Download as pdf or txt
Download as pdf or txt
You are on page 1of 499

Particulate Fluids Processing Centre

Department of Chemical and Biomolecular Engineering


The University of Melbourne

Solid-Liquid Separation in the Water


and Wastewater Industries

by

Anthony D. Stickland
BE (Hons) BSc (Hons)
June 2005

Submitted in total fulfilment of the requirements of the degree of Doctor of


Philosophy

Supervisor: Prof. Peter J. Scales

Printed on acid-free paper


Abstract

Water and wastewater treatment produce sludge that is dewatered prior to


disposal in order to reduce disposal costs. This work combined mathematical
modelling of thickeners, filters and centrifuges with accurate measurement of sludge
dewatering properties to give improvements in operation, design and control of such
devices, including increased throughput and reduced water content.

The theoretical basis of this work was the phenomenological solid-liquid


separation theory of Buscall and White, 1987, in which the local volume fraction
dependent properties of compressive yield stress, Py(φ), and hindered settling
function, R(φ) (which, when combined, give the solids diffusivity, D(φ)), describe the
material behaviour in compression. The local conservation equations were given in
vector notation, converted to one-dimension, and applied to piston-driven filtration.
A range of models of plate-and-frame filter presses and solid-bowl centrifuges were
developed. Simple visual basic programs were written to provide operators and
designers access to filter model predictions.

Filtration and settling tests were used to measure the material characteristics
for water and wastewater treatment sludges from a range of industrial sites, including
different coagulants and coagulation conditions, with weekly and seasonal variations.
The results showed that ferric-based water treatment sludges have very little
variability, while alum sludges can exhibit some changes in properties. New
qualitative and quantitative methods were developed for wastewater sludges, which
exhibit atypical filtration behaviour. The production of synthetic sewage sludge
showed that this is caused by the extracellular polymeric constituent.

The material characteristics were used to predict the throughput and cake
solids for filter presses, centrifuges and continuous thickeners. The filter models were
validated at a range of water treatment plants, and used to investigate the optimisation
and control of press performance. A modified-Darcy’s law, which was validated

i
using model predictions, was used to determine the membrane resistance from on-site
measurements.

The centrifuge models included one-dimensional transient and continuous, and


two-dimensional continuous decanting models. Algorithms for the solution of the
one-dimensional models were developed for both the un-networked and networked
cases, and the results for a simple example illustrated. The development of the
decanting centrifuge model included conversion of the conservation equations to
helical coordinates, consideration of the viscous flow of the centrate down the spiral
formed by the blades of the scroll, the movement of the solids due to differential
speed of the scroll and the sedimentation-consolidation of the solids due to the
centrifugal acceleration.

Civil engineering consolidation theory and testing methods were outlined and
compared to filtration. The relevant dewatering parameters were related via a simple
equation. The material characteristics of a kaolin sample were measured using
oedometer and pressure filtration testing and compared using the derived relationship.

Overall, this represents several important developments in the understanding


of water and wastewater sludge dewatering, especially the measurement of water
treatment sludge characteristics at a range of sites and with seasonal variations, the
development of analysis techniques to determine wastewater sludge characteristics,
and the development or adaptation of new or existing models of filters, centrifuges
and thickeners. Considerable cost savings were made possible by understanding both
the material behaviour and the dewatering device.

ii
Declaration

This is to certify that:

- The thesis comprises only my original work towards the PhD except where
indicated in the Preface;

- Due acknowledgement has been made in the text to all other material used;
and

- The thesis is less than 100,000 words in length, exclusive of tables and
bibliographies.

Anthony Stickland

June 2005

iii
Preface

Various work presented in this thesis was carried out in collaboration with
others and due reference is made within the text. This included some filtration testing
of materials by Peter Harbour, Nevil Anderson and Colin Soh of CSIRO, Peter Hillis
and Rene Frost of United Utilities, Martin Tillotson and Anna Taylor of Yorkshire
Water, and Casey Burgess, Nicholas Cotterill and Timothy Birks of the University of
Melbourne. Peter Harbour also contributed to the comparison of single-pressure
sewage sludge filtration testing results and to the development of synthetic sewage
sludges. The Visual Basic calculators were developed in collaboration with Adam
Kilcullen of the University of Melbourne. The model formulations of batch and
decanting centrifugation were developed in collaboration with Lee White of Carnegie
Melon University, Pittsburgh.

Preliminary work on synthetic sewage sludges was carried out in collaboration


with Igor Posenjak and Angela Karvouris as part of an undergraduate research project
at the University of Melbourne. The initial results were submitted as part of the
qualification of Bachelor of Engineering, prior to PhD candidature enrolment.

iv
Publications

The following papers and reports were published during this work:

Anderson, N. J., Abd Aziz, A. A., Dixon, D. R., Harbour, P. J., Scales, P. J.
and Stickland, A. D., “Sludge project: Progress report to United Utilities and
Yorkshire Water, February 2001”, CSIRO Molecular Science and the University of
Melbourne, Technical Report (2001)

Anderson, N. J., Abd Aziz, A. A., Dixon, D. R., Harbour, P. J., Scales, P. J.
and Stickland, A. D., “Sludge project: Progress report to United Utilities and
Yorkshire Water, June 2001”, CSIRO Molecular Science and the University of
Melbourne, Technical Report (2001)

Abd Aziz, A. A., Dixon, D. R., Harbour, P. J., Scales, P. J. and Stickland, A.
D., “Sludge project: Progress report to United Utilities and Yorkshire Water, February
2002”, The University of Melbourne, Technical Report (2002)

Abd Aziz, A. A., Dixon, D. R., Harbour, P. J., Scales, P. J. and Stickland, A.
D., “Sludge project: Annual progress report to United Utilities and Yorkshire Water,
June 2002”, The University of Melbourne, Technical Report (2002)

Abd Aziz, A. A., Dixon, D. R., Harbour, P. J., Scales, P. J. and Stickland, A.
D., “Sludge project: Annual progress report to United Utilities and Yorkshire Water,
December 2002”, The University of Melbourne, Technical Report (2002)

Bartlett, J., Dixon, D. R., McCausland, C., Scales, P. J., Stickland, A. D.,
Usher, S. P. and Verrelli, D. I., “Sludge project: Progress report to United Utilities and
Yorkshire Water, June 2003”, The University of Melbourne, Technical Report (2003)

De Kretser, R. G., Dixon, D. R., McCausland, C., Fisher, D. T., Kilcullen, A.


R., Scales, P. J., Stickland, A. D., Studer, L. J., Usher, S. P. and Verrelli, D. I.,

v
“Sludge project: Annual report to Yorkshire Water and United Utilities, February
2005”, The University of Melbourne, Technical Report (2005)

De Kretser, R. G., Dixon, D. R., McCausland, C., Scales, P. J., Stickland, A.


D., Usher, S. P. and Verrelli, D. I., “Sludge project: Annual report to Yorkshire Water
and United Utilities, February 2004”, The University of Melbourne, Technical Report
(2004)

De Kretser, R. G., Dixon, D. R., McCausland, C., Scales, P. J., Stickland, A.


D., Studer, L. J., Usher, S. P. and Verrelli, D. I., “Sludge project: Progress report to
Yorkshire Water and United Utilities, July 2004”, The University of Melbourne,
Technical Report (2004)

Harbour, P. J., Anderson, N. J., Abd-Aziz, A. A., Dixon, D. R., Hillis, P.,
Scales, P. J., Stickland, A. D. and Tillotson, M., “Fundamental dewatering
characteristics of potable water treatment sludges”, J. Wat. Supply: Res. Tech. - Aqua,
53 (1), 29 (2004)

Scales, P. J., de Kretser, R. G., Stickland, A. D. and Usher, S. P., “Filtration


process modelling and optimisation using fundamental theory”, Proceedings of
Filtech Europa, Düsseldorf, October 2001, 11 (2001)

Scales, P. J., Dixon, D. R., Harbour, P. J. and Stickland, A. D., “The


fundamentals of wastewater sludge characterization and filtration”, Wat. Sci. Tech., 49
(10), 67 (2004)

Stickland, A. D., “Case study: Langsett WTP plate-and-frame presses”, The


University of Melbourne, Technical Report, June 2001

Stickland, A. D., de Kretser R. G. and Scales, P. J., “Non-traditional constant


pressure filtration behaviour”, AIChE J. (accepted 2005)

Stickland, A. D., Dixon, D. R., Harbour, P. J., Scales, P. J. and Tillotson, M.,
“Thornton Steward WTP, North Yorkshire – Sludge investigation and dewatering
options”, University of Melbourne and Yorkshire Water, Technical Report, July 2002

vi
Stickland, A. D. and Hillis, P., “Hodder sludge investigation”, The University
of Melbourne and United Utilities, Technical Report, November 2001

Stickland, A. D., Scales, P. J. and Styles, J. R.; “Comparison of geotechnical


engineering consolidation and physical science filtration testing techniques for soils
and suspensions”, Geotech. Testing J. (accepted 2005)

vii
Acknowledgements

There are numerous people to thank for contributing in some way, great and
small, to the completion of this thesis. There were plenty of Peters involved in the
previous four years whom I would like to thank. In particular, my supervisor Peter
Scales has been an inspirational mentor whose honesty, intelligence and humour
meant that this was an extremely enjoyable and stimulating time. Although he may be
a fountain of wisdom (for example, “It’s all chicken shit and bitumen”), I am still not
convinced about his navigational skills. I would like to thank Peter Hillis, the project
leader from United Utilities, for his warm hospitality and friendship, and his constant
and unconditional support for the project. Go Liverpool! Much appreciation goes to
my fellow “sludger” and strummer, Peter Harbour, for all the sessions – academic,
musical or otherwise. I would also like to thank my three familial Peters – my brother
Peter Stickland, my brother-in-law Peter Ekstedt and Sophie’s father Peter O’Donnell
– for their support and interest.

There are many others who are not fortunate enough to be called Peter. First
and always first is my beloved partner, Sophie, without whose encouragement and
belief I could not have achieved anything. She can have me back now! I would like
to thank all of my family and friends for their endless encouragement, and apologise
for the twenty-minute replies when asked, “How’s the PhD going?” In particular,
thanks to my Mum and Dad for their support through 23 years of education, although
it may have seemed endless.

My gratitude and congratulations go to all my colleagues and friends in the


department for creating an environment where doing cutting edge research is a lot of
fun. I would especially like to thank my unofficial supervisor, David Dixon, for his
wisdom and insight, especially regarding the Essendon Football Club; Ian, Adam,
Will, Brendan, Dan, Ross and anyone else who been part of the PFPC Ballhandlers
juggernaut; and Sabina Zahirovic and Damien Callahan for their friendship since the
Dark Ages (undergraduate days).

viii
There are many people who have contributed to the work presented in this
thesis who deserve recognition and thanks. My appreciation goes to the project leader
from Yorkshire Water, Martin Tillotson, for his input and hard work, as well as our
other contacts, Anna Taylor, Bill Brady and Derek Wilson. My thanks go to David
Leslie, David Ridge, Rene Frost and Paul Griffiths from United Utilities, and Ralph
Woolley from Brisbane City Council. Many thanks go to Lee White at Carnegie
Melon University in Pittsburgh for the incredible experience of studying with such a
brilliant scientist, and to his vibrant wife Wrenn and the rest of the Pittsburgh crew for
their fantastic hospitality and affection. My thanks go to Alastair Martin at UMIST
for his challenging discussions, generosity and homebrew, and to our international
visitors from Ireland and Denmark – Ciaran McCausland, John Bartlett, Mogens
Hingens, Kasper Soerensen, Soeren Olesen, Peter Christensen and Kristian Keiding –
for their collaboration and friendship.

Closer to home, I would like to thank the rest of the members of the Sludge
Group – Adam Kilcullen, Ainul Aziz, Brendan Gladman, Daniel Fisher, Daniel
Lester, David Dixon, David Verrelli, Ian Olmstead, Janine Hulston, Jonathan Foong,
Lindsay Studer, Ross de Kretser, Shane Usher and Will Goodall – for their
contributions and efforts. I believe that, together, we are bringing about a paradigm
shift in many industries with regards to how they think about their sludge. My thanks
also go to Peter’s personal assistants, Nicole Robinson and Annmaree Sharkey,
without whom nothing happens, and to our former collaborators from CSIRO, Peter
Harbour, Nevil Anderson, Rob Eldridge, Brian Bolto, Florien Huber and Colin Soh.
As well as the postgraduates and researchers above, there were several fourth year
undergraduates who contributed to this work, including Angela Karvouris, Igor
Posenjak, Michael Kennedy, Casey Burgess, Nicholas Cotterill, Timothy Birks, David
Lam and Catherine Cappone. Thanks also go to John Styles from the Department of
Civil and Environmental Engineering for his contributions.

I acknowledge funding by an Australian Postgraduate Award from the


Australian Research Council and the project sponsors (United Utilities, Yorkshire
Water and Brisbane City Council) and contributions from the Particulate Fluids
Processing Centre, the Department of Chemical and Biomolecular Engineering and

ix
the University of Melbourne (through the Postgraduate Overseas Research Experience
Scholarship) towards travel expenses.

Finally, I would like to thank the reviewers. I predominantly enjoyed writing


this thesis and I hope that they enjoy it too.

x
Table of Contents
ABSTRACT........................................................................ I

DECLARATION ............................................................... III

PREFACE ........................................................................IV

PUBLICATIONS ...............................................................V

ACKNOWLEDGEMENTS ..............................................VIII

LIST OF FIGURE CAPTIONS .................................... XVIII

LIST OF TABLE CAPTIONS ................................... XXXIV

1 INTRODUCTION ............................................................ 1
1.1 Potable Water Treatment ....................................................... 3
Treatment Processes..........................................................................................................4
Sludge Composition ..........................................................................................................5
Sludge Disposal.................................................................................................................7

1.2 Wastewater Treatment............................................................ 9


Treatment Processes..........................................................................................................9
Sludge Composition ........................................................................................................11
Sludge Disposal...............................................................................................................12

1.3 Sludge Handling..................................................................... 15

1.4 Solid-Liquid Separation ........................................................ 21

1.5 Thesis Overview ..................................................................... 22

2 THEORY ...................................................................... 26

2.1 Background ............................................................................ 26

xi
2.2 Sedimentation-Consolidation Equations ............................. 32
One-Dimensional Cartesian Coordinates ........................................................................34

2.3 Filtration Theory.................................................................... 37

2.4 Constant-Pressure Piston-Driven Filtration without


Membrane Resistance............................................................................ 42
2.4.1 Initial Suspension Un-networked (φ0 < φg, t ≤ tc, Rm = 0) ...................45
2.4.2 Linear Approximation (φ0 < φ*, Rm = 0)..............................................46
Application To Experimental Characterisation ...............................................................48
Stepped-Pressure Filtration .............................................................................................50
Scaling Behaviour ...........................................................................................................51
2.4.3 Finite Difference Method (φ0 ≥ φg and φ0 < φg, t ≥ tc; Rm = 0) ............53
2.4.4 Non-Traditional Filtration ...................................................................55
Background .....................................................................................................................55
Material Characteristics ..................................................................................................56
Modelling Results ...........................................................................................................59
Conclusions.....................................................................................................................64

2.5 Constant-Pressure Piston-Driven Filtration with Membrane


Resistance................................................................................................ 65
2.5.1 Compressional Rheological Formulation............................................65
Runge-Kutta Numerical Algorithm.................................................................................72
2.5.2 Modified Darcy’s Law ........................................................................76

2.6 Soil Consolidation .................................................................. 77


2.6.1 One-dimensional Consolidation Theory .............................................78
2.6.2 Equating the Parameters......................................................................79

3 EXPERIMENTAL METHODS AND MATERIALS ....... 81


3.1 Methods................................................................................... 81
3.1.1 Background .........................................................................................81
3.1.2 Sample Preparation .............................................................................84
3.1.3 Density Measurement..........................................................................84
3.1.4 Oven drying.........................................................................................85
3.1.5 Constant-Pressure Batch Filtration......................................................86

xii
Stepped-Pressure Testing (Traditional Filtration) ...........................................................88
Single-Pressure Testing (Non-Traditional Filtration) .....................................................90
3.1.6 Batch settling.......................................................................................92
Transient Batch Settling..................................................................................................93
Equilibrium Batch Settling..............................................................................................93
3.1.7 Oedometer Testing ..............................................................................94

3.2 Calculations ............................................................................ 97


3.2.1 Water Treatment Sludge Characterisation ..........................................97
3.2.2 Wastewater Treatment Sludge Characterisation .................................99

3.3 Materials ............................................................................... 101


3.3.1 Water treatment sludges ....................................................................101
3.3.2 Wastewater treatment sludges ...........................................................102
3.3.3 Synthetic Sewage Sludges.................................................................102
3.3.4 Pulp and Paper...................................................................................104
3.3.5 Kaolin ................................................................................................104

3.4 Case Studies.......................................................................... 105

4 MATERIAL CHARACTERISATION RESULTS......... 106


4.1 Potable Water Treatment Sludges ..................................... 106
Solids Density ...............................................................................................................106
4.1.2 Ferric Water Treatment Sludges .......................................................108
Langsett Water Treatment Plant....................................................................................108
Arnfield Water Treatment Plant....................................................................................116
Other Ferric Water Treatment Sludges .........................................................................121
4.1.3 Alum Water Treatment Sludges ........................................................125
Hodder Water Treatment Plant .....................................................................................125
Huntington WTP and Oswestry WTP (Transient Batch Settling).................................129
Other Alum Water Treatment Sludges..........................................................................132
4.1.4 Water Treatment Coagulant Comparison..........................................136

4.2 Wastewater Treatment Sludges ......................................... 138


4.2.1 Filtration Scaling Behaviour .............................................................138
Experimental Validation ...............................................................................................138
Effect of Additives and Pre-Treatments........................................................................148
4.2.2 Logarithmic Fitting of Cake Compression........................................160

xiii
Theoretical Validation...................................................................................................160
Luggage Point Wastewater Treatment Plant .................................................................164
Other Wastewater Sludges ............................................................................................176
4.2.3 Synthetic Sewage Sludges.................................................................178

4.3 Other Sludges ....................................................................... 183


4.3.1 Kaolin (Consolidation Testing Methods)..........................................183
4.3.2 Pulp-and-Paper Sludges ....................................................................192
4.3.3 Overall Sludge Comparison ..............................................................195

5 PLATE-AND-FRAME FILTER PRESS MODELLING197


5.1 Model Formulations ............................................................ 197
One-Dimensional Filtration ..........................................................................................200
5.1.2 Analytical Models .............................................................................203
Fixed-Cavity Filtration..................................................................................................203
Flexible-Membrane Filtration .......................................................................................204
5.1.3 Numerical Models .............................................................................205
Fixed-Cavity Filtration..................................................................................................206
Flexible Membrane Filtration .......................................................................................213
Runge-Kutta Numerical Algorithms .............................................................................216
5.1.4 Model Results....................................................................................222
Fixed-Cavity Results.....................................................................................................222
Fill-and-Squeeze Results...............................................................................................234
5.1.5 Prediction Tools ................................................................................240
Fill-Only Calculation Program......................................................................................240
Fill-and-Squeeze Calculation Program .........................................................................244

5.2 Model Validation Case Studies........................................... 248


5.2.1 Langsett Water Treatment Plant........................................................249
Background ...................................................................................................................249
Results and Discussion..................................................................................................250
Model Validation ..........................................................................................................260
Outcomes ......................................................................................................................263
5.2.2 Hodder Water Treatment Plant .........................................................264
Background ...................................................................................................................264
Results and Discussion..................................................................................................265
Fill-Only Model Validation...........................................................................................276
Throughput Options and Recommendations.................................................................277

xiv
Outcomes ......................................................................................................................280
5.2.3 Hodder Water Treatment Plant (Pilot-Scale Trials) ..........................281
Background ...................................................................................................................281
Pilot-Scale Trial Results................................................................................................282
Flexible-Membrane Model Validation..........................................................................288
Outcomes ......................................................................................................................292
5.2.4 Thornton Steward Water Treatment Plant ........................................293
Background ...................................................................................................................293
Results...........................................................................................................................294
Filtration Options ..........................................................................................................299
Outcomes ......................................................................................................................306
5.2.5 Arnfield Water Treatment Plant........................................................308
Background ...................................................................................................................308
Fill-Only Model Predictions .........................................................................................308
Outcomes ......................................................................................................................311

5.3 Filter Press Optimisation and Control .............................. 312


Loading Time................................................................................................................313
Ramping Pressure .........................................................................................................314
Cake Compression ........................................................................................................314
5.3.2 Fill-Only Filter Press Performance ...................................................316
Feed Concentration .......................................................................................................316
Fill Pressure ..................................................................................................................321
Cavity Width .................................................................................................................323
Membrane Resistance ...................................................................................................326
5.3.3 Fill-and-Squeeze Filter Press Performance .......................................329
Feed Concentration .......................................................................................................332
Squeeze Pressure...........................................................................................................335
Handling Time ..............................................................................................................336
Optimisation and Control..............................................................................................338

5.4 Conclusions........................................................................... 339

6 CENTRIFUGE MODELLING ..................................... 341


6.1 One-Dimensional Solid-bowl Batch Centrifuge................ 343
6.1.1 Model Formulation............................................................................346
Sedimentation-Consolidation Equations .......................................................................346
Dimensionless Equations ..............................................................................................347
Case 1: Initial Suspension Un-networked (φ0 < φg).......................................................349

xv
Case 2: Initial Suspension Networked (φ0 > φg) ............................................................359
6.1.2 Results ...............................................................................................365
Case 1: Initial Suspension Un-networked (φ0 < φg).......................................................365
Case 2: Initial Suspension Networked (φ0 > φg) ............................................................367

6.2 Pseudo-One-Dimensional Solid-Bowl Continuous


Centrifuge ............................................................................................. 371
6.2.1 Model Formulation............................................................................374
Sedimentation-Consolidation Equations .......................................................................374
Case 1: Inlet Suspension Un-networked (φi < φg) .........................................................379
Case 2: Inlet Suspension Networked (φi > φg) ..............................................................382
6.2.2 Results ...............................................................................................385

6.3 Continuous Decanting Centrifuge Modelling ................... 394


6.3.1 Background .......................................................................................394
6.3.2 Model Description.............................................................................397
6.3.3 Overflow Section...............................................................................400
Helical Coordinates.......................................................................................................401
Viscous Gravity Flow ...................................................................................................403
Scrolling Velocity of Consolidation Zone.....................................................................412
Conservation Equations ................................................................................................416
Dimensionless Equations ..............................................................................................419
6.3.4 Inlet Section.......................................................................................424
6.3.5 Underflow Section.............................................................................426
6.3.6 Numerical Solution ...........................................................................428

6.4 Conclusions........................................................................... 429

7 THICKENER MODELLING........................................ 431


7.1 Water and Wastewater Sludge Thickening Predictions .. 431
7.1.1 Background .......................................................................................431
7.1.2 Results and Discussion......................................................................432
Characterisation ............................................................................................................432
Thickener Modelling.....................................................................................................432
7.1.3 Conclusions .......................................................................................436

8 CONCLUSIONS......................................................... 437

xvi
NOMENCLATURE ........................................................ 440
Latin alphabet................................................................................................................440
Greek alphabet ..............................................................................................................443
Subscripts and Superscripts...........................................................................................444

REFERENCES .............................................................. 446

xvii
List of Figure Captions

Figure 1.1.1: Schematic of a typical water treatment process........................................4


Figure 1.2.1: Schematic of biological wastewater treatment process ............................9
Figure 1.2.2: Sewage sludge matrix .............................................................................11
Figure 1.3.1: Sludge handling flowchart for water and wastewater treatment ............15
Figure 1.3.2: Thickening tank at Oswestry Water Treatment Plant, Cheshire.............16
Figure 1.3.3: Gravity belt filter at Calder Vale Incineration Plant, Yorkshire.............16
Figure 1.3.4: Cake release from the plate-and-frame filter press at North Dean
Wastewater Treatment Plant, Yorkshire ..............................................................17
Figure 1.3.5: Belt press filter at Calder Vale Incineration Plant, Yorkshire ................17
Figure 1.3.6: Decanting Centrifuge (courtesy of Alfa-Laval)......................................18
Figure 2.1.1: Volume fraction dependent material characteristics: (a) Compressive
yield stress, Py(φ); (b) Hindered settling function, R(φ).......................................29
Figure 2.2.1: Volume element of networked suspension, illustrating the forces
involved................................................................................................................32
Figure 2.3.1: Schematic of one-dimensional constant pressure filtration (φ0 < φg); (a)
Initial condition; (b) Cake formation; (c) Cake compression; (d) Equilibrium
condition...............................................................................................................37
Figure 2.3.2: Algorithms for Constant-Pressure Piston-Driven Filtration ...................41
Figure 2.4.1: Example of experimental results for traditional filtration ......................49
Figure 2.4.2: Example of experimental results for non-traditional filtration behaviour
..............................................................................................................................55
Figure 2.4.3: Solids diffusivity, D(φ) = 10-6 φ2 (1 - φ)6 m2/s........................................58
Figure 2.4.4: Plot of D(φ) showing φ∞ and the range of φ0 values used for modelling
variable initial concentration................................................................................60
Figure 2.4.5: Finite difference modelling predictions of constant pressure filtration for
φ∞ = 0.60 v/v with varying φ0 ...............................................................................60
Figure 2.4.6: Plot of D(φ) showing φ0 and the range of φ∞ values used for modelling
variable applied pressure......................................................................................61

xviii
Figure 2.4.7: Finite difference modelling predictions of constant pressure filtration for
φ0 = 0.10 v/v with varying applied pressure.........................................................62
Figure 2.4.8: Linear approximation and finite difference predictions for φ∞ = 0.6 v/v
and φ0 = 0.1 v/v ....................................................................................................63
Figure 2.4.9: φ* as a function of φ0 for a range of φ∞ values: ◊ Variable applied
pressure; □ Variable initial concentration; --- φ* = φ0 ..........................................63
Figure 2.5.1: Algorithm for piston-driven constant-pressure filtration with membrane
resistance ..............................................................................................................74
Figure 3.1.1: Photo and schematic of automated filtration rig.....................................87
Figure 3.1.2: Example of t versus V2 results for the stepped-pressure compressibility
test (Ferric water treatment sludge, Langsett WTP, 05/06/01) ............................89
Figure 3.1.3: Example of t versus V2 results for the stepped-pressure permeability test
(Ferric water treatment sludge, Langsett WTP, 05/06/01)...................................89
Figure 3.1.4: Example of dt/dV2 versus t results for the stepped-pressure permeability
test (Ferric water treatment sludge, Langsett WTP, 05/06/01) ............................90
Figure 3.1.5: Example of logarithmic curve fit to average volume fraction results
(Luggage Point WWTP, 14/02/03, 20 kPa) .........................................................91
Figure 3.1.6: Schematic of oedometer .........................................................................95
Figure 3.2.1: Flowchart of water treatment sludge characterisation using stepped-
pressure filtration and equilibrium batch settling.................................................97
Figure 3.2.2: Flowchart of water treatment sludge characterisation using stepped-
pressure filtration and transient batch settling......................................................98
Figure 3.2.3: Flowchart of wastewater treatment sludge characterisation from single-
pressure filtration and transient batch settling......................................................99
Figure 4.1.1: Effect on Py(φ) of using ρs = 2 or 3 g/cm3 (Alum water treatment sludge,
Hodder WTP, 06/11/01).....................................................................................107
Figure 4.1.2: Effect on D(φ) of using ρs = 2 or 3 g/cm3 (Alum water treatment sludge,
Hodder WTP, 06/11/01).....................................................................................107
Figure 4.1.3: Py(φ) results for stepped-pressure filtration of Langsett WTP sludge ..109
Figure 4.1.4: R(φ) results for stepped-pressure filtration of Langsett WTP sludge ...110
Figure 4.1.5: D(φ) results for stepped-pressure filtration of Langsett WTP sludge...111
Figure 4.1.6: Experimental data and model predictions for compressibility stepped -
pressure filtration test of Langsett WTP, 05/06/01 ............................................112

xix
Figure 4.1.7: Experimental data and model predictions for the permeability stepped -
pressure filtration test of Langsett WTP, 05/06/01 ............................................112
Figure 4.1.8: Transient batch settling results and analysis curve fits for Langsett WTP,
09/07/03..............................................................................................................113
Figure 4.1.9: Py(φ) results from stepped-pressure filtration and transient batch settling
for Langsett WTP, 09/07/03...............................................................................114
Figure 4.1.10: R(φ) results from stepped-pressure filtration and transient batch settling
for Langsett WTP, 09/07/03...............................................................................115
Figure 4.1.11: D(φ) results from stepped-pressure filtration and transient batch settling
for Langsett WTP, 09/07/03...............................................................................116
Figure 4.1.12: Gel point determination via equilibrium batch settling for Arnfield
WTP samples from 14/06/02..............................................................................118
Figure 4.1.13: Py(φ) results for stepped-pressure filtration of Arnfield WTP sludge 118
Figure 4.1.14: R(φ) results for stepped-pressure filtration of Arnfield WTP sludge .119
Figure 4.1.15: D(φ) results for stepped-pressure filtration of Arnfield WTP sludge .120
Figure 4.1.16: Py(φ) results for stepped-pressure filtration of ferric water treatment
sludges ................................................................................................................122
Figure 4.1.17: R(φ) results for stepped-pressure filtration of ferric water treatment
sludges ................................................................................................................123
Figure 4.1.18: D(φ) results for stepped-pressure filtration of ferric water treatment
sludges ................................................................................................................123
Figure 4.1.19: Gel point determination of Hodder WTP 30/10/01 via equilibrium
batch settling ......................................................................................................126
Figure 4.1.20: Py(φ) results for stepped-pressure filtration of Hodder WTP sludges 127
Figure 4.1.21: R(φ) results for stepped-pressure filtration of Hodder WTP sludges .127
Figure 4.1.22: D(φ) results for stepped-pressure filtration of Hodder WTP sludges .128
Figure 4.1.23: Transient batch settling results and analysis curve fits for Huntington
WTP 09/07/03 and Oswestry WTP 25/06/03.....................................................129
Figure 4.1.24: Py(φ) results and curve fits from filtration and batch settling for
Huntington WTP 09/07/03 and Oswestry WTP 25/06/03 .................................130
Figure 4.1.25: R(φ) results and curve fits from filtration and batch settling for
Huntington WTP 09/07/03 and Oswestry WTP 25/06/03 .................................131

xx
Figure 4.1.26: D(φ) results and curve fits from filtration and batch settling for
Huntington WTP 09/07/03 and Oswestry WTP 25/06/03 .................................132
Figure 4.1.27: Py(φ) results for stepped-pressure filtration of alum water treatment
sludges ................................................................................................................133
Figure 4.1.28: R(φ) results for stepped-pressure filtration of alum water treatment
sludges ................................................................................................................134
Figure 4.1.29: D(φ) results for stepped-pressure filtration of alum water treatment
sludges ................................................................................................................135
Figure 4.1.30: Overview of D(φ) for a range of coagulant types in potable water
treatment.............................................................................................................136
Figure 4.2.1: t versus V2 results for single-pressure filtration of un-flocculated Carrum
WWTP activated sludge at 200 kPa with varying h0 .........................................139
Figure 4.2.2: Adjusted time versus scaled volume results for single-pressure filtration
of un-flocculated Carrum WWTP activated sludge at 200 kPa with varying h0140
Figure 4.2.3: Raw and scaled results for single-pressure filtration of flocculated
Carrum WWTP activated sludge at 50, 100 and 200 kPa with varying h0 ........141
Figure 4.2.4: t versus V2 results for single-pressure filtration of flocculated Carrum
WWTP activated sludge at 200 kPa with varying φ0 .........................................142
Figure 4.2.5: Adjusted time versus scaled volume results for single-pressure filtration
of flocculated Carrum WWTP activated sludge at 200 kPa with varying φ0 .....143
Figure 4.2.6: Generalised material property functions used to investigate the variation
of filtration behaviour due to changes in φ0 .......................................................144
Figure 4.2.7: Variation of φ0 scaling function with solids diffusivity........................145
Figure 4.2.8: Variation of φ0 scaling function with compressive yield stress............146
Figure 4.2.9: Variation of φ0 scaling function with applied pressure, ∆P for D1P5 ..146
Figure 4.2.10: t versus V2 for constant pressure filtration of sludges from a variety of
WWTP’s.............................................................................................................149
Figure 4.2.11: t versus φ for single-pressure filtration of sludges from a variety of
WWTP’s.............................................................................................................149
Figure 4.2.12: Adjusted time versus scaled volume results for single-pressure
filtration of sludges from a variety of WWTP’s ................................................150

xxi
Figure 4.2.13: Adjusted time versus scaled volume results for single-pressure
filtration of Colac digested sludge treated with a variety of cationic
polyelectrolytes. .................................................................................................152
Figure 4.2.14: Adjusted time versus scaled volume results for single-pressure
filtration of Lilydale activated sludge conditioned with Zetag 87 at varying doses
............................................................................................................................153
Figure 4.2.15: Adjusted time versus scaled volume results for single-pressure
filtration of Colac digested sludge conditioned with Zetag 87 at varying doses154
Figure 4.2.16: Adjusted time versus scaled volume results for single-pressure
filtration of Mornington digested sludge conditioned with Zetag 87 at varying
doses ...................................................................................................................154
Figure 4.2.17: Adjusted time versus scaled volume results for single-pressure
filtration of conditioned sludges at the optimum or maximum dose of Zetag 87
............................................................................................................................155
Figure 4.2.18: Adjusted time versus scaled volume results for single-pressure
filtration of Colac digested sludge conditioned with ferric chloride at varying
doses ...................................................................................................................156
Figure 4.2.19: Adjusted time versus scaled volume results for single-pressure
filtration of Lilydale activated sludge conditioned with 4% ferric chloride and
varying doses of Zetag 87 ..................................................................................157
Figure 4.2.20: Adjusted time versus scaled volume results for single-pressure
filtration of Lilydale activated sludge conditioned with various combinations of
ferric chloride and Zetag 87 ...............................................................................157
Figure 4.2.21: Adjusted time versus scaled volume results for single-pressure
filtration of Carrum activated sludge after hydrothermal treatment, treatment
with Fenton’s reagent, and both in combination ................................................159
Figure 4.2.22: Filtration predictions for validation of logarithmic method ...............161
Figure 4.2.23: Logarithmic fits to complete and partial results for 100 kPa..............162
Figure 4.2.24: Compressive yield stress, Py(φ), and predictions from logarithmic curve
fitting of full and incomplete data sets ...............................................................162
Figure 4.2.25: Solids diffusivity, D(φ), and predictions from logarithmic curve fitting
of full and incomplete data sets..........................................................................163

xxii
Figure 4.2.26: Experimental and curve fitting results for Luggage Point M400-1-S1
............................................................................................................................167
Figure 4.2.27: Experimental and curve fitting results for Luggage Point A20-S2 ....167
Figure 4.2.28: Py(φ) results for Luggage Point WWTP from single-pressure filtration
tests.....................................................................................................................168
Figure 4.2.29: D(φ) results for Luggage Point WWTP from single-pressure filtration
tests.....................................................................................................................169
Figure 4.2.30: Transient batch settling test results for Luggage Point WWTP Sample 2
............................................................................................................................170
Figure 4.2.31: Py(φ) results and curve fits from filtration and batch settling for
Luggage Point WWTP Sample 2 .......................................................................171
Figure 4.2.32: R(φ) results and curve fits from filtration and batch settling for
Luggage Point WWTP Sample 2 .......................................................................172
Figure 4.2.33: D(φ) results and curve fits from filtration and batch settling for
Luggage Point WWTP Sample 2 .......................................................................173
Figure 4.2.34: Validation of experimental results using modelling predictions for
Luggage Point WWTP Sample 2 .......................................................................175
Figure 4.2.35: D(φ) from single-pressure filtration of a range of wastewater treatment
sludges ................................................................................................................177
Figure 4.2.36: Scaled single-pressure filtration results for Carrum WWTP sewage
sludge and alginate synthetic sludge based on the model proposed by Vesilind
............................................................................................................................180
Figure 4.2.37: Scaled single-pressure filtration results for Carrum WWTP sewage
sludge and alginate synthetic sludges with varying particle composition .........181
Figure 4.2.38: Scaled single-pressure filtration results for Carrum WWTP sewage
sludge and gelatine synthetic sludge ..................................................................182
Figure 4.3.1: Casagrande's method applied to oedometer sample 1, 100kPa ............184
Figure 4.3.2: Taylor's method applied to oedometer sample 1, 100kPa ....................185
Figure 4.3.3: cv(e) results from oedometer testing for kaolin.....................................187
Figure 4.3.4: Py(φ) results from filtration and oedometer testing of kaolin sample...188
Figure 4.3.5: D(φ) results from filtration and oedometer testing of kaolin sample ...188
Figure 4.3.6: Experimental results and model predictions for compressibility stepped-
pressure filtration test for kaolin ........................................................................190

xxiii
Figure 4.3.7: Py(φ) results for stepped-pressure filtration of pulp-and-paper sludges
............................................................................................................................192
Figure 4.3.8: R(φ) results for stepped-pressure filtration of pulp-and-paper sludges 193
Figure 4.3.9: D(φ) results for stepped-pressure filtration of pulp-and-paper sludges 194
Figure 4.3.10: D(φ) results from pressure filtration for a variety of sludges ............196
Figure 5.1.1: Schematic of plate-and-frame filter press cycle: (a) Load, (b) Fill, (c)
Squeeze, (d) Unload ...........................................................................................198
Figure 5.1.2: Applied pressure and feed flowrate with time for Hodder WTP, Press #1,
6/11/01................................................................................................................199
Figure 5.1.3: Schematic of one-dimensional pressure filtration; (a) Fixed-cavity
filtration, and (b) Flexible-membrane filtration .................................................201
Figure 5.1.4: Observed pressure rise and stepped pressure model input for Hodder
WTP, Press #1, 6/11/01......................................................................................203
Figure 5.1.5: Runge-Kutta numerical algorithm for fixed-cavity filtration with
membrane resistance ..........................................................................................218
Figure 5.1.6: Runge-Kutta numerical algorithm for flexible-membrane filtration with
membrane resistance ..........................................................................................221
Figure 5.1.7: Volume fraction distribution results for fixed-cavity filtration with
varying Rm; (φ0 = 0.004 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0) ........................223
Figure 5.1.8: Model predictions of t versus V2 for fixed-cavity filtration with varying
Rm; (φ0 = 0.004 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0) .....................................224
Figure 5.1.9: Model predictions of dt/dV2 versus t for fixed-cavity filtration with
varying Rm; (φ0 = 0.004 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0) ........................225
Figure 5.1.10: Model predictions of cake resistance for fixed-cavity filtration with
varying Rm; (φ0 = 0.004 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0) ........................226
Figure 5.1.11: Volume fraction distribution results for fixed-cavity filtration with
varying Rm; (φ0 = 0.02 v/v, h0 = 0.015 m, ∆PF = 6 bar, tp = 0) ..........................227
Figure 5.1.12: Model predictions of t versus V2 for fixed-cavity filtration with varying
Rm; (φ0 = 0.02 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0) .......................................228
Figure 5.1.13: Model predictions of dt/dV2 versus t for fixed-cavity filtration with
varying Rm; (φ0 = 0.02 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0) ..........................228
Figure 5.1.14: Model predictions of cake resistance for fixed-cavity filtration with
varying Rm; (φ0 = 0.02 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0) ..........................229

xxiv
Figure 5.1.15: Volume fraction distribution results for fixed-cavity filtration with
ramping pressure; (h0 = 0.015 m, ∆PF = 6 bar, tP = 1200 s, Rm = 0)..................230
Figure 5.1.16: Model predictions of t versus V2 for fixed-cavity filtration with
ramping pressure (h0 = 0.015 m, ∆PF = 6 bar, tp = 0) ........................................231
Figure 5.1.17: Model predictions of membrane resistance as measured by the Darcian
method for fixed-cavity filtration with varying Rm; (h0 = 0.015 m, ∆PF = 6 bar, tP
= 0) .....................................................................................................................232
Figure 5.1.18: Volume fraction distribution results for fixed-cavity filtration with
complex material characteristics (φ0 = 0.02 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0
s, Rm = 0 Pa.s/m).................................................................................................233
Figure 5.1.19: Model predictions of t versus V2 for fixed-cavity filtration with
complex material characteristics and varying Rm (φ0 = 0.02 v/v, h0 = 0.015 m,
∆PF = 6 bar, tP = 0).............................................................................................233
Figure 5.1.20: Volume fraction distribution results for flexible-membrane filtration
with membrane resistance (φ0 = 0.004 v/v, ∆PF = 6 bar, ∆PS = 10 bar; h0 = 0.015
m; Rm = 1011 Pa.s/m, tf = 3600 s)........................................................................235
Figure 5.1.21: Model predictions of t versus V2 for flexible-membrane filtration with
varying tF and Rm (φ0 = 0.004 v/v, h0 = 0.015 m, ∆PF = 6 bar, ∆PS = 10 bar, tP =
0).........................................................................................................................236
Figure 5.1.22: Model predictions of dt/dV2 versus t for flexible-membrane filtration
with varying tF and Rm (φ0 = 0.004 v/v, h0 = 0.015 m, ∆PF = 6 bar, ∆PS = 10 bar,
tP = 0)..................................................................................................................237
Figure 5.1.23: Volume fraction distribution results for flexible-membrane filtration
with membrane resistance (φ0 = 0.02 v/v, ∆PF = 6 bar, ∆PS = 10 bar, h0 = 0.015
m, Rm = 1011 Pa.s/m, tF = 3600 s, tP = 0)............................................................238
Figure 5.1.24: Model predictions of t versus V2 for flexible-membrane filtration with
varying tF and Rm (φ0 = 0.02 v/v, h0 = 0.015 m, ∆PF = 6 bar, ∆PS = 10 bar, tP = 0)
............................................................................................................................239
Figure 5.1.25: Model predictions of dt/dV2 versus t for flexible-membrane filtration
with varying tF and Rm (φ0 = 0.02 v/v, h0 = 0.015 m, ∆PF = 6 bar, ∆PS = 10 bar, tP
= 0) .....................................................................................................................239
Figure 5.1.26: Example of V versus t data with varying φ0 for fill-only array file ....241
Figure 5.1.27: Example of V versus t data with varying ∆PF for fill-only array file .241

xxv
Figure 5.1.28: Applied pressure profiles for fill-only data files.................................242
Figure 5.1.29: User interface for fill-only calculation program .................................243
Figure 5.1.30: Example of V versus t data for fill-and-squeeze data files .................245
Figure 5.1.31: User interface for fill-and-squeeze calculation program ....................246
Figure 5.2.1: Langsett WTP plate-and-frame presses: (a) Operator Bill Brady with
Press 1; (b) Press 3 .............................................................................................249
Figure 5.2.2: Layout of the membrane and cavity for Langsett WTP Presses 1 and 2
............................................................................................................................251
Figure 5.2.3: Layout of the membrane and cavity for Press 3 ...................................252
Figure 5.2.4: t versus V2 results (raw and adjusted) for Langsett WTP Press 2.........253
Figure 5.2.5: dt/dV2 versus 1/V results for Langsett WTP Press 2.............................255
Figure 5.2.6: Solids throughput versus cake solids results for Langsett WTP Press 2
............................................................................................................................255
Figure 5.2.7: t versus V2 results for Langsett WTP Press 3 .......................................257
Figure 5.2.8: Time versus cake solids for Langsett WTP Press 3..............................257
Figure 5.2.9: dt/dV2 versus 1/V results for Langsett WTP Press 3.............................258
Figure 5.2.10: Solids throughput versus time for multiple runs of Langsett WTP Press
3..........................................................................................................................259
Figure 5.2.11: Experimental data and model predictions for Langsett WTP press 2 260
Figure 5.2.12: Experimental data and model predictions for Langsett WTP press 3 262
Figure 5.2.13: Hodder Water Treatment Plant, Lancashire .......................................264
Figure 5.2.14: Raw water turbidity and colour at Hodder WTP ................................265
Figure 5.2.15: Variation of alum dose with colour at Hodder WTP..........................266
Figure 5.2.16: Variation of total suspended solids with colour of at Hodder WTP...267
Figure 5.2.17: Layout of the membrane and cavity for Filter Presses 1 and 2 ..........268
Figure 5.2.18: Fill-only press #2 at Hodder WTP (a) Press in operation; (b) Operator
Paul Griffiths unloading the press ......................................................................269
Figure 5.2.19: Observed lateral movement during filtration - Red lines represent flow;
Blue lines represent lateral compression ............................................................270
Figure 5.2.20: Solids distribution across cake............................................................271
Figure 5.2.21: Applied pressure and feed flowrate variation with time for Hodder
WTP fill-only presses.........................................................................................273
Figure 5.2.22: t versus V2 results (raw and adjusted) for Hodder WTP fill-only presses
............................................................................................................................273

xxvi
Figure 5.2.23: dt/dV2 versus 1/V for Hodder WTP fill-only presses..........................274
Figure 5.2.24: Solids throughput versus cake solids results for Hodder WTP fill-only
presses ................................................................................................................275
Figure 5.2.25: Experimental results and modelling predictions for Hodder WTP fill-
only presses ........................................................................................................276
Figure 5.2.26: Pilot-scale plate-and-frame filter press (Baker-Hughes) ....................281
Figure 5.2.27: t versus V2 results for Hodder WTP fill-and-squeeze trial set 1 .........284
Figure 5.2.28: t versus V2 results for Hodder WTP fill-and-squeeze trial set 2 .........284
Figure 5.2.29: dt/dV2 versus 1/V for Hodder WTP fill-and-squeeze trial set 1..........287
Figure 5.2.30: dt/dV2 versus 1/V for Hodder WTP fill-and-squeeze trial set 2..........287
Figure 5.2.31: Experimental and modelling results for Hodder WTP fill-and-squeeze
trail set 1 .............................................................................................................289
Figure 5.2.32: Experimental and modelling results for Hodder WTP fill-and-squeeze
trail set 2 .............................................................................................................290
Figure 5.2.33: Fill-only plate-and-frame filter press at Thornton Steward WTP ......293
Figure 5.2.34: Feed flowrate and pressure results for Thornton Steward WTP fill-only
press, 20/06/02 ...................................................................................................296
Figure 5.2.35: t versus V2 results for Thornton Steward WTP fill-only press, 20/06/02
............................................................................................................................296
Figure 5.2.36: dt/dV2 versus 1/V for Thornton Steward WTP fill-only press, 20/06/02
............................................................................................................................297
Figure 5.2.37: Solids throughput versus cake solids for Thornton Steward WTP fill-
only press, 20/06/02 ...........................................................................................299
Figure 5.2.38: t versus V2 results and model predictions for Thornton Steward WTP
fill-only press......................................................................................................300
Figure 5.2.39: Fill-only model predictions of t versus V2 for Arnfield WTP ............309
Figure 5.2.40: Fill-only model predictions of average cake solids with time for
Arnfield WTP.....................................................................................................310
Figure 5.2.41: Fill-only model predictions of suspended solids throughput with
adjusted cake solids for Arnfield WTP ..............................................................311
Figure 5.3.1: Fixed-cavity filtration predictions of average cake solids with time for
ferric water treatment sludge; ∆PF = 10 bar, d = 3 cm, Rm = 0, tH = 1 hr, tP = 30
min......................................................................................................................317

xxvii
Figure 5.3.2: Fixed-cavity filtration predictions of average cake solids with time for
alum water treatment sludge; ∆PF = 10 bar, d = 3 cm, Rm = 0, tH = 1 hr, tP = 30
min......................................................................................................................317
Figure 5.3.3: Fixed-cavity filtration predictions of solids throughput with average
cake solids for ferric water treatment sludge; ∆PF = 10 bar, d = 3 cm, Rm = 0, tH =
1 hr, tP = 30 min .................................................................................................319
Figure 5.3.4: Fixed-cavity filtration predictions of solids throughput with average
cake solids for alum water treatment sludge; ∆PF = 10 bar, d = 3 cm, Rm = 0, tH =
1 hr, tP = 30 min .................................................................................................319
Figure 5.3.5: Fixed-cavity filtration predictions of specific filtrate flowrate with
average cake solids for ferric water treatment sludge; ∆PF = 10 bar, d = 3 cm, Rm
= 0, tH = 1 hr, tP = 30 min...................................................................................320
Figure 5.3.6: Fixed-cavity filtration predictions of specific filtrate flowrate with
average cake solids for alum water treatment sludge; ∆PF = 10 bar, d = 3 cm, Rm
= 0, tH = 1 hr, tP = 30 min...................................................................................321
Figure 5.3.7: Fixed-cavity filtration predictions of t versus V2 for ferric water
treatment sludge with varying applied pressure; φ0 = 3 wt%, d = 3 cm, Rm = 0, tH
= 1 hr, tP = 30 min ..............................................................................................322
Figure 5.3.8: Fixed-cavity filtration predictions of cake solids with time for ferric
water treatment sludge with varying pressure; φ0 = 3wt%, d = 3cm, Rm = 0, tH = 1
hr, tP = 30 min ....................................................................................................322
Figure 5.3.9: Fixed-cavity filtration predictions of solids throughput with average
cake solids for ferric water treatment sludge with varying applied pressure; φ0 =
3 wt%, d = 3 cm, Rm = 0, tH = 1 hr, tP = 30 min .................................................323
Figure 5.3.10: Fixed-cavity filtration predictions of t versus V2 for ferric water
treatment sludge with varying cavity widths; φ0 = 3 wt%, ∆PF = 10 bar, Rm = 0, th
= 1 hr, tp = 30 min ..............................................................................................324
Figure 5.3.11: Fixed-cavity predictions of cake solids with time for ferric water
treatment sludge with varying cavity width; φ0 = 3 wt%, ∆P = 10 bar, Rm = 0, th =
1 hr, tp = 30 min..................................................................................................325
Figure 5.3.12: Fixed-cavity predictions of solids throughput with cake solids for ferric
water treatment sludge with varying cavity width; φ0 = 3 wt%, ∆P = 10 bar, Rm =
0, th = 1 hr, tp = 30 min.......................................................................................325

xxviii
Figure 5.3.13: Fixed-cavity filtration predictions of t versus V2 for ferric water
treatment sludge with varying membrane resistances; φ0 = 3 wt%, ∆P = 10 bar, d
= 3 cm.................................................................................................................327
Figure 5.3.14: Fixed-cavity filtration predictions of cake solids versus cycle time for
ferric water treatment sludge with varying membrane resistances; φ0 = 3 wt%,
∆P = 10 bar, d = 3 cm ........................................................................................327
Figure 5.3.15: Fixed-cavity filtration predictions of solids throughput versus cake
solids for ferric water treatment sludge with varying membrane resistances; φ0 =
3 wt%, ∆P = 10 bar, d = 3 cm ............................................................................328
Figure 5.3.16: Flexible-membrane predictions of squeeze time after a certain fill time
for ferric water treatment sludge; ∆PF = 6 bar, ∆PS = 10 bar, φ0 = 4 wt%, d = 3
cm, tH = 60 min ..................................................................................................330
Figure 5.3.17: Flexible-membrane predictions of squeeze time after a certain fill time
for alum water treatment sludge; ∆PF = 6 bar, ∆PS = 10 bar, φ0 = 4 wt%, d = 3
cm, tH = 60 min ..................................................................................................330
Figure 5.3.18: Flexible-membrane predictions of solids throughput with total cycle
time for ferric water treatment sludge; ∆PF = 6 bar, ∆PS = 10 bar, φ0 = 4 wt%, d
= 3 cm, tH = 60 min ............................................................................................331
Figure 5.3.19: Flexible-membrane predictions of solids throughput with total cycle
time for alum water treatment sludge; ∆PF = 6 bar, ∆PS = 10 bar, φ0 = 4 wt%, d =
3 cm, tH = 60 min ...............................................................................................332
Figure 5.3.20: Flexible-membrane predictions of squeeze time required to reach 30
wt% with fill time for ferric water treatment sludge; ∆PF = 6 bar, ∆PS = 10 bar, d
= 3 cm, tH = 60 min ............................................................................................333
Figure 5.3.21: Flexible-membrane predictions of squeeze time required to reach 30
wt% with fill time for alum water treatment sludge; ∆PF = 6 bar, ∆PS = 10 bar, d
= 3 cm, tH = 60 min ............................................................................................333
Figure 5.3.22: Flexible-membrane predictions of maximum solids throughput with
feed concentration for ferric water treatment sludge; Fill-and-squeeze (F&S):
∆PF = 6 bar, ∆PS = 10 bar; Fill-only (F): ∆PF = 10 bar; d = 3 cm, tH = 60 min.334
Figure 5.3.23: Flexible-membrane predictions of maximum solids throughput with
feed concentration for alum water treatment sludge; Fill-and-squeeze (F&S): ∆PF
= 6 bar, ∆PS = 10 bar; Fill-only (F): ∆PF = 10 bar; d = 3 cm, tH = 60 min ........335

xxix
Figure 5.3.24: Flexible-membrane predictions of maximum solids throughput with
feed concentration under various pressure regimes (∆PF, ∆Ps) for ferric water
treatment sludge; φF = 30 wt%, d = 3 cm, tH = 60 min ......................................336
Figure 5.3.25: Flexible-membrane predictions of solids throughput with cycle time for
ferric water treatment sludge at a range of tH; ∆PF = 6 bar, ∆PS = 10 bar, φ0 = 4
wt%, φF = 30 wt% ..............................................................................................337
Figure 5.3.26: Flexible-membrane predictions of maximum solids throughput with φ0
for ferric water treatment sludge for a range of tH; ∆PF = 6 bar, ∆PS = 10 bar, φF
= 30 wt%, d = 3 cm ............................................................................................337
Figure 6.1.1: Schematic of 1-D batch centrifugal thickening: (a) Axial view; (b)
Lateral view........................................................................................................343
Figure 6.1.2: Numerical algorithm for batch centrifugation, initial suspension un-
networked case (T ≤ Tc)......................................................................................355
Figure 6.1.3: Numerical algorithm for batch centrifugation, initial suspension un-
networked case (T > Tc)......................................................................................357
Figure 6.1.4: Numerical algorithm for batch centrifugation, initial suspension
networked case ...................................................................................................362
Figure 6.1.5: Volume fraction distribution predictions for 1-D batch centrifuge model
with different initial concentrations (φ0 < φg, rb = 0.5 m, rf = 0.25 m and ω =
2000 rpm). The annotated values correspond to the scaled time at each solution
............................................................................................................................365
Figure 6.1.6: Predictions of sedimentation and consolidation profiles for 1-D batch
centrifuge model with differing initial concentrations (φ0 < φg, rb = 0.5 m, rf =
0.25 m and ω = 2000 rpm). The annotated values correspond to lines of constant
concentration. (b) and (c) are enlarged views of (a) and (d) .............................366
Figure 6.1.7: Volume fraction distribution predictions for 1-D batch centrifuge model
with different initial concentrations (φ0 > φg, rb = 0.5 m, rf = 0.25 m and ω =
2000 rpm). The annotated values correspond to the scaled time at each solution
............................................................................................................................367
Figure 6.1.8: Predictions of sedimentation and consolidation profiles for 1-D batch
centrifuge model with differing initial concentrations (φ0 > φg, rb = 0.5 m, rf =
0.25 m and ω = 2000 rpm). The annotated values correspond to lines of constant
concentration ......................................................................................................368

xxx
Figure 6.1.9: Transient sediment and cake height results for one-dimensional batch
centrifuge model with different initial concentrations (rb = 0.5 m, rf = 0.25 m and
ω = 2000 rpm) ....................................................................................................369
Figure 6.1.10: Equilibrium volume fraction distribution results for one-dimensional
batch centrifuge model with different initial concentrations (rb = 0.5 m, rf = 0.25
m and ω = 2000 rpm) .........................................................................................370
Figure 6.2.1: Schematic of 1-D continuous centrifugal thickening: (a) Axial view; (b)
Lateral view........................................................................................................371
Figure 6.2.2: Numerical algorithm for 1-D continuous centrifugation, inlet suspension
un-networked......................................................................................................381
Figure 6.2.3: Numerical algorithm for 1-D continuous centrifugation, inlet suspension
networked ...........................................................................................................384
Figure 6.2.4: One-dimensional continuous centrifuge results of volume fraction
distribution for ferric water treatment sludge at different inlet concentrations (φi
< φg, ru = 0.5 m, ri = 0.25 m, W(r) = 0.5 m and ω = 2000 rpm).........................386
Figure 6.2.5: One-dimensional continuous centrifuge results of sediment volume
fraction distribution for ferric water treatment sludge at different inlet
concentrations (φi < φg, ru = 0.5 m, ri = 0.25 m, W(r) = 0.5 m and ω = 2000 rpm)
............................................................................................................................387
Figure 6.2.6: One-dimensional continuous centrifuge results of volume fraction
distribution for ferric water treatment sludge at different inlet concentrations (φi
> φg, ru = 0.5 m, ri = 0.25 m, W(r) = 0.5 m and ω = 2000 rpm).........................388
Figure 6.2.7: One-dimensional continuous centrifuge results of scaled cake radius
versus underflow volume fraction for ferric water treatment sludge at different φi
and ω (ru = 0.5 m, ri = 0.25 m and W(r) = 0.5 m) ..............................................389
Figure 6.2.8: One-dimensional continuous centrifuge results of scaled throughput
versus underflow volume fraction for ferric water treatment sludge at different φi
and ω (ru = 0.5 m, ri = 0.25 m and W(r) = 0.5 m) ..............................................390
Figure 6.2.9: One-dimensional continuous centrifuge results of solids throughput
versus underflow volume fraction for ferric water treatment sludge at different φi
and ω (ru = 0.5 m, ri = 0.25 m and W(r) = 0.5 m) ..............................................391

xxxi
Figure 6.2.10: One-dimensional continuous centrifuge results of suspension
throughput versus underflow volume fraction for ferric water treatment sludge at
different φi and ω (ru = 0.5 m, ri = 0.25 m and W(r) = 0.5 m) ...........................391
Figure 6.2.11: One-dimensional continuous centrifuge results of scaled cake radius
versus underflow volume fraction for ferric water treatment sludge at different φi
and ri (ru = 0.5 m, ω = 2000 rpm and W(r) = 0.5 m)..........................................392
Figure 6.2.12: One-dimensional continuous centrifuge results of scaled solids
throughput versus underflow volume fraction for ferric water treatment sludge at
different φi and ri (ru = 0.5 m, ω = 2000 rpm and W(r) = 0.5 m) .......................393
Figure 6.2.13: One-dimensional continuous centrifuge results of suspension
throughput versus underflow volume fraction for ferric water treatment sludge at
different φi and ri (ru = 0.5 m, ω = 2000 rpm and W(r) = 0.5 m) .......................393
Figure 6.3.1: Cross-section of cylindrical-conical helical-conveyor centrifuge. (Bird
Machine Co.)......................................................................................................394
Figure 6.3.2: Schematic of change of variables from (a) cylindrical coordinates to (b)
helical coordinates..............................................................................................397
Figure 6.3.3: Schematic of continuous decanting centrifuge, unwrapped along scroll
blade ...................................................................................................................398
Figure 6.3.4: Schematic of overflow section of continuous decanting centrifuge .....400
Figure 6.3.5: Schematic of scroll edge .......................................................................401
Figure 6.3.6: Viscous gravity flow across a stick surface..........................................403
Figure 6.3.7: Streamlines of gravity viscous flow across a flat surface.....................411
Figure 6.3.8: Movement of the consolidation zone in a decanting centrifuge due to the
scroll; (a) Radial view (b) View down spiral direction; (c) Transverse view ....412
Figure 6.3.9: Helical movement of solids through a continuous decanting centrifuge
............................................................................................................................414
Figure 6.3.10: Fluid and solid velocities in the radial and helical directions in the
overflow section .................................................................................................416
Figure 6.3.11: Schematic of inlet section of decanting centrifuge.............................424
Figure 6.3.12: Schematic of underflow section of decanting centrifuge ...................426
Figure 7.1.1: Thickener modelling output for Langsett thickener Sample 1 - Solids
flux versus underflow solids concentration for four different bed heights (1, 2, 5,
10 m) ..................................................................................................................434

xxxii
Figure 7.1.2: Thickener modelling output for Langsett, Huntington and Oswestry
thickener samples - Solids flux versus underflow solids concentration for bed
heights of 1 and 10 m .........................................................................................434
Figure 7.1.3: Thickener modelling output for Luggage Point Sample 2 - Solids flux
versus underflow solids concentration for four different bed heights (1, 2, 5, 10
m) .......................................................................................................................435

xxxiii
List of Table Captions

Table 1.2.1: Approximate costs of sewage sludge disposal for United Utilities, UK..13
Table 1.3.1: Dewatering devices used in water (WT) and wastewater (WWT)
treatment...............................................................................................................19
Table 1.5.1: Summary of model development presented in this work.........................24
Table 2.4.1: Material property parameters ...................................................................57
Table 3.3.1: Standard formulation for synthetic sludges ...........................................103
Table 4.1.1: Langsett WTP sludge samples ...............................................................108
Table 4.1.2: Fitting parameters for Langsett WTP averages......................................111
Table 4.1.3: Py(φ) fitting parameters for Langsett WTP, 09/07/03............................114
Table 4.1.4: Arnfield WTP sludge samples ...............................................................117
Table 4.1.5: Fitting parameters for Arnfield WTP sludges........................................120
Table 4.1.6: Other ferric water treatment sludge samples..........................................121
Table 4.1.7: Fitting parameters for average ferric water treatment sludge ................124
Table 4.1.8: Hodder WTP sludge samples .................................................................125
Table 4.1.9: Fitting parameters for average Hodder WTP sludge .............................128
Table 4.1.10: Py(φ) fitting parameters for Huntington WTP 09/07/03 and Oswestry
WTP 25/06/03 ....................................................................................................129
Table 4.1.11: Other alum water treatment sludge samples ........................................133
Table 4.1.12: Fitting parameters for average alum WTP sludge ...............................135
Table 4.2.1: Properties of the polyelectrolytes used for conditioning sludges. ........152
Table 4.2.2: Material parameters used for theoretical validation...............................160
Table 4.2.3: Pressure filtration experimental and curve fitting results for Luggage
Point WWTP ......................................................................................................165
Table 4.2.4: Py(φ) curve fitting parameters for Luggage Point WWTP Sample 2.....172
Table 4.3.1: Comparison of consolidation and filtration analyses .............................183
Table 4.3.2: Oedometer testing results using Taylor’s and Casagrande’s methods...186
Table 4.3.3: Fitting parameters for kaolin from filtration testing ..............................189
Table 4.3.4: Summary of other sludges characterised using pressure filtration ........195

xxxiv
Table 5.1.1: Layout of t and V values in array file for given φ0 and ∆PF...................245
Table 5.2.1: Filter press runs from Langsett WTP.....................................................250
Table 5.2.2: Required solids throughput at current coagulation and flocculation
conditions ...........................................................................................................267
Table 5.2.3: Fill-only filter press runs at Hodder WTP .............................................270
Table 5.2.4: Required solids throughput at increased coagulation and flocculation
dosage.................................................................................................................277
Table 5.2.5: Hodder WTP pilot-scale trials by Baker-Hughes ..................................282
Table 5.2.6: Measured versus calculated final solids concentrations for Hodder WTP
fill-and-squeeze trials .........................................................................................286
Table 5.2.7: Comparison of experimental and modelling results ..............................291
Table 5.2.8: Filter cake sample analysis from Thornton Steward WTP fill-only press,
20/06/02..............................................................................................................298
Table 5.2.9: Throughput predictions for refurbished press at Thornton Steward WTP
............................................................................................................................302
Table 5.2.10: Throughput predictions for new presses at Thornton Steward WTP...304
Table 5.2.11: Throughput predictions for refurbished press with increased D(φ) at
Thornton Steward WTP .....................................................................................305
Table 5.2.12: Throughput predictions for new press with increased D(φ) at Thornton
Steward WTP .....................................................................................................305

xxxv
1 Introduction

Water is a limited resource, and its conservation and effective use are issues of
global importance. The removal of soluble and insoluble matter and contaminants
from raw water to produce potable water is imperative for the health and safety of
communities. Industrial and domestic wastewater requires sterilisation and treatment
before safe and efficient reuse or disposal of the effluent and solid waste is achieved.
Unsafe water and poor sanitation cause in excess of 5 million deaths per year (Annan,
2000). Recognising this, most nations understand that the supply of clean drinking
water and the efficient disposal of wastewater are prerequisites for sustainable
development. Global initiatives to increase access to potable water and sewage
treatment have been created, such as the United Nations International Decade on
Water for Life (2005-2015). In September 2000, world leaders at the United Nations
Millennium Summit pledged to halve the proportion of people unable to reach or to
afford safe drinking clean water by 2015 (Annan, 2000). At the 2002 World Summit
on Sustainable Development in Johannesburg, a matching target was agreed to halve
the proportion of people lacking adequate sanitation, also by 2015 (UN, 2002).

Industrial-scale water and wastewater treatment processes produce a mixture


of solid and liquid waste, or sludge, that must be disposed of in some way. Reducing
the water content of the sludge using solid-liquid separation operations generally
decreases the cost of disposal. This thesis encompasses the fundamental
mathematical modelling of plant-scale separation devices and the accurate
measurement of the compressional rheological material properties of sludges to
predict performance and give improvements in operation, design and control of such
devices. The aim of the work was to increase throughput and reduce water content,
thereby reducing the economic, social and environmental costs of sludge handling and
disposal.

Chapter 1 outlines the processes involved in water and wastewater treatment,


the nature of the sludges produced, and the common disposal routes. Both the

1
volumes of sludge and the costs of disposal are increasing, thus the cost of sludge
handling is becoming a greater proportion of the overall cost of treatment. An
overview of the devices used in industry is given, followed by an introduction and
overview of the work performed to measure the material properties and to model the
separation devices.

Stock's Reservoir, Lancashire, United Kingdom

2
1.1 Potable Water Treatment

The principle aim of the potable water treatment industry is the delivery of high
quality drinking water to the community. The standards and regulations for the
quality of drinking water differ around the world but generally specify maximum
levels and ensure constant monitoring and documentation of a large range of
contaminants. These contaminants include (AWWA, 1999):

- A multitude of organics including natural organic matter (NOM), petroleum


by-products, dioxins, polymers, fertilizers, pesticides and herbicides;

- Pathogens such as Giardia lamblia, Legionella, Escherichia coli and


Cryptosporidium parvum;

- Heavy metals (lead, mercury, arsenic et cetera);

- Inorganics such as fluoride and nitrates;

- Disinfection by-products such as trihalomethanes; and

- Radionuclides such as uranium.

Colour, odour, pH and hardness are also regulated. Taste is also usually
monitored by water providers. Guidelines are released periodically by the World
Health Organisation (WHO, 2004) as a basis for the development of national
standards, including the Australian Drinking Water Guidelines (ADWG, 1996). The
ADWG provide an authoritative reference on the quality of drinking water and
incorporates frameworks for the management of drinking water quality, which puts
greater emphasis on prevention rather than testing to ensure water coming from
people's taps is safe. The European Union has the Drinking Water Directive (EU,
1998), although member nations may adopt stricter standards. In the United
Kingdom, the Drinking Water Inspectorate (through the Department for Environment
Food and Rural Affairs) heavily regulates the semi-privatised industry. In the United
States, the key legislation governing quality is the Safe Drinking Water Act (SDWA,

3
1974) that enables the U.S. Environmental Protection Agency to establish the
necessary regulations.

Treatment Processes

The method used for water treatment depends on the raw water quality, which
is measured in terms of its colour (dissolved matter), turbidity (suspended solids) and
concentrations of contaminants. Contaminant removal is the main goal of most
companies and is therefore the main determinant of the required process. The
schematic shown in Figure 1.1.1 outlines a typical treatment process used for waters
free of serious contamination (in which case special treatment may be necessary).

Raw Preliminary
Water Treatment

Coagulation

Thickener Overflow,
Filtrate or Centrate
Flocculation

Sludge
Clarification
Handling

Disinfection and Potable


Sludge Disposal
pH Adjustment Water

Figure 1.1.1: Schematic of a typical water treatment process

The major steps involve (Casey, 1997, Binnie, et al., 2002):

- Preliminary treatment through screening and settlement of large solids, and


disinfection and oxidation with ozone;

- pH adjustment and coagulation to precipitate dissolved material and to


agglomerate particulates. This is predominantly performed chemically using
solutions of aluminium or ferric salts, in which the cationic species neutralises

4
the charge repulsion effects of suspended particulates. ‘Sweep coagulation’
involves adding a large excess of coagulant in order to ensure contaminant
removal. ‘Enhanced coagulation’ refers to adding even more coagulant to
guarantee NOM removal. Polyelectrolytes such as polyaluminum chlorides
and polyacrylates are sometimes used in conjunction with coagulant solutions
or as the sole coagulant;

- The addition of long-chain polyelectrolytes or polymers to flocculate


particulates through bridging or electrostatic mechanisms. Flocculation at
optimum conditions of dose, pH and shear can aid solid-liquid separation by
increasing the particle size and floc density;

- The removal of solid matter from the potable water using settlement or
clarification processes. Wide ranges of processes are used, often in
conjunction with each other, such as lamellar clarifiers, gravity sand-filters,
dissolved air flotation, membrane filtration and activated carbon adsorption;

- The wash-water and sludges from these processes are combined and
concentrated using sedimentation, centrifugation and/or filtration techniques
(termed ‘sludge handling’) before disposal. The thickener overflow, centrate
or filtrate is returned to the head of the works; and

- Secondary disinfection of the potable water using chlorine, and pH adjustment


and dechlorination using sulphur dioxide prior to distribution.

If the sludge handling process operations are designed or operated poorly, the
throughput of the plant is restricted (that is, the sludge handling causes a bottleneck)
or the final waste has high water content.

Sludge Composition

Water treatment sludges generally consist of material from six sources (Casey,
1997, Binnie, et al., 2002):

- Raw water turbidity;

- Raw water colour;

5
- Dissolved metals such as iron and manganese;

- Coagulant;

- Flocculant; and

- Biological growth (such as algae in the raw water or during processing,


especially slow sand filters).

The addition of electrolytes such as the trivalent salts of aluminium and iron
causes coagulation through the destruction of colloidal stability. These act by
reducing the zeta potential of hydrophobic colloids through increases to the ionic
strength of the medium (which compresses the electric double-layer), and ion
adsorption (which reduces the net repulsive charge). Al3+ and Fe3+ are commonly
used due to their effectiveness at coagulation, low solubility and relative low cost.
They also precipitate as hydroxide flocs, which gives rise to a third mechanism of
colloidal destabilisation through physical entrapment of small particles. Long-chain
polyelectrolytes are occasionally used to coagulate colloidal particles using bridging
mechanisms, but are predominantly used as flocculants to bridge coagulated particles.

The chemical reactions due to the Al3+ or Fe3+ include dissolution of their
respective salts, followed by hydrolysation and complexation to form metal hydroxide
ions. For example, the complexation reactions for aluminium include:

Al(H2O)63+ ⇔ Al(H2O)5(OH)2+ + H+ …(1.1.1)

and

Al(H2O)5(OH)2+ ⇔ Al(H2O)4(OH)2+ + H+ …(1.1.2)

These metal hydroxides further polymerise to form dimers, trimers and higher
hydroxide complexes which ultimately form positively charged colloidal precipitates:

2Al(H2O)5(OH)2+ ⇔ Al2(H2O)8(OH)24+ + 2H2O …(1.1.3)

The net precipitation reaction in the presence of alkalinity for trivalent metal
ions is:

6
M3+ + 3HCO3- → M(OH)3(s) + 3CO2 …(1.1.4)

Aluminium hydroxide is insoluble in the narrow pH range of 5.2 to 7.2, while


ferric hydroxide is more versatile and insoluble in the pH range of 5 to 10. The
precipitated solids generally form bulky gelatinous structures with a relatively low
biodegradable organic fraction.

Sludge Disposal

Water treatment sludges are usually disposed of to landfill or down the sewer,
or occasionally used in the manufacture of concrete and bricks, for land reclamation
or as a soil additive in agriculture. The disposal routes in the United Kingdom in
1998 as a percentage of the total mass of dry solids (131 000 tonnes) were 57% to
landfill, 25% to sewer and 18% by other means (Binnie, et al., 2002), although the
ratio varies for individual companies. Sludges that are disposed of to landfill are
usually thickened and dewatered in order to reduce transport costs and landfill
charges, which are steadily increasing. The total cost of sludge handling and
dewatering can be significant (for Yorkshire Water, it is approximately 20% of the
total cost of water treatment (Appleton, 2004)), but varies greatly depending upon the
raw water, the treatment process and the disposal route.

More water treatment sludge is expected to be dewatered to offset the


increasing costs of transport and disposal. On top of this, the total amount of sludge is
generally increasing since greater demands for quality of potable water has resulted in
a trend from sweep coagulation towards enhanced coagulation, which translates as
more sludge.

Note that companies that are also responsible for sewage treatment
predominantly perform discharge of liquid wastes to sewer; therefore the disposal cost
is burdened by the wastewater operations within the same company. It is debatable
whether this is an advantage, since sewage sludges are more difficult to dewater and
water treatment sludges have little calorific or biological content, but dewatering
operations for water and wastewater are combined.

Encouragingly, increased conservation efforts are beginning to help to reduce


the demands on the quantities of potable water. In Australia, which has one of the

7
worlds highest per capita water usage, extended periods of drought across the country
has enforced restrictions cutting household usage and encouraged grey water use
instead of potable water use in irrigating gardens, parks and farms. In the United
Kingdom, the metering of houses has highlighted wastage from old pipes. In New
Zealand, where the Maori people view the release of human waste into rivers, lakes
and oceans as offensive, composting toilets for urban communities are being
commercialised (Lowe, 2004). If widely implemented, such systems reduce the
amount of potable water used to flush the toilet, and the amount of sewage sludge to
be treated.

Desalination of seawater as an alternative to the treatment of freshwater is an


economically viable option only for the driest and richest places, such as the Middle-
East, although the relative costs are falling and more potable water will be produced
using this method. For example, the state government of New South Wales plans to
build desalination plants in the near future to overcome water shortages, despite the
disincentive to improve water efficiency and high greenhouse emissions (Davies,
2004, Peatling, 2004).

8
1.2 Wastewater Treatment

The majority of developed and an increasing number of developing countries


have strict legislation covering the standards of treatment of household and industrial
wastes, with the aim of reducing the levels of dissolved and suspended organic and
inorganic matter in the effluent, including the concentrations of pollutants, pathogens
and heavy metals. The consequences of not treating raw waste prior to discharge can
be highly detrimental to the local environment and population. This section
introduces the nature of wastewater treatment and the sludge it produces, and outlines
the various disposal methods and their relative advantages and disadvantages.

Treatment Processes

A wide range of processes is used in wastewater or sewage treatment, but the


majority generally involve initial screening and primary sedimentation of the raw
sewage followed by biological treatment. Figure 1.2.1 shows a simplified schematic
of the common unit operations.

Initial
Influent
Screening

Primary
Settling

Biological
Treatment
Activated Sludge Biofilter Effluent
Recycle Recycle

Secondary
Settling
Sludge Effluent

Figure 1.2.1: Schematic of biological wastewater treatment process

9
The two main biological treatment processes for domestic sewage are aerobic
biofiltration or activation (Casey, 1997, Henze, et al., 2002):

- Aerobic biofilters (also called ‘trickling’ or ‘percolating’ filters) involve


passing sewage through a bed of coarse particles (50 – 100mm). The physical
processes of absorption and coagulation and the biological processes of
synthesis and respiration remove suspended and dissolved organics. Zones of
oxidation and nitrification exist at different depths of the bed. The influent
may be mixed with an effluent recycle to increase hydraulic loading rates.

- The activated sludge process involves aerating a mixed-liquor of influent and


recycled (or activated) settled sludge. The micro-organisms in the activated
sludge use the organic matter in the influent for nutrients. A range of
configurations exists to promote different aerobic, anaerobic and anoxic
conditions, depending on the required biochemical reactions. The effluent is
separated from the activated sludge by secondary settling.

A wide collection of physicochemical or biological treatment processes also


exists as pre-treatments or as the sole treatment process. They often involve changes
to temperature, pressure or pH in order to facilitate the biochemical reactions or to
change the nature of the sludge. Mesophilic processes (such as Mesophilic Aerobic
Digestion (MAD)) usually operate around 20°C (Metcalf and Eddy, 1991), while
thermophilic digestion operates at a temperature exceeding 45°C (LaPara and
Alleman, 1999). A recent technology, Auto Thermal Aerobic Digestion (ATAD) is
an activated sludge process operating at thermophilic temperatures, in which heat is
generated by the metabolic activities of the thermophiles present (Mason, et al., 1992,
Stentiford, 2001). Other treatments include the CAMBI process (Kepp, et al., 2000,
Neyens and Baeyens, 2003), enzymatic hydrolysis (Lutz, et al., 1993), super-critical
oxidation and pre-digestion ultra-sonication (Ødegaard, 2004).

Energy producing processes such as catalytic cracking of long carbon chains to


make fuel oil out of the sludge (Skrypsi-Mäntele, et al., 2000) or microbial fuel cells
(in which the reduction/oxidation reactions of digestion are used to generate
electricity) (Biever, 2004) have the potential to revolutionise wastewater treatment by
offsetting the cost of treatment.

10
Sludge Composition

100 µm

Figure 1.2.2: Sewage sludge matrix

A very diverse collection of micro-organisms may be present in sewage sludge


at each stage of processing and depends upon the source of the wastewater and the
method of treatment. The various methods of treatment and populations of micro-
organisms all produce a sludge containing a matrix of biological material. A
microscopic view is shown in Figure 1.2.2, illustrating the heterogeneous nature of the
matrix. Sewage sludges consist of fibrous matter, dissolved and suspended organic
and inorganic particles (such as bacterial cells, contaminants and dirt), and
extracellular polymer (ECP) (Bruus, et al., 1992, Vesilind, 1994, Novak, et al., 1999).
The ECP component consists of various proteins, organic acids, lipids and
polysaccharides, which form a polymeric network that binds the various components
together. Various flocculants and coagulants are also added to the sludge to make
dewatering easier, further complicating the sludge matrix.

In recent years there has been an increased focus on the precise role of ECP in
sludge dewatering. For many years, carbohydrate was considered the main
constituent, but recent work suggests that protein is also a very important component
(Liu and Fang, 2002). In general, the functional groups present on the ECP have a net
negative charge (Sobeck and Higgins, 2002), such that their role in flocculation and
dewatering is important and the cations (especially polyvalent bridging cations such
as Ca2+) have a large effect (Higgins and Novak, 1997, Cousin and Ganczarczyk,
1998). The ECP may affect dewatering due to steric forces, gelation or complexation.

11
It has been demonstrated (Mikkelsen and Keiding, 2002, Neyens and Baeyens,
2003) that ECP aids sedimentation due to a tendency to form larger flocs, but is
detrimental to filtration. Thus, optimum ECP concentrations and compositions exist
that are dependent on the method of dewatering. For example, ATAD may be very
useful for digestion but may produce a highly impermeable sludge. A number of
researchers have considered the possibility of changing the nature of the sludge by
manipulating feedstock and digestion conditions (Chen, et al., 2001, Houghton and
Stephenson, 2002, Sponza, 2003).

Sludge Disposal

Common disposal methods for sewage sludges are listed as follows (Priestley,
1998):

- Wastewater sludge is a potent fertiliser and can be mixed into soils as a


renewable source of nutrients;

- Sludge cakes can be transported and disposed of to landfill. Sewage sludges


are classified as hazardous wastes and are under strict landfill and transport
regulations; and

- Sewage sludges can be burnt in an incinerator in order to generate electricity,


which can offset the cost of treatment. Since the calorific value depends upon
the moisture content, there is a solids concentration at which the organic
content will burn without added fuel (or added cost) and becomes ‘auto-
thermal’. For digested sludges, this is usually about 32 to 33 weight percent
solids.

Occasionally, opportunities arise to dispose of sludge in the reclamation of


abandoned mines or used as a fertilizer prior to reforestation.

The relative benefits of the various disposal options depend upon a wide
variety of economic, social and environmental driving forces. A comparison of the
approximate economic costs of disposal per tonne of dry solids (tds) for United
Utilities is given in Table 1.2.1 (Coombes, 2004):

12
Table 1.2.1: Approximate costs of sewage sludge disposal for United Utilities, UK

Disposal Option Disposal Cost (£/tds)

Land Application (cake) 50

Land Application (liquid) 120

Landfill (cake)+ 150

Incineration (cake)++ 80
+
Includes transport, gate fees and landfill tax
++
Includes fuel oil costs

Land application of sludge cake is the cheapest option, followed by


incineration. For all disposal methods, sludge cakes are preferable to liquid sludges as
they are cheaper to transport and store, which offsets the cost of dewatering. In
addition, land application of liquid sludges is more expensive than sludge cakes since
they must be injected into soils to prevent run-off and cannot be applied in wet
weather, whereas cake is easily spread and ploughed. As such, a greater proportion of
sludge is being dewatered than in previous years. The economic costs of all disposal
methods are expected to increase in the future as transport and fuel costs increase
(Coombes, 2004).

Due to its low cost, land application was the predominant method of disposal
in Europe and was previously encouraged by government policy (EU, 1986).
However it has become controversial due to concerns from supermarkets and
consumers of a risk to public health. The voluntary ADAS agreement between
retailers and treatment operators (ADAS, 2001), which ended the application of raw
sludge to land, and legislative changes aiming to reduce the amount of nitrates in
organic fertilizers by up to 30% (EU, 1991) are likely to contribute to the significant
reduction of sludge disposed to land. For United Utilities this will be by half. The
practice is expected to be banned altogether in the future (Coombes, 2004).

In Australia, landfill is the predominant method of disposal (Priestley, 1998).


However, in Europe, the Landfill Directive (EU, 1999) was introduced to reduce the
amount of waste going to landfill by increasing the landfill tax at a rate much larger
than inflation. Thus, this disposal route is also diminishing.

13
Incineration commonly has poor community perceptions due to high dioxin
levels in gaseous emissions. However, levels in the United Kingdom have dropped
considerably due to stricter standards (EU, 2000). Although many communities still
consider incineration unsafe, it is becoming the disposal route of choice for sludge
managers in highly populated countries.

Volumes of sludge to be disposed are expected to increase as well as the costs


of disposal. In the past, sludge from coastal areas was sometimes released into nearby
oceans, but legislation in many countries has banned this practice. In the United
Kingdom the volumes of sewage sludge have increased by roughly 50% in the past
ten years due to the introduction of the Urban Wastewater Treatment Directive (EU,
1991), since more treatment sites incorporate secondary treatment. This trend is
expected to continue with the introduction of new phosphorous standards that
necessitate extra ferric dosing, which will further increase sludge volumes by 10 –
15%. Sludge volumes in developing countries are also increasing as more wastewater
treatment is performed.

Although local practices and conditions may differ from the trends highlighted
here for the United Kingdom, sludge handling and disposal is becoming a major
proportion of the cost of wastewater treatment due to increased volumes and disposal
costs. Mikkelsen and Keiding, 2002, report that 30 – 50% of the annual operating
costs of a plant can be related to dewatering alone, which is confirmed by industrial
representatives (Appleton, 2004, Coombes, 2004). Profound changes are required to
the philosophy of wastewater treatment to incorporate this change, such that the entire
process is considered in terms of the type of sludge produced and its ease of
dewatering. For example, sludges that are to be incinerated on-site do not require
disinfection, while destructive processes such as pre-treatment ultra-sonication and
ATAD may increase the efficiency of digestion, but may also produce sludge with
very poor dewaterability.

14
1.3 Sludge Handling

The water and wastewater treatment industries use a range of solid-liquid


separation devices beyond the initial clarification or settling processes to increase the
solid content of their waste sludge and reduce disposal costs. Figure 1.3.1 shows a
typical flowchart for sludge handling.

Water Sludge Handling Wastewater


Treatment Treatment
Conditioned Thickener Overflow or
GBT Filtrate
Influent
Water
Filter Backwash

Settled Sludge
Sediment or

Thickening Biological
Clarification • Thickening Tanks
• Gravity Belt Thickeners
Treatment

Dewatering
Filtrate or

Filtrate or
Centrate

Centrate

Disinfection and •Plate-and-Frame Filters


pH Adjustment •Decanting Centrifuges
•Belt Press Filters

Potable Water Sludge Disposal Effluent Disposal

Figure 1.3.1: Sludge handling flowchart for water and wastewater treatment

The sediment from clarifiers or settling tanks, or the backwash from


membrane filters is thickened using thickening tanks or gravity belt thickeners
(GBTs). The thickener overflow or GBT filtrate is recycled, while the thickened
sludge is further dewatered by pressure filters, belt filters or centrifuges. The
resulting sludge cake is then disposed, while the filtrate or centrate is returned to the
head of the works. Some water and wastewater sites also incorporate evaporative
processes to dry the resulting cake before disposal. The following list introduces the
various process equipment that are used:

15
- Thickening tanks are large tanks operated batch-wise or continuously in which
the solids are allowed to settle from the liquid and form a blanket of sludge
(see Figure 1.3.2). The clear-liquor or overflow is removed from the top of the
tank, while the thickened sludge or underflow is removed from the bottom of
the tank. Rakes or lamellar plates are used to increase efficiencies.

Figure 1.3.2: Thickening tank at Oswestry Water Treatment Plant, Cheshire

- Gravity belt thickening involves spreading the liquid sludge across a moving
semi-permeable belt or membrane (see Figure 1.3.3). The sludge settles, and
filtrate passes through the membrane. Ploughs are used at intervals to disturb
the sludge blanket. GBTs are often used as the precursor to belt press filters.
GBT’s may use a vacuum to assist dewatering.

Figure 1.3.3: Gravity belt filter at Calder Vale Incineration Plant, Yorkshire

16
- Plate-and-frame filter presses consist of concertinas of plates suspended from
a frame (see Figure 1.3.4). The indented-plates are lined with a semi-
permeable membrane and form a cavity when pressed together. Sludge is fed
at pressure to the cavities through a central feed inlet. A cake builds-up
against the membrane and water is released to the filtrate ports. Some filters
then use hydraulic or pneumatic pressure to squeeze the membranes together.

Figure 1.3.4: Cake release from the plate-and-frame filter press at North Dean Wastewater
Treatment Plant, Yorkshire

- Belt press filters are operated continuously and involve pressing the sludge
between two taut belts that pass over a series of rollers (see Figure 1.3.5). The
dewatered cake is then scraped off the belts.

Figure 1.3.5: Belt press filter at Calder Vale Incineration Plant, Yorkshire

17
- Decanting centrifuges are essentially large spinning cylinders (see Figure
1.3.6). The solids settle against the bowl wall under the centrifugal field, and
are pushed out of the centrifuge by a scroll or screw that is rotating at a
slightly different rate to the bowl.

Figure 1.3.6: Decanting Centrifuge (courtesy of Alfa-Laval)

Water and wastewater treatment sludges are both particulate suspensions.


Above a certain solids concentration (termed the gel point), a network of particles
exists that can resist a load – dewatering only occurs when the strength of the network
is overcome. Such materials are described as yield stress materials and exhibit this
behaviour in both compression and shear. In compression, the mechanisms of
dewatering involve the self-weight of the particles either due to gravity or centrifugal
acceleration, or an external applied pressure. The dominant mechanism of
compressional dewatering and the typical initial and final concentrations for the
devices commonly used in water and wastewater treatment are presented in Table
1.3.1.

The self-weight of a particulate network is dependent upon the density


difference between the solid and liquid phases and the acceleration. Thus, the
thickening of biological materials under gravity produces small pressures while the
centrifugation of inorganic suspensions produces large pressures. Many devices use
this mechanism to dewater, including clarifiers, thickeners, settling ponds, tailings
dams, gravity belt filters and centrifuges.

18
Table 1.3.1: Dewatering devices used in water (WT) and wastewater (WWT) treatment

Mechanism of Typical initial Typical final


Dewatering device
dewatering concentrations (wt%) concentrations (wt%)

Thickener (tank or
Self-weight 0.1 – 1.0 1.0 – 4.0
lamellar)

Gravity Belt Filter Self-weight 0.1 – 4.0 1.0 – 8.0

WT: 15 – 25
Decanting Centrifuge Self-weight 0.5 – 4
WWT: 15 – 30

WT: 15 – 30
Plate-and-Frame Filter Applied pressure 0.5 - 4
WWT: 20 – 35

Belt Press Filter Applied pressure 0.1 – 8.0 15 – 25

External applied pressures are used in conjunction with a semi-permeable


membrane to form a cake of material, and may be due to a number of sources:

- Positive displacement pumps in fixed-cavity plate-and-frame filtration;

- Pneumatic or hydraulic pressure during the squeeze-phase of flexible-


membrane plate-and-frame filtration;

- Vacuum pressure used in drum and disc filtration and some gravity belt
operations; and

- Belt tension in belt press filters.

Changing the nature of the suspension through chemical additives and including
shear effects in the physical dewatering mechanism generally increase the efficiency
of these devices. Polymeric flocculants are added to sludges prior to dewatering in
order to aggregate coagulated particles into larger flocs, which increases the settling
and filtration rates and reduces the fines in the overflow or filtrate. Floc formation
requires some shear to facilitate mixing and agglomeration but high levels of shear
prior to dewatering are detrimental due to degradation of the fragile structure.
Inorganic filter aids such as anhydrous lime or diatomaceous earth can also be used to
increase filtration rates.

19
The incorporation of small levels of shear during compressional dewatering can
have tremendous impacts on processing since particulate suspensions are relatively
strong in compression, but yield easily in shear processes. Examples of shear
processes include raked thickeners, belt presses, lamellar clarifiers and disc-bowl
centrifuges. Too much shear during dewatering is detrimental since the suspension
begins to mix rather than yield, as seen in the doughnut-effect in raked thickeners, or
the flocs degrade.

20
1.4 Solid-Liquid Separation

Several solid-liquid separation theories and techniques for the measurement of


fundamental sludge properties exist for the prediction of dewatering behaviour, but
are generally not used to full effect by the water and wastewater industries. Suppliers
often dictate the choice of additives and equipment. Paraphrasing a senior industry
executive, the industry is not an intelligent buyer of chemicals or devices (Ford,
2001). This is caused in part by the theoretical and experimental restrictions of the
techniques that are used, especially for sewage sludges, whose composition and
behaviour is poorly understood and unsuccessfully predicted (Novak, 2001).
Traditionally, Kynchian batch settling analysis is used to model thickeners (Kynch,
1952, Warden, 1983, Dillon, 1997) and Darcy’s approach to model filters (Tiller and
Shirato, 1964). The Kynchian theory fails to incorporate bed compression in
thickening, while Darcy’s law is less accurate for compressible materials.

Recent compressional rheological theories (Buscall and White, 1987, Bürger,


et al., 2001) have had limited application to water and wastewater sludges, but have
been used successfully in many other industries. Models of thickening, filtration and
centrifugation based upon these fundamental theories provide tools for prediction,
optimisation and trend analysis. Central to accurate prediction is the determination of
useful material properties. The traditional tests of Capillary Suction Time (CST),
Settled Volume Index (SVI) and Specific Resistance to Filtration (SRF) provide
limited information. However, filtration and batch settling techniques developed at
the University of Melbourne based upon compressional rheology measure
fundamental material properties over a wide range of solids concentrations, which can
then be used to model the material behaviour in any dewatering device.

This approach has the potential to revolutionise the water and wastewater
industries attitude to sludge handling. By providing useful experimental and
modelling tools, operators can make direct comparisons of additives and sludge types,
and accurate operational, design and control decisions. This has the potential to
minimise flocculant usage and optimise the performance of thickeners, filters and
centrifuges, thereby reducing the costs associated with the disposal of sludges.

21
1.5 Thesis Overview

The aim of the work presented in this thesis was to predict and optimise the
performance of solid-liquid separation processes in the water and wastewater
treatment industries through the use of mathematical models of sedimentation,
filtration and centrifugation and the accurate measurement of the dewatering
properties of sludges. Following this introduction are chapters specifying:

- Compressional rheological theories used;

- Material characterisation experimental methods and analysis techniques;

- Experimental results for the characterisation of the dewatering properties of a


range of sludges; and

- Mathematical development, solution and use of models of filters, centrifuges


and thickeners, including the on-site validation of the filter models.

The theory chapter begins with a review of relevant dewatering theories,


including Darcian and phenomenological filtration theories from the physical science
literature and soil consolidation theory from civil engineering. The compressional
rheology theory of Buscall and White, 1987 was used as the basis for the models, and
the general equations of conservation are covered in depth. The chapter provides a
summary of the previously developed model of piston-driven filtration, which was the
main experimental technique used here.

The filtration behaviour of sewage sludges is different to the behaviour of


water treatment sludges. The theory was used to show that this behaviour is covered
by existing theory. Based upon this, the filtration results of sewage sludges were
scaled to allow for differences in initial concentration and initial piston height.

Chapter 3 outlines the various experimental methods used, predominantly


single and multiple-pressure filtration and transient batch settling, and their analysis
techniques. It discusses issues of oven drying, density measurement, sludge
preparation and storage. Whilst previously developed methods for the analysis of

22
filtration results for inorganic sludges such as water treatment sludges were utilised,
new techniques were required for organic sludges such as sewage sludges - a new
method was developed and validated. An overview of soil consolidation analysis
methods is given. A summary of the materials used and an outline of the work
performed at several on-site case studies is included.

Chapter 4 details the results for the compressional dewatering characterisation


of sludges, including:

- Potable water treatment sludges from a wide range of sources, including


different coagulant and flocculant types;

- A wastewater treatment sludge, which was characterised over the full range of
solids concentrations using the novel analysis technique (allowing the
subsequent validation of the technique). This section also includes qualitative
comparisons of a wider range of wastewater sludges based on the scaling
behaviour and work on the development of synthetic sewage sludges; and

- A consolidated kaolin sample used in the comparison of filtration and


consolidation methods and analysis techniques.

The sludge properties are compared to a wide range of materials, such as


mineral and pulp-and-paper sludges.

The subsequent three chapters, titled Plate-and-Frame Filter Press Modelling,


Centrifuge Modelling and Thickener Modelling, detail the mathematical modelling
and solution of the dewatering device, and include ensuing validation, trend analysis
and software development. In most cases, the models were generalised such that any
functional form for the material characteristics could be used and the initial solids
concentration, φ0, could be above or below the gel point, φg. Analytical solutions
were used wherever possible in order to decrease calculation time. A summary of the
models is given in Table 1.5.1, along with the extent of their respective development.

In chapter five, analytical and numerical models of fixed-cavity and flexible-


membrane plate-and-frame filter presses are presented with validation from several
full-scale and pilot-scale test cases. The models were used on-site to optimise the

23
throughput and final cake solids concentration, to give design predictions and to
identify key control parameters. The membrane resistance was incorporated into the
filter press models allowing the investigation of the effect of fouling the filter cloths
on throughput and final solids predictions, the development of online measurement of
membrane resistance and cloth cleaning protocols. Visual Basic software was
developed to enable operators and designers to use the models without background
theoretical knowledge or programming skills.

Table 1.5.1: Summary of model development presented in this work

Validation/ Case Studies


Numerical Formulation

Analytical Formulation

Working Algorithm

Trend Analysis
Model

Fixed-cavity plate-and-frame filter (with Rm)  -+   

Fixed-cavity plate-and-frame filter (no Rm) ++ *   

+
Flexible-membrane plate-and-frame filter (with Rm)    

Flexible-membrane plate-and-frame filter (no Rm) ++ *   

1-D Batch centrifuge  -+  × ×

1-D Continuous centrifuge  ×  × ×

Decanting centrifuge # × × × ×
 Completed
× Incomplete or not attempted
- Not applicable
+
Analytical small-time approximations were formulated
++
Exact numerical solution for cake formation (φ0 < φg)
*Applies for φ0 < φ* (see Chapter 2 for definition of φ*)
#
Applies for φ0 < φi (see Chapter 6 for definition of φi)

Chapter 6 outlines the formulation of models of one-dimensional batch and


continuous centrifuges and the algorithms written to give numerical solutions. A two-
dimensional model of the overflow section of continuous decanting centrifuges was
developed using a combination of solid-liquid separation theory and fluid mechanics,

24
and an overview of the necessary theoretical and experimental steps forward for the
modelling of the inlet and underflow sections is discussed.

The third modelling chapter presents results for the application of continuous
thickening models developed from minerals industry research to the water and
wastewater industry using water and wastewater sludge characteristics.

As outlined in the conclusions, the work presented here represents several


significant developments in understanding the compressional dewatering behaviour of
water and wastewater treatment sludges, especially the comparison of ferric and alum
water treatment sludges from many sites and the development of analysis techniques
for the filtration characterisation of sewage sludges. The models of dewatering
devices are important advances in the application of solid-liquid separation theory to
full-scale plant operation and design, are either new applications of compressional
rheological theory or novel adaptations of existing models, and are relevant to a wide
range of industries.

The final chapters include the nomenclature and references.

25
2 Theory

2.1 Background

Solid-liquid suspensions are described as discrete particles or solids suspended


in a continuous fluid. Flocculated suspensions can form networked structures capable
of transmitting pressure such that the solid phase is also continuous. In order to
formulate a description of solid-liquid separation, the interactions between the solid
and liquid phases (the fluid drag) and between the particulates (the network strength)
must be understood. In general, dewatering theories describe the effect of an applied
pressure (mechanical, gravitational or centrifugal) on the particle and fluid pressures
(pp and pf respectively) in terms of the compressibility and permeability of the
suspension, where the compressibility is a measure of the equilibrium extent of
dewatering and the permeability gives a measure of the rate of dewatering.

Up until the late twentieth century, solid-liquid separation theories developed in


two parts, filtration or consolidation modelling and sedimentation modelling.
Filtration modelling is traditionally based on descriptions of the flow of liquid through
packed beds. Poiseuille, 1840, recognised that the fluid flowrate through capillaries
with circular cross-sections was proportional to the pressure drop and inversely
proportional to the viscosity of the fluid. At a similar time as Poiseuille, the local
municipal engineer at Dijon, Darcy, 1856, was developing a corresponding theory for
flow through a packed sand bed:

dV ∆p f
=K …(2.1.1)
dt ηf L

dV/dt is the liquid flowrate per unit area (L/T), K is the bed permeability (L2),
∆pf is the fluid pressure drop over the length of bed (M/LT2), ηf is the fluid viscosity
(M/LT) and L is the bed length (L).

26
Kozeny and Forchheimer, 1928, extended Poiseuille’s equation to beds of
irregular particles or channels. Assuming that the average path length varies with the
overall length, L, and using the Dupuit relation (Dupuit, 1863) to relate the mean pore
velocity to the overall velocity gives the Kozeny-Carman equation:

dV ε3 ∆p f
= …(2.1.2)
dt K ′′S 2p (1 − ε )2 η f L

ε is the voidage (specifically, the ratio of the voids to the total volume), Sp is
the specific surface area (L2) and K” is an empirical constant. Combining this with
Darcy’s law shows that the permeability of a packed bed is a function of the porosity,
specific surface area and K”. Scarlett, 1968, used a stereological description of
geometry for an irregular pore with laminar flow to give K” = π2/2.

Conventional filtration theory (Ruth, 1946, Tiller and Shirato, 1964) has
developed from these descriptions of packed bed behaviour, concentrating on how the
fluid flows through the bed. If the specific surface area can be measured, the Kozeny-
Carman equation is used to describe filtration behaviour. Otherwise, a modified
Darcy’s law is used, where the specific cake resistance per unit volume, αv (L-2), is
used instead of the permeability, K:

dV ∆p f
= …(2.1.3)
dt η f α v hb (t ) + Rm

V(t) is the specific volume of filtrate (L), hb(t) is the bed height (L) and Rm is
the membrane resistance (M/L2T). The specific cake resistance per unit mass, αm
(L/M), is also commonly used. Equation 2.1.3 is referred to as the Hagen-Poiseuille
equation (Dahlstrohm, et al., 1998). Assuming that both phases are incompressible,
hb(t) is related to V(t) by the conservation of volume:

φ0
hb (t ) = V (t ) …(2.1.4)
(φ b (t ) − φ 0 )

27
φb is the average bed solids volume fraction. Assuming that the filter cake is
incompressible such that φb and αv are constant with respect to time allows integration
of equation 2.1.3 to give:

η f α vφ 0 Rm
t= V2 + V …(2.1.5)
2 ∆p f (φ b − φ 0 ) ∆p f

Thus, for incompressible materials, t versus V2 is linear. For compressible


materials such as water and wastewater sludges, the permeability and porosity are
dependent upon the pressure difference or liquid flow rate (in other words, the
permeability and compressibility are dependent upon the solids volume fraction).

Civil engineers (Gibson, et al., 1967, Terzaghi and Peck, 1967) use Darcy’s
law to model the one-dimensional consolidation of saturated soils, by assuming that
the soils do not compress much and their properties are constant throughout the
process. Tiller and Shirato, 1964, use conventional filtration theory to describe cake
formation and Terzaghi’s model to describe cake compression during piston filtration.
Both of these theories rely on volume-average properties and a priori knowledge of
the material property functional forms. Despite its shortcomings, it is shown here that
the analytical nature of the modified-Darcy’s law (equation 2.1.5) is useful to measure
the membrane resistance from transient filtration data.

The understanding of sedimentation processes has developed in parallel to


filtration theory. Stokes, 1845, derived an expression for the terminal settling velocity
of a single spherical particle in laminar flow. However, the drag on the particle is
dependent on turbulence and shape as well as the presence of other particles.
Thickeners and centrifuges have often been modelled using the approach of Kynch,
1952, which is derived from the hindered settling behaviour of suspensions. It fails to
consider the compression or consolidation of the settled material. As well as this,
single-point measurements are used to describe material properties that change by
many orders of magnitude.

Buscall and White, 1987, developed a fundamentally rigorous


phenomenological model of dewatering in which networked particulate suspensions
are described as yield stress materials – the network has a certain strength in

28
compression, called the compressive yield stress, Py, which must be exceeded for the
structure to fail and dewater. Py varies with the number, strength and arrangement of
inter-particle bonds and is therefore a function of solids volume fraction, φ (see Figure
2.1.1(a)). For flocculated materials, the concentration at which a network forms is
given by the gel point solids volume fraction, φg. There is no physically measurable
network strength at concentrations below φg. Py(φ) → ∞ as φ approaches close
packing.

The hydrodynamic drag between the particles and the fluid is described by the
hindered settling function, R. Since the drag changes dramatically with concentration,
R is a highly non-linear function of φ, as shown in Figure 2.1.1(b). At infinite dilution
(that is, as φ → 0), R(φ) → λ/Vp, where λ is the Stoke’s drag coefficient for a single
particle of volume Vp. At concentrations up to φg, R(φ) represents the increased drag
due to the presence of surrounding particles. Beyond φg, R(φ) is the resistance to flow
through a porous network. R(φ) → ∞ as φ → 1.
Hindered Settling Function, R (φ )
Compressive Yield Stress, Py (φ )

φg φg

Solids Volume Fraction, φ (v/v) Solids Volume Fraction, φ (v/v)


(a) (b)

Figure 2.1.1: Volume fraction dependent material characteristics: (a) Compressive yield stress,
Py(φ); (b) Hindered settling function, R(φ)

Py(φ) and R(φ) are usually presented as power-law and exponential functions
(Auzerais, et al., 1990, Landman and White, 1994, Eberl, et al., 1995, Eckert, et al.,
1996, Miller, et al., 1996, Channell and Zukoski, 1997, Green and Boger, 1997). The
power-law functions predominantly used here are given by equations 2.1.6 and 2.1.7:

29
 0 φ < φg 
   
Py (φ ) =   φ 
p2

   …(2.1.6)
 p1  φ  − 1 φ ≥ φ g 
  g   

R (φ ) = r1 (1 − φ )r2 …(2.1.7)

p1, p2, r1 and r2 are fitting parameters to be determined experimentally.

The dewatering model developed by Buscall and White, 1987, balances the
forces acting upon a volume element of suspension to obtain the mass and momentum
conservation equations for the solid and liquid phases, and uses a kinetic equation to
describe the collapse of the solid network. The conservation equations are applicable
to all two-phase dewatering operations since Py(φ) and R(φ) are fundamental state
properties of the material and describe suspension dewatering behaviour over a large
range of concentrations. The theory has been used to model one-dimensional batch
(Howells, et al., 1990, Lester, et al., 2005) and continuous (Landman, et al., 1988)
settling of a suspension under gravitational acceleration, constant pressure filtration
(Landman, et al., 1991, Landman and White, 1994, 1997) and gravity filtration
(Martin, 2004). Buscall and White, 1987, also modelled the equilibrium state of a
consolidating suspension in a centrifugal tube (that is, a steady-state Cartesian
coordinate problem with centrifugal acceleration). Filtration, sedimentation and
centrifugation techniques for determining the local material properties have been
developed based upon these theories (Green, et al., 1998, de Kretser, et al., 2001,
Lester, et al., 2005).

A very similar approach is used by Bürger and co-workers (Bürger and


Concha, 1998, Bürger, 2000) to model the solid-liquid separation of flocculated
suspensions, in which the compressibility is described by the effective solid stress
function, σe(φ), and the permeability by the Kynch batch flux density function, fbk(φ).
σe(φ) is the same as Py(φ), while fbk(φ) is inversely related to R(φ).

Landman and White, 1994, first introduced a common parameter that emerges
from the mathematical development of one-dimensional models: the solids diffusivity,
D(φ). An equivalent parameter, a(φ), emerges from the work of Bürger. As a

30
diffusion coefficient, a higher D(φ) indicates a material that dewaters faster (de
Kretser, et al., 2001), and is used as an overall dewatering parameter. D(φ) is shown
here to be analogous to the coefficient of consolidation, cv, used by geotechnical
engineers.

A key conceptual difference between the classical filtration approach and


phenomenological modelling is seen in the definition of specific cake resistance and
solids pressure from the operating conditions of the filtration process rather than as
general “material properties” defined independently of their application. Despite
these conceptual differences, the underlying fundamental quantities can be directly
related to Py(φ) and R(φ), as illustrated by Landman and White, 1994.

In the work presented in this chapter, the conservation of mass and momentum
equations were developed in vector notation, which were used as the basis for all the
modelling. The vector equations were applied to the transient one-dimensional
Cartesian coordinate problem of piston-driven filtration, since this was the main
experimental technique employed here and similar formulations were used for
modelling plant-scale devices. Models for constant-pressure piston-driven filtration
with and without membrane resistance were formulated and algorithms written to
solve the problem. These were used to show that the filtration behaviour of sewage
sludges is described by phenomenological theory if the sludges have certain
characteristics, and appropriate scaling behaviour was derived. Darcy’s law for
pressure filtration was manipulated to give a measure of membrane resistance from
online measurements of filtrate data. Finally, soil consolidation theory was compared
to filtration theory.

31
2.2 Sedimentation-Consolidation Equations

The rheological model developed by Buscall and White, 1987, and reviewed
by Landman and White, 1994, balances the hydrodynamic (drag), hydrostatic
(upthrust), network pressure and acceleration (gravitational or centrifugal) forces
acting upon a volume difference of solids (see Figure 2.2.1). The network pressure is
comprised of the principal components of the stress tensor (xx, yy and zz) – the other
components (xy, xz and yz) are assumed to be small here, but must be included for
solid-liquid separation that incorporates shear processes.

Network Pressure

Interphase Drag

Solids Buoyancy

Figure 2.2.1: Volume element of networked suspension, illustrating the forces involved

The conservation of momentum equations for the solid and liquid phases are
given by equations 2.2.1 and 2.2.2 respectively:

( )
− φR(φ ) u p − u f − φ∇p f − ∇p p + φρ p g = 0 …(2.2.1)

φR(φ )(u p − u f ) − (1 − φ )∇p f + (1 − φ )ρ f g = 0 …(2.2.2)

up – uf is the local velocity of the particle relative to the fluid, ρp and ρf are the
particle and fluid densities respectively, and g is the acceleration vector. Eliminating
∇pf from equations 2.2.1 and 2.2.2 gives:


φ
1−φ
( )
R (φ ) u p − u f − ∇p p + φ∆ρg = 0 …(2.2.3)

∆ρ is the density difference between the solid and liquid phases. The first
term in equation 2.2.3 represents the drag due to the fluid-solid interaction; the second
term represents the strength of the particulate network in response to an applied
pressure, while the third term represents any gravitational force.

32
The relative velocity is eliminated from equations 2.2.1 and 2.2.2 to give the
relationship between the local fluid and pressure gradients:

∇p f + ∇p p = ρ f g + φ∆ρg …(2.2.4)

Eliminating the particle pressure gradient from equations 2.2.1 and 2.2.2 gives
an analogous form of the classical Darcy-Shirato equation in vector notation
(Landman and White, 1994):

∇p f − ρ f g =
1−φ
φ
(
R(φ ) u p − u f ) …(2.2.5)

The conservation of mass equations for the particle and fluid phases (assuming
incompressibility of both phases) are given by equations 2.2.6 and 2.2.7 respectively:

∂φ
∂t
(
= −∇. φu p ) …(2.2.6)

∂ (1 − φ )
∂t
(
= −∇. (1 − φ )u f ) …(2.2.7)

Adding these equations gives the overall conservation of mass:

(
∇. φu p + (1 − φ )u f = 0 ) …(2.2.8)

In order to ensure irreversibility, a kinetic equation describing the collapse of


the particulate network due to excess pressure is required:

 0 p p < Py (φ )
Dφ  
=  …(2.2.9)
Dt 
( )[
κ φ , p p p p − Py (φ ) ] p p ≥ Py (φ )

κ(φ,pp) is the dynamic compressibility and Dφ/Dt is the material derivative of


φ, which is defined as:

Dφ ∂φ
Dt
=
∂t
+ u p .∇φ =
∂φ
∂t
(
+ ∇. φu p − φ∇.u p ) …(2.2.10)

Substituting equations 2.2.6 and 2.2.10 into equation 2.2.9 gives:

33
 
 0 p p < Py (φ )
 
∇.u p =   …(2.2.11)
 κ φ, pp ( ) 
−
φ
[
p p − Py (φ ) ] p p ≥ Py (φ )
 

Equation 2.2.11 states that the network will collapse only when pp exceeds
Py(φ) and that compression is irreversible (that is, the material is assumed to be
inelastic). If the drainage of the suspending fluid is rate-determining rather than the
breaking and/or reformation of inter-particle bonds (that is, κ(φ) is of order (φ0/ηf) or
greater, which is most likely), then the particle pressure is always less than or equal to
the yield stress: pp ≤ Py(φ) (Buscall and White, 1987, Landman, et al., 1988). Thus,
the network only collapses when pp > Py(φ), and is then equal to Py(φ) unless pp <
Py(φ).

Providing that κ(φ) is large, it is redundant and only Py(φ) and R(φ) are needed
for modelling purposes. Several authors (Dixon, 1980, Tiller and Leu, 1980, Tiller
and Khatib, 1984) make use of a constitutive equation such as Py linking pp to φ, but
its use for flocculated materials assumes large κ(φ). The irreversibility condition is
only required for systems where pp < Py(φ), such as gravitational (Martin, 2004) and
centrifugal filtration.

The geometry of the device being modelled determines the coordinate system
for the conservation of mass and momentum equations. The one-dimensional
Cartesian equations used for filtration and settling are now presented, followed by the
formulation and solution of constant-pressure piston-driven filtration. One and two-
dimensional models of centrifugation using radial and helical coordinate systems
respectively are given in the modelling chapters.

One-Dimensional Cartesian Coordinates

The vectors up, uf and g are converted to one-dimensional scalars, up, uf and g
assuming gravitational rather than centrifugal thickening:

u p = −u p ( z , t )ẑ u f = −u f ( z , t )ẑ g = − gẑ …(2.2.12)

34
Converting equation 2.2.8 to z and integrating gives the bulk flow, q(t), which
is a constant with respect to z.

φu p + (1 − φ )u f = q(t ) …(2.2.13)

For batch thickening, q(t) is zero. For filtration or continuous thickening, q(t)
is given by the specific flowrate of filtrate, dV/dt. Rearranging equation 2.2.13 gives
the solids velocity relative to the fluid velocity:

u p − q (t )
up −u f = …(2.2.14)
1−φ

Converting equation 2.2.3 to z and using equation 2.2.14 to eliminate uf(z,t)


gives:

∂p p
φ
(
R (φ ) u p − q (t ) − ) − φ∆ρg = 0 …(2.2.15)
(1 − φ )2 ∂z

Equation 2.2.4 in one-dimension becomes:

∂p f ∂p p
∂z
+
∂z
(
= − ρ f + φ∆ρ g ) …(2.2.16)

If κ(φ) is large and the applied pressure is constant or increasing, pp = Py(φ).


Substituting this into equation 2.2.15 and combining with equation 2.2.6 eliminates
up(z,t). The resulting mixed hyperbolic-parabolic convection-diffusion equation is the
overall governing equation for one-dimensional consolidation in filtration and
thickening:

∂φ ∂ 
=  D(φ )
∂φ
+ φq (t ) + φ∆ρg
(1 − φ )2 
 …(2.2.17)
∂t ∂z  ∂z R (φ ) 

The solids diffusivity, D(φ), is defined by:

dPy (φ ) (1 − φ )2
D(φ ) = …(2.2.18)
dφ R (φ )

35
The last term in equation 2.2.17 is combined by Bürger and Concha, 1998, as
fbk(φ):

f bk (φ ) = φ∆ρg
(1 − φ )2 …(2.2.19)
R(φ )

The common initial condition for transient thickening and filtration models is
that the sludge is uniform at concentration φ0:

φ ( z ,0 ) = φ 0 …(2.2.20)

The boundary conditions and solution methods for equations 2.2.16 and 2.2.17
are dependent upon whether the model is a thickener or a filter. For filtration models,
they depend on whether:

- The gravitational settling time-scale competes with the filtration time-scale;

- The membrane resistance, Rm, is assumed to be zero;

- The applied pressure is constant or variable;

- The initial suspension is un-networked or networked (that is, φ0 < φg or φ0 >


φg); and

- The filtration cell or cavity is fixed or variable. For piston- or air-driven


filtration, the top of the sample represents a moving boundary condition, while
for the fill-stage of plate-and-frame filtration (see Chapter 5), the cavity width
is fixed and the plane of symmetry at the centre of the cavity is the boundary.

The remainder of this chapter outlines the formulation of piston-driven


filtration with constant applied pressure, since this was the predominant experimental
method used.

36
2.3 Filtration Theory

A schematic of one-dimensional constant pressure piston-driven filtration for


the initial sample un-networked (φ0 < φg) case is presented in Figure 2.3.1. The
gravitational settling effects have been assumed to be negligible, as discussed below.

t=0 0 < t < tc t > tc t→∞


∆P
∆P
∆P

h0
h(t)
φ0 h(t)
φ0
h∞
zc(t) φg < φ(z,t) < φ∞
φg < φ(z,t) < φ∞ φ∞
0 0 0 0

V(t) V(t)
(a) (b) (c) (d)

Figure 2.3.1: Schematic of one-dimensional constant pressure filtration (φ0 < φg); (a) Initial
condition; (b) Cake formation; (c) Cake compression; (d) Equilibrium condition

A homogeneous particulate suspension is loaded into a pressure cell (Figure


2.3.1(a)). The walls and piston are assumed to be impervious and frictionless, while
the membrane is impervious to the solid phase. It is assumed that the individual
particles are small compared to the filtration cell. h0 is the initial piston height. The
origin (z = 0) is defined at the membrane.

When a pressure, ∆P, is applied, the piston height, h(t), reduces, filtrate is
exuded and a cake of height zc(t) begins to build up from the membrane (Figure
2.3.1(b)). Only the case of constant ∆P is considered here. V(t) is the cumulative
specific filtrate volume with time, and is related to h(t) through the overall
conservation of volume:

V (t ) = h0 − h(t ) …(2.3.1)

37
The one-dimensional solid-liquid separation equations (2.2.16 and 2.2.17) are
used to describe the volume fraction distribution with time, φ(z,t). If the gravitational
settling rate is comparable to the filtration rate, a zone of clear liquor will form
directly below the piston and a compression wave may form between the cake and the
suspension. If the ratio of the filtration and sedimentation time-scales (Tfilt and Tsed
respectively) is small, the gravitational terms are approximately zero. Landman and
White, 1994, give the ratio as:

∆ρgφ0 h0 R(φ ∞ )  1 − φ0 
2
T filt
≈   …(2.3.2)
Tsed ∆P R (φ0 )  1 − φ ∞ 

It is assumed that this ratio is small in the following derivation of piston-


driven filtration. Integrating equation 2.2.16 shows that pf and pp add to comprise the
applied pressure, ∆P, and that their contributions vary with time and position:

p f ( z ,t ) + p p ( z ,t ) = ∆P …(2.3.3)

Assuming that the piston is impervious, up(h(t),t) is equal to uf(h(t),t).


Therefore q(t) is given by the piston speed, dh/dt. Substituting this into equation
2.2.17 and assuming no gravitational effects gives the filtration governing equation:

∂φ ∂  ∂φ dh 
=  D(φ ) − φ  …(2.3.4)
∂t ∂z  ∂z dt 

Since D(φ) is zero for φ < φg, material above the cake remains at constant
concentration, φ0. The cake has a volume fraction distribution ranging from φg at the
top of the cake to a maximum at the membrane, φ(0,t), which is dependent upon the
applied pressure and the membrane resistance, Rm. Rm is treated in a Darcian manner
(Landman, et al., 1991) such that the fluid velocity, -dh/dt, is dependent upon the
hydraulic resistance, Rm, and the fluid pressure drop across the membrane, ∇pfmem:

dh ∇p f mem
− = …(2.3.5)
dt Rm

38
Rm is defined here by the fluid viscosity, ηf, membrane width, lm, and
membrane permeability, km, and is assumed to be independent of pressure.

l mη f
Rm = …(2.3.6)
km

The membrane resistance can also be described in units of reciprocal length as


the ratio of lm and km, excluding ηf. Assuming that the membrane is thin and fully
saturated, the fluid pressure gradient is the difference between the absolute fluid
pressure at z = 0 and atmospheric pressure. If pf is defined as gauge pressure:

∇p f = p f (0 ,t ) …(2.3.7)
mem

From equation 2.3.3, pf is the difference between ∆P and pp. Thus, the volume
fraction at the membrane is given by the solution of equation 2.3.8:

Py [φ (0 ,t )] = ∆P + Rm
dh
…(2.3.8)
dt

Thus, providing that Rm is not zero, φ(0,t) is dependent upon dh/dt, which is
dependent on the solution of the governing equation. The concentration gradient at
the membrane is derived from equation 2.2.15, assuming that up(0,t) = 0:

∂φ φ (0 ,t ) dh
= …(2.3.9)
∂z 0 D[φ (0 ,t )] dt

Since the piston is impervious, the moving boundary condition at h(t) is:

∂φ
=0 …(2.3.10)
∂z h(t )

The rate of growth of the shock at zc(t) during cake formation is given by
integrating equation 2.3.4 from zc- to zc+:

dz c
=−
( )
D φ g ∂φ
+
dh
dt ( )
φ g − φ0 ∂z z c− dt
…(2.3.11)

39
Rearranging equation 2.3.11 gives the concentration gradient at zc(t):

∂φ φ g − φ0  dz c dh 
=−  − 
∂z z −
c
( )
D φ g  dt dt 
…(2.3.12)

The overall conservation of volume is given by the integral of φ(z,t):

h(t )
∫ φ ( z ,t )dz = φ 0 h0 = φ h(t ) …(2.3.13)
0

<φ> is the average solids volume fraction. During cake formation, φ(z,t) for z
> zc(t) is constant at φ0, thus equation 2.3.13 reduces to:

z c (t )
∫ φ ( z ,t )dz = φ 0 (h0 − h(t ) + z c (t )) …(2.3.14)
0

There is an internal continuous boundary at φ0 for the φ0 > φg, Rm = 0 case, but
no cake formation stage for the situation with Rm ≠ 0 since the network already spans
the filtration cell such that zc(t) is equal to h(t) for t > 0 (Landman, et al., 1991).

The cake forms until it meets the piston at time tc (h(tc) = zc(tc) = hc). After tc,
the cake rearranges and compresses (see Figure 2.3.1(c)) until pp approaches ∆P (and
pf is zero) and φ approaches the equilibrium volume fraction, φ∞ (see Figure 2.3.1(d)):

Py (φ ∞ ) = ∆P …(2.3.15)

The equilibrium piston height, h∞, is given by equation 2.3.13:

φ h
h∞ = 0 0 …(2.3.16)
φ∞

V∞ is the equilibrium specific filtrate volume, and is given by:

 φ 
V∞ = h0  1 − 0  …(2.3.17)
 φ∞ 

40
A range of analytical, numerical and approximate methods to solving the
problem exists in the literature, depending on the complexity of the problem. Various
algorithms were developed in this work, as outlined in Figure 2.3.2.

Rm = 0 Rm ≠ 0

φ0 < φ* φ0 ≥ φ*

φ0 < φg φ0 ≥ φg

Exact Similarity
Numerical Solution
(t ≤ tc)

Linear Approximation Finite Difference Iterative Runge-Kutta


Analytical Solution Numerical Solution Numerical Solution

Figure 2.3.2: Algorithms for Constant-Pressure Piston-Driven Filtration

The formulation for Rm = 0 followed the work of Landman and White, 1997,
where an approximation to D(φ) gives analytical solutions for cake formation and
compression when φ0 < φ*, where φ* is a shifted gel point such that the cake is in
formation until φ(h(tc),tc) = φ*. When φ0 ≥ φ*, a finite difference numerical method
was used. An exact solution exists for φ0 < φg up until t = tc, after which the finite
difference method was used. An iterative 4th – 5th order Runge-Kutta numerical
scheme was required for Rm ≠ 0, based upon Landman, et al., 1991. The linear
approximation was used as the basis for experimental characterisation due to its
simplicity, while the numerical models were used to validate new analysis techniques
and to verify experimental results. The numerical solution when Rm ≠ 0 was used to
model the effects of membrane resistance. Apart from the four phenomenological
algorithms presented in Figure 2.3.2, an approach to on-line measurement of Rm using
conventional filtration theory was developed and validated using the numerical model.

41
2.4 Constant-Pressure Piston-Driven Filtration
without Membrane Resistance

For the filtration of low permeability materials at high pressures with clean
filter membranes, the resistance of the cake, Rc, is almost immediately greater than
Rm, and Rm can be ignored. From equations 2.3.8 and 2.3.15, the boundary condition
at the membrane in such circumstances is given by φ∞:

φ (0 ,t ) = φ ∞ …(2.4.1)

This indicates that all solutions for Rm = 0 have an internal moving boundary
condition, given that the initial condition (equation 2.2.20) contradicts the boundary
condition at the membrane (equation 2.4.1).

Landman and White, 1997, simplify the problem by scaling the parameters z,
zc(t), h(t) and t to Z, Zc(T), H(T) and T, where:

z
Z=
h0
z (t )
Z c (T ) = c
h0
h(t ) …(2.4.2)
H (T ) =
h0

tD (φ ∞ )  φ ∞
2

T=  
h02  φ 0 

The time scaling is dependent on the properties of the cake as opposed to the
initial suspension, as used in earlier models (Landman, et al., 1991). Z is converted to
a material co-ordinate, w(Z,T), given by:

1 Z
w(Z ,T ) = ∫ φ (Z ,T )dZ …(2.4.3)
φ0 0

42
w(0,T) = 0 and w(H(T),T) = 1, such that the range 0 ≤ w ≤ 1 encompasses the
entire filtration cell and the moving external boundary condition at the piston is
removed. The void ratio, e(w,T) is substituted for φ(Z,T), where e(w,T) is defined as:

e(w,T ) =
1
−1 …(2.4.4)
φ (Z ,T )

Using equations 2.4.2 to 2.4.4 in conjunction with the overall conservation


equation (equation 2.3.13), equation 2.3.4 is restated as a simpler second order
diffusion equation:

∂e ∂  ∂e 
=  ∆(e )  …(2.4.5)
∂T ∂w  ∂w 

where

φ 2 D(φ )
∆ (e ) = …(2.4.6)
φ ∞2 D(φ ∞ )

The initial condition, in terms of void ratio, is:

e(w,0 ) = e0 =
1
−1 …(2.4.7)
φ0

The boundary condition at the membrane is:

e(0, T ) = e∞ =
1
−1 …(2.4.8)
φ∞

From equation 2.3.10, the boundary condition at w(H(T),T) = 1 is:

∂e
=0 …(2.4.9)
∂w 1

H(T) is given by the overall conservation of volume (equation 2.3.13):

1
1 + ∫ e(w,T )dw
H (T ) = 0
…(2.4.10)
1 + e0

43
This change of variables has not removed the internal moving boundary
condition (equations 2.4.7 and 2.4.8 still contradict each other). For cake formation in
the φ < φg case, the boundary condition at the top of the cake at wc(T) (equation
2.3.12) becomes:

∂e ( )
e0 − e g dwc
=
∂w w (T )
c
( )
∆ eg dT
…(2.4.11)

Several solutions for the Rm = 0 problem are used in this work and are outlined
here. Landman and White, 1997, give an exact solution for cake formation for φ0 <
φg, and an analytical solution based upon a linear approximation to ∆(e), while the full
equations are solved here using a finite difference method (FDM).

44
2.4.1 Initial Suspension Un-networked (φ0 < φg, t ≤ tc, Rm = 0)

An exact similarity solution for e(w,T) exists for cake formation (that is, for T
≤ Tc) in the φ0 < φg case (Landman and White, 1997), providing that ∆(eg) > 0. The
similarity variable takes the form of X(w,T):

T
X (w,T ) =
w
=w c …(2.4.12)
wc (T ) T

such that:

 E ( X ) w ≤ wc (T )
e(w,T ) =   …(2.4.13)
 e0 w > wc (T )

Therefore, wc(T) varies with the square-root of T (and therefore, so does H(T)
and Zc(T)). Transforming equation 2.4.5 gives a non-linear ordinary differential
equation for E(X) with appropriate boundary conditions:

 dE 
∆ (E ) dX 
X dE d
− = …(2.4.14)
2Tc dX dX

E (0 ) = e∞ …(2.4.15)

E (1) = e g …(2.4.16)

dE e0 − e g
=
dX 1 2Tc ∆ e g ( ) …(2.4.17)

E(X) is determined numerically by solving equation 2.4.14 from X = 1 to X = 0


for successive estimates of Tc until equation 2.4.15 holds. H(Tc) = Hc is then given by
the cumulative void ratio:

 1 
H c = φ0 1 + ∫ E ( X )dX  …(2.4.18)
 0 

45
2.4.2 Linear Approximation (φ0 < φ*, Rm = 0)

The non-linear problem becomes easier to solve if ∆(e) is approximated by a


step function of unit height, ∆eff:

1 e ≤ e * 
∆eff = θ (e * −e ) =   …(2.4.19)
0 e > e * 

e* is effectively a shifted gel point, such that the filtration process is in cake
formation until e(1,T) = e*. The choice of e* ensures that the approximation is scaled
so that the value at e∞ and the area under the ∆(e) curve from e∞ to e0 are preserved.
The suggested form is (Landman and White, 1997):

e0
e* = e∞ + ∫ ∆ (e )de =
1
−1 …(2.4.20)
e∞ φ*

Thus, the linear approximation is dependent upon both e0 and e∞, and is
therefore applicable for φ0 < φ*. Substituting ∆eff for ∆(e) in equation 2.4.5 gives a
simpler diffusion equation:

 ∂ 2e 
 2 e ≤ e *
∂e  ∂w 
=  …(2.4.21)
∂T 
0 e > e *
 
 

The solution of equation 2.4.21 subject to equations 2.4.7 to 2.4.9 shows that
there are two regimes of filtration behaviour – cake formation and cake compression.
For φg < φ0 < φ*, cake formation represents the time taken for the shock initially at the
membrane to propagate to the piston (that is, φ(H,Tc) = φ*). The solution during cake
formation is found by solving equation 2.4.22 for the dimensionless parameter, α (not
to be confused with the specific resistance of filtration, αv):

46
e * −e ∞
e0 − e *
= π α exp α 2 erf (α ) ( ) …(2.4.22)

α is then used to calculate Tc and Hc:

1
Tc = …(2.4.23)
4α 2

From equation 2.4.10:

e − e*
Hc = 1 − 0
1 + e0
exp α 2 ( ) …(2.4.24)

The piston height during cake formation varies with the square root of time:

H (T ) = 1 − (1 − H c )
T
; T ≤ Tc …(2.4.25)
Tc

Rearranging equation 2.4.23 shows that, during cake formation, t varies


linearly with V2:

2 2
V  Tc  φ 0 V 
t =   =   ; t ≤ tc …(2.4.26)
β  D(φ ∞ )  φ ∞ (1 − H c ) 

β is the constant of proportionality between V and t½. The solution to equation


2.4.21 during cake compression is given as a Fourier series:

 ∞ An − (n + 1 2 ) π 2 (T − Tc ) 
2
H ( T ) = φ0 1 + ∑  ; T ≥ Tc
 n = 0 (n + 1 2 )π
exp

…(2.4.27)

where An is given by:

4(e * −e∞ ) α
 πz 
cos (n + 1 2 )  dz
)erf (α ) ∫
−z2
An = exp …(2.4.28)
π
3
2
(n + 12 0  α

47
Truncating the series in equation 2.4.27 to the first term (applicable for T – Tc
>> 4/9π2) and removing the scalings shows that the filtration time (to this
approximation) varies logarithmically with filtrate volume during cake compression:

t = E1 − E 2 ln(V∞ − V ) …(2.4.29)

where

h02φ02  4  2 A0φ0 h0 
E1 = Tc + ln  …(2.4.30)
D(φ ∞ )φ ∞2  π2  π 

4 h02φ02
E2 = …(2.4.31)
π 2 D(φ ∞ )φ ∞2

Thus t can be found as a function of V and is dependent only upon parameters


determined from φ0, h0, ∆P, Py(φ), and D(φ). For a given filtration time, tf, the extent
of filtration is found by solving equation 2.3.15 for φ∞, calculating Tf, ∆(e), e0, e∞ and
e* from equations 2.4.2, 2.4.6 to 2.4.8 and 2.4.20 respectively, solving equation 2.4.22
for α and then calculating Tc from equation 2.4.23. If Tf < Tc, the process is in cake
formation and H(Tf) is given by equation 2.4.25. If Tf > Tc, the process is in cake
compression and H(Tf) is given by equation 2.4.27.

Application To Experimental Characterisation

The compressibility and permeability of particulate systems are typically


measured using a pressure filtration cell by recording the volume of filtrate with time
(de Kretser, et al., 2001). As indicated by the linear approximation, t is quadratic with
V during cake formation (equation 2.4.26), followed by a logarithmic dependency (to
first order) during cake compression (equation 2.4.29). An example of a typical
piston-driven filtration experimental curve is shown in Figure 2.4.1, with a guideline
to illustrate the quadratic behaviour.

48
Time, t (s)

Cake Cake
Formation Compression

2 2 2
(Specific Filtrate Volume) , V (m )

Figure 2.4.1: Example of experimental results for traditional filtration

The equilibrium height, h∞, measured over a range of pressures gives Py as a


function of φ∞ (Green, et al., 1998). Landman, et al., 1999, showed that, by assuming
that the solids velocity in the compact cake is negligible compared to the piston
velocity, β2, measured over a range of pressures, is combined with φ∞ to give R(φ),
according to equation 2.4.32:

2  1 
R (φ ∞ ) = (1 − φ ∞ )2
1
 − …(2.4.32)
dβ 2  φ 0 φ ∞ 
d∆P

D(φ) is then determined from equation 2.2.18. Alternatively, D(φ) can be


determined directly from β2 by combining equation 2.4.32 with equation 2.2.18:

−1
1 dβ 2  1 1 
D(φ ∞ ) =  −  …(2.4.33)
2 dφ ∞  φ 0 φ ∞ 

These equations give two alternatives for determining D(φ∞) from filtration
data, but both require differentiation of β2(∆P) or β2(φ∞). Landman, et al., 1999, also

49
derive an iterative method for determining D(φ∞) from β2(φ∞) without the need for
differentiation, but this method was not used here.

Water treatment sludges show filtration behaviour typified by long cake


formation times (up to 85% of the total time) followed by short compression times,
allowing the use of β2 methods. Indeed, the extraction of the material parameters of a
wide range of particulate suspensions from the minerals, sediments, pigments and
ceramics industries (many of them flocculated) usually relies on this behaviour. The
term ‘traditional’ is used in this work to describe relatively long cake formation times.

A less common method of extracting filtration data is curve fitting the end of
the compression region to give results of φ∞ and D(φ∞) for a single pressure filtration
run, based upon the results of Landman and White, 1997. Py(φ) and D(φ) are then
deduced from a series of single-pressure runs (an algorithm is presented in the
methods chapter). This method is therefore useful for materials that show little or no
cake formation, such as wastewater treatment sludges. This is similar to Casagrande’s
graphical log-time method for soil consolidation (Casagrande and Fadum, 1942),
although fundamental differences exist since the curve fitting method allows for large
strains.

Stepped-Pressure Filtration

The linear approximation was extended by Scales, et al., 2001, to model


stepped-pressure filtration. Filtration at the initial pressure proceeds as given by the
linear approximation (equations 2.4.26 and 2.4.29). The cake height at the end of the
time at the lowest pressure is determined, and depends on whether the process is in
cake formation or cake compression. An assumption is made with the application of
the next pressure that the existing cake is uniform and that it undergoes a
rearrangement until the solids concentration corresponds to the new applied pressure.
Standard pressure filtration then proceeds at the new applied pressure until the next
pressure step, and so on. The stepped-pressure model was used here to validate
experimental data (see Section 4.1.1) and to provide a basis for modelling the initial
ramping pressure of plate-and-frame filtration (see Section 5.2.1).

50
Scaling Behaviour

Theoretical manipulation of the time-scalings used by Landman and White,


1997, gives a method for scaling the initial piston height and solids concentration
dependencies from filtration data to make direct qualitative comparisons of results.
From the time scaling given in equation 2.4.2, the filtration time, tf, is given by:

2
h02  φ 0 
tf =   T f …(2.4.34)
D(φ ∞ )  φ ∞ 

where Tf is the dimensionless time to filtration. Tf is independent of h0, therefore tf


scales proportionally with h02. Experimental results presented in Chapter 4 validate
this. However, Tf is dependent on φ0. From equation 2.4.26, Tf during cake formation
is given in terms of average volume fraction by:

2
Tc  φ 
Tf = 1 − 0  …(2.4.35)
(1 − H c )  φ 
2 

Substituting this into equation 2.4.34 and rearranging gives:

2
tf  φ 
= c (φ 0 ) 1 − 0  …(2.4.36)
φ 02 h02  φ 

where

Tc
c (φ 0 ) = …(2.4.37)
φ ∞2 D (φ ∞ )(1 − H c )2

c(φ0) contains all the material property dependencies. If the filtration results
for samples with differing φ0 values are to be qualitatively compared, the following
proportionality must exist:

 tf  c(φ )   
2
  0  ∝  1 − φ0 
 φ 2h2
 0 0
 c φ (
 0 ,ref
)  
  φ 
…(2.4.38)

where φ0,ref is a reference concentration and c(φ0) is scaled to c(φ0,ref):

51
c(φ o )
=
Tc  φ ∞ ,ref

2
(
 D φ ∞ ,ref

)  1 − H c ,ref  2
(
c φ o ,ref )Tc ,ref  φ ∞ 
 D (φ ∞ )  1− H
 c


…(2.4.39)

The left hand side of proportionality 2.4.38 represents the scaled time, while
the squared relationship on the right hand side represents a dimensionless form of V2.
Thus, if c(φ0)/c(φ0,ref) is known, direct comparisons can be made between filtration
profiles of differing h0 and φ0. Empirical solutions for c(φ0)/c(φ0,ref) for a typical
sewage sludge are given in Section 4.2.1. The scaling method was validated using the
single pressure filtration testing results of activated sewage sludge, and then used to
qualitatively compare different sludges.

52
2.4.3 Finite Difference Method (φ0 ≥ φg and φ0 < φg, t ≥ tc; Rm = 0)

Finite difference methods (FDM) are common numerical techniques used to


solve non-linear partial differential equations (Smith, 1985). Bürger and co-workers
use it extensively in their models of solid-liquid separation (Bürger, et al., 2001), but
special care must be taken to resolve discontinuous moving boundaries and to ensure
irreversibility. Runge-Kutta numerical techniques were generally used within this
work, but the FDM was useful for modelling piston-driven cake compression with Rm
= 0 as there are no internal discontinuous boundaries and e(0,T) is independent of
dh/dt.

The FDM is applied to equation 2.4.5, with ∆(e) being fully explicit as δ(e):

∂e ∂ 2δ (e )
= …(2.4.40)
∂T ∂w 2

where

e
δ (e ) = ∫ ∆ (e )de …(2.4.41)
e0

Equation 2.4.40 is converted into N discrete time steps and J discrete lengths
of size ∆T and ∆w respectively using a central difference approximation in ∆w and a
forward difference approximation in ∆T:

e n +1, j = e n , j +
∆T
[δ (en , j+1 )− 2δ (en , j )+ δ (en , j−1 )] …(2.4.42)
∆w 2

where en,j is the void ratio at time n and position j. The initial and boundary
conditions (equations 2.4.7 to 2.4.9) in discrete form become

e0 , j = e0 …(2.4.43)

en ,0 = e∞ …(2.4.44)

53
∆T
en+ 1 ,J = en ,J − 2 [δ (en ,J ) − δ (en,J −1 )] …(2.4.45)
∆w 2

The factor of 2 in equation 2.4.45 arises due to the central difference


approximation of equation 2.4.9 rather than the backward approximation, as is
sometimes used (Bürger, et al., 2001). Equations 2.4.40 to 2.4.45 give a complete
description of one-dimensional constant pressure filtration – if all the values of en,j are
known, the equation can be solved for en+1,j.

The piston height at time n, hn, is given by the void ratio distribution (from
equation 2.4.10), which gives the filtrate volume, Vn, and the average volume fraction,
<φ>n:

 1 J 
hn = h0φ 0 1 + ∑ e nj  …(2.4.46)
 J j =1 

Vn = h0 − hn …(2.4.47)

h0φ 0
φ n
= …(2.4.48)
hn

The accuracy and required computational time of the FDM depends upon the
size of ∆w and the complexity of δ(e). J is chosen arbitrarily, but is usually greater
than one hundred for reasonable accuracy. The explicit method is convergent for time
steps less than the maximum, ∆Tmax (Smith, 1985):

∆w 2
∆Tmax = …(2.4.49)
2 max ∆ (e )

∆T is chosen such that ∆T ≤ ∆Tmax. It is relatively simple to program a


computer to perform these calculations for each time step, although the solution
requires much more computational time than the linear approximation analytical
solution. Since the latter approximates the material characteristics, it is expected to be
less accurate than the FDM in certain circumstances, as illustrated in the next section.

54
2.4.4 Non-Traditional Filtration

Background

Compared with the filtration results for ‘traditional’ materials (see Figure
2.4.1), very different filtration profiles are observed for many sludges containing high
molecular weight and cross-linked biomolecules, including wastewater treatment
sludges (Tiller and Kwon, 1998, Anderson, et al., 2002, Scales, et al., 2003). These
materials are sometimes classed as ‘super-compactable’ (Tiller and Li, 2000). In
constant pressure filtration, such sludges tend to show little or no cake formation
followed by a long compression phase, as illustrated in Figure 2.4.2. Such behaviour
is termed ‘non-traditional’ here.
Time, t

Cake
Compression

Cake
Formation

2 2
(Filtrate Volume) , V

Figure 2.4.2: Example of experimental results for non-traditional filtration behaviour

For the data shown in Figure 2.4.2, it is difficult to extract permeability or


compressibility parameter information from a single pressure run. Traditional
filtration methods of using the slope of the linear section of the plot to extract

55
parameters such as R(φ) or αv are inaccurate due to short formation times competing
with membrane resistance and initially variable pressure effects. Equilibrium
compressibility data is obtained only after extremely long filtration times and errors
due to evaporation and/or ageing of the sludge then compete with the accuracy of the
compressibility determination. Thus, multiple- or stepped-pressure techniques are
unsound. Casagrande’s log-time method for the determination of the consolidation
coefficient, cv (Casagrande and Fadum, 1940), cannot be used due to the assumption
of small strains.

In recent work, Tiller and co-workers (Tiller, et al., 2001) measured the
volume-average material properties of solidosity, permeability and specific cake
resistance using a stepped-pressure permeation technique, and defined a class of
materials as super-compactable based upon the parameters used for the functional
forms of the properties. As well as the difficulties using a multiple-pressure
technique, the application of average properties over large volume fraction changes
can be misleading (Lee, et al., 2000), and the theoretical approach required empirical
constitutive equations to be known a priori.

Aside from experimental difficulties associated with dewatering parameter


determination, the question arises as to whether extant theory is capable of predicting
such non-traditional behaviour or if extra forces are involved. An example of an
approach invoking extra forces is the recent approach used by Keiding and
Rasmussen, 2003, where a term postulated as being due to the osmotic pressure of the
filter cake was utilised in a phenomenological description of non-traditional filtration.

This section contains excerpts of a paper written (and currently under review)
to show that the mechanics of filtration for super-compactable materials such as
wastewater treatment sludges are not different than other particulate materials by
exploring the circumstances when the theory of Landman and White, 1997, solved
using the FDM, is capable of predicting non-traditional constant pressure filtration
behaviour.

Material Characteristics

Substituting equations 2.1.6 and 2.1.7 into equation 2.2.18 gives the common
functional form of D(φ):

56
 0 φ < φg 
D(φ ) =  d 2
φ ≥ φ g 
…(2.4.50)
d1φ (1 − φ )
d3

where the fitting parameters d1, d2 and d3 are given by:

p1 p 2
d1 =
r1φ gp 2
d 2 = p2 − 1 …(2.4.51)
d 3 = 2 − r2

For many particulate systems, D(φ) has been shown to be monotonically


increasing and the filtration behaviour shows the traditional long cake formation and
short cake compression (Eberl, et al., 1995, Abd Aziz, et al., 2000, de Kretser, et al.,
2001, Harbour, et al., 2004). However, for the power-law functional form given in
equation 2.4.50, D(φ) has a maximum, Dmax, at φmax:

d2 p2 − 1
φ max = = …(2.4.52)
d 3 + d 2 p 2 − r2 + 1

This work investigated the consequences of D(φ) exhibiting a turning point or


decreasing over the volume fraction range from φ0 to φ∞. The parameters used in
equations 2.1.6 and 2.4.50 are given in Table 2.4.1 and D(φ) is plotted in Figure 2.4.3.

Table 2.4.1: Material property parameters

Parameter Value

φg 0.01 v/v

p1 1.00 Pa

p2 3.00

d1 1.00 x 10-6 m2/s

d2 2.00

d3 6.00

57
1.4E-08

Solids Diffusivity, D(φ) (m /s)


1.2E-08 D max

2
1.0E-08

8.0E-09 D( φ )

6.0E-09

4.0E-09 φ max

2.0E-09

0.0E+00
0 0.2 0.4 0.6 0.8

Solids Volume Fraction, φ (v/v)


Figure 2.4.3: Solids diffusivity, D(φ) = 10-6 φ2 (1 - φ)6 m2/s

The parameters were chosen in order to demonstrate a type of behaviour rather


than to explicitly predict a given filtration time. The values represent a material with
low permeability and high compressibility such as sewage sludge, but do not describe
any particular material. From the parameters given in Table 2.4.1, φmax is 0.250 v/v.

By the rationale provided by the linear approximation (see Section 2.4.2), the
filtration behaviour is non-traditional when Tc is small relative to the overall filtration
time. Tc must always be positive when Rm = 0 since there is always a finite time taken
for the shock at (0,0) to propagate to the piston. For Rm ≠ 0 and φ0 > φg, Tc is zero.
For Tc to be greater than or equal to 0, e* must be less than or equal to e0, giving the
following inequality from equation 2.4.21:

e0
e0 − e∞ ≤ ∫ ∆ (e )de …(2.4.53)
e∞

An approximation to the integral is made using the trapezium rule:

e0
e0 − e∞
∫ ∆ (e )de ≈ (1 + ∆(e0 )) …(2.4.54)
e∞ 2

58
Thus, as a rule of thumb for Tc to be small, ∆(e0) > 1 such that ∆(e) increases
from e∞ to e0. Correspondingly, D(φ0) > (φ∞/φ0)2D(φ∞) (that is, D(φ) decreases
significantly) for the filtration behaviour to show relatively short cake formation
times. From the definition of D(φ), dPy(φ)/dφ must increase at a slower rate than R(φ)
with increasing φ, that is Py(φ) must be a weak function of φ and R(φ) a strong
function of φ.

While the linear approximation provides insight into the relative cake
formation time, it is shown later that Tc is underestimated for certain conditions when
φ* decreases towards φ0, which is precisely the area of interest in this work. As a
result, the FDM was used to solve the filtration equation.

Modelling Results

The FDM numerical solution to pressure filtration was used with the material
characteristics above to generate results of t versus V2 for two scenarios:

- Variable φ0 values with constant ∆P; and

- Variable φ∞ values with constant φ0.

For illustration purposes, V was scaled with the equilibrium specific filtrate
volume, V∞, and t was scaled with tf, the time taken to reach a fraction, f, of V∞, such
that non-traditional behaviour was observed when t/tf versus V/V∞ deviated from
linearity at early relative times. In the examples shown here, f was chosen as 0.99 and
h0 was constant at 0.04 m for all the examples. Varying h0 would not have affected
the relative formation time since both t and V2 vary with h02.

Variable Initial Volume Fraction

The FDM was used to generate filtration results for φ∞ = 0.6 v/v
(corresponding to ∆P = 216 kPa) for a range of initial concentrations, φ0 = 0.04, 0.05,
0.075, 0.10, 0.15 and 0.25 v/v. φ0 was greater than φg to avoid the discontinuity at zc.
These φ0 values were chosen to sample different portions of the D(φ) curve for use in
the model (see Figure 2.4.4), with D(φ0) values from less than D(φ∞) up to Dmax, such
that D(φ) was either increasing and then decreasing or predominantly decreasing.

59
1.4E-08
Increasing φ 0

Solids Diffusivity, D(φ) (m /s)


1.2E-08

2
1.0E-08

φ∞ = 0.60 v/v
8.0E-09

6.0E-09

4.0E-09

2.0E-09

0.0E+00
0 0.2 0.4 0.6 0.8

Solids Volume Fraction, φ (v/v)

Figure 2.4.4: Plot of D(φ) showing φ∞ and the range of φ0 values used for modelling variable initial
concentration

1
Fraction of Filtration Time, t /tf

0.8

0.6 Increasing φ 0

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
2 2
(Fraction of Total Volume) , (V /V ∞ )
Figure 2.4.5: Finite difference modelling predictions of constant pressure filtration for φ∞ = 0.60
v/v with varying φ0

60
The modelling results of t/tf versus (V/V∞)2 for the range of φ0 values are
presented in Figure 2.4.5. Guidelines are shown to illustrate the deviation from
linearity and the onset of cake compression. The results clearly show that as φ0
approached φmax and D(φ0) increased, the filtration curves had shorter formation times
and longer compression times, proving that the filtration theory can generate the non-
traditional behaviour under appropriate conditions without the need to introduce any
extra forces.

Variable Applied Pressure

The other operating variable that has an effect on the relative cake formation
time is the applied pressure. The FDM was used to generate results for constant φ0 =
0.10 v/v and variable ∆P = 63, 124, 215, 342, 511 and 728 kPa. The applied
pressures correspond to φ∞ = 0.40, 0.50, 0.60, 0.70, 0.80 and 0.90 v/v. A plot of D(φ)
is presented in Figure 2.4.6.

1.4E-08
Increasing ∆ P
Solids Diffusivity, D(φ) (m2/s)

1.2E-08

1.0E-08
φ0 = 0.10 v/v

8.0E-09

6.0E-09

4.0E-09

2.0E-09

0.0E+00
0 0.2 0.4 0.6 0.8 1

Solids Volume Fraction, φ (v/v)


Figure 2.4.6: Plot of D(φ) showing φ0 and the range of φ∞ values used for modelling variable
applied pressure

As with the case for variable φ0, the φ∞ values were chosen arbitrarily to
sample different portions of the D(φ) curve, with D(φ∞) from close to Dmax to less than

61
D(φ0). The filtration predictions for variable applied pressure are shown in Figure
2.4.7. The results show that relatively short formation times were also dependent
upon ∆P. Thus, a material that exhibits traditional filtration behaviour at low
pressures may become increasingly non-traditional as the pressure increases.

1
Fraction of Filtration Time, t /tf

0.8

0.6 Increasing ∆ P

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
2 2
(Fraction of Total Volume) , (V /V ∞ )

Figure 2.4.7: Finite difference modelling predictions of constant pressure filtration for φ0 = 0.10
v/v with varying applied pressure

Linear Approximation

The FDM was used for the above predictions instead of the linear
approximation since the latter does not apply for φ* values less than φ0. There is also
a region as φ* decreases towards φ0 where Tc is underestimated. This is illustrated in
Figure 2.4.8 which shows a comparison of the predictions from the two methods for
φ0 = 0.1 v/v, where, even though φ* > φ0, the linear approximation predicted earlier
cake compression.

62
60000

50000 Finite Difference Method


Linear Approximation
40000
Time, t (s)

30000

20000

10000

0
0 0.0004 0.0008 0.0012
2 2 2
(Specific Filtrate Volume) , V (m )
Figure 2.4.8: Linear approximation and finite difference predictions for φ∞ = 0.6 v/v and φ0 = 0.1
v/v

0.40
0.3
0.50 Equilibrium Limit, φ∞ (v/v)

0.60
0.2
φ* (v/v)

0.1
0.70

0.80
0
0 0.1 0.2 0.3

Initial Volume Fraction, φ 0 (v/v)


Figure 2.4.9: φ* as a function of φ0 for a range of φ∞ values: ◊ Variable applied pressure; □
Variable initial concentration; --- φ* = φ0

63
The linear approximation was useful, however, for providing insight into the
conditions required for relatively short formation times. A plot of φ* calculated from
equation 2.4.20 as a function of φ0 for a range of applied pressures is shown in Figure
2.4.9, where the region below the dashed line (φ* < φ0) represents non-traditional
behaviour. At constant initial concentrations, φ* decreased with increasing pressure –
thus higher pressures were expected to show shorter relative formation times. For φ0
= 0.10 v/v, φ* = φ0 when φ∞ = 0.621 v/v. At constant pressure with φ∞ = 0.60 v/v, φ*
increased slowly with increasing φ0 until φ* = φ0 = 0.120 v/v. Thus, the trends
observed were the same as those given by the full numerical solution.

Conclusions

The solid-liquid separation theory of constant pressure filtration was solved


using a finite difference method and shown to exhibit short formation times and long
compression times when D(φ) decreased over the volume fraction range of interest.
This was due to high compressibility over the solids range, such that D(φ) had a weak
dependency on dPy(φ)/dφ and a strong dependency on R(φ).

Rather than belonging just to a class of materials, such behaviour was


dependent upon both the material properties and the operating conditions. The
physical interpretation of this result was that materials that have deformable,
rearrangeable structures that contain high molecular weight and cross-linked
biomolecules exhibit non-traditional constant pressure filtration behaviour at high
pressures and/or high initial concentrations. Significantly, no extra forces were
invoked. Provided that the correct dewatering properties are obtained, the theory can
predict non-traditional behaviour – the challenge now was to more accurately
determine these properties given that permeability and compressibility information
was difficult to extract.

64
2.5 Constant-Pressure Piston-Driven Filtration
with Membrane Resistance

Incorporation of the membrane resistance into solid-liquid separation theory


affects the boundary condition at the membrane, as shown by equation 2.3.8. Instead
of a cake at equilibrium concentration forming immediately, the concentration at the
membrane increases gradually. Two models of pressure filtration with membrane
resistance were developed. The first involved the compressional rheological
description (which required numerical solution) to give predictions of filtration
performance. The second used Darcy’s law to give an approximate analytical method
for determining Rm from transient filtration data, such that Rm for industrial batch filter
presses could be measured on-line if V(t) was measured.

2.5.1 Compressional Rheological Formulation

The formulation for piston-driven filtration with membrane resistance


presented here follows the work of Landman, et al., 1991, but uses the time scaling of
Landman and White, 1997. In this formulation, the problem is kept in Cartesian
coordinates rather than using the material coordinate, w – as such, the scaling for D(φ)
is slightly different. From equation 2.3.4, the scaled governing equation for
consolidation is given by:

∂φ ∂  ∂φ dH 
=  ∆ (φ ) − φ …(2.5.1)
∂T ∂Z  ∂Z dT 

where the various scaled parameters are given by:

65
z
Z=
h0
z (t )
Z c (T ) = c
h0
h(t )
H (T ) = …(2.5.2)
h0

tD (φ ∞ )  φ ∞
2

T=  
h02  φ0 
φ02 D(φ )
∆ (φ ) =
φ ∞2 D(φ ∞ )

D(φ) is scaled with φ02 rather than φ2 since the void ratio substitution is not
used. The initial conditions are:

φ (Z ,0 ) = φ0
…(2.5.3)
H (0 ) = 1

The volume fraction at the membrane is given by the dimensionless form of


equation 2.3.8:

Py [φ (0 ,T )] dH
= 1 + βm …(2.5.4)
∆P dT

where βm is the scaled membrane resistance:

D(φ∞ )  φ∞ 
2
βm =   Rm …(2.5.5)
h0 ∆P  φ0 

βm gives an indication of the importance of the membrane resistance with


respect to the material characteristics and the operating conditions. When βm >> 1,
the membrane resistance is significant compared to the cake resistance, and dominates
the initial stages of the filtration. Equation 2.5.5 shows that the effect of the
membrane decreases when the pressure or the feed concentration are increased and
that, for materials with low diffusivity (such as sewage and water treatment sludges),
Rm must be quite large for βm > 1.

66
The concentration gradient at the membrane is required for the numerical
solution outlined below, and is derived from equation 2.3.9:

∂φ φ (0 ,T ) dH
= …(2.5.6)
∂Z Z = 0 ∆[φ (0 ,T )] dT

From equation 2.3.10, the scaled concentration gradient at the piston is:

∂φ
=0 …(2.5.7)
∂Z H (T )

The scaled global conservation (from equation 2.3.13) is given by:

H (T )
∫ φ (Z ,T )dZ = φ0 …(2.5.8)
0

The solution of equation 2.5.1 subject to the initial and boundary conditions
gives the volume fraction distribution, φ(Z,T) and the scaled piston height, H(T). The
solutions differ for φ0 < φg and φ0 ≥ φg.

Initial Suspension Un-Networked (φ0 < φg, Rm ≠ 0)

For the φ0 < φg case, an internal boundary at Zc(T) exists at the boundary
between the sediment and consolidating cake. Since the concentration is constant at
φ0 in the sedimentation region, the scaled global conservation equation becomes:

Z c (T )
∫ φ (Z ,T )dZ = φ0 (1 − H (T ) + Z c (T )) …(2.5.9)
0

From equation 2.3.12, the concentration gradient at the top of the bed is given
by:

∂φ φ g − φ0  dH dZ c 
=  − 
∂Z Z −
c
∆ φ ( )
g  dT dT 
…(2.5.10)

Therefore, dφ/dZ → -∞ or a large negative number as φ → φg, since ∆(φg) may


be either equal to or greater than 0, depending on the functional form of Py(φ).

67
Initial Suspension Networked (φ0 ≥ φg, Rm ≠ 0)

Landman, et al., 1991, show that, for the φ0 > φg case for piston-driven
filtration with membrane resistance, there is in fact no region where φ remains at φ0
for T > 0, such that Zc(T) → H(T) for T > 0. The implication for the numerical
solution is that the concentration gradient, dφ/dZ → 0 as Z → H(T), and that the exact
solution is an upper bound of possible solutions when iterating on dH/dT.

Small-Time Behaviour

The small-time solutions are used to give the initial time-step in the numerical
algorithm. The small time behaviour is derived based on the work of Landman, et al.,
1991, by introducing a similarity variable, ξ:

Z
ξ= …(2.5.11)
Tb

Substituting equation 2.5.11 into equation 2.5.1 gives:

∂φ 1 ∂  ∂φ  dH  
=  ∆ (φ ) +  bξT 2b − 1 − T b φ
dT  
…(2.5.12)
∂T T ∂ξ 
2 b ∂ξ 

The dominant behaviours at small times for both the un-networked and
networked cases when Rm ≠ 0 are determined by using a series approximation for
φ(Z,T):

φ (Z ,T ) = φ c + T aΦ (ξ ,T ) …(2.5.13)

where φc is the volume fraction at the top of the cake (that is, the maximum of φg and
φ0). Substituting equation 2.5.13 into equation 2.5.12 and dividing through by Ta-2b
highlights the time-dependent behaviour of each term:

∂ Φ
2
 ∂Φ 
∆ (φ c ) + T 2b − 1 bξ − aΦ 
∂ξ 2
 ∂ξ 
…(2.5.14)
∂Φ dH ∂Φ ∂  ∂Φ 
= T 2b +Tb − ∆ ′(φ c )T a Φ
∂T dT ∂ξ ∂ξ  ∂ξ 

68
The small-time approximations for the piston height and the boundary
condition at the membrane come from equations 2.5.4 and 2.5.6:

dH 1  Py (φ c ) 
=− 1 −  …(2.5.15)
dT βm  ∆P 

∂Φ ∂Φ
∆ (φ c ) + T bΦ (0 ,T ) − T aΦ (0 ,T )∆ ′(φ c )
dH dH
= T b − aφ c
∂ξ ξ = 0 dT dT ∂ξ ξ = 0
…(2.5.16)

Along with the magnitude of βm, the small-time solutions give an indication of
the significance of the membrane resistance, since φ(0,T) << φ∞ when Rm is large.

Initial Suspension Un-Networked (φ0 < φg, Rm ≠ 0)

The dominant behaviour at small times for the un-networked case requires a
series approximation Zc(T). For ξ to be a similarity variable, Zc(T) must be
proportional to Tb:

Z c (T ) = BT b …(2.5.17)

The solution for the φ0 < φg case is found by setting a = b = 1 and φc = φg in


equation 2.5.14:

 ∂   dH  
∆ (φ g ) ( )
∂ 2Φ ∂Φ ∂Φ
= T Φ + Φ  ∆ ′ φ g −ξ +   + T 2
∂ξ 2  ∂ξ   ∂ξ dT   ∂T

…(2.5.18)

Assuming ∆(φg) ≠ 0 and T is small, equation 2.5.18 gives:

∂ 2Φ
=0 …(2.5.19)
∂ξ 2

At small times, the flowrate is given by a scaled version of Darcy’s law, since
Py(φg) = 0 in equation 2.5.15:

69
dH 1
=− …(2.5.20)
dT βm

The top of the cake is at φg, therefore:

Φ (B ,T ) = 0 …(2.5.21)

The dominant behaviour of equation 2.5.16 gives the gradient of Φ at the


membrane:

∂Φ φg dH
= …(2.5.22)
∂ξ ξ =0
∆( φ g ) dT

Solving equation 2.5.19 subject to equations 2.5.21 and 2.5.22 and substituting
equation 2.5.20 gives the small-time behaviour of Φ(ξ,T) with one unknown, B:

φg 1
Φ (ξ ,T ) = (B − ξ ) …(2.5.23)
∆( φg ) β m

The small-time approximation of equation 2.5.10 gives the gradient at ξ = B:

∂Φ φ g − φ0  1 
=−  B +  …(2.5.24)
∂ξ ξ =B
∆( φg )  β m 

Eliminating the gradients from equations 2.5.22 and 2.5.24 gives B:

1 φ0
B= …(2.5.25)
β m φg − φ0

Thus, the small-time approximations for Zc(T) and φ(Z,T) are:

φ0
Z c (T ) =
T
…(2.5.26)
φ g − φ0 β m

  φ0 
φ (Z ,T ) = φ g 1 +
1  T
− Z  …(2.5.27)
 β m ∆( φg )  φ g − φ0 β m 


Thus, φ(0,T) initially varies linearly T:

70
−1
 1 
φ (0 ,T ) = φ g +  −
1  T
( )
…(2.5.28)
 φ0 φ g  β m2 ∆ φ g
 

Initial Suspension Networked (φ0 ≥ φg, Rm ≠ 0)

The small-time analysis for the φ0 > φg case is found by setting a = b = ½ and
φc = φ0 in equation 2.5.12 and discarding higher order terms of T:

∂ 2Φ ξ ∂Φ Φ
∆ (φ0 ) + − =0 …(2.5.29)
∂ξ 2 2 ∂ξ 2

With these values for a and b, the dominant behaviour of equation 2.5.14 is:

∂Φ φ dH
= 0 …(2.5.30)
∂ξ ξ = 0 ∆ (φ 0 ) dT

Since dφ/dZ → 0 as Z → H(T) in the limit as T → 0, equation 2.5.7 becomes:

∂Φ
=0 …(2.5.31)
∂ξ ξ → ∞

As ξ becomes large, the concentration approaches the initial concentration,


therefore:

ξ → ∞ Φ (ξ ,T ) = 0
lim
…(2.5.32)

The solution to equation 2.5.29 subject to equations 2.5.30 to 2.5.32 is:

dH  1 −( Aξ )2 
Φ (ξ ,T ) = −4 Aφ0  e − Aξerf ( Aξ ) …(2.5.33)
dT  π 

where

1
A= …(2.5.34)
2 ∆(φ0 )

71
From equation 2.5.13, the dominant behaviour of the piston speed at small
times is:

dH 1  P (φ ) 
=− 1 − y 0  …(2.5.35)
dT βm  ∆P 

Therefore, the series approximation for the volume fraction at small times for
the networked case is given by (Landman, et al., 1991):

 Py (φ0 )  T − T 


AZ
2

  AZ  
φ (Z ,T ) = φ0 1 +
 4 A
1 −  e − AZerf   …(2.5.36
 β m  ∆P  π  T  
  
)

While the initial flow rate is constant, φ(0,T) varies with the square-root of
time:

 1  P (φ )  4T 
φ (0 ,T ) = φ0 1 + 1 − y 0   …(2.5.37)
 β m  ∆P  π∆ (φ0 ) 

Runge-Kutta Numerical Algorithm

A numerical algorithm, illustrated in Figure 2.5.1, was programmed in


Mathematica®. It is capable of solving equation 2.5.1 subject to the appropriate
boundary conditions for the following scenarios:

- Initial suspension networked or un-networked;

- With or without membrane resistance.

Equation 2.5.4 shows that φ(0,T) is dependent on dH/dT, which is determined


from the governing equation, indicating an iterative solution is needed. A Runge-
Kutta shooting technique (Matthews, 1992) is presented here to solve for φ(Z,T).
Equation 2.5.1 is simplified to two coupled ordinary differential equations by
introducing the scaled solids flux, ψ(Z), and making a backward difference
approximation in time:

72
dφ 1  dH 
= ψ +φ …(2.5.38)
dZ ∆ (φ )  dT 

dψ φ − φ <
= …(2.5.39)
dZ ∆T

where φ<(Z) is the value of φ(Z) at the previous time step.

The cumulative solids volume fraction, Q(Z), is used to evaluate the


conservation of mass:

dQ
=φ …(2.5.40)
dZ

The initial time step value is determined by giving the small-time solutions to
the membrane volume fraction (equations 2.5.28 and 2.5.38) small values such as 2φc:

  1 1  
 βm 2
( )
∆ φg φg  − 
 φ0 φ g 
φ0 < φ g 
   
∆T =  −2  …(2.5.41)
πβ m ∆ (φ0 ) 
2 Py (φ0 )  
 1 −  φ0 ≥ φ g 
 ∆P  
 4  

∆T cannot be too high (otherwise the accuracy of the approximation for the
time derivative is reduced) or too small (otherwise the required computational time is
unreasonable). A reasonable maximum ∆T is 0.1, with a minimum of 0.0001. Since
the changes in volume fraction are greatest at small times, ∆T is allowed to increase at
each time step until the maximum value is reached.

At each time step, an estimate of dH/dT, dH*/dT, is iterated upon using an


interval halving method. The lower bound, dH/dTlow, for the first time step is given
by the small-time solution (equation 2.5.13), while the initial upper bound, dH/dThigh,
is zero. For a known time step, ∆T, an estimate of H(T), H*, is given by a first-order
approximation:

dH *
H * = ∆T + H< …(2.5.42)
dT

73
1-D Constant Pressure Dead-End Filtration with Membrane Resistance
Iterative 4th – 5th order Runge-Kutta Algorithm

Material Characteristics:
Compressive yield stress, Py(φ); Hindered settling function, R(φ); Solids diffusivity, D(φ); Gel point, φg
Operating Conditions:
Initial height, h0; Initial concentration, φ0; Applied pressure, ∆P; Membrane resistance, Rm
Filtration time, tf, or Stopping fraction, f

φ c = Maximum {φ0 ,φ g }; Py (φ ∞ ) = ∆P

Scalings:
z (t ) h(t ) tD(φ ∞ )  φ ∞  φ 2 D(φ ) D(φ ∞ )  φ ∞
2 2

; Z c (T ) = c ; H (T ) =  ; ∆ (φ ) = 0
z
Z= ;T =  ; βm =   Rm
h0 h0 h0 h0  φ0 
2
φ ∞2 D(φ ∞ ) h0 ∆P  φ0 

Initial Conditions:
 P (φ ) 
T = 0; φ (Z ,0 ) = φ0 ; Z c (0 ) = 0 ; H ( 0 ) = 1;
dH 1
=− 1 − y c 
dT 0 βm 
 ∆P 

Initial Time-Step Size and Iteration Bounds:


−2
P (φ ) 
; If φ0 ≥ φ g , ∆T = πβ m ∆ (φ0 )  1 − y 0  ; dH
  2 
( )
*
1 1 dH dH *
If φ0 < φ g , ∆T = β m2 ∆ φ g φ g  −   = ; =0
 φ0 φ g  4 ∆ P dT dT
    low 0 dT high

Time step, T = T< + ∆T

dH * 1  dH * dH *  * * P [φ (0 )] dH *
= + ; H = ∆T dH + H < ; y = 1 + βm ; ψ (0 ) = 0; Q (0 ) = 0
dT 2  dT high dT low  dT ∆P dT

Runge-Kutta Algorithm:
φ − φ < 
dφ 1  dH *  dψ  ∆T Z ≤ Z c<  dQ
Solve = ψ +φ ; =  and =φ
dZ ∆ (φ )  dT  dZ  φ − φ c < dZ
Z > Zc
 ∆T 

from Z = 0 until φ = φ c , > 0 or Q = φ0
dZ

Q (Z )
H test = 1 − +Z
φ0

∂φ dH * dH * dH * dH *
If H test < H * or > 0, = ; If H test > H * , =
∂Z dT high dT dT low dT

dH * φ
; H (T ) = H * ; Z c (T ) = Z ; φ = 0
dH
=
dT T dT H (T )

If φ(0,T) < f φ∞ and t < tf, proceed to next step of ∆T

dH * dH dH *
= ; =0
dT low dT dT high

Figure 2.5.1: Algorithm for piston-driven constant-pressure filtration with membrane resistance

74
H< is the value of H at the previous time step. The first-order approximation
tends to underestimate H(T) since d2H/dT2 generally monotonically increases. Higher
order approximations, which may be more accurate, are not used because they tend to
overestimate H(T), such that no solution exists for the φ0 > φg case. dH/dTlow for
subsequent steps is given by the value at the previous step.

Equations 2.5.38 to 2.5.40 are solved using 4th order Runge-Kutta steps of ∆Z
from Z = 0 (where φ(0,T) is given by equation 2.5.4 and ψ(0,T) = Q(0,T) = 0) until φ =
φc, dφ/dZ > 0 or Q = φ0. The accuracy of each step of ∆Z is kept in check using the
truncation error from the 5th order. If dφ/dZ > 0, H* is too high and dH*/dT is the
upper bound for the next iteration. If φ = φc or Q = φ0, a test value for H(T), Htest, is
given by substituting equation 2.5.40 into equation 2.5.15:

Q (Z )
H test = 1 − +Z …(2.5.43)
φ0

If Htest < H*, dH*/dT is the lower bound for the next iteration. Conversely, if
Htest > H*, dH*/dT is the upper bound for the next iteration. dH*/dT is iterated upon
until Htest ≥ H* to within the desired accuracy. Htest cannot be less than H* as this may
correspond to the stopping condition Q > φ0. Since the conservation of volume is
used to give Htest, equation 2.5.10 can be evaluated and used to check the overall
accuracy of the solution for the un-networked case.

Fundamentally, a change of variables to X = Z/Zc(T) should be made in the φ0


< φg case due to the discontinuity at Zc(T). However, since the relationship between
Zc and H is related by the solution of the governing equation (see equation 2.5.9),
using X introduces an extra iteration variable, dZc*/dT. This would increase the
computational time by an order of magnitude. An approximation is made, such that
the solids flux gradient for Zc< < Z < Zc is given by:

dψ φ − φ g
= …(2.5.44)
dZ ∆T

The algorithm was written to be as general as possible. Batch settling results


of water and wastewater treatment sludges in the next chapter show that the
compressive yield stress can vary by 6 or 7 orders of magnitude over the small

75
volume fraction range above the gel point. In the analysis software used, the
derivative of the functional form used to describe Py(φ) asymptotes to 0 as φ → φg,
and is therefore infinite at φg. In order to avoid the singularity as φ → φg for the un-
networked case, it was necessary to assume a finite positive value for ∆(φg) (such as
∆(1.0001 φg)) and to ensure that φ ≥ φg and dφ/dZ ≤ 0 within the Runge-Kutta
algorithm by reducing ∆Z.

2.5.2 Modified Darcy’s Law

The traditional method for calculating the membrane resistance in filtration


involves determining the intercept of a plot of t/V versus V (see equation 2.1.4).
However, in situations where a time lag exists between the start of the filtration run
and the application of the set pressure such as in the case of plate-and-frame filter
presses, this approach is not useful (Usher, et al., 2001).

An alternative method is to differentiate equation 2.1.5 with respect to V2:

dt η f α vφ 0 Rm 1
= + …(2.1.45)
dV 2 2 ∆P f (φ b − φ 0 ) 2∆P f V

Thus, plotting dt/dV2 versus 1/V gives a straight line with a slope that allows
calculation of Rm, independently of initial pressure fluctuations. However, equation
2.1.45 still requires φb and αv to be independent of t, such that the material is
incompressible. Thus, equation 2.1.45 is useful only after the operating pressure is
reached but before the onset of cake compression.

76
2.6 Soil Consolidation

Civil engineers use Terzaghi’s consolidation model (Terzaghi and Peck, 1967)
to analyse the settlement of structures, the stability of slopes and the design of pile
foundations. Terzaghi’s model has been extended and improved over time by various
authors (Davis and Raymond, 1965, Janbu, 1965, Mikasa, 1965, Gibson, et al., 1967,
Schiffman, et al., 1984) and reviewed by de Boer, 1996. Considering saturated soils
as highly concentrated suspensions or suspensions as high void ratio soils, the
physical situations of the consolidation and settlement of soils and the dewatering of
suspensions due to an applied pressure are essentially the same – both describe the
extrusion of water from a saturated particulate network through a porous barrier due
to the effect of an applied load.

The governing equations for Terzaghi’s consolidation model and the


phenomenological filtration model are essentially the same (Landman and White,
1994). However, the theories for describing the solid-liquid separation, the
experimental methods and the analysis techniques used to measure the material
properties of soils and suspensions have developed separately.

Both approaches use similar attributes for permeability and compressibility,


and combine them to give a single parameter that describes the process, although the
nomenclature and methods of determining the relative dewatering parameters differ.
The consolidation method matches experimental data from oedometer testing to the
theoretical predictions of the model in order to determine the coefficient of
consolidation, cv. The filtration method determines the solids diffusivity coefficient,
D, based upon the experimental data from a filtration rig, which is then used in
modelling to make predictions.

Apart from the theory of Tiller and Shirato, 1964, who included Terzaghi’s
theory to describe cake compaction, and the work undertaken by Eckert, et al., 1996,
who demonstrated the conversion of permeability-void ratio relationships from
geotechnical experiments to state functions of hindered settling velocity and
compressive yield stress, there is little collaboration between the two fields. Physical
scientists generally do not refer to soil consolidation theory or methods, although

77
users of volume-averaged filtration theories may recognise the equivalence of cv with
diffusivity, D. Likewise, more recently developed filtration theories and their
methods of extracting material parameters have not been applied to soil consolidation.

The similarities between filtration and consolidation were investigated in order


to reconcile the two approaches. In this section, the theoretical relationship between
the two parameters, cv and D, is developed, while the material characteristics of a
kaolin sample determined using both oedometer and filtration testing are presented in
Section 4.3.1. The results were equated using the developed theoretical relationship
and compared using the stepped-pressure filtration model.

2.6.1 One-dimensional Consolidation Theory

Traditionally, geotechnical engineers have used soil mechanics models to


predict the ability of a soil to support a load. As described by Craig, 1997, a fully
saturated soil subjected to an increase in applied load undergoes an increase in the
internal pore water pressure. Consolidation is the gradual reduction in volume due to
the drainage of some of the pore water. The process continues until the excess pore
water pressure (pf) has completely dissipated and the system equilibrates. At this
time, ∆P equals the effective vertical stress, σ’, in the sample. σ’ is analogous to pp.

In 1923, Terzaghi developed a one-dimensional theory of consolidation that is


considered as the foundation of modern geotechnical engineering (Schiffman, et al.,
1984). Terzaghi’s theory assumes that Darcy’s law is valid, the sample is
homogeneous and fully saturated, the mineral structure is incompressible, the strains
are small (such that the hydraulic conductivity, k, and the coefficient of volume
compressibility, mv, are constant throughout consolidation), and that there is no
secondary compression (Holtz and Kovacs, 1981). The coefficient of consolidation,
cv, is defined as the ratio of k and mv (Olson, 1986):

k
cv = …(2.6.1)
mv γ f

78
where γf is the unit weight of the fluid (kg m-2 s-2), and is the product of ρf and g. k
and mv have been shown to be functions of void ratio, e (Davis and Raymond, 1965)
and will vary significantly if the void ratio changes and strains are appreciable, as is
the case for the consolidation of high void ratio soils.

mv is defined as the fractional volume change per unit increase in effective


stress, σ’, and has units of inverse pressure (m2/N). mv is measured independently of
cv using a variation of an oedometer test in which the load is increased incrementally.
For an increase in effective vertical stress from σ’0 to σ’1, the thickness of the
material subjected to compression decreases from h0 to h1, such that:

1  h0 − h1  1  e0 − e1  1 de
mv =   =   = − …(2.6.2)
h0  σ '1 −σ ' 0  1 + e0  σ '1 −σ ' 0  1 + e0 dσ '

k arises from Darcy’s Law, which can be expressed in terms of the difference
between uf and up (Gibson, et al., 1967):

k ∂p f
e
(
(1 + e )
u f −up = − )
γ f ∂z
…(2.6.3)

k has the units of velocity rather than area per unit time and is not, strictly
speaking, a coefficient of permeability, and is called the hydraulic conductivity. k is
determined directly by means of constant head or falling head permeability tests.

cv is determined independently of k and mv by means of an oedometer test, in


which the experimental results are fitted to theoretical predictions. The analysis
techniques used here (Casagrande and Fadum, 1940, Taylor, 1942) are described in
Section 3.1.7.

2.6.2 Equating the Parameters

R(φ) can be expressed in terms of k by equating the fluid pressure gradients


given in equations 2.2.5 and 2.6.3:

79
R (φ ) =
(1 − φ )2 γ f …(2.6.4)
φ k

The compressive yield stress, Py(φ), is equivalent to the effective vertical


stress, σ’ (Eckert, et al., 1996), and is substituted along with equation 2.6.4 into the
definition of D(φ) (equation 2.2.18) to give:

k dσ ′
D(φ ) = φ …(2.6.5)
γ f dφ

The volume fraction, φ, is replaced by the void ratio, e, according to equation


2.4.4 to give the solids diffusivity as a function of void ratio:

k dσ ′
D(e ) = −(1 + e ) …(2.6.6)
γ f de

Substituting the definition of mv (equation 2.6.2) gives:

D(e ) =
(1 + e) k …(2.6.7)
(1 + e0 ) mvγ f

Substituting cv for k, γf and mv (equation 2.6.1) yields D(e) in terms of cv, e and
e0:

D (e ) =
(1 + e) c …(2.6.8)
(1 + e0 ) v

Similarly, cv can be given in terms of D(φ), φ and φ0:

φ 
cv =   D(φ ) …(2.6.9)
 φ0 

Therefore, cv(e) as determined from oedometer testing can be directly related


to D(φ) from filtration testing, as illustrated in Section 4.3.1. In addition, equation
2.6.9 also indicates that cv can be determined directly from filtration experimental
results and D can be determined directly from oedometer testing.

80
3 Experimental Methods and
Materials

This chapter outlines the experimental methods and analysis techniques used to
characterise sludges, gives a short description of the materials tested and an outline of
the work performed for onsite case studies.

3.1 Methods

In general, sludge dewatering behaviour was dependent on the solids content


and density, pH, temperature, fluid viscosity, coagulant and/or flocculant type and
dose, and shear history. The properties that needed to be measured for compressional
modelling purposes depended on the dewatering mechanism of the device being
modelled and the theoretical framework used, but they generally involved filtration,
centrifugation or settling testing.

3.1.1 Background

Traditionally, simple empirical tests such as capillary suction time (CST),


sludge volume index (SVI) and specific resistance to filtration (SRF) are used for
predicting the dewatering and determining the optimum flocculant dosage of water
and wastewater sludges. CST and SVI are used for the analysis of thickening
operations, and SRF is occasionally used for filtration operations. All the results from
these tests are dependent upon the initial solids concentration, and generally fail to
give fundamental material properties.

81
CST is obtained by examining the rate at which water can be drawn out of a
given quantity of sludge by capillary suction through a filter membrane (Baskerville
and Gale, 1968). CST continues to be used by practicing engineers to rank
suspensions according to their filterability due to its simplicity (Smiles, 1998), but is
dependent upon many non-fundamental parameters (Vesilind, 1988, Dentel, 1997),
such as the type and thickness of the filter as well as the temperature, concentration
and viscosity of the suspension and cannot be used to give an indication of the final
solids achievable in dewatering processes (Novak, et al., 1999, Chen, et al., 2001,
Kopp and Dichtl, 2001, Pan, et al., 2003). The use of CST as an indication of
dewaterability across different sludge types and solid concentrations is fundamentally
flawed (Vesilind, 1988, Vesilind and Ormeci, 2000, Harbour, et al., 2001). Thus,
CST has limited use in optimisation of flocculant dose at the test conditions, but tends
to lead to overdosing since the there is little or no increase in CST after the optimum
dose, such that the minimum is not always clear (Wu, et al., 1997), and cannot be
extrapolated to filtration behaviour or used for modelling predictions.

SVI is the percentage of solids that settle after a certain amount of time in a
batch settling test (Kopp and Dichtl, 2001). As illustrated by many authors (Kynch,
1952, Buscall and White, 1987, Howells, et al., 1990, Bürger and Concha, 1998,
Lester, et al., 2005) and detailed in Section 3.1.6, the results of a batch-settling test are
dependent upon the material properties of the suspension and the settling time, and the
extraction of relevant parameters from transient data is not simple. For example,
sludge with high permeability and low compressibility will settle to a lower height
than sludge with low permeability and high compressibility over a short period, but
vice versa over a longer period of time. Thus, SVI has limited application to
predicting settling rates in thickeners at the test conditions only.

SRF (or αv) is measured using constant pressure filtration techniques. The
modified Darcy’s law (equation 2.1.5) predicts that the slope of t with V2 gives αv. As
discussed in the theory, this volume-averaged approach assumes incompressible cakes
and so does not account for cake compaction (in fact, tests are usually not performed
to equilibrium). The data obtained by the SRF method is dependent on the starting
volume fraction and cake thickness and therefore sludges with differing starting
concentrations or solids loading cannot be directly compared (Harbour, et al., 2001).

82
Instead of these methods, the experimental methods and analysis techniques
developed at the University of Melbourne (Green, et al., 1998, de Kretser, et al.,
2001, Lester, et al., 2005) were used to determine the fundamental compressional
rheology parameters of Py(φ) (including φg), R(φ) and D(φ) for a range of water and
wastewater sludges. Since these properties vary by large orders of magnitude with
concentration, they were measured over the entire operational volume fraction range
within the device to be modelled – for example, from φ0 to φ∞ for filters and from the
feed to the underflow concentration for thickeners. Extrapolating from settling
behaviour to predict filtration behaviour or vice versa was avoided where possible. In
this work, filtration tests were used to measure high volume fraction properties, while
batch settling tests were used to measure low volume fraction properties.

This chapter outlines the laboratory techniques of constant-pressure filtration,


batch settling, oven drying and density measurement, and discusses general issues
relating to sample preparation including the laboratory conditioning of sludges using
polyelectrolytes. The results from the filtration and settling tests were analysed using
curve-fitting software in conjunction with calculations outlined below to give the
material characteristics. The methods of analysis differed for traditional and non-
traditional filtration, and separate procedures were required.

Specific issues related to these methods with regards to water and wastewater
treatment sludges are also discussed. Characterisation of the dewaterability of
biosludge samples was difficult because testing was often hampered by the slow rate
of settling and filtration. Biological activity could change the properties of the sludge
before the test was completed. In the methods presented here, the time problem was
addressed so that the relevant data was obtained before biological activity invalidated
results.

A description of soil consolidation oedometer testing is given in order to


illustrate the similarities between the civil engineering and physical science fields of
research. The oedometer test, with permeable barriers at either end, is effectively a
double-ended filtration test.

A summary of the materials that were characterised and a work outline for on-
site trials are given.

83
The methods presented here did not consider the shear properties of the sludge
or the effect of shear on the compressional properties of the sludge. Particulate
suspensions are yield-stress materials in shear as well as compression, although the
shear yield stress, σy(φ,p), is much less than the compressive yield stress (Buscall, et
al., 1987). Thus, small amounts of shear can have large impacts on the compressional
behaviour, as illustrated in the use of rakes and lamellar surfaces in thickening.
Likewise, the characterisation did not consider the drainage or drying properties of the
sludge, such as capillary pressure, residual moisture content and evaporation rates.
These de-saturation mechanisms occur in vacuum filters, washing and evaporative
processes and in the beach section of decanting centrifuges. While these effects are
essential to the modelling of these devices, the methods to measure the appropriate
fundamental material properties are either in development or do not exist.

3.1.2 Sample Preparation

Sludge samples were treated with care to ensure that the dewatering
characteristics did not change prior to experimental testing. Particular issues of
importance are listed below:

- All sludges were kept in the fridge in order to inhibit degradation and reduce
biological activity;

- Sludge samples were warmed to room temperature prior to settling or filtration


testing, since R(φ) is directly proportional to ηf; and

- The initial concentration was determined from oven-drying (see below) to


allow determination of coagulant and/or polyelectrolyte dosage.

3.1.3 Density Measurement

The solids concentration for sludges was measured as solids mass as a


percentage of the total mass (wt%) by gravimetric techniques. This unit of
measurement is unsuitable for use in filtration characterisation as it is the volume of

84
solids that influences the filtration rate. φ was calculated from the solids
concentration using ρp and ρf:

ρ f wt%
φ= …(3.1.1)
ρ p (1 − wt% ) + ρ f wt%

ρf was assumed to be 1000 kg/m3. If ρp was not known, it was measured using
a density bottle (a bottle with an accurately known volume). The weight of the bottle
when dry, full of water and full of sludge of a known mass fraction (from oven
drying) gave the mass, and therefore volume, of water in the sludge. Since the
volume of the bottle was known, the solids density could be calculated.

Small errors in solids measurement caused large errors in density values since
the sludge concentration was usually 1-2 wt%. Concentrating the sludge to higher
concentrations beforehand alleviated this accuracy problem. As long as the mass of
solids in the bottle was known then the density could be calculated.

It was practically difficult to get the sludge into the small opening of a
standard glass density bottle, especially if the sludge had been thickened to increase
the accuracy of the measurement. The use of a density cup with a larger opening
solved this problem.

3.1.4 Oven drying

The solids concentration was determined by oven drying. The first


measurement (performed in triplicate) was done on the unfiltered or unsettled sludge
and was used along with ρp to calculate the initial volume fraction, φ0. The second
measurement was to oven dry the final cake or settled solids to give the final volume
fraction, φf, again based on ρp. The procedure is outlined below:

- A small oven tray or crucible was weighed using an analytical balance


(accuracy required to 3 decimal places or greater);

- The sample and oven tray were weighed again to obtain a “damp weight”; and

85
- The cake or sludge was oven dried overnight at 80ºC, cooled in a desiccator,
and then reweighed to give the dry weight of cake.

φf was used to check φ0 based upon a volume balance. Any discrepancies were
due to the extrusion of water before the start of the filtration run, errors in the density
measurement, evaporation of water from the cake at the end of the run, soluble
material in the filtrate, air bubbles in the sludge or changes with time in the solids
concentration of the sludge after the initial measurement. The implications for
experimental protocols were that it was important to:

- Measure any loss of liquid before the start of filtration runs;

- Ascertain the amount of dissolved solids in the filtrate; and

- Minimise experimental time to reduce the loss of liquid through evaporation.

Calculations of volume fraction accounted for dissolved solids only if there


was a significant amount (Usher, 2002). There were generally very little dissolved
solids in water and wastewater sludges measured here, and it was not an important
issue.

3.1.5 Constant-Pressure Batch Filtration

A constant pressure filtration device was used in this work to determine mid to
high pressure material properties (Py(φ), R(φ) and D(φ)) for a suspension. The
filtration rig was originally designed by Green, et al., 1998 and developed further by
de Kretser, et al., 2001 and Usher, et al., 2001. A photo and a schematic of the rig are
shown in Figure 3.1.1.

Samples were loaded into the stainless steel filtration cylinder (42 mm internal
diameter), which had a semi-permeable membrane supported by a permeable sintered
metal disc at the base. Approximately 50 ml of sludge were loaded each time, giving
initial piston heights of approximately 30 mm. The filtration device used a pneumatic
cylinder to exert a pressure upon a sample through a piston and measured the piston
height with time using a linear encoder (with spatial resolution of 10 µm). A pressure

86
transducer mounted in the face of the piston measured the applied pressure, which
was controlled from the pneumatic cylinder via a feedback loop. The software
running the rig is described in detail elsewhere (de Kretser, et al., 2001). Random
errors in t, V(t) and ∆P were very small. An alternative design also employed here
used air-pressure to filter the suspension, with the volume of filtrate measured via a
balance (Abd Aziz, 2004).

Linear
Encoder Pressure
Controller

Air Line
Double Ended
Pneumatic Cylinder

Bleed
Line
Stainless
Steel Piston
Pressure
Transducer
Stainless Steel
Filtration Cylinder
Membrane
Beaker PC

Figure 3.1.1: Photo and schematic of automated filtration rig

Also used here for single-pressure runs, as an alternative to the automated


devices, was the manual filtration die designed by Nevil Anderson of CSIRO. A
heavy weight was hung from the piston of a simple filtration cell (20 mm internal
diameter) and the piston height recorded manually using callipers. The benefits of the
manual rig were the ability to reach high pressures and to perform multiple tests
simultaneously using several rigs side-by-side. However, it was labour-intensive and
less accurate (due to errors in h(t), ∆P and friction) than the automated version.

φ∞ was calculated from the final height of the suspension. Repeating the
experiment over a range of pressures generated a complete Py(φ) curve. Values for
R(φ) were found from the slope of the linear portion of t versus V2 measured over a
range of pressures to give β2 versus ∆P (see equation 2.4.34). Py(φ) and R(φ) were
combined to give an expression for the solids diffusivity, D(φ), according to equation
2.2.18. The calculation procedure is outlined in Section 3.2.1. As well as this, φ∞ and

87
D(φ∞) was deduced by curve –fitting a logarithmic function to the compression stage
of filtration. The filtration tests were either single-pressure or stepped-pressure and
were performed over a range of pressures, such as 20, 50, 100, 150, 200 and 300 kPa.

Due to the extremely compressible and impermeable nature of sewage sludges,


the time to reach the end of a single-pressure filtration run could be large such that
biological activity and evaporation influenced the results. Since the filtration time
varies with h02, loading less sludge reduced the filtration time. However, the applied
pressure of the filtration test was accurate for cake heights above approximately 2
mm. Runs proceeding beyond this were manually stopped and the test repeated with a
larger initial height. For low pressure tests, sludges at concentrations below the gel
point would settle whilst in the filtration cell, skewing the results. Therefore, such
sludges were settled prior to filtration.

Stepped-Pressure Testing (Traditional Filtration)

Following the method of de Kretser, et al., 2001, two stepped-pressure


experiments were used to determine the material parameters for traditional materials,
such as water treatment sludges. In the compressibility test for determining Py(φ∞)
(see Figure 3.1.2), a normal batch filtration at the lowest test pressure was allowed to
proceed until the piston movement dropped below a very low rate (assumed to be
close to equilibrium). The pressure was increased to the next nominated value and
then kept constant until equilibrium was again approached. Repeating the experiment
over a range of pressures generated a series of equilibrium heights, which were
converted to φ∞ values. These results were then combined with measurements of the
gel point to give a complete Py(φ) curve. The overall time was reduced by loading
less sample (on the condition that the final cake height was greater than 2 – 3 mm) or
by foregoing low-pressure measurements.

In the permeability test (see Figure 3.1.3), the time at each pressure was
truncated such that the pressure was stepped when the slope of t versus V2 was stable
(see Figure 3.1.4). The control software took an average slope over a user-defined
number of preceding data, and recognised when this was linear to within a user-
defined tolerance. The permeability test was only applicable during cake formation
and therefore large initial heights (6 – 7 cm) were useful.

88
18000

16000
300 kPa
14000
200 kPa
12000
150 kPa
Time, t (s)

10000
100 kPa
8000
50 kPa
6000

4000 20 kPa
2000

0
0 0.00005 0.0001 0.00015 0.0002
2 2 2
(Specific Filtrate Volume) , V (m )

Figure 3.1.2: Example of t versus V2 results for the stepped-pressure compressibility test (Ferric
water treatment sludge, Langsett WTP, 05/06/01)

5000
300 kPa
4500

4000

3500 200 kPa


Time, t (s)

3000
150 kPa
2500
100 kPa
2000

1500 50 kPa
1000

500 20 kPa

0
0 0.0002 0.0004 0.0006
2 2 2
(Specific Filtrate Volume) , V (m )

Figure 3.1.3: Example of t versus V2 results for the stepped-pressure permeability test (Ferric
water treatment sludge, Langsett WTP, 05/06/01)

89
3.0E+07

20 kPa
2.5E+07

2.0E+07
dt /dV (s/m )
2

50 kPa
1.5E+07

100 kPa
2

150 kPa

200 kPa

300 kPa
1.0E+07

5.0E+06

0.0E+00
0 1000 2000 3000 4000 5000
Time, t (s)

Figure 3.1.4: Example of dt/dV2 versus t results for the stepped-pressure permeability test (Ferric
water treatment sludge, Langsett WTP, 05/06/01)

As indicated by the time scale, the stepped-pressure compressibility test took


four to five hours while the permeability test took one to two hours for typical water
treatment sludges. Thus, the two tests were easily performed in a day. The random
errors for stepped-pressure filtration have been shown to be very small and the results
very reproducible (Green, et al., 1998, de Kretser, et al., 2001, Usher, 2002).

Single-Pressure Testing (Non-Traditional Filtration)

Sewage sludge filtration typically exhibited short cake formation times,


followed by long cake compression times. Thus, it was difficult to get the slope and
φ∞ values from a single pressure run, and multiple pressure runs were not useful.
However, based on the linear approximation solution given by Landman and White,
1997, φ∞ and D(φ∞) were determined by curve fitting a logarithmic function to the end
of the data from single-pressure filtration tests (de Kretser, et al., 2001, Usher, 2002).
Equation 2.4.29 was rearranged in terms of <φ>, giving equation 3.1.2 for the later
stages of cake compression:

90
 1 1 
t = E 3 − E 2 ln − 
…(3.1.2)
 φ φ∞ 

where

h02φ02  4  2 A0  4 h02φ02 ln(φ0 h0 )


E3 = Tc + ln  = E − …(3.1.3)
D(φ ∞ )φ ∞2  π 2 D(φ ∞ )φ ∞2
1
π2  π 

The last m points of the compression data were used to give material
parameter information since E2 contains an expression for D(φ∞). The method used
here involved randomly generating two hundred sets of m and φ∞ from a range of
possible values and then finding the set that gave the minimum fitting error. The
range was then reduced by 50% and the exercise repeated until the values were stable.

160000
Average volume fraction
140000
Curve Fit
Endpoint = 0.1825 v/v
120000
Time, t (s)

100000

80000

60000

40000

20000

0
0 0.05 0.1 0.15 0.2

Volume fraction, φ (v/v)

Figure 3.1.5: Example of logarithmic curve fit to average volume fraction results (Luggage Point
WWTP, 14/02/03, 20 kPa)

An example of the curve fitting is shown in Figure 3.1.5. The same technique
was applied to a range of pressures enabling functional forms for Py(φ), R(φ), and
D(φ) to be determined. The calculation procedure is outlined in Section 3.2.2. The

91
results for single-pressure filtration testing have been shown to correspond to stepped-
pressure results (Usher, et al., 2001).

3.1.6 Batch settling

Batch settling testing was used to give low volume fraction permeability and
compressibility data. In these tests, sludge was loaded into large measuring cylinders
(500ml or greater) and allowed to settle. General issues that arose during batch
settling testing included:

- The initial solids concentration: Settling tests were generally performed at


concentrations significantly less than φg, thus samples may have required
dilution prior to testing;

- Wall friction: Experiments were conducted in containers with a flat bottom


and a diameter of 4cm or above;

- Unclear interface: The inhomogeneous nature of some sludges caused an


unclear interface in batch settling, especially during the initial induction and
free-fall stages. The height was taken as an estimate of where there was a
definite discontinuity in concentration;

- Flotation of flocs: The generation of the gaseous by-products of biological


activity could cause high organic content sludges to begin to float. Such
samples were diluted in order to lower φ0 and increase the initial rate of
settling, thus reducing the time taken to perform the tests. Lowering φ0 also
ensured that the final equilibrium height of the sludge-water interface was
attained sooner; and

- Vibrations: The compression behaviour of sludges is extremely sensitive to


shear at low concentrations. As such, vibrations in the laboratory may
influence the results of the batch settling and were avoided where possible.

Two types of tests were used, transient and equilibrium.

92
Transient Batch Settling

Transient batch settling required the height of the sludge/clear-liquor interface


to be measured with time until the interface stopped falling. R(φ) for φ0 < φ < φg was
determined through a transformation of the height versus time data (Lester, et al.,
2005) using a program developed at the University of Melbourne (called “Batch
Settling Analysis Method Software” (BSAMS)) on behalf of the minerals industry.
The inputs to the software were the initial solids volume fraction (φ0), initial sludge
height (h0), transient height (h(t)) and high volume fraction Py(φ) and R(φ) data from
filtration testing.

Py(φ) was described by the following composite functional form with 9 fitting
parameters, two of which, b and φcp, require definition by the user.

 
 
 0 φ < φg 
 − k1 
  ( φcp − φ )( b + φ − φ g )  
Py (φ ) =  a1
  φ g ≤ φ < φ p  …(3.1.4)
 (φ − φg ) 
  
 − k2 
 ( φcp − φ )( b + φ − φ g 2 )  
 a2 φ > φp 
( φ − φg 2 ) 
  

This complex functional form, although cumbersome, was necessary since


Py(φ) varies by many orders of magnitude near φg, as shown in the next chapter. The
φcp and b values were manipulated until the best fit was achieved (note that this user
intervention was not ideal, and alternative functional fits are currently under
investigation). The output from the analysis included fits to Py(φ) and R(φ) from the
initial concentration of the batch settling test up to the final concentration of the
filtration compressibility test, and a measurement of φg.

Equilibrium Batch Settling

Alternatively, φg was also determined through equilibrium settling tests


(Green, et al., 1996). The solids concentration at the top of the sediment in a settling
test is φg, but the average solids concentration, <φ>, is higher than φg as some
compression of the network always occurs due to the weight of solids present. The

93
compressive pressure supplied at the base of the sediment due to the self-weight was
calculated from (Buscall and White, 1987):

p p ( 0 ) = ∆ρgφ0 h0 …(3.1.5)

Varying φ0 or h0 changed the maximum compressive force in the settled


sediment, and therefore <φ>. Extrapolating to <φ> = 0 gave the gel point.

The experimental method involved a range of settling experiments all


commencing with the same initial solids concentration, but with three different initial
heights. The suspensions were allowed to settle until the sediment interface ceased to
move. <φ> for each test was determined from the final sediment height (see equation
2.3.16), plotted against either h0 or pp(0) and linearly extrapolated to zero. A similar
technique holding h0 constant and changing φ0 (Usher, 2002) was not used here.

3.1.7 Oedometer Testing

This work used stepped-pressure filtration and oedometer testing to determine


the material properties of a kaolin sample over a range of pressures. The methods
were essentially the same – a load was applied to a particulate network and water was
extruded through a porous boundary. The only physical difference was that the
oedometer had two porous stones whereas the filtration rig had one semi-permeable
membrane. When the self-weight of the sample was insignificant, the filtration test
was essentially one half of the oedometer test.

cv was measured using oedometer testing following the ASTM Standard Test
Method for One-Dimensional Consolidation Properties of Soils (ASTM, 2001) or the
Australian Standard Methods of Testing Soil for Engineering Purposes: Part F - Soil
Strength and Consolidation Tests: F6.1 - Determination of the One-Dimensional
Consolidation Properties of a Soil (AS, 1998).

Figure 3.1.6 shows a schematic of an oedometer. The saturated, pre-


consolidated test sample was held inside a metal ring between two porous stones. The
bottom stone was fixed in position while the upper stone could move inside the metal

94
ring. The upper stone was under a metal loading cap through which a pressure could
be applied to the specimen. The entire assembly sat in a water bath to which the pore
water had free access. The deformation of the sample (and therefore the degree of
compression of the sample) was measured by means of a dial gauge above the loading
cap. The oedometer test for a single applied load was used to give results of gauge
reading (or bed height, h) versus time, t.

Load
2 cm
Porous stones

Sample

Confining ring Water

Figure 3.1.6: Schematic of oedometer

cv was determined from oedometer testing following the traditional soil


consolidation approach, where experimental oedometer results (h-t) were graphically
fitted to the theoretical solution of the consolidation governing equation (U-T, where
U is the degree of consolidation and T is the dimensionless time). Terzaghi’s theory
gives the dimensionless time, Tv, as:

cv t
Tv = …(3.1.6)
d2

where d is the nominal maximum pore drainage length (equal to half the average
thickness of a sample with two porous boundaries):

h + ht → ∞
d = t =0 …(3.1.7)
4

The initial and final heights, ht=0 and ht→∞, were determined from the sample
heights corresponding to no consolidation (U = 0) and complete consolidation (U =
1), respectively. Thus, d was dependent upon the graphical method used.

95
Casagrande’s h-log(t) (Casagrande and Fadum, 1940) and Taylor’s h-t½
(Taylor, 1942) methods were the traditional graphical methods used to calculate cv.
Other graphical methods that involve similarities between the theoretical and
experimental curves include the velocity method (using a plot of log(dh/dt) versus
log(t), Parkin, 1978), the h-t/h method (Sridharan and Prakash, 1993), the rectangular
hyperbola method (involving a plot of t versus t/h, Sridharan and Sreepada Rao, 1981)
and the inflection point method (Mesri, et al., 1999). However, these other methods
were not further investigated here.

Following Casagrande’s log-time method, a plot of bed height versus log time
was used to find the time at 50% consolidation (t50). From the theoretical linear
solution of the consolidation equation, T at U = 0.5 is 0.197. Therefore, cv is
determined by equation 3.1.8:

0.197 d 2
cv = …(3.1.8)
t 50

Casagrande’s method of marking U = 1 as the intercept of the two linear


portions of the plot generally yields a larger t50 value, and consequently a smaller cv
value than other methods (Olson, 1986).

cv was also determined using Taylor’s root-time method. The dial readings
from the oedometer test were plotted against the square root of the time. The
intercept of the experimental data with 1.15 times the slope of the linear section gave
the time at 90% consolidation (t90). The theoretical value of Tv corresponding to U =
0.9 is 0.848. Therefore, the coefficient of consolidation, cv, was calculated from:

0.848d 2
cv = …(3.1.9)
t 90

Casagrande’s and Taylor’s methods were used here to calculate cv for a kaolin
sample from two sets of oedometer results at 50, 100, 200, 400 and 800 kPa. The
void ratio for each pressure was taken as the void ratio at the height corresponding to
100% consolidation (U = 1).

96
3.2 Calculations

The analytical procedures for characterising water and wastewater sludges


differed since the former behaved in a classical manner and the latter exhibited little
or no cake formation.

3.2.1 Water Treatment Sludge Characterisation

The majority of testing of water treatment sludges presented here used


stepped-pressure filtration tests combined with equilibrium settling tests for gel point
measurement. A flow chart for the analysis procedure is shown in Figure 3.2.1.

Stepped-Pressure Equilibrium Batch


Filtration Settling

∆P, φ∞, β2 φg

 
p2 
φ
Py (φ ∞ ) = p1  ∞  − 1
 φ g  
  

Curve Fit β2(∆P) Curve Fit β2(φ)

2  1  −1
R(φ ∞ ) = (1 − φ ∞ )2
1
 − 1 dβ 2  1 
2 φ D(φ ∞ ) =
1
dβ  0 φ ∞   − 
2 dφ  φ0 φ ∞ 
d∆P

Py(φ), R(φ), D(φ)

Figure 3.2.1: Flowchart of water treatment sludge characterisation using stepped-pressure


filtration and equilibrium batch settling

The results from the stepped pressure compressibility test were fitted with a
power-law (see equation 2.1.6) using φg from equilibrium batch settling analysis to
give Py(φ). The results from the stepped-pressure permeability test were then

97
combined with Py(φ) to give R(φ) or D(φ) (see equations 2.4.32 and 2.4.33), which are
related by equation 2.2.18. This required curve fits to β2(∆P) or β2(φ) – power-laws
were used here. While the errors in β2, ∆P and φ were very small, the curve fitting
introduced some error, which is highlighted wherever necessary in the results. This
method was only applicable for modelling high-pressure filtration when settling
effects were assumed negligible.

Some water treatment sludges were also characterised for thickener modelling
purposes using transient batch settling in conjunction with stepped-pressure filtration
tests. An iterative analysis scheme (shown in Figure 3.2.2) was necessary since the
initial curve fit for Py(φ) required φg, which was only provided after settling analysis,
which required R(φ), which proceeded from Py(φ). φf from batch settling was used as
the initial estimate for φg.

Stepped-Pressure Transient Batch


Filtration Settling

∆P, φ∞, β2

h(t)
Estimate φg

 
p2 
φ
Py (φ ∞ ) = p1  ∞  − 1
 φ g  
  
Batch Settling
Curve Fit β2(∆P) Analysis

2  1 
R(φ ∞ ) = (1 − φ ∞ )2
1
 −
dβ 2  φ 0 φ ∞ 
d∆P Low R(φ), φg

dPy (1 − φ )2
D (φ ) =
dφ R(φ )

Py(φ), R(φ), D(φ)

Figure 3.2.2: Flowchart of water treatment sludge characterisation using stepped-pressure


filtration and transient batch settling

98
3.2.2 Wastewater Treatment Sludge Characterisation

Wastewater treatment sludges showed non-traditional filtration behaviour and


previously have been very difficult to characterise. Multiple single-pressure filtration
runs were required, since there was little or no cake formation. As illustrated in the
theory chapter, D(φ) at low φ was expected to be high for non-traditional materials.
Thus, transient batch settling analysis was also used.

After initially determining ρp using a density bottle and φ0 from oven-drying,


single-pressure filtration tests over a range of pressures were combined with batch
settling tests to characterise the sludge over a range of solids concentrations,
following the procedure outlined in Figure 3.2.3.

Single-Pressure Transient Batch


Filtration Settling

∆P, φ∞, D(φ∞)


h(t)
Estimate φg

 
p2 
φ
Py (φ ∞ ) = p1  ∞  − 1
 φ g   Batch Settling
  
Analysis

dPy (1 − φ ∞ )2
R(φ ∞ ) =
dφ D (φ ∞ ) Low R(φ), φg

dPy (1 − φ )2
D(φ ) =
dφ R(φ )

Py(φ), R(φ), D(φ)

Figure 3.2.3: Flowchart of wastewater treatment sludge characterisation from single-pressure


filtration and transient batch settling

The results from each filtration run were used to give φ∞ and D(φ∞) for each
pressure via the logarithmic curve fitting described in Section 3.1.5. As with the case
for transient settling analysis combined with stepped-pressure filtration, φg was first

99
estimated, and an iterative approach implemented. After fitting Py(φ∞) using a power-
law, the filtration results were manipulated to give R(φ∞) (from equation 2.2.18):

dPy (1 − φ ∞ )2
R (φ ∞ ) = …(3.2.1)
dφ D (φ ∞ )

The batch settling analysis could now proceed, giving low R(φ) data and φg. If
this value of φg was significantly different from the previous estimate of φg, it was
then used to recalculate the high R(φ) data and the batch settling analysis was repeated
until φg was stable (usually only one or two iterations were necessary). D(φ) was then
calculated from Py(φ) and R(φ) using equation 2.2.18.

100
3.3 Materials

This section outlines the source of a range of water and wastewater treatment
sludges and other materials characterised using the above methods.

3.3.1 Water treatment sludges

Ferric and alum water treatment sludges were sampled from a range of sites in
Australia and the United Kingdom, and tested either on-site or in the laboratory. The
water treatment works, including the coagulant type and the operating company (UU
– United Utilities plc, UK; YW – Yorkshire Water plc, UK; and BCC – Brisbane City
Council, Australia), are listed in alphabetical order:

- Arnfield WTP (Fe3+, UU);

- Eccup WTP (Al3+, YW);

- Gellibrand WTP (Fe3+, Barwon Water, Australia);

- Hodder WTP (Al3+, UU);

- Huntington WTP (Al3+, UU);

- Langsett WTP (Fe3+, YW);

- Oswestry WTP (Al3+, UU);

- Swan Reach WTP (Al3+/PACl, SA Water, Australia);

- Tailem Bend WTP (Al3+/PACl, SA Water, Australia);

- Thornton Steward WTP (Al3+, YW);

- Top Hill Low WTP (Al3+, YW); and

- Wybersley WTP (Fe3+, UU).

101
In addition, sludges from controlled pilot plant trials performed on water from
Auravale Lake, near Melbourne, Australia were tested.

3.3.2 Wastewater treatment sludges

Wastewater treatment sludges from a variety of treatment plants were tested


using single-pressure filtration:

- Carrum WWTP (South East Water, Australia);

- Colac WWTP (Barwon Water, Australia);

- Ellesmere Port Effluent Treatment Facility (UU);

- Lilydale WWTP (Yarra Valley Water, Australia);

- Luggage Point WWTP (BCC);

- Mornington WWTP (South East Water, Australia); and

- Oxley Creek WWTP (BCC).

Abattoir wastewater sludge from Wagstaff, Victoria, and dairy sludge from
Bonlac, Cobden, were also tested using single-pressure filtration. In addition to
pressure filtration, samples from Luggage Point WWTP were tested using transient
batch settling.

3.3.3 Synthetic Sewage Sludges

Several synthetic sludges were manufactured with the aim of producing a


synthetic sludge that mimicked the permeability and compressibility characteristics of
real wastewater sludges. Based on the procedure proposed by Ormeci and Vesilind,
2000, synthetic sludges were made using stock solutions of polystyrene particles
(approximately 0.25µm diameter, supplied at 8 wt%), unbleached Kraft process paper
fibre (see Section 3.3.4), alginate of three different molecular weight’s designated as

102
low, medium and high viscosity, and calcium chloride as a coagulant. Alginates are
polysaccharide-based polymers that can range in molecular weight from 32,000 to
200,000 Da (Gregor, et al., 1996).

In addition, commercial yeast cells were used in some experiments as a


substitute for polystyrene particles in order to examine the effect of larger, deformable
particles. The yeast cells were considered as ovoid particles 5µm in medium diameter
that behaved as deformable particles due to the flexibility of the cell walls, thereby
decreasing porosity (Meireles, et al., 2004). Commercial gelatine, a heterogeneous
mixture of water-soluble proteins (Windholz, 1983), was used as a replacement for
alginate in some experiments in order to represent the reported nature of ECP’s that
they are less charged, have shorter chain lengths and absorb to surfaces better than
alginate. FeCl3 was used as the coagulant for gelatine and neutralised with NaOH to
pH 7. LT-20 (supplied by CIBA), a high molecular weight polymeric chain, was used
as a flocculant for the gelatine-based synthetic sludge.

Stock solutions of each component were prepared by diluting the components


with water. Synthetic sludges were prepared by mixing stock solutions of fibres,
particles and ECP substitutes in a beaker using an overhead stirrer. The coagulant
was then added slowly. A typical formulation for the synthetic sludges is shown in
Table 3.3.1. The resultant suspensions consisted of macroscopically coagulated flocs
with a gel point typically around 2 to 3 wt%.

Table 3.3.1: Standard formulation for synthetic sludges

Component Description wt% dry wt%

Fibre Kraft process fibres 0.21 39.12

Particles Polystyrene 0.42 41.31

ECP substitute Alginate 0.45 19.56

Coagulant CaCl2 0.42 -

103
3.3.4 Pulp and Paper

A sample of unbleached Kraft process paper fibres was tested using stepped-
pressure filtration. Visual observation indicated that the fibres were mono-dispersed
in length of about 1 mm. They were supplied at 30 wt% and diluted to 1 to 3 wt%.
Ultrasonication of the fibre stock solutions was sometimes necessary in order to
achieve a homogeneous dispersion. The gel point of these fibres was approximately 2
wt%.

3.3.5 Kaolin

The material characteristics of a kaolin sample undergoing one-dimensional


compression were determined using both oedometer and filtration testing. The
material used for testing was an acid washed kaolin sample (specific gravity
approximately 2.65 g/cm3) supplied by BDH. The samples for oedometer testing
were prepared at approximately φ = 0.27 v/v with distilled water, pre-consolidated at
25 kPa to φ = 0.433 v/v and trimmed directly into the cell ring. For filtration testing,
the samples were prepared at φ = 0.237 v/v for ease of loading into the apparatus and
because the control algorithm (designed for suspensions) required φ0 to be less than
φ∞ to a certain extent.

104
3.4 Case Studies

Several on-site case studies were performed in order to validate the plate-and-
frame models. In chronological order, these sites were:

- Langsett WTP, Yorkshire Water;

- Hodder WTP, United Utilities;

- Arnfield WTP, United Utilities; and

- Thornton Steward WTP, Yorkshire Water.

Both fill-only and fill-and squeeze presses were in use at Langsett WTP, while
the other sites used just fill-only presses. Fill-and-squeeze pilot-scale tests from
Hodder WTP were also made available.

In general, the validation work at each site involved experimental material


characterisation and measurement of the press performance. The material
characteristics were determined using the stepped-pressure filtration techniques
outlined above. The physical dimensions of the filter presses were measured to
calculate the total volume of the cavities within the presses, Vpress, and the total
available surface area for filtration, Apress. This included consideration of the size of
the baffles, and feed and filtrate ports. For each filtration cycle, the filtrate flowrate
and applied pressure were measured with time, and the initial and final solids
measured by oven drying. Gauges that were either mounted on the inlet of the press
or on the feed pumps indicated the pressure, which was recorded periodically. The
flowrate was either measured directly using either a flowmeter or a tank level, or
indirectly based on the number of strokes of the feed pump.

The validation and optimisation results for the case studies are detailed in
Section 5.2.

105
4 Material Characterisation
Results

This chapter details the results for the measurement of the compressional
rheology material parameters of a wide range of industrial water and wastewater
treatment sludges using pressure filtration and batch settling techniques. This
includes the application and validation of new analysis techniques for wastewater
treatment sludges, covering both the scaling behaviour and fundamental material
characterisation. Also included are the results for the development of synthetic
sewages sludges, the characterisation of kaolin using both filtration and consolidation
testing methods, and an overview of the characteristics of a range of other sludges.

4.1 Potable Water Treatment Sludges

A number of water treatment sludges from industrial sources were


characterised for the purpose of predicting the sludge behaviour in dewatering. By
characterising the sludge, changes to performance can be clearly attributed to either
changes to sludge characteristics or processing changes (such as φ0, Rm and ∆P). Such
techniques can also be used to optimise additive dosage and identify floc degradation
during processing. Sludges from several industrial sites were characterised more than
once, allowing comparisons of daily and seasonal variations. Comparisons between
sites and coagulant types were also made. Some of the water treatment results are
previously published (Abd Aziz, et al., 2000, Abd Aziz, 2004, Harbour, et al., 2004).

Solids Density

ρp was assumed constant at 3 g/cm3 for both ferric and alum water treatment
sludges rather than allow for changes from sludge to sludge. Although differences of
up to 0.5 g/cm3 were observed, these measurements were inherently inaccurate and it

106
was considered more accurate to keep ρp constant. The effect of using 2 or 3 g/cm3 is
shown in Figure 4.1.2 and Figure 4.1.2, showing that ρp only effects the magnitude of
φ, and not the calculation of the magnitude of D(φ), since gravity is ignored in
pressure filtration. Therefore, by assuming constant ρp, comparisons on an equal
volume fraction basis also correspond to mass fraction trends.

200 200
3 g/cm3 data 3 g/cm3 data
Compressive Yield Stress, Py (φ )

180 180
3 g/cm3 fit 3 g/cm3 fit
160 2 g/cm3 data 160 2 g/cm3 data
2 g/cm3 fit 2 g/cm3 fit
140 140

120 120
(kPa)

100 100

80 80

60 60

40 40

20 20

0 0
0 0.02 0.04 0.06 0.08 0.1 0% 5% 10% 15% 20%

Solids Volume Fraction, φ (v/v) Solids Weight Fraction, wt%

Figure 4.1.1: Effect on Py(φ) of using ρs = 2 or 3 g/cm3 (Alum water treatment sludge, Hodder
WTP, 06/11/01)

1.E-07 1.E-07
3 g/cm3 data 3 g/cm3 data
Solids Diffusivity, D (φ ) (m /s)

3 g/cm3 fit 3 g/cm3 fit


2

2 g/cm3 data 2 g/cm3 data


2 g/cm3 fit 2 g/cm3 fit

1.E-08 1.E-08

1.E-09 1.E-09
0 0.04 0.08 0.12 0% 5% 10% 15% 20%

Solids Volume Fraction, φ (v/v) Solids Weight Fraction, wt%

Figure 4.1.2: Effect on D(φ) of using ρs = 2 or 3 g/cm3 (Alum water treatment sludge, Hodder
WTP, 06/11/01)

107
4.1.2 Ferric Water Treatment Sludges

Ferric coagulants are often used to treat raw waters with high colour and low
turbidity. High colour waters require large coagulant doses – therefore the ratio of
particles to coagulant is low and the sludge consists predominantly of the precipitated
coagulant.

Langsett Water Treatment Plant

Four sets of samples (thirteen samples in total) from Langsett WTP from
November/January 2000/01, June 2001, November 2001 and July 2003 were
characterised by a number of people using stepped-pressure filtration testing (see
Table 4.1.1). The coagulant and flocculant doses were unknown and fluctuated
depending on the raw water quality and the solids concentration. From November
2001 onwards there was a large increase in coagulant dose to ensure adequate NOM
removal, corresponding to a change from ‘sweep’ to ‘enhanced’ coagulation.

Table 4.1.1: Langsett WTP sludge samples

Sample Operator φ0 (v/v) Sample Operator φ0 (v/v)

22/11/00 MRT ~0.01 05/06/01 ADS 0.0091

03/01/01 AT 0.0096 06/06/01 ADS 0.0067

08/01/01 AT 0.0061 07/06/01 ADS 0.0098

09/01/01 AT 0.0076 11/06/01 ADS 0.0104

11/01/01 AT 0.0099 12/11/01 RDF 0.0121

17/01/01 AT 0.0093 09/07/03_1 ADS 0.0101

09/07/03_1 ADS 0.0106


MRT – Martin R. Tillotson, Yorkshire Water
AT – Anna Taylor, Yorkshire Water
ADS – Anthony D. Stickland, The University of Melbourne
RDF – Rene D. Frost, United Utilities

The samples from 07/06/01 and 11/06/01 included high pressure testing using
a manual filtration rig (at 628 kPa). All the samples were taken from the bleed valve
of the filter press feed pumps except the two samples from July 2003, which were

108
from the underflow of the two thickening tanks and were characterised using transient
batch settling in conjunction with pressure filtration.

The compressive yield stress results (see Figure 4.1.3) were, with a few
exceptions, very consistent, suggesting that the choice of ferric as the coagulant
dictated equilibrium solids concentrations at high pressures rather than plant
operations such as flocculant dose or preparation method. From transient settling of
the 09/07/03 samples, φg was 0.0070 v/v.

700
22/11/00
Compressive Yield Stress, Py (φ )

03/01/01
600
08/01/01
09/01/01
500 11/01/01
17/01/01
05/06/01
400
(kPa)

06/06/01
07/06/01
300 11/06/01
12/11/01
200 09/07/03_1
09/07/03_2
Average Py
100

0
0 0.05 0.1 0.15

Solids Volume Fraction, φ (v/v)

Figure 4.1.3: Py(φ) results for stepped-pressure filtration of Langsett WTP sludge

As with the Py(φ) results, the R(φ) results (see Figure 4.1.4) were grouped
tightly. There was a small variation between summer and winter, corresponding to
the changes in fluid viscosity due to temperature, although in England this variation
was minimal. The R values for the two samples from the thickening tanks
(09/07/03_1 and _2) were slightly below average, suggesting that there may have
been some small degradation of flocs in between thickening and filtering, or that
sampling at the press pumps involved shearing the flocs. Such results suggest further
investigation using a comprehensive ‘shear audit’, which would involve

109
characterising samples from different stages of processing at the same time to identify
floc degradation.

1E+16
Hindered Settling Function, R (φ ) Average R 22/11/00 03/01/01
08/01/01 09/01/01 11/01/01
17/01/01 05/06/01 06/06/01
07/06/01 11/06/01 12/11/01
09/07/03_1 09/07/03_2

1E+15
(Pas/m )
2

1E+14

1E+13
0 0.05 0.1 0.15

Solids Volume Fraction, φ (v/v)

Figure 4.1.4: R(φ) results for stepped-pressure filtration of Langsett WTP sludge

The D(φ) results presented in Figure 4.1.5 clearly showed that the daily and
seasonal variations were small. The samples from November/January 2000/01 were
very similar, except 11/01/01, which had a slightly higher permeability. Likewise, the
samples from Summer 2001 showed good consistency. The difference between the
thickening tanks may have been due to different shear conditions or polymer dose.

The power-law functions (equations 2.1.6, 2.1.7 and 2.4.51) used as averages
fitted the filtration data reasonably well, although in some cases they did not quite
represent the curvature of the experimental data. The parameters are given in Table
4.1.2.

110
1.0E-07

Solids Diffusivity, D (φ ) (m /s)


2

1.0E-08

Average D 22/11/00 03/01/01


08/01/01 09/01/01 11/01/01
17/01/01 05/06/01 06/06/01
07/06/01 11/06/01 12/11/01
09/07/03_1 09/07/03_2
1.0E-09
0 0.05 0.1 0.15

Solids Volume Fraction, φ (v/v)

Figure 4.1.5: D(φ) results for stepped-pressure filtration of Langsett WTP sludge

Table 4.1.2: Fitting parameters for Langsett WTP averages

Parameter Value Parameter Value Parameter Value

φg (v/v) 0.0070 r1 (Pa s/m2) 5.246 x 1013 d1 (m2/s) 3.002 x 10-5

p1 (Pa) 48.505 r2 -23.766 d2 2.248

p2 3.248 d3 25.766

Validation of Experimental Results

Comparing the raw data with model predictions of filtration validates the
experimental results. A stepped-pressure model based on the linear approximation
(Landman and White, 1997, Scales, et al., 2001) was used here. The inputs to the
model were Py(φ), D(φ), ∆P (as a stepped function of time), h0 and φ0 for Langsett
05/06/01. Figure 4.1.6 shows the comparison for the compressibility stepped-pressure
test. The extent of filtration was predicted very well, but the rate of filtration at low
pressures was too slow.

111
18000
Filtration Data 254 kPa
16000
Model Results
14000 200 kPa

12000 150 kPa


Time, t (s)

10000 100 kPa

8000 50 kPa
6000

4000 20.2 kPa

2000

0
0 0.00005 0.0001 0.00015 0.0002
2 2 2
(Specific Filtrate Volume) , V (m )

Figure 4.1.6: Experimental data and model predictions for compressibility stepped -pressure
filtration test of Langsett WTP, 05/06/01

6000
Filtration Data
5000 Model Results 300 kPa

4000
Time, t (s)

200 kPa
3000 150 kPa

100 kPa
2000
50.3 kPa
1000
20 kPa

0
0 0.0002 0.0004 0.0006
2 2 2
(Specific Filtrate Volume) , V (m )

Figure 4.1.7: Experimental data and model predictions for the permeability stepped -pressure
filtration test of Langsett WTP, 05/06/01

112
The raw experimental data and model predictions for the permeability test for
Langsett WTP 05/06/01 are shown in Figure 4.1.7. The rate of filtration was
accurately predicted except, as with the Py(φ) results, at lower pressures. This
reflected the poor fit of the power-law function of R(φ) at low volume fractions.
Other functional forms are required for modelling low-pressure filtration and
thickening (as used by the batch settling analysis software), but were adequate for the
purposes of high-pressure filtration (> 50 kPa).

Transient Batch Settling

The two samples from Langsett WTP from 09/07/03 (called “S1” and “S2”)
were diluted to 0.00303 v/v and 0.00319 v/v and allowed to settle in large, graduated
cylinders. The sediment height was recorded with time and results were analysed
using the batch settling software in conjunction with the high volume fraction results
from pressure filtration testing following the method outlined in Section 3.2.1 to give
low volume fraction R(φ) and an estimate of φg. The experimental results and analysis
curve fits are presented in Figure 4.1.8, showing good agreement.

0.16
S1 Settling Results
S1 Curve Fit
0.14
Interfacial Height, h (m)

S2 Settling Results
S2 Curve Fit
0.12

0.1

0.08

0.06

0.04
0 2000 4000 6000
Time, t (s)

Figure 4.1.8: Transient batch settling results and analysis curve fits for Langsett WTP, 09/07/03

113
The compressibility results are presented in Figure 4.1.9, showing that the
thickener samples had slightly higher Py(φ) than the Langsett average at high φ. The
results were fitted well at high φ by the power-law function and at low φ the BSAM
composite function. The interpolation between φg and the first datum point from
pressure filtration testing was subjective as the strength of the network varies in a
highly non-linear fashion. The parameters used for equation 3.1.2 for the samples
from thickener 1 and 2 are given in Table 4.1.3.

1000
Compressive Yield Stress, Py (φ )

100

10
(kPa)

0.1
Langsett (average)
S1 Filtration Results
0.01 S1 Curve Fit
S2 Filtration Results
S2 Curve Fit
0.001
φg
0 0.04 0.08 0.12

Solids Volume Fraction, φ (v/v)

Figure 4.1.9: Py(φ) results from stepped-pressure filtration and transient batch settling for
Langsett WTP, 09/07/03

Table 4.1.3: Py(φ) fitting parameters for Langsett WTP, 09/07/03

Parameter Sample 1 Sample 2 Parameter Sample 1 Sample 2

φg 0.0068 0.0071 φg2 0.0063 0.0066

a1 1.4419 1.4752 a2 1.4902 1.5277

b1 0.001 0.001 b2 0.001 0.001

k1 15.2008 16.0916 k2 16.1274 17.1916

φcp 0.4 0.4 φp 0.0184 0.0181

114
The R(φ) results for the two thickener samples from Langsett WTP 09/07/03
are shown in Figure 4.1.10 along with the average stepped-pressure filtration results
for Langsett WTP. Of the two samples, sample 2 was slightly more permeable than
sample 1 at high φ. The batch settling analysis shows that R(φ) changed by several
orders of magnitude between free settling (φ = 0) and φg, and by another three to four
orders of magnitude change as the network concentration increased beyond φg. The
power-law function that described the Langsett average, which fitted the high φ data
reasonably well, failed to describe the low φ behaviour.

1.E+16
Hindered Settling Function, R (φ )

1.E+15

1.E+14

1.E+13
(Pas/m )
2

1.E+12

1.E+11
Langsett (average)
1.E+10 S1 Filtration Results
S1 Setlling Results
S1 Curve Fit
1.E+09 S2 Filtration Results
S2 Settling Results
S2 Curve Fit
1.E+08
φg
0 0.04 0.08 0.12

Solids Volume Fraction, φ (v/v)

Figure 4.1.10: R(φ) results from stepped-pressure filtration and transient batch settling for
Langsett WTP, 09/07/03

D(φ) was calculated from dPy(φ)/dφ and R(φ) using equation 2.2.18, and
presented in Figure 4.1.11. At high φ, the settling analysis curve fits showed the same
magnitude as the filtration results, but did not match the slope of D(φ). This was due
to greater error in their high φ curve fit for Py(φ). The composite function exhibited a
maximum between φg and the lowest filtration measurement, corresponding to the
large increase in R(φ) and the sharp change in Py(φ). The exact magnitude depended
on the interpolation used.

115
1.0E-05
Langsett (average)
S1 Filtration Results

Solids Diffusivity, D (φ ) (m /s)


S1 Settling Results

2
S1 Curve Fit
S2 Filtration Results
1.0E-06 S2 Settling Results
S2 Curve Fit

1.0E-07

1.0E-08

1.0E-09
φg
0 0.04 0.08 0.12

Solids Volume Fraction, φ (v/v)

Figure 4.1.11: D(φ) results from stepped-pressure filtration and transient batch settling for
Langsett WTP, 09/07/03

Arnfield Water Treatment Plant

Arnfield WTP (United Utilities) is a ferric water treatment plant in the Peak
District that treats approximately 90 ML of raw water per day. Lime is added prior to
dewatering with a fill-only plate-and-frame press. Short filtration times and high cake
solids are observed, suggesting that the addition of lime aids in the dewatering
process. Conversely, the filter press cloths are regularly cleaned, which may
contribute to the enhanced performance.

Chemically, CaO may undergo hydrolysis, dissolution and/or complexation,


any of which could effect sludge dewatering. The compressional rheology approach
was used to quantitatively determine the effect of lime addition. The press feed (after
lime addition) was sampled in January 2001, and samples of sludge were taken before
and after lime addition in June 2002. The material characteristics of the sludges were
measured using filtration and equilibrium batch settling tests. Model predictions of
the fill-only press performance are presented in Section 5.2.5 in a case study used to
give an analysis of the benefits of adding lime to ferric water treatment sludges, and to

116
investigate the effects of regular cloth cleaning. The three samples from Arnfield
WTP are presented in Table 4.1.4.

Table 4.1.4: Arnfield WTP sludge samples

Sample Operator φ0 (v/v) φg (v/v) pH

05/01/01 (with lime) AT 0.0138 - 11.78

14/06/02 (pre-lime) ADS 0.0122 0.0094 7.01

14/06/02 (with lime) ADS 0.0131 0.0069 11.77


AT – Anna Taylor, Yorkshire Water
ADS – Anthony D. Stickland, The University of Melbourne

The solids density was measured as 2830 ± 100 kg/m3 for 14/06/02 (pre-lime)
and 2660 ± 400 kg/m3 for 14/06/02 (with lime) but was assumed to be 3000 kg/m3 for
consistency. The change in φ0 for 14/06/02 corresponded to a lime dose of 28.5
kg/tds. The amount of lime added at the plant was not controlled and was based on
four 25kg bags per filtration run, irrespective of solids concentration or feed rate.
Thus, the dose may have varied from 20 – 40 kg/tds. Increasing φ0 improves the press
performance, as illustrated in Chapter 5. By characterising the effect of lime addition
on the dewatering characteristics, the improvement in performance due to changes in
characteristics was distinguishable from changes to φ0.

φg was measured using equilibrium batch settling. The samples from the
14/06/02 were diluted to 0.00245 v/v and 0.00262 v/v, and allowed to settle to
equilibrium in flat-bottomed measuring cylinders. The results for three different
initial heights for each sample are shown in Figure 4.1.12. φg was calculated by linear
extrapolation to h0 = 0. The small drop in φg with the addition of lime (such that a
lower volume fraction was required to form a network) suggested that the lime helped
to form a more open structure.

The results for stepped-pressure compressibility tests and power-law fits (see
equation 2.1.6) are presented in Figure 4.1.13. The compressibility results for the
14/06/02 samples were limited to medium pressures due to a malfunctioning
compressor. The results show that the material before the addition of lime had lower
Py(φ) at a given φ, suggesting that lime increases the network strength of ferric sludge.

117
0.012

Final Average Volume Fraction, <φ >


0.01
φ g = 0.0094 v/v
0.008

φ g = 0.0069 v/v
(v/v)

0.006

0.004
140602 (pre-lime)
140602 (with lime)
0.002 Linear (140602 (pre-lime))
Linear (140602 (with lime))
0
0 10 20 30
Initial Height, h 0 (cm)

Figure 4.1.12: Gel point determination via equilibrium batch settling for Arnfield WTP samples
from 14/06/02

300
Compressive Yield Stress, Py (φ )

05/01/01 (with lime)


14/06/02 (pre-lime)
250 14/06/02 (with lime)

200
(kPa)

150

100

50

0
φg
0 0.05 0.1 0.15

Solids Volume Fraction, φ (v/v)

Figure 4.1.13: Py(φ) results for stepped-pressure filtration of Arnfield WTP sludge

118
The results for stepped-pressure permeability tests were combined with the
Py(φ) results to give R(φ), as shown in Figure 4.1.14, which were fitted using power-
law functions (see equation 2.1.7). The results show that the permeability of the
sludge increased slightly with the addition of lime. However, these results were not
consistent with the 05/01/01 result. This may have been due to different lime doses or
seasonal changes to sludge characteristics. Despite the different seasons, the tests
were performed at similar temperatures (~12°C) and the viscosity was unlikely to be
different.

1.E+15
Hindered Settling Function, R (φ )

05/01/01 (with lime)


14/06/02 (pre-lime)
14/06/02 (with lime)
(Pas/m )
2

1.E+14

1.E+13
0 0.05 0.1 0.15

Solids Volume Fraction, φ (v/v)

Figure 4.1.14: R(φ) results for stepped-pressure filtration of Arnfield WTP sludge

The Py(φ) and R(φ) results were combined to give D(φ), as shown in Figure
4.1.15. The 14/06/02 results show that the diffusivity increased by a factor of two
with the addition of lime. The difference between the post-lime samples suggested
that the sample from 05/01/01 might have been either under- or over-dosed, or that
there was not as much benefit from adding lime. The pre-lime sample was very
similar to other ferric sludges (see Figure 4.1.18), and both results for the addition of
lime showed an improvement over the average ferric characteristics.

119
1.E-07

Solids Diffusivity, D (φ ) (m /s)


2

1.E-08

05/01/01 (with lime)


14/06/02 (pre-lime)
14/06/02 (with lime)
1.E-09
0 0.05 0.1 0.15

Solids Volume Fraction, φ (v/v)

Figure 4.1.15: D(φ) results for stepped-pressure filtration of Arnfield WTP sludge

The parameters for the power-law functions are given in Table 4.1.2, which
were used in Section 5.2.5 for modelling the effect of lime addition on fill-only plate-
and-frame presses at Arnfield WTP.

Table 4.1.5: Fitting parameters for Arnfield WTP sludges

05/01/01 14/06/02 14/06/02


Parameter
(with lime) (pre-lime) (with lime)

p1 (Pa) 67.5731 44.2461 33.3245

p2 2.9558 3.2540 3.1051

φg (v/v) 0.0069 0.0094 0.0069

r1 (Pa s/m2) 3.20 x 1013 1.53 x 1013 1.60 x 1013

r2 -18.2098 -22.9407 -17.6344

d1 (m2/s) 1.5252 x 10-5 3.7042 x 10-5 3.3214 x 10-5

d2 1.9558 2.2540 2.1051

d3 20.2098 24.9407 19.6344

120
This analysis only gave an estimate of the impact of lime addition. David Lam
and Catherine Cappone undertook a more detailed analysis in an undergraduate
research project at the University of Melbourne in August to November 2003.
Although the results were inconclusive, the indications were that anhydrous lime
affected both alum and ferric sludges, but hydrated lime and sodium hydroxide did
not. This suggested that any benefit was due to the physical mechanism of lime
hydration and floc dehydration rather than chemical mechanisms of bond formation or
breakage due to pH change or complexation.

Other Ferric Water Treatment Sludges

Several other ferric water treatment sludge samples were characterised using
stepped-pressure filtration as summarised by Table 4.1.6. Gellibrand WTP is in
Australia, while Wybersley WTP is in the United Kingdom. Two doses of FeCl3 were
used with sludges from the Auravale pilot plant trial – 18.6 and 24.1 mg Fe/L,
corresponding to sweep and enhanced coagulation. Unlike the treatment plant
samples, the pilot plant trials did not include flocculant addition.

Table 4.1.6: Other ferric water treatment sludge samples

Ferric Dose φ0
Source Date Operator
(mg Fe/L) (v/v)

Auravale Pilot Plant 03/05/00 NJA 18.6 0.0043

Auravale Pilot Plant 02/05/00 NJA 24.1 0.0027

Gellibrand WTP 27/03/00 NJA 2.4 0.0052

Wybersley WTP 17/05/01 AT Unknown 0.0089


NJA – Nevil J. Anderson, CSIRO
AT – Anna Taylor, Yorkshire Water

The Py(φ) results for all the ferric sludges are presented in Figure 4.1.16. In
general, they had low equilibrium solids concentrations (< 0.015 v/v at 300 kPa)
compared to other sludges such as minerals tailings or wastewater sludges. There was
some variation between the ferric samples. Gellibrand WTP and Wybersley WTP had
very similar Py(φ), while the Langsett average had a higher network strength at a
given φ. The lower-dosed Auravale sample had slightly lower equilibrium solids than
the higher-dosed sample, but both showed higher values than the plant samples.

121
400
Arnfield (pre-lime)

Compressive Yield Stress, Py (φ )


350 Auravale Fe 18.6mg/L
Auravale Fe 24.1mg/L
300 Gellibrand
Langsett (average)
250 Wybersley
Ferric (average)
(kPa)
200

150

100

50

0
0 0.05 0.1 0.15

Solids Volume Fraction, φ (v/v)


Figure 4.1.16: Py(φ) results for stepped-pressure filtration of ferric water treatment sludges

The R(φ) results for the ferric sludges are presented in Figure 4.1.17. The
plant results were all similar, despite being formed under different conditions. The
high R(φ) and low Py(φ) results for Langsett WTP suggested that the flocculant was
over-dosed. The Arnfield WTP sample had lower R(φ) than all the other treatment
plant samples even prior to lime addition, which may have helped to account for the
short filtration times seen on-site. The pilot plant results were similar to the Arnfield
WTP result, further indicating that the Arnfield sample was likely to have not seen
much shear. The pilot plant results did not vary much, suggesting that R(φ) was not
sensitive to coagulant dose.

The D(φ) results for the ferric sludges are presented in Figure 4.1.18. The
results were very closely grouped together, despite originating from very different
sources and conditions. This outcome indicated that the dewatering behaviour of
ferric sludges is predominantly governed by the coagulant. Thus, differences in plant
performance are more likely to be due to operational changes (such as φ0 or ∆P) rather
than changes to the sludge characteristics.

122
1E+16
Arnfield (pre-lime)

Hindered Settling Function, R (φ )


Auravale Fe 18.6mg/L
Auravale Fe 24.1mg/L
Gellibrand
Langsett (average)
Wybersley
1E+15 Ferric (average)
(Pas/m2)

1E+14

1E+13
0 0.05 0.1 0.15 0.2

Solids Volume Fraction, φ (v/v)


Figure 4.1.17: R(φ) results for stepped-pressure filtration of ferric water treatment sludges

1.E-07
Solids Diffusivity, D (φ ) (m /s)
2

1.E-08

Arnfield (pre-lime)
Auravale Fe 18.6mg/L
Auravale Fe 24.1mg/L
Gellibrand
Langsett (average)
Wybersley
Ferric (average)
1.E-09
0 0.05 0.1 0.15 0.2

Solids Volume Fraction, φ (v/v)


Figure 4.1.18: D(φ) results for stepped-pressure filtration of ferric water treatment sludges

123
Small variations between the samples due to additive dosage and preparation
were seen. The lower ferric dose for the pilot plant showed lower D(φ) at low φ but
higher D(φ) at high φ compared to the higher ferric dose due to the differences in the
slope of Py(φ), such that the optimum dose was dependent on the dewatering device.

The parameters for the average functions are given in Table 4.1.7.

Table 4.1.7: Fitting parameters for average ferric water treatment sludge

Parameter Value Parameter Value Parameter Value

p1 (Pa) 7.1836 r1 (Pa s/m2) 5.10 x 1013 d1 (m2/s) 1.29 x 10-5

p2 3.1142 r2 -17.3077 d2 2.1142

φg (v/v) 0.0040 d3 19.3077

124
4.1.3 Alum Water Treatment Sludges

Hodder Water Treatment Plant

Hodder WTP treats the coloured (50 to 200 colour units), turbid (2 to 13 NTU)
waters of Stock’s reservoir in Bowland Forest, Lancashire. As with Langsett WTP,
numerous samples were taken from Hodder WTP from April 2001,
October/November 2001 and June 2002 and characterised by a number of people
using stepped-pressure filtration testing (see Table 4.1.8). The alum and flocculant
dosages were known in some but not all cases. The alum dose corresponded to sweep
coagulation. The flocculant was C492, a cationic polyacrylamide supplied by Cytec.

Table 4.1.8: Hodder WTP sludge samples

φ0 Alum dose Poly dose


Sample Operator
(v/v) (mg Al/L) (mg/L)

30/04/01 AT 0.0096 - -

24/10/01 RDF 0.0080 - -

30/10/01 ADS 0.0078 - -

31/10/01 ADS 0.0070 3.58 0.190

01/11/01 ADS 0.0059 3.42 0.148

05/11/01 ADS 0.0084 3.41 0.263

06/11/01 ADS 0.0074 - -

17/06/02 RDF/ADS 0.0102 2.50 0.184


AT – Anna Taylor, Yorkshire Water
RDF – Rene D. Frost, United Utilities
ADS – Anthony D. Stickland, The University of Melbourne

The gel point of the sample from the 30/10/01 was determined from
equilibrium batch settling. The sample was diluted to 0.0019 v/v and allowed to settle
to equilibrium from three initial heights. <φ> was calculated from a volume balance,
and plotted against h0 (see Figure 4.1.19). Extrapolating to h0 = 0 gave φg = 0.0032
v/v.

125
0.005

Final Average Volume Fraction, < φ >


0.004

0.003 φ g = 0.0032 v/v


(v/v)

0.002

0.001
30/10/01
Linear (30/10/01)
0
0 2 4 6 8 10
Initial Height, h 0 (cm)

Figure 4.1.19: Gel point determination of Hodder WTP 30/10/01 via equilibrium batch settling

The Py(φ) results are presented in Figure 4.1.20. Due to a malfunctioning


compressor, the results did not extend beyond 200 kPa. In general, the Hodder WTP
sludges showed very low equilibrium solids (< 0.09 v/v at 200 kPa). There was much
variability between the samples, reflecting the complexity of alum chemistry. The
results from October/November 2001 showed two distinct groups – 01/11/01 and
05/11/01, and the rest. From the limited dosage information, lower equilibrium
concentrations corresponded to higher and lower flocculant doses. Assuming
consistent preparation methods, this suggested the possibility of an optimum dose.

The R(φ) results are presented in Figure 4.1.21. Most of the variation in the
results corresponded to changes in volume fraction rather than magnitude of R(φ). Of
the samples with known dosages, the high alum / medium poly sample (31/10/01) was
slightly more permeable than the higher poly-dosed sample (05/06/01) and much
more permeable than the lower poly-dosed sample (01/11/01). The low alum /
medium polymer dosed sample (17/06/02) was more permeable than the higher alum
dosed samples.

126
250
30/04/01

Compressive Yield Stress, Py (φ )


24/10/01
30/10/01
200 31/10/01
01/11/01
05/11/01
06/11/01
150 17/06/02
(kPa)
Average

100

50

0
0 0.04 0.08 0.12

Solids Volume Fraction, φ (v/v)

Figure 4.1.20: Py(φ) results for stepped-pressure filtration of Hodder WTP sludges

1E+16
30/04/01
Hindered Settling Function, R (φ )

24/10/01
30/10/01
31/10/01
01/11/01
05/11/01
06/11/01
1E+15 17/06/02
Average
(Pas/m )
2

1E+14

1E+13
0 0.04 0.08 0.12

Solids Volume Fraction, φ (v/v)

Figure 4.1.21: R(φ) results for stepped-pressure filtration of Hodder WTP sludges

127
The Py(φ) and R(φ) results were combined to give D(φ) (see Figure 4.1.22).
The results for October/November 2001 showed a distinct change on 01/11/01, and
again on 06/11/01, indicating that the characteristics differed daily and the network
structure was sensitive to process conditions. The limited dosage information
suggested that optimum alum and flocculant doses existed and were dependent on
each other. Of the higher alum-dosed samples, the high and low flocculant-dosed
samples (01/11/01 and 05/11/01) showed higher D(φ) but lower solids than the
medium flocculant-dosed sample (31/10/01). The low alum-dosed sample (17/06/02)
exhibited slightly lower diffusivities than the other sludges at corresponding φ.

1.E-07
Solids Diffusivity, D (φ ) (m /s)
2

1.E-08
30/04/01
24/10/01
30/10/01
31/10/01
01/11/01
05/11/01
06/11/01
17/06/02
Average

1.E-09
0 0.04 0.08 0.12

Solids Volume Fraction, φ (v/v)


Figure 4.1.22: D(φ) results for stepped-pressure filtration of Hodder WTP sludges

The average Hodder power-law fitting parameters for are given in Table 4.1.9.

Table 4.1.9: Fitting parameters for average Hodder WTP sludge

Parameter Value Parameter Value Parameter Value

p1 (Pa) 0.1543 r1 (Pa s/m2) 7.60 x 1013 d1 (m2/s) 2.93 x 10-3

p2 4.6103 r2 -28.2954 d2 3.6103

φg (v/v) 0.0032 d3 30.2954

128
Huntington WTP and Oswestry WTP (Transient Batch Settling)

Samples from the thickener underflows at Huntington WTP from 09/07/03 and
Oswestry WTP from 25/06/03 were diluted to 0.00151 and 0.00182 v/v respectively
and allowed to settle in large measuring cylinders. The height of the interface was
recorded with time. The experimental results are presented in Figure 4.1.23.

0.4 0.08
Huntington Settling Results Oswestry Settling Results
0.35 0.07
Interfacial Height, h (t ) (m)

Analysis Curve Fit Analysis Curve Fit


0.3 0.06

0.25 0.05

0.2 0.04

0.15 0.03

0.1 0.02

0.05 0.01

0 0
0 5000 10000 15000 20000 0 5000 10000 15000
Time, t (s) Time, t (s)

Figure 4.1.23: Transient batch settling results and analysis curve fits for Huntington WTP
09/07/03 and Oswestry WTP 25/06/03

The batch settling results were analysed in conjunction with stepped-pressure


filtration results following the method outlined in Section 3.2.1. The compressibility
results are presented in Figure 4.1.24. φg for Huntington (0.00714 v/v) was higher
than for Oswestry (0.00424 v/v), but had lower equilibrium concentrations at higher
pressures. The parameters for Py(φ) are given in Table 4.1.10. The material
properties were used for predicting thickener operation (see Chapter 7)

Table 4.1.10: Py(φ) fitting parameters for Huntington WTP 09/07/03 and Oswestry WTP 25/06/03

Parameter Huntington Oswestry Parameter Huntington Oswestry

φg 0.00714 0.00424 φg2 0.00667 0.00401

a1 1.5037 1.3350 a2 1.5515 1.3658

b1 0.001 0.001 b2 0.001 0.001

k1 16.848 14.009 k2 17.917 14.541

φcp 0.400 0.400 φp 0.0178 0.0134

129
1000

Compressive Yield Stress, Py (φ )


100

10
(kPa)

0.1 Huntington Filtration Results


Huntington Settling Results
Huntington bSAMS Fit
Huntington Power-law Fit
Oswestry Filtration Results
0.01 Oswestry Settling Results
Oswestry bSAMS Fit
Oswestry Power-law Fit
0.001
φg
0 0.02 0.04 0.06 0.08 0.1

Solids Volume Fraction, φ (v/v)


Figure 4.1.24: Py(φ) results and curve fits from filtration and batch settling for Huntington WTP
09/07/03 and Oswestry WTP 25/06/03

The results show that a very weak network existed near φg. The interpolation
between φg and the first datum point from pressure filtration testing was subjective as
the strength of the network varied in a highly non-linear fashion. The power-law
functions fitted the high φ data well, and showed the right trends with decreasing φ.

The R(φ) results from batch settling analysis of Huntington WTP 09/07/03 and
Oswestry WTP 25/06/03 (see Figure 4.1.25) changed by three to four orders of
magnitude between free settling (φ = 0) and φg, and by another three to four orders of
magnitude change as the network concentration increased beyond φg. The power-law
functions, which fitted the high φ data reasonably well, were useless to describe the
low φ data. This illustrated the importance of measuring material properties across
the entire φ range for thickening and centrifuge operations. Of the two samples, the
Huntington sample was more permeable than the Oswestry sample across the entire
volume fraction range.

130
1.E+15

Hindered Settling Function, R (φ )


1.E+14

1.E+13

1.E+12
(Pas/m )
2

1.E+11

1.E+10
Huntington Filtration Results
Huntington Settling Results
1.E+09 Huntington bSAMS Fit
Huntington Power-law Fit
Oswestry Filtration Results
1.E+08 Oswestry Settling Results
Oswestry bSAMS Fit
Oswestry Power-law Fit
1.E+07
φg
0 0.02 0.04 0.06 0.08 0.1

Solids Volume Fraction, φ (v/v)


Figure 4.1.25: R(φ) results and curve fits from filtration and batch settling for Huntington WTP
09/07/03 and Oswestry WTP 25/06/03

D(φ) was calculated from dPy(φ)/dφ and R(φ) using equation 2.2.18, based on
the filtration results only, and the combined filtration and settling results (see Figure
4.1.26). The two functions showed reasonable agreement at high φ, but diverged
dramatically at low φ. The composite function exhibited a maximum between φg and
the lowest filtration measurement, corresponding to the extraordinary increase in R(φ)
and the sharp change in Py(φ). The exact magnitude depended on the interpolation
used. Therefore, measurement of intermediate φ data using permeation and
centrifugation testing is recommended for future material characterisation. The effect
of the discrepancy between the composite and power-law functions on high-pressure
filtration prediction was investigated (see Section 5.1.4).

131
1.0E-05
Huntington Filtration Results
Huntington Settling Results

Solids Diffusivity, D (φ ) (m /s)


Huntington bSAMS Fit
Huntington Power-law Fit

2
1.0E-06 Oswestry Filtration Results
Oswestry Settling Results
Oswestry bSAMS Fit
Oswestry Power-law Fit

1.0E-07

1.0E-08

1.0E-09

1.0E-10
0 φ g 0.02 0.04 0.06 0.08 0.1

Solids Volume Fraction, φ (v/v)


Figure 4.1.26: D(φ) results and curve fits from filtration and batch settling for Huntington WTP
09/07/03 and Oswestry WTP 25/06/03

Other Alum Water Treatment Sludges

A range of other alum water treatment sludge samples were characterised


using stepped-pressure filtration. The samples are listed in alphabetical order in Table
4.1.11. The five water treatment plant samples were all from the United Kingdom.
Three doses of alum were used with sludges from the Auravale pilot plant trial – 3.5,
5.1 and 8.0 mg Al/L.

The Py(φ) results for the alum sludges are presented in Figure 4.1.27. In
general, the alum sludges exhibited low equilibrium concentrations (< 0.1 v/v at 300
kPa), reflecting the formation of a strong network. The treatment plant sludges
showed variable Py(φ), with Tophill Low WTP and Eccup WTP compressing to the
highest solids, while Hodder WTP compressing the least. The results from the
Auravale Pilot Plant indicated that Py(φ) decreased with alum dose. These sludges
compressed to greater concentrations than the treatment plant sludges since there was
no flocculant added. The Eccup WTP sludge had the same results as the pilot plant
sludge with the corresponding alum dose.

132
Table 4.1.11: Other alum water treatment sludge samples

Alum Dose φ0
Source Date Operator
(mg Al/L) (v/v)

Auravale Pilot Plant 01/02/01 PJH 3.5 0.0040

Auravale Pilot Plant 19/04/00 NJA 5.1 0.0032

Auravale Pilot Plant 05/09/00 NJA 8.0 0.0021

Eccup WTP 10/08/01 MRT 5.5 0.0057

Huntington WTP 25/06/03 ADS - 0.0161

Huntington WTP 09/07/03 ADS - 0.0157

Oswestry WTP 25/06/03 ADS - 0.0145

Tophill Low WTP 15/02/01 AT - 0.0158


PJH – Peter J. Harbour, CSIRO
NJA – Nevil J. Anderson, CSIRO
MRT – Martin R. Tillotson, Yorkshire Water
ADS – Anthony D. Stickland, The University of Melbourne
AT – Anna Taylor, Yorkshire Water

400 Auravale 3.5mgAl/L Auravale 5.1mgAl/L


Auravale 8.0mgAl/L Eccup
Compressive Yield Stress, Py (φ )

Huntington 25/06/03 Huntington 09/07/03


350 Hodder (average) Oswestry
Tophill Low Alum (average)
300

250
(kPa)

200

150

100

50

0
0 0.05 0.1 0.15

Solids Volume Fraction, φ (v/v)


Figure 4.1.27: Py(φ) results for stepped-pressure filtration of alum water treatment sludges

133
Figure 4.1.28 shows the R(φ) results for the alum treatment sludges. There
was some variation between plants, with Hodder WTP the least permeable and
Huntington WTP (09/07/03) and Eccup WTP the most permeable at a given φ. The
two Huntington WTP samples had different R(φ) but the same Py(φ), with the earlier
sample less permeable at high φ. The pilot plant samples had lower R(φ) than the
treatment plant sludges. There was a slight decrease in R(φ) with increasing alum
dose.

1E+16
Auravale 3.5mgAl/L Auravale 5.1mgAl/L
Hindered Settling Function, R (φ )

Auravale 8.0mgAl/L Eccup


Huntington 25/06/03 Huntington 09/07/03
Hodder (average) Oswestry
Tophill Low Alum (average)
1E+15
(Pas/m )
2

1E+14

1E+13
0 0.05 0.1 0.15

Solids Volume Fraction, φ (v/v)


Figure 4.1.28: R(φ) results for stepped-pressure filtration of alum water treatment sludges

The D(φ) results for the alum sludges are presented in Figure 4.1.29. There
was a greater range of results than seen for the ferric sludges (see Figure 4.1.18). The
order from highest to lowest D(φ) for the treatment plant sludges was Huntington >
Eccup ≥ Hodder > Tophill Low > Oswestry. As an overall dewatering parameter,
D(φ) reflected both changes to permeability and compressibility. For example, of the
two Huntington samples, 09/07/03 had higher D(φ) because of lower R(φ), whereas of
the three Auravale sludges, the lowest alum dose had the highest D(φ) due to larger
dPy(φ)/dφ. The medium dose had a slightly higher D(φ) at the lowest φ, indicating

134
that the optimum dose was dependent on the applied pressure, and therefore the
dewatering device used.

1.E-07
Solids Diffusivity, D (φ ) (m /s)
2

1.E-08

Auravale 3.5mgAl/L Auravale 5.1mgAl/L


Auravale 8.0mgAl/L Eccup
Huntington 25/06/03 Huntington 09/07/03
Hodder (average) Oswestry
Tophill Low Alum (average)
1.E-09
0 0.05 0.1 0.15

Solids Volume Fraction, φ (v/v)


Figure 4.1.29: D(φ) results for stepped-pressure filtration of alum water treatment sludges

The power-law fitting parameters for the average alum sludge are given in
Table 4.1.12.

Table 4.1.12: Fitting parameters for average alum WTP sludge

Parameter Value Parameter Value Parameter Value

p1 (Pa) 9.5445 r1 (Pa s/m2) 2.86 x 1013 d1 (m2/s) 2.86 x 10-3

p2 3.4697 r2 -23.8732 d2 2.4697

φg (v/v) 0.005 d3 25.8732

135
4.1.4 Water Treatment Coagulant Comparison

The average D(φ) results for the alum and ferric water treatment sludges are
shown in Figure 4.1.30. The alum sludges exhibited slightly higher diffusivity and
lower equilibrium concentrations than ferric sludges, indicating that they generally
dewatered faster, but to a lesser extent at a fixed pressure.

1.E-07
Alum + Poly
Solids Diffusivity, D (φ ) (m /s)
2

Alum

1.E-08

Ferric

Poly-only

1.E-09
0 0.1 0.2 0.3 0.4

Solids Volume Fraction, φ (v/v)


Figure 4.1.30: Overview of D(φ) for a range of coagulant types in potable water treatment

Also shown in Figure 4.1.30 are stepped-pressure results for water treatment
sludges generated from other coagulants (Harbour, et al., 2004). Swan Reach WTP
and Tailem Bend WTP in South Australia treat raw water from the Murray River to
supply Adelaide with potable water. The water is generally highly turbid but with
little colour, and, at different times of the year, LT35 (a cationic polyelectrolyte from
Ciba Specialty Chemicals) is used as either the sole coagulant or in combination with
alum. The results were for a range of doses, with or without further conditioning with
LT22. The exact dosage was not an issue here, rather the large effect that

136
conditioning had on polyelectrolyte sludges compared to sludges where alum or ferric
was the sole coagulant.

The results where the primary coagulant was a combination of alum and
polyelectrolyte showed higher equilibrium solids concentrations but lower diffusivity
than conventional alum sludges. Adding extra flocculant increased the diffusivity by
a large amount, thereby improving the rate of filtration. The alum-only sludges did
not exhibit this large change with post-coagulation polymer addition.

The results where polymer was used as the sole coagulant showed even higher
concentrations and lower diffusivity than the alum/poly combination. This reflected
the high molecular weight, cross-linked polymeric nature of the sludge, which
appeared to resemble the ECP component of sewage sludges. As with the alum/poly
combination, adding further flocculant had a huge effect on the permeability.

Overall, the comparison of coagulant types reinforced the hypothesis that the
choice of coagulant predominantly determined the sludge dewatering properties.

137
4.2 Wastewater Treatment Sludges

There were two aims for wastewater sludge characterisation:

- Comparison of the effect of additives and pre-treatments; and

- Extraction of material characteristics for the prediction of the behaviour in


dewatering operations.

The first aim was successfully investigated using a scaling procedure to allow
qualitative comparisons of experimental data, while the second aim was achieved
using compressional curve fitting techniques.

4.2.1 Filtration Scaling Behaviour

Analysis of the pressure filtration theory shows that the filtration time is
dependent on h02 and is a function of φ0 (see Equation 2.4.36). For materials such as
wastewater treatment sludges, where fundamental characterisation information is
difficult to extract, these dependencies can be removed using a semi-empirical scaling
procedure (see Section 2.4.2). This allows direct comparisons of constant pressure
filtration profiles with varying φ0 and h0, which is important when investigating the
effects of additives and pre-treatments.

Experimental Validation

Experiments were performed on flocculated and un-flocculated activated


sewage sludge samples from Carrum WWTP, Victoria, to compare the filtration
profiles of sewage sludges under different φ0 and h0 conditions in order to investigate
the relationships between the time of filtration and these variables.

Varying Initial Height, h0

The manual filtration rig (see Section 3.1.2) was used to perform single-
pressure filtration of un-flocculated sludge at 200 kPa at three different initial heights,

138
and of flocculated sludge at 50, 100 and 200 kPa with two initial heights at each
pressure. The flocculant was Zetag 87. The solids density was determined as 1430 ±
100 kg/m3.

The t versus V2 results for the un-flocculated samples are shown in Figure
4.2.1. As indicated by the time scale, the material had a very low permeability. The
initial slope of t versus V2 was similar for each run due to constant initial
concentrations (φ0 = 0.0149 v/v). As was expected, both the required filtration time
and the equilibrium filtrate volume increased with increasing h0.

300000
h0 = 3 cm
250000 h0 = 2 cm
h0 = 1 cm
200000
Time, t (s)

150000

100000

50000

0
0 0.0002 0.0004 0.0006 0.0008 0.001
2 2 2
(Specific Filtrate Volume) , V (m )
Figure 4.2.1: t versus V2 results for single-pressure filtration of un-flocculated Carrum WWTP
activated sludge at 200 kPa with varying h0

Scaling the time and volume data with h02, the same data is plotted as t/h02
versus (1-φ0/φ)2 in Figure 4.2.2, where φ was calculated from a volume balance.
Since ∆P was constant, the final scaled volume should have been equal, such that the
data effectively collapsed to a single curve. The results showed good consistency,
indicating that the filtration time was dependent on h02 throughout cake formation and
cake compression.

139
3.0E+08
h0 = 3 cm

Adjusted Time, t/h0 (s/m )


2.5E+08 h0 = 2 cm

2 h0 = 1 cm

2.0E+08
2

1.5E+08

1.0E+08

5.0E+07

0.0E+00
0 0.2 0.4 0.6 0.8 1

(1-φ 0 /φ )2
Figure 4.2.2: Adjusted time versus scaled volume results for single-pressure filtration of un-
flocculated Carrum WWTP activated sludge at 200 kPa with varying h0

Nevil Anderson of CSIRO repeated these experiments at 50, 100 and 200 kPa
with a flocculated sample in order to confirm the hypothesis that the filtration time
varies with the square of the initial height. h0 was either 2.5 cm or 5 cm and φ0 was
0.025 v/v. The t versus V2 results are presented in Figure 4.2.3. The raw results
illustrated the large increase in filtration time and filtrate volume due to the increased
height.

The t and V2 data were scaled with h02 (see Figure 4.2.3). Biological changes
in the sludge would have introduced some error to the samples with longer filtration
times, but the results generally showed very close agreement, within the variation
expected. These results confirmed the theoretical relationship between the time of
filtration and the initial height.

140
1.4E+06 6E+08

1.2E+06
50 kPa 50 kPa

Adjusted time, t/h0 (s/m )


5E+08

2
1.0E+06
4E+08

2
Time, t (s)

8.0E+05
3E+08
6.0E+05
2E+08
4.0E+05

2.0E+05 1E+08

0.0E+00 0E+00

1.4E+06 6E+08

1.2E+06
100 kPa 100 kPa
Adjusted time, t/h0 (s/m )
2 5E+08

1.0E+06
4E+08
2
Time, t (s)

8.0E+05
3E+08
6.0E+05
2E+08
4.0E+05

1E+08
2.0E+05

0.0E+00 0E+00

1.4E+06 6E+08

1.2E+06
200 kPa 200 kPa
Adjusted time, t/h0 (s/m )

5E+08
2

1.0E+06
4E+08
2
Time, t (s)

8.0E+05
3E+08
6.0E+05
2E+08
4.0E+05

2.0E+05 1E+08

0.0E+00 0E+00
0 0.0005 0.001 0.0015 0.002 0.0025 0 0.2 0.4 0.6 0.8 1

(Specific Filtrate Volume) , V 2 2 2


(m ) (1-φ 0 /φ ) 2

Figure 4.2.3: Raw and scaled results for single-pressure filtration of flocculated Carrum WWTP
activated sludge at 50, 100 and 200 kPa with varying h0

Varying Initial Volume Fraction, φ0

While the variation due to h0 is described well by a squared relationship, the φ0


variation is more complicated. All of the φ0 dependencies are grouped in a scaling
parameter, c(φ0)/c(φ0,ref). An empirical functional form for c(φ0)/c(φ0,ref) was
determined from the linear approximation (see Section 2.4.2) using the characteristics

141
of a highly compressible and impermeable material, at an applied pressure of 200 kPa
and a reference volume fraction of 0.02 v/v:

c(φ0 ) 1
= …(4.2.1)
c(φ0 = 0.02 ) − 73.087φ − 150.77φ02 + 53.033φ
3
0 0

Modelling work presented later in this section showed that equation 4.2.1 was
fairly robust (±10%) as long as φ0 << φ∞, which is the case for sewage sludges. The
greatest inaccuracies arose for materials with large changes in Py(φ) for small changes
in φ values.

Single-pressure filtration tests were performed by Nevil Anderson, CSIRO, on


flocculated Carrum WWTP activated sludge using the manual filtration rigs. The
samples were diluted by differing amounts in order to explore the variation caused by
the initial volume fraction, φ0, on the filtration time, t. The raw experimental data is
shown in Figure 4.2.4.

5E+06
phi0 = 0.0144 v/v
phi0 = 0.0163 v/v
4E+06 phi0 = 0.0180 v/v
phi0 = 0.0203 v/v
phi0 = 0.0299 v/v
Time, t (s)

3E+06

2E+06

1E+06

0E+00
0 0.0005 0.001 0.0015 0.002 0.0025
2 2 2
(Specific Filtrate Volume) , V (m )
Figure 4.2.4: t versus V2 results for single-pressure filtration of flocculated Carrum WWTP
activated sludge at 200 kPa with varying φ0

142
As all samples had the same initial height (5 cm) and applied pressure (200
kPa), the results showed the expected trends – the samples with the higher φ0 took
longer to filter since the permeability decreased. The tests proceeded to the same final
concentration (measured by oven drying as φf = 0.456 ± 0.005 v/v), and therefore to
different volumes of filtrate since there were different total amounts of solids in the
filtration cell (see equation 2.3.17).

t was scaled by dividing by φ02 and multiplied by c(φ0)/c(φ0,ref), calculated


from equation 4.2.1. The results were plotted against (1-φ0/φ)2 (see Figure 4.2.5).

8E+09
phi0 = 0.0144 v/v
7E+09 phi0 = 0.0163 v/v
phi0 = 0.0180 v/v
c (φ0 )/c(φ0,ref )(t /φ0 ) (s)

6E+09
phi0 = 0.0203 v/v
2

5E+09 phi0 = 0.0299 v/v

4E+09

3E+09

2E+09

1E+09

0E+00
0 0.2 0.4 0.6 0.8 1

(1-φ 0 /φ )2
Figure 4.2.5: Adjusted time versus scaled volume results for single-pressure filtration of
flocculated Carrum WWTP activated sludge at 200 kPa with varying φ0

The results showed very close agreement (within experimental error) between
the three samples with average φ0. The run for the sample with the highest initial
concentration (φ0 = 0.0297 v/v) exhibited a higher adjusted time, which may have
been an artefact of experimental error, or due to the material behaviour. The sample
with the lowest φ0 (φ0 = 0.0143 v/v) showed smaller adjusted times at higher volumes
than the other results. Thus, the scaled behaviour still had a slight dependence on φ0,
suggesting that either the relationship between tf and φ0 was greater than squared, or

143
the predicted scaling values were too high. The robustness of the scaling parameter is
discussed next. Overall, these results showed that the time dependence of φ0 is close
to the theoretical relationship, and that, if c(φ0)/c(φ0,ref) is known, useful comparisons
can be made between samples with varying initial concentrations.

Variation of c(φ0)/c(φ0,ref)

The variation of the scaling function with material characteristics and applied
pressure was investigated. In order to calculate values for c(φ0)/c(φ0,ref) for sewage
sludges, the following functional forms were used:

D (φ ) = 3.582 × 10 −8 φ 2.468 …(4.2.2)

Py (φ ) = 1.465 × 10 6 (φ − 0.02 )
1.924
…(4.2.3)

These fits arose from initial estimates, and represented a highly compressible,
highly impermeable material. D(φ) and Py(φ) were compared with eight diffusivity
functions (designated as D1 through to D8) and five compressibility functions (P1
through to P5), as shown in Figure 4.2.6. The two sets combined gave 40 different
possible combinations of material properties. These functions were originally
developed to cover the complete range of behaviour of water treatment sludges, and
were used here to investigate the variability of c(φ0)/c(φ0,ref).

P1 P2 P3 P4 P5 D1 D2 D3
400 1E-07
Compressive Yield Stress, Py(φ)

Solids Diffusivity, D (φ ) (m /s)

350 D6
D4
2

300
D7
D5
1E-08
250
(kPa)

200
D8
150
1E-09 sewage
sewage
100 example example

50

0 1E-10
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5

Solids Volume Fraction, φ (v/v) Solids Volume Fraction, φ (v/v)

Figure 4.2.6: Generalised material property functions used to investigate the variation of
filtration behaviour due to changes in φ0

144
The variation of c(φ0)/c(φ0,ref = 0.01 v/v) with D(φ) at 100 kPa is presented in
Figure 4.2.7. Despite magnitudes between D1, D8 and the sewage example, there was
little difference between c(φ0)/c(φ0,ref) – at 0.025 v/v, the difference between D1P5
and D8P5 was only 6.3% while at 0.05 v/v, the difference was 19.4%. The difference
between D1P5 and the sewage example was even smaller. In overall terms, the
magnitude of D(φ) appeared to have only minor effects on c(φ0)/c(φ0,ref).

1
Scaling Function, c (φ0 )/c (φ0,ref )

0.9

0.8

0.7

0.6

0.5

0.4

0.3 D8P5

0.2 D1P5
sewage
0.1 example
0
0 0.02 0.04 0.06 0.08 0.1

Initial Volume Fraction, φ 0 (v/v)


Figure 4.2.7: Variation of φ0 scaling function with solids diffusivity

The variation of the scaling function with compressibility at 100 kPa is


presented in Figure 4.2.8. The scaling passed through a minimum and began to
increase as φ0 approached φ∞. For example, for a relatively incompressible material
such as that represented by curve P1, c(φ0)/c(φ0,ref) began to increase as φ0 neared the
compressional limit of 0.0493 v/v. However, φ0 is usually much smaller than φ∞ for
most wastewater sludges. The results for the most compressible materials (P4, P5 and
the sewage example) were similar up to relatively high φ0 (0.06 to 0.08 v/v). Overall,
there was some dependence upon the compressibility, more so than for the diffusivity,
but if the compressibility of the material was roughly known or φ0 was much smaller
than φ∞, then c(φ0)/c(φ0,ref) could be found without too much error.

145
1
D1P1 D1P2 D1P3

Scaling Function, c (φ0 )/c (φ0,ref )


0.9

0.8

0.7

0.6

0.5

0.4
D1P4
0.3
D1P5
0.2

0.1 sewage
example
0
0 0.02 0.04 0.06 0.08 0.1

Initial Volume Fraction, φ 0 (v/v)

Figure 4.2.8: Variation of φ0 scaling function with compressive yield stress

1
Scaling Function, c (φ0 )/c (φ0,ref )

0.9
20 kPa
0.8

0.7

0.6

0.5
50 kPa
0.4

0.3 100 kPa


200 kPa
0.2 300 kPa
0.1

0
0 0.02 0.04 0.06 0.08 0.1

Initial Volume Fraction, φ 0 (v/v)

Figure 4.2.9: Variation of φ0 scaling function with applied pressure, ∆P for D1P5

146
Figure 4.2.9 shows the dependence of the scaling function with applied
pressure, and thus further demonstrates the effect of the compressibility curve since
∆P was equated to Py(φ) to give φ∞. c(φ0)/c(φ0,ref) for 20 and 50 kPa began to increase
as φ0 became significant compared to φ∞.

Overall, these plots show that, if two materials have similar compressibility’s,
the applied pressures are close in terms of compressibility, or the initial volume
fractions to be reconciled are well below the minimums seen in the scaling functions,
then a value of c(φ0)/c(φ0,ref) can be assigned and a comparison between two filtration
runs with differing initial volume fractions can be made.

Conclusions

The results showed that the variation of time to filtration with initial
conditions was accounted for using theoretical considerations. By adjusting the time
by φ02, h02 and a scaling parameter c(φ0)/c(φ0,ref), and scaling V2 to (1-φ0/φ)2, direct
comparisons could be made between sludges since a sample with a higher adjusted
time had worse dewatering properties than a material with a lower adjusted time. For
example, the effects of conditioning and treatment of sludges could be qualitatively
determined. The variation of the scaling parameter with material properties and
applied pressure was investigated and shown to be relatively constant under
appropriate conditions.

By demonstrating that the theoretical relationships apply to sewage sludge


filtration, operational trends with feed concentration and cavity width arising from the
modelling of plate-and-frame presses can be applied, such that:

- Higher throughputs are achieved with higher feed solids;

- By finishing a fill stage earlier, the starting concentration for the squeeze
phase will be lower, and the filtration time will be reduced; and

- An optimum cavity width exists for a given material that allows a given cake
concentration to be reached more quickly.

147
Effect of Additives and Pre-Treatments

Nevil Anderson and Peter Harbour of CSIRO performed a range of single-


pressure filtration experiments on Colac WWTP digested sludge, Mornington WWTP
digested sludge, Lilydale WWTP activated sludge, and two different Carrum WWTP
activated sludges using the manual filtration rig. They also investigated a range of
additives (include ferric salts, cationic polymers, and combinations thereof) and
treatments such as hydrothermal treatment and oxidation using Fenton’s reagent. h0
was 5 cm and ∆P was 200 kPa in all cases.

The raw experimental results were originally treated with some caution due to
problems associated with achieving constant solids loading. Using the procedure
described in Section 2.4.2, the filtration times and filtrate volumes were scaled to
eliminate the effects of differing initial conditions such that any significant changes in
the filtration behaviour were unambiguously attributed to changes in the sludge
characteristics. Thus, comparisons between sludges, additives and treatments were
made.

Wastewater Treatment Plant Sludge Comparisons

One of the aims of characterisation of sludges was to be able to compare


sludges from different plants. Most commonly, CST or SRF are used to measure the
dewaterability of sludges. As noted earlier, these methods give an indication of the
initial rate of filtration or permeability of the sludge, but erroneous comparisons are
made without due consideration of the effect of h0 and φ0. In many cases, users of
SRF or CST techniques overcome this problem by simply dividing the filtration time
by the initial solids concentration. This practice is incorrect as the filtration time is
not a linear function of the initial solids fraction, as already demonstrated.

Figure 4.2.10 shows the t versus V2 results for a series of constant pressure
filtration tests performed on sewage sludges from a variety of wastewater treatment
plants in Australia. The initial rate of filtration is indicated by the inverse of the slope
and shows the same trends that would be obtained from SRF or CST. This data
indicates that the order of filtration behaviour, from fastest to slowest, was Lilydale >
Colac > Carrum 1 > Mornington > Carrum 2.

148
200000
Carrum 1 Carrum 2
180000 Colac Lilydale
160000 Mornington

Time, t (s) 140000

120000

100000

80000

60000

40000

20000

0
0 0.0005 0.001 0.0015 0.002
2 2 2
(Specific Filtrate Volume) , V (m )
Figure 4.2.10: t versus V2 for constant pressure filtration of sludges from a variety of WWTP’s

200000
Carrum 1
180000 Carrum 2
Colac
160000 Lilydale
Mornington
140000
Time, t (s)

120000

100000

80000

60000

40000

20000

0
0 0.02 0.04 0.06 0.08 0.1

Average Solids Volume Fraction, <φ > (v/v)


Figure 4.2.11: t versus φ for single-pressure filtration of sludges from a variety of WWTP’s

149
Another common way to analyse the data is to look at the rate of consolidation
by plotting the average solids concentration as a function of time (see Figure 4.2.11).
In this figure the rate of consolidation is indicated by dφ/dt, and the initial rate shows
the same order of filtration behaviour as Figure 4.2.10. Neither of these two methods
takes φ0 into account. The Carrum samples were exactly the same sludge but with
different φ0 (0.0143 and 0.0296 v/v respectively). Using either Figure 4.2.10 or
Figure 4.2.11 to compare the results would have lead to the wrong conclusion that
Carrum 1 was more permeable.

Using the scaling procedure to eliminate the φ0 dependencies enables more


accurate comparisons of the sludge properties. The scaled results are presented in
Figure 4.2.12 and plotted on a log-log scale. Material with lower scaled times and
higher scaled volumes filtered faster.

1E+16
Carrum 1
Carrum 2
Colac
1E+15 Lilydale
c (φ0 )/c (φ0,ref ) x t/φ0 , (s)

Mornington
2

1E+14

1E+13

1E+12

1E+11
0.0001 0.001 0.01 0.1 1

(1-φ 0 /φ )2
Figure 4.2.12: Adjusted time versus scaled volume results for single-pressure filtration of sludges
from a variety of WWTP’s

The true order of sludge filtration behaviour, fastest to slowest, is now shown
to be Lilydale > Colac > Carrum 1 > Carrum 2 > Mornington. There was still some
difference between the Carrum samples either due to experimental error or a relatively

150
small volume fraction dependence that was not totally resolved by the mathematical
treatment. However, the difference was relatively small compared to the differences
between the other sludges. At the Lilydale plant, ferrous salts were added to remove
phosphorus and the final sludge contained approximately 4% Fe by weight. It was
largely for this reason that the Lilydale sludge was by far the quickest to dewater. The
Mornington sludge was more than an order of magnitude slower to dewater than
Lilydale sludge.

Treatment and Conditioning Comparisons

The use of the scaling procedure also allowed comparisons of different


treatment conditions. When polyelectrolyte was used as a conditioning agent, the
solids consolidated and produced large quantities of free water during the
conditioning step, such that transfer of the sludge to filtration equipment could result
in changes to φ0. Addition of solids, such as ash or lime as conditioning agents, or
coagulants that form solids in situ, such as ferric salts, also changed φ0. Another
treatment that resulted in a reduction in φ0 was thermal hydrolysis. Thus, any
differences observed in filtration experiments were primarily due to changes in φ0
rather than the effect of treatments or additives. By applying the scaling procedure,
the dependence of t on φ0 was removed and direct comparisons of treatment options
were made.

Comparisons of Polyelectrolytes used for Conditioning

Figure 4.2.13 shows the effect of various cationic polyelectrolytes (see Table
4.2.1) on the filtration behaviour of Colac digested sludge. The polyelectrolytes were
compared at an equal dose of 9.4 mg/g, rather than optimised for each individual
polymer. The three best sludge conditioners at this dose were Zetag 87, Betz 1158
and MK4000. Table 4.2.1 shows that these conditioners were the two high charge,
high molecular weight polyacrylamides and the high charge, medium to high
molecular weight polymethacrylate. The cationic polystyrene showed only a small
improvement over the raw sludge, and the polyDADMAC and polyacrylate performed
moderately well.

151
1.E+09
No additive
Dow QX
Magnafloc 368

c (φ0 )/c (φ0,ref ) x t/φ0 , (s)


Percol 292
1.E+08 Diaclear MK4000
Betz 1158
2

Zetag 87

1.E+07

1.E+06

1.E+05
0.001 0.01 0.1 1

(1-φ 0 /φ )2
Figure 4.2.13: Adjusted time versus scaled volume results for single-pressure filtration of Colac
digested sludge treated with a variety of cationic polyelectrolytes.

Table 4.2.1: Properties of the polyelectrolytes used for conditioning sludges.

Flocculant Charge Density MW Description

Zetag 87 high high cationic polyacrylamide

Betz 1158 high high cationic polyacrylamide

Diaclear MK4000 high med-high cationic polymethacrylate

Magnafloc 368 high high polyDADMAC

Percol 292 medium high cationic polyacrylate

Dow QX med-high low cationic polystyrene

Optimum Polyelectrolyte Dose

The Lilydale, Colac and Mornington sludges were conditioned with various
doses of Zetag 87 in order to assess the optimum conditioner dose. Figure 4.2.14
shows the scaled filtration results for Lilydale activated sludge. The best dose

152
produced the shortest scaled filtration time and the greatest dimensionless volume.
Hence the optimum dose was 9.44 mg/g.

1.E+09
No additive
2.36 mg/g Z87
4.72 mg/g Z87
c (φ0 )/c (φ0,ref ) x t/φ0 , (s)

7.08 mg/g Z87


9.44 mg/g Z87
1.E+08 11.8 mg/g Z87
2

16.5 mg/g Z87

1.E+07

1.E+06
0.1 1

(1-φ 0 /φ )2
Figure 4.2.14: Adjusted time versus scaled volume results for single-pressure filtration of
Lilydale activated sludge conditioned with Zetag 87 at varying doses

Similarly, Figure 4.2.15 shows the data for Colac digested sludge conditioned
with Zetag 87. In this case the optimum was 9.45 mg/g, almost identical to the
Lilydale result. This result was odd since Lilydale sludge had better filtration
properties (see Figure 4.2.12). However, the filtration of the highest dose for Colac
(11.8 mg/g) was slower than the lowest dose for Colac (4.72 mg/g) and therefore may
have been an anomaly and the required dose may have been higher than 9.45 mg/g.
Whatever the case, these figures were close to site values.

The scaled data for Mornington digested sludge conditioned with Zetag 87 is
presented in Figure 4.2.16, showing that the optimum dose for Mornington was 14.3
mg/g or greater. As shown in Figure 4.2.12, the Mornington digested sludge had the
slowest dewatering behaviour of the sludges tested, thus it was reasonable that the
optimum dose was higher than for Lilydale activated sludge.

153
1.E+09
No additive
4.72 mg/g Z87
5.90 mg/g Z87

c (φ0 )/c (φ0,ref ) x t/φ0 , (s)


1.E+08 7.09 mg/g Z87
9.45 mg/g Z87
2

11.8 mg/g Z87

1.E+07

1.E+06

1.E+05
0.001 0.01 0.1 1

(1-φ 0 /φ )2
Figure 4.2.15: Adjusted time versus scaled volume results for single-pressure filtration of Colac
digested sludge conditioned with Zetag 87 at varying doses

1.E+10
No additive
7.14 mg/g Z87
8.57 mg/g Z87
1.E+09 10.0 mg/g Z87
c (φ0 )/c (φ0,ref ) x t/φ0 , (s)

11.4 mg/g Z87


14.3 mg/g Z87
2

1.E+08

1.E+07

1.E+06

1.E+05
0.001 0.01 0.1 1

(1-φ 0 /φ )2
Figure 4.2.16: Adjusted time versus scaled volume results for single-pressure filtration of
Mornington digested sludge conditioned with Zetag 87 at varying doses

154
The Colac, Lilydale and Mornington sludges with Zetag 87 doses of 9.45, 9.44
and 14.3 mg/g respectively are compared in Figure 4.2.17. The results show that the
maximum conditioned Mornington sludge was slower than the optimum conditioned
Colac sludge, which was slower than optimum conditioned Lilydale sludge, which is
the same trend as the unconditioned sludges. The trend post-conditioning was far
closer, but, even with a higher level of polymer conditioning, the slower dewatering
sludges still had worse performance compared to the faster dewatering ones, such that
the polymer conditioning did not equalise sludge dewatering performance.

1.E+09
Colac 9.45 mg/g Z87
Lilydale 9.44 mg/g Z87
Mornington 14.3 mg/g Z87
c (φ0 )/c (φ0,ref ) x t/φ0 , (s)

1.E+08
2

1.E+07

1.E+06

1.E+05
0.1 1

(1-φ 0 /φ )2
Figure 4.2.17: Adjusted time versus scaled volume results for single-pressure filtration of
conditioned sludges at the optimum or maximum dose of Zetag 87

Ferric Chloride for Sludge Conditioning.

Ferric chloride was used to condition Colac digested sludge. The scaled
results for a range of doses are presented in Figure 4.2.18, showing that, even at the
lowest dose of 7.9 wt%, there were significant improvements in dewatering behaviour
due to the additive. There was little improvement in the filtration behaviour above 14
wt%, but this level of addition was very high and impractical at full scale.

155
1.E+09
No additive
7.87% FeCl3
11.0% FeCl3

c (φ0 )/c (φ0,ref ) x t/φ0 , (s)


1.E+08 14.1% FeCl3
18.9% FeCl3
2

1.E+07

1.E+06

1.E+05
0.001 0.01 0.1 1

(1-φ 0 /φ )2
Figure 4.2.18: Adjusted time versus scaled volume results for single-pressure filtration of Colac
digested sludge conditioned with ferric chloride at varying doses

Ferric Chloride and Zetag 87 in Combination for Sludge Conditioning

The Lilydale activated sludge was conditioned with 4.0% ferric chloride in
combination with various doses of Zetag 87. The scaled results are presented in
Figure 4.2.19. With ferric, the optimum dose of Zetag 87 was 2.36 mg/g, which was
much less than the 9.44 mg/g found for the Lilydale activated sludge conditioned with
Zetag 87 only. It was likely that the ferric chloride helped to titrate the ECP and
therefore reduced the polyelectrolyte requirements substantially.

Figure 4.2.20 shows the scaled filtration results for various combinations of
ferric chloride and polyelectrolyte treatments of Lilydale activated sludge. The
polyelectrolyte alone gave the best conditioning, followed closely by the treatment
using 4% ferric chloride and 2.4 mg/g Zetag 87. Ferric chloride alone at moderate
levels did not give the same magnitude of improvement as polyelectrolyte.

156
1.E+09
No Z87
0.59 mg/g Z87
1.18 mg/g Z87

c (φ0 )/c (φ0,ref ) x t/φ0 , (s)


1.77 mg/g Z87
2.36 mg/g Z87
1.E+08 3.54 mg/g Z87
2

7.04 mg/g Z87

1.E+07

1.E+06
0.1 1

(1-φ 0 /φ )2
Figure 4.2.19: Adjusted time versus scaled volume results for single-pressure filtration of
Lilydale activated sludge conditioned with 4% ferric chloride and varying doses of Zetag 87

1.E+09
No additive
4% FeCl3
6% FeCl3
2.36 mg/g Z87
c (φ0 )/c (φ0,ref ) x t/φ0 , (s)

4% FeCl3, 2.36 mg/g Z87


7.08 mg/g Z87
1.E+08
2

4% FeCl3, 7.04 mg/g Z87


9.44 mg/g Z87

1.E+07

1.E+06
0.1 1

(1-φ 0 /φ )2
Figure 4.2.20: Adjusted time versus scaled volume results for single-pressure filtration of
Lilydale activated sludge conditioned with various combinations of ferric chloride and Zetag 87

157
Figure 4.2.20 also shows the scaled filtration data for Lilydale activated sludge
with and without 4% ferric chloride addition at three different Zetag 87 doses. When
no Zetag 87 was used, the benefits of adding 4% ferric chloride were relatively small.
At 2.36 mg/g Zetag 87 (approximately one quarter of the optimum dose for this
sludge) the polymer alone substantially improved the dewaterability. The addition of
4% ferric chloride prior to the polymer addition also improved the dewaterability. At
the higher dose of 7.04 mg/g Zetag 87 (just below the optimum dose of 9.4 mg/g) the
filtration characteristics were similar to the 2.36 mg/g Zetag 87, and 4% ferric
chloride, whereas the use of ferric chloride prior to polymer addition reduced the
effect of the polyelectrolyte. The best combination ultimately depended on the cost of
the additives.

Advanced Pre-treatments (Fenton’s reagent and Hydrothermal


Treatments)

Hydrothermal processes such as the CAMBI process improve the efficiency of


digesters and reduce the volume of sludge produced. Such processes also alter the
material characteristics. It is claimed that the CAMBI process improves the
dewaterability (Kepp, et al., 2000, Neyens and Baeyens, 2003), but improved cake
concentrations may be due to increases in φ0.

The results from a series of experiments investigating hydrothermal and


chemical oxidation treatments are presented in Figure 4.2.21. The results show that
Fenton’s reagent alone had little effect on the filtration behaviour, but hydrothermal
treatment and the combination of Fenton’s reagent and hydrothermal treatment had
measurable benefits, such that hydrothermal treatment improved the dewaterability of
sewage sludge. This laboratory process was different from the CAMBI process in
that the hydrothermally treated sludge had not been subject to further digestion.
Therefore, to fully substantiate the CAMBI claims, it would be necessary to obtain
samples from operational plants.

158
1.E+10
Untreated
Fenton's reagent
1.E+09 Hydrothermal treatment

c (φ0 )/c (φ0,ref ) x t/φ0 , (s)


2 Fenton's + Hydrothermal

1.E+08

1.E+07

1.E+06

1.E+05
0.001 0.01 0.1 1

(1-φ 0 /φ )2
Figure 4.2.21: Adjusted time versus scaled volume results for single-pressure filtration of
Carrum activated sludge after hydrothermal treatment, treatment with Fenton’s reagent, and
both in combination

Conclusions

The scaling procedure was used to scale constant-pressure filtration data to


eliminate the effects of different starting conditions. Thus, it was possible to
unambiguously compare the dewatering behaviour of different sludges, select
conditioning agents, optimise conditioning doses for sludge dewatering, compare
different conditioned sludges, and analyse the effects of other advanced pre-
treatments, such as hydrothermal treatment, on sludge dewaterability.

159
4.2.2 Logarithmic Fitting of Cake Compression

As indicated by the linear approximation solution of Landman and White,


1997, the dewatering characteristics of sludges may be extracted from a series of
single-pressure filtration runs by fitting a logarithmic function to the compression
stage of experimental data (see Section 2.4.2). This section outlines the theoretical
validation of this approach by using model generated filtration predictions, and then
details the application and experimental validation to industrial wastewater sludges.

Theoretical Validation

The FDM (see Section 2.4.3) was used to generate predictions for a series of
single-pressure runs at a range of pressures. φ∞ and D(φ∞) were extracted from these
predictions using the logarithmic curve fitting method and compared to the original
material properties. Thus, experimental errors were eliminated from the data and the
logarithmic approach was theoretically tested.

An important issue in laboratory characterisation is the required filtration time.


For non-traditional filtration, the time required to reach equilibrium may be very long.
This work gave an indication of the required filtration time by analysing the effect of
extracting parameters from incomplete data.

The material parameters for Py(φ) and D(φ) power-law functions (equations
2.1.6 and 2.4.50) used to generate filtration predictions are given in Table 4.2.2:

Table 4.2.2: Material parameters used for theoretical validation

Parameter Value Parameter Value

φg (v/v) 0.01 d1 (m2/s) 3.61 x 10-7

p1 (Pa) 10 d2 4/3

p2 7/3 d3 12

160
These functions were not representative of any particular material but
demonstrated extreme compressibility and decreasing behaviour in D(φ) (required for
non-traditional filtration, see Section 2.4.4). The filtration predictions for a range of
applied pressures (20, 50, 80, 100, 150, 200 and 300 kPa) generated using the FDM
are shown in Figure 4.2.22. As the pressure increased, both the rate and extent of
filtration increased. There was very little increase in the initial rate for pressures
above 50 kPa, but there were significant changes to the cake compression data.

500000
10 kPa 20 kPa
450000 50 kPa 80 kPa
400000 100 kPa 150 kPa
200 kPa 300 kPa
350000
Time, t (s)

300000

250000

200000

150000

100000

50000

0
0 0.0001 0.0002 0.0003 0.0004 0.0005 0.0006
2 2 2
(Specific Filtrate Volume) , V (m )
Figure 4.2.22: Filtration predictions for validation of logarithmic method

The V data was transformed to <φ> using equations 2.3.1 and 2.3.13. The t
versus <φ> results were analysed using the logarithmic method (see Section 3.1.5).
Since the fit was over an unknown number of data points, this work investigated
fitting the entire data set, and data up to 10 and 15 multiplied by tc (the time for non-
linearity in t versus V2). An example showing the logarithmic fits to the complete and
partial data for 100 kPa is given in Figure 4.2.23. The φ∞ and D(φ∞) results from
curve fitting are presented in Figure 4.2.24 and Figure 4.2.25 respectively.

161
1200000
Model pred.
Full data set
1000000 tc x 15
tc x 10

800000
Time, t (s)

600000

400000 t c x 15

t c x 10
200000

tc
0
0 0.1 0.2 0.3 0.4 0.5 0.6

Average Solids Volume Fraction, <φ > (v/v)

Figure 4.2.23: Logarithmic fits to complete and partial results for 100 kPa

400
Compressive Yield Stress, Py (φ )

Py(phi)
350 Full data set
tc x 15
300
tc x 10
250
(kPa)

200

150

100

50

0
0 0.2 0.4 0.6 0.8 1

Solids Volume Fraction, φ (v/v)

Figure 4.2.24: Compressive yield stress, Py(φ), and predictions from logarithmic curve fitting of
full and incomplete data sets

162
1.0E-08

Solids Diffusivity, D (φ ) (m /s)


1.0E-09

2 1.0E-10

1.0E-11

1.0E-12

1.0E-13
D(phi)
1.0E-14 Full data set
tc x 15
1.0E-15
tc x 10
1.0E-16
0 0.2 0.4 0.6 0.8 1

Solids Volume Fraction, φ (v/v)

Figure 4.2.25: Solids diffusivity, D(φ), and predictions from logarithmic curve fitting of full and
incomplete data sets

The results show that, if the complete data set was used for the curve fitting,
the parameters corresponded to the exact value for the lower pressures, confirming
that the extraction of dewatering parameters for sewage sludges from single-pressure
filtration experiments by curve-fitting the end of the compression region was
theoretically correct. If the run was incomplete (tc x 15 and tc x 10), the predicted
values of φ∞ were lower and D(φ∞) higher than the exact values, especially for the
higher pressures (200 and 300 kPa). These results show that, especially for high-
pressure filtration runs, the run should be performed for as long as possible. Note that
the material properties used here were extreme in terms of compressibility and
permeability, and the results demonstrated a worst-case scenario. Thus, the
logarithmic fitting method was used for real materials with confidence.

163
Luggage Point Wastewater Treatment Plant

The theory shows that curve fitting a logarithmic function to the end of the
filtration data allows extraction of material property parameters (see Section 2.4.2).
As shown in Section 2.4.4, the theory was capable of producing sewage sludge-like
behaviour if D(φ) decreased significantly from φ0 to φ∞. The aim of the experimental
work presented here was twofold:

- To demonstrate and develop protocols for the experimental measurement of


fundamental dewatering characteristics of sewage sludges; and

- To investigate the functional form of the measured characteristics.

The dewaterability of three digested sewage sludge samples from Luggage


Point WWTP in Brisbane, Australia, were characterised at the University of
Melbourne using single-pressure filtration and batch settling techniques, the first two
by Casey Burgess and Peter Harbour, and the third by Timothy Birks and Nicholas
Cotterill. New computational tools and a novel algorithm were applied to enable
complete dewaterability characterisation.

The first two samples were from the same source at the same time (14/02/03)
and were not expected to differ, while the third was from the same source later in the
year (20/08/03). The initial solids concentration, determined from oven drying, was
2.758 wt%, 2.789 wt% and 2.575 wt% respectively. From measurements of the final
cake in filtration testing, the solids density was approximately 1100 kg m-3. ρf was
assumed to be 1000 kg m-3. In order to minimise biological activity, samples were
stored in the fridge, and warmed to room temperature before use. Before testing, each
sample was dosed with Zetag 87 polymer at 9 mg/g, stirred with a magnetic stirrer for
two minutes and inverted two to three times.

Pressure Filtration Results

The filtration testing used both the manual and automated filtration rigs. A
total of 18 runs were performed using the automated rig (5 for sample 1, 7 for sample
2 and 6 for sample 3) and 11 runs using the manual rig (5 for sample 1 and 6 for

164
sample 2). The final solids concentration from each run was determined by oven
drying. Table 4.2.3 gives a summary of the results from the pressure filtration testing.

Table 4.2.3: Pressure filtration experimental and curve fitting results for Luggage Point WWTP

Pressure Measured φ0 Calculated Calculated D(φ∞)


Run ID Date
(kPa) (v/v) φ0 (v/v) φ∞ (v/v) (m2/s)
A2-S1 14/02/03 2.0 0.0251 0.0839 1.46 x 10-9
A5-S1 17/02/03 5.4 0.0610 0.1298 5.96 x 10-10
A20-S1 18/02/03 23.1 0.0207 0.1774 2.85 x 10-10
A50-S1 19/02/03 50.7 0.0191 0.2640 1.56 x 10-10
A100-S1 21/02/03 101.6 0.0233 0.3317 3.04 x 10-11
0.0251
M100-1-S1 21/02/03 100.0 0.0253 0.2917 1.69 x 10-10
M100-2-S1 21/02/03 100.0 0.0231 0.2653 1.29 x 10-10
M200-1-S1 17/02/03 200.0 0.0190 0.3598 2.06 x 10-10
M400-1-S1 19/02/03 400.0 0.0201 0.4237 1.23 x 10-10
M600-1-S1 18/02/03 600.0 0.0243 0.4423 1.06 x 10-9
A2-S2 27/02/03 2.0 0.0354 0.0889 2.22 x 10-9
A5-S2 26/02/03 5.3 0.0251 0.1291 5.65 x 10-10
A20-S2 25/02/03 20.5 0.0241 0.1825 3.91 x 10-10
A50-S2 24/02/03 50.6 0.0214 0.2115 3.24 x 10-10
A100-S2 28/02/03 100.0 0.0280 0.2956 2.01 x 10-10
A200-S2 3/03/03 200.4 0.0150 0.4124 9.87 x 10-11
A400-S2 7/03/03 378.3 0.0254 0.0145 0.3594 1.81 x 10-10
M200-2-S2 24/02/03 200.0 0.0224 0.3709 1.15 x 10-10
M200-3-S2 24/02/03 200.0 0.0225 0.4498 5.38 x 10-11
M400-2-S2 28/02/03 400.0 0.0220 0.4143 1.97 x 10-10
M400-3-S2 28/02/03 400.0 0.0203 0.3911 1.78 x 10-10
M400-4-S2 3/03/03 400.0 0.0232 0.4684 4.97 x 10-11
M400-5-S2 3/03/03 400.0 0.0207 0.4099 2.06 x 10-10
A10-S3 02/09/03 10.05 0.0242 0.1821 2.86 x 10-10
A20-S3 05/09/03 20.00 0.0246 0.2121 1.48 x 10-10
A50-S3 20/08/03 50.00 0.0292 0.2452 1.65 x 10-10
0.0235
A100-S3 19/08/03 99.79 0.0197 0.3368 8.15 x 10-11
A200-S3 09/09/03 200.0 0.0104 0.3429 1.12 x 10-10
A300-S3 26/08/03 299.9 0.0123 0.4332 7.33 x 10-11

165
The run ID indicates the filtration apparatus used (A – automated, M –
manual), the applied pressure and the sample number. The manual runs also required
a run number for each repeated pressure. The date of each test is provided to indicate
the age of the sample when tested. The automated rig was used for applied pressures
from 2 to 400 kPa, while the manual rig was used from 100 to 600 kPa.

The measured φ0 values were compared with the values calculated from the
initial and final heights and the measured final solids. The ratio of the measured and
calculated φ∞ values was the same if the final solids were calculated from the initial
solids and compared to the measured cake concentrations. Any discrepancies were
due to the extrusion of water before the start of the filtration run, errors in the density
measurement, evaporation of water from the cake at the end of the run, soluble
material in the filtrate, or changes with time in the solids concentration of the sludge
after the initial measurement. The results of the curve fitting varied significantly if
the transient volume fraction profile was calculated from the measured initial or final
solids. In this work, the final solids were used as the basis for the analysis of the
transient data.

The transient piston height data gave the filtrate volume, which was converted
to the average volume fraction using a volume balance. The cake compression end
was then fitted with a logarithmic curve (see Section 3.1.5) to give φ∞ and D(φ∞). The
number of points to be fitted was determined by minimising the curve fitting error.
The fit for M400-1-S1 is illustrated in Figure 4.2.26 and the fit of for A20-S2 in
Figure 4.2.27. Since the manual rig had limited data, the curve fitting was somewhat
subjective. Several minimum errors may have existed for differing numbers of data
points – the curve fit with the end-point closest to the measured value was chosen.
This highlights the importance of frequent data collection.

166
200000
Filtration Results
180000
Compression Curve Fit
160000
Endpoint = 0.4237 v/v
140000
Time, t (s)

120000

100000

80000

60000

40000

20000

0
0 0.1 0.2 0.3 0.4 0.5

Average Solids Volume Fraction, <φ > (v/v)


Figure 4.2.26: Experimental and curve fitting results for Luggage Point M400-1-S1

120000
Filtration Results
100000 Compression Curve Fit
Endpoint = 0.1825 v/v
80000
Time, t (s)

60000

40000

20000

0
0 0.05 0.1 0.15 0.2

Average Solids Volume Fraction, <φ > (v/v)


Figure 4.2.27: Experimental and curve fitting results for Luggage Point A20-S2

167
The Py(φ) and D(φ) data determined from pressure filtration tests are presented
in Figure 4.2.28 and Figure 4.2.29 respectively. The results showed some variability,
especially amongst the manual data. This was partly due to fewer data points and
greater experimental and curve fitting error for the manual rig. For example, M600-1-
S1 (φ∞ = 0.4338 v/v, D(φ∞) = 4.27 x 10-8 m2/s) was obviously an outlier.

1000
S1 Automated rig
Compressive Yield Stress, Py (φ )

S1 Manual rig
S2 Automated rig
S2 Manual rig
S3 Automated rig
100
(kPa)

10

1
0 0.1 0.2 0.3 0.4 0.5

Solids Volume Fraction, φ (v/v)


Figure 4.2.28: Py(φ) results for Luggage Point WWTP from single-pressure filtration tests

In general, the Py(φ) results described a highly compressible sludge, both in


terms of the magnitude of φ and dPy(φ)/dφ. At high pressures, the sewage sludges
compressed to very high volume fractions, while networks existed at low
concentrations that were capable of supporting appreciable loads. This behaviour was
consistent with materials containing high molecular weight, cross-linked biopolymers
such as the ECP component of sewage sludges.

The magnitude of the D(φ) results indicated a material with very low
permeability, while the general downward trend was significant for non-traditional
filtration behaviour.

168
1.0E-08

Solids Diffusivity, D (φ ) (m /s)


2
1.0E-09

1.0E-10
S1 Automated rig
S1 Manual rig
S2 Automated rig
S2 Manual rig
S3 Automated rig
1.0E-11
0 0.1 0.2 0.3 0.4 0.5

Solids Volume Fraction, φ (v/v)

Figure 4.2.29: D(φ) results for Luggage Point WWTP from single-pressure filtration tests

Batch Settling Results

Transient batch settling was performed using measuring cylinders by


monitoring the height of the solid-liquid interface with time. The samples were
diluted for the settling tests in order to increase the initial settling rates. The h versus t
results for a transient settling test for Luggage Point WWTP Sample 2 is presented in
Figure 4.2.30. φ0 was 0.0064 v/v and h0 was 0.17 m. These results were used to
calculate low solids compressibility and permeability data (see Section 3.1.6). The
curve fit to the batch settling results that arises from this analysis is included in Figure
4.2.30.

169
0.18

0.17 Batch settling results

Interfacial height, h (t ) (m)


Analysis curve fit
0.16

0.15

0.14

0.13

0.12

0.11

0.1

0.09

0.08
0 5000 10000 15000 20000
Time, t (s)

Figure 4.2.30: Transient batch settling test results for Luggage Point WWTP Sample 2

Calculations

The automated filtration rig results for Sample 2 were analysed using the
procedure outlined in Section 3.2.2. The batch settling results analysis required high
volume fraction R(φ) and Py(φ) data from filtration testing. Equation 3.2.1 requires a
curve fit of Py(φ) in order to calculate R(φ). Initially, φg was estimated as the diluted
φ0 value, allowing the filtration results of ∆P versus φ to be fitted to a function such as
equation 2.1.6. Thus, for the compressibility results from Sample 2, φg = 0.0064 v/v,
p1 = 40.38 Pa and p2 = 2.131. Using this fit in conjunction with the D(φ∞) results and
equation 3.2.1 gave the high volume fraction R(φ∞) (9.854x1013, 5.393x1014,
1.016x1015, 1.347x1015, 2.531x1015, 5.227x1015, 2.899x1015 Pasm-2). The batch
settling analysis could now proceed. Note that the batch settling analysis programme
requires R(φ) to be monotonically increasing, therefore, outliers were removed. For
Sample 2, the penultimate datum point was removed.

The batch settling analysis uses two user-defined variables, b and φcp, to define
the function (equation 3.1.2) to be fitted to Py(φ). They were iterated upon until the
best fit to the data was achieved. Using b = 1.00 and φcp = 0.63 v/v, R(φ) was

170
recalculated to give (2.70x1013, 1.95x1014, 5.23x1014, 8.50x1014, 3.14x1015,
2.48x1016, 6.91x1015 Pasm-2). The batch settling analysis was repeated with these
new values, giving a new fit to Py(φ) and R(φ) (see Figure 4.2.31 and Figure 4.2.32
respectively).

1000
Compressive Yield Stress, Py (φ )

100

10

1
(kPa)

0.1
Pressure Filtration
0.01 Batch settling
Curve fit
0.001
phig = 0.0111 v/v
0.0001
0 0.1 0.2 0.3 0.4 0.5

Solids Volume Fraction, φ (v/v)

Figure 4.2.31: Py(φ) results and curve fits from filtration and batch settling for Luggage Point
WWTP Sample 2

The Py(φ) results showed a dramatic drop in magnitude as φ decreased towards


φg. This indicated that a very weak network existed at low concentrations – so weak
that an applied pressure of 2 kPa was able to compress it to 0.09 v/v. The material
was extremely compressible, such that high concentrations were possible at high
pressures. The best-fit parameters for equation 3.1.2 are given in Table 4.2.4.

The R(φ) results are shown in Figure 4.2.32. At very low φ, the sewage
sludges were somewhat permeable, but increasingly impermeable due to the
formation of the network at the gel point. There was a huge increase in R(φ) of eight
orders of magnitude over the volume fraction range measured by batch settling and
pressure filtration. This effect was also seen for flocculated mineral suspensions and
water treatment sludges, but the observed change was much less than exhibited here

171
for sewage sludges. Very few engineering parameters vary by this amount within a
unit operation.

Table 4.2.4: Py(φ) curve fitting parameters for Luggage Point WWTP Sample 2

Parameter Value Parameter Value

φg 0.01110 φg2 0.00987

a1 0.004745 a2 0.005042

b1 1.00 b2 1.00

k1 2.3151 k2 2.3475

φcp 0.6300 φp 0.0921

1E+17
Hindered Settling Function, R (φ )

1E+16

1E+15

1E+14
(Pas/m )

1E+13
2

1E+12

1E+11
Pressure Filtration
1E+10
Batch Settling
1E+09 Curve fit
1E+08 phig = 0.0111 v/v
1E+07
0 0.1 0.2 0.3 0.4 0.5

Solids Volume Fraction, φ (v/v)

Figure 4.2.32: R(φ) results and curve fits from filtration and batch settling for Luggage Point
WWTP Sample 2

D(φ) was calculated from Py(φ) and R(φ) using equation 2.2.18. The results
are presented in Figure 4.2.33 along with the filtration results, showing good
agreement. The curve fit showed a maximum near the gel point, and was then
monotonically decreasing. From the definition of D(φ), R(φ) increased at a faster rate

172
than dPy(φ)/dφ with increasing φ and Py(φ) was significant at low volume fractions.
As shown in Section 2.4.4, materials with such characteristics show short cake
formation times and long compression times, as observed for sewage sludges.

1.0E-06
Filtration results
Solids Diffusivity, D (φ ) (m /s)
2

1.0E-07 Curve fit


phig = 0.0111 v/v

1.0E-08

1.0E-09

1.0E-10

1.0E-11
0 0.1 0.2 0.3 0.4 0.5

Solids Volume Fraction, φ (v/v)

Figure 4.2.33: D(φ) results and curve fits from filtration and batch settling for Luggage Point
WWTP Sample 2

Experimental Validation

The determination of D(φ) for sewage sludges was a difficult and somewhat
subjective task depending on the accuracy of Py(φ) and R(φ) data and subsequent
curve fits. Validation using model predictions was conducted to confirm that the
curve fits represented a valid dewaterability characterisation, by comparing the
experimental data with modelling predictions of batch settling and pressure filtration
tests.

The batch settling curve fit is shown along with the experimental results in
Figure 4.2.30. The model results matched the experimental results closely, showing
that the low φ material characteristics were valid. There was a small discrepancy at
intermediate times due to the limited data, highlighting the need for frequent
measurement.

173
The filtration results were validating using the Runge-Kutta numerical
algorithm described in Section 2.5.1. The inputs were the material characteristics for
Luggage Point sample 2 (Py(φ), D(φ)) and the initial conditions for each filtration run
(∆P, φ0 and h0). In the results presented here, h0 was adjusted using a volume balance
to ensure that V∞ corresponded to the measured values, thus removing the
experimental error between the measured φ0 and φ∞ values:

φ ∞V∞
h0 = …(4.2.4)
(φ ∞ − φ0 )

The experimental and modelling results for the seven pressures used with the
automated rig for Luggage Point sample 2 are presented in Figure 4.2.34.
Encouragingly, the predictions were similar to the experimental results for all the
pressures except 5 kPa. In general, the model predicted a higher initial rate of
dewatering and longer cake compression, indicating that the calculated D(φ) was too
high at high φ and too low at low φ. Given that the curve fit in Figure 4.2.33 already
varied by over four orders of magnitude, this outcome was somewhat astounding.
This indicated that testing should include permeation and centrifugation experiments
to characterise the sludge at intermediate volume fractions.

174
160000 100000
Filtration Results - 2 kPa 90000 Filtration Results - 5 kPa
140000
Model Predictions Model Predictions
80000
120000
70000
Time, t (s)
100000
60000

80000 50000

40000
60000
30000
40000
20000
20000 10000

0 0
0 0.0001 0.0002 0.0003 0.0004 0.0005 0 0.0001 0.0002 0.0003 0.0004 0.0005
100000 60000
Filtration Results - 20 kPa Filtration Results - 50 kPa
90000
Model Predictions 50000 Model Predictions
80000

70000
40000
Time, t (s)

Time, t (s)
60000

50000 30000

40000
20000
30000

20000
10000
10000

0 0
0 0.0002 0.0004 0.0006 0.0008 0 0.0001 0.0002 0.0003 0.0004 0.0005 0.0006
100000 60000

90000 Filtration Results - 100 kPa Filtration Results - 200 kPa


Model Predictions 50000 Model Predictions
80000

70000
40000
Time, t (s)

Time, t (s)

60000

50000 30000

40000
20000
30000

20000
10000
10000

0 0
0 0.0002 0.0004 0.0006 0.0008 0.001 0 0.0002 0.0004 0.0006 0.0008

100000

90000 Filtration Results - 400 kPa


Model Predictions
80000

70000
Time, t (s)

60000

50000

40000

30000

20000

10000

0
0 0.0002 0.0004 0.0006 0.0008 0.001
2 2
(Specific Filtrate Volume) , V (m2)

Figure 4.2.34: Validation of experimental results using modelling predictions for Luggage Point
WWTP Sample 2

175
Conclusions

Batch settling and pressure filtration experiments for sewage sludge samples
from Luggage Point WWTP were performed in order to measure the compressional
rheological properties. The results showed that φ∞ and D(φ∞) could be determined
from a series of single pressure filtration experiments, and that, when combined with
batch settling results, gave Py(φ) and R(φ) functions. D(φ) was determined by a novel
algorithm, identifying a peak near the gel point followed by monotonically decreasing
D(φ) with increasing φ, confirming the hypothesis that this feature exists for biosludge
materials. Comparing the experimental results with model predictions validated the
characterisation.

This method represents a significant breakthrough in the characterisation of


sewage sludges. However, the single pressure filtration results showed significant
variability, especially when using the manual rig. Great care must be taken when
performing the tests in order to minimise experimental errors due to evaporation,
biological activity, et cetera. The methods applied here were developed into a general
protocol for the full dewaterability characterisation of sewage sludges, to be utilised in
research and industry. The challenge for future work is to use such methods to
quantify, rather than just qualify, the effect of different treatments and additives.

Other Wastewater Sludges

While the full characterisation of wastewater sludges for modelling purposes


required the combination of batch settling and single-pressure filtration testing, the
results from numerous single-pressure filtration tests were analysed using the cake
compression method. Nevil Anderson and Peter Harbour of CSIRO tested a range of
wastewater sludges from various treatment plants using pressure filtration. These
include the single pressure filtration runs for conditioned and un-conditioned sludge
from Carrum, Colac, Lilydale and Mornington presented in Section 4.2.1, as well as a
sample from Oxley Creek WWTP (Queensland) and a series of tests in conjunction
with this author on un-conditioned Carrum activated sludge. As well as these, Peter
Hillis of Unitied Utilities tested a sample from Ellesmere Port WWTP in the United
Kingdom. The diffusivity results using the cake compression log fitting procedure are
presented in Figure 4.2.35.

176
Carrum Raw
1.0E-07 Carrum 10 mg/g Z87
Colac Raw
Colac 9.44 mg/g Z87
Solids Diffusivity, D (φ ) (m /s) Ellesmere Port
2
1.0E-08 Lilydale Raw
Lilydale 11 mg/g Betz 1158
Lilydale 9.44 mg/g Z87
Luggage Point 9 mg/g Z87
1.0E-09 Mornington 9.44 mg/g Z87
Oxley Creek 7 mg/g Z87

1.0E-10

1.0E-11

1.0E-12
0 0.1 0.2 0.3 0.4 0.5 0.6

Solids Volume Fraction, φ (v/v)

Figure 4.2.35: D(φ) from single-pressure filtration of a range of wastewater treatment sludges

In general, the results indicated that sewage sludges were highly compressible
suspensions with low diffusivity. The unconditioned sludge from Carrum clearly had
the lowest D(φ), while the conditioned sludges exhibited higher values. Of the Colac
sludges, conditioning improved the dewaterability by an order of magnitude. The
results for the Lilydale sludges did not show the expected trend since the
unconditioned sample had higher D (but lower φ) than the Z87 sample. This may
have been due to fitting errors caused by the limited number of data points from the
manual rig. For this same reason, the manual rig data for unconditioned Mornington
and Oxley Creek samples were not used. Overall, the results illustrated that the cake
compression logarithmic fitting method was capable of differentiating sludges and
additives.

177
4.2.3 Synthetic Sewage Sludges

The dewatering behaviours of several synthetic sludges were investigated


using constant pressure batch filtration. The aim of the work was to manufacture a
synthetic sludge that mimicked the filtration characteristics of real sewage sludges. A
biologically and chemically stable synthetic sludge can be used to investigate the
fundamental dewatering behaviour of sewage sludges without any of the inherent
problems usually encountered with laboratory testing of active sludges.

Introduction

The inherent microbial activity of both digested and activated sewage sludges
cause their physical and chemical characteristics to change constantly, which makes it
difficult to carry out extensive series of controlled experiments or to reproduce test
results. The dewatering properties may vary considerably, not only between samples,
but also within samples over relatively short time periods. There is clearly a need for
a synthetic sludge with similar physical and chemical properties to sewage sludge.
The study of a synthetic sludge also allows the possibility of identifying and
examining in detail the components of sludge that are responsible for the slow
dewaterability (Ormeci and Vesilind, 2000).

To produce synthetic sludge that mimics sewage sludge in filtration, the


physical properties should be:

- Solids density of 1100 to 1400 kg/m3;

- Extremely low permeability (long filtration time); and

- High compressibility (high equilibrium volume fraction).

The task of producing a synthetic sludge is by no means a trivial one.


Naturally, the criteria applied depend on the topic under investigation. In terms of
dewaterability the permeability and compressibility are of great importance. The
concept of making a synthetic sludge is not new and has been attempted by a number
of researchers. Sanin and Vesilind, 1996, 1999, combined polystyrene latices (used to

178
simulate bacteria), sodium alginate (used to simulate ECP), and calcium ions (to allow
for bridging and complexation of the carboxylate groups). Ormeci and Vesilind, 2000
advanced this model by adding cellulose fibres to simulate the filamentous bacteria
and fibrous particles commonly found in sewage sludges. By comparing SRF, CST,
viscosity and optimum polymer dose they proposed that the properties of the synthetic
sludge were very similar to those of real activated sewage sludge. However, the CST
values obtained were in the order of 12 to 25 seconds. With such low initial
concentrations (0.0017 v/v) and such short times there was barely time to create the
filter cake that was necessary to measure the permeability. In fact these short times
are very close to that of pure water (approximately 8 seconds with a standard collar)
and different results were more likely to be changes in filtrate viscosity than changes
in sludge behaviour.

Unlike CST, constant pressure filtration provides both rate and extent
information. The aim of this work was to compare the synthetic sludge developed by
Vesilind and co-workers (a typical formulation is outlined in Section 3.3.3) with real
sewage sludge from Carrum WWTP in Melbourne, Australia, using single-pressure
constant pressure filtration tests at 20 kPa. Qualitative comparisons were made using
the scaling procedure outlined in Section 2.4.2 in order to negate any effects due to
different starting conditions of h0 and φ0.

The components of the synthetic sludge were varied in order to elucidate the
mechanism that causes slow filtration for sewage sludges. Yeast and gelatine were
used as replacements for the polystyrene and alginate respectively.

Results and Discussion

Alginate-based synthetic sludge

Figure 4.2.36 shows the adjusted time of filtration versus scaled filtrate
volume for the Carrum WWTP sludge and the alginate-based synthetic sludge at an
applied pressure of 20 kPa. The extremely high rate of filtration for the alginate-
based synthetic sludge negated the use of higher pressures. The Carrum sludge took
approximately 100 hours to reach an equilibrium volume fraction of 0.233 v/v. This
was an extremely long time but not atypical of un-flocculated sewage sludges. In
comparison, the synthetic sludge based on the Vesilind formulation dewatered several

179
orders of magnitude too quickly and to a lesser extent when compared to the sewage
sludge.

1.E+14
Carrum WWTP Sludge
Adjusted Time, c (φ0 )/c (φ0,ref ) x t /h02
1.E+13 Alginate Synthetic Sludge

1.E+12

1.E+11
(s/m2)

1.E+10

1.E+09

1.E+08

1.E+07
0.0001 0.001 0.01 0.1 1

(1-φ 0 /φ )
2

Figure 4.2.36: Scaled single-pressure filtration results for Carrum WWTP sewage sludge and
alginate synthetic sludge based on the model proposed by Vesilind

Variation of particle composition

The scaled results for the addition of yeast particles to the mixture as a
replacement for polystyrene are shown in Figure 4.2.37. Various ratios were
considered. The results show that the addition of yeast particles decreased the
permeability by orders of magnitude. The decreased permeability was consistent with
increasing tortuosity caused by malleable particles (Meireles, et al., 2004), but did not
adequately explain the extremely low permeability of the sewage sludge.
Interestingly, the trend shown by the sludges with yeast present was to decrease
permeability when there was less polystyrene, which was consistent with the smaller
polystyrene particles filling the interstitial sites of the yeast.

180
1.E+14
Carrum WWTP Sludge
Polystyrene

Adjusted Time, c (φ0 )/c (φ0,ref ) x t /h02


1.E+13 50:50 Yeast:Polystyrene
75:25 Yeast:Polystyrene
Yeast
1.E+12

1.E+11
(s/m )
2

1.E+10

1.E+09

1.E+08

1.E+07
0.0001 0.001 0.01 0.1 1

(1-φ 0 /φ )
2

Figure 4.2.37: Scaled single-pressure filtration results for Carrum WWTP sewage sludge and
alginate synthetic sludges with varying particle composition

Variation of ECP substitute

Gelatine was investigated as a substitute for alginate in order to study the role
of the ECP substitute. Figure 4.2.38 shows the adjusted time of filtration versus the
scaled volume results for the gelatine-based synthetic sludge compared to the Carrum
WWTP sludge. The synthetic sludge had very similar results to the real sludge,
despite being of similar appearance to the alginate-based synthetic sludges.

The results show clearly that the nature of the ECP substitute had a drastic
effect on the dewaterability of synthetic sludges, indicating that it was the ECP
component of the sewage sludge that caused its poor dewaterability. The decreased
permeability, as compared to the alginate-based sludge, was due either to a gelatine-
based polymeric network being formed, or gelatine-stabilised, dispersed particles
causing membrane or cake fouling.

181
1.E+14

Adjusted Time, c (φ0 )/c (φ0,ref ) x t /h02


1.E+13

1.E+12

1.E+11
(s/m2)

1.E+10

1.E+09

Carrum WWTP Sludge


1.E+08
Gelatine Synthetic Sludge

1.E+07
0.0001 0.001 0.01 0.1 1

(1-φ 0 /φ )
2

Figure 4.2.38: Scaled single-pressure filtration results for Carrum WWTP sewage sludge and
gelatine synthetic sludge

Conclusions

The synthetic sludge formulation of Ormeci and Vesilind, 2000, produced a


synthetic sludge with inadequate filtration properties, highlighting the importance of
performing filtration analysis to quantify dewaterability rather than just CST. By
changing the components of this formulation, it was shown that the character of the
particulates has small effects on alginate-based synthetic sludges, but do not
adequately explain the poor filterability of sewage sludges. By substituting gelatine in
place of alginate as the ECP surrogate, a synthetic suspension with similar filtration
times to sewage sludge was produced. This shows that the ECP component of sewage
sludges predominantly determines its dewatering characteristics. On this basis alone,
either denaturation or titration of this component is essential to improving
dewaterability of the sludges. Gelatine is more proteinaceous than alginate,
suggesting that it is proteinaceous fraction of ECP that determines the poor
dewaterability.

182
4.3 Other Sludges

Several sludges beyond water and wastewater sludges were characterised


using pressure filtration testing, including kaolin and pulp-and-paper. The results are
presented in this section, along with an overall comparison of different sludge types.

4.3.1 Kaolin (Consolidation Testing Methods)

The material characteristics of a kaolin sample were measured using


oedometer testing (see Section 3.1.7). Once cv was determined, it was converted to
D(φ) according to the relationship developed in Section 2.6.2 (equation 2.6.9), and the
results compared to the results from filtration testing.

While it was clear that the physical circumstances of oedometer and filtration
testing were the same, significant differences between the analysis methods existed, as
outlined in Table 2. Generally, the graphical methods for consolidation analysis fit
theoretical behaviour to the raw experimental data and extract a characteristic time
value in order to determine cv, whereas the filtration method uses Py(φ) and β2(∆P) as
determined directly from the experimental results, and combines them using empirical
functional fits to give D(φ).

Table 4.3.1: Comparison of consolidation and filtration analyses

Consolidation Analysis Filtration Analysis

U = 0 determined by assuming parabolic or


Results are independent of initial fluctuations
logarithmic behaviour

U = 1 determined by fitting experimental data to φ∞ determined when dt/dV2 increases beyond a


theoretical behaviour calculated value, independently of behaviour

Fundamental values of Py(φ) and β2(∆P)


Extract t50 or t90 by assuming theoretical
determined directly from data and fitted with
behaviour to give cv
empirical curves to give D(φ)

183
cv from Oedometer testing

The same sample of kaolin, pre-consolidated at 25 kPa, was tested using two
adjacent oedometers at nominal pressures of 50, 200, 400 and 800 kPa. The results
were analysed using the two graphical methods outlined in Section 3.1.7 –
Casagrande’s log-time method (Casagrande and Fadum, 1940) and Taylor’s root-time
method (Taylor, 1942).

As an example of the application of Casagrande’s method, Figure 4.3.1 shows


a plot of bed height versus log-time for the oedometer test of kaolin at 100 kPa.

5.60

5.50 U =0, 5.4563 mm


Dial Reading, h (mm)

5.40

5.30 5.287 mm

5.20
U =0.5, 5.1734 mm
5.10 5.1177 mm

5.00
U =1, 4.8905 mm
4.90

4.80
t 50 = 274s
10 100 1000 10000
Time, t (s)

Figure 4.3.1: Casagrande's method applied to oedometer sample 1, 100kPa

The bed height corresponding to no compression (U = 0) was determined by


assuming a parabolic relationship between the dial readings at t = 100s (ht=100 =
5.2870 mm) and t = 400s (ht=400 = 5.1177 mm) to yield h0 = 5.4563 mm. The values
of 100 and 400 seconds were chosen arbitrarily. The bed height corresponding to
complete compression (U = 1) was taken as the intersection of the two linear sections,
giving 4.8905 mm. Therefore, h50 = 5.1734 mm and t50 = 274 s. The final dial
reading for Sample 1 was 2.168 mm and the final height was 14.73 mm, therefore the
difference between the dial reading and the actual sample thickness was 12.562 mm.

184
The initial and final sample thicknesses were taken at U = 0 and U = 1 respectively,
such that the average pore length, d, was 8.868 mm. From equation 3.1.8, cv = 5.625
x 10-8 m2/s.

Figure 4.3.2 shows a plot of bed height versus root-time for the same results as
above as an example of the application of Taylor’s method. A straight line was drawn
through the linear portion of the plot. The y-intercept gave the corrected zero point (U
= 0, h0 = 5.450 mm). A second line was drawn with an abscissa 1.15 times the first
line, as specified by Taylor’s method, and the intersection of this line with the
filtration data corresponded to the time for 90% consolidation, t90 = 1173s. From this
analysis, d was 8.869 mm. Therefore, from equation 3.1.9, cv = 5.686 x 10-8 m2/s.

5.60

5.50 U =0, 5.450 mm


Dial Reading, h (mm)

5.40

5.30

5.20

5.10

5.00

4.90 U =0.9, 4.956 mm

4.80
t 90 = 1173 s
0 20 40 60
1/2 1/2
Root-Time, t (s )
Figure 4.3.2: Taylor's method applied to oedometer sample 1, 100kPa

The same analysis methods were applied to all the pressures for both
oedometer tests. The results are given in Table 4.3.2, and are presented along with
average values for e and cv at each pressure in Figure 4.3.3. The results show
reasonable agreement between the two graphical methods, especially at 100 kPa,
considering the two methods can differ by up to 4 times for the same experimental
results (Duncan, 1993). The value for Taylor’s method, Sample 1 at 400 kPa (e =

185
1.0747) was assumed to be an outlier and was ignored when calculating the average
cv. Casagrande’s method generally yielded smaller values than Taylor’s method,
confirming Olson’s statement that ‘cv from root-t plots almost always exceeds cv from
log t plots’ (Olson, 1986).

Table 4.3.2: Oedometer testing results using Taylor’s and Casagrande’s methods

Taylor’s Method Casagrande’s Method

∆P, kPa e, v/v cv, m2/s e, v/v cv, m2/s

Sample 1

50 1.3561 3.97 x 10-8 1.3485 2.94 x 10-8

100 1.2612 5.69 x 10-8 1.2598 5.63 x 10-8

200 1.1636 9.47 x 10-8 1.1559 6.68 x 10-8

400 1.0747 2.96 x 10-7 1.0499 9.36 x 10-8

800 0.9401 1.44 x 10-7 0.9332 8.82 x 10-8

Sample 2

50 1.3666 5.22 x 10-8 1.3678 5.56 x 10-8

100 1.2596 5.09 x 10-8 1.2607 5.31 x 10-8

200 1.1612 8.67 x 10-8 1.1553 6.72 x 10-8

400 1.0553 1.05 x 10-7 1.0469 7.33 x 10-8

800 0.9446 1.12 x 10-7 0.9432 8.75 x 10-8

186
1.0E-06

Consolidation Coefficient, cv (m /s)


2

1.0E-07

S1 - Taylor
S1 - Casagrande
S2 - Taylor
S2 - Casagrande
Oedometer (avg)
1.0E-08
0.8 0.9 1 1.1 1.2 1.3 1.4
Void Ratio, e (v/v)

Figure 4.3.3: cv(e) results from oedometer testing for kaolin

D(φ) from Filtration Testing

The kaolin sample was also tested using stepped-pressure filtration at nominal
pressures of 25, 50, 100, 200 and 400 kPa. The compressibility data was fitted to a
power-law relationship given by equation 2.1.6. The experimental data and the
functional fit are presented along with the average compressibility results from the
oedometer tests in Figure 4.3.4. There was good agreement between the two methods
although the oedometer tests gave slightly higher final volume fractions at lower
pressures (50 and 100 kPa), and lower values at higher pressures (400 and 800 kPa).

β2(∆P) from the permeability stepped-pressure filtration test was fitted to an


exponential function and combined with Py(φ) according to equation 2.4.33 to give
D(φ). The oedometer results were converted from cv(e) to D(φ) using equation 2.6.9,
where the initial volume fraction, φ0, for pressures other than the first pressure was the
final volume fraction for the previous applied pressure. The results are presented
along with the filtration results in Figure 4.3.5.

187
900

Compressive Yield Stress, Py (φ )


S1 - Taylor
800 S1 - Casagrande
S2 - Taylor
700 S2 - Casagrande
Oedometer (avg)
600
Filtration Test
500
(kPa)

400

300

200

100

0
0.35 0.4 0.45 0.5 0.55

Solids Volume Fraction, φ (v/v)


Figure 4.3.4: Py(φ) results from filtration and oedometer testing of kaolin sample

1.0E-06
Filtration Test
Solids Diffusivity, D (φ ) (m /s)

S1 - Taylor
S1 - Casagrande
2

S2 - Taylor
S2 - Casagrande
Oedometer (avg)

1.0E-07

1.0E-08
0.35 0.4 0.45 0.5 0.55

Solids Volume Fraction, φ (v/v)


Figure 4.3.5: D(φ) results from filtration and oedometer testing of kaolin sample

188
The results for the two methods show reasonable agreement, considering that
different materials can vary by up to 4 orders of magnitude, different samples of the
same material can vary by up to one order of magnitude, and even different analysis
methods of the same data can lead to differences of up to 4 times (Duncan, 1993). In
general, the oedometer test gave slightly lower D values than the filtration test, which
may be due to high strain. The results support the hypothesis that filtration testing is
essentially the same as oedometer testing, and that variations stem from differences in
analysis rather than fundamental differences between consolidation and filtration
processes, or between cv and D(φ). If the tests are equivalent, cv and D(φ) are
interchangeable and can be used with either the geotechnical or filtration models to
predict the rate and degree of dewatering of suspensions and soils.

Comparison of Results using Filtration Model

The analytical stepped-pressure filtration model (Usher, et al., 2001) was used
to compare the results and as an example of using oedometer testing to predict
filtration behaviour. Since cv and D are equivalent parameters, they can be used
interchangeably in either consolidation or filtration modelling and should be
independent of the experimental method. The average D(φ) results from the
oedometer testing were fitted with an exponential curve for use as an input to the
model:

D (φ ) = 7.9502 × 10 −10 exp(9.5548φ ) …(4.3.1)

The parameters for equations 2.1.6 and 2.4.50 from filtration testing are given
in Table 4.3.3.

Table 4.3.3: Fitting parameters for kaolin from filtration testing

Parameter Value Parameter Value

φg (v/v) 0.2368 d1 (m2/s) 0.01053

p1 (Pa) 61.149 d2 11.041

p2 12.041 d3 4.884

189
The inputs to the model came directly from the stepped-pressure
compressibility test, allowing direct comparison with experimental data. The inputs
to the model were the applied pressure at each step, P = [25576.5, 50517.7, 100216,
200008, 319715] Pa, the filtration time at each pressure, t = [61215.1, 86289.5,
109525, 130536, 153165] s, the initial solids concentration, φ0 = 0.2369 v/v, and the
material characteristics of Py(φ) and D(φ).

Three predictions are presented along with the raw experimental results in
Figure 4.3.6. The first prediction used Py(φ) and D(φ) as determined by the filtration
testing. This prediction gave a close fit to the experimental data, except at low
pressures due to the inaccuracy of the curve fit for β2(∆P). The specific volume at the
end of each pressure was the same as the experimental data, since this data was used
to give Py(φ).

160000
Experimental results
D(phi) predictions
140000 cv predictions (Py from filtration)
cv predictions (Py from oedometer)
120000
Time, t (s)

100000

80000

60000

40000

20000

0
0 0.0001 0.0002 0.0003
2 2 2
(Specific Filtrate Volume) , V (m )
Figure 4.3.6: Experimental results and model predictions for compressibility stepped-pressure
filtration test for kaolin

The second prediction presented in Figure 4.3.6 used the cv results from the
oedometer testing and the compressibility data from the filtration testing, allowing a
direct comparison between the magnitude of cv as determined by oedometer testing

190
and D as determined by filtration testing since the specific volumes at the end of each
pressure were the same. cv under predicted the slope at low pressures, suggesting that
the filtration results were more realistic than the oedometer results for low pressures.
This was expected since the strains were large.

The third prediction presented in Figure 4.3.6 used both the cv and
compressibility results from the oedometer testing, giving an indication of the overall
differences between the two testing methods as well as giving an example of how the
oedometer test can be used to predict filtration behaviour. The compressibility data
from the oedometer testing was fitted with a power-law function:

Py (φ ) = 1.079 × 1010 (φ − 0.2 )


8.192
…(4.3.2)

The differences in end-points between this prediction and the other predictions
arose from the different compressibility data (see Figure 4.3.4). This had an added
effect on the slope of t versus V2 as shown in the variation between the second and
third predictions.

Conclusions

Terzaghi’s consolidation coefficient, cv, was measured using oedometer


testing, while the solids diffusivity, D(φ), was measured using filtration testing. The
two parameters were shown to be equivalent and related by a simple equation. The
results for a kaolin sample were measured using both techniques, and compared using
the derived equation. The results were used as inputs to a stepped-pressure filtration
model to validate the results. The two experimental methods gave reasonably similar
results and it was hypothesised that the degree of variation arose from different
methods of analysis, rather than fundamental differences in the experimental
techniques or properties being measured.

By illustrating that oedometer and filtration testing are equivalent, the


opportunity arises to use filtration testing for the determination of consolidation
properties, giving access to new methodologies and techniques. Likewise,
consolidation testing and methodologies can be used for the determination of filtration
properties.

191
4.3.2 Pulp-and-Paper Sludges

Stepped-pressure filtration tests were performed on a suspension consisting of


just the Kraft process fibres used in the manufacture of synthetic sludges (see Section
4.2.3) in order to ascertain the role of the fibres in the sludge matrix. Also presented
here is work by Colin Soh at CSIRO who characterised pulp-and-paper sludges from
two industrial sites in Australia, designated here as A and B. The exact details are
confidential, but B3 was the feed to the belt press while B1 and B2 were from
treatment ponds and had increasing amounts of biological material.

The compressibility results for the pulp-and-paper sludges are presented in


Figure 4.3.7, which show a wide variety of behaviour. The Kraft fibres were
measured from 5 to 200 kPa, while the other sludges were measured from 20 to 300
kPa. The Kraft fibres and B3, both nominally just fibrous material, compressed to
very similar equilibrium concentrations, while the suspensions with more biological
material were the most compressible.

350
Compressive Yield Stress, Py (φ )

300

250

200
(kPa)

Kraft
150 A1
A2
100 A3
B1
50 B2
B3
0
0 0.2 0.4 0.6

Solids Volume Fraction, φ (v/v)


Figure 4.3.7: Py(φ) results for stepped-pressure filtration of pulp-and-paper sludges

192
Figure 4.3.8 shows the R(φ) results for the pulp-and-paper sludges, again
showing a wide range of behaviour. The results followed the same trend as the Py(φ)
results, such that the least compressible sludges were also the least permeable, except
for the Kraft fibres, which were quite permeable compared to A3 and B3 despite
being as compressible. The outcome of these trends is shown in Figure 4.3.9, such
that D(φ) and φ both increase from sludge to sludge.

1E+15
Hindered Settling Function, R (φ )

Kraft
A1
A2
A3
B1
1E+14 B2
B3
(Pas/m )
2

1E+13

1E+12
0 0.2 0.4 0.6

Solids Volume Fraction, φ (v/v)


Figure 4.3.8: R(φ) results for stepped-pressure filtration of pulp-and-paper sludges

Each sludge exhibits a turning point in D(φ), except the Kraft fibres that are
monotonically decreasing with increasing φ. The t versus V2 results reveal
increasingly non-traditional behaviours with increasing pressures, which is consistent
with the outcome given in Section 2.4.4. The magnitude of the material
characteristics of the Kraft fibres suggests that they do not contribute significantly to
the high compressibility and low permeability of the synthetic sludges, further
confirming the hypothesis that the behaviour is predominantly determined by the ECP
substitute.

193
1.E-07

Solids Diffusivity, D (φ ) (m /s)


2

1.E-08
Kraft
A1
A2
A3
B1
B2
B3
1.E-09
0 0.2 0.4 0.6

Solids Volume Fraction, φ (v/v)


Figure 4.3.9: D(φ) results for stepped-pressure filtration of pulp-and-paper sludges

194
4.3.3 Overall Sludge Comparison

A wide variety of sludges have been characterised using the filtration


techniques developed at the University of Melbourne, as outlined in Table 4.3.4. This
list is not exhaustive, but is presented here to show the breadth of application. The
sludges varied from entirely inorganic to organic, and the solid components were
flocculated or dispersed, but all were described by compressional rheology theory.

Table 4.3.4: Summary of other sludges characterised using pressure filtration

Sludge Description Reference

Abattoir Activated wastewater sludge Unpublished, Harbour, P. J. (CSIRO)

Alumina AKP30, pH 8.3 Abd Aziz, 2004

Antarctic Tip Ferric water treatment Northcott, 2004

Calcite Unpublished, Gladman, B.

Dairy Organic waste Unpublished, Huber, F. (CSIRO)

Hematite Synthetic Bayer Liquid Usher, 2002

MIEX Ion Exchange Regeneration effluent Unpublished, Kennedy, M.

Red Mud Bauxite process liquor; 50°C Usher, 2002

Zirconia pH 6.8 de Kretser, et al., 2001

The characterisation results for these sludges and the sludges presented in this
chapter are combined in Figure 4.3.10, which is dubbed the ‘Mother-of-all-Graphs’
since it enables comparisons of all types of sludges. Generally, sludges that show
high equilibrium solids and high diffusivities (that is, in the top right-hand corner) are
the best sludges to process in terms of throughput and solids concentration. Many
mineral sludges exhibit this behaviour and, consequently, thickening tanks are
predominantly used for dewatering operations. This is in contrast to sludges that lie in
the bottom left-hand corner, which require high pressures and long processing times
to reach low concentrations. For such sludges, other dewatering mechanisms such as
de-saturation and evaporation may be useful.

195
1.E-04

Solids Diffusivity, D (φ ) (m /s) 1.E-05


2
Mineral
1.E-06
Industrial
1.E-07

1.E-08
WT Pulp&Paper
1.E-09

1.E-10
Sewage
1.E-11

1.E-12
0 0.2 0.4 0.6

Solids Volume Fraction, φ (v/v)

Figure 4.3.10: D(φ) results from pressure filtration for a variety of sludges

Figure 4.3.10 puts water and wastewater treatment sludges into context. The
alum and ferric water treatment sludges exhibited very low equilibrium solids and
medium to low diffusivities. In contrast, sewage sludges reached high concentrations,
but took a very long time to get there – their diffusivity was up to six orders of
magnitude lower than some mineral samples at comparable volume fractions (for
example, Carrum activated sludge compared to calcite). To illustrate the point,
compared to a mineral sample that dewaters in 10 seconds, a sewage sample under the
same conditions takes 107 seconds (about 4 months) to dewater.

While some variability was seen in the water treatment results presented
earlier in this Chapter between sites, additive dose and preparation conditions, the
magnitude was very small compared to the changes seen between different materials.
The limited results for flocculated wastewater sludges suggest that additives may help
improve the diffusivity by about an order of magnitude, but operators are still required
to process a highly impermeable sludge.

196
5 Plate-and-Frame Filter
Press Modelling

This chapter details the formulation of models of plate-and-frame filter presses


and subsequent validation through several case studies. The models were used to
identify key process variables and optimum operating conditions for throughput and
final cake solids, based on variations of performance due to changes to operation or
material. Visual basic tools were developed to enable operators and designers to use
the model results without requiring knowledge of the laboratory measurement, theory
or programming within the model.

5.1 Model Formulations

Plate-and-frame filter presses are used extensively in a wide range of


industries to dewater various types of sludges, both organic and inorganic materials,
including wastewater treatment, sewage, mineral, coal and dairy sludges. Such
filtration units consist of a vertical or horizontal concertina of plates. The plates are
indented, such that a cavity of width d is formed between two plates when the press is
closed. Semi-permeable membranes, or filter cloths, are mounted to the faces of the
plates. Through the introduction of a slurry under a pressure, a cake forms against the
membrane, and dewatering occurs. The membrane separation may be fixed or
variable, depending on whether or not the press has an air- or water-driven squeeze
phase.

Plate-and-frame filter presses are batch-operated devices, where tT is the total


cycle time. The batch process (see Figure 5.1.1) consists of three or more stages,
depending on the type of filter press.

197
membrane h(t)
filtrate ports h0

plate d
V V V V

feed ∆PF

∆PS

∆PS
port φ0

V V V V

filtrate ports
(a) (b) (c) (d)
(0 ≤ t ≤ tL) (tL < t ≤ tF) (tF < t ≤ tF +tS) (tF +tS< t ≤ tT)

Figure 5.1.1: Schematic of plate-and-frame filter press cycle: (a) Load, (b) Fill, (c) Squeeze, (d)
Unload

The first stage consists of loading the cavities with sludge at φ0 (v/v) (see
Figure 5.1.1(a)). Typical pumps are positive displacement types such as plunger or
progressive cavity. For flocculated materials, low shear pumps are imperative. The
loading time, tL, is dependent on the press volume, Vpress, and the pumping flowrate at
atmospheric pressure, qL:

V press
tL = …(5.1.1)
qL

It is assumed that there is no cake build-up or fluid expressed during the


loading stage since the path of least resistance is for the air to escape through the
membrane and the suspension to fill the cavity rather than for the fluid to pass through
the membrane. A small fluid pressure exists for vertical pressures as this stage
proceeds, but it is insignificant compared to the mechanical applied pressure.

Filtration begins once all the cavities are filled with sludge (see Figure
5.1.1(b)). Sludge is fed to the cavities through central inlet ports and filtrate flows out
through peripheral ports as the pumps apply the pressure. tF is the time in which the
pumps are in use, and referred to as the fill stage or fixed-cavity filtration here, but is
also sometimes called the expression stage. The applied pressure consists of an initial

198
ramping pressure phase followed by constant operating pressure, ∆PF. Therefore, the
fill consists of the loading, variable pressure (tP) and constant pressure (tCP) phases:

t F = t L + t P + tCP …(5.1.2)

The ramping pressure is often modelled using constant flowrate until ∆PF is
reached, which simplifies the equations considerably. Figure 5.1.2 shows the on-site
results from Hodder WTP, which illustrates that the flowrate is constant during
loading and at low pressure, but begins to fall before operating pressure is reached.
Fundamentally, the applied pressure of a positive displacement pump is a function of
flowrate, but this leaves the applied pressure undefined for high flowrates (for
example, a feed flowrate of 2.9 L/s could mean a pressure anywhere from 0 to 7 bar).
A compromise is made here, and the pressure is defined as a function of time.

tL tL + tP
12 3.6
P(t)
Applied Pressure, ∆P (bar)

10 q(t) 3.2

8 2.8 Feed Flowrate, q (L/s)

6 2.4

4 2.0

2 1.6

0 1.2
0 1000 2000 3000 4000 5000
Time, t (s)

Figure 5.1.2: Applied pressure and feed flowrate with time for Hodder WTP, Press #1, 6/11/01

The third stage of the plate-and-frame batch cycle is only seen for flexible
membrane-type presses, which are capable of squeezing the membranes together
using air or water pressure at ∆PS (see Figure 5.1.1(c)). The operating pressure is
reached quickly and is kept constant for a given time, tS, or until a minimum flowrate

199
is reached. Although generally not used by the water and wastewater industries, some
presses also incorporate air-drying and washing stages, but these are not considered
further here.

At the end of the batch cycle, the press is opened and the filter cake discharged
(see Figure 5.1.1(d)). Cake release from vertical presses is dependent upon many
factors, including the applied pressure, the cake weight and dryness, and the
membrane material and condition. The handling time, tH may be constant or variable
in an operating cycle, depending on individual sites. For example, tH is constant if the
press cycle is automatic or there is always an operator available to start a cycle as
soon as the previous cycle finishes (that is, shift and weekend work are available).

Thus, tT consists of tF, tS and tH:

tT = t F + t S + t H …(5.1.3)

The operation of plate-and-frame filter presses depends upon the size, shape
and pressure of the press and the material characteristics of the sludge to be processed.
By using the dewatering theory of Buscall and White, 1987, to describe the volume
fraction distribution during the fill and squeeze stages, the throughput and final cake
solids can be predicted. Two models have been developed and are described here.
The first was based on the linear approximation stepped-pressure filtration model
introduced in Section 2.4.2, while the second model used numerical methods to solve
the full governing equations and included the effect of membrane resistance.

One-Dimensional Filtration

In order to model the filtration behaviour, the consolidation is simplified to


one-dimension, z, with the origin at the membrane (see Figure 5.1.3). Since the cavity
is symmetrical, the mid-plane of the cavity at h is taken as the boundary of the
problem. This assumes that there is negligible gravitational effects and that the feed
source is at a plane at z = h rather than a point at the central inlet port.

200
h0 h(t)

∆PS
V(t)
V(t) φ = φ0 V(t)
∆PF
∆PS
z z
0 0

(a) (b)

Figure 5.1.3: Schematic of one-dimensional pressure filtration; (a) Fixed-cavity filtration, and (b)
Flexible-membrane filtration

Vpress is approximately h multiplied by the total membrane area of the press,


Apress:

V press ≈ h(t )A press …(5.1.4)

The error in the approximation arises from the size of the baffles used to keep
the membranes apart and the feed and filtrate ports. The average specific throughput
of sludge, <Q> (m/s), for a batch-operated filter-press is given by:

V (t ) + h(t )
Q = …(5.1.5)
tT

where V(t) is the specific volume of filtrate. <Q> is an averaged quantity for the cycle
with a maximum that depends upon the behaviour of V and h, which vary depending
on the stage of the filtration cycle. <Q> is not the same as the specific throughput per
run, Qrun (m/run):

Qrun = V (t ) + h(t ) …(5.1.6)

201
<Q> may have a maximum value whereas Qrun may not. The maximum Qrun
is when the most sludge has been put into the press (that is, the filtrate flowrate falls
to zero), whereas the optimum operation is when <Q> is maximised.

During fixed-cavity filtration (0 < t < tF), h is constant at h0 and the flowrate of
sludge fed to the press is equal to the flowrate of filtrate. A volumetric balance gives
the average volume fraction, <φ>:

h0
φ h0 = ∫ φdz = φ0 (V (t ) + h0 ); t L < t < t F …(5.1.7)
0

During flexible membrane filtration (tF < t < tF + tS), the cavity width is a
function of time and directly related to V(t):

h(t ) = VF + h0 − V (t ); t F < t < t F + t S …(5.1.8)

where VF is the specific volume of filtrate at the end of the fill stage. <φ> during the
squeeze phase is given by:

h(t )
φ h(t ) = ∫ φdz = φ0 (VF + h0 ); t F < t < t F + t S …(5.1.9)
0

202
5.1.2 Analytical Models

The linear approximation solution to the filtration equations by Landman and


White, 1997 (see Section 2.4.2), has been used as the basis for an analytical model of
plate-and-frame filtration. As such, the models are only applicable for φ0 < φ* and
assume that the membrane resistance is negligible.

Fixed-Cavity Filtration

In order to use the linear approximation to model the fill stage, the initial
height, h0, is assumed to be a very large number. Thus, the model is valid up until the
cake reaches the central plane of the cavity and is therefore only valid for cake
formation. The stepped-pressure approximation developed by Scales, et al., 2001 and
outlined in Section 2.4.2 is employed in order to model the variable pressure at the
start of the filtration process.

12
Observed pressure rise
Applied Pressure, ∆P (bar)

10 Stepped pressure for model

0
0 1000 2000 3000 4000 5000
Time, t (s)

Figure 5.1.4: Observed pressure rise and stepped pressure model input for Hodder WTP, Press
#1, 6/11/01

203
Figure 5.1.4 shows an example of the discretisation of the applied pressure as
the input to the analytical model. At the onset of each new pressure, the time taken
for the existing cake from the previous pressure to rearrange to a solids distribution
commensurate with the new applied pressure is determined. The average cake solids
of the existing cake is used as the initial solids, and the rearrangement time considered
as the time to reach the end of cake formation. This assumes that no material is added
to the cake for the short rearrangement. After this time, the cake formation proceeds
at the new pressure, with allowances made to the calculations for the rearranged cake.

Flexible-Membrane Filtration

The squeeze phase is modelled by assuming that the application of ∆PS is


instantaneous. The linear approximation solutions for both cake formation and cake
compression are used. Allowance is made for the cake that has built up during the fill
stage using the stepped-pressure method, such that the cake is rearranged to a
concentration distribution commensurate to the squeeze pressure rather than the fill
pressure. The method involves calculating the height of the cake at the end of the
expression phase, assuming that this cake has a constant solids distribution, and then
applying the filtration equations at the new pressure. Since the time required for this
is likely to be only a small portion of the overall filtration time, it is unlikely that these
assumptions will introduce significant errors. Once the cake has rearranged itself to
the appropriate solids distribution, the process continues as a one-dimensional
constant pressure filtration run. If the two cakes building up on either side of the
cavity meet, the process enters the cake compression regime.

204
5.1.3 Numerical Models

The analytical models:

- Are invalid if the filter cake reaches the mid-plane of the filter cavity during
the fill-stage;

- Exclude the use of complex material properties as seen for sewage sludges;
and

- Require the membrane resistance to be insignificant compared to the cake


resistance.

This section contains the details for numerical models of plate-and-frame filter
presses. The models are based on the formulation for piston-driven filtration with
membrane resistance given in Section 2.5.1. Numerical algorithms for fixed-cavity
and flexible-membrane filtration using the Runge-Kutta shooting method are outlined.
This numerical modelling represents several significant developments beyond the
capability of the analytical models, including compression during fixed-cavity
filtration, the use of complex material properties and the inclusion of membrane
resistance.

Since the filtration height is constant during fixed-cavity filtration, q(t) in


equation 2.2.13 is not given by dh/dt but by dV/dt. Thus, the relevant governing
equation (analogous to equation 2.3.4) is given by:

∂φ ∂  ∂φ dV 
=
∂t ∂z  D (φ ) ∂z + φ dt  …(5.1.10)
 

The concentration and concentration gradient at the membrane throughout the


filtration process are analogous to equations 2.3.8 and 2.3.9:

Py [φ (0 ,t )] = ∆P − Rm
dV
…(5.1.11)
dt

205
∂φ φ (0 , t ) dV
=− …(5.1.12)
∂z z =0 D[φ (0 , t )] dt

While the governing equation is the same for the fill and squeeze stages, the
scalings, initial conditions and boundary conditions at the central plane of the cavity
differ. Models for fixed-cavity and flexible-membrane filtration are now formulated
separately. As with the analytical models, a time-dependent pressure rise is
incorporated into the fill-stage.

Fixed-Cavity Filtration

During fixed-cavity filtration, sludge is fed to the filter press by positive


displacement pumps, where ∆PF is the operating pressure and tF is the total time
during the fill-stage. An initial time-dependent pressure rise is incorporated into the
fill-stage, where tP is the time taken to reach ∆PF. A linear function is used here:

 0 0 ≤ t ≤ tL 
 
 
t − tL
∆P (t ) =  ∆PF t L ≤ t ≤ t L + t p  …(5.1.13)
tp
 
 
 ∆PF t L + t p ≤ t ≤ t F 

Fundamentally, the pressure exerted by positive displacement pumps is a


function of the flowrate. Therefore, tP depends on the solution of the filtration
equation. However, since the flowrate is the iteration variable in the solution scheme
outlined below and is independent of pressure at low pressures (see Figure 5.1.2), ∆PF
is given as a function of time in order to ensure convergence of the algorithm.

The compressive limit of the sludge during the fill-stage, φ∞,F, is given by the
applied pressure:

Py (φ ∞ ,F ) = ∆PF …(5.1.14)

As t → ∞, φ(z,t) → φ∞,F, such that the equilibrium state is independent of the


membrane resistance, which affects only the initial rate of dewatering.

206
The initial conditions for the fill-stage are given by the assumptions that the
cavities have previously been loaded with sludge at a constant concentration, φ0, and
that no filtrate is exuded during this loading phase:

φ (z , t L ) = φ 0
…(5.1.15)
V (t L ) = 0

The overall global conservation of solids volume (equation 5.1.7) relates V(t)
to the volume fraction distribution, φ(z,t). As t → ∞, V(t) approaches its equilibrium
value, V∞,F:

 φ ∞ ,F 
V∞ ,F = h0  − 1 …(5.1.16)
 φ0 

The problem is simplified by scaling the parameters z, V(t), t and ∆P(t) to Z,


γ(T), T and Σ(T):

z
Z=
h0
V (t )
γ (T ) =
h0

T=
(
D φ ∞, F  φ∞, F

) 
2
 t …(5.1.17)
 
h02  φ0 
 0 0 ≤ T ≤ TL 
∆P (t ) T − TL 

Σ(T ) = = TL ≤ T ≤ TL + TP 
∆PF  T p 
 1 TL + TP ≤ T ≤ TF 

where TL, TP and TF are the scaled times for the loading, pressure rise and total fill-
stage times respectively. Using these scalings, equation 5.1.10 is restated as:

∂φ ∂  ∂φ dγ 
=  ∆ F (φ ) +φ …(5.1.18)
∂T ∂Z  ∂Z dT 

where the scaled diffusivity, ∆F(φ), is defined as:

207
φ 02
∆ F (φ ) = D(φ ) …(5.1.19)
φ ∞2 ,F D(φ ∞ ,F )

The scalings are also applied to the initial and boundary conditions. The
initial conditions become:

φ (Z , T L ) = φ 0
…(5.1.20)
γ (T L ) = 0

An appropriate scaled membrane resistance, βm,F, is:

D(φ ∞ ,F )  φ ∞ ,F
2

β m ,F =   R m …(5.1.21)
h0 ∆PF  φ 0 

Applying the scalings to equation 5.1.11 gives:

Py [φ (0 ,T )] dγ
= Σ (T ) − β m ,F …(5.1.22)
∆PF dT

When βm,F = 0, equation 5.1.22 reduces to:

Py [φ (0 ,T )]
= Σ (T ) …(5.1.23)
∆PF

such that φ(0,T) is at the equilibrium concentration for the applied pressure. The
scaled concentration gradient at the membrane is derived from equation 5.1.12:

∂φ φ (0 ,T ) dγ
=− …(5.1.24)
∂Z Z =0 ∆ F [φ (0 ,T )] dT

Equations 5.1.22 and 5.1.24 show that both φ(0,T) and ∂φ/∂Z at Z = 0 are
dependent upon dγ/dT, which is determined from the governing equation, indicating
an iterative solution is needed.

The scaled global conservation equation is derived from equation 5.1.9:

1
∫ φ (Z , T )dZ = φ0 (γ (T ) + 1); TL ≤ T ≤ TF …(5.1.25)
0

208
An internal boundary at Zc(T) may exist at the boundary between the sediment
and consolidating cake. Prior to the cake reaching the mid-plane of the cavity at Tc,
the concentration above Zc(T) is constant at φ0. Therefore, equation 5.1.25 becomes:

Z c (T )
∫ φ (Z , T )dZ = φ0 (γ (T ) + Z c (T )); TL ≤ T ≤ TF …(5.1.26)
0

Equation 5.1.26 is equivalent to equation 2.5.9, illustrating that piston-driven


filtration and fixed-cavity filtration are identical until the growing cake either hits the
piston or the mid-plane of the cavity. The concentration at Zc(T), φc, is constant for Zc
< 1, and depends on whether the feed suspension is networked or un-networked, that
is, φ0 > φg or φ0 < φg, where φg is the gel point concentration. φc = φ0 and is
continuous for the networked case while a discontinuity exists for the un-networked
case (the concentration jumps from φ(Zc+,T) = φ0 to φ(Zc-,T) = φg). These two cases
are now investigated separately.

Feed Suspension Un-Networked (φ0 < φg)

For the φ0 < φg case, the concentration gradient at the top of the bed is given
by equation 5.1.27:

∂φ φ g − φ 0  dZ c dγ 
=−  + 
∂Z Z −
c
∆ F φ g( ) dT dT 
…(5.1.27)

∆F(φg) may either be equal to or greater than 0, depending on the functional


form of Py(φ), therefore ∂φ/∂Z → -∞ or a large negative number as φ → φg. Equation
5.1.27 is redundant in the algorithm for Rm ≠ 0 presented here, since the conservation
of volume (equation 5.1.26) is used to give a test value for the iteration variable.
However, it is necessary for the small-time solution and the exact solution for the case
when Rm = 0 and tP = 0.

Small-Time Approximation

An approximation for the small-time behaviour of the φ0 < φg solution is


determined following the formulation in Section 2.5.1 using a similarity variable,
ξ(Z,T), and series expansions of Zc(T) and φ(Z,T). The results show that the flow rate

209
is constant and the membrane volume fraction and the piston and cake heights vary
linearly with time for small times:

dγ Σ(T L )
= …(5.1.28)
dT β m, F

φ0 Σ(TL )
Z c (T ) =
(φ g − φ 0 ) β m,F (T − TL ) …(5.1.29)

−1
Σ(TL )2  1 
φ (0, T ) = φ g + (T − TL )
 1 
− …(5.1.30)
( )
β m2 , F ∆ F φ g  φ 0 φ g 

This analysis assumes that ∆F(φg) ≠ 0 and βm,F ≠ 0, and is trivial when Σ(0) =
0. The small-time approximation is used to give the size of the initial time-step in the
solution algorithm by giving φ(0,T) a small value such as 2φg and Σ(0) = 1. Along
with the magnitude of βm.F, the small-time solution also gives an indication of the
significance of the membrane resistance, since φ(0,T) << φ∞. Thus, from equation
5.1.30:

β m2 , F ∆ F (φ g )  1 
T − TL << (φ ∞, F − φ g ) φ1 −
φ g 
…(5.1.31)
Σ(T L )2  0

If the right-hand side of the inequality is very small, then the volume fraction
at the membrane increases quickly and the membrane resistance is insignificant. If it
is large, then the membrane resistance dominates the filtration process such that dγ/dT
remains constant.

Similarity Solution for φ0 < φg, Rm = 0, tP = 0, Zc(T) ≤ 1

Analogous to piston-driven filtration, an exact similarity solution exists for


cake formation for the special case where the feed suspension is un-networked, the
resistance is negligible and the operating pressure is applied instantaneously. In this
case, the volume fraction at the membrane jumps immediately to the compressional
limit, φ∞,F:

210
φ (0 ,T ) = φ∞ ,F …(5.1.32)

A similarity variable, Φ(X), is substituted for φ(Z,T):

 Z 
φ (Z ,T ) = Φ   = Φ ( X ) …(5.1.33)
 Z c (T ) 

For this case (TL ≤ T ≤ TL + Tc), Zc(T) and γ(T) are exactly functions of the
square-root of time, such that:

Z c (T ) = β T − T L
dγ α …(5.1.34)
=
dT Z c (T )

α and β are constants of proportionality to be determined by the solution of the


governing equation. Substituting Φ(X), Zc(T) and γ(T) into the scaled governing
equation (equation 5.1.18) gives a second-order ordinary differential equation for
Φ(X):

d  dΦ   β 2  dΦ

 F (Φ ) + α + X =0 …(5.1.35)
dX  dX   2  dX

The relevant boundary conditions are:

dΦ φ∞ ,F
Φ (0 ) = φ∞ ,F ; =− α
dX 0 ∆F (φ∞ ,F )

Φ (1) = φ g ;

=−
(
φ g − φ0 
 α+
) β 2 
…(5.1.36)

dX 1 ( )
∆ φ g  2 

The problem is simplified to two coupled first-order equations by making a


change of variables to η and substituting G(η):

dΦ G (η )
=
dη ∆ F (Φ )
…(5.1.37)
dG  η  G (η )
=  1 − 
dη  2 B 2  ∆ F (Φ )

211
where η, A and B are:

η = A(1 − X )
β2
A =α + …(5.1.38)
2
α β
B= +
β 2

The boundary conditions become:

Φ (0 ) = φ g ; G (0 ) = φ g − φ 0
 A  …(5.1.39)
Φ ( A) = φ ∞ ,F ; G ( A) = φ ∞ ,F  1 − 
 2B 2 

∆F(Φ) may be described by many different functional forms, therefore, to


remain general, a numerical technique is used to solve equations 5.1.37 and 5.1.39 for
A and B. An iterative Runge-Kutta shooting technique, described later in this section,
is used, giving the behaviour of Zc(T) and dγ/dT from TL until Tc = (B/A)2:

Z c (T ) =
A
T − TL
B
…(5.1.40)
dγ  A  1
= B− 
dT  2B  T − TL

This solution dramatically decreases the solution time for the φ0 < φg, Rm = 0
case and gives a maximum time step value for the φ0 < φg, Rm ≠ 0 case.

Feed Suspension Networked (φ0 > φg)

For the Rm ≠ 0, φ0 > φg case for dead-end filtration, there is in fact no region
where φ remains at φ0 such that Zc(T) → 1 for T > 0 (Landman, et al., 1991). The
implication for the numerical solution of fixed cavity filtration is that ∂φ/∂Z → 0 as Z
→ 1, and that, when iterating on dγ/dT, the exact solution is an upper bound of
possible solutions (since ∂φ/∂Z ≤ 0).

212
Small-Time Approximation

The small-time approximation for the φ0 > φg case shows that the flow rate is
constant and the membrane volume fraction varies with the square-root of time for
small times:

dγ  P (φ ) 
=
1  Σ(TL ) − y 0  …(5.1.41)
dT β m, F  ∆PF 

 2Σ(T L )  P (φ )  
φ (0, T ) = φ 0 1 + 1 − y 0  T − T L  …(5.1.42)
 β m, F π∆ F (φ 0 )  ∆PF  

Setting Σ(TL) = 1 and φ(0,T) << φ∞ gives:

πβ m2 , F ∆ F (φ 0 )  φ ∞, F

2
 P (φ ) 
−2
T − T L <<  − 1  1 − y 0  …(5.1.43)
 φ   ∆PF 
4  0  

The small time approximation is used to give the initial time step in the
algorithm detailed later.

Flexible Membrane Filtration

During the squeeze stage of a plate-and-frame filter press cycle (tF < t ≤ tF +
tS), sludge is no longer fed to the press, and the membranes are pushed together using
air or water pressure at a constant applied pressure, ∆PS. The cavity is still assumed
to be symmetrical, but the width varies with time (d(t) = 2 h(t)). The initial conditions
for the squeeze stage are given by the volume fraction distribution and filtrate volume
at the end of the fill stage (φ(z,tF) and VF respectively):

V (t F ) = V F
h(t F ) = h0 …(5.1.44)
h0
∫ φ ( z , t F )dz = φ 0 (V F + h0 ) = φ F h0
0

where φF is the average volume fraction at tF. Since the initial condition for the
squeeze stage is dependent upon the fill stage, the small-time solution is unknown.

213
There is no more sludge fed to the press, thus the overall conservation of
solids volume is given by a constant (see equation 5.1.9). The specific volume of
filtrate is directly related to the cavity width through the conservation of volume (see
equation 5.1.8).

It is assumed that the pressure is applied instantaneously. The equilibrium


volume fraction during the squeeze phase, φ∞,S is given by equating Py(φ) to ∆PS:

Py (φ ∞ ,S ) = ∆PS …(5.1.45)

As t approaches infinity, the filtrate volume reaches its equilibrium value for
the squeeze phase, V∞,S:

 φ  VF 
V∞ ,S = h0 1 − 0  + 1  …(5.1.46)
 φ ∞ ,S  h0 

The scalings for the squeeze phase are:

z
Z=
h0
V (t )
γ (T ) =
h0
( ) 2 …(5.1.47)
D φ ∞, S  φ ∞, S 
T=   t ; t ≥ t F
h02  φ F 
h(t )
H (T ) =
h0

z is scaled linearly with h0 during the squeeze phase as in the fixed-cavity


formulation. Rather than subsequently changing to the material coordinate, w, as used
in the Rm = 0 case for piston-driven filtration, Z is kept as the spatial coordinate.
Substituting equation 5.2.43 into equation 5.2.3 gives the scaled governing equation
during the squeeze phase:

∂φ ∂  ∂φ dγ 
=
∂T ∂Z ∆ S (φ ) ∂Z + φ dT  …(5.1.48)
 

where

214
φ F2
∆ S (φ ) = D(φ ) …(5.1.49)
φ ∞2 ,S D(φ ∞ ,S )

The scaled conditions at the start of the squeeze-stage are:

( )
γ TF* = γ F
H (TF* ) = 1 …(5.1.50)

∫ φ (Z , TF )dZ = φ0 (γ F + 1) = φ F
1
*
0

where TF* is tF using the squeeze-stage scalings (in contrast to TF, which is tF using
the fill-stage scalings). The scaled membrane resistance during the squeeze phase,
βm,S, is:

D(φ ∞ ,S )  φ ∞ ,S
2

β m ,S =   R m …(5.1.51)
h0 ∆PS  φ F 

Substituting equation 5.1.51 into equation 5.1.11 gives:

Py [φ (0 ,T )] dγ
= 1 − β m ,S …(5.1.52)
∆PS dT

The concentration gradient at the membrane (equation 5.1.12) becomes:

∂φ φ (0 ,T ) dγ
=− …(5.1.53)
∂Z Z =0 ∆ S [φ (0 ,T )] dT

The scaled conservation during the squeeze phase is:

H (T )
∫ φ (Z ,T )dZ = φ F …(5.1.54)
0

Allowing for constant concentration above the cake shows that equation 5.1.26
holds for both the fill and squeeze stages. This shows that the cake formation is the
same for both stages, and that it is the cake compression that is affected by the fixed-
cavity or flexible-membrane conditions. The filtrate volume during the squeeze stage

215
is given by the solution of equation 5.1.48 subject to equations 5.1.50 and 5.1.52 to
5.1.54. A Runge-Kutta numerical technique is outlined below.

Runge-Kutta Numerical Algorithms

Algorithms have been written in Mathematica® that are capable of numerically


solving the governing equations for both the fill and squeeze stages, subject to the
appropriate boundary conditions, for networked and un-networked feed suspensions
and with or without membrane resistance. To ensure convergence for both the
networked and un-networked cases, ∂φ/∂Z must be allowed to vary from 0 to very
large negative numbers, but not become positive or infinite. Also, the use of a
numerical technique for the Rm = 0 case is necessary in order to model sewage
sludges, which exhibit non-traditional behaviour, and to predict compression at the
end of the fill-stage.

In order to apply the Runge-Kutta numerical method, the governing equation


(either equation 5.1.18 or 5.1.48) is simplified to two coupled ordinary differential
equations by introducing the scaled solids flux, ψ(Z) and making a backward
difference approximation in time. dψ/dZ is given by equation 2.5.39, while the
concentration gradient is given by:

dφ 1  dγ 
= ψ −φ  …(5.1.55)
dZ ∆ (φ )  dT 

Q(Z) (see equation 2.5.40) gives the cumulative conservation of solids volume.
dγ/dT is used as the iteration variable for both the fill and squeeze stages. An estimate
of dγ/dT, dγ*/dT, is iterated upon using an interval halving method. For a known time
step, ∆T, an estimate of γ, γ*, is given by a first-order approximation:

dγ *
γ * = ∆T +γ < …(5.1.56)
dT

where γ< is the value of γ at the previous time step. Since dγ/dT decreases, the first-
order approximation underestimates γ(T). Higher order approximations, which may
be more accurate, are not used because they tend to overestimate γ(T), such that no
solution exists for the φ0 > φg case.

216
Fixed-Cavity Filtration Algorithm

A flowchart of the numerical algorithm for fixed cavity filtration with


membrane resistance is shown in Figure 5.1.5. The inputs to the model are the
material characteristics (Py(φ), R(φ), D(φ) and φg) and the operating conditions (d, φ0,
Rm, ∆P(t), tL, tP and tF). After evaluating φ∞,F and V∞,F, the scalings are applied and
the conditions at TL are set.

The initial time step value is determined by giving the small-time solutions for
φ(0,T) (equations 5.1.30 and 5.1.42) small values such as 2φc and setting Σ(0) = 1:

  1 1  
( )
 β m2 ,F ∆ F φ g φ g  −
 φ0 φ g 
φ0 < φ g 
   
∆T =  2  …(5.1.57)
 πβ m ,F ∆ F (φ 0 ) 
−2
Py (φ 0 )  
 φ0 ≥ φ g 
 4  1 − ∆P 
  F  

Since the changes in volume fraction are large at small times such that ∆T is
initially very small, ∆T is allowed to increase at each time step until the maximum
value is reached. However, ∆T cannot be too high (otherwise the accuracy of the
approximations for the time derivatives is reduced) or too small (otherwise the
required computational time is unreasonable). The maximum ∆T for the φ0 < φg case
is given by the exact solution for the Rm = 0, tp = 0 case. A reasonable maximum ∆T
for the φ0 > φg case is 0.1.

For a given time step, ∆T, there is one unknown, dγ*/dT, which is solved using
an interval halving technique. γ* is then given by a first-order approximation (see
equation 2.5.42). The initial upper bound, dγ/dThigh, is given by the small-time
solution (either equation 5.1.28 or 5.1.41), while the initial lower bound, dγ/dTlow, is
zero. However, for the variable initial pressure case (where Σ(TL) ≠ 1), dγ/dT
increases initially, and dγ/dThigh is given by the inverse of βm,F.

217
1-D Constant Pressure Fixed-Cavity Filtration with Membrane Resistance
Iterative 4th – 5th order Runge-Kutta Algorithm

Material Characteristics:
Compressive yield stress, Py(φ); Hindered settling function, R(φ); Solids diffusivity, D(φ), Gel point, φg
Operating Conditions:
Cavity width, d = 2 h0; Feed concentration, φ0; Membrane resistance, Rm; Fill time, tF, Loading time, tL
Applied pressure, ∆P(t), including operating fill pressure, ∆PF and time to reach ∆PF, tP

 φ ∞ ,F
[ ]
φ c = Max φ0 ,φ g ; Py (φ ∞ , F ) = ∆PF ; V∞ , F = h0 

− 1
 φ0 

Scalings:
D (φ ∞ ,F )  φ ∞ ,F V (t ) ∆P(t ) D (φ ∞ , F )  φ ∞ , F
2 2
z  φ02 
Z= ;T =   t ; γ (T ) = ; Σ (T ) = ; ∆ F (φ ) = D (φ ); β m , F =   Rm
h0 h02  φ 0  h0 ∆PF φ ∞ ,F D (φ ∞ , F )
2 h0 ∆PF  φ0 

P (φ ) 
Initial Conditions:
dγ 1 
T = TL ; φ (Z , TL ) = φ 0 ; Z c (TL ) = 0; γ (TL ) = 0; =  Σ(TL ) − y c 

dT T β m, F  ∆PF 
L

Initial Time-Step Size and Iteration Bounds:


−2
 1  πβ 2 ∆ (φ )  P (φ ) 
If φ0 < φ g , ∆T = β m
2
( ) 
,F φ g ∆F φ g  −
1 ; If φ 0 ≥ φ g , ∆T = m , F F 0  1 − y 0 
  ∆PF 
;
dγ *
=
1
;
dγ *
=0
 φ0 φ g  4  dT high β m , F dT low

Fixed time step, T = T< + ∆T

dγ * 1  dγ * dγ *  * P [φ (0 )]
 ; γ = ∆T d γ + γ < ; y dγ *
*
= + = Σ (T ) − β m , F ; ψ (0 ) = 0 ; Q(0 ) = 0
dT 2  dT high dT low  dT ∆PF dT

Runge-Kutta Algorithm:
φ − φ < 
dφ 1  dγ *  dψ  Z ≤ Z c<  dQ
Solve = ψ −φ , =  ∆T  and =φ
dZ ∆ F (φ )   dZ  φ − φ c
Z > Z c< 
dT  dZ
 ∆T 
from Z = 0 until Z = 1, φ = φ c ,

dZ
> 0 or Q = φ0 γ * + 1 ( )
Q (Z )
γ test = −Z
φ0

∂φ dγ * dγ * dγ * dγ *
If > 0 or γ test > γ * , = ; If γ test < γ * , =
∂Z dT low dT dT high dT

dγ dγ *
= ; γ (T ) = γ * ; Z c (T ) = Z ; φ = φ0 (γ (T ) + 1)
dT T dT

If T < TF, proceed to next step of ∆T; If ∆T < ∆Tmax, ∆T = 2∆T


dγ * 1 dγ *
= ; =0
dT high β m , F dT low

Figure 5.1.5: Runge-Kutta numerical algorithm for fixed-cavity filtration with membrane
resistance

218
Equations 5.1.55, 2.5.39 and 2.5.40 are solved using 4th order Runge-Kutta
steps of ∆Z from Z = 0 (where φ(0,T) is given by equation 5.1.22 and ψ(0,T) = Q(0,T)
= 0) until φ = φc, Z = 1, dφ/dZ > 0 or Q = φ0(γ*+1). The accuracy of each step of ∆Z is
kept in check using the truncation error from the 5th order Runge-Kutta step. The
final condition arises from equation 5.1.25.

If dφ/dZ > 0, dγ*/dT is too low and is the lower bound for the next iteration. If
φ = φc, Z =1 or Q = φ0(γ*+1), a test value for γ(T), γtest, is given by Q(Z) (from
equation 5.1.26):

Q (Z )
γ test = −Z …(5.1.58)
φ0

If γtest > γ*, dγ*/dT is the lower bound for the next iteration. Conversely, if γtest
< γ*, dγ*/dT is the upper bound for the next iteration. dγ*/dT is iterated upon until γtest
= γ* to within the desired accuracy. The algorithm then proceeds to the next time step,
until T = TF.

As discussed in Section 2.5.2, a change of variables to X = Z/Zc(T) should be


made in the φ0 < φg case due to the discontinuity at Zc(T). Instead, equation 2.5.44 is
used to give the solids flux gradient for Zc(T<) < Z < Zc(T).

Depending on the functional form used to describe Py(φ), ∆(φg) may be equal
to 0 and a singularity may exist as φ → φg for the un-networked case. In such cases, it
is necessary to assume a finite positive value for ∆(φg) (such as ∆(1.0001 φg)) and to
ensure that φ ≥ φg and dφ/dZ ≤ 0 within the Runge-Kutta algorithm by reducing ∆Z.

Algorithm for Exact Similarity Solution

The exact solution for the φ0 < φg, Rm = 0, Zc(T) ≤ 1 case is given by solving
the coupled equation 5.1.37 for A and B using an iterative Runge-Kutta shooting
technique. This also gives a maximum time step value for the φ0 < φg, Rm ≠ 0 case.
For an estimate of B*, equation 5.1.37 is solved from η = 0 until Φ(η*) = φ∞,F. A test
value for B*, Btest, is given by the fourth boundary condition, G(η*):

219
η* φ ∞ ,F
Btest = …(5.1.59)
2 φ ∞ ,F − G (η * )

B* is iterated upon using an interval halving method until B* = Btest to within


the desired accuracy. Thus B = B* and A = η*, giving Zc(T) and dγ/dT up until Tc.
Beyond this time, the full-numerical solution is required.

Flexible-Membrane Filtration Algorithm

A flow chart of the flexible-membrane algorithm is shown in Figure 5.1.6.


There are small but important differences in the squeeze algorithm compared to the
fill algorithm. The initial conditions for the squeeze stage are the conditions at t = tF.
φ∞,S and V∞,S are calculated and t, D(φ) and Rm are re-scaled using the squeeze
pressure. Since the small-time solution is unknown, ∆T is set at a small number, such
as 0.01. Equations 5.1.55, 2.5.39 and 2.5.40 are solved from Z = 0 (where φ(0,T) is
given by equation 5.1.52 and ψ(0,T) = Q(0,T) = 0) until φ = φc, dφ/dZ > 0 or Q = φF.
The final condition arises from equation 5.1.54. As in the fill-stage algorithm, if
dφ/dZ > 0, dγ*/dT is too low and is the lower bound for the next iteration. If φ = φc or
Q = φF, γtest is given by equation 5.1.58. If γtest > γ*, dγ*/dT is the lower bound for the
next iteration. Conversely, if γtest < γ*, dγ*/dT is the upper bound for the next iteration.
dγ*/dT is iterated upon until γtest = γ* to within the desired accuracy, giving γ(T) and
H(T). The algorithm then proceeds to the next time step, until T ≥ TF* + TS.

220
1-D Constant Pressure Flexible-Membrane Filtration with Membrane Resistance
Iterative 4th – 5th order Runge-Kutta Algorithm

Material Characteristics:
Compressive yield stress, Py(φ); Hindered settling function, R(φ); Solids diffusivity, D(φ); Gel point, φg
Operating Conditions:
Initial cavity width, d = 2 h0; Feed concentration, φ0; Squeeze pressure, ∆PS; Membrane resistance, Rm; Squeeze time, tS

Conditions from Fixed-Cavity Filtration at tF:


h0
t = tF ; ∫ φ (z ,t F ) = h0φ F ; z c (t F ) = z c ,F ; V ( t F ) = V F
0

 
[ ]
φ c = Max φ0 ,φ g ; Py (φ ∞ ,S ) = ∆PS ; V∞ ,S = h0  1 −
φF
φ ∞ ,S


 

Scalings:
D (φ ∞ ,S )  φ ∞ ,S V (t ) h(t ) D (φ ∞ ,S )  φ ∞ , S
2 2
z  φ F2 
Z= ;T =   t ;γ (T ) = ; H (T ) = ; ∆S (φ ) = D(φ ); β m ,S =   Rm
h0 h02  φ F  h0 h 0 φ ∞ , S D (φ ∞ ,S )
2 h0 ∆PS  φ F 

[ ( )]
Initial Conditions:
D(φ ∞ ,S )  φ ∞ ,S 
( ) ( )  1 − Py φ 0 ,TF
2 *
 z V dγ
 t F ; φ Z ,TF* = φ (z ,t F ); Z c TF* = c , F ; γ ( TF* ) = F ;
1
TF* =  =
h02  φ F β  ∆PS 
 h 0 h0 dT *
TF m ,S  
dγ * dγ dγ *
∆T = 0.01; = ; =0
dT high dT T * dT low
F

Fixed time step, T = T< + ∆T

dγ * 1  dγ * dγ *  * P [φ (0 )]
; γ = ∆T dγ + γ < ; y dγ *
*
= + = 1 − β m ,S ; ψ (0 ) = 0 ; Q(0 ) = 0
dT 2  dT high dT low  dT ∆PS dT

Runge-Kutta Algorithm:
φ − φ < 
dφ 1  dγ *  dψ  Z ≤ Z c<  dQ
Solve = ψ −φ , =  ∆T  and =φ
dZ ∆S (φ ) 
 dT  dZ  φ − φ c
 <  dZ
Z > Zc
 ∆T 

from Z = 0 until φ = φ c , > 0 or Q = φ F
dZ

Q(Z )
γ test = −Z
φ0

∂φ dγ * dγ * dγ * dγ *
If > 0 or γ test > γ * , = ; If γ test < γ * , =
∂Z dT low dT dT high dT

dγ dγ * φ
= ; γ (T ) = γ * ; H (T ) = 1 − γ (T ); Z c (T ) = Z ; φ = F
dT T dT H (T )

If T < TF* + TS, proceed to next step of ∆T


dγ * dγ dγ *
= ; =0
dT high dT dT low

Figure 5.1.6: Runge-Kutta numerical algorithm for flexible-membrane filtration with membrane
resistance

221
5.1.4 Model Results

The two sets of material characteristics from Huntington 09/07/03 given in


Section 4.1.2 were used to illustrate the analytical and numerical model predictions.
The first set of results employed power-law functional forms for Py(φ), R(φ) and D(φ).
The power-law functions were simple to use and the derivative of Py(φ) exists at φg.
The second set of material characteristics used here were the output from the transient
batch settling analysis for the same sample in order to illustrate the versatility of the
numerical model.

Fixed-Cavity Results

The fixed-cavity models were used to generate predictions for the behaviour of
the fill stage of plate-and-frame filtration with the Huntington 09/07/03 material
characteristics. The operating pressure was set at 6 bar, the cavity width was 3 cm,
and the loading time was assumed to be 0. The results for un-networked (φ0 = 0.004
v/v) and networked (φ0 = 0.02 v/v) feed suspensions were determined for various
values of membrane resistance, from 0 to 1012 Pa s/m, and with or without an initial
pressure rise.

The volume fraction distribution results illustrated the solution of the


governing equations, representing the build up and subsequent consolidation of the
filter cake. A comparison of the relevant time dependent data gave an indication of
the role of the membrane resistance and ramping pressure in the scenarios presented.

Feed Suspension Un-Networked

The fixed-cavity numerical model was used to generate results with φ0 = 0.004
v/v, such that the feed suspension was un-networked. The volume fraction
10 11 12
distribution results for four values of Rm (0, 10 , 10 and 10 Pa.s/m) and tL = tP = 0
are presented in Figure 5.1.7. The discontinuity in concentration and concentration
gradient at the top of the cake is pronounced. Once the cake reached the mid-plane of
the cavity, φ(1,T) and dφ/dZ at Z = 1 were discontinuous and approached φ∞,F and 0
respectively as T → ∞. The modelling of cake compression during the fill-stage

222
represents a new application of phenomenological filtration theory and an important
advance in the modelling of the fill-stage of plate-and-frame filtration.

0.12
Rm = 0 Rm = 1010 Pa.s/m
φ∞
0.1
Volume Fraction, φ (Z ,T )

0.08

0.06

0.04

0.02
φg
0 φ0
0.12
Rm = 1011 Pa.s/m Rm = 1012 Pa.s/m φ∞
0.1
Volume Fraction, φ (Z ,T )

0.08

0.06

0.04

0.02

φg
0 φ0
0 0.2 0.4 0.6 0.8 10 0.2 0.4 0.6 0.8 1
Scaled Height, Z Scaled Height, Z

Figure 5.1.7: Volume fraction distribution results for fixed-cavity filtration with varying Rm; (φ0 =
0.004 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0)

In the approximation that the membrane resistance was negligible, φ(0,T) at


the membrane immediately jumped to φ∞,F and a high concentration cake built up
from Z = 0. For the Rm = 1010 Pa.s/m case, the volume fraction distribution was very
similar to Rm = 0 and φ(0,T) increased quickly. As the resistance was increased,
φ(0,T) increased more slowly and the growing cake was at lower concentrations. At
the highest resistance, the filter cake was at low concentrations and almost
independent of Z such that, once Zc(T) = 1, the cake behaved like a sponge.

223
10e12
2000000 10e11
40000 10e10
10e9
No Rm
30000
1600000
20000
Time, t (s)

1200000 10000

0
800000 0 0.01 0.02 0.03 0.04

400000

0
V∞,F
0 0.04 0.08 0.12 0.16
2 2 2
(Specific Volume) , V (m )

Figure 5.1.8: Model predictions of t versus V2 for fixed-cavity filtration with varying Rm; (φ0 =
0.004 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0)

The filtration predictions of t versus V2 for five values of Rm (0, 109, 1010, 1011
and 1012 Pa.s/m) and tP = 0 are shown in Figure 5.1.8. t was initially linearly
dependent upon V, corresponding to constant resistance, followed by quadratic
behaviour as the cake formed and cake resistance increased. Once the cake reached
the mid-plane of the cavity, the cake compressed and V(t) → V∞,F as t → ∞. The
results clearly show that, as the membrane resistance increased, the time taken to
reach a given filtrate volume increased. The inset shows that there was negligible
difference between 0 and 109 Pa.s/m and very little effect due to Rm at 1010 Pa.s/m.
There was considerable effect at 1011 Pa.s/m or larger, while at 1012 Pa.s/m, the
membrane resistance dominated the filtration behaviour.

A plot of the variation of dt/dV2 (as given by a first order approximation


between data points) with t is given in Figure 5.1.9. The beginning of cake
compression is illustrated by the dramatic change in the slope of t versus V2. For the
case where Rm = 0, dt/dV2 was constant throughout the cake formation process as
calculated using the similarity solution. With the introduction of the membrane

224
resistance, dt/dV2 initially decreased as the membrane resistance dominated the
process, until it reached the value for the Rm = 0 case. For the 1011 and 1012 Pa.s/m
scenarios, dt/dV2 did not reach this limit before cake compression began, further
illustrating the large effect of Rm values above 1011 Pa.s/m.

1.E+11 10e12
10e11
10e10
1.E+10
10e9
No Rm
1.E+09
dt /dV (s/m )
2

1.E+08
2

1.E+07

1.E+06

1.E+05
1.E+00 1.E+02 1.E+04 1.E+06 1.E+08

Time, t (s)

Figure 5.1.9: Model predictions of dt/dV2 versus t for fixed-cavity filtration with varying Rm; (φ0 =
0.004 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0)

A value for the cake resistance, Rc, was given by considering the flowrate as
the ratio of the applied pressure and the sum of Rm and Rc. Therefore, Rc was:

∆PF
Rc = − Rm …(5.1.60)
dV
dt

Equation 5.1.60 was applied to the model predictions for the un-networked
case to give the variation of Rc with time, as shown in Figure 5.1.10. For the Rm = 0
case, Rc jumped immediately to 109 Pa.s/m, and then increased by several orders of
magnitude. As the membrane resistance increased, Rc took longer to increase. The
time taken for Rc to equal Rm was 3.95, 390.5, 35,253 and 589,050 seconds for the
increasing values of Rm respectively. These results further illustrated that, for these

225
operating conditions and material characteristics, Rm = 109 or 1010 Pa.s/m had
negligible or small effects, Rm = 1011 Pa.s/m had a major effect and Rm = 1012 Pa.s/m
dominated the filtration behaviour.

1.E+14
No Rm
Cake Resistance, Rc (Pa.s/m)

10e9
1.E+13 10e10
10e11
10e12
1.E+12

1.E+11

1.E+10

1.E+09

1.E+08
1.E+00 1.E+02 1.E+04 1.E+06

Time, t (s)

Figure 5.1.10: Model predictions of cake resistance for fixed-cavity filtration with varying Rm; (φ0
= 0.004 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0)

Feed Suspension Networked

The fixed-cavity numerical model was used with the Huntington WTP
material characteristics to generate results for φ0 = 0.02 v/v (such that the feed
suspension was networked), ∆PF = 6 bar, h0 = 0.015 m and tP =0. The volume
fraction distribution results for four values of Rm (0, 1010, 1011 and 1012 Pa.s/m) are
presented in Figure 5.1.11. At the top of the consolidating region, the concentration
was continuous and at φ0, and the concentration gradient approached zero. The
concentration at the mid-plane of the cavity began to increase imperceptibly as soon
as the pressure was applied.

As for the un-networked case, increasing Rm decreased the cake concentrations


and the rate at which the membrane concentration increased. At the highest
resistance, the filter cake was at low concentrations and almost independent of Z.

226
0.12

φ∞
0.1
Volume Fraction, φ (Z ,T )

0.08

0.06

0.04

0.02 φ0
Rm = 0 Rm = 1010 Pa.s/m
0
0.12

φ∞
0.1
Volume Fraction, φ (Z ,T )

0.08

0.06

0.04

0.02 φ0
Rm = 1011 Pa.s/m Rm = 1012 Pa.s/m
0
0 0.2 0.4 0.6 0.8 10 0.2 0.4 0.6 0.8 1
Scaled Height, Z Scaled Height, Z

Figure 5.1.11: Volume fraction distribution results for fixed-cavity filtration with varying Rm; (φ0
= 0.02 v/v, h0 = 0.015 m, ∆PF = 6 bar, tp = 0)

The filtration predictions of t versus V2 for five values of Rm (0, 1010, 1011,
5x1011 and 1012 Pa.s/m) and tP = 0 are shown in Figure 5.1.12. t was initially linearly
dependent upon V, corresponding to constant resistance, followed by quadratic
behaviour as the resistance of the cake increased. Once φ(h0,t) began to rise
appreciably, the cake compressed and V(t) → V∞,F as t → ∞. As the membrane
resistance was increased, the time taken to reach a given filtrate volume increased.
There was little difference between 0 and 1010 Pa.s/m, while the membrane resistance
dominated the filtration behaviour at 1012 Pa.s/m.

A plot of the variation of dt/dV2 with t is given in Figure 5.1.13. For the case
where Rm = 0, dt/dV2 was virtually constant until φ(h0,t) began to increase and the
process entered cake compression.

227
240000
10e12
5x10e11
200000 10e11
10e10
No Rm
160000
Time, t (s)

120000

80000

40000

0
0 0.001 0.002 0.003 0.004 0.005
2 2 2
(Specific Volume) , V (m )

Figure 5.1.12: Model predictions of t versus V2 for fixed-cavity filtration with varying Rm; (φ0 =
0.02 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0)

1.E+10
No Rm
10e10
10e11
1.E+09 5x10e11
10e12
dt /dV (s/m )
2

1.E+08
2

1.E+07

1.E+06
1.E+00 1.E+02 1.E+04 1.E+06

Time, t (s)

Figure 5.1.13: Model predictions of dt/dV2 versus t for fixed-cavity filtration with varying Rm; (φ0
= 0.02 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0)

228
With the introduction of Rm, dt/dV2 initially decreased as the membrane
resistance dominated the process, until it reached the value for the Rm = 0 case. For
the 1011 Pa.s/m or greater scenarios, dt/dV2 did not reach this limit before cake
compression began.

Equation 5.1.60 was applied to the model predictions for the networked case
to calculate Rc. The results are shown in Figure 5.1.14. For the Rm = 0 case, Rc was
above 109 Pa.s/m almost immediately, and then increased by several orders of
magnitude during the filtration process. As the membrane resistance increased, Rc
took longer to increase. The time taken for Rc to equal Rm was 63, 5510, 47,313 and
97,784 seconds for the increasing values of Rm respectively. These results further
illustrate that, for these material characteristics, Rm = 1010 Pa.s/m had negligible
effect, Rm = 1011 Pa.s/m had a major effect and values above Rm = 5x1011 Pa.s/m
dominated the filtration behaviour.

1.E+13
10e12
5x10e11
Cake resistance, Rc (Pa.s/m)

10e11
10e10
No Rm
1.E+12

1.E+11

1.E+10

1.E+09
1.E+00 1.E+02 1.E+04 1.E+06

Time, t (s)

Figure 5.1.14: Model predictions of cake resistance for fixed-cavity filtration with varying Rm; (φ0
= 0.02 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0)

229
Ramping Pressure

The behaviour of the initial time-dependent ramping pressure was


investigated. The volume fraction distribution results for the un-networked and
networked cases with the operating conditions of ∆PF = 6 bar, tL = 0s, tP = 1200s, h0 =
0.015m and Rm = 0 are presented in Figure 5.1.15. Even though Rm = 0, φ(0,t) only
reached φ∞,F once the operating pressure was reached at tP. Comparing Figure 5.1.15
with the results incorporating Rm (Figure 5.1.7 and Figure 5.1.11) indicated that Rm
and ramping pressures were competing effects during the initial cake build-up.

0.12
φ0 = 0.004 v/v φ0 = 0.02 v/v
Solids Volume Fraction, φ (Z ,T )

φ∞,F φ∞,F
0.1

0.08

0.06

0.04

0.02 φ0
φg
0 φ0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Scaled Height, Z Scaled Height, Z

Figure 5.1.15: Volume fraction distribution results for fixed-cavity filtration with ramping
pressure; (h0 = 0.015 m, ∆PF = 6 bar, tP = 1200 s, Rm = 0)

The t versus V2 results for the two ramping pressure scenarios are presented in
Figure 5.1.16, along with the results for varying Rm with tP = 0s (see Figure 5.1.8 and
Figure 5.1.12). The results show the impact of a ramping pressure at the beginning of
the fill-stage – the results for Rm = 0, tP = 0s were effectively shifted by tP. For both
the networked and un-networked cases, an initial pressure rise of 1200s was roughly
equivalent to a membrane resistance of 1010 Pa.s/m, although Rm did continue to
influence the boundary condition at φ(0,T) well beyond tP.

230
φ0 = 0.004 v/v φ0 = 0.02 v/v
4000

3500

3000
Time, t (s)

2500

2000

1500
10e11, tp=0s 10e11, tp=0s
1000 10e10, tp=0s 10e10, tp=0s
10e9, tp=0s
No Rm, tp=0s
500 No Rm, tp=0s
No Rm, tp=1200s No Rm, tp=1200s
0
0 0.001 0.002 0.003 0.004 0 0.0002 0.0004 0.0006 0.0008
2 2 2 2 2 2
(Specific Volume) , V (m ) (Specific Volume) , V (m )

Figure 5.1.16: Model predictions of t versus V2 for fixed-cavity filtration with ramping pressure
(h0 = 0.015 m, ∆PF = 6 bar, tp = 0)

Online Measurement of Membrane Resistance

The method of measuring Rm on-site from continuous measurements of filtrate


flow based on the conventional Darcian description of cake filtration was validated
using the model generated data. The method assumed that Rm was dependent upon
the slope of dt/dV2 versus 1/V such that the concentration of the bed was not a
function of time. As the above distribution plots show, the concentration varied
greatly under all the different circumstances explored, such that the theoretical basis
of this method was flawed.

To investigate the theoretical error involved in measuring Rm in this manner,


the technique was applied to the filtration predictions for network and un-networked
feed suspensions for varying Rm. The results are plotted in Figure 5.1.17. They
clearly show that the Darcian approach consistently overestimated the membrane
resistance at early times, and was not useful during the compression stage since the
slope of dt/dV2 versus 1/V was negative. However, the Darcian method correctly
predicted the order of magnitude of Rm. As such, it was a useful tool for the quick
extraction of Rm from on-line measurements of filtrate volume, with the caveat that it
generally overestimated Rm, and any trends with time were due to theoretical
assumptions rather than real experimental effects.

231
0.004 10e12 0.02 10e12
1.E+13 0.004 10e11 0.02 10e11
0.004 10e10 0.02 10e10
0.004 10e9 0.02 5x10e11

Darcian membrane resistance 1.E+12

(Pa.s/m)
1.E+11

1.E+10

1.E+09

1.E+08
1.E+02 1.E+03 1.E+04 1.E+05 1.E+06
Time, t (s)

Figure 5.1.17: Model predictions of membrane resistance as measured by the Darcian method for
fixed-cavity filtration with varying Rm; (h0 = 0.015 m, ∆PF = 6 bar, tP = 0)

Complex Material Characteristics

The models were capable of using the complex material characteristics from
the transient settling analysis. An example of the volume fraction distribution
predictions using the Huntington WTP material characteristics is shown in Figure
5.1.18. In this case, the feed was networked, the applied pressure was instantaneous
and the membrane resistance was ignored. As with the results using power-law
functions, φ(0,T) jumped immediately to φ∞,F and a cake of high concentration began
to form. However, unlike the power-law functions, the complex functions exhibited a
peak in D(φ) at low concentrations followed by a minimum. This caused quick
consolidation at these low concentrations, and the process entered cake compression
early. Two sections of the cake were observed – the thick cake at the membrane that
determined the overall flowrate, and the thin cake that stretched to the mid-plane that
determined the cake compression.

232
0.1
φ ∞ ,F

Solids Volume Fraction, φ (Z ,T )


0.09

0.08

0.07

0.06

0.05

0.04

0.03

0.02 φ0
0.01 Rm = 0
0
0 0.2 0.4 0.6 0.8 1

Scaled Height, Z

Figure 5.1.18: Volume fraction distribution results for fixed-cavity filtration with complex
material characteristics (φ0 = 0.02 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0 s, Rm = 0 Pa.s/m)

40000

35000

30000
Time, t (s)

25000

20000

15000

10000
10e12 Power-law
10e11 Power-law
5000 10e10 Power-law
No Rm Power-law

0
0 0.001 0.002 0.003 0.004 0.005
2 2 2
(Specific Volume) , V (m )

Figure 5.1.19: Model predictions of t versus V2 for fixed-cavity filtration with complex material
characteristics and varying Rm (φ0 = 0.02 v/v, h0 = 0.015 m, ∆PF = 6 bar, tP = 0)

233
A plot of the results of t versus V2 for the complex material characteristics for
varying Rm is shown in Figure 5.1.19, along with the results using the power-law
characteristics. The results for the complex material generally showed similar
behaviour to the power-law predictions, as would be expected. The initial rate and the
final filtrate volume differed, mainly due to the differences between the high volume
fraction Py(φ) data (see Section 4.1.2). As Rm increased, the difference between the
initial rates decreased since the cake resistance became negligible compared to the
membrane resistance. These results demonstrate the use of the complex material
characteristics in the algorithms, enabling the filtration modelling of materials (such
as sewage sludges) that exhibit non-traditional behaviour.

Fill-and-Squeeze Results

The flexible-membrane algorithm was used after the fixed-cavity algorithm to


generate predictions for the behaviour of the squeeze-stage of plate-and-frame
filtration. The Huntington WTP power-law material characteristics were used with
∆PF = 6 bar, ∆PS = 10 bar, h0 = 0.015 m and tP = 0 s to generate results for un-
networked (φ0 = 0.004 v/v) and networked (φ0 = 0.02 v/v) feed suspensions for
variable values of tF (1, 2 and 3 hours) with two values of Rm (1010 and 1011 Pa.s/m).

Feed Suspension Un-Networked

The volume fraction distribution results for φ0 = 0.004 v/v, Rm = 1011 Pa.s/m
and tF = 3600 s are shown in Figure 5.1.20, which illustrate the build-up and
subsequent consolidation of the filter cake. Three types of behaviour were observed
for the un-networked case:

(a) Cake formation during the fill-stage;

(b) Cake formation during the squeeze-stage; and

(c) Cake compression during the squeeze-stage

At the beginning of the squeeze stage, Zc(T) was constant or decreasing as the
cake rearranged to the increased applied pressure. The cake did not grow much
during the squeeze-stage cake formation – the process was primarily concerned with
reducing the size of the sediment above the cake through the reduction in H(T), thus

234
the rate was determined by the permeability of the bed. In contrast, there were large
changes in cake concentration and small changes in height during cake compression.
Thus, in general, the fill stage provided the throughput, and the squeeze stage the cake
concentration for plate-and-frame operation.

0.14
(a) Fill-Stage Cake Formation (b) Squeeze-Stage Cake Formation
0.12 φ∞,S
Volume Fraction, φ (Z ,T )

0.1 φ∞,F

0.08

0.06

0.04

0.02
φg
0 φ0
0 0.05 0.1 0.15 0.2 0.25 0 0.2 0.4 0.6 0.8 1
0.14
(c) Cake Compression (d) Fill-and-Squeeze
0.12 φ∞,S
Volume Fraction, φ (Z ,T )

φ∞,F
0.1

0.08

0.06

0.04

0.02
φg
0
φ0
0 0.05 0.1 0.15 0.2 0.25 0 0.05 0.1 0.15 0.2 0.25

Scaled Height, Z Scaled Height, Z

Figure 5.1.20: Volume fraction distribution results for flexible-membrane filtration with
membrane resistance (φ0 = 0.004 v/v, ∆PF = 6 bar, ∆PS = 10 bar; h0 = 0.015 m; Rm = 1011 Pa.s/m, tf
= 3600 s)

The filtration predictions of t versus V2 for three fill times (1, 2 and 3 hours)
and two values of Rm (1010 and 1011 Pa.s/m) are shown in Figure 5.1.21. The fill-
stage behaviour was exactly the same as presented in Figure 5.1.8 and Figure 5.1.12.
Depending on the relative resistance of the membrane and the cake at tF, t was linearly
dependent upon V followed by quadratic behaviour as the cake formed and its
resistance increased, or immediately showed quadratic behaviour. Once the cake
reached the mid-plane of the cavity, the cake compressed and V(t) → V∞,S as t → ∞.
Note that V∞,S was dependent upon the preceding fill-stage.

235
14000

12000

Time, t (s) 10000

8000

6000 10e11 fill


10e11 tf=3600
4000 10e11 tf=7200
10e11 tf=10800
10e10 fill
2000 10e10 tf=3600
10e10 tf=7200
10e10 tf=10800
0
0 0.004 0.008 0.012 0.016

(Specific Volume)2, V 2 (m2)

Figure 5.1.21: Model predictions of t versus V2 for flexible-membrane filtration with varying tF
and Rm (φ0 = 0.004 v/v, h0 = 0.015 m, ∆PF = 6 bar, ∆PS = 10 bar, tP = 0)

A plot of the variation of dt/dV2 with t is shown in Figure 5.1.22. At the onset
of the squeeze phase, dt/dV2 decreased, corresponding to the increased applied
pressure. dt/dV2 was either constant, for the case where the cake resistance dominated
(Rm = 1010 Pa.s/m), or decreasing, as in the case where the membrane resistance
dominated (Rm = 1011 Pa.s/m). The beginning of cake compression in the squeeze
phase was denoted by the dramatic change in the slope of t versus V2. In general, the
compression was much faster during the squeeze stage than during the fill stage since
the height was decreasing, rather than compression caused by the solids loading.

236
10e11 fill
1.E+09 10e11 tf=3600
10e11 tf=7200
10e11 tf=10800
10e10 fill
1.E+08 10e10 tf=3600
dt /dV (s/m ) 10e10 tf=7200
10e10 tf=10800
2

1.E+07
2

1.E+06

1.E+05
0 5000 10000 15000

Time, t (s)

Figure 5.1.22: Model predictions of dt/dV2 versus t for flexible-membrane filtration with varying
tF and Rm (φ0 = 0.004 v/v, h0 = 0.015 m, ∆PF = 6 bar, ∆PS = 10 bar, tP = 0)

Feed Suspension Networked

The volume fraction distribution results for φ0 = 0.02 v/v, Rm = 1011 Pa.s/m
and tF = 3600 s are shown in Figure 5.1.23. Unlike the un-networked case, there was
no cake formation in the networked case, since the suspension was consolidating
throughout the cavity. Figure 5.1.23(a) illustrates the consolidation during the fill-
stage, while Figure 5.1.23(b) shows the rearrangement to the increased operating
pressure and subsequent cake compression during the squeeze stage. The two stages
are combined in Figure 5.1.23(c).

237
0.14
(a) Fill-Stage (b) Squeeze-Stage
Volume Fraction, φ (Z ,T ) 0.12 φ∞,S

0.1 φ∞,F

0.08

0.06

0.04

φ0 φ0
0.02

0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
0.14
(c) Fill-and-Squeeze
0.12
φ∞,S
Volume Fraction, φ (Z ,T )

φ∞,F
0.1

0.08

0.06

0.04

0.02 φ0
0
0 0.2 0.4 0.6 0.8 1
Scaled Height, Z

Figure 5.1.23: Volume fraction distribution results for flexible-membrane filtration with
membrane resistance (φ0 = 0.02 v/v, ∆PF = 6 bar, ∆PS = 10 bar, h0 = 0.015 m, Rm = 1011 Pa.s/m, tF =
3600 s, tP = 0)

The filtration predictions of t versus V2 for three fill times (1, 2 and 3 hours)
and two values of Rm (1010 and 1011 Pa.s/m) are shown in Figure 5.1.24. A plot of the
variation of dt/dV2 with t is shown in Figure 5.1.25. The results show that, as tF
increased, the required squeeze time to reach a given V increased, since the
concentration at the start of the squeeze increased. At the onset of the squeeze phase,
dt/dV2 fell, corresponding to the increased applied pressure. For the networked case,
the process was in compression, and therefore the cake resistance increased very
quickly. In contrast to the un-networked case, the onset of compression was gradual
rather than sudden.

238
20000 10e11 fill
10e11 tf=3600
10e11 tf=7200
10e11 tf=10800
10e10 fill
16000 10e10 tf=3600
10e10 tf=7200
10e10 tf=10800
Time, t (s)

12000

8000

4000

0
0 0.0005 0.001 0.0015 0.002 0.0025

(Specific Volume)2, V 2 (m2)

Figure 5.1.24: Model predictions of t versus V2 for flexible-membrane filtration with varying tF
and Rm (φ0 = 0.02 v/v, h0 = 0.015 m, ∆PF = 6 bar, ∆PS = 10 bar, tP = 0)

10e11 fill 10e10 fill


1.E+10 10e11 tf=3600 10e10 tf=3600
10e11 tf=7200 10e10 tf=7200
10e11 tf=10800 10e10 tf=10800

1.E+09
dt /dV (s/m )
2

1.E+08
2

1.E+07

1.E+06
0 5000 10000 15000 20000
Time, t (s)

Figure 5.1.25: Model predictions of dt/dV2 versus t for flexible-membrane filtration with varying
tF and Rm (φ0 = 0.02 v/v, h0 = 0.015 m, ∆PF = 6 bar, ∆PS = 10 bar, tP = 0)

239
5.1.5 Prediction Tools

Visual basic (VB) programs were developed to provide operators and


designers with quick and easy tools for optimising the operation of and making design
decisions about plate-and-frame filter presses without needing in depth understanding
of the modelling (including the mathematics and programming) or the experimental
material characterisation. The VB programs interpolated from large data files
generated from the analytical and numerical models. The data files were arrays of
information containing predictions covering a large variety of operating conditions for
a given set of material characteristics. The VB programs enabled quick and easy
access to modelling results without needing to run the models for each situation or
understanding the model inputs or internal workings.

Summaries of the documentation for the fill-only and fill-and-squeeze VB


programs are provided here to illustrate their simplicity and convenience. Information
on program installation and troubleshooting is not useful and has been omitted since
the program was intellectual property and is not provided with this thesis. Adam
Kilcullen, a research assistant at the University of Melbourne, carried out the
programming.

Fill-Only Calculation Program

A VB interpolation program was developed using the fill-only models to


generate the necessary data files.

Data File Description

Each data file contained a range of fill-only results for a particular set of
material characteristics. The material type was either from a specific treatment plant
or a given type of coagulant. Each file contained columns of V versus t data for
specified ∆PF and φ0 as predicted by the filter model, including data for five pressures
(6 - 10 bar) (see Figure 5.1.26) and six initial concentrations (0.00336 – 0.02083 v/v,
corresponding to 1 – 6 wt% if the solids density is 3000 g/L, see Figure 5.1.27) in
order of t then V for the six concentrations at the first pressure, followed by t and V
data for the six concentrations at the next pressure and so on.

240
0.4
tL tL + tP t lim

Specific Filtrate Volume, V (m)


0.3

0.2

0.1 1 wt%
2 wt%
3 wt%
4 wt%
5 wt%
6 wt%
0
0 20000 40000 60000 80000 100000
Time, t (s)

Figure 5.1.26: Example of V versus t data with varying φ0 for fill-only array file

0.5
tL tL + tP t lim
Specific Filtrate Volume, V (m)

0.4

0.3

0.2

0.1 10 bar
9 bar
8 bar
7 bar
6 bar
0
0 20000 40000 60000 80000 100000
Time, t (s)

Figure 5.1.27: Example of V versus t data with varying ∆PF for fill-only array file

241
For simplicity, the pressure profile within the press prior to reaching ∆PF (see
Figure 5.1.28) was kept constant. It was assumed that tL = tP = 1800s. The analytical
model used six linear pressure steps, while the numerical model used a linear ramping
pressure.

12
tL tL + tP
Specific Filtrate Volume, V (m)

10

10 bar
2 9 bar
8 bar
7 bar
6 bar
0
0 1000 2000 3000 4000 5000
Time, t (s)

Figure 5.1.28: Applied pressure profiles for fill-only data files

The total filtration time was adjusted to make allowances for different pressure
profiles. For example, if fast-fill pumps were used such that the loading time was ten
minutes rather than thirty, the filtration time was reduced by 20 minutes. If the
pressure rise was quicker than thirty minutes, the filtration time was reduced
appropriately.

Program Description

The fill-only VB program interface, shown in Figure 5.1.29, was divided into
four sections: Inputs, Outputs, Plate Arrangement and Throughputs. Firstly, clicking
‘Open File’ and selecting the desired file loaded the array file. The user specified ∆PF
(bar), φ0 (wt%), d (cm) and ρp (g/L) in the Inputs section. The default value for ρp

242
was 3000 g/L, while ρf was assumed to be 1000 g/L. ρp was the same as that used for
the material characterisation.

Figure 5.1.29: User interface for fill-only calculation program

Fill-only presses have only one independent variable, such that by specifying
either tF (mins) or φF (wt%) in the Outputs section defined the system. φF was related
to VF, φ0 and d by a volume balance (see equation 5.1.7). By selecting ‘Calculate’, the
program performed a quadratic interpolation of the V versus t data in the array file to
give VF (and thus φF) after the specified time, or the time required to reach a specified
φF. The Outputs cleared if either an Input was changed or an Output was selected.

The input files had time and volume limits, tlim and Vlim. tlim was a physical
limit of the data file, set at 86400s (24 hours). Vlim was a limit of the model, such that
the height of the cake was not greater than half the cavity width. The program
notified the user if either of these limits was exceeded.

The user then specified the number of cavities, N, and the cross-sectional area
of each membrane, A in the Plate Arrangement section. N was one more or one less
than the number of plates, depending on the end-plate configuration. By pressing

243
‘Update’, the program calculated the throughput as mass of dry solids per cycle, total
volume of sludge filtered and the final cake volume and mass:

Dry Solids (kg/cycle) = 2NA (d/100) φF ρp …(5.1.61)

Total Sludge Filtered (m3) = 2NA (VF + (d/100)) …(5.1.62)

Cake Volume (m3) = 2NA (d/100) …(5.1.63)

Cake Mass (kg/cycle) = 2NA (d/100) (φF ρp + (1 - φF)ρf) …(5.1.64)

The user could also specify one of N or A, and one of the Throughput values.
By selecting ‘Update’, the program calculated the other Plate Arrangement value.
‘Update’ did not work unless the Outputs were calculated. The Throughput solutions
cleared if an Input, Output or Plate Arrangement was selected.

Fill-and-Squeeze Calculation Program

A VB interpolation program was developed using the fill-and-squeeze models


to generate the necessary data files. The squeeze-stage provided an extra independent
variable, which added an extra level of complexity to the calculation program.

Data File Description

Each data file contained an array of results corresponding to a particular set of


material characteristics at a given ∆PS. Each array consisted of columns of V versus t
data as predicted by the filter models (see Figure 5.1.30) for particular φ0 and ∆PF.
Table 5.1.1 shows the layout of t and V data for m squeeze times and n fill times. The
top line corresponded to the fill-stage results, while the data in columns beneath each
result corresponded to the squeeze-stage results for that particular fill-time. The fill
times ranged up to 12000s, with squeeze times up to 7200s.

Data for five fill pressures (6 - 10 bar) and six initial concentrations (0.00336 –
0.02083 v/v, corresponding to 1 – 6 wt% if the solids density is 3000 g/L) were
contained in each file in order of t then V for the six concentrations at the first
pressure, followed by t and V data for the next pressure and so on: (∆PF,1 φ0,1; ∆PF,1
φ0,2 … ∆PF,1 φ0,6); (∆PF,2 φ0,1; ∆PF,2 φ0,2 … ∆PF,2 φ0,6); … (∆PF,5 φ0,1; ∆PF,5 φ0,2 …

244
∆PF,5 φ0,6). Each data file contained only one squeeze pressure. Different squeeze
pressures required different data files, and interpolation between squeeze pressures
was up to the user.

Specific Filtrate Volume, V (m) 0.1

0.08

0.06

0.04

1200s 2400s
0.02 3600s 4800s
6000s 7200s
8400s 9600s
10800s 12000s
Fill
0
0 5000 10000 15000 20000
Time, t (s)

Figure 5.1.30: Example of V versus t data for fill-and-squeeze data files

Table 5.1.1: Layout of t and V values in array file for given φ0 and ∆PF

tF1 VF1 tF2 VF2 … tFn VFn

tF1 + tS1,1 VS1,1 tF2 + tS2,1 VS2,1 … tFn + tSn,1 VSn,1

tF1 + tS1,2 VS1,2 tF2 + tS2,2 VS2,2 … tFn + tSn,2 VSn,2


tF1 + tS1,m VS1,m tF2 + tS2,m VS2,m … tFn + tSn,m VSn,m

The same assumptions were made regarding the loading and ramping pressure
profile as for the fill-stage.

245
Program Description

The user interface for the fill-and-squeeze calculation program is shown in


Figure 5.1.31. The interface was divided into the same four sections as the fill-only
program: Inputs, Outputs, Plate Arrangement and Throughput.

Figure 5.1.31: User interface for fill-and-squeeze calculation program

After opening the desired data file (which fixed ∆PS), the user specified ∆PF
(bar), φ0 (wt%), d (cm), ρp (g/L) and tH (mins) in the Inputs section. Beyond this
information, fill-and-squeeze presses have two independent variables, tF and tS, or a
range of dependent variables such as tT, φS, 2 h(tS), <Q> and Qrun. These variables
were related by equations 5.1.3 to 5.1.5, 5.1.8 and 5.1.9 through VF and VS, which
were determined by interpolation of the array. The user specified which variables
were to be used by clicking on the appropriate boxes. Two variables were chosen,
although some combinations were not allowed since they did not give a unique
solution. For example, defining the final cavity width and φS was allowed, but tT and
φS was not allowed since there were many combinations of tF and tS that would be
possible.

246
By selecting ‘Calculate’, the program performed an interpolation of the t and V
data to give values for the unspecified variables. The Outputs cleared if an Input was
changed or an Output was selected.

The user then gave N and A in the Plate Arrangement section. By clicking
‘Update’, the program calculated the Throughput, in terms of dry solids per cycle,
total volume of sludge filtered, final cake volume and final cake mass.

Dry Solids (kg/cycle) = 2NA hS φS ρp …(5.1.65)

Total Sludge Filtered (m3) = 2NA (VS + hS) …(5.1.66)

Cake Volume (m3) = 2NA hS …(5.1.67)

Cake Mass (kg/cycle) = 2NA hS (φS ρp + (1 - φS)ρf) …(5.1.68)

The user could also specify one of N or A, and one of the Throughput values.
By selecting ‘Update’, the program calculated the other Plate Arrangement value.
‘Update’ did not work unless the Outputs had been calculated. The Throughput
solutions cleared if an Input, Output or Plate Arrangement was selected.

Examples of using the calculation program to maximise the throughput of


plate-and-frame presses are given in Section 5.3.3.

247
5.2 Model Validation Case Studies

The models were validated using on-site case studies at Langsett WTP,
Hodder WTP (including pilot-scale trials), Thornton Steward WTP and Arnfield
WTP:

- The aim of the work at Langsett WTP was to optimise the throughput of three
presses (two fill-only and one fill-and-squeeze) in order to save operational
expenses;

- The work at Hodder WTP aimed to increase the throughput of the two fill-only
presses such that the presses were no longer the bottleneck for the works.
Predictions were made to allow for future changes in initial solids
concentration due to extra thickening and increased total suspended solids due
to extra clarifying and long-term deterioration of water quality, which showed
that the current presses were insufficient. Baker-Hughes, a press
manufacturer, performed subsequent pilot-scale trials and the results were
made available for further model validation;

- The case study at Thornton Steward WTP highlighted the importance of


membrane resistance in some instances, which initiated the development of the
numerical filtration models. The models were used to evaluate the options for
improving the press throughput;

- The case study at Arnfield WTP combined the laboratory characterisation


results and the press modelling to predict the effectiveness of the addition of
lime to ferric water treatment sludges.

248
5.2.1 Langsett Water Treatment Plant

Background

Measurements were taken at Yorkshire Water’s Langsett WTP during June


2001 in order to validate the model and to optimise the performance of the filter
presses. There were three automated plate-and-frame filter presses, fed by vertical,
positive displacement ram pumps (or ‘Willet’ pumps). Presses 1 and 2 were identical,
older fill-only presses (see Figure 5.2.1(a)) while Press 3 was a fill-and-squeeze press
(see Figure 5.2.1(b)). The fill-stage used a flux cut-off condition (that is, filtration
proceeded until the pumping rate dropped below a certain value), while the squeeze-
phase used a fixed time cut-off.

(a) (b)

Figure 5.2.1: Langsett WTP plate-and-frame presses: (a) Operator Bill Brady with Press 1; (b)
Press 3

249
Results and Discussion

A total of six filter press runs were analysed (see Table 5.2.1). Two runs were
monitored for press 2, both with 70 second per stroke cut-off conditions. Four runs
were monitored for press 3, with 60, 40, 40 and 15 second per stroke fill cut-off
conditions and 60, 60, 90 and 90 minute squeeze times respectively. The initial and
final solids concentrations for each run were determined by oven drying. The cake
was sampled across the width of the cake and away from ports and baffles.

Table 5.2.1: Filter press runs from Langsett WTP

Fill Cut-Off Squeeze Time Initial Solids Final Solids


Run ID Date
(s/stroke) tS (min) φ0 (v/v) φF or φS (v/v)
2.1 08/06/01 70 n/a 0.0098 0.0922
2.2 11/06/01 70 n/a 0.0104 0.0873
3.1 05/06/01 60 60 0.0091 0.1219
3.2 11/06/01 40 60 0.0104 0.0993
3.3 12/06/01 40 90 0.0104* 0.1042
3.4 13/06/01 15 90 0.0104* 0.0946
*
Assumed to be equal to 11/06/01

Material Characteristics

The material characterisation results for Langsett WTP, including the four
samples taken from the filter press feeds in June 2001, are presented in Section 4.1.1.
The results show that the filtration properties did not change much and that changes in
plant operating behaviour were more likely to be due to changes in operating
conditions rather than material properties.

Comparing the measured final solids to Py(φ) suggests that these filtration runs
were not reaching the compressional limit of the material. Press 2 averaged φF =
0.0898 v/v (22.8 wt%) compared with φ∞,F = 0.1236 v/v (29.7 wt%). Press 3 averaged
φS = 0.1050 v/v (26.0 wt%) compared with φ∞,S = 0.1346 v/v (31.8 wt%). Therefore,
the presses were likely to be operating in such a way that they were limited by the
permeability of the sludge. As such, changes to operating procedures that effectively
increase the permeability or the time of filtration would ensure greater final solids, or
give higher throughput for a fixed filtration time.

250
Press Dimensions

The physical dimensions of the filter presses were measured to calculate Vpress,
and Apress. Presses 1 and 2 consisted of 119 fixed frames each. Each frame had two
1150 mm x 1150 mm membranes attached, with cavities of approximately 30 mm
between the frames. The suspension was fed to the cavity through a central inlet (i.d.
= 140 mm), and the cavities were kept apart by seven conical baffles (d1 = 50 mm, d2
= 100 mm) (see Figure 5.2.2). Filter cakes formed between the end plates and the
press, such that there were 120 filter cakes. Allowing for the baffles, the central feed
inlet and the tapered edges, the volume of each cavity was 0.04106 m3, and the
surface area available for filtration was 1.452 m2 per membrane. Therefore, Vpress2 was
4.93 m3/run and Apress2 was 349 m2.

1150 mm

50 mm
15 mm
140 mm
100 mm

1230 mm
Width
= 30 mm

Figure 5.2.2: Layout of the membrane and cavity for Langsett WTP Presses 1 and 2

Press 3 consisted of 101 plates. Each plate had a 1330 mm x 1330 mm


flexible membrane. The initial membrane separation, h0, was approximately 32 mm.
The cavity width at the end of the squeeze cycle varied from 14 mm to 26 mm. As
with the other two presses, the suspension was fed to the cavity through a central inlet
(i.d. = 150 mm), and the cavities were kept apart by four conical baffles (d1 = 110
mm, d2 = 160 mm) (see Figure 5.2.3). Filter cakes did not form in between the end
plates and the press, therefore there were 100 filter cakes. Allowing for the baffles,
the central feed inlet and the tapered edges, the volume of each cavity was 0.03332
m3, and the surface area available for filtration was 1.6187 m2 per membrane.
Therefore, Vpress3 = 3.33 m3/run and Apress3 = 324 m2.

251
1000 mm

110 mm
16 mm
150 mm
160 mm

1330 mm Width
= 32 mm

Figure 5.2.3: Layout of the membrane and cavity for Press 3

Pump Volumes

The dimensions of the vertical displacement pumps were measured to give the
volume delivered per stroke, Vstroke. Four identical R&B refurbished ram pumps fed
presses 1 and 2. The diameter of the pistons was 15.2 cm and the length of each
stroke was approximately 20.3 cm, giving Vstroke2 as 3.7 L/stroke. A larger Willet ram
pump fed press 3. The diameter of the piston was 24 cm and the stroke length was 30
cm (therefore Vstroke3 = 13.6 L/stroke). The nominal volume given by the
manufacturer’s literature for the larger pump was 9 L/stroke. The difference between
the measured and delivered values introduced significant uncertainty when
performing volume balances to determine V(t).

Filtration Pressure

The pressure delivered by the pumps was monitored using pressure gauges
mounted on the press inlets. The accuracy of these gauges was questionable. After an
initial loading stage of about 20 minutes, the pressure showed a ramping pressure over
about 30 minutes, followed by constant pressure (∆PF ≈ 600 kPa). ∆PS was
approximately 900 kPa for press 3. From Py(φ) for samples 07/06/01 and 11/06/01
(see Figure 4.1.3), φ∞,F = 0.124 ± 0.006 v/v (29.7 wt%) by interpolation and φ∞,S =
0.135 ± 0.007 v/v (31.8 wt%) by extrapolation.

252
Fill-Only Results

The volume of filtrate with time was required to give the performance of filter
presses. This was measured indirectly during the fill stage from the volume of sludge
delivered to the presses, which was monitored by counting the number of pump
strokes in a given interval. The control unit automatically logged the count every
twenty minutes for presses 1 and 2 and every ten minutes for press 3. The cumulative
stroke count was converted to V(t) by calculating the volume of feed from the number
of strokes and Vstroke, subtracting the amount of material needed to fill the cavities,
Vpress, and dividing by Apress. This assumes that there is no filtrate until all the cavities
are filled.

A plot of the t versus V2 results for Press 2 is presented in Figure 5.2.4.

100000
VF2 = 0.0124 m2

VF2 = 0.0160 m2

2.1 Raw data


2.1 Adjusted
2.2 Raw data
80000 2.2 Adjusted
Time, t (s)

60000

40000

20000

tL + tP
0
0 0.01 0.02 0.03 0.04 0.05 0.06
2 2 2
(Specific Filtrate Volume) , V (m )

Figure 5.2.4: t versus V2 results (raw and adjusted) for Langsett WTP Press 2

The raw data shows very close agreement between the two runs, with only
slight variations over time. After loading the press with sludge, the transient data
shows an initially decreasing slope corresponding to constant flowrate behaviour due
to the initial ramping pressure and the effect of the membrane resistance. A linear

253
section follows, indicating that the process was in the constant pressure cake
formation regime. Run 2.2 was slightly slower than run 2.1, corresponding to the
slightly higher initial solids. Towards the end of the cycle, the results deviated from
linearity, and entered cake compression.

Figure 5.2.4 also shows VF2 as given by equation 5.1.7, using φ0, φF and h0.
The raw V results from the stroke count were adjusted using this value, such that V
corresponded to the measured solids concentrations and press dimensions. The raw
V2 data exceeded the adjusted data by a factor of four to five, suggesting significant
errors in the stroke volume or the solids concentrations. To give VF corresponding to
the raw data for runs 2.1 and 2.2, φ0 would equal 0.0053 or 0.0050 v/v or φF would
equal 0.1691 or 0.1812 v/v, respectively. The error in cake solids were small since
the cake was sampled at several places and an average calculated, and 0.1691 and
0.1812 v/v are much greater than the compressibility limit (0.1236 v/v). The feed
concentration may have varied during the run, but was unlikely to have diluted by
half. The most likely error arose from a reduced efficiency in the stroke volume. The
stroke volume needed to be less than 2.01 L/stroke for run 2.1 or 1.78 L/stroke for run
2.2 for V to be less than VF, compared to the measured value of 3.7 L/stroke.
Generally, these results showed that measuring the volume of filtrate based upon the
number of strokes was error prone.

Rm was calculated from the slope of dt/dV2 versus 1/V following the approach
outlined in Section 2.5.2 (see Figure 5.2.5). dt/dV2 was calculated as the gradient
between two data points. dt/dV2 versus 1/V was linear only after the operating
pressure was reached and before cake compression. The raw data gave Rm = (1.04 ±
0.01) x 1011 Pas/m while the adjusted results gave Rm = (2.13 ± 0.17) x 1011 Pas/m.

The average throughput results, calculated from equation 5.1.4, as a function


of average cake solids (using the adjusted results) are presented in Figure 5.2.6, where
tH = 3600s. Note that the average throughput assumed that one cycle commences as
soon as the previous cycle was finished, such that there was an increase in the number
of cycles with reduced filtration time.

254
1.2E+07
2.1 Raw data
2.1 Adjusted
1.0E+07 2.2 Raw data
2.2 Adjusted

8.0E+06
dt /dV (s/m )
2

6.0E+06
2

R m = (2.13 ± 0.17) x 1011 Pas/m


4.0E+06

2.0E+06
R m = (1.04 ± 0.01) x 1011 Pas/m

0.0E+00
0 20 40 60 80
-1
1/V (m )

Figure 5.2.5: dt/dV2 versus 1/V results for Langsett WTP Press 2

0.4
Average Solids Throughput, Qp

0.35

0.3

0.25
(kg/hr/m )
2

0.2

0.15
φF = 0.0873 v/v

φF = 0.0922 v/v

0.1

0.05 Run 2.1


Run 2.2
0
0 0.02 0.04 0.06 0.08 0.1

Average Cake Solids, <φ > (v/v)


Figure 5.2.6: Solids throughput versus cake solids results for Langsett WTP Press 2

255
The throughput results showed a maximum at low cake solids, corresponding
to high filtration rates at early times, followed by low throughputs at high cakes
solids, due to the low filtration rates at the end of the run required to reach the high
cake solids. In general, high throughputs and high cake solids were mutually
exclusive for fill-only filter presses. A drop in cake solids was needed in order to
increase the throughput of the fill-only presses at Langsett WTP. For example, based
on run 2.1, throughput increases of up to 105% were possible, but the cake solids
reduced dramatically from 0.0922 to 0.0588 v/v.

Fill-and-Squeeze Results

Figure 5.2.7 shows the t versus V2 results for press 3. The stroke data during
the fill stage was converted to V using a stroke volume of 9.0 L/stroke. The VS datum
points were calculated from equation 5.1.46 using h0, φ0, φS and VF. V during the
squeeze stage was unknown since the filtrate volume was not measured directly.
Since VS was dependent on VF and V(tF < t < tS) was unknown, the adjustment made
for the fill-only results could not be made, and the accuracy of the stroke volume was
unclear. The nominal value of 9 L/stroke was used instead of the measured value of
13.6 L/stroke, since VF > VS for the later.

The t versus V2 results during the fill-stage showed similar trends as the results
for press 2 – loading, followed by the initial non-linearity due to the ramping pressure
and membrane resistance, linear cake formation and, lastly, cake compression. The
slope of t versus V2 was expected to increase for the squeeze-stage since the pressure
was increased. This was seen for all runs except 3.4. However, since the transient
squeeze data was unknown, the initial rate may have been greater, but then decreased
as the process entered cake compression. Interestingly, run 3.3 showed slower
filtration than run 3.2, suggesting that φ0 had in fact changed.

The filtration time as a function of <φ> for press 3 is presented in Figure 5.2.8.
The results presented indicate a dramatic increase in <φ> due to the flexible-
membrane, compared to the small increase in <φ> during the fill-stage, since the
membrane separation was fixed. Thus, in the operation of fill-and-squeeze filter
presses, the fill-stage provided the throughput, while the squeeze-stage gave the cake
solids.

256
70000
Run 3.1
60000 Run 3.2
Run 3.3
Run 3.4
Time, t (s) 50000

40000

30000

VS2 = 0.0071 m2

VS2

VS2
20000

= 0.0100 m2

= 0.0105 m2
0.0034 m2
10000 VS2 =

0
0 0.004 0.008 0.012
2 2 2
(Specific Filtrate Volume) , V (m )
Figure 5.2.7: t versus V2 results for Langsett WTP Press 3

70000
Run 3.1
Run 3.2
60000 Run 3.3
Run 3.4
50000
Time, t (s)

φS = 9.93 v/v

40000

30000
φS = 10.42 v/v

φS = 12.19 v/v

20000
φS = 9.46 v/v

10000

0
0 0.05 0.1 0.15

Average Cake Solids, <φ > (v/v)

Figure 5.2.8: Time versus cake solids for Langsett WTP Press 3

257
The trends shown by the squeeze stage were governed by the long fill-times.
Theoretically, the required squeeze time to reach a given cake solids initially increases
as the fill time increases due to the higher amount of solids in the cavity at the end of
the fill-stage. The squeeze time passes through a maximum, and begins to fall as the
fill-stage cake solids begin to approach the maximum solids. The results suggest that
the fill-stage was passing this maximum, since run 3.1 reached a higher cake solids
than run 3.2 and run 3.3 surpassed run 3.4.

Rm was calculated from the slope of dt/dV2 versus 1/V during the fill-stage as
(8.46 ± 2.87) x 1010 Pas/m (see Figure 5.2.9). As for the press 2 results, linear
behaviour was seen only after the operating pressure was reached and before cake
compression. Since there was no filtrate monitoring, the resistance during the
squeeze-stage (and therefore the pressure dependence of Rm) was unknown.

1.E+07
Run 3.1
9.E+06 Run 3.2
Run 3.3
Run 3.4
8.E+06
dt /dV (s/m )
2

7.E+06
2

6.E+06

5.E+06
R m = (8.46 ± 2.87) x 1010 Pas/m
4.E+06

3.E+06
0 20 40 60 80 100
-1
1/V (m )

Figure 5.2.9: dt/dV2 versus 1/V results for Langsett WTP Press 3

Reducing the cut-off criteria for the fill-stage had a minor effect on the final
solids concentration, but significantly reduced the cycle time. Thus more cycles could
have been performed such that the throughput increased, without the loss of cake
solids. This is illustrated in Figure 5.2.10. Based on run 3.1, performing 3 runs in 48

258
hours gave a solids throughput of 8.9 kg/m2, whereas performing 7 cycles (based on
run 3.4) gave a solids throughput of 14.7 kg/m2, an increase of 65%. Run 3.2 had a
higher throughput than run 3.3 despite longer runs, since the initial filtration rate was
quicker. These results demonstrated the ability of fill-and-squeeze presses to
maximise both throughput and cake solids.

16
Run 3.1
Solids Throughput, Qp (kg/m )

Run 3.2
2

14
Run 3.3
Run 3.4
12

10

0
0 12 24 36 48
Time, t (hr)

Figure 5.2.10: Solids throughput versus time for multiple runs of Langsett WTP Press 3

259
Model Validation

The analytical and numerical models were critically examined based upon the
observations at Langsett WTP. The inputs to the models were the material
characteristics, (Py(φ) and D(φ)) and the operating conditions (∆P(t), d, φ0 and Rm).
The loading time was 20 minutes, followed by a ramping pressure of 30 minutes for
all the following predictions. The analytical model used a stepped-pressure, whereas
the numerical model used a linear pressure rise.

Fill-Only Validation

The experimental (adjusted data) and modelling results for press 2 are
presented in Figure 5.2.11. The material characteristics for the two runs were from
07/06/01 and 11/06/01 respectively, as presented in Section 4.1.1. Rm for the
numerical model was 2 x 1011 Pa.s/m.

200000

180000 2.1 Adjusted 2.2 Adjusted


Analytical model Analytical model
160000 Numerical model Numerical model

140000
Time, t (s)

120000

100000

80000
V∞,F2 = 0.0306 m2

V∞,F2 = 0.0241 m2
VF2 = 0.0160 m 2

VF2 = 0.0124 m 2

60000

40000

20000

0
0 0.008 0.016 0.024 0.032 0 0.005 0.01 0.015 0.02 0.025
2 2
(Specific Filtrate Volume)2, V (m2) (Specific Filtrate Volume)2, V (m2)

Figure 5.2.11: Experimental data and model predictions for Langsett WTP press 2

The results show that the models reasonably predicted the filtration
performance. Both models under predicted the filtration rate during the linear section,
especially for run 2.1. This suggested that the material characteristics had changed
from 7/6/01 to 8/6/01. VF for the adjusted data was less than V∞,F, although the
dramatic change in slope of the experimental data suggested that φ was approaching
φ∞. The results highlighted the importance of measurement of filtrate flowrate and
applied pressure, and accurate material characterisation at the applied pressure.

260
As well as predicting the magnitude, the models showed the appropriate
characteristics of the filtration runs. The analytical model, which did not include the
effect of the membrane resistance, showed linear behaviour in t versus V2 from tL + tP
onwards. This model also assumed no compression, and thus did not follow the
deviation from linearity at the end of the cycle.

The numerical model, which did include Rm, initially followed the
experimental data, illustrating the effect of the membrane. This model then
demonstrated linearity and deviation from linearity as V approached V∞,F, although at
higher values than the experimental data. Assuming that the experimental V data was
reasonably accurate, the results suggested early onset of cake compression. An aspect
of fixed cavity filtration that was not covered by the models was that material must
spread from the single, central feed inlet, such that there was lateral, two-dimensional
compression as well as one-dimensional normal compression. This aspect was
investigated further in the Hodder WTP case study (see section 5.2.2).

Fill-and-Squeeze Validation

The experimental and modelling results for press 3 are presented in Figure
5.2.12. The material characteristics for 05/06/01 were used to predict run 3.1, while
the characteristics for 11/06/01 were used to predict runs 3.2 to 3.4. Rm for the
numerical model was 9.61 x 1010, 6.20 x 1010, 9.79 x 1010 and 8.25 x 1010 Pas/m for
the four runs respectively.

Encouragingly, the models matched the experimental data very well, despite
the experimental difficulties outlined above. As for the press 2 results, the fill stage
modelling showed the features of ramping pressure and cake formation. dt/dV2 during
the linear sections of the experimental and model results were in close agreement,
validating the material characterisation. The numerical model results were
consistently higher than the analytical model since the former included the effect of
the membrane resistance and cake compression during the fill-stage. The modelling
of Rm matched the site data well, except for run 3.4, which appeared to have a short
loading time. However, the numerical model predicted cake compression later than
the experimental results, such that the model errors increased towards the end of the
fill-stage.

261
70000 60000

60000 50000

50000
40000
Time, t (s)

40000
30000
30000
20000
20000
Run 3.1 Run 3.2
10000 10000
Analytical model Analytical model
Numerical model Numerical model
0 0
0 0.005 0.01 0.015 0 0.004 0.008 0.012
50000 25000

40000 20000
Time, t (s)

30000 15000

20000 10000

10000 Run 3.3 5000 Run 3.4


Analytical model Analytical model
Numerical model Numerical model
0 0
0 0.002 0.004 0.006 0.008 0.01 0 0.001 0.002 0.003 0.004 0.005

(Specific Filtrate Volume)2, V 2 (m2) (Specific Filtrate Volume)2, V 2 (m2)

Figure 5.2.12: Experimental data and model predictions for Langsett WTP press 3

The models also showed information during the squeeze-stage, which was not
available experimentally, with predictions of both cake formation and compression.
dt/dV2 at the start of the squeeze and the extent of filtration depended on the cake
solids at the end of the fill-stage.

262
Outcomes

The case study at Langsett WTP showed results for both fill-only and fill-and-
squeeze presses. During the fill-stage, dt/dV2 increased, corresponding to a ramping
pressure and the effect of the membrane resistance. This was followed by constant
pressure cake formation and cake compression. The results for the squeeze-stage
were limited since there was no filtrate measurement. The fill-only results showed
that increasing the throughput was detrimental to cake solids, while the fill-and-
squeeze results illustrated that press 3 was more versatile than presses 1 and 2 such
that increased throughput was possible without any loss of cake solids. Comparisons
of the experimental results with the predictions of the analytical and numerical
filtration models showed that the models predicted the necessary features of the filter
press cycle and that the material characterisation reflected the sludge behaviour.

Upon implementation of shorter filtration cycles, the throughput of press 3


was dramatically increased and consistently high cake solids were achieved. In 2002,
Yorkshire Water performed an evaluation of the impact at the site, which concluded
that considerable operating and capital cost savings had been made due to reductions
in overtime and avoiding further capital expenditure. The throughput was improved
such that one of the fill-only presses could be taken off-line, or sludge could be
tankered from nearby treatment sites for dewatering at Langsett WTP.

263
5.2.2 Hodder Water Treatment Plant

Figure 5.2.13: Hodder Water Treatment Plant, Lancashire

Background

Investigations were carried out in October and November 2001 on filtration at


United Utilities’ Hodder WTP in Lancashire with the aim of increasing the throughput
of the two fill-only plate-and-frame presses. At this time, the presses were unable to
handle the required amount of sludge and restricted the throughput of the works,
especially during periods of high colour and turbidity.

Two staged improvements, AMP2 and AMP3, were under construction or


scheduled to be constructed, and their expected impacts on the sludge processing were
taken into consideration. The AMP2 project included extra thickening and the
concentration of the sludge was expected to increase from 1-2 wt% to 3-6 wt%. The
AMP3 project aimed to introduce an extra clarifying step that would increase the
amount of sludge to be processed due to increased coagulant and flocculant dosing for
clarifiers. Combined with an expected long-term deterioration of raw water quality,
the sludge production was expected to significantly increase.

The aim of this work was to optimise the performance of the presses under the
current conditions, to estimate the capability of the presses to handle the changes in
sludge quality and quantity caused by AMP2 and AMP3, and to put forward

264
recommendations on how to overcome any difficulties, including refurbishing the
current presses (remedial works) or purchasing new presses.

Results and Discussion

Raw Water Quality and Alum Dosing

The operators at Hodder WTP measured the colour and turbidity of the raw
water daily. A plot of the variation in turbidity with colour over the previous year is
shown in Figure 5.2.14. The turbidity ranged from 2 to 13 NTU’s and the colour
varied between 50 and 190 colour units, such that the raw water was very coloured
and quite turbid. The colour was highest during autumn and winter, and was expected
to deteriorate in the future. Normally, there is no correlation or even a negative
correlation between the two measures of water quality. It was therefore surprising to
see an increase in turbidity with colour, even though it was a weak correlation.

14

12

10
Turbidity (NTU)

R2 = 0.5326
8

0
0 50 100 150 200
Colour

Figure 5.2.14: Raw water turbidity and colour at Hodder WTP

The daily alum dose was calculated from the amount of aqueous Al2(SO4)3
used and the raw water throughput. A plot of the variation in alum dose with colour is
shown in Figure 5.2.15. The alum dose increased from 2.8 mg/L at low colour to 3.8

265
mg/L at high colour. As was expected, the results showed a reasonable correlation
between the dose and colour, and suggested that the water was either over-dosed at
low colour or under-dosed at high colour (more likely the former).

3.8
R2 = 0.7581
Alum dose (mg/L)

3.6

3.4

3.2

2.8
0 50 100 150 200
Colour

Figure 5.2.15: Variation of alum dose with colour at Hodder WTP

The standard calculation used by water treatment operators and designers to


give the total amount of suspended solids (TSS) in mg/L is:

TSS = (Colour x 0.2) + (Turbidity x 1.2) + (Al dose x 1.2) …(5.3.1)

A plot of the variation in TSS with colour is presented in Figure 5.2.16,


showing a strong correlation between the raw water quality and the solids production.
The amount of solids varied significantly from 22 to 65 mg/L. For the purposes of
prediction and optimisation, average and worst water quality were assumed to yield
35 and 65 mg/L TSS respectively.

The total amount of solid material produced by the process was calculated
from the total raw water throughput, where the current throughput was 65 ML/day,
the average throughput required was 80 ML/day, and the maximum was 105 ML/day.
Thus, there were six solids throughputs (in tonnes of dry solids, tds) to consider for

266
each dewatering option, as shown in Table 5.2.2, ranging from 2.28 tds/day at average
water quality and current raw water throughput to 6.83 tds/day at worst water quality
and maximum raw water throughput. The desired cake solids was 20 wt% or better
(φF > 0.0769 v/v).

70
Total Suspended Solids (mg/L)

60
R2 = 0.9537
50

40

30

20

10

0
0 50 100 150 200
Colour

Figure 5.2.16: Variation of total suspended solids with colour of at Hodder WTP

Table 5.2.2: Required solids throughput at current coagulation and flocculation conditions

Raw water throughput Solids throughput


(ML/day) (tds/day)

Average water quality Worst water quality

Current - 65 2.28 4.23

Average - 80 2.80 5.20

Maximum - 105 3.68 6.83

267
Material Characterisation

Six samples were taken from the feed to the presses over a range of days
during the course of the investigation. A summary of the samples and the
characterisation results are presented in Section 4.1.2, along with the
characterisation results for other alum water treatment sludges. Extrapolating the
average Hodder WTP characteristics to 10 bar gives φ∞,F = 0.0964 v/v.

Press Dimensions

The physical dimensions of the filter presses were measured to calculate Vpress
and Apress. Presses 1 and 2 were identical and consisted of 79 fixed frames each. Each
frame had two 1200 mm x 1200 mm membranes attached, forming cavities of
approximately 30 mm between the frames. The suspension was fed to the cavities
through central inlet ports (i.d. = 160 cm), and the cavities kept apart by eight conical
baffles (d1 = 5 cm, d2 = 11 cm) (see Figure 5.2.17). Filter cakes formed between the
end plates and the press, giving 80 filter cakes.

50 mm
15 mm
160 mm
110 mm

1200 mm
Width
= 30mm

Figure 5.2.17: Layout of the membrane and cavity for Filter Presses 1 and 2

Allowing for the baffles, the central feed inlet and the tapered edges, the
volume of each cavity was 0.04133 m3, and the surface area available for filtration
was 1.404 m2 per membrane. Therefore, Vpress = 3.31 m3/run and Apress = 225 m2.

268
Pump Volumes

Two vertical, positive displacement Willet ram pumps fed the filter presses at
Hodder WTP. The diameter of the pistons was 15.2 cm and the length of each stroke
was approximately 20.3 cm, giving Vstroke as 3.7 litres. The difference between the
measured and delivered values introduced errors when performing volume balances.

Press Performance

(a) (b)

Figure 5.2.18: Fill-only press #2 at Hodder WTP (a) Press in operation; (b) Operator Paul
Griffiths unloading the press

Each press cycle consisted of closing the press, starting the pumps, loading the
cavities with sludge, filtration and unloading. The presses could be shutdown at this
time for any cloth replacement, washing or general maintenance that was required.
One cycle of each press was monitored on 06/11/01, as summarised in Table 5.2.3. φ0
and φF for each cycle were determined by oven drying. φ0 was 0.00744 v/v (2.20
wt%) for both runs. φF was measured at a range of cake locations since the cakes
exhibited varied concentration. The pressure exerted by the pumps was observed over
time using the pressure gauges mounted on the surge tanks of the pumps. The volume
of sludge delivered to the presses was monitored by manually counting the number of

269
piston strokes of the displacement pumps in one-minute intervals at various times
during the entire press cycle.

Table 5.2.3: Fill-only filter press runs at Hodder WTP

Press
φ0 tL tP tF ∆ PF φF (edge) φF (centre)
(v/v) (s) (s) (s) (bar) (v/v) (v/v)

1 0.0074 1200 2700 34500 9.5 0.0763 0.0410

2 0.0074 1200 2100 37740 10 0.0809 0.0353

Solids Distribution

A minimum cake concentration was required to remove the cake from the
cloths and to transport the cake to hoppers nearby. However, different regions within
the cavity experienced different amounts of compressional forces.

Figure 5.2.19: Observed lateral movement during filtration - Red lines represent flow; Blue lines
represent lateral compression

The sludge flowed from the central feed inlet to the outermost edges of the
plates, as represented by the red lines in Figure 5.2.19. There were consolidation
forces both normal to the membrane as the sludge compacted against the membrane
and lateral to the membrane as the sludge was pushed from the inlet to the edges by

270
the feed coming in through the inlet. Therefore, there was a solids distribution both
laterally and normally. The normal distribution from membrane to membrane was
predicted by modelling while the lateral compression (represented by dotted blue lines
in Figure 5.2.19) would have required three-dimensional computational fluid
dynamics programming to predict.

The degree of lateral variation in concentration across the cake was


investigated. Samples were taken from various positions of a cake that was far from
completely compressed and was expected to show a concentration distribution. The
samples were oven dried and the solids concentration determined. The results are
presented in Figure 5.2.20.

18.7 wt% 17.1 wt%


18.4 wt%

11.0 wt%

19.2 wt%

11.5 wt%

Figure 5.2.20: Solids distribution across cake

The concentration near the edges of the cake was significantly higher than
near the feed inlet in the middle of the cake. The concentration was also higher close
to the baffles, confirming the observation that channels formed to transport the sludge
to the edges.

This experiment indicated that the measurement of the final cake solids must
be either representative of the whole cake or measured at the same position every
time. An accurate measure of the average concentration would have required

271
sampling the entire cake, which was unfeasible, or waiting until the run reached
equilibrium, which was not possible due to plant throughput requirements. Instead,
the cake was sampled from the same position consistently, giving an indication of the
maximum concentration achieved and allowing comparisons between cakes and
between runs to be made.

If the energy required to compress the cake was dependent on the applied
force multiplied by the length, the lateral compression must have been inefficient
compared to the normal compression since it acted over distances up to 0.5 m,
whereas the normal compression acted over half the width of the cavity, 0.015 m.
Thus, multiple smaller inlets distributed evenly over the plate may have reduced the
amount of lateral compression.

Fill-Only Results

The stroke-count and pressure profile results are presented in Figure 5.2.21.
Both cycles showed an initial 1200s period during which the presses filled with sludge
at constant flowrate. After this, the pressure exerted by the pumps rose to operating
pressures of 9.5 and 10 bar for presses 1 and 2 respectively. tP was 2700s for press 1
and 2100s for press 2. The feed flowrate began to decrease before the operating
pressure was reached. Constant pressure was applied for 9½ and 10½ hours for press
1 and 2 respectively, during which the flowrate monotonically decreased.

The number of strokes per minute was converted to V based on a stroke


volume of 3.7 L/stroke. The results are presented as a plot of t versus V2 in Figure
5.2.22, which shows no filtrate during loading, decreasing dt/dV2 due to the ramping
pressure and the membrane resistance and then constant dt/dV2 corresponding to
constant pressure cake formation. It was expected that press 2 would show a lower
rather than a higher gradient since the applied pressure was slightly higher. There
were small deviations from linearity suggesting that processes were entering
compression.

272
tL tL + tP
12 3.0

Applied Pressure, P (t ) (bar)


10 2.5

Feed flowrate, q( t) (L/s)


8 2.0

6 1.5

4 1.0

Press 1 P(t)
2 Press 2 P(t) 0.5
Press 1 q(t)
Press 2 q(t)
0 0.0
0 1000 2000 3000 4000 5000
Time, t (s)

Figure 5.2.21: Applied pressure and feed flowrate variation with time for Hodder WTP fill-only
presses

40000

35000 Press 1
Press 2

30000
Time, t (s)

25000

20000
VF2 = 0.0088 m2

VF2 = 0.0091 m2

15000

10000

5000 t L +t P
tL
0
0 0.002 0.004 0.006 0.008 0.01
2 2 2
(Specific Filtrate Volume) , V (m )

Figure 5.2.22: t versus V2 results (raw and adjusted) for Hodder WTP fill-only presses

273
A measure of the membrane resistance was determined from the slope of
dt/dV2 versus 1/V (see Figure 5.2.23) using the modified Darcian approach outlined in
Section 2.5.2. Rm was 4.30 x 1010 Pas/m for press 1 and 3.28 x 1010 Pas/m for press 2.
The magnitude of the Rm values indicated that the membrane resistance was not a
significant contribution to the filtration resistance, which reflected the weekly cloth
cleaning performed by the operators.

6.0.E+06

Press 1
5.5.E+06 Press 2

5.0.E+06
dt /dV (sm )
-2

4.5.E+06 R m = 3.28 x 1010 Pas/m


2

4.0.E+06

3.5.E+06
R m = 4.30 x 1010 Pas/m

3.0.E+06
0 20 40 60 80 100
-1
1/V (m )

Figure 5.2.23: dt/dV2 versus 1/V for Hodder WTP fill-only presses

The average cake solids were calculated from equation 5.1.7 using φ0, h0 and
V(t). The final cake solids for presses 1 and 2 were calculated as 0.0547 and 0.0540
v/v respectively, which were between the maximum and minimum measured results,
and were much less than φ∞,F, which shows that longer cycles could reach higher cake
solids. Adjusting the raw data such that φF equalled the maximum measured values
required stroke volumes of 5.17 and 5.55 L/stroke.

The average throughput was calculated from equation 5.1.4 with tH = 3600s,
which assumed that the press ran continuously. The results of Qp versus <φ> for
presses 1 and 2 are presented in Figure 5.2.24.

274
0.5
Press 1

Average Solids Throughput, Qp


Press 2
0.4

(kg/hr/m2) 0.3

0.2 φ0 = 0.0074 v/v

φF = 0.0540 v/v
φF = 0.0547 v/v
0.1

0
0 0.02 0.04 0.06

Average Cake Solids, <φ > (v/v)

Figure 5.2.24: Solids throughput versus cake solids results for Hodder WTP fill-only presses

The downfall of fill-only type plate-and-frame presses was that any increase in
throughput required a loss of final cake concentration. The plots show that the solids
throughput could have been increased by 68% and 87% for presses 1 and 2
respectively, but <φ> would not have been much greater than φ0 and the cake would
have been very wet. The cakes for the measured runs were below the required cake
concentration, such that the final solids had already been compromised to increase
throughput. Therefore, under the current conditions, there was no room for
improvement in throughput by changing the operational procedures used for the
presses.

The total solids throughput of each run was approximately 0.54 tds.
Therefore, the combined daily solids throughput was approximately 2.16 tds/day at an
average cake concentration of 0.0543 v/v (14.7 wt%). Therefore, the presses were a
bottleneck at all throughputs and water qualities, which conformed to the operator’s
experience.

275
Fill-Only Model Validation

The analytical and numerical models for fixed-cavity filtration were critically
examined based upon the observations at Hodder WTP. The experimental and
modelling results are presented in Figure 5.2.25. The inputs to the models were the
material characteristics (Py(φ) and D(φ) for average Hodder WTP sludge, see Table
4.1.9) and the operating conditions (∆P(t), d, φ0 and Rm). The loading time was
1200s, followed by a ramping pressure of 2700s and 2100s for presses 1 and 2
respectively. The analytical model used a stepped-pressure, whereas the numerical
model used a linear pressure rise.

The press results and model predictions show exceptionally close agreement,
especially considering the number of assumptions and possible sources of error. Both
the analytical and numerical models showed an increasing dt/dV2 during the ramping
pressure and constant dt/dV2 during constant pressure filtration. The numerical model
also showed that the effect of the membrane resistance in these cases was minor. The
deviation from linearity of the press results suggested that the presses were beginning
to enter cake compression towards the end of the runs.

40000
t F = 34500s t F = 37740s
35000
Press 2 results
VF2 = 0.0091 m2

Press 1 results Analytical model


30000
VF2 = 0.0088 m 2

Analytical model Numerical model


Numerical model
Time, t (s)

25000

20000

15000

10000

5000 t L +t P = 3900s t L +t P = 3600s


t L = 1200s t L = 1200s
0
0 0.002 0.004 0.006 0.008 0.01 0 0.002 0.004 0.006 0.008 0.01
2 2 2 2 2 2
(Specific Filtrate Volume) , V (m ) (Specific Filtrate Volume) , V (m )

Figure 5.2.25: Experimental results and modelling predictions for Hodder WTP fill-only presses

276
Throughput Options and Recommendations

Impact of AMP2

The AMP2 project was close to completion at the time of this case study. It
was expected that the extra thickening would increase the sludge quality from 1-2 to
3–6 wt%. As discussed in depth in Section 5.3.1, the fill-only plate-and-frame model
predicted the effect of varying the initial solids concentration on filtration time and
throughput, depending on the desired final solids concentration. As shown in Figure
5.3.4, increasing from 2 to 4 wt% would either double the throughput at 20 wt%
(which assumed continuous operation) or increase the cake solids to 30 wt% without
reducing throughput. Thus, if the impact of AMP2 was an increase in press
throughput by 100% to roughly 4.32 t/day, the presses could have handled all
conditions except average and maximum throughput at worst water quality.

Impact of AMP3

The AMP3 project, which was due to begin construction in 2002, aimed to
introduce an extra clarifying step that would increase the amount of sludge to be
processed due to increased coagulant and flocculant dosing for clarifiers. Combined
with an anticipated long-term deterioration of raw water quality, the sludge
production was to expected significantly increase. The required solids throughput for
a TSS increase of 50% is given in Table 5.2.4.

Table 5.2.4: Required solids throughput at increased coagulation and flocculation dosage

Raw water throughput Solids throughput


(ML/day) (tds/day)

Average water quality Worst water quality

Current - 65 3.41 6.34

Average - 80 4.20 7.80

Maximum - 105 5.51 10.24

Thus, the current presses would not have been able to cope with the sludge
production whenever the water quality was bad and would keep the plant throughput

277
below 105 ML/day at average water quality, even if AMP2 was successful (that is, the
presses were fed at 4 wt%). Therefore, the current presses needed to be either
replaced or refurbished.

Press Refurbishment

One avenue for improvement was to refurbish the current presses. Several
issues were considered with regards to remedial works:

- Incorporation of a squeeze phase would have given compression that was


more efficient and allowed optimisation of the throughput without the loss of
final solids concentration;

- A fast fill pumping process to load and pressurise the press would have
reduced the cycle time by 30 to 60 minutes, a small benefit for current
operation but significant when combined with a squeeze phase;

- New plates were necessary, since the old plates were worn and easily fouled
with silt. A lighter material such as plastic would have allowed easier
movement of the plates. A thinner cavity would have reduced the average
filtration time (see New Press design issues);

- Installation of an automated unloading mechanism to improve unloading times


and reduce manning hours (this was done previously, but the old plates slipped
and the mechanism failed); and

- Installation of an automated washing system, including filtrate measurement to


determine Rm.

These improvements would have allowed an increase in throughput, but it was


unclear whether such presses would have been able to handle the required load. The
effect of each upgrade is unknown. If AMP2 worked very well and the presses ran at
maximum throughput, they should have been able to cope with all required loads.
However, the key issue here was that the presses were already old (installed in the
summer of 1969), and such refurbishment would only have extended their operational
lifetime by about ten years.

278
New Presses

The final option investigated for improving the sludge processing at Hodder,
and the main recommendation of this work, was investing in new presses. Should the
right presses be installed, the sludge processing would be adequate for the lifetime of
the presses.

Plate-and-frame presses were inherited by the water treatment industry from


the minerals processing industry, which traditionally have much easier materials to
dewater. However, presses should be designed to suit the material. Therefore, certain
design features would be discussed with manufacturers before buying new presses,
including:

- Use a fill-and-squeeze press rather than a fill-only press, since they are more
efficient and allow optimisation of throughput without loss of solids
concentration;

- Automation, such that the press can be run continuously without the need for
shifts, and the throughput can be maximised. Automation also reduces the
requirements of the operator;

- The position and number of feed inlets, such that the minimum amount of
lateral compression is seen during the fill phase. An uneven distribution of
solids also disrupts the squeeze phase, making the compression inefficient and
resulting in an uneven cake;

- Cavity width, which should be designed to meet the material and cycle
requirements (see Section 5.3.1). By decreasing the width, the filtration time
required to meet a certain final solids concentration reduces by a squared
relationship. If the number of plates remains constant, then the volume of the
press, and therefore the throughput per cycle, will decrease linearly with the
cavity width. For example, halving the cavity width gives a quarter of the
filtration time and halves the press volume, therefore doubling the daily
throughput (assuming there is no handling time and the presses are run
continuously).

279
The required press dimensions were calculated. If two new presses were
required to have capacities to handle 80 ML/day raw water throughput at average
water quality each, or 105 ML/day at worst water quality combined, each press would
be required to process approximately 4.2 tds/day. The fill-and-squeeze results for
alum sludges give a maximum solids throughput of 0.6 kg/hr/m2 (see Figure 5.3.23)
for an initial concentration of 4 wt%, where the optimum cycle is 3 to 4 hours.
Therefore, each press would require an area of approximately 300 m2. If the plate
dimensions remained the same, these presses would require in excess of 100 plates.

Outcomes

The results from the investigation showed that the current presses were
operating at close to their maximum throughput and were a bottleneck to the process
such that raw water throughput was limited to about 65 ML/day at current water
quality. Little improvement could be made to their throughput based upon changes to
operating procedures. The increased initial solids concentration caused by AMP2
would have increased the press throughput such that the raw water throughput may
have increased to 105 ML/day at average water quality if short runs could have been
implemented (since maximum throughput requires shift work). The presses would
still have been unlikely to cope with the required load at bad water quality. However,
the increased suspended solids due to AMP3 and the expected deterioration of raw
water quality would have kept the presses as the bottleneck of the process even if
operated at their maximum capacity.

Costly refurbishment would have extended the operating lives of the existing
plate-and-frame presses for a few years. They required new plates, perhaps with a
thinner cavity, and included incorporation of a squeeze phase following the fill phase.
Other possibilities included automated unloading and washing.

The main recommendation of this investigation was to purchase new fill-and-


squeeze presses that could cope with the sludge produced at maximum throughput and
worst water quality. Rough estimates of press dimensions suggested two presses with
100 plates each. Some design and control issues were discussed, such as cavity width
and feed inlet position and number.

280
5.2.3 Hodder Water Treatment Plant (Pilot-Scale Trials)

Background

As a consequence of the Hodder WTP case study, press manufacturer Baker-


Hughes performed two sets of on-site trials of five and six runs respectively in July
and October 2002 using a portable pilot-scale filter press as part of the tender process
for new presses. The rig was an R&B Model EH 6363 M/I Filter Press with four
chambers. The plates were manufactured from PPH (ultra high density
polypropylene) with detachable membranes. The plates were of top centre feed
design with four corner filtrate ports. The normal chamber width pre-squeeze was 30
mm, with an approximate cake volume (pre-squeeze) of 5.55 L/chamber, and a
filtration area of 0.43 m2/chamber.

For each trial, the volume of filtrate and pressure were measured with time.
By measuring the filtrate directly, the transient behaviour during the squeeze stage
was assessed. The feed and cake solids concentrations were determined by oven
drying. The solids density was assumed to be 3000 kg/m3.

Figure 5.2.26: Pilot-scale plate-and-frame filter press (Baker-Hughes)

281
Pilot-Scale Trial Results

The results from the pilot-scale trials were made available for tender
evaluation and to validate the flexible-membrane filter press models. A summary of
the eleven test rig trials is given Table 5.2.5.

Table 5.2.5: Hodder WTP pilot-scale trials by Baker-Hughes

Trial
φ0 ∆ PF tF ∆ PS tS φS
(v/v) (bar)* (s) (bar)+ (s) (v/v)

1.1 0.0091 7.0 3600 9.0 3600 0.1148

1.2 0.0117 7.0 3600 9.1 3600 0.1173

1.3 0.0068 7.0 2700 9.2 3000 0.1163

1.4 0.0141 7.0 4500 9.4 3900 0.1148

1.5 0.0099 7.0 7200 9.7 3600 0.1158

2.1 0.0099 7.5 4500 10 3600 0.1143

2.2 0.0099 7.5 5400 10 3600 0.1148

2.3 0.1020 7.5 6300 10 3600 0.1138

2.4 0.0099 7.5 9000 10 3600 0.1158

2.5 0.0106 7.5 10800 10 3600 0.1148

2.6 0.0099 7.5 6300 15 3600 0.1356


*
∆PF was taken as the final pressure reached during the fill phase. The first set of trials took
approximately 6 minutes for the pressure to begin to rise, 10 minutes to begin extruding filtrate, and 35
minutes to reach fill pressure. The second set lacked this data and showed immediate filtrate extrusion
and pressure rise, taking about 30 minutes to reach operating pressure
+
∆PF for set 1 was taken as average pressure over the entire squeeze phase, although the nominal
pressure was 10 bar. For set 2, the squeeze pressure was taken as the nominal squeeze pressure (10 or
15 bar).

Out of set one, trial 1.1 and 1.2 had similar fill and squeeze conditions, with an
increased feed concentration for 1.2. Trial 1.3 had shorter fill and squeeze times and a
low initial concentration, whereas trial 1.4 had slightly longer fill and squeeze times
and a high initial concentration. Trial 1.5 had the longest fill time and was at roughly

282
average feed concentration. The squeeze pressure also varied during and between
runs. In general, many variables were changed at once in set 1.

In set two, the squeeze time was kept constant and the feed concentration
varied as little as possible in order to investigate the changes due to different fill times
(trials 2.1 to 2.5). Trial 2.6 had the same fill time as 2.3 but an increased squeeze
pressure.

The method of cake sampling was unknown, and the results may have been
erroneous unless an entire cross-section of the cake was sampled. Despite a range of
fill and squeeze times, squeeze pressures and feed concentrations, all trials with a
nominal 10 bar squeeze reached similar final cake concentrations (average value of
0.1153 v/v, or 28.1 wt%), suggesting that they were all approaching the
compressibility limit of the material. The increased squeeze pressure for trial 2.6
produced a thicker cake at 0.1356 v/v (32.0 wt%). It was unknown whether the
polymer dosing level was changed according to the solids concentration or just flow
rate, which may have led to small variations in compressibility between samples with
different initial solids concentrations.

The volume of filtrate was measured approximately every 5 minutes. The


cumulative volume was converted to specific volume, V, by dividing by Apress = 4 x
0.43 = 1.72 m2. The t versus V2 results for sets one and two are presented in Figure
5.2.27 and Figure 5.2.28 respectively.

The fill-stage results show some of the expected characteristics:

- No filtrate was exuded while the press was loaded with sludge (around 10
minutes for trial 1). Peculiarly, the pressure began to rise before filtrate was
exuded for these runs. For set 2, the pressure increased immediately – there
was no indication of the time required to load the press. This implies that
there was an error in the measurement of V for set 2;

- The loading phase was followed by a curved portion (up to approximately 35


minutes for set 1 and 10 minutes for set 2) as the pressure increased to ∆PF.
All the pressure profiles for set 2 were exactly the same, suggesting the
pressure was measured once;

283
12000
1.1 Fill 1.1 Sq
1.2 Fill 1.2 Sq
1.3 Fill 1.3 Sq
10000 1.4 Fill 1.4 Sq
1.5 Fill 1.5 Sq

8000
Time, t (s)

6000

4000

2000

0
0 0.0005 0.001 0.0015 0.002 0.0025
2 2 2
(Specific Filtrate Volume) , V (m )

Figure 5.2.27: t versus V2 results for Hodder WTP fill-and-squeeze trial set 1

16000
2.1 Fill 2.1 Sq
14000 2.2 Fill 2.2 Sq
2.3 Fill 2.3 Sq
2.4 Fill 2.4 Sq
12000 2.5 Fill 2.5 Sq
2.6 Fill 2.6 Sq
Time, t (s)

10000

8000

6000

4000

2000

0
0 0.0002 0.0004 0.0006 0.0008
2 2 2
(Specific Filtrate Volume) , V (m )

Figure 5.2.28: t versus V2 results for Hodder WTP fill-and-squeeze trial set 2

284
- At constant pressure, t versus V2 was linear, corresponding to normal cake
formation. Curiously, this was observed before operating pressure was
reached; and

- In many cases, the results show a non-linear section at the end of the fill-stage,
corresponding to cake compression. In the case of Trial 1.1, the fill-stage
entered non-linear compression before the fill pressure was reached. Trials 2.4
and 2.5 entered compression earlier than the other runs in set 2, despite similar
conditions. In general, all the curves entered compression relatively early,
possibly due to the small plates (630 x 630 mm) used and the top centre feed
design.

By measuring the filtrate directly, as opposed to indirectly by counting the


strokes of a displacement pump, the transient behaviour during the squeeze stage was
examined. The squeeze stage results show initial linear portions (sometimes very
short), representing cake formation, followed by non-linear sections signifying cake
compression. All the trials reached cake compression towards the end of the squeeze
phase, which, along with the consistent cake concentration results, suggested that the
trials were approaching the compressional limit of the material.

Theoretically, dt/dV2 is constant during cake formation and dependent on ∆PF


and φ0. The trend in set 1 during both stages was expected to be in order of initial
solids - (1.4, 1.2, 1.5, 1.1, 1.3) rather than (1.4, 1.5, 1.2, 1.3, 1.1). The trend was
consistent across both the fill and squeeze phases. Generally, set 2 had much slower
fill and squeeze results than set 1. The values should have been constant for the same
φ0 and ∆P, but showed some variation.

Theoretically, the squeeze phase should have had a lower slope than the feed
phase due to the increased applied pressure. In most cases, the slope for the squeeze
phase increased rather than decreased. Runs 2.4 and 2.5 were exceptions, while the
slope did not decrease much but remained almost constant for runs 1.5 and 2.6.

The accuracy of the filtrate volume and cake solids results were examined
using a volumetric balance over the filter press. The average volume fraction within
the cake, <φ>, is given by equation 5.1.3. Table 5.2.6 gives a summary of the

285
measured φ0, φS, VF and VS, and φ0 or φS as calculated from equation 5.1.3, using φS,
VF and VS or φ0, VF and VS, respectively. There was a significant variation between
the measured and calculated values, especially for set 2, indicating that there was a
large error in measurement of Apress, V, φ0 or φS. The feed was sampled from a batch
thickening tank, thus φ0 may have varied throughout a trial. Likewise, φS may not
have been representative of the entire cake.

Table 5.2.6: Measured versus calculated final solids concentrations for Hodder WTP fill-and-
squeeze trials

Measured φ0 Measured φS VF VS Calculated φ0 Calculated φS


Trial
(v/v) (v/v) (m) (m) (v/v) (v/v)

1.1 0.0091 0.1148 0.0401 0.0483 0.0143 0.0728

1.2 0.0117 0.1173 0.0306 0.0386 0.0180 0.0768

1.3 0.0068 0.1163 0.0262 0.0361 0.0143 0.0555

1.4 0.0141 0.1148 0.0247 0.0313 0.0244 0.0663

1.5 0.0099 0.1158 0.0345 0.0414 0.0189 0.0606

2.1 0.0099 0.1143 0.0232 0.0254 0.0383 0.0294

2.2 0.0099 0.1148 0.0240 0.0262 0.0376 0.0301

2.3 0.1020 0.1138 0.0256 0.0278 0.0359 0.0323

2.4 0.0099 0.1158 0.0249 0.0269 0.0378 0.0302

2.5 0.0106 0.1148 0.0253 0.0273 0.0370 0.0327

2.6 0.0099 0.1356 0.0250 0.0273 0.0430 0.0311

Rm was calculated from the slope of dt/dV2 versus 1/V. The method was
somewhat subjective due to the limited number of data points. The results for set one
(see Figure 5.2.29) showed fairly constant Rm, with an average of (4.88 ± 1.95) x 1010
Pas/m, suggesting a small influence due to the resistance. The dt/dV2 versus 1/V
results showed increasing gradients at high 1/V, corresponding to ramping pressure,
and negative gradients at low 1/V, indicating cake compression. Trials 1.4 and 1.5
exhibited strange behaviour as the fill-stage progressed, with the data passing through
a maximum after the onset of compression.

286
1.E+07
1.1 Fill
1.2 Fill
1.3 Fill
8.E+06 1.4 Fill
1.5 Fill

4.92x1010
dt /dV (sm )
-2

6.E+06
2

4.E+06 3.56x1010
6.54x1010 4.42x1010

2.E+06
4.95x1010

0.E+00
0 50 100 150
-1
1/V (m )

Figure 5.2.29: dt/dV2 versus 1/V for Hodder WTP fill-and-squeeze trial set 1

2.0E+07
2.1 Fill
2.2 Fill
2.3 Fill
1.6E+07 2.4 Fill
2.5 Fill
2.6 Fill
dt /dV (sm )
-2

1.2E+07 1.20x1011
1.39x1011
2

8.0E+06 2.62x1011
1.83x1011
2.10x1011
4.0E+06 2.62x1011

0.0E+00
0 20 40 60 80 100
-1
1/V (m )

Figure 5.2.30: dt/dV2 versus 1/V for Hodder WTP fill-and-squeeze trial set 2

287
The dt/dV2 versus 1/V results for set two are shown in Figure 5.2.30. In
general, the Rm values were high (1.68 ± 1.04) x 1011 Pas/m, suggesting that the
membrane exerted considerable influence on the filtration rate. The results for set two
showed strange behaviour at high 1/V, such that the gradient was negative. This
would have been possible only if the membrane was fouling drastically during the
early stages of filtration, since the filtration rate was highest at the lowest pressures.
This further suggested that the experimental results for set two were erroneous.

Flexible-Membrane Model Validation

The average Hodder WTP material characteristics (see Section 4.1.2) were
used for modelling purposes. Py(φ) was extrapolated to operating pressures using the
power-law function, giving the compressibility of the material. At 10 bar, φ∞,S =
0.0964 v/v while at 15 bar, φ∞,S = 0.1052 v/v. The high experimental results
suggested that either the compressibility results did not extrapolate based on the
power-law functional form, or that the material compressibility had changed. In
either case, the trials appeared to have reached their compressibility limit.

The material characteristics were used in the analytical and numerical models
along with the operating conditions (φ0, h0, P(t), Rm, tL, tP, tF and tS) to predict the
press performance in order to validate the models. The experimental results and
model predictions for trial set one (see Figure 5.2.31) showed that both models
predicted slower filtrate rates for all runs, especially Trials 1.1 and 1.2. The
numerical model consistently under predicted the performance compared to the
analytical model, reflecting the membrane resistance effect. The largest resistance
(Trial 1.3) had a significant effect compared to the lowest resistance (Trial 1.5).

The squeeze predictions exhibited similar characteristics to the experimental


results – cake formation followed by cake compression. The model predicted that
dt/dV2 decreased at the start of the squeeze phase since the applied pressure increased,
while the majority of the trials showed increasing dt/dV2. The model predictions for
the gradient during the squeeze phase for trials 1.3 and 1.5 were very close to the
experimental results. The numerical model predicted that the membrane resistance
sometimes had an effect at the start of the squeeze phase, especially for the cases
where tF and the corresponding cake resistance were small.

288
8000 8000
t F +t S VS2 t F +t S VS2
7000 7000

6000 6000
Time, t (s)

Time, t (s)
5000 5000

4000 tF VF2 4000 tF VF2

3000 3000
t L +t P t L +t P
2000 2000
1.1 Fill 1.2 Fill
1000 1.1 Squeeze 1000 1.2 Squeeze
Analytical model tP Analytical model tP
Numerical model Numerical model
0 0
0 0.0005 0.001 0.0015 0.002 0.0025 0 0.0004 0.0008 0.0012 0.0016
6000 9000
t F +t S VS2
t F +t S VS2 8000
5000
7000

4000 6000
Time, t (s)

Time, t (s)

5000 tF VF2
3000 tF
2 4000
VF
t L +t P
2000 3000
t L +t P
2000
1000 1.3 Fill 1.4 Fill
1.3 Squeeze 1.4 Squeeze
Analytical model tP 1000 Analytical model tP
Numerical model Numerical model
0 0
0 0.0005 0.001 0.0015 0 0.0002 0.0004 0.0006 0.0008 0.001
12000
t F +t S VS2

10000

8000
tF
Time, t (s)

VF2
6000

4000

t L +t P
2000 1.5 Fill
1.5 Squeeze
Analytical model tP
Numerical model
0
0 0.0005 0.001 0.0015 0.002
2 2
(Specific Filtrate Volume) , V (m2)

Figure 5.2.31: Experimental and modelling results for Hodder WTP fill-and-squeeze trail set 1

For set 2 (see Figure 5.2.32), the model did not predict the experimental
results successfully. The analytical model generally over predicted the press
performance, while the numerical model under predicted the performance. The

289
squeeze phase results for set 2 all showed very similar behaviour at the same initial
solids, with a corresponding decrease with increased pressure.

9000 10000
VS2 t F +t S VS2 t F +t S
8000 9000

7000 8000

7000
6000
Time, t (s)

6000
5000 VF2 tF
VF2 tF
5000
4000
4000
3000
3000
2000 t L +t P t L +t P
2.1 Fill 2000 2.2 Fill
2.1 Squeeze 2.2 Squeeze
1000 Analytical model 1000 Analytical model
Numerical model Numerical model
0 0
0 0.0005 0.001 0.0015 0 0.0004 0.0008 0.0012 0.0016
12000 14000
VS2 t F +t S
2
VS t F +t S 12000
10000

10000
8000 VF2 tF
Time, t (s)

VF2 tF 8000
6000
6000
4000
4000

2000 2.3 Fill t L +t P 2.4 Fill t L +t P


2.3 Squeeze 2000 2.4 Squeeze
Analytical model Analytical model
Numerical model Numerical model
0 0
0 0.0005 0.001 0.0015 0.002 0 0.0005 0.001 0.0015 0.002 0.0025
16000 12000
VS2 t F +t S
14000 VS2 t F +t S
10000
12000
VF2 tF
8000
Time, t (s)

10000
VF2 tF
8000 6000

6000
4000
4000
2.5 Fill 2000 2.6 Fill t L +t P
2.5 Squeeze t L +t P 2.6 Squeeze
2000 Analytical model Analytical model
Numerical model Numerical model
0 0
0 0.0005 0.001 0.0015 0.002 0.0025 0 0.0005 0.001 0.0015 0.002
2 2 2 2 2
(Specific Filtrate Volume) , V (m ) (Specific Filtrate Volume) , V (m2)

Figure 5.2.32: Experimental and modelling results for Hodder WTP fill-and-squeeze trail set 2

A summary of the experimental results and modelling predictions is presented


in Table 5.2.7, which compares the measured and predicted cake concentrations and

290
throughputs. The models consistently under predicted the final cake solids, due to the
differences in the compressive yield stress, and did not always reach the
compressibility limit. This was especially the case for the numerical model for set 2,
due to the large predicted influence of the membrane resistance.

Table 5.2.7: Comparison of experimental and modelling results

Trial
φS (expt.) φS (anal.) φS (num.) Qp (expt.) Qp (anal.) Qp (num.)
(v/v) (v/v) (v/v) (kg/m2/hr) (kg/m2/hr) (kg/m2/hr)

1.1 0.1148 0.0913 0.0940 0.4994 0.3401 0.2925

1.2 0.1173 0.0903 0.0927 0.5356 0.4029 0.3585

1.3 0.1163 0.0920 0.0946 0.3263 0.2913 0.2187

1.4 0.1148 0.0895 0.0909 0.5038 0.4481 0.4036

1.5 0.1158 0.0865 0.0861 0.3675 0.3558 0.3322

2.1 0.1143 0.0921 0.0551 0.3476 0.3801 0.2257

2.2 0.1148 0.0904 0.0774 0.3296 0.3767 0.2592

2.3 0.1138 0.0880 0.0666 0.3313 0.3797 0.2572

2.4 0.1158 0.0839 0.0728 0.2625 0.3538 0.2834

2.5 0.1148 0.0820 0.0724 0.2554 0.3551 0.3022

2.6 0.1356 0.1021 0.1043 0.3154 0.3718 0.3097

The model gave throughput results within 25% of the measured values, and
often much closer. For set one, the analytical and numerical models under predicted
the throughput by 16% and 27% respectively, while for set two, the analytical model
over predicted the throughput for by 22% and the numerical model under-predicted
the throughput by 9%. Overall, these results were reasonably close, and on this basis,
the model was considered partly successful.

The dramatic difference between the two sets indicated a systematic change in
operating conditions that was not accounted for by the models. The differences
between the model and trial results may have arisen from changed material
characteristics (due to different levels of dosing or yearly fluctuations), experimental

291
errors (initial solids, filtrate volume or membrane area measurements) or modelling
assumptions (such as the onset of non-linear compression or constant membrane
resistance).

Outcomes

Baker-Hughes conducted two sets of trials (eleven runs in total) with a pilot-
scale fill-and-squeeze plate-and-frame filter press at Hodder WTP. The results were
analysed and compared to model predictions. The trial results appeared to contain
significant errors, in particular the V, φ0 and φS measurements looked poor and
inconsistent. In general, the predictions varied from the model results. Overall, the
trial results did not validate the model, although this was because of poor results
rather than invalid model assumptions.

292
5.2.4 Thornton Steward Water Treatment Plant

Background

An investigation was carried out at Thornton Steward WTP in June 2002 to


provide options for increasing the throughput of the filtration stage. Thornton
Steward WTP is an alum water treatment plant situated in the North Yorkshire Dales
operated by Yorkshire Water. The colour of the raw water was approximately 35
colour units and the turbidity was usually between 1.05 and 1.4 N.T.U. However, the
raw water was subject to dramatic changes in character due to seasonal algal blooms.
The sludge produced by the treatment of the raw water was dosed with polymer
flocculant (Posifloc PW80), underwent batch settling, and was then filtered by a fill-
only plate-and-frame press (see Figure 5.2.33).

Figure 5.2.33: Fill-only plate-and-frame filter press at Thornton Steward WTP

During normal operation (without algal blooms), at a maximum raw water


flow of 22.7 ML/day, the press was required to process 2100 kg of dry solids per day
(if operated 5 days a week). If weekends were worked, this fell to 1500 kg ds/day.
During algal events, the required throughput increased to 2800 kg ds/day (2100 kg

293
ds/day if weekends were worked), and the character of the sludge was expected to
change. These throughput values were used as the basis for the recommendations of
this investigation.

The aim of this investigation was to consider options for increasing the
throughput of the works. General observations of pre-filtration options were made
during a site tour on 28/6/02. The performance of one press run on 20/6/02 was
monitored over time by counting the number of strokes of the 6” Willet ram pump
feeding the press and recording the pressure as shown on the pressure gauge on the
press feed. Two samples were taken from the bleed valve of the displacement pump,
one at the start of the run and one half way through the run. The solids concentration
was determined by oven drying. The material characteristics were determined using
an air driven filtration rig following the stepped-pressure technique. The gel point
was determined using equilibrium settling tests. Three samples of the filter cake were
taken and the solids concentration determined by oven drying.

Two press manufacturers, Lathams and Baker-Hughes, had put tenders


forward for press refurbishment or a new press. Using the plate-and-frame filter press
models, the estimated throughputs were calculated in order to evaluate the tenders.

Results

Pre-Filtration Performance

Three issues that were considered important to filtration performance arose


from viewing the pre-filtration treatment:

- The thickener was operated in a batch manner. This meant that when the tank
was emptying, the sludge was at 1.5 - 2 wt%, even if sludge at the bottom had
been thickened to 4 wt% or greater;

- Polyelectrolyte was added once, before the thickening operations. There was
no poly dose in between thickening and filtration; and

- Related to this, the sludge passed through at least two pumps prior to the press,
subjecting the flocs to high shear and probably causing degradation of floc

294
structure. This would have had a dramatic effect on the filterability of the
material.

All of these effects would have reduced the throughput of the filtration stage.

Feed Analysis

Two samples were taken from the bleed valve of the positive displacement
pump, at the beginning of the filtration run and after 4 hours. The average
concentration was determined by oven drying as 0.00633 v/v (1.88 wt%). φg was
determined from equilibrium settling tests to be 0.0047 v/v (1.40 wt%). The material
characteristics were determined, but, due to difficulties with the rig, the
compressibility results contained a large degree of error and model validation was not
possible. Instead, the average alum characteristics were used for prediction purposes.

The material characterisation was performed on sludge without an algal event.


During algal events, the material character was expected to change and the total
amount of sludge would increase.

Fill-Only Press Performance

The press was a fill-only type press with 80 cavities. The membranes were
1200 x 1200 mm, giving a total filtration area of 225 m2 and a press volume of 3.31
m3/run. The volume of each stroke was 3.7 L.

The number of strokes per minute and the applied pressure were measured
over the entire run. The results are presented in Figure 5.2.34. The press took
approximately 25 minutes to fill with sludge, then under 5 minutes to reach an
operating pressure of 6.5 bar. This quick pressure rise and the near constant flowrate
indicated that poor membrane permeability was dominating the filtration behaviour.
The total cycle time was 24 hours.

The stroke count results were converted to V, assuming a stroke volume of


3.7L. Figure 5.2.35 presents the results as a plot of t versus V2. The large gap in the
results was due to the press running overnight, although in that time the stroke count
per minute only changed from 4.32 to 4.26.

295
10 2.5

tL
tL + tP
Pressure

Applied Pressure, P (t ) (bar)


Flowrate

Feed Flowrate, q( t) (L/s)


8 2

6 1.5

4 1

2 0.5

0 0
0 1000 2000 3000 4000
Time, t (s)

Figure 5.2.34: Feed flowrate and pressure results for Thornton Steward WTP fill-only press,
20/06/02

90000

80000
VF2 = 0.0010 m2

70000

60000
Time, t (s)

50000

40000

30000

20000

10000
tL + tP
0
0 0.004 0.008 0.012

(Specific Filtrate Volume)2, V 2 (m2)

Figure 5.2.35: t versus V2 results for Thornton Steward WTP fill-only press, 20/06/02

296
The results were expected to show linearity during constant pressure filtration,
since the rate of filtration is normally governed by the permeability of a growing cake.
The results presented in Figure 5.2.35 show that this was not the case (the relationship
was closer to linear), suggesting that the permeability of the filter cloths was
significant and hindered the filtration process.

Figure 5.2.36 shows a plot of dt/dV2 versus 1/V. The slope of a linear fit of the
data was used to give Rm = 3.70 x 1011 Pas/m, according to the modified Darcian
approach outlined in Section 2.5.2. The value was high, but not exceedingly so.
Visual inspection of the cloths showed that they were blackened by silt build-up.
Under current operations, steam washing was rarely performed since it took two days
- the cloths had not been washed for two years. Therefore, Rm was expected to be
even higher.

1.2E+08

1.0E+08

8.0E+07
dt /dV (sm )
-2

6.0E+07
2

4.0E+07
R m = 3.70x1011 Pas/m

2.0E+07

0.0E+00
0 100 200 300 400
-1
1/V (m )

Figure 5.2.36: dt/dV2 versus 1/V for Thornton Steward WTP fill-only press, 20/06/02

Filter Cake Analysis

After twenty-four hours of filtration, the filter cakes were generally of low
concentration with significant variations across each cake and from cake to cake. The

297
cake, especially towards the centre of the plates, tended to slump off the membrane
and splatter in the awaiting bin rather than separate neatly from the membrane and fall
and shatter. This made sampling difficult; therefore, the cake concentration results
were used as a guide only. It was also observed that the sludge in the central feed
bore showed some thickening, suggesting that the sludge was not being transported
into the cavity effectively, a further indication that the press was running inefficiently.

Two samples were taken from a relatively ‘dry’ cake, one from near the
middle and the other from near the edge. The third sample was taken from near the
edge of a relatively ‘wet’ cake. The results are presented in Table 5.2.8, showing that
the current operation was failing badly – the maximum measured concentration was
12.2 wt%, whereas the desired final cake concentration was above 20 wt%.

Table 5.2.8: Filter cake sample analysis from Thornton Steward WTP fill-only press, 20/06/02

Cake Sample wt% φF (v/v)

‘dry’ cake, near edge 12.20 0.0443

‘dry’ cake, near middle 6.09 0.0211

‘wet’ cake, near edge 8.32 0.0294

Based on the measured VF and press dimensions, the final cake concentration
should have been 0.0485 v/v. Thus, the actual delivered stroke volume was less than
3.7 L/stroke. The total throughput of sludge was estimated from the stroke count to
be 0.090 kg/hr/m2 (= 488 kg ds/day), a third of the desired throughput. The
throughput of fill-only presses is directly related to the final cake solids, such that an
improvement of throughput at Thornton Steward WTP under the current operating
procedures would have required a further decrease in cake solids (see Figure 5.2.37),
which was already well below the desired concentration.

298
0.24

Solids Throughput, Qp (kg/hr/m )


2
0.2

0.16

0.12

0.08 φ0 = 0.0063 v/v

φF = 0.0485 v/v
0.04

0
0 0.01 0.02 0.03 0.04 0.05

Average Cake Solids, <φ > (v/v)

Figure 5.2.37: Solids throughput versus cake solids for Thornton Steward WTP fill-only press,
20/06/02

Filtration Options

Three options were considered for the improvement of the filter press. The
first contemplated the current press with clean cloths and improved feed
concentration. The second option required refurbishment of the press, while the third
dealt with new presses. Tenders were received from two press manufacturers, Baker
Hughes and Lathams, for the latter two options. The average alum sludge
characteristics were used in conjunction with the plate-and-frame models to predict
the outcome of such changes. It was also suggested that an extra polyelectrolyte dose
prior to filtration would have had a dramatic effect on the material characteristics. As
such, the press performances were recalculated assuming an improved D(φ).

Extrapolating the average alum characteristics to the operating pressures gave


φ∞,F = 0.1262 v/v (30.2 wt%) at 7 bar, φ∞,F = 0.1399 v/v (32.8 wt%) at 10 bar and φ∞,F
= 0.1573 v/v (35.9 wt%) at 15 bar. These results indicated that Baker Hughes and
Lathams predictions of cake solids of 30 and 25 wt% respectively were possible.

299
Existing press (7 bar, fill-only, new cloths and regular washing)

The results from the stroke count showed that the current press had problems
with cloth blinding. As well as this, it was observed that the cake concentration
improved where some of the old cloths had been replaced. If new cloths were
installed and a regular washing routine established, the overall press performance
would improve.

Model predictions were made based upon the average alum material
characteristics at two feed concentrations and with and without membrane resistance.
The t versus V2 results are presented in Figure 5.2.38, showing that the site data was
significantly slower than the average alum characteristics, even at the improved
concentration.

90000

80000

70000

60000
Time, t (s)

50000

40000

30000

20000 Site data


Rm, 1.88wt%
10000 No Rm, 1.88wt%
No Rm, 4wt%
0
0 0.01 0.02 0.03 0.04 0.05
2 2 2
(Specific Filtrate Volume) , V (m )

Figure 5.2.38: t versus V2 results and model predictions for Thornton Steward WTP fill-only
press

This difference between the average alum sludge and the Thornton Steward
WTP sludge could have been due to floc degradation or the presence of algal blooms
in the raw water. The model predictions at 1.88wt% show only minor effect due

300
membrane resistance, suggesting that the value of Rm given by the modified Darcy’s
law was too low, since all other indications were that the resistance was significant.

At 1.88wt% with no Rm, the model predicted that a 23 hour fill (1 press per
day, allowing for 1 hour handling time) would reach 23.9 wt% and have a throughput
of 673 kg ds/day, much less than the desired value of 2100 kg ds/day. With the
implementation of extra thickening, such that the feed was at 4 wt%, the model
predicted that filtration of 11 hours reached a cake concentration of 24.2 wt%. The
throughput per cycle was 593 kg ds. If shift work were available such that two runs
per day were possible, the throughput would be 1186 kg ds/day. This was still much
less than the required amount, suggesting that the current press with clean cloths
would not have handled the required load, regardless of improvements to thickening.

Refurbished press (7 bar fill, 7 bar squeeze)

Tenders were put forward investigating using the existing frame and
refurbishing the press to a fill-and-squeeze type. The fill and squeeze phases were
expected to be at 7 bar, since the frame was rated at 7 bar. The number and size of the
plates were limited by the size of the frame: 80 plates of 1200 x 1200 mm. The
assumption was made that the cloths would be washed regularly and that the cloth
permeability would be negligible.

The throughput was estimated for the following daily operating regimes:

- 2 cycles/day = 1 x 7½ hr, 1 x 16½ hr

- 3 cycles/day = 2 x 3½ hr, 1 x 17 hr

- 4 cycles/day = 3 x 2½ hr, 1 x 16½ hr

- 8 cycles/day = 8 x 3 hr

The first three regimes were based on an 8-hour working day, with short runs
during the day and one long run overnight. The last regime assumed 24-hour
manning. The optimum cycle time was about 3 hours. If shift work was employed,
the optimum regime was the last option, at 8 cycles/day. If presses were performed
over the weekend (7 days rather than 5 days per week), the daily capacity requirement

301
dropped from 2100 to 1500 kg ds / day (or from 2800 to 2000 kg ds/day during algal
events).

Lathams recommended a refurbished press with 76 chambers. The filtration


area was given as 218 m2 and the press volume as 3.298 m3. They suggested 3 cycles
per day, with two 3.5-hour cycles and one 17-hour cycle. Baker Hughes
recommended a refurbished press with 74 chambers. The filtration area was not
given; therefore, the membranes were assumed to be 1200 x 1200mm, giving a total
filtration area of 208 m2 and press volume of 3.06 m3. Baker Hughes recommended 4
cycles per day, each of 4 to 5 hours. This would have required shift work.

The fill-and-squeeze model was used to predict the performance of the


refurbished press with differing cycles times (the final concentration was fixed at
20wt%). Based upon these results, the throughput under different operating regimes
was predicted. The predictions for the Baker Hughes and Lathams refurbished
presses at feed concentrations of 1.88 and 4 wt% are given in Table 5.2.9.

Table 5.2.9: Throughput predictions for refurbished press at Thornton Steward WTP

Solids t’put Solids t’put


Manufacturer Operating regime (kg ds /day) (kg ds /day)
(1.88 wt%) (4 wt%)

2 cycles/day 913 1217

3 cycles/day 1001 1343


Baker Hughes
4 cycles/day 1052 1437

8 cycles/day 1553 2377

2 cycles/day 959 1282

3 cycles/day 1052 1414


Lathams
4 cycles/day 1105 1513

8 cycles/day 1631 2504

The model predicted that, at the current feed concentration, both designs of
refurbished presses would have failed by close to 50% of the desired throughput.
They would have just been able to process the expected throughput if operated

302
continuously, with shift work and working weekends. Even if the feed was thickened
to 4 wt%, the refurbished presses as specified by the manufacturers would have been
only able to cope with up to 72% of the desired throughput. If shift work was
employed such that the press ran continuously 5 days per week, both refurbished
presses could have coped with the normal load of 2100 kg ds/day. If weekends were
worked and the daily operating regime was 4 cycles/day, the Baker Hughes press
would nearly have handled the load (1500 kg ds/day), while the Lathams press would
have just coped. The presses could have coped with algal events if shift work and
weekends were employed (8 cycles/day, 7 days/week), assuming that the sludge
characteristics did not change to a significant extent.

New presses

The new press specifications had different squeeze pressures for each
manufacturer and were modelled separately. Baker Hughes recommended a new
press with 78 chambers, operating at ∆PF = 7 bar and ∆PS = 10 bar. The filtration
area was given as 288 m2 and the press volume as 4.397 m3. Lathams recommended a
new press with 60 chambers, operating at ∆PF = 7 bar and ∆PS = 15 bar. The
filtration area was given as 280.8 m2 and the press volume as 4.271 m3.

The model predictions of throughput for differing operating regimes are


shown in Table 5.2.10. The final concentration was fixed at 20 wt%. These results
show that, at the current feed concentration, the new press proposed by Baker Hughes
would not have handled the required load within normal working hours
(approximately 40% more plates would be required). The throughput increased
sufficiently if shift work was employed (for example, 4 cycles/day at 7 days/week; or
8 cycles/day at 5 days/week). If the feed concentration was increased significantly
and the press was run with 4 cycles/day, the proposed new press by Baker Hughes
would almost suffice. The recommendation in the tender of 2-3 cycles/day would
have given a throughput of about 1800 kg ds/day.

The new press for Lathams was also modelled. In general, the additional
squeeze pressure reduced tS by 5 – 10 minutes, allowing a small increase in
throughput (up to 5% for the shorter cycles). The press predictions in Table 5.2.10
illustrate that, at 1.88wt%, the new press proposed by Lathams would not have

303
handled the required load within normal working hours (approximately 40% more
plates were required), and shift work was necessary (4 cycles/day at 7 days/week or 8
cycles/day at 5 days/week). As for the Baker Hughes new press, if φ0 was increased
and the press was run with four cycles per day, the proposed new press by Lathams
would suffice. The tender’s recommendation of three cycles/day gave a solids
throughput of 1879 kg ds/day.

Table 5.2.10: Throughput predictions for new presses at Thornton Steward WTP

Solids t’put Solids t’put


Manufacturer Operating regime (kg ds /day) (kg ds /day)
(1.88 wt%) (4 wt%)

2 cycles/day 1277 1693

3 cycles/day 1411 1893


Baker Hughes
4 cycles/day 1493 2048

8 cycles/day 2229 3456

2 cycles/day 1252 1662

3 cycles/day 1394 1879


Lathams
4 cycles/day 1486 2054

8 cycles/day 2249 3484

Improved Flocculation

The calculations were repeated with improved solids diffusivity (double the
average alum D(φ)), as an estimation of improved flocculation. The model
predictions using this functional form for the refurbished presses are summarised in
Table 5.2.11, while the results for the new presses are shown in Table 5.2.12. The
results show that, even with improved permeability and increased feed concentration,
the refurbished designs would still have been unable to handle the required load
without resorting to shift work. Either new press would have been able to handle the
load at increased feed concentration. At 1.88 wt%, they would have been almost able
to cope, perhaps requiring the occasional Saturday. The results still suggest that

304
weekends would have to be worked during algal blooms (such that the load was 2000
kg ds/day rather than 2800 kg ds/day).

Table 5.2.11: Throughput predictions for refurbished press with increased D(φ) at Thornton
Steward WTP

Solids t’put Solids t’put


Manufacturer Operating regime (kg ds /day) (kg ds /day)
(1.88 wt%) (4 wt%)

2 cycles/day 1164 1297

3 cycles/day 1272 1558


Baker Hughes
4 cycles/day 1353 1685

8 cycles/day 2117 3234

2 cycles/day 1223 1366

3 cycles/day 1336 1641


Lathams
4 cycles/day 1421 1776

8 cycles/day 2224 3406

Table 5.2.12: Throughput predictions for new press with increased D(φ) at Thornton Steward
WTP

Solids t’put Solids t’put


Manufacturer Operating regime (kg ds /day) (kg ds /day)
(1.88 wt%) (4 wt%)

2 cycles/day 1622 1813

3 cycles/day 1783 2197


Baker Hughes
4 cycles/day 1908 2400

8 cycles/day 3018 4615

2 cycles/day 1590 1768

3 cycles/day 1757 2158


Lathams
4 cycles/day 1890 2376

8 cycles/day 3002 4578

305
Outcomes

The options for increasing the filter press throughput at Thornton Steward
WTP were investigated. Several improvements were outlined relating to the
thickening and polymer dosing prior to filtration, both of which affected the filter
feed, including running the thickener continuously rather than batch-wise, removing
high-shear pumps and excessive flow rates in between polymer dosing and filtration,
and including an extra polyelectrolyte dose prior to filtration.

The performance of the current press was monitored. It was unable to cope
with the required throughput and excess sludge was transported from site. The
performance was particularly restricted due to severe cloth blinding. Model
predictions of average alum sludge characteristics were used to show that replacing
the cloths and establishing a cloth-washing regime on the current press would not
have provided the extra throughput required. A regime for cloth replacement and
washing should have been established whichever option was followed.

Specifications were received from Lathams and Baker Hughes for the
refurbishment of the current press or for a new press. The filter press models were
used to give estimations of the capabilities of the specified presses. The results
showed that, in general, the refurbishment option would not have provided the desired
throughput due to limitations to the size and number of plates. If operations were
content to tanker untreated sludge from site or work nights and weekends during times
of increased throughput, this option may have been economic. A further disadvantage
of the refurbishment option was that the frame was already over 25 years old, and
would not have the expected lifetime of a new press.

A new press with large plates and increased squeeze pressure would have had
a greater throughput, although model predictions showed little to distinguish between
the tenders in performance terms. If φ0 were not improved, shift or weekend work
would have to be implemented during times of peak loading. If φ0 were improved,
both press designs would give close to the required throughput.

Some outstanding issues concerning material characterisation at Thornton


Steward WTP included:

306
- Material characterisation of the sludge without an algal event;

- Sample collection and analysis throughout polymer dosing, thickening and


transport to the press house to quantify the effect of floc degradation and give
insight into its prevention;

- Material characterisation of the sludge with extra polymer dosing to allow


more accurate prediction of its effect (rather than an assumed the change to
D(φ)); and

- Material characterisation of the sludge with algae to establish whether the


algae affected just the amount of sludge produced or actually changed the
material properties.

307
5.2.5 Arnfield Water Treatment Plant

Background

Lime was added to the ferric sludge at Arnfield WTP prior to dewatering with
a plate-and-frame press. The press at Arnfield WTP was a 1965 fill-only type press
with 80 plates that operated at a gauge pressure of 9.2 bar. The plates had an effective
membrane area of 1.4 m2/membrane, giving a total filtration area of 225 m2. The
cavity width was 3 cm and the press volume was 3.31 m3.

Cake samples were taken from a cycle in June 2002 to give an estimate of the
press performance. After two hours of filtration, φF was 0.0753 v/v (17.75 wt%).
Such short filtration times to reach medium cake solids suggested that the lime
addition aided the filtration performance. However, the cloths were cleaned
frequently and the feed concentration was high, which could have accounted for some
or all of this effect.

Feed samples were taken before and after lime addition. Prior to lime
addition, φ0 was 0.0122 v/v (3.58 wt%), while with lime, φ0 increased to 0.0131 v/v
(3.82 wt%). The characterisation results for one sample pre-lime and two samples
with lime are presented in Section 4.1.1, which shows an increase in D(φ) with added
lime. The material characteristics were used as inputs to the numerical fill-only
model to illustrate the effect of lime-addition on dewatering.

Fill-Only Model Predictions

The numerical fill-only model was used to predict the press performance using
the three different material characteristics. The results were compared on a
throughput basis in order to ascertain the benefits of lime addition, as opposed to just
high feed solids and clean cloths. The model inputs were D(φ), Py(φ), φ0, P(t), d and
A. It was assumed that tL = tP = 1200s and ∆PF = 9.2 bar. φ0 was 0.0122 v/v for the
pre-lime sample while φ0 was 0.0131 v/v for the two sets of characteristics with lime
added.

308
Extrapolating Py(φ) to ∆PF gave φ∞,F (pre-lime 14/06/02) = 0.1996 v/v (42.8
wt%), φ∞,F (post-lime 05/01/01) = 0.1728 v/v (38.5 wt%) and φ∞,F (post-lime
14/06/02) = 0.1858 v/v (40.6 wt%). Thus, the equilibrium limit for the cake solids
dropped slightly with the introduction of lime. However, since the cycle was not
proceeding to equilibrium, φF was dependent on the permeability rather than the
compressibility.

The model results of t versus V2 are presented in Figure 5.2.39. The material
with the largest D(φ) values (14/06/02 post-lime) filtered the quickest, even though
the initial solids were higher. The increased diffusivity overcame the detrimental
effect on filtrate rate of increased φ0 for the 05/01/01 sample compared to the
14/06/02 pre-lime sample. The model predicted that all the samples would enter cake
compression within 24 hours (86400s).

100000
05/01/01 post-lime
14/06/02 pre-lime
80000 14/06/02 post-lime
Time, t (s)

60000

40000

20000

0
0 0.01 0.02 0.03
2 2 2
(Specific Filtrate Volume) , V (m )

Figure 5.2.39: Fill-only model predictions of t versus V2 for Arnfield WTP

Figure 5.2.40 shows a plot of cake concentration with time as predicted by the
model, which illustrates that the sludge with the highest D(φ) reached higher cake
concentrations quicker. The predictions for the 14/06/02 sample with lime added

309
gave similar results to the observation that the cake concentration reached 17 wt%
after two hours. All samples reached high concentrations fairly quickly,
corresponding to the generally high feed solids concentration.

Average Cake Solids, <φ > (wt%) 40%

35%

30%

25%

20%

15%

10%
05/01/01 post-lime
5% 14/06/02 pre-lime
14/06/02 post-lime
0%
0 4 8 12 16 20 24
Time, t (hr)

Figure 5.2.40: Fill-only model predictions of average cake solids with time for Arnfield WTP

The performances of the different materials were compared on the basis of the
suspended solids throughput, which was the total throughput less the amount of added
lime. Likewise, the average cake solids were compared on the basis of the fraction of
suspended solids in the cake, since the addition of lime had to ensure higher total cake
solids to keep disposal costs the same. A plot of the average suspended throughput
versus adjusted cake solids is shown in Figure 5.2.41. Even though the sample from
05/01/01 had higher diffusivity than the pre-lime sample from 14/06/02, the increase
was barely sufficient (especially at high cake concentrations) to overcome the extra
solids throughput required to reach the same suspended solids throughput at a cake
concentration commensurate with the same total cake volume. The sample with lime
from 14/06/02 showed a significant increase in press performance due to the addition
of lime.

310
1

Average Suspended Solids


Throughput, Qp (kg/hr/m )
2
0.8

0.6

0.4

0.2 05/01/01 post-lime


14/06/02 pre-lime
14/06/02 post-lime
0
0% 10% 20% 30% 40%

Adjusted Cake Solids, <φ > (wt%)

Figure 5.2.41: Fill-only model predictions of suspended solids throughput with adjusted cake
solids for Arnfield WTP

Ultimately, the economics of lime addition were dependent on the costs


involved. Addition of lime allowed savings in operational expenses through reduced
manning hours, pumping costs and maintenance, or a greater throughput if required,
minus the cost of the large amount of lime. However, the press at Arnfield WTP was
not required to handle large throughputs, and the benefits of lime addition may not
have been fully realised.

Outcomes

This work illustrated the principle of using the models in conjunction with
material characterisation to make operating decisions. It showed that the addition of
lime to ferric sludges significantly increased the diffusivity, with corresponding
increases to the throughput of the filter press at a given cake solids. However, the
economic benefit of lime addition depended on the throughput required and the
costs/savings involved, and the lime may not have been advantageous.

311
5.3 Filter Press Optimisation and Control

The optimum processing conditions were developed in a mathematical


framework by dividing the batch cycle into its individual components and
investigating the effect on the maximum throughput. Following this, modelling work
investigating the impact of changes to processing variables of φ0, ∆P, d and Rm is
presented, along with considerations of press control.

The simplest batch cycle to be investigated was the theoretical case when there
was no press volume (therefore h = 0 and tL = 0), the operating pressure was reached
instantaneously (tP = 0), the fill stage remained in cake formation and out of
compression, and there was no squeeze phase (tS = 0). The linear approximation
result for piston-driven filtration (see section 2.4.2) shows that VF varies with the
square root of tF during cake formation (see equation 2.4.26), where the constant of
proportionality is β. The important consequence of this relationship was that the flow
rate was greatest at the start of the filtration process but then decreased as the filtration
time increased. Substituting equation 2.4.26 into equation 5.1.4 gives the average
specific throughput in terms of tF:

β tF
Q = …(5.3.1)
tF + tH

<Q> is inversely proportional to tH, therefore tH should be minimised.


Equation 5.3.1 has an extrema in <Q> where the derivative with respect to tF
vanishes, and is a maximum where the second derivative is negative. Therefore, the
optimum filtration time that gives the maximum throughput, tF,max, is equal to the
handling time:

t F ,max = t H …(5.3.2)

This important result, derived previously by Landman and White, 1997, gives
a general rule of thumb for filter press operation, that the time spent in filtration
should be equal to the handling time. The maximum average specific throughput,

312
<Q>max, is therefore determined by the material characteristics and ∆P (which give β),
and the operating procedure (tH):

β
Q max = …(5.3.3)
2 tH

Similar mathematical arguments were used to investigate the effect of tL, tP


and cake compression.

Loading Time

The time to load the press with sludge, tl, is given by equation 5.1.1. In this
case, the average throughput calculation (equation 5.1.4) includes the volume of
sludge within the press (given by h0) as well as V(t):

β t F − t L + h0
Q = …(5.3.4)
tF + tH

If all other variables are fixed, the average throughput decreases as the loading
time increases, indicating that the quicker the sludge is loaded, the better the
throughput. This ignores any detrimental shear effects that may come about from
increasing the sludge flowrate. Equating the derivative with respect to tF of equation
5.3.4 to zero gives the extrema:

2 2
h  2h h 
t F ,max = 2t L + t H + 2 0  ± 0 t L + t H +  0  …(5.3.5)
 β  β  β 

The negative sign gives a maximum for the throughput, since the second
derivative is then less than zero, indicating a maximum rather than minimum.
Compared with equation 5.3.4, this result shows that the ratio h0/β increases tF,max due
to the time required to load the press but decreases tF,max due the extra throughput
when the volume of the press is incorporated. However, this ratio will be small, and
tF,max is roughly when the time at constant pressure filtration (tF – tL) equals the sum of
the other times (tL + tH).

313
Ramping Pressure

An added complication to the fill-stage includes the effect of the pressure rise
of the fill pump. The simplest case is that the filtrate flow rate is constant during the
pressure rise. Incorporating this into equation 5.1.4 gives:

c P t P + β t F − t L − t P + h0
Q = …(5.3.6)
tF + tH

where cP is the constant of proportionality between tP and VP. tF,max, assuming


constant tP, is given by:

2
 c t + h0 
t F ,max = 2t L + 2t P + t H + 2 P P 
 β 
…(5.3.7)
2
 c t + h0   c t + h0 
− 2 P P  t L + t P + t H +  P P 
 β   β 

tP can be considered as an extra component to the loading time, given that cPtP
is combined with h0. Thus, the optimum time is increased due to the time taken for
the pressure rise but reduced due to the volume added during this stage. Since the
optimum press performance is when the pressure is maximised, the shortest possible
tP will give the highest throughput.

Cake Compression

The final level of complexity for fixed-cavity filtration is when the fill stage
enters non-linear compression after a given time. This phenomenon arises due to two
effects – the normal effect of the cake building up to the central plane of the cavity
and the lateral effect of the restriction of transport of sludge to the edges of the cavity.
Overall, compression during the fill stage is dependent upon the cavity width, initial
solids and material properties, but can generally be described by a logarithmic
equation, as illustrated by equation 2.4.29. Combining this with equation 5.1.4 gives:

314
E −t 
c P t P + β t F − t L − t P − t C + V∞ − exp 1 C  + h0
 E2 
Q =
tF + tH
…(5.3.8)

where tC is the time spent in non-linear compression. Differentiating equation 5.3.8


with respect to tC while holding all other variables (including tF) constant and
equating to zero shows that the maximum throughput is when tC is zero, given that tC
cannot be negative. Thus, cake compression is always inefficient, and is to be
avoided at all costs. In the case of fill-only filter presses, the optimum throughput will
always be before non-linear compression begins. However, a minimum cake
concentration is usually required, and the cake compression stage is usually needed to
reach this concentration. Therefore, most fill-only presses are run at below the
maximum throughput. This highlights the natural deficiencies and design faults (such
as cavity width) of such presses for use with water and wastewater sludges.

This analysis showed that the optimum cycle consists of the shortest possible
loading and pressure-rise, negligible cake compression and cake formation times
equal to the sum of the other times. The remainder of this section presents
optimisation, design and control results using the plate-and-frame models.

315
5.3.2 Fill-Only Filter Press Performance

The fixed-cavity numerical model was used with the characteristics of average
alum and ferric water treatment sludges to predict fill-only filter press performance
(based on solids throughput and final cake solids) with varying feed concentrations,
applied pressures, cavity widths and membrane resistances. The solids concentrations
are presented as mass fraction rather than volume fraction and the solids throughput as
mass per unit time per unit membrane area in order to be straightforward for operators
and designers. The solids throughput is presented as average throughput rather than
throughput per run, which assumes that the press is automated such that each cycle
commences immediately upon the completion of the previous cycle. Times are
presented as total cycle times, which included one hour handling and half an hour
loading prior to filtration. The pressure rise to reach operating pressure was fixed for
all cases at a linear rise over half an hour.

The laboratory determined material characteristics (see Chapter 4) were


extrapolated using power-law equations up to the applied pressures (6 to 12 bar). The
results shown in Chapter 4 illustrate that the ferric water treatment characteristics are
fairly constant while the alum water treatment characteristics can vary somewhat from
sludge to sludge.

Feed Concentration

The model predictions for fixed-cavity filtration of the average ferric water
treatment sludge at a range of initial concentrations are presented in Figure 5.3.1. At
10 bar, the equilibrium solids concentration is 39.6 wt%. The predictions show that
by increasing the feed concentration, higher average cake solids are reached for a
given filtration time. Since fill-only filter presses are usually operated on a daily basis
such that the cycle time is twenty-four hours, the final cake solids changes with
varying feed concentration. For example, feed concentrations of 6 wt% reach cake
solids of 33 wt% in a day whereas feed concentrations of 1 wt% reach only 17 wt%.
If consistent cake solids are desired under variable feed conditions, filtration times
must be allowed to vary.

316
40%
φ ∞ = 39.6 wt%
35%

Average Cake Solids (wt%)


30%

25%

20%

15%
6wt%
10% 5wt%
4wt%
3wt%
5% 2wt%
1wt%
0%
0 12 24 36 48
Total Cycle Time (hr)

Figure 5.3.1: Fixed-cavity filtration predictions of average cake solids with time for ferric water
treatment sludge; ∆PF = 10 bar, d = 3 cm, Rm = 0, tH = 1 hr, tP = 30 min

40%

35% φ ∞ = 32.8 wt%


Average Cake Solids (wt%)

30%

25%

20%

15%
6wt%
10% 5wt%
4wt%
3wt%
5% 2wt%
1wt%
0%
0 12 24 36 48
Total Cycle Time (hr)

Figure 5.3.2: Fixed-cavity filtration predictions of average cake solids with time for alum water
treatment sludge; ∆PF = 10 bar, d = 3 cm, Rm = 0, tH = 1 hr, tP = 30 min

317
The transient cake solids results for the average alum sludge characteristics are
shown in Figure 5.3.2. Compared to the ferric sludge, the alum sludge is slightly less
compressible (32.8 wt% at 10 bar) and slightly more permeable. As such, alum
sludges reach higher cake solids at earlier times, but lower cake solids at later times.

Fill-only filter presses are limited in terms of optimising throughput and cake
solids. Even though the throughput for a particular run is always increasing as more
material is fed to the press, the average throughput tends to decrease since the
filtration rate decreases as the cycle proceeds. By assuming that the filtrate volume
varies with the square root of the filtration time, the maximum throughput is when the
filtration time is half the total cycle time. Figure 5.3.3 and Figure 5.3.4 illustrate the
variation of average specific solids throughput with average cake solids for the ferric
and alum sludge characteristics respectively. For the operating conditions and cavity
width used, the cake solids at maximum throughput is significantly less than the
desired cake solids, such that high throughput and high cake solids are mutually
exclusive. To increase throughput, the cake solids must be reduced; conversely, to
increase cake solids, the throughput is compromised.

Figure 5.3.3 and Figure 5.3.4 also show that the average solids throughput at a
given cake concentration increases significantly with increasing feed concentration.
This is somewhat counterintuitive since the initial filtration rate decreases with
increasing concentrations due to lower permeability. However, assuming that the
press is automated, the average solids throughput increases since higher cake
concentrations are reached earlier for higher feed concentrations and more runs are
employed such that, on average, more time is spent during the high rate stages.
Overall, the best performance in terms of solids throughput and cake solids is when
the feed concentration is highest.

The results in Figure 5.3.3 and Figure 5.3.4 can be used to give membrane
area specifications for new presses and to set performance benchmarks for existing
presses. Since large variations in performance occur due to changing feed
concentrations, this parameter must be measured frequently and, through the
performance of pre-filtration thickening, kept as constant and as high as possible.

318
1
6wt%
0.9 5wt%

Solids Throughput (kg/hr/m )


2
4wt%
0.8 3wt%
2wt%
0.7 1wt%

φ∞ = 39.6 wt%
0.6

0.5

0.4

0.3

0.2

0.1

0
0% 10% 20% 30% 40%
Average Cake Solids (wt%)

Figure 5.3.3: Fixed-cavity filtration predictions of solids throughput with average cake solids for
ferric water treatment sludge; ∆PF = 10 bar, d = 3 cm, Rm = 0, tH = 1 hr, tP = 30 min

1
6wt%
0.9 5wt%
Solids Throughput (kg/hr/m )

4wt%
2

3wt%
0.8 2wt%
1wt%
0.7

0.6

0.5

0.4
φ∞ = 32.8 wt%

0.3

0.2

0.1

0
0% 10% 20% 30% 40%
Average Cake Solids (wt%)

Figure 5.3.4: Fixed-cavity filtration predictions of solids throughput with average cake solids for
alum water treatment sludge; ∆PF = 10 bar, d = 3 cm, Rm = 0, tH = 1 hr, tP = 30 min

319
In order to control the operation of fill-only filter presses, it is necessary to
measure the volume of filtrate, either directly or indirectly by measuring the volume
of feed. Measuring the volume gives the filtrate rate, which, as illustrated by Figure
5.3.5 and Figure 5.3.6, removes the sensitivity of the final cake solids to the feed
concentration. The exact flowrate will depend on the size of the press, the stroke
volume and the desired cake concentration.

1.E-04

φ∞ = 39.6 wt%
Specific Filtrate Flowrate, dV /dt

1.E-05
(m/s)

1.E-06

6wt%
1.E-07 5wt%
4wt%
3wt%
2wt%
1wt%
1.E-08
0% 10% 20% 30% 40%
Average Cake Solids (wt%)

Figure 5.3.5: Fixed-cavity filtration predictions of specific filtrate flowrate with average cake
solids for ferric water treatment sludge; ∆PF = 10 bar, d = 3 cm, Rm = 0, tH = 1 hr, tP = 30 min

320
1.E-04

Specific Filtrate Flowrate, dV /dt 1.E-05


(m/s)

1.E-06

φ∞ = 32.8 wt%
6wt%
1.E-07 5wt%
4wt%
3wt%
2wt%
1wt%
1.E-08
0% 10% 20% 30% 40%
Average Cake Solids (wt%)

Figure 5.3.6: Fixed-cavity filtration predictions of specific filtrate flowrate with average cake
solids for alum water treatment sludge; ∆PF = 10 bar, d = 3 cm, Rm = 0, tH = 1 hr, tP = 30 min

Fill Pressure

The fixed-cavity model was used to give predictions for the variation of
performance with applied pressure. The inputs to the model were the average ferric
material characteristics, a feed concentration of 3 wt% (φ0 = 0.0102 v/v) and d =
0.015 m. The material characteristics were extrapolated to the applied pressures of 6,
8, 10 and 12 bar.

Figure 5.3.7 shows the t versus V2 results. As the applied pressure is made
larger, both the rate and extent of filtration increase. However, at these pressures and
feed concentration for this material, there are only minor changes to the initial rate,
corresponding to small changes in Py(φ∞) and D(φ∞).

The variation of average cake solids versus time with pressure is presented in
Figure 5.3.8. The increase in pressure has very little effect on the filtration
performance; for example, increasing the pressure from 8 to 12 bar only gives a
couple of extra percentage cake solids after twenty-four hours of filtration.

321
1200000
12 bar
10 bar
1000000
8 bar
6 bar
800000
Time, t (s)

600000

400000

200000

0
0 0.02 0.04 0.06 0.08
2 2 2
(Specific Filtrate Volume) , V (m )

Figure 5.3.7: Fixed-cavity filtration predictions of t versus V2 for ferric water treatment sludge
with varying applied pressure; φ0 = 3 wt%, d = 3 cm, Rm = 0, tH = 1 hr, tP = 30 min

40%

35%
Average Cake Solids (wt%)

30%

25%

20%

15%

12 bar
10%
10 bar
5% 8 bar
6 bar
0%
0 24 48 72 96
Total Cycle Time (hr)

Figure 5.3.8: Fixed-cavity filtration predictions of cake solids with time for ferric water
treatment sludge with varying pressure; φ0 = 3wt%, d = 3cm, Rm = 0, tH = 1 hr, tP = 30 min

322
Figure 5.3.9 shows the solids throughput results for various applied pressures.
The benefit of increasing the pressure is minimal. Thus, for this material at these
operating conditions, increasing the pressure does not provide an avenue for
increasing the press throughput.

0.6
12 bar
Solids Throughput (kg/hr/m )

10 bar
2

0.5
8 bar
6 bar
0.4

0.3

0.2

0.1

0
0% 10% 20% 30% 40%
Average Cake Solids (wt%)

Figure 5.3.9: Fixed-cavity filtration predictions of solids throughput with average cake solids for
ferric water treatment sludge with varying applied pressure; φ0 = 3 wt%, d = 3 cm, Rm = 0, tH = 1
hr, tP = 30 min

Cavity Width

While the overwhelming majority of plate-and-frame presses have cavity


widths of about 3 cm, some manufacturers do offer presses with different cavity
widths. The numerical model has been used to investigate the dependence of fixed-
cavity filtration on cavity width. The material characteristics for the average ferric
water treatment sludge were used with constant feed concentration at 3 wt% and
constant operating pressure at 10 bar.

Figure 5.3.10 shows the model predictions of t versus V2 for cavity widths
ranging from 1 to 3 cm. While the initial rate does not vary (corresponding to
consistent cake formation), the process enters cake compression earlier for smaller

323
cavities. Since there is greater capacity per unit area, the equilibrium filtrate volume
increases with increasing cavity width.

800000
3.0 cm
700000 2.5 cm
2.0 cm
1.5 cm
600000 1.0 cm
Time, t (s)

500000

400000

300000

200000

100000

0
0 0.01 0.02 0.03 0.04 0.05 0.06
2 2 2
(Specific Filtrate Volume) , V (m )

Figure 5.3.10: Fixed-cavity filtration predictions of t versus V2 for ferric water treatment sludge
with varying cavity widths; φ0 = 3 wt%, ∆PF = 10 bar, Rm = 0, th = 1 hr, tp = 30 min

The model predictions for average cake solids with time for the various cavity
widths are presented in Figure 5.3.11. The results show that, by reducing the cavity
width, a significant increase in the cake concentration in a given filtration time is
achieved. For example, a press with 3 cm cavities reaches 20.7 wt% in a 12 hour
cycle, while a 1 cm cavity reaches 36.7 wt%.

Figure 5.3.12 shows the variation of average solids throughput with cake
solids for the range of cavity widths. At low concentrations, the wider cavities exhibit
higher throughputs, corresponding to extended cake formation times. At high cake
solids the thinner cavities show higher throughputs, since higher cake solids are
achieved quicker and extra cycles can be used.

This analysis does not include the added bonus of smaller cavities that either
more plates can be used on the same frame (with a proportional increase in
throughput) or a smaller frame can be used.

324
φ ∞ = 39.6 wt%
40%

35%

Average Cake Solids (wt%)


30%

25%

20%

15%

10% 1.0 cm
1.5 cm
2.0 cm
5% 2.5 cm
3.0 cm
0%
0 12 24 36 48
Total Cycle Time (hr)

Figure 5.3.11: Fixed-cavity predictions of cake solids with time for ferric water treatment sludge
with varying cavity width; φ0 = 3 wt%, ∆P = 10 bar, Rm = 0, th = 1 hr, tp = 30 min

0.6
φ∞ = 39.6 wt%
Solids Throughput (kg/hr/m )
2

0.5

0.4

0.3

0.2
3.0 cm
2.5 cm
0.1 2.0 cm
1.5 cm
1.0 cm
0
0% 10% 20% 30% 40%
Average Cake Solids (wt%)

Figure 5.3.12: Fixed-cavity predictions of solids throughput with cake solids for ferric water
treatment sludge with varying cavity width; φ0 = 3 wt%, ∆P = 10 bar, Rm = 0, th = 1 hr, tp = 30 min

325
Also shown in Figure 5.3.12 is the progression of the maximum throughput to
higher cake solids as the required filtration time approaches half the total cycle time.
For the 3 cm cavity, the maximum throughput is when the cake solids are 8.0 wt%,
whereas for the 1 cm cavity, the maximum throughput is when the cake solids are
20.8 wt%. Thus, an optimum cavity width exists for a given set of operation
conditions and desired final cake concentration, with the caveat that thinner cakes are
lighter and may not have good release properties (that is, the cake may stick to the
membrane) even at high final concentrations.

Membrane Resistance

The use of the numerical fixed-cavity model allows the prediction of the
impact of increased membrane resistance on the performance of fill-only filter
presses. The model was used with the average ferric characteristics at a range of Rm
values (0, 1010, 1011, 5 x 1011, 1012 and 1013 Pas/m). The other inputs were held
constant (φ0 = 3 wt%, ∆P = 10 bar, d = 3 cm and tP = 1800s).

The t versus V2 results are presented in Figure 5.3.13. Rm affects the initial
rate, but does not have an effect on the equilibrium filtrate volume. Rm values of 1010
and 1011 Pas/m have little effect, while resistances greater than 1012 Pas/m dominate
the filtration behaviour.

Figure 5.3.14 shows the change of average cake solids concentration with
cycle time for the different Rm values. Over a twenty-four hour cycle, Rm values
greater than 1011 begin to have significant detrimental effects on cake solids, with the
predicted solids dropping from 26.5 wt% for Rm = 1011 Pas/m, to 22.3 wt% for Rm = 5
x 1011 Pas/m, to 16.3 wt% for Rm = 1012 Pas/m.

The variation of solids throughput with cake solids for the range of Rm values
used is shown in Figure 5.3.15, illustrating that membrane resistances less than 1011
Pas/m have only minor effects on throughput, and cleaning is likely to be a waste of
time and money. However, Rm values greater than this show increasingly detrimental
effect on throughput, such that cleaning may be beneficial.

326
1000000

V∞2 = 0.0618 m2
900000 10e13
10e12
800000 5x10e11
10e11
10e10
700000 No Rm
Time, t (s)

600000

500000

400000

300000

200000

100000

0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
2 2 2
(Specific Filtrate Volume) , V (m )

Figure 5.3.13: Fixed-cavity filtration predictions of t versus V2 for ferric water treatment sludge
with varying membrane resistances; φ0 = 3 wt%, ∆P = 10 bar, d = 3 cm

40%
φ ∞ = 39.6 wt%
35%
Average Cake Solids (wt%)

30%

25%
No Rm
20% 10e10
10e11
5x10e11
15% 10e12
10e13
10%

5%

0%
0 12 24 36 48 60 72 84 96
Total Cycle Time (hr)

Figure 5.3.14: Fixed-cavity filtration predictions of cake solids versus cycle time for ferric water
treatment sludge with varying membrane resistances; φ0 = 3 wt%, ∆P = 10 bar, d = 3 cm

327
0.6
No Rm
10e10

Solids Throughput (kg/hr/m )


2
10e11
0.5 5x10e11
10e12
10e13
0.4

φ∞ = 39.6 wt%
0.3

0.2

0.1

0
0% 10% 20% 30% 40%
Average Cake Solids (wt%)

Figure 5.3.15: Fixed-cavity filtration predictions of solids throughput versus cake solids for ferric
water treatment sludge with varying membrane resistances; φ0 = 3 wt%, ∆P = 10 bar, d = 3 cm

328
5.3.3 Fill-and-Squeeze Filter Press Performance

The analytical flexible-membrane model was used to predict the dewatering of


ferric and alum water treatment sludges in fill-and-squeeze filter presses in order to
investigate the variation of throughput and final cake solids with feed concentration
and applied pressure. As for the fill-only predictions above, the solids concentration
is presented as mass fraction and the solids throughput as mass per unit time per unit
membrane area. The initial loading and pressure rise phases were both constant at
half an hour for all cases. The results presented here have utilised the fill-and-squeeze
calculation program (Section 5.1.5), and may therefore contain small interpolation
errors. The analytical model assumes that the membrane resistance is negligible and
that the fill-stage does not enter compression.

Fill-and-squeeze filter presses possess greater versatility compared to fill-only


filter presses. The squeeze phase allows the press to reach any desired cake
concentration below the compression limit such that there is a unique combination of
fill and squeeze times for a given initial solids concentration and applied pressure to
reach a particular cake concentration. Figure 5.3.16 shows the model results for the
variation of the required squeeze time with fill time for ferric water treatment sludge
at a feed concentration of 4 wt%. The maximum cake solids for this material at 10
bar is 39.6 wt%. The squeeze time required to reach a fixed cake solids initially
increases with fill time since the average solids concentration at the start of the
squeeze phase increases and therefore the rate of filtration decreases. The squeeze
time may then pass through a maximum and begin to fall as the fill stage nears the
required cake concentration.

Figure 5.3.17 shows the corresponding squeeze time results for alum water
treatment sludge at the same conditions. The maximum cake solids for this material
at 10 bar is 32.8 wt%, and is therefore less compressible than ferric sludge. However,
the alum sludge reaches concentrations below the compressibility limit quicker the
ferric sludge (in the range of 20 to 35 minutes compared with 30 to 55 minutes) since
it is more permeable.

329
60
35 wt%
55 30 wt%
25 wt%

Squeeze Time (min)


50 20 wt%

45

40

35

30

25
0 60 120 180 240
Fill Time (min)

Figure 5.3.16: Flexible-membrane predictions of squeeze time after a certain fill time for ferric
water treatment sludge; ∆PF = 6 bar, ∆PS = 10 bar, φ0 = 4 wt%, d = 3 cm, tH = 60 min

40
32 wt%
30 wt%
35 25 wt%
Squeeze Time (min)

20 wt%

30

25

20

15
0 60 120 180 240
Fill Time (min)

Figure 5.3.17: Flexible-membrane predictions of squeeze time after a certain fill time for alum
water treatment sludge; ∆PF = 6 bar, ∆PS = 10 bar, φ0 = 4 wt%, d = 3 cm, tH = 60 min

330
The average solids throughput is determined by the volume of sludge fed to
the press during the fill stage and the total cycle time. There is no sludge fed to the
press during the squeeze stage, which can therefore be considered as an extra
component of the handling time. Figure 5.3.18 shows the model predictions of the
variation of the average specific solids throughput with total cycle time to reach a
particular cake concentration for ferric water treatment sludge at a feed concentration
of 4 wt%. An optimum cycle time exists when the throughput is at a maximum,
which is approximately constant for the range of final solids concentrations presented.
At short cycle times, the throughput is low since the squeeze and handling times are
proportionally large compared to the fill time. The throughput is also low at high
cycle times since the rate of filtration during the fill stage is continuously falling. The
average throughput decreases slightly as the desired cake solids increases due to the
extra squeeze time required. Compared to fill-only presses, where high throughput
and high cake solids are mutually exclusive, the reduction of average throughput with
cake solids is much less pronounced for fill-and-squeeze presses, which are therefore
more versatile.

0.55
Solids Throughput (kg/hr/m )
2

0.5

0.45

35 wt%
0.4
30 wt%
25 wt%
20 wt%
0.35
60 120 180 240 300
Total Cycle Time (min)

Figure 5.3.18: Flexible-membrane predictions of solids throughput with total cycle time for ferric
water treatment sludge; ∆PF = 6 bar, ∆PS = 10 bar, φ0 = 4 wt%, d = 3 cm, tH = 60 min

331
0.65

Solids Throughput (kg/hr/m )


2
0.6

0.55

0.5

32 wt%
0.45 30 wt%
25 wt%
20 wt%
0.4
60 120 180 240 300
Total Cycle Time (min)

Figure 5.3.19: Flexible-membrane predictions of solids throughput with total cycle time for alum
water treatment sludge; ∆PF = 6 bar, ∆PS = 10 bar, φ0 = 4 wt%, d = 3 cm, tH = 60 min

The equivalent plot for alum water treatment sludge at 4 wt% is shown in
Figure 5.3.19, which also illustrates the maximum throughput. The optimum cycle
time is slightly lower for alum sludge than ferric sludge, due to the shorter required
squeeze times and the maximum throughput is increased since more sludge is fed
during the fill-stage, both consequences of the slightly higher permeability of alum
sludges compared to ferric sludges.

Feed Concentration

The flexible-membrane model was used to investigate the effect of varying the
feed concentration. Figure 5.3.20 and Figure 5.3.21 show the required squeeze time
to reach cake solids of 30 wt% for feed concentrations from 1 to 6 wt% for ferric and
alum water treatment sludge respectively. The accuracy is variable due to
interpolation errors, but the trends are reliable. The squeeze times at low fill times
increase as φ0 increases, corresponding to greater cake solids at the beginning of the
squeeze phase. The required squeeze time may pass through a maximum as the final
concentration of the fill-stage begins to approach the desired cake solids.

332
60

50

Squeeze Time (min) 40

30

20 6wt%
5wt%
4wt%
10 3wt%
2wt%
1wt%
0
0 60 120 180 240
Fill Time (min)

Figure 5.3.20: Flexible-membrane predictions of squeeze time required to reach 30 wt% with fill
time for ferric water treatment sludge; ∆PF = 6 bar, ∆PS = 10 bar, d = 3 cm, tH = 60 min

35

30
Squeeze Time (min)

25

20
6wt%
5wt%
15 4wt%
3wt%
2wt%
1wt%
10
0 60 120 180 240
Fill Time (min)

Figure 5.3.21: Flexible-membrane predictions of squeeze time required to reach 30 wt% with fill
time for alum water treatment sludge; ∆PF = 6 bar, ∆PS = 10 bar, d = 3 cm, tH = 60 min

333
As shown in Figure 5.3.18 and Figure 5.3.19, optimum cycle times exist for
given operating conditions such that the solids throughput is maximised. The model
results for the variation of maximum solids throughput with feed concentration for a
range of final cake solids for ferric and alum sludges are shown in Figure 5.3.22 and
Figure 5.3.23 respectively. The effect of the feed concentration on the average
throughput is very large – even though the filtration rate decreases with increasing
feed concentration (since the sludge is thicker), there is less water to be removed to
reach a desired cake concentration. The end result is that the highest average solids
throughput occurs when the feed concentration is maximised, and therefore the press
performance is highly dependent on the performance of pre-filtration thickening or
clarification.

0.8
F&S 20wt%
Maximum Solids Throughput

0.7 F&S 25wt%


F&S 30wt%
F&S 35wt%
0.6 F 20wt%
F 25wt%
0.5 F 30wt%
(kg/hr/m )
2

F 35wt%
0.4

0.3

0.2

0.1

0
0 1 2 3 4 5 6 7
Feed Concentration (wt%)

Figure 5.3.22: Flexible-membrane predictions of maximum solids throughput with feed


concentration for ferric water treatment sludge; Fill-and-squeeze (F&S): ∆PF = 6 bar, ∆PS = 10
bar; Fill-only (F): ∆PF = 10 bar; d = 3 cm, tH = 60 min

Comparing the two coagulant types, the alum water treatment sludge shows
roughly 20% higher throughput than the ferric sludge for all feed concentrations and
cake solids. The ferric sludge may be slower, but does reach slightly higher
concentrations. A comprehensive comparison of the two sludges must incorporate

334
thickening behaviour, since a sludge that thickens better will have a higher
throughput.

0.9
F&S 20wt%
0.8 F&S 25wt%
Maximum Solids Throughput
F&S 30wt%
F&S 35wt%
0.7 F 20wt%
F 25wt%
0.6 F 30wt%
F 32wt%
(kg/hr/m )
2

0.5

0.4

0.3

0.2

0.1

0
0 1 2 3 4 5 6 7
Feed Concentration (wt%)

Figure 5.3.23: Flexible-membrane predictions of maximum solids throughput with feed


concentration for alum water treatment sludge; Fill-and-squeeze (F&S): ∆PF = 6 bar, ∆PS = 10
bar; Fill-only (F): ∆PF = 10 bar; d = 3 cm, tH = 60 min

Also shown in Figure 5.3.22 and Figure 5.3.23 are the 10 bar fill-only results
from the previous section. Fill-only presses reach comparable throughputs to fill-and-
squeeze presses only at high feed concentrations and low cake solids when the
required squeeze time approaches zero. For low feed concentrations or high cake
solids, fill-only presses have much lower throughputs for both alum and ferric water
treatment sludge. This reflects the nature of fill-only presses, such that both
throughput and cake solids are achieved by fixed-cavity filtration, whereas for fill-
and-squeeze presses, the fill-stage gives the throughput while the squeeze phase is
used to achieve high cake solids.

Squeeze Pressure

The effect of various pressure regimes on average throughput was also


investigated. Figure 5.3.24 shows the results for ferric water treatment sludge at a

335
range of combinations of fill and squeeze pressures (6 to 10 and 10 to 15 bar
respectively). The results show that very small increases are achieved through
increasing the pressure. Increasing the squeeze pressure shortens the squeeze time,
but the decrease as a proportion of the overall cycle time is small. As illustrated in the
fill-only section, increasing the fill pressure gives a small increase in filtration rate,
but the versatility of the fill-and-squeeze press negates this effect.

0.7
Maximum Solids Throughput

0.6
(kg/hr/m )

0.5
2

0.4
6,10 bar
6,12 bar
0.3 6,15 bar
8,10 bar
10,10 bar
0.2
0 1 2 3 4 5 6 7
Feed Concentration (wt%)

Figure 5.3.24: Flexible-membrane predictions of maximum solids throughput with feed


concentration under various pressure regimes (∆PF, ∆Ps) for ferric water treatment sludge; φF =
30 wt%, d = 3 cm, tH = 60 min

Handling Time

Figure 5.3.26 shows the effect of improving the handling time on solids
throughput as a function of total cycle time for ferric water treatment sludge with a
feed concentration of 4 wt% and a desired cake solids of 30 wt%. Either automating
the filter press or enhancing cake release can improve the handling time. The
maximum throughput increases and the optimum cycle time decreases with
decreasing tH. The optimum cycle time decreases at a higher rate than the decreasing
handling time since the optimum fill time also decreases. The maximum throughput
increases since extra cycles are used and more time is spent in high rate filtration.

336
0.7

Solids Throughput (kg/hr/m )


2
0.6

0.5

15min
0.4
30min
45min
60min
0.3
60 120 180 240 300
Total Cycle Time (min)

Figure 5.3.25: Flexible-membrane predictions of solids throughput with cycle time for ferric
water treatment sludge at a range of tH; ∆PF = 6 bar, ∆PS = 10 bar, φ0 = 4 wt%, φF = 30 wt%

0.9
15min
Maximum Solids Throughput

0.8 30min
45min
0.7 60min
(kg/hr/m )
2

0.6

0.5

0.4

0.3

0.2
0 1 2 3 4 5 6 7
Feed Concentration (wt%)

Figure 5.3.26: Flexible-membrane predictions of maximum solids throughput with φ0 for ferric
water treatment sludge for a range of tH; ∆PF = 6 bar, ∆PS = 10 bar, φF = 30 wt%, d = 3 cm

337
The variation of maximum throughput with feed concentration for the
different handling times is illustrated in Figure 5.3.26, which shows that the increase
in throughput with decreasing handling times is consistent across the feed
concentration range.

Optimisation and Control

The results show that the maximum throughput of fill-and-squeeze filter


presses for different operating conditions occurs when the feed concentration is
highest, and the handling, loading and ramping pressure times are lowest. The
applied pressure and desired cake solids have only small effects on throughput,
providing that the operating procedures are adjusted accordingly. Both high
throughputs and high cake solids are achievable.

The results indicate that the optimum cycle time does not vary much with
varying feed concentration. For ferric sludges, the optimum is at roughly 210
minutes, while for alum sludges, the optimum is about 180 minutes (assuming tH = 60
min, tL = 30 min and tP = 30 min). This suggests that, for consistent operation, fill-
and-squeeze presses are operated with constant fill times. Since the fall in throughput
is more gradual with increasing cycle time rather than decreasing cycle time, presses
should be operated at cycle times slightly higher than the optimum. The required
squeeze time can then be either set as a minimum time (for example, 60 min for ferric
sludges and 35 min for alum sludges), or the filtrate can be measured and a flowrate
condition used as a cut-off criterion for the squeeze phase. The former would ensure
high cake solids, but the press would be operated at below the optimum throughput.
The latter control method is preferable, but many sites do not measure filtrate volume.

The trends predicted by the model show that, in general operation, if the cake
concentration is too low, a shorter fill time with the same squeeze time or the same fill
time with a longer squeeze will increase the cake concentration. If the cake is too thin
such that it is not heavy enough to drop from the membrane, an increase in fill time
and corresponding increase in squeeze time will give greater loading at the same cake
concentration, but with lower throughput.

338
5.4 Conclusions

This work detailed the theoretical development of models of fill-only and fill-
and-squeeze plate-and-frame filter presses, and presented analytical and numerical
algorithms to solve the governing equations. These models represented important
advances in plate-and-frame modelling. The models were capable of predicting the
behaviour with or without membrane resistance for networked and un-networked feed
solutions using the standard power-law functional forms or the complex material
characteristics that arise from batch settling analysis. The numerical models were also
able to predict cake compression during the fill-stage. The results showed that when
the membrane resistance dominated the filtration process, t was linear with V. As the
filtration progressed, the cake resistance increased and began to dominate, such that
during cake formation, t became linear with V2. The results also showed that the
ramping pressure and the membrane resistance were competing factors during the
start of the fill-stage.

The predictions were used to validate the modified Darcian approach to


measuring Rm from on-site data, including the outcome that it generally predicted the
magnitude of Rm. This enabled the development of cloth cleaning and replacement
protocols for batch-operated filtration devices. However, the approach was used with
caution due to the inherent assumptions in the theory.

Visual Basic calculation tools were outlined that use interpolation of model
generated data via a simple user interface to give quick and straightforward solutions
to plate-and-frame performance.

The models were validated and used to predict press performance at a range of
sites. The models showed reasonable agreement with the site data in most cases. The
predictions were used to evaluate options for increasing press throughput. At some
sites, substantial improvements were made, while at others, the results showed that
new presses were required. Tenders put forward by manufacturers were evaluated
based on the model results.

339
Finally, the optimisation and control of filter presses was discussed, including
the effect on performance of changing various operating conditions. High throughput
and high cake solids for fill-only presses were mutually exclusive, while fill-and-
squeeze presses exhibited greater versatility. The model results showed that the best
performances were when the initial solids were highest, and the loading, increasing
pressure and handling times were shortest.

The work also gave insight into press control by ensuring accurate filtrate
volume measurement. Options for design and guidelines for the acquisition of new
equipment provide buyers with detailed knowledge of their press requirements, rather
than relying upon manufacturers. Key aspects to control the operation of plate-and-
frame presses were identified, such as accurately measuring the filtrate volume and
ending the fill stage before non-linear compression sets in. Fill-only presses should
be controlled based on flowrate rather than time to account for changes to φ0, whereas
the fill-stage of fill-and-squeeze presses should be a fixed time and the squeeze-stage
controlled using flowrate to give high throughput and cake solids.

340
6 Centrifuge Modelling

Centrifuges are used extensively by a wide variety of industries to perform


solid-liquid separation. In liquid-solid systems, the centrifugal buoyancy due to the
phase density difference causes the solids to consolidate against the bowl of the
centrifuge. There are two basic types: thickening and filtering centrifuges, which are
the centrifugal analogies for gravity thickening and gravity filtration respectively, and
can be operated in batch, semi-continuous or continuous modes.

In thickening (or sedimentation) centrifuges, the bowl wall is solid; the


particles settle against the wall of the centrifuge and form a cake, and the liquor is
withdrawn from above the cake. Disc, decanter and tubular centrifuges are examples
of thickening centrifuges. Disc centrifuges have internal conical discs to aid
sedimentation, which is analogous to the use of lamella plates in gravity thickening.
Decanter centrifuges use an internal screw or scroll that rotates at a different speed to
the bowl to push the cake through the centrifuge, and is the predominant device used
by the water and wastewater industries. The distinction is made between solid-bowl
tubular centrifuges, where a radial coordinate is required, and laboratory tube
centrifuges, where a Cartesian coordinate with radial acceleration is used.

In filtering centrifuges, the bowl wall is semi-permeable, and the liquor is


drawn through the cake and out the walls of the centrifuge. Pendulum, pusher and
knife centrifuges are examples of filtering centrifuges.

The first tubular centrifuge models were based upon the Stoke’s settling
trajectory of a particle of a given size, called Sigma Theory (Ambler, 1952, 1959). It
was acknowledged but not considered that the consolidation of the solid phase under
centrifugal acceleration critically affected the efficiency of the centrifuge. An
empirical experience factor was introduced to allow for hindered settling and cake
consolidation.

341
Conventional filtration theory was expanded to include the effect of liquid
flow through a consolidating cake using Darcy’s law (Tiller and Shirato, 1964) and
applied to centrifugal thickening by Tiller and Hsyung, 1993. A correction factor was
introduced into Sigma Theory to include this effect by Corner-Walker, 2000, but it is
not useful for highly compressible materials. Anestis and Schneider, 1983, extended
Kynchian theory (Kynch, 1952) based on the theory of kinematic waves to model
one-dimensional batch centrifugation and showed that the sediment concentration is a
function of time for ideal suspensions.

Green, et al., 1996, used the phenomenological theory of solid-liquid


separation developed by Buscall and White, 1987, to model the equilibrium state of a
consolidated suspension in a centrifugal tube (that is, a steady-state Cartesian
coordinate problem with centrifugal acceleration). In this work, the sedimentation-
consolidation equations of phenomenological theory were applied to radial
coordinates with centrifugal acceleration. Three models of thickening centrifugation
were developed. The two simpler models described the one-dimensional batch and
continuous centrifugation of sludges (based on tubular centrifuges), while the third
model described two-dimensional decanting centrifuges. A model of disc-bowl
centrifugation needed to consider the effects of shear at the disc surface and this was
not considered further here. Likewise, models of filtering centrifuges were not
considered since they must include the reducing applied pressure due to the falling
head of fluid (such that the irreversibility condition must be invoked), the resistance
of the semi-permeable membrane, and any capillary-assisted cake compaction and
desaturation caused by fluid drainage.

342
6.1 One-Dimensional Solid-bowl Batch
Centrifuge

The solid-liquid separation performed by batch-thickening centrifugation was


modelled by formulating the mass and momentum conservation equations into radial
coordinates, applying appropriate scalings and identifying the boundary conditions for
each zone. Models are presented for both the initial suspension networked (φ0 > φg)
and the initial suspension un-networked (φ0 < φg) cases. The conservation equations
given here are analogous to the formulations of Bürger and Concha, 2001, who
incorporated a similar phenomenological approach for the φ0 < φg case to extend the
work of Anestis and Schneider, 1983, to include cake compression.

ω
ω
r
rf
Sedimentation Zone
Consolidation Zone

Clear-liquor Zone

rb rs(t)
rc(t)
rs(t) L
rc (t)
rf r
rb

(a) Axial view (b) Lateral view

Figure 6.1.1: Schematic of 1-D batch centrifugal thickening: (a) Axial view; (b) Lateral view

Axial and lateral views of one-dimensional solid-bowl batch centrifugation are


presented in Figure 6.1.1. At the start of the process (t = 0), the sludge is assumed
uniform at a concentration of φ0. The centrifugal action exerts acceleration upon the
particles and thickening begins. Three zones of behaviour are distinguishable:

343
- A consolidation zone of material at concentrations greater than φg (or greater
than φ0 in the φ0 > φg case) forms against the bowl of the centrifuge (r = rb). A
network of particles exists where the local stress on the network, pp, is equal to
the compressive yield stress of the network, Py(φ).

- A sedimentation zone exists where the particles are in free-fall. The


sedimentation zone behaves differently for networked and un-networked cases
(Landman, et al., 1988):

o For φ0 > φg, a network exists but the consolidating forces are less than
the compressive yield stress, so the zone slumps at a constant
concentration. The network is able to transmit pressure, which varies
from 0 at the top of the sediment, rs(t), to Py(φ0) at the boundary
between the sedimentation and consolidation zones, rc(t). φ(rc,t) is
continuous and at φ0. At equilibrium, the sedimentation zone still
exists, since the compressive forces do not surpass Py(φ0).

o For φ0 < φg, no network exists in the sedimentation zone and the
particle velocity is independent of Py(φ). In one-dimensional
centrifugal thickening with a constant cross section, the sedimentation
zone remains at a constant concentration, φ0 (Green, et al., 1996).
However, in radial coordinates, the increasing area has the effect of
thinning the sedimentation zone such that the volume fraction in the
sedimentation zone, φs, is a function of time. rc(t) is discontinuous,
with the volume fraction at the top of the bed, φ(rc+,t), at φg and φ(rc-,t)
= φs(t). A critical time, tc, exists when rc(t) = rs(t), that is, when all of
the particles have settled to the consolidating bed and there is no
sedimentation zone. After tc, the boundary at rc remains discontinuous,
with φ(rc+,t) = φg and φ(rc-,t) = 0. The bed compresses until the
equilibrium state is reached.

- The third zone is the clear-liquor zone from which solid material has settled (φ
= 0). rf is the fluid height, which is determined from the initial volume of
suspension per unit length, V0:

344
V0
rf = rb2 − …(6.1.1)
π

In this work, analytical steady-state solutions were used to give the


equilibrium distribution, while analytical similarity series solutions were used to give
the small-scale time dependence. The governing equations for the transient
behaviours were solved using a forward difference approximation in time and an
iterative 4th-5th order Runge-Kutta adaptive step-size numerical method in the spatial
direction.

This work did not consider the effect of internal discs, as seen in disc-bowl
centrifuges or semi-continuous operation in which the cake is allowed to build up
while the clear-liquor is continuously withdrawn.

The model presented here represents the special case of a constant diameter
co-current decanting centrifuge in which the sedimentation and consolidation zones
progress through the centrifuge at the same rate.

345
6.1.1 Model Formulation

Sedimentation-Consolidation Equations

The local conservation of mass and momentum equations for particulate


networks are given in Section 2.2.1. In one-dimensional radial coordinates with up =
up(r,t) and uf = uf(r,t), equation 2.2.8 becomes:

1 ∂
r ∂r
[(
r φu p + (1 − φ )u f = 0 )] …(6.1.2)

Integrating equation 6.1.2 with respect to r gives the particle and fluid
velocities in terms of the bulk flow, q(t):

q (t )
φu p + (1 − φ )u f = …(6.1.3)
r

q(t) is constant with respect to r. q is equal to zero for batch thickening since
the total volume is conserved, therefore equation 6.1.3 becomes:

φ
uf =− up …(6.1.4)
(1 − φ )

Converting equation 2.2.3 to a one-dimensional radial coordinate with


centrifugal acceleration, g = ω2r, and substituting equation 6.1.4 gives:

R (φ ) ∂p p
− φu p − + φ∆ρω 2 r = 0 …(6.1.5)
(1 − φ ) 2 ∂r

In the consolidation zone, the applied pressure is greater than the yield stress,
therefore pp = Py(φ). Therefore, equation 6.1.5 becomes:

R (φ ) dPy (φ ) ∂φ
− φu p − + φ∆ρω 2 r = 0 …(6.1.6)
(1 − φ ) 2 dφ ∂r

346
up(r,t) is found from the transient behaviour, which is given by converting
equation 2.2.6 to radial coordinates:

∂φ
∂t
=−
1 ∂
r ∂r
rφu p ( ) …(6.1.7)

These non-linear partial differential equations (equations 6.1.6 and 6.1.7) are
the overall governing equations for consolidation in solid-bowl batch centrifugal
thickening, giving the transient volume fraction distribution when solved using the
appropriate initial and boundary conditions. When combined, they represent a
second-order hyperbolic diffusion equation (Bürger and Concha, 2001). The initial
condition is given by the assumption that the sludge concentration is initially uniform
at φ0:

φ (r ,0 ) = φ0 …(6.1.8)

The boundary condition at the bowl wall is that the solids velocity is zero:

u p (rb , t ) = 0 …(6.1.9)

The boundary conditions at rc(t) and rs(t) depend on whether the initial
suspension is networked or un-networked, and are outlined later. The global
conservation of solids volume is:

rb V0φ0
∫ rφdr = …(6.1.10)
rs (t ) 2π

Dimensionless Equations

The sedimentation-consolidation equations are simplified by applying


appropriate scalings. The radial coordinate, r is scaled with rb to give the
dimensionless length scale, Z:

r2
Z = 1− …(6.1.11)
rb2

347
Thus, rb, rc(t), rs(t), and rf map to 0, Zc(T), Zs(T) and Zf respectively. pp(r,t)
and Py(φ) are scaled with the centrifugal force to give Pp(Z,T) and f(φ):

2
Pp = pp …(6.1.12)
∆ρω 2 rb2

f (φ ) = Py (φ )
2
…(6.1.13)
∆ρω 2 rb2

R(φ) is scaled with the initial concentration to give B(φ):

(1 − φ0 )2 R(φ )
B(φ ) = …(6.1.14)
R(φ 0 ) (1 − φ )2

Thus, the appropriate time scaling is based on the rate of sedimentation as


opposed to the rate of consolidation:

2 ∆ρω 2 (1 − φ 0 )2
T= t …(6.1.15)
R(φ 0 )

The solids velocity, up(r,t), is scaled to give a dimensionless solids flux,


ψ(Z,T):

R(φ0 )
ψ (Z ,T ) = rφu p …(6.1.16)
∆ρω rb (1 − φ0 )2
2 2

Substituting these scalings into the governing equations (6.1.6 and 6.1.7)
gives:

∂φ 1  φ ψ 
=− −
∆(φ )  B (φ ) 1 − Z 
 …(6.1.17)
∂Z

∂φ ∂ψ
= …(6.1.18)
∂T ∂Z

where the scaled diffusivity, ∆(φ), is defined as:

348
 0 φ < φg 
 
∆ (φ ) =  f ' (φ )
φ ≥ φ g 
…(6.1.19)
 B(φ )
 

The scaled version of equation 6.1.5 is used when Pp(Z,T) < f(φ):

∂Pp B(φ )
= ψ −φ …(6.1.20)
∂Z 1− Z

The initial condition, wall boundary condition and global conservation


become:

φ (Z ,0 ) = φ0 …(6.1.21)

ψ (0 ,T ) = 0 …(6.1.22)

Z s (T )
∫ φdZ = Z f φ0 …(6.1.23)
0

The boundary conditions at Zc(T) and Zs(T) and thus the method of solution,
depend on whether the initial suspension is networked or un-networked. These two
cases are now examined separately. In each case equilibrium and small time solutions
are derived and a numerical method for the general time dependent case is outlined.

Case 1: Initial Suspension Un-networked (φ0 < φg)

For the φ0 < φg case, the particle pressure in the sedimentation zone is zero,
since the concentration is below that which is necessary to transmit a stress. The top
of the cake is at φg.

Steady-state solution (φ0 < φg, T→∞)

As the time gets very large, all the solids settle from the sedimentation zone in
the φ0 < φg case. The steady-state solution, φ(Z∞), is found by setting the time
derivatives and the solids flux, ψ, in the consolidation equations to zero. Therefore,
equation 6.1.17 becomes:

349
dφ φ
=− …(6.1.24)
dZ ∞ f ' (φ )

The boundary conditions to this ordinary differential equation are that φ(0) =
φ∞ and φ(Zc∞) = φg, where the volume fraction at the bowl wall at infinite time, φ∞,
and the height of the cake at infinite time, Zc∞, are to be determined. φ∞ is found from
the global conservation:

Z c∞

( )

∫ φ Z dZ = Z f φ0 …(6.1.25)
0

Rearranging equation 6.1.24 as the gradient of f(φ) with respect to Z∞ and


integrating from 0 to Zc∞ gives the local pressure at the bowl wall at equilibrium that,
when equated to the compressive yield stress, gives φ∞:

f (φ∞ ) = Z f φ0 …(6.1.26)

If f(φ) is given as an analytical function, the volume fraction distribution at


steady-state is determined explicitly by solving equation 6.1.24. However, to remain
general, a 4th-5th order Runge-Kutta numerical technique is used here. Equation
6.1.24 is evaluated in 4th order steps of ∆Z from φ(0) = φ∞ until φ(Z∞) = φg. If a step
gives φ < φg or the error of the 5th order is outside a user-defined tolerance, ∆Z is
reduced by a halving method. When φ = φg, Z∞ = Zc∞.

Sedimentation Zone (φ0 < φg, Zs(T) < Z < Zf, T ≤ Tc)

For the initial suspension un-networked case, the volume fraction in the
sedimentation zone, φs, and the height of the sedimentation zone, Zs(T), are
determined independently of the consolidation zone. By definition, ∆(φ) = 0 for φ <
φg. Thus, from equation 6.1.20, the solids flux in the settling zone, ψs(Z,T), is given
by:

φs
ψ s (Z , T ) = (1 − Z ) ; Z s (T ) < Z < Z f …(6.1.27)
B(φ s )

Integrating with respect to Z and substituting into equation 6.1.18 gives:

350

∂φ s  φ s  ∂φ s φ
− (1 − Z )  =− s …(6.1.28)
∂T  B(φ s )  ∂Z B (φ s )

Equation 6.1.28 is a first order non-linear partial differential equation that is


solved using the method of characteristics (Hildebrand, 1952). Parameterising φs(Z,T)
to φs(ξ) and equating coefficients with equation 6.1.28 gives

B(φs )
Z = 1 − C1 (ξ ) …(6.1.29)
φs

φ s B (φ )
T =− ∫ dφ + C2 (ξ ) …(6.1.30)
φ0
φ

where the functions C1(ξ) and C2(ξ) and the lower limit of integration, φ0, depend on
the initial and boundary conditions. The solution depends on the functional form of
B(φ) from φ0 < φ < φg and the relative value of φ0, such that the discontinuity at Zc(T)
can be a shock, a fan, or a combination of both (Anestis and Schneider, 1983, Bürger
and Concha, 2001). The simplest scenario to consider involves a kinematic shock at
Zc(T) (corresponding to Case Ia by Anestis and Schneider, 1983). By using the initial
condition (equation 6.1.21), equation 6.1.28 reduces to an ordinary differential
equation, such that φs is independent of radius and is function of time only:

dφ s φ
=− s …(6.1.31)
dT B(φ s )

The initial condition for this ordinary differential equation is φs(0) = φ0. The
top of the sedimentation zone, Zs(T), is discontinuous and the velocity of the shock is
given by integrating the conservation of volume (equation 6.1.23) with respect to Z
from the top of the sediment at Zs- to the bottom of the clear-liquor zone at Zs+.
Substituting for ψs and φs, and recognising that ψ and φ are zero in the clear-liquor
zone (at Zs+) gives:

dZ s (1 − Z s )
=− …(6.1.32)
dT B(φ s )

351
Substituting equation 6.1.32 into equation 6.1.31 to eliminate T and integrating
from φs(Zf) = φ0 to φs(Zs(T)) gives an explicit relationship between Zs(T) and φs(T) that
is dependent on Zf and φ0:

φ0
Z s (T ) = 1 −
φ s (T )
(1 − Z f ) …(6.1.33)

Since B(φ) decreases as φ decreases, the sediment clarifies quicker for lower
solids concentrations. If the aim of the process is to clarify a liquid rather than
consolidate a cake, low initial solids concentrations are preferable. Zs(T) is dependent
upon the initial loading of the centrifuge as well as the sedimentation characteristics
of the solid phase. As Zf approaches unity (that is, the centrifuge is initially
completely full of suspension), Zs(T) changes very little since the particles at the
centre of the centrifuge experience little or no acceleration.

At a theoretical time, Ts, the top of the sediment reaches the bowl wall, such
that Zs(Ts) = 0. Ts represents a limit for the solution of the sedimentation zone since it
is always greater than Tc, the time when Zc(T) = Zs(T).

φ s (Ts ) = φ0 (1 − Z f ) …(6.1.34)

Given an analytical function for B(φ), equation 6.1.31 can be solved explicitly.
However, for the sake of generality, since B(φ) may be given as an interpolating or
non-analytical function, it is solved here using the 4th-5th order Runge-Kutta
numerical technique. Starting at φs(0) = φ0, equation 6.1.31 is evaluated in 4th-order
steps of ∆T. Zs is evaluated at each time step using equation 6.1.33, and the process is
repeated until Zs = 0. If a step of ∆T gives a negative value for Zs or the error of the
5th-order is too large, the step size is reduced using an interval halving method. The
accuracy of the numerical method is checked by comparing the calculated φs value
when Zs = 0 with φs(Ts) from equation 6.1.34.

352
Consolidation Zone (φ0 < φg, 0 < Z < Zc(T), T ≤ Tc)

A discontinuity exists at Zc(T) in the initial suspension un-networked case.


The velocity of the shock at Zc(T) is given by integrating equation 6.1.18 from Zc- to
Z c +:

( ) (
dZ c ψ s Z c+ , T − ψ Z c− , T
=
)
; T ≤ Tc …(6.1.35)
dT φ g − φ s (T )

The global conservation equation for the un-networked case is:

Q(Z c ,T ) + (Z s − Z c )φ s (T ) = Z f φ 0 T ≤ Tc …(6.1.36)

where Q(Z,T) is the cumulative volume balance:

∂Q
=φ …(6.1.37)
∂Z

Due to the discontinuity, a change of variables to X = Z/Zc is made such that


equations 6.1.17 and 6.1.18 become:

∂φ Z  φ ψ 
=− c  −  …(6.1.38)
∂X ∆ (φ )  B(φ ) 1 − XZ c 

∂ψ ∂φ dZ c ∂φ
= Zc −X …(6.1.39)
∂X ∂T dT ∂X

A numerical method is used to solve the non-linear partial differential


equations that govern the behaviour in the consolidation zone. A forward difference
approximation in time is made such that equations 6.1.38 and 6.1.39 become ordinary
differential equations of φ and ψ with respect to X:

dφ Z  φ ψ 
=− c  −  …(6.1.40)
dX ∆ (φ )  B(φ ) 1 − XZ c 

dψ φ ( X ,T + ∆T ) − φ ( X ,T ) dZ c dφ
= Zc −X …(6.1.41)
dX ∆T dT dX

353
The numerical scheme, illustrated in Figure 6.1.2, involves taking fixed steps
in the bowl wall volume fraction of size ∆φ. ∆φ is chosen as the fixed variable rather
than ∆T since φ(0,T) initially increases dramatically, as is shown later in the results.
For each step, there are two unknown variables, ∆T* and dZc/dT*, which are solved
iteratively using an interval halving technique.

The upper and lower bounds for these variables (dZc/dThigh, dZc/dTlow, ∆Thigh,
and ∆Tlow) for the first step are given by the small-time solution, which is derived later
in this section. For the first iteration at each step, dZc/dT* is well bounded by the
result from the previous step, dZc/dT<, and 0. The lower initial bound of ∆T* for each
step is the previous result, ∆T<. However, the upper bound for ∆T* is not well defined
and is assumed here to be no more than twice the previous result. Zc* is approximated
to order ∆T3 for each value of ∆T* and dZc/dT* by the trapezium rule:

∆T *  dZ c * dZ c 
Z *c (T + ∆T ) = +  + Z (T ) …(6.1.42)
2  dT  c
dT T
 

Equations 6.1.40 and 6.1.41 are solved from X = 0 (where φ(0,T+∆T*) = φ(0,T)
+ ∆φ and ψ(0,T+∆T*) = 0) using the 4th-5th order Runge-Kutta technique until φ = φg,
dφ/dX = 0 or X =1. If φ(X,T+∆T*) = φg for 0 ≤ X < 1, ∆T* is too large and becomes the
new upper bound for ∆T. If φ > φg, ∆T* is too small and becomes the new lower
bound for ∆T.

∆T* is iterated upon until φ(1,T+∆T*) = φg. ψ(1,T+∆T*) is then used in


equation 6.1.35 to give an iteration test value for dZc/dT, dZc/dTtest:

dZ c
= s c
( )
ψ Z * ,T + ∆T − ψ (1,T + ∆T )
…(6.1.43)
dT test φ g − φ s (T + ∆T )

If dZc/dT * > dZc/dTtest,, dZc/dT * is too high and becomes the new upper bound
for dZc/dT. If dZc/dT * < dZc/dTtest, then the estimate is too low and becomes the new
lower bound for dZc/dT. dZc/dT* is iterated upon until it is equal to the test value to
the desired precision. For each iteration of dZc/dT*, the upper and lower bounds are
reset to ∆T< and 0 respectively and ∆T* is re-evaluated.

354
1-D Solid-Bowl Batch Centrifuge Model, φ0 < φg
Iterative 4th – 5th order Runge-Kutta Algorithm for T ≤ Tc

Material Characteristics: Py(φ), R(φ), φg, ∆ρ


Operating Conditions: φ0, V0, rb, ω, f

r2 V0 (1 − φ0 )2 2 Py (φ ) R(φ ) (1 − φ0 )2 f ′(φ )
Scalings: Z = 1 − ;Zf = ; T = 2∆ρω 2 t ; f (φ ) = ; B(φ ) = ; ∆(φ ) =
rb2 πrb2 R(φ0 ) ∆ρω rb2 2 R (φ )
0 (1 − φ )
2 B (φ )

Equilibrium f (φ ) = Z φ ; Solve dφ = − φ from φ (0 ) = φ to φ Z ∞ = φ


Distribution:
∞ f 0
dZ ∞ f ′(φ )
∞ c 0 ( )
Sedimentation Solve dφ s = − φ s and Z (T ) = 1 − φ0 (1 − Z ) from φ (0 ) = φ until Z = 0
B (φ s )
s f s 0 s
Zone: dT φs

Initial Conditions: T = 0; φ (Z ,0 ) = φ0 ; Z c (0 ) = 0; Z s (0 ) = Z f

φ0  φ0 
φ (0 ,T ) = φ g 1 + ; Z c (T ) =
Small-Time T 
Approximation: 

(φ g − φ 0 ) f ′ (φ g )

(φ g − φ0 )
T

Step Size and ∆φ = (φ ∞ − φ 0 ) ; dZ c dZ


= 0; c =
φ0
; ∆Tlow = 0 ; ∆Thigh = 0.01; ∆Tmax = 0.01
Iteration Bounds: 100 dT low dT high (φ g − φ0 )

dZc/dT* dZ c * = 1  dZ c +
dZ c 

Iteration: dT 2  dT high dT low 

∆T * =
∆Thigh + ∆Tlow
2
; Z *c =
∆T *  dZ c *
2  dT
+
dZ c 
dT T 
( ) (
+ Z c (T ); φ 0 ,T + ∆T * = φ (0 ,T ) + ∆φ ; ψ 0 ,T + ∆T * = 0 )
 

dφ Z*  φ ψ  dψ φ ( X , T + ∆T *) − φ ( X , T ) dZ c * dφ
Solve =− c  − and = Z c* −X
Runge-Kutta dX ∆ (φ )  B(φ ) 1 − XZ c*  dX ∆T * dT dX
Algorithm: dφ
from X = 0 until φ = φ g , = 0 or X = 1
dX

Ifφ > φ g , ∆Tlow = ∆T * ; If X < 1, ∆Thigh = ∆T *

dZ c (
ψ Z * ,T + ∆ T * − ψ 1 ,T + ∆ T *) ( )
dT test
= s c
φ g − φ s T + ∆T * ( )
dZ c * dZ c dZ c dZ c * dZ c * dZ c dZ c dZ c *
If > , = ; If < , = ; ∆ Tlow = 0 ; ∆ T high = ∆ T <
dT dT test dT high dT dT dT test dT low dT

dZ c dZ c *
∆T < = ∆T ; ∆T = ∆T * ; T = T + ∆T ; Z c (T ) = Z *c ; =
dT dT

dZc dZ dZ
If Zc (T ) < Z s (T ), go to next step; = 0; c = c ; ∆Tlow = 0; ∆Thigh = 2∆T
dT low dT high dT
∆φ
If ∆T > ∆Tmax or Zc (T ) > Z s (T ), ∆φ =
2

Figure 6.1.2: Numerical algorithm for batch centrifugation, initial suspension un-networked case
(T ≤ Tc)

355
By evaluating Q(Z,T) with each step, equation 6.1.36 gives a check on the
accuracy of the iterative numerical technique at each time step.

After a certain time, Tc, the sedimentation zone disappears. Once a time step
gives a result of Zc(T+∆T) > Zs(T+∆T), ∆φ is halved (and the initial upper and lower
bounds for the iteration variables appropriately adjusted) and the algorithm is repeated
until Zc(T+∆T) = Zs(T+∆T) to the desired accuracy. Likewise, the accuracy of the
algorithm decreases as ∆T increases. Therefore, if ∆T > ∆Tmax, ∆φ is halved for the
next step. ∆Tmax is chosen as 0.01.

Consolidation Zone (φ0 < φg, 0 < Z < Zc(T), T > Tc)

After Tc, Zc(T) decreases since there is no material being added to the
consolidating cake. The top of the cake remains at φg since there are no compressive
forces to increase the concentration. The velocity of the shock at Zc is given by
setting φs = ψs = 0 in equation 6.1.35:

dZ c
=−
(
ψ Z c− ,T )
; T > Tc …(6.1.44)
dT φg

The global solids conservation equation when there is no sedimentation zone


is:

Q(Z c ,T ) = Z f φ 0 ; T > Tc …(6.1.45)

There is no need to make the change of variables to X since Zc is decreasing


and φ(Z,T) exists for Zc(T) < Z < Zc(T+∆T). A forward difference approximation in
time is used to convert equations 6.1.17 and 6.1.18 to two coupled ordinary
differential equations of φ and ψ with respect to Z:

dφ 1  φ ψ 
=− −
∆ (φ )  B(φ ) 1 − Z 
 …(6.1.46)
dZ

d ψ φ ( Z , T + ∆T ) − φ ( Z , T )
= …(6.1.47)
dZ ∆T

356
For a given step of ∆T, there is one unknown, ∆φ*, which is solved using an
interval halving iterative approach as outlined in Figure 6.1.3. Rather than fixing ∆φ
and iterating on ∆T* as in the previous scheme, ∆T is fixed in this formulation since
∆φ* is well bounded. The initial upper and lower bounds for ∆φ* at each time step are
(φ∞ - φ(Z,T)) and 0 respectively.

1-D Solid-Bowl Batch Centrifuge Model, φ0 < φg


Iterative 4th – 5th order Runge-Kutta Algorithm for T > Tc

T = Tc ; φ (Z ,Tc ); Z c (Tc ) = Z s (Tc ); Q (Z c ,Tc ) = Qc

Step Size and ∆T = 0.01; ∆φ low = 0; ∆φ high = φ ∞ − φ (0 ,Tc )


Iteration Bounds:

∆φ* ∆φ high + ∆φ low


∆φ * = ; φ (0 ,T + ∆T ) = φ (0 ,T ) + ∆φ * ; ψ (0 ,T + ∆T ) = 0 ; Q(0 ,T + ∆T ) = 0
Iteration: 2

dφ 1  φ ψ  dψ φ (Z , T + ∆T ) − φ (Z , T ) dQ
Solve =−  − ; = and =φ
Runge-Kutta dZ ∆(φ )  B(φ ) 1 − Z  dZ ∆T dZ
Algorithm: dφ
from Z = 0 until φ = φ g , Z = Z c (T ), Q = Qc or =0
dZ

If φ > φ g , ∆φ high = ∆φ * ; If Q < Qc , ∆φ low = ∆φ *

∆φ = ∆φ * ; T = T + ∆T ; Z c (T ) = Z

If φ (0,T ) < fφ∞ , go to next step; ∆φlow = 0; ∆φhigh = φ∞ − φ (0,T )

Figure 6.1.3: Numerical algorithm for batch centrifugation, initial suspension un-networked case
(T > Tc)

There is no need for an estimate of dZc/dT in equations 6.1.46 and 6.1.47.


Therefore, the convergence of the iteration is based on the overall solids conservation.
Equations 6.1.37, 6.1.46 and 6.1.47 are solved from Z = 0 (where φ(0,T+∆T) = φ(0,T)
+ ∆φ*, and ψ(0,T+∆T) = Q(0,T+∆T) = 0) until φ(Z,T+∆T) = φg, dφ/dZ ≥ 0 or Q ≥ Zf
φ0. If either of the last two conditions are met and φ > φg, ∆φ* is too large. If
φ(Z,T+∆T) = φg, and Q < Zf φ0, ∆φ* is too small. ∆φ* is iterated until Q = Zf φ0 to the
desired precision. The accuracy of the numerical scheme is given by comparing
equation 6.1.44 with the trapezoidal approximation of dZc/dT (see equation 6.1.42).

357
Analytical similarity series solution for small times (φ0 < φg, T << 1)

The leading order terms of the analytical similarity series solution are
determined by assuming that an infinitely thin cake of concentration φg forms at the
bowl wall at T = 0. Asymptotic results for Zc and φ are found using the following
series expansions:

Z c (T ) = βT + Ο(T 2 ) …(6.1.48)

( )
φ ( X , T ) = φ g + TΦ ( X ) + Ο T 2 …(6.1.49)

Interestingly, the first order term for Zc varies with T. This occurs with
pressure filtration with membrane resistance (Landman, et al., 1991), whereas for
batch settling (Buscall and White, 1987) and pressure filtration without membrane
resistance (Landman and White, 1997), the asymptotic results for Zc vary with T½.
Substituting equations 6.1.48 and 6.1.49 into equations 6.1.38 and 6.1.39 and
rearranging gives the second derivative of Φ(X) as a function of higher order terms of
T (assuming that ∆(φg) > 0).

d 2Φ
=0 …(6.1.50)
dX 2

The volume fraction at the top of the cake is at φg, thus Φ(1) = 0. The
boundary condition at the bowl wall is given by ψ(0,T) = 0. Substituting this into
equation 6.1.38 and eliminating higher order terms of T gives:

dΦ βφ g
=−
dX 0 f ′ φg ( ) …(6.1.51)

Thus, the solution for Φ(X) is:

βφ g
Φ (X ) =
( ) (1 − X )
f ' φg
…(6.1.52)

β is found using the equation to the shock at Zc (equation 6.1.35), which


requires small time approximations for the sedimentation zone:

358
φ s (T ) = φ0 − Ο (T ) …(6.1.53)

( )
ψ s Z c+ ,T = φ0 − Ο (T ) …(6.1.54)

An expression for ψ(1,T) is found from the small-time approximation of


equation 6.1.38:

(
ψ Z c− ,T = ) Bφ(φg ) + ∆(φβ g ) ∂∂ΦX + Ο (T ) …(6.1.55)
g X =1

Substituting equations 6.1.52 to 6.1.53 into equation 6.1.35 and eliminating


higher order terms of T gives:

φ0
β=
(φ g − φ0 ) …(6.1.56)

Thus, the height of the consolidating bed and the volume fraction at the bowl
wall (X = 0) for small times are given by:

φ0
Z c (T ) = T << 1
(φ g − φ0 ) T ; …(6.1.57)

 φ0 
φ (0 ,T ) = φ g 1 +
T
 ; T << 1
 ( ) ( )
φ g − φ 0 f ' φ g 
…(6.1.58)

Equations 6.1.57 and 6.1.58 are used to give the bounds for the first step in the
algorithm.

Case 2: Initial Suspension Networked (φ0 > φg)

The initial suspension networked case requires a different solution method to


the un-networked case. The same governing equations (6.1.17, 6.1.18 and 6.1.20) are
used (that is, there is no change in the underlying assumptions of material behaviour)
but the boundary conditions to the zones change. Unlike the φ0 < φg case, the solution
in the sedimentation zone is coupled with the consolidation zone results, since φ(Zc,T)
and ψ(Zc,T) are continuous.

359
Sedimentation Zone (φ0 > φg, Zc < Z < Zs)

The suspension slumps at constant concentration, φs = φ0, in the sedimentation


zone since Pp is less than Py(φ0). Therefore, from equation 6.1.18, the gradient of the
solids flux is zero and ψs is constant with respect to Z. Therefore, the pressure
gradient is given by equation 6.1.20. Substituting φ0 and ψs(T) gives the pressure
gradient in the sedimentation zone:

∂Pp ψ s (T )
= − φ0 …(6.1.59)
∂Z 1− Z

Integrating equation 6.1.59 from Pp(Zc,T) = f(φ0) to Pp(Zs,T) = 0 gives the


relationship between Zc and Zs:

 1 − Zs 
f (φ0 ) = ψ s (T ) ln  + φ0 (Z s − Z c ) …(6.1.60)
 1 − Zc 

The velocity of the shock at Zs(T), dZs/dT, is given by integrating equation


6.1.18 from Zs- to Zs+:

dZ s ψ (T )
=− s …(6.1.61)
dT φ0

The global solids conservation equation for the φ0 > φg case is:

Zc
∫ φdZ + (Z s − Z c )φ0 = Z f φ 0 …(6.1.62)
0

Eliminating Zs from equations 6.1.60 and 6.1.62 and substituting Q(Zc,T)


gives:

 Z f φ0 − Q(Z c ,T ) 
f (φ0 ) = ψ s (T ) ln 1 −  + Z f φ0 − Q(Z c ,T ) …(6.1.63)
 φ0 (1 − Z c (T )) 

Equation 6.1.63 couples the consolidation and sedimentation zones, since


ψs(T) = ψ(Zc-,T).

360
Steady-state solution (φ0 > φg, T→∞)

As T → ∞, the sedimentation zone will not disappear for the φ0 > φg case since
there will always be material above the cake where the compressive forces acting
upon the network are less than the strength of the network. The derivation of the
steady-state solution for the φ0 > φg case uses the same condition as for the φ0 < φg
case, that is, that the solids flux, ψ, is zero everywhere. From equation 6.1.60, the
height of the sedimentation zone at equilibrium, Zs∞ is:

f (φ0 )
Z s∞ = + Z c∞ …(6.1.64)
φ0

Setting ψ = 0 in equation 6.1.20 gives the particle pressure gradient throughout


the suspension at equilibrium:

dPp
= −φ …(6.1.65)
dZ

Integrating equation 6.1.65 from Pp(0) = f(φ∞) to Pp(Zs∞) = 0 and equating to


the conservation of volume (equation 6.1.63) gives f(φ∞) from equation 6.1.26. φ(Z∞)
in the consolidation zone is given by solving equation 6.1.24 from φ(0) = φ∞ until
φ(Zc∞) = φ0. Zs∞ is then given by equation 6.1.64.

Consolidation Zone (φ0 > φg, 0 < Z < Zc(T))

There is no need to change variables to X for the φ0 > φg case since φ(Zc(T),T)
is continuous. Therefore, φ(Z,T) in the consolidation zone is given by equations
6.1.46 and 6.1.47. The numerical algorithm for the networked case is shown in Figure
6.1.4. For a given step of ∆φ (typically chosen as 0.01 (φ∞ - φ0)), there is one
unknown, ∆T*, which is solved using an interval halving iterative approach for
successive estimates. The upper and lower bounds for ∆T*, ∆Thigh and ∆Tlow, for the
first step are given by the small-time solution, which is derived later in this section.
The initial lower bound for ∆T* for each subsequent step is the value from the
previous step ∆T< (providing that ∆φ is constant), while the initial upper bound for
∆T* for each step is chosen as 2 ∆T<.

361
1-D Solid-Bowl Batch Centrifuge Model, φ0 > φg
Iterative 4th – 5th order Runge-Kutta Algorithm

Material Characteristics: Py(φ), R(φ), φg, ∆ρ


Operating Conditions: φ0, V0, rb, ω, f

r2 V0 (1 − φ0 )2 2 Py (φ ) R(φ ) (1 − φ0 )2 f ′(φ )
Scalings: Z = 1 − ;Zf = ; T = 2∆ρω 2 t ; f (φ ) = ; B(φ ) = ; ∆ (φ ) =
rb2 πrb2 (
R φ0 ) ∆ρω rb2 2 R (φ0 ) (1 − φ )2 B(φ )

Equilibrium f (φ ) = Z φ ; Solve dφ = − φ from φ (0 ) = φ to φ Z ∞ = φ ; Z ∞ = f (φ0 ) + Z ∞


Distribution:
∞ f 0
dZ ∞ f ′(φ )
∞ c 0 s
φ0
c ( )

Initial Conditions: T = 0 ; φ (Z ,0 ) = φ 0 ; Z c (0 ) = 0; Z s (0 ) = Z f

f (φ 0 ) − f (φ ∞ ) λ
; Solve π Ae A erf ( A) =
2
λ=
(
φ 0 ln 1 − Z f ) 1− λ
for A
Small-Time
Approximation:  2 Ae A (1 − λ ) 
2

φ (0, T ) = φ 0 1 + T
 ; Z c (T ) = 2 A f ′(φ 0 ) T ; Z s (T ) = Z f − λT
 f ′(φ 0 ) 
 

2
Initial Step Size
(φ − φ )   Zf 
2
∆φ
and ∆φ = ∞ 0 ; ∆Tlow = f ′(φ0 ) 
; ∆T = f ′(φ )  ; ∆Tmax = 0.01
 2φ Ae A2 (1 − λ ) 
0 
2 A 
high
100 
Iteration Bounds:  0 

∆T*
Iteration:
∆T * =
∆Thigh + ∆Tlow
2
( ) ( ) (
; φ 0 ,T + ∆T * = φ (0 ,T ) + ∆φ ; ψ 0 ,T + ∆T * = 0 ; Q 0 ,T + ∆T * = 0 )

Solve

=−
1  φ
 − ; =
(
ψ  dψ φ Z , T + ∆T * − φ (Z , T )
and
dQ )

Runge-Kutta dZ ∆(φ )  B(φ ) 1 − Z  dZ ∆T * dZ
Algorithm: dφ
from Z = 0 until φ = φ0 or =0
dZ

 Z f φ 0 − Q (Z ) 
f test = ψ (Z ) ln 1 −  + Z f φ 0 − Q (Z )
 φ 0 (1 − Z ) 

If φ > φ 0 or f test < f (φ 0 ), ∆Tlow = ∆T * ; If f test > f (φ 0 ), ∆Thigh = ∆T *

Q (Z c )
∆T = ∆T * ; T = T + ∆T ; Z c (T ) = Z ; Z s (T ) = Z f − + Z c (T ); ψ s (T ) = ψ (Z c ,T )
φ0

If φ (0,T ) < fφ∞ , proceedto next step; ∆Tlow = ∆T ; ∆Thigh = 2∆T


∆φ ∆T
If ∆T > ∆Tmax , ∆φ = ; ∆Tlow =
2 2

Figure 6.1.4: Numerical algorithm for batch centrifugation, initial suspension networked case

362
Equations 6.1.37, 6.1.46 and 6.1.47 are solved from Z = 0 (where φ(0,T+∆T) =
φ(0,T) + ∆φ, and ψ(0,T+∆T) = Q(0,T+∆T) = 0) until φ(Z,T+∆T) = φ0 or dφ/dZ = 0. If
the latter condition is met and φ(Z,T+∆T) > φ0, ∆T* is too small. If φ(Z,T+∆T) = φ0,
the right-hand side of equation 6.1.63 is evaluated using Z, ψ(Z,T+∆T) and
Q(Z,T+∆T). If the solution is less than f(φ0), ∆T* is too small, whereas if the solution
is greater than f(φ0), ∆T* is too large. ∆T* is iterated until equation 6.1.63 is true to the
required precision. The accuracy of the numerical solution is indicated by evaluating
ψs(T) from equation 6.1.61 using an appropriate approximation to dZs/dT and
comparing to ψc(Zc-,T).

Analytical similarity series solution for small times (φ0 > φg, T << 1)

The following asymptotic results hold for Zs(T), Zc(T) and φ(Z,T):

Z s (T ) = Z f − λT − Ο T 2 ( ) …(6.1.66)

Z c (T ) = α T + Ο (T ) …(6.1.67)

 
φ (Z ,T ) = φ 0 + T Φ   + Ο (T )
Z
…(6.1.68)
 Z c (T ) 

Notice that the small time behaviour for Zc varies with T½ rather than T. Thus
all the time derivatives tend to infinity as T → 0. Substituting these in the governing
equations in terms of the similarity variable, X = Z/Zc(T) (equations 6.1.38 and 6.1.39)
and eliminating higher order terms gives the following non-linear second-order
ordinary differential equation in Φ(X):

1 d 2Φ dΦ
+X − Φ (X ) = 0 …(6.1.69)
2 2 dX
2 A dX

where

α
A= …(6.1.70)
2 f ′(φ 0 )

363
The boundary conditions for equation 6.1.70 are determined from φ(Zc,T) = φ0,
ψ(0,T) = 0 and from the global conservation (equation 6.1.62):

Φ (1) = 0 …(6.1.71)

2 Aφ 0
Φ ′(0 ) = − …(6.1.72)
f ′(φ0 )

2 Aφ 0
Φ ′(1) = − (1 − λ ) …(6.1.73)
f ′(φ0 )

The solution is:

2 Aφ 0  A2  1 − A2 X 2  
Φ (X ) = − X + (1 − λ ) π e  AXerfAX + e 
f ′(φ 0 )   π 
…(6.1.74)

Thus, the series approximation for the volume fraction for small times is given
by:

 A2
(1 − λ ) 
φ (0 ,T ) = φ0 1 + T
 2 Ae
…(6.1.75)
 f ' (φ 0 ) 
 

where A is the solution to:

2 λ
π Ae A erfA = …(6.1.76)
1− λ

Substituting the series approximations for Zc(T) and Zs(T) into equations
6.1.60 and 6.1.61 and eliminating higher order terms gives the solution for λ:

f (φ 0 ) − φ 0 Z f
λ=
(
φ 0 ln 1 − Z f ) …(6.1.77)

364
6.1.2 Results

The material characteristics for the average ferric water treatment sludge
described in Chapter 4 have been used here to illustrate the batch centrifuge models.
The material characteristics were extrapolated using the fitted power-law functional
forms for R(φ) values below 0.004 v/v and therefore represented a simplistic case
where the solution to equation 6.1.28 was always a shock at Zc(T). The results for
fixed operating conditions (rb = 0.5 m, rf = 0.25 m and ω = 2000 rpm) with a range of
initial concentrations (above and below φg) were investigated.

Case 1: Initial Suspension Un-networked (φ0 < φg)

Figure 6.1.5 shows the volume fraction distribution results for the batch
centrifuge model with φ0 = 0.001 and 0.002 v/v, such that the initial suspension was
un-networked. The results illustrate the build-up of the cake and the decrease of the
sediment volume fraction with time. The concentration jumped from φs to φg at the
cake/sediment interface.

(a) φ 0 = 0.001 v/v (b) φ 0 = 0.002 v/v


0.05 0.06
Solids Volume Fraction, φ (Z ,T )

φ ∞ = 0.0546 v/v
φ ∞ = 0.0437 v/v 0.05
0.04

0.04
Equilibrium
0.03 distribution Equilibrium
1.39 distribution
Tc = 1.34 1.42
1.3
0.03 Tc = 1.32
1.2 1.2
0.02 1.1 1.1
1.0 1.0
0.9
0.02 0.9
0.8 0.8

0.01
0.01
0.01

0.05

φg φg
0.1

0.2

0.3

0.4

0.5
0.6
0.7

0.01

0.05

0.1

0.2

0.3

0.4

0.5
0.6
0.7

φ0 φ0
0 0
0 0.01 0.02 0.03 0 0.01 0.02 0.03 0.04 0.05
Scaled Radius, Z Scaled Radius, Z

Figure 6.1.5: Volume fraction distribution predictions for 1-D batch centrifuge model with
different initial concentrations (φ0 < φg, rb = 0.5 m, rf = 0.25 m and ω = 2000 rpm). The annotated
values correspond to the scaled time at each solution

365
As T approached and passed Tc, the cake concentration approached the
equilibrium distribution, indicating that the transient numerical solution was valid.
There were only very small changes in φ after Tc, such that the sedimentation time-
scale was greater than the consolidation time-scale.

(a) φ 0 = 0.001 v/v (b) φ 0 = 0.001 v/v


0.8 0.03

0.0009

0.0008

0.0007

0.0006

0.0005

0.0004

0.0003

Zs( T)
Zf
0.7
0.004
Z s (T ) Sedimentation zone 0.010
0.6 0.015
Scaled Radius, Z

0.02 0.020
0.5 Clear-liquor zone
(φ = 0) 0.025

Zc,small( T)
0.4
Sedimentation 0.030

0.3 zone Consolidation


0.01 zone 0.035

0.2
0.0009

0.0008

0.0007

0.0006

0.0005

0.0004

0.0003

0.1 Z c (T )
0.040

Z c (T ) Consolidation zone
0.043
0 0 Tc
0 (c) φ 0 = 0.002 v/v
0.5 1 1.5 0 0.5φ 0 = 0.002 v/v
(d) 1 1.5
0.8 0.05
0.0018

0.0016

0.0014

0.0012

0.0010

0.0008

0.0006

Zs( T)
Zf
0.7
Z s (T ) 0.04 Sedimentation zone 0.004
0.010
0.6 0.015
Scaled Radius, Z

0.020

0.5 Clear-liquor zone 0.025

(φ = 0) 0.03 0.030
Zc,small( T)

0.4 0.035
Sedimentation
zone 0.02 Consolidation 0.040
0.3
zone
0.045
0.2
0.01
0.0018

0.0016

0.0014

0.0012

0.0010

0.0008

0.0006

0.050
0.1 Z c (T )
Z c (T ) Consolidation zone
0.054
0 0
Tc
0 0.5 1 1.5 0 0.5 1 1.5
Scaled Time, T Scaled Time, T

Figure 6.1.6: Predictions of sedimentation and consolidation profiles for 1-D batch centrifuge
model with differing initial concentrations (φ0 < φg, rb = 0.5 m, rf = 0.25 m and ω = 2000 rpm).
The annotated values correspond to lines of constant concentration. (b) and (c) are enlarged
views of (a) and (d)

The concentration profile results for φ0 = 0.001 and 0.002 v/v are presented in
Figure 6.1.6 (a) and (c), with enlarged views of the consolidation zone in (b) and (d)
respectively. For these material characteristics and operating conditions, φs was a
function of T but independent of Z. The enlarged views show that the small time
approximation over-predicted Zc(T) and applied only for very small values of T, and
further illustrate the small changes in concentration after Tc.

366
Case 2: Initial Suspension Networked (φ0 > φg)

Figure 6.1.7 shows the volume fraction distribution results for φ0 > φg. φ was
initially constant throughout the centrifuge at φ0. With the onset of centrifugal
acceleration, φ(0,T) increased dramatically as the solids sedimented from the liquid to
form the cake. Zc(T) passed through a maximum and began to decrease since the
acceleration was a function of radius. As φ0 was increased, the final height and
equilibrium concentration increased.

(a) φ 0 = 0.005 v/v (b) φ 0 = 0.01 v/v


0.12
Solids Volume Fraction, φ (Z ,T )

0.1
φ ∞ = 0.0916 v/v

f
Z = 0.75
Equilibrium distribution
0.08 Equilibrium distribution φ ∞ = 0.0733 v/v 0.90
0.90 0.80
Z = 0.75

0.80 0.70
f

0.70 0.60
0.06 0.60 0.50
0.50 0.40
0.40 0.30
0.30 0.20
0.04 0.20 0.10
0.10 0.05
0.05 0.01
0.01
0.02
0.90

0.80

0.70

0.60

0.50

0.40

0.30

0.20

0.10
0.05
0.01
0.90

0.80

0.70

0.60

0.50

0.40

0.30

0.20

0.10
0.05
0.01

φ0 φ0
0
(c) φ 0 = 0.015 v/v (d) φ 0 = 0.02 v/v
0.12
φ ∞ = 0.1043 v/v φ ∞ = 0.1144 v/v
Solids Volume Fraction, φ (Z ,T )

0.1
f
Z = 0.75
Z = 0.75

Equilibrium distribution
f

1.00
0.90
Equilibrium distribution
0.80
0.08 0.90
0.70
0.80
0.60
0.70
0.50
0.60
0.40
0.50
0.06 0.40
0.30
0.20
0.30
0.10
0.20
0.05
0.10
0.04 0.05
0.01

0.01
0.80

0.70

0.60

0.50

0.40

0.30

0.20

0.10
0.05
0.01
0.80

0.70

0.60

0.50

0.40

0.30

0.20

0.10
0.05
0.01

0.02
φ0
0.90

φ0
0.90

0
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8
Scaled Radius, Z Scaled Radius, Z

Figure 6.1.7: Volume fraction distribution predictions for 1-D batch centrifuge model with
different initial concentrations (φ0 > φg, rb = 0.5 m, rf = 0.25 m and ω = 2000 rpm). The annotated
values correspond to the scaled time at each solution

367
φ(Z,T) approached the equilibrium distribution as T increased, indicating that
the numerical algorithm was valid. φs was constant at φ0 throughout the process. The
maximum in Zc(T) became more pronounced as φ0 was increased, since the
consolidation time-scale increased compared to the sedimentation time-scale.

(a) φ 0 = 0.005 v/v (b) φ 0 = 0.01 v/v


0.8

0.7 Zf Zf
Z s,small (T )
0.6 Z s,small (T )
Scaled Radius, Z

0.5 Z s (T ) Clear-liquor zone Z s (T ) Clear-liquor zone


(φ = 0) (φ = 0)
0.4
Sedimentation Sedimentation
0.3 zone
zone
(φ = φ 0 ) (φ = φ 0 )
0.2
0.034 0.026 0.034 0.042 0.050 0.058
0.028 0.040 0.046 0.018 0.066
0.016 0.022
0.1 0.052
Z c (T ) 0.074
Z c (T ) 0.058
0.064 Consolidation zone 0.082
Consolidation zone 0.070
0 Z c,small (T ) Z c,small (T )
0.090

0 (c) φ 0 = 0.015 0.8


0.4 v/v 1.2 0 (d) φ 0 = 0.02 v/v
0.4 0.8 1.2
0.8

0.7 Zf Zf

Z s,small (T ) Z s,small (T )
0.6
Scaled Radius, Z

Z s (T ) Z s (T )
0.5 Clear-liquor zone Clear-liquor zone
(φ = 0) (φ = 0)
0.4
Sedimentation Sedimentation
0.3 zone zone
(φ = φ 0 ) (φ = φ 0 ) 0.038 0.047
0.056
0.029
0.052
0.2 0.028 0.036
0.044 0.060

0.020
0.065
0.1 Z c (T ) Z c (T ) 0.074
0.068 Consolidation zone 0.083
Consolidation zone 0.076
0.084 0.092
0.092 0.101
0.100 0.110
0
Z c,small (T ) Z c,small (T )
0 0.4 0.8 1.2 0 0.4 0.8 1.2
Scaled Time, T Scaled Time, T

Figure 6.1.8: Predictions of sedimentation and consolidation profiles for 1-D batch centrifuge
model with differing initial concentrations (φ0 > φg, rb = 0.5 m, rf = 0.25 m and ω = 2000 rpm).
The annotated values correspond to lines of constant concentration

The concentration profile results for the networked case are presented in
Figure 6.1.8. The small time approximations for Zc(T) were less than the numerical
solutions and applied for T values up to 0.1. Zs,small(T) were greater than the numerical
solutions, and were accurate up to considerable values of T. These results represent

368
the first reported modelling of the initial suspension networked case of batch
centrifugation.

The transient sediment and cake height results are presented in Figure 6.1.9,
which clearly show the change in behaviour of the sedimentation zone between the
networked and un-networked cases. For the φ0 > φg case, Zs(T) was initially linear,
while it was non-linear for the φ0 < φg case. Since the time scaling was based on φ0,
the Zs(T) results for the two cases were initially equal. Figure 6.1.9 also shows the
increasing cake height with increasing φ0.

0.8
Z f = 0.75
0.7

0.6 Clear-liquor zone


Scaled Radius, Z

0.5
Z s (T )
0.4

0.3 Sedimentation
zone φ0
0.2 0.02
0.015
Z c (T ) 0.01
0.1
Consolidation zone 0.005
0.002
0 0.001
0 0.5 1 1.5
Scaled Time, T

Figure 6.1.9: Transient sediment and cake height results for one-dimensional batch centrifuge
model with different initial concentrations (rb = 0.5 m, rf = 0.25 m and ω = 2000 rpm)

The equilibrium distribution results are presented in Figure 6.1.10. The


sedimentation zone did not exist and the top of the cake was at φg for the φ0 < φg case,
whereas the sediment was present and at φ0 for the φ0 > φg case. As φ0 was increased,
φ∞ increased since the solids pressure at the bowl wall increased, and Zc∞ increased
since the total amount of solids increased. Thus, in general, high solids loadings are
necessary to achieve high cake concentrations in centrifugal operations.

369
0.12

Equilibrium Distribution, φ (Z∞ )


0.1

0.08 Increasing φ 0

0.06

0.04

0.02

0.0075

0.0125

0.0175
φg
0.001

0.002

0.003
0.004
0.005

0.015
0.01

0.02
0
0 0.05 0.1 0.15 0.2
Scaled Radius, Z

Figure 6.1.10: Equilibrium volume fraction distribution results for one-dimensional batch
centrifuge model with different initial concentrations (rb = 0.5 m, rf = 0.25 m and ω = 2000 rpm)

370
6.2 Pseudo-One-Dimensional Solid-Bowl
Continuous Centrifuge

This section describes the modelling of continuous, steady-state, centrifugal


thickening. This is the centrifugal analogy to the pseudo-one-dimensional model of
continuous thickening developed by Landman, et al., 1988. Side and axial
representations of a continuous tubular solid-bowl centrifuge with variable width are
shown in Figure 6.2.1.

y
Inlet ω
ω qi, φi

ro
Sedimentation Zone
Consolidation Zone

W(r)
Clear-liquor Zone

ru ri
rc
ri
rc
ro r

ru

Overflow r
Underflow qo
qu, φu
(a) Axial view (b) Lateral view

Figure 6.2.1: Schematic of 1-D continuous centrifugal thickening: (a) Axial view; (b) Lateral view

A continuous inlet feed of concentration φi enters the centrifuge at a


cylindrical plane at radius ri and feed rate qi. The centrifuge rotates at a constant
angular velocity, ω, causing the solid phase to sediment against the bowl wall and
form a cake due to centripetal acceleration. The concentrate or underflow (using
thickening terminology) is taken from the bowl at radius ru, concentration φu, and
flowrate qu. The centrate or overflow is at radius ro (where ro ≤ ri) and flowrate qo,
and is assumed to be clear of undissolved solids (φo = 0). The width of the centrifuge

371
at radius r is W(r). A(r) is the corresponding cross-sectional area at radius, r, given
by:

A(r ) = 2πrW (r ) …(6.2.1)

As for the batch centrifuge models, the volume fraction distribution, φ(r), is
dependent on φi relative to φg. Models are presented for the inlet suspension
networked (φi > φg) and the inlet suspension un-networked (φi < φg) cases. There are
three distinguishable zones of behaviour within the centrifuge:

- A consolidation zone or cake exists against the bowl of the centrifuge where
the concentration varies from φ(ru) = φu to φ(rc) = φg (or φ(rc) = φi in the φi > φg
case). A network of particles exists where the local stress on the network, pp,
is equal to the compressive yield stress of the network, Py(φ). The solid and
liquid velocities at the bowl are assumed to be equal.

- A sedimentation zone exists where the particles are in free-fall. The top of the
sedimentation zone is assumed to be at ri. The sedimentation zone behaves
differently for the networked and un-networked cases (Landman, et al., 1988):

o When φi > φg, a network exists but the consolidating forces are less
than the compressive yield stress, so the sedimentation zone is at a
constant concentration, φi. The network is able to transmit pressure,
which varies from pp(ri) = 0 to pp(rc) = Py(φi). φ(rc) is continuous and
equal to φi.

o When φi < φg, no network exists in the sedimentation zone and the
particle velocity is independent of Py(φ). The sediment concentration,
φs, is a function of radius, and φ(rc) is discontinuous, with the top of the
bed (φ(rc+)) at φg and φ(rc-) = φs(r).

- The third zone is the clear-liquor zone from ro to ri. The solid-liquid
separation is independent of ro since there are no solids in the clear-liquor zone
to exert a particle pressure, and the fluid height is constant.

372
For any function F(r,y,t) at position (r,y), a corresponding cross-sectionally
averaged quantity F (r , t ) can be defined such that the two-dimensional problem is
simplified to one-dimension. The solid-liquid separation behaviour is determined by
the mass and momentum conservation equations, which are given in vector notation in
Section 2.2.1. In this work, the vector equations are cross-sectionally averaged to a
radial coordinate and steady-state behaviour is assumed to model one-dimensional
continuous centrifugation. The boundary conditions for the different zones of
behaviour are identified and appropriate scalings are applied to give the solids
throughput as a function of underflow concentration for given operating conditions.

This model assumes that there are no internal discs, such as those found in a
disc-bowl centrifuge. Discs are used to aid sedimentation but require an
understanding of the shear effects at the disc surface in order to model, which is
beyond the scope of this work. Likewise, transient behaviour such as start-up, shut-
down and semi-continuous operation are not considered.

373
6.2.1 Model Formulation

Sedimentation-Consolidation Equations

Global Conservation Equations

Assuming incompressibility, the global conservation equation is given in


terms of the inlet, overflow and underflow throughputs (qi, qo and qu respectively):

qi = q o + q u …(6.2.2)

The particle throughput, qp, is related to the relative concentrations of the inlet
and underflow (φi and φu). The overflow concentration, φo, is assumed to be zero.

q p = φ i qi = φ u qu …(6.2.3)

Likewise, the overall fluid throughput, qf, is given by:

q f = (1 − φ i )qi = (1 − φ u )q u + q o …(6.2.4)

Thus, for a given inlet concentration, there are two unknowns, φu and qp,
required to specify the system.

Local Equations

The formulation for the local conservation equations is outlined in Chapter 2:


the local conservation of momentum equation in vector notation is given by equation
2.2.3, the conservation of mass of the particle and fluid phases are given by equations
2.2.6 and 2.2.7 respectively, while the irreversible kinetics of collapse is described by
equation 2.2.11.

Cross-sectionally averaged equations

For any function F(r,y,t) at position (r,y), a corresponding cross-sectionally


averaged quantity F (r , t ) is defined by:

374
F (r ,t ) = (2 )
∫ F (r , y ,t )d y
1
…(6.2.5)
A(r ) A(r )

Differentiation with respect to r gives:

∂ ∂F
∂r
[ A(r )F (r ,t )] = A(r )
∂r
+ A′(r )F (r , y ,t ) y =W (r ) …(6.2.6)

Cross-sectionally averaging the conservation of particle mass (equation 2.2.6)


gives:

∂φ
∂t
+
1
A(r ) A(r )
( )
(2 )
∫ ∇ . φu p d y = 0 …(6.2.7)

The divergence theorem is used to equate this volume integral in (r,y,t) to a


surface integral in (r,t):

+
(
∂φ 1 ∂ rφu p ,r
+
A′(r ) )
rφu p , y ( )=0 …(6.2.8)
∂t r ∂r rA(r ) y =W (r )

where up,r and up,y are the radial and axial components of up. Assuming that the
particle velocity at the wall is along the wall gives the kinematic constraint:

 u p ,y  dW
  = …(6.2.9)
 u p ,r 
  y =W (r ) dr

Substituting this boundary condition into equation 6.2.8 and using equation
6.2.6 gives the cross-sectionally averaged conservation of solids mass:

A(r )
∂φ 1 ∂
+
∂t r ∂r
[
rA(r )φu p ,r = 0 ] …(6.2.10)

Likewise, cross-sectionally averaging the conservation of fluid mass (equation


2.2.7) gives:

∂ (1 − φ ) 1 ∂
A(r )
∂t
+
r ∂r
[ ]
rA(r )(1 − φ )u f ,r = 0 …(6.2.11)

375
where uf,r is the radial component of uf.

Steady-state approximation

The process is continuous and the inlet concentration is assumed to be


constant. Thus, the steady-state approximation is introduced such that all functions
are independent of time. Therefore, equations 6.2.10 and 6.2.11 reduce to:

1 d
r dr
[
rA(r )φu p ,r = 0 ] …(6.2.12)

1 d
r dr
[
rA(r )(1 − φ )u f , r = 0 ] …(6.2.13)

It is necessary to make the assumption that φ(r,y) does not vary significantly
across A(r), such that:

φ (r , y ) ≈ φ (r ) = φ (r ) …(6.2.14)

Therefore:

φu p ,r = φ (r )u p ,r …(6.2.15)

(1 − φ )u f ,r = (1 − φ (r ))u f ,r …(6.2.16)

At the underflow radius, ru, the solid and liquid velocities are assumed to be
the same and are given in terms of the solids flux, qp,

qp
u p ,r (ru ) = u f ,r (ru ) = …(6.2.17)
A(ru )φ u

Using this boundary condition to evaluate equations 6.2.12 and 6.2.13 gives:

rA(r )φ u p ,r = ru q p …(6.2.18)

(1 − φ u )
rA(r )(1 − φ )u f ,r = ru q p …(6.2.19)
φu

Cross-sectionally averaging equation 2.2.3 with centrifugal acceleration gives:

376

φ
(1 − φ )
( )
R(φ ) u p ,r − u f ,r −
dp p
dr
+ φ∆ρω 2 r = 0 …(6.2.20)

Substituting equations 6.2.18 and 6.2.19 into equation 6.2.20 to eliminate up,r
and uf,r gives a first order non-linear ordinary differential equation for pp(r):

dp p ru q p R(φ )  φ 
= φ∆ρω 2 r −  1 −  …(6.2.21)
dr rA(r ) (1 − φ )2  φ u 

Cross-sectionally averaging the kinetic condition (equation 2.2.11) and using


the result from equation 6.2.18 gives the irreversibility condition:

 0 p p < Py (φ )
dφ  
=  rA(r )
dr  [
φκ (φ ) p p − Py (φ ) ] p p ≥ Py (φ )

…(6.2.22)
 ru q p 

Dimensionless Equations

Equations 6.2.21 and 6.2.22 are simplified by applying appropriate scaling


factors based on the centrifuge dimensions, the centrifugal buoyancy and the feed
concentration to leave the underflow concentration, solids throughput and bed height
to be determined. The radius, r (including ru, rc, ri and ro), is scaled with ru to give the
dimensionless length scale, Z (including 0, Zc, Zi and Zo), such that the origin is now
at the bowl wall:

r2
Z = 1− …(6.2.23)
ru2

pp and Py(φ) are scaled with the centrifugal buoyancy of the solid phase to give
Pp and f(φ):

2pp
Pp = …(6.2.24)
∆ρω 2 ru2

2 Py (φ )
f (φ ) = …(6.2.25)
∆ρω 2 ru2

377
The hindered settling function is scaled with the inlet concentration to give
B(φ):

B(φ ) =
R(φ ) (1 − φi )2 …(6.2.26)
(1 − φ )2 R(φi )

The cross-sectional area is scaled with the underflow radius:

A(r )
α (Z ) = …(6.2.27)
A(ru )

The appropriate scaled particle throughput, Qp, and scaled dynamic


compressibility, Κ(φ), are:

R(φi )
QP = qp …(6.2.28)
∆ρω 2 ru A(ru )(1 − φi )2

r 2 R (φi )
Κ (φ ) = u κ (φ ) …(6.2.29)
4(1 − φi )2

Substituting equations 6.2.23 to 6.2.29 into equations 6.2.21 and 6.2.22 gives
the scaled particle pressure gradient, dPp/dZ, and the concentration gradient, dφ/dZ:

dPp Q p B(φ )  φ 
= −φ + 1 −  …(6.2.30)

α (Z )(1 − Z )  φu 
dZ 

 
 0 Pp < f (φ )
dφ  
=  …(6.2.31)
dZ α (Z )
 [
φΚ (φ ) Pp − f (φ ) ] 
Pp ≥ f (φ )
 Qp 

If it is assumed that the drainage of the suspending fluid is rate-determining


(rather than the breaking and/or reformation of interparticle bonds) such that Κ(φ) is
very large, then Pp ≤ f(φ). Thus, the concentration gradient is given by:

378
 
 0 Pp < f (φ )
dφ  
=  …(6.2.32)
dZ  
1  φ Qp  φ 
−  − 1 −  Pp = f (φ )
 ∆(φ )  B(φ ) α (Z )(1 − Z )  φu  

∆(φ) is defined by equation 6.1.19. The boundary conditions for this non-
linear non-homogeneous first order ordinary differential equation depend on whether
the inlet suspension is networked or un-networked. For φi < φg, the sedimentation
zone has no network strength and the top of the bed is at the gel point. For φi > φg, the
sedimentation zone has network strength but is below the yield stress of the inlet
material and the top of the bed is at the inlet concentration. These two scenarios are
now investigated separately.

Case 1: Inlet Suspension Un-networked (φi < φg)

Sedimentation Zone (Zc < Z ≤ Zi)

Since the inlet suspension is un-networked, Pp(Z) = 0 and dPp/dZ = 0 in the


sedimentation zone. Rearranging equation 6.2.30 gives:

Q p B(φ )  φ 
−φ + 1 −  = 0; Z c < Z ≤ Z i …(6.2.33)
α (Z )(1 − Z )  φu 

Rearranging equation 6.2.33 and evaluating at φ(Zi) = φi gives Qp:

 φ φ 
Q p = α (Z i )(1 − Z i ) u i  …(6.2.34)
 φu − φi 

Therefore, the solids flux for a given inlet concentration and geometry is a
function of underflow concentration only for the φi < φg case, and is independent of
the cake height, whereas Zc is dependent upon φu and Qp. The throughput is
maximum in the trivial case when φu = φg and the centrifuge acts as a clarifier; the
minimum throughput is when φu is a maximum (when Zc = Zi). Substituting equation
6.2.34 into equation 6.2.33 gives Z as a function of φ in the sedimentation zone:

379
φi  φu − φ 
α (Z )(1 − Z ) = α (Z i )(1 − Z i )B(φ )  
 φ − φ ; Z c < Z ≤ Z i …(6.2.35)
φ  u i

Consolidation Zone (0 < Z ≤ Zc)

At Zc, φ is discontinuous and jumps to φg, and dφ/dZ becomes negative.


Within the cake, the particle pressure is equal to the compressive yield stress,
therefore the volume fraction distribution, φ(Z), is given by the solution of dφ/dZ
(equation 6.2.32) upon substitution of Qp (equation 6.2.34):

dφ 1  φ α (Z i )(1 − Z i )  φ u − φ 
=−  − φi   ; 0 < Z ≤ Z c
dZ ∆ (φ )  B(φ ) α (Z )(1 − Z )  φ u − φ i 
…(6.2.36)

Equation 6.2.36 is solved numerically from φ(0) = φu to φ(Zc) = φg, where Zc is


to be determined. The algorithm for calculating the solids throughput and the height
of the cake for the complete range of possible underflow concentrations is outlined in
Figure 6.2.2. For each known value of the underflow solids (starting from φu = φi, in
steps of ∆φu), φ(Z) in the consolidation zone is found from equation 6.2.36 by using
4th order Runge-Kutta steps from Z = 0. The accuracy of each step is checked by the
5th order, and the step size reduced if it is outside a given tolerance. If φ(Z) is less
than φg or dφ/dZ is greater than 0, ∆Z is halved. The steps of ∆Z are repeated until φ =
φg or Z = Zi. If φ(Zi) is greater than φg, the underflow solids is greater than the
possible maximum, and ∆φu is reduced. To give the complete range of possible
underflow concentrations, steps of ∆φu are repeated until the top of the cake is at the
feed plane (that is, φ(Zi) = φg). The volume fraction distribution in the sediment is
found by evaluating equation 6.2.35 for Z in steps of ∆φ from φ(Zi) = φi until Z = Zc or
φ = 0. If a step of ∆φ results in Z < Zc or φ < 0, ∆φ is halved.

380
1-D Solid-Bowl Continuous Centrifuge Model, φi < φg
4th – 5th order Runge-Kutta Algorithm

Material Characteristics: Py(φ), R(φ), φg, ∆ρ


Operating Conditions: ru, ri , φi, A(r), ω

Scalings:
r2 A(r ) 2 Py (φ ) R(φ ) (1 − φ i )2 f ′(φ ) q p R(φ i )
Z = 1− ; α (Z ) = ; f (φ ) = ; B(φ ) = ; ∆(φ ) = ; Qp =
2
ru A(ru ) ∆ρω ru2 2 R (φ )
i (1 − φ )
2 B (φ ) ∆ρω ru A(ru )(1 − φ i )2
2

Step Size: ∆φu = 0.01


0th Step: φu = φg - ∆φu

φ uφi 
∆φu Step: φu = φu + ∆φu ; Q p = (1 − Z i ) φ − φ 
 u i

Consolidation Zone: φ(0) = φu; Z = 0; ∆Z = 0.01 Zi

Runge-Kutta Solve dφ = − 1  φ − α (Z i )(1 − Z i ) φ φ u − φ  until φ (Z ) = φ , dφ = 0 or Z = Z


 
∆ (φ )  B(φ ) α (Z )(1 − Z ) φ u − φ i 
i g i
Algorithm: dZ dZ

dφ ∆Z
If φ (Z ) < φ g , > 0 or Z > Z i , ∆Z =
dZ 2

∆φ u
If φ (Z i ) > φ g , ∆φ u =
2

Zc = Z; φ(0<Z<Zc)

Sedimentation Zone: Z = Zi; φ(Z) = φi; ∆φs = 0.01 φi

φ (Z ) = φ (Z ) − ∆φ s

φ φ −φ
Solve α (Z )(1 − Z ) = α (Z i )(1 − Z i )B(φ ) i u for Z until Z = Z c or φ = 0
φ φu − φi

∆φ s
If Z < Z c or φ < 0 , ∆φ s =
2

φ(Zc<Z<Zi)

Proceed to next step of ∆φu

Figure 6.2.2: Numerical algorithm for 1-D continuous centrifugation, inlet suspension un-
networked

381
Case 2: Inlet Suspension Networked (φi > φg)

Sedimentation Zone (Zc ≤ Z ≤ Zi)

The sedimentation zone for the inlet suspension networked case is able to
transmit pressure since a particle network exists, but Pp(Z) < f(φi). Therefore the
concentration remains constant at φi and dφ/dZ = 0. The pressure gradient in the
sedimentation zone is found from equation 6.2.30:

dPp Qp  φ 
= −φi + 1 − i ; Z c ≤ Z ≤ Z i …(6.2.37)
dZ α (Z )(1 − Z )  φu 

Integrating equation 6.2.37 from Pp(Zi) = 0 to Pp(Zc) = f(φi) gives an


expression for Qp:

f (φi ) − φi (Z i − Z c )
Qp = …(6.2.38)
 φ Z c 1
1 − i  ∫
 φ  α (Z )(1 − Z ) dZ
 u  Zi

Thus, the solids throughput for the φi > φg case is dependent upon the height of
the consolidating cake as well as the underflow concentration. Qp asymptotes to
infinity for the trivial case as φu approaches φi (that is, Zc = 0), and is equal to zero
when Zc is a maximum:

f (φ i )
Z c ,max = Z i − …(6.2.39)
φi

For the simpler case when the centrifuge has constant width (α(Z) = 1), the
throughput is given by:

f (φi ) − φi (Z i − Z c )
Qp = …(6.2.40)
 φ   1 − Zi 
1 − i  ln 
 φ  1 − Z 
 u  c

382
Consolidation Zone (0 < Z ≤ Zc)

At Zc, the network pressure exceeds the compressive yield stress and the
consolidation zone exists. φ(Zc) is continuous at φi. The volume fraction distribution
in the consolidation zone is given by the solution of dφ/dZ (equation 6.2.32), however
an iterative approach is required since Qp is dependent on Zc.

For the trivial case when Qp = 0, the volume fraction distribution is given by:

dφ φ
=− ; 0 ≤ Z ≤ Z c ,max …(6.2.41)
dZ f ′(φ )

The maximum possible underflow solids concentration, φu,max, is given by


numerically integrating equation 6.2.41 from φ(Zc,max) = φi to φ(0) = φu,max.

The algorithm for determining Qp, Zc and φ(0<Z<Zc) for a given φu is outlined
in Figure 6.2.3. Since Qp is a function of Zc, Zc* is introduced as an iteration variable.
The initial upper bound for Zc*, Zc,high, is Zc,max and the lower bound, Zc,low, is 0. An
interval halving method is used to iterate upon Zc*. For each estimate of Zc*, Qp is
given by equation 6.2.40, allowing equation 6.2.32 to be solved using a 4th-5th order
Runge-Kutta numerical technique from φ(0) = φu in steps of ∆Z. If φ < φi, dφ/dZ > 0
or Z > Zc*, ∆Z is halved. The steps of ∆Z are repeated until φ(Z) = φi, dφ/dZ = 0 or Z =
Zc*. If dφ/dZ = 0 or φ(Zc*) > φi, Zc* is too small and becomes the lower bound for the
next iteration. If Z < Zc*, Zc* is too large and becomes the upper bound for the next
iteration. Zc* is iterated until φ(Zc*) = φi, such that Zc = Zc*. Steps of ∆φu are repeated
from φu = φi + ∆φu until φu = φu,max.

383
1-D Solid-Bowl Continuous Centrifuge Model, φi > φg
Iterative 4th – 5th order Runge-Kutta Algorithm

Material Characteristics: Py(φ), R(φ), ∆ρ


Operating Conditions: ru, ri , φi, A(r), ω

Scalings:
r2 A(r ) 2 Py (φ ) R(φ ) (1 − φ i )2 f ′(φ ) q p R (φ i )
Z =1− ; α (Z ) = ; f (φ ) = ; B (φ ) = ; ∆(φ ) = ; Qp =
2
ru A(ru ) ∆ρω ru2 2 R(φ i ) (1 − φ )2 B (φ ) ∆ρω ru A(ru )(1 − φ i )2
2

f (φ i ) dφ φ
Z c ,max = Z i − ; Solve =− from φ (Z c ,max ) = φ i until φ (0 ) = φ u ,max
φi dZ f ′(φ )

Step Size: ∆φu = 0.01(φu,max – φi); 0th Step: φu = φi, Zc = 0

∆φu Step: φu = φ u + ∆φ u ; Z c ,high = Z c ,max ; Z c ,low = Z c

Z c* =
Z c , high + Z c, low
; Q *p =
(
f (φ i ) − φ i Z i − Z c* )
*
Zc* Iteration: 2  φi Z c 1
1 −  ∫ dZ
 φu  Z i α (Z )(1 − Z )

dφ 1  φ Q *p  φ 
Solve =−  − 1 − 
Runge-Kutta dZ ∆ (φ )  B(φ ) α (Z )(1 − Z )  φ u 
 
Algorithm: dφ
from φ (0 ) = φ u until φ (Z ) = φ i , = 0 or Z = Z c*
dZ

dφ ∆Z
If φ (Z ) < φ i , > 0 or Z > Z *c , ∆Z =
dZ 2

( )
If φ Z *c > φ i or

dZ
= 0 , Z c ,low = Z *c ; If Z < Z *c , Z c ,high = Z *c

( )
φ Z c* = φ i ; Z c = Z c* ; Q p = Q *p

If φu < φu,max, proceed to next step of ∆φu

Figure 6.2.3: Numerical algorithm for 1-D continuous centrifugation, inlet suspension networked

384
6.2.2 Results

The average ferric water treatment sludge power-law material characteristics


presented in Section 4.1.1 were used as inputs to the models for both the un-
networked and networked cases in order to illustrate the one-dimensional continuous
centrifuge model. While these characteristics were based only upon filtration data and
were not realistic below φ = 0.004 v/v, the simple functional forms were easy to use
and are traditionally used to illustrate such models.

The continuous centrifuge models were used to generate the results for a range
of inlet concentrations (from 0.001 to 0.02 v/v) and rotational rates (from 1000 to
3000 rpm) with ru = 0.5 m, ri = 0.25 m and W(r) = 0.5 m (constant width). These
dimensions represent a large tubular centrifuge rotating at medium speeds. The
volume fraction distribution results for the range of possible underflow concentrations
for the φi < φg case (φi = 0.001, 0.002, 0.003 and 0.0039 v/v) at ω = 2000 rpm are
presented in Figure 6.2.4. The results for each value of φi illustrate, in φu increments
of 0.01 v/v, the increase of the consolidation zone as φu increases, up until φ(0) =
φu,max and Zc = Zi. The results show very little difference between the different inlet
concentrations ((a) to (d)), indicating that the consolidation zone is only slightly
dependent on the inlet concentration.

An enlarged view of the sediment volume fraction distribution is shown in


Figure 6.2.5, clearly illustrating the reduction in φ as Z decreases from Zi to Zc. The
sedimentation behaviour is predominantly determined by the inlet concentration, and
is virtually unaffected by the underflow concentration, except at very low cake
heights. As illustrated later, this is because the throughput does not change much at
higher underflow solids.

385
(a) φ i = 0.001 v/v; ω = 2000 rpm (b) φ i = 0.002 v/v; ω = 2000 rpm
0.2

Zi = 0.75

Zi = 0.75
Solids Volume Fraction, φ (Z ) φ u,max = 0.1782 v/v φ u,max = 0.1770 v/v

0.15

0.1

0.05

φg φg
0
(c) φ i = 0.003 v/v; ω = 2000 rpm (d) φ i = 0.0039 v/v; ω = 2000 rpm
0.2

Zi = 0.75
Zi = 0.75
φ u,max = 0.1758 v/v φ u,max = 0.1746 v/v
Solids Volume Fraction, φ (Z )

0.15

0.1

0.05

φg φg
0
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8

Scaled Radius, Z Scaled Radius, Z

Figure 6.2.4: One-dimensional continuous centrifuge results of volume fraction distribution for
ferric water treatment sludge at different inlet concentrations (φi < φg, ru = 0.5 m, ri = 0.25 m, W(r)
= 0.5 m and ω = 2000 rpm)

386
(a) φ i = 0.001 v/v; ω = 2000 rpm (b) φ i = 0.002 v/v; ω = 2000 rpm
0.004

Zi = 0.75

Zi = 0.75
Solids Volume Fraction, φ (Z )

0.003

0.002 φi

0.001 φi

0
(c) φ i = 0.003 v/v; ω = 2000 rpm (d) φ i = 0.0039 v/v; ω = 2000 rpm
0.004
φi
Solids Volume Fraction, φ (Z )

0.003 φi

0.002

0.001

Zi = 0.75
Zi = 0.75

0
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8

Scaled Radius, Z Scaled Radius, Z

Figure 6.2.5: One-dimensional continuous centrifuge results of sediment volume fraction


distribution for ferric water treatment sludge at different inlet concentrations (φi < φg, ru = 0.5 m,
ri = 0.25 m, W(r) = 0.5 m and ω = 2000 rpm)

The volume fraction distribution results for the range of possible underflow
concentrations for the φi > φg case (φi = 0.005, 0.01, 0.015 and 0.02 v/v) at ω = 2000
rpm are presented in Figure 6.2.6. The results illustrate the build-up of the cake as φu
is increased. As Zc approaches Zc,max, increases in φu produce very small increases in
Zc. Increasing φi affects the distribution at the top of the consolidation zone.

387
(a) φ i = 0.005 v/v; ω = 2000 rpm (b) φ i = 0.01 v/v; ω = 2000 rpm
0.2
φ u,max = 0.1795 v/v φ u,max = 0.1796 v/v

Zi = 0.75

Zi = 0.75
Solids Volume Fraction, φ (Z )

0.15

0.1

0.05

φi
φi
0
(c) φ i = 0.015 v/v; ω = 2000 rpm (d) φ i = 0.02 v/v; ω = 2000 rpm
0.2
φ u,max = 0.1796 v/v φ u,max = 0.1798 v/v

Zi = 0.75
Zi = 0.75
Solids Volume Fraction, φ (Z )

0.15

0.1

0.05

φi
φi
0
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8

Scaled Radius, Z Scaled Radius, Z

Figure 6.2.6: One-dimensional continuous centrifuge results of volume fraction distribution for
ferric water treatment sludge at different inlet concentrations (φi > φg, ru = 0.5 m, ri = 0.25 m, W(r)
= 0.5 m and ω = 2000 rpm)

Figure 6.2.7 gives the scaled cake radii as functions of φu for the different inlet
concentrations at three rotational rates (ω = 1000, 2000 and 3000 rpm). At a fixed
value of ω, Zc decreases slightly with increasing φi. For φi < φg and at low ω, Zc is
effectively independent of φi, and there is only a small dependence for φi > φg. For a
given Zc, varying ω has a large effect on φu, since higher rotational speeds enable
greater compression.

388
0.8
Z i = 0.075

Increasing φ i

Scaled Cake Radius, Zc


0.6

ω = 1000 rpm

ω = 2000 rpm

ω = 3000 rpm
0.4

0.2

0
φg
0 0.1 0.2 0.3

Underflow Volume Fraction, φ u (v/v)

Figure 6.2.7: One-dimensional continuous centrifuge results of scaled cake radius versus
underflow volume fraction for ferric water treatment sludge at different φi and ω (ru = 0.5 m, ri =
0.25 m and W(r) = 0.5 m)

The scaled throughput results for the range of inlet concentrations and
rotational rates are presented in Figure 6.2.8. Qp is high at low φu and low at high φu –
the behaviour changes from low solids / high flux where the operation is limited by
the permeability of the cake to high solids / low flux where the operation is limited by
the compressibility of the cake. There is step change in behaviour between the two
modes of operation, illustrating the effect of having a sedimentation zone able to
transmit a network pressure. For φi < φg, Qp is constant except at low φu, while for φi
> φg, Qp plateaus for intermediate φu and approaches zero as φu approaches φu,max.
Therefore, the underflow concentration can be extremely sensitive to changes in
throughput. The results also show that Qp is independent of ω for φi < φg, and
independent of ω at low φu for φi > φg. Qp increases for increasing inlet
concentrations.

389
1.E-01

Increasing φ i

Scaled Solids Throughput, Qp


1.E-02

1.E-03

ω = 1000 rpm

ω = 2000 rpm

ω = 3000 rpm
1.E-04

1.E-05 φ
g
0 0.1 0.2 0.3

Underflow Volume Fraction, φ u (v/v)

Figure 6.2.8: One-dimensional continuous centrifuge results of scaled throughput versus


underflow volume fraction for ferric water treatment sludge at different φi and ω (ru = 0.5 m, ri =
0.25 m and W(r) = 0.5 m)

Figure 6.2.9 and Figure 6.2.10 show the solids and suspension throughputs as
functions of φu for the range of inlet concentrations and rotational speeds. Removing
the scaling from Qp shows that qp increases with increasing ω and increasing φi. qi
also increases with increasing ω, but φi has much less effect. For φi < φg, qi is almost
constant with changing φi, except at low φu. For φi < φg, qi converges at low φu for
different φi, and is then almost independent of φi. Therefore, in processes where the
centrate is valuable (such that the suspension throughput is the operating variable to
be maximised), the optimum φi is not necessarily the maximum φi. For φi > φg, qi
converges at low φu for different φi, and then decreases with increasing φi.

390
1.E-07
Increasing φ i

Solids Throughput, qp (m /s)


3 1.E-08

1.E-09

ω = 1000 rpm

ω = 2000 rpm

ω = 3000 rpm
1.E-10

1.E-11
φg
0 0.1 0.2 0.3

Underflow Volume Fraction, φ u (v/v)


Figure 6.2.9: One-dimensional continuous centrifuge results of solids throughput versus
underflow volume fraction for ferric water treatment sludge at different φi and ω (ru = 0.5 m, ri =
0.25 m and W(r) = 0.5 m)

1.E-05
Suspension Throughput, qi (m /s)

Increasing φ i
3

Increasing φ i

1.E-06

1.E-07
ω = 1000 rpm

ω = 2000 rpm

ω = 3000 rpm

1.E-08
φg
0 0.1 0.2 0.3

Underflow Volume Fraction, φ u (v/v)


Figure 6.2.10: One-dimensional continuous centrifuge results of suspension throughput versus
underflow volume fraction for ferric water treatment sludge at different φi and ω (ru = 0.5 m, ri =
0.25 m and W(r) = 0.5 m)

391
The variation of behaviour with different pond depths has also been
investigated by varying ri. The chosen values correspond to Zi = 0.64, 0.75 and 0.99.
Figure 6.2.11 shows the variation of scaled cake radius with varying Zi. Higher
underflow solids are achievable for greater pond depths. Below Zi, there is no
difference in Zc for the φi < φg case, since the consolidation zone is independent of the
sedimentation zone. There is a slight dependence on Zi for the φi > φg case.

1
Z i = 0.99
Scaled Cake Radius, Zc

0.8

Z i = 0.75
0.6
Z i = 0.64
φ i = 0.02 v/v
0.4

0.2
φ i = 0.001 v/v

0 φ
g
0 0.05 0.1 0.15 0.2 0.25

Underflow Volume Fraction, φ u (v/v)

Figure 6.2.11: One-dimensional continuous centrifuge results of scaled cake radius versus
underflow volume fraction for ferric water treatment sludge at different φi and ri (ru = 0.5 m, ω =
2000 rpm and W(r) = 0.5 m)

Figure 6.2.12 shows the scaled solids throughput results for the variation of
fluid height. Even though higher underflow concentrations are achievable, the
throughput decreases as the fluid height is increased, since the acceleration is
dependent on r. The discontinuity in behaviour between the networked and un-
networked cases becomes more pronounced as Zi increases. qp follows the same
trends as Qp (since equation 6.2.28 is independent of ri). The suspension throughput
results are presented in Figure 6.2.13, which show the same trends.

Combined, the results for varying ω and Zi show that the optimum conditions
for qp and φu are when ω and φi are maximised and Zi is significantly less than unity.

392
1.E-01

Increasing φ i

Scaled Solids Throughput, Qp


1.E-02

1.E-03

1.E-04

1.E-05

Zi = 0.64

Zi = 0.75

Zi = 0.99
1.E-06 φ
g
0 0.05 0.1 0.15 0.2 0.25

Underflow Volume Fraction, φ u (v/v)


Figure 6.2.12: One-dimensional continuous centrifuge results of scaled solids throughput versus
underflow volume fraction for ferric water treatment sludge at different φi and ri (ru = 0.5 m, ω =
2000 rpm and W(r) = 0.5 m)

1.E-05
Suspension Throughput, qi (m /s)
3

Increasing φ i

Increasing φ i
1.E-06
Z i = 0.64
Z i = 0.75

1.E-07
Zi = 0.99

1.E-08 φ
g
0 0.05 0.1 0.15 0.2 0.25

Underflow Volume Fraction, φ u (v/v)


Figure 6.2.13: One-dimensional continuous centrifuge results of suspension throughput versus
underflow volume fraction for ferric water treatment sludge at different φi and ri (ru = 0.5 m, ω =
2000 rpm and W(r) = 0.5 m)

393
6.3 Continuous Decanting Centrifuge Modelling

6.3.1 Background

Continuous decanting (or thickening) centrifuges are used extensively


throughout the water and wastewater treatment industries to perform solid-liquid
separation. Continuous centrifuges are useful since they generally achieve high cake
concentrations whilst maintaining high throughputs and have a small footprint.
However, failures occur, and the reasons are often unknown or misunderstood.
Successful modelling of performance allows optimisation of operation, provides a
useful tool for design purposes, and gives insight into the mechanisms of failure.

Figure 6.3.1: Cross-section of cylindrical-conical helical-conveyor centrifuge. (Bird Machine Co.)

A cross-sectional view of the centrifuge and its auxiliary parts such as the
drive motor and discharge ports is shown in Figure 6.3.1. The centrifuge itself
comprises a solid cylindrical bowl rotating at high speeds (between 5000 and 12000
rpm). The density difference between the solid and liquid phases under the
centrifugal acceleration causes the solids to collect against the bowl of the centrifuge.
Within the bowl is a helical screw or scroll conveyor that rotates at a slightly different
speed to the bowl of the centrifuge (∆ω), pushing the consolidated material in the
axial direction. The liquid discharge or overflow end of the centrifuge has adjustable
liquid ports (or a “weir”) of a given height that dictate the fluid (or “pond”) height

394
within the centrifuge. The solids discharge or underflow end of the centrifuge has a
cone section, called the “beach”, where the solids are scrolled clear of the pond. The
beach allows drainage of the solids and prevents the flow of liquid out the underflow.

Continuous decanting centrifuges can be either horizontal or vertical – there is


little difference in practice since the centrifugal forces are much greater than the force
due to gravity. The centrifuges can also be either co-current or counter-current,
depending on the inlet position. In the co-current design, the feed suspension is
introduced at the overflow end and the liquid and solids move in the same direction
axially. In the counter-current design, the feed is introduced towards the underflow
end. The liquid flows in the opposite direction to the movement of the scroll while
solids that have sedimented from the liquid are pushed back up the centrifuge by the
scroll. In both cases, the feed suspension is pre-accelerated to the bowl speed to
prevent floc degradation.

The performance of decanting centrifuges is given by the solids recovery, cake


solids content, residence time (and solids conveyance), throughput, torque, power
consumption and product destruction (Reif, et al., 1990), which are influenced to
varying degrees by the material behaviour, centrifuge design and centrifuge operation.
The important design parameters include the bowl dimensions of radius, cylinder
length, cone length and cone angle, and the scroll dimensions of scroll angle, blade
angle and pitch. The operating parameters include the throughput, feed concentration,
bowl angular velocity, differential angular velocity and pond height. The influences
of the various material, operating and design parameters on performance are highly
interconnected and are quantified through detailed modelling. Qualitative
descriptions based on performance are useful (Records, 1974) but do not give detailed
information. For example, the required clarification length (and therefore cylinder
length and scroll pitch) will depend on the settling characteristics of the feed, scroll
dimensions, throughput, bowl radius, bowl speed and pond height.

This work considered the modelling of the counter-current design that is the
prevailing device used by the water and wastewater industries, but the approach is
applicable to the co-current design. The one-dimensional models of batch and
continuous centrifugation presented in Sections 6.1 and 6.2 represented simplistic
interpretations of industrial operation. Reif, et al., 1990, presented a model of

395
decanting centrifuges that identified three important aspects: clarification, dewatering
and conveyance. The term ‘dewatering’ was used by Reif, et al., 1990, to describe the
desaturation or drainage of the cake on the beach, whereas ‘dewatering’ was used
within this work to describe the sedimentation and consolidation of suspensions. The
term ‘desaturation’ was used here to avoid confusion. Sigma theory (Ambler, 1952)
was used to describe clarification, which ignores cake consolidation; Darcy’s law was
used to describe desaturation, which does not account for capillary-assisted
compaction; and the conveyance of the solids was described using a simplistic
frictional force balance (Reif and Stahl, 1989), which is not applicable to yield-stress
materials such as water and wastewater sludges. A similar model that did not account
for desaturation was presented by Records, 1974, and expanded to include cake
consolidation by Corner-Walker, 2000. In this work, the hindered settling,
consolidation and shear yield-stress behaviour of sludges, and their effect on solid-
liquid separation and scrolling behaviour, were considered using a comprehensive
approach, since the three aspects of decanting centrifuges are inter-related.

396
6.3.2 Model Description

The approach to model the steady-state behaviour of a decanting centrifuge


involves dividing the centrifuge into three sections:

- The overflow or liquid discharge end, where the majority of the sedimentation
occurs. It is assumed that the bowl radius is constant in the overflow section;

- The inlet, where conditions may be turbulent and mixing occurs; and

- The underflow, or solids discharge, where the solids are scrolled from the
pond and up the beach (the bowl radius is variable).

The modelling presented here considers the flow of the fluid, the solids
movement due to the scroll, and the sedimentation-consolidation and desaturation of
the solids due to centrifugal acceleration. Assuming that the scroll wall is
impermeable, the liquid flows down the channel formed by the blades of the scroll,
opposite to the direction of the scrolling. An orthogonal change of variables from
{ } { }
cylindrical coordinates rˆ , θˆ , zˆ to helical coordinates rˆ , sˆ, tˆ is made, as illustrated in
Figure 6.3.2.

r̂ r̂

θ̂ ẑ ŝ

(a) (b)

Figure 6.3.2: Schematic of change of variables from (a) cylindrical coordinates to (b) helical
coordinates

ŝ is the vector in the direction of the scroll blade and t̂ is the vector normal to
ŝ . r̂ does not change in this transformation. The modelling of the centrifuge is

397
visualised by unwrapping the channel formed by the scroll, as illustrated in Figure
6.3.3. If the scroll has holes in it, as is sometimes the case, the residence time of the
liquid will be reduced since the flow is axial rather than helical – the assumption is
made that there is no variation in the t̂ direction. Likewise, there is a gap between the
scroll blade and the bowl wall. However, for yield-stress materials at steady-state, a
thin layer of consolidated material forms against the bowl wall, and the gap has no
effect on centrifuge performance other than to reduce the effective bowl radius.

q i , φi

r̂ (h)
q u , φu
qo (a) (b) (d) (f)
(c) (e) (g)

overflow inlet underflow

Figure 6.3.3: Schematic of continuous decanting centrifuge, unwrapped along scroll blade

The three sections to be modelled are the overflow (a, b and c), the inlet (d and
e) and the underflow (f, g and h). The overflow section has three zones of behaviour -
the clear-liquor zone (a) and the sedimentation zone (b) flow opposite to the direction
of the scroll blade, while the scroll pushes the cake in the consolidation zone (c) with
a velocity relative to s, vc. It is assumed that all the solids have sedimented from the
fluid and there are no solids in the overflow, since entrained solids represents failure
of operation and would require consideration of the fluid mechanics of the weir. The
inlet section consists of the mixing zone (d), where the inlet feed is mixed with fluid
flowing back from the underflow, above the consolidating cake (e). The underflow
section has a clear-liquor zone (f) but no sedimentation zone since the only solids in
this section are already in the consolidated cake (g). The cake is pushed up the beach
by the scroll. When the cake is lifted from the pond, drainage begins to occur and the
cake consolidates to the capillary pressure of the material. After that, the liquid front
recedes into the cake and the cake begins to desaturate (h).

398
Assuming incompressibility of the liquid and solid phases, overall volumetric
balances and volumetric balances for both phases over the entire centrifuge relate the
overflow, underflow and inlet fluxes (qo, qu and qi respectively) and the underflow
and inlet concentrations (φu and φi respectively). Assuming that the overflow
concentration, φo, is zero, equations 6.2.2 to 6.2.4 give the global conservation
equations.

The incomplete work presented here details the derivation of the helical
coordinate system in the overflow section and its use to formulate the fluid dynamics
of the liquid phase and the mechanics of the transportation of the solid phase. In order
to verify assumptions about the flowing phase, the Navier-Stokes equations for
viscous gravity flow are solved, showing that the fluid flow is in the helical direction
only (there is no radial component). The forces acting on the cake are resolved to
determine the scrolling velocity of the solid phase. The equations for the conservation
of momentum and mass are derived for the different zones of the overflow section,
and a numerical scheme is outlined. The approach required for the formulation of the
inlet and underflow sections is discussed.

399
6.3.3 Overflow Section

A schematic of the overflow section is shown in Figure 6.3.4. The section


consists of the clear-liquor zone from rf to rs(s), the sedimentation zone from rs(s) to
rc(s), and the consolidation zone from rc(s) to rb. s = 0 is chosen as the point at which
the sedimentation and consolidation zones disappear. This is at the same point since
the two zones cannot exist without each other. It is assumed that the weir is at
negative s, such that the overflow concentration is zero and the behaviour of the solids
is independent of the flow of liquid over the weir. The concentration of the sediment
at the inlet (at si) is φsi. φsi is not equal to φi since the sediment consists of the inlet
suspension and the fluid returned from the underflow section.

ŝ rf rs(s) rc(s) rb

qo qo,F(s) qo,F(si)
qo,C(s) qo,C(si)

0 si

Figure 6.3.4: Schematic of overflow section of continuous decanting centrifuge

Modelling of the overflow section involves consideration of the velocity of the


clear-liquor and sedimentation zones (the flowing region) through the channel formed
by the scroll blades, the movement of the consolidation zone due to the differential
scroll velocity and the solid-liquid separation due to centrifugal acceleration. Initially,
the transformation from cylindrical to helical coordinates is performed. The flowing
region is modelled using the Navier-Stokes equations, with assumptions of a slip
boundary at the fluid-air interface and a stick boundary at the sedimentation-
consolidation interface. The solid-liquid separation is modelled by formulating the
phenomenological equations into helical coordinates.

400
Helical Coordinates

This section outlines the derivation of the helical coordinate system. Figure
6.3.5(a) shows a schematic of the edge of the scroll blade as it progresses along the
length of the centrifuge. Unwrapping the cylinder a distance of one revolution (see
Figure 6.3.5(b)) gives a rectangle of length 2πrb and width P.

P
r̂ ŝ

2πrb
θ̂
ẑ β


(a) (b)

Figure 6.3.5: Schematic of scroll edge

Based upon this geometric consideration, the angle between the scroll and the
axis of rotation of the centrifuge, β, is given by:

 P 
β = arctan  …(6.3.1)

 br

The scroll angle is different to the blade angle, γ, which is the angle formed by
the blade in relation to the radial direction.

{ }
In the orthogonal transformation from cylindrical coordinates, rˆ , θˆ , zˆ , to

{ }
helical coordinates, rˆ , sˆ, tˆ , the radial component is unchanged. The transformation
of the axial and angular components is given in vector notation:

401
 P
ˆs  1  r 2π  θ 
ˆ
ˆt  =  P   …(6.3.2)
  2
2  P  − r  ˆz 
r +   2π 
 2π 

The differential operators for this transformation have been derived and the
results are presented here. A displacement vector dr is considered:

dr = drˆr + dsˆs + dtˆt = drˆr + rdθθ


ˆ + dzˆz …(6.3.4)

Equating dr in both coordinate systems gives ds and dt in terms of dθ and dz:

 2 P
ds  1  r 2π  dθ 
 dt  =  rP   …(6.3.5)
   P
2
 − r   dz 
r2 +    2π 
 2π 

Similarly, the vector operators are derived in terms of a general function, f.

f = f r ˆr + fθ θ
ˆ + f ˆz = f ˆr + f ˆs + f ˆt
z r s t …(6.3.6)

Thus

∂f ∂f ˆ ∂f
θ ∂f ∂f ∂f ˆ
∇f = ˆr + + ˆz = ˆr + ˆs + t
∂r θ , z ∂θ r , z r ∂z r ,θ ∂r s ,t ∂s r ,t ∂t r ,s

…(6.3.7)

∂f r f r 1 ∂fθ ∂f z ∂f r f r ∂f s ∂f t
∇•f = + + + = + + + …(6.3.8)
∂r r r ∂θ ∂z ∂r r ∂s ∂t

2 1 ∂  ∂f  1 ∂  1 ∂f  ∂f ∂ 2 f 1 ∂f ∂ 2 f ∂ 2 f
∇ f= r  +   + = + + +
r ∂r  ∂r  r ∂θ  r ∂φ  ∂z ∂r 2 r ∂r ∂s 2 ∂t 2
…(6.3.9)

These equations show that s behaves as an arc length, rθ, and t behaves like
the axial length, z.

402
Viscous Gravity Flow

Certain assumptions are made about the overflow section in order to model the
fluid transport. The fluid phase is assumed to flow due to viscous drainage down the
channel formed by the blades of the scroll. The flowing region is all material below
φg (that is, without a yield stress) and therefore consists of the free-liquor and
sedimentation zones. In order for this to be the case, the sedimentation-consolidation
interface is at low shear (that is, a ‘stick’ boundary condition), the top of the fluid is a
stress free surface, and there is slip at the walls of the channel.

In order to justify these assumptions, the steady state, two-dimensional


laminar flow of a Newtonian liquid across a stick surface is formulated. The Navier-
Stokes equations are scaled to the viscous terms and an order of magnitude argument
made to eliminate the inertial terms. The resulting governing equations are solved
under the appropriate boundary conditions to show that the flowing region is
governed by viscous drainage (that is, low Reynold’s numbers). An illustration of the
system is shown in Figure 6.3.6.

y
h(x)

Q
x

Figure 6.3.6: Viscous gravity flow across a stick surface

Conservation Equations

The steady-state Navier-Stokes momentum equations in the x and y directions


are:

403
∂v x ∂v x η  ∂ 2 v x ∂ 2 v x  1 ∂p
−
vx + vy = + …(6.3.10)
∂x ∂y ρ  ∂x 2 ∂y 2  ρ ∂x

η  ∂ vy ∂ vy  1 ∂p
2 2
∂v y ∂v y
vx + vy =  + − −g …(6.3.11)
∂x ∂y ρ  ∂x 2 ∂y 2  ρ ∂y
 

Assuming incompressibility, the local conservation of mass is:

∂v x ∂v y
+ =0 …(6.3.12)
∂x ∂y

The global conservation of mass is:

h( x )
− Q = ∫ v x ( x , y )dy …(6.3.13)
0

Boundary Conditions

The boundary conditions at the bottom surface are that the liquid velocity is
zero in the x-direction (a stick condition) and the surface is impenetrable:

v x ( x ,0 ) = 0 …(6.3.14)

v y ( x,0) = 0 …(6.3.15)

From equations 6.3.12 and 6.3.13, the kinematic condition at the top surface
states that flow at the top surface is in the direction of the surface:

v y ( x, h )
h' = = Tanθ …(6.3.16)
v x ( x, h )

where θ is the angle of the surface relative the horizontal plane. The top surface is
also a stress-free surface, such that the normal stress tensor is zero:

ˆ = σ nn n
σ•n ˆnˆ + σ lnˆln
ˆ =0 …(6.3.17)

404
The stress terms in the x-y direction are equated to the stress terms in the plane
of the surface using θ:

σ xx xˆ xˆ + σ xy xˆ yˆ + σ xy yˆ xˆ + σ yy yˆ yˆ
[ ]
= Sin 2θσ xx − 2 SinθCosθσ xy + Cos 2θσ yy nˆ nˆ
+ [Cos θσ − 2 SinθCosθσ + Sin θσ ] ˆlˆl
2
xx xy
2
yy

+ [− CosθSinθσ + (Cos θ − Sin θ )σ + CosθSinθσ ] nˆ ˆl


xx
2 2
xy yy

+ [− CosθSinθσ + (Cos θ − Sin θ )σ + CosθSinθσ ] ˆlnˆ


xx
2 2
xy yy

…(6.3.18)

Equating the terms in equations 6.3.17 and 6.3.18 gives:

Sin 2θσ xx − 2 SinθCosθσ xy + Cos 2θσ yy = 0 …(6.3.19)

and

( )
− CosθSinθσ xx + Cos 2θ − Sin 2θ σ xy + CosθSinθσ yy = 0 …(6.3.20)

From the Navier-Stokes equations, the stresses are given in terms of the
pressure and velocity gradients:

∂v x
σ xx = − p + 2η
∂x
∂v y
σ yy = − p + 2η …(6.3.21)
∂y
 ∂v ∂v y 
σ xy = η  x + 
 ∂y ∂x 

Substituting equations 6.3.16 and 6.3.21 into equations 6.3.19 and 6.3.20
gives:

 h′2 ∂v x 1 ∂v y  2h′  ∂v x ∂v y 
− p + 2η  + − η + =0
 1 + h′2 ∂x 1 + h′2 ∂y  1 + h′2  ∂y ∂x 

 
…(6.3.22)

and

405

2 h′  ∂v x
η − +
( )
∂v y  1 − h′2  ∂v x ∂v y 
η + =0 …(6.3.23)
2  ∂x  1 + h′2  ∂y ∂x 
1 + h′  ∂y

Subtracting equation 6.3.22 from equation 6.3.23 and invoking the


incompressibility equation (6.3.12) gives the stress-free boundary equations for the
top surface:

 1 + h′2  ∂v y
p( x, h ) = 2η   …(6.3.24)
 1 − h′ 2  ∂y
  h

∂v x ∂v 4h' ∂v y
+ y =− …(6.3.25)
∂y h ∂x h
1 - h'2 ∂y h

Dimensionless Equations

The following dimensionless variables are introduced to simplify the problem:

x
X =
Xs
y
Y=
hs
h( x )
H (X ) =
hs
…(6.3.26)
v ( x, y )
u( X , Y ) = x
Vs
X s v y ( x, y )
v( X , Y ) =
hsVs
p ( x, y )
P( X , Y ) =
ρghs

The scalings Xs, hs and Vs are to be determined. Substituting these into


equations 6.3.10 and 6.3.11 gives the scaled momentum balances:

 V 2   ∂u ∂u   ηVs X s  ∂ 2 u  h  ∂ 2 u  ∂P
2
 s  u +  + s  −
 hs g   ∂X
v
∂ =   ∂Y 2  X s
  ρghs3  ∂X  ∂X
Y 2
  
…(6.3.27)

406
2  ∂ 2 v  h 
  Vs2   ∂v ∂v   ηVs X s  ∂ 2 v 
2
∂P h 
+ 1 = − s  + + s  
 
u
 ∂X v  −   ∂Y 2  X s
∂Y  ∂   ρghs3  ∂X  
2
 Xs   s 
h g Y 
…(6.3.28)

From equation 6.3.12, the scaled local conservation of mass is:

∂u ∂v
+ =0 …(6.3.29)
∂X ∂Y

The scaled overall conservation of mass is:

H (X )
= ∫ u ( x , y )dY
Q
− …(6.3.30)
h sV s 0

The scaled boundary conditions at Y = 0 are:

u ( X ,0 ) = 0 …(6.3.31)

and

v( X ,0 ) = 0 …(6.3.32)

The scaled stress-free boundary conditions at Y = H(X) are:

  h 2 
 1 +  s
 H ′ 2 

 ηVs X s  hs    X s   ∂v
2

P( X , H ) = 2 
3 
 
   …(6.3.33)
 ρghs  X s    hs  ∂Y
2
2  H
1 −
 X   H ′ 
  s 

 
2  
∂u h   ∂v 4H ′ ∂v 
= − s   +  …(6.3.34)
∂Y H  Xs   ∂X H  hs 
2 ∂Y H 
 1 −   H ′ 2

  s
X 

The scalings are chosen so that the viscous terms are order 1 quantities:

407
ηVs X s
=1 …(6.3.35)
ρghs3

Q
=1 …(6.3.36)
h sV s

A Reynold’s number is defined as:

Vs2 ρVs hs2


Re = = …(6.3.37)
hs g ηX s

Therefore, the characteristic length scale is hs2/Xs. Substituting equations


6.3.35 to 6.3.37 into equations 6.3.27 and 6.3.28 further simplifies the problem:

2
 ∂u ∂u  ∂ 2 u  hs  ∂ 2 u ∂P
Re u +v  = +  − …(6.3.38)
 ∂X ∂Y  ∂Y 2  X s  ∂X 2 ∂X

2 2
∂v  
2
∂P h  h  ∂ 2v
+ 1 =  s  ∂ v + s 
 ∂v
− Re u + v   …(6.3.39)
∂Y  Xs   ∂Y 2  X s  ∂X
2  ∂X ∂Y  
 

Equations 6.3.30, 6.3.33 and 6.3.34 become:

H (X )
∫ u ( X , Y )dY = −1 …(6.3.40)
0

  hs 
2 
 
2  1 +   H ′ 2
 ∂v
h   Xs 
P( X , H ) = 2 Re s    …(6.3.41)
 Xs   h 
2  ∂Y H
 1 −  s  H ′ 2 
 
  Xs  

 
2  
∂u  hs   ∂v 4H ′ ∂v 
= −   +  …(6.3.42)
∂Y H  X s   ∂X H  hs 
2 ∂Y H 
 1 −   H ′ 2 
  s
X 

408
Viscous Flow

Eliminating Vs from equations 6.3.35 and 6.3.36 gives an expression for hs/Xs:

hs ηQ
= << 1 …(6.3.43)
X s ρghs3

Estimates for these parameters of η = 10-3 kg/ms, Q = 0.04 m2/s, ρ = 1000


kg/m3, g = 10 m/s2 and hs = 0.2 m, gives a ratio of hs to Xs of 5 x 10-5. The Reynold’s
number for such values is 0.05, thus the flow is laminar. Given that the acceleration
increases significantly in a centrifuge, these estimates represent an upper limit to the
conditions in the channel. These values show that the velocity component in the x-
direction is much greater than the in the y-direction, such that the lubrication theory is
invoked.

Eliminating the inertial terms (hs/Xs and Re) from equations 6.3.38 and 6.3.39
significantly simplifies the governing equations:

∂ 2u ∂P
− =0 …(6.3.44)
∂Y 2 ∂X

∂P
+1=0 …(6.3.45)
∂Y

Likewise, the boundary conditions at Y = H(X) are simplified to:

P( X , H ) = 0 …(6.3.46)

∂u
=0 …(6.3.47)
∂Y H

Integrating equation 6.3.45 with respect to Y and evaluating at Y = H(X) using


equation 6.3.46 shows that the pressure is purely hydrostatic in viscous flow:

P( X , Y ) = H ( X ) − Y …(6.3.48)

Substituting equation 6.3.48 into equation 6.3.44, integrating twice with


respect to Y and using equations 6.3.31 and 6.3.47 gives u(X,Y) in terms of H:

409
 Y
u ( X , Y ) = − H ′Y  H −  …(6.3.49)
 2

Integrating u(X,Y) with respect to Y and using equation 6.3.40 allows


calculation of H(X):

H ( X ) = (1 + 12 X )1 4 …(6.3.50)

Therefore, u(X,Y) is:

 Y
u ( X ,Y ) = − Y (1 + 12 X )1 4 − 
3
…(6.3.51)
(1 + 12 X )3 4  2

v(X,Y) is found by differentiating equation 6.3.51 with respect to X,


substituting into equation 6.3.29, integrating with respect to Y, and evaluating the
constant of integration using equation 6.3.32:

 Y
v( X ,Y ) = − Y 2 (1 + 12 X )1 4 − 
9
…(6.3.52)
(1 + 12 X ) 54  2

X is very small since Xs is very large, therefore H(X) changes very little down
the length of the channel. The dimensionless stream function, Ψ(X,Y), is calculated
from the velocity functions:

∂Ψ ∂Ψ
u=− ;v = + …(6.3.53)
∂Y ∂X

Therefore, the stream function is:

3Y 2  Y
Ψ ( X ,Y ) =  (1 + 12 X )1 4 −  …(6.3.54)
2(1 + 12 X )3 4  3

The streamlines of the fluid flow are found when the stream function is
constant. A plot of the streamlines is presented in Figure 6.3.7, which illustrates that,
when hs/Xs is very small, the streamlines vary only slightly, and the fluid flow is
considered as flow in the x-direction only.

410
1
Ψ = 1.0
0.9 Ψ = 0.9
0.8 Ψ = 0.8
Ψ = 0.7
0.7 Ψ = 0.6
0.6 Ψ = 0.5

0.5 Ψ = 0.4
Y

Ψ = 0.3
0.4
Ψ = 0.2
0.3
Ψ = 0.1
0.2

0.1
Ψ =0
0
-0.01 -0.008 -0.006 -0.004 -0.002 0

Figure 6.3.7: Streamlines of gravity viscous flow across a flat surface

In order to justify the stick boundary condition at the bottom of the channel,
the shear stress is evaluated:

 ∂v x ∂v y   ηQ 
σ xy Y =0 ˆx = η  +  ≈ −3 2  for X ≈ 0 …(6.3.55)
 ∂y ∂x  Y =0  hs 

From the estimates to η, Q and hs given above, the stress at the bottom is
approximately 0.003 Pa. This is significantly less than the shear yield stress of a
particulate network at most concentrations above the gel points. The consequence for
decanting centrifuges is that any particles that are not networked will flow down the
channel while any solids that are networked do not flow with gravity, but are pushed
through the centrifuge by the scroll. The boundary can therefore be assumed to be at
the sedimentation – consolidation interface.

411
Scrolling Velocity of Consolidation Zone

The viscous gravity flow results show that material that exhibits a shear yield
stress will not flow down the channel of the scroll blades. Instead, the consolidating
material is moved through the decanter by the movement of the scroll (see Figure
6.3.8(a)). The forces acting upon the cake are investigated in order to determine the
cake velocity.

Scroll Solid
movement movement

Scroll Position after one


blade scroll revolution
Ffr,b Fn,sc
ds
Ffr,sc
+Ptrs δ
β z
P
t = P Cosβ
β

(a)

r r

r
Scroll blade

γ
Fn,b rc
P(s) Fn ,b P(s+ds)
Fn,sc
hc
Ffr,b rb
Bowl wall t ds
(b) (c)

Figure 6.3.8: Movement of the consolidation zone in a decanting centrifuge due to the scroll; (a)
Radial view (b) View down spiral direction; (c) Transverse view

412
Assuming that there are no concentration variations in the transverse direction,
the forces acting upon an element of cake (length PCosβ, width ds and height (rb –rc),
see Figure 6.3.8) include (Records, 1974, Reif and Stahl, 1989):

- The normal force exerted by the bowl, Fn,b, which is in the negative radial
direction;

- The normal force from the scroll, Fn,sc, which is normal to the helical
coordinate;

- The frictional drag on the bowl face, Ffr,b, which is in the direction of motion
of the cake; and

- The frictional drag on the scroll walls, Ffr,sc, which is in the negative helical
direction.

Also in the negative helical direction is a transverse pressure term, Ptr, due to
concentration changes across ds. The normal force exerted by the bowl is the reaction
to the centrifugal force pushing the material against the bowl. This force is equated to
the particle pressure, pp, within the consolidating material, which varies with the depth
of material. Likewise, the normal force due to the scroll blades also depends on the
solids pressure. pp is given by the constitutive equation for the compressive yield
stress of the material, Py(φ). The frictional forces in both the radial and axial
directions depend upon the shear yield stress of the material, σy(φ), and the cross-
sectional area through which the stress is applied. From a geometric argument, these
forces are combined to give the angle of cake movement relative to the normal of the
scroll, δ:

F fr ,sc + Ptr
Cos (δ − β ) = …(6.3.56)
F fr ,b

An element of cake of width ds is considered to move as a block of material of


length t and height (rb – rc). Performing a force balance in the helical direction over
this block and equating it to the pressure gradient through the transverse faces gives
the overall movement of the cake. The area multiplied by the shear yield stress gives
the drag force on the scroll walls (see Figure 6.3.8(a)):

413
2 rb rds
F fr , sc = ∫ σ y (φ (s, r ))dr …(6.3.57)
Cosγ rc rb

The pressure force through the transverse faces is (see Figure 6.3.8(c)):

[ ]
rb
Ptr = −t ∫ p p sˆ s + ds − p p sˆ s dr …(6.3.58)
rc

The total frictional drag on the bowl face is (see Figure 6.3.8(b)):

F fr ,b = tdsσ y (φ (s ,0 )) …(6.3.59)

Equations 6.3.57 to 6.3.59 are combined using equation 6.3.56 to give δ:

rb ∂p 2 rb
∫ rσ y (φ (s, r ))dr
p
∫ dr +
rc ∂s rbtCos γ rc
Cos (δ − β ) = …(6.3.60)
σ y (φ (s,0))

Thus the angle depends on both the concentration and depth of the solids and
therefore varies along the length of the centrifuge, but generally moves in a helical
direction up the cylinder (see Figure 6.3.9).

cake movement
P

fluid flow

Figure 6.3.9: Helical movement of solids through a continuous decanting centrifuge

414
At best scrolling efficiency (δ = β), the cake moves normal to the blade of the
scroll; at worst efficiency, the cake moves circumferentially. Since the flowing region
proceeds down the channel formed by the scroll, the decanting centrifuge is more
accurately described as cross-current rather than counter-current. The velocity of the
consolidation zone relative to the axial direction, va, is given by:

∆ωPCosβCosδ
va = …(6.3.61)
2πCos (δ − β )

The corresponding cake velocity relative to the helical coordinate, vc, is given
by:

∆ωP (Cosβ − SinβTan (δ − β ))


vc = …(6.3.62)
2πTanβ

The scroll torque and power consumption can be calculated from the frictional
and normal forces acting on the scroll (Corner-Walker, 2000), but were not
considered further here due to time constraints.

415
Conservation Equations

The conservation of mass and momentum equations for solid-liquid separation


are given in vector notation in Section 2.2. The liquid and solids velocities are
separated into the radial and helical components. All material in the flowing phase
moves down the spiral with a velocity given by the fluid dynamics of the flow, v(r,s),
which is assumed to be only in the direction of -s. This assumption is valid for
viscous flow. The consolidated material moves due to the scroll at a rate relative to s,
vc. The particle and fluid velocities in the radial direction are due only to the
centrifugal buoyancy and are given by up(r,s) and uf(r,s) respectively:

u p = v + u p (r , s )rˆ …(6.3.63)

u f = v + u f (r , s )rˆ …(6.3.64)

where v is given by:

− v(r , s )sˆ r f < r < rc 


 
v=  …(6.3.65)
 v sˆ rc < r < rb 
 c

This is illustrated in Figure 6.3.10.


rf
rs(s)
-v(r,s)
up(r,s) uf(r,s)
rc(s)
up(r,s) uf(r,s) vc
rb

Figure 6.3.10: Fluid and solid velocities in the radial and helical directions in the overflow section

416
Global Conservation Equations

The overall volumetric balance around the overflow end of the centrifuge
relates the relative flowrates of the cake and flowing zones, such that qo is the
difference between the flux in the flowing zone, qo,F, and the flux in the consolidating
zone, qo,C:

rc (s ) rb 
− qo = − qo, F (s ) + qo,C (s ) = 2π  ∫ rv(r , s )dr + ∫ rvc dr  …(6.3.66)
 rf rc (s ) 
 

qo,F consists of the fluid and the particles in the flowing zone:

rc (s ) rc (s ) 
qo, F (s ) = −2π  ∫ r (1 − φ )v (r , s )dr + ∫ rφv(r , s )dr  …(6.3.67)
 rf 
 rf 

Likewise, qo,C consists of the fluid and particles in the consolidation zone:

 rb rb 
qo,C (s ) = 2π  ∫ r (1 − φ )vc dr + ∫ rφvc dr  …(6.3.68)
rc (s ) rc (s ) 

The particle balance shows that, since φo = 0, the flux of particles in the
flowing region must balance the flux of particles in the consolidation zone:

rc (s ) rb
∫ rφv (r , s )dr + ∫ rφvc dr = 0 …(6.3.69)
rf rc (s )

The corresponding fluid balance is given by subtracting equation 6.3.69 from


equation 6.3.66:

rc (s ) rb 
− qo = −2π  ∫ r (1 − φ )v (r , s )dr − ∫ r (1 − φ )vc dr  …(6.3.70)
 rf rc (s ) 
 

The volumetric flow balances are used to link the solutions of the flowing and
consolidation regions.

417
Local Conservation Equations – Solid-Liquid Separation

Substituting equations 6.3.63 to 6.3.65 into the local conservation of mass


equation (2.2.8) gives the relationship between up and uf (identical to equation 6.1.4,
as derived for one-dimensional transient centrifugation). Converting equation 2.2.3
into helical coordinates with centrifugal acceleration and substituting for up, uf and uf
gives the consolidation equation for the radial direction (equation 6.1.5), which holds
in both the flowing and consolidating zones.

Equation 2.2.6 is equal to zero since the system is at steady state. Converting
to helical coordinates and substituting for up gives the relationship between the radial
and helical velocities:

 ∂φ 
− v(r , s ) ∂s r f < r < rc 

(
1 ∂ rφu p 
=
) 
 …(6.3.71)
r ∂r  ∂φ 
 vc rc < r < rb 
 ∂s 

Equations 6.1.5 and 6.3.71 are used to describe the sedimentation and
consolidation processes in the cake and flowing zones. The boundary conditions for
the consolidation zone are that the radial solids velocity is zero at the bowl (equation
6.1.9) and the top of the cake is at φg.

Local Conservation Equations – Fluid Dynamics

The assumption is made that the fluid dynamics in the flowing zone are given
by the Navier-Stoke’s equation:

η f ∇ 2 v − ∇p f + (ρ f + φ∆ρ )ω 2 rˆr = 0 …(6.3.72)

The s- and r-components of equation 6.3.72 are:

 1 ∂  ∂v  ∂ 2v  ∂p f
ηf  r  + + =0 …(6.3.73)
 r ∂r  ∂r  ∂s 2  ∂s

∂p f

∂r
[ ]
+ ρ f + φ∆ρ ω 2 r = 0 …(6.3.74)

418
The boundary conditions for the flowing zone are analogous to the boundary
conditions given for the stress free surface and stick boundary conditions in viscous
gravity flow. The fluid pressure is ambient at the surface:

pf =0 …(6.3.75)
rf

There are no shearing forces on the surface:

∂v
=0 …(6.3.76)
∂r r
f

The bottom of the flowing phase is stuck to the cake moving at vc:

v r = −vc …(6.3.77)
c

Dimensionless Equations

To simplify the equations, r, s, q, pp, pf, Py(φ), R(φ), up, v and vc are scaled to
give Z, S, Q, Pp, Pf, f(φ), B(φ), ψ, V and α respectively. The following scalings apply
if the radial scaling is chosen as equation 6.1.11, the flux is scaled with qo and the
coefficients of the governing equations (6.1.5, 6.3.71, 6.3.73 and 6.3.74) are set to
unity (Pp and f(φ) are given by equations 6.1.12 and 6.1.13):

8qoη f
S= s …(6.3.78)
π∆ρω 2 rb6

q
Q= …(6.3.79)
qo

2pf
Pf = …(6.3.80)
∆ρω 2 rb2

4qo2η f R (φ )
B (φ ) = …(6.3.81)
π 2
∆ρ 2ω 4 rb8 (1 − φ )

419
π 2 ∆ρω 2 rb6
ψ (Z , S ) = rφu p (r , s ) …(6.3.82)
4qo2η f

πrb2
V (Z , S ) = v(r , s ) …(6.3.83)
qo

πrb2
α= vc …(6.3.84)
qo

With these scalings, the volume fraction gradient in the Z-direction in the
sedimentation and consolidation zones is given by equation 6.1.17. The
incompressibility of the particles (equation 6.3.71) becomes:

 ∂φ 
∂ψ  α ∂S 0 < Z ≤ Zc 
=  …(6.3.85)
∂Z − V ∂φ Zc < Z ≤ Z f 
 ∂S 

The s- and r-components of the Navier-Stoke’s equation become:

∂  ∂V  ∂P f
(1 − Z ) + =0 …(6.3.86)
∂Z  ∂Z  ∂S

∂P f ρf 
= − + φ  …(6.3.87)
∂Z  ∆ρ 

The scaled boundary conditions for the flowing zone (equations 6.3.75 to
6.3.77) are:

Pf =0 …(6.3.88)
Zf

∂V
=0 …(6.3.89)
∂Z Z
f

V Z = −α …(6.3.90)
c

420
Integrating the radial fluid pressure gradient (equation 6.3.87) using the
ambient pressure boundary condition (6.3.88) gives an expression for the scaled fluid
pressure:

ρf
(Z f )
Zs
P f (Z , S ) = − Z + ∫ φdZ ; Zc < Z < Z f …(6.3.91)
∆ρ Z

The upper limit for the integral is Zs rather than Zf since φ = 0 for Z > Zs.
Differentiating equation 6.3.91 with respect to S and eliminating the pressure gradient
using equation 6.3.86 gives

∂  ∂V  ρ f dZ f dZ s Z s ∂φ
 (1 − Z ) = − − φ ( Z ) − ∫ dZ …(6.3.92)
∂Z 
s
∂Z  ∆ρ dS dS Z ∂S

Integrating equation 6.3.92 with respect to Z and evaluating the stress free
boundary condition (6.3.89) gives an expression for the velocity gradient:

 ρ f dZ f dZ  Z s ∂φ
(1 − Z ) ∂V
∂Z
(
= Zf −Z 
∆ ρ
) + φ (Z s ) s  + ∫

(z − Z )dz
 dS dS  Z S

…(6.3.93)

where z is an introduced integration variable. Integrating equation 6.3.93 using the


stick boundary condition (6.3.90) gives a complex expression for the velocity field in
the flowing region.

ρ f   1− Z 
V = −α +  (
Z f ′ + Z s ′ φ Z   Z − Z c + 1 − Z f ln) 
 ∆ρ s 
  1 − Zc 
 Z s ∂φ    
 ∫  z − Z c + (1 − z ) ln 1 − z  dz Z > Zs 
  1 − Z 
 Z c ∂S   c  
Z 
 ∂φ  z − Z + (1 − z ) ln 1 − z  dz 
+ ∫  1 − Z  
∂S 
c
Z c  c 
 Zs Z < Z s 
+ ∫ ∂φ  Z − Z + (1 − z ) ln 1 − Z  dz 
 Z ∂S  c  1 − Z  
  c  
…(6.3.94)

421
An expression for the shock at Zs is given by integrating equation 6.3.85 from
just below the shock, Zs-, to just above the shock, Zs+:

dZ s
[ψ ]Z s−
Zs
=− …(6.3.95)
Z s+
V Z [φ ]
dS
s Z s−

Substituting equation 6.1.20 for ψ into equation 6.3.95 and recognising that
the particle pressure is zero in the flowing region and φ = 0 in the clear-liquor zone
gives:

dZ s (1 − Z s )
= …(6.3.96)
dS V Z B (φ ) Z −
s s

The shock at Zc is calculated in the same manner:

dZ c
[ψ ]Z c−
Zc
= …(6.3.97)
Z c+
α [φ ]
dS
Z c−

Substituting equation 6.1.17 for ψ, with the top of the bed at the gel point, φg,
above which there is no particulate network (therefore Pp = 0) gives:

dZ c
=
(1 − Z c ) ∆(φ ) ∂φ +
φg

φZ 
+
c
 …(6.3.98)
dS α (φg − φ Z + )  ∂Z B(φ g ) B(φ Z ) 
g
c  Z c− +
c 

Scaled Global Conservation Equations

Since q is scaled with qo, from equation 6.3.66, the scaled overall volumetric
balance around the overflow end of the centrifuge is:

Zf
− 1 = −Qo, F (S ) + Qo,C (S ) = ∫ V (Z , S )dZ + αZ c …(6.3.99)
Zc

The scaled flux in the flowing region, Qo,F is:

422
Z f Zf 
Qo, F (s ) = −  ∫ (1 − φ )V (Z , S )dZ + ∫ φV (Z , S )dZ  …(6.3.100)
 Z c Zc 

Likewise, the scaled flux in the consolidation zone, Qo,C is:

Z c Zc 
Qo,C (S ) = α  ∫ (1 − φ )dZ + ∫ φdZ  …(6.3.101)
 0 0 

From equation 6.3.69, the overall scaled particle balance is:

Zf Zc
∫ φV (Z , S )dZ + α ∫ φdZ = 0 …(6.3.102)
Zc 0

Subtracting equation 6.3.102 from equation 6.3.99 gives scaled fluid balance:

Zf Zc
− 1 = ∫ (1 − φ )V (Z , S )dZ + α ∫ (1 − φ )dZ …(6.3.103)
Zc 0

The volumetric flow balances are used to link the solutions of the flowing and
consolidation regions.

423
6.3.4 Inlet Section

A schematic of the inlet section is shown in Figure 6.3.11. The inlet is an


unknown region of width ∆si where the feed is mixed with the clear-liquor from the
underflow, which then proceeds to sediment in the overflow section. At the bottom of
the inlet section is the consolidated cake from the overflow section.

qi, φi

rf
qo,F(si), φsi qu,F(si +∆si)
rc
qo,C(si) qu,C(si+∆si)
rb
∆si

Figure 6.3.11: Schematic of inlet section of decanting centrifuge

The overall volumetric balance for the inlet section is:

q i = q o, F (si ) − q o,C (s i ) − q u , F (si + ∆s i ) + q u ,C (s i + ∆s i ) …(6.3.104)

where qu,F and qu,C are the flowrates of the underflow clear-liquour and consolidation
zones respectively.

∆si is expected to be small compared to the overall length of the centrifuge –


therefore the simplifying assumption is made that ∆si = 0 and that mixing is
instantaneous and complete. Thus, there is no need to consider the sedimentation or
consolidation of particles or the mechanism of mixing, and the sedimentation zone of
the overflow section at si is at constant concentration, φsi. Based on this assumption,
the concentration profile and flowrates of the consolidation zones of the overflow and
underflow sections are equal at si. The volumetric balance becomes:

424
q i = q o, F (s i ) − q u , F (s i ) …(6.3.105)

The solids balance over the upper region of the inlet section is:

rs (si )
φ i qi = φ si q o, F (si ) = −2πφ si ∫ rv(r , si )dr …(6.3.106)
rf

Using the scalings for the overflow, equation 6.3.106 becomes:

Zf
φ i Qi = −φ si ∫ V (Z , S i )dZ …(6.3.107)
Z s (Si )

The assumption is made in the overflow section that the scroll pushes all
networked material through the centrifuge, while all un-networked material flows
down the channel formed by the blades of the scroll. Therefore, this formulation is
valid for conditions when φsi < φg.

425
6.3.5 Underflow Section

The underflow section of the decanting centrifuge consists of up to four


regions, depending on the relative positions of the beach and inlet, and the dewatering
behaviour (see Figure 6.3.12).


qu, φu

rf (s) rd (s)
qu,F(si )
rc (s) rcc(s)
qu,C(si)
rb (s)
sd
si sb scap

Figure 6.3.12: Schematic of underflow section of decanting centrifuge

The constant radius region extends from the inlet to the beginning of the beach
(si < s < sb), and consists of consolidation and clear-liquor zones. There is no
sedimentation zone since there is no un-networked material. The flow field of the
clear-liquor zone is described by the Navier-Stoke’s equations, while the scroll
velocity and the radial cake compaction govern the consolidation zone.

At sb, the bowl radius becomes variable with axial position. Since the
centrifugal acceleration is dependent on radius, a reduction in bowl radius introduces
a region in the consolidation zone where pp is less than Py(φ). It is assumed that the
material is inelastic, such that the zone consists of consolidated and consolidating
regions, with a boundary at rcc(s). The formulation with varying rb can proceed by
either adding a radial component to the flow equations (that is, flow down an inclined
plane), or by changing to conical coordinates.

The third region of the underflow section begins when the fluid height equals
the cake height at scap. At this point, capillary-assisted compaction occurs where the

426
top of the cake is consolidated due to fluid drainage. At sd, the capillary pressure at
the top of the cake equals Py(φ). De-saturation occurs and the liquid front at rd
proceeds into the cake, giving the drainage region (the fourth region of underflow
section). The theory for capillary-assisted drainage and desaturation is in
development, and requires further experimental progress to measure the appropriate
material parameters.

427
6.3.6 Numerical Solution

The decanting centrifuge model was incomplete due to time constraints.


Progress to date consisted of the formulation for the overflow and inlet sections.
Once the underflow section is formulated, a numerical scheme can be devised to
predict the volume fraction distribution for a given set of operating conditions and
material characteristics.

A complex algorithm will be required, since the solution to the consolidation


zone of the overflow section is solved from S = 0, but depends on the flowing region.
In contrast, the flowing region must be solved from Si, but depends on the scrolling
velocity, which depends on the cake concentration. Thus, an iterative approach will
be needed. A possible approach is to start at Si, with φsi = φi and Zc = 0 and solve the
flowing region of the overflow section until Zs = 0. The consolidation zone is then
solved in the overflow and underflow sections. This in turn gives the flowing region
of the underflow section, which is combined with the inlet to give a new starting value
for the overflow. The calculation is repeated until the solution converges.

428
6.4 Conclusions

Three models of the centrifugation of flocculated suspension were described.


The volume fraction distribution for solid-bowl batch centrifugation was calculated
for the two cases of initial suspension networked and un-networked using a Runge-
Kutta numerical technique. The results showed that three zones of behaviour exist: a
clear liquor zone from which material has settled, a sedimentation zone, and a
consolidation zone as cake builds up from the wall of the centrifuge. The volume
fraction distribution in the settling zone was shown to be constant with respect to
radius but changing with time for the initial suspension un-networked case, and
constant for the initial suspension networked case.

Models of pseudo-one-dimensional continuous centrifugation for the inlet


suspension networked and un-networked cases were formulated and numerical
schemes for the solution of the governing equations using a Runge-Kutta technique
were outlined. The results for ferric water treatment sludge were shown. As for the
transient case, three zones of behaviour were seen. The volume fraction in the
sediment was a function of radius for the inlet suspension un-networked case, and
constant for the networked case. The model represented the centrifugal analogy of
continuous gravity thickening, such that the throughput and the bed height were
functions of the underflow concentration.

The third model presented investigated the solid-liquid separation performed


by a counter-current decanting centrifuge. The formulations for the overflow and
inlet sections were given. In order to continue this work, the underflow section is to
be formulated, a numerical scheme devised and the model validated using plant-scale
equipment. The validation will require experimental techniques and protocols for the
measurement of certain material properties, such as capillary pressure and shear yield
stress, to be developed.

Several operational conclusions were drawn from the three models regarding
decanting centrifuges. The one-dimensional models showed that high solids loading
was necessary to achieve high concentrations, especially for low density solids such
as sewage sludges. However, this would cause higher torques and power

429
consumptions and lower the extent of desaturation on the beach, such that an optimum
differential speed exists for a given throughput and pond height. Likewise, an
optimum feed concentration exists, which is a compromise between hindered settling
behaviour and cake concentration. A significant result from the decanting model is
that feed suspensions significantly above φg (such that φsi > φg) will not flow down the
scroll channel, causing catastrophic failure. Other optimums and limits for the
various operating and design parameters, such as cone angle or scroll pitch, will be
evident from the completed model.

430
7 Thickener Modelling

7.1 Water and Wastewater Sludge Thickening


Predictions

7.1.1 Background

Traditional characterisation of water and wastewater treatment sludges focuses


on either filtration testing at high solids concentrations or empirical batch settling
procedures for low solids concentrations. However, for the non-empirical modelling
of thickening, quantitative knowledge of the behaviour at low solids concentrations is
critical. In recent work for the minerals industry, Ross de Kretser, Daniel Lester and
Shane Usher from the University of Melbourne addressed these issues and significant
advances were made in the characterisation of suspensions at low solids
concentrations, culminating in the advanced batch settling analysis software used
within this work (see Section 3.1.6). The software analyses batch settling test data
and, in conjunction with high solids de-watering parameter data (obtained via
traditional methods), outputs complete Py(φ) and R(φ) curves from low to high solids
concentrations.

Numerical models of continuous thickening were also developed as part of


that work, based on the model of Howells, et al., 1990. These were coded into a
module of the batch settling analysis software, enabling complete characterisation of
the dewaterability of a sample followed by generation of thickener modelling results.
This section outlines the first application of these tools to the modelling of water and
wastewater treatment sludge thickening. The advantage of the modelling approach
employed was its capacity to deal with compression effects in the bed of settled

431
sludge – an important feature missing from the standard water industry approach used
for design and modelling of thickeners, called the WRC model (Warden, 1983,
Dillon, 1997), which is based on Kynch’s model (Kynch, 1952).

The thickener model underwent preliminary validation on pilot and plant scale
mineral processing sites both in Australia and internationally. The results were good
in thickeners with low degrees of shear (for example, no raking). However the model
makes no allowance for shear and further work in this area is underway.

7.1.2 Results and Discussion

Characterisation

Batch sedimentation tests were conducted on a ferric water treatment sludge


from Langsett WTP, alum water treatment sludges taken from Huntington WTP and
Oswestry WTP, and a wastewater sludge from Luggage Point WWTP. The results
and their analysis are presented in Sections 4.1.1, 4.1.2 and 4.2.2 respectively.
Inclusion of the batch settling data made a large difference to the Py(φ) and R(φ) at
low solids concentrations and had a significant impact on predictions.

Thickener Modelling

The thickener modelling results were determined directly from the output of
the batch settling analysis program. The thickeners were assumed to be flat-based and
straight-sided, with no raking and an internal area of 1 m2 (giving solids throughputs
per unit area). Thus, the results were generalised for any flat-bottomed thickener.
The feed concentration was the same as the initial solids concentration employed in
the batch settling experiment. The feed concentration effects only very high
throughputs and low underflow solids.

Figure 7.1.1 illustrates results from thickener modelling using the dewatering
parameter data from the batch settling analysis program. Such data plots aid in
understanding how to control or optimise operation of thickeners from an underflow
solids point of view. The graph shows the underflow solids expected for a certain

432
solids throughput (measured as the solids flux) and bed depth. The results are divided
into two sections.

- Low solids / High fluxes – in this region, operation of the thickener is


permeability limited and the depth of the bed has no impact on the underflow
solids concentration. What is limiting the underflow solids is not the stiffness
of the bed, but the rate at which liquid can escape.

- High solids / Low Fluxes – in this region, operation of the thickener is


compressibility limited and the underflow solids concentration depends on the
bed depth and the compressive force it generates relative to the strength of the
bed. As the flux is reduced, the importance of the bed permeability becomes
less.

The thickener modelling results for Langsett WTP 09/07/03 Sample 1 at a


range of bed heights are presented in Figure 7.1.1. At high fluxes, φu was independent
of hb. For example, a throughput of 10-3 tds/hr/m2 fixed φu at 0.009 v/v. At lower
fluxes, φu was dependent on hb as well as the flux. For example, at a flux of 10-4
tds/hr/m2, φu increased from 0.0101 v/v at a bed height of 1 m to 0.0131 v/v at a bed
height of 10 m. The results showed that underflow concentrations of up to 0.02 v/v
(5.8 wt%) were possible for this material, but only at low throughputs and large bed
heights.

The modelling results for the Langsett WTP, Huntington WTP and Oswestry
WTP samples at bed heights of 1 m and 10 m are presented in Figure 7.1.2, which
further illustrate the low solids / high flux and high solids / low flux behaviour of
continuous thickeners. The Huntington WTP sludge gave similar predictions to the
Langsett WTP sludge, despite being alum and ferric sludges respectively. The
Oswestry WTP sludge, which was also an alum sludge, gave predictions of lower
throughputs and lower underflow concentrations. These trends were reflected in the
material characterisation results.

433
1.E-01
Langsett (1 m)
Langsett (2 m)
Langsett (5 m)
1.E-02 Langsett (10 m)
Solids Flux (tds/hr/m )
2 Langsett (Perm. Limit)

1.E-03

1.E-04

1.E-05

φ0 φg
1.E-06
0 0.005 0.01 0.015 0.02
Underflow Solids Concentration, φ u (v/v)

Figure 7.1.1: Thickener modelling output for Langsett thickener Sample 1 - Solids flux versus
underflow solids concentration for four different bed heights (1, 2, 5, 10 m)

1.E-01
Langsett (1 m)
Langsett (10 m)
Huntington (1 m)
Solids Flux (tds/hr/m )

1.E-02
2

Huntington (10 m)
Oswestry (1 m)
Oswestry (10 m)
1.E-03

1.E-04

1.E-05

1.E-06
0 0.005 0.01 0.015 0.02
Underflow Solids Concentration, φ u (v/v)

Figure 7.1.2: Thickener modelling output for Langsett, Huntington and Oswestry thickener
samples - Solids flux versus underflow solids concentration for bed heights of 1 and 10 m

434
The thickener modelling results for Luggage Point WWTP Sample 2 are
presented in Figure 7.1.3. The material was less permeable but more compressible
than the water treatment sludges, such that higher underflow concentrations were
possible (up to 0.04 v/v for bed heights up to 10 m).

1.E-01
LP S2A (1 m)
LP S2A (2 m)
LP S2A (5 m)
1.E-02 LP S2A (10 m)
Solids Flux (tds/hr/m )
2

LP S2A (Perm. Limit)

1.E-03

1.E-04

1.E-05

φ0 φg
1.E-06
0 0.01 0.02 0.03 0.04 0.05
Underflow Solids Concentration, φ u (v/v)

Figure 7.1.3: Thickener modelling output for Luggage Point Sample 2 - Solids flux versus
underflow solids concentration for four different bed heights (1, 2, 5, 10 m)

Results such as in Figure 7.1.1 to 7.1.3 can be practically used in a number of


ways. Based on the operational flux, the sensitivity of the underflow solids
concentration to variations in feed rate can be assessed. Alternatively, if optimisation
of additives is required, batch settling characterisation of different additives and doses
can be conducted followed by thickener modelling. The curves can then be compared
and the additive that demonstrates the highest underflow solids for the operating flux
selected as the optimal. Likewise, comparisons of sludge types can be made.

435
7.1.3 Conclusions

Application of software for fully characterising the dewaterability of


suspensions via batch settling tests and continuous thickener modelling tools was
done for the first time for water and wastewater treatment sludges. The software was
developed via minerals industry research at the University of Melbourne. Sludges
from Langsett WTP, Huntington WTP, Oswestry WTP and Luggage Point WWTP
were successfully characterised from low to high solids concentrations using the
technique and the results used in a continuous thickener modelling tool. The
outcomes from this work demonstrated the potential for optimisation and control of
continuous thickeners based on the properties of the feed material to give more stable
underflow concentrations. In particular, the bed height dependence on underflow
solids was implemented successfully at Langsett WTP to increase the concentration of
the filter press feed, with corresponding increases to filter throughput, resulting in
considerable cost savings.

Traditionally, shallow, raked thickeners are used by the water and wastewater
industries. This work showed that deep thickeners with large bed heights would allow
higher underflow concentrations, providing increases to the throughput of subsequent
dewatering devices. The thickeners are generally operated continuously, although
some batch and semi-continuous operations were observed (for example, at Thornton
Steward WTP) such that the extent of dewatering depended on the settling time and
the feed concentration to subsequent filters and centrifuges varied with time. The
modelling work presented in Chapters 5 and 6 showed that the optimum operating
conditions were when the feed concentration was maximised. Thus, constant output
at medium to high φu for continuous thickeners was preferred to variable output from
high to low φu for batch thickeners.

436
8 Conclusions

Water and wastewater treatment produce sludge that is dewatered prior to


disposal. Overviews of the treatment processes, sludge composition and sludge
disposal routes for both water and wastewater treatment are given. In general, the
amount of sludge to be dewatered and the cost of disposal are expected to increase.
Common dewatering devices used in sludge handling are thickeners, filters and
centrifuges. This work applied phenomenological solid-liquid separation theory to the
mathematical modelling of dewatering devices and measured the sludge dewatering
properties in order to give improvements in operation, design and control of such
devices, including increased throughput and reduced water content, thus reducing the
costs of sludge disposal.

The theoretical basis used the local volume fraction dependent properties of
compressive yield stress, Py(φ), and hindered settling function, R(φ), to describe
dewatering (which, when combined, give the solids diffusivity, D(φ)). The local mass
and momentum balances were given in vector notation, converted to one-dimension,
and applied to piston-driven filtration. A range of solutions for differing initial
concentrations and membrane resistances were derived, and their application to
experimental characterisation discussed.

For water treatment sludges, which show traditional filtration behaviour, the
slope of the linear section of a plot of t versus V2 gives a measure of the permeability,
while a measure of the compressibility is given by the end-point. Stepped-pressure
filtration tests are useful to give material characteristics over a range of pressures in a
short time. These techniques were used to measure the material characteristics for
water treatment sludges from a range of industrial sites, including different coagulants
and coagulation conditions, with weekly and seasonal variations. The results showed
that ferric-based sludges have very little variability, while alum sludges can exhibit
some changes in properties. Water treatment sludges are less permeable and have

437
lower equilibrium solids concentrations compared to flocculated mineral and clay
suspensions.

Wastewater sludges, on the other hand, exhibit non-traditional filtration


behaviour, where the linear section is very short or non-existent and equilibrium is
reached only after a long time (and the accuracy is affected). The development of
synthetic sewage sludge showed that this is caused by the extracellular polymeric
constituent. The filtration theory was used to show that initial height and
concentration dependencies could be accounted for and qualitative comparisons
between sludges could be made. The results from a range of wastewater treatment
plants with differing additives, doses and treatments were compared using this scaling
method. The theory was also used to show that non-traditional behaviour is predicted
when D(φ) significantly decreases over the volume fraction range of interest, such that
this behaviour is dependent on both the material characteristics and the operating
conditions. A method for the measurement of the material characteristics of non-
traditional materials from single-pressure filtration tests based on the compression
stage of the filtration profile was described and theoretically validated. This method
was used to characterise a wastewater sludge sample over a range of volume fractions
– the results were validated using the filtration model. The results described a highly
compressible but impermeable material, and D(φ) exhibited a maximum at low φ.

The settling characteristics of both water and wastewater sludges were


measured using transient batch settling tests in conjunction with an analysis program
developed at the University of Melbourne. The results showed that Py(φ) and R(φ)
change by many orders of magnitude – modelling that does not allow for this will not
be accurate. The results were used with a previously developed model to predict the
throughput and underflow solids for a flat-bottomed continuous thickener.

Models of plate-and-frame filter presses were formulated and solution


algorithms outlined. The models described the initially ramping pressure, followed by
constant pressure fill and squeeze stages, and included the effect of membrane
resistance. Simple Visual Basic programs were written to provide operators and
designers access to the model predictions. The models were validated during several
on-site case studies at a range of water treatment plants, and used in conjunction with
the material characterisation results to investigate the optimisation and control of

438
press performance. The optimisation of fill-only presses was shown to be limited,
whereas fill-and-squeeze presses have greater versatility. The volume of filtrate must
be measured in order to control filter presses, while the feed concentration has the
greatest impact on performance and should also be measured. A modified-Darcy’s
law, which was validated using model predictions, was used to determine the
resistance of the membrane from on-site measurements.

Three different models of solid-bowl centrifugation were formulated,


including one-dimensional transient and continuous, and two-dimensional continuous
decanting models. Algorithms for the solution of the one-dimensional models were
developed for both the un-networked and networked cases, and the results for a
simple example illustrated. The development of the decanting centrifuge model
included conversion of the mass and momentum conservation equations to helical
coordinates, consideration of the viscous flow of the centrate down the spiral formed
by the blades of the scroll, and prediction of the movement of the solids due to
differential speed of the scroll. An overview of the work required to complete the
two-dimensional model was given.

As an aside, civil engineering consolidation theory and testing methods were


outlined and compared to filtration. The two dewatering parameters, cv and D, were
related via a simple equation. The material characteristics of a kaolin sample were
measured using oedometer and pressure filtration testing and compared using the
derived relationship. The two fields of research are very similar, which invites further
collaboration between the disciplines.

Overall, the work contained in this thesis represents several important


developments in the understanding of water and wastewater sludge dewatering,
especially the measurement of water treatment sludge characteristics at a range of
sites and with seasonal variations, the development of analysis techniques to
determine wastewater sludge characteristics, and the development of new models and
the adaptation of existing models of filter presses and centrifuges. By understanding
both the material behaviour and the dewatering device, considerable savings in the
economic, social and environmental disposal costs were made possible.

439
Nomenclature

Latin alphabet

a, b Indices of T in Rm ≠ 0 small-time approximations


A Calculation variable for φ0 > φg small-time solution
A(r) Cross-sectional area
An Fourier series for cake compression linear approximation
Apress Total filter press membrane area
B(φ) Scaled hindered settling function
cv Consolidation coefficient (consolidation theory)
d Cavity width (plate-and-frame filter)
d Nominal maximum pore drainage length (consolidation theory)
d1, d2, d3 Solids diffusivity power-law parameters
D(φ) Solids diffusivity
Dmax Maximum solids diffusivity
e Void ratio (including e0, e∞, eg and e*)
E(X) Void ratio distribution for exact similarity solution
E1, E2, E3 Logarithmic cake compression constants
f Algorithm stopping fraction
f(φ) Scaled compressive yield stress
fbk(φ) Kynch batch flux density function
g Acceleration vector
g Gravitational acceleration
G(η) Fill-stage exact similarity solution
h(t) Piston height (filtration); Interface height (sedimentation)
hb(t) Bed height
h0 Initial piston height
hc Piston height at tc
h∞ Equilibrium piston / bed height

440
H(T) Scaled piston height
Hc Scaled piston height at Tc
J Total number of spatial steps (finite difference model)
k Hydraulic conductivity (consolidation theory)
km membrane permeability
K Bed permeability (Darcy’s law)
K” Empirical constant (Kozeny-Carman equation)
lm Membrane width
L Bed length (Darcy’s law)
mv Coefficient of volume compressibility (consolidation theory)
N Total number of time steps (finite difference model)
pp , pf Particle and fluid pressures
p1 , p2 Compressive yield stress power-law parameters
P Pitch (decanting centrifuge)
Pp, Pf Scaled particle and fluid pressures
Py(φ) Compressive yield stress
q(t) Bulk flow
q Throughput
qL Pumping flowrate during press loading
Q(Z,T) Scaled cumulative solids volume
Q Scaled throughput
<Q> Average specific throughput
Qrun Specific throughput per run
r Radial coordinate
r Finite difference stability and convergence factor
r1 , r2 Hindered settling function power-law parameters
Rc Cake resistance
Rm Membrane resistance
R(φ) Hindered settling function
ŝ Helical coordinate (decanting centrifuge)
S Scaled helical coordinate (decanting centrifuge)
Sp Particle specific surface area
t Time (s)

441
t̂ Tangential helical coordinate (decanting centrifuge)
t50 Time at 50% consolidation (Casagrande’s method)
t90 Time at 90% consolidation (Taylor’s method)
tc Critical time for cake compression (filtration) or for sediment to
disappear (centrifuge)
tf Filtration time
tF Total fill-stage time
tH Handling time
tL Filter press loading time
tP Filter press ramping pressure time
tS Filter press squeeze time
tT Total filter press batch cycle time
T Scaled time
Tc Scaled critical time
Tf Scaled filtration time
Tfilt Filtration time-scale
Ts Theoretical scaled time when Zs = 0 (φ0 < φg)
Tsed Sedimentation time-scale
Tv Dimensionless consolidation time (consolidation theory)
u, u(r,t) Local velocity (m/s)
U Degree of consolidation (consolidation theory)
v, v(r,s) Bulk velocity (decanting centrifuge)
V(Z,S) Scaled velocity (decanting centrifuge)
V(t) Specific filtrate volume
V0 Initial suspension volume per unit length (batch centrifuge)
V∞ Equilibrium specific filtrate volume
Va Consolidation zone velocity relative to centrifuge axis
Vc Consolidation zone velocity relative to scroll blades
Vp Particle volume
Vpress Press volume
w Material coordinate
wc(T) Cake height in material coordinates
W(r) Centrifuge width

442
X Similarity variable (scaled spatial coordinate)
z Length coordinate; axial coordinate (decanting centrifuge)
zc(t) Cake height
Z Scaled spatial coordinate
Zc(t) Scaled cake height

Greek alphabet

α Linear approximation calculation variable (filtration)


α Small time scaled cake radius coefficient (batch centrifuge)
α Scaled cake velocity (decanting centrifuge)
αm Specific cake resistance per unit mass
αv Specific cake resistance per unit volume
α(Z) Scaled cross-sectional area (continuous centrifuge)
β Constant of proportionality between V and t½ (cake formation)
β Scroll angle (decanting centrifuge)
βm Scaled membrane resistance
δ(e) Scaled solids diffusivity (explicit)
∆(φ) Scaled solids diffusivity
∆(e) Scaled solids diffusivity (implicit)
∆eff Step function approximation to ∆(e)
∆P Applied pressure
∆T Temporal step size (finite difference and Runge-Kutta methods)
∆w Spatial step size (finite difference method)
∆ρ Density difference between solid and liquid phases
∆ω Differential scroll speed (decanting centrifuge)
ε Voidage
φ(r,t) Local solids volume fraction
<φ> Average solids volume fraction
φ* Shifted gel point (filtration linear approximation)
φ0 Initial solids volume fraction
φ∞ Equilibrium solids volume fraction
φb(T) Average bed solids volume fraction

443
φg Gel point solids volume fraction
φmax Volume fraction at Dmax
Φ(X) Small time variation of solids volume fraction
γ Blade angle (decanting centrifuge)
γf Weight of fluid (consolidation theory)
γ(T) Scaled filtrate volume (filter press model)
η Change of variables (fill-stage exact solution)
ηf Fluid viscosity
κ(φ) Dynamic compressibility
Κ(φ) Scaled dynamic compressibility
λ Small-time scaled sediment radius velocity (batch centrifuge)
λ Single particle drag coefficient
ω Angular velocity
ρ Density
θ Cylindrical angular coordinate
σy Shear yield stress
σ’ Effective stress (consolidation theory)
σe(φ) Effective solids stress function
Σ(T) Scaled applied pressure
ξ Similarity variable
ψ(Z,T) Scaled local solids flux

Subscripts and Superscripts

+/- Upper and lower discontinuity limit values


< Value at previous time step
* Estimate for iteration variable

b Bowl
c Cake
C Consolidating region (decanting centrifuge)
f Fluid / liquid
F Fill-stage (filtration); flowing region (decanting centrifuge)

444
high/low Upper and lower bounds for iteration variable
H Handling
i Inlet
L Loading (filter press)
lim Limiting values for visual basic datafiles
m Membrane
max Maximum value
n Spatial position (finite difference method)
o Overflow
p Particle / solid
P Ramping pressure (filtration)
ref Reference value
s Sediment / sedimentation zone
S Squeeze (filtration)
T Total
test Test value for iteration variable
u Underflow
0 Initial
∞ Equilibrium

445
References

Abd Aziz, A. A., PhD Thesis: Characterisation of shear upon dewaterability of


colloidal suspensions, The Department of Chemical and Biomolecular Engineering,
The University of Melbourne, (2004)

Abd Aziz, A. A., de Kretser, R. G., Dixon, D. R. and Scales, P. J., "The
characterisation of slurry dewatering", Wat. Sci. Tech., 41 (8), 9 (2000)

ADAS, "The safe sludge matrix: Guidelines for the application of sewage sludge to
agricultural land (3rd ed)", 2001)

ADWG, "Australian Drinking Water Guidelines", National Health and Medical


Research Council and the Agricultural and Resource Management Council of
Australia and New Zealand1996)

Ambler, C. M., "The evaluation of centrifuge performance", Chem. Eng. Progr., 48


(3), 150 (1952)

Ambler, C. M., "The theory of scaling up laboratory data for the sedimentation type
centrifuge", J. Biochem. Micro. Tech. Eng., 1 (2), 185 (1959)

Anderson, N. J., Dixon, D. R., Harbour, P. J. and Scales, P. J., "Complete


characterisation of thermally treated sludges", Wat. Sci. Tech., 46 (10), 51 (2002)

Anestis, G. and Schneider, W., "Application of the theory of kinematic waves to the
centrifugation of suspensions", Ing.-Arch., 53, 399 (1983)

Annan, K. A., "Millennium Report of the Secretary-General of the United Nations",


The United Nations2000)

446
Appleton, A., Personal Communication, Yorkshire Water Sludge Manager (2004)

AS, "Australian standard methods of testing soil for engineering purposes. Part F -
Soil strength and consolidation tests. F6.1 - Determination of the one-dimensional
consolidation properties of a soil", AS 1289.6.6.1-1998 F6.1 (1998)

ASTM, "Standard test method for one-dimensional consolidation properties of soils",


D2435-96 (2001)

Auzerais, F. M., Jackson, R., Russel, W. B. and Murphy, W. F., "The transient settling
of stable and flocculated dispersions", J. Fluid Mech., 221, 613 (1990)

AWWA, Water quality and treatment: A handbook of community water supplies (5th
ed), McGraw-Hill, New York (1999)

Baskerville, R. C. and Gale, R. S., "A simple automatic instrument for determining
the filterability of sewage sludges", J. Inst. Wat. Pollut. Control, 67, 233 (1968)

Biever, C., "Plugging into sewage power", New Scientist, (2438), 21 (2004)

Binnie, C., Kimber, M. and Smethurst, G., Basic water treatment (3rd ed), Thomas
Telford, London (2002)

Bruus, J. H., Nielsen, P. H. and Keiding, K., "On the stability of activated sludge flocs
with implications to dewatering", Wat. Res., 26 (12), 1597 (1992)

Bürger, R., "Phenomenological foundation and mathematical theory of sedimentation-


consolidation processes", Chem. Eng. J., 80, 177 (2000)

Bürger, R. and Concha, F., "Mathematical model and numerical simulation of the
settling of flocculated suspensions", Int. J. Multiphase Flow, 24, 1005 (1998)

447
Bürger, R. and Concha, F., "Settling velocities of particulate systems: 12. Batch
centrifugation of flocculated suspensions", Int. J. Miner. Process., 63, 115 (2001)

Bürger, R., Concha, F. and Karlsen, K. H., "Phenomenological model of filtration


processes: 1. Cake formation and expression", Chem. Eng. Sci., 56, 4537 (2001)

Buscall, R., Mills, P. D. A., Stewart, R. F., Sutton, D., White, L. R. and Yates, G. E.,
"The rheology of strongly-flocculated suspensions", J. Non-Newtonian Fluid Mech.,
24 (2), 183 (1987)

Buscall, R. and White, L. R., "The consolidation of concentrated suspensions. Part 1. -


The theory of sedimentation", J. Chem. Soc., Faraday Trans. 1, 83, 873 (1987)

Casagrande, A. and Fadum, R. E., "Notes on soil testing for engineering purposes",
Harvard Univ - Graduate School Eng - Publ (Soil Mechanics Series), 8, 268 (1940)

Casagrande, A. and Fadum, R. E., "Application of soil mechanics in designing


building foundations", Am. Soc. Civ. Engrs. - Proc., 68 (9), 1487 (1942)

Casey, T. J., Unit treatment processes in water and wastewater engineering, John
Wiley & sons, Chichester (1997)

Channell, G. M. and Zukoski, C. F., "Shear and compressive rheology of aggregated


alumina suspensions", AIChE J., 43 (7), 1700 (1997)

Chen, Y., Yang, H. and Gu, G., "Effect of acid and surfactant treatment on activated
sludge dewatering and settling", Wat. Res., 35 (11), 2615 (2001)

Coombes, R., Personal Communication, United Utilities Sludge Manager (2004)

Corner-Walker, N., "The dry solids centrifuge: conveyer torque and differential",
Filtr. Sep., 37 (8), 18 (2000)

448
Corner-Walker, N., "The dry solids decanter centrifuge: capacity scaling", Filtr. Sep.,
37 (4), 28 (2000)

Cousin, C. P. and Ganczarczyk, J. J., "Effects of salinity on physical characteristics of


activated sludge flocs", Wat. Qual. Res. J. Canada, 33 (4), 565 (1998)

Craig, R. F., Soil mechanics (6th ed), E & FN Spon, London (1997)

Dahlstrohm, D. A., Bennett, R. C., Emmett, R. C. J., Harriott, P., Laros, T., Leung,
W., McCleary, C., Miller, S. A., Morey, B., Oldshue, J. Y., Priday, G., Silverblatt, C.
E., Slottee, J. S., Smith, J. C. and Todd, D. B., "Section 18: Liquid-solid operations
and equipment" in Perry's Chemical Engineers' Handbook (7th ed), Green, D. W.,
McGraw-Hill, Sydney (1998)

Darcy, H. P. G., Les fontaines publiques de la ville de Dijon, Dalamont, Paris (1856)

Davies, A., "Saltwater solution for thirsty city: Sartor", Sydney Morning Herald, 18
September 2004, 3

Davis, E. H. and Raymond, G. P., "A non-linear theory of consolidation",


Geotechnique, 15 (2), 161 (1965)

de Boer, R., "Highlights in the historical development of the porous media theory:
Toward a consistent macroscopic theory", Appl. Mech. Rev., 49 (4), 201 (1996)

de Kretser, R. G., Scales, P. J. and Usher, S. P., "Practically applicable modelling of


cake filtration", Science & Technology of Filtration and Separations for the 21st
century, Tampa, Florida, (2001)

de Kretser, R. G., Usher, S. P., Scales, P. J., Boger, D. V. and Landman, K. A., "Rapid
filtration measurement of dewatering design and optimisation parameters", AIChE J.,
47 (8), 1758 (2001)

449
Dentel, S. K., "Evaluation and role of rheological properties in sludge management",
Wat. Sci. Tech., 36 (11), 1 (1997)

Dillon, G., "Application guide to waterworks sludge treatment and disposal", Report
No.: TT016, Water Research Council, June (1997)

Dixon, D. C., "Effect of sludge funnelling in gravity thickeners", AIChE J., 26 (3),
471 (1980)

Duncan, M. J., "Limitation of conventional analysis of consolidation settlement", J.


Geotech. Eng., 119 (9), 1333 (1993)

Dupuit, A. J. E. J., Etudes theoriques et pratiques sur le movement des eaux,


Dalamont, Paris (1863)

Eberl, M., Landman, K. A. and Scales, P. J., "Scale up procedures and test methods in
filtration: A test case on kaolin plant data", Colloids Surf. A, 103, 1 (1995)

Eckert, W. F., Masliyah, J. H., Gray, M. R. and Fedorak, P. M., "Prediction of


sedimentation and consolidation of fine tails", AIChE J., 42 (4), 960 (1996)

EU, "Council Directive 86/278/EEC on the protection of the environment, and in


particular of the soil, when sewage sludge is used in agriculture", 12 June (1986)

EU, "Council Directive 91/271/EEC concerning urban waste water treatment", 21


May (1991)

EU, "Council Directive 91/676/EEC on nitrates from agricultural sources", 12


December (1991)

EU, "Council Directive 98/83/EC on the quality of water intended for human
consumption", 3 November (1998)

EU, "Council Directive 99/31/EC on the landfill of waste", 26 April (1999)

450
EU, "Council Directive 2000/76/EC on the incineration of waste", 28 December
(2000)

Ford, R. R., Personal Communication, Professor of Innovation and Technology


Strategy at the University of Salford; formerly Research and Development Director at
United Utilities (2001)

Gibson, R. E., England, G. L. and Hussey, M. J. L., "The theory of one-dimensional


consolidation of saturated clays. 1. Finite non-linear consolidation of thin
homogenous layers", Geotechnique, 17, 261 (1967)

Green, M. D. and Boger, D. V., "Yielding of suspensions in compression", Ind. Eng.


Chem. Res., 36, 4984 (1997)

Green, M. D., Eberl, M. and Landman, K. A., "Compressive yield stress of


flocculated suspensions: determination via experiment", AIChE J., 42 (8), 2308
(1996)

Green, M. D., Landman, K. A., de Kretser, R. G. and Boger, D. V., "Pressure


filtration technique for complete characterisation of consolidating suspensions", Ind.
Eng. Chem. Res., 37 (10), 4152 (1998)

Gregor, J. E., Fenton, E., Brokenshire, G., van den Brink, P. and O'Sullivan, B.,
"Interactions of calcium and aluminium ions with alginate", Wat. Res., 30 (6), 1319
(1996)

Harbour, P. J., Anderson, N. J., Abd Aziz, A. A., Dixon, D. R., Hillis, P., Scales, P. J.,
Stickland, A. D. and Tillotson, M., "Fundamental dewatering characteristics of
potable water treatment sludges", J. Wat. Supply: Res. Tech. - Aqua, 53 (1), 29 (2004)

Harbour, P. J., Aziz, A. A. A., Scales, P. J. and Dixon, D. R., "Prediction of the
dewatering of selected inorganic sludges", Wat. Sci. Tech., 44 (10), 191 (2001)

451
Henze, M., Harremoes, P., la Cour Jansen, J. and Arvin, E., Wastewater treatment:
Biological and chemical processes (3rd ed), Springer-Verlag, Berlin (2002)

Higgins, M. J. and Novak, J. T., "Characterization of exocellular protein and its role
in bioflocculation", J. Environ. Eng., 123 (5), 479 (1997)

Hildebrand, F. B., Methods of applied mathematics (1st ed), Prentice-Hall,


Englewood Cliffs, N. J. (1952)

Holtz, R. D. and Kovacs, W. D., An introduction to geotechnical engineering,


Prentice-Hall, Englewood Cliffs, N. J. (1981)

Houghton, J. I. and Stephenson, T., "Effect of influent organic content on digested


sludge extracellular polymer content and dewaterability", Wat. Res., 36 (14), 3620
(2002)

Howells, I., Landman, K. A., Panjkov, A., Sirakoff, C. and White, L. R., "Time
dependent batch settling of flocculated suspensions", Appl. Math. Model., 14, 77
(1990)

Janbu, N., "Consolidation of clay layers based on non-linear stress strain",


Proceedings of the 6th International Conference on Soil Mechanics and Foundation
Engineering, Montreal, 83 (1965)

Keiding, K. and Rasmussen, M. R., "Osmotic effects in sludge dewatering", Adv. Env.
Res., 7 (3), 641 (2003)

Kepp, U., Machenbach, I., Weisz, N. and Solheim, O. E., "Enhanced stabilisation of
sewage sludge through thermal hydrolysis - three years of experience with full scale
plant", Wat. Sci. Tech., 42 (9), 89 (2000)

Kopp, J. and Dichtl, N., "Characterisation" in Sludge into biosolids: processing,


disposal, utilization, Spinosa, L. and Vesilind, P. A., IWA Publishing, (2001)

452
Kozeny, J. and Forchheimer, P., "Ueber kapillare leitung des wassers im boden
(Capillary flow of water in soils)", Gas- und Wasserfach, 71 (19), 437 (1928)

Kynch, G. J., "A theory of sedimentation", Trans. Faraday Soc., 48, 166 (1952)

Landman, K. A., Sirakoff, C. and White, L. R., "Dewatering of flocculated


suspensions by pressure filtration", Phys. Fluids A, 3 (6), 1495 (1991)

Landman, K. A., Stankovich, J. M. and White, L. R., "Measurement of the filtration


diffusivity D(φ) of a flocculated suspension." AIChE J., 45 (9), 1875 (1999)

Landman, K. A. and White, L. R., "Solid/liquid separation of flocculated


suspensions", Adv. Colloid Interface Sci., 51, 175 (1994)

Landman, K. A. and White, L. R., "Predicting filtration time and maximizing


throughput in a pressure filter", AIChE J., 43 (12), 3147 (1997)

Landman, K. A., White, L. R. and Buscall, R., "The continuous flow gravity
thickener: steady state behaviour", AIChE J., 34 (2), 239 (1988)

LaPara, T. M. and Alleman, J. E., "Thermophilic aerobic biological wastewater


treatment", Wat. Res., 33 (4), 895 (1999)

Lee, D. J., Ju, S. P., Kwon, J. H. and Tiller, F. M., "Filtration of highly compactible
filter cake: variable internal flow rate", AIChE J., 46 (1), 110 (2000)

Lester, D. R., Usher, S. P. and Scales, P. J., "Estimation of the hindered settling
function R(φ) from batch settling tests", AIChE J., 51 (4), 1158 (2005)

Liu, H. and Fang, H. P., "Extraction of extracellular polymeric substances (EPS) of


sludges", J. Biotechnol., 95 (3), 249 (2002)

Lowe, I., "Dewatering the closet", New Scientist, 183 (2456), 45 (2004)

453
Lutz, T., Jungschaffer, G. and Sprossler, B., "Improved sludge dewatering by
enzymatic treatment", Wat. Sci. Tech., 28 (1), 189 (1993)

Martin, A. D., "Filtration of flocculated suspensions under declining pressure", AIChE


J., 50 (7), 1418 (2004)

Mason, C. A., Haener, A. and Hamer, G., "Aerobic thermophilic waste sludge
treatment", Wat. Sci. Tech., 25 (1), 113 (1992)

Matthews, J. H., Numerical methods for mathematics, science, and engineering (2nd
ed), Prentice-Hall, New Jersey (1992)

Meireles, M., Molle, C., Clifton, M. J. and Aimar, P., "The origin of high hydraulic
resistance for filter cakes of deformable particles: Cell-bed deformation or surface-
layer effect?" Chem. Eng. Sci., 59 (24), 5819 (2004)

Mesri, G., Feng, T. W. and Shahien, M., "Coefficient of consolidation by inflection


point method", J. Geotech. Geoenviron. Eng., 125 (8), 716 (1999)

Metcalf and Eddy, I., "Biological unit processes" in Wastewater engineering:


treatment, disposal, and reuse (3rd ed), Tchobanoglous, G. and Burton, F. L.,
McGraw-Hill, New York (1991)

Mikasa, M., "The consolidation of soft clay" in Civil engineering in Japan, Japanese
Society of Civil Engineering, (1965)

Mikkelsen, L. H. and Keiding, K., "The shear sensitivity of activated sludge: An


evaluation of the possibility for a standardised floc strength test", Wat. Res., 36 (12),
2931 (2002)

Miller, K. T., Melant, R. M. and Zukoski, C. F., "Comparison of the compressive


yield response of aggregated suspensions: Pressure filtration, centrifugation, and
osmotic consolidation", J. Am. Ceram. Soc., 79 (10), 2545 (1996)

454
Neyens, E. and Baeyens, J., "A review of thermal sludge pre-treatment processes to
improve dewaterability", J. Hazard. Mater., 98 (1-3), 51 (2003)

Northcott, K. A., PhD Thesis: Development and application of particle separator


technology for the removal of particles from contaminated water in Antarctica, The
Department of Chemical and Biomolecular Engineering, The University of
Melbourne, (2004)

Novak, J. T., "Dewatering" in Sludge into biosolids: processing, disposal, utilization,


Spinosa, L. and Vesilind, P. A., IWA Publishing, (2001)

Novak, J. T., Agerbaek, M. L., Sorensen, B. L. and Hansen, J. A., "Conditioning,


filtering, and expressing waste activated sludge", J. Environ. Eng., 125 (9), 816
(1999)

Ødegaard, H., "Sludge minimization technologies - an overview", Wat. Sci. Tech., 49


(10), 31 (2004)

Olson, R. E., "State of the art: consolidation testing" in Consolidation of soils: testing
and evaluation, ASTM STP 892, Yong, R. N. and Townsend, F. C., American Society
for Testing and Materials, Philadelphia (1986)

Ormeci, B. and Vesilind, P. A., "Development of an improved synthetic sludge: a


possible surrogate for studying activated sludge dewatering characteristics", Wat.
Res., 34 (4), 1069 (2000)

Pan, J. R., Huang, C., Cherng, M., Li, K.-C. and Lin, C.-F., "Correlation between
dewatering index and dewatering performance of three mechanical dewatering
devices", Adv. Env. Res., 7 (3), 599 (2003)

Parkin, A. K., "Coefficient of consolidation by the velocity method", Geotechnique,


28 (4), 472 (1978)

455
Peatling, S., "To quench our thirst it will cost 255,500 tonnes of greenhouse gases",
Sydney Morning Herald, 1 November, 2004, 1

Poiseuille, J. L. M., Compt. Rend., 11, 961 (1840)

Priestley, A. J., "Report on sewage sludge treatment and disposal: Environmental


problems and research needs from an Australian perspective", CSIRO, Division of
Chemicals and Polymers1998)

Records, F. A., "The continuous scroll discharge decanting centrifuge", The Chemical
Engineer, (281), 41 (1974)

Reif, F. and Stahl, W., "Transportation of moist solids in decanter centrifuges", Chem.
Eng. Progr., 85 (11), 57 (1989)

Reif, F., Stahl, W. and Langeloh, T., "Optimising decanter centrifuges", Filtr. Sep., 27
(6), (1990)

Ruth, B. F., "Correlating filtration theory with industrial practice", Ind. Eng. Chem.,
38 (6), 564 (1946)

Sanin, F. D. and Vesilind, P. A., "Synthetic sludge: A physical/chemical model in


understanding bioflocculation", Wat. Env. Res., 68 (5), 927 (1996)

Sanin, F. D. and Vesilind, P. A., "A comparison of physical properties of synthetic


sludge with activated sludge", Wat. Env. Res., 71 (2), 191 (1999)

Scales, P. J., de Kretser, R. G., Stickland, A. D. and Usher, S. P., "Filtration process
modelling and optimisation using fundamental theory", Proceedings of Filtech
Europa, Dusseldorf, 11 (2001)

Scales, P. J., Dixon, D. R., Harbour, P. J. and Stickland, A. D., "The fundamentals of
waste water sludge characterization and filtration", Proceedings of the IWA Specialist

456
Conference: BIOSOLIDS 2003, Wastewater Sludge as a Resource, Trondheim,
Norway, 97 (2003)

Scarlett, B., Institute of Chemical Engineers Symposium on Filtration, London,


(1968)

Schiffman, R. L., Pane, V. and Gibson, R. E., "The theory of one-dimensional


consolidation of saturated clays: IV. An overview of nonlinear finite strain
sedimentation and consolidation", Sedimentation Consolidation Models: Predictions
and Validation: Proceedings of a Symposium / Sponsored by the ASCE Geotechnical
Engineering Division in Conjunction with the ASCE Convention in San Francisco,
California, 1 (1984)

SDWA, "The Safe Drinking Water Act", (Public Law 93-523), (1974)

Skrypsi-Mäntele, S., Bridle, T. R., Freeman, P., Luceks, A. and Ye, P. D., "Production
of high quality fuels using the enhanced EnersludgeTM process", Wat. Sci. Tech., 41
(8), 45 (2000)

Smiles, D. E., "Water flow in filter paper and capillary suction time", Chem. Eng. Sci.,
53 (12), 2211 (1998)

Smith, G. D., Numerical solution of partial differential equations: finite difference


methods (3rd ed), Oxford University Press, Oxford (1985)

Sobeck, D. C. and Higgins, M. J., "Examination of three theories for mechanisms of


cation-induced bioflocculation", Wat. Res., 36 (3), 527 (2002)

Sponza, D. T., "Investigation of extracellular polymer substances (EPS) and


physicochemical properties of different activated sludge flocs under steady-state
conditions", Enzyme Microb. Technol., 32 (3-4), 375 (2003)

457
Sridharan, A. and Prakash, K., "δ-t/δ method for the determination of the coefficient
of consolidation", Geotech. Test. J., 16 (1), 131 (1993)

Sridharan, A. and Sreepada Rao, A., "Rectangular hyperbola fitting method for one
dimensional consolidation", Geotech. Test. J., 4 (4), 161 (1981)

Stentiford, E. I., "Aerobic digestion" in Sludge into biosolids: processing, disposal,


utilization, Spinosa, L. and Vesilind, P. A., IWA Publishing, London (2001)

Stokes, G. C., Trans. Cambridge Philos. Soc., 8, 287 (1845)

Taylor, D. W., Research on consolidation of clays, Massachusetts Institute of


Technology, Dept. of Civil and Sanitary Engineering, Serial no. 82, (1942)

Terzaghi, K. and Peck, R. B., Soil Mechanics in Engineering Practice (2nd ed), John
Wiley & Sons, New York (1967)

Tiller, F. M. and Hsyung, N. B., "Unifying the theory of thickening, filtration, and
centrifugation", Wat. Sci. Tech., 28 (1), 1 (1993)

Tiller, F. M. and Khatib, Z., "Theory of sediment volumes of compressible,


particulate structures", J. Colloid Interface Sci., 100 (1), 55 (1984)

Tiller, F. M. and Kwon, J. H., "Role of porosity in filtration: XIII. Behavior of highly
compactible cakes", AIChE J., 44 (10), 2159 (1998)

Tiller, F. M. and Leu, W.-F., "Basic data fitting in filtration", J. Chin. Inst. Chem.
Eng., 11 (2), 61 (1980)

Tiller, F. M. and Li, W. P., "Strange behaviour of super-compactible filter cakes",


Chem. Process., 63 (9), 49 (2000)

458
Tiller, F. M., Li, W. P. and Lee, J. B., "Determination of the critical pressure drop for
filtration of super-compactible cakes", Wat. Sci. Tech., 44 (10), 171 (2001)

Tiller, F. M. and Shirato, M., "The role of porosity in filtration: VI. New definition of
filtration resistance", AIChE J., 10 (1), 61 (1964)

UN, "Report of the World Summit on Sustainable Development", Report No.:


A/CONF.199/20*, United Nations, 26 August - 4 September (2002)

Usher, S. P., PhD Thesis: Suspension dewatering - characterisation and optimisation,


The Department of Chemical and Biomolecular Engineering, The University of
Melbourne, (2002)

Usher, S. P., de Kretser, R. G. and Scales, P. J., "Validation of a new filtration


technique for dewaterability characterization", AIChE J., 47 (7), 1561 (2001)

Vesilind, P. A., "Capillary suction time as a fundamental measure of sludge


dewaterability", J. Wat. Pollut. Control Fed., 60, 215 (1988)

Vesilind, P. A., "The role of water in sludge dewatering", Wat. Env. Res., 66 (1), 4
(1994)

Vesilind, P. A. and Ormeci, B., "A modified capillary suction time apparatus for
measuring the filterability of super-flocculated sludges", Wat. Sci. Tech., 9, 135
(2000)

Warden, J. H., "Sludge treatment plant for waterworks", Report No.: TR 189, Water
Research Council, March (1983)

WHO, Guidelines for Drinking-Water Quality (3rd ed ed), World Health


Organisation, Geneva, Switzerland, (2004)

Windholz, M., The Merck index: an encyclopedia of chemicals, drugs, and


biologicals (10th ed), Merck, Rahway, N.J. (1983)

459
Wu, C.-C., Huang, C. and Lee, D. J., "Effects of polymer dosage on alum sludge
dewatering characteristics and physical properties", Colloids Surf. A, 122 (1-3), 89
(1997)

460
Minerva Access is the Institutional Repository of The University of Melbourne

Author/s:
Stickland, Anthony D.

Title:
Solid-liquid separation in the water and wastewater industries

Date:
2005

Citation:
Stickland, A. D. (2005). Solid-liquid separation in the water and wastewater industries. PhD
thesis, Faculty of Engineering, Chemical and Biomolecular Engineering, The University of
Melbourne.

Publication Status:
Unpublished

Persistent Link:
http://hdl.handle.net/11343/35270

File Description:
Solid-liquid separation in the water and wastewater industries

Terms and Conditions:


Terms and Conditions: Copyright in works deposited in Minerva Access is retained by the
copyright owner. The work may not be altered without permission from the copyright owner.
Readers may only download, print and save electronic copies of whole works for their own
personal non-commercial use. Any use that exceeds these limits requires permission from
the copyright owner. Attribution is essential when quoting or paraphrasing from these works.

You might also like