Mathematical Modelling of Granular Filters and Constriction-Based

University of Wollongong
Research Online
University of Wollongong Thesis Collection University of Wollongong Thesis Collections
2006
Mathematical modelling of granular filters and

constriction-based filter design criteria
Ashok K. Raut
Recommended Citation
Raut, Ashok K, Mathematical modelling of granular filters and constriction-based filter design criteria, PhD, Department of Civil
Engineering, University of Wollongong, 2006. http://ro.uow.edu.au/theses/44
Research Online is the open access institutional repository for the

University of Wollongong. For further information contact the UOW
Library: research-pubs@uow.edu.au
Mathematical Modelling of Granular Filters and
Constriction-based Filter Design Criteria
A thesis submitted in fulfilment of the
requirements for the award of the degree
Doctor of Philosophy
from the
Ashok Kumar Raut
BSc Eng (Distinction), M.Tech. (IIT Delhi)
Department of Civil Engineering
2006
DECLARATION
I, Ashok Kumar Raut, declare that this thesis, submitted in fulfilment of the requirements
for the award of Doctor of Philosophy, in the Department of Civil Engineering, University
of Wollongong, is wholly my own work unless otherwise referenced or acknowledged. The
document has not been submitted for qualifications at any other academic institution.
-------------------------
Ashok Kumar Raut
August 2006
ABSTRACT
In geotechnical engineering, a filter is designed to protect soils, called base, within or
behind a structure from erosion due to seepage. As water flows through the soil, fine
particles can be washed out, leading to internal erosion and eventually, the failure of the
structure. A correctly designed filter retains loose soil particles, thus preventing erosion. In
contrast, it allows unobstructed seepage, preventing build-up of detrimental pore pressure.
A filter is commonly a natural or manufactured sand and gravel. Filters are used in dams,
agricultural drainage, road pavements, retaining walls, canal linings, coastal protection,
landfills and so on.
First time in early 1920s, Terzaghi suggested two filter design criteria through laboratory
investigations on uniform sands. These criteria involved ratios of some specific sizes of
filter and base materials. Most subsequent studies ended up either merely investigating
validity of or extending these criteria to other soil types. Current professional guidelines are
still empirical and based on particle sizes. However, within filters, it is pores (i.e.
constrictions) that govern filtration. Consequently, these particle-based guidelines exhibit
some serious limitations.
In this thesis, limitations of current professional guidelines are comprehensively discussed.
A detailed mathematical procedure is developed to determine constriction sizes, and
subsequently, constriction-based criteria are proposed to describe filter effectiveness in
various types of soils. The proposed criteria are verified using several large-scale tests
carried out at the University of Wollongong including several test data available in the
iii
literature. Finally, an enhanced filter design guideline is suggested for the professional
practice.
iv
ACKNOWLEDGEMENTS
I am extremely grateful to my supervisor, Professor Buddhima Indraratna for his
enthusiastic guidance, invaluable help and encouragement in all aspects of this research
study. His comments, criticisms and suggestions in producing this research thesis on time
are gratefully acknowledged. His patience and availability despite his heavy workload are
greatly appreciated. Among his many strengths, his quick reading ability and do-it-now
attitude was inspiring to me.
I am equally grateful to my co-supervisor, Dr Hadi Khabbaz. His constant help and
encouragement has been my motivation to finish this research study on time. His comments,
criticisms and suggestions are greatly appreciated. I must mention that his ability to keep a
cool head is enviable indeed.
I wish to extend my sincere thanks to all technicians, particularly Mr. Alan Grant and Mr.
Ian Laird. Their continuous support and help in maintaining and operating the laboratory
equipments is acknowledged.
I would like to express my sincere thanks to my fellow postgraduate students in
Engineering, particularly Dr. Cholachat, Mr. Jayanathan and Mrs Linda for their
discussions, support and social interaction during this study. I would like to thank all
academic and non-academic members of Civil, Mining and Environmental Engineering for
their warm-hearted friendship, help and co-operation during my stay at the University of
Wollongong. I also wish to thank the University of Wollongong for providing me the
v
research scholarship for this study. My special thanks go to the academics, Dr Brett Lemass,
A/Prof. Michael Boyd, A/Prof. M. Sivakumar and A/Prof. M. Hadi for their special
supports and encouragement throughout the study.
I would like to thank Dr Mark Locke, Senior Geotechnical Engineer, SMEC Victoria. His
early guidance and suggestions helped me to start this research study in right direction. I
would also like to express my sincere thanks to my friends and relatives Dr. Hom Murti
Panta and his wife Mrs Ambika Panta, Dr. Keshav Mani Bhattrai, and his wife Mrs Punam
Panta and Mr. Abadhesh Chandra Jha for their constant support and encouragement.
Finally, I wish to express my heartfelt gratitude to my family (my wife Gita, our sons
Anjuman and Anshuman, and my brother Yogi). Without their sacrifices, guidance and
support, I would have never thought of this study. Gita has been my support and inspiration
through out this study. Even during some testing time periods, she could maintain her
energy level and kept me motivated in this research study. Her philosophy-prevention is
better than cure- kept all of us healthy all through these three years.
Thank you all.
“You cannot produce a baby in one month by impregnating nine women.”
(Murphy’s Law)
vi
LIST OF PUBLICATIONS
Refereed Journals
Raut, A. K. and Indraratna, B. (2007). “Analytical Approach to Geotextile Filter
Design”, Journal of Geotechnical and Geoenvironmental Engineering, ASCE (To be
submitted after review of the third paper)
Raut, A. K. and Indraratna, B. (2007). “Constriction-based Analysis of Internal Stability
of Granular Soils”, Journal of Geotechnical and Geoenvironmental Engineering, ASCE
(To be submitted after review of the third paper)
Raut, A. K. and Indraratna, B. (2006). “Limitations of Particle Size Approach in Filter
Design and Advancement through Constriction-Based Techniques”, Journal of
Geotechnical and Geoenvironmental Engineering, ASCE (Under review)
Indraratna, B., Raut, A.K. and Khabbaz, H. (2007). “Constriction-based Retention
Criterion for Granular Filter Design.” Journal of Geotechnical and Geoenvironmental
Engineering, ASCE (To be published in March 2007)
Indraratna, B. and Raut, A. K. (2006). “Enhanced Criterion for Base Soil Retention in
Embankment Dams.” Journal of Geotechnical and Geoenvironmental Engineering,
ASCE, Vol 132, No.12, pp.1621-1627
vii
Keynote Paper
Indraratna, B., Raut, A.K. and Locke, M. (2004). “Granular Filters in Embankment
Dams: A Conceptual Overview and Experimental Investigation.” ASCE/Geo-Institute
Proc. International Conference Geo2004: Advances in Geotechnical Engineering with
Emphasis on Dam Engineering, Irbid, Jordan, pp. 15-34
Refereed Conferences
Indraratna, B. and Raut, A.K. (2004). “Mathematical modelling of granular filters in
embankment dams.” Proc. 4th International Conference on Filters and Drainage in
Geotechnical and Environmental Engineering, Geofilters 2004, Johannesburg, SA,
pp.75-81
Raut, A.K. and Indraratna, B.(2004). ‘Constriction Size Distribution in a Granular
Filter’ Proc. 15th South East Asia Geotechnical Conference, Bangkok Thailand, pp.
409-414
viii
TABLE OF CONTENTS
ABSTRACT iii
ACKNOWLEDGEMENTS v
LIST OF PUBLICATIONS vii
LIST OF FIGURES xvi
LIST OF TABLES xxii
1 INTRODUCTION 1
1.1 Granular Filters 1
1.2 Locations of Filters in Embankment Dams 2
1.3 Erosion Mechanisms and Properties of Granular Filters 4
1.4 Factors Affecting Filtration 6
1.5 Effective and Ineffective Filters 8
1.6 Relevance of Study 11
1.7 Research Aims 13
1.8 Scope and Organization of the Thesis 14
2 LITERATURE REVIEW 18
2.1 Introduction 18
2.2 Empirical Investigations 20
2.2.1 Conventional Experimental Approaches in Filtration 20
2.2.2 No Erosion Filter Test (NEF Test) 23
2.2.3 Particle Size Ratio 26
ix
2.2.4 Broadly-Graded Base Soils and Filter Materials 30
2.2.5 Internally Unstable Base Soils 35
2.2.6 Particle Size -Permeability Relationship 37
2.2.7 Cohesive Base Soils 39
2.3 Analytical Investigations 45
2.3.1 Numerical Models 46
2.3.2 Filter Void Models 46
2.3.3 Constriction Size 50
2.3.4 Particle Infiltration Models 56
2.3.5 Particle Transport Models 59
2.4 Current Filter Design Guidelines 65
2.4.1 Lafleur Filter Design Procedure 65
2.4.2 NRCS Filter Design Procedure 66
2.5 Critical Aspects of Existing Design Practice and Research Focus 68
2.5.1 Factor of Safety in Filter Designs 69
2.5.2 Filter Effectiveness and Base Soil Grading 70
2.5.3 Representative Filter Particle Size and Filter Grading 74
2.5.4 Controlling Constriction Size and Filter Compaction 76
2.5.5 Particle Frequency 76
2.5.6 Filtration of Cohesive Base Soils 77
2.5.7 Numerical Modelling of Particle Migration 78
x
3 MATHMATICAL MODELLING OF FILTERS 80
3.1 Introduction 80
3.2 Constriction Sizes in the Densest Particle Arrangements 81
3.3 Constriction Sizes in the Loosest Particle Arrangements 84
3.4 Particle Frequency and Filter Compaction 86
3.5 Programming Algorithms 89
3.5.1 Number of Unique Groups Constituting Constrictions 90
3.5.2 Constriction Size Computation 92
3.5.3 Sorting, Cumulating and Interpolating 93
3.6 Worked-Out Example 94
3.7 CSD Program Calibration 99
3.7.1 CSDs of Uniform and Well-graded Filters 99
3.7.2 Effect of Relative Density on Constriction Size 99
3.7.3 Manual CSD Computations in Past Studies 100
3.7.4 Experimental Observations 103
3.8 Controlling Constriction Size 104
3.8.1 Analytical Concepts 105
3.8.2 Experimental Verification 109
3.9 Filter Thickness 114
3.10 Summary of Constriction Modelling 115
xi
4 SURFACE AREA CONCEPT APPLIED TO BASE SOILS 117
4.1 Introduction 117
4.2 Modelling of Base Soils 118
4.3 Development of Filter Retention Criterion 122
4.4 Verification of the Model Based on Experimental Data 123
4.4.1 Series A: Very Uniform Base Soils and Filters 123
4.4.2 Series B: Moderately –graded Base Soil and Filters 126
4.4.3 Series C: Well-graded Base Soil and Uniform Filters
128
4.4.4 Series D: Well-graded Base Soils and Filters 128
4.5 Comparison with Existing Retention Criteria 130
4.5.1 Terzaghi Criterion 130
4.5.2 Current Design Practice 134
4.6 Summary and Conclusions 135
5 STABILITY OF SELF-FILTRATION LAYER 137
5.2 PSD of Self-Filtration Layer 138
5.3 Stability of Self-filtration Layer 142
5.4 Model Verification 144
5.5 Comparison with Existing Criteria 148
5.5.1 Terzaghi Method 148
5.5.2 Current Design Implication 149
xii
5.6 Summary 151
6 SELF-FILTERING BASE FRACTION AND FILTER DESIGN 153
6.2 Model Development 155
6.3 Model Procedure Illustration 157
6.4 Model Verification 161
6.5 Model Comparison with Current Professional Practices 168
7 DISCUSSION AND ANALYSIS OF FILTER CRITERIA 177
7.2 Scope of Application of Constriction-based Design Criteria 178
7.2.1 Surface Area Filter Criterion 178
7.2.2 Self-filtration Stability Filter Criterion 180
7.2.3 Modified Base PSD Filter Criterion 183
7.3 Analysis of Controversial USBR data 184
7.4 Filtration of Gap-graded Base Soils 187
7.5 Laboratory Study of Filtration 189
7.5.1 Laboratory Approach and Failure Criteria 189
7.5.2 Laboratory Observations 192
xiii
8 ENHANCED DESIGN GUIDELINE AND ITS APPLICATIONS 196
8.2 New Filter Design Procedure 198
8.3 Practical Application 202
8.3.1 Design of Non-Reactive Barrier 203
8.3.2 Design of Reactive Barrier 206
8.4 Summary 209
9 CONCLUSIONS AND RECOMMENDATIONS 210
9.1 General Summary 210
9.2 Specific Observations 212
9.2.1 Constriction Analyses 212
9.2.2 Surface Area Concept Applied to Base Soils 214
9.2.3 Stability of Self-filtration Layer 215
9.2.4 Self-filtrating Base Fraction and Constriction-based Filter Criterion
216
9.2.5 Experimental Observations 217
9.2.6 Controversial USBR Observations 218
9.2.7 Enhanced Filter Design Guideline 218
9.3 Future Filter Research 219
9.3.1 Re-evaluation of Critical Filters in Existing Dams 219
9.3.2 Evaluation of Crack Susceptibility of Critical Filters in Existing
Dams 219
xiv
9.3.3 Clogging 220
9.3.4 Internal Stability Assessment Method 221
9.3.5 Constriction Size and Permeability 221
9.3.6 Geotextile Filters 221
9.3.7 Cyclic Behaviour of Granular Filters 222
REFERENCES 223
xv
LIST OF FIGURES
Figure 1.1 Location of filters within a typical embankment dam 4
Figure 1.2 An embankment dam being rehabilitated with a downstream filter 5
Figure 1.3 A stable base-filter interface during seepage (after Indraratna and Locke 2000) 6
Figure 1.4 Possible Outcomes of Filtration (Indraratna and Locke 2000) 11
Figure 2.1 Typical Laboratory Apparatus (Locke 2001) 22
Figure 2.2 NEF Test Apparatus (Sherard and Dunnigan, 1985) 24
Figure 2.3 Base particle sizes vs. filter permeability relationship (Indraratna et al. 1996)
39
Figure 2.4Concentrated leak through a crack in a cohesive core (after Sherard et al. 1984b).
40
Figure 2.5 One-dimensional multi-layered void network filter model (Kenney et al.1985)
48
Figure 2.6 Void Channel Model (Indraratna and Vafai 1997) 49
Figure 2.7 Pore Network Model (Witt 1993) 50
Figure 2.8 Cubic Pore Network Model (Schuler 1996) 50
Figure 2.9 Particle packing arrangement for a) dense and b) loose states (Indraratna and
Locke 2000) 51
Figure 2.10 Comparison of Filter Constriction Model (DeMello 1977) 53
Figure 2.11 Model for geometrical structure of pore constrictions (Schuler 1996) 54
xvi
Figure 2.12 Equilibrium of particle “plugging” a vertical pore channel in a) XZ plane; b)
YZ plane (Indraratna and Vafai, 1997) 62
Figure 2.13 Illustration of (a) base and filter elements and (b) generalised slurry flow
through a filter element (Indraratna and Vafai, 1997) 63
Figure 2.14 Numerical simulation of time-dependent change in filter PSD in an effective
filter (Indraratna and Vafai, 1997) 64
Figure 2.15 Numerical and laboratory simulation of particle capture with depth within a
granular filter (Indraratna and Locke 2000) 64
Figure 2.16 Filter design procedure-Lafleur method (ICOLD 1994) 66
Figure 2.17 Three base soils with identical d85 but different Cu values 72
Figure 2.18 Rock slope protection designed to prevent undermining (Cedergren 1969)
75
Figure 2.19 A base soil tested against two filters with identical D15 but different Cu values
77
Figure 3.1 The densest filter particle arrangement 83
Figure 3.2 A typical filter particle size distribution (PSD) and constriction size distribution
(CSD) showing passing probability 83
Figure 3.3 The loosest filter particle arrangement 85
Figure 3.4 Particle arrangements for α to be (a) minimum and (b) maximum 86
Figure 3.5 Flow chart for the detailed CSD computational procedure 91
Figure 3.6 Particle size distribution of a filter with 10 particle class sizes 95
xvii
Figure 3.7 A typical input data file XINPUT.DAT 96
Figure 3.8 A typical output data file XOUTPUT.DAT 97
Figure 3.9 A typical CSD program outputs for a uniform filter PSD in graphical form
98
Figure 3.10 Filter CSDs computed with the same filter PSD but with two different
descretizations (Rd = 90 %) 98
Figure 3.11 Filter PSDs and CSDs by mass (M), by number (N) and by surface area (SA)
(a) Uniform Filter, F1 (Cu=1.2, Rd=70%) (b) Non-uniform Filter, F2 (Cu=3.8,
Rd=70%) 101
Figure 3.12 CSDs of a uniform filter at various levels of compaction i.e. at different relative
densities, Rd 102
Figure 3.13 CSDs by mass of a filter in the densest state computed by using the computer
program and manually by Silveira (1965) 102
Figure 3.14 CSDs by mass of a filter in the densest state computed by using the computer
program and manually by Soria (1993) 104
Figure 3.15 Model predictions of filter constrictions against a number of experimental and
analytical results 105
Figure 3.16 Probability of forward movement and predicted depth of infiltration 109
Figure 3.17 Controlling constrictions in filters with varying uniformity coefficients (Cu)
and D15 sizes (a) Filter PSDs and (b) Filter CSDs 111
Figure 3.18 Filters used by Sherard et al. (1984a) for filtration of sand and gravel 113
xviii
Figure 3.19 Comparison between Dc35 of the current model and the upper bound of the base
particles eroded through filters as observed by Sherard et al. (1984a) 114
Figure 4.1 Base soils and filters with various uniformity coefficients (Cu) but having the
same retention ratio (D15/d85) 120
Figure 4.2 PSDSA of base soils of different uniformity coefficients 121
Figure 4.3 PSDSA of well-graded base soil B-3 121
Figure 4.4 Series A: Analysis of very uniform filters and base soil of parallel gradations (a)
PSDs of filters and base soil (b) Filter CSDs and PSDSA of base soil 125
Figure 4.5 Series A: Analysis of very uniform base soil and filters (Rd=50%) (a) PSDs of
base soil and filters (b) Filter CSDs and PSDSA of base soil 127
Figure 4.6 Series B: Analysis of moderately-graded base soil and filters (a) PSDs of base
soil and filters (b) Filter CSDs and PSDSA of base soil 129
Figure 4.7 Application of retention criteria to distinguish between effective and ineffective
filters (a) Terzaghi criterion (USACE 1953), and (b) current constriction model
133
Figure 4.8 Application of retention criteria to distinguish between effective and ineffective
filters using the current design practice with regraded base soil PSDs (NRCS 1994)
135
Figure 5.1 Dominant constrictions in various types of filters139
Figure 5.2 PSD of self-filtration layer in a typical base soil-filter combination 141
Figure 5.3 PSDs of self-filtration layers in progressively coarser filters 143
xix
Figure 5.4 Kenney and Lau (1985) procedure for internal stability assessment 143
Figure 5.5 (a) Analysis of effective uniform filter F1 with a uniform base soil and (b) an
ineffective uniform filter F2 with a uniform base soil 145
Figure 5.6 (a) Analysis of an ineffective uniform filter F1 and (b) an effective well-graded
filter F2 with a well-graded base soil 147
Figure 5.7 Comparative analysis of test results using the current model (a) Tests #1-8 (b)
Tests #9-13 (c) Tests #14-19 and (d) Tests #20-27 149
Figure 6.1 (a) Analysis of effective uniform filter F1 with a uniform base soil, and (b) an
ineffective uniform filter F2 with a uniform base soil 158
Figure 6.2 (a) Analysis of an ineffective uniform filter F1 and (b) an effective well-graded
filter F2 with a well-graded base soil 160
Figure 6.3 Comparative analyses of test results using the original Terzaghi retention
criterion (i.e. before regrading of the base soil) 162
Figure 6.4 Comparative analyses of test results using the current model 163
Figure 7.1 Linearly-graded soil having PSD with H/F=1 181
Figure 7.2 Analysis of USBR test data using the current design criteria 186
Figure 7.3 Analysis of an internally unstable gap-graded base soil 188
Figure 7.4 Large-scale filter permeameter 190
Figure 7.5 Small-scale high pressure permeameter 190
Figure 7.6 Discharge and turbidity pattern in effective tests 191
xx
Figure 7.7 Discharge and turbidity pattern in ineffective tests 192
Figure 7.8 Filter and base materials used in filtration of non-cohesive tests 193
Figure 7.9 Filter and base materials used in filtration of non-cohesive tests 194
Figure 8.1 Grading of acidity-affected soil near Bomaderry NSW within study area
204
Figure 8.2 Design of non-reactive barrier of the Permeable Reactive Barrier 204
Figure 8.3 Constriction Analysis of NRB F1 using Modified Base PSD method 205
Figure 8.4 Constriction Analysis of NRB F2 using Modified Base PSD method 205
Figure 8.5 PSDs of effective NRB F1 and potential reactive barriers F3 and F4 207
Figure 8.6 Constriction Analysis of RB F3 using Modified Base PSD method 207
Figure 8.7 Constriction Analysis of RB F4 using Modified Base PSD method 208
Figure 8.8 Gradings of suitable reactive and non-reactive barrier materials 208
xxi
LIST OF TABLES
Table 2.1 Summary of some well-known empirical filter design criteria 25
Table 2.2 Base soil category based on fines content (NRCS 1994) 68
Table 2.3 Design criteria – Maximum D15 68
Table 3.1 Filter PSD with eleven data points 96
Table 3.2 PSD of the same filter as given in Figure 3.6 but with different descretizations
97
Table 3.3 Controlling constrictions by past procedures and the current model 112
Table 4.1 Filter and base soil parameters 124
Table 4.2 Description of base and filter materials for various laboratory tests 132
Table 6.1 Description of base and filter materials for various laboratory tests 164
Table 6.2 Comparative study of existing design criteria with the current model 171
Table 7.1 Details of USBR test data including their filtration analysis 187
Table 8.1 Base soil category based on fines content 199
Table 8. 2 Design criteria – Maximum D15 199
Table 8.3 Design criteria – Minimum D15 200
Table 8. 4 Maximum and minimum particle size criteria 201
Table 8. 5 Segregation criteria 201
xxii
CHAPTER
ONE
INTRODUCTION
1.1 Granular Filters
In geotechnical engineering, a filter is a structural arrangement that is designed to protect
soils from erosion due to seepage. As water flows through a soil, fine particles can be
washed out, leading to internal erosion (i.e. piping) and eventually, the failure of the
structure. A correctly designed filter retain loose soil particles thus preventing piping, while
allowing seepage water to flow and avoiding the development of high internal pore
pressures. Filters are used in embankment dams, agriculture drainage works, road
pavements, retaining walls, canal linings, coastal protection works, landfills and so on. In
the current engineering practice, there are two types of filters: granular and geotextile. A
granular filter is typically well-graded sand or sandy gravel. This has been in practice for
almost a century now. In contrast, a geotextile is a synthetic textile material, a relatively
Chapter 1: Introduction 1
recent technological development, and has been used in professional practice for the past
thirty years.
There has been an increasing interest to replace granular filters with geotextiles because of
cheaper installation and better quality assurance. Some embankment dams have been built
with geotextile filters, and nearly thirty years later these dams are still performing well
(Faure et al. 1996). However, there is still a concern regarding the long-term performance
of geotextiles, given that a major dam usually has a design life over 100 years. Because of
these concerns, granular filters are still a preferred choice especially in important structures
such as embankment dams, where failure consequence is often enormous. This study will
focus solely on the performance of granular filters with a special emphasis on its use in
embankment dams. Hereafter, a granular filter is simply called a filter.
1.2 Locations of Filters in Embankment Dams
Filters are necessary in several locations within an embankment dam (Figure 1.1). The most
critical filter is the one placed immediately downstream of the dam core. This filter must
minimise internal erosion by retaining particles eroded from the core, while controlling the
seepage flow and forming a drainage layer to avoid saturation of the downstream
embankment. The critical filter is continued between the downstream fill and the
foundation to prevent particle movement either from the fill into the foundation or vice
versa, and transports seepage safely to the downstream toe. Core erosion can lead to rapid
failure of the dam by piping; hence, it is essential that this filter performs its function
correctly.
Within an embankment dam, filters are also often used beneath the upstream riprap to
protect fine fill from erosion due to wave action. A filter upstream of the core is required to
prevent erosion during drawdown of the reservoir. The positions of these filters are shown
in Figure 1.1. The focus of this study is the critical filter downstream of the core as
illustrated by Figure 1.2, where an old homogeneous embankment dam without any filter
has recently being repaired with a critical downstream filter, since the success of this filter
has been essential to avoid internal erosion of the dam core. The filter concepts developed
in this study apply equally to filters in other locations of dams and other geotechnical
structures, where effective drainage is an important requirement for their physical stability.
Rip-rap Self-filtration
(Gravel, Boulder)
Layer (Filter)
Filters
Drainage
Layer
(Coarse Filter)
Coarse
Coarse Core Fill Collector Drain
Fill (Fine)
GL
To Natural Source
Foundation
Figure 1.1 Location of filters within a typical embankment dam
1.3 Erosion Mechanisms and Properties of Granular Filters
Piping failure within a soil mass can result from the continuation of three fundamental
processes of particle migration causing erosion of the soil: namely backward erosion,
suffusion, and erosion through a crack. Backwards erosion is a process where seepage
forces remove particles from the exit face of the soil body. Continued erosion leads to a
backward movement (i.e. upstream) of the erosion face, as a tunnel is formed and
progresses towards the reservoir. This process occurs mainly in non-cohesive core
materials. Suffusion of internally unstable core materials involves the loss of fines from
within the structure of a soil. If a sufficient quantity of fines is eroded, then a volume
reduction of the core material may occur. Concentrated erosion requires a crack through the
core or an interface with a solid surface, such as a conduit or spillway structure. A high
velocity flow through the crack can result in erosion and transport of particles, and
subsequent enlargement of the crack. If the core material is cohesive it may be able to
maintain a roof over the crack, allowing extensive erosion. In this respect, effective filters
must possess the following properties.
Figure 1.2 An embankment dam being rehabilitated with a downstream filter
(Courtesy: SMEC-Victoria)
1. The filter must be fine enough to capture some of the larger particles of the protected core
material (also called the base soil) as shown in Figure 1.3. These captured particles block
the filter voids and subsequently retain the finer fraction of the base soil.
2. The filter must be coarse enough to allow seepage flow to pass through the filter, thus
preventing the build-up of high internal pore pressures and draining all the seepage water
from the dam, avoiding saturation of the downstream fill.
3. The filter must be non-cohesive, requiring a limit on the quantity of cohesive fines in the
filter. Cavities or cracks may form within the cohesive core soils. The protective filter
must have negligible cohesion so that it can collapse and self-heal over the crack when it
forms by any means such as hydraulic fracture, earthquake settlement, or drying of core
soils during prolonged droughts when reservoirs are operated at extremely low levels.
Base Filter
Soil
Seepage
Larger base soil
particles trapped in
Original small filter voids
Interface
Figure 1.3 A stable base-filter interface during seepage

(after Indraratna and Locke 2000)
1.4 Factors Affecting Filtration
Three properties of a filter, namely retention of particles, sufficient permeability to allow
seepage, and no cohesion, impose conflicting requirements on the filter particle size.
Designing a suitable filter to meet these functions requires knowledge of the factors
influencing base soil-filter interaction under seepage flows. The interaction of real granular
materials leads to a very complex process of particle migration in a porous medium, with
many factors involved in the process. These factors that can be broadly termed geometric,
physical, hydraulic, chemical and biological, explained briefly below.
• Geometric factors define the shape of particles and the particle size distribution of both
filter and base soil, and also the structure of the filter medium (pore constriction size and
distribution). In all empirical approaches, the particle size distribution of the base soil
and filter is considered by far the most important and hence, often the only factor
considered in the laboratory.
• Physical factors may include inter-particle friction and cohesion, particle surface
roughness, filter density, particle specific gravity etc. The seepage fluid physical
characteristics include viscosity, density and temperature effects. Consideration of
equilibrium of forces on a loose particle within a filter requires modelling of some of the
above physical properties (Indraratna and Vafai, 1997).
• Hydraulic factors include the applied total head, hydraulic gradients of seepage water
and the corresponding particle velocities and mass flow rates. Considering hydraulic
forces may allow a relaxation in filter design criteria (de Groot et al. 1993).
• Chemical factors that affect particle sizes (dispersion or flocculation) and flow
characteristics are associated with both water and soil chemistry. It is well known that
the reservoir water chemistry on limestone terrain has the beneficial effect of increased
floc sizes, leading to more economical filter design (Indraratna et al. 1996). Reddi and
Bonala (1997) have shown that pore fluid chemistry can influence the capture of clay
particles in very fine filters.
• Biological effects generally involve the change in porosity of the filter media due to
bacterial and fungal growth.
Because of a large number of parameters and complex interaction between them, it has
been necessary to simplify the problem and concentrate only on certain aspects in
modelling filter behaviour. In geotechnical engineering, most studies have considered the
particle size distribution curve of the base soils and filters as the most important parameters
(Sherard and Dunnigan 1985). Some studies have also included a simple model of the
hydraulic effects and physical properties (Indraratna and Vafai 1997; Indraratna and Locke
2000; Locke et al. 2001) in order to model time-dependent filter behaviours. Hydraulic
effects are often ignored because it is conservative to assume that the hydraulic force is
sufficiently high to mobilize particles that are not stopped by geometric constraints.
Irrespective of magnitude of hydrodynamic forces, a base particle is always captured by a
smaller void constriction. This study is founded on this fundamental premise, considering
the severest hydraulic conditions. For this reason, although hydraulic gradient and some
physical parameters of filters and base soils are important in time-dependent modelling,
final filtration results are largely influenced by geometric parameters of filters and base
soils.
1.5 Effective and Ineffective Filters
To demonstrate how a filter functions, Figure 1.3 shows a stable base soil-filter interface.
Seepage forces wash out some base soil particles into the filter, initiating the backwards
erosion process. Initially, some fine base particles may be transported completely through
the filter, but in an effective filter the larger base particles are trapped by the pore
constrictions (i.e. smallest connections between pores) of the filter material. These trapped
particles then form smaller pore constrictions, subsequently retaining smaller base particles,
and as this process continues the entire interface becomes stable without any significant
loss of base soil. After formation of a stable interface, no further particle loss occurs. This
process is called “self-filtration”. The water flow rate varies during the process, but
generally reaches a steady state as the interface is stabilised.
Filtration is controlled by the pore constrictions within the filter. A pore constriction is the
smallest, two-dimensional opening between two filter pores (i.e. void spaces). These pore
constrictions are smaller than the filter pores and are responsible for preventing the
movement of loose base particles. If a particle is smaller than all of the constrictions exiting
a pore, then it is retained within that pore, irrespective of magnitude of hydrodynamic
forces. The possible outcomes of filtration process depend on the relative size of base
particles and can be grouped into four cases shown in Figure 1.4, which are briefly
described below.
A. The pore constrictions are finer than all the base particles, and no particles penetrate the
filter. This case is excessively conservative, and often leads to insufficient permeability to
allow drainage of seepage water as the base and filter material have almost the same
particle sizes.
B. This case describes a stable base soil-filter combination. Some base material penetrates
the filter, but these particles eventually encounter a smaller pore constriction and are
captured. These retained particles form smaller pores that are able to capture further base
particles, hence a stable interface results. The formation of a stable interface is often
accompanied by a decreasing flow rate that becomes constant with time. Though
effective, this case may still be a conservative option.
C. The figure shows another base soil-filter combination, which may be stable or unstable
depending on the amount of base soil loss through the filter before the interface stabilizes.
Here, fine base particles are initially able to wash through the filter without being
impeded, while coarser particles are retained. The retained particles block the pores and
form a self-filtering interface. The degree of fine particle washout defines the failure or
success of this case. If an excessive amount of fines are eroded, and the base permeability
increases and subsidence may occur, this case may be unacceptable.
D. This is an unstable base-filter combination, where the filter is too coarse to retain any
base particles; hence, the formation of a stable interface is not possible.
A filter in a stable base soil-filter combination is called an effective filter, because the filter is
able to prevent erosion of the base soil. In contrast, a filter in an unstable base soil-filter
combination is called an ineffective filter. An effective filter lies somewhere between the case
B and the case C, depending on the base material to be protected. The case B is safe, but may
be somewhat conservative. The case C may be acceptable, but it is necessary to have detailed
knowledge of the filtration process and properties of the materials before designing a filter
near the success-failure boundary. In many applications, it may be acceptable to allow some
loss of fine material before a stable interface can be established.
Figure 1.4 Possible Outcomes of Filtration (Indraratna and Locke 2000)
1.6 Relevance of Study
The failure of dams by internal erosion and piping through inadequate filters is a real risk.
Foster et al. (1998) examined historical records of dam failures and determined that the
average probability of failure of large embankment dams is 1.2% over the life of the dams
(136 dam failures out of 11192 large embankment dams constructed up to 1986). Out of
these failures, 46% were due to piping, where the filters were either inadequate or not
present. Uncontrolled erosion of the fine core materials occurred either by seepage into the
foundation or through the dam structure, leading to the failure of the dam. The dam failures
are not limited only to older dams; 13 of the recorded piping failures occurred in the dams
constructed during or after 1980. In this respect, the development of an enhanced filter
design guideline based on the better understanding of filter behaviour still deserves a higher
priority.
The filter design criteria in practice are currently based on laboratory tests that were mostly
carried out on uniform base soil and filter materials. These criteria mostly involve specific
particle size ratios, where the system of base soil and filter is represented by some
characteristic particle sizes. Consequently, these criteria have some obvious limitations,
particularly when applied to non-uniform materials. Moreover, in filters, it is the
constriction sizes rather than the particle sizes that govern filtration (Locke et al. 2001;
Kenney et al. 1985). However, there is no straight-forward procedure developed so far to
determine the filter constriction sizes. Based on theoretical framework developed by Locke
et al (2001), this study presents an elaborated computational procedure to calculate the
constriction size distribution (CSD) of the filter. The whole procedure is coded into a
comprehensive computer program, which is calibrated against some of well-known
laboratory test data. An analytical procedure is developed to determine the controlling
constriction size of a filter, defined as the size of the largest base particle that can pass
through the filter. An analytical procedure is also established to determine the self-filtering
fraction of the base soil with respect to a given filter. Considering fundamental filtration
mechanisms such as suffusion, internal stability, self-filtration, mechanical sieve-granular
filter analogy and so on, alternative design criteria are proposed based on the filter
constriction sizes. The advantages and limitations of these criteria are discussed in relation
to existing design practice, and a new design guideline is recommended with a real-life
example of its application.
1.7 Research Aims
As discussed earlier, it is the constriction sizes that are the most important parameters in a
filter design. However, current design practices are mostly based on the particle sizes of
filter and base particles. Consequently, there are opportunities for improvement. The
primary aim of this research is to develop a filter design guideline that is based on the
constriction sizes of the filter, which addresses the limitations associated with current
design guidelines in a more rigorous manner. The major objectives of this research study
are outlined below.
• A critical review of literature and discussion of review findings for better understanding
of filtration mechanisms and identifying limitations of existing design guidelines;
• Development of an analytical computational procedure and a computer program to
determine the constriction size distribution of the filter and its controlling constriction
size;
• Development of an analytical method to determine the self-filtering fraction of the base
soil and a mathematical procedure to determine the amount of base soil retained in the
filter;
• Development of base soil retention criteria based on the filter constriction sizes and some
well-known filtration phenomena such as stability of self-filtration layer, self-filtering
fraction of base soils, and surface area concept applied to the base soils;
• Laboratory filter tests on various uniform and well-graded base and filter materials
particularly where existing guidelines exhibit limitations;
• An elaborated discussion of these criteria in relation to their advantages and limitations
over the prevailing design practices;
• Recommendation of a new design guideline based on the filter constriction sizes with a
real life worked-out example to demonstrate the scope of application and advantages over
the existing guideline.
1.8 Scope and Organization of the Thesis
Following this introduction, Chapter 2 presents a comprehensive review of the past
research studies related to the granular filters. The research findings are discussed in
relation to potential pitfalls associated with existing design practices, and the critical issues
related to this study is highlighted.
Chapter 3 is dedicated to “Mathematical Modelling of Filters”. The theoretical concepts of
the filter constriction model are described. This chapter is subdivided to describe the
development of the constriction model, concise algorithms for coding these concepts in a
comprehensive computer program, and program calibration with published research data,
and development of analytical method to determine controlling constriction size of the filter
and its validation with the results of some well-known laboratory investigations.
Chapter 4 on “Surface Area Concepts Applied to Base Soils” describes how the concept of
the surface area of a particle, which is used to describe a granular filter with advantage over
the use of mass of the particle, can also be applied to the base soil. The controlling
constriction size model developed in Chapter 3 when compared with a specific base particle
size based on surface area consideration provides a constriction-based filter retention
criterion to describe effective filters for the filtration of non-cohesive base soils. The model
is validated with well-known published research data including results of filter tests
conducted at the University of Wollongong during this research study. The model is also
compared with current design practice to discuss its relative scope and advantages.
Chapter 5 titled as “Stability of Self-filtration Layer” describes self-filtration mechanism
and examines the stability of self-filtration layer. It establishes an analytical procedure to
determine the self-filtering fraction of the base soil with respect to a given filter. Based on
fundamental principle of soil mechanics, a mathematical method is developed to measure
the mass proportion of base soil that is retained in the void spaces of self-filtering layer.
The base particles smaller than the largest effective constriction size are likely to be eroded.
It examines the stability of self-filtration layer that is formed by retention of potentially
erodible base particles into the filter. A stable base soil-filter combination forms a stable
self-filtration layer. In contrast, in the case of an unstable base soil-filter combination, the
stable self-filtration is not formed and consequently, erosion of base particles is not
prevented. In this respect, the assessment of stability of self-filtration layer gives rise to a
filter criterion to describe filter effectiveness in the case of non-cohesive base soils. The
model is verified by the same filter test data described earlier and is also compared with
current design guideline.
Chapter 6 presents the “Self-Filtering Base Fraction and Filter Design”, where theoretical
concepts of controlling constriction size of the filter and self-filtering fraction of the base
soil developed in Chapters 3 and 5 are compared to formulate an enhanced criterion to
describe filter effectiveness. The model is validated against almost a hundred sets of test
data taken from a number of well-known experimental investigations carried out in the past
including data obtained through laboratory tests carried out at the University of
Wollongong during this research study. The data sets include the tests carried out on wide
range of filter and base materials such as coarse and fine, and uniform and well-graded
filters with uniform, well-graded and broadly-graded cohesive (dispersive and non-
dispersive) and non-cohesive base soils.
Chapter 7 on “Analysis and Discussion of Filter Criteria” discusses the scope and
limitations of three filter design criteria proposed in Chapters 4, 5 and 6 in relation to the
broad range of the possible base soils. It also describes the experimental approach adopted
in order to carry out filter tests during this research study at the University of Wollongong
and discusses the results in relation to laboratory observations of various studies. Finally, it
illustrates how the current study can be successfully used to resolve controversies
associated with test results of the past studies.
Chapter 8 titled as “Enhanced Design Guideline and Its Applications” recommends an
enhanced guideline for the design of granular filters by incorporating constriction-based
retention criteria in the existing design guideline and illustrate its application with a real-
life design example of a Permeable Reactive Barrier (PRB) to be constructed as a
geotechnical measure on a trial basis to reclaim acid sulphate soil in Australia.
Chapter 9 presents the main research findings and conclusions of the research study with
the recommendation of an improved filter design guideline. It also presents some future
directions in this field.
CHAPTER
TWO
LITERATURE REVIEW
2.1 Introduction
Filters are an integral part of embankment dams, providing protection to the dam core.
The core provides an impermeable barrier, limiting seepage flows through the dam. A
filter downstream of the core is essential to protect the fine material from being eroded
by internal seepage. Sufficient evidence exists to show that concentrated leaks
commonly develop in the core of well designed and constructed dams (Sherard and
Dunnigan 1985). These leaks are usually attributed to cracking due to differential
settlement, earthquake movement, shrinkage or hydraulic fracturing. A suitable filter,
downstream of the dam core, is able to retain eroded particles from the core and seal
any concentrated leaks.
It was Terzaghi (1922) who first developed the granular filter design criteria and
proposed the following two conditions to be fulfilled to describe an effective filter.
Chapter 2: Literature Review 18

• The filter should be fine enough to prevent erosion of the base soils and prevent
piping. To ensure this, Terzaghi suggested a retention criterion given by:
D15/d85 ≤ 4-5 (2.1)
• The filter should be permeable enough to prevent excessive internal pore pressure
build-up in the dam. To ensure this, he suggested a permeability criterion given by:
D15/d15 ≥ 4 (2.2)
All through this thesis, the upper case D denotes the filter particle size and the lower
case d the base particle size, and the subscripts refer to the percentage of particles that
are finer than the size. In this way, the Terzaghi criteria impose two conflicting
requirements, suggesting that a suitable filter must be fine enough to retain the base soil
while being coarse enough to drain the seepage water.
While the Terzaghi design criteria are still used for some simplified filter designs,
extensive subsequent research has greatly improved knowledge in this field. A series of
laboratory observations on sets of base soil-filter combinations has usually led the
researchers to recommend empirical relationships for a stable combination (Sherard and
Dunnigan 1985). However, these empirical criteria can only be reliably applied to the
range of soils tested, and may have certain laboratory bias due to different testing
methods, definitions of failure etc. Most empirical criteria do not provide the designer a
confident picture of the process occurring within the dam, nor the level of safety
involved with design decisions. With increasing use of computers, many researchers are
now concentrating on more advanced numerical analysis of filtration, with particular
focus on particle movement through the filter.

This literature review is a summary and discussion of the outcomes of past studies in
this field. The review describes separately both the empirical and analytical
investigations. The final section identifies major outstanding issues associated with the
current state-of-the-art of filter design and introduces the main focus of the current
research.
2.2 Empirical Investigations
Extensive empirical research has been carried out in the last almost one hundred years
after the pioneer work of Terzaghi. This section describes the basic experimental
approaches adopted by most researchers. The results of important past empirical studies
are summarized to describe important factors affecting the filtration process. The factors
discussed here include: particle size ratios to determine design criteria; additional
requirements for broadly-graded materials; problems with internally unstable soils;
alternative criteria relating the base soil particle size to permeability; additional
problems with cohesive base soils; and various constriction sizes and their laboratory
measurement.
2.2.1 Conventional Experimental Approaches in Filtration
Experiments have typically been conducted using a vertical cylinder, called a
permeameter, similar to that shown in Figure 2.1. The cylinder is typically 150 to
300mm long and 100 to 250mm diameter. More recently, Indraratna and co-researchers
have used a much larger permeameter (500mm diameter and 1000mm high) in order to

minimise the effect of larger pores around the wall of the permeameter. This also
enables study of the extent of particle migration in the filter in a comprehensive manner,
especially in case of coarse, non-cohesive base and filter materials (Indraratna and
Locke 2000). Sherard et al. (1984a) describe a typical method for these standard tests.
In short, the experiment involves vertical flow of water through the base soil and filter.
Effluent water is collected to measure flow rates and the amount of base soil in the
effluent slurry. Often a range of hydraulic gradients are applied, varying from 0.5 to as
high as 50. This has been done to examine the effect of different water pressures and to
ensure more severe hydraulic conditions in the test than may exist within a real dam.
The apparatus is often vibrated or tapped with a rubber mallet, as this has been shown to
break up soil bridges that may form over the filter pores. Most researchers have
employed similar experimental methods; however, they have defined slightly different
criteria for success and failure in the tests. Generally, accepted failure criteria include:
• visual inspection - the base soil gradually sealing the filter or passing through the
filter;
• measurement of permeability changes throughout the test;
• measurement of the change of mass of both the soil and filter, which gives a
quantitative measurement of the movement of soil particles;
• determination of the particle size distribution difference before and after the test; and
• measurement of the change in turbidity of effluent indicating the base soil erosion.
A common alternative test method is the slurry test (Sherard et al. 1984b, Indraratna et
al. 1996). In this test the filter is compacted in the standard filtration apparatus, and then

slurry of the base soil is applied to the filter under high pressure to examine whether
individual base soil particles can move through the filter. The filter is successful if the
slurry forms a thin skin on the face of the filter, and unsuccessful if the slurry passes
through the filter.
Figure 2.1 Typical Laboratory Apparatus (Locke 2001)
Several other tests have been developed to identify particular facets of filter behaviour
or parameters affecting filtration. Examples include the crack erosion test proposed by
Maranha das Neves (1989) and physical examination of the filter pores by filling a filter
material with glue and cutting it into slices (Wittmann, 1979; Sherard et al. 1984a; Witt
1993). Some well-known design criteria based on experimental approaches are provided
in Table 2.1. While this list is by no means complete (Schuler and Brauns 1993; ICOLD
1994), it provides a good summary of the most commonly adopted criteria for dam filter
designs.

2.2.2 No Erosion Filter Test (NEF Test)
Sherard and Dunnigan (1985) realized that the standard filter test procedure described
above was not suitable for examining the filtration of fine cohesive soils, particularly
when the material is cracked. Sherard and Dunnigan (1985) proposed a new laboratory
arrangement, known as the “no-erosion” filter (NEF) test, to determine stable base soil-
filter combinations in the case of cohesive base soils. This test utilized the standard
permeameter apparatus (Figure 2.2). The filter material is compacted in the cylinder,
and a thin layer of base material compacted on top of it (typically 25mm thick). A hole
(1mm in diameter for fine soils and 5-10mm for coarse soils) is formed in the base soil.
Then a high water pressure (400 kPa) is applied to the system to investigate erosion of
the base soil through the filter. The base soil-filter combination is declared successful if
no visible erosion of the pinhole through the base soil occurred during a 20-minute test,
and a thin layer of fine base particles covered the filter interface, significantly blocking
the flow. The boundary filter particle size D15Fbdy is defined as the largest filter D15 size,
at which no visible erosion of the pinhole occurs. The filters coarser than this boundary
allow some erosion before they seal. Sherard and Dunnigan (1989) state that the NEF
test is the best test available for evaluating critical filters, located downstream of
impervious cores in embankment dams. The study also mentioned that the conditions in
the tests duplicate the most severe states that can develop inside a dam from a
concentrated erosive leak through the core discharging into a filter.

Figure 2.2 NEF Test Apparatus (Sherard and Dunnigan, 1985)

Table 2.1 Summary of some well-known empirical filter design criteria
Author Base Material Filter Criteria Comments
Terzaghi (1922) Uniform sand D15/d85 ≤ 4-5 and D15/d15 ≥ 4 Based partly on theoretical analyses and
partly on laboratory observations
Karpoff (1955) adopted Cu<4 5 < D50/d50 ≤ 10 (fine sand) 100% passing 75mm sieve, less than 5%
by USBR (1963) passing 75µm sieve, finer section of base
Cu>4 12 < D50/d50 < 58 and 12 < D15/d15< 40 and filter should have parallel grading
(natural, graded filters)
9<D50/d50<30 and 6<D15/d15<18
(crushed rock filters)
U.S. Army Corps of Uniform base with Cu<6 D15/d85 ≤ 5 and For medium to high plasticity clays
Engineers (1971) 5 < D15/d15 < 20 and D50/d50 ≤ 25 use D15F=0.4mm, also require CuF ≤20
Honjo and Veneziano Broad grading up to d95B/d75B ≤ 7 D15/d85 ≤ 5.5-0.5 d95/d75 Based on statistical analysis of compiled
(1989) previous research
Sherard and Dunnigan Fine Silts/Clays (>85% fines) D15/d85 ≤ 9 Fines content means the base soil fraction
(1985) finer than #200 sieve size (75µm) after
Silty/Clayey Sands (40-85% D15 ≤ 0.7mm regrading by #4 sieve size (4.75mm). Also,
fines) filters for fine soils (>40% fines) should
have less than 60% coarser than 4.75mm
Coarse material (<15% fines) D15/d85 ≤ 4 and maximum particle size 50mm.
Intermediate (15-40% fines) Interpolate between previous categories

based on fines content
Indraratna et al. (1996) Lateritic soil, S.E. Asia for d85 50 to 60 µm: D15/d85 < 5 to 5.5 Slurry tests on lateritic soil of South East
for d85 60 to 80 µm: D15/d85 < 4 to 5 Asia

2.2.3 Particle Size Ratio
Particle size ratios have been the most important parameters in evaluation of effective
filters. While working on the design of dams in North Africa, Terzaghi developed two
design criteria earlier given by Equations (2.1) and (2.2), partly based on laboratory
observations and partly using theoretical considerations (ICOLD 1994). Although these
criteria were developed using uniform sands as filter and base materials, they were initially
used indiscriminately to other soil types also. Numerous investigations have been carried
out since then, which considerably enhanced the understanding of filtration process. It now
seems well-accepted that base soil-filter stability can be ensured, provided an empirically
determined particle size ratio (or combination of ratios) is met, for example, D15/d85<5 or
D50/d50<30. Because so many different particle size ratios have been adopted, a discussion
of the scope of their application is necessary.
A mechanical sieve as a filter can be used to study self-filtration of the base material. Self-
filtration is the phenomenon where the coarser base particles are first retained at the filter
interface, and these coarse base particles are then able to retain some finer base particles.
Continuation of this process results in a stable, self-filtering interface, able to retain even
the finest particles. If a sieve is able to retain a soil containing both particles coarser and
finer than its aperture, then self-filtration must be occurring. Otherwise the fines would
continually move through the sieve. Many researchers using metal sieves as filters have
shown that very little soil loss occurs through the sieve when the sieve aperture is smaller
than d80. The loss is also greater when the aperture is larger than d90 (Vaughan and Soares

1982; Kwang 1990). This implies that it is acceptable to use d85 to represent the stability of
self-filtering base soils. Most of these tests used uniform base and filter materials.
Fischer and Holtz (1996), based on statistical analysis of previous research results,
suggested the particle size ratio D15/d75 accurately predicts granular soil retention
behaviour, regardless of the coefficient of uniformity of the base and filter materials.
However, there is no physical justification for adopting d75 to represent the base soil.
Kenney et al. (1985) studied the retention capability of granular filters by examining the
controlling constriction size, Dc*, which is the size of the largest base particle that can pass
through the filter. Dc* can be related to the particles sizes in the finer fraction of the filter,
determined by:
Dc*/D5 ≤ 0.25 (2.3)
Dc*/D15 ≤ 0.20 (2.4)
The study found that these relations are independent of the filter thickness exceeding
200D5. Equation (2.4) reinforces the use of D15 as a representative value of the filter size.
Witt (1993) measured the filter constriction sizes directly in laboratory from imprints of
filter particles on silicon rubbers and analysed them through probabilistic method. This
study also found that a single controlling pore size exists beyond 300D5 and is related to the
particles sizes in the finer fraction of the filter, given by:
dp* = 0.23d G (2.5)

where dG is the mean particle size, which varies from D5 to D10 for real filters and from D10
to D30 for uniform filters (Cu<3). The study ignored the filter compaction a significant
parameter in filtration analysis.
Sherard et al. (1984a) measured the dimension of filter constrictions using the Molten Wax
technique where a densely compacted gravel filter was filled with hot molten wax and was
dissected after freezing. The dimension of minimum flow channel was found to vary from
0.09D15 to 0.18D15 (i.e. the largest base particles that could pass through the filter were in
the order of 0.18D15). Similarly, it was found that the maximum dimension of flow channel
approximately varied from 0.1D15 to 0.6D15. However, there were no places along the
seepage path where the maximum dimension highly exceeded 6mm. Foster and Fell (2001)
mentioned that Sherard (unpublished 1985) also found that the maximum sizes of the base
particles ranged from 0.13D15 to 0.20D15 with a median of about 0.16D15.
The mid-size ratio, D50/d50, or similar ratios have been proposed by some researchers
including the U.S. Army Corps of Engineers (USACE, 1971) and Karpoff (1955). The
combination of this ratio with another (e.g. D15/d15) was intended to better represent the full
gradation curve of well-graded materials. However, the studies such as Sherard and
Dunnigan (1985), Honjo and Veneziano (1989) and Fischer and Holtz (1996) suggested
that the mid-size ratio does not correlate to filter performance and should not be used in the
filter design. A direct limitation on the coefficient of uniformity, Cu, may be more sensible

for broadly-graded materials. Most of tests carried out by Karpoff (1955) involved well-
graded base soils and filter materials with Cu larger than 6 and found that several tests with
D15/d85 less than 9 failed. However, Sherard et al. (1984a) contradicted the test results
suggesting that the failure criteria of Karpoff (1955), later adopted by the USBR, were too
conservative and concluded that “…we believe that all 7 of the USBR ‘failed’ tests were
actually successful”. By means of statistical analyses of hundreds of published test data,
Honjo and Veneziano (1989) suggested that “… although some of the cases judged by
Karpoff as “failures” might have been stable cases, most of them are likely to have been
very close to the critical condition”. Foster and Fell (2001) also suggested that the statistical
analysis of USBR (1955) data demonstrated that the results of these tests were not
consistent with the other data. However, such statistical analyses carry laboratory bias
associated with data and they also favour the majority of data in setting the trend. In this
respect, such analyses shift the trend towards the test results involving uniform base and
filter materials. This is because in the past, more tests were carried out using uniform
materials.
The fine size ratio, D15/d15, also known as permeability ratio, has been used to represent
both filter permeability and retention criterion. For example, U.S. Army Corps of Engineers
(1971) suggested the following relationship (Equation 2.6) for effective filters.
5 < D15/d15 < 20 (2.6)

Sherard et al. (1984a) suggested that d15 has no significant influence on the retention
properties of the needed filter. In many design situations it is necessary to ensure that the
permeability of the filter is sufficiently greater than the base soil permeability to drain the
soil and prevent the build-up of high pore pressures. However, it seems reasonable to adopt
a direct permeability relation such as filter permeability > 20 times base permeability rather
than an indirect particle size relation to meet the permeability requirement (ICOLD 1994).
Hence, the use of d15 is not recommended to represent filter performance, particularly its
retention characteristics.
Honjo and Veneziano (1989) performed a statistical analysis of extensive data from
previous research efforts. The analysis revealed that D15/d85 is the most important
parameter in predicting filter performance; accordingly, it seems reasonable to adopt D15 as
a representative filter particle size. Terzaghi retention ratio D15/d85 is also suitable for
representing the stability of a base-filter combination. Recent laboratory observations
(Foster and Fell 2001) also revealed that the safe D15/d85 ratio decreases when applied to
well-graded base soils.
2.2.4 Broadly-Graded Base Soils and Filter Materials
Broadly-graded soils having a wide range of particle sizes exist in nature in abundance.
Usually, the soils having Cu greater than 20 are considered to be broadly-graded. The soils
with Cu<20 can also be considered broadly-graded if the soil segregates during placement

(ICOLD 1994). Honjo and Veneziano (1989) examined published laboratory data of 287
experiments from 11 references, so that a statistical analysis of filter performance could be
performed. The statistical analysis determined that, in addition to the retention ratio, D15/d85,
the ratio d95/d75, called the self-healing index, is a less important but still significant
parameter. The self-healing index is related to the capability of the base soil to form a
satisfactory self-healing layer. The study proposed a new design rule for cohesionless, well-
graded and broadly-graded soils, given by:
D15 d
≤ 5.5 − 0.5 95 for d95/d75 < 7 (2.7)
d 85 d 75
It can be concluded from the above relationship that as the base soil grading becomes wider,
the safe D15/d85 reduces. However, there was no justification for d95 and d75, and their
physical significance could not be explained as is the case often associated with statistical
analyses.
A number of incidents have been reported in embankment dams with broadly-graded
cohesionless tills as the core material. Lafleur (1984) examined these materials to study the
self-filtration phenomenon. The base materials ranged in particle size from 2µm to 19mm,
with Cu of 8 to 360, containing 10 to 50% fines. An examination of the base soil particle
size distribution (PSD) showed that although the base particles coarser than 1mm make up
as much as 30% by weight of the total particles, they are so few in number that they float in
a matrix of finer soil. Hence, these coarse particles do not influence filtration.

In one test, an unstable filter resulted with the retention ratio D15/d85 equal to 2.2. Lafleur
(1984) revised this ratio considering only the fraction of base soil finer than US sieve no. 4
(4.75mm) size and found the retention ratio to be 8.4, in close agreement with experimental
data for uniform base soils (Bertram 1940). Based on this, Lafleur (1984) speculated that
self-filtration acts on base particles finer than a representative size that was ‘fortuitously’
the same as that given by the fraction passing 4.75mm. This representative size was likely
to be related to the shape of the particle size curve. Coarser particles are not numerous
enough to affect self-filtration. Lafleur et al. (1989) examined filtration of broadly-graded
cohesionless materials. The study assumed that for successful filtration, the allowable
opening size of the filter Of must be smaller than some indicative base particle size, also
called self-filtration size, dSF (Equation 2.8).
Of < dSF (2.8)
For uniform materials it is accepted that dSF is the d85 size. However, Lafleur et al. (1989)
showed that a finer indicative particle size is required for broadly-graded base soils in order
to reduce the extent of erosion before the filter seals and self-filtration occurs. Based on the
previous findings of Kenney et al. (1985), the pore opening size can be estimated from Of ≤
D15/5. Using this relation, Lafleur et al. (1989) examined stable base-filter combinations in
the laboratory, to examine dSF for various types of base soils such as broadly-graded, gap-
graded and internally unstable soils. Tests showed that the indicative base particle size, dSF,
depends on the type of base soils used. For base soils with linear grading (on the log scale)
and Cu>20, dSF ranges from d50 to d80. For gap-graded materials, dSF corresponds to the
lower fraction of the gap, since coarse particles play no part in filtration. Once the

indicative base particle size dSF is found, the original Terzaghi design criteria can be
applied, by replacing d85 with dSF.
Experiments by Lafleur et al. (1989) using wire screens as filters showed that for internally
stable base soils, there is some rearrangement of particles at the filter face, as self-filtration
occurs. This zone of particle loss and rearrangement was called the self-filtration zone. The
self-filtration zone is small for well-graded soils, but is quite extensive for broadly-graded
soils. This is because there are fewer large particles in the broadly-graded soils to initiate
self-filtration, and also a larger range of particles must be captured. Hence, more fines are
lost before the interface stabilizes, and broadly-graded base soils lose a large amount of
fines before successful filtration is established, if the Terzaghi filter criteria are applied.
This is why a finer indicative base soil size must be adopted to reduce erosion during
filtration of broadly-graded soils.
Lafleur et al. (1989) conducted laboratory tests using the filters with varying levels of
compaction i.e. relative density Rd ranging from 0 to 100%. The study revealed that some
of filters with Rd less than 50% were unsuccessful to retain the base soils whereas the same
filters at higher compaction levels were found to be effective in retaining the base soils;
hence, it was concluded that densely compacted filters behave differently from less
compacted ones, recognizing the filter compaction also a significant filtration parameter.

Broadly-graded filter materials also deserve discussion. These materials are attractive to
designers because they are able to retain fine base soils (a property controlled by the fine
filter fraction D15 or smaller), while still having coarse particles that can be easily retained
by the downstream fill of an embankment dam. This often means that a single filter can be
used, rather than multi-stage filters. Often broadly-graded filter is significantly cheaper to
manufacture than a uniform filter, or may be available in local natural deposits. However,
broadly-graded filters do have their drawbacks. Segregation during placement is a major
concern because the fine fraction of the filter material is relied upon for base soil retention.
If the fine fraction is not present due to segregation, then piping can occur. In addition, a
single layer of a broadly-graded filter material will usually have a lower permeability than
the overall permeability of a multi-stage filter. Hence, the filter may not provide the
necessary drainage properties. Clogging of broadly-graded filters is also more likely,
leading to high pore pressures in the dam core. To reduce the risk of segregation or
clogging within critical filters in embankment dams, the US Natural Resources
Conservation Service (NRCS 1994) recommends a Cu value of less than 6 for the coarse
and fine sides of the filter band (i.e. the band between the maximum and minimum
allowable gradings in the design specification). In addition, the ratio of maximum and
minimum allowable particle sizes (the width of the filter band) for the fraction finer than
D60 should be limited to 5 in order to avoid the selection of gap-graded filters.

2.2.5 Internally Unstable Base Soils
In internally unstable soils, the coarse particles form a skeleton that sustains all of the
external stress within the soil. Some fine particles are not part of the soil structure, and are
not restrained by the external stresses. These particles, under the influence of seepage or
vibration forces, can move through the skeleton of coarser particles, causing internal
erosion. This process is called suffusion. Filter design criteria relying on the coarse fraction
of the base soil (e.g. D15/d85) are not suitable for internally unstable soils since extensive
base soil loss may occur even when the coarse fraction of the base soil is retained.
Internally unstable soils generally have either a concave upward or gap-graded grading
curve. Lafleur et al. (1989) showed that internally unstable soils which are successfully
retained by a filter (i.e. the filter is fine enough to retain the loose fine base particles) may
“self clog”. This is a process where fine particles are washed out through the base material
and accumulate at the filter interface, producing a layer of low permeability and increased
pore pressures at the interface.
Kenney and Lau (1985) investigated the internal stability of soils. The study suggested that
for a material to be internally unstable, the following three conditions are necessary.
• The compacted granular material must possess a primary fabric of coarse particles which
support the imposed loads;
• Within the pores of the primary fabric, there must be loose particles which can be moved
by forces such as water flow; and

• The size of constrictions in the pore network formed by coarse particles must be larger
than some of the loose particles.
This requires a gently inclined particle size distribution (i.e. wide range of particles) or gap-
grading in the lower part of the PSD curve. Kenney and Lau (1985) described a method
based on the particle size distribution (PSD) to determine the internal stability of a soil. The
authors defined a boundary between stable and unstable behaviour, which is defined by:
H/F=1.3 (2.9)
where
H = percent of mass between two particle sizes, D and 4D;
F = percent of mass finer corresponding to particle size, D.
The test conditions used to define the above boundary were severe and considered to be
overly-conservative. For this reason, the conservative test conditions were recognized in the
closure to this paper and the above relationship was revised to:
H/F=1 (2.10)
Moreover, they suggested that the best method to determine internal stability is to perform
a hydraulic test in a permeameter. For internally unstable soils, the filter could be designed
to retain the fraction of fine base particles that are stable, using normal design criteria.
However, the problem of “self clogging” must still be considered.

Kwang (1990) performed laboratory tests on gap-graded materials having a distinct gap in
the particle size distribution. Kwang (1990) showed that instability of gap-graded materials
occurs when the particle size ratio of the gap range (the ratio of the higher to lower sizes of
the gap) is greater than four. This simple relation can be used to determine the internal
stability of gap-graded materials.
2.2.6 Particle Size -Permeability Relationship
Vaughan and Soares (1982) suggested an alternative method for designing filters for non-
cohesive soils based on a base soil size-filter permeability relationship, as shown in Figure
2.3. Since the permeability of a filter implicitly reflects the porosity and differences in
grading and grain shape, this relation should take into account all of these factors. Vaughan
and Soares (1982) suggested a design rule given by:
k = 6.7 × 10 −6 × δ 1.52 k in m/s and δ in µm (2.11)
where, δ represents the soil particle size to be retained. For an internally stable, non-
cohesive base material, δ should be taken as d85. For cohesive base soils, Vaughan and
Soares (1982) suggested that the filter should be designed to retain the mean size of the clay
flocs. Indraratna et al. (1996) developed an alternative particle size-permeability
relationship for filtration of cohesive, lateritic soils, based on the observation of slurry tests.
This relationship is based on the d85 size of the base soil, given by:

k f = 6.3 × 10 − 4 × (d 85 )
1.25
(2.12)
Indraratna et al. (1996) also found that a relationship exists between the permeability and
the size of fine soil particles for non-cohesive soils given by:
k = 0.96( D5 D10 ) 0.92 (2.13)
where, D5 and D10 are in mm, and k in cm/s. Since permeability can be related to the fine
particle sizes of the filter material and it has been shown that the fine filter particles are
responsible for filtration, the relationship suggested by Vaughan and Soares (1982),
between filter permeability and the base soil size, appears to be a plausible design criterion.
Indraratna et al. (1990) compared their experimental results examining the effect of filter
permeability on the filtration of lateritic clay with those of Vaughan and Soares (1982)
finding similar results (Figure 2.3). These observations suggested that a base soil particle
size-filter permeability relationship is adequate for the design of granular filters. However,
there are practical problems with determining the in-situ filter permeability during dam
construction, as the permeability may change appreciably, with small fluctuations in the
fines content and extent of compaction. The standard filter grain size ratios are generally
easier to control during dam construction.

1000
effective ineffective clogging
Indraratna et al. (1990)
Vaughan & Soares (1982)
100
d85B
(µm)
lateritic soil (Thailand)
10
well graded till (U.K.)

Vaughan & Soares (1982)
London clay (dispersed)

1
0.001 0.01 0.1 1
Permeability (cm/sec)
Figure 2.3 Base particle sizes vs. filter permeability relationship (Indraratna et al. 1996)
2.2.7 Cohesive Base Soils
The majority of dam cores are made of cohesive materials such as sandy or silty clays,
which is impermeable enough to prevent the loss of water from the reservoir. However,
most of design criteria described so far have been developed using non-cohesive base
materials. Many designers applied design principles developed for non-cohesive base
materials to cohesive bases, assuming that they would be conservative, since the cohesion

of the base soil reduces erosion rates. However, based on observations of dam failures,
Vaughan (2000) suggested that the rules for non-cohesive soils were invalid for two
reasons. Firstly, the cohesive forces in the clay did not prevent filter failure; rather they
allowed cracks to stay open and stable while their walls were eroded by small flow.
Secondly, segregation invalidated reliance on self-filtering to prevent loss of material. Self-
filtering is explicitly assumed in the design criteria for non-cohesive soils. Concentrated
leaks occur in most embankment dams of all types and sizes without being observed
(Sherard and Dunnigan 1985). After a crack develops, erosion of the crack by high velocity
flow may occur as shown in Figure 2.4. Eroded particles may be transported to the filter
interface. If the filter is fine enough, these particles are captured at the filter interface and
form a mud skin over the filter. This low permeability skin or filter cake reduces the flow
rate through the crack and prevents further erosion. If the filter is too coarse, eroded
particles will pass through the filter and the crack may enlarge as erosion continues.
Figure 2.4 Concentrated leak through a crack in a cohesive core

(after Sherard et al. 1984b).

Vaughan and Soares (1982) found that cracks up to a certain size and shape in a cohesive
material remained stable even when they were flooded. The study suggested that at low
flow velocities, slow erosion of these cracks may be accompanied by segregation of the
eroded debris within the crack, this segregation may result in only fine particles reaching
the filter. In the absence of coarse base soil particles, a self-filtering layer cannot form and
the finer particles are continually lost and the crack enlarges, leading to possible piping
failure. Based on this, Vaughan and Soares (1982) defined a “perfect filter” to protect a
cracked, cohesive base material. The perfect filter will retain the smallest particles that can
arise during erosion, even if they arrive at the filter interface after complete segregation.
These smallest particles are the clay flocs that form when the base material is dispersed in
the reservoir water. The perfect filter concept is a conservative approach, intended to
provide a filter that cannot allow particles to erode. Several studies including Ripley (1982)
criticized the design of some filters that would be so fine they may possess some cohesion.
This is unacceptable, as cracks can propagate through the filter. Sherard (1982) in his
discussion of the perfect filter concept mentioned that:
• Laboratory tests revealed that filters of clean sand (or gravel sand) with D15 size of
0.5mm conservatively seal concentrated leaks in fine-grained cores.
• Silt-sized particles (30-70 microns) comprise a substantial portion of all fine-grained
clayey soils. As these particles are available to seal concentrated leaks, it is not necessary
to provide a “perfect filter” to catch clay flocs of 10-20 microns size.

It is now accepted that the “perfect filter” is conservative. There is sufficient evidence
suggesting that well constructed fine sand filters can initiate self-filtration and adequately
protect most of dam cores (Sherard et al. 1984b).
The ‘critical filter concept’ suggested by Sherard and Dunnigan (1985), adopting the no-
erosion filter (NEF) test, was proposed as an alternative approach to the design of filters to
seal cracks in cohesive materials. It is a conservative test of filters to seal concentrated
leaks through cohesive soils. The empirical design criteria of Sherard and Dunnigan (1985),
presented in Table 2.1, are based on the NEF test. In most instances, these design criteria
provide a coarser filter than that required by the perfect filter concept. Vaughan (2000)
criticized the use of the NEF test to model the cohesive soils, suggesting that the test fails
to reproduce the features of crack behaviour in two ways. Firstly, it is conducted in a rigid
cylinder. If pieces of clay are washed onto the filter, the filter face can seal. Being rigidly
constrained, the filter and the walls of the crack can resist the hydraulic pressure applied,
and the filter is defined as a success. In a dam core, a crack can be sustained and re-opened
by hydraulic pressure, if this pressure is larger than the total stress. The mechanism in the
test is strain controlled, whereas the mechanism in the field is stress controlled. Secondly,
the segregation of eroded debris and migration of only fine particles is specifically avoided
in the NEF test. Sherard and Dunnigan (1985) conducted several NEF tests at very high
hydraulic gradients in vertical apparatus. These tests found the suitability of the filter to
retain particles eroded by high velocities, but did not examine the possible segregation at

low velocities. Thus, the NEF test cannot predict field behaviour if segregation of debris is
possible.
In order to study flow at low velocities, Maranha das Neves (1989) developed a ‘crack
erosion test’, where water flows at varying velocities over the flat surface of a soil sample
and then through a filter, to examine the behaviour of the simulated crack and filter. The
following observations were made.
• There was no visible segregation during transport of the eroded debris, i.e. all the eroded
material is transported and there is no preferential movement of fines;
• When erosion occurs, low flow velocities (2cm/s) are sufficient to transport sand-sized
particles to the filter surface, thus enabling self-filtration to occur at the filter interface;
• Before self-filtering is established, even the most conservative filters were not able to
retain fine particles.
Maranha das Neves (1989) concluded that segregation at low flow velocities is not a
problem in filter design; the NEF test is suitable for determining successful filters for
cohesive soils and the Sherard and Dunnigan (1985) criteria can be adopted for the design
of filters for most cohesive base soils.
The design criteria of Sherard and Dunnigan (1985) have been shown to have limitations,
particularly when applied to broadly-graded and gap-graded cohesive materials. Khor and
Woo (1989) conducted a number of NEF tests on sandy clays. The study assumed that

particles coarser than fine sand cannot be relied upon to help seal the filter, and that it is
necessary to provide a protective filter that will retain the fines but not necessarily the clay
floc-sized particles.
Indraratna et al. (1996) studied the filter requirements for a cohesive, lateritic soil, a typical
residual soil of Thailand and other parts of South East Asia. This material lacks much of the
silt-sized particle fraction usually present in other natural soils. The experiments involved
forcing slurry of base soil into the filter under high pressure, to examine the retention of
clay flocs. Based on these observations, Indraratna et al. (1996) revealed that the following
particle size ratios are appropriate for this material.
For d85= 50 to 60 µm: D15/d85 <5 to 5.5 (2.14)
For d85= 60 to 80 µm: D15/d85 <4 to 5 (2.15)
The retention ratios D15/d85 (Equations 2.14 and 2.15) for filtration of a lateritic soil are
considerably lower than those proposed by Sherard and Dunnigan (1985), for fine soils (i.e.
D15/d85 ≤9). This is most likely because the lateritic soil is low plasticity soil and behaves
like a cohensionless soil in relation to filtration. These lower safe ratios suggested that the
current filter criteria for cohesive fine soils may not be universally applicable and testing of
proposed combinations is usually necessary. The current practice of using concrete sand as
a filter material to protect fine-grained soils reflects general indistinctness in determining
correct filter materials to protect cohesive soils, for example, the Sherard and Dunnigan
(1985) requirement, D15 <0.7mm, for sandy silts and clays.

More recently, statistical analyses carried out by Foster and Fell (2001) found that the
criterion D15/d85 9 corresponds to a 50% chance of erosion whereas for a 10% chance, the
ratio is 4.60. Unlike the observations of Sherard et al. (1984b), where the study found that
the base soil-filter combinations with retention ratios much higher than 9 were successful,
Foster and Fell (2001) and Sherard and Dunnigan (1989) found that the failure boundary
varies in between 6 to 14. However, both the studies recommended D15/d85 9 as filter
criterion as a mean value of two limits. The former also recommended that for highly
dispersive base soils with more than 85% fines, D15/d85 6.4 be used whereas for base soils
with fines content in between 35 to 85%, D15 0.5mm is more appropriate. The study
ignored the effect of filter gradation and used only densely compacted filters.
2.3 Analytical Investigations
With increased use of computers in engineering research, it has been possible to perform
detailed numerical simulations of the particle movement at the base soil-filter interface.
Based on the fundamental principles of hydrodynamics such as conservation of mass and
momentum, recent years have produced a few numerical models (Indraratna and Vafai
1997; Locke et al. 2001) to simulate the complex base-filter particle interaction during
filtration. Although these analytical studies do not provide any direct criteria to distinguish
between effective and ineffective filters, they have greatly enhanced overall knowledge of
filtration.

2.3.1 Numerical Models
Within the filter, it is the voids rather than the particles that control seepage and filtration.
In particular, base particles are usually trapped by the smallest part of a connection between
two voids. The size of these constrictions is dependent on the size and packing geometry of
the filter particles. The normal approach is to determine the filter void constriction size
distribution (CSD) in some way, usually based on the filter PSD and adopt a filter void
model. These numerical models consider the movement of base particles under seepage
forces, and the mechanisms of capture of these particles within the filter. Particle capture is
usually modelled by a probabilistic comparison of base particle and filter constriction sizes.
A particle smaller than the filter constriction between pores can pass through to the next
pore where the probabilistic comparison is repeated (Silveira 1965). In this way, the
expected infiltration distance of base soil particles can be simulated by a probabilistic
analysis of the base PSD and filter CSD. Other factors that these models consider include
the amount of base particles mobilized by applied seepage forces and changes in the filter
CSD as particles are captured. Void network and particle transport models are described in
following subsections.
2.3.2 Filter Void Models
Silveira (1965) proposed a simple filter void model where base soil particles encounter
constrictions at uniform spacing in the direction of flow. If a particle is smaller than a
constriction, then it can move to the next constriction, where the comparison is repeated.

Kenney et al. (1985) also used a similar multi-layered void network model as shown in
Figure 2.5. Probability functions are used to estimate the number of confrontations with
randomly generated constrictions until a base particle is retained by a smaller constriction.
An alternative void model represents the filter voids as a series of channels of varying size
(Figure 2.6). The smallest of the pore constrictions within the pore channel governs the size
of a base particle that can pass through the pore channel.
Indraratna and Vafai (1997) adopted a model of this type. Kovacs (1981) found the model
to be a good representation of the large and small pores in a natural soil. In this model, the
minimum pore channel size, d0, is given by:
ne Dh
d 0 = 1.63 (2.16)
1 − ne α
where, Dh is a mass weighted equivalent diameter and α is a shape coefficient. Methods to
determine these values can be found in Indraratna and Vafai (1997). Locke et al (2001)
found that although this model is a good approximation to uniform filters, it overestimates
the minimum pore channel for well-graded filters. For this reason, the authors considered
the three-dimensional void network to further extend the Indraratna and Vafai’s (1997)
numerical model.

Figure 2.5 One-dimensional multi-layered void network filter model
(Kenney et al.1985)
A recent development has been the use of a three-dimensional pore network model. Witt
(1993) developed a pore model of spheres as ‘pores’ interconnected by pipes as ‘pore
constrictions’ (Figure 2.7). Particles can move from one pore to another through any of the
constrictions, provided that the particle is smaller than the constriction size connecting two
pores. The study found that from each pore, there are a number of possible exits (i.e. pore
constrictions), and the largest constriction size determines whether a particle can move
from the pore.

Pore Channel
d max
da
do
∆z
(a) (b)
Figure 2.6 Void Channel Model (Indraratna and Vafai 1997)
Schuler (1996) used a regular cubic network model of pores interconnected by six
constrictions, similar to that of Witt (1993), as shown in Figure 2.8. Schuler (1996)
determined that there were on average 5.7 constrictions per pore, and hence adopted the
cubic network with six constrictions connecting every pore.

Figure 2.7 Pore Network Model (Witt 1993)
Figure 2.8 Cubic Pore Network Model (Schuler 1996)
2.3.3 Constriction Size
All of the filter void models described above require a constriction size distribution in order
to determine the size of particles that can pass through the filter. Various modelling and

laboratory methods have been used to estimate the size of constrictions. Silveira (1965)
defined a constriction to be the largest circle that can fit between three tangent filter
particles (Figure 2.9a). Based on geometrical probabilistic approach, Federico and Musso
(1985) also developed a mathematical method to determine the constriction size formed by
three tangent particles. These models assumed that the grains are spherical, the filter is at its
maximum density, and the relative positions occupied by the grains are random. Based on
the probabilities of occurrence of each filter particle size, determined from the filter PSD, a
combination of these probabilities can lead to the probability of different constriction sizes,
and hence, a CSD curve. Since filters are not always compacted to maximum density,
Silveira et al. (1975) suggested another particle arrangement for the loose state of a soil,
where four particles combine to form a void. This model is shown in Figure 2.9b.
Constriction
Size
(a) (b)
Figure 2.9 Particle packing arrangement for a) dense and b) loose states
(Indraratna and Locke 2000)
Silveira (1965) assumed a one-dimensional void model similar to Kenney et al (1985),
which has been shown to be a good approximation for uniform filters. However, in more
broadly-graded filters, the use of the particle size distribution by mass (as determined by

sieving), to represent the frequency of filter particles, introduces errors. This is because
large particles, with a high individual mass but low number, are over-represented in the
model and produce a high number of large pores. It is unlikely that these few large particles
meet to form a large pore. De Mello (1977) showed that the Silveira (1965) model predicts
an increase in the ratio of large constriction sizes to median filter particle sizes, dv85/D50, as
the coefficient of uniformity of the soil increases (Figure 2.10), where dv85 is the filter
constriction size in the dense model, where 85% constrictions are finer than the size. It is
expected that as Cu increases, the number of small particles filling voids between the larger
particles would increase, leading to smaller constriction sizes. Hence, the PSD by mass
should not be used to model the constrictions of well-graded filters. Kenney et al. (1985)
and Federico and Musso (1993) overcame this problem by converting the PSD by mass to
the PSD by number of particles. This approach gives a better approximation of the real
CSD, but still has been shown to result in errors for broadly-graded filters. In broadly-
graded filters, the smaller constrictions are over-estimated because of numerical superiority
of fine particles. Humes (1996) and Schuler (1996) used the PSD by surface area. This is
considered to be more representative of the possible particles, which may form a
constriction. This is because although there are only a small number of larger particles, they
have a great number of contacts with other particles, due to their large surface area.

Figure 2.10 Comparison of Filter Constriction Model (DeMello 1977)
Schuler (1996) also produced a model of the size of pore constrictions using an adaptation
of the geometric method of Silveira (1965). The model is shown in Figure 2.11. Four
particles are present and form two constrictions in a variable geometric assembly.
Importantly, the model considers the effect of filter density. The angle α (Figure 2.11)
decreases with increasing density, producing smaller constriction sizes. The least dense
particle packing is considered when α=90o, while the most dense packing is the case where
α is a minimum and filter particles B and D are touching, producing two small
constrictions.

Figure 2.11 Model for geometrical structure of pore constrictions
(Schuler 1996)
Considering that the real filters are most likely to exist between two extreme states (i.e.
loose and dense states), Indraratna and Locke (2000) suggested that the actual constriction
size in a real filter can be given by:
DV ,i = DVMD ,i +
i
(1 − Rd )(DVLD ,i − DVMD,i ); i = 0,1,2,...n (2.17)
n
where, DVMD,i and DVLD,i are the 100 i ( n

)% coarsest constrictions from the most dense and
least dense constriction size distributions respectively.

Some researchers have taken a direct approach and measured the constriction sizes within a
filter, rather than estimating the sizes. Wittman (1979) filled a filter medium with resin and
then, after hardening, cut the sample into sections to examine the void size distribution.
Using this measured CSD as a probability density function, he developed a model of a flow
path in the form of a pore channel with irregular width in the direction of flow. Wittman’s
(1979) work drew some criticism, because a slice of the filter material shows a random
plane through both pores and pore constrictions, rather than the smallest part of the voids
(the pore constrictions), which form the restrictions to particle movement.
Witt (1993) measured the pore constriction size distribution of a real material; silicon
rubber was poured into the voids of gravel and cut open to reveal 3D representations of the
pores and constrictions. In this case, the truly three-dimensional shapes of the filter pores
were obtained, overcoming the problem of measuring constriction sizes in slices of the
material. This led Witt (1993) to a statistical distribution of the largest constriction size
associated with each pore. The distribution of exit sizes was found to be log-normal. The
size of effective opening was given by Equation (2.7) discussed earlier. Kenney et al.
(1985) measured the size of the largest base particles eroded through the filters, called the
controlling constriction size and found to be correlated with fine fraction of the filter as
discussed earlier (Equations 2.3 and 2.4). Furthermore, based on the probabilistic theory of
Silveira (1993), Soria et al. (1993) conducted several filtration hydraulic tests to determine
the constriction size distribution of the granular filter and non-woven geotextile and found
that for the granular filter, there is a good agreement between the theory and test results.

2.3.4 Particle Infiltration Models
The next step of analytical and/or numerical modelling is to describe the movement of base
soil through the filter void model. The two broad categories of infiltration models are
probabilistic methods and particle transport models. Probabilistic methods simulate the
probable depth of base particle penetration into the filter voids by comparing the
probability of a base particle encountering a void through which it can pass. Silveira (1965)
assumed that particles move only in the direction of flow and encounter random void
constrictions at uniform intervals. The probability p that a base particle of size d can pass
through a single, random filter void constriction is the cumulative fraction of constrictions
larger than d, which can be found directly from the CSD (i.e. p is the fraction coarser than d
from the CSD). Then the number of confrontations, n, until the particle encounters a
smaller void constriction through which it cannot move, based on a confidence level P can
be estimated by:
n=
(
ln 1 − P ) (2.18)
ln( p )
And the distance L the particle can infiltrate into the filter is given by:
L = n.s (2.19)
where, s is the distance between each confrontation with a pore constriction. In other
words, s is the distance between unit layers in the filter pore model, often called a unit step.
Silveira (1965) assumed that s is given by the mean diameter of filter grains (from the PSD
by mass). Humes (1996) improved on the Silveira (1965) model by considering the PSD by

surface area to represent the filter particles. However, Indraratna and Locke (2000) and
Locke et al. (2001) suggested that one-dimensional pore network model is conservative. In
reality, the pore network is more likely to be the one represented by three-dimensional
network suggested by Schuler (1996). In this pore network model, a base particle is
assumed to have larger probability of forward movement. Unlike in the one-dimensional
model, the base particle, if stopped by smaller constriction at the bottom of a pore, may
take sideway exits and subsequently may move downwards. Based on this assumption, the
overall probability P(F) of one step forward movement for a base particle with a passing
probability p can be given by (Locke et al. 2001; Indraratna and Locke 2000):
[ ] {[ ] }
∞
P(F ) = p + ¦ 1 − (1 − p ) (1 − p ) p 1 − (1 − p ) (1 − p )
4 3 i
(2.19)
i =0
The studies also suggested that in the Equation (2.18), p should be replaced with P(F)
providing a new relationship for particle infiltration given by:
n=
(
ln 1 − P) (2.20)
ln( P( F ))
Kenney et al. (1985) used the Silveira (1965) constriction size model and the unit layers
void model, as shown in Figure 2.5. Rather than a probabilistic analysis of the infiltration
depth of particles, the analysis then considered the probability of occurrence of minimum
constriction sizes along flow paths perpendicular to the unit layers. Kenney et al. (1985)

found that the minimum constriction size encountered along any flow path soon approached
a constant value as the filter thickness was increased. This was called the controlling
constriction size, Dc* and was found to be related to the finer fractions of filter particle
sizes, represented by Equations (2.3) and (2.4).
Using stochastic methods, based on a pore model and the size distribution of the pore
constrictions, the probability that a particle meets a constriction through which it cannot
pass in a number of confrontations can be determined. Witt (1993) used a three-
dimensional pore network model (Figure 2.7) to show that as the number of confrontations
with constrictions increases (i.e. the filter thickness increases), the minimum constriction
size of any path tends to a constant value dp* for all paths, and for a depth of penetration L,
that can be given by:
dp* = 0.27 dG for L ≈ 60 D5 (2.21)
dp* = 0.23 dG for L ≈ 300 D5 (2.22)
In the above equations, dG is the mean grain diameter of the filter (by number), usually
between approximately D5 to D10. Particles finer than dp* can pass through the filter. An
important observation, similar to that of Kenney et al. (1985), was that if L is increased
beyond 300D5 the decrease in the minimum constriction size is negligible. If there are
particles in the base soil which are larger than dp*, then these particles are retained and self
filtration is initiated (provided the base soil is internally stable).

Aberg (1993) model considers a random line drawn through a collection of particles and
compares the length of sections through particles (a grain chord) and the length of sections
through voids (a void chord). Based on the average length of these chords, the model
estimates the average void sizes, which were found to have a good correlation with the test
results of Kenney et al. (1985).
2.3.5 Particle Transport Models
Particle transport modelling has been considered extensively in both contaminant transport
and chemical filtration (Reddi et al. 2000; Reddi 1997). However, the use of these models
to describe the movement of large particles, such as through granular filters, has been
limited. The usual approach is to consider basic physical concepts such as conservation of
mass and momentum, to model the flow of a fluid containing solid particles through the
voids of a filter. Honjo and Veneziano (1989) developed a soil particle transport model
based on conservation of mass in the solid and liquid (slurry of soil and water) phases. The
model is capable of describing absorption and release of soil particles with time in different
elements of the base and filter. Various soil particle sizes can also be considered. The
model was used to demonstrate self-healing of the base soil as coarser particles collect at a
screen. In addition, internal stability was investigated using the model.
Indraratna and Vafai (1997) adopted the particle transport approach, incorporating a simple
pore channel model (Figure 2.6) that provides the geometric constraint to movement. The

model also considers the hydraulic forces required to mobilize the base soil particles. The
model was later extended by Indraratna and Locke (2000) by incorporating the cubic void
network and enhanced particle transport models. A critical hydraulic gradient, ic, is
determined based on a balance of the frictional resisting forces with the hydraulic and
gravitational disturbing forces. Figure 2.12 shows the situation considered, where the
contact friction resists erosion, unless the applied hydraulic gradient is large enough to
overcome the frictional resistance. The frictional forces (Fxz and Fyz) are assumed to be
caused by the horizontal confining stresses σx and σy acting on the projected sides of the
particle. In an irregular pore channel it is difficult to estimate the exact friction mobilized at
a given location. For the analytical model, it was postulated that the upper limit of the
frictional force is proportional to the fully mobilized lateral stress in the filter. This lateral
stress can be estimated from the overburden depth hs and hydraulic head hw (i.e. an
effective vertical stress of hsγs - hwγw). Hence the upper limit of the frictional forces is given
by:
Fxz = Fyz < 0.5πd 2 K(γ s h s − γ w h w ) tan φ ' (2.23)
A critical hydraulic gradient, ic, is determined based on a balance of the frictional resisting
forces with the hydraulic and gravitational disturbing forces, given by:
4K ' 2d
ic <
δzγ w
(γ h
s s − γ w h w ) tan φ −
3δzγ w
(γ s − γw ) (2.24)

ªπ φ' º
where, K < tan 2 « − » and φ’ is the effective friction angle between base and filter
¬4 4 ¼
particles. If seepage forces exceed the critical hydraulic gradient, and the particle is smaller
than the geometric constraint, d0, it will move. Moving particles are controlled by the
governing differential equations of mass and momentum conservation. The base-filter
interface is divided into elements as shown in Figure 2.13a. Considering the mass flow
rates in and out of a typical element shown in Figure 2.13b, and the rate of mass
accumulation within the element during a time period, dt, the principle of mass
conservation requires:
d ( ρ m u) dρ m
= (2.25)
dz dt
The principle of momentum conservation is applied to the volume of slurry, Vm, to give:
§ du du ·
¦F = ρ m m V ¨ +u ¸
© dt dz ¹
(2.26)

σ x= λ σ z σ y= λ σ z
pore channel
σz σz
∆P ∆P
δz
σx σx σy σy
direction W direction direction W direction
Fxz Fxz Fyz Fyz
y
x Fu Fu
z do z do
Figure 2.12 Equilibrium of particle “plugging” a vertical pore channel in

a) XZ plane; b) YZ plane (Indraratna and Vafai, 1997)
The external forces, ΣF, include surface forces due to hydrodynamic pressure and body
forces resulting from gravity and viscous drag. Defining R as the viscous drag per unit mass
of the slurry, the external forces can be determined by:
∂P
¦ F = − ∂Z V m − ρ m gVm + ρ m RVm (2.27)
By considering a number of elements at the base soil-filter interface, the movement of
particles is modelled by a finite difference analysis. The process predicts the change in
slurry density (ρm) and the slurry velocity (u) within the time-step, ∆t. The slurry velocity is
then incorporated in Darcy’s law to compute the corresponding hydraulic gradient at each
time-step. The flow rate is easily determined from the continuity equation, according to the
slurry velocity and element porosity. The PSD of each element at any given time is

computed based on the mass and PSD of the element at the previous time step and the
subsequent rate of erosion and retention. The PSD of each element is recalculated for each
time step and the finite difference procedure repeated. This analysis shows the gradual
change in particle size distribution (Figures 2.14 and 2.15), permeability and porosity of the
base soil and filters elements, and hence, describes what is occurring at the base - filter
interface with time for the entire particle size range. This is the only method to date that
uses the entire PSD and considers time variant changes in PSD. Although such analyses do
not provide any explicit criteria to identify effective filters, they certainly explain the
process occurring at the base soil-filter interface and give a confident picture of particle
interaction under given geo-hydraulic conditions.
inflow
(ρmu)
B1
base soil flow
B2 zi
in direction
core
Bn
i th element ∆z
F1 z i + ∆z
filter FE2
F2
∂ (ρ u)
(ρmu)+ m dz
Fn ∂z
FEn
effluent
Figure 2.13 Illustration of (a) base and filter elements and (b) generalised
slurry flow through a filter element (Indraratna and Vafai, 1997)

100
Filtration Time constant head
90
80 0 min BE1
Percent finer 70 10 min BE2
60 20 min FE1
50 30 min FE2
40 50 min
effluent flow
30
'tail' of fine base particles
20
zone of base particle
10
retention
0
0.01 0.1 1
Particle size (mm)
Figure 2.14 Numerical simulation of time-dependent change in filter PSD in an

effective filter (Indraratna and Vafai, 1997)
4.5
3.5 Minimum
Mass Loss (g/cm2)
3 Lab D15/d85=4
2.5 Filter Lab D15/d85=6
Lab D15/d85=8
2
Model D15/d85=4
1.5 Model D15/d85=6
Model D15/d85=8
1
0.5
0
0 5 10 15 20 25 30
Distance from filter interface (cm)
Figure 2.15 Numerical and laboratory simulation of particle capture with depth
within a granular filter (Indraratna and Locke 2000)

2.4 Current Filter Design Guidelines
Beginning with two simple relationships proposed by Terzaghi in early 1920s, the filter
design guidelines have evolved after numerous investigations carried out, under varying
geo-hydraulic conditions over the period of more than seventy years. These studies
considerably enhanced the knowledge of filtration. Several design guidelines were
proposed and practiced in the past. As more studies were carried out revealing more facts
about the filter behaviour, these guidelines were revised on several occasions. Some of
important guidelines adopted in the past were USACE (1955), USBR (1963), Sherard et al.
(1963), USACE (1971) etc. However, although all important outcomes of these studies
have been incorporated implicitly in the current design guidelines, the basic form of design
criteria still remains the same. ICOLD (1994) suggests the following two design procedures
as the current state of practice in the filter design.
2.4.1 Lafleur Filter Design Procedure
As discussed earlier, the studies such as Lafleur (1984) and Lafleur et al. (1989) found that
the classic Terzaghi retention criterion leads to unsafe filter designs when applied to
broadly-graded cohensionless base soils (tills). Lafleur et al. (1989), as discussed earlier,
compared the filter opening suggested by Kenney et al. (1985) with the indicative base
particle size. The indicative base size in the case of broadly-graded and gap-graded base
soils is invariably smaller than d85, and an internally unstable base soil with concave
upward PSD curve have appropriate size as fine as d20. Based on the outcome of these tests

on broadly-graded cohensionless tills, Lafleur suggested a design procedure involving the
original Terzaghi criterion where d85 is replaced by the appropriate indicative base particle
size. The whole procedure is depicted in flow-chart diagram as shown in Figure 2.16.
Figure 2.16 Filter design procedure-Lafleur method (ICOLD 1994)
2.4.2 NRCS Filter Design Procedure
This design procedure is based on the results of laboratory tests, carried out by the Natural
Resources Conservation Service (formerly the Soil Conservation Service) from 1980 to
1985 and mainly includes the works of Sherard and colleagues (Sherard et al. 1984a;

Sherard et al. 1984b; Sherard and Dunnigan 1985; and Sherard and Dunnigan 1989). This
is the most widely used filter design guideline in current practice. A recent version of the
procedure is presented in the National Engineering Handbook 1994 of the US Department
of Agriculture. The guideline requires classifying the base soils into four categories,
depending on the fines content (i.e. fraction smaller than US #200 sieve size, 0.075mm),
determined after regrading the base soil PSD curves for the particle size larger than US #4
sieve size (i.e. 4.75mm) as presented in Table 2.2. Subsequently, the maximum D15 size of
effective filters for each group is determined by the design criteria given in Table 2.3. The
NRCS guideline also imposes the constraint on the maximum size of filter particle and
uniformity coefficient Cu of the filter bands in order to prevent the segregation during filter
installation and to avoid the selection of gap-graded filters. A recent laboratory study
(Foster and Fell 2001) suggests that the limit of fines content of category 2 should be
changed to 35-85%; maximum D15 for dispersive soil in category 2 should be lowered
down to 0.5mm and required regraded retention ratio D15/d85R for category 1 dispersive
base soils to 6.4. Although the studies such as Sherard and Dunnigan (1989; 1985) and
Foster and Fell (2001) found that the tests on fine silts and clays failed with retention ratios
from 6 to 14, they still recommended D15/d85R9 as filter criterion for the soils in the
category 1, considering the average value.

Table 2.2 Base soil category based on fines content (NRCS 1994)
Base Fines Content Base Soil Description
Soil (%<0.075mm)
Category
1 >85 Fine silt and clays
2 40-85 Sands, silts, clays, and silty and clayey sands
3 15-39 Silty and clayey sands and gravel
4 <15 Sands and gravel
•
Table 2.3 Design criteria – Maximum D15

Base Soil Filter Criteria –Maximum D15
Category
1 9 d85R but not less than 0.2mm
2 0.7mm
3 § 40 − A ·
¨ ¸(4d 85 R − 0.7 mm ) + 0.7 mm but not less than 0.7mm
© 40 − 15 ¹
4 4 d85R
Note that A is fines content after regrading and d85R is regraded d85
2.5 Critical Aspects of Existing Design Practice and Research Focus
The well-known Terzaghi filter criterion, D15/d85 4 or 5, for the design of granular filters
in embankment dams was developed on the basis of laboratory results and conceptual
analyses in early 1920s and still governs the main philosophy of filtration. Filter and base
materials considered for analyses were uniform sands. Consequently, it exhibits some
serious limitations when applied to non-uniform filter and base materials, particularly to
non-uniform base soils. In the early years, this criterion was indiscriminately used to all
types of base soils. Numerous studies have been conducted after that, to better understand

fundamental mechanisms that govern filtration. Even though majority of these studies
ended up with either merely investigating the validity of these filter criteria or extending
them to other soil types, these studies have undoubtedly enhanced overall knowledge of
filtration. Now, it is a common conviction among geotechnical engineers that in filters, the
constriction sizes are more important filtration parameters than the particle sizes, that are
the only parameters considered so far in the filter design practices. The current design
guidelines endeavor to implicitly incorporate all the revelations of these studies through
different procedures such as classification of base soils into various categories based on
fines content, regrading of base PSD curve, use of smaller representative base particle sizes,
and so on. They are still purely empirical and based on the particle sizes. Lately, although
Prof. Indraratna and colleagues conducted several comprehensive constriction-based
studies and have been successful in modeling base particle migration into the filter, it has
still not been possible to incorporate these concepts in the current filter design practices.
However, these studies set up a new trend in filter design based on constriction concepts.
Critical aspects of review in relation to the current design practices are succinctly
summarized below.
2.5.1 Factor of Safety in Filter Designs
The Terzaghi filter criteria were developed on the basis of partly laboratory results and
partly conceptual analyses using uniform sands (ICOLD 1994). The controlling constriction
size in the filter is approximately given by D15/4 (Kenney et al. 1985). In this respect, the
classic Terzaghi retention criterion D15/ d854 (i.e. D15/ 4 d85 ) can be interpreted to mean

that the filter effectiveness can be ensured with substantial conservativeness in a uniform
base soil-filter combination, if there are 15% base particles larger (i.e. d85) than the
controlling constriction size of the filter (i.e. D15/4). The success-failure boundary in the
case of uniform base soils is given by D15/d85 9 (Sherard et al. 1984a), which can be
rearranged as D15/4 (2.25) d85. The number in parenthesis represents the factor of safety,
FoS, associated with the classic Terzaghi retention criterion D15/4<d85 in relation to
filtration of uniform base soils. This implies that the self-filtration can be achieved with less
than 15% (i.e. 15/2.25§ 7%) base particles larger than D15/4 (i.e. D15/4 < d93). However,
the use of the Terzaghi criterion leads to unsafe filter designs when applied to highly well-
graded (i.e. broadly-graded) base soils. In case of well-graded and broadly-graded base
soils, even more than 15 % particles larger than D15/4 are not adequate enough to ensure
filter effectiveness within an acceptable limit of soil loss. For this reason, several studies
recommended smaller representative base particle sizes such as d20, d50, d80 etc depending
on the base soil grading, particularly in the case of highly well-graded base soils (Lafleur
1989) or recommended regrading of base soil PSD for different base particle sizes (i.e.
25.4mm; 4.75mm) (Sherard et al. 1963; NRCS 1994). Further discussion about filter
effectiveness is included in the following subsections.
2.5.2 Filter Effectiveness and Base Soil Grading
There is now enough evidence (Foster and Fell 2001; Lafleur et al. 1989; Honjo and
Veneziano 1989; Karpoff 1955) that filter effectiveness diminishes as base soils become
well-graded. Diminishing filter effectiveness in the case of well-graded base soils can be

comprehensibly explained with the illustration presented in Figure 2.17, where three base
soils with identical d85 but different Cu values are tested against a mechanical sieve with its
aperture size equal to d85 size. In all cases, 85% of base particles are smaller than the sieve
aperture so some base particles smaller than sieve opening (i.e. finer fraction) are initially
lost through it before enough base particles larger than aperture (i.e. coarser fraction) are
retained to initiate self-filtration. However, it is to be noted that although the coarser
fraction in each case is the same 15%, self-filtration in the case of uniform base soil (B1) is
much quicker compared to non-uniform base soil (B2). This is because the finer fraction in
this case is much coarser i.e. particle larger than 0.7mm is 85% whereas the fraction larger
than 0.7mm in B2 is less than 10%. For this reason, initial soil loss is more in the case of
B2 and self-filtration takes longer time. For the same reason, soil loss is even more in the
case of B3 and self-filtration takes even longer. A granular filter, where the screen aperture
is characterized by the controlling constriction size of the filter, is also similar to the
aforementioned case. This explains why filters with high retention ratio (i.e. up to 9) are
effective in the case of uniform base soils whereas when the base soils are broadly-graded,
the Terzaghi criterion leads to unsafe designs. Thus, it can be concluded that each base soil
with a unique grading exhibits a unique filtration characteristic.
In relation to two current design practices described earlier, as these soils are cohensionless
and non-broadly-graded, with the largest base particle size smaller than US sieve no. 4
(4.75mm), regrading is not required i.e. d85 and d85R are identical. Filter designs for all these
base soil by both guidelines require the classic Terzaghi retention criterion, D15/d85 4 or 5,

to be satisfied. Irrespective of D15, retention ratios D15/d85 in all cases are identical,
implying that these three base soils exhibit identical filtration behaviours. However, the fact
is that although these designs may be safe, the filter effectiveness (i.e. factor of safety) is
not the same; it decreases as the base soil grading becomes wider. In this respect, it can be
concluded that the current design practices fail to simulate unique characteristics of the
base soil.
100
80
Base Soils
Percent Finer
60
40
B1
B2
B3
20
B3 B2 B1
0
0.1 1
Particle Size, D (mm)
Figure 2.17 Three base soils with identical d85 but different Cu values
The Terzaghi criterion contains a factor of safety of about 2 in the case of uniform
cohensionless base soils as discussed earlier (Sherard et al. 1984a; Kenney et al. 1985).

With well-graded base soils, this value decreases gradually. The decrease in filter
effectiveness becomes evident only when the designs become unsafe (i.e. FoS <1),
especially in the case of highly well-graded base soils. This is the reason that Foster and
Fell (2001) could not explain several laboratory observations in the case of category 4 base
soils, where the filters with regraded D15/d85 (i.e. D15/d85R) less than 4 also failed. The
empirical guidelines deal with the broadly-graded base soils separately from other well-
graded soils. It is this understanding that introduced the terminology “broadly-graded” in
order to differentiate well-graded base soils, for which the Terzaghi criterion did not hold,
from other well-graded soils because there is no such terminology as ‘broadly-graded’ in
the existing soil classification systems. Furthermore, consider the design of rock slope
protection for an embankment dam presented in Figure 2.18 (Cedergren 1989), where fine
base soils are protected by intermediate layers of sand and gravel. Each layer protects the
adjacent finer layer. The layer of rock spalls (curve 3) is protected against erosion by the
layer of coarse rocks (curve 4). All base particles of rock spalls (curve 3) are larger than
4.75mm so regrading fails to explain this design. This implies that although filter designs
by regrading may be safe in the case of most core soils, this is still not a rational approach.
A base particle of 4.75mm may be too large to influence self-filtration in a fine filter, but
not so for a coarser filter. In this respect, regrading by a fixed value in all base soils
appears to be unrealistic. Similarly, there is invariably a significant difference between the
values of d50 and d80 sizes, particularly in broadly-graded base soils. In this respect, the
Lafleur procedure (i.e. the use of d50 for a required size of d80) also makes the designs
unrealistic.

2.5.3 Representative Filter Particle Size and Filter Grading
After years of experimental investigations and conceptual analyses, D15 suggested by
Terzaghi is still found to be the most convincing parameter to represent a filter and hence
used in the current design practice. Although D15 appears to correlate to the filter
performance, this parameter is only indirectly related to actual filter behaviour. Instead, a
constriction size between the filter particles is a better indicator of filter effectiveness.
Because constriction sizes are not easily measured, most filter criteria use particle sizes,
recognizing that there is a relationship between particle and constriction sizes. Whilst this is
correct, studies such as Sherard et al. (1984a) found that a well-graded filter with Cu>20
can retain a base particle of half the size a uniform filter with the identical D15 can. This
does not mean that well-graded filters with Cu<20 behave as a uniform filter. In fact, filters
with unique grading exhibit unique characteristics. The change is gradual as Cu value
increases.
Indraratna et al. (1996) found the filter permeability having high correlation with finer
particle sizes such D5 and D10. Kenney et al. (1985) found that the controlling constriction
size bears a good correlation with both D5 and D15. In this respect, it can be concluded that
the representative size varies from D5 to D15 or further depending upon the filter gradation.
Although the current design practice (NRCS 1994) exercises control over the filter
gradation in order to prevent segregation during installation and avoid selection of gap-
graded filters, it does not distinguish between uniform and well-graded filters in relation to
their retention characteristics. For example, consider the illustration given in Figure 2.19

where a base soil is tested against two filters with identical D15 values but with different
PSD curves. According to the current design practice, both filters are equally acceptable
irrespective of whether d85 is regraded or not regraded. Thus, the use of D15 involving well-
graded filters may sometimes make designs unduly conservative. Indraratna et al (1990)
found that if the filters are excessively graded, the risk of clogging is introduced.
Figure 2.18 Rock slope protection designed to prevent undermining (Cedergren 1989)

2.5.4 Controlling Constriction Size and Filter Compaction
Past studies (Lafleur et al. 1989) revealed that an ineffective filter at lower density can be
effective at a higher level of compaction. The reason is obvious. The filter constriction or
void sizes change with the degree of its compaction but the particle sizes do not. Besides,
Kenney et al. (1985) suggested two relationships based on D5 and D15 to estimate the
controlling constriction size in a filter. It is not clear which one is more appropriate if two
estimates are significantly different. This implies that with the use of D15 as a representative
filter particle size, the current design practices ignore the filter gradation and its level of
compaction.
2.5.5 Particle Frequency
Although the use of the particle size distribution by mass (as determined by sieving)
appears reasonable to represent the frequency of filter particles in uniform filters, it
introduces errors resulting in over-representation of large pore constrictions in well-graded
filters (DeMello 1977). The particle size distribution by number of filter particles
eliminates large pore constrictions. However, it introduces errors resulting in over-
estimation of smaller pore constrictions due to numerical superiority of fine particles in
well-graded filters (Kenney et al. 1985). In this respect, the use of the particle size
distribution by surface area of filter particles best describes the filtration behaviours of
well-graded filters. This is because although there are only a small number of larger
particles in well-graded filters, they have a great number of contacts with other particles

due to their large surface area (Humes 1996). For this reason, the particle size distribution
by mass should not be used to model the well-graded filters. This aspect is not considered
in the current design practice.
100
F1 F2
85%
80
Base Soil
Percent Finer
60
40
20 15% Filters
0
0.01 0.1 1 10 100
Figure 2.19 A base soil tested against two filters with identical
D15 but different Cu values
2.5.6 Filtration of Cohesive Base Soils
Sherard et al. (1984b) found that in the case of fine silts and clays, the success-failure
boundary was found to be D15/d859; none of the tests with retention ration less than 9
failed; some tests were found stable even with retention ratio as high as 50. Considering the
lowest observed value as the success-failure boundary, Sherard and Dunnigan (1985)

recommended D15/d859 as the design criterion for fine silts and clays. In contrast, Sherard
and Dunnigan (1989) found that the tests failed with D15/d85 from 6 to 14. However, the
study recommended the same D15/d859 as the design criterion considering the average
value of two limits. Foster and Fell (2001) also obtained observations similar to Sherard
and Dunnigan (1989) and still recommended the same D15/d859 as the design criterion,
considering the mean of two limiting values rather than the lowest value (i.e. 6). Sherard
and colleagues did not notice any significant difference between dispersive and non-
dispersive soils in relation to their filtration behaviours. However, Foster and Fell (2001)
recommended using the smaller boundary given by D15/d856.4 for dispersive soils in
category 1 and D150.5mm for dispersive soils in category 2. As discussed earlier, Lafleur
design procedure (ICOLD 1994) recommends D150.4mm for non-dispersive cohesive base
soils and D150.2mm for dispersive base soils. This implies that there is ambiguity in the
design procedure for cohesive base soils and needs further clarification.
2.5.7 Numerical Modelling of Particle Migration
There is no doubt that the numerical models such as Indraratna and Vafai (1997) and Locke
et al. (2001) are the initial framework of simulation of base and filter particle interaction
and represent the ultimate goal a geotechnical engineer involved in the filtration analysis
would wish to achieve. Because of a large number of parameters and complex interaction
of particles, these models are still not final and should be updated as more investigations
enhance the knowledge of filtration. As discussed earlier, although these models do not
provide any direct filter criteria to evaluate filter effectiveness, they certainly provide a

picture of whole process occurring within the dam. In fact, these models form the basis of
constriction-based study this research study is all about. However, updating these numerical
models is not the focus of this research study.
All of the above critical aspects of the current design practices are addressed in this
research study by extending the constriction concepts of the filter established earlier by the
studies such as Indraratna and Vafai (1997), Indraratna and Locke (2000) and Locke et al.
(2001).

CHAPTER
THREE
MATHMATICAL MODELLING OF FILTERS
3.1 Introduction
After years of experimental investigations and conceptual analyses, D15 suggested by
Terzaghi is still found to be the most convincing parameter to represent a filter and hence
used in the current design practice. Here, D15 is the filter particle size where 15% of the
partilces are finer than the size. Although D15 appears to correlate to the filter
performance, this parameter is only indirectly related to actual filter behaviour (Fischer
and Holtz 1996). As discussed in Chapter 2, there are some obvious limitations
associated with the use of D15. Instead, a criterion based on a constriction size among the
filter particles may be a better parameter (Kenney et al. 1985; Indraratna and Vafai 1997;
Locke et al. 2001). However, because constriction sizes are not easily measured, most
filter criteria consider particle sizes, recognising that there is a relationship between
particle and constriction sizes. The use of D15 as a representative filter particle size
ignores filter compaction and the surface area of the particle, which are found to be
significant parameters in filtration analyses.
Chapter 3: Mathematical Modelling of Filters 80

Indraratna and Locke (2000) and Locke et al. (2001) formulated the basis of
mathematical modelling of filters. Based on the previous work on the constriction sizes,
these studies developed a detailed analytical procedure to compute the constriction size
distribution (CSD) of the filter. This research study primarily extends the same
theoretical concepts to describe filters in more detail. Underlying theoretical concepts are
discussed in depth and coded into a comprehensive computer program. The program is
calibrated against some well-known published laboratory data. Finally, an analytical
model is presented to determine the controlling constriction size of the filter, which is
also verified by past laboratory observations.
3.2 Constriction Sizes in the Densest Particle Arrangements
In a real granular filter, particles exist in a group of three or four, representing the densest
and the loosest arrangements, respectively (Silveira 1965; Silveira et al. 1975; Giroud
1996). Silveira (1965) assumed that in a filter at maximum density, only the densest
arrangements exist, and defined the constriction size DcD as the diameter of largest circle
that can fit within three tangent filter particles (Figure 3.1), which can be given by:
((D1 + D2 + D3 ) D1 D2 D3 )1 / 2 = (( D1 + D2 + DcD )D1 D2 DcD )1 / 2

+ (( D1 + DcD + D3 ) D1 DcD D3 )
1/ 2
(3.1)
+ (( DcD + D2 + D3 ) DcD D2 D3 )
1/ 2
Alternatively, the constriction size DcD in the densest arrangements can also be computed
by (Humes 1996):

2
ª§ 2 · § 2 ·º
2 2 2 2
§ 2 · § 2 · § 2 · § 2 · · § 2 · § 2
¨¨ ¸¸ + ¨¨ ¸¸ + ¨¨ ¸¸ + ¨¨ ¸¸ = 0.5«¨¨ ¸¸ + ¨¨ ¸¸ + ¨¨ ¸¸ + ¨¨ ¸¸» (3.2)
© D1 ¹ © D2 ¹ © D3 ¹ © DcD ¹ ¬© D1 ¹ © D2 ¹ © D3 ¹ © DcD ¹¼
The frequency (i.e. probability of occurrence) pc of the constriction size DcD depends on
individual frequencies of these three particles constituting the group, and can be given
by:
3!
pc = p1 p2 p3 (3.3)
r1!r2 !r3!
where, p1, p2, and p3 are the frequencies of the particles 1, 2, and 3, respectively and r1,
r2 and r3 = 0,1,2, or 3 depending upon the number of times particles of the same size
appear in the group, such that r1+r2+r3 = 3. If the filter particle size distribution (PSD) is
divided into n descretized particle sizes D1, D2, D3, … , Dn as shown in Figure 3.2, the
total number of unique groups constituting the constrictions C n ,3 is mathematically
given by:
(n + 3 − 1)!
C n,3 = (3.4)
3!(n − 1)!
For example, consider that a filter PSD as shown in Figure 3.2 is divided into 10 different
particle sizes (n = 10). Then, there are 220 unique groups of three particles for which
Dc’s and pc’s can be determined. Subsequently, Dc’s along with pc’s are sorted in order of
increasing size and pc is cumulated to give the constriction size distribution in the densest
model i.e. CSD(D). From this, one can interpolate Dc corresponding to any value of

frequency Pc of interest. Ideally, more than 10 divisions are not required even in the case
of a very well-graded filter.
D1 Constriction
Size (D
(DccD))
D2
D3
Figure 3.1 The densest filter particle arrangement

pmn
p=1-Pc
pm...
Percent Finer
Filter CSD
Filter PSD
by Mass
pm2
Pc
pm1
Dc D1 D2 D... Dn
Constriction Size, Dc; Particle Size, D
Figure 3.2 A typical filter particle size distribution (PSD) and constriction
size distribution (CSD) showing passing probability p ( = 1 − Pc )

3.3 Constriction Sizes in the Loosest Particle Arrangements
As a real filter is not always compacted to its maximum density, which implies that the
densest constriction model is conservative. For a loose particle arrangement, according to
Silveria (1975) the constriction space, Sc, between four particles (Figure 3.3) is given by:
Sc =
1
8
( (
(D1 + D2 )(D1 + D4 )sin α + (D2 + D3 )(D2 + D4 )sin γ − αD12 + β D22 + γD32 + δD42 )) (3.5)
where the angles β, γ and δ can be related to α by plane geometry. The angle α varies
between αmin and αmax (Figures 3.4a and 3.4b). For a particular value of α between these
two extreme values, when the value of Sc is maximum, the corresponding constriction
size in the loosest arrangement based on equivalent diameter DcL is given by:
4S c , max
DcL = (3.6)
π
However, unlike the densest model, constrictions do not usually form on a plane through
the centres of all four particles. Schuler (1996) suggested that the mean of all possible
chord lengths through the spherical particle be used to represent apparent particle
diameter, which is about 0.82 × actual diameter. Silveira (1975) ignored this distortion.

D1
D2
DD
DcLccD
D4
D3 SV
Sc
Figure 3.3 The loosest filter particle arrangement
Again, the frequency pc of the constriction size DcL depends upon the individual
frequencies of four filter particles constituting the group. This can be given by:
4!
pc = p1 p2 p3 p4 (3.7)
r1! r2 ! r3! r4 !
where p1, p2, p3, and p4 are the frequencies of the particles 1, 2, 3, and 4, respectively,
and r1, r2, r3 and r4 = 0, 1, 2, 3 or 4 depending upon the number of times particles of the
same size appear in the arrangement, such that r1+r2+r3 +r4 = 4. The total number of
unique groups constituting the constrictions, C n , 4 , is given by:
(n + 4 − 1)!
Cn, 4 = (3.8)
4!(n − 1)!
For 10 different particle sizes, for example, there are 715 unique groups of four particles.
In a similar fashion as explained earlier, the constriction size distributions in the loosest
model, CSD(L), can also be determined.

D1
D1
α m in α ma x
D2 D2
D4 D4
D3
D3
(a) (b)
Figure 3.4 Particle arrangements for α to be (a) minimum and (b) maximum
3.4 Particle Frequency and Filter Compaction
For simplicity, most researchers have used the densest CSD for constriction-based
analyses, where the filter PSD either based on mass or based on number of particles has
been used. As explained by Locke et al. (2001), although PSD by mass obtained through
sieve analysis is accepted as a good representation of CSD for uniform filters, the use of
PSD by mass introduces some errors in well-graded filters. This is because large particles
with a high individual mass but low in number will be over-represented, as it is unlikely
that these few large particles will meet together to form a large constriction. In a similar
manner, the PSD by number over-represents the finer constrictions. Humes (1996)
suggested that although there are only a small number of large particles, they impose
significant contact with other particles due to their larger surface area, and showed that
the filter PSD based on surface area is the best option for filtration analysis. If a filter
material is composed of n diameters, D1, D2, D3, … , Dn and their mass frequencies are
pm1, pm2, pm3,…, pmn, respectively (Figure 3.2), then their respective frequencies by

surface area ( p SAi ) can be obtained as follows (see also Humes 1996). Consider that the
specific gravity of the filter particles are the same and
Volume of filter particles =V
Mass/volume frequency of filter particles in the class i
(as shown in Figure 3.2) = pmi
Total volume of filter particles in the class i = pmi .V
( p mi .V )
Total number of filter particles in the class i, Ni =
(πD i
3
/6 )
Total surface area of the particles in the class i, SAi = Ni. πDi ( 2
)
( pmi .V )
=
(πD 3
/ 6)
(πD ) i
2
p mi
= 6V .
Di
i =n
Total surface area of the filter particles in total volume V = ¦ SA
i =1
i
i=n
p mi
= 6V. ¦D i =1 i
Particle frequency by surface area of the filter particles
SAi
in class i, pSAi = i=n
¦ SA
i =1
i
p mi
D
= i =n i
p mi
¦
i =1 Di

Thus, given the particle size distribution based on mass or volume after sieving (Figure
3.2), the particle size distribution based on surface area can be determined by Equation
(3.9).
p mi
D
p SAi = i=n i (3.9)
p mi
¦
i =1 Di
Similarly, their frequencies by number ( p Ni ) can be obtained as follows.
i =n
Total number of filter particles in volume V = ¦N
i =1
i
Ni
Particle frequency by number of particles in class i, pNi = i=n
¦N
i =1
i
p mi
3
Di
= i=n
p mi
¦D
i =1
3
i
Thus, given the particle size distribution by mass or volume after sieving (Figure 3.2),
the particle size distribution by number of filter particles can be also determined by
Equation (3.10).
pmi
p Ni = i = nDi (3.10)
p mi
¦
i =1 Di

Besides, real filters are likely to exist in between two extreme states-the loosest and the
densest. Schuler (1996) found that the CSD curves of a granular material have similar
shapes at different relative densities. Giroud (1996) found that the densest particle
arrangements exist in certain locations even in medium dense soils; hence the smallest
constriction size remains the same, regardless of filter density. Based on these two
observations, irrespective of whether the CSD is determined by mass, number or surface
area, the actual constriction size Dc for any given relative density Rd is given by
Equation (3.11) (Locke et al. 2001).
Dc = DcD + Pc (1 − Rd )(DcL − DcD ) (3.11)
where, Dc is the actual constriction size for a given value of the percent finer Pc; DcD
and DcL are the constriction sizes in the densest and the loosest models, respectively for
the same Pc. The authors have incorporated these theoretical concepts in a
comprehensive computer program, which computer the filter CSD by all three frequency
considerations such as mass, number and surface area in order to study the contrast
between various frequency concepts. However, to compute the actual filter CSD, only
the surface area option should be used. Accordingly, only the CSD computation
procedure by surface area is depicted in the flow chart diagram presented in Figure 3.5.
3.5 Programming Algorithms
The formulation of algorithms is an important routine step in any program development,
particularly in complex engineering numerical problems like this study. The
computational aspect of constriction sizes has not been dealt with in detail in the past

studies, Although Indraratna and Locke (2000) and Locke et al. (2001) established a
computational procedure for determining the CSD, these studies do not elaborate on
algorithm aspects. Moreover, the procedure was developed as a Visual Basic macro
subroutine of another program developed on SPREADSHEET platform. It is difficult to
use it as a stand alone program for CSD computation. In this study, an exclusive program
for CSD computation has been developed. Some important steps depicted in the flow
chart (Figure 3.5) are briefly described in this section.
3.5.1 Number of Unique Groups Constituting Constrictions
If the filter PSD is divided into n descretized particle sizes, the total number of possible
unique groups constituting the constriction is n3 or n4 depending on the dense or loose
model respectively. However, with repetitions permitted, only the combinations of n
particles taken m at a time C nr,m given by Equation (3.12) (i.e. the generalized form of
Equations (3.4) and (3.8)), are unique in terms of effective constriction area and the rest
are permutations. The program must be able to exclude these permutations.
(n + m − 1)!
Cnr, m = (3.12)
m!(n − 1)!

Input:
Filter PSD by mass (after sieve analysis)
Filter Relative Density Rd
Pc (percent finer) values of interest
Descretize filter PSD into classes of particle size

Calculate mean particle size of each class (geometric)
Calculate mass frequency of each class
Calculate Filter PSD by surface area
Calculate apparent particle size
Calculate Cn,3 Calculate Cn,4

Calculate Dc’s Calculate Dc’s
Calculate corresponding pc’s Calculate corresponding pc’s
Sort Dc’s in ascending order Sort Dc’s in ascending order

Cumulate pc’s (check: sum of pc’s =1) Cumulate pc’s (check: sum of pc’s =1)
Constriction Size Distribution Constriction Size Distribution

(Densest) (Loosest)
Interpolate Dc’s for Pc’s of interest Interpolate Dc’s for Pc’s of interest
(DcD) (DcL)
Combine Dc’s based on relative density Rd to compute Dc,Pc’s for Pc’s of interest
Output:
Filter Constriction Size Distribution (CSD)
Figure 3.5 Flow chart for the detailed CSD computational procedure

3.5.2 Constriction Size Computation
For the densest arrangement, Equation (3.1) is solved for the constriction size DcD by the
bisection method of iteration between two extreme values given by the constriction size
formed by three tangent particles of the smallest particle size in the group and the
constriction size formed by three tangent particles of the biggest particle size in the
group. The iteration is carried out to an accuracy of 10-6mm. For the loosest arrangement,
the angle α corresponding to the biggest particle in the group is assumed and
corresponding angles β, γ and δ are calculated. Subsequently, the constriction space Sc is
calculated by Equation (3.5). It is to be noted that the cosine functions must be used to
calculate angles, particularly when these angles are likely to be obtuse angles. In order to
find the maximum value of Sc, the angle α is increased by some increment and the next
value of Sc is calculated. The iteration is continued until Sc is maximised, and
subsequently DcL will be determined. There are two possible extreme values of α namely
αmin and αmax, as shown in Figures 3.4a and 3.4b. These values can be calculated by
Equations 3.13-3.15.
1/ 2
§α · ª D2 D4 º
tan ¨ min ¸ = « » (3.13)
© 2 ¹ D (
¬ 1 1 D + D 2 + D )
4 ¼
α max = α1 + α 2 , where
1/ 2
§α · ª D2 D3 º
tan ¨ 1 ¸ = « » (3.14)
© ¹ ¬ 1 1
2 D ( D + D 2 + D )
3 ¼

1/ 2
§α · ª D3 D 4 º
(3.15)
tan ¨ 2 ¸ = « »
© 2 ¹ ¬ D1 (D1 + D3 + D4 ) ¼
Silveira et al. (1975) suggested that for all practical purposes, the increment in α could be
equal to 2o. This is true in the case of uniform filters. However, sometimes in well-graded
filters, the difference between αmin and αmax can be less than 2o. This can lead to a less
accurate estimation of constriction space. Consequently, depending upon the difference
between αmin and αmax, smaller incremental value is taken. If the difference is less than 1o,
the increment is taken to be equal to one tenth of the difference otherwise one tenth of a
degree.
3.5.3 Sorting, Cumulating and Interpolating
After calculating the constriction sizes and their respective probabilities of occurrence in
both densest and loosest constriction models, the constriction sizes are sorted in
ascending order. Subsequently, the probabilities are cumulated resulting in the smallest
constriction size with zero percent finer and the biggest constriction size with 100 percent
finer. Hence, CSDs for both densest and loosest constriction models with number of data
points given by Equations (3.4) and (3.8). The constriction sizes are then interpolated for
some desired values of percent finer from both densest and loosest CSDs, resulting in
CSDs with less number of representative data, relatively more data points towards
extremities of the distributions. Because of invariably large number of data, for any value
falling between two data points, the straight line variation is considered for interpolation.
The constriction sizes at a given value of percent finer in two CSDs are finally combined

to incorporate the effect of filter compaction based on Equation (3.11), resulting in the
final CSD of the filter with desired number of data points.
All sub-programs used in the program are developed as a fool-proof program and they
are checked for numerical values obtainable with exclusively analytical considerations.
For example, the constriction sizes for groups such as D1D1D1, D1D1D1D1, D2D2D2,
D2D2D2D2 etc. can be calculated based on purely analytical considerations whereas other
values are checked in relation to these values.
3.6 Worked-Out Example
A worked-out example is given here to demonstrate the CSD computational procedure
based on the computer program developed by the author as described earlier. Consider a
filter given in Figure 3.6. The filter PSD is represented by eleven descretized data points
as given in Table 3.1a (i.e. ten class sizes) and also shown by label markers in Figure 3.6.
The program inputs are given through an input data file called XINPUT.DAT and
outputs are appended in an output data file called XOUTPUT.DAT. Typical input and
output files for the above filter at relative density of Rd = 70% are given in Figures 3.7
and 3.8, respectively. The values of input parameters are given in a single column
without any space. While descretizing the filter PSD, it is to be noted that the first and
last class size should be kept as small as possible so that the mean class size does not
deviate much from the smallest and the largest particle sizes i.e. extreme data points. In
this way, the program also avoids extrapolation for constriction size, very close to
minimum possible constriction size. The program is capable of interpolation only. A
typical program output in graphical form is represented by Figure 3.9 where the CSD of

the given filter is calculated for a relative density of 70%. The typical outputs clearly
demonstrate that the constriction size distributions in the densest and the loosest states of
filter are first determined and then combined together by Equation (3.11) resulting in the
actual filter CSD for a given relative density of filter.
100
80
Percent Finer
60
Filter PSD
40
20
0
0.1 1 10
Particle Size D (mm)
Figure 3.6 Particle size distribution of a filter with 10 particle class sizes

Figure 3.7 A typical input data file XINPUT.DAT
Table 3.1 Filter PSD with eleven data points

Particle Size 1.000 1.004 1.041 1.084 1.129 1.200 1.275 1.355 1.440 1.494 1.500
D (mm)
Percent Finer 0.00 1.00 10.00 20.00 30.00 45.00 60.00 75.00 90.00 99.00 100.00
(%)

Figure 3.8 A typical output data file XOUTPUT.DAT
The question may arise whether the CSD remains the same if the descretization of filter
PSD is different from one given in Table 3.1. In fact, the CSD remains the same unless
the descretized data points modify the filter PSD. This can be demonstrated by
considering different class sizes, given in Table 3.2. Figure 3.10 shows the CSDs of the
filter determined by considering two different descretizations. As expected, it is clearly
demonstrated that descretizations are immaterial so long as the descretized data points do
not modify the filter PSD.
Table 3.2 PSD of the same filter as given in Figure 3.6 but with different descretizations
Particle Size 1.000 1.004 1.020 1.063 1.107 1.176 1.250 1.328 1.411 1.494 1.500
D (mm)
Percent Finer 0.00 1.00 5.00 15.00 25.00 40.00 55.00 70.00 85.00 99.00 100.00
(%)

100
80
Percent Finer CSD (Rd = 70%)

60 Filter PSD
CSD (Densest)
40 CSD (Loosest)
20
0
0.1 1 10
Constriction Size D c ; Particle Size D (mm)
Figure 3.9 A typical CSD program outputs for a uniform filter PSD in graphical form
100
80
Percent Finer
60 CSD 1
CSD 2
Filter PSD
40
20
0
0.1 1 10
Figure 3.10 Filter CSDs computed with the same filter PSD but with two different
descretizations (Rd = 90 %)

3.7 CSD Program Calibration
The computer program is calibrated against established analytical principles (qualitative
verification) and laboratory observations (quantitative verification). Following
subsections compare the model against some well-known laboratory results and
analytical outcomes of past studies.
3.7.1 CSDs of Uniform and Well-graded Filters
For a relative density of 70%, Figure 3.11 illustrates the filter PSD and calculated CSD
curves for uniform and well-graded filters based on mass, number of particles and
surface area. For the uniform filter (Figure 3.11a), the difference among the above three
methods is insignificant, whereas the difference in the PSD and CSD curves for the well-
graded filter (Figure 3.11b) is considerable, depending on the method used. As expected,
this outcome is well in agreement with past research outcomes (Locke et al. 2001;
DeMello 1977; Humes 1996; Schuler 1996; Kenney et al. 1985). For the reasons
explained earlier, the authors have used the surface area method for subsequent
constriction analyses.
3.7.2 Effect of Relative Density on Constriction Size
The CSDs of a uniform filter are determined for various levels of compaction i.e. for
various values of relative density Rd. Typical outputs for various values of relative
density Rd are shown in Figure 3.12. The densest and the loosest CSDs are plotted as
boundaries, and three specific CSD curves corresponding to relative densities (Rd) of 0%,
50% and 70% are determined and plotted. The model outcomes agree well in the research
findings of Giroud (1996) and Schuler (1996) where the former found that the densest

particle arrangements exist even in the loosely compacted filters, while the latter found
that the shape of CSD curves at various compaction levels are similar.
3.7.3 Manual CSD Computations in Past Studies
Silveira (1965) manually calculated, for the first time, the CSD of a filter by mass for the
densest CSD model. The study divided the filter PSD in only five particle size classes in
order to minimise the amount of calculations and considered all different possible dense
particle arrangements. The current study used the computer program to compute the
densest CSD of the same filter based on mass for the densest state and plotted in Figure
3.13 with the manually calculated CSD. In the similar fashion, Soria et al. (1993) also
determined the filter CSD by mass of another filter in the densest state. This is plotted in
Figure 3.14 with the CSD curve determined by using the computer program. As
illustrated, the computed values are in good agreement with the calculated values in these
past studies, validating the model performance.

100
(a) M =Mass
N =Number
80 SA =Surface Area
Percent Finer
60
PSD(M)
Filter CSDs PSD(SA)
PSD(N)
CSD(M)
40 CSD(SA)
CSD(N)
20 Filter PSDs
0
0.1 1 10
Constriction Size, Dc (mm); Particle Size, D (mm)
100
(b)
Filter CSDs
80 N N
SA SA
M
M
Percent Finer
60
40
M=Mass
N=Number
SA=Surface Area
20
Filter PSDs
0
0.1 1 10
Figure 3.11 Filter PSDs and CSDs by mass (M), by number (N) and by surface
area (SA) (a) Uniform Filter, F1 (Cu=1.2, Rd=70%) (b) Non-uniform Filter, F2
(Cu=3.8, Rd=70%)

100
Surface Area
Rd=1.0 L = Loosest
80 0.7
D = Densest
CSD (D) 0.5

0
CSD (L)
Percent Finer 60
40
20
0
0.1 1
Figure 3.12 CSDs of a uniform filter at various levels of compaction

i.e. at different relative densities, Rd
100
CSD (Densest, Mass) CSD( Densest, Mass)
(Silveira) (Current Model)
80
Filter PSD
Percent Finer
60
40
Filter CSDs CSD (Silveria, 1965)

20
Filter PSD
CSD(Current Model)
0
0.01 0.1 1 10
Constriction Size,D c (mm); Particle Size,D (mm)
Figure 3.13 CSDs by mass of a filter in the densest state computed by using the
computer program and manually by Silveira (1965)

3.7.4 Experimental Observations
Based on the probabilistic filtration theory proposed by Silveira (1993), Soria et al.
(1993) carried out several filter experiments by considering the filters of various
thicknesses. Based on laboratory observations, the study back-calculated the filter CSD
for a given PSD. Because of limitations of laboratory equipments, only the middle
fraction of the filter CSD could be determined. Relative density (Rd) of all filters was in
the order of 90%. Humes (1996) calculated the CSD by surface area (SA) for the filters
in the densest state (D) for the same data (i.e. denoted as CSD(D)SA in Figure 3.14). The
CSDs computed by the authors are also shown for comparison. Figure 3.14 illustrates
that these model predictions are in good agreement with the densest model of Humes
(1996) and the experimental findings of Soria et al. (1993) at Rd =90%. All of the above
comparisons and analyses clearly illustrate that the current model is a realistic
representation of the constriction sizes in a granular filter.

100
CSD (Densest, Mass) CSD (Densest, Mass)
(Soria) (Current Model)
80 Filter PSD
Percent Finer
60
40
PSD
20 Filter CSDs
CSD(Soria, 1993)
CSD (Model)
0
0.01 0.1 1 10
Constriction Size, D c (mm); Particle Size, D (mm)
Figure 3.14 CSDs by mass of a filter in the densest state computed by using the
computer program and manually by Soria (1993)
3.8 Controlling Constriction Size
Kenney et al. (1985) used a multi-layered one-dimensional constriction model to
analytically investigate the size of controlling constriction in a filter, defined as the size
of the largest base soil particle that can potentially infiltrate through the filter. Although
this model is a good approximation for uniform filters and provides a sound
understanding of the fundamental filtration mechanisms, it considers the flow channels to
be independent, which is unrealistically restrictive in relation to flows through porous
media. As discussed earlier in Chapter 2, Kenney et al. (1985) suggested two criteria to
evaluate the controlling constriction size in a filter. Because of the fact that these criteria
are based on the specific particle sizes of the filter such as D5 and D15, they have inherent
limitations in relation to filter relative density, Rd, and coefficient of uniformity, Cu.
Moreover, it is not clear which value can be considered to be more realistic, in the case

that the controlling constriction sizes estimated by two criteria are found to be
significantly different. This necessitates a more realistic method to determine the size of
the controlling constriction in a filter. The following subsections deal with the controlling
constriction size of the filter and include the model development and its validation versus
the past laboratory observations.
100
D = Densest State
SA = Surface Area
CSD (Experimental)
80 (Soria et al., 1993)
(Rd=90%)
CSD (Current Model)
Percent Finer
60
CSD(D)SA
(Humes, 1996)
40
CSD(D)SA
(Current Model)
20 Filter PSD
(Soria et al., 1993)
0
0.1 1
Figure 3.15 Model predictions of filter constrictions against a

number of experimental and analytical results
3.8.1 Analytical Concepts
Considering the 3D pore network model and the possible sideways exits available for the
base soil particles, Locke et al. (2001) found an increased value of probability of forward
movement P (F ) corresponding to the value of passing probability p (percentage of

larger constrictions = 1 − Pc ), as shown earlier in Figure 3.2. The relevant algebraic
relationship between P(F) and p is given by Equation (3.16).
[ ] {[ ] }
∞
P(F ) = p + ¦ 1 − (1 − p ) (1 − p ) p 1 − (1 − p ) (1 − p )
4 3 i
(3.16)
i =0
The above equation can be simplified based on the limiting sum of the infinite geometric
series as follows. On rearrangement of terms, Equation (3.16) can be written as:
[ ] {[ ] }
∞
P(F ) = p + 1 − (1 − p ) (1 − p ) p ¦ 1 − (1 − p ) (1 − p )
4 3 i
(3.17)
i =0
Substituting values of i=0, 1, 2….to infinity and by further simplification, Equation
(3.18) and (3.19) can be obtained as follows.
{[
ª 1 − (1 − p )3 (1 − p ) 0
«
] º
»
}
{[
« + 1 − (1 − p )3 (1 − p ) 1
«
] »
»
}
[
P(F ) = p + 1 − (1 − p )
4
] « {[
(1 − p ) p + 1 − (1 − p ) (1 − p )
«
3
]
2
»
»
} (3.18)
« + ............................. »
« ∞»
« {[
+ 1 − (1 − p ) (1 − p ) »
3
] }
¬ ¼

ª1 º
« »
{[
« + 1 − (1 − p ) (1 − p ) »
3
] 1
}
« 2»
[
P(F ) = p + 1 − (1 − p )
4
] {[
(1 − p ) p « + 1 − (1 − p )3 (1 − p ) » ] } (3.19)
« + ............................. »
« »
« + to ∞ terms »
¬« ¼»
{[ ]
As p is always less than or equal to 1, the term 1 − (1 − p ) (1 − p ) is always less than 1
3
}
and consequently, the series represented by the terms inside the large square bracket can
be considered as infinite geometric series with common ratio of 1 − (1 − p ) (1 − p ) . On {[ 3

] }
simplification of Equation (3.19), one may finally obtain Equation (3.20).
P (F ) = p +
{1 − (1 − p ) }(1 − p ) p
4
1 − {1 − (1 − p ) }(1 − p )
3
(3.20)
Replacing p by ( 1 − Pc ) (Figure 3.2) in Equation (3.20) and simplifying it further, gives
Equation (3.21).
(1 − Pc )(1 + Pc − Pc )
4 5
P( F ) = (3.21)
(1 − Pc + Pc )
4
For a confidence level of P , the number of layers (forward exits) n that a base particle
with a passing probability, p, can infiltrate in the filter before it is captured by a smaller
constriction is given by Equation (3.22) (Locke et al. 2001; Silveira 1965).
n=
(
ln 1 − P ) (3.22)
ln P(F )

The corresponding distance S that the same base particle travels in the filter having a
mean particle size, Dm, can now be determined by Equation (3.23).
S=
(
ln 1 − P )
.Dm = n.Dm (3.23)
ln P(F )
where,
1
Dm = (3.24)
n p SA,i
¦i =1 Di
Figure 3.16 represents the relationships given by Equations (3.21) and (3.22) for a
confidence level ( P ) of 95%. Indraratna and Locke (2000), and Locke et al. (2001)
found a constriction model less sensitive to the choice of P , and adopted P = 95%. It is
noted that the value of n becomes exceedingly high when P ( F ) in Equation (3.22)
approaches unity as Pc < 35%. In other words, the rapidly increasing nature of n-curve
for Pc < 35% clearly indicates that any further increase in filter thickness beyond the
value of 225Dm does not contribute to base soil retention significantly. For all practical
purposes, a base soil particle smaller than Dc 35 may not be retained by a granular filter,
unless the constrictions become progressively finer by self-filtration. Therefore, the
authors propose that the controlling constriction in a granular filter can be given by the
specific constriction size ( Dc 35 ).

1
1000
Prob. of Forward Movement, P(F)

0.8
800
No. of Layers, n
0.6
n 600
( P = 95%) P(F)
0.4
400
n = 225
0.2 200
Pc = 35%
0 0
20 40 60 80 100
Percent Finer (Pc)
Figure 3.16 Probability of forward movement and predicted depth of infiltration
3.8.2 Experimental Verification
A number of past studies (e.g. Kenney et al. 1985; Witt 1993; Sherard et al. 1984a;
Foster and Fell 2001) adopted various experimental procedures to determine the size of
the largest base particles that can potentially penetrate a given filter. This subsection
basically compares the sizes of the controlling constrictions in various filters estimated
by the Dc35, and various other procedures.
The particle size distributions for five filters (F1-F5) with varying Cu and D15 values are
shown in Figure 3. 17a. In the controlling constriction analysis of Kenney et al. (1985),
the uniform filters such as F1, F4, and F5 were compacted to a relative density of about
70%, whereas the non-uniform filters F2 and F3 were compacted to Rd = 90%. The CSDs
of these filters are computed for similar values of Rd and presented in Figure 3. 17b. Witt

(1993) experimentally measured the constriction sizes from imprints of filter particles on
silicon rubbers. The controlling constrictions of these filters have been calculated by two
different procedures and compared with the Dc 35 values in Table 3.3. In general, the
authors’ Dc35 values except for F3 are similar to the Kenney et al. (1985)
recommendation of Dc* ≤ 0.20 D15, but they are consistently smaller than the value of
dp* (effective opening size) given by Witt (1993). One possible reason for this deviation
is that Witt’s approach for calculating dp* does not include the role of Rd. The
controlling constriction size in well-graded filter F3 ( Cu =7) is significantly smaller than
that in F2 ( Cu = 3.8), and this explains the discrepancy between Kenney’s Dc* = 0.210
mm and the authors’ Dc35 = 0.178 mm for the filter F3 (Table 2). It is of interest to note
that for less uniform or well-graded filters, as in the case of F2 (Cu = 3.8) and F3 (Cu =
7.0), Kenney’s second recommendation based on D5 (i.e. Dc* ≤ 0.25 D5) is in perfect
agreement with the authors’ Dc35 as indicated in Table 3.3. This also verifies that as
expected, the controlling constriction sizes in non-uniform filters are smaller than those
in uniform filters for the same D15 and for a given level of compaction (Sherard et al.
1984a).

100
F2
(a)
(Cu=3.8)
F5
(Cu=1.3)
80 F1
(Cu=1.2)
F3
(Cu=7.0)
Percent Finer
60 F4
(Cu=1.2)
40
20
D15
0
0.1 1 10
100
(b)
80
Percent Finer
60 F5 F4
F3
F2
F1
40
Dc35(F3)
20
Dc35(F2)
Dc35(F4)
Dc35(F5) Dc35(F1)
0
0.01 0.1 1
Constriction Size, Dc (mm)
Figure 3.17 Controlling constrictions in filters with varying uniformity

coefficients (Cu) and D15 sizes (a) Filter PSDs and (b) Filter CSDs

Table 3.3 Controlling constrictions by past procedures and the current model
Filters
Studies Filter Parameters
F1 F2 F3 F4 F5
0.20 D 15 0.213 0.209 0.210 0.095 0.023
Kenney et al.(1985)
0.25 D 5 0.255 0.200 0.177 0.115 0.028
Witt (1993) dp* 0.240 0.210 0.198 0.108 0.026
Foster and Fell (2001) Median (0.16D 15 ) 0.167 0.167 0.168 0.076 0.019
Sherard et al.(1984a) 0.18D 15 0.191 0.188 0.189 0.086 0.021
D c35 0.215 0.200 0.178 0.096 0.024
Current Study
D c35 (R d =100% ) 0.181 0.188 0.167 0.081 0.020
dp*=effective opening size
+ all dimensions in mm
Foster and Fell (2001) measured the size of base particles washed through the filters, and
found an upper bound value of 0.20D15 with a median size of 0.16D15. Using molten wax
technique, Sherard et al. (1984a) measured the dimension of minimum flow channel and
found 0.18D15 as an upper bound for the effective opening size. Both these studies used
highly compacted filters with a relative density (Rd) approaching 100%. The Dc35 data
presented in this study (Table 2) include Rd of about 70% for uniform filters and 90% for
well-graded filters, similar to those used by Kenney et al. (1985). However, the Dc35
values based on Rd =100% can be directly compared with those of Foster and Fell (2001)
and Sherard et al. (1984a). In Table 3.2, the values of Dc35 at Rd of 100% are also
tabulated and compared with the median size (0.16D15) of eroded base particles proposed
by Foster and Fell (2001), and the upper bound (0.18D15) of Sherard et al. (1984a). It is
seen that the values of Dc35 (at Rd = 100%) are only slightly larger than the median size of
eroding base particles (0.16D15), and very close to the above stated upper bounds.
Similarly, Figure 3.18 presents all fourteen filters used by Sherard et al. (1984a) for
filtration analysis of sand and gravel. Most of these filters are uniform with Cu<3, except

the filter #1, which is well-graded with Cu=6.5. The CSDs for all these filters are
computed using Rd=90% and compared with the upper bound of base particles size
eroded through the filters (i.e. 0.18D15), as shown in Figure 3.19. The comparison shows
high correlation between these two calculations, showing a trend of Dc35=0.18D15. As
expected, the filter #1 being well-graded shows comparatively a smaller value of Dc35,
resulting in the slightly inaccurate trend. It is to be noted that the correlation will change
if the CSDs are computed at a compaction level other than Rd=90%. Bigger values of
Dc35 are expected at a lower compaction levels whereas small values at a higher levels.
100
80
Percent Finer
60 9
7 8
40
20 10 4 5 13
14
12
1 2 3 11
0
0.1 1 10 100
Figure 3.18 Filters used by Sherard et al. (1984a) for filtration of sand and gravel

3
(0.18D 15 ) = 0.9964(D c35 ) - 0.0066
R2 = 0.9933
2
0.18D15 (mm)
0
0 1 2 3
D c35 (mm)
Figure 3.19 Comparison between Dc35 of the current model and the upper bound of the
base particles eroded through filters as observed by Sherard et al. (1984a)
3.9 Filter Thickness
As explained earlier (Figure 3.15), the current model suggests a minimum filter thickness
should be about 225Dm for a 95% confidence level. Here, Dm is the mean particle size of
the filter. A smaller thickness can be obtained for a reduced confidence level or if a
coarser base soil is considered. Given that the computation of Dm is based on surface area
approach, it varies from D5 to D15 in most practical dam filters. In this respect, a filter
thickness of 225Dm is in agreement with the laboratory observations of 300D5 to 300D10
as suggested by Witt (1993) and 200D5 (Kenney et al. 1985). For typical filter gradations
(e.g. ICOLD, 1994), all these values vary in the range of 40-60mm and may be used as
preliminary guidance in the design of filters. Some may argue that the laboratory
observations show a value smaller than 225Dm. This is because in practice, the

laboratory values are based on eye estimations, where the researchers look at the amount
of base particles retained in the filters after tests. Then they decide that filter beyond a
certain depth is ‘practically’ redundant because there is no significant amount of base
particles retained beyond this depth. This method is subjective and very often, involves
personal bias. There is no established guideline, to date in literature, to determine the
filter thickness. In this respect, this proposition is analytical and bears a good agreement
with well-known laboratory observations. In practice, the thickness of dam filters is
usually much greater than the above mentioned values. For both construction feasibility
and structural stability, the actual thickness of dam filters often exceeds 500 mm
(ICOLD, 1994). The proposed value can be used in preliminary filter designs as a
minimum filter thickness.
3.10 Summary of Constriction Modelling
The use of D15 introduces some deficiencies in filter designs, which need to be rectified.
This can be done by introducing constriction concepts in the current design procedures.
In order to do so, it is important to determine the constriction size distribution of the
filter. This chapter is dedicated to various aspects of constriction modelling, which can be
summarized as follows.
• Despite the fact that the particle frequency based on mass or number of particles
introduces errors in well-graded filters, it is still a normal practice. The filter is
modelled the best by the surface area approach rather than by mass or number of
particles, which should be used in the current practice.
• The constriction model originally developed by Locke et al. (2001) and now
extended by this study, describes the filter in a more reasonable manner.

• The one-dimensional void network models, originally suggested by Silveira (1965)
are restrictive, where the base particles are assumed to be retained in the pore if
stopped at bottom exit. In this respect, the three-dimensional void network adopted
by Locke et al. (2001) is more reasonable, where the base particles are assumed to
take a side exit through larger constriction if stopped at the bottom exit.
• In contrast to Kenney et al. (1985) empirical procedures, Dc35 is an analytical and
more accurate method to estimate the controlling constriction size in the filter. The
Dc35 method distinguishes between uniform and well-graded filters by estimating
comparatively smaller in well-graded filters.
• Minimum filter thickness given by 225Dm is a reasonable estimation and can be
used in preliminary filter designs.

CHAPTER
FOUR
SURFACE AREA CONCEPT APPLIED TO BASE SOILS
4.1 Introduction
As discussed earlier in chapter 2, although known to be conservative and originally
developed for cohensionless uniform base soil and filter materials, the well-known
Terzaghi retention criterion (USACE 1953), D15/d85 4-5 is still used for some simplified
designs. Here, D15 is the filter particle size where 15% particles are finer than the size
and d85 is the base particle size where 85% particles are finer than the size. Several past
studies (e.g. Sherard et al. 1984a; Bertram 1940) revealed that filters even with higher
values of D15/d85 such as 9 can be effective, especially in the case of uniform base soils.
In contrast, studies conducted by Lafleur (1984) demonstrated that some filters with
retention ratios smaller than 5 involving non-uniform or well-graded base soils were
ineffective. To address this effect of diminishing filter effectiveness in the case of well-
graded base soils, the current design practice (NRCS 1994) recommends the use of d85
after regrading the base soil PSD for particles larger than 4.75mm (the number 4 ASTM
standard sieve ) i.e., d85R rather than the conventional d85 without regrading. ICOLD
(1994) also suggests the use of a smaller base soil representative size such as d50.
Chapter 4: Surface Area Concepts Applied to Base Soils 117

However, these practices are purely empirical. As illustrated in the summary section of
literature review, these practices fail to describe the filtration in many cases.
Honjo and Veneziano (1989) carried out a statistical analysis on various test data and
found that the reliability of filters diminished for non-uniform base soils. However, such
statistical analyses do not explain the fundamental physics of filtration, and are not
always free from bias inherent in experimental procedures. For example, consider three
different base soils (B1, B2 and B3) having the same d 85 tested against three different
filters (F1, F2 and F3) having the same D15 (Figure 4.1). All base soils have the largest
base soil particle sizes smaller than 4.75mm so these base soil and filter arrangements
have identical D15/d85 or D15/d85R ratios. The question is whether these base soil-filter
systems have similar filtration characteristics in terms of mass retention and flow rates.
In other words, is the D15/d85 or D15/d85R ratio on its own adequate to describe the filter
effectiveness? Locke et al. (2001) highlighted that the evaluation of filter effectiveness
based on the constriction size distribution is more appropriate than the sole use of particle
sizes. Moreover, it is now clear that the particle frequency based on surface area is the
best approach to model the porous granular media. This chapter presents a novel filter
criterion based on new constriction concepts discussed earlier, particularly applying the
surface area method to the base soils.
4.2 Modelling of Base Soils
It is evident from Figures 1.2 and 2.18 that finer layers are protected against erosion by
adjacent coarser layers. In this respect, the same granular soil can sometimes
simultaneously function both as a filter and a base soil depending on the location of

placement. As explained and illustrated in Chapter 3, there is no doubt that filters are best
modelled by the PSD based on the surface area of particles. Eroded base soil particles
are transported to the filter, making the filter constrictions smaller upon retention, and
thereby initiating self-filtration. In this respect, the base soil can also be modelled by the
PSD based on surface area similar to filters. Hereafter, all filter design parameters D15,
d85 and PSD based on surface area will be denoted by D15SA, d85SA and PSDSA,
respectively. Consider three base soils having the same d85 by mass of 0.80mm and
different Cu values (Figure 4.1) filtered through a mechanical sieve of aperture equal to
0.8mm (Figure 4.2). Except for the very uniform base soil B1 where d85SA = d85, for less
uniform soils B2 and B3, the values of d85SA are less than that of d85 or the sieve aperture.
In other words, although the three base soils have the same d85 (Figure 4.1), only the base
soil B1 has effectively 15% base soil particles larger than the sieve opening (Figure 4.2).
As the base soil becomes less uniform, increasingly smaller amounts of base particles
remain larger than the sieve aperture of 0.80mm (i.e d85). In general, it may be concluded
that d85SA should become considerably smaller from d85SA = 0.80mm to d85SA = 0.37mm
as the uniformity coefficient (Cu) of the base soil increases from 1.4 to 4.0. This explains
why the filter effectiveness tends to decrease as the base soil becomes increasingly non-
uniform as observed in various past studies (Honjo and Veneziano 1989; Lafleur 1984;
Foster and Fell 2001).
As another example, consider a well-graded base soil tested by Lafleur (1984). The study
found that the base soil particles larger than the sieve No. 4 size (4.75mm) do not
influence filtration. The PSDSA of the base soil B-3 is computed and plotted in Figure
4.3. It is noted that based on the PSDSA curve, the larger base soil particles have

insignificant representation. In fact, as these larger particles ‘float’ in the matrix of finer
grains, the influence of these large base particles on self-filtration is insignificant.
100
d85
80 F1 F2 F3
Base Soils
Percent Finer
60
40 Filters
20
B3 B2 B1
D15
0
0.01 0.1 1 10 100
Figure 4.1 Base soils and filters with various uniformity coefficients
(Cu) but having the same retention ratio (D15/d85)

100
85%
80 PSDs of
Base Soils
by Surface Area
Percent Finer d85SA
60 of
B1-B3
40
20 B3SA B2SA B1SA

(Cu=4.0) (Cu=2.7) (Cu=1.1)
Sieve
Aperture
0
0.1 1
Figure 4.2 PSDSA of base soils of different uniformity coefficients
100
85%
80
PSDSA (B-3)
Percent Finer
60
PSD(B-3)
by mass
(Cu=9)
40
20
d85SA Sieve #4
(4.75 mm) d85
0
0.01 0.1 1 10 100
Figure 4.3 PSDSA of well-graded base soil B-3

4.3 Development of Filter Retention Criterion
Several past studies including Honjo and Veneziano (1989) investigated the filtration
process using mechanical sieves as filters, and they revealed that the sieve can be
effective in retaining the base soils only if there are at least 15% of base soil particles
larger than the sieve aperture. These investigations were mainly carried out on uniform
base soils and filters. Lafleur (1984) showed that self-filtration takes a longer time in the
case of non-uniform base soils and that the soil loss is excessive if the filter is designed
based on the Terzaghi’s criterion. As shown in Figure 4.2, the use of conventional d85 for
non-uniform soils does not ensure that at least 15% of the base soil particles are retained,
whereas the use of d85SA (which is generally smaller than d85) invariably satisfies this
condition. Although a granular filter of randomly compacted particles is more complex
than a regular mechanical sieve, it can still be considered as equivalent to a sieve with
apertures equal to the controlling constriction size (Dc35). Thus, for an effective base soil-
filter combination, Dc35 must be smaller than d85SA to ensure that at least 15% base
particles are available to initiate and sustain self-filtration, hence:
Dc 35
<1 (4.1)
d 85 SA
The above constriction-based criterion for base soil retention is comprehensive as it takes
into consideration an array of fundamental filter parameters including PSD, CSD, Cu and
Rd, in comparison with the single filter grain size of D15 in the Terzaghi’s criterion. As
the surface area concept is applicable to cohensionless granular soils, this criterion is
applicable exclusively to the category 4 base soils i.e. gravels and sands (NRCS 1994).

4.4 Verification of the Model Based on Experimental Data
4.4.1 Series A: Very Uniform Base Soils and Filters
As a preliminary example, Figure 4.4 presents three very uniform sand filters (F4, F5, and
F6) and a base soil (fine sand) all of parallel gradation (Cu=1.2). The relevant filter and
base soil parameters are also given in Table 4.1. It is noted that with such a uniform base
soil, the filter with a retention ratio of 5 or less is more effective at higher relative
densities. Because Dc 35 (F4) < d85SA, F4 is considered effective, whereas F6 is deemed to
be ineffective as Dc 35 (F6)> d85SA. It is shown that the magnitude of Dc 35 (F5) is almost the
same as d85SA, hence, the filter F5 will probably be effective only at higher relative
densities. The above constriction-based analysis agrees with the applicability of USACE
(1953) and Sherard et al. (1984a) criteria for uniform base and filter materials, where
effective filters are safely characterized by D15/d85 < 5.

Table 4.1 Filter and base soil parameters
Series A: (a)Very Uniform Base Soils (Fine Sand) and Filters (Current Study)
Base/Filter Rd D15 d85 Dc35 d85SA Laboratory

Cu D15/d85 Dc35/d85SA D15/d85R
Materials (%) (mm) (mm) (mm) (mm) Assessment
Base soil - 1.2 - 0.116 - - 0.116 - - -

Filter F4 70 1.2 0.465 - 4.0 0.096 - 0.83 4.0 Effective
(b) Uniform Base Soils (Lateritic) and Filters (Indraratna et al. 1996)
Base soil - 1.29 - 0.044 - - 0.041 - -
Coarse Filter 50 1.47 0.680 - 15.45 0.161 - 3.95 15.45 Ineffective
Medium Filter 50 1.45 0.228 - 5.18 0.054 - 1.33 5.18 Effective
Fine Filter 50 1.28 0.120 - 2.72 0.027 - 0.67 2.72 Effective
Series B: Moderately –graded Base Soil and Filters (Indraratna and Vafai, 1997)
Base Soil - 2.86 - 0.42 - - 0.301 -
F1 90 2.87 0.75 - 1.78 0.147 - 0.49 0.49 Effective
F2 90 2.87 4.21 - 9.94 0.849 - 2.82 2.82 Ineffective
Series C: Well-graded Base Soil and Uniform Filters (Current Study)
Base soil - 8.75 - 1.8 - - 0.39 - - -
Filter F2 70 1.2 7.3 - 4.06 1.47 - 3.76 4.87 Ineffective
Series D: Well-graded Base Soils and Filters (Lafleur 1984)
Base soil B-3 - 9.0 - 7.00 - - 0.159 - - -
Filter F-1 70 25.0 0.26 - 0.04 0.031 - 0.20 0.37 Effective
Filter F-5 70 1.9 15.0 - 2.14 3.236 - 20.34 21.43 Ineffective

100
d85 (a)
85%
80 F6
F5
Percent Finer
60 F4
Base Soil
(Cu=1.2)
40
Filters
(Cu=1.2)
20
15%
D15(F4) D15(F6)
0
0.1 1
Effective Ineffective:Dc35>d85SA
100
(b)
d85SA
85%
80 PSDSA
(Base) F4 F5 F6
Percent Finer
60
40 35%
20
Dc35(F4) Dc35(F5)
Dc35(F6)
0
0.1
Figure 4.4 Series A: Analysis of very uniform filters and base soil of
parallel gradations (a) PSDs of filters and base soil (b) Filter CSDs and
PSDSA of base soil

Indraratna et al. (1996) conducted a series of tests on a uniform lateritic residual base soil
against three uniform sand filters (fine, medium and coarse) (Figure 4.5a and Table 4.1).
The filter CSDs and the PSDSA of the base soil are computed and presented in Figure
4.5b. For the fine sand filter, Dc 35 (fine) < d85SA; hence, it is expected to be effective. The
coarse sand filter is ineffective as Dc 35 (coarse)> d85SA. These predictions are in total
agreement with the laboratory observations made by Indraratna et al. (1996). However,
the current model classifies the medium sand filter also as ineffective in the retention of
lateritic base soil. Laboratory observations indicated that a very uniform ( C u <1.5)
medium sand filter often took a much longer time to establish self-filtration compared to
a uniform fine sand filter, while uniform coarse sand filters could not establish a self-
filtering interface at all (Indraratna et al., 1996). However, uniform medium sand filters
could be made more effective when compacted to a much higher relative density
exceeding 70% (Dilema 1990).
4.4.2 Series B: Moderately –graded Base Soil and Filters
In this series, data is taken from Vafai (1997), where the base soil consisted of fine beach
sand from Wollongong. Two parallel-graded filters (F1 and F2) consisted of medium-
grained sand and river gravel respectively. The relevant filter and base soil parameters,
and laboratory results are shown in Figures 4.6a and 4.6b, and Table 4.1. In relation to
the current model, the filter CSDs and the PSDSA of the base soil are computed and
plotted in Figure 4.6b. It is shown that Dc 35 (F1) < d85SA, and Dc 35 (F2) > d85SA,
classifying F1 as effective and F2 as ineffective. These predictions are in accordance
with the experimental observations reported by Indraratna and Vafai (1997).

100
(a) d85
85%
80 Base Soil
Coarse
(Cu=1.3)
(Cu=1.4)
Medium
Percent Finer
Filter D15/d85 (Cu=1.4)
60 Coarse 15.5
Medium 5.2 Fine
Fine 2.7 (Cu=1.3)
40
20 15%
D15(Fine)
0
0.001 0.01 0.1 1
100
(b) d85SA
85%
80 PSDSA Fine Medium Coarse

(Base)
Percent Finer
60
40 35%
20
Dc35(Coarse)
0
0.01 0.1 1
Figure 4.5 Series A: Analysis of very uniform base soil and filters
(Rd=50%) (a) PSDs of base soil and filters (b) Filter CSDs and PSDSA of
base soil

4.4.3 Series C: Well-graded Base Soil and Uniform Filters
In this series of base soil and filters, data is taken from the laboratory tests conducted
during this research study. The base soil mainly consisted of well-graded silty sand, and
two selected filters F1 and F2 consisted of uniform sub-rounded gravel. The relevant
filter and base soil parameters, and the laboratory evaluations are summarized in Table
4.1. As described earlier for Series A and B, by comparing Dc35 with d85SA it can be
predicted that F1 is effective and F2 is ineffective confirming the laboratory
observations. The corresponding graphical plots are omitted for the brevity of
presentation.
4.4.4 Series D: Well-graded Base Soils and Filters
Finally, another example is examined based on Lafleur (1984) filtration tests. They
carried out several tests involving broadly-graded cohensionless tills as base soils. The
filters consisted of sand and gravel sizes (see Table 4.1). By comparing Dc35 with d85SA, it
is clear that the filter F-1 is effective and F-5 is ineffective. The Terzaghi criterion will
classify F-5 as effective as its retention ratio ( D15 / d 85 = 2.2) is less than 5. This again
demonstrates that for highly well-graded base soils, the Terzaghi criterion may be
unreliable as this retention criterion ( D15 / d 85 < 5) was developed on the basis of testing
uniform materials (USACE 1953).
The above experimental results and CSD-based analyses demonstrate that the validity of
the proposed constriction-based model (Dc35/ d85SA < 1) is consistent with most
laboratory observations. Given the original PSDs of a filter and a base soil, and the
expected level of filter compaction in the field, the current model can reliably predict

whether the filter is effective or not, for a range of both uniform and well-graded
materials.
100
(a)
d85
85%
80 Filter D15/d85
F1 1.7
F2 9.9
Percent Finer
60
Base Soil
(Cu=3)
40
F1 F2
(Cu=3.0) (Cu=3.0)
20 15%
D15(F1) D15(F2)
0
0.1 1 10
100
(b) d85SA
85%
80 PSDSA F1 F2
(Base)
Percent Finer
60
40 35%
20
Dc35(F1) Dc35(F2)
0
0.1 1
Figure 4.6 Series B: Analysis of moderately-graded base soil and filters

(a) PSDs of base soil and filters (b) Filter CSDs and PSDSA of base soil

4.5 Comparison with Existing Retention Criteria
The use of CSD and the PSD by surface area instead of PSD by mass is a novel feature of
the current model. With the increase in Cu of the base soil, the reliability of D15/d85 < 5
(USACE 1953) criterion for effective filters becomes questionable. For instance, uniform
filters can often become ineffective for highly well-graded base soils. The proposed filter
criterion eliminates this limitation by employing d85SA instead of d85, and comparing it
with the specific constriction size Dc35. While ICOLD (1994) and NRCS (1994) suggest
the possible benefits of using a finer representative particle size for well-graded base
soils or regrading, the current model incorporates the non-uniformity of a given base soil
in a more comprehensive and quantifiable manner. The existing design practices (NRCS
1994) is the extension of the original Terzaghi criterion (USBR 1953) through regrading
the base soil PSD by US sieve #4 size (i.e. 4.75mm). In this respect, it is relevant to make
comparison with these existing criteria.
4.5.1 Terzaghi Criterion
Twenty-seven sets of experimental data from past studies for both effective and
ineffective filters including those discussed earlier in the section 4.4 are plotted together
in Figure 4.7. Well-known particle-size based retention criteria (USACE 1953) are also
shown in Figure 4.7a to examine the demarcation between the effective and ineffective
filters. In contrast, the constriction size based model is shown in Figure 4.7b for the same
data. Figure 4.7a demonstrates that for highly well-graded cohesionless tills in which the
conventional (by mass) d85 size is usually much larger than d85SA, a cluster of coarse and
uniform ineffective filters fall in the predicted effective zone. This perhaps justifies the
introduction of an additional USACE (1953) constraint, D15 < 0.40 mm, to ensure

effective filtration. These contradictory tests are invariably related to well-graded base
materials, verifying the limitations of the Terzaghi retention criteria. In comparison,
Figure 4.7b verifies that the current model can more successfully separate the effective
from the ineffective filters for a wide range of base and filter materials, and without
being unduly conservative. It is noted that only a few filters (very close to the boundary)
experimentally evaluated to be effective fall on the predicted ineffective zone. It is
understood that for such borderline filters, where Dc35 is only slightly larger than d85SA,
the actual retention efficiency cannot be assured with absolute confidence. For instance,
some uniform filters may take a long time to establish a self-filtering layer thereby losing
a significant amount of base soil initially, but stabilizing after a much longer period of
time. With regard to test #12, the laboratory observation was inconclusive. Although this
test was declared to be effective, there was substantial base soil loss before the filter
stabilized. This is also clear from the fact that the test #11 involving a finer filter was
declared to be ineffective based on excessive soil loss. It is encouraging to note that in
Figure 4.7b, none of the experimentally proven ineffective filters fall in the predicted
effective zone. In this respect, the proposed criterion of Dc35/ d85SA <1 seems to benefit
from an inherent safety factor.

Table 4. 2 Description of base and filter materials for various laboratory tests
Base Soil Filter
Test # + +
D 15 /d 85 Laboratory Reference
Notation C u d 85 Notation Cu D 15 Observation
1 Base Soil 2.86 0.42 F1 2.87 0.75 1.78 Effective Indraratna and Vafai (1997)
2 F2 2.87 4.21 9.94 Ineffective
3 Base Soil 11.40 1.18 F-1-40 1.30 5.10 4.32 Ineffective Current study
4 Base Soil 9.33 1.40 F1 1.20 5.00 3.57 Ineffective
5 F2 5.23 1.40 0.18 Effective
6 Base Soil 1.29 0.04 Coarse 1.47 0.68 15.45 Ineffective Indraratna et al. (1996)
7 Medium 1.45 0.23 5.18 Effective
8 Fine 1.28 0.12 2.72 Effective
9 B-3 8.89 7.00 F1 25.0 0.26 0.04 Effective Lafleur (1984)
10 F2 8.00 1.00 0.14 Effective
11 F3 4.29 3.20 0.46 Ineffective
12 F4 2.39 7.30 1.04 Effective
13 F5 1.85 15.00 2.14 Ineffective
14 Base Soil 1.2 0.4 F1 1.44 2.50 6.25 Effective Current study
15 F2 1.18 1.30 3.25 Effective
16 Base Soil 1.18 0.28 F3 1.17 3.60 12.86 Ineffective
17 Base Soil 1.28 0.116 F4 1.28 0.47 4.00 Effective
18 F5 1.28 0.58 5.00 Effective
19 F6 1.28 0.69 6.00 Effective
20 B1 2.63 1.55 F1 1.2 7.3 4.71 Ineffective
21 F2 3.12 1.40 0.90 Effective
22 B2 5.2 1.28 F1 1.20 7.30 5.70 Ineffective
23 F2 4.47 0.78 0.61 Effective
24 B3 5.85 0.75 F1 1.20 7.30 9.73 Ineffective
25 F2 3.12 1.40 1.87 Effective
26 B4 8.75 1.8 F1 1.20 7.30 4.06 Ineffective
27 F2 3.12 1.40 0.78 Effective
* Values in % where H /F is minimum in the range of 0-20% or 0-30%, whichever is applicable
+
Values in mm

100
Lab Assessment (a)
Effective
13
Ineffective
24
10
ve 22 20 26 12
fe cti 16 2
In ef 3 4 11
14 25 21
D15 (mm)
15 5 27 10
1 6 19 1
18 23
7 17 9
e
t iv
ec
Eff
8
0.1
=5
/d 85
D15
Note: Refer to Table 4.2 for details of test numbers
0.01
0.01 0.1 1 10
d 85 (mm)
10
Lab Assessment (b)
Effective 22 =1
13 A
Ineffective / d 85S
12
24 D c35
26 20
1 3
e 2 e
c tiv 11 16 c tiv
fe 15
ffe
Dc35 (mm)
ef E
In 27 21
25 14
6 19 10 1 23
18
0.1 17
7
8 9
0.01
0.01 0.1 1 10
d 85SA (mm)
Figure 4.7 Application of retention criteria to distinguish between effective and
ineffective filters (a) Terzaghi criterion (USACE 1953), and (b) current constriction
model

4.5.2 Current Design Practice
The use of d85SA in the current model and d85R in the NRCS (1994) design guidelines
provides two alternative solutions to the same filtration problem. The current model
cannot be directly compared with the NRCS (1994) guidelines where the filter
boundaries vary depending on the percentage of fines in the base soils. However, the
current model is developed for cohesionless soils (Figure 4.7b) so it can be compared to
the available criterion, D15/d85R 4 (NRCS 1994), for base soils in Category 4 (Figure
4.8). Figure 4.8 represents d85R on the horizontal axis, and the boundary, D15/d85R = 4,
demarcates the effective filters from the ineffective ones. It can be seen that while the
regraded boundary also applies well for cohensionless soils used in this analysis, the
authors’ criterion based on constriction size Dc35 is equally acceptable (Figure 4.7b).
However, as discussed earlier in literature review section, the regraded criterion has
obvious limitation with coarser base soils. Moreover, a key advantage of the proposed
Dc35 approach is that regrading of base soil is not needed. Also, the current method is
based on analytical principles capturing the surface area and constriction size concepts
rather than a purely empirical technique. The fact that the current criterion holds for
cohensionless granular base soils further validates the theory that the frequency by
surface area of particles best models the porous granular media.

100
Lab Assessment
Effective
Ineffective 13
10
e 12 24 22 26
iv
ec t 16 2 20
ff
I ne
11 34
14
21
D15 (mm)
25
6 15 5 27
1 19 1 10
18 23
7 17
e 9
c t iv
0.1
8
E ffe
=4
/ d 85R
D1 5
0.01
0.01 0.1 1 10
d 85R (mm)
Figure 4.8 Application of retention criteria to distinguish between effective and

ineffective filters using the current design practice with regraded base soil PSDs
(NRCS 1994)
4.6 Summary and Conclusions
Filter criteria employed in practice are often based on laboratory tests that were carried
out on uniform filter and base materials. Most of these empirical criteria invariably
involve some characteristic particle sizes by mass, typically the Terzaghi retention ratio,
D15 / d 85 . They have obvious limitations especially when well-graded base soils are
tested with coarse uniform filters. Unless regrading is carried out (NRCS 1994), the
original D15 / d 85 criterion does not hold for many well-graded base soils. Limitations are
clearly illustrated in literature review section. In this study, in lieu of regrading, the
authors have proposed an alternative filter criterion based on the controlling constriction
and surface area concepts applied to base soils, especially those that are well-graded.

Similar to mechanical sieves, it is the constriction (opening) sizes rather than the particle
sizes that influence filtration in granular filters (Kenney et al. 1985; Locke et al. 2001). In
this study, the representative filter constriction size is proposed to be Dc35. Also, by
surface area consideration of the base soil, d85SA is found to be a more appropriate
representative parameter compared to the conventional d 85 by mass. It is demonstrated
that the proposed criterion, Dc35/d85SA <1, for identifying effective filters is more realistic,
whereby the size of prescribed base particle size (d85SA) is directly compared with the
size of controlling filter constriction (Dc35).
For a wide range of base and filter materials, the current model is shown to successfully
separate the effective from the ineffective filters without being unduly conservative.
Unlike D15 the main advantage of Dc35 is that it is sensitive to the uniformity coefficient
(Cu) of the filter particle size distribution (PSD) as well as to its relative density (Rd).
It is important to note that the proposed constriction-based criterion for effective filters
(Dc35/d85SA <1) is developed for cohensionless base soils. Surface area concept was
basically developed and verified to model the porous granular media such as sand and
gravel filters. Because of high specific surface area of very fine clay particles, the base
particles are over-represented when the concept is applied to cohesive base soils.
However, it was found to be reasonable to lateritic base soils. This is because lateritic
base soils have uniform grading, where as illustrated in chapter 3, the choice of
frequency considerations, whether it is mass, number or surface area, does not make a
difference. In this respect, it can be said that this criterion is applicable to mainly
category 4 base soils (i.e. gravels and sands) and any base soil with uniform grading.

CHAPTER
FIVE
STABILITY OF SELF-FILTRATION LAYER
5.1 Introduction
It is now well accepted that self-filtration is the most important phenomenon in effective
filtration. Coarser base particles are captured by the filter thereby making the
constrictions smaller, which in turn capture smaller base particles. In effective filters,
the process is continued until the filter stabilises and no further base particle can move
into filter. In contrast, in ineffective filters, either the filter does not stabilise and the
base particles continue to wash through the filter or it stabilises after a significant
amount of base soil is lost through the filter. In this respect, an effective self-filtration
layer is formed by the capture of erodible base particles into the filter, as illustrated by
Figure 1.3. The question that arises is what fraction of base particles is erodible. There
is sufficient evidence (Lafleur 1984; Sherard and Dunnigan 1985; Lafleur et al. 1989)
suggesting that the larger base particles do not influence filtration. What size is too
large?
Chapter 5: Stability of Self-filtration Layer 137

This chapter provides a rigorous analytical model to determine the self-filtering fraction
of the base soil with respect to a given filter. It also develops an analytical procedure to
determine the mass fraction of the base soil retained by a given filter, providing the
particle size distribution (PSD) curve of the self-filtration layer. The stability of self-
filtration layer is evaluated based on the constriction concepts, which gives an enhanced
filter design criterion to describe effective filters. The model is verified using the same
test data as employed in Chapter 4.
5.2 PSD of Self-Filtration Layer
Potentially erodible base particles are transported to the filter by hydrodynamic forces.
As suggested by Kenney et al. (1985), base particles larger than controlling constriction
size are initially captured by constrictions, producing finer constrictions, which then
progressively retain smaller base particles. In this manner, a self-filtration layer is
formed immediately downstream from the base soil-filter interface. Figure 5.1 presents
five filters F1 to F5 used by Lafleur (1984), where constriction size distributions
(CSDs) are computed for a relative density of 70% based on the method described by
Locke et al. (2001). As the size of the largest particles, D100, in all filters is the same, the
size of the largest constriction, Dc100, is also expected to be the same. However, in well-
graded filters, the sizes of dominant constrictions should be considerably smaller than
Dc100. For instance, Figure 5.1 shows that Dc95 is more appropriate for distinguishing
between filters F1 to F5 at the upper end of the coarse constrictions. The choice of Dc95
is further justified by both analytical and laboratory observations.

100
95%
D c95 (F1) D c100 D 100

Filter CSDs
80
F5
Percent Finer
60 F3 F4
F1 F2
40 Base Soil PSD
Filter PSDs
20
F3 F4 F5
F1 F2
0
0.01 0.1 1 10 100
Constriction Size D c / Particle Size D (mm)
Figure 5.1 Dominant constrictions in various types of filters
For a base particle of Dc95 size, p = 5%. P(F) can be calculated by Equation (3.20) and it
comes to be 6.01%. For a confidence level of P =95%, ‘n’ can be calculated by
Equation (3.22) and it comes to be 1.06 (§1). This means that for any base particle
larger than Dc95, ‘n’ will be less than 1, which means that this particle can not penetrate
even a single layer of filter and consequently can not move into the filter. Because CSD
model is already verified based on laboratory observations (Locke et al. 2001) and as
illustrated in Chapter 3, the authors are confident that this proposition is correct and
realistic. Accordingly there is a 95% chance that a base particle larger than Dc95 cannot
penetrate a single layer of the filter and therefore would not influence self-filtration.
This modification of the base soil PSD also explains why the coarser particle fraction
could be ignored in filter designs that involve well-graded and internally unstable gap-
graded base soils (Lafleur et al.1989). Thus it is clear that the self-filtration is initiated

by retention of the base particles larger than Dc35 and smaller than Dc95 at various filter
depths, and gradually finer (i.e. <Dc35) base particles are retained resulting in stable self-
filtration layer. In other words, it can be concluded that the PSD of the self-filtration
layer is formed by filter particles and the base particles finer than the constriction size
Dc95. Now the question arises, what will be the relative mass proportions of the filter
and base particles in the self-filtration layer?
Kenney and Lau (1985) mentioned that the captured base particles usually remain in a
loose state within the filter pores, resulting in a net porosity (nB) of about 0.40. The
initial filter porosity (nF) depends on the field compaction. Assuming that the specific
gravity of the base and filter particles is the same and considering a unit volume of self-
filtration layer:
Volume of filter particles, VSF = 1-nF
Volume of base particles, VSB = (1- nB)nF
Fraction of filter particles in self-filtration layer, PF = VSF/(VSF +VSB)
= (1- nF)/(1- nF . nB)
Fraction of base particles in self-filtration layer, PB = VSB/(VSF + VSB)
= [(1- nB).nF]/(1- nF . nB)
PF / PB ratio = (1/nF -1)/(1-nB)
Once the mass proportions are determined, employing the PF /PB ratio, the PSD of self-
filtration layer can be obtained by combining the PSD of the base soil (modified by
disregarding any base particles larger than Dc95) and the PSD of the filter.

In order to illustrate the computation procedure for determining the PSD of the self-
filtration layer, the particle size and constriction size distributions of the filter F5 and
the base soil from Figure 5.1 are re-plotted in Figure 5.2. The CSD is computed
following the method of Locke et al. (2001) and Dc95 is 6 mm. The modified PSD of the
base soil is then calculated by ignoring all base particles larger than 6 mm, which is
presented in Figure 5.2. Knowing the relative density (Rd = 70%), the equivalent
porosity nF is calculated to be about 36%. As mentioned earlier a value of 40% is
considered for nB. Subsequently PF and PB are calculated as described earlier and found
to be 74.20% and 25.80%, respectively. Finally the PSD of the self-filtration layer
(Figure 5.2) is obtained by combining the filter PSD and the modified base soil PSD in
the ratio of PF : PB (approx. 3:1).
100
95%
D c95
Modified Base Soil PSD
80
Rd 70.00 %
nB 40.00 %
Percent Finer
nF 36.70 %
60 PF 74.20 % Filter CSD
PB 25.80 %
Original Base Soil PSD

40
Self-filtration Layer PSD
20
Filter PSD
0
0.01 0.1 1 10 100
Figure 5.2 PSD of self-filtration layer in a typical base soil-filter combination

5.3 Stability of Self-filtration Layer
In order to illustrate the base soil and filter interaction in the self-filtration layer, the
PSDs of the self-filtration layers corresponding to the filters F1 to F5 (Figure 5.1) are
determined and plotted in Figure 5.3. The progressively widening ‘gaps’ in the PSD
curves of self-filtration layers corresponding to the coarser filters (F3, F4 and F5) imply
their internally unstable, gap-graded nature. These coarser filters may not be able to
retain the potentially erodible fine base particles, which will probably render them
ineffective. By contrast, the self-filtration layers of the finer filters (F1 and F2) do not
have ‘gaps’, hence, they represent internally stable soils. These filters are most likely to
retain the potentially erodible base particles, thereby considered to be effective. The
internal stability of a self-filtration layer can be examined using the Kenney and Lau
(1985) method succinctly presented in Figure 5.4. As illustrated in Figure 5.4, H is the
percent of mass between two particle sizes D and 4D, and F is the percent of mass finer
than the particle size D. An evaluation of the internal stability of filters based on self-
filtration leads to a rigorous model for identifying effective filters. The proposed
approach uses the largest dominant constriction size, Dc95, for disregarding coarser
particles, which do not influence filtration. This approach is more comprehensive than
the Terzaghi method of using particle size ratios, especially for well-graded soils. In this
respect, the aim of the proposed model is to use the derived PSD curve for the self-
filtration layer to determine the H/F ratio, in order to examine the stability of a given
base soil-filter system.

100
Self-filtration Layer PSDs

80
The shape of self-filtration layer PSD curves clearly
shows that as expected, the stability of self-filtration
layer decreases with progressively coarser filters.
Percent Finer
60
40
F1 F2 F3 F4 F5
20
0
0.01 0.1 1 10 100
Figure 5.3 PSDs of self-filtration layers in progressively coarser filters
H/F 1 Stable Grading

Particle Size
Percent Finer
Distribution H/F < 1 Unstable Grading

Curve
H Provided
F 30% for uniform coarser part (Cu<3), and
F 20% for widely-graded coarser part (Cu>3)
F
D 4D
Particle Size
Figure 5.4 Kenney and Lau (1985) procedure for internal stability assessment

5.4 Model Verification
Data from several filter tests conducted during this research study and by others were
analysed using the current model. A few examples of which are considered here as
illustrations. Indraratna and Vafai (1997) carried out two large-scale tests using
Wollongong beach sand as the base soil and sub-rounded river pebbles as the filter
material. Both base and filter materials are uniform with Cu just less than 3. The filters
were compacted to a relative density of 90%, i.e. an equivalent porosity nF of about
31%. The retention ratios D15/d85 in these tests F1 and F2 are 1.78 and 9.94,
respectively. The laboratory observations indicated that the filter F1 was effective
whereas F2 was ineffective. The values of PF and PB were determined to be 78.37% and
21.63%, respectively. The filter CSDs were calculated and the corresponding
constriction sizes Dc95 are 0.32 mm and 1.71 mm for F1 and F2, respectively. The
modified PSD of the base soil and the PSDs of the self-filtration layers for these two
filters are presented in Figures 5.5a and 5.5b. The internal stability of the layers was
subsequently checked by calculating the H/F ratios in the range of F = 0-30%. This
relatively larger range of F = 0-30% was considered because the coarser part of the PSD
of the self-filtration layer is predominantly composed of uniform filter grains (Cu=2.87).
The analysis shows that for F1, the minimum H/F ratio is 1.40 at F = 9.01 with
corresponding H = 12.59. For F2, the minimum H/F ratio is 0.017 at F = 21.63 and H =
0.37. For F1, H/F > 1 in the range of F=0-30% indicates that the PSD of self-filtration
layer is internally stable, resulting in an effective filter. For F2, H/F < 1 in the range
F=0-30%, which suggests that a stable self-filtration layer could not be formed,
resulting in an ineffective filter. Thus the model predictions confirm the laboratory
observations.

100 95%
5.0
D c95
D 15 /d 85 1.78 Original Base
80 H /F (Min) 1.40 H /F Soil PSD 4.0
(C u = 2.86) (a)
Percent Finer
60 3.0
Filter CSD Filter PSD
H /F
(C u = 2.87)
40 2.0
Modified Base
H /F (Min)= 1.40
Soil PSD
H /F = 1
20 1.0
Self-filtration
Layer PSD
0 0.0
0.01 0.1 1 10
Constriction Size D c /Particle Size D (mm)
100 95%
5.0
D 15 /d 85 9.94 D c95
H /F
80 H /F (Min) 0.017 4.0
Original Base Filter CSD (b)
Soil PSD
Percent Finer
60 (C u = 2.86) 3.0 H /F
Modified Base
Soil PSD
40 Filter PSD 2.0
Self-filtration (C u =2.87)
Layer PSD
H /F = 1
20 1.0
0 0.0
0.01 0.1 1 10 100
Constriction Size D c /Particle Size D (mm)
Figure 5.5 (a) Analysis of effective uniform filter F1 with a uniform base soil and
(b) an ineffective uniform filter F2 with a uniform base soil

Several filter tests using well-graded base soils were tested against uniform and well-
graded filters during this study. The well-graded base soil (Cu = 9.33) was prepared by
mixing clean quarry sands of different uniform sizes with a non-plastic sandy silt soil at
50:50 proportions. Similarly the well-graded filter was prepared by mixing clean quarry
sands of different uniform sizes with river pebbles of various uniform sizes. The
porosity of filters was determined in relation to a compacted relative density of 70% i.e.
equivalent porosity nF of 36%. For this well-graded base soil (Cu = 9.33), the PSD and
CSD analysis is illustrated for F1 and F2 filters in Figures 5.6a and 5.6b, respectively.
Cu for the filters F1 and F2 are 1.20 and 5.23, respectively. The corresponding retention
ratios D15/d85 in these tests are 3.57 and 0.18, respectively. The values of PF and PB
were calculated to be 74.18% and 25.82%, respectively. The filter CSDs were
determined and the constriction sizes Dc95 were calculated to be 1.46 mm and 0.16 mm
for F1 and F2, respectively. The modified PSDs of the base soil and the PSD of the self-
filtration layers of these filters are also presented in Figures 5.6a and 5.6b. The internal
stability of the layers was examined by calculating H/F ratios in the range of F = 0-30%
for F1 and in the range of F =0-20% for F2. A smaller range F = 0-20% was considered
for F2 because the coarser part of the PSD of the self-filtration layer is predominantly
well-graded (Cu>3). The analysis shows that for F1, the minimum H/F ratio is 0.06 at F
= 24.27 with corresponding H =1.55, and for F2, the minimum H/F ratio is 1.26 at F =
4.43 and H = 5.57, confirming the laboratory observations that the filter F1 was
ineffective and F2 effective. The above examples verify that the authors’ approach can
successfully distinguish between effective and ineffective filters.

100 95% 5.0
D 15 /d 85 3.57 Original Base Soil PSD D c95

H /F (Min) 0.06 (C u =9.33)
80 4.0
Percent Finer
Filter CSD
60 3.0
(a)
H /F
H /F
40 2.0
H /F = 1
20 1.0
Self-filtration Layer PSD Filter PSD
(C u =1.20)
0 0.0
0.001 0.01 0.1 1 10
100 95% D c95 5.0

Original Base
Filter CSD
Soil PSD
80 D 15 /d 85 0.18
(C u =9.33) 4.0
H /F (Min) 1.26
Percent Finer
60 3.0 H/F
Filter PSD
(C u =5.23)
40 (b) 2.0
H /F
H /F = 1
20 1.0
Self-filtration Layer PSD
0 0.0
0.001 0.01 0.1 1 10
Figure 5.6 (a) Analysis of an ineffective uniform filter F1, and (b) an effective
well-graded filter F2 with a well-graded base soil

5.5 Comparison with Existing Criteria
5.5.1 Terzaghi Method
Since the scope of Kenney and Lau (1985) method for assessment of internal stability is
limited to cohensionless granular soils, the same twenty-seven sets of test data, as used
in Chapter 4, are reanalysed using the current approach. The analyses using the
Terzaghi’s criterion are already illustrated in Figures 4.7a and the test details are
tabulated in Table 4.2. Data were mostly taken from the tests involving well-graded
base soils, where most conventional particle based criteria exhibit limitations. For
example, Figure 4.7a clearly shows that some filters involving retention ratios D15/d85
well below 4-5 failed to retain the well-graded base soils but still plot in the effective
zone. Figure 8, however, based on the current constriction-based approach, clearly
illustrates that none of the failed tests plot on the effective zone (H/F>1) established by
the model. It is to be noted that in order to enhance clarity of a large number of data
points, Figure 5.7 has been divided into four parts (Figures 5.7(a)-5.7(d)). A few data
points showing limited erosion, such as #7 (Indraratna et al. 1996; Figure 5.8 (a)), #12
(Lafleur 1984; Figure 5.8(b)), and #14 and #19 (Figure 5.8(c)) cross the H/F=1
boundary to the ineffective zone, albeit considered effective in laboratory tests. Except
the test #12, three other tests take a relatively longer time to establish self-filtration
compared to the other effective filters, and are still effective (Indraratna et al. 1996).
However, as expected for the reason described in section 4.5.1, the test #12 is again
found to be ineffective so this is considered to be a failed test in subsequent analysis.
Data points #14 and #19 represent the recent tests involving uniform base soils. Similar
to the observations discussed above, these tests also showed limited erosion and

relatively longer self-filtration time before the filters attended some stability. In Figure
5.7, the authors have used different symbols for these points to indicate ‘limited
erosion’.
40
(a) (b)
1
30 5
8 7
8
6
H 20 1 9
1 H /F=
=1
2 H/F 13
3
10
10 11
4 7 11
4 6 13 12
5 2
3
0
40
(c) (d)
21 20 26
30
15
=1
20
23 H/F
H 17 20 25
1
H/F= 14 27
24 22
10 22
18 19 26 24
16
0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
F F
Effective Effective (Limited Erosion) Ineffective
Figure 5.7 Comparative analysis of test results using the current model (a) Tests #1-8
(b) Tests #9-13 (c) Tests #14-19 and (d) Tests #20-27
(Refer to Table 4.2 for details of test numbers)
5.5.2 Current Design Implication
The analysis based on the current guideline is already given in Chapter 4 illustrated by
Figure 4.8. Regrading of base soil (NRCS 1994) and the proposed Dc95 criterion based
on self-filtration and internal stability are two alternatives to address the same

limitations of the original Terzaghi filtration approach. The current model cannot be
directly compared with the NRCS (1994) guidelines where the filter boundaries vary
depending on the percentage of fines in the base soils. However, since Kenney and Lau
(1985) internal stability method is based on cohensionless soils, the current model can
be compared to the regraded criterion (i.e. D15/d85R 4) for cohensionless base soils
(Figure 4.8). As described earlier in Section 4.5.2, Figure 4.8 represents d85R on the
horizontal axis, and the boundary, D15/d85R = 4, demarcates the effective filters from the
ineffective ones. It can be seen that while the regraded boundary applies well for
cohensionless soils used in the current analysis, the proposed Dc95 model, employing
H/F technique, is quite reasonable (Figure 5.7).
The key advantage of the proposed approach based on Dc95 is that regrading of base soil
is not required. Also, as the H/F ratio of the proposed method inherently includes
internal stability, the designer is not required to carry out a prior analysis to examine the
internal stability of the base soil. Moreover, plotting the self-filtration PSDs (Figure 5.3)
where a ‘gap’ is evident in all ineffective base soil-filter combinations will certainly
boost the designer’s confidence. Thus, Dc95 forms a rational basis for regrading the base
soil grading curve in order to determine its self-filtering fraction with respect to a given
filter. This explains the fact that a base particle of 4.75mm size may be too large for a
finer filter but not for a coarser filter. In this respect, the proposed model is essentially
more comprehensive because it takes into account filter compaction, porosity and
coefficient of uniformity (Cu). In addition it considers self-filtration and internal
stability to enhance the rigor in assessing filter effectiveness.

5.6 Summary
When eroded base particles are transported to the filter, only coarser particles larger
than the controlling constriction size are initially captured. These finer constrictions
progressively retain finer base particles to form a self-filtration layer. Base particles
larger than the constriction size Dc95 do not influence the process of self-filtration
because they do not penetrate the filter. Therefore, the constriction size Dc95 is a
reasonable cut-off value, and the base soil PSD modified accordingly is more realistic in
the analysis of filtration. Hereafter, Dc95 is called the self-filtering constriction size.
Mass retained in the self-filtration layer depends on the initial porosity of the filter and
the subsequent porosity of the self-filtration layer. The PSD of the self-filtration layer
can be determined by combining the initial filter PSD and the modified base soil PSD
incorporating Dc95. In effective filters potentially erodible base particles must form a
stable self-filtration layer that is not gap-graded or concave upward.
An assessment of the internal stability of the layer on the basis of H/F ratios gives rise
to a rigorous analytical model to successfully identify effective filters. Considering the
test data discussed in this study, the prediction of filter effectiveness based on the
current constriction-based approach is accurate in relation to various combinations of
base and filter materials for uniform and well-graded base soils. The current model
provides a more rational and rigorous procedure for filter design by eliminating the
obvious limitations of conventional particle size criteria based on the D15/d85 ratio alone.
As illustrated by Figure 5.3, where progressively widening ‘gap’ is evident in a coarser

filter, drawing a PSD curve of the self-filtration layer will certainly boost the designer’s
confidence.
The current method employs the Kenney and Lau (1985) method of internal stability
assessment. In this regard, the current method is expected to be associated with the
limitation of this assessment procedure. This initial analysis of the study found the H/F
ratio equal to 1.3 as the boundary between stable and unstable soils. However, the
results showed that this ratio was conservative and hence it was amended to 1 without
much elaboration. Is this method still a conservative approach? This issue will be
discussed later in chapter 7.

CHAPTER
SIX
SELF-FILTERING BASE FRACTION AND FILTER DESIGN
6.1 Introduction
As discussed earlier in Chapter 2, the famous Terzaghi filter criterion, D15/d85 4-5, was
developed partly on the basis of theoretical analysis and partly laboratory observations
carried out using uniform sands as base and filter materials. Consequently, it has some
obvious serious limitations associated with non-uniform base and filter materials,
particularly with well-graded and broadly-graded base soils. The problem was identified
a long time ago in the early 1950s (Karpoff 1955; Sherard et al. 1963). Since then,
several studies extended the original Terzaghi criterion, assigning a smaller value to d85
depending on the amount of fines (i.e. <ASTM No. 200 sieve size) and gravel (i.e.
>ASTM No.4 sieve size) contents, in order to make the criterion applicable to other soil
types, particularly broadly-graded base soils. Sherard et al. (1963) suggested the use of
d85 after regrading the base soil for the base particles larger than 1” (i.e. 25.4 mm).
USBR (1963) recommended the d85 after regrading by ASTM No. 4 sieve size (i.e.
4.75mm). In current practices, as described earlier, ICOLD (1994) recommends either
Chapter 6: Self-filtering Base Fraction and Filter Design 153

the use of the d85 after regrading by ASTM No. 4 sieve size or the use of d50 or d20
depending upon the base soil grading as described in Section 2.4. However, as
illustrated in Section 2.5, these methods are empirical and still have some serious
limitations. As described earlier, Figures 2.17 and 2.18 clearly illustrate that how the
current design guidelines fail to explain the filtration of such soils. Because of these
limitations, associated with the current guidelines, Foster and Fell (2001) found that in
spite of regrading, the filter effectiveness diminishes as the fines content of base soils
increases. They could not explain some of laboratory observations where the tests
involving category 4 (NRCS 1994) base soils with regraded retention ratio, D15/d85R,
less than or equal to 4 failed to provide effective filters.
As described in previous chapters, Dc35 and Dc95 are two important constriction sizes in
filtration. The former, Dc35, is called the controlling constriction size of the filter, and
represents the largest base particle that can wash through a filter i.e. the largest flow
channel in the filter. The latter, Dc95, is known as the self-filtering constriction size and
represents the largest effective constriction size in the filter i.e. the base particles larger
than this size do not enter into the filter and for this reason, they are said to be not
erodible and do not influence self-filtration. In an effective filtration, base particles
smaller than the controlling constriction size Dc35 are initially lost through the filter
during filtration and those in between Dc35 and Dc95 are eroded from the base soil and
retained at various depths depending upon the particle sizes. These larger base particles
initiate self-filtration upon retention. The Dc95 forms a rational approach to separate self-
filtering fraction of a base soil with respect to a given filter. The base particles larger
than this size are not erodible and are retained at the base soil-filter interface itself, and

hence do not influence self-filtration. In this respect, it is clear that in a granular filter,
self-filtration is initiated by retention of base particles of the size in between Dc35 and
Dc95. Past studies (Lafleur et al. 1989; Sherard et al. 1989) mentioned that only the self-
filtering fraction of base soils be considered while applying the original Terzaghi
retention criterion to well-graded base soils. Earlier, two different filter design criteria
were developed using these constriction sizes combined with the surface area concepts
applied to base soils and stability of self-filtration layer. However, as discussed earlier,
the scope of these criteria is limited to the cohensionless base soils of category 4, which
contain less than 15% fines (NRCS 1994).
In this section, a filter design criterion is presented on the basis of self-filtering fraction
of base soils, Dc95 and the controlling constriction size, Dc35. The model is verified using
laboratory test results of past studies as well as experiments carried out at the University
of Wollongong during this research study.
6.2 Model Development
Well-known Terzaghi filter design criterion, D15/d85 4-5, was developed through
investigations carried out on uniform filter and base materials, particularly uniform
natural sand. As revealed earlier in chapter 3 including studies such as Kenney et al.
(1985), Sherard et al. (1984a), and Foster and Fell (2001), the controlling constriction
size in the medium to highly compacted uniform filters approximately ranges in
between D15/5 to D15/4. In this respect, the Terzaghi criterion may be interpreted to
mean that the filter effectiveness is ensured with substantial conservativeness if there

are 15% base particles larger (i.e. d85) than the controlling constriction size of the filter (
i.e. D15/4 or D15/5), which can now be determined more realistically by the constriction
size Dc35. However, the use of d85 does not ensure filter effectiveness in the case of
filtration of well-graded base soils for two reasons explained earlier in Chapter 2.
Firstly, the finer base fraction (i.e. fraction smaller than D15/4 or D15/5) is comparatively
finer in well-graded base soils requiring more time to complete self-filtration and hence
resulting in more base soil loss through the filter. Secondly, not all base particles of
coarser fraction (i.e. fraction larger than D15/4 or D15/5) influence self-filtration. In this
regard, as revealed earlier in Chapter 5, the use of self-filtering constriction parameter
Dc95 is a more rational approach to determine the self-filtering fraction of base soils.
This also explains why regrading is not required i.e. use of d85 is justified in the case of
uniform base and filter materials. It can be seen in Figure 4.4b that in all cases, Dc95 is
larger than d100 (i.e. the largest base particle size) thus requiring no modification of base
soil PSD. For this reason, the classic Terzaghi filter criterion holds well in the case of
uniform base and filter materials. Accordingly, it can be concluded that the filter
effectiveness can be ensured with substantial conservativeness if d85mod is larger than
Dc35, resulting in the following constriction-based filter retention criterion where d85mod
is defined as the d85 of the base soil after modifying the base soil PSD for the particle
size larger than the self-filtering constriction size, Dc95.
Dc35 d85mod (6.1)
Similar to other filter retention criteria developed earlier in Chapters 4 and 5, this
criterion is also based on sound analytical principles. However, unlike the other two,

this is not constrained by any factor that restricts its application to a particular base soil.
The following sections demonstrate the model procedure with a number of examples
using some well-known laboratory results and subsequently, the model is compared
with the Terzaghi original criterion and the existing design guidelines.
6.3 Model Procedure Illustration
Although data from several filter tests carried out during this research study and by
others are analysed in the subsequent section using the current model, a few examples
of which are considered here in detail to illustrate the model procedure.
The same examples as discussed in Section 5.4 are considered here. Indraratna and
Vafai (1997) carried out two sets of tests using large-scale equipment on Wollongong
beach sand as the base soil and sub-rounded river pebbles as the filter material. Both
base and filter materials were uniform with Cu slightly less than 3. The filters were
compacted to a relative density of 90%. The retention ratios, D15/d85, in these tests (F1
and F2) have been 1.78 and 9.94, respectively. The laboratory observations indicated
that the filter F1 was effective and F2 ineffective. The filter CSDs were calculated and
the corresponding constriction sizes Dc35 are 0.147mm and 0.849mm, and Dc95 are
0.32mm and 1.71mm for F1 and F2, respectively. The PSDs of the base and filter
materials along with the modified PSD of the base soil for these two filters are
presented in Figures 6.1a and 6.1b. Because Dc 35 (F1) < d85mod, F1 is considered
effective, whereas F2 is deemed to be ineffective as Dc 35 (F2)> d85mod, confirming the
laboratory observations.

100 95%
d 85mod
85%
80 D 15 /d 85 1.78 Original Base

D c35 /d 85mod 0.46 Soil PSD (a)
(C u = 2.86)
Percent Finer
60 Filter PSD
Filter CSD
Modified Base Soil (C u = 2.87)
PSD
40 35%
20
D c35 D c95
0
0.01 0.1 1 10
Constriction Size D c ;Particle Size D (mm)
100 95%
d 85 =d 85mod
85%
80 D 15 /d 85 9.94
D c35 /d 85mod 2.00 Filter CSD
(b)
Percent Finer
Original/Modified
60 Base Soil PSD
(C u = 2.86) Filter PSD
(C u = 2.87)
40 35%
20
D c35 D c95
0
0.01 0.1 1 10 100
Figure 6.1 (a) Analysis of effective uniform filter F1 with a uniform base soil, and
(b) an ineffective uniform filter F2 with a uniform base soil

Several filter tests were conducted during this research study using well-graded base
soils tested against uniform and well-graded filters. The well-graded base soil (Cu =
9.33) was prepared by mixing clean quarry sands of different uniform sizes with a non-
plastic sandy silt soil at 50:50 proportions. Similarly the well-graded filter was prepared
by mixing clean quarry sands of different uniform sizes with river pebbles of various
uniform sizes. The porosity of filters was determined in relation to a compacted relative
density of 70%. For this well-graded base soil (Cu = 9.33), the PSD and CSD analysis
are illustrated for F1 and F2 filters in Figures 5.6a and 5.6b, respectively. Cu for the
filters F1 and F2 are 1.20 and 5.23, respectively. The corresponding retention ratios
D15/d85 in these tests are 3.57 and 0.18, respectively. The filter CSDs were determined
and the constriction sizes Dc35 were calculated to be 1.264mm and 0.051mm, and Dc95
are 1.46mm and 0.16mm for F1 and F2, respectively. The PSDs of the base and filter
materials along with the modified PSDs of the base soil are also presented in Figures
6.2a and 6.2b. Because Dc 35 (F1) > d85mod, F1 is considered ineffective, whereas F2 is
deemed to be effective as Dc 35 (F2) < d85mod, confirming the laboratory observations that
the filter F1 was ineffective and F2 effective. The above examples demonstrate the
model procedure.

100 95%
85% Original Base Soil PSD D c95

(C u = 9.33)
80
D 15 /d 85 3.57
D c35 /d 85mod 1.62
Percent Finer
Filter CSD
60
40 35%
Filter PSD
(a) (C u = 1.20)
20
d 85mod D c35
0
0.001 0.01 0.1 1 10
100 95%
d 85mod
Original Base Soil PSD
Filter CSD
(C u = 9.33)
80
D 15 /d 85 0.18
D c35 /d 85mod 0.27
Percent Finer
60
Filter PSD
(C u = 5.23)
40
(b)
20
D c35 D c95
0
0.001 0.01 0.1 1 10
Figure 6.2 (a) Analysis of an ineffective uniform filter F1 and (b) an effective well-
graded filter F2 with a well-graded base soil

6.4 Model Verification
Eighty-three sets of test data involving base soils ranging from uniform sands, broadly-
graded tills, lateritic residual soils to dispersive and non-dispersive cohesive soils were
taken from some past well-known published filter investigations as well as results of
laboratory tests conducted at the University of Wollongong during this research study.
Data was taken mostly from the tests involving well-graded non-cohesive and cohesive
(dispersive and non-dispersive) base soils, where most conventional particle based
criteria exhibit limitations. Data were analyzed using the current approach and Terzaghi
retention criterion. The results are tabulated in Table 6.1 and also presented in Figures
6.3 and 6.4. For example, Figure 6.3 clearly shows that some filters involving retention
ratios D15/d85 well below 4-5 failed to retain the well-graded base soils but still plot in
the effective zone. Figure 6.4, however, based on the current constriction-based
approach, clearly illustrates that none of the failed tests plot on the effective zone (i.e.
Dc35/d85mod<1) established by the model. A few effective data points showing limited
erosion plot on the ineffective zone, albeit considered effective in laboratory tests. As
mentioned by Indraratna and Vafai (1996) and Lafleur (1984), these tests take a much
longer time to establish self-filtration compared to the effective filters, and are normally
associated with some initial loss of the base soil through the filter. In this regard, the
current model benefits from an inherent conservativeness.

100
10
ve
fecti
f
Ine
D15 (mm)
ve
cti
E ffe Laboratory Observations
0.1 Eff./Ineff.
Lafleur/et al. (1984; 1989)
Indraratna and Vafai (1997)
Current Study
0.01
0.01 0.1 1 10 100
d 85 (mm)
Figure 6.3 Comparative analyses of test results using the original Terzaghi retention criterion
(i.e. before regrading of the base soil)

10
1
e
ctiv
ffe e
Ine ctiv
E ffe
Dc35 (mm)
0.1
Laboratory Observations
Eff./Ineff.
Lafleur/et al. (1984; 1989)
Indraratna and Vafai (1997)
Current Study
0.01
0.01 0.1 1 10
d 85mod (mm)
Figure 6.4 Comparative analyses of test results using the current model
(Refer to Table 6.1 for details of test numbers)

Table 6.1 Description of base and filter materials for various laboratory tests
Base Soil Filter D 15 D c35 Laboratory

Test # + + Reference
Notation Type Cu d 85 d 85mod Cu R d (%) D 15 Dc35 Dc95 /d 85 /d 85mod Observation
1 B-3 Linearly-graded 8.89 7.00 0.06 25.00 70 0.26 0.03 0.11 0.04 0.56 Effective Lafleur
2 tills 7.00 0.35 8.00 70 1.00 0.17 0.58 0.14 0.47 Effective (1984)
3 7.00 0.54 4.29 70 3.20 0.67 1.92 0.46 1.24 Ineffective
4 7.00 0.58 2.39 70 7.30 1.59 3.66 1.04 2.75 Ineffective
5 7.00 0.90 1.85 70 15.00 3.24 6.03 2.14 3.60 Ineffective
6 B1 Gap-graded tills 625.00 19.00 0.07 25.00 4 0.26 0.05 0.20 0.01 0.64 Effective Lafleur et al.
7 19.00 0.07 10.00 100 1.00 0.14 0.42 0.05 2.11 Ineffective (1989)
8 19.00 0.08 4.29 84 3.20 0.61 1.71 0.17 7.63 Ineffective
9 19.00 0.18 2.39 70 7.30 1.59 3.66 0.38 8.85 Ineffective
10 19.00 0.18 1.85 98 15.00 2.74 4.22 0.79 15.20 Ineffective
11 B2 Linearly-graded 86.96 7.50 0.06 25.00 70 0.26 0.03 0.11 0.04 0.52 Effective
12 tills 7.50 0.28 10.00 24 1.00 0.21 0.88 0.14 0.74 Effective
13 7.50 0.60 4.29 65 3.20 0.68 2.11 0.46 1.14 Effective
14 7.50 0.92 2.39 80 7.30 1.50 3.27 1.04 1.63 Ineffective
15 7.50 1.90 1.85 92 15.00 2.84 4.61 2.14 1.49 Ineffective
16 B3 Broadly-graded 39.13 1.50 0.08 25.00 70 0.26 0.03 0.11 0.17 0.41 Effective
17 tills where large 1.50 0.32 10.00 23 1.00 0.21 0.89 0.67 0.65 Effective
18 particles float in 1.50 0.42 4.29 24 3.20 0.84 2.98 2.13 2.01 Ineffective
19 fine soil matrix 1.50 0.52 2.39 57 7.30 1.71 4.18 4.87 3.29 Ineffective
20 1.50 0.62 1.85 57 15.00 3.45 6.87 10.00 5.57 Ineffective
21 B4 Gap-graded tills 333.33 17.00 0.07 25.00 70 0.26 0.03 0.11 0.02 0.46 Effective
22 17.00 0.30 10.00 70 1.00 0.17 0.55 0.06 0.55 Effective
23 17.00 0.85 4.29 52 3.20 0.74 2.39 0.19 0.87 Effective
24 17.00 0.85 2.39 54 7.30 1.74 4.29 0.43 2.05 Ineffective
25 17.00 0.85 1.85 98 15.00 3.07 5.45 0.88 3.61 Ineffective

Table 6.1 Description of base and filter materials for various laboratory tests (contd.)
27 tills 11.00 0.28 10.00 70 1.00 0.17 0.58 0.09 0.59 Effective
28 11.00 0.85 4.29 66 3.20 0.68 2.09 0.29 0.80 Effective
29 11.00 1.32 2.39 73 7.30 1.57 3.54 0.66 1.19 Effective
30 11.00 1.90 1.85 73 15.00 3.17 5.83 1.36 1.67 Ineffective
33 particles float in 3.00 0.45 4.29 63 3.20 0.69 2.16 1.07 1.54 Effective
35 3.00 0.55 1.85 100 15.00 2.70 4.09 5.00 4.91 Ineffective
37 tills 20.00 0.42 10.00 0 1.00 0.23 1.03 0.05 0.55 Effective
38 20.00 0.85 4.29 34 3.20 0.81 2.77 0.16 0.95 Effective
39 20.00 1.32 2.39 65 7.30 1.64 3.86 0.37 1.24 Ineffective
40 20.00 1.90 1.85 65 15.00 3.31 6.35 0.75 1.74 Ineffective
43 particles float in 6.50 0.80 4.29 67 3.20 0.68 2.07 0.46 0.85 Effective
44 fine soil matrix 6.50 1.50 2.39 51 7.30 1.77 4.41 1.04 1.18 Effective
45 6.50 2.40 1.85 53 15.00 3.52 7.12 2.14 1.47 Ineffective
48 particles float in 1.20 0.80 4.29 8 3.20 0.91 3.32 2.67 1.13 Ineffective
50 1.20 0.78 1.85 46 15.00 3.64 7.58 12.50 4.67 Ineffective

51 Base Soil South East 1.29 0.04 0.04 1.47 50 0.68 0.16 0.29 15.45 3.67 Ineffective Indraratna et al.
52 Asian Lateritic 0.04 0.04 1.45 50 0.23 0.05 0.09 5.18 1.29 Effective (1996)
53 soil 0.04 0.04 1.28 50 0.12 0.03 0.05 2.72 0.68 Effective
54 Base Soil Non-cohesive 2.86 0.42 0.32 2.87 90 0.75 0.15 0.32 1.78 0.46 Effective Indraratna & Vafai
55 0.42 0.42 2.87 90 4.21 0.85 1.71 9.94 2.00 Ineffective (1997)
56 Base Soil Non-cohesive 11.40 1.18 0.50 1.30 70 5.10 1.07 1.60 4.32 2.14 Ineffective Current study
57 9.33 1.40 0.78 1.20 70 5.00 1.26 1.46 3.57 1.62 Ineffective
58 1.40 0.19 5.23 70 1.40 0.05 0.16 0.18 0.27 Effective
59 1.19 0.40 0.40 1.44 70 1.30 0.25 0.34 3.25 0.62 Effective
60 0.40 0.40 1.18 70 2.50 0.45 0.55 6.25 1.11 Effective
61 1.18 0.28 0.28 1.17 70 3.60 0.64 0.78 12.86 2.27 Ineffective
62 1.28 0.12 0.12 1.28 70 0.47 0.10 0.14 4.01 0.83 Effective
63 0.12 0.12 1.28 70 0.58 0.12 0.18 5.00 1.03 Effective
64 0.12 0.12 1.28 70 0.70 0.14 0.21 6.02 1.24 Effective
65 B1 Non-cohesive 2.64 1.55 1.30 1.21 70 7.30 1.48 2.11 4.71 1.13 Ineffective
66 1.55 0.67 3.12 70 1.40 0.31 0.78 0.90 0.46 Effective
68 1.28 0.37 4.48 70 0.78 0.16 0.47 0.61 0.44 Effective
70 0.75 0.51 3.12 70 1.40 0.31 0.78 1.87 0.60 Effective


Test # + Reference
Notation Type Cu d 85+ d 85mod Cu R d (%) D 15 Dc35 Dc95 /d 85 /d 85mod Observation
72 1.80 0.50 3.12 70 1.40 0.31 0.78 0.78 0.62 Effective
73 Base Soil Cohesive 21.43 0.41 0.41 1.26 70 5.20 1.07 1.60 12.70 2.62 Ineffective
74 0.42 0.36 2.88 70 2.50 0.49 0.79 6.11 1.37 Ineffective
75 0.41 0.25 1.94 70 0.92 0.20 0.41 2.25 0.82 Effective
76 33.52 0.26 0.22 2.06 70 0.92 0.19 0.39 3.57 0.88 Effective
77 0.26 0.26 1.25 70 5.10 1.04 1.55 19.77 4.02 Ineffective
78 0.26 0.26 1.92 70 2.10 0.43 0.83 8.14 1.66 Ineffective
79 Base Soil Dispersive 20.00 0.05 0.05 1.38 70 1.35 0.28 0.45 27.00 5.63 Ineffective
80 0.04 1.94 70 0.75 0.17 0.33 15.00 3.80 Ineffective
81 0.04 2.78 70 0.42 0.09 0.22 8.30 2.23 Ineffective
82 0.04 4.10 70 0.23 0.05 0.13 4.56 1.29 Effective
83 0.03 5.71 70 0.13 0.02 0.07 2.50 0.83 Effective

6.5 Model Comparison with Current Professional Practices
All eighty-three data sets are analyzed based on the two well-known current design
guidelines applied in professional practices, namely NRCS (1994) and Lafleur procedure
outlined in ICOLD (1994) as described earlier in Section 2.4. As the current model cannot
be compared directly with these guidelines in the manner as it is compared with the classic
Terzaghi filter criterion, all relevant parameters are calculated and tabulated in Table 6.2. It
is interesting to note that the model predictions are largely in agreement with the
evaluations of the professional guidelines. Only 3 out of 83 data sets considered in this
analysis, which were observed to be ineffective in the laboratory, were adjudged to be
effective when the filter effectiveness is assessed using the professional guidelines,
resulting in unsafe designs.
Out of forty-four effective tests, Lafleur procedure evaluated only twenty-three tests
correctly (i.e. effective tests as effective and remaining successful tests as ineffective). In
this respect, Lafleur procedure can obviously be considered as a conservative approach. As
discussed earlier in Section 2.5, it is too conservative to use d50 as a self-filtering size dSF
even for those base soils for which d80 is a more appropriate size. However, on the other
side, an ineffective test (i.e. test #57) is assessed to be effective by this procedure. This is
because some base soils with Cu less than 20 may exhibit a filtration characteristic of
broadly-graded base soils (Lafleur 1984). The base soil of this test has the largest base
particle size smaller than 4.75mm so this is less likely to be segregated. It has still a large

Cu value of about 10. Moreover, the self-filtering base particle size is not a fixed parameter
for a given base soil. It also depends upon the size and grading of the filters.
In a similar manner, NRCS procedure assessed twenty-seven effective tests correctly. In
this respect, this can also be considered as a conservative approach similar to Lafleur
procedure. However, there are three ineffective tests (i.e. #39, #57, and #81) which are
assessed to be effective, resulting in unsafe designs. In the case of test #39, which is a
combination of a linearly broadly-graded base soil with a high Cu value of 100 and a
uniform coarse filter with D15 of 7.3mm, regraded Cu (=35) is still large enough for the base
soil to be called a broadly-graded base soil (i.e. Cu >20) and for this reason, the regraded
d85 (i.e. d85R) is still the larger than the actual self-filtering size. Moreover, in some uniform
coarse filters, the actual representative filter particle size is larger than D15. Similarly, in the
case of test #57, the largest base particle size is just smaller than 4.75mm so the NRCS
guidelines exempt regrading. However, the soil is significantly well-graded with relatively
a high Cu value of 10 and still a category 4 base soil with less than 15% fines content. This
clearly exhibits the limitation of regrading by 4.75mm size. Foster and Fell (2001) also
recorded similar observations. Lastly, with regard to test #81, the recommended criterion of
effective filters is D15/d85R 9. This boundary is usually too large for the base soils in this
category, particularly in the case of dispersive soils. Both Lafleur (ICOLD 1994) and Foster
and Fell (2001) recommend much finer boundaries (i.e. D15 <0.2mm or D15/d85R 6.4).
Besides, NRCS (1994) guideline does not elaborate about filtration of internally unstable

base soils such as gap-graded and other unstable base soils. However, in this analysis,
normal procedure has been applied to the finer fraction of the base soils.

Table 6.2 Comparative study of existing design criteria with the current model (¥=effective, X=ineffective)
Max D 15 Results
Test # D 15 D c35 d 85 d 85R d SF d 85mod %Fines Current Reference
NRCS Lafleur NRCS Lafleur Lab
Model
1 0.26 0.03 7.00 0.70 0.28 0.06 24.00 2.04 1.12 ¥ ¥ ¥ ¥ Lafleur
2 1.00 0.17 7.00 0.70 0.28 0.35 24.00 2.04 1.12 ¥ ¥ ¥ ¥ (1984)
3 3.20 0.67 7.00 0.70 0.28 0.54 24.00 2.04 1.12 x x x x
4 7.30 1.59 7.00 0.70 0.28 0.58 24.00 2.04 1.12 x x x x
5 15.00 3.24 7.00 0.70 0.28 0.90 24.00 2.04 1.12 x x x x
6 0.26 0.05 19.00 0.06 0.10 0.07 70.00 0.70 0.40 ¥ ¥ ¥ ¥ Lafleur et al.
7 1.00 0.14 19.00 0.06 0.10 0.07 70.00 0.70 0.40 x x x x (1989)
8 3.20 0.61 19.00 0.06 0.10 0.08 70.00 0.70 0.40 x x x x
9 7.30 1.59 19.00 0.06 0.10 0.18 70.00 0.70 0.40 x x x x
10 15.00 2.74 19.00 0.06 0.10 0.18 70.00 0.70 0.40 x x x x
11 0.26 0.03 7.50 0.85 0.13 0.06 57.00 0.70 0.52 ¥ ¥ ¥ ¥
12 1.00 0.21 7.50 0.85 0.13 0.28 57.00 0.70 0.52 x x ¥ ¥
13 3.20 0.68 7.50 0.85 0.13 0.60 57.00 0.70 0.52 x x x ¥
14 7.30 1.50 7.50 0.85 0.13 0.92 57.00 0.70 0.52 x x x x
15 15.00 2.84 7.50 0.85 0.13 1.90 57.00 0.70 0.52 x x x x
16 0.26 0.03 1.50 0.60 0.09 0.08 52.00 0.70 0.36 ¥ ¥ ¥ ¥
17 1.00 0.21 1.50 0.60 0.09 0.32 52.00 0.70 0.36 x x ¥ ¥
18 3.20 0.84 1.50 0.60 0.09 0.42 52.00 0.70 0.36 x x x x
19 7.30 1.71 1.50 0.60 0.09 0.52 52.00 0.70 0.36 x x x x
20 15.00 3.45 1.50 0.60 0.09 0.62 52.00 0.70 0.36 x x x x
21 0.26 0.03 17.00 0.21 0.04 0.07 68.00 0.70 0.15 ¥ x ¥ ¥
22 1.00 0.17 17.00 0.21 0.04 0.30 68.00 0.70 0.15 x x ¥ ¥
23 3.20 0.74 17.00 0.21 0.04 0.85 68.00 0.70 0.15 x x ¥ ¥
24 7.30 1.74 17.00 0.21 0.04 0.85 68.00 0.70 0.15 x x x x
25 15.00 3.07 17.00 0.21 0.04 0.85 68.00 0.70 0.15 x x x x

Table 6.2 Comparative study of the current model with existing design criteria (contd.)
Max D 15 Results
Model
26 0.26 0.03 11.00 1.80 0.47 0.08 39.00 0.96 1.88 ¥ ¥ ¥ ¥
27 1.00 0.17 11.00 1.80 0.47 0.28 39.00 0.96 1.88 x ¥ ¥ ¥
28 3.20 0.68 11.00 1.80 0.47 0.85 39.00 0.96 1.88 x x ¥ ¥
29 7.30 1.57 11.00 1.80 0.47 1.32 39.00 0.96 1.88 x x x ¥
30 15.00 3.17 11.00 1.80 0.47 1.90 39.00 0.96 1.88 x x x x
31 0.26 0.03 3.00 0.70 0.12 0.09 42.00 0.70 0.48 ¥ ¥ ¥ ¥
32 1.00 0.20 3.00 0.70 0.12 0.28 42.00 0.70 0.48 x x ¥ ¥
33 3.20 0.69 3.00 0.70 0.12 0.45 42.00 0.70 0.48 x x x ¥
34 7.30 1.79 3.00 0.70 0.12 0.60 42.00 0.70 0.48 x x x x
35 15.00 2.70 3.00 0.70 0.12 0.55 42.00 0.70 0.48 x x x x
36 0.26 0.03 20.00 2.80 1.80 0.07 23.00 7.84 7.20 ¥ ¥ ¥ ¥
37 1.00 0.23 20.00 2.80 1.80 0.42 23.00 7.84 7.20 ¥ ¥ ¥ ¥
38 3.20 0.81 20.00 2.80 1.80 0.85 23.00 7.84 7.20 ¥ ¥ ¥ ¥
39 7.30 1.64 20.00 2.80 1.80 1.32 23.00 7.84 7.20 ¥ x x x
40 15.00 3.31 20.00 2.80 1.80 1.90 23.00 7.84 7.20 x x x x
41 0.26 0.03 6.50 2.00 0.50 0.08 22.00 5.96 2.00 ¥ ¥ ¥ ¥
42 1.00 0.23 6.50 2.00 0.50 0.60 22.00 5.96 2.00 ¥ ¥ ¥ ¥
43 3.20 0.68 6.50 2.00 0.50 0.80 22.00 5.96 2.00 ¥ x ¥ ¥
44 7.30 1.77 6.50 2.00 0.50 1.50 22.00 5.96 2.00 x x x ¥
45 15.00 3.52 6.50 2.00 0.50 2.40 22.00 5.96 2.00 x x x x
46 0.26 0.04 1.20 0.70 0.21 0.14 22.00 2.21 0.84 ¥ ¥ ¥ ¥
47 1.00 0.22 1.20 0.70 0.21 0.48 22.00 2.21 0.84 ¥ x ¥ ¥
48 3.20 0.91 1.20 0.70 0.21 0.80 22.00 2.21 0.84 x x x x
49 7.30 1.74 1.20 0.70 0.21 0.70 22.00 2.21 0.84 x x x x
50 15.00 3.64 1.20 0.70 0.21 0.78 22.00 2.21 0.84 x x x x

Max D 15 Results
Model
51 0.68 0.16 0.04 0.04 0.04 0.04 100.00 0.40 0.18 x x x x Indraratna et al.
52 0.23 0.05 0.04 0.04 0.04 0.04 100.00 0.40 0.18 ¥ x x ¥ (1996)
53 0.12 0.03 0.04 0.04 0.04 0.04 100.00 0.40 0.18 ¥ ¥ ¥ ¥
54 0.75 0.15 0.42 0.42 0.42 0.32 0.00 1.68 1.68 ¥ ¥ ¥ ¥ Indraratna &Vafai
55 4.21 0.85 0.42 0.42 0.42 0.42 0.00 1.68 1.68 x x x x (1997)
56 5.10 1.07 1.18 1.18 1.18 0.50 13.00 4.72 4.72 x x x x Current study
57 5.00 1.26 1.40 1.40 1.40 0.78 12.00 5.60 5.60 ¥ ¥ x x
58 1.40 0.05 1.40 1.40 1.40 0.19 12.00 5.60 5.60 ¥ ¥ ¥ ¥
59 1.30 0.25 0.40 0.40 0.40 0.40 0.00 1.60 1.60 ¥ ¥ ¥ ¥
60 2.50 0.45 0.40 0.40 0.40 0.40 0.00 1.60 1.60 x x x ¥
61 3.60 0.64 0.28 0.28 0.28 0.28 0.00 1.12 1.12 x x x x
62 0.47 0.10 0.12 0.12 0.12 0.12 0.00 0.46 0.46 x x ¥ ¥
63 0.58 0.12 0.12 0.12 0.12 0.12 0.00 0.46 0.46 x x x ¥
64 0.70 0.14 0.12 0.12 0.12 0.12 0.00 0.46 0.46 x x x ¥
65 7.30 1.48 1.55 1.55 1.55 1.30 0.00 6.20 6.20 x x x x
66 1.40 0.31 1.55 1.55 1.55 0.67 0.00 6.20 6.20 ¥ ¥ ¥ ¥
67 7.30 1.48 1.28 1.28 1.28 1.20 0.00 5.12 5.12 x x x x
68 0.78 0.16 1.28 1.28 1.28 0.37 0.00 5.12 5.12 ¥ ¥ ¥ ¥
69 7.30 1.48 0.75 0.69 0.75 0.65 15.00 2.76 3.00 x x x x

Max D 15 Results
Model
70 1.40 0.31 0.75 0.69 0.75 0.51 15.00 2.76 3.00 ¥ ¥ ¥ ¥
71 7.30 1.48 1.80 1.50 1.80 1.15 10.00 6.00 7.20 x x x x
72 1.40 0.31 1.80 1.50 1.80 0.50 10.00 6.00 7.20 ¥ ¥ ¥ ¥
73 5.20 1.07 0.41 0.41 0.41 0.41 47.00 0.70 0.40 x x x x
74 2.50 0.49 0.42 0.41 0.41 0.36 47.00 0.70 0.40 x x x x
75 0.92 0.20 0.41 0.41 0.41 0.25 47.00 0.70 0.40 x x ¥ ¥
76 0.92 0.19 0.26 0.26 0.26 0.22 47.00 0.70 0.40 x x ¥ ¥
77 5.10 1.04 0.26 0.26 0.26 0.26 47.00 0.70 0.40 x x x x
78 2.10 0.43 0.26 0.26 0.26 0.26 47.00 0.70 0.40 x x x x
79 1.35 0.28 0.05 0.05 0.05 0.05 93.00 0.45 0.20 x x x x
80 0.75 0.17 0.05 0.05 0.05 0.04 93.00 0.45 0.20 x x x x
81 0.42 0.09 0.05 0.05 0.05 0.04 93.00 0.45 0.20 ¥ x x x
82 0.23 0.05 0.05 0.05 0.05 0.04 93.00 0.45 0.20 ¥ x x ¥
83 0.13 0.02 0.05 0.05 0.05 0.03 93.00 0.45 0.20 ¥ ¥ ¥ ¥

In comparison, the current model evaluates thirty-five effective tests correctly and
discards the remaining nine effective tests as failed. Most of these tests were associated
either with very uniform base soils or with tests that, though called to be effective,
showed limited erosion, especially in the beginning of filtration. In this respect, the
current model is more comprehensive and based on sound analytical principles, hence,
takes care of most of issues or deficiencies that exist in the current professional
practices.
The current design practices are based on empirical considerations and thus suffer from
serious limitations. Regrading by a specific base particle size, namely the ASTM No. 4
sieve size (i.e. 4.75mm) or the use of smaller self-filtration base size such as d50 is
considered based on assumption that the self-filtering base particle size is a fixed
parameter. However, the actual self-filtering base size is a variable filter parameter,
which depends on both the grading, compaction and size of filter materials and can be
given by the self-filtering constriction size, Dc95. It is the fixed value of self-filtering
fraction that introduces imbalance in the filter designs based on the existing professional
guidelines. On one side, this makes some designs conservative (i.e. over-safe designs)
whereas on the other side, the same, results in some unsafe designs. In addition, the use
of D15 as the representative filter particle size often makes some designs, particularly
involving well-graded filters, conservative whereas the same design may result in unsafe

designs in the case of uniform coarse filters. In contrast, the current model is based on
analytical principles and provides reasonably conservative designs.
NRCS (1994) does not discuss in detail about the filtration of internally unstable gap-
graded or other base soils. In the above analysis, the normal NRCS procedure is applied
to the finer fraction of the base soils. The current model also considers only the finer
fraction of the base soils. Further discussion, in this relation is included in Chapter 7
entitled “Discussion and Analysis”.

CHAPTER
SEVEN
DISCUSSION AND ANALYSIS OF FILTER CRITERIA
7.1 Introduction
According to the earlier discussions in Chapter 2, it is clear that certain deficiencies exist
in the filter designs that are based on current professional practices. This is largely
because the current design guidelines are purely empirical and based on the particle sizes
of the filter. There is enough evidence to support the theory that the constriction sizes are
more important parameters than the particle sizes. Based on sound analytical principles
and laboratory observations, the current study has suggested three constriction-based
filter design criteria. In this respect, it is relevant to discuss in detail the relative
advantages and scope of application of these criteria in relation to the existing
professional practices.
As discussed earlier in Section 2.5, there are some ambiguities associated with the
filtration of cohesive soils. There are several test observations on the filtration of
broadly-graded cohensionless base soils with very high Cu values usually larger than 20.
There are not many test observations on well-graded cohensionless base soils with a
relatively smaller Cu value (i.e. 20). Karpoff (1955), later adopted by USBR, conducted
Chapter 7: Discussion and Analysis of Filter Criteria 177

several tests on such base soils. However, the recent studies such as Sherard et al.
(1984a), Fell and Foster (2001) reported that the USBR failure criteria were very
conservative; Majority of USBR tests, which were evaluated to be ineffective in the
laboratory, did not fail. In contrast, Honjo and Veneziano (1989) indicated through
statistical analyses of some 287 test data from the published research that although some
of USBR test observations might have been conservative, most of them were very close
to the success-failure boundary. In this respect, a series of tests were conducted during
this study to examine filtration of well-graded cohensionless, and dispersive and non-
dispersive cohesive base soils. Controversial USBR data are analysed using the current
filter design criteria in order to resolve the issues. Laboratory approach and its findings,
and failure criteria used in this laboratory study are also discussed.
7.2 Scope of Application of Constriction-based Design Criteria
7.2.1 Surface Area Filter Criterion
The studies such as Humes (1996) and Schuler (1996) showed that the filter PSD based
on surface area of particles is more realistic in assessing filter effectiveness. Some
detailed analyses and discussions in Chapter 3 further highlighted the relevance and
appropriateness of the investigation in regard to the filter. The surface area concept was
successfully applied to the granular filter materials such as sands and gravels. It is clear
from Figures 1.2 and 2.18 that there are several instances where these base soils may be
sands and gravels. This reasoning led to the application of surface area concept to the
base soils to investigate filter effectiveness. It was found that the filter design criterion
thus developed (i.e. Dc35<d85SA) can be successfully applied to assess the filtration of the
cohensionless base soils, particularly the Category 4 base soils (i.e. sands and gravels)

(NRCS 1994) where the fines content (i.e. <ASTM No. 200 sieve size) is less than 15%.
This finding further validates the application of theory of surface area in modelling the
granular filter. The issue of application of this concept to finer base soils such as silts and
clays, hence, must be considered.
As described in Section 3.4, if a soil PSD is divided into n different classes based on the
particle sizes Di, and pmi represents the corresponding mass fraction of a particular class i,
then the corresponding surface area fraction, pSAi, is given by:
p mi
D
p SAi = i=n i (7.1)
p mi
¦
i =1 Di
The above equation has been reproduced here from Section 3.4 for convenience in
presentation and can be re-written as:
§ 6
p mi ·
¨¨ ¸¸
= i =n © ¹
6 Di
p SAi (7.2)
p mi
¦
i =1 Di
6
The fraction in parenthesis, , represents the specific surface area of the particle.
Di
Because of extremely small particle sizes of cohesive soils (i.e. high specific surface
area), particularly clays, finer particles are over-represented by the surface area
considerations, resulting in unreasonably small d85SA. This makes the filter designs

involving fine soils conservative. However, it was seen earlier that the surface area-based
filter criterion can successfully demarcate the success-failure boundary in lateritic soils
where fines content (i.e. particles less than 75 µm) is 100%. This is because the lateritic
soil is mostly uniformly-graded silts (Figure 4.5a) where as described previously in
Chapter 3, various frequency considerations do not make any significant difference. This
further substantiates the claims made by the studies such as Sherard et al. (1984b) and
Sherard and Dunnigan (1985) that the plasticity of the soils is not an important parameter
in relation to the filtration of the soils. It is the grading of the base soils that govern their
filtration characteristics. However, the plasticity of the core soils certainly determines the
mode of erosion in embankment dams. Cracking of core is a common erosion mechanism
in dams involving cohesive core soils. In contrast, cracking is not evident in non-plastic
cohensionless core soils.
7.2.2 Self-filtration Stability Filter Criterion
As described in Chapter 5, it is found that the self-filtering constriction size, Dc95, forms a
rational basis to determine the fraction of the base soils that influences filtration. This
concept combined with fundamental principles of soil mechanics provides a procedure to
determine the PSD of self-filtration layer. The assessment of internal stability of the
self-filtration layer gives rise to a filter criterion. Although this criterion is based on
sound analytical considerations, Kenney and Lau (1985) procedure is used to evaluate the
stability. This procedure is mainly developed and verified on the basis of granular soil
filtration. For this reason, the scope of this criterion is also limited to the filtration of
sands and gravels as illustrated in Section 5.5.

As discussed in Chapter 2, Kenney and Lau (1985) initially suggested the relationship,
H/F=1.3, as stability criterion. Later, it was amended to H/F=1, simply quoting that the
conditions used in deriving the earlier relationship were very severe. Therefore, it is
important to assess the degree of conservativeness associated with this stability criterion.
Consider a soil as shown in Figure 7.1, the PSD of which is log-linear such that it
maintains H/F=1 for any value of particle size D. The slope of PSD curve can be
obtained by two ways. Firstly, determine the slope by considering two points D10 and
D60, and say it S1. Then,
Soil PSD
Percent Finer
F4D
H=FD
FD
F=FD
D 4D
Particle Size D
Figure 7.1 Linearly-graded soil having PSD with H/F=1
60 − 10
S1 = (7.3)
log(D60 ) − log(D10 )

Equation (7.3) can be simplified to:
50
S1 = (7.4)
log(C u )
Secondly, determine the slope by considering two points D and 4D, and say it S2. Then,
F4 D − FD
S2 = (7.5)
log(4 D ) − log(D )
Equation (7.5) can be simplified to:
FD
S2 = (7.5)
log(4 )
In order for the soil to be internally stable, S1 S2. This provides:
50 FD
≥ (7.6)
log(C u ) log(4 )
The procedure also prescribes that in well-graded stable soils, H/F1 must hold in the
range F<20. In this respect, putting FD equal to 20 in Equation (7.6) and solving it
reveals that Cu32. This implies that all log-linear soils with Cu larger than 32 will be
internally stable. This seems to be unrealistic, making the stability assessment procedure
conservative. For this reason, the filter criterion based on this approach may sometimes
lead to slightly conservative designs, particularly when applied to broadly-graded base

soils. This criterion can be enhanced by incorporating more realistic stability assessment
method. However, as discussed earlier in Chapter 5, drawing the PSD of self-filtration
layer alone provides a valuable picture of filter effectiveness and will certainly boost the
designer’s confidence.
7.2.3 Modified Base PSD Filter Criterion
There are four major limitations associated with the current professional practices, as
listed below.
• They do not differentiate between uniform and well-graded filters provided they
have identical D15 values.
• They ignore the filter compaction.
• They are unrealistic with well-graded and broadly-graded base soils.
• They are ambiguous in relation to filtration of cohesive base soils.
As described in Section 3.8, the use of Dc35 as the controlling constriction size of the
filter in the filter criterion, Dc35<d85mod, successfully addresses the first two issues
whereas the use of Dc95 as self-filtering constriction size resolves remaining issues. In
this regard, the filter criterion, Dc35<d85mod, convincingly addresses these limitations in
the current practices and provides a simple, yet comprehensive design method. As
illustrated earlier in Sections 6.4 and 6.5, this method convincingly demarcates the
boundary in most core soils used in dams and drainage structures, without being
unreasonably conservative. The fact that a single filter design criterion can be applied to
all soil types also substantiates the earlier observations that plasticity of the soil is not a
significant parameter in relation to its filtration behaviour (Indraratna et al. 1996; Sherard
et al. 1984a; Foster and Fell 2001).

7.3 Analysis of Controversial USBR data
Karpoff (1955), later adopted by USBR, conducted a series of filtration tests on different
types of uniform and well-graded base soils ranging from clayey silt to medium sand
using uniform and well-graded filters. Laboratory observations showed that the tests with
retention ration, D15/d85, from 4 to 11 failed to retain the base soils. Sherard et al. (1984a)
also conducted several tests on uniform sands and found that none of tests with retention
ratio, D15/d85, less than 9 failed. This contradicted the USBR observations. It was
reported that the study reproduced some of USBR tests and found that they did not fail.
Sherard et al. (1984a) argued that the USBR failure criteria are too conservative and in
reality, none of the alleged failed tests actually failed.
Honyo and Veneziano (1989) conducted statistical analyses of published laboratory test
data using stepwise regression and found that the filter effectiveness decreases as non-
uniformity of base soils increases i.e. base soils becomes well-graded. In this respect, the
study suggested that although some of failed USBR tests might not have failed, most of
them are likely to be close to critical conditions i.e. on boundary. However, more
recently, Foster and Fell (2001) also conducted statistical analysis of laboratory tests data
from various sources and found that the USBR data were not compatible with other data.
The study argued that the USBR conducted conventional filtration tests on low
permeability soils where erosion did not occur at all or it took place on preferential paths
along the wall of permeameter. For this reason, the study discarded these data from the
analysis.

Repeating laboratory tests to verify other similar tests may not be an appropriate
approach because laboratory approaches very often involve some kinds of biases.
Statistical analyses often do not explain the process and are biased to the kind of data
which has majority. In the past, more laboratory investigations were carried out on
uniform base and filter materials. For this reason, the analyses may favor the results of
tests on uniform materials. In this respect, it can be said that the controversy regarding
the USBR data exists.
Now, when rational design criteria have been developed, it is possible to resolve this
controversy. The filter criteria based on surface area and self-filtration can not be used
for this because they are limited in scope to cohensionless soils only. The filter criterion
based on modified base PSD, Dc35<d85mod, is used to analyse the USBR data. The results
are shown in Figure 7.2 and also tabulated in Table 7.1. The results reveal that some of
failed tests such as #18, #19 and #24 with Dc35/d85mod values of 0.37, 0.33 and 0.80
respectively did not fail, substantiating the findings of the above studies. However, it is
not true that all tests did not fail.
By now, it is clear that the tests on uniform base soils with retention ratio, D15/d85, over 9
are most likely to fail, whereas this ratio decreases in the case of more widely-graded
base soils and is found to be well below 4 in the case of broadly-graded base soils. In this
respect, failure of tests #5, #8 and #11, where the ratio D15/d85 is above 10 and Dc35/d85mod
values almost equal to 2 or more than 2 in all cases, is not contradictory to any
established principles of filtration. Regarding test #4, where base soil is well-graded with
Cu value equal to 7 with D15/d85 ratio equal to 5.5 and Dc35/d85mod value well above 1, it is
certainly not a case that can be considered as a successful test with confidence. Some of

effective test data plot on the ineffective region. Most of them include uniform base soils,
where such results are predictable, considering the inherent conservativeness of the
current filter criterion as discussed earlier in Section 6.4 and 6.5. Thus, it can be
concluded that the USBR test criteria are conservative; however, except three tests #18,
#19 and #24, all other USBR test observations are realistic.
10
Note: Refer to Table 7.1 for details of test numbers
=1
A
5
/d 85S
e D c3
tiv 11
1 f fe c 10 24
Ine
Dc35 (mm)
8 12
5 e
9 tiv
ec
17
7
2 4 Ef f
0.1 6
3
18
15 19 23 Laboratory Observations
1 1416 20
21 22 Effective
13
Ineffective
0.01
0.01 0.1 1 10
d 85mod (mm)
Figure 7.2 Analysis of USBR test data using the current design criteria

Table 7.1 Details of USBR test data including their filtration analysis

Test #
Notation Cu d 85+ d 85mod Cu R d (%) D 15 + Dc35 Dc95 /d 85 /d 85mod Observation
1 Series A 7.00 0.12 0.05 1.15 70 0.17 0.03 0.06 1.42 0.75 Effective
2 0.07 1.15 70 0.31 0.06 0.09 2.61 0.91 Effective
3 0.10 1.33 70 0.45 0.08 0.13 3.75 0.85 Effective
4 0.12 1.75 70 0.66 0.14 0.22 5.50 1.14 Ineffective
5 0.12 1.70 70 1.49 0.29 0.45 12.42 2.39 Ineffective
6 Series B 1.38 0.13 0.12 1.27 70 0.41 0.08 0.13 3.15 0.72 Effective
7 0.13 1.43 70 0.66 0.14 0.22 5.08 1.06 Effective
8 0.13 1.52 70 1.49 0.29 0.45 11.46 2.21 Ineffective
9 Series C 1.54 0.62 0.43 1.38 70 1.38 0.30 0.47 2.23 0.70 Effective
10 0.62 1.29 70 5.20 1.04 1.48 8.39 1.67 Effective
11 0.62 1.78 70 6.50 1.16 1.85 10.48 1.87 Ineffective
12 Series D 11.40 2.10 0.45 1.93 70 1.40 0.32 0.71 0.67 0.72 Effective
13 Series A1 7.00 0.13 0.04 3.33 70 0.10 0.02 0.05 0.75 0.49 Effective
14 0.06 4.55 70 0.13 0.03 0.07 0.98 0.43 Effective
15 0.08 6.67 70 0.20 0.03 0.11 1.51 0.37 Effective
16 0.06 9.17 70 0.15 0.03 0.08 1.13 0.42 Effective
17 0.13 12.00 70 0.35 0.04 0.16 2.64 0.27 Effective
18 0.13 7.09 70 0.78 0.05 0.35 5.86 0.37 Ineffective
19 Series B1 7.00 0.13 0.11 19.51 70 0.72 0.04 0.21 5.43 0.33 Ineffective
20 0.09 20.50 70 0.35 0.03 0.12 2.64 0.31 Effective
21 0.06 18.33 70 0.17 0.02 0.08 1.28 0.41 Effective
22 0.07 7.00 70 0.13 0.02 0.08 0.98 0.35 Effective
23 Series C1 22.73 1.50 0.16 27.50 70 1.50 0.04 0.22 1.00 0.27 Effective
24 1.20 5.60 70 6.10 0.96 3.69 4.07 0.80 Ineffective
7.4 Filtration of Gap-graded Base Soils
While evaluating filter effectiveness in the case of internally unstable gap-graded or
concave upward base soils, it is customary to consider only the fine fraction of the gap.
However, it is not necessary to do so when self-filtration stability filter criterion is used.
For example, consider a gap-graded base soil is tested against two filters (Figure 7.3).
The gap is more than 4 so the soil is internally unstable (Kenney and Lau 1985). Filters
are assumed to be compacted to 70% relative density. For rough estimation, retention
ratios, D15/d85, are calculated for both filters, considering only the fine fraction of the
base soil and found to be 2.4 and 12.3 respectively. This implies that finer filter is
effective whereas the coarser filter is ineffective in retaining the base soil.

100 95
85
80 SL=Self-filtration Layer
Rd=70%
Fine Fraction
Percent Finer
CSD (C)
60 CSD (F)
Coarse
SL-PSD (F)
40 35% PSD (Base)
SL-PSD (C)
20 15%
Fine Filter (F) Coarse Filter (C)

0
0.01 0.1 1 10
Figure 7.3 Analysis of an internally unstable gap-graded base soil
The PSDs of self-filtration layers in the case of both coarse and fine filters are
determined according to the procedure described in Chapter 5. The analyses are also
shown in Figure 7.3. It is clear that the gap in finer self-filtration layer PSD is smaller
than 4 so when stability criterion H/F=1 is applied to evaluate the stability of this self-
filtration layer, it is effective. In contrast, the gap in coarser self-filtration layer is larger
than 4, implying that the coarse filter is ineffective. This illustrates that any prior
assessment of internal stability of the base soil is not necessary when the self-filtration
stability filter criterion is used to determine the filter effectiveness. Similarly, the
modified base PSD filter criterion does not need any prior assessment of internal stability
of the base soil if Dc95 lies in the gap or below it. However, the surface area filter
criterion does not have this advantage.

7.5 Laboratory Study of Filtration
7.5.1 Laboratory Approach and Failure Criteria
As mentioned in last three chapters, in addition to using published data from literature,
several laboratory tests were also carried out during this study at the University of
Wollongong. Base soils used were both cohesive and non-cohesive. They were uniform
and well-graded. Filters used were fine sand to coarse gravel. They were also uniform
and well-graded. Two different laboratory arrangements were used for fine and coarse
soils. For filtration of non-cohesive soils where filters were coarse, the large-scale
permeameter as shown in Figure 7.4 was used, whereas for fine cohesive soil filtration,
smaller permeameter as shown in Figure 7.5 was used. Dry filter materials were first
placed layer by layer and compacted using a procedure similar to the one Karpoff (1955)
used to simulate light rolling of field installations. Density and void ratios were measured
in few cases and relative densities were determined. Relative densities were found to be
very close to 70%. For this reason, a relative density of 70% has been used in all tests
carried out at the University of Wollongong. Base soil, usually 30-60mm thick, was
placed on the top of the filter using similar procedure. Base soil density or void ratio was
not measured. However, in the case of cohesive soil filtration tests, a hole ranging from
3mm to 6mm were made to simulate core cracking as prescribed by Sherard et al.
(1984b).

Figure 7.4 Large-scale filter permeameter
Figure 7.5 Small-scale high pressure permeameter

A hydraulic head of 250-300 kPa across the base soil-filter arrangement was applied.
Mild vibration was imparted to the permeameter by using rubber mallet at the interval of
3-5 minutes. Discharge and effluent turbidity were measured.
Effective and ineffective tests were decided based on the criteria described in Section
2.2.1 using the approach used by Indraratna et al.(1996). A typical discharge and effluent
turbidity in effective and ineffective tests as observed in two of tests are shown in Figure
7.6 and 7.7. It is clear from Figure 7.6 that in effective tests, discharge decreases and
stabilises with time while turbidity quickly reduces to that of clean tap water, showing no
more erosion. In contrast, in ineffective tests (Figure 7.7), discharge does not decrease
appreciably. Sometimes, it increases rather. Turbidity reduces but continues for quite
some time. Higher values of turbidity were measured on vibration, as shown by peaks.
150 6
5
Discharage (mL/s)
Turbidity (NTU)
100 4
50 2
0 0
0 5 10 15 20 25 30
Time (min)
Figure 7.6 Discharge and turbidity pattern in effective tests

250 100
200 80
Discharg
Discharge (mL/s)
Turbidity (NTU)
150 60
100 40
Turbidity
50 20
0 0
0 10 20
Time (min)
Figure 7.7 Discharge and turbidity pattern in ineffective tests
7.5.2 Laboratory Observations
Several tests on broadly-graded cohensionless base soils were conducted by the studies
such as Lafleur (1984) and Lafleur (1989). Most base soils used in these studies have
Cu>20. Karpoff (1955) conducted several tests on well-graded base soils Cu<20.
However, because of some controversy with regard to test procedure, it was not possible
to use these data to validate the filter criteria developed in this study. For this reason,
several tests were carried out on well-graded cohensionless base soils. Filter and base
materials used are given in Figure 7.8. The aim of these tests was to investigate if tests
with D15/d85 less than 9 fail in the case of well-graded base soils. It is clear from tests
such as #67 and #71 in Table 6.1 that the tests with D15/d85 ranging from 4 to 6 fail if the
base soils are not uniform, confirming the trend observed by Karpoff (1955), Honjo and
Veneziano (1989) and Foster and Fell (2001).

100
85%
80
Filters
Percent Finer
Base Soils
60
40
20 15%
0
0.01 0.1 1 10
Figure 7.8 Filter and base materials used in filtration of non-cohesive tests
Similarly, eleven tests on cohesive soils were conducted using the filter and base
materials shown in Figure 7.9. The aim was to investigate if any test with D15/d85>15 (i.e.
unusually high value) is found to be effective as found by Sherard (1984b), and also what
value of D15 is conservative for successful filtration of dispersive soils. It was found that
none of the tests larger than D15/d85>15 were effective, confirming the observations of
Sherard and Dunnigan (1985) and Foster and Fell (2001). In contrast, few tests (#74, #78
and #81) with D15/d85 in the range of 6 to 8 failed to retain cohesive base soils. In this
respect, D15/d85<9 as filter design criterion appears unrealistic. With regard to filtration of
dispersive soils, although only five tests were carried out, D15<0.2mm (ICOLD 1994)
and D15/d85 <6.4 (Foster and Fell 2001) were found to be reasonable. More importantly,
these tests reveal that the plasticity is not an important factor in relation to filtration
characteristics of the base soils.

100
85
%
Percent Finer 80 Dispersive
Filters
60
Base Soils
Non-dispersive
40
20 15
%
0
0.0001 0.001 0.01 0.1 1 10
Figure 7.9 Filter and base materials used in filtration of non-cohesive tests
Major outcomes of discussions and analyses in preceding sections on different aspects of
filtration are briefly summarized below.
• The filter retention criterion based on surface area is limited in application to
filtration of granular base soils such as sand and gravel. This validates the theory
that the PSD based on surface area of particles models the porous granular media in
a more rational manner.
• Kenney and Lau (1985) method for assessment of internal stability of soil is still
conservative. This makes the filter criterion based on stability of self-filtration layer
also slightly conservative when applied to highly broadly-graded base soils.
However, drawing the PSD of self-filtration layer alone will certainly boost the
designer’s confidence, particularly in critical filters.

• The filter design criterion based on the modified PSD of base soil is based on
sound analytical principles and is realistic in filtration of most base soils. This
should be used to improve the current professional filter design guidelines. A new
filter design guideline is given in Chapter 8.
• Failure criteria used by Karpoff (1955) were slightly conservative. This resulted in
some unrealistic observations. However, not all observations were contradictory as
suggested by some studies (Sherard et al. 1984a). The fact that the filter
effectiveness decreases with non-uniformity of base soils is certainly reflected by
these test observations.
• Laboratory observations on well-graded cohensionless soils clearly showed that
some filters with retention ratio, D15/d85, less than 9 fail to retain the soils. This
confirms the trend that the filter effectiveness diminishes in non-uniform base soils.
Tests on cohesive soils revealed that some base soils with retention ratio, D15/d85,
less than 9 fail. This suggests that the NRCS recommendation of filter boundary
equal to D15/d85 =9 in the Category 1 base soils is slightly coarser and hence
unrealistic (Foster and Fell 2001).

CHAPTER
EIGHT
ENHANCED DESIGN GUIDELINE AND ITS
APPLICATIONS
8.1 Introduction
As described in Sections 2.4 and 6.5, the two existing filter design guidelines in
professional practices are similar in approach so they suffer from the same limitations. The
Lafleur procedure recommends the use of a smaller representative base particle size such as
d50. The NRCS procedure adopts the use of d85 after regrading where base particles larger
than ASTM No. 4 sieve size (i.e. 4.75mm) are ignored. As discussed earlier in Chapter 6,
on one hand, these guidelines are conservative for some base soils; on the other hand, they
are unsafe for some others. However, the NRCS procedure is relatively elaborated and also
recommends controls over the grading of filter materials, in order to avoid selection of gap-
graded filter and prevent possible segregation of filters during installation.
Chapter 8: Enhanced Design Guideline and Its Applications in Practice 196

Like many other civil engineering designs, the filters are also designed in two ways. If the
potential filter materials are available at some natural sites, a common occurrence, these
guidelines are used to evaluate their effectiveness with respect to the available base soil. In
contrast, if the potential filter materials are not available before the designs are carried out,
these guidelines are used to prescribe the suitability band of the filters with respect to the
given base soil. The filter materials are compared against the prescribed band later when
they are available. In the NRCS guideline, the lower boundary of filter suitability band is
based on the permeability criterion, whereas its upper boundary is based on the retention
criteria. For the Category 1 base soils as described earlier in Chapters 2 and 7, the
prescribed upper boundary D15/d85R9 is coarser than reasonable. For the Category 2 base
soils, the boundary D150.7mm is found to be coarser for some base soils too. Similarly, the
boundary D15/d85R 4 is also found to be coarser for some Category 4 base soils. Thus, it
can be said that the upper boundary prescribed by the NRCS guideline is invariably coarser
and for this reason, it is most likely that the potential filters lie inside the prescribed band.
However, it is to be noted that for the reasons discussed earlier, not all filters lying in the
filter band determined by the existing NRCS guideline are effective. Once the approximate
filter band is calculated on the basis of the present NRCS guideline, the potential filter can
be chosen from within the filter band and the new constriction-based filter criterion can be
used for further assessment of filter effectiveness. For very coarse base soils where
substantial part of base soil PSD lies beyond ASTM No. 4 sieve size (i.e. 4.75mm), it is
recommended that the original Terzaghi retention criterion can be used to demarcate the
upper boundary. In summary, despite several limitations, the existing NRCS guideline is

still relevant and can be used a preliminary guide to identify the potential filter materials. A
potential filter is determined using the current NRCS guideline with slight modification for
very coarse base soils. Then the filter effectiveness of the chosen filter is finally evaluated
using the constriction-based filter design criterion. This makes new filter design guideline
more realistic.
8.2 New Filter Design Procedure
In the above respect, a new filter design procedure is proposed and can be summarized in
the following steps. For more details, refer to the NRCS (1994) design guideline.
Step 1
Plot the PSD curve of the base soil. Analyze it for its internal stability. If the base soil is
stable, proceed or consider only the finer fraction of the base soil and proceed to step 2.
Step 2
Proceed to Step 4 if the base soil contains no gravel (i.e. the base particles are smaller than
ASTM No. 4 sieve size). Proceed to Step 5 if the base soil contains all gravel (i.e. all
particles are larger than ASTM No. 4 sieve size).
Step 3
Regrade the base soil PSD for base particles larger than ASTM No. 4 sieve size.

Step 4
Place the base soil in a category determined by the percent finer than ASTM No. 200 sieve
size (i.e. 0.075mm) from regraded PSD, according to Table 8.1.
Table 8.1 Base soil category based on fines content

Base Fines Content Base Soil Description
Soil (%<0.075mm)
Category
1 >85 Fine silt and clays
2 40-85 Sands, silts, clays, and silty and clayey sands
3 15-39 Silty and clayey sands and gravel
4 <15 Sands and gravel
Step 5
To satisfy filtration (i.e. retention) requirements, determine the maximum D15 size for the
filter in accordance with Table 8.2.
Table 8. 2 Design criteria – Maximum D15

Base Soil Filter Criteria –Maximum D15
Category
1 9 d85R but not less than 0.2mm
2 0.7mm
3 § 40 − A · ½
®¨ ¸(4d 85 R − 0.7 ) + 0.7 ¾mm but not less than 0.7mm
¯© 40 − 15 ¹ ¿
4 4 d85R or 4d85 if all base particles >ASTM No. 4 sieve size
Note that A is fines content after regrading and d85R is regraded d85

Step 6
To satisfy the drainage (i.e. permeability) requirements, determine the minimum D15 in
accordance with Table 8.3.
Table 8.3 Design criteria – Minimum D15

Base Soil Category Permeability Criteria –Minimum D15
All categories 4 d85 but not less than 0.1mm
Step 7
The width of the allowable filter design band must be kept relatively narrow to prevent the
use of possibly gap-graded filters. Adjust maximum and minimum D15 sizes for the filter
band determined in Steps 5 and 6 so that ratio of maximum D15 to minimum D15 is 5 or less
at any given percent finer of 60 or less.
Step 8
The designed filter band must not have an extremely broad range of particle sizes to
prevent the use of possibly gap-graded filters. Adjust the limits of design filter band so that
the coarse and fine sides have a coefficient of uniformity Cu less than 6.
Step 9
Determine minimum D5 and maximum D100 sizes of the filter in accordance with Table 8.4.

Step 10
To minimize segregation during filter installations, the relationship between the maximum
D90 and the minimum D10 of the filter is important. Minimum D10 can be determined by
dividing minimum D15 by 1.2. Determine D90 in accordance with Table 8.5.
Table 8. 4 Maximum and minimum particle size criteria

Base Soil Category Maximum D100 Minimum D5
All categories 75mm ASTM No. 200 sieve size
Table 8. 5 Segregation criteria

Base Soil Category If D10 is: (mm) Then, maximum D90 is: (mm)
All categories <0.5 20
0.5-1.0 25
1.0-2.0 30
2.0-5.0 40
5.0-10.0 50
>10.0 60
Step 11
After the approximate upper and lower boundaries of filter band are determined using
above 10 steps, all selected filters within the band may not be safe. Apply any of three
constriction-based filter criteria developed in this study depending on the scope of their
application to evaluate suitability of chosen filter that lies within the filter band using

expected level of field compaction. The use of filter criterion based on the modified base
soil PSD (i.e. Dc35<d85mod) should be preferred.
Step 12
Determine the PSD curve of self-filtration layer for visual examination of suitability of
filter. Usually a conservative filter has a smaller gap whereas a less conservative filter has a
relatively wider gap. This will give a further confidence to the designer.
8.3 Practical Application
Under the joint-venture of the University of Wollongong and the Shoalhaven City Council,
a pilot study program has been initiated to use a civil engineering measure to improve acid
sulphate soil near Bomaderry, NSW. The civil engineering measure to enhance acid
sulphate soil in this way is called a Permeable Reactive Barrier (PRB). In the PRB method,
the acid sulphate water in soil is intercepted by an alkaline barrier and reacts with alkaline
reactive material in the barrier to reduce acidity in the affected soil. In this respect, a
longitudinal trench is excavated across the natural slope of the ground. It is filled with
natural gravel or concrete gravel mixed with oyster shells (i.e. alkaline reactive material).
This barrier made of gravel-sized material is highly permeable. For this reason, the method
is called a Permeable Reactive Barrier. In most instances, it is too coarse to prevent
adjacent soils from migrating into the barrier so a finer layer of sand is placed in between
adjacent soil and the reactive barrier to prevent the soil from entering into the trench and

eventually clogging the reactive barrier. This is sometimes called a Non-Reactive Barrier
(NRB).
8.3.1 Design of Non-Reactive Barrier
A team of engineers from the University of Wollongong visited the site, and a number of
soil samples were collected. The soil PSD was determined using the Malvern Particle Size
Analyzer and found to be clayey silt of Category 1. The soil PSD is shown in Figure 8.1.
Approximate band of non-reactive barrier material is determined using the guideline
procedure described earlier. Two NRB materials are chosen within the band, namely F1
and F2 as gradings of potential non-reactive materials, as shown in Figure 8.2. Usually, it is
either natural sand or manufactured concrete sand. According to existing design procedure,
both filters are acceptable because they lie within the filter band. Check these filters with
new constriction-based filter criterion Dc35<d85mod.
The filter density is assumed to be 70 % corresponding to light rolling compaction (ICOLD
1994). Parameters Dc35, Dc95 and modified base soil PSD for NRB F1 are determined and
plotted in Figure 8.3. As Dc35<d85mod, NRB F1 is effective. Dc35/d85mod ratio is found to be
0.66. Similarly, parameters Dc35, Dc95 and modified base soil PSD for NRB F2 are
determined and plotted in Figure 8.4. As Dc35>d85mod, NRB F2 is ineffective, although it lies
within the same band. Dc35/d85mod ratio is found to be 2.38. In this respect, the existing
NRCS guideline is greatly enhanced by incorporating the constriction-based filter criteria.

100
85%
80
Percent Finer
60
Acid Sulphate Soil

40
20
0 2µm 75µm
0.0001 0.001 0.01 0.1 1
Particle Size (mm)
Figure 8.1 Grading of acidity-affected soil near Bomaderry NSW within study area
100
85%
80
Finer Limit
Percent Finer
Soil
60
40
Coarser Limit
Reactive Barrier F1 Reactive Barrier F2

20 15%
0
0.0001 0.001 0.01 0.1 1 10 100
Figure 8.2 Design of non-reactive barrier of the Permeable Reactive Barrier

100 95%
85%
80
Mod. Soil PSD

Percent Finer
60
CSD (F1)
Soil PSD
40 35%
Non-Reactive Barrier F1
20
D c35 d 85mod
0
0.0001 0.001 0.01 0.1 1 10
Constriction Size Dc ; Particle Size D (mm)
Figure 8.3 Constriction Analysis of NRB F1 using Modified Base PSD method
100 95%
85%
80
Soil PSD
Percent Finer
60
CSD (F2)
Mod. Soil PSD
40 35%
20
d 85mod
D c35
0
0.0001 0.001 0.01 0.1 1 10
Constriction Size Dc ; Particle Size D (mm)
Figure 8. 4 Constriction Analysis of NRB F2 using Modified Base PSD method

8.3.2 Design of Reactive Barrier
The non-reactive barrier must be stable against the permeable reactive barrier. This is
usually a uniform natural gravel or concrete gravel ranging in size from 5mm to 20mm
depending upon the grading of adjacent affected soil. In this respect, a few uniform
gradings in this range can be selected and checked by new design criterion to identify the
suitable grading of potential reactive barrier materials. For simplicity, just two gradings of
potential reactive barrier materials, namely F3 and F4 are considered, as shown in Figure
8.5. The constriction analyses of both reactive barriers F3 and F4 are carried out and shown
in Figures 8.6 and 8.7. Analyses show that reactive barrier F4 is suitable; F3 is too coarse
for non-reactive barrier F1. Check for permeability criterion for F4. D15(F4)/d15(F1) is
found to be 40 (i.e.>4), which is quite acceptable. In this way, Figure 8.8 shows the
gradings of suitable reactive and non-reactive barriers. Several sets of such reactive and
non-reactive barriers are possible depending on the choice of designer. A designer in this
case may choose potential non-reactive barrier materials within the band different from
previously chosen NRB F1 and F2 in this analysis. The PSD of self-filtration layers in both
cases can be drawn for further confidence in selection of these barrier materials. In
summary, constriction-based filter design criteria can be applied to practical filter design
problems. Furthermore, their inclusion in the existing NRCS (1994) guideline makes the
designs more realistic.

100
15%
80
Percent Finer
60 F4
40 F3
Reactive Barriers
20 15%
0
0.01 0.1 1 10 100
Figure 8.5 PSDs of effective NRB F1 and potential reactive barriers F3 and F4
100 95%
85%
80
Percent Finer
60 CSD(F3)
Mod. Soil PSD
40 35%
Soil PSD
F3
20 d 85mod D c35
Reactive Barrier
0
0.01 0.1 1 10 100
Figure 8. 6 Constriction Analysis of RB F3 using Modified Base PSD method

100 95%
85%
80
Percent Finer
60
Mod. Soil PSD CSD(F4)
40 35%
Soil PSD F4
20
D c35 d 85mod Reactive Barrier
0
0.01 0.1 1 10 100
Figure 8. 7 Constriction Analysis of RB F4 using Modified Base PSD method
100
85%
80
Percent Finer
60 Non-Reactive Barrier
40 Acid Sulphate Soil

PSD
Reactive Barrier
20 35%
0
0.0001 0.001 0.01 0.1 1 10 100
Figure 8.8 Gradings of suitable reactive and non-reactive barrier materials

8.4 Summary
The particle size-based existing filter design guidelines are greatly enhanced by inclusion of
new constriction-based filter criteria in the procedure. These existing guidelines modified
for very coarse base soils are easy to use when compared with complex constriction-based
computations and analyses. These existing guidelines are still relevant and can be used as a
preliminary guide to identify the potential filter materials, especially when the filter
materials are not readily available for assessment. The use of the proposed constriction-
based criteria along with plotting the PSD of self-filtration layer will certainly boost the
designer’s confidence.

CHAPTER
NINE
CONCLUSIONS AND RECOMMENDATIONS
9.1 General Summary
In geotechnical engineering, a filter is designed to protect eroding base soils within or
behind a structure from erosion due to seepage. As water flows through the soil, fine
particles can be washed out, leading to internal erosion and eventually, the failure of the
structure. A correctly designed filter retains loose soil particles, thus preventing erosion. It
also allows seepage, preventing build-up of detrimental pore pressure. A filter is commonly
natural or manufactured sand and gravel or a geotextile. Filters are used in dams,
agricultural drainage, road pavements, retaining walls, canal linings, coastal protection,
landfills and so on. A geotextile filter is easier to install and a better quality assurance can
be achieved. However, it has a shorter performance history. This research study focuses on
granular filters.
Chapter 9: Conclusions and Recommendations 210

First time in early 1920s, Terzaghi suggested filter design criterion, D15/d854-5, based
some conceptual analyses and laboratory investigations on uniform sands. Most subsequent
studies ended up either merely investigating validity of or extending this criterion to other
soil types, particularly broadly-graded base soils. In the current professional practices, the
extension of the Terzaghi filter criterion is mainly accomplished by either regrading the
base soil grading curve for the particle size larger than the ASTM No. 4 sieve size (i.e.
4.75mm) or the use of a smaller representative base particle size such as d50 in broadly-
graded base soils. Despite the realisation that within filters, it is the constriction sizes rather
than the particles sizes that govern filtration, the current design guidelines are still based on
empirical considerations and particle sizes. As a consequence, these guidelines exhibit
some serious limitations.
In this study, a mathematical procedure has been established to determine the constriction
size distribution (CSD) of a filter. The procedure has been coded into a comprehensive
computer program to develop a CSD computation tool. Filter design criteria have been
suggested based on the constriction analyses of some fundamental principles of filtration.
These criteria have been verified against some well-known test data from literature as well
as the tests carried out during this research study. The performance of these filter criteria
has been compared in detail with the existing design guidelines. Finally, an enhanced
design guideline has been suggested by incorporating the constriction-based filter criteria in
the existing guideline.

9.2 Specific Observations
According to the theoretical and experimental results of this study, including the new
constriction-based criterion, the following concluding remarks and observations can be
drawn:
9.2.1 Constriction Analyses
1. Since during filtration the base particles infiltrate into the filter through the
constrictions, it is the constriction sizes within filters rather than the particle sizes
that govern filtration. Because the existing design guidelines are still based on
particle sizes, these guidelines exhibit some obvious limitations.
2. The particle size distribution (PSD) based on the surface area of the filter particles
model the filter in the best manner compared to those based on the mass or number
of the particles. If the PSD by mass of a filter as obtained after sieving is given, the
PSDs by surface area and number can be calculated mathematically (Chapter 3).
3. Like PSDs, the constriction size distributions (CSDs) of a uniform filter are
independent of the choice of particle frequencies (i.e. mass, number or surface). In
contrast, the CSDs of a well-graded filter are significantly different in three cases,
validating the previous finding. In this regard, the use of D15 in the existing
practices to represent a filter fails to simulate the unique filtration characteristics of
an individual filter.
4. The controlling constriction size of a granular filter is defined as the size of the
largest base particle that can penetrate a given filter, and can be given by the

constriction size, Dc35. Here, Dc35 is the constriction size of the filter where 35%
constrictions are finer than the size. This is a very important filter parameter and
provides a measure of the largest flow channel in the filter. This can also be
compared with the apparent opening size (AOS) of a geotextile and the aperture of a
mechanical sieve. In contrast to Kenney et al. (1985) empirical procedures, Dc35
forms a rigorous approach to estimate the controlling constriction size in the filter.
The Dc35 method distinguishes between uniform and well-graded filters by
estimating comparatively smaller in well-graded and also between loose and
compacted filters by estimating relatively smaller constrictions in the compacted
ones.
5. The self-filtering constriction size of a filter is defined as the size such that a base
particle larger than this size cannot move into the filter, and can be given by Dc95.
Here, Dc95 is defined as the constriction size of the filter where 95% of the
constrictions are finer than the size. This is another important filter parameter that
determines the self-filtering fraction of a base soil with respect to a given filter.
Base particles larger than this size do not influence self-filtration. This explains why
the coarser fraction of internally unstable gap-graded base soils is ignored or why
regrading is necessary in broadly-graded base soils when the Terzaghi criterion is
used. However, although regrading the base soil grading by a specific size such as
the ASTM No. 4 sieve size (NRCS 1994) appears to successfully describe filter
effectiveness in practical dam core soils, the NRCS approach is not realistic for two
reasons. Firstly, it fails to simulate the unique filtration characteristics of an

individual base soil. Secondly, it fails to describe the filtration of coarse base soils
where the whole grading curve lies beyond this size. Similarly, the use of d50 does
not simulate individual filtration properties of the base soils either. Moreover, it is a
conservative approach and very often makes filter designs unduly conservative
(Chapter 6). In this regard, the use of Dc95 forms a comprehensive method to
determine the self-filtering base fraction.
9.2.2 Surface Area Concept Applied to Base Soils
1. Very often, the base soils are cohensionless granular materials. In all filters, the
eroded base particles are transported to the filter where they are captured by filter
constrictions, and thus become an integral part of the filter; hence, like filters, the
surface area concept can also be applied to the granular base soils.
2. A granular filter can be compared with a mechanical sieve with its aperture
equivalent to the controlling constriction size, Dc35 of the filter. A granular filter-
mechanical sieve analogy provides a constriction-based filter criterion, Dc35 d85SA,
where d85SA is the base particle size where 85% base particles by surface area are
finer than the size. This filter criterion successfully separates effective filters from
ineffective counterparts in cohesionless granular soils (Chapter 4).
3. This surface area approach is based on analytical principles and obviates the need of
regrading. This further validates the theory that the surface area option is a more
realistic approach to describe the porous granular media. Because of large specific
surface area of fine clay and silt particles, the d85SA becomes unreasonably small.

This results in unduly conservative designs. Nevertheless, it distinctly describes the
filter boundary in the lateritic residual soils. This is because the soils predominantly
consist of uniform silt particles, where the choice of particle frequencies is
immaterial.
9.2.3 Stability of Self-filtration Layer
1. With the use of self-filtering constriction size, Dc95, it is possible to mathematically
determine the self-filtering fraction of a base soil with respect to a given filter. Once
the self-filtering fraction is determined, it is also possible to analytically calculate
the grading curve of self-filtration layer. This layer consists of the filter particles
and the base particles finer than the self-filtering constriction size, Dc95, in the
proportion defined by this study based on fundamental principles of soil mechanics
(Chapter 5). Plotting the grading curve of self-filtration layer, where “gaps”
gradually widen in the progressively coarser filters, provides a confident picture of
self-filtration process.
2. The assessment of the internal stability of self-filtration layer grading curve
procedure provides a comprehensive filter criterion. This criterion also successfully
demarcates the boundary between effective and ineffective filters in the granular
soils.
3. The Kenney and Lau (1985) stability assessment procedure is slightly conservative
in describing internal stability of broadly-graded base soils. As a result, the scope of
application of the proposed criterion is also limited. With a more realistic stability

assessment procedure, this filter criterion can be further enhanced to describe the
filtration in most base soils. Nevertheless, the plotting the grading curve of self-
filtration layer alone will definitely boost the designer’s confidence, particularly in
the critical filters.
4. This criterion is based on analytical principles. In contrast to the existing design
guidelines, it obviates regrading so it can describe filtration even in very coarse
filters.
9.2.4 Self-filtrating Base Fraction and Constriction-based Filter Criterion
1. The comparison of the controlling constriction size, Dc95, with the d85 of the self-
filtering fraction of the base soil (i.e. d85mod) provides a simple but comprehensive
filter design criterion, Dc35 d85mod. This criterion successfully describes the filter
effectiveness in most base soils and filter materials without being overly
conservative (Chapter 6).
2. In contrast to the existing design guidelines, this criterion is based on sound
analytical principles. The comparison of the proposed criterion with the classic
Terzaghi criterion reveals that in uniform base soils, Dc95 is invariably greater than
the largest base particle size, d100 and thus does not require regrading. In well-
graded base soils, Dc95 is usually smaller than d100, thus requiring regrading. As a
result, the classic Terzaghi criterion fails to describe filtration in well-graded and
broadly-graded base soils. However, regrading by a specific size as recommended

by the existing guidelines is empirical and unrealistic in many base soils,
particularly very coarse base soils.
3. The comparison of the proposed criterion with the existing design guidelines reveal
that the proposed criterion leads to more realistic designs without being
conservative and still maintaining a reasonable factor of safety. On the contrary, the
existing guidelines sometimes lead to over-conservative designs whereas some
unsafe ones are also likely.
4. Irrespective of the plasticity of base soils, the proposed criterion describes filtration
in most of them. This further substantiates the theory that the plasticity is not a
signification filtration parameter of the base soil.
5. Compared to the existing guidelines, the proposed criterion is computationally
complex. However, considering the capability of most computers today, the
computing task will not be difficult.
9.2.5 Experimental Observations
1. Laboratory tests carried out during this study on well-graded cohensionless soils (Cu
<20) show that some filters with retention ratio, D15/d85, of about 4-6 fail to retain
soils, confirming the trend that the filter effectiveness diminish in non-uniform base
soils.
2. Observations on cohesive soils reveal that some filters with retention ratio, D15/d85,
less than 9 failed. This suggests that the filter boundary, D15/d85=9, in the Category
1 cohesive soils (NRCS 1994) is slightly coarser and hence unrealistic.

9.2.6 Controversial USBR Observations
1. Reproducing a test to verify previous laboratory results of a similar test does not
appear to be a realistic approach. This is because very often, such laboratory tests
suffer from personal and/or laboratory biases. The statistical analyses favour the
data, which has majority. In the past, most laboratory investigations were carried
with uniform base and filter materials.
2. The analysis of the data using the proposed constriction-based criterion, Dc35
d85mod, clearly reveals that the USBR failure criteria were slightly conservative. A
few tests, which were assessed to be ineffective in the laboratory, actually did not
fail. However, majority of USBR test results were in agreement with established
principles of filtration. The tests successfully simulate the diminishing filter
effectiveness in well-graded base soils.
9.2.7 Enhanced Filter Design Guideline
1. The particle size-based existing filter design guidelines are greatly enhanced by
inclusion of new constriction-based filter criteria in the procedure.
2. These guidelines modified for very coarse base soils are easy to use when compared
with complex constriction-based computations and analyses. Their use should be
continued as a preliminary guide to identify the potential filter materials, especially
when the filter materials are not readily available for assessment.

3. The use of the constriction-based criteria developed in this study along with plotting
the PSD of self-filtration layer will certainly boost the designer’s confidence.
9.3 Future Filter Research
9.3.1 Re-evaluation of Critical Filters in Existing Dams
This study has clearly illustrated that the existing filter design guidelines in professional
practices exhibit some serious limitations, particularly in relation of filtration of well-
graded and broadly-graded base soils. There is some ambiguity with regard to the filtration
of cohesive base soils too. For these reasons, it is likely that some filter designs in existing
dams carried out on the basis of these guidelines might be unsatisfactory. In this respect,
reassessment of critical filters in existing dams in Australia or elsewhere may be a sensible
research project in the near future.
9.3.2 Evaluation of Crack Susceptibility of Critical Filters in Existing Dams
The existing design guidelines recommend that the filter materials must be non-cohesive
enough so that any crack in cohesive dam cores would not propagate through the filter. In
this regard, the filter materials must not contain more than 5% fines (i.e. particles <ASTM
No. 200 sieve size). During filtration of core materials, fine clay and silt particles are
eroded and retained in the filter. It is likely that by retaining these fine clay and silt size
base particles, the filters may themselves be cohesive and consequently susceptible to

cracking. With this research study, it is now possible to determine the amount of fines that
is retained within the self-filtration layer. In this respect, it is possible that a methodology
can be developed to determine a minimum amount of fines that makes a filter cohesive and
hence prone to cracking. Subsequently, this methodology can be applied to evaluate crack
susceptibility of the critical filters in existing dams. The constriction-based study of crack
susceptibility of critical filters in embankment dams is likely to be a good future project.
9.3.3 Clogging
There are evidences which show that the filters are physically clogged (Indraratna et al.
1990). They are mostly well-graded and broadly-graded filters, and the filters behind
suffosive and gap-graded base soils. However, a systematic approach in this regard is
lacking and so far, there is no guideline that can predict that a given filter is likely to be
clogged with respect to a given base soil. In this regard, a systematic and exclusive study of
clogging is an appropriate area to explore.
9.3.4 Internal Stability Assessment Method
Kenney and Lau (1985) developed a graphical procedure to evaluate the internal stability of
a cohensionless soil. This has been in professional practice since then. Earlier, it was
criticised to be overly conservative. It was amended in favour of criticism. However, the
analyses included in this study clearly illustrates that this procedure is still unrealistic and

conservative. With the constriction-based analyses, it is possible to develop an enhanced
method to evaluate the internal stability of a given soil.
9.3.5 Constriction Size and Permeability
Permeability of a granular soil is easily measured in the laboratory. Most past studies
attempted to relate the permeability to the finer particle sizes. However, there is no doubt
that the permeability will have a better correlation with the constriction of the granular soil.
With the development of the tool to compute the constriction size of a granular soil, a more
realistic relationship can be established between the constriction size and the permeability.
A realistic estimate of permeability of a granular soil can provide an alternative procedure
for the design of filters and drainage.
9.3.6 Geotextile Filters
Like the existing professional guidelines for the design of granular filters, the existing
guidelines for the design of geotextile filters (Carroll 1983; Luetich 1992) are also
empirical. The constriction-based approach developed during this research study can be
also extended to the design of geotextile filters.

9.3.7 Cyclic Behaviour of Granular Filters
Majority of granular filter studies in the past were carried out under static conditions where
a base soil was tested against a filter under mild to severe hydraulic conditions. Mild
vibration is sometimes imparted to the permeameter (filter equipment) in order to disturb
the bridging of base particles over a filter constriction. However, filters such as capping
layer of rail tracks or under the road pavements are often subjected to significant cyclic
loads. Under such conditions, granular filters may behave differently. Extending the main
concepts, developed in this study to the cyclic behaviour of granular filters, would be an
interesting research in design of modern road and rail tracks.

REFERENCES
Aberg (1993). “Washout of Grains from Filtered Sand and Gravel Materials”, Journal of
Geotechnical Engineering Division, ASCE, Vol. 119(1), pp. 36-53.
Bertram, G. E. (1940). “An experimental investigation of protective filters” Report,
Harvard Graduate School of Engineering, Publication No. 267, Vol.6.
Caroll, R.G. Jr. (1983). “Geotextile Filter Design” Engineering Fabrics in Transportation
Construction, TRR 916, TBR, Washington, DC, pp.44-53
Cedergren, H. R. (1989). “Seepage, Drainage, and Flow Nets”, 3rd Edition, Wiley, New
York
Dilema, E. L. G. (1990). “Development of permeability-floc size criteria for granular filter
design.” MSc Thesis, No. GT-89-4, Asian Institute of Technology, Bangkok, Thailand,
DeMello V. (1977). “Reflections on Design Decisions of Practical Significance to

Embankment Dams”, Geotechnique, Vol. 27(3), pp. 279-355.
Faure Y. and Gendrin P. (1989) “Filtration of a Granular Medium by Textiles” Powders

and Grains, Biarez & Gourves (Eds.) Balkema, Rotterdam, pp. 425-432.
Faure Y., Farkouth B., Delmas P., and Nancey A. (1996) “Valcros Dam: Summary of Tests
and Analysis of Filter Criteria” Geofilters ’96, Comptes Rendus Proceedings, Lafleur &
Rollin (Eds.), Bitech Publications, Canada
Federico F. and Musso A. (1985) “Pore size distribution in filtration analyses”, XI ICSM
E.S. Francis (Ed.), Vol III, pp 1207-1212
References 223
Federico F. and Musso A. (1993). “Some Advances in the Geometric-probabilistic Method
for Filter Design”, Proc. Filters in Geotechnical and Hydraulic Engineering, Brauns,
Heibaum & Schuler (eds.), 1993 Balkema, Rotterdam, pp.75-82
Fischer G. and Holtz R. (1996). “A Critical Review of Granular Soil Filter Retention
Criteria”, Proc. Geofilters ’96, Comptes Rendus Proceedings, Lafleur & Rollin (Eds.),
Bitech Publictions, Canada. pp. 409-418.
Foster, M. and Fell, R. (2001). “Assessing embankment dam filters that do not satisfy
design criteria” Journal of Geotechnical and Geoenvironmental Engineering., ASCE,
Vol. 127(5), pp.398-407
Foster M., Fell R., and Spannagle M. (1998). “Analysis of Embankment Dam Incidents”
UniCiv Report No. R-374, University of New South Wales, Australia.
Giroud J. (1996). “Granular Filters and Geotextile Filters”, Proc. Geofilters ’96, Comptes
Rendus Proceedings, Lafleur & Rollin (Eds.), Bitech Publictions, Canada, pp. 565-680.
de Groot M., Bakker K., Verheij H. (1993). “Design of Geometrically Open Filters in
Hydraulic Structures”, Proc. Filters in Geotechnical and Hydraulic Engineering,
Brauns, Heibaum & Schuler (eds.), 1993 Balkema, Rotterdam, pp.143-154
Honjo Y. and Veneziano D. (1989). “Improved Filter Criterion for Cohesionless Soils”,
Journal of Geotechnical Engineering Division, ASCE, Vol. 115, No. 1, pp 75-94.
Humes C. (1996). “A New Approach to Compute the Void-Size Distribution Curves of

Protective Filters” Proc. GeoFilters ’96, Lafleur & Mlynarek (Eds.), Bitech Publishing,
Canada, pp. 57-66.
ICOLD (1994). “Embankment Dams – Filters and Drains” Bulletin No. 95, ICOLD, France.
Indraratna B. Dilema E., Nutalaya P. (1990). “Design of Granular Filters for a Lateritic
Residual Soil”, Dam Engineering Vol. 1(3), AIT Bangkok, Thailand, pp. 201-220.
References 224
Indraratna, B. and Locke, M. (2000). “Analytical modeling and experimental verification of
granular filter behaviour.” Keynote Paper, Filters and Drainage in Geotech. & GeoEnv.
Eng., W. Wolski & J. Mlynarek (Eds.), Balkema, Rotterdam, pp.3-26.
Indraratna B., Vafai F., Dilema E. (1996). “An Experimental Study of the Filtration of a
Lateritic Clay Slurry by Sand Filters”, Proc. Institution of Civil Engineering
Geotechnical Engineering, Vol 119, pp. 75-83.
Indraratna B. and Vafai F. (1997). “Analytical Model for Particle Migration within Base
Soil - Filter System”, Journal of Geotechnical and Geoenvironmental Engineering,
ASCE, Vol 123, No. 2, pp. 100-109.
Karpoff K./USBR (1955). “The Use of Laboratory Tests to Develop Design Criteria for
Protective Filter”, Proc. 58th Annual Meeting ASTM, pp. 1183-1198.
Kenney T., Chahal R., Chiu E., Ofoegbu G., Omange G., Ume C. (1985). “Controlling
Constriction Sizes of Granular Filters”, Canadian Geotechnical Journal, Vol. 22, pp.
32-43
Kenney T. and Lau D. (1985). “Internal Stability of Granular Filters”, Canadian

Geotechnical Journal, Vol. 22, pp. 215-225.
Khor C. and Woo H. (1989) “Investigation of Crushed Rock Filters for Dam
Embankment”, Jour of Geotechnical Engineering, ASCE, Vol. 115(3), pp. 399-412.
Kovacs G. (1981) Seepage Hydraulics, Elsevier Publishing, Amsterdam, 570p.
Kwang T. (1990) “ Improvement of Dam Filter Criterion for Cohesionless Base Soil” M.
Eng. Thesis, Asian Institute of Technology, Bangkok, Thailand.
Lafleur J. (1984). “Filter Testing of Broadly Graded Cohesionless Tills” Canadian

Geotechnical Journal, Vol. 21, pp. 634-643.
References 225
Lafleur J., Mlynarek J., Rollin A. (1989). “Filtration of Broadly Graded Cohesionless
Soils”, Journal of Geotechnical Engineering Division, ASCE, Vol 115, No. 12, pp.
1747-1768.
Luettich, S.M., Giroud, J.P., and Bachus, R.C. (1992). “Geotextile Filter Design Guide”
Journal of Geotextiles and Geomembranes, Vol. 11, No.4-6, 1992, pp.19-34
Locke, M., Indraratna, B., and Adikari, G. (2001). “Time-dependent particle transport
through granular filters.” Journal of Geotechnical and Geoenvironmental Engineering,
Vol.127(6), pp.521-529.
Maranha das Neves E. (1989). “Analysis of Crack Erosion in Dam Cores. The Crack
Erosion Test” De Mello Volume, pp. 284-298
Natural Resources Conservation Services (1994). “Gradation design of sand and gravel
filters”, Chapter 26, Part 633 National Engineering Handbook, USDA
Reddi L. and Bonala M. (1997). “Analytical Solution for Fine Particle Accumulation in
Soil Filters”, Journal of Geotechnical Engineering Division, ASCE, Vol. 123, No. 12,
pp. 1143-1152.
Reddi L. Ming X., Hajra M., Lee I. (2000). “Permeability Reduction of Soil Filters due to
Physical Clogging” Journal of Geotechnical and Geoenvironmental Engineering,
ASCE, Vol. 126(3), pp. 236-246.
Ripley C. (1982). “Design of Filters For Clay Cores for Dams” Discussion by Vaughan and
Soares (1982), Journal of Geotechnical Engineering Division, ASCE, pp 1193-1195.
Schuler U. (1996). “Scattering of the Composition of Soils. An Aspect for the Stability of
Granular Filters”, Proc. Geofilters ’96, Comptes Rendus Proceedings, Lafleur & Rollin
(Eds.), Bitech Publictions, Canada. pp. 21-34
References 226
Schuler, U. and Brauns, J. (1993). “Behaviour of Coarse and Well-Graded Filters”, Proc.
Filters in Geotechnical and Hydraulic Engineering, Brauns, Heibum and Schuler (eds.),
Balkema, Rotterdam, pp.3-18
Sherard J. (1982) Discussion of “Design of Filters For Clay Cores for Dams” by Vaughan
and Soares (1982), Jour. Geotechnical Engng Div. ASCE, pp 1195-1197.
Sherard J., Dunnigan L., Talbot J. (1984a). “Basic Properties of Sand and Gravel Filters”
Journal of Geotechnical Engineering Division, ASCE, Vol 110, No. GT6, June 1984, pp
684-700.
Sherard J., Dunnigan L., Talbot J. (1984b) “Filters for Silts and Clays” Journal of
Geotechnical Engineering Division, ASCE, Vol. 110(6), pp. 701-718.
Sherard J. and Dunnigan L. (1985). “Filters and Leakage Control in Embankment Dams”
Proc. Symposium on Seepage and Leakage from Dams and Impoundments, R.L. Volpe
and W.E. Kelly eds., ASCE 1985, pp 1-30.
Sherard J. and Dunnigan L. (1989). “Critical Filters for Impervious Soils” Journal of
Geotechnical Engineering Division, ASCE, Vol 115, No. GT7, July1989, pp 927-947.
Sherard, J.L., Woodward, R.J., Cizienski, S.F., and Clevenger, W.A. (1963). Earth and
Earth-Rock Dams, John Wiley and Sons, New York
Silveira A. (1965). “An Analysis of the Problem of Washing through in Protective Filters”,
Proc. 6th Int. Conf. Soil Mechanics and Foundation Engineering, Canada, Vol. 2, pp.
551-555.
Silveira A., de Lorena Peixoto T., Nogueira J. (1975). “On Void Size Distribution of
Granular Materials”, Proc. 5th Pan-American Conf. Soil Mechanics and Foundation
Engineering, pp. 161-176.
Silveira A. (1993). “A Method for Determining the Void Size Distribution Curve for Filter
Materials” Proc. Filters in Geotechnical and Hydraulic Engineering, Brauns, Heibaum
& Schuler (eds.), 1993 Balkema, Rotterdam, pp. 71-74.
References 227
Soria M., Aramaki R., Viviani E. (1993). “Experimental Determination of Void Size
Curves” Proc. Filters in Geotechnical and Hydraulic Engineering, Brauns, Heibaum &
Schuler (eds.), 1993 Balkema, Rotterdam, pp. 43-48.
Terzaghi K. (1922). “Der Grundbruch an Stauwerken und Seine Verhuntung Forcheimer-

Nummer Wasserkr, 17, pp. 445-449, quoted by Vafai (1996).
USACE (1955) “Drainage and erosion control-subsurface drainage facilities for airfields”
Engineering Manual, Military Construction, Part XIII, Ch 2, Washington DC
USBR (1955). “The Use of Laboratory Tests to Develop Design Criteria for Protective
Filter”, Proc. 58th Annual Meeting ASTM, pp. 1183-1198 adopted from Karpoff
K.(1955)
U.S. Bureau of Reclamation, USBR, (1963). Earth Manual, First Edition (revised), US
Govt. Printing Office, Washington D.C., 751p.
USACE (1953). “Investigation of Filter Requirements for Underdrains.” Tech. Memo., No.
3-360, U.S. Waterways Experiment Station, Vicksburg, Mississippi.
U.S. Army Corps of Engineers, USACE, (1971). “Dewatering and Groundwater Control
for Deep Excavations”, Technical Memorandum No. 5-818-5 (April), Office of Chief of
Engineers, US Army, Washington D.C.
Vaughan P. (2000) “Filter Design for dam cores of clay, a retrospect” Filters and Drainage
in Geotechnical and Environmental Engineering, Wolski & Mlynarek Eds., Balkema,
Rotterdam, pp. 189-196.
Vaughan P. and Soares H. (1982). “Design of Filters for Clay Cores of Dams”, Journal of
Geotechnical Engineering Division, ASCE, Vol. 108, pp. 17-31.
References 228
Vafai F. (1996). “Analytical Modelling and Laboratory Studies of Particle Transport in
Filter Media”, PhD Thesis, University of Wollongong, NSW, Australia.
Witt K. (1993). “Reliability Study of Granular Filters”, Proc. Filters in Geotechnical and
Hydraulic Engineering, Brauns, Heibaum & Schuler (eds.), 1993 Balkema, Rotterdam,
pp.35-42
Wittmann L. (1979) “The Process of Soil Filtration – its Physics and the Approach in
Engineering Practice” Proc. 7th EuroConf. Soil Mech. and Fnd. Engng, pp. 303-310.
References 229

Mathematical Modelling of Granular Filters and Constriction-Based

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mathematical Modelling of Granular Filters and Constriction-Based

Uploaded by

Copyright:

Available Formats

University of Wollongong

Mathematical modelling of granular filters and

Research Online is the open access institutional repository for the

Constriction-based Filter Design Criteria

A thesis submitted in fulfilment of the

requirements for the award of the degree

Ashok Kumar Raut

BSc Eng (Distinction), M.Tech. (IIT Delhi)

Department of Civil Engineering

of Wollongong, is wholly my own work unless otherwise referenced or acknowledged. The

contrast, it allows unobstructed seepage, preventing build-up of detrimental pore pressure.

landfills and so on.

constrictions) that govern filtration. Consequently, these particle-based guidelines exhibit

some serious limitations.

In this thesis, limitations of current professional guidelines are comprehensively discussed.

A detailed mathematical procedure is developed to determine constriction sizes, and

subsequently, constriction-based criteria are proposed to describe filter effectiveness in

I am extremely grateful to my supervisor, Professor Buddhima Indraratna for his

attitude was inspiring to me.

I am equally grateful to my co-supervisor, Dr Hadi Khabbaz. His constant help and

cool head is enviable indeed.

I would like to express my sincere thanks to my fellow postgraduate students in

supports and encouragement throughout the study.

Thank you all.

“You cannot produce a baby in one month by impregnating nine women.”

 Raut, A. K. and Indraratna, B. (2007). “Analytical Approach to Geotextile Filter

Design”, Journal of Geotechnical and Geoenvironmental Engineering, ASCE (To be

submitted after review of the third paper)

 Raut, A. K. and Indraratna, B. (2007). “Constriction-based Analysis of Internal Stability

of Granular Soils”, Journal of Geotechnical and Geoenvironmental Engineering, ASCE

(To be submitted after review of the third paper)

 Raut, A. K. and Indraratna, B. (2006). “Limitations of Particle Size Approach in Filter

Design and Advancement through Constriction-Based Techniques”, Journal of

Geotechnical and Geoenvironmental Engineering, ASCE (Under review)

 Indraratna, B., Raut, A.K. and Khabbaz, H. (2007). “Constriction-based Retention

Criterion for Granular Filter Design.” Journal of Geotechnical and Geoenvironmental

Engineering, ASCE (To be published in March 2007)

Embankment Dams.” Journal of Geotechnical and Geoenvironmental Engineering,

ASCE, Vol 132, No.12, pp.1621-1627

Dams: A Conceptual Overview and Experimental Investigation.” ASCE/Geo-Institute

Proc. International Conference Geo2004: Advances in Geotechnical Engineering with

Emphasis on Dam Engineering, Irbid, Jordan, pp. 15-34

 Indraratna, B. and Raut, A.K. (2004). “Mathematical modelling of granular filters in

embankment dams.” Proc. 4th International Conference on Filters and Drainage in

Geotechnical and Environmental Engineering, Geofilters 2004, Johannesburg, SA,

 Raut, A.K. and Indraratna, B.(2004). ‘Constriction Size Distribution in a Granular

LIST OF PUBLICATIONS vii

LIST OF FIGURES xvi

LIST OF TABLES xxii

1.1 Granular Filters 1

1.2 Locations of Filters in Embankment Dams 2

1.3 Erosion Mechanisms and Properties of Granular Filters 4

1.4 Factors Affecting Filtration 6

1.5 Effective and Ineffective Filters 8

1.6 Relevance of Study 11

1.7 Research Aims 13

1.8 Scope and Organization of the Thesis 14

2.2 Empirical Investigations 20

2.2.1 Conventional Experimental Approaches in Filtration 20

2.2.2 No Erosion Filter Test (NEF Test) 23

2.2.3 Particle Size Ratio 26

2.2.5 Internally Unstable Base Soils 35

2.2.6 Particle Size -Permeability Relationship 37

Raut, A. K. and Indraratna, B. (2007). “Analytical Approach to Geotextile Filter

Raut, A. K. and Indraratna, B. (2007). “Constriction-based Analysis of Internal Stability

Raut, A. K. and Indraratna, B. (2006). “Limitations of Particle Size Approach in Filter

Indraratna, B., Raut, A.K. and Khabbaz, H. (2007). “Constriction-based Retention

Indraratna, B. and Raut, A.K. (2004). “Mathematical modelling of granular filters in

Raut, A.K. and Indraratna, B.(2004). ‘Constriction Size Distribution in a Granular