You are on page 1of 279

NOVEL MODEL FORMULATION FOR

VIRUS REMOVAL BY TANGENTIAL FLOW


ULTRAFILTRATION
By

Angayar Kumari Pavanasam

A thesis submitted in fulfilment of the requirement for the

degree of

Doctor of Philosophy

School of Chemical Engineering

The University of New South Wales

Sydney 2052 AUSTRALIA

March 2011
STATEMENT OF ORIGINALITY

‘I hereby declare that this submission is my own work and to the best of my knowledge

it contains no materials previously published or written by another person, or substantial

proportions of material which have been accepted for the award of any other degree or

diploma at UNSW or any other educational institution, except where due acknowledge-

ment is made in the thesis. Any contribution made to the research by others, with whom

I have worked at UNSW or elsewhere, is explicitly acknowledged in the thesis. I also

declare that the intellectual content of this thesis is the product of my own work, except

to the extent that assistance from others in the project’s design and conception or in style,

presentation and linguistic expression is acknowledged.’

COPYRIGHT STATEMENT

‘I hereby grant the University of New South Wales or its agents the right to archive and to

make available my thesis or dissertation in whole or part in the University libraries in all

forms of media, now or here after known, subject to the provisions of the Copyright Act

1968. I retain all proprietary rights, such as patent rights. I also retain the right to use in

future works (such as articles or books) all or part of this thesis or dissertation.

I also authorise University Microfilms to use the 350 word abstract of my thesis in

Dissertation Abstract International. I have either used no substantial portions of copyright

material in my thesis or I have obtained permission to use copyright material; where per-

mission has not been granted I have applied/will apply for a partial restriction of the digital

copy of my thesis or dissertation’.


AUTHENTICITY STATEMENT

‘I certify that the Library deposit digital copy is a direct equivalent of the final officially

approved version of my thesis. No emendation of content has occurred and if there are any

minor variations in formatting, they are the result of the conversion to digital format.’

Signed: .......................... Date: .........


Dedicated to

My teachers &

Mom - Gnanam, Dad - Pavanasam, Sister - Selvi, & Brother - Prabhu


ACKNOWLEDGEMENTS

The journey of Ph.D. is a constant discovery and transformation of one’s untapped potential

and ability in the chosen field. There are many instrumental persons who made my journey

a memorable one. I am extremely glad to thank my supervisor Professor Vicki Chen for

the total freedom she bestowed on me and for her timely guidance. Her support at the

critical time helped me a lot to push myself to complete on time.

The continued discussions with Dr. Ali Abbas, my supervisor at the University of Sydney,

about population balance theory was thought provoking which got deciphered into this

thesis. His constant motivation, ever-encouraging words, and access to gPROMS made it

possible for me to take every step ahead. He helped me immensely in analyzing, inter-

preting and presenting the results precisely. My visits to University of Sydney was always

like visiting my home. This is not just for the sake of acknowledging but it is a heartiest

thankyou to Dr. Ali. He remained a constant source of inspiration since the day one of this

journey.

I would like to thank A/Professor Greg Leslie, Dr. Pierre LeClech, and Professor Rose

Amal for being a part of my review committee and for their valuable comments which

helped in shaping up the thesis. I would like to thank the UNESCO Center for Membrane

Science and Technology and the School of Chemical Engineering for giving me the great

opportunity to learn and prosper.

Thanks Eisham for the cross flow rig without which I would have struggled to carry out

the experiments. Maylim, thankyou so much for the training on particle size measurement.

I would like to thank Dorothy and Rebaya for helping me in ICP analysis and Katie for
ii

the imaging analysis. Thanks to all my UNESCO membrane mates. Very special thanks to

Dr. Alice Antony for the lunch time discussions, lunches and constant calls - I owe a lot to

her especially for the initial settling-in recipes. I would like to thank all the technical and

administrative staff of the Chemical Engineering School in general and Deyan in particular

for facilitating my access to laboratory equipments and the chemicals, and Ling for her

timely reminders regarding the reviews and courses.

The time and support offered by Sashi during the illness time, I cannot thank her enough

for that. The regular reminders about my deadline always came from my friends. Thanks

to Chandra and Simit for their kind and moral support and Kiet for the computer support.

Ofcourse, my special thanks to Raghav and Jose from Univ. of Sydney for their thoughtful

help throughout - it meant a lot. I am running short of words to express my sincere gratitude

to my friends - distance never mattered - Prof. Raman (India), Prof. Raghavan (India), Anu

(US) and Suganthi (US), Dr. Sundar (US), Dr. Bala (Germany), Dr. Kumaran (India), and

Balaji (Dubai) for their words of confidence and motivation. Balaji (Singapore), eternal

thanks for those (silent) words of wisdom and memories.

It is a heavenly bliss to have two wonderful and dedicated souls as my parents. My mother

known for her intelligence and my father a superb human being and for instilling the sense

of discipline and conviction, I am blessed to be their daughter. My sister Selvi and brother

Prabhu are my constant source of inspiration and lot more. Thanks for being very patient

with all my tantrums.

My namaskarams with thanks to the almighty for showering only good things on me!.

ANGAYAR KUMARI PAVANASAM

August 2010
ABSTRACT

Products of human and animal origin (biologicals and pharmaceuticals) and drinking water

pose unique health and safety problems, especially due to viral contaminations, and have

traditionally been a key concern both for regulators and industrial operations. They have

also proven to be a stumbling block for early product research and development and inex-

perienced small start-up firms. Membrane filtration, being non-invasive, non-destructive

and a robust technique, is one of the most preferred choices for virus filtration with ul-

trafiltration being an efficient process for removing macromolecules, colloids, endotoxins,

viruses, and bacteria.

The challenge, in virus filtration, is to remove virus particles from products to close to

zero levels especially in emerging water and heavily regulated multi-billion biopharmaceu-

tical industries. In virus capture or clearance operations, the virus particles are distributed

in the range of 15nm - 300nm depending upon their protein coat and family, hence it is crit-

ical to study impact of particle size in ultrafiltration. A number of previous studies have

been concerned with particle characteristics and their relation to flux decline behavior.

However the relationship between, on one hand, particle properties of the feed and operat-

ing parameters like transmembrane pressure and cross flowrate and on the other hand, the

filtration efficiency (log reduction value), are still not adequately established.

This thesis investigates the effect of feed particle size, transmembrane pressure and

cross flowrate on the filtration efficiency for tangential flow ultrafiltration. An empirical

model is developed to quantify the dependencies of the above said parameters. For this

study, a cross flow ultrafiltration rig (built in-house) is used with 30 and 100 kDa polyether
iv

sulphone membranes for three different particle sizes of silica (model virus particles) at

three different transmembrane pressures (20-60 kPa) and three cross flowrates (0.3-1.0

L/min). The investigation shows that among feed particle size, transmembrane pressure

and cross flowrate, feed particle size and transmembrane pressure are significant param-

eters in controlling the filtration efficiency. In the studied experimental range, higher log

reduction values are obtained at lower transmembrane pressure (20 kPa) and at higher

cross flowrate (1 L/min). The experimental statistical analysis also gives an insight on the

cross interactions of the feed particle size, transmembrane pressure and cross flowrate. It

shows that the effect of transmembrane pressure and feed particle size seems to be more

significant than that of the interaction of cross flowrate with feed particle size.

The experimental investigations identified that feed particle size is one of the key in-

fluential factors on the filtration efficiency. Far less attention is given to this effect in the

literature both experimentally and theoretically. This thesis attempts to fill this gap by

developing a rigorous model formulation for tangential flow ultrafiltration to predict the

evolution of particle size in the filtered stream. Population balance theory is employed to

describe the population density of each particle size class in the output streams. In this

formulation, the population balance equation is coupled with filtration kinetic constitutive

relations and with mass balance equations and solved using discretisation method. The

model predicts permeate particle size distribution and the log reduction value as primary

model outputs. A novel approach in the form of optimization-based parameter estimation

is employed to estimate filtration kinetic parameters. This approach which directly uses

the population balance model, as well as particle size data, represents a stark deviation

from previous filtration kinetic calculations that only utilize flux data and no models that
v

describe the particle state. As such, this approach offers more comprehensive description

of filtration kinetics incorporating the effects of particle size. The model is validated and is

found to be in good agreement with experimental results. The model serves as a predictive

tool for filtration efficiency and filtrate particle size distribution. Overall, this model-based

prediction and estimation capability serves well and facilitates the design, operation and

scale-up of ultrafiltration processes.

Keywords: Ultrafiltration, population balance, virus filtration


AUTHOR’S PUBLICATIONS

Publications based on this work

Refereed Journal Articles


(i) Angayar K Pavanasam, Ali Abbas, "Ultrafiltration and Virus Removal: A Mini Re-

view of Recent Patents", Recent Patents on Chemical Engineering, 151-156, 1(2),

2008.

(ii) Angayar K Pavanasam, Ali Abbas, Vicki Chen, "Influence of Particle Size and Oper-

ating Parameters on Virus Ultrafiltration Efficiency", Water Science and Technology:

Water Supply, 31-38, 11(1), 2011.

(iii) Angayar K Pavanasam, Ali Abbas, Vicki Chen, "An Empirical Modeling of Particle

Size and Polydispersity of Model Virus Filtration", Biotechnology and Bioprocess

Engineering (Submitted).

(iv) Angayar K Pavanasam, Ali Abbas, "Predictive Model Formulation and Simulation of

Tangential Flow Ultrafiltration: A Novel Approach", Chemical Engineering Science

(Under review for submission).

Conference Articles
(v) Angayar K Pavanasam, Ali Abbas, Santosh Ansumali, Vicki Chen, "A Population

Balance Model for Ultrafiltration Processes", Engineering With Membranes (EWM

2008), Portugal, May 2008.


vii

(vi) Angayar K Pavanasam, Ali Abbas, Santosh Ansumali, Vicki Chen, "Modeling virus

ultrafiltration processes: A population balance approach, International Congress on

Membranes and Membrane Processes (ICOM 2008),Hawaii, July 2008.

(vii) Angayar K Pavanasam, Ali Abbas, Vicki Chen, "Influence of particle size and oper-

ating parameters on virus ultrafiltration efficiency, International Water Association -

Membrane Technology Conference (IWA-MTC 2009), Beijing, September 2009.

(viii) Angayar K Pavanasam, Ali Abbas, Vicki Chen, "Modeling Particle Size in Ultrafil-

tration, EuroMembrane (EM 2009), France, September 2009.

(ix) Angayar K Pavanasam, Ali Abbas, Vicki Chen, "Particle Size in Membrane Filtra-

tion: Novel Strategy for Better Insights", Membrane Society of Australasia - Stu-

dents Conference, Wollongong, Australia, February 2010.

Award Articles
(x) Angayar Kumari Pavanasam, Ali Abbas, Vicki Chen, "Novel Model Formulation for

Virus Removal by Tangential Flow Ultrafiltration", Dean’s Award for Excellence in

Postgraduate Research, University of New South Wales, Sydney, November 2008.

(xi) Angayar Kumari Pavanasam, Ali Abbas, Vicki Chen, "Population Balance Modeling

Approach: Better Insights into Virus Removal Mechanism", Dean’s Award for Ex-

cellence in Postgraduate Research, University of New South Wales, Sydney, October

2009.
GLOSSARY

AeDNV Aedes aegypti DensoNucleosisVirus

AF4 Asymmetric Flow Field Flow Fractionation

ALV Avian Leukemia Virus

ATCC American Type Culture Collection

BHK Born Hamster Kidney cells

BVDV Bovine Viral Diarrhea Virus

B19 Human Parvovirus

CFR Cross Flow Rate

CHO Chinese Hamser Ovary

CJD CreutzfeldJacob disease

CMR Comprehensive Microbial Resource

CPV Canine Parvo Virus

DEF/NFF Dead End Filtration/Normal Flow Filtration

DLS/MALS Dynamic Light Scattering/MultiAngle Light Scattering

EBV Epstein-Barr Virus (also known human herpesvirus 4)

EMC Encephalo MyoCarditis

FCS Fetal Calf Serum

FDA Food and Drug Administration

FFF Field Flow Fractionation

FPS Feed Particle Size


ix

GMP Good Manufacturing Practices

HAV Hepatitis A Virus

HBV Hepatitis B Virus

HCV Hepatitis C Virus

HIV Human Immunideficiency Virus

HPTFF High Performance Tangential Flow Filtration

HSA Human Serum Albumin

ICH International Conference on Harmonization

IgG Immunoglobulin G

LRV/LTR Log Reduction Value/Log Titre Reduction

MAB Monoclonal AntiBody

MLV Murine Lukemia Virus

MVM Minute Virus of Mice

MWCO Molecular Weight Cut Off

PBS Phosphate Buffered Saline

PCR Polymerase Chain Reaction

PPV Porcine Parvo Virus

PSD Particle Size Distribution

PSR PSeudo Rabies Virus

PVDF Poly Vinylidene Fluoride

RTP Recombinant Therapeutic Protein

SV40 Simian Virus 40

TFF/CFF Tangential Flow Filtration/Cross Flow Filtration


x

TMP Transmembrane Pressure

UF Ultrafiltration

USP United States Pharmacopeia

VSV Vesicular Stomatis Virus

Symbols
A - Membrane area, m2

B, G - Nucleation and growth rates, #/m3 min

C f - Feed concentration, weight of solute per weight of the suspension, w/w%

C p - Permeate concentration - weight of solute per weight of the suspension, w/w%

Cr - Retentate concentration, weight of solute per weight of the suspension, w/w%

D(t) - Permeate particle size, m

F - Filtration kinetics

F f - Feed flow rate, L/min

F p - Permeate flow rate, L/min

Fr - Retentate flow rate, L/min

k1 , k2 , k3 - Kinetic parameters

L - Membrane thickness, m

La - Average particle size, nm

L pa - Particle size, nm

L po - Average pore size, nm

N - Number of measurements taken during experiment

NE - Number of experiments

NMi j - Number of measurements of the jth variable in the ith experiment


xi

NVi - Number of variables measured in the ith experiment

n p - Particle number density in permeate stream, #/m3

nr - Particle number density in retentate stream, #/m3

n f - Particle number density in feed stream, #/m3

P pa - Probability of particle permeation

t - Time, min

U - Manipulated variables

V - System volume, m3

σ - Standard deviation

σ2i jk - Variance of kth measurement of variable j in experiment i

ω2i jk - Standard deviation of the measurement error

θ - Vector of parameters

φ - Objective function
Contents

ACKNOWLEDGEMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . i

ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii

AUTHOR’S PUBLICATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . vi

GLOSSARY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii

1 Introduction 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Literature Review 7

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.2 Virus Contamination . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.2.1 In the context of biopharmaceutical industry . . . . . . . . . . . . 12

2.2.2 In the context of water industry . . . . . . . . . . . . . . . . . . 14

2.3 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.4 Regulatory Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2.4.1 Regulations governing biopharmaceutical industry . . . . . . . . 16

2.4.2 Regulations governing water industry . . . . . . . . . . . . . . . 20

2.5 Virus Filtration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.5.1 Membrane modes and configurations . . . . . . . . . . . . . . . 23

2.5.1.1 Normal flow filtration . . . . . . . . . . . . . . . . . . 23

2.5.1.2 Tangential flow filtration . . . . . . . . . . . . . . . . . 26


Contents xiii

2.5.1.3 High performance tangential flow filtration . . . . . . . 28

2.5.1.4 Field flow fractionation . . . . . . . . . . . . . . . . . 29

2.5.1.5 Remarks on membrane modes and configurations . . . 31

2.5.2 Membrane modules . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.5.2.1 Flat sheet . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.5.2.2 Spiral wound . . . . . . . . . . . . . . . . . . . . . . . 34

2.5.2.3 Hollow fibre . . . . . . . . . . . . . . . . . . . . . . . 34

2.5.3 Particle characterization . . . . . . . . . . . . . . . . . . . . . . 36

2.5.3.1 Dynamic light scattering . . . . . . . . . . . . . . . . . 36

2.5.3.2 Imaging and image analysis . . . . . . . . . . . . . . . 41

2.5.3.3 Asymmetrical flow field flow fractionation and multi-

angle light scattering . . . . . . . . . . . . . . . . . . . 42

2.5.3.4 Electro-spray differential mobility analysis . . . . . . . 42

2.6 Economics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

2.7 Patent Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

2.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

3 Experimental Section 68

3.1 Membrane Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.2 Characterization Techniques . . . . . . . . . . . . . . . . . . . . . . . . 72

3.2.1 Particle concentration measurement by inductively coupled

plasma technique . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.2.2 Particle morphology studies by transmission electron microscopy 74


Contents xiv

3.2.3 Particle size and distribution studies by dynamic light scattering

technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

3.3 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

4 Influence of Particle Size and Size Distribution on Filtration Efficiency and

Separation Mechanism 82

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

4.2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

4.3 Cross-Flow Filtration Experimental Details . . . . . . . . . . . . . . . . 89

4.4 Results and Discussion: Filtration Efficiency Study . . . . . . . . . . . . 89

4.4.1 Experimental design and surface response modeling . . . . . . . 90

4.4.2 Pearson’s correlation analysis . . . . . . . . . . . . . . . . . . . 92

4.4.3 Model prediction results . . . . . . . . . . . . . . . . . . . . . . 95

4.4.4 Effect of process parameters and input condition on filtration effi-

ciency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

4.4.4.1 Effect of TMP on filtration efficiency . . . . . . . . . . 98

4.4.4.2 Effect of CFR on filtration efficiency . . . . . . . . . . 99

4.4.4.3 Effect of particle and pore size on filtration efficiency . 101

4.4.5 Concluding remarks for filtration efficiency study . . . . . . . . . 105

4.5 Results and Discussion - Influence of Particle Size and Size Distribution

on Filtration Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . 107

4.5.1 Study of particle size characterization based on particle concentration109

4.5.2 Experimental design and statistical analysis . . . . . . . . . . . . 112


Contents xv

4.5.3 Model prediction results . . . . . . . . . . . . . . . . . . . . . . 116

4.5.3.1 Effect of TMP on permeate particle size and polydisper-

sity index . . . . . . . . . . . . . . . . . . . . . . . . 116

4.5.3.2 Effect of CFR on permeate particle size and polydisper-

sity index . . . . . . . . . . . . . . . . . . . . . . . . 117

4.5.4 Particle size distribution (PSD) analysis . . . . . . . . . . . . . . 119

4.5.4.1 Temporal effect . . . . . . . . . . . . . . . . . . . . . 119

4.5.4.2 Effect of TMP on particle size distribution in permeate

streams . . . . . . . . . . . . . . . . . . . . . . . . . . 120

4.5.4.3 Effect of CFR on particle size distribution in permeate

streams . . . . . . . . . . . . . . . . . . . . . . . . . . 123

4.5.4.4 Modeling particle size distribution . . . . . . . . . . . 125

4.5.4.5 Size polydispersity index and its effects . . . . . . . . . 126

4.5.5 Concluding remarks for particle size and distribution investigation

section . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

5 Modeling and Simulation 137

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

5.2 Earlier Modeling patterns . . . . . . . . . . . . . . . . . . . . . . . . . . 142

5.2.1 Macro-scale modeling: A quick review . . . . . . . . . . . . . . 143

5.2.1.1 Flux decline prediction models . . . . . . . . . . . . . 145

5.2.2 Particle property models . . . . . . . . . . . . . . . . . . . . . . 147

5.3 Novel Model Formulation: Population Balance Approach . . . . . . . . . 150


Contents xvi

5.3.1 Statement of the problem . . . . . . . . . . . . . . . . . . . . . . 151

5.3.2 Modeling of size distribution of suspended particles through a

cross-flow UF process . . . . . . . . . . . . . . . . . . . . . . . 153

5.3.2.1 Input and output streams . . . . . . . . . . . . . . . . 154

5.3.2.2 Constitutive equations . . . . . . . . . . . . . . . . . . 155

5.3.2.3 Permeation probability . . . . . . . . . . . . . . . . . 156

5.3.2.4 Expression of solid permeate flow . . . . . . . . . . . . 157

5.3.2.5 Balance equations . . . . . . . . . . . . . . . . . . . . 157

5.4 Model Solution and Simulation Results . . . . . . . . . . . . . . . . . . 160

5.4.1 Discretisation of the model . . . . . . . . . . . . . . . . . . . . . 161

5.4.1.1 Discretisation issues . . . . . . . . . . . . . . . . . . . 161

5.4.1.2 The internal coordinate and its significance . . . . . . . 163

5.4.1.3 The method of moments . . . . . . . . . . . . . . . . . 166

5.4.1.3.1 Different mean sizes . . . . . . . . . . . . . . 169

5.4.1.4 Number and volume distributions . . . . . . . . . . . . 171

5.4.1.5 Useful quantity predicted from the above model . . . . 171

5.4.2 Simulation using gPROMS . . . . . . . . . . . . . . . . . . . . . 172

5.4.2.1 Model entities of the entire process . . . . . . . . . . . 173

5.4.2.2 Parameters and variables employed in the model devel-

opment . . . . . . . . . . . . . . . . . . . . . . . . . . 174

5.4.3 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . 174

5.4.3.1 Representative particle size distribution in output streams 174

5.4.3.2 Effect of feed particle size on particle size distribution . 177


Contents xvii

5.4.3.3 Effect of transmembrane pressure drop on the particle

size distribution . . . . . . . . . . . . . . . . . . . . . 178

5.4.3.4 Effect of cross flow rate on particle size distribution . . 178

5.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

6 Optimization-based Parameter Estimation 191

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192

6.2 Overview of Parameter Estimation . . . . . . . . . . . . . . . . . . . . . 196

6.3 Filtration Kinetic Parameter Estimation . . . . . . . . . . . . . . . . . . 197

6.4 Estimation Results and Discussion . . . . . . . . . . . . . . . . . . . . . 200

6.4.1 Parameter estimation summary . . . . . . . . . . . . . . . . . . . 200

6.4.2 Overlay plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202

6.5 Model Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202

6.5.1 Validation against literature data . . . . . . . . . . . . . . . . . . 208

6.6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211

7 Conclusions and Future Directions 214

7.1 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214

7.2 Recommendations for Future Directions . . . . . . . . . . . . . . . . . . 218

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
List of Figures

1.1 A typical bioprocess flowchart with indicative operations . . . . . . . . . 2

2.1 Different types of virus particles giving a snapshot of size and shape of

viruses (Pic. courtesy: antibac2k.com, iah.bbsrc.ac.uk.com, innovations-

report.de). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.2 Virus Clearance is a critical step in biotechnolgical/biopharmaceutical

manufacturing. It ensures removal of viruses found to have contaminated

the bioprocess. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.3 Schematic representation of dead end flow (NFF), cross flow (TFF) and

flow field fractionation (FFF) modes. . . . . . . . . . . . . . . . . . . . . 24

2.4 TFF modules for small and large scale applications. (Courtesy: Millipore

Corporation) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2.5 Particle size distribution of high molecular weight, low molecular weight,

and virus like particle (MIX) solutions by DLS [Lipin et al., 2008]. . . . . 38

2.6 Particle size distribution in the retentate (right) and permeate (left) streams

through 0.1 μ membrane during the study of HAV by TFF system. . . . . 39

2.7 Particle size distribution in the retentate (right) and permeate (left) streams

through 0.2 μ membrane during the study of HAV by TFF system. . . . . 40

2.8 Particle size distribution of diethylaminoethyl (DEAE) and packaged virus

like particles by dynamic light scattering technique. . . . . . . . . . . . . 40


List of Figures xix

2.9 Membrane filtration’s place in the Pharmaceutical/Biotech industry -

World Market Share . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

3.1 Tangential flow UF experimental setup employed for this study . . . . . . 69

3.2 An outline of various methods employed for the analysis of silica samples. 72

3.3 TEM Micrograph of small silica particles at x200K magnification. . . . . 74

3.4 TEM Micrograph of medium silica particles at x100K magnification. . . . 75

3.5 TEM Micrograph of big silica particles at x200K magnification. . . . . . 75

3.6 Schematic diagram depicting the main components in a typical dynamic

light scattering setup. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

4.1 Schematic diagram of the tangential flow filtration experimental setup. . . 88

4.2 Cross flow filtration: Sample flowing parallel to the membrane surface. . . 89

4.3 Log reduction values: predicted vs experimental. . . . . . . . . . . . . . 94

4.4 Rankit Plot showing the residuals are normally distributed. . . . . . . . . 94

4.5 Model prediction showing the effect of cross flowrate at 20 kPa on LRV. . 95

4.6 Model prediction showing the effect of cross flowrate at 40 kPa on LRV. . 96

4.7 Model prediction showing the effect of cross flowrate at 60 kPa on LRV. . 96

4.8 Model prediction vs experimental trend for varying TMP at 0.6 L/min. . . 97

4.9 Experimental results showing the effect of TMP at 0.3 L/min on LRV (lines

drawn to indicate trends). . . . . . . . . . . . . . . . . . . . . . . . . . . 99

4.10 Experimental results showing the effect of TMP at 0.6 L/min on LRV (lines

drawn to indicate trends). . . . . . . . . . . . . . . . . . . . . . . . . . . 100


List of Figures xx

4.11 Experimental results showing the effect of TMP at 1 L/min on LRV (lines

drawn to indicate trends). . . . . . . . . . . . . . . . . . . . . . . . . . . 100

4.12 Model prediction showing the effect of TMP at 0.6 L/min on LRV. . . . . 101

4.13 Experimental results showing the effect of feedflow rate at 40 kPa on LRV

(lines drawn to indicate trends). . . . . . . . . . . . . . . . . . . . . . . . 102

4.14 A comparison of experimental results (points joined by dotted lines to in-

dicate the trend) with our model and Ferry-Faxen model. . . . . . . . . . 103

4.15 Efficiencies (LRV) for the two MWCO membranes as a function of the

PPSR. Data for lower TMP (20 kPa) and higher flowrate (1 L/min). . . . 105

4.16 Two different representation of particle size distribution of the original

feed sample: multimodal size distribution (top) and log normal distribu-

tion (bottom). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

4.17 Illustration of the correlation function - Exponential decay of auto-

correlation function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

4.18 Counts per second versus concentration plot for three different particles.

This shows the range of suitable concentration below 800 kcps threshold

as specified by the BIC particle sizer. . . . . . . . . . . . . . . . . . . . . 111

4.19 Observed vs Predicted: Permeate Particle Size. . . . . . . . . . . . . . . 115

4.20 Observed vs Predicted: Permeate PDI. . . . . . . . . . . . . . . . . . . . 116

4.21 Experimental (points joined by dashed lines to indicate the trend) Vs

Model Prediction: Effect of TMP at the middle CFR (0.6 L/min) on the

permeate particle size. . . . . . . . . . . . . . . . . . . . . . . . . . . . 117


List of Figures xxi

4.22 Experimental (points joined by dashed lines to indicate the trend) Vs

Model Prediction: Effect of TMP at the middle CFR (0.6 L/min) on the

permeate PDI. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

4.23 Experimental (points joined by dashed lines to indicate the trend) Vs

Model Prediction: Effect of CFR at the middle TMP (40 kPa) on the per-

meate particle size. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

4.24 Experimental (points joined by dashed lines to indicate the trend) Vs

Model Prediction: Effect of CFR at the middle TMP (40 kPa) on the per-

meate PDI. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

4.25 Representative distribution of feed stream sample of all three particles. . . 120

4.26 An illustration of the particle size distribution evolution in the permeate

stream at TMP=20 kPa. . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

4.27 An illustration of the particle size distribution evolution in the permeate

stream at TMP=60 kPa. . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

4.28 Particle size distribution of the permeate stream at various TMPs (20-60

kPa), time=9 mins, CFR=0.6 L/min. . . . . . . . . . . . . . . . . . . . . 122

4.29 Particle size distribution of the retentate stream at various TMPs (20-60

kPa), time=9 mins, CFR=0.6 L/min. . . . . . . . . . . . . . . . . . . . . 123

4.30 Particle size distribution of the permeate stream at various CFRs (0.3-1

L/min), time=9 mins, TMP=40 kPa. . . . . . . . . . . . . . . . . . . . . 124

4.31 Particle size distribution of the retentate stream at various CFRs (0.3-1

L/min), time=9 mins, TMP=40 kPa. . . . . . . . . . . . . . . . . . . . . 124


List of Figures xxii

4.32 Representative particle size distribution of observed (points joined by

dotted lines to show the trend) and predicted plots for smaller perme-

ate particles at (a)-(c) at TMP=20,40,60 kPa and 0.3 L/min (d)-(f)at

TMP=20,40,60 kPa and 0.6 L/min (g)-(i)at TMP=20,40,60 kPa and 1 L/min.127

4.33 Representative particle size distribution of observed (points joined by

dotted lines to show the trend) and predicted plots for medium perme-

ate particles at (a)-(c) at TMP=20,40,60 kPa and 0.3 L/min (d)-(f)at

TMP=20,40,60 kPa and 0.6 L/min (g)-(i)at TMP=20,40,60 kPa and 1 L/min.128

4.34 Representative particle size distribution of observed (points joined by

dotted lines to show the trend) and predicted plots for bigger perme-

ate particles at (a)-(c) at TMP=20,40,60 kPa and 0.3 L/min (d)-(f)at

TMP=20,40,60 kPa and 0.6 L/min (g)-(i)at TMP=20,40,60 kPa and 1 L/min.129

4.35 Effect of feed and permeate PDIs on filtration efficiency at middle TMP

(40 kPa) and CFR (0.6 L/min). . . . . . . . . . . . . . . . . . . . . . . . 130

5.1 Illustrations of (a) dead-end filtration and (b) Tangential flow filtration. . . 142

5.2 A schematic diagram of the tangential flow filtration: Particles sieving

through the membrane (permeate) and those being excluded (retentate). . 152

5.3 A schematic diagram of the membrane filtration and its relationship with

the proposed PBE model. . . . . . . . . . . . . . . . . . . . . . . . . . . 152

5.4 An overview of the model encapsulating the set of balances involved in the

model formulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153


List of Figures xxiii

5.5 Plot showing the behavior of the probability of particles permeating

through the membrane: Smaller the particles higher the probability and

vice-versa. Also the plot shows that the particles approaching membrane

pore reflects zero probability of particle permeation. . . . . . . . . . . . . 158

5.6 Solid permeate flow indicating the bounding condition of no particle flow

of bigger particles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

5.7 The discretisation of the density function and the relationship between size

class, and density. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

5.8 Plots indicating the significance of the geometric constant (a-c) and num-

ber of size classes (d-f) for the model discretisation. . . . . . . . . . . . . 167

5.9 Representative plot indicating the relation of particle size and the number

of size classes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168

5.10 Representative particle size distribution in the permeate at TMP=90 kPa

and CFR=0.045 L/min. . . . . . . . . . . . . . . . . . . . . . . . . . . . 176

5.11 Representative particle size distribution in the retentate at TMP=90 kPa

and CFR=0.045 L/min. . . . . . . . . . . . . . . . . . . . . . . . . . . . 176

5.12 Particle size distribution of three range of feed particle sizes. . . . . . . . 177

5.13 Particle size distribution of permeate for three feed particle size ranges. . 177

5.14 Permeate particle size distribution with changing transmembrane pressure

at CFR=0.045 L/min. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178

5.15 Retentate particle size distribution with changing transmembrane pressure

at CFR=0.045 L/min. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179


List of Figures xxiv

5.16 Permeate particle size distribution with changing crossflow rate at

TMP=90 kPa. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180

5.17 Retentate particle size distribution with changing crossflow rate at

TMP=90 kPa. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180

6.1 Flowchart showing the various steps involved in parameter estimation:

pre-estimation and post-estimation checks in arriving at the accurate values. 195

6.2 Comparison of measured values and predicted values during the parameter

estimation step. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203

6.3 Comparison of measured values and predicted values during the parameter

estimation step. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204

6.4 Simulation results of permeate particle size distribution for bigger sized

particles (140nm avg.) at TMP=40 kPa and CFR=0.6 L/min. . . . . . . . 205

6.5 Simulation results of retentate particle size distribution for bigger sized

particles (140nm avg.) at TMP=40 kPa and CFR=0.6 L/min. . . . . . . . 205

6.6 Simulation results of permeate particle size distribution for medium sized

particles (80nm avg.) at TMP=40 kPa and CFR=0.6 L/min. . . . . . . . . 206

6.7 Simulation results of retentate particle size distribution medium sized par-

ticles (80nm avg.) at TMP=40 kPa and CFR=0.6 L/min. . . . . . . . . . 206

6.8 Model predictions vs. experimental results of filtration efficiency at

TMP=40 kPa and CFR=0.6 L/min. . . . . . . . . . . . . . . . . . . . . . 207

6.9 Model predictions vs. experimental results of permeate particle size for all

three particle size ranges at TMP=40 kPa and CFR=0.6 L/min. . . . . . . 208
List of Figures xxv

6.10 Model predictions vs. experimental results of permeate particle size distri-

bution at TMP=40 kPa and CFR=0.6 L/min of bigger sized particles. . . . 209

6.11 Model predictions vs. experimental results of permeate particle size distri-

bution at TMP=40 kPa and CFR=0.6 L/min of medium sized particles. . . 209

6.12 Model predictions vs. experimental results of permeate particle size distri-

bution at TMP=40 kPa and CFR=0.6 L/min of smaller sized particles. . . 210
List of Tables

2.1 Virus Characteristics: Type, family and size used in virus clearance study. 11

2.2 Reported incidents of virus contamination. . . . . . . . . . . . . . . . . . 13

2.3 Virus contamination and their cell lines in the production of biological

products. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.4 Virus contamination and their sources resulting in a possible outbreak in

drinking water. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.5 Timeline of global regulations for viral safety of licensed products. . . . . 19

2.6 Summary of various operating modes of FFF. . . . . . . . . . . . . . . . 31

2.7 Typical cross flowrates for various membrane configurations. . . . . . . . 31

2.8 Summary of commercially available membrane modules for virus removal. 35

2.9 Summary of various analytical techniques available for virus particle char-

acterization studies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

2.10 A mini review of patents on the use of UF for the removal of virus particles. 47

3.1 Characteristics of Silica Colloid. . . . . . . . . . . . . . . . . . . . . . . 71

4.1 Full Factorial, 3 level experimental design for cross flow UF of silica par-

ticles using 100 kDa MWCO membrane (b : coded variables). . . . . . . . 91

4.2 Pearson’s correlation analysis. . . . . . . . . . . . . . . . . . . . . . . . 93

4.3 Analysis of variance summary. . . . . . . . . . . . . . . . . . . . . . . . 93


List of Tables xxvii

4.4 Efficiencies (LRV) for the two MWCO membranes against particles of

different size ranges. Data for lower TMP (20 kPa) and higher flowrate (1

L/min). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

4.5 Full Factorial, 3 level experimental design for cross flow UF of three par-

ticle size ranges of silica colloid. . . . . . . . . . . . . . . . . . . . . . . 113

4.6 Summary of Coefficients, Analysis of variance (ANOVA). . . . . . . . . 114

5.1 Summary of various models considering feed particle size as a factor to

study the permeate flux. . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

5.2 Summary of the partitioned sections available for the model entity. . . . . 175

5.3 Simplified summary of the partitioned sections available for the process

entity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

5.4 Input data for the simulation of the model. . . . . . . . . . . . . . . . . . 175

5.5 Various input conditions utilised for the model simulations. . . . . . . . . 175

5.6 LRV at various TMPs and crossflow rates of smaller particles (80 nm avg.). 180

6.1 The types of parameter estimation problems. . . . . . . . . . . . . . . . . 194

6.2 Specification of the variance model in the gEST file. . . . . . . . . . . . . 200

6.3 Parameter estimation summary showing the number of experiments used

for estimating the parameters, the variables D(t) and C(t) measured, vari-

ance model employed and the objective function contribution. . . . . . . 201

6.4 Model parameter estimates. . . . . . . . . . . . . . . . . . . . . . . . . . 201

6.5 Correlation matrix. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202

6.6 Lack of fit test. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202


List of Tables xxviii

6.7 Validation of the model with the Sano et al.’s [Sano et al., 2006] results. . 210
Chapter 1

Introduction

1.1 Introduction
Products of human and animal origin products (biologicals and pharmaceuticals) and

drinking water pose unique health and safety problems, especially due to viral contami-

nations, and have traditionally been a key concern both for regulators and industrial op-

erations. They have also proven to be a stumbling block for early product research and

development and inexperienced small start-up firms. In a biopharmaceutical industry, a

typical bioprocess consists of various unit operations including media preparation, cell

harvesting, followed by three or four stages of purification using a series of chromatogra-

phy columns and ultrafiltration or diafiltration steps. Those operations performed before

chromatographic operation are referred to as upstream while those post-chromatographic

are referred to as downstream processing (Fig. 1.1).

Membrane filtration, being non-invasive, non destructive and a robust technique, is

the most preferred choice for virus removal and which is used extensively throughout the

production, purification, and formulation of biotechnology products, including recombi-

nant proteins and nucleic acids [Brandwein and Aranha-Creado, 2000; Davis, 2008]. Over

the past few years, viral safe products have been manufactured by the development and

availability of biocompatible viral ultrafiltration systems. Ultrafiltration is an efficient pro-


1.1. Introduction 2

Figure 1.1: A typical bioprocess flowchart with indicative operations

cess for removing macromolecules, colloids, endotoxins, pyrogens, viruses, and bacteria

[van Reis and Zydney, 2007]. It uses membranes of pore size ranging up to 1000 kDa

MWCO and these systems are specifically designed to allow proteins and retain viruses,

and typically more than 4 logs of virus removal are achieved while ensuring good pro-

tein permeability and recovery [Arkhangelsky et al., 2008; Hirasaki et al., 2002; Manabe,

1996]. During filtration, viruses smaller than the pore size carried along with the fluid,

pass through the membrane pores and when the pore size is smaller than the virus parti-

cles, then the particles get retained. Size exclusion is considered the main mechanism of

clearance by filtration.

Viral retentive membrane filtration systems have been evolving tremendously owing to

their efficient removal of numerous bacterial and mammalian viruses, including human and

murine retroviruses, simian virus (SV40), various hepatitis viruses and parvoviruses such

as minute virus of mice and B19. New research offers promising data on retention of trans-

missible spongiform encephalopathy (TSE) agents known as prions. The total US market
1.1. Introduction 3

value for ultrafiltration is expected to reach US$908 million by 2011, following its steady

growth from US$579 million in 2005 and US$635 dollars million in 2006, according to

ultrafiltration membrane markets, MST044B from BCC Research [Hanft, 2010]. These

data suggest an even wider application of membrane filtration to enhance viral safety in

manufacturing biological and biopharmaceutical products, including the ability to scale

reliably and efficiently from the lab to production.

Quality by Design highlights the importance of identifying and controlling all the pa-

rameters that can influence final product performance: Specifically, International Confer-

ence on Harmonization (ICH) Topic Q6A identifies particle size as a potentially important

variable [ICH, December 2000]. The specific need for a particle size specification is deter-

mined by assessing if and how this parameter affects the performance in purification and

fractionation. In virus capture or clearance operations, the virus particles are distributed

in the range of 15nm - 300nm depending upon its protein coat and its family, hence the

particle size study in ultrafiltration becomes inevitable because of its distribution nature.

Filtration is most effective when operated as tangential flow (TFF) or otherwise known

as cross-flow mode and is being exploited in several virus removal and purification oper-

ations [Guo et al., 2009; Michalsky et al., 2009] as it is less expensive compared to other

techniques like chromatography and are easier to operate. But extensive experimentation is

required prior to design and implementation especially for biotechnological applications.

Although various efficient experimental techniques for flux prediction have been presented

and used, there still is a lack of theoretical studies particularly those aiming at predicting

filtration efficiency in terms of log reduction value. Thus there is strong incentive for new

approaches to predict final permeate quality of tangentially ultrafiltered products.


1.1. Introduction 4

While others have attempted modelling systems such as deep bed filtration [Santos and

Bedrikovetsky, 2004], this work, we believe, is a maiden attempt to illustrate the appli-

cation of the population balance framework to predict the final product efficiency in the

context of tangential flow ultrafiltration (UF). The main objective of this research is to de-

velop a novel model for the TFF based on the population balance theory. Such predictive

modelling will serve as a promising tool for optimizing and controlling the TFF process.

The innovation of this work lies in the above fact, whereas previous ultrafiltration models

have been dominated by flux modeling techniques, i.e. at the macroscale.

The thesis is organized as follows: first, a literature survey covering various sources of

virus contamination and different regulatory requirements is presented. This is followed by

an overview of the UF process and its modes of operation. Particle characterization, a quick

economic sketch and a mini patent review on virus removal are then presented. Chapter

3 outlines the experimental setup and the methods employed for particle size and particle

concentration ultrafiltration studies. Chapter 4 gives a detailed account on the undertaken

filtration efficiency study and the influence of feed particle size and size distribution on

separation mechanism. Chapter 5 discusses already available models before introducing

our modeling approach based on population balances. The model simulation results are

presented to illustrate the applicability of our model to simulate the virus filtration mech-

anism. Chapter 6 presents optimization-based parameter estimation and discusses its nov-

elty in the field of filtration in general, and specifically shows the usefulness of such an

approach in the area of ultrafiltration by way of validation against our experimental data.

Finally, Chapter 7 summarizes the significant findings and recommends future directions

of this work.
Bibliography 5

Bibliography
Arkhangelsky, E., Steubing, B., Ben-Dov, E., Kushmaro, A., Gitis, V., 2008. Influence of

ph and ionic strength on transmission of plasmid dna through ultrafiltration membranes.

Desalination 227 (1-3), 111–119. 2

Brandwein, H., Aranha-Creado, H., 2000. Membrane filtration for virus removal. Devel-

opments in biological standardization 102, 157–163. 1

Davis, K., 2008. Biopharmaceutical manufacturing: The challenge of global regulatory

compliance. Pharmaceutical Technology 32 (6), 60–68. 1

Guo, Y., Cheng, A., Wang, M., Zhou, Y., 2009. Purification of anatid herpesvirus 1 par-

ticles by tangential-flow ultrafiltration and sucrose gradient ultracentrifugation. Journal

of Virological Methods 161 (1), 1 – 6. 3

Hanft, S., 2010. Ultrafiltration membranes: Technologies and the u.s. market. BCC Report

ID: MST044C. 3

Hirasaki, T., Yokogi, M., Kono, A., Yamamoto, N., Manabe, S., 2002. Removal and deter-

mination of dispersion state of bacteriophage [phi]x174 in aqueous solution by cupram-

monium regenerated cellulose microporous hollow fiber membrane (bmm®). Journal of

Membrane Science 201 (1-2), 95 – 102. 2

ICH, December 2000. Ich topic q6a specifications: Test procedures and acceptance criteria

for new drug substances and new drug products. Chemical Substances. 3
Bibliography 6

Manabe, S., 1996. Removal of virus through novel membrane filtration method. Develop-

ments in biological standardization 88, 81–90. 2

Michalsky, R., Passarelli, A., Pfromm, P., Czermak, P., 2009. Purification of the bac-

ulovirus autographa californica m nucleopolyhedrovirus by tangential flow ultrafiltra-

tion. Desalination 245 (1-3), 694 – 700. 3

Santos, A., Bedrikovetsky, P., 2004. Size exclusion during particle suspension transport in

porous media: stochastic and averaged equations. Computational and Applied Mathe-

matics 23 N. 2-3, 259–284. 4

van Reis, R., Zydney, A., 2007. Bioprocess membrane technology. Journal of Membrane

Science 297 (1-2), 16–50. 2


Chapter 2

Literature Review

Contents
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.2 Virus Contamination . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.2.1 In the context of biopharmaceutical industry . . . . . . . . . . . . . 12

2.2.2 In the context of water industry . . . . . . . . . . . . . . . . . . . 14

2.3 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.4 Regulatory Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2.4.1 Regulations governing biopharmaceutical industry . . . . . . . . . 16

2.4.2 Regulations governing water industry . . . . . . . . . . . . . . . . 20

2.5 Virus Filtration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.5.1 Membrane modes and configurations . . . . . . . . . . . . . . . . 23

2.5.1.1 Normal flow filtration . . . . . . . . . . . . . . . . . . . 23

2.5.1.2 Tangential flow filtration . . . . . . . . . . . . . . . . . . 26

2.5.1.3 High performance tangential flow filtration . . . . . . . . 28

2.5.1.4 Field flow fractionation . . . . . . . . . . . . . . . . . . 29

2.5.1.5 Remarks on membrane modes and configurations . . . . 31


8

2.5.2 Membrane modules . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.5.2.1 Flat sheet . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.5.2.2 Spiral wound . . . . . . . . . . . . . . . . . . . . . . . . 34

2.5.2.3 Hollow fibre . . . . . . . . . . . . . . . . . . . . . . . . 34

2.5.3 Particle characterization . . . . . . . . . . . . . . . . . . . . . . . 36

2.5.3.1 Dynamic light scattering . . . . . . . . . . . . . . . . . . 36

2.5.3.2 Imaging and image analysis . . . . . . . . . . . . . . . . 41

2.5.3.3 Asymmetrical flow field flow fractionation and multi-

angle light scattering . . . . . . . . . . . . . . . . . . . . 42

2.5.3.4 Electro-spray differential mobility analysis . . . . . . . . 42

2.6 Economics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

2.7 Patent Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

2.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

Preface
Ultrafiltration (UF) is having an increasing role in virus filtration and is becoming a

significant operation for virus removal or virus purification/concentration operations. It is

mainly employed in the biopharmaceutical and water industry to meet strict regulations

(i.e. those of the Food and Drug Administration’s and Safe Drinking Water Act respec-

tively) requiring validation of viral clearance. In order to obtain the final products safe
2.1. Introduction 9

from viral contamination, the proper insight of the source material and the subsequent pro-

duction and purification stages are essential. This review covers various aspects including

possible ways of virus contamination, reviewing various regulatory agencies’ guidelines,

mechanism and modes of virus UF, and particle size characterization techniques. It finally

assesses the economic significance of virus UF before finishing with a review of commer-

cial patent literature in the field. The content of this chapter has been published in Recent

Patents on Chemical Engineering, 2008 [i].

2.1 Introduction
Viruses are simple life forms, potentially pathogenic organisms with a small number

of genes (DNA or RNA) encased in a coat of protein (capsid) or a lipid membrane, i.e.

RNA or DNA enveloped/non-enveloped types, and normally fall in the size range 15nm

- 300nm. Table 2.1 shows types of viruses and the sizes which are normally involved in

virus filtration studies (Fig. 2.1). Viruses are incapable of independent replication and

need a host cell where newly replicated viral particles infect other cells. In some cases, the

viral DNA becomes an integral part of the host cell chromosome.

There are many reported incidents where some products, including drinking water,

vaccines, hormones and blood clotting factors, have resulted in the transmission of infec-

tious disease. Hence, virus filtration is a major concern in the biopharmaceutical, blood

products and water industries [Burnouf and Radosevich, 2003; Charcosset, 2006; Hensgen

et al., 2010; Huang and Peterson, 2001; Soluk et al., 2008] as risk assessment and safety

assurance are critical elements of the starting materials. Globally, inadequate supply of

water and increased consumption of contaminated drinking water create a big dent on the

health safety and developing countries are the worst hit. This necessitates the policy mak-
2.1. Introduction 10

Figure 2.1: Different types of virus particles giving a snapshot of size and shape of viruses
(Pic. courtesy: antibac2k.com, iah.bbsrc.ac.uk.com, innovations-report.de).
2.1. Introduction 11

Table 2.1: Virus Characteristics: Type, family and size used in virus clearance study.
Type Viruses Family Size (nm)
RNA enveloped HIV Lentivirus 80-110
Xenotropic, ecotropic, Retrovirus 80-110
recombinant murine retro-
viruses
Bovine diarrhea virus Flavivirus 50-70
Vesicular stomatits virus Rhabdovirus 70-150
RNA non enveloped Reovirus Reovirus 60-80
Polio virus Picornavirus 25-30
EMC virus Picornavirus 25-30
Hepatitits A virus Picornavirus 25-30
DNA enveloped Pseudorabies virus Herpesvirus 120-200
Herpes simplex virus Herpesvirus 120-200
DNA non enveloped Adenovirus Adenovirus 70-90
SV-40 virus Papovavirus 45-55
Porcine parvovirus Parvovirus 18-24
Minute virus of mice Parvovirus 18-24

ers and production companies to quantitatively assess the health risk of existing or planned

catchment, treatment, and distribution system of drinking water [Henley, 2009].

Virus filtration is one of the emerging markets for membranes, because it is considered

a robust and effective virus clearance technology and a common unit operation in various

industries. Early work in this field dates back to 1971 when virus retention using mem-

branes was first attempted [Scutt, 1971]. The impetus behind the continuous evolution of

virus filtration technology is due to two important factors; (i) liability of virus contami-

nation is large and there are numerous cases where virus contamination has led to serious

illness, and (ii) viruses must be removed, in any case, say drinking water or during pro-

duction in a biotech process. In virus filtration, it is very important that attention is paid to

meet the regulatory requirements, selection criteria, optimization, and validation of virus

filtration steps and to process scale-up.

Among various filtration processes available, UF is emerging as a powerful tool for

virus filtration. Number of works demonstrate UF being employed for virus purifica-
2.2. Virus Contamination 12

tion/concentration studies where the degree of recovery of virus is less critical in compar-

ison with virus removal operation. Viruses purified by UF include Aedes aegypti, Minute

Virus of Mice, Autographa californica M nucleopolyhedrovirus which are industrially im-

portant vectors [Guo et al., 2009; Hensgen et al., 2010; Michalsky et al., 2009]. In virus

removal or protein separation studies, the main aim is to achieve higher retention of viruses

with complete recovery of protein product. Say, during the manufacturing of diaspirin

crosslinked haemoglobin (DCLHb), UF is identified to be one of the crucial stages where

filtration of red blood cell haemolysate through hollow fibre membrane cartridges of 500

kDa results in stroma free haemoglobin [Azari et al., 2000; Ogle and Azari, 2001]. Thus

virus filtration is efficiently achieved by UF and which is dependent upon the contaminant

source, the regulatory requirements, and the configuration.

In this chapter, we discuss various aspects of virus filtration giving an overview of the

possible ways of virus contamination, reviewing various regulatory agencies’ guidelines,

mechanism and modes of virus UF, and types of virus filters. We also assess the economic

significance of virus UF before finishing with a mini review of commercial patent literature

in the field. This chapter is organized to cover the above outlined aspects of virus filtration

considering biopharmaceutical and water industry as major players.

2.2 Virus Contamination

2.2.1 In the context of biopharmaceutical industry


The occurrence of biopharmaceutical contamination by viruses happens at various lev-

els starting from the source material to the final product preparation. Throughout the his-

tory of biologics administered for human use, one can find numerous examples of products
2.2. Virus Contamination 13

that have been contaminated with potential human pathogens. Table 2.2 summarizes some

of the instances of viral contamination. The risk of the contamination is considered based

on the source, i.e. if its animal derived, human blood or plasma derived. It is very sur-

prising to note that the diphtheria epidemic paved the way for the inception of regulatory

affairs for health care industries and the maiden one dates back to 1902 Biologics Control

Act [Bren, Jan-Feb 2006]. With the advent of the biotechnology industry and the use of

continuous animal cell culture to produce therapeutic monoclonal antibodies, viral vac-

cines and recombinant proteins for human use, there are many concerns relating to the

potential transmission of animal viruses present in the cell lines, particularly retroviruses,

to humans [Garnick, 1989]. For this reason, regulatory authorities both in the United States

and in Europe formulated guidelines designed to minimize any potential risk of viral trans-

mission. This involved extensive testing of the cell banks, unpurified bulk material, and

final product for the presence of infectious agents. In addition, it was requested that the

production processes used in the purification of these biopharmaceuticals were tested for

their ability to remove or inactivate any potential infectious virus contamination.

Table 2.2: Reported incidents of virus contamination.


Disease (Year) Contaminated bio- Virus Possible Source
logical
Diptheria (1900’s) Antitoxin Unknown Horse
Yellow Fever (World Vaccine HBV as HSA Human blood
War II)
1950’s and 1960’s Vaccine ALV Embryonated hens
– Poliovirus Vaccine SV40 Monkey primary cell
lines
Restricted growth syn- – CJD Human growth hor-
drome (1970’s) mones
1980-2000 MABs, RTP HIV, HAV, HBV, HCV Human blood / plasma
– Veterinary Vaccine Bluetongue Unknown

Despite the stringent controls that are already in place, specific instances of virus con-
2.2. Virus Contamination 14

tamination of biotechnology production processes still occur. In each case, contamination

is thought to be either adventitious, that is, being introduced from an external source such

as the medium / serum used in the culture process or from an operator through a break-

down in current Good Manufacturing Practice (cGMP) procedures. Viral contamination

of cultured cells is associated with several problems including detection of virus particles,

capacity to replicate, risk for the operators/patients/non-contaminated cell cultures. Thus,

it is becoming evident that all biological products are prone to virus contamination (Table

2.3). These are thus characterized and rigorously tested for the presence of retroviral or

latent viruses and require purification and validation for complete removal. For a complete

review on the various sources of virus contamination, their cell lines and the final products

and viral detection techniques, the interested readers are referred to the article by Merten

[Merten, 2002].

Table 2.3: Virus contamination and their cell lines in the production of biological products.
Virus Cell line Product Contamination
Retrovirus Hybridoma Monoclonal anti- Endogenous
body / recombi-
nant protein
Retrovirus BHK Recombinant Endogenous
protein / vaccines
EBV Namalva Interferons EBV transforma-
tion

2.2.2 In the context of water industry


Traditionally, the human health impact of drinking water is assessed via epidemiology

and it is generally considered that pathogens pose higher health risks than chemicals. Ac-

cording to such studies, the consumption of drinking water is associated with waterborne

outbreaks of diseases such as Hepatitis A & E, gastroenteritis, Cryptosporidium, and Gia-

rdia [Grabow, 1996, 1997; Guthmann et al., 2006; Teunis et al., 1997]. Enteric viruses are
2.3. Definitions 15

increasingly seen as the most important pathogens in drinking water in developed regions

(from a health outcome point of view), versus Cryptosporidium that is recognized to cause

the majority of outbreaks [Ashbolt, 2004]. Table 2.4 gives an overview of various viruses

capable of contaminating drinking water and the possible outbreaks.

Table 2.4: Virus contamination and their sources resulting in a possible outbreak in drink-
ing water.
Disease and Transmission Virus Contaminant Source
Adenovirus infection Adenovirus Self manifestation in improperly
treated water
Gastroenteritis Calicivirus, Enteric Aden- Self manifestation in improperly
ovirus, and Parvovirus treated water
SARS (Severe Acute Respi- Coronavirus Self manifestation in improperly
ratory Syndrome) treated water
Hepatitis A Hepatitis A virus (HAV) Self manifestation in improperly
treated water
Poliomyelitis (Polio) Poliovirus Through the feces of infected indi-
viduals
Polyomavirus infection Two of Polyomavirus: JC Self manifestation in improperly
virus and BK virus treated water

It is reported that there were 642 outbreaks of waterborne disease in the United States

(1971-1996 period) of which 58% of it is associated with ground water sources, especially,

the identified agent is virus which accounted to 9% of the outbreaks [E.P.A., 2000]. As

per 1999-2000 report, 70% infectious disease outbreaks are attributed to groundwater, and

virus is identified as one of those [Lee et al., 2002]. Thus, virus contamination is becoming

increasingly a threat, especially with new virus candidates evolving (like H5N1), to a safe

drinking water.

2.3 Definitions
Several terms relevant to the UF process are defined here to facilitate understanding of

discussions that follow. Primarily are the synonymous terms Log Reduction Value (LRV)

or Log Titer Reduction (LTR) (Equation 2.1) defined as the logarithmic10 of the ratio of
2.4. Regulatory Requirements 16

the virus concentration (titer) in the feed (C f ) and the virus concentration (titer) in the

permeate stream (C p ). In some literature, LRV is also referred to as reduction factor.

 
Cf
LRV = log10 (2.1)
Cp

The filtration recovery (S) refers to the ratio of the permeate flow (Q p ) to the feed flow

(Q f ):
Qp
S = (2.2)
Qf

Another term is the retention or rejection (R) coefficient of a membrane which is expressed

as:
Cp
R=1− (2.3)
Cf

2.4 Regulatory Requirements


Stringent regulatory requirements are to be met during the course of processing stage,

and the regulations are quickly summarized for biopharmaceutical industry and water in-

dustry.

2.4.1 Regulations governing biopharmaceutical industry


To ensure the safety of biopharmaceutical/biological products for human use, the bio-

pharmaceutical companies need to implement adequate technologies in their manufactur-

ing process and demonstrate the capability of their processes to remove or inactivate con-

taminants based on a process-specific virus clearance strategy [Ray and Tarrach, 2008].
2.4. Regulatory Requirements 17

Another major challenge to enter the global market is to follow geographic based regula-

tions.

Various regulations, prior to harmonizing, could be quickly summarized as:

• U.S. Food and Drug Administration (FDA) requires extensive information on facil-

ities and validation processes. FDA’s Center for Biologics Evaluation and Research

is responsible for most of the drug products for virus contamination. They mainly

target to protect and enhance public health through regulation of biological products

including blood, vaccines, therapeutics and related drugs and devices according to

statutory authorities.

• The European Agency for the Evaluation of Medicinal Products looks for less vali-

dation data but rigorous on inspection data. It has various centers but the Center for

Proprietary Medicinal Products is concerned with virus clearance validation.

• Japanese Ministry of Health, Labor and Welfare has a role in the international con-

ference on harmonization towards quality, safety and efficacy.

• The Paul Ehrlich Institute provides advisory functions at the national (Federal Gov-

ernment, German states) and international level (World Health Organization, Euro-

pean Medicines Agency, European Commission, Council of Europe and others) for

the safety of biological medicinal products. It mainly establishes validation practices

for virus clearance.

Tracing the inception of these regulations on viral clearance in the family map shows

that the European Union was the first to provide written guidance on biopharmaceutical
2.4. Regulatory Requirements 18

Figure 2.2: Virus Clearance is a critical step in biotechnolgical/biopharmaceutical manu-


facturing. It ensures removal of viruses found to have contaminated the bioprocess.

manufacturers (Table 2.5). Various guidelines now exist that detail the requirements nec-

essary to ensure these products are free from viral contamination ([Hesse and Wagner,

2000]). Manufacturers should select and test source materials for the absence of viruses,

check for the capability of the production process to remove or inactivate viruses and at

each and every stage of the production till the final product stage (Fig. 2.2).

The process of harmonizing global regulatory requirements began more than 20 years

ago (1990), when industry groups and regulatory authorities in the US, Western Europe,

and Japan formed the International Conference on Harmonization (ICH) of Technical Re-

quirements for the Registration of Pharmaceuticals for Human Use to discuss scientific and
2.4. Regulatory Requirements 19

Table 2.5: Timeline of global regulations for viral safety of licensed products.
Year Agency
1989 European Union - First guide for viral clearance for bio-pharmaceutical
manufacturers
1991a Committee for Proprietary Medical Products - First published document
1991b Committee for Proprietary Medical Products - Human blood and plasma
derived products
1993 US FDA’s center for Biologics Evaluation and Research - General
framework for process evaluation studies
1994 FDA - Guide for viral clearance and validations
1994 Paul Erlich Institute - Recommendation for viral clearance from plasma
derived products
1996 Committee for Proprietary Medical Products - Viral validation studies
1996 International Conference on Harmonization - Viral safety evaluation of
biotechnology products from human/animal origin cell lines

technical aspects of pharmaceutical product registration. According to these regulations,

the manufacturers need to meet the safety margin, i.e. less than 1 virus particle per 106

doses. It also depends on the type of viruses and the product. For example, endogenous

retroviruses log reduction value should be in the range of 12-18 log clearance whereas for

adventitious viruses it is around 6 log removal. Paul Ehrlich’s recommendations state that

at least 4 log clearance should be achieved in one of two robust steps in a process while a

total of 6 log clearance should be achieved in the overall process. This being with the log

clearance of 10 for enveloped viruses and 4 for nonenveloped viruses. So, depending upon

the respective regulatory requirements, either type of retrovirus clearance or parvovirus

clearance is to be chosen [EMEA, 2006].

To summarize, the major guidelines for viral safety stipulate that a licensed product

must be assured by three complimentary approaches:

• The source/raw materials and the cell lines are to be tested thoroughly for viral con-

tamination.

• The capacity of downstream processing of clear infectious viruses are to be assessed.


2.4. Regulatory Requirements 20

• Each and every stage of the product manufacturing is to be tested for virus contami-

nation and virus clearance levels.

These indicators are aimed to serve as good guidelines, however the committee for

proprietary medical product guidelines emphasize more on the robustness of steps rather

than clearance values. Hence, each process is reviewed on a case-by-case basis by the

regulatory authorities [Trijzelaar, 1993].

2.4.2 Regulations governing water industry


In 1974, the US congress passed the Safe Drinking Water Act, applicable for all public

water systems, which published enforceable regulations. These regulations formulated

by Environmental Protection Agency (EPA), applicable for virus contamination, can be

enumerated under two categories : (i) Surface Water Treatment Rule (SWTR) and (ii)

Ground Water Rule (GWR).

• The Surface Water Treatment Rule (54 FR 27486; 29 June 1989) highlights the

importance of safe public water systems that use surface water, or groundwater under

the direct influence of surface water, against water borne pathogens, specifically

including Giardia lamblia, viruses, and Legionella [Berger et al., 2009].

• The Groundwater Rule (71 FR 65574; 8 November 2006) applies to all public water

systems that use groundwater (about 142 000 systems). However, systems that use

groundwater that are under the direct influence of surface water must instead comply

with the SWTR [Berger et al., 2009]. Under the GWR, the State will determine

whether a system is fecally contaminated or is vulnerable to fecal.


2.5. Virus Filtration 21

Both the rules (SWTR and GWR) require systems to use sufficient water treatment (re-

moval, inactivation, or both) to reduce enteric viruses by at least 99.99% of their source

water densities. Apart from the above US regulations, WHO’s Guidelines for drinking wa-

ter quality (2006) gives a comprehensive strategy for ensuring safe drinking water [WHO,

2008]. Other notable countries providing such guidelines include Canada (www.hc-

sc.gc.ca), Australia (www.waterquality.crc.org.au), and New Zealand (www.mfe.govt.nz).

In addition, the US Environmental Protection Agency (USEPA) has a constant vigilance

on the contaminant candidate and potential ones include adenoviruses, caliciviruses, cox-

sackieviruses, echoviruses posing a threat to virologically safe water. It is also identified

that the modeling tools for such water borne pathogens are in its infancy.

2.5 Virus Filtration


Products of human/animal origin products (biologicals and pharmaceuticals) and

drinking water pose unique safety problems and have traditionally been a key concern

both for regulators and industry as already discussed. It has also proven to be a stumbling

block for early product research and development, and inexperienced small start-up firms.

Filtration, being non-invasive, non-destructive and a robust mechanism, is the most

preferred choice for virus removal and purification [Brandwein and Aranha-Creado, 2000;

Hensgen et al., 2010; Kitis et al., 2003; Michalsky et al., 2009]. Over the past few years,

a major leap towards viral safe products has been made, and accomplished by the de-

velopment and availability of biocompatible viral UF systems. Membranes of pore size

ranging upto 1000 kDa are utilized and these UF systems typically provide >4 logs of

virus removal in protein-viral mixtures ensuring good protein permeability and recovery

[Arkhangelsky et al., 2008; Hirasaki et al., 2002; Manabe, 1996] and in the case of water
2.5. Virus Filtration 22

treatment, log removal values range between 1.5-10 which exhibit tremendous application

of UF membranes for virus removal.

In the case of biopharmaceutical applications, D’Alisa et al. [Chandra et al., 2002] are

the first to show the potential of UF for a range of therapeutics and varied virus removal

efficiencies. Their paper showed results of single filtration of a factor IX concentrate pro-

ducing a 7.3 log reduction in vaccinia virus, 11.2 log reduction in HIV, 7.05 log reduction

in VSV, 4.19 log reduction in Sindbis virus, and 3.90 log reduction in EMC virus. Thus vi-

ral retentive membrane filtration systems demonstrate removal of numerous bacterial and

mammalian viruses, including human and murine retroviruses, SV40 virus and various

hepatitis viruses and parvoviruses such as MMV and B19. New research offers promis-

ing data on retention of transmissible spongiform encephalopathy (TSE) agents known as

prions [Terrab and Pawlak, 2007]. These data suggest an even wider application of mem-

brane filtration to enhance viral safety in manufacturing biological and biopharmaceutical

products, including the ability to scale reliably and efficiently from the lab to production.

In water treatment applications, more than 4.44 log removal of influenza viruses

(H5N1) and more than 3.85 log removal of MS2 phage through a hollow fibre polyether

sulphone membranes of 150kDa are reported [Lenes et al., 2010]. MS2 phage is an ac-

cepted model virus particle for enteric viruses such as poliovirus, hepatitis A or calicivirus

and for highly-pathogenic influenza viruses as well. UF membranes exhibited more than

7.1 log removal of MS2 virus when pre-coagulated / flocculated [Fiksdal and Leiknes,

2006]. Similar results have been reported with MS2 removal achieving more than 6 log

removal with 100 kDa membranes by Otaki et al. [Otaki et al., 1998].
2.5. Virus Filtration 23

2.5.1 Membrane modes and configurations


A number of examples of filtration work found in the literature aimed at removal and

concentrating viruses in final products using two main modes, namely normal flow filtra-

tion (NFF) (otherwise known as dead end flow filtration (DEF)) and cross flow filtration

(CFF) (otherwise known as tangential flow filtration (TFF)). Recently, two technologies

are evolving namely high performance tangential flow filtration (HPTFF) and field flow

fractionation (FFF) (Fig. 2.3). All the above modes differ in their operation and perfor-

mance characteristics. The challenging aspect of separation of viruses from the product

to be recovered lies in the fact when the size of viruses is similar to the size of proteins,

such as in the case of parvovirus clearance from the protein immunoglobulin G (IgG). The

membrane configuration in this case should be chosen to allow the separation of viruses

while maintaining the passage of proteins. Hence, the selection of the membrane mode of

operation, membrane type and configuration are crucial and specific to each case.

2.5.1.1 Normal flow filtration

Normal flow filtration is mostly employed for virus filtration and especially for labora-

tory scale applications [Bohonak and Zydney, 2005]. In NFF, the feed passes through the

membrane perpendicularly and hence there is more tendency for particle accumulation on

the membrane surface. When particles, in this NFF configuration, are trapped within the

pores or body of filter medium, it is called depth filtration or clarifying filtration, whereas

when the particles are stopped at the surface of the medium and accumulated as a cake of

increasing thickness, the separation is called cake filtration [Russell et al., 2007]. Normal

flow filtration for virus removal involves the following sequence of steps: loading the filter

modules within their housings, flushing, pre-use integrity testing, flushing, sanitization,
2.5. Virus Filtration 24

Figure 2.3: Schematic representation of dead end flow (NFF), cross flow (TFF) and flow
field fractionation (FFF) modes.

buffer flushing, challenge fluid processing, product recovery flushing, cleaning, post-use

integrity testing and module removal. The product quality in terms of permeability, LRV,

and retention characteristics is influenced by several mechanisms while operated in NFF

mode, including:

• Adsorption of protein products or other components in the fluid constricting the

pores, leading to reduction in the permeability and increase in the retention of larger

components.

• Particle capture associated with cake formation on the surface reduces the perme-

ability and retention is more complex and is normally conceived as initial increase

in passage followed by a drop in passage because of the balancing effect of prefer-


2.5. Virus Filtration 25

ential plugging and protein and or virus formation on the surface as a cake layer.

• Dynamic changes in the passage during filtration due to the concentration polariza-

tion effect leading to reduction in hydraulic driving force and protein precipitation.

The above effects can change the flow patterns within the membrane structure, in

some cases, reorienting the overall pore size distribution and a reduction in LRV

with throughput. The protein recovery could be well enhanced in NFF by buffer

rinse which is carried out at either constant transmembrane pressure (TMP) or fil-

trate flux.

Number of works illustrates the application of NFF mode for virus filtration. For exam-

ple, PlanovaT M 15N/35N (Asahi Kasei Medical Co. Ltd., Japan) constructed of cuprammo-

nium regenerated cellulose hollow fibers of 15nm and 35nm pore size showed more than

5.7 log removal for both the enveloped (eg. HIV, BVDV, PSR) and non-enveloped viruses

(eg. bovine parvovirus, poliovirus, SV40, reovirus). This is operated in dead end mode

while concentrating Factor IX and Factor XI [Radosevich et al., 1994]. The UF mem-

branes DV20/50 (Pall Corporation, USA) demonstrated virus removal of 50nm and larger

size viruses with higher LRVs (> 6 LRV) whereas smaller viruses like bacteriophage and

poliovirus were retained less in a complex medium [Brandwein and Aranha-Creado, 2000;

Roberts, 1997]. This is achieved using a polyvinylidene fluoride membrane of 20nm and

50nm pore size, operated in dead end mode for processing cell culture medium or phos-

phate buffered saline albumin. Proteins as large as IgGs were able to permeate through

these filters showing good recovery rates. Similarly, the Millipore Viresolve NFP is a nor-

mal flow parvovirus removal filter with a patented membrane structure for strength and

high flow. The filter retains >4 logarithmic reduction value (LRV) of 20 nm diameter
2.5. Virus Filtration 26

parvovirus.

2.5.1.2 Tangential flow filtration

Tangential flow or cross flow filtration involves passing of the fluid across the surface

of the filter medium, in a direction tangent to the filter. Industrial UF processes are usually

carried out in cross flow mode, the advantage being the lower extent of concentration po-

larization, filter plugging, high permeability and component passage [Belfort et al., 1994;

Guo et al., 2009; Powell and Timperman, 2005]. In addition to that, the cross flow mode

allows recirculation of the retentate stream which offers operational advantages [Ghosh,

2008]. During TFF operation, depending upon the particle membrane interaction, fluc-

tuations in LRV as a function of process variations maybe observed. It is claimed that a

region of high protein concentration known as boundary layer results which aids in retain-

ing viruses leading to improved LRV. UF in this mode finds a variety of applications in

concentrating (removal of solvent from protein solutions), fractionation (protein-protein

separation), clarification/removal (removal of particles from protein solutions), and desalt-

ing (removal of low molecular weight compounds from protein solutions).

The fluid flow dynamics in a typical TFF system used for virus filtration are controlled

by the pressure gradients, flowrate, fluid properties and if applicable, the polarization of

the membrane by solute or solids. A TFF system can be operated in constant TMP mode

or in constant flux mode. The TMP is the average driving force for the feed to pass through

the membrane and is given by:

P f + Pr
T MP = − Pp (2.4)
2

where P f is the feed inlet pressure, Pr is the retentate pressure and P p is the permeate
2.5. Virus Filtration 27

pressure. In a TFF system, the permeate side pressure is commonly not restricted and

hence can be considered to be at atmospheric pressure. Generally, the fluid flow through

the pores is laminar, and therefore proportional to the TMP. The nature of the flow through

the pores is dependent on the flow regime (i.e. on Reynolds number) of the retentate flow

and on the presence of screens disrupting the steady flow. The performance of the virus

removal filters which is graded by LRV is typically more influenced by the membrane

configuration, operating range and feedstock.

Number of researchers have employed cross flow filtration for virus filtration studies
TM
and have reported good removal efficiency. For example, Planova 20N/75N (Asahi

Kasei Medical Co. Ltd., Japan) membranes are operated in tangential mode for prothrom-

bin complex concentrates and IVIG resulting in more than 6 log removals with both small

and large virus particles. Similarly, ViresolveT M modules (Millipore Corporation, USA)

demonstrated a good range of removal starting 3.5 LRV with 70nm pore size membranes

which is a patented composite PVDF membrane that is specifically geared towards virus

removal applications. The membrane is designed so that not only a high LRV (>4) is

achieved in combination with a high protein recovery (>98 percent), but also that the ap-

plication aspects are incorporated: water wettable, a simple and predictable scale-up and

the process flux is competitive and economical. Aranhe-Creado et al. [Aranha-Creado

and Fennington Jr., 1997] showed the use of dead end in conjunction with tangential flow

to achieve > 10 log removal with the above membranes where some virus aggregation

is found to happen. Zydney [Ho and Zydney, 2003] and coworkers suggested that the

Millipores’ Viresolve membrane is efficient enough for the study of virus filtration but

concluded that membrane morphology and orientation play a critical role in determining
2.5. Virus Filtration 28

the overall performance of virus filtration membranes. It is quite evident from the above

survey that the filtration modes are operated in both the modes extensively but the choice

is dependent on the feedstock, operating parameters, expected log removal values and the

cost measures.

2.5.1.3 High performance tangential flow filtration

High performance tangential flow filtration, first introduced by Robert van Reis in

1997, is an emerging technology which enables simultaneous protein purification, con-

centration, and buffer exchange in a single unit operation [van Reis et al., 1997]. It can

potentially be used throughout the downstream purification process to remove specific im-

purities, clear viruses, and/or eliminate protein oligomers. HPTFF is a two dimensional

unit operation that exploits both size and charge mechanisms. Several factors make this

process a natural choice: Operating in the pressure-dependent flux regime, optimized oper-

ating parameters and input conditions lead to highest selectivity, resolution and throughput

[van Reis, 1996, 2000]. Charged membranes accentuate the resolution of both charged

and neutral solutes [van Reis et al., 1999]. In this mode, the permeate collected is re-

turned to the permeate side of the module such that the permeate flows co-current to the

feed. This aids in maintaining constant TMP throughout the module leading to finer frac-

tionation of proteins. Many researchers have worked on this, for example, Grzenia et al.

[Grzenia et al., 2008, 2007] demonstrated the use of both the modes i.e. TFF and HPTFF

for virus (AeDNV) purification and reported that with 100 kDa membrane, TFF exhibited

complete rejection whereas in HPTFF, smaller viral fragments permeate through the mem-

brane. But with respect to protein passage, HPTFF possessed better passage for proteins

that are similar in size with the molecular weight cutoff of the membranes and better sep-
2.5. Virus Filtration 29

aration factors for proteins with size differences less than an order of magnitude. Hensgen

et al. employed HPTFF for purification of Minute Virus of Mice using 50,100 and 300

kDa Sartorius membranes, in which they have virus particles being excluded completely

in 50 & 100 kDa membranes whereas 300 kDa is found to pass virus particles through the

membrane [Hensgen et al., 2010]. They have also reported more flux decay with 300 kDa

than 50 & 100 kDa which they have attributed to the virus particles’ access to the pores and

the cell culture medium. For a comprehensive review on HPTFF, the readers are referred

to the article by Zydney et al. [Zydney and van Reis, 2001].

2.5.1.4 Field flow fractionation

Field flow fractionation is a highly efficient technique for separating and estimating

physical parameters of different materials including biopolymers, biological cells, mi-

croorganisms, and colloidal and solid particles. FFF technique is briefly discussed here

as it is considered to be the fingerprint of UF owing to similar sieving mechanisms, foulant

membrane interactions, and membrane relative sizes [van de Ven et al., 2009]. The FFF

technique is successfully employed to fractionate macromolecules in the wider range of

molecular weights (103 - 1016 ) and particle sizes (10−3 - 102 μm) [Giddings, 1993]. The

separation occurs by differential retention in a stream of liquid flowing through a thin trape-

zoidal channel by the application of an external field. Broad methods of separation are af-

forded via FFF and are classified depending on the nature of the external field, i.e. whether

it is sedimentation, thermal, electrical, cross-flow, magnetic, or other field [Schimpf et al.,

2000]. The field is applied at right angles to the flow forcing particles to move towards the

walls of the channel at different velocities which in turn cause the separation [Giddings,

1993; Ratanathanawongs and Giddings, 1993]. Due to the parabolic velocity profile gen-
2.5. Virus Filtration 30

erated in the channel by the combined effects of field and flow, larger particles stay close to

the membrane whereas the smaller particles diffuse further away from the membrane and

are carried faster along the channel [Caldwell et al., 1980]. Ideally, the laminae occupied

by each particle type would be held essentially at a fixed position within the parabolic flow

profile. More realistically, Brownian motion and other disturbances cause some broaden-

ing of the regions. Several different types of distributions, arise out of different operating

modes, and are dependent on the balancing of primary and opposing forces (equilibrium

distribution) which establishes the selectivity, resolution and elution time. Various operat-

ing modes and the relevant opposing forces are summarized in Table 2.6. Among all the

techniques listed, the cross flow FFF is considered the most universal technique and is used

to fractionate different synthetic polymers and particles of natural samples. The retention

is dependent on the fields applied and defined as:

Cross-flow:
6D
R= (2.5)
Uw

Electric:
6D
R= (2.6)
μEw

Thermal:
6T
R= (2.7)
αT (dT/dx)w

where R is the retention, D is the diffusion coefficient, U is the cross flow velocity, w is
2.5. Virus Filtration 31

the channel thickness, μ relates to the electrophoretic mobility, E relates to electric field

strength, αT is the thermal diffusion and T is the temperature.

Table 2.6: Summary of various operating modes of FFF.


Operating mode Opposing force Sample distribution
Normal (up to 1 μ) Diffusion Exponential
Steric (1-100 μ) Repulsion from wall Thin layer at wall
Lift hyperlayer Hydrodynamic lift Thin layer above wall
forces
Gradient hyperlayer Secondary gradients Thin layer above wall
Cyclical Friction coefficient Oscillating
Chromatographic Surface forces Thin layer at wall
hybrid

Effect of cross flow

The effect of cross-flow in FFF which plays a major role in achieving an optimized sep-

aration is well covered by Giddings [Giddings et al., 1987]. However, it is briefly discussed

here. Cross flowrate is a key parameter in the design and operation of a cross flow mode

as it has a major influence on the concentration gradient buildup near the membrane wall

and the proper range depends on the configuration (Table 2.7). Increase in cross-flow im-

proves the permeate flux due to increasing sweeping effect resulting in less concentration

polarization. Whereas, it has a relatively less effect on the retention data.

Table 2.7: Typical cross flowrates for various membrane configurations.


Configuration cross flow velocity (ap-
proximate)
Flat sheet <3 m/s
Hollow fiber <5 m/s
Spiral (1 cm by 16 cm) <0.4 L/s

2.5.1.5 Remarks on membrane modes and configurations

All the above modes find extensive applications in virus filtration and the choice is de-

pendent upon the virus dimensions and characteristics, protein recovery, filter capacity and

process issues such as ease of use, as well as process economics. The NFF modes have
2.5. Virus Filtration 32

fewer components, are easier to operate, lower in capital costs (however the cost is depen-

dent on larger membrane areas if plugging is rapid). The thickness of the membrane in this

mode may be important when considering rejection. The TFF mode has the capacity to

suppress cake formation and the decline in flux thereby larger feed volumes can be filtered

before cleaning and or disposal. The thickness of the membrane is of secondary impor-

tance as rejection is primarily on the surface. The TFF mode is robust and is the most

preferred mode, for industrial applications, with respect to consistency during the entire

course of filtration and while considering capacity to handle different feedstocks without

plugging. If size exclusion alone is aimed at, more emphasis is on the virus and membrane

characteristics. On the other hand, while trying to achieve higher LRV and or retention,

more consideration is put on operating parameters, feed properties and relevant input con-

ditions. The HPTFF mode finds enormous number of applications in the purification of

small virus particles (like parvovirus) where contaminating host cell proteins and DNA

may be of similar size. The FFF mode has an unique advantage of characterizing the sam-

ples in their native form. The promise from this technology see it continuously attracting

interest and advancing. This method is still considered to be commercially sensitive in the

field of bioseparations. Compared to thermal and centrifugal force, flow field is considered

to be the leading technique and when coupled with other characterization techniques will

serve as a powerful separation method for protein and virus purification and is likely to

replace the other techniques, in future.

2.5.2 Membrane modules


The UF membranes employed for virus filtration are cast into different modules: flat

sheets, spiral wound, and hollow fibre membranes, the latter traditionally being employed
2.5. Virus Filtration 33

in major industries including bio-pharmaceutical and water industry [Li et al., 2008] (Table

2.8). Many of these modules are also available either as cartridges or capsules which may

use pleated structures to maximize filter area for high particle retention capacity while still

allowing high flowrates. The selection, design, and operation of membrane modules are

highly dictated by the techno-economic factors. These include cost of supporting materials

and enclosures, power consumption, ease of cleaning and replaceability, good mass trans-

fer characteristics, feed stream components, product volume, and scaling needs [Gagne

and Vaccaro, 2003; Schwinge et al., 2004].

2.5.2.1 Flat sheet

Flat sheet membranes are most sought after module because of ease of use and ease of

cleaning. They have relatively high packing density, provide good cross flow performance,

can handle viscous feeds and suspensions, and are amenable to incorporating turbulence-

promoting spacers. Limitations to this design are labour intensive fabrication, low mem-

brane area to module volume ratio, high hold up volume of the feed side, and lower pack-

ing densities compared to hollow fibre membranes. Several basic flat sheet membranes are

connected within a cassette in series or in parallel which have a wider application range

from small to large scale operations in virus filtration (Fig. 2.4). DiLeo et al. demonstrated

filtration of mammalian virus spiked protein solutions in flat sheet module and reported ef-

ficient flowrates with reproducibility and high retention [DiLeo et al., 1993]. Biesert et

al. attempted removing both enveloped and non-enveloped viruses from protein solutions

using flat sheet modules and obtained a high level of log removal with various viruses

[Biesert et al., 1997]. Similar results obtained with flat sheet modules at small and large

scales are reported elsewhere [Brough et al., 2002; Choi and Kim, 2008].
2.5. Virus Filtration 34

Figure 2.4: TFF modules for small and large scale applications. (Courtesy: Millipore
Corporation)

2.5.2.2 Spiral wound

Spiral wound membranes are difficult to clean and sterilize, are subject to plugging,

have a limited range of scalability, and their long flow distance makes high flow with low

pressure drop difficult. However they offer relatively high packing density, are inexpen-

sive and very successful for particulate free process streams. Generally, nanofiltration and

reverse osmosis popularly employ spiral wound configuration than UF and MF [Pearce,

2007]. But literatures are available with UF spiral wound configuration like Lute et al.

reporting efficient virus removal with ultipors DV20 membranes in spiral wound configu-

ration [Lute et al., 2008].

2.5.2.3 Hollow fibre

Hollow fibre membranes are self supporting, do not need spacers, have low dead vol-

umes on the feed side, have low pumping cost, have greater membrane surface area per

volume. However their disadvantage is their lower mass transfer coefficients and possi-

bility of fiber breakage. The inside-out hollow fibre membranes are the preferred ones as

they afford controlled flow hydraulics and cross flow along the membrane surface limits

membrane fouling . In this mode, the feed enters one end of the fiber, retentate leave the
2.5. Virus Filtration 35

other, while the permeate pass through the outside of the membrane fiber. Hollow fibre

is reported to be widely employed in biopharmaceutical industries wherein all types of

viruses are efficiently removed. Viruses larger than 35 nm in size (eg. BVDV, HIV) are

removed using hollow fibre Planova 35N membranes with 5.9 - 7.3 log removal whereas

significant removal of bigger viruses is easily accomplished using Planova 15N. Smaller

viruses (eg. PPV) showed 4.4 log removal with Planova 15N owing to very close virus

and pore size. Hirasaki et al. demonstrated that hollow fibre Planova 20N provides high

levels of parvovirus and porcine removal from human IgG solutions with minimal fouling

[Hirasaki et al., 2006].

Table 2.8: Summary of commercially available membrane modules for virus removal.
Name (Manufacturer) Material Module type Pore size, nm LRV
Viresolve 70 (Milli- Hydrophilic PVDF Flat sheet TFF – PV > 3.5
pore)
SV40 > 5.6
SBV > 7.4
RV > 7.2
MLV > 6.6
HIV > 8.5
Viresolve 180 (Milli- Hydrophilic PVDF Flat sheet TFF – PV > 2
pore)
HAV > 4
SV > 4.5
RV > 5.5
MLV > 6
Planova 15 N (Asahi Hydrophilic cupram- Hollow fiber - 15 ± 2 PPV > 4.4
Kasei) monium regenerated NFF
cellulose
PV > 7.8
Planova 20N (Asahi Hydrophilic cupram- Hollow fiber - 19 ± 2 PPV > 4.2
Kasei) monium regenerated NFF
cellulose
ECM > 5.4
Planova 35N (Asahi Hydrophilic cupram- Hollow fiber - 35 ± 2 BVDV > 5.9
Kasei) monium regenerated NFF
cellulose
HIV > 7.3
Virosart Polyether sulphone Spiral wound - 20 RV > 6
CPV(Sartorius) NFF
Bacteriophage >
4
Ultipor DV20 (Pall) Hydrophilic PVDF Spiral wound - > 20 Bacteriophage >
NFF 3
Ultipor DV50 (Pall) Hydrophilic PVDF Spiral wound - > 50 Bacteriophage >
NFF 6
2.5. Virus Filtration 36

Based on the above survey (Table 2.8), it is evident that all the three modes find numer-

ous applications for varying virus particles. But literature indicates hollow fibre is largely

employed in industrial applications [Li et al., 2008] whereas for small/lab scale, flat sheet

membranes maybe employed.

2.5.3 Particle characterization


Quality by Design highlights the importance of identifying and controlling all the pa-

rameters that can influence final product performance. Specifically, ICH Topic Q6A iden-

tifies particle size as a potentially important variable [ICH, December 2000]. The specific

need for a particle size specification is determined by assessing if and how this parameter

affects purification and separation performances. In virus clearance or purification opera-

tions, the virus particles are distributed in the range of 15nm - 300nm depending upon virus

protein coat and virus family. Various analytical techniques are available for virus particle

characterization which are summarized in Table 2.9. The prominent methods being either

light scattering and/or image analysis. However, field-flow fractionation coupled with mul-

tiple angle light scattering and electrospray differential mobility analysis (ES-DMA) are

slowly emerging as a powerful technique for virus particle characterization [Chuan et al.,

2008; EP, 2006; USP, 2008].

2.5.3.1 Dynamic light scattering

Particle size distribution is typically presented as number weighted (number of particles

in each size class) or volume/mass weighted (volume/mass of particles in each size class)

or intensity weighted distributions. If coarser particles are predominant, then volume/mass

weighted distribution is the appropriate presentation. Number based distributions are sen-

sitive to presence of finer particles and more so when dispersion dominates [USP, 2008].
2.5. Virus Filtration 37

Table 2.9: Summary of various analytical techniques available for virus particle character-
ization studies.
Method Brief description (manufacturer exam- References
ple)
Electron microscopy [Kuo, 2007; Lengyel et al.,
2008; Pelchen-Matthews and
Marsh, 2007]
- Transmission Electron (TEM) TEM – Particle morphology and size [Gillock et al., 1997]
(JOEL)
- Scanning Electron/Probe (SEM/SPM) SEM/SPM – Particle morphology and [Johnson and Chiu, 2000]
size
- Cryo Electron Tomography (CET) CET – 3D molecular structure [Harris et al., 2006]
Light scattering [Citkowicz et al., 2008]
- Dynamic Light Scattering (DLS) DLS – Measures stoke’s (hydrodynamic) [Tsoka et al., 1999]
radius of particles (Malvern, Brookhaven)
- Multi Angle Light Scattering (MALS) MALS – Measures root mean square ra- [Shortt et al., 1996]
dius, molar/molecular mass (Wyatt Tech
Corp –Heleos)
Asymmetric Flow Field Flow Fractiona- Analytical tool for characterization [Pease III et al., 2009]
tion (AF4) of biological samples (Wyatt Tech
Corp—Eclipse)
Electrospray differential mobility analy- Particle size distribution (TSI Inc) [Pease III et al., 2009]
sis (ES-DMA)
Diode Array Detector (DAD) Quantification and identification of parti- [Citkowicz et al., 2008]
cles (Agilent 1100)
Fluorescent Detector (FD) Quantifies and identifies particles like [Citkowicz et al., 2008]
proteins, DNA impurities (Agilent 1100)
Refractive Index Detector (RID) Measures absolute refractive index (RI) [Citkowicz et al., 2008]
and dn/dc (differential RI) of the solution
(Wyatt Tech Corp - Optilab REX)
2.5. Virus Filtration 38

In optical laser based particle characterization, repeatability and reproducibility are ex-

acting requirements but which depends on the choice of suitable sample concentration

and dispersion conditions. The US and European pharmacopeias specify a target repro-

ducibility of 10% for volume median diameters of particles >10μm in diameter and 20%

for those smaller than this [EP, 2006; USP, 2008]. Dynamic light scattering (DLS) thus

finds extensive application in particle size studies and especially for size distribution anal-

ysis. For example, Lipin et al. [Lipin et al., 2008] reported the use of DLS for particle

size distribution characterization of viral particles obtained during protein purification by

chromatographic separation column (Fig. 2.5). They showed the versatility of the DLS

technique in quantifying changes in size distribution that result from changes made to the

process.

Figure 2.5: Particle size distribution of high molecular weight, low molecular weight, and
virus like particle (MIX) solutions by DLS [Lipin et al., 2008].

In another study [Wickramasinghe et al., 2005], tangential flow UF is used to capture


2.5. Virus Filtration 39

human influenza virus particles using 100 and 300 kDa UF membranes. Particle size dis-

tributions were measured successfully in the permeate and retentate streams using laser

diffraction light scattering demonstrating fractionation of the original suspension (Figs.

2.6,2.7). In their study, Citkowicz et al. [Citkowicz et al., 2008] elaborated on the charac-

terization techniques for gene therapy from human polyoma JC virus, demonstrating the

online and offline use of DLS technique for measurement of particle size to the order of

25nm (Fig. 2.8).

Figure 2.6: Particle size distribution in the retentate (right) and permeate (left) streams
through 0.1 μ membrane during the study of HAV by TFF system.

It is very surprising to note that though particle size plays a significant role in the

membrane filtration, very limited literature is available relating the particle size and final

product quality. All the works cited above do not give or relate the particle properties

on the filtration efficiency values. Especially, quantifying the dependency of particle size

parameter on the filtration mechanism and studying its influence is still an unearthed area.
2.5. Virus Filtration 40

Figure 2.7: Particle size distribution in the retentate (right) and permeate (left) streams
through 0.2 μ membrane during the study of HAV by TFF system.

Figure 2.8: Particle size distribution of diethylaminoethyl (DEAE) and packaged virus like
particles by dynamic light scattering technique.
2.5. Virus Filtration 41

Hence, to bridge this knowledge gap we investigate and quantify this effect which will

lead to a better understanding and modeling particle size distribution, which is still in its

infancy.

2.5.3.2 Imaging and image analysis

Image analysis is another excellent choice as it provides statistically relevant particle

size and shape data in addition to detailed visual insight about shape and size. Several

imaging techniques are available for the study of viral size and morphology including

transmission electron microscopy, scanning electron/probe microscopy, atomic force mi-

croscopy, cyto electron microscopy and electron tomography. The choice of imaging tech-

nique is dependent on cell lines used, and safety aspects [Brandenburg and Zhuang, 2007;

Gaczynska and Osmulski, 2008; Kolin and Wiseman, 2007; Lengyel et al., 2008; Pelchen-

Matthews and Marsh, 2007]. Transmission electron microscopy (TEM) is the most com-

monly used technique for virus particle studies as it readily allows visualization of particle

morphology and size. However its limited resolution and possible artifacts caused by sam-

ple preparation may limit its accuracy. The use of cryo electron tomographic approach

to analyze and quantitate the structural complexity of viruses give an insight about the

structural information and bridges a critical gap in virus imaging analysis. Alongside the

imaging techniques, image processing which refers to the algorithms and software that

analyse the imaging data has been evolving rapidly in recent years. Several applications

are now commercially available including the popular freeware ImageJ (image process-

ing software inspired by National Institute of Health, USA) [Sun et al., 2007]. These can

perform several techniques for particle analysis including projected particle pixel area and

chord length measurement. Other algorithms such as that developed by Korath et al. [Ko-
2.5. Virus Filtration 42

rath et al., 2008] can perform more complex analyzes including high-fidelity separation of

touching particles.

Thus, image analysis gives an accurate information about the morphology of the virus

particles. Whereas for size distribution studies, image analysis method becomes quite

tedious and hence light scattering method is found to be more suitable.

2.5.3.3 Asymmetrical flow field flow fractionation and multi-angle light scattering

The availability of commercial field flow fractionation (FFF) equipment coupled with

sophisticated analytical instruments like multi-angle light scattering (MALS) has started

slowly penetrating into various biotechnological applications especially for the analysis of

viruses and virus like particles for vaccines and gene therapy [Wei et al., 2007]. Inspite of

FFF’s presence for more than 30 years, the technique is yet to gain momentum at various

levels viz. academic, commercial and research [Roda et al., 2009] for virus analysis appli-

cations. Specifically, the hybrid technique, i.e. asymmetric flow FFF (AF4) coupled with

MALS, is extremely useful for fractionation of wider range of particles of different species,

and is expected to make a big way in fractionation and removal. Chuan et al. [Chuan et al.,

2008] demonstrated the use of AF4-MALS technique for characterization of virus-like

particles giving cumulative size distribution at optimized condition and could, using such

data, show the influence of cross flowrates on fractionation. Pease et al. [Pease III et al.,

2009] demonstrated AF4-MALS as a valuable method to characterize virus-like particle

distributions rapidly and quantitatively.

2.5.3.4 Electro-spray differential mobility analysis

One of the emerging techniques for virus sample characterization is the electro-

spray differential mobility analysis (ES-DMA) which is capable of analyzing viruses like
2.6. Economics 43

MS2,T2, T4 and adenoviruses [Cole et al., 2009; Hogan Jr. et al., 2005; Thomas et al.,

2004]. In this technique, the concentration of samples are obtained. Samples are elec-

trosprayed into a gas stream where particles are detected using a condensation particle

counter via various stages including evaporation, charge neutralization, size separation.

This method is preferable for spherical viruses because of its ability to retain their viability

after electrospraying thus helping consistent sizing measurements. Cole et al. [Cole et al.,

2009] demonstrated the concentration measurement of the virus particles MS2, PP7 and

φX174 by ES-DMA and recommended PP7 to be used for virus filter standardization. ES-

DMA technique also provides information on the aggregation state of the sample and has

a wide range of concentration measurement from small viruses including MS2 and larger

ones including adenovirus (used for gene therapy) [Hogan Jr. et al., 2006].

Thus, to summarize, virus particle characterization maybe largely achieved by image

and dynamic light scattering analysis. Electron microscopic methods can accurately pro-

vide size and morphology information of the virus particles, however imaging techniques

are prone to image artifacts and are not optimised for high throughput analysis. Whereas,

dynamic light scattering technique provides size and size distribution with a fair level of

accuracy, but when the sample available is limited, this method is restricted as the sample

concentration sufficient to obscure the laser needs to be determined. A more recent hybrid

technique namely AF4 with MALS is expected to replace the current techniques in a big

way, leading to a more qualitative and accurate separation and characterization.

2.6 Economics
UF technology is constantly evolving with newer types of membranes and materials to

meet the regulations, hence variedly employed in the booming pharmaceutical and biotech
2.6. Economics 44

fields (Fig. 2.9 [Vacura, 2008]). In general, the market for membrane technology used in

biotechnological discovery, development and commercial production, is expected to rise at

a compound annual growth rate (CAGR) of 7.4% to $908 million in 2011 [Hanft, 2010].

Specifically, UF a multi-million dollar market, is growing at an annual average growth rate

(AAGR) of 7.6%. The growth of the UF method for bioprocessing expands because of the

constant demand of protein purification, gene therapy, vaccine production, juice and wine

industry, pharmaceutical grade water wherein virus filtration is a vital operation.

Figure 2.9: Membrane filtration’s place in the Pharmaceutical/Biotech industry - World


Market Share 2008 ($ million) (UPW-UltraPure Water, MaF-Macro Filtration, SAC-
Sedimentation and Centrifugation).

According to the updated technical market research report on UF Membrane Markets -

MST044C (May 2010) - from BCC Research, there is a clear indication of an accelerated

growth rate of UF. Specifically, the US market value for UF was $882 million in 2009

and is expected to reach $932 million by the end of 2010. As UF is found to be the most

appropriate technology for producing dairy and juice products, protein fractionation, corn

and other mill product processing, nutraceuticals manufacture, and waste water treatment,
2.6. Economics 45

the challenge is to obtain most efficient membranes but at reasonably reduced cost. The

capital costs involved in membrane filtration industry are higher whereas resources like

company space and energy consumption are less compared to other chemical processes.

This enables companies to reduce costs by upwards of 65%.

The success of biotechnology for bulk product manufacturing, thus, heavily depends

on engineering solutions in downstream processing in which separation and purification

have a crucial role to play. Interestingly, biotechnology sector is expected to grow robustly

over the next 5 years. Growth forecast for the period 2010-2015 is expected to rise at a

compound annual growth rate of 10.2% [Hanft, 2010]. Whereas in drinking water treat-

ment, UF is becoming competitive pretreatment system for RO, where the feed water is

not too high in terms of organic content. And UF has become the preferred alternative to

conventional technology to remove water borne pathogens in the preparation of drinking

water [Clever et al., 2000; Laine et al., 2000; Rautenbach and Vobenkaul, 2001]. Consis-

tently, UF seems to efficiently achieve current water regulation values for turbidity, Giardia

and virus removal. UF applications, in full-scale plants, represent about 74% of the total

installed capacity. This is due to the fact of no loss in the performance in terms of pro-

duction capacity, membrane material over time and water quality. UF proves to be quite

economic including power, membrane replacement, consumable, maintenance, capital and

labor costs, treating surface water costs around $ 0.235 per m3 reports Drouiche et al.

[Drouiche et al., 2001].

These data clearly confirm that UF technology is gaining a stronghold position meet-

ing the international acceptable standards in biotechnology and water sectors. With this

tremendous growth in the upward direction, regulatory authorities depend specifically on


2.7. Patent Review 46

well-designed, optimized process and cost-efficient polymers to remove contaminants at

the end product.

2.7 Patent Review


Accelerated developments are taking place in the manufacturing methods of UF mem-

branes, their configurations and material makeup towards achievement of appropriate re-

tention values, throughput and permeability for virus removal. Viruses, either those oc-

curring in nature, or recombinant versions thereof, are extensively used in the field of

gene therapy and vaccinations. Purification of such viruses, being a vital operation, is

efficiently achieved by UF and widely employed in industries. It is observed that high re-

tention values are obtained with porous polyvinylidene fluoride membranes and cellulosic

membranes [Degen et al., 1998b; DiLeo et al., 1991; Handlin et al., 2009; Masayuki and

Shin-ichi, 2009]. Defect free cellulosic UF membranes made from microporous polymeric

substrates, are capable of maintaining desirable flux values as well as good retention per-

formance [Tucelli and McGrath, 1996; Walter et al., 2010]. Wilson’s patent [Wilson and

Mikhail, 2007] includes an additional step of autoclaving the membrane in boiling water

and/or steam during the fabrication stage. This provides a caustic resistance and which can

be of hollow fiber or sheet type. These hydrophobic polysulfone composite polymer mem-

branes, having its surface rendered hydrophilic with a hydroxyalkyl cellulose, are capable

of achieving higher LRV value and throughput. Miranda’s patent claims that purification

of viruses is best achieved by applying back pressure in the permeate side [Miranda, 2009].

A patent review covering inventions on the use of UF for virus removal is summarized in

Table 2.10.

A TFF system has been considered by van Reis [van Reis, 1996, 2000] to separate
2.7. Patent Review 47

Table 2.10: A mini review of patents on the use of UF for the removal of virus particles.
Year/US Patent Patent Title Patent Inventor(s) Patent Assignee
Number
2009/20090123989 Virus purification using UF Weggeman Miranda Crucell Holland B.V.
(Leiden, NL)
2007/7226429 Method for removal of viruses from Tullis Richard H Aethlon Medical Inc.
blood by lectin affinity hemodialy- (San Diego, CA)
sis
2007/7223585 Viral purification methods Coffey Matthew Oncolytics Biotech Inc.
(Calgary, CA)
2006/7118675 Process for removing protein aggre- Siwak Martin, An Millipore Corporation
gates and virus from a protein solu- Hong, Cormier Jason (Billerica, MA)
tion R, Kinzlmaier Dana
2006/7108791 High resolution virus removal Tkacik Gabriel, Kazan Millipore Corporation
methodology and filtration capsule Greg (Billerica, MA)
useful
2002/6365395 Process for removing protein aggre- Antoniou Millipore Corporation
gates and virus from a protein solu-
tion
2000/6054051 Tangential flow filtration systems van Reis Genentech Inc.(South
Sanfransisco, CA)
1998/5736051 Polyvinylidene fluoride membrane Degen Peter John, Sip- Pall Corporation (East
and method for removing viruses sas Ioannis P, Rapis- Hills, NY)
from solutions arda Gregory C Gregg
Joseph
1998/5731164 Method of Checking the rate of re- Becker Gerhard, Lar- Sanorell Pharma
moval of pyrogenic substances, in son Paul Marcel, Heidl GmbH and Co.
particular viruses, from organic ma- Reiner (Baiersbronn, DE)
terial
1997/5645984 Process for depleting viruses in so- Nader Warner Sanorell Pharma
lutions and determining their deple- GmbH and Co.
tion rate (Baiersbronn, DE)
1991/5017292 Membrane Process and system for DiLeo AJ, Allegrezza Millipore Corporation
isolating virus from solution AE, Burke ET (Bedford, MA)

species of interest in the range 1 to 1000 kDa MWCO (mostly targeting biological fluids)

from mixtures. The flux is found to be dependent on transmembrane pressure until the

pressure reaches a transition point beyond which this dependency drops off. This patented

filter is claimed to be capable of filtering similar sized specimens and from thereon, TFF

is reported to being employed in number of purification stages for high selectivity and

throughput [John et al., 2003]. Later, high performance tangential flow filtration emerged

to yield a high-purity product in protein purification [Robert et al., 2008]. David et al.’s

[David and William, 2001] method employs filtration aid together with low concentra-
2.7. Patent Review 48

tion of metal ions in place of nucleases for commercial scale purification of encapsulated

viruses from cell culture. In these methods, UF plays an important role in separating

viruses from DNA and RNA species present in the cell culture. The filtration is aided not

by enzymatic nucleases but rather by filtration aids such as diatomaceous earth. This offers

advantages for commercial scale purification of encapsulated viruses particularly as spec-

ification tests that would have to be designed to demonstrate the removal (i.e. absence)

of the nuclease(s) in the final product would be eliminated. Similarly, a tangential flow

system with the hollow type UF membrane is recommended as the most preferable and

robust method for virus concentration [Matthew, 2007].

Matthew’s improved method for virus purification from cell culture involves a simple

extraction step in which a detergent is directly added to the cell culture where the resulting

viruses are suitable for clinical administration to mammals, this being a major step towards

cell therapy [Matthew, 2003]. Tullis utilized hollow fiber membranes (200-500 nm pore

size) for virus (110 nm dia) removal from the blood stream [Tullis, 2007]. The retention

level of min 3 LRV of virus removal from protein solutions was achieved using normal flow

or DEF filtration but with one or more UF membranes [Martin et al., 2006; Tkacik and

Kazan, 2006]. Stefan’s virus filtration method for solutions containing macromolecules

suggests that the DEF technique offers economic advantages due to simpler equipment,

simpler operating procedures and reduction in the loss of macromolecules [Stefan, 1996].

It facilitated the virus filtration process in terms of residence time reduction and optimized

yield in comparison with the TFF as DEF involves placing the macromolecule contain-

ing solution in a pressure vessel prior to filtration and pressing the solution through the

membrane with the aid of a pressure source.


2.7. Patent Review 49

The method of using multilayered membranes [Degen et al., 1998a; Tkacik, 2007;

Tkacik and Kazan, 2006] with at least one membrane being oriented tight side down stream

showed satisfactory levels of virus removal but the protein passage and flux are low. Such

multilayered membranes can be UF membranes supported with monomer surface coat-

ing or two anisotropic membranes with one membrane juxtaposed with at least one other

membrane so that substantially all the skin surface of one membrane is in intimate contact

with all of the skin of the other membrane. Thus the membranes made by cocasting or se-

quentially casting plurality of polymer solutions substantially prevent the passage of virus

particles but permit the passage of protein [Tkacik et al., 2010]. Other such membranes

can be comprised of a tight side with porous support which is characterized by having

asymmetric structure free of macro-voids when exposed to protein solution, exhibiting

low protein binding and these are normally used to filter viruses from protein streams. In

order to increase the filtration rate during the UF of viruses, the viruses to be removed

are increased in size by incubation with a high molecular weight receptor, binding prefer-

ably a specific antibody so that the separation effect is improved and large pore diameter

membranes can be chosen [Nowak and Bernhardt, 2002].

From this mini-review of patents, a silver bullet solution was not exactly found but it is

arrived that a suitable membrane and membrane configuration should be selected in tandem

to obtain the required retention performance along with complete protein product recovery.

As such there is a body of knowledge in the patent literature that one ought to consult in

the process of choosing a more suitable filtration setup. It can be seen that there is an art

at play here where the skilled artisan may select through experimental campaigns the opti-

mal combination of membrane, membrane configuration, operating conditions, processing


2.8. Conclusion 50

steps, and so on. Thus well-defined deterministic knowledge for virus filtration by UF is

still outstanding. Having said that, there is a wide scope of research to be done in the area

of virus filtration providing ample opportunities for scientific inventors in this field. For

example, there is a rapidly increasing interest in the development of cost-efficient tech-

nologies for industrial separation of whey protein, a heterogeneous mixture of individual

protein fractions that have their own unique nutritional, functional, physiological and neu-

traceutical properties. Unfortunately, the protein separation / purification is a demanding

process due to the complexity of proteins themselves and their biological environments.

2.8 Conclusion
Although the basic scientific principles behind membrane technology have been around

since the 1950s, it was not until the 1970s that membrane technology was exploited for pu-

rification and or removal of virus particles. In this review, virus filtration is found to be

quite a significant downstream operation faced with challenges including meeting strin-

gent regulations, variability in virus contamination sources and optimal configuration and

operation.

• UF applicable for virus filtration can be operated in both modes, i.e. normal flow

and tangential flow. Two other techniques viz. HPTFF and FFF are largely evolving,

specifically FFF is finding broader applications as it is capable of handling the

streams in native form for characterization. But mostly, in industrial operations,

TFF is the preferred mode because of lower extent of concentration polarization and

filter plugging, high permeability, and operational advantages.


2.8. Conclusion 51

• Quality by Design highlights the importance of particle size as one of the influ-

encing parameters on the final product quality. Survey on particle size study in

membrane filtration, specifically for virus filtration operations, shows that only

limited literature is available on the effect of particle size along with the operating

parameters on the final product quality. With new virus candidates outbreak and

wider range of virus particles, it is important to study the size and its distributions’

role in the final permeate quality which will lead to an indepth understanding to

achieve better efficiency. Hence, it is identified that an urgent focus is needed to

develop a model for prediction of particle size and its distribution.

• Increasing demand of efficient UF virus filters in the bioseparation and water

treatment fields is causing a tremendous growth in UF membrane market. The

challenge lies in optimizing yields and reducing costs with the newer technology

and its development.

Membrane filtration for virus filtration operations is a fertile field, and researchers are

continuously proposing new and improved ways for employing UF as an efficient mech-

anism for producing virus free biologicals and drinking water. For UF to be still more

attractive, achieving higher filtration efficiency will be the key solution. Though many

factors influence the filtration efficiency, it is identified that particle size and its interplay

along with the operating parameters has not been addressed in the literature. Hence, this

thesis aims to fill the knowledge gap of particle size and size distribution effect along with

the operating parameters (TMP, CFR) on filtration efficiency. This is achieved by formu-
Bibliography 52

lating a novel model based on population balance theory, first of its kinds, which describes

the evolution of particle size distribution in the filtration product streams.

Bibliography
Aranha-Creado, H., Fennington Jr., G. J., 1997. Cumulative viral titer reduction demon-

strated by sequential challenge of a tangential flow membrane filtration system and a

direct flow pleated filter cartridge. PDA Journal of Pharmaceutical Science and Tech-

nology 51 (5), 208–212. 27

Arkhangelsky, E., Steubing, B., Ben-Dov, E., Kushmaro, A., Gitis, V., 2008. Influence of

ph and ionic strength on transmission of plasmid dna through ultrafiltration membranes.

Desalination 227 (1-3), 111–119. 21

Ashbolt, N. J., 2004. Risk analysis of drinking water microbial contamination versus dis-

infection by-products (dbps). Toxicology 198 (1-3), 255 – 262. 15

Azari, M., Boose, J., Burhop, K., et al., 2000. Evaluation and validation of virus removal by

ultrafiltration during the production of diaspirin crosslinked haemoglobin. Biologicals

28, 81–94. 12

Belfort, G., Davis, R. H., Zydney, A. L., 1994. The behavior of suspensions and macro-

molecular solutions in crossflow microfiltration. Journal of Membrane Science 96 (1-2),

1–58. 26

Berger, P., Clark, R., Reasoner, D., Rice, E., Santo Domingo, J., 2009. Water, drinking,

121 – 137. 20
Bibliography 53

Biesert, L., Hoffer, L., Josic, D., Suhartono, H., Gärtner, T., Schwinn, H., 1997. Viral

safety of a new factor ix (octanine f®): Combination of solvent/detergent (sd) treatment

with an ultrafiltration step. Infusionstherapie und Transfusionsmedizin 24 (4), 202. 33

Bohonak, D., Zydney, A., 2005. Compaction and permeability effects with virus filtration

membranes. Journal of Membrane Science 254, 71–79. 23

Brandenburg, B., Zhuang, X., 2007. Virus trafficking - learning from single-virus tracking.

Nature Reviews Microbiology 5 (3), 197–208. 41

Brandwein, H., Aranha-Creado, H., 2000. Membrane filtration for virus removal. Devel-

opments in biological standardization 102, 157–163. 21, 25

Bren, L., Jan-Feb 2006. The road to the biotech revolution: Highlights of 100 years of

biologics regulation. FDA Consumer magazine. 13

Brough, H., Antoniou, C., Carter, J., Jakubik, J., Xu, Y., Lutz, H., 2002. Performance of a

novel viresolve nfr virus filter. Biotechnology progress 18 (4), 782–795. 33

Burnouf, T., Radosevich, M., 2003. Nanofiltration of plasma-derived biopharmaceutical

products. Haemophilia 9 (1), 24–37. 9

Caldwell, K., Nguyen, T., Giddings, J., Mazzone, H., 1980. Field-flow fractionation of

alkali-liberated nuclear polyhedrosis virus from gypsy moth lymantria dispar linnaeus.

Journal of Virological Methods 1 (5), 241–256. 30

Chandra, S., Groener, A., Feldman, F., 2002. Effectiveness of alternative treatments for

reducing potential viral contaminants from plasma-derived products. Thrombosis Re-

search 105 (5), 391–400. 22


Bibliography 54

Charcosset, C., 2006. Membrane process in pharmaceutical and biotechnological applica-

tions. ITBM-RBM 27 (1), 1–7. 9

Choi, Y. W., Kim, I. S., 2008. Viral clearance during the manufacture of urokinase from

human urine. Biotechnology and Bioprocess Engineering 13 (1). 33

Chuan, Y. P., Fan, Y. Y., Lua, L., Middelberg, A. P. J., 2008. Quantitative analysis of

virus-like particle size and distribution by field-flow fractionation. Biotechnology and

bioengineering 99 (6), 1425–1433. 36, 42

Citkowicz, A., Petry, H., Harkins, R. N., Ast, O., Cashion, L., Goldmann, C., Bringmann,

P., Plummer, K., Larsen, B. R., 2008. Characterization of virus-like particle assembly

for dna delivery using asymmetrical flow field-flow fractionation and light scattering.

Analytical Biochemistry 376 (2), 163–172. 37, 39

Clever, M., Jordt, F., Knauf, R., Raebiger, N., Ruedebusch, M., Hilker-Scheibel, R., 2000.

Process water production from river water by ultrafiltration and reverse osmosis. Desali-

nation 131 (1-3), 325–336. 45

Cole, K. D., III, L. F. P., Tsai, D.-H., Singh, T., Lute, S., Brorson, K. A., Wang, L.,

2009. Particle concentration measurement of virus samples using electrospray differen-

tial mobility analysis and quantitative amino acid analysis. Journal of Chromatography

A 1216 (30), 5715 – 5722. 43

David, M., William, O., 2001. Method using filtration aids for the separation of virus

vectors from nucleic acids and other cellular contaminants. WO/2001/048155. 47


Bibliography 55

Degen, P., Sipsas, I., Rapisarda, G., Gregg, J., 1998a. Filtration medium. US Patents

5788862. 49

Degen, P., Sipsas, I., Rapisarda, G., Gregg, J., 1998b. Polyvinylidene fluoride membrane

and method for removing viruses from solutions. US Patents 5736051. 46

DiLeo, A., Allegrezza, A., Burke, E., 1991. Membrane process and system for isolating

virus from solution. US Patent: 5017292. 46

DiLeo, A. J., Vacante, D. A., Deane, E. F., 1993. Size exclusion removal of model mam-

malian viruses using a unique membrane system, part ii: Module qualification and pro-

cess simulation. Biologicals 21 (3), 287–296. 33

Drouiche, M., Lounici, H., Belhocine, D., Grib, H., Piron, D., Mameri, N., 2001. Eco-

nomic study of the treatment of surface water by small ultrafiltration units. Water SA

27 (2), 199–204. 45

EMEA, 2006. European commission: Emea guideline on virus safety evaluation of

biotechnological investigational medicinal products. London. 19

EP, 2006. Laser diffraction measurement of particle size. European Pharmacopoeia, Chap-

ter 2.9.31 http://www.edqm.eu/, 4429. 36, 38

E.P.A., 2000. Environmental protection agency proposed rules. Federal Register 65,

30193–30274. 15

Fiksdal, L., Leiknes, T., 2006. The effect of coagulation with mf/uf membrane filtration

for the removal of virus in drinking water. Journal of Membrane Science 279 (1-2), 364

– 371. 22
Bibliography 56

Gaczynska, M., Osmulski, P. A., 2008. Afm of biological complexes: What can we learn?

Current Opinion in Colloid and Interface Science 13 (5), 351–367. 41

Gagne, D., Vaccaro, G., 2003. Sulaibiya water reuse project well underway. Water and

Wastewater International 18 (9), 14–18. 33

Garnick, R. L., 1989. Safety aspects in the quality control of recombinant products from

mammalian cell culture. Journal of Pharmaceutical and Biomedical Analysis 7 (2), 255

– 266. 13

Ghosh, R., 2008. Handbook of Membrane Separations Ed. by Anil K Pabby and Syed S H

Rizvi and Ana Maria Sastre. CRC Press, London. 26

Giddings, J., 1993. Field flow fractionation - analysis of macromolecular, colloidal and

particulate materials. Science 260 (5113), 1456–1465. 29

Giddings, J., Chen, X., Wahlund, K., Myers, M., 1987. Fast particle separation by

flow/steric field-flow fractionation. Analytical Chemistry 59, 1957–1962. 31

Gillock, E. T., Rottinghaus, S., Chang, D., Cai, X., Smiley, S. A., An, K., Consigli, R. A.,

1997. Polyomavirus major capsid protein vp1 is capable of packaging cellular dna when

expressed in the baculovirus system. Journal of virology 71 (4), 2857–2865. 37

Grabow, W. O. K., 1996. Waterborne diseases: Update on water quality assessment and

control. Water SA 22 (1), 193–202. 14

Grabow, W. O. K., 1997. Hepatitis viruses in water: Update on risk and control. Water SA

23 (4), 379–386. 14
Bibliography 57

Grzenia, D. L., Carlson, J. O., Wickramasinghe, S. R., 2008. Tangential flow filtration for

virus purification. Journal of Membrane Science 321 (2), 373–380. 28

Grzenia, D. L., Wickramasinghe, S. R., Carlson, J. O., 2007. Ultrafiltration of parvovirus.

Separation Science and Technology 42 (11), 2387–2403. 28

Guo, Y., Cheng, A., Wang, M., Zhou, Y., 2009. Purification of anatid herpesvirus 1 par-

ticles by tangential-flow ultrafiltration and sucrose gradient ultracentrifugation. Journal

of Virological Methods 161 (1), 1 – 6. 12, 26

Guthmann, J., Klovstad, H., Boccia, D., Hamid, N., Pinoges, L., Nizou, J., Tatay, M.,

Diaz, F., Moren, A., Grais, R. F., Ciglenecki, I., Nicand, E., Guerin, P. J., 2006. A large

outbreak of hepatitis e among a displaced population in darfur, sudan, 2004: The role of

water treatment methods. Clinical Infectious Diseases 42 (12), 1685–1691. 14

Handlin, J., Trenor, S., Dado, G., 2009. Method for making sulfonated block copolymers,

methods for making membranes from such block copolymers and membrane structures.

US Patents: 2009/0280255. 46

Hanft, S., 2010. Ultrafiltration membranes: Technologies and the u.s. market. BCC Report

ID: MST044C. 44, 45

Harris, A., Cardone, G., Winkler, D. C., Heymann, J. B., Brecher, M., White, J. M.,

Steven, A. C., 2006. Influenza virus pleiomorphy characterized by cryoelectron tomog-

raphy. Proceedings of the National Academy of Sciences of the United States of America

103 (50), 19123–19127. 37


Bibliography 58

Henley, M., 2009. Biocontrol markets show continued gowth. Ultrapure Water 26 (3), 10–

13. 11

Hensgen, M., Czermak, P., Carlson, J., Wickramasinghe, S., 2010. Purification of minute

virus of mice using high performance tangential flow filtration. Desalination 250 (3),

1121 – 1124. 9, 12, 21, 29

Hesse, F., Wagner, R., 2000. Developments and improvements in the manufacturing of

human therapeutics with mammalian cell cultures. Trends in Biotechnology 18 (4), 173

– 180. 18

Hirasaki, T., Yamaguchi, K., Yanagida, K., Okuyama, K., 2006. Removal of small viruses

(parvovirus) from igg solution by virus removal filter planova20n. Journal of Membrane

Science 278 (1-2), 3 – 9. 35

Hirasaki, T., Yokogi, M., Kono, A., Yamamoto, N., Manabe, S., 2002. Removal and deter-

mination of dispersion state of bacteriophage [phi]x174 in aqueous solution by cupram-

monium regenerated cellulose microporous hollow fiber membrane (bmm®). Journal of

Membrane Science 201 (1-2), 95 – 102. 21

Ho, C., Zydney, A., 2003. Effect of membrane morphology on system capacity during

normal flow microfiltration. Biotech. Bioeng. 83 No.5, 537–543. 27

Hogan Jr., C. J., Kettleson, E. M., Lee, M., Ramaswami, B., Angenent, L. T., Biswas, P.,

2005. Sampling methodologies and dosage assessment techniques for submicrometre

and ultrafine virus aerosol particles. Journal of applied microbiology 99 (6), 1422–1434.

43
Bibliography 59

Hogan Jr., C. J., Kettleson, E. M., Ramaswami, B., Chen, D., Biswas, P., 2006. Charge

reduced electrospray size spectrometry of mega- and gigadalton complexes: Whole

viruses and virus fragments. Analytical Chemistry 78 (3), 844–852. 43

Huang, P., Peterson, J., 2001. Scaleup and virus clearance studies on virus filtration in

monoclonal antibody manufacture. Membrane separations in biotechnology, 327– 350.

ICH, December 2000. Ich topic q6a specifications: Test procedures and acceptance criteria

for new drug substances and new drug products. Chemical Substances. 36

John, K., Ann, L., Aaron, G., Brian, C., 2003. Methods of adenovirus purification.

WO/2003/097797. 47

Johnson, J., Chiu, W., 2000. Structures of virus and virus-like particles. Current Opinion

in Structural Biology 10 (2), 229 – 235. 37

Kitis, M., Lozier, J. C., Kim, J. H., Mi, B., Marinas, B. J., 2003. Evaluation of biologic

and non-biologic methods for assessing virus removal by and integrity of high pressure

membrane systems. Vol. 3. 21

Kolin, D. L., Wiseman, P. W., 2007. Advances in image correlation spectroscopy: Measur-

ing number densities, aggregation states, and dynamics of fluorescently labeled macro-

molecules in cells. Cell biochemistry and biophysics 49 (3), 141–164. 41

Korath, J. M., Abbas, A., Romagnoli, J. A., 2008. A clustering approach for the separa-

tion of touching edges in particle images. Particle and Particle Systems Characterization

25 (2), 142–153. 41
Bibliography 60

Kuo, J., 2007. Electron microscopy methods and protocols in molecular biology. Humana

Press, NewJersey. 37

Laine, J., Vial, D., Moulart, P., 2000. Status after 10 years of operation - overview of uf

technology today. Desalination 131, 17–25. 45

Lee, S. H., Levy, D. A., Craun, G. F., Beach, M. J., Calderon, R. L., 2002. Surveillance

for waterborne-disease outbreaks—united states 1999-2000. Morb. Mortal. Wkly. Rep.

Surveill. Summ. 51, 1–47. 15

Lenes, D., Deboosere, N., Menard-Szczebara, F., Jossent, J., Alexandre, V., Machinal, C.,

Vialette, M., 2010. Assessment of the removal and inactivation of influenza viruses h5n1

and h1n1 by drinking water treatment. Water Research 44 (8), 2473 – 2486. 22

Lengyel, J. S., Milne, J. L. S., Subramaniam, S., 2008. Electron tomography in nanoparti-

cle imaging and analysis. Nanomedicine 3 (1), 125–131. 37, 41

Li, N., Fane, A., Winston, W., Matsuura, T., 2008. Advanced Membrane Technology and

Applications. John Wiley and Sons, New Jersey. 33, 36

Lipin, D. I., Lua, L. H. L., Middelberg, A., 2008. Quaternary size distribution of soluble

aggregates of glutathione-s-transferase-purified viral protein as determined by asymmet-

rical flow field flow fractionation and dynamic light scattering. Journal of Chromatogra-

phy A 1190 (1-2), 204–214. 38

Lute, S., Riordan, W., Pease III, L. F., Tsai, D. ., Levy, R., Haque, M., Martin, J., Moroe,

I., Sato, T., Morgan, M., Krishnan, M., Campbell, J., Genest, P., Dolan, S., Tarrach, K.,

Meyer, A., Aranha, H., Bailey, M., Bender, J., Carter, J., Chen, Q., Dowd, C., Jani, R.,
Bibliography 61

Jen, D., Kidd, S., Meltzer, T., Remington, K., Rice, I., Romero, C., Sato, T., Jornitz,

M., Sekura, C. M., Sofer, G., Specht, R., Tarrach, K., Wojciechowski, P., Zachariah,

M. R., Tarlov, M. J., Etzel, M., Brorson, K., 2008. A consensus rating method for small

virus-retentive filters. i. method development. PDA Journal of Pharmaceutical Science

and Technology 62 (5), 318–333. 34

Manabe, S., 1996. Removal of virus through novel membrane filtration method. Develop-

ments in biological standardization 88, 81–90. 21

Martin, S., Hong, A., Jason, C., Dana, K., 2006. Process for removing protein aggregates

and virus from a protein solution. US Patents: 7118675. 48

Masayuki, H., Shin-ichi, M., 2009. Fluoresin polymer separation membrane and process

for producing the same. EP: 2047898. 46

Matthew, C., 2003. Viral purification methods. WO03/093463. 48

Matthew, C., 2007. Improved viral purification methods. US Patent: 7223585. 48

Merten, O., 2002. Virus contaminations of cell cultures - a biotechnological view. Cy-

totechnology 39 (2), 91–116. 14

Michalsky, R., Passarelli, A., Pfromm, P., Czermak, P., 2009. Purification of the bac-

ulovirus autographa californica m nucleopolyhedrovirus by tangential flow ultrafiltra-

tion. Desalination 245 (1-3), 694 – 700. 12, 21

Miranda, W., 2009. Virus purification using ultrafiltration. US Patents: 20090123989. 46


Bibliography 62

Nowak, T., Bernhardt, G., 2002. Removal of viruses from protein solutions by ultrafiltra-

tion. US Patents: 6391657. 49

Ogle, K. F., Azari, M. R., 2001. Virus removal by ultrafiltration: A case study with di-

aspirin crosslinked hemoglobin (dclhb). Membrane Separations in Biotechnology, 299–

326. 12

Otaki, M., Yaho, K., Ohgaki, S., 1998. Virus removal in a membrane separation process.

Water Science Technology 37, 107–116. 22

Pearce, G., 2007. Water and wastewater filtration: Membrane module format. Filtration

and Separation 44 (4), 31 – 33. 34

Pease III, L., Lipin, D., Tsai, D.-., Zachariah, M., Lua, L., Tarlov, M., Middelberg, A.,

2009. Quantitative characterization of virus-like particles by asymmetrical flow field

flow fractionation, electrospray differential mobility analysis, and transmission electron

microscopy. Biotechnology and bioengineering 102 (3), 845–855. 37, 42

Pelchen-Matthews, A., Marsh, M., 2007. Electron Microscopy Analysis of Viral Morpho-

genesis. Vol. 2007. 37, 41

Powell, M., Timperman, A., 2005. Quantitative analysis of protein recovery from dilute,

large volume samples by tangential flow ultrafiltration. Journal of Membrane Science

252 (1-2), 227 – 236. 26

Radosevich, M., Appourchaux, P., Huart, J., Burnouf, T., 1994. Nanofiltration, a new spe-

cific virus elimination method applied to high-purity factor ix and factor xi concentrates.

Vox sanguinis 67 (2), 132–138. 25


Bibliography 63

Ratanathanawongs, S., Giddings, J., 1993. Particle size analysis using flow field flow frac-

tionation. ACS Symposium Series 521, 13–29. 29

Rautenbach, R., Vobenkaul, K., 2001. Pressure driven membrane processes-the answer

to the need of a growing world population for quality water supply and waste water

disposal. Separation and Purification Technology 22-23, 193–208. 45

Ray, S., Tarrach, K., 2008. Virus clearance strategy using a three-tier orthogonal technol-

ogy platform. BioPharm International 21 (9), 50–59. 16

Robert, F., Deborah, F., Benedicte, L., van Reis, R., 2008. Non-affinity purification of

proteins. US Patents:7323553. 47

Roberts, P., 1997. Efficient removal of viruses by a novel polyvinylidene fluoride mem-

brane filter. Journal of virological methods 65 (1), 27–31. 25

Roda, B., Zattoni, A., Reschiglian, P., Moon, M., Mirasoli, M., Michelini, E., Roda,

A., 2009. Field-flow fractionation in bioanalysis: A review of recent trends. Analytica

Chimica ACTA 635 (2), 132–143. 42

Russell, E., Wang, A., Rathore, A. S., 2007. Process Scale Bioseparations for the Biophar-

maceutical Industry Ed. by Abhinav A. Shukla and Mark R. Etzel and Shishir Gadam.

CRC Press. 23

Schimpf, M., Caldwell, K., Giddings, J., 2000. Field flow fractionation handbook. John

Wiley and Sons, NewYork. 29

Schwinge, J., Wiley, D., Fane, A., 2004. Novel spacer design improves observed flux.

Journal of Membrane Science 229, 53. 33


Bibliography 64

Scutt, J., 1971. Virus retention by membrane filters. Water Research 5, 183–185. 11

Shortt, D. W., Roessner, D., Wyatt, P. J., 1996. Absolute measurement of diameter distri-

butions of particles using a multi-angle light scattering photometer coupled with flow

field-flow fractionation 21. 37

Soluk, L., Price, H., Sinclair, C., Atalla-Mikhail, D., Genereux, M., 2008. Pathogen safety

of intravenous rh immunoglobulin liquid and other immune globulin products: En-

hanced nanofiltration and manufacturing process overview. American Journal of Thera-

peutics 15 (5), 435–443. 9

Stefan, W., 1996. Filtration. WO/1996/000237. 48

Sun, W., Chen, T., Chen, C., Li, J., 2007. A study on membrane morphology by digital

image processing. Journal of Membrane Science 305 (1-2), 93–102. 41

Terrab, A., Pawlak, D., 2007. Validation of a prion and virus purification process in the

manufacture of bovine thrombin. Journal of Applied Research 7 (3), 217–226. 22

Teunis, P. F. M., Medema, G. J., Kruidenier, L., Havelaar, A. H., 1997. Assessment of the

risk of infection by cryptosporidium or giardia in drinking water from a surface water

source. Water research 31 (6), 1333–1346. 14

Thomas, J. J., Bothner, B., Traina, J., Benner, W. H., Siuzdak, G., 2004. Electrospray ion

mobility spectrometry of intact viruses. Spectroscopy 18 (1), 31–36. 43

Tkacik, G., 2007. Virus removal devices. EP:1218093. 49


Bibliography 65

Tkacik, G., Goddard, P., Kools, W., Nitin, S., 2010. Multilayer ultrafiltration composite

membranes and methods of making. EP:2163296. 49

Tkacik, G., Kazan, G., 2006. High resolution virus removal methodology and filtration

capsule. US Patents: 7108791. 48, 49

Trijzelaar, B., 1993. Regulatory affairs and biotechnology in europe: Introduction into

good regulatory practice - validation of virus removal and inactivation. Biotherapy 6 (2),

93–102. 20

Tsoka, S., Holwill, I., Hoare, M., 1999. Virus-like particle analysis in yeast homogenate

using a laser light- scattering assay. Biotechnology and bioengineering 63 (3), 290–297.

37

Tucelli, R., McGrath, P., 1996. Cellulosic ultrafiltration membrane. US Patents: 5522991.

46

Tullis, R., 2007. Method for removal of viruses from blood by lectin affinity hemodialysis.

US Patents: 7226429. 48

USP, 2008. Light diffraction measurements of particle size: General chapter. US Pharma-

copoeia. 36, 38

Vacura, K., 2008. Filtration industry overview: World pharma and biotechnology market

2008. Filtration and Separation 45 (8), 22–24. 44

van de Ven, W., Punt, I., Kemperman, A., Wessling, M., 2009. Unraveling ultrafiltration of

polysaccharides with flow field flow fractionation. Journal of Membrane Science 338 (1-

2), 67 – 74. 29
Bibliography 66

van Reis, R., 1996. Tangential flow filtration process and apparatus. US Patents 5490937.

28, 46

van Reis, R., 2000. Tangential flow filtration system. US Patents 6054051. 28, 46

van Reis, R., Brake, J., Charkoudian, J., Burns, D., Zydney, A., 1999. High performance

tangential flow filtration using charged membranes. Journal of Membrane Science 159,

133. 28

van Reis, R., Gadam, S., Frautschy, L. N., Orlando, S., Goodrich, E. M., Saksena, S.,

Kuriyel, R., Simpson, C. M., Pearl, S., Zydney, A. L., 1997. High performance tangen-

tial flow filtration. Biotechnology and bioengineering 56 (1), 71–82. 28

Walter, K., Noah, M., Scott, S., Jason, N., Jennifer, B., 2010. Water flux polymer mem-

branes. WO/2010/051150. 46

Wei, Z., Mcevoy, M., Razinkov, V., Polozova, A., Li, E., Casas-Finet, J., Tous, G. I.,

Balu, P., Pan, A. A., Mehta, H., Schenerman, M. A., 2007. Biophysical characteriza-

tion of influenza virus subpopulations using field flow fractionation and multiangle light

scattering: Correlation of particle counts, size distribution and infectivity. Journal of

virological methods 144 (1-2), 122–132. 42

WHO, 2008. Guidelines for drinking-water quality. World Health Organization: Geneva.

21

Wickramasinghe, S., B, K., Zimmermann, A., Thom, V., Reichl, U., 2005. Tangential

flow microfiltration and ultrafiltration for human influenza a virus concentration and

purification. Biotechnology and BioEngineering 92 No.2, 199–208. 38


Bibliography 67

Wilson, M., Mikhail, K., 2007. Ultrafiltration membranes rendered hydrophilic by hy-

droalkyl cellulose and method of making and use of such membranes. EP: 1775016A1.

46

Zydney, A., van Reis, R., 2001. Membrane separations in biotechnology. Current Opinion

in Biotechnology 12, 208–211. 29


Chapter 3

Experimental Section

Contents
3.1 Membrane Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.2 Characterization Techniques . . . . . . . . . . . . . . . . . . . . . . . . 72

3.2.1 Particle concentration measurement by inductively coupled plasma

technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.2.2 Particle morphology studies by transmission electron microscopy . 74

3.2.3 Particle size and distribution studies by dynamic light scattering

technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

3.3 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

Preface
This chapter gives a description of the cross flow rig details and materials used to

carry out the experiments. With a brief theoretical background, the methods employed for

concentration, particle size and morphology studies are also presented in this chapter.
3.1. Membrane Module 69

3.1 Membrane Module


The cross flow module employed in this work is shown in Figure 3.1. The whole

length of the module is approximately 264mm and is constructed in-house from perspex

blocks. The pressure transducer is inserted at the top half of the module. While running

the filtration experiment, bolts and wing nuts are used to fasten the two halves together.

The module also comprises a series of perforated aluminium plates to support the UF

membrane strip.

Figure 3.1: Tangential flow UF experimental setup employed for this study

All the experiments are conducted using this non-commercial experimental prototype

perspex cross flow module of membrane channel dimensions 2 mm*25 mm*210 mm. The

required feed pressure is obtained using the peristaltic feed pump (Master Flex Model

7529-00 by Cole Parmer). The pressure required is obtained by adjusting a valve situated

at the module outlet. A personal computer is used to record the data through the pressure
3.1. Membrane Module 70

transducers, flow transducer, a type-T thermocouple located upstream of the module.

Pressure transducers
The pressure transducers employed in the membrane configuration are PDCR 810 from

Druck (Davidsons, Australia). They produce an output voltage of 100 mV for 350 mbar

(i.e. corresponding to 35 kPa) and above, but which could be used for a range by a factor

of 6 (i.e. approximately 200 kPa) for this model with no change in the calibration factor.

The pressure measured and the raw voltage measured by the sensor are related as follows:

C f s Vm
Pressure(measured) = (3.1)
Vex CF

C f s - full scale capacity i.e. max. pressure which the transducer should receive

Vex - excitation voltage - recommended input voltage

Vm - raw voltage measured by the sensor

CF - calibration factor

The maximum deviation in the values obtained will be 0.1% and this means, it could

be ±2.1 kPa deviation in output for a pressure of 35 kPa.

Flow sensors
Flow sensor employed is from Radiospare (RS Stock number: 256-225) which could

measure upto 9.0 L/min and is suited for various liquids. The accuracy level obtained in

this sensor is 1% full scale (FS) and the repeatability is ±0.25 %.


3.1. Membrane Module 71

Membrane material
The membranes used for the entire experimental investigation are the poly ether sulfone

(PES) membranes. They are procured from Pall Corp from lots 8220B, 7265G. These UF

membranes are asymmetric in nature. The molecular weight cut off of the UF membranes

is 30,000 and 100,000 Da. Initially, a careful study is carried out with the samples to assess

the membranes so they do not completely reject all the particles as we are interested in

particle size fractionation and hence 100000 Da membrane is employed. Whereas, 30,000

Da membrane is employed just to compare the membrane effect on the filtration efficiency.

Sample material
Silica particles used for this study are procured from Nissan Chemicals, US and the

properties are reported in Table 3.1. The particles obtained are used as purchased condition

and the pH (8.5-10) is quite stable in this range and consistent as well.

Table 3.1: Characteristics of Silica Colloid.


SnowTex ZL SnowTex 20L SnowTex ST-50
Density, kg/m3 1305 1130 1380
Mass, gm 500 525 500
wt% SiO2 40.5 20.5 48
Concn, g/l 522 231.45 662.4
Mass, gm (for 1 1.814 4.08 1.43
g/l)
Particle size, nm 130-140 70-100 20-50
Refractive Index 1.46 1.46 1.46

The methods employed for the analysis are discussed in the following sections of this

chapter.
3.2. Characterization Techniques 72

3.2 Characterization Techniques


A brief outline of the methods employed for particle concentration, size and morphol-

ogy study is sketched here (Fig. 3.2). Particle concentration measurement by inductively

coupled plasma technique, particle morphology study by microscopic technique, whereas

and, particle size measurement by dynamic light scattering technique are discussed in the

following sections of this chapter.

Figure 3.2: An outline of various methods employed for the analysis of silica samples.

3.2.1 Particle concentration measurement by inductively coupled

plasma technique
Inductively Coupled Plasma - Optical Emission Spectroscopy (ICP-OES) is one of

several techniques available for elemental analysis which is capable of multi element de-

termination in solutions. It makes use of the fact that the atoms of elements can take up

energy from an inductively coupled plasma (a plasma is an electrically neutral, highly ion-

ized gas that consists of ions, electrons, and atoms), are thereby excited, and fall back
3.2. Characterization Techniques 73

into their ground state again emitting a characteristic radiation. The identification of this

radiation permits the qualitative analysis of a sample. A quantitative determination takes

place on the basis of the proportionality of radiation intensity and element concentration

in calibration and analysis samples.

In this technique, samples are fed through an auto sampler for unattended operation to

an argon plasma via a nebuliser. The Segmented-array, Charged-coupled-device Detector

(SCD) capable of detecting over 5000 emission lines in the range of 167 to 782nm, quanti-

fies atomic emissions in the plasma and WinLab32 software records the values diligently.

The standards are prepared at 0.5, 1, 10, 100 ppm and the internal standard used for

this sample is Yttrium. The sampler is washed with 2% HNO3 solution (nitric acid) and

the standards are tested. Subsequently, the samples are acquired with the aid of autosam-

pler and are fed to an argon plasma via a nebuliser and spray chamber using the integral

peristaltic pump. Though various nebulisers and spray chambers are available, but the

choice is highly dependent on the nature and quantity of the sample to be analysed. In this

analysis of silica, gemcone high dissolved solid nebuliser with cyclonic spray chamber is

employed. A solid-state radio frequency (RF) generator supplies RF energy at 1300W to a

coil around the horizontally mounted quartz torch to maintain the argon plasma, the (rela-

tively cool) end of which is removed by a shear gas - a thin ’wall’ of high-velocity air (for

samples run at axial view). Light from the plasma and the various atomic emissions therein

passes through windows ’viewing’ the plasma radially into the thermally stabilised, argon-

purged optical compartment housing an Echelle polychromator to separate the light into

its component wavelengths. This polychromator comprises diffraction gratings, mirrors,

lenses and visible and ultra-violet detector sets each comprising arrays of charge-coupled
3.2. Characterization Techniques 74

devices and allows simultaneous quantification of atomic emissions in the plasma.

In this work, Perkin Elmer OPTIMA 7300 DV ICP is employed to measure the concen-

tration of silica samples (feed and permeate). Samples are fed through an S10 auto sampler

to an argon plasma (15 L/min gas flow) via a nebuliser at a gas flowrate of 0.7 L/min. In

order to find out the instrument detection limit, three replicates are run of calibrated blank

and the detection limit is normally three times the standard deviation of 3 replicates. The

software records the values diligently which gives a ready to use concentration values.

3.2.2 Particle morphology studies by transmission electron mi-

croscopy
The transmission electron microscope (TEM), JOEL 1400 with 100 keV accelerating

voltage is used for the image analysis. This imaging study is mainly conducted to see the

morphology of silica particles. The silica sample is prepared by depositing silica nanopar-

ticles on 300 mesh copper grids and left for drying. The dried samples are then mounted

on the sample holder of the TEM to view the images. The TEM images clearly confirm the

silica particles are mostly spherical in nature (Figs. 3.3, 3.4, 3.5) which makes it suitable

for particle size study through light scattering technique.

Figure 3.3: TEM Micrograph of small silica particles at x200K magnification.


3.2. Characterization Techniques 75

Figure 3.4: TEM Micrograph of medium silica particles at x100K magnification.

Figure 3.5: TEM Micrograph of big silica particles at x200K magnification.

3.2.3 Particle size and distribution studies by dynamic light scattering

technique

Dynamic light scattering - Theory

The dynamic light scattering method is popularly employed to estimate particle size

and its distribution [Berne and Pecora, 1976]. It is also known as photon correlation spec-

troscopy and quasi-elastic light scattering, is non-invasive, highly sensitive and requires

very little sample volume (≤ 400 μl). In this technique, the fluctuations in monochromatic

light scattered from an ensemble of Brownian particles are measured with time. This fluc-

tuation contains information about the dynamics of the scattering particles. The angular

dependence of the scattering intensity arises from different positions on the same particle

and if the particle is large enough to accommodate multiple photon scattering, it is known

as Mie scattering. If the angular dependence of the scattering intensity is lost which hap-
3.2. Characterization Techniques 76

pens when the particle size is smaller than the wavelength of the incident light, multiple

photon scattering is avoided and is known as Rayleigh scattering. Here, the scattered light

intensity I(t) at a time t and I(t+τ) that at a time t+τ are measured. The similarity be-

tween two signals over a period of time is measured by autocorrelator. The autocorrelation

function C(τ) (Equations 3.2 & 3.3), and the normalized autocorrelation function, g’(τ)

(Equation 3.4), are calculated as follows:

C(τ) =< I(t) · I(t + τ) > (3.2)

C(τ) = Ae−2Γ + B (3.3)

C(τ) − C(∞)
g (τ) = (3.4)
C(0) − C(∞)

where Γ is the decay rate or relaxation rate (Equation 3.5) which is due to the transla-

tional, diffusive motion of the particles, D is the diffusion coefficient and q the scattering

wave vector.

Γ = Dq2 (3.5)

The scattering wave factor is given by the Equation 3.6:

q = 4πn/λo sin(θ/2) (3.6)

λo is the wavelength of the incident light, θ is the scattering angle, n is the refractive
3.2. Characterization Techniques 77

index of the solution. Thus if the fluctuation is measured and correlated, the diffusion

coefficient and the particle size can be obtained from the fitted parameter. In the case of

mono-dispersed particles, it is observed that the field-field autocorrelation function (given

in Equation 3.3) decays exponentially with a decay rate proportional to the diffusion coef-

ficient of the particles [Boschel et al., 2003; Frisken, 2001; Patty and Frisken, 2006] and

it produces its own unique autocorrelation function i.e. single exponential decay. In the

case of poly-disperse systems, the autocorrelation function is described by a distribution

function.

 ∞

g (τ) = G(Γ)exp(−Γτ)dΓ (3.7)
0

where G(Γ) is the normalized distribution function.

The diffusion coefficient depends on the inter-particle interaction, the viscosity (η) of

the solvent and temperature of the solution respectively. The effective diameter (hydrody-

namic or stokes diameter), d, and its speed due to Brownian motion can be simplified to

the Stokes-Einstein relation as follows:

kB T
d= (3.8)
3πηDo

where Do is the free particle diffusion coefficient which is obtained by linear extrapo-

lation of the D values of the diluted samples to zero concentration and kB is Boltzmann’s

constant.

Polydispersity index (PDI) is another useful measure which is obtained from the DLS

data. It is defined as the width of the decay rate and is obtained from Equation 3.9:
3.2. Characterization Techniques 78

PDI = μ2 /Γ2 (3.9)

where μ2 is the second moment and Γ, the decay or relaxation rate, depends on diffusion

coefficient and scattering wave vector (Equation 3.5). A PDI of 1 indicates large variations

in particle size and a reported value of 0 means that there is no variation in size. Thus,

this analysis yields two important values - a mean value for the size and a width parameter

known as the polydispersity index. Apart from the effective diameter and the polydispersity

index, the dynamic light scattering software uses algorithms to extract the decay rates for a

number of size classes to produce a size distribution. The basic measurement thus obtained

is intensity and all other distributions are generated from this using Mie theory.

In essence, the sample materials scatter incoming laser light. Due to the random motion

of these particles, the scattered light intensity fluctuates in time. Processing the fluctuating

signal with a state-of-the-art digital auto-correlator yields the particles diffusion coefficient,

from which the equivalent spherical particle size is calculated using the Stokes-Einstein

equation (Equation 3.8).

Instrumentation

A typical dynamic light scattering (DLS) system comprises six main components as

shown in Fig. 3.6. A laser is used to provide a light source to illuminate the sample

particles within a cell. Most of the laser beam passes straight through the sample, but

some are scattered by the particles within the sample. A photodiode detector or photo

multiplier is used to measure the intensity of the scattered light. As a particle scatters light

in all directions, theoretically its possible to place the detector in any position to detect

the scattering. The intensity of the scattered light must be within a specific range for the
3.2. Characterization Techniques 79

detector to successfully measure it. If too much light is detected then the detector will

become overloaded. To overcome this, an attenuator is used to reduce the intensity of the

laser and hence reduce the intensity of the detected light. In the case of smaller particles or

low concentration, the amount of light striking the detector should be maximized so that

the attenuator will allow more laser light through to the sample. For samples that scatter

more light i.e. larger particles or higher concentration, the amount of light striking the

detector must be decreased. The scattering intensity signal for the detector is passed to a

digital signal processing board called a correlator. The correlator compares the scattering

intensity at successive time intervals to derive the rate at which the intensity is varying.

This correlator information is then passed to a computer where the software analyzes the

data and derive size information.

Figure 3.6: Schematic diagram depicting the main components in a typical dynamic light
scattering setup.
3.3. Concluding Remarks 80

Particle size measurement by dynamic light scattering using Brookhaven particle

sizer

Brookhaven (BIC) 90Plus particle sizer is employed to measure the particle size with a

scattering angle of 90o , laser wavelength of 635 nm and the intensity at 35 mW, TurboCorr

digital autocorrelator. The particle shape was consistently spherical (from TEM images)

and hence the laser diffraction method response can be regarded as reliable. The particle

sizing software generates results in terms of the intensity, number, area and volume size

distributions along with derived statistics namely mean size and PDI. In this work, the data

presented throughout with respect to the particle diameter (average) has a standard error of

less than 3% of the mean of three repeated runs which indicates excellent repeatability.

3.3 Concluding Remarks


This chapter outlined the cross flow experimental setup as well as the silica material

being employed as model virus particles for the experimental investigation. The methods

employed to analyse the results were discussed which included dynamic light scattering

technique for particle size analysis, transmission electron microscopy for particle morpho-

logical and inductively coupled plasma technique for particle concentration. Results and

discussions will be presented in Chapter 4.

Bibliography
Berne, B., Pecora, R., 1976. Dynamic Light Scattering with applications to chemistry,

biology and physics. Wiley. 75

Boschel, D., Janich, M., Roggendorf, H., 2003. Size distribution of colloidal silica in
Bibliography 81

sodium silicate solutions investigated by dynamic light scattering and viscosity mea-

surement. Journal of Colloid and Interface Science 267, 360–368. 77

Frisken, B. J., 2001. Revisiting the method of cumulants for the analysis of dynamic light

scattering data. Applied Optics 40 No.:24, 4087–4091. 77

Patty, P., Frisken, B., 2006. Direct determination of the number weighted mean radius and

polydispersity from dynamic light scattering data. Applied Optics 45 no.10, 2209–2216.

77
Chapter 4

Influence of Particle Size and Size

Distribution on Filtration Efficiency and

Separation Mechanism

Contents
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

4.2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

4.3 Cross-Flow Filtration Experimental Details . . . . . . . . . . . . . . . . 89

4.4 Results and Discussion: Filtration Efficiency Study . . . . . . . . . . . . 89

4.4.1 Experimental design and surface response modeling . . . . . . . . 90

4.4.2 Pearson’s correlation analysis . . . . . . . . . . . . . . . . . . . . 92

4.4.3 Model prediction results . . . . . . . . . . . . . . . . . . . . . . . 95

4.4.4 Effect of process parameters and input condition on filtration efficiency 98

4.4.4.1 Effect of TMP on filtration efficiency . . . . . . . . . . . 98

4.4.4.2 Effect of CFR on filtration efficiency . . . . . . . . . . . 99

4.4.4.3 Effect of particle and pore size on filtration efficiency . . 101


83

4.4.5 Concluding remarks for filtration efficiency study . . . . . . . . . . 105

4.5 Results and Discussion - Influence of Particle Size and Size Distribution

on Filtration Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . 107

4.5.1 Study of particle size characterization based on particle concentration 109

4.5.2 Experimental design and statistical analysis . . . . . . . . . . . . . 112

4.5.3 Model prediction results . . . . . . . . . . . . . . . . . . . . . . . 116

4.5.3.1 Effect of TMP on permeate particle size and polydisper-

sity index . . . . . . . . . . . . . . . . . . . . . . . . . 116

4.5.3.2 Effect of CFR on permeate particle size and polydispersity

index . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

4.5.4 Particle size distribution (PSD) analysis . . . . . . . . . . . . . . . 119

4.5.4.1 Temporal effect . . . . . . . . . . . . . . . . . . . . . . 119

4.5.4.2 Effect of TMP on particle size distribution in permeate

streams . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

4.5.4.3 Effect of CFR on particle size distribution in permeate

streams . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

4.5.4.4 Modeling particle size distribution . . . . . . . . . . . . 125

4.5.4.5 Size polydispersity index and its effects . . . . . . . . . . 126

4.5.5 Concluding remarks for particle size and distribution investigation

section . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
84

Preface
There is limited work which analyzes and quantifies the effects of operating parameters

and particle size on final efficiency in general filtration operations, and this lack of work

is specifically more pronounced in tangential flow virus filtration. The effects of feed par-

ticle size, distribution and polydispersity are discussed to investigate this influence on the

filtration process and product property, aiming at filling the knowledge gap. This chapter

is divided into two sections considering the above key discussions, i.e. filtration efficiency

and particle size effects.

Research Questions:

1. How critical are the operating parameters and the allied input conditions in affecting

the filtration efficiency?

2. What is the role of cross interaction of the process parameters and their influence on

the filtration efficiency?

3. Could an empirical model be developed to quantify filtration efficiency via a mathe-

matical relation that takes into account the operating parameters?

The above questions necessitate the investigation of the following which is covered

under the Section 4.5 of this chapter.

1. What is the significance of feed particle size along with the operating parameters on

the permeate particle size and thereby the separation/fractionation?

2. What is the role of polydispersity index of feed on the permeate quality?


4.1. Introduction 85

Section 4.4 of this chapter has been presented in International Water Association -

Membrane Technology Conference, Beijing, 2009 [vii] and also published in Water Science

and Technology: Water Supply [ii]. Section 4.5 of this chapter is included in manuscript

submitted to (Biotechnology and Bioprocess Engineering [iii]).

4.1 Introduction
Water and wastewater treatment operations must strictly meet various standards to-

wards microbiological, toxicological, organoleptic, and operational regulatory limits. Spe-

cial attention is paid to the removal of microorganisms (bacteria, viruses, protozoa, cysts)

apart from organic and inorganic colloids, for a hygienically safe and microbiologically

stable drinking water [Graveland, 1998; Kitis et al., 2003; Lenes et al., 2010]. Specifi-

cally, virus removal is a critical operation as quantitative risk assessment has become a

mandatory step to project expected health risks of water treatment and distribution sys-

tems. There are various conventional methods employed for virus removal, but the use

of membrane filtration technology is observing a tremendous growth. This is due to the

concern over the quality of drinking water, regulatory pressure, the rapidly declining costs

of membrane systems, and consumer pressures for non-chemical treatment [Bodzek and

Konieczny, 1998; Laine et al., 2000; Pearce, 2007].

UF is recognized by the water industry as a very attractive technology for producing

drinking water especially since it is applicable to treating water sources that are of lower

quality, like ground water [Adham et al., 1998; Arnal et al., 2009]. The removal of virus

particles through UF has proven to be effective as evidenced by the wide acceptance of this

technology in the water industry. Several parameters influence the UF operation includ-

ing the membrane modules and materials, the operating conditions as well as parameters
4.1. Introduction 86

associated with the water source being treated. For example, tangential flow mode maybe

preferred for dilute feed where in the concentration polarization and cake formation are

expected to be negligible. As the virus particles range from parvovirus (18-24 nm), re-

ovirus (60-80 nm) to pseudorabies virus (120-200 nm) and can be either enveloped or

non-enveloped, the removal efficiency of virus particles is largely influenced by the virus

types as well. Though knowledge of virus particle behaviour in filtration is growing con-

tinuously, the passage/retention of such particles across a membrane is still not completely

understood [Chang et al., 1996, 2008].

Various studies have demonstrated how UF can be efficient in virus removal, with

removal efficiency ranging from 1.5 to 10 log reduction value (LRV) through various

membranes. For example, complete removal of poliovirus is achieved with 30 kDa UF

membranes [Madaeni et al., 1995] whereas MS2 removal with 100 kDa membrane could

achieve more than 6 log removal [Otaki et al., 1998]. In contrast, the removal of coli phage

Qβ and T4 was incomplete through a 20-40 kDa UF membranes showing 2.5 log removal

[Urase et al., 1998]. These works and others aim at just arriving at the log reduction values

but they do not address exactly the relationships between filtration efficiency and operating

parameters such as transmembrane pressure (TMP), membrane pore dimension or opera-

tion mode (dead end or cross flow modes). On the other hand, these works mainly try

to understand the underlying UF mechanisms through various models that are formulated

to describe permeate flux and fouling [Oh et al., 2007; Song and Elimelech, 1995] where

permeate flux is typically based on the different mass transfer mechanisms involved. There

still seems to be a lack of literature works that address the profound influence of virus par-

ticle size (PS) on the mobility and passage of particles across membranes and hence on the
4.2. Materials and Methods 87

final outcome of the filtered product [Han et al., 2003]. In the case of virus filtration where

the removal efficiency is measured in terms of log reduction value (LRV), it is of interest

to understand the influence of particle size along with other operating parameters on this

efficiency indicator.

We study virus removal using surrogate particles through cross flow UF aiming primar-

ily at understanding how model virus particle size influences the LRV, under the Section

4.4 of this chapter. We also examine the interaction of particle size effects with other vari-

able effects like those of TMP and feed flowrate (or otherwise termed cross flowrate in

cross flow mode). A systematic experimental programme based on statistical surface re-

sponse modelling approach is conducted. Two types of viruses are targeted by the model

particles, norovirus and enteric adenovirus. Norovirus has a particle size range of 27-38

nm and is the most common ecological agent for gastroenteritis outbreaks in all ages and

has significant public health impact worldwide. Enteric adenovirus falls in the range of

70-100 nm and is usually detected in the summer seasons.

4.2 Materials and Methods


A schematic diagram of the laboratory scale model of tangential flow filtration unit

used is shown in Fig. 4.1. The feed solution is held in a 10 litres tank and is fed into

the inlet of the membrane module by a peristaltic pump for which the speed can be ad-

justed and the feed suspension passing through the membrane is collected separately as

permeate and the particles which fail to pass through is collected as retentate. From the

pressure transducers, the pressure values are recorded and the transmembrane pressures

are evaluated using Equation 2.4. In this work, there are two response factors measured

i.e. particle diameter and particle concentration. With respect to particle diameter, the data
4.2. Materials and Methods 88

presented throughout corresponds to the average diameter. This is obtained by calculat-

ing the mean of three repeated runs and it is found that the standard error is less than 3%

of the mean of three repeated runs. This indicates excellent reproducibility. With regard

to the concentration measurement (for LRV calculation), the instrument detection limit is

three times the standard deviation of three replicates of calibration blank which is less than

1% demonstrating the reliability of the concentration data. As far as experimental data,

initially number of experiments was performed to identify the range of transmembrane

pressure, and cross flow rate in order to obtain reasonable permeate data. During these ini-

tial runs, experiments were run to check the repeatability and were found that the variation

in the data was less than 3% which demonstrated the repeatability.

Figure 4.1: Schematic diagram of the tangential flow filtration experimental setup.

Silica particles used for this study are procured from Nissan Chemicals, US and the

properties are summarized in Table 3.1 of Chapter 3. The particles obtained are used as

purchased condition. Three different particle size ranges are used for this study: smaller
4.3. Cross-Flow Filtration Experimental Details 89

particles (20-50 nm), medium particles (70-100 nm), and bigger particles (130-140 nm).

4.3 Cross-Flow Filtration Experimental Details


It is common to find UF operations working in the cross flow mode (alternatively

termed tangential flow). This mode of operation is presented in Fig. 4.2 which depicts

the mobility, across the membrane, of particles either permeating through the membrane

or getting retained and being convectively collected as retentate.

Figure 4.2: Cross flow filtration: Sample flowing parallel to the membrane surface.

Cross flow filtration has received attention for its fouling reduction capabilities in re-

moving molecules or particles from the membrane surface by the tangential flow of liquid.

There are mainly three related (input) variables that play an important role in the effective-

ness of filtration namely transmembrane pressure, cross flow, and feed particle size. Key

performance (output) variable considered for this investigation is the log reduction value

which is discussed in the next section.

4.4 Results and Discussion: Filtration Efficiency Study


Various terms have been used to represent the microbial removal efficiency of mem-

brane filters. Initially, Beta ratio was the term used to describe the retention of organisms

from hydraulic fluids of fine particles [Leahy and Sullivan, 1978; Reti, 1977]. Subse-

quently, the term titer reduction T R - the ratio of particles in the influent stream to those in

the effluent, the class definition of filter efficiency was used in membrane filtration [Pall and
4.4. Results and Discussion: Filtration Efficiency Study 90

Kimbauer, 1978]. Currently, the term in common use is log reduction value (LRV) adapted

by Health Industry Manufacturing Association (HIMA) to standardize the terminology.

The LRV is the logarithm to the base of 10 of the ratio of the organisms in the influent

stream to those that emerge in the filtrate/permeate. In order to determine the LRV, it is rec-

ommended that the original influent stream along with the input conditions are chosen such

that it causes some passage of organisms through the filtration system for investigational

purposes. Whereas, few applications, like water purification industry and biotechnology

industry, demand the undesirable particles to be minimum or close to zilch in the permeate

stream. So, these indicators are typically dependent on the type of application, operation

(removal or concentrating), and meeting respective regulatory requirements, Thus to sum-

marize, the efficiency is best described in the filtration system as log reduction value (LRV)

or alternatively as filtration efficiency. Thus LRV being a direct indicator of virus filtration

is defined as:

Cf
LRV = log10 (4.1)
Cp

where C f and C p are the concentrations of the feed and permeate streams respectively.

This equation suggests that for higher LRV values corresponding to higher removal effi-

ciencies, it is desired to have the value of C p much smaller than C f . This term, LRV or

filtration efficiency, is constantly recalled at various sections in the whole of this thesis.

4.4.1 Experimental design and surface response modeling


The experimental design used for this regression modeling of cross flow UF of silica

particles was carried out using three factors viz. transmembrane pressure (TMP), cross
4.4. Results and Discussion: Filtration Efficiency Study 91

flowrate (CFR) (process variables) and the particle size (FPS) of the feed stream (input

condition). These three factors signify the independent variables that are varied in the

experiments. Statistical experimental design is used in this investigation to map the system

and thus approximate the filtration efficiency by a quadratic model. Such model allows us

to (a) analyse and understand in more detail how the varied factors influence the response

and (b) make predictions of filtration efficiency under different operating conditions. A full

factorial three level experimental design is used to analyse the data. The range of values

actually used for experimentation and their respective coded values are shown in Table 4.1.

The experimental LRVs obtained are also listed in Table 4.1.

Table 4.1: Full Factorial, 3 level experimental design for cross flow UF of silica particles
using 100 kDa MWCO membrane (b : coded variables).
Sample ID Factors (input variables) Response
Transmembrane Pressure Cross flowrate Particle Size
TMP, Levelb CFR, Levelb FPS, nm Levelb LRV
kPa X1 L/min X2 X3
1 20 -1 1 1 Small -1 2.26
2 20 -1 1 1 Medium 0 2.6
3 20 -1 1 1 Big 1 2.71
4 40 0 1 1 Small -1 2.16
5 40 0 1 1 Medium 0 2.41
6 40 0 1 1 Big 1 2.58
7 60 1 1 1 Small -1 1.63
8 60 1 1 1 Medium 0 2.34
9 60 1 1 1 Big 1 2.41
10 20 -1 0.6 0 Small -1 2.24
11 20 -1 0.6 0 Medium 0 2.56
12 20 -1 0.6 0 Big 1 2.66
13 40 0 0.6 0 Small -1 2.12
14 40 0 0.6 0 Medium 0 2.35
15 40 0 0.6 0 Big 1 2.54
16 60 1 0.6 0 Small -1 1.51
17 60 1 0.6 0 Medium 0 2.19
18 60 1 0.6 0 Big 1 2.02
19 20 -1 0.3 -1 Small -1 2.18
20 20 -1 0.3 -1 Medium 0 2.26
21 20 -1 0.3 -1 Big 1 2.4
22 40 0 0.3 -1 Small -1 2.14
23 40 0 0.3 -1 Medium 0 2.2
24 40 0 0.3 -1 Big 1 2.4
25 60 1 0.3 -1 Small -1 1.44
26 60 1 0.3 -1 Medium 0 1.91
27 60 1 0.3 -1 Big 1 1.98

In the surface response modeling exercise of this section, only one response is con-
4.4. Results and Discussion: Filtration Efficiency Study 92

sidered that being the filtration efficiency i.e. LRV. A multilinear regression model was

fitted to the data considering the three independent variables and the one dependent re-

sponse. This model matched the experimental data well compared to other regression

models. Also, ANOVA analysis indicates that the standard error of estimate is smaller

than the response values and the model predictions are normally distributed, hence this

model is chosen. Thus, the resulting empirical model after eliminating insignificant terms

is given by the Equation 4.2. The equation indicates that X1 (TMP) has a negative influ-

ence whereas X2 and X3 (CFR and FPS) have a positive influence on LRV. It also indicates

that the cross interaction between TMP and FPS significantly affects LRV.

LRV = 2.42 − 0.247X1 + 0.122X2 + 0.223X3 + 0.062X1X3 − 0.139X12 − 0.126X32 (4.2)

4.4.2 Pearson’s correlation analysis


Pearson’s correlation is employed to arrive at the relationship of the independent vari-

ables on LRV (Table 4.2); here increase in transmembrane pressure results in decrease in

LRV (negative correlation: -0.629) whereas increases in the other two factors i.e. cross

flowrate (positive correlation: 0.310) and particle size (positive correlation: 0.569), shows

to have positive influence on the filtration efficiency. The Pearson correlation suggests

that the cross flowrate has a relatively less effect on the LRV (significance value is 0.116)

whereas the TMP and FPS (significance values are less than 0.05) play a significant role

on the control of the filtration efficiency.

The analysis of variance (ANOVA) (Table 4.3) indicates the significance of the multi-
4.4. Results and Discussion: Filtration Efficiency Study 93

Table 4.2: Pearson’s correlation analysis.


Coefficient Significance
LRV 1 -
TMP -0.629 0
CFR 0.31 0.116
FPS 0.569 0.002
T MP ∗ CFR 0.055 0.783
CFR ∗ FPS 0.109 0.588
T MP ∗ FPS 0.128 0.524
T MP2 -0.204 0.307
CFR2 -0.03 0.881
FPS 2 -0.185 0.356
TMP*CFR*FPS 0.002 0.992

ple regression model. The mean square for the regression model in comparison with the

residual demonstrates that the regression model is highly significant and also supported by

the F test value (19.251) at P=0 (i.e. P is less than 0.001). The summary section gives

the coefficient of determination (R2 = 0.923) which indicates that 92.3% of the variation in

LRV is explained by variation in the independent variables (TMP, CFR, FPS), and the R

value (0.961) indicates a strong correlation between LRV and the factors. The mean value

of LRV (2.23) is much larger than the standard error of estimate (0.115) which clearly

defines the suitability of the multiple regression model for the variance of LRV, within the

tested independent variable ranges.

Table 4.3: Analysis of variance summary.


ANOVA
Source Sum Sq. D.F. Mean Sq. F Prob.
Regression 2.56 10 0.256 19.251 0
Residual 0.213 16 0.013
Total 2.772 26
Summary
R2 R Adj. R2 Standard Error of Estimate
0.923 0.961 0.875 0.115

The quadratic regression model developed based on the experimental design data is

given in Equation 4.2. Through student’s t test, three other terms representing the TMP

and FPS quadratic terms and the TMP-FPS interaction term were, besides the three inde-
4.4. Results and Discussion: Filtration Efficiency Study 94

Figure 4.3: Log reduction values: predicted vs experimental.

Figure 4.4: Rankit Plot showing the residuals are normally distributed.
4.4. Results and Discussion: Filtration Efficiency Study 95

pendent factors, identified to be significant.

The model is validated statistically as the predicted R2 (0.923) (shown in parity plot of

Fig. 4.3) is in agreement with the adjusted coefficient of determination (0.875). The Rankit

plot (Fig. 4.4) indicates that the residuals are normally distributed and therefore very low

probability of non-conformance suggesting only minor departure of the residuals from the

normal.

4.4.3 Model prediction results


The model (Equation 4.2) developed based on the surface response method demon-

strates the effect of TMP, CFR and FPS on the filtration efficiency. The model is used to

simulate under different TMP and CFR conditions of 20-60 kPa and 0.3-1.0 L/min range.

Figs. 4.5, 4.6 & 4.7 are shown to demonstrate the model predictions for varying CFR at all

three TMPs.

Figure 4.5: Model prediction showing the effect of cross flowrate at 20 kPa on LRV.

The trend of the predicted plots clearly indicates that model agrees well with experi-

mental results (Fig. 4.8). For comparison purposes, the plots are shown at 0.6 L/min cross
4.4. Results and Discussion: Filtration Efficiency Study 96

Figure 4.6: Model prediction showing the effect of cross flowrate at 40 kPa on LRV.

Figure 4.7: Model prediction showing the effect of cross flowrate at 60 kPa on LRV.
4.4. Results and Discussion: Filtration Efficiency Study 97

flowrate and for varying TMP. Specifically, it is noted that smaller and bigger particles

agree very well but in the case of medium particles deviation is observed nevertheless it

differs by 10% which is well within the acceptable range. From the statistical point of

view, all statistical estimators (i.e. R2 is in agreement with adj. R2 , F has high value, Prob.

has low value) in our empirical model demonstrate the model’s validity. At the same time,

it is important to note that the standard error of estimate is 0.115 which has led to the

slight deviation. Also, in the model only the interaction effects of significant terms have

been retained (i.e. TMP and FPS) while ignoring the interaction effects between the other

factors (i.e. TMP and CFR, CFR and FPS) according to the results of Pearson’s correlation

analysis. This may also be a reason for the slight deviation. Thus, this model could well

be employed to predict the filtration efficiency for a given set of operating conditions in

the studied experimental range.

Figure 4.8: Model prediction vs experimental trend for varying TMP at 0.6 L/min.
4.4. Results and Discussion: Filtration Efficiency Study 98

4.4.4 Effect of process parameters and input condition on filtration

efficiency
The present study as mentioned earlier is initiated to investigate the relationship be-

tween selected parameters (TMP, CFR, FPS) and filtration efficiency. The reason behind

the variation of filtration efficiency is due to the partial permeation of particles with UF

membranes. It can be explained based on the following reasons: (i) literature indicates that

UF membranes come with their specified pore size range but normally it does not include

the presence of abnormally large pores [Amirilargani et al., 2009; Sun et al., 2007; Urase

et al., 1998]. At the same time, pore size distribution parameters (like pore size, standard

deviation) actually vary depending upon the zones viz. zone of long tail of big pores, zone

of standard linearization of pore size distribution, and zone of sieving effect. (ii) one other

possible explanation is due to lack of membrane integrity due to the presence of pinholes

[Kitis et al., 2003]. Thus, the interaction between the varying particle sizes and the pore

size is a big challenge in arriving at the final filtration efficiency values. Hence, initiating

a study on the effect of the particle size alongwith the process parameters will be a big

step towards a better understanding of filtration efficiency mechanism. Incorporating the

membrane pore variations and its uncertainties will certainly be a complete compliment in

knowing the separation mechanism but which is not covered in this thesis.

4.4.4.1 Effect of TMP on filtration efficiency

The experimental results of tangential flow UF of silica samples indicate that the UF

polyethersulfone membranes form an efficient or selective barrier depending upon the par-

ticle size and varying operating parameters. Runs were carried out at different TMPs (20,
4.4. Results and Discussion: Filtration Efficiency Study 99

40, 60 kPa). Increasing TMP results in a reduction in removal efficiency. It is likely that

the higher TMP enhances the membrane pore elongation and thereby allowing more par-

ticles to pass through the membrane thereby decreasing the LRV (Figs. 4.9, 4.10, 4.11 &

4.12).

Figure 4.9: Experimental results showing the effect of TMP at 0.3 L/min on LRV (lines
drawn to indicate trends).

4.4.4.2 Effect of CFR on filtration efficiency

Similarly, runs were carried out at three cross flowrates (0.3, 0.6, 1 L/min). Increasing

the feed flowrates was found to increase the filtration efficiency (Fig. 4.13). It is also

important to note that CFR has less influence on the filtration efficiency than TMP and

FPS. This effect could be attributed to the fact that at higher feed flowrate, the residence

time of particles across the surface of the membrane is reduced limiting the probability

of particle permeation and hence reducing the concentration of particles in the permeate

stream. The improvement in LRV is observed at lower TMP and higher flowrates which

indicates the importance of selection of optimal TMP and flowrate values to achieve better
4.4. Results and Discussion: Filtration Efficiency Study 100

Figure 4.10: Experimental results showing the effect of TMP at 0.6 L/min on LRV (lines
drawn to indicate trends).

Figure 4.11: Experimental results showing the effect of TMP at 1 L/min on LRV (lines
drawn to indicate trends).
4.4. Results and Discussion: Filtration Efficiency Study 101

Figure 4.12: Model prediction showing the effect of TMP at 0.6 L/min on LRV.

efficiency.

4.4.4.3 Effect of particle and pore size on filtration efficiency

The UF membrane is characterized by the presence of numerous pores of varying sizes

and the final permeate quality is greatly influenced by the physical dimensions of the pore

and the particle size. In UF membranes its conventional to use MWCO wherein the par-

ticles that have molecular weights lesser than the rated membrane pore molecular weight

cutoff (MWCO) point (or dimensions smaller than the pore diameter) is expected to pass

through or fractionate which highly influences the LRV/filtration efficiency.

For a better insight, it is necessary to use a model that relates the filtration efficiency

and particle to pore size (MWCO in the case of UF) ratio. Typically, Ferry-Faxen equa-

tion (Equation 4.3) describes the theoretical relation between the rejection coefficient and

particle diameter to pore rating [Chao and Tojo, 1987; Torras et al., 2006] and is given as:
4.4. Results and Discussion: Filtration Efficiency Study 102

Figure 4.13: Experimental results showing the effect of feedflow rate at 40 kPa on LRV
(lines drawn to indicate trends).

1−R = (1−PPS R)2 [1−0.104PPS R−5.21PPS R2 +4.19PPS R3 +4.18PPS R4 −3.04PPS R5 ]

(4.3)

where R is the retention coefficient and PPSR is the particle-to-pore size ratio. Combining

the definition of R and LRV, we get

 
1
LRV = log10 (4.4)
1−R

For comparison studies, Equation 4.3 is modified to suit our study by introducing the

above relation (Equation 4.4). However, it is important to note that Ferry-Faxen model

describes the retention characteristics considering only the particle and pore radius but fails

to include the operating parameters which have a great deal of influence on the retention

mechanism. Hence, Ferry Faxen model though qualitatively agrees with our experimental
4.4. Results and Discussion: Filtration Efficiency Study 103

results, but quantitatively deviates (Fig. 4.14).

Figure 4.14: A comparison of experimental results (points joined by dotted lines to indicate
the trend) with our model and Ferry-Faxen model.

This necessitates to extend the model to incorporate the effect of TMP and CFR apart

from PPSR on the final filtration efficiency. Our data and statistical investigation have

clearly demonstrated that LRV depends not only on the feed particle size but also on TMP

and CFR as well which has already been discussed. Hence, here we have developed an

empirical model to describe the relationship between log reduction value and the operating

parameters (TMP and CFR) along with the PPSR (Equation 4.5). A multi-linear regression

model thus obtained is based on our experimental results and is given as:

LRV = 2.15 − 0.01T MP + 0.34CFR + 0.67PPS R (4.5)

Thus, in order to validate the above model, a lower MWCO (30,000 Da) membrane

was used in a further test to assess the level of improvement in particle removal. Fil-

tration efficiencies for the two membranes are shown in Table 4.4 for the three different
4.4. Results and Discussion: Filtration Efficiency Study 104

particle size ranges (small, medium, big) and in Fig. 4.15 as a function of the particle-to-

pore-size ratio (PPSR). For a given membrane, the data shows that LRV increases as the

particle size range being filtered increases. This is expected since UF is a size-exclusion

particle removal process where the membrane would reject the particles more efficiently

if the particles are bigger in size. The interaction between particles and the membrane is

demonstrated in Fig. 4.15 showing the filtration efficiency behavior with PPSR. This figure

clearly shows the improvement in LRV (hence increased particle removal) brought about

by the 30 kDa MWCO membrane, but more significantly, it shows continuity across the

two membranes in the LRV trend under increasing PPSR; both membranes observing very

similar LRV values (2.8 and 2.71) at a PPSR near 0.84. Here we note the value in report-

ing LRV data against PPSR in helping to compare membranes of different pore sizes. It

is also important to note the observed deviation between our model and the experimental

data showing that at higher PPSR values, the interaction between particles and pores are

said to be statistically insignificant but which the model is incapable of defining, hence the

deviation.

Table 4.4: Efficiencies (LRV) for the two MWCO membranes against particles of different
size ranges. Data for lower TMP (20 kPa) and higher flowrate (1 L/min).

Membrane pore size, Da Particle size category LRV


30,000 MWCO Smaller Particles 2.8
Medium Particles 4.1
Bigger Particles 4.6
100,000 MWCO Smaller Particles 2.26
Medium Particles 2.6
Bigger Particles 2.71

From the above study, it drives to an insight that the blocking of pores is statistically

insignificant if PPSR is greater than 1. Thereby, bigger particles efficiently being rejected
4.4. Results and Discussion: Filtration Efficiency Study 105

Figure 4.15: Efficiencies (LRV) for the two MWCO membranes as a function of the PPSR.
Data for lower TMP (20 kPa) and higher flowrate (1 L/min).

which increases the LRV. Thus, this study is a significant leap proposing a predictive em-

pirical model (Equation 4.5) to achieve a target LRV value for a given TMP and CFR with

a known virus particle size range.

4.4.5 Concluding remarks for filtration efficiency study


The model results (Figs. 4.12 & 4.6) and the experimental results (Figs. 4.10 & 4.13)

indicated that particle size, besides the operating parameters of TMP and flowrate, has a

great deal of influence on surrogate virus particle removal in UF. The proper selection of

operating parameters coupled with proper membrane selection will result in better rejec-

tion and/or separation. Having identified the effect of particle size as a more prominent

factor in the filtration process, a detailed investigation on the feed particle size on the per-

meate quality was carried out which is discussed in the next section. This investigation

showed that among the two operating parameters considered, TMP plays a major role in

controlling the filtration efficiency in comparison with flowrate. In the studied experimen-
4.4. Results and Discussion: Filtration Efficiency Study 106

tal range, higher LRV values were obtained at lower TMP (20 kPa) and at higher feed

flowrate (1 L/min). Further, the effect on LRV of the interaction between TMP and FPS

seems to be more significant than that of the interaction of flowrate with FPS. The empir-

ical statistical model developed serves as a powerful tool for studying the impact of the

key operating parameters on the permeate quality (Figs. 4.9 & 4.13) and can be employed

for optimisation of operating parameters to achieve maximum efficiency. The above find-

ings are hereby summarized thus answering the research questions posed at the start of the

chapter:

• The feed particle size, transmembrane pressure and cross flowrate have a significant

role in the final filtration efficiency. Among the above three parameters considered,

FPS and TMP are seen to be major key factors (with very high significance based

on the Pearsons correlation analysis) with CFR’s significance being relatively less.

It is very insightful to recognise that the interplay of parameters plays an equally

important role as well. It has been identified that interplay of TMP and FPS are

more influential than TMP and CFR or CFR and FPS. This necessitated a further

investigation on the feed particle size on the separation or fractionation mechanism.

This is presented in Section 4.5

• The effect of above parameters were formulated in an empirical model development.

The model results obtained were compared with the experimental results and found

to be in excellent agreement. Hence, the model obtained from this study can be

employed to predict the filtration efficiency given the TMP and CFR and the feed

particle property within the studied experimental range.


4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 107
• The above emprirical modeling study demonstrated the dependencies of these pa-

rameters on the filtration efficiency and this serves as an excellent approach/method

to analyse the effects of these parameters on product efficiency for filtration pro-

cesses in general and in the context of virus UF in specific. Moreover, this approach

can be applied for scaled-up processes.

4.5 Results and Discussion - Influence of Particle Size and

Size Distribution on Filtration Mechanism


The efficiency of tangential flow filtration is highly influenced by operating parameters

and the feed particle size as has been demonstrated in the study presented in the previ-

ous section (i.e. Section 4.4). In this section, particle size and distribution are primarily

considered and their effects along with the effects of the operating parameters on the per-

meate particle size are investigated. Model virus particles (silica particles) employed for

this study fall in a similar range of sizes mimicing viruses including reovirus, pseudorabies

virus, adenovirus, herpes simplex viruses, and murine retro viruses. Thus, the use of these

model particles is to give a generic validation for virus separation by UF.

Other than application in the water industry, UF membranes are targeted for purifica-

tion of biological solutions wherein they are employed to fractionate proteins with similar

sizes, achieving almost a complete separation/rejection of virus particles or a particular

protein [Lightfoot and Moscariello, 2004; Saxena et al., 2009]. Hence, it is considered to

be a complex mechanism and a combination of fractionation (i.e. similar sized particles

getting separated) and/or rejection (smaller particles being separated from bigger ones).

Previously, it was assumed to be difficult to separate particles if they differ by less than
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 108
an order of magnitude of similar sizes. Whereas recent studies clearly indicate that by

carefully choosing the operating conditions, suitable mechanism (i.e. protein fractionation

or viral clearance) could be achieved [van de Ven et al., 2009; van Reis et al., 1997]. For

example, fractionation of proteins is efficiently achieved by carefully selecting the ionic

strength and pH [Arkhangelsky et al., 2008; Saksena and Zydney, 1994]. Similarly, elec-

trostatic and electrokinetic interactions have been shown to have a great deal of influence

on the fractionation mechanism of proteins up to 100 fold [Pujar and Zydney, 1994]. As

far as the authors know, very limited literature is available to explain or understand frac-

tionation and separation mechanism considering virus particle size and distribution as well

as virus particle interactions with the UF operating parameters. Specifically, the study with

respect to the filtrate stream is still lacking. Hence, it is very vital to bridge the knowledge

gap in order to meet the regulation requirements of the end products [Fischer and Raasch,

1985; Lu and Hwang, 1995; Zhou et al., 2008]. Number of literatures address the flux de-

cline behavior considering particle characteristics of the feed [Bakhshayeshi and Zydney,

2008; Foley et al., 1995b; Hwang et al., 1998; Lu and Ju, 1989; Mackley and Sherman,

1992b; Saxena et al., 2009]. Whereas, the relationship between the particle properties of

feed and permeate alongwith the filtration efficiency (LRV) is still not adequately estab-

lished. Hence, the approach of correlating these factors will help to initially observe the

dependencies and then to understand the underlying mechanisms.

This section targets to cover the following objectives (1) to examine the size distri-

bution of the model virus particles: pre-filtration and post-filtration by the dynamic light

scattering technique (2) to investigate the impact of the size distribution on the filtration

mechanism and (3) empirical model development for permeate particle size (PS p ) and
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 109
polydispersity index (PDIP ) which aids in identifying specific dependencies of the UF

efficiency on feed particle size as well as on operational conditions.

4.5.1 Study of particle size characterization based on particle concen-

tration
Initially, a careful study is conducted with the silica samples (recall Table 3.1) to es-

tablish the most appropriate concentration for the measurement of particle size and the

distribution of the pre-filtration and post-filtration samples. According to the manufac-

turer’s specification (Brookhaven (BIC) 90 Plus Particle Sizer), one of the indicators of

the quality of the data being collected is the average count rate which is a measure of the

signal intensity in kilo counts per second (kcps). The auto-correlation plot along with the

average count rate guides the analysis as they are the closest parameters to the raw/actual

data.

The particle sizing software is capable of producing results in terms of multimodal

distribution and log normal distribution (Fig. 4.16). The log normal distribution has been

widely used until the multimodal algorithm was available and lognormal distribution offers

a simplified representation. On the contrary, multimodal size distribution gives more infor-

mation like the groups of particles of varying sizes and BIC employs the Non-Negatively

Constrained Least Squares (NNLS) algorithm for data representation.

In this section, the data presented throughout refers to the effective diameter with re-

spect to the particle size and the multimodal size distribution with respect to the particle

size distribution. The standard error of the average diameter is less than 3% of the mean

of the three repeated runs which indicates excellent repeatability. The correlation function

plot immediately gives an idea about the sensitivity to the range of particles and applicabil-
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 110

Figure 4.16: Two different representation of particle size distribution of the original feed
sample: multimodal size distribution (top) and log normal distribution (bottom).

ity for the test sample (Fig. 4.17). In order to maintain the uniformity and consistency for

the analysis, the particle size distributions are presented in terms of intensity weighted dis-

tributions as it forms the basic measurement technique. The intensity weighted distribution

is defined as the intensity of light scattered by the suspension of particles with diameter

(say d) which depends on size, scattering angle, index of refraction, and wavelength.

The characterization of silica particles at various concentrations is carried out to find

out the suitability and reliability for the experimental runs. It is observed that at very high

concentrations, the instrument reported a very high count rate (of Mcps order) which indi-

cates those values are unreliable for all the three particle size ranges. From the measured

values of reliable signal intensity, it is concluded that the recommended concentration for

the measurement is to be less than 5 g/L for smaller particles, 10 g/L for bigger particles

and 50 g/L for medium particles (Fig. 4.18).


4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 111

Figure 4.17: Illustration of the correlation function - Exponential decay of auto-correlation


function.

Figure 4.18: Counts per second versus concentration plot for three different particles. This
shows the range of suitable concentration below 800 kcps threshold as specified by the
BIC particle sizer.
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 112

4.5.2 Experimental design and statistical analysis


The experimental design used for this regression modeling of cross flow UF of silica

particles is carried out using three process variables viz. transmembrane pressure (TMP),

cross flowrate (CFR), feed particle size (FPS). These three factors represent the indepen-

dent variables that are controlled in the experiments. Statistical experimental design is used

in this investigation to map the system and thus approximate the responses by a regression

model. Such a model will allow us to (a) analyze and understand in more detail how the

factors influence the permeate particle size and polydispersity index, (b) make predictions

of permeate particle size and polydispersity index under different operating conditions in

the studied experimental range. The range of values actually used for experimentation and

their respective coded values are shown in Table 4.5. In this study, the responses are the

permeate particle size and polydispersity index. A multiple regression model is used to

establish an empirical relationship between the factors and the responses. The cross flow

unit is operated in the range of 20-60 kPa transmembrane pressure, with cross flowrate in

the range of 0.3-1.0 L/min for the feed particle size range 35.6-142.9nm.

In the surface response modeling exercise of this study, the resulting multiple regres-

sion model giving the relationship of permeate particle size (PS p ) and the permeate PDI

(PDI p ) with the operating parameters (TMP, CFR and FPS) after eliminating insignificant

terms are given as follows (Equations 4.6 & 4.7):

PS p = a0 + a1 T MP + a2CFR + a3 FPS + a4CFR2 + a5 FPS (4.6)


4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 113

Table 4.5: Full Factorial, 3 level experimental design for cross flow UF of three particle
size ranges of silica colloid.
Factors (input parameters) Responses (Permeate stream)
Transmembrane Pressure (TMP) Cross Flowrate (CFR) Feed Particle Size (FPS) Observed
Expt. TMP, kPa Levelb X1 CFR, Levelb X2 FPS, nm Levelb X3 Dia, nm PDI
No. L/min
1 20 -1 1 1 46.5 -1 42.4 0.27
2 40 0 1 1 46.5 -1 37.7 0.22
3 60 1 1 1 46.5 -1 36.3 0.22
4 20 -1 0.6 0 46.5 -1 42.2 0.28
5 40 0 0.6 0 46.5 -1 38.6 0.24
6 60 1 0.6 0 46.5 -1 35.1 0.17
7 20 -1 0.3 -1 46.5 -1 38.5 0.22
8 40 0 0.3 -1 46.5 -1 37 0.22
9 60 1 0.3 -1 46.5 -1 29.8 0.22
10 20 -1 1 1 92.4 0 88.4 0.18
11 40 0 1 1 92.4 0 86.6 0.16
12 60 1 1 1 92.4 0 83.8 0.14
13 20 -1 0.6 0 92.4 0 86.2 0.15
14 40 0 0.6 0 92.4 0 85.9 0.14
15 60 1 0.6 0 92.4 0 82 0.13
16 20 -1 0.3 -1 92.4 0 82.9 0.10
17 40 0 0.3 -1 92.4 0 81.2 0.10
18 60 1 0.3 -1 92.4 0 76.9 0.06
19 20 -1 1 1 144.6 1 146.8 0.07
20 40 0 1 1 144.6 1 141.4 0.01
21 60 1 1 1 144.6 1 139.7 0.01
22 20 -1 0.6 0 144.6 1 143.4 0.05
23 40 0 0.6 0 144.6 1 142.4 0.03
24 60 1 0.6 0 144.6 1 139.5 0.05
25 20 -1 0.3 -1 144.6 1 142.9 0.02
26 40 0 0.3 -1 144.6 1 139.5 0.01
27 60 1 0.3 -1 144.6 1 135.7 0.002
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 114

PDI p = a0 + a1 T MP + a2CFR + a3 FPS (4.7)

The coefficients (ai ) of the regression model are listed in Table 4.6 which indicates for

permeate particle size, the resulting model is a quadratic regression model, whereas PDI

is linearly dependent on the operating parameters. The p-values in Table 4.6 also provide

a check on the significance of each of the coefficients where a coefficient with a smaller

p-value is more significant. The analysis of variance (ANOVA) test results are also shown

in Table 4.6 indicating the goodness of the model as the value of standard deviation (SD) of

regression is much larger than the standard deviation of the residuals. The sum of square

(SS) and mean squares (MS) for the regression models in comparison with the residual

demonstrates that the regression correlation is highly significant. The F ratio for permeate

particle size (6639.74) and permeate PDI (133.73) are very much higher than Fcritical at

95% confidence value (i.e. 98.88 for permeate particle size and 0.25 for permeate PDI),

thereby, confirming the significance of the regression correlation (Table 4.6).

Table 4.6: Summary of Coefficients, Analysis of variance (ANOVA).


Permeate Particle Size Permeate PDI
Coefficients Values SE p-value Values SE p-value
Constant (a0 ) 87.863 0.532 3.42*10−34 0.127 0.004 5.28*10−20
TMP (a1 ) -3.053 0.286 6.10*10−10 -0.019 0.005 0.001
CFR (a2 ) 2.148 0.286 2.23*10−07 0.016 0.005 0.004
FPS(a3 ) 51.876 0.286 4.86*10−35 -0.099 0.005 9.67*10−16
CFR*CFR(a4 ) -1.625 0.507 0.004 – – –
FPS*FPS (a5 ) 2.720 0.497 1.99*10−05 – – –
ANOVA - Permeate Particle Size
Parameters DF SS MS F SD p-value
Total Corrected 26 48920.5 1881.56 43.377
Regression 5 48889.6 9777.92 6639.74 98.883 0
Residual 21 30.925 1.473 1.214
ANOVA - Permeate PDI
Parameters DF SS MS F SD p-value
Total Corrected 26 0.202 0.008 0.088
Regression 3 0.191 0.0637 133.173 0.252 0
Residual 23 0.0110 0.001 0.0219
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 115
The model predictions under the conditions of experiments as given in the experimental

design in Table 4.5 is plotted with observed values for both the responses (Figs. 4.19 &

4.20). From these parity plots, the correlation coefficient (R2 ) of the model is determined

to be in the range of 0.95 (permeate PDI) and 0.99 (permeate particle size) indicating that

the model is suitable for representing the factor-response relationships within the range

studied.

Figure 4.19: Observed vs Predicted: Permeate Particle Size.

Based on the surface response methodology, the correlation established by the estima-

tion of coefficients (Table 4.6) leads to the following insights: It is found that TMP (p

= 6.1*10−10 ) is more significant in comparison with CFR (p = 2.23*10−07 ) for permeate

particle size whereas TMP & CFR (p = 0.005) are equally significant for permeate PDI.

Table 4.6 also demonstrates, in all cases, the standard error (SE) is small and TMP, CFR

and FPS play a significant role along with interplay of these factors on the final product
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 116

Figure 4.20: Observed vs Predicted: Permeate PDI.

permeate particle size. Whereas for permeate PDI, TMP, CFR and FPS play a crucial role

and are linearly related.

4.5.3 Model prediction results


The empirical model (Equations 4.6 & 4.7) developed based on the surface response

method is employed to predict the behavior of changes on the final permeate particle size

and PDI. The following sections discuss the model prediction at varying TMP (20-60 kPa)

and CFR (0.3-1 L/min) for the three particles considered. It is quite important to note that

the empirical model developed agrees well with the experimental results thereby clearly

becoming an efficient tool for predicting the responses.

4.5.3.1 Effect of TMP on permeate particle size and polydispersity index

From the model results, it is found that with increase in TMP, the permeate particle

size decreases and there is a decline in permeate PDI as well (Figs. 4.21 & 4.22). At
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 117
higher TMP, more driving force and hence there is likely to be more of particle permeation

through the membrane. But at the same time, because of the combination of hydrodynamic

and colloidal forces among the particles, more of bigger particles fail to pass through the

membrane. Also, mixture of particles occupies the site of membrane pores which highly

hinders the bigger particles permeation. But interestingly, particles of slightly smaller sizes

pass through the membrane with ease.

Figure 4.21: Experimental (points joined by dashed lines to indicate the trend) Vs Model
Prediction: Effect of TMP at the middle CFR (0.6 L/min) on the permeate particle size.

4.5.3.2 Effect of CFR on permeate particle size and polydispersity index

In the case of increase in CFR, both the permeate particle size and PDI increases (Figs.

4.23 & 4.24). At higher rate of flow, pore narrowing of larger pores is less thereby making

bigger particles to pass through, but because of the lesser residence time particle concen-

tration is less.

Thus, from the above model predictions, it is clearly evident that the empirical model

developed could be used as a predictive tool to estimate the permeate particle property.
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 118

Figure 4.22: Experimental (points joined by dashed lines to indicate the trend) Vs Model
Prediction: Effect of TMP at the middle CFR (0.6 L/min) on the permeate PDI.

Figure 4.23: Experimental (points joined by dashed lines to indicate the trend) Vs Model
Prediction: Effect of CFR at the middle TMP (40 kPa) on the permeate particle size.
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 119

Figure 4.24: Experimental (points joined by dashed lines to indicate the trend) Vs Model
Prediction: Effect of CFR at the middle TMP (40 kPa) on the permeate PDI.

4.5.4 Particle size distribution (PSD) analysis


The experimental results of tangential flow UF of silica particles demonstrate the mech-

anism of separation and or fractionation of particles through UF polyether sulfone mem-

branes operated at varying system parameters. Experimental runs reported here are car-

ried out in the range of 20-60 kPa transmembrane pressure (TMP) and 0.3-1 L/min cross

flowrate (CFR). In Fig. 4.25, the particle size distribution of the original silica particles

employed for this study is depicted. The effective diameters of these particles are found to

lie between 35.6 nm to 142.9 nm range with standard errors of maximum 4.2nm.

4.5.4.1 Temporal effect

During tangential flow of particles along the membrane, multiple particles arrive at the

pore entrance, so initially the flow happens instantly and consequently the particle size

distribution decreases (Figs. 4.26 & 4.27). Thus, in general, particle permeation over time

is found to shift to left, however the dynamic change is not very rapid. The particle size
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 120

Figure 4.25: Representative distribution of feed stream sample of all three particles.

distribution evolution behavior is due to the fact with time the particle blocking the pores

maybe possible but it is insignificant which is quite evident from the plots. This particle

size distribution evolution can be related and agrees well with the available literatures of

the particle size distribution evolution on the membrane surface [Belfort and Nagata, 1985;

Chang et al., 2008].

4.5.4.2 Effect of TMP on particle size distribution in permeate streams

To study the influence of transmembrane pressure on the particle permeation, bigger

particles ultrafiltered in cross flow mode at 20-60 kPa pressure range are analysed for

particle size distribution. Based on the above study, particle permeation study can be

explained as two mechanisms depending on the transmembrane pressure change (Figs.

4.28 & 4.29):

• At lower and medium transmembrane pressures (i.e. 20, 40 kPa), it is found that

the particles are more towards separation mechanism i.e. smaller particle size range

are found to be present in the permeate stream and bigger particles in the retentate
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 121

Figure 4.26: An illustration of the particle size distribution evolution in the permeate
stream at TMP=20 kPa.

Figure 4.27: An illustration of the particle size distribution evolution in the permeate
stream at TMP=60 kPa.
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 122
stream.

• At higher transmembrane pressure (i.e. 60 kPa), the particles of similar sizes (frac-

tionation) are found in both the streams (permeate and retentate), along with smaller

particles permeate through the membrane. Thus at higher TMP, both separation and

fractionation are taking place.

Figure 4.28: Particle size distribution of the permeate stream at various TMPs (20-60 kPa),
time=9 mins, CFR=0.6 L/min.

These observations are in line with the results of the filtration efficiency results reported

in the previous section. It shows that at lower TMP, the log reduction value is high which

means that less of particle concentration in the permeate stream. It is also confirmed that

as summarised above, at lower and medium TMPs (LRV = 2.66, 2.54 respectively), the

trend is similar but with slight increase in the distribution which is clearly evidenced in

the minor drop in LRV value (0.12 LRV). Whereas, at higher TMP the distribution in-

tensity is greater and a mixture of small and similar sized particles are observed. This is

directly related to more of permeate concentration and hence there is a higher decline in
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 123

Figure 4.29: Particle size distribution of the retentate stream at various TMPs (20-60 kPa),
time=9 mins, CFR=0.6 L/min.

the filtration efficiency values (0.5 LRV). Thus, at higher TMP particles experience frac-

tionation and separation which has a direct influence of decreasing the filtration efficiency

values. Whereas, at lower TMP, particles experience more of rejection mechanism and

hence better filtration efficiency values.

4.5.4.3 Effect of CFR on particle size distribution in permeate streams

In order to understand the effect of cross flowrate on the final permeate particle size

distribution, cross flow experiments are analysed at three different rates ranging 0.3 -1

L/min. It is found that with change in cross flowrate, the particle size distributions change

is equally pronounced in the permeate stream. But there is a clear indication of particles

separating than fractionating when there is a change in the cross flow velocity.

With increase in cross flowrate, smaller particle permeation and bigger particle rejec-

tion is quite distinctively evident from the particle size distribution (Fig. 4.31). Thus, in

the case of change in cross flowrate, the filtration mechanism is more influenced by the
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 124

Figure 4.30: Particle size distribution of the permeate stream at various CFRs (0.3-1
L/min), time=9 mins, TMP=40 kPa.

Figure 4.31: Particle size distribution of the retentate stream at various CFRs (0.3-1 L/min),
time=9 mins, TMP=40 kPa.
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 125
separation mechanism than the fractionation in the studied cross flowrate range.

The above particle size distribution based study agrees well with the filtration effi-

ciency studies as indicated in the previous section. It is reported that with increase in cross

flowrate (from 0.3 to 1 L/min), the log reduction value increases from 2.4 to 2.58 LRV

which account only for a minor improvement which is evidenced in the small change in

PSD as well. This filtration efficiency increase is well supported by the results of Ra-

machandran and Fogler [Ramachandran and Fogler, 1999] where with increasing flow the

efficiency of particle retention has increased which has been attributed to the hydrodynamic

bridging effect.

4.5.4.4 Modeling particle size distribution

In this section, particle size distribution is modeled from the predicted values of perme-

ate particle size and PDI. The particle size distribution is best described by the following

equation:

 
 1 (x − x̄)2
I(x) = √ exp − (4.8)
2πσ2 2σ2

where 
I(x) is the intensity distribution at x, σ is the standard deviation which is ob-

tained from Equation 4.9 and x̄ is average particle size.

σ
PDI = (4.9)

The modeled particle size distribution is plotted against the experimental value for

comparison. It is evident that the predicted PSD, is overly consistent with the experimental

PSDs (Figs. 4.32, 4.33, 4.34). Thus, for a set of input conditions (TMP, CFR, FPS), using
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 126
Equations 4.6 & 4.7 permeate particle size and PDI could be computed. And further,

Equation 4.8 could be employed to plot the PSD of the permeate stream. Thus, in this way

the permeate PSD could be predicted within the range of TMP, CFR and FPS studied.

4.5.4.5 Size polydispersity index and its effects

The polydispersity index (PDI) is a measure of the nonuniformities that exist in the

particle size distribution. Three feed particle size ranges used for the experimental runs are

analysed for the polydispersity index effect. It is observed that smaller particles have higher

PDI (0.208) than the medium particle (0.151) and bigger particles (0.053). Further, the

tangential flow UF study of these particles clearly indicated that smaller particles exhibited

lower filtration efficiency (Fig. 4.35). Therefore the higher the polydispersity index of

original suspension the lower the filtration efficiencies. In order to demonstrate the effect

of PDI on the filtered product, LRV values at 40kPa TMP and 0.6 L/min crossflow velocity

are reported here (Fig. 4.35). It is also worth noting down that permeate PDI shows the

same effect as well on the filtration efficiency.

In the case of medium and smaller particles with a drop of 0.05 in PDI, the increase in

LRV is 0.23. Whereas in the bigger particles with less than 0.1 PDI (i.e. PDI=0.053), there

is an increase of 0.42 LRV in comparison with the smaller particles. This trend indicates

that if the original feed suspension has particles with broader distribution of sizes, the

potential barrier between particles is more prevalent and these inter-particle interactions

play a significant role in fractionation of particles, especially if particle size is less than

100 nm and the PDI is more than 0.1. Thus, the lower the feed and permeate PDIs, the

higher the LRV. However, it is found that feed PDI is statistically highly significant on LRV

(with p-value less than 10−4 ) whereas permeate PDI has slightly less influence on LRV (p-
Filtration Mechanism
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on

Figure 4.32: Representative particle size distribution of observed (points joined by dotted lines to show the trend) and predicted
127

plots for smaller permeate particles at (a)-(c) at TMP=20,40,60 kPa and 0.3 L/min (d)-(f)at TMP=20,40,60 kPa and 0.6 L/min
(g)-(i)at TMP=20,40,60 kPa and 1 L/min.
Filtration Mechanism
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on

Figure 4.33: Representative particle size distribution of observed (points joined by dotted lines to show the trend) and predicted
plots for medium permeate particles at (a)-(c) at TMP=20,40,60 kPa and 0.3 L/min (d)-(f)at TMP=20,40,60 kPa and 0.6 L/min
128

(g)-(i)at TMP=20,40,60 kPa and 1 L/min.


Filtration Mechanism
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on

Figure 4.34: Representative particle size distribution of observed (points joined by dotted lines to show the trend) and predicted
129

plots for bigger permeate particles at (a)-(c) at TMP=20,40,60 kPa and 0.3 L/min (d)-(f)at TMP=20,40,60 kPa and 0.6 L/min
(g)-(i)at TMP=20,40,60 kPa and 1 L/min.
4.5. Results and Discussion - Influence of Particle Size and Size Distribution on
Filtration Mechanism 130

Figure 4.35: Effect of feed and permeate PDIs on filtration efficiency at middle TMP (40
kPa) and CFR (0.6 L/min).

value = 0.033). Thus, it is deduced from the above study that increase in feed PDI and

permeate PDI, both, lead to decrease in filtration efficiency. The permeate PDI is highly

influenced and or controlled by the operating conditions and input parameters which is

governed by Equation 4.7. This reflects on the complex interactions existing between the

operating conditions.

4.5.5 Concluding remarks for particle size and distribution investiga-

tion section
Tangential flow UF of three size ranges of silica particles (as model virus particles) was

investigated to understand the rejection and or fractionation of particles considering the ef-

fects of feed particle size and transmembrane pressure and cross flowrate. For fractionation

of particles, suitable transmembrane pressure is to be chosen whereas if the particles are

to be separated, cross flow velocity is the key factor for better efficiency of membrane fil-

tration. From the experimental range employed in this study, it was found that the lower
Bibliography 131

the transmembrane pressure, the more separation occurs, while at higher transmembrane

pressure, a combination of separation and fractionation occurs. Whereas with change in

cross flow velocity, separation mechanism dominates and has good control over filtration

efficiency. Also the polydispersity study showed that the lower the dispersion of particles,

the higher the filtration efficiency attributed to the lower levels of inter particle forces.

Detailed statistical investigation indicated that permeate particle size follow a polynomial

dependency on the operating parameters whereas permeate PDI varies linearly with the

operating parameters. The particle level study covered in this article offers opportunities

for academic and industrial researchers to gain fundamental insights into how particle size

properties are linked to permeate quality.

Bibliography
Adham, S., Jacangelo, J., Laine, J., 1998. Low pressure membranes - assessing integrity.

Desalination 119, 73–77. 85

Amirilargani, M., Sadrzadeh, M., Mohammadi, T., 2009. Synthesis and characterization

of polyethersulfone membranes. Journal of Polymer Research, 1–15. 98

Arkhangelsky, E., Steubing, B., Ben-Dov, E., Kushmaro, A., Gitis, V., 2008. Influence of

ph and ionic strength on transmission of plasmid dna through ultrafiltration membranes.

Desalination 227 (1-3), 111–119. 108

Arnal, J., Garcia-Fayos, B., Verdu, G., Lora, J., 2009. Ultrafiltration as an alternative mem-

brane technology to obtain safe drinking water from surface water: 10 years of experi-

ence on the scope of the aquapot project. Desalination 248, 34–41. 85


Bibliography 132

Bakhshayeshi, M., Zydney, A. L., 2008. Effect of solution ph on protein transmission and

membrane capacity during virus filtration. Biotechnology and bioengineering 100 (1),

108–117. 108

Belfort, G., Nagata, N., 1985. Fluid mechanics and cross-flow filtration: so me thoughts.

Desalination 53, 57–79. 120

Bodzek, M., Konieczny, K., 1998. Comparison of various membrane types and module

configurations in the treatment of natural water by means of low-pressure membrane

methods. Separation and Purification Technology 14 (1-3), 69–78. 85

Chang, J., Tsai, L., Vigneswaran, S., 1996. Experimental investigation of the effect of

particle size distribution of suspended particles on microfiltration. Water Science Tech-

nology 34, 133–140. 86

Chang, J., Vigneswaran, S., Kandaamy, J., Tsai, L., 2008. Effect of pore size and particle

size distribution on granular bed filtration and microfiltration. Separation Science and

Technology 43 (7), 1771–1784. 86, 120

Chao, A., Tojo, S., 1987. Permeate quality of ultrafiltration process. Journal of Environ-

mental Engineering 113 (2), 383–394. 101

Fischer, E., Raasch, J., 1985. Cross-flow filtration. German chemical engineering 8 (4),

211–216. 108

Foley, G., Malone, D. M., MacLoughlin, F., 1995. Modelling the effects of particle poly-

dispersity in crossffow filtration. Journal of Membrane Science 99 (1), 77–88. 108


Bibliography 133

Graveland, A., 1998. Particle and micro-organism removal in conventional and advanced

treatment technology. Water Science and Technology 37 (10), 125–134. 85

Han, B., Akeprathumchai, S., Wickramasinghe, S., 2003. Flocculation of biological cells:

Experiment vs theory. AIChE Journal 49, 1687–1701. 87

Hwang, K., Liu, H., Lu, W., 1998. Local properties of cake in cross-flow microfiltration of

submicron particles. Journal of Membrane Science 138 (2), 181–192. 108

Kitis, M., Lozier, J. C., Kim, J. H., Mi, B., Marinas, B. J., 2003. Evaluation of biologic

and non-biologic methods for assessing virus removal by and integrity of high pressure

membrane systems. Vol. 3. 85, 98

Laine, J., Vial, D., Moulart, P., 2000. Status after 10 years of operation - overview of uf

technology today. Desalination 131, 17–25. 85

Leahy, T. J., Sullivan, M. J., 1978. Validation of bacterial retention capabilities of mem-

brane filters. Pharm.Technol. 2 (11), 65–75. 89

Lenes, D., Deboosere, N., Menard-Szczebara, F., Jossent, J., Alexandre, V., Machinal, C.,

Vialette, M., 2010. Assessment of the removal and inactivation of influenza viruses h5n1

and h1n1 by drinking water treatment. Water Research 44 (8), 2473 – 2486. 85

Lightfoot, E., Moscariello, J., 2004. Bioseparations. Biotechnology and BioEngineering

87 No.3, 259–273. 107

Lu, W.-M., Hwang, K.-J., 1995. Cake formation in 2-d cross-flow filtration. AICHE Jour-

nal 41 (6), 1443–1455. 108


Bibliography 134

Lu, W.-M., Ju, S.-C., 1989. Selective particle deposition in crossflow filtration. Separation

Science and Technology 24 (7-8), 517–540. 108

Mackley, M. R., Sherman, N. E., 1992. Cross-flow cake filtration mechanisms and kinetics.

Chemical Engineering Science 47 (12), 3067–3084. 108

Madaeni, S., Fane, A., Grohmann, G., 1995. Virus removal from water and wastewater

using membranes. Journal of Membrane Science 102(1-3), 65–75. 86

Oh, H. K., Takizawa, S., Ohgaki, S., Katayama, H., Oguma, K., Yu, M., 2007. Removal of

organics and viruses using hybrid ceramic mf system without draining pac. Desalination

202 (1-3), 191–198. 86

Otaki, M., Yaho, K., Ohgaki, S., 1998. Virus removal in a membrane separation process.

Water Science Technology 37, 107–116. 86

Pall, D. P., Kimbauer, E. A., 1978. Fifty Second Colloid and Surface Science Symposium

of Tennessee. 89

Pearce, G., 2007. Water and wastewater filtration: Membrane module format. Filtration

and Separation 44 (4), 31 – 33. 85

Pujar, N. S., Zydney, A. L., 1994. Electrostatic and electrokinetic interactions during pro-

tein transport through narrow pore membranes. Industrial and Engineering Chemistry

Research 33 (10), 2473–2482. 108

Ramachandran, V., Fogler, H. S., 1999. Plugging by hydrodynamic bridging during flow

of stable colloidal particles within cylindrical pores. Journal of Fluid Mechanics 385,

129–156. 125
Bibliography 135

Reti, A., 1977. An assessment of test criteria for evaluating the performance and integrity

of sterilizing filters. Bulletin of the Parenteral Drug Association 31 (4), 187–194. 89

Saksena, S., Zydney, A. L., 1994. Effect of solution ph and ionic strength on the separa-

tion of albumin from immunoglobulins (igg) by selective filtration. Biotechnology and

bioengineering 43 (10), 960–968. 108

Saxena, A., Tripathi, B. P., Kumar, M., Shahi, V. K., 2009. Membrane-based techniques

for the separation and purification of proteins: An overview. Advances in Colloid and

Interface Science 145 (1-2), 1 – 22. 107, 108

Song, L., Elimelech, M., 1995. Particle deposition onto a permeable surface in laminar

flow. Journal of Colloid and Interface Science 173 (1), 165–180. 86

Sun, W., Chen, T., Chen, C., Li, J., 2007. A study on membrane morphology by digital

image processing. Journal of Membrane Science 305 (1-2), 93–102. 98

Torras, C., Ferrando, F., Paltakari, J., Garcia-Valls, R., 2006. Performance, morphology

and tensile characterization of activated carbon composite membranes for the synthesis

of enzyme membrane reactors. Journal of Membrane Science 282 (1-2), 149–161. 101

Urase, T., Yamamoto, K., Ohgaki, S., 1998. Effect of pore structure of membranes and

module configuration on virus retention. Journal of Membrane Science 115(1) (10), 21–

29. 86, 98

van de Ven, W., Punt, I., Kemperman, A., Wessling, M., 2009. Unraveling ultrafiltration of

polysaccharides with flow field flow fractionation. Journal of Membrane Science 338 (1-

2), 67 – 74. 108


Bibliography 136

van Reis, R., Gadam, S., Frautschy, L. N., Orlando, S., Goodrich, E. M., Saksena, S.,

Kuriyel, R., Simpson, C. M., Pearl, S., Zydney, A. L., 1997. High performance tangen-

tial flow filtration. Biotechnology and bioengineering 56 (1), 71–82. 108

Zhou, J. X., Solamo, F., Hong, T., Shearer, M., Tressel, T., 2008. Viral clearance using dis-

posable systems in monoclonal antibody commercial downstream processing. Biotech-

nology and bioengineering 100 (3), 488–496. 108


Chapter 5

Modeling and Simulation

Contents
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

5.2 Earlier Modeling patterns . . . . . . . . . . . . . . . . . . . . . . . . . . 142

5.2.1 Macro-scale modeling: A quick review . . . . . . . . . . . . . . . 143

5.2.1.1 Flux decline prediction models . . . . . . . . . . . . . . 145

5.2.2 Particle property models . . . . . . . . . . . . . . . . . . . . . . . 147

5.3 Novel Model Formulation: Population Balance Approach . . . . . . . . 150

5.3.1 Statement of the problem . . . . . . . . . . . . . . . . . . . . . . . 151

5.3.2 Modeling of size distribution of suspended particles through a cross-

flow UF process . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

5.3.2.1 Input and output streams . . . . . . . . . . . . . . . . . 154

5.3.2.2 Constitutive equations . . . . . . . . . . . . . . . . . . . 155

5.3.2.3 Permeation probability . . . . . . . . . . . . . . . . . . 156

5.3.2.4 Expression of solid permeate flow . . . . . . . . . . . . . 157

5.3.2.5 Balance equations . . . . . . . . . . . . . . . . . . . . . 157

5.4 Model Solution and Simulation Results . . . . . . . . . . . . . . . . . . 160


138

5.4.1 Discretisation of the model . . . . . . . . . . . . . . . . . . . . . . 161

5.4.1.1 Discretisation issues . . . . . . . . . . . . . . . . . . . . 161

5.4.1.2 The internal coordinate and its significance . . . . . . . . 163

5.4.1.3 The method of moments . . . . . . . . . . . . . . . . . . 166

5.4.1.3.1 Different mean sizes . . . . . . . . . . . . . . . 169

5.4.1.4 Number and volume distributions . . . . . . . . . . . . . 171

5.4.1.5 Useful quantity predicted from the above model . . . . . 171

5.4.2 Simulation using gPROMS . . . . . . . . . . . . . . . . . . . . . . 172

5.4.2.1 Model entities of the entire process . . . . . . . . . . . . 173

5.4.2.2 Parameters and variables employed in the model develop-

ment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

5.4.3 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . 174

5.4.3.1 Representative particle size distribution in output streams 174

5.4.3.2 Effect of feed particle size on particle size distribution . . 177

5.4.3.3 Effect of transmembrane pressure drop on the particle size

distribution . . . . . . . . . . . . . . . . . . . . . . . . . 178

5.4.3.4 Effect of cross flow rate on particle size distribution . . . 178

5.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
139

Preface
The ultimate purpose of modeling filtration processes is the mathematical understand-

ing of the dynamic behavior of the process. A validated model can be implemented in

simulation, optimization and control of the process, overall helping to achieve desired

product characteristics. Typical filtration models found in literature focus on flux predic-

tions and fouling mechanisms. Our comprehensive filtration model, based on population

balances, is novel approach in the field of virus filtration. It takes into account individual

particles and their particle size property, equations of conservation, and kinetic submodel.

The model developed is capable of successfully predicting the permeate particle size and

size distribution as well as final filtration efficiency. This represents a big leap in under-

standing filtration mechanisms and process behaviour and importantly can be applied in

scale-up studies. This novel model demonstrates a promising step to optimize and control

the process to achieve the required regulations. This chapter details the model formulation,

model solution and the model parametric analysis.

Research Questions:

1. Is the population balance framework, which has a wide spectrum of applications in

other fields of chemical engineering problems like coagulation, precipitation, crys-

tallisation, a suitable approach for modeling the filtration process?

2. Is it suitable to describe particle permeation by a stochastic kinetic model?

The content of this chapter has been presented in Engineering With Membranes, Portu-

gal [v] and International Congress on Membranes and Membrane Processes, Hawaii [vi].

Manuscript under review for submission in Chemical Engineering Science journal [iv].
5.1. Introduction 140

5.1 Introduction
Process modeling is an important avenue towards understanding and optimizing design

and operations of an unit operation [Ahmad et al., 2006; Bodzek and Konieczny, 1994;

Mohammadi et al., 2005]. While there is a constant evolution in mathematical model

formulations to predict the performance of membrane filtration processes in terms of per-

meate flux and fouling mechanisms, the same cannot be said about models for prediction

of permeate particle size and filtration efficiency. While others have attempted modelling

systems such as deep bed filtration [Santos and Bedrikovetsky, 2004], this work, we be-

lieve, is a maiden attempt to illustrate the application of the population balance framework

to predict the final product efficiency in the context of tangential flow ultrafiltration (UF).

The main objective of this research is to develop a novel model for the TFF based on

the population balance theory. Such predictive modelling will serve as a promising tool

for optimizing and controlling the TFF process. The innovation of this work lies in the

above fact, whereas previous ultrafiltration models have been dominated by flux modeling

techniques, i.e. at the macroscale.

The framework of population balances to model particulate processes was proposed

in the seminal work by Hulburt, Katz [Hulburt and Katz, 1964] and has since penetrating

into various engineering problems and applications. Randolph [Randolph, 1964; Randolph

and White, 1977] give a comprehensive insight on population balance theory. The use of

population balance models (PBM) has seen a significant growth in the last 15 years or so.

The population balance equation (PBE) offers its utility for describing the complex nature

of particle systems distributed along the particle size domain. Its applicability to a wider
5.1. Introduction 141

range of systems is well explained by Ramakrishna [Ramakrishna, 2000]. These applica-

tions include crystallisation, [Gerstlauer et al., 2006; Rawlings et al., 1992] polymerisation

[Alvarez et al., 1994], fluidized bed [Tan et al., 2005] to name a few. The uniqueness of

the population balance method lies in the fact that the processes are described in terms

of internal coordinates and external coordinates; the internal coordinate referring to size,

chemical composition whereas the external coordinate describes the position in space of

the entity concerned. For the model to mimic the exact way as the process, the right choice

of phase coordinates is a primary factor. The internal coordinate, hence, is defined in terms

of any size related variable (diameter or volume) and, say, if volume is considered as the

internal coordinate the accumulation also is given in terms of the volume range.

The solution of PBEs can be obtained by any one of the methods, viz. analytical meth-

ods (including analytical solutions and similarity solutions), method of moments, method

of weighted residuals and methods such as finite element method or direct discretisation.

There is an extensive literature on the solution and dynamical analysis of particulate pro-

cesses [Gelbard and Seinfeld, 1978; Ramakrishna, 2000; Randolph, 1964].

The detailed investigation carried out in studying the influential parameters on the fil-

tration mechanism (which is already presented in Chapter 4), clearly indicates that apart

from various operating parameters like, cross flow velocity, and transmembrane pressure,

particle size plays an important role in challenging the filtration efficiency. It is more com-

mon to find typical UF operations working in the cross-flow mode (alternatively termed

tangential flow) for bio-technological applications in comparison with the dead-end fil-

tration mode (Fig. 5.1) [Meltzer and Jornitz, 2007]. Cross-flow filtration has received

attention for its fouling reduction capabilities in removing molecules or particles from the
5.2. Earlier Modeling patterns 142

membrane surface by the tangential flow of liquid. Currently, various theories and mod-

els exist to describe the transport mechanism (mass/fluid/momentum) for tangential flow

filtration mechanism [Lee and Clark, 1998; Mohammadi et al., 2005], but none have ad-

dressed the particle size distribution (PSD) and further still none have attempted describing

the transport mechanism as a stochastic one. This work aims at extending the population

balance modeling concept to UF processes thereby describing the evolution of the particle

size distribution in the filtration product streams. The status-quo in modeling, i.e. flux

predictions, would be completed by size distribution modeling.

Figure 5.1: Illustrations of (a) dead-end filtration and (b) Tangential flow filtration.

5.2 Earlier Modeling patterns


Numerous process models have been presented in the literatures that describe flux de-

cline, fluid dynamics, and fouling characteristics. The key performance (output) variables

in UF include the rejection coefficient, log reduction value, the permeate flux and the

transmission characteristics. There are mainly three related (input/operational) variables


5.2. Earlier Modeling patterns 143

that play a significant role in the effectiveness of filtration; change in the pressure drop

across the layer, fluid flow rate along the membrane, and the feed suspension properties.

With respect to the modeling objective, the UF models can be classified into two cate-

gories [Driscoll, 2001]. The first category includes models that are related to the filtration

process design and equipment including the membrane. In this category, models are pre-

dominantly targeted at flux decline prediction under the effect of various membrane and

process parameters. Process models that are members of the second category are aimed

at particle suspension properties, more specifically, at particle polydispersity parameters.

For the second category of models, the prediction of flux decline is a secondary modeling

objective and so the first (more conventional) category of models is in effect a subset of

the second category. But the first category of models is predominantly found in the litera-

ture whereas the second category is still in its infancy demanding more attention. Hence,

this work tries to address the modeling technique at the particle scale which will help in

optimisation and controlling the process. Before presenting our model, let us have a quick

review of already available models in order to capture the essence of existing modeling

techniques to highlight the uniqueness of the current work.

5.2.1 Macro-scale modeling: A quick review


Many realistic models for UF are available in the literature which are macroscopic in

nature, predicting flux and fouling phenomenon, based on cake layer formation, concen-

tration polarization, shear induced diffusion, and adsorption [Belfort et al., 1994; Bowen

and Sharif, 2002; Ghosh, 2008; Lee and Clark, 1998; Richard Bowen and Williams, 2001].

These models, viz. analytical, semi-analytical, and numerical comprehensively predict the

flux decline behavior and thereby membrane performance. In membrane separation pro-
5.2. Earlier Modeling patterns 144

cesses, another promising modeling technique is based on fluid dynamics which is well

addressed by Belfort [Belfort, 1989; Belfort and Nagata, 1985]. The fluid flow through

membranes is a combination of free flow and flow through porous media where the free

flow is best described by Navier-Stokes equation and the flow through the porous media

is represented by Darcy’s law for non-inertial, incompressible flows with small porosity.

Development of these models have found broader applicability and computational fluid

dynamics (CFD) is the most sought after technique. These models are chiefly categorized

as either solubilisation diffusion models or concentration polarization models [Ghidossi

et al., 2006; Lee et al., 1996] and normally coupled with mass transfer equations. Lee

et al. [Lee et al., 1996] modeled cross-flow UF through CFD technique where the ef-

fects of the particle size, concentration and membrane dimension on the flux decline were

explained. Huang and Morrissey’s [Huang and Morrissey, 1999] work used a numeri-

cal model to reproduce the process of concentration polarization and to model the solute

concentrations on the membrane surface. Nassehi [Nassehi, 1998] used the finite element

method and presented a more robust simulation technique compared to other previous

works. Pak et al. [Pak et al., 2008] used the finite volume method and studied the ef-

fect of various physical parameters on the growth of concentration polarization layer in

a two-dimensional convective-diffusion equations coupled with resistance-in-series model

for permeation transport. Damak et al. [Damak et al., 2004] used finite difference method

to simulate a laminar, incompressible and isothermal flow in a cylindrical tube with a per-

meable wall. Though improvement of these models is in a continuous trend with regard

to understanding the flux decline and fouling, debate still exists in understanding the ba-

sics of polarization phenomenon. With better understanding, the processes become more
5.2. Earlier Modeling patterns 145

complex and the complexity of these models is considerably reduced by the use of CFD

technique which in turn helps minimizing the number of experiments. These models are

usually solved using commercially available packages (eg. Fluent, CFX) and has the con-

venience to analyze the effects of operating parameters on the membrane performance and

many authors confirm the benefits of using CFD to optimize membrane systems and pro-

cesses [Belfort and Nagata, 1985; Ghidossi et al., 2006; Kleinstreuer and Belfort, 1984].

The handicap in these models is that the permeate particle size and its distribution deter-

mination are not accounted for. In reality, particles interact with each other and with the

membrane pores impacting on the filtration efficiency and mechanism.

5.2.1.1 Flux decline prediction models

It is well established that flux declines with time during an UF process, such decline

being attributed to membrane fouling (see [Song, 1998] and citations within). There has

been a number of works that reviewed flux prediction models over the years [Chudacek

and Fane, 1984; Kim, 2007; Lee et al., 1996; Sarkar et al., 2009]. Three main theories

have been used in these modeling techniques leading to three types of models; (1)

Gel-polarization model [Rautenbach and Schock, 1988; Wakeman and Akay, 1994;

Wijmans et al., 1984], (2) Osmotic pressure model [Ahmad et al., 2006; Kim, 2007] and

(3) the resistance model [Sarkar et al., 2009; Zaamouche et al., 2009]. The description of

the flux in the first type of models is in terms of the bulk, gel, and permeate concentrations

(Cb , Cg and Cp ):

J = f (Re, S c, Cb , Cp , Cg ) (5.1)
5.2. Earlier Modeling patterns 146

Unknown parameters within the flux equation are experimentally determined. In the

second type of flux models, the flux is a function of transmembrane (ΔPTM ) and osmotic

pressures (ΔΠ) as well as concentration at the membrane (Cm ):

J = f (ΔPTM , ΔΠ, Cm ) (5.2)

This equation is popularly written in the modified Hagen-Poiseuille form (original refer-

ence deriving the H-P equation, perhaps Bird et al. [Bird et al., 2006]) and gives the flux

as superficial permeate velocity. Viscosity is also considered which can be corrected for

non-water permeates.

Lastly, the third type of flux model describes the flux as being dependent on resistances

against the pressure driving force. Typically three resistances in series are used, being the

resistances across the gel (Rg ), the fouling (Rf ) and the membrane (Rm ) layers:

J = f (ΔPTM , Ri ) (5.3)

While the above three types of models for describing flux have been prevalent in the liter-

ature, other approaches have also been used to an extent including the film model, those

models that use a combination of the above three flux modeling methods [Bolton et al.,

2006; Mohammadi et al., 2005; Zhang and Song, 2000] and those that extend the descrip-

tion to other flux dependencies, such as on pH and electrostatic forces.


5.2. Earlier Modeling patterns 147

5.2.2 Particle property models


In the previous section, we looked at past UF models, those aimed at flux prediction.

Flux modeling has received much attention, presumably because this is a macroscale vari-

able that is easily measurable and understood. We now look at a second category of UF

models, those with descriptions extended to particle properties particularly particle size

and size distribution. Particle size distribution (PSD) is an important variable in UF be-

cause this is directly responsible for a greater impact on filterability [Roorda et al., 2004].

The emphasis on particle size leads to descriptions at the particle scale, as such it is prod-

uct property focused which is important to ensure product quality, purity and successful

operation of an UF system. Moreover PSD prediction can provide information about the

numbers of particles in the permeate and retentate which is of significance in the UF of

virus suspensions, where quantification of virus numbers is directly required [Alspach and

Allgeier, 2003; Brandenburg and Zhuang, 2007]. Future UF operations may well become

routinely characterized by alternative variables like PSD. Unlike flux, PSD was previously

not easy to measure particularly at the ultra size range wherein UF processes operate, but

with advances in analytical instrumentation, it is nowadays possible to measure PSD us-

ing techniques like dynamic light scattering (DLS). In spite of the importance of PSD, UF

models have not addressed nor incorporated this variable sufficiently. The following is a

review of previous related works.

Chang and Hwang [Chang and Hwang, 1995] developed a mathematical model which

predicts the unsteady state permeate flux of crossflow MF of polydispersed solution. The

PSD of the cake formation was obtained and they have observed that the permeate flux

increases with an increase in the mean particle size of the cake. Whereas, mean particle size
5.2. Earlier Modeling patterns 148

of the cake increases with increase in the mean particle size of the original solution. Vyas

et al. [Vyas et al., 2000] estimated permeate flux, internal and surface fouling, porosity and

the PSD of the feed particles of lactalbumin. Special emphasis was given to the influence

of the feed properties on the performance of the microfiltration process. Wakeman’s work

[Wakeman, 1994] discussed the effect of parameters such as particle size, feed suspension

concentration, pH, transmembrane pressure and crossflow velocity on the development of

the particle layer and the corresponding permeate flux. His subsequent work states that it is

qualitatively and quantitatively difficult to predict the effect of particle shape. Barauh et al.

[Baruah et al., 2005] framed a global model that accounts for pH, ionic strength, sieving

through membrane cake, effect of hydrodynamics to predict the performance of crossflow

UF processes. They also highlighted the difficulty in using molecular dynamics to solve

the MF/UF process because of the complexity of the mechanism of the process. They

attributed the complexity involved in modeling MF to the complex nature of the fluids

and the difficulty in specifying the behavior of their suspended and dissolved components.

Benjamin McCoy’s work [McCoy, 1995] dealt with the membrane and mixture parameters

and showed how these parameters influence the sieving coefficient. For a steady-state UF

process with a well-mixed upstream concentration distribution, the theory predicts how

much the downstream concentration distribution shifts toward smaller solutes as larger

solutes are hindered or rejected by the membrane.

The modeling approaches, considering particle size as one of the factors, can be sum-

marized as (i) force or torque balance models describing the particle size and its deposition

rate on the membrane surface based on variety of forces acting on the particle [Chang and

Hwang, 1995; Hwang et al., 1998; Lu and Hwang, 1995] (ii) hydrodynamic models de-
5.2. Earlier Modeling patterns 149

tailing the mechanisms of particle deposition and cake formation in the microfiltration of

micron particles [Belfort and Nagata, 1985] (Table 5.1). The major limitation of such

models formulated, so far, are not capable of predicting the permeate quality say in terms

of number density or concentration values [Stoller, 2009]. It is very important and cru-

cial to note that the models incorporating the particle size and its distribution is limited to

handling the input/feed suspension [Chang et al., 1996, 2008; Foley et al., 1995b].

Table 5.1: Summary of various models considering feed particle size as a factor to study
the permeate flux.

Trajectory Analysis (Force or Torque)


Brownian diffusion [Porter, Cross flow rate promote back transport alongwith
1990] shear induced radial migration of particles.
Shear induced diffusion Local particle concentration by balancing the
[Romero and Davis, 1990] convective motion of particles with the shear-
induced diffusion.
Inertial lift [Belfort et al., 1994; Lateral migration theory to explain flux paradox.
Green and Belfort, 1980]
Dean vortices [Winzeler and Higher performance with introduction of Dean
Belfort, 1993] vortices.
Frictional drag, Brownian force Cake porosity, specific cake resistance, effect of
[Tarabara et al., 2002] cross-flow rate and transmembrane pressure on fil-
trate rate.
Lateral migration velocity Permeation drag, inertial lift and sedimentation
[Chellam and Wiesner, 1992] dominate lateral migration in far field region
whereas van der Waals force is important near the
membrane wall.

In the current presentation, we discuss and present an extended particle population

UF model for virus filtration. The interesting aspect of virus filtration is: in the case of

virus purification, the degree of recovery or total removal of virus is less critical. Viruses

purified by UF include Aedes aegypti, Minute Virus of Mice, Autographa californica M

nucleopolyhedrovirus which are industrially important vectors [Guo et al., 2009; Hensgen

et al., 2010; Michalsky et al., 2009]. Whereas, in the case of virus removal or protein
5.3. Novel Model Formulation: Population Balance Approach 150

separation studies, the main aim is to achieve complete removal of viruses with recovery of

protein product. Typically, these feed suspensions are very dilute and hence cake formation

and concentration polarization are assumed to be negligible. Thus this modeling is based

on the condition where such negligible concentration polarization and cake formation are

expected.

The particle population focus founding this proposed model is statistical in its nature

and is consequently more informative about the particle phase and specifically the evolu-

tion of the PSD in the product streams. Such description which is extensible to particle

properties other than PSD has been lacking in the literature where previous works have

been more focused on flux determinations [Mohammadi et al., 2005; Song, 1998]. The

details of our model are presented and discussed in the subsequent sections.

5.3 Novel Model Formulation: Population Balance Ap-

proach
The process of UF, can be classified as a particulate one with the solid particulate

phase suspended in the carrier liquid phase. A population balance allows one to describe

the evolution of individuals (particles in the UF case) in the entire population, which is

generally defined with the forces driving the population dynamics. The evolution of each

individual is described by internal coordinates such as size, age or velocity. Size is a

typical property used to describe the variable changes to individual particles. For purposes

of statistical simplification, we describe particles in classes rather than individual particles,

where each particle class has a range of sizes carrying the particles that fall in that size

range. Consequently the output of such models is distributions across the time and particle
5.3. Novel Model Formulation: Population Balance Approach 151

size domains providing continuous and clear picture of the particle state. Thus, a detailed

mechanistic model is proposed which utilizes the population balance background along

with the stochastic behavior of particles, in the following sections. This developed model

has a promising future, as a tool for the optimization, control and design of UF processes

for virus removal and purification operations. This model is quite unique as it is capable

of predicting precisely the virus population in the output streams and the final filtration

efficiency as well.

5.3.1 Statement of the problem


In cross-flow UF process, the suspension (i.e. feed stream) while passing through the

membrane will have these random phenomena: (i) particles permeate through the mem-

brane if the particle size (pa) is smaller than the pore size (po) (ii) not all particles smaller

than the pore size pass through, (iii) particles bigger than the pores (and a fraction of

smaller particles as well) are dragged by the shear rate in the tangential mode as reten-

tate. The above mechanism is based on the fact that the UF is size exclusion i.e. sieving

mechanism.

To model the UF process, the UF system is described physically as the one which has

the feed suspension as input stream and the permeate and retentate as the output streams,

each has liquid and particles and the separation boundary is the membrane as shown in

Fig. 5.2. The focus of the PBE model in predicting size properties in the filtration process

is shown in Fig. 5.3.

The modeling problem is addressed via the following steps:

• The formulated model and the accompanying allied equations are presented along-

with the initial conditions.


5.3. Novel Model Formulation: Population Balance Approach 152

Figure 5.2: A schematic diagram of the tangential flow filtration: Particles sieving through
the membrane (permeate) and those being excluded (retentate).

Figure 5.3: A schematic diagram of the membrane filtration and its relationship with the
proposed PBE model.
5.3. Novel Model Formulation: Population Balance Approach 153

• The operating parameters and the input conditions of the tangential flow filtration

process are defined.

• The model is solved and simulated using gPROMS (Process Systems Enterprise,

UK).

Figure 5.4 is a summary of the model and its interplay with other equations and phe-

nomena.

Figure 5.4: An overview of the model encapsulating the set of balances involved in the
model formulation.

5.3.2 Modeling of size distribution of suspended particles through a

cross-flow UF process
The population balance equation (PBE) is a continuity statement written in terms of

the number density function. The macroscopic population balance of tangential flow UF is

modeled based on the approach proposed by Randolph and Larson [Randolph and Larson,

1988]:

∂n 
= B−G+ Fk nk /V (5.4)
∂t k
5.3. Novel Model Formulation: Population Balance Approach 154

where n and nk are the number distributions (no./m3 ), F is the flow rate of either the

input or output streams (L/min), G is the growth term, B is the nucleation term, V is the

volume of the original suspension (L).

In the tangential flow UF process, it is assumed that there is no aggregation, growth

and/or nucleation taking place hence the first two terms of the right hand side of the equa-

tion are ignored. Whereas the driving force being transmembrane pressure is defined in

terms of filtration kinetics. Incorporating these, the above equation reduces to

∂nr 
=F+ Fk nk /V (5.5)
∂t k

where nr is the number density in the retentate stream, F is the filtration kinetics, Fk ,

nk are the flow rate and number density respectively, which is defined in the following

section. It is quite evident that the number density in retentate stream is zero initially i.e.

at time t=0.

5.3.2.1 Input and output streams

In the cross-flow UF process, the cross flow rate (otherwise feed flow rate) is chosen

as the reference flow and hence the total flow streams in the system could be well thought

of as input streams with feed flow rate (F f ) and the outlet streams being the permeate and

retentate streams (F p , Fr ) and each streams are proportional to the distribution function

[Randolph and Larson, 1988].

Inlet (i) and outlet (o) streams are thus defined by the respective equations as follows:

  
Fk nk pa = Fi ni pa − Fo no pa (5.6)
pa pa pa
5.3. Novel Model Formulation: Population Balance Approach 155

where Fi & Fo are the flow rates of the inlet and outlet streams respectively. And ni &

no are the number densities of the inlet and outlet streams respectively. The inlet and outlet

streams are given as:

 
Fi ni pa = F f n f pa (5.7)
pa pa

  
Fo no pa = F p n p pa + Fr nr pa (5.8)
pa pa pa

where n p & nr are the number densities of the permeate and retentate streams respec-

tively.

The above equation is a modeling simplification that has a profound effect in reducing

the complexity of the set of equations that satisfactorily describe the system. Physically,

this modeling simplification indicates that particle classification occurs as size split at the

output streams and thus this modeling simplification is well justified [Randolph, 1964;

Randolph and White, 1977].

The particle number density of the feed suspension is defined as:

exp(−((L pa − L¯pa )2 /2σ2 ))


n f pa = √ (5.9)
2πσV

where L pa & L¯pa are the particle size and average particle size respectively, σ is the

standard deviation.

5.3.2.2 Constitutive equations

It is a well established fact that the driving force of the UF process is the transmem-

brane pressure drop and therefore, the kinetics of the model is defined as power law which
5.3. Novel Model Formulation: Population Balance Approach 156

includes the stochastic behavior of particles as well:


k
F= k1 T MPk2 P pa (5.10)
pa=i

where TMP is the transmembrane pressure drop, F is the filtration kinetics, k1 and k2

are kinetic parameters, and P pa is the stochastic behavior of particles which is defined in

the next section. Thus, this approach of defining kinetic model is quite novel as it is capable

of capturing the random behavior of the particles influenced by the driving force. Initially,

the model is simulated with guess values to verify it is solving well [gPROMS, 2009a] and

finally, for validation purposes the kinetic parameters are obtained via the experimental

results.

5.3.2.3 Permeation probability

The filtration mechanism, using stochastic behavior, can be conceived of random par-

ticle behavior in the system: if the particles of a range of size class (in the suspension)

approaching the membrane (particle size lesser than the pore size) has a diameter lesser

than the pore, then it is very likely that the particle will permeate through the pore. In

this way each individual particle smaller than the membrane pore has a probability (P pa )

of passing through the pore. This stochastic behavior has a significant role in the above

defined filtration kinetics. The probability distribution is defined as follows:

La 2 La 2
P pa = [ ] 2
+ [ ]2 (5.11)
La + L pa
2 2
La + L pa L pa−1
2

This relationship holds valid if the particle size is smaller than the pore, otherwise, the

probability approaches zero. There are occasions where the data for particle size distri-
5.3. Novel Model Formulation: Population Balance Approach 157

bution is limited, in those cases, the average particle size of the feed stream is considered

to explain the mechanism. The same assumption maybe extended for the pore size of the

membrane, and where the pore size distribution is unavailable, the average pore size maybe

considered and this approach is employed in this work.

Now, the above discussed stochastic behavior of the particles permeating is demon-

strated here for model depiction. From Figure (Fig. 5.5), it is evident that with increasing

particle size the probability with which the particles permeate decreases and when it ex-

ceeds the average pore size, it approaches zero.

5.3.2.4 Expression of solid permeate flow

An extension of the stochastic behavior coupled with the kinetic rate definition as de-

fined above, the solid permeate flow is typically solved with number density of the perme-

ate particles. And if L pa is smaller than L po , then the particle permeation is correlated with

the number density distribution and thus, giving a good indication of the model behavior of

the permeate particles. This has an explicit bounding condition of no permeation when the

particles approaching the pore are larger which is seen in the simulation result of particle

permeate flow (Fig. 5.6).

5.3.2.5 Balance equations

The overall conservation in terms of overall and component mass balance equations

are written as follows with the assumption that the density of all the streams is constant,

true for dilute systems.

F f n f = F p n p + Fr nr (5.12)
5.3. Novel Model Formulation: Population Balance Approach 158

Figure 5.5: Plot showing the behavior of the probability of particles permeating through
the membrane: Smaller the particles higher the probability and vice-versa. Also the plot
shows that the particles approaching membrane pore reflects zero probability of particle
permeation.
5.3. Novel Model Formulation: Population Balance Approach 159

Figure 5.6: Solid permeate flow indicating the bounding condition of no particle flow of
bigger particles.

F f = F p + Fr (5.13)

The number conservation is thought of as the total number of particles of a given size

entering-in (n f pa ) either permeates through the membrane (n p pa ) and the rest are excluded

as retentate (nr pa ) which is given as:

n p pa = n f pa − nr pa (5.14)

Finally, the permeate concentration of the particles in the suspension is obtained as

particle permeation through the total permeate rate.

Particle and pore dimension

The size domain is defined after the generalization method given in [Abbas and Ro-

magnoli, 2007], where the general equation describing the size of a particle (Equation

5.15) is given by:


5.4. Model Solution and Simulation Results 160

L pa = ab pa (5.15)

where a is the minimum detectable size and b is the geometric constant obtained based

on the maximum particle size present in the system. The pore size of the membrane L po is

assumed to be constant in the model formulated and the solute permeate flow is obtained

when the particle to pore size ratio is less than one. Thus to consolidate, the original pop-

ulation balance equation (Equation 5.5) coupled with the allied equations, i.e. Equations

5.6, 5.12, 5.13, 5.10 & 5.14) are solved to simulate the model and the approach to solve

the model is presented in the following section.

5.4 Model Solution and Simulation Results


The novel model formulated which is based on the population balance framework cou-

pled with the stochastic behavior of particles is, in its raw form, a partial differential equa-

tion in time and size. The solution methods for such models broadly include analytical

solution and numerical technique. But analytical solutions exist only for few cases, and

thus numerical techniques are resorted to. In order to ease the solution of population bal-

ance equations, various numerical techniques are developed. These include include Monte

Carlo method [Ramakrishna, 2000; Rosner et al., 2003], method of characteristics [Ku-

mar and Ramakrishna, 1996], discretised methods like finite volume or finite element or

finite difference [Marchisio et al., 2003], standard method of moments [Hulburt and Katz,

1964], quadrature method of moments [Gimbun et al., 2009] and direct quadrature method

of moments [Su et al., 2008].


5.4. Model Solution and Simulation Results 161

5.4.1 Discretisation of the model


The discretisation of a typical population balance equation is well covered by several

researchers as mentioned above. In this work, a discretisation numerical method is chosen

which needs to a relatively low computational charge as the complex balance equation is

solved as a set of ordinary differential equations. Furthermore, this method also allows to

calculate the particle size distribution with a great precision when using a large number of

discretised classes. Thus, the original model equation (Equation 5.5) is discretised with

respect to the size domain into a normal differential equation. This method overcomes the

disadvantage of other methods as it allows the kinetic rate to be discretised along the size

domain as well.

5.4.1.1 Discretisation issues

Discretisation of the population balance equation transforms it into a family of linear

differential equations by discretising along the size domain. Thus the original PBE results

into an equation with reference to the independent size variable (L pa ):

∂nr pa 
= F pa + Fk nk pa /V, pa = 1, 2, 3, ......n (5.16)
∂t k

The above equation becomes a system of ordinary differential equation as the time

derivative is only an ordinary one. Hence,

dnr pa 
= F pa + Fk nk pa /V, pa = 1, 2, 3, ......n (5.17)
dt k

which is the discretised form of Equation 5.16. Here, since the density function is

along the size axis and is defined as classes, the width of the class is just the difference
5.4. Model Solution and Simulation Results 162

between the immediate adjacent sizes of that interval and is defined as follows:

φ pa = L pa − L pa−1 (5.18)

The scale of the size axis showing the relationship between the class, density, is shown

in Fig. 5.7. Thus the discretisation of such size axis is chosen or defined according to

the input conditions. In this work, the change in size axis is kept constant (i.e. ΔL pa =

constant).

Figure 5.7: The discretisation of the density function and the relationship between size
class, and density.

Thus, Equation 5.17 takes the form as follows:

dnr1 
= F1 + Fk nk1 /V (5.19)
dt k

dnri 
= Fi + Fk nki /V (5.20)
dt k

for i = 2,3...n-1

dnrn 
= Fn + Fk nkn /V (5.21)
dt k
5.4. Model Solution and Simulation Results 163

The permeate flow is given as solid permeate flow which is obtained from the kinetics

and the liquid permeate flow is obtained from the transmembrane pressure drop equation.

Thus, the total permeate flow rate is the sum of solid and liquid permeate flowrates.

5.4.1.2 The internal coordinate and its significance

The internal coordinate which is the size axis plays a significant role in the simulation

aspect. Proper choice of the particle size range dictates the sensitivity and the step size

gives a broader and or narrower type of the simulation results, at the same time it gives the

finer discretization of the problem in itself.

The general equation describing the particle size could well be expressed as either

arithmetic (Equation 5.22) or geometric progression (Equation 5.23) as:

L pa = L1 + (n − 1)b∀pa = 1, 2, ....n (5.22)

where L1 is the lowest size of the particle which the measurement could be obtained,

b is the arithmetic constant to be determined and L pa is the size of the respective path size

range. Ln is the largest particle size present in the system.

L pa = ab pa ∀pa = 1, 2, ....n (5.23)

where a is the lowest size of the particle which the measurement could be obtained, b

is the geometric constant to be determined and L pa is the size of the respective path size

range. Ln is the largest particle size present in the system. Therefore, the smallest size is

given by the following equation as per the above equation:


5.4. Model Solution and Simulation Results 164

L0 = a (5.24)

In this work, it is realized that the geometric progression suited appropriate to the

measurement, hence Equation 5.23 is employed throughout the simulation results. Also,

geometric series is found to be the most suitable for determining the sequence of size

ranges according to number of researchers [Abbas and Romagnoli, 2007].

Thus the size characterization of the chosen range is given as:

φ pa = ab pa − ab pa−1 (5.25)

In other words, the midpoint of the discrete particle size range (S pa ) is given by the

following equation (Equation 5.26) and which gives the average of the particle size:

ab pa − ab pa−1
S pa = (5.26)
2

As shown in the Equation 5.23, the discrete particle size range could well be adjusted

based on two parameters viz. geometric constant, b, and the number of discrete particle

size range, n. Increasing the geometric constant will influence two factors viz. computa-

tional time and the accuracy of the results. But it is found that changing the discretisation

size of the particle affects the simulation time and also has a considerable impact on the

accuracy of the results. Hence, enough care is taken while choosing the constant bearing

the fact of maximum particle size available in the suspension.

Varying the number of size ranges has a direct effect on the upper limit of the particle

size range. This is achieved by choosing the lower and upper limit of the particle size
5.4. Model Solution and Simulation Results 165

expected or known. Accordingly, the value of n is obtained and higher the number of

particle size classes will slightly increase the simulation time whereas it also has a less

effect on the accuracy.

To illustrate the above, let us consider the following cases:

Case 1 : Calculating the number of particle size class:

Considering the minimum detectable particle size is 10 nm, maximum particle size

range is 200 nm with geometric constant 1.032, then the number of particle size class is

obtained from Equation 5.23 as:


200
bn = (5.27)
10

1.032n = 20 (5.28)

leading to n = 95.

Case 2 : Calculating the geometric constant:

In order to vary the particle sizes in itself, then accordingly the geometric constant is

adjusted. For example, considering the number of particle size classes as say 100, with the

min. particle size being 10 nm, but for 2 max. particle sizes i.e. 150 nm and 300 nm, we

get,

  100
1
150
b= = 1.0275 (5.29)
10

  100
1
300
b= = 1.0346 (5.30)
10
5.4. Model Solution and Simulation Results 166

Based on the above examples, the two parameters are fixed depending upon the input

suspension stream conditions. In total, this method generalizes the particle size discretisa-

tion to any system desired.

Now, the role of particle size class and the geometric constant is demonstrated in Fig.

5.8 which clearly indicates the significant effect on the size distribution. It is recalled that

increase in both size class range and the geometric constant increases the simulation time.

Figs. 5.8 (a-c) show the distribution behavior with increasing geometric constant whereas

Figs. 5.8 (d-f) show the effect of number of size class ranges. These plots demonstrate

that the geometric progression has a major role than the number of size classes chosen.

Hence, proper choice of both the parameters is an important criteria in discretising the

model formulation and a particle size distribution is shown to depict the relation of the

particle size and the number of size classes of the original suspension (Fig. 5.9).

5.4.1.3 The method of moments

The time evolution of population balance models is generally computational expensive,

depending upon the complexity and nature of the problem, a suitable method is chosen.

The method of moments proposed by Hulburt et al. [Hulburt and Katz, 1964] is a popu-

lar technique which is used extensively by number of researchers. This method converts

the partial differential equation that is population balance equation into a set of coupled

ordinary differential equations. This method is chosen as only low order moments of the

number density function is of interest.

The population density is a function of particle sizes present in the suspension and the

number density function is used to express the number of particles, n pa in a size class L pa

to L pa+1 per unit volume of the suspension. The jth moment of the number density function
5.4. Model Solution and Simulation Results

Figure 5.8: Plots indicating the significance of the geometric constant (a-c) and number of size classes (d-f) for the model discreti-
sation.
167
5.4. Model Solution and Simulation Results

Figure 5.9: Representative plot indicating the relation of particle size and the number of size classes.
168
5.4. Model Solution and Simulation Results 169

(Equation 5.31) is described as follows:


μj = L pa j ndL pa (5.31)

Thus, the zeroth moment, μ0 of the distribution at j=0 is expressed as (Equation 5.32):


μ0 = L pa 0 ndL pa (5.32)

The first, second and third moments (Equations 5.33, 5.34 & 5.35) also provide infor-

mation about the distribution and are as follows:


μ1 = L pa 1 ndL pa (5.33)


μ2 = L pa 2 ndL pa (5.34)


μ3 = L pa 3 ndL pa (5.35)

Thus, the above equations give the various moments of number density functions which

could well be employed to evaluate further information which is discussed in the following

section.

5.4.1.3.1 Different mean sizes The moments obtained as explained in the previous sec-

tion leads to various definitions of mean size of particles. Generally, three types of means

are employed viz. number weighted mean diameter (arithmetic mean), Ln (Equation 5.36),

the surface area weighted mean diameter (Sauter mean diameter), La (Equation 5.37), and
5.4. Model Solution and Simulation Results 170

the volume weighted mean diameter (De Brouckere mean diameter), Lw (Equation 5.38).

μ2
Ln = (5.36)
μ1

μ3
La = (5.37)
μ2

μ4
Lw = (5.38)
μ3

It is important that the modeling takes into account the type of measurement being

used where validation is a subsequent step to modeling. Number of techniques are avail-

able to measure particle sizes in various forms as mentioned above i.e. based on number,

length, volume. Thus, depending upon the requirement, one can compute particle size

respectively. To consolidate, the generalized form is given as:

  j−k1
μj
L j,k = (5.39)
μk

It is important to select the correct mean size to represent the samples of particles

for the particular application. In this work, measurements are made by the Brookhaven

Particle Size Analyser (BIC 90 Plus) which gives broad range of measurements including

intensity weighted, number, area and volume weighted particle sizes. For validation study

which will be discussed in Chapter 6, the number based distribution is employed as per the
5.4. Model Solution and Simulation Results 171

definition given (Equation 5.40):


μ2 L2pa n j dL pa
Ln = = (5.40)
μ1 L pa n j dL pa

5.4.1.4 Number and volume distributions

The previous section gives a glimpse of various methods of representing the particle

sizes, but it is crucial to choose the right way of presenting the data especially when the

model validation is the key aim. The discretised population balance equation given above

represents the number density of particles (n pa ) appearing in the particle size range φ pa .

Considering S as the characteristic size of the particles in the size range and αv the volu-

metric shape factor (kv ) then the volume of particles is:

kv πS pa
ΔV pa = Δn pa (5.41)
6

Thus, having known one candidate, it is easier to calculate the other distribution. Thus

this model depending upon the actual data available, it has the advantage of converting into

the required distribution form.

5.4.1.5 Useful quantity predicted from the above model

The above formulated model clearly determines the size and size distribution profiles

which has never been attempted to date for tangential flow filtration considering the basic

balance equations and transmembrane as the driving force. But it will be very useful and

applicable to translate the above data to a measureable and or useful quantity. This work

being on virus filtration, hence the useful quantity reported here is the log reduction value

which is the critical deciding factor of accept/reject of final product. Primarily are the
5.4. Model Solution and Simulation Results 172

synonymous terms Log Reduction Value (LRV) or Log Titer Reduction (LTR) which refer

to the virus reduction factor (Equation 5.42) defined as the logarithmic10 of the ratio of

the virus concentration (titer) in the feed (C f ) and the virus concentration (titer) in the

permeate stream (C p ). In some literature, LRV is also referred to as reduction factor.

 
Cf
LRV = log10 (5.42)
Cp

5.4.2 Simulation using gPROMS


gPROMS (General PROcess Modeling System) is a numerical modeling package pro-

duced by Process Systems Enterprise Limited. gPROMS finds extensive applications with

process systems involving systems of algebraic, differential and partial differential equa-

tions, and is optimised for these systems. gPROMS uses a modular structure for the devel-

opment of entities including model, process, estimation and experimental entities. Model

entities represent the inner working of the process and can be built in a single entity or in

a modular form. Process entities provide the operating conditions for the system, such as

in initial conditions, control parameters, and other solver information. Estimation and ex-

periment entities are used for parameter estimation and optimisation procedures, with the

estimation entity containing the guesses and solution parameters for the procedure and the

experiment entity containing the experimental data. Flows of data between model entities

are represented using streams which simply carry values from one model to another. The

gPROMS user guide is a good reference point for the descriptions of the modeling process
5.4. Model Solution and Simulation Results 173

and a number of useful examples [gPROMS, 2004].

gPROMS is employed to solve the set of differential and algebraic equations for the

formulated model of filtration. The choice of this software is obvious because of its capa-

bility of solving distributed parameter systems and being a simple and friendly language

with strong numerical capabilities.

5.4.2.1 Model entities of the entire process

Each model entity has the equations that govern the behavior of the model, and incase

if more than one model is used, the streams used to connect the models are described as

well. The model entity also has the information on the solution method for the partial

differential equations if the model has partial differential equations (PDEs). The sections

used in the simulation of the models are Parameter, Distribution Domain, Unit, Variable,

Stream, Set, Boundary and Equation.

The Parameter sections lists any global parameters that are referred to in the models

that are defined in the Process entity. The Distribution Domain section defines any domains

required for the distributed variable and PDEs in the model and gives the upper and lower

limits on this domain. The Unit section defines any submodels that will be used within

the model and gives these models an identifying name that will be used within the model

equations. Units can either by singular or an array of units. The variable section lays out all

the variables that will be used within the model. The Stream section of the model defines

any Streams entering or leaving the model entity, and states the variables that are contained

within the Stream. The Set section of the model entity defines the approximation method

to be used for any distribution domains used in the model. This includes the method

of approximation for the derivatives (eg. forward difference method), the order of the
5.4. Model Solution and Simulation Results 174

solution (first, second order derivatives, etc.) and the number of solution points to use. The

Equation section of the model entity has all the algebraic, differential or PDEs equations

and it has the flexibility to solve simultaneous equations [gPROMS, 2009a,b].

5.4.2.2 Parameters and variables employed in the model development

Various types of parameters and variables could be defined in gPROMS and are glob-

ally defined that do not change during the entire simulation of the process. Parameters

are defined as being either real numbers, integers or logical values. Parameters can also

be defined as arrays of values. Variables in gPROMS can be grouped as controlled vari-

ables which are defined by the process entity and describe variables that can be directly

controlled or process variables which can vary as the simulation proceeds. All variables in

gPROMS are defined using a unit type which are defined by the user in the variable type

entities section. Variables can also be defined as arrays of values, or as a distributed vari-

able for use in PDEs. Distributed variables are variables that vary continuously along a de-

fined domain and use distributed domains described in the relevant section of the gPROMS

model [gPROMS, 2009a,b]. Table 5.2 contains a summary of the partitioned sections avail-

able for the model entity. Table 5.3 contains simplified summary of the partitioned sections

available for the process entity.

5.4.3 Simulation results


Values used for model simulation are tabulated in Table 5.4 and the parameters used in

combination for the analysis of results is summarised as three cases in Table 5.5.

5.4.3.1 Representative particle size distribution in output streams

The distributions reported (both permeate and retentate streams) throughout this chap-

ter represents the number distribution percentage normalised to the original feed stream.
5.4. Model Solution and Simulation Results 175

Table 5.2: Summary of the partitioned sections available for the model entity.
PARAMETER
Parameter declarations
VARIABLE
Variable declarations
DISTRIBUTION DOMAIN
Domain declarations
BOUNDARY
Boundary equations
TOPOLOGY
Unit connection equations
EQUATION
Model equations
INITIAL
Equations

Table 5.3: Simplified summary of the partitioned sections available for the process entity.
UNIT
Process equipment declarations
SET
Parameter value setting
INITIAL
Initial condition specfications
SOLUTION PARAMETERS
Model based activities specifications
SCHEDULE
Operating policy specifications

Table 5.4: Input data for the simulation of the model.


Parameters Values
Membrane pore size (average) 100 nm
Membrane area 0.0018 m2
Particle size of feed (average) 80-120 nm
Standard deviation of feed particle size 30 nm
Minimum feed particle size 10 nm
Weight percent concentration of particles to suspension in feed 0.1%
Feed flow rate 0.03 - 0.06 L/min
Transmembrane pressure drop 60-120 kPa

Table 5.5: Various input conditions utilised for the model simulations.
TMP, kPa CFR, L/min FPS, nm
Case 1 60 0.045 80,100,120
Case 2 60,90,120 0.06 100
Case 3 60 0.03,0.045,0.06 100
5.4. Model Solution and Simulation Results 176

Figs. 5.10 & 5.11 show the permeate and retentate particle size distributions simulated for

the input conditions reported as in Table 5.5 with crossflow rate at 0.03 L/min (case 3). It is

clearly evident that during the initial stages of filtration particles easily permeate and over

time, particles are more of fractionating than separating. The effect of particle transport

across and or through the membrane is investigated with changing operating parameters

which is discussed as case studies in the next sections.

Figure 5.10: Representative particle size distribution in the permeate at TMP=90 kPa and
CFR=0.045 L/min.

Figure 5.11: Representative particle size distribution in the retentate at TMP=90 kPa and
CFR=0.045 L/min.
5.4. Model Solution and Simulation Results 177

5.4.3.2 Effect of feed particle size on particle size distribution

Case 1: With increase in average particle size, the distribution peak shifts right (Fig.

5.12). For a given pore size, if the particle size increases, the amount of particles permeat-

ing through the membrane is getting restricted relatively and vice versa.

Figure 5.12: Particle size distribution of three range of feed particle sizes.

Figure 5.13: Particle size distribution of permeate for three feed particle size ranges.
5.4. Model Solution and Simulation Results 178

5.4.3.3 Effect of transmembrane pressure drop on the particle size distribution

Case 2: At higher transmembrane pressures, it is found that peak of the number distri-

bution shifts upwards in the permeate stream (Fig. 5.14) and decreases in retentate stream

(Fig. 5.15). When there is an increase in TMP, there is a possibility for the pore elongation

during which more of the particles permeate which is leading to the increase in the peak.

This has a direct influence on the final filtration efficiency. It is observed that with increase

in TMP, there is a decrease in the LRV values (Table 5.6). At higher TMP, as can been

seen from the permeate particle size distribution, more particle permeation hence higher

the permeate concentration which decreases the log reduction value (Table 5.6).

Figure 5.14: Permeate particle size distribution with changing transmembrane pressure at
CFR=0.045 L/min.

5.4.3.4 Effect of cross flow rate on particle size distribution

Case 3: With increase in crossflow rate, there is a drop in the peak value of permeate

streams whereas increase in the retentate streams (Figs. 5.16, 5.17). Under conditions of

low cross flow velocity, residence time of the particles on the membrane surface is high
5.4. Model Solution and Simulation Results 179

Figure 5.15: Retentate particle size distribution with changing transmembrane pressure at
CFR=0.045 L/min.

and it is found that mixture of particles permeate easily. But with increase in cross flow

rate, the more fraction of particles get convected in the retentate - this is attributed to the

fact that back transport of finer particles occur from the membrane surface on to the cross

flow. Also, at high cross flow rate, less of residence time leading to less of permeation

drag through the membrane. The previous literatures also indicate that there is a paradox

of both increase and decrease of particle drag along and across the membrane when there

is an increase in the cross flow velocity in tangential flow filtration [Foley et al., 1995a].

Nevertheless, in this study it is observed that with increase in the cross flow velocity,

the particles are tangentially carried along the membrane as retentate resulting in a better

rejection of bigger particles. This is reflected in the log reduction value with increase in

the cross flow rate (Table 5.6), there is an increase in the log reduction value meaning less

of particle permeation which is confirmed in the permeate particle size distribution.


5.4. Model Solution and Simulation Results 180

Figure 5.16: Permeate particle size distribution with changing crossflow rate at TMP=90
kPa.

Figure 5.17: Retentate particle size distribution with changing crossflow rate at TMP=90
kPa.

Table 5.6: LRV at various TMPs and crossflow rates of smaller particles (80 nm avg.).
TMP, kPa LRV CFR, L/min LRV
60 1.16 0.03 0.99
90 1.03 0.045 1.03
120 0.86 0.06 1.21
5.5. Concluding Remarks 181

5.5 Concluding Remarks


Literature suggests several models are available for flux prediction and most models

do not take into account the influence of feed particle size and size distribution. We have

shown in this and previous chapters that feed particle size has a great deal of influence on

the final quality of the filtered product. A novel model formulation based on population

balances was developed to predict the particle size and its distribution in the filtered

product streams as well as the prediction of filtration flux. This modeling effort was

solved by a discretisation method and facilitated the understanding of the behaviour of

the filtration process under the effects of operating parameters such as cross flow rate,

pressure, and feed particle size. The validation of the model will be the subject of the

following chapter, in a parameter estimation exercise that uses data presented in Chapter

4.

Bibliography
Abbas, A., Romagnoli, J. A., 2007. Multiscale modeling, simulation and validation of

batch cooling crystallization. Separation and Purification Technology 53 (2), 153–163.

159, 164

Ahmad, A., Chong, M., Bhatia, S., 2006. Ultrafiltration modeling of multiple solutes sys-

tem for continuous cross-flow process. Chemical Engineering Science 61, 5057–5069.

140, 145

Alspach, B., Allgeier, S., 2003. Application of ultrafiltration for virus removal using the
Bibliography 182

lt2eswtr regulatory framework. Proceedings, AWWA Water Quality Technology Con-

ference. 147

Alvarez, J., Alvarez, J., Hernandez, M., 1994. A population balance approach for the de-

scription of particle size distribution in suspension polymerization reactors. Chemical

Engineering Science 49, 99–113. 141

Baruah, G., Venkiteshwaran, A., Belfort, G., 2005. Global model for optimizing crossflow

microfiltration and ultrafiltration processes: A new predictive and design tool. Biotech-

nol. Prog. 21, 1013–1025. 148

Belfort, G., 1989. Fluid mechanics in membrane filtration: recent developments. Journal

of Membrane Science 40, 123–147. 144

Belfort, G., Davis, R. H., Zydney, A. L., 1994. The behavior of suspensions and macro-

molecular solutions in crossflow microfiltration. Journal of Membrane Science 96 (1-2),

1–58. 143, 149

Belfort, G., Nagata, N., 1985. Fluid mechanics and cross-flow filtration: so me thoughts.

Desalination 53, 57–79. 144, 145, 149

Bird, R., Stewart, W., Lightfoot, E., 2006. Transport Phenomena. Wiley. 146

Bodzek, M., Konieczny, K., 1994. Optimization of the ultrafiltration of latex wastewaters.

Desalination 94 (3), 289–306. 140

Bolton, G., LaCasse, D., Kuriyel, R., 2006. Combined models of membrane fouling: De-

velopment and application to microfiltration and ultrafiltration of biological fluids. Jour-

nal of Membrane Science 277 (1-2), 75–84. 146


Bibliography 183

Bowen, W. R., Sharif, A. O., 2002. Prediction of optimum membrane design: Pore en-

trance shape and surface potential. Colloids and Surfaces A: Physicochemical and En-

gineering Aspects 201 (1-3), 207–217. 143

Brandenburg, B., Zhuang, X., 2007. Virus trafficking - learning from single-virus tracking.

Nature Reviews Microbiology 5 (3), 197–208. 147

Chang, D., Hwang, S., 1995. Unsteady-state permeate flux of crossflow microfiltration:

Effect of particle size distribution. Separation Science and Technology 30, 2917–2931.

147, 148

Chang, J., Tsai, L., Vigneswaran, S., 1996. Experimental investigation of the effect of

particle size distribution of suspended particles on microfiltration. Water Science Tech-

nology 34, 133–140. 149

Chang, J., Vigneswaran, S., Kandaamy, J., Tsai, L., 2008. Effect of pore size and particle

size distribution on granular bed filtration and microfiltration. Separation Science and

Technology 43 (7), 1771–1784. 149

Chellam, S., Wiesner, M. R., 1992. Particle transport in clean membrane filters in laminar

flow. Environmental Science and Technology 26(8), 1611–1621. 149

Chudacek, M. W., Fane, A. G., 1984. The dynamics of polarisation in unstirred and stirred

ultrafiltration. Journal of Membrane Science 21 (2), 145–160. 145

Damak, K., Ayadi, A., Zeghmati, B., Schmitz, P., 2004. A new navier-stokes and darcy’s

law combined model for fluid flow in crossflow filtration tubular membranes. Desalina-

tion 161 (1), 67 – 77. 144


Bibliography 184

Driscoll, K., 2001. Development of a process simulator for the ultrafiltration/diafiltration

process. PhD Thesis University of Arkansas. 143

Foley, G., Malone, D., MacLoughlin, F., 1995a. Modelling the effects of particle polydis-

persity in crossflow filtration. Journal of Membrane Science 99, 77–88. 179

Foley, G., Malone, D. M., MacLoughlin, F., 1995b. Modelling the effects of particle poly-

dispersity in crossffow filtration. Journal of Membrane Science 99 (1), 77–88. 149

Gelbard, F., Seinfeld, J., 1978. Numerical solution of the dynamic equation for particulate

processes. J. Comp. Phys. 28, 357–375. 141

Gerstlauer, A., Ghan, C., Zhou, H., Schreiber, M., 2006. Application of population bal-

ances in the chemical industry - current status and future needs. Chemical Engineering

Science 61, 205–217. 141

Ghidossi, R., Veyret, D., Moulin, P., 2006. Computational fluid dynamics applied to mem-

branes: State of the art and opportunities. Chemical Engineering and Processing 45,

437–454. 144, 145

Ghosh, R., 2008. Handbook of Membrane Separations Ed. by Anil K Pabby and Syed S H

Rizvi and Ana Maria Sastre. CRC Press, London. 143

Gimbun, J., Nagy, Z. K., Rielly, C. D., 2009. Simultaneous quadrature method of moments

for the solution of population balance equations, using a differential algebraic equation

framework. Industrial and Engineering Chemistry Research 48 (16), 7798–7812. 160

gPROMS, 2004. General PROcess Modeling System Introductory User Guide, v2.3.1. PSE

Enterprise, UK. 173


Bibliography 185

gPROMS, 2009a. General PROcess Modeling System Model Builder Guide, v3.2.0. PSE

Enterprise, UK. 156, 174

gPROMS, 2009b. General PROcess Modeling System Model Developer Guide, v3.2.0.

PSE Enterprise, UK. 174

Green, G., Belfort, G., 1980. Fouling of ultrafiltration membranes: Lateral migration and

the particle trajectory model. Desalination 35 (1 -2-3), 129–147. 149

Guo, Y., Cheng, A., Wang, M., Zhou, Y., 2009. Purification of anatid herpesvirus 1 par-

ticles by tangential-flow ultrafiltration and sucrose gradient ultracentrifugation. Journal

of Virological Methods 161 (1), 1 – 6. 149

Hensgen, M., Czermak, P., Carlson, J., Wickramasinghe, S., 2010. Purification of minute

virus of mice using high performance tangential flow filtration. Desalination 250 (3),

1121 – 1124. 149

Huang, L., Morrissey, M., 1999. Finite element analysis as a tool for crossflow membrane

filter simulation. J. Membr. Sci. 155, 19–30. 144

Hulburt, H., Katz, S., 1964. Some problems in particle technology. Chemical Engineering

Science 19, 555–574. 140, 160, 166

Hwang, K., Liu, H., Lu, W., 1998. Local properties of cake in cross-flow microfiltration of

submicron particles. Journal of Membrane Science 138 (2), 181–192. 148

Kim, A. S., 2007. Permeate flux inflection due to concentration polarization in crossflow

membrane filtration: A novel analytic approach. European Physical Journal E 24 (4),

331–341. 145
Bibliography 186

Kleinstreuer, C., Belfort, G., 1984. Mathematical modeling of fluid flow and solute dis-

tribution in pressure driven membrane modules. Synth Membr Processes, Fundam and

Water Appl, 131–190. 145

Kumar, S., Ramakrishna, D., 1996. On the solution of population balance equations by

discretization -ii: a moving pivot technique. Chem. Eng. Sci. 51, 1333–1342. 160

Lee, Y., Clark, M., 1998. Modeling of flux decline during crossflow ultrafiltration of col-

loidal suspensions. Journal of Membrane Science 149 No. 2, 181–202. 142, 143

Lee, Y., Clark, M. M., Kim, B. J., 1996. A numerical model for flux decline during cross-

flow uf. ACS Division of Environmental Chemistry, Preprints 36 (2), 122–123. 144,

145

Lu, W.-M., Hwang, K.-J., 1995. Cake formation in 2-d cross-flow filtration. AICHE Jour-

nal 41 (6), 1443–1455. 148

Marchisio, D. L., Pikturna, J. T., Fox, R. O., Vigil, R. D., Barresi, A. A., 2003. Quadrature

method of moments for population-balance equations. AICHE Journal 49 (5), 1266–

1276. 160

McCoy, B., 1995. Membrane sieving of a continuous polydisperse mixture through dis-

tributed pores. Separation Science and Technology 30 N. 4, 487–507. 148

Meltzer, T., Jornitz, M., 2007. Filtration and Purification in the Biopharmaceutical Indus-

try. CRC Press. 141

Michalsky, R., Passarelli, A., Pfromm, P., Czermak, P., 2009. Purification of the bac-
Bibliography 187

ulovirus autographa californica m nucleopolyhedrovirus by tangential flow ultrafiltra-

tion. Desalination 245 (1-3), 694 – 700. 149

Mohammadi, T., Kohpeyma, A., Sadrzadeh, M., 2005. Mathematical modeling of flux

decline in ultrafiltration. Desalination 184, 367–375. 140, 142, 146, 150

Nassehi, V., 1998. Modeling of combined navier-stokes and darcy flows in crossflow mem-

brane filtration. Chemical Engineering Science 53, 1253–1265. 144

Pak, A., Mohammadi, T., Hosseinalipour, S., Allahdini, V., 2008. Cfd modeling of porous

membranes. Desalination 222, 482–488. 144

Porter, M., 1990. Handbook of Industrial Membrane Technology. Noyes. 149

Ramakrishna, D., 2000. Population Balances: Theory and applications to particulate sys-

temsin Engineering. Academic Press. 141, 160

Randolph, A., 1964. Population balance for countable entities. Canadian Journal of Chem-

ical Engineering 42(6), 280. 140, 141, 155

Randolph, A., Larson, M., 1988. Theory of Particulate Processes: Analysis and Techniques

of Continuous Crystallization. Academic Press. 153, 154

Randolph, A., White, E., 1977. Modeling size dispersion in the prediction of crystal size

distribution. Chemical Engineering Science 32, 1067–1076. 140, 155

Rautenbach, R., Schock, G., 1988. Ultrafiltration of macromolecular solutions and cross-

flow microfiltration of colloidal suspensions. a contribution to permeate flux calcula-

tions. Journal of Membrane Science 36 (C), 231–242. 145


Bibliography 188

Rawlings, J., Witkowski, W., Eaton, J., 1992. Modeling and control of crystallisers. Pow-

der Technology 69, 3–9. 141

Richard Bowen, W., Williams, P. M., 2001. Prediction of the rate of cross-flow ultrafiltra-

tion of colloids with concentration-dependent diffusion coefficient and viscosity - theory

and experiment. Chemical Engineering Science 56 (10), 3083–3099. 143

Romero, C., Davis, R. H., 1990. Transient model of crossflow microfiltration. Chemical

Engineering Science 45 No.1, 13–25. 149

Roorda, J. H., te Poele, S., van der Graaf, J. H. J. M., 2004. The role of microparticles in

dead-end ultrafiltration of wwtp-effluent. Water science and technology 50 (12), 87–94.

147

Rosner, D. E., McGraw, R., Tandon, P., 2003. Multivariate population balances via mo-

ment and monte carlo simulation methods: An important sol reaction engineering bi-

variate example and "mixed" moments for the estimation of deposition, scavenging, and

optical properties for populations of nonspherical suspended particles. Industrial and

Engineering Chemistry Research 42 (12), 2699–2711. 160

Santos, A., Bedrikovetsky, P., 2004. Size exclusion during particle suspension transport in

porous media: stochastic and averaged equations. Computational and Applied Mathe-

matics 23 N. 2-3, 259–284. 140

Sarkar, B., DasGupta, S., De, S., 2009. Flux decline during electric field-assisted cross-

flow ultrafiltration of mosambi (citrus sinensis (l.) osbeck) juice. Journal of Membrane

Science 331 (1-2), 75–83. 145


Bibliography 189

Song, L., 1998. Flux decline in crossflow microfiltration and ultrafiltration: mechanisms

and membrane fouling. Journal of Membrane Science 139, 183–200. 145, 150

Stoller, M., 2009. On the effect of flocculation as pretreatment process and particle size

distribution for membrane fouling reduction. Desalination 240(1-3), 209–217. 149

Su, J., Gu, Z., Li, Y., Feng, S., Xu, X. Y., 2008. An adaptive direct quadrature method of

moment for population balance equations. AICHE Journal 54 (11), 2872–2887. 160

Tan, H., Goldschmidt, M., Boerefijn, R., Hounslow, M., Salman, D., Kuipers, J., 2005.

Population balance modeling of fluidized bed melt granulation: An overview. Chemical

Engineering Research and Design 83, 871–880. 141

Tarabara, V., Hovinga, R., Wiesner, M., 2002. Constant transmembrane pressure vs con-

stant permeate flux: Effect of particle size on crossflow membrane filtration. Environ-

mental Engineering Science 19, 343–355. 149

Vyas, H., Bennett, R., Marshall, A., 2000. Influence of feed properties on membrane foul-

ing in crossflow microfiltration of particulate suspensions. International Dairy Journal

10, 855–861. 148

Wakeman, R., 1994. Visualisation of cake formation in crossflow microfiltration. Transac-

tions in Industrial Chemical Engineering 72, 530–540. 148

Wakeman, R. J., Akay, G., 1994. Flux decay and rejection during micro- and ultra-filtration

of hydrophobically modified water-soluble polymers. Journal of Membrane Science

91 (1-2), 145–152. 145


Bibliography 190

Wijmans, J., Nakao, S., Smolders, C., 1984. Flux limitation in ultrafiltration: Osmotic

pressure model and gel layer model. Journal of Membrane Science 20, 115–124. 145

Winzeler, H. B., Belfort, G., 1993. Enhanced performance for pressure-driven membrane

processes: The argument for fluid instabilities. Journal of Membrane Science 80, 35–48.

149

Zaamouche, R., Beicha, A., Sulaiman, N. M., 2009. Cross-flow ultrafiltration model based

on concentration polarization. Journal of Chemical Engineering of Japan 42 (2), 107–

110. 145

Zhang, M., Song, L., 2000. Mechanisms and parameters affecting flux decline in cross-flow

microfiltration and ultrafiltration of colloids. Environmental Science and Technology

34 (17), 3767–3773. 146


Chapter 6

Optimization-based Parameter

Estimation

Contents
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192

6.2 Overview of Parameter Estimation . . . . . . . . . . . . . . . . . . . . . 196

6.3 Filtration Kinetic Parameter Estimation . . . . . . . . . . . . . . . . . . 197

6.4 Estimation Results and Discussion . . . . . . . . . . . . . . . . . . . . . 200

6.4.1 Parameter estimation summary . . . . . . . . . . . . . . . . . . . . 200

6.4.2 Overlay plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202

6.5 Model Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202

6.5.1 Validation against literature data . . . . . . . . . . . . . . . . . . . 208

6.6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
6.1. Introduction 192

Preface
Kinetics of filtration flux together with associated mechanisms are well covered in the

literature. Whereas the kinetics related to the stochastic behavior of particle permeation

has not received much attention. The unknown parameters that emerge from the filtration

model presented in Chapter 5 are estimated using maximum likelihood theory in the cur-

rent chapter. Such method is an optimization-based approach that utilizes the non-linear

model and experimental data of the filtration and particle states. First, an overview of

parameter estimation is given before presenting details of the estimation problem. The

results of the parameter estimation exercise are then presented and finally the model with

its estimated parameters is validated against data from literature.

Research Questions:

1. What are the feasibility of estimation of filtration kinetic parameters via

optimization-based parameter estimation techniques?

2. Can the model be validated with (i) our experimental results and (ii) available data

from others’ works?

3. What potential does this kinetic parameter estimation approach have for filtration

kinetics identification in general?

6.1 Introduction
Several membrane transport models have been proposed for flux decline and fouling

mechanisms based on film theory, solution diffusion model, concentration polarization

models to name a few. All these models invariably employ simple linear or nonlinear meth-
6.1. Introduction 193

ods or trial and error to estimate the parameters that emerge from those models (among

others [Cassano et al., 2007; Denisov, 1994; Murthy and Chaudhari, 2008]). For example,

the kinetics of cake layer formation during cross flow filtration is studied by Knyaz’kova

and the pore blocking model is solved by linear dependence of flux-volume [Knyaz’kova,

1995]. Similar studies have been reported elsewhere for cross flow filtration [Baruah et al.,

2005; Jiao and Sharma, 1994; Mackley and Sherman, 1992a]. The simple models used in

filtration kinetic studies do not capture the true kinetics and are limited because temporal

data of particle size are rarely used. Our model presented in Chapter 5 defines the filtration

kinetics relating to the stochastic behavior of particle permeation. The model is itself part

of the parameter estimation formulation and along with size and concentration temporal

data is used to estimate unknown parameters. This approach is unique, providing a sys-

tematic way to determine kinetic parameters and thus bridges a gap towards more accurate

estimation of filtration kinetics.

Parameter estimation, also called data regression is an important step involved in the

formulation and validation of mathematical models. The model kinetic system contains

parameters (recall Chapter 5) that need to be identified via a parameter estimation step,

the latter being an important prerequisite step for simulation, dynamic optimization, and

model-based control. Essentially, the parameter estimation problem is an optimization one

where the model is fitted to the experimental data. Once the parameters are estimated, the

identified model maybe used for predictive simulation analysis where the various process

aspects may be elucidated, and used for optimization and control scheme development. A

number of efficient and robust methods have been developed depending upon the structure

of objective functions. Various criteria relating to parameter estimation problems include:


6.1. Introduction 194

• Model framework: defined for algebraic and differential equation models as well as

for linear or nonlinear models.

• Objective function selection: refers to a scalar function with dependence to the cho-

sen parameters. The choice of the objective function dictates the value of the esti-

mates as well as the statistical properties.

• Solution techniques: developed for specific algorithms and methods to minimize or

maximize the objective function. Generally, the parameter estimation problem is

solved by any one of the five optimisation problems (Table 6.1).

• Statistical properties: uncertainty in model parameters as well as in calculated val-

ues.

• Model adequacy: validations to how good the model responds to the system.

Table 6.1: The types of parameter estimation problems.


Parameter Estimation Methods Description
Choosing the best function to min- The sum of the differences between the model and experi-
imise mental data is minimised. This is known as method of least
squares.
Minimisation of the chosen func- Minimisation of the chosen function with respect to the pa-
tion rameters where the value(s) of estimated parameter(s) at the
minimum represents the optimum.
Optimal design of experiments I Optimal design of experiments to obtain the best parameter
estimates. Usually the model structure is known.
Optimal design of experiments II Optimal design of experiments to choose a model structure
from several competing models
Infinite number of model structures Determination of the model structure when there is a lack
of data which is considered to be the toughest parameter
estimation methods. This is known as system identification
problem.

The key feature of our approach of parameters estimation is using maximum likeli-

hood function. It determines values for the unknown parameters in order to maximize the
6.1. Introduction 195

probability that the mathematical model will predict the values obtained from the exper-

iments. The complete details of this technique are available in [gPROMS, 2004, 2009a].

The overall scheme employed in a typical parameter estimation strategy is shown in Fig.

6.1.

Figure 6.1: Flowchart showing the various steps involved in parameter estimation: pre-
estimation and post-estimation checks in arriving at the accurate values.

It is a known fact that the parameter estimation is extremely sensitive to initial guesses
6.2. Overview of Parameter Estimation 196

provided for the unknown parameters. If the initial guesses are not close to the final solu-

tion, the estimation procedure fails. It is thus suggested that first the model parameters are

determined crudely such that simulated trends follow the experimental results. In the rest

of this chapter, optimization-based parameter estimation is presented with specific focus

on our filtration kinetics problem.

6.2 Overview of Parameter Estimation


The model developed and described in Chapter 5 has unknown parameters (k1 ,k2 ,k3 )

to be estimated. Generally, models are greatly improved and understood by experimental

studies. The data obtained from the experimental investigation discussed in Chapter 4 are

employed for this estimation exercise.

The following details are specified in a parameter estimation exercise:

• The initial condition in the experimental run

For the model developed, the number density of the particles at time t=0 is zero.

• The duration of the experimental run

The experiment is carried out for 600 secs

• Values of time-independent quantity

The operating parameters i.e. transmembrane pressure (range 20-60 kPa), cross-

flow velocity (0.3-1 L/min)

The input parameters i.e. feed particle size (average particle size of 3 different

ranges lying 35.6-142.9nm.)

• Time variant output quantities


6.3. Filtration Kinetic Parameter Estimation 197

Permeate concentration and particle size

Parameter estimation problem is solved using gEST tool of gPROMS package. The

MXLKHD solver, an indirect solver, is used where the global optimum is found by ap-

plying sequential quadratic programming method. This solver takes advantage of the DA-

SOLV solver for solution of the model equations.

6.3 Filtration Kinetic Parameter Estimation


The filtration model described in Chapter 5 and the equations constituting the model

can be represented by Y(t) = [D(t), C(t)], U = [TMP, CFR] as the manipulated variables

and θ representing the vector of parameters to be identified which in this case is [k1 , k2 ,

k3 ]. The problem can be formulated as follows:

f (Y(t), U, θ) = 0 (6.1)

Equation 6.1 is accompanied by a set of initial conditions for the differential equation

of the model and these can be collated into the vector Y(0) = y, where y is a set of values

in this case representing [D(0), C(0)]. For each experiment the measured variables are

denoted as follows:

Ŷ(t) = [D(t),
ˆ C(t)]
ˆ (6.2)

Two criteria are generally used in parameter estimation, namely the least squares or

maximum likelihood. Maximum likelihood is a mathematical optimization technique that

attempts to find a best fit to a set of data by minimizing the error between the experimen-

tal measurements and predicted values. The least squares is effectively a special case of
6.3. Filtration Kinetic Parameter Estimation 198

maximum likelihood method.

Considering the problem specified the maximum likelihood criterion that describes the

highest probability of the model predicting the real data is given by the following objective

function:

⎧   ⎡ ⎤⎫
N 1 ⎪ NE NVi N Mi j ⎢⎢
⎨ ⎢ (Yˆi jk − Yi jk )2 ⎥⎥⎥⎪⎬
Φ(θ) = ln(2π) + min
. ⎪
⎩ ⎢⎣ln(σ i jk ) +
2 ⎥⎦⎪
⎭ (6.3)
2 2 i=1 j=1 k=1 σ i jk
2

where N is the number of measurements taken, NE is the number of experiments, NVi

is the number of variables measured in the ith experiment, N Mi j is the number of measure-

ments of the jth variable in the ith experiment, σ2i jk is the variance of the kth measurement of

variable j in experiment i. The variance σ2i jk can be described by any of a number of models

available when the error structure of the data is known [gPROMS, 2009b]. These are ei-

ther homoscedastic where the variance error is assumed to be constant, or heteroscedastic

where the variance error is a function of the measured and predicted values and may be

proportional to Ŷ(t)γ/2 and Y(t)γ/2 , where γ is a parameter determined in the optimisation

range which is between 0 and 1.

In the homoscedastic case, used in this work, the variance error may be determined by

analyzing similar sets of experimental data. The value of the variance can be specified in

the estimation as:

σ2i jk = ω2i jk (6.4)

where ω is the constant standard deviation of the measurement error and the objective
6.3. Filtration Kinetic Parameter Estimation 199

function is given as follows:

⎧  ⎡ ⎤⎫

⎨ NVi N Mi j ⎢⎢⎢ (Yˆi jk − Yi jk )2 ⎥⎥⎥⎪⎬
Φ(θ) = min ⎩ j=1 k=1 ⎢⎣ln(ω i jk ) + ⎥⎦⎪
2
. ⎪ ω2 i jk ⎭ (6.5)

For some other cases, more information is available about the data, and the error struc-

ture may be known to depend on the magnitude of the measured or predicted values. For

example, it may be known that Ŷ(t) increases with increase in the variance of Ŷ(t). For

these cases, the heteroscedastic model of variance is employed:

σ2i jk = ω2i jk Ŷi2jk (6.6)

In the above context, the maximum likelihood criterion becomes a weighted least

squares problem as follows:

⎧ ⎡ ⎤⎫

⎨NVi N Mi j ⎢⎢⎢⎢
⎪ (Yˆi jk − Yi jk )2 ⎥⎥⎥⎪ ⎪
Φ(θ) = min
. ⎪ ⎢⎣ln(ω i jk Ŷi jk ) +
2 2 ⎥⎥⎦⎬
⎪ (6.7)

⎩ j=1 k=1 ω2 i jk Ŷi2jk ⎪ ⎭

By this definition, the error weights, ωi jk , are chosen to reflect the relative level of

confidence in the measurements. This leads to the minimisation of the objective function

with more weight given to the data that has less error. For pure heteroscedastic variance

models, the structure of the error takes the more complex form as

σ2i jk = ω2i jk [Ŷi2jk ]γ (6.8)

If γ has the value of 0, Equation 6.8 reduces to Equation 6.4 and at the other extreme,

where γ has the value 1, Equation 6.8 becomes equal to Equation 6.6.
6.4. Estimation Results and Discussion 200

For our problem, the measured variables D(t) and C(t) are described by constant rel-

ative variance model on the measured values. The choice of this model is justified as the

measurement errors in both variables are constant and do not depend on the magnitude of

the measurement. It is possible using the gEST parameter estimation facility in gPROMS

to specify the parameter estimation problem according to any one type of variance models

mentioned above. gEST also allows for the determination of ω along with θ as part of the

optimisation of the estimation. The specification of the variance model is a simple pro-

cedure in gPROMS and can typically be shown as in Table 6.2 for the above mentioned

measurements of parameter estimation problem.

Table 6.2: Specification of the variance model in the gEST file.


Measure
Permeate.Mean2 [i.e. D(t)]
CONSTANT-RELATIVE-VARIANCE (1 : 0.01 : 3)
Measure
Permeate.Cp [i.e. C(t)]
CONSTANT-RELATIVE-VARIANCE (1 : 0.01 : 3)

6.4 Estimation Results and Discussion

6.4.1 Parameter estimation summary


Table 6.3 shows the summary of the parameter estimation stage indicating the constant

relative variance model being employed for the measured variables (D(t) and C(t)). The

objective function contributions to each experimental variable associated in the kinetic

parameter estimation is also listed. It shows that the objective function of both D(t) and

C(t) are of the same order for the measured experimental data.

Table 6.4 summarizes the estimated values of the model parameters. As mentioned

earlier, constant relative variance model is employed for the parameters estimates along
6.4. Estimation Results and Discussion 201

Table 6.3: Parameter estimation summary showing the number of experiments used for
estimating the parameters, the variables D(t) and C(t) measured, variance model employed
and the objective function contribution.
Variables Measured Variance Model Objective Function Contribution
Experiment 1: PEB1
PERMEATE.MEAN2 CONSTANT RELATIVE VARIANCE -64.442
PERMEATE.CP CONSTANT RELATIVE VARIANCE -51.841
Experiment 2: PEB2
PERMEATE.MEAN2 CONSTANT RELATIVE VARIANCE -64.414
PERMEATE.CP CONSTANT RELATIVE VARIANCE -51.133
Experiment 3: PEB3
PERMEATE.MEAN2 CONSTANT RELATIVE VARIANCE -64.389
PERMEATE.CP CONSTANT RELATIVE VARIANCE -48.078
Experiment 4: PEM2
PERMEATE.MEAN2 CONSTANT RELATIVE VARIANCE -48.349
PERMEATE.CP CONSTANT RELATIVE VARIANCE -34.983
Experiment 5: PEM3
PERMEATE.MEAN2 CONSTANT RELATIVE VARIANCE -48.304
PERMEATE.CP CONSTANT RELATIVE VARIANCE -35.012
Experiment 6: PES1
PERMEATE.MEAN2 CONSTANT RELATIVE VARIANCE -64.441
PERMEATE.CP CONSTANT RELATIVE VARIANCE -52.217
Experiment 7: PES2
PERMEATE.MEAN2 CONSTANT RELATIVE VARIANCE -64.375
PERMEATE.CP CONSTANT RELATIVE VARIANCE -52.258
Experiment 8: PES3
PERMEATE.MEAN2 CONSTANT RELATIVE VARIANCE -64.357
PERMEATE.CP CONSTANT RELATIVE VARIANCE -46.990

with with initial guess of ω = 0.1 and lower and upper bounds are 0.01 and 3.0 respectively.

Table 6.4: Model parameter estimates.


Parameter Optimal Estimate
k1 75.9
k2 -4.20
k3 2.91*10−3

Table 6.5 shows the correlation matrix, R, which is calculated from the variance co-

variance matrix (Vi j ) as follows:

Vi j
Ri j =  ,i  j (6.9)
Vii V j j

Ri j = 1, i = j (6.10)
6.5. Model Validation 202

Values close to 1 in the off diagonals indicate a high correlation of the corresponding

factors and vice versa. So, in our work, it is found that k1 , k2 , and k3 are not very highly

correlated.

Table 6.5: Correlation matrix.


Parameter Optimal Value Parameter Number 1 2 3
k1 75.9 1 1.00 -.999 0.516
k2 -4.2 2 -.999 1.00 -0.520
k3 2.91*10−3 3 0.516 -0.520 1.00

6.4.2 Overlay plots


As explained earlier, the estimation problem is solved by the gEST facility and the

optimal solution points of the parameters are recorded in Table 6.4. The estimated values

obtained are found to fit the experimental values satisfactorily which is reflected in Table

6.6, i.e. weighted residual value is much less than χ2 -value.

Table 6.6: Lack of fit test.

Weighted Residual χ2 value (95%) Comment


34.5 66.339 Good fit: weighted residual less than χ2 -value

6.5 Model Validation


The optimal estimated parameters thus obtained (Table 6.4) are employed for model

prediction as shown in Figs. 6.2 & 6.3. The model fit is satisfactory whereas the mismatch

observed between the measured values and the predicted values are attributed to experi-

mental and characterization errors, but unmodelled mechanisms could also contribute to

this mismatch.

The model is now validated with the optimally estimated parameters, and simulation

analysis can be carried out. First, the model simulations of the output streams are shown
6.5. Model Validation
203

Figure 6.2: Comparison of measured values and predicted values during the parameter estimation step.
6.5. Model Validation
204

Figure 6.3: Comparison of measured values and predicted values during the parameter estimation step.
6.5. Model Validation 205

for three different particle sizes (Figs. 6.4, 6.5, 6.6 & 6.7). It is important to note that such

model simulation results in the area of filtration, mechanistic understanding of particle size

distribution considering the operating parameters (TMP and CFR) and the input conditions

(FPS) are unique to this study. Experimental results discussed in Chapter 4 are employed

for validation purposes. Specifically, the model predictions of filtration efficiency and

permeate particle size distribution are tested. The results of this validation are presented

next.

Figure 6.4: Simulation results of permeate particle size distribution for bigger sized parti-
cles (140nm avg.) at TMP=40 kPa and CFR=0.6 L/min.

Figure 6.5: Simulation results of retentate particle size distribution for bigger sized parti-
cles (140nm avg.) at TMP=40 kPa and CFR=0.6 L/min.
6.5. Model Validation 206

Figure 6.6: Simulation results of permeate particle size distribution for medium sized par-
ticles (80nm avg.) at TMP=40 kPa and CFR=0.6 L/min.

Figure 6.7: Simulation results of retentate particle size distribution medium sized particles
(80nm avg.) at TMP=40 kPa and CFR=0.6 L/min.
6.5. Model Validation 207

The filtration efficiency values are plotted to compare the model predictions with the

experimental values. It is clearly evident that model predictions agree well qualitatively

with the experimental results but it is important to put forth the fact that variation is ob-

served in the bigger particles (Fig. 6.8). The deviation in the results may be attributed to

the fact that the model suffers in calculating the solid permeate concentration as it is more

dependent on the estimated parameters. More experiments will help to resolve this.

Figure 6.8: Model predictions vs. experimental results of filtration efficiency at TMP=40
kPa and CFR=0.6 L/min.

Now, the mean diameter obtained from the model and the experimental results are

compared and it is quite evident that both agree well (Fig. 6.9). The deviation is more

pronounced in the case of smaller particles in comparison with the medium and bigger

particles owing to the fact that the smaller particles easily pass through the membrane

wherein particle sizes are represented in a distributed form but pore size is not defined

as a distribution but in the actual scenario it has a distribution. Literature also reports

that the membrane has bigger pores which are not covered in the average pore calculation

[Amirilargani et al., 2009; Sun et al., 2007]. The model suffers this limitation of non
6.5. Model Validation 208

inclusion of pore size distribution which is not addressed in this thesis.

Figure 6.9: Model predictions vs. experimental results of permeate particle size for all
three particle size ranges at TMP=40 kPa and CFR=0.6 L/min.

The model’s ability to predict particle size distribution is also validated (Figs. 6.10,

6.11 & 6.12). The model demonstrates a fair ability in predicting the permeate particle

size distribution. The variation observed in the particle size distribution of the smaller

particles is due to model’s inability to predict the multimodal distribution. This result

represents a milestone in UF modeling and answers the research questions posed at the start

of this chapter, that is, that the optimisation-based estimation can be a feasible approach

for filtration kinetics calculations towards validated UF models.

6.5.1 Validation against literature data


In this section, the model is validated with other researchers’ literature data. Though

a number of research works are available in virus filtration, as mentioned earlier, very

limited data is available with respect to the particle size focus. Here, Sano et al.’s [Sano

et al., 2006] experimental results are used to validate the model for the filtration efficiency

prediction. Sano’s work does not deal with the particle size distribution but it is exploited
6.5. Model Validation 209

Figure 6.10: Model predictions vs. experimental results of permeate particle size distribu-
tion at TMP=40 kPa and CFR=0.6 L/min of bigger sized particles.

Figure 6.11: Model predictions vs. experimental results of permeate particle size distribu-
tion at TMP=40 kPa and CFR=0.6 L/min of medium sized particles.
6.5. Model Validation 210

Figure 6.12: Model predictions vs. experimental results of permeate particle size distribu-
tion at TMP=40 kPa and CFR=0.6 L/min of smaller sized particles.

for the log reduction value prediction results only. In their work, removal of indigenous

noroviruses through microfiltration and UF membranes of sewage sludge and waste water

were studied. Table 6.7 gives the data used for predicting the filtration efficiency.

Table 6.7: Validation of the model with the Sano et al.’s [Sano et al., 2006] results.
Sample Sample Membrane Norovirus size, nm TMP, kPa LRV (Sano et al.) LRV (Our model)
Case 1 Sewage Sludge MF (0.1 μ) 20-30 50.00 4.00 4.05
Case 2 Waste water MF (0.1 μ) 20-30 50.00 2.17 2.00
Case 3 Sewage Sludge UF (100 kDa) 20-30 100.00 5.07 4.60
Case 4 Waste water UF (100 kDa) 20-30 100.00 3.19 3.90

It clearly shows that the model predicts quite precisely in two cases (i.e. Case 1 and 2)

whereas, it differs by close to 10 % in Cases 3 & 4. These results are a clear indication of

the viability of the model for filtration efficiency predictions knowing the input condition

(original particle sizes of the feed stream) and for any operating parameters. The particle

size distribution validation becomes difficult because of the lack of literature data.

To conclude, the current novel model successfully predicts the more vital indicator of

filtration, i.e. log reduction value of virus filtration, the permeate particle size and distri-
6.6. Concluding Remarks 211

bution which have a significant role especially during complete virus removal operation

and/or fractionation of proteins during the filtration mechanism. A number of limitations

in the current form of the model needs to be highlighted: (i) pore size distribution is not

modelled and (ii) multimodal size distribution prediction is lacking. From the above vali-

dation studies, it is very clear that the model can serve as promising tool for optimization

studies which are capable of predicting the optimum operating parameters to maximize the

filtration efficiency and to obtain the optimum permeate particle size.

6.6 Concluding Remarks


This chapter has successfully estimated the unknown parameters of the model pre-

sented in Chapter 5. This estimation was based on maximum likelihood theory and repre-

sents a unique approach in the field of filtration kinetic identification. The model predic-

tions are shown to be in very good agreement with the experimental results, thus verifying

the validity of the model for log reduction value and the permeate particle size prediction.

The model was also found to be in close agreement with filtration efficiency literature data.

The validated model demonstrated its prediction capability through simulations of the per-

meate particle size distribution under various TMP and CFR conditions. The model is thus

proposed as a key tool for scale-up of UF operations.

Bibliography
Amirilargani, M., Sadrzadeh, M., Mohammadi, T., 2009. Synthesis and characterization

of polyethersulfone membranes. Journal of Polymer Research, 1–15. 207

Baruah, G., Venkiteshwaran, A., Belfort, G., 2005. Global model for optimizing crossflow

microfiltration and ultrafiltration processes: A new predictive and design tool. Biotech-
Bibliography 212

nol. Prog. 21, 1013–1025. 193

Cassano, A., Marchio, M., Drioli, E., 2007. Clarification of blood orange juice by ul-

trafiltration: Analyses of operating parameters, membrane fouling and juice quality.

Desalination 212, 15–27. 193

Denisov, G., 1994. Theory of concentration polarization in cross-flow ultrafiltration: Gel-

layer model and osmotic-pressure model. Journal of Membrane Science 91 No. 1-2,

173–187. 193

gPROMS, 2004. General PROcess Modeling System Introductory User Guide, v2.3.1. PSE

Enterprise, UK. 195

gPROMS, 2009a. General PROcess Modeling System Model Builder Guide, v3.2.0. PSE

Enterprise, UK. 195

gPROMS, 2009b. General PROcess Modeling System Model Developer Guide, v3.2.0.

PSE Enterprise, UK. 198

Jiao, D., Sharma, M. M., 1994. Mechanism of cake buildup in crossflow filtration of col-

loidal suspensions. Journal of colloid and interface science 162 (2), 454–462. 193

Knyaz’kova, T. V., 1995. Immobilization of polymers on cellulose acetate membranes.

Catalysis Today 25 (3-4), 255 – 262. 193

Mackley, M., Sherman, N., 1992. Cross-flow cake filtration mechanisms and kinetics.

Chemical Engineering Science 47 (12), 3067 – 3084. 193


Bibliography 213

Murthy, Z., Chaudhari, L., 2008. Application of nanofiltration for the rejection of nickel

ions from aqueous solutions and estimation of membrane transport parameters. Journal

of hazardous materials 160 (1), 70–77. 193

Sano, D., Ueki, Y., Watanabe, T., Omura, T., 2006. Membrane separation of indigenous

noroviruses from sewage sludge and treated wastewater. Water Science and Technology

54 (3), 77–82. xxviii, 208, 210

Sun, W., Chen, T., Chen, C., Li, J., 2007. A study on membrane morphology by digital

image processing. Journal of Membrane Science 305 (1-2), 93–102. 207


Chapter 7

Conclusions and Future Directions

Contents
7.1 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214

7.2 Recommendations for Future Directions . . . . . . . . . . . . . . . . . . 218

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220

7.1 Concluding Remarks


Tangential flow UF of silica particles (model virus particles) was investigated to iden-

tify the significant factors that affect the filtration efficiency. Factors considered were feed

particle size and the operating parameters transmembrane pressure and cross flowrate. Ex-

perimental results indicate that feed particle size has a great deal of influence on surrogate

virus particle log reduction efficiency in UF. The proper selection of operating parameters

coupled with proper membrane selection will result in better filtration efficiency. Hav-

ing identified the feed particle size as a prominent factor for the filtration mechanism, a

detailed investigation on the feed particle size on the permeate quality was then carried

out. This filtration efficiency investigation showed that among the two operating param-

eters considered, transmembrane pressure plays a major role in controlling the filtration
7.1. Concluding Remarks 215

efficiency in comparison with flowrate. In the studied experimental range, higher LRV

values are obtained at lower TMP (20 kPa) and at higher feed flowrate (1 L/min). Further,

the effect on LRV of the interaction between TMP and FPS seems to be more significant

than that of the interaction of flowrate with FPS. The empirical statistical model developed

serves as a powerful tool for studying the impact of the key operating parameters on the

permeate quality (Figs. 4.9 & 4.13) and can be employed for optimisation of operating

parameters to achieve maximum efficiency. The above findings are hereby summarized:

• The feed particle size, transmembrane pressure and cross flowrate have a significant

role in the final filtration efficiency. Among the above three parameters considered,

FPS and TMP are seen to be major key factors with CFR’s significance being rel-

atively less. It is very insightful to note that the interplay of parameters plays an

equally important role as well. It has been identified that interplay of TMP and FPS

are more influential than TMP and CFR or CFR and FPS. This necessitates a further

investigation on the feed particle size on the separation or fractionation mechanism.

• The effect of above factors was quantified in an empirical model development. The

model results obtained were compared with experimental results and found to be

in in excellent agreement. Hence, the model obtained from this study could, given

TMP and CFR, be employed to predict the filtration efficiency and the feed particle

property within the studied experimental range.

• The above study provided clear indications about the dependencies of filtration effi-

ciency on these factors. This will serve as an excellent approach/method to analyze

the effects of influential parameters on product efficiency for larger scale operations.
7.1. Concluding Remarks 216

In a second experimental investigation, the impact of size effect on the filtration mech-

anism (i.e. rejection and/or fractionation) was considered. For fractionation of particles,

suitable transmembrane pressure should to be chosen whereas if the particles are to be

separated, cross flow velocity is the key factor for better efficiency of membrane filtration.

From the experimental range employed in this study, it was found that the lower the trans-

membrane pressure, the more of separation, while the higher the transmembrane pressure,

a combination of separation and fractionation occurs. Whereas with change in cross flow

velocity, separation mechanism dominates and has good control over filtration efficiency.

Also the polydispersity study showed that the lower the dispersion of particles, the higher

the filtration efficiency attributed to the lower levels of inter particle forces. Detailed sta-

tistical investigation indicated that permeate particle size follow a polynomial dependency

on the operating parameters whereas permeate PDI varies linearly with the operating pa-

rameters. The particle level study covered in this article offers opportunities for academic

and industrial researchers to gain fundamental insights into how particle size properties are

linked to permeate quality.

The ultimate purpose of modeling filtration processes is the mathematical understand-

ing of the dynamic behavior of the process. A validated model can be implemented in sim-

ulation, optimization and control of the process, overall helping to achieve desired product

characteristics. Typical filtration models found in literature focus on flux predictions and

fouling mechanisms. Our comprehensive filtration model, based on population balances,

is novel approach in the field of virus filtration. It takes into account individual parti-

cles and their particle size property, equations of conservation, and kinetic submodel. The

model developed is capable of successfully predicting the permeate particle size and size
7.1. Concluding Remarks 217

distribution as well as final filtration efficiency. This represents a big leap in understanding

filtration mechanisms and process behaviour and importantly can be applied in scale-up

studies. It is also important to note down the limitations of the model when applied for

industrial settings:

• The model assumes that fouling is negligible which may be a limiting factor in in-

dustrial applications.

• The input parameters are to be characterized precisely which are the input conditions

in the model which have a direct influence on the accuracy of final predicted values.

For example, measuring particle size in the feed is a challenge and therefore it limits

the use of this methodology in the industrial setting.

• Another issue with regard to the feed particle size, in industrial setting, there may

be some particles which are beyond the defined range of particle size in the model.

This may affect the final results.

• Information technology issue is another challenge in implementing the model in the

industrial setting. It needs to be overcome as the model will need to receive data

from the process on the fly (i.e. in real time). The speed of the model here becomes

a limitation; it needs to be recorded in a faster algorithm.

• During inline testing in industrial settings, intermittent manual monitoring may be

required to check the particle size measurement and concentration in order to avoid

some erratic behavior.

Finally, the last part of this study successfully estimates the unknown filtration kinetic

parameters based on maximum likelihood theory. This optimisation-based approach uses


7.2. Recommendations for Future Directions 218

the developed non-linear model. The model was shown to provide a good fit to the experi-

mental data. Minor mismatch between model output and experimental data was observed

in quantitative behavior of the filtration efficiency values for the bigger feed particles. The

model predictions in the form of particle size distribution agree well with the experimental

results for medium and bigger particles, whereas smaller particles present some mismatch.

This was attributed to experimental errors as well as unmodelled filtration mechanisms and

highlights the need for more experimental data as well as the importance of incorporating

multimodal distribution forms in the model. The model was then shown to be closely

matching with literature data on LRV. The optimisation-based parameter estimation ap-

proach is regarded as a milestone in filtration kinetic parameter estimation which lends a

stong hand in developing validated fitlration models more complex than those that solely

depend on filtration flux kinetics.

7.2 Recommendations for Future Directions


The first part of this study in Chapter 4 involving the filtration of three different particles

can be extended to include:

• A more extensive investigation in which bigger and smaller feed particles are mixed.

This will provide more understanding about the effect of particle-particle interaction

on the final filtration efficiency.

• The role played by the membrane pore size distribution on particle-pore interactions

and the impact this has on the mechanism of separation and/or fractionation.

The second part in Chapter 4 elucidating the role of transmembrane pressure and the

cross flow velocity on the separation and fractionation mechanisms respectively can be
7.2. Recommendations for Future Directions 219

further improvised:

• It will be insightful to conduct similar set of experiments with other types of mem-

branes.

• In this work, model virus particles were employed for the investigation, but it will

be useful to employ other model viruses like bacteriophage to examine the effect of

virus characteristics on the filtration efficiency behavior.

• Virus surface charge characteristics can be studied to analyse for any effect on filtra-

tion efficiency.

Chapter 5 demonstrates that the novel model developed successfully predicts the parti-

cle size distribution of the output streams. But the mismatch between the model simulation

output and the experimental results indicates that other factors need to be accounted in the

modeling which includes:

• Pore size distribution of the membrane plays a role in affecting the final filtrate par-

ticle size distribution and hence inclusion of pore size distribution may have a sig-

nificant impact in arriving at exact predictions.

• Defining multi-modal distributions as well pore blockage mechanisms may improve

the model significantly.

• Virus particle aggregation and breakage are not considered in our current model,

hence it is highly recommended to incorporate these factors in the future models.


Bibliography 220

Bibliography
Abbas, A., Romagnoli, J. A., 2007. Multiscale modeling, simulation and validation of

batch cooling crystallization. Separation and Purification Technology 53 (2), 153–163.

Adham, S., Jacangelo, J., Laine, J., 1998. Low pressure membranes - assessing integrity.

Desalination 119, 73–77.

Ahmad, A., Chong, M., Bhatia, S., 2006. Ultrafiltration modeling of multiple solutes sys-

tem for continuous cross-flow process. Chemical Engineering Science 61, 5057–5069.

Alspach, B., Allgeier, S., 2003. Application of ultrafiltration for virus removal using the

lt2eswtr regulatory framework. Proceedings, AWWA Water Quality Technology Con-

ference.

Alvarez, J., Alvarez, J., Hernandez, M., 1994. A population balance approach for the de-

scription of particle size distribution in suspension polymerization reactors. Chemical

Engineering Science 49, 99–113.

Amirilargani, M., Sadrzadeh, M., Mohammadi, T., 2009. Synthesis and characterization

of polyethersulfone membranes. Journal of Polymer Research, 1–15.

Aranha-Creado, H., Fennington Jr., G. J., 1997. Cumulative viral titer reduction demon-

strated by sequential challenge of a tangential flow membrane filtration system and a

direct flow pleated filter cartridge. PDA Journal of Pharmaceutical Science and Tech-

nology 51 (5), 208–212.

Arkhangelsky, E., Steubing, B., Ben-Dov, E., Kushmaro, A., Gitis, V., 2008. Influence of
Bibliography 221

ph and ionic strength on transmission of plasmid dna through ultrafiltration membranes.

Desalination 227 (1-3), 111–119.

Arnal, J., Garcia-Fayos, B., Verdu, G., Lora, J., 2009. Ultrafiltration as an alternative mem-

brane technology to obtain safe drinking water from surface water: 10 years of experi-

ence on the scope of the aquapot project. Desalination 248, 34–41.

Ashbolt, N. J., 2004. Risk analysis of drinking water microbial contamination versus dis-

infection by-products (dbps). Toxicology 198 (1-3), 255 – 262.

Azari, M., Boose, J., Burhop, K., et al., 2000. Evaluation and validation of virus removal by

ultrafiltration during the production of diaspirin crosslinked haemoglobin. Biologicals

28, 81–94.

Bakhshayeshi, M., Zydney, A. L., 2008. Effect of solution ph on protein transmission and

membrane capacity during virus filtration. Biotechnology and bioengineering 100 (1),

108–117.

Baruah, G., Venkiteshwaran, A., Belfort, G., 2005. Global model for optimizing crossflow

microfiltration and ultrafiltration processes: A new predictive and design tool. Biotech-

nol. Prog. 21, 1013–1025.

Belfort, G., 1989. Fluid mechanics in membrane filtration: recent developments. Journal

of Membrane Science 40, 123–147.

Belfort, G., Davis, R. H., Zydney, A. L., 1994. The behavior of suspensions and macro-

molecular solutions in crossflow microfiltration. Journal of Membrane Science 96 (1-2),

1–58.
Bibliography 222

Belfort, G., Nagata, N., 1985. Fluid mechanics and cross-flow filtration: so me thoughts.

Desalination 53, 57–79.

Berger, P., Clark, R., Reasoner, D., Rice, E., Santo Domingo, J., 2009. Water, drinking,

121 – 137.

Berne, B., Pecora, R., 1976. Dynamic Light Scattering with applications to chemistry,

biology and physics. Wiley.

Biesert, L., Hoffer, L., Josic, D., Suhartono, H., Gärtner, T., Schwinn, H., 1997. Viral

safety of a new factor ix (octanine f®): Combination of solvent/detergent (sd) treatment

with an ultrafiltration step. Infusionstherapie und Transfusionsmedizin 24 (4), 202.

Bird, R., Stewart, W., Lightfoot, E., 2006. Transport Phenomena. Wiley.

Bodzek, M., Konieczny, K., 1994. Optimization of the ultrafiltration of latex wastewaters.

Desalination 94 (3), 289–306.

Bodzek, M., Konieczny, K., 1998. Comparison of various membrane types and module

configurations in the treatment of natural water by means of low-pressure membrane

methods. Separation and Purification Technology 14 (1-3), 69–78.

Bohonak, D., Zydney, A., 2005. Compaction and permeability effects with virus filtration

membranes. Journal of Membrane Science 254, 71–79.

Bolton, G., LaCasse, D., Kuriyel, R., 2006. Combined models of membrane fouling: De-

velopment and application to microfiltration and ultrafiltration of biological fluids. Jour-

nal of Membrane Science 277 (1-2), 75–84.


Bibliography 223

Boschel, D., Janich, M., Roggendorf, H., 2003. Size distribution of colloidal silica in

sodium silicate solutions investigated by dynamic light scattering and viscosity mea-

surement. Journal of Colloid and Interface Science 267, 360–368.

Bowen, W. R., Sharif, A. O., 2002. Prediction of optimum membrane design: Pore en-

trance shape and surface potential. Colloids and Surfaces A: Physicochemical and En-

gineering Aspects 201 (1-3), 207–217.

Brandenburg, B., Zhuang, X., 2007. Virus trafficking - learning from single-virus tracking.

Nature Reviews Microbiology 5 (3), 197–208.

Brandwein, H., Aranha-Creado, H., 2000. Membrane filtration for virus removal. Devel-

opments in biological standardization 102, 157–163.

Bren, L., Jan-Feb 2006. The road to the biotech revolution: Highlights of 100 years of

biologics regulation. FDA Consumer magazine.

Brough, H., Antoniou, C., Carter, J., Jakubik, J., Xu, Y., Lutz, H., 2002. Performance of a

novel viresolve nfr virus filter. Biotechnology progress 18 (4), 782–795.

Burnouf, T., Radosevich, M., 2003. Nanofiltration of plasma-derived biopharmaceutical

products. Haemophilia 9 (1), 24–37.

Caldwell, K., Nguyen, T., Giddings, J., Mazzone, H., 1980. Field-flow fractionation of

alkali-liberated nuclear polyhedrosis virus from gypsy moth lymantria dispar linnaeus.

Journal of Virological Methods 1 (5), 241–256.

Cassano, A., Marchio, M., Drioli, E., 2007. Clarification of blood orange juice by ul-
Bibliography 224

trafiltration: Analyses of operating parameters, membrane fouling and juice quality.

Desalination 212, 15–27.

Chandra, S., Groener, A., Feldman, F., 2002. Effectiveness of alternative treatments for

reducing potential viral contaminants from plasma-derived products. Thrombosis Re-

search 105 (5), 391–400.

Chang, D., Hwang, S., 1995. Unsteady-state permeate flux of crossflow microfiltration:

Effect of particle size distribution. Separation Science and Technology 30, 2917–2931.

Chang, J., Tsai, L., Vigneswaran, S., 1996. Experimental investigation of the effect of

particle size distribution of suspended particles on microfiltration. Water Science Tech-

nology 34, 133–140.

Chang, J., Vigneswaran, S., Kandaamy, J., Tsai, L., 2008. Effect of pore size and particle

size distribution on granular bed filtration and microfiltration. Separation Science and

Technology 43 (7), 1771–1784.

Chao, A., Tojo, S., 1987. Permeate quality of ultrafiltration process. Journal of Environ-

mental Engineering 113 (2), 383–394.

Charcosset, C., 2006. Membrane process in pharmaceutical and biotechnological applica-

tions. ITBM-RBM 27 (1), 1–7.

Chellam, S., Wiesner, M. R., 1992. Particle transport in clean membrane filters in laminar

flow. Environmental Science and Technology 26(8), 1611–1621.

Choi, Y. W., Kim, I. S., 2008. Viral clearance during the manufacture of urokinase from

human urine. Biotechnology and Bioprocess Engineering 13 (1).


Bibliography 225

Chuan, Y. P., Fan, Y. Y., Lua, L., Middelberg, A. P. J., 2008. Quantitative analysis of

virus-like particle size and distribution by field-flow fractionation. Biotechnology and

bioengineering 99 (6), 1425–1433.

Chudacek, M. W., Fane, A. G., 1984. The dynamics of polarisation in unstirred and stirred

ultrafiltration. Journal of Membrane Science 21 (2), 145–160.

Citkowicz, A., Petry, H., Harkins, R. N., Ast, O., Cashion, L., Goldmann, C., Bringmann,

P., Plummer, K., Larsen, B. R., 2008. Characterization of virus-like particle assembly

for dna delivery using asymmetrical flow field-flow fractionation and light scattering.

Analytical Biochemistry 376 (2), 163–172.

Clever, M., Jordt, F., Knauf, R., Raebiger, N., Ruedebusch, M., Hilker-Scheibel, R., 2000.

Process water production from river water by ultrafiltration and reverse osmosis. Desali-

nation 131 (1-3), 325–336.

Cole, K. D., III, L. F. P., Tsai, D.-H., Singh, T., Lute, S., Brorson, K. A., Wang, L.,

2009. Particle concentration measurement of virus samples using electrospray differen-

tial mobility analysis and quantitative amino acid analysis. Journal of Chromatography

A 1216 (30), 5715 – 5722.

Damak, K., Ayadi, A., Zeghmati, B., Schmitz, P., 2004. A new navier-stokes and darcy’s

law combined model for fluid flow in crossflow filtration tubular membranes. Desalina-

tion 161 (1), 67 – 77.

David, M., William, O., 2001. Method using filtration aids for the separation of virus

vectors from nucleic acids and other cellular contaminants. WO/2001/048155.


Bibliography 226

Davis, K., 2008. Biopharmaceutical manufacturing: The challenge of global regulatory

compliance. Pharmaceutical Technology 32 (6), 60–68.

Degen, P., Sipsas, I., Rapisarda, G., Gregg, J., 1998a. Filtration medium. US Patents

5788862.

Degen, P., Sipsas, I., Rapisarda, G., Gregg, J., 1998b. Polyvinylidene fluoride membrane

and method for removing viruses from solutions. US Patents 5736051.

Denisov, G., 1994. Theory of concentration polarization in cross-flow ultrafiltration: Gel-

layer model and osmotic-pressure model. Journal of Membrane Science 91 No. 1-2,

173–187.

DiLeo, A., Allegrezza, A., Burke, E., 1991. Membrane process and system for isolating

virus from solution. US Patent: 5017292.

DiLeo, A. J., Vacante, D. A., Deane, E. F., 1993. Size exclusion removal of model mam-

malian viruses using a unique membrane system, part ii: Module qualification and pro-

cess simulation. Biologicals 21 (3), 287–296.

Driscoll, K., 2001. Development of a process simulator for the ultrafiltration/diafiltration

process. PhD Thesis University of Arkansas.

Drouiche, M., Lounici, H., Belhocine, D., Grib, H., Piron, D., Mameri, N., 2001. Eco-

nomic study of the treatment of surface water by small ultrafiltration units. Water SA

27 (2), 199–204.

EMEA, 2006. European commission: Emea guideline on virus safety evaluation of

biotechnological investigational medicinal products. London.


Bibliography 227

EP, 2006. Laser diffraction measurement of particle size. European Pharmacopoeia, Chap-

ter 2.9.31 http://www.edqm.eu/, 4429.

E.P.A., 2000. Environmental protection agency proposed rules. Federal Register 65,

30193–30274.

Fiksdal, L., Leiknes, T., 2006. The effect of coagulation with mf/uf membrane filtration

for the removal of virus in drinking water. Journal of Membrane Science 279 (1-2), 364

– 371.

Fischer, E., Raasch, J., 1985. Cross-flow filtration. German chemical engineering 8 (4),

211–216.

Foley, G., Malone, D., MacLoughlin, F., 1995a. Modelling the effects of particle polydis-

persity in crossflow filtration. Journal of Membrane Science 99, 77–88.

Foley, G., Malone, D. M., MacLoughlin, F., 1995b. Modelling the effects of particle poly-

dispersity in crossffow filtration. Journal of Membrane Science 99 (1), 77–88.

Frisken, B. J., 2001. Revisiting the method of cumulants for the analysis of dynamic light

scattering data. Applied Optics 40 No.:24, 4087–4091.

Gaczynska, M., Osmulski, P. A., 2008. Afm of biological complexes: What can we learn?

Current Opinion in Colloid and Interface Science 13 (5), 351–367.

Gagne, D., Vaccaro, G., 2003. Sulaibiya water reuse project well underway. Water and

Wastewater International 18 (9), 14–18.


Bibliography 228

Garnick, R. L., 1989. Safety aspects in the quality control of recombinant products from

mammalian cell culture. Journal of Pharmaceutical and Biomedical Analysis 7 (2), 255

– 266.

Gelbard, F., Seinfeld, J., 1978. Numerical solution of the dynamic equation for particulate

processes. J. Comp. Phys. 28, 357–375.

Gerstlauer, A., Ghan, C., Zhou, H., Schreiber, M., 2006. Application of population bal-

ances in the chemical industry - current status and future needs. Chemical Engineering

Science 61, 205–217.

Ghidossi, R., Veyret, D., Moulin, P., 2006. Computational fluid dynamics applied to mem-

branes: State of the art and opportunities. Chemical Engineering and Processing 45,

437–454.

Ghosh, R., 2008. Handbook of Membrane Separations Ed. by Anil K Pabby and Syed S H

Rizvi and Ana Maria Sastre. CRC Press, London.

Giddings, J., 1993. Field flow fractionation - analysis of macromolecular, colloidal and

particulate materials. Science 260 (5113), 1456–1465.

Giddings, J., Chen, X., Wahlund, K., Myers, M., 1987. Fast particle separation by

flow/steric field-flow fractionation. Analytical Chemistry 59, 1957–1962.

Gillock, E. T., Rottinghaus, S., Chang, D., Cai, X., Smiley, S. A., An, K., Consigli, R. A.,

1997. Polyomavirus major capsid protein vp1 is capable of packaging cellular dna when

expressed in the baculovirus system. Journal of virology 71 (4), 2857–2865.


Bibliography 229

Gimbun, J., Nagy, Z. K., Rielly, C. D., 2009. Simultaneous quadrature method of moments

for the solution of population balance equations, using a differential algebraic equation

framework. Industrial and Engineering Chemistry Research 48 (16), 7798–7812.

gPROMS, 2004. General PROcess Modeling System Introductory User Guide, v2.3.1. PSE

Enterprise, UK.

gPROMS, 2009a. General PROcess Modeling System Model Builder Guide, v3.2.0. PSE

Enterprise, UK.

gPROMS, 2009b. General PROcess Modeling System Model Developer Guide, v3.2.0.

PSE Enterprise, UK.

Grabow, W. O. K., 1996. Waterborne diseases: Update on water quality assessment and

control. Water SA 22 (1), 193–202.

Grabow, W. O. K., 1997. Hepatitis viruses in water: Update on risk and control. Water SA

23 (4), 379–386.

Graveland, A., 1998. Particle and micro-organism removal in conventional and advanced

treatment technology. Water Science and Technology 37 (10), 125–134.

Green, G., Belfort, G., 1980. Fouling of ultrafiltration membranes: Lateral migration and

the particle trajectory model. Desalination 35 (1 -2-3), 129–147.

Grzenia, D. L., Carlson, J. O., Wickramasinghe, S. R., 2008. Tangential flow filtration for

virus purification. Journal of Membrane Science 321 (2), 373–380.


Bibliography 230

Grzenia, D. L., Wickramasinghe, S. R., Carlson, J. O., 2007. Ultrafiltration of parvovirus.

Separation Science and Technology 42 (11), 2387–2403.

Guo, Y., Cheng, A., Wang, M., Zhou, Y., 2009. Purification of anatid herpesvirus 1 par-

ticles by tangential-flow ultrafiltration and sucrose gradient ultracentrifugation. Journal

of Virological Methods 161 (1), 1 – 6.

Guthmann, J., Klovstad, H., Boccia, D., Hamid, N., Pinoges, L., Nizou, J., Tatay, M.,

Diaz, F., Moren, A., Grais, R. F., Ciglenecki, I., Nicand, E., Guerin, P. J., 2006. A large

outbreak of hepatitis e among a displaced population in darfur, sudan, 2004: The role of

water treatment methods. Clinical Infectious Diseases 42 (12), 1685–1691.

Han, B., Akeprathumchai, S., Wickramasinghe, S., 2003. Flocculation of biological cells:

Experiment vs theory. AIChE Journal 49, 1687–1701.

Handlin, J., Trenor, S., Dado, G., 2009. Method for making sulfonated block copolymers,

methods for making membranes from such block copolymers and membrane structures.

US Patents: 2009/0280255.

Hanft, S., 2010. Ultrafiltration membranes: Technologies and the u.s. market. BCC Report

ID: MST044C.

Harris, A., Cardone, G., Winkler, D. C., Heymann, J. B., Brecher, M., White, J. M.,

Steven, A. C., 2006. Influenza virus pleiomorphy characterized by cryoelectron tomog-

raphy. Proceedings of the National Academy of Sciences of the United States of America

103 (50), 19123–19127.


Bibliography 231

Henley, M., 2009. Biocontrol markets show continued gowth. Ultrapure Water 26 (3), 10–

13.

Hensgen, M., Czermak, P., Carlson, J., Wickramasinghe, S., 2010. Purification of minute

virus of mice using high performance tangential flow filtration. Desalination 250 (3),

1121 – 1124.

Hesse, F., Wagner, R., 2000. Developments and improvements in the manufacturing of

human therapeutics with mammalian cell cultures. Trends in Biotechnology 18 (4), 173

– 180.

Hirasaki, T., Yamaguchi, K., Yanagida, K., Okuyama, K., 2006. Removal of small viruses

(parvovirus) from igg solution by virus removal filter planova20n. Journal of Membrane

Science 278 (1-2), 3 – 9.

Hirasaki, T., Yokogi, M., Kono, A., Yamamoto, N., Manabe, S., 2002. Removal and deter-

mination of dispersion state of bacteriophage [phi]x174 in aqueous solution by cupram-

monium regenerated cellulose microporous hollow fiber membrane (bmm®). Journal of

Membrane Science 201 (1-2), 95 – 102.

Ho, C., Zydney, A., 2003. Effect of membrane morphology on system capacity during

normal flow microfiltration. Biotech. Bioeng. 83 No.5, 537–543.

Hogan Jr., C. J., Kettleson, E. M., Lee, M., Ramaswami, B., Angenent, L. T., Biswas, P.,

2005. Sampling methodologies and dosage assessment techniques for submicrometre

and ultrafine virus aerosol particles. Journal of applied microbiology 99 (6), 1422–1434.
Bibliography 232

Hogan Jr., C. J., Kettleson, E. M., Ramaswami, B., Chen, D., Biswas, P., 2006. Charge

reduced electrospray size spectrometry of mega- and gigadalton complexes: Whole

viruses and virus fragments. Analytical Chemistry 78 (3), 844–852.

Huang, L., Morrissey, M., 1999. Finite element analysis as a tool for crossflow membrane

filter simulation. J. Membr. Sci. 155, 19–30.

Huang, P., Peterson, J., 2001. Scaleup and virus clearance studies on virus filtration in

monoclonal antibody manufacture. Membrane separations in biotechnology, 327– 350.

Hulburt, H., Katz, S., 1964. Some problems in particle technology. Chemical Engineering

Science 19, 555–574.

Hwang, K., Liu, H., Lu, W., 1998. Local properties of cake in cross-flow microfiltration of

submicron particles. Journal of Membrane Science 138 (2), 181–192.

ICH, December 2000. Ich topic q6a specifications: Test procedures and acceptance criteria

for new drug substances and new drug products. Chemical Substances.

Jiao, D., Sharma, M. M., 1994. Mechanism of cake buildup in crossflow filtration of col-

loidal suspensions. Journal of colloid and interface science 162 (2), 454–462.

John, K., Ann, L., Aaron, G., Brian, C., 2003. Methods of adenovirus purification.

WO/2003/097797.

Johnson, J., Chiu, W., 2000. Structures of virus and virus-like particles. Current Opinion

in Structural Biology 10 (2), 229 – 235.


Bibliography 233

Kim, A. S., 2007. Permeate flux inflection due to concentration polarization in crossflow

membrane filtration: A novel analytic approach. European Physical Journal E 24 (4),

331–341.

Kitis, M., Lozier, J. C., Kim, J. H., Mi, B., Marinas, B. J., 2003. Evaluation of biologic

and non-biologic methods for assessing virus removal by and integrity of high pressure

membrane systems. Vol. 3.

Kleinstreuer, C., Belfort, G., 1984. Mathematical modeling of fluid flow and solute dis-

tribution in pressure driven membrane modules. Synth Membr Processes, Fundam and

Water Appl, 131–190.

Knyaz’kova, T. V., 1995. Immobilization of polymers on cellulose acetate membranes.

Catalysis Today 25 (3-4), 255 – 262.

Kolin, D. L., Wiseman, P. W., 2007. Advances in image correlation spectroscopy: Measur-

ing number densities, aggregation states, and dynamics of fluorescently labeled macro-

molecules in cells. Cell biochemistry and biophysics 49 (3), 141–164.

Korath, J. M., Abbas, A., Romagnoli, J. A., 2008. A clustering approach for the separa-

tion of touching edges in particle images. Particle and Particle Systems Characterization

25 (2), 142–153.

Kumar, S., Ramakrishna, D., 1996. On the solution of population balance equations by

discretization -ii: a moving pivot technique. Chem. Eng. Sci. 51, 1333–1342.

Kuo, J., 2007. Electron microscopy methods and protocols in molecular biology. Humana

Press, NewJersey.
Bibliography 234

Laine, J., Vial, D., Moulart, P., 2000. Status after 10 years of operation - overview of uf

technology today. Desalination 131, 17–25.

Leahy, T. J., Sullivan, M. J., 1978. Validation of bacterial retention capabilities of mem-

brane filters. Pharm.Technol. 2 (11), 65–75.

Lee, S. H., Levy, D. A., Craun, G. F., Beach, M. J., Calderon, R. L., 2002. Surveillance

for waterborne-disease outbreaks—united states 1999-2000. Morb. Mortal. Wkly. Rep.

Surveill. Summ. 51, 1–47.

Lee, Y., Clark, M., 1998. Modeling of flux decline during crossflow ultrafiltration of col-

loidal suspensions. Journal of Membrane Science 149 No. 2, 181–202.

Lee, Y., Clark, M. M., Kim, B. J., 1996. A numerical model for flux decline during cross-

flow uf. ACS Division of Environmental Chemistry, Preprints 36 (2), 122–123.

Lenes, D., Deboosere, N., Menard-Szczebara, F., Jossent, J., Alexandre, V., Machinal, C.,

Vialette, M., 2010. Assessment of the removal and inactivation of influenza viruses h5n1

and h1n1 by drinking water treatment. Water Research 44 (8), 2473 – 2486.

Lengyel, J. S., Milne, J. L. S., Subramaniam, S., 2008. Electron tomography in nanoparti-

cle imaging and analysis. Nanomedicine 3 (1), 125–131.

Li, N., Fane, A., Winston, W., Matsuura, T., 2008. Advanced Membrane Technology and

Applications. John Wiley and Sons, New Jersey.

Lightfoot, E., Moscariello, J., 2004. Bioseparations. Biotechnology and BioEngineering

87 No.3, 259–273.
Bibliography 235

Lipin, D. I., Lua, L. H. L., Middelberg, A., 2008. Quaternary size distribution of soluble

aggregates of glutathione-s-transferase-purified viral protein as determined by asymmet-

rical flow field flow fractionation and dynamic light scattering. Journal of Chromatogra-

phy A 1190 (1-2), 204–214.

Lu, W.-M., Hwang, K.-J., 1995. Cake formation in 2-d cross-flow filtration. AICHE Jour-

nal 41 (6), 1443–1455.

Lu, W.-M., Ju, S.-C., 1989. Selective particle deposition in crossflow filtration. Separation

Science and Technology 24 (7-8), 517–540.

Lute, S., Riordan, W., Pease III, L. F., Tsai, D. ., Levy, R., Haque, M., Martin, J., Moroe,

I., Sato, T., Morgan, M., Krishnan, M., Campbell, J., Genest, P., Dolan, S., Tarrach, K.,

Meyer, A., Aranha, H., Bailey, M., Bender, J., Carter, J., Chen, Q., Dowd, C., Jani, R.,

Jen, D., Kidd, S., Meltzer, T., Remington, K., Rice, I., Romero, C., Sato, T., Jornitz,

M., Sekura, C. M., Sofer, G., Specht, R., Tarrach, K., Wojciechowski, P., Zachariah,

M. R., Tarlov, M. J., Etzel, M., Brorson, K., 2008. A consensus rating method for small

virus-retentive filters. i. method development. PDA Journal of Pharmaceutical Science

and Technology 62 (5), 318–333.

Mackley, M., Sherman, N., 1992a. Cross-flow cake filtration mechanisms and kinetics.

Chemical Engineering Science 47 (12), 3067 – 3084.

Mackley, M. R., Sherman, N. E., 1992b. Cross-flow cake filtration mechanisms and kinet-

ics. Chemical Engineering Science 47 (12), 3067–3084.


Bibliography 236

Madaeni, S., Fane, A., Grohmann, G., 1995. Virus removal from water and wastewater

using membranes. Journal of Membrane Science 102(1-3), 65–75.

Manabe, S., 1996. Removal of virus through novel membrane filtration method. Develop-

ments in biological standardization 88, 81–90.

Marchisio, D. L., Pikturna, J. T., Fox, R. O., Vigil, R. D., Barresi, A. A., 2003. Quadrature

method of moments for population-balance equations. AICHE Journal 49 (5), 1266–

1276.

Martin, S., Hong, A., Jason, C., Dana, K., 2006. Process for removing protein aggregates

and virus from a protein solution. US Patents: 7118675.

Masayuki, H., Shin-ichi, M., 2009. Fluoresin polymer separation membrane and process

for producing the same. EP: 2047898.

Matthew, C., 2003. Viral purification methods. WO03/093463.

Matthew, C., 2007. Improved viral purification methods. US Patent: 7223585.

McCoy, B., 1995. Membrane sieving of a continuous polydisperse mixture through dis-

tributed pores. Separation Science and Technology 30 N. 4, 487–507.

Meltzer, T., Jornitz, M., 2007. Filtration and Purification in the Biopharmaceutical Indus-

try. CRC Press.

Merten, O., 2002. Virus contaminations of cell cultures - a biotechnological view. Cy-

totechnology 39 (2), 91–116.


Bibliography 237

Michalsky, R., Passarelli, A., Pfromm, P., Czermak, P., 2009. Purification of the bac-

ulovirus autographa californica m nucleopolyhedrovirus by tangential flow ultrafiltra-

tion. Desalination 245 (1-3), 694 – 700.

Miranda, W., 2009. Virus purification using ultrafiltration. US Patents: 20090123989.

Mohammadi, T., Kohpeyma, A., Sadrzadeh, M., 2005. Mathematical modeling of flux

decline in ultrafiltration. Desalination 184, 367–375.

Murthy, Z., Chaudhari, L., 2008. Application of nanofiltration for the rejection of nickel

ions from aqueous solutions and estimation of membrane transport parameters. Journal

of hazardous materials 160 (1), 70–77.

Nassehi, V., 1998. Modeling of combined navier-stokes and darcy flows in crossflow mem-

brane filtration. Chemical Engineering Science 53, 1253–1265.

Nowak, T., Bernhardt, G., 2002. Removal of viruses from protein solutions by ultrafiltra-

tion. US Patents: 6391657.

Ogle, K. F., Azari, M. R., 2001. Virus removal by ultrafiltration: A case study with di-

aspirin crosslinked hemoglobin (dclhb). Membrane Separations in Biotechnology, 299–

326.

Oh, H. K., Takizawa, S., Ohgaki, S., Katayama, H., Oguma, K., Yu, M., 2007. Removal of

organics and viruses using hybrid ceramic mf system without draining pac. Desalination

202 (1-3), 191–198.

Otaki, M., Yaho, K., Ohgaki, S., 1998. Virus removal in a membrane separation process.

Water Science Technology 37, 107–116.


Bibliography 238

Pak, A., Mohammadi, T., Hosseinalipour, S., Allahdini, V., 2008. Cfd modeling of porous

membranes. Desalination 222, 482–488.

Pall, D. P., Kimbauer, E. A., 1978. Fifty Second Colloid and Surface Science Symposium

of Tennessee.

Patty, P., Frisken, B., 2006. Direct determination of the number weighted mean radius and

polydispersity from dynamic light scattering data. Applied Optics 45 no.10, 2209–2216.

Pearce, G., 2007. Water and wastewater filtration: Membrane module format. Filtration

and Separation 44 (4), 31 – 33.

Pease III, L., Lipin, D., Tsai, D.-., Zachariah, M., Lua, L., Tarlov, M., Middelberg, A.,

2009. Quantitative characterization of virus-like particles by asymmetrical flow field

flow fractionation, electrospray differential mobility analysis, and transmission electron

microscopy. Biotechnology and bioengineering 102 (3), 845–855.

Pelchen-Matthews, A., Marsh, M., 2007. Electron Microscopy Analysis of Viral Morpho-

genesis. Vol. 2007.

Porter, M., 1990. Handbook of Industrial Membrane Technology. Noyes.

Powell, M., Timperman, A., 2005. Quantitative analysis of protein recovery from dilute,

large volume samples by tangential flow ultrafiltration. Journal of Membrane Science

252 (1-2), 227 – 236.

Pujar, N. S., Zydney, A. L., 1994. Electrostatic and electrokinetic interactions during pro-

tein transport through narrow pore membranes. Industrial and Engineering Chemistry

Research 33 (10), 2473–2482.


Bibliography 239

Radosevich, M., Appourchaux, P., Huart, J., Burnouf, T., 1994. Nanofiltration, a new spe-

cific virus elimination method applied to high-purity factor ix and factor xi concentrates.

Vox sanguinis 67 (2), 132–138.

Ramachandran, V., Fogler, H. S., 1999. Plugging by hydrodynamic bridging during flow

of stable colloidal particles within cylindrical pores. Journal of Fluid Mechanics 385,

129–156.

Ramakrishna, D., 2000. Population Balances: Theory and applications to particulate sys-

temsin Engineering. Academic Press.

Randolph, A., 1964. Population balance for countable entities. Canadian Journal of Chem-

ical Engineering 42(6), 280.

Randolph, A., Larson, M., 1988. Theory of Particulate Processes: Analysis and Techniques

of Continuous Crystallization. Academic Press.

Randolph, A., White, E., 1977. Modeling size dispersion in the prediction of crystal size

distribution. Chemical Engineering Science 32, 1067–1076.

Ratanathanawongs, S., Giddings, J., 1993. Particle size analysis using flow field flow frac-

tionation. ACS Symposium Series 521, 13–29.

Rautenbach, R., Schock, G., 1988. Ultrafiltration of macromolecular solutions and cross-

flow microfiltration of colloidal suspensions. a contribution to permeate flux calcula-

tions. Journal of Membrane Science 36 (C), 231–242.

Rautenbach, R., Vobenkaul, K., 2001. Pressure driven membrane processes-the answer
Bibliography 240

to the need of a growing world population for quality water supply and waste water

disposal. Separation and Purification Technology 22-23, 193–208.

Rawlings, J., Witkowski, W., Eaton, J., 1992. Modeling and control of crystallisers. Pow-

der Technology 69, 3–9.

Ray, S., Tarrach, K., 2008. Virus clearance strategy using a three-tier orthogonal technol-

ogy platform. BioPharm International 21 (9), 50–59.

Reti, A., 1977. An assessment of test criteria for evaluating the performance and integrity

of sterilizing filters. Bulletin of the Parenteral Drug Association 31 (4), 187–194.

Richard Bowen, W., Williams, P. M., 2001. Prediction of the rate of cross-flow ultrafiltra-

tion of colloids with concentration-dependent diffusion coefficient and viscosity - theory

and experiment. Chemical Engineering Science 56 (10), 3083–3099.

Robert, F., Deborah, F., Benedicte, L., van Reis, R., 2008. Non-affinity purification of

proteins. US Patents:7323553.

Roberts, P., 1997. Efficient removal of viruses by a novel polyvinylidene fluoride mem-

brane filter. Journal of virological methods 65 (1), 27–31.

Roda, B., Zattoni, A., Reschiglian, P., Moon, M., Mirasoli, M., Michelini, E., Roda,

A., 2009. Field-flow fractionation in bioanalysis: A review of recent trends. Analytica

Chimica ACTA 635 (2), 132–143.

Romero, C., Davis, R. H., 1990. Transient model of crossflow microfiltration. Chemical

Engineering Science 45 No.1, 13–25.


Bibliography 241

Roorda, J. H., te Poele, S., van der Graaf, J. H. J. M., 2004. The role of microparticles in

dead-end ultrafiltration of wwtp-effluent. Water science and technology 50 (12), 87–94.

Rosner, D. E., McGraw, R., Tandon, P., 2003. Multivariate population balances via mo-

ment and monte carlo simulation methods: An important sol reaction engineering bi-

variate example and "mixed" moments for the estimation of deposition, scavenging, and

optical properties for populations of nonspherical suspended particles. Industrial and

Engineering Chemistry Research 42 (12), 2699–2711.

Russell, E., Wang, A., Rathore, A. S., 2007. Process Scale Bioseparations for the Biophar-

maceutical Industry Ed. by Abhinav A. Shukla and Mark R. Etzel and Shishir Gadam.

CRC Press.

Saksena, S., Zydney, A. L., 1994. Effect of solution ph and ionic strength on the separa-

tion of albumin from immunoglobulins (igg) by selective filtration. Biotechnology and

bioengineering 43 (10), 960–968.

Sano, D., Ueki, Y., Watanabe, T., Omura, T., 2006. Membrane separation of indigenous

noroviruses from sewage sludge and treated wastewater. Water Science and Technology

54 (3), 77–82.

Santos, A., Bedrikovetsky, P., 2004. Size exclusion during particle suspension transport in

porous media: stochastic and averaged equations. Computational and Applied Mathe-

matics 23 N. 2-3, 259–284.

Sarkar, B., DasGupta, S., De, S., 2009. Flux decline during electric field-assisted cross-
Bibliography 242

flow ultrafiltration of mosambi (citrus sinensis (l.) osbeck) juice. Journal of Membrane

Science 331 (1-2), 75–83.

Saxena, A., Tripathi, B. P., Kumar, M., Shahi, V. K., 2009. Membrane-based techniques

for the separation and purification of proteins: An overview. Advances in Colloid and

Interface Science 145 (1-2), 1 – 22.

Schimpf, M., Caldwell, K., Giddings, J., 2000. Field flow fractionation handbook. John

Wiley and Sons, NewYork.

Schwinge, J., Wiley, D., Fane, A., 2004. Novel spacer design improves observed flux.

Journal of Membrane Science 229, 53.

Scutt, J., 1971. Virus retention by membrane filters. Water Research 5, 183–185.

Shortt, D. W., Roessner, D., Wyatt, P. J., 1996. Absolute measurement of diameter distri-

butions of particles using a multi-angle light scattering photometer coupled with flow

field-flow fractionation 21.

Soluk, L., Price, H., Sinclair, C., Atalla-Mikhail, D., Genereux, M., 2008. Pathogen safety

of intravenous rh immunoglobulin liquid and other immune globulin products: En-

hanced nanofiltration and manufacturing process overview. American Journal of Thera-

peutics 15 (5), 435–443.

Song, L., 1998. Flux decline in crossflow microfiltration and ultrafiltration: mechanisms

and membrane fouling. Journal of Membrane Science 139, 183–200.

Song, L., Elimelech, M., 1995. Particle deposition onto a permeable surface in laminar

flow. Journal of Colloid and Interface Science 173 (1), 165–180.


Bibliography 243

Stefan, W., 1996. Filtration. WO/1996/000237.

Stoller, M., 2009. On the effect of flocculation as pretreatment process and particle size

distribution for membrane fouling reduction. Desalination 240(1-3), 209–217.

Su, J., Gu, Z., Li, Y., Feng, S., Xu, X. Y., 2008. An adaptive direct quadrature method of

moment for population balance equations. AICHE Journal 54 (11), 2872–2887.

Sun, W., Chen, T., Chen, C., Li, J., 2007. A study on membrane morphology by digital

image processing. Journal of Membrane Science 305 (1-2), 93–102.

Tan, H., Goldschmidt, M., Boerefijn, R., Hounslow, M., Salman, D., Kuipers, J., 2005.

Population balance modeling of fluidized bed melt granulation: An overview. Chemical

Engineering Research and Design 83, 871–880.

Tarabara, V., Hovinga, R., Wiesner, M., 2002. Constant transmembrane pressure vs con-

stant permeate flux: Effect of particle size on crossflow membrane filtration. Environ-

mental Engineering Science 19, 343–355.

Terrab, A., Pawlak, D., 2007. Validation of a prion and virus purification process in the

manufacture of bovine thrombin. Journal of Applied Research 7 (3), 217–226.

Teunis, P. F. M., Medema, G. J., Kruidenier, L., Havelaar, A. H., 1997. Assessment of the

risk of infection by cryptosporidium or giardia in drinking water from a surface water

source. Water research 31 (6), 1333–1346.

Thomas, J. J., Bothner, B., Traina, J., Benner, W. H., Siuzdak, G., 2004. Electrospray ion

mobility spectrometry of intact viruses. Spectroscopy 18 (1), 31–36.


Bibliography 244

Tkacik, G., 2007. Virus removal devices. EP:1218093.

Tkacik, G., Goddard, P., Kools, W., Nitin, S., 2010. Multilayer ultrafiltration composite

membranes and methods of making. EP:2163296.

Tkacik, G., Kazan, G., 2006. High resolution virus removal methodology and filtration

capsule. US Patents: 7108791.

Torras, C., Ferrando, F., Paltakari, J., Garcia-Valls, R., 2006. Performance, morphology

and tensile characterization of activated carbon composite membranes for the synthesis

of enzyme membrane reactors. Journal of Membrane Science 282 (1-2), 149–161.

Trijzelaar, B., 1993. Regulatory affairs and biotechnology in europe: Introduction into

good regulatory practice - validation of virus removal and inactivation. Biotherapy 6 (2),

93–102.

Tsoka, S., Holwill, I., Hoare, M., 1999. Virus-like particle analysis in yeast homogenate

using a laser light- scattering assay. Biotechnology and bioengineering 63 (3), 290–297.

Tucelli, R., McGrath, P., 1996. Cellulosic ultrafiltration membrane. US Patents: 5522991.

Tullis, R., 2007. Method for removal of viruses from blood by lectin affinity hemodialysis.

US Patents: 7226429.

Urase, T., Yamamoto, K., Ohgaki, S., 1998. Effect of pore structure of membranes and

module configuration on virus retention. Journal of Membrane Science 115(1) (10), 21–

29.
Bibliography 245

USP, 2008. Light diffraction measurements of particle size: General chapter. US Pharma-

copoeia.

Vacura, K., 2008. Filtration industry overview: World pharma and biotechnology market

2008. Filtration and Separation 45 (8), 22–24.

van de Ven, W., Punt, I., Kemperman, A., Wessling, M., 2009. Unraveling ultrafiltration of

polysaccharides with flow field flow fractionation. Journal of Membrane Science 338 (1-

2), 67 – 74.

van Reis, R., 1996. Tangential flow filtration process and apparatus. US Patents 5490937.

van Reis, R., 2000. Tangential flow filtration system. US Patents 6054051.

van Reis, R., Brake, J., Charkoudian, J., Burns, D., Zydney, A., 1999. High performance

tangential flow filtration using charged membranes. Journal of Membrane Science 159,

133.

van Reis, R., Gadam, S., Frautschy, L. N., Orlando, S., Goodrich, E. M., Saksena, S.,

Kuriyel, R., Simpson, C. M., Pearl, S., Zydney, A. L., 1997. High performance tangen-

tial flow filtration. Biotechnology and bioengineering 56 (1), 71–82.

van Reis, R., Zydney, A., 2007. Bioprocess membrane technology. Journal of Membrane

Science 297 (1-2), 16–50.

Vyas, H., Bennett, R., Marshall, A., 2000. Influence of feed properties on membrane foul-

ing in crossflow microfiltration of particulate suspensions. International Dairy Journal

10, 855–861.
Bibliography 246

Wakeman, R., 1994. Visualisation of cake formation in crossflow microfiltration. Transac-

tions in Industrial Chemical Engineering 72, 530–540.

Wakeman, R. J., Akay, G., 1994. Flux decay and rejection during micro- and ultra-filtration

of hydrophobically modified water-soluble polymers. Journal of Membrane Science

91 (1-2), 145–152.

Walter, K., Noah, M., Scott, S., Jason, N., Jennifer, B., 2010. Water flux polymer mem-

branes. WO/2010/051150.

Wei, Z., Mcevoy, M., Razinkov, V., Polozova, A., Li, E., Casas-Finet, J., Tous, G. I.,

Balu, P., Pan, A. A., Mehta, H., Schenerman, M. A., 2007. Biophysical characteriza-

tion of influenza virus subpopulations using field flow fractionation and multiangle light

scattering: Correlation of particle counts, size distribution and infectivity. Journal of

virological methods 144 (1-2), 122–132.

WHO, 2008. Guidelines for drinking-water quality. World Health Organization: Geneva.

Wickramasinghe, S., B, K., Zimmermann, A., Thom, V., Reichl, U., 2005. Tangential

flow microfiltration and ultrafiltration for human influenza a virus concentration and

purification. Biotechnology and BioEngineering 92 No.2, 199–208.

Wijmans, J., Nakao, S., Smolders, C., 1984. Flux limitation in ultrafiltration: Osmotic

pressure model and gel layer model. Journal of Membrane Science 20, 115–124.

Wilson, M., Mikhail, K., 2007. Ultrafiltration membranes rendered hydrophilic by hy-

droalkyl cellulose and method of making and use of such membranes. EP: 1775016A1.
Bibliography 247

Winzeler, H. B., Belfort, G., 1993. Enhanced performance for pressure-driven membrane

processes: The argument for fluid instabilities. Journal of Membrane Science 80, 35–48.

Zaamouche, R., Beicha, A., Sulaiman, N. M., 2009. Cross-flow ultrafiltration model based

on concentration polarization. Journal of Chemical Engineering of Japan 42 (2), 107–

110.

Zhang, M., Song, L., 2000. Mechanisms and parameters affecting flux decline in cross-flow

microfiltration and ultrafiltration of colloids. Environmental Science and Technology

34 (17), 3767–3773.

Zhou, J. X., Solamo, F., Hong, T., Shearer, M., Tressel, T., 2008. Viral clearance using dis-

posable systems in monoclonal antibody commercial downstream processing. Biotech-

nology and bioengineering 100 (3), 488–496.

Zydney, A., van Reis, R., 2001. Membrane separations in biotechnology. Current Opinion

in Biotechnology 12, 208–211.

You might also like