The ChemSep Book

Second Edition

Harry A. Kooijman
Amsterdam, The Netherlands
[email protected]

Ross Taylor
Clarkson University, Potsdam, New York
[email protected]
Copyright
c 2007 by H.A. Kooijman and R. Taylor. All rights reserved.

This publication is public domain. No responsibility is assumed by the authors for any injury
and/or damage to persons or property as a matter of products liability, negligence, or otherwise,
or from any use or operation of any methods, products, instructions, information, or idea’s
contained in the material herein. The use of general descriptive names, registered names, trade-
marks, etc. in this publication does not imply, even in absence of specific statement, that such
names are exempt from the relevant protective laws and regulation and therefore free for general
use.
Contents

Preface 7

Part A. Using ChemSep 1

1 A Simple Equilibrium Stage Column 3

1.1 A Short Tour of the ChemSep Interface . . . . . . . . . . . . . . . . . . . . . 3
1.2 Distillation of Benzene and Toluene . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.4 Real stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.5 Parametric Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
1.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2 Nonequilibrium Columns 45
2.1 Equilibrium and Nonequilibrium Models . . . . . . . . . . . . . . . . . . . . 45
2.2 A Sieve Tray Column . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.3 A Depropanizer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.4 A Packed Absorber . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.5 Extractive Distillation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.7 Model Uncertaincies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

3 CAPE-OPEN Flowsheet Simulations with ChemSep 79

3.1 Using the COCO Simulator . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.2 Flowsheet Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.3 Di-Ethyl Ether from Ethanol-Water Feed . . . . . . . . . . . . . . . . . . . . 103

4 Analysis of Separation Processes 109

4.1 Plot Data File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.2 Parametric Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.3 Column Rating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.4 Stream Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.5 Phase Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.6 Residue Curve Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.7 Property Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

3
 4.8 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

5 Databanks 131
5.1 Pure Component Data (PCD) . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.2 Pure Component Libraries (LIB) . . . . . . . . . . . . . . . . . . . . . . . . 131
5.3 Group Contribution Data (GCD) . . . . . . . . . . . . . . . . . . . . . . . . 132
5.4 Interaction Parameter Data (IPD) . . . . . . . . . . . . . . . . . . . . . . . . 133
5.5 Column Internals Parameter Libraries (INP) . . . . . . . . . . . . . . . . . . 135

6 Tools and Settings 137

6.1 Commandline Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
6.2 Interface Settings and Files . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.3 Tools and Tool Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.4 Design Mode Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
6.5 ChemSep Internals and Model Development . . . . . . . . . . . . . . . . . . 146

7 Reactive Distillation 171

8 Extraction 173

9 Three Phase Distillation 175

Part B. Technical Reference 177

10 Property Models 179

10.1 Thermodynamic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
10.2 Physical Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193

11 Numerical Solution of Systems of Equations 221

11.1 Repeated Substitution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
11.2 Newton’s Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
11.3 Homotopy-Continuation Methods . . . . . . . . . . . . . . . . . . . . . . . 224

12 Flash Calculations 237

12.1 Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
12.2 Solution of the Flash Equations . . . . . . . . . . . . . . . . . . . . . . . . . 239
12.3 The Rachford-Rice Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 240

13 Equilibrium Stage Simulation 243

13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
13.2 The Equilibrium Stage Model . . . . . . . . . . . . . . . . . . . . . . . . . . 245
13.3 Tearing Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
13.4 Simultaneous Correction Methods . . . . . . . . . . . . . . . . . . . . . . . 262
13.5 Continuation Methods for Difficult Problems . . . . . . . . . . . . . . . . . . 272
13.6 Relaxation Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
13.7 Three Phase Distillation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
13.8 Reactive Distillation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
14 Nonequilibrium Column Modelling 299
14.1 The Model Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
14.2 Model Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
14.3 Flow Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
14.4 The Design Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308

15 Design Models for Trayed Columns 313

15.1 Mass Transfer Coefficient Correlations . . . . . . . . . . . . . . . . . . . . . 313
15.2 Pressure Drop Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
15.3 Entrainment and Weeping . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
15.4 Packing Flooding and Minimum Wetting . . . . . . . . . . . . . . . . . . . . 319
15.5 Column Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
15.6 Comparison with Experimental Distillation Data . . . . . . . . . . . . . . . . 323

16 Design Models for Packed Columns 339

16.1 Mass Transfer Coefficient Correlations . . . . . . . . . . . . . . . . . . . . . 339
16.2 Pressure Drop Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
16.3 Entrainment and Weeping . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
16.4 Packing Flooding and Minimum Wetting . . . . . . . . . . . . . . . . . . . . 348
16.5 Column Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349

17 Design Models for Extraction Columns 357

17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
17.2 Sieve Tray Columns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
17.3 Packed Columns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
17.4 Rotating Disk Contactors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
17.5 Spray Columns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
17.6 Modeling Backflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370

18 Nonequilibrium Distillation Dynamics 375

References 377

Index 377
Preface

This book is a companion to ChemSep, a set of programs that perform multicomponent separa-
tion processes simulations. It is not designed to be a textbook for teaching separation processes.
Instead, the reader will find a guided tour on how to use ChemSep, many examples solved with
the help of ChemSep, and a complete technical reference with detailed descriptions of all the
models and algorithms used in the programs. We hope this book provides the ChemSep user
with all the information needed to make good use of the program.

The ChemSep project was started in February 1988 at the University of Technology Delft in
the Netherlands, by Arno Haket, Harry Kooijman, and Ross Taylor. ChemSep was designed to
be easy to use by students with no experience of engineering software, while having sufficient
flexibility and power to appeal to expert users. In pursuit of these objectives ChemSep features
a menu-driven, user-friendly interface with an integrated help system and an autopilot mode
that leads the novice user through the data input phase. In March of 1991, when the nonequi-
librium model was added, the programs and user interface were completely rewritten. Over the
years various extensions of the nonequilibrium model were published and added: an extrac-
tor model, various dynamic models, a (reactive) cell model, a three phase distillation model,
and a total reflux model (for simulating distillation test data). In 2004 a start was made with
a new windows GUI that is still being further developed to include all the column models. In
2005 ChemSep became Cape-Open compliant. As a result it is now possible to use the program
inside flowsheeting tools such as Aspen Plus (Aspentech), PRO/II (SimSci/Escor) and COCO
(Amsterchem). In 2006 we made a LITE version of the Cape-Open version of ChemSep avail-
able so that anybody can test-drive the program. The LITE version is included in the COCO
flowsheeting package (http://www.cocosimulator.org).

We continue to develop the simulators to enable more accurate simulation of industrial chemical
and petroleum column operations. We do this by developing and adding published thermody-
namic and physical property models as well as by developing and implementing of published
mass transfer models. Please visit the ChemSep website at http://www.chemsep.org for
more information. Since ChemSep is a work in progress, so must be this book. We welcome
your suggestions on how the material presented here can be improved.

Harry Kooijman and Ross Taylor

7
Part A
Using ChemSep
2
Chapter 1

A Simple Equilibrium Stage

Column

In this chapter we take you on a quick tour of ChemSep by showing you how to input the
specifications for the simulation of a simple distillation column for a simple binary separation.
We analyse the results of this example by looking at the various plots ChemSep offers, including
McCabe-Thiele diagrams.

1.1 A Short Tour of the ChemSep Interface

When you start ChemSep you first obtain an introduction screen that disappears after a couple
of seconds. ChemSep starts automatically with a ‘new’ problem where you are asked to enter a
descriptive title for the simulation. The user interface is divided a main menu, a tool bar, a tree
which lists various input sections on the left and the first input section (with a tabbed screen)
for the title and optional comments on the right, previous/next buttons, and a status line.

Note that in the tree list there are icons next to the items. In the case of the ‘title’ there is a gray
checkmark in front meaning that this item is optional. Other items have a red cross meaning
that they still require further specification. The interface shows you the next item that need to
be specified by labeling or coloring it red (the actual color can be customized). Here, as soon
as we type something in for the title input box, the checkmarks in front of ‘title’ (both in the
tree list as well as on the tab) change from gray to green, which means that the specification has
been completed for this tabscreen.

In ChemSep we follow the phylosophy that we only show you what you need to see. Tabsheets

3
 Figure 1.1: ChemSep opens with a new problem.

with specification have either a red cross (action required) or a green checkmark (no further
action required). Tabsheets without icons contain results or optional inputs. Note that the cross
next to the results only indicates whether or not any results are present. You can navigate
through the sheets by means of the next(F6) and previous (F5) buttons on the bottom, or by
clicking on the tabs, or by clicking on the items in the tree listing on the left of the interface.

When you inspect the main menu you notice there is no ‘window’ item in the main menu. This
is because for each problem you want to solve, we instantiate a separate window. This has the
advantage that you can switch problems with the Alt-Tab windows key, similarly as other MS
Office products behave. Each of the windows behaves completely autonomically, that is, they
do not share memory space and define each their own problem. Do note that when you save
the interface settings, you do this globally. However, as the problems already running are not
affected, you would need to close all other windows before you’d see the changes in the other
windows. The definitions of the units of measure, path information, and the solve options are
stored in the global options settings as well as locally in the sep-file for the problem.

4
1.1.1 Keys

Just like in any other windows program, ChemSep assigns several default function key behav-
iors: F1 calls up (context sensitive) help and F10 selects the main menu. You can navigate
in menu’s with the arrow keys. Use Enter to select and Escape to go back up one level. F5
and F6 have been programmed as shortcuts for the selection of the previous and next tabsheets
to quickly navigate through the problem specification. ChemSep also allows you to define the
functionality of the other function keys, see the Tools menu.

1.1.2 File Menu

The ChemSep file menu has the usual entries such as New, Open, Save, Save as, and Close.
These behave just like most programs define these functions, execpt for that ‘new’ and ‘open’
both open a new window to start the new problem. The ‘close’ option returns the current
window to the state of a ‘new’ problem (and will ask you to save the problem when any changes
would get lost in the process). This is consistent with the fact that ChemSep uses a different
window for each problem it opens. To enable people to open a file in the current window, the
filemenu of ChemSep contains the options ‘Load’ and ‘Reload’. The latter reloads the currently
opened file. This is particularly handy when you want to shed any changes that you made and
want to go back to the problem as it was saved to disk.

Furthermore, the file menu contains the usual print, print setup, and exit menu items. Extra,
however, is the option to export results to reports in specific formats (for example HTML). Also
extra (in comparison to regular windows based programs) are options to view or edit a file on
disk, and to open a window with a operating system (OS) shell for typing commands on the
commandline.

1.1.3 Edit Menu

The edit menu has the usual cut/copy/paste/delete options, but also contains options to switch
to the previous and next input screen, just like the previous and next buttons below the input
screen. When a solved problem has been loaded, the edit menu also contains an option to
generate a new problem using one of the output streams in the simulation. This option allows
the quick evaluation of a separation distillation train.

1.1.4 Solve Menu

The solve menu includes options to check the input on completeness and to solve the problem.
ChemSep includes a quick solve that will bypass any checks for experts only. The method used

5
to solve and to inialize the problem are set in the solve options which you can set by clicking
on ‘solve option’ in the tree list.

Figure 1.2: Solve options input screen.

1.1.5 Analysis Menu

The analysis menu contains various options to investigate the behavior of the mixture as defined
in the input. You can view phase diagrams, residue curve maps, and diffusion coefficients for
binary and ternary systems. As soon as the problem has been solved the parametric study option
is available which allows you to study the effect of each of the input variables. One or more
parameters can be varied over a specified range. This is useful for optimisation studies. When
a column was solved you can rate it to determine the diameter using a general tray capacity
function or transfer column information to vendor software for detailed hardware design.

1.1.6 Tools Menu

The tools menu allows you to set the interface settings and save/load these. It allows you to add
your own tools to ChemSep. For example, our units conversion tool is a separate program that
is added to the user interface. In the interface settings you define the formating of the input and
results and the interactive behavior of the interface. Note that these global settings are saved
together with the units of measure, path, and solve options of the problem you have loaded.
These will then be used as defaults when you start a new problem!

6
1.1.7 Input

Whereever you need to enter any numeric input, ChemSep shows a ‘*’ to indicate a value is
not specified. Where possible, ChemSep will supply a default during the simulation so that you
do not need to make the specification. However, some values need to be entered and this is
indicated by the automatic checking of the input; the values that are required will be indicated
in red. When you select the field and press Enter ChemSep will provide you with a default
value, if it can. Or it might provide you with a selection list of models from which you have to
chose.

ChemSep can process algebraic calculations wherever numerical input is required. This is useful
since, if you don’t know the actual numerical value that should be entered but you know how
to calculate it, you may enter the calculation. Numerical formulae may include the four basic
arithmetic operations, +, -, *, and /. Operations may be nested within parentheses () as well.
When you have typed in the formula, press Enter to evaluate the result and Enter a second time
to accept that result. ChemSep does not remember formula entries, only the final result so you
may edit the formula until you press Enter. Here are some examples of numerical formula
entry:

3 Enter
5-2 Enter
(2-1) * (5-2) Enter

All of these result in the number 3. Formula entry can be useful in the feed spreadsheet where
you are asked to enter the component flows. Perhaps you know the total flow rate and the
mole fractions rather than the component flows. Instead of using your calculator to compute
the component flows, you can let ChemSep do the calculations for you. By way of an example
consider a column with a feed flow of 573 mol/s containing 36.5 mole percent ethanol, the rest
being water. The component feed flows of ethanol and water could be typed in as:

573 * 0.365 Enter

573 * (1 - 0.365) Enter

1.1.8 Units of Measure

ChemSep data entry fields also accept units. This feature is particularly useful if you know a
quantity in some units other than the current set of units. You don’t have to change the global
units settings for this, simply type in the numerical value of the quantity and follow it with the
units. The number will then be displayed in the units defined globally for the problem. For
example, what if the default flow units are kmol/s but we know the feed flows in lbmol/h?

7
Simply type in the feed flow as, for example, 375 lbmol/h followed by Enter and ChemSep will
automatically convert the number to the correct value in the default set of units. You can use this
feature in any data entry field in ChemSep. Spaces are ignored when evaluating the expression
with units. Any combination of the units listed under Other Units may be used.

ChemSep checks the dimensions of the units you enter and displays a warning message if they
do not have the correct dimensions. If the dimensions are correct a message with the conversion
is displayed (this behavior can be switched off in the interface settings). A numerical formula
and a unit string can be entered in the same field at the same time. All results of formula and
unit entry are displayed in the default set of units. Note that all internal calculations are carried
out in SI units and that unit entries are case sensitive when using their abbreviations.

Units that ChemSep recognises are shown in Table 1.1. ChemSep also recognizes the standard
prefixes for multiples of 10. For example, mmol/s is recognized as mol/1000 / s. The complete
list of prefixes and their abbeviations are listed in Table 1.2.

All units and their prefixes can be typed in data entry fields as well as selected to be part of the
default unit set, see Figure 1.3.

Figure 1.3: Units of measure specification for mass flows.

To change the units select the ‘units’ item from the list tree. Now click on any of the units and
you see a pull-down list of the available units. This is just a sample list, you can enter any units
by entering this from hand in the selection field. After selecting a unit or after pressing Enter
while editing a unit definition a dimension analysis is performed to check its correctness.

8
 Table 1.1: Units of measure

Abbreviation: Unit: Abbreviation: Unit:

kg kilogram lbf pound force
m meter kgf kilogram force
s seconds atm atmosphere
K Kelvin psia lbf/square inch
kmol kilomole psig lbf/sq in gauge
rad radian yd yard
N Newton ton ton
Pa Pascal USton US ton
J Joule oz ounce
W Watt lbmol poundmole
o
degree mol mole
lb pound erg erg
g gram dyn dyne
min minute P Poise
h hour mi mile
o
day day F Fahrenheit
o
UKgal UK gallon C Celcius
USgal US gallon F Fahrenheit
l liter C Celcius
cal calorie R Rankine
Btu British Thermal Unit bbl barrel
in inch - dimensionless
” inch % percent
ft feet %% per thousand
torr torr ppm parts/million
bar bar ppb parts/billion
barg bar in gauge

1.2 Distillation of Benzene and Toluene

Now let us start with a simple binary distillation column. As title enter ”Distillation of Benzene
and Toluene”. You’ll notice the title input window is now fully specified. So use the next
button to move to the component entry. In this window you notice the default library (here
chemsep1.pcd) is selected and that on the left list the components in this PCD library are listed
by name. Now type the names of the components you want to select in the ‘Find’ inputbox.
When you enter ”benzene” you will notice there are many other components listed. This is
because the interface uses a synonyms file to search and that all these components contain an

9
 Table 1.2: Prefixes that can be used on any unit

Short: Full: Factor: Short: Full: Factor:

T Tera 1012 d deci 10−1
G Giga 109 c centi 10−2
M Mega 106 m milli 10−3
k kilo 103 µ micro 10−6
h hecto 102 n nano 10−9
da deka 10 p pico 10−12
f femto 10−15

benzene ring. You can use the down and up arrow keys to select the actual component you want,
in this case Benzene, and then select it by pressing Enter. This will also clear the find string so
we can go ahead with selecting the next component, Toluene. Note that as soon as you defined
the first component, the interface marked the component selection as complete (and assigned
the green checkmark to the tab). Of course, for a flash problem one component selected suffices
to do calculation, but for a distillation problem we need to select at least two components. Select
Toluene as 2nd component.

Figure 1.4: Selection of the component Benzene.

Alternatively you can select a component with the right arrow ‘Add’ button. Once the com-
ponents are selected you can reorder them with the Up and Down arrow buttons. Delete com-
ponents by using the left arrow ‘Delete’ button. You can select other component libraries by

10
clicking the browse button. ChemSep names its binary libraries *.PCD and its text libraries
*.PCT. To modify or inspect the component libraries use the pure component data manager,
in the main menu under ‘Databanks’. Note that the components are identified by their names.
We would like instead to use the letters ”B” and ”T” as identifiers. That can easily be done
by editing the identifier fields for the selected components in the list on the right. We are then
finished with the components and press F6.

Figure 1.5: Selection of the type of simulation.

The next step is to select the type of simulation: in this case we are selecting an ‘Equilibrium
column’. As soon as we make this selection we need to specify the configuration of our dis-
tillation column. Here the interface supplies us with all kinds of defaults and we keep picking
these: Simple distillation; Total condenser; Partial reboiler; 10 stages; feed on stage 5. Note
that for a simple distillation column there are no sidestreams or pumparounds allowed. If you
would want to enter those select ‘Complex column’ as kind of operation. As soon as we entered
the feed stage we have fully specified the column configuration and a schematic representation
of the column is shown on the right, see Figure 1.6.

Figure 1.6: Column configuration for the BT column.

Having finished with the specification of the operation we continue by expanding the ‘Proper-
ties’ in the tree list on the left by clicking the little plus. We see that there are three sub entries:
Thermodynamics, Physical properties and Reactions. The latter is already completed as no
reactions are assumed per default.

We have to specify the thermodynamic models used to simulate the vapor-liquid equilibria on

11
 Figure 1.7: Thermodynamic models.

the stages in the column. Process simulation makes extensive use of thermodynamic properties,
and they can have a profound influence on the simulation results. Selecting the right models
sometimes requires insight and experience from the user. It is your responsibility to assess
whether or not the thermodynamic model you have selected is adequate for your needs. The
main specification here is the selection of the K-value model. Other model specifications depend
on the type of K-value model selected. K-values describe the ratio of vapour (yi ) to iquid
(xi ) mole fractions: Ki = yi /xi ; for each component. The various models employ different
assumptions and models for the calculation of the vapour and liquid fugacities from which the
equilibrium phase compositions are derived via the equilibrium relation (which states that at
equilibrium the component fugacities in both phases are equal, see Chapter 10).

Figure 1.8: Context sensitive help.

Upon selecting a K-value model the items in the list for K-models are explained further when
you press F1 for the online help, see Figure 1.8. You can browse the help by selecting the links
that are visible in the various help screens. Use Esc to get out of the online help and select

12
the DECHEMA K-model, UNIFAC for the activity coefficients, and Antoine for the vapor
pressures. These are reasonably good model selections for the fairly ideal Benzene/Toluene
system of our example. Use as enthalpy model ‘none’. This means that the enthalpies for the
liquid and vapor are assumed constant and equal for all the components. This assumption is
often referred to as ‘Equimolar Overflow’ or ‘Constant Molar Overflow’ (CMO). It implies that
the vapor and liquid flows up and down the column are constant from stage to stage (except
around stages that involve feed or product streams).

After selecting the thermodynamic models we need to load the data needed by these models.
As Antoine parameters are already part of the pure component data loaded, and the UNIFAC
model is a group contribution method for which we know the component’s groups, no further
input is required. This is shown in the area where the thermodynamic model parameters are
to be entered with a message that no further input is required. The tab sheet also has a green
checkmark indicating it is finished. However, other models most likely will require specification
of model parameters.

Using F6 we switch to the next input screen. This brings us to the physical properties where we
can define models for densities, viscosities, thermal conductivities, surface tension, and diffu-
sivities. ChemSep provides a default set of model selections so no user input is required. If you
want your own selection, uncheck the checkbox ”Use default methods” and all the selections
will become visible. This is not needed now and we use F6 again for the next input screen,
which is reactions. As benzene and toluene do not react with each other, we can skip this item
and move to the next item in the input: Feeds.

Figure 1.9: Feed state specification.

The next item in the menu is the input of the column feed data. The feed stage has already been
specified in the operation spreadsheet,

Note that ChemSep already specified a name for the feed we entered during the specification of
the column operation. It numbers the feeds in sequence of the feed stage numbers we supplied
but it can be changed here if necessary. Two different types of specifications can be given to fix
the feed state: the temperature and pressure or the vapor fraction and the pressure (of the feed).

13
This specification is necessary to determine how much of the feed is vapor or liquid. This helps
to determines the internal flow rates.

In this example we want to specify a ‘saturated feed’ and we do this by selecting the vapor
fraction specification. When a vapor fraction of zero is specified, the feed is taken to be at its
boiling point (in the simulation a flash calculation is performed to calculate the boiling point
of the feed mixture). Of course, we must specify a pressure to determine the feed temperature.
Here our column operates just above atmospheric pressure in order to prevent ingress of any air
in case any leak might develop.

We enter ”1.1 atm” and we are asked to confirm the conversion to the default units for the
pressure, namely in bar. If you enter the wrong type of units ChemSep will warn you and let
you try again. Next we need to set the feed composition and total flow rate. Here we want
a equimolar feed with a total mass flow rate of 1 kg/s. To do this we first enter equimolar
feedrates of Benzene (B) and Toluene (T). Any amount will do as long as we enter the same
rates for the two components. Then we enter ”1 kg/s” in the total flowrate field. As the interface
knows the molecular weights of the components it can do the conversion of the mass to the
molar flowrate. The 1 kg/s corresponds to 11.7471 mol/s as total feed flow.

Figure 1.10: The feed’s total massflow rate specification of 1 kg/s is automatically
converted to molar flows.

As our simple distillation column only has one feed, we are done and can move to the next input
screen: expand the tree for the ‘specifications’. There are five tab screens listed for this input:
Analysis, Pressures, Heaters/Coolers, Efficiencies, and Column specifications. The analysis
screen shows us the degrees of freedom for the equilibrium staged column and therefore what
all needs to be specified.

We see that we need to specify the pressures on each stage and we do this by selecting the
Pressures tab after we opened the Specifications. Since we configured the column to have a
condenser ChemSep asks us to specify pressure in the condenser first.

There you will notice that each item that needs to be specified is shown as an asterisk (*) in

14
 Figure 1.11: Degrees of freedom analysis for the BT column.

Figure 1.12: The pressures tab sheet.

ChemSep. Upon selecting such a field, the asterisk usually is replaced by a default value or
else a value that is consistent with the other specifications that already have been made. In this
example we already have entered a feed pressure and thus ChemSep assumes that the condenser
pressure equals this pressure. Normally, the condenser is not located on top of the column but
at some lower elevation. There is, therefore, a considerable length of pipe between the column
and the condenser over which a certain pressure drop exists. Thus, it is likely that the condenser
has a lower pressure than the top of the column but to keep this problem simple we assume that
the pressure is the same throughout the column as well as in the condenser. Thus, we accept the
defaults ChemSep provides for the condenser and column top pressures and select the constant
pressure specification.

The degrees of freedom also tells us that on each stage we need to specify a heat duty, repre-
senting either a heater when positive or a cooler when negative. These duties are optional, and
you therefore iwould need to ”insert” such a cooler or heater. When dong so the interface asks
you to specify the stage number, name of the heat exchanger, and its duty. As columns typically
operate at temperatures above ambiant, there is typically also a heat loss over the column. This
is modeled as a separate duty that gets subtracted from each stage, except for the condenser
and reboiler. The default name for this the heat loss is set to ‘Qcolumn’ and you only need to

15
 Figure 1.13: Column heat loss and stage coolers/heaters.

specify a value for it. In this example we will assume the column is insulated well enough that
the heat loss is negligable and therefore we assume the default value of zero.

Figure 1.14: Efficiencies.

After the specifications of the heaters and coolers the next sheet brings us to the stage Efficien-
cies. A default stage efficiency for all the stages is to be specified. This value is used for all
stages in the column except the condenser and reboiler, which are assumed to operate at equi-
librium. We can specify stage specification for specific stages by inserting and deleting stage
efficiencies. These are Murphree vapour efficiencies (see /chaprefSEXrealstages). Here we use
the default efficiency value of one.

Figure 1.15: Condenser specifications.

The next and also the last specification tabsheet is that for the column product specifications.
As we have equipped our column with both a condenser and a reboiler, we have to make two

16
additional specifications that are going to fix the heat duties of these heat exchangers and that
will determine the internal flow rates as well as the overall mass balance by means of the distil-
late and bottoms flows. Again, we see that ChemSep has provided default names for the product
flow names (Top, respectively Bottom) and heat duties of the condenser and reboiler (indicated
with a Q). For the top specification there is a whole list of specifications which we are allowed
to make. The first is the reflux ratio, RR, a very common specification that determines the ratio
between the distillate top product and the flow that is refluxed back to the column. ChemSep
provides a default value for the reflux ratio of two, however, this value will depend very much
on the application and system at hand. This means that the ratio of refluxed liquid flow rate
(R) to the product distillate flow (D) is two (twice as much liquid is returned to the top of the
column as reflux compared to the liquid drawn as distillate product).

For the reboiler select the total bottoms flow rate, B. As we fed an equimolar mixture to the
column, we want to draw equal bottoms as distillate flows, so we specify a bottoms flow of
5.874 mol/s. The operation specifications spreadsheet shows both the these specifications. The
combination of reflux ratio and bottoms flow rate is one that gives the most rapid convergence is
our experience. You can make other combinations but not all will convergence very quickly, and
some combinations might even give rise to impossible specifications. That’s why we advocate
the RR-B combination for each problem you start with. After you have solved the problem with
this specification you can see what the column does and when necessary switch to a different
set of specifications.

Figure 1.16: Column specification for the Benzene-Toluene example.

After the column specifications have been set all the red crosses in the input tree list have
become green checkmarks, meaning we have successfully completed the input specification for
our binary distillation example and can go ahead with solving the column. We can do so by
pressing Alt-S or clicking the little calculator on the toolbar. This will also check our input and
if we forgot something bring us to the window where more input is required or where there is
a problem with the specifications. To solve the problem we must also save it into a simulation
file in order to pass the problem on to the simulator programs. They are separate programs that
read the problem from the simulation file and append the results to the same file. Thus, we are

17
prompted for a valid filename and path. The default extension (.sep) is automatically appended
by ChemSep. After the simulation filename is specified, say ”bt1”, the input is saved and the
column simulator called.

Figure 1.17: Solving the benzene-toluene column.

During the solve process a window shows the progress of the calculations in the form of mes-
sages of reading the problem and while solving a sum of errors for each iteration. This error
sum is a norm for the convergence obtained and needs to become less than the accuracy set
under the solve options (for a column typically 10−3 ). When the norm is small enough the
process ends. Click the window away with the ”Done” button.

1.3 Results

1.3.1 Tables

After the simulation has been completed successfully ChemSep will reread the sep-file and
enable the results tabsheets: Tables, Graphs, McCabe-Thiele, and FUG, the latter is an acronym
for Fenske-Underwood-Gilliand. The interface will show the tables tab with the mass and
energy balances for the column (Fig. 1.18) as default table. Here we can see in whether the
overall and component mass balance have been met. The overall balances should be equal to
zero or close to 0 (within the criterion used to converge the simulation). Also, the component
balances must be (approximately) zero. When this is not the case, the solution found is not a
physically meaningful solution. It is important to always check this.

18
To see a different table, select it from pull down list of the table selector. For example the
T/P/Flow profiles. When selecting this table we see that the reflux ratio we specified was indeed
also obtained and that together with the bottoms flow rate specification this implied a boilup
ratio of three (within the tolerance we supplied at the solve options). We see that the pressures
are all the same throughout the whole column – as we specified – and the temperatures vary
from 85o C in the condenser to 111o C in the reboiler. As we add heat in the reboiler and
remove heat in the condenser, this is the temperature profile one would expect for a distillation
column.

Figure 1.18: Mass and energy balances.

Figure 1.19: Profiles for the temperature, pressure, and column flows.

Note that we can use one of the buttons above the table to export the table into an Microsoft
Excel table or an ASCII table we can edit in a text editor. Alternatively, we can copy the table
to the clipboard. The nice thing is that the table can also be edited from hand in the interface
before we do that. Of course, as soon as we select a new table with the table selector we loose
any changes we make.

The next table (Fig. 1.20) shows the compositions and K values for benzene. Here we see that
the benzene concentrates in the top of the column and that its K-values are larger than one.
As we recall, the K value was defined as the ratio of the vapor over the liquid mole fraction,
this means that the relative amounts of benzene are higher in the vapor than in the liquid. This
is why the fraction of benzene increases as the vapor rises up the column, condensing at 92

19
mole% whereas the feed was only 50%.

Figure 1.20: Profile for the compositions and K values for benzene.

Figure 1.21: Stream table.

The stream table shows a complete overview of the feed(s) and products of the distillation col-
umn. It shows pressure, vapor fraction, temperature, enthalpy, molar and mass flows for the
(component) flows of each stream. In this table we see that the feed is equimolar; both com-
ponent mole fractions are 0.5. The distillate top product and the bottoms product compositions
are almost exactly mirrors of each other in respect to the mole fractions. As there is a small
difference in molecular weights between the components, the weight fractions are somewhat
different. Note that Figure 1.21 we have edited the stream table. Just place the mouse in the
table and start editing.

Two more tables are available: a summary of the specifications and a table showing the output
from the column simulator which was written to the run window. The specification summary
is a very useful table for it contains the complete problem specification of the simulation, with
this table anyone can reproduce the simulation. It also includes the little schematic drawing of
the column flowsheet to show the column configuration.

20
 Figure 1.22: ChemSep simulator output.

1.3.2 Graphs

ChemSep has a rich set of plotting options for the column simulation results. It is much easier
to analyze the results in graphical than in table format (as is true for most other problems). The
graphs tab sheet has a selector that shows the different standard graphs that are available. When
you select a graph it will be displayed using the default graph program, GNUplot. To copy this
graph to the clipboard or to print it, just right-click the GNUplot graph window and you get a
selection of options. Use Alt-F4 to click the graph window away. For a detailed discussion of
GNUplot see the chapter on GNUplot.

Figure 1.23: Graph menu.

The first two kinds of graphs listed in the graph picklist are the liquid and vapor mole fraction
profiles that show how the components are distributed over the column. The default orientation
in ChemSep is to plot the stages vertically, with the condenser at the top and the reboiler on the
bottom. One can imagine the profile projected over the column standing straight up on the left
axis.

21
If we want to change any plot setting, simply select the graph settings checkbox on the right
hand side. This will show the current settings of the graph, the horizontal and vertical axii,
and the data sets plotted in the selected graph. To plot the graph after making changes in any
of these settings, use the ‘Display’ button. With the settings visible we can change the colors,
point types, axes, etc. of the plot as well as add new profiles to be plotted (possibly using the
same or a second axis).

Figure 1.24: Liquid compositions.

Looking at the compositions in Figure 1.24 we observe that benzene concentrates in the top
of the column and toluene in the bottom. This is to be expected as benzene boils at 80 and
toluene at 110.6 o C at atmospheric pressure. Note that the composition profiles show an almost
symmetric behavior. This is an indication that benzene and toluene form a relatively ideal
mixture.

The components in the mixture divide over the liquid and vapor phase as is dictated by the
K-value, which is defined as the ratio of the mole fractions in the two phases present. As we
selected to use the DECHEMA K-model the K-values are given by
yi γi p∗i
Ki ≡ = (1.1)
xi p
As the pressure is constant over the whole column and benzene and toluene form an relatively
ideal mixture, the activity coefficients are close to unity. Therefore, the K values are dictated
mostly by the behavior of the component vapor pressures, which increase with increasing tem-
peratures. Figure 1.25 shows that the K-values and temperatures increase from top to bottom.

22
 Figure 1.25: K values plot.

As the acivity coefficients have very little influence here it must be caused by an increasing
temperature. In Figure 1.26 we indeed see that this is the case. The temperature is increasing
from top to bottom because in the top we have more benzene boiling at 80 degrees and in the
bottom we have more toluene boiling at 110 degrees.

The ratio of the K-values is the relative volatility:

Ki
αij ≡ (1.2)
Kj
where we usually select for component i the lighter component (which evaporates quicker, ergo,
it has a larger vapor pressure and thus a larger K-value). This ratio indicates how easy it is to
separate two components from each other. In our example – using the DECHEMA model – this
ratio becomes:
γi p∗i
αij = (1.3)
γj p∗j
When the activity coefficients are both close to unity (true for ideal mixtures) the relative volatil-
ity is just the ratio of the component vapor pressures. As both vapor pressures increase with
temperature, the relative volatility is often only a weak function of temperature. This is es-
pecially so when the boiling points of the two components are close. In our case the relative
volatility varies from 2.25 in the reboiler to 2.6 in the condenser (Figure 1.27).

However, the relative volatility is not the only thing that influences the separation, so can the

23
 Figure 1.26: Temperature profile.

internal vapor and liquid column flows, as they are responsible for moving components through
the column. Figure 1.28 shows that in this example the vapor flow rates are constant in the
whole column whereas the liquid flows are discontinuous at stage 5.

This can be explained by the fact that the feed enters as a saturated liquid on stage 5. At
this stage we therefore see an increase in the liquid internal flowrate. As the feed is at its
boiling point, no vapor is condensed, nor is any vapor added at the feed stage. The only vapor
generation is done in the reboiler. The difference in the liquid from stage 1 (the condenser) and
the vapor from stage 2 (which enters the condenser) equals the distillate flowrate. We see that
this difference is half the size of the reflux flow from the condenser (the liquid flow stage 1);
thus, the reflux ratio is equal to 2.

The stripping factor is defined as the ratio of the component vapor flow going to the stage above
and the component liquid flow going to the stage below:
yi V Ki V
Si ≡ = (1.4)
xi L L
Therefore, it is the stripping factor actually indicating the direction a component is moving
(up or down) in the column. If it is larger than one, this means the component flow upwards
in the column is larger than the component flow downwards at this stage. As we can see in
Figure 1.29, the stripping factor for benzene is larger than unity at all points in the column,
hence, the composition of benzene is decreasing monotoniously from the top to the bottom of

24
 Figure 1.27: Relative volatility.

the column.

Figure 1.30 shows that the mass transfer rates for the two components are symmetrically oppo-
site. In this plot the mass transfer rates are defined to be positive for condensing components.
That is why the mass transfer rates for benzene are negative as it evaporates at each stage in the
column (note the condenser and reboiler are not shown!). The fact that the profiles for benzene
and toluene are symmetric is obvious when we realize that we are separating a binary mixture
under equimolar overflow conditions: if one component evaporates the same amount of the
other component needs to condense.

In another useful diagram we plot the ratio of the liquid mole fractions:
xi
Rij ≡ (1.5)
xj
see Figure 1.31. This plot shows the actual separation achieved between components i and j.
This ratio is plotted on a logarithmic scale as when each stage removes a certain fraction of a
component, the composition of a component will decrease exponentially to zero.

This plot can be used to check whether a feed is properly placed to the column. If the ratio
shows a abrupt change in direction, this is an indication that the separation could be improved
by moving the feed to a different stage. In our case the curve is changing its direction in
a continuous manner. A better way to determine the feed placement is the McCabe-Thiele

25
 Figure 1.28: Internal flows.

analysis.

1.3.3 The McCabe-Thiele Diagram

The McCabe-Thiele diagram was used by McCabe and Thiele (1923) to provide a graphical
solution to the distillation equations. It is based on a binary x-y diagram, see Figure 1.33. For
ideal binary mixtures where the relative volatility is constant the vapor mole fraction can be
calculated from:
K1 y1 /x1 y1 (1 − x1 )
α12 = = = (1.6)
K2 y2 /x2 x1 (1 − y1 )
solving for y1 this gives:
α12 x1
y1 = (1.7)
1 + x1 (α12 − 1)
This is the curved line in the diagram. Of course, ChemSep uses the actual K-model to calculate
the equilibrium curve and in this case the relative volatility is not constant, as we saw in Fig-
ure 1.27. In the x-y diagram we add the y = x line as an auxiliary line. The larger the relative
volatility, the further away the equilibrium line is from the y = x line and the fewer stages are
required to separate the two components.

At x = 0.5 another small line is drawn that represents the feed or ‘q’ line. A number indicates

26
 Figure 1.29: Stripping factors.

the feed stage (5). Here ‘q’ stand for the ‘quality’ or liquid fraction in the feed, L/F . Using the
component balance for the feed divided by the feed flow F we find for z (the feed overall mole
fraction):
L V
z =x +y (1.8)
F F
substituting the total molar balance V = F − L and q = L/F gives

z = qx + (1 − q)y (1.9)

which can be rewritten to determine y as

1 q
y= z− x (1.10)
(1 − q) (1 − q)

when q = 1 Eq. (1.9) represents a vertical line at x = z, as is shown in our example. From
Eq. (1.10) we also see that the q-line crosses the y = x line at x = z.

The other solid lines in the diagram are the operating lines. These lines define the molar bal-
ances for the two column sections in our column. If we look at the top (rectifying) part of the
column, we have an internal vapor column flow V entering a stage, and an internal liquid flow
L leaving the stage. For the condenser (stage 1) this is:

V2 = L1 + D (1.11)

27
 Figure 1.30: Mass transfer rates.

The reflux ratio is defined as R = L1 /D. We can write for the stage below the condenser:

V3 = L2 + D (1.12)

When the the vapor and liquid enthalpies are constant, the amounts of liquid and vapor do not
change (if there are no feeds!); thus, we can write:

V = L + D = (R + 1)D (1.13)

The same is true for the component molar balances:

V y = Lx + DxD (1.14)

This is a straight line in our diagram with:

L D R xD
y= x + xD = x+ (1.15)
V V R+1 R+1
This is a straight line with slope L/V (or R/(R + 1)) which crosses the y = x line at xD .

Now the same derivation can be done for the lower part of the column where bottoms flow B
with composition xB and the L0 and V 0 internal flows give:

L0 B
y= x − 0 xB (1.16)
V0 V

28
 Figure 1.31: Key ratios.

Figure 1.32: McCabe-Thiele tabsheet.

This is also a straight line with slope L0 /V 0 which crosses the y = x line at xB . Now it
becomes clear why we used the y = x line as our auxiliary line: the two operating lines and the
q line cross this line at the feed and top/bottom compositions, respectively.

The dotted lines drawn in between the equilibrium line and the operating lines represent the
stages. We start drawing at the reboiler composition xB on the y = x line. The reboiler
generates a vapor with a vapor composition of yB , which is the y value on the equilibrium line
at x = xB . In other words, we draw a dotted line straight up to the equilibrium line. The liquid
composition from the stage above can be obtained by moving horizontally to the operating line,
as the operating line determined the mass balance for the internal flows. Then, for the stage

29
 Figure 1.33: McCabe-Thiele diagram.

above we draw again a vertical line to the equilibrium line and repeat this process. When we
pass the feed stage we switch from using the stripping operating line to using the rectification
operating line. A total condenser does not generate another triangle as all the vapor is condensed
into a liquid and no further separation is achieved.

If instead of a saturated liquid feed we want to specify a saturated vapor feed we need to go
back to the feeds specification by clicking on the ‘Feeds’ item in the tree on the left. Specify a
vapor fraction of 1 and use Alt-S to initiate the checks and solving. You are now asked whether
to overwrite the simulation file. Don’t, instead save the new problem under a different name,
say bt2.sep.

The resulting McCabe-Thiele diagram looks quite different. Observe that the distillate and
bottoms compositions are closer to the feed composition. To view what is going on around the
feed line we will make an enlargement of this area: select the graph settings checkbox in the
McCabe-Thiele tabsheet and enter xmin = 0.2, xmax = 0.6, ymin = 0.3, and ymax = 0.7.
When we then select the ‘Display’ button we see that the feed is now a vapor feed as the ‘q-line’
is a horizontal line. The slope of the q-line is given by q/(1 − q) which is zero if q = 0. This
results in a misplaced feed and we have to keep stepping off stages on the stripping operating
line when it would have been more advantageous to step off stages on the rectifying operating
line. The two stages in the column are much less effective as they could be and hence the
column products have lower purities.

30
 Figure 1.34: McCabe-Thiele diagram for column with a vapor feed.

If the vapor fraction is specified as 0.5, we can see that the feed line has a slope of -1 and that
the feed is again much better placed (but still not optimally). In this case stage 6 would be the
optimal feed stage.

The feed quality can also be defined as:

L0 − L
q≡ (1.17)
F
which has the advantage that the feed quality can also be larger than unity or smaller than
zero. This occurs when subcooled liquid or superheated vapor feeds are introduced in the
column. A subcooled liquid needs to be heated up to its boiling point, in which it will condense
a certain amount of vapor generating a higher internal liquid flow L0 . The opposite is true for a
superheated vapor. For a subcooled liquid the slope of the feed line is even positive, the same
for a superheated vapor.

Subcooled liquids are more common than superheated vapors for distillation columns for the
simple fact that distillation feeds are often liquids (from storage) and a pre-heater represents an
extra piece of equipment. The same task can be taken over by the reboiler, albeit at a higher
temperature than the feed, requiring energy at a higher temperature (as well as shifting the
operating line in the stripping section of the column).

When we change the reflux ratio of the column from 2 to 1 (Fig. 1.36), we see a change in the

31
Figure 1.35: McCabe-Thiele diagram for a feed with equal amounts of vapor and
liquid.

McCabe-Thiele diagram, namely that the operating lines have different slopes. As was shown
before, the slopes of the operating lines are given by the ratios of the internal flows. For the
rectification line we had a slope equal to R/(R + 1). For R = 2 this gives a slope of 2/3 and
for R = 1 only a slope of 1/2.

We also see that the operating lines intersect each other almost on the equilibrium line. When
they intersect the equilibrium line we would need an infinite number of stages to switch from
the stripping to the rectification operating lines. Therefore, this is called a ‘pinch-point’, as the
stages get pinched between the equilibrium and the operating lines. The reflux ratio at which
the operating lines intersect the point at which the feed line crosses the equilibrium line is called
the minimum reflux ratio. At that point the column has an infinite number of stages.

On the other hand, if we increase the reflux ratio to 10 (/figrefFEXrr10) we observe that the
operating lines lie closer to the y = x line and that the stepping of stages leads to larger
changes in x resulting in a better separation. If we further increase the reflux the operating lines
move closer to the y = x line until, at total reflux, they coincide with the y = x line. Of course,
at total reflux (R = ∞), all the vapor overhead is condensed and returned to the column so no
product can be drawn (and hence, there is no production). At this point the separation uses the
lowest number of stages to achieve a specific separation.

32
 Figure 1.36: McCabe-Thiele diagram with a reflux ratio of 1.

We see that there are two extremes for the separation for a given set of distillate and bottom
compositions: the minimum reflux ratio at the pinch point and the minimum number of stages
at total reflux. Real operation must lie somewhere between these two extremes. Typically,
distillation columns operate at reflux ratios that are 1.1 and 1.5 times the minimum reflux ratio.
Reset the reflux ratio to 2.

1.4 Real stages

Previously we assumped constant equimolar overflow (CMO) causing the flow rates to only
change at a feed or side draw in the column. In a real column, however, the various components
have different latent heats of vaporization as a result we do not have equal moles being con-
densed and vaporized on each stage. When this assumption is relaxed, we must calculate the
enthalpies of the internal and feed/product streams in and to/from the column. In order to do so
we must select a different enthalpy model: in the thermodynamic models under ‘Properties’ we
select the ‘Excess’ enthalpy model.

When we redo our column simulation (using the feed at bubble point and a reflux ratio of 2)
and look at the flow profiles we see that the internal flows now do change from stage to stage.
Figure 1.39 shows the difference between using no enthalpy and the excess enthalpy models.

33
 Figure 1.37: McCabe-Thiele diagram with a reflux ratio of 10.

Figure 1.38: Selection of the Excess enthalpy model.

As the internal flows change slightly, so do the slopes of the operating lines. The operating
lines in the McCabe-Thiele plot now become curved instead of straight lines (in this case the
curvature is rather small so we have left out this plot).

When we select the Excess enthalpy model we are able to construct a Ponchon-Savarit diagram;
in ChemSep it is actually just a H-xy plot (Figure 1.40). Note that the vapor enthalpies are
positive and liquid enthalpies are negative, the difference being the heat of vaporization. Vapor
enthalpies are positive because the reference point for enthalpies is taken as gas at 298 K. As the

34
Figure 1.39: Difference in internal flows for no enthalpy model (dotted lines, CMO)
and ‘Excess’ enthalpy model (solid lines).

column temperatures are higher than 25o C, the vapor enthalpies are larger than 0 (see 10.1.7
for how enthalpies are calculated). It is important to know what the reference point is when you
are using enthalpies!

Another aspects of real column behavior is that the stages do not operate at equilibrium because
of mass transfer limitations. So we make use of a stage efficiency that describes the degree of
which equilibrium is approached. The most commonly used efficiency definition is the Mur-
phree stage efficiency:
actual change in vapor
EMV ≡ (1.18)
change in vapor for equilibrium stage
or
yj − yj+1
EjM V ≡ (1.19)
yj∗ − yj+1
where yi∗ is the vapor composition in equilibrium with the liquid composition xi on the stage j.
With this definition the equilibrium relations are

0 = EiM V Ki xi − (1 − EiM V )yi,j+1 − yi (1.20)

Note that when the efficiency is one this reduces to

0 = Ki xi − yi (1.21)

35
 Figure 1.40: H-xy plot.

which equals our definition of the K-values.

In actual columns each stage achieves only a partial separation and the efficiency is defined
as the achieved fraction of what is theoretically possible. In most cases the Murphree stage
efficiency is taken to be equal for all components. In a binary system, this theoretically correct.
However, this is not the case for multicomponent systems. In fact, in the next chapter we will
see the limitations of this assumption.

Stage efficiencies can be entered in the appropriate input spreadsheet under the Specifications
– Efficiencies tab. Here, a default efficiency is defined for all the stages except for the con-
denser and reboiler (both are assumed to be at equilibrium). In an actual column the efficiencies
change from stage to stage. Let’s assume a stage efficiency of 50% and resolve the problem,
see Fig. 1.41. When we compare Figure ?? with Figure 1.33 we see that the separation is worse
when the Murphree efficiency is smaller than one — as can be expected, of course. Note that
efficiencies can also change the optimal feed location and/or quality.

In the McCabe-Thiele plot the ratio in Eq. (1.18) represents a fraction of the vertical line that
is drawn from the operating line to the equilibrium line. Each stage only achieves a separation
equal to this fraction of the (vertical) distance between the two lines. The reboiler and condenser
are exceptions: they are assumed to operate at equilibrium.

36
Figure 1.41: McCabe-Thiele diagram for a column with a saturated liquid feed, reflux
ratio of 2, and a Murphree stage efficiency of 0.5.

1.5 Parametric Study

When the column simulation converged we often want to study the influence of our column
specifications. For example, the reflux ratio has a profound influence on the column operation
with respect to product purities as well as condenser temperature. ChemSep allow you to do
this automatically via the Parametric Study option under the Analysis menu. Figure 1.42 shows
how this can be done. We selected to vary the reflux ratio for our initial problem over a range
of 1 to 5. We have to select each variable we vary as well as the start and end value. All input
variables are varied simultaneously from their start to end values with increments determined
by the number of steps and their individually set start and end values. When you want to select
a new variable ChemSep shows you a dropdown list of all the variables you can vary.

Each result we want to see we need to select as well, so that the results are stored after each
simulation and can be used for plotting at a later stage. Again we can select result variables
with a dropdown list. Note that there are more variables to pick from than that can be varied as
specifications. Figure 1.43 shows the plotted results for when we varied the reflux ratio. The
mole fraction of the heavy component in the distillate is drawn on the left axis on a logarithmic
scale. Note that the purity of the distillate increases with increasing reflux ratio. Also note that
the condenser temperature asymptotically reaches the boiling point of Benzene at the specified

37
 Figure 1.42: Parametric study varying the reflux ratio from 1 to 5.

pressure (1 atm).

1.6 Exercises

1.6.1 High Pressure Distillation

A ‘C2 splitter’ is a high pressure column used to separate the important chemical ethylene from
ethane. The two components boil at very low temperatures, so the column must be operated
at as high a pressure as possible in order to be able to condense the column overhead. As the
relative volatility is reduced at higher pressures, this makes the separation even more difficult.
At high pressures that approach the critical point we need to use an equation of state to describe
the phase equilibria. The vapor phase becomes much denser and the density difference between
vapor and liquid is much smaller, hence, the physical separation also becomes more difficult.
At the critical point the density difference vanishes and computational problems will manifest
themselves at such high pressures. The column pressure is determined by two considerations:

• it should be as high as possible to reduce refrigeration costs for the reflux

• it must be below the critical pressure of both components

38
Figure 1.43: Composition of heavy component in the distillate and condenser temper-
ature as function of the reflux ratio.

• it must be low enough so that the density difference between vapour and liquid is large
enough to cause an efficient phase separation in the downcomers of the distillation trays

In practice the overhead temperature would be between -30 and -10 o F . Design a column to
separate ethylene and ethane. The feed, which consists of 100 mol/s of both components is at
its boiling point at the column pressure. The ethylene top product has to be very pure, with a
mole fraction of ethane of less than 0.001i, in order to be polyethylne grade quality. We also
wish to recover at least 90% of the ethylene in the feed.

1.6.2 Column Sequence

Design a sequence of columns to separate the following mixture: 14.08 lbmol/h n-C4, 19.53
lbmol/h n-C5, 24.78 lbmol/h n-C6, and 33.94 lbmol/h n-C8. The column pressure is 1.7
atm and all product streams must be 98.5% pure.

39
1.6.3 Propanol Removal from Ethanol - Water Mixture using a
Side Drawoff

Ethanol from fermentation contains traces of any number of components. These often have to
be removed down to very low levels because of their smell. This is not always easy.

Design a distillation column (using ChemSep) to separate a feed of 10 mol/s ethanol and
89 mol/s water. These two components form an azeotrope with a mole fraction of about 0.89
ethanol (this is at atmospheric pressure, the azeotrope mole fraction changes with pressure). The
top product should have a composition close to the azeotropic composition, the bottom product
should be almost pure water. Use DECHEMA/UNIFAC/Antoine/Excess for the thermodynamic
models.

Add a trace of 1-propanol (1 mol/s) to the feed of the column created above. Rerun the simula-
tion and look at the liquid composition profiles. Why is there a maximum in the mole fraction of
1-propanol? To help answer this question look at the K-value profiles and answer the questions
below.

1. What are the normal boiling points of ethanol, 1-propanol, and water? (Use the compo-
nent search engine and select components by boiling point).
2. Do you think that a mixture of ethanol and 1-propanol is almost ideal or strongly non-
ideal? Why?
3. Do you think that a mixture of 1-propanol and water is almost ideal or strongly non-
ideal? Why?
4. How do you think you might test your answers to questions 2 and 3?
5. In the column created in part 2 above is the 1-propanol in the top part of the column
(where ethanol is the major component) trying to go up or trying to go down? Why?
6. In the column created in part 2 above is the 1-propanol in the bottom part of the column
(where ethanol is the major component) trying to go up or trying to go down? Why?

After some time 1-propanol can accumulate in the middle of the column and it may cause all
kinds of problems. We can avoid this by removing the propanol with a side-draw. Redesign the
column so that you remove all (or nearly all) of the propanol with a side stream. At the same
time try to maximize the propanol concentration in the side draw (it can be much higher than in
the feed).

1.6.4 Extractive Distillation

Toluene and n-heptane are difficult to separate by simple distillation. They can, however, be
separated quite easily by extractive distillation using phenol as a solvent. Design a facility to
process a stream containing 200 lbmol/h toluene and 200 lbmol/h of n-heptane that is at 200

40
o
F and 20 psia. The design objective is to obtain a heptane product of at least 99 mole% and
a maximum of 2.0 lbmol/h of heptane in the toluene product. Phenol is available at 220 o F
and 20 psia. In view of the high cost of the solvent, your design should provide for solvent
recovery and recycle.

1.6.5 Azeotropic Distillation

Azeotropic distillation columns can be very difficult to model in the sense that getting the equa-
tions to converge can be very hard to do. Some simulation programs can have serious difficulties
with these problems. Here we investigate the very well known problem of separating ethanol
(1) from water (3) using benzene (2) as an entrainer.

Set up an azeotropic distillation column with 41 stages, a partial reboiler, and no condenser.
There is a feed containing 85.64 kmol/h ethanol and 14.36 kmol/h water at a pressure of
1.027 bar and a temperature of 311 K to stage 5. In the absence of a condenser we simulate the
reflux flow with a second feed to stage 1 containing 103.21 kmol/h of ethanol, 215.8 kmol/h
benzene, and 56.29 kmol/h water at a pressure of 1.013 bar and a temperature of 298 K. The
specifications provided here are very similar to a version of this problem studied by Prokopakis
and Seider (AIChEJ, Vol. 29, p. 49, 1983). We will make the following modifications: The
bottoms flow rate is 83.0 mol/h and the number of stages is 41.

There are no less than three solutions for this set of specifications; your task is to find them.
These calculations are VERY sensitive to the VLE models and parameters. Here we will use
the DECHEMA / UNIQUAC Q’ / Antoine combination of models.

Since these calculations can be hard to converge it is advisable to limit the changes in flows
and compositions between iterations. Start by pressing F6 and go to Solve options. In that
spreadsheet set the default number of iterations to 100 (but don’t be alarmed if you sometimes
need more than this), the flow limit to 0.3 and the composition step to 0.1. This helps to
stop the flows, temperatures and mole fractions from taking on ridiculous values during the
computations.

Solution number 1 can be found by letting ChemSep do what it wants. Study the profiles
carefully. Is this a good way to operate the column?

A second solution can be found as follows.

1. Set the efficiency to 0.1 and solve the problem using the Automatic initialization.
2. Go to Solve options and change the initialization option to Old Results.
3. Change the efficiency. Go back to step 1 and solve the problem again.

Look at how the profiles change. Compare the final set of profiles with those of the first case.
Which is more desirable?

41
Now try and find the third solution. You can use a method very similar to that used above, just
change the value of the efficiency used in step 1. Good luck.

Try repeating this exercise with UNIFAC instead of UNIQUAC as the activity model. Are the
profiles different? Can you find multiple solutions for this case? Think about the consequences
for design and operation of a plant that displays this kind of behavior. Note that some columns
have even more than three steady state solutions. Examples with 5 and 9 are also known!

1.6.6 Separation of Benzene and n-Pentane using Liquid-Liquid

Extraction

An equimolar mixture of n-pentane (1) and benzene (2), at a total flow rate of 100 mol/s is
to be separated by liquid-liquid extraction. This operation might be carried out in equipment
very much like distillation columns except that there is no reboiler or condenser. Our extractor
is to operate at 50 o C. It is required to recover 99.98% of the benzene. Further the purity of
the extract is to be such that not more than 2.5 mole% n-pentane can be present in the extract
product. The solvent is sulfolane (3).

Determine the solvent flow rate and the number of equilibrium stages required to achieve the
recovery and purity targets specified above. Assume that the optimum design corresponds to
the use of the lowest possible solvent to feed ratio for a specified number of stages. In addition,
answer the following questions:

1. What is the chemical formula of sulfolane?

2. What is it’s structure?
3. What is the boiling point of sulfolane?

Table 1.3: UNIQUAC interaction parameters (K) for the LLX example.

i j Aij Aji
1 2 179.86 -95.7
1 3 375.93 247.77
2 3 131.51 -6.28

1.6.7 Absorber Design

A gas stream flowing at 80 mol/s contains ammonia (mole fraction 0.416) and air (mole frac-
tion 0.584). The temperature of the gas stream is 20 o C and the pressure is 1 atm. The ammonia
content must be reduced to ppm levels before the air stream can be vented to the atmosphere.

42
Fortunately, ammonia is readily absorbed in water so it has been decided to build a simple
absorption column to process the stream in question.

Your task, to design the column, is complicated by the fact that the design program available to
you does not have the most appropriate thermodynamic models for this system (Henry’s law).
The staff thermodynamicist has suggested that simple polynomial models (which are available
in your software system) will suffice for the K-values and enthalpies over the temperature range
of interest. Analysis of some ancient (graphical) data in the company files has shown that the
K-values can be correlated by a simple expression of the form:

Kim = αi + βi T (1.22)

where T is the temperature in Kelvin. The enthalpy of the vapor and liquid phases can be
calculated from similar polynomials:

HiV = Ai + Bi T (1.23)
HiL = ai + bi T (1.24)

All the parameters are given in Table 1.4.

Table 1.4: K value and enthalpy parameters (J/kmol) for the absorber

Component αi βi m Ai Bi ai bi
Ammonia -32.285 0.1137 1 3.57e7 3.52e4 0 7.5e4
Water -1.231833 0.00427325 1 4.3486e7 3.35e4 0 7.5e4
Air -191835 900 1 0 2.92e4 0 7.5e4

In addition to providing a column design adequate to the task at hand, please explain

1. The physical significance of the parameters in the K-value and enthalpy correlations.
2. The shape of the temperature profile for the optimum column design.

References

J.D. Seader, E.J. Henley, Separation Process Principles, Chapter 11.7, Wiley, New York (1998),
pp. 631–641.

43
44
Chapter 2

Nonequilibrium Columns

In this chapter we introduce the modelling of multicomponent separation processes by using the
nonequilibrium model. We start with the benzene-toluene distillation column that we encoun-
tered in the preceding chapter, followed by a depropanizer with sieve trays, a nonisothermal
packed absorber where acetone is absorbed in water, and finally the extractive distillation of
methyl cyclohexane using phenol as a solvent in an ‘old’ column with bubble cap trays. These
examples illustrate that when working with systems with more than two compounds we need
abandon the use of Murphree efficiencies and switch to a full multicomponent mass transfer
model instead. And in doing so we must start using the Baur efficiency as performance indica-
tor for our column internals hardware.

2.1 Equilibrium and Nonequilibrium Models

Most mathematical descriptions of a distillation column begins with the concept of ‘stages’:
fairly large parts of the column that have well defined boundaries (Figure 2.1). For a tray
column, we can regard a single tray as a stage. There are well defined flows entering and
leaving one of our stages. These can be the main vapor and liquid flows through the column, or
feed or drawoff flows. There is no natural boundary to the stages in packed columns (unless we
consider packed sections as stages). Even so, chemical engineers often try to subdivide packed
columns into stages, sometimes with ludicrous results.

Conceptually, the simplest kind of stage is the ‘equilibrium stage’ or ‘theoretical plate’. For
such a stage (which does not exist in reality), the (vapor and liquid) streams leaving the stage
are assumed to be in equilibrium with each other (Figure 2.2). This implies that the chemical
potential of any component has the same value in vapor and liquid, and also that vapour and
liquid have the same temperature.

45
 Figure 2.1: Plant, separation equipment, and stages.

Figure 2.2: The equilibrium stage model.

To describe an equilibrium stage model, we ‘only’ need phase equilibrium data (thermodynam-
ics) and mass and energy balances. The model is beautifully consistent, and has attracted the
attention of generations of engineers and thermodynamicists. With the equilibrium stage model
we can simulate many of the properties of real systems.

The equilibrium stage model it has one great weakness: it has no direct connection to real
equipment. Of course, engineers have found ways of coupling our theoretical models to practi-
cal equipment. The two concepts mostly used for this purpose are (Figure 2.3):

• the (Murphree vapor) ‘efficiency’ of a tray or plate and

46
 • the ‘Height Equivalent to a Theoretical Plate’ (HETP) for packings.

When using a Murphree stage efficiency, we assume that the change in vapor composition on
a real tray is a certain fraction of that obtained in an equilibrium stage. This allows us to
approximate the behaviour of real trays. The concept works quite well for binary separations;
the two components there have equal efficiencies. Also the efficiencies are often fairly constant
along a column, and therefore an efficiency is a useful way of summarising practical experience.
It often has a value of about 0.7.

For packings, we look at the composition of the liquid at a certain height. We then assume that
further up in the column, we will be able to find vapor with a composition that is in equilibrium
with the liquid considered. The difference in height between these two points is the HETP.
Also here, this works fairly well for binary mixtures. The HETP’s are the same for the two
components and they are often fairly constant along a column. Experience shows that the value
of the HETP is usually a few tenths of a meter.

Figure 2.3: Tray efficiency and ‘Height Equivalent to a Theoretical Plate’ lead to real
equipment.

Although the efficiency and the HETP have been useful for summarising experience on trays
and packings, these concepts have two serious deficiencies:

• they are not easily related to the construction and behaviour of equipment, and
• in multicomponent mixtures, their behaviour can be extremely confusing.

2.1.1 ‘Exact’ models

If we do not wish to use efficiencies or HETP’s, what should we do? The dream of the modern
engineer is to subdivide each piece of equipment into zillions of little elements dx.dy.dz, each
of which will contain part of one of the phases in the equipment (Figure 2.4). We should then

47
set up difference forms of the Navier-Stokes equations to define the hydraulic flows, the Fick or
Maxwell-Stefan equations to calculate diffusion flow, the Fourier equations for energy transfer
together with all of the required balances and boundary conditions. This vast array of equations
will then to be solved by our brilliant parallel supercomputer algorithms, and they will tell us
everything we wish to know. Such a model can be used for trays, for packings or for any other
piece of equipment. It will deliver beautiful and impressive multidimensional color plots. It will
describe unsteady, turbulent, chaotic two phase flow, the breaking up and coalescence of drops,
their moving and deforming boundaries, interfacial motion and mass transfer, incomplete and
irregular wetting of packings, gas and liquid entrainment...

Unfortunately, there are still a few problems. For a realistic simulation of the important phe-
nomena in a column, our guess is that we would need elements with dimensions of about one
micrometer. That is one million elements per meter, or 1018 elements per m3 of column. This
number is several times the number of bytes on a multi-gigabyte hard disk. So it looks as if we
still have a long way to go before we can simulate the details of multiphase flow. Of course this
also means that the subject is a challenging one, and possibly a good topic to let our students
and young engineers sharpen their teeth. However, with this approach, we should not expect
complete solutions to practical problems in the near future.

Figure 2.4: Exact models: the dream of the modern engineer.

2.1.2 The Nonequilibrium Model

The nonequilibrium model is less ambitious. It contains a smaller number of elements per
m3 than the ‘exact’ models above. Say, five, instead of the 1018 that we were considering.
We choose these elements such that they allow a simple simulation of what we think are the
important characteristics of our mass transfer equipment:

• the equilibrium distribution of all components between the two phases (thermodynam-
ics),
• the effects of large scale flow and mixing patterns, and
• the effects of diffusional heat and mass transfer resistance near phase interfaces.

48
How few elements we choose, is a balancing act. With too few elements our model will not
be realistic, with too many elements it will be impossible to obtain the model parameters (al-
though., with 1018 elements, you do not need any model parameters).

The motion of the gas and liquid on and around the trays or packing has a large effect on equip-
ment performance. The ideal flow pattern in separation equipment usually is countercurrent
plug flow. Modern structured packings give patterns that are fairly close to this ideal. However,
even there, phenomena like turbulence, differences between the channels, and maldistribution
of liquid or vapor at feed points cause deviations from ideality (Figure 2.5).

Figure 2.5: Even in a structured packing, flow is not ideal.

Entrainment of liquid by the vapor, and of vapor by the liquid, causes backmixing. In large, long
column sections, irregularities of the flow have been found to develop, especially near walls and
due to the contacting planes between packing elements.

On trays, the vapor is often assumed to pass upwards in plug flow. The liquid has a cross flow
pattern; at the sides of columns there can more or less stagnant zones. On short tray passes the
overall effect is often similar to that of complete mixing of the liquid. If we have a spray of
drops on the tray, there will be a range of drop sizes and the behaviour of small and large drops
may differ considerably. We may expect small drops to approach equilibrium more rapidly than
large drops. Something similar applies to the bubbles of different sizes in a froth.

The simplest flow model is that where both phases are assumed to be ideally mixed (Figure 2.6).
This mixed flow model only contains two bulk phases with convective flows in and out. In
addition it contains three other elements: the phase interface and two mass transfer resistances
on either side of the interface. We discuss these elements in the next section.

Flow models that are a bit more realistic are (see Figure 2.7):

• on a tray: liquid mixed, vapor in plug flow, no entrainment/backmixing,

• in a packing: both phases in plug flow, no maldistribution.

49
 Figure 2.6: The simplest nonequilibrium model.

Even these models are not always adequate.

Figure 2.7: More realistic flow models.

There are many different ways to describe more complicated flow patterns (Figure 2.8). We
can approximate plug flow by stacking a number of mixed elements. When a phase consists of
coarse and fine parts (such as large and small drops), we may split it in two pseudo phases. We
may expect a convective exchange between these pseudo phases. Entrainment and backmixing
can be simulated by allowing extra flows going in and out of each phase. Finally, maldistribution
can be simulated using parallel stages with different flow ratio’s. The problem with these more
complicated models is, that we do not know their parameters well enough.

At the interface between the two phases, we assume (also in a nonequilibrium model) that
equilibrium exists. (In absorption with a chemical reaction, this may not be true, but we do not
consider that here).

The same equations that we use in equilibrium stage calculations are used to yield the distri-
bution coefficients of all components at the interface. In nonequilibrium calculations we also

50
 Figure 2.8: Complicated flow patterns using simple elements.

use the same equations to calculate the driving forces for mass transfer (which are the chemical
potential gradients, as we shall see below).

An essential part of a nonequilibrium model is the description of mass and heat transfer between
the phases. In most equipment, flow of both phases is highly turbulent (this applies both to the
continuous and the dispersed phase). Turbulence provides a rapid levelling of concentration
and temperature differences. So on not-too-large scales (say up to ten centimeters), the concen-
trations and temperatures in the bulk of fluids can be taken as homogeneous. However, near
phase interfaces, turbulence dies out (Figure 2.9). Eddies do not pass across interfaces. Near
these interfaces, only three transport mechanisms remain, all of which are much slower than
turbulence:

• ‘drift’: convective transport of both heat and matter due to an overall displacement of
matter across an interface,

• diffusion: the transport of matter due to composition gradients and molecular motion
within the fluid, and

• conduction: the transport of energy by a temperature gradient and molecular motion.

The layers (or ‘films’) where turbulence does not dominate transport, are very thin: roughly one
tenth of a millimeter in gases and ten micrometers in liquids.

The traditional description of mass transfer begins with Fick’s law. This states that flux of a
component (in a binary mixture) is proportional to the concentration gradient of that component
and a diffusivity. In addition, the component may also be carried along by drift:

dxi
Ni = diffusion + drift = −ct Di + Ntotal xi (2.1)
dz

51
For most engineering applications, a difference approximation of this equation suffices:

Di
Ni = −kibinary ct ∆xi + Ntotal xi kibinary = (2.2)
δ

The parameter kibinary is a binary mass transfer coefficient.

Figure 2.9: Mass transfer resistance: the ‘film’ near interfaces.

Unfortunately, the description above does not work well for mixtures with more than two com-
ponents. (And most mixtures do contain more than two components!) This is for two reasons:

• it does not take the interactions between the different components into account, and
• it does not allow for the fact that the driving force for mass transfer is not the concentra-
tion gradient, but the gradient of the chemical potential.

The first effect is often important, especially for the behavior of trace components. The second
effect can be important is systems where liquid resistance dominates.

If we take the trouble to make a nonequilibrium model of a column, we can best use a fun-
damental multicomponent model of mass transfer through the interface. A good choice are
the Maxwell-Stefan equations which say that the driving force on i will cause it to move with
respect to the mixture. This motion is counteracted by friction with the other species j.

The driving force is the chemical potential gradient:

dµi
(driving force on i) = − (2.3)
dz
This is a real driving force, with units of Newtons per mole. If we have equilibrium, we have no
gradients, no driving forces and no motion. The chemical potential contains contributions due
to mole the fraction and the activity coefficient of i:

µi = RT ln(γi xi ) + constant (2.4)

The contribution of the mole fraction gradient is similar to that of the concentration in the Fick
equation. However, that equation does not consider the effect of non-ideality via the activity
coefficient.

52
The friction force between i and j (per mole of i) is proportional to the local mole fraction of
j, a friction coefficient ζij and the difference in drift velocities between the two components:
(friction)ij = xj ζij (ui − uj ) (2.5)
When we equate the driving force with the sum of all friction forces on i, we obtain the
Maxwell-Stefan equations:
n
dµi X
− = xj ζij (ui − uj ) (2.6)
dz
j=1

These equations allow a consistent description of diffusion in mixtures with any number of
components.

Usually we are not interested in species velocities through the interface, but in their fluxes (flow
rates per unit area):
Ni = c i u i (2.7)
and we normally use multicomponent diffusion coefficients instead of friction coefficients. Fi-
nally, we will use a difference form of the resulting equations that contain the multicomponent
mass transfer coefficients:
Dij
kij = (2.8)
δ
We will not write out these equations; the important thing to understand, is that the multicom-
ponent diffusivities and mass transfer coefficients can be estimated fairly well from binary data.

In addition to the mass transfer equations, we need heat transfer equations. These require esti-
mates of the heat transfer coefficients in both phases, and they must incorporate heat transport
through the interface due to drift.

2.1.3 Practicalities

We have restricted ourselves to a brief (and incomplete) summary of modern approaches to

the modelling of separation processes like distillation and absorption. Chapter 13 describes the
equilibrium stage model, providing all the equations and discussing in some detail many of the
methods used to solve the model equations. Chapter 14 provides more detail for the nonequi-
librium model. The numerical methods needed to solve the model equations are described in
Chapter 11.

We only need two kinds of physical properties for the equilibrium stage model: Equilibrium ra-
tios, also known as K-values, and enthalpies. The nonequilibrium model requires more physical
properties. For this model we need, in addition to the thermodynamic data identified above, all
properties that are required to describe the flow, heat transfer and mass transfer in the equipment,
such as: densities, viscosities, surface tension, interfacial area, diffusion coefficients, thermal
conductivity, heat and mass transfer coefficients, and so on.

Equilibrium and nonequilibrium models also differ greatly in how detailed a description of the
equipment is required. In the equilibrium model, we only need to give the number of stages and

53
 Figure 2.10: The NE-model contains a ‘drawing’ of the column.

the location of the different feed and product withdrawal points. The nonequilibrium model,
in contrast, needs to contain a complete ‘drawing’ of the column (Figure 2.10): the number of
trays in the different sections, the diameters of the sections, the tray heights, the downcomer
configurations, the hole diameters, the free areas, the weir heights, and so on and so on... When
we are designing a column (see Chapter 14), we can leave it to the program to determine all
of these parameters. However, when we are simulating an existing column, we must provide
this information. The additional detail also allows us to see the effects of changing equipment
design parameters. This is an important extension of the classical equilibrium stage model. The
nonequilibrium models also are applicable to multicomponent separations; they do not need
efficiencies or HETP’s which are not easily predicted for multicomponent mixtures. Finally,
ChemSep’s nonequilibrium models are quite easy to use. Let us illustrate that by using the
exampple of the previosu chapter.

2.2 A Sieve Tray Column

Using the same benzene-toluene example of the previous chapter we now want to simulate a
sieve tray column using the nonequilibrium column model. This column model is selected in
the operation menu of ChemSep.

Upon selection of the nonequilibrium model the column configuration is specified in the same
way as done for an equilibrium-stage column. However, instead of specifying the stage effi-
ciencies for the column, we need to specify the column layout design (already marked with a
red cross as the design specifications are incomplete). This design encompasses the type of
internals, such as trays or packing, but also the parameters of the column internals, such as the
sieve tray hole diameter and the column diameter.

Upon entering the design spreadsheet the ChemSep interface automatically finds the appropriate

54
 Figure 2.11: Selecting the nonequilibrium column model.

Figure 2.12: Column design.

sections in the column. Here, a column section means any part of the column between feeds
and/or product streams. In our simple example, there is a rectification section (stages 2-4) and a
stripping section (stages 5-9). For each section we need to select the type of internal (Fig. 2.13)
to be installed as well as the layout of the internal. The column layout design can in principle
be different for each stage in the column, however, in practice, the design is taken to be constant
within a section of the column. Sometimes, the same design is used for the whole column.

Here we will use sieve trays as column internal. Once we selected this type of internal we are
prompted for the layout parameters specific to this type of internal, see Figure ??. The most
important parameter in this list is the column diameter, however, all of the other parameters play
a role in determining the mass transfer and the hydraulic performance of the internal. Some of
the parameters are also interrelated, such as the various area’s and the column diameter, and
must be set in a consistent manner. Note that you can enter the parameters in any units, but that
they will always be converted to the units as selected in global units settings.

55
 Figure 2.13: Column internals type selection.

Now we come to an inherent problem of using nonequilibrium models: how do we know the
column diameter for the sieve tray so that the tray will operate satisfactorily? The trays will
weep when the flows are too small or flood when they are too large. To size the trays we first
need to know the internal vapor and liquid flowrates and have estimates of the physical prop-
erties such as the densities. However, these depend on the simulation results themselves: the
vapor density is dependent on the temperature and the composition of the vapor. The nonequi-
librium model allows us to simulate an existing column where we know the layout, a simulation
in rating mode, but how can we simulate a column for which we do not have a column layout?

ChemSep solves this dilemma through the built-in design mode. When we do not specify the
column diameter of a section (that is, we leave it as ”*”), the design mode is invoked automat-
ically and a built-in design method computes a layout for the selected type of internal ”on the
fly”, during the simulation. The interface shows per section which mode is selected next to
the section number: each the design or the rating mode. The user can specify start values for
certain layout parameters but the design methods are written in such a manner that they do not
require any user input. In other words, it is only necessary to specifie the type of internals for
each column section and ChemSep’s design mode will compute valid layouts for each section
while it solves the problem. This makes it very easy to use the nonequilibrium model and it can
also be used in flowsheet calculations where flowrates are not know a priori. The design mode
is explained in detail in Section 16.5.

As here we do not yet know anything about the physical design of our benzene-toluene column,
we will use the design mode for both sections and leave all the layout parameters as unspecified
(as ”*”).

Besides the specification of internals type and the actual layout parameters, we also need to
select the various models that determine how the performance of the internal is calculated.
These include correlations for the Mass Transfer Coefficients (MTC’s), liquid phase resistance,
vapor and liquid flow models, pressure drop, liquid entrainment, and holdups. In order to
facilitate the design mode a design method and parameters for this design method also need to
be selected.

ChemSep assigns defaults for these models depending on the type of internals that the user

56
 Figure 2.14: Sieve tray layout specification.

Figure 2.15: Sieve tray performance and design models selection.

selected. Usually this provides a selection that will ”work” and provide a converging solution.
In this case we see that the AIChE mass transfer model has been selected and that we include the
liquid phase resistance in the calculation of the mass transfer rates. The phases are considered
to be completely mixed, the pressure drop is fixed (that is not computed from the internal flows
and tray hydraulics), we do not correct for liquid entrainment, and the tray is designed on a
fraction of flood basis.

For sieve trays, we may choose from several different mass transfer coefficient correlations;
these are described in Chapter 16. For now, select the AIChE correlation. We recommend this
correlation for your first approach to the simulation for it is well behaved as long as the liquid
flows do not become very small. It is also a conservative und thus an inherently save method
for predicting the mass transfer efficiency of a tray.

For the vapor flow on the tray, we can choose to model the flows as completely mixed or in plug
flow. This choice of flow model has a rather large influence on the mass transfer performance
of a sieve tray. Plug flow significantly enhances the mass transfer on a sieve tray. Typically,
the vapor is assumed to flow in a plug flow manner through the froth. However, such a choice

57
 Figure 2.16: Mass transfer coefficient correlations for sieve trays.

Figure 2.17: Vapor flow model.

does not apply for the liquid flow on a tray. Liquid on small sieve trays (say less than 0.5m) is
most likely to be completely mixed, whereas the liquid flow over large trays approaches plug
flow. Therefore a dispersion model for the liquid phase is actually best but this approach is more
demanding computationally.

The plug flow model actually requires a simulation of infinite number of cells on the tray. This
can be approximated by averaging the incoming and outgoing vapor (y) or liquid (x) composi-
tions in the mass transfer rate calculations, or alternatively, with a theoretical model. ChemSep
provides a whole list of dispersion models that are typically combined with specific MTC mod-
els. For example, the AIChE model was developed using the Gautreaux-O’Connell dispersions
model. Care must be taken to make a sensible combination as ChemSep allows you to combine
any set of models!

Pressure drops can either be assumed fixed or calculated from the hydraulic operation of the
sieve tray. In the latter case the pressures on each stage no longer are fixed during the simulation
but they are allowed to change depending to the operating conditions of the trays. In this
example we choose to have the pressure drops estimated by the program.

Two different types of tray design methods are available for the sieve trays: a design based on a
specified fraction of flooding or a design method based on a specified maximum pressure drop.
Trayed columns often are designed on a specified fraction of flooding, a typical design value is

58
 Figure 2.18: Pressure drop model.

Figure 2.19: Selecting the design method.

75% fraction of flood (which is ChemSep’s default). This means that the internal flowrates may
increase by about 1/75% before the column would start to flood. Packed columns are typically
designed for a specified pressure drop, however.

In this example we will first simulate the column using most of the default model selections
of ChemSep. We therefore specify the stripping section also to contain sieve trays. It is not
necessary for all sections to have identical internals; we are free to combine trayed and packed
sections within one column. This might be useful to maintain the same diameter throughout
the whole column and thus lower the column cost. However, we do want to make use of a
new feature that the nonequilibrium model has over the regular equilibrium model: we want the
simulator to compute the actual pressure drop over the trays to see the pressure profile over the
column.

Finally, an overall system factor is used in the tray design of the column. This is a general safety
factor and it is typically associated with the foaming tendency of the mixture in the column. To
see some commonly used values, press F1 to obtain help on the selection of the system factor.
If we use the design mode, the smaller the system factor, the larger our column diameter will
be. Here, we choose a system factor of 1.0.

59
 Figure 2.20: Help on system factor.

The final design spreadsheet shows us in one glance the selected type of internals and all the
selected models and methods for all the sections (that can be displayed).

Figure 2.21: Final design specifications spreadsheet.

Now we have made a complete specification of the required input and we can return to the input
menu and use the solve option to check our input. We will see that there is an inconsistency
in the pressure drop specification: previously we had specified a fixed pressure in the whole
column but in the design specifications we selected to estimate the pressure drops for the sieve
trays. When we use the solve option, the ChemSep’s check on the input will prompt us to accept
or to correct this.

60
 Figure 2.22: Checking the input.

We can correct this by also selecting to use the estimated pressure drops in the column pressures
specification, see Figure 2.23.

Figure 2.23: Selecting estimated pressure drops.

Now we can try to solve our problem again and we will see that there are no more input prob-
lems (assuming you typed correctly all the input mentioned above). Now the nonequilibrium
simulator is called and the problem is solved. During the calculation we see that the design
mode is on and that after the initialisation of the internal flows the program designs the two
sections. The problem solves without there being a need to redesign the trays. The problem
rapidly converges in 4 iterations. If we look at the output tables we see an extra table for the
internals layout as calculated by the selected design method, see Figure 2.24.

In this case we see that the tray designs for both sections are close in diameter and that in
principle we could use just one the same diameters. The actual difference for the top and bottom
section is reflected in the downcomer area which is due to the large difference in liquid flowrates
between the sections. Note that the tray designs have equal tray spacing, this is because the tray
design starts with using certain default settings for many tray design parameters. You find these
under tools — Design mode settings, see Figure 2.25. For a definition of these defaults, see
Section ??.

The user can manipulate the tray design by initializing some of the design parameters for the
specific trays in the design window. In this case these values will be used and only when
necessary they will be overwritten by the design method. Alternatively you could change the
defaults in the tdesign.def file but then this will be reflecting all designs. Of course, the tray
design the simulator makes is not an optimal design but should be seen as a starting point

61
 Figure 2.24: Internal design table showing the column layout for the sieve trays.

Figure 2.25: Design mode settings from the standard tdesign.def file.

for a feasible tray design. Further optimization for tray flexibility, minimum pressure drop or
maximum mass transfer efficiency is most of the time possible.

Using these tray designs, the nonequilibrium model calculated the mass transfer rates and all the
column variables. We may backcalculate the Murphree component efficiencies for each stage.
Figure 2.26 shows the efficiencies where the mass transfer rates were computed with the AIChE
MTC model and mixed vapor and liquid flow models.

62
 Figure 2.26: Predicted Murphree component stage efficiencies.

We see that the Murphree efficiencies for benzene and toluene are equal at about a value of
0.5. The resulting McCabe-Thiele diagram is therefore quite similar to the one we obtained by
using the equilibrium model with fixed efficiencies, Figure ??. Actually, normal efficiencies for
a column like this are higher. We need to use the plug flow models to properly simulate this
column, bringing the efficiencies to 61-73%.

Sieve trays have a limited operation range and the calculated fraction of flood (FF) and weeping
fraction (WF) are also computed during the simulation. These are plotted in the operating limits
plot, one of the extra available plots when we do a nonequilibrium simulation. This plot shows
the actual operating range by plotting WF and the inverse of FF. When the weeping line is larger
than one the trays weep, when the flooding line is smaller than one the trays are flooded.

The pressure profile for the sieve tray column is shown in Figure 2.29 where we see that the
pressure drop over the 8 trays is about 0.05 bar or about 6 mbar/tray. Note that the pressure
is constant between the condenser and the column. Normally, this is not the case as there
is a substantial amount of piping between the top of the column and the condenser, with an
associated pressure drop.

For the design of trayed columns the actual vapor loading and the flow parameter velue are
important.

63
 Figure 2.27: McCabe-Thiele diagram for the sieve tray column.

Mass transfer in the column is computed from number of transfer units for both the vapor and
liquid phase.

2.3 A Depropanizer

The ”depropanizer” is a high pressure column in an oil refinery (Figure 2.32). For our example
the feed contains four components: pentane, butane, propane and ethane (in a real column
there would be more, but this is enough to show most of the behavior of this column). These
components form relatively ideal solutions and the heat effects will be small. The column is to
split the feed between butane and propane.

We will use the Peng Robinson equation of state to describe the phase equilibria and use ‘Ex-
cess’ enthalpies. We choose the top temperature at 40o C (to allow cooling with ambient air).
With the ‘Flash’ unit of ChemSep, we find that we need a pressure of 22 bar. With a total con-
denser, 28 sieve trays, a partial reboiler and a boiling point feed on stage 16, combined with a
reflux ratio of 2, we obtain the desired separation (This is not necessarily an optimal design, but

64
 Figure 2.28: Operating limits for the sieve trays.

we may change it later). It takes about two minutes to get the required data into the program.
Using the ‘design mode’ the program sizes the column and we need only specify the type of
internal used, in this case sieve trays with a tray spacing of 0.5 m. We specify the design param-
eters at 60% fraction of flood and 50% fraction of weeping to obtain a more flexible design. We
have used the AIChE MTC model using plug flow for the vapor and mixed flow for the liquid.
The run converges without difficulty. Output graphs show flow rates (Figure 2.33) which are
almost constant in the two sections, except for a sharp increase of the liquid flow at the feed
tray.

The flows of the components vary strongly along the column (Figure 2.34), with ethane and
propane being concentrated in the top and the other components in the bottom. The two minor
components (ethane and pentane) have constant flow rates in the greater part of the two sections,
the rates only changing rapidly at the ends of the column. The propane flow generally decreases
as we pass down the column, while butane increases. There are however several irregularities.
These are mainly because propane is displaced by the minor components in certain parts of the
column. Similar remarks can be made on butane. We see that the trays above the feed stage do
not provide much separation between butane and propane; this is an indication that we should
put the feed on a higher tray.

The transfer rates of the components on the different trays differ wildly (Figure 2.35), showing
how complicated such a system is. In the greater part of the column the transfer rates of ethane

65
 Figure 2.29: Pressure profile for the sieve tray column.

and propane are negative, which means that they are moving into the vapor. The opposite holds
for butane and pentane.

We can get out any amount of detailed information on the tray layout in the column, see Fig-
ure 2.36). You can immediately find out what are the effects of modifying the design (for exam-
ple the tray height, or the free areas) by running the program again. Or you can check whether
all trays are operating within their hydraulic limits, and how this changes if you modify the
design. This is one of the more interesting aspects of nonequilibrium models.

The nonequilibrium model does not use efficiencies. However, from its results, we can calculate
the efficiencies of each component on each tray. These are shown in Figure 2.37. Have a close
look at these efficiencies. The first thing to be noted is that all four efficiencies are different. The
second things is that they vary wildly along the column: from negative (!) to values far larger
than the 100% maximum expected for a tray with a mixed liquid. Remember that these are
results for a simple, almost ideal mixture; they are typical of any multicomponent separation.
Most of the efficiencies are between 50 and 95% though and those of the two key components
propane and n-butane are around 80%.

Our experience is, that many engineers have difficulty in believing these results. They know
from experience that binary efficiencies are fairly constant and well behaved. So we have
added one calculation for a similar column, but now with a binary feed of only propane and

66
 Figure 2.30: Flow parameter and vapor loading of the trays in the column.

butane (Figure 2.38). Here, indeed, we see that the efficiencies of the two components are
equal, and that they have a more or less constant value of about 80%. For columns with large
flow path lengths (over 1 m), we need to model the liquid flow with the plug flow model. If we
do this for our depropanizer we observe that the component efficiencies lie closer to eachother
and are somewhat higher, on average 90%. The backcalculated overall efficiencies of actual
depropanizers is around 80-90%.

From this example we see that experience with binary mixtures does not translate to multi-
component mixtures in a straightforward manner. For a binary, an efficiency may be a useful
way of summarizing experience. However in multicomponent mixtures, efficiencies are hardly
predictable. They are different for all components and may vary from minus infinity, through
zero, up to plus infinity. In our experience, efficiencies in multicomponent separations can be
thoroughly confusing.

There are many other aspects of binary distillation, that do not translate well into multicompo-
nent mixtures. Figure 2.39 shows the McCabe-Thiele diagram of the two key components (also
provided by ChemSep). At first, it looks much like that of a binary mixture. A closer inspection
shows that the operating lines are strongly curved. In this example they even run through a small
loop near the feed stage. There are other differences between this example and a proper binary.
Here, although the feed is at its boiling point, the feed line is not vertical. A close look at the

67
Figure 2.31: Number of Transfer Units for the vapor (red) and liquid (green, the higher
values).

Figure 2.32: A depropanizer.

diagram shows that the ‘stairs’ do not extend up to the equilibrium line. This is understandable;
we are dealing with a nonequilibrium model. However, we have seen that the efficiencies of the

68
 Figure 2.33: Flow rates in the depropanizer.

Figure 2.34: Compositions of the components in the depropanizer.

propane and butane are not equal, so the height of each step cannot be a direct measure of the
vapor efficiency as it is in a binary mixture.

2.4 A Packed Absorber

Our next example (Figure 2.40) seems to be a simple problem, suitable for an introductory
lecture in a separation processes course. We are to absorb acetone (10 mol/s) from a stream
of nitrogen (100 mol/s) at 80o C so that the leaving vapor mole fraction of acetone must be

69
 Figure 2.35: Mass transfer rates of the components on each stage.

Figure 2.36: The tray layout of the depropanizer.

below 0.1%. The solvent is 880 mol/s water at 30o C, and the 10 m high column is filled with
Raschig rings (as it should be in any good, old fashioned exercise).

For this exercise we use the DECHEMA K-model using the NRTL activity coefficient model,
Antoine vapor pressures, and the ‘Excess’ enthalpy model. The NRTL parameters for the
acetone-water binary are in the ChemSep databank. The parameters for the binaries involv-
ing nitrogen must be set to a large value (104 ) so that no nitrogen disolves (with αij = 0.3).
The Onda correlations are used for the mass transfer coefficients. Both phases were taken to be

70
 Figure 2.37: Murphree vapor efficiencies of the components in the depropanizer.

Figure 2.38: With a binary mixture, efficiencies are equal and well behaved.

in plug flow and the column consisted of 30 elements, each of one third of a meter. We obtained
the water flowrate by starting at a large value (say ten times the vapor flow) and then reducing
it until our specification was met. This resulted into a column of 1.5 m in diameter (at 75%
fraction of flood).

Figure 2.41 shows the mass transfer rates in the different sections. We see that acetone absorbs
throughout the column and that nitrogen (which is hardly soluble in water) absorbs so little that
this is not visible in the plot. The most spectacular component is water. Water condenses in
the upper part of the column, but evaporates with high rates in the bottom part of the column.
This means that there is a large internal water recycle in the column, which carries a substantial
amount of heat. All this shows up in the temperature profile (Figure 2.42). We see notable

71
 Figure 2.39: The pseudo-binary McCabe-Thiele diagram of the depropanizer.

Figure 2.40: The Acetone absorber.

differences in temperature between gas and liquid, as is to be expected. We could have also
plotted the interface temperature but it is so close to the liquid temperature that it would be
hard to distinguish the two. Actually, the interface temperature is higher than the bulk liquid
temperature in the top of the column, but lower in the bottom. This means that the direction
of heat transfer changes (at just about the position of the temperature bulge). The situation is
obviously far more complicated than the simple textbook problem that we had in mind!

The nonequilibrium model does not use HETP’s. However, from its results, we can calculate
the HETP’s of each component in each element, these are shown in Figure 2.43. The first thing
to be noted is that all three HETP’s are different. The second is, that they vary wildly along the
column. Nitrogen has an extreme value that is negative (!) and water an extremely large value,
+78 m, in the top of the columne (you may argue that you are not interested in the HETP of
nitrogen, but the point is that an HETP can be negative). Even the HETP of acetone varies from

72
 Figure 2.41: Mass transfer rates of the components in the absorber.

Figure 2.42: Temperature profiles in the absorber.

0.6 to 0.8 m. Remember that these are results for what you thought was a simple introductory
example.

From this example we see that experience with binary mixtures does not translate to multi-
component mixtures in a straightforward manner. For a binary, an HETP may be a useful way
of summarizing experience. However in multicomponent mixtures, HETP’s are different for
all components and may vary from minus infinity, through zero, up to plus infinity. In our
experience, HETP’s in multicomponent separations can be thoroughly confusing.

73
 Figure 2.43: HETP’s of the components in the absorber.

2.5 Extractive Distillation

The last example considers the simulation of a strongly nonideal mixture with varying internal
flowratios. Methyl-cyclohexane (MCH) and toluene have close boiling points; they cannot be
separated by conventional distillation. However, they do differ in polarity, MCH being the least
polar of the two. We separate them by adding a large excess of a polar, nonvolatile solvent (here
phenol) to the liquid. (Figure 2.44) shows this extractive distillation column. Phenol is added
on a few trays below the top of the column. The activity of MCH in phenol is larger than that
of toluene, so it becomes the most volatile and ends up in the top of the column.

In this problem the UNIFAC model is used to describe the (strongly nonideal) vapor liquid equi-
libria. As column internals we use bubble cap trays and the AIChE method for the mass transfer
coefficients. Also here, we need to start with an excess of solvent to obtain convergence after
which the amount of solvent is reduced until the profiles shown in Figure 2.45 were obtained.
This is not an optimal design, but it can be improved. Large nonidealities do not pose any ex-
ceptional problems to the nonequilibrium model. For this column the tray efficiencies are all
over the place.

2.6 Exercises

2.6.1 Separating Four n-Alkanes in One Column

This example came from Henley and Seader (exercise 15.2, 1981) and involves the separation of
four components into products of at least 96% purity in one column, using multiple sidestreams.

74
Figure 2.44: Extractive distillation of methylcyclohexane (MCH) and toluene using
phenol as solvent.

Figure 2.45: Compositions of the components in the extractive distillation.

A feed of n-C4 (14.08 lbmol/h), n-C5 (19.53), n-C6 (24.78), and n-C8 (39.94 lbmol/h) enters
a column at 25 psia and 150 o F . The columns condenser operates at 20 psia and we let
the nonequilibrium model estimate the pressures in the column. Use the Peng-Robinson EOS
for the thermodynamics and the complete enthalpy model. Design a column configuration
operating at 75% of flood using Sieve trays. How do you specify the column? Why? What
happens with the component efficiencies? Could this column be modelled with an overall stage
efficiency? (Hint: sidestreams do not need to be only liquid streams).

75
Do the sections need to be of different design or could one general column layout suffice? Try
using different internals in various sections: switch to valve trays, 1 inch metal Pall rings, and
Sulzer BX structured packing (you can do this within one simulation!). Compare and comment
on the operating regions and the resulting column diameters you find for these internals. Are
they realistic? Check the influence of using plug flow models and try out different MTC and
pressure drop models. Put your results in a table to make comparisons.

2.6.2 Azeotropic Distillation

In this example we simulate the separation of benzene (boiling at 80.2o C) and cyclohexane
(80.8o C), which form an azeotrope which boils at 77.4o C. Acetone (56.4o C) is used as en-
trainer. Acetone does not form an azeotrope with benzene, but does so with cyclohexane that
boils at 53.1o C (thus, a minimum boiling azeotrope which will go over the top). The acetone
and the pure cyclohexane are recovered from this azeotrope with a water wash, followed by a
simple distillation that produces water and acetone for recycling. Figure 2.46 shows the whole
process (see also Smith, 1963, p. 405).

Azeotropic Distillation
cyclohexane

water
makeup
Water-wash
(extraction)

feed

Azeotropic
Distillation

Distillation

benzene acetone
makeup

Figure 2.46: Azeotropic distillation of benzene and cyclohexane with acetone as en-
trainer, and a water wash for entrainer recovery.

Here we will only look at the azeotropic column. Since the acetone is recovered by means of a
water wash there will be some water (3 mol%) and cyclohexane (2 mol%) in the recycle feed
to the azeotropic column as well. The recycled acetone is fed to the same stage as the equimolar
feed of 10 mol/s cyclohexane and 10 mol/s benzene (simulations with the nonequilibrium

76
require actual feedflows as actual tray performances must be calculated). The flowrate of the
acetone entrainer is varied to obtain the optimum separation, in this case it is set at 35 mol/s.
The column is simulated with 38 sieve trays, a total condenser and partial reboiler, with the feed
entering on the 14th tray (counting from the top). To model the thermodynamic equilibrium the
DECHEMA – UNIFAC model was used and for the mass transfer the AIChE model. No column
layout parameters were specified, these were calculated by the program for a column to operate
at 75 % of flooding. The reflux ratio was set at 4 and the bottoms flowrate was specified to be
10 mol/s.

A 99.5 mol% pure benzene bottom product is obtained and a top product containing mainly
cyclohexane and acetone with a benzene content that can result in cyclohexane purity of 99
mol% or better. The column was split into two sections, one above the combined feed tray and
the other one from the feed tray to the last tray in the column. Answer the following questions:

1. Look at the resulting tray layouts. Do we need different designs in the two sections?
2. Check at the Benzene concentration profile. Explain what happens in the top section of
the column
3. Plot the component efficiencies. Explain the behavior of the component efficiency of
Acetone. Could this column also be simulated using the equilibrium model with a con-
stant efficiency? Explain.
4. Inspect the McCabe-Thiele plot, can you improve the column configuration? Try!
5. A second steady-state solution was found (with the same specifications) where the bot-
tom product contains a lot of cyclohexane and the acetone concentrations in the middle
of the column are much lower. Can you find both the two steady states?

2.7 Model Uncertaincies

The nonequilibrium model improves our description of separation processes. This does not
mean that there are no problems left to be solved in distillation. Complicated mathematical
models do not solve problems; they may only improve your solution. The results are only as
good as the important parts of the model. During development of nonequilibrium models it
has become clear that our knowledge of the details of what happens inside a column remains
incomplete. There are several areas where current models and correlations (at least those in the
open literature) can be improved:

• correlations for interfacial areas,

• correlations for mass and heat transfer coefficients,
• minimum load correlations, and
• liquid flow models and the description of maldistribution in packed columns.

As distillation research focuses on these area’s the nonequilibrium model will benefit from the
model improvements and users will obtain more accurate simulations results.

77
References

E.J. Henley, J.D. Seader, Equilibrium Stage Separation Operations in Chemical Engineering.
Wiley, New York (1981).

B.D. Smith, Design of Equilibrium Staged Processes, McGraw-Hill, New York (1963)

R. Taylor and R. Krishna, Multicomponent Mass Transfer, John Wiley and Sons, New York
1993.

J.A. Wesselingh and R. Krishna, Mass Transfer, Ellis Horwood, Chichester 1990.

78
 Chapter 3

CAPE-OPEN Flowsheet
Simulations with ChemSep

ChemSep is fully CAPE-OPEN compliant and can be used as a ”plug-in” in all compliant flow-
sheeting packages. This chapter illustrates the use of ChemSep column simulations in the
COCO Simulator (www.cocosimulator.org) that is available for free. The chapter starts with
a small introduction of COCO and then discusses examples ranging from simple separation
trains up to the simulation of complete chemical plants.

3.1 Using the COCO Simulator

When you open the COCO simulator via the start menu you’ll find an empty flowsheet loaded
in the GUI, called COFE. The GUI has the usual menu entries (File, Edit, Window, Help) of
multiple document application in MS windows. The other items are:

Insert : to add a new stream, unit operation, or object (line, text, bitmap) to the flowsheet
Flowsheet : to configure, validate, and solve a flowsheet, edit a unit operation or to make reports
Plot : to generate physical property plots or plot results of a specific unit operation
View : to set aspects of the graphical representation of the flowsheet, i.e. the zoom level, grids,
stream and unit labels, etc.

The COCO window is split over a menu and toolbar, an area for the flowsheet graphs, and a log
screen displayed on the bottom over the status line. Multiple flowsheets can be loaded simul-
taneously and these are completely separate idententies. COCO comes with its own packages

79
for thermodynamics and physical properties, called TEA (derived from the ChemSep code and
Pure Component Data PCD files); standard unit operations, COUSCOUS (which include mix-
ers, splitters, heat exchangers, pumps, etc.); and reactions, CORN. Let us now start to setup a
simple Benzene/Toluene/p-Xylene (BTX) separation train in COCO using ChemSep columns.

Figure 3.1: The COCO Simulator with a new stream.

Our goal is to separate an equimolar stream of BTX of 80 tonnes/day (25 degrees C, 1 atm) into
99% pure Benzene, Toluene, and p-Xylene. Start COCO and maximize the default flowsheet
it opens with. We then insert a stream by selecting the menu ”Insert-Stream” (or Ctrl-I) and
clicking on the flowsheet area once, then draging the cursor a bit to the right and clicking again,
see Figure 3.1.

Note that COCO issued a warning, as can be seen in the flowsheet log: the stream we added
does not have any material assigned to it. So before we go on we must configure the materials as
well as thermodynamic and physical property model that will be used to describe the materials
in the flowsheet. Press Ctrl-K or go to the Flowsheet menu and select the Configure option,
this will bring up the flowsheet configuration window, Figure 3.2.

You’ll see that there are many parts of the flowsheet to configure but we just need to fill in two:
Compounds and Property packs. Select the Property packs tab (default for a new flowsheet) and
click on the Add button that appears to the right. This brings up a window where you can select
the type and version of property packages that are available on the current machine. If you

80
 Figure 3.2: Empty configuration window.

just installed COCO (and no other CAPE-OPEN compliant software), you only get a selection
between the two different versions of the package that comes along with COCO, namely TEA,
Figure 3.3.

Figure 3.3: Property package selection window.

If you installed ASPEN PLUS, PRO/II, or other CAPE-OPEN software more packages will
be listed. This is very useful when flowsheets are re-used: once a property package is de-
fined, it can be used in other flowsheets, even in flowsheets of other software packages as long
as the CAPE-OPEN standards are followed. For the present purposes we will use TEA as it
comes with COCO. Both versions listed will work (for more information on versions of the
thermo/properties standard consult the CAPE OPEN web site, www.colan.org). For now click
OK to chose ”TEA (CAPE-OPEN 1.0)” which will bring up another window, Figure 3.4.

81
 Figure 3.4: Property package selection window.

The left part of this window shows a list of CAPE OPEN property packages that were created
earlier. None of these packages are useful to us, so we must first create a new package. Click
on the Edit pack button that appears towards the upper right of this window. This will bring up
the TEA window with previously defined packages, Figure 3.5.

Figure 3.5: TEA Thermo window.

We want to define a new template, so click on Create template button, on the upper right.
Finally, here we are able to actually define compounds and property models to compute physical
equilibria. Type in something that describes the compounds and models for the name of the
property pack template, Figure 3.6.

Use a name that helps you to identify the template for later use and also for use by others. Be
aware that templates are machine specific, i.e. they are stored in a general area where any user
can access all templates defined. This is a strong point for sharing information but could be
a risk for confidential projects. To add actual compounds select the compounds tab. The list
will be empty. Click on the Add button to select components for the new property template,
Figure 3.7. COCO shows the compounds with their overall formula, mole weight, and CAS
numbers (so you can differentiate between isomers). You can pick compounds from this list by
double clicking or you can search all the components with a benzene ring by typing ”benzene”
in the filter edit box on the bottom. Besides Benzene you also need to select ”Toluene” and

82
 Figure 3.6: Define template for the BTX property package.

”p-Xylene” as compounds Figure 3.8. Note that these are read from a ChemSep PCD file called
”DefaultComponentLib.pcd” that is part of the COCO distribution.

Figure 3.7: Component selection.

To complete the BTX template we need to choose the models that will be used to estimate ther-
modynamic and physical properties. Click on the Property Calculations tab. COCO shows
there has not been any model selection made yet, as indicated by the ”N/A” for the three prop-
erties that are present by default, Figure 3.9. Note that TEA allows us to exactly define specific

83
 Figure 3.8: Selected BTX compounds.

sets of properties.

Figure 3.9: Empty model selection.

For a ChemSep simulation we will require property models for the (log) fugacity coefficients,
enthalpy, entropy, heat capacity, density, viscosity, thermal conductivity, and surface tension.
The fugacity coefficients may be computed with a variety of other models. Here we will want
to use an activity coefficient model with a vapor pressure model. Use the Add property button
to add the ”Activity coefficient” property to the template, Figure 3.10.

84
 Figure 3.10: Property selection.

Note that adding the activity coefficient model automatically adds the vapor pressure model.
Similarly, if one select ”surface tension” COCO also adds a surface tension property for pure
compounds. Go ahead and add all the required properties mentioned above that TEA must
support in order to be able to solve a column in ChemSep, Figure 3.11.

Figure 3.11: The required set of properties for ChemSep (activity coefficient model).

Note that in Figure 3.11 we see COCO has preselected certain models for the viscosity for
example. It also assigned the ”LeeKessler” method for the vapor pressure. However, we want
to use the Antoine model for the vapor pressures. To change this click the selected model to get
a list of the available methods, Figure 3.12.

In order to use the property package we must make a complete selection of methods for each of
the properties we selected to be part of the template. For the activity coefficients we will make
use of the UNIFAC-VL method which is a predictive method that does not require further input.
Use Figure 3.13 to fully define the property package we will use for the BTX example. Note

85
 Figure 3.12: Vapor pressure methods.

that some of the properties still have ”N/A” as they are not applicable to the phase in question.
Note that by default TEA is making a template with vapour and liquid phases. If you would
want to have other phases present, the phase definitions need to be adapted. We won’t need that
for this example.

Figure 3.13: Property methods for the BTX example (UNIFAC+Antoine).

86
Once we finished the property template for the BTX example, click the OK button and we will
see the new template is added to the available set of templates. Close the window displaying
these templates and the window displaying the avalable property packs is also updated and
includes our new BTX package. In other words, making a template doesn’t necessarily add the
template to the Flowsheet. Thus we must select the BTX package and click the OK button to
acually add this package to the flowsheet. COCO will also confirm whether this package is to
be made the default template. Click Yes, Figure 3.14 and close the configuration window.

Figure 3.14: Selection of the BTX property package.

When we now double click on the stream ”1” we previously added to the flowsheet, we see that
COCO indeed displays a definition with the three compounds we defined to be part of the BTX
property package.

We now want to define our feed. We would like to do this in convenient units, therefore click
on the units string for the pressure to display the available units of measure for pressure. Switch
from Pascal to Atmosphere (atm), Figure 3.15.

Similarly, define the temperature to degrees Celcius and the flow to tonnes/day and then finish
the feed definintion for our BTX example, Figure 3.16. Once the stream is fully defined COCO
will automatically compute the stream and the properties that were defined in the property
package. Note that it will just show the properties that are turned on for display in the Properties
tab of the flowsheet configuration. Take some time to browse through all the properties. Note

87
 Figure 3.15: Define units of measure for pressure.

that at 1 atmosphere and 25 Celcius the feed is a liquid. You can select some more convenient
units for the density for example, use kg/m3 . When done, close the window. Notice the color
of the feed has changed from brown to black. This means the feed is now fully defined and you
can solve the flowsheet (the menu button is now enabled).

Figure 3.16: Define BTX feed.

Solve the flowsheet by pressing F5 and notice how the feed changes color from black to green.
All units that are solved are colored green. An unit or stream for which solving resulted in an
error turns red. If the flowsheet does not converge, all items in it will be black, as in specification
complete. A non-defined stream is colored brown.

88
 Figure 3.17: Solved flowsheet with BTX feed.

Once the flowsheet is solved COCO will display the stream status when you hover with the
mouse over it, see Figure 3.17. We now add a distillation column by pressing Ctrl-U (or select
Unit Operation from the Insert menu). You can select from a list of registered CAPE-OPEN
unit operations available on the current machine, see Figure 3.18. You can select a ChemSep
distillation column by opening the list of separator unit operations and selecting ChemSepUO.

You will get a confirmation that the new unit operation is added without any ports hooked up
to it. Then a default distillation icon will be shown on the flowsheet. It’s displayed in red and
you can move it to any place that is safe on the flowsheet. After you placed the column on the
flowsheet with a click, the color changes to brown, indicating its specification is incomplete. To
make specifications for this distillation column, double click it. That opens the unit operation
(UO) in the generic editor of COCO, Figure 3.18. To directly open the GUI of the unit operation
itself instead of COCO’s generic editor, hold the Alt-key while double clicking the column.

This is a generic editor and only shows the variables that the operation exports. These will
be stored in a persistently manner when you save and open flowsheets. In our column case
these include flags as to which type of flashes to use on feeds and products, or how to log
CAPE-OPEN calls. To specify the column we want to launch the GUI of ChemSep. To do
this, click the Show GUI button. Currently, ChemSep uses a wrapper for CAPE-OPEN, called
ChemSepUO.DLL. It is the GUI of the wrapper which is launched by COCO, see Figure 3.19.

89
 Figure 3.18: Inserting a new unit operation.

The options available in this wrapper are explained later. All you need to do for now is to click
the ChemSep Interface button to launch the ChemSep GUI in CAPE-OPEN modus.

Figure 3.19: Editing a Unit Operation (UO) in COCO.

When the ChemSep GUI opens in CAPE-OPEN mode several features are hidden or disabled.
We continue with setting up our BTX column as follows:

• Select metric units for this problem

• Use a simple distillation column configuration with 10 stages and a feed to stage 5
• Specify a constant pressure of 1 atm in the column
• There are no column heat duty or heaters/coolers and a stage efficienycy of unity
• Specify a Reflux Ratio (RR) of two and the fraction of feed flow recovered in the bottom
equal to the expected components; i.e. 2/3

90
 Figure 3.20: Setting simple distillation column configuration.

Once the problem is setup, exit the GUI and save the column specifications. As we now have
defined that the column has a feed and two products, we can also start connecting them in the
flowsheet. Add our feed to the column and add top and bottom streams. You’ll notice that the
color of the column changes from brown to black when you do this. COCO noticed that the
problem is specified and enabled also the solve button. You can now solve the BTX column.
Do so by clicking the solve button.

This makes COCO call the wrapper which in turn calls ChemSep’s simulators. During the
time that the column simulator actually solves the column, the icon is displayed with a orange
background. Once the BTX column is solved, it is displayed as green. If there are warnings
there will be a warning label attached where you can click on to see what was wrong. If solving
failed, the column will be shown in red.

After the column has been solved, go an edit it again by double clicking while holding the Alt-
key. Note that this skips COCO’s editor and jumps straight to the GUI of the ChemSep wrapper.
The simulator’s output is shown in the status window, Figure 3.22, in green when the column is
solved.

Now the column is solved you can click the status tab and look at the results report. This
will give a summary of all the tabular output that can be seen in the ChemSep GUI. The same

91
 Figure 3.21: Connecting top and bottom product streams.

Figure 3.22: Solved unit operations with last output.

information can be inserted as the report of a unit operation on the flowsheet. Similarly, you
can add stream tables for all the streams in a converged flowsheet.

92
Add a stream table now by selecting Insert-Stream report which shows the mole fractions of
the three components in our BTX example. We can even edit the table by double clicking it and
adding the density as property to report to the stream table, see Figure 3.23.

Figure 3.23: Flowsheet with densities.

Notice that the separation isn’t good enough to produce 99% pure Benzene as top product. To
reach this purity we can add more stages in the column, here 20 stages (with the feed in the
middle) suffices to reach the required purity. Of course, we could have also increased the reflux
ratio we specified. These are process considerations, of course, since more stages increases the
height of the column whereas more reflux increases its diameter and energy usage. To determine
these aspects you must rate the column (e.g. by means of the Rating tab in the ChemSep GUI).

Now we have successfully separated Benzene from the ternary BTX, we still need to separate
the bottoms of this column into pure Toluene and p-Xylene. To do so we must add a second
column which will take as a feed the bottoms of the first column. Go through the same steps
as above but while specifying the second column use more stages (30) and a higher reflux ratio
(3). Connect the bottoms of the first column to this second column, and attach top and bottom
product streams. Resolve the flowsheet and add the products to our stream table. The result is
shown in Figure 3.24.

93
Note that in this separation train there was no recycle and thus the flowsheet does not need
to iterate to obtain the final solution. When you re-solved the first column you might also
notice a significant speed-up. The reason is that COCO remembers the old state of the column
and switches the column simulator automatically to a mode where it re-uses the old results as
initialization. This generally saves the user a lot of time but sometimes these old results can
lead to the convergence to a wrong solution. Therefore, the wrapper has a button to delete them
under the Options tab.

Figure 3.24: Separation train for BTX with two columns.

3.2 Flowsheet Examples

Here we discuss several examples that have been modelled with COCO and ChemSep. We
will not supply step by step setup as the previous BTX example but instead just discuss the
necessary details to specify and simulate the problem. The flowsheets are distributed on the
download page ChemSep, http://www.chemsep.org/downloads/. Note that here we
introduce the flowsheets using equilibrium column models. Of course, better descriptions of
actual column performance are obtained with the nonequilibrium column model. We show this

94
in a bit more detail with one particular example where we illustrate that ChemSep’s integrated
design method enables us to use the nonequilibrium column model while still changing the
flowsheet. The design method eliminates the need to know the column layout a priori: it is
continuously adapted while the column model is solved. In most of these examples ChemSep
solves the columns without needing specific guesses with respect to temperatures or flows: its
automatic initialization suffices. Note that while iterating to find the solution of the flowsheet,
old results are used as initialization. This speeds up the convergence when the flowsheet close
to being solved (but can sometimes result in a column not being solved).

3.2.1 Extractive Distillation of MCH and Toluene with Phenol (PH)

We need to separate an equimolar mixture of methylcyclohexane (MCH) and Toluene which

form an azeotrope together. We can still separate the two components by means of extractive
distillation using Phenol as the solvent. The flowsheet is shown in Figure 3.25.

Figure 3.25: Extractive Distillation.

In the fist column we introduce the 50/50mole% feed of MCH and Toluene. Above this feed
the solvent is introduced into the column on a stage close to the top. The feedstage for this
depends on the difference in boiling point between the solvent and that if the light component,
here MCH and the feed stage is set to stage 5.

will be seen that the process involves two distillation columns, a heat exchanger, and a make-
up stream. Valve trays are used for both the columns using the design mode nonequilibrium
model. The phenol recycle is cooled to 100o C. For a high purity of the products the solvent

95
feed to MCH/toluene feed ratio as well as the reflux ratio needs to be sufficiently high (for the
extractive column). We need the make-up unit to regulate the amount of phenol in the feed to
the first column.

3.2.2 Distillation with a Heterogeneous Azeotrope (BW)

Water and n-butanol form an azeotrope and cannot be seaparated by conventional distillation.
However, at not too high temperature, the liquid mixture splits into two liquid phases; an aque-
ous phase with very little butanol and a butanol phase containing a substantial amount of water.
We can use a decanter to separate the two liquid phases. A feed that is predominantly water
but contaminated with butanol (1 mole%) can be separated into two pure products using such
a decanter. The flowsheet is shown in Figure ??. The first column produces a bottom product
that is water, and the vapor overhead goes to a condenser/cooler and then to the decanter. There
we obtain water- and a butanol-rich streams that get recycled to the columns. In a second distil-
lation column we produce butanol as the butanol-rich recycle contains much more butanol than
the butanol-water azeotrope. Note that we use flash units to simulate the coolers of the vapor
overheads of the columns (which, therefore, do not have any condensers). The liquid-liquid
decanter is simulated using a single equilibrium-stage extraction unit. Since we need to esti-
mate the water and butanol-rich recycle flows, it is best to simulate these columns first with a
condenser. The solutions so obtained can be used as a basis for estimating the recycle flows.

3.2.3 Distillation of a Pressure Sensitive Azeotrope (MA)

Methanol and acetone form a minimum temperature azeotrope but the composition of this
azeotrope is sensitive to the pressure. We can make use of this to separate the two compo-
nents into pure products by operating two columns at different pressures.

here show an example of the azeotropes of methanol-acetone in two overlapping Txy diagrams
at two different pressures using chemsep

This type of azeotropic distillation is rare as the azeotrope composition needs to be sensitive
to the pressure in order to obtain a recycle stream that is not unreasonably large, and that the
pressures are such that no special equipment is required. The Methanol and Acetone system
is such a system and the separation shown in Figure 3.26 consists of two columns, the first
operating at 1 bar and the second at 6 bar. The recycle flow is 59% of the feed flowrate (when
the feed is equimolar). Note that the feed coming from the second column at higher pressure
can be fed directly back to the first column as ChemSep takes care of flashing the feed at 1 bar.
In the actual process this stream would be pumped and flashing would occur over the feed flow
control valve.

96
 Figure 3.26: Azeotropic Distillation at Two Pressures.

3.2.4 Petyluk Columns (PETYLUK)

Three different species can be separated using two simple distillation columns (as long as there
are no azeotropes to complicate things). An alternative configuration that lowers the energy
consumption of a conventional separation train is to use the so-called Petlyuk column shown in
Figure ?? that involves only one condenser and reboiler.

The feed to be separated goes to the first column. Since the first column has no condenser or
reboiler it needs a vapor feed to the bottom and a liquid feed to the top. These streams are
provided by the second column from which they are, in fact, withdrawn as sidestreams. The
products of the first column go into the second column at the the same sidestream stages.

In the example that comes in the flowsheet directory of ChemSep we included a Petyluk column
combination for the separation of n-propanol, isobutanol, and n-butanol. The ChemSep model
involves just two columns; there are no extra flashes, make-up streams, or stream splitters. To
get this flowsheet to run requires that you first solve both the columns separately; this is not
easy. The second column needed a user supplied initialization information in order to obtain a
solution. Subsequent convergence of the flowsheet (or of this column by itself) can be obtained
more easily by using the old results as initialization.

3.2.5 Liquid-Liquid Extraction with Solvent Recovery

(BP)

Extraction can be used to separate aromatics from paraffins. This is a common type of sep-
aration in the refining of crude oil. Figure ?? shows a simplified process flowsheet in which

97
we separate benzene from n-pentane, using a liquid-liquid extraction column with sulfolane as
solvent. The extractor is a rotating disk contactor (RDC) operating at 50o C. The solvent is
recovered from the extract by simple distillation. The UNIQUAC model is used for the activity
coefficients in the extractor; parameters for this model are given in Table 3.1. The extraction
is sensitive to the temperature as can be seen from the interaction parameters; this fact can be
used to influence the separation. Note that sulfolane only melts at around 27o C and this limits
the temperature for the extraction.

Table 3.1: UNIQUAC LLE interaction parameters (K)

Components i-j Aij Aji

25o C:
pentane-benzene 295.38 -171.63
pentane-sulfolane 532.04 187.84
benzene-sulfolane 151.02 -36.08
50o C:
pentane-benzene 179.86 -95.70
pentane-sulfolane 375.93 247.77
benzene-sulfolane 131.51 -6.28

The RDC extraction column does not achieve such a good separation as ChemSep suggests,
largely because backflow in the RDC has been ignored. In practice, the feed would not be to
the bottom of the extractor but more towards the middle. The aromatics concentration would be
increased by recycling some of the distillate of the second column to the extractor.

3.2.6 The Separation of Air

Distillation has long been used for air separation. The three major components of air - nitrogen,
oxygen and argon - do not form any azeotropes and can be separated by simple distillation,
albeit carried out at cryogenic temperatures because these compounds boil at very low temper-
atures (77.5 K, 90.3 K, and 87.3 K respectively, at atmospheric pressure). In fact, the relative
volatility between oxygen and argon is close to unity, especially in the pure argon region. Cool-
ing is achieved through the Joule-Thomson effect: if we expand higher pressure air it cools
down and can be partially liquefied. To simulate this process we use the Peng Robinson EOS.
Interactions parameters can be obtained from DECHEMA where various data sets are at hand,
each with a different fit. Here we used the following binary parameters: N2-O2 = -0.0133,
N2-Ar = -0.0026, and O2-Ar = 0.0089. Although these parameters are small, they have a pro-
found effect on the design of air separation columns. For example the separation of Oxygen and
Argon is difficult due to the reduced relative volatility in the pure Argon region.

The cryogenic air separation process is a tight integration of heat exchangers and separation
columns which is completely driven by the compression of the air at the inlet of the unit. The

98
 Figure 3.27: Cryogenic Distillation of Air.

air inlet stream is cooled cooled to ambient temperatures with cooling water and further cooled
against product and waste streams leaving the plant until partially liquefied. The liquefied fra-
cion varies between 25 to 35 mol%) depending on cooling water temperatures and compressor
outlet pressure. This stream enters the bottom of a column that produces pure liquid nitro-
gen (impurities ¡ 1ppm) as overhead. The column has roughly 40 equilibrium stages including
condenser and operates at ”high” pressure. The cooling medium in the condenser is the oxy-
gen bottom product produced in a second column. Thus, this second column must operate at
a lower pressure to make the reboiler colder than the condenser of the high pressure column
(typically by 1 Kelvin only). Thus, the pressures of both columns are linked mainly by the
vapour pressures of Nitrogen and Oxygen and the temperature approach (heat exchanger size)
of the combined condenser/reboiler. In order for the oxygen to get out of the plant the bottom
of the low pressure column is typically around 1.3 bar which sets a pressure of at least 5.7 bar
for the high pressure column. This requires the inlet compressor to deliver air at slightly higher
pressures (here we chose 6.1 bar). These two columns share the same column shell to minimize
the temperature difference between the condensing nitrogen and evaporating oxygen.

The liquid bottom product of the High Pressure Column (HPC) is rich in oxygen. When this
Rich Liquid (RL) is reduced in pressure the Joule-Thomson (JT) effect causes this rich liquid to
cool further. The rich liquid is fed to the second Low Pressure Column (LPC). If Argon is also
produced this refrigeration is used to create reflux for a side-rectifier that separates Argon from

99
Oxygen. Because Argon is the intermediate boiler it builds up between the RL feed and the
oxygen produced in the reboiler. Argon concentration increases from 1mol% in the RL to up to
13 mol% at the vapor draw to the side rectifier. The side rectifier operates with a high reflux rate
to further distill the Argon. This liquid reflux of this side rectifier is returned at the same location
as the draw into the low pressure column. When the side rectifier is equiped with structured
packing enough equilibrium stages can be included to produce pure Argon (xO 2 < 1ppm)
while the condenser still operates above atmospheric to ensure there is no ingress of air through
leaks.

When no Argon is produced the refrigeration produced by the JT effect of both the liquid nitro-
gen and the RL can be suffice run the whole plant. However, while producing Argon additional
refrigeration will be needed. This can be obtained by expanding an additional flow of air in via
an expander, using the JT effect, that feeds directly into the low pressure column. To obtain a
more effective refrigeration this air is compressed in a second stage to a higher pressure, where
effectively the extra refrigeration is obtained via the cooling water (but at a higher temperature).
Of course, since this air is poorer in oxygen than the Rich Liquid (RL) fed to the Low Pres-
sure Column (LPC) is, it enters the LPC at a higher feed location. Important to realize is that
a certain part of the compressed air cannot be separated and has to leave the plant as a waste
stream. This waste stream leaves the low pressure column as a side vapor draw. Finally, the
low pressure column is refluxed with with liquid nitrogen from the high pressure column, after
having been flashed and cooled by the JT effect. The liquid oxygen bottoms is gasified and
heated against the incoming air, just as the pure gaseous nitrogen of the top of the LPC and the
waste air stream are. The products of the Air Separation Unit (ASU) are thus Gaseous Nitrogen
(GN2) and Oxygen (GO2), as well as liquid Argon (LAr).

With the application of structured packings in ASU’s (in 1980-1990) it became possible to
”pack” enough separation in the columns without incurring a too high a column pressure drop.
It enabled the production of high purity Argon without the need to use a reactive oxygen removal
from the crude Argon with Hydrogen. With 70 to 80 equilibrium stages in the LP column the
oxygen product purity ranges from 97.5 to 99.5mol%. For both Nitrogen and Argon purities
are typically 99.999% (impurities less than 1 ppm). The Argon column is equiped with 100 to
120 equilibrium stages (see also Comp. Chem. Eng., Vol. 30 (2006) pp. 1436-1446). The low
pressure high capacity structured packings first introduced in 1999 by Sulzer further eased the
design of these packed columns with over 200 combined equilbrium stages.

Cryogenic air separation by distillation involves tightly (thermally) linked cycles of material as
well as energy where the cost of heat integration is balanced against the cost of air compression
(heat integration shows itself in capital cost whereas compression cost is a production cost).
This is further complicated by the many two phase flows between the various (tall!) columns
and heat exchangers. Since pumping of pure Oxygen is something typically avoided and the
flow regime determines the pipe pressure drop controlling the flow regime during design is
very important. The tightness of the coupling of the HPC and LPC via the combined con-
denser/reboiler largely sets the operating pressures. With temperature differences in the heat
exchangers of just one or a few degrees Kelvin there is significant interaction such that any
disturbance directly affects one of the column controlled variables.

100
 Figure 3.28: Simplied ASU.

The ASU is simulated with an air feed at 30 degrees Celcius and 1 atmosphere. Air is taken to
consist of Nitrogen (N2), Oxygen (O2), and Argon (Ar) only (normally there is also moisture
in the air as well as traces of other gases such as Carbon dioxide and other Noble gases such
as Neon and Krypton). The gas composition is taken as 0.7812 N2, 0.2095 O2, and 0.0093
Ar. This Air is compressed to 6.1 bara in three stages using a compression ratio per stage of
1.9. The compressors are only 80% efficient and water coolers are needed to cool the inflowing
air to 30 degrees C. This air flow is then split into two: 85% is cooled to a temperature of
99.4 Kelvin against streams leaving the ASU where 31 mole% gets liquified. The other 15% is
cooled only to 140 K to be generating power and refrigeration in an expander. When this stream
is entering the low pressure column it has cooled down to about 103 K. The split-ratio of these
two streams is used to control the refrigeration in the ASU. To stabalize the plant the percentage
of liquification in the main air stream to the high pressure column is controlled. That percentage
is a direct function of the temperature as at that point the air still has the atmospheric condition
and the splitter can be regulated with by measuring the exit temperature directly.

The high pressure column is to produce liquid nitrogen that is used to provide reflux in the low
pressure column and hence it must as pure enough as the GN2 product needs to be. Hence
the impurity of Argon is used to control the reflux ratio. In the simulation we set the Argon
composition to 0.7ppm and the top pressure to 5.6 bar. In reality, the top pressure is controlled
by the reboiler temperature of the low pressure column. As we fix the composition of that

101
 Figure 3.29: ASU in COCO.

column needs to be 98.5 mole% Oxygen, the temperature in the reboiler is largely a function of
the top pressure and allowed pressure drop in that column. The top of the low pressure column
is specified at 1.2 bar since the GN2 overhead still needs to be able to pass through the heat
exchangers out the ASU. To obtain the desired oxygen purity we need 70-80 stages in the low
pressure column and with a typical pressure drop of 2 mbar per HETP we get a bottom pressure
of at least 1.35 bar. This allows for a comfortable 2.6 degrees temperature difference in teh
condenser/reboiler of the two columns.

In this example we’ll use a simplified process where the second compression stage and the
subcooler are omitted at the expense of lower recoveries. The flowsheet was simulated with one
tear stream: the returning liquid stream from the Argon column back to the low pressure column.
Normally, the Argon column bottoms liquid flows back to the low pressure column by means
of gravity. This is simulated by using a pump that raises the pressure of the bottoms liquid,
as otherwise COCO warns us that the bottoms liquid would flow against a pressure gradient
(the Argon column operates at a lower pressure than the low pressure column!). Also note that
there is a vapor and a liquid drawn from the condenser of the high pressure column. This is
done for control purposes and allows the HPC condenser duty to be matched by a controller to
LPC reboiler duty. The combined stream is fed to the top of the low pressure column. The rich

102
liquid bottom product of the high pressure column is flashed where the heat input is set equal
to that of the condenser duty of the Argon side rectifier. The gaseous product streams are all
heated up against the inlet air flow to recover heat. Heat leaks are added to the columns from the
environment as well as pressure drops in the columns over the column internals, which raises
the required inlet pressure for the ASU.

Here the three heat integrations in the ASU is shown: the cooling of the feed against the product
streams in the main heat exchanger, the LP condenser against the HP reboiler, and the conden-
sation of Argon against evaporization of rich liquid.

Easy McCabe-Thiele plots let the user quickly evaluate the process and the feed and draw
locations to/from the low pressure column.

3.3 Di-Ethyl Ether from Ethanol-Water Feed

We consider the production of diethylether from an ethanol-water (85%-15%) feed, as shown

in Figure ??. The input file for this flowsheet, ep.fs, can be found in the fs directory of the
ChemSep distribution. Use the load option to load this file, after which you can peruse it using
the edit option.

In this process ethanol is converted to diethylether in the Reactor which has a conversion of
50%. The reaction is:
2C2 H5 OH → (C2 H5 )2 O + H2 O (3.1)
Diethylether (99.5 mol%), is recovered in a distillation column Sep-1. The ethanol is recovered
in a second distillation column and recycled Sep-2. We also obtain water (containing 1 mol%
ethanol) as a byproduct.

The flowsheet input file defines the unit operations and streams in the flowsheet and consists of
four parts. The first part consists of one section where all the components, units, streams, feeds,
and stream estimates are declared. It also contains the output filename, executable directory, and
the method and accuracy used in solving the flowsheet. The section starts with the [Flowsheet]
identifier.

[Flowsheet]
Comment=Ethylether Production
Components=Water,Ethanol,Diethylether
Units=Mixer,Reactor,Sep-1,Sep-2
Streams=1,2,3,4,5,6,7,8
Feeds=1
Estimates=7
Output=ep.fs
ExeDir=c:\cs\exes\

103
MaxIter=20
Method=Direct
Accuracy=0.001
! this is a comment

The different sections of the input file are separated by empty lines. You can add (non empty)
comment lines in the input file by, for example, using any punctuation character to start the line.
The flowsheet program looks for specific keywords and if none are found the line is considered
to be a comment. These comment lines will not get copied to the output file. The keywords are
the identifiers left of the equal signs in the above flowsheet section. They are not case-sensitive
and you may enter them in any order (some restrictions do apply, though). The keywords can
be different in each section.

Comma’s are used to separate the various components, streams, units, feeds, or stream esti-
mates, and therefore, they can not be part of a name. This is important as IUPAC component
names sometimes use comma’s. In that case omit the comma’s in the component name. It is
essential that the component names appear in the same order in the sep files for each unit!
The case of the names you specify is not important but the names in the flowsheet section are
used in the rest of the report.

In our example the unit operations are a mixer, reactor, and two columns (Sep-1 and Sep-2)
where we separate the products. We have identified the streams that connect these units using
numbers from 1 to 8 (names may also be used). There is only one feed that must be specified,
1, and we must estimate the flow and composition of the recycle stream (7). Stream 2 does
not actually exist in this process, but is needed here since the flash program that simulates the
mixer produces a vapor product stream (with no flow). Our output file is ep.fs but the flowsheet
program fs2 writes the output to the edit window. To save your results, you would need to use
the save option in the main menu.

The maximum number of flowsheet iterations allowed here is fixed at 20. Method refers to the
method used to converge the flowsheet calculations. In this case Direct substitution is used, with
a convergence tolerance of 0.001. To test for convergence the differences between stream vari-
ables (temperature, pressure, flows, compositions, enthalpies) are compared between flowsheet
iterations. The absolute differences between these variables or their relative differences

 new value − old value 

 
c= (3.2)
old value


are compared to the convergence criterion given in the input file. Absolute differences are
normalized otherwise the large values for the enthalpies or pressures would cause convergence
problems. The normalization factors are 102 for temperatures, 105 for pressures, and 107 for
enthalpies.

The next part of the input file describes each of the units in turn. There are four units in our
example. Each unit section starts with the [Unit] identifier and has five keywords: Name, File,
Type, Inlets, and Outlets.

104
[Unit]
Name=Mixer
File=ep-mix.sep
Type=go-col2.exe
Inlets=1,7
Outlets=2,3

[Unit]
Name=Reactor
File=ep-reac.sep
Type=reactors.exe
Inlets=3
Outlets=4

[Unit]
Name=Sep-1
File=ep-sep1.sep
Type=go-neq2.exe
Inlets=4
Outlets=5,6

[Unit]
Name=Sep-2
File=ep-sep2.sep
Type=go-neq2.exe
Inlets=6
Outlets=7,8

The specified unit name must match the one given in the flowsheet section. The unit file is
the associated data file, for most units it is the Sep-file that was created by ChemSep. The unit
type is the program that must be run to simulate this specific unit. For a ChemSep flash that
is go-col2.exe, for an equilibrium-stage column go-col2.exe, and for a nonequilibrium column
go-neq2.exe. This allows you to make your own unit simulation program that reads a data file
with feed and product section as in a Sep-file. In the case of the reactor this is handled by the
reactors.exe program (which is described below).

The units are connected by inlets and outlet streams that were declared in the flowsheet section
above. Be sure to use the same names (case is not important) as otherwise the flowsheet will be
incorrect or incomplete. If you do not want to run the simulator for a specific unit, omit the unit
file name and its type. The flowsheet program will consequently not execute anything for this
unit.

[Feed]

105
Name=1
Temperature=40C
Pressure=1.01325bar
Rate=20mol/s
Z=0.15,0.85,0

The next part of the input file consists of the [Feed] sections where the feed streams are defined.
The specified feed name must match the one given in the flowsheet section (again, case is not
important). Furthermore, the feed stream pressure, temperature, flow rate, and composition
must be specified. Append units when specifying temperatures, pressures, and flows. The
default units (which can be omitted) are temperatures in o C, pressure in bar (absolute), and
flows in mol/s (see Section ?? how to change the default units). The flowrate and compositions
can also be set by specifying the component flows, for example for the definition of the feed
above we could use as well

[Feed]
Name=Feed
Temperature=40C
Pressure=1atm
F=3,17

F is the list of component flows (default units mol/s). If all the component flows are speci-
fied, the total flow rate and compositions are computed. Other possible feed specifications are
vapor fraction or enthalpy (in J/kmol). The flowsheeting program uses the pressure and en-
thalpy specification internally. Only this type of specification guarantees proper handling of the
streams. For this reason one must be careful when including reactive distillation columns, for
the enthalpy in these column simulations includes the heats of formation (all units must switch
these on, see Section ??).

All streams are reset at the start of the simulation. Only the specifications supplied in the input
files are written to the unit file. If stream values are reset the values already present in the
unit-file will be used.

The last part consists of the [Estimate] section where estimates of stream variables (flow, tem-
perature, pressure, or composition) can be provided. This part is required only if you specified
streams under the estimates keyword in the flowsheet section. Stream estimates have the same
input format as feeds (except for the different identifier, of course). In our case we estimate the
recycle stream 7 to start with a better value of the flowrate to the reactor:

[Estimate]
Name=7
Temperature=80C

106
Pressure=1.01325bar
Rate=10mol/s
Z=0.85,0.15,0

Stream estimates are required for any recycles in the flowsheet because otherwise one of the
unit operations will have undefined feed specifications. However, information on the recycle
streams is likely to be unavailable at the start of a simulation. By specifying Rate=0 as an
estimate for the recycle stream the user will be able to start simulating the flowsheet. Upon
completion or interruption of the calculations, a new estimate will be written by the flowsheet
simulator.

The files that make up our example all start with ‘ep’ (ep.fs, ep-mix.sep, ep-reac.sep, ep-
sep1.sep, ep-sep2.sep) to group them together. In the main menu of the flowsheet program
you will also find a units option which shows a listing of the units in the flowsheet (Figure ??).
Selecting one of the units will start the ChemSep interface and load the associated datafile as
sep-file. This facilitates quick editing of the unit’s specifications.

107
108
Chapter 4

Analysis of Separation
Processes

In this chapter we will use the various tools ChemSep provides to analyze the specified prob-
lem: for example parametric studies to compare experimental data with simulations or thermo-
dynamic model predictions with measured VLE data. Or plotting stream properties as function
of temperature, pressure, or heat content as well as the rating of the column simulated using
vendor tools. These analysis tools are only available when either the components and models
have been defined or the problem has been solved before.

4.1 Plot Data File

Here we will explain how you can plot any data, for example experimental data that was pub-
lished in a journal article or a book. As an example we will use data on Sulzer BX packing
which was published in Chemical Engineering Progress, January 1998 (p.60). This paper was
written by by Mike Lockett and the data originated from the Fractionation Research Institute
FRI. The article showed ”measured” HETP’s as a function of the packing F-factor at four dif-
ferent pressures. In particular, we use a scan for Lockett’s plot of HETP’s for the o/p-Xylene
system at 1 atmosphere, Figure 1a. We could read the values from this graph by hand but to get
a little more precise numbers we use the tool ”ScanIt” to obtain the datapoints. ScanIt (available
for free) lets you pick the origin and two axis end points of the Figure to determine the scale
of the scan. After this you can select each point with the mouse and ScanIt will mark it and
compute the x,y values of the original datapoint.

With the ”Copy data” button we can copy the data to the clipboard. We open a Notepad text

109
 Figure 4.1: Using ScanIt to obtain the data points in the Figure.

editing window and paste the data in there. We also add some descriptions in this file:

# fullref= Chemical Engineering Progress, January 1998, p. 60, Fig. 1a

# author= Michael J. Lockett
# title= Easily predict structured-packing HETP
# url= http://www.aiche.org/CEP/Issues/1998-01/
# fileref= cep/cep1998-01p60.pdf
# subref= fig1a
# type= structured
# vendor= Sulzer
# name= BX
# system= xylenes
# pressure= 96 [kPa]
# location= FRI

110
# diameter= 1.22 [m]
# bedheight= 3 [m]
# xbot_ave= 0.3
# data_y= ave.HETP [m]
# data_x= F-factor [(Pa.s)ˆ0.5]
# data_type= experiments
# copyleft= Harry Kooijman, ChemSep.org

this basically gives a complete reference to the article. If possible, we refer to the publishers
URL of the paper. Note that this is not always possible, as in this case, as CEP does not have
an electronic versions of their older publications. In this case we make a link in the same way
as it works for the current publication of the journal. Important to realize is that any datalines
starting with a ”#” is not a data line. All these entries are providing meta information about
the data that follows. Does the ata represent experimental values or model predictions? What
was measured as a function of what, in what units? What chemical system was used, at what
pressure, and in how large a column? What was the (average) molefraction in the bottom of the
column of the more volatile compound? We can add a title and a label with:

# plottitle=Sulzer BX o/p-Xylene 730 Torr @ FRI (1.22m ID) CEPjan1998p60fig1a

# plotstage=Vertical
# plotcommands=set label "ChemSep" at 3,0.3 right

The title gives packing type, test system and pressure, column ID, and reference. The label is
just an example of text which is right alligned. There is also another ways of making labels, as
we will see later. We can now also define the X- and Y-axis with

# x1-title=F-factor
# x1-start=0
# x1-end=3
# x1-ticint=1
# x1-stics=4
# x1-log=Off
# y1-title=Average HETP (m)
# y1-start=0
# y1-end=0.3
# y1-ticint=0.05
# y1-stics=4
# y1-log=Off

Note that ChemSep can draw 4 axii, and that the default x- and y-axis are x1 and y1. This is the
notation used by GNUplot, the tool ChemSep uses to make its plots. To make define the color
and style of the datapoints we can use:

111
# label= exp.
# x-axis=1
# y-axis=1
# x-units=
# y-units=
# color=1
# style=1
# thick=1
# point=6
# flipp=0

which defines the datapoints as red filled circles labeled as ”experimental” points. Next we
paste the content of the buffer with the data collected by ScanIt:

0.242 0.145
0.306 0.0955
0.345 0.1
0.385 0.115
0.398 0.168
0.444 0.126
0.512 0.126
0.581 0.155
0.659 0.158
0.707 0.155
0.765 0.15
0.825 0.172
0.97 0.178
0.989 0.163
1.12 0.219
1.15 0.173
1.26 0.176
1.3 0.176
1.45 0.199
1.45 0.229
1.45 0.183
1.52 0.175
1.55 0.188
1.78 0.191
1.87 0.206
2.03 0.216
2.13 0.219
2.22 0.222
2.41 0.194

112
We save this as a text file with a descriptive name, for example cep1998janp60Fig1a.txt, where
the name uses the abbreviated journal name, followed by the volume number, a small ”p” for
page, and then the page number. This usually generates an unique name. When the journal
does not use volume numbers but year and month instead, we use those, with the month as
three small letters. For conferences we use an abreviated conference name, year, and a session
number or a page number of the proceedings.

The resulting text file can directly be plotted with ChemSep using Analysis - Plot Data File
menu-item. The plot has the same axii as it was originally published with as well as labels in
the plot. When multiple experimental data sets are shown in one graph it is better to separate
them into separate files and call them for example fig1a, fig1b, fig1c etc. This will allow them to
be used separately in model comparison studies as we will see below. You can always combine
them into one figure by adding another text file, for example as ”fig1”, by appended the files
into that file. Make sure you give each set a different label and color/point-type so that they
remain distinguable! Our example will look like Figure 4.2.

Figure 4.2: Showing the distillation test data with the ”Plot data file”.

113
4.2 Parametric Study

The Sulzer BX packing data as plotted with the Plot Data File option above can be compared
with with predictions of mass transfer coefficient (MTC) correlations in ChemSep. Lockett also
made such a comparison with the mass transfer model of Bravo, Rocha, and Fair (Hydrocarbon
Processing, January 1985, p. 91). Here we discuss how to do the same using the ChemSep
Analysis - Parametric Study option. To make this comparison you need to setup a simulation
of the distillation tests in ChemSep for the system in question. We like to use the same filename
as the dataset and add to it the model name (abbreviation), for example for the BRF85 model
cep1998janp60Fig1a BRF85.sep. The easiest way to get this done is by searching our online
database for a simulation (sep) file with the same distillation test mixture (and pressure) and
adapt it for the right packing, bed height, and MTC model.

Figure 4.3: Setting up the o/p-xylene simulation with the BRF’85 MTC model.

It is important to realize that ”an HETP” can not measured as such but that it is computed by
simulating the distillation tests with an equilibrium stage simulator while matching measured
bottoms and tops compositions. To get a good match one stage is to be taken to have a fractional
stage efficiency (between 0 and 1). The HETP is then resulting from the bed height divided by
the number of stages in the simulation (of course, excluding condenser and reboiler. If the
vapor velocity in the bed changes it is important to use the location with the highest flows for
computing the F-factor. When we have a converged simulation and verified it is correct we need

114
to simulate it at various flow rates using Analysis - Parametric Study option in ChemSep. We
vary the reflux flow to the column to simulate the tests at different vapor loadings. You might
need to play a little with the liquid reflux rates to obtain the right range for the test data. Note
that we chose to plot the average HETP, by using AHETP1. This computes an average HETP
over the whole bed of section number 1. We plotted it as function of the F-factor on stage 35,
which resides in the middle of our bed.

Figure 4.4: Using Parametric Study to simulate the column at different vapor loadings
(and hence F factors) and calculating the average HETP.

Then plot the data points again with the ”Plot data file” menu item and switch on the ”Graph
settings”. This will allow you to add a data-set wth the ”Add set” button. Here we plot the
second and third column in the Parametric Study, by using the identifiers PS2 and PS3, on axii
x1 and y1 for the 6 points generated in the Parametric Study (adapt this if you use a different
number and indices for the variables).

When we now click the ”Display” button we obtain the experimental HETP and the model
predicted HETP in one graph. We see that there is quite a good agreement with Lockett’s
publicated model predictions for the BRF’85 model. Deviations stem from using the average
HETP, different physical properties and the use of a full nonequilibrium model. It is important

115
Figure 4.5: Use ”graph settings” to add the BRF’85 prediction next to the experimental
data.

that we use a sufficient number of stages to integrate the packed bed and the correct flow models
are chosen in the simulation, of course.

To compare varous MTC models we need to be able to also compute the standard deviations
in this plot. It would also be nice if we can visualize the model deviations. We can! Simply
click the ”Calc Dev’s” button and these deviations will be computed for all data points. Note
that the range of the parametric study must be large enough for this to work. ChemSep uses
rational inter- and extra-polation to compute the model HETP’s for each data point. The overall
standard deviation in HETP and relative HETP is also added as label to the graph. This is all
written to a new data file which includes the deviations as individual line pieces, and ChemSep
asks you to specify a new name for this file. Typicaly we append a digit for the text file, for
example: Structured/Sulzer/BX/cep1998janp60Fig1a1.txt. When you inspect this file you see
also that ChemSep appends a label with the deviation and the relative error:

# string 0 0.3 s=0.024 r=0.143 b=+0.012

This allows you to further add other labels to the graph. The label may contain spaces but

116
Figure 4.6: The BRF’85 prediction with the FRI experimental data for Sulzer BX
packing with o/p-Xyelene at 1 atm shows a good agreement.

special characters might not print properly.

We are now ready to compare the results for different models and select the best MTC model
for this packing (and distillation system). We welcome anybody to contribute to our online
database of distillation tests. We try to cover all industrially relevant packings. Obviously, this
only works if the vendor makes its test data available or someone publishes measured distillation
tests on the packings.

4.3 Column Rating

ChemSep can perform a simple column tray rating to determine tray and overall column cost es-
timates (using tray correlations from Economopoulous, 198x, and cost correlations by Douglas,
1988). This preliminary rating can then be used to initialize software of industrial column inter-
nals vendors, for example SulCol by Sulzer Chemtech, that do more detailed column internals
rating using the vendors own (proprietary) capacity correlations.

117
Figure 4.7: The BRF’85 model shows a relative standard deviation of 14 per cent or
24mm on an HETP that increases from 120mm at low vapor loadings to 200mm at
high vapor loadings. Note that the bias is positive (+12mm), meaning that the BRF’85
model is 7 per cent optimistic in its prediction of the HETP!.

Figure 4.8: Column rating.

Note that in Figure 4.8 the column is split into two sections automatically by ChemSep, as the
feed to the column is used to automatically split up the column in a rectification and a stripping
section, that due to their different internal traffic have different diameters. You can insert or
remove any section, in cause this automatic division is not sufficient.

118
4.4 Stream Curves

Stream curves are often used to perform sizing of heat exchangers for condensers, reboilers, and
heaters/coolers for feeds and product streams. To size these exchangers it is very important to
determine the temperature and physical properties (especially densities, viscosities, and thermal
conductivities) as function of the heat content of the stream flowing through the exchanger.

4.5 Phase Diagrams

In ChemSep it is quick and easy to analyse the VLE of a mixture at hand. It is even possible to
make comparisons with measured VLE data. We’ll use binary VLE data on Acetone and Ben-
zene measured 25 degrees Celcius as an example. It was published by A. Tasic, B. Djordjevic,
D. Grozdanic, N. Afgan, D. Malic, in Chem. Eng. Sci. (1978), 33, pp. 189-197. Many binary
VLE data sets published in literature are collected in the thermodynamic libraries. The most
well known of these libraries is the DECHEMA data series of which there is also an online
version, called Detherm. This library includes the data of the Dortmund Daten Bank (DDB).

We will start with a new file in ChemSep and enter the VLE Data in the comments section of the
sep file. You can find the VLE data in our set of binary vle (bvle) files using the library indices
of acetone, 1051, and that of benzene, 501. The data file is then called ”bvle 1051 501 t25.txt”.
The order of the components is set to increasing normal boiling points. Since acetone boils at
a lower temperature than benzene it comes first. The component indices are separated with an
underscore since the uses of spaces in file names makes the use of commandline tools difficult.
As a last part of the file we add an indicator for the conditions at which the measurements were
taken, here at constant temperature of 25 C. We can use copy-paste for entering the data as text
in the comments field:

# file=bvle_1051_501_t25.txt
# ref=ces33p189
# title=P-xy data for acetone (1051) / benzene (501) at 25 C in torr
# fullref=A.Tasic, B.Djordjevic, D.Grozdanic, N.Afgan, D.Malic, Chem.Eng.Sci. (1978), 33, p
p x y
95.05 0 0
116.65 0.082 0.248
137.2 0.185 0.416
156.3 0.307 0.55
169.3 0.401 0.632
182.1 0.505 0.703
191.2 0.59 0.755
203.2 0.704 0.824
213.4 0.8055 0.882

119
220.95 0.8955 0.936
222.35 0.909 0.944
225.05 0.932 0.958
230.4 1 1

Note that we separated the data here by tabs. ChemSep’s import option for binary VLE data also
allows spaces, comma’s, or semicolons as separators. Important is that there is a line describing
the type of data in each column. When data is only available in graphical form you can use the
tool ScanIt to obtain the datapoints. ScanIt lets you pick the origin and two axis end points to
determine the scale of the scan. After this you can select each point with the mouse and ScanIt
will mark it and compute the x,y values of the original datapoint. With the ”Copy data” button
you can copy the data to the clipboard and from there to a sep file or a text file. Be sure to always
include the complete reference to the original article besides a descriptive title. If possible, we
refer to the publishers URL of the publication. We use specific set of journal abbreviations and
refer to the volume and first page of the publication, in this case ”ces33p189”. The title and
reference are put on lines that start with a ”#” to indicate the line does not contain data. By
including lines with the file name, short and full reference, and title, it is easier to automatically
scan a lot of files for specific information. It takes a little effort to do this systematically for
each file but the pay-back is huge when many of these small files are compiled into free online
resources. If you send us your files without references they are of little use to us as we can not
make them available to others.

When we want to compare the VLE data to predicted values by thermodynamic models we
need to define the components as well as make a model selection and provide the interaction
parameters for the models. Select ”Components” in the tree of the input navigator on the left
hand side. This will open a pane where we can select the compounds. ChemSep shows the
compounds in the default component library (a ”PCD” file) and you can select any of these by
double-clicking a compound. Typing a (part of a ) name in the ”find” box will narrow down the
list of components to only those with the particular string in the name (clicking the ”advanced”
checkbox allows for more sophisticated searched). We add the components in the order of their
boiling point, first acetone and then benzene:

Next we click on the ”properties” in the input navigation tree to select the thermodynamic
models. To describe Vapor-Liquid equilibria ChemSep uses K-values, which define the ratio of
the vapor to the liquid compositions (K=y/x). For ideal mixtures the K value equals the ratio of
the component vapor pressure to that of the system. For nonideal systems such as the system at
hand, this is insufficient, and an activity coefficient is needed to describe the non-ideality of the
liquid mixture fugacity. The resulting model for the K-values is called the DECHEMA model,
since it is also used in the DECHEMA data book series. As activity model we will first select
a predictive model, UNIFAC, which will require no further input. As vapor pressure model we
select the Antoine model.

To display our calculations in the same units of measure, select ”units” in the navigator. Our
experimental data was in degrees Celcius and torr (mm Hg). Since torr is not part of the standard
list of pressure units you must type it in the edit box. ChemSep allows you to do so and thus

120
 Figure 4.9: Entering VLE data in a new file.

compose all kinds of personal units of measure. ChemSep knows most chemical engineering
units and all the prefices to scale them.

Now save the file under a descriptive name, for example ”bvle 1051 501 t25.sep”. Then go to
Analysis - Phase diagrams - Binary menu option. This opens a window to generate our pxy
diagram. Select the type of diagram (pxy) and phase equilibrium (VLE) as well the condition
(temperature = 25C) and click on calculate to generate the phase diagram:

Now we want to make a comparison of the model predictions with the experimental data. For
this we must load the data either from a file or from the comments section. For both you must
click the ”file open” icon next to data. In our case we must select the ”read from comments
field” file type and select to open any file. This loads the following information onto the data
line (all in one line):

pxy;
p=95.05,x=0,y=0;
p=116.65,x=0.082,y=0.248;

121
 Figure 4.10: VLE model predictions.

p=137.2,x=0.185,y=0.416;
p=156.3,x=0.307,y=0.55;
p=169.3,x=0.401,y=0.632;
p=182.1,x=0.505,y=0.703;
p=191.2,x=0.59,y=0.755;
p=203.2,x=0.704,y=0.824;
p=213.4,x=0.8055,y=0.882;
p=220.95,x=0.8955,y=0.936;
p=222.35,x=0.909,y=0.944;
p=225.05,x=0.932,y=0.958;
p=230.4,x=1,y=1;

We see the data is processed: the pxy identifier is repeated and then used for identifying each
data item, separated by comma’s. The data points themselves are separated by semicolons.
Note that the original data did not have units, so no conversions are done. The diagram now has
the points shown as entered:

122
Figure 4.11: Comparing experimental data with UNIFAC/Antoine model predictions.

To quantify the deviations between the model and the experimental points we can compute
the standard deviations by clicking the ”Data Dev.” button. This computes for each point the
pressure q and vapor fraction z that is in equilibrium with the measured liquid composition x
and the specified temperature 25 C. ChemSep adds these to the data line:

pxy;
p=95.05,x=0,y=0;
p=116.65,x=0.082,y=0.248,q=116.8901,z=0.231314;
p=137.2,x=0.185,y=0.416,q=137.7293,z=0.414939;
p=156.3,x=0.307,y=0.55,q=158.3805,z=0.557008;
p=169.3,x=0.401,y=0.632,q=171.8393,z=0.636984;
p=182.1,x=0.505,y=0.703,q=184.7208,z=0.708714;
p=191.2,x=0.59,y=0.755,q=193.9699,z=0.759586;
p=203.2,x=0.704,y=0.824,q=204.9321,z=0.822478;
p=213.4,x=0.8055,y=0.882,q=213.554,z=0.877809;

123
p=220.95,x=0.8955,y=0.936,q=220.3915,z=0.930066;
p=222.35,x=0.909,y=0.944,q=221.3489,z=0.938402;
p=225.05,x=0.932,y=0.958,q=222.9353,z=0.95301;
p=230.4,x=1,y=1;

It also computes the relative standard deviations (only for the binary points) and adds them to the
title of the diagram. Here we see that the UNIFAC+Antoine models predict vapor compositions
(sy) with 0.72% and the pressure (sp) with 1.01%. This is actually very good. Actually most of
the deviation in the pxy diagram is caused by the Antoine model not capable of predicting the
exact pure component bubble pressures.

Figure 4.12: Calculating standard deviations in y and p.

Once you are happy with the way your VLE diagrams looks like use the Plot button to generate
a GNUplot diagram. From GNUplot you can bring the plot into any other windows program
supporting MS meta-graphics, such as Word and powerpoint, by pressing Alt-Space in GNUplot
and selecting Options - Copy to Clipboard.

Though the fit of the predictive UNIFAC model looks very good, we see that close to pure
acetone the dew and bubble point lines start to curve at a decreasing slope, indicating that we

124
 Figure 4.13: VLE diagram in GNUplot.

are close to an azeotrope. Actually, the data does not seem to have this tendency. When we
go back to the thermodynamic model selections and select the UNIQUAC model and load the
parameters from the library,

we will obtain a fit that is less good - deviations in pressure increase by more than a factor two
- but the curvature of the dew and bubble point lines is correct:

Thus, lower relative deviations of pressure and mole fractions aren’t necessarily indicative for a
qualitative better description of the binary VLE. This is due to the errors in the pure component
vapour pressures. Errors over half of the concentration range are introduced when a pure com-
ponent bubble point temperature is off by just one or two degrees! We leave it to the reader as
an exercise to select different models for the vapor pressure to see how the predictions can vary.

125
 Figure 4.14: UNIQUAC fit of the VLE data.

4.6 Residue Curve Maps

In ternary or multi-component mixtures, the distillation path in the column can be analysed by
means of ternary Residue Curve Maps (RCM).

4.7 Property Diagrams

Just as ChemSep allows an easy and quick way of plotting VLE diagrams, it also allows you to
do the same with any physical property. Here we will show how to plot binary diffusivities as
they would occur in a distillation column. We add the following data to the comments area in
the GUI:

# binary Diffusivities at 40 C
xA

126
0.000 1.700; 0.024 1.510; 0.100 1.000; 0.144 0.780; 0.200 0.680
0.254 0.635; 0.300 0.610; 0.400 0.640; 0.500 0.730; 0.590 0.850
0.600 0.865; 0.680 1.020; 0.700 1.060; 0.792 1.260; 0.800 1.275
0.880 1.440; 0.900 1.475; 0.960 1.570; 1.000 1.640
__

# binary Diffusivities at 58 C
xA
0.000 2.400; 0.024 2.143; 0.100 1.500; 0.144 1.220; 0.200 1.020
0.254 0.931; 0.300 0.930; 0.400 1.020; 0.500 1.185; 0.590 1.360
0.600 1.380; 0.680 1.550; 0.700 1.590; 0.792 1.792; 0.800 1.810
0.880 2.010; 0.900 2.060; 0.960 2.240; 1.000 2.360
__

Note that these represent data for property ”A” where everytime the data is presented as pair of
x and A values. Multiple values can be entered per line, they are separated with semicolons.
The ” ” denotes the end of the data. It is good to realize that property A will be the property
plotted with the left axis, and B using the right axis.

# Activity coefficients at 1 atm using Antoine vapor pressure

# Antoine Ethanol 8.1122, 1592.864, 226.184
# Antoine Water 8.07131, 1730.63, 233.426
xAB
0.015 6.192 0.997
0.040 3.564 1.042
0.045 4.218 1.032
0.121 2.830 1.065
0.200 2.202 1.091
0.201 2.215 1.061
0.223 1.956 1.174
0.235 2.002 1.149
0.274 1.777 1.170
0.319 1.606 1.231
0.376 1.440 1.313
0.393 1.423 1.310
0.496 1.227 1.503
0.548 1.148 1.629
0.650 1.085 1.781
0.826 1.018 2.148
0.849 1.013 2.185
__

127
 Figure 4.15: Liquid Fick diffusivities of ethanol-water at 40 degrees Celcius.

4.8 References

Economopoulous, 198x,

J.M. Douglas, Conceptual Design of Chemical Processes, McGraw-Hill, New York, (1988).

128
Figure 4.16: Liquid Fick diffusivities of ethanol-water at 40 and 58 degrees Celcius
described with the Kooijman (2002) method.

129
130
Chapter 5

Databanks

5.1 Pure Component Data (PCD)

Pure component data can be stored in binary or in text formatted files. The interface can inspect
and add/change/remove any components data. Of course, the binary data files are much smaller
and as they have a fixed format file the changes in an individual compound can be directly
written to the file, whereas text files require a sequential writing process than will require more
time when the data files become large. The default PCD library supplied with ChemSep is
chemsep1.pcd.

5.2 Pure Component Libraries (LIB)

Polynomial K-value and enthalpy correlation coefficients as well as extended Antoine coeffi-
cients are stored as component LIBraries (LIB files), which are ASCII files as well. The default
LIB files are

• EANTOINE.LIB (Extended Antoine)

• H-POLY.LIB (Example polynomial enthalpy coefficients)
• K-POLY.LIB (Example polynomial K-value coefficients)

Here the interface starts reading the file after the [LIB] keyword. Again a comment is read from
the next line (after the ‘=’) and then the data starts with a line for each component (first the
library index followed by the coefficients). For example the extended Antoine file (with data
from Prausnitz et al., 1980) starts like:

131
 Figure 5.1: PCDman.

[LIB]
Comment=Extended Antoine Prausnitz et al.
#
# ID A B C D E F G
902 3.15799e+01 -3.2848e+2 0. 0. -2.5980e+0 0. 2.0 Hydrogen
905 3.48329e+01 -9.5124e+2 0. 0. -2.5980e+0 0. 2.0 Nitrogen
901 3.57639e+01 -1.1660e+3 0. 0. -2.5980e+0 0. 2.0 Oxygen

5.3 Group Contribution Data (GCD)

Group Component Data (GCD) files are binary files containing data for group contribution
method such as UNIFAC or ASOG. There are three UNIFAC files (UNIFACRQ.GCD, UNI-
FACVL.GCD and UNIFACLL.GCD) for Vapor-Liquid and Liquid-Liquid systems. These files
need not be changed, unless for example the UNIFAC group tables have changed. Several
GCD-files are used for the estimation of pure component data in ChemSep. There are also text
versions of the GCD-files, they have the GCT file extension. They can be converted from bi-
nary to text format and vice versa with the GClib utility. This program also allows editing of
the data, see Figure 5.2.

132
 Figure 5.2: Editing the UNIFAC R and Q group contributions.

The GCD manager allows one to change the R and Q group contributions for each group. Note
that there is a third column which contains the molecular weights. Many calculations require
the molecular weight of the groups so this was added as a group parameter. UNIFAC uses
subgroups and main groups, for which both numbers can be edited. Note that the group names
are not used in the calculations. After exiting the GCD manager will prompt you whether you
want to save the group contribution data (in the same type of file format as it was read).

For UNIFAC, there are also the group interaction parameters to calculate the residual terms of
the liquid activity coefficients. Thus, a second type of GCD file can be edited; this file contains
the interaction parameters for all of the main groups. Figure 5.3 shows the GCD manager edit
window for this file. Note that interaction parameters that are unknown are set to zero so that the
residual part also becomes zero and that these interactions do not contribute to liquid activity
coefficients.

5.4 Interaction Parameter Data (IPD)

The Interaction Parameter Data (IPD) files are ASCII text files with the interaction parameters
for activity coefficient models and equations of state. Currently we have the following IPD files
for liquid activity coefficient models:

133
 Figure 5.3: Editing the UNIFAC group interaction parameters.

• MARGULES.IPD
• VANLAAR.IPD
• WILSON.IPD
• NRTL.IPD
• UNIQUAC.IPD
• UNIQUACP.IPD (UNIQUAC Q’)

and for equations of state:

• PR.IPD (Peng-Robinson)
• SRK.IPD (Soave-Redlich-Kwong)
• HAYDENO.IPD (Hayden O’Connell Virial)

These files are in plain text format file so that the user can add more data (it is probably better
to backup the original files or to use a new name for the extended files). The files may be edited
using any ASCII editor, or the built-in editor in ChemSep.

The actual data in these files comes after the line with the [IPD] keyword. The next line contains
the file-comment after the ‘=’, which will be used as a header in the interface. Next comes a
line (only for activity coefficient models) with the interaction parameters units. Then comes the

134
actual data, one line per binary pair. The first two numbers are the component library indices,
then the interaction parameters. Appended text is optional but will also be displayed. As an
example we include the first relevant lines of the NRTL.IPD file:

[IPD]
Comment=DECHEMA NRTL data @ 1atm.
Units=cal/mol
#
1101 1921 -189.0469 792.8020 0.2999 Methanol/Water p61 1/1a

Lines starting with a ‘#’ are comment lines which may appear anywhere. Since the interface
will only start reading the file from the [IPD] keyword on, you can start the file with some
text describing where the data was obtained and remarks on who/when/how changed the file.
Most of the IPD files contain data from the DECHEMA series, a very extensive collection of
interaction parameters. The Hayden O’Connell virial and the UNIQUAC Q’ parameters are
from Prausnitz et al. (1980).

5.5 Column Internals Parameter Libraries (INP)

135
136
Chapter 6

Tools and Settings

This chapter discusses the various tools and options ChemSep has. This includes discussions
of the files that ChemSep uses to define menu options, units of measure, nonequilibrium model
column internals and models, design mode defaults, etc. This allows the user to tailor the
interface and the program to the users specific needs.

6.1 Commandline Options

The following commandline options exist for the GUI (WinCS.exe):

/debug=logfile switch debugging on to logfile

/memdebug switch memory logging on
/granu=XX set memory granularity to XX bytes
-oFilename load options from filename. cnf extension is added if needed
-install generates the chemsep.cnf in the directory where wincs resides and registers Chem-
SepUO.dll when present
-hintsfile saves a hintsfile in ’chemsep.hnt ’ after loading the existing hintsfile. Can be used to
extract all widgets in the GUI by name
-pPath override paths. When path contains ’
bin’ then smart substitutions are made for the help, pcd, ipd, ild paths and the temporary
path is set to the ’WINTMPDIR’ as well as default synonyms file name is selected
-phlpPath override help path with the specified path string
-ppcdPath override pure component data path with the specified path string

137
-pipdPath override property data path with the specified path string
-pildPath override internals layout path with the specified path string
-pbinPath override binary executables path with the specified path string
-ptmpPath override temporary path with the specified path string
-pscrPath override scripts path with the specified path string
-LPCDfile sets the PCD file for displaying components to select from
-Cape-Open switches GUI into CO mode
-NN match CO components by name instead of CAS number
-CC component check & save the sep file w/o starting the GUI
-CHECK perform check on supplied sep-file and save under name.chk w/o starting
GUI
-new=name generate new sep-file with specified name
-getvar=name write value of a results variable in the specified sep-file w/o starting GUI
-out=name specify output filename
-setvar=name set value of an input variable in the specified sep-file w/o starting GUI
-value=value string with value to set
-units=units string with units of value to set
-getvardescriptions=name output results variable names of the specified sep-file w/o starting
GUI
-setvardescriptions=name output input variable names of the specified sep-file w/o starting
GUI
-getCOvars=name write CAPE-OPEN in- and output variables names of the specified sep-file
w/o starting the GUI
-plot=XX generate plot number XX as would show in the GUI but w/o starting GUI
-gout=gif/png/emf/svg/ps select file format of graphical output
-bvleplot generate binary phase vle diagram
-bvledev=name calculate and write binary vle deviations from data to file
-cs do not show about box or startup screen but start GUI immediately

6.2 Interface Settings and Files

The chemsep interface has a general settings screen, Fig. 6.1. This screen allows the user to set
the following defaults:

Numerical Input Number of significant digits, E notation, trailing zeroes, automatic units con-
versions, and the color of the input field that needs to be specified or is incorrect.

138
Hints The file with descriptions of the fields as well as all buttons, menu options, etc. When
the user want to use a different language he can select the english, german, or dutch
versions. The user may specify a different time that hints remains visible or select them
to not be shown.
Component selection Components can be stored in the sep file instead of loading them from
the binary PCD which is the default behavior (to shorten the sep-file). Also the default
identified can be selected (Name, Structure, or Formula).
File The user can specify his/her own file viewer and editor to be used
Printing A left margin and a line spacing may be used for printing tables
Definitions Here the user can inspect the various nonequilibrium models that are loaded at
startup of the interface as well as defined in the ChemSep Model development environ-
ment (see below). Only the nonequilibrium model requires these definitions which are
stored in the CHEMSEP.DEF file. The main definion types are Operations, Modeltypes,
Internaltypes, Models, and Internals. The interface can reload the definitions from a
different file. The user can opt for automatic assignment of column sections or a mode
where (s)he needs to set them by hand.
Sep-files When you solve a problem ChemSep saves the new specifications to a file with the
.sep extension that is read by the simulations programs. By default ChemSep creates a
backup of the old file using the backup extension specified here. If you want to keep
more than one backup change the number of backups, the maximum is nine. These are
numbered 1 through 9 using the name of the sep-file together with a special character to
separate the name from the backup number.
Output The user can select the number of columns in the stream table or the number of
columns in the table with the column sections. When these are set to zero all the streams
and column sections are put next to eachother, instead of wrapping them. The user can
also switch the internal computation of the physical properties on (the values are also
supplied by the simulator). Furthermore, the GUI font can be set here and program
which is used for producing the graphs. By default this is GNUplot in a normal window.
Alternatively, you can let GNUplot generate a GIF file and view that file in the browser.
A delay for the running of the external graph tool can be set so that the files needed for
the graphing are not deleted before the tool has finished running. There is also a possi-
bility to not generate a graph but to write an XML file instead that later can be imported
under ”imported data”. This allows the user to make overlapping graphs.

The CHEMSEP.CNF file is used to store the default configuration, such as the tools definitions,
directory structure, selected models, and solve options. CHEMSEP.CNF is the default config-
uration file which will be loaded upon startup of the interface. If none is found in the current
directory, the file supplied in the original distribution is used. This allows one to have multiple
CHEMSEP.CNF files in different directories, which automatically configure the interface to (a)
specific problem(s).

The CHEMSEP.SYN is a file containing synonyms for over 1000 compounds. While searching
for a special component name you can use synonyms if you have selected to do so in the options
interface spreadsheet. You must select this file as your synonyms file. The synonym search does

139
 Figure 6.1: Interface settings.

not work while typing in a searchlist for a synonyms name. You will have to issue a search under
the synonyms name! The synonyms file is an ASCII file you can modify to your needs.

The CHEMSEP.UDF file contains the definitions for the units of measure and the unit conver-
sions in ChemSep. The first line must have the number of following lines with on each a unit
definition. Such a unit definition consists of 15 characters (from column 1) with the unit abbre-
viation (take care, these are case sensitive!) followed by 15 characters (from column 16) with
the full unit name (not case sensitive). Then, from column 31 the offset-factor (fo) and multi-
plication factor (fm) come and finally the reference unit. The conversion is done according the
following formula:
Number (in Reference units) = fm * ( Number (in Units) - fo )
For example 22o C = 1.0(22 − −273.15) = 295.15K. You can inspect ASCII text files with
ChemSep’s file viewer (F7) or make simple modifications with edit-file. However, we strongly
suggest you do not change the original data files that come with ChemSep. We carefully selected
the data in these files and other users might use your changed data and obtain erroneous results.
We urge you first to copy the file to another name before you change or add anything to these
files. Errors in the unit conversion can be very frustrating to find so it is good to check some
results if you have changed or added a unit definition. To discourage people to make changes,
CHEMSEP.UDF is set to be a read-only file!

140
6.3 Tools and Tool Configuration

ChemSep allows the configuration of the tools menu to include any auxilary tool to be directly
called from within the interface. Likewise the toolbar is fully configurable. You can move
actions down or up the list and any ChemSep menu option with an icon can be added or removed
from the toolbar. Fig. 6.2 shows the tools and toolbar configuration.

Figure 6.2: Tools.

Each menu item can have a shortcut key, which can be edited to personalize the GUI to your
needs. Items frequently used could be programmed as function keys. Note that no checking is
done on multiple entries of the same shortcut key. The GUI will only activate the first menu
item it finds for such a shortcut.

External tools can be extremely handy for processing the files ChemSep creates as well as to
start utility programs (plotting tools, editors, etc.). By default ChemSep configures the separate
unit conversion tool and notepad as external tools. The latter is set to load the sep-file with the
specifications. Each external tool appears as a menu item under the tools menu. Tools can be
added, edited, moved, or deleted.

The editing of the configuration for an external tool is shown in Fig. 6.3. Each tool has a title
that is used in ChemSep’s Tools menu and it must be associated with an external executable
file. The user can specify a shortcut key if so desired. Also the directory where the tool should
be started from can be specified. If ommited, the starting directory equals that of the sep-file.
Finally a list of arguments can be specified. For this argument list macro’s can be inserted to fill
in the name of the currently loaded sep-file, or a specific path.

141
 Figure 6.3: Tool Configuration of Tools.

The use of the $SAVE macro can cause the interface to save the current problem prior to the
execution of the tool, and the $LOAD macro can cause the interface to load the sep file again
after the executable has finished. the $PROMPT macro causes a window to popup and ask the
user for input which will be inserted as arguments at the position of the macro.

6.4 Design Mode Settings

Tray design is done using many default values for the parameters that determine the tray layout,
all these values are set in the design mode. However, they also can be loaded from a configura-
tion file that resides in the directory where the sep file resides. The latter allows designs tailored
to specific operations (distillation or absorption, for example). The nonequilibrium design mode
checks the existance of a local TDESIGN.DEF definitions file. If it is there, the tray design pa-
rameters are read from this file in a fixed format. TDESIGN.DEF files can be generated and
altered by selecting ”Tools — Design Mode Settings — Save” option, see Fig. 6.4. The settings
of different DEF files can also be loaded into the interface.

Here we’ll briefly discuss each of the screens that define the settings for the built-in design
method for trays. Care must be taken in setting the design parameters as the layout design
process is an iterative process which needs termination with a useful design in order to guarantee
the convergence of the simulation problem.

Each TDESIGN.DEF has a general description line. The tray design algorithm adapts the layout

142
 Figure 6.4: General design mode settings.

in steps of certain sizes to obtain either a certain fraction of flood or the pressure drop to be less
than a certain maximum. Three different step sizes are used: large, medium, and small. Each
are defined as a fractional stepsize. The stepsizes applied to the tray layout and the required
number of iterations for the design method to come to an acceptable tray design are correlated:
the smaller the steps, the more iterations are required.

The simulator only triggers a re-design when the flows on a tray change by a certain amount.
Basically this allows the design to deviate from the selected fraction of flood or maximum
pressure drop. By default this is 5%. Be careful not to set this allowed deviation too small
otherwise the method will always exit after the maximum number of iterations is reached with a
layout that is most likely to be unbalanced. The re-design can be done inbetween each iteration
or, alternatively, only after the simulation has converged. The default method is the ”continuous
redesign” as this provides the best performance/convergence. If a simulation does not converge
in design mode, switching this to the discontinuous redesign mode might help.

The design method varies the column diameter, active area (and thus implicitely the downcomer
area), open (”free”) area ratio, weir height, and tray spacing to meet the specified fraction of
flood as well as a minimum turn down ratio. It uses a double iteration loop with the free area
ratio in the inner loop, using a convergence criterion. If the design method doesn’t converge this
criterion can be loosened. In addition to the targetted fraction of flood some other limits must
be met: a maximum liquid fractional entrainment, a maximum fractional vapor entrainment, a
maximum ratio of froth height over tray spacing (jet flood), and a maximum allowed pressure
drop over the tray. To prevent the design mode from hanging in an endless loop only a certain
maximum number of iterations is allowed.

For bubble cap and valve tray columns two different default settings are used for the bubble
caps and valves: one for small columns, another for large columns. The column diameter
where this switch is made can be set as well. Various switches can be set: whether to generate

143
tray parameter output and to generate warnings and error messages

Figure 6.5: Sieve tray design settings.

Several correlations for predicting sieve tray capacity can be selected: Treybal, Ward, Economopolous,
Lygeros and Magoulas, or Ogboja and Kuye. The latter is the default. Lower limits, default val-
ues, and upper limits are set for the sieve tray hole diameter, tray thickness, tray spacing, and
the free area ratio can be given to provide the most suitable starting point of the tray design and
to adhere to company guidelines.

Figure 6.6: Bubble-cap tray design settings.

As tray capacity correlation only the Ogboja and Kuye method can be used bubble-cap trays.
We have a different default bubble cap disk diameter in small and large columns a value where
we switch from small to large bubble-caps. Furthermore the user can specify a lower limit,
default value, and upper limit can be set for the bubble cap hole diameter, bubble cap skirt
height, slot height, and the pitch between caps.

There are three correlations for predicting valve tray capacity: Koch, Glitsch (default) and
Ogboja and Kuye. Defaults can be given for the coefficients K for a closed and open valve, the
valve eddy loss coefficient, the valve weight ratio legless/with legs, the default valve density,
the default valve thickness, and the default valve diameter (for both small and large columns).

For the tray weirs a lower limit, default value, and upper limit can be set for the weir height.
The weir height is also limited to be less than a certain fraction of the tray spacing. The design
method employs a maximum liquid weir load to determine the number of passes on a tray.
Liquid flow passes are added until the calculated liquid load is less than this limit. For northed

144
 Figure 6.7: Valve tray design settings.

Figure 6.8: Tray weirs design settings.

weirs a default notch depth can be specified and for serrated weirs a default serration depth and
serration angle. For circular weirs a diameter as fraction of the weir length can be defined.

Figure 6.9: Downcomer design settings.

Downcomer area is sized according to Koch or Glitsch rules or is taken as a fixed 12% of the
cross-sectional area. A maximum downcomer velocity check according to Koch or Glitsch can
be selected, or none. The downcomer average liquid fraction is specified as well as the minimum
downcomer residence time. When the weir height is less than the downcomer clearance, the
downcomer clearance is adjusted to maintain a certain downcomer liquid seal. A lower limit,

145
default value, and upper limit can be set for the downcomer clearance.

6.5 ChemSep Internals and Model Development

ChemSep allows the development of models for the proper nonequilibrium simulation of sep-
aration processes via dynamically linked libraries (DLL’s). The user is free to program these
models in C or in Fortran and can specify up to 50 model parameters which can consist of
model switches, for example type selections, or model coefficients. Models can be written to
compute mass transfer rates, pressure drops, liquid mixing and holdup, as well as entrainment or
backmixing. Similarly, the user can generate new types of column internals that are completely
compatible with the existing column internal types. Up to 50 hardware parameters can be sup-
plied to describe the column internal. A design method can be added (just like any model) that
enables the engineer to simulate non-existing new columns still to be designed. The following
types of models can be added:

• Mass Tansfer Coefficient (MTC)

• Pressure Drop (PD)
• Liquid Holdup (HL)
• Entrainment/backmixing (ENTR/BACK)
• Design method (DSGN)
• Heat transfer Coefficient (HTC)

The interface offers a step by step process that guides the user through the process of:

1. defining the model/internal parameters

2. writing the model code,
3. compiling the model code,
4. writing code to test and validate the model,
5. compiling the test-code,
6. running the test-code to test in a debugger,
7. compiling the code as a DLL,
8. compiling the test-code to run with the DLL,
9. running the test-code with the DLL in a debugger,
10. enable the model definitions in the interface,
11. load an example test sep-file that uses the model,
12. running the example test sep-file with/without running the debugger.

146
The writing of the DLL is simplified by use of templates which can be used to create a new
model or internal. The C/Fortran templates contain the required routines as empty routines.
The actual code only needs to be added inside these routines. The develop environment can
contain several models/internals next to eachother that can be selected by means of a selector,
see Figure 6.10.

Figure 6.10: ChemSep Model Development.

We will discuss here how define model/internal parameters, what routines are required (and
what their function), and then illustrate the model development with several examples.

6.5.1 Model Definitions Files

ChemSep reads a definitions file (CHEMSEP.DEF) at startup, where models for the mass trans-
fer coefficients, pressure drop, flow models, entrainment, and holdup are defined. This alleviates
the problem of having to adapt the ChemSep interface upon any addition of or modification to a
model. The definition file also includes the definitions of any non-standard type of internal. In
case no definitions file is found, the nonequilibrium part of ChemSep is disabled.

The definitions file must start with ‘[ChemSep Definitions]’ followed by a line which defines
the version (the current version is 3). Only after these two lines content may appear. Everything
after a ‘#’ is regarded as comment. Comments may appear on any content line, and may also be
appended after any content. There are five different definitions types that are read from the defi-

147
nition file: [InternalType], [Operation], [ModelType], [Internal], and [Model]. Other entries
are ignored. The five different types of definitions may be mixed throughout the definitions file.
A definition may have no blank lines in it as entries are ended by a blank line!

The [InternalType] defines the nature of an internal, that is whether it is continuous or discrete.
The [Operation] defines the nature of the operation, that is whether we are contacting a vapor
and a liquid (VL) or two liquids (LL). The [ModelType] defines the type of models, that is
whether the model describes mass transfer coefficients, pressure drops, etc. Each of these has
the following fields: ID, Name, and Short, for example:

[Operation]
ID=1
Name=Vapor-Liquid
Short=VL

The [Internal] definitions link these types together: its ‘Type’ must be one of the defined inter-
nal types, and its ‘Operation’ one of the defined operation types. It also defines which ‘Models’
need to be specified for the internal as well as which layout ‘Parameters’ the internal has. These
parameters and models are separated by semicolons (;) or comma’s (,). The parameters might
include a units string in parentheses and a default value (with different units) after a colon (:).
Note that multiple ‘Parameters’ entries are allowed but that its total length is limited to 255
characters (these rules also apply for the ‘Models’ and ‘Internals’ entries).

A parameter entry may also contain a list in square brackets, separated by vertical lines. In
such cases the interface will prompt the user to select the parameter from the list of items. The
first item in the list will result in the value 1, the second in 2, etc. A default item may also
be specified using the colon and the numeric value of the parameter. For the model entries,
either ID numbers or short notation may be used, as long as they are defined ModelTypes. For
example the definition for a baffle tray may be:

[Internal]
ID=10
Name=Baffle tray
Type=Discrete
Operation=VL
Short=Baffle
Models=MTC,PD,VF,LF,DSGN
Parameters=Column diameter (m):120cm;Baffle area (%):50
Parameters=Baffle spacing (m)
Parameters=Baffle type[Disk & donut|Single flow|Multiflow]:2
# default baffle type = Single

The [Model] definitions has instead of a ‘Models’ entry the ‘Internals’ entry, where the types of

148
internals for which the model can be used are defined. It also has the ‘Parameters’ where model
parameters can be defined. For the definition of a baffle tray above the parameters are a column
diameter (with meters as units and a default value of 1.2m), the baffle area (as percentage,
default 50%), the baffle spacing (in meters, without any default), and the baffle type (with a
pick list between 3 different types where the default is the single flow). Internals parameters
can be stored and reloaded to/from test files (*.inp) to enable the selection of standard types of
trays and/or packings.

The models that are required for the baffle tray are a mass transfer coefficient model (MTC),
a pressure drop model (PD), models for the vapor and liquid flow (VF and LF), as well as a
design method (DSGN). Note that the design method is also handled as a model. Thus, more
than one design method can be implemented. A sample design method definition is:

[Model]
ID=1
Name=Fraction of flood
Type=DSGN
Operation=VL
Short=%flood
Internals=Bubble cap;Sieve;Valve
Parameters=Fraction of flooding:75%;Fraction of weeping:70%

Model parameters can be loaded from model parameter files (*.mop). Note that a model def-
inition may also be repeated (with the same id, name, type, and operation) so to add another
internalstype to the list to which the model applies. Short fields are optional, and have a max-
imum length of ten characters, used for displaying selected models etc. ID fields associate
a unique number (in 1 . . . 255) to the definition. Only for the internal and model definitions
non-unique numbers are allowed. When the interface reads the definitions file it uses the Short
descriptions to assign the ID’s for the Operations, InternalTypes and ModelTypes, see Table 6.1
(the Short descriptions used by the interface are in parenthesis). Thus, you will have to use
these Short descriptions but are free to change the ID numbers or names.

There are standard definitions for internal types that are built into the interface. These do not
require the parameters entry to be specified. If they are specified, they -override- the internal
built-in types. For example, the definition of the bubble cap tray internal is:

[Internal]
ID=1
Name=Bubble cap tray
Type=Discrete
Short=Bubble cap
Operation=VL
Models=MTC,PD,VF,LF,ENTR

149
 Table 6.1: Interface Assigned Types

Operation InternalType ModelType

Vapor-Liquid (VL) Discrete (DIT) Mass transfer coefficient (MTC)
Liquid-Liquid (LL) Continuous (CIT) Pressure drop (PD)
Vapor flow (VF)
Liquid flow (LF)
Entrainment (ENTR)
Holdup (HOLD)
Light liquid flow (LLF)
Heavy liquid flow (HLF)
Backmixing (BACK)
Design method (DSGN)

The built-in types are listed in Table 6.2. This table also lists defined vapor and liquid flow
models and models for entrainment.

Table 6.2: Assigned Internals and Vapor flow, Liquid flow, and Entrainment Models

Internal Vapor Flow / Liquid Flow / Entrainment /

Light liquid Heavy liquid Backmixing
Bubble cap tray (1) Mixed (1) Mixed (1) None (1)
Sieve tray (2) Plug flow (2) Plug flow (2) Estimated (2)
Valve tray (3)
Dumped packing (4)
Structured packing (5)
Equilibrium stage (6)
RDC compartment (7)
Spray column stage (8)

An example of a model definition is:

[Model]
ID=1
Name=AIChE
Type=MTC
Operation=VL
Internals=Bubble cap,Sieve tray,Valve tray

150
Assigned models for Mass Transfer Coefficient and Pressure Drop models are listed in Table 6.3.
Model parameters may be read from model parameter libraries (*.mop).

Table 6.3: Assigned MTC and PD Models

Mass Transfer Coefficient Pressure Drop

AIChE (1) Fixed (1)
Chan Fair (2) Estimated (2)
Zuiderweg (3) Ludwig 1979 (3)
Hughmark (4) Leva GPDC (4)
Harris (5) Billet-Schultes 1992 (5)
Onda et al. 1968 (6) Bravo-Rocha-Fair 1986 (6)
Bravo-Fair 1982 (7) Stichlmair-Bravo-Fair 1989 (7)
Bravo-Rocha-Fair 1985 (8) Bravo-Rocha-Fair 1992 (8)
Bubble-Jet (9)
Bravo-Rocha-Fair 1992 (10)
Billet-Schultes 1992 (11)
Nawrocki et al. 1991 (12)
Chen-Chuang (13)
Handlos-Baron-Treybal (14)
Seibert-Fair (15)
Kronig-Brink-Rowe (16)
Rose-Kintner-Garner-Tayeban (17)
Sherwood (20)

6.5.2 Required DLL Routines

The DLL used by the simulator is always called MODINT.DLL and the development environ-
ment copies the selected DLL in and out of the binary executables directory. The DLL must
contain the following routines:

UsrFap Appends output to unit to specified filename

UsrFcl Close output to unit
UsrDin Assigns default column internal parameters and sets the number of model parameters
and defines which represent the internals diameter (D) and height (H)
UsrDot Outputs the column internal’s parameters
UsrDes Designs a column internal type
UsrRed Re-designs a section of one specific column internal type

151
UsrAt Computes internal interfacial area and residence times
UsrMTC Computes mass transfer coefficients (MTCs) for a column internal/model
UsrRep Report/calculate pressure drop, entrainment, fraction of flooding/weeping
UsrHTC Computes heat transfer coefficients (HTCs) for a column internal/model

The first two routines, UsrFap and UsrFcl, are used for opening and closing file output and do
not need to be modified. The UsrDin, UsrDot, UsrDes, and UsrRed routines are only needed
for programming a new type of column internal. As the simulator cannot know what the internal
parameters represent the user must supply his/her own routines to set default values, how the
parameters are named in the sep-file, and also how the internal is to be designed as well as how
different internals sitting in one column section are to be rationalized into one design (i.e. to
have the same diameter and parameter values). The UsrAt and the UsrMTC routines can com-
pute interfacial area and residence times as well as the mass transfer rates. The UsrRep routine
can compute pressure drop, liquid holdup, or entrainment and expresses the hydraulic behavior
of a column internal in terms of fraction of flood and fraction of weep (for trays. For other
internals this represents of lowest operation without impairing the mass transfer performanc of
the device).

The use of these routines is differently. To make a new mass transfer coefficient (MTC) model
for existing column internals requires just implementation of the UsrAt and the UsrMTC rou-
tines. Below we discuss such an example with how we can program the AIChE MTC model
which can employ different models for the liquid on the tray. To make a new pressure drop
(PD) model we just need to implement a new UsrRep routine. We will show this in an example
that models the pressure drop of Sulzer I-Rings. To make a new internal we need all routines.
We show this by implementing baffle trays with design and performance models as published
by Fair et al..

6.5.3 AIChE MTC model

The model defintions for our new AIChE model are:

# BEGIN INTERNALS TEMPLATE

# MTC (new definition)
[Model]
ID=150
Name=AIChE model
Type=MTC
Operation=VL
Short=MTC
Internals=Bubble cap;Sieve;Valve
Parameters=Hydraulic model[Original|Bennett]

152
parameters=Tray type[Bubble-Cap|Sieve|Fit]
Parameters=Debug[Off|On]
Parameters=Cv:1;Cl:1
# END TEMPLATE

The lines starting with # are comment lines. In this case we use them to mark the start and
end of the definitions template. Here we defined our new model of type MTC, to be used for
columns with vapor and liquid phases, and that can be used for the existing column internals
Bubble-Cap, Sieve, and Valve trays. There are 5 model parameters: the hydraulic model, the
type of tray, a debug flag, and two model coefficients, Cv and Cl. The hydraulic model is a
selector with two choices: Original and Bennett. The tray type is a selector with three choices
and the debug flag a selector with two choices (off/on). The model coefficients have the default
values of unity.

The UsrAt routine is used here to compute the interfacial area used in the mass transfer cal-
culations. we will discuss it in detail here. it starts with the call declaration that may not be
changed:

subroutine UsRAt(InType,iMTC,j,n,m,Parms,PMTC,
+ Vmf,Lmf,RhoV,RhoL,VisV,VisL,MolwtV,MolwtL,Sigma,
+ Atot,Avap,Aliq,Tvap,Tliq,iColD,iColH)
integer InType,iMTC,j,n,m,iColD,iColH
double precision Parms(n,m),PMTC(n,m),
+ Vmf,Lmf,RhoV,RhoL,VisV,VisL,MolwtV,MolwtL,Sigma,
+ Atot,Avap,Aliq,Tvap,Tliq
c ------------------------------------------------------------------

Next we declare the local variables and common blocks we are going to use:

c Local variables:
double precision Dc,Ts,Z,Active,Ahol,Pitch,W,hw,free,Xarea,Vol,
+ Vm,V,Uv,Lm,L,Ul,Ql,cc,Flv,
+ Aint,alfa,Hcl,Eps,Hf,
+ psi,HclHZ,Term,C,aeB,HclB,aeG,HclG
c Common block:
double precision A_t,A_l,A_v,T_v,T_l,Fs,Qlw
common /areat/ A_t,A_l,A_v,T_v,T_l,Fs,Qlw
c ------------------------------------------------------------------

The routine starts with making sure we execute the model only for the correct types of column
internals (bubble cap, sieve, and valve trays have the identifiers 1, 2, and 3) and the right mass

153
transfer model (our id is 150). If we do not filter for the right id’s we might overwrite calcula-
tions by other models! To signal we handled the call we set the iColD and iColH variables that
identify which parameters are the diameter and height of the column’s internal. If we do not do
this the simulator halts the execution.

c
if (InType .ge. 1 .and. InType .le. 3) then
if (iMTC .eq. 150) then
c
c set {iColD,iColH} to signal we handled call
c
iColD = 1
iColH = 2

Next we convert the tray dimensions we need for the computations. We take them from the
Params array and put them into local variables. The tray dimentsions are ordered by the pa-
rameters list for the sieve trays in the chemsep.def file. Strictly speaking this conversion is not
necessary as we can also access the tray dimension by directly refering to the Params array but
this technique helps to prevent programming mistakes with the array indices. For example if
the order of the tray parameters were to change it is easier to also propagate such changes when
we concentrate the conversion into local variables in blocks like this.

c
c Get tray layout parameters:
c (Parms from Sieve tray definition in chemsep.def):
c
c Parameters=Column diameter (m)
c Parameters=Tray spacing (m)
c Parameters=Number of flow passes
c Parameters=Liquid flow path length (m)
c Parameters=Active area (m2)
c Parameters=Total hole area (m2)
c Parameters=Downcomer area (m2)
c Parameters=Hole diameter (m)
c Parameters=Hole pitch (m)
c Parameters=Weir length (m)
c Parameters=Weir height (m)
c Parameters=Weir type[Segmental|Notched|Serrated|Circular]
c Parameters=Notch depth/Weir diameter (m)
c Parameters=Serration angle (rad)
c Parameters=Downcomer clearance (m)
c Parameters=Deck thickness (m)

154
c Parameters=Downcomer sloping (%)
c
Dc = Parms(j,iColD)
Ts = parms(j,iColH)
Z = Parms(j,4)
Active = Parms(j,5)
Ahol = Parms(j,6)
Pitch = Parms(j,9)
W = Parms(j,10)
hw = Parms(j,11)

Next we start to compute some tray geometry values such as the cross-sectional area and tray
volume, mass and superficial velocities, and vapor load factor (cc) and flow parameter (Flv).

c
c computed tray geometry
c
free = Ahol / Active
Xarea = 3.141d0 * Dc**2 / 4.0d0
Vol = Ts * Xarea
c
c compute mass, mass/area and superficial velocities, liquid volumetric flow
c
Vm = Vmf * MolwtV
V = Vm / Active
Uv = V / RhoV
Lm = Lmf * MolwtL
L = Lm / Active
Ul = L / RhoL
Ql = Lm / RhoL
Qlw = Ql / W
c
c C and F factor, flow parameter
c
cc = Uv * dsqrt(RhoV / (RhoL-RHoV))
Fs = dsqrt(RhoV) * Uv
Flv = (Lm / Vm) * dsqrt(RhoV / RhoL)

Next we compute the specific interfacial area (m2 /m3 ) using a correlation by Zuiderweg and
the liquid holdup and residence time by either the Bennett or the Colwell/Gerster models, de-
pendent on selection set by the hydraulic model parameter, PMTC(j,1):

155
c
c Zuiderweg correlation for clear liquid (m) and interfacial area (m2/m3)
c
Psi = Qlw / Uv * dsqrt (RhoL / RhoV)
HclHZ = 0.6d0 * (Psi * Pitch * hw * hw) ** 0.25D0
Term = Uv * Uv * RhoV * HclHZ * Flv / sigma
Aint = 43.d0 / free**0.3d0 * Term**0.53d0
c
if (PMTC(j,1) .eq. 2.d0) then
c
c Bennett’s liquid holdup model
c
C = 0.5d0 + 0.438d0 * dexp( -137.8d0 * hw )
aeB = dexp( -12.55d0 * cc**0.91d0 )
HclB = aeB*(hw + C * (Qlw/aeB)**0.67d0)
alfa = aeB
Hcl = HclB
Tliq = Hcl * Active / Ql
c
else
c
c Gerster et al. liquid height
c
HclG = 0.040d0 + 0.29d0*hw - 0.0135d0*Fs + 2.45d0*QlW
c
c Colwell model for liquid density
c
aeG = cc**2 / (9.81d0 * hclG)
aeG = 12.6d0 * aeG**0.4d0 / free**0.25d0
aeG = 1.d0/ (1.d0+aeG)
alfa = aeG
Hcl = HclG
Tliq = Hcl * Z * W / Ql
endif
c
c Vapour holdup fraction (eps) and froth height, residence times
c based on the Selected hydraulic model
c
Eps = 1.d0 - Alfa
Hf = Hcl / Alfa
Tvap = Eps * Hf / Uv
c

Next we compute the specific interfacial area per volume liquid and per volume vapor as well
as the total interfacial area. The results are stored in a common block so we can pick them up

156
again in the mass transfer rates calculation. Finally, dependent on the debug flag PMTC(j,3) we
write out some debug messages to standard output (STDOUT = unit 6) if this is requested:

c
c Interfacial Area’s
c
Aliq = Aint / (Alfa * Hf)
Avap = Aint / (Eps * Hf)
Atot = Aint * Active
c
c Store area’s & residence times in common block for MTC calcs
c
A_t = Atot
A_l = Aliq
A_v = Avap
T_v = Tvap
T_l = Tliq
c
c Debug messages
c
if (PMTC(j,3) .gt. 1.d0) then
write (6,*) ’Area/residence times AIChE ’,j
write (6,203) Atot,’ Atot’
write (6,203) Avap,’ Avap’
write (6,203) Aliq,’ Aliq’
write (6,203) Tvap,’ Tvap’
write (6,203) Tliq,’ Tliq’
write (6,203) Hf,’ Hf’
203 format (1H , 1PE12.5,(A))
endif
c
endif
endif
c
return
end

For the routine that computes the actual mass transfer coefficients we do the start with the
declaration of the routine, the local variables and the common block, and then verify again that
the calculations are to be done only for the correct internals and models type (on the basis of
the id’s in InType and iMTC). We set the iColD and iColH to signal we compute the MTC’s
and obtain the internals dimensions we need from the Params array:

157
 subroutine UsrMTC(InType,iMTC,j,n,m,Parms,PMTC,
+ Vmf,Lmf,Dv,Dl,RhoV,RhoL,VisV,VisL,ScV,ScL,
+ CpV,CpL,ConV,ConL,MolwtV,MolwtL,Sigma,
+ KKV,KKL,iColD,iColH)
integer InType,iMTC,j,n,m,iColD,iColH
double precision Parms(n,m),PMTC(n,m),
+ Vmf,Lmf,Dv,Dl,RhoV,RhoL,VisV,VisL,ScV,ScL,
+ CpV,CpL,ConV,ConL,MolwtV,MolwtL,Sigma,
+ KKV,KKL
c ------------------------------------------------------------------
c Local variables:
double precision Xarea,Vol,Uv,ql,cc,C,ae,hl,Fss,tll,tv
+ Dc,Ts,hw,W,Z,Ahol,Activ
c Common block:
double precision A_t,A_l,A_v,T_v,T_l,Fs,Qlw
common /areat/ A_t,A_l,A_v,T_v,T_l,Fs,Qlw
c ------------------------------------------------------------------
c
if (InType .ge. 1 .and. InType .le. 3) then
if (iMTC .eq. 150) then
c
c set {iColD,iColH} to signal we handled call
c
iColD = 1
iColH = 2
c
hw = Parms(j,11)

Next we compute the number of transfer units (NTU) from the AIChE correlations. Note that a
different equation is used for the liquid MTC dependent on the tray type parameter, PMTC(j,2).
The number of transfer units are scaled by the model parameters PMTC(j,4) (for the vapor)
and PMTC(j,5) (for the liquid). Mass transfer coefficients are computed by dividing the NTU’s
by the product of the interfacial area and the residence time (of each particular phase). Again,
debug messages are written when requested by the debug flag parameter, PMTC(j,3):

c
c Number of Transfer Units: NTUv, NTUl
c
KKV = (0.77d0 + 4.57d0*hw -0.238d0*Fs + 104.8d0*Qlw)
KKV = KKV / dsqrt(ScV)
if (PMTC(j,2) .eq. 1.d0) then
c Bubble-Cap
KKL = 20320d0*dsqrt(Dl)*(0.21d0*Fs+0.15d0)*t_l
else if (PMTC(j,2) .eq. 2.d0) then

158
c Sieve
KKL = 19700d0*dsqrt(Dl)*(0.4d0*Fs+0.17d0)*t_l
else if (PMTC(j,2) .eq. 3.d0) then
c Fit with coefficients
KKV = PMTC(j,4) * dsqrt(T_v/ScV)
KKL = PMTC(j,5) * dsqrt(T_v/ScL)
endif
c
c Mass Transfer Coefficients: Kv, Kl in (m/s)
c
KKV = KKV / (A_v * T_v)
KKL = KKL / (A_l * T_l)
c
c Debug messages
c
if (PMTC(j,3) .gt. 1.d0) then
write (6,*) ’MTC AIChE ’,j
write (6,203) KKV,’ kv’
write (6,203) KKL,’ kl’
203 format (1H , 1PE12.5,(A))
endif
c
endif
endif
c
return
end

The routine UsrHTC from the template we do not need to adapt as we do not want to change the
heat transfer coefficients ChemSep computes itself from the average mass transfer coefficeints.

When running through the debugger:

6.5.4 I-Rings Pressure Drop Model

-here the model must come-

6.5.5 Baffle Tray Internal

In this example we define a complete new column internal type, the baffle tray. We start defining
a new internal. Watch out to use a unique identifier number for the baffle tray (here we used
20) that’s not been used before. Indicate what type of internal it is (a discrete tray internal for

159
Figure 6.11: Stepping through the code for the AIChE MTC model in the debugger
allows you to see each calculation. All the variables can be inspected in the ”Locals”
watch window..

vapor and liquid operation only) as well as what kind of Models the internal requires: mass
transfer coefficients (MTC), pressure drop (PD), vapor and liquid flow (VF, LF) and design
mode (DSGN) models.

# BAFFLE-v1.0-BaffleTrays
# implementation Baffle trays

160
# INTERNAL (new definition)
[Internal]
ID=20
Name=Baffle tray
Type=Discrete
Operation=VL
Short=Baffle
Models=MTC,PD,VF,LF,DSGN
Parameters=Column diameter (m):1;Baffle spacing (m):0.5
Parameters=Baffle area (%):50;Baffle flow type[Single|Double|Multi]:1
Parameters=Debug[Off|On]:1

Then for each of the required models, specify these:

# PD (add to existing definition)

[Model]
ID=1
Name=Fixed
Type=PD
Operation=VL
Internals=Baffle

# VF (add to existing definition)

[Model]
ID=1
Name=Mixed flow
Type=VF
Operation=VL
Internals=Baffle

# LF (add to existing definition)

[Model]
ID=1
Name=Mixed flow
Type=LF
Operation=VL
Internals=Baffle

# DSGN (new definition)

[Model]
ID=20
Name=Fraction of flood
Type=DSGN
Operation=VL

161
Short=%flood
Internals=Baffle
Parameters=Fraction of flood (%):75
Parameters=Debug[Off|On]:1

# MTC (new definition)

[Model]
ID=210
Name=Baffle
Type=MTC
Operation=VL
Short=Baffle
Internals=Baffle
Parameters=C_v:0.0025;C_l:1.0
Parameters=Debug[Off|On|Extra|Wait]

# BAFFLE-v1.0-BaffleTrays

Note that here we have assigned pressure drop and flow models that use existing models (for
example the fixed pressure drop model where the pressures are specified in the interface).

We then write our UsrDin routine that sets the number of parameters of the baffle tray as well
as assigns default values (so to make it work out of the box — most of the time)

subroutine UsrDin(InType,nParms,ParmsD,j,n,m,iColD,iColH)
integer InType, nParms, j, n, m, iColD, iColH
double precision ParmsD(n,m)
c ------------------------------------------------------------------
include ’baffle.i’
c
if (InType .eq. iTbafl) then
nParms=5
c Parameters:
c Column diameter (m):1
iColD=jColD
ParmsD(j,jColD)=1.d0
c Baffle spacing (m):0.6096
iColH=jColH
ParmsD(j,jColH)=0.6096d0
c Baffle area (%):50
ParmsD(j,jArea)=0.5d0
c Baffle flow type[Single|Double|Multi]:1
ParmsD(j,jFtype)=1.d0
c Debug[Off|On]:1

162
 ParmsD(j,jDebug)=1.d0
endif
return
end

Note how we used an ”include file” called ’baffle.i’ where we defined constants such as the
ID of the baffle tray, ITbafl, that we use in the code. We also defined constants jColD, jColH,
jArea, jFtype, and jDebug to access the baffle tray parameters from the array with internal
parameters. This makes it easier to keep track of things and to move the sequence of the baffle
tray parameters (we’d only need to make changes in the include and defintions files). Next we
must write an UsrDot routine that outputs the internal to our sep-files. This is done by writing a
line with the tray type ID followed by a description and then an equal sign and the two numbers,
the operation type and the number of parameters the new internal has. In our case the operation
type is VL (=1) and we have five parameters that describe the tray. Thus ”= 1 5” is appended to
the description. Each next line contains a parameter value followed by a description (which is
ignored but enhances the readability of the sep-file):

subroutine UsrDot(iUnit,InType,j,n,m,Parms)
integer iUnit, InType, j,n,m
double precision Parms(n,m)
c ------------------------------------------------------------------
include ’baffle.i’
c
if (InType .eq. iTbafl) then
c write "tray-type description = operation-type number-internal-parms"
write (iUnit,201) InType, ’ Discrete Baffle tray = 1 5’
write (iUnit,203) Parms(j,jColD), ’ Column diameter (m)’
write (iUnit,203) Parms(j,jColH), ’ Baffle spacing (m)’
write (iUnit,203) Parms(j,jArea), ’ Baffle area (-)’
write (iUnit,203) Parms(j,jFtype),’ Baffle flow type’
write (iUnit,203) Parms(j,jDebug),’ Debug’
201 format (1H , i5,7X,(A))
203 format (1H , 1PE12.5,(A))
endif
return
end

subroutine UsrDes(InType, iDesgn, Dparms, j,n,m,

+ Vmf, Lmf, RhoV, RhoL,
+ VisV, VisL, MolwtV, MolwtL,
+ Sigma, SF,
+ Parms, iColD, iColH)

163
 integer InType, iDesgn, j,n,m,
+ iColD, iColH
double precision Vmf, Lmf, RhoV, RhoL,
+ VisV, VisL, MolwtV, MolwtL,
+ Sigma, SF,
+ Dparms(n,m), Parms(n,m)
c ------------------------------------------------------------------
include ’baffle.i’
c Local variables:
double precision foot, one, two, Hbdef, Csbdef, Abdef
parameter (foot=0.3048d0, one=1.d0, two=2.d0)
parameter (Hbdef=two*foot, Csbdef=0.65d0*foot, Abdef=0.5d0)
double precision Csbw,Sqrtrr,Uw,Ac
c
if (InType .eq. iTbafl) then
if (iDesgn .eq. iDbafl) then
iColD = jColD
iColH = jColH
c Flow type must be single baffle
Parms(j,jFtype) = one
c Set default open area if not set properly
if (Parms(j,jArea) .lt. 0.1d0 .or.
+ Parms(j,jArea) .gt. 0.9d0) then
Parms(j,jArea) = Abdef
endif
c Set default baffle spacing to
if (Parms(j,jColH) .lt. 0.01d0 .or.
+ Parms(j,jColH) .gt. 1.0d1) then
Parms(j,jColH) = Hbdef
endif
c Calculate vapor flooding velocity based on open area and
c scaled by baffle spacing to the power one half
Csbw = Csbdef * ( Parms(j,jColH) / Hbdef )**0.5d0
Sqrtrr = (RhoV/(RhoL-RhoV))**0.5d0
Uw = Csbw / Sqrtrr
c Derate vapor velocity with fraction of flood and system factor
if ((Dparms(j,jFF) .gt. 0.01d0) .and.
+ (Dparms(j,jFF) .le. one ) ) then
Uw = Dparms(j,jFF) * SF * Uw
else
Uw = 0.75d0 * SF * Uw
endif
c Calculate open, cross sectional area, and column diameter
Ac = (Vmf * MolwtV / RhoV) / Uw
Ac = Ac / Parms(j,jArea)

164
 Parms(j,jColD) = sqrt(4.d0*Ac/3.141d0)
c When Debugging show results
if (Parms(j,jDebug) .eq. 2.d0) then
write (io,*) ’Designing baffle tray ’,j
write (io,*) Dparms(j,jFF),’ Fraction of flood @ Design’
call UsrDot(io,InType,j,n,m,Parms)
endif
endif
endif
return
end

subroutine UsrRed(InType,n,m,jstart,jend,Parms,iFlow,iColD,iColH)
integer InType,n,m,jstart,jend,iFlow,iColD,iColH
double precision Parms(n,m)
c ------------------------------------------------------------------
include ’baffle.i’
c Local variables:
integer jd,jh,k
double precision dmax,hmax
c
if (inType .eq. iTbafl) then
c find the stages with largest diameter and spacing
j=jstart
jd=j
jh=j
iColD=jColD
iColH=jColH
dmax=Parms(j,iColD)
hmax=Parms(j,iColH)
do j=jstart+1,jend
if (Parms(j,jColD) .gt. dmax) then
dmax=Parms(j,jColD)
jd=j
endif
if (Parms(j,jColH) .gt. hmax) then
hmax=Parms(j,jColH)
jh=j
endif
enddo
c set design to stage with largest diameter
do j=jstart,jend
do k=1,5
Parms(j,k)=Parms(jd,k)

165
 enddo
enddo
c When Debugging show results
if (Parms(jd,5) .eq. 2.d0) then
write (io,*) ’Rationalizing baffle trays ’,jstart,’-’,jend
call UsrDot(io,InType,jd,n,m,Parms)
endif
endif
return
end

subroutine UsRAt(InType,iMTC,j,n,m,Parms,PMTC,
+ Vmf,Lmf,RhoV,RhoL,VisV,VisL,MolwtV,MolwtL,Sigma,
+ Atot,Avap,Aliq,Tvap,Tliq,iColD,iColH)
integer InType,iMTC,j,n,m,iColD,iColH
double precision Parms(n,m),PMTC(n,m),
+ Vmf,Lmf,RhoV,RhoL,VisV,VisL,MolwtV,MolwtL,Sigma,
+ Atot,Avap,Aliq,Tvap,Tliq
c ------------------------------------------------------------------
include ’baffle.i’
c Local variables:
double precision Aint,Xarea,Vol,Uv,Ul,Rev
c
if (InType .eq. iTbafl) then
if (iMTC .eq. iMbafl) then
c
iColD = jColD
iColH = jColH
c
Aint = 25.d0
Xarea = 3.141d0 * Parms(j,jColD)**2 / 4.0d0
Vol = Parms(j,jColH) * Xarea
Atot = Aint * Vol
Avap = Aint
Tvap = Vol * (RhoV/MolwtV) / Vmf
Aliq = Aint
Tliq = Vol * (RhoL/MolwtL) / Lmf
c
if (PMTC(j,3) .gt. 1.d0) then
write (io,*) ’Area/residence times Baffle tray ’,j
write (io,203) Atot,’ Atot’
write (io,203) Avap,’ Avap’
write (io,203) Aliq,’ Aliq’
write (io,203) Tvap,’ Tvap’

166
 write (io,203) Tliq,’ Tliq’
203 format (1H , 1PE12.5,(A))
endif
c
endif
endif
c
return
end

subroutine UsrMTC(InType,iMTC,j,n,m,Parms,PMTC,
+ Vmf,Lmf,Dv,Dl,RhoV,RhoL,VisV,VisL,ScV,ScL,
+ CpV,CpL,ConV,ConL,MolwtV,MolwtL,Sigma,
+ KKV,KKL,iColD,iColH)
integer InType,iMTC,j,n,m,iColD,iColH
double precision Parms(n,m),PMTC(n,m),
+ Vmf,Lmf,Dv,Dl,RhoV,RhoL,VisV,VisL,ScV,ScL,
+ CpV,CpL,ConV,ConL,MolwtV,MolwtL,Sigma,
+ KKV,KKL
c ------------------------------------------------------------------
include ’baffle.i’
c Local variables:
double precision Aint,Xarea,Vol,Uv,Ul,Rev
c
if (InType .eq. iTbafl) then
if (iMTC .eq. iMbafl) then
c
iColD = jColD
iColH = jColH
c
Xarea = 3.141d0 * Parms(j,jColD)**2 / 4.0d0
Vol = Parms(j,jColH) * Xarea
c
Uv = Vmf * MolwtV / Xarea / RhoV
Ul = Lmf * MolwtL / Xarea / RhoL
c
cc Rev = 0.0001d0 * Parms(j,jColD) * Parms(j,jArea)
cc ReV = ReV * RhoV * Uv / VisV
cc ShV = PMTC(j,jCv) * Rev**0.8d0 * ScV**0.33d0
cc ShL = PMTC(j,jCl) * ScL**0.5d0
c
c HTU in ft
c
KKV = ( (CpV/4.1868d3) /

167
 + (PMTC(j,jCv) * (7.3734d2*RhoL*Ul)**0.44d0 ) ) *
+ ( ScV / (CpV*VisV/ConV) )**(2.d0/3.d0)
c
c Kv in m/s
c
KKV = Uv / (0.3048d0 * KKV * Aint)
c
c What this is set to doesn’t matter as the liquid transfer
c model should be ignored! So let set it to a LARGE number
c in case this wasn’t done!
c
KKL = 1.d0
c
if (PMTC(j,jDbg) .gt. 1.d0) then
write (io,*) ’MTC Baffle tray ’,j
write (io,203) KKV,’ kv’
write (io,203) KKL,’ kl’
203 format (1H , 1PE12.5,(A))
if (PMTC(j,3) .gt. 2.d0) then
write (io,*) ’PMTC ’,PMTC(j,1),PMTC(j,2),PMTC(j,3)
call UsrDot(io,InType,j,n,m,Parms)
if (PMTC(j,3) .gt. 3.d0) then
read(*,*) ii
endif
endif
endif
c
endif
endif
c
return
end

The UsrRep routine computes the performance of the tray. In this case we set the pressure drop
to 1 Pascal for the baffle tray. Of course, this is incorrect and needs to be replaced by actual
code computing the pressure drop as a function of the vapor and liquid traffic on the tray.

subroutine UsrRep(iOut,iReprt,InType,j,n,m,iColD,iColH,
+ Parms,Pmtc,Ppd,Phold,Pvf,Plf,Pentr,
+ Vmf, Lmf, RhoV, RhoL,
+ VisV, VisL, MolwtV, MolwtL,
+ CpV, CpL, ConV, ConL, Sigma, sf,
+ Pdrop,HoldUp,Phil,FloodF,WeepF,
+ QQv,QQl,TTv,TTi,TTl)

168
 integer iOut,iReprt,InType,j,n,m,iColD,iColH
double precision Parms(n,m),Pmtc(n,m),Ppd(n,m),Phold(n,m),
+ Pvf(n,m),Plf(n,m),Pentr(n,m),
+ Vmf,Lmf,RhoV,RhoL,VisV,VisL,
+ CpV,CpL,ConV,ConL,MolwtV,MolwtL,Sigma,
+ Pdrop,HoldUp,Phil,FloodF,WeepF,
+ QQv,QQl,TTv,TTi,TTl
c ------------------------------------------------------------------
include ’baffle.i’
c
double precision a
c
if (InType .eq. iTbafl) then
c
c Baffle
c ------------
c
iColD = j1ColD
iColH = j1ColH
c
Pdrop = 1.d0
c
if (iReprt .eq. 2) then

write(iOut,*) ’TestOut ’, 0.01d0*DBLE(j)

endif
c
endif
c
return
end

The UsrHTC, UsrFap, and UsrFcl routines we leave unchanged.

169
170
Chapter 7

Reactive Distillation

171
Chapter 8

Extraction

173
Chapter 9

Three Phase Distillation

175
Part B
Technical Reference
178
Chapter 10

Property Models

This chapter contains a catalog of all of the thermodynamic and physical property
models available in ChemSep. The formulae for most methods are given here but
additional reading is available in the following sources:

A: R.C. Reid, J.M. Prausnitz and B.E. Poling, The Properties of Gases and Liquids,
4th Ed., McGraw-Hill, New York (1988).

B: S.M. Walas, Phase Equilibria in Chemical Engineering, Butterworth Publish-

ers, London (1985).

References to these works are between parentheses, by combining the letter A or B

with the page number, for example (A43). Some of the methods described below are
referred to by the name given to them in Danner and Daubert (1983).

10.1 Thermodynamic Properties

These properties are required for both the equilibrium and the nonequilibrium models.
They determine the equilibrium mole fractions of the phases that are present in the
column as well as the terms in the energy balance. Typically, they have a profound
influence on the simulation. Selecting the right models will require insight from the
user; there are no models that cover all systems and operations.

179
10.1.1 K-value models

The distribution ratio of component i, the K-value, is defined as the ratio of the mole
fraction in the vapor over the mole fraction in the liquid phase:
yi
Ki = (10.1)
xi
Of course, for extraction columns, where two liquid phases are present the K-value
equals the ratio of the liquid mole fractions in each liquid.

The available K-models for VLE are:

Ideal (A251, B548) K-values for ideal mixtures are given by Raoult’s law:

Pi∗
Ki = (10.2)
P
where Pi∗ is the vapor pressure of component i and P the system pressure.

EOS (A319, B301) K-values are calculated from the ratio of fugacity coefficients:

φLi
Ki = (10.3)
φVi

where the fugacity coefficients (φ) are calculated from an equation of state. This
model is recommended for separations involving mixtures of hydrocarbons and
light gases (hydrogen, carbondioxide, nitrogen, etc.) at low and high pressures.
It is not recommended for nonideal chemical mixtures at low pressures. The
EOS must be able to predict vapor as well as liquid fugacity coefficients.

Gamma-Phi (A250, B301) K-Values are calculated using an activity coefficient model
for the liquid phase and an equation of state for the vapor:

γi φ∗i Pi∗ PFi

Ki = (10.4)
φVi P

This option should be used when dealing with nonideal fluid mixtures. It should
not be selected for separations at high pressures.

DECHEMA (B301) K-values are calculated from a simplified form of the complete
Gamma-Phi model in which the vapor phase fugacity coefficient and Poynting
correction factor are assumed equal to unity:

γi Pi∗
Ki = (10.5)
P

180
 This is the form of the K-value model used in the DECHEMA compilations
of equilibrium data (Hence the name given to this menu option). DECHEMA
uses the Antoine equation to compute the vapor pressures but ChemSep allows
you to choose other vapor pressure models if you wish. This option should be
used when dealing with non-ideal fluid mixtures. It should not be selected for
separations at high pressures.

Chao-Seader (B303) The Chao-Seader method is widely used for mixtures of hydro-
carbons and light gases. It is not recommended for nonideal mixtures. The
method uses the Regular solution model for the liquid phase and the Redlich
Kwong EOS for the vapor phase. An alternative choice would be the Equation
of State option.

Polynomial (B11) Calculate K-value as function of the absolute temperature (Kelvin):

1/mi
Ki = Ai + Bi T + Ci T 2 + Di T 3 + Ei T 4 (10.6)

You must supply the coefficients A through E and the exponent m.

There is currently only one LLE model available in ChemSep. In this model the K-
value model is equal to the ratio of the activity coefficients in the two liquid phases:

xIi γiII
Ki = = (10.7)
xII
i γiI

For supercritical extractors an alternative is to use cubic equation of states and define
the K-value as the ratio of the fugacity coefficients, just as described above.

10.1.2 Activity coefficient models

Here we discuss the activity coefficient models available in ChemSep. For an in depth
discussion of these models see the standard references. For the calculation of activity
coefficients and their derivatives (for diffusion calculations) see Kooijman and Taylor
(1991). Most of these models require interaction parameters that can be loaded from
libraries.

Ideal For an ideal system the activity coefficient of all species is unity (ln γi = 0).

Regular (A284, B217) The regular solution model is due to Scatchard and Hilde-
brand. It is probably the simplest model of liquid mixtures. The model uses the

181
 Flory-Huggins modification. The activity coefficient is given by:
c
X
θi = Vi / xk Vk (10.8)
k
 2
c
Vi  X
ln γi∗ = δi − xj δ j θ j  (10.9)
RT j

ln γi = ln γi∗ + ln θi + 1 − θi (10.10)

where δi is called the solubility parameter and Vi the molal volume of compo-
nent i (both available in the PCD-file). This model also is incorporated into the
Chao-Seader method of estimating K-values.

Margules (A256, B184) The ‘three suffix’ or two parameter form of the Margules
equation is implemented in ChemSep:

ln γi = (Aij + 2(Aji − Aij )xi ) x2j (10.11)

It can only be used for binary mixtures (i=1, j=2 and i=2, j=1).

Van Laar (A256, B189) The Van Laar equation is

Aij
ln γi =  2 (10.12)
Aij xi
1+ Aji xj

It can only be used for binary mixtures (i=1, j=2 and i=2, j=1).

Wilson (A274, B192) The Wilson equation was proposed by G.M. Wilson in 1964. It
is a 1two parameter equation’. That means that two interaction parameters per
binary pair are needed to estimate the activity coefficients in a multicomponent
mixture. For mixtures that do NOT form two liquids, the Wilson equation is,
on average, the most accurate of the methods used to predict equilibria in multi-
component mixtures (Reference B). However, for aqueous mixtures the NRTL
model usually is superior.

Λij = (Vj /Vi ) exp(−(µij − µii )/RT ) (10.13)

Xc
Si = xj Λij (10.14)
j=1
c
X
ln γi = 1 − ln(Si ) − xk Λki /Sk (10.15)
k=1

The two interaction parameters per binary pair of components: (µij − µii ) and
(µji − µii ).

182
NRTL (A274, B201) The NRTL equation due to Renon and Prausnitz is a three pa-
rameter equation. Unlike the original Wilson equation it may also be used for
liquid-liquid equilibrium calculations.

τij = (gij − gii )/RT (10.16)

Gij = exp(−αij τij ) (10.17)
Xc
Si = xj Gji (10.18)
j=1
Xc
Ci = xj Gji τji (10.19)
j=1
c
X
ln γi = Ci /Si + xk Gik (τik − Ck /Sk )/Sk (10.20)
k=1

The interaction parameters per binary are aij = (gij − gii ), aji = (gji − gii ),
and αij (only one α is required as αij = αji ). The temperature dependent
version computes the interaction parameters with a reference temperature of
T o = 25o C = 298K and:
 
o T
aij = aij 1 − o (10.21)
T

UNIQUAC (A274, B205) UNIQUAC stands for Universal Quasi Chemical and is a
very widely used model of liquid mixtures that reduces, under certain assump-
tions, to almost all of the other models mentioned in this section. Like the
Wilson equation, it is a two parameter equation but is capable of predicting
liquid-liquid equilibria as well as vapor-liquid equilibria. Two versions of the
UNIQUAC model are available in ChemSep: Original and q-prime. Original is
the default option and is to be used if you have obtained interaction parameters
from DECHEMA. The q-prime (q 0 ) form of UNIQUAC is recommended for
alcohol mixtures. An additional pure component parameter, q 0 , is needed. Both
versions are identical if q 0 = q.
c
X
r = xi ri (10.22)
i=1
Xc
q = xi qi (10.23)
i=1
φi = xi ri /r (10.24)
θi = xi qi /q (10.25)
τji = exp(−(λji − λii )/RT ) (10.26)

183
 c
X
Si = θj τji (10.27)
j=1
   
 z  φi z θi
ln γiC = 1 − qi ln + qi ln (10.28)
2 xi 2 xi
 
ri z ri qi
− + q −
r 2 r q
c
!
X θk τik
ln γiR = qi 1 − ln(Si ) − (10.29)
Sk
k=1
ln γi = ln γiC + ln γiR (10.30)

Here, φi is the segment fraction (similar to the volume fraction) and θi is the area
fraction of component i. The interaction parameters per binary are (λij − λii )
and (λji − λii ). The parameters ri and qi are read from the component database
(PCD file).

UNIFAC (A314, B219) UNIFAC is a group contribution method that is used to pre-
dict equilibria in systems for which NO experimental equilibrium data exist.
The method is based on the UNIQUAC equation, but is completely predictive
in the sense that it does not require interaction parameters. Instead, these param-
eters are estimated from group contributions of all the molecules in the mixture.
UNIFAC uses a set of main groups, each of which have their own interaction
parameters aij (and aji ) with which the τij are computed:

τij = exp(−aij /T ) (10.31)

These main groups are further divided into subgroups (k) that have their own
Rk and Qk values (these parameters are obtained from the van der Waals group
volume and surface areas). The ri and qi for component i are computed by
summing the product of the number of group k in i with Rk , respectively Qk :
X
ri = νki Rk (10.32)
k
X
qi = νki Qk (10.33)
k

The combinatorial part can then calculated using the UNIQUAC equations. In
the residual part of the activity coefficient the segment fraction (θ) is replaced
by:
c
X
Q = Xi Qi (10.34)
i=1
θi = Xi Qi /Q (10.35)

184
 where Xi is the mole fraction of group i in the mixture. Summations and resid-
ual parts are then computed for group k instead of component i (using Eqs.
10.27, 10.29). The residual part of the activity coefficient for component i is
then computed with:
X  
ln γiR = νki ln γkR − ln γkR,i (10.36)
k

where the last term, ln γkR,i , represents the residual part in a reference solution
containing only molecules of type i. This normalization makes ln γiR become
zero when the mixture contains only molecules i.
If you select one of the other models but fail to specify a complete set of the
interaction parameters, then UNIFAC is used to compute any unspecified pa-
rameters.

ASOG (A313, B219) ASOG is a group contribution method similar to UNIFAC but
based on the Wilson equation. It was developed before UNIFAC but is less
widely used because of the comparative lack of fitted group interaction param-
eters.

10.1.3 Vapor pressure models

Antoine (A208, B11) The Antoine Equation is:

Bi
ln Pi∗ = Ai − (10.37)
T + Ci

Note the natural logarithm. This option should be selected if you are using ac-
tivity coefficient models with parameters from the DECHEMA series. Antoine
parameters are available in the ChemSep data files and need not be loaded. If
you wish to use different parameters then select the Extended Antoine model
and enter the parameters yourself.

Extended Antoine (B11) An equation with more parameters that is sometimes known
as the extended Antoine equation is incorporated in ChemSep

Bi
ln Pi∗ = Ai + + Di T + Ei ln T + Fi TiG (10.38)
Ci + T

The parameters A through G must be supplied by the user. A library of pa-

rameters for some common chemicals is provided with ChemSep in the file
EANTOINE.LIB.

185
DIPPR (B11) The Design Institute for Physical Property Research (DIPPR) has pub-
lished a correlation for the vapor pressure that takes the form:
Bi
ln Pi∗ = Ai + + Ci ln T + Di TiE (10.39)
T
Riedel (B523) The Riedel equation is best suited to nonpolar mixtures:
ζT = 36/Tr + 96.7 log Tr − 35 − Tr6 (10.40)
6
ζT b = 36/Trb + 96.7 log Trb − 35 − Trb (10.41)
φ = 0.118ζT − 7 log Tr (10.42)
ψ = 0.0364ζT − log Tr (10.43)
0.136ζT b + log Pc − 5.01
α = (10.44)
0.0364ζT b − log Trb
log Pr∗ = −φ − (α − 7)ψ (10.45)

Lee-Kesler (A207, B69) Lee and Kesler used a Pitzer expansion to obtain:
ln Pi∗ = f (0) + ωi f (1) (10.46)
6.09648
f (0) = 5.92714 − − 1.28862 ln Tr + 0.169347Tr6 (10.47)
Tr
15.6875
f (1) = 15.2518 − − 13.4721 ln Tr + 0.43577Tr6 (10.48)
Tr
where Tr = T /TCi . Both the Riedel and Lee-Kesler models are recommended
for hydrocarbon mixtures in particular.

10.1.4 Equations of State (EOS)

Three types of equations of state may be selected in ChemSep; Ideal Gas, Virial, and
Cubic EOS. The ideal gas EOS is, of course:
RT
P = (10.49)
V
The fugacity coefficient of an ideal gas mixture (B3) is unity.

The Virial and cubic EOS are discussed in the sections below.

10.1.5 Virial EOS

The two-term virial equation:

RT BRT
P = + (10.50)
V V

186
where B is the second virial coefficient. Different methods are available for estimating
this coefficient.

Hayden-O’Connell (B39) Hayden and O’Connell have provided a method of pre-

dicting the second virial coefficient for multicomponent vapor mixtures. The
method is quite complicated (see Prausnitz et al., 1980) but is well suited to
ideal and nonideal systems at low pressures. You must input the association
parameters. A library of association parameters is provided with ChemSep in
the file HAYDENO.IPD.
Tsonopoulous (B45) The method of Tsonopoulous for estimating virial coefficients
is recommended for hydrocarbon mixtures at low pressures. It is based on a
correlation due to Pitzer.
c X
X c
B = yi yj Bij (10.51)
i=1 j=1
 
(0) (0)
Bij = RTc,ij Pc,ij Bij + ωij Bij (10.52)
(0) 0.33 0.1385 0.0121 0.000607
Bij = 0.1445 − − − − (10.53)
Tr Tr2 Tr3 Tr8
(1) 0.331 0.423 −0.0008
Bij = 0.0637 + − (10.54)
Tr2 Tr3 Tr8
ωi + ωj
ωij = (10.55)
2
Zci + Zcj
Zc,ij = (10.56)
2
1/3 1/3
1/3 Vci Vcj
Vc,ij = (10.57)
2 p
Tc,ij = (1 − kij ) Tci Tcj (10.58)
Zc,ij RTc,ij
Pc,ij = (10.59)
Vc,ij
Binary interaction parameters kij must be supplied by the user. For paraffins
kij can be estimated from:
p
8 Vci Vcj
kij = 1 − 1/3 1/3
(10.60)
(Vci + Vcj )3

DIPPR The Design Institute for Physical Property Research (DIPPR) has published
a correlation for the second virial coefficient.
Chemical theory This is an extension of the Hayden O’Connell virial model that takes
into account the association between molecules (see Prausnitz et al., 1980).

187
 Since the mole fractions are a function of the association, an iterative method
(here Newton’s method) must be used to obtain them in order to compute the
virial coefficients.

10.1.6 Cubic EOS

Van der Waals (A43, B15) The Van der Waals (VdW) Equation was the first cubic
equation of state.
RT a
P = − (10.61)
V −b V2
with
27R2 Tci
2
ai = (10.62)
64Pci
RTci
bi = (10.63)
8Pci
and the mixing rules:
c X
X c
a = yi yj aij (10.64)
i=1 j=1
√
aij = ai aj (10.65)
c
X
b = yi bi (10.66)
i=1

The VdW equation should not be used to determine properties of liquid phases,
thus it may not be selected for the EOS K-value model. The basic equation has
served as a starting point for many other EOS.
Redlich Kwong (A43, B43) The Redlich Kwong (RK) equation is used in the Chao-
Seader method of computing thermodynamic properties. The RK equation
should not be used to determine properties of liquid phases (ChemSep does not
allow it’s selection as the EOS K-value model.
RT a
P = −√ (10.67)
V −b T V (V + b)
with
Ωa R2 Tci
2.5
ai = (10.68)
Pci
Ωa = 0.42748 (10.69)
Ωb RTci
bi = (10.70)
Pci
Ωb = 0.08664 (10.71)

188
 and the mixing rules:
c X
X c
a = yi yj aij (10.72)
i=1 j=1
√
aij (1 − kij ) ai aj
= (10.73)
Xc
b = yi bi (10.74)
i=1

where kij is a binary interaction parameter (original RK: kij = 0).

Soave Redlich Kwong (A43, B52) Soave’s modification of the Redlich Kwong (SRK)
EOS is a very widely used method of computing thermodynamic properties.
The SRK EOS is most suitable for computing properties of hydrocarbon mix-
tures.
RT a
P = − (10.75)
V − b V (V + b)
with

ai = ai (Tci )αi (10.76)

Ωa R2 Tci 2
ai (Tci ) = (10.77)
Pci
Ωa = 0.42747 (10.78)
 p 2
2
αi = 1 + (0.480 + 1.574ωi − 0.176ωi )(1 − Tri ) (10.79)
Ωb RTci
bi = (10.80)
Pci
Ωb = 0.08664 (10.81)

and the mixing rules:

c X
X c
a = yi yj aij (10.82)
i=1 j=1
√
aij (1 − kij ) ai aj
= (10.83)
Xc
b = yi bi (10.84)
i=1

API SRK EOS (B53) Graboski and Daubert modified the coefficients in the SRK EOS
and provided a special relation for hydrogen. This modification of the SRK EOS
has been recommended by the American Petroleum Institute (API), hence the
name of this option. It uses the same equations as the SRK except that:
 p 2
αi = 1 + (0.48508 + 1.55171ωi − 0.15613ωi2 )(1 − Tri ) (10.85)

189
 with a special relation for hydrogen:

αi = 1.202e−0.30288Tri (10.86)

Peng Robinson EOS (A43, B54) The Peng-Robinson equation is another cubic EOS
that owes it’s origins to the RK and SRK EOS. The PR EOS, however, gives
improved predictions of liquid phase densities.

RT a
P = − (10.87)
V − b V (V + b) + b(V − b)

The parameters are given by the same equations as for the SRK EOS with the
following differences.

Ωa = 0.45724 (10.88)
 p 2
αi = 1 + (0.37464 + 1.5422ωi − 0.26992ωi2 )(1 − Tr ) (10.89)
Ωb = 0.07880 (10.90)

10.1.7 Enthalpy

The mixture enthalpy can be calculated from the component enthalpies with:
c
X
H= zi Hi (10.91)
i=1

where zi is the mole fraction of species i in the phase in question. The component
enthalpy may be expressed as the sum of the ideal contribution and an excess enthalpy:

Hi = Hiid + Hiex (10.92)

where Hiex is the excess enthalpy and Hiid is the ideal contribution. The ideal gas
enthalpy of component i is given by:
Z T
g,id
Hiid = Cp,i dT + HfTorm,i
0
+ Href,i (10.93)
T0

id
where Cp,i is the ideal gas heat capacity of component i, T0 is the enthalpy reference
temperature, HfTorm,i
0
the enthalpy of formation at T0 , and T is the actual temperature
at which the enthalpy is to be calculated. Note that there is no composition dependence
in the ideal gas enthalpy! The normal reference temperature (T0 ) in /cs/ is set to 298
K (see the ”References” menu under the properties selection).

190
The reference enthalpy Href depends on the model selected. If an no EOS model is
used then we need to correct for the state. If we select the vapor as reference state (the
default), the reference enthalpies are:
v
Href,i = 0 (10.94)
l
Href,i = −∆Hivap (T ) (10.95)
or, when we choose the liquid as reference state:
v
Href,i = ∆Hivap (T0 ) (10.96)
l
Href,i = ∆Hivap (T0 ) − ∆Hivap (T ) (10.97)
where the latent heat of vaporization (∆Hivap )
is computed with a temperature corre-
lation (see the section on physical properties below) or, if no correlation parameters
are known, using the method of Pitzer et al.:
∆Hivap (T ) = RTc,i (7.08(1 − Tr,i )0.354 + 10.95(1 − Tr,i )0.456 ωi ) (10.98)
If an EOS model is used for computing the enthalpies, we do not need to correct for
the state and consequently Href = 0.

The enthalpies of formation are normally not needed, as they normally cancel out.
These terms are large compared to the other contributions and make the enthalpy bal-
ance less accurate, so they can be ignored. Of course, when reactions are present they
need to be included:
Z T0
g,id
HfTorm,i
0
= Hf298K
orm,i + Cp,i dT (10.99)
298K

If an equation of state model is used for the K-values in ChemSep, the excess enthalpy
of both phases is calculated from an equation of state as follows:
 
ex 2 ∂ ln φi
Hi = −RT (10.100)
∂T
If an activity coefficient model is used for the liquid phase, then the component excess
enthalpy is computed from:
 
ex 2 ∂ ln γi
Hi = −RT (10.101)
∂T
ChemSep incorporates the following methods for estimating the enthalpy:

None No enthalpy balance is used in the calculations. WARNING: the use of this
model with subcooled and superheated feeds or for columns with heat addition
or removal on some of the stages will give incorrect results. The heat duties of
the condenser and reboiler will be reported as zero since there is no basis for
calculating them.

191
Ideal (B152) In this model the enthalpy is computed from the ideal gas contribution
and the excess enthalpy is assumed to be zero.

Excess (B518) This model includes the ideal enthalpy as above. The excess enthalpy
is calculated from the activity coefficient model or the temperature derivative of
the fugacity coefficients dependent on the choice of the model for the K-values,
and is added to the ideal part.

Polynomial vapor as well as liquid enthalpy are calculated as functions of the absolute
temperature (K). Both the enthalpies use the following function:

Hi = Ai + Bi T + Ci T 2 + Di T 3 (10.102)

You must enter the coefficients A through D in the ‘Load Data’ option of the
Properties menu for vapor and liquid enthalpy for each component.

EOS the enthalpies are always computed using a EOS model, even when the K-values
are computed with another model.

10.1.8 Entropy

The mixture entropy can be calculated from the component entropies in a similar
manner as the enthalpy:
Xc
S= zi Si (10.103)
i=1

where zi is the mole fraction of species i in the phase in question. The component
entropy may be expressed as the sum of the component entropy plus the contributions
of ideal and excess mixing entropy:

Si = Siid + Siex (10.104)

where Siex is the excess entropy and Siid is the ideal contribution for component i. The
ideal (gas) entropy of component i is given by:
T g,id
Cp,i
Z  
p
Siid = S0,i + dT − R ln − R ln zi + Sref,i (10.105)
T0 T p0
id
where Cp,i is the ideal gas heat capacity of component i, T0 is the reference tempera-
ture, p0 the reference pressure, S0,i the absolute entropy of component i at T0 and p0 ,
zi the mole fraction of i, and T and p are the actual temperature and pressure at which
the entropy is to be calculated. The entropy reference point is fixed as vapor at 298 K
and 1 atm because the absolute component entropies are defined at this temperature

192
and pressure. These component absolute entropies are stored in the pure component
database. The reference entropy is only added for liquids when we do not calculate
the entropy using an EOS; it corrects for the entropy of vaporization:

l ∆Hvap,i (Tb oil)

Sref,i =− (10.106)
Tboil

where ∆Hvap,i (Tb oil) is the vaporizations enthalpy of component i at Tboil , the com-
ponent normal boiling temperature. Excess mixture entropies are computed with
 
ex ∂ ln γi
Si = −R T + ln γi (10.107)
∂T

when an activity coefficient model is used. If an equation of state is used γi is replaced

by φi , the fugacity coefficient of i.

10.1.9 Exergy

The exergy (Ex) of a mixture or stream can be calculated from the enthalpy and en-
tropy with
Ex = H − Tsurr S (10.108)
where Tsurr is the surroundings temperature of the process (this temperature is nor-
mally set to 298 K but can be changed under the ”References” menu in the properties
selection). Note that with the inclusion of the enthalpy, the exergy in ChemSep is also
a relative quantity with respect to the enthalpy reference point.

Exergy is sometimes also referred to as ”availability”, it indicates the work that can
be extracted from a stream that is brought to equilibrium with its surrounding state.
Therefore, exergies provide useful insights in the thermodynamic efficiency of pro-
cesses. The second law of thermodynamics states that for any real, irreversal process
there is a nett entropy production (∆Sirr ), for which work (either direct or indirect in
the form of exergy) is lost. The lost work equals To ∆Sirr and the higher this amount
relative to the work/exergy input, the lower is the thermodynamic efficiency.

10.2 Physical Properties

These properties are not needed for equilibrium stage calculations but are required for
apparatus design and nonequilibrium simulations. They include: density, viscosity,
thermal conductivity, diffusivity, and surface tension.

193
 Table 10.1: Temperature correlations

Equation Parameter(s) Formula

number
2 A,B A + BT
3 A-C A + BT + CT 2
4 A-D A +BT + CT 2+ DT 3
B
10 A-C exp A − C+T
100 A-E A + BT + CT 2 + DT 3 + ET
4
101 A-E exp A + B T + C ln T + DT
E
B 2
102 A-D AT /(1 + C/T +
D/T )
103 A-D A + B exp − TCD
C106 D1016 E1018
104 A-E A+ B T + T3 + T8 + T9
D
105 A-D A/B (1+(1−T /C) )
2 3
106 A-E A(1 − Tr )(B+CTr +DTr +ETr )
107 A-E A + B( C C 2 D D 2
T / sinh( T )) + D( T / cosh( T ))
2 3 4
114 A-D AT + BT /2 + CT /3 + DT /4
117 A-E AT + B( C C E E
T )/ tanh( T ) − D( T )/ tanh( T )

Many different equations are available in ChemSep to evaluate physical properties

over a certain temperature range. These temperature correlations are assigned a unique
number in the range of 0-255 (see Table 10.1). For each up to 5 parameters (A-E) are
available. Table 10.2 shows which pure component properties can be modeled with
temperature correlations and their typical correlation number.

Any of the equations may be used for any of the physical properties but, of course,
some formulas were specifically developed for prediction of particular properties. Be-
sides the parameters A-E the temperature limits of the correlation must also be present.
If the temperature specified falls out of the temperature range of a correlation (or the
temperature limits are missing/incomplete) normally an alternative (default) method
will be used automatically.

Physical properties models can be selected manually or the automatic selection can
be used (this is the default). Below we discuss the models for calculating physical
properties for pure components and mixtures, for vapor and liquid phases. ChemSep
will make it’s own choices of model when none has been selected by the user (by
leaving the selection as a *). Depending on the range, phase, conditions, data avail-
ability, and required property ChemSep will guess the best model to use. ChemSep
does allow you to pick default models, and will use them if the model’s range is valid.

194
 Table 10.2: Component properties with the typical correlation number

Liquid density 105

vapor pressure 10 or 101
Heat of vaporisation 106
Liquid heat capacity 100
Ideal gas heat capacity 4 or 107
Second virial coefficient 104
Liquid viscosity 2 or 101
vapor viscosity 102
Liquid thermal conductivity 100
vapor thermal conductivity 102
Surface tension 106

Table 10.3: Default physical property correlations

Mixture liquid density Rackett

Component liquid density Polynomial
vapor density Cubic EOS
Mixture liquid viscosity Molar averaging
Component vapor viscosity Polynomial/Letsou-Stiel
Mixture vapor viscosity Brokaw
Component vapor viscosity Polynomial
Mixture liquid thermal conductivity Molar average
Component liquid thermal conductivity Polynomial
Mixture vapor thermal conductivity Molar average
Component vapor thermal conductivity Polynomial/9B-3
Liquid diffusivity Kooijman-Taylor/Wilke-Chang
vapor diffusivity Fuller et al.
Mixture surface tension Molar average
Component surface tension Polynomial
Liquid-liquid interfacial tension Jufu et al.

In case a property cannot be computed with a specific model it will use an estimation
method or a fixed estimate (it is a good habit to check predicted physical properties
when possible).

Certain methods require mixture (critical) properties, commonly used mixing rules

195
are:
c
X
Tc,m = xi Tc,i (10.109)
i=1
Xc
Vc,m = xi Vc,i (10.110)
i=1
Xc
Zc,m = xi Zc,i (10.111)
i=1
Pc,m = Zc,m RTc,m /Vc,m (10.112)
Xc
Mm = x i Mi (10.113)
i=1

These relations will be referred as the ‘normal’ or ‘conventional’ mixing rules. Re-
duced properties may be calculated from:

Tr = T /Tc (10.114)
Pr = P/Pc (10.115)
Vr = V /Vc (10.116)

unless specified otherwise.

10.2.1 Liquid density

Mixture liquid densities (in kmol/m3 ) may be estimated from:

Equation of State The Peng-Robinson equation of state can be used to calculate the
mixture compressibility directly from pure component critical properties and
mixture parameters, from which the density can be calculated easily. Use this
method if some components in the mixture are supercritical.

Amagat’s law
c
1 X xi
= (10.117)
ρL
m ρL
i=1 i

where the pure component liquid densities, ρL

i , are computed as discussed be-
low.

Rackett (A67,89) This method requires pure component critical temperatures, pres-
sures, molecular weights and Racket parameters (the critical compressibilities

196
 are used if ZR,i is not known):
c
X
Tc,m = xi Tc,i (10.118)
i=1
Xc
ZR,m = xi ZR,i (10.119)
i=1
T
Tr = (10.120)
Tc,m
(1+(1−Tr )2/7 )
Fz = ZR,m (10.121)
c
X xi Tc,i
A = (10.122)
i=1
M i Pc,i
c
X
ρL
m = 1/ARFz xi M i (10.123)
i=1

If the reduced temperature, Tr , is greater than unity a default value of 50 kmol/m3

is used.

Yen-Woods Mixture critical temperature, volume, and compressibility are calculated

with the ‘normal’ mixing rules. If the reduced temperature, Tr = T /Tc,m , is
greater than unity a default value of 50 kmol/m3 is used, otherwise the density
is calculated from:
T∗ = (1 − Tr )1/3 (10.124)
A = 17.4425 − 214.578Zc + 989.625 ∗ Zc2 − 1522.06Zc3 (10.125)
Zc ≤ 0.26 : B = −3.28257 + 13.6377Zc + 107.4844Zc2 − 384.211Zc3
(10.126)
Zc > 0.26 : B = 60.20901 − 402.063Zc + 501Zc2 + 641Zc3 (10.127)
1 + AT∗ + BT∗2 + (0.93 − B)T∗4
ρL
m = (10.128)
Vc
Hankinson-Thompson (A55-66,89,90) The methods of Hankinson and Thomson (AIChE
J, 25, 653, 1979) and Thomson et al. (AIChE J, 28, 671, 1982) are summarised
below:
c c c
!
1 X X X
Vm∗ = xi Vi∗ + 3( xi Vi∗ 2/3 )( xi Vi∗ 1/3 ) (10.129)
4 i=1 i=1 i=1

c X
X c
Tc,m = xi xj Vij∗ Tc,ij /Vm∗ (10.130)
i=i j=i

197
 c
X
ωSRK,m = xi ωSRK,i (10.131)
i=1
Zc,m = 0.291 − 0.08ωSRK,i (10.132)
Pc,m = Zc,m RTc,m /Vm∗ (10.133)
If the reduced temperature is larger than unity a default value of 50 kmol/m3
is used, otherwise the saturated liquid volume (Vs ) is calculated from:
Vs (0) (δ)
= VR (1 − ωSRK,m VR ) (10.134)
Vm∗
(0)
VR = 1 + a(1 − Tr )1/3 + b(1 − Tr )2/3 + c(1 − Tr ) + d(1 − Tr )4/3 (10.135)
(δ) e + f Tr + gTr2 + hTr3
VR = (10.136)
(Tr − 1.00001)
a=-1.52816 e=-0.296123
b= 1.43907 f= 0.386914
where
c=-0.81446 g=-0.0427258
d= 0.190454 h=-0.0480645
The molar density is the inverse of the liquid molar volume.
For the density of compressed liquids the saturated liquid volume is corrected
as follows (Thomson et al., AIChE J, 28, 671, 1982):
 
β+P
V = Vs 1 − c ln (10.137)
β + Pvpm

β/Pc = −1+a(1−Tr )1/3 +b(1−Tr )2/3 +d(1−Tr )+e(1−Tr )4/3 (10.138)

2
e = exp(f + gωSRK,m + hωSRK,m ) (10.139)
c = j + kωSRK (10.140)
a=-9.070217 g= 0.250047
b= 62.45326 h= 1.14188
where
d=-135.1102 j= 0.0861488
f= 4.79594 k= 0.0344483
with the vapor pressure from the generalized Riedel equations:
Pvpm = Pc,m Prm (10.141)
(0) (1)
log Prm = Prm + ωSRK,m Prm (10.142)
(0)
Prm = 5.8031817 log Trm + 0.07608141α (10.143)
(1)
Prm = 4.86601 log Trm + 0.03721754α (10.144)
6
α = 35 − 36/Trm − 96.736 log Trm + Trm (10.145)
Trm = T /Tc,m (10.146)

198
 This method should be used for reduced temperatures from 0.25 up to the criti-
cal point.

Pure component liquid densities are computed from the Peng-Robinson EOS for tem-
peratures above a components critical temperature, otherwise by one of the following
methods:

Polynomial When within the temperature range, a polynomial is the default way for
calculating component liquid densities.

Rackett
(1+(1−Tr )2/7 )
Fz = ZR (10.147)
ρL
m = Pc /RTc Fz (10.148)

COSTALD The Hankinson and Thompson method described as above but with pure
component parameters.

The pure component liquid densities are corrected for pressure effects with the correc-
tion of Thomson et al. (1982) as described for the Hankinson and Thompson method
for mixtures.

10.2.2 Vapor density

Vapor densities are computed with the equation of state selected for the thermody-
namic properties (possible selections are Ideal gas EOS, Virial EOS, and Cubic EOS).

10.2.3 Liquid Heat Capacity

The mixture liquid heat capacity is the molar average of the component liquid heat
capacities, that often are computed from a temperature correlation. Alternatively the
liquid heat capacity could be computed from a corresponding states method and the
ideal gas capacity. ChemSep uses a temperature correlation for all temperatures to
prevent problems arising from using different liquid heat capacity methods in the same
column (that especially trouble nonequilibrium models). Liquid heat capacities could
also be computed from selected thermodynamic models to circumvent this problem.

199
10.2.4 Vapor Heat Capacity

The mixture vapor heat capacity is the molar average of the component vapor heat
capacities, that are computed from the ideal gas heat capacity 4 parameter polynomial
(A657). If no parameters for this correlation are present, the vapor heat capacity
temperature correlation is used (if within the temperature range).

10.2.5 Liquid Viscosity

Mixture liquid viscosity are computed from the pure component liquid viscosities:
c
X
L
ln ηm = zi ln ηiL (10.149)
i=1

where zi are either the mole fractions (for molar averaging, the default) or, alterna-
tively, the weight fractions for mass averaging. A better method is due to Teja and
Rice (1981, A479). However, this method requires interaction parameters. Here,
a different mixing rule (for Tcij Vcij ) is used which improves the model predictions
with unity interaction coefficients:
c
X
ωm = xi ωi (10.150)
i=1
Xc
Mm = x i Mi (10.151)
i=1
Xc X c
Vcm = xi xj Vcij (10.152)
i=1 j=1
 3
1/3 1/3
Vci + Vcj
Vcij = (10.153)
Pc P8c
i=1 j=1 xi xj Tcij Vcij
Tcm = (10.154)
Vcm
Tci Vci + Tcj Vcj
Tcij Vcij = ψij (10.155)
2
where ψij is set to unity for all components. The liquid viscosity of the mixture is
computed from two reference components
 
ωm − ω1
ln(m ηm ) = ln(1 η1 ) + (ln(2 η2 ) − ln(1 η1 )) (10.156)
ω2 − ω1

200
with  defined as
2/3
V
i = √ ci (10.157)
Tci Mi
and the reference component vioscosities are evaluated at T.Tci /Tcm . Component
liquid viscosities are calculated from the liquid viscosity temperature correlation if the
temperature is within the valid range. Otherwise the component viscosity is computed
from the Letsou-Stiel method (1973, see A471):
1/6
2173.424Tc,i
ξ = √ 2/3
(10.158)
Mi Pc,i
ξ (0) = (1.5174 − 2.135Tr + 0.75Tr2 )10−5 (10.159)
ξ (1)
= (4.2552 − 7.674Tr + 3.4Tr2 )10−5 (10.160)
ηiL = (ξ (0)
+ ωξ (1)
)/ξ (10.161)

Alternatively, the simple temperature correlation given in Reid et al. (A439) can be
used:
log η = A + B/T (10.162)
A high pressure correction by Lucas (A436) is used to correct for the influence of the
pressure on the liquid viscosity:

1 + D(∆Pr /2.118)A
η= ηSL (10.163)
1 + Cωi ∆Pr
where ηSL is the viscosity of the saturated liquid at Pvp , and

∆Pr = (P − Pvp )/Pci (10.164)

A = 0.9991 − 4.674 × 10−4 /(1.0523Tr−0.03877 − 1.0513)

(10.165)

C = −0.07921 + 2.1616Tr − 13.4040Tr2 + 44.1706Tr3 (10.166)

−84.8291Tr4 + 96.1209Tr5 + − 59.8127Tr6 15.6719Tr7

D = 0.3257/(1.0039 − Tr2.573 )0.2906 − 0.2086

(10.167)

10.2.6 Vapor Viscosity

Mixture vapor viscosities are computed from component viscosites as follows:

c
x ηL
Pi i
X
L
ηm = (10.168)
i=1
xi φij

201
where the interaction parameters φij can be calculated by Wilke’s method (1950):
p
1 + ηi /ηj (Mi /Mj )1/4 )2
φij = ( p (10.169)
8(1 + Mi /Mj )

or by Brokaw’s method: q
φij = SA ηi /ηj (10.170)
 1/4
4
sm = (10.171)
(1 + Mj /Mi )(1 + Mi /Mj )
(Mi /Mj − (Mi /Mj )0.45 ) (1 + (Mi /Mj )0.45 )
 
sm
A= p 1+ +√
Mi /Mj 2(1 + Mi /Mj ) sm(1 + Mi /Mj )
(10.172)
If the Lennard-Jones energy parameter,  (in Kelvin), and the Stockmayer polar pa-
rameter, δ, are known, S is calculated from:
p
1 + (T /i )(T /j ) + δi δj /4
S=p q (10.173)
1 + T /i + δi2 /4 1 + T /j + δj2 /4

otherwise it is approximated by S = 1.  and δ can be estimated from:

3.6
 = 65.3Tc,i Zc,i (10.174)
2
µ
δ = 1.744 × 105 9 (10.175)
Vb,i Tb,i

where µ is the dipole moment in Debye. Vapor viscosities are a function of pres-
sure and a correction normally is applied. Mixture properties are computed using the
‘normal’ mixing rules. The high pressure viscosity can be obtained from:

ρc = 1/Vc,m (10.176)
ρr = ρ/ρc (10.177)
p
1/6 2/3
ξ = 2173.4241Tc,m / Mm Pc,m (10.178)
A = exp(1.4439ρr ) − exp(−1.111ρ1.85
r ) (10.179)
−7
B = 1.08 × 10 A/ξ (10.180)
ηhp = η+B (10.181)

where ρ is the vapor mixture molar density.

Both the Wilke and Brokaw methods require pure component viscosities. These are
normally obtained from the vapor viscosity temperature correlations, as long as the
temperature is within the valid temperature range. If not, then the viscosity can be

202
 Table 10.4: Constants for the Yoon-Thodos method

Hydrogen Helium Others

a=47.65 a=52.57 a=46.1
b=0.657 b=0.656 b=0.618
c=20.0 c=18.9 c=20.4
d=-0.858 d=-1.144 d=-0.449
e=19.0 e=17.9 e=19.4
f=-3.995 f=-5.182 f=-4.058

computed with the Chapman-Enskog kinetic theory (see Hirschfelder et al. 1954 and
A391-393):
T∗ = T / (10.182)
∗ −b ∗ ∗
Ωv = a(T ) + c/ exp(dT ) + e/ exp(f T ) (10.183)
V − 2 2 ∗
η = 26.69 × 10 7M T /σ (Ωv + 0.2δ /T ) (10.184)
where the collision integral constants are a = 1.16145, b = 0.14874, c = 0.52487,
d = 0.77320, e = 2.16178, and f = 2.43787. The viscosity may also be computed
with the Yoon and Thodos method:
1/6 2/3
p
ξi = 2173.4241Tc,i / Mi Pc,i (10.185)
1 + aTrb − c exp(dT − r) + e exp(f Tr )
ηiV = (10.186)
108 ξ
where the constants a − f are given in Table 10.4.

Another method for calculating the vapor viscosity is the Lucas (A397) method:
η = 10−7 (0.807Tr0.618 − 0.357 exp(−0.449Tr ) + (10.187)
0.340 exp(−4.058Tr ) + 0.018)Fpo Fqo /ξ (10.188)
 1/6
Tc
ξ = 0.176 (10.189)
M 3 (10−5 Pc )4
where Fpo and Fqo are polarity and quantum correction factors. The polarity correction
depends on the reduced dipole moment:
(µ/3.336 × 10−30 )2 (10−5 Pc )
µr = 52.46 (10.190)
Tc2
If µr is smaller than 0.022 then the correction factor is unity; if it is smaller than 0.075
it is given by
Fpo = 1 + 30.55(0.292 − Zc )1.72 (10.191)

203
else by

Fpo = 1 + 30.55(0.292 − Zc )1.72 (0.96 + 0.1(Tr − 0.7)) (10.192)

The quantum correction is only used for quantum gases He, H2 , and D2 ,

1/M 
Fqo = 1.22Q0.15 1 + 0.00385 Tr − 12)2 sign(Tr − 12) (10.193)

where Q = 1.38 (He), Q = 0.76 (H2 ), Q = 0.52 (D2 ). There is also a specific
correction for high pressures (A421) by Lucas.

η = Y F p Fq η o (10.194)

aPre
Y =1+ (10.195)
bPrf + (1 + cPrd )−1
1 + (Fpo − 1)Y −3
Fp = (10.196)
Fpo

1 + (Fqo − 1) Y −1 − 0.007(ln Y )4


Fq = (10.197)
Fqo
where η o refers to the low-pressure viscosity (note that the original Lucas method has
a different rule for Y if Tr is below unity, however, this introduces a discontinuity that
is avoided here). The parameters a through f are evaluated with:

1.245 × 10−3
a = exp 5.1726Tr−0.3286 (10.198)
Tr
b = a(1.6553Tr − 1.2723) (10.199)
0.4489
c = exp 3.0578Tr−37.7332 (10.200)
Tr
1.7368
d = exp 2.2310Tr−7.6351 (10.201)
Tr
e = 1.3088 (10.202)
f = 0.9425 exp −0.1853Tr0.4489 (10.203)

where, in case Tr is below unity, Tr is taken to be unity. For mixtures the Lucas model
uses the following mixing rules:
c
X
Tcm = yi Tci (10.204)
i=1
Xc
Vcm = yi Vci (10.205)
i=1

204
 c
X
Zcm = yi Zci (10.206)
i=1
Pcm = RTcm Zcm /Vcm (10.207)
Xc
Mm = yi Mi (10.208)
i=1
Xc
o o
Fpm = yi Fpi (10.209)
i=1
c
X
o o
Fqm = A yi Fqi (10.210)
i=1

where A is a correction factor depending on the molecular weights of the components

in the mixture. Let H denote the component of highest molecular weight and L of
lowest, then if MH /ML > 9:
A = 1 − 0.01(MH /ML )0.57 (10.211)
else A = 1. The mixture vapor viscosity is computed with the Lucas method as for a
o o
component that has the mixture properties Tcm , Pcm , Mm , Fpm , and Fqm . Therefore,
the method is not interpolative in the same way as are the techniques of Wilke and
Brokaw (that is, the method does not necessarily lead to pure component viscosity ηi
when all yj = 0 except yi = 1).

10.2.7 Liquid Thermal Conductivity

The mixture liquid thermal conductivity, λL

m (W/m K), can be computed using the
following methods from the component liquid thermal conductivities:

Molar average This is the default method (and the simplest):

c
X
λL
m = xi λL
i (10.212)
i=1

DIPPR procedure 9I
c
X
Fv,i = xi / xi /ρL
i (10.213)
i=1
λij = 2/(1/λi + 1/λj ) (10.214)
Xc X c
λL
m = Fv,i Fv,j λij (10.215)
i=1 j=1

205
DIPPR procedure 9H
c
1 X wi
= (10.216)
(λL
p 2
λL
m i=1 i )

where wi is the weight fraction of component i.

A correction is applied for high pressures:

λhp = 0.63Tr1.2 Pr /(30 + Pr ) + 0.98 + 0.0079Pr Tr1.4 λ

(10.217)

The mixture parameters are computed by the ‘normal’ mixing rules. Pure component
liquid thermal conductivities are calculated from one of the following methods:

Polynomial The temperature correlation normally is used as long as the temperature

is in the valid range and no other method is explicitly selected.

Pachaiyappan et al.

f = 3 + 20(1 − Tr )2/3 (10.218)

2/3
b = 3 + 20(1 − 273.15/Tc,i ) (10.219)
−4
λi = c10 Mix ρL
i (f /b) (10.220)

for straight chain hydrocarbons c = 1.811 and x = 1.001 else c = 4.407 and
x = 0.7717.

Latini et al. (see A549,550):

A(1 − Tr )0.38
λL
i = 1/6
(10.221)
Tr
A∗ Tbα
A = (10.222)
Miβ Tcγ

where parameters A∗ , α, β, and γ depend on the class of the component as

shown in Table 10.5.

10.2.8 Vapor Thermal Conductivity

Molar average The mixture vapor thermal conductivity is computed from the pure
component thermal conductivities as follows:
c
X
λVm = xi λVi (10.223)
i=1

206
 Table 10.5: Parameters for the Latini equation for liquid thermal conductivity

Family A∗ α β γ
Saturated hydrocarbons 0.0035 1.2 0.5 0.167
Olefins 0.0361 1.2 1.0 0.167
Cycloparaffins 0.0310 1.2 1.0 0.167
Aromatics 0.0346 1.2 1.0 0.167
Alcohols, phenols 0.00339 1.2 0.5 0.167
Acids (organic) 0.00319 1.2 0.5 0.167
Ketones 0.00383 1.2 0.5 0.167
Esters 0.0415 1.2 1.0 0.167
Ethers 0.0385 1.2 1.0 0.167
Refrigerants:
R20, R21, R22, R23 0.562 0.0 0.5 -0.167
Others 0.494 0.0 0.5 -0.167

Kinetic theory This is DIPPR procedure 9D:

c
x λV
Pc i i
X
λVm = (10.224)
i=1 j=1 xj φij

where interaction parameters φij are computed from:

s
ηi Mj 3/4 T + 1.5Tb,i 2 T + 1.52 Tb,i Tb,j
p
φij = 0.25(1 + ) (10.225)
ηj M i T + 1.5Tb,j T + 1.5Tb,i

Note that the component viscosities are required for this evaluation.

If the system pressure is larger than 1 atmosphere a corection is applied. Mixture pa-
rameters are computed using the ‘normal’ mixing rules. Critical and reduced densities
are computed from:

1
ρc = (10.226)
Vc,m
ρ
ρr = (10.227)
ρc

If the reduced density is below 0.5 then a = 2.702, b = 0.535, and c = −1; if the
reduced density is within [0.5, 2] then a = 2.528, b = 0.67, and c = −1.069; other-
wise a = 0.574, b = 1.155, and c = 2.016. The high pressure thermal conductivity

207
correction then is calculated from:

a10−8 (exp(bρr ) + c)
∆λ =  √ 1/6
 (10.228)
Mm Tc,m 5
2/3 Zc,m
Pc,m

and added to the thermal conductivity calculated at low pressure.

Pure component vapor thermal conductivities are estimated from the following meth-
ods:

Polynomial The temperature correlation is normally used as long as the temperature

is in the valid range and no other method is explicitly selected.

DIPPR procedure 9B-3 This method is the default in case the temperature is out of
the range of the temperature correlation:

λVi = (1.15(Cp − R) + 16903.36)ηiV /Mi (10.229)

DIPPR procedure 9B-2 This method is recommended for linear molecules:

λVi = (1.3(Cp − R) + 14644 − 2928.8/Tr )ηiV /Mi (10.230)

DIPPR procedure 9B-1 This method is suitable for monatomic gases only:

λVi = 2.5(Cp − R)ηiV /Mi (10.231)

Misic-Thodos 2 This method is used for methane and cyclic compounds below Tr =
1:
1/6
2173.424Tc,i
ξ = √ 2/3
(10.232)
Mi Pc,i
λi = 4.91 × 10−7 Tr Cp /ξ (10.233)

Misic-Thodos 1 This is the Misic-Thodos method for all other compounds:

1/6
2173.424Tc,i
ξ = √ 2/3
(10.234)
Mi Pc,i
λi = 11.05 × 10−8 (14.52Tr − 5.14)1/6 Cp /ξ (10.235)

208
10.2.9 Liquid Diffusivity

Generalized Maxwell-Stefan binary diffusion coefficients−

D ij are computed from the
Kooijman-Taylor (1991) correlation
c
Y
−
D ij = − kij )xk
(D (10.236)
k=1

o
where the Dij are the diffusion coefficients at infinite dilution and where

D kij
− = o
Dij , k=j (10.237)
D kij
− = o
Dji , k=i (10.238)
q
D kij
− = Dik o Do ,
jk k 6= i, k 6= j (10.239)

k
An alternative is the method of Wesselingh and Krishna (1990) where Dij is calcu-
lated from: q
−D kij = Dij
o Do ,
ji k=6 i, k 6= j (10.240)
o
Liquid binary diffusion coefficients at infinite dilution (Dij ) normally are computed
by the Wilke-Chang method unless selected otherwise. The following models are
available:

Wilke-Chang The Wilke and Chang (1955, see A598) method is

√
o −16 φb Mb T
Dab = 1.1728 × 10 0.6 (10.241)
ηb Vb,a

where φb is the association factor for the solvent (2.26 for water, 1.9 for methanol,
1.5 for ethanol and 1.0 for unassociated solvents).

Hayduk-Laudie The Hayduk and Laudie (1974) method is

−1.14 −0.589
o
Daw = 8.62 × 10−14 ηw Vb,a (10.242)

Hayduk-Minhas aqueous Estimates diffusivity of solute a in water, proposed by Hay-

duk and Minhas (1982, see also A602):
−0.19
o
Daw = (3.36Vb,a −3.65)10−13 (1000ηw )(0.00958/Vb,a −1.12) T 1.52 (10.243)

Hayduk-Minhas non-aqueous Estimates diffusivity of solute a in polar and non-polar

solvent b (not water), proposed by Hayduk and Minhas (1982, see also A603):
o
Dab = 4.3637 × 10−18 ηb−0.19 ra0.2 rb−0.4 T 1.7 (10.244)

209
Hayduk-Minhas paraffins Estimates diffusion coefficients for mixtures of normal paraf-
fins from Hayduk-Minhas correlation (equation 7 of their paper) as corrected by
Siddiqi and Lucas (1986, see also A602):
−071
o
Dab = 9.859 × 10−14 Vb,a (1000ηb )(0.0102/Vb,a −0791) (10.245)

Siddiqi-Lucas aqueous Estimates diffusivity of solute a in water, proposed by Siddiqi

and Lucas (1986):
−0.5473 −1.026
o
Daw = 5.6795 × 10−16 Vb,a ηw T (10.246)

Siddiqi-Lucas Estimates diffusivity of solute a in solvent b (not water), proposed by

Siddiqi and Lucas (1986):
−0.45 0.265 −0.907
o
Dab = 5.2383 × 10−15 Vb,a Vb,B ηb T (10.247)

Umesi-Danner Estimates diffusivity of solute a in solvent b:

T rb
o
Dab = 5.927 × 10−12 2/3
(10.248)
ηb ra

Tyn-Calus correlation Estimates diffusivity of solute a in solvent b (see A600):

!1/6  
0.6
o −12 Vb,a Pb T
Dab = 8.93 × 10 2 (10.249)
Vb,B Pa ηb

where Pi is the Parachor of the pure component (at TnB ).

Modified Tyn-Calus correlation This method uses the above correlation where the
Parachor is calculated from:
1/4
Pi = Vb,i σi (10.250)

The surface tension is computed with the Brock and Bird corresponding states
method. This makes the Parachor temperature dependent, in contrast to the
regular Tyn-Calus method.

10.2.10 Vapor Diffusivity

Generalized Maxwell-Stefan binary diffusion coefficients Dij are equal to the normal
binary diffusion coefficients (since the gas is considered an ideal system for which the
thermodynamic matrix is the identity matrix). Normally these are computed using the
Fuller-Schettler-Giddings method (see A587) but, if the Fuller volume parameters are
missing, the Wilke-Lee modification of the Chapman-Enskog kinetic theory is used.

210
 Table 10.6: Fuller diffusion volumes

Atomic and structural diffusion volume increments

C 15.9 F 14.7
H 2.31 Cl 21.0
O 6.11 Br 21.9
N 4.54 I 29.8
Ring -18.3 S 22.9
Diffusion volumes of simple molecules
He 2.67 CO 18.0
Ne 5.98 CO2 26.9
Ar 16.2 N2O 35.9
Kr 24.5 NH3 20.7
Xe 32.7 H2O 13.1
H2 6.12 SF6 71.3
D2 6.84 Cl2 38.2
N2 18.5 Br2 69.0
O2 16.3 SO2 41.8
Air 19.7

Fuller et al. This method was developed by Fuller et al. (1966,1969):

1.75
p
V −2 T (1/Ma + 1/Mb )
Dab = 1.013 × 10 √ √ (10.251)
P ( Va + 3 Vb )2
3

where Va and Vb are the Fuller molecular diffusion volumes, calculated by sum-
ming the atomic contributions from Table 10.6. This table also includes the
diffusion volumes for some simple molecules.
Chapman-Enskog kinetic theory The binary diffusion coefficient in gas/vapor mix-
tures may be estimated from a simplified kinetic theory.

V T 3/2 sqrt1/Ma + 1/Mb

Dab =C (10.252)
P σab ΩD
where constant C = 1.883 × 10−2 .
The average collision diameter and energy parameter are:
σab = (σa + σb )/2 (10.253)
√
ab = a + b (10.254)
The diffusion collision integral is
ΩD = a(T ∗ )−b + c/ exp(dT ∗ ) + e/ exp(f T ∗ ) + g/ exp(hT ∗ ) (10.255)

211
 where
T ∗ = T /ab (10.256)
The collision integral constants are a = 1.06036, b = 0.1561, c = 0.193,
d = 0.47635, e = 1.03587, f = 1.52996, g = 1.76474, and h = 3.89411. If
the Stockmayer polar parameters are known the integral must be corrected as
follows:
0.19δa δb
ΩD,c = ΩD + (10.257)
T∗
Wilke-Lee Wilke and Lee (1955, see A587) proposed a modified version of the kinetic
theory method described above with
p
C = 2.1987 × 10−2 − 5.07 × 10−3 1/Ma + 1/Mb (10.258)

Saksena-Saxena Saksena and Saxena, Ind.Eng.Chem.Fundam., Vol. 11 (1972) p. 420

p
Tc,ab = Tc,a Tc,b (10.259)
Tc,ab
Dij = 1.2246DF uller /(1 + 0.4364 )/10.1325/T 0.25(10.260)
T
Nain-Ferron Nain and Ferron, Ind.Eng.Chem.Fundam., Vol. 11 (1972) p. 420.
p
Tc,ab = Tc,a Tc,b (10.261)
Tc,ab
Dij = DF uller /(1 + 0.1759 ) (10.262)
T
Slattery-Bird Slattery and Bird, AIChE J. 4 (1958) p. 137-142.

1
(1 × 10−10 pc,a pc,b 3
pt = (10.263)
1 × 10−5 p
tt = (Tc,a Tc,b )( 5/12 − 1.823/2) (10.264)
1 1
fm = + (10.265)
Ma Mb
p
Dij = 2.745 × 10−8 T 1.823 pt tt fm (10.266)

10.2.11 Surface Tension

Mixture surface tension for VLE are computed with:

Molar avarage This is the default method:

c
X
σm = xi σi (10.267)
i=1

212
Winterfeld et al. This method, by Winterfeld et al. (1978), is DIPPR Procedure 7C:

√
Pc  Pc 
i=1 (xi /ρL 2
i ) + j=1 (xi xj σi σj /ρL L
i ρj )
σm = (10.268)
L 2
Pc

i=1 (xi /ρi )

Digulio-Teja This method evaluates the component surface tensions at the compo-
nents normal boiling points (σb,i ) and computes the mixture critical temper-
ature, normal boiling temperature and the mixture surface tension at normal
boiling temperature with the following mixing rules:
c
X
Tc,m = xi Tc,i (10.269)
i=1
Xc
Tb,m = xi Tb,i (10.270)
i=1
Xc
σb,m = xi σb,i (10.271)
i=1

Then it corrects the σb , m with:

(1/Tr − 1)
T∗ = (10.272)
(1/Trb − 1)
T
σ = 1.002855(T ∗ )1.118091 σr (10.273)
Tb

Pure component surface tensions are determined only for temperatures below the com-
ponent’s critical temperature, otherwise it is assumed that the component does not
contribute to the mixture surface tension (i.e. σi = 0). The following methods are
available:

Polynomial The temperature correlation is normally used as long as the temperature

is in the valid range and no other method is explicitly selected.

Brock-Bird

Tbr = Tb,i /Tc,i (10.274)

 
Tbr (ln(Pc,i ) − 11.526)
Q = 0.1207 1 + − 0.281 (10.275)
(1 − Tr )
2/3 1/3
σi = 4.6 × 10−7 Pc,i Tc,i Q(1 − Tr )11/9 (10.276)

213
Lielmezs-Herrick This method by Lielmezs and Herrick (1986) uses a polynomial but
evaluates it at the reduced normal boiling temperature and corrects the resulting
σr with:

(1/Tr − 1)
T∗ = (10.277)
(1/Trb − 1)
T
σ = 1.002855(T ∗ )1.118091 σr (10.278)
Tb

surface tensionSastri-Rao Sastri and Rao (Chem. Eng. J., 59 (1996) pp. 181-186

Tb oil
Tb,r = (10.279)
Tc
  11
(Tc − T ) 9
Tm = (10.280)
(TC − Tb )
p −1.85 0
σ = 0.158 × 10−3 10−5 pc Tb,r Tb .35Tm (10.281)

10.2.12 Liquid-Liquid Interfacial Tension

This property is required only for simulating liquid-liquid extraction operations with
the nonequilibrium model. The calculated surface tensions for both liquid phases and
the interfacial tension, σ 0 , is computed from
√
σ 0 = σ1 + σ2 − 1.1 σ1 σ2 (10.282)

This method generally over-predicts the interfacial tension for aqueous systems. We
use a general method from Jufu et al. (1986):

X = − ln(x001 + x02 + x3r ) (10.283)

KRT X
σ0 = (10.284)
Aw0 exp(X)(x001 q1 + x02 q2 + x3r q3 )

with Aw0 = 2.5 × 105 (m2 /mol), R = 8.3144 (J/mol/K), K = 0.9414, and qi
is the UNIQUAC surface area parameters of the components i. The components are
ordered in such a manner that component 1 and 2 are the dominating components in
the two liquid phases. Then the rest of the components are lumped into one mole
fraction, x3 . This lumped mole fraction is taken for the phase that has the largest x3
(the richest). q3 is the molar averaged q for that phase for all components except 1 and
2.

214
Symbol List

Latin Symbols

a, b cubic EOS parameters

B second virial coefficient (m3/kmol)
c number of components
Cp mass heat capacity (J/kg/K)
Dij binary diffusion coefficient (m2 /s)
−
D ij binary Maxwell-Stafan diffusion coefficient (m2 /s)
o
Dij infinite dilution binary diffusion coefficient (m2 /s)
Ki K-value of component i, equilibrium ratio (Ki = yi /xi )
kij binary interaction coefficient (for EOS)
M molecular mass (kg/kmol)
R gas constant = 8134 (J/kmol/K)
r radius of gyration (Angstrom)
P pressure (P a)
P ∗ , Pvap vapor pressure (P a)
Pi Parachor (m3 kg 1/4 /s1/2 ) of component i
PF Poynting correction
q UNIQUAC surface area parameter
T temperature (K)
Tr reduced temperature (Tr = T /Tc )
Tb normal boiling temperature (K)
V molar volume (m3 /kmol)
Vb molar volume at normal boiling point (m3 /kmol)
Vs saturated molar volume (m3 /kmol)
w weight fraction (of component)
x liquid mole fraction (of component)
y vapor mole fraction (of component)
Z compressibility
ZR Racket parameter

Greek Symbols

α attractive parameter in EOS

ω acentric factor
Ωa , Ωb EOS parameters

215
Ωv vollision integral for viscosity
ΩD collision integral for diffusion
γ activity coefficient
δ Stockmayer parameter
 molecular energy parameter (K)
λ thermal conductivity (W/m/K)
ρ molar density (kmol/m3 )
η viscosity (P a.s)
φi fugacity coefficient of component i
φs association factor for solvent s (Hayduk-Laudie)
φi j interaction parameter for viscosities
φi fugacity coefficient of component i
φ∗ pure fugacity coefficient at saturation
σ surface tension (N/m)
collision diameter (Angstrom)
σb surface tension at Tb (N/m)
σ0 liquid-liquid interfacial tension (N/m)
µ dipole moment (Debye)
ξ inverse viscosity (defined in text)

Superscripts

L liquid
V,G vapor, gas
∗ saturated liquid,
T /

Subscripts

0 reference
b at normal boiling point
c critical
i of component i
j of component j
m mixture
r reduced
s saturated liquid

216
Abbreviations

EOS equation of state

RK Redlich-Kwong
SRK Soave Redlich-Kwong
PR Peng-Robinson

References

Wikipedia contains several useful pages:

http://en.wikipedia.org/wiki/Debye-H%C3%BCckel_equation
http://en.wikipedia.org/wiki/Activity_%28chemistry%29
http://en.wikipedia.org/wiki/Fugacity
http://en.wikipedia.org/wiki/Chemical_potential

R.P. Danner and T.E. Daubert, Manual for Predicting Chemical Process Design Data,
AIChE, New York (1983).

R. Digulio, A.S. Teja, “Correlation and Prediction of the Surface Tensions of Mix-
tures”, Chem. Eng. J., Vol. 38 (1988) pp. 205–208.

E.N. Fuller, K. Ensley, J.C. Giddings, “A New Method for Prediction of Binary Gas-
Phase Diffusion Coefficients”, Ind. Eng. Chem., Vol. 58 (1966) pp. 19–27.

E.N. Fuller, P.D. Schettler, J.C. Giddings, “Diffusion of Halogenated Hydrocarbons

in Helium. The Effect of Structure on Collision Cross sections”, J. Phys. Chem., Vol.
73 (1969) pp. 3679–3685.

W. Hayduk, H. Laudie, “Prediction of Diffusion Coefficients for Nonelectrolytes in

Dilute Aqueous Solutions”, AIChE J., Vol. 20 (1974) pp. 611–615.

W. Hayduk, B.S. Minhas, “Correlations for Prediction of Molecular Diffusivities in

Liquids”, Can. J. Chem. Eng., 60, 295-299 (1982); Correction, Vol. 61 (1983) p. 132.

217
J.O. Hirschfelder, C.F. Curtis, R.B. Bird, Molecular Theory of Gases and Liquids,
Wiley, New York (1954).

F. Jufu, L. Buqiang, W. Zihao, “Estimation of Fluid-Fluid Interfacial Tensions of Mul-

ticomponent Mixtures”, Chem. Eng. Sci., Vol. 41, No. 10 (1986) pp. 2673–2679.

H.A. Kooijman, R. Taylor, “On the estimation of Diffusion Coefficients in Multicom-

ponent Liquid Systems”, Ind. Eng. Chem. Res., Vol. 30, No. 6 (1991) pp. 1217–1222.

A. Letsou, L.I. Stiel, “Viscosity of Saturated Nonpolar Liquids at Elevated Pressure”,

AIChE J., Vol. 19 (1973) pp. 409–414.

J. Lielmezs, T.A. Herrick, “New Surface Tension Correlation for Liquids”, Chem.
Eng. J., Vol. 32 (1986) pp. 165-169.

J.M. Prausnitz, T. Anderson, E. Grens, C. Eckert, R. Hsieh, J. O’Connell, Com-

puter Calculations for Multicomponent Vapor-Liquid and Liquid-Liquid Equilibria,
Prentice-Hall (1980).

J.S. Rowlinson, Liquids and Liquid Mixtures, 2nd Ed., Butterworth, London (1969).

R.C. Reid, J.M. Prausnitz, T.K. Sherwood, Properties of Gases and Liquids, 3rd Ed.,
McGraw-Hill, New York (1977).

R.C. Reid, J.M. Prausnitz and B.E. Poling, The Properties of Gases and Liquids, 4th
Ed., McGraw-Hill, New York (1988).

M.A. Siddiqi, K. Lucas, “Correlations for Prediction of Diffusion in Liquids”, Can. J.

Chem. Eng., Vol. 64 (1986) pp. 839–843.

R. Taylor, H.A. Kooijman, “Composition Derivatives of Activity Coefficient Models

(For the estimation of Thermodynamic Factors in Diffusion)”, Chem. Eng. Comm.,
Vol. 102 (1991) pp. 87–106.

A.S. Teja, P. Rice, “Generalized Corresponding States Method for the Viscosities of
Liquid Mixtures”, Ind. Eng. Chem. Fundam., Vol. 20 (1981) pp. 77–81.

S.M. Walas, Phase Equilibria in Chemical Engineering, Butterworth Publishers, Lon-

don (1985).

C.R. Wilke, “A Viscosity Equation for gas Mixtures”, J. Chem. Phys., Vol. 18 (1950)
pp. 517–519.

218
C.R. Wilke, P. Chang, “Correlation of Diffusion Coefficients in Dilute Solutions”,
AIChE J., Vol. 1 (1955) pp. 264-270.

C.R. Wilke, C.Y. Lee, “Estimation of Diffusion Coefficients for Gases and Vapors”,
Ind. Eng. Chem., Vol. 47 (1955) pp. 1253–1257.

P.H. Winterfeld, L.E. Scriven, H.T. Davis, “An Approximate Theory of Interfacial
Tensions of Multicomponent Systems: Applications to Binary Liquid-Vapor Ten-
sions”, AIChE J., Vol. 24 (1978) pp. 1010–1014.

219
220
Chapter 11

Numerical Solution of Systems

of Equations

In this chapter we discuss the numerical methods used in ChemSep to solve the non-
linear equations that model separation processes. For more in-depth treatment of nu-
merical methods see, for example, Ortega and Rheinbolt (1970), Pozrikidis (1998),
and Garcia and Zangwill (1981).

11.1 Repeated Substitution

To use repeated substitution the equations to be solved must be expressed in the form

G(x) = x (11.1)

where G is a vector consisting of all the equations to be solved and x is the vector of
variables. The procedure for solving the equations is summarized in Table 11.1.

11.2 Newton’s Method

Newton’s method is one of the most versatile and robust methods for solving sys-
tems of nonlinear equations. It is, in fact, the method of choice in ChemSep. To use

221
 Table 11.1: Repeated Substitution

1. Set iteration counter, k, to zero. Estimate xk

2. Compute G(xk )

3. Set xk + 1 = G(xk )

4. If not converged, increment k and return to step 2

Newton’s method, the equations to be solved are written in the form

F(x) = 0 (11.2)
where F is a vector consisting of all the equations to be solved and x is, again, the
vector of variables. A Taylor series expansion of the function vector around the point
x0 at which the functions are evaluated gives
F(x) = F(x0 ) = J(x − x0 ) + O(|x − x0 | +) (11.3)
where J is the Jacobian matrix of partial derivatives of F with respect to the indepen-
dent variables x:
∂Fi
Jij = (11.4)
∂xj
If x is the actual solution to the system of equations, then F(x) = 0. Additionally, if
the O(x − x0 ) term is neglected then equation (11.3) can be rearranged to give
J(x − x0 ) = −F(x0 ) (11.5)
Here the only unknown is the value of the vector x. Thus, by solving the linear
system of equations (11.5) it is possible to obtain a value for this vector. If the new
vector, x, obtained in this way does not actually satisfy the set of equations, F, then
the procedure can be repeated using the just calculated x as a new x0 . The entire
procedure is summarized in Table 11.2.

Newton’s method works extremely well if the initial estimate of the solution vector
is ”near” the actual solution. Outside this ”near” region, the method may converge
slowly, oscillate, or diverge. The question to which there is no simple answer is, ”What
exactly is it near?”. Having said this it must also be stated that Newton’s method is a
much more reliable procedure than repeated substitution.

An important issue not addressed here is how to tell when the problem has been solved;
that is, when convergence has been obtained. A proper discussion of this topic is
beyond the scope of this work. What actually is done in ChemSep is discussed briefly
in later chapters.

222
 Table 11.2: Newton’s Method

1. Set iteration counter, k, to zero. Estimate xk

2. Compute F(xk ) and J(xk )

3. Solve Equation (11.5) for xk + 1

4. Check for convergence. If not converged, increment k and return to step 2

11.2.1 Quasi-Newton Methods

To evaluate the Jacobian, one must obtain the partial derivative of each function with
respect to every variable. This can be a time consuming step even if many of these
derivatives are known to be zero. In order to reduce the time required to calculate the
necessary derivative information for the Jacobian, a class of methods, termed quasi-
Newton methods, has been developed. In these methods, approximations to the Ja-
cobian are made and updated through the use of formulae derived to satisfy certain
constraints which have been chosen to force the approximate Jacobian to mimic the
behavior of the actual Jacobian. One such update formula is given by Broyden (1965):

(yk − Bk sk )sTk
Bk+1 = Bk + (11.6)
sTk sk

where Bk is the k-th approximation to the Jacobian; sk is the k-th change to the
variable vector, (xk+1 −xk ); yk is the k-th change to the function vector, (F(xk+1 )−
F(xk )). These methods, which make use of the step changes in the function and
variable vectors, may also be designed to preserve known sparsity patterns within the
matrix (Schubert, 1970).

Although these methods can considerably reduce the amount of computer time nec-
essary to evaluate the Jacobian, they also generally increase the number of iterations
necessary to obtain convergence to an extent where the quasi-Newton solution pro-
cedure may very well end up taking more time than the unaltered Newton’s method.
Furthermore, the methods tend to reduce the domain of convergence from that of
Newton’s method. That is, even though Newton’s method worked from a given initial
estimate of the solution vector, a quasi-Newton method may fail.

223
11.3 Homotopy-Continuation Methods

Homotopy continuation methods form a class of methods for solving difficult sys-
tems of nonlinear algebraic equations. In addition they can be used for tracking the
solutions to nonlinear algebraic equations in terms of a parameter that occurs in the
equations. The basic idea is to follow a path that is traced from a solution to a simple
problem to the solution of the difficult problem.

The homotopy path is implicitly defined by the homotopy equation:

H(x, t) = 0 (11.7)

This represents a set of n continuously differentiable equations in terms of n + 1

variables.

For the systems under consideration here, the homotopy function is a continuous
blending of two functions F(x) and G(x), in terms of a parameter t. In the fol-
lowing, F(x) = 0 represents the original - presumably difficult - problem, G(x) = 0
is the easy problem. The class of equations we will be looking at is that of linear
convex homotopies given by:

H(x, t) = t · F(x) + (1 − t) · G(x) (11.8)

This equation reduces to G(x) = 0 for t = 0 and to F(x) = 0 for t = 1. If F(x)

and G(x) are n-dimensional problems, then, as will be shown later, H(x, t) describes
a curve in n + 1 dimensional space going through G(x) = 0 and F(x) = 0. The
choice of G(x) can be quite arbitrary, as long as the problem has the same dimension
as the original problem. Most commonly used is the Newton homotopy method, where
G(x) is given by:
G(x) = F(x) − F(x0 ) (11.9)
Leading to the homotopy equation:

H(x, t) = F(x) − (1 − t) · F(x0 ) (11.10)

Here x0 is an arbitrary starting guess for F(x). This method does not require us to
solve an easy problem first, because the solution is already known (x0 ). Another im-
portant feature is that by specifying different starting guesses, we can generate differ-
ent paths that can lead to different solutions. This way multiple solutions to F(x) = 0
(if they exist) can be found with relative ease.

A second method that has the same convenient features is the fixed point homotopy
method. In this case G(x) is given by:

G(x) = (x − x0 ) (11.11)

224
The homotopy equation is then given by:

H(x, t) = t · F(x) + (1 − t)(x − x0 ) (11.12)

A third approach is the so-called ‘linear’ homotopy. Here G(x) can be a simplified
version of the problem posed by F(x) = 0. Often, in large big systems of nonlinear
equations, the number of equations that actually cause problems is quite limited. By
replacing the equations causing the problems by simpler ones, a system of equations
may be obtained that is solved with relative ease. This can then serve as our starting
point for the homotopy curve.

11.3.1 Existence of Paths

As stated before, the homotopy equation describes a curve in n + 1 dimensional space

going through the zero’s of F(x) and G(x). Therefore, the solution to F(x) = 0 can
be obtained by following the curve from t = 0 and G(x) = 0 to t = 1. How this
is done exactly will be treated later. We will follow the proof given by Garcia and
Zangwill (1981) to show that such a path exists.

A homotopy equation is given by: H : Rn+1 → Rn and we are interested in the

solutions to:
H(x, t) = 0 (11.13)
These solutions are given by:

H−1 = {(x, t)|H(x, t) = 0} (11.14)

This is the set of all solutions (x, t) ∈ Rn+1 for which H(x, t) = 0. This does not
say anything about the location of (x, t) and these points could be anywhere. They
are, however, included in H−1 . The key to making sure that H−1 consists solely of
paths is given by the implicit function theorem. For that we will examine the solutions
in H−1 more precisely. We take an arbitrary point (x̃, t̃) ∈ H−1 so that:

H(x̃, t̃) = 0 (11.15)

If H is continuously differentiable, we can make a linear approximation near (x̃, t̃):

n
∂Hi (x̃, t̃) ∂Hi (x̃, t̃)
Hi (x, t) ∼
X
= Hi (x̃, t̃) + · (xj − x̃j ) + · (t − t̃) (11.16)
j=1
∂xj ∂t

or, in vector form:

H(x, t) = Hx · (x − x̃) + Ht · (t − t̃) (11.17)

225
where Hx is a Jacobian matrix defined by
 ∂H1 ∂H1
···

∂x1 ∂xn
 · · 
Hx = 
 ·
 (11.18)
· 
∂Hn ∂Hn
∂x1 ··· ∂xn

and  ∂H1 
∂t
 · 
Ht = 
 ·

 (11.19)
∂Hn
∂t

Now we want to show that the points (x, t) in H−1 that are near (x̃, t̃) are on a path
through (x̃, t̃). By definition we find that all points in H−1 must satisfy:

H(x, t) = 0 (11.20)

Assuming that H is actually linear near (x̃, t̃) we get from Eq. (11.17) and Eq. (11.20):

0 = Hx · (x − x̃) + Ht · (t − t̃) (11.21)

Assuming that Hx is invertible at x̃, t̃ we get:

(x − x̃) = −H−1
x · Ht · (t − t̃) (11.22)

This is a system of n linear equations and n + 1 variables (also referred to as the

Homotopy Differential Equations, HDE’s, when written as (dx/dt) = −Hx−1 Ht ).
The solutions must therefore form a straight line. In general H will not be linear, but
H can be approximated by a linearised system as done above. And as long as H is
continuous differentiable and Hx is invertible the homotopy equation describes a path
through (x̃, t̃). In addition, this path will be smooth and continuously differentiable.
All this is summarized by the implicit function theorem:

IMPLICIT FUNCTION THEOREM

Let H : Rn+1 → Rn be continuously differentiable, (x̃, t̃) ∈ H and
Hx are invertible. Thus, in the neighborhood of (x̃, t̃) all points (x, t)
that satisfy H(x, t) = 0 are on a single continuously differentiable path
through (x̃, t̃)

We will now use this theorem for determining a method for following the path. First
we will replace the original variables (x, t) by y. The homotopy equation can, there-
fore, be written as
H(y) = 0 (11.23)

226
Next, we define the partial Jacobian H−i to be the n × n matrix that is formed after
removing the i − th column from the original Jacobian matrix:

 ∂H1 ∂H1 ∂H1 ∂H1


∂y1 ··· ∂yi−1 ∂yi+1 ··· ∂yn+1
 · · · ·
 
H−i =  (11.24)

 · · · ·


∂Hn ∂Hn ∂Hn ∂Hn
∂y1 ··· ∂yi−1 ∂yi+1 ··· ∂yn+1

Note here that

Hx = H−t = H−(n+1) (11.25)
According to the implicit function theorem, a single continuously differentiable path
exists for (x, t) ∈ H−1 in the neighborhood of (x̃, t̃). A unique path is guaranteed
by this theorem, although several H−i may be invertible. Proof of this is relatively
simple and is left as an exercise to the reader. At least one invertible H−i matrix can
be found if the original (n + 1) × n matrix is of full rank. This leads to the following
theorem:

PATH THEOREM
Let H : Rn+1 → Rn be continuously differentiable and suppose that for
every y ∈ H−1 , the jacobian is of full rank, then H−1 consists only of
continuously differentiable paths.

This theorem forms the basis of the proof that paths exist. This is quite convenient,
since if we want to follow a path, it would be nice to know that it actually exists.

11.3.2 Path Following Methods

In the next section we will focus on the Newton homotopy method. Similar derivations
can be made for other methods. Recalling the original Newton homotopy equation

H(x, t) = F(x) − (1 − t) · F(x0 ) (11.26)

The simplest approach to following the path is given by the following method:

Algorithm 1

1. Set initial values for x0

2. Increase value of t

227
 3. Evaluate H(x, t)

4. Solve H(x, t), using x0 as a starting guess

5. Update solution to x0

6. If t = 1, return, else go to 2

This approach might work for simple problems, but is somewhat naı̈ve. It does not
allow for the homotopy path to turn back on itself. What this means will be shown
later.

A better approach for curve tracking (where we are actually following a path) is given
by the differential arc length continuation method. The basis of this lies in the implicit
function theorem. We have shown that H(x, t) represents a path in n + 1 dimensional
space and, in case we want to find the solution to F(x), we only have to follow that
path from t = 0 and G(x) = 0 to t = 1 and F(x) = 0.

In other words, we have to evaluate how the variables change along the curve, and
correct each accordingly (including the continuation parameter t), as we walk along
the curve. A natural way of doing this is by differentiating the homotopy equations
with respect to the arc length of the homotopy curve. This results in the following
equation (for the Newton homotopy method):
   
∂H(x, t) ∂F(x) ∂x ∂t
= · + F(x0 ) · (11.27)
∂s ∂x ∂s ∂s

The changes of the variables are restricted by the change in the arc length: a multidi-
mensional Pythagorean theorem:

(dx1 )2 + (dx2 )2 + . . . + (dxn )2 + (dt)2 = (ds)2 (11.28)

Which can be rewritten to:

 2  2  2
∂x1 ∂x2 ∂t
+ + ... + =1 (11.29)
∂s ∂s ∂s

There are a few things to be noted here. F(x0 ) is an n-dimensional vector and may
be added to the Jacobian matrix leading to the following expression:
∂F1 ∂F1   ∂x1
···
 
∂x1 ∂xn F1 (x0 ) ∂s
∂H(x, t)  · · ·   ·
 ·  ∂x

=
 ·
=0 (11.30)
∂s · ·  
∂s
n 
∂Fn ∂Fn ∂t
∂x1 ··· ∂xn Fn (x0 ) ∂s

228
This is where the importance of the path theorem comes into its own. In order to be
able to solve the above system of equations, the matrix should be of full rank. Then
we can choose a partial Jacobian and solve the system of equations that results:

∂x1
 
∂s
 ∂F1 ∂F1 ∂F1 ∂F1
 ·   ∂F1
··· ··· F1 (x0 )   
∂x1 ∂xi−1 ∂xi+1 ∂xn  ∂xi−1  ∂xi
  ∂s
 · · · · ·  ·  ∂xi
 
  ∂xi+1
· ∂s  = − ·

 · · · · ·  ·  ∂s


 ·

∂Fn ∂Fn ∂Fn ∂F1 ∂Fn
··· ···

∂x1 ∂xi−1 ∂xi+1 ∂xn Fn (x0 )  ∂xn  ∂xi
 ∂s 
∂t
∂s
(11.31)
Note here that xi is the independent variable. This system may be solved using any
sparse matrix solver to give:

∂x1
 
β1
 
∂s
 ·
  ·
   
1 0  ∂xi−1
  
∂s
  βi−1 
·    ∂xi
  
· ∂xi+1
 = − βi+1 · (11.32)
 
 ·   ∂s   ∂s
·

 ·
  
0 1
  
 ∂xn 
 ∂s   βn 
∂t βn+1
∂s

This is, in fact, an expression for the changes of all variables along the curve in terms
of the change in xi . The total change in xi along the curve can now be evaluated using
equation Eq. (11.29) rewritten as:
 2  2 2   2  2  2
∂xi ∂x1 ∂xi−1
∂xi+1 ∂xn ∂t
= 1− −. . .−
− −. . .− −
∂s ∂s ∂s
∂s ∂s ∂s
(11.33)
Now we can substitute Eq. (11.32) into this equation and solve for the derivative of xi
giving:
∂xi 1
=q (11.34)
∂s n+1
1 + k=16=i βk2
P

Substituting Eq. (11.34) back into Eq. (11.32) then results in a system of ordinary dif-
ferential equations that can be integrated using some numerical integration method.

Before we do this however we should pay some attention to selection criteria for the
independent variable. Let us assume that we have not made a good choice for the
independent variable so that after solving the matrix equation we end up with a matrix

229
that looks like this
∂x1
 
β1
 
∂s
 ·
  ·
   
1 c1  ∂xi−1
  
∂s
  βi−1 
·    ∂xi
  
· ∂xi+1
 = − βi+1 · (11.35)
 
 1 cn−1   ∂s   ∂s
·

 ·
  
0 0 0
  
 ∂xn 
 ∂s   βn 
∂t βn+1
∂s

This is a singular matrix. This system cannot be solved for the independent variable.
We can, however, swap the last column of the matrix with the residual vector, thereby
obtaining the following equation system:
   ∂x1   
1 β1 ∂s c1
·   ·
 ·  ∂x  = −  ·  · ∂t
   
 (11.36)
 1 βn   n 
∂s
 cn  ∂s
∂xi
0 0 βn+1 ∂s
0

This system can be solved very easily, and here we can also see what a singular Jaco-
bian means: The independent variable does not change along the curve at this point
on the curve. We must keep this in mind when picking the independent variable. It is
preferable to choose the variable that changes the most along the curve at this point.
In practice we will use the variable that changed the most at the previous point. This
should be a good enough approximation since it is very unlikely that the parameter
that changed the most in the previous step will not change at all in the following step.

Integrating the system of differential equations over (part of) the curve will give us an
estimate of the following point on the curve. Often, Euler’s method is used.
 
∂x
xn+1 = xn + h · (11.37)
∂s
Where h is the integration stepsize. It is likely that this new estimate is someway off
the homotopy curve. Usually a Newton correction step is used to get the estimate back
on the line, before the next extrapolation step is done.

A possible algorithm for a Newton homotopy method is given by:

1. Calculate the function vector

2. Calculate Jacobian.

3. Choose an independent variable (typically t for the first step)

230
 4. Calculate the derivatives of the variables with respect to the arc length

5. Determine the variable that changes most to use as independent variable next
step

6. Integrate the differential equations using Eulers method

7. Correct the newly obtained estimate with Newton’s method

8. Adjust the stepsize of the integration.

9. If t is smaller than 1, go to 4.

Step 8 in this algorithm has not yet had any attention, and a few things must also be
said with respect to item 6.

In Step 6 we use Euler’s method for the integration of the system of differential equa-
tions. This is the simplest method for integrating a system of differential equations,
but it is not the most accurate method. However, whereas accuracy is important in
solving ODE’s, the errors in the continuation methods are determined solely by the
termination criterion of the corrector step. Since we are following a curve, we need
only a global estimate of where the next point lies on the curve, and by a correction
step, this estimated point may be moved arbitrarily close to the curve. Thus, we need
only to make sure that the estimate is within the domain of convergence for the cor-
rector - usually Newton’s - method). Higher order integration methods are therefore
not often used, as the cost of calculation increases considerably with these methods,
and the advantage often is negligible, because the error caused by the integration is
reduced in the correction step.

11.3.3 Step Sizes

The most simple minded step size strategy would be to specify a fixed step size. The
drawback of this method is that the step size has to be small enough to get through
highly curved regions of the homotopy path. However, this strategy leads to an un-
necessarily large number of steps in the straighter sections. A benefit of this method
is that no calculation time is lost in calculating a new stepsize.

We would like to specify the step size in such a way that the number of integration
steps and correction iterations is minimized simultaneously. However there is a clear
tradeoff here: A very small stepsize will result in estimates that will be relatively
close to the homotopy curve and therefore we do not need many correction steps per
integration step. On the other hand, specifying a large stepsize will reduce the number

231
 Step 1: The Angle of predictor n is compared to Step 2: The predicted point is moved back onto the
angle of predictor n-1. If the angle is within a curve. If the contraction ratio exceeds an ‘ide l’
maximum angle, Pn is accepted as new predictor ratio, the corrector is rejected

Pn-1
Pn Pn-1 Pn

Step 3: The predictor step is halved, and a new point Step 4: A new predictor is evaluated, and the angle
is calculated. If the contraction ratio criterion is with the previous predictor is checked. If this angle is
satisfied,the point is accepted. too large, the new predictor and point Pn are rejected.

Pn-1
Pn
Pn
Pn-1

Step 5: A new predictor is generated from Pn-1 and a Step 6: Once the new estimate is corrected, a new
new estimate is calculated. predictor is calculated and checked for the angle with
the previous predictor.

C1
Pn-1 Pn
Pn-1

Figure 11.1: Stepsize for homotopy method.

of steps to be taken. The expected number of correction steps will, however, increase
as each new estimated point will probably be further away from the curve. In some

232
cases they might even move outside the domain of convergence for Newton’s method.

A principal aspect of a continuation method should be a technique for assessing if the

corrector may be expected to converge when started from the predicted point. Den
Heijer and Rheinboldt (1981), however, proved that this is theoretically impossible.
At best we can find a method that maximizes the probability of convergence, while
allowing for the occasional failure in the corrector. This means that the predictor will
have to be adjusted.

The algorithm for choosing step sizes that is implemented in ChemSep is similar to the
one used by Woodman (1989). The algorithm is based on two criteria: (1) the angle
of two successive predictors (2), the ratio of convergence of the corrector iteration.

In the Euler integration step, the a new estimate of a point on the curve is generated
by adding the predictor to the point found after the previous corrector:
 n+1
n+1 n n+1 ∂x
x =x +h · (11.38)
∂s
where n represents the step counter. The stepzize hn+1 is determined from:

hn+1 = β · hn (11.39)

where the β is calculated from:

( s )
αIn θI
β = min 2, , (11.40)
α1n θn

α1n (the contraction ratio) and θn (the predictor angle change) are defined as
nres,1
α1n = (11.41)
nres,0
and
 n+1 T  n 
n ∂x ∂x
θ = arccos[ ] (11.42)
∂s ∂s
In the definition of α1n , nres,i is the residual error of the equations at the end of the i-th
Newton corrector during the n-th predictor-corrector step. αIn is an ‘ideal’ contraction
ratio, given by:
! pp−1
m −1
m
n n p−1 ( f )
αI = (res,0 ) · (11.43)
(nres,0 )pm
where p is the assumed order of convergence (2 for second order convergence, and
typically 1.5 for superlinear convergence), f is the maximum required error for con-
vergence of the Newton step, and m is the maximum number of Newton iterations to

233
convergence. In addition, the algorithm is accompanied by a decision structure illus-
trated in Figure 11.1. If a point has been accepted in the previous run, a new predictor
is generated, and the angle between the two successive predictors is checked. If the
angle is within a maximum allowable angle, the predictor is accepted. Subsequently
the point is moved back onto the homotopy curve by means of a Newton correction
step. If the iteration history suggests that convergence will not be attained within a
desired number of iterations, the predicted point is rejected, the step size is halved,
and a new estimate is generated. Halving the step size is done until the corrector
gives an iteration history that is adequate for convergence within the desired number
of iterations, or until a minimum allowable stepsize is attained. Once the estimate has
been corrected a new predictor is generated and the angle between the new and the
previous predictor is evaluated. If this value exceeds a maximum allowable value, the
previously corrected point is rejected and a new estimate is generated based on the old
predictor, while the step size is halved. If both the angle between successive predictors
and the iteration history of the corrector are within the allowable limits, the step size
is increased, according to Eq. (11.39) and Eq. (11.40). In addition, the step size is not
allowed to become smaller than a minimum allowable stepsize and it is not allowed to
become larger than a maximum allowable stepsize.

11.3.4 Problems and Pitfalls

Homotopy continuation methods sometimes are claimed to be methods that will guar-
antee a solution. This is not true. There are several situations, in which the continua-
tion method will fail.

• The augmented Jacobian is not of full rank. In this case, no invertible Jacobian
can be obtained, and the system cannot be solved. The points at which the Jaco-
bian is not of full rank may be termed ‘singular’. At these points, the homotopy
path may, for instance, split up into two (or more) branches that can lead to
multiple solutions to the problem. This is normally only a problem if either the
predictor or the corrector is located very closely to the singular point.

• Jumping over and skipping parts of the homotopy curve can take place if the
homotopy curve is highly curved, and the parameters for the step size algorithm
(minimum allowable step size, maximum allowable stepsize, maximum allow-
able angle between successive predictors) allow this. What can happen here
is illustrated in Figure 11.2. An example homotopy curve is given in the top
left of the figure, and a good curve tracking routine should follow the curve as
depicted in the top right of the figure. If a bad step size algorithm or a badly
tuned method are used, any of the four scenarios depicted in the figure may then

234
Homotopy Curve Desired Situation

Scenario 1: loop is skipped Scenario 2: Loop is entered in the wrong direction

P6
P5

P2 P3 P2 P3
P4 P5 P4
P1 P1

Scenario 3a: Circulation of the solution Scenario 3b: Circulation of the solution

P1 P1

P5
P2 P3 P2
P3 P5
P4 P4

Figure 11.2: Problems and pitfalls.

occur. The loop may be skipped entirely, or entered the in the wrong direction.
In addition, failures may occur in exiting the loop, also resulting in no solution.

235
References

Broyden, C.G., “A Class of Methods for Solving Nonlinear Simultaneous Equations”,

Math. Comp., Vol. 19 (1965) p. 577.

Garcia, C.B. and W.I. Zangwill; Pathways to Solutions, Fixed Points, and Equilibria,
Prentice Hall, Inc.; Englewood, NJ (1981).

Ortega, J.M., W.C. Rheinboldt, Iterative Solution of Nonlinear Equations in Several

Variables, Academic Press, New York (1970).

Pozrikidis, C. Numerical Computation in Science and Engineering, Oxford University

Press, Oxford (1998).

Schubert, L.K., “Modification of a Quasi-Newton Method for Nonlinear Equations

with a Sparse Jacobian”, Math. Comp., Vol. 24 (1970) p. 27.

Woodman, M.R., “Simulation of the Distillation of Three Phase Mixtures”, Ph.D.

Thesis, Cambridge University (1989).

236
Chapter 12

Flash Calculations

A flash is a one stage operation where a (multiple phase) feed is ‘flashed’ to a cer-
tain temperature and/or pressure and the resulting phases are separated. The flash in
ChemSep deals only with two different phases leaving, a vapor and a liquid. Liquid-
Liquid or multiphase Vapor-Liquid-Liquid flashes are currently not yet supported in
ChemSep. For more information see the references given at the end of this chapter.

V,y

Flash
F,z

L,x

Figure 12.1: Schematic diagram of an equilibrium flash.

237
12.1 Equations

A schematic diagram of an equilibrium flash is drawn in Figure 12.1. The vapor and
liquid streams leaving the flash are assumed to be in equilibrium with each other. The
equations that model equilibrium flashes are summarized below:

• The Total Material Balance:

V +L−F =0 (12.1)

• The Component Material Balances:

V yi + Lxi − F zi = 0 (12.2)

• The EQuilibriuM relations:

Ki xi − yi = 0 (12.3)

• The SUMmation equation:

c
X
(yi − xi ) = 0 (12.4)
i=1

• The Heat (or entHalpy) balance:

V H V + LH L − F H F + Q = 0 (12.5)

where F is the molar feed rate with component mole fractions zi . V and L are the
leaving vapor and liquid flows with mole fractions yi and xi , respectively. Equilibrium
ratios Ki and enthalpies H are computed from property models as discussed in other
chapters. Q is defined as the heat added to the feed before the flash.

If we count the equations listed, we will find that there are 2c + 3 equations, where c
is the number of components. The variables appearing in these equations are:

• c vapor mole fractions, yi

• c liquid mole fractions, xi

• vapor flow rate, V

• liquid flow, L

238
 • temperature, T

• pressure, p

• heat duty, Q

Since we have 2c + 3 equations, two of the 2c + 5 variables above must be specified.

ChemSep allows the following nine flash specifications:

PT: pressure and temperature

PV: pressure and vapor flow

PL: pressure and liquid flow

PQ: pressure and heat duty

TV: temperature and vapor flow

TL: temperature and liquid flow

TQ: temperature and heat duty

VQ: vapor flow and heat duty

LQ: liquid flow and heat duty

12.2 Solution of the Flash Equations

ChemSep uses Newton’s method for solving flash problems as well as simpler bubble
and dew point calculations. The vector of variables used in the PQ-FLASH is:

XT = (V, y1 , y2 . . . yc , T, x1 , x2 . . . xc , L) (12.6)

the vector of functions, (F), is:

FT = (T M B, CM B1 , CM B2 . . . CM Bc , H,
EQM1 , EQM2 . . . Ec , SU M ) (12.7)

The structure of the Jacobian matrix J is shown below:

239
 V y T x L
TMB 1 1
CMB | \ \ |
H x - x - x
EQM # | #
SUM - -

The symbols used in this diagram are as follows:

x single matrix element

1 single element with a value of unity

| vertical column of c elements

\ diagonal with c elements

# square submatrix of order c

- row submatrix with c elements

12.3 The Rachford-Rice Equation

Another effective method for solving the flash equations was developed by Rachford
and Rice (1952). Their method combines all the material balances, equilibrium re-
lations and summation equations into one equation. To obtain the RR equation the
component material balance is used to express xi :
F zi
xi = (12.8)
L + V Ki
On eliminating L using the total molar balance L = F − V we find
zi
xi = (12.9)
1 + (Ki − 1) VF
A similar equation can be derived for the vapor mole fractions:
Ki zi
yi = (12.10)
1 + (Ki − 1) VF
The RR equation is based on subtracting the mole fraction summation equations from
each other: X X
xi − yi = 0 (12.11)

240
which becomes:
c
X (Ki − 1)zi
=0 (12.12)
i=1
1 + (Ki − 1) VF
Of course, this equation can only be solved if the K-values differ from unity. Given
pressure and temperature, the K-values can be calculated and then the vapor to feed
flow ratio is determined from the root of the RR equation. The RR function is mono-
tonic in the ratio of vapor flow over the feed flow, V /F . This is convenient, for this
ratio is also bounded between 0 to 1.

This method can be selected by editing the sep file and typing -1 before method under
the Solve Options section heading (Newton’s method is 1).

References

Henley, E.J., J.D. Seader, Equilibrium-Stage Separation Operations in Chemical En-

gineering, Wiley (1981).

King, C.J., Separation Processes, Second Edition, McGraw Hill (1980).

Rachford, H.H. Jr., J.D. Rice, J. Petrol. Technol., 4, No. 10 (1952) p. 19.

241
242
Chapter 13

Equilibrium Stage Simulation

In this chapter we give the reader a description of the equilibrium stage model for
multistage, multicomponent separation processes. We also provide an overview of the
methods which used to solve the system of equations.

This chapter is adapted from material written by David Vickery and Ross Taylor in
1987 at Clarkson University. Ross used this material during his sabbatical year at
the University of Delft (The Netherlands) to teach an advanced separation processes
course. Since then corrections have been made to bring the review more up to date,
although developments continue to take place and no review of ths field can ever be
complete.

13.1 Introduction

Multicomponent separation processes like distillation, absorption and extraction have

been modelled using the equilibrium stage concept for almost a century. The equations
that model equilibrium stages are called the MESH equations. The M equations are
the Material balance equations, the E equations are the Equilibrium relations, the S
equations are the Summation equations and the H equations are the entHalpy balance
equations.

The MESH equations for the interior stages of a column together with equations for
the reboiler and condenser (if they are needed) are solved together with any speci-
fication equations to yield, for each stage, the vapor mole fractions; the liquid mole

243
fractions; the stage temperature and the vapor and liquid flow rates.

Since the late 1950’s, hardly a year has gone by without the publication of at least
one (and usually more than one) new algorithm for solving the equilibrium stage
model equations. One of the incentives for the continued activity has always been
(and remains) a desire to solve problems with which existing methods have trouble.
The evolution of algorithms for solving the MESH equations has been influenced by,
among other things: the availability (or lack) of sufficient computer storage and power,
the development of mathematical techniques that can be exploited, the complexity of
physical property (K-value and enthalpy) correlations and the form of the model equa-
tions being solved. Developments to about 1980 have been described in a number of
textbooks (see, for example, Holland, 1963, 1975, 1981; King, 1980; Henley and
Seader, 1981; Seader and Henley, 1998) and a number of reviews (see, for example,
Wang and Wang, 1981; Haas, 1992). Seader (1985a) has written an interesting history
of equilibrium stage simulation.

We may identify several classes of methods of solving the MESH equations:

1. Short cut methods in which a great many simplifying assumptions are made in
order to obtain approximate solutions of the MESH equations for certain special
cases (for example, total or minimum reflux operation). We shall not discuss
such methods in this review.

2. Tearing Methods in which the equations are divided into groups and solved
separately.

3. Simultaneous Correction Methods in which all of the equations are solved si-
multaneously using Newton’s method (or a variant thereof).

4. Relaxation Methods in which the MESH equations are cast in unsteady state
form and integrated numerically until the steady state solution has been found.

5. Continuation Methods which are intended for solving difficult problems and
are the subject of much current research interest. We can expect to see further
developments in this area in the next few years.

6. Collocation Methods that have been found useful for solving systems of par-
tial differential equations have been adapted to solve both steady state and dy-
namic equilibrium stage model problems. These methods have some potential
for certain types of simulation problems but have not yet become part of the
mainstream of equilibrium stage simulation and are not discussed further. The
reader is referred to papers by Stewart et al. (1985), Swatrz and Stewart (1986),
Cho and Joseph (1983, 1984), and by Seferlis and Hrymak (1994a,b).

244
Most algorithms for solving the MESH equations in use today fall into categories 2
and 3; tearing methods and simultaneous correction methods. Following sections on
each of these two classes of method we discuss continuation methods in Section 13.5
and relaxation methods in Section 13.6. We conclude with a short discussion of three
phase distillation.

13.2 The Equilibrium Stage Model

The equilibrium stage model was first used by Sorel (1893) to describe the rectifica-
tion of alcohol. Since that time it has been used to model all manner of separation pro-
cesses: distillation (including rectification, stripping, simple (single feed, two product
columns), complex (multiple feed, multiple product columns), extractive, azeotropic
and petroleum refinery distillation), absorption, stripping, liquid-liquid and supercrit-
ical extraction.

Vj L j-1

Wj

Stage j
F
j

Uj
V j+1 Lj

Figure 13.1: Schematic diagram of an equilibrium stage.

In order to better understand the methods that have been proposed for solving the equa-
tions modelling equilibrium stage operations, some familiarity with the basic model
is necessary. A schematic diagram of an equilibrium stage is shown in Figure 13.1.
Vapor from the stage below and liquid from a stage above are brought into contact on
the stage together with any fresh or recycle feeds. The vapor and liquid streams leav-
ing the stage are assumed to be in equilibrium with each other. A complete separation
process is modeled as a sequence of s of these Equilibrium Stages (Figure 13.2).

245
 Vj

W1

Stage 1
F
1 U1

Vj L1

W j-1

Stage j-1
F
j-1 Uj-1

Vj L j-1

Wj

Stage j
F
j Uj

Vj+1 Lj

W j+1

Stage j+1
F
j+1 Uj+1

Vs L j+1

Ws

Stage s
F
s

Ls

Figure 13.2: A sequence of stages modeling a separation column.

13.2.1 The MESH Equations (The 2c+3 Formulation)

The equations that model equilibrium stages are termed the MESH equations, MESH
being an acronym referring to the different types of equations that form the mathemat-
ical model. The M equations are the Material balance equations, of which there are
two types: the Total Material Balance:

MjT ≡ 1 + rjV Vj + 1 + rjL Lj

−Vj+1 − Lj−1 − Fj = 0 (13.1)

and the Component Material Balances:

Mij ≡ 1 + rjV Vj yij + 1 + rjL Lj xij

246
 −Vj+1 yi,j+1 − Lj−1 xi,j−1 − Fj zij = 0 (13.2)

r is the ratio of sidestream flow to interstage flow:

rjV = Wj /Vj rjL = Uj /Lj (13.3)

The E equations are the Equilibrium relations:

Eij ≡ Kij xij − yij = 0 (13.4)

where the Kij are the K-values for species i on stage j. The estimation of the K-values
from fugacity and activity coefficients is discussed at length in chapters ?? and 10.

The S equations are the Summation equations:

c
X
SjV ≡ yij − 1 = 0 (13.5)
i=1
Xc
SjL ≡ xij − 1 = 0 (13.6)
i=1

and the H equations are the Heat balance equations:

1 + rjV Vj HjV + 1 + rjL Lj HjL + Qj

Hij ≡
V L
−Vj+1 Hj+1 − Lj−1 Hj−1 − Fj HjF = 0 (13.7)

where the superscripted ”H’s” are the enthalpies of the appropriate phase. If we count
the equations listed, we will find that there are 2c + 4 equations per stage. How-
ever, only 2c + 3 of these equations are independent. These independent equations
are generally taken to be the c component mass balance equations, the c equilibrium
relations, the enthalpy balance and two more equations. These two equations can be
the two summation equations or the total mass balance and one of the summation
equations (or an equivalent form). The 2c + 3 unknown variables determined by the
equations are the c vapor mole fractions, yij ; the c liquid mole fractions, xij ; the stage
temperature, Tj and the vapor and liquid flow, Vj and Lj . Thus, for a column of s
stages, we must solve s(2c + 3) equations. Table 13.1 shows how we may easily end
up having to solve hundreds or even thousands of equations.

The first entry in this table corresponds to a simple binary problem that could easily
be solved graphically. The second and third are fairly typical of the size of problem
encountered in azeotropic and extractive distillation processes. The last two entries
are typical of problems encountered in simulating hydrocarbon and petroleum mixture
separation operations.

247
 Table 13.1: Number of equations for equilibrium stage model

Number of Number of Number of

components stages equations
c s s(2c + 3)
2 10 70
3 20 180
5 50 650
10 30 690
40 100 8300

13.2.2 The 2c + 1 Formulation

An alternative form of the MESH equations is used in many algorithms. In this varia-
tion, we make use of the component flow defined by:
vij = Vj yij or lij = Lj xij (13.8)
to decrease the number of equations and variables for each stage by two. In terms of
the component flow, the MESH equations can be written as:
Mij ≡ 1 + rjV vij + 1 + rjL lij

−vi,j+1 − li,j−1 − Fij = 0 (13.9)

lij vij
Eij ≡ Kij Pc − Pc =0 (13.10)
k=1 lkj k=1 vkj
c c

X
X
Hij ≡ 1 + rjV vkj HjV + 1 + rjL vkj HjL + Qj
k=1 k=1
c
X c
X
V L
− vk,j+1 Hj+1 − lk,j−1 Hj−1 − Fj HjF = 0 (13.11)
k=1 k=1

Since the total vapor and liquid flow rates are, by definition, the sum of the component
flow rates of the respective phases, the summation equations and total mass balance
equation are satisfied automatically. Thus, the number of equations and variables per
stage has been reduced from 2c + 3 to 2c + 1.

13.2.3 The c + 3 Formulation

The number of unknown variables per stage can be reduced to only c + 3 if we use
the equilibrium relations (13.4) to eliminate the vapor phase mole fractions from the

248
component mass balances (13.2):

M Eij ≡ + 1 + rjV Vj Kij + 1 + rjL Lj xij + Fj zij





−Vj+1 Ki,j+1 xi,j+1 − Lj−1 xi,j−1 = 0 (13.12)

and from the summation equation (13.6)

c
X
SjV ≡ Kij xij − 1 = 0 (13.13)
i=1

which is familiar to us from bubble point calculations. In this formulation of the

MESH equations the vapor phase mole fractions are no longer independent variables
but are defined by equation (13.4). This formulation of the MESH equations has been
used in quite a number of algorithms. It is less useful if vapor phase nonideality is
important.

13.2.4 Special Stages

The MESH equations can be applied as written to any of the interior stages of a col-
umn. In addition to these stages, the reboiler and condenser (if they are included) for
the column must be considered (see Figure 13.2). These stages differ from the other
stages in the column in that they are a heat source or sink for the column. The MESH
equations of the preceding subsections may be used to model these stages exactly as
you would any other stage in the column. For a total condenser at the top of a dis-
tillation column, for example, the distillate is U1 and the reflux ratio is 1/r1L . For a
partial condenser, the vapor product is V1 and the ”reflux ratio” is L1 /V1 . Finally, for
a partial reboiler at the base of a column, the bottoms flow rate is LN . These spe-
cial stages add degrees of freedom to the separation problem and therefore additional
specification equations are required. Common specifications are:

1. the flow rate of the distillate/bottoms product stream,

2. the mole fraction of a given component in either the distillate or bottoms product
stream,

3. component flow rate in either the distillate or bottoms product stream,

4. a reflux/reboil ratio or rate

5. the temperature of the condenser or reboiler

6. a heat duty to the condenser or reboiler.

249
In the case of a total condenser, the vapor phase compositions used in the calculation
of the equilibrium relations and the summation equations are those that would be in
equilibrium with the liquid stream that actually exists. That is for the total condenser,
the vapor composition used in the equilibrium relations is the vapor composition de-
termined during a bubble point calculation based on the actual pressure and liquid
compositions found in the condenser. At the same time, these compositions are not
used in the component mass balances since there is no vapor stream from a total con-
denser.

13.2.5 Nonequilibrium Stages

In actual operation the trays of a distillation column rarely, if ever, operate at equi-
librium despite attempts to approach this condition by proper design and choice of
operating conditions. The degree of separation is, in fact, determined as much by
mass and energy transfer between the phases being contacted on a tray or within sec-
tions of a packed column as it is by thermodynamic equilibrium considerations. The
usual way of dealing with departures from equilibrium in multistage towers is through
the use of stage and/or overall efficiencies.

There are many different definitions of stage efficiency: Murphree (1925), Hausen
(1953), vaporization (Holland, 1975) and generalized Hausen (Standart, 1965). There
is by no means a consensus on which is best. Arguments for and against various possi-
bilities are presented by, among others, Standart (1965, 1971), Holland and McMahon
(1970) and by Medina et al. (1978, 1979). Possibly the most soundly based definition,
the generalised Hausen efficiency of Standart (1965) are the most difficult to use (see,
however, Fletcher, 1987), the least soundly based, the Murphree efficiency is the one
most widely used in separation process calculations because it is easily combined with
the equilibrium relations:
MV MV
Eij ≡ Eij Kij xij − yij − (1 − Eij )yi,j+1 = 0 (13.14)
MV
where Eij is the Murphree vapor efficiency for component i on stage j. In the 2c+1
formulation of the MESH equations, the equilibrium relations read as follows
MV M
Eij Kij lij vij (1 − Eij )vi,j+1
Eij ≡ Pc − Pc + Pc =0 (13.15)
l
k=1 kj k=1 kjv v
k=1 k,j+1

Efficiencies are not normally incorporated in the c + 3 formulation for reasons that
will be discussed later.

Whichever definition of efficiency is adopted, it must either be specified in advance

or calculated from an equation derived by dividing by some reference separation the

250
actual separation obtained from a model of the transport processes taking place on
a tray. Common practice is to use a single value for all of these efficiencies, one
calculated (or specified) in terms of the plate efficiency of the key components. In
a c component system there are c − 1 independent component efficiencies and there
are sound theoretical reasons as well as experimental evidence for not assuming the
component efficiencies to be alike (see Taylor and Krishna, 1993 for a review of the
literature on this topic).

13.2.6 Numerical Solution of the MESH Equations. Preliminary

Considerations

The MESH equations form a set of nonlinear, algebraic equations (algebraic in the
sense that no derivatives or integrals are involved) and must, therefore, be solved by
some iterative process. Their are two steps in the development of an algorithm for
solving systems of nonlinear equations:

• The selection of particular numerical methods

• The selection of the order in which the equations are to be solved

Almost every one of the many numerical methods that have been devised for solving
systems of nonlinear equations has been used to solve the MESH equations. However,
most of the more widely used methods are based on just two methods: repeated sub-
stitution and Newton’s method (or one of its relatives). These methods are described
in detail in Chapter 11.

13.3 Tearing Methods

Equation Tearing involves breaking a system of nonlinear equations into small groups
and solving each group of equations in turn. When solving any subset of the complete
set of equations, only a corresponding number of variables, the ”tear” variables, can
be determined. In order to start the calculations, therefore, it is necessary to assume
values for the remaining variables. The ”torn” set of equations is then solved for the
”tear” variables assuming that the values assigned to all other variables are correct.
Successive groups of equations and variables are torn or decoupled from the full set
of equations and variables until all the variables have been updated. At this point, the
process starts over and is repeated until all the equations are satisfied simultaneously.

251
A great many tearing methods for solving the MESH equations have been proposed.
According to Friday and Smith (1964), tearing methods may be analysed in the fol-
lowing terms:

1. The order in which the equations are grouped.

2. The order in which each group of equations is solved.

3. The selection of which variables are to be computed from which equation.

Subsequent issues are:

4. The method(s) used to solve the M and E equations.

5. The method of calculating the new stage temperatures.

6. The method of calculating the new flow rates, Vj and Lj

The first methods for solving the MESH equations involve keeping the equations for
a given stage together; the equations for one stage are solved then the equations for
a second stage are solved and so on until all the equations for the column have been
solved. These are called Stage to Stage methods and are discussed in Section 13.3.1
below. Most modern methods group all the equations of a given type together; for
example, the M equations or the H equations, for all the stages at once. These methods
are discussed in Section 13.3.2.

13.3.1 Stage-to-Stage Methods

The papers by Lewis and Matheson (1932) and Thiele and Geddes (1933) represent
the first attempts at solving the MESH equations for multicomponent systems numer-
ically (graphical methods for binary systems had already been developed by Ponchon,
Savarit and by McCabe and Thiele). At that time the computer had yet to be invented
and since, modelling a column could require hundreds or even thousands of equations,
it was necessary to break the MESH equations in smaller subsets if hand calculations
were to be feasible.

In 1932, Lewis and Matheson proposed an algorithm in which the MESH equations
were solved in a ”tray-by-tray” approach. The calculations were for a design problem,
a problem in which the recoveries of key components are specified and the number
of trays above and below the feed must be determined. In their approach, the MESH
equations of one stage were solved to determine either the stage temperature, the flow
rate and composition of the entering vapor stream and the flow rate and composition

252
of the exiting liquid stream (in the top or enriching section of the column) or the stage
temperature, the flow rate and composition of the entering liquid stream and the flow
rate and composition of the exiting vapor stream (in the bottom or stripping section of
the column).

Based on the specified recoveries, estimates of the product stream compositions and
flow rates were made and these estimates were used to start the calculations which then
proceeded along two paths. With the conditions of the distillate known (assumed),
it is possible to use the MESH equations describing the condenser to determine the
composition and flow rate of the liquid reflux to the column and the vapor entering
the condenser (and leaving the top stage of the column itself). Since the liquid leaving
this stage is in equilibrium with this vapor stream, it follows that a simple (or not so
simple) dew point calculation will give the composition of the liquid stream leaving
this stage. If equimolar overflow is assumed then the flow rates throughout the column
are already known and the composition of the vapor stream entering the stage can
be determined quite straightforwardly from the component material balances for the
stage. If equimolar overflow does not apply then it is necessary to compute the flow
rate of the leaving liquid stream and the flow rate and composition of the vapor stream
entering from the stage below by solving the component material balances together
with the total material balance and energy balance. This is a much more involved
calculation.

Now we know the composition of the vapor leaving the next stage lower down the
column and we can repeat the entire procedure for this next stage and so we continue
until the feed stage is reached. At this point, this set of calculations is terminated
and a corresponding set of calculations moving from the bottom of the column up to
the feed stage is carried out. The extent to which these progressions of calculations
do not agree hopefully provides a basis for adjusting the estimates of the end condi-
tions. Then the top-down, bottom-up calculations can be repeated until the midcolumn
matching achieves the desired accuracy. When this method was first proposed there
was no systematic method for making the adjustments in the end conditions.

The following year, Thiele and Geddes (1933) proposed an approach which was not
only a tray-by-tray method but also one in which each type of equation (M-E-S-H) was
solved in turn. In this method, the number of stages above and below the feed along
with the reflux ratio and the distillate flow rate were given as column specifications (a
form of the operating problem). Values for the stage temperatures and vapor flow rates
were assumed and the M and E equations were solved tray-to-tray for the compositions
throughout the column (K-values were assumed to be functions of temperature only).
Then the stage temperatures and vapor flow rates were re-estimated using the S+E and
H equations respectively.

253
Holland and coworkers (see Holland, 1963) developed the ”theta-method” in order to
improve the convergence properties of the Thiele - Geddes algorithm by forcing the
component material balances for the column as a whole to close on every iteration.
This allowed faster, surer convergence than the original procedure. The theta method
could also be applied to the Lewis - Matheson method described above for adjusting
the estimated distillate composition. In both cases, the theta method significantly
improves the reliability of these methods.

Despite their essential simplicity and appeal from the pedagogical viewpoint, it must
be said that stage to stage calculation procedures are not used now as often as they
used to be. It was the development of the digital computer that led, in the first place,
to the ”routine” use of these methods and, subsequently, to the discovery of problems
which these, the earliest numerical methods for solving the MESH equations, could
not solve. One drawback to stage to stage methods is that they are sometimes suscepti-
ble to the propagation of truncation errors caused by the use of finite length arithmetic
(eight digits on most computers unless double precision is used). This is why stage-
to-stage calculations with trace components are (very) difficult. Holland (1963) gives
a nice illustration of this problem for those interested in finding out more.

13.3.2 Matrix Methods

Matrix techniques were first used in separation process calculations by Amundson

and Pontinen (1958) who demonstrated that the combined M+E equations (the c + 3
formulation) could be conveniently written in matrix form. We define a column matrix
of combined material balance and equilibrium relations F as

FT = (M Ei,1
c c
, . . . , M Ei,j−1 c
, M Ei,j c
, M Ei,j+1 c T
, . . . , M Ei,s ) (13.16)

c
where M Eij is Eq. (13.12). We also define a column matrix of liquid phase mole
fractions (X) by

XT = (xi,1 , . . . , xi,j−1 , xi,j , xi,j+1 , . . . , xi,s )T (13.17)

Now, we may write the vector function F in matrix form as follows:

ABC x=R (13.18)

254
This system of equations has the structure shown below
B1 C 1 X1 R1
    
 A2 B 2 C2   X2   R2
    
 
 ··· ··· · · · · · ·  ···   ··· 
    
 Aj Bj Cj   Xj  =  Rj  (13.19)
    

 ··· ··· ··· ···   ···   ···
   

 An−1 Bn−1 Cn−1   Xn−1   Rn−1 
An Bn Xn Rn
where the coefficient matrix ABC has three adjacent diagonals with coefficients:
Aj = Lj−1 (13.20)
Bj = −(Vj Ki j + Lj ) (13.21)
Cj = Vj Ki j (13.22)
The right hand side matrix R has elements
Rj = −Fj zij (13.23)
Note that there is one of these tridiagonal matrix equations for each component in the
mixture. A specialized form of Gaussian elimination known as the Thomas Algorithm
(Lapidus, 1962) can be used to solve the tridiagonal system (13.18) very efficiently.
This form of elimination procedure is used to minimize the amount of storage and cal-
culations for solving the linear system of equations and is summarized in Table 13.2.

The first step of any method in this category is to estimate the flow rates and K-values.
The three other independent equations for each stage; i.e. the total mass balances,
the enthalpy balances and the summation equations; are subsequently used to deter-
mine the three remaining independent variables for each stage; the vapor flow rates,
the liquid flow rates and the stage temperatures. When all these variables have been
re-estimated, the coefficients in the system of combined M and E equations are recal-
culated and the process is repeated until all the model equations are simultaneously
satisfied.

Three quite different approaches to the problem of computing the flow rates and tem-
peratures have evolved:

1. In the Bubble Point (BP) method, the S+E equations are used to determine the
stage temperatures from a bubble point calculation. The vapor and liquid flow
rates are computed from the energy balances and total material balances.
2. In the Sum Rates (SR) method the vapor and liquid flow rates are computed
directly from the summation equations and the stage temperatures determined
from the energy balances.

255
 Table 13.2: The Thomas algorithm

• Elimination procedure:

For the first equation (j = 1):

C1
C1 =
B1
R1
R1 =
B1
For j = 2 to N − 1:

Dj = Bj − Aj Cj−1
Cj
Cj =
Dj
Rj − Aj Rj−1
Rj =
D
For the last equation (j = N ):

DN = BN − AN CN −1
RN − AN RN −1
RN =
DN

• Backward substitution procedure:

For the last variable:

XN = RN

For j = N − 1 down to 1:

Xj = Rj − Cj Xj+1

3. In the Newton-Raphson methods the bubble point equations and energy bal-
ances are solved simultaneously for the stage temperatures and vapor flow rates;
the liquid flow rates follow from the total material balances.

256
13.3.3 The BP Method

The BP method was introduced by Amundson and Pontinen (1958) who used matrix
inversion to solve equations (13.12). A significant improvement in the BP method
equations was introduced by Wang and Henke (1966) who introduced the Thomas
Algorithm (Lapidus, 1962) to solve the tridiagonal system (4.1). Holland and more
coworkers (1974,1975) have combined some of these ideas with the ”theta-method”
of convergence acceleration (Holland, 1963) to develop a ”modified theta-method”.
The ”theta” and ”modified theta” methods themselves have been further refined by
Pierucci et al. (1983) and have even been cast in a form where the ”theta” correction
factor is replaced with a temperature dependent parameter, the T-method (Pierucci and
Ranzi, 1982; Pierucci et al., 1983). The BP method of solving the MESH equations is
summarized in Table 13.3. Sridhar (1990) has developed a modified BP method using
the same principles as Sridhar and Lucia (1990) for the SR method (see below) and
obtained similar improvements in reliability wjen compared to traditional BP methods.

13.3.4 The SR Method

The SR algorithm summarized in Table 13.4 is due to Sujata (1961) and Friday (1963)
actually, Sujata (1961) used the 2c+1 version of the MESH equations in his work
and the tridiagonal system of combined M+E equations was written in terms of the
component flow rates in the liquid phase. Burningham and Otto (1967) incorporated
the Thomas algorithm in their implementation of the SR method. Sridhar and Lucia
(1990) developed a modified SR method based on insights provided by rigorous anal-
ysis that provides an analytical expression for the Newton acceleration of the inner
loop as well as partial derivatives of the total vapor flow with respect to temperature
needed in the outer loop. The result is a method that is capable of solving problems
involving narrow, intermediate and wide-boiling mixtures alike.

13.3.5 Newton Raphson Methods

Tierney and coworkers (Tierney and Yanosik, 1969; Tierney and Bruno, 1967; Bruno
et al., 1972) and Billingsley and Boynton (1971)) avoided the problem of assigning ei-
ther the vapor flow rates or temperatures to either the summation equations or enthalpy
balances by solving both sets of equations for both sets of variables simultaneously us-
ing a multidimensional form of Newton’s method (order 2N to be precise). The liquid
flow rates are eliminated from these equations using the total material balances. Any
composition dependence of the K-values and enthalpies is ignored in the application

257
of Newton’s method of which more in the next section.

The ”multi-theta method” of Holland (1975) is similar to the Tierney and Yanosik
method except that the stage temperatures and flow ratios (Vj / Lj ) are taken to be the
independent variables. Tomich (1970) has also worked with a 2N Newton Raphson
method and used Broyden’s quasi Newton method to update the Jacobian after just
one iteration with the full Newton method.

Other applications of Newton’s method to solving distillation problems (within the

context of a tearing algorithm) are due to Newman (1963) and Boynton (1970).

13.3.6 Inside-Out Methods

One interesting attempt to alleviate the difficulties mentioned resulted in the so called
”inside-out” (IO) algorithms of Boston and Sullivan (1974), Boston and Britt (1978),
and Boston (1980). In these methods, complicated equilibrium and enthalpy expres-
sions were replaced by simple models and the iteration variables T , V , x and y were
replaced by variables within the models which are relatively free of interactions with
each other. However, the resulting algorithms are quite problem specific requiring
new software for each type of specification, flash problem, etc. This is not to say
that other decoupling methods are not at all application specific. IO methods for most
of the common equilibrium stage separation processes have been developed; indeed,
IO methods have replaced many of the older algorithms in commercial simulation
programs, the modified version of Russell (1983) being the basis for some of them.
Readers are referred to Seader and Henley (1998) for complete details of the some-
what lengthy algorithm.

13.3.7 Nonequilibrium Stages

Murphree efficiencies are not normally included in the matrix methods described in
the preceding section because if Equation (13.4) is used to eliminate the yij from
the component material balances (13.2) the upper triangular portion of the coefficient
matrix is filled out (Huber, 1977). We could use Equation (13.4) to eliminate the xij
from the component mass balances and solve the resulting tridiagonal system for the
vapor phase compositions instead but this is not recommended (King, 1980, p 472).
Rather than combine Equations (13.4) and (13.2) into a single set of equations, we
define a column matrix of discrepancy functions

FT = (Mi,1
c c
, Ei,1 , . . . , Mi,j−1 c
, Ei,j−1 , Mi,j , Ei,j ,

258
 Table 13.3: Bubble Point Method for Solving the MESH Equations

Begin Outer Loop

1. Initialize the vapor and liquid flow rates using the total mass balance and the
enthalpy balance with an assumption of constant molar flows (flow rates change
from one stage to the next only upon the introduction of a feed).

Begin Inner Loop

2. Choose an initial temperature profile for the column.

3. Evaluate the K-values for each stage (if this is the first iteration, either use an
ideal K-value [temperature dependence only] or assume that all mole fractions
are the same).

4. Set up the linearized systems of equations (one system for each component) as
defined by equations (13.12) and solve for the liquid phase compositions.

5. For each stage j, normalize the liquid compositions.

6. Using the normalized liquid compositions, solve the summation equation for
each stage [in the form (xi − yi )] for the stage temperatures. If there is no
change since the last iteration of the inner loop proceed to Step 7, otherwise
return to Step 3.

End Inner Loop

7. Use the enthalpy and total mass balances to re-estimate the vapor and liquid
flow rates. If there is no change since the last outer loop iteration, end; otherwise
return to Step 4 (the K-values will not have changed since the last inner loop
iteration).

End Outer Loop

c c
Mi,j+1 , Ei,j+1 , . . . , Mi,s , Ei,s )T (13.24)
c
where Mij is the component material balance Equation (13.2) and Ei j represents
equation (13.4). Each pair of equations depends on only four mole fractions: xi,j−1 ,
yij , xij and yi,j+1 . Thus, if we define a column matrix of mole fractions X by

XT = (yi,1 , xi,1 , . . . , yi,j−1 , xi,j−1 , yi,j , xi,j ,

259
 Table 13.4: Sum Rates Method for Solving the MESH Equations

1. Estimate the vapor flow rates and calculate the liquid flow rates using the total
mass balances. Assume constant molar flows from stage to stage (flow rates
change from one stage to the next only upon the introduction of a feed).

2. Choose an initial temperature profile for the column.

3. Evaluate the K-values for each stage (if this is the first iteration, either use an
ideal K-value [temperature dependence only] or assume that all mole fractions
are the same as the combined feed composition).

4. Set up the tridiagonal systems of equations (one system for each component) as
defined by equations (13.12) and solve for the liquid phase compositions.

5. For each stage, sum the liquid compositions. Compute new liquid flow rates by
multiplying old value by the sums computed in step 5. Compute new vapor flow
rates from the total mass balances.

6. Normalize the xi,j . Compute vapor phase compositions from the equilibrium
relations and then normalize the vapor mole fractions.

7. Use the enthalpy balances to re-estimate the stage temperatures. Compare new
stage temperatures and flow rates with the previous values. If there is no change
since the last iteration, end; otherwise return to Step 3.

yi,j+1 , xi,j+1 , . . . , yi,s , xi,s )T (13.25)

we may write
ABCD X−R=0 (13.26)
With the equations and variables ordered in this way, the coefficient matrix ABCD has
four adjacent diagonals with the additional diagonal in the upper triangular portion of
the matrix:

B1 C1 D1 X1 R1
    

 A2 B2 C2 D2 
 X2  
  R2 


 ··· ··· ··· ··· 
 ···  
  ··· 

 Aj Bj Cj Dj  Xj = Rj  (13.27)
    

 ··· ··· ··· ··· 
 ···  
  ··· 

 An−1 Bn−1 Cn−1  Xn−1   Rn−1 
An Bn Xn Rn

260
where n (=2s) is the order of the matrix. On the odd numbered rows of ABCD the
elements are given by

A2j−1 = −Lj B2j−1 = Wj + Vj

C2j−1 = Uj + Lj D2j−1 = −Vj+1 (13.28)

(these elements being derived from the component material balances) whereas on the
even numbered rows we have .
MV
A2j = −1 B2j = Eij Kij
MV
C2j = 1 − Eij D2j = 0 (13.29)

which belong to the equilibrium relations. For a total condenser set E M V = 0 For a
total reboiler, set E M V = 0 and B2 j = 1 The right hand side matrix R has elements

R2j−1 = Fj zi,j R2j = 0 (13.30)

The index j corresponds to the stage number and stages are numbered down from the
top. This linear system of equations can be solved for the xij and yij very easily using
Gaussian elimination.

We have used this method of solving the component material balances and efficiency
equations in place of the corresponding steps in the BP method for distillation simu-
lations. While not as fast as the conventional tridiagonal matrix algorithm (the quad-
diagonal matrix has twice the number of rows as the tridiagonal matrix in the Wang
and Henke method), it permits the straightforward solution of problems in which the
stage efficiency is not unity. We estimate that the quad-diagonal algorithm will be
faster than Huber’s (1977) method of dealing with stage efficiencies (with an upper
Hessenberg matrix of order s) if the number of stages exceeds nine.

13.3.8 Summary

Equation tearing involves iteratively solving a subset of the complete set of model
equations for a subset (the tear variables) of the complete set of variables. Within the
loop that determines how the tear variables are to be re-estimated, all other equations
are solved exactly. If the remaining set of equations requires iteration (or, possibly,
further tearing) the result is a series of nested iteration loops.

There are essentially two ways of tearing the equilibrium stage model equations: by
stage or by type. Stage to stage methods were the first to be developed but have
largely been abandoned in favor of the by-type methods which make use of matrix

261
formulations of the equations. By their very nature, matrix methods are limited to
solving operating problems; design problems can be tackled by changing the problem
specification, solving the operating problem that results and seeing whether it meets
the design requirements. This limitation of the matrix methods (when compared to
the stage to stage methods) is offset by the following:

1. There is no limit on the number of components. Stage to stage methods work

best for two component systems and least well for systems with a great many
components for which it is hard to make good estimates of the distillate compo-
sition.

2. There is no limit on the number of feed streams. Feed may even be introduced
directly into the reboiler or condenser should the need arise without changing
the form of (4.1). There is no limit on the number of sidestreams that may be
withdrawn from the column (multiple feeds and sidedraws cause problems for
stage to stage methods).

3. Most important of all is the fact that the combined mass balance and equilibrium
relations depend on the liquid phase composition on only three adjacent stages:
xi,j−1 , xij and xi,j+1 . This means that the very efficient Thomas algorithm for
computing the liquid phase composition from Equation (13.12) can be used.

There are three major sub-types of matrix method for solving equilibrium stage sepa-
ration process problems: the BP, SR and Newton-Raphson methods. The BP formu-
lation supposedly works ”well” for narrow boiling systems and the SR formulation
for wide boiling systems typical in absorption and stripping operations (see Friday
and Smith (1964) for the first discussion of this point). Two recent papers by Sridhar
and Lucia (1989, 1990a) have analyzed these classic tearing algorithms in some de-
tail. Sridhar and Lucia (1990b) have presented a modified sum-rates method that can
be used to solve problems involving narrow boiling mixtures as well as wide boiling
systems.

Inside-out methods are widely used in commercial simulation programs.

13.4 Simultaneous Correction Methods

One problem with tearing methods is the number of times physical properties must
be evaluated (several times per outer loop iteration) if temperature and composition
dependent physical properties are used. It is the physical properties calculations that
generally dominate the computational cost of chemical process simulation problems

262
(Westerberg et al., 1979; Baden and Michelson, 1987). A second problem can arise
if any of the iteration loops are hard to converge. One might well wonder what is the
point of requiring subsets of the complete set of equations to be satisfied exactly (by
iteration) when the estimate of the tear variables is not the correct solution. Might
it not be a better strategy to solve all of the MESH equations simultaneously using
Newton’s method (or one of its relatives)? By guessing enough unknown variables
to permit all other quantities in the MESH equations to be calculated explicitly and
by not requiring that any of the equations be satisfied until complete convergence has
been achieved we might avoid both of these problems with tearing methods.

It is not completely clear who first implemented a simultaneous correction method

for solving multicomponent distillation and absorption problems. As is so often the
case, it would appear that the problem was being tackled by a number of people in-
dependently. Simultaneous solution of all the MESH equations was suggested as a
method of last resort by Friday and Smith (1964) in a classic paper analysing the rea-
sons why other algorithms fail. They did not, however, implement such a technique.
The two best known and most frequently cited papers are those of Goldstein and Stan-
field (1970) and Naphtali and Sandholm (1971), the latter providing more details of an
application of Newton’s method described by Naphtali at an AIChE meeting in May
1965. Naphtali (1965), Goldstein and Stanfield (1970) and Wang and Wang (1980)
cite a communication by Wang and Oleson (1964), but their work is not in the public
domain. As described by Naphtali (1965), the method of Wang and Oleson employs
just 2 independent equations and variables per stage and is not, therefore, a full simul-
taneous correction method.

To the best of our knowledge, a method to solve all the MESH equations for all stages
at once using Newton’s method was first implemented by Whitehouse (1964) (see,
also, Stainthorp and Whitehouse, 1967). Among other things, Whitehouse’s code
allowed for specifications of purity, T , V , L or Q on any stage. Interlinked systems of
columns and nonideal solutions could be dealt with even though no examples of the
latter type were solved by Whitehouse.

It is interesting to note that the early work in this field was done prior to the publica-
tion of many of the methods of estimating thermodynamic properties taken for granted
today (NRTL, UNIQUAC and UNIFAC methods for liquid phase activity coefficients
and SRK, PR and other equations of state) and before some algorithms widely used
today were developed. Since the pioneering work of Whitehouse, Naphtali and Sand-
holm and Goldstein and Stanfield, many others have employed Newton’s method or
one of its relatives to solve the MESH equations (Roche, 1971; Ishii and Otto, 1973;
Shah and Bishnoi, 1978; Christiansen et al., 1979; Magnussen et al., 1979; Ferraris
and Morbidelli, 1982 to name but a few).

263
Simultaneous correction procedures have shown themselves to be generally fast and
reliable, having a locally quadratic convergence rate in the case of Newton’s method,
and these methods much less sensitive to difficulties associated with nonideal solutions
than are tearing methods. Extensions to the basic method to include complex column
configurations, interlinked columns (see Section 8), nonstandard specifications (Fer-
raris, 1981), and applications to column design (Ricker and Grens, 1976) result in
only minor changes in the algorithm. In addition, simultaneous correction procedures
can easily incorporate stage efficiencies within the calculations (something that is not
always possible with other algorithms).

13.4.1 Designing a Simultaneous Solution Procedure

Seader (1986) lists a number of things to be taken into consideration when designing a
simultaneous correction method; a revised and extended list follows and is discussed
in more detail below.

1. What equations should be used?

2. What variables should be used?

3. How should the equations be ordered?

4. Hhow should the variables be ordered?

5. How should the linearized equations be solved?

6. Should J be updated on each iteration or should it be held constant for a number

of iterations or should it be approximated using quasi-Newton methods. Should
derivatives of physical properties be retained in the calculation of J?

7. Should flexibility in specifications be provided and, if so, how?

8. What criterion should we use to determine convergence?

9. How should the initial guess be obtained?

10. What techniques should we use to improve reliability?

13.4.2 What Equations and Variables Should be Used?

The possible choices include: the 2c + 3 formulation, the 2c + 1 formulation, the c + 3

formulation or some other, equivalent set of equations?

264
The variables that go along with these alternative formulations are (Lj , Vj , Tj , yij ,
xij ) in the 2c+3 formulation, (lj , vj , Tj ) in the 2c+1 formulation and (Lj , Vj , Tj , xij )
in the c + 3 formulation. Goldstein and Stanfield (1970) used the c + 3 formulation of
the MESH equations while Naphtali (1965) and Naphtali and Sandholm (1971) used
the 2c + 1 formulation.

The advantage of the c + 3 formulation is that the much smaller number of equations
means that much less computer time will be required to solve the linearized system of
equations. The disadvantage is that the vapor phase composition dependence of the K-
values cannot be properly accounted for. This limits this formulation to systems which
exhibit ideal behavior in the vapor phase. Departures from equilibrium normally are
handled through some form of stage efficiency factor. Both the 2c + 1 and 2c + 3
formulations are easily modified to handle stage efficiencies without losing some of
the highly desirable Jacobian structures we note below. The same cannot be said of
the c + 3 formulation.

In the 2c + 1 formulation, the material balance equations are linear and will, if New-
ton’s method is used, be satisfied on every iteration after the first. Thus, the problem
reduces to solving the equilibrium relations and energy balance equations. In the
2c + 3 formulation the material balance equations are quadratic and will require a few
iterations before being satisfied. The 2c+1 formulation is probably the best choice
for single columns. For interlinked systems the 2c + 3 formulation may be better
depending on just how much flexibility in specifications is allowed.

13.4.3 How should the Equations and Variables be Ordered?

Goldstein and Stanfield (1970) grouped the equations by type (M+E-S-H) (as also,
and somewhat earlier, did Whitehouse, 1964), whereas Naphtali and Sandholm (1971)
grouped the equations and variables by stage. Grouping the equations and variables
by stage is preferred for systems with more stages than components (practically all
distillation and many absorption and extraction problems) while grouping by type is
preferred for systems with more components than stages (some gas absorption prob-
lems).

13.4.4 Computation of the Jacobian

To evaluate the Jacobian, one must obtain the partial derivative of each function with
respect to every variable. Part of the appeal of the grouping by stage formulation is that
for standard specifications (and some others), the Jacobian matrix is block tridiagonal

265
in structure:

B1 C1
A2 B2 C2
A3 B3 C3
J = . . .
. . .
Am−1 Bm−1 Cm−1
Am Bm

in which each entry A, B, C is a matrix in its own right. The Aj submatrices contain
partial derivatives of the equations for the j-th stage with respect to the variables for
stage j − 1. The Bj submatrices contain partial derivatives of the equations for the
j-th stage with respect to the variables for the j-th stage. Finally, the Cj submatri-
ces contain partial derivatives of the equations for the j-th stage with respect to the
variables for stage j + 1.

A B C
V y T x L V y T x L V y T x L
TMB 1 1 1 1
CMB \ | | \ \ | | \
H x - x x - x - x x - x
EQM # | # \
SUM - -

The symbols used in these diagrams are as follows:

x single matrix element

1 single element with a value of unity

| vertical column of c elements

\ diagonal with c elements

# square submatrix of order c

- row submatrix with c elements

The elements of Bj include partial derivatives of K-values with respect to temperature

and composition. Since it is rather a painful experience to differentiate, for example,
the UNIQUAC equations with respect to temperature and composition, in many SC

266
codes this differentiation is done numerically. This can be an extremely time con-
suming step. However, neglect of these derivatives is not recommended unless one
is dealing with nearly ideal solutions, since, to do so, will almost certainly lead to
an increase in the required number of iterations or even to failure. Michelson (1986)
and Baden and Michelson (1987) have shown that coding analytical derivatives of the
complicated physical property models is well worth while as this leads to substantial
reductions in the computer time required for this portion of the calculations.

Almost all of the partial derivatives needed in ChemSep are computed from analyti-
cal expressions. The exception is the temperature derivatives of the excess enthalpy
which requires a second differentiation with respect to temperature of the activity and
fugacity coefficient models. In the case of non-standard specifications (which allow
”wild” specification such as the temperature on a stage in the middle of a column,
instead at one of its ends) the specification equation must be calculated using finite
differencing.

In order to reduce the time required to calculate the necessary derivative information
for the Jacobian, a class of methods, termed quasi-Newton methods, has been used. In
these methods, approximations to the Jacobian are made and updated through the use
of formulae derived to satisfy certain constraints which have been chosen to force the
approximate Jacobian to mimic the behavior of the actual Jacobian. Experience with
quasi-Newton methods has been reported by Gallun and Holland (1980), Lucia and
coworkers (Lucia and Macchietto, 1983; Lucia and Westman, 1983; Westman, Lucia
and Miller, 1984, Venkat and Lucia, 1987) and others. It is probably fair to say that
pure quasi-Newton methods are not especially useful for separation process problems.

In an effort both to decrease the number of iterations required by a pure quasi-Newton

method and to try to maintain the region of convergence of Newton’s method, Lucia
and coworkers have developed what they call a hybrid approach to the solution of sets
of model equations. In this hybrid approach, any derivative information which is easy
(i.e., inexpensive) to calculate is included in a ”computed” part of the Jacobian, C,
while derivative information which is difficult or expensive to obtain is included in an
”approximated” part of the Jacobian, A

J=C+A (13.31)

In equilibrium stage calculations A includes derivatives of activity and fugacity co-

efficients with respect to composition. On each iteration in the solution procedure,
C is calculated exactly and A is updated using some appropriate quasi-Newton for-
mula. The result is a method that, while taking more iterations than Newton’s method
to solve a given problem, generally takes about 50% less computer time to converge
(when compared to numerical differentiation but not when compared to analytical dif-
ferentiation) and has nearly the same region of convergence as Newton’s method.

267
Another possibility is to selectively retain the derivatives of physical properties in the
calculation of the Jacobian. For dynamic simulations this has shown to be counterpro-
ductive (Kooijman, 1995) and therefore ChemSep doesn’t support such a technique.

13.4.5 How Should the Linearized Equations be Solved?

The Jacobian matrix above is of order s(2c + 3). Thus, the number of elements in the
matrix is s2 (2c + 3)2 . Since the vast majority of the elements are zero it is absolutely
essential to take account of the sparsity of J when solving the linearized equations;
straightforward matrix inversion is totally impractical (not to say numerically impos-
sible). Software for solving sparse linear systems is described by Seader (1985b).

Linear systems with a block tridiagonal structure may efficiently be solved using a
generalized form of the Thomas algorithm, one where the scalar arguments are re-
placed by matrices and divisions are replaced by premultiplication by inverse matri-
ces. The steps of this algorithm are given by Henley and Seader (1981). Still further
improvements in the block elimination algorithm for solution of separation process
problems can be effected if we take advantage of the that the special structure of the
submatrices A, B and C, in fact, A and C are nearly empty (Baden, 1984; Baden and
Michelson, 1987).

If the column has pumparounds extra matrices will be present which are not on the di-
agonal and the use of block tridiagonal methods becomes less straight forward. Simi-
lar problems arise with non-standard specifications that are not on the variables of the
condenser and reboiler. When we solve multiple interlinked columns (currently not
supported by ChemSep) a special ordering is required to maintain the diagonal struc-
ture of the jacobian. Therefore, ChemSep now uses a sparse Gaussian elimination
solver (NSPIV) to solve the linear system of equations.

13.4.6 The Convergence Criterion

Convergence of Newton’s method in ChemSep is assessed by looking at the residu-

als of the equations that make up the model. These function-values are squared and
summed to give one single convergence parameter: the sum of squares (SS). In Chem-
Sep we compare the square root of this sum of squares with a user specified conver-
gence criterion. Typical values for this convergence criterion vary between 10−6 for
equilibrium column and flash calculations to 10−3 for nonequilibrium column calcu-
lations.

268
As ChemSep does not scale the equations it is possible that this sum of squares remains
too large for such small convergence criterions. As this is not always foreseeable, we
extended the convergence test in ChemSep with a check on the change in the vari-
ables. If the absolute relative value of the change of each variable is smaller than the
convergence criterion, ChemSep also considers the problem solved.

Finally, in order to prevent calculations from continuing endlessly, there is a maxi-

mum number of iterations after which the program pauses. ChemSep will then ask
the user whether to continue the calculations, and for how many additional iterations.
Typically, we find that 30 Newton iterations suffices to solve most separation prob-
lems at hand. Often, when more iterations are needed, there are other problems that
prevent the method from converging. For example, specifications can be impossible to
attain, or the iterations oscillate between two or more sets of values, neither of which
represents a solution to the model equations.

13.4.7 How Should the Initial Guess be Obtained?

In order to obtain convergence, Newton’s method requires that reasonable initial es-
timates be provided for all N (2c + 3) independent variables. It is obviously imprac-
tical to expect the user of a SC method to guess this number of quantities. Thus, the
designer of a computer code implementing a SC method must provide one or more
methods of generating initial estimates of all the unknown variables from, at most,
one or two user supplied values of end stage temperatures and flow rates.

There is no standard method for generating initial estimates of the flow rates, tem-
peratures and compositions. Temperatures are often easy to estimate using linear
interpolation between two (or more) guessed values. Flow rates, too, can be esti-
mated fairly easily by assuming constant molar flows from stage to stage. Estimating
the compositions of the vapor and liquid phases is a little more difficult. Possible
approaches include performing one (or more) iterations of a plate to plate calculation
procedure (Burton and Morton, 1987) or a tridiagonal matrix method such as the Wang
and Henke algorithm for distillation or the sum rates method for absorption (see, for
example, Wayburn and Seader, 1983). Alternative procedures that have been used
include:

1. short cut calculation procedures (Venkat and Lucia, 1987)

2. relaxation techniques (Ketchum, 1979)

3. simply setting the composition of each stream equal to the composition of the
combined feeds (Furzer, 1986)

269
 4. flashing the combined feeds at the average column pressure and assuming the
resulting composition of the vapor and liquid streams hold for each stage.

ChemSep uses an automatic initialization procedure that does not require the user to
make any estimates. Flow rates are estimated assuming constant molar flows from
stage to stage. If the bottoms flow rate and reflux ratio are NOT specified and cannot
be estimated from the specifications that are supplied then the bottoms flow rate is arbi-
trarily assigned a value of half the total feed flow and the reflux ratio is given a default
value of 2. This, of course, could cause serious convergence problems. Therefore, op-
tional guesses have been added to the specifications to circumvent this problem (you
need to toggle an interface switch to make them visible).

Mole fractions of both phases and temperatures are estimated by carrying out a few
iterations of the Bubble Point method described above. The first time through this
procedure we use K-values estimated assuming ideal solution behavior at the col-
umn average pressure and an estimate of the boiling point of the combined feeds.
(This eliminates the normal requirement of estimates of end stage temperatures). Two
more iterations of the BP method are carried out using the (possibly nonideal) K-value
model actually selected.

For columns without condenser and reboiler a different method is used for the initial-
ization. The composition of the liquid streams are set equal to the top liquid feed com-
position and the composition of the vapor equal to the bottom vapor feed composition.
Temperatures in the column are set equal to the temperature of the first feed specified.
For columns with either a condenser or a reboiler - but not both - the compositions are
initialized to the overall compositions of all feeds combined, and the temperatures to
the bubble point at the column pressure and the overall feed compositions.

Currently, no special methods are employed for columns with pumparounds. Columns
with large pumparound flows may require the flows to be estimated by the user, or, to
repetitively solve the problem using the old results as the initialization and increasing
the pumparound flow.

13.4.8 Reliability

SC methods are far more reliable and versatile than most other methods. The same
method will solve distillation, gas absorption and liquid extraction problems. It must
be admitted though, that although the probability that Newton’s method will converge
from the automatic initial estimates is quite high, there is no guarantee of conver-
gence. The difficulty of supplying good initial estimates is most severe for problems

270
involving strongly nonideal mixtures, interlinked systems of columns and nonstandard
specifications.

Several methods have been used to improve the reliability of Newton’s method; damp-
ing the Newton step (Naphtali and Sandholm, 1971), use of steepest descent (ascent)
formulations for some of the iterations (Powell, 1970), and combination with relax-
ation procedures (Ketchum, 1979); none of which has proven to be completely satis-
factory. The methods most recently proposed for assisting convergence of Newton’s
method are continuation methods discussed in a later section.

In default mode, ChemSep does NOT use any of these techniques, other than a check
to make sure that all quantities remain positive. Mole fractions, for example, are not
permitted to take on negative values. The user does have the option of supplying
damping factors.

13.4.9 Damping Factors

The Newton’s method in ChemSep has some extra features that will enhance the con-
vergence to the solution. Newton’s method computes a new estimate of the solution
based on the current Jacobian and function vector. However, the new solution vector
might be physically meaningless, for example if a mole fraction becomes less than
zero or larger than unity. Also, the new solution vector might represent too large a
change in stage temperatures or flows for the method to be stable. To alleviate these
problems ChemSep allows damping of the Newton iteration. Temperature changes
are limited to a maximum number of degrees per iteration (the default to 10 K) and
flow changes up to a maximum fraction of the old flows (the default fraction is 0.5).
The compositions require a special type of damping. If a mole fraction is going to be
negative or larger than unity, the change is limited to half the distance to the relevant
extreme value. In addition, if a damping factor is specified, the maximum change
in composition equals the factor (the default factor is 1 allowing a change over the
whole mol fraction range). This type of damping has proven to be very effective. The
damping factors can be found under the Options - Solve options menu.

If, for some reason, your column simulation does not converge, changing the damping
factors might help. If you know the iteration history (by either limiting the number of
iterations or by printing out the intermediate answers. Both can be done in the Options
- Solve options menu) you can adapt the damping factors so the column simulation
might converge. Note that convergence tends to be slower when you start to apply
extra damping by making the factors smaller, Newton method loses its effectiveness
when damped. Also, damping does not guarantee convergence.

271
13.4.10 User Initialization

For difficult problems it might be necessary for the user to provide initial temperature,
pressure, or flow profiles.

The user can specify either temperature or flow profiles, or both. The only requirement
is that values for the first and last stages are provided. Missing temperatures on inter-
mediate stages are computed by linear interpolation, missing flows are computed on a
constant flow from stage to stage basis. Therefore, it is better to specify the flows of
the first and last two stages in case a condenser and reboiler are present. Composition
profiles are computed through the method described above, however, temperatures
are not computed using the bubble point calculations. If both user specified tempera-
ture and flow profiles are incomplete ChemSep switches to the automatic initialization
method.

13.4.11 Initialization with Old Results

In some cases it is useful to use the converged results of a previous run as the initial
guess for a new - similar - problem (for example, when the bottoms flow rate and
reflux rate are not specified and cannot be estimated, and the automatic initialization
always uses a reflux ratio of 2). This is a very straightforward way of specifying the
initial guess as long as the number of components remains the same. Care must be
taken when feed or product specifications or locations are changed. The results are
interpolated if the number of stages is changed.

13.5 Continuation Methods for Difficult Problems

An unconstrained Newton’s method should converge at least 75% of all separation

process problems with no particular difficulty. Adding bells and whistles in the form
of damping factors, line searches and the like will increase the probability of success
to around 95%. However, the problems that Newton’s method fails to solve demand
much more than 5% of the engineers time as he/she attempts to make the method
converge. Continuation methods introduced in Section 11.3 are intended to alleviate
this problem. There are three categories of continuation method that have been used
for solving the equilibrium stage MESH equations:

1. Mathematical methods which place the MESH equations into a homotopy equa-
tion of purely mathematical origin. In this category we find the Newton homo-

272
 topy used by Wayburn and Seader (1983), Bhargava and Hlavacek (1984) and
many others.

2. Physical continuation methods in which the nature of the equations being solved
is exploited in some way. In this category we have the thermodynamic homo-
topies of Vickery and Taylor (1986a,b) and a related method due to Frantz and
Van Brunt (1986a,b). Parametric continuation methods also belong in this cate-
gory (Jelinek et al. (1973) and others).

3. Proprietary continuation methods, about which very little can be said (Byrne
and Baird, 1983; 1985).

We might also note that continuation methods may be combined in the same program.

The first to use the Newton homotopy for separation process problems were Salgovic
et al. (1981). They integrated the HDEs with a high order differential equation solv-
ing technique and sufficiently small step sizes to ensure that no significant buildup
of error had occurred (i.e. x(1) was indeed the solution to F(x) = 0). They found
that while this method was reliable, the number of function evaluations became quite
large. Salgovic et al. also solved their test problems by switching to Newton’s method
to solve the final problem after a prescribed number of Euler steps was used to inte-
grate the HDEs. No correction was used for of the intermediate problems; Newton’s
method was used only to solve the final problem. This method was found to be quite
effective as long as the differential equations were solved to an appropriate accuracy.
Unfortunately, the appropriate number of steps and the size of each step could not be
determined a priori.

Wayburn and Seader (1983) used the Newton homotopy in their solution of the MESH
equations for interlinked distillation columns. Initial estimates of the solution were ob-
tained by assuming temperature and flow rate profiles throughout the columns and then
obtaining composition estimates from one iteration of a Wang-Henke type method (see
Wayburn and Seader, 1983). If Newton’s method failed to solve the given problem
from any of the starting guesses, the Newton homotopy was used. They parameter-
ized the problem in terms of the arclength of the curve being traced out in the solution
space and used an Euler step to proceed from one intermediate problem to the next and
corrected the resulting estimate of the solution vector for the functions using Newton’s
method in the plane orthogonal to the unit normal vector at the point being considered.
The problems they solved involve the ”ideal” system benzene - toluene - o-xylene and
included some problems with nonstandard specifications.

273
Using these same procedures, Chavez et al. (1986) were able to find multiple solutions
for some systems of interlinked columns; a Petlyuk system and a fractionator with
side stripper used to separate a mixture of benzene, toluene and o-xylene. Problem
specifications included the purity of the three product streams as well as a bottoms flow
rate for the Petlyuk configuration. While it was not possible to find all the solutions at
a given reflux ratio through use of Newton’s method alone, the homotopy method was
successful in all cases. Lin et al. (1987) have shown how all these solutions may be
computed from a single starting point.

Bhargava and Hlavacek (1984) have made use of an implementation of the Newton
homotopy that they found to be both effective and efficient for the nonideal problems
with which they dealt. In it, they do only one Newton correction on the problems
leading to the final problem to be solved. The estimate of the solution to the next
problem is taken to be the result of that iteration; there is no attempt to use a predictor
step to obtain the solution to the next problem. In their paper, they found that an even
step length in the range 0.2-0.5 was sufficient to solve their problems. However, this
suggestion was based upon the results of some problems which could be solved by
Newton’s method without any particular difficulty.

Kovach and Seider (1986) used the Newton homotopy for solving azeotropic distilla-
tion problems in conjunction with another homotopy for the liquid-liquid equilibrium
and three phase equilibrium problems arising in the distillation calculations. They
used the Newton homotopy to find a solution to the azeotropic distillation problems
in much the same way that Wayburn and Seader did. From this solution they defined
another homotopy which allowed them to look for multiple solutions to the MESH
equations and effectively to perform parametric sensitivity calculations. This second
homotopy as well as the homotopies used for the liquid-liquid equilibrium calcula-
tions fall more appropriately in the section on physically based homotopies and will
be discussed there.

Burton and Morton (1987) (see, also, Burton, 1986) modified an equation oriented
flowsheeting system so that the Newton homotopy could be called upon. A number
of example problems involving distillation were solved using essentially the methods
described by Wayburn and Seader (1983). The method was not able to find solutions
in every case.

Ellis et al. (1986) used Algorithm 6.4 with a parameterized form of the Newton ho-
motopy. They integrated the HDEs using Euler’s method and a semi-implicit Runge-
Kutta method, both with Newton correction. In addition, Gear’s method was applied
directly to solve the homotopy equation H(x, t) = 0. Step size was adjusted as the
integration proceeded. The results of three test problems show that Euler’s method
required fewest steps but Gear’s method took the shortest time. The performance of

274
the Runge-Kutta method was disappointing.

Frantz and Van Brunt (1986a,b) and Vickery and Taylor (1986a,b) have also used the
Newton homotopy (more on their work below).

13.5.1 Drawbacks to Mathematical Homotopies

The homotopy functions, H(x, t), used in most algorithms are purely mathematical
contrivances; they do not model a real separation process except when the continuation
parameter takes on its final value. This, of itself, is not a serious difficulty; after all, the
function is being used only as a means of obtaining the solution to a problem which
is of physical origin. However, difficulties can arise if elements of x take on negative
values at intermediate values of the continuation parameter. Again, of itself, this is
not always a serious matter. However, most methods for estimating, for example,
activity coefficients, demand at least nonnegative input and possibly normalized mole
fractions in order to return real values. If, for example, one or more of the mole
fractions were negative, a subroutine implementing such a method would be unable
to deliver meaningful values. This situation was encountered by Vickery and Taylor
(1986a,b) and by Burton and Morton (1987). Ellis et al. (1986) report negative total
flow rates (a much less serious problem) at intermediate values of the continuation
parameter.

13.5.2 Physical Continuation Methods

In order to be certain that a continuation method does not encounter difficulties with
physically meaningless values of variables while solving intermediate problems, the
method should be designed so that the equations H(x, t) are physically meaningful
for any value of the continuation parameter. Parametric continuation methods auto-
matically meet this requirement, mathematical homotopies do not. A logical first step
in designing a physically based homotopy function is to look for the reasons why algo-
rithms used to solve the MESH equations fail. It is the nonlinearity of the equilibrium
ratios or ”K-values” and the enthalpies that are responsible for the majority of con-
vergence difficulties with Newton’s method. Enthalpies are normally expressed as the
sum of an ideal enthalpy and an ”excess” term that accounts for the nonidealities in
the phase under consideration. The Ki may be estimated from.

yi φL γi p ∗
Ki = = Vi = V i (13.32)
xi φi φi p

275
The vapor pressure p∗i is a relatively simple function of temperature. Fugacity coeffi-
cients normally are calculated from equations of state (EOS), e.g. the Redlich-Kwong
and Peng-Robinson EOS. The nonlinearities in these terms arise from the temperature
dependence of the EOS and a composition dependence that manifests itself primar-
ily through the rules used for the evaluation of mixture parameters. The liquid phase
activity coefficient is obtained from a complicated function of composition and tem-
perature, e.g. the van Laar, Wilson, NRTL, UNIQUAC or UNIFAC equations (see, for
example, Walas, 1985).

13.5.3 Thermodynamic Homotopies

Vickery and Taylor (1986a,b) have designed a homotopy in which the thermodynamic
quantities are modified, simplifying them for the initial problem so that the K-values
and enthalpies are much easier to deal with. Then, as the homotopy parameter is
increased, the thermodynamic quantities are brought back to their original forms, until,
finally, the original problem has been solved. Here, we generalize their formulation
by writing
(1−g(t)) g(t)
Ki (T, x, y, t) = Ki,simple Ki,actual (13.33)
(1−g(t)) g(t)
Hi (T, x, y, t) = Hi,simple Hi,actual (13.34)

where Ki,simple is a simple model for the K-value and Ki,actual is the rigorous ex-
pression above. Hsimple is a simple model for the enthalpy and Hactual is the rigorous
model. The functions g(t) and w(t) are zero at t = 0 and one at t = 1 but are other-
wise arbitrary. Possible choices for Ki,simple include:

1. unity

2. a real, positive number, different for each component

3. an ideal solution K-value

Vickery and Taylor took g(t) = w(t) = t and Ki,simple = p∗i /p with p∗i given by the
Antoine equation. Powers et al. (1986) have used Ki,simple = 1 in another, related,
application of these ideas. Choices for Hsimple are discussed by Vickery and Taylor
(1986a).

The evaluation of Hx and Ht is not quite as straightforward as it is in the linear ho-

motopies. None the less, these quantities are quite easily calculated and do not require
much extra work. The square matrix Hx is still identically equal to the Jacobian

276
used in the Newton-Raphson iterations. The evaluation of Ht is discussed by Vickery
(1988).

Vickery and Taylor (1986a,b) found the thermodynamic homotopy to be rather more
effective than the simple implementation of the Newton homotopy (Algorithm 6.4)
for solving distillation problems involving strongly nonideal systems. Fewer steps
and fewer iterations were needed by the thermodynamic homotopy and the Newton
homotopy occasionally failed (for the reasons discussed above). Fidkowski et al.
(1993) used a linear combination of ideal and nonideal K-values as part of a homotopy
method for finding all the azeotropic pooints in multicoponent mixtures.

Frantz and Van Brunt (1986) used analogous arguments in developing physically
based homotopies for application to hydrometallurgical solvent extraction models. In
their work, they describe the development of custom imbedding techniques which
allow a physically realistic model for the chemical equilibrium in the process flow-
sheets they consider. The method they develop is shown to be superior to any of the
mathematical homotopies they employed.

Since they were dealing with heterogeneous azeotropic systems, Kovach and Seider
(1986) had to consider liquid-liquid phase splitting. To solve these problems they
would start by assuming the two phases formed as almost pure binaries with the re-
maining components added as the homotopy parameter is changed. At the same time,
the isoactivity of only the component being added is enforced at all parameter values.
The isoactivity equations are solved in a manner reminiscent of the Newton homotopy.
Kovach and Seider also showed that the two phase envelope for ternary systems could
be traced quite easily by parameterizing the feed composition and following the phase
split as the homotopy parameter is changed.

13.5.4 Parametric Continuation Methods

The reflux ratio or bottoms flow rate are parameters that have been used in parametric
solutions of the MESH equations (Jelinek et al., 1973). Parametric continuation has
also been used to detect multiple solutions of the MESH equations (Ellis et al., 1986;
Kovach and Seider, 1986; Burton, 1986).

13.5.5 An Efficiency Based Continuation Method

A parameter occurring naturally in the MESH equations that makes a good continua-
tion parameter is the stage efficiency, E M V . The maximum separation possible on a

277
 Table 13.5: Efficiency continuation

1. Choose initial (nonzero) value of E M V and compute initial vector x

2. Solve F(x, E M V ) = 0 at current value of E M V using Newton’s method

3. If E M V is not unity, solve the homotopy differential equations

dx/dE M V = −Hx−1 H

4. Adjust E M V , use Euler’s method to approximate x at E M V + ∆E M V and

initialize x for next problem in the sequence. Return to step 2.

given stage is obtained with an efficiency of unity. On the other hand, for a vanish-
ingly small stage efficiency the stage performs no separation worth mentioning and
the streams leaving the stage have essentially the same flow rates, composition and
temperature as the combined feeds. This fact can be exploited in a simple yet ”effi-
cient” continuation method for solving difficult equilibrium stage separation process
problems

Muller (1979) appears to have been the first to use the Murphree efficiency as a con-
tinuation parameter. Each problem in the sequence was solved using a stage to stage
procedure. Sereno (1985) used the efficiency as a continuation parameter for solving
liquid liquid extraction problems. Vickery et al. (1988) have used the efficiency as a
continuation parameter in conjunction with Newton’s method and the ”HDEs” (with
t ≡ E M V ) to solve multicomponent distillation problems in single and interlinked
systems of columns. The algorithm of Vickery et al. (1988) is summarized in Ta-
ble 13.5. The initialization procedure, Step 1, described in detail by Vickery et al.
(1988), is completely automatic and does not require any guesswork on the part of the
user. Although this method properly belongs to the class of parametric continuation
methods, it was developed for solving difficult problems, not for parametric sensitivity
studies. In this sense, it is closer in spirit to homotopy continuation methods.

The efficient continuation method is very effective at solving difficult problems in-
volving ”standard specifications”; it cannot, without modification, handle problems in
which a product stream purity is specified. Part of the reason that this method is so
effective is that the nonlinearities of the equilibrium relations are again reduced. The
K-values are now multiplied (for the first problem) by a small number - the efficiency.
Also, since the compositions vary little from tray to tray, the temperatures, excess en-
thalpies and activity coefficients do not change significantly from tray to tray. These
two factors combine to make the MESH equations much easier to solve at very low
values of E M V .

278
13.5.6 Combined Physical Homotopies

An efficiency-based continuation method is very effective at solving problems involv-

ing standard specifications but cannot handle problems in which a product stream pu-
rity is specified. Woodman (1989, 1990) used effiency-based continuation to solve a
thermodynamically ideal problem involving standard specifications. Thermodynamic
continuation was then used to bring in the solution nonidealities in a controlled way.
Finally, parametric continuation was invoked to solve any problems that involved non-
standard specifications.

13.5.7 Summary

If a well designed code for solving the MESH equations using Newton’s method is
available, then a simple implementation of the Newton homotopy, the thermodynamic
homotopy or the efficiency based continuation method can be implemented very eas-
ily. A little extra effort is required to implement the HDEs and still more work is
needed if a sophisticated step size algorithm and/or arc length parameterization is de-
sired. However, the end result of all of this work will be a code that rarely will fail, as
long as the problem being solved actually has a solution.

We would like to stress that Newton’s method should usually be tried before resorting
to a continuation method (at least until a continuation method is designed that can
outperform Newton’s method on problems where both work).

13.6 Relaxation Methods

Relaxation techniques differ from both tearing algorithms and simultaneous conver-
gence methods in that the relaxation techniques do not solve steady state problems.
Rather, at least one set of the MESH equations is cast in an unsteady state form. The
equations are then integrated from some initial state (guess) until successive values of
the variables do not change.

The first published results of application of the relaxation technique were given by
Rose, Sweeney, and Schrodt (1958). In their method, the M equations alone were cast
in an unsteady state form. Compositions were then calculated using Euler’s method
and the temperatures and flow rates calculated from the S and H equations. An im-
provement to the method was proposed by Ball (1961) in which Euler’s method was
replaced by an implicit integration scheme which led to the formulation of a tridiago-

279
nal set of algebraic equations. The H equations were written in unsteady state form by
Vermeuil and Oleson (1971). Other authors have extended the idea to more complex
columns and higher order integration schemes.

The appeal of the relaxation technique lies in the extreme stability of the method. As
long as the time step stays within the region of stability for the differential equation
solver, the method should converge to a solution. However, this convergence is gen-
erally a very slow process, slowing even more as the solution is approached. For this
reason, the method is used (only) for problems which are very difficult to converge
or for situations where a knowledge of how the steady state is achieved is important.
Ketchum (1979) has proposed a compromise in which a relaxation technique is used
in conjunction with a simultaneous correction procedure which gradually takes over
as the rate of convergence due to relaxation becomes small.

13.7 Three Phase Distillation

There are many commercially important operations involving the distillation of mix-
tures which may form two distinct liquid phases. Steam distillation (the injection
of live or ”open” steam into the base of a distillation column) is widely used in the
petrochemical industries for the recovery and purification of light aromatics such as
benzene, toluene and xylene, for the recovery of gasoline by steam distillation from
absorber oils and for the separation of many complex petroleum mixtures into frac-
tions of narrow boiling range. Many of these systems will form two liquid phases
on some or all of the trays in a distillation column; a water rich phase and a hydro-
carbon rich phase. Azeotropic and extractive distillation are used in the chemical
industry for separating mixtures that are hard to separate by fractional distillation.
Many azeotropic systems form two-liquid phases either in a decanter or, additionally,
on some of the stages in a column.

Examples of such processes include the distillation of many alcohol/water mixtures

such as butanol-water, propanol-butanol-water (Block and Hegner, 1976), dehydration
of ethanol using, as the entrainer, benzene (Ross and Seider, 1979; Prokopakis and Sei-
der, 1983), trichloroethylene, diethyl ether, pentane, gasoline fractions or trimethylpen-
tane (Furzer, 1985); dehydration of secondary butyl alcohol (Kovach and Seider,
1986b); dehydration of acetic acid and water using butyl acetate (Fuchs et al., 1983);
dehydration of isopropanol with cyclohexane and the distillation of vinyl acetate -
acrylonitrile - water and acetonitrile - acrylonitrile - water mixtures (Ferraris and Mor-
bidelli, 1982). Furfural and water is used as a separating agent in the extractive dis-
tillation of butane/butadiene mixtures, a process that assumed enormous commercial
importance in the 1940’s for the production of butadiene for use in the manufacture of

280
synthetic rubber (Grohse 1948 and others). Grohse reported the existence of two (and
sometimes even three) liquid phases on the trays of such columns.

The separation of fermentation products (Pucci et al., 1986); the production of fluoro-
carbons by reaction with carbon tetrachloride and hydrogen fluoride (Ross and Seider,
1980), the distillation of light hydrocarbons with ammonia (see van Winkle, 1967),
the cryogenic distillation of nitrogen/natural gas mixtures (Yu et al., 1969) are other
distillation processes that involves two liquid phases.

13.7.1 Equilibrium Stage Model of Three Phase Distillation

Design and simulation of three-phase distillation processes has also been carried out
using equilibrium stage approaches and there has been a good deal of interest shown in
this topic in the last few years. This resurgence of interest in the computer aided design
and simulation of these operations has followed the development of thermodynamic
models that are able to accurately account for the high degree of nonideality exhibited
by these systems. The extension of the mathematical model to three-phase systems
is quite straightforward, it is obtaining the numerical solution that is ”difficult” (more
on this below). The equations conventionally used to model the behavior of a two-
phase stage are the same as those presented in Section 2. For those stages with three
phases, the MESH equations are written as described in section 2 with some minor
modifications:

1. Extra terms are added to the mass balance equations to allow for one more
liquid stream into the stage and one more liquid stream out of the stage. the
Total Material Balance:

Mj ≡ Vj + L0j + L00j − Vj+1 − L0j−1 − L00j−1 − Fj = 0 (13.35)

and the Component Material Balances:

Mij ≡ Vj yij + L0j x0ij + L00j x00ij − Vj+1 yi,j+1

−L0j−1 x0i,j−1 − L00j−1 x00i,j−1 − Fj zij = 0 (13.36)

2. Two new sets of c equilibrium relations are added. The second liquid phase is
in equilibrium with both the vapor phase and the other liquid phase.
0 0 0
Eij = Kij xij − yij = 0 (13.37)
00 00 00
Eij = Kij xij − yij = 0 (13.38)
∗ ∗ 00
Eij = Kij xij − yij = 0 (13.39)

281
 However, a total of only 2c equilibrium relations are independent since
∗ 00 0
Kij = Kij /Kij (13.40)

3. One more summation equation is added to deal with the second liquid phase.
c
X
SjV = yij − 1 = 0 (13.41)
i=1
c
0 X
SjL = x0ij − 1 = 0 (13.42)
i=1
c
00 X
SjL = x00ij − 1 = 0 (13.43)
i=1

4. Extra terms are added to the enthalpy balance to allow for one more liquid
stream entering the stage and one more liquid stream exiting the stage.
0 00
Hij V
≡ Vj Hij + L0j Hij
L
+ L00j Hij
L V
− Vj+1 Hi,j+1
0 00
−L0j−1 Hi,j−1
L
− L00j−1 Hi,j−1
L F
− Fj Hij + Qj = 0 (13.44)

This brings the total number of independent equations describing a stage with three
phases on it to 3c + 4 equations as compared to the 2c + 3 required to describe a two
phase separation stage.

These equations must be solved for all stages in the column in order to predict the
product compositions and flow rates. However, the numerical solution of these equa-
tions is a decidedly nontrivial task. There are a number of reasons for this:

1. The very high nonlinearity of the thermodynamic quantities (the K-values and
enthalpies) means that the numerical computations are unusually sensitive to
estimates of the temperatures and compositions and convergence often is very
difficult to obtain.

2. Unlike two phase separation process problems, the number of equations and
variables necessary to determine the operating condition of a column is not
known until the problem has been solved. The reason for this is that even though
the process is called ”three phase” distillation, there will almost certainly be a
number (probably a majority) of stages with only one liquid phase present.

3. The set of model equations listed above always admits the (”trivial”) two-phase
solution; that is, one vapor and one liquid phase.

282
Many different algorithms for solving this set of equations have been proposed (Block
and Hegner, 1976; Boston and Shah, 1979; Ross and Seider, 1980; Prokopakis et al,
1981; Ferraris and Morbidelli, 1981, 1982; Kinoshita et al, 1983; Schuil and Bool,
1985; Pucci et al., 1986; Kovach and Seider, 1986; Baden, 1984; Baden and Michel-
son, 1987). All of these methods are, in one sense or another, extensions of methods
that have been found useful for solving two-phase distillation problems; for exam-
ple, simultaneous solution of all of the model equations using Newton’s method (Fer-
raris and Morbidelli, 1981,1982; Baden, 1984; Baden and Michelson, 1987; Cairns
and Furzer, 1990), a multistage flash algorithm (Ferraris and Morbidelli, 1981 again),
”bubble-point” type methods (Ferraris and Morbidelli, 1981, Block and Hegner, 1976;
Kinoshita et al, 1983), ”inside-out” methods in which the property evaluations are put
in an outer iteration loop rather than in an inner loop as in the ”bubble point” type
methods (Boston and Shah, 1979; Ross and Seider, 1980; Prokopakis et al, 1981;
Schuil and Bool, 1985) and homotopy-continuation methods (Kovach and Seider,
1986a,b), Woodman et al. (1989).

Since the number of liquid phases present on each tray in the separation unit is not
generally known, the number of equations necessary for modelling the unit cannot be
determined until the actual operating conditions of the column have been determined.
In some algorithms, for example, the SC algorithm of Ferraris and Morbidelli, it is
necessary to specify in advance which stages have two phases and which stages have
three phases. This is an unsatisfactory feature of their method that makes convergence
a rather uncertain adventure.

13.7.2 The Mixed K-value Model

Schuil and Bool (1985) extended the work of Niedzwieki et al. (1980) so that a three
phase separation process unit could be modelled using existing two phase separation
process computer programs. A similar approach is used in the SC method of Baden
and Michelson (1987).

The basic idea is to treat the two liquid phases as a single phase with thermodynamic
properties defined by an appropriate averaging of the two liquid phases present. If
the stage being considered actually has only one liquid phase then the thermodynamic
properties are those of the actual liquid. In either situation, the number of equations
used to model a single stage is always known.

The two quantities for which a suitable averaging must be determined are the equi-
librium ratios and the liquid phase enthalpy. Considering first the equilibrium ratios,

283
recall that the K-value for species i is defined by
yi
Ki (T, x, y) = (13.45)
xi
In this case, we will take xi to refer to the overall mole fraction of component i in the
two liquid phases.

To determine xi in terms of the two existing liquid phase compositions, x0i and x00i , the
following expression is used.
x0i L0 + x00i L00
xi = (13.46)
L0 + L00
By defining Ki0 to be yi /x0i and K”i to be yi /x”i , equations (13.45) and (13.46) can
be combined to yield
(L0 + L00 )Ki0 Ki00
Ki = (13.47)
Ki0 L0 + Ki00 L00
or if α is defined to be the fraction of the liquid that is found in the first (’) phase
(L0 /L00 )
Ki0 Ki00
Ki = (13.48)
0
αKi + (1 − α)Ki00
Recalling that the K-value definition
yi γi p∗
Ki (T, x, y) ≡ = Vi (13.49)
xi φi p
it can be seen that the only term forming the K-value which depends upon the con-
ditions in the liquid phase is the liquid phase activity coefficient. The pure phase
information as well as the vapor phase fugacity coefficient are the same for both Ki0
and Ki00 . With this information, it is possible to show that equation (13.48) effectively
reduces to
γi0 γi00
γi = (13.50)
0
αγi + (1 − α)γi00
and the mixed K-value can be given by (13.48) using this mixed activity coefficient.

In order to calculate a mixed liquid phase enthalpy, start with

(L0 + L00 )H = L0 H 0 + L00 H 00 (13.51)

Then, dividing through by (L0 + L”) gives

H = H 0 + (1 − α)H 00 (13.52)

These two phase combined properties replace the single liquid phase properties in a
two phase separation process algorithm.

284
13.7.3 Efficiency of Three Phase Distillation

The equilibrium stage model, so widely used in distillation simulation and design,
does not, of course, represent reality in that few stages actually operate at equilibrium.
The usual way out of this difficulty is either to use overall efficiencies or to combine an
equation for ”stage efficiency” with the equilibrium relations. Stage efficiencies have
a more fundamental basis than overall efficiencies (which are really nothing more
than a disguised form of rules of thumb and operating experience) and, for two phase
systems, can be estimated from appropriate correlations (e,g. the AIChE method;
King, 1980). However, no methods have been developed that can be used to estimate
the stage efficiency of a tray with two liquid phases on it.

Very few experimental studies on this interesting and, from this point of view, badly
overlooked area of distillation technology have been published. Schoenborn et al.
(1941) report overall column efficiencies for the distillation of trichloroethylene, toluene
and water in laboratory scale columns. An investigation into the efficiency of the ex-
tractive distillation of butane/butadiene mixtures using furfural and furfural/water as
the extractive agent was carried out by Grohse (1948). He found that a second liq-
uid phase was present whenever the water weight fraction was in the range 6 to 9%.
Both groups found that the overall efficiency was not affected by the number of liquid
phases present (two or three). However, in the discussion that accompanies Schoen-
born’s paper, D.B. Keyes, G.G. Brown and D.F. Othmer separately warn against gen-
eralizing the finding of Schoenborn et al. stating that in some cases the presence of a
second liquid phase adversely affects efficiency, sometimes it has no effect at all and
on some occasions it actually helps the separating ability of the column. Brown writes
that “the efficiency goes all to pot if you don’t keep the water off the plates” and “in
some cases the tray efficiency closely approaches zero”.

Furzer (1985) and Cairns and Furzer (1987, 1990a) give tray to tray composition data
for the dehydration of ethanol using 224-trimethylpentane. These experiments were
carried out at total reflux. Data for only one experiment are given in the first paper
but Furzer (1985) does present a graphical summary of the efficiency calculated for
ten similar experiments. From these graphs it can be seen that the overall efficiency
for the two phase and three phase regions are not always the same; efficiencies of
these two regimes show no consistent trends and fluctuates with the column F-factor.
Moreover, these efficiencies are, as implied by Furzer, dependent on the model used
to estimate the activity coefficients (UNIFAC in this case). This finding is not so very
surprising; efficiencies of strongly nonideal multicomponent fluids can be extremely
sensitive to the methods used to estimate the thermodynamic properties (K-values).

This seems to cast further doubt on the suitability of the equilibrium stage - stage
efficiency approach for such systems. Kovach and Seider (1986b) have reported some

285
plant data for the dehydration of secondary butyl alcohol. They simulated the column
with an equilibrium stage model assuming an overall column efficiency of around
80%.

Schoenborn et al (1941) carried out some of their experiments in a glass column so

that the action on the trays could be observed. According to Schoenborn et al., good
mixing of the two liquids was observed and although the two phases could readily
be distinguished, it was impossible to tell which phase (if any) was dispersed in the
other. When samples of the liquid were withdrawn it appeared that the solvent phase
was usually dispersed in the water phase. Grohse (1948) does not provide much detail
on the nature of the dispersion except to imply that they too had good mixing. Furzer
(1985) states that an emulsion existed on the trays in his column. Davies and Porter
(1987) have recently published photographs that show that three phase systems tend
to foam in the region around the binodal curve. It would also appear that, in some
instances at least, the two liquid phases lie one above the other.

Only Seider and coworkers (Ross and Seider, 1980; Prokopakis et al, 1981; Kovach
and Seider, 1986a) and Pucci et al (1986) have even attempted to handle departures
from equilibrium on three phase trays. Ross and Sieder have done this by introducing
two Murphree-type stage efficiencies. In their calculations Ross and Seider provide
arbitrary values for these stage efficiencies and, also for lack of any better information,
assume that all efficiencies are equal, i.e. the two liquid phases are in equilibrium with
each other. Two particularly interesting conclusions to arise from the study by Ross
and Seider are

1. ”The Murphree tray efficiency has an important effect on the location of 2 liquid
phases and, in some cases whether two liquid phases appear”.

2. The extension to account for two liquid phases not in equilibrium is nontrivial.

Since a second liquid phase may seriously affect the performance of the column, it is
important to be able to accurately locate those trays on which a second liquid phase
forms (so that it can be withdrawn by a sidestream if desired). Ross and Seider’s
calculations suggest that this cannot be done reliably, given that there is no sufficiently
extensive base of stage efficiency data for this kind of distillation process.

13.8 Reactive Distillation

Much of the literature on RD modelling by equilibrium stage models is concerned pri-

marily with the development of methods for solving the steady state EQ stage model.

286
For the most part such methods are more or less straightforward extensions of methods
that had been developed for solving conventional distillation problems. The number of
examples that illustrate most papers usually is rather limited, both in number as well
as in the type of RD process considered (most often it is an esterification reaction).
Only rarely is there any attempt to compare the results of simulations to experimental
data (some exceptions are noted below). More and more of the more recent modelling
studies are carried out using one or other commercial simulation package: Aspen
Plus, Pro/II, HYSYS, and SpeedUp are the packages mentioned most often in the pa-
pers discussed in what follows. A comprehensive review of RD modelling has been
given by Taylor and Krishna (2000).

Symbol List

Latin Symbols

c number of components
EMV Murphree vapor efficiency
E equilibrium or efficiency equation
f feed component flow rate (kmol/s)
F feed flow rate (kmol/s)
(F ) discrepancy function vector or
column matrix of mass balance and efficiency equations
G(x) ”Easy” function
H enthalpy (J/kmol)
H heat balance equation (J/s)
H homotopy function
Ht derivative of homotopy function with respect to t
[J] Jacobian matrix
K K-value
l liquid component flow rate (kmol/s)
L liquid flow rate (kmol/s)
M component material balance
p pressure (P a)
Q heat load on stage (J/s)
r ratio of side stream flow to interstage flow
R rate equation
s number of stages
S summation equation
t homotopy parameter

287
U liquid sidestream flow (kmol/s)
v vapor component flow rate (kmol/s)
V vapor flow rate (kmol/s)
W vapor sidestream flow (kmol/s)
(X) column matrix of independent variables, or
vector of variables
x liquid phase mole fractions
y vapor phase mole fractions
z feed mole fractions

Greek Symbols

γ activity coefficient
[Γ] matrix of thermodynamic factors
φ fugacity coefficient

Subscripts

i component number
j stage number
k iteration number

Superscripts

* saturated
F pertains to feed
L pertains to liquid phase
V pertains to vapor phase

Matrix Notation

[] square matrix
() vector

288
References

Amundson, N.R., and A.J. Pontinen, “Multicomponent Distillation Calculations on a

Large Digital Computer”, Ind. Eng. Chem., Vol 50 (1958) p. 730.

Amundson, N.R., and A.J. Pontinen and J.W. Tierney, AIChE J., Vol. 5 (1959) p. 730.

Ball, W.E., “Computer Programs for Distillation”, paper presented at the AIChE 44-th
National Mtg., New Orleans, Feb. (1961).

Bhargava, R. and V. Hlavacek; “Experience with Adopting One-Parameter Imbedding

Methods Toward Calculation of Countercurrent Separation Processes”, Chem. Eng.
Commun., Vol. 28 (1984) p. 165.

Billingsley, D.S., and G.W. Boynton, “Iterative Methods for Solving Problems in Mul-
ticomponent Distillation at the Steady State”, AIChE J., Vol. 17, No. 1 (1971) p. 65.

Block, U. and B. Hegner, “Development and Application of a Simulation Model for

Three Phase Distillation”, AIChE J., Vol. 22 (1976) p. 582.

Boston, J.F., and S.L. Sullivan, Jr., “A New Class of Solution Methods for Multicom-
ponent, Multistage Separation Processes”, Can. J. Chem. Eng., Vol. 52, 52 (1974) p.
52.

Boston, J.F., and H.I. Britt, “A Radically Different Formulation and Solution of the
Single-Stage Flash Problem”, Comp. Chem. Eng., Vol. 2 (1978) p. 109.

Boston, J.F., “Inside-Out Algorithms for Multicomponent Separation Process Calcula-

tions” in ”Computer Applications in Chemical Engineering Process Design and Sim-
ulation”, ACS Symposium Series No., 124 (1981) p. 135.

Boston, J.F. and V.B. Shah, “An Algorithm for Rigorous Distillation Calculations with
Two Liquid Phases”, Paper presented at AIChE National Meeting, Houston, April
(1979).

Broyden, C.G., “A Class of Methods for Solving Nonlinear Simultaneous Equations”,

Math. Comp., Vol. 19 (1965) p. 577.

Bruno, J.W., J.L. Yanosik and J.W. Tierney, “Distillation Calculations with Nonideal
Mixtures”, Adv. Chem. Number 115, 122, Am. Chem. Soc. (1972).

Burningham, D.W., and F.D. Otto, “Which Computer Design for Absorbers?”, Hy-
droc. Process., Vol. 40, 10 (1967) p. 163.

289
Burton, P.J., PhD Thesis in Chemical Engineering, Cambridge University (1986).

Burton, P.J. and W. Morton, “Differential Arclength Homotopy Continuation in Equa-

tion Oriented Simulation”, Proc. Conf. on The Use of Computers in Chemical Engi-
neering (CEF 87), p59 (1987).

Byrne, G.D. and L.A. Baird, “Research on the Convergence of a Distillation Code”,
paper presented at Spring National AIChE Meeting, March 1983, Houston, TX.

Byrne, G.D. and L.A. Baird, “Distillation Calculations Using a Locally Parameterized
Continuation Method”, Comp. Chem. Eng., Vol. 9, 6 (1985) p. 593.

Cairns, B.P. and I.A. Furzer, “Three Phase Azeotropic Distillation. Experimental Re-
sults”, I. Chem. E. Symp. Ser., 104 (1987) p. B505.

Cairns, B.P. and I.A. Furzer, “Multicomponent Three Phase Azeotropic Distillation.
1. Extensive Experimental Data and Simulation Results”, Ind. Eng. Chem. Res., Vol.
29 (1990a) pp. 1349–1363.

Cairns, B.P. and I.A. Furzer, “Multicomponent Three Phase Azeotropic Distillation.
2. Phase Stability and Phase Splitting Algorithms”, Ind. Eng. Chem. Res., Vol. 29
(1990b) pp. 1364–1382.

Cairns, B.P. and I.A. Furzer, “Multicomponent Three Phase Azeotropic Distillation.
3. Modern Thermodynamic Models and Multiple Solutions”, Ind. Eng. Chem. Res.,
Vol. 29 (1990b) pp. 1364–1382.

Chavez C, R., J.D. Seader and T.L. Wayburn; “Multiple Steady State Solutions for
Interlinked Separation Systems”, Ind. Eng. Chem. Fundam., Vol. 25, (1986) p. 566.

Cho, Y.S. and B. Joseph, “Reduced Order Steady State and Dynamic Models for Sep-
aration Processes”, AIChE J., Vol. 29 (1983) pp. 261–269, 270–276.

Cho, Y.S. and B. Joseph, “Reduced Order Models for Separation Columns - III: Ap-
plication to Columns with Multiple Feeds and Sidestreams”, Comput. Chem. Eng.,
Vol. 8, (1984) pp. 81–90.

Christiansen, L.J., M.L. Michelson and A. Fredenslund, Comput. Chem. Eng., Vol. 3
(1979) p. 535.

Ellis, M.F., R. Koshy, G. Mijares, A. Gomez-Munoz and C.D. Holland, “Use of Multi-
point Algorithms and Continuation Methods in the Solution of Distillation Problems”,
Comput. Chem. Eng., Vol. 10 (1986) p. 433.

290
Ferraris, G.B., “Interlinked, Multistaged Separators with Nonstandard Specifications
Solved by the Newton-Raphson Method”, AIChE J., Vol. 27, 1 (1981) p. 163.

Ferraris, G.B. and M. Morbidelli, “Distillation Models for Two Partially Immiscible
Liquids”, AIChE J., Vol. 27 (1981) p. 881.

Ferraris, G.B. and M. Morbidelli, “An Approximate Mathematical Model for Three
Phase Multistage Separators”, AIChE J., Vol. 27 (1982) p. 881.

Ferraris, G.B. and M. Morbidelli, “Mathematical Modelling of Multistaged Separa-

tors with Mixtures whose Components Have Largely Different Volatilities”, Comput.
Chem. Enmg., Vol. 6 (1982) p. 303.

Fletcher, J.P. “A Method for the Rigorous Calculation of Distillation Columns Using
a Generalized Efficiency Model”, Proc 4-th International Symposium on Distillation
A437, I. Chem. E., Brighton (1987)

Frantz, R.W., L.N. O’Quinn and V. Van Brunt, “Rate Process Based Continuation Ap-
plied to Hydrometallurgical Solvent Extraction”, Paper presented at AIChE National
Meeting, Miami Beach, November (1986a).

Frantz, R.W., L.N. O’Quinn and V. Van Brunt, “Stability of a Steady State Hydromet-
allurgical Solvent Extraction Model”, Paper presented at AIChE National Meeting,
Miami Beach, November (1986b).

Friday, Ph.D. Thesis, Purdue University (1963).

Friday, J.R. and B.D. Smith, “An Analysis of the Equilibrium Stage Separations Prob-
lem - Formulation and Convergence”, AIChE J., Vol. 10 (1964) p. 698.

Fuchs, R., M. Gipser and J. Gaube, “Calculation of Ternary Vapor Liquid Liquid
Equilibria for Design of Three Phase Distillation”, Fluid Phase Equilibria, Vol. 14
(1983) p. 325.

Furzer, I.A., “Ethanol Dehydration Efficiencies Using UNIFAC”, AIChE J., Vol. 31
(1985) p. 1389.

Gallun, S.E. and C.D. Holland, “Solve More Distillation Problems — Part V: For
Highly Nonideal Solutions”, Hydroc. Process., Vol. 55, 1 (1976) p. 137.

Gallun, S.E. and C.D. Holland, “A Modification of Broyden’s Method for the Solution
of Sparse Systems with Application to Distillation Problems Described by Nonideal
Thermodynamic Functions”, Comput. Chem. Eng., Vol. 4 (1980) p. 93.

291
Grohse E.W., “Plate Efficiencies in the Separation of C4 Hydrocarbons by Extractive
Distillation”, PhD Dissertation, University of Delaware (1948).

Goldstein, R.P. and R.B. Stanfield, “Flexible Method for the Solution of Distillation
Design Problems using the Newton-Raphson Technique”, Ind. Eng. Chem. Process
Des. Dev., 9, (1970) p. 78.

Haas, J.R. “Rigorous Distillation Calculations”, Chapter 4 in Kister, H. Distillation

Design, McGraw-Hill, New York (1992).

Hausen, H. “The Definition of the Degree of Exchange on Rectifying Plates for Binary
and Ternary Mixtures”, Chemie Ingr. Tech., Vol. 25 (1953) pp. 595–597.

Henley, E.J., and J.D. Seader, Equilibrium-Stage Separation Operations in Chemical

Engineering, Wiley (1981).

Hidalgo S., R., A. Correa V., A. Gomez M. and J.D. Seader, “An Optimal Arrange-
ment of Simultaneous Linearized Equations for General Systems of Interlinked, Mul-
tistaged Separators”, AIChE J., Vol. 26, 4 (1980) p. 585.

Hlavacek, V. and P. van Rompay, “Calculation of Parametric Dependence and Finite-

Difference Methods”, AIChE J., Vol. 28, 6 (1982) p. 1033.

Hlavacek, V. and P. van Rompay, “Simulation of Countercurrent Separation Processes

via a Global Approach”, Comp. Chem. Eng., Vol. 9, 4 (1985) p. 343.

Hofeling, B.S. and J.D. Seader, “A Modified Naphtali-Sandholm Method for General
Systems of Interlinked, Multistage Separators”, AIChE J., Vol. 24, 6 (1978) p. 1131.

Holland, C.D., Multicomponent Distillation, Prentice Hall, Inc., NJ (1963).

Holland, C.D. and McMahon, K.S. “Comparison of Vaporization Efficiencies with

Murphree — type Efficiencies in Distillation — I.”, Chem. Eng. Sci., Vol. 25 (1972)
pp. 431–436.

Holland, C.D., Fundamentals and Modelling of Separation Processes, Prentice Hall,

Inc., NJ (1975).

Holland, C.D., Fundamentals of Multicomponent Distillation, McGraw-Hill, Inc.;

New York (1981).

Holland, C.D., and G.P. Pendon, “Solve More Distillation Problems — Part I: Im-
provements Give Exact Answers”, Hydroc. Process., Vol. 53, 7 (1974) p. 148.

292
Holland, C.D., G.P. Pendon, and S.E. Gallun, “Solve More Distillation Problems —
Part III: Application to Absorbers”, Hydroc. Process., Vol. 54, 1 (1975) p. 101.

Huber, W.F., Hydroc. Process., August (1977) p. 121.

Ishii, Y. and F.D. Otto, “A General Algorithm for Multistage Multicomponent Sepa-
ration Calculations”, Can J. Chem. Eng., Vol. 51 (1973) p. 601.

Jelinek, J., V. Hlavacek and M. Kubicek, “Calculation of Multistage Countercurrent

Separation Processes — I. Multicomponent Multistage Rectification by Differentia-
tion with respect to an Actual Parameter”, Chem. Eng. Sci., Vol. 28 (1973) p. 1555.

Krishna, R., R.Taylor, “Multicomponent Mass Transfer — Theory and Applications”,

Chapter 7 of Handbook of Heat and Mass Transfer Operations, Volume II, Gulf Pub-
lishing Company, Houston (1986).

Ketchum, R.G., “A Combined Relaxation-Newton Method as a Global Approach to

the Computation of Thermal Separation Processes”, Chem. Eng. Sci., Vol. 34 (1979)
p. 387.

King, C.J., Separation Processes, Second Edition, McGraw Hill (1980).

Kinoshita, M., I. Hashimoto and T. Takamatsu, “A New Simulation Procedure for

Multicomponent Distillation Column Processing of Nonideal Solutions or Reactive
Solutions”, J. Chem. Eng. Japan, Vol. 16, 5 (1983) p. 370.

Kovach, J.W. and W.D. Seider, “Heterogeneous Azeotropic Distillation — Homotopy-

Continuation Methods”, Paper presented at AIChE National Meeting, Miami Beach,
November (1986a).

Kovach, J.W. and W.D. Seider, “Heterogeneous Azeotropic Distillation — Experi-

mental and Simulation Results”, Paper presented at AIChE National Meeting, Miami
Beach, November (1986b).

Kubicek, M., V. Hlavacek and F. Prochaska, “Global Modular Newton - Raphson

Technique for Simulation of an Interconnected Plant Applied to Complex Rectifica-
tion Columns”, Chem. Eng. Sci., Vol. 31 (1976) p. 277.

Lapidus, L., Digital Computation for Chemical Engineers, McGraw Hill (1962).

Lewis, W.K. and G.L. Matheson, “Studies in Distillation: Design of Rectifying Columns
for Natural And Refinery Gasoline”, Ind. Eng. Chem., Vol. 24, 5 (1932) p. 496.

Lin, W-J., J.D. Seader and T.L. Wayburn, “Computing Multiple Solutions to Systems

293
of Interlinked Separation Columns”, AIChE J., Vol. 33 (1987) p. 886.

Lucia, A., and S. Macchietto, “A New Approach to the Approximation of Quanti-

ties Involving Physical Properties Derivatives in Equation-Oriented Process Design”,
AIChE J., Vol. 29 (1983) p. 705.

Lucia, A. and K.R. Westman , “Low-Cost Solutions to Multistage, Multicomponent

Separation Problems by a Hybrid Fixed Point Algorithm”, Proceedings of the Second
International Conference on Foundations of Computer-Aided Process Design, A.W.
Westerberg and H.H. Chen, eds., CACHE (1984) p. 741.

Magnussen, T., M.L. Michelson and A. Fredenslund, “Azeotropic Distillation Using

UNIFAC”, I. Chem. E. Symp. Ser, 56, Proc. International Symposium on Distillation,
London, (1979).

Medina, A.G., Ashton, N. and McDermott, C. “Murphree and Vaporization Efficien-

cies in Multicomponent Distillation”, Chem. Eng. Sci., Vol. 33 (1978) pp. 331–339.

Medina, A.G., Ashton, N. and McDermott, C. “Hausen and Murphree Efficiencies

in Binary and Multicomponent Distillation”, Chem. Eng. Sci., Vol. 34 (1979) pp.
1105–1112.

Muller, F.R., PhD Thesis in Chemical Engineering, ETH Zurich (1979).

Murphree, E.V. “Rectifying Column Calculations”, Ind. Eng. Chem., Vol. 17 (1925)
pp. 747–750; 960–964.

Naphtali, L.M., “The Distillation Column as a Large System”, presented at AIChE

56-th National Mtg., San Francisco, May 16, (1965).

Naphtali, L.M. and D.P. Sandholm, “Multicomponent Separation Calculations by Lin-

earization”, AIChE J., Vol. 17, 1 (1971) p. 148.

Newman, J.S., Hydrocarbon Processing, Petrol. Refiner., Vol. 42, 4 (1963) p. 141.

Niedzwiecki, J.L., R.D. Springer and R.G. Wolfe, “Multicomponent Distillation in the
Presence of Free Water”, Chem. Eng. Prog., Vol. 76 (1980) p. 57.

Pierucci, S.J., E.M. Ranzi, G.E. Bardi and M. Dente; “T-Method Computes Distilla-
tion”, Hydroc. Process., Sept. (1981) p. 179.

Pierucci, S. and E. Ranzi, “T Method for Distillation Columns”, Chem. Eng. Com-
mun., Vol. 14 (1982) p. 1.

294
Pierucci, S.J., E.M. Ranzi and G.E. Biardi, “Corrected Flowrates Estimation by Using
( Convergence Promoter for Distillation Columns”, AIChE J., Vol. 29, 1 (1983) p.
113.

Powell, M.J.D., “A Hybrid Method for Nonlinear Equations”, in Numerical Methods

for Nonlinear Algebraic Equations, P. Rabinowitz, Ed., Gordon and Breach (1970).

Powers, M.F., D.J. Vickery and R. Taylor, “Simulation of Azeotropic and Extrac-
tive Distillation Operations Using a Nonequilibrium Stage Model”, Paper presented at
AIChE National Meeting, Miami Beach, November (1986).

Pucci, A., P. Mikitenko and L. Asselineau, “Three Phase Distillation, Simulation and
Application to the Separation of Fermentation Products”, Chem. Eng. Sci., Vol. 41
(1986) p. 485.

Roche, E.C., “General Design Algorithm for Multistage Counter Current Equilibrium
Processes”, Brit. Chem. Eng., Vol. 16 (1971) p. 821.

Rose, A., R.F. Sweeney, and V.N. Schrodt, “Continuous Distillation Calculations by
Relaxation Method”, Ind. Eng. Chem., Vol. 50, 5 (1958) p. 737.

Ross, B.A. and W.D. Seider, “Simulation of Three Phase Distillation Towers”, Com-
put. Chem. Eng., Vol. 5 (1981) p. 7.

Russell, R.A., “A Flexible and Reliable Method solves single tower and crude distil-
lation problems”, Chem. Eng October 17 (1983) pp. 53–59.

Saeger, R. and P.R. Bishnoi, “A Modified ‘Inside-Out’ Algorithm for Simulation of

Multistage Multicomponent Separation Processes Using the UNIFAC Group Contri-
bution Method”, Can. J. Chem. Eng., Vol. 64 (1986) pp. 759–767.

Salgovic, A., V. Hlavacek and J. Ilavsky, “Global Simulation of Countercurrent Sep-

aration Processes via One-Parameter Imbedding Techniques”, Chem. Eng. Sci., Vol.
36, 10 (1981) p. 1599.

Schoenborn, E.M., J.H. Koffolt and J.R. Withrow, “Rectification in the Presence of an
Insoluble Component”, Trans AIChE (1941).

Schuil J.A. and K.K. Bool, “Three-Phase Flash and Distillation”, Comp. Chem. Eng.,
Vol. 9, 3 (1985) p. 295.

Seader, J.D., “The BC (Before Computers) and AD of Equilibrium Stage Operations”,

Chem. Eng. Ed., Spring, 88, (1985a).

295
Seader, J.D., “Computer Modeling of Chemical Processes”, AICheE Monograph Se-
ries 15, Vol. 81 (1985b).

Seader, J.D. and E.J. Henley Separation Process Principles, Wiley, New York (1998).

Seader, J.D., C.C. Rafael and T.L. Wayburn; “Multiple Solutions to Systems of In-
terlinked Distillation Columns by Differential Homotopy Continuation”, paper 63f,
AIChE National Meeting, San Francisco, CA; Nov. (1984).

Sereno, A.M., Doctoral Thesis, University of Porto, Portugal (1985).

Shah, M.K. and P.R. Bishnoi, Can. J. Chem. Eng., Vol. 56 (1978) p. 478.

Smith. B.D., Design of Equilibrium Stage Processes, McGraw-Hill, New York (1964).

Sridhar, L.N. and A. Lucia, “Analysis of Multicomponent Separation Processes: Fixed

Temperature and Fixed Pressure Profiles” Ind. Eng. Chem. Res., Vol. 29 (1989) pp.
1668–1675.

Sridhar, L.N. and A. Lucia, “Analysis and Algorithms of Multicomponent Separation

Processes”, Ind. Eng. Chem. Res., Vol. 29 (1990a) pp. 793-803.

Sridhar, L.N. and A. Lucia, “Tearing Algorithms for Separation Process Simulation”,
Comp. Chem. Eng., Vol. 14 (1990b) pp. 901–905.

Stadtherr, M.A. and M.A. Malachowski, “On Efficient Solution of Complex Systems
of Interlinked Multistage Separators”, Comp. Chem. Eng., Vol. 6, 2 (1982) p. 121.

Stainthorp, F.P., and P.A. Whitehouse, “General Computer Programs for Multi Stage
Counter Current Separation Problems — I: Formulation of the Problem and Method
of Solution”, I. Chem. E. Symp. Series, Vol. 23 (1967) p. 181.

Standart, G.L. “Studies on Distillation — V. Generalized Definition of a Theoretical

Plate or Stage of Contacting Equipment”, Chem. Eng. Sci., Vol. 20 (1965) pp. 611–
622.

Standart, G.L. “Comparison of Murphree Efficiencies with Vaporization Efficiencies”,

Chem. Eng. Sci., Vol. 26 (1971) pp. 985–988.

Stewart, W.E. “Simulation of Fractionation by Orthogonal Collocation”, Chem. Eng.

Sci., Vol. 40 (1985) pp. 409–421.

Sujata, A.D., “Absorber-Stripper Calculations Made Easier”, Hydroc. Process., Vol.

40, 12 (1961) p. 137.

296
Swartz, C.L.E. and W.E. Stewart, “A Collocation Approach to Distillation Column
Design”, AIChE J., Vol. 32, (1986) pp. 1832–1838.

Taylor, R. and R. Krishna, “Modelling Reactive Distillation”, Chem. Eng. Sci., Vol.
55 (2000).

Thiele, E.W., and R.L. Geddes, “Computation of Distillation Apparatus for Hydrocar-
bon Mixtures”, Ind. Eng. Chem., Vol. 25 (1933) p. 289.

Tierney, J.W. and J.A. Bruno, “Equilibrium Stage Calculations”, AIChE J., Vol. 13, 3
(1967) p. 556.

Tierney, J.W., and J.L. Yanosik, “Simultaneous Flow and Temperature Correction in
the Equilibrium Stage Problem”, AIChE J., Vol. 15, 6, (1969) p. 897.

Tomich, T.F., “A New Simulation Method for Equilibrium Stage Processes”, AIChE
J., Vol. 16, 2 (1970) p. 229.

Van Winkle, M., Distillation, McGraw-Hill Book Co., New York, NY (1967).

Verneuil, V.S., and A.P. Oleson, “Steady-state Distillation via Transient-state Calcula-
tions”, presented at 161-st ACS National Mtg., Los Angeles, March (1971).

Vickery, D.J. and R. Taylor, “Path Following Algorithms to the Solution of Multicom-
ponent, Multistage Separation Process Problems”, AIChE J., Vol. 32, 4 (1986a) p.
547.

Vickery, D.J. and R. Taylor, “A Thermodynamic Continuation Method for the Solu-
tion of Multicomponent, Multistage Separation Process Problems”, paper 50e AIChE
Spring Nat. Mtg., New Orleans, April (1986b)

Vickery, D.J., J.J. Ferrari and R. Taylor, “An ’Efficient’ Continuation Method for
the Solution of Difficult Equilibrium Stage Separation Process Problems”, Comput.
Chem. Eng., (1987)

Vickery D.J., PhD Thesis in Chemical Engineering, Clarkson University (1987)

Waggoner, R.C. and C.D. Holland, “Solution of Problems Involving Conventional

and Complex Distillation Columns at Unsteady State Operation”, AIChE J., Vol. 11,
1 (1965) p. 113.

Wang, J.C. and G.E. Henke, “Tridiagonal Matrix for Distillation”, Hydroc. Process.,
Vol. 45, 8 (1966) p. 155.

297
Wang, J.C. and Y.L. Wang; “A Review on the Modeling and Simulation of Multi-
Stage Separation Processes”, Foundations of Computer-Aided Chemical Process De-
sign, Vol. II; R.S.H. Mah and W.D. Seider, eds.; Engineering Foundation; (1981) p.
121.

Wang, Y.L., and A.P. Oleson, cited as private communication in Wang and Wang,
1981, (1964).

Walas, S.M. Phase Equilibria in Chemical Engineering, Butterworth Publishers, (1985).

Wayburn, T.L. Ph.D. Thesis, Univ. of Utah (1983).

Wayburn, T.L. and J.D. Seader; “Solutions of Systems of Interlinked Distillation

Columns by Differential Homotopy-Continuation Methods”, Proceedings of the Sec-
ond International Conference on Foundations of Computer-Aided Process Design,
A.W. Westerberg and H.H. Chen, eds., CACHE (1983) p. 765.

Wayburn, T.L., and J.D. Seader, “Degree Theory and Homotopy: Tools for Computer-
Aided Process Design”, paper 27d, AIChE Nat. Meeting, Nov. 25-30, San Francisco,
CA, Nov. (1984).

Whitehouse, P.A., PhD Thesis in Chemical Engineering, University of Manchester

Institute of Science and Technology, April (1964).

Westerberg, A.W., H.P Hutchison, R.L. Motard and P. Winter, Process Flowsheeting,
Cambridge University Press, (1979).

Westman, K.R., A. Lucia and D. Miller, “Flash and Distillation Calculations by a

Newton-like Method”, Comp. Chem. Eng, Vol. 8 (3/4), (1984) p. 219.

Woodman, M.R., W. Morton and W.R. Paterson, “Simulation of the Distillation of

Three Phase Mixtures: A new Algorithm Predicts New Phenomena”, Paper presented
at CHISA, Prague, (1990).

Yu, P., I.M. Elshayal and B.C-Y. Lu, “Liquid-liquid-vapor Equilibria in the Nitrogen
Methane Ethane System”, Can. J. Chem. Eng., Vol. 47 (1969) p. 495.

298
Chapter 14

Nonequilibrium Column
Modelling

The nonequilibrium model and the model equations are described in detail in three
chapters. In this chapter we focus our attention on those aspects of the nonequilibrium
model that pertain to all such operations. Subsequent chapters include details of the
model that are specific to gas/vapor-liquid separations (distillation and absorption) and
to liquid-liquid extraction.

14.1 The Model Equations

A second generation nonequilibrium model was developed by Taylor and coworkers

and is described in detail by Taylor et al. (1994) and Taylor and Krishna (1993).
It can be used to simulate trayed columns as well as packed columns. A schematic
diagram of a nonequilibrium stage is shown in Figure 14.1. This stage may represent
one (or more than one) tray in a trayed column or a section of packing in a packed
column. The vertical wavy line in the middle of the diagram represents the interface
between the two phases which may be vapor and liquid (distillation), gas and liquid
(absorption) or two liquids (extraction).

Figure 14.1 also serves to introduce the notation used in writing down the equations
that model the behavior of this nonequilibrium stage. The flow rates of the light and
heavy phases leaving the j-th stage are denoted by Vj and Lj respectively. The mole

299
 V
j

L vapor
j-1 side draw
x y
i,j-1 i,j
HL HV
j-1 j
TL TV
j-1 j

Stage j
N V
QjL Qj

E
fL fV
i,j i,j
HL HV
j j
x V
i,j j+1
HL y
i,j+1
j
TL HV
j j+1
TV
j+1
liquid
side draw
L
j

Figure 14.1: Schematic diagram of a nonequilibrium stage (after Taylor and Krishna,
1993).

fractions in these streams are yij and xij . The Nij are the molar fluxes of species
i on stage j. When multiplied by the area available for interphase mass transfer we
obtain the rates of interphase mass transfer. The temperatures of the two phases are
not assumed to be equal and we must allow for heat transfer as well as mass transfer
across the interface.

If Figure 14.1 represents a single tray then the term φL j is the fractional heavy phase
entrainment defined as the ratio of the moles of heavy phase entrained in the light
phase in stage j to the moles of downwards flowing heavy phase from stage j. Sim-
ilarly, φVj is the ratio of the light phase entrained in the heavy phase leaving stage j.
For packed columns, this term represents axial dispersion. Weeping in tray columns
may be accounted for with a similar term.

The component material balance equations for each phase may be written as follows:

V
Mij ≡ (1 + rjV + φVj )Vj yij − Vj+1 yi,j+1 − φVj−1 Vj−1 yi,j−1

300
 n
X
V
−fij − GVijν + Nij
ν=1
= 0 i = 1, 2, . . . , c (14.1)
L
Mij ≡ (1 + rjL + φL L
j )Lj xij − Lj−1 xi,j−1 − φj+1 Lj+1 xi,j+1
Xn
L
−fij − GLijν − Nij
ν=1
= 0 i = 1, 2, . . . , c (14.2)
where Gijν is the interlinked flow rate for component i from stage ν to stage j, and n
is the number of total stages (trays or sections of packing). The last terms in Equations
(14.1) and (14.2) are the mass transfer rates (in kmol/s), where mass transfer from
the “V” phase to the “L” phase is defined as positive. At the V/L interface we have
continuity of mass and, thus, the mass transfer rates in both phases must be equal.

The total material balances for the two phases are obtained by summing Equations
(14.1) and (14.2) over the component index i.
MtVj ≡ (1 + rjV + φVj )Vj − Vj+1 − φVj−1 Vj−1
c X
X n
−FjV − GVijν + Ntj = 0 (14.3)
i=1 ν=1

MtLj ≡ (1 + rjL + φL L
j )Lj − Lj−1 − φj+1 Lj+1
c X
X n
−FjL − GL
ijν − Ntj = 0 (14.4)
i=1 ν=1
Pc
Fj denotes the total feed flow rate for stage j, Fj = i=1 fij .

Here total flow rates and mole fractions are used as independent variables and total
as well as component material balances are included in the set of independent model
equations. In the nonequilibrium model of Krishnamurthy and Taylor (1985) compo-
nent flow rates were treated as variables.

The nonequilibrium model uses two sets of rate equations for each stage:
V V
Rij ≡ Nij − Nij =0 i = 1, 2 . . . , c − 1 (14.5)
L L
Rij ≡ Nij − Nij =0 i = 1, 2 . . . , c − 1 (14.6)
where Nij is the mass transfer rate of component i on stage j. The mass transfer rate
in each phase is computed from a diffusive and a convective contribution with
V
Nij = aIj Jij
V
+ yij Ntj (14.7)
L
Nij = aIj Jij
L
+ xij Ntj (14.8)

301
where aIj is the total interfacial area for stage j and Ntj is the total mass transfer rate
Pc
on stage j (Ntj = i=1 Nij ). The diffusion fluxes J are given by (in matrix form):

(J V ) = cVt [k V ](y V − y I ) (14.9)

L
(J ) = cL L
t [k ](x
I − xL ) (14.10)

where (y V − y I ) and (xI − xL ) are the average mole fraction difference between the
bulk and the interface mole fractions (Note that the fluxes are multiplied by the inter-
facial area to obtain mass transfer rates). The average mole fraction differences are
calculated depends on the chosen flow model (more on this topic later). The matrices
of mass transfer coefficients, [k], are calculated from

[k P ] = [RP ]−1 [ΓP ] (14.11)

where [ΓP ] is a matrix of thermodynamic factors for phase P . For systems where an
activity coefficient model is used for the phase equilibrium properties the thermody-
namic factor matrix Γ (order c − 1) is defined by
 
∂ ln γi
Γij = δij + xi (14.12)
∂xj T,P,xk ,k6=j=1...c−1

If equations of state are used γi is replaced by φi . Expressions for the composition

derivatives of ln γi are given by Taylor and Kooijman (1991). The rate matrix [R]
(order c − 1) is a matrix of mass transfer resistances calculated from the following
formulae:
c
P zi X zk
Rii = P
+ P
(14.13)
kic k=1,k6=i kik
!
P 1 1
Rij = −zi P
− P (14.14)
kij kic

P
where kij are binary pair mass transfer coefficients for phase P . Mass transfer co-
efficients, kij , are computed from empirical models given in the next two chapters
(see, also, Taylor and Krishna, 1993). Equations (14.13) and (14.14) are suggested
by the Maxwell-Stefan equations that describe mass transfer in multicomponent sys-
tems (see Taylor and Krishna, 1993). The matrix of thermodynamic factors appears
because the fundamental driving force for mass transfer is the chemical potential gra-
dient and not the mole fraction or concentration gradient. This matrix is calculated
from an appropriate thermodynamic model. Thus, the mass transfer coefficients are
of central importance in the nonequilibrium model; indeed, sometimes it is possible to
predict different behavior of a column by selecting different mass transfer coefficient
correlations.

302
Note that there are c(c−1)/2 binary pair Maxwell-Stefan diffusion coefficients and c−
1 times c−1 elements in the [RP ] and [k P ] matrices and, therefore, only c−1 equations
per set of rate equations. This is the result of the fact that diffusion calculations only
yield relative transfer rates. We will need an extra equation that will ”bootstrap” the
mass transfer rates: the energy balance for the interface.

The energy balance equations on stage j are written for each phase as follows:

EjV ≡ (1 + rjV + φVj )Vj HjV − Vj+1 Hj+1

V
− φVj−1 Vj−1 Hj−1
V

Xn
−FjV HjV F − GVjν Hjν
V
+ QVj + eVj = 0 (14.15)
ν=1

EjL ≡ (1 + rjL + φL L L L L
j )Lj Hj − Lj−1 Hj−1 − φj+1 Lj+1 Hj+1
Xn
−FjL HjLF − GL L L L
jν Hjν + Qj − ej = 0 (14.16)
ν=1

where Gjν is the interlink flow rate from stage ν to stage j. The last term in the left-
hand-side of Equations (14.15) and (14.16), ej , represents the energy transfer rates for
the two phases and are defined by
c
X
eVj = aIj hV (T V I
−T )+ V
Nij V
H̄ij (14.17)
i=1
Xc
eL
j = aIj hL (T I − T L ) + L L
Nij H̄ij (14.18)
i=1

where H̄ij are the partial molar enthalpies of component i for stage j. We also have
continuity of the energy fluxes across the V/L interface which gives the interface en-
ergy balance:
EjI ≡ eVj − eLj =0 (14.19)

where hV and hL are the vapor and liquid heat transfer coefficients respectively, and
T V , T I , and T L the light phase, interface, and heavy phase temperatures. For the
calculation of vapor phase heat transfer coefficients the Chilton-Colburn analogy be-
tween mass and heat transfer is used:

hV = kρCp Le2/3 (14.20)

where
λ Sc
Le = = (14.21)
DCp ρ Pr

303
For the calculation of the liquid phase heat transfer coefficients a penetration model is
used:
√
hL = kρCp Le (14.22)

where k is the average mass transfer coefficient and D the average diffusion coeffi-
cient.

In the nonequilibrium model of Krishnamurthy and Taylor (1985) the pressure was
taken to be specified on all stages. However, column pressure drop is a function of
tray or packing type and design and column operating conditions, information that is
required for or available during the solution of the nonequilibrium model equations. It
is, therefore, quite straightforward to add an hydraulic equation to the set of indepen-
dent equations for each stage and to make the pressure of each stage (tray or packed
section) an unknown variable. The stage is assumed to be at mechanical equilibrium
so pVj = pL j = pj .

In the second generation model, the pressure of the top tray (or top of the packing) is
specified along with the pressure of any condenser. The pressure of trays (or packed
sections) below the topmost are calculated from the pressure of the stage above and the
pressure drop on that tray (or over that packed section). If the column has a condenser
(which is numbered as stage 1 here) the hydraulic equations are expressed as follows:

P1 ≡ pc − p1 = 0 (14.23)

P2 ≡ pspec − p2 = 0 (14.24)

Pj ≡ pj − pj−1 − (∆pj−1 ) = 0 j = 3, 4, . . . , n (14.25)

where pc is the specified condenser pressure, pspec is the specified pressure of the tray
or section of packing at the top of the column, and ∆pj−1 is the pressure drop per tray
or section of packing from section/stage j − 1 to section/stage j. If the top stage is not
a condenser, the hydraulic equations are expressed as

P1 ≡ pspec − p1 = 0 (14.26)

Pj ≡ pj − pj−1 − (∆pj−1 ) = 0 j = 2, 3, . . . , n (14.27)

In general we may consider the pressure drop to be a function of the internal flows,
the fluid densities, and equipment design parameters.

∆pj−1 = f (Vj−1 , Lj−1 , ρVj−1 , ρL

j−1 , Design) (14.28)

The calculation of the pressure drop is discussed in the next two chapters.

304
Phase equilibrium is assumed to exist only at the interface with the mole fractions in
both phases related by:
QIij ≡ Kij xIij − yij
I
=0 i = 1, 2, . . . , c (14.29)
where Kij is the equilibrium ratio for component i on stage j. The Kij are evaluated
at the (calculated) temperature, pressure, and mole fractions at the interface.

The mole fractions must sum to unity in each phase:

c
X
SjV ≡ yij − 1 = 0 (14.30)
i=1
c
X
SjL ≡ xij − 1 = 0 (14.31)
i=1
as well as at the interface:
c
X
SjV I ≡ I
yij −1=0 (14.32)
i=1
c
X
SjLI ≡ xIij − 1 = 0 (14.33)
i=1

14.1.1 Degrees of Freedom and Specifications

Table 14.1 lists the type and number of equations for the nonequilibrium model. The
model has 5c + 6 equations and variables, where c is the number of components.
Nonequilibrium and equilibrium models require similar specifications. Feed flows
and their thermal condition must be specified for both models, as must the column
configuration (number of stages, feed and sidestream locations etc.). Additional spec-
ifications that are the same for both simulation models include the specification of, for
example, reflux ratios or bottom product flow rates if the column is equipped with a
condenser and/or a reboiler. The specification of the pressure on each stage is neces-
sary if the pressure drop is not computed; if it is, only the top stage pressure needs
be specified (the pressure of all other stages being determined from the pressure drop
equations that are part of the model described above).

14.1.2 Solving the Model Equations

The nonequilibrium model equations are solved simultaneously using Newton’s method
as described in Section 11.2. This requires initial estimates for all the variables. Chem-

305
 Table 14.1: Nonequilibrium model equations type and number

Equation Number
Material balances 2c + 2
Energy balances 3
transfer Rate equations 2c − 2
Summations equations 2
Hydraulic equation 1
interface eQuilibrium relations c
Total MERSHQ 5c + 6

Table 14.2: Currently supported column internals

Bubble-cap trays
Sieve trays
Valve trays (including double weight valves)
Dumped packings
Structured packings
Equilibrium stage (with Murphree stage efficiency)
Rotating Disk Contactor (RDC) compartment (for extraction)

Sep uses the same automatic initialization procedure for the nonequilibrium model
and the equilibrium model. The temperatures of the vapor, interface, and liquid are
set equal to each other. Mass and energy transfer rates are initialized as zero and the
interface mole fractions are set equal to the bulk mole fractions as estimated for the
equilibrium stage model. Pressure drops initially are assumed to be zero.

14.2 Model Requirements

A nonequilibrium simulation cannot proceed without some knowledge of the column

type and the internals column design. Equipment design details may be needed in or-
der to calculate the mass transfer coefficients. Table 14.2 lists the currently supported
types of column internals.

The nonequilibrium model requires the evaluation of many more physical properties
and of the heat and mass transfer coefficients. For the evaluation of the heat and mass
transfer coefficients, pressure drop, and the entrainment/weeping flows a nonequilib-

306
rium simulation needs the following:

• Column internals type

• Column internals layout or design mode parameters (such as fraction of flooding

etc.)

• Mass transfer coefficient model

• Flow model for both phases

• Entrainment and weeping models

• Pressure drop model

Flow models are discussed below. Mass transfer coefficient models are at the heart of
the nonequilibrium model. All of the models incorporated in ChemSep are from the
published literature. The choice of mass transfer coefficient can influence the results
of a simulation. For this reason, we have tried – in the following two chapters that
deal with distillation (and absorption) operations and with liquid-liquid extraction – to
document the origin of each method in order to guide you in selecting models.

14.3 Flow Models

For the calculation of the diffusion fluxes the average mole fraction difference between
the bulk and the interface mole fractions were required (see Equations 14.9 and 14.10).
How these average mole fraction differences are computed depends on the flow model.
Two flow models are discussed: mixed flow and plug flow.

14.3.1 Mixed flow

If we assume both phases are present in a completely mixed state, we can use

(y V − y I ) = (y V − y I ) (14.34)
I L
(xI − xL ) = (x − x ) (14.35)

This keeps the rate equations (relatively) simple and only a function of the mole frac-
tions leaving the current stage. However, on a tray where the vapor bubbles through
a liquid which flows from one downcomer to the opposite downcomer this model is

307
not accurate. Indeed, only for very small diameter columns will the mixed flow model
give reasonable results. The mixed model is the simplest flow model and is the easiest
to converge. Convergence to the true column profiles for packed columns by using
increasing number of stages can be quite slow using the mixed flow model.

14.3.2 Plug flow

In the plug flow model we assume that the vapor or liquid moves in plug flow (without
mixing) through the froth. This complicates the rate equations so much that no exact
solution is possible. The mass transfer rate equations need to be integrated over the
froth. Kooijman and Taylor (1995) derived expressions for the average vapor and
liquid compositions assuming constant mass transfer coefficients and that the interface
compositions is constant (it isn’t, but its ”average” value is obtained):

(y V − y I ) = Ω[−NV ](y V − y I ) (14.36)

L I L
(xI − xL ) = Ω[−N ](x − x ) (14.37)

where the mole fractions refer to the leaving streams and the number of mass transfer
units (N) for the vapor and liquid are defined as:

NV = cVt k V aV hf Ab /V (14.38)
L
N = cL L L
t k a hf Ab /L (14.39)

Ω[M ] is a matrix function defined as

Ω[M ] = [exp[M ] − [I]][M ]−1 [exp[M ]]−1 = [exp[−M ] − [I]][−M ]−1 (14.40)

The efficiencies predicted for tray columns are more accurately represented with this
model. The plug flow model can also be used for packed columns, providing much
faster convergence to the true column profiles when compared to the mixed flow
model.

Currently no correction terms is applied to the plug flow model to correct for the
change in mole fractions over the integration (as is discussed by Kooijman and Taylor,
1995).

14.4 The Design Mode

The nonequilibrium column simulator has an optional design mode to automatically

assign equipment design parameters. The user just needs to select one of the types

308
of internals (for each section in the column). The design-mode is activated simply by
failing to specify the column diameter (leaving it as a ”default” with ”*”) for a specific
section. Other layout parameters can be specified but they may be changed by the
design mode. Packed bed heights must be specified since this parameter determines
the desired separation and the capacity.

The internal layout is determined after the flows have been estimated. Each stage in
the column is designed separately and independently of adjacent stages. The different
designs then are rationalized so that all stages within a particular section of the column
have the same layout. During each iteration (that is, an update of the flows) each
stage is re-designed only if the flowrates have changed by more than by a certain
fraction (that can be specified). Only sections with re-designed stages are rationalized
again. After convergence a complete design of any trayed or packed section in the
column is obtained. Different types of column internal can be freely mixed in a column
simulation/design.

The different design methods employed generate a column-design that might not be
optimal from an engineers viewpoint. They must be seen as starting points for the
actual design layouts. Also, the design does not include calculations to determine
equipment support details, or the thicknesses of trays or the column.

Symbol List

Latin Symbols

c Number of components,
Molar concentration (kmol/m3 )
D Binary diffusivity coefficient (m2 /s)
e Energy transfer rate (J/s)
fij Component i feed flow to stage j (kmol/s)
Fj Total feed flow rate to stage j (kmol/s)
G Interlinked flow rate (kmol/s)
h Heat transfer coefficient (J/m2 /K/s)
H Molar enthalpy (J/kmol)
H̄i Partial molar enthalpy of component i (J/kmol)
J Molar diffusion flux (kmol/m2 s)
k Binary mass transfer coefficient (m/s)
Ki K-value or equilibrium ratio component i: Ki = yi /xi
L Liquid flow rate (kmol/s)

309
Le Lewis number (Le = Sc/Pr)
M Mass flow rate (kg/s)
N Mass transfer rate (kmol/s)
n Number of stages
p Pressure (P a)
∆p Pressure drop (P a)
Q Heat input (J/s)
r Ratio sidestream to internal flow
[R] Matrix defined by (14.13) and (14.14)
T Temperature (K)
V Vapor flow rate (kmol/s)
x Liquid mole fraction
y Vapor mole fraction
z Mole fraction

Greek Symbols

[Γ] Thermodynamic matrix

λ Heat conductivity (W/m/K)

Superscripts

I Interfacial
L Heavy phase
P Phase P
V Light phase

Subscripts

i component i
j stage j,
component j
t total
ν from interlinking stage ν

310
References

H.A. Kooijman, R. Taylor, “Modelling Mass Transfer in Multicomponent Distilla-

tion”, Chem. Eng. J., Vol. 57, No. 2 (1995) pp. 177–188.

R. Krishnamurthy, R. Taylor, “A Nonequilibrium Stage Model of Multicomponent

Separation Processes. Part I: Model Description and Method of Solution”, AIChE J.,
Vol. 31, No. 3 (1985) pp. 449–455.

R. Taylor, H.A. Kooijman, “Composition derivatives of Activity Models (for the es-
timation of Thermodynamic Factors in Diffusion)”, Chem. Eng. Comm., Vol. 102
(1991) pp. 87–106.

R. Taylor, H.A. Kooijman, J-S. Hung, “A second generation nonequilibrium model

for computer simulation of multicomponent separation processes”, Comput. Chem.
Engng., Vol. 18, No. 3 (1994) pp. 205–217.

R. Taylor, R. Krishna, Multicomponent Mass Transfer, Wiley, New York (1993).

311
312
Chapter 15

Design Models for Trayed

Columns

This chapter serves as a guide to the correlations in ChemSep for estimating mass
transfer coefficients and the pressure drop for distillation (and absorption) trayed columns.
The methods used in Design Mode to determine column design parameters also are
discussed in detail. The chapter ends with the comparison of experimental distillation
data and nonequilibrium model simulations.

15.1 Mass Transfer Coefficient Correlations

Table 16.1 provides a summary of the available correlations for trays and packings;
the various correlations are discussed in more detail below. Recommended models
are shown in boldface.

Binary mass transfer coefficients (MTC’s) can be computed from the Number of
Transfer Units (NTU’s = N) by:
kV = NV /tV aV (15.1)
L L L
k = N /tL a (15.2)
where the vapor and liquid areas are calculated with
aV = ad /hf (15.3)
L
a = ad /αhf (15.4)

313
 Table 15.1: Available mass transfer coefficient correlations per internals type

Bubble-Cap Sieve Valve

tray tray tray
AIChE AIChE AIChE
Hughmark Chan-Fair
Zuiderweg
Chen-Chuang
Harris
Bubble-Jet

the interfacial area density may be estimated from Zuiderweg’s (1982) method (see
below).

AIChE One of the oldest methods for estimating numbers of transfer units came from
the AIChE tray efficiency research program of the 1950s. The correlations can
be used for sieve trays, valve trays, and bubble-cap trays.
p
NV = (0.776 + 4.57hw − 0.238Fs + 104.8QL /Wl )/ ScV (15.5)
√
NL = 19700 DL (0.4Fs + 0.17)tL (15.6)
where
q
Fs = us ρVt (15.7)
ScV = η V /ρVt DV (15.8)
tL = hL ZWl /QL (15.9)
The clear liquid height hL is computed from a correlation due to Bennett et al.
(1983):
hL = αe hw + C(QL /αe Wl )0.67


(15.10)
αe = exp(−12.55(us (ρV /(ρL − ρV ))0.5 )0.91 ) (15.11)
C = 0.50 + 0.438 exp(−137.8hw ) (15.12)

Chan-Fair The number of transfer units for the vapor phase is:
√ p
NV = (10300 − 8670F F )F F DV tV / hL (15.13)
tV = (1 − αe )hL /(αe us ) (15.14)
The AIChE correlation is used for the number of transfer units for the liquid
phase. (hL and αe also are computed with the correlation of Bennett et al.).

314
Zuiderweg The vapor phase mass transfer coefficient is

k V = 0.13/ρVt − 0.065/(ρVt )2 (15.15)

Note that k V is independent of the diffusion coefficient. The liquid mass trans-
fer coefficient is computed from either:

k L = 2.6 10−5 (η L )−0.25 (15.16)

or
k L = 0.024(DL )0.25 (15.17)
The interfacial area in the spray regime is computed from:
0.37
Us2 ρVt hL F P

40
ad hf = (15.18)
φ0.3 σ
and in the froth-emulsion regime from:
0.37
Us2 ρVt hL F P

43
ad hf = (15.19)
φ0.3 σ
The transition from the spray to mixed froth-emulsion flow is described by:

F P > 3bhL (15.20)

where b is the weir length per unit bubbling area:

b = Wl /Ab (15.21)

The clear liquid height is given by:

0.5
hL = 0.6hw (pF P/b)0.25 (15.22)

Hughmark The numbers of transfer units are given by:

r
ρL
NV = (0.051 + 0.0105Fs ) (15.23)
Fs
p
NL = (−44 + 10.7747 104 QL /Wl + 127.1457Fs ) DL Abub /QL (15.24)

Harris The numbers of transfer units are given by:

0.3 + 15tG
NV = √ (15.25)
ScG
5 + 10tL (1 + 0.17(0.82Fs − 1)(39.3hw + 2))
NL = √ (15.26)
ScL

315
Chen-Chuang The numbers of transfer units for the vapor is:
hl
tV = (15.27)
us
√
Fs = Us ρV (15.28)
2 1/3 p
 
1 ρL Fs
NV = 11 0.1 β 0.14 DV tV (15.29)
ηL σ2

and for the liquid

ρL
tL = tV (15.30)
ρV
1/3 
ρL Fs2
 p
L 1 V
N = 14 0.1 0.14 D L tL (15.31)
ηL β σ2 L

Bubble-Jet This is a combination of a theoretical model described by Taylor and Kr-

ishna (1993) and empirical models (Prado, 1986, Prado and Fair, 1990) to de-
termine bubble sizes and velocities.

15.2 Pressure Drop Models

There are as many ways to compute tray pressure drops as there to estimate mass
transfer coefficients. For packings there is a move away from the use of generalized
pressure drop charts (GPDC) to more theoretically based correlations. We have cho-
sen to employ the most recently published models. For packings there are at least 7
methods available (see Table 16.2). For packings operating above the loading point
(F F > 0.7) we advise the use of models that take the correction for the liquid holdup
into account, such as the SBF-89. However, disadvantage is that these models can
have complex (imaginary) solutions, especially at high fractions of flood. This can
cause non-cnvergence! The Lev-92 model can be an alternative for it includes a de-
pendence on the liquid flow rate to simulate the increased pressure drop at loading
conditions. When the pressure drop is specified as fixed, it is assumed zero.

15.2.1 Tray pressure drop Estimation

The liquid heights on the trays are evaluated from the tray pressure drop calculations.
The wet tray pressure drop liquid height is calculated from:

hwt = hd + hl (15.32)

316
 Table 15.2: Pressure drop correlations per internals type

Bubble-Cap Sieve Valve

tray tray tray
Fixed Fixed Fixed
Estimated Estimated Estimated

where hd is the dry tray pressure drop liquid height and hl the liquid height:
hlg
hl = hcl + hr + (15.33)
2
The clear liquid height, hcl , is calculated with
hcl = αhw + how (15.34)
where the liquid fraction of the froth, α, is computed with the Barker and Self (1962)
correlation:
0.37hw + 0.012Fs + 1.78QL /Wl + 0.024
α= (15.35)
1.06hw + 0.035Fs + 4.82QL /Wl + 0.035
The choice of correlation for the liquid fraction turns out to be important as certain
correlations are dynamically unstable. The height of liquid over the weir, how , is
computed by various correlations for different types of tray weirs (see Perry’s Hand-
book, 1984) and a weir factor (Fw ) correction (see Smith, pp. 487) is employed. For
example for a segmental weir:
 2/3
QL
how = 0.664Fw (15.36)
Wl
Wl
w = (15.37)
Dc
w2
Fw3 = √ (15.38)
1 − (Fw w( 1.68Q
W 2.5
) + 1 − w2 )2
L 2/3
l

where QL is the volumetric flow over the weir per weir length. The residual height,
hr , is only taken into account for sieve trays. Bennett’s method (see Lockett, 1986,
pp. 81) is:
   2/3  1/3
6 σ ρL − ρV
hr = (15.39)
1.27ρL g dh
Dry tray pressure, hd , is calculated with:
ρG 2
hd = K u (15.40)
ρL h
ξ
K = (15.41)
2g

317
where the orifice coefficient ξ for sieve trays is computed as described by Stichlmair
and Mersmann (1978). For valve trays we use the method of Klein (1982) as described
in Kister (1992, pp. 309–312) where K is given for the cases with the valves closed
or open. It is extended for double weight valve trays as discussed by Lockett (1986,
pp. 82–86). The dry tray pressure drop is corrected for liquid fractional entrainment.

The froth density is computed from

hcl
hf = (15.42)
α
The liquid gradient, hlg , is estimated from a method due to Fair (Lockett, 1986, pp.
72):
W hf
Rh = (15.43)
W + 2hf
QL
Uf = (15.44)
W hcl
Rh Uf ρL
Ref = (15.45)
ηL
f = 7 × 104 hw Ref−1.06 (15.46)
Zf Uf2
hlg = (15.47)
gRh
where W is the average flow-path width for liquid flow, and Z the flow path length.
The height of liquid at the tray inlet is:
s  2 
2αh2f

2 QL 1 1
hi = − + (15.48)
g Wl hcl hc 3

where hc is the height of the clearance under the downcomer. The pressure loss under
downcomer (expressed as a liquid height) is
  2
1 QL
hudc = (15.49)
2g Cd Wl hc

where Cd = 0.56 according to Koch design rules. The height of liquid in the down-
comer can now be calculated with the summation:

hdb = hwt + hi + hudc (15.50)

Liquid heights on bubble-cap trays are estimated from a method given in Perry’s
(1984) handbook and in Smith (1963). The liquid fraction of the froth is computed
according to Kastanek (1970).

318
15.3 Entrainment and Weeping

Entrainment and weeping flows (for trays only) change the internal liquid flows and
influence the performance of the column internals. ChemSep currently does not sup-
port the handling of these flows. This is due to the fact that few entrainment models
behave properly. Neither is the effect of the entrainment and weeping flows on the
mass transfer properly taken into account.

Entrainment can be estimated from the fractional liquid entrainment from Hunt’s cor-
relation and from Figure 5.11 of Lockett (1986) for sieve trays:
   3.2
0.073 Uv
φL = 7.75 10−5 Mv (15.51)
σ Ts − 2.5hcl
The weep factor is estimated from a figure from Smith (1963, p. 548), that was fitted
to the following correlation
0.135 φ ln(34(Hw + How ) + 1)
WF = (15.52)
(Hd + Hr )
where φ is the open area ratio.

15.4 Packing Flooding and Minimum Wetting

The fraction of flooding for packings is computed by dividing the superficial gas ve-
locity by the gas velocity at flood. The latter is found in an iterative process from the
pressure drop correlation (keeping the liquid to vapor ratio constant!) and the speci-
fied pressure drop of flood. If no flood pressure drop is specified it is computed from
the packing factor with the Kister and Gill correlation. If no pressure drop model is
selected, the Leva’s method is used, which also has the packing factor as only param-
eter.

The minimum operating conditions, similar to the weep point for tray operation, is set
to the minimum wetting rate predicted by Schmidt (1979). Schmidt calculates a liquid
falling number

CL = 1633.6(0.062428ρL )(1000σ)3 /(1000ηL )4 (15.53)

and the minimum wetting rate is (in U Sgal/min/f t2 ):

2/9
0.3182CL (1 − cos φ)2/3
QM W = p (15.54)
(1 − TL )ap /0.3048

319
where φ is the liquid contact angle and TL is the shear stress number:
TL = 0.9F F −2.8 (15.55)
with F F as the fraction of flood. The contact angle for metals is 10o (0.1745 rad).
To obtain contact angles for other packings the contact angle is assumed inverse pro-
portional to the critical surface tension, σc . For metals we have σc = 0.075, so the
contact angle is computed with
 
0.075
φ = 0.1745 (15.56)
σc
With this approach we obtaim higher minimum wetting rates for plastic packings than
for metal packings. The minimum operating or ‘weep’ fraction W F is then taken as
QM W
WF = (15.57)
uL
where uL is the superficial liquid flow.

15.5 Column Design

Tray layout parameters that specify a complete design (for the calculation of mass
transfer coefficients and pressure drops) are shown in Table 15.3. For packings only
the column diameter and bed height are design parameters, other parameters (such as
void fraction, nominal packing diameter, etc.) are fixed once the type of packing has
been selected. The packed bed height must be specified since it determines the desired
separation and the capacity.

A very important parameter in tray column design is the system factor (SF). It repre-
sents the uncertainty in design correlations with regard to phenomena that are currently
still not properly modeled, such as foaming.

Different design methods can be employed:

Fraction of flooding; this is the standard design method for trays, we have employed
a modified version of the method published by Barnicki and Davis (1989).
Pressure drop; this is the usual design method for packed columns, but is very useful
as well for tray design with pressure drop constraints.

other design methods can be thought of: ones that minimize pressure drop and cost,
or maximize flexibility and efficiency. ChemSep has a modular structure to allow
different design methods to be implemented.

320
 Table 15.3: Tray layout data

General (sieve) tray layout data:

Column diameter Active area
Number of flow passes Total hole area
Tray spacing Downcomer area
Liquid flow path length Weir length
Hole diameter Weir height
Hole pitch Deck thickness
Downcomer clearance
Additional data for bubble caps:
Cap diameter Slot area
Slot height Riser area
Skirt clearance Annual area
Additional data for valves:
Closed Loss K Open Loss K
Eddy Loss C Ratio Valve Legs
Valve Density Valve Thickness
Fraction Heavy Valves Heavy Valve Thickness

15.5.1 Tray Column Design: Fraction of flooding

The first task in this approach to tray design is to assign all layout parameters to con-
sistent values corresponding to the required capacity defined by the fraction of flood-
ing and current flowrates. These defaults function as starting points for subsequent
designs.

The initial free area ratio is taken to be 15% of the active area. The active area is de-
termined from a capacity factor calculation with internals specific methods (for sieve
and bubble-cap trays the default is Fair’s correlation by Ogboja and Kuye (19), and
the Glitsch method is used for valve trays). The tray spacing is initially set to the
default value (of 0.5m) and the downcomer area is calculated according to a method
in the Glitsch manual (limited by a minimum time residence check). From the com-
bined areas the column diameter is computed. The number of liquid passes on a tray
is initially set by the column diameter; under 5f t one pass, under 8f t two, 10f t three,
under 13f t four, else five passes. With the number of passes and the column diameter
the total weir length is computed. Once the weir length is determined the liquid weir
load is checked, if too high the number of passes is incremented and a new weir length
is evaluated until the weir load is below a specified maximum.

321
Initial weir height is taken as 2”, but limited to a maximum of 15 % of the tray spacing.
For notched or serrated weirs the notch depth is a third of the weir height. For serrated
weirs the angle of serration is 45o . Circular weirs have diameters 0.9 times the weir
length. The hole diameter is set to 3/16” for sieve trays and tray thickness 0.43 times
the hole diameter (or 1/10”). The hole pitch is computed from the free area ratio and
hole diameter according to a triangular pitch. The default downcomer clearance is 1.5”
but is limited by the maximum allowed downcomer velocity according to the Glitch
method, de-rated with the system factor. The clearance is set to be at least 0.5” lower
than the weir height to maintain a positive liquid seal but is limited to a minimum of
0.5”.

For bubble-cap trays the cap diameter is 3” for column diameters below 4.5 f t and
4” for above. The hole diameter can vary between 60% to 71% of the cap diameter,
and default taken as 70%. Default skirt clearance is 1” with minimum of 0.5” and
maximum of 1.5”. The slot height can vary in between 0.5” and 1.5”, where the
default is 1” for cap diameters below 3.5” and 1.25” for larger cap diameters. The
pitch can vary from 1.25” to half the flow path length (minimum number of rows is
two), with the default value set to 1.25”.

Valve trays are initialized to be Venturi orifice uncaged, carbon steel valves of 3mm
thick with 3 legs (see Kister, 1992, p312). The hole diameter is 1” for column smaller
than 4.5f t, otherwise 2”. No double weight valves are present.

The second task in the fraction of flooding method consists of finding the proper free
area ratio (β = Ah /Ab = hole area / active area) so that no weeping occurs. This
ratio can vary between a minimum of 5% (for stable operation) and a maximum of
20%. To test whether weeping occurs, we use the correlation of Lockett and Banik
(1984): Frhole > 2/3. The method requires all liquid heights to be evaluated at weep
rate conditions. This task is ignored for bubble-cap trays. The weep test is done at
weeping conditions, with a weep factor at 60 % (this can be changed). Calculation of
the liquid heights was described in Section 16.2. If weeping occurs at the lower bound
for the free area ratio, a flag is set for the final task to adapt the design.

The final task consists of evaluating all liquid heights at normal conditions and to do
a number of checks:

• vapor distribution (for bubble-cap trays),

• weeping (for sieve trays/valve trays),

• hydraulic flooding,

• excessive liquid entrainment,

322
 • froth height limit, and
• excessive pressure drop

If a check fails, the design is modified to correct the problem, according to the ad-
justments shown in Table 15.4 after which new areas are calculated from the capacity
correlations. Part of this task is also to keep the layout parameters within certain
bounds to maintain a proper tray design. Finally, the number of iterations for the
design method is checked against a maximum (default 30) to prevent a continuous
loop.

The adjustment factors f1 , f2 , and f3 are percentile in/decrements, normally set at 5, 2,

and 1 %. These factors – together with all the default, lower, and upper settings that are
used in the design routine – are stored in a “design file” (TDESIGN.DEF, described in
Section ??) that can be tailored to handle specific kinds of designs and columns. This
allows the selection of different methods for capacity and hydrodynamic calculations
as well. Also the fraction that the flows need to change before a re-design is issued can
be changed in this manner together with other design criteria. The design file must be
in the current directory for the nonequilibrium program to use it, otherwise the normal
defaults will be used. To obtain the best results, it would be best to have different
TDESIGN.DEF files for distillation and absorption. However, a compromise between
these operations was struck and ChemSep currently has one file to handle both.

15.5.2 Tray Column Design: Pressure drop

The same design routine for a tray design at a specific fraction of flood may also be
used to design a tray with a certain maximum pressure drop, using a default fraction of
flood of 75%. Note that this method of designing trays does not fix the pressure drop:
it only applies a maximum allowed pressure drop over the tray. If the tray design
results in a pressure drop larger than that specified the layout is adjusted according
to the steps for excessive pressure drop per Table 15.4. To obtain a layout which has
a lower pressure drop is to raise the bubbling area, in effect lowering the fraction of
flood. However, the weir height is also lowered and the hole diameter is increased
(albeit with a smaller factor).

15.6 Comparison with Experimental Distillation Data

Here, we illustrate the performance of the nonequilibrium model by simulating sev-

eral systems for which experimental data is available in the literature. We focus our

323
 Table 15.4: Tray design checks and adjustments

Problem Test Adjustments

Bubble cap vapor distribution hlg /hd > 0.5 p + f1 ,
hskirt + f2 ,
hslot + f3 ,
dh − f3
Weeping Frh /(2/3) < 1 − fa
f ree < 0.05
Ab < Abf : Ab = Abf
Wf + f1
else: Ab − f 1
dh − f3
hw − f3
tv + f2 (vt)
Hydrodynamic (downcomer) flooding) Ts < hdb /F F Ts + f1
Ad + f 1
hw + f2
hc + f3
Excessive liquid entrainment Ab + f 1
Ts + f1
dh − f2
hw − f3
Froth height limit hf > 0.75Ts Ab + f 1
Ts + f2
hw − f3
Excessive pressure drop gρhwt > ∆pmax Ab + f 1
hw − f1
dh + f2
p + f1 (bc)
hskirt + f2 (bc)
hslot + f3 (bc)
Excessive vapor entrainment Ad + f 1

attention on ternary systems; only in mixtures with more than two species can the
molecular interaction between them influence the mass transfer resulting in different
individual component Murphree efficiencies.

The complete process of comparing experimental data with the current version of
ChemSep has been automated and provides a test for the correct behavior of the sim-

324
ulator. We hope to build up an extensive database of distillation experiments and
welcome additional data. The experimental literature data is entered in a database file
that consists of keywords and the data. Then we carry out simulations and store each
simulation in a sep-file. With the use of some utility programs we convert the simula-
tion information and compare it to the experimental measured values. We can create
parity plots and calculate average and maximum errors.

15.6.1 Ternary Distillation Experiments

The experiments discussed below were carried out at total reflux in columns with
bubble-cap trays.

Vogelpohl (1979) has reported some results for the distillation of two non-ideal sys-
tems: acetone, methanol, and water as well as methanol, isopropanol, and water. The
experiments were done in a column with 38 bubble-cap trays of 0.3 m in diameter.
Due to the ease of separation, only up to 13 trays were active for the experimental
runs. The experimental data clearly shows that the component Murphree efficien-
cies are unequal; indeed, in the acetone-methanol-water system the composition of
methanol passes through a maximum in the column and the efficiency for this compo-
nent becomes unbounded. Vogelpohl shows that the assumption of equal component
efficiencies gives rise to large differences between the predicted and measured com-
position profiles.

The simulations were done using the total reflux mode of the nonequilibrium model.
For total reflux specifications one must use a column configuration with a condenser
and reboiler plus one feed. Either the distillate flow is set to zero and a reboiled vapor
flow is specified, or, the bottoms flow is set to zero and a reflux flowrate is specified.
Only these specifications of the column operation will trigger the total reflux mode
of the column simulator. In this mode the feed specifications are employed as the
specification of the vapor or liquid compositions on a specific stage. The compositions
(and stage) that are to be fixed are set by the feed compositions and stage. Thus, if
we know the compositions of the vapor leaving the reboiler we specify a feed to the
reboiler and set the feed component flows equal to the known mole fractions. To
indicate that these are vapor mole fractions I set the vapor fraction of the feed as 1.
On the other hand, if the liquid reboiler compositions are known we would specify the
feed vapor fraction as 0.

For Vogelpohl’s data the compositions of the vapor leaving the (total) reboiler were
specified to match the measured values. The reboiler vapor flowrate was specified at
0.65 kmol/h resulting in an effective F-factor around 0.1. The calculated fraction of
flood was around 20 %. Table 15.5 shows the UNIQUAC interaction parameters used

325
for the acetone, methanol, and water column. These interaction parameters were fitted
to binary VLE data. The bubble-cap trays were specified with 0.3 m tray diameter,
tray spacing as 0.2 m, bubbling area of 0.06008 m2 , a weir height of 0.03 m, and a
flow path width of 0.24 m. The other layout parameters are assigned/computed by the
simulator. The AIChE MTC model was used together with the vapor plug flow and
liquid mixed flow models.

Table 15.5: UNIQUAC interaction parameters (cal/mol) for the Acetone (1) -
Methanol (2) - Water (3) system

Components i-j Aij Aji

acetone - methanol 403.8524 -84.2364
acetone - water 698.7989 -110.382
methanol - water -337.129 549.2958

R3(1) Rn run number n (1) acetone

(2) methanol
0.8
R2(1)
R1(2)

0.6
Mole fraction

R2(2)

0.4 R1(1)

0.2 R3(2)

0
0 2 4 6 8 10
Stage Number

Figure 15.1: Experimental compositions (points) and the predicted composition pro-
files (lines) for the acetone-methanol-water system.

"Acetone"
"Methanol"
0.8 "Water"
Simulated compositions

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
Experimental compositions

Figure 15.2: Parity plot for the acetone-methanol-water system.

For the methanol, isopropanol, and water column UNIQUAC interaction parameters
were taken from DECHEMA (p. 575) that were fitted to ternary VLE data, see Ta-

326
 1

Rn run number n (1) methanol

(2) isopropanol
0.8

0.6

Mole fraction
R3(2)
0.4 R1(2)

0.2 R3(1) R1(1)

0
0 2 4 6 8 10 12 14
Stage Number

Figure 15.3: Experimental compositions (points) and the predicted composition pro-
files (lines) for the methanol-isopropanol-water system.

"Methanol"
Isopropanol
0.8 "Water"
Simulated compositions

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
Experimental compositions

Figure 15.4: Parity plot for the methanol-isopropanol-water system.

ble 15.6. The same bubble-cap tray layout, column specifications, and model selec-
tions were used as for the acetone-methanol-water column.

Figure 15.1 shows the measured compositions and the simulated mole fraction profiles
for the acetone-methanol-water system. Figure 15.2 shows that the nonequilibrium
model does an excellant job in predicting the mole fractions for this very nonideal
system. Figure 15.3 and Figure 15.4 show that the methanol-isopropanol-water sys-
tem is predicted less well. We suspect this may be due to the interaction parameters
of the activity coefficients. These simulations are extremely sensitive to these param-
eters, especially as isopropanol is going through a maximum in concentration. The
average and maximum discrepancies in the predicted and experimentally measured
compositions are shown in Table 15.8.

Ternary distillation experiments using acetone, methanol, and ethanol were performed
by Free and Hutchison (1960) in a column with 7 bubble-cap trays of 0.1016 m in di-
ameter. They also find that equal Murphree efficiencies cannot explain the bevavior
of this system. Twelve runs were conducted covering different regions of the ternary
composition triangle. The column they used had a tray diameter of 0.1016 m, bub-

327
Table 15.6: UNIQUAC interaction parameters (cal/mol) for the Methanol (1) - Iso-
propanol (2) - Water (3) system

Components i-j Aij Aji

methanol - isopropanol 754.216 -457.107
methanol - water 670.563 -442.713
isopropanol - water 355.536 28.874

bling area of 0.00689 m2 , 0.015 m weir height

p and a flow path length of 0.0813 m.
The F-factor was assumed to be around 0.9 kg/m/s and a vapor boilup of 0.65
kmol/h was specified, resulting in fraction of flood around 35 %. Table 15.7 shows
the UNIQUAC Q prime interaction parameters were used (Prausnitz et al., 1980). The
UNIQUAC Q prime activity coefficient model gives excellent results for these exper-
iments. Figure 15.5 shows the data and the predicted column profiles for some of the
experimental runs. The agreement is excellent, as can also be seen from the parity
plot, Figure 15.6.
1

Rn run number n (1) acetone

(2) methanol
0.8 R11(1)
R4(2)
R9(1)
0.6
Mole fraction

R4(1)
0.4

R9(2)
0.2 R11(2)

0
0 1 2 3 4 5 6 7 8
Stage Number

Figure 15.5: Comparison of experimental compositions (points) and the predicted

composition profiles (lines) for the acetone-methanol-ethanol system.

Table 15.7: UNIQUAC interaction parameters (cal/mol) for the Acetone (1) -
Methanol (2) - Ethanol (3) system

Components i-j Aij Aji

acetone - methanol 359.10 -96.90
acetone - ethanol 404.49 -131.25
methanol - ethanol 660.19 -292.39

All these simulations were done by specifying the bottom compositions in the column,
just as was done by Krishnamurthy and Taylor (1985). The errors in Table 15.8 would

328
 1

"Acetone"
"Methanol"
0.8 "Ethanol"

Simulated compositions
0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
Experimental compositions

Figure 15.6: Comparison of experimental and predicted mole fractions for the
acetone-methanol-ethanol system.

be reduced if we would specify the compositions in the middle of the column, as the
errors in the predicted mole fractions often accumulate from stage to stage.

Table 15.8: Summary of the average and maximum discrepancies between model
prediction and experimental measurement

System No. of No. of Average Maximum

Runs Samples Error Error
acetone - methanol - water 3 22 0.015 0.070
methanol - isopropanol - water 3 27 0.024 0.087
acetone - methanol - ethanol 12 67 0.010 0.051

15.6.2 Significance of Multicomponent Interaction Effects

Figure 15.7 is a comparison for the acetone-methanol-water experiments of the nonequi-

librium model with an ‘equal diffusivity’ model. This model uses a single average
value of the diffusion coefficients for each component. It yields equal component stage
efficiencies and, therefore, corresponds to the conventional equilibrium stage model
with efficiencies. The comparison clearly shows that the predicted top compositions of
the ‘equal diffusivity’ model are qualitatively different from the experimental values.

This can be explained by inspecting the back-calculated Murphree component effi-

ciencies of the nonequilibrium model, see Figure 15.8. We see that the efficiencies of
water and acetone differ by about 10%! This explains the observed 10% difference
in top compositions. The methanol efficiencies show an even more weird behavior:
they become unbounded between plate 6 and 7. When we look at the compositions,
we see that the reason for this is that methanol goes through a maximum around stage

329
6. Whenever a component goes through an extreme in composition (the driving force
becomes zero) and there is still mass transfer occurring (however little) than the effi-
ciencies are unbounded. If we had more stages in the column, we would observe that
the methanol efficiency below this maximum stays at the higher value of the Murphree
efficiency of water. Apparently, the direction of the mass transfer is important!
1

Rn run number n (1) acetone

(2) methanol
0.8

R1(2)

0.6

Mole fraction
0.4 R1(1)

0.2

0
0 2 4 6 8 10
Stage Number

Figure 15.7: Experimental compositions (points) and the predicted composition pro-
files for the acetone-methanol-water system using the nonequilibrium model (solid
lines) and the ‘equal diffusivity’ model (broken lines).

water
0.8
Murphree Efficiency

0.6 acetone

0.4 methanol

0.2

0
1 2 3 4 5 6 7 8
Stage Number

Figure 15.8: Back-calculated Murphree component efficiencies for the acetone-

methanol-water system from the nonequilibrium simulation.

15.6.3 Binary Distillation Experiments: Mass Transfer Coefficients

and Flow Models

To illustrate the behavior of the various tray MTC models as well as the different
flow models we include here some comparisons with experimental data of the FRI by
Yanagi and Sakata (1979, 1981). Two systems were used in these tests: the cyclo-
hexane - n-heptane system at pressures of 28, 34, and 165 kP a, and the i-butane -
n-butane system at pressures of 1138, 2056, and 2756 kP a. The experiments were
carried out in sieve tray columns operated at total reflux. Nonequilibrium simulations

330
 Table 15.9: Sieve Plate Dimensions of FRI Column

Column diameter (m) 1.2

Tray spacing (m) 0.61
Sieve plate material 316SS
Plate thickness (mm) 1.5
Hole diameter (mm) 12.7
Hole pitch (mm) 30.2
Weir length (m) 0.94
Weir height (mm) 25.4,50.8
Downcomer clearance (mm) 22,38
Effective bubbling area (m2 ) 0.859
Hole area (m2 ) 0.118

were done for a column with the same tray design (the design parameters are sum-
marized in Table 15.9) at total reflux (Kooijman, 1995). The Murphree efficiencies
were calculated for each component on each tray from the results of a simulation and
averaged (over those trays not adjacent to condenser/reboiler). Simulations were done
using different combinations of flow models:

• Mixed vapor - Mixed liquid

• Plug flow vapor - Mixed liquid
• Plug flow vapor - Plug flow liquid

Simulations were carried out with flows that go from 20% to 100% of flooding. Fig-
ure 15.9 shows some of these results with the Chan and Fair (1984) mass transfer
coefficients correlations. The experimental efficiencies show a decline at low and
high fractions of flooding, probably due to weeping and liquid entrainment. The Chan
and Fair model includes a quadratic dependence of NV on the fraction of flooding
in order to account for the decrease in mass transfer at both low and high fractions of
flooding. For this reason the Chan and Fair method usually describes the mass transfer
(and hence, the efficiencies) better than the other mass transfer coefficient models.

Note that the Mixed-Mixed flow model underpredicts the efficiencies, as is true for
the Plug flow vapor - Mixed flow liquid model. The Plug-Plug flow model fits the
experimental efficiencies quite well. This should not be too surprising as this data
was actually used in the development of the Chan and Fair correlations. We are,
however, using the correlations in a nonequilibrium model rather than in an efficiency
calculation.

331
 100

80

Murphree Efficiency (%)

60

40

20

0
0 20 40 60 80 100
Fraction of flooding (%)

Figure 15.9: Murphree efficiencies for different flow models for a (8% hole area) sieve
tray column with the cyclohexane - n-heptane system operating at 165 kP a. Mixed-
Mixed flow (thick dotted line), Plug flow vapor - Mixed flow Liquid (thick dashed
line), and Plug-Plug flow (solid line). Mass transfer coefficients from the Chan and
Fair correlation..

Four different methods for estimating the binary mass transfer coefficients for sieve
trays were tested: AIChE (1958), Chan and Fair (1984), Zuiderweg (1982), and Chen
and Chuang (1994). Figure 15.10 shows the results for the i-butane/n-butane tests
using the plug-plug flow model. In general, the AIChE and the Chan-Fair correlations
behave in similarly, except for the strong dependence on the fraction of flooding of
the Chan-Fair model. This is not surprising since both use the same expression for the
liquid mass transfer coefficients. The Zuiderweg and Chen-Chuang models predict
higher efficiencies as they have higher values for the liquid number of transfer units
than the first two correlations. For this test they perform well; in many other tests
these two models tend to overpredict the Murphree efficiencies.
120

100
Murphree Efficiency (%)

80

60

40

20

0
0 20 40 60 80 100
Fraction of flooding (%)

Figure 15.10: Murphree efficiencies for different Mass Transfer Coefficient models
for a (14% hole area) sieve tray column with i-butane - n-butane system operating at
1138 kP a. Chan and Fair (thick dotted line), AIChE (thick dashed line), Zuiderweg
(dashed line), and Chen and Chuang (solid line)..

332
The Chan-Fair correlation tends to describes the overall behavior of the Murphree ef-
ficiencies better than the other methods considered. However, it’s formulation causes
it always to have a maximum efficiency at 60% fraction of flood. This limits the
model to sieve trays only and to the range of fractions of flooding where the quadratic
term is positive (0-1.2). Presumably, the fall-off in tray performance at low and high
fractions of flooding is due to increases in weeping or entrainment at these extreme
flows. It is not clear to us that mass transfer coefficient correlations should account for
these effects. Rather, we suggest that these effects should be separated. We have also
encountered situations where the Chan and Fair correlations provide negative mass
transfer coefficients because the flows are outside it’s range. Not only are negative
mass transfer coefficients physically meaningless, they may prevent the program that
implements our nonequilibrium model from converging to a solution! Despite these
problems with the Chan and Fair method we think its limitations are less serious (from
our perspective) than are the limitations of other methods and, for now, it is our method
of choice.

Symbol List

Latin Symbols

ad interfacial area density (m2 /m3 )

a area (m2 )
Ah hole area (m2 )
Ab , Abub bubbling area (m2 )
Ad downcomer area (m2 )
B packing base (m)
c number of components,
molar concentration (kmol/m3 )
Ca Capillary number
dh hole diameter (m)
deq equivalent diameter
D binary diffusivity coefficient (m2 /s)
Dc column diameter (m)
De eddy dispersion coefficient (m2 /s)
f1 , f2 , f3 design adjustment factors
Fp packing factor (1/m)
√
Fs F factor Fs = Uv ρV (kg 0.5 /m0.5 /s)
FF fraction of flooding p
FP flow parameter F P = ML /MV ρVt /ρL t

333
Fr Froude number
g gravitational constant, 9.81 (m/s2 )
hc clearance height under downcomer (m)
hcl clear liquid height (m)
hd dry tray pressure drop height (m)
hdb downcomer backup liquid height (m)
hf froth height (m)
hi liquid height at tray inlet (m)
hlg liquid gradient pressure drop height (m)
hl , hL liquid pressure drop height (m)
how height of liquid over weir (m)
hr residual pressure drop liquid height (m)
hwt wet tray pressure drop liquid height (m)
hw weir height (m)
hudc liquid height pressure loss under downcomer (m)
k binary mass transfer coefficient (m/s)
Le Lewis number (Le = Sc/Pr)
Mw molecular weight (kg/kmol)
N number of transfer units, NTU
P perimeter (m)
p hole pitch (m),
pressure (P a)
∆p pressure drop (P a)
∆Pmax maximum design pressure drop (P a/tray or P a/m)
Pr Prandtl number
Q volumetric flow (m3 /s)
Re Reynolds number
S packing side (m)
Sc Schmidt number
SF system derating factor
t residence time (s)
tv valve thickness (m)
T temperature (K)
Ts tray spacing (m)
u, U velocity (m/s)
V vapor flow rate (kmol/s)
We Weber number
Wl weir length (m)
x liquid mole fraction
y vapor mole fraction
z mole fraction

334
Z tray flow path length (m)

Greek Symbols

α fraction liquid in froth

β fractional free area β = Ah /Ab ,
 void fraction
Γ liquid flow per perimeter
φ fractional entrainment
ρ density (kg/m3 )
σ surface tension (N/m)
η viscosity (P a s)
λ thermal conductivity (W/m/K)

Superscripts

I interface
L liquid
P phase P
V vapor

Subscripts

b, bub bubbling
c critical, contact
ef f effective
f l, f lood at flooding conditions
i component i
j stage j,
component j
h hole
MW minimum wetting
p packing
spec specified

335
t total

Abbreviations

bc bubble-caps

References

Bubble Tray Design Manual: Prediction of Fractionation Efficiency, AIChE, New

York (1958).

P.E. Barker, M.F. Self, “The evaluation of Liquid Mixing Effects on a Sieve Plate using
Unsteady and Steady-State Tracer Techniques”, Chem. Eng. Sci., Vol. 17, (1962) p.
541.

S.D. Barnicki, J.F. Davis, “Designing Sieve-Tray Columns, Part 1: Tray Design”,
Chem. Engng., Vol. 96, No. 10 (1989) pp. 140–146.

S.D. Barnicki, J.F. Davis, “Designing Sieve-Tray Columns, Part 2: Column Design
and Verification”, Chem. Engng., November (1989) pp. 202–212.

D.L. Bennett, R. Agrawal, P.J. Cook, “New Pressure Drop Correlation fo Sieve Tray
Distillation Columns”, AIChE J., Vol. 29 (1983) pp. 434–442.

R. Billet, M. Schultes, ”Advantage in correlating packed column perfomance”, IChemE.

Symp. Ser. No. 128 (1992) p. B129.

J.L Bravo, J.R. Fair, ”Generalized Correlation for Mass Transfer in Packed Distillation
Columns”, Ind. Eng. Chem. Process Des. Dev., Vol. 21 (1982) pp. 162–170.

J.L. Bravo, J.A. Rocha, J.R. Fair, ”Mass Transfer in Gauze Packings”, Hydrocarbon
Processing, January (1985) pp. 91–95.

J.L. Bravo, J.A. Rocha, J.R. Fair, ”Pressure Drop in Structured Packings”, Hydrocar-
bon Processing, March (1986) pp. 45–48.

J.L. Bravo, J.A. Rocha, J.R. Fair, ”A Comprehensive Model for the Performance of
Columns Containing Structured Packings”, I. ChemE. Symp. Ser., No. 128 (1992) pp.

336
A439–A457.

H. Chan, J.R. Fair, “Prediction of Point Efficiencies on Sieve Trays. 1. Binary Sys-
tems”, Ind. Eng. Chem. Process Des. Dev., Vol. bf 23 (1984) pp. 814–819.

G.X. Chen and K.T. Chuang, “Prediction of Efficiencies for Sieve Trays in Distilla-
tion”, Ind. Eng. Chem. Res., Vol. 32 (1993) pp. 701–708.

I.J. Harris, “ Optimum Design of Sieve Trays”, Brit. Chem. Engng., Vol. 10, No. 6
(1965) p. 377.

G.A. Hughmark, “Models for Vapour Phase and Liquid Phase Mass Transfer on Dis-
tillation Trays”, AIChE J., Vol. 17, No. 6 (1971) p. 1295.

F. Kastanek, “Efficiencies of Different Types of Distillation Plate”, Coll. Czech.

Chem. Comm., Vol. 35 (1970) p. 1170.

H.Z. Kister, Distillation Design, McGraw-Hill, New York (1992).

G.F. Klein, Chem. Engng., May 3 (1982) p. 81.

M. Leva, Tower Packings and Packed Tower Design, The U.S. Stoneware Company
(1951).

M. Leva, “Reconsider Packed-Tower Pressure-Drop correlations”, Chem. Eng. Prog.

January (1992) p. 65.

M.J. Lockett, Distillation Tray Fundamentals, Cambridge University Press (1986).

M.J. Lockett, S. Banik, “Weeping from Sieve Trays”, Ind. Eng. Chem. Process Des.
Dev., Vol. bf 25 (1986) p. 561.

E.E. Ludwig, Applied Process Design for Chemical and Petrochemical Plants, Vol. 2,
2nd Ed., Gulf Pub. Co., Houston, TX, (1979).

O. Ogboja, A. Kuye, “A Procedure for the Design and Optimisation of Sieve Trays”,
Trans. I. Chem. E., Vol. 68, Part A (1990) pp. 445-452.

K. Onda, H. Takeuchi, Y. Okumoto, ”Mass Transfer Coefficients Between Gas and

Liquid Phases in Packed Columns”, J. Chem. Eng. Jap., Vol. 1, No. 1 (1968) p. 56.

M. Prado, J.R. Fair, “Fundamental Model for the Prediction of Sieve Tray Efficiency”,
Ind. Eng. Chem. Res., Vol. 29, No. 6 (1990) pp. 1031–1042.

M. Prado, PhD Thesis, Austin, TX (1986).

337
R.H. Perry and D. Green,Perry’s Chemical Engineering Handbook, 6th edition, sec-
tion 18, Liquid-Gas System (1984) pp. 18-8 – 18-12.

B.D. Smith, Design of Equilibrium Staged Processes, McGraw-Hill, New York (1963).

J. Stichlmair, A. Mersmann, “Dimensioning Plate Columns for Absorption and Recti-

fication”, Chem. Ing. Tech., Vol. 45, No. 5 (1978) p. 242.

J. Stichlmair, J.L. Bravo, J.R. Fair, Gas. Sep. Purif., Vol. 3, (1989) p. 19.

R. Taylor, R. Krishna, Multicomponent Mass Transfer, Wiley NY (1993)

P.C. Wankat, Separations in Chemical Engineering - Equilibrium Staged Separations,

Elsevier (1988) pp. 420–428.

F.J. Zuiderweg, “Sieve Trays - A View of the State of the Art”, Chem. Eng. Sci., Vol.
37 (1982) pp. 1441–1461.

338
Chapter 16

Design Models for Packed

Columns

This chapter serves as a guide to the correlations in ChemSep for estimating mass
transfer coefficients and the pressure drop in packed distillation (and absorption) columns.
The methods used in Design Mode to determine column design parameters also are
discussed in detail. The chapter ends with the comparison of experimental distillation
data and nonequilibrium model simulations.

16.1 Mass Transfer Coefficient Correlations

Table 16.1 provides a summary of the available correlations for trays and packings;
the various correlations are discussed in more detail below. Recommended models
are shown in boldface.

16.1.1 Random Packings

OTO-68 Onda et al. (1968) [parameters ap , dp , σc ] developed correlations of mass

transfer coefficients for gas absorption, desorption, and vaporization in random
packings. The vapor phase mass transfer coefficient is obtained from

k V = ARe0.7 0.333
V ScV (ap DV )(ap dp )−2 (16.1)

339
 Table 16.1: Available mass transfer coefficient correlations per internals type

Dumped Structured
packing packing
Onda 68 Bravo 85
Bravo 82 Bravo 92
Billet 92 Billet 92
... ...

where A = 2 if dp < 0.012 and A = 5.23 otherwise. Vapor and liquid veloci-
ties are calculated from
uV = V MwV /ρV At (16.2)
uL = LMwL /ρL At (16.3)
and Reynolds and Schmidt numbers defined by:
ρV uV
ReV = (16.4)
(η V ap )
ρL uL
ReL = (16.5)
(η L ap )
ηV
ScV = (16.6)
(ρVt DV )
ηL
ScL = (16.7)
(ρL L
t D )
The liquid phase mass transfer coefficient is
k L = 0.0051(ReIL )2/3 Sc−0.5
L (ap dp )0.4 (η L g/ρL )1/3 (16.8)
where ReIL is the liquid Reynolds number based on the interfacial area density
ρL uL
ReIL = (16.9)
(η L ad )
The interfacial area density, ad (m2 /m3 ), is computed from
−0.05
ad = ap 1 − exp −1.45(σc /σ)0.75 Re0.1 0.2



L FrL WeL (16.10)
where
ap u2L
FrL = (16.11)
g
ρL u2L
WeL = (16.12)
ap σ

340
BF-82 Bravo and Fair (1982) [parameters ap , dp , σc ] used the correlations of Onda
et al. for the estimation of mass transfer coefficients for distillation in random
packings but proposed an alternative relation for the interfacial area density:
√
ad = 19.78(CaL ReV )0.392 σH −0.4 ap (16.13)

where H is the height of the packed section and CaL is the capillary number

CaL = uL η L /ρL σ (16.14)

Since the interfacial area density is used in the calculation for the liquid phase
Reynolds number the Bravo and Fair method will predict different mass transfer
coefficients for the liquid phase.
BS-92 Billet and Schultes (1992) [parameters ap , , Cf l , Ch , Cp , Cv , Cl ] describe
an advanced empirical/theoretical model which is dependent on the pressure
drop/holdup calculation (Ch , Cp , Cf l ). The correlation can be used for both
random and structured packings. Vapor and liquid phase coefficients are ad-
justed by parameters Cv and Cl , bringing the total number of parameters to
five. There are trends in the parameters that can be observed from the tabulated
data. Unfortunately, no such generalization was done by Billet, making use of
the model dependent on the availability of the parameters or experimental data.
The mass transfer coefficients are computed from
 1/6 s L  1/3
gρ l D uL
k L = Cl (16.15)
ηl dh ap
 r
1 a V
k V = Cv √ D (ReV )3/4 (ScV )1/3 (16.16)
 − ht dh
with Reynolds and Schmidt numbers calculated as in the method of Onda et al..
The hydraulic diameter dh is

dh = 4/ap (16.17)

and the liquid holdup fraction, ht , is calculated as described below under the
pressure drop section. The interfacial area density is given by:

ad = ap (1.5/ ap dh )(uL dh ρL /η L )−0.2 (u2L ρL dh /σ)0.75 (u2L /gdh )−0.45

p

(16.18)

16.1.2 Structured Packings

BRF-85 Bravo et al. (1985) [parameters ap , , B, hc , S, Deq , θ] published correla-
tions for structured packings. This method is based on the assumption that the

341
 surface is completely wetted and that the interfacial area density is equal to the
specific packing surface: ad = ap . The Sherwood number for the vapor phase
is
ShV = 0.0338ReV0.8 Sc0.333
V (16.19)
and is defined by
k V deq
ShV = (16.20)
DV
The equivalent diameter of a channel is given by

deq = Bhc [1/(B + 2S) + 1/2S] (16.21)

where B is the base of the triangle (channel base), S is the corrugation spacing
(channel side), and hc is the height of the triangle (crimp height). The vapor
phase Reynolds number is defined by

deq ρV (uV,ef f + uL,ef f )

ReV = (16.22)
ηV
The effective velocity of vapor through the channel, uV e , is

uV,ef f = uV /( sin θ) (16.23)

(uV is the superficial vapor velocity,  the void fraction, and θ the angle of the
channel with respect to the horizontal). The effective velocity of the liquid is
1/3
(ρL )2 g

3Γ
uL,ef f = L (16.24)
2ρ 3η L Γ

where Γ is the liquid flow rate per unit of perimeter

Γ = ρL uL /P (16.25)

where P is the perimeter per unit cross-sectional area, computed from

P = (4S + B)/Bhc (16.26)

The penetration model is used to predict the liquid phase mass transfer coeffi-
cients with the exposure time assumed to be the time required for the liquid to
flow between corrugations (a distance equal to the channel side):

tL = S/uL,ef f (16.27)
s
DL
kL = 2 (16.28)
πtL

342
BRF-92 Bravo et al. (1992) [parameters ap , , S, θ, Fse , K2 , Ce , dP dzf lood ] de-
veloped a theoretical model for modern structured packings. Four parameters
can be supplied. However, the authors advise using a fixed value for the surface
renewal correction (Ce ), normally 0.9. They provide a relation for parameter
K2 as well:
K2 = 0.614 + 71.35S (16.29)
The mass transfer calculations depend on the pressure drop and holdup calcu-
lation. The effective area can be adjusted with the surface enhancement factor
Fse , and the liquid resistance with a correction on the surface renewal following
the penetration model (parameter Ce ). Effective velocities are computed with
uL
uL,ef f = (16.30)
ht sin θ
uV
uG,ef f = (16.31)
(1 − ht ) sin θ
where ht is the fractional liquid holdup (see below at the section on pressure
drop calculation). Reynolds numbers and liquid mass transfer coefficient is now
calculated as in Bravo et al. (1985) but with

tL = Ce S/uL,ef f (16.32)

The vapor phase mass transfer coefficient is obtained from

DV 1/3
k V = 0.054( )Re0.8
V ScV (16.33)
S
where the equivalent diameter is replaced with the channel side S and a different
coefficient is used. The assumption of a completely wetted packing is dropped.
Instead, the interfacial area density is given by

ad = Ft Fse ap (16.34)
0.15 0
29.12(WeL FrL ) S .359
Ft = (16.35)
Re0.2
L 
0.6 (sin θ)0.3 (1 − 0.93 cos γ)

where cos γ is equal to 0.9 for σ < 0.0453, otherwise it is computed from

cos γ = 5.211 10−16.835σ (16.36)

Note that a switch point different from that used by Bravo et al. (1992) is
employed to guarantee continuity in cos γ.

BS-92 Billet and Schultes (1992) [parameters ap , , Cf l , Ch , Cp , Cv , Cl ] developed

a model for both random and structured packings, see the section on random
packings above.

343
 Table 16.2: Pressure drop correlations per internals type

Dumped Structured
packing packing
Fixed Fixed
Ludwig 79 Billet 92
Leva 92 Bravo 86
Billet 92 Stichlmair 89
Stichlmair 89 Bravo 92
... ...

16.2 Pressure Drop Models

There are as many ways to compute tray pressure drops as there to estimate mass
transfer coefficients. For packings there is a move away from the use of generalized
pressure drop charts (GPDC) to more theoretically based correlations. We have cho-
sen to employ the most recently published models. For packings there are at least 7
methods available (see Table 16.2). For packings operating above the loading point
(F F > 0.7) we advise the use of models that take the correction for the liquid holdup
into account, such as the SBF-89. However, disadvantage is that these models can
have complex (imaginary) solutions, especially at high fractions of flood. This can
cause non-cnvergence! The Lev-92 model can be an alternative for it includes a de-
pendence on the liquid flow rate to simulate the increased pressure drop at loading
conditions. When the pressure drop is specified as fixed, it is assumed zero.

16.2.1 Random packing Pressure Drop Correlations

For packings the vapor and liquid mass flow per cross sectional area (kg/m2 s) and
velocities (m/s) are:

La = LM L /At (16.37)
Va = V M V /At (16.38)
L
uL = La /ρ (16.39)
V
uV = Va /ρ (16.40)

Lud-79 Ludwig (1979) [parameters A, B] supplied a simple empirical equation for

344
 the pressure drop requiring two fitted parameters (see Wankat, 1988, 420–428):
∆p (0.2048Va )2 B(0.06243La )
= 3.281 242A 10 (16.41)
∆z (0.06243ρV )
where 3.281 242, 0.2048, and 0.06243 are conversion factors so that we can use
A and B parameters from Wankat. Its accuracy is limited since the influence
of physical properties such as viscosity or surface tension on A and B are not
included. Even more, the fitted parameters can be flow regime dependent. The
loading regime is not well described with the simple exponential term.
Lev-92 Leva (1992) [parameter Fp ] devised a modified version of the Generalized
Pressure Drop Correlation (GPDC) presented long ago by Leva (1951). The
GPDC has been the standard design method for decades. Some modifications
that were actually simplifications made the GPDC lose its popularity. The func-
tion worked back from the GPDC and limiting (La = 0) behavior is (in SI
units):
∆p 100.035La φ
= 22.3Fp (η L )0.2 φVa2 (16.42)
∆z gρV
with φ = ρwater /ρl = 1000/ρl . This expression is similar to the Ludwig
(1979) equation with corrections for the influence of the liquid density and vis-
cosity. The only parameter is the packing factor Fp that can be obtained from
dry pressure drop experiments (see Leva, 1992) or estimated from the specific
packing area over the void fraction cubed. Again, the loading regime is not well
described with the simple exponential term. This model is the default method
for random packings if no model is specified, since it requires only the packing
factor.
SBF-89 Stichlmair et al. (1989) [parameters ap , , C1 , C2 , C3 ] published a semi-
empirical method from an analogy of the friction of a bed of particles and the
pressure drop. It contains a correction for the actual void fraction corrected for
the holdup, that is dependent on the pressure drop. It is, therefore, an iterative
method. It is suitable for both random and structured packings, but there are
few published parameters for structured packings. The pressure drop is
∆p
= 0.75f0 (1 − p )ρV ∗ UV2 /(dp 4.65
p ) (16.43)
∆z
where the void fraction of the irrigated bed, equivalent packing diameter, Reynolds
number, and friction factor for a single particle are:
p =  − ht (16.44)
dp = 6(1 − p )ap (16.45)
V V
ReV = uV dp ρ /η (16.46)
p
f0 = C1 /ReV + C2 / ReV + C3 (16.47)

345
 The iterations are started by assuming a dry bed for which p =  and the holdup
fraction is computed with the liquid Froude number:

FrL = u2L ap /g4.65 (16.48)

1/3
ht = 0.555FrL (16.49)

The liquid holdup is limited to 0.5 in order to handle flooding.

BS-92 Billet and Schultes (1992) and Billet’s monograph (1979) [parameters a, ,
Cf l , Ch , Cp ] include a model with a comprehensive list of packing data and
fitted parameters. The method corrects for the holdup change in the loading
regime but employs an empirical exponential term, and is not iterative.
Packing dimension, hydraulic diameter and F-factor are

dp = 6(1 − )/ap (16.50)

dh = 4/ap (16.51)
p
Fs = uV ρV (16.52)

Liquid Reynolds and Froude number are

ReL = uL ρL /η L ap (16.53)
FrL = u2L ap /g (16.54)

If ReL < 5 then

0.15 0.1
q = Ch ReL FrL (16.55)
else
q = 0.85Ch Re0.25
L FrL
0.1
(16.56)

!1/3
12η L a2p uL
hl,1 = (16.57)
ρL g
hl,2 = hl,1 q 2/3 (16.58)
 L
0.05
η ρw
hl,f l = 0.3741 (16.59)
η w ρL

s 0.2
ρL ηL
 
uL
f l = (16.60)
uV ρV ηV
f l = g/(Cf2l f−0.39
l ) (16.61)
q q q √
uv,f l = 2g/f l ( − hl,f l )1.5 hl,f l /ap ρL /ρV /  (16.62)

346
 if uV > uV,f l then ht = hl,f l else
ht = hl,2 + (hl,f l − hl,2 )(uV /uV,f l )13 (16.63)
The pressure drop is then
K1 = 1 + (2/3)(1/1 − )(dp /Dc ) (16.64)
V V
ReV = uV dp ρ /(1 − )η K1 (16.65)
φl1 = Cp (64/ReV + 1.8/Re0.08
V )
exp(ReL /200)(ht /hl,1 )0.3 (16.66)
∆p
= φl1 (ap /( − ht )3 )(Fs2 /2)K1 (16.67)
∆z

16.2.2 Structured packing Pressure Drop Correlations

BRF-86 Bravo et al. [parameters , S, sin(θ), C3 ] compute the pressure drop from
an empirical correlation with one fitted parameter, called C3 . This model is
unsuitable for pressure drop correlations in the loading regime (F F > 0.7).
The pressure drop per height of packing is:
∆p 1
= (0.171 + 92.7/ReV )(ρV u2V,ef f /deq )( √ )5 (16.68)
∆z (1 − C3 Fr)
where
uV,ef f = uV /( sin θ) (16.69)
V
deq ρ uV,ef f
ReV = (16.70)
ηV
2
FrL = uL /deq g (16.71)

SBF-92 Stichlmair et al. (1989) [parameters a, , C1 , C2 , C3 ] published a semi-

empirical method, see the section on pressure drop of random packed columns
above.
BRF-92 Bravo et al. (1992) [parameters ap , , S, θ, K2 , dP dzf lood ] developed a
theoretical model developed for modern structured packings. Two parameters
need to be supplied for pressure drop calculations, however, the K2 parameter
was fitted by the authors. The pressure of flooding (dP dzf lood ) can be easily
obtained from data or via Kister’s correlation and the packing factor. The model
includes an iterative method with a dependence of the liquid holdup on the
pressure drop (and vice versa). The Weber, Froude, and Reynolds numbers are
WeL = u2L ρL S/σ (16.72)
FrL = u2L /(Sg) (16.73)
L L
ReL = uL Sρ /η (16.74)

347
 The effective g (as a function of ht ) is obtained from:

(ρL − ρV )
 
dP dZ
gef f = 1− g (16.75)
dP dZf lood ρL

Then Ft (see above), ht , and dPdZ are computed

2/3  1/3
3η L uL

4Ft
ht = V
(16.76)
S ρ sin θgef f
0.177ρV
A = (16.77)
S2 (sin θ)2
88.774η V
B = (16.78)
S 2  sin θ
 5
∆p 1
= (Au2V + BuV ) (16.79)
∆z 1 − K2 ht

The calculation is repeated until pressure drop converges or when it becomes

larger than the pressure drop at flood. There can be problems converging this
method.

BS-92 Billet and Schultes (1992) and Billet’s monograph (1979?) [ap , , Cf l, Ch ,
Cp ] See the section on pressure drop of random packed columns above.

16.3 Entrainment and Weeping

Entrainment can occur in packings and cause a rapid loss of efficiency due to back-
mixing but currently is not accounted for. Weeping does not occur in packed columns.

16.4 Packing Flooding and Minimum Wetting

The fraction of flooding for packings is computed by dividing the superficial gas ve-
locity by the gas velocity at flood. The latter is found in an iterative process from the
pressure drop correlation (keeping the liquid to vapor ratio constant!) and the speci-
fied pressure drop of flood. If no flood pressure drop is specified it is computed from
the packing factor with the Kister and Gill correlation. If no pressure drop model is
selected, the Leva’s method is used, which also has the packing factor as only param-
eter.

348
The minimum operating conditions, similar to the weep point for tray operation, is set
to the minimum wetting rate predicted by Schmidt (1979). Schmidt calculates a liquid
falling number

CL = 1633.6(0.062428ρL )(1000σ)3 /(1000ηL )4 (16.80)

and the minimum wetting rate is (in U Sgal/min/f t2 ):

2/9
0.3182CL (1 − cos φ)2/3
QM W = p (16.81)
(1 − TL )ap /0.3048
where φ is the liquid contact angle and TL is the shear stress number:

TL = 0.9F F −2.8 (16.82)

with F F as the fraction of flood. The contact angle for metals is 10o (0.1745 rad).
To obtain contact angles for other packings the contact angle is assumed inverse pro-
portional to the critical surface tension, σc . For metals we have σc = 0.075, so the
contact angle is computed with
 
0.075
φ = 0.1745 (16.83)
σc
With this approach we obtaim higher minimum wetting rates for plastic packings than
for metal packings. The minimum operating or ‘weep’ fraction W F is then taken as
QM W
WF = (16.84)
uL
where uL is the superficial liquid flow.

16.5 Column Design

For packings only the column diameter and bed height are design parameters, other
parameters (such as void fraction, nominal packing diameter, etc.) are fixed once the
type of packing has been selected. The packed bed height must be specified since it
determines the desired separation and the capacity.

16.5.1 Packed Column Design: Fraction of flooding

For packed columns only the column diameter is to be estimated. Default packing
data are used for all parameters that are not specified; values of 1” metal Pall rings

349
for random packed sections and of Koch Flexipack 2 (316SS) for structured sections.
To determine the packed column diameter, the diameter that gives rise to the flooding
pressure drop (as specified) is computed using the selected pressure drop model. The
resulting diameter is corrected for the fraction of flooding and the system factor:

Dc,f lood
Dc = √ (16.85)
F F SF

This makes the resulting column diameter depend on the selected pressure drop model.
If no pressure drop model is selected the Leva (1992) model is selected (which is a
function only of the packing factor). If no pressure drop at flood is specified, it is
estimated with Kister and Gill correlation (1992):

∆Pf lood = 0.115Fp0.7 (16.86)

(with pressure drop as liquid height and English units!). This correlation is a function
only of the packing factor but has been tested on a wide range of packings and an
accuracy of 15 %. As long as the packing factor is known, this design method will not
fail.

16.5.2 Packed Column Design: Pressure drop

Tray design based on a specified pressure drop is done as discussed above but with a
default fraction of flooding of 75 %. However, the specified pressure drop functions
as a maximum allowed pressure drop per tray. No adjustment is done if the pressure
drop is below this specified pressure drop.

Packed column design automatically finds the diameter leading to the specified pres-
sure drop (with the selected pressure drop model). This is done by using a linear search
technique as the different packing pressure drop correlations can behave quite errati-
cally. The maximum allowed pressure drop is the flooding pressure drop as specified
or computed from Kister’s correlation and the packing factor. If the pressure drop is
specified to be very low the column diameter might converge to unrealistic diameters.
A zero or larger than flooding pressure drop specification results in a 70 % fraction of
flooding design.

350
Symbol List

Latin Symbols

ad interfacial area density (m2 /m3 )

a area (m2 )
Ah hole area (m2 )
Ab , Abub bubbling area (m2 )
Ad downcomer area (m2 )
B packing base (m)
c number of components,
molar concentration (kmol/m3 )
Ca Capillary number
dh hole diameter (m)
deq equivalent diameter
D binary diffusivity coefficient (m2 /s)
Dc column diameter (m)
De eddy dispersion coefficient (m2 /s)
f1 , f2 , f3 design adjustment factors
Fp packing factor (1/m)
√
Fs F factor Fs = Uv ρV (kg 0.5 /m0.5 /s)
FF fraction of flooding p
FP flow parameter F P = ML /MV ρVt /ρL t
Fr Froude number
g gravitational constant, 9.81 (m/s2 )
hc clearance height under downcomer (m)
hcl clear liquid height (m)
hd dry tray pressure drop height (m)
hdb downcomer backup liquid height (m)
hf froth height (m)
hi liquid height at tray inlet (m)
hlg liquid gradient pressure drop height (m)
hl , hL liquid pressure drop height (m)
how height of liquid over weir (m)
hr residual pressure drop liquid height (m)
hwt wet tray pressure drop liquid height (m)
hw weir height (m)
hudc liquid height pressure loss under downcomer (m)
k binary mass transfer coefficient (m/s)
Le Lewis number (Le = Sc/Pr)

351
Mw molecular weight (kg/kmol)
N number of transfer units, NTU
P perimeter (m)
p hole pitch (m),
pressure (P a)
∆p pressure drop (P a)
∆Pmax maximum design pressure drop (P a/tray or P a/m)
Pr Prandtl number
Q volumetric flow (m3 /s)
Re Reynolds number
S packing side (m)
Sc Schmidt number
SF system derating factor
t residence time (s)
tv valve thickness (m)
T temperature (K)
Ts tray spacing (m)
u, U velocity (m/s)
V vapor flow rate (kmol/s)
We Weber number
Wl weir length (m)
x liquid mole fraction
y vapor mole fraction
z mole fraction
Z tray flow path length (m)

Greek Symbols

α fraction liquid in froth

β fractional free area β = Ah /Ab ,
 void fraction
Γ liquid flow per perimeter
φ fractional entrainment
ρ density (kg/m3 )
σ surface tension (N/m)
η viscosity (P a s)
λ thermal conductivity (W/m/K)

352
Superscripts

I interface
L liquid
P phase P
V vapor

Subscripts

b, bub bubbling
c critical, contact
ef f effective
f l, f lood at flooding conditions
i component i
j stage j,
component j
h hole
MW minimum wetting
p packing
spec specified
t total

Abbreviations

bc bubble-caps

References

Bubble Tray Design Manual: Prediction of Fractionation Efficiency, AIChE, New

York (1958).

P.E. Barker, M.F. Self, “The evaluation of Liquid Mixing Effects on a Sieve Plate using
Unsteady and Steady-State Tracer Techniques”, Chem. Eng. Sci., Vol. 17, (1962) p.
541.

353
S.D. Barnicki, J.F. Davis, “Designing Sieve-Tray Columns, Part 1: Tray Design”,
Chem. Engng., Vol. 96, No. 10 (1989) pp. 140–146.

S.D. Barnicki, J.F. Davis, “Designing Sieve-Tray Columns, Part 2: Column Design
and Verification”, Chem. Engng., November (1989) pp. 202–212.

D.L. Bennett, R. Agrawal, P.J. Cook, “New Pressure Drop Correlation fo Sieve Tray
Distillation Columns”, AIChE J., Vol. 29 (1983) pp. 434–442.

R. Billet, M. Schultes, ”Advantage in correlating packed column perfomance”, IChemE.

Symp. Ser. No. 128 (1992) p. B129.

J.L Bravo, J.R. Fair, ”Generalized Correlation for Mass Transfer in Packed Distillation
Columns”, Ind. Eng. Chem. Process Des. Dev., Vol. 21 (1982) pp. 162–170.

J.L. Bravo, J.A. Rocha, J.R. Fair, ”Mass Transfer in Gauze Packings”, Hydrocarbon
Processing, January (1985) pp. 91–95.

J.L. Bravo, J.A. Rocha, J.R. Fair, ”Pressure Drop in Structured Packings”, Hydrocar-
bon Processing, March (1986) pp. 45–48.

J.L. Bravo, J.A. Rocha, J.R. Fair, ”A Comprehensive Model for the Performance of
Columns Containing Structured Packings”, I. ChemE. Symp. Ser., No. 128 (1992) pp.
A439–A457.

H. Chan, J.R. Fair, “Prediction of Point Efficiencies on Sieve Trays. 1. Binary Sys-
tems”, Ind. Eng. Chem. Process Des. Dev., Vol. bf 23 (1984) pp. 814–819.

G.X. Chen and K.T. Chuang, “Prediction of Efficiencies for Sieve Trays in Distilla-
tion”, Ind. Eng. Chem. Res., Vol. 32 (1993) pp. 701–708.

I.J. Harris, “ Optimum Design of Sieve Trays”, Brit. Chem. Engng., Vol. 10, No. 6
(1965) p. 377.

G.A. Hughmark, “Models for Vapour Phase and Liquid Phase Mass Transfer on Dis-
tillation Trays”, AIChE J., Vol. 17, No. 6 (1971) p. 1295.

F. Kastanek, “Efficiencies of Different Types of Distillation Plate”, Coll. Czech.

Chem. Comm., Vol. 35 (1970) p. 1170.

H.Z. Kister, Distillation Design, McGraw-Hill, New York (1992).

G.F. Klein, Chem. Engng., May 3 (1982) p. 81.

M. Leva, Tower Packings and Packed Tower Design, The U.S. Stoneware Company

354
(1951).

M. Leva, “Reconsider Packed-Tower Pressure-Drop correlations”, Chem. Eng. Prog.

January (1992) p. 65.

M.J. Lockett, Distillation Tray Fundamentals, Cambridge University Press (1986).

M.J. Lockett, S. Banik, “Weeping from Sieve Trays”, Ind. Eng. Chem. Process Des.
Dev., Vol. bf 25 (1986) p. 561.

E.E. Ludwig, Applied Process Design for Chemical and Petrochemical Plants, Vol. 2,
2nd Ed., Gulf Pub. Co., Houston, TX, (1979).

O. Ogboja, A. Kuye, “A Procedure for the Design and Optimisation of Sieve Trays”,
Trans. I. Chem. E., Vol. 68, Part A (1990) pp. 445-452.

K. Onda, H. Takeuchi, Y. Okumoto, ”Mass Transfer Coefficients Between Gas and

Liquid Phases in Packed Columns”, J. Chem. Eng. Jap., Vol. 1, No. 1 (1968) p. 56.

M. Prado, J.R. Fair, “Fundamental Model for the Prediction of Sieve Tray Efficiency”,
Ind. Eng. Chem. Res., Vol. 29, No. 6 (1990) pp. 1031–1042.

M. Prado, PhD Thesis, Austin, TX (1986).

R.H. Perry and D. Green,Perry’s Chemical Engineering Handbook, 6th edition, sec-
tion 18, Liquid-Gas System (1984) pp. 18-8 – 18-12.

B.D. Smith, Design of Equilibrium Staged Processes, McGraw-Hill, New York (1963).

J. Stichlmair, A. Mersmann, “Dimensioning Plate Columns for Absorption and Recti-

fication”, Chem. Ing. Tech., Vol. 45, No. 5 (1978) p. 242.

J. Stichlmair, J.L. Bravo, J.R. Fair, Gas. Sep. Purif., Vol. 3, (1989) p. 19.

R. Taylor, R. Krishna, Multicomponent Mass Transfer, Wiley NY (1993)

P.C. Wankat, Separations in Chemical Engineering - Equilibrium Staged Separations,

Elsevier (1988) pp. 420–428.

F.J. Zuiderweg, “Sieve Trays - A View of the State of the Art”, Chem. Eng. Sci., Vol.
37 (1982) pp. 1441–1461.

355
356
Chapter 17

Design Models for Extraction

Columns

This chapter deals with the application of the nonequilibrium model to the modelling
of extraction columns. In such operations the two phases present are both liquids
instead of a liquid and a vapor as in the case of distillation, stripping, or absorption.
This requires fundamentally different mass transfer coefficients and flow models, as
well as a completely new design method.

17.1 Introduction

Nonequilibrium modelling of extraction uses essentially the same model as described

in Chapter 14 (see Lao et al., 1989). In extraction, of course we have light and a heavy
liquid phases, where the light liquid behaves as the vapor with, of course, liquid-like
properties. If the heavy phase (L) is lighter than the light phase (V) the program stops.
However, either phase (that is, L or V) can be the dispersed phase. The user must
specify which is the disperse phase, since this changes the design and the calculation
of MTC’s. Currently sieve trays, structured and random packed columns, rotating
disc contacters, and spray columns are supported as internals (as well as equilibrium
stages with a specified stage efficiency). The K-values must be the liquid-liquid ex-
traction model, which uses activity coefficients. The energy balance can be ignored
(Enthalpy=None) or included. In case it is ignored the column temperature is dictated
by that of the feeds, and linear interpolation is used to provide a column temperature

357
profile. A specific temperature profile can be imposed if the energy balance is ignored
and the user initializes all of the temperatures. Default values for the total interfacial
area and mass transfer coefficients are: Ai = 100 m2 , kd = 10−5 and kc = 10−4
m/s. Mass transfer in coalescencing layers and jet zones is neglected (they could
be modeled by a special stage for packed/RDC columns). Thus, only the drop rise
zone is taken into account for mass transfer. Current limitations of the nonequilibrium
extraction model are:

• No efficiencies are back-calculated (yet),

• Limited number of mass transfer coefficient correlation,

• No comparisons of simulations with experiment have been performed,

• Backflow and dispersion calculations do not work properly (yet).

The nonequilibrium extraction work still requires more work. There is an abundance
of correlations for extraction but no benchmarks have been done to indicate what are
the right models.

17.2 Sieve Tray Columns

ChemSep will attempt to design the extraction column if no design is specified. This
sieve tray design method is adapted from course notes by R. Krishna.

17.2.1 Design

The default free area ratio is 5%, tray spacing is 0.4m, and the clearance under the
downcomer a quarter of the tray spacing. There is no weir and the hole diameter is set
by default to:
r
σ
x = (17.1)
∆ρg
dh = 1.8x (17.2)

but dh is limited (if supplied) by:

0.5x < dh < πx (17.3)

358
and the practical limits (overriding):

3mm < dh < 8mm (17.4)

The hole velocity is computed with:

∆ρgd2h
Eo = (17.5)
σ
We = 4.33Eo−0.26 (17.6)
s
Weσ
Uh = (17.7)
ρd d h

If the hole velocity is less than 0.15 (m/s) then its design value is kept at 0.15 (m/s).
The Froude number is computed from

Uh2
Fr = (17.8)
gdh

For Eo is less than 0.4 the Sauter mean droplet diameter is computed by:
 0.67 !
−0.4 ∆ρ
dp = Eo 2.13 + exp(−0.13Fr) dh (17.9)
ρd

otherwise
dp = Eo−0.42 1.24 + exp(−Fr0.42 ) dh


(17.10)
The hole area is
Qd
Ah = (17.11)
Uh
The ratio of the hole area over the active area (free area ratio, f ) is limited to lie
between 1 and 20 %.
Aa = Ah /f (17.12)
The hole pitch can be computed if the hole diameter and free area ratio are known.
The downcomer velocity can be computed if a minimum droplet diameter, dmin , is
assumed that should not be entrained. This diameter is taken to be 0.5 mm and the
downcomer velocity is the velocity of the continuous phase, Uc :
0.33
(g∆ρ)2

Uc = 0.249dmin (17.13)
ρc η c

With Uc known we can compute the downcomer area:

Ad = Qc /Uc (17.14)

359
The total area is equal to two downcomer areas plus the active area and 0.5 % area for
support etc.:
At = (Aa + 2Ad )/0.995 (17.15)
With the total tray area known the column diameter can be computed. The net area
for the disperse phase, An , and the disperse velocity, Ud , are:

An = AA + Ad (17.16)
Qd
Ud = (17.17)
An
Next, the dispersed phase velocity holdup and slip velocity are computed. The slip
velocity (Vs ) is estimated to be one tenth of the dispersed phase velocity. This makes
the dispersed phase holdup equal to one tenth since it is defined as
Ud
φd = (17.18)
Vs
The slip velocity (a function of the dispersed phase holdup and needs to be obtained
iteratively) can be calculated from:
s   1.834
∆ρ (1 − φd )
Vs = 2.725gdp (17.19)
ρc (1 + φ0.33
d )

After the dispersed phase holdup is computed (it depends on Vs ), it is checked to

ensure that it lies between 1 and 20 % for standard operational conditions. If it is too
small the free area ratio is increased, if it is too large the free area ratio is decreased
(each by 5 %) until it is within the desired range.

The Weber number

We = ρd Uh2 dp /σ (17.20)
must be larger than 2 to ensure all holes produce drops (i.e. to avoid inactive holes,
see Seibert and Fair, 1988).

The height of the coalesced layer is (Treybal, 1980):

(Uh2 − Ud2 )ρd 4.5Uc ρc 6σ

hc = 2 + + (17.21)
2gCd ∆ρ 2g∆ρ dp g∆ρ

(with Cd = 0.67). The first term is height to overcome flow through the orrifices. The
second term is for friction losses due to contraction/expansion on entry/exit (0.5 + 1.0
velocity heads) and change of direction (2 times 1.5 velocity heads). The third term is
for the interfacial tension effects at the holes. The height needs to be larger than 4 cm
(to ensure safe operation). If not, then the hole diameter is decreased by 5 % and we
repeat the procedure from the hole velocity calculation, Eq. (17.5).

360
This design is for a one pass sieve tray, and the flow path length, Lf is computed from
geometric relations. The weir length is (segmental downcomer):

Wl = Aa /Lf (17.22)

The tray thickness is defaulted to a tenth of an inch. To prevent entrainment of

droplets, the flow under the downcomer is only allowed to be 50 % higher than the
downcomer velocity. If higher, then the downcomer clearance is enlarged until this
requirement is met. The tray spacing is adjusted so that the coalesced layer and coa-
lescence zone divided over the length of the downcomer equals the fraction of flooding
(multiplied with the system factor).

The reported fraction of flooding is to the ratio of the height of the coalesced layer over
the height of the downcomer (according to Seibert and Fair the flooding calculation is
within 20 %). The lower operating limit is the ratio of two over the Weber number (to
guarantee proper droplet formation).

17.2.2 Mass Transfer Coefficients

The ‘Handlos-Baron-Treybal’ method is used. The hole velocity Uh , Eo, F r, net area
An , Sauter mean drop diameter dp , disperse velocity Ud , slip velocity Vs , disperse
phase holdup φd , hc , and hz are computed as above (but with fixed design parameters).
The interfacial area per unit of volume is

6φd
Ai = (17.23)
dp

and the drop rise zone (where mass transfer is assumed to take place):

hdrop = ts − hc (17.24)

where ts is the tray spacing. The volume for mass transfer on a tray is

Vi = An hdrop (17.25)

The mass transfer coefficients for transport from the dispersed phase are (Handlos and
Baron, 1957):
0.00375Vs
kd = (17.26)
(1 + ηd /ηc )
and for transport from the continuous phase are (Treybal, 1963):

kc,ij = 0.725Rec−0.43 (1 − φd )Vs Nuc−0.58 (17.27)

361
with

Rec = dp Vs ρc /ηc (17.28)

c
Nuc = ηc /ρc Dij (17.29)

Note that kd is not a function of the diffusion coefficient and, thus, is the same for all
components.

17.3 Packed Columns

Column design and calculation of mass transfer coefficients is done the same way for
structured packings and random packings, following the methods and correlations as
outlined by Seibert and Fair (1988).

17.3.1 Design

For mass transfer from the continuous phase to the dispersed phase we have x = 1 for
the calculation of the Sauter mean drop diameter:
r
σ
dp = 1.15x (17.30)
∆ρg

The slip velocity of a single droplet at zero disperse phase holdup is given by
s
o 4∆ρgdp
Vs = (17.31)
3ρc Cd

where Cd = 0.38 (for high values of Reynolds). Static disperse holdup is:
0.076ap dp
φds = (17.32)
ξ
where ap is the packing area and ξ the packing void fraction. The static holdup area
and total area are:

as = 60.076ap (17.33)
a = ap + as (17.34)

The tortuosity is defined as

adp
ζ= (17.35)
2

362
The superficial velocity of the continuous phase at the flood point is
πζ
e = cos( ) (17.36)
4
0.192ξVso
Ucf = (17.37)
(1.08 + (Qd /Qc )/e2 )
This needs to be corrected for the fraction of flooding (and system factor):

Uc = SF × F F × Ucf (17.38)

to give the net area

Qc
An = (17.39)
Uc
from which the packed column diameter (Dc ) can be calculated.

The reported fraction of flooding is the quotient of computed Uc to Ucf as discussed

above. The dispersion coefficients are given by (see Section 21 of Perry and Green,
1984):
Ed Vc
log = 0.046 + 0.301 (17.40)
Vd dp Vd
Ec Vc
log = 0.161 + 0.347 (17.41)
Vc dp Vd

where dp is the packing diameter.

17.3.2 Mass Transfer Coefficients

The method of Seibert and Fair (1988) is used. The phase velocities are computed by
Qc
Uc = (17.42)
An
Qd
Ud = (17.43)
An
The drop diameter dp , slip velocity Vso , area a, static holdup area as , and tortuosity ζ
are calculated as above. Then the dispersed phase holdup, φd , is determined iteratively
(starting at 0.1.) from:
−6φd
f (φd ) = exp( ) (17.44)
π
Ud
φd = (17.45)
ξ(Vso f (φd ) − Uc )e2

363
Then the slip velocity is

Vs = Vso f (φd )e + (1 − e)Uc (17.46)

since Ud = Vso f (φd ). The mass transfer coefficient for the dispersed phase is com-
puted by:
√
Scd
φ = (17.47)
(1 + ηd /ηc )
0.023Vs
φ > 6 : kd,ij = √ (17.48)
Scd
0.00375Vs
φ < 6 : kd = (17.49)
(1 + ηd /ηc )
ηd
Scd = (17.50)
ρd Dd,ij
(17.51)

If φ is larger than 6 the Laddha and Degaleesan correlation is used otherwise the
Handlos-Baron method. For the mass transfer coefficient in the continuous phase:

Shc = 0.698Re0.5 0.4

c Scc (1 − φd ) (17.52)
Shc Dc,ij
kc,ij = (17.53)
dp
ρc V s d p
Rec = (17.54)
ηc
ηc
Scc = (17.55)
ρc Dc,ij

The interfacial area per unit volume is:

6ξφd
ai = (17.56)
dp

The total interfacial area in a stage is the stage height times the net area times the
interfacial area per unit volume:

ai,tot = ai An hstage (17.57)

17.4 Rotating Disk Contactors

This design method is based on material in the Handbook of Solvent Extraction (Chap-
ter 13.1) and course notes by R. Krishna.

364
17.4.1 Design

The phase ratio α is

Qd
α= (17.58)
Qc
The maximum stable drop diameter is

(g∆ρ)5/21 σ 6/21
u0 = 0.9 10/21 1/21
(17.59)
ρc ρd
σ
dp,max = (17.60)
ρc u20

A stable drop diameter is selected as half of the maximum diameter

dp = 0.5dp,max (17.61)

and the require power input (Pi = N 3 R5 /HD2 ) is computed

  0.6 2.5
σ
 0.25 ρc
e =  (17.62)

dp


πe
Pi = (17.63)
4Cp

(Cp = 0.03 for Re > 105 ). If no column diameter is known, an estimate is made from
assuming a cross-sectional area for a combined velocity of 0.05 m/s with:

Ac = (Qc + Qd )/0.05 (17.64)

The required rotational speed (using these standard ratios) is

0.1

!0.33
Pi 0.65
N= (17.65)
Dc2

The slip velocity can be calculated using a correlation from Kung and Beckman
(1961):
 0.9  2.3  0.9  2.6 
V s ηc ∆ρ S H R g 
= (17.66)
σ ρc R R D RN 2
The dispersed phase holdup at flood is determined from
√
α2 + 8α − 3α
φd = (17.67)
4(1 − α)

365
from which the continuous phase velocity at flood can be determined with

Uc,f = Vs (1 − φd )2 (1 − 2φd ) (17.68)

Correction for fraction of flooding (and system factor) gives

Uc = SF × F F × Ucf (17.69)

from which the column area and diameter can be calculated

Qc
Ac = (17.70)
Uc

The rotor diameter R, stator diameter S, and the height of the compartment have
standard ratios with respect to the column diameter (Dc )

R = 0.6Dc (17.71)
S = 0.7Dc (17.72)
H = 0.1Dc (17.73)

These ratios help determine the size of the column. Below a Reynolds number of 105
Cp becomes a function of the Reynolds number. Normally RDC’s are operated in the
regime above 105 so the Reynolds number is computed by

ρd N R 2
Red = (17.74)
ηd

and a smaller diameter is selected (and the calculations repeated) if necessary. On re-
design the layout of the stage with the largest diameter is used for the entire section.

The reported fraction of flooding is the quotient of computed Uc over Ucf as discussed
above. The operating velocity is proportional to the slip velocity and so inverse pro-
portional to the square of the rotation speed. One of the design rules was to keep
the dispersed phase Reynolds number larger than 105 so the lower operating limit is
defined as:
 5 2
10
(17.75)
Red
Stemerding et al. (1963) gave a correlation for the axial dispersion coefficient for the
continuous phase
Ec
= 0.5 + 0.012N R(S/D)2 /Vc (17.76)
Vc H
The dispersed phase dispersion coefficient is set to twice this number.

366
17.4.2 Mass Transfer Coefficients

The method of ‘Kronig-Brink-Rowe’ is used. Phase ratio α, energy input Pi (from N ,

R, H, and Dc ) are computed as above. The drop diameter is computed from

Cp =
0.03 (17.77)
4Cp Pi
e = (17.78)
π
0.6
0.25 (σ/e)
dp = (17.79)
ρ0.4
c

The dispersed phase holdup φd is calculated iteratively and the slip velocity is deter-
mined as described above (with 17.66). The mass transfer coefficients are:

Shd = 10.0 (17.80)

Shd Dd,ij
kd,ij = (17.81)
dp
Shc = 2 + 0.42Re0.62
c Sc0.36
c (17.82)
Shc Dc,ij
kc,ij = (17.83)
dp

with

ρc d1.33
p e0.33
Rec = (17.84)
ηc
ηc
Scc = (17.85)
ρc Dc,ij

The interfacial area per unit volume is

6φd
ai = (17.86)
dp

Alternatively, the ”Rose-Kintner-Garner-Tayeban” method can be used:

p p
Shc = 0.6 Rec Scc (17.87)
b = d0.225
p /1.242 (17.88)
8σb n(n + 1)(n − 1)(n + 2)
ω = (17.89)
dp (n + 1)ρd + nρc
p
kd = 0.45 Dd,ij ω (17.90)

where n = 2, and dp is in cm for the calculation of b and ω.

367
17.5 Spray Columns

This design method is adapted from Jordan (1968) and Lo et al. (1983).

17.5.1 Design

The height of a stage in a spray column is set to the default value of 0.4 m and the
hole diameter in the distributor to 0.005 m. The hole velocity (Uo ) in the distributor
is set to 0.1 m/s from which the total hole area is then:
Ao = Qd /Uo (17.91)
The droplet diameter can be calculated from (Vedaiyan et al, 1972):
 2 −0.067 r
Uo σ
dp = 1.592 (17.92)
2gdo g∆ρ
The flood velocity of the continuous phase is (Treybal, 1963):
0.3894∆ρ0.28
Ucf =  √ √ 2 (17.93)
0.2165ηc0.075 ρc + 0.2670d0.056
p ρd α
where α = Qd /Qc . The dispersed phase holdup at flood is
√
α2 + 8α − 3α
φdf = (17.94)
4(1 − α)
The velocity of the continuous phase is then
Uc = F F × SF × Ucf (17.95)
and the column area
Ac = Qc /Uc (17.96)
from which the column diameter can be calculated. (The column area must also be
larger than the total hole area, if not, the column area is set to four times the hole area.)

The fraction of flooding reported is calculated as

Uc
FF = (17.97)
SF × Ucf
where Ucf is computed as in the spray column design and Uc = Ac /Qc . No lower
operating limit is calculated. The dispersion coefficient for the continuous phase is
(see Section 21 of Perry and Green, 1984):
Ec p
= 7.2 Ud Dc (17.98)
Vc H

368
Since the dispersion coefficient for the dispersed phase is unknown it is set equal to
that for the continuous phase.

17.5.2 Mass Transfer Coefficients

The transition drop size below which droplets become stagnent is calculated from
ρ2c σ 4
P = (17.99)
gηc4 ∆ρ
r
σ
dp,t = 7.25 (17.100)
g∆ρP 0.15
The drop terminal velocity is (Satish et al., 1974):
 2 −0.082  1/4
Uo σg∆ρ
Vt = 1.088 (17.101)
2gdo ρ2c
With the continuous operating and flood velocities the fraction of flooding is calcu-
lated and then the dispersed phase holdup is
φd = F F φdf (17.102)
and the slip velocity
Vs = (1 − φd )Vt (17.103)
If the drops are stagnent (dp < dp,t ) the dispersed phase MTC is computed from
kd,ij = 18.9Dd,ij /dp (17.104)
otherwise the Handlos-Baron correlation (1957) is used:
0.00375Vs
kd = (17.105)
(1 + ηd /ηc )
For the continuous phase MTC we use (see Section 21 of Perry and Green, 1984)
kc = 0.725Rec−0.43 Scc−0.58 (1 − φd )Vs (17.106)
where
Rec = dp Vs ρc /ηc (17.107)
ηc
Scc = (17.108)
ρc Dc,ij
The interfacial area for mass transfer per unit of volume is
6φd
Ai = (17.109)
dp

369
17.6 Modeling Backflow

The backflows in an extraction column may be estimated from the dispersion coeffi-
cients with:
Ed
αd = − 0.5 (17.110)
Vd H
Ec
αc = − 0.5 (17.111)
Vc H
where α is the fractional backflow (”entrainment”) in the stage, and H is the stage
height.

For spray columns (see Section 21 of Perry and Green, 1984):

p
Ec = 7.2 Vd Dc (17.112)
For packed columns:
Ed Vc
log = 0.046 + 0.301 (17.113)
Vd dF Vd
Ec Vc
log = 0.161 + 0.347 (17.114)
Vc dF Vd
For a RDC:
 
S
Ec = 0.5HVc + 0.012RN H (17.115)
Dc
Ed = F Ec (17.116)
where F is calculated from
3.3
4.2105

Vd
F = (17.117)
Dc2 h
F must be greater than or equal to one. Krishna uses:
 
0.5HUc S
Ec = + 0.012RN H (17.118)
(1 − Φd ) Dc
 
0.5HUd S
Ed = + 0.024RN H (17.119)
Φd Dc

Symbol List

Latin Symbols

ap packing area per unit volume (m2 /m3 )

370
as static holdup area per unit volume (m2 /m3 )
Aa total tray active area (m2 )
Ad downcomer area (m2 )
Ai interfacial area per unit volume (m2 /m3 )
Ah total tray hole area (m2 )
An netto tray area (m2 ), An = Aa + Ad
At total tray area (m2 )
Cd drag coefficient
D binary diffusion coefficient (m2 /s)
Dc column diameter (m)
de effective drop diameter (m) ?
dh hole diameter (m)
dmin minimum droplet diameter (m)
dp Sauter mean drop diameter (m)
Eo Eotvos number (∆ρgdh /σ)
f free area ratio (Ah /Aa )
F molar flow (kmol/s)
Fr Froude number (Uh2 /g/dh )
FF fraction of flooding
g, gc gravitational constant, 9.81 (m/s2 )
H RDC compartment height (m)
hc height of coalesced layer (m)
h, hdrop height of drop rising zone (m)
hstage stage height for packed column (m)
k binary mass transfer coefficient (m/s)
Mw molecular weight (kg/kmol)
N rotation speed (rad/s)
Nu Nusselt number
Pe Peclet number
Pi power input (?)
Q volumetric flow (m3 /s)
R rotor diameter (m)
Re Reynolds number
S inner stator diameter (m)
Sc Schmidt number
Sh Sherwood number
SF system derating factor
t contact time (s)
ts tray spacing (m)
Uc ,Ud continuous, dispersed phase velocity (m/s)
Ucf continuous phase superficial velocity at flood (m/s)

371
Uh hole diameter (m/s)
Vi tray volume for interfacial mass transport (m3 )
Vs slip velocity (m/s)
Vso slip velocity at zero disperse phase holdup (m/s)
We Weber number (ρd Uh2 dp /σ)
Wl weir length (m)

Greek Symbols

α phase ratio (Qd /Qc )

ρ mass density (kg/m3 )
φd disperse phase holdup fraction
φds static disperse phase holdup fraction
σ interfacial tension (N/m)
η liquid viscosity (P a.s)
µ kinematic viscosity (η/ρ)
ζ tortuosity
κ
δ

Subscripts

c continuous phase
d disperse phase,
downcomer
i interface,
component i
j component j

References and Further Reading

F.H. Garner, M. Tayeban, Anal. Real Soc. Espan. Fis. Quim. (Madrid), B56 (1960)
pp. 479.

372
A.E. Handlos, T. Baron, “Mass and Heat Transfer from Drops in Liquid-Liquid Ex-
traction”, AIChE J., Vol. 3 (1957) pp. 127–136.

D.G. Jordan, Chemical Process Development, Part 2, John Wiley, New York (1968).

R. Krishna, Design of Liquid-Liquid Extraction Columns, University of Amsterdam,

NL (1993).

R. Kronig, J.C. Brink, ”The Theory of Extraction from Falling Droplets”, Appl. Sci
Res., Vol. A2 (1950) pp. 142–154.

M. Lao, J. P. Kingsley, R. Krishnamurthy and R. Taylor, ”A Nonequilibrium Stage

Model of Multicomponent Separation Processes VI: Simulation of Liquid-Liquid Ex-
traction”, Chem. Eng. Comm., Vol. 86 (1989) pp. 73–89.

G.S. Laddha, T.E. Degaleesan, Transport Phenomena in Liquid Extraction, McGraw-

Hill (1978).

T.C. Lo, M.H.I. Baird, C. Hanson, Handbook of Solvent Extraction, John Wiley, NY
(1983).

R.H. Perry and D. Green, Perry’s Chemical Engineering Handbook, 6th Ed. (1984).

P.N. Rowe, K.T. Claxton, J.B. Lewis, Trans. Inst. Chem. Eng., 43, T14 (1965).

P.M. Rose, R.C. Kinter, “Mass Transfer from Large Oscillating Drops”, AIChE J., Vol.
12, No. 3 (1966) p. 530.

E.Y. Kung, R.B. Beckman, “Dispersed-Phase Holdup in a Rotating Disk Extraction

Column”, AIChE J., Vol. 7, No. 2 (1961) pp. 319-324.

L. Satish, T.E. Degaleesan, G.S. Laddha, Indian Chem. Eng., Vol. 16 (1974) p. 36.

A.F. Seibert, J.R. Fair, ”Hydrodynamics and Mass Transfer in Spray and Packed
Liquid-Liquid Extraction Columns”, Ind. Eng. Chem. Res., 27, No.3 (1988) pp.
470–481.

S. Stemerding, E.C. Lumb, J. Lips, “Axiale Vermischung in einer Drehscheiben-

Extraktions Kolonne”, Chem. Ing. Tech, Vol. 35 (1963) pp. 844–850.

R.E. Treybal, Mass Transfer Operations, 3rd ed., McGraw-Hill, New York (1980)

R.E. Treybal, Liquid Extraction, 2nd ed., McGraw-Hill, New York (1963).

S. Vedaiyan, T.E. Degaleesan, G.S. Laddha, H.E. Hoelscher, “Some Performance

373
Characteristics of Spray Columns”, AIChE J., Vol. 18 (1972) pp. 161–168.

374
Chapter 18

Nonequilibrium Distillation
Dynamics

375
376