You are on page 1of 262

Kinetics of adsorption from liquid phase on activated

carbon
Kouyoumdjiev, M.S.

DOI:
10.6100/IR387873

Published: 01/01/1992

Document Version
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)

Please check the document version of this publication:

• A submitted manuscript is the author's version of the article upon submission and before peer-review. There can be important differences
between the submitted version and the official published version of record. People interested in the research are advised to contact the
author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication

Citation for published version (APA):


Kouyoumdjiev, M. S. (1992). Kinetics of adsorption from liquid phase on activated carbon Eindhoven:
Technische Universiteit Eindhoven DOI: 10.6100/IR387873

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal ?
Take down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.

Download date: 02. may. 2018


KINETICS OF ADSORPTION
FROM LIQUID PHASE
ON ACTIVATED CARBON
KINETICS OF ADSORPTION
FROM LIQUID PHASE
ON ACTIVATED CARBON

PROEFSCHRIFf

ter verkrijging van de graad van doctor aan de


Technische Universiteit Eindhoven, op gezag van
de Rector Magnificus, prof. dr. J.H. van Lint,
voor een commissie aangewezen door het College
van Dekanen in het openbaar te verdedigen op
vrijdag 18 december 1992 om 16.00 uur

door

Marcho Stefanov Kouyoumdjiev


geboren te Sofia

druk: wibro dissertatiodrukkerij, helmond.


Dit proefschrift is goedgekeurd door de promotoren:

prof.dr.ir. P.J.A.M. Kerkhof


prof.dr. R. Krishna
English translation of the official text on the title page

KINETICS OF ADSORPTION FROM LIQUID


PHASE ON ACTIVATED CARBON

THESIS

to obtain the degree of doctor at the Eindhoven


University of Technology by the authority of the
Rector Magnificus, prof.dr. J.H. van Lint, to
be defended in public in the presence of a
committee nominated by the council of Deans on
Friday, December 18th 1992 at 16:00 hrs

by

Marcho Stefanov Kouyoumdjiev


bom in Sofia
This thesis has been approved by the promoters:

prof.dr.ir. P.J.A.M. Kerkhof


prof.dr. R. Krishna
To Selia, to my parents
ACKNOWLEDGMENTS

The research work presented in this thesis was carried out in the
Laboratory of Chemica! Process Technology at the Eindhoven University of
Technology. I would like to thank all memhers of this laboratory, working
in the Fr-hal, for their support and contributions to this thesis.
I would like to express my gratitude especially to prof. Kerkhof for the
opportunity he gave me to work at the Eindhoven University of Technology
and for the regular discussions and valuable comments during my work on the
thesis. I am particularly indebted to Marlus Vorstman for his kindness and
help in the four years of the research study. The on-line constructive
discussions which I had almost daily with Marlus were an invaluable souree
of ideas and a solid support during my work.
The assistance of all the technica! staff of the Fr-hal is greatly
appreciated and I want to mention especially Chris Luyk who contributed
substantially to the construction of the experimental installation.
Significant contributions to this work were made by the graduale students
Sandra Vedder, Paul Steenbergen, Paul Laimböck, Heinie Voncken, Marc
Donker and Marco Ligthart, for which I sincerely thank them.
During the course of this thesis many other people have helped or advised
me about my work and have contributed directly or indirectly to its
successful completion and I want to mention especially Toine Ketelaars,
Gerben Mooiweer, S. Rienstra and Kostadin Paev.
Thanks are due to Norit BV for the more-than-enough supply of activaled
carbon with which I performed all experiments described in this thesis. I
would also like to thank Wim van Lier for the valuable advise and remarks
during our frequent discussions.
Last but not least I want to thank Unilever Research Laboratorium in
Vlaardingen for the onderstanding and technica! support during the last
weeks of the completion of this thesis.

Marcho Kouyoumdjiev
22 October 1992
SUMMARY

Adsorption is one of the most important processes used in industry for the
separation of solutes from a fluid stream. To design adsorption equipment it
is very important to know the adsorption capacity and the rate of adsorption.
Usually the capacity is represented by an isotherm based on measured data. The
rate of adsorption depends, among others, on the rate of transport to the
outer surface of the carbon particles and also on the rate of transport inside
the particles.

In this thesis the kinetics of adsorption on activated carbon are investigated


and single-solute kinetic parameters are used to describe (predict)
multisolute kinetics. This is achieved by studying the overall rate of
adsorption of several organic compounds from dilute aqueous solutions on
granular activaled carbon in a batch system with special focus on the
dependenee of diffusion inside the partiele on initial organic compound(s)
concentration, amount of carbon added, type of organic compound used, type of
adsorption system (single- or two-solute) and partiele size (for a limited
number of experiments). Several mathematica! models are developed and the
simulation results are compared to experimental data.

In order to obtain experimental data kinetic experiments were performed. This


was done by means of adding a certain amount of carbon to a fixed volume of
solution and following the concentration change of the component(s) in time.
The shape of the concentration decay curve curve depends on several factors,
some of which (mentioned in the previous paragraph) were chosen for variations
in this study.

Three single-solute systems (4-isopropylphenol, p-nitroaniline and


nitrobenzene) and the combined two solutes systems (4-isopropylphenoV
p-nitroaniline and nitrobenzene/p-nitroaniline) are used. These compounds were
chosen because they are representative for pollutants occurring in industrial
waste water streams and because their concentrations could be accurately
measured by the analytical metbod used (UV-spectrophotometry).
Equilibrium experiments were also performed, for all systems under
investigation. The single-solute data is successfully fitted with a
Radke-Prausnitz isotherm. An improved version of the Fritz & Schlünder
isotherm equation, proposed in this study, is correlated to the two-solute
equilibrium data and is used for calculating local equilibrium in the
multisolute kinetic models.

Two basically different diffusion models are developed to predict the behavior
of the investigated adsorption systems under different conditions. Both models
take into account an extemal film resistance and inside the partiele
two-intraparticle diffusion mechanisms in parallel: pore liquid diffusion and
surface diffusion. The rate of the adsorption step is considered fast compared
to the rates of the diffusional steps and local equilibrium is assumed
between pore liquid and adsorbent.

In literature the intra-partiele kinetics of adsorption has been described


most often with a Fick diffusion equation which is the basis of one of the
models used in this study.
Another way of descrihing intra-partiele kinetics of adsorption is based on
the generalized Maxwell-Stefan formulation in which the gradient of the
chemical potential is taken as the driving force for diffusion. Such a model,
derived for adsorption from the gas phase, was recently proposed by Krishna
and has already been successfully applied by Hu and Do. The main advantage
compared to the Fickian description is, apart from being thermodynarnically
correct, the potential to apply single-solute diffusivities in
multisolutekinetics. The second basic model used in this thesis extends the
model of Kris~a to liquid phase adsorption.

Apart from the two basic diffusion models a third, more simple model was also
used in this thesis. In this model mass transport inside the partiele is
modeled by means of a constant effective diffusivity. This constant
diffusivity is a lumped parameter and does not depend on the driving force
formulation (Fickian or Maxwell-Stefan). Although this model is clearly far
from the real physical picture inside the carbon particle, it is quite often
encountered in literature and therefore was used for comparison with the other
models in this study.
The models were solved numerically by using a fmite difference method. All
model results were compared to the respective experimental concentratien decay
curves with the help of an optimization procedure and from the best-fit model
curves the values of the different types of diffusivities were determined.

In the case of single-solute systems model simulations showed that the spread
in diffusivity values for the constant diffusion model is too big for the
model to have any predictive value. The Fickian model showed a relative
standard deviation of the average diffusivity value similar to the one for the
constant diffusivity model which indicates that this model has no predielive
value either.
The Maxwell-Stefan model, applied for the first time to adsorption from liquid
phase, showed a significantly smaller spread around the average diffusivity
value than the other two models. This means that the model bas a real
predictive value for the single-component adsorption systems and may have the
potential to apply single-solute diffusivities in multicomponent kinetics.

In the case of two-solute systems the constant diffusion model fits reasonably
well the experimental 4-isopropylphenol/p-nitroaniline results and to a lesser
extent the nitrobenzene/p-nitroaniline ones. It can be used for simulations,
however, as the values of the constant diffusivities clearly deviate from the
single-solute values two-solute experiments will be needed to make model
predictions.
The multisolute Maxwell-Stefan model gives a fair predietien of the
experimental concentration decay curves and gives ranges for the diffusivity
values close to the single-solute ones for components 4-isopropylphenol and
p-nitroaniline. The diffusivity range for nitrobenzene, although not large, is
rather different from the single-solute diffusivity range. This model can be
used to predict fairly well multisolute kinetics based on single-solute
diffusivity data.
SAMENVATTING

Adsorptie is een van de belangrijkste processen voor het (af)scheiden van


opgeloste stoffen uit vloeistofstromen. Voor het ontwerp van
adsorptieapparatuur is kennis omtrent de adsorptiecapaciteit en de snelheid
van adsorptie een vereiste. Gewoonlijk wordt de capaciteit beschreven door een
adsorptie-isotherm welke verkregen is uit metingen. De snelheid van adsorptie
wordt onder meer bepaald door de transportsnelheid aan de buitenzijde van de
absorbensdeeltjes en van de transportsnelheid binnen de deeltjes.

In dit proefschrift wordt een studie beschreven naar de kinetiek van adsorptie
op aktieve kool met als doel te komen tot een zodanige beschrijving van de
kinetiek dat de difffusiecoëfficiënten verkregen uit experimenten met één
opgeloste component gebruikt kunnen worden voor het voorspellen van de van de
kinetiek in meercomponenten systemen. Daartoe zijn ladingsgewijs metingen van
de adsorptiesnelheid aan granulaire aktieve kool uitgevoerd vanuit verdunde
oplossingen van verschillende organische componenten in water. Hierbij wordt
gekeken naar de invloed van een aantal factoren op de diffusie binnen het
deeltje: de beginconcentratie(s) van de opgeloste component(en), de
hoeveelheid toegevoegde kool, soort van organische component, keuze van één-
of tweecomponentensysteem en - bij een beperkt aantal experimenten- de
deeltjesgroottte . Er zijn verschillende wiskundige modellen ontwikkeld en de
uitkomsten van de berekeningen met deze modellen zijn vergeleken met de
experimentele gegevens.

Ter verkrijging van experimentele gegevens zijn kinetische experimenten


uitgevoerd. Daarbij werd een bepaalde hoeveelheid kool toegevoegd aan een vast
volume oplossing en werd het concentratieverloop als functie van de tijd
gemeten. De vorm van de zogenaamde concentratievervalcurve hangt af van een
aantal factoren waarvan er een aantal (genoemd in de vorige alinea) gekozen
werden als te onderzoeken variabelen.

Onderzocht zijn drie systemen met één opgeloste component (4-isopropylfenol,


p-nitroaniline en nitrobenzeen) en twee gecombineerde tweecomponentensystemen
(4-isopropylfenoVp-nitroaniline en nitrobenzeen/p-nitroaniline). Deze
componenten zijn gekozen omdat ze representatief zijn voor verontreinigingen
die in industrieel afvalwater voorkomen en omdat hun concentraties nauwkeurig
te meten zijn met de toegepaste analysemethode (UV-spectrofotometrie).

Naast kinetische experimenten zijn voor alle betrokken systemen ook


evenwichtsexperimenten uitgevoerd. In dit onderzoek is een verbeterde versie
van de Pritz & Schlünder vergelijking voor de evenwiehts isotherm voorgesteld.
Deze vergelijking is gebruikt voor het beshcfijven van de 'twee-componenten'
systemen en toegepast voor het berekenen van het lokaal evenwicht in de
kinetische modellen.

Er zijn twee wezenlijk verschillende diffusiemodellen ontwikkeld om het gedrag


van de onderzochte adsorptiesystemen onder verschillende condities te kunnen
beschrijven(voorspellen). Beide modellen gaan uit van een uitwendige
filmweerstand en binnen het deeltje twee parallel verlopende
diffusiemechanismen: diffusie in de porievloeistof en oppervlaktediffusie. De
snelheid van de eigenlijke adsorptiestap wordt groot verondersteld ten
opzichte van de diffusiesnelheden; daarom is er in alle modellen van uitgegaan
dat binnen het deeltje evenwicht bestaat tussen porievloeistof en het
adsorbans op die plaats.

In de literatuur wordt de adsorptiekinetiek binnen het deeltje meestal


beschreven met behulp van de diffusievergelijking van Piek; deze vormt ook de
grondslag voor een van de in deze studie gebruikte modellen.
Een andere wijze van beschrijven van de kinetiek is gebaseerd op de
veralgemeende Maxwell-Stefan formulering waarbij de chemische potentiaal als
drijvende kracht voor diffusie wordt genomen. Krishna heeft recent een hierop
gebaseerd model voorgesteld voor diffusie uit de gasfase; dit model is
inmiddels met succes toegepast door Hu en Do. Het belagrijkste voordeel van de
Maxwell-Stefan beschrijving in vergelijking tot die volgens Piek is - naast de
thermodynamische fundering dat diffusiecoëfficiënten verkregen uit
experimenten met een enkele opgeloste stof gebruikt kunen worden voor het
beschrijven van de kinetiek in meercomponent systemen. Het tweede model dat in
dit proefschrift wordt toegepast kan worden gezien als een uitbreiding van het
model van Krishna tot adsorptie vanuit oplossingen.

Naast de twee bovengenoemde modellen is ook nog een derde, eenvoudiger, model
toegepast. In dit model wordt het massatransport binen het deeltje beschreven
met behulp van een constante effectieve diffusiecoëfficiënt. Deze constante
diffusiecoëfficiënt is een 'lumped parameter' en hangt niet af van de keuze
van defmitie van de drijvende kracht (volgens Fick of Maxwell-Stefan). Omdat
het model vaak in de literatuur wordt gebruikt is het in deze studie ter
vergelijking met de andere twee modellen meegenomen, ondanks het feit dat het
model duidelijk ver afstaat van de fysische realiteit binnen het deeltje.

De modellen zijn numeriek opgelost door middel van een 'fmite-difference'


methode. Alle modelresultaten zijn met behulp van een optimalisatieprogramma
vergeleken met de bijbehorende experimentele curves van het
concentratieverloop . De diffusiecoëfficiënt behorende bij de curve die het
beste overeen kwam met de experimentele curve geldt als waarde voor dat
experiment volgens het betreffende model.

Voor de 'ééncomponentsystemen' bleek uit de modelsimulaties dat de spreiding


in de waarden van de diffusiecoëfficiënten erg groot was voor zowel het model
met constante effectieve diffusiecoëfficiënt als voor het model gebaseerd op
de wet van Fick. Dit betekent dat beide modellen geen voorspellende waarde
hebben.
Het Stefan-Maxwell model, dat hier voor het eerst op adsorptie vanuit
vloeistoffen is toegepast, vertoonde een veel kleinere relatieve spreiding
rond de gemiddelde waarde van de diffusiecoëfficiënt. Dit betekent dat dit
model wel degelijk voorspellende waarde heeft voor de systemen met slechts één
opgeloste component en het opent de mogelijkheid de gevonden waarden van de
diffusiecoëfficiënt toe te passen in systemen met meer componenten.

Voor de 'meercomponentensystemen' blijkt het model met de constante


diffusiecoëfficiënt de experimentele resultaten redelijk goed te kunnen
beschrijven in het systeem 4-isopropylfenol/p-nitroaniline en in iets
mindere mate in het systeem nitrobenzeen/p-nitroaniline. Het model zou dus
gebruikt kunnen worden voor simulatieberekeningen. Aangezien de waarden van de
gevonden diffusiecoëfficiënten echter duidelijk afwijken van de (niet
constante) waarden uit de ééncomponent systemen zijn experimenten met het
meercomponentensysteem vereist.
Het meercomponenten Maxwell-Stefan model geeft een redelijke voorspelling van
het experimentele concentratieverloop en levert waarden voor de
diffusiecoëfficiënten die dicht in de buurt liggen van de waarden uit het
ééncomponenten systeem voor zowel 4-isopropylfenol als p-nitroaniline. De
waarden van de diffusiecoëfficiënt voor nitrobenzeen in het tweecomponenten
systeem verschillen onderling niet veel maar wijken wel af van die in het
ééncomponentsysteem.
Dit model maakt het mogelijk het kinetische gedrag van meercomponenten
systemen redelijk te voorspellen met behulp van diffusiecoefficienten die zijn
verkregen uit experimenten met één opgeloste component.
CONTENTS

Chapter 1. Introduetion 1

1.1 Activated carbon 1


1.2 Historica! background 2
1.3 Liquid phase applications 3
1.4 Objectives of this study 4

Chapter 2. Equations for liquid phase adsorption equilibria 7

2.1 Introduetion 7
2.2 Adsorption isotherrns for single-component systems 8
2.3 Adsorption isotherrns for multicomponent systems 11
2.4 Some remarks on irreversibility of adsorption 18

Chapter 3. Model formulation for adsorption kinetics 20

3.1 Introduetion 20
3.2 Basic adsorption models - literature review 21
3.3 Models based on Pickian diffusivity 23
3.3.1 Introduetion 23
3.3.2 Physical description of general model 24
3.3.3 Matbematkal description of Pickian model 27
3.3.4 Pull set of equations for constant
diffusivity (HPMC) model 34
3.3.5 Pull set of equations for Pickian model 34
3.3.6 Two-solute constant diffusivity model 36
3.4 Models based on Maxwell-Stefan diffusivity 36
3.4.1 Introduetion 36
3.4.2 Theoretica! background 37
3.4.3 Maxwell-Stefan pore diffusion 40
3.4.4 Relation between Pickian and Maxwell-Stefan
diffusivities 45
3.4.5 Maxwell-Stefan surface diffusion 48
3.4.6 Single-solute model equations 58
3.4.7 Two-solute model equations 59

Chapter 4. Experimental methods and results for equilibria 62

4.1 Matenals and instanation 62


4.1.1 Adsorbates and adsorbate mixtures 62
4.1.2 Water 64
4.1.3 Adsorbent 64
4.1.4 Equipment for equilibrium experiments 65
4.2 Experiments and analytica! methods 66
4.2.1 Preparation of standard solutions 66
4.2.2 Carbon pretreatment and preparation 67
4.2.3 Procedure for absorbance measurements 67
4.2.4 Procedure for obtaining equilibrium data 68
4.2.5 Remarks on experimental and instrumental errors 70
4.3 Correlation of adsorption equilibrium data 72
4.3.1 Single-solute systems 72
4.3.2 Two-solute systems 78
4.4 Some notes on irreversibility of adsorption 88
4.5 Conclusions 90

Chapter 5 Experimental and numerical methods for kinetics 92

5.1 Batch kinetic experiments 92


5.1.1 Experimental setup 92
5 .1.2 Procedure for obtaining kinetic adsorption data 96
5 .1.3 Ex perimental results 98
5.1.4 Some remarks on experimental reproducibility 103
5.2 Numerical solution, parameter fitting and optimization 104
5.2.1 Numerical salution and algorithms 104
5.2.2 Comparison of analytica! and numerical solutions 106
5.2.3 Parameter fitting and optimization 109
5.2.4 Influence of equilibrium on parameter fitting 118
5.3 Conclusions 122

Chapter 6. Estimation of rate parameters for single-solute


systems 123

6.1 Estimation of external mass transfer coefficient 123


6.2 Model calculations for single-solute systems 127
6.2.1 Introduetion 127
6.2.2 Model with constant diffusivity 128
6.2.3 Fickian model with variabie diffusivity 134
6.2.4 Transition model based on the pore liquid
concentration gradient 140
6.2.5 Maxwell-Stefan model with variabie diffusivity 142
6.3 Some additional remarks and discussions 150
6.4 Conclusions 157

Chapter 7. Estimation of rate parameters for two-solute


systems 159

7.1 Estimation of external mass transfer coefficient 159


7.2 Model calculations for two-solute systems 160
7.2.1 Constant diffusivity model 160
7.2.2 Maxwell-Stefan model without cross-diffusion
coefficients 164
7.3 Conclusions and recommendations 168

Chapter 8. Epilogue 170

8.1 General conclusions 170


8.2 Recommendations for future work 170

Appendix 4.1
Experimental equilibrium data for single-solute PNA and IPP systems 172
Appendix 4.2
Experimental equilibrium data for two-solute systems 174

Appendix 5.1
Experimental conditions for kinetic single-solute systems 176

Appendix 5.2
Experimental conditions for kinetic two-solute systems 179

Appendix 5.3
Some important experimental errors and reproducibility
measurements 181

Appendix 5.4
Numerical solution of differential equations used
in this stud y 186

Appendix 5.5
Flowsheet of general algorithm for single-solute
model with variabie diffusion coefficient 192

Appendix 5.6
Pascal program for two-component Maxwell-Stefan model 195

Appendix 5.7
Comparison of analytica} and numerical solutions for
special cases of the constant diffusivity models
(some additional details) 208

Appendix 6.1
Experimentally determined values for extemal
mass transfer coefficient 214

Appendix 6.2
Transition model based on the pore concentration gradient
(see also section 6.2.4) 216
Appendix 7.1
Experimental values for extemal mass transfer coefficients 224

References
List of symbols
Curriculum vitae
Chapter I 1

INTRODUCTION

Adsorption is a surface phenomenon by which a solute is removed from one phase


and accumulated at the surface of a second phase. The material being adsorbed
is the adsorbate, and the adsorbing material is called the adsorbent.

1.1. ACTIV ATED CARBON

Some of the most widely used types of adsorbents nowadays are activated
carbons. They are non-hazardous, processed, carbonaceous products, having a
porous structure and a large intemal surface area. These materials cao adsorb
a wide variety of substances, i.e. they are able to attract molecules to their
intemal surface and are therefore often used in processes of purifying,
decolorizing, deodorizing, detoxicating and separating.
Although it is now recognized that the pore structure is the most important
property of activated carbon, it was formerly believed that the carbon had to
he activated by chemica! and heat treatment before it could remove color,
hence the name activated carbon. It is now known that the removal of
impurities from gases and liquids by activated carbon is by adsorption, and
the activation process simply increases the intemal surface area of the
carbon and hence the number of sites available for adsorption.

Activated carbons cao he prepared in the laboratory from a large number of


materials but those most commonly used in commercial practice are peat, coal,
lignite, wood and coconut shell. The residues from carbonization have a large
pore volume, and as this is derived from very small diameter pores the
intemal surface area is high. Activated carbons have internat surface areas
in the range 500 1500 square meters per gram and it is this large area which
makes them effective adsorbents.

Th ere are two main types of activation processes: steam


activation and chemica! activation.
Steam activated carbons are produced by a two-stage process. Firstly the
material is carbonized and a coke is produced, the pores of which are either
too small or constricted for it to he a useful adsorbent. The next stage is a
Chapter I 2

process of enlarging the pore structure so that an accessible internat surface


is created. This is achieved by reacting the semi-product with steam between
900°C and I000°C. At this temperature the rate determining factor is the
chemical reaction between carbon and steam. This reaction takes place at the
internat surface of the carbon, removing carbon from the pore walls and
thereby enlarging them.
Chemically activated carbons are produced by mixing a chemical with a
carbonaceous material, usually wood, and carbonizing the resulting mixture.
The carbonization temperature is relatively low, e.g. 400°C 500°C and the
chemical is usually reeavered for use after carbonization. The chemieals
normally used are phosphoric acid and zinc chloride solutions which swell the
wood and open up the cellulose structure. On carbonization the chemica! acts
as a support and does not allow the resulting char to shrink.
The majority of activated carbon used throughout the world is produced by
steam activation.

1.2. HlSTORICAL BACKGROUND

Activated carbon is one of the oldest and best known adsorbents with a wide
range of dornestic and industrial applications. Charcoal, the forerunner of
modern activated carbon was used by ancient Egyptians for medicinat purposes.
The properties of carbon to adsorb gases were first reported by Scheele in the
18th century. In 1790 Lowitz established the first use of powdered charcoal
for the removal of bad tastes and odors from water on an empirica! basis.
The first application of activated carbon in a water treatment plant, for
removing chlorophenolic substances, was reported in the United States, in
1929. By some estimates by 1939 there were already 400 such plants, using
powdered carbon to reduce odors in drinking water [Faust, 1987].
Nowadays activated carbon is used for solvent recovery and air purification,
in food processing and chemica! industries; in the purification of many
chemical and foodstuff products; for the recovery of gold, silver and other
inorganic products and in the treatment of dornestic and industrial waste
waters. Nearly 80% (220 000 tons/yr) of the total world production of
activaLed carbon is consumed for liquid phase applications where both granular
and powder activated carbons are used [Bansal, 1988].
Introduction 3

1.3. LIQUID PHASE APPLICATIONS

Since 1930 powdered activated carbon has been used as the main adsorbent for
cleaning ground water and thus making it suitable for drinking. However,
during the last hundred years the consumption of potable water bas
significantly increased and so it became necessary to use other sources, such
as surface waters and process them into a form suitable for drinking purposes.
The presence of pesticides and other toxic substances in the surface waters
led to the need for much larger amounts of activated carbon and for radkal
improvement of the existing methods and technologies for potable water
treatment. At the same time it became clear that not only the water for
drinking had to be purified, but also the enormous amounts of waste waters
produced had to be processed, before being discharged into the environment.
Using granular instead of powdered activated carbon proved out to be an
economically viabie and efficient solution. This led to the building of
numerous granular carbon treatment facilities in many parts of the world. To
make them work efficiently it was necessary to define and develop suitable
design parameters and to study the kinetics of adsorption from dilute
solutions.

In many cases the use of granular carbons, unlike powdered carbons, involves
the regeneration of the carbon. Regeneration is the treatment of the carbon to
remove the adsorbed impurities so that it can be re-used. The three most
commonly used ways to regenerale used carbons are: steam regeneration this
can be used if the adsorbed products are volatile, i.e. if they can be steam
distilled; chemical regenerafion in certain cases the adsorbed impurity can
be desorbed from the surface by treatment with a chemical. Caustic soda has
been used successfully with eertaio organic acid purification systems; thermal
regeneration this is the most widely used metbod particularly in the large
tonnage outiets in the sugar, glucose and water industries. This involves the
removal of the carbon from the packed column and buming off the impurities
under controlled conditions in a regeneration kiln.
There are several important design parameters in the construction of a
granular carbon plant used for potable water production or waste water
treatment (EPA report, 1973]:
type of carbon filters used (open gravity or closed pressure filters)
Chapter I 4

- mode of operation of the filters (fixed bed, fluidized, etc.)


- contiguration of the filters (in series, in parallel, etc.)
- contact time between treated liquid and carbon, fluid velocity,
pressure drop
- type of carbon used (particle size, pore structure)
- reactivation facilities (local or extemal)

The overall rate of adsorption can be defined as the rate at which the
adsorbable component is transferred from the bulk phase outside the carbon
particles to the intemal adsorption sites [Van Lier, 1989]. The overall rate
of adsorption depends on such factors as:
- the type of carbon which is used
- the nature and relative concentrations of the substances to be removed
- the pH and temperature of the water
- the presence of compounds which have an influence on the adsorption
process, although they do not have to be removed

The overall rate of adsorption is very important for the design because it
determines the optimal values of the linear liquid velocity and contact time.

1.4. OBJECTIVES OF TUIS STUDY

The main objective of this study is to investigate the kinetics (i.e. overall
rate) of adsorption on activated carbon and to try to use single-component
kinetic parameters to describe (predict) multicomponent kinetics. This
objective is achieved by studying the overall rate of adsorption of several
organic compounds from dilute aqueous solutions on granular activated carbon
in a batch system, with special focus on the dependenee of diffusion inside
the partiele on initial organic compound(s) concentration, amount of carbon
added, type of organic compound used and type of adsorption system (single- or
two-component). Several mathematica! models are developed and the simulation
results are compared to experimental data.

For all experiments only one type of activated carbon is used. In several
cases the influence of the partiele size and partiele size distribution are
Introduetion 5

also investigated. The one-component experimental systems used are


nitrobenzene(NBZ)/water, p-nitroaniline(PNA)/water and 4-isopropylphenol(IPP)/
water. The two-component systems are nitrobenzene/p-nitroaniline/water
(NBZ/PNA) and 4-isopropylphenol/p-nitroaniline/water (IPP/PNA).
The above-mentioned organic compounds and respective mixtures were chosen
because they are often present m waste waters and because their
concentrations (absorbances) could be accurately measured by the analytical
metbod used (UV-spectrophotometry).

A review of existing equilibrium models is presented in chapter 2. Empirica!


and theoretically based isotherm equations are discussed and several improved
versions for some of the isotherm equations are being proposed. In the last
part of the chapter some remarks are made on adsorption irreversibility

In the first part of chapter 3 a literature review of kinetic batch models is


given. Later in the same chapter a detailed denvation and explanation of the
mathematica! models used in this study is presented.

Chapter 4 contains a description of the experimental methods and matenals


used for equilibrium experiments. The correlations of the obtained data with
different isotherm models are also presented in this chapter.

Chapter 5 is dedicated to estimation of single-solute kinetic parameters. In


the beginning of the chapter , the experimental methods and results from the
kinetic experiments are given and also some remarks are made on experimental
errors and reproducibility. Later in the same chapter results from the
numerical model calculations are compared with analytica! solutions from
literature. The last part of chapter 5 includes a description of the
optimization procedure used and a discussion of the sensitivity of different
fitting parameters.

Chapter 6 contains the results of all model calculations for the three
single-component systems used in this study. Comparisons between the different
models and between the experimental systems are made and the validity and
predictive value of the models are discussed.
Chapter I 6

In chapter 7, some data and discussions are presented about the two-component
model calculations, tagether with comparisons between model and experimental
curves.

Finally, chapter 8 contains general conclusions and recommendations for future


work.

Some chapters have Appendices which contain additional information and details
about the subjects under discussion. The Appendices carry the number of the
respective chapter, but are all placed at the end of this thesis.
Chapter ll 7

EQUATIONS FOR LIQUID PHASE ADSORPTION EQUILIBRIA

2.1. INTRODUCTION

In order to describe the overall rate of adsorption on activated carbon it is


necessary to understand the mass transfer processes, and also the equilibrium
behavior of the systems involved. This chapter deals mainly with liquid phase
adsorption equilibrium to the extent necessary to describe the rate behavior
of the single- and two-component systems used in this study.

In principle, the process of adsorption involves the concentration of the


adsorbate on the surface of a solid or, less frequently, of a liquid. Two
types of adsorption can be distinguished, depending on the nature of the
forces involved. In chemical adsorption a single layer of molecules, atoms or
ions is attached to the adsorbent surface by chemica! bonds. In physical
adsorption adsorbed molecules are held by the weaker van der W aais' forces and
multilayer formation is possible.

Physical adsorption of single-gases on solid adsorbent surfaces is well


described in literature. The first serious attempt to give a fundamental
description of monolayer localized adsorption was made by Langmuir in 1918.
Since that time numerous other papers were published, extending the ideas of
Langmuir to multilayer adsorption on homogeneous solid surfaces [Brunauer,
1938], others assuming a semi-mobile adsorption model [Patrykiejew, 1984],
and still others descrihing the single-gas adsorption by
statistical-mechanical means [Steele, 1974], [Sokolowski, 1981].

There is also a significant amount of publications available in the field of


multicomponent gas adsorption, although here the overall physical picture is
less clear. The most complete compilation of existing data on equilibrium of
pure gases and gas mixtures has been publisbed by Valenzuela and Myers in
1989. This same pubHeation contains also a lot of data on liquid mixtures. On
the theoretica} side, in 1965 Myers and Prausnitz applied for the frrst time
solution thermodynamics to mixed-gas adsorption, foliowed by further advances
in the same field [Sircar, 1973], [Myers, 1987]. An alternative thermodynamic
Chapter !I 8

description of mixed-gas adsorption is based on the vacancy solution theory


[Suwanayuen, 1980a,b], [Fukuchi, 1982], [Talu, 1987].
In many theoretica! studies in literature it is assumed that the surface of
the adsorbent is homogeneaus and that there are no lateral interactions
between the adsorbed molecules. This is not the case for such a strongly
heterogeneaus adsorbent as activated carbon and it makes the task of finding a
fundamentally correct and at the same time universally applied isotherm
equation very difficult.

A different type of adsorption is the adsorption from liquid solutions. Here


it is much more complicated than for gases to formulate a unified theory,
because there are many liquid mixtures with very different physical and
chemical properties. Another complication arrises from the strong possibility
for interactions in the liquid phase between different components, as well as
the complex structure of the adsorbent [Jaroniec, 1985].
Adsorption from dilute solutions is a special case of liquid adsorption. In
principle, the isotherm equations for adsorption from dilute solutions can be
derived from the general isotherm equations in liquid solutions which describe
the whole concentration range (Jaroniec, 1988]. It often happens, however,
that researchers use gas-adsorption isotherm equations and change the relative
pressure into solute concentration. From a mathematica! point of view the
isotherm equations can be identical, which does not necessarily meao that the
mechanisms involved in the two processes are the same (Jaroniec, 1983].
In spite of these difficulties, there are several empirica! and semi-empirica!
isotherm equations which give in many cases a good description of experimental
liquid-phase adsorption results [Radke, 1972a], [Jossens, 1978], (McKay,
1989].

Further in this chapter a short description will be given of several commonly


used isotherms, some of which have been applied in this study. As noted, most
discussions will be in terros of liquid phase adsorption.

2.2. ADSORPTION ISOTHERMS FOR SINGLE-COMPONENT SYSTEMS

The Henry isotherm


The most simple type of isotherm equation is the one described by Henry's
Adsorption equilibria 9

law. At very low sol ute concentrations the molecules of the adsorbate do not
interact with each other and do not compete for adsorption sites. In that case
the equilibrium relationship between the concentrations in the fluid and in
the adsorbed phases will have a linear form. This is referred to as Henry's
law, where the constant of proportionality is the adsorption equilibrium
constant Kh [Ruthven, 1984]:

Q =KC
h p'
(2.1)

where Q and C can be expressed as molecules or moles or weight per weight


p
or per unit volume in the adsorbed and fluid phases.

TheLangmuir isotherm
The first isotherm model which assumed monolayer coverage of the adsorbent
surface was proposed by Langmuir in 1918. It contains several very important
assumptions:
1. All molecules are adsorbed on definite sites of the adsorbent surface.
2. Each site can be occupied by only one molecule.
3. The adsorption energy of all sites is equal.
4. When adsorbed molecules occupy neighboring sites there is no
interaction between them.
Initially this isotherm was derived by Langmuir kinetically while later, it
was also derived on basis of statistica} mechanics, thermodynamics and the
Maxwell-Boltzmann distribution law [Young, 1962].

The most commonly used form of the Langmuir isotherm is the following:

Q
m
b cP
Q= (2.2)
1 + b c
p

where C is the equilibrium concentration of the component; Q is the


p m
concentration in the adsorbed phase required to have monolayer coverage of the
whole surface; and b is a constant related to the heat of adsorption.
At very low concentrations the term b·C is much smaller than unity and then
p
Eq. (2.2) reduces to:
Chapter ll JO

Q Qm b c'P (2.3)

which is exactly the form of Henry's law. On the other hand from Eq. (2.2) it
follows that Q approaches Q asymptotically as C goes to infinity which is
m p
the expected behavior for monolayer adsorption.
Although for many systems the assumptions of the Langmuir isotherm do not hold
it can be still very useful as a theoretica! basis for studying adsorption
equilibrium.

The Freundlich isotherm


The Freundlîch equation [Freundlich, 1926] is one of the best known
mathematica! descriptions of adsorption equilibrium. lt bas the following
general form:

(2.4)

where Fr and Nf (Nf > 0) are constants characteristic of the system.

Although this is an empirica! isotherm it still gives some information about


the heterogeneity of the adsorbent surface [Jaroniec, 1983]. A drawback is
that it does not reduce to a straight line (that is, to Henry's law) at very
low concentrations.

The BET isotherm


The isotherm model developed by Brunauer, Emmett and Teller (BET) in 1938
extends the concept of monolayer adsorption to the formation of multilayers on
the adsorbent surface. It makes several basic assumptions, the most important
of which is that each molecule in the frrst adsorbed layer is providing one
site for the. second and subsequent layers. The molecules of these layers are
considered to behave as the saturated liquid while the equilibrium constant
for the first layer of molecules in contact with the surface of the adsorbent
is different [Ruthven, 1984]. The most widely used form of the BET equation
reads as follows:

Q B C
Q= m p (2.5)
(C-C) [1+ (B-l)C /C]
s p p s
Adsorption equilibria 11

where cs is the solubility of the component in water at the relevant


temperature and B is a constant.
The BET isotherm model is applied in one of the main methods for the
determination of the specific surface area of microporons solids.

The Radke-Prausnitz isotherm


An empirica! relation with three parameters was proposed by Radke and
Prausnitz [1972a] to describe equilibrium data from liquid phase over a wide
concentration range:

Q
KC
= ___r__,_p_ __
(2.6)
1 + ( K /F )C 1-Nr
r r p

where Kr , Fr and Nr are constants and Nr < 1.

Although empirical, the Radke-Prausnitz equation bas several important


properties which make it suitable for use in many adsorption systems. As
already mentioned it can describe fairly well the equilibrium dependenee over
a very wide concentration range. At low concentrations it reduces to a linear
isotherm, described by Bq. (2.1). At high solute concentrations it becomes the
Freundlich isotherm, Bq. (2.4), and for the special case of Nr = 0 it becomes
the Langmuir isotherm (Bq. (2.2)). The three parameters are estimated by a
non-linear statistica! fit of the equation to the experimental data.
The Radke-Prausnitz isotherm was used in this study to correlate the
single-component experimental equilibrium data.

2.3. ADSORPTION ISOTHERMS FOR MULTICOMPONENT SYSTEMS

Many practical adsorption systems, especially those in potable and waste water
treatment, contain more than one component. That is why the problems of
correlating and predicting multisolute equilibria from single-solute data is
of primary importance when dealing with such systems. Several isotherm models
will be considered in this section.
Chapter IJ 12

Models based on the isotherm


The classica! Langmuir model for multicomponent adsorption is based on the
same assumptions as the single-component model. The resulting equation has
been derived for the first time by Butler and Oclcrent [1930] and reads as
follows:

Q ' b, c'
Q. = ___m_:_,•_ _•_P'-'·-'- (2.7)
1 k
+:Eb, c '
j l p,t I

where k is the number of components, i is the component number and the


constants b are obtained from single-component data. This model is still often
used although recent studies [McKay, 1989] show that in many cases it can not
give a good correlation of the experimental data.

In 1973 Jain and Snoyeink introduced a two-component model, which is a


modified version of Langmuir's model:

Q =
(Q
rn,l
- Q
m ,2
) b
I
cp,l +
Q
m,2
b
l
cp,l
(2.8)
l 1+ b c
I p, I
1 +bC
I p, 1
+bC
2 p,2

(2.9)

It is a semi-competitive model as it assumes that part of the adsorption takes


place without competition due to the fact that not all adsorption sites are
available to all solutes. In quantitative terms this means that a number of
sites corresponding to Qm,l Qm,2 (where Qm,l > Qm,2) can accept only
molecules from component 1 while all components compete for the remaining
sites, which are equal to Q . These equations can be very suitable for
m,2
components with different molecular sizes or with different chemica!
properties relative to the adsorbent. A drawback of this model is that it is
valid only for two-component systems.

The above described roodels use only single-component data and do not take into
account the possible interactions between the components while the adsorption
Adsorption equilibria 13

takes place. It is very probable that in real adsorption systems there are
various interactions which have to be accounted for when modeling the process.
This reasoning was the basis for the introduetion of another, modified version
of the Langmuir model, in which an interaction term, 11, appears. In this
model, proposed by Schay [1957] the interaction terrus 11. are evaluated by
I

correlating the model to multicomponent experimental data. He proposed the


following correlation:

Qm, . b. (C p, . I 'llJ
1 1 t 1
Q (2.10)
k
+I b. (C . I 11)
j=l J p.j J

where 11 is a term which accounts for possible interactions between the


components and is specific for each solute.

Models based on the Radke-Prausnitz isotherm


Mathews [1975] extended the Radke-Prausnitz isotherm to multicomponent
adsorption and proposed the following equation:

Q' = ~----,.--~----"--···--
1 k
(2.11)
1 +I (K , I F ,) C ~-Nrj
j= r,J r,J p,J
1

In this equation the constants are calculated from the single-component


systems which are described by Eq. (2.6).
Por some of the systems used, Mathews achieved better results by using Eq.
(2.11) than by using Eqs. (2.8) - (2.9). However, other systems could not be
adequately described even by Eq. (2.11), so he applied an interaction term and
proposed another equation:

(2.12)
k
1
+ L (K ,/ F .) (C . I '11) -Nrj
j I r ,J rJ p,J J

The constants Kr , F,r Nr for each component are obtained in the same way as
for Eq. (2.11). The terrus 11. are determined by miniruizing the sum of squares
1
Chapter /1 14

(SOS) of the difference between the relative experimental and calculated


adsorbed phase concentrations for each component.

One of the basic conditions a multicomponent equilibrium model should fulfill


is that when the concentrations of all but one components are zero it should
reduce to the single-component form of the equation. This is not the case with
the isotherm model described by Eq. (2.12) which raises questions about its
validity, especially in systems where some solute concentrations in the
adsorbed phase can be very low. lt may also lead to erroneous results when
using the equation in a ldnetic model where the calculations always start with
a zero adsorbed phase concentration.
Due to the above reasons, an improved version of Eq. (2.12) is proposed in
this study:

K
r i
(2.13)
k
1 + (K . I F .) C . +
r,1 r,1 p,1 j
L= (K r ,J.I Fr,J.) (C p,J. I TJ/-NrJ
J
1
( j * î)
In contrast to Eq. (2.12), Eq. (2.13) reduces to the single-component form
when all but one concentrations are zero. A discussion and correlation with
some experimental results are given in chapter 4.

Models based on the Freundlich isotherm


Fritz and Schlünder [1974] proposed a general empirica! multisolute
equation, which has the following form:

A.10 C p 1. Bio
Q.1 = k
t
(2.15)
E + LA ..
I IJ
cPJBij

where A, B and E are constants.

For specific values of the constants Eq. (2.15) reduces to the Mathews
equation (Eq. (2.11)) or to the Langmuir muitkomponent model (Eq. (2.7)).
Fritz and Schlünder showed a method for evaluating the constants for a
two-component system when the single-component data could be fitted with the
Adsorption equilibria 15

Freundlich isotherm. In this case the equations are:

F C Nf, I
+ B
II
f, I p,l
Q
I
= (2.16a)
c p •I B
I 1 +A
12
c p,2 B
12

N
f. 2
+ 8
22
~= (2.16b)
c p,2 8
22 + A 21 cp • I B
21

where Ff and Nr are the single-solute Freundlich constants, while the A and B
constauts are determined by correlating the multicomponent equilibrium data.

The use of Eqs. (2.16) in kinetic models calculations presents eertaio


· difficulties. As mentioned the Freundlich isotherm does not reduce to a linear
form at low concentrations and this is also valid for its multicomponent form,
i.e. Eqs. (2.16). This may lead to erroneous results when calculating the
profiles inside the partiele in the multicomponent kinetic models and
consecutively to the deterrnination of the wrong diffusion coefficients. For
this reason a new, extended form of Eqs. (2.16) is proposed in this study in
which an additional term is introduced, accounting for the linear part of the
isotherm at concentrations close to zero:

1 1 c p ,I B I + A
I
c p,2 D I
:::: + (2.17a)
Ql K
r, I
c p,l F cp • I N
r , I
+ B
I
r, I

1 1 c p,2 B 2 + A
2
cp • ID2
= +--··· (2.17b)
Q2 K
r, 2
c p,2 F c p,2 N
r , 2
+ B2
r, 2

The transformation from Eqs. (2.16) to (2.17) is analogous to the one from
Freundlich to Radke-Prausnitz isotherm for a single-solute system. Note,
however, that in Eqs. (2.17a) - (2.17b) all the single-solute parameters used
are from the Radke-Prausnitz isotherm, while Eqs. (2.16) are based on
Chapter IJ 16

Freundlich single-solute parameters.


lsotherm Eq. (2.17) was used in this study for calculating the local
equilibrium in the kinetic models. Details are presented in chapter 4.

Model based on the ideal adsorption salution theory (lAST)


The technique of predicting multisolute data on the basis of single-solute
experiments using the concept of an ideal adsorbed solution was proposed for
the first time by Myers and Prausnitz [1965]. The main idea is that in an
ideal salution the partial pressure of an adsorbed component is given by the
product of its mole fraction in the adsorbed phase and the pressure which it
would exert as a pure adsorbed component at the same temperature and spreading
pressure as those of the mixture. This thermadynamie spreading pressure can be
calculated from the experimental adsorption isotherms and does not depend on
any particular physical model of the adsorbed phase. lnitially this model was
derived for gas mixtures and was extended later to dilute liquid mixtures by
Radke and Prausnitz [ 1972b]. Si nee that time the lAST has been used by
numerous researchers to predict multicomponent equilibria from
single-component isotherm parameters [ûkazaki, 1980], [Fritz, 1981],
[Crittenden, 1985], [Baudu, 1989]. However, predicted adsorption equilibria
are not always found to be in good agreement with experimental data.
Extensions and modifications of the lAST were proposed, where more pronounced
deviations from ideality were taken into account [Lee, 1988], [Seidel, 1989],
[Gamba, 1990], [Myers, 1991].

The basic assumptions made in the ideal adsorption salution model in the case
of liquid mixtures are: 1) the liquid solution in question is dilute, 2) the
adsorbed phase forms an ideal solution, 3) the adsorbent is thermodynamically
inert, and 4) the available surface area is identical for all solutes. For
these systems the model is based on the assumption of a hypothetical ideal
adsorbed phase salution which contains only one component i, and this salution
has the same spreading pressure n as the adsorbed phase solution containing
all the components. n is defrned as the difference between the interfacial
tension of the pure solvent-solid interface and that of the solution-solid
interface at the same temperature:

0=0" 0" (2.18)


pure solv.-solid solution-solid
Adsorption equilibria 17

The model can be described by several basic equations and can predict
muitkomponent behavior from single-component data [Radke, 1972b]:

n
<2r = :L Q. 1
(2.19)
i =l

z.1 = Q.I I o
"'î
(2.20)

ci = zi C i
0
(2.21)

n z.
= L-1- (2.22)
.
I l
Qo.
1

TI 0 A
TI A
m s
= ---
i
= JCo i
Qo
_i
0
dCo. (2.23)
RT RT 0 c
where
TI0I :spreading pressure of single solute [N·m- 1]
I1 :spreading pressure of mixture [N·m- 1]
m

The two basic equations needed to calculate multi-solute equilibria are (2.21)
and (2.22). Equation (2.22) indicates that the total invariant adsorption is
determined by the adsorptions of the single solutes at identical values of
spreading pressure and temperature. Equation (2.21) results from consiclering
the equilibrium between the adsorbed and liquid phases.
In order to apply Eqs. (2.21) and (2.22) it is necessary to know the spreading
pressores of the different singly-adsorbing solutes in the mixture. These
pressores are calculated from Eq. (2.23) using the data from experimental
single-component isotherms. As I1 is determined only by these isothenns no
theoretica! model is needed to describe the adsorption equilibria.
There is, however, an important condition to be fulfilled. As the integration
in Eq. (2.23) begins at zero concentration it is necessary to have
experimental data for the whole range of loadings from zero to in order to Q?
calculate TI accurately enough.
The lAS theory is applied most often to two-component systems, but extending
it to multicomponent systems is straightforward and easy. There is no
Chapter 11 18

restrietion whether the components are below or above their melting points, or
whether or not they are miscible in all proportions with the solvent [Radke,
1972b].

Although the lAS model is thermodynamically consistent and extensively used,


its assumptions have limited applicability and its theory cannot provide a
general method for the prediction of binary or multicomponent equilibria from
single-component data [Ruthven, 1984]. In addition to that, as already pointed
out, it is necessary to have single-component data at very low concentrations
in order to calculate the spreading pressure. The model requires extensive
computational effort, especially in the case when the spreading pressure
integration (Eq. (2.24)) cannot be solved analytically. Another objection 1s
that at high surface loadings the assumption for an ideal adsorbed phase is
probably not correct.

In view of the foregoing the lAS model was not used in this study and more
simple, empirica! relations were applied in the kinetic models.

2.4 SOME REMARKS ON IRREVERSIBILITY OF ADSORPTION

All isotherm equations discussed so far are based on the assumption that
adsorption is a physical process and no chemica} reactions occur on the
adsorbent surface. However, all carbonaceous materials, even graphite and
diamond, contain surface functional groups. On activaled carbon these surface
groups are mostly located at the edges of graphite-like basal planes and there
are probably electronk interactions between them [Rivin, 1971]. Therefore, it
is possible that under certain conditions, the components which are adsorbed
on the activated carbon would react with these surface groups, leading to the
formation of strong chemica! honds. In 1969 Mattson proved the formation of
charge-transfer complexes between adsorbates and surface functional groups.

In adsorption literature most authors use the term "irreversibility" to


designale the occurrence of chemisorption [Suzuki, 1978], [Yonge, 1985],
[Vidic, 1990]. There are various studies which strongly suggest the existence
Adsorption equilibria 19

of strong irreversibility (chemisorption). Recently, Grant [1990], through


systematic sequence of experiments, determined that oxidative coupling of
phenolic compounds on carbon surfaces is a plausible explanation for
irreversible adsorption. Earlier Coughlin [1968] showed that oxidation of the
activated carbon surface prior to adsorption lowered the capacity of the
carbon for phenol and nitrobenzene. Other researchers however [Thakkar, 1987]
succeeded in fully recovering several phenolic compounds from activated carbon
and thus concluded that there was little, if any, irreversible adsorption in
this case.

A different concept for irreversibility was proposed in 1988 by Kerkhof. He


suggested that in some cases physical rather than chemical adsorption can be
the cause for irreversibility and that capillary crystallization may be the
mechanism by which it is occurring. However, forther studies are necessary in
order to get more conclusive evidence about this model.

In the present study limited attention is given to the problems of


irreversibility. The kinetic models developed are based on the assumption that
adsorption is fully reversible and there is no special term accounting for
possible chemica} reactions. Although the discussed literature sourees offer
some evidence about irreversible adsorption it is still the case that almost
all existing models do not take it into consideration and for most practical
applications these roodels can predict fairly well the process.

A very limited amount of desorption experiments were performed with 4-IPP to


try to determine the extent of irreversibility, if any, at different
temperatures. Details are given in chapter 4.
Chapter III 20

MODEL FORMULATION FOR ADSORPTION KINETICS

3.1. INTRODUCTION

Descrihing adsorption on a porous solid requires an understanding not only of


equilibrium behavior but also of mass transfer (transport) phenomena.
In principle, adsorption kinetics can be determined by several processes:
- transfer of molecules from the bulk phase to the outer surface of the
partiele through a fluid boundary layer surrounding the partiele (external
mass transfer).
- diffusion of molecules in the liquid in the pores (pore diffusion)
- diffusion of already adsorbed molecules along the surface of the pores
(surface diffusion).
- the elementary processes of adsorption and desorption.

One or several of these processes can be much slower than the others and in
that case they determine the overall rate of adsorption. On the other hand
since adsorption is exotherrnic and the heat of sorption must be removed by
heat transfer there is, in general, a difference in temperature between the
adsorbent partiele and the bulk fluid when adsorption takes place. How
important this temperature difference is depends on the relative rates of heat
and mass transfer. However for adsorption from liquid systems it will be
fairly accurate to assume that heat transfer is sufficiently rapid, so that
temperature gradients in and around the partiele are negligible [Ruthven,
1984].

A convenient way to describe the intrapartiele transport in an adsorbent like


activated carbon is to consider it as a diffusive process and to express it
according to Fick's 1 s 1 and 2"d laws, which will be considered in more detail
later in this chapter. This is the most widely used form and almost all models
in literature are based on Fick's mathematica! representation. However, since
the true driving force of a diffusion process is the gradient of the chemical
potential and not the concentration gradient an alternative model is discussed
in Section 3.4, based on the Generalized Maxwell-Stefan equations.
All models used in this study consider the adsorbent partiele as having a
Adsorption kinetics models 21

homogeneaus and isotropie pore structure. However, in order to distinguish


between the two types of flux-driving force defmitions, the Fickian model
will be referred to as Homogeneaus Partiele Fickian Model (HPMF) and the MS
model will be referred to as Homogeneaus Partiele Maxwell-Stefan Model (HPMS).
As it will become elear from the mathematica! derivations in this chapter both
types of models inelude a variabie apparent diffusivity (D app ), but constant
pore (D ) and surface (D ) diffusion coefficients.
p s

In addition, two other models are also used in this study. The first one,
named Homogeneaus Partiele Constant Diffusivity Model (HPMC) assumes a
constant D . The second one, called Homogeneaus Partiele Transition Model
app
(HPMT) assumes a variabie D and is based on rather empirica!
app
considerations. This latest HPMT model is discussed in chapter 6 and Appendix
6.2.

3.2. BASIC ADSORPTION MODELS LITERATURE REVIEW

Extensive efforts were made during the last few decades to try to model the
adsorption of single and multicomponent systems of gases and, to a lesser
extent, of liquids. This short review will deal mainly with studies of mass
transfer on activated carbon from aqueous solutions, however, work of a more
general nature, ineluding other adsorbents and solvents, have also been
included where they are relevant to the present study. A much more extensive
literature review on kinetics of adsorption on activated carbon has been
published by Van Lier [1989].
The models reviewed in this survey are classified on basis of the type of mass
transfer resistances (one or more) that they take into account. This
elassification is commonly used in adsorption literature and gives a good
basis for comparison between different classes of models.

In earlier adsorption studies models were proposed which considered only the
resistance to extemal mass transfer as important. One such model was
published by Misic [1971]. There, adsorption of benzene in carbon slurries was
modeled and the intrapartiele transport resistance was considered negligible
(for the specific system conditions), while the resistance to mass transfer
through the liquid immediately surrounding the partiele appeared to be
Chapter lil 22

significant. The concentration at the outer surface of the partiele was


directly related to the concentration inside the partiele by an adsorption
isothenn.

In later studies the mass transport was considered to be governed by pore


diffusion [Kocirik, 1973], [Furusawa, 1974], [Smith, 1985]. Models were
developed in which the adsorbent particles were regarded as consisting of a
solid phase interspersed with small pores. The adsorbate diffuses into the
pore liquid phase and adsorbs on the pore walls.
When both pore diffusion and external mass transfer were taken into account
adsorption kinetic models with two types of resistances were fonnulated
[Suzuki, 1974], [Neretnieks, 1976a], [McKay, 1985], [McKay, 1988].

It appeared, however, that all models considering pore diffusion as the only
intrapartiele mass transport mechanism had a serious drawback. Various results
publisbed in literature showed values for the pore diffusivity higher than
those for the respective free liquid diffusivity [Van Lier, 1989], [McKay,
1985], [Ramos, 1981].
These fmdings indicated that there was probably another transport mechanism
inside the adsorbent partiele which contributed significantly to the value of
the diffusion coefficient. Models were fonnulated where surface instead of
pore diffusion was considered as the rate lirniting intrapartiele mass
transport mechanism. The surface diffusion model assumes phase separation to
be occurring by adsorption on the pore walls and subsequent movement of the
adsorbed molecules along the pore wall surface. Some of these models neglect a
possible external mass transfer contribution [Sudo, 1978], [Korniyama, 1974],
[Suzuki, 1982], [Kozhusko, 1989] while others consider both k and D
f s
contributions to the governing transport processes [Mathews, 1975],
[Neretnieks, 1976b], [Ramos, 1981], [Pirbazari, 1981], [Muraki, 1982],
[Mathews, 1983], rrraegner, 1989].

In another development, Ying [1979] devised single-solute models that


included the effect of biologica! activity that can occur on activaled carbon.
Arve [1987] proposed a model for biospecific adsorption (affinity
chromatography) in a batch system. The model accounts for film and diffusional
mass transfer resistances as well as for the rates of interaction between
Adsorption kinetics models 23

adsorbates and ligands.


The most complete type of homogeneaus partiele models accounts for all three
possible types of mass transfer resistances and includes in its equations the
kf' D p and D s coefficients. Such models were developed and solved numerically
by Liapis [1977], Mansour [1985], Fettig [1987] and many other ·researchers
[Van Lier, 1989], [Moon, 1983], [Neretnieks, 1976b,c].

From a theoretica! point of view a good model should account for all possible
mass transfer resistances and give a realistic description of the adsorption
processes. For this reason the mathematica! models solved and used in this
study inelude all three types of resistances discussed above while in some
cases, in order to compare different models, some simplifying assumptions are
being made.

3.3. MODELS BASEDON FICKIAN DIFFUSIVITY

3.3.1. INTRODUCTION

As noted in Section 3.1, there is a basic difference between the Fickian and
Maxwell-Stefan's way of expressing the diffusive process in an adsorbent
particle. This has a direct influence on the mathematica! equations with which
different models with variabie diffusivity (D ) are described. However, if a
app
constant apparent diffusion coefficient (D ) is assumed (HPMC model), the
app
equations are exactly the same for both ways of expressing the fluxes.

In this Section, 3.3, the Fickian formulation for diffusion will be used to
derive the Fickian (HPMF) model, that includes extemal, pore and surface
diffusion mechanisms and also the constant D (HPMC) model will be
app
presented. For each model a tentative physical picture will be given, foliowed
by a mathematica! derivation.
Befare that, however, a more general description of the physical model will be
presented which is valid for all models used in this study, independent of the
definitions of the fluxes. It is referred to as general Homogeneaus Partiele
Model.
Chapter 11/ 24

3.3.2. PHYSICAL DESCRIPTION OF GENERAL MODEL

As already noted, activated carbon is a highly microporous adsorbent. lts


structure consists of elementary microcrystallites of graphite, which are
stacked together in random orientation and it is the spaces between the
crystals which form the micropores [Bansal, 1988]. The actual pore size
distribution . depends primarily on the specific conditions of manufacturing of
the carbon, which means that it will be different for the different types of
carbon.
Descrihing correctly this structure in mathematica! terms 1s rather
complicated and leads to the introduetion of several parameters which are
difficult to estimate in a realistic way. Por these reasons, in the general
model presented below, a number of simplifications are made which permit an
easier description of the processes in and around the particle.
In all models in this study three potential resistances to mass transfer are
considered important for the overall rate of adsorption. The models describe a

PARTICLE
c bulk I
I
OI
I
qHJ
I
OI
dI
Figure 3.1 Schematic adsorbent partiele structure with principal mass
transfer processes: external mass transfer (/), internal mass
transport (pore + surface diffusion) (11), actual adsorption (/1/).
Adsorption kinetics models 25

batch system but, apart from the difference in the boundary condition at the
extemal surface, the intrapartiele diffusion is essentially the same for a
partiele in a flxed bed as for a partiele in a batch vessel. The whole process
is illustrated in Fig. 3 .1.

The transfer of the solute molecules through the extemal fluid film is
considered as the first step of the adsorption process. The second step is the
internat mass transport which consists of two processes acting in parallel -
pore diffusion ánd surface diffusion. The third step is the actual adsorption
step which is considered to be fast compared to the diffusion processes, thus,
it is not a rate-limiting factor.

In order to make the physical picture of the model more clear some tentative
explanation about the nature of the above-defined diffusivities will be given.

Extemal mass transfer


The extemal mass transfer is assumed to occur through a thin liquid film
surrounding the partiele and is described by a linear driving force equation
with rate constant k r·

Pore diffusion
Pore diffusion is mainly molecular diffusion which involves a flux through the
liquid phase within the pores. It occurs essentially by the same mechanism as
in gases, but the methods of correlation and prediction are not so accurate
due to a less well developed fundamental theory [Reid, 1986] and to the
usually complicated structure of the adsorbent.
Due to the random orientation of the pores and the unequal pore diameters the
diffusion coefficients in the pore liquid will be smaller than they would be
in a straight cylindrical pore. Both influences are commonly accounted for by
a "tortuosity factor" ('t) [Ruthven, 1984]:

Dp = Df I t (3.1)

where D f is the free liquid diffusivity.


This "tortuosity factor" is difficult to estimate because it depends, for
example in the case of activated carbon, on the way the adsorbent has been
Chapter 111 26

produced and activated and may differ substantially from one carbon to
another. Dullien [1975] reported experimentally detennined values within the
range 2 - 6. Satterfield [1968] studied the diffusion properties of a great
number of commercially manufactured pelleted catalysts. Almost all tortuosity
factors found were in the range from 3 to 7, however, these values include
also the porosity factor (e ) for the particle.
pa

There is a general correlation between tortuosity and porosity (tortuosity


varies inversely with porosity) and higher tortuosity factors are found for
low-porosity pellets [Ruthven, 1984]. The activated carbon pellets used in
this study had a measured porosity (e ) of 0.57 m3 pore volume I m3 partiele
pa
volume. This value is rather low compared to porosity values for many other
types of carbons, publisbed in literature [Van Lier, 1989].
In view of the foregoing and due to the fact that no experimental data about
the tortuosity factor was available, the value of 7 was used for 't in this
study.

Surface diffusion
Surface diffusion is considered to take place after the adsorption step as the
adsorbed molecules move along the pore walls. It also means that the adsorbed
molecules will possess a certain degree of mobility along the walls of the
pores. Although this mobility will probably be much smaller than in the liquid
phase, the concentration is much higher, which means that a serious
contribution to the flux is possible.
As in the case of D . the surface diffusivity, D , will also be an effective
p s
one. Although tortuosity factors can be defmed in a similar way as for pore
diffusion, the values for the two factors can be different [Satterfield,
1970]. This was confrrmed by Horiguchi [1971] who found evidence that solid
surfaces have irregularities of molecular size which would not influence
molecules diffusing in the pore volume but which increase the path of
molecules diffusing along the surface. It should be pointed out that
estimating D with an equation similar to Eq. (3.1) is not realistic, because
s
in the case of surface diffusivity there is no term, equivalent to the D f in
molecular diffusion, which could be experimentally detennined.
Adsorption ldnetics models 27

As noted, in the general model (i.e. in all models) the actual adsorption step
is assumed to he infinitely fast and fully reversible. This means that on
local level there is equilibrium between the adsorbate molecules on the pore
walls and those in the pore liquid and that the overall rate of adsorption
depends only on the rates of mass transfer.
Some additional assumptions were also made in order to properly formulate the
model in its present form. These assumptions are:
- all particles are considered homogeneaus and isotropie with respect to
their pore structure
- all particles are of spherical shape and have the same radius
pore diffusivity is independent of concentration
- surface diffusivity can he independent or dependent on the solid
phase concentration (carbon loading)
- the whole adsorption process is isothermal

3.3.3. MATHEMATICAL DESCRIPTION OF FICKIAN MODEL

In order to have a complete mathematica! description of the Fickian


(HPMF) model several equations are necessary:
mass balance of adsorbate inside the partiele
mass balance of adsorbate in the bulk solution or overall mass balance of
adsorbate
- initia! and boundary conditions
- equation for the adsorption isotherm

The mass balance in the bulk solution and the initia! and boundary conditions
are different for a packed-bed or batch systems, while the mass balance inside
the partiele and the isotherm hold in both cases. As already noted, in this
study all models are applied to batch adsorption systems.

Mass balance of adsorbate inside the porous partiele


To derive the differential equation descrihing the mass balance inside the
particle, the mass balance over a shell of the spherical carbon partiele is
considered. The partiele consists of pores with a volume fraction of e pa and
Chapter III 28

solid carbon part with a volume fraction of 1-e .


pa
First, to describe the molecular diffusion in the pores, a relationship is
used where the driving force is the concentration gradient in the pore volume.
The flux can be represented as follows:

acp
JP = - e D (3.2)
pa p ar

where
J : kg of adsorbate diffusing in the positive direction of r in
p
the fluid phase per unit time per unit of partiele area
perpendicular to r, i.e. mass flux with respect to
1 2
rnass-average velocity (kg s- m )

C : concentration of adsorbate in the pore of the actsorbent


p
partiele, kg of adsorbate per unit pore volume (kg m- 3 )
r : radial coordinate of the actsorbent partiele as measured from
the center of the partiele (m)
D : effective diffusivity for adsorbate in the fluid phase within the
p
2 1
pore of an actsorbent partiele (pore diffusivity) (m s- )
3 3
e : porosity of partiele (m of pore volume per m of
pa
partiele volume)

The pore diffusion process, expressed by Eq. (3.2), as well as the one for
surface diffusion further in this section, are both in tenns of the Fick
fonnulation, which is a well known constitutive relation for the diffusion
flux, j.
It may be useful to note bere that according to the defmition by Bird [1960]
the mass flux, n., with respect to stationary axis is given by:
I

n = Oln + J. (3.3)
i t i

where j. is the mass flux of species i with respect to rnass-average velocity,


I

Ol. - mass fraction of species i and n. and n are the mass fluxes of species i
I I t
and the total mixture, respectively, with respect to a stationary axis.
However, it can be shown that for dilute-solution systems the diffusion mass
flux n. with respect to a stationary coordinate will be equal to the diffusion
I
Adsorption kinetics models 29

mass flux j.1 with respect to the rnass-average velocity. This is due to the
fact that in n.1 the term containing the mass flux of the .dissolved component,
resulting from the bulk motion of the fluid, is much smaller than the flux
resulting from diffusion.

As a next step in the model description it is assumed that after being


initially adsorbed the molecules have a certain mobility and move along the
surface. So this surface migration may be represented in terms of the gradient
of the amount of adsorbate on the pore walls surface. The flux can be
expressed as follows:

aQ
j = (1 -E )p D - (3.4)
s pa c s ar

where
js kg of adsorbate diffusing in the positive direction
of r in the adsorbed phase per unit time per unit of
partiele surface, i.e. mass flux with respect to
rnass-average velocity (kg s · 1m. 2 )
Q : concentration of adsorbate in the adsorbed phase or carbon
loading (kg of adsorbate per kg of partiele) (kg kg- 1)
D effective diffusivity for adsorbate in the adsorbed
phase (m 2 s · 1)
pc : carbon density (this is the density of carbon in
water) (kg m · 3)

Here D is an effective surface diffusivity in which tortuosity 1s accounted


s
for. Note that p
~
= (1 - E )·p
pa c
, where p
~
is the partiele density.

Further, the mass-balance on the adsorbate in the element of volume from r to


r+& over the period of time from t to t+Ät is given by:

t+Ät

I [ 4m2(j/ js) Ir,t - 4m2(j/ js) r+&,t] dt I =


t
Chapter lil 30

r+~r
2
4m [cpa Cp+ (1-E ) p QJ j
I r
[
pa pa t+~t
,r

[ Epa Cp + (1-Epa ) ppa QJ It,rJ dr

From the niean value theorems for integral and differential calculus [Kreyszig,
1988], foliowed by the limiting process where & and ~t go to zero and also
taking into account that r and t were arbitrarily chosen it follows:

a [r
2
(jp + j)] a [cpa Cp + (1-E
pa
) pc Q]
r2 (3.5)
ar at

Then we have:

a
r
2 ar [r\jp + js)] = ~t[ Epa Cp + (1-Epa ) pc Q] (3.6)

When j and j from Eqs. (3.4) and (3.6) are substituted in Eq. (3.8) we get:
p s

ac ac aQ
- 1 a [ r 2 rE D _P + (1-€ )p D aQ ]] =E _P + (1-E )p (3.7)
r 2 ar l pa p ar pá c s ar pa at pa ca t

As local equilibrium has been assumed, C and Q are related to each other via
P aQ
the adsorption isotherm. When the chain rule is applied to the terms - and
aQ at
it follows:
ar
aQ aQ ac p
. aQ aQ· ac p
---- and = ---- (3.8)
ar ac p ar at ac p at

where ~8 •
p
can be considered as the derivative of the adsorption isotherm at

pore liquid concentra ti on C and Q * is the carbon toading in equilibrium with


p
Adsorption kinetics models 31

the pore concentration.


Equation (3.8) is valid for a single-component system, where a
single-component isotherm is used. However, when we have systems with two
components (i.e. comp.l - comp. 2 - water) the carbon loading of each
component will be a function of the pore concentrations of both components
(via the isotherm). Therefore, for these systems, Eq. (3.8) transforms to:

aQ I aQ*I ac
_ _p,l
aQ*I ac
_ _p,2
= + (3.9)
ar ac p ,I ar ac p,2 ar

and

aQ I aQ*I aQ*I ac
_ _p,2
'I
= + (3.10)
at ac at ac p,2 at
p' I

A similar set of equations will hold for the loading of the second component.

Substitution of Eq. (3.8) in Eq. (3.7) for the single solute system results
in:

ac 1 a
_ _P

at
1 20
ar [ r app ar ] (3.11)
[ 1 + {1-€ pa )/€ pa pc

where D , the apparent diffusivity, is given by the expression:


app

(1-e ) aQ*
D = DP + _=--""p_a p D (3.12)
app epa c s ac p

On basis of Eq. (3.7) different models can be formulated, two of which are
used in this study.
In the first (HPMC) model D is considered to be concentration independent.
app
The argument for such a choice was put forward for the first time by Barrer
[1967] and such a model was used also by Van Lier [1989]. The reasoning behind
such a choice is the following: When the adsorption isotherm equation used is
Chapter I/I 32

highly non-linear and of the "favorable" type, which is the case with the
Radke-Prausnitz isotherm used in this study, then the derivative will not be a
constant and will deercase with increasing carbon loading. When it is assumed
that the surface diffusion coefficient increases with increasing loading, then
the opposite changes of D and the isotherm slope will balance each other [Van
s
Lier, 1989]. D is considered to be practically concentration independent and
p
then the term D remains as a whole constant.
app
The described model is mathematically identical to a model with constant D .
p
This becomes obvious from Eq. (3.7) when the term containing D is nottaken
s
into account.

In the second (HPMF) model (Fickian) a concentration dependent D is


app
assumed. In this model both D and D are constant, while D is still
s p app
variable, due to the changing isotherm slope (cf. Eq.(3.12)). In Section 3.3.5
a more detailed description of the Fickian model equations will be presented.

Overall mass balance of the adsorbate in a batch


In addition to the mass balance inside the partiele the overall mass balance
of the adsorbate has to be considered. This mass balance reads:

R
41t J pa
(E
pa
2
Cp + (1-epa )p cQ) r dr (3.13)
0

where:
C:: : initial concentration of the adsorbate in the bulk (kg m -3)
Cb : concentration of the adsorbate in the bulk (kg m -3)
V : volume of the liquid in the batch system (m 3)
M : amount of adsorbent in the batch system (kg)
kf : extemal mass transfer coefficient (m s- 1)
R : partiele radius (m).
pa

Initial and conditions


Two initial and two boundary conditions are necessary m order to make
Adsorption kinetics models 33

complete the model description. These conditions are slightly different for
the constant diffusion and Fickian models.

First, the amount of adsorbate teaving the bulk solution is equal to the
amount that enters the carbon particles. This can he expressed as:

mAk
sf
(C b Cp IR ) = j IR
-mAsp - mAs js IR (3.14)
p p p

2 1
where A is the specific outer surface of the adsorbent (m kg' )
s

Substitution of Eqs. (3.2), (3.4) and taking into account the expression for
D in Eq. (3.12) gives:
app

k f (C b - cp IR ) (3.15)
pa

Equation (3.15) couples the flux of the adsorbate leaving the bulk solution to
the flux of the adsorbate entering the particle. This expression is the
boundary condition of the differential equation descrihing the mass balance
inside the partiele for r = Rpa. In the case of the Fickian model the apparent
diffusivity in Eq. (3 .15) . is variable, while for the HPMC model it is
constant.

The other boundary condition of this differential equation at r =0 follows


from the symmetry of the spherical particle:

ac
PI - (3.16)
ar o

The initial conditions (at t = 0) are:

(3.17)

cp = Q = 0 for 0 < r < R


pa
(3.18)
Chapter /I/ 34

3.3.4. FULL SET OF EQUATIONS FOR HPMC MODEL

The two mass balance equations which, together with the boundary conditions,
Eqs. (3.15) and (3.16), the initial conditions, Eqs. (3.17) and (3.18) and
the chosen adsorption isotherm fully describe the HPMC model, are:

ac p D
app 1 a [r 2 ac P J (3.19)
at
(1-E pa )/E pa P c aa8*) r p
2
ar ar
R
m 1
pa
4
1t
-
R3 p V
41t J <E
pa
C
p
+ (1-E )p Q
pá c
2
) r dr (3.20)
pa pa 0

In the constant diffusivity (HPMC) model and in all other models discussed in
this study highly non-linear equilibrium relationships are used. This makes it
practically impossible to find an analytica! solution and so it is necessary
to use some kind of numerical method. More details are given in chapter 5.

3.3.5. FULL SET OF EQUATIONS FOR FICKIAN MODEL

In the Fickian (HPMF) model the diffusion coefficient D , as defined in Eq.


app
(3.12), is not constant anymore, but is a function of the carbon loading. This
makes it necessary to have a somewhat different mass balance inside the
partiele and also changes the outer boundary condition (3.15), which becomes:

(3.21)

The full set of equations descrihing the HPMF model, together with the
boundary conditions (3.21) and (3.16), the initial conditions (3.17) and
(3.18) and the chosen adsorption isotherm, reads:
Adsorption kinetics models 35

ac =
_ _P
1
2
ac
[ r 0 app (Q) _
ar
P ] (3.22)
at aag • J
[ 1 + (1-e pa )/e pa pc
p

R
C =C
b
0
b
m
41tR3p
1
V
41t J pa
(e C + (1-e )p Q ) r dr
pa p pa c
2
(3.23)
pa pa 0

An interesting point for discussion is the concentration dependenee of D .


app
As mentioned, D , as defmed in the Fickian model, is variabie even when D
app p
and D are both constant, due to the changing isotherm slope. At the same time
s
the surface diffusivity, D, itself may be a function of different parameters.
s
This last assumption has been a subject of extensive studies by many
researchers. Numerous empirica! and semi-empirical relations have been
proposed where D was correlated as a function of adsorption capacity
s
[Komiyama, 1974], the adsorbed phase concentration [Suzuki, 1982],
[Neretnieks, 1976c], the fmal equilibrium amount adsorbed [Sudo, 1978], the
pore liquid concentration [Lin, 1990] or the degree of saturation [Mamchenko,
1979]. Others tried to derive thermodynamically a relation between D and Q
s
based on the Freundlich isotherm where D and Q were determined for each
s
adsorbate individually [Muraki, 1982].

In the present study a limited amount of calculations were performed with a


special version of the HPMF model where a power law dependenee of D on the
s
carbon loading was used:

D
s
= D
so
Qa (3.24)

where:
2 1
D : surface diffusivity for a carbon loading equal to one (m s- )
so
: exponent (-)
Chapter 111 36

In this case the equation for the apparent diffusivity becomes:


(1-e ) aQ·
D =D + pa p D Qa (3.25)
app p € pa c sO aC
p

The simulation results are discussed in chapter 6.

3.3.6. TWO-SOLUTE CONSTANT DIFFUSIVITY MODEL

Earlier in this chapter, when discussing single-solute adsorption, it was


implicitly assumed that the second solute in the system, i.e. water, does not
affect diffusion on the intemal pore surface (surface diffusion). The term
"two-solute system" is used in this study to designate the presence of two
different organic substances in dilute water solution.

The set of equations for the two-solute Constant diffusivity (HPMC) model look
basically like Eqs. (3.22)-(3.23) and therefore will not be repeated bere. The
main difference is that a muitkomponent isotherm should be used which
influences the calculations of the partial derivatives of the pore
concentrations with respect to time and radius (compare Eqs. (3.8) with (3.9)
and (3.10)). In this model possible interactions between the solutes during
diffusion (pore or surface) are not taken into account (no cross-diffusion
coefficients).

3.4. MODELS BASED ON MAXWELL-STEF AN DIFFUSIVITY

3.4.1. INTRODUCTION

In Section 3.3 the Fick formulation for diffusion was used to derive a
mathematica} model, the Fickian (HPMF) model, that included extemal, pore and
surface diffusion mechanisms. In this section the same physical picture and
assumptions for the partiele structure will be used, but a model will be
derived by adopting a different approach in formulating the driving forces
Adsorption kinetics models 37

inside the porous particle. In this latest approach the theory of irreversible
thermodynamics is used as a basis for defming the necessary equations, which
are in the form known as the generalized Maxwell-Stefan equations. Later in
the chapter the model will be extended to multicomponent systems.

3.4.2. THEORETICAL BACKGROUND

The theory of Irreversible Thermodynamics identifies the various driving


forces which can cause relative motion of molecular species and also provides
a relationship between the driving forces exerted on the species and the
veloeities at which these species are caused to move in relation to the
mixture [Krishna, 1987].
Even when using the theory of Irreversible Thermodynamics there are different
ways of formulating the flux-driving force equations. One possible way is
based on the generalized Maxwell-Stefan equations, which approach bas been
extensively used by Krishna [Krishna, 1977], [Krishna, 1979] and [Standart,
1978]. Another way is the Onsager formulation of irreversible thermodynamics,
introduced by Ash [1967]. Ho wever, neither the generalization of the Fickian
concept (used in the HPMF model), nor the Onsager approach offer a systematic
way for prediction of multicomponent mass transfer on basis of
single-component mass transfer [Krishna, 1990]. In the present study the
Maxwell-Stefan approach was chosen as an alternative to Fick's concept and
comparisons are made between these conceptually different models.

A systematic and at the same time practically-oriented application of the


Maxwell-Stefan equations to single and multicomponent mass transfer is due to
[Krishna, 1986]. This same author publisbed in 1987 a unified theory of
separation processes based on Irreversible Thermodynamics (IT) [Krishna,
1987]. Before deriving the concrete flux equations for the Maxwell-Stefan
model used m this study, a few basic points of the Irreversible
Thermodynamics theory will be discussed, which are necessary for the better
understanding of the model equations .

The main goal of Irreversible Thermodynamics 1s to extend the classica!


thermodynamics theory and study systems where non-equilibrium processes are
taking place. One important postulate from the theory of Irreversible
Chapter Ili 38

Thermodynamics concerns the state of "local equilibrium" [Krishna, 1986]:


"Departures from local equilibrium are sufficiently small that all
thermadynamie state quantities may be defmed locally by the same relations as
for systems at equilibrium".
This postulate is necessary in order to relate the entropy to such quantities
which can be measured, i.e. temperature, pressure and composition.
Another important postulate states that "the fluxes are linearly and
homogeneously related to the driving force" [Lightfoot, 1974]. This postulate
bas been used by Lightfoot to propose an expression for the overall driving
force.
In the case of separation processes, e.g. adsorption, the theory of
Irreversible Thermodynamics can be used to define the different driving forces
which can cause relative motion of molecular species. It can also give the
relationship between these driving forces and the veloeities with which the
species move in relation to the mixture.
One very important relationship in the theory of lrreversible Thermodynamics
gives an expression for the rate of entropy production which is caused by the
relative motion of species during a separation process [Krishna, 1987]:

n
cr [ c RT d .. (u .. u ) ~ 0 (3.26)
T i=l t ) I

where:
cr rate of entropy production (J m-3 s- 1 K 1)
T : absolute temperature (K)
c : molar concentration of mixture (kmol m- 3)
t
1 1
R : universa} gas constant (J kmor K )

d :generalized driving force for the motion of


species i relative to the mixture (m- 1)
u. :velocity of diffusing species i with
I

respect to stationary axis (m s- 1)


u : average molar velocity of mixture with
respect to stationary axis (m s" 1)
n : number of species in multicomponent mixture
Adsorption kinetics models 39

In Eq. (3.26) the term cRTd. gives the driving force which is exerted on
I I
component i per unit volume of the mixture, which driving force moves
component i with respect to the mixture with a relative velocity (u.- u).
I

When there is no driving force there is no relative motion and the mixture is
at thermodynamic equilibrium:

a =0 (equilibrium) (3.27)

The theory of lrreversible Thermodynamics considers the overall driving force,


as defmed in Eq. (3.26), to he the sum of three driving forces which come
from: a) chemica! potential gradients; b) pressure gradients; c) electrostatic
potential gradients for electrically charged species. An expression for the
overall driving force has been proposed by Lightfoot [1974]:

x . ( <p. - 00.) F
d = - -1 V J.t + -
1 1
--- Vp + x.z. V<I> (3.28)
i RT T,p i c RT 1
I RT
t

where:
1
x.I : mole fraction of species i (mol mor )

p : system pressure (N m-2)


zi : charge of species i (-)
F : constant of Faraday (C kmor 1)
J.ti : molar chemical potential of species i (J kmor 1)
<I> : electrostatic potential [V]
3 3
<pi : volume fraction of species i (m m · )
1
oo. : mass fraction of species i (kg kg· )
I

As already noted, the flux-driving force relationship based on lrreversible


Thermodynamics can he formulated in different ways. When the Maxwell-Stefan
approach is used this relationship is called the constitutive relation and can
he written in two equivalent forms [Krishna, 1987]:

n - u.) n (x.N. - x.N.)


d.
I
=I j=l D
I
= L I J •
D
J I
' i ;;: 1, 2, .... , n (3.29)
ij j "''
j~i
ct ij
j~i
Chapter /IJ 40

where:
N.,N. : molar fluxes of species i, j, with respect to
l J
stationary axes (kmol m- 2s- 1)
Maxwell-Stefan diffusivity of i-j pair in
·
muI tlcomponent ·
nuxture ( m2s·I)
cI molar concentration of mixture (kmol m· 3 )

In principle, Eqs. (3.26) and (3.28) are the two general equations m the
Maxwell-Stefan approach and form the basis of the unified theory of separation
processes, put forward by Krishna. These equations will be applied to the
specific systems modeled in the present study and the results will be compared
with the Fickian models.

As noted, the MS-type of model will be referred to as the Homogeneous Partiele


Maxwell-Stefan Model (HPMS), in contrast to the Fickian one (HPMF). Both
models assume a variabie apparent diffusivity but constant pore and surface
diffusion coefficients. The model with constant D is exactly the same for
app
both descriptions and the sets of equations were already given in Section
3.3.4.
First, a description of the physical picture behind the MS-driving force-flux
equations will be given, foliowed by a mathematica! denvation of the
equations, both for pore and surface diffusion.
Note, once again, that systems which contain one organic component and water
are referred to as single-solute systems, while those containing two organics
plus water are called two-solute ones. When discussing general equations and
systems the terms binary or temary mixtures are used to describe any system
containing two or three species, respectively.

3.4.3. MAXWELL-STEF AN PORE DIFFUSION

The physical basis of development of the Maxwell-Stefan diffusion equations


for bulk (pore) diffusion can be presented in a relatively simple way. Note
that our organics-water systems are very dilute, so for the so-called
two-solute systems (i.e. solute 1 - solute 2 - water), when descrihing pore
Adsorption kinetics models 41

diffusion, we assume that no interactions occur between the molecules of the


two organic compounds. The only possible interactions are solute 1 - water and
solute 2 - water. Keeping this in mind we can use the following schematic
picture of molecular interactions, for example between solute 1 and water:

V2

V1

Figure 3.2. A simpte physical model for descrihing the molecular


interactions between water and organic molecules.
Adapted from artiele by Krishna (1992).

We assume that the only driving force in the mixture is the gradient in the
chemical potential. The reason for this assumption will be discussed later, in
the paragraph containing the mathematica! description. Now, as the molecules
of solute 1 move through the aqueous salution they meet resistance which is
caused by the hydrodynamic drag exerted on them. Although the drag on an
individual molecule is not large, the total friction can be significant due to
the enormous amount of molecules in one mole. The friction between the
molecules of water and solute 1 is assumed to be proportional to the velocity
difference between the two species.
This physical description was proposed by Krishna [1987]. It fits rather well
to our systems in the case of pore diffusion, however, things become more
complicated for surface diffusion.
Chapter 1// 42

The starting point in applying the Maxwell-Stefan approach will be the use of
Eqs. (3.28) and (3.29). For the adsorbent-adsorbate systems modeled in the
present study it seems Jogical to assume that pressure gradients and
electrostatic potential gradients would be sufficiently small so as to ignore
them. Thîs means that in Eq. (3.28) the terrns Vp and V<I> cancel and the only
driving force will be the gradient in the chemica! potential:

x
d, __i V 1..1. (3.30)
I RT T.p i

In further notations the subscripts T and p will be omitted from the term
VT.p1..1.I.• although the conditions remain the same (T and p are assumed to be
constant).
Hence, for a temary mixture, on the basis of Eqs. (3.30) and (3.29) it holds:

x x x (u u2) xlx3(ul- u3)


I 2 1
I v~..~. = (3.31)
RT I DI 2 DI 3

x x x (u u I) x x (u - u)
_2 v~..~. 2 I 2 2 3 2
(3.32)
2
RT D 21 D 23

x x x (u - u ) x x (u - u2)
v~..~. 3 I 3 1 3 2 3
-··-3
3 = (3.33)
RT D3 I D 32

On the basis of the Gibbs-Duhem equation (Walas, 1985], at constant T and P,


we have:
0 (3.34)

which means that only two of the three Eqs. (3.32) - (3.34) are independent.
And while for a binary system this restrietion is sufficient to prove that D 12
= D , for a temary system it is necessary to postulate that [Krishna, 1986]:
21
Adsorption kinetics models 43

D21 .• D
l3
= D
31
·
'
D
23
= D
32
(3.35)

In terms of the physical picture presenled above and in view of Eqs (3.31)
(3.33) the friction between any two molecular species, for example 1 and 2,
per mole of 1, is proportional to the concentra ti on of molecular species 1 and
to the velocity difference between 1 and 2. The same holds for species 1-3.
Further, the driving force on species 1 is balanced by the friction between
&'Pecies 1-2 and 1-3 .
It is also evident from Eqs. (3.31)-(3.33) that in order to obtain the
absolute veloeities of all three species, it is necessary to add a different
type of relation to the two existing independent ones. In literature this
additional relation is known as the boot-strap relation and gives the
veloeities within the chosen frame of reference [Wesselingh and Krishna,
1991].

Let us deal for a moment with another type of "physical picture problem" which
is also important for the correct interpretation of the model results. As
already noted, the systems used in this study are called single-solute and
two-solute systems. However, in reality, all of these systems contain one more
component, water. Let us assume for a moment that we have a temary mixture
where components 1 and 2 are the organic substances in a dilute water solution
and component 3 is the water and let us choose Eqs. (3.31) and (3.33) as the
two independent equations. Taking into account the physical picture of the
homogeneous particle, as described in Section 3.3.2, let us consider first
diffusion in the pore volume of the adsorbent partiele which can be treated as
molecular diffusion. As we are dealing with a dilute solution, it appears
reasonable to suppose that the molar fractions of solutes 1 and 2, x 1 and x 2 ,
will be very much smaller than the molar fraction of water, x • On the other
3
hand, the veloeities u and u will be of comparable magnitude, while the
1 2
velocity of water, u , will be much smaller. This means that the following
3
will hold for the right-hand sides of Eqs. (3.31) and (3.33):

x x (u - u ) << x x (u u ) ; (3.36)
121 2 131 3

x2x1(u2 - u l ) << x23


x (u 2- u3) (3.37)
Chapter lil 44

Eqs .. (3.36) and (3.37) indicate that the mutual influence of solutes 1 and 2
can he neglected compared to the friction with water.
Hence, for the temary systems used in the present study (referred to as
two-solute systems), we can assume that pore diffusion is described by a
reduced form of Eqs. (3.31) and (3.33), namely:

x x x (u - u )
Vj.! I 3 I 3
d _I (3.38)
I 1
RT D
I3

x xx(u u )
d = - -2 Vj.!
2 3 2 3
(3.39)
2 RT 2

where D and D are the individual free liquid diffusivities of solutes 1


13 23
and 2 in water, corrected with a tortuosity factor.

For the case of binary mixtures (referred to as single-solute systems)


imptementing the Gihbs-Duhem equation leads to the sum of the driving forces
being zero and, consequently, to the two diffusion coefficients being equal.
This conclusion, however, is not so ohvious and requires some more attention.

One of commonly used forms of the Gibbs-Duhem equation reads as follows:

I x.I Vil.=
1
o (3.40)

This equation was already used for a temary system (cf. (3.34)). It will he
shown bere how Eq. (3.40) can he applied to a binary system.
For binary mixtures Eq. (3.40) can he used in the following form (d is used
instead of V for simplicity) [Walas, 1985]:

dj.!
2
= - x 1dj.!/(1
I
- x )
I
(3.41)

In turn, Eq. (3.43) can he used also in another form. namely


Adsorption kinetics models 45

all2
x
1
+ x2 = 0 (3.42)
ax ax
1 I

On the other hand dJ.! (or VJ.1 ) in Eq. (3.38) and dJ.! in Eq. (3.41) are not the
I I I
same, as the first one is the differential with respect to mole fraction and
the second one is the differential with respect to a space coordinate (for our
systems this is the radial coordinate, r). But as the chemical potential is a
function not only of r but also of x, we can write:

all• all2 all I ax l all2 ax I


x1 + x2 x -+x
ar ar 1
ax I ar 2 ax ar
1
(3.43)
ax 1
= [ x aJ.11 + x2 aJ.12 )
1 ax ax 1 ar
I

which, in view of Eq. (3.42), proves that from Gibbs-Duhem equation, for a
binary mixture, it follows directly that the sum of the two driving forces is
zero.
Therefore, there will be only one independent flux-driving force equation in a
binary system and taking this into account, Eq. (3.30) reduces to:

x x x (u u)
d1 = 1 VJ.1
RT 1
= 1 2

D
1 2
(3.44)
12

where the subscript 1 is for the organic compound and 2 for the water. For
pore diffusion. D will be equal to free liquid diffusivity of the solute,
12
corrected for the tortuosity factor.

3.4.4. RELATION BETWEEN FICKIAN AND MAXWELL-STEFAN


DIFFUSIVITIES

It is not difficult to prove that the two diffusion coefficients, D (MS) and D
(Fickian), are related through a special factor. This is true for any system,
Chapter lil 46

that is why it will be shown for a binary system, described by Eq. (3.44) and
the derivations will be made in general terros with vector notation.

From thermodynamics it follows that the chemica} potential f.l of component 1


I
(or 2) can be given by the equilibrium relation:

0
f.l 1 . = f.l 1 + RT ln(a I ) (3.45)

0
where a is the activity of component 1 in equilibrium with the mixture and f.l
I I
is the chemical potential of component i at the chosen standard state.
Further, using the d notation for the driving force and mass defmitions instead
I
of molar ones, from Eq. (3.44) we get:

(3.46)

1 - ro , part of the numerator in Eq. (3.46) can be rewritten as


I

ro2(u 2- u I) = ro22
u - ro21
u

(1 - ro 1)u 1 = ro2u2 + ro 1u I - u I

u- u 1 (3.47)

where u is the rnass-average velocity.

Further, taking into account that ro C I Cp, t (where Cp, t is the total
I p,l
mass concentration in the pore volume) and substituting Eq. (3.47) in Eq.
(3.46) we find:

- C
p, t
D
12
d
I
= C
p ,I
(u - u)
I
j (3.48)
1

where j is the mass diffusion flux, according to the basic definition by Bird
1
[1960].
Adsorption kinetics models 47

When the defmition for d from Eq. (3.44) is used, by keeping the vector
1
notation and using mass defmitions, substitution in Eq. (3.48) gives:

rol
j
I
= - C
p, t
D
12
-
RT
VJ.!
I
(3.49)

Further, on basis of Eqs. (3.49) and (3.45) we have:

j
1
= - C
p. t
D
12
ro
I
Vln(a )
I
(3.50)

Por part of Eq. (3.50) the following holds:

Vro I + ro I Vln(y)
I
(3.51)

where y is the activity coefficient based on weight fractions which in the


I
general case is not constant.
Then, from Eq. (3.51) it follows:

ro VIn( a) = Vro [ 1 + aln( 'Y ,) ] (3.52)


1 1 1
aIn( ro I )

Finally, taking into account the general form of Fick's first law

jI = - Cp, t
D12Vro I (3.53)

then substituting Eq. (3.52) in Eq. (3.50) and comparing the result with Eq.
(3.53), the relation between the Fickian and MS diffusivities becomes obvious:

(3.54)

The term in front of the MS diffusivity is designated by r and is called a


thermodynamic factor . Evidence has been found that for highly non-ideal
Chapter liJ 48

mixtures the r factor is normally a strong function of the mixture composition


and vanishes in the region of the critica} point [Krishna, 1987]. According to
the interpretation given by Krishna, the Maxwell-Stefan diffusivity has a
fundamental physical meaning of an inverse drag and is more easy to interpret
and predict than the Fickian diffusivity. The Fick diffusion coefficient, D 12,
combines two different concepts which are drag effects and thermadynamie
non-ideality effects and is much more concentration dependent
If we are dealing with a dilute solution, i.e. the activity coefficient, y ,
I
is constant, we have:

D,z = D 12 (3.55)

where the two coefficients are identical.


This last assumption is made in the models in this study with regard to pore
diffusion.

3.4.5. MAXWELL·STEFAN SURFACE DIFFUSION

For the dilute-solution systems used in this study, for diffusion in the pore
volume, it is reasonable to assume that Eq. (3.55) is valid. This is certainly
not the case for surface diffusion where high concentrations of the organic
compounds on the pore wall surface exist.
Two slightly different approaches in descrihing surface diffusion will be
discussed below, the main difference being the way in which the concept of
vacant sites (on the carbon pore wall) is treated.

A serious attempt to describe multicomponent surface diffusion in adsorption


from gas phase based on the Maxwell-Stefan equations is due to Krishna [1990].
The approach considers the vacant sites as the (n+ l)th component in the
diffusing mixture. All vacant sites are treated as equivalent, independent of
their location on the adsorbent surface. The analogs of the bulk phase mole
fractions are called occupancies or fractional coverages. The surface mole
fractions, x8 , are related to the fractional coverages {}. by:
I I
Adsorption ldnetics models 49

x~
I
= {), I 1
t)
t
(3.56)

where:
{),I fractional surface occupancy of component i (-)
t) fractional surface occupancy by total mixture (-)
t

This type of physical description for surface diffusion is based on the


so-called hopping model (cf. Gilliland [1974]).

Further, assuming that the adsorbent sites represent a separate component, the
driving force which acts on the adsorbed component i may be expressed in terms
of the Maxwell-Stefan formulations in the following way [Krishna, 1990]:

{),
I Vj.t~ + , i == 1,2, .. n (3.57)
RT
1
C Ds
t, S Î, V

where:
Vj.t~I
1
: surface molar chemica] potential of component i (J kmor )
8
Ni : surface molar flux of component i with respect
to stationary axes (kmol m-' s- 1)
Ns : surface molar flux of component v (vacant sites) with
V

respect to stationary axes (krnol m- 's- 1)


ct,s total molar saturation concentration of
surface (kmol m-2)

; counter-sorption Maxwell-Stefan diffusivity (m 2 s- 1)

: single-component MS surface diffusivity (m 2 s- 1)

which equation is analogous to Bq. (3.31), valid for pore diffusion.

In order to apply the above theory to liquid adsorption it is necessary to


discuss further the type of chemica} potential used in Eq. (3.57). By analogy
with pore diffusion, the j.t used bere should be the respective surface chemica]
potential of the adsorbed species, defined as:
Chapter 1/I 50

(3.58)

where:

Jls : molar surface chemica] potential of adsorbed


I
1
species (J kmor )

0
Jl:'I : molar surface chemical potential of adsorbed species
at the chosen standard state (J kmor 1)
asl : activity of component i in equilibrium with the mixture
on the surface (-)

One of the basic assumptions for all models investigated is the existence of
local equilibrium between the pore fluid and the surface and they are
considered as two distinct phases. From equilibrium thermodynamics it follows
that for a transfer of dn. moles between two phases at the same T and P, the
I

change in Gibbs energy is:

(3.59)

where n is the number of moles of the species in the system and G is the
t
Gibbs energy of the total mass of the system, i.e. G nG, and G is the
t
specific Gibbs energy.
Since G bas a minimum at equilibrium, its derivative will be zero:

(aG 1 an,)
t ' T,p,n
= 0 (3.60)

and from Eq. (3.59) it follows:

(3.61)

The above stated equality means that in the Maxwell-Stefan equations for
surface diffusion the gradient of the surface Jl can be exchanged for the
gradient of Jl in the pores.

Further, in view of Eq. (3.29) and changing to mass definitions, the


Adsorption kinetics models 51

surface mass fluxes n~ (kg m·


1
) of the diffusing components can be defmed
I

as:
0
8

i
= ct's -ö.u.
1 1
(3.62)

2
where C Is the total mass saturation concentration of the surface (kg·m- ).
t. s
The same should be valid for the vacancy flux, n , as well. Therefore, taking
V
into account Eqs. (3.62) and (3.61), rewriting Eq. (3.57) for solute 1 in a
two-solute system and dropping the vector notation, we obtain:

-ö 2 (u I - u2 )
(3.63)
RT dr

where:
ui' u 2 : velocity of diffusing component 1, 2, respectively,
within a stationary coordinate (m s- 1)
uv : velocity of "diffusion" of vacant sites v with
respect to a stationary coordinate (m s- 1)

The system we are consirlering consists of two solutes, so there is an equation


for solute 2, analogous to (3.63). In the presentations that follow only the
equation for solute 1 will be given, but all derivations are also valid for
solute 2.

The fractional surface occupancy for component i, -ö,, can be expressed in


I

terms of the previously defined adsorbent loading, Q.:


I

Qi
-ö. = Qtot
(3.64)
sat

where Q. is the loading for component i and Q101 is the total loading of the
• sat
mixture when all adsorption sites are covered ( Q 101 reflects the saturation
sat
sorbent capacity).
The term -ö represents the fraction of un-occupied, vacant, sites and we have
V
Chapter /// 52

'Ö = 1 - 'Ö - 'Ö - . .. -'Ö . (3.65)


V I 2 n

where 'Ö. was defined in Eq. (3.64).


I

Then, in view of Eqs. (3.64) - (3.65), for the driving force of solute 1 (in
the two-solute system) we obtain:

Qtot Q Q
sa t - I - 2
(3.66)
RT dr Qtot
sat

Hence, in this way it might be possible on the basis of information on kinetic


single-species adsorption and binary adsorption isotherm, to predict binary
sorption kinetics. This would be an important advantage of using the
Maxwell-Stefan formulations for surface diffusion.

Rearranging Eq. (3.66) gives:

Ds Qto t
I ,V s a t
(u - u ) = (3.67)
I V
RT Qtot _ Q _Q dr
sat I 2

It was argued by Krishna [1990] that for surface diffusion the vacancy, v, bas
a vanishingly small molar mass and, although the vacancy flux is considered
non-zero, there will be no contribution to the species mass-balance and the j
and n fluxes will be identical.

In view of the foregoing, by writing a Maxwell-Stefan surface flux equation


for solute 1 in an explicit form, by taking into account the porosity of the
adsorbent partiele as previously defined, and by using Eq. (3.62) we obtain:

Ql
= (1 _ E ) p Qtot u (3.68)
ns,l = Js,l pa c sa t Qtot I
sat
Adsorption kinetics models 53

Substituting Eq. (3.67) in Eq. (3.68) and assuming uv 0, we obtain:

QIO t
sa t
Q ------- (3.69)
1 Qtot _ Q _ Q dr
sat I 2

At the same time, from Eq. (3.45) and under the assumption that i. is constant
I
(we deal with very dilute solutions), we frnd:

dm
= RT _l_
(I}
p ,l
(3.70)
dr p,l dr

and when substituting Eq. (3.70) in Eq. (3.69) and tak.ing into account that
wp.l C
p,l
I C p,t , where C p,t is the total mass concentration in the pore
volume, we obtain:
Qto t D s d(C IC )
sa t I ,V RT l (3.71)
- (1-E ) p Q p,l p,t
pa c I Qtot
sat
(C
p, 1
IC p,t) dr

and, finally:

dC
5 p,l
- (1-E ) p D (3.72)
pa c l,V Qtot dr
sa I

Equation (3.72) represents one possible way to express the surface flux in
terrus of Maxwell-Stefan diffusivities. This definition of the flux is based on
several assumptions, the most important being the already described concept of
vacant sites and vacant site fluxes. Due to the use of Eq. (3.61) we obtained
above an equation with a gradient in the liquid phase concentration, whereas
Krishna uses a gradient in the carbon loading, combined with a thermodynamic
factor. This is not a basic difference and it affects only the way of writing
the equations.
Chapter Ili 54

A somewhat different way of descrihing the multicomponent surface diffusion


based on the Maxwell-Stefan equations is used for the model calculations in
this study. This approach considers the carbon pore wall as the (n+ l)th
component in the diffusing mixture which is analogous to the model of
Wesselingh and Krishna [1991] for membrane processes.
In this second approach, the concept of vacant sites is not considered and it
is assumed that when the adsorbed organic molecules move along the surface
there is always friction between them and the pore wall surface.

In view of the foregoing the driving force which acts on the adsorbed
component i may be expressed in terms of the Maxwell-Stefan formulations in
the following way:

x~ n (x~ N~- x~ N~) (x 5 N~ xs Ns)


l Vj.< = L J
1

Ds
J + c J

Ds
c
,i = 1,2, .. n (3.73)
RT j=r c I, s ij
c i ,c
j:;éi
t ' s

where:
Vj.t~
1
: surface molar chemica! potential of component i (J kn10r )
1
5
N : surface molar flux of component i with respect
i
to stationary axes (kmol m- 1 s- 1)
N5 : surface molar flux of component c (carbon walls) with
c
respect to stationary axes (kmol m- 1s' 1)
c : total molar saturation surface concentration (kmol m·2)
l,s
2
D
5
: counter-sorption Maxwell-Stefan diffusivity (m s- 1)
ij

D
5
: single-component Maxwell-Stefan surface diffusivity (m2s- 1)
i te

Further, in view of Eq. (3.73) and changing to mass definitions, the surface
mass fluxes of the diffusing component nsn (kg m- 1s- 1) of the diffusing
i
components can be defmed as:

c I, S
s
x.J u.1 (3.74)

At the same time, it is obvious that there will be no "carbon flux" if we fix
the axes to the carbon, i.e. the N5 term in Eq. (3.73) will be zero. Taking
c
Adsorption kinetics models 55

this into account and in view of Eqs. (3.61) and (3.74), rewriting Bq. (3.73)
for solute 1 in a two-solute system and dropping the vector notation, we
ohtain

x 8 (u - u ) x u
2
= 1
Ds
2 +-c_l
Ds
(3.75)
RT dr
12 I ,c

where:
u 1, u :velocity of diffusîng component 1, 2, respectively,
2
within a stationary coordinate (m s- 1)

Here the friction hetween water and organic molecules on the carbon surface is
neglected compared to the friction hetween organic molecules and surface.

By analogy with the earlier mentioned memhrane model, we can regard the carbon
as having a constant mole fraction and then we can write:

x
c
= (3.76)

where D 8I is the single-component Maxwell-Stefan diffusivity of component 1 as


used in the model calculations in this study (m2 s- ').

Further, we have two possihilities for interpreting the possible frictions (on
the surface) hetween the molecules of the different solutes in the system.

Firstly, we can assume that different organic molecules do not interact with
each other and in view of Eq. (3.70) we get the following equation for the
velocity:

Ds Ds d(Cp,l /C p,l )
I I 1
u=--- =------- (3.77)
1
RT dr RT (cp,l JC p,t) dr

Then, inserting the newly-defined u I from Eq. (3.77) in Bq. (3.68), we have
Chapter 111 56

for the surface flux:

j
s. I
= - (1-e )
pa
pc D 8I (3.78)
dr

By following the same line of reasoning an analogous equation is obtained for


solute 2:

j
s.2
= - (1-e )
pa
pc D25 (3.79)
dr

Eqs. (3.78) - (3.79) are the fmal form for the Maxwell-Stefan-surface fluxes,
as used in the model (HPMS) without cross-diffusion coefficients.

Secondly, we can assume that there are interactions between the different
organic molecules in the system, so the first term on the right-hand side of
Eq. (3.75) cannot be neglected anymore. Using matrix and vector notatior.
convenience we can write the following equation for both solutes (on basis of
(3.75 and 3.76):

- 1 I RT (VJ.t) = [B] (u) (3.80)

where (u) represents the column vector of velocities, while [B] is a matrix
defmed as:

/D 8I 2 + l/D 8I

l
X
2
[B] = (3.81)
[ - x
I
/D 812

From Eq. (3.80) we obtain for the velocity (u)

1
(u) = 1/RT [BT (VJ.t) (3.82)

If we now switch to mass quantities and write the explicit surface flux
equations in a manner analogous to Eq. (3.78), we obtain Eqs. (3.83):
Adsorption kinetics models 57

1 dC p,l dC
1
js,l = - (1 - e ) p Q [ R (11) + B-1
(12)
p,2 ]
pa pa 1
c p,l dr c p,2 dr

1 dC p,l 1 dC
j
s,2
= - (1 ep) ppa Q2 [ B-~21) c dr
+RI
(22)
c p,2 dr
p,2 ]

p,l

1 1 1 1
where B- ,B- ,B- and B- are the elements of the inverted matrix
( 11) (12) (21} (22}
1
[BT .This is the final form for the Maxwell-Stefan-surface fluxes, which can
be used in a model with cross-diffusion coefficients. Such a model has not
been used in this study.

lf we make now a comparison between the expressions for the surface Fickian
and Maxwell-Stefan diffusivities (cf. Eqs. (3.4 and 3.78)) and in view of Eq.
(3.8), for the relation between the two types of diffusivities we obtain

D (3.84)

where the term in front of D 8 represents the thermodynamic factor r for


surface diffusion.
Equation (3.84) shows that, indeed, the Fickian surface diffusivity is
strongly concentration dependent, while the Maxwell-Stefan diffusivity will
have a much weaker dependenee on the carbon loading. This has been proven
experimentally by Eic [1988] who found that for diffusion of benzene in
zeolites the surface Maxwell-Stefan diffusivity does not depend on the surface
occupancies, i.e. the ratio QJQ . However, it has been also reported [Pope,
sat
1967] that the Fickian surface diffusivity of SO in a mieraporous plug
2
increased sharply with surface occupancy, while the surface Maxwell-Stefan
diffusivity calculated from this data was also found to be dependent on
surface coverage, decreasing with increasing surface coverage [Krishna, 199()].
This was attributed by the author to possible interactions between the
adsorbed S0 molecules, leading to decrease in Ds with increasing surface
2
coverage.
Chapter 1/l 58

In the present study a constant Ds is assumed for the Maxwell-Stefan model.


The results from the model simulations are given in chapter 7.

3.4.6. SINGLE-SOLUTE MODEL EQUATIONS

Except for the definition of the fluxes, which is fundamentally different, the
model equations for the Maxwell-Stefan (HPMS) model are identical to those of
the Fickian model, with the same initia! and boundary conditions and
equilibrium isotherms. Therefore, only the final set of equations will be
given here. Note that in the Maxwell-Stefan model, as in the Fickian one, the
pore and surface diffusivities, DP and D 8 , are assumed to be constant.

The set of model equations for the single-solute Maxwell-Stefan model,


combined with the boundary, initia! conditions and respecti ve adsorption
isotherm (thus, same as for Fickian model) and with Maxwell-Stefan diffusion
coefficients, is:

ac
p = ] (3.85)
at *
aQL
[ 1 + ( 1- epa )/e pa pc a J
p

R
pa
cb = 0
C -
b
41t J pa p pa c
2
(e C + (1-e )p Q) r dr (3.86)
0

where D"PP is given by the expression:

(1
(3.87)

and is a function of the concentration inside the particle. As noted, both


diffusion coefficients, DP and D8 , are assumed to be constant.
Adsorption kinetics models 59

The initia} and boundary conditions for the Maxwell-Stefan model are:

(at t=O) (3.88)


C
p
=Q =0 for 0 < r < R
~

k (C - c
t b P
IR )
pa
= e o•PP(Q)
pa ar
IR
pa
(3.89)
ec
__P
8Q

ar ar lo = 0

3.4.7. TWO-SOLUTE MODEL EQUATIONS

The two types of Maxwell-Stefan model equations discussed above (with and
without cross-diffusion coefficients) differ only in the surface fluxes,
therefore one set of equations will be given and the difference between the
two models will be indicated.

For convenience the equations for the mass balance balance in the partiele
will be expressed in vector and matrix notation:

a
at [
cp (1-€ )
+--'-"pQ
€ c
pa
l = r' V [ r' [D''~ vc,] (3.90)

where [o•PP] is an apparent diffusivity matrix which is in a form camparabie


to a diffusion matrix on basis of Fick's definitions. In the
Maxwell-Stefan-model without cross diffusion coefficient [Dapp] is:
Chapter 1/1 60

Q; Ds 0
[ D: + A L
=
I
Q.
]
p,l
[DaPP] (3.91)
0 DP + A 2 Ds
2 L p,2
2

For the Maxwell-Stefan model with cross-diffusion coefficients holds:

Q. Q.
DP + A I RI
A
1 B-1
1 -c-p,l (11) ··c- (12)
p,l
[Dapp] =
Q:
.
Q2 I
(3.92)

B-I d+ 8
A
L p,l
{ 2 I) 2
A -c-p,2 D 2 B ( 22)

where ( 1-Epa)/Epa pc =A const.

The equations for the overall mass balance are:

R
m
3
l
V
41t I pa(E C pa p
+ (1-E )p c Q ) r dr
pa
2
(3.93)
4/3 1t R pa p pa 0

The systems (3.90) and (3.93) are coupled with the following initial
conditions (at t 0):

(3.94)

cp = Q = 0 for 0 < r < R


pa
(3.95)

and the following boundary conditions:

[k~ (eb - epi Rpa ) = e


pa
rn•PPl vc
p
IR (3.96)
pa
Adsorption kinetics models 61

vc p IR = 0
= VQI R 0
= 0 (3.97)
pa pa

where

kf,l
[k~ = (3.98)
[ 0

To the above equations it is necessary to add the multicomponent adsorption


isotherm equations. In this study basically two types were used and compared -
the improved Radke-Prausnitz isotherm (cf. Eq.(2.13)) and the extended Fritz &
Schlünder isotherm (cf. Eqs.(2.17a)-(2.17b).

The single and two-solute models presented in this chapter were fitted to the
respective experimental concentradon decay curves and the results are
discussed in chapters 6 and 7.
Chapter IV 62

EXPERIMENTAL METHOOS AND RESULTS FOR EQUILIBRIA

4.1. MATERIALS AND INSTALLATION

In this section the structure and properties of the adsorbates and their
mixtures will be presented, followed by a description of the properties of the
activated carbon used. The equipment employed for obtaining equilibrium data
will also be discussed here.

4.1.1. ADSORBATES AND ADSORBATE MIXTURES

Structure and properties


Three adsorbates were chosen for this study: 4-isopropylphenol (IPP), p-nitro-
aniline (PNA) and nitrobenzene (NBZ). Their structural formulas can be
represented as follows:

OH

IPP PNA
óNBZ

Phenols and benzene derivatives are common pollutants in potable and waste
water, and they are most effectively cleaned by activated carbon.
P-nitroaniline is an intermediale in the production of water soluble dyes, and
activated carbon is used in treating dye wastes.

The characteristic properties of the adsorbates are given in Table 4.1.


Components IPP and PNA are quite similar in molecular weight while the
molecule of nitrobenzene is somewhat smaller. However, some calculations about
the exact molecular structures were performed, using a molecular rnadeling
program [Janssen, 1992] which indicated that the molecule of IPP differs
substantially from the other two types of molecules in the space distribution
Equilibrium results and methods 63

of its functional groups. The two methyl groups of IPP are perpendicular to
the plane of the benzene ring, while the molecules of PNA and NBZ are almost
fully placed in the benzene plane. IPP has the biggest molecule, with a
distance between the H atom of the OH group and the furthest H atom of the CH
3
groups equal to 0.80 nm. For the PNA molecule the biggest distance between two
atoms (0 and H) is 0.71 nm and for NBZ 0.61 nm. Although these differences
are not very big, consirlering the fact that the microporous region of the
average type of activaled carbon starts below 2 nm, they might still play a
significant role in the diffusion mechanism within the particle.

Table 4.1. Properties of the selected adsorbates

Adsorbate Mol.weigh( S()IÜbllit}/. Pensity Btdk diffu~vity . M.p.


. ..---s (kQII<mol) •·•· (I<Qim3) (kg/rn3) (m2/s) (Celsius)

IPP 136.19 1.7 990 6.98 62


PNA 138.13 0.5 1424 9.26 149
NBZ 123.10 1.9 1204 9.25 6

*S o lub i! ity
from Landolt- Borns te in; Bulk di ffus ivi ty measured at 21. 9°C,
f o r IPP - by P. Rut ten, Delft Uni vers ity of Technology, for PNA and
NBZ- byF. Kievit, Eindhoven University of Technology

The IPP and PNA used in this study have a purity of 98 and 99+ %,
respectively, and were both obtained from Janssen Chimica. NBZ is of p.a.
quality and was delivered by Fluka.
Two sets of two-component systems were used 4-isopropylphenoVp-nitroaniline
(IPP/PNA) and nitrobenzene/p-nitroaniline (NBZ/PNA).
The concentrations of all three individual compounds and the above given
mixtures were measured with a UV spectrophotometer. PNA and NBZ were chosen
not only because they appear in polluted water streams, but also on basis of
their non-reactivity between themselves and with IPP. Another important
consideration was that the organic compounds in the two-component systems can
be determined independently of each other by UV spectrophotometry.
The possible chemica} reactivity of the three substances with the activaled
Chapter IV 64

carbon surface is difficult to estimate. A limited amount of results on


studying possible irreversible adsorption of IPP on the carbon are given in
Section 4.4.

4.1.2. WATER

The water used in this study was initially demineralized tap water, filtered
through a Milli-Q filter system. The original tap water could not he used
since it contains inorganic salts and organic components that may affect the
adsorplion capacity of the activaled carbon and interfere with the
spectrophotometric determination of the adsorbate concentration.
Later, double-distilled water was used, both for equilibrium and kinetic
experiments.

4.1.3. ADSORBENT

Only one type of activaled carbon was used in this study. This is an extruded
activated carbon, type RBWl, manufactured by Norit, BV for removal of organic
pollutants from industrial as well as municipal waste waters. However, it was
not a real commercial product as special precautions were taken during the
activation process. The total amount of carbon, a rather small part of which
was actually used for experiments, was produced in one single batch in order
to achieve as much uniformity m properties as possible, for all carbon
pellets.
This latest precaution was necessary because, in principle, activated carbons
produced from the same raw material, but by different manufacturing processes
can differ fundamentally in pore structure. In addition to that, another basic
characteristic of activaled carbon, pore size distribution, may still be
different for different particles. This is due to the fact that when there are
a series of activaled carbons produced by exactly the same manufacturing
process and from the same raw material, but with different residence times in
the kiln, differences in pore size distribution are he observed [Van Lier,
1989]. The physical properties of the activaled carbon used are lisled in
Table 4.2.
As noted, lhis carbon is in extrudate form. The lenglh of the pellets varles
between 1 and 4 mm. The diameter given in Table 4.2. is the diameter of a
Equilibrium results and methods 65

sphere having the same surface-to-volume ratio as the average value of the
carbon sample. It can be calculated from the following formula:

dav = 3 uI (4.1)
u+ I
where:
dav :average surface-to-volume diameter (m)
u :average diameter of cylinder (m)
I :average length of cylinder (m)

Table 4.2. Physical properties of the activated carbon

Type RW81
Manufacturer NoritBV
Carbon density (kg/m3) 1840
Surface-to-volume mean diameter (1 E-3 m)
e porosity (m3/m3) *

* 3 3
Por os i ty is m of pore volume per m of partiele volume

4.1.4. EQUIPMENT FOR EQUILIBRIUM EXPERIMENTS

Shaking machine
The equilibrium adsorption isotherm data were obtained using a Lab-Line Orbit
Environ-Shaker, Model 3528-1. This shaking unit permits the use of
temperatures from slightly above ambient to approximately 60°C. The heating
agent is air, which makes it clean and easy to operate. As most equilibrium
experiments in this study were performed at 20 ± 0.5°C, an optional cooling
coil was used to achieve stabie temperature controL

The adsorbate-adsorbent mixtures were put in 250 ml stoppered Erlenmeyer


flasks and were placed on the shaking platform, which has special vessel clips
to accommodate for up to 25 flasks. The closed flasks stayed as a minimum
about 3-4 weeks to ensure that equilibrium had been reached.
Chapter IV 66

Spectrophotometer
In the present study, absorption spectrophotometry was used for quantitative
determination of the solute concentrations. A Beekman DU-64 single-beam
speetrophalometer was used for all kinetic and most of the equilibrium
experiments. 1t has two sources: visible (325 - 900 nm, a Tungsten halogen
lamp) and ultraviolet (200 324 nm, a deuterium lamp).
Por equilibria the absorbances were measured in Helma stationary cells, type
110 QSO, with a pathlength of 10 mm. Por most of the kinetic experiments 10 mm
Helma flow-thru cells, type 114 QSO, were used, but for several batch runs it
was necessary to utilize 3 mm cells (same type).
In several cases, for some equilibrium measurements at rather high low
concentrations, another type of single-beam spectrophotometer, Zeiss PMQ 11,
was used, tagether with a 50 mm cell. A very limited amount of equilibrium
data was analyzed by an HPLC.
All stationary and flow-thru cells used in this research were made of quartz
glass.

4.2. EXPERIMENT AL AND ANALYTICAL METHODS

4.2.1. PREPARATION OF STANDARD SOLUTIONS

A separate procedure was devised for preparing the standard solutions used in
this research. All other solutions, destined for use in experiments, were
prepared by careful dilution of a standard solution. The weighing of all
necessary amounts was done on a Sartonus Analytic A 200S balance.
Isopropylphenol in water and p-nitroaniline in water standard solutions were
prepared by weighing a speeifie amount of IPP and PNA in a small weighing
bottle. After weighing water was added to the bottle(s) and the mixture(s) was
transferred to a lL volumetrie flask. Usually, it was necessary to use
additionally about 300 mL more solvent in order to fully rinse the weighing
bottle. After that enough water was added to bring the volume of the
volumetrie flask to lL. The flask had to he shaken vigorously and then stirred
for about two days to help the dissalution of the IPP (PNA) crystals.
In the case of nitrobenzene basically the same procedure was followed, except
that only few hours were necessary to obtain a homogeneaus solution.
Equilibrium results and methods 67

4.2.2. CARBON PRETREATMENT AND PREPARATION

As noted, the activated carbon used in this study was supplied by Norit BV.
Initially, it contained a lot of inorganic salts that interfere with the
analytica! measurements, because when dissolved in water they absorb light in
the UV region. Therefore, prior to use, the activated carbon was wasbed with
distilled water outside a distillation column over a 5 to 7 weeks period.
After this washing the carbon was dried at 105°C and was stored in special
closed metal tins. The carbon treated in this way was used to prepare samples
both for the equilibrium and kinetic experiments.
In order to decrease the time necessary to reach equilibrium the original
pelleted activated carbon (PAC) was ground to powder form. This results in
shorter diffusion paths, and thus, less time was needed to reach equilibrium.
The crushed carbon was then sieved and the smallest sieve fraction ( d < 63 J.lm)
was used for the equilibrium experiments.
However, as this last fraction contains also very small particles that could
easily interfere with the analytica! measurements, the carbon sample was
wasbed three times with boiling water and decanted after each washing step,
allowing sufficient time for the larger particles to settle. In some cases the
smallest fines had to be removed directly from the water surface, where they
formed a visible layer. After this procedure, the carbon was dried for 24
hours in an oven at 105°C and was then placed in stoppered glass bottles,
ready for equilibrium experiments.

4.2.3. PROCEDURE FOR ABSORBANCE MEASUREMENTS

Measurements of the individual absorption spectra of IPP, PNA and NBZ


indicated that isopropylphenol has maxima at 219 nm and at 275 nm, PNA bas
also two maxima, at 380 and 225 nm and a minimum at 275 nm. Nitrobenzene bas a
maximum at 267 and a minimum at 226 nm. These wavelength extrema give the
possibility of measuring accurately not only the absorbances in the
single-solute systems, but also those in the two-solute ones - IPP/PNA and
NBZ/PNA.
The actual measurement of the absorbance of a solution had to be preceded by
several important steps. First, if the spectrophotometer was off, after
putting it on at least 3 hours had to pass before achieving stable absorbance
Chapter IV 68

values. Then the stationary cells (sample and reference) had to be carefully
rinsed in and outside, preferably with double-distilled water. After
calibrating the spectrophotometer with the reference cell full of water and
after filling the sample cell with solution, the actual measurement could
start. However, before putting the sample cell in the cell bolder, the
external walls of the cell were carefully wiped with a piece of tissue paper
to remove any water or dirt left on them.
Finally, the absorbanee of the sample was read and record ed. In view of
achieving higher accuracy, several consecutive readings of the same salution
were made and the average value was calculated.

4.2.4. PROCEDURE FOR OBTAINING EQUILIBRIUM DATA

The beginning of the actual equilibrium experiment was normally preceded by


the steps described in Sections 4.2.1 and 4.2.2. However, the washing of the
activated carbon from inorganic matcrials with distilled water was performed
only twice, invalving very big amounts of carbon that proved to he sufficient
for all experiments.
The degree of accuracy with which the procedure for obtaining equilibrium data
is performed can affect significantly the obtained overall results, not only
for equilibrium, but also for the kinetic models. Therefore, a more detailed
description of some of the steps involved will be presented here.

The experimental metbod for obtaining equilibrium data used in this research
was the standard static bottie procedure in which each isotherm point is
developed by using different quantities of carbon.
Firstly, the activated carbon samples were prepared and put in 250 mL
Erlenmeyer flasks.
Secondly, a salution with the desired concentration was prepared, by diluting
an already available standard solution. A separate 200-mL aliquot of the same
salution was transferred to an isotherm flask containing no carbon. This flask
served as a blank to check for possible adsorbate evaporation or adsorption
onto the glass walls during the equilibration period.
Further on, all isotherm flasks were stoppered with caps, placed in the
Lab-Line Environ-Shaker and agitated at 220 rpm until equilibrium was
attained.
Equilibrium results and methods 69

Following equilibration, samples for adsorbate analysis were prepared by


filtration to separate the liquid and solid phases. This is a very important
step, since the presence of carbon particles in the sample can strongly
influence the absorbance and lead to completely invalidated results. For this
reason two different methods had to be used in this study.
Initially, samples were taken from the equilibrium flasks with a glass
syringe, taking out both solution and carbon. Then the carbon particles were
separated from the solution by replacing the syringe needle by a teflon filter
holder (Gelman, 13 mm plastic filter holder) containing a glass fiber filter
(Gelman, 13 mm, type NE, with a pore size distribution between 0.2 and 10 11m)
and by pressing the solution through the filter into a stationary quartz cell.
After measuring, the equilibrium flask was placed back in the shaker and left
there for one more week. After this week, another measurement was performed,
to check if equilibrium had really been reached.
According to the second method, which was employed later, the separation was
performed by using a membrane filter (Schleicher and Schüll, diameter 47 mm,
pore diameter 0.45 11m). The content of the Erlenmeyer flask was filtered as a
whole under suction through a glass filter bolder, containing the membrane
filter.
There were several reasons for the better performance of the second method.
One important reason was that when using the first procedure glass filter
fibers were found in the filtered sample, influencing the measured absorbance
values. Another drawback of the first metbod was that each time a sample was
withdrawn from the flask the volume and concentrations of carbon and solute(s)
changed and this was is not easy to account for.
Next, before filling the cell with the sample to be measured, it was always
flusbed three times with water and three times with the solution.
In order to calculate concentrations of sample solutions from the measured
absorbances, it was necessary to have a calibration curve for each solute and
at each specific wavelength. The calibration curve provides the mathematica!
relation between the absorbance and the concentration of a solution.

Finally, the mass of sol u te adsorbed on the activated carbon may be calculated
by using a mass balance on the adsorbate. This can be represented in the
following convenient form:
Chapter IV 70

) V
Q (4.2)
M

where M is the absolute amount of carbon used (kg), C and C - the initia]
0 p
and final concentrations (kg.m- 3) and V is the volume of the solution (m 3).

4.2.5. REMARKS ON EXPERIMENTAL AND INSTRUMENTAL ERRORS

In principle, there are several types of errors that can be made while
preparing and performing equilibrium experiments. They all have to be taken
into account when estimating how accurate and trustworthy the final results
are. However, experience showed that obtaining one final value as a sum of all
errors is not a realistic goal and it is necessary at each step of the
experimental procedure to make an independent error calculation.

The first type of error that will be discussed bere is the error in the
concentration measurements and the calculated carbon loading. Using the
propagation law, the relative standard deviations of the above parameters were
calculated for every equilibrium experiment performed in this study . Things
are complicated when dealing with the two-component systems, since in certain
cases the absorbance of one component has to be subtracted from the absorbance
of the second one. A detailed description of the formulas involved is
presented in a report by Voncken [1992]. Note that for low concentrations the
relative error in the concentration ts quite large, while for high
concentrations the error in the calculated carbon loading can be significant.
In order to increase the accuracy, three different samples were taken from
each equilibrium flask and the average absorbance value was used for
calculations.

In Table 4.3, an example of a set of results is given, containing data for the
single-solute equilibrium experiments for NBZ, together with the calculated
relative standard deviations (RSD) in the initia] concentration (SC iC 0
),

final
.
concentration (SCp ICp ) and carbon loading (SQ/Q).
Equilibrium results and methods 71

Table 4.3. Adsorption isotherm data for NBZ in water on activated carbon

3 24.8 25.74 0.28


4 42.6 51.46 0.14
5 39.1 51.46 0.14
6 37.5 51.46 0.14 5.118 1.6 0.2471 1.03
7 36.2 51.46 0.14 5.561 1.5 0.2536 1.03
8 103.4 153.2 0.054 9.124 0.9 0.2789 1.01
9 93.1 153.2 0.054 13.84 0.64 0.2996 1.01
10 77.9 153.2 0.054 24.09 0.3317 1.01
11 57.4 153.2 0.054 47.07 0.3701 1.03

Table 4.3 shows that the RSD's in the concentration are in the range 0.05
6.7% , while the ones in the carbon toading are between 1.01 and 1.05%. The
RSD range for the concentration is representative for all 3 components,
however, the RSD in carbon toading for some IPP experiments was higher than
2%. This is due to the fact that IPP was the frrst investigated component
while later, for PNA and NBZ, the experimental procedure was improved.

From Table 4.3 it can be also seen that tow concentrations have a high RSD,
while the relative standard deviation in the carbon loading becomes important
at high concentrations. The latter is strongly influenced by the high RSD in
the amount of carbon added. This is due to the fact that, in order to obtain
equilibrium points in the high concentration region, smalt amounts of carbon
were used, which results in high relative errors in the carbon dose.
The reason why the RSD in the concentration does not practically influence the
RSD of the loading is that in most experiments performed in this study, the
difference between starting and final concentrations was relatively big. The
exact way in which this influences the results can be found in the report by
Voncken.
Chapter IV 72

The foregoing observations mean that in order to achieve better experimental


accuracy it is necessary to improve, above all, the procedure of carbon dose
weighing. If exposed to air before and during the weighing, the carbon adsorbs
water vapor and increases its actual weight. Measurements showed that 60
minutes of exposure changed the weight of the sample by 4%. Therefore,
precantions were taken to reduce as much as possible the time that the samples
had to stay in open air.

Arwther type of error that may strongly influence the final equilibrium
results 1s the error made when filtering the already equilibrated
solution-carbon mixture, i.e. the step before measuring the actual absorbance.
lt was found out that when using the second method (see Section 4.2.4)
significant adsorption may occur on the filter, changing thus the actual
residual concentration in the sample. To solve this problem a different type
of filter had to be used which practically did not adsorb any of the
components.

All of the foregoing errors accuruulate during the whole equilibrium procedure
and affect the final results. None of these errors can be neglected if the
goal is to obtain accurate and reprodoeibie equilibrium data.

4.3. CORRELATION OF ADSORPTION EQUILIBRIUM DATA

4.3.1. SINGLE-SOLUTE SYSTEMS

Radke-Prausnitz isotherm equation


The single-component equilibrium tests in this study were performed on three
solute-adsorbent combinations. As already noted, three different organic
compounds were used, IPP, PNA and NBZ, and only one type of activated carbon,
Norit RBWl. All experiments were carried out at 20 ± 0.5°C.
The experimental isotherms for IPP, PNA and NBZ are plotted in Figures 4.1 and
4.2, while the exact values for IPP and PNA (for NBZ see Table 4.3) are given
in Appendix 5.1. The solid lines in the figures represent the best fit to the
experimental points using the Radke-Prausnitz equation (cf. Eq. (2.6)). The
parameters K, N and F were estimated by using the non-linear regression
r r r
procedure of the Statistica! Arialysis Subroutine (SAS) program package.
Equilibrium results and methods 73

0.5

0.4
0
0
d 0.3
z
ë5
4:
0 0.2
..J
z
0
(I) 0.1
a:
4:
ü
0.0
0 20 40 60 80 100 120 140

EOUILIBRILM CONCENTRA TION. Cp (mg/1]

D IPPe + PNAe IPPm - - PNAm

Fig. 4.1. RP isotherm and experimental equilibrium data for PNA and /PP

0.5
0\
E
Öl
.5 0.4
0
d 0.3
z
ë5
4:
0 0.2
..J
z
0
(I) 0.1
~
ü
0.0
0

EOUILIBRILM CONCENTRATION. Cp (mg/1]

+ NBZe - - NBZm

Fig. 4.2. Radke-Prausnitz isotherm and experimental equilibrium data for NBZ
Chapter IV 74

The fitting criteria used was the minimization of the value of S, defined as:

1
s =n L
n

k=l
[
(4.3)

where Qexp and Qcalc are the experimental and calculated carbon loadings
respectively, and n is the number of data points.
The optimal values for all three components are listed in Table 4.4 while
Table 4.5 contains the statistics of the SAS fit for component IPP (as an
example). The parameters were calculated, using [kg.m-3] and [kg.kg- 1] for the
bulk phase concentration and carbon loading, respectively.

Table 4.4. Fitting parameters for Radke-Prausnitz isotherm

PNA 0.456 0.091 754


NBZ 0.631 0.171 1042

Table 4.5. Statistica! data for SAS fit for component IPP
Parameter Estimate Asymptotic Asymptotic 95o/o
Std. Error Confidence Interval
Low.er. Upper

F 0.33415 0.00638 0.31642 0.35187


N 0.05656 0.00599 0.03992 0.07320
K 580.94370 41.63909 465.33661 696.55079

Asymptotîc Correlation Matrix


Corr F N K
F 1 0.9756 0.6510
N 0.9756 1 0.7140
K 0.6510 0.7104 1
Equilibrium results and methods 75

In order to get a better impression of the difference between the calculated


and experimentally determined equilibrium values, this difference can be
represented in relative form. This is shown in Fig. 4.3 for NBZ, where the RSD
in the experimentally determined carbon loading is also given.

3
N
m
z 2 <> <>
d
g fl
<>
<>
fi
~ 0
<> <>
a. <>
x
Q) -1
0 <>
~
x -2
Q)
0 0
I
a. -3
Q 0 20 40 60

Cp . NBZ [mg/l]

-a- (SQ/Q) -+- (SQ/Q) o (Qrp-Qexp)


error est. error est. /Qexp

Figure 4.3. Relative difference between the Radke-Prausnitz isotherm fit and
the experimentally determined data ( <>) and the RSD of the
experimentally determined carbon toading (-&.and -+-i (see Tab. 4.3)

Figure 4.3 shows that for concentrations above 10 mg/1 the Radke-Prausnitz
isotherm describes well this adsorbent-adsorbate system (within the limits of
the experimentally determined error), however, below that value some
deviations are observed. In addition to that, the RSD in the concentration,
which is the largest at low concentrations, is not accounted for in Fig. 4.3.
It should be also noted that in the very low concentrations region a small
error in the concentration measurement would strongly influence the calculated
value of the carbon loading.
Despite these deviations, the Radke-Prausnitz isotherm gives, as a whole, a
reasonable description of the single-solute equilibrium da~a of all three
components used.
Chapter IV 76

In Fig. 4.4 the Radke-Prausnitz isotherm fits for all three solutes are given
and, for easier comparison, the equilibrium concentration on the x-axis is
expressed as the ratio of the actual over the solubility concentrations of the
respective solute.

0.30

0
E
......
Ol
.s 0.20

0.10

0.00 l . . . - - - - - - - - 1 . - - - - - - - - ' - - - - - - - - - - '


0.00 0.01 0.02 0.03

Cp I Csol

Figure 4.4. Radke-Prausnitz isotherms for IPP, PNA and NBZ.

Figure 4.4 indicates that practically over the whole concentration range
(measured in this study) the adsorption capacity of the carbon for NBZ is the
highest one. It appears also that at low concentrations there is a substantial
difference between the capacity of the adsorbent for IPP and PNA.

Some additional remarks about the NBZ isotherm parameters


The equilibrium isotherm parameters for NBZ, presented in Table 4.4, correlate
well with the experimental equilibrium data. However, when these parameters
were used for kinetic model calculations, serious deviations occurred between
the experimental and model kinetic curves (examples are given in chapter 6).
These deviations were especially pronounced in the later part of the curves,
where equilibrium had been practically reached.
Because of these deviations the NBZ kinetic experiments in which equilibrium
Equilibrium results and methods 77

had been reached were used to determine additional equilibrium concentration


and carbon loading points. The newly obtained data points (seven) were
correlated independently from the previous data, using again the RP isotherm.
The results, presented in Fig. 4.5, show a clear difference between the
"equilibrium" equilibrium and "kinetic" equilibrium data points. According to
Fig. 4.5 the granular activated carbon has a lower equilibrium capacity than
the powdered one (they differ more than 8%).

0.50

0
0.40
--- ---- ---
----- ----
~
ëi
<(
0_J
0.30

0.20 I
/
/
;'
---
I
I
z
0
en 0.10
a:
<(
u
0.00
0 20 40 60

EOUIL. CONCENTRATION. Cp [mg/1]

"kinetic" •equil." o kinetic


isotherm isotherm eq. pointe

Figure 4.5. Equilibrium data points determined from equilibrium and


kinetic experiments, logether with RP isotherm fits, NBZ

Several possible explanations for this occurrence may be considered. Firstly,


it could be due to differences in the intemal surface area and pore volume of
the granular (used for kinetics) and powdered (used for equilibrium) activated
carbon samples. Secondly, the difference could be caused by the sieving of the
crushed carbon particles, which means that in some cases samples with
non-representative fractions were used for equilibrium. A third possibility is
that real equilibrium bas not been reached in the kinetic experiments.
Finally, a more simple explanation is that some systematic errors may have
occurred either during the equilibrium or the kinetic experiments.
Chapter IV 78

All four possibilities were investigated and details about the techniques and
methods are available in a report by Ligthart [1992]. The obtained results did
not confirm the first two hypothesis, which leaves open the third and fourth
possibilities, i.e. real equilibrium had not been reached and/or experimental
errors occurred. Although all experimental procedures were rechecked and no
obvious experimental error was found, part of the observed deviations might
have been due to inaccuracies in zero point corrections during and after the
kinetic experiments. More .details can he found in the above mentioned report.
lt should be noted here that there was no "kinetic" equilibrium data for the
other two solutes, IPP and PNA, because no kinetic curves were available in
which equilibrium had been reached. This makes it difficult to estimate
whether the same differences would occur as in the case of NBZ. However, model
calculations showed that the isotherm parameters have the biggest influence in
the later part of the kinetic curve, close to equilibrium. There are no such
experimental curves available for IPP (very slowly diffusing) and very few for
PNA (slower than NBZ), which means that the influence of potential
inaccuracies in the isotherm parameters on the model calculations for these
solutes will be less than for NBZ.

4.3.2. TWO-SOLUTE SYSTEMS

The experimental procedure described for single-solute adsorption was also


used in developing equilibrium data for the two-solute systems. The results
for nitrobenzene/para-nitroaniline (NBZ/PNA) and isopropylphenol/para-nitro-
aniline (IPP/PNA) are given in graphical form in Figures 4.6.a/b - 4.8.a/b,
respectively, and in tabulated form in Appendix 4.2.

The estimated RSD's for the two-solute systems are listed in the above
mentioned tables in Appendix 4.2. The values show that for most experiments
the RSD in the carbon loading is around 1%, while the average RSD in the
concentrations is different for the different components. In order to decrease
these deviations and improve experimental accuracy and reproducibility, the
same precautions and measures have to be taken as for the single-solute
systems and it is not necessary to repeat them bere.
Equilibrium results and methods 79

N 0.30
co
z 0 + + + +
0
0
0
ei 0.25 0
0
"' "' "'
E
......
0
0
"' "' "' "' "'
g 0.20 "'
0 di>
0
"'
20 0.15
"'<> 0
0 <> 0

<(
0 0.10
...J
z
0 0.05
co
([
<(
u 0.00
0 20 40 60 80 100

EOUIL. CONCENTRAT ION, Cp [mg/1], I'BZ


0 4:1 + 3:1 <> 1:1 6 2:1

Figure 4.6.a. Equilibrium results for NBZ in the system NBZIPNA with given
ratio of [~BZ : ~NA.

~ 0.30

i~ 0.25

0.20
0

~ 0.15 <> "'At:, "'


0 ~ "' "'

i
6
...J

u
0.10

0.05

0.00
r
~ r:P

~-------L------~~------~------~------~
0 20 40 60 80 100

EOUIL. CONCENTRA TION. Cp [mg/1], PNA

0 4:1 + 3:1 <> 1:1 6 2:1

Figure 4.6.b. Equilibrium results for PNA in the system NBZ!PNA with given
• NBZ ".PNA
ratw of C : L .
0 0
Chapter IV 80

0. 0.30
Q, 0
0 0
·ei 0.25
0
+ +
0

.....E ++ + ++ +
~ 0 0 0
0.20 0 0
a
0 0.15
z
ö<(
0 0.10
....J

0
z
0.05
f
CD
er<(
0 0.00
0 20 40 60 80 100

EOUIL. CONCENTRA TION. Cp [rng/1]. IPP


0 2:1 + 4:1 0 8:1 à 1:4

Figure 4.7.a. Equilibrium results for IPP in the system IPP/PNA with given
. ëNA
ratw of C IPP : .
0 0

~
0.
0.30
t>.
t>.

t>.

~
t>.a
0.25
.....
OI
t>.t>.

.§.
0.20 fö
a ~

~ 0.15
o o
0
ö<( 00

0 0.10 ~ +i)
'lt+
....J 0
fo
öCD 0.05 f
er<(
0 0.00
0 20 40 60 80 100

EOUIL. CONCENTRA TION. Cp [mg/1), PNA

0 2:1 + 4:1 0 8:1 à 1:4

Figure 4.7.b. Equilibrium results for PNA in the system IPP!PNA with given
. IPP
ratw of C : c.roJ'NA
0
.
0
Equilibrium results and methods 81

As noted in the previous section, by using the Radke-Prausnitz isotherm very


good correlations of the single-solute experimental data were obtained.
Therefore, initially, the multisolute form of this isotherm was applied to the
two-solute systems. Equation (2.11) was used for this purpose, but it was
found inadequate to describe the data for any of the above systems.

The next step was to apply the improved form of the multisolute
Radke-Prausnitz (RP) isotherm (cf. Eq. (2.13)) to the experimental data. As
mentioned, m contrast to Eq. (2.12), Eq. (2.13) reduces to the
single-component form when all but one concentrations are zero.
Despite the fact that this last equation, (2.13 ), is physically more correct
and logical, the results obtained for the correlation of both two-component
systems were still not satisfactory.

In view of the foregoing problems a third empirica! multisolute model, based


on the Fritz and Schlünder (FS) isotherm, was applied to the experimental data
(cf. Eqs. (2.17a,b)). An extension to the classical FS model was added in this
equation which accounts for concentrations close to zero and all single-solute
parameters are from the RP isotherm.
Table 4.5 contains the values for the optimization parameter S (cf. Eq. (4.3))
which were calculated in SAS by fitting isotherm equations (2.13) and
(2.17a,b) to the experimental data.

Table 4.5. Values of S, obtained by correlating the results for the


two-component systems, by using the improved RP and the extended
FS isotherm equations

1? i .•..• 1 • NBZJPNA system •··. > · îPPl'fN~$Y~t~.ro i~


/> ...•..... •· >·.·.•· S(nbz)"1E+3. S(pna)*11:t~ §(ipp)~.{~3< §(poa)"lg.fa

Eq. 2.13 11.0 43.0 47.0 18.0


Eq. 2.17alb 1.86 0.45 5.23 1.43
Chapter IV 82

The results shown in Table 4.5 indicate that the extended FS equation
correlates the experimental results in a better way than the improved RP
isotherm. This may partially be explained by the large number of
muitkomponent correlation parameters (3 per component) present in the
extended FS, in contrast to 1 parameter per component for the improved RP
isotherm.
Note also that there are about 60 equilibrium data points for each two-solute
system that have been successfully fitted with the extended FS equation which
gives a strong indication that these isotherm equations can be used in kinetic
models calculations.
The correlation parameters obtained with Eq. 2.17a/b are listed in Table 4.6.

Table 4.6. Fitting parameters for e.x:tended FS isotherm for IPP!PNA and
NBZ!PNA systems

System Comp, Krp Frp . · · . Nrp A·..... ij .•. •. . . tt


m~S/kg m~SNJkg"N [~) .· [-])
m"SD/kQ"'D bi

NBZ/PNA NBZ 1042 0.631 0.17 1.067 0.863 0.89


NBZ/PNA PNA 754 0.456 0.09 0.888 0.991 1.11
IPP/PNA IPP 581 0.334 0.06 0.645 5.836 1.08
IPP/PNA PNA 754 0.4561 0.09 0.632 0.448 0.59

The deviation of the predicted from the experimental equilibrium data is shown
in Figures 4.8.a/b 4.9.a/b on p.83. In these figures the ordinale represents
two terms: the estîmated error (on basis of error analysis) in the
experimentally determined carbon loadings and the relative difference between
the calculated (FS isotherm) and experimental loadings. The abscissa is the
equilibrium concentration of the particular solute in the mixture. It is clear
that the two-component fits show larger deviations than the single-component
ones.

Considerable insight into the effect of adding a second component to a


single-solute adsorbate-adsorbent system can be obtained by plotting on one
graph the isotherms of one component, both in the single and in the binary
systems. These graphs are given in Figures 4.9 - 4.10 on p.84. The single-
Equilibrium results and methods 83
N
(!)
z 15
.......
'$.
10 +
0....._ +
0
(/) 5
_,..+ +
+
d. + ...±
+
f/)
n....O ...±
~ 0
·""" '!' + +
a. +
x + +
+ +
(j) -5 + + +
0....._ + +
a.x
(j)
-10
0

-
I
f/)
-15
Q 0 20 40 60 80 100

Cp [mo/11 NBZ
-e- (SQ/Q) + (Qfs-Qexp) ...,._ (SQ/Q)
error est. /Qexp error est.
Figure 4.8.a. Relative errors in the experimental loadings (solid lines) and
the calculated, with the extended FS isotherm, loadings (+). for
NBZ in NBZ/PNA system.

z~
a. 15
'$.
10
0....._
0
(/) 5
+ +
d.
f/)
;1;.

~ 0
+ +
r+-+
~ + +
+
(j) -5
0....._
a.x -10
(j)
0

-
I
f/)
-15
Q 0 20 40 60 80 100

Cp [mg/1). PNA

-e- (SQ/Q) + (Qfs-Qexp) ~ (SQ/Q)


error est. /Qexp /Qexp

Figure 4.8.b. Relative errors in the experimental loadings (solid lines) and in
the calculated, with the extended FS isotherm, loadings (+). for
PNA in NBZIPNA system.
Chapter IV 84

<(
0.4
~
PNA
1..... 0.3
g
0

~ 0.2
ëi
<(
0_J
N3Z:PNA=4: 1
0.1
öffi
rr IPP: PNA
<(
u
0.0
0 10 20 30

EOUIL. CONCENTRA TION. Cp [mg/1]. PNA

Figure 4.9. Adsorption isotherms for PNA in the systems PNA (single-
component), NBZ/PNA and IPP!PNA. In the two binary systems
there is a constant ratio of 1:4 between the equilibrium
(final) concentrations CPNA : c~BZ and ëNA : CIPP
p p p p

0.4 . - - - - - - - - - - - - - - - - - - -
N
~ NBZ
.......

1 0.3
IPP

'iëi 0.2
<(
0
_J

ö 0.1 IPP : PNA 4 : 1

~
u N3Z : PNA 4 : 1
0.0 or:__ _ _ _ _ __j.___ _ _ _ ___._ _ _ _ _ ___,
0 10 20 30

EOUIL. CONCENTRA TION. Cp [mgJIL NBZ

Figure 4.10. Adsorption isotherr.ns for NBZ in the systems NBZ (single-
component) and NBZ/PNA, andfor 1PP in the systems IPP (single-
component) and IPP/PNA. In the two binary systems there is a
constant ratio of 1:4 between the equilibrium (final)
concentrations CNA : C~BZ and CNA : ciPP
p p p p
Equilibrium results and methods 85

solute isotherms are calculated with the Radke-Prausnitz equation, while the
two-component ones by applying the extended Fritz and Schlünder equation.

Analyzing Fig. 4.9 reveals that the adsorption of PNA is strongly influenced
by the presence of a second component, in this case NBZ or IPP. Both NBZ and
IPP strongly compete with PNA and reduce its uptake, but obviously when IPP is
present the interference is stronger.
In Fig. 4.10 a similar trend is observed when adding a second component (PNA)
to a single-solute system, containing NBZ or IPP. Moreover; the presence of
PNA reduces the carbon capacity for NBZ more than the capacity for IPP. This
is in accordance with the interactions noted in Fig. 4.9.

In principle, the behavior of each solute in a multicomponent adsorption


system is governed by the physical and chemical properties of the adsorbates
and the adsorbent. These properties have an effect on the adsorbate-adsorbent
interactions and may influence the actual state of thermadynamie equilibrium.
The main objective in this research, with respect to adsorption equilibria, is
to find a suitable semi-empirical isotherm equation, which can be correlated
successfully to the experimental data and consequently, provide a suitable
mathematica! formulation for the local equilibrium condition in the kinetic
models.
Nevertheless, it may be interesting to note that the actual adsorption of two
organic species from water solution on activaled carbon may depend on the
accessibility of certain pores to only one of the adsorbates. This may partly
explain the higher carbon capacity for NBZ (cf. Fig. 4.4), which has the
smallest molecule. In addition, according to the data presented in Appendix
4.2 the total loadings in both two-component systems are higher than the
individual loadings in the single-solute systems. This indicates that probably
more surface is occupied when two different components adsorb simultaneously.

Finally in this section a non-traditional way of presenting the two-component


equilibrium isotherms is proposed. The common way of plotting these isotherms
in literature is drawing two-dimensional plots, as shown in Figs. 4.9 - 4.10.
However, as in reality the loading of each solute is a function of the
liquid-phase concentrations of both solutes, a better view can be achieved by
making three-dimensional surface plots. These plots, made for the
Chapter IV

0.30

0.25

0.20

0.15

0.10
100
90
80
70
60
20 50 c
40
JIJ IPP fmg/1 j
20
8 10
4 2 0
C PNA fmg/l) 0

Figure 4.11.a. 3-D plot of the carbon loading of IPP as a f•nction of both
solute concentrations in the system IPP!PNA, calculated with
the extended FS isotherm.

Q PNA [mg/mg)

o.Jo
0.25

0.20

o.Js
0.10

C PNA [mgll)

flgure 4.1l.b. 3-D plot of the carbon loading of PNA as a function of both
solute concentrations in the system IPP!PNA, calculated with
the extended FS isotherm.
Equilibrium results and methods 87

Q NBZ [mg/mg]

0.30

0.25

0.20

0.15
100
0.10 90
80
70
60
50 c
40
3CI NBZ [mg!l]
50 20
10
15 10 0
5 0
C PNA [mg/l]

Figure 4.12.a. 3-D plot of the carbon loading of NBZ as a function of both
solute concentrations in the system NBZIPNA, calculated with
the extended FS isotherm.

Q PNA

0.30

0.25

0.20

0.15

0.10

0 C NBZ [mg/1)

Figure 4.12.b. 3-D plot of the carbon toading of PNA as a function of both
solute concentratîons in the system NBZ!PNA, calculated with
the extended FS isotherm.
Chapter IV 88

two-component systems in this study, together with the experimental results,


are shown in Figs. 4.ll.a/b - 4.12.a/b. The surface area is calculated with
the extended Fritz & Schlünder isotherm. In the "b" figures the PNA loading is
depicted and in the "a" figures the loading of the other solutes. The vertical
lines on the graphs conneet the experimentally determined points ( situated on
the top end of the line) and the cl/c2 plane. In Fig. 4.1l.a, for example, for
PNA concentration equal to zero, the line of the single-solute IPP isotherm is
obtained in the plane QIPP/ cfPP. A plot of total loading for NBZ/PNA,
calculated with the extended F&S isotherm, is given in Fig. 4.13.

Q tot al [mg/mg]

0.45
0.40
0.35
0.30
0.25
0.20
0.15

5045
40
3530
2520
C PNA [mg/1]
1
\o

Figure 4.13. 3-D plot of the total carbon toading as a function of both
solute concentrations in the system NBZ/PNA, calculated with
the extended FS isotherm.

4.4. SOME NOTES ON IRREVERSIBILITY OF ADSORPTION

In the present study a limited amount of experiments were performed to check


for possible irreversible effects in the adsorption of IPP. The experiments
included adsorption and desorption of isopropylphenol in water solution on
Norit RBWl activated carbon.
Equilibrium results and methods 89

of IPP from water solution at 25°C


First, the adsorption isotherm of IPP in water solution was determined at
25°C. A similar procedure to the one described in Section 4.2.4 was used,
although problems occurred when taking samples for measurements. In some cases
it was oot possible to retain all carbon fines on the filter and also filter
glass fibers were detected in the already filtered samples.
In order to calculate the carbon loading the mass balance equation (Eq. 4.2)
was used. The results obtained are given in Fig. 4.14. The calculated loading
is in good agreement with the IPP equilibrium experiments previously described
and both sets of results were obtained independently of each other.

of IPP in water solution at 25°C


The loaded carbon samples were dried in open air at room temperature and then
their weight was measured. Further on, using the already obtained loading
values the pure carbon amount (without IPP) was calculated. A specific amount
of double-distilled water was added to several flasks containing the loaded
carbon samples. The erlenmeyers were placed in the shaking machine at 25°C.
After an average contact time of 10 days the concentrations in the bulk
solution were measured, using the already described techniques.

3.00
'öi
'V
g 2.50
s 0, n
n .ro c
a 2.00
+
+
.fO
~ 1.50
~
0
...J 1.00 J

3
aJ 0.50
J

~
()
0.00
0 20 40 60 80 100 120 140

EOUJL. CONCENTRA TION. Cp [mg/1]


0 ads + des - - RP-iaoth.

Figure 4.14. Adsorption and desarpfion of JPP from water salution


Chapter IV 90

In order to better visualize the difference between adsorption and desorption,


the experimental desorption points are also plotted in Fig. 4.14, together
with the Radke-Prausnitz isotherm fit.
Figure 4.14 shows that the largest difference between the adsorption and
desorption points is less than 8%. On basis of these experimental results it
may be concluded that at 25°C there is no significant irreversibility for the
analyzed system. However, these results have to be treated with caution, as
the measured desorption points cover only a limited concentration range.

4.5. CONCLUSIONS

Detailed information has been obtained about the single-solute adsorption


equilibria of IPP, PNA and NBZ from water salution on activated carbon at
20°C. Priority was given to performing accurate and reproducible experiments
and statistica! analysis was used to calculate the errors involved. On basis
of the estimated errors, special procedures were designed in order to minimize
the relative standard deviations in the concentration and carbon loading.

The equilibrium data of all three single solutes were successfully fitted with
the Radke-Prausnitz equation. This made it possible to use this equation in
all single-component kinetic models calculations.

Several multicomponent isotherm equations were applied to the equilibrium data


of the IPP/PNA and NBZ/PNA systems. The final outcome indicated that the
extended Fritz & Schlünder isotherm, proposed in this study, gave the best
correlation of the experimental data.

The total adsorption capacity of the activated carbon in the two-component


systems was higher than its capacity for the respective single components in
in the single-solute systems. This could mean that more pores were available
for adsorption in the bi-solute than in the single-solute systems.

A non-traditional way of presenting two-component equilibrium data has been


Equilibrium results and methods 91

proposed. The 3-D plots of the component carbon loadings as function of the
the two solutes concentrations give a better picture of the complex
adsorption isotherm than the common 2-D figures.

A limited amount of experiments were performed to check for possible


irreversible effects when adsorbing IPP on activated carbon at 25°C. Results
indicated that such effects are probably not significant at this temperatures
although, due to the limited number of equilibrium points, there was no
conclusive evidence available to this respect for the full concentration range
(0 100 mg/1).
Chapter V 92

EXPERIMENTAL AND NUMERICAL METHOOS FOR KINETICS

5.1. BATCH KINETIC EXPERIMENTS

The main goal of the batch kinetic experiments is to measure the rate of
adsorption. This is done by means of adding a eertaio amount of carbon to a
fixed volume of salution and following the concentration change of the
component(s) in the bulk salution in time. The dependency obtained is called
a concentration decay curve (COC). This concentration decay curve is compared
to the model calculations with the help of an optimization procedure and from
the best-fit model curve the values of the different types of diffusivities
are determined.
The shape of the COC for each adsorbent-adsorbate system depends on several
factors, some of which were chosen for varlation in this study. These were the
initia} concentration of the solution, amount of carbon used (carbon dose),
type of organic component and presence of a second component. All other
parameters, like partiele size, temperature, stirrer speed, shape and
dimensions of stirrer, type of vessel, etc. were kept constant. It should be
noted, however, that in a limited number of experiments also the partiele size
was varied.
All experiments were performed with one type of carbon, described already in
Chapter 4.

5.1.1. EXPERIMENT AL SETUP

The experimental setup used for performing kinetic experiments is shown in


Figure 5.1. In order to increase the amount of experiments 4 vessels, each
with its own circulation system, were installed, but all using the same
spectrophotometer, water bath, computer and printer. This permitted the
running of four independent experiments at the same time. The different parts
of one such instaBation will be described below.

Batch vessel
The batch vessels used were 3.5 liter glass vessels (except for vessel 1, with
a volume of 3 liters) with 4 built-in baffles. Each vessel was covered with a
Experimental and numerical methods for kinetics 93

lil

VI [

~
IÏI I IV
0
-
f r----

I
y V
I
VIl
1--

VIII

IX
I 0
[i.
/
·"'
Fig. 5.1. Scheme of experimental setup for batch kinetic experiments, con-
sisting of a batch vessel (/), water-bath (I/), stirrer (I/1),
filter (IV), pump (V), airtrap {VI), spectrophotometer (VII),
personaf computer (VIII) and printer (IX).

five-neek glass head, having an opening in the middle for a stirrer to pass,
one outlet, two inlets and a free neck. The conneetion between the vessel and
the head was done with a stainless-steel damp. A detailed picture of the
vessel is shown in Figure 5.2.

Water-bath
The batch vessel was placed in a thennostated water-bath and the temperature
was held constant at 20 ± 0.5°C. The temperature was controlled by a contact
thennometer, placed on a heater/cooler and monitored independently with a
thennocouple.

Stirrer
A teflon stirrer-blade was used with a diameter of 50 mm and height of 35 mm.
The material chosen was teflon because, firstly, it adsorbed neither IPP nor
Chapter V 94

... =====;t

Fig. 5.2. Batch vessel (I) tagether with stirrer (111) and filter (IV).

PNA or NBZ and, secondly, no breaking of carbon particles occurred. A


stainless steel stirrer rod and teflon stoppers, provided with hearings for
the stirrer were used. The stirrer speed was fixed at ± 467 rpm. The
stirrer-blade was placed low in the vessel (20 mm above the bottom) to prevent
the formation of air-bubbles and to promate complete partiele suspension.

Filter
At the inlet of the sucking line of the pump a sintered glass filter was
placed to prevent carbon particles from entering the other parts of the
installation. This was especially important for the flow-thru cells in the
spectrophotometer, where the presence of carbon fines could seriously alter
the measurements. The diameter of the filter was 25 mm, with pore diameters in
the range of 100-160 J.Ull. A new filter was placed befare each experiment as the
cleaning of used filters proved almast impossible.

Pump
The pump used was a magnetically-driven gear pump with teflon hearings and
teflon or nickel gears. The average capacity applied in the kinetic runs was
0.13 Vmin.

An airtrap was placed in the pressure line of the pump to prevent air-bubbles
from getting into the flow-thru cells. As in the case of carbon particles, the
Experimental and numerical metlwds for kinetics 95

presence of air-bubbles can influence the absorbance values.


In the airtrap the inlet stream was split in a stream free of air bubbles
going to the spectrophotometer and an air-bubbles-containing stream sent back
to the vessel.

Spectrophotometer
The spectrophotometer, DU-64, was the same one used for equilibrium
experiments and already described in chapter 4. Several additional options of
the apparatus will be noted bere, relevant mainly to the killetic measurements.
The spectrophotometer has a programming mode and four program storage areas. A
built-in RS-232 interface accessory makes possible communication between the
DU-64 and a PC. The spectrophotometer has also a program controlled seven
position cell bolder.
The cuvettes used were Helma Flow-thru Compact Cells, types 176.700-QS with
pathlengths of 10 and 3 mm.
Each batch vessel had its own 10 mm cell or, in a number of cases, its own
combination of 3 and 10 mm cells. In all cases the liquid content of each
system remained constant.

Personal computer
A personal computer was used for the on-line storage and for the processing of
data. A special communication program (DU) made it possible to exchange
programs between the spectrophotometer and the PC. With the help of a data
communications program (PROCOM) the output data from the spectrophotometer
was received as an ASCII file.

Printer
Both computer and spectrophotometer were connected to a printer for additional
output of data when necessary.

Piping
All piping used in the instanation was made of stainless steel. The inner
diameter of the sucking line of the pump was 3 mm, all other lines had an
intemal diameter of 2 mm. At several places, where a mobile conneetion was
Chapter V 96

necessary, polymer tubing was used. It was made of a non-adsorbing material


(viton). Aexible teflon lines connected the stainless steel pipelines to the
flow-thru cells of the spectrophotometer.

5.1.2. PROCEDURE FOR OBTAINING KINETIC ADSORPTION DATA

Preparation of activated carbon dose


As already noted, the original extruded activated carbon received from Norit
BV had to be wasbed carefully for removing all inorganic salts. The basic
pretreatment procedures were discussed in chapter 4, so they will not be
repeated here. In this section only the steps which are directly related to
the kinetic experiments will be described.

Firstly, the bulk portion of wasbed carbon had to be divided several times
into smaller and smaller portions by sample splitting. This was done to ensure
that all portions had the same partiele size distribution and the same
distribution of degree of activation of the carbon. After that the carbon
portions for kinetic use were stored in closed botties and the smallest
samples contained the amount of carbon needed for one experiment.

Secondly, just before the start of an experiment the pores of the activated
carbon dose had to be saturated with water in order to prevent any mass
transfer Iimitations occurring at the beginning of the adsorption process.
This was done by taking the already dried and weighed carbon sample and
placing it in boiling water for about 30 minutes.

Finally, after the boiling procedure and before placing the carbon dose in the
batch vessel, it was left to cool for a few minutes in a closed bottie (to
avoid adsorption of any kind of vapor).

Determination of Concentration
Concentration decay curves were obtained by following the change in the
concentration(s) of the component(s) in the adsorber in time. The actual
measurement procedure was highly automated, controlled by a OU-program in the
Experimental and numerical methods for kinetics 97

spectrophotometer. Several such programs were written for the needs of the
kinetic experiments in this study. The DU programs were made in such a way
that the absorbance at specified wavelengtbs was measured every 5 minutes for
the first 10 hours (the strongest concentration decrease is in this part of
the COC) and every 30 rninutes for the rest of the time (typically 70-80
hours). To achieve higher accuracy each recorded measurement value was
calculated as the average absorbance of 50 successive measurements for the
same point in time. As an additional precaution the standard deviation of each
measured value was also recorded.
It should be noted here that making one measurement was related actually to
determining the absorbance in two cells (the flow-thru cell and the reference
cell) at three, sametimes four, different wavelengths. Because of deviations
in the measured absorbance with time due to the stability of the
spectrophotometer (according to specifications the drift was within ± 0.003
Absorbance Units/hr) the absorbance of the reference cell was also measured
for each point in time. The absorbances of the other cells had to be corrected
for the absorbance deviations in the reference cuvette. In many cases an extra
stationary cell filled with salution was included in the programmed
measurements to check the stability of the spectrophotometer. During the whole
experiment the corrected absorbance of such a cell should remain constant
within the accuracy of the apparatus.

For the single-component IPP experiments the absorbances at 219 and 275 nm
were used while for PNA the specific wavelengtbs at which absorbances were
measured were 225 and 380 nm. For NBZ these wavelengtbs were 226 and
267 nm.
The absorbance forthetwo-component mixture IPP/PNA was measured at 275 nm to
determine the IPP concentration in the salution and at 380 nm to deterrnine the
PNA concentration. In the system NBZ/PNA the absorbance of NBZ was measured
at 226 and 267 nm and that of PNA at 380 nm.
For all experimental runs the absorbance at 500 nm was also measured and, as
the scans of all experimental solutes showed no absorbance at this wavelength,
this was used as a check for undesired processes occurring in the vessel and
for the presence of unwanted particles or air-bubbles in the cell.
Chapter V 98

After carrying out calibrations the actual measurement was started. The ready
carbon close was added to the solution in the batch vessel through a funnel
placed in the free neck of the head. This was clone after one measurement of
the pure solution was performed which made it possible to determine the exact
starting concentration.
After each experiment the whole installation was flusbed with water and the
flow-thru cells were checked for possible pollution. This was clone by filling
each cell with water and measuring the absorbances at the wavelengtbs
specified above. lf no pollulion occurred the obtained values had to be the
same as those measured before the start of the experiment, otherwise the
measured values had to be corrected accordingly.
Additional details about the experimental methods and techniques are available
in the reports of Vedder [1990] and Steenbergen [1991].

5.1.3. EXPERIMENT AL RESUL TS

A significant number of experiments were performed in this study both for the
single and two-component systems and many of these experiments were executed
twice in order to check the level of reproducibility. ln the presentation that
follows and in the Appendices only those results are given which had a
reasonable degree of accuracy and reproducibility (the concrete values will be
discussed later) and were considered trustworthy enough to compare with the
model calculations.
Complete tables of experimental conditions for all single- and multicomponent
systems are presented in Appendices 5.1 and 5.2, respectively, while the
detailed discussions and comparisons with the models are in chapters 6 and 7.
Below, some examples of ex perimental results will be presented in order to
better understand the numerical solution and sensitivity analysis later in
this chapter.

As already noted, the main objective of this study was to investigate the
influence of several system conditions on the rate of adsorption. The
effect of the following variables on the shape of the kinetic experimental
curve has been studied:
Experimental and numerical methods for kinetics 99

initia! concentration, C"


b
- amount of carbon, m
type of organic compound
- presence of a second component
A limited arnount of experiments were performed with a smaller partiele size
(R ).
pa

Other system conditions and variables were kept exactly the same for all rate
experiments. These were:
- type of carbon
the hydrodynamic conditions
- temperature

Some examples of experimental decay curves for single-solute systems


Figure 5.3 shows an example of the influence of the initia! concentration on
the concentration decay curve for the single-component IPP system, while

40

lz 30 <>
Carbon dose exp.
0 0
0
(mgll)
0
f= o.,
<(
o.,.,
a:
1-
z
w
0
20 +
......... 50.5
37

ö0 50.7
10 23
:::(
:l
(IJ 49.9
22
0
0 10 20 30 40 50 60

TIME (hours)

Figure 5.3. The influence of variations in the initial concentration (C:) on


the concentration decay curve of PNA.
Chapter V 100

60

::::::
OI 50 Co (m~/1) Carbon dose exp.
É
(mg/1)
bf= 40
<(
(( 175 52
I-
zw 30
0
0
ü
z 0
0
0
0 20 0

...........
00
ü 00
Oo 350
29
y:
_J
10 51.26
ITl 525 53

0
0 10 20 30 40 50 60

TIME (hours)

Figure 5.4. The influence of variations in the amount of carbon (m) on


the concentration decay curve of IPP.

30.-----------------------------------------~

26 0

Rpa (mm)
0
+

22 0
....
18 ·.. ....... · · · · · · · · · · -~ - -· · · · · · · · · · · · · · · · · · · · · · ·~-~- -
++ 000000:
0.26

. . . --=======~~l'm:--------!
•.... OooOoooOOooo.oo. 48
14
+++++++++++++t++++++++++++++++

0.16 47

10 L--------L--------~------~---------L------~

0 10 20 30 40 50

TIME {hours)

Figure 5.5. The influence of variations in the carbon partiele size on


the concentration decay curve of PNA.
Experimental and numerical methods for kinetics 101

Fig. 5.4 illustrates the effect of the amount of carbon on the CDC in the same
system.
An example of a set of experimental concentration decay curves for different
partiele sizes for the single-component PNA system are shown in Fig. 5.5.

Figure 5.6 represents an example of two pairs (one for NBZ and one for PNA) of
duplicate experiments, which were used for a reproducibility check. The points
on the residue graphs represent the difference between the bulk concentrations
of a pair of duplicate experiments divided by the absolute concentration of
one of them.

0.025 . - - - - - - - - - - - - - - - - - - - - - ,

0.015

NBZ
0.005

-0.005

-0.015

10 20 30 40 50 60 70

TltvE (hours)

Figure 5.6. Residue graphs for two pairs of duplicate experiments for
single-component PNA and NBZ. The two duplicate experiments
in each pair are designated as exp. 1 and exp. 2.

Some examples of experimental decay curves for multisolute systems


Figures 5.7a/b illustrate the influence of variations m the initial
concentration of component NBZ on its own concentration decay and on the CDC
of PNA in the system NBZ/PNA.
Chapter V 102

100 .-------------------------------------------~

Co (mg/1) exp.
80
(PNA)

60
ö ......
+
f=
<(
[(
1-
40 ...
+...+-++.+,
••+++++++++++++ 14.9 TC202
z
w
f'++++++t+-4-++++++
++++++++++++++++.........++++++++++++++++++-++++++++++++++

3u 20
14.8 TC203

0 ~--------L---------~--------L---------L-------~
0 10 20 30 40 50

TIME (hours)

Figure 5.7.a. The influence of variations in the initia[ concentration


of NBZ on its concentration decay curve in the system
NBZ!PNA.

20 r---------------------------------------------.
<(
z
n.
15 Co (mg/1) exp.

î
öf=
<(
[(
1-
z
w
3
u

0 10 20 30 40 50

TIME (hours)

Figure 5.7.b. The influence of variations in the initia[ concentration


of NBZ on the concentration decay curve of PNA in
the system NBZIPNA.
Experimental and numerical methods for kinetics 103

5.1.4. SOME REMARKS ON EXPERIMENTAL REPRODUCIBILITY

When performing kinetic experiments various types of errors can be introduced


into the system through handling, analysis and other procedures, which may
strongly influence the quality of the obtained results. Some of these errors
were already mentioned in chapter 4, while others, related to the experimental
concentration, volume and time, are discussed in some detail in Appendix 5.3.

Two main criteria are used in this study to estimate the quality of the fit
between model and experimental curves. This quality of the fit is described by
the Sum of Squares (SOS) between the two curves, as defined later in Section
5.2.3 (Bq. 5.6).
The first criterion is based on a statistica! analysis of the error in the
concentration measurements and is defined as the Sum of Squares (SOS) between
the curves which describe the upper and lower limit of the 95% confidence
interval:

2
SOS E [ cupper I Cob - C lower curve,
. I Co ]
b
I J. (5.1)
std curve,i 1

= 0

where j is the number of experimentally determined data points, CÜ is the


b
initial concentration. Note that all symbols "SOS" (used bere and later on)
denote actually the respective sum of squares divided by the number of points.

Calculations of SOS for several experiments showed a range of ( 1.0 -


std
4
4.5)·10- , which range was used for comparison with the SOS between model and
experimental curves. Details about the SOS calculations are available in a
std
report by Ligthart [1992].

The second criterion 1s based on the reproducibility of the kinetic


experiments which was repeatedly verified by performing duplicate experiments
for given sets of system conditions. The relative SOS 's, calculated between
two duplicate curves, are defined as:
Chapter V 104

SOS = (5.2)
rep

where:
: number of experimental points
: the experimentally determined concentration (dimensionless)
in the ith experimental point, exp. 1
: the experimentally determined concentradon of the same
: component (dimensionless) in the ith experimental point, exp.2

It should he noted bere that, as it is difficult to exactly reproduce the


initia! concentrations and carbon doses, the influence of the occurring
differences is also included in the calculated SOS .
rep

Calculations of SOS rep for several sets of duplicate experiments (single- and
two-component) showed a range of (0.06 - 7 .0} w·'t. ho wever, with respect to
the order of magnitude, most of the values obtained are closer to the upper
boundary of the range (cf. Table A.5.3.3), with an average value for SOS
rep
4
equal to 2·10· • This range includes the range of the SOS , which makes the
std
comparison with the Sums of Squares between model and experimental curves
easier.

5.2. NUMERICAL SOLUTION, PARAMETER FITTING AND


OPTIMIZATION

5.2.1. NUMERICAL SOLUTION AND ALGORITHMS

Single-component models
The basic equations of the different models were developed in chapter 3. Each
set of single-component model equations consists of a partial differential
equation and an ordinary integral equation, together with the respective
initial and boundary conditions and equilibrium isotherms. The partial
differential equation, which represents physically the shell mass balance over
the partiele is non-linear and second order with respect to the radius, r and
first order with respect to the time, t (cf. Eq. 3.7).
Experimental and numerical methods for kinetics 105

Due to tbe fact tbat tbe isotherm equation used (cf. Eq. (2.6)) is bigbly
non-linear the systems of single-component model equations could not be solved
analytically and it was necessary to use a numerical metbod for solving them.
The same was valid for tbe two-component model equations.

In literature, various numerical libraries can be found, containing ready


algoritbms and programs for solving differential equations. However, it proved
impossible to solve the sets of coupled differential equations used in tbis
study witb ready modules and, moreover, wben working witb sucb programs it is
often difficult to understand wbat is really happening during tbe
calculations.
Due to the foregoing reasons a well-known numerical metbod, the fmite
difference method, was cbosen and several algoritbms and programs were made
for solving the systems of equations. The basis of this metbod is tbe exchange
of the initial differential equations for their approximate valnes wbicb are
described by the valnes of tbe functions in separate, discrete points. As a
result of this change the set of differential equations is transformed into a
system of algebraic equations.
In tbe present study a semi-implicit type of fmite difference metbod was
used, applying an algorithm proposed by Berkovski [1976]. Some details
conceming tbe specific application of the numerical metbod to tbe systems of
equations are given in Appendix 5.4.

Appendix 5.5 contains tbe basic programming algoritbm used for the
single-component models witb variabie diffusivity. The programming language
used was VAX Pascal and the programs were run on a VAX VMS system.
Furtber details about tbe algorithms and Pascal programs for the
single-component models can be found in a report by Laimböck [1991].

Multicomponent models
The numerical metbod used for solving tbe sets of differential equations for
tbe two component models was the same one as mentioned above. However,
constructing the programing algorithm proved to be mucb more complicated tban
in tbe case of single-solute models, due to the interactions between
components, which introduces cross-diffusion coefficients and multicomponent
Chapter V 106

equilibrium isotherm equations. The developed algorithm permits calculating


concentration decay curves with or without cross-diffusion coefficients,
depending on the chosen model. In this study only the multicomponent
Maxwell-Stefan model without cross-diffusion coefficients was used and the
respective Pascal program is listed in Appendix 5.6.
Complete algorithms and programs for the two-component models are available in
a report by Donker [1992].

5.2.2. COMPARISON OF ANALYTICALAND NUMERICAL SOLUTIONS

Mass balances outside the


Before discussing the comparisons between analytica! and numerical solutions,
which are used to verify the correctness of the model calculations, it is
necessary to make a distinction between two different types of mass balances
outside the particle, used in this study.

The mass balance outside the partiele for a batch system can be derived in two
different ways: (A) as an overall mass balance of the adsorbate, (B) as a mass
balance of the adsorbate in the bulk solution. The first type of mass balance
was already presented in chapter 3 and is the one used in most model
calculations in this study:

R
m
cb 41t J pa
(€
pa
Cp + (1-€ )p
pa c
2
Q ) r dr (3.13)
0

In Equation (3.13) the overall mass balance is updated by integrating the C


p
and Q profiles along the radial coordinate of the partiele for the whole
partiele volume at each time leveL Because of the equilibrium assumed the Q
profile can be calculated from the C profile with the isotherm relation. In
p
this way the uptake of the component of one partiele is known from this
integral at each time leveL The total uptake per unit of batch volume is
equal to the uptake of one partiele multiplied by the number of particles per
unit of batch volume.
Experimental and numerical methods for kinetics 107

For the second type of mass balance it holds that the changes in the amount of
the adsorbate in the bulk per unit of time is equal to the amount of adsorbate
teaving the bulk solution:

ac
V _ b ;;; - m k A (C - C 1R ) (5.3)
at t s b P pa

with:
A
3 (5.4)
(1-E )p R
pa c pa

where:
2
A : specific outer surface of the adsorbent (m kg· \

Although the overall mass balance (Eq. 3.13) was used in most model
calculations, the other mass balance (Eq. 5.3) was also applied in some
cases and used in the comparisons with analytical solutions.

Comparisons with analytica! solutions


For special cases of the constant diffusion single-component models the
numerical solutions obtained were compared to analytical solutions reported in
literature. This was done to check the accuracy and reliability of the
programs used for solving the models.
In addition to that, the programs for solving the single-component models with
variabie D were compared to the ones with constant diffusivity, by setting
app
the variabie D to be constant.
app

The flrst analytica} solution used for comparison is based on the following
model simplifications:
L The concentration in the bulk solution, C , is constant (infmite
bath
bath).
2. The adsorption isotherm is a linear function: Q* = Kh ·C•

The analytica} solution for this simplified case is available from Crank
[1956].
Chapter V 108

The other analytica! solution used for comparison 1s based on the following
model simplifications:
1. No extemal mass transfer resistance: kr ~ "" .
2. The adsorption isotherm is again a linear function (as m Case 1):
Q* = Kh ·C.

This analytica! solution (Ruthven [1984]) is based on the loading Q. Details


about the foregoing analytica! solutions, the equations and methods used in
the comparisons can be found in Appendix 5.7 while some of the final results
will be discussed below.

Table 5.2 Relative difference in bulk concentration between numerical and


analytica/ salution (Ruthven) for several adsorption times when
using component mass balance (53) and overall mass balance
(3.13).

0.25 6.28 0.065


2.50 1.24 0.026
5.00 0.72 0.0018
7.50 0.53 0.0018
10.00 0.45 0.0018

Table 5.2 shows the relative difference in the amount of component adsorbed is
given at five different times after starting the adsorption, for both types of
mass balances outside the particle. The relative difference is defined as
follows:

(1-C ) - (1-C
an num
) c oum - can
r.d. = = (5.5)
(1-C ) 1 - c
an an
Experimental and numerical methods for kinetics 109

where:
Cnum : the numerically calculated dimensionless bulk concentration,
Can : the analytically calculated dimensionless bulk concentration.

Table 5.2 shows that in the case of mass balance (5.3) the relative difference
for short adsorption times is considerable. Results also indicate that with
mass balance (3.13) the relative difference is about 100 times smaller than
with mass balance (5.3) and rather close to the analytica} solution. The
reason may be that (3.13) is less sensitive to the steep concentration
profiles, which occur especially in the beginning of the adsorption run at
r"" R pa .
In view of the foregoing and for some other reasons, discussed in Appendix
5.7, the overall mass balance (3.13) was used in most model calculations.

Additional verifica tions


The model calculations described above were performed with a constant time
step of 180 seconds and a variabie space step grid of 185 steps. Some
additional numerical runs were also performed with a variabie time step, using
a very fine grid at the beginning.
The results showed no differences between the constant and variabie-time-step
numerical concentration decay curves for the fastest diffusing component, NBZ.
For the slowest diffusing one, IPP, some slight differences were observed
during the first 10 experimental hours. Details can be found in a report by
Donker [1992].

As the comparisons with analytica! solutions were carried out with simplified
versions, they do not prove completely that the normal (non-simplified) models
are solved correctly. Nevertheless, these checks give a good indication about
the correctness of the numerical solution and programs used.

5.2.3. PARAMETER FITTING AND OPTIMIZATION

Optimization procedure
In principle, the quality of a fit can always be expressed with a so-called
objective function f(P), whiclt depends on the different parameters to be
Chapter V 110

optimized: \,x 2, .. ,xN. These parameters are the coordinates of point P. This
means that the dimension of the objective function is equal to the number of
fit-parameters to be optimized. The objective function can be defmed in
different ways.
In this study the objective function has been defined as the Sum Of Squares
(SOS) of the difference between the experimental dimensionless bulk
concentration and the dimensionless bulk concentration calculated from the
model at times t. (j
J
= 1,2, ..,J) divided by the maximum time level number J:

j =J 2
f(P) = SOS = E ( cexp,l, I Cb0 C
num,t.
(P) I Cb0 ) IJ (5.6)
j =1 J
J

where:
j time level number;
J maximum time level number;
cexp,t, numerically calculated bulk liquid concentration at t.;
J
J
cnum,t. experimentally measured bulk liquid concentration at t.;
J
J
co initia! bulk liquid concentration, t = 0.
b

For multicomponent adsorption (where N is the number of components) the


objective function has been defmed as:

f(P) =SOS
n=N

[
J.
J = 0
=J
L (C n , exp,t /
J
c: - C n,num,t(P) I
J
c: ) 2
IJ
l (5.7)

n = 1

The best fit will be found for the model concentration decay curve which has
the lowest SOS. However, to really estimate how good or bad is the fit
characterized by a given SOS value, it is necessary to have a comparison with
the actual experimental reproducibility and also with the standard deviation
in the concentration measurements. The two criteria used for this comparison
Experimental and numerical methods for kinetics 111

were already discussed m section 5.1.4, while results are presented m


chapters 6 and 7.

Going back to the optimization procedure, in order to fmd the lowest SOS the
objective function, Eq. (5.6), bas to be minimized. There are many different
methods for function minimization discussed in literature. The easiest
procedure to find the minimum is (of course) just calculating f(P) for many
different points P. In case of a two dimensional minimization (two parameters
are varied to find an optimal fit) a grid of different points P (x ,x ) can be
1 2
made. The objective function can be visualized with a three dimensional
profile or with a contour picture in the grid domain. From a contour picture
the minimum (if present in the domain) can be easily found.

A much more sophisticated metbod is the conjugated gradient minimization


metbod [Press, 1989]. Besides the function value this metbod of successive
line minimizations uses also the gradient of the objective function in a
certain N-dimensional point P and the starting point for determining the new
line minimization direction is a second order Taylor approximation of the
objective function in a particular point P.
More details about this metbod can be obtained m the literature souree cited
above. It should be noted bere that the special optimization routines from
Numerical Recipes in Pascal were combined with the numerical solution programs
and were run on a VAX VMS system.

Results from parameter search and optimization


The initial parameter-fitting of the constant diffusion models was performed
by using a two-dimensional objective function, P (as defmed in Eq. (5.6),
where the fit-parameters were k and D . An example of a set of results for
f app
single-component IPP experiments is shown in Table 5.3.
The k values have been determined experimentally from the initial part of
f,exp
the concentration decay curve and details about the procedure and results are
given in chapter 6.

From the data presented in Table 5.3 it becomes clear that the two-dimensional
optimization (2-D in the table) gives in some cases very high values for kt'
Chapter V 112

It is obvious that external mass transfer coefficients in the range of 10· 1


m/s (cf. exp 52) are unrealistic and although, for all experiments in Table
5.3, the minimization procedure successfully found the P functions minima, the
results have no physical value. Because of these values a contour picture was
made to get more information about the behavior of the objective function P.
Figure 5.8 shows the contours of the objective function of a single-component
IPP experiment (exp. 32) when optimizing simultaneously k and D in a model
. f app
with constant diffusion coefficient. From the valley parallel to the kf axis
it can be seen that the SOS is very sensitive to changes in D and hardly
app

Table 5.3 R esults of 1-dimensional {1-D) and 2 -dimensional (2 -D) minimization


runs with 4-IPP experiments.

exp eb m R k time opt kf D Bi SOS


pa f,exp app
2
·104 ·105 ·105 ·1010 ·105
(mg/1) (mg/1) (m) (m/s) (hrs) (m/s) (m /s) (-)

26 14.81 51.5 6.0 4.84 71.0 2-D 11131 5.45 93.5 14.0
1-D 4.68 6.65 34.7
29.0 2-D 12.29 7.07 3.12
27 14.78 50.0 6.0 4.11 71.0 2-D 50.06 6.68 64.8 7.83
1-D 4.68 8.22 19.5
29.0 2-D 6.50 9.07 2.00
28 48.44 175.4 6.0 5.92 68.5 2-D 27.68 2.60 240 28.0
1-D 4.68 2.80 38.2
29 48.30 350.0 6.0 4.12 69.0 2-D 6.47 4.42 98.2 7.60
1-D 4.68 4.58 8.45
30 15.28 49.9 6.0 4.97 29.0 2-D 14.48 6.80 77.0 2.90
1-D 4.68 8.73 9.52
31 14.82 51.4 6.0 4.53 29.0 2-D 9.20 7.28 65.5 2.44
1-D 4.68 8.82 6.42
32 15.21 51.8 6.0 3.37 29.0 2-D 23.82 4.54 64.1 3.37
1-D 4.68 5.13 4.95
52 51.85 174.4 6.0 4.62 65.5 2-D 1377 3.37 144 15.4
1-D 4.68 3.68 26.2
53 51.26 524.9 6.0 3.79 65.5 2-D 3.76 7.81 51.1 8.01
1-D 4.68 7.32 10.6
Experimental and numerical methods for kinetics 113

sensitive to changes in k( In practical terms this signifies that the model


calculated concentration decay curve is strongly affected by even small
changes in D and at the same time is practically insensitive to changes in
app
kf (unless kf is very small). It also means that in this case the intemal
resistance to mass transfer is the rate-limiting factor (given by D ).
app

0.000020 0.000115 0.000210 0.000305 0.0004


KF
SOS 0.000047 0.000084 0.000121
0.000158 - - - 0.000194 0.000231
0.000268

Figure 5.8. Contour plot of the objective function for single-component IPP
experiment 32 (Bi = 64.1).

A common way to express the ratio of extemal to intemal mass transfer


resistances is by defining a Biot number:

krR
Bi = (5.8)
e pa Dapp

When D is constant each set of experimental conditions bas a unique value


app
of Bi. On this basis three ranges for the Biot numbers can he distinguished:
Chapter V 114

High Bi numbers. This indicates that intemal resistances to mass


transfer are the rate-limiting factors.
- Intermediale Bi numbers. This means that both extemal and intemal
resistances to mass transfer are important.
- Low Bi numbers. This shows that the resistance to extemal mass transfer
is limiting the overall rate of adsorption.
These ranges of the Biot number were found to be different for the various
adsorbent-adsorbate systems. For example, for the single-component IPP kinetic
experiments (calculated with the experimentally determined kf (see Section
6.1) and with the D found from 1-dimensional optimization) Bi varies from
app
37 to 231.

Similar ranges for the Bi number and types of surface plots were found hy
Traegner [1989]. He found that for batch kinetic experiments the range of Bi
numbers is usually between 1 and 100, where mass transport is influenced hy
both extemal and intrapartiele mass transport resistances. He concluded also
that for Bi >> 100 the numerical values for kf should be rejected as
inaccurate and that for Bi << 1 the values for the diffusion coefficient are
not reliable. For the region I < Bi < 100 the author recommends the use of
two-dimensional optimization for determining simultaneously the extemal mass
transfer coefficient and the diffusivity.

However, results ohtained in the present study indicate that even in the
intermediale range of Bi it is not advisable to use two-dimensional
minimizations for k and D . Table 5.3 shows that even for kinetic
f app
experiments with low Biot numbers most of the 2-D runs result in
unrealistically high values for kf, which have no physical meaning. These
unrealistic values for kf show that the optimization procedure is forcing the
k , mayhe in order to compensate for changes in D or for other deviations.
f ~
It is also an indication that prohably all kinetic experiments with IPP will
have a SOS ohjective function like the one shown in Fig. 5.8 (exp. 32), which
means that for this component intemal mass transport is the rate limiting
process. lt is reasonable to conclude, then, that the minimization procedure
finds a minimum in the valley, which gives an unrealistic value for kr due to
the insensitivity of the model curve to changes in k.f For this reason all IPP
Experimental and numerical methods for kinetics 115

kinetic experiments were fitted by one-dimensional minimizations (varying only


the diffusion coefficient) and an experimentally determined, fixed value for
the external mass transfer coefficient (k = 4.68·10- 5 m/s) was used (cf.
f
chapter 6).
Several values for D are listed in Table 5.3 (designated by "1-D"), which
app
show that the use of a lower value for k is compensated by a higher value for
f
D (up to 30% higher).
app

A similar analysis was performed on the results obtained from optimizing the
PNA kinetic experiments. The kinetic experiments with PNA showed not only
contour pictures with valleys parallel to the k -axis (like the 4-IPP
f
experiments) but also contour pictures with valleys parallel to the D -axis
app
and contour pictures of intermediate situations (cf. Fig. (5.9)).

Figure 5.9 shows the contour pictures of several PNA experiments with
different Biot numbers. From exp. 51, with Bi = 3.75 (Fig. 5.9.c) it can be
seen from the contours of the objective function that both the external and
internal resistances to mass transfer are important. On increasing the Biot
number the minimum gradually changes into a valley parallel to the kf axis
This is illustrated in Fig. (5.9.b), (exp. 19) and in Fig. (5.9.a), (exp. 20)
and it indicates that intrapartiele mass transport becomes the main
rate-limiting process.
This predicted behavior was confirmed by performing model simulations with
another model, the Maxwell-Stefan model which uses a constant MS-surface
diffusivity, D (cf. chapter 3), and for the same experimental conditions (cf.
s
exp. 20). Firstly, only the external mass transfer coefficient, kf, was varied
(Fig. 5.10.a), secondly, parameter variation only for D was used (Fig.
s
5.10.b). As predicted by the contour plot (cf. Fig. 5.9.a) the concentration
decay curves in Fig. (5.11.a) show that changes in the value of kf do not
affect the shape of the concentration decay curve. This also means that the
variations of kf will not affect the predicted value for the diffusion
coefficient.
At the same time, Fig. (5.10.a) shows that variations in D have a serious
s
influence on the shape of the concentration decay curve. This is confirnled by
the contour plot in Fig. (5.9.a).
Chapter V 116

3.05E-9
Bi = 26.3
a

2.l.E-9

l..l.SE:-9

0.000030 0.0001.72 O.OOÓ3l.5 0.000457 0.000600


KF
DAPP
4.575E-9 Bi= 15.3

3.l.5E-9

.000024 .000043 .000062 .000081. .0001.00


KF

Bi= 3.15
d

l..525E-8 7.625E-8

l..05E-8 5.25E-8

5.75E-9 2.875E-8
' ''

Bi = 0.521
l. E- 9 +----,:-------..--- - - - - - l SE- 9 --------,··-~--···------,
.000042 KF .000061. .000080 .000042 KF .000061.

Figure 5.9. Contour plot of the objective function for single-camp. PNA for
several experiments with different Biot numbers. The contour
lines have values of SOS similar to those in Fig. 5.8.
Experimental and numerical methods for kinetics ll7

1.0
Kf (m/s)

-
8
~
z
0.9

0.8
1 = 6E-5

2 = 6.5E-5

0
f= 3 = 7.6E-5
<(
er.
1- 0.7 4 .. 8.2E-5
z
~
z
0
u 0.6

0.5
0 20 40 60 80

TIME (hotrsl

Fig. S.lO.a. Maxwell-Stefan model calculations for one set of experimental


~rt~~~~~~;;;:;et;~j ;~!r!e;;~o::tan':(b.be!n /%;!]~' J~J~
1.0
Ds (m2/sl

ê
-
u
~
z
0.9 1 - 2.5E-14

2 .. 5E-14

0 0.8
f= 3 = 7.SE-14
~
1- 4 - 9.8E-14
z
UJ
0.7
u
5u 0.6

0.5
0 20 40 60 80

TIME <hotrsl
Fig. S.lO.b. Maxwell-Stefan model calculations for one set of experimental
conditions (ex;p. 20) where only Ds has been varied. while all
_other parameters were kept constant (kt = 7e-5 mis).
Chapter V 118

Back to the constant D model, when the Biot number decreases the minimum of
app
the objective function transforms into a valley parallel to the D - axis,
app
which indicates that external mass transfer becomes the rate-limiting process
(Fig. (5.9.d)). It is obvious that the parameter value representing the mass
transfer process which is not rate determining (k for high Bi, D for low
f app
Bi) and which is determined from a two-dimensional minimization run will not
be reliable. For these reasons and because an experimentally determined value
for kr was available, only one-dimensional optimizations were performed for
the PNA experiments.
The same type of optimization procedures were used for the third component,
NBZ and the results will be discussed in chapter 6.

In the case of the two-component systems it was necessary to use


two-dimensional minimizations, due to the presence of two diffusion
coefficients. The procedure used for the two-component systems was the same as
for the single-component ones, therefore it will not be discussed further.

5.2.4. INFLUENCE OF EQUILIBRIUM ON PARAMETER FITTING

The results from parameter fitting with different single-component models are
presented in chapter 6. In this section, however, in view of the discussion
that follows, a limited amount of calculation results will be presented,
obtained by applying the Maxwell-Stefan model to the NBZ system.

As already noted in chapter 4 (cf. Section 4.3.1), in the case of NBZ,


differences ~ere observed between the equilibrium points obtained from
equilibrium data measurements and the equilibrium points found from those
kinetic experiments which had reached equilibrium. The reason for
investigating this problem were the poor fits obtained when applying the
Maxwell-Stefan model to NBZ experiments. lbese fits were especially poor in
the last part of the concentratien decay curve, where concentration
practically does not change anymore. An example is given in Fig. 5.11.
Experimental and numerical methods for kinetics 119

1.00

0
0
..... 0.90
f1
z
0 0.80
i=
<(
((
1- 0.70
z
~
z
0 0.60
0

0.50
0 10 20 30 40 50 60 70

TltvE (hoursl

Experiment + MS-model t:,. MS-model


isoth. 1 isoth. 11

Figure 5.11. Experimental curve and Maxwell-Stefan model


curve with adsorption isotherm 11, exp. 208

Table 5.4. Calculated surface diffusivities with Maxwell-


Stefan model using isotherms I and /1, NBZ

Ads.isotherm I Ads.isotherm 11
SOS Ds(MS)
*lE+4 *1E+13
[m21sJ

200 24.89 51.30 1.59 0.42


201 20.00 1.64
202 24.50 0.55
203 24.56 0.21

209 2.39 0.38


210 50.34 2.07 0.61 3.26 0.04
211 23.54 2.72 0.23
212 12.06 2.96 3.08
Chapter V 120

Figure 5.11 shows big differences between the model predictions with isotherm
1 and 2. It is also clear that although the experiment lasted for more than 60
hours, equilibrium had been reached already after 40 hours. In the long
equilibrium part of the curve the value of D has practically no influence,
app
while the most important factors then are the equilibrium isotherm parameters.

In view of the foregoing it was decided to use the equilibrium points from
those kinetic experiments where full equilibrium had been reached. The
isotherm parameters obtained were already presented in chapter 4. Further,
optimization runs with the MS-model were performed for all NBZ experiments and
the results obtained with both isotherms are listed in Table 5.4.

From Table 5.4 it can be seen that when using isotherm 2 (kinetic) the SOS's
for practically all experiments (except 201) are significantly smaller than
those with isotherm 1. This shows indeed that the main reason for the bad fits
were the isotherm parameters and the improverneut which occurs when applying
isotherm 1 is quite obvious in Fig. 5.11.

Another important question which arises from the foregoing discussion is the
usefulness of fitting concentration decay curves with long equilibrium parts.
As mentioned earlier, these equilibrium parts do not give any valuable
information about the diffusion coefficient and do not contribute toward
better understanding of the kinetics of the adsorption process.
Several arguments may be raised here. On one hand, using such experiments may
lead to erroneous results for the diffusivity as the SOS, which is minimized,
is calculated over the whole curve, including the equilibrium part. The
'
optimization procedure may try to improve the fit of the equilibrium part at
the expense of the steeper parts of the curve, which are the ones determined
by the kinetics. On the other hand, if the adsorption isotherm used is
correct, the equilibrium part of the COC should not influence the optimal
value of the intrapartiele diffusivity.
In order to check if the above is indeed true the MS-model with isotherm 2 was
applied twice to the same experiments, with and without the equilibrium part.
The results are presented in Table 5.5.
Experimental and numerical methods far kinetics 121

Table 5.5. Camparison between MS-model calculations for NBZ


experiments with and without equilibrium part
with eq. without
part eq.part
E)(p; Co m time Ds(MS} SOS time Ds(MS) SOS
....
.. .· .... *1E+1S *1E+4 *1E+13 "'1E+4

... (mg/JJ (frlg/l) . [hours) [mals) (bours] (m2/s)

200 24.89 51.30 64.0 2.15 0.42 47.0 2.73 0.21


203 24.56 25.48 63.0 1.92 0.21 32.0 2.35 0.26
206 59.44 245.15 67.5 1.97 2.41 45.0 2.51 1.36
207 59.20 125.45 63.5 2.99 0.49 35.0 2.99 0.84
208 59.78 75.40
~3.42 0.22 35.0 3.42 0.35
50.34 24.44 3.26 0.04 25.0 .· 3.26 0.07

From Table 5.5 it can be seen that the Ds for exps. 207, 208 and 210 is
practically unchanged, while for exps. 200, 203 and 206 it increases for the
calculations without equilibrium part. As a whole, these results show that the
MS-model with isotherm 2 behaves as expected and the higher SOS's in Table
5.5 can be explained by the fact that the good description of the equilibrium
part is not included in the calculated value for the SOS.
For the above reasons it was decided to shorten those experiments where clear
equilibrium had been reached and do not use these parts in the mathematica}
modeling.

Finally, it should be noted that the foregoing problems arise only in the case
of NBZ, which reaches equilibrium faster than IPP and PNA and for which
component most of the available experiments have reached equilibrium. For
component IPP there are no experiments available in which equilibrium had been
reached and for component PNA there are very few such experiments. As it was
already shown the inaccurate equilibrium parameters influence the results of
the fitting mainly in the long equilibrium part of the curve. As such a part
is not present in the cases of IPP and PNA, it may be expected that the
diffusivity values obtained for these two components will be less sensitive to
variations in the equilibrium parameters.
Chapter V 122

5.3. CONCLUSIONS

A large number of adsorption kinetic experiments has been performed for all
single- and two-component systems used in this study. The main conditions
varied were the initial concentration, amount of carbon and type of organic
component and the experiments were fairly reproducible.

The comparisons of the numerical solutions (using component mass balance) with
the analytica! ones showed that for short adsorption times a deviation existed
between them. For Jonger adsorption times the agreement was much better.
The replacement of the component mass balance over the bulk salution with the
overall mass balance produced a much better agreement between the numerically
and analytically calculated concentratien decay curves, also for short
adsorption times.

From the two-dimensional optimizations performed and the contour plots made it
becomes clear that the determination of the fit parameter which describes the
non-rate limiting mass transfer process is unreliable. It can also be
concluded that the concentratien decay curve practically does not contain
information about the mass transfer process which is not rate limiting. This
means that kinetic experiments with high Biot numbers are not suitable to
determine kf and experiments with low Biot numbers are unsuitable for
determining the diffusion coefficient.
In the two-dimensional optimizations with experiments with high Biot numbers
the kf parameters could be used to compensate the bad description of the
internal mass transfer processes, producing unrealistic values for k . Oue to
' f
this occurrence the optimal value of the diffusion coefficient from a
two-dimensional fit will be less reliable than the one found from a
one-dimensional fit with a realistic fixed kf .

The numerical concentratien decay curves (COC) are quite sensitive to even
small variations in the equilibrium isotherm parameters. This influence is
present over the whole COC, but is especially strong in the part of the curve
where equilibrium has been reached. It is of little value to fit the
equilibrium part of the COC, since it gives no information about diffusion.
Chapter VI 123
ESTIMATION OF RATE PARAMETERS FOR
SINGLE-SOLUTE SYSTEMS

6.1. ESTIMATION OF EXTERNAL MASS TRANSFER COEFFICIENT

In genera!, the value of the extemal mass transfer coefficient cao be


evaluated by various methods. One such metbod is based upon using a
single-resistance model, which assumes that only extemal mass transfer plays
a role and neglects the influence of any intrapartiele resistances throughout
the whole adsorption process. In this type of model, already mentioned in
chapter 3, the values obtained for k are normally lower than the
f
experimentally determined ones, due to the assumption that all resistances are
concentraled in the film layer. It was already shown in chapter 5 that for
experiments with Biot numbers in the intermediale range where both extemal
and intemal resistances to mass transfer are important and even more for low
Biot numbers, using a lower-than-the-real-one value for kf produces
immediately a higher value for the intrapartiele diffusivity.

Van Lier [1989] applied another metbod for evaluating kf based on the
assumption that in the very beginning of the process extemal mass transfer
controls the rate of adsorption. The equation required for calculating the
extemal mass transfer coefficient is the mass balance of the component in the
bulk solution:

(5.3)

When the limit is taken for t -? 0 we have

ae m k A
lim
t-?0 at
b = -
V
f s [ li m
t-?o
eb limC
1-70 p
IR)
pa
(6.1)

Assuming that in the beginning the concentration at the extemal surface of


the partiele cao be neglected with respect to the bulk concentration, i.e.
ep IR « eb , Eq. (5.3) reduces to:
pa

mkf A se
____ (6.2)
b
V
Chapter VI 124

Using the initia! condition that at t = 0 we have eb C: and


integrating Eq. (6.2) we obtain

m k A
In C = - _ _f _ _s t + In C' (6.3)
b b
V

The value of k can now be obtained from the slope of the plot of ln C versus
f b
t for low villues of t.

Table 6.1. Calculated external mass transfer coefficients for


several single-solute IPP experiments

24.16
24.16 232.8

49.05 176.82 0.60 0.74


49.92 351.4 0.60 0.76
5 49.03 349.9 0.60 0.74

9 97.28 175.2 0.60

7 50.00 167.5 0.26 0.68


8 48.99 350.4 0.26 0.71

The foregoing method was also used in this study. Table 6.1 gives an example
of values of the extemal mass transfer coefficient for several
single-component IPP experiments, estimated by using this method. All values
were evaluated for two different time intervals, 15 and 30 min, starting
always at t = 0.
Table 6.1 shows that the shorter time interval (15 min) gave kf values 18% to
33% larger than the longer one (30 min). This means that intemal diffusion
most likely has already an influence in this part of the curve (after 15 min)
Estimation of single-solute kinetic parameters 125

and that the real kf values are probably slightly higher than the values given
in the table. Table 6.1 shows also the influence of the partiele radius on the
kf values, which influence will he discussed in more details later in this
section.

The experimentally determined external mass transfer coefficients for the


three single-component systems, based on the 15 rninutes interval, are
presented in Tables A 6.1.1, A 6.1.2, A 6.1.3 in Appendix 6.1.

As the geometry of the particles, the stirrer speed and the temperature of the
solution were the same for most experiments (although several IPP and PNA
experiments were performed at a slightly higher temperature 23°C instead of
20°C), it can he expected that the values of kf would he practically equal for
all experiments for a given component. This is indeed the case for NBZ (cf.
Table A 6.1.1, Appendix 6.1)) where the kf value for most experiments is in
the range of (6.5 7)·10-5 rn/s.
The values for PNA (cf. Table A 6.1.3, Appendix 6.1) show, for particles with
R = 0.6 mrn, a sirnilar range as those for NBZ. The results for IPP (cf. Table
pa
A 6.1.2, Appendix 6.1) gave a larger spread but as the contour plots in
chapter 5 showed it is quite reasonable to choose for the model calcultions
one fixed kf value.

An alternative way of estimating the values of the external mass transfer


coefficient is to use hydrodynarnic correlations from literature. There are
several correlations descrihing continuous mass transfer for solid particles
in a stirred vessel, which could be used for the calculation of the external
mass transfer. Two of these relations have been considered in this study:

4
- -
Huige
-
[1972] (for e dpa I v 3 < 106):

(6.4)

- Letterman [1974]:
Sh = 2 + 0.64 (e d4 I v3)0.I97 Sco.33 (6.5)
pa
Chapter VI 126

where:
Sh : k d I D
f pa f
Sc :v/Df
E : stirrer power input per kg (m 2 .s-3)
2 1
V : kinematic visoosity (m .s- )
d : partiele diameter (m)
pa
2 1
Dr : free liquid diffusivity (m .s- )

The stirrer input was determined independently and had a value of 0.32 m2/s3 ;
for the kinematic viscosity the value for water was used (V = 10-6 m2/s). The
values for the free liquid diffusivities were measured separately and were
already presented in chapter 4 (cf. Table 4.1).
Applying correlations (6.4) and (6.5) to the data for each component results
in the average values for kf listed in Table 6.2.

Table 6.2. Calculated average values for "J [mis]. using hydrodynamic
correlations, for partiele diameter 0.60 mm.

kf.(NBZ)
*1e+5

7.4
9.0

The above correlations describe also the relation between kr and the partiele
diameter. This relation has been compared to the experimental values found for
kf for carbon' particles with different partiele sizes. The results obtained
for PNA are shown in Fig. 6.1. lt is interesting to note that the experimental
values show the same trend as the correlations.

Figure 6.1 shows also certain differences between the batch vessels which may
be due to differences in the haffles of vessels 1 and 2, resulting in more
turbulence in vessel 2 and smaller extemal mass transfer coefficients. To
prevent this and also the formation of air bubbles during mixing, a larger
amount of salution was used in vessel 2 (2.5 instead of 2 L).
Estimation of single-solute kinetic parameters 127

13

12

11 ..

10
11)
'
~ 9

~ 6
il
7

4'
0 0.2 0.4 0.6 0.6
Rpalmml
o ve ..... JI o veaael2 Hui ge

Figure 6.1 The external mass transfer coefficient as a function of partiele


size: experimental and according to correlations

Finally, it should be reminded once again that in most model calculations a


fixed value for kf has been used for each component. The kf value for a given
solute was obtained by averaging all experimental values from the
single-solute experimental concentration decay curves of the respective
adsorbate.

6.2. MODEL CALCULATIONS FOR SINGLE-SOLUTE SYSTEMS

6.2.1. INTRODUCTION

Several mathematica! models were used in this study to simulate the adsorption
behavior of the organic compounds on activated carbon. A detailed denvation
of three of these models was already given in chapter 3. A fourth model,
referred to as the Homogeneous Partiele Transition Model (HPMT) has not been
discussed previously, as it has a rather empirica! basis. Details about this
model are given in Appendix 6.2.
In the following paragraphs the calculations of all models will be compared
and the results discussed for each of the three components used in this study.
Chapter VI 128

6.2.2. MODEL WITH CONSTANT DIFFUSIVITY (HPMC)

As noted in chapter 3, the model with a constant D (or D ) is identical


app app
for the cases of Fickian and Maxwell-Stefan diffusivities, and consequently,
only one type of diffusion coefficient can be obtained from this model. This
model with a constant D is reffered to as the Homogeneaus Partiele Model
app
Constant (HPMC).
The HPMC model was used for simulations for all three single-component systems
investigated in this study. Part of the results of 1-dimensional and
2-dimensional minimization runs with IPP experiments with the constant D
app
model (HPMC) were already given in chapter 5 (cf. Table 5.3). The results of
another set of 1-dimensional runs with the same model are presented in Table
6.3.
The 1-dimensional minimizations in Table 6.3 were performed with fixed values
of k(

Table 6.3. R esults of I -dimensional minimization runs for several


lPP experiments
exp ëo m time kf Dapp Bi SÖS
*tE+5 *1E+10 *1E+4
.. (mg/1] .•. (mg/1] .. [hrs] (mis] (m2/s] 1-1

1 24.16 168.8 19.5 5.9 9.47 63.6 1.55


2 49.05 176.8 69.0 5.9 3.92 125.5 5.38
3 24.16 232.8 109.5 5.9 8.25 92.3 4.08
4 49.95, 351.4 41.5 5.9 5.23 96.7 3.81
5 49.03 349.9 95.0 5.9 4.87 106.6 11.2
6 49.14 157.5 109.5 5.9 2.42 234.4 16.5
7 50.00 167.5 91.5 5.9 6.11 106.6 7.73
8 48.99 350.4 70.5 5.9 18.57 34.6 0.99
9 97.28 175.2 109.5 5.9 2.19 197.8 4.06

The results from model calculations with PNA are presented in Table 6.4. In
several instances 1-dimensional minimizations were carried out besides the
2-dimensional ones, in case the latter resulted in unrealistically high values
Estimation of single-solute kinetic parameters 129

Table 6.4. Results of 1- and 2-dimensional minimization runs


for several PNA experiments

20 30.34 48.6 2-D 37.64 1.64 26.3 2.16


1-D 7.00 2.44 3.40
37 29.88 50.5 2-D 14.48 3.99 16.6 0.75
1-D 7.00 1.60
19.89 50.2 2-D 10.55 .4 0.50
1-D 7.00 1.90
9.80 100.9 2-D 6.27 0.09
42 9.92 74.6 2-D 6.44 0.05
1-D 7.00 41.94 0.57 0.17
43 9.86 49.6 2-D 6.12 25.90 2.41 0.15

Table 6.5. Results of 1-dimensional minimization runs


for NBZ experiments

51.30 7.0
32.76 7.0 0.63
51.40 7.0 9.59 7.11 5.64
25.48 7.0 8.51 7.99 2.10
206 59.44 245.15 7.0 30.75 2.22 8.11
207 59.20 125.45 7.0 7.28 9.33 6.36
208 59.78 75.40 7.0 7.48 9.12 3.36
209 30.04 124.10 7.0 13.04 22 2.66
210 50.34 24.44 7.0 6.63 10 0.80
75.00 7.0 21.78 3.12 0.29
24.44 7.0 29.27 2.34 1.74
Chapter VI 130

for kf' In the same table the Biot numbers are also given, calculated with
k f.exp and Dapp of the 2-dimensional minimizations.
The results of model calculations with the HPMC model for the single-component
NBZ system are listed in Table 6.5.

Below, Figure 6.2a/b shows an experimental concentration decay curve and two
HPMC curves (1-D and 2-D) plus the respective residue plots for an IPP
experiment. These fits are characteristic for all IPP experiments, fitted with
the HPMC modeL Although the IPP experiments are rather insensitive to
variations in kf, the decay curve obtained from the 1-dimensional minimization
run (with preset, fixed k) gave a higher SOS than the decay curve obtained
from a 2-D minimization. This is, of course, logica! and it can be seen from
Table 5.3 and also in Fig. 6.2A. In Fig. 6.2B a plot of the residues is given
which shows that the initial part of the 1-D model curve gives a much better

b 0.025

+ + +

..
1-Q

!
0.015 .... ++ ++ ••••••+
....
a
u j
0.005 ..
+ ~ 'f/9vVVVVV9Vf!hr+++
Vf/'1/VVV VVV'$..-,
~.
~
-O.oo5

. V +•• vv9Vv
•. Vvvv v

ê
!
+++
+
-... .
V v9
.....
ü
~ 1.0
-0.015
-....
z -o.025
10 20 30
0 0
i= TllvE <nou-sl
<i 0.9
a:
1-
zw
ü
z 0.8
8
0.7 L __ _ _ _ ___J_ _ _ _ _ ____,__ _ _ _ _ ___,

0 10 20 30

TIME (hours)

exp "V 2-D + 1-D


Figure 6.2 a Dimensionless experimental and HPMC model curves
(1-D and 2-D) for IPP experiment (No. 30)
b Residue graphs of 1-D and 2-D model curves
Estimation of single-solute kinetic parameters 131

agreement with the experimental curve in contrast to the 2-D curve, which
indicates that the fixed kr value (kf = 4.68e-5 m/s) is indeed more realistic
that the one found from the 2-D minimization. Fig. 6.2b shows also that the
bad agreement of the later part of the 1-D model curve is responsible for the
high total SOS. If we assume that the experiments are correct, then the
disagreement would be due to a bad description of the intrapartiele mass
transport by the HPMC. It is also clear that the 2-D optimization produces a
lower total SOS due to the fitting of kf' which not only gives an unrealistic
k but also a less reliable value for D
f app
Table 6.3 shows that the range of optimal apparent diffusivities, found by
perforrning 1-dimensional optimizations (with a fixed kf value) is (2.2
18)·10' 10 m2/s. The other set of IPP experiments, listed in Table 5.3, shows a
range for D of (2.8 - 9.0)·10- 10 m2/s from 1-dimensional optimizations
app
and a range for D of (2.6 - 8.6)·10' 10 m2/s from 2-dimensional ones.
app
These rather large ranges indicate that the HPMC does not describe well the
overall rate of adsorption. Another indication for that is the poor quality of
the fits. It is clear, then, that because of these factors the HPMC model has
little predictive value for the single-component IPP system.

It may be of interest to compare the ranges of Biot numbers for the three
components under investigation. The comparison shows that Biot numbers for NBZ
(2.1 - 9.3) are much smaller than those for IPP (34 234) and within the
range of those for PNA (0.5 - 26). This indicates that both for NBZ and PNA
the rate of adsorption will depend more strongly on the resistance to extemal
mass transfer than for IPP. Most NBZ and PNA experiments with lower Biot
numbers gave better fits and it means that the HPMC can model the far simpter
extemal mass transfer better than the intrapartiele one.

Tables 6.4 and 6.5 indicate that the ranges of optimal apparent diffusivities
2
found for PNA and NBZ are (1.7 - 38.6)·10·9 m2/s and (6.6 30.8)·10'9 m /s,
respectively. It is clear that both ranges are rather large and one reason
for this may be the urneliabieD 's calculated from experiments with minimum
app
valleys parallel to the D axis. Another reason probably is, as in the case
app
of IPP, that the overall rate of adsorption cannot be described well with the
HPMC model which uses a constant diffusion coefficient.
Chapter VI 132

1.6 . - - - - - - - - - - - - - - - - - - - - - - - - - ,

(j)
N 1.2
..5

0.8
V + +
+
+
+
V+
0.4 V
+
+ V

0.0 L __ _ __L__ _ __.l__ _ ____.__ _ _ _...___ ______;

0.07 0.13 0.18 0.24 0.29 0.35

Co/m (kg/kg)

'V Exp. set 1 + Exp. set 2

Figure 6.3. The apparent diffusivity, D , as a function of the cfJtm ratio


app b
for 1PP experiments (set 1 see Table 6.3; set 2 Table 5.3).

2.5

(j) 2.0
.......
(\1
..5
1.5
ro
I V
w
T"" V
V
.! 1.0
a.
a.
(\1
0
0.5 V
vj
'11
VVV
V V V

0.0
0.0 0.3 0.6 0.9 1.2 1.5

Co/m (kg/kg)

Figure 6.4. The apparent diffusivity, D , as a function of the cfJ1m


app b
ratio for PNA experiments
Estimation of single-solute kinetic parameters 133

In order to check whether experimental conditions influence the values found


for Dapp, values from different experiments can presented as a function of the
0
corresponding C /m ratios. Figs. 6.3 - 6.4 show this relation for sets of IPP
b
and PNA experiments.
The first set of IPP experiments in Fig. 6.3 shows a slight trend of
decreasing D with increasing r!/m ratio. The second set, however, does not
app b
show this trend, and it even seems that for several experiments (cf. exps. 28,
29, 53 in Table 5.3) the apparent diffusivity decreases with an increasing
amount of carbon, m, and a constant C0 , i.e. with a decreasing r!/m ratio.
b b
The same trend, as shown by the first set of IPP experiments, is also present
in the cases of components PNA (cf. Fig. 6.4) and NBZ and is even more
pronounced than for IPP.

Another way of looking for indications of a possible concentration dependenee


of the D is plotting the apparent diffusivity values from the different
app
experiments versus the corresponding ultimate carbon loading, Q 00 ( at t=oo).

5.0

-
ûl
(\J
s
4.0

3.0
ro
I
w
,....
.! 2.0
a.
a.
1\S V
0 vv
1.0
V

""" V ~V
0.0
0.10 0.14 0.18 0.22 0.26 0.30

Q(inf) (kg/kg)

Figure 6.5. The apparent diffusivity, D


app
, as a function of the ultimate
loading, Q00 , for PNA experiments

The ultimate carbon loading, Q00 , bas been calculated via Eq. (6.6) where the
uptake of component in the pore liquid of the carbon partiele is neglected:
Chapter VI /34

m Q00 = V (c"b - C00 ) (6.6)

An example of the results obtained for PNA is given m Fig. 6.5 where
0
different Cb/m ratios or Q00 values may represent different loading situations
for the different kinetic experiments. Another possibility is that the same
c:/m ratio in two experiments represents different loading situations or that
the same Q 00 has an entirely different loading history. The relation

shown in Fig 6.5 indicates that the actual diffusivity probably depends on the
carbon loading, as previously suggested.

Although all model calculations presented show that the HPMC does not describe
well any of the three single-component systems under examination the results
obtained still give some valuable information about the possible physical
picture inside the carbon partiele during adsorption. As already noted the
apparent diffusivity consists of a pore liquid diffusion and surface diffusion
contributions. At the same time, due to constriction and tortuosity factors it
can be expected that the pore diffusivity will be lower than the corresponding
free liquid diffusivity, D f (cf. Table 4.1). By taking these factors into
account it becomes clear from the values presented in Tables 6.3 - 6.5 that
the term containing the surface diffusivity (cf. Eq. 3.14) will be at least of
the same order of magnitude as, or even up to three magnitudes higher than,
the one containing the pore liquid diffusivity. This means that a model with a
variabie D with more emphasis on a good description of the surface
app •
diffusion term and not neglecting the possible variations in D ·(BQ !aC ) may
s p
give a better description of the IPP, PNA and NBZ adsorption systems.

6.2.3. FICKIAN (HPMF) MODEL WITH VARIABLE DIFFUSIVITY

As mentioned already in chapter 3, the diffusion coefficient inside the


particle, as defined in Eq. (3.12), can be variabie even when D and D are
p
constant, due to the changing isotherm slope. At the same time the surface
diffusivity, D , itself may be a function of different parameters.
s

Initially, a relation was used in this study for D in the HPMF model, a power
s
law dependenee of D on the carbon loading:
s
Estimation of single-solute kinetic parameters 135

(3.24)

and D is given by:


app

(1-e ) a aQ
Dapp "" Dp + e pa pc D so Q - - (3.25)
pa ac p
A series of 2-dimensional minimizations were performed, where the optimized
parameters were D and a. The results for several IPP, · PNA and NBZ
so
experiments are presented in Tables 6.6, 6.7 and 6.8, respectively. lt should
be noted that in all model calculations that follow a fixed, averaged value
for kf has been used for each component.

Table 6.6. Results from 2-D optimizations with Fickian


model with variabie diffusivity, IPP

27 14.78 50.0 0.30 0.093


29 48.30 350.0 0.16 0.25
52 51.85 174.7 0.17 0.24
54 .39 653.6 0.17 0.23

The analysis of IPP results in Table 6.6 shows that the optima! values of
parameter a are relatively close to zero, which more or less eliminates the
influence of the term Qa from Eq. (3.25). This trend is even more pronounced
in the case of PNA (cf. Table 6.7) where all optimal a's are in effect zero.
The limited amount of NBZ experiments optimized in this way and given in Table
6.8 show similar results.

There are two possible explanations for the foregoing results.


Firstly, the minimization procedure may have found a local minimum in the
a-D domain while the real minimum is situated in a point far away from the
so
Chapter VI 136

Table 6.7. Results from 2-D optimizations with Fickian


model with variabie diffusivity, PNA
...
! Ëxp~
·.· ......
Co m Ot;Q \ •••• > ~lfa•········· r SOS
II··...·····
• "1E-t-'IS I \•····•·•.· 1;:+4
I i .•. .[mg/1) [mg/1) (m~sJ
·.·•· ...

..... ·.···· .·········.····


.
...•. ···•···

17 29.49 98.75 1.33 0.043 2.07


20 30.34 48.64 0.93 0.070= 3.56
22 10.21 49.92 4.17 0.070 0.56
.08 50.72 1.62 0.032 1.34
37 29.88 50.60 1.69 0.032 1.44
38 19.89 50.16 1.52 0.034 1.64
1 9.80 100.9 7.47 0.003 1.80
9.86 49.56 7.13 0.003 1.11

Table 6.8. Results from 2-D optimizations with Fickian


model with variabie diffusivity, NBZ

24.89 51.30 0.35 0.64


24.56 25.48 0.41 0.34
'
23.54 75.00 0.02 0.59

local one. In that case the real optima! a's (and D 's) may be fully
so
different from the ones presented above.
Secondly, it is possible that these are the real minima of the SOS parameter ,
in which case the power law expression used to describe the concentration
dependenee of the surface diffusivity (cf. Eq. (3.24) becomes obsolete. This
would mean that, given the varlation in a for IPP and NBZ, the predictive
value of the model becomes questionable.
Estimation of single-solute kinetic parameters 137

In order to investigate whether the first hypothesis is true several


1-dimensional optimizations were performed per experiment by inputting each
time a fixed, but different value for a and optimizing only D . This
SQ

procedure was applied to several experiments for each component. Results


showed that moving away from the optimal a's, found from 2-D optimizations,
increased in all cases the SOS's and therefore, decreased the quality of the
fit. As the results of these 1-D minimizations were used only as a check and
were not relevant to the following stages of HPMF modeling, they will not be
discussed further. More details about the use of such 1-D and 2-D
minimizations are given in Appendix 6.3, where also another empirical loading
dependenee of D , the Transition (HPMT) model , is discussed.
s

On basis of these results, it was concluded that the minima found by the 2-D
optimizations are probably the real ones and consequently, a better approach
would be to use a Fickian model with a constant surface diffusion coefficient.
This still means that D is variable, due to the presence of the term
* app
aQ /BCp in Eq. (3.25), in contrast (for a=O) to the Transition (HPMT) model
which will be briefly discussed in the following section.

The results in Tables 6.6 - 6.8 show that, although in most cases the optimal
a is relatively close to zero, it is not exactly zero. Moreover, for different
experiments of a given component, especially in the cases of IPP and NBZ, a
can vary more than 100%. This makes the comparison between the D values
SQ

rather dubious, as they have been calculated with different a's.

In view of the foregoing, additional 1-dimensional minimizations were executed


for IPP and NBZ, optimizing D and fixing the a to be always equal to zero.
SQ

As noted, this makes the HPMF a Fickian model with variabie D , but constant
app
surface diffusion coefficient. No such additional runs were performed for PNA,
as the previous values obtained for a were very small, and therefore
practically zero.

Tables 6.9 and 6.10 list the results from these additional 1-D minimizations
for IPP and NBZ, respectively.
Chapter VI 138

Table 6.9. Results from 1-D optimizations with Fickian model


with constant surface diffusivity for IPP

14.78 2.00
48.30 8.10
14.82 0.56
51.85 174.7 4.97
51.26 525.2
99.39 653.6
98.11 172.0

Table 6.9 shows that the range of the Fickian surface diffusion coefficients
for IPP is (1.7 - 27.7)·10' 15 m% with a relative standard deviation of 66%.
Surprisingly, this is even higher than the relative standard deviation (48%)

1. 1

0 1.0
ü
'
Q
0.9
ö
i=
<{ 0.8
a:
1-
z
w 0.7
ü
ö
ü 0.6

0.5
0 20 40 60 80

TIME {hOL.f"S)

- - exp + HPMF

Figure 6.6. Dimensionless experimental and HPMF model


concentration decay curves for exp. 56, IPP
Estimation of single-solute kinetic parameters 139

of the D app 's range obtained from the constant diffusivity model (cf. Tables
5.3 and 6.3). An example of an experimental and HPMF model dimensionless
concentration decay curves for IPP is given in Fig. 6.6.
The curves on Fig. 6.6 clearly show that the HPMF model does not describe very
well the overall rate of adsorption of IPP and that the same trends are
observed as in the case of the constant diffusivity model: initially the model

1.2

0 1.0
u....._
~
z 0.8
0
i=
<( 0.6
([
1-
w
z
0.4
uz
0
u 0.2

0.0
0 10 20 30 40 50 60

TIME (hou-s)

- - exp + HPMF

Figure 6.7. Dimensionless experimental and HPMF model


concentration decay curves for exp. 38, PNA

curve runs too high, which means that the diffusivity value used in the model
is too low, while later the curve is too low, meaning that the diffusivity is
too high compared to the real situation as described by the experimental
curve.
The example of experimental and model concentration decay curves for PNA,
given in Fig. 6.7, shows a similar, although more difficult to see, trend.
However the absolute value of this systematic deviation between experimental
and model curves is far less pronounced for PNA than it is for IPP.
The range of diffusivities for PNA was already given in Table 6.7. As
mentioned earlier, the very small a's were considered to be practically zero,
therefore no 1-dimensional optimization were performed. Here, in contrast to
Chapter VI 140

component IPP, the relative standard deviation of D (83%) 1s lower than the
s
one for the constant diffusivity (93%).

Table 6.10. Resultsfrom 1-D optimizations with Fickian (HPMF)


model with constant surface diffusivity for NBZ

.. Exp. Co m Ds(Fick) SOS


.. .. *1E+13 *1E+4
: .::.
[mg/1] [mg/1) [m2/s)

200 24.89 51.30 7.15 0.77


201 20.00 32.76 0.1.5 21.43
202 24.50 51.40 6.05 1.19
203 24.56 25.48 10.73 1.22
206 59.44 245.15 5.23 : 1.46
207 59.20 125.45 7.27 2.49
208 59.78 75.40 7.01.· 0.39
209 30.04 124.10 8.45 1.02
210 50.34 24.44 3.78 0.08
211 23.54 75.00 5.90 9.40
212 12.06 24.44 3.97 5.36

The results for NBZ, listed in Table 6.10 show a slightly different picture
than that for PNA. The range of Fickian surface diffusivities is (0.15 -
2
10.73)·10- 13 m /s, with a relative standard deviation of 46%. This RSD is
almast twice as low as the PNA one and is camparabie to RSD for IPP.
From the ranges of Fickian surface diffusivity it becomes clear that the HPMF
model does not have predictive value for any of the three single-component
systems investigated and is necessary to look for a better model description.

6.2.4. TRANSITION MODEL (HPMT) BASED ON TUE PORE


LIQUID CONCENTRATION GRADIENT

The third model can be classified as an intermediate one between models based
on Fickian and Maxwell-Stefan diffusivities. Due to its empirica! background
this model has not been discussed separately in chapter 3, therefore its basis
Estimation of single-solute kinetic parameters 141

and the obtained results are described more extensively in Appendix 6.2. In
the present section only some general trends will be briefly discussed.

In the HPMT model the concentration dependenee of the surface diffusivity is


included in the apparent diffusivity expression in the following way:

(l.
(1-f ) [
DT
app
= D +
p
----:::-'-pa_
t
pa
p DT
c so ) (6.7)

where:
DT :surface diffusivity for a carbon toading equal to zero [m2 .s- 1]
so
Qref
:constant [kg.kg- 1]
(l. :constant [-]

The model was fitted to several IPP and PNA experiments. Initially, a limited
number of 2-dimensional minimizations were performed, where the optimized
parameters were DT and a.. Most of them gave a value of a approximately 0
so
which, in view of (6.7), means that D is constant. An example of some 2-D
app
results is given in Table A 6.2.1 in Appendix 6.2.
From contour plots of the a.-D domain it was also found that the two
so
parameters were dependent on each other. Several experiments per component
were chosen and for each one a number of 1-dimensional minimizations were
performed, optimizing D and keeping a constant, but different for each run.
so
The results are presented in Tables A 6.2.2 and A 6.2.3 in the same Appendix.
The most surprising phenomena observed was that both for IPP and PNA for one
constant value of a (a.=3.5), rather narrow ranges for D were found. For
so
example, in the case of PNA the D range was (1.30 1.65)·10- 10 m2/s which,
so
compared to the ranges of D obtained from the HPMC model, bas a
app
significantly smaller spread.
The foregoing observations and the results and discussion presented in
Appendix 6.2 indicate that the HPMT model has a good predictive value and fits
reasonably well the different experiments for IPP and PNA single-component
adsorption systems. However, with regard to the model fits with the
experimental data, the same basic trend was observed as in the case of the
HPMC: a too low D for short adsorption times and a too high D for longer
app app
adsorption times. An additional point is that the model is based on empirica!
Chapter VI 142

considerations which makes it rather tricky for use in the modeling of


complicated adsorption systems.

6.2.5. MAXWELL-STEF AN MODEL WITH VARIABLE DIFFUSIVITY


(HPMS)

The fourth model is based on the Maxwell-Stefan defmitions of the flux


equations and includes also a variabie diffusion coefficient. This model is
referred to as the HPMS or simply MS-model. The D"PP used in this model is
defmed by Eq. (3.87):

(1-€ ) Q.
D"PP DP + pa p Ds __ (3.87)
€ pa c c
p

where DP and D 5 are constant.

This model was used for simulations for all three single-component systems
investigated in this study. Only 1-dîmensional optimizations were perforrned
5
(kf was fîxed) , the optimized parameter being the MS surface diffusivity, D •
To make the parallel between different models easîer, the results for the MS
model will be presented together with the diffusivities obtained from two of
the previously described models, the HPMC (Dapp const) and the HPMF (D s =
const, Fick). As mentioned earlier, the transition model (HPMT) is based on
rather empiri~al considerations and therefore it is not given in these tables.

The results for the IPP, PNA and NBZ single-component systems are listed in
Tables 6.11, 6.12, 6.13 (and in Table 6.3 which contains some additional
results for IPP obtained with the HPMC model).
Estimation of single-solute kinetic parameters 143

Table 6.11. Results of model simulations for IPP experiments

iji()6~. 9.18 0.14


.JÛ>.~~l 0.42 0.29
0~()769 5. 15 0.30
9u)o1a 10.1 0.10
5.20 0.15
56 98.11 2.10 0.57

Table 6.12. Results of model simulations for PNA experiments

tiJ}!. Çt? >m .. ti?~~~ ~9~/ q~(f!9~i alfa $.$~ gs(M§J (söS!èóim
I t[\) . ' t1Ê+~•·iè.f4 ~f~tt3 1 i 11.ét4 •1g-1-1~> ~1E~ i /
I r ) îro9llr ·tm911llrÓ2isf. · · · ·..·. . · · •·• •· • Ini2islt···.·. •. ·• ·•••••·······1 / i c tm~sf I i· ......·.··•······· t
111 29.49 98.75 4;!Jt o.95 r t;33 o.043 2.01 6~557 3.6o o.3o
20 30.34 48.64 2~44 3.4o rQ:9~. 0.070 3.56 '0=405 0.38 o.62
..

22 10.21 49.92 13.53 0.06 <4~17 0.070 0.56 0.621 0.17 0.20
23 20.08 50.72 14.23 1.90 1.62. 0.032 1.34 o;s:hf 1.70 0.40
37 29.88 50.60 •. 4/?'( 1.60 . . . . 1.69 0.032 1.44 ·...():??~ 1.6
~89 50.16 5~19 1.90 1.52 0.034 1.64 0~520. 1.1
41 9.80 100.9 44~1$ 0.12 ··.·1.41 0.003 1.80 . . 1.17() 0.12 0.10
42 9.92 74.64 41;94 0.17 7:29 0.003 1.43 1!148• 0.30 0.13
43 9~ 49.56 26.57 0.15 7.13 0.003 1 .1 1 0.~()~ 0.07 0.20
Chapter VI 144

Table 6.13. Results of model simulations for NBZ experiments

0.49
0.61
0.48
0.96
0.24
0.47
208 0.79
209 0.24
2.06
0.31
0.49

Analysis of the results in the tables shows that the diffusivity ranges for
8
D constant and D are large, in contrast to the Maxwell-Stefan range of
app F
diffusivities.
This is the case for all three components as can be seen from Table 6.14 in
which ranges of the diffusivities wîth various models are summarized tagether
with the relative standard deviations (RSD) of the diffusion coefficients and
the averaged values for the SOS in the model optimizations. From this table it
is clear that for all three solutes the RSD of the diffusivity is smallest if
the MS-model is applied. This implies that the MS diffusivity has the best
predictive value.

Table 6.14. Summary of model results for IPP, PNA and NBZ
cornp. . cönstDapp model
Range RSD SOS
: Fickian model
Range RSD SOS
MS-model
Range RSb SOS
*1E+9 .. *1E+4 *1E+13 *1E+4 *1E+13 *1E+4

IPP
PNA
0.2-0.9
2-44
48%
93%
5.0
1.1
0.02-0.3
0.9-7.5
66%
83%
5.4 0.06-0.00
1.7 0.4-1.0
Bêt
38%
5.0
1.0
NBZ 17-30 60% 3.1 0.2-10 46% 4.1 2.3-3.3 20% 0.8
Estimation of single-solute kinetic parameters 145

As it can be seen from the values for the SOS' s for the different models a
lower spread in the diffusivities does not necessarily result in a better fit
of the model: all three models provide nearly the same value for the averaged
SOS for IPP and PNA, while in the case of NBZ the MS-model gives a clear
improvement. At the same time, the fits between model and experiment for PNA
and NBZ are obviously better than those for IPP.
The averaged value of the SOS for the IPP experiments is higher than the
4
average for all sets of duplicate experiments SOS rep= 2·10 (cf. Sec. 5.1.4).
Therefore, it seems that even the MS-model does not fully describe the
kinetics for this solute. The same condusion emerges from the deviations
between model and experimental curves which all show the same systematic
deviation.
On the other hand, the SOS values of the PNA model fits are in most cases less
than the average SOS and within the SOS range (cf. Sec. 5.1.4), which
rep std
means that for this component the models describe reasonably well the
concentration decay curves.
0.050
b

1.2 ]
a
1
0 ~
0..... 1.0
~
z ~ -0.030
0 0.8
i= -Q.Q50
<(
a:1- 0 20 40 60 80 100 120

z 0.6 Tt..E ihoozsl


w
0
z
0 0.4
0

0.2
0 20 40 60 80 100 120

TIME (hoiJS)

- - exp + MS-model

Figure 6.8. a Dimensionless Experimental and MS-model curves


for PNA, exp. 38
b Residue graphof MS-model
Chapter VI 146

The largest differences observed between the SOS's, and consequently, between
the quality of the different model fits are m the case of the
single-component NBZ system. It is easy to see from the results presented in
Table 6.13 that for most experiments the Maxwell-Stefan model gives lower SOS
than the other two models. This is also obvious from the average SOS values
listed in Table 6.14.
An example of experimental and MS-model curves for component PNA is shown in
Fig. 6.8.

As already noted, the relative standard deviations gi ven in Table 6.12 show
that for all three components the ranges of MS diffusivities (D 5) are the
smallest on es of all models and consequently, it may be reasonable to assume
that the MS diffusion coefficient of each solute is practically constant.
Thus, if the model can describe the forms of the different experimental
concentration decay curves for a given component with one single value for the
diffusivity then it has a real predictive value.
In order to verify the predictiveness of the MS model the average value of D
s
was calculated for each component and was then used as input for all model
calculations for the given component. The results obtained included also the
SOS value for each experiment which was compared to the respective
SOS of the fit with the optimized D .
s

Table 6.15. The SOS from minimization runs and SOS with
an average value for MS diffusivity. IPP

(ingll)

50.0
350.0
51.44 0.0931 0.42 0.081 1.21
174.7 0.0760 5.15 0.081 5.64
525.15 0.1018 10.70 0.081 19.30
653.6 0.0785 5.20 0.081 9.67
172.00 0.0615 2.10 0.081 3.81
Estimation of single-solute kinetic parameters 147

Table 6.15 illustrates the calculation results for IPP. The data presented
shows that the SOS's of the curves calculated with an average D 8 are larger
4
than the average SOS which, as mentioned earlier, was 2-10- • This means ,
rep
that even the MS model cannot predict satisfactorily the overall rate of
adsorption of IPP.
Nevertheless, the MS-model, compared to the other models investigated in this
study, is the best one to predict the behavior of the single-component IPP
system and the value for Ds could be used, according to theory, for input in
multicomponent model calculations.

The results from MS-model calculations with an average Ds for component PNA
are presented in Table 6.16. The data shows that the SOS's from the model runs
5
with optimal D 's are for most cases below the average experimental SOS rep
8
value. However, the SOS values from calculations with the average D are
mainly above the SOS (although not for all experiments). This indicates
rep
that a better description of the COC' s can still be achieved and the model may
need further improvement.

Table 6.16. The SOS from minimization runs and SOS with
an average value for MS diffusivity, PNA

I Average
11·········•.• • .•.• .•. · Optlmizad
1
Êxp.· · •• êó ·•· · m I Qs(MS)
1

sos f?s(MS) SOS


.· .....•........ ~···
I>
.... > .•. •• ~1E+13 ~1E+4 ·. ~1.Ett~ *1E+4
I (lllg/1) (m9Jtl (m2/sJ I ..•. [m2/sJ I
f?• . .· · · •· ·•·•
17 29.49 98.75 0.557 3.60 ().73 18.0
....
20 30.34 48.64 0.405 ·..... 0.3 0.73 23.0
22 10.21 49.92 0.627 ·.. 0.17 ..... Q,l$ 1.0
23 20.08 50.72 0.576 1.70 ().73 11.0
~29.88 50.60 0.779 1.60 . 0.73 1.6
38 19.89 50.16 0.520 2.10 0.73 14.0
41 9.80 100.9 1.170 0.12 0.73 1.0
42
43
9.92
9.86
74.64
49.56
1.148
0.963
0.30
0.07
0.73
0.73 ·R
2.7
Chapter VI 148

A limited number of single-component PNA experiments were performed with with


a different carbon partiele size. The experimental conditions and the results
of the Maxwell-Stefan model fits to the experimental data are presenled in
Table 6.17.

Table 6.17. Results of MS model simulations with different partiele sizes


for component PNA

Rpa Kf Ds(MS) SOS


*1e+5 *1e+14 *1e+4
[mm] (m/s] [m2/s] [-]

0.41 7 7.26 2.73


0.41 10 6.71 2.08
0.41 11 6.61 2.04

0.26 7 6.56 2.89

0.26 9 6.96 2.45


0.26 11 5.64 2.49

0.16 7 4.59 2.24

0.16 9 4.07 2.34

0.16 11 3.83 2.43

As the value of the external mass transfer coefficient is directly related to


'
the partiele diameter (cf. p. 125), other (higher) values for k were used m
f
these model calculations.
The results in Table 6.17 show that for partiele radii of 0.41 and 0.26 mm the
Maxwell-Stefan diffusivity is close to the calculated average D value for PNA
s
(cf. Table 6.16). The values obtained for the partiele radius of 0.16 mm are
lower, however, they are within the range (lower limit) of optimal
diffusivities found for the standard used partiele size of 0.60 mm radius.
Although these results are based on a limited number of experiments they still
give an additional indication for the validity of the MS model for a
different type of system conditions.
Estimation of single-solute kinetic parameters 149

Table 6.18 The SOS from minimization runs and SOS with
an average value for MS diffusivity, NBZ

200 24.89 51.30


201 20.00 32.76
202
203
206
207
208
209
210
211
212

Table 6.18 presents the sum of squares of model runs for NBZ with optimal D 8
values and with an average D" value. As expected, the SOS's of runs with the
5
average D are larger than SOS's with optimal diffusivities. Por this
component the comparison with the SOS (2·10- 4 ) clearly indicates that the
rep
model can describe very well the NBZ concentration decay curve for different
experimental conditions. It also means that for NBZ the MS-model has a very
good predictive value and gives the best adsorption system description
compared with the other two components. The average value of D 8 may be used,
according to theory, for further input in multicomponent systems involving NBZ.

Table 6.19. Average values for surface MS-diffusivities for IPP, PNA, NBZ
.. .....·. <r IPP PNA NBZ
·Average
.•• . .. .
Ds(MS)·.
.
. ..
0.08 0.7 2.6
*1E+13
(1112/s)
Chapter VI /50

6.3. SOME ADDITIONAL REMARKS AND DISCUSSlONS

Diffusion behavior of components


Comparing the MS surface diffusivity values of the three components (cf. Table
6.19) shows that according to the MS-model, as it was for the other models,
NBZ exhibits the highest diffusivity, while IPP the lowest one. lt is
interesting to note that on the average the NBZ and IPP MS-diffusivities
differ by a factor more than 30. This means that NBZ is the fastest diffusing
component which can he seen also from the pore liquid concentration profiles
inside the particle, shown later in this section.
Component PNA is diffusing somewhat slower than NBZ, although the difference
in diffusivities (about 4 times) is not so pronounced as between NBZ and IPP.
At the same time, PNA is diffusing 10 times faster than IPP, according to the
MS-model.
As mentioned previously, the diffusivity values, obtained from the other
models, show somewhat different ratios for the different models. On one hand,
the general trend remains always the same - NBZ is the fastest diffusing
solute, then comes PNA and IPP is the slowest diffusing one. On the other
hand, for the HPMC model the D range of NBZ falls within the D range of
app app
PNA, while for the HPMF and HPMS models this is not the case, so in this
respect the different models give a difference.

This trend in diffusivity values is very interesting but not easy to explain.
One possible reason for the observed trend may he the size of the NBZ molecule
which is the smallest of all three components. Furthermore, although IPP and
PNA are quite similar in molecular weight, in view of the structures
presented in Appendix D, it becomes clear that the molecule of IPP differs
substantially from the other two types of molecules. As already noted, the two
methyl groups of IPP are perpendicular to the plane of the benzene ring, while
the molecules of PNA and NBZ are almost fully placed in the benzene ring
plane. It was previously shown (cf. chapter 4) that IPP has the biggest
molecule, hence, these differences, although not very big, may still play a
role in the diffusion process.

Another possible explanation for the low surface dîffusion coefficient


exhibited by IPP may he, as already noted, related to the so-called
Estimation of single-solute kinetic parameters 151

irreversible adsorption of phenolic compounds, reported by Yonge and Grant [4,


5]. As shown by Grant, polycondensation may occur and this may seriously
influence diffusion behavior.

Concentration dependenee of diffusivities


It was shown previously that both the Fickian and Maxwell-Stefan models used
in this study assume a variabie apparent diffusivity, while the actual surface
diffusion coefficients remain constant. Defmed in this way the two
expressions for D are directly comparable (cf. Eqs (3.87) and (3.12)), the
app *
only difference being in the way the ratio between Q and C is expressed.
p
However, on basis of the presented results , it becomes clear that the
assumption of a constant D for the Fickian model is not realistic, there are
s
significant variations in the values for the different experiments, and
consequently, the model has no real predictive value.
In contrast, the MS model gives a very narrow range of surface diffusivities,
which means that it can be used to predict the adsorption systems behavior in
different conditions. These results are in line with the theoretica! basis of
the two models, which was discussed in detail in chapter 3 and according to
which the MS diffusivity should be much less concentration dependent than the
corresponding Fickian diffusivity. This is well illustrated by the foregoing
results. For the Fickian model these results mean that it is necessary to look
for an additional expression for D which has to account for the strong
s
concentration dependenee of this parameter. One such expression was applied to
a limited amount of experiments, but there are also other expressions proposed
in literature (cf. chapter 3). The serious disadvantage of all these relations
is that they are empirica! and have no fundamental basis, as it is not known
what is the exact relation between the surface diffusivity and carbon toading
(or pore liquid concentration). Therefore the translation to multicomponent
diffusion coefficients will be very complicated - it is not possible to
predict how the diffusion coefficient will depend on the loading of a second
component.

In view of the foregoing, it would be interesting to analyze once more, now


for all models together, the influence of the initia} concentration and amount
of carbon on the values of the diffusion coefficients. It was shown earlier in
Chapter VI 152

this chapter that when applying the constant diffusivity model (HPMC),
experiments with higher C0 /m ratios and Q had a trend of decreasing D .
b oo app
A high CO/m
b
ratio and Q00 indicate that the average loading during the
adsorption experiment is higher than in the case of an experiment with a low
C:lm ratio and ~· It is possible that the (local) surface diffusivity
(Fickian or Maxwell-Stefan) depends on the local loading (or pore liquid
concentratioil).
If we now look at the diffusivity data for PNA (Table 6.12), we see a clear
0
trend of decreasing MS-diffusivity values with increasing C /m ratio, similar
b
to the trend observed for the constant diffusivity (HPMC) model.Although the
spread in the diffusivity values is quite different depending on the model the
observed trend for PNA remains interesting and shows that even the MS
diffusivity is not totally concentration (or loading) independent and may
require the inclusion of some kind of relation for certain adsorption systems.
The foregoing trend is also partially present in the case of NBZ (cf. Table
6.18) however no real conclusions can be drawn on basis of this data. For IPP
no such trend was observed for the diffusivity values of the Fickian or
Maxwell-Stefan models.

Deviations between and model curves


An interesting point for discussion is the systematic deviation between
experimental and model curves which is more or less present in all model
curves for all three components. Practically in all cases, initially, the
model curve runs too high, which is equivalent to the diffusivity in the model
being too low, while later on the curve is too low, equivalent to a model
diffusivity too high compared to the real situation as described by the
experimental curve.

There may be several reasons for this behavior.


Firstly, in the case of IPP this deviation may be due to the occurrence of
irreversibility, which involves additional physical and chemica! processes,
not accounted for in any of the morleis used in this study.
This, however, is less likely to be the case for PNA and NBZ.

Secondly, the real form of the carbon particles used for experiments is
cylindrical (extrudates), therefore not spherical as all models assume.
Estimation of single-solute kinetic parameters 153

Although the volume-to-surface mean diameter applied in the model calculations


gives a reasonable approximation of the actual form, this difference may
still be one of the reasons for the systematic deviations observed.

In the third place, a reason for the deviations may be the actual carbon
partiele inhomogeneity. One of the basic assumptions of all models applied was
that the carbon particles are homogeneous and isotropic, in the sense that the
pore size distribution is the same everywhere in the partiele and that there
isn 't a preferential directions of pores. Although this may be, to a certain
extent, a reasonable assumption for extruded types of carbons, it still is an
approximation of the reality and can easily be a souree of systematic
deviations between model and experiment.

In the fourth place, as already mentioned in previous chapters, in a series of


activated carbons produced by the same manufacturing process from the same raw
material but with varying residence times in the kiln, differences in pore
size distribution are observed. These differences have a gradual and
systematic character and are usually referred to as differences in degree of
activation. Although for all experiments performed in this study one type of
activated carbon was used, coming from one batch product, it is very likely
that different particles had different residence times in the kiln and
consequently, this may lead to certain deviations between model and
experimental curves.

Some concentration
It is interesting to compare the concentration profiles inside the partiele as
calculated by different models for the different components. Figures 6.9, 6.10
and 6.11 show several such concentration profiles calculated at 2 and 32 hours
for IPP, PNA and NBZ, respectively. Figure 6.12 gives the concentration
profiles from the same model calculations for 16 and 64 hours for PNA. The
profiles were calculated with the optimal diffusion coefficients for the
respective experiments and models, using the constant diffusivity (HPMC) model
and the Maxwell-Stefan (HPMS) model.
Figure 6.13 presents the carbon toading profiles for PNA, which have been
calculated via the equilibrium isotherm from the pore liquid concentration
profiles given in Fig. 6.10.
Chapter VI 154

1.0
HPM:S 2 h r s -
HPMC 2 hrs -···-
0.9
H~S 32 h r s -
HPMC 3 2 hrs -·-··
0.8

0.7
0
u
u 0.6

~
_;
!= 0.5
;;!
................/ '
~ 0.4

s 0.3

0.2
.............../ ...................
0.1

0.0
0.6 o. 7 0.8 0.9 1.0
RBLATIVB PARTICLE RADIUS

Figure 6.9. Pore liquid concentration profiles inside the partiele


calculated with optima! diffusivities, IPP ( exp. 52)

1.0
HPMS 2 hrs-
HPM: 2 hro ......
0.9
HPMS 32 hro -
HPMC 32 hro .....
0.8

0. 7
0
!2
u 0.6

~
. != 0.5

~ 0.4

s 0.3 _______ ... ----------··


-·-·

0.2 .•'

0.1

0.0
0.4 0.5 0.6 0.7 0.8 0.9 1.0
RBLATIVB PARTICLB RADIUS

Figure 6.10. Pore liquid concentration profiles inside the partiele


calculated with optima! diffusivities, PNA ( exp. 37)
Estimation of single-solute kinetic parameters 155

1.0 r---~-----r----~----r----,-----r----~----~--~~--~
HPMS 2 h n -
KPM::: 2 hre -·-··
0.9 HPMS 32 h r e -
H-= 32 hra ---·-
0.8

0.7 r:-----------
0.6

o.s

0. 4

0.3

0 2

0.1

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Q


RBLATIVE PARTICLE RADIUS

Figure 6.11. Pore liquid concentration profiles inside the partiele


calcu/ated with optima/ diffusivities, NBZ (exp. 210)

1 0
HAlS
HPIC
16 hr•-
16 hr• --···
0.9 HPMS 64 h r a -
HPIC 64 hre #···~
0.8

0.7
0
!:-:
~ 0.6

~
;:: 0 5
~
~ 0. 4

~
V 0 .J

0.2

0.1

o.s 0.6 0.7 0.8 0.9 l.O


RBLATIVE PARTICLE RADIUS

Figure 6.12. Pore liquid concentration profiles inside the partiele


calculated with optima[ diffusivities, PNA (at 16, 64 h.)
Chapter VI 156

1.0
HPMS 2 h r a -
0.9 HPM: 2 hre --···
HPMS 32 hre -
HPN: 32 hra ··--·
0.8

~
<
u
0.7

0.6
""'
< ä: 0.5

"~ 0.4
!i
0 0.3
..,<5
0.2

0.1

0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0


RBLATIVB PARTICLE RADIUS

Figure 6.13. Carbon toading (Q) profiles inside the partiele


calculated with optima/ diffusivities, PNA ( exp. 37)

lf we consider one set of pore liquid concentration profiles, for example in


Fig. 6.10, we can see that all profiles start at the outer partiele surface,
situated on the right-hand side of the graph. Following any of these profiles
shows that the C drops further to zero at a given point inside the
p,k
partiele, due to the internat resistances described by the apparent
diffusivity (constant or variable). If we now compare the profiles of any
model at 2, 16, 32 and 64 hours (cf. also Fig. 6.12), we can see that the
unloaded partiele zone has moved inwards and at all times there is an unloaded
part shrinking' in size.
The foregoing observations and figures illustrate the fact that the external
mass transfer and internat diffusion coefficients are the main parameters
descrihing the profile of a concentration decay curve.
Further, comparing the pore liquid concentration profiles of the three
components confirms that NBZ has the highest diffusion coefficient, while IPP
the lowest one, independent of which model was used.

If we look at the profiles of PNA for the two models, HPMC and HPMS, we can
see (cf. Figs. 6.10 and 6.12) that the profiles of the constant diffusivity
Estimation of single-solute kinetic parameters 157

model are almost linear, while for the MS-model initially they sharply deseend
and later flatten out. This illustrates the fact that the HPMC model uses a
constant diffusion coefficient, which is basically an average value of the
real variabie diffusivity, used by the MS-model.
Another comparison, this time between the pore liquid (C ) and pore wall (Q)
p
concentrations (cf. Figs. 6.10 and 6.13) shows that near the outer surface the
C profile is much steeper than the Q profile, while later on it becomes the
p
opposite. This is due to the highly non-linear form of the equilibrium
isotherm, which at higher C concentrations is rather flat and becomes very
p
steep at low concentrations.

6.4. CONCLUSIONS

Four different kinetic models have been applied to the single-component


systems under investigation.

All models account for the influence of the resistance to extemal mass
transfer. The extemal mass transfer coefficient for each component was
calculated from the initial part of the experimental decay curve. The spread
in the k1values for each component is small and the average values obtained
for the different components are rather close to each other and in accordance
with literature correlations.

- Results from modeling showed that diffusivity values obtained for components
NBZ and PNA are rather sensitive to variations in kr, while the diffusion
coefficient of IPP is not influenced, even by large k 1 variations.

The spread in the diffusivity values for the constant diffusion model
(HPMC) is too big for the model to have any real predictive value. Results for
all three components also show a rather clear tendency of decreasing apparent
diffusivities (where each D was obtained from a different experiment) with
app
increasing C~m ratio or carbon loading, Q. This indicates that the apparent
diffusivity is most likely concentration dependent.
Chapter VI 158

- The Fickian variabie diffusivity (HPMF model) showed a relative standard


deviation similar to the one for the constant diffusivity model, so it can be
concluded that for the systems used in this study the Fickian model has no
predictive value either.

- The Transition model (HPMT) was applied to components IPP and PNA and
produced a rather small spread in the surface diffusivity values for both of
them. However, as the model is based on empirica! considerations it cannot be
used in a consistent and systematic way to predict kinetic behavior of
different adsorption systems.

The Maxwell-Stefan (HPMS) model was applied for the first time to
adsorption from liquid systems. The results show a significantly smaller
spread in the MS diffusion coefficients compared to the Fickian ones. This
indicates that the MS-model with constant surface diffusivity bas a real
predielive value and consequently, according to theory, the average
diffusivities for each component could be used in multicomponent model
calculations.

A general trend in the diffusivity values of the three components under


investigation bas been found, where NBZ shows the highest diffusivity, while
IPP - the lowest one. This trend follows from the results of all four models
applied in this study. Although a number of plausible explanations have been
considered it remains difficult to predict which one is dosest to the
physical reality inside the particle.

- There is a systematic deviation between experimental and model curves which


is more or less present in all model curves. This might be due to a number of
factors, e.g. the assumption of a spherical particle, actual carbon partiele
inhomogeneity, differences in the degree of activation.
Chapter VIl 159

ESTIMATION OF RATE PARAMETERS FOR


TWO-SOLUTE SYSTEMS

7.1. ESTIMATION OF EXTERNAL MASS TRANSFER COEFFICIENT

The metbod used in this study for determining the extemal mass transfer
coefficients is based on the assumption that in the very beginning of the
process extemal mass transfer controls the rate of adsorption. This metbod
was already described in chapter 6.
Table 7.1 gives an example of values of the extemal mass transfer
coefficient for several two-eomponent NBZ/PNA experiments. The kf values for
the rest of the NBZ/PNA experiments, as well as those for the IPP/PNA system
are listed in Appendix 7.1.

Table 7.1. Calculated external mass transfer coefficients for several


two-component NBZ/PNA experiments

TC200 58.9 15.0


TC201 88.7 15.2
TC202 59.1 14.9
TC203 29.7 14.7
TC204 58.0 14.6 124.7
TC205 58.3 14.7 74.24

It is interesting to note that the kf values obtained for the different


components are rather constant, even for significant variations in system
conditions (cf. for example Table A.7.1.1). This indicates that both the
metbod used and the experiments performed have reasonable accuracy and
reproducibility.
Chapter VII 160

Due to the fact that only diluted solutions were used, one may expect the
extemal mass transfer coefficients for a given component in the single and
two-component systems to be equal. Analyzing the k values for NBZ and PNA
f
in Table 7.1 (and Table A.7.1.2) shows ranges of (5.2 - 6.2) and (5.2
7 .5)·10· 5 m/s, respectively, which are slightly lower than the ranges found
in the single-solute NBZ and PNA systems (cf. Appendix 6.1). On the other
hand, from the IPP/PNA data in Table A.7.1.1 it can be seen that k (PNA),
f
range (3.9 - 4.9)·10'5 m/s (except for exp. TC116), is elearly lower than in
the system NBZ/PNA. The values for IPP, range (2.7 3.6)·10- 5 m/s (except
for exp. TC116), are about 30% lower than the single-component values.

The foregoing results can be explained in terms of the limitations of the


metbod used for evaluating kt" The metbod is based on the assumption that
the concentration at the outer surface of the partiele is zero which implies
that when a solute molecule reaches the extemal core of the partiele it
will diffuse immediately inside. We assume this is true for the first 15
min. and use the experimental concentration decay curve points to calculate
kf. However, in two-solute systems, due to the presence of a second
component this mass transfer inside the partiele will be slower, which will
give a higher outer surface concentration at an earlier point in time. This,
in turn, leads to stronger deviations from the assumption of the outside
concentration being zero.
In view of the foregoing the single-solute kf values were used for the
two-component model calculations.

7.2. MODEL' CALCULATIONS FOR TWO-SOLUTE SYSTEMS

7.2.1. CONSTANT DIFFUSIVITY (HPMC) MODEL

This model was used for simulations both for the NBZIPNA and IPP/PNA
systems. Tables 7.2 and 7.3 give the variations in the starting
concentrations, amounts of carbon, constant D 's values and the total Sum
app
of Squares (SOS ) for each experiment. Note that this SOS is the
tot tot
average value of the individual SOS's of the two solutes for a given
experiment ( SOS tot = [SOS c ompl + SOS comp2 ] I 2).
Estimation of two-solute rate parameters 161

Table 7.4 summarizes the D and SOS ranges from the HPMC single- and
app
two-solute model runs, and gives also the all-experimental SOS
reproducibility range.

Table 7.2. Results of HPMC model runs for NBZ!PNA system


.· ..
Exp. Co(NBZ) Co(PNA) m Dapp(nbz) Dapp(pna) SQ§(tot)
...
.. . .· ~1E+9 *1E+9 *1E+4
; I· (rTig/1] (mgJIJ 1 (mgJIJ (lll~S) (m2/s] .......

TC200 58.9 15.0 172.8 5.51 .. 11.34 19.4


.
TC201 88.7 15.2 125.2 . 3.01 ; 4.07 11.5
TC202 59.1 14.9 124.6 I 6.85 14.02 12.2
'
TC204 58.0 14.6 124.7 ..•• 6.89 15,13 12.0
TC205 58.3 14.7 74.2 5.36 8.59 7.5
TC211 59.2 29.5 172.3 2~07 .. 7.63 15.4
TC212 30.3 29.7 172.8 6.52 8.91 17.3
TC213 89.0 29.6 173.7 2.24 6.73 11.1
TC215 59.2 14.8 173.4 ·. 5.15 5.15. 9.6
TC218 43.8 43.5 173.9 . 4.81 6;92 12.9

Table 7.3. Results of HPMC model runs for 1PP!PNA system

m .. Oapp(IPP) Df!pP{PNÄl SOS(tót)

F .... r .......: <~ ·.··


•::::.·. > i 1 JmgllJ ····:... Jrn911J (mg/IL :• (ïll2/s]·· ·· !o12lsl

TC101 61.32 14.98 124.90 Ö.16;!... 0.836< 3.29


TC102 60.97 15.07 125.12 0;146 o;s2a 5.2o
k .....
TC104 61.78 15.20 75.16 0.159 /().1$4 > 2.38
TC105 62.00 10.04 125.00 I 0.199> .. 0,948 4.61
TC106 61.42 15.05 175.52 0.208 ·.·.·. 1.075 4.34
TC107 61.37 20.03 125.45 0~201 3.31
TC111 92.48 15.11 125.40 1.98
TC115 49.27 24.13 125.00 0.204 I. 0.744 2.24
TC116 26.20 24.65 125.00 0.166 I f~09~ 7.74
Chapter VI/ 162

The valnes for D (NBZ) and D (PNA) given in Table 7.2 show a relative
app app
standard deviation of 38 and 41%, respectively. For neither of the
components can a clear trend be detected between D app and initial
concentration and/or carbon dose.

Table 7.4 Single- and two-solute ranges for D and SOS, HPMC model
app

[)app~1E+9 7-30 0.15-0.21 0.5-1.1 2.1-6.7 5.2-15.1


SOS*1E+4 0.6-8.1 (tot) 2.0-7.7 (tot) 7.5-19.4
SOS(rep) 0.06-6.99

Camparing the diffusivity valnes from the NBZ/PNA system with the respective
single-component D valnes (cf. Table 7 .4) shows that for NBZ the
app
two-solnte valnes are lower, while for PNA they fall within the
single-solute range. As the HPMC model is rather simplified, it is difficult
to draw any major conclusions about the possible interactions between the
two components.
From the SOS range for NBZ/PNA and the SOS given in Table 7.4 it
tot rep
becomes clear that the SOS values are an order of magnitude higher than
tot
the respective SOS ones. These resnlts indicate that for the NBZ/PNA
rep
system the HPMC model does not give a good description of the overall rate
of adsorption.

The IPP/PNA data presented in Table 7.3 show a relative standard deviation
of 13 and 20% for D (IPP) and D (PNA) , respectively. There is no clear
app app
trend between the initial concentration and/or amonnt of carbon for the Dapp
of component IPP; the valne of D for PNA seems highest at low values of
app
CJm and lowest at the highest total or IPP loading. This tendency is the
same as for D in the single-component systems.
app

Table 7.4 indicates that the SOS range for the IPP/PNA system is slightly
tot
higher than the SOS reprodncibility range. An example of experimental and
model concentration decay curves for IPP/PNA is given in Fig. la/b. From
Bsdmation of two-solute rate parameters 163

0,05

b
~
a
1.2 r - - - - - - -
! -Q/:)1

0
~ -o.o3
0
Q 1.0 -0.05
0 10 20 30 4() 50 60 70 80
z TllvE O'lou"sl
0
i=
<( 0.8
a:
1-
z
~
z 0.6
0
0

0.4 L--~---L--~--~-~L--~---L-~

0 10 20 30 40 50 60 70 80

TIME (hoLrS)

o exiPP + exPNA mod - - mod

Figure 7.1 a Experimental and HPMC model curvesjor 1PPIPNA system, exp 101
b Residue graphs for 1PP and PNA model curves

residue graphs in Fig 1b it appears that for this experiment the IPP
experimental curve is slightly better described by the model than the PNA
one. Results also showed that for most experiments in Table 7.3 the SOS for
PNA is bigger than the one for IPP (in Table 7.3 only the total SOS is
given).
Comparing the values for D from the two-solute IPP/PNA system with those
app
from the single-solute ones (cf. Table 7.4) shows that the two-solute values
are significantly lower, about twice for IPP and over tenfold for PNA. It
should be also noted that the D of IPP in the two-solute systems is of
app
order of the pore liquid diffusivity, which could mean that for IPP it the
surface diffusivity plays only a minor role.
The foregoing results indicate that although the HPMC model is far from
sophisticated and cannot be based on single-solute diffusivity data, due to
relatively small spread in the two-solute diffusivities it may be used for
Chapter VII 164

practical purposes to predict kinetic behavior for the IPP/PNA system and to
a lesser extent for NBZ/PNA. As the values of D clearly deviate from the
app
single-component values two-component experiments will be needed to make
such predictions.

7.2.2. MAXWELL-STEFAN (HPMS) MODEL WITHOUT


CROSS-DIFFUSION COEFFICIENTS

The Maxwell-Stefan model without cross-diffusion coefficients was used for


calculations for both two-component systems. The optimized diffusivities and
corresponding SOS's for the NBZ/PNA and IPP/PNA systems are listed in Tables
7.6 and 7.7. The tables give also the starting concentrations and carbon
concentrations for the experiments, as well as the total SOS's. For all
model calculations performed the average single solute values of DMS values
s
were used as starting optimization parameters.

Table 7.5 contains the DMS and SOS ranges from the HPMS single- and
s
two-solute model runs, and the same table gives also the all-experimental
SOS reproducibility range.

Table 7.5 Single- and two-solute ranges for DMS and SOS, HPMS model
s

Range of IPP/PNA sys.


NBZ IPP PNA

Ds*1E+14 4-10 23-33 0.5-0.7 1.0-7.8 5-16 6-11


SOS*1E+4 0.1-1.7 0.2-2.1 (tot) 2.2-8.3 (tot) 1-17
SOS(rep) 0.06-7.0

The NBZ Maxwell-Stefan surface diffusivity values in Table 7.6 (next page)
have a relative standard deviation of 39%. This spread of DMs is not too
s
large, however, the range is different from the single-solute HPMS
diffusivity range for NBZ (cf. Table 7.5).
Estimation of two-solute rate parameters 165

Table 7.6. Results of HPMS model runs for NBZ!PNA system


Exp. Co{NBZ) Co(PNA) m Ds(MS) Ds(MS) SOS(tot)
NBZ PNA
~1E+13 ~1E+13 ~1E+4

[mg/1] [mg/1) (mg/1] [m2/s] (m2/s]

TC200 58.9 15.0 172.8 1.06 0.77 11.6


TC201 88.7 15.2 125.2 1.35 0.92 3.7
TC202 59.1 14.9 124.6 1.55 1.08 3.9
TC203 29.7 14.7 124.3 1.08 0.89 4.2
TC204 58.0 14.6 124.7 1.51 1.06 3.9
TC205 58.3 14.7, 74.2 1.51 1.13 1.2
TC207 58.9 59.7 171.2 0.64 0.74 5.1
TC211 59.2 29.5 172.3 0.63 0.63 17.1
TC212 30.3 29.7 172.8 0.58 0.69 14.4
TC213 89.0 29.6 173.7 1.01 0.89 5.4
TC215 59.2 14.8 173.4 1.27 0.85 10.9
TC218 43.8 43.5 173.9 0.47 0.62 12.0

Table 7.6 shows quite different results for the other component in this
system, PNA. The DMs for PNA varies within (0.6 - 1.1)·10- 13 m2/s and the
s
RSD (21 %) is significantly lower than the one for NBZ. It is interesting to
note that the average Maxwell-Stefan diffusivity for PNA from the
single-solute HPMS model is 0.7·10- 13 m 2 /s and most of the PNA values in
Table 7.6 are very close to this value. This is also confirmed by the
single- and two-solute DMs ranges given in Table 7.5. This indicates that in
s
the case of PNA the single-component MS-diffusion coefficient can be used to
predict the kinetic behavior of the solute in a multicomponent mixture.
The SOS values in Table 7.6 and the SOS and SOS ranges in Table 7.5
tot tot rep
indicate that for some NBZ/PNA experiments the quality of the model fit is
not good enough. Figure 7.2 presents an example of HPMS model and
experimental concentration decay curves for NBZ/PNA. The residue plot shows
in the beginning a somewhat better description of the experimental curve by
NBZ for this experiment, which proved to be the case for most of the other
experiments as well (in Table 7.6 only the total SOS is given).
Chapter VIl 166

0.05
b
0.03
~
0.01
a ! -0.01
1.2
! -o.o3
0
0....... -o.05
Q 1.0 0 20 40 60 80

z TINE O'lou"sl
0
i=
<{ 0.8
a:
1-
z
w
0
z 0.6
0
0

0.4
0 20 40 60 80

TIME (hours)
"'V exNBZ mod + exPNA mod

Figure 7.2 a Experimental and HPMC model curvesJor NBZ!PNA system


b Residue graphs for NBZ and PNA model curves, exp 201

Table 7.7 contains the experimental data and model calculations for the
IPP/PNA system. The range of Maxwell-Stefan surface diffusivities for IPP
has a relative standard deviation of 12%. This range corresponds rather well
to the single-component IPP diffusivity range (cf. Table 7.5), although the
8
DM 's from the two-solute calculations are slightly lower.
s

The range of DMs values for PNA presented in Table 7.7 has a relative
s
standard deviation of 53%, which is rather large. However, as Table 7.5
shows, the two-solute PNA range of DMs covers most of the single-solute
s
range and the predicted two-component MS-diffusivities are slightly lower
than the single-component ones.
Estimation of two-solute rate parameters 167

Table 7.7. Results of HPMS model runs for IPPIPNA system

EXP· Ço(IPP) Co(PNA) m I)S{MS) O$(M$J


.......
I
: .. IPP h· PNA ...•....
.......... ...
....
· .. ·.· ... I ·.. *1.E-t14 *lE+14 *lE+4
[mg/1) .... lm911J !m9fl1·.··. [m2Js} ~ Jrn2lsl ..

TC102 60.97 15.07 125.15 0.66 1.33 6.39


TC104 61.78 15.20 75.16 OAS 4.73 6.12
TC106 61.42 15.05 175.52 0.62 ·. I 7.78 4.74
TC107 61.37 [20.03 125.45 OJ72 5.62 3.51
3.65
~111
92.48 15.11 125.40 0.56 0"98
115 49.27 24.13 125.00 0.57 4.29 2.21
116 26.20 24.65 125.00 0,60 6.97 8.34

The SOS range for the IPP/PNA system is compared to the all-experimental
lol
SOS in Table 7.5. Although the SOS values are somewhat higher, the
rep lot
range is clearly smaller than for the NBZ/PNA system.

There are several points conceming the Maxwell-Stefan model calculations


which require some further attention.

First, results indicate that the ranges obtained for PNA are practically
identical (NBZ/PNA system) or very close (IPP/PNA system) to the MS
single-solute diffusivity range. Secondly, the DMS range for IPP in the
s
IPP/PNA system also corresponds rather well to the single-component IPP
diffusivities. Thirdly, in the case of NBZ, the spread in diffusivity values
is not high, but the values as a whole are 2.5 - 3 times lower than the
single-solute NBZ diffusion coefficients.
Conceming the SOS values, many experiments are within the SOS range,
rep
however, others show much higher values. It is also clear that the
systematic deviation between model and experimental curves is still present
(as in the single-solute fits).
The foregoing observations indicate that in the cases of components IPP and
PNA the results correspond reasonably well to the theoretica! bases of the
Chapter VI/ 168

Maxwell-Stefan equations, as outlined in chapter 3. This is not the case for


NBZ, where the two~solute diffusivities are much lower than predicted by the
single-solute simulations.

There could be several reasons for the observed behavior.


First, it should be reminded that this is the frrst time the MS equations,
based on the chemica! potential as driving force, are applied to adsorption
from liquid phase and it appears that the two-component MS theory may need
to be developed further.

Secondly, it can be expected that when calculating two-component model


curves the spread in the optimized values and the variations in the SOS's
will be larger due to the fact that deviations in one curve cause also
deviations in the other one (via equilibrium).

Thirdly, it is interesting to observe that in both systems there is a faster


and a slower diffusing component, as predicted by the diffusivity values
from the single-solute model calculations (some NBZ!PNA experiments make an
exception). If we consider the system IPP/PNA, for example, it may be
speculated that after both components enter initially in the outer shells of
the adsorbent particle, later, the faster diffusing solute (PNA) has to
"pass through" the already adsorbed IPP in order to penetrate fmther inside
the particle. the same mechanism may be assumed for the other system (the
faster NBZ "passes through" the slower PNA).
The above observations may indicate that, contrary to the assumptions of the
HPMS model used in this chapter, there are "friction" and "drag" forces
between the ~omponents in a two-component system and consequently, the
cross-diffusion coefficients in the MS model equations cannot be neglected.

7.3. CONCLUSIONS AND RECOMMENDATIONS

Two different models have been applied to the two-component systems under
investigation.

The multicomponent constant D model (HPMC) fits reasonably well the


app
Estimation of two-solute rate parameters 169

experimental IPP/PNA results and to a lesser extent the NBZ/PNA ones. As it


is oversimplifying the physical reality, this model has no real theoretica!
value, but can be used for predicting kinetic behavior in practical
situations. Because the values of D clearly deviate from the
app
single-component values two-component experiments will be needed to make
such predictions.

The multicomponent Maxwell-Stefan model (HPMS) without cross-diffusion


coefficients was applied for the first time to adsorption from liquid
solutions. The results for the two systems investigated differ slightly, but
the model gives a fair prediction of the experimental concentration decay
curves and gives ranges for the DMS values close to the single-solute ones

for components IPP and PNA. The range for DMS of NBZ, although not too
s
large, is quite different from the single solute NBZ diffusivity range.
This model can be used for gaining further theoretica! insight into the
processes taking place within the adsorbent particle, as well as for more
qualitative and practical estimations of the overall rate of adsorption.

Although the results look promising, in order to prove the applicability of


the HPMS model, further experimental data and model calculations are
necessary, invalving other solutes and system conditions. It may be also
useful to perform model calculations with the HPMS model with
cross-diffusion coefficients to try to imprave the results and obtain more
information about the mechanisms of adsorption.
Chapter 8 170

EPILOGUE

8.1. GENERAL CONCLUSIONS

A new kinetic model is proposed for the description of adsorption from liquid
phase based on the Maxwell-Stefan formulations in which the gradient of the
chemical potential is taken as the driving force for diffusion.

Model calculations for the single-solute p-isopropylphenol, p-nitroaniline and


nitrobenzene systems showed that the Maxwell-Stefan description is clearly
better than the Fickian one and that the model has a real predielive value.

Optimization runs with the multisolute Maxwell-Stefan model without


cross-diffusion coefficient showed that the two-solute diffusivity value,
compared to the single-solute one, was somewhat lower for nitrobenzene and
equal for p-nitroaniline and p-isopropylphenol. This model can be used to
predict fairly well multisolute kinetics based on single-solute diffusivity
data.

To describe two-solute equilibrium a new, extended form of the Fritz &


Schlünder equation is proposed which uses parameters from the single-solute
Radke-Prausnitz isotherm and additional 3 parameters per component, which have
to be fitted to two-solute equilibrium data. The proposed equation reduces to
a linear isotherm when the concentrations of both solutes are very low. It
fitted well to the experimental equilibrium data and was used for calculating
local equilibri~m in the kinetic models.

8.2. RECOMMENDATIONS FOR FUTURE WORK

Although the results from the multisolute Maxwell-Stefan model look promising,
in order to prove the applicability of the model, further experimental data
and model calculations are necessary, in volving other solutes and different
system conditions.
Chapter 8 171

Another possibility is to perfonn model calculations with a model with


cross-diffusion coefficients to try to improve the results and obtain more
infonnation about the mechanisms of adsorption.

The models presented in this thesis are all based on the assumption of a
homogeneons adsorbent partiele with respect to pore structure and pore size
distribution. An alternative approach could be to develop and solve a model
assuming a biporous structure of the adsorbent particle, which model will be
more close to the physical reality.

Another area of interest is modeling of fixed-bed adsorption systems which are


widely used in water treatment. It would be very intriguing to see if the
diffusivity data obtained from the batch kinetic Maxwell-Stefan models can be
used to predict intra-partiele mass transport in the case of column systems.
Appendix 4.1 172

Experimental equilibrium data for single-solute PNA and IPP systems

Table A.4.1.1. Equilibrium data for single-solute PNA system

Exp V m Co Cp(fin) Q
I·····
I [IJ ' [mg) [mg/1) (mg/1) [mg/mg}

1 0.2 9.8 98.61 80.95 0.361


2 0.2 20.2 98.61 63.51 0.347
3 0.2 40.2 98.61 31.29 0.335
4 0.2 40.4 98.61 31.21 0.334
5 0.2 60.3 98.61 10.97 0.291
6 0.2 81.1 98.61 2.36 0.237
7 0.2 100.1 98.61 1.06 0.195
8 0.2 100.3 98.61 1.06 0.194
9 0.2 119.7 100.14 0.57 0.166
10 0.2 139.9 100.14 0.27 0.143
11 0.2 149.6 100.14 0.44 0.133
12 0.2 150.3 100.14 0.33 0.133
13 0.2 159.6 100.14 0.47 0.125
14 0.2 180.7 100.14 0.34 0.111
15 0.2 200.5 100.14 0.41 0.099
16 0.2 49.8 100.14 106.51 0.369
17 0.2 100.3 100.14 31.35 0.333
18 0.2 151.1 100.14 5.05 0.256
19 0.2 199.2 100.14 1.19 0.198
20 0.2 200.1 100.14 1.23 0.197
21 0.2 250.2 100.14 0.76 0.158
22 0.2 300.6 100.14 0.59 0.132
Appendix 4.1 173

Table A.4.12. Equilibrium data for single-solute IPP system

149.1 100.06 0.59 0.134


118.7 99.08 1.19
95.5 99.08 1.91 0.204
200.22 2.48 0.196
200.22 2.98 0.22
200.22 4.46 0.245
79.4 100.06, 6.05 0.237
8 0.2 59.8 100.06 21.89 0.261
9 0.2 50.4 100.06 32.29 0.27
10 0.2 39.9 100.06 44.36 0.279
11 0.2 104.3 200.22 51.96 0.284
12 0.2 20.1 100.06 71.23
13 0.2 9.7 100.06 85.85
.5 200.22 104.65 0.311
200.22 131.29 0.308
Appendix 4.2 174

Experimental equilibrium data for two-solute systems

Table A.4.2.1. Equilibrium data and error estimate for IPPIPNA system

SCo/Co Cp
(%) [o/o) (nig/1)

53.3 103.5 0.071 12.39 0.29 0.1415 1.0


2 59.9 103.5 .09 0.071 9.989 0.32 0.1339 1.0
3 75.8 103.5 0.015 22.93 .09 0.071 5.492 0.53 0.1177 1.0
4 28.2 49.93 0.03 21.82 0.14 6.444 0.45 0.131 1.0
5 36 49.93 0.03 14.18 24.92 0.14 3.704 0.82 0.1179 1.0
6 36.7 104.0 0.015 57.07 25.26 0.12 7.153
7 46.67 25.26 0.12 5.09
25.26 0.12 1.333
12.48 0.28 2.011
12.48 0.28 1.471 2.27 0.0701 1.
12.48 0.28 3.619 0.84 0.0895 1.1
12 14.6 50.41 12.48 0.28 5.196 0.56 0.0998 1.1
13 21.4 50.41 0.03 25.62 0.28 3.291 0.94 0.0859 1.1
14 10.3 100.0 0.015 85.44 0.26 7.731 0.38 0.0971 1.3
15 20.1 100.0 0.015 73.04 0.26 4.948 0.59 0.0775 1.1
16 30.1 100.0 0.015 60.19 0.07 0.2646 1.01 12.73 0.26
17 40.2 100.0 0.015 45.26 0.1 0.2724 1.01 12.73 0.26
18 50 100.0 0.015 32.78 0.14 0.269 1.01 12.73 0.26
19 62.2 100.0 0.015 19.02 0.08 0.2605 1.01 12.73 0.26
20 49 12.5 1.5 0.099 38.35 0.0511 1.84 49.23 0.052 1.481 1.82 0.2009 1.0
21 42.2 12.5 1.5 0.361 10.48 0.0593 1.87 49.23 0.052 3.085 0.84 0.2332 1.0
12:5 1.5 49.23 0.052 5.024 0.51 0.2557 1.0
80 24.67 0.73 96.32 0.03 5.484 0.47 0.2407 1.0
76 24.67 0.73 96.32 0.03 6.506 0.4 0.2534 1.0
73 24.67 0.73 96.32 0.03 8.842 0.32 0.2638 1.0
69 24.67 0.73 96.32 0.03 9.761 0
67 24.67 0.73 96.32 0.03 12
Appendix 4.2 175

Table A.4.2.2. Equilibrium data and error estimate for NBZ/PNA system
·.· ..·
.·.. N~;~z·· P~A·······•·\·.········••·············••
··•··. . ............ \
.o· .·. sQ(ä
. ···
Eic:. m Co SCo/Co Cp SCp/Cp a SQ/Q Gei SCci/C.o Cp $dPJc~
...
(mg] [mg/1] [%) [mgll] (o/o] (mg/mg] [%] (mg/1] [%] [mg/1]
..··.1%1······· (mg/mg] ]%]·

1 102.6 100.1 0.24 3.784 0.77 0.1879 1.04 25.29 0.12 0.290 2.69 0.0487 1.01
2 78.1 100.1 0.24 10.25 2.31 0.2303 1.07 25.29 0.12 1.08 0.66 0.062 1.01
3 68.5 100.1 0.24 15.92 1.51 0.246 1.08 25.29 0.12 1.914 0.37 0.0683 1.01
4 63.1 100.1 0.24 19.9 1.24 0.2545 1.09 25.29 0.12 2.595 1.22 0.0719 1.02
5 55.5 100.1 0.24 28.29 0.94 0.2591 1.12 25.29 0.12 4.084 0.73 0.0764 1.02
6 50.9 100.1 0.24 31.47 0.76 0.27 1.12 25.29 0.12 4.898 0.6 0.0801 1.03
7 41.5 100.1 0.24 42.6 0.56 0.2775 1.17 25.29 0.12 7.436 0.39 0.0861 1.03
8 36.3 100.1 0.24 48.89 0.49 0.2826 1.2 25.29 0.12 8.991 0.34 0.0898 1.04
9 106.2 225.6 0.11 75.23 '0.32 0.2832 1.03 75.65 0.091 18.26 0.17 0.1081 1.01
10 100.5 225.6 0.11 81.42 0.3 0.287 1.03 75.65 0.091 20.64 0.15 0.1095 1.01
11 93.7 225.6 0.11 89.31 0.27 0.2909 1.04 75.65 0.091 23.2 0.15 0.112 1.02
12 88.3 225.6 0.11 97.18 0.25 0.2909 1.04 75.65 0.091 25.59 0.14 0.1134 1.02
13 86.6 48.22 0.5 1.867 1.28 0.107 1.13 49.7 0.09 0.603 1.24 0.1134 1.01
14 69.1 48.22 0.5 4.863 4.88 0.1255 1.27 49.7 0.09 1.923 0.37 0.1383 1.01
15 55.6 48.22 0.5 9.686 2.44 0.1386 1.33 49.7 0.09 4.834 0.61 0.1614 1.01
16 47.1 48.22 0.5 14.75 1.63 0.1421 1.43 49.7 0.09 8.719 0.34 0.174 1.01
17 38.5 48.22 0.5 20 1.18 0.1466 1.56 49.7 0.09 13.41 0.22 0.1886 1.02
18 143.8 99.19 0.24 1.654 1.44 0.1357 1.04 50.82 0.09 0.224 3.52 0.0704 1.01
19 112.4 99.19 0.24 4.473 5.31 0.1685 1.07 50.82 0.09 0.816 0.89 0.089 1.01
20 98.8 99.19 0.24 7.657 3.09 0.1853 1.07 50.82 0.09 1.609 2.06 0.0996 1.01
21 47.8 99.19 0.24 43.51 0.55 0.233 1.18 50.82 0.09 16.57 0.18 0.1433 1.02
22 97.2 99.27 0.24 8.309 2.85 0.1872 1.07 51.8 0.09 1.867 0.38 0.1027 1.01
23 75.9 99.27 0.24 18.77 1.3 0.2121 1.09 51.8 0.09 5.419 0.54 0.1222 1.01
24 64.5 99.27 0.24 27.44 0.86 0.2228 1.11 51.8 0.09 9.28 0.33 0.1318 1.01
25 57 99.27 0.24 34.46 0.69 0.2274 1.13 51.8 0.09 12.58 0.28 0.1376 1.02
26 50.2 99.27 0.24 40.56 0.59 0.2339 1.16 51.8 0.09 15.31 0.19 0.1454 1.02
27 41 99.27 0.24 50.63 0.48 0.2373 1.23 51.8 0.09 21.01 0.15 0.1502 1.02
28 60.4 150.1 0.16 73.87 0.33 0.2525 1.1 77.86 0.092 31.51 0.1 0.1535 1.02
29 54.3 150.1 0.16 80.97 0.3 0.2547 1.12 77.86 0.092 35.02 0.1 0.1578 1.02
30 49.4 150.1 0.16 87.19 0.27 0.2548 1.14 77.86 0.092 38.48 0.09 0.1594 1.03
Appendix 5.1 176

Experimental conditions for kinetic single-solute systems

Table A.5.1.1. Experimental conditions for single-solute 1PP experiments

E~p. Vessel Co m Rpa Time


[mg/1) [mg/1) [mm) [hrs)

1 1 24.16 168.83 0.60 19.5


2 1 49.05 176.82 0.60 69.0
3 1 24.16 232.8 0.60 109.5
4 1 49.92 351.4 0.60 41.5
5 1 49.03 349.9 0.60 95.0
6 1 49.14 157.5 0.60 109.5
7 1 50.00 167.5 0.26 91.5
8 1 48.99 350.4 0.26 70.5
9 1 97.28 175.2 0.60 109.5
24 1 14.74 51.9 0.60 68.5
25 1 14.94 51.4 0.60 68.5
26 1 14.81 51.5 0.60 71.0
27 1 14.78 50.0 0.60 71.0
28 1 48.44 175.4 0.60 68.5
29 1 48.30 350.0 0.60 69.0
30 1 15.28 49.9 0.60 29.0
31 2 14.82 51.4 0.60 29.0
32 3 15.21 51.8 0.60 29.0
52 1 51.85 174.7 0.60 65.5
53 3 51.26 524.9 0.60 65.5
54 1 99.39 653.6 0.60 62.0
55 2 99.17 526.1 0.60 62.0
56 1 98.11 172.1 0.60 64.5
Appendix 5.1 177

Table A.5.1.2. Experimental conditions for single-solute PNA experiments

Exp. Vessel Co m Rpa Time


[mg/1] (mg/1) [mm) (hrs]

17 1 29.49 98.81 0.60 64.5


18 2 29.71 152.22 0.60 64.5
19 1 29.03 100.6 0.60 87.0
20 2 30.34 48.6 0.60 87.0
21 3 30.41 24.8 0.60 87.0
22 2 10.21 49.9 0.60 50.0
23 2 20.08 50.7 0.60 50.0
33 1 29.44 40.1 0.41 65.5
34 2 28.87 50.1 0.16 65.5
35 1 29.n 50.2 0.66 69.0
36 3 30.05 50.0 0.60 69.0
37 1 29.88 50.5 0.60 109.5
38 2 19.89 50.2 0.60 109.5
39 1 29.32 50.1 0.65 64.5
40 1 29.74 50.3 0.60 64.5
41 1 9.80 100.9 0.60 51.0
42 2 9.92 74.6 0.60 65.0
43 2 9.86 49.6 0.60 65.0
44 1 29.22 50.1 0.26 35.0
45 3 29.48 50.5 0.41 68.0
46 2 31.17 49.9 0.16 15.0
47 1 28.03 50.8 0.16 56.0
48 2 28.27 50.8 0.26 56.0
49 1 27.96 50.7 0.65 66.0
50 3 28.27 50.4 0.55 66.0
51 2 9.95 50.0 0.60 68.0
Appendu 5.1 178

Table A.5.1.3. Experimental conditions for single-solute NBZ experiments

200 24.89 51.30 0.60 64.0


201 2 20.00 32.76 0.60 64.0
202 24.50 51.40 0.60 63.0
203 2 24. 0.60 63.0
204
205 2 12.05 24.50
206 59.44
207 59.20
208 2 59.78
209 30.04
210 2 50.34
211 23.54 75.00 60.5
212 2 12.06 24.44 60.5
Appendix 5.2 179

Experimental conditions for kinetic two-solute systems

Table A.5.2.1. Experimental conditions for two-solute IPP!PNA experiments

Exp.

TC101 61.32 14.98 124.90 0.60 69.5


TC102 2 60.97 15.07 125.12 0.60 69.5
TC104 2 61.78 15.20 75.16 0.60 66.0
TC105 62.00 10.04 125.00 0.60 69.0
06 2 61.42 15.05 175.52 0.60 69.0
61.37 20.03 125.45 0.60 68.5
92.48 15.11 65.5
49.27 24.13 50.0
26.20 24.65 68.0
Appendix 5.2 180

Table A.S.2.2. Experimental conditions for two-solute NBZ!PNA experiments

~p..· i ."' Çit(~~Z) ··çö<e~Al I (w.<·. ume


.......... '•'ii•·· ./
•"'~··
••.... ?) i }> . <. < i (Î'Ii9/ll. >• .•. • <[rl19Ï!L< . @Q/1] • · ·•· •·i•••· · [nu:ri]
f:tPi\········· [hh>l''

Të2oOT 2 58.9 15.0 172.8 0.60 71.5


TC201 3 88.7 15.2 125.2 0.60 71.5
TC202 2 I s9.1 14.9 124.6 0.60 50.0
TC203 3 29.7 14.7 124.3 0.60 50.0
TC204 1
58.0 14.6 124.7 1 o.6o 71.0
TC205
TC206
2
1
58.3
58.9
=p,g 74.24
213.6
0.60
0.60
71.0
99.0
TC207 1 1 58.9 59.7 171.2 0.60 99.0
TC210 2 58.9 14.9 172.5 0.60 109.5
TC211 3 59.2 29.5 172.3 0.60 1109.5
TC212 3 30.3 :::: 29.7 172.8 0.60 109.5
TC213 2 89.0 29.6 173.7 0.60 72.5
TC214 3 44.7 44.3 171.7 0.60 72.5
TC215 2 0.60 84.0


59.2

~
TC217
3
3
43.6
58.1
.9
.2
0.60= 84.0
0.60 65.5
TC218 2 43.8 .9 0.60 75.0
Appendix 5.3 181

Some important experimental errors and reproducibility measurements

Zero-point error
One important factor, which may contribute substantially to the total
experimental error is the varlation in the zero-point of the flow-thru cells
during a kinetic experiment. In the course of an experiment which typically
lasts 2-3 days, the zero-point can change sigificantly, due to varlous
impurities accumulating gradually on the walls of the flow-thru cell. For this
reason the zero point values of the cell were always measured twice, before
and after an experiment. Assuming that the values obtained are the lowest and
highest ones for the given run and that the change in the zero point in time
is linear, the absolute error introduced by this change can be estimated using
Eq. (A.5.3.1):

dC z.p. (A.5.3.1)

where:
dCz.p. : absolute error in the concentration measurement due to the
1
: change in the zero point [mg·r ]
1
k1 : calibration constant [mg·r ]

A (C=O) : average zero point during experiment [-]


A =0(C=O): zero point at time t=O [-]
1

The varlation that this error brings in the concentration decay curve can be
estimated with the sum of squares as defined in the following equation:

dC
sosz.p.- [ c~·P·] (A.5.3.2)
b

where:
SOS :varlation in the concentration decay curve due to the
z.p.
:change in the zero point [-]
0
C :initial concentration [mg·r']
b
Appendix 5.3 182

This way of defming the SOS makes it possible to compare its values with
z.p.
the SOS values calculated from the difference between experimental and model
curves (cf. Eq. (5.6)).

As the zero point at the beginning of the experiment is known the error in the
initia! concentration can be deterrnined in the following way:

dC
0
b
=k 1
(A-(C=O) - A
t=O
(C=O) ) (A.5.3.3)

where:
0
·dC
• b :error in initial concentration due to an error in the average
1
:zero point [mgT ]

Table A.5.3.1. Variations in the concentration decay curve and errors


in the initia/ concentration due to a varying zero-
point, NBZ system

Exp. i
Co l dCo, z.p. "c z;p.• I SOSz.p.
[mg/IJ *1E+5 H

200 24.9 +0.30 +/-0.15 3.6


201 20.0 +0.22 +/-0.11 3.0
202 24.5 1 -0.14 +/-0.068 o.n
203 24.6 -0.14 +/-0.068 0.76
206 59.4 +0.27 +/-0.13 0.46
207 59.2 -0.054 +1-0.027 0.21
208 59.8 +0.027 +/-0.014 0.0055
209 30.0 +0.14 +/-0.059 3.9
210 50.3 -0.027 +/-0.014 0.0077
211 23.5 +0.22 +/-0.11 ~
212 12.06 +0.057 +/-0.029 0.058
Appendix 5.3 183

Tables A.5.3.1 and A.5.3.2 contain the SOS , d~ and dC values for
z.p. b,z.p. z.p.
the several single-solute NBZ and two-solute NBZ/PNA experiments. The tables
show that each experiment has a different value for the error in the initial
concentration and it is possible to correct the C~values for this error.

Table A.S.3.2. Variations in the concentration decay curve and errors


in the initial concentration due to a varying zero-
point, NBZ!PNA system

.c~ ••
EX:p. •••
dÇo;z.p. .••·•·•.·... <dCz;p...., . ..... SOSz.p.
.. · .·••.·••·••.•·.· · (JTig/1).
-
. ... •• • >••••. }.
i TT [Q19111 . i ...... "1E+5 [-]
•·. ·•· .·•.· ·. ]Q1glll
I/ . Nàz PNA .. NBZ PNA
··········
~ez 1·•.-.·.PNA •. NB~~
200 58.85 14.98 -0.11 -0.02 +/-0.09 +/-0.02 0.23 I o.18
201 88.65 I 15.19 -0.21 -0.05 +/-0.14 +/-0.04 0.25 0.69
202 59.13 14.86 +0.32 +0.14 +/-0.21 +/-0.15 1.26 10.2
203 29.67 14.70 +0.054 +0.11 +/-D.04 +/-0.07 0.18 2.27
204 58.01 14.58 -0.03 -0.04 +/-0.03 +/-0.03 0.03 0.42
205 58.27 14.74 -0.03 0.0 +/-0.03 +1-0.0 0.03 0.0

Error in volume
At the start of each experiment, in addition to the solution in the batch
vessel, there are always about 25 ml of pure water in the piping of the
installation. Also, when adding the carbon to the vessel another 5 ml of water
are added. This means that the the real volume of the experiment is (V +
0
0.03) liters. As the carbon concentration in the model calculations is
determined from the volume of the solution and the added amount of carbon,
using V0 instead of (V + 0.3) liters gives an error of 1% in the carbon
0
concentration. Although this has very little influence on the calculated
curve, consiclering the fact that all types of errors accumulate, it is
advisable to use a corrected volume in the model calculations.
Appendix 5.3 184

Error in time
For all experiments performed the first absorbance measurement was used to
determine the actual initial concentration. The activated carbon dose was
added just before the second measurement (time interval between two
measurements - 5 min.). This meant that, in reality, the experiment started at
time t=5 min, which was accounted for when comparing the experimental and
model ,curves.

Results from experimental reproducibility ~measurements


The changes in the concentration decay curve, resulting from the cumulative
effect of the different types of errors, were expressed with the help of a
SOS , defined by the following equation (also given in chapter 5):
rep

j c c. 2
SOS !\[c~J ( 1,1 ) ] (A.5.3.4)
rep J L c o
i=O b,l
c 0
b,l

where:
SOS :the sum of squares of the relative differences between the
rep
concentration decay curves of two duplicate experiments
('rep' stands for reproducibility)
:number of experimental points
:the experimentally determined concentration (dimensionless)
in the ith experimental point, exp. 1
:the experimentally determined concentration of the same
component (dimensionless) in the ith experimental point, exp. 2

This way of defining the SOS makes it possible to compare its values with
rep
the SOS values calculated from the difference between experimental and model
curves.

For every single- and two-component system used in this study at least one
duplicate experiment has been performed, while for the system NBZIPNA several
different duplo's were run. Table A.5.3.3 contains the typical values obtained
for SOS for the different systems.
rep
Appendix 5.3 185

Table A.5.3.3. SOSrep values from duplicate experiments for the different
systems used in the present study.

0.03, 6.91, 1.39


NBZ/PNA ' 1.92, 4.99, 0.26, 0.14
IPP/PNA IPP 0.52, 1.03
IPP/PNA PNA 0.11, 1.12

The average SOS rep , calculated on basis of the values in Table A.5.3.3 is
4
2·10- • This value was used , together with the range for SOS std (cf. section
5.L4) for comparison with the Sum of Squares between modeland experimental
curves (divided by the number of data points).
Appendix 5.4 186

Numerical solution of differential equations used in this study

The basic equations of the different models are developed in Chapter 3. Each
set of single-solute model equations consists of a partial differential
equation and an ordinary integral equation, tagether with the respective
initia! and . boundary conditions and equilibrium isotherms. The partial
differential equation, which represents physically the shell mass balance over
the partiele is non-linear and second order with respect to the radius, r and
frrst order with respect to the time, t (cf. Eq. 3.11).
Due to the fact that the isotherm equations used were highly non-linear the
systems of model equations could not be solved analytically and it was
necessary to use a numerical finite difference metbod for solving them.

For solving the equations a numerical algorithm was used, proposed by


Berkovskii [1976], where the space level is represented with an algebraic
equation of the concentrations of the previous, the present and the next space
level at any time level j+ 1 :

A ci+l + B d+ 1 + S ci+l = F k 1,2,3, ... ,K-1 (A.5.4.1)


k k-1 k k k k+l k

with the boundary conditions:

C=aC+b CK=aC +bK (A.5.4.2)


0 0 I 0 K K-1

where:
j the time level number,
k the space level number,
K the maximum space level number,
Ak, Bk. sk. Fk, ao, bo, aK, bK : coefficients which can be found from
the finite difference representation of the differential equation
of the intemal mass balance.

In this way there will be a set of K equations and solving it would be very
time consuming especially if the amount of levels is considerable. So it is
necessary to take into account the specific nature of this system. The matrix
Appendix 5.4 187

of the coefficients of this set of algebraic equations has a three-diagonal


forrn which makes it possible to use some special metbod of solution. One
possible metbod to solve this type of matrices will be dicussed below.

The solution of the system (A.5.4.1) - (A.5.4.2) is found by writing ck for


all the defined values of k in the same way as C has been expressed:
0

ci+l = a ci+l + ~ (A.5.4.3)


k k k+l k

Here a k and ~ are unknown coefficients. Equation (A.5.4.3) is valid for all
k
defined values of k, so it is also possible to write:

ei+! = a ei+! + ~ ' (A.5.4.4)


k-1 k- I k k-1

Substituting Eq. (A.5.4.4) in (A.5.4.1) gives:

A (a ei+' + ~ ) +B ei+' + s ei+' = F (A.5.4.5)


k k-1 k k-1 k k k k+l k

Writing Equation (A.5.4.5) in the forrn of equation (A.5.4.3) results in:

sk F k - Ak~k- I
ei+'=
k
ei+'
k+l
+ (A.5.4.6)
A a
k k- I
+ Bk A a
k k- I
+ Bk

A comparison of equations (A.5.4.3) and (A.5.4.6) gives the relations for the
coefficients ak and ~k:

Fk -A ~k-1
ak =- ~k = (A.5.4.7)
Aa
k k-1
+B
k
A a
k- I
+ Bk

From the boundary condition with index k = 0 follows that a = a and ~0 = b0 •


0 0
Equation (A.5.4.7) makes it possible to calculate consecutively a 1, ~ 1 , a 2 ,
~
2
,.... , a k-1 , ~
k-1
. Because Eq. (A.5.4.3) holds for all defined values of k,
the equation becomes for k = K-1:

ci+l = a ci+l + A (A.5.4.8)


K-1 K-1 K 1-'K-1
Appendix 5.4 188

This relation should be fulfilled independently of the other boundary


condition relating these parameters.
Combining Eq. (A.5.4.8) and the boundary condition with index k = K (cf.
Eq.(A.5.4.2)) gives:

ci+l = (A.5.4.9)
K
-a a
K K-1

ci+l is known now and when Eq. (A.5.4.7) for k = k-1 is used ci+ 1• ci+ 1, ... ,
K K-1 K-2
C~+l, C~+l can be found consecutively. In this way the system of K equations
is solved.
In order to find the A , B , S , and F coefficients in Equation (A.5.4.1) the
k k k k
differential Equation (3.11) (Chapter 3) has to be written in finite
difference form. The finite difference form of the general equation
(A.5.4.10), according to Berkovskii [1976], is given by Equations (A.5.4.11)
and (A.5.4.12):

ae = a ( r q X(r,t) ar
ae )
at rq
ar + ftr,t) (A.5.4.10)

2 h q e j+l_ e i+l
k+l k
= [
a r
( k
+ - k+-I ]
k+l 2 h
k+ I

ei+l - ei+!
k k-1
(A.5.4.11)
h

for r = 0, by using L'Hopital's rule it follows

2 (1 + q)
= a ( ei+l_ ei+l ) (A.5.4.12)
I I 0
't

where:
x(r,t) and f(r,t) : functions of radial coordinate r and time t,
k : space level,
Appendix 5.4 189

j : time level,
hk : space step between space level k-1 and k,
't : constant time step between two time levels,
q : parameter to indicate the configuration of the system:
q 0 : slab,
q = 1 : cylinder,
q 2 : sphere,
2
(h + h )
<pk =[1+ k k+l ] /.j+l/2 for q = 2,
48 r! k

For q = 2 (sphere), X(r,t) D = variabie and the term j(r,t) combined with
~~· the general Equation (A-.5~!:10) is similar to Bq. (3.11).

For a variabie D the terms a and a can be expressed as follows:


app k k+l

= 1 ( D (ei+') + D (ei+l ) )
ak 2 app p,k app p.k-1 (A.5.4.13)

a = 1 ( D (ei+l ) + D (ei+') ) (A.5.4.14)


k+l 2 app p,k+l app p,k

If a variabie grid is applied, the finite difference form of equation (3.11)


will be given by:

ei+l - eï 2 d h 2 ei+l ei+'


p,k p,k = --:------ [a (r +~ J p,k+l- p,k
t r~ (hk + hk+,) k+l k 2 hk+ I

2
hk
- a k (rk -
2
] hk
I
] (A.5.4.15)

eï+l - ei 1 .
p,O p,O
for r = 0: = 6d a (eJ+l
h2 I p,l
ei+l)
p,O (A.5.4.16)
't
1
Appendix 5.4 190

ei+l _ ei
for r = 0: p,O p,O
= 6 d (A.5.4.16)
t

with: d= (A.5.4.17)
[ 1 + (1-e pa )/epa pc :8· )
p

The variabie grid along the radial coordinate of the partiele can be
illustrated with the following scheme:

center rk surface
-----~----f-H
r =0 2 k I K
K I

Further on, rearranging Eqs. (A.5.4.15) and (A.5.4.16) in the form of Eq.
(A.5.4.1) gives the following coefficients in the algorithm:

1 2 d h 2
Ak =- a ( r- ~)
2
h r (h + h ) k k 2
k k k k+l

2 d h 2
h k ) 2)
Bk = t1 +
r
2
(hk + hk+l)
[ 1
h
a( r + ~) + _!_ a(r
k+l k
2 h k k 2
k k+l k

1
sk =-
h
k+l
(A.5.4.18)
ei
F
k
= - tp,k
- (k 1,2, .. K-l)

B = 1 + 6d a , (k = 0, boundary)
A 0 = 0,
0 h2 k+l
I
Appendix 5.4 191

ei
~
't
(k = 0, boundary).

Knowing from Eq. (A.5.4.7) for = 0 that o:0 = - k 0


and SJB ~
0 = FJB0,it is
possible to calculate o; and ~ and subsequently all o; and ~ for k = 1 to
0 0 k k
K-1 can be calculated with the same Eq. (A.5.4.7).
From the boundary condition at r = R , ei+ 1 can be calculated by making this
p p,K
equation discrete:

e o ( ei+l ei+ I ) = h k( e - ei+ I) (A.5.4.19)


pa app p,K p,K-1 k f b p,K

Together with Eq. (A.5.4.8) this gives:

Bii+l h
___k-:--:-- ~ e + ~
D (e +I) R b K I
app p,K pa
ei+l (A.5.4.20)
p,K Bi +I h
k k
-0:
K
+
D (ei+') R
ap p p,K pa

where Bi kR
f pa
IepaD app .

By using Eq. (A.5.4.3) the whole concentration profile for the time level j+ 1
can be calculated.
Finally the overall mass balance of the component has to be considered for
completing the approximation. The equation for the overall mass balance can be
integrated by using the well-known trapezoidal rule and the unknown bulk
concentration at j+ 1 time level can be expressed as:

R
m
R3 p
1 4n
V
J pa (e
pa
ei+ I+ (1-e )p Qi+')rzdr (A.5.4.21)
p pa c
0
pa pa

Por the value of e~+ 2 all the calculations above have to be repeated.
Appendix 5.5 192

Flowsheet of general algorithm for single-solute model with variabie


diffusion coefficient

In Fig. A.5.5.1 a flowsheet diagram is given of the algorithm of the single


component type of model with variabie diffusivity (this is also the flowsheet
of the constant diffusion model, with some minor differences). Starting with a
new time level j+l, the first new pore liquid concentration profile ci+l (new)
• p,k,
is calculated with a slope of the isotherm aQ /BC and D (var) at C1 and
p . app p~
with the bulk'
concentration of the previous time level C.
b
Bulk concentration
ei and aQ*!aC and D at ei serve as initial values for the new time
b p app p,k .
level calculation. The reason for taking these initia} values is that the C1 +1
* .I l1·f
profile is needed to calculate aQ /BC and D at C1+ and to calculate C1+ .
* p . fPP p,k . I b
On the other hand, aQ /BC and D at CJ+ and bulk concentration C1+ are
. p app p,k b
needed to calculate the C 1+1 profile. To overcome this problem two convergence
p,k .
1
loops (loop I and loop 2) are applied in the algorithm. After C1 + (new) is
renamed in ei+ I' the calculation of the ei+ I (new) profile is :e;eated, but
P•• k . I p ,k
now with a aQ /aC and D at the C1+ (new) of the previous iteration, which
. Ip a pp p ,k . I
is now called C1p,k
+ . This repeated calculation of C1p,k
+ (new) profiles continues

until the sum of the squares of the differences between the new calculated
profile cj+l(new) and the previous calculated cj+l profile is less than a
p,k p,k
certain input value for the precision (erl ). The result of these
stop
1
iterations is a ci+ profile aQ*!aC and
calculated with a D at
. p,k p app
1
approximately C 1+ .
p,k
These iterations are called loop 1 in Fig. A.5 .4.1.

After updating the mass balance of the component in the bulk solution, which
is of course a function of the last calculated ci+l profile, loop 1 is
. I p,k
repeated with the newly calculated C1+ (new). Again, the overall mass balance
b .
of the component is updated with the C1+1 profile (calculated in loop I) and a
. I p,k . I
new C 1 + (new) replaces the previous one, which is renamed in C 1 + • This
b b
repetition of ci+l(new) calculations stops when the square of the difference
b . .
between the new calculated c~+ 1 (new) and the previous c~+l is less than a
eertaio input value for the precision (er2 ). This loop around loop 1 is
stop
Appendix 5.5 193

[begin J
-r-
=0 0 '
C
p ' k
0
Cb
0
= 1' l
r
I j = + 1

ei+ 1 ei in i ti a I values for new


b b
t i me I evel calculation s
ei+ 1 ei
p' k p 'k

~ r
k= 0

[~] p
ei+
p' 0'
1 D
• p p
(Ci+ 1
p '0
), ao, Po
' r
space
level Ek + 1
I
loop r
[~] p
ci+l
p' k'
D
• p p
(Ci+ 1 ) ,
p 'k at' 13k

r
no {~~
r yes
~ (new)l
r
[k = k~
space
-r
level
loop ~ (new)l
r
no {~~
r yes
[ erl =~ < ei+
p ' k
1 ( new) - ei + 1
p ' k
) 2

I
r
loop 1 I
ei+ 1
p' k
= ci+
p ' k
1
(new)
]
r
I er 1 < e r 1
no I s t op I
1
yes r
Appendix 5.5 194

time
Ie vel
loop

Figure A.S.S.l. Flowsheet of general algorithm for single-component model


with variabie diffusion coefficient

called loop 2. The result after loop 2 is a cJ+l profile calculated with a
• . I p,k
aQ /aC and D at approximately C1+ and with a bulk concentration of
P apf p,k .
approximately C~ . After printing j+ 1 and C~+\ the whole procedure is
1

repeated for the next time leveL


Appendix 5.6 195

program AAI(input,output);

{Two-component Maxwell-Stefan model)


{This program calculates the CDC's with matrix calculations. It solves the
differential equations simultaneously. This makes it possible to include
dql/dc2's and cross-diffusion coefficients. )

CON ST
ncorop =2;
ndim = 2;
ftol = l.Od-3;
=
cexmax 500;
jkrmax = 500;

TYPE
extended = double;
real = double;
RealArrayNP = ARRAY [1 •. ndim] OF extended;
arrayldr ARRAY [l .. ncomp] OF real;
array2dr = ARRAY [1 .. ncomp] OF arrayldr;
arrayldi = ARRAY [l .. ncomp] OF integer;
array I db = ARRAY [l .. ncomp] OF boolean;
cexarr ARRAY [O .• cexmax] OF arrayldr;

VAR
LinminNcom : integer;
LinminPcom,LinminXicom : RealArrayNP;
fret,sos extended;
i ter ,n,l,lmax,k,fparam,jkr integer;
p,st,dx RealArrayNP;
cex cexarr;
finp,foutpopt,foutpcur ,foutpinf ,finp2 text;
inputfilename,infofilename,filename,compnaam : string(25) value ";
singlerun boolean;

7.include 'FNCl' {This is the function containing the numerical metbod


calculations. It is listed after the main program )

FUNCTION fnc(VAR x:RealArrayNP):extended;


var teller:integer;
BEGIN
sos:= fncl(x};
for teller:=l to ndim do
begin
wri te(f outpopt,exp{x[ teller]),' ');
write{exp(x[teller]):l2,' ');
end;
write{foutpopt,sos,' ');
write(sos:l2,' ');
for teller:=l to ndim do
begin
write(foutpopt,x[teller],' ');
write(x[teller]:l2,' ');
end;
Appendix 5.6 196

writeln(foutpopt);
writeln;
if singlerun then
begin
close(foutpopt);halt;
end;
fnc:=sos
END;
PROCEDURE dfnc(VAR x,df:RealArrayNP);
VAR
n:integer;
xn:RealArrayNP;
sosd, testvar:extended;

BEGIN
sosd := sos;
for n:=l to ndim do
BEGIN
xn:=x;
xn[n] := ln(l+dx[n])+x[n]; { x[nJ•(l+dx[n]); }
testvar:=fncl(xn);
df[n] := (testvar-sosd)/ln(l+dx[n]);
END;
END;

7.include 'FlDIM.PAS'
7.include 'DFlDIM.P AS'
7.include 'MNBRAK.PAS' { These are optimizationprocedures from
7.include 'DBRENT.PAS' "Numerical Recipesin Pascal" }
7.include 'DLINMIN.PAS'
7.include 'FRPRMNLP AS'

BEGIN
fparam:=O;
write('Enter input-filename (return for default): ');readln(filename);
if filename=" then filename:='lx910308';
inputfilename := 'mexp:'+filename; {change!}
writeln('inputfilename = '" ,inputfilename, "");
writeln;
filename:=";
write('Enter output-filename (return for default): ');readln(filename);
if filename=" then filename:='dflt';
filename:='mres:'+filename;
writeln('outputfilename = '" ,filename, "");
writeln;

open(finp,inputfilename+' .red' ,old);


reset(finp);
1:=0;
while not eof(finp) do
begin
readln(finp,cex[l, I] ,cex[l ,2]);
1:=1+1;
end;
Appendix 5.6 197

lmax:=l-1;
close(finp);

writeln('Total experiment length:' ,lmax:4,' HALF hours. ');


write('Enter new experiment length IN HALF HOURS (0 for number above): ');
readln(jkr);
if jkr=O then jkr:=lmax;
writeln;
writeln('Experimental length is ',lmax/2:6:1,' hours');
writeln('Calculated length will be ',jkr/2:6:1,' hours');
writeln('SOS is calculated over smallest of the two');
writeln;
if jkr<lmax then lmax:= jkr;

for n:=1 to ndim do


begin
write('Enter starting value for parameter' ,n:2,': ');
readln(st[n]); {s~arting values for param}
dx[n]:=1d-4; {value of differentiation interval}
writeln('Differentiation interval will be set to' ,dx[n]:9);
writeln;
p[n]:= In (st[n]);
end;

open(foutpopt,filename+' .opt' ,history:=new ,sharing:=readonly);


rewrite(foutpopt);

frprmn(p,ndim,ftol,iter ,fret);

close(foutpopt);

open(foutpinf ,infofilename,old);
extend(foutpinf);
writeln(foutpinf);
writeln(foutpinf, 'resultaten: ');
writeln(foutpinf, 'iteraties= ',iter:3);
writeln(foutpinf,'Minimum found at: ');
for n:=1 to ndim do
begin
writeln(foutpinf, 'opt. value' ,n:2,' = ',exp(p[n]));
end;
writeln(foutpinf, 'Minimum function value = ',fret);
close(foutpinf);

END.

FUNCTION FNCI(var x:RealArrayNP):extended;

CONST
ikr=46;
epspa=0.57d0;
rp::0.0006d0;
rop=799;
pi=3.1415926535897932d0;
Appendix 5.6 198

initval=ld-34; {initia! conc. value}


cpprint=true; {print cp-profiles? }

TYPE
cbarr =ARRAY (O .. jkrmax] OF arrayldr;
cparr array [O .. ikr] of arrayldr;
matrarr = array [O .. ikr] of array2dr;
timevar = packed array (1..14] of char;

VAR
cpold,cpnew ,cpmid, bet a cparr;
cb,cpikr cbarr;
af ,bs, bf ,ad,bd,k,cO,dp,ds,fk,iv ,erl,sos,mm arrayldr;
cbpr ,cbpr2,dummyv ,cbb,cba arrayldr;
alfa matrarr;
ak,bk,sk,b,al,a2,ii,kf ,dummym array2dr;
vol : array [O .. ikr] of real;
h : array [I..ikr] of real;
p : arrayldi;
i,j,n,o,noi,printind,fail : integer;
pk,qk,cd,fr extended;
ml,tau,m,v : extended;
r,som,dl2 : extended;
sing,accuracy : boolean;
time, tt, timestep,increase,maxtimestep : extended;
eval,evalinit,evalstep,stepcounter : integer;

7.include 'procs.pas'
7.include 'prtime.pas'

function q(m:integer; c:arrayldr ):extended;


var den,num,bog,tog,cl,c2: extended;
begin

cl:=c(m];
c2:=c[3-m];

bog:=bf[m]+bs(m]+l;
tog:=bf[m]+bs[m];
num:=k[m]*af[m]•ci••bog;
den:=af[mJ•(clUtog)+k[m]*cl•((clUbs[m])+ad[m]•(c2Ubd[m]));
q:=num/den;
end;

function dq(m,n:integer; c:arrayldr ):extended;


var den,pwdql,mog,nog,A, W ,dB,B,cl,c2:extended;
begin

cl:=c[m];
c2:=c[3-m];

if m=n then
begin
den:=(cl**bs[m])+ad[m]•(c2**bd[m));
mog:=bf[m)+2•bs[m]-l;
nog:=bf[m)+bs[m]-1;
Appendix 5.6 199

pwdql:=af(m]lllbf[m]lll(clllllllmog);
den:=denlllden;
d8::;:(pwdql+af(m]lllad(m]lll(clllllllnog)lll(c2llllllbd[m])lll(bf[m]+bs[m]))/(den);
8:=af(m]lll(clllllll(bf[m]+bs[m]))/((clllllllbs[m])+ad[m]lll(c2lll*bd[m]));
A:=l/(k[m]lllcllllcl);
W:=A+d8/(8lll8);
dq:=q(m,c)lllq(m,c)liiW;
end
else
begin
a:=ad[m]lllbd[m]lll(c2lll*(bd[m]-l))/(af[m]lll(clllllll(bf[m)+bs[m])));
dq:=-q(m,c)lllq(m,c)llla;
end;
end;

function dapp(i,j:integer;c:arrayldr):extended;
var xl,x2,d:extended;
begin
if i<> j then dapp:=O
el se
dapp:=dp(i]+mllllds[i]lllq(i,c)/c[j];
end;

BEGIN
singlerun:=false;
diaga(l,ncomp,O,O,kf);
open (finp,inputfilename+' .dat' ,old);
reset(finp);
readln(finp,m);
readln(finp, v);
for i:=l to ncomp do
begin
readln(finp,cO[i]);
end;
readln(finp,compnaam);
close(finp);
open(finp2, 'mexp: '+compnaam+' .eql' ,old);
reset(finp2);
readln(finp2,dl2);
for i:=l to ncomp do
begin
readln(finp2,compnaam);
readln(finp2,k[i]);
readln(finp2,af[i]);
readln(finp2,bf[i]);
readln(finp2, bs[i]);
readln(finp2,ad[i]);
readln(finp2,bd[i]);
open(finp, 'mexp: '+compnaam+' .dat' ,old);
reset(finp);
readln(finp,mm[i]);
readln(finp,kf(i,i]);
readln(finp,ds[i]);
readln(finp,dp[i]);
Appendix 5.6 200

close(finp);
end;
close(finp2);

ds[l] := exp(x[l]);
ds[21 := exp(x[2]);

for n:=l to ncomp do


begin
for i:=O to ikr do
begin
epnew[i,nJ:=initvai•eO[n];
end;
ebpr[n]:=cO[n]; { previous eb }
cbpr2[n] :=eO[n]; { 2nd previous eb }
end;
cb(O]:=cbpr;
cpold:=epnew;

h[461:=0.0000ld0;
h[ 45]:=0.00ld0;
for i:=44 downto 2 do
begin
h[i]:=h[i+l]+O.OOldO;
end;
h[l]:=0.00999d0;

{ INITIATETIME-VARIABLES}
timestep:=O.OOldO;
maxtimestep:=900d0;
inerease:= 1. 03d 0;
evalinit:=O;
evalstep:=l800;
eval:=evalinit+evalstep;
time:=O;
{ END INITIA TE }

{ integral mass balanee


pk:=3*m/(4*pi•rop•v);
vol[O]:= pbpi/6•((h[l])**3);
r:=O;
for i:= 1 to ikr-1 do
begin
r:=r+h[i];
vol[i]:=pk•4/3•pi•(((r+h[i+l]/2)**3)-((r-h[i]/2)**3));
end;
voUikr ]:=pb4/3•pi•(l-((l-h[ikr ]/2)••3));
ml:=rop/epspa;
printind:=l;
stepeounter:=O;
j:=I;

{ CREATE INFO-FILE }
if fparam = 0 then
begin
Appendix 5.6 201

infofilename := filename+' .inf';


open(foutpinf ,infofilename,new);
rewri te(foutpinf);
writeln(foutpinf, 'inputfile: ',inputfilename);
write(foutpinf, 'optimize: ');
if singlerun then writeln(foutpinf,'NO') else writeln(foutpinf,'YES');
writeln(foutpinO;
writeln(foutpinf, 'number of time steps = ',jkr:S);
writeln(foutpinf, 'number of place steps = ',ikr:S);
writeln(foutpinf, 'radius partiele = ',rp!l!l000000:6:0,' micrometer');
writeln(foutpinf,'mass carbon = ',m!I!I000000:7:1,' milligram');
writeln(foutpinf,'volume = ',v!l!l000:7:l,' liter');
writeln(foutpinf ,' initial timestep = ', timestep:l0:4,' sec');
writeln(foutpinf, 'timestep increase = ',(increase-1)!1!100:8:2,' 7.');
writeln(foutpinf, 'maximum timestep = ',maxtimestep:6:0,' sec');
writeln(foutpinf);
writeln(foutpinf, 'cO[l] ',c0[1]!1!1000:7:2,' milligram/liter');
writeln(f outpinf, 'Kf[l, l] = ',Kf(l,l]:IO);
writeln(f outpinf, 'Dp[l] = ',dp[l]:IO);
writeln(foutpinf, 'Ds[l] = ',ds(l]:lO);
writeln(foutpinf);
writeln(foutpinf, 'c0[2] = ',c0(2]!1!1000:7:2,' milligram/liter');
writeln(foutpinf, 'Kf[2,2] ',Kf[2,2]:10);
writeln(f outpinf, 'Dp[2] = ',dp[21:10);
writeln(foutpinf, 'Ds[2] ',ds(2]:10);
writeln(foutpinf);
writeln(foutpinf, 'isothermll]: ');
writeln(foutpinf, 'k[l] = ',kUI:S:O);
writeln(foutpinf ,'af[l] = ',af(l]:IO:S);
writeln(foutpinf ,'bf[l] = ',bf(l]:lO:S);
writeln(foutpinf, 'bs[l] = ',bs[l]:lO:S);
writeln(foutpinf ,'ad[l] ',ad[l]:lO:S);
writeln(foutpinf, 'bd[l] = ',bd[11:10:5);
writeln(foutpinf);
wri teln(f outpinf, 'isotherm(2]:' );
writeln(foutpinf ,'k[2] = ',k[2] :5:0);
writeln(foutpinf, 'af(2] = ',af[2]:10:5);
writeln(foutpinf, 'bf[2] = ',bf[2]:10:5);
writeln(foutpinf ,'bs(2] = ',bs[2]:10:5);
writeln(foutpinf ,'ad(2] = ',ad[2]:10:5);
writeln(foutpinf, 'bd[2] ',bd[2]:10:5);
close(foutpinf);
end;
fparam:=fparam+ 1;
{END CREATE INFO-FILE)

while j<= jkr do


{START CALCULATION FOR THIS TIME LEVEL)
begin

{TIME CALCULATIONS)
if eval-time>timestep then
begin
tt:=timestep;
time:=time+timestep;
end
Appendix 5.6 202

else
begin
tt:=eval-time;
time:=eval;
end;
{END TIME CALCULATIONS}

{ LINEAR ESTIMATION OF NEXT CB AND CPNEW }


for n:=l to ncomp do
begin
for i:=O to ikr do
begin
pk:=cpnew[i,n]-cpold[i,n];
cpold[i,n]:=cpnew[i,n];
cpnewli,n]:=cpnewli,n]+pk;
end;
cb[j,n]:=2•cbpr[n]-cbpr2(n];
end;
{END LINEAR ESTIMATION}

noi:=O;
tau:=tt/rp/rp;
stepcounter:=stepcounter+ I;

REPEAT

for n:=l to ncomp do


begin
erlln]:=O;
for i:=O to ikr do cpmid[i,n]:=(cpold(i,n]+cpnew(i,n])/2;
end;
noi:=noi+l;
{START CALCULATION OF CPNEW}
{INNER BOUNDARY CONDITION}
for n:=l to ncomp do
BEGIN
for o:=l to ncomp do
begin
b(o,n]:=ml•dq(o,n,cpmid(O]);
a2[o,n]:=(dapp(o,n,cpmid[l])+dapp(o,n,cpmid[0]))/2;
end;
b[n ,nl:=b[n,n]+l;
end;

fr:=I;
dec(l,ncomp,b,p,sing,cd,fr); { invert b }
if (cd>IOOOOOO) or (cd<O.OOOOOI) then writeln(j:S,0:4,noi:4,cd:l0,
cpmid[i,Il:IO,cpmid(i,2]:10};
decinvU,ncomp,b,p,fail);
amaalb(l,ncomp,b,a2,sk); { calculate sO }
alfaaU,ncomp,-6/(h[ll•b[l]),sk,dummym);
sk:=dummym;
alfav(l,ncomp,l/tau,cpold[O],fk); { calculate fO }
diaga(l,ncomp,l/tau,O, bk); { calculate bO }
Appendix 5.6 203

aminb(l,ncomp, bk,sk,dummym);
bk:=dummym;
fr:=l;
dec(l,ncomp,bk,p,sing,cd,fr); { invert bk }
decinv(l,ncomp,bk,p,fail);
amaalbU,ncomp,bk,sk,alfa[O)); { calculate alfa }
alfaaO,ncomp,-l,alfa[O],dummym);
alfa[O]:=dummym;
amaalvU,ncomp,bk,fk,beta[O]); { calculate beta }
{END INNER BOUNDARY CONDITION}

r := 0;
for i:=l TO ikr-1 do
begin
r := r+h[i];

al:=a2;
for n:=l to ncomp do
BEGIN
for o:=l to ncomp do
begin
b[o,n]:=ml*dq(o,n,cpmid(i]);
a2(o,n]:=(dapp(o,n,cpmid[i+l])+dapp(o,n,cpmid(i]))/2;
end;
b[n,n]:=b(n,n]+l;
end;

pk:=2/(r•r•(hlil+hli+l]));
qk:=pk•(r-hlil/2)•(r-h[i]/2)/h[i);
pk:=pk•(r+hli+ll/2)•(r+hli+ll/2)/hli+ll;

fr:=l;
dec(l,ncomp,b,p,sing,cd,fr); { invert b }
if i mod 10=0 then
begin
if (cd>IOOOOOO) or (cd<O.OOOOOl) then writeln(j:5,i:4,noi:4,cd:IO,
cpmid[i,I]:IO,cpmid[i,2):10);
end;
decinv(l,ncomp,b,p,fail);
amaalbU,ncomp, b,a2,sk); { calculate sk }
alfaa(l,ncomp, -pk,sk,dummym);
sk:=dummym;
amaalb(l,ncomp,b,al,ak); { calculate ak }
alfaa(l,ncomp, -qk,ak,dummym);
ak:=dummym;
alfav(l,ncomp,l/tau,cpold(i],fk); { calculate fk }
diaga(l,ncomp,l/tau,O,bk); { calculate bk }
aminbU,ncomp,bk,sk,dummym);
aminb(l,ncomp,dummym,ak,bk);
amaalb(l,ncomp,ak,alfa[i -l] ,dummym); {calculate ii (ii=helpmatrix)}
aplusb(l,ncomp,dummym,bk,ii); { ii=A[kl•alfa(k-ll+B(k] }
fr:=l;
decU,ncomp,ii,p,sing,cd,fr); { invert ii }
decinvU,ncomp,ii,p,fail);
amaalb(l,ncomp,ii,sk,dummym); { calculate alfa }
alfaaU,ncomp,-l,dummym,alfali]);
Appendix 5.6 204

amaalv(l,ncomp,ak,betali-ll,iv); {calculate beta (iv=helpvector) }


uminv(l,ncomp,fk,iv ,dummyv);
amaalvU,ncomp,ii,dummyv ,beta[i]);
end;

{ OUTER BOUNDARY CONDITION }


diaga(l,ncomp,l,O,ii); { al,a2,ak,ii: helpmatrices }
aminbU,ncomp,ii,alfa[ikr-J],dummym); { fk,iv: helpveetors }
for n:=l to ncomp do
begin
for o:=l to ncomp do
begin
al[ o ,n] :=epspa•dapp(o, n,cpmid[ikr]);
end;
end;
amaalb(l,ncomp,al,dummym,ii);
alfaa(l,ncomp,h[ikr J•rp,kf ,a2);
aplusb(l,ncomp,ii,a2,dummym);
ii:=dummym;
fr:=l;
dec(l,ncomp, ii, p, sing, cd, fr); { invert ii }
decinv(l,ncomp,ii,p,fail);
amaalv(l,ncomp,al,beta[ikr-l],fk);
amaalv(l,ncomp,a2,cb[j],iv);
uplusv(l,ncomp,iv,fk,dummyv);
amaalvU,ncomp,ii,dummyv,iv);
for n:=l to ncomp do
begin
er l[n] :=er l[n]+sqr((i v[n]-cpnew[ikr ,n])/i v[n]);
end;
cpnew[ikr] :=i v;
{END OUTER BOUNDARY CONDITION}

for i:=ikr-1 downto 0 do


begin
amaalv(l,ncomp,alfa[i],cpnew[i+l],iv);
uplusv(l,ncomp,iv,beta(i),dummyv);
iv:=dummyv;
for n:=l to ncomp do
begin
erl[n]:=erl[n]+sqr({iv[n]-cpnew[i,n])/iv[n]);
end;
cpnew[i]:=iv;
end;

{ END CALCULATION OF CPNEW }

{START CALCULATION OF CB}


for n:=l to ncomp do
begin
som:=O;
for i:=O to ikr do
begin
som: =som+vol [i J•( epspalfcpnew[i ,n]+rop*q(n, cpnew[i]));
end;
iv[n]:=cO(n]-som;
Ap;Jendix 5.6 205

end;
cba:=iv;
for n:=l to ncomp do
begin
iv[n]:=kf[n,nl•J•m•tt/rp/rop/v;
cbb[n] :=(cbpr[n]+i v[nl•cpnew[ikr ,n])/(iv[n]+ 1);
end;

iv:=cba;
for n:=l to ncomp do
begin
erl[n]:=erl[n]+sqr((iv[n]-cb[j,n])/iv[n]);
end;
cb[j]:=iv;
{END CALCULATION OF CB}

accuracy:=true;
for n:=l to ncomp do
begin
erl[n]:=erl[n]/(ikr+2);
accuracy:=accuracy and (erlln]<O.OOOl);
end;
UNTIL accuracy;

{ FLUX CHECKING }
uminvU,ncomp,cb[j],cpnew[ikr],iv);
amaalvU,ncomp,kf ,iv ,dummyv);

uminvU,ncomp,cbpr,cb[j],iv);
for o:=l to ncomp do
begin
iv[o]:=iv[o]•v/tt•rp•rop/3/m;
end;
{ END FLUX CHECKING }

cpikr[j]:=cpnew[ikr];
cbpr2:=cbpr;
cbpr:=cb[j];

{ OUTPUT TO SCREEN }
{if (stepcounter mod 10=0)
or (stepcounter<10) then writeln(j:4,noi:4,stepcounter:S,prtime(time),
erl[l] :10,er1[2] :10,cb[j,1]:12,cb[j,2]:12,'
dummyv[l]:l2,iv[l]:l2,' ',dummyv[2]:12,iv[2]:12); }
{ END OUTPUTTO SCREEN }

{ PRINTCP-PROFILES}
if cpprint then
begin
if time=3600*printind then
begin
if printind>99 then printind:=O;
open(foutpcur,filename +chr(48+(printind div IOO))
+chr(48+((printind-100•(printind div 100)) div 10))
+chr(48+(printind mod 10))+' .pep' ,unknown);
Appendix 5.6 206

rewrite(foutpcur);
r:=O;i:=O;
writeln(foutpcur,r:l2,cpnew[i,J]/c0[1):12,cpnew[i,2)/c0[2]:12);

for i:=l to ikr do


begin
r:=r+h[i);
writeln(foutpcur,r:12,cpnew[i,l]/c0[1):12,cpnew[i,2]/c0[2]:12);
end;
close(foutpcur );
printind:=printind•2;
end;
end;
{ END PRINTCP-PROFILES }

{TIME CALCULATIONS}
timestep:=timestep•increase;
if timestep>maxtimestep then timestep:=maxtimestep;
if time=eval then
begin
eval:=eval +evalstep;
j:=j+l;
end;
{END TIME CALCULATIONS}

end;
{END CALCULATION FOR THIS TIME LEVEL}

{ CALCULATE SOS}
pk:=O;
for n:=l to ncomp do
begin
sos[n]:=O;
for 1:=0 to lmax do
begin
sos[nl:=sos[n)+sqr(cb[l,n)/cO[n)-cex[l,n));
end;
sos[n):=sos[n)/lmax;
pk:=pk+sos[n);
writeln('SOS component' ,n:2,' = ',sos[n));
end;
fncl:=pk/ncomp;
{ END CALCULATE SOS }

{ CREATE CURVE-FILE}
open(foutpcur,filename+' .cur' ,unknown);
rewrite(foutpcur);
writeln(foutpcur,'tt comp 1: cb_exp cb_cal cp_ikr residue_pp comp 2: idem');
writeln(foutpcur,'tt cOlli= ',cO[l],' gram/liter');
writeln(foutpcur,'tt c0[2)= ',c0[2],' gram/liter');
writeln(foutpcur ,'# Ds[l)= ',ds[l]);
writeln(foutpcur,'# Ds[2)= ',ds[2));
writeln(foutpcur ,'tt sos[l]= ',sos[l]);
writeln(foutpcur,'# sos[2]= ',sos[2]);
1:=0;
while l<=jkr do
Appendix 5.6 207

begin
if l<=lmax then writeln(foutpcur ,l/2:10,
cex[l,11:12, cb[l,1]/cO[l]:l2, cpikr(l,l]/cO[l]:l2,
cb[l,1]/cO[l]-cex[l,1]:12,
cex[l,2]:12, cb[l,2]/c0[2]:12, cpikr[l,2]/c0[21:12,
cb[l,2]/c0[2) -cex[l,2] :12)
else writeln(foutpcur ,l/2:10,
0:12, cb[l,l]/c0(1]:12, cpikr[l,l]/cO[l]:l2, 0:12,
0:12, cb[l,2]/c0[2]:12, cpikr[l,2]/c0[2]:12, ' 0:12);
'
1:=1+1;
end;
cl ose(foutpeur);
{END CREATE CURVE-FILE}

end;
Appendix 5.7 208

Comparison of analytica) and numerical solutions for


special cases of the constant-ditlusion models
(some additional details)

In order to check the accuracy and reliability of the numerical solution and
programs used in this study the single-component constant diffusion model was
compared, for special cases, to analytica! solutions reported in literature.

Further, the programs for solving the single-component models with variabie
D were compared to the ones with constant diffusivity, by setting the
app
variabie D to be constant as well.
app
As all checks were carried out with simplified versions, they do not prove
completely that the normal (non-simplified) models are solved correctly.
Nevertheless, these checks give a serious indication about the correctness of
the numerical solution and programs used.

Case I
A particular case of the general constant diffusivity model is when the
concentration in the bulk solution is constant (infmite bath) and the
adsorption isotherm is a linear function (which means that the derivative
aQ*taC is constant).
p

Due to the first simplification it was decided to compare the pore liquid
concentration profiles inside the particle.

If we consider now the aQ·/aC in the right term of the partiele differential
p
equation we can replace it by a constant value, K . Because of that the whole
h
part outside the differential term becomes a constant which constant part can
be defined as a constant diffusion coefficient D' . Then the modified
app
partiele differential equation will be:

ac
p = D'app
at
(A.6.2.1)
Appendix 5.7 209

where:

D
app
D' = = constant
app
[ 1 + (1-E pa )IE pa pc K h]

The boundary conditions are:

ac
pI -
ar 0
(A.5.6.2)
ac
k (C
r bath
- cp IR ) = E D _P
pa app a
IR
pa r
pa

and the initial condition: at t = 0 is: C


p
=Q =0 for 0 < r < R
pa
at t ~ 0 is: C = constant
bath

The analytica! salution of this simplified set of equations has the following
form (Crank [1990]):

n=oo
c )l sin ()l /;)
2
F~
_P = 2 Bi L n n
--
2
1 [
1- e
-)ln
J (A.6.2.3)
cbath I; n=1 )l sin )l - Bi cos )l )ln
n n n

where:
3
C : the concentration of the component in the infmite bath (kg/m ),
bath
Bi = R k I D E ,
pa f app pa
2
Fo = D'
app
tI R ,
pa

=riRpa
)l are the positive roots of the characteristic equation :
n

)l = (1- Bi) tan )l . (A.6.2.4)


n n
Appendix 5.7 210

Several concentration profiles, each one for a different Fourier time, were
compared. The profiles corresponded very well, except for low Fourier times.
Mter 12 minutes of adsorption the frrst profiles were compared. The
difference in the pore concentrations at R was about 4% (relative to the
pa
concentration of the component in the infinite bath). After 3 hours of
adsorption this difference had been reduced to 0.5% and after 10 hours and
more it was already less than 0.1 %. For all the adsorption times considered
the difference between the analytically and numerically calculated
concentratien profiles diminished, going from r = Rpa to r = 0.
The D and Bi values used in the analytica! salution (A.6.2.3) were 8·10- 10
app
2
m /s and 23.077, respectively.

Case 2
The other analytica! solution used for comparison is based on the following
model simplifications:
1. No extemal mass transfer resistance: kr -7 oo •

2. The adsorption isotherm is again a linear function (as in Case 1).

This analytica! salution (Ruthven [1984], cf. Chapter 5) is based on the


loading Q. Therefore, in order to make the desired comparison, a slightly
different model had to be used, based not on the pore liquid concentration,
C , but on the carbon loading, Q. However, as the algorithm and numerical
p
solution were exactly the same as for the original models (Maxwell-Stefan and
Fickian models based on C ) the comparison gave valuable information
p
conceming the accuracy and reliability of the calculation programs used.
No extensive derivations of this model will be given here as it is not
relevant for this study, only the fmal set of equations will be presented:

8Q
(A.5.6.5)
at
8C
V b = - m A (1-€ ) p DQ 8
Q IR (A.5.6.6)
8 t s pa c app 8 r
pa
Appendix 5.7 211

aQ
The boundary conditions are: lo = o and Ql
ar R
pa

and the initial conditions (t = 0) are:

Q = 0 for 0 < r < Rpa

• is in equilibrium with the bulk concentration C b.


Qb

The analytica! solution of the system of Eqs. (A.5.6.5), (A.5.6.6), the


boundary conditions, the initial conditions and a linear isotherm is given
by:

p 2 t I R2
00
exp( -D'Q )
app n pa
= 1- 6 [ 2
(A.5.6.7)
Qoo 9A/ ( 1-A) + (1-A)p
n =1 n

where:
p are the positive roots of:
n
3p n
3 + (1/ A - 1) pn2 '

Qav : the average loading of the carbon partiele at time t (kg.kg · 1)


Q : the ultimate loading (kg.kg- 1)
00

A : the fraction of the component ultimately adsorbed: (C7 c:)/C~.


with Coo : the ultimate bulk concentration.
b

Equation (5.3) (Chapter 5) is used to fmd the bulk concentration at time t


from the analytica! solution (A.5.6.7):

e
Q ..!!!._ [ pa + 1] (A.5.6.8)
av V K (1-e )p
h pa c
Appendix 5.7 212

From Eq. (A.5.6.8) it can be seen that the total component uptake consists of
the uptake by the pore liquid and the uptake by the carbon. The uptake by the
carbon is of course much larger.
The value of A in analytica! solution (A.5.6.7) can be found by substituting
C = Coo and Qav = K h Coo in (A.5.6.8). The values used in the check runs were:
K = 10 m 3/kg, A = 0.3334 and D'Q = 6.8868·10. 14 m2/s.
h app
In a time interval of 0 to 10 hours the numerically calculated bulk
concentration decay curve was compared with the analytically calculated curve.

In Table 5.2 (cf. Chapter 5) the relative difference in the amount of


component adsorbed is given at five different times after starting the
adsorption. The relative difference is defined as follows:

(1-C ) - (1-C
an num
) cnum - can
r.d. = (5.5)
(1-C ) 1 - c
an an

where:
cnum the numerically calculated dimensionless bulk concentration,
can the analytically calculated dimensionless bulk concentration.

Table 5.2 (cf. Chapter 5) shows that for short adsorption times the relative
difference is considerable. The difference between the numerically and
analytically calculated profiles inside the partiele (the first comparison)
was also large in the beginning of the adsorption run. These differences for
short adsorption times can be explained with the equations for updating the
mass balance of the component in the bulk solution used in the models. In this
case the amount of component which leaves the bulk solution in one time step
is calculated with Eq. (5.3). This mass balance depends on the gradient of the
C profile at R via the boundary condition. Using the finite difference
p pa
metbod the gradient at R is approximated on basis of two consecutive space
pa

steps. However, the gradient is very high at R in the beginning of the


pa
adsorption experiment. As a result a poor approximation of the gradient is
obtained because the space step is not infinitely small. The calculated
Appendix 5.7 213

gradient at R is too small, which means that too little component moves into
pa
the partiele in one time step. The bulk concentradon decreases less rapidly
as a function of time in the case of the numerial salution and the relative
difference is therefore positive. Because the relative difference ts
considerable for short adsorption times the mass balance of the component in
the bulk solution, Eq. (5.3) was replaced by the overall mass balance of the
component (cf. Eq. (3.13)).

Another reason for using mass balance (3.13) instead of mass balance (5.3) is
that when using (5.3) the bulk concentration is only indirectly linked to the
concentration profiles inside the particles. The errors made m the
approximation of the gradient at each time level accuroulate and the bulk
concentration may drift away from the actual uptake after a number of time
steps. When Eq. (3.13) is used, the bulk concentration is coupled to the
uptake of component by the particles at every time level during the adsorption
run.
Appendix 6.1 214
Experimentally determined value for external mass transfer coefficient
Table A.6.1.1. Experimentally determined values for external mass transfer
coefficient for component NBZ
\.g)cp~)· Çcl ''''. öï I Rp~ kf

1>······'·········, il *:fE+S
I .. ,. i .· .
(Olg/1)> •••• [qlg/1) [rtl] [m/s)..
,.>,······ ., I
'

,.......
200 24.89 51.30 0.60 ·., . . 7.Q7 ....
••••

201 20.00 32.76 0.60 . . ,• ~.4!)••'•,•·


202 24.50 51.40 0.60 .s;55,
203 24.56 25.48 0.60 .' < 8~87

206 59.44 245.15 0.60 .6.53


207 59.20 125.45 0.60 6191
...
208 59.78 75.40 0.60 6.58
209 30.04 124.10 0.60 '6~59
210 50.34 24.44 0.60 . 5~62

211 23.54 75.00 0.60 6.53


..
212 12.06 24.44 0.60 6.63.

Table A.6.1.2. Experimentally determined values for external mass transfer


coefficient for component IPP
Exp. Co
· ...
.m Rpa
., ' .... ,.
··.·· kf ..
.
,..... ,
I'
*1Ë+5 <
,·'..·
,.,.
.... , [rngtl] .. [ing/1] ... [mm] [öïh>] .· ..

24 14.74 51.9 0.65 4.12


25 14.94 51.4 0.55 5~11 ·.··
14.81
~t8o~f
26 51.5 0.60
··,······,·
27 14.78 50.0 0.60 . ,4lt1 '...
28 48.44 175.4 0.60 3;58
..
29 48.30 350.0 0.60 ·î:12.
30 15.28 49.9 0.60 4.97
31 14.82 51.4 0.60 4.53
32 15.21 51.8 0.60 4.36
52 51.85 174.7 0.60 4.62
53 51.26 524.9 0.60 3.79
54 99.39 653.6 0.60 4.39
56 98.11 172.1 0.60 4.56
Appendix 6.1 215

Table A.6.1.3. Experimentally determined values for external mass transfer


coefficient for component PNA

18 29.71
19 29.03 100.6 0.60 L .. ·FÜiS)
20 30.34
21 30.41
22 10.21
23 20.08
33 29.44
34 28.87 50.1
36 30.05 50.0
37 29.88 50.5
38 19.89 50.2
39 29.32 50.1
40 29.74 50.3
41 9.80 100.9
42 9.92 74.6
43 9.86 49.6 o.6o ••·. >6~4!S
44 29.22 50.1
45 29.48 50.5 0.41 . 8:63
............
46 31.17 49.9 0.16 ······
s.•HL .. •

47 28.03 50.8
....... ~·ojfi·...
48 28.27 50.8 0.26 ·~--...
:. :.

49 27.96 50.7 0.65 •... )7;32

~---:-~--~--29-~~-~-+-~-·. · · ·• ·: ~· · ·
,.•
Appendix 6.2 216

TRANSITION MODEL (HPMT) BASED ON TUE PORE LIQUID


CONCENTRATION GRADIENT

(see Section 6.2.4)

The HPMT model, which uses relation (6. 7) (cf. Chapter 6) for D was fitted
app
to several IPP and PNA experiments. Initially, a limited number of
2-dimensional minimizations were performed, where the optimized parameters
were DTso and a. Most of them gave a value of a approximately 0 which, in view
of Eq. (6.7), means that D is constant. An example of some 2-D results is
app
given in Table A.6.2.1.

Table A.6.2.1. Results from 2-D minimizations for several experiments

IPP 31 5.63 0.028 0.64


IPP 52 2.35 0.130 2.73
PNA 20 12.8 12 3.44

lt was also found out that the two parameters were dependent on each other
which is illustrated by Figure A.6.2.1. The contour pictures shown were made
in the a-D domain of objective functions of two experiments.
so

In view of the contour plots presenled in Fig. A.6.2.1, it would be


interesting to know how the valleys of different adsorption experiments of a
component are situated in the a-D domain. If all valleys cross each other at
so
one point of the domain, than that point of intersection will be defined by
one set of a D which are the optimum values for all experiments. The most
so
obvious way to fmd out if this is the case is to make contour plots of all
experiments fora given component. However, this procedure can be very tedious
and time-consuming and for this reason another option was used in this study.
Several experiments per component were chosen and for each one a number of
1-dimensional minimizations were performed, optimizing D and keeping a
so
Appendix 6.2 217

3.75E-ll

2.5E:-ll 2.5E-J.3

l.25E-ll l.2:0E:-l3

.0000000~~~~----~~--~ .0000000~--~~~--~~----~
0.0 ALFA 0.5 0.0 ALFA 0.:0

Figure A.6.2.1. a Objective function of PNA exp. 43 when Eq.(6.7) is used


b Objective function of IPP exp. 52 when Eq.(6.7) is used

constant, but different for each run. The optimal D 's found in combination
so
with the fixed a's indicate where the valley is situated in the a-D domain.
so

The results from the 1-dimensional fits are presented in Tables A.6.2.2 and
A.6.2.3. They show that a Q equal to 0.344 kg/kg was used for all IPP and
ref
PNA experiments. In addition, for a few IPP experiments, a value of 0.285 for
Qre1 was also applied. Note that the choice of Qref was rather arbitrary and
this reference loading does not have a real physical meaning. Some attempts
were made to link Q to the carbon loading of the component at saturation
ref
concentration, however, such an assumption involves extrapolating Q values for
a concentration range far beyond the one covered by the equilibrium isotherm
experiments.

The results presented in Tables A.6.2.2 and A.6.2.3 give also an indication
about the possible reasons for most 2-dimensional optimizations ending up at
w:::O. It should be noted that parameter a cannot become negative because the
fit-parameters were scaled logarithmically. This means that if the SOS in a
Appendix 6.2 218

Table A.6.2.2. Results of 1-dimensional minimization runs


for /PP experiments, HPMT model
.. J
Exp. Co m Qsat .··.· Dso SOS
.... *1E+4
. ·· .
·.·•·• : ... : ·.·.

(1119111 .· (lllgiJ) : 1

1kQ/kQJ •.: ... ·.· f812/S)

27 14.78 50.0 0.344 -1.5 1.02E-13 1.75


0.344 -1.0 1.73E-13 1.88
0.3 4.89E-13 1.95
0.344 .0 1.31E-12 2.03
0.344 1.5 2.10E-12 2.05
0.$44 3.5 L12E-n . 2;01 •·:·
0.344 4.0 1.61 1.88

0.285 -2.0 5.12E-14 9.77


0.285 -1.0 1.74E-13 4.66
0.285 0.0 4.89E-13 1.97
0. 0 2.01 E-12 2.06
0. 5 2.84E....:11 1.88

29 48.30 350.0 0. 2.0 1.91 E-14 1.06


0.344 -1.0 6.83E-14 5.09
0.344 0.0 2.34E-13 8.45
0.344 1.0 7.57E-13 2.41
0.344 3.5 9.43E-12 g,gg
0. .5 2.07E-11 12.5

0.285 -2.0 3.39E-15~


0.285 -1.0 i 3.02E-14
0.285 0.0 2.34E-13 0.85
0.285 1.0 1.41 E-12 5.81

~
2.0 5.98E-12 12.5
3.5 2.82E-11 15.6
Appendix 6.2 219

Table A.6.2.2. ( continued)

§~$i.
. >Q~ti•········· r>. i;ilfil··········· i'b!L6ll••············ Il}f1ê+4>
Co ..•.•. }~ti
<!=XP···•·.·.·••·······
< .•. . . .• )i >···· i ....... i i ~w•.· · ·:;·,· · <i••·· I •· ·.
••••••• •••• •••••••••••••••••••••••••
f . rmi
·······.·.·.···
é' i. , i• •
'f
·.;·

............... i . . [fu9/tl·······. .·•. i<············)


lfl1giiJ ]kgl~g]
'>>. r:
... · · · '...• ····<

31 14.82 51.4 0.344 -1.5 1.10E-13 0.68


0.344 -1.0 1.87E-13 0.67
4 0.0 5.31 E-I.:J v.64
0.344 1.0 1.42E-12 0.58
~,~ ....... ••··r~<·•·•·•·••• !~~~Ä~±if. [?ó;~~(
0.344 4.5 2.41E-11 0.36

5 -2.0 2.01E-14 0.70


0.285 -1.0 1.09E-13 0.72
0.2 0.0 5.31E-13 0.64
0.285 1.0 2.19E-12 0.51
0.285 2.0 7.32E-12 0.40
• o:285 · .·• / S;~ . 2.996-11 Q;33

52 51.85 174.7 0. -1.5 2.19E-14 1.69


0.344 -1.0 .36E-14 1.94
0.344 0.0 1.71E-13 2.62
0.344 1.0 6.33E-11 3.45
()..~4 ···s,s. 1.04E-l1 5.33

0.285 -2.0 1.03E-15 5.33


0.285 -1.0 1.46E-14 9.77
0.285 0.0 1.71E-13 2.62
0.285 1.0 1.42E-12 4.66
0.285 3.5 3,79E;..;U 5.47·.•.
••••••••

53 51.26 524.9 0.344 -2.0 3.99E-14 5.33


0.344 0.0 4.26E-13 1.06
0.344 3.5 1.44E··4'fi .· i 1á.ij i'
Appendix 6.2 220

Table A.6.2.3. R esults of 1-dimensional minimization runs


for PNA experiments, HPMT model
i ..........
~. ·.ro•····
i>
' t:)(p. ~., ·epi ··· alfa.••·• 0~ i iS()§
u ....> ...•.... · · ·•· ·•·• · •..•••
. . Osat .. ,
·~ · I·
f •••••.• < .. ... *1E+4 .......· · ·
·•· •••••••••••••••
·

..... ...
I
. [mD/11 ImfliJ kg/kg] .· .. ...······· [rn2TsJ .••.•. 1. •. I I

17 29.49 98.8 0.344 -1.0 6.80E-13 0.17


0.344 0.0 3.00E-12 0.95
0.344 1.0 1.15E-11 2.35
.•. ·.~).~.:1, ...... 3;5
· .. ·.
1.G5E-'10 · .· 3.59
0.344 4.5 3.79E-10 1.79

20 30.34 48.6 0.344 -1.5 7.43E-14 2.60


I
0.344 -1.0 1.99E-13 2.87
I 0.344 0.0 1.27E-12 3.44
..·.· .. ·
0.344 .•.[ 1.35.E-tO i2,66 ...
i 3.5 . •·.·.·
0.344 4.5 3.21E-10 1.63

38 19.89 50.2 0.344 -2.0 1.84E-13 0.87


0.344 0.0 3.20E-12 1.93
0.344 1.0 1.14E-11 2.30
0.344 ·..•.. 3.5 .......... 1.S3E.. 10 . 0.90
0.344 4.5 3.61E-10 1.26

41 9.80 100.9 0.344 -1.0 2.38E-11 0.18


0.344 0.0 3.55E-11 0.16
0.344 1.0 5.23E-11 0.13
0.344 3.5 1.30E-10 0.14
!
0.344 4.5 1.84E-10 0.20

42 9.92 74.6 0.344 -1.0 1.73E-11 0.14


0.344 0.0 2.89E-11 0.17
0.344 1.0 4.71E-11 0.22
0.344 3.5 1.47E-10 0.55
0.344 4.5 2.24E-10 0.81
Appendix 6.2 221

valley decreases in the direction of lower a, the minirnization procedure will


will end up in the valley where a is 0. The values obtained from the 1-D
optirnization runs show indeed that for a number of experiments (although not
for all) the SOS decreases in the direction of lower a.

If we take IPP experiment 52 as an example, results in Table A.6.2.1 show that


4
the 2-D minirnization gave an optimal a of 0.13 and a SOS equal to 2.73·10- .
On the other hand, the data in Table 6.2.2 indicates that the above SOS is not
the lowest one, therefore it is not the minimum of the valley. One possible
explanation is that the gradient in the valley became too low and the
minirnization procedure stopped somewhere in the valley , without going to the
exact minimum. This may always happen when the minirnization procedure
encounters long, stretched valleys. However, it is not so important to know
the exact minimum of the valley but rather to find out where the valleys of
different experiments come together. Only then will the model be able to
predict the behaviour of the system under investigation.
Figure A.6.2.2 shows the logarithm of D as a function of a for several PNA
so
experiments. Figure A.6.2.3 gives the same dependence, but for component IPP.

-9.0

-10.0

11.0
~
~
.Q -12.0

-13.0

14.0
-3 -2 -1 0 2 3 4 5

alfa

~ e17 ~ e20 ~ e38 ~ e42 ~ e41

Figure A.6.2.2. Valleys of several PNA experiments for Qref = 0.344 kg/kg
Appendix 6.2 222

In this way of presenting the results the valleys of the different experiments
are given as lines.

When analyzing Fig. A.6.2.2 perhaps the most surprising phenomena is the
existence of one intersection point for all valleys. For a:;() and D ""
so
10
1'.5-10- m2/s the HPMT gives a reasonably good fit for all PNA experiments
listed in Table A.6.2.3. The range of D
so
for a = 3.5 is (1.30 - 1.65)·10. 10,
which compared to the ranges of D obtained from the HPMC is rather
app
narrow.

In the case of IPP there isn't a distinct point of intersection, although the
valleys come close together for a=2 when Qref = 0.285 kg/kg and for a""3.5
when Q = 0.344 kg/kg (cf. Fig. A.6.2.3). For the IPP experiments the range
ref
of D , for a:::3.5 and 0 0.344, is (0.94 - 1.45}10. 11 m2 /s which is
so "'ref
nevertheless smaller than the D range.
app

-10.0

-10.8

0!I) -11.6
Q

~
Ol
.Q -12.4 ~
~~+~
-13.2 +~
+4
-14.0
-3 -2 -1 0 2 3 4 5

alfa

-'V- exp27 - +- exp29 o exp31 - ~- exp52

Figure A.6.2.3. Valleys of several IPP experiments for Qref = 0344 kg/kg

In Fig. A.6.2.4 two model fits for PNA are shown, one with the HPMC, the other
with the HPMT. From the residues in Fig. A.6.2.4b it becomes clear that the
Appendix 6.2 223

a
1.2 .--------------5
1.1 ~ -o.010
0
u...... 1.0 -0.020
~
z 0.9 -o.030 '-----'------'------'-----'------''-----'
0 0 20 100 120
i= 0.8
<(
a: 0.7
I-
z
w 0.6
u
ö
u
0.5

0.4
0.3
0 20 40 60 80 100 120

TIME {hours)

exp. 38 + Dap var. o Dap const.


{HPMTl {HPMCl
Figure A.6.2.4. a Experimental curve, HPMT model curve (D = 5.46.10- 10 .
so
a = 3.5 and Qref 0344) and HPMC curve of PNA
b Residue graphs of the two models, exp. 38

HPMf curve fits better than the HPMC curve which means that not only the
diffusion coefficient range is narrower in the case of HPMf, but also the
quality of the fit can he better.
The IPP data presented in Table A.6.2.2 shows that for some experiments the
HPMf gives better fits, while for others they are worse than those with the
HPMC model.

The foregoing results and discussion indicate that the HPMf has a good
predictive value and fits reasonably well the different experiments for 1PP
and PNA single-component adsorption systems. However, as shown, it is basedon
rather empirica! considerations which makes it rather tricky to use for
modeling of complicated adsorption systems.
Appendix 7.1 224

Experimental values for external mass transfer coefficients


Table A.7.1.1. Experimentally determined values for external mass transfer
coefficients for system NBZ!PNA

TC206 58.9 15.4 213.6 99.0


TC207 58.9 59.7 171.2 99.0
TC210 58.9 172.5 109.5
TC211 59.2 29.5 172.3 109.5
TC212 30.3 29.7 172.8 109.5
TC213 89.0 29.6 173.7 72.5
TC214 44.7 44.3 171.7
TC215 59.2 14.8 173.4
TC216 169.9
TC217 172.2
TC218 43 173.9 75.0

Table A.7.1.2. Experimentally determined values for extemal mass transfer


coefficients for system IPPIPNA

TC101 61.32 14.98


TC102 60.97 15.07
TC104 61.78 15.20 75.16 66.0
TC105 62.00 10.04 125.00 69.0
TC106 15.05 175.52 69.0
TC107 20.03 125.45 68.5 2.78
TC111 92.48 15.11 125.40 65.5 2.74
TC115 3.62 4.55
TC116 4.13 5.43
REFERENCES

Arve, B.H., Liapis, A.I. "Modeling and Analysis of Biospecific Adsorption


in a Finite Bath", AIChE J., 33, 179 (1987).

Ash, R., Barrer, R.M. "Mechanisms of Surface Flow", Surface Sci., 8, 461
(1967).

Bansal, R. Ch., Donnet, J.B., Stoeckli, Fr. "Active Carbon", Marcel


Dekker, lnc., New York and Basel (1988).

Barrer, R.M. "Surface and Volume Flow in Porous Media", m


"The Solid-Gas
Interface", vol. 2, Flood, E.A. (Ed.), Marcel Dekker, New York (1967).

Berkovskii, B.M., Nogotov, E.F. "Difference methods in Heat Transfer


Problems", Publishing House 'Nauka i Technika', Minsk (1976). In Russian.

Bird, R.B., Stewart, W.E., Lightfoot, E.N. "Transport Phenomena", New


York (1960).

Brunauer, S., Emmett, P.H., Teller, E. "Adsorption of Gases in


Multimolecular layers", J. Amer. Chem. Soc., 60, 309 (1938).

Butler, J.A.V., Ockrent, C. "Studies in Electrocapillarity. lil", J. Phys.


Chem., 34, 2841 (1930).

Baudu, H., LeCloirec, P., Martin, G. "Modelisations des lsothermes


d' Adsorption sur Charbon Actif de Composés Aromatiques en Solution Aqueuse",
The Chem. Eng. J., 41, 81 (1989).

Cooney, O.O., Strusi, F.P. "Analytica! Description of Fixed-Bed Sorption of


Two Langmuir Solutes under Nonequilibrium Conditions", Ind. Eng. Chem. Fund.,
11, 123 (1972).

Costa, E., Sotelo, J.L., Calleja, G., Marron, C. "Adsorption of Binary and
Temary Hydrocarbon Mixtures on Activated Carbon", AIChE J., 27, 5 (1985).

Coughlin, R.W., Ezra, E.S. "Role of Surface Acidity in the Adsorption of


Organic Pollutants on the Surface of Carbon", Env. Sci. Technol., 2,
291 (1968).

Crank, J., "The Mathernaties of Diffusion", 2"d ed., Oxford University Press
(1990).

Crittenden, J.C., Luft, P., Haud, O.W., Oravitz, J.L., Loper, S.W., Art,
M. "Prediction of Multicomponent Adsorption Equilibria Using Ideal Adsorbed
Solution Theory", Env. Sci. Technol., 19, 1037 (1985).

Do, D.D., Rice, R.G. "Validity of the Parabolic Profile Assumption in


Adsorption Studies", AIChE J., 32, 149 (1986).
Relerences

Donker, M. Graduate Report, Eindhoven University of Technology (1992).

Driel, J. "Surface Chemistry of Activated Carbon", in "Activated


Carbon ...a Fascinating Material", Norit NV, the Netherlands (1983).

Duri, B.A.A.K. Al "Mass Transfer Processes in Single and Multicomponent


Batch Adsorption Systems" Boston Spa, British Library (1988).

Eic, M., Goddard, M., Ruthven, D.M. "Diffusion of Benzene in NaX and
natura! faujasite", Zeolites, 8, 327 (1988).

US Environmental Proteetion Agency. "Process Design Manual for Carbon


Adsorption", Cincinnati, Ohio (1973).

Faust, S.D., Aly, 0. "Adsorption Processes for Water Treatment", Butterworth


Publishers, MA 02180 (1987).

Fettig, J., Sontheimer, H. "Kinetics of Adsorption on Activated Carbons: I.


Single-Solute Systems", J. of Env. Eng., 113, 764 (1987).

Freundlich, H. Colloid and Capillary Chemistry, Metbeun and Co., London


(1926).

Fritz, W., Schlünder, E.U. "Simultaneous Adsorption Equilibria of Organic


Solutes in Dilute Aqueous Solutions on Activated Carbon", Chem.Eng. Sci., 29,
1279 (1974).

Fritz, W., Schlünder, E.U. "Competitive Adsorption of Two Dissolved Organics


onto Acti vated Carbon I. Adsorption Equilibria", Chem. Eng. Sci.,36, 721
(1981).

Fukuchi, K. "Application of Vacancy Solution Theory to Adsorption from Dil u te


Aqueous Solutions", J. of Chem. Eng. of J., 15, 316 (1982).

Furusawa, I., Smith, J.M. "Intraparticle Mass Transport in Slurries by


Dynamic Adsorption Studies", AIChE J., 20, 88 (1974).

Gamba, G., Rota, R., Carrá, S., Morbidelli, M. "Adsorption Equilibria of


Nonideal Multicomponent Systems at Saturation", AIChE J., 36, 1736 (1990).

Gilliland, E.R., Baddour, R.P., Perkinson, G.P., Sladek, KJ. "Diffusion on


Surfaces. I. Effect of Concentration on the Diffusivity of Physically Adsorbed
Gases", Ind. Eng. Chem Fund., 13, 95 (1974).

Grant, T.M., King, J.C. "Mechanism of Irreversible Adsorption of Phenolic


Compounds by Activated Carbons", Ind. Eng. Chem. Res., 29, 264 (1990).

Hand, D.W., Ctittenden, J.C., Thacker, W.E. "User-orientated Batch Reactor


Solutions to the Homogeneaus Surface Diffusion Model", J. Env. Eng. Div.,
ASCE, 109, 82 (1983).

Helferich, F., Klein, G. "Multicomponent Chromatography", Marcel Dekker, New


York, NY (1970).
References

Horiguchi, Y., Hudgins, R.R., Silveston, P.L. Can. J. Chem. Eng., 49, 76
(1971).

Hsieh, J.S.C., Turian, R.M., Tien, C. "Multicomponent Liquid Phase


Adsorption in Fixed Beds", AIChE J., 23, 263 (1977).

Huige, N.J.J. "Nucleation and Growth of lee Crystals from Water and Sugar
Solutions in continuous stirred-tank crystallizers", Ph.D. Thesis, Eindhoven
University of Technology (1972).

Jain, J.S., Snoyeink, V.L. "Adsorption from Bisolute Systems on Active


Carbon", J. Water Poll. Control Fed., 45, 2463 (1973).

Janssen, R. Personnel comrnunication. (1992).

Jaroniec, M. "Physical Adsorption on Heterogeneaus Solids", Advances


Colloid Int. Sci., 18, 149 (1983).

Jaroniec, M., Brauer, P. Surface Sci. Reports, 6, 65 (1985).

Jaroniec, Madey, R. "Physical Adsorption on Heterogeneaus Solids", Elsevier,


Amsterdam (1988).

Jossens, L., Prausnitz, J.M., Fritz, W., Schlünder, E.U., Myers, A.L.
"Thermodynamics of Multi-Solute Adsorption from Dilute Aqueous Solutions",
Chem. Eng. Sci., 33, 1097 (1978).

Kerkhof, P.J.A.M. Personnal communication. (1988).

Kocirik, M., Zikánová, P., Dubsky, J.Ind. Eng. Chem. Fund., 12(4), 440
(1973).

Komiyama, H., Smith, J.M. "Surface Diffusion in Liquid-Filled Pores", AIChE


J., 20, 1110 (1974).

Kozhusko, V.V., Lukin, V.D. "Kinetics of Adsorption in the Concentration


Dependenee of the Diffusion Coefficient", Russsian J. of Phys. Chem., 62,
1994 (1989).

Kreyszig, E. "Advanced Engineering Mathematics", 6th ed., Chichester: Wiley


(1988).

Krishna, R. "A Generalized Film ModelforMass-Transfer in Non-Ideal Fluid


Mixtures", Chem. Eng. Sci., 32, 659 (1977).

Krishna, R., Standart, G.L. "Mass and Energy Transfer in Mlticomponent


Systems", Chem Eng. Comrn., 3, 201 (1979).

Krishna, R., Taylor, R. "Multicomponent Mass Transfer: Theory and


Applications", in "Handbook for Heat and Mass Transfer Operations", ed. by
N.P. Cheremisinoff, Chapter 7, Vol. 2., Gulf Publishing Corporation, Houston
(1986).
Reierences

Krishna, R. "A Unified Theory of Separation Processes Based on Irreversible


Thennodynamics", Chem. Eng. Comm., 59, 33 (1987).

Krishna, R. "Multicomponent Surface Diffusion of Adsorbed Species: A


Description Based on the Generalized Maxwell-Stefan Equations", Chem. Eng.
Sci., 45, 1779 (1990).

Krishna, R. "Problems and Pitfalls in the use of the Fick Pormulation for
intraoarticle diffusion", to he published, (1992).
Laïmbock, P. Graduate report, Eindhoven University of Technology (1991).
Landolt-Boernstein, "Zahlenwerte Und Funktionen aus Naturwissenschaften und
Technik", II. Band, Teil 2, b, p. 3-451, 3-447.

Langmuir, J. "The Adsorption of Gases on Planes of Glass Mica and Platinum".


J. Amer. Chem. Soc., 40, 1361 (1918).
Lee, C.S., Connell, J.P. "Statistica! Mechanics of Partially-Mobile
Adsorption of Gases on Homogeneous Solid Surfaces", J. Colloid Int. Sci., 41,
415 (1972).

Lee, C.S., Belfort, G. "Thermodynarnics of Multilayer Solute Adsorption from


Dilute Aqueous Solutions. 1. The Use of an Equation of State", lnd. Eng.
Chem. Res., 27, 951 (1988).

Letterman, R.D .. , Quon, J.E., Gemmell, J. Water Pollut. Contr. Feder.,46,


2536 (1974).

Liapis, A.I., Rippin, D.W. "A General Model for the Simulation of
Multi-Component Adsorption From a Finite-Bath", Chem. Eng. Sci., 32, 619,
(1977).

Lier, W. van "Mass transfer to activated carbon in aqueous solutions", Ph.D.


Thesis, Delft University of Technology ( 1989).

Lightfoot, E.N. "Transport Phenomena and Living Systems", John Wiley, New
York (1974).
Ligthart, M. Graduate report, Eindhoven University of Technology (1992).
Lin, S.H. "Effectiveness Factors For Concentration-Dependent Diffusion", J.
of Chem. Eng. of J., 23, 635 (1990).
Mamchenko, A.V., Marutovskii, R.M., Koganovskii, A.M. '' Dependenee of the
Effective Diffusion Coefficient of Adsorbed Molecules on the Adsorption
Conditions", Russian J. of Phys. Chem., 53, 738 (1979).
Mansour, A.R., Shalalam, A.B., Sotari, M.A. "Parametric Sensitivity Study
of Multicomponent Adsorption in Agitated Tanks", Sep. Sci. and Technol., 20,
1 (1985).
References

Mathews, A.P. "Mathematica! Modeling of Multicomponent Adsorption inBatch


Reactors", Ph.D. Thesis, Univ. of Michigan, Ann Arbor, MI (1975).

Mathews, A.P. "Adsorption in an Agitated Slurry of Polydisperse Particles",


AIChE Symp. Ser., "Adsorption and Ion Exchange", No.230, Vol. 79, 18
(1983).

Mattson, J.S., Mark, H.B., Malbin, M.D., Weber, W.J, Crittenden, J.C."Surface
Chemistry of Active Carbon: Specific Adsorption of Phenols",J. of Coll. and
lnterf. Sci., 31, 131 (1969).
McKay, G., Bino, M.J. "Application of Two Resistance Mass Transfer Model to
Adsorption Systems", Chem. Eng. Res. Des., 63, May (1985).
McKay, G., Bino, M.J. "Adsorption of Pollutants from Wastewater onto
Activated Carbon Based on External Mass Transfer and Pore Diffusion", Wat.
Res., 22, 279 (1988).
McKay, G., Al Duri, B. "Prediction of Multicomponent Adsorption Equilibrium
Data Using Empirica! Correlations", The Chem. Eng. J., 41, 9 (1989).
Misic, D.M., Smith, J.M. "Adsorption of Benzene in Carbon Slurries", Ind.
Eng. Chem. Fund., 10, 380 (1971).
Moon, H., Lee, W.K. "Intraparticle Diffusion in Liquid-Phase Adsorption of
Phenols with Activated Carbon in Finite Bath Adsorber", J. of Coll. and Intf.
Sci., 96, 162 {1983).
Muraki, M., Iwashima, Y., Hayakawa, T. "Rate of Liquid-Phase Adsorption on
Activated Carbon in the Stirred Tank", J. of Chem. Eng. of Japan., 15, 34,
(1982).

Myers, A.L., Prausnitz, J .M. "Thermodynamics of Mixed-Gas Adsorption", AIChE


J., 11, 121 (1965).

Myers, A.L. in Fundamentals of Adsorption, Proceedings of Second


International Confer. Eng. Foundation, New York (1987).
Myers, A.L. "Equation for Equilibrium Adsorption from Liquid Mixtures" ,in
Fundamentals of Adsorption, Proceedings of Third International Confer. United
Eng. Trustees, Inc., New York (1991).
Neretnieks, I. "Adsorption in Finite Bath and Countercurrent Flow with
Systems Having a Nonlinear Isotherm", Chem. Eng. Sci., 31, 107 (1976a).
Neretnieks, I. "Adsorption in Finite Bath and Countercurrent Flow with
Systems Ha ving a Concentration Dependent Coefficient of Diffusion", Chem. Eng.
Sci., 31, 465 (1976b).
Neretnieks, I. "Analysis of Some Adsorption Experiments with Activated
Carbon", Chem. Eng. Sci., 31, 1029 (1976c).
References

Okazaki, M., Hiroyuki, K., Toie, R. "Prediction of Liquid Phase Adsorption


Equilibria of Multi-Solutes in Water", J. of Chem. Eng. of Japan, 13, 286
(1980).

Patrykiejew, A., Jaroniec, M. "Partially Mobile Adsorption of Gases on Solid


Surfaces", Advances Colloid Int. Sci., 20, 273 (1984).

Pirbazari, M., Weber, W.J. Jr., "Adsorption of Polychlorinated Biphenyls


from Water by Activated Carbon", in "Chemistry in Water Reuse", ed. by
Cooper, N.J., Ann Arbor Sci., (1981).

Pope, C.G. "Flow of Sulphur Dioxide over the Surface of Spheron 6(2700)
Graphitized Carbob Black", Trans. Faraday Soc., 63, 734 (1967).

Press, W.H., Flannery, B.P., Teukolsky, S.A., Vetterling, W.T. "Numerical


Recipes in Pascal", Cambridge University Press, Cambridge (1989).

Ramos, R.L. "Surface Diffusion in Liquid-Filled Pores of Activated


Carbon", PhD Thesis, Ohio State Univ., Columbus (1981).

Radke, C.J., Prausnitz, J.M. "Adsorption of Organic Compounds from Dilute


Aqueous Solutions on Activated Carbon", Ind. Eng. Chem. Fund., 11, 445
(1972a).

Radke, C.J., Prausnitz, J.M. "Thermodynamics of Multi-Solute Adsorption from


Dilute Liquid Solutions", AIChE J., 18, 761 (1972b).

Reid, R.C., Prausnitz, J.M., Sherwood, T.K. "The Properties of Gases and
Liquids", 4th edition, McGraw-Hill, New York (1986).

Rice, R.G. Approximate Solution for Batch, Packed Tube and Radial Flow
Adsorbers - Comparisons with Experiment", Chem. Eng Sci., 37, 83 (1982).

Richter, E., Schultz, W., Myers, A.L. "Effect of Adsorption Equation on


Predietien of Muitkomponent Adsorption Equilibria by the lAS Theory", Chem.
Eng. Sci., 44, 1609 (1989).

Rivin, D. Rubber Chem. Technol., 44, 307 (1971).

Ruthven, D.M. "Principles of Adsorption and Adsorption Processes",


Wiley-Interscience Publication, New York (1984).

Satterfield, C.N., Cadle, J. Ind. Eng. Chem. Process Des. Dev., 7, 256
(1968).

Satterfield, C.N. "Mass Transfer in Heterogeneous Catalysis", MIT Press,


Cambridge, Mass. (1970).

Schay, G.J., Fejes, F.P., Szethmary, J. "Studies on the Adsorption of Gas


Mixtures, I. Statistica! Theory of Physical Adsorption of the Langnmuir type
in Multicomponent Systems", Acta Chem. Acad. Sci. Hung.,12, 299 (1957).
Heferences

Seidel, A., Gelbin, D. "On Applying the Ideal Adsorbed Solution Theory to
Multicomponent Adsorption Equilibria of Dissolved Organic Compounds on
Activated Carbon", Chem. Eng. Sci., 43, 79 (1988).

Sircar, S., Myers, A.L. "Surface Potential Theory of Multilauyer Adsorption


from Gas Mixtures", Chem. Eng. Sci., 28, 489 (1973).

Smith, O.M., Keiler, J.F. "Nonlinear Sorption Effects on the Determination


of Diffusion/Sorption Parameters", lnd. Eng. Chem. Fund., 24(4), 497 (1985).

Sokolowski, S. "High Temperature Physical Adsorption of Gases on Solids: A


Comparison of Two-Dimensional and Three-Dimensional Treatments", Advances
Colloid Int. Sci., 15, 71 (1981).

Standart, G.L., Taylor, R., Krishna, R. "The Maxwell-Stefan Pormulation of


Irreversible Thermodynamics for Simultaneous Heat and Mass Transfer", Chem.
Eng. Comm., 3, 277 (1979).

Steele, W.A. "The Interaction of Gases with Solid Surfaces",Pergamon Press,


Oxford (1974).

Steenbergen, P.A. Graduate Report, Eindhoven University of Technology,


(1991).
Sudo, Y., Misic, O.M., Suzuki, M. "Concentration Dependenee of Effective
Surface Diffusion Coefficients in Aqueous Phase Adsorption on Activated
Carbon", Chem. Eng. Sci., 33, 1290 (1978).

Suzuki, M., Kawazoe, K. "Batch Measurement of Adsorption Rate in an


Agitated Tank", J. of Chem. Eng. of J., 7, 346 (1974).

Suzuki, M., Misic, O.M., Koyama, 0., Kawazoe, K. "Study of Thermal


Regeneration of Spent Activated Carbons: Thermogravimetrie Measurement of
Various Single Component Organics loaded on Activated Carbon", Chem. Eng.
Sci., 33, 271 (1978).

Suzuki, M., Fujii, T. "Concentration Dependenee of Surface Diffusion


Coefficient of Propionic Acid in Activated Carbon Particles", AIChE J., 28,
380 (1982).
Suwanayuen, S., Danner, R.P. "A Gas Adsorption lsotherm Equation Basedon
Vacancy Solution Theory", AIChE J., 26, 68 (1980).

Suwanayuen, S., Danner, R.P. "A Gas Adsorption Isotherm Equation Based on
Vacancy Solution Theory", AIChE J., 26, 76 (1980).

Talu, 0., Kabel, R.L. "Isosteric Heat of Adsorption and The Vacancy Solution
Model", AIChE J., 33, 510 (1987).

Thacker, W.E., Crittenden, J.C., Snoyeink, V.L. "Modeling of \Adsorber


Performance: Variabie lnfluent Concentration and Comparison of Adsorpents"J.
Water Poll. Control Fed., 56, 243 (1984).
References

Thakkar, S., Manes, M. "Adsorptive Displacement Analysis of Many-Component


Priority Pollutants on Activated Carbons", Env. Sci. Technol., 21, 546
(1987).

Thomas, H.C. "Heterogeneous Ion Exchange in a Flow System", J. Amer. Cbem.


Soc., 66, 1664 (1944).

Traegner, U.K., Suidan, M.T. "Parameter Evaluation for Carbon Adsorption",


J. of Env. Eng., 115, 109 (1989).

Valenzuela, D., Myers, A.L. "Adsorption Equilibrium Data Handbook", Prentice


Hall, New Jersey (1989).

Vedder, H.G.M. Graduate Report, Eindhoven University of Technology,


(1990).

Vidic, R., Suidan, M.T., Traegner, U.K., Nakhla, G.F. "Adsorption


isotherms: illusive capacity and role of oxygen", Wat. Res., 24,
1187 (1990).

Voncken, H. Graduate Report, Eindhoven University of Technology, (1992).

Walas, S.M. "Phase Equilibria in Cbemical Engineering", Butterworth


Publishers, MA (1985).

Wesselingh, J.A., Krishna, R. "Mass Transfer", London: Ellis Horwood, 2nd


ed. (1991).

Ying, W., Weber, W.J. "Bio-Physical Adsorption Model Systems for Wastewater
Treatment", J. Water Poll. Control Fed., 51, 2661, (1979)

Yonge, D., Keinath, T.M., Poznanska, K., Jiang, Z.P. "Single-solute


Irreversible Adsorption on Granular Activated Carbon", Env. Sci. Technol.,l9,
690 (1985).

Yonge, D., Keinath, T.M. "The Effects of Non-Ideal Completion on


Multi-Component Adsorption Equilibria", J. Water Poll. Control Fed., 58,
77 (1986).

Y oung, D .M., Crowell, A.D. "Physical Adsorption of Gases", Butterworth, London


(1962).
LIST OF SYMBOLS*

a activity of component i in equilibrium [-]

-b -b -3(b -b )
A A parameters in the two-component Fritz and [kg 12 kg ll m 12 11]
12' 21
Schlünder isotherm, Eq. (2.16)

-D -B -3(D -B)
1 1 l ]
At' A 2 parameters in extended Fr.& Schl. isotherm, [kg l kg m

Eq. (2.17)

A 2 -1
specific outer surface of the adsorbent [m .kg ]

bll, bl2' exponents in the two-component Fritz and [-]


b b Schlünder isotherm, Eq. (2.16)
21' 22

B 1 'B2 ' exponents in two-component extended Fritz [-]


Dt , D2 and Schlünder isotherm, Eq. (2.17)

can analytically calculated dimensionless bulk


liquid concentration, Chapter (5) [-]

cob' <~) initial concentration in bulk solution [kg.m-3]


(of component i)

eb, (Cb) concentration in bulk solution [kg.m-3]


(of component i)
1
é'i single-solute liquid phase concentration for [moi.r ]

component i, calculated at the spreading pres-


sure of the mixture, Chapter (2)

c. liquid phase concentration for [moi.r


1
]

component i (2)

cnurn numerically calculated dimensionless bulk

* Nwnbers in parentheses refer 10 the chapter or equation in which the syrnbol is used.
When no such reference is given, syrnbols are comrnon 10 several chapters or equations.
Sorne of the symbols which appear only once in the text are not given bere.
lîquid concentration, Chapter (5) [-]

C, (Cp,1.) concentration in pore liquid [kg.m-3]


p
(of component i) at equilibrium

cp,t mass concentration of mixture [kg.m-3]


in pore liquid

ct total molar concentration of mixture [kmol.m-3]

c solubility of solute in water [kg.m-3]


s

ct,s total molar saturation concentration [kmol.m-2]


of surface, Eq. (3.57)

ct,s total mass saturation concentration [kg.m-2]


of surface

d.1 generalized driving force for the motion [m-I]


of species i relative to the mixture

0 app Fick apparent diffusivity [m2.s-l]

0p Fick pore diffusivity [m2.s-I]

0 Fick surface diffusivity [m2s-1]

0 Fick surface diffusivity of carbon loading [m2.s-1]


so
equal to one, Eq. (3.24)
OT Surface diffusivity of carbon loading [m2.s-1]
so
equal to zero, Eq. (6.7)

of Free liquid diffusivity [m2.s-1]

Dapp(D~PP) GMS apparent diffusivity (of component i) [m2.s-I]


I

DP (DP) GMS pore diffusivity (of component i) [m2.s-1]


1

Ds (Ds) GMS surface diffusivity (of component i) [m2.s-1]


I

Ds Ds [m2.s-1]
Î, V ' .I, C GMS surface diffusivity of component v, c
(vacant sites, carbon)

D GMS diffusivity of i-j pair in multicom- [m2.s-1]


IJ
List of symbols

ponent mixture
F constant of Faraday, Eq. (3.28)
-1 -N -3N
parameter (of component i) [kg.kg kg fm t]

in the Freundlich isotherm

J time level number

j p , (j p, .) mass flux (of component i) in pore volume [kg.s-1 .m-2]


1

expressed per m2 shell area

J (j .) mass flux (of component i) on surface [kg.s-1 .m-2]


S S, I

of pore walls expressed per m2 shell area

Kh parameter in the Henri isotherm

kf external mass transfer coefficient

K r, (Kr,1.) parameter (of component i)


in the Radke-Prausnitz isotherm

m amount of adsorbent [mg.l-1]

M total amount of adsorbent (in batch system) [kg]

N.,l N.J molar fluxes of species i, j with respect [kmol.m · 2 s- 1]


to stationary axes

surface molar fluxes of species i, j , v


(vacant sites) and c (carbon walls) with
respect to stationary axes

Nt' (Nt) exponent (of component i) [-]


in the Freundlich isotherm

N, (N .) exponent (of component i) [-]


r r,1
in the Radke-Prausnitz isotherm
s
n.,1 n.1
sn
surface mass flux of species i, as defmed in [kg.m -2 s -1]
equations (3.64) and (3.74), respectively

n.,I n. mass fluxes of species i, j with respect [kg.m -2 s - ']


J
to stationary axes

p system pressure [N.m-2 ]


p fit parameter vector [-]

Q, (Q.) carbon loading or mass solid phase [kg.ki'l


I

concentration (of component i)

• (Q.)
Q, • carbon loading or mass solid phase [kg.ki
1
]
I

concentration (of component i) at equilibrium

Qav average loading at time t [kg.ki'l

Qm monolayer capacity [kg.ki'l

Qo 1
single-solute solid phase concentration for [mol.ki ]
I

component i, calculated at the spreading pres-


sure of the mixture (2)

Q"!I molar solid phase concentration [mol.kg _,]


for component i

1
QT total surface concentration (2) [mol.ki ]

1
Qoo the ultimate loading [kg.ki ]

Qlol total loading of mixture at saturation [kg.ki'J


sal

r radial coordinate [m]

R universa! gas constant [J.mor 1K 1]

R partiele radius [m]


pa

T absolute temperature [K]

U., U. velocity of diffusing species i, j with [m.s- 1]


I J
respect to stationary axes

u• average molar velocity of mixture with [m.s- 1]


respect to stationary axes
List of symbols

V volume of liquid in batch system [m3]

V vacant sites
1
x.I mole fraction of species i [mol.mor ]

xs , xs surface mole fraction of species i, c [mol.mol -I]


c
(carbon walls)

z.I charge of species i [-]

GREEK SYMBOLS

a exponent [-]

-öi, (-öt) fractîonal surface occupancy by component i [-]


(total mixture)

epa partiele porosity [-]

A fraction of the component ultimately adsorbed [-]

Tl (Tl i) interaction parameter (for solute i) [-]

J..l.(p. s) 1
molar (pore, surface) chemica) potential [J.kmor ]
I

of species i

J..l.n positive root of characteristic function,

n spreading pressure (2) [N.m- 1]


no spreading pressure of single solute [N.m- 1]
l

n spreading pressure of mixture, Eq.(2.24) [N.m- 1]


m

pc density of carbon in water [kg.m-3]

ppa partiele density [kg.m-3]


't general tortuosity factor [-]
(j interfacial tension (2), or [N.m- 1]
rate of entropy production (3) [J.m -3s-IK-I]

ai volume fraction of species i [-]

e.o. 1
mass fraction of species i [-]

<I> electrostatic potential [V]

DIMENSIONLESS NUMBERS

Bi Biot number

Fo Fourier number

*Boldface symbols in the text are vectors.

LIST OF ABBREVIATIONS

NBZ nitrobenzene

PNA p-nitroaniline

IPP p-isopropylphenol

SOS sum of squares divided by the number of points

COC concentration decay curve

RSD relative standard deviation

HPMC .homogeneons partiele model constant (diffusi vity)

HPMF homogeneons partiele model Fickian (variable diffusivity)

HPMS homogeneons partiele Maxwell-Stefan (model) (variable diffusivity)

HPMT homogeneons partiele model Transition


Curriculum Vitae
'St/.
Marcho Kouyoumdjiev was bom on September the 13th, 1961 in Sofia, Bulgaria.
From 1976 to 1980 he studied in the French gymnasium in Sofia where he
graduated with cum laude. From 1980 to 1982 he did obligatory military
service. In 1982 he began his study in the Faculty of Organic Chemistry and
Technology at the Technica! University - Sofia. His graduation work was
performed at the Department of Chemical Engineering. In 1987 he passed his
fmal graduation exam with cum laude. From 1987 to 1988 he worked as a
research assistant in the Department of Chemical Engineering · at the Technica!
University- Sofia. From October 1988 until September 1992 he was workingas a
Ph. D. research assistant at the Laboratory of Chernical Process Technology at
Eindhoven University of Technology. The research work described in this thesis
was performed in the group of Prof. dr. ir. PJ.A.M. Kerkhof.
Since september the 1 8 1 , 1992 he is working at Unilever Research Laboratorium
in Vlaardingen.
STELLINGEN
behorende bij het proefschrift van M.S. Kouyoumdjiev

1. Using Maxwell-Stefan instead of Fickian defmitions for the surface flux in


the adsorbent partiele is thermodynamically more correct and gives a
diffusion coefficient which is much less concentration dependent

This thesis, chapter 6

2. Literature data on adsorption from activated carbon should be treated with


great caution as it is often not clear what are the exact physical
properties of the carbon, what experimental procedure has been foliowed and
what is the reproducibility of the results.

This thesis, chapters 3,5

3. The influence of thermodynamic activity should receive more attention in


the modeling of mass transport processes during ultrafiltration of
multicomponent systems.

N. Papamichael, M.-R. Kula J. Membrane Sci. 30 ( 1987) 259 - 272

4. The moisture dependenee of the mechanica] properties of the material is


essential in the modeling of stress. This phenomena was unjustly neglected
by Hasatani et al.

Ketelaars A.AJ., 'Drying deformable media: kinetics, shrinkage and


stresses', Ph.D. Thesis, Eindhoven University of Technology, chapter 6,
1992.

Hasatani M. and Y. ltaya, 'Deformation characteristic of ceramics during


drying', Drying '92, Ed. A.S. Mujumdar, Elsevier Science Publisher,
Amsterdam, 190 - 202, 1992.
5. Models of fluidized beds using deterministic descriptions of the
hydrodynamics of this system will always give a varianee of about 20%. The
reason is that bubble dynamics are more stochastic than deterministic,
imp~ying that in order to decrease the varianee stochastic models have to

be used or developed.

C. van Lare 'Mass Transfer in Gas Fluidized Beds: Sealing, Modeling and
Partiele Size lnfluence', Ph.D. Thesis, Eindhoven University of
Technology, chapter 7, 1991.

6. Different people like different music. What IS really universa! is the


passion for music, not the music itself.

7. Even a mere "Were are here, too" answer from space would, by its very
existence, offer hope that advanced civilizations need not self-destruct
with their nuclear toys. If they made it, maybe we can, too.

8. The conservalion of nature will succeed only if it takes place in tandem


with economie growth, not at the expense of growth.

9. Once again the volatile Balkans could provide the spark for conflicts
engulfmg millions of Europeans.

10. The GATI trade talks now oppose Europe to America but the underlying
conflict seems to be between French farmers and the rest of the world.

11 All major corporations are shifting their activities in the direction of


core technologies. Those companies that are already strong in core
technologies will specialire in these fields and hardly diversify in other
areas.

You might also like