Professional Documents
Culture Documents
net/publication/324182477
CITATIONS READS
0 2,667
1 author:
Binous Housam
University of Carthage
70 PUBLICATIONS 432 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Binous Housam on 04 April 2018.
HABILITATION
February 2016
1
Habilitation Dr. Housam Binous 2015
2
Habilitation Dr. Housam Binous 2015
Acknowledgment
3
Habilitation Dr. Housam Binous 2015
Table of Contents
Transport Phenomena…p-69
Engineering Applications…p-129
Application to Distillation…p-146
4
Habilitation Dr. Housam Binous 2015
The present report starts with a chronological description of my research achievement while
KFUPM (Dhahran). Later, seven chapters describe the main research axis, which I have grown
to like the most, namely the application of numerical techniques to solve chemical engineering
problems. These problems are related to transport phenomena (momentum – heat – mass
transfer), distillation (standard, reactive and extractive), chemical reaction engineering and
thermodynamics.
California at Davis and a year as a trainee at Shell Tunisia (in 1999), I started my academic
career in Tunisia, first as a lecturer and later as an assistant professor both at the National
Institute of Applied Sciences and Technology in Tunis (INSAT). Under the supervision of
Professor Naceur Ayed, I have been involved in some research in industrial organic chemistry
have presented a poster in Budapest, Hungary, on textile dyes with special emphasis on the
imperial purple. In the Fall semester of the same year (i.e., in the year 2000), I have presented
a talk on the characterization of the imperial purple dye, with various analytical techniques
such as IR spectroscopy and XRD, at Stanford, USA. My collaboration with Professor Naceur
Ayed at INSAT resulted in the publication of two papers (one in the famous “revue
d’Archeometrie” in 2001 and the other in Applied Physics A in 2006). Meanwhile, I have
continued perfecting my chemical engineering and computational skills. This led to a series of
pedagogical papers on the usage of the state-of-the-art mathematical and process design
software such as Hysys©, Matlab® and MATHEMATICA©. These peer-reviewed and abstracted
5
Habilitation Dr. Housam Binous 2015
(scopus) papers were mainly single-author papers and dealt with distillation dynamic and
control, liquid-liquid extraction, residue curve maps for reactive and non-reactive mixtures,
flow of non-Newtonian fluids (power-law, Carreau, and Bingham fluids), cubic equations of
state (both the Peng-Robinson and the Soave-Redlich-Kwong equation of states) and gas
permeation in nonporous membranes. These papers have been published from 2005 until 2012
Engineering Education and Computer in Education Journal. During the period from 2007-
2010, I have supervised or co-supervised 5 Masters’ students at INSAT, all of whom graduated
with the “mention très bien”. Mrs Manel Abdessamia was under the supervision of Professor
Naceur Ayed and I in 2007. She performed her thesis work on pharmaceuticals. My
contribution was to help Mrs. Manel with the experimental design part of her project. Then,
Mrs. Zakia Nasri worked entirely under my supervision in 2007-2008. The project’s idea
originated from Professor Jean-Pierre Corriou (previously at INP Mancy). In this research
project, the dynamics and control of a 27-stages multicomponent high-pressure column were
distillation column were compared to those provided by HYSYS©. Mrs. Nasri and I wrote three
peer-reviewed papers: one research paper in the Chemical Engineering of Japan Journal in
2007 and two pedagogical papers (one in Chemical Engineering Education appeared in 2009
and the other in Computer Applications in Engineering Education published was accepted in
2009 and published in 2012). Mrs. Iness Wada and Mrs. Manel Selmi have conducted under
my sole supervision research on the modelling of reactive distillation columns for the
production and decomposition of MTBE as well as the production of methyl acetate and
ethylene glycol. Mrs Wada used gPROMS while Mrs Selmi used MATHEMATICA©. Finally, Mr.
Yosra Messaoudi worked on a thesis project supervised by Professor Mohammed Gargouri and
I. The project dealt with the optimization of the production of bio-ethanol. My contribution to
6
Habilitation Dr. Housam Binous 2015
the work was to guide Mrs. Yosra on how to implement, using MATHEMATICA©, the
separations problems using MATHEMATICA© were given in the summer of 2004 in Istanbul,
Turkey.
In the period 2008-2009, a joint work with Dr. Hasseine and his students, all from the
University of Biskra, Algeria, led to one publication in the International Journal of Materials
Sciences in 2008 and a conference paper in 2009 in the 8th World Congress of Chemical
Engineering, Montréal, Canada. The journal paper was on the study of thermocouples while the
Starting 2010, I have joined King Fahd University of Petroleum & Minerals in Dhahran in
Saudi Arabia with the rank of visiting Associate Professor in the Chemical Engineering
Department. I also left the research group of Professor Naceur Ayed at INSAT and joined the
research team of Professor Ahmed Bellagi at ENIM in order to focus more on chemical
engineering. In the period from 2010 to 2015, I have supervised or co-supervised 3 students
enrolled in the professional Masters sponsored by ARANCO under the joint responsibility of
Institut Français du Pétrole (IFP) and King Fahd University of Petroleum & Minerals
(KFUPM). I have also supervised 2 graduate students in the research Masters’ program at
KFUPM. These two students have both graduated in the Fall of 2015.
I have further expanded my skills in numerical methods during this period (2010-2015). A
peer-reviewed paper on the homotopy continuation method was published in 2010 in the
Chemical Engineering Journal of Japan. This was the fruit of a collaboration with Professor
Brian Higgins from the University of California at Davis, a collaboration which started as early
7
Habilitation Dr. Housam Binous 2015
as 2007. Along the same lines, a conference paper on arc-length continuation was delivered in
Maastricht, Holland in 2008. In 2014, Professor Higgins and I have published a technical paper,
technical magazine called Convertech & e-Print. This paper describes several interactive
programs using MATHEMATICA© that can be useful to initiate students and professional to some
Two papers on the experimental and theoretical study of binary and multi-component batch
and Professor Ahmed Bellagi. The first was published in 2014 and the second in 2015.
Additional work with these two chemical engineering faculties is currently under way. Indeed,
we are investigating binary and ternary mixtures containing acetic acid, a chemical compound,
which dimerizes in the gas-phase as well as a selection of immiscible and partially miscible
multicomponent batch distillation at the annual meeting of the AIChE conference in the autumn
In 2010, I started a collaboration with Professor Abdullah Shaikh. A peer-reviewed paper on steady-
state multiplicity in chemical reactor appeared in 2013 in Chemical Engineering & Technology
as a result of this collaboration. In 2015, Dr. Shaikh presented another joint work in the
Statistics. A paper regarding this work entitled: Bifurcation of Equilibrium Solutions in a Class
of Chemically Reactive Systems has recently been submitted for publication. In 2015,
technique in the KFUPM graduate program. This paper studied several chemical engineering
problems in the fields of distillation (both reactive and non-reactive), thermodynamics and
8
Habilitation Dr. Housam Binous 2015
chemical reaction engineering. Also, in 2015, a peer-reviewed paper, written by Binous, Shaikh
and Bellagi, was published in Computer Applications in Engineering Education Journal. This
Phenomena Problems With Matlab and Mathematica. Another manuscript written in 2015 by
Binous, Kaddech and Bellagi using the same Chebyshev orthogonal collocation technique but
for two-dimensional chemical engineering problem has been accepted for publication in 2015.
Also, recently, in June 2015, a paper written with Professor Bellagi and entitled: Orthogonal
Education Journal. This later paper focus on the solution of various graduate-level chemical
Since 2014, I have started a collaboration with Professor Mozahar Hossain at KFUPM. Dr.
modelling and chemical reaction equilibria calculations using the Gibbs free energy
in Engineering Education Journal in 2015. Authors of this paper, entitled: Haber Process and
Steam-Coal Gasification: Two Standard Thermodynamic Problems Elucidated Using Two Distinct
Approaches, include Ahmed Aheed a graduate student from Sudan who is currently working on
his Masters’ thesis under the direction of Dr. Hossain and I at KFUPM. I also have presented the
preliminary results of the research project of Mr. Aheed at the annual meeting of the AIChE
conference in 2015 in Atlanta, USA. Finally, Mr. Afees, a graduate student has worked also under
our supervision (Dr. Hossain and I), has completed his research thesis on the oxidative
dehydrogenation of propane to propylene and has written a manuscript, which was accepted in
2016 in Catalysis Science and Technology (impact factor 5.426). This manuscript is entitled:
9
Habilitation Dr. Housam Binous 2015
oxygen.
A funded project in underway in collaboration with Dr. Ali Matar from the University of Jordan
and KFUPM. In this project, which started in the Fall of 2014, we first worked (Dr. Al-Matar
and I) on particle swarm optimization. A pedagogical paper has been submitted in October
2015 to Education for Chemical Engineers Journal (Elsevier) on this topic. The ultimate goal
of this project is the determination using molecular simulations of vapor-liquid phase equilibria
diagram for UF6 using simplified temperature dependent interaction potential. Dr. Ali Al-
project (project number: IN131040) which is funded by the Deanship of Scientific Research of
KFUPM. More recently, in March 2016, a paper entitled: Vapor - liquid phase equilibria
diagram for uranium hexafluoride (UF6) using simplified temperature dependent interaction
potential (TDIP), was accepted in the Journal of Radioanalytical and Nuclear Chemistry
In collaboration with Dr. Naim Faqir and Dr. Eid Al-Mutairi from the University of Jordan and
columns have been performed starting January 2011. In addition, the control of binary
distillation column have been studied. A special emphasis to dual column product purity
control was given in this work. A pedagogical peer-reviewed paper has been published in 2014
containing the main results that we have obtained. This paper appeared in Computer
A funded project where Dr. Abdullah Al-Shammari was the principal investigator (an expert
in process control) with team members: Dr. Faqir from the University of Jordan, Dr.
Shamsuzzoha previously at KFUPM (now in Emirates) and I, started in 2011. The outcome of
this project was the publication of a book chapter in Advances in Engineering Research, Volume
10
Habilitation Dr. Housam Binous 2015
8, Nova Publishers, 2014. The subject of this book chapter was distillation column dynamics and
control with special emphasis on model predictive control using the corresponding Matlab ®
Two Book chapters were published by Binous in 2013 and by Binous and Bellagi in 2013. The
first book chapter was about the computation of residue curves and their application in the
conceptual design of distillation systems. This book chapter appeared as chapter 6 of the book
Nova Publishers, 2013. The other chapter, written by Binous and bellagi, appeared in Advances
in Systems Engineering Research, Nova Publishers, 2013. This book chapter was entitled:
Distillation Techniques using Mathematica©. This paper was a rigorous study (non-CMO case
where both the mass and enthalpy balance equations are considered) of standard distillation
problems as well as advanced distillation techniques such as extractive and reactive distillation.
Since last year, I have initiation a collaboration with Dr. Nayef Al-Saifi from KFUPM. In this
research cooperation, we investigated the systematic determination of all the roots of several
forms of the SAFT equation of state (a field of interest of Dr. Nayef) using arc-length
the spring of 2015, Dr. Nayef’s student (Mr. Isa) presented our work in a poster session at
SAFT 2015 at RICE University, USA and Dr. Nayef presented a talk at the 28th European
preparation.
A collaboration between Professor Bellagi, Professor Shaikh and I is underway since the Spring
of 2015. Our goal was to study a laminar flow reactor operating under various conditions:
isothermal, adiabatic, non-isothermal with isothermal cooling jacket, and non-isothermal with
non-isothermal cooling jacket. We have found perfect agreement between the results obtained
11
Habilitation Dr. Housam Binous 2015
with COMSOL™ and with Chebyshev orthogonal collocation and MATHEMATICA©. The
approach using MATHEMATICA© is superior since we can study other situations such as the
determination of the optimal linear or exponential temperature profiles of the cooling jacket.
In December 2015, a collaboration with professor Hasseine from the University of Biskra
(Algeria) started and I have performed particle swarm optimization calculations to determine
binary interaction of the NRTL model. The outcome of this cooperation was a paper published
in Fluid Phase Equilibria in January 2016 (impact factor 2.20). This paper is entitled: Liquid–
liquid equilibrium data for water + formic acid + solvent (butyl acetate, ethyl acetate, and
12
Habilitation Dr. Housam Binous 2015
Abstract
The chapter describes the arc-length continuation method; a mathematical technique that can
solve a wide variety of chemical engineering problems. Several case studies involving
distillation, reaction engineering and thermodynamics are treated with the help of this method.
The approach is powerful enough to elucidate situations when several turning points occur.
Introduction
Natural parameterization of branches using arc-length is one of the main advantages of arc-
length continuation over other techniques. This particular feature of arc-length continuation
allows solution of problems even in the presence of turning points and/or at bifurcations
Several authors have used arc-length continuation to solve important chemical engineering
problems (e.g., problems related to distillation and chemical reaction engineering) since the
pioneering work of Kubíček and Holodniok (1984) on chemical reactors. Indeed, Lin et al.
determine the boiling temperature and composition of azeotropes for any multi-component
mixture. Later, Venimadhavan and co-workers (1994 and 1995), Okasinski and Doherty (1997)
and Chen et al. (2002) have used this very same technique to study reactive distillation. In
particular, Okasinski and Doherty (1997) were interested in the appearance/disappearance and
the composition of reactive azeotropes for several hypothetical mixtures as well as few
13
Habilitation Dr. Housam Binous 2015
industrially relevant mixtures. The majority of the results obtained, applying arc-length
In the chapter, we show how the MATHEMATICA© (version 9.0) commands NDSolve and
WhenEvent can be put into advantage to track solutions of simple and complex systems of
nonlinear algebraic equation in parameter space even when turning points are present. It is our
belief that even though AUTO clearly has superior capabilities than MATHEMATICA© for these
types of problems, learning AUTO requires special training (works on LINUX platform and
uses a Fortran-like syntax) and much more time than available in a one-semester course. The
powerful software AUTO seems more appropriate to an M.S. or a Ph.D. research project rather
than to three-month term projects and/or to quick (e.g., one-hour period) illustrations of few
fundamental points in the classroom or in the computer lab. On the other hand, the approach
proposed in the chapter requires much less effort to master and easily allows instructors to
Several pedagogical chapters have dealt with thermodynamics (Baur et al., 1998; Liu, 2011;
Cruz-Peragón, 2012), separation processes such as distillation (Liu and Peng, 2007; Castrellón
et al., 2011; Granjoet al., 2012; Binous et al., 2014) and chemical reaction engineering (Porto
and Ogeda, 1998; Rodrigues, Soares and Secchi, 2010). However, to our knowledge, none has
The chapter starts with a general introduction of the arc-length continuation method. The
following section shows how various chemical engineering case studies are solved with the
help of this technique. The first test problem presents a new approach to compute vapor-liquid
equilibrium data. The next two examples are related to distillation. Then, two case studies
14
Habilitation Dr. Housam Binous 2015
𝒇(𝒖; 𝛼) = 0 (1)
Now suppose we have found a solution to Eq. (1), say by Newton's method for = 𝛼0 . We
𝒇 (𝒖𝟎 ; 𝛼0 ) = 0 (3)
If we assume that the solution 𝒖is an analytic function of 𝛼, i.e,𝒖 = 𝒖 (𝛼)is continuous and
𝒖 (𝛼), on the solution curve by constructing a Taylor series expansion about 𝒖𝟎 . If the solution
In order to solve Eq. (5), we need an additional equation. The appropriate equation is given
𝑑𝒖 𝑑𝒖 𝑑𝛼 2
. + ( 𝑑𝑠 ) = 1 (6)
𝑑𝑠 𝑑𝑠
Thus, Eqs. (5) and (6) represent 𝑁 + 1differential algebraic equations (DAEs) in terms of the
𝑁 + 1 unknowns. Examples treated in the chapter show that this DAEs can be readily solved
with the built-in MATHEMATICA© command NDSolve even when there are several turning
𝜕𝑓 𝜕𝑓
𝒇(𝒖𝟎 , 𝛼0 ) = 0,(𝜕𝑢) = 0and(𝜕𝛼) ≠0 (7)
𝒖𝟎 ,𝛼0 𝒖𝟎 ,𝛼0
15
Habilitation Dr. Housam Binous 2015
𝑑𝑢
Plainly, the tangent to the curve𝑑𝛼 is not definedat the turning point.
When designing a separation unit, such as distillation or absorption columns, one needs good
vapor-liquid equilibrium data. Such data is either determined experimentally (for new
Raoult’s law (for common mixtures). Consider two ubiquitous binary mixtures: (1) ethanol and
water and (2) ethanol and ethyl acetate. We compute the isobaric vapor-liquid diagram as well
as the equilibrium curve at user-set values of the total pressure 𝑃 (expressed in mmHg). The
modified Raoult's law is used along with the van Laar model for activity coefficient prediction
and Antoine equation for vapor pressure versus temperature relationship. Both systems present
a positive pressure-sensitive azeotrope (i.e., a point where the bubble and dew curves intersect
and where the tangent to these two curves is horizontal). When present, this azeotrope is
indicated on the equilibrium curve by a red dot (see Figure 1a and 1b for 𝑃 = 760𝑚𝑚𝐻𝑔).
The loci of the azeotrope versus pressure 𝑃 are given in a separate plot (see Figure 2a and 2b).
One particular feature of the present calculation is that it uses the arc-length continuation
method to find the bubble/dew point temperatures versus liquid/vapor phase compositions of
ethanol. Even though there are no turning points for this particular problem, we take advantage
the azeotropes; indeed they verify 𝑇′(𝑠) = 0, where 𝑠 is the arc-length parameter.
action that occurs when the event triggers it for equations in NDSolve.
16
Habilitation Dr. Housam Binous 2015
17
Habilitation Dr. Housam Binous 2015
Plainly, for vapor-liquid equilibrium data computations, the nonlinear equation 𝑓(𝑇, 𝑥) = 0,
where 𝑥 is the liquid mole fraction of ethanol and 𝑇 is the bubble temperature, isgiven by:
where 𝑃𝑖𝑠𝑎𝑡 (𝑇) and 𝛾𝑖 (𝑥, 𝑇) are the vapor pressure and the activity coefficient of component 𝑖,
mixture). This equation is the well-known bubble point equation derived from Dalton's law and
the modified Raoult's law. We introduce an arc-length parameter 𝑠. The nonlinear algebraic
equation becomes 𝑓(𝑇(𝑠), 𝑥(𝑠)) = 0. We use the built-in MATHEMATICA© function NDSolve
to solve this equation together with the differential equation (called the arc-length constraint)
𝑇′ (𝑠)2 + 𝑥′ (𝑠)2 = 1 in order to find 𝑥(𝑠) and 𝑇(𝑠). A simple consistent initial condition is
found by taking 𝑥(0) = 1 and 𝑇(0) equal to the boiling temperature of pure ethanol at 𝑃.
Finally, 𝑦(𝑠) (i.e., the vapor-phase mole fraction of ethanol) is obtained from the Modified
Raoult’s law:
𝑦 𝑃 = 𝑃1𝑠𝑎𝑡 𝛾1 𝑥 (9).
Distillation is the most widely used separation technique in the chemical industry. It is one the
most important topics covered in the Chemical Engineering curriculum in most graduate and
undergraduate programs.
Consider an ideal equimolar binary mixture of benzene and toluene at 760 mmHg. This mixture
is fed to a 10-stage column with a total condenser and partial reboiler. The feed stage is stage
7. Column stages are counted from the top. The feed flow rate is taken equal to 10 kmol/hr.
The values of the reflux ratio,𝑅, and the reboil ratio,𝑆, are chosen equal to 3 and 2.5,
respectively. In Figure 3, the top (in dashed blue) and bottom (in dashed red) product purities
are given versus the thermal quality of two-phase (0 < 𝑞 < 1), superheated (𝑞 < 0) and
18
Habilitation Dr. Housam Binous 2015
19
Habilitation Dr. Housam Binous 2015
The code uses a rigorous approach, which takes into account both energy and mass balances
(i.e., the so-called MESH equations). For the governing equations, the reader is directed to
Henley and Seader (1981). Again, the computation is based on the arc-length continuation
technique but the number of DAEs is of course much larger (i.e., all MESH equations plus the
auxiliary equation; a total of 66 equations). In this case, vector 𝒖 contains: (1) all temperatures
and liquid and vapor mole fractions of benzene in all stages of the column, (2) liquid and vapor
flow rates in all the column including distillate and bottom streams(3) the feed stream quality,
temperature and compositions of each phase of the feed stream (when 0 < 𝑞 < 1)and (4) the
reboiler and condenser heat duties. In addition, parameter 𝛼is 𝑞: the feed thermal quality. A
relatively tedious approach would be to solve a system of algebraic equations using for example
a built-in MATHEMATICA© command such as FindRoot for a particular value of 𝑞. This latest
method would oblige us to have, for each particular value of 𝑞, a good initial guess for the
equation solver: FindRoot. Here, the advantage of the arc-length continuation approach is
that we find all column profiles for all values of 𝑞 with just a single run of our code. In addition,
NDSolve uses a DAEs solver that can compute, for most problems, a consistent initial
condition automatically. Figure 4 depicts the liquid-phase composition profiles. The blue and
All results found here show good agreement with those given by HYSYS ®
(http://www.aspentech.com/hysys/). This is particularly true despite the fact that the authors
use a slightly simplified version of the liquid-phase and vapor-phase enthalpies versus
temperature.
Indeed, the distillation column temperature profiles for this calculation (the open blue circles)
and HYSYS® result (the filled red dots) are compared(see Figure 5), for 𝑞 = −0.2, 𝑅 = 3,
and𝑆 = 2.5.
20
Habilitation Dr. Housam Binous 2015
Figure 3 - Top (in dashed blue) and bottom (in dashed red) product purities versus the
thermal quality
21
Habilitation Dr. Housam Binous 2015
Reactive distillation has been adopted as the industry standard for the production of bulk and
such technique, that combined chemical reaction and distillation, over conventional processes
are numerous and range from huge process simplification to near complete conversions of
Consider a mixture of four components 𝐴, 𝐵, 𝐶, and 𝐷 with relative volatilities 𝛼𝐶𝐷 = 7, 𝛼𝐴𝐷 =
𝐶 + 𝐷 with reaction rate = 𝑘 (𝑥𝐴 𝑥𝐵 − (𝑥𝐶 𝑥𝐷 )/𝐾𝑒𝑞 ) where 𝑘 is the forward reaction rate
constant and 𝐾𝑒𝑞 is the reaction equilibrium constant. This mixture is fed to a reactive
distillation column with a total number of theoretical plates equal to 17. The feed stage location
is stage number 10, the reactive stages are from stage 6 to stage 14, and the feed composition
is equimolar in 𝐴 and 𝐵(i.e. the feed is composed of 50 mole % 𝐴 and 50 mole % 𝐵).
The feed, a saturated liquid, has a flow rate of 100 kmol/hr. For simplicity, constant molal
overflow (CMO) is assumed and heat effects are neglected. Governing equations can be found
in the review chapter by Taylor and R. Krishna (2000). Again, the computation is based on the
arc-length continuation technique and the number of DAEs is 60. In this case, vector 𝒖
contains: (1) liquid mole fractions in all stages of the column for the three components: 𝐴, 𝐵
and𝐶, (2) the liquid and vapor molar flow rates in the different sections of the column including
distillate and bottom streams. In addition, parameter 𝛼 is 𝐷𝑎: the Damköhler number. This
(𝐻/𝐿)
dimensionless number is defined by 𝐷𝑎 = , where 𝐻 is the liquid hold-up in a plate, 𝑘 is
(1/𝑘)
the forward reaction rate constant, and 𝐿 is the liquid flow rate in each section of the
column.One could solve a system of algebraic equations using for example a built-in
MATHEMATICA© command such as FindRoot for a particular value of 𝐷𝑎. The arc-length
22
Habilitation Dr. Housam Binous 2015
continuation approach is superior since the column profile is obtained for all values of 𝐷𝑎 with
just a single run of our code. In addition, NDSolve uses a DAEs solver that can find, for the
majority of problems, a consistent initial condition automatically. Figure 6 shows the purities
of the distillate and bottom (i.e., mole fractions of 𝐶(in blue) at the top of the column and 𝐷 (in
red) at the bottom of the column) versus the Damköhler number. Figure 7 shows the liquid-
phase compositions versus plate number for components𝐴, 𝐵, 𝐶, and 𝐷 (in red, blue, green,
and magenta, respectively) for 𝐷𝑎 = 5. The reactive zone is shown in light blue, while the
stripping and rectifying zones are indicated in light red and light green, respectively.
Figure 5 -Temperature profile for 𝑞 = −0.2, 𝑅 = 3, and 𝑠 = 2.5 (this calculation: the open
23
Habilitation Dr. Housam Binous 2015
It is clear from Figure 7 that this reactive distillation setup can achieve conversions near 100%
Consider a cascade of two continuous stirred tank reactors with recycle with exothermic first-
Figure 6 - Purities of the distillate (i.e., mole fractions of 𝐶(in solid blue) at the top of the
column) and bottom (i.e., mole fraction of𝐷 (in dashed red) at the bottom of the column)
24
Habilitation Dr. Housam Binous 2015
The steady state is described by 4 nonlinear equations for the reaction conversions {𝑋1 , 𝑋2 } and
dimensionless temperature {𝑇1 , 𝑇2 }where the parameter 𝐷𝑎 is the Damköhler number. The
dimensionless equations that describe the steady state are given by Kubíček and Marek (1984),
25
Habilitation Dr. Housam Binous 2015
𝑇1
𝑓1 (𝑋1 , 𝑋2 , 𝑇1 , 𝑇2 ) = 𝐷𝑎 𝑒 1+0.001 𝑇1 (1 − 𝑋1 ) − 𝑋1 = 0 (10)
𝑇1
𝑓2 (𝑋1 , 𝑋2 , 𝑇1 , 𝑇2 ) = 22 𝐷𝑎𝑒 1+0.001 𝑇1 (1 − 𝑋1 ) − 3 𝑇1 = 0 (11)
𝑇2
𝑓3 (𝑋1 , 𝑋2 , 𝑇1 , 𝑇2 ) = 𝑋1 + 𝐷𝑎 𝑒 1+0.001 𝑇2 (1 − 𝑋2 ) − 𝑋2 = 0 (12)
𝑇2
𝑓4 (𝑋1 , 𝑋2 , 𝑇1 , 𝑇2 ) = 22 𝐷𝑎 𝑒 1+0.001 𝑇2 (1 − 𝑋2 ) + 𝑇1 − 3 𝑇2 = 0 (13)
The derivation of the above equations (which represent the species and energy balances for
each tank) is given by Kubíček et al (1980). A steady solution for 𝐷𝑎 = 0.01 is 𝑋1 (0.01) =
Figure 8 shows the dependence of 𝑇2 on the parameter 𝐷𝑎. Though not shown, one can readily
plot the dependence of the remaining variables 𝑋1 , 𝑋2 and 𝑇1 on 𝐷𝑎. The solution path has 6
turning points (indicated in green in Figure 8). For 𝐷𝑎 = 0.039 there are 7 steady state
solutions! (Indicated in blue in Figure 8). The results are in quantitative agreement with
Kubíček and Marek's calculations (see Table 1). The method is based also on arc-length
continuation with 5 DAEs. Both the steady states (for a particular value of 𝐷𝑎) and the turning
points are readily obtained using the built-in MATHEMATICA© command WhenEvent. Not all
these steady state solutions are stable to small perturbations. The stability of the steady states
is determined by the sign of the real part of the eigenvalues of the Jacobian matrix (Varma and
Morbidelli, 1997).
Internal transport processes (i.e., the effectiveness factor) are covered in great detail in
Heterogeneous Catalysis.
26
Habilitation Dr. Housam Binous 2015
The governing equations (Beers, 2007), obtained from mole and enthalpy balances, and the
boundary conditions, obtained from the symmetry condition and the known values of
𝑑 𝑑𝑐A
(𝑟 2 𝐷A ) − 𝑟 2 𝑘𝑐A = 0 (15)
𝑑𝑟 𝑑𝑟
𝑑 𝑑𝑇
(𝑟 2 𝜆 𝑑𝑟 ) + 𝑟 2 (−𝛥𝐻)𝑘𝑐A = 0, (16)
𝑑𝑟
𝑑𝑐A 𝑑𝑇
(𝑟 = 0) = 0 and (𝑟 = 0) = 0, (17)
𝑑𝑟 𝑑𝑟
where 𝑅 is the radius of the spherical pellet, 𝜆 is the effective thermal conductivity of the
pellet, 𝑘 is the reaction rate constant, 𝐷𝐴 is the effective binary diffusivity of 𝐴 within the
27
Habilitation Dr. Housam Binous 2015
radial position, and 𝑪𝑨𝒔 and 𝑇𝑆 are the concentration and temperature at the surface of the
𝑑 𝑑𝜙A 𝛾𝛽(1−𝜙 )
[𝜉 2 ] − 𝜉 2 𝛷2 exp[1+𝛽(1−𝜙A )]𝜙A = 0, (20)
𝑑𝜉 𝑑𝜉 A
𝑟 𝑪𝐀 (𝒓)
with 𝜙A (1) = 1,𝜙A ′(0) = 0, 𝜉 = 𝑅 and 𝜙A = .
𝐂𝐀𝐒
𝑘(𝑇S ) 𝐷A 𝛥𝐻𝑪𝐀𝐒 𝐸
Φ = 𝑅√ ,𝛽 =− , and 𝛾 = 𝑅𝑇𝑎 , where Φ, the Thiele modulus, is a measure of
𝐷A 𝜆𝑇S s
internal mass transfer resistance. WhenΦ ≪ 1, mass transfer resistance is negligible and
diffusion is very fast. Large values of 𝛾, the dimensionless activation energy, mean that the
reaction rate is very sensitive to temperature. Typical values of 𝛾 are 1, 5, 10, 15, or 30. Figure
28
Habilitation Dr. Housam Binous 2015
9 displays the effectiveness factor (red curve) 𝜂, a measure of the total reaction rate inside the
pellet compared to its value at the pellet surface, versus the Thiele modulus Φ, for various
values of parameter 𝛽. It is clear that the curve has two turning points, which lead to steady
where
1
𝛾𝛽(1−𝜙 (𝜉))
𝜂 = 3 ∫ 𝜙A (𝜉)exp[1+𝛽(1−𝜙A (𝜉))]𝜉 2 𝑑𝜉. (22)
A
0
The solution is based on the arc-length continuation method and the Chebyshev orthogonal
42. In this case, vector 𝒖 contains: all the dimensionless compositions of species 𝐴at the 41
collocation points. In addition, parameter 𝛼 is Φ: the Thiele modulus. For large values of Φ
and for 𝛽 > 0.24, our method performs poorly and we use the fact that the effectiveness factor
𝜂 is inversely proportional to the Thiele modulus Φ (see blue dashed line) (Petersen, 1962).
The parameter, 𝛽, which you can set, is a measure of the relative importance of the heat of
reaction to conduction. Large values of 𝛽 mean that there is significant internal heating; thus
for Φ ≈ 1 the effectiveness factor will be greater than 1. This means that the rate of reaction is
higher inside the pellet compared to its value at the surface of the pellet in the absence of
reactions (Δ𝐻 > 0) and smaller values of the effectiveness factor 𝜂. When Φ ≫ 1, the internal
transport resistance becomes large (a depleted zone inside the pellet forms and the
concentration of species 𝐴 is low). Thus, the reaction rate inside the pellet is smaller than its
value at the pellet surface and consequently the effectiveness factor becomes very small
29
Habilitation Dr. Housam Binous 2015
shown in Figures 10a and 10b, up to three solutions (shown in blue, magenta and brown) can
Figure 9 - Effectiveness factor 𝜂versus the Thiele modulus Φ, for 𝛽 = 0,6 and 𝛾 = 30.
Consider the following nonlinear chemical reaction scheme described by Schlögl (1972):
𝐴 + 2𝑋 ⇆ 2𝑋+𝑌 (23)
𝑋 ⇆ 𝐵 (24)
Eq. 23 is the autocatalytic step and concentrations of the species 𝐴 and 𝐵 are assumed constant.
One can show that this nonlinear chemical system obeys the nondimensionless equation:
𝑑𝑥
= −𝑥 3 + 3 𝑥 2 − 𝜆 𝑥 + 𝜇 , (25)
𝑑𝑡
𝑑𝑥
Figure 11 shows the steady-state values, obtained when = 0, versus μ for λ = 1. As can be
𝑑𝑡
30
Habilitation Dr. Housam Binous 2015
The two turning points are computed and indicated by the green dots. These turning points are
obtained with the help of the built-in MATHEMATICA© command WhenEvent. For some
31
Habilitation Dr. Housam Binous 2015
values of μ there are three steady states. A stability analysis can show that only two of these
are stable. The dotted curve in Figure 11 is the loci of unstable steady-states.
Solution of systems of nonlinear equations are covered in any required graduate-level course
on numerical methods. Such equations arise naturally when one studies the steady-state
𝑑𝑥
= 𝑓1 (𝑥, 𝑦) = 𝑥 − 𝑦 𝑐𝑜𝑠(𝑦) 𝑧1 = 0 and (26)
𝑑𝑡
𝑑𝑦
= 𝑓2 (𝑥, 𝑦) = −𝑦 + 𝑥 𝑠𝑖𝑛(𝑥) + 𝑧2 = 0, (27)
𝑑𝑡
The arc-length continuation method combined with Rahimian’s approach (2011) allows us to
find all steady-states of this two-dimensional dynamical system in the domain [−12,12] ×
32
Habilitation Dr. Housam Binous 2015
[−12,12]. The roots loci are indicated in Figure 12 by the red dots. The problem considered is
described as follows: 𝐻(𝑡(𝑠), 𝑥(𝑠), 𝑦(𝑠)) = 1 − 𝑡(𝑠) + 𝑓1(𝑥(𝑠), 𝑦(𝑠)) = 0 (the function
𝐻(𝑡, 𝑥, 𝑦) was first proposed by Rahimian et al. (2011)), 𝑓2(𝑥(𝑠), 𝑦(𝑠)) = 0, and (𝑡′(𝑠))2 +
(𝑥′(𝑠))2 + (𝑦′(𝑠))2 = 1 (i.e., the auxiliary equation). Using the built-in MATHEMATICA©
function WhenEvent all roots of the system are readily obtained when 𝑡(𝑠) = 1. A list of all
Semenov developed a model that describes the thermal ignition phenomenon quantitatively
(Nicolis, 2005). Assume that the temperature, 𝑇, of the reacting system is constant and uniform
across the volume of the system and assume that the walls of the system and the ambient
The steady-state temperature is obtained by setting the heat production term (i.e., the left-hand
side of the equation) equal to the heat losses term, which is mainly due to convection between
the system and the surroundings (i.e., the right-hand side of the equation):
−1
𝑒 𝜖𝑇 = γ (T − 1) (28)
where 𝑇 is a dimensionless temperature, 𝛾 depends on the overall heat transfer coefficient, and
The system presents multiple steady states, as you can see from the bifurcation diagram (see
Figure 14). Figure 15 displays the heat losses (shown in magenta) and the heat production term
(shown in blue); it is clear that three steady states (i.e., three intersection points identified by
The bifurcation diagram is obtained using two different techniques: (1) using the arc length
continuation method; and (2) using the MATHEMATICA© built-in command ContourPlot.
Perfect agreement is obtained using these two methods. Finally, turning points are shown by
33
Habilitation Dr. Housam Binous 2015
The Frank–Kamenetskii problem relates to the self-heating of a reactive solid (Higgin and
Binous, 2010). When the heat generated by a reaction is balanced by conduction in a one-
dimensional slab of combustible material, the nonlinear boundary value problem (BVP):
admits up to two solutions. Here, 𝑢 is the dimensionless temperature and ℎ is the heat transfer
coefficient.
34
Habilitation Dr. Housam Binous 2015
35
Habilitation Dr. Housam Binous 2015
Figure 15 – The heat losses (shown in magenta) and the heat production term (shown in blue)
1/2) 𝜃/2]/𝑐𝑜𝑠ℎ(𝜃/4)), where 𝜃 is one of the two solutions of the transcendental equation
We use the arc length continuation method and the Chebyshev orthogonal collocation
technique (with 𝑁 + 1 = 21 collocation points) to track the solutions, 𝑢(𝑥), in the 𝛼 parameter
space.
36
Habilitation Dr. Housam Binous 2015
𝑁+1
The plot of the norm of the solution ‖𝑢‖ = √∑𝑖=1 𝑢(𝑥𝑖 )2 versus 𝛼 clearly indicates that there
can be up to two solutions. This plot is shown in Figure 16. These two solutions are plotted in
𝑁+1
Figure 16 – The norm of the solution ‖𝑢‖ = √∑𝑖=1 𝑢(𝑥𝑖 )2 versus 𝛼
37
Habilitation Dr. Housam Binous 2015
Seader's function admits multiple real roots (up to 14 roots for 𝑢1 = 0.99999) while Bessel's
Here, using Seader's approach (Rahimian et al., 2011) and the arc length continuation
Figure 17 – The two solutions (in blue and magenta) for 𝛼 = 2.5
The problem considered is described as follows: 𝐻(𝑡(𝑠), 𝑥(𝑠)) = 1 − 𝑡(𝑠) + 𝑓(𝑥(𝑠)) = 0 (the
function 𝐻(𝑡, 𝑥) was first proposed in [1]) and (𝑡′ (𝑠))2 + (𝑥′ (𝑠))2 = 1 (i.e., the auxiliary
equation). Using the built-in MATHEMATICA© function WhenEvent, all roots of 𝑓(𝑥) are
A list of all roots is provided for both test functions (see tables 3 and 4). When you compare
the present approach for the Bessel function of the first kind with the built-in MATHEMATICA©
The Peng–Robinson equation of state is used to compute the correction factor due to gas-phase
Haber process.
𝜙NH3
1.5 𝜙 0.5 , where 𝜙NH3 , 𝜙N2 , and 𝜙H2 are the fugacity coefficients of the species involved; they
𝜙H 2 N2
are calculated using the Peng–Robinson equation of state. In addition, the extent of reaction is
computed. Figure 20 and 21 shows these two quantities versus pressure for temperatures
39
Habilitation Dr. Housam Binous 2015
The extent of reaction is high and approaches one at very high pressure and low temperature,
as expected from Le Chatelier's principle. 𝐾𝑣 is close to one at low pressure and is considerably
smaller than one for high pressure, which indicates that this correction factor must be taken
into account at high pressure. Here, the method of arc-length continuation is applied to solve
Conclusion
Arc-length continuation has been used to readily solve several nontrivial chemical engineering
41
Habilitation Dr. Housam Binous 2015
42
Habilitation Dr. Housam Binous 2015
As illustrated in the study cases 𝐸 and 𝐼 (i.e., the Weisz and Hicks Problem and the Frank–
collocation can help almost any graduate student and/or researcher solve problems that may
References
[1] R. Baur, J. Bailey, B.Brol, A.Tatusko and R. Taylor, “Maple and the art of
[3] B. W. Bequette, Process Dynamics, Modeling, Analysis, and Simulation, Upper Saddle
[4] H. Binous, E. Al-Mutairi and N. Faqir, “Study of the separation of simple binary and ternary
2014 pp.87-98.
[5]T. Castrellón, D. C. Botía, R. Gómez, G. Orozco and I. D. Gil, “Using process simulators in
the study, design, and control of distillation columns for undergraduate chemical engineering
[6] F. Chen, R. S. Huss, M. F. Doherty and M. F. Malone, “Multiple steady states in reactive
distillation: kinetic effects, ”Computers and Chemical Engineering,26(1), 2002 pp. 81–93.
[8] E. Doedel, AUTO: Software for Continuation and Bifurcation Problemsin Ordinary
43
Habilitation Dr. Housam Binous 2015
[12] B. G. Higgins and H. Binous, "A Simple Method for Tracking Turning Points in Parameter
Applications of Bifurcation Theory, P. H. Rabinowitz, ed., New York: Academic Press, 1977.
[15] M. Kubíček and M. Marek, Computational Methods in Bifurcation Theory and Dissipative
equations describing heat and mass transfer in chemical reactors,” Chemical Engineering
[18] S. X. Liu and M. Peng, “The simulation of the simple batch distillation of multiple-
44
Habilitation Dr. Housam Binous 2015
[20] M. F. Malone and M. F. Doherty, “Reactive Distillation,” Ind. Eng. Chem. Res., 39(11),
[24] L. M. Porto and R. H. Ogeda, “Java Applets for Chemical Reaction Engineering,”
[25] S. K. Rahimian, F. Jalali, J. D. Seader, and R. E. White, "A New Homotopy for Seeking
all Real Roots of a Nonlinear Equation," Computers and Chemical Engineering, 35(3), 2011
pp. 403–411.
pp.607-618.
[28] R. Seydel, Practical Bifurcation and Stability Analysis- From Equilibrium to Chaos, 2nd
45
Habilitation Dr. Housam Binous 2015
Residue Curve Maps for Reactive Distillation,” AIChE J., 40(11), 1994 pp. 1814-1824.
[33] P. B. Weisz and J. S. Hicks, “The Behaviour of Porous Catalyst Particles in View of
Internal Mass and Heat Diffusion Effects,” Chemical Engineering Science, 17(4), 1962 pp.
265–275.
46
Habilitation Dr. Housam Binous 2015
Abstract
We present in this chapter several important aspects of the application of orthogonal collocation
The presented method of resolution (i.e., orthogonal collocation) of such problems can be
research problem.
Introduction
A quick search in Scopus for all papers having orthogonal collocation in title–abstract–key-
words entry is given by the red curve in Figure 1. Although, there is in recent years on average
the blue curve in Figure 1. Thus, the present chapter’s objective is to modestly increase the
awareness about the potential use of this technique to solve complex chemical engineering
problems.
Several pedagogical papers dealt with chemical engineering problems using many
mathematical and spreadsheet softwares such as Maple, Matlab®, MATHEMATICA©, and Excel.
To cite only a few, papers in fields such as (1) thermodynamics [1-3], (2) separations [4-8], (3)
chemical reaction engineering [9, 10], and (4) numerical methods [11, 12] have been published
47
Habilitation Dr. Housam Binous 2015
in the recent years. Nevertheless, none of them uses orthogonal collocation except Ref. 12,
which deals with the special case of the Chebyshev polynomials. We hope that the present
chapter will fill at least partly this lack of literature on the subject, which hinders the
chemical engineering papers are shown by the red and blue curves, respectively.
The Chapter structure is a follows. First, we start by giving a few theoretical considerations
about orthogonal collocation mainly following the description of Ramirez [13]. Then, we
thoroughly illustrate the technique for two simple problems: (1) a symmetrical problem and (2)
48
Habilitation Dr. Housam Binous 2015
engineering cases studies. All of these later can be solved using orthogonal collocation. A list
Finally, we conclude the chapter by sharing our vision on how this versatile numerical
technique could be introduced in the chemical engineering graduate-level curriculum and serve
The method of orthogonal collocation uses the roots of orthogonal polynomials as collocation
The two polynomial forms, considered in the present paper, are: (1) the shifted Legendre
polynomials and (2) the Jacobi polynomials. The shifted Legendre polynomials are applied
when the problem has no symmetry. On the other hand, the Jacobi polynomials are used for
problems with symmetry. Here, we take the weighting function, 𝑤(𝑥), to be equal to 1 for the
shifted Legendre polynomials case and 1 − 𝑥 2 for the Jacobi polynomials case.
Following the description of Ramirez [13], let us take a trial function written as:
𝑦(𝑥) = ∑𝑁𝑝+1
𝑖=0 𝑑𝑖 𝑥
𝑖−1
(1)
49
Habilitation Dr. Housam Binous 2015
𝑑𝒚 𝑑2 𝒚
= 𝐴 𝒚 and =𝐵𝒚 (5)
𝑑𝑥 𝑑𝑥 2
where 𝐴 = 𝐶 𝑄 −1 and 𝐵 = 𝐷 𝑄 −1 . The square matrixes, of order 𝑁𝑝+2, 𝐶, 𝐷 and 𝑄 are defined
𝑦(𝑥) = ∑𝑁𝑝+1
𝑖=1 𝑑𝑖 𝑥
2𝑖−2
(6)
𝑁𝑝+1
𝑦′(𝑥𝑗 ) = ∑𝑖=1 𝑑𝑖 (2𝑖 − 2) 𝑥𝑗2𝑖−3 , and (8)
𝑑𝒚 𝑑2 𝒚
= 𝐴 𝒚 and 𝑑𝑥 2 = 𝐵 𝒚 (10)
𝑑𝑥
defined as follows: 𝑄𝑖𝑗 = 𝑥𝑗2𝑖−2 , 𝐶𝑖𝑗 = (2𝑖 − 2) 𝑥𝑗2𝑖−3 and 𝐷𝑖𝑗 = (2𝑖 − 2) (2𝑖 − 3)𝑥𝑗2𝑖−4.
50
Habilitation Dr. Housam Binous 2015
a- A symmetrical problem
𝑑2 𝑦 𝑑𝑦 2
+ (𝑑𝑥 ) = 1, 0 ≤ 𝑥 ≤ 1 (11)
𝑑𝑥 2
𝑑𝑦
with (𝑑𝑥 ) = 0 and 𝑦(𝑥 = 1) = 2. Let us consider the roots of the Jacobi Polynomial of
𝑥=0
𝑑𝒚
order 𝑁𝑝 = 5 (see Table 1). From the section above, the following relations apply: =𝐴𝒚
𝑑𝑥
𝑑2 𝒚
and = 𝐵 𝒚 where matrices 𝐴 and 𝐵 are given in Appendix 1. Now it is possible to find the
𝑑𝑥 2
vector 𝒚 = (𝑦1 , 𝑦2 , 𝑦3 , 𝑦4 , 𝑦5 , 𝑦6 ), which contains the values of the function 𝑦(𝑥) at the
collocation points and at the boundary 𝑥 = 1. A simple degree of freedom analysis indicate
that we have six unknowns and six equations. One of the equations is the obvious boundary
condition 𝑦6 = 𝑦(𝑥 = 1) = 2. Thus, one has to solve the following system of nonlinear
Np+1 Np+1 2
∑ 𝐵𝑗,𝑘 𝑦𝑘 + (∑ 𝐴𝑗,𝑘 𝑦𝑘 ) = 1, and (12)
𝑘=1 𝑘=1
𝑦6 = 2 (13)
Figure 2 shows that perfect agreement is obtained between the orthogonal collocation
technique (see the red squares in Figure 2 and Table 1 for the corresponding values of the
component of vector 𝒚) and the exact solution (shown as a blue curve in Figure 2) obtained
51
Habilitation Dr. Housam Binous 2015
𝒙𝒊 𝒚𝒊
0.136553 1.57551
0.399531 1.64399
0.632876 1.75439
0.819279 1.86988
0.944899 1.95869
52
Habilitation Dr. Housam Binous 2015
b- A non-symmetrical problem
𝜕𝑇 𝜕2 𝑇
= 𝜕𝑥 2 , 0 ≤ 𝑥 ≤ 1 (14)
𝜕𝑡
with initial condition and boundary conditions given by: 𝑇(𝑥, 𝑡 = 0) = 1 and 𝑇(𝑥 = 0, 𝑡) =
𝑇(𝑥 = 1, 𝑡) = 0. This problem describes the transient heat conduction in a bar. Let us consider
the roots of the shifted Legendre Polynomial of order 𝑁𝑝 = 5 (see Table 2). As described
𝑑2 𝑻
above, in the section on theoretical considerations, the following relation applies: =𝐵𝑻
𝑑𝑥 2
where matrix 𝐵 is given in Appendix 2. Now, it is possible to find the vector 𝑻(𝑡) =
(𝑇1 (𝑡), 𝑇2 (𝑡), 𝑇3 (𝑡), 𝑇4 (𝑡), 𝑇5 (𝑡), 𝑇6 (𝑡), 𝑇7 (𝑡) ), which contains the values of the function 𝑇(𝑥)
at the collocation points and at both boundaries (i.e., at 𝑥 = 0 and 𝑥 = 1). A simple degree of
freedom analysis indicate that we have seven unknowns functions and seven first-order
differential equations. Two of the equations are the obvious from the boundary conditions:
𝑑𝑇1 𝑑𝑇7
= 0 𝑎𝑛𝑑 = 0 . Thus, one has to solve the following system of linear coupled first-order
𝑑𝑡 𝑑𝑡
differential equations again using the built-in function of MATHEMATICA© called NDSolve:
Np+2
𝑑𝑇𝑗
=∑ 𝐵𝑗,𝑘 𝑇𝑘 , for 𝑗 = 2. . 𝑁𝑝 + 1, and (15)
𝑑𝑡 𝑘=1
𝑑𝑇1 𝑑𝑇𝑁𝑝+2
= 0 𝑎𝑛𝑑 = 0. (16)
𝑑𝑡 𝑑𝑡
𝑇𝑁𝑝+2 (𝑡 = 0) = 0.
Figure 3 shows that perfect agreement is obtained between the orthogonal collocation
technique (see the red dots in Figure 3 and Table 2 for the corresponding values of the
53
Habilitation Dr. Housam Binous 2015
component of vector 𝑻) and the exact solution (shown as a blue curve in Figure 3) obtained
using NDSolve. Here, the above comparison is performed for a value of time equal to 0.1.
𝒙𝒊 𝑻𝒊
0.0469101 0.069715
0.230765 0.314738
0.5 0.474451
0.769235 0.314738
0.95309 0.069715
54
Habilitation Dr. Housam Binous 2015
Consider acetone and methanol diffusing concurrently through a stagnant air column (see
The mole fractions of all three components are obtained by solving the species balance equation
augmented with the Stefan–Maxwell equations for the diffusion fluxes [14, 15]. The Stefan–
𝑑𝑦𝑖 𝑦𝑖 𝑁𝑗 −𝑦𝑗 𝑁𝑖
𝑑𝑧
=∑ 𝐶𝐷𝑖,𝑗
for 𝑖 = 1, 2, … , 𝑛, (17)
𝑗≠𝑖
where 𝑁𝑖 is the flux of species 𝑖 (expressed in 𝑚𝑜𝑙/(𝑚2 𝑠𝑒𝑐)). Binary diffusivities are equal
to 𝐷13 = 0.1372, 𝐷23 = 0.1991, and 𝐷12 = 0.0848 (all expressed in cm2/sec). The resulting
55
Habilitation Dr. Housam Binous 2015
nonlinear boundary value problem can be solved using either a shooting method or by
implementing an orthogonal collocation technique. In this present work, we use both methods.
Figure 5 displays the steady state concentration profiles of the three species in the diffusion
column. The solid curves were obtained with a shooting technique using the terminal
and 𝑦𝑚𝑒𝑡ℎ𝑎𝑛𝑜𝑙 (𝑧 = 1) = 0 and NDSolve. The blue, orange and green colors correspond to
acetone, methanol, and air, respectively. The colored dots shows the solution obtained using
the orthogonal collocation technique for user-set values for the number of interior collocation
points.
Figure 5 – Composition profile versus dimensionless vertical position (acetone, methanol and
air are shown in blue, orange and green; respectively)
Here, the roots of the shifted Legendre polynomials have to be employed since the problem is
56
Habilitation Dr. Housam Binous 2015
methods. From the calculation, we find that Nacetone=0.001768 and Nmethanol=0.003102, while
To quantify the accuracy of the collocation method, we compute the Euclidean norm of the
3
𝑁𝑝
error for 𝑁𝑝 collocation points: 𝜖 = √∑ (∑𝑖=1 |𝑦𝑗 (𝑧𝑖 ) − 𝑦𝑗,𝑒 (𝑧𝑖 )|2 ), where 𝑦𝑗 and 𝑦𝑗,𝑒 are the
𝑗=1
mole fraction of a species 𝑗 with 𝑗 = 1, 2, and 3 obtained with orthogonal collocation and
NDSolve, respectively. This error is displayed for any value of 𝑁𝑝. As the number of
collocation points is increased, the Euclidean norm is reduced. For 𝑁𝑝 = 15, 𝜖 ∼ 𝑂(10−7 ).
In the second case, we describe the start-up of Poiseuille flow in a Newtonian fluid. The
governing equation in dimensionless form for Poiseuille flow for a Newtonian fluid in a tube
is given by:
𝜕𝑣 1 𝜕 𝜕𝑣
= 4 + 𝜉 𝜕𝜉 (𝜉 𝜕𝜉 ), (18)
𝜕𝜏
where the dimensionless time is 𝜏 = (𝑡 𝜈)/𝑅 2 , the dimensionless radial position is 𝜉 = 𝑟/𝑅,
and the dimensionless velocity is 𝑣 = 𝑉𝑧 (𝑟)/𝑉𝑧,𝑚𝑎𝑥 . The boundary and initial conditions are:
The kinematic viscosity is 𝜈 = 𝜇/𝜌 where 𝜇 is the fluid's viscosity and 𝜌 its density, the radius
of the pipe is 𝑅, and 𝑉𝑧,𝑚𝑎𝑥 = −(1/(4 𝜇)) 𝜕𝑝/𝜕𝑧 𝑅 2 is the velocity at the center of the pipe at
steady state.
The velocity is determined using either NDSolve (solid colored curves) or orthogonal
collocation (colored dots). Figure 6 displays the dimensionless velocity profile versus the
0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.6, and 0.8). The code allows users to change the
number and location of interior points used in the orthogonal collocation method. The
collocation points are the roots of the Jacobi polynomials. Indeed, the problem, at hand here,
57
Habilitation Dr. Housam Binous 2015
is symmetrical. The black curve shows the typical parabolic velocity profile obtained for
(𝜏 = 0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.6, and 0.8)
To quantify the accuracy of the collocation method, we compute the Euclidean norm ϵ of the
𝑁𝑝
error for 𝑁𝑝 collocation points, given by 𝜖 2 = ∑𝑖=1 |𝑣(𝜏, 𝜉𝑖 ) − 𝑣𝑒 (𝜏, 𝜉𝑖 )|2 where 𝑣𝑒 is the
solution obtained using NDSolve. This error is plotted as a function of time on the error plot
(see Figure 7). The red dots denote the times used in velocity versus radial position plot. As
58
Habilitation Dr. Housam Binous 2015
the number of collocation points is increased, the Euclidean norm in the error plot is reduced.
Figure 7 – Euclidian norm of the error between the orthogonal collocation result
conductivity
𝜕𝜃 𝜕2 𝜃 𝜕𝜃
= 𝑓(𝜃, 𝛿) 𝜕𝑥 2 + 𝑓′(𝜃, 𝛿)(𝜕𝑥 )2 , (20)
𝜕𝜏
59
Habilitation Dr. Housam Binous 2015
where 𝑓(𝜃, 𝛿) is the dimensionless thermal diffusivity and 𝛿 is an empirical parameter. The
variables 𝜏, 𝜃, and 𝑥 are the dimensionless time, temperature, and position. The initial
Figure 8 – dimensionless temperature versus the dimensionless position in the slab for 𝜏 =
0.01, 0.05, 0.2, and 0.5 and for the three forms of 𝑓(𝜃, 𝛿)
Figure 8 displays the dimensionless temperature versus the dimensionless position in the slab
for 𝜏 = 0.01, 0.05, 0.2, and 0.5 (the orange, green, red, and cyan curves, respectively). The
colored dots correspond to the solution using orthogonal collocation. The MATHEMATICA© code
allows users to specify the number of collocation points for the calculation as well as the
functional dependency for the thermal diffusivity through the parameter δ. The colored curves
60
Habilitation Dr. Housam Binous 2015
correspond to the dimensionless temperature obtained using NDSolve. The solid curves
correspond to 𝑓(𝜃, 𝛿) = 𝑒𝑥𝑝(𝛿 𝜃); the dashed curves correspond to 𝑓(𝜃, 𝛿) = 1 + 𝛿 𝜃, and
the dotted curves are obtained using 𝑓(𝜃, 𝛿) = 1. The calculations show that for small values
of δ the linearized form for the thermal diffusivity is adequate for short time periods, but less
so for large time periods. Again, in this example there is symmetry, the location of the
(ODEs or elliptic PDEs) or initial value problems (parabolic PDEs) are often encountered in
chemical engineering applications. A popular numerical method for solving such problems is
orthogonal collocation, which reduces the problem to either a system of nonlinear algebraic
equations or a system of nonlinear ODEs in time [17, 18]. The accuracy of the method depends
In this case study, we evaluate the accuracy of the collocation method for studying a family of
The reaction takes place in a spherical liquid drop. Species 𝐶 is absorbed, then diffuses and
undergoes a decomposition reaction in the interior of the drop. We consider the following
possible rates of reaction, 𝑟𝑒 , for the decomposition reaction. The four models are:
61
Habilitation Dr. Housam Binous 2015
The concentration of species 𝐶 is taken to be a function of 𝑟 only. (In the following IC, BC1,
and BC2 stand for initial and boundary conditions.) The number balance for 𝐶 implies:
𝜕𝑐 1 𝜕 𝜕𝑐
= 𝐷(𝑟 2 𝜕𝑟 (𝑟 2 𝜕𝑟)) + 𝑟𝑒 , 0 ⩽ 𝑟 ⩽ 𝑅, (22)
𝜕𝑡
IC: 𝑐(0, 𝑟) = 0,
BC1: (𝜕𝑐/𝜕𝑟)𝑟=0 = 0,
BC2: 𝑐 = 𝑐𝑒𝑞 at 𝑟 = 𝑅,
where 𝑅 is the radius of the drop, taken to be constant during the reaction, and 𝑐𝑒𝑞 is the
dimensionless form, the PDE for the first-order reaction kinetic becomes:
𝜕𝐶 1 𝜕 𝜕𝐶
= 𝜉2 𝜕𝜉 (𝜉 2 𝜕𝜉 ) − 𝐷𝑎 𝑐, 0 ⩽ 𝜉 ⩽ 1, (23)
𝜕𝜏
IC: 𝐶(0, 𝜉) = 0,
BC1: (𝜕𝐶/𝜕𝜉)𝜉=0 = 0,
BC2: 𝐶 = 1 at 𝜉 = 1.
For convenience, we set 𝑘1 𝑐eq = 1 so that the problem depends on a single parameter 𝐷𝑎 =
𝑛−1 2 ⁄
𝑘𝑐eq 𝑅 𝐷, the Damköhler number (where 𝑛 = 1 for kinetic models 1, 3, and 4 and 𝑛 = 2
The orthogonal collocation method is based on Jacobi polynomials (i.e., the problem exhibits
symmetry) that reduce the PDE to a system of nonlinear ODEs that are then solved at the
collocation points using NDSolve. It is possible to can vary the number of interior collocation
points and the value of the Damköhler number. Figures 9 displays the dimensionless
concentration versus the dimensionless radial position for various values of the dimensionless
time using different colors and for the first-order reaction model. The other model present
similar results, which are not given in the present Chapterto keep the treatise as concise as
62
Habilitation Dr. Housam Binous 2015
possible. The colored dots represent the solutions at the collocation points. The colored curves
are the predictions found when the MATHEMATICA© command called NDSolve is applied to
the nonlinear PDE directly. The blue curve shows the behavior of the concentration at a steady
state. This curve was obtained by solving the corresponding boundary value problem using a
shooting technique. It is found that a steady state is reached for 𝜏 > 0.5 for the range of
Figure 9 – dimensionless concentration versus the dimensionless radial position for various
values of the dimensionless time and for the first-order reaction model
63
Habilitation Dr. Housam Binous 2015
Figure 10 – Euclidian norm of the error between the orthogonal collocation result
To quantify the accuracy of the collocation method we compute the Euclidean norm of the
𝑁𝑝
error for 𝑁𝑝 collocation points given by: 𝜖 2 = ∑𝑖=1 |𝐶 (𝜏, 𝜉𝑖 ) − 𝐶𝑒 (𝜏, 𝜉𝑖 )|2 where 𝐶 and 𝐶𝑒 are
the solutions obtained using orthognal collocation and NDSolve. This error is plotted, in
Figure 10 for the first-order reaction model, as a function of time on the error plot. The red dots
denote the times used in concentration versus radial position plots. As the number of
collocation points is increased, the Euclidian norm in the error plot is reduced. For 𝑁𝑝 = 10
64
Habilitation Dr. Housam Binous 2015
Conclusion
The method of orthogonal collocation present in this Chapter can conveniently solve any split
process. Here, the solution of four graduate-level problems have been given and compared
whenever possible to what other technique provide as solution. The author taught for four years
at KFUPM this technique in the graduate-level course on numerical methods called CHE 560.
Several term-projects employing this technique have been assigned to students. These projects
References
[1] R. Baur, J. Bailey, B. Brol, A. Tatusko and R. Taylor, “Maple and the art of
[4] H. Binous, E. Al-Mutairi and N. Faqir, “Study of the separation of simple binary and
[5] H. Binous and M. A. Al-Harthi, Simple batch distillation of a binary mixture, Computer
[6] T. Castrellón, D. C. Botía, R. Gómez, G. Orozco and I. D. Gil, “Using process simulators
in the study, design, and control of distillation columns for undergraduate chemical
65
Habilitation Dr. Housam Binous 2015
630.
[8] S. X. Liu and M. Peng, “The simulation of the simple batch distillation of multiple-
[9] L. M. Porto and R. H. Ogeda, “Java applets for chemical reaction engineering,” Computer
607–618.
[13] W. F. Ramirez, Computational Methods for Process Simulation, 2nd ed., Oxford:
Butterworth–Heinemann, 1997.
[14] J. Benitez, Principles and Modern Applications of Mass Transfer Operations, New
66
Habilitation Dr. Housam Binous 2015
Conductivity by the Orthogonal Collocation Method,” Letters in Heat and Mass Transfer,
1980.
[18] R. G. Rice and D. D. Do, Applied Mathematics and Modeling for Chemical Engineers,
67
Habilitation Dr. Housam Binous 2015
68
Habilitation Dr. Housam Binous 2015
Abstract
We present in this chapter an alternative numerical method for the resolution of transport
transport phenomena problems involving partial differential equations and split boundary
value problems can now readily be mastered. A description of several sample problems and
Introduction
Transport Phenomena and Numerical Methods are typical study materials in most graduate
programs in chemical engineering. We need not elaborate on the different phases in the
teaching of the Numerical Methods course; however it is fair to suggest that there are three
distinct phases: (1) Development of Fortran programs (e.g., Carnahan et al. 1969; Dorn and
McCracken, 1972); (2) Application of IMSL codes; and (3) Application of Commercial
Software codes. The great revolution in computational power has allowed most users to focus
on the third phase, where software codes, such as Matlab®, Mathcad, or MATHEMATICA© can
phenomena. The integration of such software codes in chemical engineering education and
research, both at the undergraduate and graduate levels, has been discussed in the literature
in several articles; see for example, Loughlin and Shaikh (2001), and Parulekar (2006).
69
Habilitation Dr. Housam Binous 2015
In the chapter, we address ourselves to the numerical solution of highly nonlinear problems
finally,
A set of seven typical transport phenomena problems is discussed in the chapter. We begin
with a short description of the Chebyshev orthogonal collocation method and illustrate the
In the discrete Chebyshev–Gauss–Lobatto orthogonal collocation method, the grid points are
defined by
and represent the locations of the extrema of the first kind Chebyshev polynomials, 𝑇𝑁 (𝑥).
70
Habilitation Dr. Housam Binous 2015
with
2𝑁2 +1
𝑑00 = 6
2𝑁2 +1
𝑑𝑁𝑁 = − 6
𝑗𝑦
𝑑𝑗𝑗 = − 2(1−𝑦 2 ) for 1 ≤ 𝑗 ≤ 𝑁 − 1.
{ 𝑗
𝑐 (−1)𝑗+𝑘
𝑑𝑗𝑘 = 𝑐 𝑗 (𝑦 −𝑦
𝑘 𝑗 𝑘)
with
𝑐𝑚 = 1 for 1 ≤ 𝑚 ≤ 𝑁 − 1 and 𝑐0 = 𝑐𝑁 = 2.
If 𝑣 is the vector formed by the values of the function 𝑢(𝑦) at the locations 𝑦𝑗 , 𝑗 = 0, … , 𝑁,
the values of the approximations 𝑣′ and 𝑣′′ of the derivates 𝑢′ and 𝑢′′ of u at the grid points
𝑣′ = 𝐷 𝑣 ; 𝑣′′ = 𝐷2 𝑣.
where𝐷 is the differentiation matrix. The MATHEMATICA© code for building 𝐷 is given in
Appendix 1. A similar code, this time based on Matlab®, can be found in the textbook by
Trefethen (2000).
𝑢 𝑢’ − 𝑢’’ = 1 (3)
71
Habilitation Dr. Housam Binous 2015
. 𝑢(−1) = 0, 𝑢(1) = 2.
𝑢𝑗 (∑𝑁 𝑁
𝑘=0 𝑑𝑗𝑘 𝑢𝑘 ) − ∑𝑘=0 𝑏𝑗𝑘 𝑢𝑘 = 1, 𝑗 = 1, … , (𝑁 − 1)
{ 𝑢0 = 2 (4)
𝑢𝑁 = 0
where (𝑑𝑗𝑘 )0≤𝑗,𝑘⩽𝑁 are the elements of the derivation matrix D given in the previous section
and (𝑏𝑗𝑘 )0≤𝑗,𝑘⩽𝑁 the elements of 𝐷2 = 𝐷 • 𝐷. Such system of nonlinear algebraic equations
is readily solved using MATHEMATICA© and Matlab® built-in functions FindRoot and
A solution of this problem can be found using the shooting technique. The data obtained,
using this later method, are also presented in Table 1 for comparison purposes.
𝜕𝑢 𝜕2 𝑢
= (5)
𝜕𝑡 𝜕𝑦 2
72
Habilitation Dr. Housam Binous 2015
𝒋 𝒚𝒋 𝒖𝒋 (a) 𝒖𝒋 (b)
0 1.0000 2.0000 2.
1 0.9877 1.9837 1.9836
2 0.9511 1.9366 1.9365
3 0.8910 1.8634 1.8634
4 0.8090 1.7704 1.7704
5 0.7071 1.6634 1.6634
6 0.5878 1.5471 1.5470
7 0.4540 1.4243 1.4243
8 0.3090 1.2969 1.2969
9 0.1564 1.1656 1.1656
10 0.0000 1.0310 1.0309
11 -0.1564 0.8938 0.8937
12 -0.3090 0.7552 0.7551
13 -0.4540 0.6170 0.6170
14 -0.5878 0.4822 0.4822
15 -0.7071 0.3545 0.3544
16 -0.8090 0.2385 0.2385
17 -0.8910 0.1399 0.1398
18 -0.9511 0.0642 0.0641
19 -0.9877 0.0164 0.0163
20 -1.0000 0 -2.4087*10-21
𝑢(𝑡 = 0, 𝑦) = 1
𝑢(𝑡, 𝑦 = 0) = 0
73
Habilitation Dr. Housam Binous 2015
𝑢(𝑡, 𝑦 = 1) = 0
This is the well-known one-dimensional diffusion equation. It is possible to find the solution
𝑢′𝑗 (𝑡) = 4 ∑𝑁
𝑘=0 𝑏𝑗𝑘 𝑢𝑘 (𝑡), 𝑗 = 1, … , 𝑁 − 1
{ 𝑢0 (𝑡) = 0 (6)
𝑢𝑁 (𝑡) = 0
where 𝐷2 = 𝐷 • 𝐷 = (𝑏𝑗𝑘 )0≤𝑗,𝑘⩽𝑁 is the square of the derivative matrix, 𝐷. The initial
algebraic equations is readily solved using MATHEMATICA© and Matlab® built-in functions
NDSolve and ode15s, respectively. The solution, 𝑢, at 𝑡 = 0.0126 and at the collocation
This transport problem admits an analytical solution derived using the separation of variable
74
Habilitation Dr. Housam Binous 2015
The data obtained, using the analytical solution, are also presented in Table 2 for comparison
purposes.
𝒋 𝒚𝒋 𝒖𝒋 (a) 𝒖𝒋 (b)
0 1.0000 0 0.
1 0.9938 0.0307 0.0309
2 0.9755 0.1218 0.1225
3 0.9455 0.2670 0.2686
4 0.9045 0.4500 0.4525
5 0.8536 0.6407 0.6437
6 0.7939 0.8029 0.8058
7 0.7270 0.9124 0.9145
8 0.6545 0.9693 0.9704
9 0.5782 0.9914 0.9918
10 0.5000 0.9965 0.9967
11 0.4218 0.9914 0.9918
12 0.3455 0.9693 0.9704
13 0.2730 0.9124 0.9145
14 0.2061 0.8029 0.8058
15 0.1464 0.6407 0.6437
16 0.0955 0.4500 0.4525
17 0.0545 0.2670 0.2686
18 0.0245 0.1218 0.1225
19 0.0062 0.0307 0.03093
20 0 0 0.
Case Studies
75
Habilitation Dr. Housam Binous 2015
Consider as a first case the unsteady evaporation of a liquid A in non-diffusing gas B. The
migration of particles A in the gas phase is described by the equation (Arnold, 1944):
where 𝐷𝐴𝐵 is the gas diffusion coefficient, 𝑐𝐴0 , the interfacial gas-phase concentration,
𝑡 = 0, 𝑐𝐴 (𝑧, 0) = 0
condition) 𝑧 = ∞, 𝑐𝐴 (∞, 𝑡) = 0
𝑐𝐴 (𝜉) 1−erf(𝜉−𝜙)
= , (9)
𝑐𝐴0 1+erf(𝜙)
𝑧
where 𝜉 = and 𝜙, a solution of the nonlinear equation:
√4𝐷𝐴𝐵 𝑡
𝑐𝐴0 2
1 𝑐𝑡 𝑒 −𝜙
𝜙= 𝑐 . (10)
√𝜋 (1− 𝐴0 ) (1+erf(𝜙))
𝑐𝑡
DAEs:
2 𝑁 𝑁 𝑁
2 𝐷𝐴𝐵 2 2
𝑐′𝐴,𝑗 (𝑡) = 𝐷𝐴𝐵 ( ) (∑ 𝑏𝑗𝑘 𝑐𝐴,𝑘 (𝑡)) + 𝑐 ( ) (∑ 𝑑0𝑘 𝑐𝐴,𝑘 (𝑡)) ( ) (∑ 𝑑𝑗𝑘 𝑐𝐴,𝑘 (𝑡)),
𝐿 1 − 𝑐𝐴0 𝐿 𝑘=0 𝐿
𝑘=0 𝑘=0
𝑡
for 𝑗 = 1, … , 𝑁-1
𝑐𝐴,0 (𝑡) = 0
76
Habilitation Dr. Housam Binous 2015
where (𝑑𝑗𝑘 )0≤𝑗,𝑘⩽𝑁 are the elements of the derivative matrix D, (𝑏𝑗𝑘)0≤𝑗,𝑘⩽𝑁 , the elements of
for 𝑗 = 0, … , 𝑁.
The numerical solution, 𝑐𝐴 (𝑧, 𝑡), thus obtained (red dots in Figure 3) can be compared to the
analytical solution represented by the blue curve on the same figure. Excellent agreement
programs with sliders, check boxes, controls, etc. Here, one can vary the values of 𝑡, 𝐷𝐴𝐵 ,
77
Habilitation Dr. Housam Binous 2015
𝑐𝐴0 as well as the number of the Chebyshev collocation points, (𝑁 + 1), and instantaneously
𝜕𝑐 𝜕2 𝑐 𝜕𝑐
𝜕𝑡
= 𝐷𝑓 𝜕𝑥2 − 𝑣 𝜕𝑥, (11)
where 𝐷𝑓 is the diffusion coefficient, 𝑣, the velocity, 𝑐, the concentration of the diffusing
𝑡 = 0, 𝑐(𝑥, 0) = 0,
𝑥 = 0, 𝑐(0, 𝑡) = 1,
𝑥 = ∞, 𝑐(∞, 𝑡) = 0.
Equation (11) has an analytical solution that can be readily be found using Laplace
transforms:
1 𝑥−𝑣𝑡 𝑥+𝑣𝑡
𝑐 (𝑥, 𝑡) = 2 (erfc [2√𝐷𝑡] + 𝑒 (𝑣𝑥)⁄𝐷 erfc [2√𝐷𝑡]). (12)
In Figure 4 the evolution of the concentration 𝑐(𝑥, 𝑡) represented by the analytical solution
(blue curve) and the numerical solution 𝑐(𝑡, 0 ≤ 𝑥 ≤ 𝐿 = 10) obtained using the
As noted in the previous case study, the built-in MATHEMATICA© command Manipulate
78
Habilitation Dr. Housam Binous 2015
governed by a non-linear ODE known as the Falkner-Skan equation (Deen, 1998), named
79
Habilitation Dr. Housam Binous 2015
Equation (13) admits only a numerical solution which requires the application of an
Applying the shooting technique leads to the blue curve while the red dots represent the
velocity obtained using the Chebyshev orthogonal collocation method by solving a system
𝑁
∑ 𝑑0𝑘 𝑓𝑘 = 1
𝑘=0
𝑁
∑ 𝑑𝑁𝑘 𝑓𝑘 = 0
𝑘=0
𝑓𝑁 = 0
2
3 𝑁 2 𝑁 𝑁
2 2 2
( ) (∑ 𝑒𝑗𝑘 𝑓𝑘 ) + ( ) 𝑓𝑗 (∑ 𝑏𝑗𝑘 𝑓𝑘 ) + 𝛽 1 − (( ) ∑ 𝑑𝑗𝑘 𝑓𝑘 ) =0
𝐿 𝐿 𝐿
𝑘=0 𝑘=0 𝑘=0
( )
{ for 𝑗 = 1, … , 𝑁 − 2
where (𝑑𝑗𝑘 )0≤𝑗,𝑘⩽𝑁 are the elements of the derivative matrix D, (𝑏𝑗𝑘)0≤𝑗,𝑘⩽𝑁 , the elements of
The well-known Blasius equation (flow past a flat plate with a wedge angle, 𝛽 = 0) appears
as a particular case in this study. Planar stagnation flow is also found by setting the wedge
attack angle 𝛽 = 𝜋.
Newtonian fluid (such as water), the viscosity is independent of how fast it is stirred, but for a
80
Habilitation Dr. Housam Binous 2015
𝜋
Figure 5 – Velocity profile for flow past a wedge (𝛽 = ).
4
It gets either easier or harder to stir faster for different types of non-Newtonian fluids. Different
constitutive equations, giving rise to various models of non-Newtonian fluids, have been
is assumed, where 𝑛 is the power-law exponent and 𝜅 is the power-law consistency index.
Dilatant or shear-thickening fluids correspond to the case where the exponent in this equation
is positive, while pseudo-plastic or shear-thinning fluids are obtained when 𝑛 < 1. The
viscosity decreases with strain rate for 𝑛 < 1, which is the case for pseudo-plastic fluids (also
called shear-thinning fluids). On the other hand, dilatant fluids are shear thickening. If n=1,
the Newtonian fluid behavior can be recovered. The power-law consistency index is chosen
81
Habilitation Dr. Housam Binous 2015
to be 𝜅 = 1.5 10−5 . For the pseudo-plastic fluid, the velocity profile is flatter near the center,
where it resembles plug flow, and is steeper near the wall, where it has a higher velocity than
the Newtonian fluid or the dilatant fluid. Thus, convective energy transport is higher for
pipe of a power-law fluid, an analytical expression is available (Binous, 2007 and Wilkes,
2006). According to the Carreau model for non-Newtonian fluids, first proposed by Pierre
Carreau,
. . 𝑛−1
𝜇(𝛾) − 𝜇∞ = (𝜇0 − 𝜇∞ )(1 + 𝜆2 𝛾 2 ) 2 (15)
For 𝜆 𝛾 ≪ 1, this reduces to a Newtonian fluid with 𝜇 = 𝜇0. For 𝜆 𝛾 ≫ 1, we obtain a power-
.
law fluid with 𝜇 = 𝜅 𝛾 𝑛−1 .
The infinite-shear viscosity 𝜇∞ and the zero-shear viscosity 𝜇0 of the Carreau fluid are taken
equal to 0 and 2.04 × 10−3 , respectively. The relaxation parameter 𝜆 is set equal to 0.2. For
the flow of a Carreau fluid in a pipe, only numerical solutions are available. The velocity
profile versus radial position is obtained for the steady-state laminar flow of power-law and
ΔP
Carreau fluids in a pipe. The pipe radius is 𝑅 = 1 and the applied pressure gradient is 𝐿
= 1.
For both the power-law and Carreau fluids, the red dots represent the solutions obtained using
the Chebyshev collocation technique (see Figure 6a and 6b). The velocity profile for a power-
law fluid (the blue curve) is obtained from the analytical solution given by (Binous, 2007 and
Wilkes, 2006):
𝛥𝑃
𝑣 (𝑟) = (2𝐿𝜅)1⁄𝑛 (𝑅1+1⁄𝑛 − 𝑟 1+1⁄𝑛 )⁄(1 + 1⁄𝑛).
(16)
The velocity profile for a Carreau fluid (the blue curve) is obtained using the shooting
technique and the built-in MATHEMATICA© function NDSolve. For both fluids, you can vary
the exponent.
82
Habilitation Dr. Housam Binous 2015
(a)
(b)
Figure 6 – Velocity profile for (a) the power-law fluid and (b) the Carreau fluid.
83
Habilitation Dr. Housam Binous 2015
Consider the free convection past a vertical flat plate or wall maintained at a constant
temperature 𝑇𝑤 with 𝑇𝑤 > 𝑇∞ , where 𝑇∞ is the fluid temperature far from the wall.
The governing equations for such a problem in steady state (Deen, 1998; Glasgow, 2010)
are:
2
𝐹(3) + 3𝐹𝐹′ − 2 (𝐹′ ) + 𝛩 = 0
𝛩(2) + 3 Pr 𝐹 𝛩′ = 0
with the B. C. (17)
𝐹(0) = 𝐹′ (0) = 0
𝐹′ (∞) = 0
{ 𝛩(0) = 1; 𝛩(∞) = 0
𝑇−𝑇∞ 𝜈
where 𝛩 = 𝑇 is the dimensionless temperature, Pr = 𝛼, the Prandtl number (a
𝑤 −𝑇∞
dimensionless number giving the rate of viscous momentum transfer relative to heat
𝜓
𝐹(𝜂) = 𝑔𝛽(𝑇𝑤−𝑇∝ ) 1/4 3/4
4𝜈[ ] 𝑥
4𝜈2
Gr𝑥 1/4 𝑦
𝜂=[ ]
4 𝑥
𝑔𝛽ΔT𝑥3
The derivative 𝐹 ′ of 𝐹 is proportional to the fluid velocity. Gr𝑥 = 𝜈2
is the Grashof
number. To solve this split boundary value problem one has to apply an appropriate
numerical procedure such as the shooting and/or relaxation technique. In this study we make
84
Habilitation Dr. Housam Binous 2015
(a)
(b)
Figure 7–(a) Temperature profile and (b) velocity profile for free convection past an
85
Habilitation Dr. Housam Binous 2015
The obtained temperature and velocity profiles vs. 𝜂 for Pr = 10 are represented in figures
7a and 7b as red dots for the results of the Chebyshev integration approach and blue curves
for the shooting method. The Chebyshev collocation procedure requires the solution of a
As can be noted, only qualitative agreement is obtained for the velocity profile using the
shooting and the orthogonal collocation approach. One however can verify that if the Prandtl
number Pr is reduced, the dimensionless temperature curve extends farther away from the
The governing equation, in dimensionless form, for Poiseuille flow for a Newtonian fluid in
𝜕𝑣 1 𝜕 𝜕𝑣
𝜕𝜏
= 4 + 𝜉 𝜕𝜉 (𝜉 𝜕𝜉),
(18)
𝑡𝜈 𝑟
where the dimensionless time is 𝜏 = , the dimensionless radial position,𝜉 = 𝑅, and the
𝑅2
𝑉 (𝑟)
dimensionless velocity, 𝑣 = 𝑉 𝑧 .
𝑧,max
𝑣 (𝜉 = 1, 𝑡) = 0,
𝜕𝑣
𝜕𝜉
(𝜉 = 0, 𝑡) = 0,
𝑣 (𝜉, 𝑡 = 0) = 0.
𝜇
𝜈 = 𝜌 is the kinematic viscosity with 𝜇 the fluid's viscosity and 𝜌, its density. The radius of
1 𝜕𝑝
the pipe is 𝑅, and 𝑣𝑧,max = − 4𝜇 𝜕𝑧 𝑅2 is the velocity at the center of the pipe at steady state.
86
Habilitation Dr. Housam Binous 2015
Figure 8 depicts the velocity profile obtained using Matlab® at 𝜏 = 0.1359. Here one has to
𝒌 𝝀𝒌
1 2.4048
2 5.5201
3 8.6537
4 11.7915
5 14.9309
6 18.0711
7 21.2116
2
𝑟 2
𝜈 (𝜆𝑘 )
𝑉𝑧 (𝑟) 𝑟 − 𝑡
𝑣=𝑉 = (1 − (𝑅) ) − ∑∞
𝑘=1 𝐴𝑘 𝐽0 (𝜆𝑘 𝑅) 𝑒 𝑅2
𝑧,max
87
Habilitation Dr. Housam Binous 2015
where
2 𝑅 𝑟 2
𝐴𝑘 = 2 ∫0 𝑟 𝐽0 (𝜆𝑘 𝑅) (𝑅 − 𝑟2 )𝑑𝑟
(𝑅2 𝐽1 (𝜆𝑘 ))
and 𝜆𝑘 are the roots of the zeroth order Bessel function of the 1st kind (i.e., 𝐽0 (𝜆𝑘 ) = 0).
𝒋 𝒚𝒋 𝒖𝒋 (a) 𝒖𝒋 (b)
0 1.0000 0 -0.0000
1 0.9938 0.0084 0.0084
2 0.9755 0.0326 0.0326
3 0.9455 0.0706 0.0706
4 0.9045 0.1188 0.1188
5 0.8536 0.1732 0.1731
6 0.7939 0.2295 0.2294
7 0.7270 0.2839 0.2838
8 0.6545 0.3335 0.3333
9 0.5782 0.3764 0.3762
10 0.5000 0.4117 0.4114
11 0.4218 0.4394 0.4390
12 0.3455 0.4602 0.4598
13 0.2730 0.4751 0.4747
14 0.2061 0.4852 0.4847
15 0.1464 0.4916 0.4911
16 0.0955 0.4952 0.4947
17 0.0545 0.4970 0.4965
18 0.0245 0.4977 0.4972
19 0.0062 0.4979 0.4973
20 0 0.4979 0.4973
Table 4 –Values at the collocation points of the solution
88
Habilitation Dr. Housam Binous 2015
Table 3 gives a list of these values and Table 4 shows a comparison between the analytical
This aim of this section is to show how the Couette flow in a slot develops. Initially the fluid
and both walls are stationary. To start the flow the lower wall is brought to a constant velocity,
𝑢(0) = 𝑈 = 1 m/s.
𝜕𝑢 𝜕2 𝑢
𝜕𝑡
= 𝜈 𝜕𝑦2
(19)
(𝜈 = 0.1 ; 𝑈 = 1 m/s)
89
Habilitation Dr. Housam Binous 2015
The steady-state solution is the linear velocity profile given by 𝑢(𝑦) = 𝑈 (1 − 𝑦). The
analytical solution of equation (19) can be found by applying the separation of variables
2 1 2 2
𝑢(𝑡, 𝑦) = 𝑈 (1 − 𝑦) − 𝜋 𝑈 ∑∞
𝑘=1 𝑘 sin(𝑘 𝜋 𝑦) 𝑒
−𝑘 𝜋 𝜈𝑡
𝒋 𝒚𝒋 𝒖𝒋 (a) 𝒖𝒋 (b)
0 1.0000 0 -0.0000
1 0.9938 0.0000 0.0000
2 0.9755 0.0000 0.0000
3 0.9455 0.0000 0.0000
4 0.9045 0.0000 0.0000
5 0.8536 0.0000 0.0000
6 0.7939 0.0000 0.0000
7 0.7270 0.0000 0.0000
8 0.6545 0.0000 0.0000
9 0.5782 0.0003 0.0003
10 0.5000 0.0016 0.0016
11 0.4218 0.0079 0.0078
12 0.3455 0.0295 0.0292
13 0.2730 0.0854 0.0849
14 0.2061 0.1940 0.1933
15 0.1464 0.3561 0.3554
16 0.0955 0.5474 0.5467
17 0.0545 0.7313 0.7309
18 0.0245 0.8775 0.8773
19 0.0062 0.9691 0.9690
20 0 1.0000 1.0000
Table 5 – Values at the collocation points of the solution
90
Habilitation Dr. Housam Binous 2015
Using Matlab®, we plot in Figure 9 the velocity profile at 𝑡 = 0.1255 and 𝜈 = 0.1. Here also
one has to solve a system of DAEs using the built-in Matlab® command ode15s. A
comparison between the numerical and analytical results (truncated up to 𝑘 = 100) for 𝑡 =
0.1255 and 𝜈 = 0.1 is given in Table 5. For very small kinematic viscosities, the response is
slower. This result is to be expected since the momentum transport is governed here solely
cm2⁄s), which is just like molecular diffusion and heat conduction in gases a very slow
process.
1 𝜕𝑢 𝜕2 𝑢
= 𝜕𝑥 2 , with 𝑡 ≥ 0 and 0 ≤ 𝑥 ≤ 1. (20)
𝛼 𝜕𝑡
This equation represents heat conduction in a rod. The boundary conditions are such that the
Without loss of generality, one can take the thermal diffusivity, 𝛼, equal to 1 𝑐𝑚2 /𝑠. The initial
condition is given by
𝑢(𝑥, 0) = 1 for 0 ≤ 𝑥 ≤ 1.
The temperature can be found using either NDSolve (solid colored curve) or the Chebyshev
collocation technique (colored dots). Both methods give the same results, which are plotted
in the same diagram (see Figure 10) at various values of time, 𝑡 ranging from 0.01 to 0.1 with
a span of 0.01. You can set the number of interior points used by the Chebyshev collocation
method.
91
Habilitation Dr. Housam Binous 2015
Figure 10 – Temperature profile for 𝑡 ranging from 0.01 to 0.1 with a span of 0.01
1 𝜕𝑢 𝜕2 𝑢 𝜕2 𝑢
= 𝜕𝑥 2 + 𝜕𝑦 2 , with 𝑡 ≥ 0 and 0 ≤ 𝑥, 𝑦 ≤ 1, (21)
𝛼 𝜕𝑡
which represents heat conduction in a two-dimensional domain. The boundary conditions are
such that the temperature, 𝑢, is equal to 0 on all the edges of the domain:
Without loss of generality, one can take the thermal diffusivity, 𝛼, equal to 1 𝑐𝑚2 /𝑠. The initial
condition is given by
𝑢(𝑥, 𝑦, 0) = 1 for 0 ≤ 𝑥, 𝑦 ≤ 1.
92
Habilitation Dr. Housam Binous 2015
Figure 11 – Three contours of the dimensionless temperature, 𝑢, (i.e., 0.25, 0.5, and 0.75)
at 𝑡 = 0.02
Figure 12 – Contour plot of the solution obtained using the Chebyshev collocation method
93
Habilitation Dr. Housam Binous 2015
The dimensionless temperature, 𝑢, can be found using either NDSolve or the Chebyshev
collocation technique. The data given at 𝑥 = 𝑦 = 1/2 and shown in Table 6, indicate that the
two methods give the same results. Also, three contours of the dimensionless temperature, 𝑢,
(i.e., 0.25, 0.5, and 0.75) at 𝑡 = 0.02 are shown in Figure 11 using the red curves and dashed
black curves for the solution obtained with NDSolve and Chebyshev collocation,
respectively.
Again, perfect agreement is observed. Finally, you can set the value of the dimensionless
time, and the contour plot tab lets you display the contour plot of the solution obtained using
Concluding Remarks
In this chapter, we have described how two software packages have been gainfully employed
presenting details of some of the highly nonlinear problems that were selected as illustrative
examples.
In the recent past, the solving of transport phenomena revolved around two issues: setting up
mathematical models and associated initial and boundary conditions, and derivation of
analytical solutions mostly for asymptotic cases of the mathematical models. Nowadays, the
integration of powerful and rather user-friendly numerical codes, such as Matlab® and
94
Habilitation Dr. Housam Binous 2015
Table 6 – The dimensionless temperature, 𝑢(0.5,0.5, 𝑡), found using either NDSolve or the
Chebyshev collocation technique
References
(1944).
64 (2007).
95
Habilitation Dr. Housam Binous 2015
[3] R. B. Bird, W. E. Stewart, and E. N. Lightfoot, Transport Phenomena, 2nd ed., Wiley,
[4] B. Carnahan, H. A. Luther and J. O. Wilkes, Applied Numerical Methods, Wiley, New
York (1969).
[5] W. M. Deen, Analysis of Transport Phenomena, Oxford University Press, New York
(1998).
[6] W. S. Dorn and D. D. McCracken, Numerical methods with Fortran IV- Case Studies,
[7] V. M. Falkner and S. W. Skan, Aeronautical Research Council Reports and Memoranda,
York (2010).
[11] E. V. Martínez,
http://www.mechens.org/wiki/index.php?title=Transient_Plane_Couette_Flow, accessed
March 2014.
96
Habilitation Dr. Housam Binous 2015
[15] J. O. Wilkes, Fluid Mechanics for Chemical Engineers, 2nd ed., Prentice Hall, Upper
Appendix 1
The implementation can be tested for 𝑁𝑝 = 5 through comparison with the results given in
97
Habilitation Dr. Housam Binous 2015
𝐷 = DerivativeMatrix[5]
𝐷 2 = DerivativeMatrix[5] • DerivativeMatrix[5]
98
Habilitation Dr. Housam Binous 2015
Abstract
The present chapter describes how to apply the Chebyshev orthogonal collocation technique to
solve steady-state and unsteady-steady two-dimensional problems. All problems are solved
using one single computer algebra, MATHEMATICA©. The problems include: (1) steady-state
heat transfer in a rectangular bar, (2) steady-state flow in a rectangular duct, (2) steady-state
heat transfer in a cooling cylindrical pin fin, (4) steady-state heat conduction in an annulus, (5)
unsteady-state heat transfer in a rectangular bar, and finally (6) unsteady-state diffusion
reaction system. Whenever possible the results obtained with orthogonal collocation are
compared to the analytical solution in order to validate the applied numerical technique.
Introduction
Many interesting problems in transport phenomena and generally engineering education are
governed by partial differential equations (PDE) such as the heat equation, the Laplace
equation and the wave equation. These equations are often solved analytically using separation
of variables in graduate-level courses. However, these courses frequently do not address the
more realistic nonlinear problems, which admit only numerical solutions. In addition, in many
cases, it is actually nowadays simpler and faster to find a numerical solution to a particular
problem. The objective of the present chapter is to show how one can readily solve partial
HYSYS or EXCEL is common in chemical engineering pedagogical literature [1-12], only one
99
Habilitation Dr. Housam Binous 2015
paper applies this technique[12], restricted however to only one-dimensional problems. The
authors hope that the present treatment of two-dimensional problems will motivate more
chemical engineering students, faculties, and professionals to use this versatile orthogonal
collocation method.
3- Conclusion giving our thought about this technique as a teaching tool in the chemical
The spectral collocation method using Chebyshev orthogonal polynomial basis is known for
its efficiency in term of accuracy in regard to the relatively small number of collocation points
used to solve many engineering problems: the associated error is of the order of 𝑂(1/𝑁 𝑁 ),
which is better than finite difference or finite element methods [16-17]. The main difficulty in
the transition from a one-dimensional case to a two-dimensional case lies in building the
derivative matrix. Let 𝐷𝑁 (N is the number of collocation points) be the first order derivative
matrix for a one-dimensional problems, the partial first order derivative matrix relative to the
first space variable is obtained by performing the Kronecker product of 𝐼𝑁 and 𝐷𝑁 denotes by
𝐼𝑁 ⊗ 𝐷𝑁 . On the other hand, the second space variable first order derivative matrix is obtained
procedure is used to compute the p order derivative for 2D problems by replacing 𝐷𝑁 by 𝐷𝑁𝑝
100
Habilitation Dr. Housam Binous 2015
where 𝐷𝑁𝑝 is the one dimensional p order derivative for 1D problem. For the mixed derivatives
(of order p with respect to the first variable and q with respect to the second variable) can be
𝑐𝑘 (−1)𝑗+𝑘
𝑑𝑘𝑗 = , for 𝑗 ≠ 𝑘
𝑐𝑗 (𝜉𝑘 − 𝜉𝑗 )
−𝜉𝑘
𝐷𝑁 = [𝑑𝑖𝑗 ]0 ≤ 𝑖, 𝑗 ≤ 𝑁 𝑑𝑘𝑘 =− , for 𝑘 ≠ 0 and 𝑘 ≠ 𝑁
2(1 − 𝜉𝑘 2 )
2𝑁 2 + 1
{ 𝑑00 = = −𝑑𝑁𝑁
6
where,
2 𝑘 = 0 or 𝑁
𝑐𝑘 = {
1 otherwise
The two dimensional spectral collocation method consists in finding the values of the unknown
𝑖𝜋 𝑗𝜋
𝑐𝑜𝑠 ( 𝑁 ) and 𝜂𝑗 = 𝑐𝑜𝑠 ( 𝑁 ) where 0 ≤ 𝑖, 𝑗 ≤ 𝑁 with 𝜉𝑖 and 𝜂𝑗 belonging to the mapped
numerical domain [−1,1]. If the physical space is a the rectangle [−𝐴𝑥 , 𝐴𝑥 ] × [−𝐴𝑦 , 𝐴𝑦 ], the
𝑥(𝜉) = 𝐴𝑥 𝜉
{
𝑦(𝜂) = 𝐴𝑦 𝜂
𝑁 𝑁
101
Habilitation Dr. Housam Binous 2015
where 𝑇𝑖 and 𝑇𝑗 are the Chebyshev orthogonal polynomial of degree 𝑁. The derivatives of
𝜕𝑃𝑁𝑁 𝜓(𝜉, 𝜂) 1
= ( 𝐼𝑁 ⊗ 𝐷𝑁 ) 𝑃𝑁𝑁 𝜓(𝜉, 𝜂)
𝜕𝜉 𝐴𝑥
𝜕𝑃𝑁𝑁 𝜓(𝜉, 𝜂) 1
=( 𝐷 ⊗ 𝐼𝑁 ) 𝑃𝑁𝑁 𝜓(𝜉, 𝜂)
𝜕𝜂 𝐴𝑦 𝑁
This method make possible the resolution of any partial differential equation by finding the
expansion coefficients, 𝑎𝑖𝑗 , which allow us to find the values of the unknown function 𝜓 on
1 2
1 2
∆= ( 𝐼𝑁 ⊗ 𝐷𝑁 ) + ( 𝐷 ⊗ 𝐼𝑁 )
𝐴2𝑥 𝐴2𝑦 𝑁
The above presentation makes clear that the derivative matrix 𝐷𝑁 plays a key role in the
implementation of the spectral collocation technique. Its calculation, for any number of nodes
for N =5 the matrices 𝐷5 and 𝐷52 with page 7 of Ref. [15]. To evaluate the partial derivative
matrices, the built-in functions IdentityMatrix and KroneckerProduct are used, for
instance and in the case of N =3 in both directions (𝑥 and 𝑦), the first order partial derivate
sequences respectively,
KroneckerProduct[IdentityMatrix[4], dM[3]]
KroneckerProduct[dM[3],IdentityMatrix[4]]
102
Habilitation Dr. Housam Binous 2015
and the second order partial derivate matrices relative to 𝑥, 𝔻2𝑥 = 𝐼𝑁 ⊗ 𝐷𝑁2 and to 𝑦,
KroneckerProduct[IdentityMatrix[4], dM[3].dM[3]]
KroneckerProduct[dM[3].dM[3],IdentityMatrix[4]]
The results of these calculations can be compared with those given in page 79 of Ref. [15].
Let us consider a bar of rectangular cross section subject to both temperature and heat flux
boundary conditions (see Figure 1). The governing equation and boundary conditions in
𝜕 2𝜃 𝜕 2𝜃
+ =0
𝜕𝑥 2 𝜕𝑦 2
With 0 ≤ 𝑥 ≤ 𝐿 and 0 ≤ 𝑦 ≤ 𝑊
𝜕𝜃 𝑞𝑤𝑎𝑙𝑙
(𝜕𝑥 ) =− (Continuity of heat flux at 𝑥 = 0 boundary)
𝑥=0 𝑘
For this case study an analytical solution can be obtained using the separation of variables
103
Habilitation Dr. Housam Binous 2015
Figure 1 – Case study 1: Boundary conditions at the edges of the rectangular bar cross
section.
∞
2 𝜃𝑏 sin(𝜆𝑛 𝐿) sinh(𝜆𝑛 𝑦)
𝜃=∑ [
𝐿 𝜆𝑛 sinh(𝜆𝑛 𝑊)
𝑛=0
with
(2 𝑛+1) 𝜋
𝜆𝑛 = for 𝑛 = 0, 1, 2, …
2𝐿
collocation points deliver the temperature distribution in the bar represented in Figure 2. The
solution steps are as follows. First, the spatial variables (i.e., 𝑥 and 𝑦) are discretized to
transform the partial differential equation (PDE) into a systems of 1681linear algebraic
equations where the unknowns are the values of 𝜃 at the nodes. This linear equations system is
then readily solved using the built-in command NSolve. Table 1 shows that the numerical
104
Habilitation Dr. Housam Binous 2015
technique and the analytical solution agree almost perfectly. The summation in the analytical
0 -1.6156 10-9 0
0.00615583 0.75737 0.757661
0.0244717 3.00066 3.00096
0.0544967 6.5918 6.5921
0.0954915 11.1934 11.1937
0.146447 16.299 16.2993
0.206107 21.3945 21.3948
0.273005 26.0737 26.074
0.345492 30.0519 30.0522
0.421783 33.1408 33.1411
0.5 35.2237 35.224
0.578217 36.2405 36.2408
0.654508 36.1814 36.1817
0.726995 35.0925 35.0928
0.793893 33.0908 33.0911
0.853553 30.392 30.3923
0.904508 27.3389 27.3392
0.945503 24.3912 24.3915
0.975528 22.0123 22.0126
0.993844 20.5087 20.5097
1. 20. 20.0312
105
Habilitation Dr. Housam Binous 2015
Consider the fluid flow through a rectangular duct (see Figure 3). The flow velocity in the
cross section obeys the following governing equation with the associated boundary
conditions:
𝜕 2 𝑣𝑧 𝜕 2 𝑣𝑧 𝜕𝑃
2
+ 2
= ( ) /𝜇
𝜕𝑥 𝜕𝑦 𝜕𝑧
With 0 ≤ 𝑥 ≤ 𝐿 and 0 ≤ 𝑦 ≤ 𝑊
𝜕𝑣
( 𝜕𝑥𝑧 ) = 0 (Symmetry condition)
𝑥=0
𝜕𝑣
( 𝜕𝑦𝑧 ) = 0 (Symmetry condition)
𝑦=0
106
Habilitation Dr. Housam Binous 2015
Figure 3 – Case study 2: Boundary conditions for the rectangular duct problem
(Only the domain [0, 𝐿] × [0, 𝑊]is considered).
The separation of variables technique yields the following analytical solution (Ref. [16]):
∞
2 −𝜕𝑃 (−1)𝑛 cosh(𝜆𝑛 𝑦)
𝑣𝑧 = ( )∑ 3 [1 − ] cos(𝜆𝑛 𝑥)
𝜇 𝜕𝑧 𝐿 𝜆𝑛 cosh(𝜆𝑛 𝑊)
𝑛=0
with
(𝑛+1/2) 𝜋
𝜆𝑛 = for 𝑛 = 0, 1, 2, …
𝐿
collocation points. The procedure is similar to that applied in case study 1: Discretization of
the spatial variables to transform the PDE into a system of 1681 linear algebraic equations
where the unknowns are now the values of 𝑣𝑧 at the nodes followed by its numerical solution
using the built-in MATHEMATICA© command NSolve. Figure 4 gives the velocity distribution
in the rectangular duct. The comparison between both solution techniques (Table 2) makes
107
Habilitation Dr. Housam Binous 2015
clear that the numerical techniques and the analytical solution agree very well. Again, the
Figure 4 – Case study 2: Velocity distribution 𝑣𝑧 (𝑥, 𝑦) obtained using Chebyshev orthogonal
collocation
𝜕𝑃
( )
𝜕𝑧
(𝐿 = 1, 𝑊 = 1, and = −10)
𝜇
Let us consider now a problem with a cylindrical geometry. In electronic systems, a fin is a
heat sink or a passive heat exchanger that cools a device by dissipating heat into the surrounding
medium (i.e., air). A cylindrical pin fin, used to maximize heat transfer to a fluid between two
walls, is depicted in Figure 5. The walls are at a high temperature 𝑇𝑤 . The fluid flowing over
the pin has a free stream temperature 𝑇∞ . The heat transfer coefficient between pin wall and
108
Habilitation Dr. Housam Binous 2015
0 2.91088 2.91088
0.00615583 2.91079 2.91079
0.0244717 2.90941 2.90941
0.0544967 2.90357 2.90357
0.0954915 2.88841 2.88841
0.146447 2.8579 2.8579
0.206107 2.80553 2.80553
0.273005 2.72495 2.72495
0.345492 2.61066 2.61066
0.421783 2.45868 2.45868
0.5 2.26719 2.26719
0.578217 2.0372 2.0372
0.654508 1.77313 1.77313
0.726995 1.48319 1.48319
0.793893 1.17947 1.17947
0.853553 0.877396 0.877396
0.904508 0.594772 0.594772
0.945503 0.350128 0.350128
0.975528 0.160816 0.160816
0.993844 0.0410119 0.0410119
1. 0 0
1 𝜕 𝜕𝜙 𝜕 2𝜙
(𝑟 )+ 2 =0
𝑟 𝜕𝑟 𝜕𝑟 𝜕𝑧
0 ≤ 𝑟 ≤ 𝑟0 and 0 ≤ 𝑧 ≤ 𝐿
109
Habilitation Dr. Housam Binous 2015
𝜕𝜙
( 𝜕𝑧 ) = 0 (Axial symmetry condition)
𝑧=0
𝜕𝜙
( 𝜕𝑟 ) = 0 (Radial symmetry condition)
𝑟=0
𝜕𝜙 ℎ
( 𝜕𝑟 + 𝑘 𝜙) = 0 (Continuity of heat flux at the boundary fin/surrounding air)
𝑟=𝑟0
The analytical solution of the differential equation obtained using the separation of variables
∞
2 Bi 𝐽0 (𝜆𝑛 𝑟)cosh(𝜆𝑛 𝑧)
𝜙=∑
cosh(𝜆𝑛 𝐿) 𝐽0 (𝜆𝑛 𝑟0 )[(𝜆𝑛 𝑟0 )2 + Bi2 ]
𝑛=1
ℎ 𝑟0
Bi =
𝑘
𝑓(𝜆) = (𝜆 𝑟0 )𝐽1 (𝜆 𝑟0 ) − 𝐽0 (𝜆 𝑟0 ) Bi
A plot of 𝑓(𝜆) vs. 𝜆 is shown in Figure 7. To find out all zeroes of 𝑓(𝜆) the systematic graphical
approach developed by Wagon [17] is adopted (See Table 3 for the first 16 roots of this
equation).
110
Habilitation Dr. Housam Binous 2015
Figure 6 – Case study 3: Boundary conditions for the cylindrical pin fin problem
(Only the domain [0, 𝐿] × [0, 𝑟0 ] is considered).
111
Habilitation Dr. Housam Binous 2015
𝒏 𝝀𝒏
1 1.25578
2 4.07948
3 7.1558
4 10.271
5 13.3984
6 16.5312
7 19.6667
8 22.804
9 25.9422
10 29.0812
11 32.2207
12 35.3606
13 38.5007
14 41.6411
15 44.7817
16 47.9223
The numerical solution of this 2-D problem obtained using our MATHEMATICA©
depicted in Figure 8. Again, the consecutive steps in the solution procedure are the same as in
both foregoing case studies. First, one discretizes the spatial variables 𝑧 and 𝑟 to transform the
partial differential equation (PDE) into a system of 1681 linear algebraic equations, where the
112
Habilitation Dr. Housam Binous 2015
(For 𝐿 = 1, 𝑟0 = 1, and Bi = 1)
113
Habilitation Dr. Housam Binous 2015
0 0.632547 0.632547
0.00615583 0.632539 0.632539
0.0244717 0.632421 0.632421
0.0544967 0.63192 0.63192
0.0954915 0.630621 0.630621
0.146447 0.628013 0.628013
0.206107 0.623555 0.623555
0.273005 0.616743 0.616743
0.345492 0.607178 0.607178
0.421783 0.594632 0.594632
0.5 0.579107 0.579107
0.578217 0.560873 0.560873
0.654508 0.540486 0.540486
0.726995 0.518768 0.518768
0.793893 0.496748 0.496748
0.853553 0.475574 0.475574
0.904508 0.456401 0.456401
0.945503 0.440289 0.440289
0.975528 0.428122 0.428122
0.993844 0.420554 0.420554
1. 0.417987 0.417987
for 𝐿 = 1, 𝑟0 = 1, and Bi = 1.
Then, this is followed by the resolution of the algebraic system. The comparison of the
numerical and analytical solutions for a particular value of the axial position 𝑧 (Table 4) shows
114
Habilitation Dr. Housam Binous 2015
that the results are in almost perfect concordance. The summation in the analytical solution was
truncated up to 𝑛 = 16.
All three problems treated above admit a straightforward analytical solution using separation
of variables. Let us consider a bit trickier problem: steady-state heat conduction in an annulus
with periodic boundary condition, which analytical solution may be more difficult to derive. In
1𝜕 𝜕𝑢 1 𝜕 2𝑢
(𝑟 )+ 2 2 = 0
𝑟 𝜕𝑟 𝜕𝑟 𝑟 𝜕𝜃
𝜋 𝜋
2 ≤ 𝑟 ≤ 4 and − 2 ≤ 𝜃 ≤ 2
𝜕𝑢
(𝜕𝜃) 𝜋 = 0 (Symmetry condition)
𝜃=
2
𝜕𝑢
(𝜕𝜃) 𝜋 = 0 (Symmetry condition)
𝜃=−
2
The dimensionless temperature, 𝑢, can be found using either directly the MATHEMATICA© built-
115
Habilitation Dr. Housam Binous 2015
Figure 9 – Case study 4: Geometry and boundary conditions of the heat transfer in an annulus
problem
116
Habilitation Dr. Housam Binous 2015
In the latter case and with 𝑁𝑝 = 31 collocation points in both spatial directions 𝑟 and 𝜃, the
discretization yields a system of 961 algebraic equations where the unknowns are the values of
2. 0 0
2.02185 0.0135411 0.0134794
2.08645 0.0531012 0.0531242
2.19098 0.117569 0.117529
2.33087 0.209848 0.20979
2.5 0.339275 0.339125
2.69098 0.520965 0.520787
2.89547 0.772515 0.772214
3.10453 1.10861 1.10829
3.30902 1.53406 1.53364
3.5 2.03661 2.03602
3.66913 2.58221 2.58157
3.80902 3.1157 3.11502
3.91355 3.56867 3.56793
3.97815 3.87382 3.87317
4. 3.98161 3.98085
Figure 10 gives the contour plot of the obtained solution. As can be noted by comparing the
results of both numerical solution methods for 𝜃 = −0.92329 in Table 5, the temperature
117
Habilitation Dr. Housam Binous 2015
For an infinitely long quadratic bar, the transient two-dimensional heat conduction is described
1 𝜕𝑢 𝜕 2 𝑢 𝜕 2 𝑢
= +
𝛼 𝜕𝑡 𝜕𝑥 2 𝜕𝑦 2
𝑡 ≥ 0 and 0 ≤ 𝑥, 𝑦 ≤ 1.
Without loss of generality, one can take the thermal diffusivity, 𝛼, equal to 1 cm2 /s. The
associated boundary conditions are set as follows: The dimensionless temperature, 𝑢, is equal
𝑢(𝑥, 𝑦, 0) = 1 for 0 ≤ 𝑥, 𝑦 ≤ 1.
It is then a transient cooling process of a solid bar. The dimensionless temperature distribution
in the solid can be found using either directly the MATHEMATICA© built-in command NDSolve
or the Chebyshev collocation technique. If we adopt the latter method with a number of
transform the PDE into a systems of 441 linear and coupled first-order ordinary differential
equations (ODEs) where the unknowns are the time dependent values of 𝑢𝑖,𝑗 (𝑡) at the nodes
using the built-in MATHEMATICA© command NDSolve. As illustration of the results of this
numerical procedure a typical map of the obtained dimensionless isotherms (for 𝑡 = 0.1) is
depicted in Figure 11. Table 6 on the other hand compares the dimensionless temperature data
118
Habilitation Dr. Housam Binous 2015
obtained for 𝑡 = 0.1 and at 𝑥 = 0.421782 by both numerical methods considered, NDSolve
and the Chebyshev collocation. As can be noted the two methods give almost same results.
0 0 0
0.00615583 0.00422514 0.00422414
0.0244717 0.016781 0.016777
0.0544967 0.037224 0.0372151
0.0954915 0.0645645 0.0645498
0.146447 0.0970036 0.0969833
0.206107 0.131769 0.131743
0.273005 0.165202 0.165168
0.345492 0.193177 0.193141
0.421783 0.211828 0.211786
0.5 0.218382 0.218343
0.578217 0.211828 0.211786
0.654508 0.193177 0.193141
0.726995 0.165202 0.165168
0.793893 0.131769 0.131743
0.853553 0.0970036 0.0969833
0.904508 0.0645645 0.0645498
0.945503 0.037224 0.0372151
0.975528 0.016781 0.016777
0.993844 0.00422514 0.00422414
1. 0 0
(𝑥 = 0.421782 , 0 ≤ 𝑦 ≤ 1, 𝑡 = 0.1).
119
Habilitation Dr. Housam Binous 2015
Figure 11 – Case study 5: Contour plot of the dimensionless temperature at time 𝑡 = 0.028.
6- Reaction-Diffusion System
In the last decades, developmental biologists have extensively used the reaction–diffusion
model to explain the pattern formation in living organisms. Turing proposed the original model
in 1952 [18]. The model is based on the idea that the pattern formation results from two
fundamental mechanisms: (1) coupled catalytic and autocatalytic reactions in a space element
between two species, an activator and an inhibitor, and (2) transfer of the interacting species to
and from the neighboring space elements through a diffusional transport mechanism. Under
appropriate reaction and diffusion conditions, a periodic pattern is formed from an initially
Examples of pattern formation can be found in Biology, Chemistry (the famous Belousov–
To illustrate the mechanism of pattern formation let us consider the following hypothetical
2𝐴 + 𝐶 → 3𝐴 (R1)
120
Habilitation Dr. Housam Binous 2015
2𝐴 + 𝐷 → 2𝐴 + 𝐵 (R2)
𝐵+𝐶 ⇄ 𝐸 (R3)
𝐴 → 𝐹 (R4)
𝐵 → 𝐺 (R5)
The species 𝐷, 𝐸, 𝐹 and 𝐺 are supposed so abundant in the reaction mixture that their respective
concentrations [𝐷], [𝐸], [𝐹] and [𝐺] can be considered constant. The activator 𝐴 reacts with
species 𝐶 to produce more 𝐴 by the autocatalytic reaction R1– 𝐴 is reactant, catalyst and
product– but it also promotes the production of the inhibitor 𝐵 by the catalytic reaction R2– 𝐴
is here catalyst. Both activator and inhibitor decay with time (Reactions R4 and R5).
Because of the equilibrium reaction R3, the concentrations of 𝐵 and 𝐶 are such that: [𝐵] ∝
1/[𝐶]. Denoting by 𝑘𝑗 the reaction constant of reaction 𝑗 we have for the reaction rates of 𝐴
and 𝐵
𝑟𝐵 = 𝑘2 [𝐴]2 − 𝑘5 [𝐵]
As can be noted from the last expression of the reaction rate of 𝐴, 𝑟𝐴 , the species 𝐵 inhibits the
activator production, hence its name: the larger its concentration the lower is the production
rate of 𝐴.
Introducing the variables 𝑢 ∝ [𝐴] and 𝑣 ∝ [𝐵] the governing equations of the reaction-
𝜕𝑢 𝜕 2𝑢 𝜕 2𝑢
= 𝒟𝐴 ( 2 + 2 ) + 𝑢2 ⁄𝑣 − 𝛽𝑢
𝜕𝑡 𝜕𝑥 𝜕𝑦
𝜕𝑣 𝜕 2𝑣 𝜕 2𝑣
= 𝒟𝐵 ( 2 + 2 ) + 𝑢2 − 𝑣
𝜕𝑡 𝜕𝑥 𝜕𝑦
121
Habilitation Dr. Housam Binous 2015
With 0 ≤ 𝑥, 𝑦 ≤ 1. In the present study we set 𝛽 = 1. For the formation of spatial patterns the
diffusion rates of activator and inhibitor should be very different: We set for the diffusion
coefficients respectively 𝒟𝐴 = 0.0005 cm2 /s and 𝒟𝐵 = 0.01 cm2 /s. For the numerical
solution of the ODEs system, we adopt further (1) periodic boundary conditions as well as (2)
𝑢(𝑥, 𝑦, 𝑡 = 0) = 𝜓𝑢𝑥𝑦 ,
Where 𝜓𝑢𝑥𝑦 and 𝜓𝑣𝑥𝑦 are numbers taken randomly in the interval [0, 1].
The Chebyshev orthogonal collocation method applied with 𝑁𝑝 = 31 nodes in both spatial
directions transforms the system of two coupled nonlinear PDEs into a system of 1922
nonlinear coupled first-order ordinary differential equations. This system of ODEs is solved
in a couple of seconds using the built-in MATHEMATICA© command NDSolve. Figures 12a
and 12b show as illustration of the obtained results the 2D inhibitor concentration distribution
It is interesting to note in particular the emergence of Turing patterns, similar to the “leopard
Conclusion
We have solved in this chapter several complex two-dimensional linear and nonlinear chemical
engineering problems using the Chebyshev orthogonal collocation method. Both steady-state
and unsteady-state problems were considered. This numerical technique is used in combination
with an appropriate computer algebra; in the present work we adopted the software
MATHEMATICA© however one can easy implement the same solution methodology with
Matlab®.
122
Habilitation Dr. Housam Binous 2015
(a) 𝑡 = 0
(b) 𝑡 = 100
123
Habilitation Dr. Housam Binous 2015
times.
The author usually assign four-month term-projects to their students in the graduate-level
projects constitute up to 30 % of the students’ final grade. This assignment is more profitable,
if students tackle problems that are more realistic rather than limit themselves to the few
transform.
Finally, the various MATHEMATICA© codes written for the six case studies are available upon
References
[1] R. Baur, J. Bailey, B. Brol, A. Tatusko and R. Taylor, “Maple and the art of
pp.223-234.
[3] N. Brauner and M. Shacham, “Numerical experiments in fluid mechanics with a tank
pp.175-183.
[4] H. Binous, E. Al-Mutairi and N. Faqir, “Study of the separation of simple binary and
124
Habilitation Dr. Housam Binous 2015
simulators in the study, design, and control of distillation columns for undergraduate
[8] S. X. Liu and M. Peng, “The simulation of the simple batch distillation of multiple-
pp. 607–618.
125
Habilitation Dr. Housam Binous 2015
[14] J. P. Boyd, Chebyshev and Fourier Spectral Methods: Second Edition, Mineola: Dover,
2001.
[16] T. D. Bennett, Transport by Advection and Diffusion, New York: Wiley, 2013.
[18] A.M. Turing, The chemical basis of morphogenesis., Phils. Trans. R. Soc. B, 237(641),
[19] H. Meinhardt, The Algorythmic Beauty of Sea Shells. Berlin: Springer-Verlag, 1995.
pp. 112–123.
[21] J.D. Murray, Mathematical Biology I: An introduction., Third Edition, New York:
Springer, 2001.
[22] J.D. Murray, Mathematical Biology I: Spatial Models and Biomedical Applications.,
126
Habilitation Dr. Housam Binous 2015
Appendix
1- Derivative matrix 𝑫𝑵 :
dM Np_ : Module d, i, j, k, y, c ,
y i_ : Cos i ;
Np
y i
d i_, i_ : If i 0 && i Np, ;
2
2 1 y i
c i_ : If i 0 Or i Np, 1, 2 ;
j k
c j 1
d j_, k_ : If j k, ;
c k y j y k
2 Np2 1
d 0, 0 ; d Np, Np d 0, 0 ;
6
Np
1
d 0, Np ; d Np, 0 d 0, Np ;
2
Table d i, j , i, 0, Np , j, 0, Np N
127
Habilitation Dr. Housam Binous 2015
128
Habilitation Dr. Housam Binous 2015
The present chapter shows how the Particle Swarm Optimization technique, PSO, can
problems. Four case studies are considered: (1) the behavior of a family of one-dimensional
functions, which exhibit rough topology featuring several local minima and a global minimum,
(2) the search of the minimum of the Rosenbrock banana function, (3) the prediction of the
exit concentration of chemical species obeying the van de Vusse mechanism in a continuously
stirred-tank reactor (CSTR) and (4) the determination of the interaction parameters for binary
mixtures containing acetone using the Wilson activity coefficient model and appropriate
examples, we show that PSO is a versatile method of optimization that can readily be
Introduction
Particle Swarm Optimization (PSO) is a stochastic population based optimization method that
belongs to swarm intelligence algorithms. It was first introduced by Kennedy and Eberhart [1]
in 1995 as a technique for optimization that does not suffer from the shortcomings of Genetic
Algorithms (GA). Later, the method was modified using inertia weighting by Shi and
Eberhart[2] to consider different balances and compromises between local and global search
ability. PSO is inherently tied to artificial life and evolutionary computation concepts. PSO
was intended to simulate the social behavior of flocks of birds, swarms of bees or ant colonies,
or schools of fish. However, many researchers from other disciplines and fields had introduced
129
Habilitation Dr. Housam Binous 2015
The basic concept of PSO is very simple in nature; in which many potential solutions (particles
or candidates) are generated at random and allowed to evolve through the solution hyperspace
using the particle coordinates and velocities. The collection of particles is called a swarm. The
convergence to the solution is tied to two components of the algorithm: the individual and the
a) The cognitive component (pbest) which is determined by noting that each individual
particle keeps track of its coordinates in the solution hyperspace which are associated
b) The social component (gbest) the overall best value, and its coordinates, obtained thus
Two recent condensed reviews of PSO and its variants are given in Sedighizadeh and
Masehian[4] and Rini et al[5]. PSO has seen an increasing publication rate since 2003 as shown
in Figure 1 and Figure 2. Year 2015 is estimated based on half of the month of January scaled
to the whole year. These figures are generated based on Scopus search of all articles that have
“Particle Swarm Optimization” keywords in their title. Since its introduction in 1995, PSO took
about eight years until it has built enough momentum and gained attention from researchers in
various disciplines and fields. The field appears to be mature with a steady publication rate of
about 1800 article per year. The Chemical Engineering (CHE) discipline lagged about 3 years
to have enough momentum of publications behind other disciplines (see Figure 3). However,
it is clear that the number of publications is on the rise compared to a mature number of
publications for all disciplines. Many CHE applications have been published recently
including, but not limited to, phase stability and splitting[6, 7], estimation of kinetic
parameters[8], optimization of liquid desiccant regenerators [9] and utility systems in chemical
process plants[10]. Therefore, it is the intention of the current article to introduce the PSO
technique and illustrate some of its potential uses and applications in the CHE discipline.
130
Habilitation Dr. Housam Binous 2015
However, this work is not intended as a comprehensive coverage of all potential applications
of PSO. The applications were developed using MATHEMATICA©. This computer algebra was
selected due to the simplicity of coding as well as the powerful computational kernel.
Nevertheless, the coding can easily be ported to other programs e.g., Matlab®. Four case studies
are considered: (1) the behavior of a family of one-dimensional functions, which exhibit rough
topology featuring several local minima and a global minimum, (2) the search of the minimum
of the Rosenbrock banana function, (3) the prediction of the exit concentration of chemical
species obeying the van de Vusse mechanism in a continuously stirred-tank reactor (CSTR)
and (4) the determination of the interaction parameters for binary mixtures containing acetone
using the Wilson activity coefficient model and appropriate experimental vapor-liquid
2500
2000
Published PSO articles
1500
1000
500
0
1995 2000 2005 2010 2015
Year
131
Habilitation Dr. Housam Binous 2015
The chapter starts by a concise presentation of the Particle Swarm Optimization technique.
Then, the four case studies are treated. Finally, we conclude with our experience with PSO in
16000
Cumulative published PSO articles
12800
9600
6400
3200
0
1995 2000 2005 2010 2015
Year
We adopt the inertia weighting method combined with velocity clamping, without further
elaboration on mathematical details which can be found elsewhere[4, 5]. In PSO, each particle
or potential solution has two state variables, which need to be updated: its current coordinates
(position) in the search hyperspace xi (t ) and its current velocity vi (t ) where t is a discrete time
step or iteration counter. The index i refers to particle number. PSO is initialized with a group
of random particles (solutions) and then searches for optima by evolving to update the new
generations.
132
Habilitation Dr. Housam Binous 2015
25
20
CHE PSO articles
15
10
0
1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015
Year
In all iterations, each particle is updated by following two "best" values: its own best (pbest)
and the global best (gbest). The equations used to update the velocities and positions are given
below:
xi (t 1) xi (t ) vi (t 1) (2)
Where Pbest ,i is particle i personal best position achieved up to the current time (iteration), gbest
is the global best position achieved up to the current time (iteration) by the population of
particles. R1 and R2 are two random numbers of the same dimensionality as the problem
133
Habilitation Dr. Housam Binous 2015
concerned extracted from a uniform distribution in the interval [0,1]. The coefficients c1 and c2
are balance factors between the self-knowledge and social knowledge effects in transferring
the particle towards the target. Usually the value 2 is assigned for both factors or c1 c2 4 is
another alternative assignment [2, 4]. The inertia weight factor, controls the velocity of the
particle towards the target. If it is too high then the algorithms may face stability issues, and if
it is set to low of a value then the algorithm may be too slow in converging to solution. The
a) Inertia, which makes the particle move at the same velocity. This is related to
diversification in which the particle searches for new regions of solution to identify the
individual particle position or returns it to a previous position, better than the current.
c) Social knowledge: Makes the particle follow the best neighbors’ direction towards the
target.
The basic algorithm for the PSO is illustrated in Figure 4. The value of the objective function
Particle Swarm Optimization technique is used to find the minimum of the following one-
134
Habilitation Dr. Housam Binous 2015
with x[-10,10]. This function exhibits a plethora of local minima and a single global
minimum. Thus, a naïve nonlinear programming approach may fail to find the global
minimum. Here, one can vary the parameters 𝑎 and 𝑏 with the help of sliders and
MATHEMATICA© built-in command Manipulate. For the global minimum of f(x), perfect
agreement is found using either the MATHEMATICA© built-in command NMinimize (the green
open square) or PSO (the red dots) with the appropriate number of iterations. Figure 5 displays
135
Habilitation Dr. Housam Binous 2015
the initial positions of the swarm of particles as well as the results after 439 iterations. It is
found that all particles are attracted to the global minimum i.e., all particle collapse to the point
shown in red.
Figure 5 – PSO with a population size equal to 50 and a 1D test function with a=0.1
and b=2. Left panel is the initial distribution of particles and right panel is the result
PSO is used to find the minimum of the following 2D test function (the Rosenbrock banana
non-convex function used as a performance test problem for optimization algorithms, was first
introduced by Howard H. Rosenbrock in 1960[11]. Figure 6 displays the initial positions of the
136
Habilitation Dr. Housam Binous 2015
swarm of particles as well as the results after 275 iterations. It is found that all particle are
attracted to the global minimum i.e., all particles collapse to the point {1,1}. Perfect agreement
is found between the MATHEMATICA© built-in command NMinimize (blue dot at {1,1}) or
PSO (red dots). It should be noted that this command (i.e., NMinimize) has an option called
Method, which allows users to choose among various optimization algorithms such as:
Nelder-Mead, Differential Evolution, Simulated Annealing and Random Search. One can vary
the number of iterations as well as the swarm size. However, it should be noticed that these
two parameters have complementary effects. Here, the problem is two-dimensional, but
Figure 6 – PSO with a population size equal to 50 and the Rosenbrock banana test
function. Left panel is initial distribution of particles and right panel is the result
137
Habilitation Dr. Housam Binous 2015
Consider the following complex reaction scheme (called the van de Vusse mechanism[12]),
taking place in a continuous stirred-tank reactor (CSTR) with a residence time and rate
k1 k2
A BC (4)
k3
2 A D (5)
The governing rate equations are given by the following set of ordinary differential equations:
dCB 1
CB k1C A k2CB (7)
dt
dCC 1
CC k2CB (8)
dt
dCD 1 1
CD k3C A2 (9)
dt 2
The feed to the reactor is pure A ( CA,i 1 ). Initially the reactor contains only species A. Thus,
CA 1, CB CC CD 0 (10)
determines the values of the residence time and the rate constants using a set of experimental
data has been developed. The method of solution is based on particle swarm optimization. A
plot of the exit concentrations of the chemical species A, B, C, and D versus time (in blue, red,
138
Habilitation Dr. Housam Binous 2015
magenta, and green, respectively) is given for the experimental data and the theoretical model
in Figure 7. One can increase the noise of the experimental data to study the influence on the
set of rate constants and residence time. The resultant values , k1, k2 and k3 are indicated on
the plot in blue and red, respectively. The value of the fitness is indicated in green in Figure
7. PSO finds the best solution after a number of iterations and for a chosen size of the swarm,
both of which one can vary. Notice, however, their complementary effect. Here, the problem
straightforward. It is seen that the fit is almost perfect (i.e., the values of the fitness are very
close to zero) when the noise of the experimental data is small (left panel of Figure 7). A
similar result is observed for large swarm sizes and number of iterations.
Left panel shows PSO fitting results with smooth experimental data, while right panel
139
Habilitation Dr. Housam Binous 2015
Finally, when the value of the noise of the experimental data is large, the fit is not perfect and
the fitness value is significantly larger than zero, as shown in the right panel of Figure 7.
equilibria (VLE) data for nine binary mixtures containing acetone plus one of the following
propanol or 1-hexene. It uses the Wilson model, the Modified Antoine equation and the
It is clear that good vapor-liquid equilibrium prediction is essential in order to obtain reliable
simulation results for the batch or continuous distillation. In the following, we present more
details about this important matter, which is ubiquitous in chemical separation science.
Vapor pressure of a pure fluid, a function of temperature only, is usually estimated using
Bi
ln Pi sat Ai Di ln T EiT Fi (11)
Ci T
Where Ai , Bi , Ci , Di , Ei , and Fi are fluid specific constants, and temperature is given in Kelvin
For mixtures at low pressures, the vapor phase can be approximated by an ideal gas mixture.
yi P xi Pi sat i (12)
140
Habilitation Dr. Housam Binous 2015
For the calculation of the activity coefficient 𝛾𝑖 of component i in the liquid phase, the Wilson
Nc Nc
xA
ln k ln x j Akj 1 Nc i ik (13)
j 1 i 1
x j Aij
j 1
Aij are the binary interaction parameters, which depend on the molar volumes ( vi and v j ) and
vj ii
Aij exp ij (14)
vi RT
This model allows good prediction of vapor-liquid equilibria as long as no phase splitting is
expected or present.
The Wilson interaction parameters 12 and 21 (expressed in kcal/mol) for each of the nine
binary mixtures, are obtained using PSO by fitting experimental VLE data taken from the
Dortmund Data Bank [15]). For every binary mixture, the isobaric VLE diagram and the VLE
curve (continuous lines) at P = 101 kPa are generated and compared to the experimental data
shown using the filled squares (see Figure 8 and Figure 9, respectively).
Conclusion
This chapter illustrated the use of particle swarm optimization technique to diverse applications
in the field of chemical engineering. The method is versatile as well as easy to grasp by
like programs e.g. MS Excel ™. Students can be productive in short time and focus on the
formulation of their chemical engineering research problem itself rather than the optimization
technique and its potential cumbersome coding. The technique is not limited to the algebraic
141
Habilitation Dr. Housam Binous 2015
applications, rather it can be used in ordinary differential equations (e.g., simple kinetic
models), partial differential equations (e.g., adsorption and catalysis kinetic parameter fitting).
parameters).
Figure 8 – Wilson’s activity coefficient model fitting of VLE data for the binary system:
chloroform and acetone at P = 101 kPa. PSO predictions are solid red curve and
142
Habilitation Dr. Housam Binous 2015
Figure 9 – Wilson’s activity coefficient model fitting of isobaric Txy data for the binary
system: 2-propanol and acetone at P = 101 kPa. PSO predictions are solid continuous
The students using computer algebra packages (e.g., MATHEMATICA©) are able to explore the
basic features of various optimization problems. This might include, but not restricted to:
1. The rate of convergence to the optimum using the parameters of PSO: inertia weights
143
Habilitation Dr. Housam Binous 2015
in standard textbooks.
References
1. Eberhart, R.C. and J. Kennedy. A new optimizer using particle swarm theory. in
Springer. p. 760-766.
1(5): p. 486-502.
5. Rini, D.P., S.M. Shamsuddin, and S.S. Yuhaniz, Particle swarm optimization:
144
Habilitation Dr. Housam Binous 2015
the calculation of phase stability analyses for nonreactive and reactive mixtures.
8. Ng, S.T., et al., Using Particle Swarm Optimization for Estimating Kinetics Parameters
2014.
10. Dai, W., et al. Particle Swarm Optimization algorithm for optimization of utility systems
11. Rosenbrock, H.H., An automatic method for finding the greatest or least value of a
12. Van de Vusse, J., Plug-flow type reactor versus tank reactor. Chemical Engineering
13. Sandler, S.I., Chemical, biochemical, and engineering thermodynamics. Vol. 4. 2006:
14. Wilson, G.M., Vapor-liquid equilibrium. XI. A new expression for the excess free
energy of mixing. Journal of the American Chemical Society, 1964. 86(2): p. 127-130.
15. Vapor-Liquid Equilibrium Data. Dortmund Data Bank 2014 Nov 13]; Available from:
www.ddbst.com/en/EED/VLE/VLEindex.php.
145
Habilitation Dr. Housam Binous 2015
Abstract
The simulation of the steady-state operations of a distillation unit requires the resolution of
the so-called MESH equations (i.e., Mass balance, Equilibrium, Summation and Enthalpy
equations). These MESH equations result in systems of nonlinear algebraic equations often
containing several hundreds of equations. These equations can be readily solved using the
built-in command of MATHEMATICA© called FindRoot. This command uses either the
secant method or Newton’s method. The author has successfully solved many distillation
problems (both binary and multicomponent systems) using this approach. Here, many
examples, including reactive and extractive distillation problems, are discussed thoroughly
Introduction
Several case studies are presented in this Chapter. We start with two industrially important
problems: the NLG distillation train and the extractive distillation technique applied to an
acetone – methanol mixture, which exhibits an azeotrope. Two additional problems concerning
binary and ternary mixtures of aromatic components (i.e., benzene – toluene – xylene) are
treated and illustrate simple yet important concepts of rigorous distillation simulations such as
the optimal feed location, heavy and light keys, and the graphical method called Ponchon-
Savarit technique. Then, three important applications of reactive distillation are treated
extensively using both MATHEMATICA© and Aspen-HYSYS®. The author concludes by giving
few insights regarding his experience using MATHEMATICA© to solve large systems of nonlinear
algebraic equations in matters of minutes, if not seconds, using any common modern personal computer.
146
Habilitation Dr. Housam Binous 2015
In a natural gas processing plant, the raw gas is pre-treated to remove liquid water and
condensate, on one side, and the accompanying impurities that may interfere with the following
separation processes and/or are undesirable in the final products, on the other side. Important
steps in this pre-treatment are acid gas and sulfur compounds removal (CO2, H2S, mercaptans,
etc.), dehydration and eventually mercury removal and finally nitrogen rejection if the N2
content of the gas is high. In the next separation unit, the demethanizer, the dry sweet gas is
cooled to separate the sales gas−containing mainly methane and less than 0.1 mol% pentane
and heavier hydrocarbons−from the rest, raw Natural Gas Liquids. NGL is a mixture of ethane
and heavier hydrocarbons (propane, butane, pentane, hexane and heptane) and constitutes an
important feedstock for petrochemical plants and facilities (steam crackers) producing olefins
and aromatics. As the components of the NGL often have higher sales value compared to the
gas itself [1] they are separated using a train of distillation columns in which the various
hydrocarbons are boiled off one by one [2, 3, 4]. The fractionation columns are called:
This first case study deals with the simulation of an NGL train. Raw NGL is to be separated in
almost pure ethane, i-butane, n-butane, and a mixture composed of the propane and the rest of
the dissolved ethane. The bottom product of the last fractionation unit is natural gasoline.
A feed, with a composition given in Table 1 where C7* is a pseudo-component that lumps
together all components heavier than n-hexane, is to be separated by an NGL train (see Figure
Deisobutanizer operating at decreasing pressures: 2600 kPa, 1700 kPa, 1000 kPa and 700 kPa,
147
Habilitation Dr. Housam Binous 2015
respectively. This pressure selection ensures, first, the separation of almost all components
heavier than ethane in the Deethanizer and allows, second, the usage of cooling water in the
First, the components are chosen from the software’s databank and the equation of state (i.e.,
PR EOS) is used as fluid package. Then, the distillation columns for the Deethanizer,
Depropanizer, Debutanizer and Deisobutanizer are added to the Process Flow Diagram. For
the Deethanizer column, when the feed stream specifications are entered and the required
column settings are selected properly, the column simulation can be performed and results are
obtained. Next, in order to decrease the high pressure of the bottom stream of the Deethanizer
(i.e., 2600 kPa), which serves as feed stream to the next column (i.e., the Depropanizer)
operating at 1700 kPa, an expansion valve is incorporated in order to reduce the pressure to
900 kPa.
Methane 0.005
Ethane 0.37
Propane 0.26
i-butane 0.072
n-butane 0.148
n-pentane 0.085
n-hexane 0.04
C7* 0.02
148
Habilitation Dr. Housam Binous 2015
i-Butane to
pipeline/
storage
Demethanizer
Propane to
Ethane to pipeline/
pipeline storage
NGL
treating
Deisobutanizer
Dehydration n-Butane to
pipeline/
storage
Deethanizer Depropanizer
Debutanizer
Natural
gasoline to
storage
149
Habilitation Dr. Housam Binous 2015
Similarly, simulations can be performed for the Depropanizer and later for the rest of the
distillation train (i.e., the Debutanizer and the Deisobutanizer). Process Flow Diagram (PFD)
In the four sections below, we present simulation results using MATHEMATICA©. This latter
approach has superior pedagogical and research benefits. Indeed, Aspen-HYSYS© is used as a
black box that does not disclose any of the governing equations and thermodynamic
relationships used. On the other hand, building all from starch using MATHEMATICA© can
be very rewarding and informative since one learns about equations of state, enthalpy departure
function for enthalpy, MESH (i.e., mass, equilibrium, summation and enthalpy) equations,
150
Habilitation Dr. Housam Binous 2015
considered below have 40 equilibrium stages and in all four cases, the feed is entering at stage
20 counting from the top. The idea, behind the choice of these configurations, is to provide
Let us start by giving the Mass and Enthalpy Balance Equations as follows:
F L f 1 V f 1 L f V f 0 (B-1)
F hF L f 1 h f 1 V f 1 H f 1 L f h f V f H f 0 (B-3)
Lk 1 Vk 1 Lk Vk 0 (B-4)
Lk 1 hk 1 Vk 1 H k 1 Lk hk Vk H k 0 (B-6)
V1 ( L0 D) 0 (B-7)
151
Habilitation Dr. Housam Binous 2015
V1 H 1 ( L0 D) h0 Q C 0 (B-9)
LN B VN 1 0 (B-10)
LN hN B hN 1 VN 1 H N 1 Q B 0 (B-12)
Figures 3-5 show the steady-state composition and temperature profiles as well as the liquid
and vapor flow rates inside the Deethanizer. Feed composition is indicated in Table 1 and feed
flow rate was chosen equal to 100 kmole/hr. In this operation ethane is separated out as
The reflux ratio and the mole % of ethane in the distillate have been set equal to 3 and 98.5,
respectively. The distillate product has a flow rate equal to 33.33 kmole/hr and contains: 1.5
mole % methane and 98.5 mole % ethane. The bottom product from Deethanizer enters into
the next column (i.e., the Depropanizer). The bottom product has a flow rate equal to 66.66
kmole/hr and its composition is as follows: 6.25 mole % ethane, 39.0 mole % propane, 10.8
mole % i-butane, 22.2 mole % n-butane, 12.75 mole % n-pentane, 6.0 mole % n-hexane and 3
mole % C7*. The cooling and heating duties of the condenser and reboiler are equal to 1.167
106 and 1.542 106 kJ/hr, respectively. Figures 3-5 show that the results of our
obtained with Aspen-HYSYS® shown by the open circles (i.e., ○). For both methods (i.e.,
calculations with MATHEMATICA© and Aspen-HYSYS®) and for all four separation units of
(a) yellow, green, cyan and blue curves correspond to methane, ethane, propane and
i-butane, respectively.
(b) red, magenta, purple and blue curves correspond to C7*, n-hexane, n-pentane
and n-butane, respectively.
Figure 3 – Steady-state composition profile for the Deethanizer
(♦ and ○ represent results using MATHEMATICA© and Aspen-HYSYS®)
153
Habilitation Dr. Housam Binous 2015
MESH equations for each one of the four separation units result in a set of 800 nonlinear
algebraic equations, which are readily solved using MATHEMATICA© in less than five
minutes with an Intel® Core™ 2 Duo CPU T9600 2.80 GHz RAM 4 GB.
The bottom stream of the Deethanizer serves as feed stream to the Depropanizer. The
overhead product of this column is propane rich and is condensed in the condenser. The
reflux ratio and the mole % of propane in the distillate have been set equal to 5 and 85.0,
respectively.
154
Habilitation Dr. Housam Binous 2015
Figure 6 shows the steady-state composition profile of this distillation column (i.e., the
Depropanizer). The distillate stream has a flow rate equal to 30.58 kmole/hr and contains: 13.64
mole % ethane, 85 mole % propane and 1.33 mole % i-butane. The bottom stream has a flow
rate equal to 36.07 kmole/hr and is composed of 18.82 mole % i-butane, 41 mole % n-butane,
23.55 mole % n-pentane, 11.08 mole % n-hexane and 5.55 mole % C7*. The cooling and heating
duties of the condenser and reboiler are equal to 2.429 106 and 2.348 106 kJ/hr, respectively.
155
Habilitation Dr. Housam Binous 2015
The bottom stream of the Depropanizer serves as feed stream to the Debutanizer. The reflux
ratio and the mole % of n-butane in the residue have been set equal to 5 and 0.1, respectively.
(a) yellow, green, cyan and blue curves correspond to methane, ethane, propane and
i-butane, respectively.
(b) red, magenta, purple and blue curves correspond to C7*, n-hexane, n-pentane
and n-butane, respectively.
Figure 6 – Steady-state composition profile for the Depropanizer
(♦ and ○ represent results using MATHEMATICA© and Aspen-HYSYS®)
156
Habilitation Dr. Housam Binous 2015
Figure 7 shows the steady-state composition profile of this distillation column (i.e., the
Debutanizer). In this column i-butane and n-butane are separated out as overhead product. The
distillate stream has a flow rate equal to 21.56 kmole/hr and contains: 31.48 mole % i-butane
and 68.51 mole % n-butane. The bottom stream has a flow rate equal to 36.07 kmole/hr and is
composed of 58.56 mole % n-pentane, 27.55 mole % n-hexane and 13.87 mole % C7*. The
cooling and heating duties of the condenser and reboiler are equal to 2.200 10 6 and 2.070 106
The distillate stream of the Debutanizer serves as a feed stream of the Deisobutanizer. Figure
8 gives the steady-state composition of i-butane and n-butane. The mole % of i-butane in the
residue and the mole % of n-butane in the distillate are both set equal to 1.
(a) yellow, green, cyan and blue curves correspond to methane, ethane, propane and
i-butane, respectively (yellow and green curves are hidden by cyan curve).
157
Habilitation Dr. Housam Binous 2015
(b) red, magenta, purple and blue curves correspond to C7*, n-hexane, n-pentane
and n-butane, respectively.
Figure 7 – Steady-state composition profile for the Debutanizer
(♦ and ○ represent results using MATHEMATICA© and Aspen-HYSYS®)
Figure 8 – steady-state composition profile inside the Deisobutanizer (orange and green
curves correspond to i-butane and n-butane, respectively;
♦ and ○ represent results using MATHEMATICA© and Aspen-HYSYS®)
158
Habilitation Dr. Housam Binous 2015
The distillate stream has a flow rate equal to 6.70 kmole/hr and contains: 98.98 mole % i-butane
and 1.01 mole % n-butane. The bottom stream has a flow rate equal to 14.86 kmole/hr and is
composed of 1 mole % i-butane and 98.99 mole % n-butane. The cooling and heating duties of
the condenser and reboiler are equal to 3.560 106 and 3.461 106 kJ/hr, respectively. Again,
For this case, we consider the same Depropanizer but add pressure drops equal to 10 kPa for
(a) yellow, green, cyan and blue curves correspond to methane, ethane, propane and
i-butane, respectively.
159
Habilitation Dr. Housam Binous 2015
(b) red, magenta, purple and blue curves correspond to C7*, n-hexane, n-pentane
and n-butane, respectively.
Figure 9 – Steady-state composition profile for the Depropanizer: case with pressure drop
effect
(♦ and ○ represent results using MATHEMATICA© and Aspen-HYSYS®)
The reflux ratio and the mole % of propane in the distillate have been set equal to 5 and 85.0,
respectively. Figure 9 shows the steady-state composition profile of this distillation column
(i.e., the Depropanizer). The pressure profile is depicted in Figure 10. The distillate stream has
a flow rate equal to 30.58 kmole/hr and contains: 13.61 mole % ethane, 85 mole % propane
and 1.37 mole % i-butane. The bottom stream has a flow rate equal to 36.07 kmole/hr and is
composed of 18.79 mole % i-butane, 41.01 mole % n-butane, 23.56 mole % n-pentane, 11.08
mole % n-hexane and 5.56 mole % C7*. The cooling and heating duties of the condenser and
reboiler are equal to 2.665 106 and 2.533 106 kJ/hr, respectively. Again, perfect agreement was
160
Habilitation Dr. Housam Binous 2015
Figure 10 – pressure profile inside the Depropanizer: case with pressure drop effects
Here, we consider the Deisobutanizer of section (d) with 40 stages all of which have a liquid
Murphree efficiency equal to 75 %. Murphree efficiency relates equilibrium and actual mole
fractions [5]. The reflux and reboil ratios are set equal to 10 and 4. The distillate stream has a
flow rate equal to 6.291 kmole/hr and contains: 87.02 mole % i-butane and 12.95 mole % n-
butane. The bottom stream has a flow rate equal to 15.271 kmole/hr and is composed of 8.53
mole % i-butane and 91.46 mole % n-butane. The cooling and heating duties of the condenser
and reboiler are equal to 1.290 106 and 1.190 106 kJ/hr, respectively. Again, perfect
161
Habilitation Dr. Housam Binous 2015
Extractive distillation (see Figure 12) is one of many techniques that are successfully applied
This method is usually more economical than heterogeneous azeotropic distillation and applies
even if the azeotrope is not pressure-sensitive. In this section, the author choose to study several
162
Habilitation Dr. Housam Binous 2015
Rectifying
section
E B A
(Entrainer)
Extractive Entrainer
Extractive
distillation A+E recovery
section
column column
A+B
(Binary Stripping
mixture) section
chlorobenzene (case study 2). Computations of residue curve maps (RCMs), pseudo-
equilibrium plots and isovolatility diagrams confirm that both solvents are suitable for
used as an entrainer, pure methanol will exit in the overhead stream of the extractive distillation
column. Steady-state simulations of the entrainer recovery column are also presented.
a partial reboiler, and a total condenser. The same column has been used by Luyben and Chien
[6], to separate acetone and methanol with water as an entrainer. Pure entrainer at 320 K is fed
to the column at a flow rate equal to 1100.03 kmole/hr at stage 24 (counting from the top). The
lower feed is composed of an equimolar mixture of acetone and methanol. The lower feed, at
320 K, is located at stage 39 (counting from the top) and has a flow rate equal to 540 kmole/hr.
163
Habilitation Dr. Housam Binous 2015
In Figures 13-15, we present the solution of the MESH equations (material, equilibrium,
summations, and heat). The steady-state composition profiles is shown in Figure 13 for a reflux
ratio and a distillate draw rate equal to 3.44 and 270.7 kmole/hr, respectively. The distillate is
almost pure acetone (99.15 mole % acetone) and the bottom is mainly composed of methanol
and water (19.69 mole % methanol and 80.31 mole % water). The condenser and reboiler heat
duties are 3.74664 107 kJ/hr and 4.15071 107 kJ/hr, respectively. For this problem, we have
solved a system of 399 nonlinear algebraic equations using MATHEMATICA© in less than 1.1
seconds. Finally, a comparison of the present calculation with Aspen-HYSYS® shows excellent
agreement (see Figures 13-15 where the open circles represent the results obtained with Aspen-
HYSYS®).
164
Habilitation Dr. Housam Binous 2015
Figure 15 – Steady-state liquid and vapor flow rate profile inside the extractive
distillation column for the water case
(♦ and ○ represent results using MATHEMATICA© and Aspen-HYSYS®)
165
Habilitation Dr. Housam Binous 2015
This comparison proves that one can perform sophisticated calculations using
MATHEMATICA©, a widely available computer software, even if one does not have access
HYSYS® or Aspen Plus®… Color panes were used, in Figures 13-15 and in the appropriate
figures thereafter, to indicate the different distillation column sections (i.e., the extractive,
rectifying and stripping sections indicated with the red, green and blue color panes,
respectively).
a partial reboiler, and a total condenser. A similar column (Luyben and Chien [6]) is used to
separate acetone and methanol with chlorobenzene as an entrainer. Pure entrainer at 320 K is
fed to the column at a flow rate, 1902.4 kmole/hr, at stage 16 (counting from the top). The
lower feed is composed of an equimolar mixture of acetone and methanol. The lower feed, at
320 K, is located at stage 32 (counting from the top) and has a flow rate equal to 540 kmole/hr.
The MESH equations (mass, equilibrium, summation, and heat) are solved and displayed in
Figure 10. The steady-state composition profile is given in Figure 16 for a reflux ratio and a
distillate flow rate equal to 1.55 and 271.2 kmole/hr, respectively. The distillate is almost pure
methanol (97.15 mole % methanol) and the bottom is mainly composed of acetone and
phenomenon takes place here since it is the intermediate-boiling component (i.e., methanol)
that exits in the overhead stream of the distillation column. The condenser and reboiler heat
duties are 2.38776 107 kJ/hr and 4.37609 107 kJ/hr, respectively. For this problem, we have
solved a system of 315 nonlinear algebraic equations using MATHEMATICA© in less than
2.24 second. Again, a comparison of the present calculation with Aspen-HYSYS® shows
166
Habilitation Dr. Housam Binous 2015
excellent agreement (see Figure 16 where the open circles represent the results obtained with
Aspen-HYSYS®).
reboiler, and a total condenser is considered. A similar column (Luyben and Chien [6]) is used
to separate methanol from water. In all their treatment, Luyben and Chien [6] used Aspen
Plus®. The residue of the extractive distillation column is fed to the entrainer (i.e., water)
recovery column at a flow rate, 1369 kmole/hr, at stage 13 (counting from the top). The MESH
equations (mass, equilibrium, summation, and heat) are solved and displayed in Figure 17. The
167
Habilitation Dr. Housam Binous 2015
steady-state composition profile is given in Figure 17 for a reflux ratio and a distillate flow rate
equal to 1.61 and 271.1 kmole/hr, respectively. The distillate is almost pure methanol (99.06
mole % methanol) and the bottom is mainly composed of chlorobenzene (99.99 mole %
chlorobenzene). The condenser and reboiler heat duties are 2.49642 107 kJ/hr and 2.60653 107
kJ/hr, respectively. For this problem, we have solved a system of 182 nonlinear algebraic
equations using MATHEMATICA© in less than 0.43 second. Again, a comparison of the
present calculation with Aspen-HYSYS® shows excellent agreement (see Figure 17 where the
168
Habilitation Dr. Housam Binous 2015
reboiler, and a total condenser, is considered in this section. A similar column (Luyben and
Chien, [6]) is used to separate acetone from chlorobenzene. The residue of the extractive
distillation column is fed to the entrainer (i.e., chlorobenzene) recovery column at a flow rate,
2169 kmole/hr, at stage 9 (counting from the top). The MESH equations (mass, equilibrium,
summation, and heat) are solved and displayed in Figure 18. The steady-state composition
profile is given in Figure 18 for a reflux ratio and a distillate flow rate equal to 3.52 and 271.2
169
Habilitation Dr. Housam Binous 2015
kmole/hr, respectively. The distillate is almost pure acetone (99.49 mole % acetone) and the
condenser and reboiler heat duties are 3.82424 107 kJ/hr and 5.01705 107 kJ/hr, respectively.
For this problem, we have solved a system of 126 nonlinear algebraic equations using
MATHEMATICA© in less than 2.47 seconds. Again, a comparison of the present calculation
with Aspen-HYSYS® shows excellent agreement (see Figure 18 where the open circles
Consider an ideal equimolar binary mixture composed of benzene and toluene at 760 mmHg.
This mixture, with a thermal quality equal to 0.5, is fed to a 10 stages column with total
condenser and partial reboiler. Feed stage is plate number 7. Feed flow rate is taken equal to
10 kmol/hr. Figure 19 and 20 show the temperature and composition profiles using a rigorous
approach, which includes both the energy and mass balance equations. All results found in the
temperature profiles. Distillate and bottom purities are found equal to 97.6 and 6.2 % mole
benzene, respectively. Figure 21 shows that the numerical simulation gives data similar to the
123864) and (0.062 , -78971) for the rectifying and stripping sections, respectively. The molar
flow rate profiles for both the liquid and vapor phases are also displayed in Figure 22. Non
Constant Molal Overflow calculation is justified by the fact that these flow rates are not
constant in both the rectifying and the stripping sections of the column. Calculations show that
for R=3 and S=2.5, stage 7 is the optimal feed location. Indeed, if one changes the feed location,
while keeping the same values for R and S, the cooling and heating duties will be higher (see
Table 3). It is the believe of the authors that optimal feed location is a concept difficult to grasp
170
Habilitation Dr. Housam Binous 2015
by students unless some kind of numerical proof is given, which can be done in class in a matter
of minutes using our MATHEMATICA® simulation. For this example, we have found cooling
and heating duties equal to Qc =-592708 kJ/hr and Qr = 435526 kJ/hr, respectively for R=3
and S=2.5. A simple observation of the Ponchon-Savrit diagram confirms that optimal feed
plate location is stage 7 since the line joining the operating points (shown by magenta dots)
and the feed point (indicated with a red dot) is just after stage 7. If one takes a close look at the
Ponchon-Savarit diagram and applies the lever-arm rule, he will clearly see that the feed’s
vapor fraction is 50%. Study of this binary distillation problem involves solving a system of
MATHEMATICA®. Thus, such calculations using any computer algebra along with the
simulation using Aspen-HYSYS® can be readily introduced in two one-hour lectures to junior
171
Habilitation Dr. Housam Binous 2015
This teaching strategy was tested successfully at KFUPM during the fall semester of 2010.
The possibility to move around sliders (see Figure 22) and select various values of the reboil
and reflux ratios are also put into advantage in this problem to discuss issues such as optimal
feed location, bottom and distillate compositions and difference point positions. Finally, how
segments) and tie lines (red segments) can be readily understood by students as depicted in
Figure 21.
172
Habilitation Dr. Housam Binous 2015
Consider a ternary mixture composed of benzene (23.33 % mole), toluene (33.33 % mole), and
p-xylene (43.34 % mole) at 101.325 kPa. This mixture, with a thermal quality equal to 1.0 (i.e.,
a saturated liquid), is fed to a 19-stage column with a total condenser and a partial reboiler. The
feed enters at stage 10. The feed flow rate is set equal to 1 kmol/hr. One can computes the
temperature (see Figure 23) and composition (see Figure 24) profiles using a rigorous
approach, which includes both the energy and mass balances thanks to MATHEMATICA®.
The blue, magenta, and brown curves, shown in the composition profile, correspond to
benzene, toluene, and p-xylene compositions, respectively. For R=4 and S=4, the results found
in the present calculation using MATHEMATICA® show perfect agreement with those given
colored dots.
173
Habilitation Dr. Housam Binous 2015
Figure 22 – Profile for liquid and vapor flow rates in blue and orange, respectively
174
Habilitation Dr. Housam Binous 2015
The temperature profiles for both binary (see Figure 19 of the previous section) and ternary
(see Figure 23) mixtures are similar and exhibit a monotonic increase going from condenser to
reboiler. On the other hand, for the composition profiles of the binary (see Figure 20 of the
previous section) and ternary (see Figure 24) mixtures, there are very clear differences. Indeed,
for the ternary system, the light-key, LK, and heavy-key, HK, are benzene and toluene,
and appears only in the residue. The LK component presents maxima in the composition profile
while no maximum is present for composition profile of the binary separation case. All
components must be present at the feed stage and there is a discontinuity in the composition
130
temperature in °C
120
110
100
5 10 15 20
stage number
175
Habilitation Dr. Housam Binous 2015
Finally, HNK composition goes through a plateau region. Heat and cooling duties were found
equal to Qc = -59817.1 kJ/hr and Qr = 60997.2 kJ/hr, respectively. The authors assumed
constant values for the liquid and vapor heat capacities of benzene, toluene and p-xylene,
profiles for any user-selected values of the reflux and reboil ratios as shown in Figure 25. The
calculation for this particular ternary separation problem involves solving 147 non-linear
Figure 24 – Composition profile for benzene (blue), toluene (magenta) and p-xylene (brown).
Aspen-HYSYS® and MATHEMATICA® results are shown by ● and ─, respectively.
176
Habilitation Dr. Housam Binous 2015
Figure 25 – Sliders allowing instantaneous calculation if the reboil and reflux ratios are
modified
5- Reactive distillation
usually handled in distinct devices. A simpler design integrating reaction and separation by
distillation or distillation/liquid extraction in the same unit is in many instances possible and
realized in one piece of equipment, a reactive distillation setup. The first applications of this
177
Habilitation Dr. Housam Binous 2015
With the continuous removal of reaction products, reactive distillation, when possible, offers
the advantages of higher yields, energy savings and reduced capital costs [10].
reaction section and a stripping section. A set of reactive trays or reactive packing is used as
reaction section.
Three applications of this innovative technology will be handled in the following: (1)
Methyl-t-butyl ether, MTBE, is used as gasoline additive to increase its octane rating. It also
allows the use of smaller amounts of lead alkyl compounds as anti-knock agents. It is further
found that with this additive, the emission of carbon monoxide and nitrous oxides on
combustion of gasoline is largely reduced [11]. The production reaction of methyl-t-butyl ether
takes place in the liquid phase in the presence of acidic ion-exchange resin as catalyst in a fixed
bed reactor [12]. A simplified flow sheet of the process is represented in Figure 26.
This case study deals with the simulation of the catalytic synthesis of methyl tert-butyl ether
using reactive distillation. The objective is to achieve a high purity product, 99 % or better.
178
Habilitation Dr. Housam Binous 2015
As a numerical example [13] we consider a reactive distillation column with 16 stages, a partial
reboiler (stage 17) and a total condenser (stage 0). The reactive section where the synthesis
reaction is taking place ranges from stage 3 to 9. The feed is a liquid mixture of 66.67% mole
of n-butane as inert solvent and 33.33% mole of i-butene on one side, and pure methanol on
the other side. It is introduced at stage 9. The operation pressure is set to 11.2 bars.
For the determination of the thermodynamic properties of the fluids and the calculation of the
liquid-vapor equilibria the Wilson model [14, 15, 16, 17] is used for the liquid phase and the
The details of the MATHEMATICA© model of the process can be found in reference [18]. We
just emphasize here some results of this interactive program. As expected, no MTBE is formed
for very low Damköhler numbers (molar hold-up near zero) and the only species exiting the
179
Habilitation Dr. Housam Binous 2015
As an exercise, the reader can run the MATHEMATICA© program to examine the effect of the
Finally, at high Damköhler numbers, almost pure MTBE (99% purity) is obtained as a bottom
product and a mixture mainly composed of n-butane (88 mol%) as distillate (Figure 27). This
composition is close to the binary nonreactive azeotrope mixture of methanol and n-butane.
Figure 28 presents a simplified flow sheet of the production of MTBE in a reactive distillation
column as simulated with Aspen-HYSYS® with similar operating conditions. As can be noted
by comparing Figure 27 and 28, the simulation results obtained using MATHEMATICA© are
in good concordance with those of Aspen-HYSYS®. In particular, the bottom product meets the
180
Habilitation Dr. Housam Binous 2015
Figure 28 – Aspen-HYSYS® PDF of the reactive distillation column for the production of
MTBE.
Besides its major application as gasoline additive, MTBE offers also a method of selectively
removing i-butene from a C4-mixture. Its subsequent decomposition by cracking yields highly
In the present case study we simulate a reactive distillation column that produces i-butene and
methanol from the decomposition of methyl tert-butyl ether (MTBE), the reverse reaction of
The reaction is taking place in the liquid phase in the presence of acid catalyst. Distilling the
decomposition products under reflux yields an overhead fraction composed mainly of i-butene
181
Habilitation Dr. Housam Binous 2015
The sieve tray column has 16 stages, a partial reboiler, and a total condenser. The column is
fed with pure MTBE at stage 8. The reactive stages range from 6 to 11.
An asymmetric thermodynamic approach (−) is used to predict the properties of the fluid
mixture and the calculation of the liquid-vapor equilibria: Wilson model for the liquid phase
and ideal gas law for the vapor phase. The operation pressure is set to 11.2 bars. The details of
the kinetics of this equilibrium limited reaction can be found in references [20, 21].
The MATHEMATICA© interactive simulation model of this reactive distillation column can
be found in reference [21]. Figures 29 and 30 are generated with this program. The reader is
encouraged to run the program in order to study the effect of the operating conditions on the
For low Damköhler numbers (low molar holdup), the conversion is low. Figure 29 shows the
composition profiles vs. stage number for the case of a molar holdup of 20 kmol. As can be
noted unconverted MBTE is found in the bottoms mixed with the product methanol.
Figure 29 – MTBE decomposition: Profiles of the compositions in the column in the case of a
182
Habilitation Dr. Housam Binous 2015
At high Damköhler numbers, we get almost pure methanol as bottoms and pure i-butene in the
distillate (Figure 30). As Figure 31 depicts, similar results are obtained using Aspen-HYSYS®.
Figure 30 – MTBE decomposition: Compositions in the for a molar holdup of 500 kmol.
Figure 31 – Aspen-HYSYS® model of the reactive distillation column for the decomposition of
183
Habilitation Dr. Housam Binous 2015
We consider at last case the disproportionation reaction, known also as a metathesis, of cis-2-
H3C
2
H3C CH3 + H3C CH3
CH3
The ternary mixture is subject to an equilibrium-limited chemical reaction with reaction rate
[22, 23]
x x
r 0.5 k xC2 A B
K eq
6000
k 3553.6 exp
RT
pressure of 3 atmospheres with 13 plates; the feed stage location is stage 5, the reactive stages
The following simulations are made with the simplifying assumption of constant molar
184
Habilitation Dr. Housam Binous 2015
The MATHEMATICA© interactive simulation model of this reactive distillation column can
be found in reference [24]. Figures 32, 33 and 34 illustrate some results of the calculations,
namely:
the composition profile vs. plate number for the three components of the ternary system
(A in red, B in blue, C in green and the reactive zone in light blue) (Figure 32),
the ternary diagram with the composition (mole fraction) of B vs. the composition of A.
The feed composition is shown by a dot and the reactive stages 2 to 7 are displayed (Figure
34).
It is worth noting that for high hold-up values (i.e., large Damköhler numbers) one recovers
the equilibrium case shown by the dashed curve in the ternary diagram.
We note again the good concordance between the simulations results of the
MATHEMATICA© program with those obtained with the Aspen-HYSYS® software (Figure
35).
Conclusion
Several distillation problems have been studied in this Chapter. A common thread to all these
problems is the need to write MESH equations and solve them using MATHEMATICA©.
Whenever possible the solutions have been compared to those obtained by the Aspen-HYSYS®
software. The author has advised three Master students at INSAT who worked on distillation
dynamics and control using MATHEMATICA© and Simulink®, reactive distillation using
185
Habilitation Dr. Housam Binous 2015
Figure 32 – Compositions in the metathesis column (Reflux ratio: 14; Molar holdup: 500
kmol).
186
Habilitation Dr. Housam Binous 2015
Figure 35 – Aspen-HYSYS® model of the reactive distillation column for the cis-2-pentene
187
Habilitation Dr. Housam Binous 2015
References
[1] Devold H. Oil and Gas Production Handbook, Edition 2.3, Oslo: ABB, 2010.
[2] Kidnay, A. J. & Parrish, W. R. Fundamentals of Natural Gas Processing, Boca Raton:
[3] Maddox R. N. & Erbar J. H. Gas Conditioning and Processing, Campbell Petroleum
[4] Younger, A. H. & Eng., P. Natural Gas Processing Principles and Technology, Part II,
[5] Wankat, P. C. Separation Process Engineering, 2nd Edition, Upper Saddle River:
[6] W. L. Luyben and I.-L. Chien, Design and Control of Distillation Systems for
Separating Azeotropes, Wiley, Hoboken, New Jersey (2010).
[7] Smith, L. A. US Patent 4 849 569, 1989; US Patent, 5 446 223, 1995.
[10] Process Modeling Using HYSYS With Chemical Industry Focus, ASPEN HYSYS
Documentation, 2004.
[11] Adams, J. M., Clement, D. E. & Graham, S. H. Clays and Clay Minerals, Vol. 30, No.
[12] Speight, J.G Chemical and Process Design Handbook, McGraw-Hill, Inc., 2002.
[13] Chen, F., Huss, R. S., Malone, M. F. & Doherty, M. F. "Simulation of Kinetic Effects
2472, 2000.
188
Habilitation Dr. Housam Binous 2015
[14] Wilson, G. M. “Vapor-Liquid Equilibrium XI: a new expression for the excess free
energy of mixing”, Journal of the American Chemical Society, Vol. 86, pp. 127-130,
1964.
[15] Sandler, S. I. Chemical Engineering Thermodynamics, 3rd Edition, John Wiley and
Sons, 1999.
[16] Poling, B. E., Prausnitz, J. M. & O’Connell, J. P. The properties of Gases and Liquids,
Computer Simulations, Boris Nemanjic and Navenka Svetozar Editors, Nova Science
[18] Binous, H., Selmi, M., Wada, I., Allouche, S. & Bellagi, A. “Methyl Tert-Butyl Ether
Project:
http://demonstrations.wolfram.com/MethylTertButylEtherMTBESynthesisWithAReac
tiveDistillationUn/
[20] Huang K. & Wang S. J. “Design and Control of a Methyl Tertiary Butyl Ether (MTBE)
Decomposition Reactive Distillation Column”, Ind. Eng. Chem. Res., 46(8), pp. 2508–
2519, 2007.
[21] Binous H., Selmi M., Wada I. Allouche S. & Bellagi A. “Methyl Tert-Butyl Ether
Project:
http://demonstrations.wolfram.com/MethylTertButylEtherMTBEDecompositionWith
AReactiveDistillation/
189
Habilitation Dr. Housam Binous 2015
5049–5068, 2005.
Othmer Encyclopedia of Chemical Technology, New York: John Wiley & Sons, 2004.
[24] Binous, H., Selmi, M., Wada, I. Allouche, S. & Bellagi, A. “Production of Cis2-Butene
Project,
http://demonstrations.wolfram.com/ProductionOfCis2ButeneAndCis2HexeneByCis2
PenteneDisproportion/
190
Habilitation Dr. Housam Binous 2015
Abstract
In this chapter, the author present four problems drawn from basic sciences (Physics,
Chemistry, Biology and Mathematics) and that illustrate several important aspects of
nonlinear dynamics such as chaotic and periodic behavior. These problems are: (1) the
Lorenz system, which is very useful in atmospheric science and the study of climate
predictions, (2) the Belousov–Zhabotinsky reaction, which is the first discovered reaction
that exhibits oscillatory behavior (i.e., the so-called chemical clock), (3) the predator and prey
system (i.e., a population dynamics example), and finally (4) the study of the behavior of
to these four problems, two problems related to the chemical engineering field are elucidate
(both involving autocatalytic reactions). Then, one problem from the field of biochemical
Introduction
Nonlinear dynamics is a multi-disciplinary field that covers not only Mathematics, but also
Physics, Chemistry and Biology. Here, the author shows few examples drawn from these four
scientific fields of study. After, three examples of nonlinear behavior in chemical and
biochemical engineering are presented. Several figures are displayed that illustrate important
concepts in nonlinear dynamics such as time-series plots, phase portrait diagrams, power
spectrum figures, Lorenz maps and bifurcation diagrams… All the calculations and
191
Habilitation Dr. Housam Binous 2015
𝑑𝑥
= 𝜎(𝑦 − 𝑥),
𝑑𝑡
𝑑𝑦
= 𝑟𝑥 − 𝑥𝑧 − 𝑦, and
𝑑𝑡
𝑑𝑧
= 𝑥𝑦 − 𝑏𝑧.
𝑑𝑡
Figure 1 – Power spectrum of the Lorenz system for 𝜎 = 16, 𝑟 = 45.92, and 𝑏 = 4.
For a particular selection of model parameters 𝜎, 𝑟, and 𝑏, you can observe periodic behavior,
period doubling, or chaotic behavior. With the help of MATHEMATICA©, one can illustrates
several important concepts of nonlinear dynamics, such as the time-series plot, the phase-space
diagram, the power spectrum, and the autocorrelation function plot. In addition, an estimate of
192
Habilitation Dr. Housam Binous 2015
Figure 2 – Phase portrait of the Lorenz system for 𝜎 = 16, 𝑟 = 45.92, and 𝑏 = 4.
Figure 3 – Time series of the Lorenz system for 𝜎 = 16, 𝑟 = 45.92, and 𝑏 = 4.
193
Habilitation Dr. Housam Binous 2015
For 𝜎 = 16, 𝑟 = 45.92, and 𝑏 = 4, you can observe chaotic behavior, which is confirmed by
the power spectrum diagram (see Figure 1). The phase-space diagram (see Figure 2) is that of
a strange attractor. In addition, the estimate of the maximum Lyapunov exponent is close to
1.453 (shown in Figure 3 – the time series plot or 𝑥(𝑡) vs. 𝑡). A positive Lyapunov exponent is
For 𝜎 = 19.8, 𝑟 = 56, and 𝑏 = 1, you can observe periodic behavior (see Figure 4 – the time
series plot or 𝑥(𝑡) vs. 𝑡), which is confirmed by the power spectrum diagram (see Figure 5).
The phase-space diagram, shown in Figure 6, is that of a limit cycle. In addition, the estimate
of the maximum Lyapunov exponent is approximately equal to zero. Thus, all Lyapunov
exponents are less than zero, which is a further indication of periodic behavior.
Figure 4 – Time series of the Lorenz system for 𝜎 = 19.8, 𝑟 = 56, and 𝑏 = 1.
194
Habilitation Dr. Housam Binous 2015
Figure 5 – Power spectrum of the Lorenz system for 𝜎 = 19.8, 𝑟 = 56, and 𝑏 = 1.
Figure 6 – Phase portrait of the Lorenz system for 𝜎 = 19.8, 𝑟 = 56, and 𝑏 = 1.
195
Habilitation Dr. Housam Binous 2015
Figure 7 shows the loci of the maxima for the time series 𝑧(𝑡) vs. 𝑡 (for 𝜎 = 16, 𝑟 =
45.92, and 𝑏 = 4). The maxima 𝑧𝑛 are readily found using the built-in function
extension gives the Lorenz map (see Figure 8), 𝑧𝑛+1 versus 𝑧𝑛 . By plotting the 𝑦 = 𝑥 reference
line in green, it becomes clear that the absolute value of the slope of the Lorenz map is greater
than 1. The Lorenz map shows that there is a well-defined relation between successive peaks.
Indeed, one can obtain a good estimate of the (𝑛 + 1)𝑡ℎ peak knowing the 𝑛𝑡ℎ peak. Thus, one
can conclude that the Lorenz system is predictable in the short term. In general, chaotic
solutions are predictable in the short term but unpredictable in the long term.
Figure 7 – The loci of the maxima for 𝑧(𝑡) vs. 𝑡 for 𝜎 = 16, 𝑟 = 45.92, and 𝑏 = 4.
196
Habilitation Dr. Housam Binous 2015
In this section, the oscillatory behavior of the Belousov–Zhabotinsky reaction (BZ reaction) is
shown. This reaction was modeled by Field, Körös and Noyes (FKN) in the early 1970s [2].
𝐴 + 𝑌 → 𝑋 + 𝑃,
𝑋 + 𝑌 → 2𝑃,
𝐴 + 𝑋 → 2𝑋 + 2𝑍,
2𝑋 → 𝐴 + 𝑃, and
1
𝐵 + 𝑍 → 2 𝑓 𝑍.
𝑃 = [𝐻𝑂𝐵𝑟].
197
Habilitation Dr. Housam Binous 2015
The law of mass action yields the following nondimensional 3×3 system of nonlinear ODEs:
𝑑𝑥
𝜖 𝑑𝜏 = 𝑞𝑦 − 𝑥𝑦 + 𝑥(1 − 𝑥),
𝑑𝑦
𝛿 = −𝑞𝑦 − 𝑥𝑦 + 𝑓𝑧, and
𝑑𝜏
𝑑𝑧
=𝑥−𝑧
𝑑𝜏
𝑋 𝑌 𝑍
where 𝑥 = 𝑋 , 𝑦 = 𝑌 , and 𝑧 = 𝑍 ; ϵ = 4 × 10−2, 𝛿 = 4 × 10−4 , 𝑞 = 8 × 10−4 and 𝑓 ∈
0 0 0
[0.5024,2.41].
When the 𝐶𝑒(𝐼𝑉) concentration reaches a maximum the solution turns blue (see Figure 9). This
concentration oscillates and when it is very low, the solution in the beaker turns red (see Figure
10).
198
Habilitation Dr. Housam Binous 2015
199
Habilitation Dr. Housam Binous 2015
Let assume that 𝜖 = 0.0099, 𝛿 = 2.4802 × 10−5 , 𝑞 = 3.1746 × 10−5 , and 𝑓 ∈ [1.0,3.0].
There is periodic behavior for 𝑓 ≤ 2.41, which is confirmed by (1) the sustained oscillations
in the time series plot in Figure 11 and (2) the limit-cycle in the time-delay reconstruction
diagram in Figure 12 (both plots are shown in red). This latter diagram is obtained by plotting
𝑥𝑡+𝜅 versus 𝑥𝑡 with 𝜅 arbitrarily taken equal to 0.01. A transition occurs at approximately 𝑓 =
2.42 with time series exhibiting damped oscillation for 𝑓 ≥ 2.42 (plot shown in blue in Figure
13) and the time-delay reconstruction diagram collapsing to a single blue point (see Figure 14).
200
Habilitation Dr. Housam Binous 2015
201
Habilitation Dr. Housam Binous 2015
This section describes the predator-prey model with two species: sharks and sardines. In this
population dynamics example, sharks prey on sardines. Sardines feed on other abundant
species. Thus, in the absence of predators, the sardine population, 𝑥(𝑡), grows at an exponential
rate. In the absence of sardines, the sharks die out also at a rate proportional to their population
size (i.e., sharks population or 𝑦(𝑡) is a decaying exponential). The number of encounters
between the species is proportional to the product of their populations or 𝑥(𝑡) × 𝑦(𝑡). These
Thus, the shark and sardine populations, 𝑥(𝑡) and 𝑦(𝑡) obey the following nonlinear system of
equations:
𝑑𝑥
= 𝑎𝑥 − 𝑏𝑥𝑦
𝑑𝑡
202
Habilitation Dr. Housam Binous 2015
𝑑𝑦
= −𝑐𝑦 + 𝑑𝑥𝑦
𝑑𝑡
For specific values of these four parameters (e.g., 𝑎 = 𝑏 = 𝑐 = 𝑑 = 1), one can observe a limit
cycle (see Figure 15) and periodic oscillations of the species population sizes (i.e., 𝑥(𝑡) and
𝑦(𝑡)) as shown in Figure 16. This behavior was first observed by the Hudson Bay Company,
which counted the number in thousands of snowshoe hares and lynxes in Canada from 1845
203
Habilitation Dr. Housam Binous 2015
.
𝑥 = 𝑎1 𝑥 + 𝑏1 𝑦, and
.
𝑦 = 𝑎2 𝑥 + 𝑏2 𝑦.
𝑎1 𝑏1
and define the matrix 𝑀 = ( ).
𝑎2 𝑏2
In this section, plots the phase portrait and the vector field of directions around the critical point
(0,0) are displayed in Figure 17-20 when the steady-state is a stable node, a center, a stable
This steady-state is indicated by the green dot in the phase plane diagram. In addition, the
eigenvalues of 𝑀, the trace 𝑡𝑟(𝑀), the determinant |𝑀|, and 𝛥 = tr 2 (𝑀) − 4 |𝑀| are
computed.
A plot of the curve obeying the equation 𝛥 = 0 is obtained if you select the stability character
tab (see Figure 21). This parabola is indicated in solid and dashed black curves for positive and
204
Habilitation Dr. Housam Binous 2015
205
Habilitation Dr. Housam Binous 2015
Figure 21 – The six regions giving the stability character of the steady-state (0, 0)
206
Habilitation Dr. Housam Binous 2015
The stability character of the steady-state is obtained by looking at the position of the black dot
relative to the different colored regions (numbered 1–6). The various possibilities are given in
Table 1.
System
207
Habilitation Dr. Housam Binous 2015
In the study of nonlinear dynamics, it is useful to first study simple systems that exhibit periodic
Steady-state is shown in green. Initial point is shown in cyan. Trajectory is shown in red.
These equations describe the autocatalytic reaction of two intermediate species 𝑥 and 𝑦 in an
isothermal batch reactor, when the system is far from equilibrium [1]. In this context, the steady
208
Habilitation Dr. Housam Binous 2015
state referred to below is a pseudo steady state, and is applicable when the precursor reactant
The unique steady state is given by 𝑥𝑆 = 𝑏 and 𝑦𝑆 = 𝑏⁄(𝑎 + 𝑏 2 ). This steady state is at the
position of the green dot in the phase portrait diagram (see Figure 22). It appears as the
intersection of the dotted blue and green curves, which are the level curves given by 𝑓1 (𝑥, 𝑦) =
0 and 𝑓2 (𝑥, 𝑦) = 0.
209
Habilitation Dr. Housam Binous 2015
210
Habilitation Dr. Housam Binous 2015
The stability of steady state to small disturbances can be assessed by determining the
A Hopf bifurcation occurs when a complex conjugate pair of eigenvalues crosses the imaginary
axis.
Figure 1 shows the phase portrait with the vector field of directions around the critical point
(𝑥𝑆 , 𝑦𝑆 ). In addition, the eigenvalues of 𝐽, the trace 𝑡𝑟(𝐽), the determinant |𝐽| and 𝛥 = tr(𝐽))2 −
4 |𝐽| are computed. The time series are also plotted in a separate diagram (see the blue and red
211
Habilitation Dr. Housam Binous 2015
Finally, the real and imaginary parts of the two eigenvalues 𝜆1 and 𝜆2 are plotted (in blue and
red, respectively) versus 𝑏 for user-specified values of the parameter 𝑎 (see Figure 24-25).
To observe a Hopf bifurcation, you can set 𝑎 = 0.08 and vary 𝑏. As you can see in the Figures
26-28, there are two Hopf bifurcation points. Indeed, for 𝑏 = 0.3, the steady state is a stable
focus (see Figure 26); for 𝑏 = 0.5, the trajectory (see Figure 27) is attracted to a stable periodic
solution (called a limit cycle); and finally for 𝑏 = 1.0, the steady state is again a stable focus
(see Figure 28). For 𝑎 = 0.08, the two Hopf bifurcation points are obtained at 𝑏 =
Finally, Figure 29 (obtained for larger values of 𝑎, for instance, 𝑎 = 0.65), the Hopf bifurcation
disappears and there are only transitions from a stable node to a stable focus and then back to
stable node.
212
Habilitation Dr. Housam Binous 2015
213
Habilitation Dr. Housam Binous 2015
dynamical behavior [6]. Indeed, period doubling and chaos are found when the bifurcation
The following reaction system is theoretical. The autocatalator's steps are the following:
reaction rate
𝑷→𝑨 𝑘0 𝑝0
𝑷+𝑪→𝑨+𝑪 𝑘𝑐 𝑝0 𝑐
𝑨→𝑩 𝑘𝑢 𝑎
𝑨 + 𝟐𝑩 → 𝟑𝑩 𝑘1 𝑎𝑏 2
𝑩→𝑪 𝑘2 𝑏
𝑪→𝑫 𝑘3 𝑐
Here 𝑃 is a chemical precursor with constant concentration, 𝐷 is the final product, 𝐴, 𝐵, and 𝐶
are intermediate chemical species, 𝑘0 , 𝑘𝑐 , 𝑘𝑢 , 𝑘1 , 𝑘2 , and 𝑘3 are rate constants for the reactions,
autocatalytic reaction is the following step: 𝐴 + 2𝐵 → 3𝐵, with 𝐵 catalyzing its own
formation. This step introduces the nonlinear term 𝑘1 𝑎𝑏 2 in the governing equations that is
The rate equations for the three intermediate species are usually written in the form:
𝑑𝑎
= 𝑘0 𝑝0 + 𝑘𝑐 𝑝0 𝑐 − 𝑘𝑢 𝑎 − 𝑘1 𝑎𝑏 2 ,
𝑑𝑡
214
Habilitation Dr. Housam Binous 2015
𝑑𝑏
= 𝑘𝑢 𝑎 + 𝑘1 𝑎𝑏 2 − 𝑘2 𝑏, and
𝑑𝑡
𝑑𝑐
= 𝑘2 𝑏 − 𝑘3 𝑐.
𝑑𝑡
𝑑𝛼
= 𝜅 + 𝜇𝛾 − 𝛼𝛽 2 − 𝛼,
𝑑𝜏
𝑑𝛽
𝜎 𝑑𝜏 = 𝛼𝛽 2 + 𝛼 − 𝛽, and
𝑑𝛾
𝛿 𝑑𝜏 = 𝛽 − 𝛾.
Here 𝛼, 𝛽, and 𝛾 are the dimensionless concentrations of 𝐴, 𝐵, and 𝐶, and the four parameters
𝜅, 𝜇, 𝛿, and 𝜎 depend on the rates of the individual reactions and the concentration of the
precursor.
Figure 30 – Time series of the three-variable autocatalator for for 𝜇 = 0.1 (period one
behavior).
The figures below illustrates the dynamics of the concentrations α, β, and γ for various values
of the bifurcation parameter 𝜇. The time series option gives a plot of log β versus time (see
215
Habilitation Dr. Housam Binous 2015
Figure 30 – here we see periodic behavior for 𝜇 = 0.1), while the phase-space option gives a
three-dimensional parametric plot of 𝑙𝑜𝑔 𝛼 − 𝑙𝑜𝑔 𝛽 − 𝑙𝑜𝑔 𝛾 (see Figure 31 – here we see a
One should try the following values of μ: 0.1, 0.14, 0.15, 0.152, and 0.157 to observe period 1,
216
Habilitation Dr. Housam Binous 2015
Figure 32 – Time series of the three-variable autocatalator for for 𝜇 = 0.14 (period 2
behavior).
Figure 33 – Time series of the three-variable autocatalator for for 𝜇 = 0.15 (period 4
behavior).
217
Habilitation Dr. Housam Binous 2015
Figure 34 – Time series of the three-variable autocatalator for for 𝜇 = 0.152 (period 6
behavior).
Figure 35 – Time series of the three-variable autocatalator for for 𝜇 = 0.157 (period 5
behavior).
218
Habilitation Dr. Housam Binous 2015
Figure 36 – Time series of the three-variable autocatalator for for 𝜇 = 0.153 (chaotic
behavior).
219
Habilitation Dr. Housam Binous 2015
For μ = 0.153, chaos is obtained and the 𝛼 − 𝛽 − 𝛾 phase-space graph is that of a strange
attractor (see Figures 36-37). The loci of the maxima is given in Figure 38 and this calculation
is the basis of the determination of the bifurcation diagram (see Figure 39). When μ is large
enough, you can observe a reversed sequence leading back to period 1 behavior. These results
are confirmed by the bifurcation diagram (a remerging Feigenbaum tree) given in Figure 39.
(period 4 behavior).
220
Habilitation Dr. Housam Binous 2015
Figure 40 plots the loci of the maxima and minima of the time series 𝛼(𝑡). Such information
is used to map the next maximum or minimum point. For particular values of the bifurcation
parameter 𝜇 (for example, 𝜇 = 0.154), the result is chemical chaos, and a map of the next
maximum/minimum shows the typical behavior of a chaotic attractor (see Figure 41-42). On
221
Habilitation Dr. Housam Binous 2015
the other hand, when periodic behavior is observed (e.g., for 𝜇 = 0.114), a map of the next
maximum/minimum displays only a small number of isolated points (see Figure 43-44).
Figure 40 – Loci of the maxima (red points) and minima (green points) of the three-variable
autocatalator for 𝜇 = 0.145 (period 2 behavior).
222
Habilitation Dr. Housam Binous 2015
223
Habilitation Dr. Housam Binous 2015
224
Habilitation Dr. Housam Binous 2015
In 1942, Monod proposed the following form of the specific growth coefficient:
𝜇max 𝑥2
𝜇Monod =
𝑘𝑚 + 𝑥2
The specific growth coefficient with the substrate inhibition (labeled SI in the text thereafter)
is given by
𝜇max 𝑥2
𝜇SI = .
𝑘𝑚 + 𝑥2 + 𝑘1 𝑥22
𝑑𝑥1
= (𝜇 − 𝐷)𝑥1, and
𝑑𝑡
𝑑𝑥2
= 𝐷(𝑥2,𝑓 − 𝑥2 ) − 𝜇 𝑥1 ⁄𝑌,
𝑑𝑡
where 𝑥1 is the biomass concentration, 𝑥2 is the substrate concentration, 𝐷 is the dilution rate,
𝑌 is the yield, 𝑥2,𝑓 is the feed substrate concentration, and 𝜇 is the specific growth coefficient.
The steady states are the solutions of the following system of equations:
(𝜇 − 𝐷)𝑥1 = 0,
𝐷(𝑥2,𝑓 − 𝑥2 ) − 𝜇 𝑥1 ⁄𝑌 = 0.
The trivial solution is obtained for 𝑥1 = 0 and 𝑥2 = 𝑥2,𝑓 . This corresponds to a situation where
there are no cells left in the reactor, a phenomena called wash out.
Here, the author finds the nontrivial steady states and shows the bifurcation diagram (𝑥2 versus
For the Monod case (see Figure 45-46), there is a single nontrivial steady state if 𝐷 < 𝜇max .
This steady state is stable. On the other hand, the trivial steady state is either stable (when 𝐷 >
225
Habilitation Dr. Housam Binous 2015
Figure 45 – Bifurcation diagram for 𝐷 > 𝜇max – a single stable steady-state – the wash out
case
Figure 46 – Bifurcation diagram for 𝐷 < 𝜇max – two steady-states – the wash out is an
unstable steady-state – the nontrivial steady-state is stable
226
Habilitation Dr. Housam Binous 2015
For the SI model, there are two nontrivial steady states if the value of 𝑫 is in the pink region
Figure 47 – Bifurcation diagram – two nontrivial steady-states (magenta and cyan dots
correspond to the stable and unstable steady-states) – the wash out (green dot) is a stable
steady-state
227
Habilitation Dr. Housam Binous 2015
𝒅𝝁
(low value of 𝒙𝟐 ) is stable because 𝒅𝒙 > 𝟎. The cyan dot (intermediate value of 𝒙𝟐 ) is unstable
𝟐
𝒅𝝁 𝝁𝐦𝐚𝐱 𝒙𝟐,𝒇
(a saddle point) because < 𝟎. The trivial solution is either stable (𝑫 > 𝒌 𝟐 ) or
𝒅𝒙𝟐 𝒎 +𝒙𝟐,𝒇 +𝒌𝟏 𝒙𝟐,𝒇
𝒅𝝁
indicated by the magenta dot (low value of 𝒙𝟐 ). This steady state is stable because > 𝟎.
𝒅𝒙𝟐
The other value of 𝒙𝟐 verifies 𝒙𝟐 > 𝒙𝟐,𝒇 , thus 𝒙𝟏 < 𝟎 (i.e., this solution is not feasible). Finally,
𝛍𝐦𝐚𝐱
if 𝑫 > , nontrivial solutions are not possible.
(𝟏+𝟐√𝐤𝟏𝐤𝐦)
Let us see how the phase portrait looks like for both the Monod and SI models. Figure 5-6 are
relevant to the Monod case where there are two steady-states: (1) the trivial steady-state or the
so-called wash out case shown by the green dot in Figure 49, and (2) the nontrivial steady-state
shown by the blue dot in the same figure. Figure 50 shows that the trivial steady-state is
unstable (i.e., one eigenvalue is real and positive) while the nontrivial steady-state is stable
(i.e., both eigenvalues are real and negative). Figure 51-52 are relevant to the SI model where
there are three steady-states: (1) the trivial steady-state or the so-called wash out case shown
by the green dot in Figure 51, and (2) two nontrivial steady-state shown by the blue and red dot
in the same figure. Figure 52 shows that the trivial steady-state is stable (i.e., both eigenvalues
are real and negative) while only one of the two nontrivial steady-states is stable (i.e., both
eigenvalues are real and negative). It is worth mentioning at this point that all the results
obtained concerning the stability of the steady-states using eigenvalues calculations (see Figure
6 and 8) are consistent with the conclusions found previously from the bifurcation diagrams.
228
Habilitation Dr. Housam Binous 2015
Figure 49 – Phase portrait for the Monod model with two steady-states
The nontrivial steady-state is indicated by the blue dot
Figure 50 – List of the steady-states and their eigenvalues for the Monod model
229
Habilitation Dr. Housam Binous 2015
Figure 52 – List of the steady-states and their eigenvalues for the SI model
230
Habilitation Dr. Housam Binous 2015
Conclusion
In the present Chapter, author showed many calculations typical of nonlinear dynamics and
chaos, period doubling, limit cycles, Lorenz map and next maxima plots, Hopf bifurcations and
illustrated by several examples such as the Lorenz attractor, the Oregonator... All these
concepts have been taught for the past two years by the author in the KFUPM graduate-level
course on nonlinear dynamics in chemical and biochemical systems called CHE 565.
References
Oscillations, Waves, Patterns, and Chaos, New York: Oxford University Press, 1998.
[5] P. Gray and S. K. Scott, Chemical Oscillations and Instabilities: Non-Linear Chemical
[6] B. Peng, S. K. Scott, and K. Showalter, "Period Doubling and Chaos in a Three-Variable
[7] B. W. Bequette, Process Dynamics, Modeling, Analysis, and Simulation, Upper Saddle
231
Habilitation Dr. Housam Binous 2015
In the present report, a selection of state-of-the-art numerical techniques have been presented.
A special emphasis was have been placed on their application to solve chemical engineering
problems in areas such as transport phenomena, distillation, chemical reaction engineering and
using Matlab® or Maple. It is my belief that the code developed for the solution of these
advanced chemical engineering problem can help both graduate students as well as
The perspectives of my contribution are numerous. Indeed, I plan to solve other advanced
chemical engineering problems using the methodology explained in this report. Whenever
possible, comparison with commercial software such as gPROMS and COMSOL will be
performed. I also hope to learn additional techniques such as the finite element method, the
Monte Carlo and the molecular dynamics methods in the next couple of years and to implement
232