . 279

4.5.3.3. Connection between Matlab and Optimizer . . . . . . . . 280

4.5.4. Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281

4.5.5. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291

4.5.6. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294

5. Conclusiones 301

6. Contribuciones Cientficas 309

6.1. Publicaciones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311

6.2. Libros . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311

6.3. Asistencia a Congresos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312

6.4. Contribuciones a Congresos . . . . . . . . . . . . . . . . . . . . . . . . . . . 312

XIII
INDICE DE TABLAS

3.2.1. Tipos de propiedades fisicoqumicas en un simulador de procesos qumi-


cos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

3.2.2.
Principales Modelos Termodinamicos disponibles en los simuladores co-
merciales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

3.2.2.
Principales Modelos Termodinamicos disponibles en los simuladores co-
merciales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

3.2.2.
Principales Modelos Termodinamicos disponibles en los simuladores co-
merciales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3.4.1. de operadores logicos


Transformacion a restricciones MILP . . . . . . . . 85

3.4.2. Esquema Reformulaciones M-Grande y Envolvente Convexa de un Pro-


blema Lineal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

3.4.3. Esquema Reformulaciones M-Grande y Envolvente Convexa de un Pro-


blema No Lineal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

4.1.1. Characteristics of different feeds to each column . . . . . . . . . . . . . . 120

4.1.2. Characteristics of short cut columns and conventional columns . . . . . 124

4.1.3. Results obtained with systems 3 compounds . . . . . . . . . . . . . . . . . 124

XV
4.1.4. Results of difficult separation systems with 3 compounds . . . . . . . . . . 128

4.1.5. Results for the systems with 4 compounds . . . . . . . . . . . . . . . . . . . 131

4.2.1. Specifications for the hot and cold utilities . . . . . . . . . . . . . . . . . . 154

4.2.2. Streams specifications for the separation system . . . . . . . . . . . . . . 155

4.2.3. Conventional Petlyuk Distillation column: Capital & Energy Cost . . . . . 156

4.2.4. Conventional Petlyuk Distillation with VRC: Capital and Energy Cost . . . 157

4.2.5. Specifications for the streams in the separation system . . . . . . . . . . . 159

4.2.6. Conventional Petlyuk Distillation column: Capital and Energy Cost . . . . 160

4.2.7. Conventional Petlyuk Distillation with RVRC: Capital and Energy Cost . . 160

4.3.1. Test problem 1 (fixed temperatures) . . . . . . . . . . . . . . . . . . . . . . 204

4.3.2. Results test problem 1 (fixed temperatures) . . . . . . . . . . . . . . . . . . 205

4.3.3. Test problems 24 (variable temperatures) . . . . . . . . . . . . . . . . . . 206

4.3.4. Results test problems 24 (variable temperatures) . . . . . . . . . . . . . . 207

4.3.5. Parameters solution test problems 24) . . . . . . . . . . . . . . . . . . . . 208

4.3.6. Flows and temperatures of problem columns . . . . . . . . . . . . . . . . 210

4.3.7. Solution of problem columns . . . . . . . . . . . . . . . . . . . . . . . . . . 211

4.3.8. Molar Flow rate of the feed (kmol/h) . . . . . . . . . . . . . . . . . . . . . 212

4.3.9. Bounds, constraints and solution of the natural gas problem . . . . . . . . 213

4.4.1. Summary of operating parameters . . . . . . . . . . . . . . . . . . . . . . 238

4.4.2. Utilities Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239

4.4.3. Data for examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251

4.4.4. Results for example 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253

4.4.5. Numerical statistics for the three instance . . . . . . . . . . . . . . . . . . 254

XVI
4.4.6. Results for example 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256

4.4.7. Results for example 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258

4.5.1. Temperature and pressure dependence of the carbon monoxide and


carbon dioxide equilibrium conversions . . . . . . . . . . . . . . . . . . . . 283

4.5.2. Synthesis gas streams data . . . . . . . . . . . . . . . . . . . . . . . . . . . 284

4.5.3. Computational Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292

4.5.4. Results: Characteristic of equipment in optimal configuration . . . . . . . 293

XVII
INDICE DE FIGURAS

3.1.1. consumo energetico


Evolucion mundial [1] . . . . . . . . . . . . . . . . . 18

3.1.2. del consumo mundial de energa . . . . . . . . . . . . . . . . 19


Distribucion

3.1.3.
Consumo Energetico
Mundial en los ultimos 20 a
nos . . . . . . . . . . . . 19

3.1.4.
Consumo Energetico Mundial detallada en 2011 . . . . . . . . . . . . . . 21

3.1.5.
Consumo per capita (Tep) en 2011 [2] . . . . . . . . . . . . . . . . . . . . 22

3.1.6. Concentraciones CO2 (ppm) en cada uno de los escenarios . . . . . . 25

3.1.7. Calentamiento Global en cada uno de los escenarios . . . . . . . . . . . 26

3.2.1. Problemas en Ingeniera de Procesos: a) Dise c) Opti-


no, b) Simulacion,
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
mizacion

3.2.2. de un simulador modular-secuencial . . . . . . . 41


Esquema de operacion

3.2.3.
Esquema de calculo en cada uno de los equipos . . . . . . . . . . . . . 47

3.3.1. esquematica
Representacion simplificada de un problema de sntesis . 50

3.4.1. Convexa y No Convexa . . . . . . . . . . . . . . . . . 72


Ejemplo de Region

3.4.2. Esquema del algoritmo de Aproximaciones Exteriores . . . . . . . . . . . 79

3.4.3. de la relajacion
Representacion utilizando M-Grande y Envolvente Con-
vexa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

XIX
4.1.1. (a, b and e) Equivalent configurations. (c, d and f) Equivalent configu-
rations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

4.1.2. Petlyuk configuration and the thermodynamically equivalent divided


wall column . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

4.1.3. Details of the connection between columns, cyclic system simulation 118

4.1.4. Details of the connection between columns, acyclic system simulation119

4.1.5. Residue curve maps of the four component systems studied: (a) i-butane,
n-butane, cyclobutane; (b) n-hexane, n-heptane, n-octane; (c) benze-
ne, toluene, p-xylene; (d) methanol, ethanol, butanol . . . . . . . . . . . 121

4.1.6. Calculation of tray columns using short cut distillation . . . . . . . . . . 122

4.1.7. Simulations of (a) acyclic and (b) cyclic system configurations . . . . . . 123

4.1.8. Comparison of flows for BTX separation: (a) liquid flow-column 1; (b) va-
por flow-column 1; (c) liquid flow-column 2; (d) vapor flow-column 2 . . 125

4.1.9. Residue curve map (azeotropic mixture) . . . . . . . . . . . . . . . . . . . 126

4.1.10. Simulations of acyclic (a) and cyclic (b) system configurations (azeotro-
pic mixture) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

4.1.11. Curve residue map acetoneacetic acidacetic anhydride (AAA) . . . 128

4.1.12. Comparison of flows for AAA separation: (a) liquid flow-column 1; (b)
vapor flow-column 1; (c) liquid flow-column 2; (d) vapor flow-column 2 . 129

4.1.13. Thermodynamically equivalents configurations-system 4 compounds: (a)


configuration 1; (b) configuration 2; (c) configuration 3 . . . . . . . . . . 130

4.1.14. Configuration separation system 5 compounds . . . . . . . . . . . . . . . 131

4.1.15. Comparison of flows for 5 compounds separation: (a) liquid flow-column


1; (b) vapor flow-column 1; (c) liquid flow-column 2; (d) vapor flow-
column 2; (e) liquid flow-column 3; (f) vapor flow-column 3; (g) liquid
flow-column 4; (h) vapor flow-column 4 . . . . . . . . . . . . . . . . . . . . 132

XX
4.2.1. Construction of the Petlyuk c), and Divided Wall columns d) by decom-
position into basic tasks b) and further elimination of heat exchangers . 145

4.2.2. Equivalent configurations for a thermal coupling . . . . . . . . . . . . . . 146

4.2.3. Petlyuk configuration with an intermediate reboiler (a) and an interme-


diate condenser (b) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

4.2.4. Scheme of a standard vapor recompression cycle . . . . . . . . . . . . . 149

4.2.5. Scheme of a Vapor Recompression Cycle using the excess of Vapor in


rectifying section of column 3 . . . . . . . . . . . . . . . . . . . . . . . . . 150

4.2.6. Scheme of a Reverse Vapor Recompression Cycle using the excess of


liquid in the stripping section of column 2 . . . . . . . . . . . . . . . . . . . 152

4.2.7. Flowsheet for the simulation of the Petlyuk configuration column . . . . . 155

4.2.8. Simulation of Petlyuk configuration column with VRC . . . . . . . . . . . . 156

4.2.9. Flowsheet for the direct sequence with and without VRC and for the
indirect sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

4.2.10. Capital cost of equipment and annualized energy and total costs for
the distillation systems studied in example 1 . . . . . . . . . . . . . . . . . 159

4.2.11. Flowsheet for the simulation of the Petlyuk column with RVRC . . . . . . . 161

4.2.12. Simulation of indirect sequence with RVRC . . . . . . . . . . . . . . . . . . 162

4.2.13. Annualized capital cost, energy costs and total costs for the distillation
systems studied in example 2 . . . . . . . . . . . . . . . . . . . . . . . . . . 162

4.3.1. Illustration of ordered temperatures and intervals . . . . . . . . . . . . . . 187

4.3.2. All the relative positions of a stream in relation to a given temperature


interval . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188

4.3.3. Heat cascade diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194

4.3.4. Logical relationships for hot streams . . . . . . . . . . . . . . . . . . . . . . 197

XXI
4.3.5. Minimum and maximum values for the inlet temperatures of hot and
cold streams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200

4.3.6. Sharp distillation for separation of a 4-component mixture . . . . . . . . . 209

4.3.7. Process flow diagram for the processing of the natural gas . . . . . . . . 212

4.4.1. Scheme of the modeling framework . . . . . . . . . . . . . . . . . . . . . 227

4.4.2. Superstructure of the utility system plant. It is possible to include N turbi-


nes in parallel of each class . . . . . . . . . . . . . . . . . . . . . . . . . . 236

4.4.3. Scheme of the actual implementation of a gas turbine in Aspen-Hysys . 239

4.4.4. Left, calculated and predicted values for the kriging model in the gas
turbine. Right, % kriging error in 500 random points . . . . . . . . . . . . . 241

4.4.5. Scheme of the implementation of a Boiler in Aspen Hysys. Dotted line


shows the information flow that is different of mass flow . . . . . . . . . . . 243

4.4.6. Scheme of a deaerator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248

4.4.7. Optimal solution for case study, instance 1 . . . . . . . . . . . . . . . . . . 252

4.4.8. Optimal solution for case study, instance 2 . . . . . . . . . . . . . . . . . . 257

4.4.9. Optimal solution for case study, instance 3 . . . . . . . . . . . . . . . . . . 259

4.5.1. Scheme of the optimization modeling framework . . . . . . . . . . . . . . 273

4.5.2. Methanol simplified process flowsheet . . . . . . . . . . . . . . . . . . . . 281

4.5.3. Superstructure for the Methanol synthesis process in Aspen-Hysys . . . . . 283

4.5.4. Optimal configuration of the case study . . . . . . . . . . . . . . . . . . . 291

XXII
CAPITULO 1

Resumen

1
Captulo 1. Resumen

Captulo 1. Resumen
moderna, la disponibilidad de energa esta
En la civilizacion fuertemente ligada
de vida del ser humano. En realidad
al nivel de bienestar, a la salud y a la duracion
vivimos en una sociedad que se poda denominar como energvora. A
no a a
no, el

consumo energetico
mundial marca maximos
historicos, como consecuencia de que

las necesidades energeticas de la humanidad no dejan de crecer. El principal motivo
de pases densamente poblados
se debe a la relativamente nueva industrializacion
como China, Brasil o India, lo que incorpora continuamente millones de nuevos con-

sumidores al mercado energetico. Dicha cantidad adicional de energa, necesaria
para suplir el incremento de la demanda mundial, se obtiene en su inmensa mayora
gas y petroleo.
a partir de combustibles basados en el carbono: carbon, Todo ello

provoca que las emisiones de contaminantes a la atmosfera (sobretodo pero no so-
lamente, de CO2 ) se incrementen a
no a a
no. El aumento de las emisiones supone un
importante problema a nivel global y es responsable de numerosas consecuencias
como son el continuo calentamiento del planeta, variaciones tanto en las cantidades
reduccion
como los patrones de precipitacion, de los casquetes polares, elevacion

del nivel del mar, etc. . . Finalmente, y si no se pone remedio al continuo aumento de
un impacto directo y a nivel mundial en los recursos hdricos
dichas emisiones, tendra
de animales y plantas y finalmente en
disponibles, la cadena alimentaria, la extincion
la futura existencia de la raza humana.

Hoy en da existen diferentes alternativas que posibilitaran cumplir con el conti-



nuo aumento de las necesidades energeticas sin incumplir el objetivo medioambien-

tal, manteniendo o incluso disminuyendo las emisiones atmosfericas. Las opciones
ampliamente estudiadas son el desarrollo y utilizacion
mas de energas renovables,
de captura y almacenamiento de CO2 o la mejora de la eficiencia
la instauracion

energetica [1]. Diversos estudios [2-4] muestran que, con las tecnologas actuales, la
presente y futura de energas renovables funcionando a maxima
instalacion capaci-
dad no bastara para proporcionar toda la energa adicional necesaria para asegurar

el crecimiento economico mundial. Todo ello indica que el principal campo que pue-

de llevar a la humanidad al cumplimiento, tanto de los objetivos economicos como

medioambientales, es la mejora de la eficiencia energetica.

3
1.Resumen


El consumo energetico del sector industrial a nivel global representa aproximada-

mente el 28 % del consumo energetico total y dentro de este sector, la industria qumi-

ca representa aproximadamente el 20 %. Si se realiza un simple calculo de porcenta-

jes, el consumo energetico de la industria qumica en el mundo representa aproxima-
damente el 5,6 % de toda la energa consumida en el mundo. Esto unido al hecho de
de
que la energa consumida es generada principalmente mediante la combustion

combustibles fosiles, hacen que cualquier mejora que permita reducir su consumo

energeticos, un importante impacto tanto en el coste economico
supondra como
medioambiental global.

El objetivo principal de la presente tesis es el desarrollo de herramientas de simu-


y optimizacion
lacion de procesos qumicos que mejoren la eficiencia energetica
de
estos. El trabajo desarrollado a lo largo de periodo doctoral puede agruparse en tres
grupos:

y mejoran la eficiencia en columnas


Herramientas que facilitan la simulacion

de destilacion

simultanea
Un modelo alternativo para la optimizacion de ca-
y la integracion
lor, que mantiene el concepto de intervalo de la temperatura, para redes de
intercambio de calor

rigurosa de problemas
Nuevas estructuras de modelado para la optimizacion
sinergica
de sntesis de procesos mediante una combinacion de simuladores
Disyuntiva Generalizada (GDP)1
de procesos con la formulacion

compuesta por dos trabajos, cada uno de


La primera parte de esta tesis esta
en revistas cientficas internacionales. El
los cuales ha dado lugar a una publicacion

primer trabajo de los realizados parte de la idea de mejorar y facilitar la simulacion

de secuencias de columnas con acoplamiento termico (TCD) 2 . El problema de la
de este tipo de columnas radica en el gran numero
simulacion de recirculaciones
existentes entre las diferentes columnas como consecuencia de los acoplamientos

termicos. y el
Esta caracterstica ralentiza y dificulta considerablemente la simulacion
1 Generalized Disjunctive Problem)
Acronimo de Ingles
2 Thermally Coupled Distillation
Acronimo del ingles

4
Captulo 1. Resumen


estudio de los sistemas TCD, necesitandose, en la mayora de los casos, buenos valo-

res iniciales proximos final para converger el sistema. La idea basica
a la solucion para
solventar este problema es la de evitar o sustituir la estructura cclica (corrientes recir-
acclica.
culadas) en el diagrama de flujo de los sistemas TCD, por una configuracion
aportada parte de los estudios realizados por Carlberg y Westerberg [5,
La solucion

6], quienes probaron, utilizando modelos aproximados (Metodo de Underwood [7]),

que un acoplamiento termico
puede ser sustituido por una unica corriente de vapor
de las secciones que conecte. En
sobrecalentado o lquido subenfriado en funcion
este trabajo se demuestra que dicha corriente de vapor sobrecalentado o lquido
subenfriado puede ser sustituida por dos corrientes una de materia y otra de energa,
de tal forma que se consigue eliminar las corrientes recirculadas en los acoplamien-

tos termicos. es aplicada a secuencias de columnas TCD
Esta nueva configuracion
de diversas mezclas como hidrocarburos, alcoholes, compuestos
para la separacion

aromaticos, etc. . . . Los resultados obtenidos muestran errores medios por debajo del
original con la
2 % para mezclas de 3 componentes, al comparar la configuracion
desarrollada en este trabajo.
configuracion


El segundo de los trabajo tiene como objetivo el aumento de la eficiencia energeti-
ca en una columna TCD. Las columnas TCD presentan diferentes caractersticas que

siendo teoricamente
hacen que su funcionamiento, aun que mini-
la configuracion

miza el consumo energetico con las
del proceso, sea ineficiente en comparacion
convencionales. La idea de este trabajo es la de aumentar
columnas de destilacion

la eficiencia energetica de las columnas TCD convirtiendo la ineficiencia inherente
de las secuencias TCD -exceso de flujo de vapor/lquido en ciertas secciones como

resultado de la transferencia llevada a cabo en los acoplamientos termicos- en una
ventaja, consiguiendo disminuir la energa y costes del sistema. La estrategia consis-
de la columna que, como
te en extraer el exceso de vapor (lquido) de la seccion

consecuencia de los acoplamientos termicos,
opera en condiciones suboptimas. La
de vapor
corriente de vapor (lquido) extrada se utiliza en un ciclo de recompresion
de vapor inverso
(VRC, Vapor Recompression Cycle) o un ciclo de recompresion
(RVRC, Reverse Vapor Recompression Cycle), reduciendo de esta forma la can-
la extraccion
tidad de calor aportado al sistema externamente. Ademas, de la co-

rriente (lquida o vapor en exceso) restaura las condiciones optimas de funcionamien-

5
1.Resumen

de la columna afectada y consecuentemente reduce el diametro


to de la seccion

de la columna. Los resultados obtenidos muestran ahorros energeticos
y economicos
de esta estrategia entorno un 20-40 %.
consecuentes de la aplicacion

Tras la anterior serie de trabajos, el siguiente bloque de la tesis se basa en el desa-


simultanea
rrollo de una estrategia alternativa para la optimizacion
y la integracion
de calor, manteniendo el concepto de intervalo de la temperatura, para redes de
ha derivado en un pu-
intercambio de calor. Comentar que dicho trabajo tambien
en revistas cientficas. La gran mayora de los metodos
blicacion desarrollados en el
energetica
campo de integracion asumen que las temperaturas de entrada y salida
a una red de intercambiadores de calor son fijas y conocidas a priori. En otras pala-
de calor se lleva a cabo solo
bras, la integracion despues
de haber fijado todas las
del proceso. Sin embargo, es sabido que la optimizacion
condiciones de operacion

simultanea de calor puede producir
de las condiciones del proceso y la integracion

importantes ahorros en el coste total del proceso [8]. Por lo que sabemos, el unico
e integracion
trabajo relacionado con la optimizacion de calor simultanea
fue desa-
rrollado por Duran y Grossmann [8]. La idea de este trabajo surge del hecho de que
el modelo de Duran y Grossmann no mantiene el concepto de intervalo de la tem-
peratura con la que muchos dise
nadores se encuentran familiarizados, debido a la

prevalencia de la tecnologa del analisis del pinch. En este apartado se pretende
simultanea
desarrollar una estrategia alternativa para la optimizacion y la integra-
de calor. La idea basica
cion implcita de todas las
es llevar a cabo una ordenacion
temperaturas del sistema, de esa manera los intervalos de temperatura quedan tam-
implcitamente definidos. El modelo se compone de tres partes: en la primera,
bien
todas las temperaturas de entrada del sistema son implcitamente ordenadas, en el
segundo, se calcula la cantidad de calor intercambiado por cada corriente en cada
intervalo de temperatura, y en la tercera, se aplica el modelo de transbordo desarro-
llado por Papoulias y Grossmann [9] con el fin de calcular los servicios calientes y fros
se deben a
a aportar al sistema. Ademas nadir un un etapa de preprocesado con el

fin de fijar ciertas variables binarias, y un conjunto de relaciones logicas que reducen

considerablemente el espacio de busqueda. Los resultados obtenidos con el nuevo
modelo muestran resultados competitivos frente al modelo de Duran y Grossmann.

6
Captulo 1. Resumen


Finalmente, y como ultima parte del trabajo doctoral, se lleva a cabo una serie
de trabajos con el fin de desarrollar nuevas estructuras de modelado para la opti-
rigurosa de problemas de sntesis de procesos, utilizando una combinacion
mizacion
GDP. Esta investigacion
de simuladores de procesos junto a la formulacion ha dado
en revistas internacionales. El
como resultado dos trabajos actualmente en revision
primero de los trabajos presenta una estructura de modelado para tratar problemas

de sntesis, que incluyen diferentes modelos matematicos para calcular o simular di-
ferentes operaciones unitarias. Los avances en el dise y sntesis de
no, optimizacion

procesos qumicos en los ultimos a
nos han sido impresionantes en todos los niveles,
del diagrama de flu-
desde operaciones unitarias individuales hasta la optimizacion
jo completo. Hoy en da existen numerosos modelos para calcular, dise
nar o simular

practicamente cualquier proceso qumico. Dentro de los diferentes modelos existen-
tes se pueden distinguir entre modelos en forma de ecuaciones, modelos en forma
de caja negra con o sin acceso a las derivadas, etc. . . En definitiva, la mejor alter-
nativa de dise y utilizacion
no es la seleccion del mejor modelo disponible venga de
donde venga independientemente de su forma y caractersticas. Con todo ello, la
idea principal del trabajo es el desarrollo de una estructura de modelado que sea

capaz de trabajar simultaneamente con modelos con comportamiento y tratamien-

to numerico distintos, de tal forma que se seleccione el mejor modelo en cada caso
para cada unidad. En este trabajo toda la estructura de modelado se desarrolla uti-
disyuntiva, porque esta estructura es adecuada para hacer
lizando la programacion
frente a este tipo de problemas, ya que permite encapsular cada modelo y utilizar
de sus caractersticas. La estructura llevada a cabo
un enfoque diferente en funcion
utiliza disyunciones para capturar cada uno de los modelos de tal forma que se apro-
vechen las ventajas o minimizan los inconvenientes de cada uno. De esta manera es
disponibles y simultaneamen-
posible utilizar las derivadas algebraicas cuando estan

te implementar metodos robustos para estimar las derivadas de otros modelos sin la
necesidad de interferir con el resto de modelos, y de esta forma mejorar el tiempo

de calculo computacional. A modo de ejemplo, el modelo es aplicado a una plan-
de vapor donde ciertas unidades son calculadas a traves
ta de generacion de un
simulador de procesos y otras son explcitamente representadas por ecuaciones. El
resultado final es un sistema hbrido que incluye modelos implcitos procedentes de

diferentes fuentes y con diferentes comportamientos numericos.

7
1.REFERENCIAS


El segundo y ultimo trabajo desarrollado presenta un nuevo marco de modelado

para tratar los problemas de sntesis. En este, las diferentes alternativas del proceso
son modeladas en forma de una superestructura, y se resuelve utilizando una combi-
sinergica
nacion GDP.
de simuladores de procesos comerciales junto a la formulacion
de pro-
Hasta donde conocemos, no existe ninguna herramienta para la resolucion
blemas donde la superestructura del proceso queda dibujada directamente sobre

la interfaz grafica del simulador de procesos (GUI). La estructura del modelo se cen-
tra en construir un problema GDP que es conectado con un simulador de procesos

(Hysys Aspen), donde tanto la topologa como los parametros continuos del proceso

son optimizados. Esta metodologa permite incluir facilmente restricciones y relacio-

nes logicas de soluciones
entre las alternativas, de tal forma que facilite la obtencion
factibles. La herramienta propuesta utiliza el algoritmo de aproximaciones exteriores

basada en la logica y por lo tanto no es necesario volver a formular el problema co-
de
mo un MINLP. Como ejemplo se presenta la sntesis de una planta de produccion
metanol, En este caso se presenta una superestructura que incluye todas las alterna-
incluyendo diferentes alimentos, diferentes sistemas de compresion
tivas de interes, y

diferentes condiciones de reaccion.

Referencias

[1] Europea.
Abadia, J. Energa y Sostenibilidad. La Energa y el Medio Ambiente en la Union
ENDESA, 2006.
Inf. tec.
[2] 2009 ASPO-USA Peak
Nelder, C. Efficiency: The Way to Close the Renewables Gap. Inf. tec.
Oil Conference, 2009.
[3] Trainer, F.E. Renewable Energy Cannot Sustain a Consumer Society. Springer London, Limi-
ted, 2007.
[4] Agency, International Energy. World Energy Outlook 2012. http://www.worldenergyoutlook.
org//. 2012.
[5] Carlberg, N.A. y Westerberg, A.W. ((Temperature-heat diagrams for complex columns. 2.
Underwoods method for side strippers and enrichers)). En: Industrial & Engineering Che-

mistry Research 28.9 (1989), pags. 1379-1386.
[6] Carlberg, N.A. y Westerberg, A.W. ((Temperature-heat diagrams for complex columns. 3.
Underwoods method for the Petlyuk configuration)). En: Industrial & Engineering Chemistry

Research 28.9 (1989), pags. 1386-1397.

8
Captulo 1. Resumen

[7] Underwood, A.J.V. ((Fractional Distillation of Multicomponent Mixtures)). En: Industrial & En-

gineering Chemistry 41.12 (1949), pags. 2844-2847.
[8] Duran, M.A. y Grossmann, I.E. ((Simultaneous optimization and heat integration of chemical
processes)). En: AIChE Journal 32.1 (1986), pags.
123-138.
[9] Papoulias, S.A. y Grossmann, I.E. ((A structural optimization approach in process synthesis
II: Heat recovery networks)). En: Computers & Chemical Engineering 7.6 (1983), pags.
707
-721.

9
CAPITULO 2

Estructura de la Memoria y Objetivos

11
Captulo 2. Estructura y Objetivos

Captulo 2. Estructura de la Memoria y Objetivos


general. en la que se reali-
La presente memoria se inicia con una introduccion
za una peque de la problematica
na presentacion
energetica se
mundial. Ademas
energetica
analiza la situacion actual en el mundo y las principales alternativas y

mejoras energeticas que pueden conseguir suplir las cada vez mayores demandas

energeticas respetando los acuerdos propuestos por el IPCC (International Panel on

Climate Change) de tal forma que se llegue a controlar el cambio climatico. A con-
y como consecuencia del gran uso de los simuladores comerciales a lo
tinuacion,
largo de la tesis, se realiza una peque del uso de los simuladores co-
na introduccion
a la arquitectura de estos.
merciales, prestando especial atencion Para acabar esta
y debido a que los simuladores utilizados durante este trabajo son simuladores
seccion
modulares-secuenciales se explica el estructura y funcionamiento. Seguidamente, se
presenta un peque
no resumen del campo de la sntesis de procesos qumicos, en el
que se describen las diferentes metodologas desarrolladas a lo largo de la historia y
se finaliza con las prospectivas de futuro en este campo. Finalmente, y para concluir la
general, se presenta un resumen de la programacion
introduccion matematica
en la
sntesis de procesos en el que se comenta brevemente los diferentes tipos de proble-

mas a resolver y la metodologa seguida en cada caso, prestando especial atencion
disyuntiva generalizada dado su gran peso a lo largo de esta tesis.
a la programacion

general, el siguiente captulo presenta los re-


Una vez realizada la introduccion
sultados obtenidos. Estos se presentan en forma de publicaciones, seguidos de un
apartado adicional que incluyen los resultados hasta la fecha no publicados. Cada
una de las publicaciones va precedida por una peque Finalmente,
na introduccion.
el captulo 5 de esta memoria corresponde a las conclusiones obtenidas. Comentar
la bibliografa utilizada en cada uno de los captulos se incluye al final de cada
que
uno de estos.

Los objetivos principales de esta tesis puede dividirse en tres grupos diferentes:

y mejoran la eficiencia en columnas


1. Herramientas que facilitan la simulacion

de destilacion

13
2.Estructura y Objetivos

para secuencias de co-


(a) Desarrollo de una nueva estructura de simulacion
lumnas TCD que elimine las corrientes recirculadas


(b) Desarrollo de una nueva estrategia para aumentar la eficiencia energeti-
ca de las columnas TCD, aprovechando la ineficiencia generada por el
exceso de flujo de vapor/lquido en ciertas secciones como resultado de

la transferencia llevada a cabo en los acoplamientos termicos

simultanea
2. Un modelo alternativo para la optimizacion de ca-
y la integracion
lor, que mantiene el concepto de intervalo de la temperatura, para redes de
intercambio de calor

rigurosa de problemas
3. Nuevas estructuras de modelado para la optimizacion
sinergica
de sntesis de procesos mediante una combinacion de simuladores
GDP
de procesos con la formulacion


(a) Desarrollo de una nueva estrategia de modelado que permita la utilizacion

simultanea
de diferentes modelos numericos de las
para la representacion
diferentes operaciones unitarias en un diagrama de flujo, de forma que se
aprovechen las ventajas de cada uno de ellos

(b) Desarrollo de un nuevo marco de modelado para problemas rigurosos don-


de la topologa del proceso no se encuentre fijada a priori, quedando el
proceso representado en forma de superestructura

14
CAPITULO 3


Introduccion

15

Captulo 3. Introduccion


Captulo 3. Introduccion
fuertemente ligada al nivel de
En la actualidad, la disponibilidad de energa esta
de vida del ser humano. En realidad vivimos
bienestar, a la salud y a la duracion
en una sociedad que puede ser denominada energvora. En esta sociedad, los
pobres muestran los consumos mas
pases mas bajos de energa, mientras que los
ricos utilizan grandes cantidades de la misma. Sin embargo este escenario
pases mas
cambiando de forma drastica.
esta en los proximos
Dicho cambio se acentuara a
nos,
precisamente los pases en vas de desarrollo quienes experimenten con
donde seran
mayor rapidez un aumento en su consumo de energa debido al incremento que
tanto en sus poblaciones como en sus economas.
tendran

3.1 Energa en el Mundo

del hombre (Homo Sapiens Sapiens), aproximadamente hace


Desde la aparicion
100.000 a sobre el planeta
nos hasta la mitad del siglo XIX, el ser humano se expandio

sin superar los mil millones de habitantes y limitando su consumo energetico a sus

necesidades basicas calefaccion,
(coccion, higiene e iluminacion).
Fue a partir de
Industrial, cuando se produjo el
la segunda mitad del siglo XVIII, con la 1 Revolucion

principal cambio energetico del modelo energetico
en la historia, con la instauracion
industrial (Figura 3.1.1) [1]. Este cambio se produjo por numerosas causas, pero una
fundamental y decisiva: el aprovechamiento de una fuente de energa primaria, que
la sustitucion
implico de dos fuentes gratuitas y de uso libre, como el agua y el viento,

por otra de caracter industrial. Con ello se intensifico
comercial, como el carbon el

consumo de combustibles fosiles, ya que por
principalmente el consumo de carbon,

aquella epoca
el petroleo
era apenas utilizado extrayendose de campos de Baku
(Azerbajan) y Alsacia (Francia).

A mediados del siglo XIX se producen una serie de descubrimientos que revolu-
cionan la vida del hombre como son el descubrimiento del electromagnetismo o el
interna. En distintas partes del mundo
invento de la turbina y el motor de combustion
de petroleo:
comienza la moderna industria de extraccion F.N. Semyenov en Asphe-

17
3.1.Energa en el Mundo

consumo energetico
Figura 3.1.1: Evolucion mundial [1]

ron,Baku (1844), I.Lukasiewicz en Bobrka, Polonia (1858) y E.Drake en Titusville, Pensil-


vania (1859).

E stas, junto a muchas otras mejoras tecnologicas,



derivaron en la gran explosion

energetica experimentada a principios del siglo XX, siendo la energa el principal im-
pulsor del desarrollo industrial y facilitador de gran parte de las actividades humanas.

de electricidad crecio
Durante el siglo XX la produccion desde cero hasta alcan-
zar los 15.000 TWh en el a
no 2000. El petroleo y el consumo de derivados llega a las
3.500 MTep 1
(millones de toneladas equivalentes de petroleo) mun-
y la poblacion
dial crece de 1.600 millones de habitantes en 1.900 a 7.000 millones de habitantes en
2011. Concretamente en el a
no 2011, el consumo energetico en
global se establecio
de los fuentes energeticas
12.274 MTep [2]. La distribucion consumidas se muestra la

Figura 3.1.2, representando los combustibles fosiles el 87 % de la energa mundial con-
sumida.

3.1.1. energetica
Situacion actual


El consumo energetico un 2,5 % en el a
mundial se incremento no 2011, pasando

de 11.978 MTep hasta 12.274 MTep (Figura 3.1.3), localizandose principalmente dicho
1
1 Tep equivale a 41.87106 kJ

18

Captulo 3. Introduccion

del consumo mundial de energa


Figura 3.1.2: Distribucion

aumento en las economas emergentes, representando China por si sola el 71 % de


una
dicho crecimiento. El conjunto de pases pertenecientes a la OCDE presento
del 0,8 % en su consumo energetico,
disminucion por el contrario los pases no perte-
necientes a la OCDE presentaron un incremento del 5,3 %.


Figura 3.1.3: Consumo Energetico
Mundial en los ultimos 20 a
nos

Actualmente, el mundo se encuentra inmerso en un estado excepcionalmente



inestable, tanto poltica como medioambientalmente, lo que provoca la aparicion

19
3.1.Energa en el Mundo

energetica
de una nueva situacion mundial, con repercusiones directas en todos y

cada uno de los recursos energeticos.


La Primavera Arabe, a principios del 2011, y la Guerra Civil Libia sacudieron

enormemente el mercado energetico mundial, principalmente el relacionado con el

petroleo. del petroleo
Dichos acontecimientos derivaron en una inmensa reduccion
norteafricano (1,2 millones de barriles/da aproximadamente) y el consecuente au-
en Oriente Medio. Esta incertidumbre en la oferta y princi-
mento de la produccion
palmente la inestabilidad poltica de la zona produjo que el precio medio anual del

barril de petroleo un maximo
alcanzara
historico (111,26 $/barril), un 40 % superior al
a
no 2010. Todas estas caractersticas produjeron que el consumo de petroleo creciera

unicamente
un 0,6 %, el menor de la ultima
decada [3].

el concepto de seguridad
Por otro lado, el accidente de Fukushima tambaleo
en Japon
en la energa nuclear, no solo sino a nivel mundial. Todo ello derivo
en una
drastica
reduccion
de la energa nuclear (-4,3 %), especialmente localizada en Japon
(44 %) y Alemania (23 %).

se incremento
El consumo de carbon un 5,4 % respecto al a
no 2010, suponien-
do dicha cifra el mayor incremento registrado en la historia, llegando a representar el

30,3 % del consumo energetico mundial. Este importante aumento tiene como causa,
es el combustible seleccionado mayoritariamente para
el hecho de que el carbon

cubrir el incremento anual de la demanda energetica. Ello es debido a que de todos

los combustibles fosiles, es por mucho el mas
el carbon abundante en el mundo y el

mejor distribuido geograficamente, de 100 pases en
pudiendo encontrarse en mas
sea la principal fuente
todos los continentes. Este hecho ha llevado a que el carbon

energetica
elegida para cubrir el incremento de la demanda energetica, especial-
mente en los pases en vas de desarrollo (principalmente China e India).

un 2,2 % a nivel mundial.


Por otro lado, el mercado del gas natural se incremento

Como en el resto de fuente energeticas, crecio
el pas donde mas el consumo fue
(11,6 %). Ademas,
en China (21,5 %), seguido de Arabia Saudi (13,2 %) y Japon la pro-
de gas natural en Estados Unidos se incremento
duccion sustancialmente en el ultimo

a
no (7,7 %) inducido por tecnologas de exploracion-producci que estan
on liberando

petroleo espo-
ligero en formaciones compactas y gas de esquisto. Todo ello esta

20

Captulo 3. Introduccion


leando la actividad economica -con el abaratamiento de los precios del gas y de la
electricidad que ofrecen un margen competitivo a la industria. Finalmente es impor-
del consumo de gas natural en la Union
tante comentar la fuerte reduccion Europea
se debe principalmente a la crisis economica,
(9,9 %). Dicha reduccion los altos pre-
cios del gas importado y el crecimiento de las energas renovables.


Estas ultimas
lograron un incremento global del 17.7 %, basandose principalmente

en la energa eolica
con un aumento del 25.8 %, representando esta de la mitad
mas
de energas renovables. En el caso de la energa solar se produjo un
de la produccion
aumento del 86.3 %, indicando este dato el correcto camino seguido por esta fuente

energetica.

energetica
Las Figura 3.1.4 y 3.1.5 muestran un resumen de la situacion mundial
respecto al a
y su variacion
no 2010, y el consumo energetico
per capita en el a
no
2011, respectivamente.


Figura 3.1.4: Consumo Energetico Mundial detallada en 2011

21
3.1.Energa en el Mundo


Figura 3.1.5: Consumo per capita (Tep) en 2011 [2]

3.1.2.
Futuro de la Energa vs. Cambio Climatico


Las necesidades energeticas de la humanidad no dejan de crecer. La relativa-
de pases densamente poblados como China, Brasil o
mente nueva industrializacion
incorporando millones de consumidores cada a
India esta
no al mercado energetico.
Dicha cantidad adicional de energa, necesaria para suplir el incremento de la de-
manda mundial, se obtiene en su inmensa mayora a partir de combustibles basados
gas y petroleo
en el carbono: carbon, (Figura 3.1.2). Todo ello provoca que las emi-

siones de contaminantes a la atmosfera (sobretodo pero no solamente, de CO2 ) se
incrementen a
no a a
no. El aumento de las emisiones es responsable de numerosas
consecuencias como son el continuo calentamiento del planeta, variaciones tanto
reduccion
en las cantidades como los patrones de precipitacion, de los casquetes
del nivel del mar, aumento del nivel de acidez de los oceanos,
polares, elevacion
etc. . . Finalmente, y si no se pone remedio al continuo aumento de dichas emisio-
un impacto directo y a nivel mundial en los recursos hdricos disponibles,
nes, tendra
de animales y plantas y finalmente en la futura
la cadena alimentaria, la extincion
existencia de la raza humana.

22

Captulo 3. Introduccion


El clima mundial se ha mantenido particularmente estable en los ultimos 10.000
a humana. Hoy en da empiezan
nos, lo que ha permitido el desarrollo de la civilizacion
cambiando [4]. Este cambio es uno de
a aparecer claros signos de que el clima esta
importantes al que se enfrenta la humanidad. La concentracion
los desafos mas de

dioxido
de carbono (CO2 ) en la atmosfera
ha alcanzado una cifra record a nivel
mundial, 391 ppm [5]. Esto significa un crecimiento de casi el 40 por ciento desde la
industrial (280 ppm) y la cifra mas
revolucion alta de los ultimos
650.000 a
nos. Como
consecuencia, la temperatura media global se ha incrementado en 0,8C desde la

epoca preindustrial [6].

el Intergovernmental Panel on Climate Change (IPCC) co-


En 1988 se constituyo

mo herramienta internacional contra el cambio climatico. El objetivo de este grupo

es el analisis cientfica, tecnica
de la informacion
y socioeconomica relevante para
de los elementos cientficos relativos al cambio climatico
la compresion de origen an-

tropogenico, as como sus posibles repercusiones , riesgos y sus posibilidades de ate-
y de adaptacion
nuacion al mismo. Dicho grupo esta
constituido por dos importantes
organismos mundiales, Unites Nations Environment Programme (UNEP) y World Me-
teorological Organization (WMO) [7].

es el establecimiento de posibles esce-


Una de las misiones de esta organizacion
de los niveles de contaminacion
narios futuros referentes la evolucion atmosferica,


cambio climatico, etc. . . Los principales escenarios presentados son los siguientes [8]:


A1 Mundo futuro con un rapido
crecimiento economico, mun-
una poblacion

dial que alcanza su valor maximo hacia mediados de siglo y disminuye poste-

riormente, una rapida de tecnologas nuevas y mas
introduccion eficientes. La
familia A1 se desarrolla en tres grupos que describen direcciones alternativas

del cambio tecnologico en el sistema de energa. Los tres grupos A1 se dife-
tecnologica:
rencian en su orientacion intensiva de combustibles de
utilizacion
(A1FI), utilizacion
origen fosil de fuentes de energa renovables (A1T), o utiliza-
equilibrada de todo tipo de fuentes (A1B)
cion


A2 Mundo muy heterogeneo cuyas principales caractersticas son la autosufi-
de las identidades locales y una poblacion
ciencia, la conservacion mundial

23
3.1.Energa en el Mundo


en continuo crecimiento. El crecimiento economico por habitante as como el

cambio tecnologico mas
estan fragmentados y son mas
lentos que en otras
lneas evolutivas

mundial alcanza su valor maximo


B1 Mundo donde la poblacion hacia media-

dos de siglo y disminuye posteriormente. Similar al modelo A1, pero con rapidos

cambios de las estructuras economicas orientados a una economa de servi-
acompa
cios y de informacion, menos intensiva de los
nados de una utilizacion
de tecnologas limpias con un aprovechamien-
materiales y de la introduccion
to eficaz de los recursos.

B2 Mundo en el que predominan las soluciones locales a la sostenibilidad eco-



nomica, aumenta pro-
social y medioambiental. Es un mundo cuya poblacion
gresivamente a una ritmo menor que en A2, con unos niveles de desarrollo

economico
intermedios, y con un cambio tecnologico
menos rapido
y mas
diverso que en las lneas evolutivas B1 y A1.

establecer el obje-
En el Protocolo de Kyoto (1996), las Naciones Unidas decidio
tivo de intentar mantener el calentamiento global por debajo de los 2C (cuantifi-

cado desde la epoca industrial) [9]. Aunque recientemente,
anterior a la revolucion
y desde un punto de vista cientfico, se ha establecido como lmite seguro para el
desarrollo tanto de la vida animal, vegetal como la vida humana, mantener el calen-

tamiento maximo
entre 1 y 1,5C. Cualquier calentamiento superior a este presentara

consecuencias catastroficas para la vida global del planeta [10].

Diferentes predicciones se han realizado a partir de los diferentes escenarios pre-


del calentamiento global. En la Figura
sentados con el fin de estudiar la evolucion
de las concentraciones de CO2 en el siglo XXI en ca-
3.1.6 se muestra la evolucion
da uno de los escenarios presentados. En la Figura 3.1.7 se muestra el calentamiento
producido en cada uno de dichos escenarios.

Como se puede observar en todos los escenarios propuestos, el objetivo marcado



de 1,5C se sobrepasa con creces. El unico caso, donde se conseguira cumplir dicho
objetivo, es aquel que consiguiera mantener constantes los niveles de CO2 corres-
pondientes al a de
no 2000. Esto significa que se deben mantener la concentracion

24

Captulo 3. Introduccion

Figura 3.1.6: Concentraciones CO2 (ppm) en cada uno de los escenarios

(391 ppm) se encuentra


CO2 entre 350-400 ppm [11]. Actualmente la concentracion

muy proxima
al valor maximo deseado.

La pregunta que el mundo debe hacerse es: es posible mantener el crecimiento



economico esperado respetando los niveles de emisiones de sustancias contamina-
tes?

tardar en 2015, la cantidad de emisiones


Para lograrlo es imperativo que a mas
paulatina (actualmente las emisiones aumentan a
inicie una disminucion no a a
no)
hasta quedar por debajo de las 23 GT (23 GigaToneladas) de CO2 desde el a
no 2030

en adelante. Actualmente se vierten en la atmosfera 27 GT de CO2 [3]. El problema
es que el crecimiento de la economa mundial exige cada a
no un aumento en la
de energa de un 1,8 % sobre el acumulado total producido. Por lo que
produccion
para el a que estar produciendo alrededor de un 60 % mas
no meta 2030 habra de
energa (aproximadamente 18.000 MTep). Actualmente en todo el mundo se produ-
cen 12.200 MTep por a
no y se emiten 27 GT de CO2 . Si nada cambiara, para suplir las

necesidades energeticas del 2030 la humanidad emitira 43 GT de CO2 ese a
no, el do-

25
3.1.Energa en el Mundo

Figura 3.1.7: Calentamiento Global en cada uno de los escenarios

esperada. En conclusion,
ble de la contaminacion para evitar el colapso ambiental

y al mismo tiempo mantener el crecimiento economico, para el 2030 la humanidad
un 60 % mas
requerira de produccion
energetica
pero contaminando un 15 % menos
de lo que lo hace actualmente.

3.1.3.
Principales Mejoras Energeticas


Las necesidades energeticas adicionales se cuantifican aproximadamente en 550
MTep/a
no. Existen diferentes alternativas que posibilitaran cumplir el objetivo energeti-
ampliamente estudiadas son el
co sin incumplir el medioambiental. Las opciones mas
de energas renovables, la instauracion
desarrollo y utilizacion de captura y almace-

namiento de CO2 o la mejora de la eficiencia energetica [12].

26

Captulo 3. Introduccion

Diversos estudios [3, 13, 14] muestran que, con las tecnologas actuales, la insta-
presente y futura de energas renovables funcionando a maxima
lacion capacidad
no bastara para proporcionar toda la energa adicional necesaria para asegurar el

crecimiento economico mundial. Las energas renovables seran capaces de propor-
cionar aproximadamente el 40 % de la energa adicional necesaria. La tecnologa de
captura y almacenamiento de CO2 se encuentra, en la actualidad, en plena fase de
y su implantacion
investigacion en la industria no esta
muy extendida. La captura y
de CO2 requiere mucha energa y aumentara las necesidades de com-
compresion
en un 25-40 % [15]. Se estima que, adicionando
bustibles de una central de carbon

estos y otros costos del sistema de captura de CO2 , el coste de la energa de las

nuevas centrales electricas entre aumente entre un 21 % y un 91 % [15, 16]. El IPCC
estima que la economa potencial de esta tecnologa podra encontrarse entre el
del carbono hasta 2100 [15]. En conclusion,
10 % y el 55 % del total de mitigacion esta

tecnologa no se puede considerar hoy en da como una tecnica importante en la
de las emisiones de CO2 , aun
reduccion siendo especialmente prometedora en la

lucha contra el cambio climatico en el futuro.

Todo ello indica que el principal campo que puede llevar a la humanidad al cum-

plimiento, tanto de los objetivos economicos como medioambientales, es la mejora

de la eficiencia energetica. E sta es un campo que tiene aplicaciones en todos y ca-

da uno de los sectores economicos, aunque es en la industria, en el transporte y en el

acondicionamiento de edificios donde mayor ahorro energetico es posible obtener
mediante la mejora de la eficiencia.


El consumo energetico del sector industrial a nivel global representa aproximada-

mente el 28 % del consumo energetico total y dentro de este sector, la industria qumi-

ca representa aproximadamente el 20 %. Si se realiza un simple calculo de porcenta-

jes, el consumo energetico de la industria qumica en el mundo representa aproxima-
damente el 5,6 % de toda la energa consumida en el mundo (unos 0,84 TW/a
no).


Ahora bien, al analizar el consumo energetico de la industria qumica, se observa

que los principales consumidores energeticos son los procesos de separa-
en esta
de compuestos. Concretamente, Mix y col. [17] determinaron que el 60 % de
cion
la energa total consumida en la industria qumica y petroqumica era empleada

27
3.1.Energa en el Mundo

El proceso de destilacion
en procesos de destilacion. es uno de los mas
importan-

tes metodos de compuestos. Se estima que unicamente
de separacion en Estados
[18], lo que
Unidos existen operando alrededor de 40000 columnas de destilacion
y
representa aproximadamente el 90 % de la totalidad de procesos de separacion
de compuestos. En conclusion,
purificacion se estima que en el proceso de destila-
se consume aproximadamente el 3 % de la energa mundial [19, 20]. Esto unido
cion

al hecho de que esta energa es generada principalmente mediante la combustion

de combustibles fosiles, hacen que cualquier mejora que permita reducir el consumo

energetico un importante impacto tanto en el coste economico
de estas, supondra
como medioambiental global.

28

Captulo 3. Introduccion

3.2 Simuladores de Procesos

dise
La simulacion, de una planta qumica, constituida por varias
no y optimizacion

unidades interconectadas por corrientes de proceso, forman el nucleo de activida-
des realizadas en la Ingeniera de Procesos. El correcto desarrollo de dichas tareas re-
de multiples
quiere de la realizacion subtareas, donde se puede encontrar balances

de materia y energa, dimensionado de equipos, o calculo y
de coste de instalacion
entre otras.
operacion,


Historicamente, de dichas tareas ha sido un proceso complicado y
la realizacion
tedioso, debido a que los modelos desarrollados para representar los procesos qumi-
cos son enormemente complejos. Afortunadamente el desarrollo de los ordenadores
facilitar en gran medida el estudio de estos procesos. Hoy en da, existen nu-
permitio

merosos programas informaticos con la capacidad de realizar estas tareas, especial-
de procesos qumicos. Este tipo de programas
mente los destinados a la simulacion
se denominan Simuladores de Procesos Qumicos.


Un simulador de procesos es capaz de realizar un buen numero de actividades,

las principales son el calculo de las propiedades fsicas tanto
riguroso y descripcion
de compuestos puros como mezclas, el modelado riguroso de unidades de proce-
de grandes sistemas de ecuaciones algebraicas y
sos y la capacidad de resolucion
a traves
diferenciales. Ademas, de un simulador de procesos es posible obtener una
lo que facilita de forma sustancial
imagen simple y global del proceso en ejecucion,
tecnica
la comprension de la planta de proceso estudiada, y sobre esta base servir
para la mejora continua del proceso o el desarrollo de nuevos procesos.

Dentro de la Ingeniera de Procesos podemos diferenciar tres tipos diferentes de


de dise
problemas: problemas de simulacion, (Figura 3.2.1).
no y de optimizacion

es aquel donde todas las corrientes de entrada a


Problema de Simulacion:

una determinada unidad, as como los parametros que definen dicha unidad,
completamente especificados. En otras palabras, un simulador calcula
estan
de las entradas y los parametros
todas las salidas de una unidad en funcion de

29
3.2.Simuladores de procesos

dicha unidad (No hay grados de libertad). Normalmente este tipo de proble-
mas presentan estructuras entrada-salida fijas

Problema de Dise salvo por la diferen-


no: es similar al problema de simulacion,

cia de que, en este, ciertas variables de dise
no de las unidades de proceso
son desconocidas (por ejemplo, volumen de reactor, no de pisos de una co-

lumna, area de un intercambiador de calor, . . . ). Para resolver el problema es
necesario a
nadir restricciones a diversas variables en las corrientes de salida,
de tal forma que estas restricciones compensen los grados de libertad extras
en el problema

es aquel donde se pretende evaluar el rendimiento


Problema de Optimizacion:
tanto de variables de dise
del proceso en funcion no como de variables de las
corrientes de entrada con el fin de minimizar su comportamiento de acuerdo

a criterios economicos, ambientales o sociales. En este tipo de problemas pue-

de presentar tanto ecuaciones de igualdad como desigualdad y su numero

puede ser igual o superior al numero de variables a calcular

Todos estos tipos de problemas son representativos de estructuras rgidas entrada-


y
salida. Si se considera una estructura abierta con diversas alternativas de operacion
de equipos para cada una de las tareas del sistema, el problema se transforma en un

Problema de Sntesis. Mediante este optima
se pretende determinar la configuracion
aspecto clave del sistema, ya sea su economa, su
que minimice o maximice algun
impacto ambiental, etc. La dificultad de este tipo de problema radica en el hecho de
del equipo seleccionado. Una herramienta
que las ecuaciones cambian en funcion
de este tipo de problemas es la utilizacion
que facilita enormemente la resolucion de
simuladores de procesos

Las principales ventajas derivadas del uso de simuladores de procesos por orde-
nador son las siguientes:

Facilidad en el estudio completo y sencillo de todas las corrientes de materia y


energa del proceso

y separacion
Favorecer el estudio de formacion de subproductos e impurezas

30

Captulo 3. Introduccion

(a)

(b) (c)

Figura 3.2.1: Problemas en Ingeniera de Procesos: a) Dise c) Optimi-


no, b) Simulacion,

zacion

del comportamiento de la planta ante cambios en las corrientes


Evaluacion
de entrada (alimentos) o la demanda de productos

Mejora de la seguridad y el control de procesos

del rendimiento economico


Optimizacion de la planta

31
3.2.Simuladores de procesos

3.2.1. Historica
Revision

por ordenador comenzo


La historia de la simulacion en 1946 con la presentacion

del primer ordenador de uso general, el ENIAC 2 [21]. El primer uso de este fue llevado

a cabo por el matematico en el modelado del proce-
John von Neumann y se centro
nuclear en el Proyecto Manhattan. Los intentos iniciales de simu-
so de detonacion
por ordenador de sistemas de ingeniera qumica, se realizaron a principios de
lacion

la decada
de los 50 y se limitaron casi exclusivamente a las computadoras analogicas

[22]. A finales de la decada significantes
de los 50 tuvo lugar uno de los avances mas
digital con el desarrollo de lenguajes de programacion
en simulacion orientados a
procedimientos tales como FORTRAN 3 . Gracias a este lenguaje, en 1961 y gracias al
el que puede ser considerado el antecesor de
Profesor Paul T. Shannon se desarrollo
los simuladores de procesos, el programa PACER 4 . El primer caso practico
(1964) fue
de una planta de acido
la simulacion
sulfurico de 300 t/da. El
con una produccion
simultanea
problema simulado requera la resolucion de 500 ecuaciones y un total de

1000 variables asociadas a corrientes y 200 parametros de equipos. El siguiente paso
de procesos se dio en 1966 cuando
llevado a cabo en el campo de la simulacion
un programa generico
la empresa Simulation Science comercializo
de simulacion
Este programa fue el nucleo
de columnas de destilacion.
del paquete de simulacion
de procesos PROCESS. Tres a tarde, en 1969, la empresa ChemShare pre-
nos mas
DESIGN, un programa de simulacion
sento de procesos para aplicaciones de gas y

petroleo.
En la decada
siguiente aparecieron los primeros programas de simulacion
orientados a ecuaciones. Algunos de ellos fueron SPEEDUP (Imperial Collage, Lon-
dres, Reino Unido) y TISFLO (DSM, Pases Bajos). En 1976, el Departamento de Energa
de EE.UU. y el MIT lanzaron conjuntamente el proyecto ASPEN, el cual dio lugar en
1985 al programa ASPEN PLUS lanzado por AspenTech.


La decada
de los 80 fue una decada en el tema de la
de continua evolucion

informatica. Varios hitos revolucionaron considerablemente este campo, uno fue la
de los ordenadores personales, otro fue la aparicion
aparicion de las redes mundia-
2
Acronimo de Electronic Numerical Integrator And Computer, fue construido en la Universidad de Pensilvania
por John Presper Eckert y John William Mauchly, pesaba 27 toneladas y ocupaba una superficie de 167 m2
3 FORmula TRANslator. Es un lenguaje de programacion para el desarrollo de aplicaciones
Proviene del ingles

matematicas de alto nivel.
y cientficas, fue el primer lenguaje de programacion
4
Acronimo de Process Assembly Case Evaluator Routine

32

Captulo 3. Introduccion


les de intercambio informatico (precursoras de internet), todo esto junto a la llegada
de las estaciones de trabajo (1985) y de un nuevo sistema operativo multitarea, UNIX,

produjo una autentica en la computacion
revolucion cientfica. Pocas fueron las com-
nas de software cientfico que sobrevivieron a estos cambios profundos. La capa-
pa
a esta nueva era se baso
cidad de adaptacion principalmente en la reutilizacion
de
las viejas rutinas desarrolladas en FORTRAN en el entorno de una nueva programa-
orientada a objetos. Precisamente en esta epoca
cion
se produjo la actualizacion
del simulador PROCESS para convertirse en PRO II.


A principios de los 90, el dominio de los ordenadores personales se consolido.
la relativa estabilizacion
Ademas de los sistemas operativos (dominados por Windows
el desarrollo de una nueva generacion
y UNIX) permitio de simuladores de procesos.

En estos nuevos simuladores, la interfaz grafica a ser la parte central
de usuario paso

del desarrollo informatico. hasta convertirse en
En este periodo HYSYM se desarrollo
el simulador HYSYS. Durante la segunda mitad de los a
nos 90 se produjo la consolida-
de la simulacion
cion de procesos, y las principales compa
nas de software cientfico.
A principios del siglo XXI aparecieron los simuladores de procesos libres, COCO 5 .

de procesos qumicos esta


Hoy en da, la comunidad dedicada a la simulacion a-
de diferentes componentes de software a traves
vanzando hacia la interconexion del
proyecto CAPE-OPEN (CO) [23]. La red de laboratorios CAPE-OPEN (COLaN) es una
internacional para el gestion
organizacion de las normas que deben aplicarse en el
de procesos libre (CO). El simulador COCO explora la idea del
software de simulacion
proyecto CO, es libre, gratuito, posee una estructura modular que opera de forma
secuencial y trabaja en estado estacionario. Como se trata de un entorno de mo-
delado de diagramas de flujo abierto, permite al usuario a
nadir nuevas operaciones

unitarias o paquetes termodinamicos con total libertad. COCO comprende cuatro
componentes principales:


COFE (CAPE-OPEN Flowsheet Environment): La interfaz grafica donde realizar
del diagrama de flujo, consta ademas
la representacion de algoritmos de re-
secuencial con corrientes de corte automatizadas. Este componente
solucion
puede ser combinado junto con otros simuladores
5 Cape Open to Cape Open
Acronimo del ingles

33
3.2.Simuladores de procesos

TEA (Thermodynamics for Engineering Applications): esta basado en el librera



termodinamica de ChemSep LITE, el cual es un simulador libre de equilibrio
y de extraccion
para columnas de destilacion lquido-lquido [24]. Incluye una
de 430 compuestos qumicos. El paquete ofrece mas
base de datos de mas

de 100 metodos
de calculo de propiedades con sus derivadas analticas o

numericas


COUSCOUS: es el paquete basico de operaciones unitarias que incluye CO-
CO. Consta en un conjunto de operaciones unitarias tales como, mezcladores
y divisores de corrientes, intercambiadores de calor, compresores, bombas y
reactores

CORN (CAPE-OPEN Reaction Numerics): es el paquete incluido en COCO para



simular reacciones cineticas y de equilibrio

3.2.2. Arquitectura de los Simuladores Qumicos

3.2.2.1.
Clasificacion

Los modelos de procesos pueden clasificarse en tres grupos principales, depen-


temporal
diendo del tipo de variable (modelos continuos o discretos), la evolucion

(dinamicos
o estacionario), y el grado de incertidumbre (deterministas o estocasti-
cos).

el tipo de variables:
Segun

Los modelos discretos hacen referencia a sistemas que pueden tomar cierto
conjuntos de valores discretos en instantes discretos del tiempo. Es decir,
un sistema discreto es aquel en el que las variables de estado cambian ins-

tantaneamente en puntos separados en el tiempo. El seguimiento de los cam-
de que
bios de estado requiere la identificacion es lo que causa el cambio y
cuando lo causa, lo que se denomina suceso, las ecuaciones del modelo se

convierten entonces en las ecuaciones y relaciones logicas que determinan

las condiciones en que tiene lugar la ocurrencia de un suceso. La simulacion

34

Captulo 3. Introduccion

discreta consiste en el seguimiento de los cambios de estado del sistema que


tienen lugar como consecuencia de la ocurrencia de una secuencia de suce-
sos.

Los modelos continuos son aquellos donde las variables implicadas son conti-
nuas, es decir aquella que puede tomar cualquier valor dentro de un rango
determinado, cambiando de forma continua en el tiempo. Generalmente, es-
te tipo de variables suele estar asociada a ciertas propiedades fsicas como
etc. . .
son temperatura, flujos, presion,

temporal:
la evolucion
Segun


Los modelos dinamicos son aquellos donde algunas de las variables asociadas
son dependientes del tiempo, es decir evolucionan temporalmente, no per-

maneces estaticas. Los problemas tpicos que utilizan este tipo de modelado
son los problemas de control de procesos, de arranque y parada de procesos,
dise
no de procesos en discontinuo, etc. . .


Los modelos estaticos son aquellos en los que las variables no dependen del
tiempo. Aunque este comportamiento no suele darse en la vida real, numero-
sos procesos son dise
nados para permanecer en estado estacionario y utilizar
el control para conseguir tal fin. Por ello, los principales problemas donde sue-
len emplearse estos modelos son problemas de equilibrio, dise
no de procesos
en continuo,etc. . .

el grado de incertidumbre:
Segun

Los modelos deterministas son aquellos que predicen el futuro como algo fijo,
donde la incertidumbre y el azar no tienen cabida. En estos modelos, una mis-
invariablemente la misma salida. Este tipo de modelos
ma entrada producira
de leyes fsicas mediante
han sido utilizados generalmente para la aplicacion

ecuaciones algebraicas o diferenciales, estudio de velocidades de reaccion,
etc. . .

35
3.2.Simuladores de procesos


Los modelos estocasticos representativos del mundo real. En estos mo-
son mas
delos, a diferencia de los modelos deterministas, se considera que el valor de

alguno de los parametros ciertas variables se ven afecta-
es incierto. Ademas,
distribuciones de probabilidad que hace que los resultados obteni-
das segun
dos no sean fijos sino que deben representarse como un valor medio indicando
estandar
su desviacion
o su varianza. Este tipo de modelo son cada vez mas
realistas como puede ser las fluctua-
utilizados para representar modelos mas
ciones del mercado, etc. . .

pensados para modelos


La mayora de simuladores de procesos disponibles estan
de su evolucion
deterministas, continuos y en funcion temporal, pueden funcionar de

forma correcta tanto para sistemas estacionarios como dinamicos.

muy empleada se basa en dos aspectos importantes del simula-


Otra clasificacion
dor, por un lado como esta configurado y descrito, y por otro como resuelve las ecua-
los simuladores
ciones generadas por el modelo. Dentro del modo de configuracion,

se clasifican como modulares o no modulares, y dentro del metodo se
de resolucion

clasifican como secuenciales y simultaneos. De acuerdo a esto, los simuladores de
procesos se pueden clasificar en tres tipos concretos; modular-secuencial, orientado

a ecuaciones (no modular simultaneo)
y modular-simultaneo.

Un simulador modular-secuencial se basa en el concepto de modularidad, con-



siderando cada uno de los equipos como una unidad de calculo independiente.
Las ecuaciones asociadas a cada equipo (balances de materia y energa, ecuacio-
nes de equilibrio, etc. . . ) son agrupadas en un modulo o subrutina computacional.
As cada modulo calcula las corrientes de salida del equipo a partir de las corrientes

de entrada y los parametros de dise
no (independientemente de donde proceda o

donde se dirige la informacion), sirviendo estas salidas como corrientes de entrada

para el siguiente equipo y as sucesivamente. Este metodo es similar al llevado a ca-

bo tradicionalmente cuando se realizan los calculos manualmente. Esta estrategia de

calculo es utilizada en la mayora de los simuladores de estado estacionario: Aspen,
Chemcad, Hysys, Prosim, Design II o SuperPro.


Los simuladores orientados a ecuaciones describen el proceso total como un uni-
co conjunto de ecuaciones que integra todas las ecuaciones relativas a todos los

36

Captulo 3. Introduccion


equipos y corrientes del proceso y se resuelve de forma simultanea, obteniendo di-
rectamente el valor de todas las variables desconocidas. Esta metodologa es mas
flexible que el Modular-Secuencial, sin embargo requiere mas esfuerzo de programa-
y se consumen mas recursos computacionales. Algunos simuladores que siguen
cion
esta estrategia son Ascend, Abacus o gProms.


Un simulador modular-simultaneo combina la modularidad de los equipos para
agrupar sus ecuaciones propias con los algoritmos altamente eficientes de los simu-
ladores orientados a ecuaciones. Estos utilizan la estructura modular tradicional, pero

a su vez se resuelve un sistema de ecuaciones simultaneas incluyendo a todas las
variables de las corrientes. La diferencia principal consiste en que el sistema de ecua-
facil
ciones a resolver es un sistema simplificado de mas que el sistema de
solucion

ecuaciones que se resolvera en el metodo orientado a ecuaciones. Especficamen-

te, estos simuladores utilizan modelos rigurosos a nivel de las unidades de operacion
los cuales son resueltos secuencialmente y modelos lineales a nivel del diagrama de
flujo, los cuales son resueltos de forma global. Dichos simuladores aprovechan algunas

ventajas del metodo incorporacion
modular-secuencial, tales como la facil de modu-

diagnostico
los unitarios del sistema o la facilidad de localizacion, de fallos,
y solucion

y por otro lado aprovecha la ventaja del metodo orientado a ecuaciones, de un
eficiente de las ecuaciones para que el esquema de calculo
manejo mas modular

funcione mejor en procesos complejos, con multiples recirculaciones o con severas
restricciones de dise simulador de
no. Comercialmente, no existe hoy en da ningun
procesos en el mercado que utilice esta estrategia.

extendido
Debido a que dentro de las opciones comerciales el simulador mas
el simulador utilizado posteriormente en todos
es el modular-secuencial y ademas

los calculos de esta tesis ha sido un simulador modular-secuencial (concretamente
se va a proceder a comentar mas
Aspen-Hysys), a continuacion detalladamente los
rese
aspectos mas nables de su arquitectura.

37
3.2.Simuladores de procesos

3.2.3. Simulador Modular-Secuencial

de una operacion
La simulacion en estado estacionario de una planta qumica
puede ser representada por una sistema de ecuaciones algebraicas no lineales:

f (x) = 0 (3.2.1)

donde f es un vector de funciones y x es un vector de variables que describen


unitaria particular como los parametros
tanto las entradas y salidas de una operacion
de dise de reflujo en
no de tal unidad (por ejemplo, volumen de reactor, relacion
etc. . . ).
columnas de destilacion,

de una plan-
El sistema de ecuaciones generado durante el proceso de simulacion
ta qumica puede contener miles de ecuaciones y variables. Este hecho provoca que
pueda ser extremadamente compleja considerando un unico
su resolucion modelo
para mejorar este problema es la que plantean
para toda la planta. Una solucion

los simuladores modulares, esta de ecuaciones
se basa en realizar una agrupacion
y variables en cada uno de los equipos de la planta. Esto hace que el problema a
resolver se reescriba como:

fi (x) = 0, i = 1, 2, . . . , nu (3.2.2)

siendo fi un subconjunto de vectores f que agrupa las funciones asociadas al



equipo i y nu es el numero total de equipos de la planta.

Estructura interna de un simulador modular comercial La estructura de un simulador


modular-secuencial puede organizarse en tres funciones o secciones perfectamente
diferenciadas:


1. La logica
central o logica general del simulador

38

Captulo 3. Introduccion

encargada de la estimacion
2. Seccion de las propiedades fisicoqumicas


3. La biblioteca de modulos
de equipos, es decir cada uno de los modulos que

representan el comportamiento de valvulas, intercambiadores, sistema de des-
sumadores, divisores, flash, compresores, etc.
tilacion,


Logica general del simulador

a su vez comprende principalmente las siguientes subseccio-


Esta primera seccion
nes:

de entrada
La seccion

de salida de resultados
La seccion


La logica
general de administracion

EL sistema de entrada/salida de datos, es una parte fundamental de todo simu-


lador de uso general. Dicho sistema debe caracterizarse por su flexibilidad y amiga-
bilidad al usuario, facilitando a un usuario inexperto, o poco conocedor del tema,

un facil estos sistemas poseen la capacidad de chequear
uso del sistema. Ademas

al maximo de datos, a los efectos de
las inconsistencias o errores en la introduccion

minimizar el tiempo gastado inutilmente De esta
al emprender tareas de simulacion.
manera, el sumatorio de fracciones molares distintas de la unidad, las temperaturas
absolutas negativas, valores inconsistentes dentro del contexto del problema, grados

de libertad violados al especificar parametros de equipos, etc., son inmediatamente
que emite mensajes de
notificados al usuario por medio de un sistema de validacion
error o advertencia.


La logica propiamente dicha es la que esta
general de administracion encargada

de administrar los distintos procesos que deben ejecutarse para lograr la simulacion
compuestos
de un proceso dado. Generalmente, los diagramas de procesos estan

por un numero
importante tanto de equipos como de corrientes. El metodo de re-
de estos diagramas depende en gran medida del grado de complejidad
solucion
simple a resolver en un simulador modular-secuencial es
que presente. El caso mas

39
3.2.Simuladores de procesos

fluye en
un sistema acclico, sin corrientes recirculadas, ya que el flujo de informacion

una unica Sin embargo, los procesos qumicos en la vida real implican el
direccion.
uso de numerosas corrientes recirculadas, principalmente en sistemas integrados (re-
de reactivos, reutilizacion
cuperacion de servicios energeticos,
etc. . . ). Este aspecto
del diagrama de flujo siendo imposible su solucion
complica mucho la resolucion di-

recta, haciendose de un metodo
necesaria la utilizacion iterativo con el fin de resolver
Para la implementacion
la simulacion. de estos metodos
es necesario tener en cuenta
operaciones unitarias pueden
una serie de consideraciones previas tales como, que

ser resueltas simultaneamente? Esta caracterstica adquiere el nombre de Particio-
orden deben ser resueltos los
nado en el campo de la Ingeniera Qumica. En que
diferentes grupos que forman el diagrama? Este aspecto es conocido como Orden
de Precedencia. Y finalmente, cuales son las corrientes claves (corrientes de corte)
facilitan la resolucion
que de cada una de las particiones obtenidas? Este aspecto es
conocido como Rasgado.

esta parte del simulador debe procesar el diagrama de flujo, decidir


En conclusion,
si puede resolverse en una secuencia lineal o si existen reciclos, seleccionar sobre
variables debera
que iterarse, determinar en funcion
de las corrientes de corte el
resueltos los equipos, etc. Existen numerosos algoritmos para
orden en el cual seran

llevar a cabo estas tareas. La mayora de estos fueron desarrollados en las decadas

de los 60 y 70. Los metodos de particionado y orden de precedencia desarrollados se

basan en la idea de la busqueda muy relacionados con la teora
de caminos y estan

de grafos. Un resumen detallado de estos metodos se muestra en los trabajos de
Sargent y Westerberg [25], Ledet and Himmelbrau [26], Tarjan [27] o Kehat y Shacham
[28]. Una vez determinadas las diferentes particiones, el siguiente paso es determinar

el mnimo numero Los
de corrientes de corte necesarias para la correcta resolucion.
principales algoritmos detallados en la literatura son los propuestos por Christensen y
Rudd [29], Barkley y Motard [30], Pho y Lapidus [31] o Upadhye y Grens [32].

necesaria (numero
Una vez determinada la informacion de particiones, orden de
precedencia y corrientes de corte) se lleva a cabo el proceso iterativo para resolver la
El problema radica en que los metodos
simulacion. exacta
iterativos hallan la solucion
en un numero
(si eso ocurre) solo infinito de iteraciones. Esto desde un punto de vista

practico
es totalmente inviable, ya que requerira un tiempo infinito de calculo. La

40

Captulo 3. Introduccion

a este problema es el empleo de metodos


solucion de convergencia que limiten el

numero de la simulacion.
de iteraciones necesarias para la resolucion Los principales

metodos
detallados en la bibliografa son el metodo directa, el metodo
de sustitucion

de Wegstein, el metodo
de Broyden y los metodos de Newton-Raphson.

Finalmente, una vez realizado el proceso iterativo de acuerdo al criterio de conver-


retener los resultados,
gencia definido, procede a detener el proceso de simulacion,
es decir almacenar en el lugar correspondiente todos los valores convergidos de las

corrientes del proceso, los valores y parametros internos de los equipos, por ejemplo
los perfiles internos de torres, composiciones, caudales, temperatura y presiones de
cada etapa, etc. La Figura 3.2.2 muestra un esquema de las tareas realizadas por

dicho modulo. esta-
Como se puede observar en el esquema existe una etapa mas
blecida para el caso que se quisiera optimizar el dimensionado de los equipos del
incluira
proceso. En este caso, el proceso de iteracion un bucle exterior donde se pro-
a modificar las dimensiones de los equipos con el fin de conseguir el optimo
cedera
del proceso.

de un simulador modular-secuencial
Figura 3.2.2: Esquema de operacion

41
3.2.Simuladores de procesos

encargada de la estimacion
Seccion de las propiedades fisicoqumicas

de pro-
En un simulador de procesos de alcance general, el sistema de estimacion
piedades fisicoqumicas es prioritario. No basta con que un simulador cuente con los

mejores algoritmos de calculo, rapidos
los mas y estables, ya que si se realiza una
incorrecta del metodo
eleccion de propiedades los resultados obte-
de prediccion
tendran
nidos en la simulacion poco o nada que ver con la realidad. Dado que se
simular diversos tipos de mezclas, de comportamiento ideal y no ideal, siste-
deberan

mas bifasicos
y/o trifasicos,
potencial presencia de electrolitos
y/o solidos, constantes
de equilibrio en reacciones qumicas, etc; resulta indispensable disponer de un banco
de un numero
de modelos para la estimacion importante de propiedades fisicoqumi-
cas en todas las situaciones mencionadas. Un peque
no resumen de las diferentes
de procesos qumicos se
propiedades fisicoqumicas que intervienen en la simulacion
muestra en la Tabla 3.2.1.

Tabla 3.2.1: Tipos de propiedades fisicoqumicas en un simulador de procesos qumicos

Tipo Propiedades Especficas

de vapor, coeficientes actividad y fugacidad, solubilidad (Constantes


Presion
Equilibrio de fase
de Henry, coeficientes de Ostwald o Bunsen), parametros
de interaccion
y ebullicion
binarias, puntos de fusion
Densidad, Volumen Molar, Compresibilidad,Propiedades Crticas, Factor
Comportamiento PVT

Acentrico


Propiedades Termicas
Capacidad Calorfica, Calor Latente, Conductividad Ionica, Entalpa, Entropa

Propiedades de Transporte
Viscosidad, Conductividad Termica,
Coeficientes de Difusion

Constante de equilibrio, Constantes de Asociacion/Disociaci Entalpa de
on,
Equilibrio qumico Entalpa de Combustion,
Formacion, Calor de Reaccion,
Energa libre de Gibbs
Velocidad de reaccion
de formacion,
Propiedades de Interfase Superficial
Tension

Propiedades Moleculares
Coeficientes viriales, Radio y Volumen Ionico, Peso Molecular, Momento dipolar

Lmite maximo
Lmites de Explosion, y mnimo de Inflamabilidad, Temperatura de
Propiedades de Seguridad
Toxicidad, Concentracion
Inflamacion, TLV

de un proceso requiere dispo-


Como ya se ha comentado, cualquier simulacion

ner del valor de las propiedades fisicoqumicas y termodinamicas de las mezclas de
compuestos que circulan, o han de circular, entre los distintos equipos de la planta, en
presion
todas las condiciones de composicion, y temperatura que puedan llegarse a
de la misma. Esto, obviamente, es practicamente
verificar en la operacion imposible

y se debe hacer uso de tecnicas que permitan estimar esos valores.
de prediccion

42

Captulo 3. Introduccion

de procesos qumicos incluye procesos tales como destilacion,


La simulacion eva-
extraccion,
poracion, etc. . . . El dise
no de equipos para estas operaciones requiere
exacto posible de las propiedades parciales de equilibrio
un conocimiento lo mas
el calculo
de los componentes de una mezcla. Ademas, de los balances de energa
de las entalpas de mezclas. La prediccion
es inadecuado sin la estimacion de estas
de modelos termodinamicos.
propiedades se lleva a cabo a traves


Los modelos termodinamicos disponibles en un simulador de procesos son numero-

sos. Este hecho se debe a que no existe hoy en da un unico
modelo termodinamico
que represente fielmente el comportamiento de todos los compuestos y mezclas po-
sibles en todo el rango de presiones y temperaturas, existiendo numerosos modelos

diferentes validos para ciertas mezclas de compuestos en unas determinadas con-

diciones. Los modelos termodinamicos se clasifican en cinco categoras diferentes,
modelos que utilizan Ecuaciones de Estado, modelos que utilizan Coeficientes de Ac-
de Vapor, modelos Semiempricos y modelos Especficos.
tividad, modelos de Presion
Un peque
no resumen de los principales modelos termodinamicos se muestran en la
Tabla 3.2.2. Comentar que debido a que el estudio de los diferentes modelos no forma
parte importante de este trabajo no se va a proceder a presentar cada uno de ellos
informacion
detalladamente. Para mas sobre los diferentes modelos termodinamicos,


vease [33-35]


Tabla 3.2.2: Principales Modelos Termodinamicos disponibles en los simuladores co-
merciales
Categora Nombre Caractersticas


Este metodo combina las leyes de Raoult y Henry. Esta
recomendado para sistemas cercanos a la idealidad
Modelos Ideales Ideal
(operaciones a vaco, bajas temperaturas, hidrocarburos y
gases ligeros

Ideal para los calculos
de VLE y calculo de la densidad de
de
Ecuacion a gran
lquidos para sistemas de hidrocarburos. Aplicacion
Peng-Robinson (PR)
Estado rango de P y T. Indicado para aplicaciones petroqumicas

(petroleo, gas, etc)
Peng-Robinson-Stryjec- del modelo PR para su aplicacion
Extension en sistemas
Vera (PRSV) moderadamente no ideales
Resultados comparables con el modelo PR, pero en un rango
Soave-Redlich-Kwong menor de P y T. No recomendable para sistemas no ideales
(SRK)
tales como alcoholes, acidos, u otros. Mejores resultados que
PR para sistemas polares

43
3.2.Simuladores de procesos


Tabla 3.2.2: Principales Modelos Termodinamicos disponibles en los simuladores co-
merciales
Categora Nombre Caractersticas

exacto para sustancias no polares.


Modelo general mas
Lee-Kesler-Plocker
Recomendado para torres de etileno
del modelo SRK que mejora los calculos
Extension VLE para
Kabadi-Danner
sistemas Agua-Hidrocarburos

Modelos aplicados para aguas acidas que contienen H2 S,
Sour PR/Sour SRK
CO2 o NH3 a bajas y medias presiones

del modelo RK que mejoran los calculos


Extension VLE para
Zudkevitch Joffee

sistemas de hidrocarburos y sistemas que contienen hidrogeno

Schwartzentruber- Recomendado para sistemas altamente no ideales a altas


Renon temperaturas y presiones, como puede ser sntesis de metanol
supercrtica
y extraccion

Primer modelo basado en coeficientes de actividad.


Coeficientes de
Wilson recomendado para sistema altamente no ideales. No
Esta
Actividad
puede ser utilizado en sistemas con equilibrio lquido-lquido

NRTL (Non Random del modelo de Wilson. Capaz de representar VLE, LLE,
Extension
Two Liquid) VLLE. Recomendado para sistemas altamente no ideales

Recomendado para sistemas altamente no ideales,


Van Laar especialmente aquellos que muestran LLE con importantes
desviaciones de la ley de Raoult
sin base teorica
Ecuacion e
para la estimacion
pero muy util
Margules de datos. No aconsejable para representar
interpolacion
sistemas lquido-lquido inmiscibles

Capaz de representar VLE, LLE, VLLE. Se emplea en sistemas


UNIQUAC poco ideales con un amplio rango de compuestos como

agua, alcoholes, nitrilos, aminas, esteres, etc. . .

Antoine con comportamiento ideal


Modelo aplicable a baja presion
de Vapor
Presion
Estrictamente aplicable a sistemas de hidrocarburos pesados a
Braun K10
bajas presiones
Estrictamente aplicable a sistemas de hidrocarburos pesados a
Esso Tabular
bajas presiones
Modelo semiemprico para hidrocarburos pesados para
Semiempricos Chao-Seader temperaturas entre -18 y 260C y presiones menores a 10000
kPa
del modelo Chao-Seader, adapatado a altas
Extension
Grayson-Streed concentraciones de H2 y temperaturas entre -18 y 425C y
presiones menores a 20000 kPa
44

Captulo 3. Introduccion


Tabla 3.2.2: Principales Modelos Termodinamicos disponibles en los simuladores co-
merciales
Categora Nombre Caractersticas


Modelo termodinamico para sistemas de procesos con aminas
Especficos Amine Package
(MEA, DEA, TEA, MDEA, DGA y DIPA)


Modelo termodinamico del
especfico para la simulacion
ASME Steam
comportamiento del agua y vapor en un amplio rango de P y T


Modelo termodinamico de
especifico para la simulacion
Glycol Package mezclas agua-TEG (trietilglicol). recomendado para el proceso
de gas natural
de deshidratacion
de propiedades de equilibrio de fases
Modelo para prediccion
OLI Electrolyte y reacciones de compuestos electrolitos en disoluciones
acuosas
del modelo NRTL para su aplicacion
Extension a disoluciones
Electrolyte NRTL
acuosas de especies electrolticas
BWR
Modelo modificado de la ecuacion
MBWR a hidrocarburos
(Benedict-Webb-Rubin) para su aplicacion
ligeros y gases


Desafortunadamente, no existe un modelo termodinamico
perfecto, ya que ningun

modelo teorico es capaz de simular eficientemente y de forma realista todos los siste-
mas posibles en todas las condiciones imaginadas. Alternativamente, simuladores co-

mo gPROMS ofrecen un nuevo sistema para predecir las propiedades termodinami-
cas de mezclas complejas basadas en modelos moleculares y considerando sus in-

teracciones con otras moleculas, denominado Teora Estadstica de Fluidos Asociados

[36, 37]. La tendencia en este campo se dirige hacia simulaciones de dinamica mo-

lecular que puedan predecir las propiedades fsicas y termodinamicas utilizan-
solo

do distribuciones moleculares microscopicos.
El fundamento teorico de este enfoque
es la teora de las soluciones de Kirkwood-Buff. Aunque este es un paso prometedor
para proporcionar simuladores de procesos totalmente fiables, actualmente la imple-
de esta nueva metodologa en los simuladores comerciales se encuentra
mentacion

totalmente descartada debido a sus prohibitivos tiempo de calculo. Como ejemplo

sencillo indicar que el calculo de la energa libre de Gibbs de una mezcla de me-

45
3.2.Simuladores de procesos

fija utilizando recursos de computacion


tanol y agua a una concentracion de alto
rendimiento se realiza empleando alrededor de 8 horas [38].


Biblioteca de Modulos de Equipos en un Simulador Modular Secuencial

industrial la materia prima que ingresa (corrientes de entrada


En toda instalacion
en el simulador) sufre gran cantidad de transformaciones hasta alcanzar los productos
deseados (corrientes de salida). Los equipos de proceso (intercambiadores de calor,
reactores, etc.) son los responsables de las transformaciones y
torres de separacion,
constituyen la maquinaria de la planta industrial.

de los diferentes equipos en un simulador es un proceso que requiere


La simulacion
de numerosas subtareas y calculos.
la realizacion En la Figura 3.2.3 se observa un
al
esquema general de funcionamiento en el que se indica el flujo de informacion

resolver un equipo o modulo.

de un equipo concreto
Como se puede observar en dicha figura, la simulacion
de una serie de pasos, concretamente tres. En primer lugar se
implica la resolucion
de todos los datos necesarios tanto los parametros
encuentra la introduccion como
se encuentra la resolucion
las corrientes de entrada al equipo. A continuacion del
de todas las ecuaciones que
equipo, la cual se lleva a cabo mediante la resolucion
describen el comportamiento del equipo: Dentro de las diferentes ecuaciones que
describen el comportamiento de cada uno de los equipos se encuentran:

Balances de materia

Balances de energa

Balances de cantidad de movimiento


Relaciones termodinamicas

Ecuaciones especificas del equipo estudiado


Finalmente y como ultimo y almacenamiento
paso se encuentra la recuperacion

de todos los datos obtenidos para su posterior utilizacion.

46

Captulo 3. Introduccion


Figura 3.2.3: Esquema de calculo en cada uno de los equipos

3.2.4. Ventajas e inconvenientes de un simulador secuencial mo-


dular

La gran ventaja del enfoque modular secuencial es su gran robustez y confiabi-


de muchos a
lidad. Despues
nos de trabajo, el modelo matematico de cada uni-

dad ha sido muy estudiado, y existen por lo tanto programas que realizan el calculo
de manera sumamente eficiente. Sobre el sistema de ecuaciones que representa la
del equipo, dado que se conoce la forma de cada expresion,
operacion se realizan
las transformaciones necesarias para reordenarlo y lograr la mejor estructura y lue-

go, de acuerdo a lo obtenido, se aplica un metodo De
adecuado de resolucion.
este modo se aprovechan las propiedades especiales de un sistema de ecuaciones

47
3.2.Simuladores de procesos


determinado para acelerar la convergencia de los modulos. se conoce el
Ademas,

comportamiento fsico del modulo y los resultados posibles del mismo, lo cual puede
verificacion
ser aprovechado para inicializacion, de resultados, consistencia, etc.


Otra importante ventaja es la facilidad de incorporar nuevos modulos a la simu-
En principio la confeccion
lacion. de la nueva subrutina se puede hacer aparte, sin
afectar el resto del simulador, y realizar sobre la misma todos los tests y verificaciones
en cada uno se puede realizar un analisis
necesarias. Tambien sobre la consistencia y
realidad fsica de los datos que se ingresan y de los resultados obtenidos.

Si bien las principales ventajas de estos simuladores radican en el hecho de poseer


la cual coincide con
una estructura determinada a priori para el flujo de informacion,
responsable de las principales desven-
el flujo material en el proceso. Ello es tambien
en la resolucion,
tajas ya que a partir de la rigidez que tiene el flujo de informacion
no es posible plantear ciertos problemas sin caer en complicadas estructuras itera-

tivas que requieren tiempos de computos de
no realistas. Por ejemplo, la resolucion
problemas de dise
no en los cuales se imponen algunas especificaciones o condicio-
nes sobre el valor de una corriente o sobre el resultado de un equipo puede resultar
extremadamente costosa. Otro tipo de problemas donde se presentas importantes
La evaluacion
dificultades es en los problemas de optimizacion. de los puntos requeri-
de un proceso implica resolver repetidamente el sistema de
dos para la optimizacion
ecuaciones que representa la planta.

48

Captulo 3. Introduccion

3.3 Sntesis y Dise


no de procesos

entretenida de la ingeniera,
La sntesis de procesos es, con diferencia, la parte mas
siendo aquella donde se desarrolla, determina y construye la estructura y condiciones
de un nuevo proceso qumico.
de operacion

El dise
no y la sntesis de procesos parte originalmente del concepto de operacion
que
unitaria. Este concepto fue introducido por Little en 1915 [39], cuando establecio
un proceso qumico puede ser representado por una serie de operaciones unitarias.

Hasta finales de la decada de los 60s, dicho concepto fue la clave del dise
no de

procesos. El termino sntesis de procesosfue propuesto como tal por Rudd et al. en
1973 [40], y desde entonces esta disciplina ha contribuido enormemente al desarro-
llo, dise de los procesos qumicos, siendo considerada actualmente
no y operacion
como la piedra angular del dise
no de estos Harmsen [46], los ahorros
[41-45]. Segun

economicos totales en la industria se cifran entre un 20 y un 60 % al utilizar la sntesis
de procesos en el dise
no de los mismos.

Tradicionalmente, la sntesis de procesos se puede describir como una metodo-



loga sistematica de la configuracion
cuyo objetivo es la obtencion optima
de un

proceso, con el fin de maximizar el rendimiento economico,
la eficiencia energetica
o bien minimizar el impacto medioambiental del mismo. Para ello a partir de todas las
posibilidades imaginables (o disponibles) del proceso -diferentes tipos de alimentos y
operaciones unitarias- se debe obtener un dise
no optimo con el fin de conseguir los
productos deseados, cumpliendo los requisitos establecidos. La Figura 3.3.1 muestra
un esquema simplificado de un problema de sntesis.

3.3.1. Metodologas en la Sntesis de Procesos


Desde la decada de los 70, numerosos y significativos avances se han realizado en

el area de la sntesis de procesos qumicos [41, 44]. Las diferentes metodologas em-

pleadas en la sntesis de procesos, con el fin de minimizar costes energeticos y totales,

se clasifican en tres grupos: los metodos basados en reglas heursticas (aquellas ba-

sadas en la experiencia y el conocimiento), los metodos
basados en la optimizacion

49
3.3.Sntesis y Dise
no de procesos

esquematica
Figura 3.3.1: Representacion simplificada de un problema de sntesis


matematica
(dada una superestructura, esta se optimiza mediante el empleo de pro-

gramas matematicos
lineales o no), y finalmente, los metodos hbridos (aquellos que

combinan aportaciones de ambas metodos,
heurstico y matematico)[45].

3.3.1.1.
Metodos basados en reglas heursticas


Los metodos heursticos se basan en la experiencia acumulada a lo largo del a
nos

por investigadores e ingenieros. El primer intento de desarrollar un metodo
sistematico
basado en la heurstica fue realizado en 1971 por Siirola y Ruud [47]. El objetivo de

este de mezclas multicomponente.


fue el desarrollo de secuencias de separacion
Desde entonces, esta metodologa ha sido utilizada por otros autores en posteriores
investigaciones [48].


No fue hasta la decada la principal metodologa
de los 80, cuando se desarrollo

dentro de este campo: los metodos
heursticos jerarquizados. Esta tecnica descom-
pone el problema completo de dise donde
no en una serie de niveles de decision,
de reglas heursticas, se va estableciendo la configuracion
sucesivamente y a traves
adecuada del proceso. Esta metodologa fue propuesta en 1985 por Douglas
mas

[42, 49], aunque no es unica en la literatura, Smith y Linnhoff [50] propusieron un mo-
delo similar estructurado en capas. Los niveles del modelo presentado por Douglas se
del tipo de decision
clasifican en funcion tomada:

50

Captulo 3. Introduccion

Continuo o Discontinuo
1. Modo de Operacion:

En este primer nivel se decide si el proceso se va a desarrollar en continuo


o discontinuo. Los procesos en discontinuo son caractersticos en la industria

farmaceutica,
plastica
y alimentaria. Estos unicamente son seleccionados si se
cumple al menos una de las siguientes caractersticas:

El producto debe estar en el mercado en un espacio muy corto de tiem-


po. Esto puede ocurrir, por ejemplo, en un carrera industrial por ocupar el
mercado generado por un nuevo producto

Cuando la demanda anual de un producto se cubre con unos pocos



dias de produccion

Cuando el proceso se encuentra en una fase preliminar expuesta a nu-


merosas variaciones y mejoras

Cuando el precio del producto es inmensamente superior al coste de



fabricacion

Cuando el producto presenta una vida corta en el mercado antes de


que otro producto lo sustituya


En la practica y en la gran mayora de procesos industriales, los procesos en
econo-
continuo suelen ser seleccionados, ya que suelen ser generalmente mas
con los costes de operacion
micos en relacion e infraestructuras, incluso en los
procesos industriales a peque
na escala.

2. Estructura Entrada-Salida (Alimentos y Productos)

En este nivel se estudian las diferentes corrientes de alimentos y productos que


se debe considerar la existencia de
intervienen en el proceso global. Ademas

subproductos e inertes, y como estos pueden influir en los diferentes procesos
importante es la recuperacion
qumicos del proceso. Una cuestion de los dife-
de purgas con el fin de evitar la
rentes compuestos, as como la introduccion
en el sistema de inertes o subproductos.
acumulacion

51
3.3.Sntesis y Dise
no de procesos


3. Numero de reactores y Corrientes Recirculadas

En este punto se estudian tanto las diferentes corrientes recirculadas del pro-
se estudia
ceso como los reactores que son empleados en el mismo. Ade-mas
el efecto que presentan, a nivel global, las diferentes condiciones de opera-
de cada uno de los reactores. Estas incluyen desde el efecto diluyente
cion
en el
de los inertes en las corrientes de alimentos, hasta el efecto de la presion
equilibrio qumico alcanzado en cada reactor.


4. Sistema de Separacion

Generalmente, los alimentos no reaccionados se suelen recuperar y recircular


a los diferentes reactores para minimizar el coste en reactivos. Por otro lado, los
productos y subproductos deben ser separados y extrados del proceso. En este
seleccionada como el lugar de
nivel se determina tanto el tipo de separacion

emplazamiento de esta.

de corrien-
Este nivel se descompone a su vez en dos subniveles: recuperacion
de corrientes gaseosas. El proceso de recuperacion
tes lquidas y recuperacion
de corrientes gaseosas suele ser el mas
y separacion costoso economicamen-

te, y por ello se debe estudiar profundamente el efecto de introducir purgas en
sin coste pero con
el sistema, ya que una purga es un proceso de separacion
de
importante efectos en el proceso. Por otro lado, el proceso de separacion

corrientes lquidas suele llevarse a cabo en columnas de destilacion.

5. Red de Intercambiadores de Calor


En esta ultima etapa, y una vez se han tomado todas las decisiones previas del
dise
no del proceso, se pretende establecer y dise
nar la red de intercambiado-
res de calor del sistema con el fin de minimizar los costes de servicios calientes
y fros del sistema.

La metodologa descrita por Douglas es una estrategia donde la toma de deci-


siones se realiza de arriba a abajo. E sta permite la rapida
de diagramas
localizacion
optima
de flujo con buenas prestaciones, cercanos a la configuracion del sistema. Sin
de este metodo,
embargo, la principal limitacion debido a su naturaleza secuencial,

52

Captulo 3. Introduccion

es que se tiende a ignorar las interacciones entre los distintos niveles, lo que produ-

ce que los metodos del dise
heursticos jerarquizados no garanticen la obtencion no

optimo del proceso.


Los metodos heursticos han sido utilizados en numerosas aplicaciones, como son
[48, 51], diagramas de procesos [47], reduc-
la sntesis de secuencias de separacion
de residuos [52] o dise
cion
no de procesos metalurgicos [53]. Un ejemplo detallado
de esta metodologa fue presentada por Douglas [49], en el caso de
de la utilizacion
de tolueno. Posteriormente,
la sntesis de benceno a partir de la hidrodealquilacion
la metodologa presentada por Douglas ha sido extendida incluyendo un nuevo nivel
externo en el modelo y hace referencia al tratamien-
[54]. Este nuevo nivel es el mas
to de aguas del proceso, de tal forma se que se minimice tanto el uso y gasto en
tratamiento de aguas de un proceso.

3.3.1.2.
Metodos matematica
basados en programacion

Conceptualmente, un problema de sntesis de procesos puede ser descrito de la


siguiente forma: a partir de un conjunto de entradas (generalmente materias primas)
y salidas (productos) con una serie de especificaciones definidas, desarrollar un dia-
grama de flujo de proceso con el fin de convertir las entradas en salidas deseadas
cumpliendo las especificaciones deseadas y optimizando un determinado objetivo o
objetivo, ya sea economica,
funcion medioambiental, etc.

matematica
La metodologa basada en programacion se centra en considerar
de ecuaciones matematicas,
explcitamente, a traves las relaciones existentes entre

los diferentes subsistemas y el balance economico del proceso. La sntesis de procesos
de una configuracion
implica la seleccion o topologa, as como sus parametros
de
dise
no. Es decir, uno tiene que determinar, por una parte, que unidades de proceso

deben integrar el diagrama de flujo y como deben estar interconectadas, y por otro
lado se tiene que determinar los tama
nos y las condiciones de funcionamiento de las

unidades. Desde un punto de vista matematico, la primera parte implica claramente

la toma de decisiones discretas, mientras que la segunda implica hacer una eleccion
en un espacio continuo. Por todo ello, el problema de la sntesis corresponde a un

53
3.3.Sntesis y Dise
no de procesos

discreto/continuo que matematicamente


problema de optimizacion queda definido
por un problema MINLP6 [55].

de esta metodologa consta de tres etapas [56].


La aplicacion

Paso 1 Desarrollo de una superestructura 7 que contenga todas las alternativas


existentes para las diversas partes del proceso, a partir de las cuales se
el dise
establecera
no optimo del proceso

Paso 2 La superestructura es modelada como un problema MINLP:

Z = min C(y, x)

s.t. h(y, x) = 0 (3.3.1)

g(y, x) 0

y [0, 1]m , x Rn

Siendo:

y: vector de variables binarias asociadas a la toma de decisiones en


el dise
no del proceso, y toma valor 1 cuando la unidad de proceso aso-
ciada a dicha variable es seleccionada y 0 cuando no es seleccionada

x : vector de variables continuas asociadas a las diferentes condicio-


del proceso (presiones, temperatura, flujos, etc. . . )
nes de operacion

objetivo del problema


C (y, x ): funcion

h(y, x ): ecuaciones de proceso

g(y, x ): especificaciones de dise


no


Aunque desde un punto de vista teorico no es estrictamente necesario,
es siempre posible modelar un problema de tal forma que las variables
binarias aparezcan siempre de forma lineal, lo cual desde un punto de

vista numerico del problema.
facilita mucho la solucion
6 Mixed Integer Nonlinear Programming Problem
Acronimo del ingles
7
Una superestructura se puede definir como un diagrama de flujo que incluye todas las alternativas existen-
de las diferentes tareas de que consta un proceso (alimentos, equipos, condiciones de
tes para la realizacion

operacion,...)
estando interconectadas entre s, de tal forma que incluya la estructura optima del sistema

54

Captulo 3. Introduccion

del correspondiente problema MINLP proporciona el dise


Paso 3 La resolucion no

optimo del proceso

Esta metodologa general fue ya propuesta en 1973 por Umeda et al. [57]. El pro-

blema descrito inicialmente contena unicamente variables continuas, siendo formu-
de tecnicas
lado como un NLP, y resuelto a traves
de busqueda directa. Posterior-
mente, Papoulias y Grossmann (1983) [58] formularon el problema como un MILP ma-
nejando variables binarias y resolviendo dicho problema computacionalmente utili-
y acotamiento. Fue Duran y Grossmann [59] quie-
zando algoritmos de ramificacion
nes formularon explcitamente el problema como un MINLP y posteriormente Kocis y
Grossmann [60] quienes aplicaron satisfactoriamente dicha metodologa a un caso
concreto.

de la metodologa descrita son los pasos 1


Los pasos cruciales en la aplicacion
de la superestructura) y 3 (resolucion
(generacion del problema MINLP). Aunque el
del problema MINLP) es una etapa extremadamente importante
paso 2 (formulacion
debido a que la forma en que se modela el problema tiene una gran importancia en
(numero
la resolucion
de ecuaciones, de variables, tiempo de calculo,
etc) de este
por parte del algoritmo utilizado.

Existen diferentes problemas y caractersticas importantes a tener en cuenta en


cada uno de los diferentes pasos de esta metodologa. En primer lugar, la construc-
de superestructuras en los procesos qumicos es una tarea altamente compleja, si
cion

bien el desarrollo de estas puede realizarse de forma sistematica para ciertos sistemas
peque
nos (por ejemplo, redes de intercambio de calor [61]), el desarrollo de superes-
tedioso y complicado. Dos enfoques
tructuras en procesos generales es mucho mas
principales se han desarrollado con el fin de mejorar y facilitar estas tarea, por un la-

do se encuentra el metodo
axiomatico propuesto por Friedler et al. [62], y por otro los

metodos estado-tarea y estado-equipos propuestos por Yeomans y Grossmann [63].

del problema es importante tener en cuenta que los


En cuanto a la formulacion
modelos de sntesis se pueden formular en tres niveles principales de detalle: (a) mo-
delos agregados, los cuales se concentran en ciertas caractersticas del sistema co-
mo puede ser los flujos de energa (por ejemplo, el modelo LP de transbordo para

55
3.3.Sntesis y Dise
no de procesos

redes de intercambio de calor propuesto por Papoulias y Grossmann [58], o los mo-
delos NLP para intercambiadores de materia y energa propuesto por Papalexandri y
de costes
Pistikopoulos [64]), (b) modelos aproximados, que implican la optimizacion
y de operacion)
(costes de inversion pero donde la modelizacion
de las unidades de
de modelos no lineales simples (por ejemplo, el modelo
procesos es realizada a traves
de diagramas de flujo de proceso propuesto por Kocis y
MINLP para la optimizacion
Grossmann [60]); (c) modelos rigurosos, que se basan en superestructuras detalladas
e implican modelos rigurosos y complejos para predecir el comportamiento de las

diferentes unidades de proceso (por ejemplo, la sntesis de secuencias de destilacion
propuesta por Smith y Pantelides [65], y los modelos GDP propuestos por Grossmann,
Aguirre y Barttfeld [66]).

de este tipo de problemas, debido a su enorme com-


Finalmente, la resolucion

plejidad (gran numero de ecuaciones y variables, uso de variables continuas y ente-
ras,...), se encuentra profundamente ligada al uso de ordenadores. Cabe destacar

que el continuo desarrollo de estos ha permitido incrementar considerablemente su

capacidad de calculo,
lograndose con ello resolver problemas extraordinariamente

grandes en tiempos insignificantes. Posteriormente, en el captulo 3.4 Programacion

matematica
en la sntesis de procesos se muestra un analisis en profundidad de los
de cada uno de los tipos de
diferentes algoritmos desarrollados para la resolucion
matematica.
problemas que se pueden presentar en programacion

3.3.1.3.
Metodos hbridos


Cada uno de los metodos de sntesis de procesos presentados -heursticos, des-
jerarquizada, programacion
composicion matematica-
poseen ciertas ventajas y des-

ventajas comparados entre s [67, 68]. Sin embargo, estos metodos pueden ser con-
siderados como competitivos y complementarios, ya que se refieren a diferentes as-
pectos del dise
no. Por lo tanto, un camino seguido en la sntesis de procesos es la
de varios de estos metodos
combinacion con el fin de aprovechar las ventajas que
un modelo que
cada uno de ellos presenta. Por ejemplo, Daichendt [69] presento
matematica
combinaba la programacion jerarquizada para
con la descomposicion
dise
nar grandes procesos qumicos. Gani et al. [70, 71] propusieron una metodologa

56

Captulo 3. Introduccion

matematica
hbrida que combinaba la programacion
con aspectos termodinamicos
para el dise multicolumna. Generalmente, la metodo-
no de sistemas de destilacion

loga hbrida conlleva a reducir el coste computacional necesario para la resolucion
de los distintos problemas.


Numerosas modelos hbridos se han desarrollado y aplicado con exito a un gran

numero de procesos qumicos como redes intercambiadoras de calor, secuencias de
redes complejas de reactores o redes de agua [72-75].
separacion,

3.3.2. Principales Avances en la Sntesis de Procesos Qumicos

En la sntesis de procesos qumicos, un ingeniero debe desarrollar, identificar o de-


que propor-
terminar cual es el diagrama de flujo y las condiciones de operacion
optima.
cionan la solucion
Dicho proceso implica el manejo de un numero enorme-
mente grande de alternativas para un proceso qumico determinado. Con el fin de
facilitar esta tarea, numerosos avances en el campo de la sntesis de procesos se han

desarrollado a lo largo de las ultimas
decadas. se muestran los prin-
A continuacion

cipales metodos
desarrollados en cada una de las principales areas de la ingeniera
qumica.

Redes de intercambiadores de calor

estudiado es el dise
Dentro de la sntesis de procesos, el problema mas no de redes
de intercambiadores de calor, para lo cual se han desarrollado numerosas metodo-

logas y codigos de software. Los procesos de dise
no de redes de intercambiadores

de calor se basa en la busqueda optima
de la configuracion que permita la recu-
y reutilizacion
peracion de la maxima

cantidad de energa, minimizando simultanea-
mente el coste anual derivado del consumo de servicios calientes y fros. La principal
metodologa desarrollada y que establece las bases de la sntesis de redes de in-

tercambiadores es el analisis de la aplicacion
del punto de pinch. Ademas a redes
de calor, dicha metodologa puede ser utilizada tanto en redes de intercambio de

materia [76] como en las redes de agua [77]. Diferentes metodos que combinan la
matematica
programacion
y el analisis de punto de pinch han sido desarrollados pa-
los servicios
ra la sntesis de redes de intercambiadores de calor incluyendo ademas

57
3.3.Sntesis y Dise
no de procesos

calientes o fros suministrados al sistema. Por ejemplo, Duran y Grossmann [59] propu-
sieron un modelo MINLP combinado con la tecnologa del punto pinch, mediante la

cual es posible calcular la red optima
que minimiza el coste en servicios energeticos
como punto de partida para numerosos trabajos reali-
externos. Dicho trabajo sirvio
zados tanto en este campo como en redes de suministro de agua. Recientemente,
global unido
Ahmetovic y Grossmann [78] han propuesto un modelo de optimizacion
a una superestructura general para la sntesis de redes de agua, el cual puede incluir

multiples fuentes de agua, diferentes procesos que utilizan agua, operaciones de pre-
tratamiento y diferentes tratamientos de agua residuales. Dicha metodologa ha sido
aplicada satisfactoriamente en diferentes procesos industriales [79, 80].


Secuencia de separacion


La sntesis optima es un campo que ha recibido
de las secuencias de separacion
debido a que,
especial atencion, gran parte del coste de los procesos qumicos se
de compuestos. Por lo que facilmente
deriva de las etapas de separacion se llega a
de que, la optimizacion
la conclusion de la etapa de separacion
implica importantes

ahorros energeticos
y economicos en la totalidad del proceso. Casi la totalidad de
las metodologas explicadas en el apartado 3.3.1 han sido aplicadas a la sntesis de
En la bibliografa pueden encontrarse numerosas revisio-
secuencias de destilacion.
nes que explican en detalle cada una de ellas [44, 81]. De todas las metodologas
matematica
existentes, la programacion utilizada. Actualmen-
es la herramienta mas

te, un gran numero de excelentes contribuciones han sido realizadas entorno a la
reactiva y los procesos de separacion
destilacion hbridos, debidos a la baja inversion

necesaria y los altas cantidades de reactivos implicadas. Ciric fue el primero en desa-
rrollar un modelo MINLP para el dise
no optimo reactiva
de columnas de destilacion
[82]. Recientemente, Ropotar et al. [83] han propuesto un modelo hbrido para la ob-
de la configuracion
tencion y las condiciones de operacion
optimas
para un sistema
cataltica. Franke [84] ha propuesto un proceso de sntesis en tres pasos
de destilacion
destilacion-cristalizaci
para procesos hbridos de separacion,
on.

la sntesis de secuencias de columnas complejas de destilacion


Ademas, para la
de mezclas multicomponente ha ganado especial atencion
separacion con el paso

del tiempo [85-87], debido a los importantes ahorros energeticos logrados en compa-

58

Captulo 3. Introduccion

con el uso de columnas de destilacion


racion convencionales. Un campo importante
dentro de las secuencias de columnas complejas es el que estudia las columnas de
acopladas termicamente
destilacion (TCD) 8 . Agrawal [88] fue quien
establecio
las

bases teoricas
de los sistemas con acoplamiento termico total. Desde el punto de

vista estructural, los sistemas que incluyen columnas con acoplamientos termicos son
complejos que los sistemas de columnas convencionales. Si se conside-
mucho mas
ran todas las posibles alternativas existentes entre las columnas convencionales y las

columnas con acoplamiento termico total, el problemas llega a ser particularmente
complejo. Diferentes metodologas han sido propuestas con el objetivo de sintetizar

todas estas alternativas. Caballero et al. [89-93] han propuesto un metodo para la
sistematica
generacion matematica
de todas las alternativas usando programacion

y relaciones logicas, mientras que Agrawal [94, 95] y Rong et al. [96-98] propusieron

un metodo conceptual alternativa.
con el mismo fin utilizando enumeracion

Redes de reactores

El desarrollo de metodologas para el dise


no optimo de redes de reactores ha si-
do menos estudiado que la sntesis de redes de intercambio de calor o sistemas de
Generalmente los sistemas de reaccion
separacion. son descritos por ecuaciones al-
gebraicas diferenciales, las cuales suelen ser fuertemente no lineales y no convexas.

Tanto los metodos matematica
heursticos como de programacion han sido aplica-
exitosa es quizas
dos a la sntesis de redes de reactores. La metodologa mas el meto-

matematica
do hbrido que combina la programacion
y el concepto geometrico de

zona alcanzable (La zona alcanzable, desde un punto de vista practico, se puede
interpretar como el total de posibles condiciones que pueden presentarse a la salida
cualesquiera que sea su configuracion)
del sistema de reaccion [99]. Por otro lado, un
importante tema que debe ser considerado a la hora de dise
nar redes de reactores

es la operatividad. As pues, Kokossis [100] considero un analisis de estabilidad en la

sntesis de redes de reactores con el fin de asegurar la operatividad del sistema opti-
acoplado
mo final. Zhao y Marquardt [101] presentaron un proceso de optimizacion

con el analisis
dinamico no lineal para la sntesis de redes de reactores, consiguiendo
con ello garantizar de forma robusta la estabilidad y operatividad del sistema.
8 Thermally Coupled Distillation
Acronimo del ingles

59
3.3.Sntesis y Dise
no de procesos

Diagrama de flujo del proceso global

En el caso de un proceso global, que incluya todas y cada una de sus etapas,
del diagrama de flujo optimo
el objetivo de la sntesis de procesos es la obtencion
optimas.
del proceso total y sus condiciones de operacion En este sentido, tanto la
heurstica jerarquizada como la programacion
descomposicion matematica
han si-

do probadas como utiles
y necesarias. Durante la ultima
decada,
la programacion

matematica, MINLP y GDP, han sido ampliamente utili-
especialmente la formulacion
zadas en el dise
no de procesos globales. Kravanja [67] propuso una estrategia jerar-
quizada multietapa acoplado a un modelo MINLP con el fin de sintetizar la configu-
optima
racion de un proceso total. Diferentes representaciones han sido propuestas
con el fin optimizar un proceso global, como pueden ser la estructura de modela-
estado-equipo [65], redes estado-tarea y
do general [64, 102-105], la representacion
redes estado-equipo [63].

3.3.3. Prospectivas de la Sntesis de Procesos

La industria de procesos es aquella que se encarga del dise


no de los procesos
y sntesis de materiales. Los primeros pasos de esta industria fueron
de fabricacion

dirigidos a las necesidades basicas
de la epoca, la agricultura y la metalurgia anti-
de las necesidades de la sociedad derivo
gua. La evolucion en la evolucion
cons-
tante de la industria de procesos, abarcando campos tales como los materiales
(ladrillo, cemento,. . . ), tintes, pigmentos, jabones, licores, combus-
de construccion

tibles,. . . Originalmente, los procesos desarrollados unicamente implicaban el aisla-
de sustancias naturales, aunque con el paso de tiempo estas
miento o separacion

tecnicas com-
fueron evolucionando hacia transformaciones fsicas y qumicas mas
plejas con el fin de obtener nuevos materiales.

de la industria de procesos puede ser explicado con numerosos ejem-


La evolucion
plos, un ejemplo claro es la industria de los polmeros.

El campo de la industria de polmeros se ha desarrollado principalmente durante



los ultimos 75 a
nos. Los primeros polmeros sintetizados industrialmente procedan o
o el acetato de celulosa ambos
derivaban de compuestos naturales, como el rayon

60

Captulo 3. Introduccion

sintetizados a partir de la celulosa. Posteriormente, nuevas rutas de sntesis de polme-



ros fueron desarrolladas utilizando diferentes monomeros como materias primas. Estas
nuevas rutas dieron lugar al descubrimiento de nuevos compuestos como poliamidas

(nylons), poliesteres, policarbonatos, poliuretanos, poliestirenos o poliolefinas. Las ma-
terias primas empleadas en estas nuevas rutas eran y son obtenidas principalmente

de productos derivados del petroleo y el gas natural, como etileno, propileno, CO, H2 ,

compuestos aromaticos,. . . Una caracterstica de estos nuevos materiales es el gran

numero de aplicaciones que presentan. El descubrimiento y sntesis de estos com-

puestos trajo consigo una epoca dorada del desarrollo industrial, numerosos avances

fueron realizados en un gran numero
de industrias, como la farmaceutica, agrcola,
de los avances industriales, la introduccion
alimentaria, etc. . . Ademas de estos nue-
de los materiales de fabricacion
vos materiales en la vida cotidiana, en sustitucion

de numerosos equipos y aparatos tradicionalmente construidos en base metalica,
y facilito el acceso de la sociedad a numerosas tecnologas.
abarato

Aproximadamente hace 25 a
nos, la velocidad de crecimiento del mercado de
a ralentizarse como consecuencia de que la capacidad de
los polmeros empezo
de metales por polmeros estaba ya muy extendida. Ademas
sustitucion dos acon-
tecimientos importantes tuvieron lugar. Por un lado, el aumento constante del precio

del petroleo en que el coste de la energa y las materias primas fueran cada
derivo
medioambiental por parte
vez mayores. Y por otro lado, la aumento de la regulacion
en un aumento de la inversion
de los gobiernos derivo en la tecnologa con el obje-
tivo de cumplir la nueva normativa. Todos estos acontecimientos produjeron que los

margenes de beneficios obtenidos por estas industrias fueran cada vez menores y co-
mo consecuencia numerosas industrias apostaron por nuevos dise
nos con el objeto
de reducir los costes.

todo ello unido a la saturacion


Ademas, de los mercados locales condujo a la
industria a dirigir sus esfuerzos fuera de los lmites donde normalmente trabajaban,

focalizando su crecimiento en la globalizacion. Como consecuencia, las amplia-
ciones realizadas en las principales industrias fueron destinadas principalmente a la
de productos. Los polticos, ante esta nueva tendencia mundial, facili-
exportacion
taron las exportaciones reduciendo las barreras al comercio internacional. Y en los

casos donde el comercio transoceanico resultaba poco competitivo, numerosas em-

61
3.3.Sntesis y Dise
no de procesos

e implantacion
presas decidieron ampliar sus mercados mediante la construccion de

nuevas plantas a lo largo y ancho de todo el globo terraqueo. El resultado de todo

ello es que mientras los mercados han crecido, la competencia es ahora mayor, mas
dura y a escala global.

por tanto se ha convertido en todo un reto para la industria en


La globalizacion
muchos mas
general y la industria qumica en particular. En un futuro previsible, habra
jugadores en el sector de la industria de procesos qumicos, aunque el mercado pro-
creciendo, debido al aumento de la calidad de vida tanto
bablemente continuara
en los pases desarrollados como principalmente en los pases en vas de desarrollo.
sobre la industria con una estructura
Generalmente, la ventaja competitiva recaera

de costes superior (mayor inversion). Otro aspecto competitivo especialmente impor-
la capacidad para desarrollar y poner en practica
tante sera
rapidamente nueva
tecnologa capaz de adaptarse a las necesidades del mercado.

enfrentarse la industria qumica seran


Sin embargo, los desafos a los que debera
almacenamiento y uso),
principalmente, las materias primas, la energa (produccion,
de residuos y de gases de efecto invernadero o la velocidad de desarrollo
la gestion
de nuevas tecnologas o procesos para satisfacer las cada vez mayores demandas
del mercado mundial.

de la sntesis de pro-
La pregunta en este punto es, cual debe ser la evolucion
cesos para hacer frente satisfactoriamente a todos estos desafos? Algunas de las

multiples evoluciones en los diferentes campos de que se compone la sntesis de pro-
cesos fueron propuestas por Barnicki y Siirola [41].

Replanteamiento del uso de materias primas y procesos tradicionales

con
La gran competencia existente en el campo de la industria qumica obligara
toda seguridad al desarrollo de nuevos procesos de sntesis para poder competir en

el mercado global. Hoy en da, gran numero de procesos qumicos utilizan como ma-

terias primas compuestos derivados del petroleo y el gas natural. El empleo de estas
relegado a un segundo plano, debido principal-
materias primas en el futuro se vera

mente a su menor competitividad economica ya sea por su progresivo agotamiento

62

Captulo 3. Introduccion


o la baja disponibilidad geografica en muchas regiones mundiales. El descenso en
la obligatoria necesidad de sustituirlas por
el uso de estas materias primas implicara
(recurso fosil
otras. Los principales candidatos pueden ser el uso del carbon mas abun-

dante y ampliamente distribuido geograficamente), el empleo de materia vegetal
(especialmente cultivada con este fin) o el uso de diferentes productos intermedia-

rios que actualmente no son competitivos pero ante el nuevo panorama economico
pueden transformarse en importante puntos de partidas de numerosos procesos.

que probable de la disponibilidad de materias primas provocara


El cambio mas
que numerosos procesos deban ser redise
nados, con el fin de adaptarse satisfactoria-
Algunos de estos cambios ya son conocidos, aunque
mente a esta nueva situacion.
deben ser analizados, investigados y desarrollados. Uno
la gran mayora de ellos aun
rese
de los aspectos mas protagonismo adquirira,
nables, y que mas sera
el continuo
aumento del uso de catalizadores y biocatalizadores, adaptados a cada proceso en
particular.

y desarrollo de nuevos
El objetivo fundamental de la sntesis de procesos (invencion
procesos adaptados a la escala deseada, que se desarrollen de manera segura, me-

dioambientalmente responsable, y siendo a su vez eficiente y economicos respecto
invariable. Aunque es muy proba-
al resto de procesos competidores) se mantendra
ble que diversos factores del dise
no se vean alterados radicalmente en el futuro. La
fuente y coste de la energa, los efectos sobre el clima, los materiales de construc-
y la cartera de operaciones unitarias puede cambiar drasticamente
cion en el futuro.
Como resultado, muchas de las heursticas empleadas actualmente en el dise
no de
procesos pueden no ser aplicables en el futuro. Por lo que, a la luz de los cambios
necesario volver a evaluar continuamente las heursticas de dise
producidos, sera no
antes de ser utilizadas.

sistematica
Generacion de alternativas

sistematica
Numerosos aspectos deben ser considerados en la generacion de al-
ternativas en el dise
no de un proceso qumico. Actualmente, los metodos de sntesis
matematica
de procesos basados en la experiencia y la programacion pura actual-
utilizados, y en el futuro se cree que este tipo de metodos
mente son los mas
seguiran

63
3.3.Sntesis y Dise
no de procesos


siendo los metodos eficaces y ampliamente utilizados para la generacion
mas de
alternativas de procesos.

Una alternativa interesante a desarrollar en el futuro es el uso de la inteligencia


artificial (IA) para el desarrollo de alternativas. Dentro de la inteligencia artificial, nu-
merosos avances se han realizado en numerosos campos como pueden ser los video-
automatica
juegos, la demostracion en robots y otros
de teoremas, la planificacion

dispositivos automaticos, de es-
etc. . . . El principal problema de la implementacion
ta tecnologa en el estudio de los procesos qumicos radica en el hecho de que la
mayora de los procesos qumicos presentan recirculaciones y recuperaciones de co-
de
rrientes, tanto de materia como de energa, lo que implica la retroalimentacion
Sin embargo, este tipo de proceso en general no han sido suficiente-
informacion.
mente tenidos en cuenta en IA, dado que no son procesos comunes en los campos
de esta.
tpicos de aplicacion En el futuro, este campo puede llegar a convertirse en

una importante area en el campo de la IA.
de investigacion

El dise
no de los procesos qumicos es una tarea muy complicada. Todas las pro-
puestas realizadas en el pasado han enfocado el problema de dise
no en base a la
jerarquizada en una serie de subproblemas, tales como subproble-
descomposicion
de separacion
mas de reaccion, y purificacion,
subproblemas energeticos,
etc. . . Tal
formulados actualmente estos subproblemas no son totalmente inde-
como estan

pendientes, sino que interactuan de maneras a menudo complejas. El rendimiento
y especialmente la optimalidad del proceso de dise
no y sntesis de procesos se ve
claramente influenciado por esta dependencia.

cambiando a lo largo del


La ingeniera de procesos qumicos ha ido y continuara
tiempo. Ocasionalmente, nuevas operaciones unitarias han sido desarrolladas y nue-

vos fenomenos son explotados para realizar y mejorar procesos tanto nuevos como ya
y otros metodos
existentes. Hoy en da la destilacion basados en gran
de separacion
y cristalizacion,
medida en el equilibrio de fases, como la extraccion han alcanzado
como en aplicacion.
altos niveles de madurez, tanto en comprension Estas opera-
siendo pilares de la industria en
ciones unitarias han sido y probablemente seguiran

el futuro previsible. Sin embargo, es probable que muchos metodos
de separacion,
basados en procesos todava relativamente inmaduros, como por ejemplo la per-

64

Captulo 3. Introduccion


meabilidad en membranas, la cromatografa a gran escala, el intercambio ionico o
en la industria en los
la electroforesis, adquieran una mayor capacidad de aplicacion

proximos a
nos.

Por otro lado, la tecnologa tiende a intensificar los procesos [106, 107], lo que lleva
dise
a pensar casi con total seguridad, que en un futuro los equipos seran nados para

actuar con mayores fuerzas impulsoras de los procesos, mayores areas por unidad de
implicara
volumen, etc. . . Dicho grado de intensificacion nuevas operaciones unita-
rias combinando varias de ellas, por ejemplo reacciones y separaciones. Un ejemplo
o integracion
claro de este proceso de intensificacion es la destilacion
reactiva [108],
la cual ya ha comenzado a ser aplicada ampliamente. Existen otros conceptos, co-
empezando a ser explotados. En el futuro, la
mo reactores de membrana, que estan
sistematica
generacion de procesos y las heursticas de dise ser actuali-
no deberan
zadas con el fin de introducir todas estas nuevas tecnologas, operaciones unitarias,
etc. . .


Desde el punto de vista computacional, existen diversos metodos de sntesis de
procesos que se han automatizado, mientras que muchos otros aun no lo han sido.
de ciertos procesos son adecua-
Esto se debe a que el manejo y representacion
dos para el uso del ser humano, mientras que su manejo computacional no han si-
do implementados de forma adecuada. Un ejemplo claro son las representaciones

graficas. Mientras que el ser humano es capaz de manejarlas exitosamente, los or-
lejos de manejar dichas habilidades. Esto unido a la
denadores se encuentran aun
del estudio de procesos ha obligado a que estos metodos
evolucion pasen de su for-

ma grafica a ser codificados para su estudio en profundidad. Este hecho se refleja
en numerosos casos de la historia de la ingeniera qumica. Un ejemplo claro de es-
de la destilacion.
te comportamiento se representa en la evolucion Inicialmente, la
se basaba casi por completo en metodos
sntesis de procesos de destilacion
graficos,
dimensiones limitadas, como son los diagramas de McCabe-Thiele. Con el desarrollo

computacional, dichos metodos
evolucionaron a modelos numericos basado en el
equilibrio entre fases, pudiendo estos ser codificados para su uso computacional y
mucho mas
consiguiendo estudiar sistemas de destilacion complejos. Una evolucion

similar esta teniendo lugar con los mapas de curvas de residuo. Sistemas de hasta tres

y cuatro componentes pueden ser representados en el espacio grafico 3D y maneja-

65
3.3.Sntesis y Dise
no de procesos

complejos con mayor


dos con facilidad por el ser humano, en cambio sistemas mas

numero derivando en el desa-
de componentes son difcilmente analizables. Ello esta

rrollo de nuevas tecnicas y computacion.
de automatizacion

de superestructuras
Optimizacion

de superestructuras ha sido considerada, desde hace mucho tiem-


La optimizacion
po, como una materia clave dentro de la sntesis de procesos. Cualquier proceso
qumico puede ser representado mediante una superestructura, la cual muestra, esta-
blece y resume todas y cada una de las alternativas posibles (equipos y condiciones)
de este proble-
en cada una de las etapas de un proceso establecido. La resolucion
simultanea
ma implica la optimizacion
tanto de la estructura del proceso (numero y

tipo de equipos) como las de las condiciones de operacion.

Este tipo de metodologa presenta dos importantes problemas. En primer lugar, la


optima
configuracion se encuentra
solamente puede ser obtenida s dicha opcion
es como generar su-
considerada en la superestructura inicial. Por lo que la cuestion
de la solucion
perestructuras que garanticen la inclusion optima.
En segundo lugar,
de esta metodologa genera problemas matematicos
la implementacion muy com-
plejos y de gran tama radica en como resolver este tipo de
no, por lo que la cuestion
problemas.

de la superestructura se llevara
En un futuro, la generacion a cabo de forma sis-

tematica en cada una de las etapas del proceso, para posteriormente combinando
probables generar la superestructura
todas las alternativas generadas o solo las mas
especialmente practica
completa del proceso. Esta forma de operar sera
y facilmen-
de superestructuras asistidas por ordenador.
te aplicable en la generacion

Los primeros trabajos realizados en la sntesis de procesos mostraron que imple-


de dicha metodologa implicaba la resolucion
mentacion de problemas matemati-

cos extremadamente complejos. Generalmente, los problemas obtenidos son pro-

blemas MINLP, y en la practica suelen ser problemas no convexos, con discontinuida-
matematica
des,. . . problemas muy difciles de resolver. La representacion apro-
mas
la mas
piada y que sera utilizada en el futuro para este tipo de problemas es la Pro-
Disyuntiva Generalizada (GDP) [109]. La principal ventaja respecto al resto
gramacion

66

Captulo 3. Introduccion


radica en el uso de la logica, de problemas de sntesis de
lo que facilita la resolucion

procesos. Este avance, junto con la mejora constante de la capacidad de calculo
de superestructuras sea
computacional, hace que la metodologa de optimizacion
potencialmente practica
una opcion en la sntesis de procesos.

de los criterios de optimizacion


Redefinicion

El dise
no de procesos ha sido siempre evaluada desde diferentes puntos de vista,

incluyendo aspectos economicos,
impacto medioambiental, consumo energetico,
control, flexibilidad, etc. . . Sin embargo, en el futuro es muy probable que los factores
importantes que estos. Factores que incluyen
sociales puedan llegar a ser incluso mas

sostenibilidad, ciclo de vida, cambio climatico,
uso de mano de obra o minimizacion
de riesgos y seguridad. La principal dificultad en el estudio de estos factores es su
ya que muchos de ellos son conceptos abstractos. Uno de los princi-
cuantificacion,
pales retos es el desarrollo de metodologas para cuantificar todos los aspectos claves
en la sociedad.

de problemas con mas


La tendencia actual de la sociedad implica la resolucion

de un objetivo (por ejemplo se buscan procesos economicos y respetuosos medioam-
de uso comun
bientalmente), por lo que en un futuro seran los problemas multiobjeti-
de diagramas de Pareto.
vo y la evaluacion

de diferentes campos dentro de la sntesis de procesos


Integracion

La puesta en marcha de un nuevo proceso qumico es un tarea complicada que


de muchos pasos. En el futuro, en el dise
precisa de la realizacion no y sntesis de
acentuar el proceso de integracion
procesos se debera e interactuacion
de las dife-
rentes partes que conforman dicho proceso.


En la practica, los diagramas de flujos de procesos son evaluados de forma se-
cuencial. Generalmente, un proceso qumico es desarrollado e inventado y posterior-
mente dicho dise
no es analizado y evaluado respecto a diferentes factores. Dichos

factores incluyen: 1) comportamiento dinamico, 2) flexibilidad (intervalos de condi-
y especificaciones donde el proceso es aplicable), 3) robustez
ciones de operacion

67
3.3.Sntesis y Dise
no de procesos

(tolerancia a los cambios en la calidad de materias primas y otras alteraciones en


el proceso) y 4) control (capacidad de mantener el proceso en un estado estable y
producir productos en las especificaciones deseadas a pesar de que tengan lugar
perturbaciones en el sistema). En general, los diagramas de control son dise
nados
posteriormente al dise de las
no del proceso. La tendencia futura es la integracion
etapas de flexibilidad y control dentro del proceso de dise
no y no como una etapa
posterior a ella.

Por otro lado, actualmente dentro de la metodologa de desarrollo de cualquier


del
proceso qumico, el primer paso que tiene lugar es el estudio y determinacion
y sus condiciones: catalizadores requeri-
proceso qumico (secuencia de reaccion
dos, temperaturas, concentraciones, purezas, o las necesidades de disolvente). Una
vez dicho proceso qumico ha sido estudiado en profundidad se procede al dise
no

de este como proceso industrial y no es hasta esta etapa
para su implementacion
cuando entran en juego la ingeniera de procesos, En la mayora de ocasiones, im-

portantes parametros son fijados en la etapa qumica sin tener en cuenta el posterior
dise
no, con lo que el resultado final suele ser un dise
no suboptimo. Esto representa
un claro problema en un mundo donde cada da la competencia en el mercado
mas
global esta acentuada, y cualquier mejora de los procesos puede establecer la
diferencia entre que un proceso triunfe o fracase.

una integracion
Por todo ello, una tendencia clara de futuro sera cada vez ma-
yor entre la qumica y la ingeniera de los procesos qumicos. Las consideraciones

tecnicas un mayor impacto en decisiones tomadas en etapas tempranas
tendran
la retroalimentacion
de establecimiento de la qumica del proceso. Aumentara en-
se llevara
tre ambas disciplinas. Esta retroalimentacion a cabo por los avances en

la qumica computacional, combinatorial, la mejora de los metodos
de prediccion
de propiedades fsicas, as como mejoras en el modelado de process qumicos. En un
futuro no muy lejano se espera que experimentos virtuales de procesos qumicos com-
binados sntesis y modelado eficaz de procesos, puedan llegar a reducir temporal y

economicamente de un nuevo proceso qumico.
el coste de implementacion

68

Captulo 3. Introduccion

3.4 matematica
Programacion en la sntesis de procesos


Desde un punto de vista practico, puede describirse co-
la tarea de optimizacion
a este
mo: dado un sistema o proceso, determinar la mejor solucion que satisfaga
de un proceso qumico es un
una serie de restricciones asociadas. La optimizacion
proceso especialmente complejo y nada trivial, el cual depende de las caractersti-
o proceso qumico puede ser descrito a
cas presentes en el proceso. Toda operacion
de un modelo matematico.
traves

Para el dise de cualquier sistema, el primer paso a realizar es la


no y operacion
de un modelo predictivo que describa el comportamiento del sistema.
obtencion
El cual traslade a un conjunto de ecuaciones e inecuaciones tanto las relaciones
analticas que describen al sistema como las restricciones asociadas. Conjuntamente
objetivo debe ser especificada de tal forma que proporcione
al modelo, una funcion
una medida cuantitativa de la propiedad a optimizar.


Los modelos matematicos desarrollados pueden incluir variables continuas y/o bi-
del tipo de
narias y estar descritos por ecuaciones lineales y/o no lineales. En funcion

ecuaciones y variables se pueden establecer diferentes formulaciones matematicas,
descritas brevemente a continuacion.
las cuales seran

3.4.1. Lineal
Programacion

(minimizar o maximizar)
Un problema lineal puede definirse como la optimizacion
lineal, denominada funcion
de una funcion objetivo, de tal forma que las variables
esten
de dicha funcion sujetas a una serie de restricciones asociadas a un sistema de
ecuaciones e/o inecuaciones lineales [110].

69
matematica
3.4.Programacion

Un Problema Lineal (LP) 9 puede expresarse en la siguiente forma:

min (o max) C T x

s.a. Ax = b (3.4.1)

x0


donde x es el vector de las variables a resolver (por ejemplo parametros de flujos,
temperatura), A es una matriz de coeficientes conocidos, y c y b son vectores
presion,
C T x es denominada funcion
de coeficientes conocidos. La expresion objetivo, y las
factible.
restricciones Ax = bforman la region

lineal se pueden expresar de forma


Aunque todos los problemas de programacion

estandar,
en la practica
esto es innecesario. Por ejemplo, aunque esta representacion
exige a todas las variables tomar valores no negativos, la mayora de los problemas

de LP permiten lmites generales l x u donde l y u son los vectores conocidos de
los lmites inferiores y superiores, pudiendo expresarse estos como y/o . De esta
forma una variable puede ser totalmente libre de forma explcita, aunque implcita-
mente las restricciones en la matriz A imponen lmites a las variables.

de este tipo de problemas se realiza a traves


La resolucion de dos tecnicas
prin-

cipalmente, el metodo
simplex y los metodos
de punto interior. El metodo simplex,
introducido por Dantzig (1957)[110-112] hace aproximadamente 50 a
nos, genera so-

luciones basicas computadas fijando muchas de las variables a sus lmites extremos
con el objetivo de reducir las restricciones Ax = b a un sistema cuadrado que se pue-

de resolver para los valores unicos
de las variables restantes. Las soluciones basicas
factible definida por Ax = b y x 0. En
representan el lmite extremo de la region

el metodo simplex las soluciones alternativas se encuentran siempre en un punto ex-
factible, en comparacion
tremo de la region con los metodos
de punto interior, que
factible. Este ultimo
visita los puntos dentro del interior de la region
metodo se deriva

de las tecnicas no lineal que fue desarrollado y popularizado
para la programacion
9 Linear Programming
Acronimo del ingles

70

Captulo 3. Introduccion

en los a a la programa-
nos sesenta por Fiacco y McCormick, aunque su aplicacion
lineal empezo
cion con el analisis
innovador de Karmarkar [113, 114].

lineal ha demostrado ser valiosa para el modelado y la solucion


La programacion
(o planning), programacion
de diversos problemas de planificacion (o schedulling),
y dise
asignacion, no. Las principales industrias que hacen uso de la LP y sus extensio-
nes, aparte de las de qumicas, incluyen las de transporte, energa, telecomunicacio-
nes, etc. . . .

3.4.2. No Lineal
Programacion

no lineal (NLP) 10 tiene como


lineal, la programacion
Al igual que la programacion
(maximizar o minimizar) de una funcion
objetivo la optimizacion objetivo sujeta a res-
La diferencia entre ambos radica en el hecho
tricciones en las variables de decision.
objetivo y/o las restricciones asociadas son ecuaciones no lineales.
de que la funcion


A diferencia de lo que ocurre con los LP en los cuales el optimo siempre corres-

ponde a un vertice factible, lo cual implica un numero
de la region finito de puntos

de busqueda. optima
La solucion en los NLP puede ser localizada en cualquiera de
factible, lo que aumenta la dificultad de resolucion
los infinitos puntos de la region de
los problemas NLP.

No lineal (NLP) puede expresarse de forma general


Un problema de Programacion
como [115]:

min (o max) f (x)

s.a. h(x) = 0
(3.4.2)
g(x) 0

x X Rn

no lineal ha sido y es un campo extensamente estu-


El campo de la programacion

diado, donde muchos investigadores han aportado un gran numero de metodologas
10 Non Linear Programming
Acronimo del ingles

71
matematica
3.4.Programacion

de problemas. Ello es debido a que en el mundo real, la gran mayora


de resolucion
de problemas a resolver son descritos por funciones no lineales.

Los problemas no lineales pueden ser convexos o no convexos (Figura 3.4.1) de-
objetivo. Un problema
bido al estado funcional de las restricciones y/o la funcion
convexa, y el
convexo es un caso donde todas las restricciones forman una region
convexa en el caso de minimizar, o una funcion
objetivo es una funcion concava
en
el caso de maximizar. Las funciones lineales son convexas, y por tanto los problemas
lineal son problemas convexos. En estos problemas, la region
de la programacion fac-
de las restricciones convexas) es una region
tible (interseccion convexa. Por extension,


un problema no convexo es cualquier problema donde el objetivo o cualquiera de
las restricciones es no convexa.

(a) (b)

Convexa y No Convexa
Figura 3.4.1: Ejemplo de Region

Esta caracterstica de los problemas no lineales es de suma importancia, ya que

unicamente objetivo y region


los problemas convexos (funcion factible convexas)
obtenida sea un optimo
aseguran que la solucion global. Esto se produce como con-
factible convexa unicamente
secuencia de que dentro de una region puede existir
y en conclusion
una solucion, esta
con la solucion
coincidira optima
global del pro-

blema. En cambio, en los problemas no convexos pueden existir multiples regiones

factibles y multiples
puntos localmente optimos con lo que
dentro de cada region,
optima
no es posible asegurar la solucion global del problema. Por lo tanto, un pro-

blema no lineal con un gran numero de variables y restricciones puede requerir un
es el optimo
enorme cantidad de tiempo determinar que una solucion global para

todas las regiones factibles o simplemente que este es no factible. Esto hace que tales
problemas sean generalmente muy difciles de resolver.

72

Captulo 3. Introduccion


Los metodos de problemas no lineales se pueden clasificar en dos
de resolucion

grandes grupos, aquellos basados en derivadas y los que no utilizan derivadas (meto-
dos metaheursticos).


Metodos metaheursticos


Los metodos que no requieren de informacion
de optimizacion de las derivadas

son denominados como metodos
de busqueda
directa o metaheurstica. Estos meto-

dos presentan como principales ventajas, la facil y el bajo nivel ne-
implementacion
a resolver. En particular, estos
cesario de conocimiento del problema de optimizacion

metodos previos en nuevas aplica-
son adecuados para estudios de optimizacion
ciones (evitando de invertir en un primer de desarrollo un gran esfuerzo en construir
sofisticado y estrategias de solucion
modelos mas mas
complejas) y en aquellos pro-
de las
blemas donde sea extremadamente costoso o difcil conseguir informacion
derivadas.


Los primeros metodos
de busqueda
directa fueron desarrollados en las decadas

de los 60 y 70, incluyendo los metodos
one-at-a-time o los metodos basados en
dise
nos experimentales. Dentro de estos metodos
se encuentra el metodo
de patron

de busqueda
de Hooke y Jeeves [116], el metodo de Nelder y Mead [117], o los

metodos
de busqueda aleatoria adaptativa de Luus y Jaakola [118]. Posteriormente

a estos, diferentes metodos no basados en derivadas han sido propuestos, dentro

de ellos podemos encontrar los algoritmos geneticos
o el metodo de enfriamiento
simulado, entre otros.


Los algoritmos geneticos fueron propuestos inicialmente por Holland [119], y son lla-
biologica
mados as porque se inspiran en la evolucion
y su base genetico-molecular.
de individuos sometiendola
Estos algoritmos hacen evolucionar una poblacion a ac-

ciones aleatorias semejantes a las que actuan biologica
en la evolucion (mutaciones

y recombinaciones geneticas), a una seleccion
as como tambien de acuerdo con
del cual se decide cuales
criterio, en funcion
algun aptos, se-
son los individuos mas

leccionados, y cuales los menos aptos, que son descartados.


El metodo
de enfriamiento simulado [120] es un algoritmo de busqueda meta-
global. Dicho metodo
heurstica para problemas de optimizacion se inspira en el pro-

ceso de recocido del acero y ceramicas,
una tecnica que consiste en calentar y

73
matematica
3.4.Programacion


luego enfriar lentamente el material para variar sus propiedades fsicas. El metodo es
del algoritmo Metropolis-Hastings, un metodo
una adaptacion de Montecarlo utiliza-

do para generar muestras de estados de un sistema termodinamico.


Metodos basados en derivadas


Estos metodos se diferencian de los del grupo anterior porque utilizan la informa-
de las derivadas de las funciones del proceso. Estos metodos
cion generalmente con-

vergen en un menor numero de iteraciones, requieren que las funciones sean diferen-
el calculo
ciables, y ademas de la derivadas pueda llevarse a cabo ya sea en forma

analtica o numerica.


Dentro de los algoritmos para resolver problemas NLP, los principales metodos utili-

zados son los metodos cuadratica
de gradiente reducido y la programacion sucesiva
(SQP) 11 [121-124].


La idea basica de la SQP se centra en resolver las condiciones necesarias de op-

timalidad de KarushKuhnTucker utilizando el metodo
de Newton (o metodos cuasi

Newton). La gran ventaja de los metodos cuadratica
de programacion sucesiva es

que convergen de forma muy rapida
con muy poco numero de las
de evaluacion

funciones. Sin embargo para problemas con un elevado numero de grados de liber-
sea mejores los metodos
tad quizas de gradiente reducido, en tanto que reducen la
dimensionalidad del problema a resolver.


Los metodos del domi-
de gradiente reducido se basan en reducir la dimension

nio del problema representando todas las variables en termino de un subconjunto
de variables independientes. Dicho de otro modo, considerando un problema con m

restricciones en forma de igualdad (forma estandar), y n variables (n m), la idea es
de las restantes y resolver
despejar m variables (tantas como restricciones) en funcion
el problema como un problema sin restricciones. Una vez reducido la dimensionalidad

de problema, el primer paso que sigue el metodo es proponer una estrategia para
factible y de mejora. Una vez calculada esta,
calcular una direccion y siguiendo la

estrategia general es necesario realizar la busqueda univariable. El principal proble-
ma de esta metodologa, tal y como fue propuesto originalmente por Wolfe [125], es
11 Successive Quadratic Programming
Acronimo del ingles

74

Captulo 3. Introduccion

de variables de forma algebraica, al


que no siempre es posible hacer la eliminacion

menos con ecuaciones no lineales. Para solucionar esto y poder ampliar dicho meto-

do a problemas NLP, dicho metodo
ha sido extendido denominandose
Metodo del
Gradiente Reducido Generalizado (GRG). La principal novedad pasa por que en las
de
restricciones no lineales, para representar las m variables dependientes en funcion
de cada una de las
las (n m) variables independientes, se realiza una linealizacion
de Taylor de primer
restricciones en el punto factible actual mediante una expansion
orden.

no lineal,
Existen numerosos trabajos en la bibliografa entorno a la programacion
completos los realizados por Biegler et al. [126], Floudas
siendo algunos de los mas
[127] o Griva et al. [128].

3.4.3. Lineal Mixta con Variables Enteras y Continuas


Programacion


Un gran numero descritos por ecuaciones lineales
de modelos de optimizacion

pueden contener tanto variables continuas como enteras. Estos modelos matemati-
Lineal Mixta (MILP)
cos se denominan como Programacion 12 . El uso de variables en-
que pueden ser definidos
teras extiende el alcance de los problemas de optimizacion
y resueltos. Una amplia gama de aplicaciones se puede modelar como problemas
MILP. Estas aplicaciones han atrado muchas otras aplicaciones en el campo de in-
de operaciones y especficamente en la ingeniera qumica, habiendo
vestigacion
significativa sobre todo en las areas
recibido una atencion de sntesis de procesos,
y control.
planificacion

Un problema MILP puede ser descrito de forma general como:

min(omax) cT x + dT y

s.a. Ax + By b
(3.4.3)
x X Rn

y {0, 1}m
12 Mixed Integer Linear Programming
Acronimo del ingles

75
matematica
3.4.Programacion


donde x es un vector de n variables continuas que representa parametros como
etc. . . , y es un vector de q variables binarias, c y d son
flujos, temperatura, presion,

vectores de los parametros,
b es un vector de terminos independientes, y A y B son
matrices con dimensiones apropiadas.


Los metodos fundamentalmente basados en el metodo
para resolver MILPs estan

clasico y Acotamiento (BB) 13 [129, 130], donde cada subproblema
de Ramificacion

lineal se resuelve utilizando el metodo simplex .


La metodologa presentada en este metodo
empieza encontrando la solucion

optima de integridad de las varia-


del problema relajado, eliminando la condicion
bles binarias, de tal forma que puedan adquirir valores fraccionarios. Si en esta prime-
las variables de decision
ra solucion, relajadas adquieren valores enteros, entonces
iterativa y se habra alcanzado la solucion.
no se requiera mas solucion Sin embargo,
ocurre en un numero
esto solo muy reducido de casos (por ejemplo en los problemas
y lo normal es que algunas de las variables tomen valores fracciona-
de asignacion)
rios. En cualquier caso, este nodo inicial produce una cota inferior global del proble-

ma. De entre las variables que no han tomada valores enteros, el metodo BB escoge

una variable (de acuerdo a una serie de reglas predeterminadas) ramificandola y
creando dos nuevos sub-problemas. Estos sub-problemas se resuelven y el proceso se
que presenta todas las variables binarias
repite, hasta que se encuentra una solucion
entera, esta
con valores enteros (0 o 1). Cuando en un nodo se obtiene una solucion
optima
es un lmite superior a la solucion del problema, de tal forma que todas las
objetivo no necesitan evaluarse (aco-
ramas abiertas con valor superior de la funcion
continua
tamiento). La enumeracion hasta que la diferencia entre las cotas inferior y
dentro de una tolerancia o bien no existen ramas abiertas.
superior estan

El peor caso que se puede encontrar es aquel donde para finalizar el algoritmo
de todos los nodos del arbol.
es necesario la enumeracion Este caso es un caso ex-

tremo e implicara el consumo de importantes recursos de calculo
para la obtencion
al problema. Con el fin de mitigar el crecimiento exponencial en la
de la solucion
de MILPs se han desarrollado dos importantes tecnicas,
solucion
las tecnicas de pre-
de planos de corte. El preprocesamiento se basa
procesamiento y la introduccion
13 Branch and Bound
Acronimo del Ingles

76

Captulo 3. Introduccion

de de tecnicas
en la utilizacion automatica
de eliminacion de variables y restriccio-
de lmites, reformulacion
nes, reduccion de restricciones y fijado a priori de algunas
variables enteras. Los planos de corte son restricciones extra a
nadidas al problema,
que tienen el efecto de reducir la
bien en el nodo inicial o dentro de la enumeracion,
factible del problema sin comprometer ninguna de las soluciones enteras del
region
mismo.

3.4.4. No Lineal Mixta con Variables Enteras y Conti-


Programacion
nuas

no lineal de procesos qumicos


Una amplia gama de problemas de optimizacion
es
involucra variables enteras o discretas. Esta clase de problemas de optimizacion
No lineal Mixta (MINLP)14 . Las variables enteras pue-
modelado como Programacion
den ser usadas para modelar por ejemplo sucesiones de eventos, candidatos alter-
nativos, existencia o no-existencia de unidades.

Un problema MINLP puede ser descrito de forma general como:

min (o max) z = f (x, y)

s.a. gj (x, y) 0 jJ
(3.4.4)
x X Rn

y {0, 1}m

Donde se asume que f y g son funciones convexas y diferenciables, X esta da-


do por el conjunto convexo X = {x | xL x xU , Ax b} y J es un ndice de
desigualdades.


Numerosos metodos
numericos han sido desarrollados para resolver este tipo de

problemas (MINLP). Un primer grupo de metodos incluyen los algoritmos de ramifica-
y acotamiento (BB). Estos son generalmente una extension
cion directa del metodo

BB para problemas lineales, donde la diferencia radica en cada uno de los nodos
14 Mixed Integer Nonlinear Programming
Acronimo del ingles

77
matematica
3.4.Programacion

de problemas no lineales (NLP) (por ejem-


se emplea un algoritmo para la resolucion

plo, metodo cuadratica
de gradiente reducido, programacion
sucesiva o metodo
de punto interior) [131, 132]. Un segundo grupo incluyen los algoritmos de descom-
posicion de benders (GBD) 15 [127, 133], y los metodos
tales como la descomposicion
de aproximaciones exteriores (OA) 16 [59, 134].


El algoritmo OA fue propuesto por Duran y Grossmann [59]. En este,
la resolucion
de una secuencia iterativa de subproblemas NLP
del problema es realizada a traves

y subproblemas Master (que suele ser un MILP). Los subproblemas NLP, con valores fijos
de las variables enteras 0-1, proporcionan la cota superior del problema, mientas que
los problemas Master proporcionan la cota inferior del problema. La convergencia
del algoritmo se logra cuando ambos lmites -superior e inferior- difieren en un valor
inferior a una tolerancia establecida. La Figura 3.4.2 muestra un esquema simplificado
del algoritmo de aproximaciones exteriores.

de Benders Generalizada se puede considerar un caso parti-


La descomposicion
cular del algoritmo de las aproximaciones exteriores. Las principales diferencias sur-
del problema Master
gen en la definicion
(MILP). Si bien en el metodo OA todas las

desigualdades son consideradas, en el metodo de Benders uni-
de descomposicion

camente se consideran las desigualdades activas, ignorandose el resto. Este hecho
del problema Master
hace que el coste computacional requerido para la resolucion
sea menor al necesario al utilizar OA. Por otro lado, los lmites inferiores predichos en
el problema relajado son inferiores o iguales a los obtenidos utilizando OA. Este hecho

es responsable de que, generalmente, el metodo GDB suela necesitar de un mayor

numero con el metodo
de iteraciones para resolver el problema, en comparacion
OA.

intermedia entre los metodos


En una situacion y acotamiento y los
de ramificacion

metodos se encuentran los metodos
de descomposicion LP-NLP [135]. En este caso en

lugar de resolver el problema Master hasta optimalidad, tan pronto como se encuen-
entera se resuelve un NLP y se actualizan todos los nodos abiertos
tra una solucion

del arbol
de busqueda.
Finalmente se encuentran los metodos de plano de corte ex-
15 Generalized Benders Decomposition
Acronimo del ingles
16 Outer Approximation
Acronimo del ingles

78

Captulo 3. Introduccion

Figura 3.4.2: Esquema del algoritmo de Aproximaciones Exteriores

tendido (ECP) 17 [136-138], desarrollados a partir de los metodos



de descomposicion.

En estos,
los subproblemas NLP no son resueltos, el algoritmo se reduce a la solucion
iterativa de problemas MILP a
nadiendo linealizaciones sucesivas a aquella restriccion
violada en el punto predicho. La convergencia se obtiene cuando la violacion
mas

maxima cae dentro de una tolerancia especificada. Cabe se
nalar que todos estos

metodos
unicamente optima
garantizan la solucion global de los problemas MINLP si
se cumple que todas las funciones no lineales fueran convexas.

17
Acronimo del inglesExtended Cutting Plane

79
matematica
3.4.Programacion

3.4.5. General Disyuntiva


Programacion

Como ya se ha comentado anteriormente, la sntesis y dise


no de procesos qumi-
cos es una tarea altamente compleja que requiere el desarrollo de modelos de pro-
matematica,
gramacion los cuales suelen incluir variables enteras y continuas, donde
las variables enteras juegan un papel clave. Estas variables son normalmente varia-
de una unidad o asignacion
bles binarias con 1 indicando la seleccion de una
tarea, y 0 indicando la ausencia de la unidad o tarea.

Una herramienta que ha demostrado ser altamente eficiente en el manejo y mo-


de sistemas es la Programacion
delado de problemas de sntesis y optimizacion Dis-
yuntiva Generalizada (GDP) 18 . Dicha metodologa aprovecha el conocimiento de la

estructura y topologa de los procesos qumicos as como las relaciones logicas entre
util
las diferentes unidades del proceso, proporcionando informacion sobre el con-

junto de soluciones factibles y mejorando de esta forma la eficiencia de busqueda
disyuntiva originalmente pro-
combinatorial[139-146]. Partiendo de la programacion
para
puesta por Balas [139], Raman y Grossmann [142] propusieron una formulacion
la sntesis de procesos, la cual facilita el modelado de problemas lineales continuos y
del uso de expresiones simbolicas,
enteros a traves la cual incluye de forma general

variables Booleanas, disyunciones y proposiciones logicas.


La estructura tpica de este tipo de problemas se muestra a continuacion.
18 Generalized Disjunctive Programming
Acronimo del ingles

80

Captulo 3. Introduccion

X
M in Z= ck + f (x)
k

s.t. r(x) 0

s(x) = 0

Yik (3.4.5)

gik (x) 0 k K


iDk
ck = ik

(Y ) = T rue

x Rn , c Rm , Y {true, f alse}m

Siendo:

x : vector de variables continuas asociadas al proceso

los cuales son activados en el


ck : costes fijos asociados a cada disyuncion,
sea verdadera
caso que la disyuncion


Yik : variables booleanas que establecen si un termino es
en la disyuncion

verdadero y por lo tanto deben cumplirse las ecuaciones dentro del termino


(Y ): relaciones logicas,
en forma de logica de proposiciones, que involu-
cran solamente a variables booleanas

esta
Una disyuncion compuesta por un numero de terminos
(condiciones) Dk uni-

dos por un operador O exclusivo (). Cada termino esta compuesto por una varia-
ble Booleana Yik , y un conjunto de restricciones no lineales gik (x). La principal ca-
es que cuando uno de sus terminos
racterstica de una disyuncion Yik se cumple (Yik

es verdadero), todas las restricciones asociados con este son activadas gik (x) 0 y
ck = ik . De lo contrario, las correspondientes restricciones son ignoradas. Cabe des-
disyuntiva dada, las variables binarias 0-1 no se encuen-
tacar, que en la formulacion
tran incluidas explcitamente en los modelos y que todas las decisiones son realizadas
de variables booleanas.
a traves

81
matematica
3.4.Programacion

En los problemas de sntesis de procesos, gik (x) y hik (x) son ecuaciones de balan-
ce de calor y materia, o especificaciones del proceso, y ik son los cargos o costes.
En numerosos trabajos realizados [135, 143], se ha supuesto que las funciones f (x),
gik (x) y r(x) son convexas (PDG Convexo), con lmites inferiores y superiores, y que
o solucion
las disyunciones dan lugar a una region factible. Pero en realidad, muchos
problemas de sntesis de procesos en ingeniera qumica incluyen restricciones y fun-
ciones objetivo muy complejas no convexas, lo que da lugar a PDG No convexa. Si no

se usan metodos de estas funciones no convexas o
potentes para la convexificacion
de la influencia de estas,
para la disminucion las soluciones del problema de sntesis

puede ser suboptimas
multiples o incluso, no factibles. Cuando las restricciones gik (x),
objetivo f (x) son funciones lineales, el problema se convier-
r(x), hik (x) y la funcion
Disyuntiva Lineal, y si cualquiera de las funciones es no lineal, el
te en Programacion
Disyuntiva No Lineal.
problema se denomina Programacion

3.4.5.1.
Metodos para Problemas GDP
de Resolucion


Los metodos de problemas GDP se pueden clasificar en dos ca-
de resolucion

tegoras, los metodos directa (sin necesidad de ser reformulados) y los
de resolucion

metodos de los problemas GDP a problemas MILP/MINLP.
que realizan la reformulacion


Metodos directa
de resolucion


Diferentes metodos directa de problemas
han sido desarrollados para la resolucion

GDP. Concretamente, dos metodos
han sido presentados, por un lado el metodo de
y acotamiento (BB) desarrollado por Lee y Grossmann [146] basado en
ramificacion

el metodo disyuntivo Branch and Bound propuesto por Beaumont [140], y por otro el

metodo
de Aproximaciones Exteriores basadas en la logica (LBOA) 19 propuesto por
Turkay y Grossmann [143].


La idea basica
del metodo directa en los terminos
BB se basa en la ramificacion
de envolvente convexa al
de las disyunciones, mientras es aplicada la reformulacion

resto de disyunciones. Este metodo proporciona resultados similares a los obtenidos al
de envolvente convexa. La
resolver el problema como MINLP utilizando reformulacion
19 Logic Based Outer Approximations
Acronimo del ingles

82

Captulo 3. Introduccion

diferencia radica en el hecho de que el tama


no de los problemas a resolver en cada

nodo son menores, con ello se consigue mejorar la robustez del metodo respecto al
problema MINLP.


En el caso del metodo
LBOA, el procedimiento es similar al metodo OA descrito

anteriormente. La idea basica consiste en resolver iterativamente el problema Master
el lmite inferior de la solu-
dado por el problema GDP Lineal (el cual proporcionara
y un subproblema NLP (el cual proporcionara
cion) el lmite superior de la solucion).

La principal diferencia entre ambos radica en el subproblema NLP resuelto en cada
Tal y como describieron Turkay y Grossmann [143], dado unos valores fijos de
iteracion.
las variables Booleanas, Yki = verdadero, Yki = f also donde i 6= i, el correspondiente
subproblema NLP queda definido de la siguiente forma:

M in Z = f (x)

s.t. r(x) 0

gki (x) 0 Yki = verdadero k K, i Dk (3.4.6)


lo up
x xx

x Rn , ck RI , Yki {true, f alse}


Como se puede observar las unicas restricciones consideradas en el subproble-

ma NLP son aquellas pertenecientes a los terminos (Yki =
activos de la disyuncion

verdadero). Las restricciones pertenecientes a los terminos inactivos son descartadas.
en el tama
Esto permite conseguir una sustancial reduccion no del problema en com-
con el metodo
paracion OA para problemas reformulados como MINLP. El problema
Master OA disyuntivo a resolver, se define como:

83
matematica
3.4.Programacion

M in Z=

s.t.

f (xl ) + f (xl )T (x xl )
l = 1,2. . . . , L
r(xl ) + r(xl )T (x xl ) 0

Yki (3.4.7)
kK
iDk gki (xl ) + gki (xl )T (x xl ) 0 l Lki

(Y ) = T rue

xlo x xup

R1 , x Rn , ck R1 , Yki {true, f alse}

Cabe se
nalar que antes de aplicar el problema Master descrito es necesario re-
solver varios subproblemas (SNLP) para diferentes valores de las variables booleanas
lineal de cada uno de los terminos
Yik a fin de producir una aproximacion i Dk en

las disyunciones k K (de tal forma que todos los terminos de las diferentes disyun-
ciones se encuentren activos al menos una vez). Como muestran Turkay y Grossmann
del menor numero
[143], la seleccion del pro-
de subproblemas equivale a la solucion
blema de conjunto de cobertura, el cual es de peque
no tama
no y facil de resolver.
Finalmente, debe tenerse en cuenta que el problema Master puede ser reformulado

como un MILP mediante el uso del metodo de la M grande o la envolvente convexa,
o bien resolverse directamente mediante un problema de BB disyuntivo.

de GDP a MILP/MINLP
Reformulacion


A pesar de la existencia de los metodos directa, generalmente estos
de resolucion
problemas son reformulados como problemas MILP o MINLP con el fin de aprovechar
la disponibilidad de los algoritmos MINLP presentes en las diferentes herramientas de
modelado (GAMS, TOMLAB, . . . ).

Con el fin de describir detalladamente cada uno de las transformaciones realiza-



das en los distintos metodos, el problema general GDP (3.4.5) se divide en 4 partes,
objetivo, restricciones globales, disyunciones y proposiciones logicas.
funcion

84

Captulo 3. Introduccion

de un problema GDP en un MILP o MINLP requiere unicamente


La transformacion

transformar la parte correspondiente a las disyunciones y las proposiciones logicas,

manteniendose objetivo como las restricciones globales. A
invariable tanto la funcion
de explicara
continuacion brevemente cada uno de dichas transformaciones.

de proposiciones logicas
Transformacion a restricciones MILP


Las proposiciones logicas de variables Booleanas deben ser
descritas a traves
transformadas a ecuaciones algebraicas compuestas por variables binarias, con el
MINLP de problema original. Cada
fin de poder ser incorporadas a la reformulacion

uno de los operadores logicos disyuntiva es transforma-
utilizados en la programacion
do en un conjunto de restricciones lineales (MILP), tal y como se describe en la Tabla
3.4.1:

de operadores logicos
Tabla 3.4.1: Transformacion a restricciones MILP


Operador Logico Booleana
Expresion MILP
Restriccion
(O) Y1 Y2 . . . Yn y1 + y2+...+yn 1
(O exclusivo) Y1 Y2 . . . Yn y1 + y2+...+yn =1
y1 = 1
y2 = 1
(Y) Y1 Y2 . . . Yn
...
yn = 1
(No) Y1 1 y1
(Si) Y1 Y2 y1 y2
(Si y solo Si) Y1 Y2 y1 = y2

con el fin de facilitar la transformacion


Ademas, de una proposicion
logica
en
lineal (reemplazando las variables booleanas por variables binarias),
una ecuacion
se deben emplear el siguiente procedimiento de forma recursiva [147]:

de implicaciones
(1) Eliminacion Y1 Y2
Y1 Y2
(2) Ley de De Morgan (Y1 Y2 )
(Y1 ) (Y2 )
(Y1 Y2 )
(Y1 ) (Y2 )
(3) Regla Distributiva (Y1 Y2 ) Y3
(Y1 Y2 ) (Y2 Y3 )

85
matematica
3.4.Programacion

logica.
Existen numerosas formas de representar un proposicion comunes
Las mas
y usadas son las dos formas extremas [141]: la Forma Normal Conjuntiva (CNF) 20 y la
Forma Normal Disyuntiva (DNF) 21 .

en-
La diferencia entre ambas formas consiste en que la CNF describe la relacion

tre las unidades de proceso o tareas en la superestructura y cada clausula en ella se
debe cumplir para el problema de que se trata, mientras que la DNF describe una
factible dentro de la superestructura y solamente una de las condicio-
configuracion
nes debe ser satisfecha.

la forma CNF. Esta forma se


En este tipo de problemas es de particular interes
logica
caracteriza por representar una proposicion de casos
como la conjuncion
logica
Q1 Q2 . . . Qn , donde cada caso corresponde a una proposicion con-

teniendo unicamente operadores tipo O. Como ya se ha comentado, para que
la forma CNF sea verdadera, cada uno de los casos que la componen deben ser
verdaderos. Utilizando las reglas booleanas comentadas previamente, el proceso de
sistematica
transformacion logica
de cualquier proposicion en una forma CNF se con-
de las siguientes etapas [148-150]:
sigue a traves

por su disyuncion
Paso 1. Reemplazar la implicacion equivalente

al interior del
Paso 2. Aplicar las leyes de De Morgan para mover la negacion

parentesis

Paso 3. Aplicar la regla distributiva para conseguir que dentro de las disyunciones

unicamente presenten operadores O


Una vez obtenida la forma CNF, el ultimo paso es traducir dicha forma a una serie
de restricciones algebraicas con el fin de ser introducidas en el problema a resolver.
un ejemplo es presentado con el fin de ilustrar el procedimiento des-
A continuacion,
crito.

logica
Consideramos la proposicion [148]:
20 Conjunctive Normal Form
Acronimo del ingles
21 Disjunctive Normal Form
Acronimo del Ingles

86

Captulo 3. Introduccion

(Y1 Y2 ) Y3 (Y4 Y5 ) (3.4.8)

por su disyuncion
Paso 1. Reemplazar la implicacion equivalente

[(Y1 Y2 ) Y3 ] (Y4 Y5 ) (3.4.9)

al interior del
Paso 2. Aplicar las leyes de De Morgan para mover la negacion

parentesis

[(Y1 Y2 ) Y3 ] Y4 Y5
(3.4.10)
[(Y1 Y2 ) Y3 ] Y4 Y5

Paso 3. Aplicar la regla distributiva para conseguir que dentro de las disyunciones

unicamente presenten operadores O

(Y1 Y2 Y4 Y5 ) (Y3 Y4 Y5 ) (3.4.11)

de la forma CNF a su equivalente conjunto de restricciones


Paso 4. Transformacion
lineales

y1 + y2 y4 y5 1
(3.4.12)
y3 y4 y5 0

87
matematica
3.4.Programacion

de disyunciones a restricciones MILP/MINLP


Transformacion

de las disyunciones involucra su reformulacion


La relajacion a uno de los modelos
matematica
de la programacion anteriormente descritos (MILP, MINLP) dependiendo
tpi-
de si el problema GDP original es lineal o no lineal. La estrategia de relajacion

ca se lleva a cabo mediante el metodo
de M-Grande (Big-M)[130], o el metodo de
Envolvente Convexa (CH)22 [146].


Metodo de M-Grande [151]:

Disjunction GDP Restricciones M-Grande

gki (x) M ki (1 yki )



Yki k K, i Dk
kK X
iDk gki (x) 0 yki = 1 kK
iDk

radica en el hecho de que cuando un termino


La clave de esta formulacion de
se encuentra activo (yki = 1), las restricciones asociadas a esta se vuel-
la disyuncion

ven activas (gki (x) 0). En los casos inactivos (yki = 0) y siendo el parametro M
lo suficientemente grande, las restricciones asociadas a estas se transforman en re-

sentido en el proceso. Indicar que la ecuacion
dundantes, dejando de tener ningun
P
y
iDk ki = 1 asegura que en cada disyunci
on
unicamente un t
ermino sea activo
(solamente una variable binaria yki en cada disyuncion
en cada solucion k adquiera
valor 1).


Metodo de Envolvente convexa [151]:

de CH puede ser diferenciada para los casos lineales y no linea-


La reformulacion
lineal no es mas
les, aunque la formulacion que un caso particular de la formulacion

no lineal.

22 Convex Hull
Acronimo del ingles

88

Captulo 3. Introduccion

Disjunction GDP Restricciones Envolv. Convexa No Lineal

X
x= ki
iDk kK

Yki yki gki ( ki /yki ) 0 k K, i Dk
kK
iDk gki (x) 0 xlo yki ki xup yki k K, i Dk
X
yki = 1 kK
iDk

No Lineal:

es la disgregacion
La principal caracterstica de esta formulacion de las variables
continuas x en un conjunto de variables ki tantas como terminos
contenga cada
En este caso, es necesario introducir lmites superiores e inferiores de cada
disyuncion.
variable xlo yki ki xup yki . Con ello se consigue que si yki se encuentra activa
(yki = 1), dicha variable disgregada adquiera un valor comprendido entre dichos
lmites, en cambio si yki se encuentra inactiva (yki = 0) se fuerza a que la variable dis-
gregada ki adquiera valor 0. Indicar que al igual que en el caso de la reformulacion

P

M-Grande, la ecuacion iDk yki = 1 asegura que en cada disyuncion unicamente


un termino (solamente una variable binaria yki en cada
sea activo en cada solucion
k adquiera valor 1). Finalmente, hay que indicar que en problemas no li-
disyuncion
de las restricciones en el metodo,
neales y como consecuencia de la formulacion en

los casos donde yki = 0 pueden tener lugar singularidades que dificultan la resolucion
de dicho problema. este hecho puede mitigarse transformando la forma general de
[152]:
las restricciones por la siguiente aproximacion

ki
yki gki ( ki /yki ) ((1 )yki + )gki ( ) gki (0)(1 yki ) (3.4.13)
(1 )yki +

Donde es un peque
no numero finito (approx. 105 ). esta aproximacion
es con-
vexa si gki (x) es convexa y produce un valor exacto tanto para yki = 0 como yki = 1.

89
matematica
3.4.Programacion

de un
Finalmente, las Tablas 3.4.2 y 3.4.3 muestran un resumen de la reformulacion

problema GDP utilizando tanto el metodo
de MGrande como el metodo de Envol-
vente Convexa en un problema lineal y no lineal respectivamente.

90
Tabla 3.4.2: Esquema Reformulaciones M-Grande y Envolvente Convexa de un Problema Lineal
Problema GDP Lineal

Disjunction GDP M-Grande Envolvente Convexa

minz = f (x) minz = f (x) minz = f (x)

s.a.r(x) 0 s.a.r(x) 0 s.a.r(x) 0

X
x= ki
iDk kK

Aki x aki + M ki (1 yki )
Yki k K, i Dk Aki ki aki y ki k K, i Dk
kK X
iDk Aki x aki yki = 1 kK xlo yki ki xup yki k K, i Dk
iDk
X
yki = 1 kK
iDk

(Y ) = V erdadero Hx h Hx h

xlo x xup , x Rn xlo x xup , x Rn x Rn

Yki {V erdadero, F also} k K, i Dk yki {0, 1} k K, i Dk yki {0, 1} k K, i Dk


91
Captulo 3. Introduccion
92
Tabla 3.4.3: Esquema Reformulaciones M-Grande y Envolvente Convexa de un Problema No Lineal
Problema GDP No Lineal

Disjunction GDP M-Grande Envolvente Convexa


3.4.Programacion

minz = f (x) minz = f (x) minz = f (x)


s.a.r(x) 0 s.a.r(x) 0 s.a.r(x) 0


matematica

X
x= ki
iDk kK

Yki gki (x) M ki (1 yki ) k K, i Dk k K, i Dk
yki gki ( ki /yki ) 0
kK X
iDk gki (x) 0 yki = 1 kK lo ki up k K, i Dk
x yki x yki
iDk
X
yki = 1 kK
iDk

(Y ) = V erdadero Hx h Hx h
lo up n lo up n
x xx , xR x xx , xR x Rn

Yki {V erdadero, F also} k K, i Dk yki {0, 1} k K, i Dk yki {0, 1} k K, i Dk



Captulo 3. Introduccion


Como se puede deducir a partir de los metodos CH
presentados, la reformulacion

presenta un numero M-Grande, como con-
mayor de variables que la reformulacion
de variables llevada a cabo. Ademas
secuencia de la disgregacion dicho metodo


aumenta el numero de restricciones respecto a la M-Grande (dos nuevas ecuacio-
extra que relaciona la variable
nes por cada variable disgregada y una ecuacion

original con las variables disgregadas). Todo ello lleva a concluir que el metodo CH

genera problemas mayores que el metodo M-Grande. Sin embargo, tal y como de-
CH consigue
mostraron Grossmann y Lee [153]y Vecchietti et al. [154], la formulacion
eficientemente (mejores relajaciones) cuando las variables
acotar el problema mas

binarias son relajadas, reduciendo el espacio de busqueda. Esta caracterstica es de
gran importancia ya que la eficacia de los algoritmos MiLP/MINLP depende conside-
rablemente de la calidad de estas relajaciones. En la figura 3.4.3 [151] se muestran

las relajaciones previstas por ambos metodos en el espacio de las variables continuas

para un problema GDP lineal compuesto por dos disyunciones (dos terminos por dis-

yuncion). factible (cuadrados ra-
Como se puede observar, mientras que esta region

yados) es la misma para ambas formulaciones, las relajaciones continuas (area azul)

son diferentes, observandose empleando CH respecto
una considerable reduccion
a M-Grande.

de la relajacion
Figura 3.4.3: Representacion utilizando M-Grande y Envolvente Con-
vexa

93
matematica
3.4.Programacion

3.4.6. Algoritmos computacionales


En las ultimas
dos decadas,
el numero
de codigos de problemas
para la resolucion

MINLP se ha incrementado considerablemente. El codigo DICOPT [155] es un algorit-
mo utilizado en problemas MINLP basado en el algoritmo de aproximaciones exte-
riores con algunas diferencias y disponible en el sistema de modelado GAMS [156].

Como caractersticas principales, este codigo utiliza un subproblema NLP relajado
del problema Master (MILP) correspondiente a
para generar la primera linealizacion
gracias a lo cual el usuario no necesita especificar un valor en-
la primera iteracion,
y debido a que el lmite inferior riguroso del problema Master
tero inicial. Ademas no

puede ser garantizado, la busqueda optima
de la solucion de los problemas no con-
vexos se considera finalizada cuando no se encuentra mejora en los problemas NLP
factibles en dos iteraciones sucesivas. Esta heurstica funciona razonablemente bien

en la mayora de los casos. Un codigo similar a DICOPT (basado en OA) pero imple-

mentado en AIMMS [157] es el programa AAOA. Algunos codigos que implementan

los metodos y acotamiento resolviendo subproblemas NLP en cada
de ramificacion
uno de los nodos incluyen: el MINLP BB basado en el algoritmo SQP [131] (Sequential
Quadratic Programming) e implementado en AMPL y el programa SBB implementa-

do en GAMS. El codigo
-ECP, disponible en GAMS, implementa el metodo de plano
de Wes-
de corte extendido de Westerlund y Pettersson [136] incluyendo la extension
[138]. El codigo
terlund y Porn implementa los metodos
MINOPT [158] tambien OA y
dinamica
GBD, y los aplica a optimizacion de estos,
con variables enteras. Ademas

existen codigos de uso libre como Bonmin [159], el cual implementa los algoritmos
del LP-NLP en un unica
BB, OA y una extension
estructura. Otro codigo libre es FilMINT
del algoritmo LP-NLP basado en el metodo
[160], el cual implementa una variacion
y acotamiento.
de ramificacion

de los problemas MINLP no es una tarea


Como ya se ha comentado, la resolucion

trivial, especialmente comparado con los problemas MILP. La principal problematica
de los subproblemas NLP, y por ello se han realizado grandes
radica en la resolucion

esfuerzos con el fin de mejorar dicho problema. Algunos de los principales codigos

computacionales desarrollados incluyen CONOPT [161] que implementa el metodo

de gradiente reducido, SNOPT [162] que implementa el metodo
de programacion

94

Captulo 3. Introduccion


cuadratica
sucesiva, y KNITRO [163] e IPOPT [164] los cuales implementan el metodo

de punto interior. El desarrollo de estos metodos hace posible actualmente resolver
modelos con miles de variables.


Sin embargo, y como consecuencia de que todos estos
se basan en el metodo
de Newton, son necesarios buenos puntos de partida para garantizar la convergen-

cia. Los metodos
descritos hasta ahora solamente garantizan el optimo global si el
modelo NLP es convexo. Sin embargo, la mayora de los modelos NLP asociados a los

procesos industriales no son convexos, por ello es necesario aplicar metodos de op-
global rigurosa -basados en los metodos
timizacion y acotamiento-
de ramificacion

para asegurar el optimo
global [165]. Algunos de estos codigos son: BARON [166],
de
LINDOGlobal [167], GloMIQO [168] y Couenne [169]. Sin embargo la aplicacion

los metodos global en problemas grandes (miles de ecuaciones y
de optimizacion
variables) llega a ser muy costoso desde el punto de vista computacional. En conse-
cuencia, y teniendo en cuenta, tanto el problema de robustez de los algoritmos NLP
global para los
como el alto coste computacional de los algoritmos de optimizacion
del problema
modelos MINLP, un enfoque que se utiliza a menudo es la reformulacion
MINLP a un problema MILP utilizando, por ejemplo, linealizaciones exactas o el uso de

aproximaciones lineales a trozos. El problema es que esta tecnica puede llevar-
solo
aproximada
se a cabo en relativamente pocos casos. En definitiva, y como solucion

en la mayora de los casos, se suele utilizar metodos como DICOPT o SBB para la re-
de problemas MINLP no convexos, aunque dichos metodos
solucion no aseguren la
optima.
solucion

Referencias

[1]
Gas, Petrotecnia. Instituto Argentino del Petroleo y del. Siglo XX: el siglo de la Energa.

Panorama energetico del mundo y de la Argentina. Actualidad y futuro. http : / / www .
petrotecnia.com.ar. 2009.
[2] Petroleum, British. Statistical Review of World Energy 2012. http://www.bp.com/ statistical-
review. 2012.
[3] Agency, International Energy. World Energy Outlook 2012. http://www.worldenergyoutlook.
org//. 2012.
[4] Copenhagen, University of. International Scientific Congress Climate Change: Global Risks,
Challenges & Decisions Synthesis Report, IARU (International Alliance of Research Uni-
University of Copenhagen, 2009.
versities). Inf. tec.

95
3.4.REFERENCIAS

[5] WMO. WMO Greenhouse Gas Bulletin, The State of Greenhouse Gases in the Atmosphere
World Meteorological Organization,
Using Global Observations through 2011 No 7. Inf. tec.
2011.
[6] World Meteoro-
WMO. WMO statement on the status of the global climate in 2011. Inf. tec.
logical Organization, 2011.
[7] Climate Change (IPCC), Intergovernmental Panel on. http://www.ipcc.ch.
[8] Climate Change (IPCC), Intergovernmental Panel on. Informe Especial del IPCC: Escena-
rios de Emisiones. http://www.ipcc.ch/pdf/special-reports/spm/sres-sp.pdf. 2001.
[9] Nations, United. Kyoto Protocol to the United Nations Framework Convention on Climatwe
United Nations, 1998.
Change. Inf. tec.
[10] Climate Change (UNFCCC), United Nations Framework Convention on. Copenhagen Ac-
cord, 18 December 2009. 2009.
[11] Science, EU Climate Change Expert Group. The 2C target, Information Reference Docu-
European Commission, 2008.
ment. Inf. tec.
[12] Europea.
Abadia, J. Energa y Sostenibilidad. La Energa y el Medio Ambiente en la Union
ENDESA, 2006.
Inf. tec.
[13] 2009 ASPO-USA Peak
Nelder, C. Efficiency: The Way to Close the Renewables Gap. Inf. tec.
Oil Conference, 2009.
[14] Trainer, F.E. Renewable Energy Cannot Sustain a Consumer Society. Springer London, Limi-
ted, 2007.
[15] Metz, B. y Climate Change. Working Group III., Intergovernmental Panel on. Carbon Dioxi-
de Capture and Storage: Special Report of the Intergovernmental Panel on Climate Chan-
ge. Cambridge University Press, 2005.
[16] Council, Coal Utilization Research e Institute, Electric Power Research. The CURC-EPRI 2012
Coal Technology Roadmap). Inf. tec. Coal Utilization Research Council, 2012.
[17] MIx, T.J., Dueck, J.S. y Weinberg, M. ((Energy conservation in distillation)). En: Chemical En-

gineering Progress 74 (1978), pags. 49-55.
[18] Humprey, J. ((Separation processes: Playing a critical role)). En: Chemical Engineering Pro-

gress 91 (1995), pags. 31-41.
[19] Engelien, H.K. y Skogestad, S. ((Selecting appropriate control variables for a heat-integrated
sistillation system with prefractionator)). En: Computers & Chemical Engineering 28 (2004),

pags. 683-691.
[20] Humphrey, J.L. y Siebert, A.F. ((Separation technologies: an opportunity for energy savings)).

En: Chemical Engineering Progress 91 (1992), pags. 31-41.
[21] Goldstine, H.H. The Computer from Pascal to von Neumann. Princeton University Press,
1993.
[22] White, J.A. Development of a Modified Pacer Simulation Program for Use in a Chemical
Engineering Curriculum. University of Kansas, Chemical y Petroleum Engineering, 1968.
[23] Braunschweig, B.L. y col. ((Process Modelling: The promise of open software architectures)).

En: Chemical Engineering Progress 96(9) (2000), pags. 65 -76.
[24] Kooijman, H.A. y Taylor, R. The ChemSep Book. Books on Demand, 2006.
[25] Sargent, R.W.H y Westerberg, A.W. ((SPEEDUP in Chemical Engineering Design)). En: Tran-

sactions of the Institution of Chemical Engineers 42 (1964), pags. 190-197.

96

Captulo 3. Introduccion

[26] Ledet, W.P. e Himmelblau, D.M. ((Decomposition Procedures for the Solving of Large Scale
Systems)). En: ed. por Thomas B. Drew Giles R. Cokelet, John W. Hoopes y Vermeulen, Theo-

dore. Vol. 8. Advances in Chemical Engineering. Academic Press, 1970, pags. 185 -254.
[27] Tarjan, R. ((Depth-First Search and Linear Graph Algorithms)). En: SIAM Journal on Compu-

ting 1.2 (1972), pags. 146-160.
[28] Kehat, E y Shacham, M. ((Chemical process simulation programs-2)). En: Process Techno-

logy International 18.3 (1973), pags. 115-118.
[29] Christensen, J.H. y Rudd, D.F. ((Structuring design computations)). En: AIChE Journal 15.1

(1969), pags. 94-100.
[30] Barkley, R.W. y Motard, R.L. ((Decomposition of nets)). En: The Chemical Engineering Journal

3.0 (1972), pags. 265 -275.
[31] Pho, T.K. y Lapidus, L. ((Topics in computer-aided design: Part I. An optimum tearing algo-
rithm for recycle systems)). En: AIChE Journal 19.6 (1973), pags.
1170-1181.
[32] Upadhye, R.S. y Grens, E.A. ((Selection of decompositions for chemical process simulation)).

En: AIChE Journal 21.1 (1975), pags. 136-143.
[33] De Hemptinne, J.C., Ledanois, J.M. y Mougin, P. Select Thermodynamic Models for Process
Simulation: A Practical Guide Using a Three Steps Methodology. Editions Technip, 2012.
[34] Kontogeorgis, G.M. y Folas, G.K. Thermodynamic Models for Industrial Applications: From
Classical and Advanced Mixing Rules to Association Theories. Wiley, 2009.
[35]
Aspen Technology, Inc. Aspen Physical Property System - Physical Property Models. Inf. tec.
Aspen Technology, Inc., 2010.
[36] Galindo, A. y col. ((The thermodynamics of mixtures and the corresponding mixing rules in
the SAFT-VR approach for potentials of variable range)). En: Molecular Physics 93.2 (1998),

pags. 241-252.
[37] Gil-Villegas, A. y col. ((Statistical associating fluid theory for chain molecules with attractive
potentials of variable range)). En: J. Chem. Phys 106.10 (1997), pags. 4168-4186.
[38] Mukherji, D. y col. ((Kirkwood-Buff Analysis of Liquid Mixtures in an Open Boundary Simula-
tion)). En: Journal of Chemical Theory and Computation 8.2 (2012), pags. 375-379.
[39] King, C.J. ((From Unit Operations to Separation Processes)). En: Separation & Purification

Reviews 29.2 (2000), pags. 233-245.
[40] Rudd, D.F., Powers, G.J. y Siirola, J.J. Process Synthesis. Prentice Hall International Series in
the Physical and Chemical Engineering Sciences Series. Prentice-Hall, 1973.
[41] Barnicki, S.D. y Siirola, J.J. ((Process synthesis prospective)). En: Computers & Chemical En-
gineering 28.4 (2004), pags. 441 -446.
[42] Douglas, J. M. ((A hierarchical decision procedure for process synthesis)). En: AIChE Journal

31.3 (1985), pags. 353-362.
[43] Nishida, N., Stephanopoulos, G. y Westerberg, A.W. ((A review of process synthesis)). En:

AIChE Journal 27.3 (1981), pags. 321-351.
[44] Westerberg, A.W. ((A retrospective on design and process synthesis)). En: Computers & Che-

mical Engineering 28.4 (2004), pags. 447 -458.
[45] Yuan, Z., Chen, B. y Gani, R. ((Applications of process synthesis: Moving from conventional
chemical processes towards biorefinery processes)). En: Computers & Chemical Enginee-

ring 49.0 (2013), pags. 217 -229.

97
3.4.REFERENCIAS

[46] Harmsen, J. y col. ((Industrially applied process synthesis method creates synergy between
economy and sustainability)). En: Foundations of Computer-Aided Process Design. Ed. por
Malone, M. F., Trainham, J. A. y Carnahan, B. Vol. 323. AIChE Symposium Series. 2000,

pags. 364-366.
[47] Siirola, J.J. y Rudd, D.F. ((Computer-Aided Synthesis of Chemical Process Designs. From
Reaction Path Data to the Process Task Network)). En: Industrial & Engineering Chemistry
Fundamentals 10.3 (1971), pags. 353-362.
[48] Seader, J.D. y Westerberg, A.W. ((A combined heuristic and evolutionary strategy for synt-
hesis of simple separation sequences)). En: AIChE Journal 23.6 (1977), pags.
951-954.
[49] Douglas, J.M. Conceptual design of chemical processes. McGraw-Hill chemical enginee-
ring series. McGraw-Hill, 1988.
[50] Smith, R. y Linnhoff, B. ((The design of separators in the context of overall processes)). En:
Chemical Engineering Research and Design 66 (1988), pags. 195-228.
[51] Nath, R. y Motard, R. L. ((Evolutionary synthesis of separation processes)). En: AIChE Journal

27.4 (1981), pags. 578-587.
[52] Douglas, J.M. ((Process synthesis for waste minimization)). En: Industrial & Engineering Che-

mistry Research 31.1 (1992), pags. 238-243.
[53] Linninger, A.A. ((Metallurgical Process Design A Tribute to Douglas Conceptual Design Ap-
proach)). En: Industrial & Engineering Chemistry Research 41.16 (2002), pags.
3797-3805.
[54] Smith, R. Chemical process design and integration. Wiley, 2005.
[55] Grossmann, I.E. ((Mixed-integer programming approach for the synthesis of integrated pro-
cess flowsheets)). En: Computers & Chemical Engineering 9.5 (1985), pags.
463 -482.
[56] Grossmann, I.E. ((MINLP optimization strategies and algorithms for process synthesis)). En:

Proc. FOCAPD Meeting (1990), pags. 105 -132.
[57] Umeda, T., Hirai, A. e Ichikawa, A. ((Synthesis of optimal processing system by an integrated
approach)). En: Chemical Engineering Science 27.4 (1972), pags. 795 -804.
[58] Papoulias, S.A. y Grossmann, I.E. ((A structural optimization approach in process synthesis
II: Heat recovery networks)). En: Computers & Chemical Engineering 7.6 (1983), pags.
707
-721.
[59] Duran, M.A. y Grossmann, I.E. ((An outer-approximation algorithm for a class of mixed-
integer nonlinear programs)). En: Mathematical Programming 36 (3 1986), pags.
307-339.
[60] Kocis, G.R. y Grossmann, I.E. ((Relaxation strategy for the structural optimization of process
flowsheets)). En: Industrial & Engineering Chemistry Research 26.9 (1987), pags.
1869-1880.
[61] Yee, T.F. y Grossmann, I.E. ((Simultaneous optimization models for heat integration - II. Heat
exchanger network synthesis)). En: Comput. Chem. Eng. 14.10 (1990), pags. 1165 -1184.
[62] Friedler, F. y col. ((Graph-theoretic approach to process synthesis: Polynomial algorithm
for maximal structure generation)). En: Computers & Chemical Engineering 17.9 (1993),

pags. 929 -942.
[63] Yeomans, H. y Grossmann, I.E. ((A systematic modeling framework of superstructure optimi-
zation in process synthesis)). En: Computers & Chemical Engineering 23.6 (1999), pags.
709
-731.
[64] Papalexandri, K.P. y Pistikopoulos, E.N. ((Generalized modular representation framework for
process synthesis)). En: AIChE Journal 42.4 (1996), pags.
1010-1032.

98

Captulo 3. Introduccion

[65] Smith, E.M.B. y Pantelides, C.C. ((Design of reaction/separation networks using detailed
models)). En: Computers & Chemical Engineering 19, Supplement 1.0 (1995), pags.
83 -88.
[66] Grossmann, I.E., Aguirre, P.A. y Barttfeld, M. ((Optimal synthesis of complex distillation co-
lumns using rigorous models)). En: Comput. Chem. Eng. 29.6 (2005), pags. 1203 -1215.
[67] Bedenik, N.I., Pahor, B. y Kravanja, Z. ((An integrated strategy for the hierarchical multi-
level MINLP synthesis of overall process flowsheets using the combined synthesis/analysis
approach)). En: Computers & Chemical Engineering 28.5 (2004), pags. 693 -706.
[68] Ponce-Ortega, J.M. y col. ((A property-based approach to the synthesis of material con-
servation networks with economic and environmental objectives)). En: AIChE Journal 57.9

(2011), pags. 2369-2387.
[69] Daichendt, M.M. y Grossmann, I.E. ((Integration of hierarchical decomposition and mathe-
matical programming for the synthesis of process flowsheets)). En: Computers & Chemical

Engineering 22.1-2 (1998), pags. 147 -175.
[70] Jaksland, C.A., Gani, R. y Lien, K.M. ((Separation process design and synthesis based on
thermodynamic insights)). En: Chemical Engineering Science 50.3 (1995), pags.
511 -530.
[71] Hostrup, M. y col. ((Integration of thermodynamic insights and MINLP optimization for the
synthesis, design and analysis of process flowsheets)). En: Computers & Chemical Enginee-
ring 23, Supplement.0 (1999), S23 -S26.
[72] Kravanja, Z. y Grossmann, I.E. ((Multilevel-hierarchical MINLP synthesis of process flowsheets)).
En: Computers & Chemical Engineering 21, Supplement.0 (1997), S421 -S426.
[73] Verheyen, W. y Zhang, N. ((Design of flexible heat exchanger network for multi-period ope-
ration)). En: Chemical Engineering Science 61.23 (2006), pags.
7730 -7753.
[74] Tay, D.H.S. y col. ((Synthesis of an integrated biorefinery via the CHO ternary diagram)).
En: Clean Technologies and Environmental Policy 13.4 (2011), pags. 567-579.
[75] Tay, D.H.S. y col. ((Fuzzy Optimization Approach for the Synthesis of a Sustainable Integrated
Biorefinery)). En: Industrial & Engineering Chemistry Research 50.3 (2011), pags.
1652-1665.
[76] El-Halwagi, M.M. y Manousiouthakis, V. ((Simultaneous synthesis of mass-exchange and re-
generation networks)). En: AIChE Journal 36.8 (1990), pags.
1209-1219.
[77] Foo, D.C.Y. ((Stateofthe-Art Review of Pinch Analysis Techniques for Water Network Synt-
hesis)). En: Industrial & Engineering Chemistry Research 48.11 (2009), pags.
5125-5159.
[78] Ahmetovic, E. y Grossmann, I.E. ((Global superstructure optimization for the design of inte-
grated process water networks)). En: AIChE Journal 57.2 (2011), pags.
434-457.
[79] Ahmetovic, E., Martn, M. y Grossmann, I.E. ((Optimization of Energy and Water Consum-
ption in Corn-Based Ethanol Plants)). En: Industrial & Engineering Chemistry Research 49.17

(2010), pags. 7972-7982.
[80] Martn, M. y Grossmann, I.E. ((Process Optimization of FTDiesel Production from Lignocellu-
losic Switchgrass)). En: Ind. Eng. Chem. Res.r 50.23 (2011), pags.
13485-13499.
[81] Anderson, J.L. Process synthesis. Advances in Chemical Engineering Series. ACADEMIC
PressINC, 1998.
[82] Ciric, A.R. y Gu, D. ((Synthesis of nonequilibrium reactive distillation processes by MINLP
optimization)). En: AIChE Journal 40.9 (1994), pags.
1479-1487.
[83] Ropotar, M. y col. ((MINLP Synthesis of Reactive Distillation Using a Disjunctive, Hybrid Mo-
del)). En: 10th International Symposium on Process Systems Engineering: Part A. Ed. por Brito

99
3.4.REFERENCIAS

Alves, Caludio Augusto Oller do Nascimento Rita Maria de y Biscaia, Evaristo Chalbaud.

Vol. 27. Computer Aided Chemical Engineering. Elsevier, 2009, pags. 543 -548.
[84] Franke, M.B. y col. ((Design and optimization of a hybrid distillation/melt crystallization pro-
cess)). En: AIChE Journal 54.11 (2008), pags.
2925-2942.
[85] Kraemer, K., Kossack, S. y Marquardt, W. ((Efficient Optimization-Based Design of Distillation
Processes for Homogeneous Azeotropic Mixtures)). En: Industrial & Engineering Chemistry

Research 48.14 (2009), pags. 6749-6764.
[86] Giridhar, A. y Agrawal, R. ((Synthesis of distillation configurations: I. Characteristics of a good
search space)). En: Computers & Chemical Engineering 34.1 (2010), pags. 73 -83.
[87] Giridhar, A. y Agrawal, R. ((Synthesis of distillation configurations. II: A search formulation for
basic configurations)). En: Computers & Chemical Engineering 34.1 (2010), pags. 84 -95.
[88] Agrawal, R. ((Synthesis of Distillation Column Configurations for a Multicomponent Separa-
tion)). En: Industrial & Engineering Chemistry Research 35.4 (1996), pags.
1059-1071.
[89] Caballero, J.A. y Grossmann, I.E. ((Generalized Disjunctive Programming Model for the Op-
timal Synthesis of Thermally Linked Distillation Columns)). En: Industrial & Engineering Che-

mistry Research 40.10 (2001), pags. 2260-2274.
[90] Caballero, J.A. y Grossmann, I.E. ((Logicbased methods for generating and optimizing
thermally coupled distillation systems)). En: European Symposium on Computer Aided Pro-
cess Engineering-12 35th European Symposium of the Working Party on Computer Aided
Process Engineering. Ed. por Grievink, Johan y Schijndel, Jan van. Vol. 10. Computer Aided

Chemical Engineering. Elsevier, 2002, pags. 169 -174.
[91] Caballero, J.A. y Grossmann, I.E. ((Thermodynamically equivalent configurations for ther-
mally coupled distillation)). En: AIChE Journal 49.11 (2003), pags.
2864-2884.
[92] Caballero, J.A. y Grossmann, I.E. ((Design of distillation sequences: from conventional to
fully thermally coupled distillation systems)). En: Computers & Chemical Engineering 28.11

(2004), pags. 2307 -2329.
[93] Caballero, J.A. y Grossmann, I.E. ((Structural Considerations and Modeling in the Synthesis
of Heat-IntegratedaThermally Coupled Distillation Sequences)). En: Industrial & Enginee-

ring Chemistry Research 45.25 (2006), pags. 8454-8474.
[94] Agrawal, R. ((A Method to Draw Fully Thermally Coupled Distillation Column Configura-
tions for Multicomponent Distillation)). En: Chemical Engineering Research and Design 78.3

(2000), pags. 454 -464.
[95] Agrawal, R. ((Synthesis of multicomponent distillation column configurations)). En: AIChE

Journal 49.2 (2003), pags. 379-401.
[96]
Rong, B.G., Kraslawski, A. y Nystrom, L. ((The synthesis of thermally coupled distillation flows-
heets for separations of five-component mixtures)). En: Computers & Chemical Engineering

24.27 (2000), pags. 247 -252.
[97]
Rong, B.G., Kraslawski, A. y Nystrom, L. ((Design and synthesis of multicomponent thermally
coupled distillation flowsheets)). En: Comput. Chem. Eng. 25.4-6 (2001), pags.
807-820.
[98] Rong, B.G., Kraslawski, A. y Turunen, I. ((Synthesis of Functionally Distinct Thermally Coupled
Configurations for Quaternary Distillations)). En: Industrial & Engineering Chemistry Research

42.6 (2003), pags. 1204-1214.
[99] Lakshmanan, A. y Biegler, L.T. ((Synthesis of Optimal Chemical Reactor Networks)). En: In-

dustrial & Engineering Chemistry Research 35.4 (1996), pags. 1344-1353.

100

Captulo 3. Introduccion

[100] Kokossis, A.C. y Floudas, C.A. ((Stability in optimal design: Synthesis of complex reactor
networks)). En: AIChE Journal 40.5 (1994), pags.
849-861.
[101] Zhao, X. y Marquardt, W. ((Constructive nonlinear dynamics for reactor network synthe-
sis with guaranteed robust stability)). En: 21st European Symposium on Computer Aided
Process Engineering. Ed. por E.N. Pistikopoulos, M.C. Georgiadis y Kokossis, A.C. Vol. 29.
Computer Aided Chemical Engineering. Elsevier, 2011, pags. 387 -391.
[102] dAnterroches, L. y Gani, R. ((Group contribution based process flowsheet synthesis, design
and modelling)). En: Fluid Phase Equilibria 228-229.0 (2005), pags.
141 -146.
[103] Proios, P., Goula, N.F. y Pistikopoulos, E.N. ((Generalized modular framework for the synthesis
of heat integrated distillation column sequences)). En: Chemical Engineering Science 60.17

(2005), pags. 4678 -4701.
[104] Alvarado-Morales, M. y col. ((Biorefining: Computer aided tools for sustainable design and
analysis of bioethanol production)). En: Chemical Engineering Research and Design 87.9

(2009), pags. 1171 -1183.
[105] Alvarado-Morales, M. y col. ((A model-based methodology for simultaneous design and
control of a bioethanol production process)). En: Computers & Chemical Engineering 34.12

(2010), pags. 2043 -2061.
[106] Stankiewicz, A. y Moulijn, J.A. Re-Engineering the Chemical Processing Plant: Process Inten-
sification. Chemical Industries. Taylor & Francis, 2003.
[107] Becht, S. y col. ((An industrial view of process intensification)). En: Chemical Engineering

and Processing: Process Intensification 48.1 (2009), pags. 329 -332.
[108] Harmsen, G.J. ((Reactive distillation: The front-runner of industrial process intensification:
A full review of commercial applications, research, scale-up, design and operation)). En:

Chemical Engineering and Processing: Process Intensification 46.9 (2007), pags. 774 -780.
[109] Grossmann, I.E., Yeomans, H. y Kravanja, Z. ((A rigorous disjunctive optimization model for
simultaneous flowsheet optimization and heat integration)). En: Computers & Chemical
Engineering 22, Supplement 1.0 (1998), S157 -S164.
[110] Dantzig, G.B. y Thapa, M.N. Linear Programming 1: Introduction. Linear Programming. Sprin-
ger, 1997.
[111] Dantzig, G.B., Orden, A. y Wolfe, P. ((The generalized simplex method for minimizing a li-
near form under linear inequality restraints)). En: Pacific Journal of Mathematics 5.2 (1955),

pags. 183-195.
[112] Nash, J.C. ((The (Dantzig) simplex method for linear programming)). En: Computing in Scien-

ce Engineering 2.1 (2000), pags. 29-31.
[113] Karmarkar, N. ((A new polynomial-time algorithm for linear programming)). En: Proceedings
of the sixteenth annual ACM symposium on Theory of computing. STOC 84. New York, NY,

USA: ACM, 1984, pags. 302-311.
[114] Adler, I. y col. ((An implementation of Karmarkars algorithm for linear programming)). En-

glish. En: Mathematical Programming 44.1-3 (1989), pags. 297-335.
[115] Caballero, J.A., Odjo, A. y Grossmann, I.E. ((Flowsheet optimization with complex cost and
size functions using process simulators)). En: AIChE Journal 53.9 (2007), pags.
2351-2366.
[116] Hooke, R. y Jeeves, T.A. ((Direct Search Solution of Numerical and Statistical Problems)).

En: J. ACM 8.2 (abr. de 1961), pags. 212-229.

101
3.4.REFERENCIAS

[117] Nelder, J.A. y Mead, R. ((A Simplex Method for Function Minimization)). En: The Computer

Journal 7.4 (1965), pags. 308-313.
[118] Luus, R. y Jaakola, T.H.I. ((A direct approach to optimization of a complex system)). En:

AIChE Journal 19.3 (1973), pags. 645-646.
[119] Holland, J.H. Adaptation in natural and artificial systems: an introductory analysis with ap-
plications to biology, control, and artificial intelligence. University of Michigan Press, 1975.
[120] Laarhoven, P.J. van y Aarts, E.H. Simulated Annealing: Theory and Applications. Mathema-
tics and Its Applications. Springer, 1987.
[121] Biegler, L.T. y Grossmann, I.E. ((Retrospective on optimization)). En: Computers & Chemical

Engineering 28.8 (2004), pags. 1169 -1192.
[122] Binder, T. y col. ((Introduction to Model Based Optimization of Chemical Processes on Mo-
ving Horizons)). English. En: Online Optimization of Large Scale Systems. Springer Berlin Hei-
delberg, 2001, pags. 295-339.
[123] Han, S.P. ((Superlinearly convergent variable metric algorithms for general nonlinear pro-
gramming problems)). English. En: Mathematical Programming 11.1 (1976), pags. 263-282.
[124] Powell, M.J.D. ((A fast algorithm for nonlinearly constrained optimization calculations)). En:
Numerical Analysis. Ed. por Watson, G.A. Vol. 630. Lecture Notes in Mathematics. Springer

Berlin Heidelberg, 1978, pags. 144-157.
[125] Frank, M. y Wolfe, P. ((An algorithm for quadratic programming)). En: Naval Research Logis-

tics Quarterly 3.1-2 (1956), pags. 95-110.
[126] Biegler, L.T. Nonlinear Programming: Concepts, Algorithms, and Applications to Chemical
Processes. SIAM e-books. Society for Industrial y Applied Mathematics (SIAM, 3600 Market
Street, Floor 6, Philadelphia, PA 19104), 2010.
[127] Floudas, C.A. Nonlinear and Mixed-Integer Optimization: Fundamentals and Applications.
Topics in Chemical Engineering. Oxford University Press, USA, 1995.
[128] Griva, I.N.S.S.A. Linear and Nonlinear Optimization, 2nd Edition. Society for Industrial y Ap-
plied Mathematics (SIAM, 3600 Market Street, Floor 6, Philadelphia, PA 19104), 2009.
[129] Dakin, R.J. ((A tree-search algorithm for mixed integer programming problems)). En: The

Computer Journal 8.3 (1965), pags. 250-255.
[130] Nemhauser, G.L. y Wolsey, L.A. Integer and combinatorial optimization. Wiley-Interscience
series in discrete mathematics and optimization. Wiley, 1988.
[131] Leyffer, S. ((Integrating SQP and Branch-and-Bound for Mixed Integer Nonlinear Program-
ming)). En: Computational Optimization and Applications 18 (3 2001), pags.
295-309.
[132] Gupta, O.K. y Ravindran, A. ((Branch and Bound Experiments in Convex Nonlinear Integer
Programming)). En: Management Science 31.12 (1985), pags.
1533-1546.
[133] Geoffrion, A.M. ((Generalized Benders decomposition)). En: Journal of Optimization Theory

and Applications 10 (4 1972), pags. 237-260.
[134] Fletcher, R. y Leyffer, S. ((Solving mixed integer nonlinear programs by outer approximation)).
En: Mathematical Programming 66.1-3 (1994), pags. 327-349.
[135] Quesada, I. y Grossmann, I.E. ((An LP/NLP based branch and bound algorithm for convex
MINLP optimization problems)). En: Computers & Chemical Engineering 16 (1992), pags.
937
-947.

102

Captulo 3. Introduccion

[136] Westerlund, T. y Pettersson, F. ((An extended cutting plane method for solving convex MINLP
problems)). En: Computers & Chemical Engineering 19, Supplement 1.0 (1995), pags. 131
-136.
[137] Westerlund, T. y col. ((An extended cutting plane method for a class of non-convex MINLP
problems)). En: Computers & Chemical Engineering 22.3 (1998), pags.
357 -365.
[138]
Westerlund, T. y Porn, R. ((Solving Pseudo-Convex Mixed Integer Optimization Problems by
Cutting Plane Techniques)). En: Optimization and Engineering 3.3 (2002), pags.
253-280.
[139] Balas, E. ((Disjunctive Programming and a Hierarchy of Relaxations for Discrete Optimization
Problems)). En: SIAM Journal on Algebraic Discrete Methods 6.3 (1985), pags.
466-486.
[140] Beaumont, N. ((An algorithm for disjunctive programs)). En: European Journal of Operatio-

nal Research 48.3 (1990), pags. 362 -371.
[141] Raman, R. y Grossmann, I.E. ((Symbolic integration of logic in mixed-integer linear program-
ming techniques for process synthesis)). En: Computers & Chemical Engineering 17.9 (1993),

pags. 909 -927.
[142] Raman, R. y Grossmann, I.E. ((Modelling and computational techniques for logic based
integer programming)). En: Computers & Chemical Engineering 18.7 (1994), pags.
563 -578.
[143] Turkay, M. y Grossmann, I.E. ((Logic-based MINLP algorithms for the optimal synthesis of
process networks)). En: Computers & Chemical Engineering 20.8 (1996), pags.
959 -978.
[144] Hooker, J.N. y Osorio, M.A. ((Mixed Logical Linear Programming)). En: Discrete Applied Mat-

hematics 9697 (1999), pags. 395-442.
[145] Hooker, J. Logic-Based Methods for Optimization: Combining Optimization and Constraint
Satisfaction. Wiley Series in Discrete Mathematics and Optimization. Wiley, 2000.
[146] Lee, S. y Grossmann, I.E. ((New algorithms for nonlinear generalized disjunctive program-
ming)). En: Computers & Chemical Engineering 24.910 (2000), pags.
2125-2141.
[147] Williams, H.P. Model Building in Mathematical Programming. Wiley, 2013.
[148] Biegler, L.T., Grossmann, I.E. y Westerberg, A.W. Systematic methods of chemical process
design. Prentice-Hall international series in the physical and chemical engineering scien-
ces. Prentice Hall PTR, 1997.
[149] Clocksin, W.F. y Mellish, C.S. Programming in Prolog. Springer Berlin Heidelberg, 2003.
[150] Raman, R. y Grossmann, I.E. ((Relation between MILP modelling and logical inference for
chemical process synthesis)). En: Computers & Chemical Engineering 15.2 (1991), pags.
73
-84.
[151] Grossmann, I.E. y Trespalacios, F. ((Systematic modeling of discrete-continuous optimization
models through generalized disjunctive programming)). En: AIChE Journal (2013), n/a-n/a.
[152] Sawaya, N. y Grossmann, I.E. ((Thesis: Reformulations, Relaxations and Cutting Planes for
Generalized Disjunctive Programming)). En: Carnegie Mellon University (2006).
[153] Grossmann, I.E. y Lee, S. ((Generalized Convex Disjunctive Programming: Nonlinear Con-
vex Hull Relaxation)). English. En: Computational Optimization and Applications 26.1 (2003),

pags. 83-100.
[154] Vecchietti, A., Lee, A. y Grossmann, I.E. ((Modeling of discrete/continuous optimization pro-
blems: characterization and formulation of disjunctions and their relaxations)). En: Compu-
ters & Chemical Engineering 27.3 (2003), pags. 433 -448.

103
3.4.REFERENCIAS

[155] Viswanathan, J. y Grossmann, I.E. ((A combined penalty function and outer-approximation
method for {MINLP} optimization)). En: Computers & Chemical Engineering 14.7 (1990),

pags. 769 -782.
[156] Brooke, A. y col. GAMS: A Users Guide. GAMS Development Corporation, 1998.
[157] Bisschop, J. y Roelofs, M. Aimms Language Reference. Lulu Enterprises Incorporated, 2006.
[158] Schweiger, C.A. y Floudas, C.A. ((Process Synthesis, Design, And Control: A Mixed-Integer
Optimal Control Framework)). En: Proceedings of DYCOPS-5 on Dynamics and Control of

Process Systems (1998), pags. 189-194.
[159] Bonami, P. y col. ((An algorithmic framework for convex mixed integer nonlinear programs)).

En: Discret. Optim. 5.2 (mayo de 2008), pags. 186-204.
[160] Abhishek, K., Leyffer, S. y Linderoth, J. ((FilMINT: An Outer Approximation-Based Solver for
Convex Mixed-Integer Nonlinear Programs)). En: INFORMS Journal on Computing 22.4 (2010),

pags. 555-567.
[161] Drud, A. ((CONOPT: A GRG code for large sparse dynamic nonlinear optimization pro-
blems)). En: Mathematical Programming 31.2 (1985), pags.
153-191.
[162] Gill, P., Murray, W. y Saunders, M. ((SNOPT: An SQP Algorithm for Large-Scale Constrained
Optimization)). En: SIAM Journal on Optimization 12.4 (2002), pags.
979-1006.
[163] Byrd, R., Hribar, M. y Nocedal, J. ((An Interior Point Algorithm for Large-Scale Nonlinear Pro-
gramming)). En: SIAM Journal on Optimization 9.4 (1999), pags. 877-900.
[164]
Wachter, A. y Biegler, L.T. ((On the implementation of an interior-point filter line-search algo-
rithm for large-scale nonlinear programming)). En: Math. Program. 106.1 (mayo de 2006),

pags. 25-57.
[165] Tawarmalani, M. y Sahinidis, N.V. Convexification and Global Optimization in Continuous
and Mixed-Integer Nonlinear Programming: Theory, Algorithms, Software, and Applica-
tions. Nonconvex Optimization and Its Applications. Springer, 2002.
[166] Sahinidis, N.V. ((BARON: A general purpose global optimization software package)). En:

Journal of Global Optimization 8.2 (1996), pags. 201-205.
[167] Inc., Lindo Systems. ((LINDOGlobal Solver)). En: (2010).
[168] Misener, R. y Floudas, C.A. ((GloMIQO: Global mixed-integer quadratic optimizer)). En: Jour-

nal of Global Optimization (mar. de 2012), pags. 1-48.
[169] Belotti, P. y col. ((Branching and bounds tighteningtechniques for non-convex MINLP)). En:

Optimization Methods Software 24.4-5 (ago. de 2009), pags. 597-634.

104
CAPITULO 4

Resultados

105
Captulo 4. Resultados

Captulo 4. Resultados
Los resultados de la presente tesis se recogen en este captulo. En total se presen-
internacional y dos ultimos
tan tres trabajos publicados en diversas revistas de difusion

trabajos pendientes de publicacion.

I: Strategies for the robust simulation of thermally coupled disti-


Publicacion
M.A., Javaloyes, J., Caballero, J.A., Gross-
llation sequences Navarro-Amoros,
mann, I.E. (2012) Computers & Chemical Engineering, 36, pp.149-159

II: A new technique for recovering energy in thermally coupled dis-


Publicacion
M.A., Ruiz-Femenia,
tillation using vapor recompression cycles. Navarro-Amoros,
R., Caballero, J.A., (2013) AIChE Journal, Article published online: Jun-2013

III: An alternative disjunctive optimization model for heat integra-


Publicacion
M.A., Caballero,
tion with variable temperatures. Autores: Navarro-Amoros,
J.A., Ruiz-Femenia, R., Grossmann, I.E. (2013) Computers & Chemical Enginee-
ring, 56, pp. 12-26

Anexo I: Integration of different models in the design of chemical processes:


en Applied
Application to the design of a Power Plant. Enviado y en revision
Energy.

Anexo II: Integration of modular process simulators under the Generalized Dis-
junctive Programming framework for the structural flowsheet optimization. En-
en Computers & Chemical Engineering.
viado y en revision

se presentan las publicaciones que forman esta tesis doctoral, ca-


A continuacion
en la que se destacan las
da una de ellas precedida por una breve introduccion
relevantes obtenidas.
conclusiones mas

107
Captulo 4. Resultados

4.1 I: Strategies for the robust simulation of ther-


Publicacion
mally coupled distillation sequences

energa consumen
Uno de los procesos dentro de la industria qumica que mas
Concretamente, el consumo energetico
es la destilacion.
de esta se cifra entorno al

3 % del consumo energetico
mundial [1]. Generalmente, los sistemas de destilacion
emplean columnas convencionales (un alimento, dos productos, caldera y conden-
sador. El principal problema radica en el hecho de que dichas configuraciones son

energeticamente
ineficientes, debido a la irreversibilidad termodinamica asociada a
como las zonas de cabeza
la mezcla de corrientes tanto en la zona de alimentacion
(condensador) y cola (caldera) de la columna [2]. Esta ineficiencia es intrnseca a
de mezclas que presenten componentes con volatilidades in-
cualquier separacion

termedias a la de los compuestos a separar. Estudios teoricos desarrollados por Petlyuk

et al. [2] mostraron que dicha ineficiencia puede ser mejorada mediante la sustitucion
de ciertos intercambiadores (calderas o condensadores) por una serie de corrientes

entre varias columnas del sistema(Acoplamiento Termico) (TCD) 1 . Diferentes inves-
tigadores [3-7] han mostrado que los sistemas TCD son capaces de conseguir ahorros
de energa de alrededor del 30 % comparados con los sistemas de columnas con-
Ademas,
vencionales de destilacion. Halvorsen y Skogestad [8-10] probaron que el

consumo energetico de una mezcla ideal de N compo-
mnimo para la separacion

nentes era obtenido mediante configuraciones con acoplamiento termico total.

com-
Desde un punto de vista estructural, las configuraciones TCD son mucho mas

plejas que las construidas unicamente con columnas convencionales. El estudio de
este tipo de configuraciones suele ser extraordinariamente complejo debido al gran

numero de
de recirculaciones existentes entre columnas. Generalmente, la resolucion
de numerosos calculos
este tipo de sistemas implica la realizacion
iterativos. Ademas
la mayora de los estudios sobre secuencias de columnas TCD han sido realizados uti-

lizando modelos aproximados, como es el metodo Fenske-Underwood-Gilliland [11].
Y aunque estos modelos son lo suficientemente precisos para identificar y compa-
1 Thermally Coupled Distillation
Acronimo del ingles

109
4.1.Strategies for the robust simulation of TCD

rar diferentes alternativas, se necesitan modelos para verificar la exactitud de estos


modelos aproximados y poder realizar dise detallados.
nos mas


Un herramienta basica en el estudio de este tipo de sistemas son los simulado-

res de procesos qumicos. Generalmente, la mayora incluyen modulos de columnas
incluso columnas tipo Petlyuk. Sin embargo, la simulacion
de destilacion, de sistemas
columnas es bastante complicado. La presencia del
TCD compuestos por dos o mas

gran numero de recirculaciones existentes son responsables de la gran dificultad de

convergencia de estos sistemas en un simulador, necesitandose, en la mayora de los

casos, buenos valores iniciales proximos final para converger el sistema.
a la solucion
la presencia de un gran numero
En conclusion, de recirculaciones ralentiza y dificulta
y el estudio de los sistemas TCD.
considerablemente la simulacion

El objetivo del presente trabajo es el desarrollo de una nueva estrategia que me-
de este tipo de sistemas de destilacion.
jore y facilite la simulacion La idea basica
es
la de evitar o sustituir la estructura cclica (corrientes recirculadas) en el diagrama de
acclica.
flujo de los sistemas TCD, por una configuracion

La idea parte de los estudios realizados por Carlberg y Westerberg [12, 13], quienes

probaron, utilizando modelos aproximados (Metodo de Underwood), que en un siste-
de una columna
ma TCD, las dos corrientes que conectan el sector de rectificacion

A con una columna B son equivalentes a una unica corriente de vapor sobrecalenta-
es que
do, cuyo flujo es el flujo neto entre ambas. El problema de esta configuracion
el grado de sobrecalentamiento y/o subenfriamiento pueden ser extremadamente
altos, llegando a presentar valores sin sentido fsico, y como consecuencia, esta es-
aportada en este
trategia no siempre se puede aplicar en un simulador. La solucion
de las corrientes sobrecalentada y subenfriada por una com-
trabajo es la sustitucion
de una corriente de materia y una corriente de energa, evitandose
binacion los pro-

blemas de sobrecalentamiento y/o subenfriamiento y consiguiendo la configuracion
acclica deseada.

Para estudiar y validar la estrategia propuesta se estudiaron diferentes sistemas



con diferente numero de componentes. Los sistemas estudiados incluyen: mezclas de

hidrocarburos (C4s-C5s-C6s), aromaticos
(BTX), alcoholes, sistemas azeotropicos no
ideales (acetona, benceno, cloroformo) y sistemas que contienen 4 y 5 componen-

110
Captulo 4. Resultados

se estudiaron varias configuraciones termodinamicamente


tes. Ademas equivalentes
de diferentes columnas. Todos los ejemplos han sido realizados utilizando como si-
mulador, Aspen-Hysys. Finalmente, para comprobar el correcto funcionamiento de
sistema acclico respecto al sistema cclico, se comparan tanto los calores necesarios
en condensador y caldera como los flujos internos de las corrientes en cada una de
las columnas del sistema.

acclica) son muy proxi-


En todos los casos estudiados, los resultados (configuracion
rigurosa (configuracion
mos a los obtenidos al realizar la simulacion cclica) del siste-
ma, presentando errores medios por debajo del 2 % para mezclas de 3 componentes.

El error medio se incrementa ligeramente al incrementar el numero de componentes,
del error como consecuencia del aumento
debido principalmente a la propagacion

de los acoplamientos termicos en el sistema. Sin embargo, en todos los casos, esta
estrategia proporciona excelentes resultados comparados con los obtenidos con la
rigurosa. Por otro parte, esta estrategia proporciona excelentes puntos de
simulacion
rigurosa. En conclusion,
partida son obtenidos para la simulacion esta nueva tecni-

de mezclas de 3, 4 o 5 componentes mediante
ca permite estudiar la separacion
acopladas termicamente
secuencias de columnas de destilacion
de forma rapida y

sencilla comparados con los metodos
actuales de simulacion.

111
4.1.Strategies for the robust simulation of TCD

Strategies for the robust simulation of thermally coupled distillation


sequences

A. Caballero, Ignacio E. Grossmann


Miguel A. Navarro, Juan Javaloyes, Jose

Computers & Chemical Engineering, 36 (2012), pp.149-159

Abstract

This paper presents a novel strategy for the simulation of thermally coupled dis-
tillation sequences using process simulators. First, we show that the two side stream
connections involved in a thermal couple can be accurately substituted by a combi-
nation of a material stream and heat flow; enabling a sequence of thermally coupled
distillation columns to be simulated without recycle streams, similar to conventional
simulations of zeotropic distillation sequences. In fact, using this method, a sequence
of thermally coupled distillation columns is not more difficult to converge than other
distillation systems without recycles. Furthermore, in most cases, this approach introdu-
ces negligible errors, and provides excellent starting points for rigorous simulations of
actual thermally coupled systems with recycle streams. Different examples, including
mixtures of hydrocarbons (C4sC5sC6s), aromatics (BTX), alcohols, non-ideal azeotro-
pic systems (acetone, benzene, chloroform) and systems involving 4 or 5 components
are presented. In addition, various thermodynamically equivalent configurations, co-
rresponding to different alternatives for implementing this approach, are discussed.

Keywords

Distillation, Simulation, Thermally coupled distillation

112
Captulo 4. Resultados

4.1.1. Introduction

Sustainable process development encourages the pursuit of designs that make


efficient use of energy. Distillation processes consume approximately 3 % of the total
global energy [1]. However, separation sequences using conventional columns (a sin-
gle feed, two product streams, condenser and reboiler) suffer from an inherent ineffi-
ciency, caused by the thermodynamic irreversibility associated with stream mixing at
the feed, top, and bottom of the column [2]. This inefficiency is intrinsic to any sepa-
ration that involves an intermediate boiling component, and can be generalized for
an N-component mixture. Theoretical studies developed by Petlyuk et al. [2] showed
that this inefficiency can be improved by removing some heat exchangers, and by in-
troducing thermal coupling between the columns. In fact, the fully thermally coupled
configuration for separating a three component mixture is named a Petlyuk configu-
ration, in honor of F. Petlyuk. However, since the advent of that pioneer work, interest
in thermally coupled distillation (TCD) declined until the end of the 1980s, when the
operation of a divided wall column (DWC) by BASF [14] renewed interest in thermally
coupled distillation. Moreover, several researchers [3-7, 15] have since shown that TCDs
are typically capable of achieving a 30 % reduction in energy use compared to con-
ventional systems. In addition, Halvorsen and Skogestad [8-10] proved that the mi-
nimum energy consumption for an ideal N component mixture is obtainable in fully
thermally coupled configurations.

However, from a structural point of view, thermally coupled configurations are much
more complex than sequences of conventional columns. In a seminal paper, Agrawal
[16] established the theoretical basis behind fully thermally coupled (FTC) systems. If
we take into account all possible alternatives, from conventional to FTC distillation sys-
tems, including all intermediate alternatives, the situation becomes even more com-
plex. A method for the systematic generation of all of these alternatives was proposed
by Caballero and Grossmann [17, 18], using logical relationships and mathematical
programming; while Agrawal [19, 20], and Rong et al. [21-23], among others, proposed
a method to accomplish the same task using alternative conceptual enumeration. In
addition, different researchers have proposed alternatives that were not previously
considered, such as: structures with a reduced number of column sections [24]; struc-

113
4.1.Strategies for the robust simulation of TCD

tures with duplicate key components [23]; structures with duplicate separation tasks
[25]; and modifications designed to enable better control, while maintaining nearly
the same level of performance over a large range of operational conditions [26, 27].
Other proposed modifications include a reduction in column interconnections, which
was hypothesized to provide better controllability [28]. However, later studies showed
that reducing the number of interconnections does not necessarily provide opera-
tional advantages over modified and simpler structural designs [29, 30]. In particular,

Segovia-Hernandez et al. [30] compared the control properties of six alternative ther-
mally coupled distillation schemes to the Petlyuk system; and found that reducing the
number of interconnections in the Petlyuk configuration does not necessarily improve
its controllability. The magnitude of the problem can be understood by considering
that there are more than 105 alternatives for separating a 5 component mixture into
its pure components [17, 31].

An important reason why TCD systems were forgotten, in spite of their clear ad-
vantages, is the lack of confidence in the controllability of these systems. However,
the controllability of TCD systems is similar to, and in some cases better than, systems
consisting only of conventional columns. The controllability of these systems has been
analyzed by many researchers [32-42].

Most of the studies performed on TCD sequences were conducted using shortcut
models, such as the FenskeUnderwoodGilliland method [11]. Although these ap-
proaches are accurate enough to identify and compare promising alternatives; a
more rigorous simulation is necessary to verify the accuracy of these shortcut models,
and enable more detailed designs.

Most chemical process simulators include side columns, or even Petlyuk-like confi-
gurations. However, simulation of thermally coupled systems involving more than two
columns (and in some cases even with two columns) is difficult, because the two si-
de flows connecting the columns produce systems with a large number of recycle
streams (in a modular simulator these recycles are converged through tear streams).
Whatever the method used to converge the cyclic structure of the flowsheet (e.g. fi-
xed point, Newton or quasi-Newton methods), good initial values approximating the
final solution are mandatory to converge the system, while maintaining product spe-

114
Captulo 4. Resultados

cifications. The presence of a large number of tear streams slows down the simulation,
making convergence difficult. For the remainder of the present study, we present a
very straightforward procedure that enables simulation of these systems in such a way
that they can be simulated with the same degree of difficulty as a sequence of con-
ventional columns.

It should be noted that, in this paper, we only deal with the problem of simulation,
and not with the design or optimization of a TCD system. However, it is not difficult
(although this is outside of the scope of this paper) to synthesize a TCD sequence
using the procedure presented by Caballero, Milan-Ya
nez, and Grossmann [43].

4.1.2. Application of the proposed simulation strategy: acyclic sys-


tem simulation

The basic idea behind the simulation strategy presented in the present study is to
avoid the recycle structure that appears in TCD systems in a modular simulator. This
idea is based on the works by Carlberg and Westerberg [12, 13], who proved, in the
context of Underwoods shortcut method, that in a TCD system, the two side streams
connecting the rectifying section of column 1 (see Fig. 4.1.1a) with column 2 are equi-
valent to a superheated vapor stream, whose flow is the net flow (i.e. the difference
between vapor exiting the column and liquid entering the column) (Fig. 4.1.1b). If the
two side streams connect the stripping section of the first column with the second co-
lumn, then they are equivalent to a single subcooled liquid stream, whose flow is the
net flow (in this case the liquid minus the vapor flows) (see Fig. 4.1.1c and d).

However, in general, this approach cannot be implemented in modular process


simulators, because the degree of superheating and/or subcooling can be so large
that it might produce results without physical meaning, and thus the simulator may fail
to converge.

A simple example will illustrate this last point. Assume that we have a mixture of ben-
zene, toluene and p-xylene, with a molar fraction composition of 0.3, 0.4 and 0.3. If we
want to use a Petlyuk configuration, which is thermodynamically equivalent to a divi-
ded wall column, we can simulate this system using two (or maybe three) conventional

115
4.1.Strategies for the robust simulation of TCD

Figura 4.1.1: (a, b and e) Equivalent configurations. (c, d and f) Equivalent configura-
tions

columns (see Fig. 4.1.2). A complete discussion of thermodynamically equivalent con-


figurations and their implications with respect to cost and operability for systems with
3 or more components can be found in Agrawal [26], Caballero and Grossmann [27]
and Agrawal and Fidkowski [44].

However, to illustrate this point, we will focus only on the first column. Assume also
that we have 20 theoretical trays, with the feed in tray 10 (a non-optimized column). In
order to avoid recycled streams, we follow the Carlberg and Westerberg strategy and
use a conventional column with a partial reboiler to obtain a saturated vapor distillate
stream. Using a feed flowrate of 100 kmol/h as a basis for calculation, and specifying
99.9 % benzene recovery in the top of the column and 99.9 % p-xylene recovery in the

116
Captulo 4. Resultados

Figura 4.1.2: Petlyuk configuration and the thermodynamically equivalent divided wall
column

bottom stream, we obtain by Aspen-Hysys(using a Peng Robinson equation of state,


a constant pressure of 1 atm and default parameters):

Distillate: Saturated vapor, 45.84 kmol/h; molar fractions of benzene, toluene and
xylene, 0.6538, 0.3456 and 0.0006, respectively; temperature 93.8C.

Bottoms: Saturated liquid, 54.16 kmol/h; molar fractions of benzene, toluene and
xylene, 0.0005, 0.4461 and 0.5534, respectively; temperature 123.6C.

The heat flow that must be removed in the condenser is 478.7 kW. And the heat
flow added in the reboiler is 918.5 kW.

With the above data and a simple energy balance, it is easy to calculate that, ac-
cording to the procedure suggested by Carlberg and Westerberg, the temperature
of the equivalent distillate stream should be 350C! Thus, Hysys fails to properly calcu-
late the temperature of the bottoms stream, because the energy balance predicts
temperatures below .273.15C (0 K!), which of course has no physical meaning.

Fortunately, it is possible to solve this problem by substituting the superheating or


subcooling streams with a combination of a material stream and an energy stream.
In the rectifying section, the material stream is vapor at its dew point and the energy
stream is equivalent to the energy removed if we include a partial condenser to pro-

117
4.1.Strategies for the robust simulation of TCD

vide reflux to the first column (see Fig. 4.1.1e). In the stripping section, the material
stream is liquid at its bubble point, and the energy stream is equivalent to the energy
added if we include a reboiler to provide vapor to the first column (see Fig. 4.1.1f).

The main source of error encountered when using the Carlberg and Westerberg
approximation is the assumption that the introduction of a liquid stream in a column
does not modify the flow of the vapor stream; and that the introduction of a vapor
stream does not modify the flow of the liquid stream. (These assumptions are discussed
in the original paper, and for further details the interested reader is referred to the
original work [12, 13]. Consider, for example, that in Fig. 4.1.3, the only possibility for
stream V1C1 to mix with stream V1C2 without modifying stream LC2
2 is if the net stream
V1C1 + V1C2 is in equilibrium with LC2
2 . Fortunately, this situation is generally found in any
thermally coupled system.

Figura 4.1.3: Details of the connection between columns, cyclic system simulation

The introduction of an energy stream and a saturated stream is completely equi-


valent to the Carlberg and Westerberg methodology, if the heat of vaporization is
independent of the composition, and if the heat of mixture can be neglected. Consi-
der, for example Fig. 4.1.4, where the introduction of an energy stream results in partial
vaporization of the liquid stream LC2
1 . Fortunately, all of the implied vapor and liquid
streams are either in equilibrium (an ideal case in which the Carlberg and Westerberg
approach does not introduce any error) or very close to equilibrium (generally found
in real situations) and therefore the dependence of the heat of vaporization on the

118
Captulo 4. Resultados

Figura 4.1.4: Details of the connection between columns, acyclic system simulation

composition and heat of mixture can be neglected. As will be demonstrated in the


remainder of the paper, even in the worst possible scenario, the values obtained with
this technique provide excellent initial points to enable convergence of the rigorous
simulations of the original system.

It is worth noting that some researchers have proposed the use of a single saturated
stream to simulate a thermal couple, rather than the two bidirectional streams (vapor
and liquid streams) [7] that characterize a thermal couple. However, this approach
introduces considerable errors and should be avoided. A detailed discussion can be

found in Vaca, Jimenez,
and Alvarez-Ram
rez [45].

It is important to remark that if good initial values are available for each of the
tear streams (vapor or liquid streams in the bidirectional streams that form a thermal
couple), for example, if those initial values come from any previous process synthesis
step, the procedure proposed here is not really necessary, and we can converge di-
rectly the actual system. Note, however, that if a change is introduced in the system
(i.e. changes in the feed composition, feed state, pressure, etc.) is not uncommon that
the system presents convergence difficulties. In this case the proposed approach is an
easy strategy to recover convergence.

119
4.1.Strategies for the robust simulation of TCD

4.1.3. Examples and results

In this section, a range of different examples are presented: including mixtures of


hydrocarbons (C4sC5sC6s), aromatics (BTX), alcohols, non-ideal azeotropic systems
(acetone, benzene, chloroform) and systems involving 4 or 5 components. In addition,
various thermodynamic equivalent configurations, corresponding to different alterna-
tives for implementing this approach are also discussed. All simulations were performed
using ASPEN-HYSYS. Reboiler and condenser duties, as well as vapor/liquid internal
flows were studied to compare the results between cyclic and acyclic simulations.

4.1.3.1. Separations of three component mixtures

Four systems involving the separation of a three component mixture were studied:
a mixture of aromatics (benzene, toluene, p-xylene); alcohols (methanol, ethanol, bu-
tanol); hydrocarbons (n-hexane, n-heptane, n-octane); and a difficult separation of a
mixture of three compounds with similar volatilities (i-butane, n-butane, cyclobutane).
As can be seen in the residue curve maps (see Fig. 4.1.5), in all cases, the separations
are theoretically easy to carry out.

Furthermore, the methodology is the same in all the cases. We will only discuss
the first case in detail, but the other three cases are solved in a similar manner. First,
we calculated the number of trays and the feed tray required in each column for a
desired separation. To do this, we use a shortcut model: either UnderwoodFenske for
near ideal systems; or simple trial and error for non-ideal systems. Note that we are
not interested in optimizing the column, but only in developing an easy and reliable
simulation (see Fig. 4.1.6 and Table 4.1.1).

Tabla 4.1.1: Characteristics of different feeds to each column

Molar Flow Composition


P(atm) T(C)
(kmol/h) Benzene Toluene pXylene
Feed 1.0 103.3 200.00 0.30 0.40 0.30
D1 1.0 95.3 97.51 0.61 0.38 6.2 104
R1 1.0 124.7 102.49 5.8 104 0.41 0.58

120
Captulo 4. Resultados

(a) (b)

(c) (d)

Figura 4.1.5: Residue curve maps of the four component systems studied: (a) i-butane,
n-butane, cyclobutane; (b) n-hexane, n-heptane, n-octane; (c) benzene, toluene, p-
xylene; (d) methanol, ethanol, butanol

Next, we simulated the acyclic sequence, where each thermal couple is substi-
tuted by a mass and energy stream, using conventional distillation columns (see Fig.
4.1.7a). In this configuration, we connect the mass and energy streams that leave the
condenser of the first column in the same tray of the second column. In the same way,
the mass and energy streams that leave the reboiler of the first column are connected
to the same tray of the second column. Finally, we converge the acyclic sequence.

121
4.1.Strategies for the robust simulation of TCD

Figura 4.1.6: Calculation of tray columns using short cut distillation

In this first example, we assume that the first column has 17 trays, with the feed in
tray 9. The distillate stream (saturated vapor) and the energy stream are introduced
in the second column in tray 14, which is simulated with 68 trays. The bottom stream
from the first column (saturated liquid) is introduced in tray 51. From this tray (tray 51),
an energy stream with exactly the same energy consumed in the reboiler of the first
column is withdrawn. Specifications used to converge the first column were 0.9999 re-
covery of benzene in the distillate and 0.9999 recovery of xylene in the bottoms; while
for the second column we specified a 0.999 recovery for each component. Toluene is
withdrawn from tray 31.

The results from this acyclic simulation were then used as initial points in the actual
system (with a cyclic structure). In particular, the initial flow rate of the liquid stream
entering the top of the first column (tear stream Liq to C1 in Fig. 4.1.7b) is assumed to
be the flow entering the first column from the partial condenser in Fig. 4.1.7a (136.022
kmol/h). The composition of this stream can then be read from the liquid composition
exiting tray 14 in column 2 of Fig. 4.1.7a. A similar procedure is followed to estimate the
flow and composition of the vapor stream Vap to C1 in Fig. 4.1.7b.

122
Captulo 4. Resultados

(a)

(b)

Figura 4.1.7: Simulations of (a) acyclic and (b) cyclic system configurations

A list of the results obtained using the above, and the characteristics of the different
columns are shown in Table 4.1.2.

It is important to note that the distillate of the first column is equivalent to a satu-
rated vapor stream plus an energy stream; hence, we are adding heat to the upper
part of the second column. However, the bottoms stream is equivalent to a saturated
liquid stream minus a heat stream, and therefore, we are actually removing heat from
the second column. In addition, the sign of the energy stream added must be taken

123
4.1.Strategies for the robust simulation of TCD

Tabla 4.1.2: Characteristics of short cut columns and conventional columns

Condenser Reboiler Energy


N Trays Feed Tray
Energy (kWh) (kWh)
Short cut columns
SC1 17 9 1777.9 2710.3
SC2 31 14 5685.0 4830.2
SC3 37 20 4561.6 4583.7
Conventional Columns - acyclic system
C1 17 9 - -
Tray D1-14
C2 68 3214.7 3315.3
Tray R151
Conventional Columns - cyclic system
C1 17 9 - -
Tray D114
C2 68 3220.7 3319.6
Tray R151

into account, since a negative sign in Hysys means that we are removing heat, which
is in fact what we want in the lower part of the second column. This is achieved in
Hysys by means of a set operation that simply multiplies the reboiler heat load in the
first column by minus one.

The results obtained were very good in all cases: average errors in internal flows
and energy consumption were lower than 1.5 % and 1 %, respectively; and in both
simulations the maximum errors were between 3 % and 5 %. The results obtained in
these systems are shown in Table 4.1.3.

Tabla 4.1.3: Results obtained with systems 3 compounds

Internal flow Energy


3 Component mixtures
Max. error Average error St. deviation Max. error
i-Butane - n-butane - cyclobutane 3.42 % 0.74 % 0.87 % 0.04 %
Hexane heptane octane 3.47 % 0.96 % 0.96 % 0.08 %
Methanol ethanol butanol 3.94 % 0.86 % 0.92 % 0.08 %
Benzene toluene xylene 4.55 % 1.46 % 1.40 % 0.19 %

As a representative example, internal flows for the benzenetoluenexylene 3-component


separation are shown in Fig. 4.1.8.

124
Captulo 4. Resultados

(a) (b)

(c) (d)

Figura 4.1.8: Comparison of flows for BTX separation: (a) liquid flow-column 1; (b) vapor
flow-column 1; (c) liquid flow-column 2; (d) vapor flow-column 2

4.1.3.2. Difficult separations of three component mixtures

Here we refer to difficult separations not in the usual sense (i.e. mixtures of com-
ponents with very similar volatilities), but rather systems that present other theoretical
difficulties, such as azeotropic systems (benzene acetone chloroform), multicom-
ponent mixtures that must be separated in groups (i.e. groups of C3s, C4s and C5s), or
non-ideal mixtures (acetone acetic acid acetic anhydride).

To perform the separation of an azeotropic mixture, we first studied its residue curve
map to determine a feasible separation route. In the following example (benzene
acetone chloroform), there is a distillation boundary. A possible sequence to obtain
the three pure components is shown in Figs. 4.1.9 and 4.1.10.

125
4.1.Strategies for the robust simulation of TCD

Figura 4.1.9: Residue curve map (azeotropic mixture)

Some remarks on this last example. The separation benzeneacetonechloroform


can be performed using a system involving only two columns. However, this example
has been selected to study the error propagation in a very non-ideal system, and the-
refore the three column sequence is presented here. On other side, it is known that
Benzene cannot be used due to environmental concerns (carcinogenic and tumori-
genic), and it can be substituted by other solvents like 1-hexanal or amyl methyl ether
[46]. We have selected this example because is a well documented and easy to re-
produce case of study; see for example the text book by Doherty and Malone [47].

The simulations performed here are different from those presented in the previous
sections, because of the requirements of using three columns to obtain pure compo-
nents, and the need to recycle the azeotropic stream and mix it with the feed stream.

However, this simulation has the same level of difficulty as the previous examples;
because, although the number of columns is increased, the number of recycle streams
remains constant. Therefore, the complexity of the problem is not increased and con-
sequently the error obtained is similar to that calculated previously.

126
Captulo 4. Resultados

(a)

(b)

Figura 4.1.10: Simulations of acyclic (a) and cyclic (b) system configurations (azeotro-
pic mixture)

The multi-component separation of groups of C3s, C4s and C5s can be compared
with a simple separation of a mixture of 3 components (as in the previous examples).
As such, the methodology is the same as that used previously. However, the error ob-
tained is slightly larger than the error associated with a 3-component mixture because
increasing the number of components increases the complexity of the separation.

Note that the case of a 3-component mixture acetone acetic acid acetic an-
hydride (AAA) is special, as can be seen from its residue curve map (Fig. 4.1.11).

In particular, all residue curves are partly over the acetic acidacetic anhydride
binary mixture, which is problematic because any small change in the concentration
of acetone in the residue causes significant changes in the concentrations of acid

127
4.1.Strategies for the robust simulation of TCD

Figura 4.1.11: Curve residue map acetoneacetic acidacetic anhydride (AAA)

and anhydride. As a consequence, the small error introduced by the substitution of


the two streams in a thermal couple with a material stream and an energy stream,
can induce large errors compared to rigorous and approximate simulations.

As expected, the results obtained were worse than those from the previously des-
cribed examples. However, in all the cases, the outcomes obtained using our novel
strategy still provided excellent initial points for use in rigorous simulations. Furthermore,
the average errors in internal flow and energy consumption were less than 4 % and 1 %,
respectively; and the maximum error between both simulations was between 4 % and
18 %. The results obtained in these systems are shown in Table 4.1.4.

Tabla 4.1.4: Results of difficult separation systems with 3 compounds


Internal Flow Energy
3 Compounds
Max. Error Average Error St. deviation Max. Error
Azeotropic Distillation 4.04 % 1.09 % 0.98 % 0.36 %
C4sC5sC6s 6.72 % 1.43 % 1.72 % 0.05 %
Acetone-acetic acid-acetic anhydride 18.40 % 3.91 % 4.61 % 0.91 %

128
Captulo 4. Resultados

The internal flows for the worst case studied (acetoneacetic acidacetic anhydri-
de) are shown in Fig. 4.1.12.

(a) (b)

(c) (d)

Figura 4.1.12: Comparison of flows for AAA separation: (a) liquid flow-column 1; (b)
vapor flow-column 1; (c) liquid flow-column 2; (d) vapor flow-column 2

129
4.1.Strategies for the robust simulation of TCD

4.1.3.3. Separation of 4 and 5 component mixtures

Application of our proposed strategy to the separation of 4 component mixtures


was also studied for butanepentanehexaneheptane, in a sequence with 16 ther-
modynamically equivalents configurations. In this case, 3 configurations were studied
using the same methodology (see Fig. 4.1.13).

(a) (b) (c)

Figura 4.1.13: Thermodynamically equivalents configurations-system 4 compounds: (a)


configuration 1; (b) configuration 2; (c) configuration 3

Finally, the separation of a 5 component mixture was evaluated. In this case, only
one configuration was studied using the same methodology described above (see
Fig. 4.1.14).

Although the results obtained were worse than any of the previously studied cases,
they still remained very good. As expected,when the number of columns and recy-
cle streams are increased, small errors propagate throughout the system. However, in
any case, the results obtained with our proposed strategy still provide excellent initial
points for use in rigorous simulations. The average errors in internal flow and energy
consumption were less than 5and 1 %, respectively, for the separation of a 4 compo-
nent mixture; and were less than 25 % and 1 %, respectively, for the separation of a 5
component mixture. The results obtained in these systems are shown in Table 4.1.5.

Internal flows for the worst case studied (separation of a 5 component mixture) are
depicted in Fig. 4.1.15.

130
Captulo 4. Resultados

Figura 4.1.14: Configuration separation system 5 compounds

Tabla 4.1.5: Results for the systems with 4 compounds

Internal Flow Energy

Max. Error Average Error St. deviation Max. Error


4 compounds (butane-pentanehexaneheptane)
Configuration 1 21.16 % 4.88 % 6.00 % 0.41 %
Configuration 2 22.42 % 5.03 % 6.16 % 0.12 %
Configuration 3 22.42 % 4.86 % 6.07 % 0.41 %
5 compounds (butane-pentanehexaneheptaneoctane)
Configuration 1 79.27 % 23.61 % 31.46 % 0.19 %

4.1.4. Conclusions

We evaluated the application of a novel strategy for the simulation of thermally


coupled distillation sequences using process simulators. Several case studies were pre-
sented to demonstrate that the results obtained with this acyclic sequence technique
are very close to those obtained with recycle calculations, with average errors below
2 % for 3 component mixtures. The average error increases slightly with the number
of components, due to error propagation as a consequence of the larger number
of thermally coupled columns in the system. However, in all cases, acyclic simula-
tion produces excellent results, which are comparable to those of the actual system.

131
4.1.Strategies for the robust simulation of TCD

(a) (b)

(c) (d)

(e) (f)

(g) (h)

Figura 4.1.15: Comparison of flows for 5 compounds separation: (a) liquid flow-column
1; (b) vapor flow-column 1; (c) liquid flow-column 2; (d) vapor flow-column 2; (e) liquid
flow-column 3; (f) vapor flow-column 3; (g) liquid flow-column 4; (h) vapor flow-column
4

132
Captulo 4. Resultados

Furthermore, this new strategy yields very good starting points to converge rigorous
simulations of these systems. In conclusion, our proposed technique enables thermally
coupled distillation systems for the separation of 3, 4 or 5 component mixtures to be
studied rapidly and easily.

Acknowledgements

The authors gratefully acknowledge financial support from the Spanish Ministerio

de Ciencia e Innovacion under project CTQ2009-14420-C02-01.

Referencias

[1] Soave, G. y Feliu, J.A. ((Saving energy in distillation towers by feed splitting)). En: Applied

Thermal Engineering 22.8 (2002), pags. 889 -896.
[2] Petlyuk, F.B., Platonov, V.M. y Slavinsk, D. ((Thermodynamically optimal method for separa-
ting multicomponent mixtures)). En: International Chemical Engineering 5.3 (1965), pag.
555.
[3] Fidkowski, Z.T. ((Distillation configurations and their energy requirements)). En: AIChE Journal

52.6 (2006), pags. 2098-2106.
[4] Fidkowski, Z.T. y Agrawal, R. ((Multicomponent thermally coupled systems of distillation co-
lumns at minimum reflux)). En: AIChE Journal 47.12 (2001), pags.
2713-2724.
[5] Fidkowski, Z.T. y Krolikowski, L. ((Minimum energy requirements of thermally coupled distilla-
tion systems)). En: AIChE Journal 33.4 (1987), pags.
643-653.
[6] Shah, P.B. y Kokossis, A.C. ((New synthesis framework for the optimization of complex distilla-
tion systems)). En: AIChE Journal 48.3 (2002), pags.
527-550.
[7] Triantafyllou, C. y Smith, R. ((Design and optimisation of fully thermally coupled distillation
columns)). En: Chemical Engineering Research and Design 70.A2 (1992), pags. 118-131.
[8] Halvorsen, I.J. y Skogestad, S. ((Minimum Energy Consumption in Multicomponent Distilla-
tion. 1. Vmin Diagram for a Two-Product Column)). En: Industrial & Engineering Chemistry

Research 42.3 (2003), pags. 596-604.
[9] Halvorsen, I.J. y Skogestad, S. ((Minimum Energy Consumption in Multicomponent Distilla-
tion. 2. Three-Product Petlyuk Arrangements)). En: Industrial & Engineering Chemistry Re-

search 42.3 (2003), pags. 605-615.
[10] Halvorsen, I.J. y Skogestad, S. ((Minimum Energy Consumption in Multicomponent Distilla-
tion. 3. More Than Three Products and Generalized Petlyuk Arrangements)). En: Industrial &

Engineering Chemistry Research 42.3 (2003), pags. 616-629.
[11] Underwood, A.J.V. ((Fractional Distillation of Multicomponent Mixtures)). En: Industrial & En-

gineering Chemistry 41.12 (1949), pags. 2844-2847.
[12] Carlberg, N.A. y Westerberg, A.W. ((Temperature-heat diagrams for complex columns. 2.
Underwoods method for side strippers and enrichers)). En: Industrial & Engineering Che-

mistry Research 28.9 (1989), pags. 1379-1386.

133
4.1.REFERENCIAS

[13] Carlberg, N.A. y Westerberg, A.W. ((Temperature-heat diagrams for complex columns. 3.
Underwoods method for the Petlyuk configuration)). En: Industrial & Engineering Chemistry

Research 28.9 (1989), pags. 1386-1397.
[14] Schultz, M.A. y col. ((Reduce costs with dividing wall columns)). En: Chemical Engineering

Progress 98.5 (2002), pags. 64-71.
[15] Ropotar, M. y col. ((MINLP Synthesis of Reactive Distillation Using a Disjunctive, Hybrid Mo-
del)). En: 10th International Symposium on Process Systems Engineering: Part A. Ed. por Brito
Alves, Caludio Augusto Oller do Nascimento Rita Maria de y Biscaia, Evaristo Chalbaud.
Vol. 27. Computer Aided Chemical Engineering. Elsevier, 2009, pags. 543 -548.
[16] Agrawal, R. ((Synthesis of Distillation Column Configurations for a Multicomponent Separa-
tion)). En: Industrial & Engineering Chemistry Research 35.4 (1996), pags.
1059-1071.
[17] Caballero, J.A. y Grossmann, I.E. ((Structural Considerations and Modeling in the Synthesis
of Heat-IntegratedaThermally Coupled Distillation Sequences)). En: Industrial & Enginee-

ring Chemistry Research 45.25 (2006), pags. 8454-8474.
[18] Caballero, J.A. y Grossmann, I.E. ((Design of distillation sequences: from conventional to
fully thermally coupled distillation systems)). En: Computers & Chemical Engineering 28.11

(2004), pags. 2307 -2329.
[19] Agrawal, R. ((Synthesis of multicomponent distillation column configurations)). En: AIChE

Journal 49.2 (2003), pags. 379-401.
[20] Agrawal, R. ((A Method to Draw Fully Thermally Coupled Distillation Column Configura-
tions for Multicomponent Distillation)). En: Chemical Engineering Research and Design 78.3

(2000), pags. 454 -464.
[21]
Rong, B.G., Kraslawski, A. y Nystrom, L. ((Design and synthesis of multicomponent thermally
coupled distillation flowsheets)). En: Comput. Chem. Eng. 25.4-6 (2001), pags.
807-820.
[22]
Rong, B.G., Kraslawski, A. y Nystrom, L. ((The synthesis of thermally coupled distillation flows-
heets for separations of five-component mixtures)). En: Computers & Chemical Engineering

24.27 (2000), pags. 247 -252.
[23] Rong, B.G., Kraslawski, A. y Turunen, I. ((Synthesis of Functionally Distinct Thermally Coupled
Configurations for Quaternary Distillations)). En: Industrial & Engineering Chemistry Research

42.6 (2003), pags. 1204-1214.
[24] Kaibel, G. ((Distillation columns with vertical partitions)). En: Chemical Engineering & Tech-
nology 10.1 (1987), pags. 92-98.
[25] Grossmann, I.E., Aguirre, P.A. y Barttfeld, M. ((Optimal synthesis of complex distillation co-
lumns using rigorous models)). En: Comput. Chem. Eng. 29.6 (2005), pags. 1203 -1215.
[26] Agrawal, R. ((More Operable Fully Thermally Coupled Distillation Column Configurations for
Multicomponent Distillation)). En: Chemical Engineering Research and Design 77.6 (1999),

pags. 543 -553.
[27] Caballero, J.A. y Grossmann, I.E. ((Thermodynamically equivalent configurations for ther-
mally coupled distillation)). En: AIChE Journal 49.11 (2003), pags.
2864-2884.
[28] Agrawal, R. ((Thermally coupled distillation with reduced number of intercolumn vapor
transfers)). En: AIChE Journal 46.11 (2000), pags.
2198-2210.
[29]
Jimenez, A. y col. ((Design and Energy Performance of Alternative Schemes to the Petlyuk
Distillation System)). En: Chemical Engineering Research and Design 81.5 (2003), pags.
518
-524.

134
Captulo 4. Resultados

[30]
Segovia-Hernandez,
J.G., Hernandez,
S. y Jimenez, A. ((Analysis of dynamic properties of
alternative sequences to the Petlyuk column)). En: Computers & Chemical Engineering

29.6 (2005), pags. 1389 -1399.
[31] Giridhar, A. y Agrawal, R. ((Synthesis of distillation configurations: I. Characteristics of a good
search space)). En: Computers & Chemical Engineering 34.1 (2010), pags. 73 -83.
[32] Alatiqi, I.M. y Luyben, W.L. ((Control of a complex sidestream column/stripper distillation
configuration)). En: Industrial & Engineering Chemistry Process Design and Development

25.3 (1986), pags. 762-767.
[33]
Jimenez, A. y col. ((Analysis of Control Properties of Conventional and Nonconventional
Distillation Sequences)). En: Ind. Eng. Chem. Res. 40.17 (2001), pags.
3757-3761.
[34] Luyben, W.L. Distillation Design and Control Using Aspen Simulation. Wiley, 2006.
[35]
Segovia-Hernandez, J.G. y col. ((Control of Thermally Coupled Distillation Arrangements
with Dynamic Estimation of Load Disturbances)). En: Industrial & Engineering Chemistry Re-

search 46.2 (2007), pags. 546-558.
[36]
Segovia-Hernandez,
J.G., Hernandez,
S. y Jimenez, A. ((Control Behaviour of Thermally Cou-
pled Distillation Sequences)). En: Chem. Eng. Res. Des. 80.7 (2002), pags.
783 -789.
[37]
Segovia-Hernandez, J.G. y col. ((A comparison of the feedback control behavior between
thermally coupled and conventional distillation schemes)). En: Computers & Chemical En-

gineering 28.5 (2004), pags. 811 -819.
[38]
Segovia-Hernandez, J.G. y col. ((Controllability Analysis of Thermally Coupled Distillation
Systems: Five-Component Mixtures)). En: Industrial & Engineering Chemistry Research 46.1

(2007), pags. 211-219.
[39] na, A. y Puigjaner, L. ((Controllability of Different Multicomponent Distillation
Serra, M., Espu
Arrangements)). En: Ind. Eng. Chem. Res. 42.8 (2003), pags. 1773-1782.
[40] na, A. y Puigjaner, L. ((Control and optimization of the divided wall column)).
Serra, M., Espu
En: Chemical Engineering and Processing: Process Intensification 38.4-6 (1999), pags. 549
-562.
[41] Serra, M. y col. ((Analysis of different control possibilities for the divided wall column: feed-
back diagonal and dynamic matrix control)). En: Computers & Chemical Engineering 25.4-

6 (2001), pags. 859 -866.
[42] Serra, M. y col. ((Study of the divided wall column controllability: influence of design and
operation)). En: Computers & Chemical Engineering 24.2-7 (2000), pags. 901 -907.
[43]
Caballero, J.A., Milan-Yanez, D. y Grossmann, I.E. ((Rigorous Design of Distillation Columns:
Integration of Disjunctive Programming and Process Simulators)). En: Industrial & Enginee-

ring Chemistry Research 44.17 (2005), pags. 6760-6775.
[44] Agrawal, R. y Fidkowski, Z.T. ((More operable arrangements of fully thermally coupled disti-
llation columns)). En: AIChE Journal 44.11 (1998), pags.
2565-2568.
[45]
Vaca, M., Jimenez-Guti
errez,
A. y Ramrez, J. Alvarez. ((A Note on the Controllability of Two
Short-Cut Designs for a Class of Thermally Coupled Distillation Sequence)). En: Industrial &

Engineering Chemistry Research 48.4 (2009), pags. 2283-2289.
[46] Hostrup, M., Harper, P.M. y Gani, R. ((Design of environmentally benign processes: integra-
tion of solvent design and separation process synthesis)). En: Computers & Chemical Engi-

neering 23.10 (1999), pags. 1395 -1414.

135
4.1.REFERENCIAS

[47] Doherty, M.F. y Malone, M.F. Conceptual design of distillation systems. McGraw-Hill chemi-
cal engineering series. McGraw-Hill, 2001.

136
Captulo 4. Resultados

4.2 II: A new technique for recovering energy in


Publicacion
thermally coupled distillation using vapor recompression
cycles


Como ya se ha comentado en el Articulo 1, uno de los procesos de separacion
ampliamente utilizados y que mas
mas energa consumen es la destilacion.
El consu-

mo energetico se cuantifica en aproxi-
relacionado con los procesos de destilacion
madamente el 60 % del total de energa consumida en la industria qumica y petro-
consuma tal cantidad de energa se debe a
qumica. El hecho de que la destilacion

que es un proceso altamente ineficiente desde un punto de vista energetico.

y su baja eficiencia, a lo largo de


Como resultado de la importancia de destilacion
los a
nos se han propuesto un gran numero de alternativas en un intento de aumentar
Algunas
su eficiencia y/o recuperar parte de la energa degradada por destilacion.

de las principales mejoras energeticas de calderas y
desarrolladas son la utilizacion
condensadores intermedios, las columnas acopladas con bombas de calor y siste-
de vapor, los metodos
mas de compresion las
de reflujo secundario y vaporizacion,
con integracion
columnas de destilacion de calor, las columnas de destilacion
con
de calor interna o la destilacion
integracion Acoplada Termicamente
(TCD).

El consumo mnimo de energa en una secuencia de columnas, sin incluir equipo


adicional (por ejemplo compresores) se obtiene siempre en secuencias con acopla-

miento termico total. Aunque debido a diversos motivos no se puede concluir que
las configuraciones obtenidas empleando columnas TCD mejoren siempre a las se-

cuencias de columnas convencionales o parcialmente acopladas termicamente. Los
principales motivos responsables de este hecho son:

En las columnas TCD, la energa es aportada y eliminada en las peores condi-



ciones, aportada a la temperatura maxima del sistema en la caldera y elimi-
nada a la temperatura mnima en el condensador.

asocia-
En los sistemas FTC, el flujo mnimo de vapor que limita el sistema esta
mas
do a la separacion difcil [1, 2]. El hecho que los acoplamientos termicos

137
4.2.New vapor recompression cycles in TCD

transfieran corrientes de vapor y lquido entre las columnas del sistema, y que
el flujo de vapor se vea limitado por ciertas separaciones, provoca que algu-

nas secciones de columna vean aumentado su diametro
respecto al diametro

optimo


En los sistemas FTC, el numero total de secciones de columnas es superior al ob-
tenido en sistemas con columnas convencionales. Diferentes autores han estu-

diado y discutido el tema relacionado con el numero de secciones necesarias
dada [3-7]
para llevar a cabo una separacion

de columnas de acoplamiento termico


La operacion compli-
es mucho mas

cada que con columnas convencionales debido al gran numero de interco-
nexiones entre las columnas

El objetivo del presente trabajo es el desarrollo de una nueva estrategia que con-
vierte la ineficiencia inherente de las secuencias TCD -exceso de flujo de vapor/lqui-
do en ciertas secciones como resultado de la transferencia llevada a cabo en los

acoplamientos termicos- en una ventaja, consiguiendo disminuir la energa y costes

del sistema. La idea basica consiste en extraer el exceso de vapor (lquido) de la sec-
de la columna que, como consecuencia de los acoplamientos termicos,
cion opera

en condiciones suboptimas. La corriente de vapor (lquido) extrada se utiliza en un
de vapor (VRC) o un ciclo de recompresion
ciclo de recompresion de vapor inverso
(RVRC), reduciendo de esta forma la cantidad de calor aportado al sistema exter-
la extraccion
namente. Ademas, de la corriente (lquida o vapor en exceso) restaura

las condiciones optimas de la columna afectada y
de funcionamiento de la seccion

consecuentemente reduce el diametro de la columna. Cabe comentar que debido
de los ciclos de compresion
a que la instalacion requiere de equipos bastantes caros,

tales como compresores, puede darse el caso de que el ahorro energetico consegui-
do en la planta no se vea compensado con el coste de los nuevos equipos, es decir,

se pueden dar casos donde el coste economico del ciclo de com-
de la instalacion
sea superior al ahorro energetico
presion conseguido. Por ello es necesario calcular
y cuantificar el coste adicional que tiene lugar como consecuencia de la compra e
de los equipos de que consta el ciclo de compresion.
instalacion

138
Captulo 4. Resultados

de ci-
La estrategia propuesta es aplicada a varios ejemplos de implementacion
clos VRC y RVRC en columnas TCD. Los sistemas estudiados son por un lado la separa-
de una mezcla de compuestos aromaticos
cion de una
y por otro lado la separacion
mezcla de hidrocarburos ligeros.


Los resultados obtenidos muestran ahorros economicos consecuentes de la apli-
de esta estrategia entorno un 20-40 %. Mediante esta propuesta, se demues-
cacion
tra que es posible convertir el problema de la ineficiencia inherente a las columnas
de la cual mejorar el funcionamiento de estos sistemas.
TCD en una ventaja, a traves
no se puede aplicar siempre en todos los sis-
Desgraciadamente, esta configuracion
Esta configuracion
temas de separacion. es potencialmente util
en los casos donde
de volatilidades de la mezcla a separar muestre diferencias importan-
la distribucion
tes entre un grupo de compuestos (con volatilidades similares entre ellos) respecto a
de un ciclo RVRC es uni-
otro. Finalmente, y como regla general, la implementacion
camente interesante en el caso de que el condensador trabaje a temperaturas muy
por debajo de la ambiente, donde el agua no pueda ser utilizado como servicio fro.

139
4.2.New vapor recompression cycles in TCD

A New Technique for Recovering Energy in Thermally Coupled


Distillation using Vapor Recompression Cycles

Ruben
Miguel A. Navarro-Amoros, Ruiz-Femenia, Jose
A. Caballero

DOI 10.1002/aic.14137

Published online in Wiley Online Library (wileyonlinelibrary.com)

Abstract

Even though it has been proved that a fully thermally coupled distillation (TCD)
system minimizes the energy used by a sequence of columns, it is well known that
vapor/liquid transfers between different sections produce an unavoidable excess of
vapor (liquid) in some of them, increasing both the investment and operating costs.
In this paper we propose to take advantage of this situation by extracting the ex-
tra vapor/liquid and subjecting it to a direct/reverse vapor compression cycle. This
new arrangement restores the optimal operating conditions of some of the affected
sections with energy savings of around 20-30 % compared with conventional TCD co-
lumns.

Various examples, including the direct and reverse vapor recompression cycles,
are presented. Furthermore, in each example, all possible modes of distillation (direct,
indirect and Petlyuk distillation) with and without vapor recompression cycles (VRC)
are compared to ensure that this approach delivers the best results.

Keywords

Distillation, Design (process simulation), Process Synthesis, Simulation, Process

140
Captulo 4. Resultados

4.2.1. Introduction

We live in a society where any activity, whether work or leisure, requires the con-
sumption of large amounts of energy. Present global energy consumption is estimated
to be 12,000 MToe (million tones oil equivalent), with a 50 % increase expected over
the next 30 years [8]. Energy consumption in the industrial sector represents approxi-
mately 28 % of global energy use. Within this sector, the chemical industry accounts
for approximately 20 % which represents about 5.6 % (0.90 TW/year) of the total energy
consumed in the world. In the chemical industry, separation processes are the most
energy intensive. Among the different separation techniques, distillation is the most
important and commonly used in the chemical and petrochemical industry. Distilla-
tion handles more than 90 % of separations [9] and this trend seems unlikely to change
in the near future. Mix et al. [10] calculated that distillation processes consumes about
60 % of the total energy in the chemical and petrochemical industry. In conclusion, it is
estimated that distillation processes accounts for about 3 % of global energy use [11,
12]. Therefore, any improvement in distillation efficiency will likely have an important
global effect on energy consumption.

In spite of its widespread use, distillation is energetically inefficient [13-15]. Heat


(used as separating agent) is conventionally supplied in the reboiler at the highest
process temperature, and removed in the condenser where the temperature is mini-
mum. Therefore, the heat recovered in the condenser cannot be reused for heating
other parts of the distillation unit.

As a consequence of the importance of distillation and its low efficiency, a lar-


ge number of alternatives have been proposed in an attempt to increase efficiency
and/or recover part of the energy degraded by distillation.

To improve the thermal efficiency of a distillation column a good number of in-


teresting alternatives have been proposed. Without embarking on a comprehensive
review, these include:

a Intercoolersinterheaters in order to reduce the reboiler or condenser heat loads


the case of condensers in the rectifying section are of special interest if refrigeration

141
4.2.New vapor recompression cycles in TCD

at sub-ambient temperature is needed in the condenser [16-20]. The objective is


enhanced reversible operation of the column

b Heat pumps and vapor recompression schemes [21-25]

c Secondary reflux and vaporization [26]

d Multiple-effect heat cascading for distillation columns [27-30]. Where the conden-
sing temperature is higher than the reboiling temperature, the condensing over-
head vapors of one distillation column can serve as the reboiler duty for another
column. This creates the equivalent of a multi-effect evaporator system, except that
distillation columns are used, rather than direct evaporation

e Heat integration [4, 31, 32]. The idea is to exchange heat between the condenser
and the reboiler of different distillation columns. The operating pressure of some
columns can be adjusted to obtain the adequate driving force

f Internally Heat Integrated Distillation Columns (HIDiC) [33-38]. Here rectifying and
stripping sections are separated. The rectifying section is compressed and conse-
quently temperature increases- and rectifying and stripping sections are totally or
partially heat integrated, again in an attempt to approximate reversible distillation
behavior

g Thermally coupled distillation (TCD)[3-6, 39-46]. Some heat exchangers (reboiler or


condensers) are removed and replaced by thermal couples. TCD produces a large
number of alternative distillation column sequences (much larger than classical co-
lumn sequencing which can be considered a special case of TCD). This introduces
a large number of new and interesting alternatives, but at the cost of increasing
the complexity of the problem

Excellent reviews of the various energy-efficient distillation techniques can be found


in Jana [47], Fonyo et al. [48] and Nakaiwa el al. [36].

Even though a Fully Thermally Coupled (FTC) configuration minimizes energy con-
sumption, it cannot be concluded that FTC configurations are always superior when

142
Captulo 4. Resultados

compared to sequences of simple columns or to partially thermally coupled configu-


rations (PTC). Instead, the optimum configuration will depend on the specific mixture
and feed conditions for at least the following reasons:

(1) The energy is supplied and removed under the worst conditions: supplied at the
highest temperature in the reboiler, and removed at the lowest temperature in the
condenser. In most cases, this precludes the use of lower cost utilities, i.e. medium
or low-pressure steam in the reboiler. This is especially true for the condenser if the
most volatile component must be condensed at sub-ambient temperature, which
rules out making use of cooling water

(2) In FTC systems the minimum vapor flow is that of the most difficult separation [1,
2]. The thermal couples transfer the vapor between columns and, therefore, some
column sections could have large diameters

(3) In FTC sequences the total number of column sections is larger than in the case
of sequences of simple columns. A detailed discussion on the number of column
sections needed for a given separation can be found in the following references
[3-7]

(4) Operation is also more difficult due to the large number of interconnections bet-
ween the columns

In this paper we propose to convert the inherent inefficiency of some TCD sequen-
ces the excess vapor/liquid flows in certain sections resulting from the vapor liquid
transfer by thermal couples in an advantage by further integrating the system, de-
creasing the energy and investment costs in the process. The basic idea involves with-
drawing the excess vapor (liquid) from the column section that, as a consequence of
thermal couples, operates sub-optimally. This restores the optimal operating conditions
of the affected column section and consequently reduces the column diameter and
utility consumption. The resulting vapor (liquid) stream is used in a Vapor Recompres-
sion Cycle (VRC) or a Reverse-Vapor recompression cycle (RVRC), reducing the utility
consumption at the cost of introducing a compressor.

143
4.2.New vapor recompression cycles in TCD

The rest of the paper is organized as follows. The source of inefficiency in TCD sys-
tems that motivated this work is explained in detail. At this point, the two configurations
based on the vapor recompression cycle are introduced. The capabilities of the ap-
proach proposed are then illustrated through two case studies, one of which corres-
ponds to an excess of vapor stream in the stripping section and the other to an excess
of liquid stream in the rectifying section. Finally, at the end of the paper, we discuss the
conclusions that can be drawn from this work.

4.2.2. Inefficiency in thermally coupled distillation systems

The literature documents various approaches to the design of TCD systems, ran-
ging from three component [49-52] to N component mixtures [4-7, 53, 54]. In order
to clearly illustrate the problem, for the sake of simplicity, but without loss of genera-
lity we will focus on a three component Petlyuk arrangement (See Figure 4.2.1). The
extension to more complex TCD arrangements is straightforward and based on the sa-
me principles. Consider a three component mixture (ABC) having no azeotropes and
sorted by decreasing relative volatilities (A is the most volatile and C the least). Inde-
pendently of design procedure, it is possible to identify three separation tasks for this
system: Separate AB from BC (AB/BC) where component B is optimally distributed bet-
ween condenser and reboiler. Separate A from B, and separate C from D (See Figure
4.2.1).

Assume that each one of the three separation tasks is designed and optimized
independently. In that case, the first separation task (AB/BC) is usually operated in
transitions split, also called preferred separation, which guarantees minimum vapor
flow inside the column. Then, columns for separations A/B and B/C can be designed.

To design the A/B and B/C separations it is important to notice that the connection
with the first column is through a pair of streams (a vapor and a liquid stream). A good
way of dealing with this involves following the procedure described by Carlberg &
Westerberg [55]. They showed, in the context of an ideal system, that the two streams
that connect the rectifying section of the first column (AB/BC) with the second column
(A/B) can be substituted by a single stream whose flow is the net flow (V1 -L1 referred to
Figure 4.2.2), and whose thermal state is superheated vapor. Similarly, the two streams

144
Captulo 4. Resultados

Figura 4.2.1: Construction of the Petlyuk c), and Divided Wall columns d) by decom-
position into basic tasks b) and further elimination of heat exchangers

that connect the stripping section of the first column with the third column (B/C) can
be substituted by a single stream whose flow is the net flow (L2 -V2 , referred to Figure
4.2.2), which is sub-cooled. Carlberg and Westerberg 56 also showed how to calculate
the degree of superheating or subcooling of these equivalent streams. Navarro et al
[56] showed that the degree of superheating or subcooling can be so high that this
approach cannot be directly implemented in a process simulator. However, they also
showed that a superheated (subcooled) stream is equivalent to a saturated stream
plus (minus) an energy stream (See Figure 4.2.2c). Following this approach they pro-
ved that it is possible to simulate complex non-ideal TCD systems without introducing
recycling (avoiding tear streams and complex iterations), and with errors in heat loads
and internal vapor flows that average 2-3 %, but rarely over 5 %. This is the approach
we will follow in this paper.

145
4.2.New vapor recompression cycles in TCD

Figura 4.2.2: Equivalent configurations for a thermal coupling

The simulation of a Petlyuk column entails that mass balances be satisfied in all
thermal couplings. While the couplings between columns 1 and 2 and between 1
and 3 are achieved by side stream extraction, the connection between columns 2
and 3 is made directly between them. As the system must exhibit the same behavior
before and after the coupling, both columns must operate at the same internal flows,
V2C2 = V1C3 . Unfortunately, this situation rarely occurs. On the contrary, the vapor flows
of the optimized columns are different, either V2C2 > V1C3 or V2C2 < V1C3 . It is necessary
to increase the internal flows either in column 2 or column 3 depending on which
column dominates (See Figure 4.2.1b). For example if V2C2 > V1C3 , then V1C3 must be
increased by V = V2C2 V1C3 to make both flows equal (LC3
1 must also be increased
by V ), which increases the diameter of column 3, the reboiler duty, and as a result its
fixed and variable costs. Similarly, if the dominant column the column with largest
internal flows- is column 3, the adjustment of flows in column 2 increases the condenser
duty and the diameter of column 2.

It is worth remarking that instead of following a sequential approach it is possible


to optimize the three columns simultaneously in order to mitigate the effect of flow
imbalances in the connection points (i.e. the distribution of flows through the thermal
couples could reduce excess duties if some of the separations took place under sub-

146
Captulo 4. Resultados

optimal conditions compared to the isolated separation tasks). In any event, except
in some special cases, at least one separation task always occurs under sub-optimal
conditions (larger internal flows and therefore larger duties) compared to operation in
isolation from the rest of the system.

There are at least two alternatives to try to improve the efficiency of the above
mentioned TCD systems:

a Introduce a heat exchanger at the connection point (See Figure 4.2.3) -a reboiler if
the dominant column is column 2, and a condenser if the dominant column is co-
lumn 3. In that way, all the column sections are operated in their optimal conditions.
An energy balance shows that duty of the new heat exchanger is comparable to
the extra duty in the original Petlyuk due to the overflows (equal if we assume cons-
tant enthalpies). There is an added advantage; the temperature at which the heat
must be supplied (removed) is between the temperature in the condenser and the
temperature in the reboiler. Therefore, in some situations we can use a cheaper uti-
lity (i.e. if we have to heat we will do it at lower temperature than in the reboiler; if
we have to cool we will do it at a higher temperature than in the condenser). Of
course, we also have to purchase the new heat exchanger. An economic analysis
is necessary to determine the best option

b The ideal situation would be to withdraw the extra vapor (liquid) and use it as a hot
(cold) stream elsewhere in the process and then return the condensed (vaporized)
stream to the column. Unfortunately this is not always possible

In this paper we propose to use this extra vapor in a Vapor Recompression Cycle
(VRC) or the extra liquid in a Reverse Vapor Recompression Cycle (RVRC).

4.2.3. Vapor recompression and reverse vapor recompression cy-


cles in thermally coupled distillation

In a Standard Vapor Recompression Cycle (SVRC) (See Figure 4.2.4) the vapor lea-
ving the top of the column is compressed isentropically, causing also an increase in

147
4.2.New vapor recompression cycles in TCD

(a) (b)

Figura 4.2.3: Petlyuk configuration with an intermediate reboiler (a) and an interme-
diate condenser (b)

temperature that provides the driving force for heat transfer to the bottom liquid. At
the same time it elevates the dew point of the overhead vapor, allowing its latent heat
to be used at a higher temperature. This overhead vapor stream is heat exchanged
with the liquid leaving the bottom of the column, which partially (or totally) vaporizes
the latter and, in turn, provides vapor load to the column and condenses by itself (to-
tally or partially). This condensed stream is then subcooled in the condenser to such
an extent that when its pressure is dropped back down to the column pressure, it does
not vaporize. Ideally, energy is only added in the compressor and removed in the con-
denser. The major drawback of SVRCs is that the difference in temperatures between
condenser and reboiler should not be too large (say a maximum of around 30C),
otherwise the compression ratio becomes too large multistage compression might
then become necessary- and compressors are some of the most expensive devices
in chemical processes. A similar situation could appear in HICiC [57]. If the temperatu-
re difference between the rectifying and stripping sections is large, then the pressure

148
Captulo 4. Resultados

increase of the rectifying section to heat integrate both sections must be also large.
Again the cost of compressor could dominate the system performance.

Figura 4.2.4: Scheme of a standard vapor recompression cycle

If in the Petlyuk configuration there is an excess of vapor in the rectifying section


of column 3 it is possible to extract this excess and isentropically compress it until the
temperature is high enough to permit heat exchange with the bottom stream. Some
saturated vapors condense on isentropic compression (a detailed explanation can
be found in the work by Felbab et al [22]), and in this case some preheating before
compression might be necessary. In contrast to SVRC, we cannot expect all the heat
duty in the reboiler to be supplied by the compressed vapor, and therefore the hot
utility cannot be removed, but merely reduced. The condensed vapor is subcooled if
needed and then expanded to the column pressure. A part of this stream is the inter-
mediate product (See Figure 4.2.5) and the rest is returned to the column to provide
the extra reflux needed by column 3 (the reflux to column 3 is provided by the liquid
entering from the stripping section of column 2 plus the returned liquid after the VR).

There are some aspects of the new VRC that deserve special attention:

149
4.2.New vapor recompression cycles in TCD

Figura 4.2.5: Scheme of a Vapor Recompression Cycle using the excess of Vapor in
rectifying section of column 3

1. The difference in temperature between the reboiler and the vapor withdrawn
from the column is lower than between condenser and reboiler; as a conse-
quence it is possible implement this new VRC in columns where SVRC is not
economically viable

2. There is a synergic effect between the TCD (Petlyuk column in the example) and
the VRC: TCD is returned to its optimal operating conditions (i.e. smaller column
diameters, reduced condenser duties) and energy is saved with the VRC. So we
have two sources of improvement working together: the improved efficiency of
the TCD and the energy savings of the VRC

3. The synergic effect commented above enhances the performance of the new
VRC scheme. Therefore, depending on the reduction in capital and operating
costs in the TCD scheme it might be of interest to implement VRC with larger
temperature differences. But of course this is case dependent and a detailed
economic analysis is necessary

150
Captulo 4. Resultados

4. There are two reasons why the VRC is expected to perform better as the mass
imbalance in the connection between sections becomes larger: 1) The extent
to which heat is exchanged in the reboiler increases with the vapor flow with-
drawn from the column and 2) the return of some column sections to their opti-
mal operating conditions have a larger impact on the total cost. It bears men-
tioning here that even though the Petlyuk configuration achieves the lowest
total energy consumption, if certain sections operate far from optimally, it is li-
kely that the Petlyuk configuration will not be the best from a total cost point of
view. Using a VRC avoids this inefficiency, but in general deciding which column
sequence is the best (with or without VRCs) requires a detailed study of all the
alternatives

5. The performance of the VRC depends on the relative utility costs of both the
heating in the reboiler and the power needed to drive the compressor and the
new equipment, but mainly the compressor. A detailed economic analysis is
always necessary

6. The direct VRC is not always possible. It requires that column 3 be dominant

If the dominant column is column 2, it is still possible to use a Reverse Vapor Recom-
pression Cycle (RVRC). In this case the excess saturated liquid is withdrawn as a side
stream from the connection point between columns 2 and 3 (See Figure 4.2.6). Part
of this liquid is the product with intermediate volatility (B); the rest is expanded, until its
temperature reaches a value low enough to be used as a cooling utility in the con-
denser. As in the case of VRC, we cannot expect all the cooling duty to be supplied
by the latent heat of this stream; therefore, we must resort to using some external co-
oling utility. After the stream is completely vaporized, it is introduced into a compressor,
where the pressure is isentropically increased until it is restored to the operating pres-
sure of the column. In this case, because the compressor efficiency is less than 100 %,
(typically around 75 %) overheating occurs in the outlet stream of the compressor and
thus guarantees the appearance of a vapor stream in the compressor output. Finally,
this vapor is introduced in the column at the same stage from which it was extracted,
and this way provides the required extra vapor flow that ensures correct behavior in
the rectifying section of the column.

151
4.2.New vapor recompression cycles in TCD

Figura 4.2.6: Scheme of a Reverse Vapor Recompression Cycle using the excess of
liquid in the stripping section of column 2

If the condenser is cooled by water or air (or any other cheap utility), it is obvious
that a RVRC is economically unfavorable even in the case of a significant reduction
in the heat utility and column diameter (an intermediate reboiler would likely produce
the same effect). However, if sub-ambient cooling is needed in the condenser (refri-
geration cycles are expensive) RVRC becomes an attractive alternative: in general,
the lower the condenser temperature, the better the expected performance of the
RVRC.

The VRC and RVRC can be implemented in any thermally coupled sequence in
which certain sections are not operating optimally. A detailed description of how to
design and optimize TCD sequences can be found, for example in the following re-

152
Captulo 4. Resultados

ferences [4-6, 58, 59]. From those models it is easy to identify the affected column
sections. Although the Petlyuk configuration is convenient for illustrating VRC cycles, it
is in general not put into practice in industry because of difficulty of operation. Espe-
cially challenging is the control of vapor flows between the columns. Vapor AB flows
from the top of the prefractionator (column 1) to the upper feed level in the product
column (column 2). This implies that the pressure in the top of the prefractionator is
greater than the pressure in the upper section of the product column. On the other
hand, vapor BC has to be transferred from the lower section of the product column
to the bottom of the prefractionator (column 3). Therefore, the pressure in the bottom
of the prefractionator has to be lower than the pressure in the lower section of the
product column [41]. The pressure in the prefractionator is neither uniformly higher nor
uniformly lower than the pressure in the product column. However, the Petlyuk arran-
gement has two thermodynamically equivalent configurations without that problem
[41, 60] and it is also equivalent to a Divided Wall Column (DWC) (See Figure 4.2.1). In
other words, the Petlyuk arrangement is convenient for simulation purposes although
the actual implementation could be any of the thermodynamically equivalent confi-
gurations, including the DWC. This can be extended to more complex TCD arrange-
ments (sequences involving N components, N3). Caballero & Grossmann [60]proved
that any sequence of separation tasks can always be arranged in a set of distillation
columns in which the vapor flows always from higher to lower pressures, although for
simulation purposes any of the thermodynamically equivalent configurations can be
used.

4.2.4. Examples and implementation details

In this section, two examples corresponding to the VRC and RVRC configurations
are shown. The installation of a VRC or RVRC involves the use of expensive equipment,
(i.e. compressors); and a detailed economic analysis must be performed to clearly
establish the advantages of a given design. Therefore, it is necessary to estimate the
additional cost due to the purchase and installation of this new equipment. The equip-
ment cost is calculated using correlations from the literature, and to this end we use
the correlations given by Turton et al. [61]. Finally, we update the prices to 2012 using

153
4.2.New vapor recompression cycles in TCD

the C
hemical Engineering Plant Cost Index(CEPCI). The annual cost of the equipment
is calculated for a time horizon (n = 10 years) and an annual interest rate (i) of 8 % [62]
using the following expression,

i(1 + i)n
Annualized capital cost = capital cost (4.2.1)
(1 + i)n 1

where:

i = fractional interest rate per year

n = number of years

All the simulations were done using the sequential modular simulator ASPEN-HYSYS
with the Soave-Redlich-Kwong (SRK) equation of state and default values. The hot and
cold utilities used are shown in Table 4.2.1.

Tabla 4.2.1: Specifications for the hot and cold utilities

Tin (C) Tout (C) Cost ($/GJ)*


Steam
Atmospheric Pressure (AP) (1 bar) 100 100 6,67
Low Pressure (LP) (6 bar) 160 160 7,78
Medium Pressure (MP) (11 bar) 184 184 8,22
High Pressure (HP) (42 bar) 254 254 9,83
Water 20 40 0,354
Refrigeration
Low Temperature (LT) -20 -20 7,89
Very Low Temperature (VLT) -50 -50 13,11
* All prices are referred to 2002

4.2.4.1. Example 1. VRC

The first case study deals with the separation of a mixture of aromatics (p-xylene,
cumene, 1,2,4-trimethylbenzene). We follow a sequential approach to optimize each
of the individual column sections. The main stream specifications are shown in Table
4.2.2.

154
Captulo 4. Resultados

Tabla 4.2.2: Streams specifications for the separation system

Molar Flow Composition


P(atm) T(C)
(kmol/h)
p-xylene cumene 1,2,4-trimethylbenzene
Feed ABC 1,00 153,7 200,00 0,3000 0,3000 0,4000
Product A 1,00 139,1 60,00 0,9998 0,0002 0,0000
Product B 1,00 153,7 60,03 0,001 0,9977 0,0013
Product C 1,00 169,4 79,97 0,000 0,0006 0,9994

First, a short-cut distillation model is used to find the number of equilibrium sta-
ges and the feed tray location of each column. Then we simulate and calculate the
energy consumption and costs associated with the separation using a conventional
Petlyuk column. The flowsheet for the simulated Petlyuk column is shown in Figure 4.2.7.

Figura 4.2.7: Flowsheet for the simulation of the Petlyuk configuration column

The next step is to study the effect of introducing the VRC into the above Petlyuk
column. With this aim, we simulate and calculate the energy consumption and cost
associated with this system. A minimum temperature difference of 15C is assumed
between the compressed vapor stream and the liquid stream entering the reboiler.
An adiabatic efficiency of 75 % is assumed in the compressor. The flowsheet for this
configuration is shown in Figure 4.2.8.

155
4.2.New vapor recompression cycles in TCD

Figura 4.2.8: Simulation of Petlyuk configuration column with VRC

The results obtained for both systems are shown in Tables 4.2.3 and 4.2.4.

Tabla 4.2.3: Conventional Petlyuk Distillation column: Capital & Energy Cost
EQUIPMENT
COLUMNS CONDENSER REBOILER
Column 1 Column 2
V(m3 ) 95,2 316,7 A (m2 ) 222,1 A (m2 ) 1273,1
Cost ( C) 359994 981392 Cost ( C) 128751 Cost ( C) 359280
Annual Cost Annual Cost Annual Cost
53650 146256 19188 53543
( C/year) ( C/year) ( C/year)

TOTAL ANNUAL COST ENERGIA


Equipment 272637 CONDENSER CALDERA
Energy 2175186 Energy (kw) 8772 Energy (kw) 8826
Total Cost 2447823 Cold Utility Water Hot Utility MP Steam
Energy Cost Energy Cost
89284 2085902
( C/year) ( C/year)

As expected, the introduction of the VRC in the conventional Petlyuk column gene-
rates significant energy savings. It is remarkable that the energy savings in the reboiler
are over 18 % (0.38 MM C/year). There is also a comparable reduction in the energy
consumed by the condenser. As we anticipated, the installation of the VRC increases
capital cost (equipment), exactly by 22 % (0.06 MM C/year), but the global energy

156
Captulo 4. Resultados

Tabla 4.2.4: Conventional Petlyuk Distillation with VRC: Capital and Energy Cost
EQUIPMENT
COLUMNS CONDENSER REBOILER
Column 1 Column 2
V(m3 ) 95,2 278.2 A (m2 ) 183,1 A (m2 ) 1042,5
Cost (C) 359994 1002380 Cost ( C) 127119 Cost ( C) 336620
Annual Cost Annual Cost Annual Cost
53650 149384 18945 50166
( C/year) ( C/year) ( C/year)
HEATER HEAT EXCHANGER COOLER
A (m2 ) 10,0 A (m2 ) 534,0 A (m2 ) 7,6
Cost (C) 15861 Cost ( C) 219988 Cost ( C) 15051
Annual Cost Annual Cost Annual Cost
2364 32785 2243
( C/year) ( C/year) ( C/year)
COMPRESSOR
Cost (C) 151432
Annual Cost
22568
( C/year)

ENERGY
COMPRESOR CONDENSER REBOILER
Energy (kW) 145 Energy (kW) 7231 Energy (kW) 7227
Utility Electricity Cold Utility Water Hot Utility MP Steam
Annual Cost Annual Cost Annual Cost
69368 73595 1708072
( C/year) ( C/year) ( C/year)
COOLER HEATER
TOTAL ANNUAL COST Energy (kw) 338 Energy (kw) 299
Equipment 332104 Cold Utility Water Hot Utility HP Steam
Energy 1938855 Energy Cost Energy Cost
3441 84378
Total Cost 2270959 ( C/year) ( C/year)

cost is reduced by 11 % (0.24 MM C/year). The capital investment of the equipment


is recovered in fewer than three years of operation. If we assume that annual capital
investment payments are spread equally over time, the annual cost decreases by 0.18
MM C per year.

The Petlyuk configuration is not the only sequence for separating a three com-
ponent mixture. There are two more distillation sequences, with or without thermal
couples, capable of separating a three component mixture into three relatively pure
products using simple columns: the direct and indirect sequence. A direct separation
sequence with a thermal couple is thermodynamically equivalent to a column with
a side stripper, while the indirect separation sequence with a thermal couple is equi-
valent to a column with a side enricher [63]. Furthermore, it is possible to use VRC in
any of these configurations (for this mixture, the VRC is only recommended for direct
distillation). To enable a comparison of results with those of the Petlyuk sequence with

157
4.2.New vapor recompression cycles in TCD

VRC, we evaluate the same separation task using the direct, indirect and direct with
VRC sequences (Figure 4.2.9).

Figura 4.2.9: Flowsheet for the direct sequence with and without VRC and for the indi-
rect sequence

The results obtained for each configuration are presented in detail in Appendix A.
The outcomes of the simulations are summarized in Figure 4.2.10. These results show
that the configuration achieving the lowest total annual cost is the Petlyuk Distillation
with VRC. It is noteworthy that the energy savings outweigh the additional cost asso-
ciated with the purchase and installation of VRC for the Petlyuk column and the direct
distillation sequence (Table A.3).

158
Captulo 4. Resultados

Figura 4.2.10: Capital cost of equipment and annualized energy and total costs for the
distillation systems studied in example 1

4.2.4.2. Example 2: RVRC

In this example we study the separation of three light hydrocarbons: ethylene, etha-
ne and propane. Again, we use the sequential optimization simulation methodology
for a Petlyuk/DWC. Table 4.2.5 shows the main specifications for the streams involved
in the simulation.

Tabla 4.2.5: Specifications for the streams in the separation system

Molar Flow Composition


P(atm) T(C)
(kmol/h)
Ethylene Ethane Propane
Feed ABC 1,00 1,5 2000,00 0,3000 0,3000 0,4000
Product A 1,00 -28,7 600,8 0,9977 0,0023 0,0000
Product B 1,00 -7,2 599,4 0,0009 0,9977 0,0014
Product C 1,00 57,1 799,8 0,0000 0,0007 0,9993

First, a conventional Petlyuk arrangement is simulated and optimized (Figure 4.2.7).


The energy consumption and costs associated with this separation are shown in Table
4.2.6. Then, we study the effect of introducing the RVRC using the flowsheet shown
in Figure 4.2.11. A summary of the results obtained are presented in Table 4.2.7. As in

159
4.2.New vapor recompression cycles in TCD

the previous example we assume a minimum approach temperature of 15C for heat
exchange and a 75 % adiabatic efficiency in the compressor.

Tabla 4.2.6: Conventional Petlyuk Distillation column: Capital and Energy Cost
EQUIPMENT
COLUMNS CONDENSER REBOILER
Column 1 Column 2
V(m3 ) 49,6 287,4 A (m2 ) 1804,0 A (m2 ) 817.1
Cost ( C) 521794 2941774 Cost ( C) 471174 Cost ( C) 269591
Annual Cost Annual Cost Annual Cost
77763 438411 70219 40177
( C/year) (C/year) (C/year)

TOTAL ANNUAL COST ENERGY


Equipment 626570 CONDENSER REBOILER
Energy 8430906 Energy (kw) 13940,5 Energy (kw) 16561,6
Total Cost 9057476 Cold Utility VLT Refrigerant Hot Utility AP Steam
Energy Cost Energy Cost
5254756 3176150
(C/year) (C/year)

Tabla 4.2.7: Conventional Petlyuk Distillation with RVRC: Capital and Energy Cost
EQUIPMENT
COLUMNS CONDENSER REBOILER
Column 1 Column 2
V(m3 ) 49,6 257,7 A (m2 ) 1168,5 A (m2 ) 512,7
Cost ( C) 521794 2354008 Cost ( C) 343114 Cost ( C) 202340
Annual Cost Annual Cost Annual Cost
77763 350817 51134 30155
( C/year) ( C/year) ( C/year)
COMPRESSOR HEAT EXCHANGER
A (m2 ) 6844,1
Cost ( C) 946032 Cost ( C) 1474823
Annual Cost Annual Cost
140987 219792
( C/year) ( C/year)

ENERGY
COMPRESSOR CONDENSER REBOILER
Energy (kW) 1365 Energy (kW) 9011 Energy (kW) 10422
Utility Electricity Cold Utility VLT Refrigerant Hot Utility AP Steam
Annual Cost Annual Cost Annual Cost
654055 3396472 1998676
( C/year) ( C/year) ( C/year)

TOTAL ANNUAL COST


Equipment 870647
Energy 6049203
Total Cost 6919850

The results lead to similar conclusions as in the first case study. First, the introduction
of the RVRC in the conventional Petlyuk column generates significant energy savings.
Indeed, the energy savings in the reboiler are greater than 35 % (1.18 MM C/year).
This reduction in energy consumption is also matched by the condenser. As expected,
the installation of the RVRC increases capital cost by 38 % (0.24 MM C/year), but the
global energy cost falls by 28 % (2.38 MM C/year). The capital investment could be

160
Captulo 4. Resultados

Figura 4.2.11: Flowsheet for the simulation of the Petlyuk column with RVRC

recovered in the first year of operation, and after subtracting the fraction of the total
depreciable capital of the equipment, the annual cost decreases by 2.14 MM C per
year.

Again, for this second case study, there are also a number of distillation sequen-
ces for separating this mixture. Furthermore, it is possible to use RVRC in any of these
configurations (for this mixture, the RVRC is only recommended for indirect distillation).
For comparison purposes, we simulate the same separation task with the direct and
indirect sequences (Figure 4.2.9a and 4.2.9b, respectively) and the indirect sequence
with RVRC (Figure 4.2.12).

The results obtained for each configuration are presented in detail in Appendix A.
The costs associated with each sequence are shown in Figure 4.2.13. The results show
that the Petlyuk with RVRC configuration achieves the lowest total annual cost. As in
the previous case study, the savings in energy outweigh the additional cost associa-
ted with the purchase and installation of RVRC in both the Petlyuk column as well as
indirect sequence. The lowest energy costs are again obtained by using the Petlyuk
column in combination with RVRC (Table A.6).

161
4.2.New vapor recompression cycles in TCD

Figura 4.2.12: Simulation of indirect sequence with RVRC

Figura 4.2.13: Annualized capital cost, energy costs and total costs for the distillation
systems studied in example 2

162
Captulo 4. Resultados

4.2.5. Conclusions

Thermally coupled distillation is an attractive alternative to conventional column


sequencing because the energy and under some circumstances (i.e. Divided Wall
Columns) the investment costs are considerably reduced. Typical values of around 10
to 50 % are frequently reported. However, even though FTC systems minimize energy
consumption, in general we cannot be certain that they minimize total annual cost for
two main reasons: The heat is added or removed under the worst conditions, at the
maximum and minimum temperatures in the system; certain sections must be opera-
ted far away from their optimal conditions, in order to satisfy the mass balances intro-
duced by thermal couples that increase some column section diameters and utilities
consumption.

On the other hand, the large degree of integration -in a fully thermally coupled
system it is possible to separate an N component mixture using only a condenser and
a reboiler- prevent the installation of standard vapor recompression cycles because
the temperature differences between condenser and reboiler are very large.

We have shown that it is possible to take advantage of the inherent inefficiency


of TCD systems, by withdrawing the excess vapor (liquid) inside certain column sec-
tions and using it in a Vapor Recompression Cycle (or a reverse vapor recompression
cycle). The benefit is twofold; on one hand, the optimal operating conditions of the
TCD system are recovered and therefore the diameter of some column sections and
the utilities consumption are reduced. On the other, the VRC or RVRC allow further
reductions in utilities consumption. In any case a detailed economic analysis is neces-
sary because vapor recompression cycles entail using expensive equipment. But the
examples presented in this paper show that economic savings can be very significant
(20-40 %).

It is clear that the new VRC cannot always be used. For example, consider the
case of a three component mixture ABC. The most profitable situation occurs when
the difference in volatilities between A and B are very small in comparison with the
difference between B and C. (The reverse is true in the case of RVRC). Indeed, this

163
4.2.REFERENCIAS

volatility distribution causes both vapor and liquid flows between the coupled sections
of the Petlyuk column to be very different.

As a general rule, the case of RVRC is only of interest if the condenser is working at
subambient temperature (i.e. water cannot be used as cooling utility). Cooling with
water is usually cheap and the extra cost of the compressor will likely make the RVRC
sub-optimal. However, refrigeration is expensive (usually much more so than heating),
and in that case the economic incentive of a RVRC is larger than VRC.

Acknowledgements

The authors would like to acknowledge financial support from the Spanish Ministerio
(PPQ, CTQ200914420-C02-02 and CTQ2012 37039-C02-02).
de Ciencias e Innovacion

Referencias

[1] Halvorsen, I.J. y Skogestad, S. ((Minimum Energy Consumption in Multicomponent Distilla-


tion. 2. Three-Product Petlyuk Arrangements)). En: Industrial & Engineering Chemistry Re-

search 42.3 (2003), pags. 605-615.
[2] Halvorsen, I.J. y Skogestad, S. ((Minimum Energy Consumption in Multicomponent Distilla-
tion. 3. More Than Three Products and Generalized Petlyuk Arrangements)). En: Industrial &

Engineering Chemistry Research 42.3 (2003), pags. 616-629.
[3] Agrawal, R. ((Synthesis of Distillation Column Configurations for a Multicomponent Separa-
tion)). En: Industrial & Engineering Chemistry Research 35.4 (1996), pags.
1059-1071.
[4] Caballero, J.A. y Grossmann, I.E. ((Structural Considerations and Modeling in the Synthesis
of Heat-IntegratedaThermally Coupled Distillation Sequences)). En: Industrial & Enginee-

ring Chemistry Research 45.25 (2006), pags. 8454-8474.
[5] Caballero, J.A. y Grossmann, I.E. ((Design of distillation sequences: from conventional to
fully thermally coupled distillation systems)). En: Computers & Chemical Engineering 28.11

(2004), pags. 2307 -2329.
[6] Caballero, J.A. y Grossmann, I.E. ((Generalized Disjunctive Programming Model for the Op-
timal Synthesis of Thermally Linked Distillation Columns)). En: Industrial & Engineering Che-

mistry Research 40.10 (2001), pags. 2260-2274.
[7] Caballero, J.A. y Grossmann, I.E. ((Synthesis of complex thermally coupled distillation sys-
tems including divided wall columns)). En: AIChE Journal 59.4 (2013), pags.
1139-1159.
[8] (U.S.), Energy Information Administration. International Energy Outlook 2011: With Projec-
tions to 2035. INTERNATIONAL ENERGY OUTLOOK. Bernan Assoc, 2011.
[9] Humprey, J. ((Separation processes: Playing a critical role)). En: Chemical Engineering Pro-

gress 91 (1995), pags. 31-41.
[10] MIx, T.J., Dueck, J.S. y Weinberg, M. ((Energy conservation in distillation)). En: Chemical En-

gineering Progress 74 (1978), pags. 49-55.

164
Captulo 4. Resultados

[11] Engelien, H.K. y Skogestad, S. ((Selecting appropriate control variables for a heat-integrated
sistillation system with prefractionator)). En: Computers & Chemical Engineering 28 (2004),

pags. 683-691.
[12] Humphrey, J.L. y Siebert, A.F. ((Separation technologies: an opportunity for energy savings)).

En: Chemical Engineering Progress 91 (1992), pags. 31-41.
[13] Flower, J.R. y Jackson, R. ((Energy Requirements in the Separation of Mixtures by Distillation)).
En: Trans. Instn Chem. Engrs. 42 (1964), T249 -T258.
[14] Freshwater, D.C. ((Thermal Economy in Distillation)). En: Trans. Instn Chem. Engrs. 29 (1951),

pags. 149 -160.
[15] King, C.J. Separation processes. McGraw-Hill chemical engineering series. McGraw-Hill,
1980.
[16] Agrawal, R. y Fidkowski, Z.T. ((On the Use of Intermediate Reboilers in the Rectifying Section
and Condensers in the Stripping Section of a Distillation Column)). En: Industrial & Enginee-

ring Chemistry Research 35.8 (1996), pags. 2801-2807.
[17] Agrawal, R. y Herron, D.M. ((Intermediate reboiler and condenser arrangement for binary
distillation columns)). En: AIChE Journal 44.6 (1998), pags.
1316-1324.
[18] Fonyo, Z. ((Thermodynamic analysis of rectification I. Reversible model of rectification)). En:

Int. Chem. Eng. 14 (1974), pags. 18-27.
[19] Le Goff, P., Cachot, T. y Rivero, R. ((Exergy analysis of distillation processes)). En: Chemical
Engineering & Technology 19.6 (1996), pags. 478-485.
[20] Terranova, B.E. y Westerberg, A.W. ((Temperature-heat diagrams for complex columns. 1.
Intercooled/interheated distillation columns)). En: Industrial & Engineering Chemistry Re-

search 28.9 (1989), pags. 1374-1379.
[21] Brousse, E., Claudel, B. y Jallut, C. ((Modelling and optimization of the steady state ope-
ration of a vapour recompression distillation column)). En: Chemical Engineering Science

40.11 (1985), pags. 2073 -2078.
[22] Felbab, N. y col. ((Vapor recompression for efficient distillation. 1. A new synthesis perspec-
tive on standard configurations)). En: AIChE Journal (2013), n/a-n/a.
[23] Ferre, J.A., Castells, F. y Flores, J. ((Optimization of a distillation column with a direct vapor
recompression heat pump)). En: Industrial & Engineering Chemistry Process Design and
Development 24.1 (1985), pags. 128-132.
[24] Fonyo, Z. y Benko, N. ((Comparison of Various Heat Pump Assisted Distillation Configura-
tions)). En: Chemical Engineering Research and Design 76.3 (1998), pags.
348 -360.
[25] Harwardt, A. y Marquardt, W. ((Heat-integrated distillation columns: Vapor recompression
or internal heat integration?)) En: AIChE Journal 58.12 (2012), pags.
3740-3750.
[26] Mah, R.S.H., Nicholas, J.J. y Wodnik, R.B. ((Distillation with secondary reflux and vaporization:
A comparative evaluation)). En: AIChE Journal 23.5 (1977), pags. 651-658.
[27] Agrawal, R. ((Thermally coupled distillation with reduced number of intercolumn vapor
transfers)). En: AIChE Journal 46.11 (2000), pags.
2198-2210.
[28] Andrecovich, M.J. y Westerberg, A.W. ((A Simple Synthesis Method Based on Utility Boun-
ding for Heat-Integrated Distillation Sequences)). En: AIChE J. 31.3 (1985), pags.
363-375.
[29] Engelien, H.K. y Skogestad, S. ((Minimum energy diagrams for multieffect distillation arran-
gements)). En: AIChE Journal 51.6 (2005), pags.
1714-1725.

165
4.2.REFERENCIAS

[30] Westerberg, A.W. y Andrecovich, M.J. ((Utility bounds for nonconstant QDT for heat integra-
ted distillation sequence synthesis)). En: AIChE Journal 31.9 (1985), pags.
1475-1479.
[31] Andrecovich, M.J. y Westerberg, A.W. ((An MILP formulation for heat-integrated distillation
sequence synthesis)). En: AIChE Journal 31.9 (1985), pags.
1461-1474.
[32] Floudas, C.A. y Paules IV, G.E. ((A mixed-integer nonlinear programming formulation for the
synthesis of heat-integrated distillation sequences)). En: Computers & Chemical Enginee-

ring 12.6 (1988), pags. 531 -546.
[33] Chen, H., Huang, K. y Wang, S. ((A novel simplified configuration for an ideal heat-integrated
distillation column (ideal HIDiC))). En: Separation and Purification Technology 73.2 (2010),

pags. 230 -242.
[34] Huang, K. y col. ((A totally heat-integrated distillation column (THIDiC) a
C the effect of
feed pre-heating by distillate)). En: Applied Thermal Engineering 28.8-9 (2008), pags.
856
-864.
[35] Mane, A. y Jana, A.K. ((A New Intensified Heat Integration in Distillation Column)). En: Indus-

trial & Engineering Chemistry Research 49.19 (2010), pags. 9534-9541.
[36] Nakaiwa, M. y col. ((Internally Heat-Integrated Distillation Columns: A Review)). En: Chemi-
cal Engineering Research and Design 81.1 (2003), pags. 162 -177.
[37] Nakaiwa, M. y col. ((A new configuration of ideal heat integrated distillation columns (HI-
DiC))). En: Computers & Chemical Engineering 24.2-7 (2000), pags.
239 -245.
[38] Suphanit, B. ((Optimal heat distribution in the internally heat-integrated distillation column
(HIDiC))). En: Energy 36.7 (2011), pags.
4171 -4181.
[39] Agrawal, R. ((A Method to Draw Fully Thermally Coupled Distillation Column Configura-
tions for Multicomponent Distillation)). En: Chemical Engineering Research and Design 78.3

(2000), pags. 454 -464.
[40] Agrawal, R. y Fidkowski, Z. ((Thermodynamically Efficient Systems for Ternary Distillation)). En:

Industrial & Engineering Chemistry Research 38.5 (1999), pags. 2065-2074.
[41] Agrawal, R. y Fidkowski, Z.T. ((More operable arrangements of fully thermally coupled disti-
llation columns)). En: AIChE Journal 44.11 (1998), pags.
2565-2568.
[42] Giridhar, A. y Agrawal, R. ((Synthesis of distillation configurations: I. Characteristics of a good
search space)). En: Computers & Chemical Engineering 34.1 (2010), pags. 73 -83.
[43] Giridhar, A. y Agrawal, R. ((Synthesis of distillation configurations. II: A search formulation for
basic configurations)). En: Computers & Chemical Engineering 34.1 (2010), pags. 84 -95.
[44] Petlyuk, F.B., Platonov, V.M. y Slavinsk, D. ((Thermodynamically optimal method for separa-
ting multicomponent mixtures)). En: International Chemical Engineering 5.3 (1965), pag.
555.
[45]
Rong, B.G., Kraslawski, A. y Nystrom, L. ((Design and synthesis of multicomponent thermally
coupled distillation flowsheets)). En: Comput. Chem. Eng. 25.4-6 (2001), pags.
807-820.
[46]
Rong, B.G., Kraslawski, A. y Nystrom, L. ((The synthesis of thermally coupled distillation flows-
heets for separations of five-component mixtures)). En: Computers & Chemical Engineering

24.27 (2000), pags. 247 -252.
[47] Jana, A.K. ((Heat integrated distillation operation)). En: Appl. Energy 87.5 (2010), 1477 -
1494.
[48] Fonyo, Z. y col. ((Energy savings of integrated and coupled distillation systems)). En: Com-
puters & Chemical Engineering 23, Supplement.0 (1999), S89 -S92.

166
Captulo 4. Resultados

[49] Amminudin, K.A. y Smith, R. ((Design and Optimization of Fully Thermally Coupled Distillation
Columns: Part 2: Application of Dividing Wall Columns in Retrofit)). En: Chemical Enginee-

ring Research and Design 79.7 (2001), pags. 716 -724.
[50] Amminudin, K.A. y col. ((Design and Optimization of Fully Thermally Coupled Distillation
Columns: Part 1: Preliminary Design and Optimization Methodology)). En: Chemical Engi-

neering Research and Design 79.7 (2001), pags. 701 -715.
[51]
Hernandez,
S. y Jimenez, A. ((Design of energy-efficient Petlyuk systems)). En: Computers &

Chemical Engineering 23.8 (1999), pags. 1005 -1010.
[52] Triantafyllou, C. y Smith, R. ((Design and optimisation of fully thermally coupled distillation
columns)). En: Chemical Engineering Research and Design 70.A2 (1992), pags. 118-131.
[53] Dunnebier, G. y Pantelides, C.C. ((Optimal Design of Thermally Coupled Distillation Co-
lumns)). En: Industrial & Engineering Chemistry Research 38.1 (1999), pags.
162-176.
[54] Kim, Y.H. ((Rigorous design of extended fully thermally coupled distillation columns)). En:

Chemical Engineering Journal 89.1-3 (2002), pags. 89 -99.
[55] Carlberg, N.A. y Westerberg, A.W. ((Temperature-heat diagrams for complex columns. 3.
Underwoods method for the Petlyuk configuration)). En: Industrial & Engineering Chemistry

Research 28.9 (1989), pags. 1386-1397.
[56] Navarro, M.A. y col. ((Strategies for the robust simulation of thermally coupled distillation
sequences)). En: Computers & Chemical Engineering 36.0 (2012), pags. 149 -159.
[57]
Cabrera-Ruiz, J., Jimenez-Guti
errez,
A. y Segovia-Hernandez, J.G. ((Assessment of the Im-
plementation of Heat-Integrated Distillation Columns for the Separation of Ternary Mixtu-
res)). En: Industrial & Engineering Chemistry Research 50.4 (2011), pags.
2176-2181.
[58] Caballero, J.A. y Grossmann, I.E. ((Logicbased methods for generating and optimizing
thermally coupled distillation systems)). En: European Symposium on Computer Aided Pro-
cess Engineering-12 35th European Symposium of the Working Party on Computer Aided
Process Engineering. Ed. por Grievink, Johan y Schijndel, Jan van. Vol. 10. Computer Aided

Chemical Engineering. Elsevier, 2002, pags. 169 -174.
[59] Caballero, J.A. y Grossmann, I.E. ((Logic-Sequential Approach to the Synthesis of Complex
Thermally Coupled Distillation Systems.)) En: 21st European Symposium on Computer Aided
Process Engineering. Ed. por E.N. Pistikopoulos, M.C. Georgiadis y Kokossis, A.C. Vol. 29.
Computer Aided Chemical Engineering. Elsevier, 2011, pags. 211 -215.
[60] Caballero, J.A. y Grossmann, I.E. ((Thermodynamically equivalent configurations for ther-
mally coupled distillation)). En: AIChE Journal 49.11 (2003), pags.
2864-2884.
[61] Turton, R. y col. Analysis, Synthesis and Design of Chemical Processes. Prentice Hall Interna-
tional Series in the Physical and Chemical Engineering Sciences. Pearson Education, 2008.
[62] Smith, R. Chemical process design and integration. Wiley, 2005.
[63] Carlberg, N.A. y Westerberg, A.W. ((Temperature-heat diagrams for complex columns. 2.
Underwoods method for side strippers and enrichers)). En: Industrial & Engineering Che-

mistry Research 28.9 (1989), pags. 1379-1386.

Appendix 1. Detailed Results of all Examples

The following tables contain the detailed results for all studied configurations (direct,
indirect, Petlyuk distillation with and without VRC).

167
4.2.REFERENCIAS

Results of example 1: Vapor recompression configuration

Table A1. Annual Capital Cost for all Configurations (e/year)

CAPITAL COST
Column 1 Column 1
Annual Cost Annual Cost
V(m3 ) Cost ( C) V(m3 ) Cost (C)
( C/year) ( C/year)
Direct Dist 201,6 728181 108520 205,0 754207 112399
Indirect Dist 189.7 687087 102396 192.6 706815 105336
Petlyuk Dist 95,2 360010 53652 316.7 1136213 169329
Petlyuk Dist with VRC 95.2 360010 53652 278.2 1002326 149376
Direct Dist with VRC 201,6 728181 108520 192.6 706815 105336
Condenser (Column 1) Reboiler (Column 1)
Annual Cost Annual Cost
A(m2 ) Cost ( C) A(m2 ) Cost (C)
( C/year) ( C/year)
Direct Dist 189.0 119893 17868 - - -
Indirect Dist - - - 1109.4 326302 48614
Petlyuk Dist - - - - - -
Petlyuk Dist with VRC - - - - - -
Direct Dist with VRC - - - 189.0 119694 17838
Condenser (Column 2) Reboiler (Column 2)
Annual Cost Annual Cost
A(m2 ) Cost ( C) A(m2 ) Cost (C)
( C/year) ( C/year)
Direct Dist 101.4 93491 13933 1738.5 451580 67299
Indirect Dist 300.8 149166 22230 293.3 147277 21949
Petlyuk Dist 222.1 128757 19189 1273.1 359296 53546
Petlyuk Dist with VRC 183.1 127125 18945 1042.5 336634 50168
Direct Dist with VRC 1047.3 313532 46726 - - -
Heater) Cooler
Annual Cost Annual Cost
A(m2 ) Cost ( C) A(m2 ) Cost (C)
( C/year) ( C/year)
Direct Dist - - - - - -
Indirect Dist - - - - - -
Petlyuk Dist - - - - - -
Petlyuk Dist with VRC 10.0 15862 2364 7.6 15051 3243
Direct Dist with VRC 27.4 66558 9919 23.0 64784 9655
Heat Exchanger) Compressor
Annual Cost Annual Cost
A(m2 ) Cost ( C) Energy (kw) Cost ( C)
( C/year) ( C/year)
Direct Dist - - - - - -
Indirect Dist - - - - - -
Petlyuk Dist - - - - - -
Petlyuk Dist with VRC 534.0 219997 32786 144.8 151439 22569
Direct Dist with VRC 1680.6 440206 65604 438.4 395890 58999

168
Captulo 4. Resultados

Table A2. Annual Energy Cost for all Studied Configurations (e/year)

ENERGY COST
Column 1 Column 1
Condenser (Column 1) Reboiler (Column 1)
Annual Cost Annual Cost
Energy (kw) Utility Energy (kw) Utility
( C/year) ( C/year)
Direct Dist 7465.1 Water 75982 - - -
Indirect Dist - - - 7690.6 MP Steam 1817633
Petlyuk Dist - - - - - -
Petlyuk Dist with VRC - - - - - -
Direct Dist with VRC - - - 7260.3 MP Steam 1715935
Condenser (Column 2) Reboiler (Column 2)
Annual Cost Annual Cost
Energy (kw) Utility Energy (kw) Utility
( C/year) ( C/year)
Direct Dist 4543.6 Water 46246 12051.8 MP Steam 2848379
Indirect Dist 11878.1 Water 120899 4212.2 MP Steam 995522
Petlyuk Dist 8772.0 Water 89284 8825.7 MP Steam 2085902
Petlyuk Dist with VRC 7230.6 Water 73595 7227.1 MP Steam 1708072
Direct Dist with VRC 7465.1 Water 75982 - - -
Heater) Cooler
Annual Cost Annual Cost
Energy (kw) Utility Energy (kw) Utility
( C/year) ( C/year)
Direct Dist - - - - - -
Indirect Dist - - - - - -
Petlyuk Dist - - - - - -
Petlyuk Dist with VRC 298.5 HP Steam 84378 338.1 Water 3441
Direct Dist with VRC 832.6 HP Steam 235314 1030.9 Water 10493
Compressor)
Annual Cost
Energy (kw) Utility
( C/year)
Direct Dist - - -
Indirect Dist - - -
Petlyuk Dist - - -
Petlyuk Dist with VRC 144.8 Electricity 69368
Direct Dist with VRC 438.4 Electricity 210097

Table A3. Total Annual Cost for all Studied Configurations (e/year)

TOTAL COST
Capital Cost ( C/year) Energy Cost ( C/year) Total Annual Cost ( C/year)
Direct Dist 320019 2970608 3290627
Indirect Dist 300525 2934054 3234579
Petlyuk Dist 295716 2175186 2470901
Petlyuk Dist with VRC 332104 1938855 2270959
Direct Dist with VRC 422597 2247821 2670418

169
4.2.REFERENCIAS

Results of example 2: The reverse vapor recompression configuration

Table A4. Annual Capital Cost for all Configurations (e/year)

CAPITAL COST
Column 1 Column 1
Annual Cost Annual Cost
V(m3 ) Cost ( C) V(m3 ) Cost ( C)
( C/year) ( C/year)
Direct Dist 199.3 1733135 258288 72.8 835298 124484
Indirect Dist 68.0 732040 109096 268.7 2490896 371217
Petlyuk Dist 49.6 521794 77763 287.4 2941774 438411
Petlyuk Dist with VRC 49.6 521794 77763 257.7 2354008 350817
Indirect Dist with VRC 68.0 732040 109096 261.0 2686405 400354
Condenser (Column 1) Reboiler (Column 1)
Annual Cost Annual Cost
A(m2 ) Cost ( C) A(m2 ) Cost ( C)
( C/year) ( C/year)
Direct Dist 1546.6 419777 62559 - - -
Indirect Dist - - - 586.6 219153 32660
Petlyuk Dist - - - - - -
Petlyuk Dist with VRC - - - - - -
Indirect Dist with VRC - - - 588.3 219530 32716
Condenser (Column 2) Reboiler (Column 2)
Annual Cost Annual Cost
A(m2 ) Cost ( C) A(m2 ) Cost ( C)
( C/year) ( C/year)
Direct Dist 912.4 289843 43195 847.2 276035 41137
Indirect Dist 2215.5 552593 82353 921.8 291827 43491
Petlyuk Dist 1804.0 471174 70219 817.1 269591 40177
Petlyuk Dist with VRC 1168.5 343114 51134 512.7 202340 30155
Indirect Dist with VRC 1623.5 435186 64856 - - -
Heat Exchanger) Compressor
Annual Cost Annual Cost
A(m2 ) Cost ( C) Energy (kw) Cost ( C)
( C/year) ( C/year)
Direct Dist - - - - - -
Indirect Dist - - - - - -
Petlyuk Dist - - - - - -
Petlyuk Dist with VRC 6844.1 1474823 219792 1364.9 946032 140987
Indirect Dist with VRC 8886.3 1899397 283066 1658.5 1085927 161835

170
Captulo 4. Resultados

Table A5. Annual Energy Cost for all Studied Configurations (e/year)

ENERGY COST
Column 1 Column 1
Condenser (Column 1) Reboiler (Column 1)
Annual Cost Annual Cost
Energy (kw) Utility Energy (kw) Utility
( C/year) ( C/year)
Direct Dist 11945.6 Very Low Temp 450819 - - -
Indirect Dist - - - 11920.2 Atm Steam 2286026
Petlyuk Dist - - - - - -
Petlyuk Dist with VRC - - - - - -
Indirect Dist with VRC - - - 11954.3 Atm Steam 2292561
Condenser (Column 2) Reboiler (Column 2)
Annual Cost Annual Cost
Energy (kw) Utility Energy (kw) Utility
( C/year) ( C/year)
Direct Dist 4257.6 Low Temp 965855 17213.4 Atm Steam 3301154
Indirect Dist 17092.4 Very Low Temp 6442843 6261.8 Water 63735
Petlyuk Dist 13940.5 Very Low Temp 5254756 16561.6 Atm Steam 3176150
Petlyuk Dist with VRC 9010.6 Very Low Temp 3396472 10421.8 Atm Steam 1998676
Indirect Dist with VRC 12525.0 Very Low Temp 4721212 - - -
Compressor)
Annual Cost
Energy (kw) Utility
( C/year)
Direct Dist - - -
Indirect Dist - - -
Petlyuk Dist - - -
Petlyuk Dist with VRC 1364.9 Electricity 654055
Indirect Dist with VRC 1658.5 Electricity 794768

Table A6. Total Annual Cost for all Studied Configurations (e/year)

TOTAL COST
Capital Cost ( C/year) Energy Cost ( C/year) Total Annual Cost ( C/year)
Direct Dist 529664 8769829 9299493
Indirect Dist 638816 8792604 9431420
Petlyuk Dist 626570 8430906 9057476
Petlyuk Dist with RVRC 870647 6049203 6919850
Indirect Dist with RVRC 1051922 7808540 8860463

171
Captulo 4. Resultados

4.3 III: An alternative disjunctive optimization mo-


Publicacion
del for heat integration with variable temperatures

Una de las principales preocupaciones de la industria de procesos qumicos es el



consumo de energa. Esto es debido a que el consumo energetico junto a las materias
al coste total de un producto. Co-
primas suele representar la principal contribucion

mo consecuencia, la clave para aumentar los beneficios economicos de un proceso

radica en mejorar su eficiencia, minimizando el consumo y las perdidas de energa y

aumentando la eficiencia energetica del proceso.

A lo largo de la historia, se han realizado numerosos avances en el campo de la


de procesos con el fin de mejorar la eficiencia energetica.
integracion el
Ademas,
aumento continuo de los costes de energa ha obligado a las empresas qumicas y
petroqumicas a encontrar nuevas formas de reducir el consumo de energa, adqui-
riendo una especial importancia la mejora de las redes de intercambiadores de calor.
de calor fueron
Los primeros investigadores en introducir el concepto de integracion
Linnhoff y Flower [1].


Entre todas las tecnicas de calor, hay
desarrolladas en el campo de la integracion

dos enfoques principales: el analisis del pinch, y los modelos basados
de localizacion
matematica
en programacion
[2]. Debido a su simplicidad, el analisis
de localizacion

del pinch es la tecnica ampliamente utilizada, ya que su aplicacion
mas es muy sen-
la eficacia y aplicabilidad de esta tecnica
cilla incluso a grandes problemas. Ademas
ha sido demostrada con numerosos ejemplos y problemas.


La gran mayora de los metodos energeti-
desarrollados en el campo de integracion
ca asumen que las temperaturas de entrada y salida son fijas y conocidas a priori. En
de calor se lleva a cabo solo
otras palabras, la integracion despues
de haber fijado
del proceso. Sin embargo, es sabido que la opti-
todas las condiciones de operacion
simultanea
mizacion de calor puede
de las condiciones del proceso y la integracion
producir importantes ahorros en el coste total del proceso [3]. Por lo que sabemos,

el unico e integracion
trabajo relacionado con la optimizacion de calor simultaneos

173
4.3.Alternative Model for heat integration

utilizando el concepto de pinch fue desarrollado por Duran y Grossmann [3] , e imple-
mentado como un MINLP por Grossmann et al [4].

El presente trabajo propone un modelo alternativo al propuesto por Duran y Gross-


y la integracion
mann para la optimizacion de calor simultaneas,
que mantiene el
concepto de intervalo de la temperatura con la que muchos dise
nadores se encuen-

tran familiarizados, debido a la prevalencia de la tecnologa del analisis del pinch. La

idea basica implcita de todas las temperaturas
es llevar a cabo una ordenacion
implci-
del sistema, de esa manera los intervalos de temperatura quedan tambien
tamente definidos. El modelo se compone de tres partes: en la primera, todas las
temperaturas de entrada del sistema son implcitamente ordenadas, en el segundo,
se calcula la cantidad de calor intercambiado por cada corriente en cada intervalo
de temperatura, y en la tercera, se aplica el modelo de transbordo desarrollado por
Papoulias y Grossmann [5] con el fin de calcular los servicios calientes y fros a aportar
al sistema.


Tal y como hemos descrito el modelo, este opera de forma correcta pero el coste
computacional requerido es enorme comparado con el modelo de Duran y Gross-
mann. Para solucionar dicho problema se procede a realizar un etapa de prepro-
cesado con el fin de fijar ciertas variables binarias y posteriormente se adicionan

un conjunto de relaciones logicas que reducen considerablemente el espacio de

busqueda.

a una serie de problemas con el fin de de-


La metodologa propuesta se aplico
mostrar su correcta funcionamiento. Los problemas estudiados incluyen problemas
de
que implican con temperaturas fijos y variables (MILP), un problema de integracion
(donde las temperaturas de las co-
calor en secuencia de columnas de destilacion
rrientes dependen de las condiciones de funcionamiento de las columnas) (MINLP),
propuesto y
y finalmente un problemas donde se combina el modelo de integracion
simultanea
la optimizacion de un diagrama de flujo, el cual se resuelve mediante un
simulador de procesos (ASPEN-HYSYS).

En todos los casos estudiados, el modelo propuesto se muestra competitivo frente


al modelo propuesto por Grossmann et al [4]. Comentar que aunque los problemas
generados por este modelo son mucho mayores a los obtenidos por Grossmann et

174
Captulo 4. Resultados

del preprocesado y las relaciones logicas


al., la aplicacion consigue reducir consi-
derablemente el tama
no del problema inicial, consiguiendo resultados competitivos

tanto en el tiempo de calculo,
como en el gap de relajacion. el modelo
Ademas,
propuesto mantiene el concepto de intervalo de temperatura en contraste con los
modelos existentes. Esta caracterstica es interesante para aquellos ingenieros e inves-

tigadores familiarizados con el concepto clasico de intervalo de temperatura en la
de calor. Por ultimo,
integracion
el modelo propuesto elimina las dificultades numeri-
max.
cas asociados con el uso de la funcion

175
4.3.Alternative Model for heat integration

An alternative disjunctive optimization model for heat integration


with variable temperatures

Jose
Miguel A. Navarro-Amoros, A. Caballero, Ruben
Ruiz-Femenia, Ignacio E. Grossmann

Computers & Chemical Engineering, 56 (2013), pp.12-26

Abstract

This paper presents an alternative model to deal with the problem of optimal energy
consumption minimization of non-isothermal systems with variable inlet and outlet tem-
peratures. The model is based on an implicit temperature ordering and the transship-
ment model proposed by Papoulias and Grossmann [5]. It is supplemented with a
set of logical relationships related to the relative position of the inlet temperatures of
process streams and the dynamic temperature intervals. In the extreme situation of
fixed inlet and outlet temperatures, the model reduces to the transshipment model.
Several examples with fixed and variable temperatures are presented to illustrate the
models performance.

Keywords

Heat integration, Disjunctive model, MILP, MINLP, Logic disjunctions

4.3.1. Introduction

A major concern of the chemical process industry is energy consumption. Energy


consumption and raw materials together usually constitute the main contribution to
the total cost of a product. To address this problem there has been an important deve-
lopment in process integration in parallel to the continuous evolution of the technolo-
gies for improving process plant design. Therefore, the key to increasing the economic

176
Captulo 4. Resultados

benefits and efficiency of a plant involve minimizing energy consumption and energy
losses, and increasing energy efficiency.

The first approaches for process integration were based on heuristic methods. The
practical implementation of heuristic methods relies on a set of rules derived from
both, first principles and designers experience for creating and screening process
alternatives. These heuristic concepts were progressively integrated into other met-
hods (e.g. graphical and thermodynamic insights), and were finally systematized by
different researchers,resulting in what is known today as conceptual design [6-9]. In its
more basic form the problem is decomposed into a set of levels of increasing detail.
For example, Douglas [6, 7] proposed decomposing the problem into five levels: 1.
batch vs. continuous, 2. input output structure, 3. recycle structure, 4. gas and liquid
sepa-ration and 5. heat integration. These levels were extended by Smith [8] to include
waste disposal, water networks, environmental considerations, etc.

The main limitation of this sequential approach is that decisions in the outer levels
constrain the alternatives in the most inner levels. The simultaneous design of reaction
separation and heat integration can significantly improve process performance.The
reaction can have a large impact on both raw materials and energy usage (in most
cases mainly on raw material consumption).The separation, depending on what the
conversion and/or selectivity are, can also have a great impact on raw material con-
sumption(e.g. the non-reacted feed is very difficult to separate), as well as on energy.
And finally heat integration has a major impact on energy consumption. In this paper
our focus is on latter.

Continually increasing energy costs has forced chemical and petrochemical com-
panies to find new ways of decreasing energy consumption by improving especially
heat exchanger networks.The first researchers to introduce the concept of heat inte-
gration were Linnhoff and Flower in 1978 [1]. They developed the concept of pinch
analysis whose initial objective was to determine the minimum utilities requirements
of a process, and to identify the best possible degree of heat recovery achievable
asa function of the minimum temperature difference inside the heat exchanger net-
work. In 1983, Linnhoff and Hindmarsh [10] demonstrated that it is possible to save a
significant part of all the energy required by a plant.

177
4.3.Alternative Model for heat integration

Since those initial developments, there has been rapid growth in the field of heat
integration. An excellent review with annotated bibliography can be found in the
work by Furman and Sahinidis [11]. Moreover, Morar and Agachi [2] presented a com-
prehensive review of heat integration techniques.

Among all the techniques that have been developed for heat integration, there
are two main approaches: pinch analysis, and the mathematical programming based
models [2]. Because of its simplicity, pinch analysis is the more widely used technique,
since it can be easily applied by hand even to large problems. It has demonstrated ef-
ficiency and applicability in many problems on industrial energy-savings. Pinch analysis
can be divided into three steps [1, 10, 12-14]:

1. Given a minimum heat recovery approach temperature (HRAT),generate by


graphical or tabular means the hot and cold composite curves and read in
the minimum utilities for the entire system. It is also possible to determine which
among all the avail-able utilities are the most adequate, using the grand com-
posite curve

2. Estimate the minimum number of heat exchangers

3. Estimate the total area and total cost (or synthesize the net by using heuristic
rules)

4. Repeat steps 13 using different values of HRAT until the mini-mum cost of the
net is obtained

Methods based on mathematical programming initially followed a sequential ap-


proach, imitating the pinch analysis:determination of minimum utilities, which takes the
form of a transshipment LP model [5], or a transportation LP model [15]; determination
of the minimum number of heat exchangers that can be formulated as a mixed in-
teger linear programming problem(MILP); and finally, generation of the actual heat
exchangers net-work, which is an NLP problem [16].

While the sequential targeting and optimization approach have the advantage
of decomposing the synthesis problem into smaller problems, it has the disadvantage

178
Captulo 4. Resultados

that the trade-offs among energy, number of units and area are not rigorously taken
into account. The reason for is that the optimization problem:

min: Area Cost + Fixed Cost Units + Utility Cost (4.3.1)

is approximated by a problem that can be stated conceptually as follows [17]:

min:Area Cost

s.t. min: Number of Units (4.3.2)

s.t. Minimum utility cost

To avoid the limitations of the sequential approach, different researchers proposed


instead simultaneous approaches to solving the problem. For example, Floudas and
Ciric [18] used a superstructure based approach (they called hyper-structure), formu-
lated as a MINLP problem, to simultaneously optimize the number of heat exchangers
and the investment cost. Later, Ciric and Floudas [19] extended the hyper-structure
to the optimization of the total annual cost of the network (simultaneous optimization
of utilities and investment costs). However, the most successful simultaneous model is
probably due to Yee and Grossmann [20-22]. The model is based on a superstructure
composed of a set of stages in which each hot stream can exchange heat with all the
cold streams;the mixture of streams at the end of each stage is assumed to be isother-
mal. Although this model does not possess some of the additional features present in
other simultaneous methods, it performs better numerically because the non-linearities
are only in the objective function while the constraints are all linear. Substantial impro-
vements have been achieved and reported when simultaneous models were used
instead of sequential ones [17].

All the methods discussed thus far assume that the inlet and out-let temperatures
are fixed and known a priori. In other words, the heat integration is performed only
after all the process operation conditions have been fixed. But it is known that the si-

179
4.3.Alternative Model for heat integration

multaneous optimization of the process conditions and heat integration can result in
important savings in total cost [3]. Including the detailed design of the heat exchan-
ger net-work (HEN) in the detailed process optimization tends to produce large non-
convex MINLP problems. However, if we assume that the dominant cost associated
with the HEN is the energy, and that the inlet and outlet temperatures could also have
a synergistic effect on the rest of the process, it is justified to simultaneously optimize
the process operating conditions (specifically including the stream temperatures) and
the cost of external hot and cold utilities that result from maximum heat integration.

As far as we know, the only research dealing with simultaneous optimization and
heat integration focusing on process and energy cost minimization but not including
the design of HEN are due to Duran and Grossmann [3] and Grossmann et al. [4].
However, a number of special models have been developed for specific applications,
e.g. Floudas and Paules [23] or Raman and Grossmann [24], for heat integration in
distillation systems,adapted later by Caballero and Grossmann [25, 26] to complex
systems. Taking into account that the pinch point coincides with the inlet temperature
of any of the hot or cold streams, and assuming constant heat capacity flowrates,
these authors showed that the criteria for selecting the correct pinch corresponding
to minimum feasible heating and cooling involved selecting among all the pinch
candidates the one that exhibited the largest heating and cooling (pinch location
method). In that way Duran and Grossmann [3] did not need the concept of tem-
perature interval to locate the pinch point. The final mathematical formulation results
in anon-differentiable optimization problem (due to the max opera-tors that appear
in the model). Initially, a smooth approximation was proposed to solve the problem.
Later, Grossmann et al. [4] proposed a rigorous disjunctive formulation of the pinch
location method that uses logic disjunctions to explicitly model the relative place-
ment of streams for various potential pinch locations, and explicitly considers the non-
isothermal and isothermal streams as separate cases.

In this paper we propose an alternative model for simultaneous optimization and


heat integration, which retains the concept of temperature interval with which many
designers are familiar due to the prevalence of the pinch technology. The basic idea
is to perform an implicit ordering of all the temperatures; in that way the temperature

180
Captulo 4. Resultados

intervals are also implicitly defined, and the classicalequations for the transshipment
[5] can be easily incorporated in the model.

4.3.2. Problem statement

We address the following problem. Given:

A superstructure of alternative process flowsheets

A set nH of hot process streams to be cooled and a set nC of cold process


streams to be heated

The available utilities, their temperatures, and their costs per unit of heat (pro-
vided or removed) determine the optimal process flowsheet that minimizes the
utility cost

The flowrates, inlet and outlet temperatures for all these process streams are not
fixed, and must therefore be optimized.

We will assume that the heat flow rates of the hot and cold process streams are
constant, and that the inlet and outlet temperatures are bounded by upper and lo-
wer limits. We formulate this minimum cost problem as a linear disjunctive program-
ming problem and rewrite it as a MILP with the aid of a Hull Reformulation(HR). The
nonlinearities that appear are due to the remaining equations describing the process,
but not to those specifically related with the heat integration. In the next section a
comprehensive description of the model is presented.

4.3.3. Simultaneous heat and process optimization: model formula-


tion

As in the models presented by Duran and Grossmann [3] and Grossmann et al. [4],
we implicitly assume that the simultaneous optimization and heat integration strategy,
together with the dominant energy cost, yield important economic benefits despite
the fact that the investment costs related to the HEN are not taken into account. A

181
4.3.Alternative Model for heat integration

fixed heat recovery approach temperature (HRAT) must be assumed, and therefore it
must be optimized in an outer loop using a targeting approach or a detailed design
of the network.

With all these points in mind, we introduce the following index sets for the proposed
model:

ST: k: k is a stream

HOT: i: i is a hot stream

COLD: j: j is cold stream

HU: m: m is a hot utility

CU: n: n is a cold utility

TI: w: w is a temperature interval

The set ST (ST = HOT COLD) is an ordered set formed by the nH hot streams follo-
wed by the nC cold streams.

From a conceptual point of view the model can be written as an MINLP problem
of the following structure:

X X
HU HU CU CU
min F (xp , xh ) + Cm Qm + Cn Qn
mHU nCU

s.t. h(xp , xh ) = 0

g(xp , xh ) 0 (4.3.3)
Axh b

QHU CU
m , Qn R+ , m HU, n CU

xh R n , x p R n

where QHU CU are the heat load of hot utility m and cold utility n, respecti-
m and Qn
HU and C CU are the unit cost of hot utility m and cold utility n, respectively.
vely, and Cm n

182
Captulo 4. Resultados

The vector variable xp represents process parameters such as pressures, flowra-


tes, and equipment sizes, and structural parameters usually represented by 0-1 binary
variables. The vector variable xh represents the flowrates and temperatures of the
pro-cess streams that undergo either cooling or heating, and also the binary variables
relating to energy integration used in the model equations.

The vectors of constraint h and g represent the material and energy balances, de-
sign specifications or structural relationships in the synthesis problem. The linear equa-
tions Axh b are used specifically for the heat integration presented in the next sec-
tion.

The model is composed of three different parts: in the first,all inlet temperatures of
the system are implicitly ordered; in the second, the heat exchange is calculated for
each stream in each temperature interval; and in the third, the previously calculated
heat exchanges of each interval are connected in descending order via the transs-
hipment model.

4.3.3.1. Implicit ordering of temperatures

The first step involves implicitly ordering the temperatures that determine the tem-
perature intervals. It is worth noting that under the assumption of constant heat ca-
pacity flowrates, the pinch point always appears between an inlet temperature of a
hot ora cold stream and any temperature of a cold or a hot stream, respectively. The-
refore, we need only consider inlet temperatures when calculating the pinch point or
temperature intervals.

Working with the actual temperatures leads to a pinch point that separates the
hot composite curve and the cold composite curve by a temperature interval that
exactly equals the value of HRAT .If we decrease the temperature of the hot streams
by HRAT /2 and increase the temperature of the cold streams by HRAT /2, then we
can obtain a pinch point that corresponds to a single temperature instead of an inter-
val:

183
4.3.Alternative Model for heat integration

in ,in HRAT
TH = TH (4.3.4)
2
out ,out HRAT
TH = TH (4.3.5)
2
in ,in HRAT
TC = TC + (4.3.6)
2
out ,out HRAT
TC = TC + (4.3.7)
2

in and T out are the shifted inlet and outlet temperatures for the hot streams,
where TH H
in
TC out are the corresponding shifted inlet and outlet temperatures for the cold
and TC
streams, and the superscript * denotes an actual temperature. The shifted tempera-
tures must be constrained to lie within lower and upper bounds:

,in HRAT in ,in HRAT


TH TH TH
2 2
,out HRAT out ,out HRAT
TH TH TH
2 2
(4.3.8)
,in HRAT in ,in HRAT
TC + TC TC +
2 2
,out HRAT out ,out HRAT
TC + TC TC +
2 2

It is also convenient to define an auxiliary temperature that stores the inlet tempe-
ratures of all the streams:


T in , if k = i
H
Tkin = k = 1, . . . , nH , . . . , nH + nC (4.3.9)
T in , if k = n + j
C H

In Eq. 4.3.9 the indices k, i, j, are ordered sets and also denote position inside their
respective sets.

The sorted temperatures correspond to a permutation of the positions of the actual


temperatures in the set of streams ST ; the set of variables Tkord
0 store the sorted values

184
Captulo 4. Resultados

of all the inlet temperatures in descending order. We accomplished this with the aid
of the definition of the Boolean variable Yk,k0 , which is true if the inlet temperature in
position k occupies position k0 after the ordering, and false otherwise.

The disjunctive model is formulated as follows:


Yk,k0
k ST (4.3.10)
k0 ST Tkin = Tkout
0

Yk,k0 k ST
k0 ST
(4.3.11)
Yk,k0 k0 ST
kST

ord
Tk1 Tkord k \ {1} ST (4.3.12)

Eq. 4.3.10 assigns to the ordered temperature k the value of the initial temperature
k. The equations in 4.3.11 assign a fixed position to an ordered temperature, implying
that a given ordered position can only be occupied by one stream at a time. Eq.
4.3.12 forces the descending ordering of temperatures.

Eqs. 4.3.10 and 4.3.11 can be reformulated in terms of binary variables(in this paper
we use capital letters for the Boolean and small letters for the binary variables). The
binary variable takes value 1 if the Boolean variable is True and 0 otherwise. The HR
formulation for the disjunctions in Eq. 4.3.10 is:

in,D ord,D
Tk,k 0 = Tk0 ,k k0 k ST (4.3.13)
in,D
X
Tkin = Tk,k 0 k ST (4.3.14)
k0 ST

Tkord,D
X
Tkord = 0 ,k k0 ST (4.3.15)
kST

in,D
T yk,k0 Tk,k 0 T yk,k0

k0 k ST (4.3.16)
ord,D
T yk,k0 Tk,k 0 T yk,k0

185
4.3.Alternative Model for heat integration

where each of the two variables Tkin and Tkord


0 appearing in the terms of disjunction
in,D
k are disaggregated into nC + nH new variables, Tk,k 0 and Tkord,D
0 ,k as described
by Eqs. 4.3.14 and 4.3.15. In Eq. 4.3.16, we use upper and lower bounds and binary
in,D ord,D
variables yk,k0 to force the disaggregated variables Tk,k 0 , Tk0 ,k to zero when the
initial temperature k is not assigned to the ordered temperature k0 .

The assignment logic propositions, Eq. 4.3.11, are written into algebraic form using
binary variables, as follows:

X
yk,k0 = 1 k ST
k0 ST
X (4.3.17)
yk,k0 = 1 k0 ST
kST

Once the temperatures have been sorted, the ordered inlet temperatures (toget-
her with the maximum and minimum possible temperatures, T max and T min , respec-
tively) define the temperature intervals (Fig. 4.3.1). The total number of temperature
intervals (NI) is equal to the number of hot streams plus the number of cold streams
plus one.

U P ) and lower (T LO )temperatures as


Each interval is characterized by its upper (Tw w

follows:

T1U P = Tmax
UP
Tw+1 = Tkord w, k = 1, . . . , nH + nC
(4.3.18)
LO
Tw = Tkord w, k = 1, . . . , nH + nC
LO
TN I = Tmin

186
Captulo 4. Resultados

Figura 4.3.1: Illustration of ordered temperatures and intervals

4.3.3.2. Heat transfer within each temperature interval

For a given temperature interval and a given stream, two cases arise where the
stream does not exchange heat in that interval.The first case occurs when the tempe-
rature interval is above the inlet temperature for a hot stream or the outlet temperature
for a cold stream (cases H2 and C1 in Fig. 4.3.2). The other case of no heat exchan-
ge occurs when the temperature interval is below the outlet temperature for a hot
stream or the inlet temperature for a cold stream (cases H1 and C2 in Fig. 4.3.2). We
must explicitly take into account these two alternatives.

A given stream can interchange heat within a certain interval in two ways. For a
hot stream, if the outlet temperature of the stream is below the lower temperature
of the interval, then this stream transfers heat throughout the whole interval (case H3
in Fig. 4.3.2).Alternatively, if the outlet temperature of the hot stream is greater than

187
4.3.Alternative Model for heat integration

Figura 4.3.2: All the relative positions of a stream in relation to a given temperature
interval

the lower temperature of the interval, the stream exchanges heat only in a part of
the interval (case H4 in Fig. 4.3.2). Similarly, a cold stream transfers heat throughout
the entire interval if its outlet temperature is higher than the upper temperature of the
interval(case C3 in Fig. 4.3.2). Likewise if the outlet temperature of the cold stream is
lower than the upper temperature of the interval, then this stream exchanges heat
only in a part of the interval (case C4 in Fig. 4.3.2).

To model these alternatives, we write the linear disjunctions 4.3.19 and 4.3.20, which
require definition of the following Boolean variables:


C
Zj,w
C
Zj,w
C3
Zj,w C4
Zj,w
C1
Zj,w C2
Zj,w j COLD



QC = F CP C (T U P T LO )
j,w j w w

C C
Qj,w = F CPj (TjCout LO


Tw )

(4.3.19)
QC QC w TI

j,w = 0 j,w = 0

TjCout Tw UP TjCin TwLO
TjCout Tw
LO TjCin Tw
UP

TjCin TwLO U P T Cout T LO
Tw j w

188
Captulo 4. Resultados


H
Zi,w
Z H
i,w

H3
Zi,w H4
Zi,w
H1
Zi,w H2
Zi,w i HOT



QH = F CP H (T U P T LO )
i,w i w w

H H U P
Qi,w = F CPi (Tw Ti Hout

)

(4.3.20)
QH QH w TI

i,w = 0 i,w = 0

TiHin TwUP TiHin TwUP
TiHin Tw
LO TiHout Tw
UP

TiHout Tw LO U P T Hout T LO
Tw i w

H
Zi,w True if the hot stream i exchanges heat in the temperature interval w

C
Zi,w True if the cold stream j exchanges heat in the temperature interval w

H1 True if the inlet temperature of the hot stream i is below the temperature of
Zi,w
interval w

H2 True if the outlet temperature of the hot stream i is above the temperature inter-
Zi,w
val w

H3 True if the hot stream i traverses the temperature interval w


Zi,w

H4 True if the outlet temperature of the hot stream i is inside the temperature interval
Zi,w
w

C1 True if the outlet temperature of the cold stream j is below the temperature
Zj,w
interval w

C2 True if the inlet temperature of the cold stream j is above the temperature inter-
Zj,w
val w

C3 True if the cold stream j traverses the temperature interval w


Zj,w

C4 True if the outlet temperature of the cold stream j is inside the temperature inter-
Zj,w
val w

In disjunctions 4.3.19 and 4.3.20 it is assumed that heat flow rates F CPiH ,F CPjC are
known and constant (it is possible to include the heat flowrates as variables, but in this
case the problem becomes bilinear). QH C
i,w and Qj,w are the heat flows exchanged by
hot and cold streams in the interval w, respectively.

Furthermore, a complete description of the alternatives demands making use of


the following logical expressions:

189
4.3.Alternative Model for heat integration

H H3 H4
Zi,w Zi,w Zi,w , i HOT ; w T I (4.3.21)
H H1 H2
Zi,w Zi,w Zi,w , i HOT ; w T I (4.3.22)
C C3 C4
Zj,w Zj,w Zj,w , j COLD; w T I (4.3.23)
C C1 C2
Zj,w Zj,w Zj,w , j COLD; w T I (4.3.24)

Disjunctions 4.3.19 and 4.3.20 can be rewritten in terms of binary variables with the
aid of HR. For the hot streams Eq. 4.3.19 the reformulation reads as follows:

190
Captulo 4. Resultados


QH QH
P
i,w = i,w,d


dHD




UP = UP
P
Tw Ti,w,d


dHD




LO = LO
P
Tw Ti,w,d w T I, i HOT (4.3.25)
dHD


TiHin Hin
P
= Ti,w,d




dHD



TiHout = Hout
P
Ti,w,d



dHD

QH H UP LO
i,w,h3 = F CPi (Ti,w,h3 Ti,w,h3 )



Hin U P
Ti,w,h3 Ti,w,h3 i HOT, w T I (4.3.26)


T Hout T LO


i,w,h3 i,w,h3

QH H UP Hout
i,w,h4 = F CPi (Ti,w,h4 Ti,w,h4 )



T Hin T U P

i,w,h4 i,w,h4
i HOT, w T I (4.3.27)
Hout T U P
Ti,w,h4
i,w,h4




Hout T LO
Ti,w,h4

i,w,h4

QH
i,w,h1 = 0

i HOT, w T I (4.3.28)
Hin T LO
Ti,w,h1
i,w,h1

QHi,w,h2 = 0

i HOT, w T I (4.3.29)
T Hout T U P
i,w,h2

i,w,h2

QH d
i,w,d Qi zi,w




d
Tw zi,w U P T zd
Ti,w,d


w i,w


d
Tw zi,w LO T z d
Ti,w,d i HOT, w T I (4.3.30)
w i,w


TiHin zi,w
d Hin T Hin z d
Ti,w,d


i i,w



TiHout d Hout T Hout z d
zi,w Ti,w,d

i i,w

HR requires introducing disaggregated variables for QH UP LO Hin , and


i,w , Tw , Tw , Ti
TiHout , which we label with subscript d, the index of the set HD = H1, H2, H3, H4 (Eq.
4.3.25). In Eq. 4.3.30, an upper and lower bound is assigned to each disaggregated
d
variable, so that when zi,w = 1 the variables can take values between these bounds,
d
and if zi,w = 0 then QH UP LO Hin Hout
i,w,d = Ti,w,d = Ti,w,d = Ti,w,d = Ti,w,d = 0. Note that the bounds

191
4.3.Alternative Model for heat integration

for the disaggregated variables, those variables which define the temperature interval
U P and T LO ), depend on the interval itself.
w (that is Tw w


QC QC
P
j,w = j,w,d


dCD




UP = UP
P
Tw Tj,w,d


dCD




LO = LO
P
Tw Tj,w,d w T I, j COLD (4.3.31)
dCD


TjCin = Cin
P
Tj,w,d




dCD



TjCout = Cout
P
Tj,w,d



dCD

QC C UP LO
j,w,c3 = F CPj (Tj,w,c3 Tj,w,c3 )



Cin LO
Tj,w,c3 Tj,w,c3 j COLD, w T I (4.3.32)


T Cout T U P


j,w,c3 j,w,c3

QC C Cout
j,w,c4 = F CPj (Tj,w,c4
LO )
Tj,w,c4


Cin T LO
Tj,w,c4


j,w,c4
j COLD, w T I (4.3.33)
Cout T U P
Tj,w,c4
j,w,c4



Cout T LO
Tj,w,c4

j,w,c4

QC
j,w,c1 = 0

j COLD, w T I (4.3.34)
Cout T LO
Tj,w,c1
j,w,c1

QCj,w,c2=0
j COLD, w T I (4.3.35)
Cin T U P
Tj,w,c2
j,w,c2

QC d
j,w,d Qj zj,w




d U P T zd

Tw zj,w Tj,w,d

w j,w



d LO T z d
Tj,w,d
Tw zj,w w j,w j COLD, w T I (4.3.36)

TjCin zj,w
d Cin T Cin z d

Tj,w,d j j,w






TjCout zj,w
d Cout T Cout z d
Tj,w,d


j j,w

where analogously to the hot case, we define the set CD = C1, C2,C3, C4 to achie-
ve disaggregation of the variables.

192
Captulo 4. Resultados

Finally, the logical expressions 4.3.214.3.24 are converted into linear inequality cons-
traints as follows:

h h3 h4
zi,w = zi,w + zi,w , i HOT ; w T I (4.3.37)
h h1 h2
1 zi,w = zi,w + zi,w , i HOT ; w T I (4.3.38)
c c3 c4
zj,w = zj,w + zj,w , j COLD; w T I (4.3.39)
c c1 c2
1 zj,w = zj,w + Zj,w , j COLD; w T I (4.3.40)

4.3.3.3. Transshipment model

In order to calculate the heat and cool utilities as well as the heat exchanged by
the entire system, we use a heat cascade diagram,following the conventional ap-
proach in pinch analysis (Fig. 4.3.3).The mathematical formulation of the problem was
proposed by Papoulias and Grossmann [5]. The heat cascade diagram can be re-
garded as a transshipment problem. In this model, hot streams are treated as source
nodes, and cold streams as destination nodes.Heat can then be regarded as a com-
modity that must be transferred from the sources to the destinations through a number
intermediate warehouses corresponding to the temperature intervals that guaran-
tee feasible heat exchange. When not all of the heat can be allocated to the desti-
nations (cold streams) in a given temperature interval, the excess is cascaded down
to lower temperature intervals through the heat residuals.

The equations for the transshipment model can be formulated as follows. We have
w temperature intervals that are numbered from the top to bottom. The heat balances
of the model are:

193
4.3.Alternative Model for heat integration

Figura 4.3.3: Heat cascade diagram

X X
QHU + QH
i,1 = R1 + QC
j,1
iHOT jCOLD
X X
Rw1 + QH
i,w = Rw + QC
j,w w = 2, . . . , W 1
iHOT jCOLD (4.3.41)
X X
RW 1 + QH
i,w = Q
CU
+ QC
j,W
iHOT jCOLD

RW = 0

This model correctly predicts the minimum utility for variable inlet temperatures, but
at a high cost in terms of computational performance. The reasons for this have to
do with the large number of combinations of binary variables that lead to infeasible
subproblems and a relaxation gap that is too large. These problems can be overcome
by first including a preprocessing step to fix a priori some binary variables (this prepro-

194
Captulo 4. Resultados

cessing is explained in the next section); and by adding a set of logical relationships
among the Yk,k0 Boolean variables, which deal with the arrangement of the inlet tem-
peratures, and the Z Boolean variables, which are related to the heat transferred by a
stream within a certain temperature interval.

4.3.3.4. Logical relationships relating reordering and heat exchange binary variables

To make the mathematical formulation as simple as possible, it is convenient to


recall that the sets HOT (hot streams), COLD (cold streams), ST (streams = HOT COLD)
and TI (Temperature intervals)are ordered sets. For simplicity, it is also convenient to
define the following additional sets:

P HIk,i Streams in sets k ST, i HOT if k = i

P CIk,j Streams in sets k ST, j COLD if k = nH + j

Hot streams

1. If the inlet temperature of the hot stream i (i P HIk,i ) is assigned to position


k0 then this stream does not exchange heat in any temperature interval above
that defined by the position k0 (see Fig. 4.3.4a)

H1
Yk,k0 Zi,w (k, i) P HIk,i , w IT, k0 ST, w k0 (4.3.42)

2. If the inlet temperature of the hot stream i (i P HIk,i ) is assigned to position k0


then this stream exchanges heat in the temperature interval starting at that po-
sition (counting top to down). The stream either traverses that interval or finishes
somewhere inside it (see Fig. 4.3.4b)

H3 Z H4
Yk,k0 Zi,w i,w (k, i) P HIk,i , w IT, k0 ST, w = k0 + 1 (4.3.43)

195
4.3.Alternative Model for heat integration

3. If the hot stream i (i P HIk,i ) does not exchange heat in a given interval then
the inlet temperature of that stream cannot be assigned to position k0 , which
defines the upper value of the interval (see Fig. 4.3.4c)

H Y
Zi,w k,k0 (k, i) P HIk,i , w IT, k0 ST, w = k0 + 1 (4.3.44)

4. If the outlet temperature of the hot stream i (i P HIk,i ) is inside a temperature


interval w then the inlet temperature of that stream cannot be in any interval
below w (see Fig. 4.3.4d)

H4 Y
Zi,w k,k0 (k, i) P HIk,i , w IT, k0 ST, k0 w (4.3.45)

Cold streams

5. If the inlet temperature of the cold stream j (j P CIk,j ) is assigned to position


k0 then this stream does not exchange heat in any temperature interval below
that defined by the position k0

C2
Yk,k0 Zj,w (k, j) P CIk,j , w IT, k0 ST, w k0 + 1 (4.3.46)

6. If the inlet temperature of the cold stream j (j P CIk,j ) is assigned to posi-


tion k0 then this stream exchanges heat in the temperature interval finishing at
that position (counting top to down). The stream either traverses that interval or
finishes somewhere inside it.

C3
Yk,k0 Zj,w C4
Zj,w (k, j) P CIk,j , w IT, k0 ST, w = k0 (4.3.47)

196
Captulo 4. Resultados

Figura 4.3.4: Logical relationships for hot streams

7. If the cold stream j (j P CIk,j ) does not exchange heat in a given interval
then the inlet temperature of that stream cannot be assigned to the position k0 ,
which defines the lower value of the interval:

C
Zj,w Yk,k0 (k, j) P CIk,j , w IT, k0 ST, k0 = w (4.3.48)

197
4.3.Alternative Model for heat integration

8. if the outlet temperature of the cold stream j (j P CIk,j ) is inside a temperatu-


re interval w then the inlet temperature of that stream cannot be in any interval
above w

C4
Zj,w Yk,k0 (k, j) P CIk,j , w IT, k0 ST, k0 w 1 (4.3.49)

The scheme of these logical relationships for the cold streams is similar to that of
the hot streams shown in Fig. 4.3.4. However, in the case of cold streams the heat
exchange is produced in the opposite direction (from cold to hot temperatures).

4.3.3.5. Logical relationships involving heat exchange intervals

1. If a hot/cold stream is below the temperature interval w, then it is also below all
the intervals above w

H1 H1
Zi,w Zi,w1 i HOT ; w(w > 1) T I (4.3.50)
C1 C1
Zj,w Zi,w1 j COLD; w(w > 1) T I (4.3.51)

2. If a hot/cold stream is above the temperature interval w, then it is also above


all the intervals below w

H2 H2
Zi,w Zi,w+1 i HOT ; w(w 6= W ) T I (4.3.52)
C2 C2
Zj,w Zi,w+1 j COLD; w(w 6= W ) T I (4.3.53)

3. If a hot/cold stream finishes the heat exchange in the interval w, then that
stream does not exchange heat in the temperature intervals below/above w

198
Captulo 4. Resultados

H4 H2
Zi,w Zi,w+1 i HOT ; w(w 6= W ) T I (4.3.54)
C4 C2
Zj,w Zi,w1 j COLD; w(w > 1) T I (4.3.55)

Note that although Eqs. 4.3.54 and 4.3.55 only force the variables Z H2 , Z C2 to
be true in the interval immediately above/below w, Eqs. 4.3.52 and 4.3.53 then
become active and force these variables to be active in the rest of the intervals
above/below the interval w

4. If a hot stream exchanges heat in a temperature interval w,then in the following


interval (top to bottom) it either continues exchanging heat, or finishes exchan-
ging heat, or does not exchange heat at all

H3 H3 H4 H2
Zi,w Zi,w+1 Zi,w+1 Zi,w+1 i HOT, w(w 6= W ) T I (4.3.56)

5. If a cold stream exchanges heat in a temperature interval w,then in the fo-


llowing interval (bottom to top) it either continues exchanging heat or finishes
exchanging heat, or does not exchange heat at all.

C3 C3 C4 C2
Zj,w Zj,w1 Zj,w1 Zj,w1 j COLD, w(w > 1) T I (4.3.57)

4.3.3.6. Preprocessing

The objective of preprocessing is to reduce the size of the problem: reducing the
number of variables by assigning fixed values to certain binary variables (yk,kk ) and by
extension to some z variables; or by limiting the number of intervals that a given stream
can be assigned to. A key characteristic of the present model is that the complexity
of the resulting model depends on the degree of overlapping in the inlet temperature
intervals, for both the hot and cold streams. Two extreme situations can arise: the first
occurs when all the inlet temperatures are fixed; in this case all the binary variables

199
4.3.Alternative Model for heat integration

can be fixed a priori and the model becomes an LP model (no integer variables): it
reduces to the transshipment problem [5]. The other extreme occurs when all the inlet
temperatures overlap; in this case it is not possible to do any a priori assignment. As
important as fixing some assignments is restricting the intervals of values to which a
given inlet stream temperature can be assigned is just as important.

Consider, for example, Fig. 4.3.5, which corresponds to test 3 in example 2 (see Ta-
ble 3). It shows the minimum and maximum values for the inlet temperatures of both
hot and cold streams;the streams are arranged by upper bound, and have been
drawn accordingly. In this way it is possible verify that stream H1 can only be assigned
to position 1; streams C3, C4, and H2 only to positions2, 3 or 4; stream H4 to positions 5
or 6; stream C2 to positions 5, 6or 7; stream H3 to positions 6 or 7, and finally, stream C1
to position8.

Figura 4.3.5: Minimum and maximum values for the inlet temperatures of hot and cold
streams

Note that if two inlet temperatures are equal we obtain a degenerate interval of
temperatures (the upper and lower bound temperatures of the interval are the same).

200
Captulo 4. Resultados

Note also that, in Fig. 4.3.5, we have drawn the positions for the assignments according
to the upper temperatures of the inlet streams.Therefore, positions 5 and 6 appear
to overlap but this does not mean that they necessarily have to in the final optimal
solutions.

Based on the foregoing analysis, it now becomes straightforward to set some values
of the Z H and Z C variables equal to zero just by inspecting the outlet temperatures
of the various streams.

Finally, once some inlet temperatures are constrained to lie in a reduced number of
intervals, the upper and lower bounds of temperatures for that interval can be adjus-
ted to obtain a tighter model(i.e. adjusted to the minimum and maximum of the inlet
temperatures of the streams that could start in that interval).

4.3.3.7. Final remarks, extensions and performance of the disjunctive model

There are several important considerations and extensions of the above model that
deserve a more detailed discussion.

The first important point pertains to the models performance.In other words, when
and why could we expect a better performance from this model versus the pinch
location method? Specifically, the comparison is with the disjunctive implementation
by Grossmann et al. [4]:

1. The logical relationships in Sections 4.3.3.4 and 4.3.3.5, as well as the preproces-
sing stage, can be removed from the model. The optimal solution remains the
same, however under these conditions the performance is really bad (e.g. very
poor relaxation gap and large CPU times). In that case increasing the number
of equations will improve the models performance.

2. How the model performs is directly related to the degree of overlapping of the
interval of possible variation of the inlet temperatures: the lower the extent of
overlapping the better the performance of the model (it is important to remark
that the overlap is only in the variation of inlet temperatures and not between
the inlet and outlet temperatures). Test problems 3 and 4 (see the Examples

201
4.3.Alternative Model for heat integration

section) show that the relaxation gap drastically decreases with overlapping
degree. The relaxation gap in these examples is much better than that which
results from using the disjunctive implementation of the pinch location method
by Grossmann et al. An extreme situation occurs when there is no overlapping
at all; in that case all the binary variables and most of the variables are fixed
and the model is an LP or NLP.

3. Related to the previous point, it is worth noting that although the difference
between inlet and outlet temperatures of a stream can be large, in industrial
practice the bounds for the inlet temperature are subject to physical and prac-
tical constraints that in most situations prevent large intervals. Therefore, a small
degree of overlapping is not uncommon.

4. The large number of constraints (compared with the pinch location method) is
only a minor problem, for two main reasons:(1) all the constraints are linear, and
modern solvers can deal very efficiently with very large sets of linear equations;
(2) the preprocessing step in any modern solver drastically reduces the number
of equations and variables that the model must effectively solve. For example, in
test problem 3 see Section 4.3.4 the problem size is reduced from 5375 equa-
tions and 2125 variables to 675 equations and 283 variables: a reduction in size
of more than 85 %. The examples show that the effective number of equations
and variables in both the pinch location and proposed model are on the same
order of magnitude.

It is also of interest to show how the model can be extended to take into account
some situations that were not specifically addressed before: isothermal streams, multi-
ple utilities or forbid-den matches.

Isothermal streams. The easiest way to deal with these streams is to assume a small
fictitious temperature difference between the inlet and outlet temperatures (say 1,
0.1, 0.01C) and to calculate the correct heat flow rate. Alternatively, it is possible to
define a temperature interval of T = 0C, in which only the isothermal stream can
exchange heat. This requires changing the equations for the heat exchange of the
affected streams.

202
Captulo 4. Resultados

Multiple utilities. If the utility is an isothermal stream (e.g. HP, MP, LP saturated steam),
it is possible to introduce the utility by using either of the two approaches for isother-
mal streams mentioned in the paragraph immediately above; however, the objective
function must now take into account the cost of the utility. Note that the variable will
be the total heat exchanged by the utility, or the heat flow rate (mass flow rate if the
heat capacity is constant). In any case, the model continues to be linear. If the utility is
a non-isothermal stream (e.g. refrigeration water) with fixed inlet and outlet tempera-
tures, it can be introduced as any other stream, but in this case the temperatures are
fixed and the heat flow rate is variable. Again, linearity is retained.

Forbidden matches. In the model presented, it is implicitly assumed that all the hot
streams can exchange heat with all the cold streams. The implementation of forbid-
den matches is straightforward simply by changing the domain of equations involving
heat exchange (i.e. by defining a set of permitted matches and defining the equa-
tions over this set).

Both the pinch location method and the proposed model assume that the pinch
point can be located at any of the inlet temperatures of the process streams. This is
always true only if the heat flow rate in any stream is assumed to be constant, or vi-
ceversa: the inlet and outlet temperatures are constant and the heat flow rate can
change. However, if both the temperatures and flows can vary at the same time, we
cannot be sure that the pinch point will be located at an inlet temperature (e.g. when
the heat capacity depends on the temperature). A typical approach to overcome
this problem is to use a piecewise linear approximation, but if there is a complex rela-
tionship between the heat flow rate and temperatures it is not always possible to use
such an approach; consequently,both models must be used with caution.

4.3.4. Examples and results

In this section, a number of different examples are presented.They include problems


involving: fixed and variable stream temperatures (MILP); heat integration of distillation
columns, in which the temperature of the streams depends on the operating condi-
tions of the columns (MINLP); and flowsheet process optimization and heat integration
using the hybrid simulation optimization approach, in which the process is solved by

203
4.3.Alternative Model for heat integration

a commercial pro-cess simulator (ASPEN-HYSYS), and the heat integration model is in


equation form.

All calculations were performed on GAMS (REFERENCIA MCCARL, 2010) and MATLAB-
TOMLAB (REFERENCIA HOLMSTROM, 1999), while simulations were per-formed on ASPEN-
HYSYS.

4.3.4.1. Process with fixed streams conditions (MILP)

First, in order to validate the model, we solve the easiest possible example, in which
temperatures (both inlet and outlet) and heat flow rates are known and constant (see
Table 4.3.1). The objective is to determine the minimum utilities cost (min CH QHot +
CC QCold ).From a mathematical point of view, this problem involves binary and con-
tinuous variables, and all the equations used in the model are linear; so this problem
must be solved as a mixed integer linear problem (MILP), which was done on GAMS-
CPLEX. Finally, to test the behavior of the model, we compare its results with those
obtained by the pinch location method according to the disjunctive implementation
proposed by Grossmann et al. [4].

Tabla 4.3.1: Test problem 1 (fixed temperatures)


Test problem 1: non-isothermal, 6 hot and 6 cold streams
Hot stream FCp (kW/C) Inlet T (C) Outlet T (C) Cold stream FCp (kW/C) Inlet T (C) Outlet T (C)
H1 1,00 280 100 C1 0,50 30 200
H2 3,00 200 80 C2 1,50 60 90
H3 1,00 220 150 C3 2,00 70 170
H4 2,00 210 90 C4 3,00 110 230
H5 1,00 250 180 C5 1,50 90 140
H6 2,00 270 120 C6 4,00 120 250
Price of steam: 80 $ kg/kW
Price of cooling water: 20 $ kg/kW.

Under these conditions the temperature intervals are fixed and can be calculated
a priori, and the problem becomes one of classical heat integration, which can be
solved either by hand (e.g. problem table by Linnhoff (REFERENCIA LINNHOFF, 1982;
[1]) or by using the transshipment problem pro-posed by Papoulias and Grossmann
[5]. The proposed model can capture this fact in the preprocessing stage, and then

204
Captulo 4. Resultados

all binary variables will be fixed. Therefore, Table 4.3.2 contains results which have been
obtained with the preprocessing stage deactivated.

Tabla 4.3.2: Results test problem 1 (fixed temperatures)

Results test problem 1


Non-isothermal Grossmann et al. (G.Y.K.) Present work
N equations 1622 6059
N variables 614 1169
N binary variables 432 900
CPU time (s)a 0,271 0,353
Iterations 0 0
Heating req. (kW/kg) 80,00 80,00
Cooling req. (kW/kg) 15,00 15,00
Optimal solution ($) 6700,00 6700,00
a Pentium Dual-Core E5300 2.60 GHz, using CPLEX 12.4.0.0 for MILP

Table 4.3.2 shows, for both methods described above, the solution to this example
and some parameters relevant to the problem.

As expected, the computational results show that the number of variables, in par-
ticular binary variables, is considerably larger in the proposed model versus the pinch
location method. However, as mentioned earlier, both models exhibit comparable
computational performance (see Table 4.3.2).

Finally, as a check on the proposed model, we find that both models obtain exactly
the same solution ($6700).

4.3.4.2. Process with variable stream conditions (MILP)

The following examples (test problems 2 and 3) illustrate the case in which there is
a range of inlet and outlet temperatures for hot and cold streams under the assum-
ption that this variation does not have an impact on process performance. In other
words,we have an extra degree of freedom of variable temperatures for the heat
integration that does affect the rest of the process. Whereas this is a valid approxima-
tion in many systems, our main interest here focuses on testing the affected part of the

205
4.3.Alternative Model for heat integration

model. All the problems were solved for a heat recovery temperature (HRAT) of 10K.
For comparison purposes, the test problems were also solved by the pinch location
method.

The main objective of these examples is to show how the over-lapping degree
in the possible range of inlet temperatures affects the models performance. As in
the previous case, the objective function minimizes the utility cost (min C HU QHU +
C CU QCU ). In all cases the heat flow rates are assumed to be constant (see Table
4.3.3).

Tabla 4.3.3: Test problems 24 (variable temperatures)


Test problem 2: non-isothermal, 3 hot and 3 cold streams
FCp Interval Inlet Interval Outlet FCp Interval Inlet Interval Outlet
Hot stream Cold stream
(kW/C) T (C) T (C) (kW/C) T (C) T (C)
H1 0,15 180 - 260 30 - 50 C1 0,20 15 - 135 170 - 190
H2 0,50 120 - 220 75 - 95 C2 0,30 110 - 190 225 - 235
H3 0,10 110 - 155 90 - 100 C3 0,15 70 - 130 140 - 150

Test problem 3: non-isothermal, 4 hot and 4 cold streams


FCp Interval Inlet Interval Outlet FCp Interval Inlet Interval Outlet
Hot stream Cold stream
(kW/C) T (C) T (C) (kW/C) T (C) T (C)
H1 0,15 230 - 260 30 - 50 C1 0,20 10 - 25 170 - 190
H2 0,50 135 - 155 110 - 150 C2 0,30 90 - 110 180 - 225
H3 0,25 80 - 100 20 - 30 C3 0,15 125 - 160 225 - 235
H4 0,30 110 - 120 80 - 100 C4 0,40 130 - 150 250 - 280

Test problem 4 (non lineal): non-isothermal, 3 hot and 3 cold streams


FCp Interval Inlet Interval Outlet FCp Interval Inlet Interval Outlet
Hot stream Cold stream
(kW/C) T (C) T (C) (kW/C) T (C) T (C)
H1 0,15 180 - 200 30 - 50 C1 0,20 15 - 25 170 - 190
H2 0,50 120 - 140 75 - 95 C2 0,30 110 - 140 225 - 235
H3 0,25 110 - 155 90 - 100 C3 0,15 70 - 100 140 - 150
Price of steam: 80 $ kg/kW
Price of cooling water: 20 $ kg/kW.

In test problem 2 the bounds of almost all the inlet temperatures overlap and the-
refore the preprocessing stage has a low impact on the models performance. In both
the pinch location method and the proposed approach the relaxation gap is really
bad (objective function equal to zero).

However, in test problem 3, the overlapping of the bounds of inlet temperatures is


moderate, see Fig. 4.3.5. The optimal solution is $4060 but the relaxation gap of the
proposed model is much better than that obtained by the pinch location method

206
Captulo 4. Resultados

(3124 vs 620).Results and relevant parameters for the test problems are shown in Tables
4.3.4 and 4.3.5, respectively.

Tabla 4.3.4: Results test problems 24 (variable temperatures)

G.Y.K. model Present work


FCp
Stream Opt. Inlet Opt. Outlet Opt. Inlet Opt. Outlet
(kW/C)
Temp. (C) Temp. (C) Temp. (C) Temp. (C)
Optimal solution test problem 2: non-isothermal, 3 hot and 3 cold streams
H1 0,15 260,00 50,00 260,00 50,00
H2 0,50 210,00 95,00 210,00 95,00
H3 0,10 110,00 100,00 110,00 100,00
C1 0,20 15,00 190,00 15,00 190,00
C2 0,30 110,00 225,00 110,00 225,00
C3 0,15 70,00 150,00 70,00 150,00
Optimal solution test problem 3: non-isothermal, 4 hot and 4 cold streams
H1 0,15 260,00 50,00 260,00 50,00
H2 0,50 155,00 120,50 155,00 126,50
H3 0,25 80,00 30,00 80,00 30,00
H4 0,30 110,00 100,00 120,00 100,00
C1 0,20 10,00 170,00 10,00 170,00
C2 0,30 90,00 180,00 90,00 180,00
C3 0,15 160,00 225,00 160,00 225,00
C4 0,40 150,00 250,00 150,00 250,00
Optimal solution test problem 4 (non lineal): non-isothermal, 3 hot and 3 cold streams
H1 0,15 196,00 41,50 196,00 41,50
H2 0,50 135,00 90,00 130,00 90,00
H3 0,10 131,50 94,75 136,50 94,75
C1 0,20 18,00 172,00 18,00 172,00
C2 0,30 137,00 225,00 137,00 225,00
C3 0,15 89,75 140,00 84,75 140,00

It is worth noticing the important reduction in the number of equations and varia-
bles in the preprocessing stage: the problem size is reduced from 5375 equations and
2125 variables to 675equations and 283 variables. This represents a reduction in size of
more than 85 %.

Any modification to the operating conditions of a process that is running optimally


with respect to a given objective function, intro-duces a penalty into that function. Test
problem 4 tries to capture in a simple way the tradeoff between improvements due

207
4.3.Alternative Model for heat integration

Tabla 4.3.5: Parameters solution test problems 24)


Test Problem 2 Test Problem 3 Test Problem 4
Results
G.Y.K. model Present work G.Y.K. model Present work G.Y.K. model Present work
Initial Problem
N equations 416 3047 730 5375 416 3047
N variables 171 1247 291 2125 171 1247
N binary variables 108 216 192 307 108 169

Reduced MIP problem


N equations 225 839 142 675 92 416
N variables 92 329 68 283 46 174
CPU time (s)a 0,453 0,749 0,108 0,218 0,678 3,332
Heating req. (kW/kg) 0,00 0,00 49,50 49,50 28,900 29,250
Cooling req. (kW/kg) 8,50 8,50 5,00 5,00 12,763 11,113

Optimal solution ($) 170,00 170,00 4060,00 4060,00 2918,625 2903,625


Solution of relaxed problem 0,00 0,00 620,00 3124,02 767,000 1904,725
GAP ( %) 100,00 100,00 84,7 23,1 73,7 34,4
a Pentium Dual-Core E5300 2.60 GHz, using CPLEX 12.4.0.0 for MILP

to energy savings and deviations from the optimal point of the process when heat
integration is not considered. To this end, we use a set of values for the inlet and outlet
temperatures that corresponds to temperatures at the optimal operating conditions
when heat integration is not taken into account. The objective function then consists
of two terms; the first one pertains to the cost of utilities, and the second one penalizes
the deviation of temperature from a given set value:

X
min : Cost of utilities + ((Tkin T Mkin )2 + (Tkout T Mkout ))2 (4.3.58)
kST

where the TM are the optimal temperatures of the non-heat integrated process
(here we have taken the central value between the upper and lower bounds of the
inlet and outlet temperatures).

Again, the overlapping degree in this example is moderate;preprocessing and lo-


gical relationships reduce the relaxation gap as compared to the pinch location met-
hod (1904 vs 767). Data,results and other relevant parameters for this example are
shown in Tables 4.3.3-4.3.5, respectively.

208
Captulo 4. Resultados

4.3.4.3. Heat integration of distillation columns (MINLP)

This example corresponds to the general case in which the heat integration cannot
be isolated from the rest of the process, and therefore the temperatures of the streams
depend on the remaining operating conditions for the process. Here, we studied the
heat integration of a distillation column sequence: in particular, the separation of a
mixture of 4 components using sharp distillation(Fig. 4.3.6). In this case, the temperature
of hot and cold streams depends on the pressure inside each column and on the
purity specifications.

Figura 4.3.6: Sharp distillation for separation of a 4-component mixture

We assume that design of the columns is not the main objective, and that the in-
vestment cost is not significantly changed with respect to optimal operation without
heat integration. In that way we can focus on the energy integration, and at the same
time simplify the problem. The sharp distillation has as main feature that the compo-
nents of the mixture are essentially separated completely between distillate and bot-
toms. In a sharp distillation, we know in advance what the composition of the products
in each column is. Using this property, we study the temperature range of the various
distillates and bottoms expected in each column as a function of the column pressure
(pressure range: 13 atm), by means of a chemical process simulator (ASPEN-HYSYS).
Then, the temperature of each stream is fitted to a quadratic polynomial in terms of
column pressure, which is the expression we finally used in the model. At the same
time, the heat flow rate (F CP ) is not constant and depends on the column pressure;

209
4.3.Alternative Model for heat integration

in this case a linear relationship is assumed. The resulting problem is a MINLP problem,
solved on GAMS-DICOPT.

As an example, we consider the separation of 200 kmol/h of an equimolar mixture


of benzene, toluene, p-xylene and -methyl-styrene (molar fraction of 0.25 benzene,
0.25 toluene, 0.25 p-xylene and 0.25 -methyl-styrene). We assume that all the ben-
zene and toluene is obtained in the distillate of column 1, while all the p-xylene and
-methyl-styrene are obtained in the bottoms of column 1. The benzene and tolue-
ne are obtained in the distillate and bottoms of column 2, respectively. Finally, the
p-xylene and -methyl-styrene are obtained in the distillate and bottoms of column
3, respectively. The flows and temperatures of the hot and cold streams are shown in
Table 4.3.6.

Tabla 4.3.6: Flows and temperatures of problem columns

Problem columns: pressure range: 13 atm


Stream FCp (kW/C) Inlet Temp (C) Outlet Temp (C)
Column 1 H1 (Condenser) 292,95 - 381,18 95,95 - 136,07 89,46 - 130,08
C1 (Reboiler) 441,11 - 562,02 157,67 - 203,36 162,34 - 207,72

Column 2 H2 (Condenser) 1102,01 - 1242,24 77,47 - 116,30 76,47 - 115,30


C2 (Reboiler) 1121,98 - 1263,15 117,64 - 159,51 118,64 - 160,51

Column 2 H3 (Condenser) 1621,92 - 1812,95 136,18 - 180,17 135,18 - 179,17


C3 (Reboiler) 1642,63 - 1834,28 173,60 - 220,67 174,60 - 221,67
Price of steam: 80 $ kg/kW
Price of cooling water: 20 $ kg/kW.

Table 4.3.7 collects the solution to this example and some relevant information re-
garding the computational results.

The CPU times taken by both models are on the same order of magnitude. This
implies that both models are of comparable complexity and perform in a similar way.
Moreover, the results that have been obtained are different. The main reason for this is
that the problem is nonlinear and non-convex, and therefore a global optimal solution
cannot be guaranteed. However, the optimal solutions that have been obtained are

210
Captulo 4. Resultados

Tabla 4.3.7: Solution of problem columns

Results problem columns


G.Y.K. model Present work)
N equation 467 3068
N variables 232 1262
N binary variables 108 207
CPU time (s)a 0,81 1,50

Columns
pressure (atm) C1: 1,000 C1: 1,000
C2: 1,893 C2: 1,000
C3: 1,919 C3: 1,926

Heating req. (kW/kg) 3267,36 3204,36


Cooling req. (kW/kg) 3067.30 3004,72
Optimal solution ($) 322735,01 316443,52
a Pentium Dual-Core E5300 2.60 GHz, using CPLEX 12.4.0.0 for MILP

quite close. From the point of view of relaxation, in this case both models perform
poorly (objective function value equal to zero in both cases).

4.3.4.4. Hybrid simulation-optimization process (MINLP)

Finally, we considered a problem in which mathematical programming with the


explicit equations is combined with the robust models that are available in commer-
cial chemical process simulators. The process is implemented in the simulator and the
resulting system is interfaced with MATLABvia ActiveX controls; it is optimized using
MATLAB-TOMLABand a proprietary implementation of the outer approximation al-
gorithm [27-29] that has been adapted to MATLAB. The problem is a mixed-integer
nonlinear programming (MINLP) problem in which some of the equations appear in
an implicit form (equations in the process simulator), while the others appear explicitly.

Heat integration of the problem proposed by Seider, Seader, and Lewin [30] was
selected as case study of a problem of this type(see Fig. 4.3.7). A natural gas stream

211
4.3.Alternative Model for heat integration

is processed at 5000 kmol/h, 20C, 10 atm, and with the composition shown in Table
4.3.8. The gaseous product is required to be at 20 atm, with at least 4900 kmol/h of
nC4 and lighter species with a combined mole percentage of at least 99.5 %.

Figura 4.3.7: Process flow diagram for the processing of the natural gas

Tabla 4.3.8: Molar Flow rate of the feed (kmol/h)

Component Feed (kmol/h)


N2 105,50
C1 4139,00
C2 435,50
C3 205,50
nC4 70,50
nC5 28,50
nC6 16,50

Total 5000,00

In this example the process simulator, HYSYS, performs the calculation at the flows-
heet level, including all mass and energy balances and all estimations of properties.
As in the previous example,to avoid the behavior of the process interfering with the
performance of the energy integration model, we assume that the cost of the pro-
cess is not substantially affected; as a result, it is not taken into account. Therefore, we
only perform heat integration of the process, and because of this, the objective of
this problem is to minimize the heat supplied by the hot and cold utilities. The streams

212
Captulo 4. Resultados

affected by the heat integration were all inlet and outlet streams of the heat exchan-
gers. The temperature bounds for all streams, the main constraints, the optimal solution
and some information on the procedure used to obtain the solution are shown in Table
4.3.9.

Tabla 4.3.9: Bounds, constraints and solution of the natural gas problem

Stream Problem limits Solution Temp


Temp range (K)
(K)
H1 In HExc1 361 - 361 361
Out HExc1 303 - 343 343
H2 In HExc2 303 - 343 343
Out HExc2 203 - 253 223,10
H3 In HExc3 203 - 253 223,10
Out HExc3 273 - 333 333
H4 In HExc4 203 - 253 223,10
Out HExc4 273 - 333 329,71

Restrictions
Molar flow light product (kmol/h) > 4930 4954,67
Molar frac. (nC4 + lighter) in light product > 0,995 0,998
Molar flow heavy product (kmol/h) > 35 45,33
Molar frac. (nC5 + nC6) in heavy product > 0,75 0,75

Solution parameters
N equations 290
N variables 195
N binary variables 116
N iterations 68

Heat solution
Heating req. (kW) 0,0
Cooling req. (kW) 1740,2
Optimal solution (KW) 1740,2

213
4.3.Alternative Model for heat integration

The optimal solution satisfies all the proposed constraints. Furthermore, the heat
integration of the system eliminates the need for hot utilities; only cold utilities are nee-
ded to satisfy the requirements of the process. The CPU time is not displayed in Table
4.3.9 because the simulator uses up most of the time to converge the system, and
therefore this parameter is not representative of the model.

In this example, there is little overlapping among the inlet temperatures of the pro-
cess streams. Both models reach the optimal solution in the relaxed MINLP problem.

This example shows that the proposed approach is robust and flexible and can be
implemented by adding some equations to any existing model with minor modifica-
tions to the original model.

4.3.5. Conclusions

An alternative model has been proposed for the simultaneous optimization and
energy integration of non-isothermal systems.Even though this results in a larger pro-
blem than is typically encountered in other models, a preprocessing stage for the
data and the inclusion of some logical relationships substantially reduces the size of
the initial problem. In fact, this alternative model competes successfully with the dis-
junctive implementation of the pinch location method by Grossmann et al. [4] from
the point of view of CPU time and relaxation gap.

The performance of the model is directly related to the overlap-ping degree of


the bounds of the inlet temperatures for the process streams. With moderate to small
overlapping the proposed model improves the relaxation gap with respect to the dis-
junctive implementation of the pinch location method.

In addition, the alternative model manages to maintain the temperature interval


concept in contrast to existing models. This feature is interesting to some engineers and
researchers that are familiar with the classical concept of temperature interval in heat
integration. Finally, the proposed model eliminates the numerical difficulties associated
with the use of the max function or smoothing approximation function, and therefore
provides a robust alternative model.

214
Captulo 4. Resultados

Acknowledgements

The authors gratefully acknowledge financial support from the Spanish Ministerio

de Ciencia e Innovacion under project CTQ2012-37039-C02-02.

Referencias

[1] Linnhoff, B. y Flower, J.R. ((Synthesis of heat exchanger networks: I. Systematic generation
of energy optimal networks)). En: AIChE Journal 24.4 (1978), pags.
633-642.
[2] Morar, M. y Agachi, P.S. ((Review: Important contributions in development and improve-
ment of the heat integration techniques)). En: Computers & Chemical Engineering 34.8

(2010), pags. 1171 -1179.
[3] Duran, M.A. y Grossmann, I.E. ((Simultaneous optimization and heat integration of chemical
processes)). En: AIChE Journal 32.1 (1986), pags.
123-138.
[4] Grossmann, I.E., Yeomans, H. y Kravanja, Z. ((A rigorous disjunctive optimization model for
simultaneous flowsheet optimization and heat integration)). En: Computers & Chemical
Engineering 22, Supplement 1.0 (1998), S157 -S164.
[5] Papoulias, S.A. y Grossmann, I.E. ((A structural optimization approach in process synthesis
II: Heat recovery networks)). En: Computers & Chemical Engineering 7.6 (1983), pags.
707
-721.
[6] Douglas, J. M. ((A hierarchical decision procedure for process synthesis)). En: AIChE Journal

31.3 (1985), pags. 353-362.
[7] Douglas, J.M. Conceptual design of chemical processes. McGraw-Hill chemical enginee-
ring series. McGraw-Hill, 1988.
[8] Smith, R. Chemical process design and integration. Wiley, 2005.
[9] Smith, R. y Linnhoff, B. ((The design of separators in the context of overall processes)). En:
Chemical Engineering Research and Design 66 (1988), pags. 195-228.
[10] Linnhoff, B. e Hindmarsh, E. ((The pinch design method for heat exchanger networks)). En:

Chemical Engineering Science 38.5 (1983), pags. 745 -763.
[11] Furman, K.C. y Sahinidis, N.V. ((A Critical Review and Annotated Bibliography for Heat Ex-
changer Network Synthesis in the 20th Century)). En: Industrial & Engineering Chemistry

Research 41.10 (2002), pags. 2335-2370.
[12] Ahmad, S., Linnhoff, B. y Smith, R. ((Cost optimum heat exchanger networks-2. targets and
design for detailed capital cost models)). En: Computers & Chemical Engineering 14.7

(1990), pags. 751 -767.
[13] Linnhoff, B. y Ahmad, S. ((Cost optimum heat exchanger networks-1. Minimum energy and
capital using simple models for capital cost)). En: Computers & Chemical Engineering 14.7

(1990), pags. 729 -750.
[14] Linnhoff, B. y Flower, J.R. ((Synthesis of heat exchanger networks: II. Evolutionary generation
of networks with various criteria of optimality)). En: AIChE Journal 24.4 (1978), pags.
642-654.

215
4.3.REFERENCIAS

[15] Cerda, J. y Westerburg, A.W. ((Synthesizing heat exchanger networks having restricted
stream / stream matches using transportation problem formulations)). En: Chemical En-

gineering Science 38.10 (1983), pags. 1723 -1740.
[16] Floudas, C.A., Ciric, A.R. y Grossmann, I.E. ((Automatic synthesis of optimum heat exchan-
ger network configurations)). En: AIChE Journal 32.2 (1986), pags.
276-290.
[17] Biegler, L.T., Grossmann, I.E. y Westerberg, A.W. Systematic methods of chemical process
design. Prentice-Hall international series in the physical and chemical engineering scien-
ces. Prentice Hall PTR, 1997.
[18] Floudas, C.A. y Ciric, A.R. ((Strategies for overcoming uncertainties in heat exchanger net-
work synthesis)). En: Computers & Chemical Engineering 13.10 (1989), pags. 1133 -1152.
[19] Ciric, A.R. y Floudas, C.A. ((Heat exchanger network synthesis without decomposition)). En:

Computers & Chemical Engineering 15.6 (1991), pags. 385 -396.
[20] Yee, T.F., Grossmann, I.E. y Kravanja, Z. ((Simultaneous optimization models for heat integra-
tion - I. Area and energy targeting and modeling of multi-stream exchangers)). En: Com-
puters & Chemical Engineering 14.10 (1990), pags. 1151 -1164.
[21] Yee, T.F. y Grossmann, I.E. ((Simultaneous optimization models for heat integration - II. Heat
exchanger network synthesis)). En: Comput. Chem. Eng. 14.10 (1990), pags. 1165 -1184.
[22] Yee, T.F., Grossmann, I.E. y Kravanja, Z. ((Simultaneous optimization models for heat integra-
tion - III. Process and heat exchanger network optimization)). En: Computers & Chemical

Engineering 14.11 (1990), pags. 1185 -1200.
[23] Floudas, C.A. y Paules IV, G.E. ((A mixed-integer nonlinear programming formulation for the
synthesis of heat-integrated distillation sequences)). En: Computers & Chemical Enginee-

ring 12.6 (1988), pags. 531 -546.
[24] Raman, R. y Grossmann, I.E. ((Symbolic integration of logic in mixed-integer linear program-
ming techniques for process synthesis)). En: Computers & Chemical Engineering 17.9 (1993),

pags. 909 -927.
[25] Caballero, J.A. y Grossmann, I.E. ((Structural Considerations and Modeling in the Synthesis
of Heat-IntegratedaThermally Coupled Distillation Sequences)). En: Industrial & Enginee-

ring Chemistry Research 45.25 (2006), pags. 8454-8474.
[26] Caballero, J.A. y Grossmann, I.E. ((Aggregated Models for Integrated Distillation Systems)).

En: Industrial & Engineering Chemistry Research 38.6 (1999), pags. 2330-2344.
[27] Duran, M.A. y Grossmann, I.E. ((An outer-approximation algorithm for a class of mixed-
integer nonlinear programs)). En: Mathematical Programming 36 (3 1986), pags.
307-339.
[28] Kocis, G.R. y Grossmann, I.E. ((Relaxation strategy for the structural optimization of process
flowsheets)). En: Industrial & Engineering Chemistry Research 26.9 (1987), pags.
1869-1880.
[29] Viswanathan, J. y Grossmann, I.E. ((A combined penalty function and outer-approximation
method for {MINLP} optimization)). En: Computers & Chemical Engineering 14.7 (1990),

pags. 769 -782.
[30] Seider, W.D., Seader, J.D. y Lewin, D.R. Process design principles: synthesis, analysis, and
evaluation. Wiley, 1999.

216
Captulo 4. Resultados

4.4 Anexo I: Integration of different models in the design of


chemical processes: Application to the design of a Power
Plant

Debido a los avances en la sntesis y el dise


no de los procesos qumicos, son cada
necesarios modelos matematicos
vez mas complejos con el objetivo de mejorar
mas
de nuevas alternativas de forma precisa y fiable. Gene-
la capacidad de deteccion

ralmente y como consecuencia de la complejidad de los modelos no es practico, e

incluso no es posible escribir un modelo matematico que cada vez que se tiene que
dise
realizar una nueva simulacion, no o sntesis de un proceso. Por lo tanto, la mejor
radica en poder utilizar el mejor modelo matematico
solucion disponible en cada uno
de los problemas.

Los avances en el dise y sntesis de procesos qumicos en los ulti-


no, optimizacion
mos a
nos han sido impresionantes en todos los niveles, desde operaciones unitarias
del diagrama de flujo completo. Sin embargo, la
individuales hasta la optimizacion
mayora de estos dise
nos se basan en metodos aproximados o, en ciertas suposicio-
nes que deben ser verificadas mediante un simulador de procesos qumicos. En al-
gunas situaciones, los modelos aproximados no son adecuados ni lo suficientemente
relevantes del proceso real.
precisos para captar los aspectos mas

en un proceso de dise
En cualquier caso, la mejor opcion y utili-
no es la seleccion
del mejor modelo disponible venga de donde venga. En funcion
zacion de su origen
y sus caractersticas, los modelos seleccionados pueden ser:

Modelos en forma de ecuaciones (con o sin variables enteras). En estas con-


disponibles, o son faciles
diciones, las primeras y segundas derivadas estan de
obtener

Modelos definidos por una estructura de entrada y salida sin acceso interno
numerica
al modelo, aunque la informacion de las derivadas (por lo menos
de las variables de
las primeras derivadas) se encuentra disponible, en funcion
entrada

217
4.4.Integration of different models in design of chemical processes

Modelos en forma de caja negra con una estructura rgida entrada - salida,
de las derivadas no esta
donde la informacion disponible. En este caso se pue-
de las derivadas puede ser calcula-
de diferenciar dos casos: si la informacion
por perturbaciones, o si el modelo introduce ruido numerico
da con precision

Modelos demasiado complejos para ser incluido directamente en cualquier


o problema de sntesis
optimizacion

El presente trabajo presenta una estructura de modelado para tratar problemas


de sntesis, que incluyen diferentes modelos con algunas de las caractersticas ante-
matematica.
riormente mencionadas utilizando un marco de programacion En parti-
disyuntiva como una estructura
cular, en este trabajo se selecciona la programacion
adecuada para hacer frente a este tipo de problemas, ya que permite encapsular
de sus caractersticas. La es-
cada modelo y utilizar un enfoque diferente en funcion
tructura llevada a cabo utiliza disyunciones para capturar cada uno de los modelos
de tal forma que se aprovechan las ventajas o minimizan los inconvenientes de cada
dispo-
uno. De esta manera es posible utilizar las derivadas algebraicas cuando estan

nibles y simultaneamente
implementar metodos robustos para estimar las derivadas
de otros modelos sin la necesidad de interferir con el resto de modelos, y de esta for-

ma mejorar el tiempo de calculo computacional. La conectividad entre los diferentes
El modelo se resuelve mediante algoritmos
modelos se realiza en forma de ecuacion.

basados en la logica sin reformularlo como un problema MINLP.

de vapor
A modo de ejemplo se presenta la sntesis de un planta de generacion

en la que sus diferentes componentes se simulan con modulos que presentan diferen-

te comportamiento numerico. El objetivo de este trabajo es mostrar que es posible
sintetizar un sistema complejo con diferentes modelos de manera eficiente.

El resultado final es un sistema hbrido que incluye modelos implcitos procedentes



de diferentes fuentes y con diferentes comportamientos numericos. se de-
Ademas
muestra que el problema de sntesis se puede escribir como un problema GDP y resol-

verse sin reformularlo a MINLP. Conceptualmente, el modelado GDP permite una facil
a nivel de usuario y, al mismo tiempo al encapsular cada sub-modelo
formulacion

facilita su estudio numerico.

218
Captulo 4. Resultados

Integration of different models in the design of chemical processes:


Application to the design of a Power Plant

A. Caballero, Miguel A. Navarro-Amoros,


Jose Ruben
Ruiz-Femenia, Ignacio E. Grossmann

Article sent to Applied Energy. Under Review

Abstract

Despite the widely availability of sophisticated tools for simulation, optimization and
synthesis of chemical processes, the user is frequently interested in using the best avai-
lable model that can come from different sources and is usually in form of a black
box with a rigid input-output structure. In this paper we propose to deal with all these
models using generalized disjunctive programming to capture the numerical charac-
teristics of each model (in equation form, modular, noisy, etc.) and to deal with each
of them according to its characteristics. The result is a hybrid modular -equation based
approach that allows synthesizing complex process using different models in a robust
and reliable way. The methodology is applied to the synthesis of a power plant for gi-
ven steam and power demand. Different instances of a case study are presented to
illustrate the capabilities of the proposed approach.

Keywords

Process synthesis, Generalized Disjunctive Programming, Utility Systems, Modular 23


Optimization, kriging.

4.4.1. Introduction

With advances in the synthesis and design of chemical processes there is an in-
creasing need for more complex mathematical models for screening alternatives with
accurate and reliable process models. Due to the complexity of the models it is not
practical, and perhaps not even possible to write a mathematical model each time
we need to use it for a new simulation, design or synthesis problem. Instead we would
like to reuse the best available mathematical model [1].

219
4.4.Integration of different models in design of chemical processes

The model of a chemical plant can be theoretically represented by a system of


nonlinear algebraic equations of the form:

f (x) = 0 (4.4.1)

where f is a vector of functions and x is a vector of variables. The variables represent


flow rates, compositions, temperatures, pressures, etc., and the functions are obtained
from physical and chemical principles concerning conservation of mass and energy,
chemical equilibrium, kinetics and transport phenomena. Modeling a chemical plant
can involve hundreds of thousands of equations and variables but be extremely spar-
se. Typically, each equation involves only around 4 to 10 variables. Sophisticated large-
scale numerical solvers can exploit the sparsity pattern in a very effective way. In so-
me cases it is possible to write and solve the complete set of equations directly using
general modeling systems (e.g. GAMS [2] , AMPL [3]) or chemical engineering orien-
ted modeling systems (e.g. ASCEND, gPROMS) that include databases of chemical
properties and thermodynamics. As the model becomes more complex more spe-
cialized knowledge is required to for example provide good initializations, and avoid
physically meaningless solutions. But even in this case, it is not uncommon that the
designer includes existing models that come from different sources and with different
numerical behavior (e.g. a third party model of a unit operation, a surrogate model
from experimental results, or a model from computational fluid dynamics).

We can consider four types of problems [4, 5] depending on the data specified
and the final objective of the model user:

In a Simulation Problem, the feeds and design variables of each unit must be spe-
cified. The unknowns are the variables related with the additional (product) streams.
They usually have a rigid input output structure, but at the same time are robust and
reliable.

A Design Problem is similar to the simulation problem, but some of the design varia-
bles (i.e. reactor volume; number of trays in a distillation column, etc) are unspecified.

220
Captulo 4. Resultados

Some constraints are then imposed on some of the stream variables to satisfy the extra
degrees of freedom.

In an Optimization Problem some variables associated with the feed and equip-
ment design can be left unspecified, in this case a performance function must be
added to the model defined by equation 4.4.1. Equality and inequality constraints
may be added to the model.

In a Synthesis Problem besides the optimal operating conditions (feed and design
variables) we are interested in which is the best configuration from a structural point
of view (combination of unit operation or technologies) for a given objective. We ha-
ve the added difficulty of solving a problem in which the set of equations change
depending on the selected equipment.

From a formal point of view including different sub-models from different sources is
equivalent to divide the original model given by equation 4.4.1 in smaller sub-problems
[5]:

fi (x) = 0, i = 1, 2, . . . , n (4.4.2)

where n is the number of sub-problems in which the original problem is split. This is
basically the idea followed by modular chemical process simulators in which a given
unit operation is represented by one of the subsets in equation 4.4.2. The advantages
of the modular approach are:

Different sub-modules can be prepared and tested separately.

The solution methods can be specifically designed for that module, e.g. the
model of a distillation column or a complex reactor. Therefore, the module is
robust and reliable.

Because of the rigid requirements, data can be easily checked for consistency
and completeness.

221
4.4.Integration of different models in design of chemical processes

The modular structure allows adding new modules easily.

Due to these advantages, it is not surprising that modular simulation is still the domi-
nant approach. However, when we move to design, optimization or synthesis problems
the modular approach loses some of its attractiveness. The straightforward approach
consists of performing simulations trying to satisfy the design or optimization objectives.
However, repeated solutions of the simulation rapidly lead to long computer times.
Therefore, the design and optimization is usually performed in an equation based en-
vironment (all equations solved simultaneously) using general modeling systems or field
specific modeling systems [6]. In the case of synthesis the model takes the form of a
Mixed-integer (Non)Linear Programming problem [4, 7, 8] (MINLP) where discrete de-
cisions are related to integer (binary) variables or a Generalized Disjunctive Program-
ming Problem (GDP) [9].

Developments in the design, optimization and synthesis of chemical processes for


the last years have been impressive at all the levels from individual unit operations,
subsystems, and complete flowsheet optimization (See for instance the following books
[7, 10-13]). However, most of these designs rely on shortcut or aggregated methods
[14] or in some assumptions that must be verified using a chemical process simulator.

In some situations the shortcut model are neither adequate nor accurate enough
to capture the more relevant aspects of the actual process. In other cases, the best
models available come from third party modelers and could have very different cha-
racteristics and numerical behavior, and usually in the form of black-boxes in which
the final user has no access to the original code. For example a legacy code that
contains classified or proprietary code, a complex model from computational fluid
dynamics, or simply a unit operation in a commercial process simulator. Usually these
models have been tested and validated so they are trusted and of course the desig-
ner is interested in use them whenever possible.

Trying to use chemical process simulators as an implicit model for solving synthe-
sis problems through a MINLP approach is not something new. Harsh et al [15] de-
veloped an interface with a MINLP and FLOWTRAN, for the retrofit of an ammonia
process. Diwekar et al [16], presented a process synthesizer using Aspen Plus. They pre-
sent some small problems and the structural optimization of the hydrodealquilation of

222
Captulo 4. Resultados

toluene process. Their algorithm is basically an implementation of the modeling and


decompositions strategy [17]. Reneaume et al. [18] point out that in a given constraint
h(d, s) 0 (d is a vector of decision variables and s are variables of interest calcula-
ted by the process simulator) the s variables depend implicitly on the d variables. But
this implicit function varies depending on what are the structural decisions. In other
words, s depends on the structural decisions and therefore the linearization of a given
constraint (e.g. in the construction of the MILP Master Problem) can lead to the linea-
rization of different functions, and consequently failure of the algorithm. They solved
the problem by adding pseudo-torn streams whose function is explicitly separate the
dependency of s variables from the structural decisions. Diaz & Bandoni [19] use a
process simulator specifically designed for ethylene plants, SISER [20], for the structural
design of an ethylene plant using a combination of different types of models (rigorous
and simplified) that includes correlations and results checking with an actual ethylene
plant. Caballero et al. [21, 22] proposed a specific algorithm for the rigorous design of
distillation columns, combining process simulators and a modified version of the outer
approximation algorithm [23-25]. Latter Brunet el al. [26] used that algorithm in the op-
timization of distillation columns in ammonia water absorption cooling cycle. They also
extended the approach for considering Life Cycle Assessment (LCA) using multiobjec-
tive optimization in which the topology in the process simulator remains unchanged
(except by the optimization of distillation columns).

In all the works previously mentioned, the complete model is solved with a process
simulator interfaced with a MINLP solver. Caballero et al [27] presented and hybrid
approach for the design of hybrid distillation vapor membrane separation systems, in
which the differential and algebraic equations of the membrane modules are calcu-
lated in equation form. Caballero & Grossmann [28] presented a detailed modeling
framework that combines a process simulator with complex algebraic (equation ba-
sed) models involving both continuous and discrete variables, although the topology
of the flowsheet was not modified. A similar approach was followed by Brunet el al.
[29, 30] who extended the methodology to consider multiobjective optimization by
including LCA in biotechnological processes in which the reactions have complex ki-
netics that cannot be solved by the process simulator. Finally, we should do note that
if we have access to the code for all the external modules, it is possible to automate a

223
4.4.Integration of different models in design of chemical processes

set of procedures that generate code for performing operations like automatic diffe-
rentiation, sparsity pattern determination or discontinuity function evaluation. Tolsma
et al [31] incorporated these procedures in the modeling system ABACUS II.

It is worth comment that deterministic optimization methods, like the approach


proposed in this paper, are not the only alternative for dealing with these problems.
Stochastic methods have proved to be a good alternative for solving hybrid simu-
lation optimization problems. Although there is a vast literature about metaheuristic
optimization, the combination with chemical process simulators is relatively recent. So-
me examples include: Dantus and High [32] who used Aspen Pluswith a simulated
annealing for multi-objective optimization under uncertainty for waste minimization.
Leboreiro & Acevedo [33] presented a genetic algorithm framework for the synthesis
of distillation sequences using process simulators. Gutierrez-Antonio & Briones-Ramirez
[34, 35] presented a multi-objective optimization approach for determining the Pareto
front (minimum number of stages vs. minimum reflux ratio) for Petlyuk columns using
genetic algorithms and the Aspen Plus chemical process simulator. Latter the met-
hodology was extended to more complex distillation sequences [36]. Odjo et al. [37],
presented a general framework for the synthesis of chemical processes using an Hybrid
approach based on simulation and genetic algorithms. Eslick and Miller [38] used an
integrated approach that included Aspen plus and spreadsheets models for the multi-
objective retrofit of a coal power plant with CO2 capture and compression process.
Torres et al [39] presented an automatic environmental and evaluation methodology
for the hybrid simulation-synthesis of chemical processes. Latter they use the tool for
the environmental and economic optimization of a complex sour water plant using
direct search and genetic algorithms [40].

The contributions of hybrid simulation-optimization systems using deterministic opti-


mization are no so extensive. The reasons are obvious, the derivative free optimization
use more or less complex strategies for reaching the optimum of a performance fun-
ction by repeated evaluations (simulations) of a model, and in general they are easy
to implement or widely available. However, they present some major drawbacks: the
number of function evaluations can be large, and therefore if the number of degrees
of freedom is large or the time to solve a flowsheet is large (i.e. complex recycle struc-
ture) the computational time becomes prohibitive. It is not easy to deal with external

224
Captulo 4. Resultados

constrains, in particular inequalities in terms of dependent variables. A detailed re-


view of techniques for dealing with constraints in metaheuristics can be found in the
following references [41-44].

In any case, the designer would like to use the best available model wherever it co-
mes from. Depending on their origin and its characteristics, for this paper the following
cases are relevant:

The model is in equation form (with or without integer variables). In these condi-
tions first and second derivatives are available (or easy to obtain)

We have not access to the model that comes with an input output structure,
but numerical derivative information (at least first derivatives) in terms of input
variables is available. This is the case of some thermodynamic packages, which
calculate the properties and in most case the derivatives. Or models in which
automatic differentiation can be used [31]

The model is a black box with a rigid input - output structure. Derivative informa-
tion is not available. Here, it is convenient to differentiate two cases: whether
derivative information can be accurately calculated by perturbation of the in-
put or design variables, or if the model introduces some numerical noise. This is
the case of models in a modular chemical process simulator. All the units are
accurate enough for simulation purposes, but some of them introduce some
numerical noise (e.g. some reactors or distillation columns) that prevent the ac-
curate determination of derivative information [45, 46]

The model is too complex to be directly included in any optimization or synthesis


problem. This is the case, for example, of computational fluid dynamics models
that can take hours of CPU time

In this paper we present a modeling framework for dealing with synthesis problems
that include models with some of the previous mentioned characteristics using a mat-
hematical programming framework. In particular, we believe that disjunctive program-
ming is a framework that is very well suited for dealing with these kinds of problems be-
cause it allows .encapsulating.each model and use a different approach depending

225
4.4.Integration of different models in design of chemical processes

of its characteristics. A model environment was developed that captures each of the
models through disjunctions and takes advantage -or minimizes the drawbacks- of
each model. In that way it is possible to use algebraic derivatives when they are avai-
lable (or automatic differentiation), implement robust methods for estimating derivati-
ves for some sub-models without interfering with the rest of the models, and then sa-
ving computational time. The connectivity between those models is in equation form
to avoid the implicit relation between design and calculated variables [18]. Logical
relationships including strong relations between alternatives (i.e. ensure only feasible
solutions) and soft relations (i.e. designer preferences) can be easily added. The model
is solved using logic based algorithms without reformulating it as an MINLP [8, 47].

In the rest of the paper we first describe the modeling framework and comment
its capabilities. Then, we introduce a brief description of the logic based outer ap-
proximation and its modifications for dealing with implicit models. The capabilities of
the proposed approach are discussed with a case study, the design of a utility sys-
tem power plant that has been decomposed in its constitutive elements, each one
with a different numerical treatment. And finally, numerical results and conclusions are
presented.

4.4.2. Modeling framework

We present a new modeling environment that is able of dealing with mathematical


models. To that end, the first step consists of developing a tool that is able to deal with
different kind of models in a Generalized Disjunctive Programming environment. Figure
4.4.1 shows a scheme of the actual implementation.

In order to deal simultaneously with different numerical behavior the first stage is to
evaluate each of the sub-models in order to answer the following questions:

1. Is the model flexible in the selection of decision (independent) variables? This


question is important because in some situations the correct selection of deci-
sion variables allows breaking (or at least) reducing the recycle structure of a
problem in order to simplify the optimization (and also the simulation)

226
Captulo 4. Resultados

Figura 4.4.1: Scheme of the modeling framework

2. Does the model provide information about derivatives respect to the indepen-
dent variables? If the answer is yes, then we can generate the Jacobian matrix,
or at least part of it. This is the case of some thermodynamic packages that

227
4.4.Integration of different models in design of chemical processes

provide information both for the property and its derivatives with respect some
variables. If the answer is not, then we have different alternatives

2.1. Can derivative information be obtained easily and accurately, for exam-
ple by using complex arithmetic? [48]. This is the case of some modu-
les developed in computer packages that allow complex variables like
Matlab (used in this work). If the answer to this question is not then we
need to implement a finite differences scheme by perturbing the inde-
pendent variables

2.2. Some models introduce numerical noise, e.g. the solution varies slightly
with identical initial values. This is common in systems that solve complex
numerical equations within a finite tolerance. In process simulators this
behavior arises in distillation columns, chemical reactors or other com-
plex operations. If the numerical noise is relatively small, it is still possible
to implement a finite difference approach by increasing the perturbation
of independent variables, but at the expense of getting approximations
of the Jacobian that eventually could produce unexpected behaviors in
the solver. Some important additional consideration must be taken into
account if derivatives are calculated by finite differences:

2.2.1. A finite difference scheme with a noisy model should never be used
inside a recycle (material or calculations) because recycles act like
noise amplifiers. This is a very common problem in process flowsheet
simulation. If this is the case the tolerances for closing recycles must
be at least a couple of orders of magnitude smaller than the pertur-
bation factors. A much better approach consists of letting the NLP
solver to converge the recycles. Although the number of variables
seen by the NLP solver increases and the number of explicit equality
constraints also increases, in general the model is more robust and
usually the computational time does not increase (e.g. by avoiding
converge all the recycles each time the simulation is call). This is the
approach we follow in this work

228
Captulo 4. Resultados

2.2.2. If the model cannot be solved fast (say in fractions of second) the ti-
me necessary for calculating derivatives could be very large making
the optimization impractical

3. If the model is very noisy or the computational time is too large for allowing a
practical implementation, then we cannot use it directly. In this case we can
use a shortcut or aggregated model, but of course this is what we want avoid
from the beginning. Alternatively, it is possible to use a surrogate model. These
include, among other, polynomial correlations; splines; neural networks; radial
basis functions, kriging models, etc. We have implemented the kriging models
[49]. Detailed description about the use of kriging models in optimization can
be found in the following references [46, 50-54]. Here we follow the implemen-
tation proposed by Caballero & Grossmann [46] which can handle both noisy
or deterministic systems by implementing a interpolating or not interpolating
approach depending on the characteristics of the model. It also uses an adap-
tative approach by contracting or moving the domain in successive iterations if
necessary in order to assure accurate results. Finally there is explicit treatment of
constraints in the case of noisy systems, and accuracy of the model is controlled
by cross-validation.

Once all the models have been characterized, we create a model composed by
sub-models with different characteristics that must work efficiently. We have develo-
ped a modeling system that can deal with all these models with the following charac-
teristics:

The complete modeling system is developed in Matlab [55]

Indexing capacities for both algebraic equations and implicit models

Use of Boolean variables, disjunctions and logic equations. Allowing the direct
formulation of the problem as a disjunctive problem without MINLP reformulation

Specific differentiation methods for each sub-model (algebraic, automatic dif-


ferentiation, finite differences based on complex variables; etc)

229
4.4.Integration of different models in design of chemical processes

Determination of sparsity pattern for each individual model and calculation of


the global Jacobian Matrix

Interfaced with different commercial solvers for NLP, LP, MILP models through
Matlab-Tomlab [55, 56], and with our implementations of a simple Branch and
bound algorithm, the outer approximation algorithm [23-25], the LP-NLP based
branch and bound algorithm [57] for MINLP models, and disjunctive solvers wit-
hout MINLP reformulation [47]

Communication with process simulators and other third party models, except
those developed in Matlab, is done by the Windows COM capabilities except
for those developed in Matlab

The following remarks deserve special attention. It is assumed that all the implicit
models are continuous and differentiable. Even in noisy systems, the underlying mo-
del is assumed to be continuous and differentiable. It is not infrequent that computer
models include max/min.operators; Ifsentences, etc. that can destroy the differen-
tiability and continuity assumptions. If the user is able to anticipate this behavior a
correct MINLP reformulation or a disjunctive model can be developed. In other ca-
se, the numerical behavior of the model can produce difficulties. All MINLP and GDP
algorithms require convexity to guarantee convergence to a global optimal solution.
In an implicit model it is difficult to prove convexity, even in if model is convex, since
in general we cannot assure convexity, there is no guarantee that a global optimum
solution can be found.

4.4.3. Logic Based Algorithms

In a hybrid model involving implicit and explicit equations and design (indepen-
dent) and dependent variables it is convenient to explicitly take into account that
differences. The disjunctive model we solve in this paper can be written as follows.

230
Captulo 4. Resultados

min f (xD , xI )

s.t. xD = rI (xI )

rE (xD , xI ) = 0

sE (xD , xI ) 0

Yi

(4.4.3)

Yi
x = h (x )
D Ii Ii
i D


h (x , x ) = 0 xIi = 0
Ei D Ii
gEi (xD , xIi ) 0

(Y ) = T rue

xI X Rn

Y {T rue, F alse}p

where xD is a vector of dependent variables (e.g calculated by the implicit mo-


dels) for which the designer has not direct control over these variables. xI is a vector
of independent variables for which the user has a complete control. The index I
makes reference to implicit equations calculated by third party models (Process Simu-
lator, property estimation modules, etc) and the index E makes reference to explicit
equations. Note that we have introduced dependent variables in explicit equations
(for example in rE (xD , xI ) = 0; orsE (xD , xI ) 0 ), implicitly this formulation involves se-
quential function evaluation, first the implicit models and then the explicit equations.
An alternative approach consists in adding a new subset of independent variables
and explicitly include the relation with the dependent variables : xjI = xjD . In this work
we follow the first approach.

In the model given by equation 4.4.3 we only allow two term disjunctions which
is the case of process networks and synthesis problems. This is not a major limitation
because an N term disjunction can be reformulated as N disjunctions with 2 terms.

231
4.4.Integration of different models in design of chemical processes

4.4.3.1. Initial and primal subproblems

To solve the problem given by equation 4.4.3 we use a version of the logic based
outer approximation algorithm [47] or a logic version of the LP-NLP based branch and
bound algorithm [57]. The first step consists in initializing all the units (sub-models) inside
the disjunctions. Here we have different alternatives. The first one consists of selecting
a basic feasible flowsheet, optimize it and then perform a sub-lagrangian optimization
of the non-existing units in that initial flowsheet. This is the Modeling and Decomposi-
tion (MD) strategy [17]. The second approach consists in selecting a minimum set of
feasible flowsheets in such a way that all the disjunctions be true at least once. This sub-
set can be selected by solving a set covering problem [47] with the constraints given
in form of logic relationships in equation 4.4.3. In some situations the sub-lagrangian
optimization cannot be easily performed. This is the case, for example, of distillation
columns in a process simulator, in which the optimization of non-existing trays is equi-
valent to optimize the complete column. In that case, Caballero et al [22] propose
performing a simulation of the non-existing configurations using the optimal values of
the initial base case, and extend this procedure in the rest of major iterations. Brunet
et al [22, 29, 30, 58] extended this approach to other systems but distillation columns.

In the iteration k only a feasible NLP problem is solved that correspond to fixed va-
lues of boolean variables given by the master problem. The major difference between
the logic versions of the Outer Approximation and LP-NLP based BB is that in the first
the master is solved to optimality, however, in the second one the NLP is solved when
an integer solution is found. In this last case, the tree generated by the master problem
is updated with new linearizations from the last NLP.

It is worth remarking that when a given configuration (set of Boolean variables that
produce a feasible solution) is selected the NLP involves only the common variables
and the variables inside the existing disjunctions. The rest of variables are ignored. Mat-
hematically the non-existing variables are fixed to zero, but that is done a posteriori,
and therefore the solver is only viewing a reduced set of variables which increases the
robustness of the optimization.

232
Captulo 4. Resultados

4.4.3.2. Master subproblem

The objective of the master problem is to provide a new set of Boolean variables
that produce better results than in previous solution. Here we present a tailored master
linear disjunctive problem.

We define the following index set for the iteration k in the algorithm:

{m | mis a feasible configuration already visited by the algorthm}

X
min : + (u1E + u2E + (u1Ei + u2Ei ))
iD

f (xkI , xkD ) + 5xI f (xkI , xkD )T (xI xkI ) + k
P
m objm


 

k) u1E
sign(rE rE (xkI , xkD ) + 5xI rE (xkI , xkD )T (xI xkI ) + m rEm
k
P






sE (xkI , xkD ) + 5xI sE (xkI , xkD )T (xI xkI ) + skEm u1E
P
m




Yi


s.t. k = 1...K
Yi (4.4.4)

 

sign(hkE ) hEi (xkIi , xkD ) + 5xI hEi (xkIi , xkD )T (xIi xkIi ) + m hkEi,m ukEi i D
P


xIi = 0




gEi (xkIi , xkD ) + 5xI gEi (xkIi , xkD )T (xIi xkIi ) + m gEi,m
k
P
0





(Y ) = T rue




u1E , u2E , ukEi 0

xI X R n

Y {T rue, F alse}p

The disjunction model 4.4.4 is obtained by linearization in terms of independent va-


riables in the last point obtained by the NLP solver, either in the initialization or in a given
iteration k. The term correspond to the difference between the objective function at
a given iteration k of the NLP and the objective function associated to a given topo-
k ; g k ; hk
logy. The terms rEm k
Em Ei,m ; gEi,m are the difference between the values
of a given constraint for the new topology and their value in the original NLPk pro-
blem. These terms are no needed (fixed to zero) if the modeling and decomposition
or the logic based outer approximation are used. Only in the third alternative, when
no sub-optimization of non-existing disjunctions is used, these terms must be included.

233
4.4.Integration of different models in design of chemical processes

Variables u00 are positive slack variables, that appear as penalties in the objective
function using an exact penalty ( is the penalty factor). These variables are intro-
duced to ensure that the only possibility for an infeasible master problem is due to
logical relationships, and at the same time is a heuristic to try to minimize the effect of
non-convexities [25].

The disjunctive master problem in 4.4.4 can be solved as a MILP problem using the
convex hull reformulation [59].

If the original problem is convex, the master problem in the MD and logic OA base
versions yields a lower bound to the optimal solution. Therefore, the optimal solution is
found when in a given iteration, the optimal solution of the primal and master problems
are inside a given tolerance (when they cross each other if a canonical cut is added in
each master problem [47]). In non-convex problems, like those solved in this work, the
master problem cannot be ensured to yield a lower bound. The search is terminated
when in two consecutive iterations the NLP does obtain an improvement [25]. This is just
a heuristic based on experience. To improve the quality of the solution in non-convex
problems. One could perform a fixed number of iterations (e.g. 10) but at the price of
increasing the computational time to solve the problem.

4.4.4. Case Study

As an example we present the synthesis of a utility system in which it is assumed that


different components are simulated with modules that have different numerical beha-
vior. There are different commercial tools that can perform the design very efficiently
(Aspen Utility Planner; Arianeby ProSim) using databases and cost correlations
obtained directly from industrial applications. The objective in this work is to show that
it is possible synthesizing a complex system with different models efficiently.

There are a many published on the optimization and design of utility systems. [60-63]
which have relied on simplified linear models. Rigorous models for power production
of steam turbines have been developed by Mavromatis and Kokossis [64, 65]. Bruno
et al. [66] proposed a superstructure optimization formulated as a MINLP based on
the previous work by Papoulias & Grossmann. Manninen & Zhu [67] decomposed the

234
Captulo 4. Resultados

problem in a Master problem which indicates major structural features of a design and
flowsheet level in which an exergy analysis identifies relevant modification options.
Varbanov et al. [68] developed more accurate models for steam and gas turbines,
that provide a better description of part-load performance.

In this work we use a superstructure inspired in the previous work by Bruno et al. [66].
Figure 4.4.2 shows the superstructure that includes:

A gas turbine with a heat recovery steam generator (HRSG) with no further fuel
addition. The gas turbine can satisfy the electricity demands where exhaust ga-
ses can be used in a HRSG boiler to provide vapor to a very high pressure (VHP)
header.

High pressure boilers fired by fuel. In all the boilers it is necessary a blowdown, in
this paper we assume a fixed blowdown rate of 3 % of the steam entering the
boiler.

Waste heat boiler that operates at medium pressure and recovers heat from
process flue gases or other units such as reactors.

We assume four pressure headers VHP, HP, MP and LP. Besides we have to consi-
der a vacuum level for the condensation turbines and an atmospheric level that
recovers the water returned from the process (previously treated). The demine-
ralized water to compensate plant losses and consumptions is also assumed to
operate at atmospheric pressure.

A deaerator is included to remove the dissolved gases by steam stripping be-


fore feeding the water into boilers. Although an extra boiler feedwater can be
considered. In this case we assume that the deaerator works with LP steam from
the plant.

Steam turbines are used to generate electricity and to satisfy mechanical po-
wer. Here we consider the following turbines.

Backpressure turbines working between any two pressure levels (VHP-HP;


VHP-MP, VHP-LP; HP-MP, HP-LP, MP-LP).

235
4.4.Integration of different models in design of chemical processes

Extraction backpressure turbines exhausting from VHP to HP and MP and


form HP to MP and LP.

Condensing turbines from HP and MP.

Extracting condensing turbines from MP to LP and vacuum.

The steam is distributed to steam consumers, to steam turbines or to the next


pressure header through letdown valves.

Electric motors can be used to meet the required power demands.

Finally utility pumps are included to change the pressure in all the liquid streams.

Figura 4.4.2: Superstructure of the utility system plant. It is possible to include N turbines
in parallel of each class

Before including a description of the actual implementation of each unit and its
characteristics, because we are using a modular approach, it is convenient to analy-

236
Captulo 4. Resultados

ze the information flow in the flowsheet (or superstructure). If we assume that the in-
formation flow coincides with the mass flow, which is the usual approach in modular
simulation, then any valid flowsheet must include a recycle involving all the unit opera-
tions. For example, we should assume a water mass flowrate entering the boilers and
the fresh water entering the deaerator, and after calculating the entire flowsheet the
water exiting from the deaerator should be equal to the assume mass flowrate ente-
ring the boilers. As commented before, it is convenient to let the optimizer converge
this recycle in order to minimize the effects of numerical noise. However, if we are able
to change the direction of the information flow we could solve a given flowsheet with
small recycles involving a single unit operation that can be efficiently solved by the im-
plicit model. In this case it is possible just by reversing the calculation in the boilers: The
water (steam) mass flowrate is calculated in terms of the fuel flowrate and operating
conditions. In the deaerator it is then possible to calculate the fresh water mass flow-
rate and no recycles appear in the flowsheet. This kind of analysis can be automated,
see for example the book by Westerberg et al [69].

Following is a description of the main equipment involved in the power plant and
their particular models for implementation.

Gas turbine and HRSG

The gas turbine is introduced in the model as a kriging metamodel. To generate


data for creating the kriging model [45, 46, 49, 52, 53] a gas turbine model was created
in Aspen Hysys. See Figure 4.4.3. The model is formed by a compressor in which air at
ambient conditions is used. The compressed air is introduced together with pressurized
fuel (natural gas) in a Gibs reactor that simulates the combustion chamber of the gas
turbine. The gases exiting from the combustion chamber are introduced in a turbine
where they are expanded to atmospheric pressure. The work generated by the gas
turbine is used to move the compressor and to generate electricity which is simulated
by an energy balance. The exhaust gases are sent to a HRSG that is used to generate
more steam, and simulated as a regular heat exchanger. All relevant data about the
different equipment are given in Table 4.4.1. Table 4.4.2 shows utilities data.

It is possible to include more complex models for efficiencies, for example using the
Willans line [64, 65, 68]. However, the parameters depend both on the particular turbi-

237
4.4.Integration of different models in design of chemical processes

Tabla 4.4.1: Summary of operating parameters


Unit Operating Parameters
Headers VHP Pressure 100 bar
HP Pressure 40 bar
MP Pressure 20 bar
LP Pressure 3 bar
Condenser Pressure 1.01325 bar (1 atm)
Vacuum Pressure 0.1 bar (10 kPa)
Vacuum Temperature 46.01o C

Gas Turbine Compressor efficiency 83 %


Turbine efficiency 85 %
Mechanical efficiency 98 %
Compression ratio (r) 4 r 20
Compression Temperature out
Tcombustion 1200o C
Exhaust Gas Temperature TExhaust 600o C
Peng Robinson (combustion chamber)
Thermodynamics
ASME Steam for steam lines

HRSG Min approach Temp 30o C


Stack Temperature 160o C

Boilers Efficiency 90 %
Approach temperature 30o C
Blowdown rate 3%
Excess combustion air 10 %
Peng Robinson Hysys default (combustion chamber)
Thermodynamics
ASME Steam for steam lines

Steam Turbines Efficiency is = c( aW


b+W
t
)0,95 (W in kW)
t
Backpressure Pin 40 bar; a = 427,0992; b = 865,5034; c = 0,8217;
Pin 20 bar; a = 378,0419; b = 758,8181; c = 0,8223;
Pin 1 bar; a = 181,9821; b = 381,1312; c = 0,8150;
Condensation Pin 40 bar; a = 313,3561; b = 648,2201; c = 0,7714;
Pin 20 bar; a = 244,8585; b = 528,1410; c = 0,7769;
Pin 1 bar; a = 142,4190; b = 323,1106; c = 0,7700;

Deaerator Pressure 1.01325 bar (1 atm)


Vent 5 % LP introduced

238
Captulo 4. Resultados

Figura 4.4.3: Scheme of the actual implementation of a gas turbine in Aspen-Hysys

Tabla 4.4.2: Utilities Data

Utility
Fuel Natural Gas
Composition (wt fraction)
Methane 0.8405
Ethane 0.1278
Propane 0.0203
n-Butane 0.0033
Ethylene 0.0081
Temperature 25o C
Pressure 101.325 kPa (1 atm)
Cost 0.23$ /(m3std ) = 2.5792 M$ /(year tone fuel)*

Cooling Water Temperature 25 - 35o C


Pressure 101.325 kPa (1 atm)
Cost 19.1952 106 M$/(kW year)

Demineralized Water Temperature 25 o C


Pressure 101.325 kPa (1 atm)
Cost 0.02 M$/(t/h year)

*Calculated for 8000 h of operation year (F uel = 0.7134 kg/(m3std ))

239
4.4.Integration of different models in design of chemical processes

ne and on the operating conditions, as we do not know, a priori, the particular turbine
assume a fixed efficiency is a reasonable assumption [70]. For fixed efficiencies values
the gas turbine can be calculated with three independent variables. We have used
the air to fuel ratio (wt.) fixed between 40 and 100. The compressor output pressure
(between 400 and 2000 kPa that correspond to compression ratios between 4 and 20.
And the fuel mass flowrate.

The model as implemented in Aspen Hysys is slightly noisy and although it slows
down a bit the optimization it can be used as is. However, as commented above, for
illustration purposes, it was substituted by a kriging metamodel. The kriging approach
has the advantage of allowing a very fast evaluation of the model and it is possible
to get numerical derivatives without error (except by the computer precision) by using
complex derivatives.

To calibrate the kriging we have used 100 points distributed using a min-max ap-
proach (minimizing the maximum distance between two points) using 1000 kg/h of fuel
as calculation basis which in fact leaves the kriging model with two degrees of free-
dom. In this case the maximum error introduced by the kriging interpolation is under
the 0.8 %. Figure 4.4.4 shows the values predicted by the kriging and the calculated
ones and the error for 500 random points.

240
Captulo 4. Resultados

Figura 4.4.4: Left, calculated and predicted values for the kriging model in the gas
turbine. Right, % kriging error in 500 random points

241
4.4.Integration of different models in design of chemical processes

Therefore, the model for the gas turbine can be written as follows:

out out stack out


[Wgt , Tcombustion , Tturbine , QHRSG , THRSG ] = Kriging T urbine(AFgt , Pcompressor , F uelgt )

40 AF 100
out
400 Pcompressor 2000(kP a)
(4.4.5)
out
Tcombustion 1200(o C)
out
Tturbine 600(o C)
stack
THRSG 160(o C)

where:

AFgt Gas Turbine air to fuel ratio (weight basis)


out
Pcompressor Output pressure from the compressor in the gas turbine (kPa)
F uelgt Gas turbine mass fuel flowrate (kg/h)
Wgt Effective work provided by the gas turbine

out Temperature of the output stream from the combustion chamber


Tcombustion
in gas turbine)
out
Tturbine Temperature of the exhaust gases from the gas turbine
QHRSG Maximum heat available from the Heat Recovery Steam Generator
stack
THRSG Temperature for the gases that exit from the HRSG

For the HRSG we have to decide at which pressure we want to generate the ex-
tra steam and which is the minimum approach temperature between the gases and
steam. With an energy balance we can calculate the extra steam mass flowrate and
its conditions (pressure and temperature). This calculation is performed using Hysys as
a thermodynamic calculator for estimating specific water and steam enthalpies at the
inlet and outlet conditions. Alternatively, it is possible to generate a flowsheet in Aspen
Hysys to perform this calculation, but in this way we do not need to interact with the
graphical interface and the calculations are faster. The model for the HRSG can be
written as follows

242
Captulo 4. Resultados

Steam Steam out


[FHRSG , THRSG ] = HRSG(QHRSG , THRSG , PHRSG , THRSG ) (4.4.6)

where:

PHRSG Steam pressure in the HRSG (kPa)


Steam
FHRSG Steam flowrate exiting from HRSG (kg/h)
Steam
THRSG Temperature of the vapor exiting from the HRSG (o C)
THRSG Minimum approach temperature in the HRSG (o C)

Boilers

For the boilers we follow a similar approach to that followed in the gas turbine. A
flowsheet is built in Aspen Hysys to rigorously simulate a boiler. In this case we take
advantage of the Aspen Hysys capability for transmitting information from one unit to
another as soon as all the degrees of freedom of a unit have been provided by the
user See Figure 4.4.5.

Figura 4.4.5: Scheme of the implementation of a Boiler in Aspen Hysys. Dotted line
shows the information flow that is different of mass flow

243
4.4.Integration of different models in design of chemical processes

The design variables, see Figure 4.4.5, are the fuel flowrate and the temperature
of the gases after heat exchange with the water to generate steam. The calculation
sequence is as follows:

A fuel and an air stream (10 % excess) are introduced in a Gibbs reactor that simu-
lates the combustion of fuel in the boiler. The combustion gases exchange heat with
water at high pressure to generate steam. Instead of using a single heat exchanger,
we use two heat exchangers in which the heat is transmitted with a given efficiency
(90 %). Note that if the output gas temperature is known (design variable), the amount
of heat available for generating steam is then also known. The temperature of steam
is fixed by assuming a minimum temperature difference between the stack tempe-
rature of combustion gases (exit temperature) and the steam. With this information
Aspen-Hysys can calculate the steam flowrate, the blowdown, the high pressure wa-
ter flowrate and the pump. See Figure 4.4.5.

The model can be developed for a given mass flow rate base (i.e. 100 kg/h). In
those conditions if the steam pressure is known the complete model depends just on
one design variable. In our case the temperature of combustion gases after heat ex-
change. Efficiency and approach temperature are fixed parameters. See Table 4.4.1
for all the relevant data for boilers.

Like in the case of gas turbines the model is slightly noisy, and it can be used as is,
because it is possible to get accurate derivatives by a finite difference approximation.
However, it is much more efficient, from a numerical point of view, to use a polynomial
or cubic spline interpolation for which the error is negligible and we have analytical
derivatives. This is the approach we use in this work.

Conceptually, the model for the boiler can be written as follows:

Steam Steam Gas


[Fboiler , Tboiler ] = Boiler(Tboiler , F uelboiler , Pboiler )
(4.4.7)
Gas
TSteam dew (Pboiler ) + T Tboiler 1900(o C)

244
Captulo 4. Resultados

where:

Pboiler Steam pressure in the boiler


F uelboiler Mass flow rate of fuel consumed by the boiler
Gas
Tboiler Temperature of the gases exiting from the boiler
Steam
Tboiler Boiler steam temperature
Steam
Fboiler Boiler steam mass flow rate

Steam Turbines

A steam turbine converts the energy of steam stream into power. The total power
for the expansion is further split into power used by the shaft and energy losses (me-
chanical friction, heat losses and kinetic losses). The three main factors affecting the
turbine performance are the turbine size (maximum power), the pressure drop across
the turbine and the operating load.

An ideal steam turbine assumes an isentropic expansion. Any model of an actual


turbine captures the real behavior by means of the isentropic efficiency:

Hin Hout
IS = (4.4.8)
Hin HIS

In the literature there are different correlations that capture the isentropic efficiency
in terms of the actual load and pressure difference [12, 66, 68]. These correlations vary
depending on the manufacturer and the turbine size and model. However, according
to Varbanov et al [68] although the machine efficiency varies noticeably in a non-
linear way with the load, it retains relatively high values, that go from 88 % for smaller
turbines to 96 % for the larger ones. Bruno et al. [66] presented correlations for turbines
working at their maximum load that relates the steam turbine efficiency with the inlet
pressure, the shaft power and exhaust conditions (condensing and non-condensing
turbines). In this work we include those correlations plus a fixed 0.95 efficiency to take
into account a possible deviation from full load operation.

For calculating the ideal turbine we can use a turbine model as implemented in
Aspen Hysys or using Aspen Hysys as a thermodynamic calculator server. We follow this

245
4.4.Integration of different models in design of chemical processes

second approach because we have a better control over the model and we do need
to interact with the Aspen Hysys graphical interface that slow down the calculations.

A complex turbine can be decomposed into its corresponding basic components


involving simple turbine units, splitters and mixers. Any minor differences between the
extraction turbine and the decomposed model can be compensated for later in the
design by small adjustments in the specification in terms of efficiency [12].

For condensation turbines, we use cooling water as cold utility. Therefore, the pres-
sure of the vacuum header is fixed to allow enough of a temperature difference bet-
ween the condensing steam and the cooling water.

Conceptually, the model for a turbine can be written as follows:

[WT urbine , FTSteamOut


urbine , FTSteamOut
urbine , Qcond ] =

= T urbine(FTSteamIn SteamIn SteamIn Out


urbine , PT urbine , TT urbine , PT urbine , T ype)
SteamOut
FT urbine k k FTSteamIn
urbine
X (4.4.9)
k = 1
kT urbineOutputs

FTSteamIn SteamIn SteamIn


urbine FT urbine FT urbine

k k k

where:

FTSteamIn
urbine Mass steam flowrate entering the turbine
PTSteamIn
urbine Pressure of the inlet steam
TTSteamIn
urbine Temperature of the inlet steam
Outk
PT urbine Pressure of each one of the output streams
T ype Type of steam turbine: back pressure or condensing
WT urbine Turbine power
SteamOutk
FT urbine Mass steam flowrate in each turbine output
SteamOut
TT urbine k Temperature of each output streams
Qcond Condenser heat load (only for condensing turbines)

246
Captulo 4. Resultados

Vapor Headers

A vapor header collects the steam (water) streams coming from different sources:
Boilers, HRSG, steam turbines, letdown valves, etc at different temperatures but at the
same pressure and distribute it where necessary (process, steam turbines, letdown
valves, deaerator,. . . ).

The header conditions (temperature if the pressure is fixed) can be calculated just
by mass and energy balances of all the streams entering to the header (it is equiva-
lent to a mixer). The enthalpy of each stream is either determined from the unit that
generated or calculated by Aspen Hysys.

Finally a simple mass balance relates the input streams with the output streams
(including process demands) in each header.

Valves

A letdown valve is simulated as an iso-enthalpic expansion between two pressure


levels. The inlet conditions are calculated by the corresponding header. Therefore,
the only computation needed is the calculation of the outlet temperature for a given
pressure and enthalpy (PH-flash).

Deaerator

For simulating the deaerator we follow the approach described by Smith [12]. The
mass flow and conditions (temperature or enthalpy) of the following water (steam)
streams are known: water returned from the process, water coming from the atmosp-
heric or vacuum headers, and water exiting from the deaerator to the boilers or HRSG.

Assuming that the deaerator works with LP steam extracted from the LP main, and
that a fixed percentage (5 %) of the steam is vented, we can calculate the treated
water makeup and the LP steam requirements (See Figure 4.4.6) by mass and energy
balances

247
4.4.Integration of different models in design of chemical processes

LP
X
Steam W ater
FT W + Fi + FDeaerator = Fboilers + FV ent
iW aterReturn
LP
X
HT W FT W + Steam
Hk Fk + HLPSteam FDeaerator = Hboilers Fboilers + HV ent FV ent(4.4.10)
W ater W ater

kW aterReturn
LP
Steam
FV ent = FDeaerator

where F makes reference to mass flow rates. H makes reference to specific enthal-
pies.

TM Reference to Treated Water Makeup


Set of all the water returning to Power plant: condensate,
W ater Return
from process, etc.)

Fraction of LP steam vented

Note that enthalpies for the water to boilers (and HRSG) and Vent streams are cal-
culated in the deaerator operating conditions.

Figura 4.4.6: Scheme of a deaerator

For the economic evaluation we use the total annualized cost (TAC). To annualize
the capital costs we use the following relationship [12]:

248
Captulo 4. Resultados

T AC = Operating Costs + f (Capital Cost


i(1 + i)N (4.4.11)
f =
(1 + i)N 1

where i is the interest rate per year and N is the number of years. Data fuel, co-
oling water and other costs are given in Table 4.4.2. To estimate the capital costs we
use the correlations given by Turton el al [13], except for the gas turbine where the
purchased cost was correlated from manufactures provided by Nye Thermodynamics
Corporation:

GasT urbineCost(M $) = 0,7886(WGT )0,7395 (WGT in M W ) (4.4.12)

The final model is formed by an objective function (TAC minimization) subject to


fixed equipment equations: vapor headers, letdown valves and deaerator. Note that
for the valves a disjunction is not needed because the flow going through a valve
takes the value zero if the valve does not exist. Conditional equipment equations that
are included in two term disjunctions are: Gas Turbine (YGT ), HRSG (YHRSG ), Boilers
(Yboilersk ) and Steam Turbines (YST ). Where the Y makes reference to the boolean
variable related with each equipment.

We must include the following logical relationships:

If the HRSG exists then the gas turbine must exist:

YHRSG YGT (4.4.13)

If a gas turbine is not selected then the HRSG cannot be selected

249
4.4.Integration of different models in design of chemical processes

YGT YHRSG (4.4.14)

At least the HRSG or a boiler must be selected

YHRSG ( Yboilersk ) (4.4.15)


kboilers

The mechanical power must be supplied by given equipment (Steam turbines or


electric motors). In that case, we need to assign that power to the equipment among
those available. To that end we define the new boolean variable:

Zeq,pd True if the equipment eq satisfies the demand pd

The following disjunctions explicitly force that relationship


Zeq,pd
(4.4.16)
Weq = P owerDemandpd

The following logical relationships correctly ensure feasible assignments:

If the mechanical power demand pd is assigned to equipment eq then the


equipment eq must be selected:

Zeq,pd Yeq (4.4.17)

Each mechanical power demand must be assigned

(Zeq,pd )eq (4.4.18)


pd

250
Captulo 4. Resultados

Although it is not strictly necessary, if we want force that a given equipment be


assigned to a single mechanical power demand we can include the following rela-
tionship:

at most(Zeq,pd )eq (4.4.19)


pd

Note that steam demands are included in the mass balances in each steam hea-
der, and therefore it is not explicitly necessary include a boolean variable for each
steam demand.

As a final remark, it is worth mentioning that adding new equipment (i.e. more gas
turbines, or boilers) modify the equations of any equipment or adding any other is
straightforward. Due to the modular approach, the modification of equipment, ad-
ding or removing alternatives does not modify the structure of the rest of the model.

4.4.5. Examples

Here we present three instances of the model that illustrates the versatility of both
the case study and the modeling framework. All the relevant data for each of the
three instances are given in Table 4.4.3.

Tabla 4.4.3: Data for examples

Instance 1 Instance 2 Instance 3


VHP Steam Demand (t/h) 1 - -
HP Steam Demand (t/h) 20 10 2
MP Steam Demand (t/h) 5 - -
LP Steam Demand (t/h) 10 5 5
Water Returned from process (t/h) 10.8 4.5 2.1
Electricity demand (MW) 50 5 0.6
2000 500
Drivers demands (kW) - 2500 1000
3500

251
4.4.Integration of different models in design of chemical processes

The first instance we have a power plant with high power (electricity) demand (50
MW) and different demands of VHP (100 bar) stream 1 t/h; HP (40 bar) steam 20 t/h; MP
steam (20 bar) 5 t/h and LP steam (3 bar) 10 t/h. In this first case there is no mechanical
power demand. The optimal solution includes a gas turbine with HRSG, and four steam
turbines working between different pressure levels. See Figure 4.4.7.

Figura 4.4.7: Optimal solution for case study, instance 1

In this example it is worth remarking the perfect steam balance. The operating con-
ditions of the gas turbine and HRSG are optimized in such a way that the steam pro-
duced in the HRSG is used, without any excess, for supplying the steam demands (in-
cluding the LP steam in the deaerator) and the extra power demand not satisfied by
the gas turbine. Table 4.4.4 shows with detail the results of this example.

252
Captulo 4. Resultados

Tabla 4.4.4: Results for example 1


Equipment
Gas Turbine Power (MW) 37.9 MW
Fuel (t/h) 8.48
Compression ratio 30
Air to fuel ratio 53.86
Exhaust gas temperature (o C) 575.08
Installed cost (M$) 40.58
HRSG Water inlet temperature (o C) 80
Steam mass flow rate (t/h) 68.99
Installed cost (M$) 3.42
Backpressure turbine 1 PIn Pout (bar) VHP(100)-HP(40)
Work (kW) 2.619
Efficiency 0.781
Steam Flow (t/h) 41.71
Installed cost (M$) 1.81
Backpressure turbine 2 PIn Pout (bar) VHP(100)-MP(20)
Work (kW) 2.691
Efficiency 0.781
Steam Flow (t/h) 26.28
Installed cost (M$) 1.82
Backpressure turbine 3 PIn Pout (bar) HP(40)-LP(3)
Work (kW) 2.778
Efficiency 0.78
Steam Flow (t/h) 21.71
Installed cost (M$) 1.83
Condensing Turbine PIn Pout (bar) MP(20)-Vacuum(0.1)
Work (kW) 4.008
Efficiency 0.738
Steam Flow (t/h) 21.28
Installed cost (M$) 1.96
Condenser heat load (MW) 13.33
Condenser Installed cost (M$) 0.39
Deaerator LP Steam flow rate (t/h) 11.71
Fresh water makeup (t/h) 25.7
Vent (t/h) 0.59
Water from atm. header (t/h) 32.08
Installed cost (M$) 0.070
Costs Total installed Cost (M$) 51.89

Annualizing factor (f)


0.1874
(8 years 10 % interest)
Fuel Cost (M$/year) 21.86
Fresh water cost (M$/year) 0.516
Cooling water cost (M$/year) 0.136
Total utilities cost (M$/year) 22.51

TAC (M$/year) 32.24


Note: cost of pipes, pumps and valves is not included

253
4.4.Integration of different models in design of chemical processes

From a computational point of view, the most complex non-linear problem, that
in which all the units exist (using for initialization, that is a feasible solution, although
of course not the optimal one) the problem is solved in around 30 seconds of CPU
time, although this time depends on the initial values. The rest of the NLP sub-problems
are solved typically in CPU times that range between 5 and 20 seconds. (Intel(R) Co-
re(TM)2Quad CPU 2.4GHz 2.39 GHz. RAM 8 GB under Windows 7). The MILPs where
solved in all the cases in less than 1 second of CPU time. Table 4.4.5

Tabla 4.4.5: Numerical statistics for the three instance

Instance 1 Instance 2 Instance 3


N of Boolean variables 1 - -
N of independent variables 20 10 2
N of linear explicit equations(1) 5 - -
N of non-linear explicit equations(2) 10 5 5
N of implicit blocks 10.8 4.5 2.1
Detailed iterations for instance 1(3)
Iterations(4) Objective CPU time (s) Solver
Initialization (all units exist). NLP 33.54 35.3 CONOPT
MILP Master 32.84 0.13 CPLEX
Iteration 2: NLP 32.69 7.9 CONOPT
Iteration 2: Master-MILP 33.09 0.16 CPLEX
Iteration 3: NLP 32.30 7.7 CONOPT
Iteration 3: Master-MILP 33.15 0.14 CPLEX
Iteration 4: NLP 32.24 6.6 CONOPT
Iteration 4: Master-MILP 33.26 0.13 CPLEX
Iteration 5: NLP 32.50 6.1 CONOPT
Iteration 5: Master-MILP 33.29 0.16 CPLEX
Iteration 6: NLP 32.48 8.9 CONOPT
Iteration 6: Master-MILP 33.35 0.34 CPLEX
Iteration 7: NLP 32.43 6.6 CONOPT
Iteration 7: Master-MILP 33.36 0.16 CPLEX
Iteration 8: NLP 32.62 8.4 CONOPT
(1)
Logical relationships not included
(2)
Includes both independent and dependent variables
(3)
Numerical performance for instances 2 and 3 is similar
(4)
Fixed number of iterations to 8. Note also that the Master is not providing lower bounds due to slack variables
(Intel(R) Core(TM)2Quad CPU 2.4GHz 2.39 GHz. RAM 8 GB under Windows 7)

All the numerical test we have performed show that for a given problem with a
fixed configuration, the optimal solution obtained is not too sensitive to the initial va-
lues. In other words, there is reasonable evidence that the solution is the global one

254
Captulo 4. Resultados

(although of course we cannot guarantee it). However, in the master problem slack
variables become active in the firsts iterations, and as a consequence we cannot
guarantee that the solution showed be the global optimum. Nevertheless, numerical
test allowing a large number of major iterations (i.e. 25 major iterations) show that the
solution obtained, if not the optimal is at least a good solution (difference lower than
10-15 % in the worst cases).

In the second instance there are mechanical power demands (three rotating equip-
ment requires 2000, 2500 and 3500 kW respectively) that must be satisfied by steam
turbines, and power (electricity) demand 5 MW. Moreover the process plant requires
10 t/h of HP steam and 5 t/h of LP steam.

The optimal solution includes a gas turbine, HRSG, three steam turbines to provide
mechanical power (one of those is a condensing turbine).See Figure 4.4.8 and Table
4.4.6. At difference of the first example, the steam needed for the turbines and to
satisfy the process demands results in electricity generated that is larger than needed,
and which can be sold to the grid.

It is possible to include a boiler to produce the extra steam without the electricity
excess but the economic balance shows that the extra cost of fuel and boiler is eco-
nomically inferior.

The computational performance is similar to the first example. Nevertheless it is in-


teresting remark that for the initialization it is not practical solve a set of feasible NLPs
that includes at least once all the alternatives. Due to the assignment equations of
turbines to mechanical demands, the number of alternatives is too large. However,
what we really need to initialize are the steam turbines. Therefore, for the initialization
of the MILP master we simple assume that all the steam turbines exist subject to the
power generated by all of them must be larger than the total power (mechanical +
electricity).

The third instance is similar to the second one but the mechanical power demands
(500 kW, 1000 kW) and electricity demand (600 kW) are smaller. There are also de-
mands of HP steam (2 t/h) and LP steam (5 t/h). The optimal solution includes a Boiler,
2 backpressure steam turbines, a condensing turbines, and the deaerator. There are

255
4.4.Integration of different models in design of chemical processes

Tabla 4.4.6: Results for example 2


Equipment
Gas Turbine Power (MW) 10.96MW
Fuel (t/h) 4.09
Compression ratio 4
Air to fuel ratio 61.67
Exhaust gas temperature (o C) 600
Installed cost (M$) 16.21
HRSG Water inlet temperature (o C) 80
Steam mass flow rate (t/h) 39.44
Installed cost (M$) 2.15
Extraction turbine 1 PIn Pout (bar) VHP(100)-MP(20)
Work (kW) 2.500
Steam Flow (t/h) 33.74
Steam Flow Out (t/h) 26.99 - 6.75
Efficiency 0.781 - 0.784
Installed cost (M$) 2.50
Extraction turbine 2 PIn Pout (bar) HP(40)-MP(20)-LP(3)
Work (kW) 2.000
Steam Flow (t/h) 19.71
Steam Flow Out (t/h) 8.02-11.70
Efficiency 0.781-0.782
Installed cost (M$) 2.70
Condensing Turbine PIn Pout (bar) MP(20)-Vacuum(0.1)
Work (kW) 3.500
Steam Flow (t/h) 17.75
Efficiency 0.738
Installed cost (M$) 1.92
Condenser heat load (MW) 11.40
Condenser Installed cost (M$) 0.348
Deaerator LP Steam flow rate (t/h) 6.70
Fresh water makeup (t/h) 10.84
Vent (t/h) 0.33
Water from atm. header (t/h) 22.25
Installed cost (M$) 0.052
Costs Total installed Cost (M$) 25.88

Annualizing factor (f)


0.1874
(8 years 10 % interest)
Fuel Cost (M$/year) 10.53
Fresh water cost (M$/year) 0.216
Cooling water cost (M$/year) 0.116
Total utilities cost (M$/year) 10.86

TAC (M$/year) 15.71

256
Captulo 4. Resultados

Figura 4.4.8: Optimal solution for case study, instance 2

also two letdown valves to distribute the steam among the steam mains and the steam
turbines. In this case, the small demand of electricity does not justify the gas turbine
high cost. See Figure 4.4.9 and Table 4.4.7 for a detailed summary of the results.

257
4.4.Integration of different models in design of chemical processes

Tabla 4.4.7: Results for example 2


Equipment
Boiler Fuel (t/h) 1.28
Temperature steam (o C) 502.6
Heat load (MW) 13.47
Steam flow rate (t/h) 14.43
Blowdown flow rate (t/h) 0.43
Installed cost (M$) 0.942
Back pressure turbine 1 PIn Pout (bar) VHP(100)-MP(40)
Work (kW) 600
Steam Flow (t/h) 10.27
Efficiency 0.782
Installed cost (M$) 1.08
Back pressure turbine 2 PIn Pout (bar) MP(20)-LP(3)
Work (kW) 1.000
Steam Flow In (t/h) 10.15
Efficiency 0.782
Installed cost (M$) 1.36
Condensing Turbine PIn Pout (bar) MP(40)-Vacuum(0.1)
Work (kW) 500
Steam Flow (t/h) 2.29
Efficiency 0.734
Installed cost (M$) 1.01
Condenser heat load (MW) 1.44
Condenser Installed cost (M$) 0.124
Deaerator LP Steam flow rate (t/h) 2.45
Fresh water makeup (t/h) 5.02
Vent (t/h) 0.12
Water from atm. header (t/h) 7.09
Installed cost (M$) 0.0317
Costs Total installed Cost (M$) 4.56

Annualizing factor (f)


0.1874
(8 years 10 % interest)
Fuel Cost (M$/year) 3.30
Fresh water cost (M$/year) 0.101
Cooling water cost (M$/year) 0.015
Total utilities cost (M$/year) 3.42

TAC (M$/year) 4.27

258
Captulo 4. Resultados

Figura 4.4.9: Optimal solution for case study, instance 3

4.4.6. Conclusions

This paper describes how models from very different sources and different numeri-
cal behavior can be incorporated as modules to a general synthesis framework that
can capture their numerical performance and implement numerical approaches tai-
lored to the need of each model. Some of the capabilities of the proposed approach
are the following:

Structural analysis of the input-output structure of each model for selecting de-
sign variables and reducing as much as possible the recycle structure of the
problem

Determination of sparsity pattern for each model to accelerate the calculation


of the Jacobian matrix

259
4.4.REFERENCIAS

Different approaches for calculating the derivative information that includes,


analytical; complex variables based differentiation, finite differences

Possibility for dealing with noisy systems through adaptive surrogate models (kri-
ging)

Disjunctive modeling that facilitates the conceptual modeling

Soft and hard logic relations easily included in the model

The final result is a hybrid system that includes implicit models coming from different
sources and with different numerical behavior, and models or constraints in equation
form. The basic idea is that the best available model can be used in a rigorous synt-
hesis framework when shortcut models are not adequate.

It has been shown that the synthesis problem can be written as a General Disjun-
ctive Programming problem and solved without MINLP reformulation using the logic
based outer approximation of LP/NLP based branch and bound. Conceptually the
GDP approach allows an easy model formulation for the final user and at the same
time encapsulates each sub-model facilitating the numerical study of its characteris-
tics.

The synthesis of a utility system, in which each of the components has been trea-
ted as a sub-model with different characteristics has been used as a case study. Future
work will concentrate on producing a new design and synthesis tool that can be ap-
plied to many processes.

Acknowledgements

The authors wish to acknowledge support from the Spanish Ministry of Science and
Innovation (CTQ2012-37039-C02-02).

Referencias

[1] Braunschweig, B.L. y col. ((Process Modelling: The promise of open software architectures)).

En: Chemical Engineering Progress 96(9) (2000), pags. 65 -76.

260
Captulo 4. Resultados

[2] Rosenthal, R.E. GAMS: A Users Guide. GAMS Development Corporation, 2013.
[3] Fourer, R., Gay, D.M. y Kernighan, B.W. AMPL: a modeling language for mathematical pro-
gramming. Thomson/Brooks/Cole, 2003.
[4] Grossmann, I.E., Caballero, J.A. y Yeomans, H. ((Mathematical programming approaches
to the synthesis of chemical process systems)). English. En: Korean Journal of Chemical

Engineering 16.4 (1999), pags. 407-426.
[5] Shacham, M. y col. ((Equation oriented approach to process flowsheeting)). En: Computers

& Chemical Engineering 6.2 (1982), pags. 79 -95.
[6] Marquardt, W. ((Trends in computer-aided process modeling)). En: Computers & Chemical

Engineering 20.6-7 (1996), pags. 591 -609.
[7] Biegler, L.T., Grossmann, I.E. y Westerberg, A.W. Systematic methods of chemical process
design. Prentice-Hall international series in the physical and chemical engineering scien-
ces. Prentice Hall PTR, 1997.
[8] Grossmann, I.E. ((Review of Nonlinear Mixed-Integer and Disjunctive Programming Techni-
ques)). English. En: Optimization and Engineering 3.3 (2002), pags.
227-252.
[9] Lee, S. y Grossmann, I.E. ((Logic Based Modeling and Solution of Nonlinear Discrete Conti-
nuous Optimization Problems)). English. En: Annals of Operations Res. 139.1 (2005), 267 -
288.
[10] Dimian, A.C., Bildea, C.S. y Kiss, A.A. Integrated Design and Simulation of Chemical Proces-
ses. Computer Aided Chemical Engineering. Elsevier Science, 2003.
[11] Luyben, W.L. Principles and Case Studies of Simultaneous Design. Wiley, 2011.
[12] Smith, R. Chemical process design and integration. Wiley, 2005.
[13] Turton, R. y col. Analysis, Synthesis and Design of Chemical Processes. Prentice Hall Interna-
tional Series in the Physical and Chemical Engineering Sciences. Pearson Education, 2008.
[14] Caballero, J.A. y Grossmann, I.E. ((Aggregated Models for Integrated Distillation Systems)).

En: Industrial & Engineering Chemistry Research 38.6 (1999), pags. 2330-2344.
[15] Harsh, M.G., Saderne, P. y Biegler, L.T. ((A mixed integer flowsheet optimization strategy for
a Cthe debottlenecking problem)). En: Computers & Chemical Enginee-
process retrofits

ring 13.8 (1989), pags. 947 -957.
[16] Diwekar, U.M., Grossmann, I.E. y Rubin, E.S. ((An MINLP process synthesizer for a sequential
modular simulator)). En: Ind. Eng. Chem. Res. 31.1 (1992), pags.
313-322.
[17] Kocis, G.R. y Grossmann, I.E. ((A modelling and decomposition strategy for the minlp optimi-
zation of process flowsheets)). En: Computers & Chemical Engineering 13.7 (1989), pags.
797
-819.
[18] Reneaume, J.M.F., Koehret, B.M. y Joulia, X.L. ((Optimal Process Synthesis in a Modular Si-
mulator Environment: New Formulation of the Mixed-Integer Nonlinear Programming Pro-
blem)). En: Industrial & Engineering Chemistry Research 34.12 (1995), pags.
4378-4394.
[19] Daz, M.S. y Bandoni, J.A. ((A mixed integer optimization strategy for a large scale chemical
plant in operation)). En: Computers & Chemical Engineering 20.5 (1996), pags. 531 -545.
[20] Ilacqua, A. y Serrati, A. ((Simulador a medida de una plnata de etileno para estudios de

sensibilidad parametrica ). En: XVI Congreso Interamericano de Ingeniera
y optimizacion)
Qumica. Buenos Aires, Argentina (1991).
[21]
Caballero, J.A., Milan-Yanez, D. y Grossmann, I.E. ((Optimal synthesis of distillation columns:
Integration of process simulators in a disjunctive programming environment)). En: European

261
4.4.REFERENCIAS

Symposium on Computer-Aided Process Engineering-15, 38th European Symposium of the


Working Party on Computer Aided Process Engineering. Ed. por Puigjaner, Luis y Espu na,

Antonio. Vol. 20. Computer Aided Chemical Engineering. Elsevier, 2005, pags. 715 -720.
[22]
Caballero, J.A., Milan-Yanez, D. y Grossmann, I.E. ((Rigorous Design of Distillation Columns:
Integration of Disjunctive Programming and Process Simulators)). En: Industrial & Enginee-

ring Chemistry Research 44.17 (2005), pags. 6760-6775.
[23] Duran, M.A. y Grossmann, I.E. ((An outer-approximation algorithm for a class of mixed-
integer nonlinear programs)). En: Mathematical Programming 36 (3 1986), pags.
307-339.
[24] Kocis, G.R. y Grossmann, I.E. ((Relaxation strategy for the structural optimization of process
flowsheets)). En: Industrial & Engineering Chemistry Research 26.9 (1987), pags.
1869-1880.
[25] Viswanathan, J. y Grossmann, I.E. ((A combined penalty function and outer-approximation
method for {MINLP} optimization)). En: Computers & Chemical Engineering 14.7 (1990),

pags. 769 -782.
[26] Brunet, R. y col. ((Minimization of the LCA impact of thermodynamic cycles using a combi-
ned simulation-optimization approach)). En: Appl. Therm. Eng. 48.0 (2012), pags.
367 -377.
[27] Caballero, J.A. y col. ((Design of Hybrid Distillation-Vapor Membrane Separation Systems)).

En: Industrial & Engineering Chemistry Research 48.20 (2009), pags. 9151-9162.
[28] Caballero, J.A., Odjo, A. y Grossmann, I.E. ((Flowsheet optimization with complex cost and
size functions using process simulators)). En: AIChE Journal 53.9 (2007), pags.
2351-2366.
[29] Brunet, R. y col. ((Cleaner Design of Single-Product Biotechnological Facilities through the
Integration of Process Simulation, Multiobjective Optimization, Life Cycle Assessment, and
Principal Component Analysis)). En: Industrial & Engineering Chemistry Research 51.1 (2012),

pags. 410-424.
[30] Brunet, R. y col. ((Hybrid simulation-optimization based approach for the optimal design of
single-product biotechnological processes)). En: Computers & Chemical Engineering 37.0

(2012), pags. 125 -135.
[31] Tolsma, J.E., Clabaugh, J.A. y Barton, P.I. ((Symbolic Incorporation of External Procedures
into Process Modeling Environments)). En: Industrial & Engineering Chemistry Research 41.16

(2002), pags. 3867-3876.
[32] Dantus, M.M. e High, K.A. ((Evaluation of waste minimization alternatives under uncertainty:
a multiobjective optimization approach)). En: Computers & Chemical Engineering 23.10

(1999), pags. 1493 -1508.
[33] Leboreiro, J. y Acevedo, J. ((Processes synthesis and design of distillation sequences using
modular simulators: a genetic algorithm framework)). En: Computers & Chemical Enginee-

ring 28.8 (2004), pags. 1223 -1236.
[34] Guterrez-Antonio, C. y Briones-Ramrez, A. ((Pareto front of ideal Petlyuk sequences using a
multiobjective genetic algorithm with constraints)). En: Computers & Chemical Engineering

33.2 (2009), pags. 454 -464.
[35]
Guterrez-Antonio, C., Briones-Ramrez, A. y Jimenez-Guti
errez, A. ((Optimization of Petlyuk
sequences using a multi objective genetic algorithm with constraints)). En: Computers &
Chemical Engineering 35.2 (2011), pags. 236 -244.
[36] Vazquez-Castillo, J.A. y col. ((Design and optimization, using genetic algorithms, of inten-
sified distillation systems for a class of quaternary mixtures)). En: Computers & Chemical
Engineering 33.11 (2009), pags. 1841 -1850.

262
Captulo 4. Resultados

[37] Odjo, A.O. y col. ((Disjunctive-Genetic Programming Approach to Synthesis of Process Net-
works)). En: Industrial & Engineering Chemistry Research 50.10 (2011), pags.
6213-6228.
[38] Eslick, J.C. y Miller, D.C. ((A multi-objective analysis for the retrofit of a pulverized coal po-
wer plant with a {CO2} capture and compression process)). En: Computers & Chemical
Engineering 35.8 (2011), pags. 1488 -1500.
[39] Torres, C.M. y col. ((Evaluation Tool for the Environmental Design of Chemical Processes)).

En: Industrial & Engineering Chemistry Research 50.23 (2011), pags. 13466-13474.
[40] Torres, C.M. y col. ((An automated environmental and economic evaluation methodology
for the optimization of a sour water stripping plant)). En: Journal of Cleaner Production 44.0

(2013), pags. 56 -68.
[41] Coath, G. y Halgamuge, S. ((A comparison of constraint-handling methods for the appli-
cation of particle swarm optimization to constrained nonlinear optimization problems)). En:
Evolutionary Computation, 2003. CEC 03. The 2003 Congress on. Vol. 4. 2003, 2419 - 2425.
[42] Michalewicz, Z. ((A Survey of Constraint Handling Techniques in Evolutionary Computation
Methods.)) En: Evolutionary Programming 4 (1995), pags.
135-155.
[43] Michalewicz, Z. y col. ((Evolutionary algorithms for constrained engineering problems)). En:

Computers & Industrial Engineering 30.4 (1996), pags. 851 -870.
[44] Michalewicz, Z. y Schoenauer, M. ((Evolutionary algorithms for constrained parameter opti-
mization problems)). En: Evol. Comput. 4.1 (1996), pags.
1-32.
[45] Caballero, J.A. y Grossmann, I.E. ((Rigorous flowsheet optimization using process simulators
and surrogate models)). En: 18th European Symposium on Computer Aided Process Engi-
neering. Ed. por Braunschweig, Bertrand y Joulia, Xavier. Vol. 25. Computer Aided Chemi-

cal Engineering. Elsevier, 2008, pags. 551 -556.
[46] Caballero, J.A. y Grossmann, I.E. ((An algorithm for the use of surrogate models in modular
flowsheet optimization)). En: AIChE Journal 54.10 (2008), pags.
2633-2650.
[47] Turkay, M. y Grossmann, I.E. ((Logic-based MINLP algorithms for the optimal synthesis of
process networks)). En: Computers & Chemical Engineering 20.8 (1996), pags.
959 -978.
[48] Squire, W. y Trapp, G. ((Using Complex Variables to Estimate Derivatives of Real Functions)).

En: SIAM Review 40.1 (1998), pags. 110-112.
[49] Krige, D.G. ((A Statistical Approach to Some Mine Valuation and Allied Problems on the
Witwatersrand)). En: Masters Thesis, University of Witwatersrand (1951).
[50] Davis, E. y Ierapetritou, M. ((A kriging method for the solution of nonlinear programs with
black-box functions)). En: AIChE Journal 53.8 (2007), pags.
2001-2012.
[51] Henao, Carlos A. y Maravelias, Christos T. ((Surrogate-based superstructure optimization
framework)). En: AIChE Journal 57.5 (2011), pags.
1216-1232.
[52] Jones, D.R. ((A Taxonomy of Global Optimization Methods Based on Response Surfaces)).

English. En: Journal of Global Optimization 21.4 (2001), pags. 345-383.
[53] Jones, D.R., Schonlau, M. y Welch, W.J. ((Efficient Global Optimization of Expensive Black-
Box Functions)). English. En: Journal of Global Optimization 13.4 (1998), pags.
455-492.
[54] Sasena, M.J. Flexibility and efficiency enhancements for constrained global design optimi-
zation with kriging approximations. University of Michigan, 2002.
[55] MATLAB. The Language of Technical Computing. The Mathworks Inc., 2006.
[56]
Holmstrom, Kenneth. ((The TOMLAB optimization environment in Matlab)). En: Adv. Model

Optim. 1 (1999), pags. 47-69.

263
4.4.REFERENCIAS

[57] Quesada, I. y Grossmann, I.E. ((An LP/NLP based branch and bound algorithm for convex
MINLP optimization problems)). En: Computers & Chemical Engineering 16 (1992), pags.
937
-947.
[58] Brunet, R. y col. ((Combined simulation-optimization methodology for the design of en-
vironmental conscious absorption systems)). En: Computers & Chemical Engineering 46.0

(2012), pags. 205 -216.
[59] Raman, R. y Grossmann, I.E. ((Modelling and computational techniques for logic based
integer programming)). En: Computers & Chemical Engineering 18.7 (1994), pags.
563 -578.
[60] Dhole, V.R. y Linnhoff, B. ((Total site targets for fuel, co-generation, emissions, and cooling)).
En: Computers & Chemical Engineering 17, Supplement 1.0 (1993), S101 -S109.
[61] Iyer, R.R. y Grossmann, I.E. ((Synthesis and operational planning of utility systems for multipe-
riod operation)). En: Computers & Chemical Engineering 22.7-8 (1998), pags. 979 -993.
[62] Nishio, M. y col. ((A Thermodynamic Approach to Steam-Power System Design)). En: Indus-

trial & Engineering Chemistry Process Design and Development 19.2 (1980), pags. 306-312.
[63] Papoulias, S.A. y Grossmann, I.E. ((A structural optimization approach in process synthesis I:
Utility systems)). En: Computers & Chemical Engineering 7.6 (1983), pags.
695 -706.
[64] Mavromatis, S.P. y Kokossis, A.C. ((Conceptual optimisation of utility networks for operational
variations - I. targets and level optimisation)). En: Chemical Engineering Science 53.8 (1998),

pags. 1585 -1608.
[65] Mavromatis, S.P. y Kokossis, A.C. ((Conceptual optimisation of utility networks for operational
variations - II. Network development and optimisation)). En: Chemical Engineering Science

53.8 (1998), pags. 1609 -1630.
[66] Bruno, J.C. y col. ((A Rigorous MINLP Model for the Optimal Synthesis and Operation of Utility
Plants)). En: Chemical Engineering Research and Design 76.3 (1998), pags.
246 -258.
[67] Manninen, J. y Zhu, X.X. ((Level-by-level flowsheet synthesis methodology for thermal system
design)). En: AIChE Journal 47.1 (2001), pags.
142-159.
[68] Varbanov, P.S., Doyle, S. y Smith, R. ((Modelling and Optimization of Utility Systems)). En: Che-
mical Engineering Research and Design 82.5 (2004), pags. 561 -578.
[69] Westerberg, A.W. y col. Process Flowsheeting. Cambridge University Press, 1979.
[70] Marshman, D.J. y col. ((Energy optimization in a pulp and paper mill cogeneration facility)).

En: Applied Energy 87.11 (2010), pags. 3514 -3525.

264
Captulo 4. Resultados

4.5 Anexo II: Integration of modular process simulators under


the Generalized Disjunctive Programming framework for
the structural flowsheet optimization

de procesos qumicos se centra en optimizar para-


Generalmente, la optimizacion
metros de proceso continuos (temperaturas, presiones, caudales, composiciones, etc),
mientras se mantiene fija la topologa (estructura) del diagrama de flujo. Una herra-
mienta muy popular para realizar esta tarea son los simuladores de procesos modu-
pero pier-
lares, los cuales se adaptan perfectamente a los problemas de simulacion,
y de sntesis. Ademas,
den parte de su atractivo para los problemas de optimizacion
la sntesis de proceso qumico suele exigir la toma de decisiones relacionadas con la
de variables enteras, lo que trans-
topologa del proceso, lo que implica la inclusion
forma los problemas en problemas MINLP. Este hecho presenta tanto oportunidades
como desafos para los investigadores a la hora de desarrollar nuevas herramientas
que faciliten la sntesis de procesos qumicos.

de los procesos qumicos en los que no se mantiene fija la topo-


La optimizacion
de procesos.
loga de diagrama de flujo es un desafo en el campo de optimizacion
de modelos basados en ecuaciones. Este
Por lo general, esta tarea se realiza a traves

enfoque presenta varios problemas, por un lado el calculo de propieda-
o estimacion
des fsicas de forma rigurosa suele ser una tarea muy tediosa y complicada, y por otro
estos modelos suelen poseer tama
nos enormes (con miles de ecuaciones y variables)

lo que dificultad su resolucion.

En este trabajo se presenta un nuevo marco de modelado para tratar los proble-

mas de sntesis. En este, las diferentes alternativas del proceso se modelan en forma
sinergica
de una superestructura, y se resuelve utilizando una combinacion de simula-
GDP. Hasta donde conocemos,
dores de procesos comerciales junto a la formulacion
de problemas donde la superestruc-
no existe ninguna herramienta para la resolucion

tura del proceso queda dibujada directamente sobre la interfaz grafica del simulador
de procesos (GUI). La estructura del modelo se centra en construir un problema GDP
que es conectado con un simulador de procesos (Hysys Aspen), donde se optimiza

tanto la topologa como los parametros continuos del proceso. Esta metodologa per-

265
4.5.Integration of simulators under the GDP framework


mite incluir facilmente
restricciones y relaciones logicas entre las alternativas, de tal
de soluciones factibles. La herramienta propuesta uti-
forma que facilite la obtencion

liza el algoritmo de aproximaciones exteriores basada en la logica y por lo tanto no
es necesario volver a formular el problema como un MINLP.

de meta-
A modo de ejemplo se presenta la sntesis de un planta de produccion
nol. En este caso se presenta una superestructura que incluye todas las alternativas
incluyendo diferentes alimentos, diferentes sistemas de compresion
de interes, y dife-
El objetivo de este ejemplo es demostrar el compor-
rentes condiciones de reaccion.
tamiento de la metodologa cuando existen diferentes alternativas para cada una
de las operaciones unitarias (disyunciones).

la metodologa propuesta presenta un nuevo enfoque que com-


En conclusion,
GDP con los beneficios de los simuladores de
bina la flexibilidad de la formulacion
proceso comerciales, donde la principal novedad radica en el hecho de que la su-
integrada directamente en la interfaz grafica
perestructura del proceso esta del simu-
lador. Esta herramienta mejora el modelado basado completamente en ecuaciones

ya que el calculo riguroso de las propiedades fsicas y de las operaciones unitarias lo
el uso de la formulacion
llevan a cabo los simuladores comerciales. Ademas GDP uni-
da a los simuladores permite solucionar problemas de procesos donde la topologa
no se encuentra totalmente fijada.

266
Captulo 4. Resultados

Integration of modular process simulators under the Generalized


Disjunctive Programming framework for the structural flowsheet
optimization

Ruben
Miguel A. Navarro-Amoros, Ruiz-Femenia, Jose
A. Caballero

Article sent to Computers & Chemical Engineering. Under Review

Abstract

The optimization of chemical processes where the flowsheet topology is not kept
fixed is a challenging discrete-continuous optimization problem. Usually, this task has
been performed through equation based models. This approach presents several pro-
blems, as tedious and complicated component properties estimation or the handling
of huge problems (with thousands of equations and variables). We propose a GDP ap-
proach as an alternative to the MINLP models coupled with a flowsheet program. The
novelty of this approach relies on using a commercial modular process simulator where
the superstructure is drawn directly on the graphical use interface of the simulator. This
methodology takes advantage of modular process simulators (specially tailored nu-
merical methods, reliability, and robustness) and the flexibility of the GDP formulation
for the modeling and solution. The optimization tool proposed is successfully applied
to the synthesis of a methanol plant where different alternatives are available for the
streams, equipment and process conditions.

Keywords

Process synthesis, Generalized Disjunctive Programming, Modular simulators, logic-


based optimization algorithm

4.5.1. Introduction

One common approach for the optimization of real chemical process handles con-
tinuous process parameters (temperatures, pressures, flowrates, compositions, etc.) as
the unique optimization variables while the flowsheet topology is kept fixed. A popular

267
4.5.Integration of simulators under the GDP framework

tool to perform this task are the process simulators based on the modular architecture,
which are perfectly suited for simulation problems but loses part of its attractiveness
for optimization or synthesis problems. In addition, chemical process synthesis also de-
mands to make decisions related to process topology, which implies the inclusion of
integer variables as free variables in the model, leading to a Mixed- Integer Nonlinear
Programming (MINLP) problem [1, 2]. This fact presents both opportunities and cha-
llenges for researchers to develop new tools that facilitate the synthesis of chemical
plants to chemical engineers.

The Generalized Disjunctive Programming (GDP) modeling framework introduced


by Raman and Grossmann [3] has brought to Process System Engineering (PSE) com-
munity the powerful framework of the disjunctive programming, which was originally
developed by Balas [4, 5] as an alternative representation of mixed-integer program-
ming problems. GDP allows to model chemical plant synthesis problems through the
use of higher level of logic constructs [3, 6] that make the formulation step more in-
tuitive and systematic, retaining in the model the underlying logical structure of the
problem. GDP represents problems in terms of Boolean and continuous variables, allo-
wing the representation of constraints as algebraic equations, disjunctions and logic
propositions [7]. The development of GDP in the chemical engineering community has
led to the development of customized algorithms that exploit this alternative modeling
framework. In particular, Turkay and Grossmann [8] extended the outer approximation
(OA) algorithm [9] for MINLPs into a logical-equivalent algorithm. Later, Lee and Gross-
mann [10] developed a disjunctive branch and bound.

GDP techniques have been successfully incorporated to many types of PSE opti-
mization problems such as process flowsheet synthesis, design of distillation columns,
scheduling and design of batch processes. In 1996, Turkay and Grossmann published
a paper in which they proposed a GDP algorithm for structural flowsheet optimization
problem and tested on several examples, including the synthesis of a vinyl chloride
monomer process consisting of 32 units. Process synthesis with heat integration was al-
so solved using disjunctions and logic propositions by Grossmann and coworkers [11].
One year later, Caballero and Grossmann [12] reported an aggregated model for the
synthesis of heat-integrated distillation columns modeled as a generalized disjunctive
program. Later, a disjunctive programming model was also applied to the synthesis

268
Captulo 4. Resultados

of distillation column sequences [13]. In all these works, the problem is entirely descri-
bed on explicit equations by a general modeling language system, like GAMS [14],
and usually relies on simplified models (i.e, shortcut or aggregated methods) for the
unit operations in the flowsheet and for the prediction of the physical properties of the
components (e.g., for the vapor-liquid equilibrium). The first feature of this approach
leads to difficulties, when modeling the problem, in the initialization step, which may
converted into a daunting task. On the other hand, the use of simplified models for the
unit operations could be not accurate enough to capture key aspects of a real che-
mical process plant. Moreover, using simplified physical property models can predict
inaccurate thermodynamic properties, leading to misleading results.

The disadvantages listed in the last paragraph can be overcome by incorporating


process simulators to the synthesis problem. Flowsheeting software provides realistic
simulations and hence an optimal solution closer to the real implementation as they
offer tailored numerical techniques developed for converging the different units and
provides an extensive component database and reliable physical property methods.
The usage of chemical process simulators as an implicit model for solving synthesis pro-
blems through a MINLP approach is not new. Harsh et al. [15] developed an interface
with a MINLP and FLOWTRAN, for the retrofit of an ammonia process. Diwekar et al [16]
proposed a MINLP synthesizer using Aspen Plus. Diaz and Bandoni [17] used a MINLP
formulation with an existing ad-hoc process simulator for the optimization of a real
ethylene plant. Caballero et al. [18] proposed a superstructure-based optimization al-
gorithm for the rigorous design of distillation columns that combines a process simulator
(Aspen HYSYS) with explicit equations. Latter Brunet et al. [19] used the same algorithm
for the optimization of an ammonia-water absorption cooling cycle implemented in
Aspen Plus. Flowsheet process optimization with heat integration has also been perfor-
med using an hybrid simulation optimization approach, in which the process is solved
by a commercial process simulator (Aspen HYSYS), and the heat integration model is
in equation form [20]. All these works are based on the augmented penalty/equality
relaxation outer-approximation algorithm [21]. Other process simulators (SuperPro) has
also been coupled with a multi-objective Matlab optimizer [22].

Another approach for the synthesis problem combines process simulators with me-
taheuristic algorithms. Although metaheuristic algorithms are not able to guarantee

269
4.5.Integration of simulators under the GDP framework

the optimality of the solutions found, they can find solutions for some real-world pro-
blems that exhibit high levels of complexity [23]. Perhaps the most serious disadvanta-
ges of metaheuristic algorithms are that the number of function evaluations to conver-
ge could be large, and as well as they exhibit poor performance in highly constrained
systems. A considerable amount of literature supports the integration of a process si-
mulator with an external optimizer based on metaheuristic algorithms. Gross and Roo-
sen [24] demonstrated the suitability of a genetic algorithm coupled with the process
simulator Aspen Plus to optimize arbitrary flowsheets. Leboreiro and Acevedo [25] al-
so succeeded in problems where deterministic mathematical algorithms had failed,
using an optimization framework for the synthesis of complex distillation sequences
based on a modified GA coupled with Aspen Plus. The same combination of process
simulator and metaheuristic algorithm is adopted by Vazquez-Castillo et al. [26] to
address the optimization of five distillation sequences. Subsequent works used a multi-
objective GA [27] for the optimization of thermally coupled distillation systems [28-30],
and for the retrofit of a subcritical pulverized coal power plant with an MEA-based car-
bon capture and CO2 compression system [31]. Finally, Odjo et al. [32] also presented
a general framework for the synthesis of chemical processes using a hybrid approach
with Hysys and genetic algorithms.

In this paper we present a new modeling framework for dealing with superstructure-
based synthesis problems that exploits the synergistic combination of commercial pro-
cess simulators with GDP formulation and their corresponding logic-based solution al-
gorithms. As far as these authors know, it has not been reported a simulation optimi-
zation tool for solving the synthesis of chemical plants whose superstructure is drawn
directly on the process simulator graphical user interface (GUI). We achieve this aim
by developing a GDP modeling system that interfaces with a process simulator (Aspen
Hysys) at the NLP step to optimize the structure and parameters of a methanol plant
based on a superstructure which involves alternative equipment, process conditions
and stream configurations. Our methodology allows easily including soft constraints
and logical relationships among alternatives, which ensure feasible solutions. The pro-
posed tool uses the logic based Outer Approximation algorithm and hence it is not
required to reformulate the problem as an MINLP.

270
Captulo 4. Resultados

The remainder of this article is organized as follows. The problem statement is first for-
mally expressed. Then the methodology is introduced. In this section, the logic based
outer approximation algorithm, integration of the process simulator in the algorithm
and the connection with the external optimization solver are described. The propo-
sed simulation-optimization framework is illustrated through a case study based on a
methanol plant in the next section, where the superstructure and the disjunctions are
presented. In this section, the results are also briefly described. Finally, we draw the
conclusions from this work.

4.5.2. Problem statement

Given a superstructure for the synthesis of chemical process plant, with some spe-
cifications fixed, determine the optimal process flowsheet that leads to the maximum
value of an economic indicator. The solution must include both topological and ope-
rational (temperatures, pressures, flow rates) information.

4.5.3. Methodology

We have developed a modeling system with the following characteristics:

1. The complete modeling system is developed in Matlab [33].

2. Indexing capacities for both algebraic equations and implicit models.

3. Use of Boolean variables, disjunctions and logic propositions. Allowing the direct
formulation of the problem as a disjunctive problem without MINLP reformula-
tion.

4. Interfaced with different commercial solvers for NLP, LP, MILP models through
Matlab- Tomlab [34], and with homemade implementations of the logic based
Outer Approximation algorithm [8].

5. Communication with process simulators and other third party models, except
those developed in Matlab, is done by the Windows COM capabilities.

271
4.5.Integration of simulators under the GDP framework

Figure 4.5.1 shows and scheme of the tool we have implemented within Matlab
environment with the Aspen HYSYS process simulator embedded, and allowing the
user to model the optimization problem under the principles of GDP. To illustrate how
our tool is used, we have added a video in the supplementary material section that
describes the entirely process step by step.

4.5.3.1. Logic Based Outer Approximation algorithm with an embedded process si-
mulator

As mentioned above, we use the Logic-based Outer Approximation algorithm [8]


to fully exploit the structure of the GDP representation of our problem. The Logic-Based
OA shares the main idea of the traditional OA for MINLP, which is to solve iteratively a
master problem given by a linear GDP, leading to a lower bound of the solution (z LB ),
and an NLP subproblem, which provides an upper bound (z U B ). The general structure
of a nonconvex GDP formulation is as follows:

min z = f (x)
x,Yik

s.t. h(x) = 0

g(x) 0

Yik

rik (x) 0

kK
(4.5.1)
iDk
sik (x) = 0

(Y ) = T rue

xlo x xup

x Rn , Yik {T rue, F alse}, i Dk , k K

where x is a vector of continuous variables representing pressures, temperatures


and flow rates of the streams in a process flowsheet superstructure. The objective fun-
ction is normally a cost function of the continuous variables. The common equality set
of constraints h(x) = 0 represents equipment rating equations and mass and energy

272
Captulo 4. Resultados

Figura 4.5.1: Scheme of the optimization modeling framework

balances; and the common set of inequalities g(x) 0 , represent design specifi-
cations. Both sets of constraints must hold true regardless the discrete decisions. The
underlying alternatives in the superstructure are represented in the continuous space

273
4.5.Integration of simulators under the GDP framework

by a set of disjunctions k K, each of which contains i Dk terms. Each term of


the disjunction represents the potential existence of an equipment (or stream) i for
performing a processing task k , and has associated a Boolean variable Yik and a
set of constraints rik 0 and sik = 0 , which are normally associated with the invest-
ment and operations cost, the energy and mass balance, and physical and chemical
equilibrium for the particular equipment . When the term is not active (Yik = F alse),
the corresponding constraints are ignored. Finally, the symbolic equation (Y ) = T rue
represents the set of logic propositions that relates the Boolean variables, which in PSE
generally indicates the logic implications among the equipment to define a feasible
topology for the process flowsheet.

It is noteworthy that for the general GDP formulation 4.5.1, the terms in the disjun-
ctions are linked by an inclusive OR logical operator (i.e., A B is true if A or B or
both are true). Nevertheless, we have implemented a logic-based OA algorithm that
requires to reformulate the disjunctive part of the problem as a set of special type of
disjunctions, each of which contains only two terms and in one of them all the varia-
bles are set to zero. Fortunately, any disjunction in its general form has a straightforward
reformulation to the special 2 terms disjunction (see Appendix 1).

As we are dealing with a process simulator embedded in a GDP formulation, it is


convenient to define a partition of x into dependent xD and independent (or design)
variables xI . The latter is the set of optimization variables and its dimension is equal
to the degrees of freedom of the nonlinear problem obtained when the binary varia-
bles are fixed. By this partition the common equality constraint h(x) can be solved for

the dependent variables xD given a vector of independent variables xI , xD = h(xI )
. In an analogous manner, for each equipment i assigned to a task k the depen-
dent variables associated to it can be expressed as functions of the decision variables

xD = s ik (xI ). In this work, dependent variables xD cannot explicitly written in terms of
decision variables, but they are implicitly calculated at the process simulator, and then
are used at the optimization level to evaluate the objective function and the common
and particular constraints. Accordingly, the GDP problem 4.5.1 can be rewritten as:

274
Captulo 4. Resultados

min z = f (xD , xI )
xD ,xI ,Yik

s.t. xD = hImplicit (xI )

h(xD , xI ) = 0

g(xD , xI ) 0

Yik
xD = sImplicit (xI )

(4.5.2)
ik
kK


iDk r (xD , xI ) 0
ik
sik (xD , xI ) = 0

(Y ) = T rue

xI,lo xI xI,up

xD RnD , xI RnI , Yik {T rue, F alse}, i Dk , k K

Note that in 4.5.2 as we introduce dependent variables in explicit equations (for


example in h(xD , xI ) = 0 or in g(xD , xI ) 0), a sequential function evaluation is
required, first the implicit models are solved and then the explicit constraints can be
evaluated.

Logic-based NLP subproblem

For fixed values of the Boolean variables Yik (i.e, given a flowsheet configuration)
the corresponding NLP subproblem is as follows:

275
4.5.Integration of simulators under the GDP framework

min z U B (Yik ) = f (x)


xD ,xI

s.t. xD = hImplicit (xI )

h(xD , xI ) = 0

g(xD , xI ) 0
(4.5.3)
xD = sImplicit
ik (xI )



rik (xD , xI ) 0 f or Yik = T rue, i Dk , k K


D I
s (x , x ) = 0


ik

xI,lo xI xI,up

xD RnD , xI RnI , Yik {T rue, F alse}, i Dk , k K

It is worthy to emphasize that only the constraints that belong to the selected equip-
l = T rue) are imposed. This leads
ment or stream (i.e. associate Boolean variable Yik
to a substantial reduction in the size of the NLP subproblem compared to the direct
application of the traditional OA method on the MINLP reformulation.

Logic-based Master problem

Assuming that L NLP subproblems are solved in which sets of linearizations are ge-
nerated for the objective function and the common constraints and particular cons-
l = T rue , we define the following
traints in the subsets of disjunction terms Lik = l : Yik
disjunctive OA master problem:

276
Captulo 4. Resultados

min z LB =
xI,Yik

s.t.

f (xD,l , xI,l ) + 5xl f (xD,l , xI,l )T (xl xI,l )



th,l [h(xD,l , xI,l ) + 5xl h(xD,l , xI,l )T (xl xI,l )] 0 l = 1, . . . , L


g(xD,l , xI,l ) + 5xl g(xD,l , xI,l )T (xl xI,l ) 0

(4.5.4)

Yik

rik (xD,l , xI,l ) + 5xl rik (xD,l , xI,l )T (xl xI,l ) 0 kK


iDk l Lik
ts,l [s (xD,l , xI,l ) + 5 l s (xD,l , xI,l )T (xl xI,l )] 0
ik x ik

(Y ) = T rue

xI,lo x xI,up

xD RnD , xI RnI , Yik {T rue, F alse}, i Dk , k K

where th and ts are assigned with either the value 1, 0 or 1, depending the sign
of the Lagrange multiplier of the corresponding nonlinear constraint. The constraints of
a particular disjunction term (equipment i for task k) are only included in the master
problem if the corresponding Boolean variable Yik is True, whereas linearizations of
temporally inactive terms are simply discarded. Again this property constitutes a major
difference to the standard OA method. Note that the master MILP 4.5.4 problem is not
a function of the dependent variables.

MILP reformulation of the master problem

The master problem of the logic-based OA algorithm can be reformulated as an


MILP using either Big-M (BM) or Hull Reformulation (HR) formulations. We apply the tigh-
ter formulation, that is HR, and then the disjunctions of the GDP master problem are
reformulated as follows:


rik (xD,l , xI,l )yik + 5xl rik (xD,l , xI,l )T (xl,i xI,l ) urik
l Lik , i Dk
ts,l

D,l , xI,l )y D,l , xI,l )T (xl,i xI,l )] us
ik [sik (x ik + 5xl sik (x

ik
k = 1, . . . , K

(4.5.5)
xI = iDk xI,i
P

xI,lo yik xI,i xI,up yik , i Dk

277
4.5.Integration of simulators under the GDP framework

where the superscripts lo and up denote the lower and upper bounds, respecti-
vely. Each vector of independent variables xI related to a potential equipment i is
disaggregated into as many new vector variables xI,i as the number of alternative
equipments (or streams) for potentially performing task k. The upper and lower bounds
applied for all the disaggregated variables with the binary variables yik are used to
force the variables to zero when the mode i in not selected for performing task stage
k.

The exclusive OR logic operator in the disjunction part of the GDP mater problem
4.5.4 is transformed into the following linear constraint:

X
yik = 1, k = 1, . . . , K (4.5.6)
iDk

Furthermore, we add a set of binary cuts [35] to exclude the previous solution for
the binary variables:

X X
yik yik | B l | 1, l = 1, . . . , L (4.5.7)
i,kB l i,kN l

where B l is the subset defined for each NLP subproblem that stores the binary
variables ysm with a value of 1 , and N l is the subset that collects the remaining binary
variables for that NLP subproblem; i.e., B l = s, m : Ysm
l = T rue and N l = s, m : Y l =
sm

F alse.

To avoid infeasible master problems caused by the nonconvexity of the GDP pro-
blem 4.5.1, we relax the linearized constraints in 4.5.5 by introducing positive slack va-
riables urik and usik , respectively. As the common constraints in Eq. 4.5.2 can be also
nonlinear, we add the slack variables uh and ug to the RHS of the common linearized
constraints in Eq. 4.5.4. These slack variables are included in the objective function th-
r , w s , w h and w g chosen to be sufficiently large.
rough a penalty term with weights wik ik
Accordingly, the objective function of the Master Problem is rewritten as:

278
Captulo 4. Resultados

XX XX
L r T r s T s
min ZLB =+ (wik ) uik + (wik ) uik + (wh )T uh + (wg )T ug (4.5.8)
xI ,Yik
i k i k

Some important remarks deserve special attention. GDP algorithms require con-
vexity to guaranty convergence to a global optimal solution. In an implicit model it
is difficult prove convexity, even in the case the model be convex, but in general we
must assume non convexity. Therefore, it is no guarantee to find a global solution, and
only locally optimal solutions should be expected.

4.5.3.2. Connection between Matlab and Aspen Hysys

The developers of Aspen HYSYS also followed the paradigm shift in the develop-
ment of process simulators, from procedural to objectedoriented programming. Ac-
cordingly, Aspen HYSYS is programmed with 32-bit C++ [36], which gives it the ability
to lend its functionalities to be used in other application software. That makes Aspen
HYSYS a very powerful and useful tool in the design of our hybrid framework. We use the
binary-interface standard Component Object Model (COM), by Microsoft, to interact
with Aspen HYSYS through the objects exposed by the developers of Aspen HYSYS. We
utilize Matlab as an automation client to access these objects and interact with Aspen
HYSYS, which works as an automation server (see Figure 4.5.1). By writing Matlab code,
it is possible to send and receive information to and from the process simulator. Thus,
the exposed objects make possible to perform nearly any action that is accomplished
through the Aspen HYSYS graphical user interface, allowing us to use Aspen HYSYS as
a calculation engine. According to the objected-oriented programming nomencla-
ture [37], in Aspen Hysys, the functions defined within each object (e.g., reactor) are
called methods (i.e., governing equations), and the variables contained in an object
are known as properties (e.g., feed composition).

279
4.5.Integration of simulators under the GDP framework

4.5.3.3. Connection between Matlab and Optimizer

We use the TOMLAB optimization environment, which provides an interface bet-


ween the Matlab model and the available optimization solvers. This tool allows us to
standardize the model definition and then use all the available solvers regardless the
different syntax required for each solver. We do not need to make a specific interface
routine for each optimization solver. We use the CPLEX version 12.2.0.0 solver for the
MILP problems and the CONOPT solver for the NLP supbroblems. The latter solver is ba-
sed on the Generalized Reduced Gradient (GRD), which is suitable for models where
feasibility is difficult to achieve.

As mentioned above, the modeling framework proposed does not require to rew-
rite the problem as an MINLP, allowing for direct application of solution methods to
problems formulated as GDP. To this aim, we implement the logic based OA algo-
rithm with the special feature that allows to use also implicit models (i.e., models inside
a process simulators). An implicit model can be treated as a black box with a rigid
input-output structure whose derivative information is not available. The models in a
modular chemical process simulator are accurate enough for simulation purposes, but
they could introduce some numerical noise (i.e., the solution varies slightly with iden-
tical initial values) that prevent the accurate determination of derivative information
[38]. We capture the gradient information by a finite difference approach with a per-
turbation size that balances and minimizes the error due to noise and the error in the
approximation of the Jacobian. As the perturbation size increases, the error in the ap-
proximation of the Jacobian becomes significant. On the other hand, the response of
a small perturbation may be corrupted by convergence noise. Here, it is appropriate
to mention that we do not follow this approach for recycles in the flowsheet, because
recycles behave like noise amplifiers. In this case, we let the NLP solver to converge the
recycles. Although, both the number of variables handles by the NLP solver and the
number of explicit equality constraints increases, in general the model is more robust
and usually the computational time does not increase because it is not required to
converge all the recycles each time the simulator is called.

280
Captulo 4. Resultados

4.5.4. Case Study

As an example to illustrate the correct behavior of the proposed methodology, we


present the case of the synthesis of methanol. This process has been studied extensi-
vely in the past [39-45]. Figure 2 shows a simplified flowsheet of the methanol process
using syngas as feed stream. Conventional methanol production uses a feed stock of
reformed methane that contains hydrogen, carbon monoxide and carbon dioxide in
a ratio of NH2 /(2NCO +3NCO2 ) close to the stoichiometric ratio of unity. The chemistry
of the methanol process involves a lot of reactions, but only three reactions are signifi-
cant, the synthesis of methanol from carbon monoxide (R-1), the synthesis of methanol
from carbon dioxide (R-2), and the water gas shift reaction (R-3).

CO + 2H2 CH3 OH o
4 Hrxn = 94,5kJ/mol (R-1)

CO2 + 3H2 CH3 OH + H2 O o


4 Hrxn = 53,0kJ/mol (R-2)

CO + H2 O CO2 + H2 o
4 Hrxn = 41,21kJ/mol (R-3)

Figura 4.5.2: Methanol simplified process flowsheet

The objective of this example is to maximize the profit of the process. We consider
two available feeds with different characteristics and prize (See Table 4.5.2). We also
consider two products in the process, a principal (and desired) product with a high
sale prize (0.25 C/kg) and a subproduct (purge stream) with a low sale price (0.018

281
4.5.Integration of simulators under the GDP framework

C/kg). The equipment cost is calculated using correlations from the literature, and to
this end we use the correlations given by Turton et al. [46], and also the prize of all used
utilities are obtained from this reference. Finally, we update the prices to 2012 using the
C
hemical Engineering Plant Cost Index(CEPCI). The annual cost of the equipment is
calculated for a time horizon (n) of 10 years and an interest rate per year (i) of 8 % [47]
using the following expression,

i(1 + i)n
Annualized capital cost = capital cost (4.5.9)
(1 + i)n 1

The simulation is performed using the sequential modular simulator Aspen-HYSYS


with the Soave-Redlich-Kwong (SRK) equation of state and default values.

To check the capabilities of our simulation-optimization tool, we build a superstruc-


ture of the methanol process that includes all the alternatives of interest (Figure 4.5.3).
Note that the aim of this example is to demonstrate the behavior of the methodo-
logy when different alternatives exist for each of unit operations (disjunctions). The key
of this process is the reactor operation. The formation of methanol, as a typical he-
terogeneously catalyzed reaction, can be described by absorption-desorption me-
chanism (Langmuir-Hinshelwood or Eley-Rideal). The synthesis of methanol is a pressu-
re and temperature dependent process. The carbon monoxide and carbon dioxide
conversions up to attainment of equilibrium are shown as a function of pressure and
temperature in Table 4.5.1.

In order to clearly illustrate the problem, for the sake of simplicity, but without loss of
generality we will focus on two operating conditions, one specifically working at 200C
and 50 atm (called Low Conversion Condition) and the other working at 200C and
100 atm (called High Conversion Condition). Note that this problem can be formula-
ted for the reactor operating in any combination of pressure and temperature, which
increases the number of alternatives.

282
Captulo 4. Resultados

Tabla 4.5.1: Temperature and pressure dependence of the carbon monoxide and car-
bon dioxide equilibrium conversions

CO conversion* CO2 conversion*


Temp (C)
50 atm 100 atm 300 atm 50 atm 100 atm 300 atm
200 96.3 99.0 99.9 28.6 83.0 99.5

250 73.0 90.6 99.0 14.4 45.1 92.4


300 25.4 60.7 92.8 14.1 22.3 71.0
350 -2.3 16.7 71.9 9.8 23.1 50.0
400 -12.8 -7.3 34.1 27.7 29.3 41.0
* Negative sign denote CO formation via Equation (R-3): CO + H2 O  CO2 + H2

Figura 4.5.3: Superstructure for the Methanol synthesis process in Aspen-Hysys

The superstructure presents in Figure 4.5.3 include the following alternatives:

Feed stream

Two different synthesis gas streams are available, both containing the reactants H2 ,
CO2 and CO, and a small amount of inert, CH4 . The characteristics of different feed
streams are shown in Table 4.5.2. For this example, we consider the option to select
only one feed stream. This is modeled with the following OR exclusive disjunction:

283
4.5.Integration of simulators under the GDP framework


Y1 Y2

F M F F F eed
F eed Cost = A
F M F F F eed
F eed Cost = B (4.5.10)

A B


F eed,LB F eed,U B F eed,LB F eed,U B
FA F F eed FA FB F F eed FB

F eed,U B
where F F eed is the molar flow rate of the synthesis gas feed stream selected. FA
F eed,LB F eed,U B F eed,LB
and FA , or FB and FB are the upper and lower bounds on the
F and M F are the average molecular weight of
availability of each feed stream; MA B
F and F are the costs of the
the syngas feed streams of type A and B, respectively; A B
syngas of type A and B, respectively; and we consider 8000 h of operation per year
( ).

Tabla 4.5.2: Synthesis gas streams data

Feed Min Molar Flow Max Molar Flow P Cost Composition


Stream (kmol/s) (kmol/s) (atm) ( C/kg) H2 CO CO2 CH4
Feed A 0.05 1.44 20 0.026 0.70 0.26 0.02 0.02
Feed B 0.05 1.89 20 0.020 0.75 0.17 0.05 0.03

As commented above, in the logic based outer approximation, we need two term
disjunctions in which one of them forces all the variables to be zero (or simply to be
discarded in the NLP problem). This can be done as follows. In the Appendix 1) we
include a general reformulation for an n-term disjunction. For this particular case we
reformulate 4.5.10 as follows:

284
Captulo 4. Resultados


Y1
Y1
F eed Cost = 0
F M F F F eed A


F eed CostA = A A A A

F F eed = 0
F eed,LB F eed F eed,U B A
FA FA FA

A = 0

Y2
Y2
F eed Cost = 0
F M F F F eed B


F eed CostB = B B B B (4.5.11)

F F eed = 0
F eed,LB F eed F eed,U B B
FB FB FB

B = 0

= A + B

F F eed = FA
F eed F eed
+ FB

F eed Cost = F eed CostA + F eed CostB

Y1 Y2

The rest of disjunctions are also reformulated as two term disjunctions even though
it is not explicitly stated in the text.

Feed Compression system

The feed enters the process at low pressure (20 atm) and must be compressed to
a higher pressure where reaction is feasible (in this case, we consider two possible
operation conditions, 50 or 100 atm). For compression, we assume the choice bet-
ween a single compressor (single stage compression) or a system consisting of two
compressor with intermediate cooling (two-stage compression). Furthermore, in each
compression system, the gas pressure can be increased to 50 bar or to a higher va-
lue of 100 bar. To model the latter choice for the single-stage compression alternative,
we define the Boolean variables Y3 and Y4 to operate either at low or high output
pressure respectively, and write following disjunction:


Y3 Y4
(4.5.12)
P F C,out = 50 atm P F C,out = 100 atm

285
4.5.Integration of simulators under the GDP framework

where P F C,out is the pressure of the stream leaving the feed compression system (In
Figure 4.5.3 corresponds with the pressure of the stream leaving compressor K-100).For
the case of two-stage compression with intermediate cooling, we also formulate a
disjunction with two terms, one corresponding to the low pressure (Y5 ) and the other
for the high pressure (Y6 ):


Y5 Y6

P F C,out = 50 atm P F C,out = 100 atm
(4.5.13)


30 atm P F C,intermediate 50 atm 40 atm P F C,intermediate 64 atm

40o C T F C,intermediate 70o C 40o C T F C,intermediate 60o C

where P F C,intermediate and T F C,intermediate are the intermediate pressure and


temperatures, respectively (In Figure 4.5.3, P F C,intermediate corresponds with the pres-
sure of the stream leaving compressor K-101, and T F C,intermediate corresponds with
the temperature of the stream leaving heat exchanger E-100) and P F C,out is the pres-
sure of the stream leaving the feed compression system (In Figure 4.5.3 corresponds
with the pressure of the stream leaving compressor K-102).

In the single stage compression 4.5.12 and the two-stage compression system 4.5.13
disjunctions, we link the two terms of each disjunction with the so-called logical opera-
tor at most one (a variation of the OR operator equivalent to A B) to not force
to select one of the two terms (i.e. the two Boolean variables associated with the low
and high pressure can be simultaneously false).

To avoid the combination with the four Boolean variables (Y3 , Y4 , Y5 and Y6 ) being
simultaneously false, which has no physical meaning as the gas feed must be com-
pressed, we add the following constraint to our optimization problem:

Y3 Y4 Y5 Y6 (4.5.14)

286
Captulo 4. Resultados

Reactor + flash units

The gas reaction (Eqs. R-1, R-2 y R-3) takes place in an high conversion expensi-
ve reactor or in a less expensive reactor working at lower conversion. The difference
between them concerns the pressure at which the reactions are produced. While the
expensive reactor works at 100 atm (high conversion), the other works at 50 atm (low
conversion). To model the latter choice reactor alternative, we define the Boolean va-
riables Y7 and Y8 to operate either at high or low conversion conditions respectively.
The characterization of each reactor is totally defined by the specification of degree
of conversion of the different compounds (CO and CO2 ). Note that the reaction R-3
is negligible under these operating conditions as compared with the other reactions.
The reactor is cooled using water at ambient conditions. The capital cost of the reac-
tor depends on its volume (for a detailed description of the volume calculation see
Appendix 2).

The next step in the process is the separation system. The vapor stream leaving
the reactor contains the desired product (methanol) and high concentration of light
components, as CO or CO2 . Therefore, a flash tank is used to remove most of the light
components and obtain methanol with desired composition. The combination of pres-
sure and temperature required in the flash unit for the desired methanol purity (molar
fraction 90 %) is attained by an expansion valve and a water-cooled heat exchan-
ger. Note that the lower and upper bounds of the pressure in flash unit are assigned
to the minimum (pressure of feed stream) and the maximum pressure of the system
(pressure in the reactor), respectively. Furthermore, the lower bound of temperature in
flash unit is 40C because of the use of water as refrigerant in heat exchanger, and the
upper bound is 140C due to higher values do not allow to reach the desired product
composition.

The choice of reactor and flash conditions are formulated with the following OR
exclusive disjunction:

287
4.5.Integration of simulators under the GDP framework


Y7 Y8


Xi,CO = 0,99
Xi,CO = 0,96



X2,CO2 = 0,83
X2,CO2 = 0,29
(4.5.15)

20 P F LASH 100 20 P F LASH 50

40 T F LASH 140 40 T F LASH 140

where X1,CO and X2,CO2 are the degree of conversion of CO and CO2 in reac-
tions R-1 and R-2, respectively. P F LASH corresponds with the pressure of the stream
leaving expansion valve V- 100 for disjunction Y7 and valve V-101 for disjunction Y8 ,
and T F LASH corresponds with the pressure of the stream leaving heat exchanger
E-101 for disjunction Y7 and valve E-103 for disjunction Y8 .

Heating/cooling before Reactor

The two operating conditions in the reactor, previously selected, implies that the
temperature of its inlet stream must be 200C. To get this, the resulting stream from the
sum of compressed feed stream and the recycled stream must be heated or cooled.
In this case, and after a previously sensitivity study, we know that only in the case of
using the single compressor at 100 atm, the temperature of the stream exceeds 200C,
and must be cooled. In all other cases, the stream is lower than 200C, and must be
heated.

To model this situation and guide the system to the correct choice, we define the
Boolean variables Y13 and Y14 to select a heater or a cooler, respectively. All the pre-
vious situations can be formulated as Booleans expressions:

Single stage compression until 50 atm implies heating: Y3 Y13


Single stage compression until 100 atm implies cooling: Y4 Y14

Two-stage compression with intermediate cooling until 50 atm implies


Y5 Y13
heating:

Two-stage compression with intermediate cooling until 100 atm implies


Y6 Y13
heating:

288
Captulo 4. Resultados

Note that the specification of the temperature of the outlet stream of the heat
exchanger (200C) is specified in the simulators. To simulate the cooler, we use a water-
cooled heat exchanger using water at ambient conditions as refrigerant. To simulate
the heater, we use a heat exchanger using high pressure steam as hot utility.

Recycled stream compression system

The recycled stream in the process must be compressed until the operation pressure
of the selected reactor (50 or 100 atm). As in the feed compression system, we assume
the choice between a single compressor (single-stage compression) or a system with
two compressor with intermediate cooling (two-stage compression). Furthermore, in
each system, the gas pressure can be increased to 50 bar or to a higher value of 100
bar. To model the latter choice for the single-stage compression alternative, we define
the Boolean variables Y9 and Y10 to operate either at low or high output pressure
respectively and we use the following disjunction:


Y9 Y10
(4.5.16)
P RC,out = 50 atm P RC,out = 100 atm

where P RC,out is the pressure of the stream leaving the feed compression system
(In Figure 4.5.3 corresponds with the pressure of the stream leaving compressor K-103).

For the case of two-stage compression with intermediate cooling, we also formula-
te a disjunction with two terms, one corresponding to the low pressure (Y11 ) and the
other for the high pressure (Y12 ):


Y11 Y12

P RC,out = 50 atm P RC,out = 100 atm
(4.5.17)


30 atm P RC,intermediate 50 atm 40 atm P RC,intermediate 64 atm

40o C T RC,intermediate 70o C 40o C T RC,intermediate 60o C

289
4.5.Integration of simulators under the GDP framework

where P RC,intermediate and T RC,intermediate are the intermediate pressure and


temperatures, respectively (In Figure 4.5.3, P RC,intermediate corresponds with the pres-
sure of the stream leaving compressor K-105, and T RC,intermediate corresponds with
the temperature of the stream leaving heat exchanger E-104) and P RC,out is the pres-
sure of the stream leaving the feed compression system (In Figure 4.5.3 corresponds
with the pressure of the stream leaving compressor K-104).

As in feed compression system, the two disjunctions require the following logic pro-
positions between the Boolean variables to ensure that at most one term is selected
in each disjunction:

at most Y9 , Y10 is equivalent to [Y9 Y10 ]


(4.5.18)
at most Y11 , Y12 is equivalent to [Y11 Y12 ]

To avoid the combination with the four Boolean variables (Y9 , Y10 , Y11 and Y12 )
being simultaneously false, which has no physical meaning as the gas feed must be
compressed, we add the following constraint to our optimization problem:

Y9 Y10 Y11 Y12 (4.5.19)

Apart from the disjunctions, the superstructure has other important characteristics
which must be commented:

Vent-to-recycle Split

An important characteristic of this example is the vent-to-recycle split. The vapor


stream leaving the flash unit contains both useful compounds (reactants as H2 , CO
and CO2 ) and waste products (inerts, CH4 and H2 O). In this situation, the split ratio
has an important effect in the global process. While higher flows in recycled stream
increase the recovery and reuse of reactants, the concentrations of inert products in
the system also increase, and as results, the compression costs increase. In addition,

290
Captulo 4. Resultados

high flows of vent stream reduce the compression cost in the system, but increase the
losses of reactants (hydrogen, carbon monoxide, and carbon dioxide).

In our case study, we define the split variable as an independent variable, which is
controlled by the optimizer.

4.5.5. Results

The optimal configuration is shown in Figure 4.5.4 and the computational results are
shown in Table 4.5.3.

Figura 4.5.4: Optimal configuration of the case study

As we discussed above, the first step in the methodology is the initialization of all
the units inside the disjunctions. In this case, this consists in selecting a minimum set
of feasible flowsheets in such a way that all the terms in the disjunctions be true at
least once. In this example, and the compression system entails 2 disjunctions (with 4
terms or alternatives in total), we need to solve 4 initial subproblem to cover all the
terms. Then, a Master problem is solved to provide a new set of Boolean variables that
produce better results than in previous solution. From Master results, we solved a NLP
problem to obtain the better feasible solution. To avoid that the algorithm stops early
due to the non-convex constraints, we use a stopping criterion based on the heuristic:

291
4.5.Integration of simulators under the GDP framework

Tabla 4.5.3: Computational Results

N of Boolean variables 14
N of independent variables 29
N of linear explicit equations 51
N of non-linear explicit equations 12
N of implicit blocks 14
Iterations* Objective CPU time (s) Solver
Initialization (4 Problems)
Init Problem 1 NLP -560.64 22.20 CONOPT
Init Problem 2 NLP -575.46 12.40 CONOPT
Init Problem 3 NLP -562.63 12.28 CONOPT
Init Problem 4 NLP -574.49 7.52 CONOPT
MILP Master 110.54 0.09 CPLEX
Iteration 5 NLP -573.82 19.97 CONOPT
* Pentium Dual-Core E5300 2.60GHz

stop as two consecutive NLP subproblem worsen. For our particular case, the optimal
solution is found at the initial problem 2 with an objective value of 57,55 MM C/year.

The optimal solution selects the low conversion reactor. In this case, the low cost in
the compression system (50 atm) compensate the lower conversion obtained with this
reactor. The main characteristics of the selected equipment are shown in Table 4.5.4.

292
Captulo 4. Resultados

Tabla 4.5.4: Results: Characteristic of equipment in optimal configuration


Compression System Recycled Stream Compression System
Single Single
Type Type
Compression Compression
Inlet Pressure (atm) 20,0 Inlet Pressure (atm) 36,7
Compression ratio 2,5 Compression ratio 1,4
Power (MW) 6,662 Power (MW) 0,07
Cost Cost
Capital Cost (C) 822338 Capital Cost ( C) 51006
Annualized Capital Cost (C/year) 122553 Annualized Capital Cost ( C/year) 7601
Electricity Cost (C/year) 2309701 Electricity Cost ( C/year) 24198

Heating or Cooling Reactor System


Low Conversion
Type Heater Type
Reactor
Area (m2 ) 85,6 Volume (m3) 66,5
Inlet - Outlet Temp (C) 143,4 - 200 Cold Utility Water
Hot Utility Steam HP(41atm) Water flow (kg/h) 1527336
Steam Flow (kg/h) 6687,2 Energy (MW) 34,040
Energy (MW) 3,198 COST
COST Capital Cost (C) 49037
Capital Cost (C) 95792 Annualized Capital Cost ( C/year) 7308
Annualized Capital Cost (C/year) 14276 Catalyst Cost ( C/year) 173506
Utility Cost (C/year) 856503 Utility Cost ( C/year) 328283

Heating or Cooling
Area (m2 ) 636.5
Inlet - Outlet Temp (C) 196.7 - 59.9
Cold Utility Water
Water Flow (kg/h) 799778
Energy (MW) 19.710
COST
Capital Cost (C) 191253
Annualized Capital Cost (C/year) 28502
Utility Cost (C/year) 190083

293
4.5.Integration of simulators under the GDP framework

4.5.6. Conclusions

It has been shown that the synthesis problem can be written as a General Disjun-
ctive Programming problem and solved without MINLP reformulation using the logic
based outer approximation with a process simulator embedded. Conceptually the
GDP approach facilitates the model formulation for the final user retaining in the mo-
del the underlying logical structure of the problem. We proposed a novel approach
that combines the flexibility of the GDP formulation with the benefits of the commercial
process simulators, where the novelty relies on the advantage that the superstructure
of the process is directly built in the graphical user interface of the simulator. Our tool
improves the approach that uses a full equation based model in a GDP formulation,
due to process simulators facilitate the task to predict the thermophysical behavior of
the process using rigorous models. Furthermore, our framework offers advantages over
the approach that indeed integrates a simulator but does not take profit of a GDP for-
mulation (the problem is reformulated into an MINLP) or fixes the topology of the flows-
heet. The proposed approach is illustrated through a case study for the production of
methanol, where some constraints are fixed and we propose several alternatives for
some streams, tasks (one and two stages compression system; two types of reactors),
and process conditions (low and high pressures). We have proven that the proposed
systematic approach provides a and intuitive way for the chemical engineers to synt-
hesize rigorously chemical processes. Finally, we illustrate our tool with a video that
shows the proposed methodology for the methanol case study (see supplementary
material).

Acknowledgements

The authors wish to acknowledge support from the Spanish Ministry of Science and
Innovation (CTQ2012-37039-C02-02).

294
Captulo 4. Resultados

Referencias

[1] Biegler, L.T., Grossmann, I.E. y Westerberg, A.W. Systematic methods of chemical process
design. Prentice-Hall international series in the physical and chemical engineering scien-
ces. Prentice Hall PTR, 1997.
[2] Grossmann, I.E. ((Review of Nonlinear Mixed-Integer and Disjunctive Programming Techni-
ques)). English. En: Optimization and Engineering 3.3 (2002), pags.
227-252.
[3] Raman, R. y Grossmann, I.E. ((Modelling and computational techniques for logic based
integer programming)). En: Computers & Chemical Engineering 18.7 (1994), pags.
563 -578.
[4] Balas, E. ((Disjunctive programming)). En: Annals of Discrete Mathematics 5 (1979), pags.
3-51.
[5] Balas, E. ((Disjunctive programming: Properties of the convex hull of feasible points)). En:

Discrete Applied Mathematics 89.1-3 (1998), pags. 3 -44.
[6] Hooker, J.N. y Osorio, M.A. ((Mixed Logical Linear Programming)). En: Discrete Applied Mat-

hematics 9697 (1999), pags. 395-442.
[7] Beaumont, N. ((An algorithm for disjunctive programs)). En: European Journal of Operatio-

nal Research 48.3 (1990), pags. 362 -371.
[8] Turkay, M. y Grossmann, I.E. ((Logic-based MINLP algorithms for the optimal synthesis of
process networks)). En: Computers & Chemical Engineering 20.8 (1996), pags.
959 -978.
[9] Duran, M.A. y Grossmann, I.E. ((An outer-approximation algorithm for a class of mixed-
integer nonlinear programs)). En: Mathematical Programming 36 (3 1986), pags.
307-339.
[10] Lee, S. y Grossmann, I.E. ((New algorithms for nonlinear generalized disjunctive program-
ming)). En: Computers & Chemical Engineering 24.910 (2000), pags.
2125-2141.
[11] Grossmann, I.E., Yeomans, H. y Kravanja, Z. ((A rigorous disjunctive optimization model for
simultaneous flowsheet optimization and heat integration)). En: Computers & Chemical
Engineering 22, Supplement 1.0 (1998), S157 -S164.
[12] Caballero, J.A. y Grossmann, I.E. ((Aggregated Models for Integrated Distillation Systems)).

En: Industrial & Engineering Chemistry Research 38.6 (1999), pags. 2330-2344.
[13] Yeomans, H. y Grossmann, I.E. ((Disjunctive Programming Models for the Optimal Design
of Distillation Columns and Separation Sequences)). En: Industrial & Engineering Chemistry

Research 39.6 (2000), pags. 1637-1648.
[14] Rosenthal, R.E. GAMS: A Users Guide. GAMS Development Corporation, 2013.
[15] Harsh, M.G., Saderne, P. y Biegler, L.T. ((A mixed integer flowsheet optimization strategy for
a Cthe debottlenecking problem)). En: Computers & Chemical Enginee-
process retrofits

ring 13.8 (1989), pags. 947 -957.
[16] Diwekar, U.M., Grossmann, I.E. y Rubin, E.S. ((An MINLP process synthesizer for a sequential
modular simulator)). En: Ind. Eng. Chem. Res. 31.1 (1992), pags.
313-322.
[17] Daz, M.S. y Bandoni, J.A. ((A mixed integer optimization strategy for a large scale chemical
plant in operation)). En: Computers & Chemical Engineering 20.5 (1996), pags. 531 -545.
[18]
Caballero, J.A., Milan-Yanez, D. y Grossmann, I.E. ((Rigorous Design of Distillation Columns:
Integration of Disjunctive Programming and Process Simulators)). En: Industrial & Enginee-

ring Chemistry Research 44.17 (2005), pags. 6760-6775.
[19] Brunet, R. y col. ((Minimization of the LCA impact of thermodynamic cycles using a combi-
ned simulation-optimization approach)). En: Appl. Therm. Eng. 48.0 (2012), pags.
367 -377.

295
4.5.REFERENCIAS

[20] Navarro-Amoros, M.A. y col. ((An alternative disjunctive optimization model for heat inte-
gration with variable temperatures)). En: Computers & Chemical Engineering 56.0 (2013),

pags. 12 -26.
[21] Viswanathan, J. y Grossmann, I.E. ((A combined penalty function and outer-approximation
method for {MINLP} optimization)). En: Computers & Chemical Engineering 14.7 (1990),

pags. 769 -782.
[22] Taras, S. y Woinaroschy, A. ((An interactive multi-objective optimization framework for sus-
tainable design of bioprocesses)). En: Computers & Chemical Engineering 43.0 (2012),

pags. 10 -22.
[23] Gendreau, M. y Potvin, J.Y. Handbook of Metaheuristics. International series in operations
research & management science. Springer, 2010.
[24] Gross, B. y Roosen, P. ((Total process optimization in chemical engineering with evolutionary
algorithms)). En: Computers & Chemical Engineering 22, Supplement 1.0 (1998), S229 -S236.
[25] Leboreiro, J. y Acevedo, J. ((Processes synthesis and design of distillation sequences using
modular simulators: a genetic algorithm framework)). En: Computers & Chemical Enginee-

ring 28.8 (2004), pags. 1223 -1236.
[26] Vazquez-Castillo, J.A. y col. ((Design and optimization, using genetic algorithms, of inten-
sified distillation systems for a class of quaternary mixtures)). En: Computers & Chemical
Engineering 33.11 (2009), pags. 1841 -1850.
[27] Guterrez-Antonio, C. y Briones-Ramrez, A. ((Pareto front of ideal Petlyuk sequences using a
multiobjective genetic algorithm with constraints)). En: Computers & Chemical Engineering

33.2 (2009), pags. 454 -464.
[28] Bravo-Bravo, C. y col. ((Extractive Dividing Wall Column: Design and Optimization)). En: In-

dustrial & Engineering Chemistry Research 49.8 (2010), pags. 3672-3688.
[29]
Guterrez-Antonio, C., Briones-Ramrez, A. y Jimenez-Guti
errez, A. ((Optimization of Petlyuk
sequences using a multi objective genetic algorithm with constraints)). En: Computers &
Chemical Engineering 35.2 (2011), pags. 236 -244.
[30] Cortez-Gonzalez, J. y col. ((Optimal design of distillation systems with less than N - 1 columns
for a class of four component mixtures)). En: Chemical Engineering Research and Design

90.10 (2012), pags. 1425 -1447.
[31] Eslick, J.C. y Miller, D.C. ((A multi-objective analysis for the retrofit of a pulverized coal po-
wer plant with a {CO2} capture and compression process)). En: Computers & Chemical
Engineering 35.8 (2011), pags. 1488 -1500.
[32] Odjo, A.O. y col. ((Disjunctive-Genetic Programming Approach to Synthesis of Process Net-
works)). En: Industrial & Engineering Chemistry Research 50.10 (2011), pags.
6213-6228.
[33] MATLAB. The Language of Technical Computing. The Mathworks Inc., 2006.
[34]
Holmstrom,
K., Goran, 268.
A.O. y Edwall, M.M. Users Guide for Tomlab 7. 1999, pag.
[35] Balas, E. y Jeroslow, R. ((Canonical Cuts on the Unit Hypercube)). En: SIAM Journal on Ap-

plied Mathematics 23.1 (1972), pags. 61-69.
[36] Bhutani, N., Ray, A.K. y Rangaiah, G.P. ((Modeling, Simulation, and Multi-objective Optimi-
zation of an Industrial Hydrocracking Unit)). En: Industrial & Engineering Chemistry Research

45.4 (2006), pags. 1354-1372.
[37] Booch, G. y col. Object-Oriented Analysis and Design with Applications. Pearson Educa-
tion, 2007.

296
Captulo 4. Resultados

[38] Biegler, L.T. y Hughes, R.R. ((Infeasible path optimization with sequential modular simula-
tors)). En: AIChE Journal 28.6 (1982), pags.
994-1002.
[39] Bell, D.A., Towler, B.F. y Fan, M. Coal Gasification and Its Applications. Elsevier Science, 2010.
[40] Ghiotti, G. y Boccuzzi, F. ((Chemical and Physical Properties of Copper-Based Catalysts for
CO Shift Reaction and Methanol Synthesis)). En: Catalysis Reviews 29.2-3 (1987), 151 - 182.
[41] Klier, K. ((Methanol synthesis)). En: Adv. Catal 31 (1982), pags.
243-313.
[42] Kung, H.H. ((Methanol Synthesis)). En: Catalysis Reviews 22.2 (1980), pags.
235-259.
[43] Lange, J.P. ((Methanol synthesis: a short review of technology improvements)). En: Catalysis

Today 64.1-2 (2001), pags. 3-8.
[44] Luyben, W.L. ((Design and Control of a Methanol Reactor/Column Process)). En: Industrial

& Engineering Chemistry Research 49.13 (2010), pags. 6150-6163.
[45] Skrzypek, J. y col. Methanol synthesis: science and engineering. Polish Scientific Publishers,
1994.
[46] Turton, R. y col. Analysis, Synthesis and Design of Chemical Processes. Prentice Hall Interna-
tional Series in the Physical and Chemical Engineering Sciences. Pearson Education, 2008.
[47] Smith, R. Chemical process design and integration. Wiley, 2005.
[48] Mayra, O. y Leiviska, K. ((Modelling in methanol synthesis)). En: CHEMICAL ENGINEERING 17
(2009).

Appendix 1. Reformulation of a N term disjunction in N two term disjunctions

Consider the N term disjunction (N = card(D) disjunction)


Yi

hi (x) = 0
(4.5.20)


iD gi (x) 0

Up
xLo
i x xi

it can be rewritten as N disjunctions with two terms each as follows:

297
4.5.REFERENCIAS


Yi

hi (zi ) = 0 Yi

iD



gi (zi ) 0
zi = 0
xLo
i zi xU
i
p
(4.5.21)

Yi
iD
X
x= Zi
iD

Appendix 2. Calculation of reactor volume

The equipment cost is calculated using correlations from the literature, and to this
end we use the correlations given by Turton et al. (Turton et al., 2008). The capital
cost of the reactor is calculated based on his volume. This parameter is calculated as
follow. For a continuous stirred flow rector (CSTR) a molar flow balance for component
j gives the volume of the reactorV . Thus, for the hydrogen component, the resulting
expression to calculate the volume of the reactor is:

in
FH2 FH2
V = (4.5.22)
rH2

where rH2 is the rate of disappearance of H2 and FH2 is the molar flow rate of H2
at the output stream, which is calculated according to the following expression:

X Fkin Xik
Fj = Fjin + ij , j, k KCi (4.5.23)
i
ik

where ij are the stoichiometric coefficient for component j in reaction i , and


Xik assesses the extent of reaction i with respect a key component k for that reaction
(KCi is the subset of components that contains the key component k for the reaction

298
Captulo 4. Resultados

i). Specifically, the conversions of the reaction R-1 with respect carbon monoxide and
reaction R-2 with respect to carbon dioxide are:

in F
FCO CO
X1,CO = in
(4.5.24)
FCO
in
FCO FCO2
2
X2,CO2 = in
(4.5.25)
FCO 2

The reaction rate with respect H2 term in Eq. 4.5.22 is calculated from reaction rates
with respect to methanol for both reactions (rCH3 OH,i ) according to the following
expression:

X rCH3 OH,i
rH2 = acat i,H2 (4.5.26)
i
CH3 OH,i

where acat is the mass of the catalyst per unit reactor volume, 1.154 ton catm3
[48].

The rate for the methanol synthesis from carbon monoxide reaction with respect
to methanol (rCH3 OH,i ) is obtained with the following kinetic model [39], which uses
partial pressures of the components (Pj ):

" #
3/2 PCH3 OH
k1 KCO PCO PH2 eq 1/2
K1 PH
2
rCH3 OH,1 = 1/2
(4.5.27)
1 + KCO PCO + KCO2 PCO2 PH2 + k2 PH2 O

where Pj = Fj /F P for all j, and K1eq is the equilibrium constants for the synthesis
of methanol from carbon monoxide R-1 reaction. In Eq. 4.5.27 the kinetic constants k1 ,
k2 , KCO and KCO2 are calculated as follows:

299
4.5.REFERENCIAS

109900
k1 = 2,69x107 exp (4.5.28)
RT
11 104500
k2 = 4,13x10 exp (4.5.29)
RT
7 58100
KCO = 7,99x10 exp (4.5.30)
RT
67400
KCO2 = 1,02x107 exp (4.5.31)
RT

where R is the gas law constant in Jmol1 K1 and T temperature in K.

There is an analogous expression for the rate of reaction R-2 with respect to metha-
nol:

" #
3/2 PCH3 OH PH2 O
k3 KCO2 PCO PH2 eq 3/2
K2 PH
2
rCH3 OH,2 = 1/2
(4.5.32)
1 + KCO PCO + KCO2 PCO2 PH2 + k2 PH2 O

where K2eq is the equilibrium constant for the synthesis of methanol from carbon
dioxide R-2 reaction, and k3 is the kinetic constant evaluated with the expression:

65200
k3 = 4,36x102 exp (4.5.33)
RT

Knowing the volume of the reactor, we calculate the capital cost of the reactor is
calculated using the correlations provides by Turton et al. [46].

300
CAPITULO 5

Conclusiones

301
Captulo 5. Conclusiones

Captulo 5. Conclusiones
Considerando los resultados expuestos a lo largo de la memoria de la presente
se destacan las siguientes conclusiones:
investigacion,

Dentro del primer objetivo general de la tesis se desarrollan dos herramientas


y la eficiencia energetica
que facilitan y mejoran sustancialmente la simulacion
de las columnas TCD.

1 muestra
La estrategia desarrollada y presentada dentro de la publicacion
una gran utilidad a la hora de simular las secuencias TCD. En ella se obser-

va como es posible simular una secuencia de columnas TCD de forma facil y
sencilla de forma totalmente acclica. Ello se consigue mediante una sencilla
de las corrientes implicadas en los acoplamientos termicos
sustitucion por dos
corrientes, una de materia y otra de energa. Los resultados obtenidos (configu-
acclica) son muy proximos
racion riguro-
a los obtenidos al realizar la simulacion
cclica) del sistema, presentando errores medios por debajo
sa (configuracion
del 2 % para mezclas de 3 componentes. El error medio se incrementa ligera-

mente al incrementar el numero de componentes, debido principalmente a la
del error como consecuencia del aumento de los acoplamien-
propagacion

tos termicos en el sistema. Sin embargo, en todos los casos, esta estrategia
proporciona excelentes resultados comparados con los obtenidos con la simu-
rigurosa. Finalmente, se obtienen excelentes puntos de partida para la
lacion
rigurosa de este tipo de sistemas. En conclusion,
simulacion esta nueva tecnica

de mezclas de 3, 4 o 5 componentes median-
permite estudiar la separacion
acopladas termicamente
te secuencias de columnas de destilacion de forma

rapida
y sencilla comparados con los metodos
actuales de simulacion.

2 se presenta una configuracion


Por otro lado, en la publicacion novedosa que
aprovecha la ineficiencia inherente a las columnas TCD -exceso de flujo de
vapor/lquido en ciertas secciones como resultado de la transferencia lleva-

da a cabo en los acoplamientos termicos-
transformandola en una ventaja,

consiguiendo disminuir gracias a esta la energa y costes del sistema. El pro-
del exceso de vapor/lquido
cedimiento presentado se basa en la extraccion

303
5.0.Conclusiones

que se trasfiere entre ciertas secciones de las columnas, con el fin de introdu-
de vapor, si la corriente extrada es gaseosa o en
cirlos en ciclo de compresion
de vapor inverso si la corriente extrada es lquida. La
un ciclo de compresion
ventaja conseguida con este procedimiento es doble. En primer lugar, se recu-

peran las condiciones optimas de todas las secciones de las columnas TCD, lo

que hace que los diametros y los consumos de servicios externos se reduzcan.
de un ciclo VRC o RVRC permite recuperar
Y en segundo lugar, la introduccion
el consumo de
mas
energa del sistema, con lo que se consiguen reducir aun
servicios externos. En cualquier caso, y debido al alto precio de alguno de los
del VRC o RVRC, es necesario un estu-
equipos necesarios para la instalacion

dio economico de es-
detallado para confirmar la idoneidad de la instalacion

ta tecnologa. Como se ha presentado anteriormente, los ahorros economicos
de esta propuesta pueden llegar a ser signifi-
consecuencia de la aplicacion
cativos, entorno a un 20-40 %.

no es rentable en todos los sistemas


Desgraciadamente, esta configuracion
E sta es potencialmente util
de separacion.
en los casos donde la distribucion
de volatilidades de la mezcla a separar muestra diferencias importantes en-
tre un grupo de compuestos (con volatilidades similares entre ellos) respecto
generara que los flujos internos tanto lquido como vapor
a otro. Esta situacion
sean muy diferentes en las diversas secciones de la columna tipo Petlyuk, consi-
de un ciclo de compresion
guiendo las condiciones ideales para la instalacion
de vapor.

de un ciclo RVRC es uni-


Finalmente, y como regla general, la implementacion
camente interesante en el caso de que el condensador trabaje a tempera-
turas muy por debajo de la ambiente, donde el agua no pueda ser utiliza-
do como servicio fro. El agua es un servicio normalmente barato, por lo que
de esta genera ahorros economicos
la reduccion relativamente bajos que no
de los equipos necesarios para la instala-
compensan el coste de adquisicion
del RVRC. Sin embargo, cuando se utilizan refrigerantes por debajo de la
cion

temperatura ambiente, normalmente caros, el ahorro energetico s que com-
de los equipos, siendo recomendable la instalacion
pensa la adquisicion de
este tipo de ciclo.

304
Captulo 5. Conclusiones

Dentro del segundo objetivo general de la tesis, se ha desarrollado una me-


todologa alternativa a la propuesta por Duran y Grossmann [1], y posterior-
simultanea
mente reformulada por Grossmann et al. [2], para la optimizacion
de calor de sistemas de procesos no isotermos. Los resultados
y la integracion
obtenidos muestran que el modelo desarrollado en este trabajo se compor-
al modelo de Grossmann et al.[2].
ta de forma competitiva en comparacion
Una caracterstica importante del modelo presentado es que los problemas

generados son mucho mayores (mayor numero de variables y ecuaciones). Es-
te hecho a priori podra ser considerado como una desventaja frente a otros
de una etapa de preprocesado y la introduccion
modelos, pero la aplicacion

de un conjunto de relaciones logicas consigue reducir considerablemente la
factible del problema inicial, consiguiendo resultados competitivos tan-
region

to en el tiempo de calculo, del gap. Ademas
como en el grado de relajacion

y como principal diferencia respecto a otros modelos es que este mantiene el
concepto de intervalo de temperatura. Esta caracterstica es interesante para

aquellos ingenieros e investigadores familiarizados con el concepto clasico de
de calor. Por ultimo,
intervalo de temperatura en la integracion el modelo pro-

puesto elimina las dificultades numericas
asociados con el uso de la funcion
max.


El ultimo objetivo general de la tesis ha permitido desarrollar dos metodologas
rigurosa de problemas de sntesis de procesos,
alternativas para la optimizacion

ambos trabajos se encuentran en fase de revision.

El primero de los trabajos, presentado en el anexo 1, presenta una estructura de


modelado para tratar problemas de sntesis, que incluyen diferentes modelos

matematicos para calcular o simular diferentes operaciones unitarias. Debido
a que hoy en da, existen numerosos y distintos modelos para representar un
proceso qumico, el modelo desarrollado se presenta como una herramienta
adecuada cuando diferentes modelos (basados en ecuaciones, tipo caja ne-

gra, ettc. . . ) se emplean de manera simultanea de un
durante la resolucion
de un sistema. La estrategia utilizada en este tra-
problema de optimizacion
disyuntiva, ya que esta estructura es ade-
bajo se basa en la programacion
cuada para hacer frente a este tipo de problemas, permitiendo encapsular

305
5.0.Conclusiones

de sus caractersticas.
cada modelo y utilizar un enfoque diferente en funcion
La estructura llevada a cabo utiliza disyunciones para capturar cada uno de
los modelos de tal forma que se aprovechan las ventajas o minimizan los in-
convenientes de cada modelo. De esta manera es posible utilizar las derivadas
disponibles y simultaneamente
algebraicas cuando estan
implementar meto-
dos robustos para estimar las derivadas de otros modelos sin la necesidad de

interferir con el resto de modelos, y de esta forma mejorar el tiempo de calculo
computacional. Algunas de las capacidades del modelo propuestos son:


El analisis de la estructura de entrada-salida de cada modelo para la
de las variables de dise
seleccion no y reducir tanto como sea posible las
recirculaciones en la estructura de problema

del patron
Determinacion de dispersion
en cada modelo para acelerar

el calculo de la matriz jacobiana

de diferentes metodos
La aplicacion
para el calculo de
de la informacion

las derivadas, como pueden ser analticos, diferencias finitas o derivacion

numerica utilizando variables complejas.


Posibilidad de trabajar con sistemas con ruido numerico de me-
a traves
tamodelos adaptativos (kriging)

disyuntiva facilita el modelado conceptual del proceso


La programacion


Gran facilidad de introducir relaciones logicas en el modelo

El resultado final es un sistema hbrido que incluye modelos implcitos proceden-



tes de diferentes fuentes y con diferentes comportamientos numericos.

Finalmente, el anexo 2 muestra una nueva estrategia de modelado para tratar


los problemas de sntesis. La metodologa propuesta es adecuada para la reso-
de problemas donde la topologa del mismo no se encuentra definida,
lucion
existiendo diferentes alternativas para cada una de las operaciones unitarias
del sistema. La estrategia presentada genera mediante una superestructura
del sistema todas las alternativas de los diferentes procesos, y posteriormente
sinergica
resuelve el problema mediante una combinacion de simuladores jun-
GDP. Conceptualmente la programacion
to a la programacion GDP facilita la

306
Captulo 5. Conclusiones

del modelo para el usuario conservando la estructura logica


formulacion sub-
yacente del problema. El modelo propuesto combina la flexibilidad de la for-
GDP con los beneficios de los simuladores de proceso comerciales,
mulacion
donde un avance importante radica en el hecho de que la superestructura
integrada directamente en la interfaz grafica
del proceso esta del simulador.
esta estrategia presenta una serie de ventajas frente a los modelos
Ademas,
basados completamente en ecuaciones. Por un lado el uso de los simuladores

de procesos facilita la tarea de simular el comportamiento termodinamico de
las mezclas y compuestos constituyentes del sistema en estudio. Y por otro lado,
matematica
el uso de la programacion consigue reducir el tama
no del proble-
ello es debido a que unicamente
ma NLP a resolver en cada iteracion, las
ecuaciones pertenecientes a las alternativas activas son consideradas para la
de los subproblemas NLP. Esta caracterstica permite obtener subpro-
resolucion
sencillos, consiguiendo facilitar su resolucion
blemas mas reduciendo el tiempo

de calculo
necesario. Finalmente comentar que la metodologa sistematica
propuesta proporciona una forma intuitiva para los ingenieros qumicos para
sintetizar procesos qumicos de forma rigurosa.

307
CAPITULO 6

Contribuciones Cientficas

309
Captulo 6. Contribuciones Cientficas

Captulo 6. Contribuciones Cientficas

6.1 Publicaciones

Navarro, M.A.; Ruiz-Femenia, R.; Caballero, J.A. Integration of modular process


simulators under the Generalized Disjunctive Programming framework for the
structural flowsheet optimization. Sent to Computers & Chemical Engineering
Jul-2013. Status: Under review

Caballero, J.A.; Navarro, M.A.; Ruiz-Femenia, R.; Grossmann, I.E. Integration of


different models in the design of chemical processes: Application to the design
of a power plant. Sent to Applied Energy Jul-2013. Status: Under review

Navarro, M.A.; Ruiz-Femenia, R.; Caballero, J.A. A new technique for recovering
energy in thermally coupled distillation using vapor recompression cycles. AIChE
Journal (2013). Article published online: Jun-2013

Navarro, M.A.; Caballero, J.A.; Grossmann, I.E. An alternative disjunctive opti-


mization model for heat integration with variable temperatures. Computers &
Chemical Engineering, 56 (2013) pp. 12-26

Navarro, M.A.; Javaloyes, J; Caballero, J.A.; Grossmann, I.E. Strategies for the ro-
bust simulation of thermally coupled distillation sequences. Computers & Che-
mical Engineering, 36 (2012) pp. 149-159

6.2 Libros

Caballero, J.A.; Navarro, M.A.; Grossmann, I.E. Hybrid Simulation-Optimization


Logic Based Algorithms for the Rigorous Design of Chemical Process. 22nd Eu-
ropean symposium on Computer Aided Process Engineering, 30 (A) (2012) pp.
582-586. Lockhart I.D.; Fairweather M. Elsevier Science. ISBN:978-0-444-59431-0

Navarro, M.A.; Caballero, J.A.; Grossmann, I.E. Heat Integration in non-Isothermal


Systems Using an Alternative Disjunctive Optimization Model. 22nd European

311
6.4.6.3. ASISTENCIA A CONGRESOS

symposium on Computer Aided Process Engineering, 30 (A) (2012) pp. 1282-


1286. Lockhart I.D.; Fairweather M. Elsevier Science. ISBN: 978-0-444-59431-0

Navarro, M.A.; Caballero, J.A.; Grossmann, I.E. Strategies for the robust simula-
tion of thermally coupled distillation sequences. 21th European symposium on
Computer Aided Process Engineering, 29 (A) (2011) pp. 196-200. Pistikopoulos,
E.N.; Georgiadis, M.C.; Kokosis, A.C. Elsevier Science. ISBN: 978-0-444-53895-6

6.3 Asistencia a Congresos

American Institute of Chemical Engineers (AICHE). 2012 AIChE Annual Mee-


ting. Pittsburgh (EEUU) 28/10/2012 - 2/11/2012

na (ANQUE). International Congress


Nacional de Qumicos de Espa
Asociacion
of Chemical Engineering 2012 (ANQUE -ICCE 2012). Sevilla 24-27/06/2012

Institution of Chemical Engineers (IChemE). .European Symposium on Computer


Aided Process Engineering (ESCAPE 22). Londres 17-20/06/2012

Sociedad Espa
nola de Qumica Industrial e Ingenieria Qumica 12th Medite-
rranean Congress of Chemical Engineering. Barcelona. 15-18/11/2011

Institution of Chemical Engineers (IChemE). .European Symposium on Compu-


ter Aided Process Engineering (ESCAPE 21). Chalkidiki (Grecia) 29/05/2011 -
1/06/2011

6.4 Contribuciones a Congresos

Caballero, J.A.; Navarro, M.A.; Reyes-Labarta, J.A. New Vapor Compression Al-
ternatives in Thermally Coupled Distillation. Tipo: Oral. 2012 AIChE Annual Mee-
ting. Pittsburgh (Estados Unidos) 28/10/2012 - 2/11/2012

312
Captulo 6. Contribuciones Cientficas

Reyes-Labarta, J.A.; Navarro, M.A.; Caballero, J.A. A Hybrid Simulation Opti-


mization Approach for the Design of Internally Heat Integrated Distillation Co-
lumns. Tipo: Oral. 2012 AIChE Annual Meeting. Pittsburgh (Estados Unidos)
28/10/2012 - 2/11/2012

Navarro, M.A.; Caballero, J.A. Generalized Disjunctive Programming (GDP) for


structural flowsheet optimization using commercial simulators: synthesis of met-
hanol. Tipo: Oral. International Congress of Chemical Engineering 2012 (ANQUE
-ICCE 2012). Sevilla (Espa
na) 24-27/06/2012

Navarro, M.A.; Caballero, J.A.; Grossmann, I.E. Heat Integration in non-Isothermal



Systems Using an Alternative Disjunctive Optimization Model. Tipo: Poster. 22nd
European symposium on Computer Aided Process Engineering (ESCAPE 22).
Londres (Reino Unido) 17-20/06/2012

Caballero, J.A.; Navarro, M.A.; Grossmann, I.E. Hybrid Simulation-Optimization



Logic Based Algorithms for the Rigorous Design of Chemical Process. Tipo: Poster.
22nd European symposium on Computer Aided Process Engineering (ESCAPE
22). Londres (Reino Unido) 17-20/06/2012

Javaloyes, J.; Ruiz-Femenia, R.; Navarro, M.A.; Caballero, J.A. Rigorous design of
complex distillation columns using process simulators and stochastic optimiza-
tion algorithms. Tipo: Oral. 12th Mediterranean Congress of Chemical Enginee-
ring. Barcelona (Espa
na) 15-18/11/2011

Navarro, M.A.; Javaloyes, J.; Caballero, J.A. An alternative disjunctive optimi-



zation model for heat integration in non-isothermal systems.Tipo: Poster. 12th
Mediterranean Congress of Chemical Engineering. Barcelona (Espa
na)
15-18/11/2011

Navarro, M.A.; Caballero, J.A.; Grossmann, I.E. Strategies for the Robust Simu-

lation of Thermally Coupled Distillation Sequences. Tipo: Oral y poster. 21th
European symposium on Computer Aided Process Engineering (ESCAPE 21).
Chalkidiki (Grecia) 29/05/2011 - 1/06/2011

313

Related Interests